]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/core/dev.c
printk: ipv4 address digits printed in reverse order
[net-next-2.6.git] / net / core / dev.c
CommitLineData
1da177e4
LT
1/*
2 * NET3 Protocol independent device support routines.
3 *
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.
8 *
9 * Derived from the non IP parts of dev.c 1.0.19
02c30a84 10 * Authors: Ross Biro
1da177e4
LT
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
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>
21 *
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
73 */
74
75#include <asm/uaccess.h>
76#include <asm/system.h>
77#include <linux/bitops.h>
4fc268d2 78#include <linux/capability.h>
1da177e4
LT
79#include <linux/cpu.h>
80#include <linux/types.h>
81#include <linux/kernel.h>
82#include <linux/sched.h>
4a3e2f71 83#include <linux/mutex.h>
1da177e4
LT
84#include <linux/string.h>
85#include <linux/mm.h>
86#include <linux/socket.h>
87#include <linux/sockios.h>
88#include <linux/errno.h>
89#include <linux/interrupt.h>
90#include <linux/if_ether.h>
91#include <linux/netdevice.h>
92#include <linux/etherdevice.h>
0187bdfb 93#include <linux/ethtool.h>
1da177e4
LT
94#include <linux/notifier.h>
95#include <linux/skbuff.h>
457c4cbc 96#include <net/net_namespace.h>
1da177e4
LT
97#include <net/sock.h>
98#include <linux/rtnetlink.h>
99#include <linux/proc_fs.h>
100#include <linux/seq_file.h>
101#include <linux/stat.h>
102#include <linux/if_bridge.h>
b863ceb7 103#include <linux/if_macvlan.h>
1da177e4
LT
104#include <net/dst.h>
105#include <net/pkt_sched.h>
106#include <net/checksum.h>
107#include <linux/highmem.h>
108#include <linux/init.h>
109#include <linux/kmod.h>
110#include <linux/module.h>
111#include <linux/kallsyms.h>
112#include <linux/netpoll.h>
113#include <linux/rcupdate.h>
114#include <linux/delay.h>
295f4a1f 115#include <net/wext.h>
1da177e4 116#include <net/iw_handler.h>
1da177e4 117#include <asm/current.h>
5bdb9886 118#include <linux/audit.h>
db217334 119#include <linux/dmaengine.h>
f6a78bfc 120#include <linux/err.h>
c7fa9d18 121#include <linux/ctype.h>
723e98b7 122#include <linux/if_arp.h>
6de329e2 123#include <linux/if_vlan.h>
8f0f2223 124#include <linux/ip.h>
ad55dcaf 125#include <net/ip.h>
8f0f2223
DM
126#include <linux/ipv6.h>
127#include <linux/in.h>
b6b2fed1
DM
128#include <linux/jhash.h>
129#include <linux/random.h>
1da177e4 130
342709ef
PE
131#include "net-sysfs.h"
132
1da177e4
LT
133/*
134 * The list of packet types we will receive (as opposed to discard)
135 * and the routines to invoke.
136 *
137 * Why 16. Because with 16 the only overlap we get on a hash of the
138 * low nibble of the protocol value is RARP/SNAP/X.25.
139 *
140 * NOTE: That is no longer true with the addition of VLAN tags. Not
141 * sure which should go first, but I bet it won't make much
142 * difference if we are running VLANs. The good news is that
143 * this protocol won't be in the list unless compiled in, so
3041a069 144 * the average user (w/out VLANs) will not be adversely affected.
1da177e4
LT
145 * --BLG
146 *
147 * 0800 IP
148 * 8100 802.1Q VLAN
149 * 0001 802.3
150 * 0002 AX.25
151 * 0004 802.2
152 * 8035 RARP
153 * 0005 SNAP
154 * 0805 X.25
155 * 0806 ARP
156 * 8137 IPX
157 * 0009 Localtalk
158 * 86DD IPv6
159 */
160
82d8a867
PE
161#define PTYPE_HASH_SIZE (16)
162#define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
163
1da177e4 164static DEFINE_SPINLOCK(ptype_lock);
82d8a867 165static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
6b2bedc3 166static struct list_head ptype_all __read_mostly; /* Taps */
1da177e4 167
db217334 168#ifdef CONFIG_NET_DMA
d379b01e
DW
169struct net_dma {
170 struct dma_client client;
171 spinlock_t lock;
172 cpumask_t channel_mask;
0c0b0aca 173 struct dma_chan **channels;
d379b01e
DW
174};
175
176static enum dma_state_client
177netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
178 enum dma_state state);
179
180static struct net_dma net_dma = {
181 .client = {
182 .event_callback = netdev_dma_event,
183 },
184};
db217334
CL
185#endif
186
1da177e4 187/*
7562f876 188 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
1da177e4
LT
189 * semaphore.
190 *
191 * Pure readers hold dev_base_lock for reading.
192 *
193 * Writers must hold the rtnl semaphore while they loop through the
7562f876 194 * dev_base_head list, and hold dev_base_lock for writing when they do the
1da177e4
LT
195 * actual updates. This allows pure readers to access the list even
196 * while a writer is preparing to update it.
197 *
198 * To put it another way, dev_base_lock is held for writing only to
199 * protect against pure readers; the rtnl semaphore provides the
200 * protection against other writers.
201 *
202 * See, for example usages, register_netdevice() and
203 * unregister_netdevice(), which must be called with the rtnl
204 * semaphore held.
205 */
1da177e4
LT
206DEFINE_RWLOCK(dev_base_lock);
207
1da177e4
LT
208EXPORT_SYMBOL(dev_base_lock);
209
210#define NETDEV_HASHBITS 8
881d966b 211#define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
1da177e4 212
881d966b 213static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
1da177e4
LT
214{
215 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
881d966b 216 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
1da177e4
LT
217}
218
881d966b 219static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
1da177e4 220{
881d966b 221 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
1da177e4
LT
222}
223
ce286d32
EB
224/* Device list insertion */
225static int list_netdevice(struct net_device *dev)
226{
c346dca1 227 struct net *net = dev_net(dev);
ce286d32
EB
228
229 ASSERT_RTNL();
230
231 write_lock_bh(&dev_base_lock);
232 list_add_tail(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
235 write_unlock_bh(&dev_base_lock);
236 return 0;
237}
238
239/* Device list removal */
240static void unlist_netdevice(struct net_device *dev)
241{
242 ASSERT_RTNL();
243
244 /* Unlink dev from the device chain */
245 write_lock_bh(&dev_base_lock);
246 list_del(&dev->dev_list);
247 hlist_del(&dev->name_hlist);
248 hlist_del(&dev->index_hlist);
249 write_unlock_bh(&dev_base_lock);
250}
251
1da177e4
LT
252/*
253 * Our notifier list
254 */
255
f07d5b94 256static RAW_NOTIFIER_HEAD(netdev_chain);
1da177e4
LT
257
258/*
259 * Device drivers call our routines to queue packets here. We empty the
260 * queue in the local softnet handler.
261 */
bea3348e
SH
262
263DEFINE_PER_CPU(struct softnet_data, softnet_data);
1da177e4 264
cf508b12 265#ifdef CONFIG_LOCKDEP
723e98b7 266/*
c773e847 267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
723e98b7
JP
268 * according to dev->type
269 */
270static const unsigned short netdev_lock_type[] =
271 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
272 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
273 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
274 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
275 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
276 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
277 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
278 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
279 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
280 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
281 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
282 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
283 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
284 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
285 ARPHRD_NONE};
286
287static const char *netdev_lock_name[] =
288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
300 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
301 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
302 "_xmit_NONE"};
303
304static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
cf508b12 305static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
723e98b7
JP
306
307static inline unsigned short netdev_lock_pos(unsigned short dev_type)
308{
309 int i;
310
311 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
312 if (netdev_lock_type[i] == dev_type)
313 return i;
314 /* the last key is used by default */
315 return ARRAY_SIZE(netdev_lock_type) - 1;
316}
317
cf508b12
DM
318static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
319 unsigned short dev_type)
723e98b7
JP
320{
321 int i;
322
323 i = netdev_lock_pos(dev_type);
324 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
325 netdev_lock_name[i]);
326}
cf508b12
DM
327
328static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
329{
330 int i;
331
332 i = netdev_lock_pos(dev->type);
333 lockdep_set_class_and_name(&dev->addr_list_lock,
334 &netdev_addr_lock_key[i],
335 netdev_lock_name[i]);
336}
723e98b7 337#else
cf508b12
DM
338static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
340{
341}
342static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
723e98b7
JP
343{
344}
345#endif
1da177e4
LT
346
347/*******************************************************************************
348
349 Protocol management and registration routines
350
351*******************************************************************************/
352
1da177e4
LT
353/*
354 * Add a protocol ID to the list. Now that the input handler is
355 * smarter we can dispense with all the messy stuff that used to be
356 * here.
357 *
358 * BEWARE!!! Protocol handlers, mangling input packets,
359 * MUST BE last in hash buckets and checking protocol handlers
360 * MUST start from promiscuous ptype_all chain in net_bh.
361 * It is true now, do not change it.
362 * Explanation follows: if protocol handler, mangling packet, will
363 * be the first on list, it is not able to sense, that packet
364 * is cloned and should be copied-on-write, so that it will
365 * change it and subsequent readers will get broken packet.
366 * --ANK (980803)
367 */
368
369/**
370 * dev_add_pack - add packet handler
371 * @pt: packet type declaration
372 *
373 * Add a protocol handler to the networking stack. The passed &packet_type
374 * is linked into kernel lists and may not be freed until it has been
375 * removed from the kernel lists.
376 *
4ec93edb 377 * This call does not sleep therefore it can not
1da177e4
LT
378 * guarantee all CPU's that are in middle of receiving packets
379 * will see the new packet type (until the next received packet).
380 */
381
382void dev_add_pack(struct packet_type *pt)
383{
384 int hash;
385
386 spin_lock_bh(&ptype_lock);
9be9a6b9 387 if (pt->type == htons(ETH_P_ALL))
1da177e4 388 list_add_rcu(&pt->list, &ptype_all);
9be9a6b9 389 else {
82d8a867 390 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
1da177e4
LT
391 list_add_rcu(&pt->list, &ptype_base[hash]);
392 }
393 spin_unlock_bh(&ptype_lock);
394}
395
1da177e4
LT
396/**
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
399 *
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
4ec93edb 403 * returns.
1da177e4
LT
404 *
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
408 */
409void __dev_remove_pack(struct packet_type *pt)
410{
411 struct list_head *head;
412 struct packet_type *pt1;
413
414 spin_lock_bh(&ptype_lock);
415
9be9a6b9 416 if (pt->type == htons(ETH_P_ALL))
1da177e4 417 head = &ptype_all;
9be9a6b9 418 else
82d8a867 419 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
1da177e4
LT
420
421 list_for_each_entry(pt1, head, list) {
422 if (pt == pt1) {
423 list_del_rcu(&pt->list);
424 goto out;
425 }
426 }
427
428 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
429out:
430 spin_unlock_bh(&ptype_lock);
431}
432/**
433 * dev_remove_pack - remove packet handler
434 * @pt: packet type declaration
435 *
436 * Remove a protocol handler that was previously added to the kernel
437 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
438 * from the kernel lists and can be freed or reused once this function
439 * returns.
440 *
441 * This call sleeps to guarantee that no CPU is looking at the packet
442 * type after return.
443 */
444void dev_remove_pack(struct packet_type *pt)
445{
446 __dev_remove_pack(pt);
4ec93edb 447
1da177e4
LT
448 synchronize_net();
449}
450
451/******************************************************************************
452
453 Device Boot-time Settings Routines
454
455*******************************************************************************/
456
457/* Boot time configuration table */
458static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
459
460/**
461 * netdev_boot_setup_add - add new setup entry
462 * @name: name of the device
463 * @map: configured settings for the device
464 *
465 * Adds new setup entry to the dev_boot_setup list. The function
466 * returns 0 on error and 1 on success. This is a generic routine to
467 * all netdevices.
468 */
469static int netdev_boot_setup_add(char *name, struct ifmap *map)
470{
471 struct netdev_boot_setup *s;
472 int i;
473
474 s = dev_boot_setup;
475 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
476 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
477 memset(s[i].name, 0, sizeof(s[i].name));
93b3cff9 478 strlcpy(s[i].name, name, IFNAMSIZ);
1da177e4
LT
479 memcpy(&s[i].map, map, sizeof(s[i].map));
480 break;
481 }
482 }
483
484 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
485}
486
487/**
488 * netdev_boot_setup_check - check boot time settings
489 * @dev: the netdevice
490 *
491 * Check boot time settings for the device.
492 * The found settings are set for the device to be used
493 * later in the device probing.
494 * Returns 0 if no settings found, 1 if they are.
495 */
496int netdev_boot_setup_check(struct net_device *dev)
497{
498 struct netdev_boot_setup *s = dev_boot_setup;
499 int i;
500
501 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
502 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
93b3cff9 503 !strcmp(dev->name, s[i].name)) {
1da177e4
LT
504 dev->irq = s[i].map.irq;
505 dev->base_addr = s[i].map.base_addr;
506 dev->mem_start = s[i].map.mem_start;
507 dev->mem_end = s[i].map.mem_end;
508 return 1;
509 }
510 }
511 return 0;
512}
513
514
515/**
516 * netdev_boot_base - get address from boot time settings
517 * @prefix: prefix for network device
518 * @unit: id for network device
519 *
520 * Check boot time settings for the base address of device.
521 * The found settings are set for the device to be used
522 * later in the device probing.
523 * Returns 0 if no settings found.
524 */
525unsigned long netdev_boot_base(const char *prefix, int unit)
526{
527 const struct netdev_boot_setup *s = dev_boot_setup;
528 char name[IFNAMSIZ];
529 int i;
530
531 sprintf(name, "%s%d", prefix, unit);
532
533 /*
534 * If device already registered then return base of 1
535 * to indicate not to probe for this interface
536 */
881d966b 537 if (__dev_get_by_name(&init_net, name))
1da177e4
LT
538 return 1;
539
540 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
541 if (!strcmp(name, s[i].name))
542 return s[i].map.base_addr;
543 return 0;
544}
545
546/*
547 * Saves at boot time configured settings for any netdevice.
548 */
549int __init netdev_boot_setup(char *str)
550{
551 int ints[5];
552 struct ifmap map;
553
554 str = get_options(str, ARRAY_SIZE(ints), ints);
555 if (!str || !*str)
556 return 0;
557
558 /* Save settings */
559 memset(&map, 0, sizeof(map));
560 if (ints[0] > 0)
561 map.irq = ints[1];
562 if (ints[0] > 1)
563 map.base_addr = ints[2];
564 if (ints[0] > 2)
565 map.mem_start = ints[3];
566 if (ints[0] > 3)
567 map.mem_end = ints[4];
568
569 /* Add new entry to the list */
570 return netdev_boot_setup_add(str, &map);
571}
572
573__setup("netdev=", netdev_boot_setup);
574
575/*******************************************************************************
576
577 Device Interface Subroutines
578
579*******************************************************************************/
580
581/**
582 * __dev_get_by_name - find a device by its name
c4ea43c5 583 * @net: the applicable net namespace
1da177e4
LT
584 * @name: name to find
585 *
586 * Find an interface by name. Must be called under RTNL semaphore
587 * or @dev_base_lock. If the name is found a pointer to the device
588 * is returned. If the name is not found then %NULL is returned. The
589 * reference counters are not incremented so the caller must be
590 * careful with locks.
591 */
592
881d966b 593struct net_device *__dev_get_by_name(struct net *net, const char *name)
1da177e4
LT
594{
595 struct hlist_node *p;
596
881d966b 597 hlist_for_each(p, dev_name_hash(net, name)) {
1da177e4
LT
598 struct net_device *dev
599 = hlist_entry(p, struct net_device, name_hlist);
600 if (!strncmp(dev->name, name, IFNAMSIZ))
601 return dev;
602 }
603 return NULL;
604}
605
606/**
607 * dev_get_by_name - find a device by its name
c4ea43c5 608 * @net: the applicable net namespace
1da177e4
LT
609 * @name: name to find
610 *
611 * Find an interface by name. This can be called from any
612 * context and does its own locking. The returned handle has
613 * the usage count incremented and the caller must use dev_put() to
614 * release it when it is no longer needed. %NULL is returned if no
615 * matching device is found.
616 */
617
881d966b 618struct net_device *dev_get_by_name(struct net *net, const char *name)
1da177e4
LT
619{
620 struct net_device *dev;
621
622 read_lock(&dev_base_lock);
881d966b 623 dev = __dev_get_by_name(net, name);
1da177e4
LT
624 if (dev)
625 dev_hold(dev);
626 read_unlock(&dev_base_lock);
627 return dev;
628}
629
630/**
631 * __dev_get_by_index - find a device by its ifindex
c4ea43c5 632 * @net: the applicable net namespace
1da177e4
LT
633 * @ifindex: index of device
634 *
635 * Search for an interface by index. Returns %NULL if the device
636 * is not found or a pointer to the device. The device has not
637 * had its reference counter increased so the caller must be careful
638 * about locking. The caller must hold either the RTNL semaphore
639 * or @dev_base_lock.
640 */
641
881d966b 642struct net_device *__dev_get_by_index(struct net *net, int ifindex)
1da177e4
LT
643{
644 struct hlist_node *p;
645
881d966b 646 hlist_for_each(p, dev_index_hash(net, ifindex)) {
1da177e4
LT
647 struct net_device *dev
648 = hlist_entry(p, struct net_device, index_hlist);
649 if (dev->ifindex == ifindex)
650 return dev;
651 }
652 return NULL;
653}
654
655
656/**
657 * dev_get_by_index - find a device by its ifindex
c4ea43c5 658 * @net: the applicable net namespace
1da177e4
LT
659 * @ifindex: index of device
660 *
661 * Search for an interface by index. Returns NULL if the device
662 * is not found or a pointer to the device. The device returned has
663 * had a reference added and the pointer is safe until the user calls
664 * dev_put to indicate they have finished with it.
665 */
666
881d966b 667struct net_device *dev_get_by_index(struct net *net, int ifindex)
1da177e4
LT
668{
669 struct net_device *dev;
670
671 read_lock(&dev_base_lock);
881d966b 672 dev = __dev_get_by_index(net, ifindex);
1da177e4
LT
673 if (dev)
674 dev_hold(dev);
675 read_unlock(&dev_base_lock);
676 return dev;
677}
678
679/**
680 * dev_getbyhwaddr - find a device by its hardware address
c4ea43c5 681 * @net: the applicable net namespace
1da177e4
LT
682 * @type: media type of device
683 * @ha: hardware address
684 *
685 * Search for an interface by MAC address. Returns NULL if the device
686 * is not found or a pointer to the device. The caller must hold the
687 * rtnl semaphore. The returned device has not had its ref count increased
688 * and the caller must therefore be careful about locking
689 *
690 * BUGS:
691 * If the API was consistent this would be __dev_get_by_hwaddr
692 */
693
881d966b 694struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
1da177e4
LT
695{
696 struct net_device *dev;
697
698 ASSERT_RTNL();
699
81103a52 700 for_each_netdev(net, dev)
1da177e4
LT
701 if (dev->type == type &&
702 !memcmp(dev->dev_addr, ha, dev->addr_len))
7562f876
PE
703 return dev;
704
705 return NULL;
1da177e4
LT
706}
707
cf309e3f
JF
708EXPORT_SYMBOL(dev_getbyhwaddr);
709
881d966b 710struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
1da177e4
LT
711{
712 struct net_device *dev;
713
4e9cac2b 714 ASSERT_RTNL();
881d966b 715 for_each_netdev(net, dev)
4e9cac2b 716 if (dev->type == type)
7562f876
PE
717 return dev;
718
719 return NULL;
4e9cac2b
PM
720}
721
722EXPORT_SYMBOL(__dev_getfirstbyhwtype);
723
881d966b 724struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
4e9cac2b
PM
725{
726 struct net_device *dev;
727
728 rtnl_lock();
881d966b 729 dev = __dev_getfirstbyhwtype(net, type);
4e9cac2b
PM
730 if (dev)
731 dev_hold(dev);
1da177e4
LT
732 rtnl_unlock();
733 return dev;
734}
735
736EXPORT_SYMBOL(dev_getfirstbyhwtype);
737
738/**
739 * dev_get_by_flags - find any device with given flags
c4ea43c5 740 * @net: the applicable net namespace
1da177e4
LT
741 * @if_flags: IFF_* values
742 * @mask: bitmask of bits in if_flags to check
743 *
744 * Search for any interface with the given flags. Returns NULL if a device
4ec93edb 745 * is not found or a pointer to the device. The device returned has
1da177e4
LT
746 * had a reference added and the pointer is safe until the user calls
747 * dev_put to indicate they have finished with it.
748 */
749
881d966b 750struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
1da177e4 751{
7562f876 752 struct net_device *dev, *ret;
1da177e4 753
7562f876 754 ret = NULL;
1da177e4 755 read_lock(&dev_base_lock);
881d966b 756 for_each_netdev(net, dev) {
1da177e4
LT
757 if (((dev->flags ^ if_flags) & mask) == 0) {
758 dev_hold(dev);
7562f876 759 ret = dev;
1da177e4
LT
760 break;
761 }
762 }
763 read_unlock(&dev_base_lock);
7562f876 764 return ret;
1da177e4
LT
765}
766
767/**
768 * dev_valid_name - check if name is okay for network device
769 * @name: name string
770 *
771 * Network device names need to be valid file names to
c7fa9d18
DM
772 * to allow sysfs to work. We also disallow any kind of
773 * whitespace.
1da177e4 774 */
c2373ee9 775int dev_valid_name(const char *name)
1da177e4 776{
c7fa9d18
DM
777 if (*name == '\0')
778 return 0;
b6fe17d6
SH
779 if (strlen(name) >= IFNAMSIZ)
780 return 0;
c7fa9d18
DM
781 if (!strcmp(name, ".") || !strcmp(name, ".."))
782 return 0;
783
784 while (*name) {
785 if (*name == '/' || isspace(*name))
786 return 0;
787 name++;
788 }
789 return 1;
1da177e4
LT
790}
791
792/**
b267b179
EB
793 * __dev_alloc_name - allocate a name for a device
794 * @net: network namespace to allocate the device name in
1da177e4 795 * @name: name format string
b267b179 796 * @buf: scratch buffer and result name string
1da177e4
LT
797 *
798 * Passed a format string - eg "lt%d" it will try and find a suitable
3041a069
SH
799 * id. It scans list of devices to build up a free map, then chooses
800 * the first empty slot. The caller must hold the dev_base or rtnl lock
801 * while allocating the name and adding the device in order to avoid
802 * duplicates.
803 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
804 * Returns the number of the unit assigned or a negative errno code.
1da177e4
LT
805 */
806
b267b179 807static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1da177e4
LT
808{
809 int i = 0;
1da177e4
LT
810 const char *p;
811 const int max_netdevices = 8*PAGE_SIZE;
cfcabdcc 812 unsigned long *inuse;
1da177e4
LT
813 struct net_device *d;
814
815 p = strnchr(name, IFNAMSIZ-1, '%');
816 if (p) {
817 /*
818 * Verify the string as this thing may have come from
819 * the user. There must be either one "%d" and no other "%"
820 * characters.
821 */
822 if (p[1] != 'd' || strchr(p + 2, '%'))
823 return -EINVAL;
824
825 /* Use one page as a bit array of possible slots */
cfcabdcc 826 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1da177e4
LT
827 if (!inuse)
828 return -ENOMEM;
829
881d966b 830 for_each_netdev(net, d) {
1da177e4
LT
831 if (!sscanf(d->name, name, &i))
832 continue;
833 if (i < 0 || i >= max_netdevices)
834 continue;
835
836 /* avoid cases where sscanf is not exact inverse of printf */
b267b179 837 snprintf(buf, IFNAMSIZ, name, i);
1da177e4
LT
838 if (!strncmp(buf, d->name, IFNAMSIZ))
839 set_bit(i, inuse);
840 }
841
842 i = find_first_zero_bit(inuse, max_netdevices);
843 free_page((unsigned long) inuse);
844 }
845
b267b179
EB
846 snprintf(buf, IFNAMSIZ, name, i);
847 if (!__dev_get_by_name(net, buf))
1da177e4 848 return i;
1da177e4
LT
849
850 /* It is possible to run out of possible slots
851 * when the name is long and there isn't enough space left
852 * for the digits, or if all bits are used.
853 */
854 return -ENFILE;
855}
856
b267b179
EB
857/**
858 * dev_alloc_name - allocate a name for a device
859 * @dev: device
860 * @name: name format string
861 *
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
866 * duplicates.
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
869 */
870
871int dev_alloc_name(struct net_device *dev, const char *name)
872{
873 char buf[IFNAMSIZ];
874 struct net *net;
875 int ret;
876
c346dca1
YH
877 BUG_ON(!dev_net(dev));
878 net = dev_net(dev);
b267b179
EB
879 ret = __dev_alloc_name(net, name, buf);
880 if (ret >= 0)
881 strlcpy(dev->name, buf, IFNAMSIZ);
882 return ret;
883}
884
1da177e4
LT
885
886/**
887 * dev_change_name - change name of a device
888 * @dev: device
889 * @newname: name (or format string) must be at least IFNAMSIZ
890 *
891 * Change name of a device, can pass format strings "eth%d".
892 * for wildcarding.
893 */
cf04a4c7 894int dev_change_name(struct net_device *dev, const char *newname)
1da177e4 895{
fcc5a03a 896 char oldname[IFNAMSIZ];
1da177e4 897 int err = 0;
fcc5a03a 898 int ret;
881d966b 899 struct net *net;
1da177e4
LT
900
901 ASSERT_RTNL();
c346dca1 902 BUG_ON(!dev_net(dev));
1da177e4 903
c346dca1 904 net = dev_net(dev);
1da177e4
LT
905 if (dev->flags & IFF_UP)
906 return -EBUSY;
907
908 if (!dev_valid_name(newname))
909 return -EINVAL;
910
c8d90dca
SH
911 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
912 return 0;
913
fcc5a03a
HX
914 memcpy(oldname, dev->name, IFNAMSIZ);
915
1da177e4
LT
916 if (strchr(newname, '%')) {
917 err = dev_alloc_name(dev, newname);
918 if (err < 0)
919 return err;
1da177e4 920 }
881d966b 921 else if (__dev_get_by_name(net, newname))
1da177e4
LT
922 return -EEXIST;
923 else
924 strlcpy(dev->name, newname, IFNAMSIZ);
925
fcc5a03a 926rollback:
3891845e
EB
927 /* For now only devices in the initial network namespace
928 * are in sysfs.
929 */
930 if (net == &init_net) {
931 ret = device_rename(&dev->dev, dev->name);
932 if (ret) {
933 memcpy(dev->name, oldname, IFNAMSIZ);
934 return ret;
935 }
dcc99773 936 }
7f988eab
HX
937
938 write_lock_bh(&dev_base_lock);
92749821 939 hlist_del(&dev->name_hlist);
881d966b 940 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
7f988eab
HX
941 write_unlock_bh(&dev_base_lock);
942
056925ab 943 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
fcc5a03a
HX
944 ret = notifier_to_errno(ret);
945
946 if (ret) {
947 if (err) {
948 printk(KERN_ERR
949 "%s: name change rollback failed: %d.\n",
950 dev->name, ret);
951 } else {
952 err = ret;
953 memcpy(dev->name, oldname, IFNAMSIZ);
954 goto rollback;
955 }
956 }
1da177e4
LT
957
958 return err;
959}
960
0b815a1a
SH
961/**
962 * dev_set_alias - change ifalias of a device
963 * @dev: device
964 * @alias: name up to IFALIASZ
f0db275a 965 * @len: limit of bytes to copy from info
0b815a1a
SH
966 *
967 * Set ifalias for a device,
968 */
969int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
970{
971 ASSERT_RTNL();
972
973 if (len >= IFALIASZ)
974 return -EINVAL;
975
96ca4a2c
OH
976 if (!len) {
977 if (dev->ifalias) {
978 kfree(dev->ifalias);
979 dev->ifalias = NULL;
980 }
981 return 0;
982 }
983
0b815a1a
SH
984 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
985 if (!dev->ifalias)
986 return -ENOMEM;
987
988 strlcpy(dev->ifalias, alias, len+1);
989 return len;
990}
991
992
d8a33ac4 993/**
3041a069 994 * netdev_features_change - device changes features
d8a33ac4
SH
995 * @dev: device to cause notification
996 *
997 * Called to indicate a device has changed features.
998 */
999void netdev_features_change(struct net_device *dev)
1000{
056925ab 1001 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
d8a33ac4
SH
1002}
1003EXPORT_SYMBOL(netdev_features_change);
1004
1da177e4
LT
1005/**
1006 * netdev_state_change - device changes state
1007 * @dev: device to cause notification
1008 *
1009 * Called to indicate a device has changed state. This function calls
1010 * the notifier chains for netdev_chain and sends a NEWLINK message
1011 * to the routing socket.
1012 */
1013void netdev_state_change(struct net_device *dev)
1014{
1015 if (dev->flags & IFF_UP) {
056925ab 1016 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1da177e4
LT
1017 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1018 }
1019}
1020
c1da4ac7
OG
1021void netdev_bonding_change(struct net_device *dev)
1022{
1023 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1024}
1025EXPORT_SYMBOL(netdev_bonding_change);
1026
1da177e4
LT
1027/**
1028 * dev_load - load a network module
c4ea43c5 1029 * @net: the applicable net namespace
1da177e4
LT
1030 * @name: name of interface
1031 *
1032 * If a network interface is not present and the process has suitable
1033 * privileges this function loads the module. If module loading is not
1034 * available in this kernel then it becomes a nop.
1035 */
1036
881d966b 1037void dev_load(struct net *net, const char *name)
1da177e4 1038{
4ec93edb 1039 struct net_device *dev;
1da177e4
LT
1040
1041 read_lock(&dev_base_lock);
881d966b 1042 dev = __dev_get_by_name(net, name);
1da177e4
LT
1043 read_unlock(&dev_base_lock);
1044
1045 if (!dev && capable(CAP_SYS_MODULE))
1046 request_module("%s", name);
1047}
1048
1da177e4
LT
1049/**
1050 * dev_open - prepare an interface for use.
1051 * @dev: device to open
1052 *
1053 * Takes a device from down to up state. The device's private open
1054 * function is invoked and then the multicast lists are loaded. Finally
1055 * the device is moved into the up state and a %NETDEV_UP message is
1056 * sent to the netdev notifier chain.
1057 *
1058 * Calling this function on an active interface is a nop. On a failure
1059 * a negative errno code is returned.
1060 */
1061int dev_open(struct net_device *dev)
1062{
1063 int ret = 0;
1064
e46b66bc
BH
1065 ASSERT_RTNL();
1066
1da177e4
LT
1067 /*
1068 * Is it already up?
1069 */
1070
1071 if (dev->flags & IFF_UP)
1072 return 0;
1073
1074 /*
1075 * Is it even present?
1076 */
1077 if (!netif_device_present(dev))
1078 return -ENODEV;
1079
1080 /*
1081 * Call device private open method
1082 */
1083 set_bit(__LINK_STATE_START, &dev->state);
bada339b
JG
1084
1085 if (dev->validate_addr)
1086 ret = dev->validate_addr(dev);
1087
1088 if (!ret && dev->open)
1da177e4 1089 ret = dev->open(dev);
1da177e4 1090
4ec93edb 1091 /*
1da177e4
LT
1092 * If it went open OK then:
1093 */
1094
bada339b
JG
1095 if (ret)
1096 clear_bit(__LINK_STATE_START, &dev->state);
1097 else {
1da177e4
LT
1098 /*
1099 * Set the flags.
1100 */
1101 dev->flags |= IFF_UP;
1102
1103 /*
1104 * Initialize multicasting status
1105 */
4417da66 1106 dev_set_rx_mode(dev);
1da177e4
LT
1107
1108 /*
1109 * Wakeup transmit queue engine
1110 */
1111 dev_activate(dev);
1112
1113 /*
1114 * ... and announce new interface.
1115 */
056925ab 1116 call_netdevice_notifiers(NETDEV_UP, dev);
1da177e4 1117 }
bada339b 1118
1da177e4
LT
1119 return ret;
1120}
1121
1122/**
1123 * dev_close - shutdown an interface.
1124 * @dev: device to shutdown
1125 *
1126 * This function moves an active device into down state. A
1127 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1128 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1129 * chain.
1130 */
1131int dev_close(struct net_device *dev)
1132{
e46b66bc
BH
1133 ASSERT_RTNL();
1134
9d5010db
DM
1135 might_sleep();
1136
1da177e4
LT
1137 if (!(dev->flags & IFF_UP))
1138 return 0;
1139
1140 /*
1141 * Tell people we are going down, so that they can
1142 * prepare to death, when device is still operating.
1143 */
056925ab 1144 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1da177e4 1145
1da177e4
LT
1146 clear_bit(__LINK_STATE_START, &dev->state);
1147
1148 /* Synchronize to scheduled poll. We cannot touch poll list,
bea3348e
SH
1149 * it can be even on different cpu. So just clear netif_running().
1150 *
1151 * dev->stop() will invoke napi_disable() on all of it's
1152 * napi_struct instances on this device.
1153 */
1da177e4 1154 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1da177e4 1155
d8b2a4d2
ML
1156 dev_deactivate(dev);
1157
1da177e4
LT
1158 /*
1159 * Call the device specific close. This cannot fail.
1160 * Only if device is UP
1161 *
1162 * We allow it to be called even after a DETACH hot-plug
1163 * event.
1164 */
1165 if (dev->stop)
1166 dev->stop(dev);
1167
1168 /*
1169 * Device is now down.
1170 */
1171
1172 dev->flags &= ~IFF_UP;
1173
1174 /*
1175 * Tell people we are down
1176 */
056925ab 1177 call_netdevice_notifiers(NETDEV_DOWN, dev);
1da177e4
LT
1178
1179 return 0;
1180}
1181
1182
0187bdfb
BH
1183/**
1184 * dev_disable_lro - disable Large Receive Offload on a device
1185 * @dev: device
1186 *
1187 * Disable Large Receive Offload (LRO) on a net device. Must be
1188 * called under RTNL. This is needed if received packets may be
1189 * forwarded to another interface.
1190 */
1191void dev_disable_lro(struct net_device *dev)
1192{
1193 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1194 dev->ethtool_ops->set_flags) {
1195 u32 flags = dev->ethtool_ops->get_flags(dev);
1196 if (flags & ETH_FLAG_LRO) {
1197 flags &= ~ETH_FLAG_LRO;
1198 dev->ethtool_ops->set_flags(dev, flags);
1199 }
1200 }
1201 WARN_ON(dev->features & NETIF_F_LRO);
1202}
1203EXPORT_SYMBOL(dev_disable_lro);
1204
1205
881d966b
EB
1206static int dev_boot_phase = 1;
1207
1da177e4
LT
1208/*
1209 * Device change register/unregister. These are not inline or static
1210 * as we export them to the world.
1211 */
1212
1213/**
1214 * register_netdevice_notifier - register a network notifier block
1215 * @nb: notifier
1216 *
1217 * Register a notifier to be called when network device events occur.
1218 * The notifier passed is linked into the kernel structures and must
1219 * not be reused until it has been unregistered. A negative errno code
1220 * is returned on a failure.
1221 *
1222 * When registered all registration and up events are replayed
4ec93edb 1223 * to the new notifier to allow device to have a race free
1da177e4
LT
1224 * view of the network device list.
1225 */
1226
1227int register_netdevice_notifier(struct notifier_block *nb)
1228{
1229 struct net_device *dev;
fcc5a03a 1230 struct net_device *last;
881d966b 1231 struct net *net;
1da177e4
LT
1232 int err;
1233
1234 rtnl_lock();
f07d5b94 1235 err = raw_notifier_chain_register(&netdev_chain, nb);
fcc5a03a
HX
1236 if (err)
1237 goto unlock;
881d966b
EB
1238 if (dev_boot_phase)
1239 goto unlock;
1240 for_each_net(net) {
1241 for_each_netdev(net, dev) {
1242 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1243 err = notifier_to_errno(err);
1244 if (err)
1245 goto rollback;
1246
1247 if (!(dev->flags & IFF_UP))
1248 continue;
1da177e4 1249
881d966b
EB
1250 nb->notifier_call(nb, NETDEV_UP, dev);
1251 }
1da177e4 1252 }
fcc5a03a
HX
1253
1254unlock:
1da177e4
LT
1255 rtnl_unlock();
1256 return err;
fcc5a03a
HX
1257
1258rollback:
1259 last = dev;
881d966b
EB
1260 for_each_net(net) {
1261 for_each_netdev(net, dev) {
1262 if (dev == last)
1263 break;
fcc5a03a 1264
881d966b
EB
1265 if (dev->flags & IFF_UP) {
1266 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1267 nb->notifier_call(nb, NETDEV_DOWN, dev);
1268 }
1269 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
fcc5a03a 1270 }
fcc5a03a 1271 }
c67625a1
PE
1272
1273 raw_notifier_chain_unregister(&netdev_chain, nb);
fcc5a03a 1274 goto unlock;
1da177e4
LT
1275}
1276
1277/**
1278 * unregister_netdevice_notifier - unregister a network notifier block
1279 * @nb: notifier
1280 *
1281 * Unregister a notifier previously registered by
1282 * register_netdevice_notifier(). The notifier is unlinked into the
1283 * kernel structures and may then be reused. A negative errno code
1284 * is returned on a failure.
1285 */
1286
1287int unregister_netdevice_notifier(struct notifier_block *nb)
1288{
9f514950
HX
1289 int err;
1290
1291 rtnl_lock();
f07d5b94 1292 err = raw_notifier_chain_unregister(&netdev_chain, nb);
9f514950
HX
1293 rtnl_unlock();
1294 return err;
1da177e4
LT
1295}
1296
1297/**
1298 * call_netdevice_notifiers - call all network notifier blocks
1299 * @val: value passed unmodified to notifier function
c4ea43c5 1300 * @dev: net_device pointer passed unmodified to notifier function
1da177e4
LT
1301 *
1302 * Call all network notifier blocks. Parameters and return value
f07d5b94 1303 * are as for raw_notifier_call_chain().
1da177e4
LT
1304 */
1305
ad7379d4 1306int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1da177e4 1307{
ad7379d4 1308 return raw_notifier_call_chain(&netdev_chain, val, dev);
1da177e4
LT
1309}
1310
1311/* When > 0 there are consumers of rx skb time stamps */
1312static atomic_t netstamp_needed = ATOMIC_INIT(0);
1313
1314void net_enable_timestamp(void)
1315{
1316 atomic_inc(&netstamp_needed);
1317}
1318
1319void net_disable_timestamp(void)
1320{
1321 atomic_dec(&netstamp_needed);
1322}
1323
a61bbcf2 1324static inline void net_timestamp(struct sk_buff *skb)
1da177e4
LT
1325{
1326 if (atomic_read(&netstamp_needed))
a61bbcf2 1327 __net_timestamp(skb);
b7aa0bf7
ED
1328 else
1329 skb->tstamp.tv64 = 0;
1da177e4
LT
1330}
1331
1332/*
1333 * Support routine. Sends outgoing frames to any network
1334 * taps currently in use.
1335 */
1336
f6a78bfc 1337static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1da177e4
LT
1338{
1339 struct packet_type *ptype;
a61bbcf2
PM
1340
1341 net_timestamp(skb);
1da177e4
LT
1342
1343 rcu_read_lock();
1344 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1345 /* Never send packets back to the socket
1346 * they originated from - MvS (miquels@drinkel.ow.org)
1347 */
1348 if ((ptype->dev == dev || !ptype->dev) &&
1349 (ptype->af_packet_priv == NULL ||
1350 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1351 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1352 if (!skb2)
1353 break;
1354
1355 /* skb->nh should be correctly
1356 set by sender, so that the second statement is
1357 just protection against buggy protocols.
1358 */
459a98ed 1359 skb_reset_mac_header(skb2);
1da177e4 1360
d56f90a7 1361 if (skb_network_header(skb2) < skb2->data ||
27a884dc 1362 skb2->network_header > skb2->tail) {
1da177e4
LT
1363 if (net_ratelimit())
1364 printk(KERN_CRIT "protocol %04x is "
1365 "buggy, dev %s\n",
1366 skb2->protocol, dev->name);
c1d2bbe1 1367 skb_reset_network_header(skb2);
1da177e4
LT
1368 }
1369
b0e380b1 1370 skb2->transport_header = skb2->network_header;
1da177e4 1371 skb2->pkt_type = PACKET_OUTGOING;
f2ccd8fa 1372 ptype->func(skb2, skb->dev, ptype, skb->dev);
1da177e4
LT
1373 }
1374 }
1375 rcu_read_unlock();
1376}
1377
56079431 1378
def82a1d 1379static inline void __netif_reschedule(struct Qdisc *q)
56079431 1380{
def82a1d
JP
1381 struct softnet_data *sd;
1382 unsigned long flags;
56079431 1383
def82a1d
JP
1384 local_irq_save(flags);
1385 sd = &__get_cpu_var(softnet_data);
1386 q->next_sched = sd->output_queue;
1387 sd->output_queue = q;
1388 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1389 local_irq_restore(flags);
1390}
1391
1392void __netif_schedule(struct Qdisc *q)
1393{
1394 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1395 __netif_reschedule(q);
56079431
DV
1396}
1397EXPORT_SYMBOL(__netif_schedule);
1398
bea3348e 1399void dev_kfree_skb_irq(struct sk_buff *skb)
56079431 1400{
bea3348e
SH
1401 if (atomic_dec_and_test(&skb->users)) {
1402 struct softnet_data *sd;
1403 unsigned long flags;
56079431 1404
bea3348e
SH
1405 local_irq_save(flags);
1406 sd = &__get_cpu_var(softnet_data);
1407 skb->next = sd->completion_queue;
1408 sd->completion_queue = skb;
1409 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1410 local_irq_restore(flags);
1411 }
56079431 1412}
bea3348e 1413EXPORT_SYMBOL(dev_kfree_skb_irq);
56079431
DV
1414
1415void dev_kfree_skb_any(struct sk_buff *skb)
1416{
1417 if (in_irq() || irqs_disabled())
1418 dev_kfree_skb_irq(skb);
1419 else
1420 dev_kfree_skb(skb);
1421}
1422EXPORT_SYMBOL(dev_kfree_skb_any);
1423
1424
bea3348e
SH
1425/**
1426 * netif_device_detach - mark device as removed
1427 * @dev: network device
1428 *
1429 * Mark device as removed from system and therefore no longer available.
1430 */
56079431
DV
1431void netif_device_detach(struct net_device *dev)
1432{
1433 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1434 netif_running(dev)) {
1435 netif_stop_queue(dev);
1436 }
1437}
1438EXPORT_SYMBOL(netif_device_detach);
1439
bea3348e
SH
1440/**
1441 * netif_device_attach - mark device as attached
1442 * @dev: network device
1443 *
1444 * Mark device as attached from system and restart if needed.
1445 */
56079431
DV
1446void netif_device_attach(struct net_device *dev)
1447{
1448 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1449 netif_running(dev)) {
1450 netif_wake_queue(dev);
4ec93edb 1451 __netdev_watchdog_up(dev);
56079431
DV
1452 }
1453}
1454EXPORT_SYMBOL(netif_device_attach);
1455
6de329e2
BH
1456static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1457{
1458 return ((features & NETIF_F_GEN_CSUM) ||
1459 ((features & NETIF_F_IP_CSUM) &&
1460 protocol == htons(ETH_P_IP)) ||
1461 ((features & NETIF_F_IPV6_CSUM) &&
1462 protocol == htons(ETH_P_IPV6)));
1463}
1464
1465static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1466{
1467 if (can_checksum_protocol(dev->features, skb->protocol))
1468 return true;
1469
1470 if (skb->protocol == htons(ETH_P_8021Q)) {
1471 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1472 if (can_checksum_protocol(dev->features & dev->vlan_features,
1473 veh->h_vlan_encapsulated_proto))
1474 return true;
1475 }
1476
1477 return false;
1478}
56079431 1479
1da177e4
LT
1480/*
1481 * Invalidate hardware checksum when packet is to be mangled, and
1482 * complete checksum manually on outgoing path.
1483 */
84fa7933 1484int skb_checksum_help(struct sk_buff *skb)
1da177e4 1485{
d3bc23e7 1486 __wsum csum;
663ead3b 1487 int ret = 0, offset;
1da177e4 1488
84fa7933 1489 if (skb->ip_summed == CHECKSUM_COMPLETE)
a430a43d
HX
1490 goto out_set_summed;
1491
1492 if (unlikely(skb_shinfo(skb)->gso_size)) {
a430a43d
HX
1493 /* Let GSO fix up the checksum. */
1494 goto out_set_summed;
1da177e4
LT
1495 }
1496
a030847e
HX
1497 offset = skb->csum_start - skb_headroom(skb);
1498 BUG_ON(offset >= skb_headlen(skb));
1499 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1500
1501 offset += skb->csum_offset;
1502 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1503
1504 if (skb_cloned(skb) &&
1505 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1da177e4
LT
1506 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1507 if (ret)
1508 goto out;
1509 }
1510
a030847e 1511 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
a430a43d 1512out_set_summed:
1da177e4 1513 skb->ip_summed = CHECKSUM_NONE;
4ec93edb 1514out:
1da177e4
LT
1515 return ret;
1516}
1517
f6a78bfc
HX
1518/**
1519 * skb_gso_segment - Perform segmentation on skb.
1520 * @skb: buffer to segment
576a30eb 1521 * @features: features for the output path (see dev->features)
f6a78bfc
HX
1522 *
1523 * This function segments the given skb and returns a list of segments.
576a30eb
HX
1524 *
1525 * It may return NULL if the skb requires no segmentation. This is
1526 * only possible when GSO is used for verifying header integrity.
f6a78bfc 1527 */
576a30eb 1528struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
f6a78bfc
HX
1529{
1530 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1531 struct packet_type *ptype;
252e3346 1532 __be16 type = skb->protocol;
a430a43d 1533 int err;
f6a78bfc
HX
1534
1535 BUG_ON(skb_shinfo(skb)->frag_list);
f6a78bfc 1536
459a98ed 1537 skb_reset_mac_header(skb);
b0e380b1 1538 skb->mac_len = skb->network_header - skb->mac_header;
f6a78bfc
HX
1539 __skb_pull(skb, skb->mac_len);
1540
f9d106a6 1541 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
a430a43d
HX
1542 if (skb_header_cloned(skb) &&
1543 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1544 return ERR_PTR(err);
1545 }
1546
f6a78bfc 1547 rcu_read_lock();
82d8a867
PE
1548 list_for_each_entry_rcu(ptype,
1549 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
f6a78bfc 1550 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
84fa7933 1551 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
a430a43d
HX
1552 err = ptype->gso_send_check(skb);
1553 segs = ERR_PTR(err);
1554 if (err || skb_gso_ok(skb, features))
1555 break;
d56f90a7
ACM
1556 __skb_push(skb, (skb->data -
1557 skb_network_header(skb)));
a430a43d 1558 }
576a30eb 1559 segs = ptype->gso_segment(skb, features);
f6a78bfc
HX
1560 break;
1561 }
1562 }
1563 rcu_read_unlock();
1564
98e399f8 1565 __skb_push(skb, skb->data - skb_mac_header(skb));
576a30eb 1566
f6a78bfc
HX
1567 return segs;
1568}
1569
1570EXPORT_SYMBOL(skb_gso_segment);
1571
fb286bb2
HX
1572/* Take action when hardware reception checksum errors are detected. */
1573#ifdef CONFIG_BUG
1574void netdev_rx_csum_fault(struct net_device *dev)
1575{
1576 if (net_ratelimit()) {
4ec93edb 1577 printk(KERN_ERR "%s: hw csum failure.\n",
246a4212 1578 dev ? dev->name : "<unknown>");
fb286bb2
HX
1579 dump_stack();
1580 }
1581}
1582EXPORT_SYMBOL(netdev_rx_csum_fault);
1583#endif
1584
1da177e4
LT
1585/* Actually, we should eliminate this check as soon as we know, that:
1586 * 1. IOMMU is present and allows to map all the memory.
1587 * 2. No high memory really exists on this machine.
1588 */
1589
1590static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1591{
3d3a8533 1592#ifdef CONFIG_HIGHMEM
1da177e4
LT
1593 int i;
1594
1595 if (dev->features & NETIF_F_HIGHDMA)
1596 return 0;
1597
1598 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1599 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1600 return 1;
1601
3d3a8533 1602#endif
1da177e4
LT
1603 return 0;
1604}
1da177e4 1605
f6a78bfc
HX
1606struct dev_gso_cb {
1607 void (*destructor)(struct sk_buff *skb);
1608};
1609
1610#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1611
1612static void dev_gso_skb_destructor(struct sk_buff *skb)
1613{
1614 struct dev_gso_cb *cb;
1615
1616 do {
1617 struct sk_buff *nskb = skb->next;
1618
1619 skb->next = nskb->next;
1620 nskb->next = NULL;
1621 kfree_skb(nskb);
1622 } while (skb->next);
1623
1624 cb = DEV_GSO_CB(skb);
1625 if (cb->destructor)
1626 cb->destructor(skb);
1627}
1628
1629/**
1630 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1631 * @skb: buffer to segment
1632 *
1633 * This function segments the given skb and stores the list of segments
1634 * in skb->next.
1635 */
1636static int dev_gso_segment(struct sk_buff *skb)
1637{
1638 struct net_device *dev = skb->dev;
1639 struct sk_buff *segs;
576a30eb
HX
1640 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1641 NETIF_F_SG : 0);
1642
1643 segs = skb_gso_segment(skb, features);
1644
1645 /* Verifying header integrity only. */
1646 if (!segs)
1647 return 0;
f6a78bfc 1648
801678c5 1649 if (IS_ERR(segs))
f6a78bfc
HX
1650 return PTR_ERR(segs);
1651
1652 skb->next = segs;
1653 DEV_GSO_CB(skb)->destructor = skb->destructor;
1654 skb->destructor = dev_gso_skb_destructor;
1655
1656 return 0;
1657}
1658
fd2ea0a7
DM
1659int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1660 struct netdev_queue *txq)
f6a78bfc
HX
1661{
1662 if (likely(!skb->next)) {
9be9a6b9 1663 if (!list_empty(&ptype_all))
f6a78bfc
HX
1664 dev_queue_xmit_nit(skb, dev);
1665
576a30eb
HX
1666 if (netif_needs_gso(dev, skb)) {
1667 if (unlikely(dev_gso_segment(skb)))
1668 goto out_kfree_skb;
1669 if (skb->next)
1670 goto gso;
1671 }
f6a78bfc 1672
576a30eb 1673 return dev->hard_start_xmit(skb, dev);
f6a78bfc
HX
1674 }
1675
576a30eb 1676gso:
f6a78bfc
HX
1677 do {
1678 struct sk_buff *nskb = skb->next;
1679 int rc;
1680
1681 skb->next = nskb->next;
1682 nskb->next = NULL;
1683 rc = dev->hard_start_xmit(nskb, dev);
1684 if (unlikely(rc)) {
f54d9e8d 1685 nskb->next = skb->next;
f6a78bfc
HX
1686 skb->next = nskb;
1687 return rc;
1688 }
fd2ea0a7 1689 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
f54d9e8d 1690 return NETDEV_TX_BUSY;
f6a78bfc 1691 } while (skb->next);
4ec93edb 1692
f6a78bfc
HX
1693 skb->destructor = DEV_GSO_CB(skb)->destructor;
1694
1695out_kfree_skb:
1696 kfree_skb(skb);
1697 return 0;
1698}
1699
b6b2fed1
DM
1700static u32 simple_tx_hashrnd;
1701static int simple_tx_hashrnd_initialized = 0;
1702
8f0f2223
DM
1703static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb)
1704{
b6b2fed1
DM
1705 u32 addr1, addr2, ports;
1706 u32 hash, ihl;
ad55dcaf 1707 u8 ip_proto = 0;
b6b2fed1
DM
1708
1709 if (unlikely(!simple_tx_hashrnd_initialized)) {
1710 get_random_bytes(&simple_tx_hashrnd, 4);
1711 simple_tx_hashrnd_initialized = 1;
1712 }
8f0f2223
DM
1713
1714 switch (skb->protocol) {
60678040 1715 case htons(ETH_P_IP):
ad55dcaf
AD
1716 if (!(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)))
1717 ip_proto = ip_hdr(skb)->protocol;
b6b2fed1
DM
1718 addr1 = ip_hdr(skb)->saddr;
1719 addr2 = ip_hdr(skb)->daddr;
8f0f2223 1720 ihl = ip_hdr(skb)->ihl;
8f0f2223 1721 break;
60678040 1722 case htons(ETH_P_IPV6):
8f0f2223 1723 ip_proto = ipv6_hdr(skb)->nexthdr;
b6b2fed1
DM
1724 addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3];
1725 addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3];
8f0f2223 1726 ihl = (40 >> 2);
8f0f2223
DM
1727 break;
1728 default:
1729 return 0;
1730 }
1731
8f0f2223
DM
1732
1733 switch (ip_proto) {
1734 case IPPROTO_TCP:
1735 case IPPROTO_UDP:
1736 case IPPROTO_DCCP:
1737 case IPPROTO_ESP:
1738 case IPPROTO_AH:
1739 case IPPROTO_SCTP:
1740 case IPPROTO_UDPLITE:
b6b2fed1 1741 ports = *((u32 *) (skb_network_header(skb) + (ihl * 4)));
8f0f2223
DM
1742 break;
1743
1744 default:
b6b2fed1 1745 ports = 0;
8f0f2223
DM
1746 break;
1747 }
1748
b6b2fed1
DM
1749 hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd);
1750
1751 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
8f0f2223
DM
1752}
1753
e8a0464c
DM
1754static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1755 struct sk_buff *skb)
1756{
fd2ea0a7
DM
1757 u16 queue_index = 0;
1758
eae792b7
DM
1759 if (dev->select_queue)
1760 queue_index = dev->select_queue(dev, skb);
8f0f2223
DM
1761 else if (dev->real_num_tx_queues > 1)
1762 queue_index = simple_tx_hash(dev, skb);
eae792b7 1763
fd2ea0a7
DM
1764 skb_set_queue_mapping(skb, queue_index);
1765 return netdev_get_tx_queue(dev, queue_index);
e8a0464c
DM
1766}
1767
d29f749e
DJ
1768/**
1769 * dev_queue_xmit - transmit a buffer
1770 * @skb: buffer to transmit
1771 *
1772 * Queue a buffer for transmission to a network device. The caller must
1773 * have set the device and priority and built the buffer before calling
1774 * this function. The function can be called from an interrupt.
1775 *
1776 * A negative errno code is returned on a failure. A success does not
1777 * guarantee the frame will be transmitted as it may be dropped due
1778 * to congestion or traffic shaping.
1779 *
1780 * -----------------------------------------------------------------------------------
1781 * I notice this method can also return errors from the queue disciplines,
1782 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1783 * be positive.
1784 *
1785 * Regardless of the return value, the skb is consumed, so it is currently
1786 * difficult to retry a send to this method. (You can bump the ref count
1787 * before sending to hold a reference for retry if you are careful.)
1788 *
1789 * When calling this method, interrupts MUST be enabled. This is because
1790 * the BH enable code must have IRQs enabled so that it will not deadlock.
1791 * --BLG
1792 */
1da177e4
LT
1793int dev_queue_xmit(struct sk_buff *skb)
1794{
1795 struct net_device *dev = skb->dev;
dc2b4847 1796 struct netdev_queue *txq;
1da177e4
LT
1797 struct Qdisc *q;
1798 int rc = -ENOMEM;
1799
f6a78bfc
HX
1800 /* GSO will handle the following emulations directly. */
1801 if (netif_needs_gso(dev, skb))
1802 goto gso;
1803
1da177e4
LT
1804 if (skb_shinfo(skb)->frag_list &&
1805 !(dev->features & NETIF_F_FRAGLIST) &&
364c6bad 1806 __skb_linearize(skb))
1da177e4
LT
1807 goto out_kfree_skb;
1808
1809 /* Fragmented skb is linearized if device does not support SG,
1810 * or if at least one of fragments is in highmem and device
1811 * does not support DMA from it.
1812 */
1813 if (skb_shinfo(skb)->nr_frags &&
1814 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
364c6bad 1815 __skb_linearize(skb))
1da177e4
LT
1816 goto out_kfree_skb;
1817
1818 /* If packet is not checksummed and device does not support
1819 * checksumming for this protocol, complete checksumming here.
1820 */
663ead3b
HX
1821 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1822 skb_set_transport_header(skb, skb->csum_start -
1823 skb_headroom(skb));
6de329e2
BH
1824 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1825 goto out_kfree_skb;
663ead3b 1826 }
1da177e4 1827
f6a78bfc 1828gso:
4ec93edb
YH
1829 /* Disable soft irqs for various locks below. Also
1830 * stops preemption for RCU.
1da177e4 1831 */
4ec93edb 1832 rcu_read_lock_bh();
1da177e4 1833
eae792b7 1834 txq = dev_pick_tx(dev, skb);
b0e1e646 1835 q = rcu_dereference(txq->qdisc);
37437bb2 1836
1da177e4
LT
1837#ifdef CONFIG_NET_CLS_ACT
1838 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1839#endif
1840 if (q->enqueue) {
5fb66229 1841 spinlock_t *root_lock = qdisc_lock(q);
37437bb2
DM
1842
1843 spin_lock(root_lock);
1844
a9312ae8 1845 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
96d20316 1846 kfree_skb(skb);
a9312ae8 1847 rc = NET_XMIT_DROP;
96d20316
DM
1848 } else {
1849 rc = qdisc_enqueue_root(skb, q);
1850 qdisc_run(q);
a9312ae8 1851 }
37437bb2
DM
1852 spin_unlock(root_lock);
1853
37437bb2 1854 goto out;
1da177e4
LT
1855 }
1856
1857 /* The device has no queue. Common case for software devices:
1858 loopback, all the sorts of tunnels...
1859
932ff279
HX
1860 Really, it is unlikely that netif_tx_lock protection is necessary
1861 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1da177e4
LT
1862 counters.)
1863 However, it is possible, that they rely on protection
1864 made by us here.
1865
1866 Check this and shot the lock. It is not prone from deadlocks.
1867 Either shot noqueue qdisc, it is even simpler 8)
1868 */
1869 if (dev->flags & IFF_UP) {
1870 int cpu = smp_processor_id(); /* ok because BHs are off */
1871
c773e847 1872 if (txq->xmit_lock_owner != cpu) {
1da177e4 1873
c773e847 1874 HARD_TX_LOCK(dev, txq, cpu);
1da177e4 1875
fd2ea0a7 1876 if (!netif_tx_queue_stopped(txq)) {
1da177e4 1877 rc = 0;
fd2ea0a7 1878 if (!dev_hard_start_xmit(skb, dev, txq)) {
c773e847 1879 HARD_TX_UNLOCK(dev, txq);
1da177e4
LT
1880 goto out;
1881 }
1882 }
c773e847 1883 HARD_TX_UNLOCK(dev, txq);
1da177e4
LT
1884 if (net_ratelimit())
1885 printk(KERN_CRIT "Virtual device %s asks to "
1886 "queue packet!\n", dev->name);
1887 } else {
1888 /* Recursion is detected! It is possible,
1889 * unfortunately */
1890 if (net_ratelimit())
1891 printk(KERN_CRIT "Dead loop on virtual device "
1892 "%s, fix it urgently!\n", dev->name);
1893 }
1894 }
1895
1896 rc = -ENETDOWN;
d4828d85 1897 rcu_read_unlock_bh();
1da177e4
LT
1898
1899out_kfree_skb:
1900 kfree_skb(skb);
1901 return rc;
1902out:
d4828d85 1903 rcu_read_unlock_bh();
1da177e4
LT
1904 return rc;
1905}
1906
1907
1908/*=======================================================================
1909 Receiver routines
1910 =======================================================================*/
1911
6b2bedc3
SH
1912int netdev_max_backlog __read_mostly = 1000;
1913int netdev_budget __read_mostly = 300;
1914int weight_p __read_mostly = 64; /* old backlog weight */
1da177e4
LT
1915
1916DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1917
1918
1da177e4
LT
1919/**
1920 * netif_rx - post buffer to the network code
1921 * @skb: buffer to post
1922 *
1923 * This function receives a packet from a device driver and queues it for
1924 * the upper (protocol) levels to process. It always succeeds. The buffer
1925 * may be dropped during processing for congestion control or by the
1926 * protocol layers.
1927 *
1928 * return values:
1929 * NET_RX_SUCCESS (no congestion)
1da177e4
LT
1930 * NET_RX_DROP (packet was dropped)
1931 *
1932 */
1933
1934int netif_rx(struct sk_buff *skb)
1935{
1da177e4
LT
1936 struct softnet_data *queue;
1937 unsigned long flags;
1938
1939 /* if netpoll wants it, pretend we never saw it */
1940 if (netpoll_rx(skb))
1941 return NET_RX_DROP;
1942
b7aa0bf7 1943 if (!skb->tstamp.tv64)
a61bbcf2 1944 net_timestamp(skb);
1da177e4
LT
1945
1946 /*
1947 * The code is rearranged so that the path is the most
1948 * short when CPU is congested, but is still operating.
1949 */
1950 local_irq_save(flags);
1da177e4
LT
1951 queue = &__get_cpu_var(softnet_data);
1952
1953 __get_cpu_var(netdev_rx_stat).total++;
1954 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1955 if (queue->input_pkt_queue.qlen) {
1da177e4 1956enqueue:
1da177e4 1957 __skb_queue_tail(&queue->input_pkt_queue, skb);
1da177e4 1958 local_irq_restore(flags);
34008d8c 1959 return NET_RX_SUCCESS;
1da177e4
LT
1960 }
1961
bea3348e 1962 napi_schedule(&queue->backlog);
1da177e4
LT
1963 goto enqueue;
1964 }
1965
1da177e4
LT
1966 __get_cpu_var(netdev_rx_stat).dropped++;
1967 local_irq_restore(flags);
1968
1969 kfree_skb(skb);
1970 return NET_RX_DROP;
1971}
1972
1973int netif_rx_ni(struct sk_buff *skb)
1974{
1975 int err;
1976
1977 preempt_disable();
1978 err = netif_rx(skb);
1979 if (local_softirq_pending())
1980 do_softirq();
1981 preempt_enable();
1982
1983 return err;
1984}
1985
1986EXPORT_SYMBOL(netif_rx_ni);
1987
1da177e4
LT
1988static void net_tx_action(struct softirq_action *h)
1989{
1990 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1991
1992 if (sd->completion_queue) {
1993 struct sk_buff *clist;
1994
1995 local_irq_disable();
1996 clist = sd->completion_queue;
1997 sd->completion_queue = NULL;
1998 local_irq_enable();
1999
2000 while (clist) {
2001 struct sk_buff *skb = clist;
2002 clist = clist->next;
2003
547b792c 2004 WARN_ON(atomic_read(&skb->users));
1da177e4
LT
2005 __kfree_skb(skb);
2006 }
2007 }
2008
2009 if (sd->output_queue) {
37437bb2 2010 struct Qdisc *head;
1da177e4
LT
2011
2012 local_irq_disable();
2013 head = sd->output_queue;
2014 sd->output_queue = NULL;
2015 local_irq_enable();
2016
2017 while (head) {
37437bb2
DM
2018 struct Qdisc *q = head;
2019 spinlock_t *root_lock;
2020
1da177e4
LT
2021 head = head->next_sched;
2022
5fb66229 2023 root_lock = qdisc_lock(q);
37437bb2 2024 if (spin_trylock(root_lock)) {
def82a1d
JP
2025 smp_mb__before_clear_bit();
2026 clear_bit(__QDISC_STATE_SCHED,
2027 &q->state);
37437bb2
DM
2028 qdisc_run(q);
2029 spin_unlock(root_lock);
1da177e4 2030 } else {
195648bb 2031 if (!test_bit(__QDISC_STATE_DEACTIVATED,
e8a83e10 2032 &q->state)) {
195648bb 2033 __netif_reschedule(q);
e8a83e10
JP
2034 } else {
2035 smp_mb__before_clear_bit();
2036 clear_bit(__QDISC_STATE_SCHED,
2037 &q->state);
2038 }
1da177e4
LT
2039 }
2040 }
2041 }
2042}
2043
6f05f629
SH
2044static inline int deliver_skb(struct sk_buff *skb,
2045 struct packet_type *pt_prev,
2046 struct net_device *orig_dev)
1da177e4
LT
2047{
2048 atomic_inc(&skb->users);
f2ccd8fa 2049 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
2050}
2051
2052#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
6229e362 2053/* These hooks defined here for ATM */
1da177e4
LT
2054struct net_bridge;
2055struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2056 unsigned char *addr);
6229e362 2057void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1da177e4 2058
6229e362
SH
2059/*
2060 * If bridge module is loaded call bridging hook.
2061 * returns NULL if packet was consumed.
2062 */
2063struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2064 struct sk_buff *skb) __read_mostly;
2065static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2066 struct packet_type **pt_prev, int *ret,
2067 struct net_device *orig_dev)
1da177e4
LT
2068{
2069 struct net_bridge_port *port;
2070
6229e362
SH
2071 if (skb->pkt_type == PACKET_LOOPBACK ||
2072 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2073 return skb;
1da177e4
LT
2074
2075 if (*pt_prev) {
6229e362 2076 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1da177e4 2077 *pt_prev = NULL;
4ec93edb
YH
2078 }
2079
6229e362 2080 return br_handle_frame_hook(port, skb);
1da177e4
LT
2081}
2082#else
6229e362 2083#define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1da177e4
LT
2084#endif
2085
b863ceb7
PM
2086#if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2087struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2088EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2089
2090static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2091 struct packet_type **pt_prev,
2092 int *ret,
2093 struct net_device *orig_dev)
2094{
2095 if (skb->dev->macvlan_port == NULL)
2096 return skb;
2097
2098 if (*pt_prev) {
2099 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2100 *pt_prev = NULL;
2101 }
2102 return macvlan_handle_frame_hook(skb);
2103}
2104#else
2105#define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2106#endif
2107
1da177e4
LT
2108#ifdef CONFIG_NET_CLS_ACT
2109/* TODO: Maybe we should just force sch_ingress to be compiled in
2110 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2111 * a compare and 2 stores extra right now if we dont have it on
2112 * but have CONFIG_NET_CLS_ACT
4ec93edb 2113 * NOTE: This doesnt stop any functionality; if you dont have
1da177e4
LT
2114 * the ingress scheduler, you just cant add policies on ingress.
2115 *
2116 */
4ec93edb 2117static int ing_filter(struct sk_buff *skb)
1da177e4 2118{
1da177e4 2119 struct net_device *dev = skb->dev;
f697c3e8 2120 u32 ttl = G_TC_RTTL(skb->tc_verd);
555353cf
DM
2121 struct netdev_queue *rxq;
2122 int result = TC_ACT_OK;
2123 struct Qdisc *q;
4ec93edb 2124
f697c3e8
HX
2125 if (MAX_RED_LOOP < ttl++) {
2126 printk(KERN_WARNING
2127 "Redir loop detected Dropping packet (%d->%d)\n",
2128 skb->iif, dev->ifindex);
2129 return TC_ACT_SHOT;
2130 }
1da177e4 2131
f697c3e8
HX
2132 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2133 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
1da177e4 2134
555353cf
DM
2135 rxq = &dev->rx_queue;
2136
83874000 2137 q = rxq->qdisc;
8d50b53d 2138 if (q != &noop_qdisc) {
83874000 2139 spin_lock(qdisc_lock(q));
a9312ae8
DM
2140 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2141 result = qdisc_enqueue_root(skb, q);
83874000
DM
2142 spin_unlock(qdisc_lock(q));
2143 }
f697c3e8
HX
2144
2145 return result;
2146}
86e65da9 2147
f697c3e8
HX
2148static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2149 struct packet_type **pt_prev,
2150 int *ret, struct net_device *orig_dev)
2151{
8d50b53d 2152 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
f697c3e8 2153 goto out;
1da177e4 2154
f697c3e8
HX
2155 if (*pt_prev) {
2156 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2157 *pt_prev = NULL;
2158 } else {
2159 /* Huh? Why does turning on AF_PACKET affect this? */
2160 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1da177e4
LT
2161 }
2162
f697c3e8
HX
2163 switch (ing_filter(skb)) {
2164 case TC_ACT_SHOT:
2165 case TC_ACT_STOLEN:
2166 kfree_skb(skb);
2167 return NULL;
2168 }
2169
2170out:
2171 skb->tc_verd = 0;
2172 return skb;
1da177e4
LT
2173}
2174#endif
2175
bc1d0411
PM
2176/*
2177 * netif_nit_deliver - deliver received packets to network taps
2178 * @skb: buffer
2179 *
2180 * This function is used to deliver incoming packets to network
2181 * taps. It should be used when the normal netif_receive_skb path
2182 * is bypassed, for example because of VLAN acceleration.
2183 */
2184void netif_nit_deliver(struct sk_buff *skb)
2185{
2186 struct packet_type *ptype;
2187
2188 if (list_empty(&ptype_all))
2189 return;
2190
2191 skb_reset_network_header(skb);
2192 skb_reset_transport_header(skb);
2193 skb->mac_len = skb->network_header - skb->mac_header;
2194
2195 rcu_read_lock();
2196 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2197 if (!ptype->dev || ptype->dev == skb->dev)
2198 deliver_skb(skb, ptype, skb->dev);
2199 }
2200 rcu_read_unlock();
2201}
2202
3b582cc1
SH
2203/**
2204 * netif_receive_skb - process receive buffer from network
2205 * @skb: buffer to process
2206 *
2207 * netif_receive_skb() is the main receive data processing function.
2208 * It always succeeds. The buffer may be dropped during processing
2209 * for congestion control or by the protocol layers.
2210 *
2211 * This function may only be called from softirq context and interrupts
2212 * should be enabled.
2213 *
2214 * Return values (usually ignored):
2215 * NET_RX_SUCCESS: no congestion
2216 * NET_RX_DROP: packet was dropped
2217 */
1da177e4
LT
2218int netif_receive_skb(struct sk_buff *skb)
2219{
2220 struct packet_type *ptype, *pt_prev;
f2ccd8fa 2221 struct net_device *orig_dev;
0d7a3681 2222 struct net_device *null_or_orig;
1da177e4 2223 int ret = NET_RX_DROP;
252e3346 2224 __be16 type;
1da177e4
LT
2225
2226 /* if we've gotten here through NAPI, check netpoll */
bea3348e 2227 if (netpoll_receive_skb(skb))
1da177e4
LT
2228 return NET_RX_DROP;
2229
b7aa0bf7 2230 if (!skb->tstamp.tv64)
a61bbcf2 2231 net_timestamp(skb);
1da177e4 2232
c01003c2
PM
2233 if (!skb->iif)
2234 skb->iif = skb->dev->ifindex;
86e65da9 2235
0d7a3681 2236 null_or_orig = NULL;
cc9bd5ce
JE
2237 orig_dev = skb->dev;
2238 if (orig_dev->master) {
0d7a3681
JE
2239 if (skb_bond_should_drop(skb))
2240 null_or_orig = orig_dev; /* deliver only exact match */
2241 else
2242 skb->dev = orig_dev->master;
cc9bd5ce 2243 }
8f903c70 2244
1da177e4
LT
2245 __get_cpu_var(netdev_rx_stat).total++;
2246
c1d2bbe1 2247 skb_reset_network_header(skb);
badff6d0 2248 skb_reset_transport_header(skb);
b0e380b1 2249 skb->mac_len = skb->network_header - skb->mac_header;
1da177e4
LT
2250
2251 pt_prev = NULL;
2252
2253 rcu_read_lock();
2254
b9f75f45
EB
2255 /* Don't receive packets in an exiting network namespace */
2256 if (!net_alive(dev_net(skb->dev)))
2257 goto out;
2258
1da177e4
LT
2259#ifdef CONFIG_NET_CLS_ACT
2260 if (skb->tc_verd & TC_NCLS) {
2261 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2262 goto ncls;
2263 }
2264#endif
2265
2266 list_for_each_entry_rcu(ptype, &ptype_all, list) {
f982307f
JE
2267 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2268 ptype->dev == orig_dev) {
4ec93edb 2269 if (pt_prev)
f2ccd8fa 2270 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
2271 pt_prev = ptype;
2272 }
2273 }
2274
2275#ifdef CONFIG_NET_CLS_ACT
f697c3e8
HX
2276 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2277 if (!skb)
1da177e4 2278 goto out;
1da177e4
LT
2279ncls:
2280#endif
2281
6229e362 2282 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
b863ceb7
PM
2283 if (!skb)
2284 goto out;
2285 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
6229e362 2286 if (!skb)
1da177e4
LT
2287 goto out;
2288
2289 type = skb->protocol;
82d8a867
PE
2290 list_for_each_entry_rcu(ptype,
2291 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1da177e4 2292 if (ptype->type == type &&
f982307f
JE
2293 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2294 ptype->dev == orig_dev)) {
4ec93edb 2295 if (pt_prev)
f2ccd8fa 2296 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
2297 pt_prev = ptype;
2298 }
2299 }
2300
2301 if (pt_prev) {
f2ccd8fa 2302 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
2303 } else {
2304 kfree_skb(skb);
2305 /* Jamal, now you will not able to escape explaining
2306 * me how you were going to use this. :-)
2307 */
2308 ret = NET_RX_DROP;
2309 }
2310
2311out:
2312 rcu_read_unlock();
2313 return ret;
2314}
2315
6e583ce5
SH
2316/* Network device is going away, flush any packets still pending */
2317static void flush_backlog(void *arg)
2318{
2319 struct net_device *dev = arg;
2320 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2321 struct sk_buff *skb, *tmp;
2322
2323 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2324 if (skb->dev == dev) {
2325 __skb_unlink(skb, &queue->input_pkt_queue);
2326 kfree_skb(skb);
2327 }
2328}
2329
bea3348e 2330static int process_backlog(struct napi_struct *napi, int quota)
1da177e4
LT
2331{
2332 int work = 0;
1da177e4
LT
2333 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2334 unsigned long start_time = jiffies;
2335
bea3348e
SH
2336 napi->weight = weight_p;
2337 do {
1da177e4 2338 struct sk_buff *skb;
1da177e4
LT
2339
2340 local_irq_disable();
2341 skb = __skb_dequeue(&queue->input_pkt_queue);
bea3348e
SH
2342 if (!skb) {
2343 __napi_complete(napi);
2344 local_irq_enable();
2345 break;
2346 }
1da177e4
LT
2347 local_irq_enable();
2348
1da177e4 2349 netif_receive_skb(skb);
bea3348e 2350 } while (++work < quota && jiffies == start_time);
1da177e4 2351
bea3348e
SH
2352 return work;
2353}
1da177e4 2354
bea3348e
SH
2355/**
2356 * __napi_schedule - schedule for receive
c4ea43c5 2357 * @n: entry to schedule
bea3348e
SH
2358 *
2359 * The entry's receive function will be scheduled to run
2360 */
b5606c2d 2361void __napi_schedule(struct napi_struct *n)
bea3348e
SH
2362{
2363 unsigned long flags;
1da177e4 2364
bea3348e
SH
2365 local_irq_save(flags);
2366 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2367 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2368 local_irq_restore(flags);
1da177e4 2369}
bea3348e
SH
2370EXPORT_SYMBOL(__napi_schedule);
2371
1da177e4
LT
2372
2373static void net_rx_action(struct softirq_action *h)
2374{
bea3348e 2375 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
1da177e4 2376 unsigned long start_time = jiffies;
51b0bded 2377 int budget = netdev_budget;
53fb95d3
MM
2378 void *have;
2379
1da177e4
LT
2380 local_irq_disable();
2381
bea3348e
SH
2382 while (!list_empty(list)) {
2383 struct napi_struct *n;
2384 int work, weight;
1da177e4 2385
bea3348e
SH
2386 /* If softirq window is exhuasted then punt.
2387 *
2388 * Note that this is a slight policy change from the
2389 * previous NAPI code, which would allow up to 2
2390 * jiffies to pass before breaking out. The test
2391 * used to be "jiffies - start_time > 1".
2392 */
2393 if (unlikely(budget <= 0 || jiffies != start_time))
1da177e4
LT
2394 goto softnet_break;
2395
2396 local_irq_enable();
2397
bea3348e
SH
2398 /* Even though interrupts have been re-enabled, this
2399 * access is safe because interrupts can only add new
2400 * entries to the tail of this list, and only ->poll()
2401 * calls can remove this head entry from the list.
2402 */
2403 n = list_entry(list->next, struct napi_struct, poll_list);
1da177e4 2404
bea3348e
SH
2405 have = netpoll_poll_lock(n);
2406
2407 weight = n->weight;
2408
0a7606c1
DM
2409 /* This NAPI_STATE_SCHED test is for avoiding a race
2410 * with netpoll's poll_napi(). Only the entity which
2411 * obtains the lock and sees NAPI_STATE_SCHED set will
2412 * actually make the ->poll() call. Therefore we avoid
2413 * accidently calling ->poll() when NAPI is not scheduled.
2414 */
2415 work = 0;
2416 if (test_bit(NAPI_STATE_SCHED, &n->state))
2417 work = n->poll(n, weight);
bea3348e
SH
2418
2419 WARN_ON_ONCE(work > weight);
2420
2421 budget -= work;
2422
2423 local_irq_disable();
2424
2425 /* Drivers must not modify the NAPI state if they
2426 * consume the entire weight. In such cases this code
2427 * still "owns" the NAPI instance and therefore can
2428 * move the instance around on the list at-will.
2429 */
fed17f30
DM
2430 if (unlikely(work == weight)) {
2431 if (unlikely(napi_disable_pending(n)))
2432 __napi_complete(n);
2433 else
2434 list_move_tail(&n->poll_list, list);
2435 }
bea3348e
SH
2436
2437 netpoll_poll_unlock(have);
1da177e4
LT
2438 }
2439out:
515e06c4 2440 local_irq_enable();
bea3348e 2441
db217334
CL
2442#ifdef CONFIG_NET_DMA
2443 /*
2444 * There may not be any more sk_buffs coming right now, so push
2445 * any pending DMA copies to hardware
2446 */
d379b01e
DW
2447 if (!cpus_empty(net_dma.channel_mask)) {
2448 int chan_idx;
0e12f848 2449 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
d379b01e
DW
2450 struct dma_chan *chan = net_dma.channels[chan_idx];
2451 if (chan)
2452 dma_async_memcpy_issue_pending(chan);
2453 }
db217334
CL
2454 }
2455#endif
bea3348e 2456
1da177e4
LT
2457 return;
2458
2459softnet_break:
2460 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2461 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2462 goto out;
2463}
2464
2465static gifconf_func_t * gifconf_list [NPROTO];
2466
2467/**
2468 * register_gifconf - register a SIOCGIF handler
2469 * @family: Address family
2470 * @gifconf: Function handler
2471 *
2472 * Register protocol dependent address dumping routines. The handler
2473 * that is passed must not be freed or reused until it has been replaced
2474 * by another handler.
2475 */
2476int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2477{
2478 if (family >= NPROTO)
2479 return -EINVAL;
2480 gifconf_list[family] = gifconf;
2481 return 0;
2482}
2483
2484
2485/*
2486 * Map an interface index to its name (SIOCGIFNAME)
2487 */
2488
2489/*
2490 * We need this ioctl for efficient implementation of the
2491 * if_indextoname() function required by the IPv6 API. Without
2492 * it, we would have to search all the interfaces to find a
2493 * match. --pb
2494 */
2495
881d966b 2496static int dev_ifname(struct net *net, struct ifreq __user *arg)
1da177e4
LT
2497{
2498 struct net_device *dev;
2499 struct ifreq ifr;
2500
2501 /*
2502 * Fetch the caller's info block.
2503 */
2504
2505 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2506 return -EFAULT;
2507
2508 read_lock(&dev_base_lock);
881d966b 2509 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
1da177e4
LT
2510 if (!dev) {
2511 read_unlock(&dev_base_lock);
2512 return -ENODEV;
2513 }
2514
2515 strcpy(ifr.ifr_name, dev->name);
2516 read_unlock(&dev_base_lock);
2517
2518 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2519 return -EFAULT;
2520 return 0;
2521}
2522
2523/*
2524 * Perform a SIOCGIFCONF call. This structure will change
2525 * size eventually, and there is nothing I can do about it.
2526 * Thus we will need a 'compatibility mode'.
2527 */
2528
881d966b 2529static int dev_ifconf(struct net *net, char __user *arg)
1da177e4
LT
2530{
2531 struct ifconf ifc;
2532 struct net_device *dev;
2533 char __user *pos;
2534 int len;
2535 int total;
2536 int i;
2537
2538 /*
2539 * Fetch the caller's info block.
2540 */
2541
2542 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2543 return -EFAULT;
2544
2545 pos = ifc.ifc_buf;
2546 len = ifc.ifc_len;
2547
2548 /*
2549 * Loop over the interfaces, and write an info block for each.
2550 */
2551
2552 total = 0;
881d966b 2553 for_each_netdev(net, dev) {
1da177e4
LT
2554 for (i = 0; i < NPROTO; i++) {
2555 if (gifconf_list[i]) {
2556 int done;
2557 if (!pos)
2558 done = gifconf_list[i](dev, NULL, 0);
2559 else
2560 done = gifconf_list[i](dev, pos + total,
2561 len - total);
2562 if (done < 0)
2563 return -EFAULT;
2564 total += done;
2565 }
2566 }
4ec93edb 2567 }
1da177e4
LT
2568
2569 /*
2570 * All done. Write the updated control block back to the caller.
2571 */
2572 ifc.ifc_len = total;
2573
2574 /*
2575 * Both BSD and Solaris return 0 here, so we do too.
2576 */
2577 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2578}
2579
2580#ifdef CONFIG_PROC_FS
2581/*
2582 * This is invoked by the /proc filesystem handler to display a device
2583 * in detail.
2584 */
7562f876 2585void *dev_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2586 __acquires(dev_base_lock)
1da177e4 2587{
e372c414 2588 struct net *net = seq_file_net(seq);
7562f876 2589 loff_t off;
1da177e4 2590 struct net_device *dev;
1da177e4 2591
7562f876
PE
2592 read_lock(&dev_base_lock);
2593 if (!*pos)
2594 return SEQ_START_TOKEN;
1da177e4 2595
7562f876 2596 off = 1;
881d966b 2597 for_each_netdev(net, dev)
7562f876
PE
2598 if (off++ == *pos)
2599 return dev;
1da177e4 2600
7562f876 2601 return NULL;
1da177e4
LT
2602}
2603
2604void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2605{
e372c414 2606 struct net *net = seq_file_net(seq);
1da177e4 2607 ++*pos;
7562f876 2608 return v == SEQ_START_TOKEN ?
881d966b 2609 first_net_device(net) : next_net_device((struct net_device *)v);
1da177e4
LT
2610}
2611
2612void dev_seq_stop(struct seq_file *seq, void *v)
9a429c49 2613 __releases(dev_base_lock)
1da177e4
LT
2614{
2615 read_unlock(&dev_base_lock);
2616}
2617
2618static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2619{
c45d286e 2620 struct net_device_stats *stats = dev->get_stats(dev);
1da177e4 2621
5a1b5898
RR
2622 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2623 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2624 dev->name, stats->rx_bytes, stats->rx_packets,
2625 stats->rx_errors,
2626 stats->rx_dropped + stats->rx_missed_errors,
2627 stats->rx_fifo_errors,
2628 stats->rx_length_errors + stats->rx_over_errors +
2629 stats->rx_crc_errors + stats->rx_frame_errors,
2630 stats->rx_compressed, stats->multicast,
2631 stats->tx_bytes, stats->tx_packets,
2632 stats->tx_errors, stats->tx_dropped,
2633 stats->tx_fifo_errors, stats->collisions,
2634 stats->tx_carrier_errors +
2635 stats->tx_aborted_errors +
2636 stats->tx_window_errors +
2637 stats->tx_heartbeat_errors,
2638 stats->tx_compressed);
1da177e4
LT
2639}
2640
2641/*
2642 * Called from the PROCfs module. This now uses the new arbitrary sized
2643 * /proc/net interface to create /proc/net/dev
2644 */
2645static int dev_seq_show(struct seq_file *seq, void *v)
2646{
2647 if (v == SEQ_START_TOKEN)
2648 seq_puts(seq, "Inter-| Receive "
2649 " | Transmit\n"
2650 " face |bytes packets errs drop fifo frame "
2651 "compressed multicast|bytes packets errs "
2652 "drop fifo colls carrier compressed\n");
2653 else
2654 dev_seq_printf_stats(seq, v);
2655 return 0;
2656}
2657
2658static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2659{
2660 struct netif_rx_stats *rc = NULL;
2661
0c0b0aca 2662 while (*pos < nr_cpu_ids)
4ec93edb 2663 if (cpu_online(*pos)) {
1da177e4
LT
2664 rc = &per_cpu(netdev_rx_stat, *pos);
2665 break;
2666 } else
2667 ++*pos;
2668 return rc;
2669}
2670
2671static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2672{
2673 return softnet_get_online(pos);
2674}
2675
2676static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2677{
2678 ++*pos;
2679 return softnet_get_online(pos);
2680}
2681
2682static void softnet_seq_stop(struct seq_file *seq, void *v)
2683{
2684}
2685
2686static int softnet_seq_show(struct seq_file *seq, void *v)
2687{
2688 struct netif_rx_stats *s = v;
2689
2690 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
31aa02c5 2691 s->total, s->dropped, s->time_squeeze, 0,
c1ebcdb8
SH
2692 0, 0, 0, 0, /* was fastroute */
2693 s->cpu_collision );
1da177e4
LT
2694 return 0;
2695}
2696
f690808e 2697static const struct seq_operations dev_seq_ops = {
1da177e4
LT
2698 .start = dev_seq_start,
2699 .next = dev_seq_next,
2700 .stop = dev_seq_stop,
2701 .show = dev_seq_show,
2702};
2703
2704static int dev_seq_open(struct inode *inode, struct file *file)
2705{
e372c414
DL
2706 return seq_open_net(inode, file, &dev_seq_ops,
2707 sizeof(struct seq_net_private));
1da177e4
LT
2708}
2709
9a32144e 2710static const struct file_operations dev_seq_fops = {
1da177e4
LT
2711 .owner = THIS_MODULE,
2712 .open = dev_seq_open,
2713 .read = seq_read,
2714 .llseek = seq_lseek,
e372c414 2715 .release = seq_release_net,
1da177e4
LT
2716};
2717
f690808e 2718static const struct seq_operations softnet_seq_ops = {
1da177e4
LT
2719 .start = softnet_seq_start,
2720 .next = softnet_seq_next,
2721 .stop = softnet_seq_stop,
2722 .show = softnet_seq_show,
2723};
2724
2725static int softnet_seq_open(struct inode *inode, struct file *file)
2726{
2727 return seq_open(file, &softnet_seq_ops);
2728}
2729
9a32144e 2730static const struct file_operations softnet_seq_fops = {
1da177e4
LT
2731 .owner = THIS_MODULE,
2732 .open = softnet_seq_open,
2733 .read = seq_read,
2734 .llseek = seq_lseek,
2735 .release = seq_release,
2736};
2737
0e1256ff
SH
2738static void *ptype_get_idx(loff_t pos)
2739{
2740 struct packet_type *pt = NULL;
2741 loff_t i = 0;
2742 int t;
2743
2744 list_for_each_entry_rcu(pt, &ptype_all, list) {
2745 if (i == pos)
2746 return pt;
2747 ++i;
2748 }
2749
82d8a867 2750 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
0e1256ff
SH
2751 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2752 if (i == pos)
2753 return pt;
2754 ++i;
2755 }
2756 }
2757 return NULL;
2758}
2759
2760static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
72348a42 2761 __acquires(RCU)
0e1256ff
SH
2762{
2763 rcu_read_lock();
2764 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2765}
2766
2767static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2768{
2769 struct packet_type *pt;
2770 struct list_head *nxt;
2771 int hash;
2772
2773 ++*pos;
2774 if (v == SEQ_START_TOKEN)
2775 return ptype_get_idx(0);
2776
2777 pt = v;
2778 nxt = pt->list.next;
2779 if (pt->type == htons(ETH_P_ALL)) {
2780 if (nxt != &ptype_all)
2781 goto found;
2782 hash = 0;
2783 nxt = ptype_base[0].next;
2784 } else
82d8a867 2785 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
0e1256ff
SH
2786
2787 while (nxt == &ptype_base[hash]) {
82d8a867 2788 if (++hash >= PTYPE_HASH_SIZE)
0e1256ff
SH
2789 return NULL;
2790 nxt = ptype_base[hash].next;
2791 }
2792found:
2793 return list_entry(nxt, struct packet_type, list);
2794}
2795
2796static void ptype_seq_stop(struct seq_file *seq, void *v)
72348a42 2797 __releases(RCU)
0e1256ff
SH
2798{
2799 rcu_read_unlock();
2800}
2801
2802static void ptype_seq_decode(struct seq_file *seq, void *sym)
2803{
2804#ifdef CONFIG_KALLSYMS
2805 unsigned long offset = 0, symsize;
2806 const char *symname;
2807 char *modname;
2808 char namebuf[128];
2809
2810 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2811 &modname, namebuf);
2812
2813 if (symname) {
2814 char *delim = ":";
2815
2816 if (!modname)
2817 modname = delim = "";
2818 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2819 symname, offset);
2820 return;
2821 }
2822#endif
2823
2824 seq_printf(seq, "[%p]", sym);
2825}
2826
2827static int ptype_seq_show(struct seq_file *seq, void *v)
2828{
2829 struct packet_type *pt = v;
2830
2831 if (v == SEQ_START_TOKEN)
2832 seq_puts(seq, "Type Device Function\n");
c346dca1 2833 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
0e1256ff
SH
2834 if (pt->type == htons(ETH_P_ALL))
2835 seq_puts(seq, "ALL ");
2836 else
2837 seq_printf(seq, "%04x", ntohs(pt->type));
2838
2839 seq_printf(seq, " %-8s ",
2840 pt->dev ? pt->dev->name : "");
2841 ptype_seq_decode(seq, pt->func);
2842 seq_putc(seq, '\n');
2843 }
2844
2845 return 0;
2846}
2847
2848static const struct seq_operations ptype_seq_ops = {
2849 .start = ptype_seq_start,
2850 .next = ptype_seq_next,
2851 .stop = ptype_seq_stop,
2852 .show = ptype_seq_show,
2853};
2854
2855static int ptype_seq_open(struct inode *inode, struct file *file)
2856{
2feb27db
PE
2857 return seq_open_net(inode, file, &ptype_seq_ops,
2858 sizeof(struct seq_net_private));
0e1256ff
SH
2859}
2860
2861static const struct file_operations ptype_seq_fops = {
2862 .owner = THIS_MODULE,
2863 .open = ptype_seq_open,
2864 .read = seq_read,
2865 .llseek = seq_lseek,
2feb27db 2866 .release = seq_release_net,
0e1256ff
SH
2867};
2868
2869
4665079c 2870static int __net_init dev_proc_net_init(struct net *net)
1da177e4
LT
2871{
2872 int rc = -ENOMEM;
2873
881d966b 2874 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
1da177e4 2875 goto out;
881d966b 2876 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
1da177e4 2877 goto out_dev;
881d966b 2878 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
457c4cbc 2879 goto out_softnet;
0e1256ff 2880
881d966b 2881 if (wext_proc_init(net))
457c4cbc 2882 goto out_ptype;
1da177e4
LT
2883 rc = 0;
2884out:
2885 return rc;
457c4cbc 2886out_ptype:
881d966b 2887 proc_net_remove(net, "ptype");
1da177e4 2888out_softnet:
881d966b 2889 proc_net_remove(net, "softnet_stat");
1da177e4 2890out_dev:
881d966b 2891 proc_net_remove(net, "dev");
1da177e4
LT
2892 goto out;
2893}
881d966b 2894
4665079c 2895static void __net_exit dev_proc_net_exit(struct net *net)
881d966b
EB
2896{
2897 wext_proc_exit(net);
2898
2899 proc_net_remove(net, "ptype");
2900 proc_net_remove(net, "softnet_stat");
2901 proc_net_remove(net, "dev");
2902}
2903
022cbae6 2904static struct pernet_operations __net_initdata dev_proc_ops = {
881d966b
EB
2905 .init = dev_proc_net_init,
2906 .exit = dev_proc_net_exit,
2907};
2908
2909static int __init dev_proc_init(void)
2910{
2911 return register_pernet_subsys(&dev_proc_ops);
2912}
1da177e4
LT
2913#else
2914#define dev_proc_init() 0
2915#endif /* CONFIG_PROC_FS */
2916
2917
2918/**
2919 * netdev_set_master - set up master/slave pair
2920 * @slave: slave device
2921 * @master: new master device
2922 *
2923 * Changes the master device of the slave. Pass %NULL to break the
2924 * bonding. The caller must hold the RTNL semaphore. On a failure
2925 * a negative errno code is returned. On success the reference counts
2926 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2927 * function returns zero.
2928 */
2929int netdev_set_master(struct net_device *slave, struct net_device *master)
2930{
2931 struct net_device *old = slave->master;
2932
2933 ASSERT_RTNL();
2934
2935 if (master) {
2936 if (old)
2937 return -EBUSY;
2938 dev_hold(master);
2939 }
2940
2941 slave->master = master;
4ec93edb 2942
1da177e4
LT
2943 synchronize_net();
2944
2945 if (old)
2946 dev_put(old);
2947
2948 if (master)
2949 slave->flags |= IFF_SLAVE;
2950 else
2951 slave->flags &= ~IFF_SLAVE;
2952
2953 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2954 return 0;
2955}
2956
b6c40d68
PM
2957static void dev_change_rx_flags(struct net_device *dev, int flags)
2958{
2959 if (dev->flags & IFF_UP && dev->change_rx_flags)
2960 dev->change_rx_flags(dev, flags);
2961}
2962
dad9b335 2963static int __dev_set_promiscuity(struct net_device *dev, int inc)
1da177e4
LT
2964{
2965 unsigned short old_flags = dev->flags;
2966
24023451
PM
2967 ASSERT_RTNL();
2968
dad9b335
WC
2969 dev->flags |= IFF_PROMISC;
2970 dev->promiscuity += inc;
2971 if (dev->promiscuity == 0) {
2972 /*
2973 * Avoid overflow.
2974 * If inc causes overflow, untouch promisc and return error.
2975 */
2976 if (inc < 0)
2977 dev->flags &= ~IFF_PROMISC;
2978 else {
2979 dev->promiscuity -= inc;
2980 printk(KERN_WARNING "%s: promiscuity touches roof, "
2981 "set promiscuity failed, promiscuity feature "
2982 "of device might be broken.\n", dev->name);
2983 return -EOVERFLOW;
2984 }
2985 }
52609c0b 2986 if (dev->flags != old_flags) {
1da177e4
LT
2987 printk(KERN_INFO "device %s %s promiscuous mode\n",
2988 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4ec93edb 2989 "left");
7759db82
KHK
2990 if (audit_enabled)
2991 audit_log(current->audit_context, GFP_ATOMIC,
2992 AUDIT_ANOM_PROMISCUOUS,
2993 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2994 dev->name, (dev->flags & IFF_PROMISC),
2995 (old_flags & IFF_PROMISC),
2996 audit_get_loginuid(current),
2997 current->uid, current->gid,
2998 audit_get_sessionid(current));
24023451 2999
b6c40d68 3000 dev_change_rx_flags(dev, IFF_PROMISC);
1da177e4 3001 }
dad9b335 3002 return 0;
1da177e4
LT
3003}
3004
4417da66
PM
3005/**
3006 * dev_set_promiscuity - update promiscuity count on a device
3007 * @dev: device
3008 * @inc: modifier
3009 *
3010 * Add or remove promiscuity from a device. While the count in the device
3011 * remains above zero the interface remains promiscuous. Once it hits zero
3012 * the device reverts back to normal filtering operation. A negative inc
3013 * value is used to drop promiscuity on the device.
dad9b335 3014 * Return 0 if successful or a negative errno code on error.
4417da66 3015 */
dad9b335 3016int dev_set_promiscuity(struct net_device *dev, int inc)
4417da66
PM
3017{
3018 unsigned short old_flags = dev->flags;
dad9b335 3019 int err;
4417da66 3020
dad9b335 3021 err = __dev_set_promiscuity(dev, inc);
4b5a698e 3022 if (err < 0)
dad9b335 3023 return err;
4417da66
PM
3024 if (dev->flags != old_flags)
3025 dev_set_rx_mode(dev);
dad9b335 3026 return err;
4417da66
PM
3027}
3028
1da177e4
LT
3029/**
3030 * dev_set_allmulti - update allmulti count on a device
3031 * @dev: device
3032 * @inc: modifier
3033 *
3034 * Add or remove reception of all multicast frames to a device. While the
3035 * count in the device remains above zero the interface remains listening
3036 * to all interfaces. Once it hits zero the device reverts back to normal
3037 * filtering operation. A negative @inc value is used to drop the counter
3038 * when releasing a resource needing all multicasts.
dad9b335 3039 * Return 0 if successful or a negative errno code on error.
1da177e4
LT
3040 */
3041
dad9b335 3042int dev_set_allmulti(struct net_device *dev, int inc)
1da177e4
LT
3043{
3044 unsigned short old_flags = dev->flags;
3045
24023451
PM
3046 ASSERT_RTNL();
3047
1da177e4 3048 dev->flags |= IFF_ALLMULTI;
dad9b335
WC
3049 dev->allmulti += inc;
3050 if (dev->allmulti == 0) {
3051 /*
3052 * Avoid overflow.
3053 * If inc causes overflow, untouch allmulti and return error.
3054 */
3055 if (inc < 0)
3056 dev->flags &= ~IFF_ALLMULTI;
3057 else {
3058 dev->allmulti -= inc;
3059 printk(KERN_WARNING "%s: allmulti touches roof, "
3060 "set allmulti failed, allmulti feature of "
3061 "device might be broken.\n", dev->name);
3062 return -EOVERFLOW;
3063 }
3064 }
24023451 3065 if (dev->flags ^ old_flags) {
b6c40d68 3066 dev_change_rx_flags(dev, IFF_ALLMULTI);
4417da66 3067 dev_set_rx_mode(dev);
24023451 3068 }
dad9b335 3069 return 0;
4417da66
PM
3070}
3071
3072/*
3073 * Upload unicast and multicast address lists to device and
3074 * configure RX filtering. When the device doesn't support unicast
53ccaae1 3075 * filtering it is put in promiscuous mode while unicast addresses
4417da66
PM
3076 * are present.
3077 */
3078void __dev_set_rx_mode(struct net_device *dev)
3079{
3080 /* dev_open will call this function so the list will stay sane. */
3081 if (!(dev->flags&IFF_UP))
3082 return;
3083
3084 if (!netif_device_present(dev))
40b77c94 3085 return;
4417da66
PM
3086
3087 if (dev->set_rx_mode)
3088 dev->set_rx_mode(dev);
3089 else {
3090 /* Unicast addresses changes may only happen under the rtnl,
3091 * therefore calling __dev_set_promiscuity here is safe.
3092 */
3093 if (dev->uc_count > 0 && !dev->uc_promisc) {
3094 __dev_set_promiscuity(dev, 1);
3095 dev->uc_promisc = 1;
3096 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3097 __dev_set_promiscuity(dev, -1);
3098 dev->uc_promisc = 0;
3099 }
3100
3101 if (dev->set_multicast_list)
3102 dev->set_multicast_list(dev);
3103 }
3104}
3105
3106void dev_set_rx_mode(struct net_device *dev)
3107{
b9e40857 3108 netif_addr_lock_bh(dev);
4417da66 3109 __dev_set_rx_mode(dev);
b9e40857 3110 netif_addr_unlock_bh(dev);
1da177e4
LT
3111}
3112
61cbc2fc
PM
3113int __dev_addr_delete(struct dev_addr_list **list, int *count,
3114 void *addr, int alen, int glbl)
bf742482
PM
3115{
3116 struct dev_addr_list *da;
3117
3118 for (; (da = *list) != NULL; list = &da->next) {
3119 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3120 alen == da->da_addrlen) {
3121 if (glbl) {
3122 int old_glbl = da->da_gusers;
3123 da->da_gusers = 0;
3124 if (old_glbl == 0)
3125 break;
3126 }
3127 if (--da->da_users)
3128 return 0;
3129
3130 *list = da->next;
3131 kfree(da);
61cbc2fc 3132 (*count)--;
bf742482
PM
3133 return 0;
3134 }
3135 }
3136 return -ENOENT;
3137}
3138
61cbc2fc
PM
3139int __dev_addr_add(struct dev_addr_list **list, int *count,
3140 void *addr, int alen, int glbl)
bf742482
PM
3141{
3142 struct dev_addr_list *da;
3143
3144 for (da = *list; da != NULL; da = da->next) {
3145 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3146 da->da_addrlen == alen) {
3147 if (glbl) {
3148 int old_glbl = da->da_gusers;
3149 da->da_gusers = 1;
3150 if (old_glbl)
3151 return 0;
3152 }
3153 da->da_users++;
3154 return 0;
3155 }
3156 }
3157
12aa343a 3158 da = kzalloc(sizeof(*da), GFP_ATOMIC);
bf742482
PM
3159 if (da == NULL)
3160 return -ENOMEM;
3161 memcpy(da->da_addr, addr, alen);
3162 da->da_addrlen = alen;
3163 da->da_users = 1;
3164 da->da_gusers = glbl ? 1 : 0;
3165 da->next = *list;
3166 *list = da;
61cbc2fc 3167 (*count)++;
bf742482
PM
3168 return 0;
3169}
3170
4417da66
PM
3171/**
3172 * dev_unicast_delete - Release secondary unicast address.
3173 * @dev: device
0ed72ec4
RD
3174 * @addr: address to delete
3175 * @alen: length of @addr
4417da66
PM
3176 *
3177 * Release reference to a secondary unicast address and remove it
0ed72ec4 3178 * from the device if the reference count drops to zero.
4417da66
PM
3179 *
3180 * The caller must hold the rtnl_mutex.
3181 */
3182int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3183{
3184 int err;
3185
3186 ASSERT_RTNL();
3187
b9e40857 3188 netif_addr_lock_bh(dev);
61cbc2fc
PM
3189 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3190 if (!err)
4417da66 3191 __dev_set_rx_mode(dev);
b9e40857 3192 netif_addr_unlock_bh(dev);
4417da66
PM
3193 return err;
3194}
3195EXPORT_SYMBOL(dev_unicast_delete);
3196
3197/**
3198 * dev_unicast_add - add a secondary unicast address
3199 * @dev: device
5dbaec5d 3200 * @addr: address to add
0ed72ec4 3201 * @alen: length of @addr
4417da66
PM
3202 *
3203 * Add a secondary unicast address to the device or increase
3204 * the reference count if it already exists.
3205 *
3206 * The caller must hold the rtnl_mutex.
3207 */
3208int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3209{
3210 int err;
3211
3212 ASSERT_RTNL();
3213
b9e40857 3214 netif_addr_lock_bh(dev);
61cbc2fc
PM
3215 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3216 if (!err)
4417da66 3217 __dev_set_rx_mode(dev);
b9e40857 3218 netif_addr_unlock_bh(dev);
4417da66
PM
3219 return err;
3220}
3221EXPORT_SYMBOL(dev_unicast_add);
3222
e83a2ea8
CL
3223int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3224 struct dev_addr_list **from, int *from_count)
3225{
3226 struct dev_addr_list *da, *next;
3227 int err = 0;
3228
3229 da = *from;
3230 while (da != NULL) {
3231 next = da->next;
3232 if (!da->da_synced) {
3233 err = __dev_addr_add(to, to_count,
3234 da->da_addr, da->da_addrlen, 0);
3235 if (err < 0)
3236 break;
3237 da->da_synced = 1;
3238 da->da_users++;
3239 } else if (da->da_users == 1) {
3240 __dev_addr_delete(to, to_count,
3241 da->da_addr, da->da_addrlen, 0);
3242 __dev_addr_delete(from, from_count,
3243 da->da_addr, da->da_addrlen, 0);
3244 }
3245 da = next;
3246 }
3247 return err;
3248}
3249
3250void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3251 struct dev_addr_list **from, int *from_count)
3252{
3253 struct dev_addr_list *da, *next;
3254
3255 da = *from;
3256 while (da != NULL) {
3257 next = da->next;
3258 if (da->da_synced) {
3259 __dev_addr_delete(to, to_count,
3260 da->da_addr, da->da_addrlen, 0);
3261 da->da_synced = 0;
3262 __dev_addr_delete(from, from_count,
3263 da->da_addr, da->da_addrlen, 0);
3264 }
3265 da = next;
3266 }
3267}
3268
3269/**
3270 * dev_unicast_sync - Synchronize device's unicast list to another device
3271 * @to: destination device
3272 * @from: source device
3273 *
3274 * Add newly added addresses to the destination device and release
3275 * addresses that have no users left. The source device must be
3276 * locked by netif_tx_lock_bh.
3277 *
3278 * This function is intended to be called from the dev->set_rx_mode
3279 * function of layered software devices.
3280 */
3281int dev_unicast_sync(struct net_device *to, struct net_device *from)
3282{
3283 int err = 0;
3284
b9e40857 3285 netif_addr_lock_bh(to);
e83a2ea8
CL
3286 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3287 &from->uc_list, &from->uc_count);
3288 if (!err)
3289 __dev_set_rx_mode(to);
b9e40857 3290 netif_addr_unlock_bh(to);
e83a2ea8
CL
3291 return err;
3292}
3293EXPORT_SYMBOL(dev_unicast_sync);
3294
3295/**
bc2cda1e 3296 * dev_unicast_unsync - Remove synchronized addresses from the destination device
e83a2ea8
CL
3297 * @to: destination device
3298 * @from: source device
3299 *
3300 * Remove all addresses that were added to the destination device by
3301 * dev_unicast_sync(). This function is intended to be called from the
3302 * dev->stop function of layered software devices.
3303 */
3304void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3305{
b9e40857 3306 netif_addr_lock_bh(from);
e308a5d8 3307 netif_addr_lock(to);
e83a2ea8
CL
3308
3309 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3310 &from->uc_list, &from->uc_count);
3311 __dev_set_rx_mode(to);
3312
e308a5d8 3313 netif_addr_unlock(to);
b9e40857 3314 netif_addr_unlock_bh(from);
e83a2ea8
CL
3315}
3316EXPORT_SYMBOL(dev_unicast_unsync);
3317
12972621
DC
3318static void __dev_addr_discard(struct dev_addr_list **list)
3319{
3320 struct dev_addr_list *tmp;
3321
3322 while (*list != NULL) {
3323 tmp = *list;
3324 *list = tmp->next;
3325 if (tmp->da_users > tmp->da_gusers)
3326 printk("__dev_addr_discard: address leakage! "
3327 "da_users=%d\n", tmp->da_users);
3328 kfree(tmp);
3329 }
3330}
3331
26cc2522 3332static void dev_addr_discard(struct net_device *dev)
4417da66 3333{
b9e40857 3334 netif_addr_lock_bh(dev);
26cc2522 3335
4417da66
PM
3336 __dev_addr_discard(&dev->uc_list);
3337 dev->uc_count = 0;
4417da66 3338
456ad75c
DC
3339 __dev_addr_discard(&dev->mc_list);
3340 dev->mc_count = 0;
26cc2522 3341
b9e40857 3342 netif_addr_unlock_bh(dev);
456ad75c
DC
3343}
3344
f0db275a
SH
3345/**
3346 * dev_get_flags - get flags reported to userspace
3347 * @dev: device
3348 *
3349 * Get the combination of flag bits exported through APIs to userspace.
3350 */
1da177e4
LT
3351unsigned dev_get_flags(const struct net_device *dev)
3352{
3353 unsigned flags;
3354
3355 flags = (dev->flags & ~(IFF_PROMISC |
3356 IFF_ALLMULTI |
b00055aa
SR
3357 IFF_RUNNING |
3358 IFF_LOWER_UP |
3359 IFF_DORMANT)) |
1da177e4
LT
3360 (dev->gflags & (IFF_PROMISC |
3361 IFF_ALLMULTI));
3362
b00055aa
SR
3363 if (netif_running(dev)) {
3364 if (netif_oper_up(dev))
3365 flags |= IFF_RUNNING;
3366 if (netif_carrier_ok(dev))
3367 flags |= IFF_LOWER_UP;
3368 if (netif_dormant(dev))
3369 flags |= IFF_DORMANT;
3370 }
1da177e4
LT
3371
3372 return flags;
3373}
3374
f0db275a
SH
3375/**
3376 * dev_change_flags - change device settings
3377 * @dev: device
3378 * @flags: device state flags
3379 *
3380 * Change settings on device based state flags. The flags are
3381 * in the userspace exported format.
3382 */
1da177e4
LT
3383int dev_change_flags(struct net_device *dev, unsigned flags)
3384{
7c355f53 3385 int ret, changes;
1da177e4
LT
3386 int old_flags = dev->flags;
3387
24023451
PM
3388 ASSERT_RTNL();
3389
1da177e4
LT
3390 /*
3391 * Set the flags on our device.
3392 */
3393
3394 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3395 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3396 IFF_AUTOMEDIA)) |
3397 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3398 IFF_ALLMULTI));
3399
3400 /*
3401 * Load in the correct multicast list now the flags have changed.
3402 */
3403
b6c40d68
PM
3404 if ((old_flags ^ flags) & IFF_MULTICAST)
3405 dev_change_rx_flags(dev, IFF_MULTICAST);
24023451 3406
4417da66 3407 dev_set_rx_mode(dev);
1da177e4
LT
3408
3409 /*
3410 * Have we downed the interface. We handle IFF_UP ourselves
3411 * according to user attempts to set it, rather than blindly
3412 * setting it.
3413 */
3414
3415 ret = 0;
3416 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3417 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3418
3419 if (!ret)
4417da66 3420 dev_set_rx_mode(dev);
1da177e4
LT
3421 }
3422
3423 if (dev->flags & IFF_UP &&
3424 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3425 IFF_VOLATILE)))
056925ab 3426 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1da177e4
LT
3427
3428 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3429 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3430 dev->gflags ^= IFF_PROMISC;
3431 dev_set_promiscuity(dev, inc);
3432 }
3433
3434 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3435 is important. Some (broken) drivers set IFF_PROMISC, when
3436 IFF_ALLMULTI is requested not asking us and not reporting.
3437 */
3438 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3439 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3440 dev->gflags ^= IFF_ALLMULTI;
3441 dev_set_allmulti(dev, inc);
3442 }
3443
7c355f53
TG
3444 /* Exclude state transition flags, already notified */
3445 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3446 if (changes)
3447 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
1da177e4
LT
3448
3449 return ret;
3450}
3451
f0db275a
SH
3452/**
3453 * dev_set_mtu - Change maximum transfer unit
3454 * @dev: device
3455 * @new_mtu: new transfer unit
3456 *
3457 * Change the maximum transfer size of the network device.
3458 */
1da177e4
LT
3459int dev_set_mtu(struct net_device *dev, int new_mtu)
3460{
3461 int err;
3462
3463 if (new_mtu == dev->mtu)
3464 return 0;
3465
3466 /* MTU must be positive. */
3467 if (new_mtu < 0)
3468 return -EINVAL;
3469
3470 if (!netif_device_present(dev))
3471 return -ENODEV;
3472
3473 err = 0;
3474 if (dev->change_mtu)
3475 err = dev->change_mtu(dev, new_mtu);
3476 else
3477 dev->mtu = new_mtu;
3478 if (!err && dev->flags & IFF_UP)
056925ab 3479 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
1da177e4
LT
3480 return err;
3481}
3482
f0db275a
SH
3483/**
3484 * dev_set_mac_address - Change Media Access Control Address
3485 * @dev: device
3486 * @sa: new address
3487 *
3488 * Change the hardware (MAC) address of the device
3489 */
1da177e4
LT
3490int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3491{
3492 int err;
3493
3494 if (!dev->set_mac_address)
3495 return -EOPNOTSUPP;
3496 if (sa->sa_family != dev->type)
3497 return -EINVAL;
3498 if (!netif_device_present(dev))
3499 return -ENODEV;
3500 err = dev->set_mac_address(dev, sa);
3501 if (!err)
056925ab 3502 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
1da177e4
LT
3503 return err;
3504}
3505
3506/*
14e3e079 3507 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
1da177e4 3508 */
14e3e079 3509static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
1da177e4
LT
3510{
3511 int err;
881d966b 3512 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
1da177e4
LT
3513
3514 if (!dev)
3515 return -ENODEV;
3516
3517 switch (cmd) {
3518 case SIOCGIFFLAGS: /* Get interface flags */
3519 ifr->ifr_flags = dev_get_flags(dev);
3520 return 0;
3521
1da177e4
LT
3522 case SIOCGIFMETRIC: /* Get the metric on the interface
3523 (currently unused) */
3524 ifr->ifr_metric = 0;
3525 return 0;
3526
1da177e4
LT
3527 case SIOCGIFMTU: /* Get the MTU of a device */
3528 ifr->ifr_mtu = dev->mtu;
3529 return 0;
3530
1da177e4
LT
3531 case SIOCGIFHWADDR:
3532 if (!dev->addr_len)
3533 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3534 else
3535 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3536 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3537 ifr->ifr_hwaddr.sa_family = dev->type;
3538 return 0;
3539
14e3e079
JG
3540 case SIOCGIFSLAVE:
3541 err = -EINVAL;
3542 break;
3543
3544 case SIOCGIFMAP:
3545 ifr->ifr_map.mem_start = dev->mem_start;
3546 ifr->ifr_map.mem_end = dev->mem_end;
3547 ifr->ifr_map.base_addr = dev->base_addr;
3548 ifr->ifr_map.irq = dev->irq;
3549 ifr->ifr_map.dma = dev->dma;
3550 ifr->ifr_map.port = dev->if_port;
3551 return 0;
3552
3553 case SIOCGIFINDEX:
3554 ifr->ifr_ifindex = dev->ifindex;
3555 return 0;
3556
3557 case SIOCGIFTXQLEN:
3558 ifr->ifr_qlen = dev->tx_queue_len;
3559 return 0;
3560
3561 default:
3562 /* dev_ioctl() should ensure this case
3563 * is never reached
3564 */
3565 WARN_ON(1);
3566 err = -EINVAL;
3567 break;
3568
3569 }
3570 return err;
3571}
3572
3573/*
3574 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3575 */
3576static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3577{
3578 int err;
3579 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3580
3581 if (!dev)
3582 return -ENODEV;
3583
3584 switch (cmd) {
3585 case SIOCSIFFLAGS: /* Set interface flags */
3586 return dev_change_flags(dev, ifr->ifr_flags);
3587
3588 case SIOCSIFMETRIC: /* Set the metric on the interface
3589 (currently unused) */
3590 return -EOPNOTSUPP;
3591
3592 case SIOCSIFMTU: /* Set the MTU of a device */
3593 return dev_set_mtu(dev, ifr->ifr_mtu);
3594
1da177e4
LT
3595 case SIOCSIFHWADDR:
3596 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3597
3598 case SIOCSIFHWBROADCAST:
3599 if (ifr->ifr_hwaddr.sa_family != dev->type)
3600 return -EINVAL;
3601 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3602 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
056925ab 3603 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
1da177e4
LT
3604 return 0;
3605
1da177e4
LT
3606 case SIOCSIFMAP:
3607 if (dev->set_config) {
3608 if (!netif_device_present(dev))
3609 return -ENODEV;
3610 return dev->set_config(dev, &ifr->ifr_map);
3611 }
3612 return -EOPNOTSUPP;
3613
3614 case SIOCADDMULTI:
61ee6bd4 3615 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
1da177e4
LT
3616 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3617 return -EINVAL;
3618 if (!netif_device_present(dev))
3619 return -ENODEV;
3620 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3621 dev->addr_len, 1);
3622
3623 case SIOCDELMULTI:
61ee6bd4 3624 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
1da177e4
LT
3625 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3626 return -EINVAL;
3627 if (!netif_device_present(dev))
3628 return -ENODEV;
3629 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3630 dev->addr_len, 1);
3631
1da177e4
LT
3632 case SIOCSIFTXQLEN:
3633 if (ifr->ifr_qlen < 0)
3634 return -EINVAL;
3635 dev->tx_queue_len = ifr->ifr_qlen;
3636 return 0;
3637
3638 case SIOCSIFNAME:
3639 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3640 return dev_change_name(dev, ifr->ifr_newname);
3641
3642 /*
3643 * Unknown or private ioctl
3644 */
3645
3646 default:
3647 if ((cmd >= SIOCDEVPRIVATE &&
3648 cmd <= SIOCDEVPRIVATE + 15) ||
3649 cmd == SIOCBONDENSLAVE ||
3650 cmd == SIOCBONDRELEASE ||
3651 cmd == SIOCBONDSETHWADDR ||
3652 cmd == SIOCBONDSLAVEINFOQUERY ||
3653 cmd == SIOCBONDINFOQUERY ||
3654 cmd == SIOCBONDCHANGEACTIVE ||
3655 cmd == SIOCGMIIPHY ||
3656 cmd == SIOCGMIIREG ||
3657 cmd == SIOCSMIIREG ||
3658 cmd == SIOCBRADDIF ||
3659 cmd == SIOCBRDELIF ||
3660 cmd == SIOCWANDEV) {
3661 err = -EOPNOTSUPP;
3662 if (dev->do_ioctl) {
3663 if (netif_device_present(dev))
3664 err = dev->do_ioctl(dev, ifr,
3665 cmd);
3666 else
3667 err = -ENODEV;
3668 }
3669 } else
3670 err = -EINVAL;
3671
3672 }
3673 return err;
3674}
3675
3676/*
3677 * This function handles all "interface"-type I/O control requests. The actual
3678 * 'doing' part of this is dev_ifsioc above.
3679 */
3680
3681/**
3682 * dev_ioctl - network device ioctl
c4ea43c5 3683 * @net: the applicable net namespace
1da177e4
LT
3684 * @cmd: command to issue
3685 * @arg: pointer to a struct ifreq in user space
3686 *
3687 * Issue ioctl functions to devices. This is normally called by the
3688 * user space syscall interfaces but can sometimes be useful for
3689 * other purposes. The return value is the return from the syscall if
3690 * positive or a negative errno code on error.
3691 */
3692
881d966b 3693int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
1da177e4
LT
3694{
3695 struct ifreq ifr;
3696 int ret;
3697 char *colon;
3698
3699 /* One special case: SIOCGIFCONF takes ifconf argument
3700 and requires shared lock, because it sleeps writing
3701 to user space.
3702 */
3703
3704 if (cmd == SIOCGIFCONF) {
6756ae4b 3705 rtnl_lock();
881d966b 3706 ret = dev_ifconf(net, (char __user *) arg);
6756ae4b 3707 rtnl_unlock();
1da177e4
LT
3708 return ret;
3709 }
3710 if (cmd == SIOCGIFNAME)
881d966b 3711 return dev_ifname(net, (struct ifreq __user *)arg);
1da177e4
LT
3712
3713 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3714 return -EFAULT;
3715
3716 ifr.ifr_name[IFNAMSIZ-1] = 0;
3717
3718 colon = strchr(ifr.ifr_name, ':');
3719 if (colon)
3720 *colon = 0;
3721
3722 /*
3723 * See which interface the caller is talking about.
3724 */
3725
3726 switch (cmd) {
3727 /*
3728 * These ioctl calls:
3729 * - can be done by all.
3730 * - atomic and do not require locking.
3731 * - return a value
3732 */
3733 case SIOCGIFFLAGS:
3734 case SIOCGIFMETRIC:
3735 case SIOCGIFMTU:
3736 case SIOCGIFHWADDR:
3737 case SIOCGIFSLAVE:
3738 case SIOCGIFMAP:
3739 case SIOCGIFINDEX:
3740 case SIOCGIFTXQLEN:
881d966b 3741 dev_load(net, ifr.ifr_name);
1da177e4 3742 read_lock(&dev_base_lock);
14e3e079 3743 ret = dev_ifsioc_locked(net, &ifr, cmd);
1da177e4
LT
3744 read_unlock(&dev_base_lock);
3745 if (!ret) {
3746 if (colon)
3747 *colon = ':';
3748 if (copy_to_user(arg, &ifr,
3749 sizeof(struct ifreq)))
3750 ret = -EFAULT;
3751 }
3752 return ret;
3753
3754 case SIOCETHTOOL:
881d966b 3755 dev_load(net, ifr.ifr_name);
1da177e4 3756 rtnl_lock();
881d966b 3757 ret = dev_ethtool(net, &ifr);
1da177e4
LT
3758 rtnl_unlock();
3759 if (!ret) {
3760 if (colon)
3761 *colon = ':';
3762 if (copy_to_user(arg, &ifr,
3763 sizeof(struct ifreq)))
3764 ret = -EFAULT;
3765 }
3766 return ret;
3767
3768 /*
3769 * These ioctl calls:
3770 * - require superuser power.
3771 * - require strict serialization.
3772 * - return a value
3773 */
3774 case SIOCGMIIPHY:
3775 case SIOCGMIIREG:
3776 case SIOCSIFNAME:
3777 if (!capable(CAP_NET_ADMIN))
3778 return -EPERM;
881d966b 3779 dev_load(net, ifr.ifr_name);
1da177e4 3780 rtnl_lock();
881d966b 3781 ret = dev_ifsioc(net, &ifr, cmd);
1da177e4
LT
3782 rtnl_unlock();
3783 if (!ret) {
3784 if (colon)
3785 *colon = ':';
3786 if (copy_to_user(arg, &ifr,
3787 sizeof(struct ifreq)))
3788 ret = -EFAULT;
3789 }
3790 return ret;
3791
3792 /*
3793 * These ioctl calls:
3794 * - require superuser power.
3795 * - require strict serialization.
3796 * - do not return a value
3797 */
3798 case SIOCSIFFLAGS:
3799 case SIOCSIFMETRIC:
3800 case SIOCSIFMTU:
3801 case SIOCSIFMAP:
3802 case SIOCSIFHWADDR:
3803 case SIOCSIFSLAVE:
3804 case SIOCADDMULTI:
3805 case SIOCDELMULTI:
3806 case SIOCSIFHWBROADCAST:
3807 case SIOCSIFTXQLEN:
3808 case SIOCSMIIREG:
3809 case SIOCBONDENSLAVE:
3810 case SIOCBONDRELEASE:
3811 case SIOCBONDSETHWADDR:
1da177e4
LT
3812 case SIOCBONDCHANGEACTIVE:
3813 case SIOCBRADDIF:
3814 case SIOCBRDELIF:
3815 if (!capable(CAP_NET_ADMIN))
3816 return -EPERM;
cabcac0b
TG
3817 /* fall through */
3818 case SIOCBONDSLAVEINFOQUERY:
3819 case SIOCBONDINFOQUERY:
881d966b 3820 dev_load(net, ifr.ifr_name);
1da177e4 3821 rtnl_lock();
881d966b 3822 ret = dev_ifsioc(net, &ifr, cmd);
1da177e4
LT
3823 rtnl_unlock();
3824 return ret;
3825
3826 case SIOCGIFMEM:
3827 /* Get the per device memory space. We can add this but
3828 * currently do not support it */
3829 case SIOCSIFMEM:
3830 /* Set the per device memory buffer space.
3831 * Not applicable in our case */
3832 case SIOCSIFLINK:
3833 return -EINVAL;
3834
3835 /*
3836 * Unknown or private ioctl.
3837 */
3838 default:
3839 if (cmd == SIOCWANDEV ||
3840 (cmd >= SIOCDEVPRIVATE &&
3841 cmd <= SIOCDEVPRIVATE + 15)) {
881d966b 3842 dev_load(net, ifr.ifr_name);
1da177e4 3843 rtnl_lock();
881d966b 3844 ret = dev_ifsioc(net, &ifr, cmd);
1da177e4
LT
3845 rtnl_unlock();
3846 if (!ret && copy_to_user(arg, &ifr,
3847 sizeof(struct ifreq)))
3848 ret = -EFAULT;
3849 return ret;
3850 }
1da177e4 3851 /* Take care of Wireless Extensions */
295f4a1f 3852 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
881d966b 3853 return wext_handle_ioctl(net, &ifr, cmd, arg);
1da177e4
LT
3854 return -EINVAL;
3855 }
3856}
3857
3858
3859/**
3860 * dev_new_index - allocate an ifindex
c4ea43c5 3861 * @net: the applicable net namespace
1da177e4
LT
3862 *
3863 * Returns a suitable unique value for a new device interface
3864 * number. The caller must hold the rtnl semaphore or the
3865 * dev_base_lock to be sure it remains unique.
3866 */
881d966b 3867static int dev_new_index(struct net *net)
1da177e4
LT
3868{
3869 static int ifindex;
3870 for (;;) {
3871 if (++ifindex <= 0)
3872 ifindex = 1;
881d966b 3873 if (!__dev_get_by_index(net, ifindex))
1da177e4
LT
3874 return ifindex;
3875 }
3876}
3877
1da177e4 3878/* Delayed registration/unregisteration */
3b5b34fd 3879static LIST_HEAD(net_todo_list);
1da177e4 3880
6f05f629 3881static void net_set_todo(struct net_device *dev)
1da177e4 3882{
1da177e4 3883 list_add_tail(&dev->todo_list, &net_todo_list);
1da177e4
LT
3884}
3885
93ee31f1
DL
3886static void rollback_registered(struct net_device *dev)
3887{
3888 BUG_ON(dev_boot_phase);
3889 ASSERT_RTNL();
3890
3891 /* Some devices call without registering for initialization unwind. */
3892 if (dev->reg_state == NETREG_UNINITIALIZED) {
3893 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3894 "was registered\n", dev->name, dev);
3895
3896 WARN_ON(1);
3897 return;
3898 }
3899
3900 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3901
3902 /* If device is running, close it first. */
3903 dev_close(dev);
3904
3905 /* And unlink it from device chain. */
3906 unlist_netdevice(dev);
3907
3908 dev->reg_state = NETREG_UNREGISTERING;
3909
3910 synchronize_net();
3911
3912 /* Shutdown queueing discipline. */
3913 dev_shutdown(dev);
3914
3915
3916 /* Notify protocols, that we are about to destroy
3917 this device. They should clean all the things.
3918 */
3919 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3920
3921 /*
3922 * Flush the unicast and multicast chains
3923 */
3924 dev_addr_discard(dev);
3925
3926 if (dev->uninit)
3927 dev->uninit(dev);
3928
3929 /* Notifier chain MUST detach us from master device. */
547b792c 3930 WARN_ON(dev->master);
93ee31f1
DL
3931
3932 /* Remove entries from kobject tree */
3933 netdev_unregister_kobject(dev);
3934
3935 synchronize_net();
3936
3937 dev_put(dev);
3938}
3939
e8a0464c
DM
3940static void __netdev_init_queue_locks_one(struct net_device *dev,
3941 struct netdev_queue *dev_queue,
3942 void *_unused)
c773e847
DM
3943{
3944 spin_lock_init(&dev_queue->_xmit_lock);
cf508b12 3945 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
c773e847
DM
3946 dev_queue->xmit_lock_owner = -1;
3947}
3948
3949static void netdev_init_queue_locks(struct net_device *dev)
3950{
e8a0464c
DM
3951 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3952 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
c773e847
DM
3953}
3954
b63365a2
HX
3955unsigned long netdev_fix_features(unsigned long features, const char *name)
3956{
3957 /* Fix illegal SG+CSUM combinations. */
3958 if ((features & NETIF_F_SG) &&
3959 !(features & NETIF_F_ALL_CSUM)) {
3960 if (name)
3961 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
3962 "checksum feature.\n", name);
3963 features &= ~NETIF_F_SG;
3964 }
3965
3966 /* TSO requires that SG is present as well. */
3967 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
3968 if (name)
3969 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
3970 "SG feature.\n", name);
3971 features &= ~NETIF_F_TSO;
3972 }
3973
3974 if (features & NETIF_F_UFO) {
3975 if (!(features & NETIF_F_GEN_CSUM)) {
3976 if (name)
3977 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3978 "since no NETIF_F_HW_CSUM feature.\n",
3979 name);
3980 features &= ~NETIF_F_UFO;
3981 }
3982
3983 if (!(features & NETIF_F_SG)) {
3984 if (name)
3985 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3986 "since no NETIF_F_SG feature.\n", name);
3987 features &= ~NETIF_F_UFO;
3988 }
3989 }
3990
3991 return features;
3992}
3993EXPORT_SYMBOL(netdev_fix_features);
3994
1da177e4
LT
3995/**
3996 * register_netdevice - register a network device
3997 * @dev: device to register
3998 *
3999 * Take a completed network device structure and add it to the kernel
4000 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4001 * chain. 0 is returned on success. A negative errno code is returned
4002 * on a failure to set up the device, or if the name is a duplicate.
4003 *
4004 * Callers must hold the rtnl semaphore. You may want
4005 * register_netdev() instead of this.
4006 *
4007 * BUGS:
4008 * The locking appears insufficient to guarantee two parallel registers
4009 * will not get the same name.
4010 */
4011
4012int register_netdevice(struct net_device *dev)
4013{
4014 struct hlist_head *head;
4015 struct hlist_node *p;
4016 int ret;
881d966b 4017 struct net *net;
1da177e4
LT
4018
4019 BUG_ON(dev_boot_phase);
4020 ASSERT_RTNL();
4021
b17a7c17
SH
4022 might_sleep();
4023
1da177e4
LT
4024 /* When net_device's are persistent, this will be fatal. */
4025 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
c346dca1
YH
4026 BUG_ON(!dev_net(dev));
4027 net = dev_net(dev);
1da177e4 4028
f1f28aa3 4029 spin_lock_init(&dev->addr_list_lock);
cf508b12 4030 netdev_set_addr_lockdep_class(dev);
c773e847 4031 netdev_init_queue_locks(dev);
1da177e4 4032
1da177e4
LT
4033 dev->iflink = -1;
4034
4035 /* Init, if this function is available */
4036 if (dev->init) {
4037 ret = dev->init(dev);
4038 if (ret) {
4039 if (ret > 0)
4040 ret = -EIO;
90833aa4 4041 goto out;
1da177e4
LT
4042 }
4043 }
4ec93edb 4044
1da177e4
LT
4045 if (!dev_valid_name(dev->name)) {
4046 ret = -EINVAL;
7ce1b0ed 4047 goto err_uninit;
1da177e4
LT
4048 }
4049
881d966b 4050 dev->ifindex = dev_new_index(net);
1da177e4
LT
4051 if (dev->iflink == -1)
4052 dev->iflink = dev->ifindex;
4053
4054 /* Check for existence of name */
881d966b 4055 head = dev_name_hash(net, dev->name);
1da177e4
LT
4056 hlist_for_each(p, head) {
4057 struct net_device *d
4058 = hlist_entry(p, struct net_device, name_hlist);
4059 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4060 ret = -EEXIST;
7ce1b0ed 4061 goto err_uninit;
1da177e4 4062 }
4ec93edb 4063 }
1da177e4 4064
d212f87b
SH
4065 /* Fix illegal checksum combinations */
4066 if ((dev->features & NETIF_F_HW_CSUM) &&
4067 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4068 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4069 dev->name);
4070 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4071 }
4072
4073 if ((dev->features & NETIF_F_NO_CSUM) &&
4074 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4075 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4076 dev->name);
4077 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4078 }
4079
b63365a2 4080 dev->features = netdev_fix_features(dev->features, dev->name);
1da177e4 4081
e5a4a72d
LB
4082 /* Enable software GSO if SG is supported. */
4083 if (dev->features & NETIF_F_SG)
4084 dev->features |= NETIF_F_GSO;
4085
aaf8cdc3 4086 netdev_initialize_kobject(dev);
8b41d188 4087 ret = netdev_register_kobject(dev);
b17a7c17 4088 if (ret)
7ce1b0ed 4089 goto err_uninit;
b17a7c17
SH
4090 dev->reg_state = NETREG_REGISTERED;
4091
1da177e4
LT
4092 /*
4093 * Default initial state at registry is that the
4094 * device is present.
4095 */
4096
4097 set_bit(__LINK_STATE_PRESENT, &dev->state);
4098
1da177e4 4099 dev_init_scheduler(dev);
1da177e4 4100 dev_hold(dev);
ce286d32 4101 list_netdevice(dev);
1da177e4
LT
4102
4103 /* Notify protocols, that a new device appeared. */
056925ab 4104 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
fcc5a03a 4105 ret = notifier_to_errno(ret);
93ee31f1
DL
4106 if (ret) {
4107 rollback_registered(dev);
4108 dev->reg_state = NETREG_UNREGISTERED;
4109 }
1da177e4
LT
4110
4111out:
4112 return ret;
7ce1b0ed
HX
4113
4114err_uninit:
4115 if (dev->uninit)
4116 dev->uninit(dev);
4117 goto out;
1da177e4
LT
4118}
4119
4120/**
4121 * register_netdev - register a network device
4122 * @dev: device to register
4123 *
4124 * Take a completed network device structure and add it to the kernel
4125 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4126 * chain. 0 is returned on success. A negative errno code is returned
4127 * on a failure to set up the device, or if the name is a duplicate.
4128 *
38b4da38 4129 * This is a wrapper around register_netdevice that takes the rtnl semaphore
1da177e4
LT
4130 * and expands the device name if you passed a format string to
4131 * alloc_netdev.
4132 */
4133int register_netdev(struct net_device *dev)
4134{
4135 int err;
4136
4137 rtnl_lock();
4138
4139 /*
4140 * If the name is a format string the caller wants us to do a
4141 * name allocation.
4142 */
4143 if (strchr(dev->name, '%')) {
4144 err = dev_alloc_name(dev, dev->name);
4145 if (err < 0)
4146 goto out;
4147 }
4ec93edb 4148
1da177e4
LT
4149 err = register_netdevice(dev);
4150out:
4151 rtnl_unlock();
4152 return err;
4153}
4154EXPORT_SYMBOL(register_netdev);
4155
4156/*
4157 * netdev_wait_allrefs - wait until all references are gone.
4158 *
4159 * This is called when unregistering network devices.
4160 *
4161 * Any protocol or device that holds a reference should register
4162 * for netdevice notification, and cleanup and put back the
4163 * reference if they receive an UNREGISTER event.
4164 * We can get stuck here if buggy protocols don't correctly
4ec93edb 4165 * call dev_put.
1da177e4
LT
4166 */
4167static void netdev_wait_allrefs(struct net_device *dev)
4168{
4169 unsigned long rebroadcast_time, warning_time;
4170
4171 rebroadcast_time = warning_time = jiffies;
4172 while (atomic_read(&dev->refcnt) != 0) {
4173 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6756ae4b 4174 rtnl_lock();
1da177e4
LT
4175
4176 /* Rebroadcast unregister notification */
056925ab 4177 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
1da177e4
LT
4178
4179 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4180 &dev->state)) {
4181 /* We must not have linkwatch events
4182 * pending on unregister. If this
4183 * happens, we simply run the queue
4184 * unscheduled, resulting in a noop
4185 * for this device.
4186 */
4187 linkwatch_run_queue();
4188 }
4189
6756ae4b 4190 __rtnl_unlock();
1da177e4
LT
4191
4192 rebroadcast_time = jiffies;
4193 }
4194
4195 msleep(250);
4196
4197 if (time_after(jiffies, warning_time + 10 * HZ)) {
4198 printk(KERN_EMERG "unregister_netdevice: "
4199 "waiting for %s to become free. Usage "
4200 "count = %d\n",
4201 dev->name, atomic_read(&dev->refcnt));
4202 warning_time = jiffies;
4203 }
4204 }
4205}
4206
4207/* The sequence is:
4208 *
4209 * rtnl_lock();
4210 * ...
4211 * register_netdevice(x1);
4212 * register_netdevice(x2);
4213 * ...
4214 * unregister_netdevice(y1);
4215 * unregister_netdevice(y2);
4216 * ...
4217 * rtnl_unlock();
4218 * free_netdev(y1);
4219 * free_netdev(y2);
4220 *
58ec3b4d 4221 * We are invoked by rtnl_unlock().
1da177e4 4222 * This allows us to deal with problems:
b17a7c17 4223 * 1) We can delete sysfs objects which invoke hotplug
1da177e4
LT
4224 * without deadlocking with linkwatch via keventd.
4225 * 2) Since we run with the RTNL semaphore not held, we can sleep
4226 * safely in order to wait for the netdev refcnt to drop to zero.
58ec3b4d
HX
4227 *
4228 * We must not return until all unregister events added during
4229 * the interval the lock was held have been completed.
1da177e4 4230 */
1da177e4
LT
4231void netdev_run_todo(void)
4232{
626ab0e6 4233 struct list_head list;
1da177e4 4234
1da177e4 4235 /* Snapshot list, allow later requests */
626ab0e6 4236 list_replace_init(&net_todo_list, &list);
58ec3b4d
HX
4237
4238 __rtnl_unlock();
626ab0e6 4239
1da177e4
LT
4240 while (!list_empty(&list)) {
4241 struct net_device *dev
4242 = list_entry(list.next, struct net_device, todo_list);
4243 list_del(&dev->todo_list);
4244
b17a7c17
SH
4245 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4246 printk(KERN_ERR "network todo '%s' but state %d\n",
4247 dev->name, dev->reg_state);
4248 dump_stack();
4249 continue;
4250 }
1da177e4 4251
b17a7c17 4252 dev->reg_state = NETREG_UNREGISTERED;
1da177e4 4253
6e583ce5
SH
4254 on_each_cpu(flush_backlog, dev, 1);
4255
b17a7c17 4256 netdev_wait_allrefs(dev);
1da177e4 4257
b17a7c17
SH
4258 /* paranoia */
4259 BUG_ON(atomic_read(&dev->refcnt));
547b792c
IJ
4260 WARN_ON(dev->ip_ptr);
4261 WARN_ON(dev->ip6_ptr);
4262 WARN_ON(dev->dn_ptr);
1da177e4 4263
b17a7c17
SH
4264 if (dev->destructor)
4265 dev->destructor(dev);
9093bbb2
SH
4266
4267 /* Free network device */
4268 kobject_put(&dev->dev.kobj);
1da177e4 4269 }
1da177e4
LT
4270}
4271
5a1b5898 4272static struct net_device_stats *internal_stats(struct net_device *dev)
c45d286e 4273{
5a1b5898 4274 return &dev->stats;
c45d286e
RR
4275}
4276
dc2b4847 4277static void netdev_init_one_queue(struct net_device *dev,
e8a0464c
DM
4278 struct netdev_queue *queue,
4279 void *_unused)
dc2b4847 4280{
dc2b4847
DM
4281 queue->dev = dev;
4282}
4283
bb949fbd
DM
4284static void netdev_init_queues(struct net_device *dev)
4285{
e8a0464c
DM
4286 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4287 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
c3f26a26 4288 spin_lock_init(&dev->tx_global_lock);
bb949fbd
DM
4289}
4290
1da177e4 4291/**
f25f4e44 4292 * alloc_netdev_mq - allocate network device
1da177e4
LT
4293 * @sizeof_priv: size of private data to allocate space for
4294 * @name: device name format string
4295 * @setup: callback to initialize device
f25f4e44 4296 * @queue_count: the number of subqueues to allocate
1da177e4
LT
4297 *
4298 * Allocates a struct net_device with private data area for driver use
f25f4e44
PWJ
4299 * and performs basic initialization. Also allocates subquue structs
4300 * for each queue on the device at the end of the netdevice.
1da177e4 4301 */
f25f4e44
PWJ
4302struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4303 void (*setup)(struct net_device *), unsigned int queue_count)
1da177e4 4304{
e8a0464c 4305 struct netdev_queue *tx;
1da177e4 4306 struct net_device *dev;
7943986c 4307 size_t alloc_size;
e8a0464c 4308 void *p;
1da177e4 4309
b6fe17d6
SH
4310 BUG_ON(strlen(name) >= sizeof(dev->name));
4311
fd2ea0a7 4312 alloc_size = sizeof(struct net_device);
d1643d24
AD
4313 if (sizeof_priv) {
4314 /* ensure 32-byte alignment of private area */
4315 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4316 alloc_size += sizeof_priv;
4317 }
4318 /* ensure 32-byte alignment of whole construct */
4319 alloc_size += NETDEV_ALIGN_CONST;
1da177e4 4320
31380de9 4321 p = kzalloc(alloc_size, GFP_KERNEL);
1da177e4 4322 if (!p) {
b6fe17d6 4323 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
1da177e4
LT
4324 return NULL;
4325 }
1da177e4 4326
7943986c 4327 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
e8a0464c
DM
4328 if (!tx) {
4329 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4330 "tx qdiscs.\n");
4331 kfree(p);
4332 return NULL;
4333 }
4334
1da177e4
LT
4335 dev = (struct net_device *)
4336 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4337 dev->padded = (char *)dev - (char *)p;
c346dca1 4338 dev_net_set(dev, &init_net);
1da177e4 4339
e8a0464c
DM
4340 dev->_tx = tx;
4341 dev->num_tx_queues = queue_count;
fd2ea0a7 4342 dev->real_num_tx_queues = queue_count;
e8a0464c 4343
f25f4e44
PWJ
4344 if (sizeof_priv) {
4345 dev->priv = ((char *)dev +
fd2ea0a7 4346 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
f25f4e44
PWJ
4347 & ~NETDEV_ALIGN_CONST));
4348 }
4349
82cc1a7a 4350 dev->gso_max_size = GSO_MAX_SIZE;
1da177e4 4351
bb949fbd
DM
4352 netdev_init_queues(dev);
4353
5a1b5898 4354 dev->get_stats = internal_stats;
bea3348e 4355 netpoll_netdev_init(dev);
1da177e4
LT
4356 setup(dev);
4357 strcpy(dev->name, name);
4358 return dev;
4359}
f25f4e44 4360EXPORT_SYMBOL(alloc_netdev_mq);
1da177e4
LT
4361
4362/**
4363 * free_netdev - free network device
4364 * @dev: device
4365 *
4ec93edb
YH
4366 * This function does the last stage of destroying an allocated device
4367 * interface. The reference to the device object is released.
1da177e4
LT
4368 * If this is the last reference then it will be freed.
4369 */
4370void free_netdev(struct net_device *dev)
4371{
f3005d7f
DL
4372 release_net(dev_net(dev));
4373
e8a0464c
DM
4374 kfree(dev->_tx);
4375
3041a069 4376 /* Compatibility with error handling in drivers */
1da177e4
LT
4377 if (dev->reg_state == NETREG_UNINITIALIZED) {
4378 kfree((char *)dev - dev->padded);
4379 return;
4380 }
4381
4382 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4383 dev->reg_state = NETREG_RELEASED;
4384
43cb76d9
GKH
4385 /* will free via device release */
4386 put_device(&dev->dev);
1da177e4 4387}
4ec93edb 4388
f0db275a
SH
4389/**
4390 * synchronize_net - Synchronize with packet receive processing
4391 *
4392 * Wait for packets currently being received to be done.
4393 * Does not block later packets from starting.
4394 */
4ec93edb 4395void synchronize_net(void)
1da177e4
LT
4396{
4397 might_sleep();
fbd568a3 4398 synchronize_rcu();
1da177e4
LT
4399}
4400
4401/**
4402 * unregister_netdevice - remove device from the kernel
4403 * @dev: device
4404 *
4405 * This function shuts down a device interface and removes it
d59b54b1 4406 * from the kernel tables.
1da177e4
LT
4407 *
4408 * Callers must hold the rtnl semaphore. You may want
4409 * unregister_netdev() instead of this.
4410 */
4411
22f8cde5 4412void unregister_netdevice(struct net_device *dev)
1da177e4 4413{
a6620712
HX
4414 ASSERT_RTNL();
4415
93ee31f1 4416 rollback_registered(dev);
1da177e4
LT
4417 /* Finish processing unregister after unlock */
4418 net_set_todo(dev);
1da177e4
LT
4419}
4420
4421/**
4422 * unregister_netdev - remove device from the kernel
4423 * @dev: device
4424 *
4425 * This function shuts down a device interface and removes it
d59b54b1 4426 * from the kernel tables.
1da177e4
LT
4427 *
4428 * This is just a wrapper for unregister_netdevice that takes
4429 * the rtnl semaphore. In general you want to use this and not
4430 * unregister_netdevice.
4431 */
4432void unregister_netdev(struct net_device *dev)
4433{
4434 rtnl_lock();
4435 unregister_netdevice(dev);
4436 rtnl_unlock();
4437}
4438
4439EXPORT_SYMBOL(unregister_netdev);
4440
ce286d32
EB
4441/**
4442 * dev_change_net_namespace - move device to different nethost namespace
4443 * @dev: device
4444 * @net: network namespace
4445 * @pat: If not NULL name pattern to try if the current device name
4446 * is already taken in the destination network namespace.
4447 *
4448 * This function shuts down a device interface and moves it
4449 * to a new network namespace. On success 0 is returned, on
4450 * a failure a netagive errno code is returned.
4451 *
4452 * Callers must hold the rtnl semaphore.
4453 */
4454
4455int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4456{
4457 char buf[IFNAMSIZ];
4458 const char *destname;
4459 int err;
4460
4461 ASSERT_RTNL();
4462
4463 /* Don't allow namespace local devices to be moved. */
4464 err = -EINVAL;
4465 if (dev->features & NETIF_F_NETNS_LOCAL)
4466 goto out;
4467
3891845e
EB
4468#ifdef CONFIG_SYSFS
4469 /* Don't allow real devices to be moved when sysfs
4470 * is enabled.
4471 */
4472 err = -EINVAL;
4473 if (dev->dev.parent)
4474 goto out;
4475#endif
4476
ce286d32
EB
4477 /* Ensure the device has been registrered */
4478 err = -EINVAL;
4479 if (dev->reg_state != NETREG_REGISTERED)
4480 goto out;
4481
4482 /* Get out if there is nothing todo */
4483 err = 0;
878628fb 4484 if (net_eq(dev_net(dev), net))
ce286d32
EB
4485 goto out;
4486
4487 /* Pick the destination device name, and ensure
4488 * we can use it in the destination network namespace.
4489 */
4490 err = -EEXIST;
4491 destname = dev->name;
4492 if (__dev_get_by_name(net, destname)) {
4493 /* We get here if we can't use the current device name */
4494 if (!pat)
4495 goto out;
4496 if (!dev_valid_name(pat))
4497 goto out;
4498 if (strchr(pat, '%')) {
4499 if (__dev_alloc_name(net, pat, buf) < 0)
4500 goto out;
4501 destname = buf;
4502 } else
4503 destname = pat;
4504 if (__dev_get_by_name(net, destname))
4505 goto out;
4506 }
4507
4508 /*
4509 * And now a mini version of register_netdevice unregister_netdevice.
4510 */
4511
4512 /* If device is running close it first. */
9b772652 4513 dev_close(dev);
ce286d32
EB
4514
4515 /* And unlink it from device chain */
4516 err = -ENODEV;
4517 unlist_netdevice(dev);
4518
4519 synchronize_net();
4520
4521 /* Shutdown queueing discipline. */
4522 dev_shutdown(dev);
4523
4524 /* Notify protocols, that we are about to destroy
4525 this device. They should clean all the things.
4526 */
4527 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4528
4529 /*
4530 * Flush the unicast and multicast chains
4531 */
4532 dev_addr_discard(dev);
4533
3891845e
EB
4534 netdev_unregister_kobject(dev);
4535
ce286d32 4536 /* Actually switch the network namespace */
c346dca1 4537 dev_net_set(dev, net);
ce286d32
EB
4538
4539 /* Assign the new device name */
4540 if (destname != dev->name)
4541 strcpy(dev->name, destname);
4542
4543 /* If there is an ifindex conflict assign a new one */
4544 if (__dev_get_by_index(net, dev->ifindex)) {
4545 int iflink = (dev->iflink == dev->ifindex);
4546 dev->ifindex = dev_new_index(net);
4547 if (iflink)
4548 dev->iflink = dev->ifindex;
4549 }
4550
8b41d188 4551 /* Fixup kobjects */
aaf8cdc3 4552 err = netdev_register_kobject(dev);
8b41d188 4553 WARN_ON(err);
ce286d32
EB
4554
4555 /* Add the device back in the hashes */
4556 list_netdevice(dev);
4557
4558 /* Notify protocols, that a new device appeared. */
4559 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4560
4561 synchronize_net();
4562 err = 0;
4563out:
4564 return err;
4565}
4566
1da177e4
LT
4567static int dev_cpu_callback(struct notifier_block *nfb,
4568 unsigned long action,
4569 void *ocpu)
4570{
4571 struct sk_buff **list_skb;
37437bb2 4572 struct Qdisc **list_net;
1da177e4
LT
4573 struct sk_buff *skb;
4574 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4575 struct softnet_data *sd, *oldsd;
4576
8bb78442 4577 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
1da177e4
LT
4578 return NOTIFY_OK;
4579
4580 local_irq_disable();
4581 cpu = smp_processor_id();
4582 sd = &per_cpu(softnet_data, cpu);
4583 oldsd = &per_cpu(softnet_data, oldcpu);
4584
4585 /* Find end of our completion_queue. */
4586 list_skb = &sd->completion_queue;
4587 while (*list_skb)
4588 list_skb = &(*list_skb)->next;
4589 /* Append completion queue from offline CPU. */
4590 *list_skb = oldsd->completion_queue;
4591 oldsd->completion_queue = NULL;
4592
4593 /* Find end of our output_queue. */
4594 list_net = &sd->output_queue;
4595 while (*list_net)
4596 list_net = &(*list_net)->next_sched;
4597 /* Append output queue from offline CPU. */
4598 *list_net = oldsd->output_queue;
4599 oldsd->output_queue = NULL;
4600
4601 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4602 local_irq_enable();
4603
4604 /* Process offline CPU's input_pkt_queue */
4605 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4606 netif_rx(skb);
4607
4608 return NOTIFY_OK;
4609}
1da177e4 4610
db217334
CL
4611#ifdef CONFIG_NET_DMA
4612/**
0ed72ec4
RD
4613 * net_dma_rebalance - try to maintain one DMA channel per CPU
4614 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4615 *
4616 * This is called when the number of channels allocated to the net_dma client
4617 * changes. The net_dma client tries to have one DMA channel per CPU.
db217334 4618 */
d379b01e
DW
4619
4620static void net_dma_rebalance(struct net_dma *net_dma)
db217334 4621{
d379b01e 4622 unsigned int cpu, i, n, chan_idx;
db217334
CL
4623 struct dma_chan *chan;
4624
d379b01e 4625 if (cpus_empty(net_dma->channel_mask)) {
db217334 4626 for_each_online_cpu(cpu)
29bbd72d 4627 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
db217334
CL
4628 return;
4629 }
4630
4631 i = 0;
4632 cpu = first_cpu(cpu_online_map);
4633
0e12f848 4634 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
d379b01e
DW
4635 chan = net_dma->channels[chan_idx];
4636
4637 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4638 + (i < (num_online_cpus() %
4639 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
db217334
CL
4640
4641 while(n) {
29bbd72d 4642 per_cpu(softnet_data, cpu).net_dma = chan;
db217334
CL
4643 cpu = next_cpu(cpu, cpu_online_map);
4644 n--;
4645 }
4646 i++;
4647 }
db217334
CL
4648}
4649
4650/**
4651 * netdev_dma_event - event callback for the net_dma_client
4652 * @client: should always be net_dma_client
f4b8ea78 4653 * @chan: DMA channel for the event
0ed72ec4 4654 * @state: DMA state to be handled
db217334 4655 */
d379b01e
DW
4656static enum dma_state_client
4657netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4658 enum dma_state state)
4659{
4660 int i, found = 0, pos = -1;
4661 struct net_dma *net_dma =
4662 container_of(client, struct net_dma, client);
4663 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4664
4665 spin_lock(&net_dma->lock);
4666 switch (state) {
4667 case DMA_RESOURCE_AVAILABLE:
0c0b0aca 4668 for (i = 0; i < nr_cpu_ids; i++)
d379b01e
DW
4669 if (net_dma->channels[i] == chan) {
4670 found = 1;
4671 break;
4672 } else if (net_dma->channels[i] == NULL && pos < 0)
4673 pos = i;
4674
4675 if (!found && pos >= 0) {
4676 ack = DMA_ACK;
4677 net_dma->channels[pos] = chan;
4678 cpu_set(pos, net_dma->channel_mask);
4679 net_dma_rebalance(net_dma);
4680 }
db217334
CL
4681 break;
4682 case DMA_RESOURCE_REMOVED:
0c0b0aca 4683 for (i = 0; i < nr_cpu_ids; i++)
d379b01e
DW
4684 if (net_dma->channels[i] == chan) {
4685 found = 1;
4686 pos = i;
4687 break;
4688 }
4689
4690 if (found) {
4691 ack = DMA_ACK;
4692 cpu_clear(pos, net_dma->channel_mask);
4693 net_dma->channels[i] = NULL;
4694 net_dma_rebalance(net_dma);
4695 }
db217334
CL
4696 break;
4697 default:
4698 break;
4699 }
d379b01e
DW
4700 spin_unlock(&net_dma->lock);
4701
4702 return ack;
db217334
CL
4703}
4704
4705/**
f0db275a 4706 * netdev_dma_register - register the networking subsystem as a DMA client
db217334
CL
4707 */
4708static int __init netdev_dma_register(void)
4709{
0c0b0aca
MT
4710 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4711 GFP_KERNEL);
4712 if (unlikely(!net_dma.channels)) {
4713 printk(KERN_NOTICE
4714 "netdev_dma: no memory for net_dma.channels\n");
4715 return -ENOMEM;
4716 }
d379b01e
DW
4717 spin_lock_init(&net_dma.lock);
4718 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4719 dma_async_client_register(&net_dma.client);
4720 dma_async_client_chan_request(&net_dma.client);
db217334
CL
4721 return 0;
4722}
4723
4724#else
4725static int __init netdev_dma_register(void) { return -ENODEV; }
4726#endif /* CONFIG_NET_DMA */
1da177e4 4727
7f353bf2 4728/**
b63365a2
HX
4729 * netdev_increment_features - increment feature set by one
4730 * @all: current feature set
4731 * @one: new feature set
4732 * @mask: mask feature set
7f353bf2
HX
4733 *
4734 * Computes a new feature set after adding a device with feature set
b63365a2
HX
4735 * @one to the master device with current feature set @all. Will not
4736 * enable anything that is off in @mask. Returns the new feature set.
7f353bf2 4737 */
b63365a2
HX
4738unsigned long netdev_increment_features(unsigned long all, unsigned long one,
4739 unsigned long mask)
4740{
4741 /* If device needs checksumming, downgrade to it. */
4742 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4743 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
4744 else if (mask & NETIF_F_ALL_CSUM) {
4745 /* If one device supports v4/v6 checksumming, set for all. */
4746 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
4747 !(all & NETIF_F_GEN_CSUM)) {
4748 all &= ~NETIF_F_ALL_CSUM;
4749 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
4750 }
e2a6b852 4751
b63365a2
HX
4752 /* If one device supports hw checksumming, set for all. */
4753 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
4754 all &= ~NETIF_F_ALL_CSUM;
4755 all |= NETIF_F_HW_CSUM;
4756 }
4757 }
7f353bf2 4758
b63365a2 4759 one |= NETIF_F_ALL_CSUM;
7f353bf2 4760
b63365a2
HX
4761 one |= all & NETIF_F_ONE_FOR_ALL;
4762 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
4763 all |= one & mask & NETIF_F_ONE_FOR_ALL;
7f353bf2
HX
4764
4765 return all;
4766}
b63365a2 4767EXPORT_SYMBOL(netdev_increment_features);
7f353bf2 4768
30d97d35
PE
4769static struct hlist_head *netdev_create_hash(void)
4770{
4771 int i;
4772 struct hlist_head *hash;
4773
4774 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4775 if (hash != NULL)
4776 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4777 INIT_HLIST_HEAD(&hash[i]);
4778
4779 return hash;
4780}
4781
881d966b 4782/* Initialize per network namespace state */
4665079c 4783static int __net_init netdev_init(struct net *net)
881d966b 4784{
881d966b 4785 INIT_LIST_HEAD(&net->dev_base_head);
881d966b 4786
30d97d35
PE
4787 net->dev_name_head = netdev_create_hash();
4788 if (net->dev_name_head == NULL)
4789 goto err_name;
881d966b 4790
30d97d35
PE
4791 net->dev_index_head = netdev_create_hash();
4792 if (net->dev_index_head == NULL)
4793 goto err_idx;
881d966b
EB
4794
4795 return 0;
30d97d35
PE
4796
4797err_idx:
4798 kfree(net->dev_name_head);
4799err_name:
4800 return -ENOMEM;
881d966b
EB
4801}
4802
f0db275a
SH
4803/**
4804 * netdev_drivername - network driver for the device
4805 * @dev: network device
4806 * @buffer: buffer for resulting name
4807 * @len: size of buffer
4808 *
4809 * Determine network driver for device.
4810 */
cf04a4c7 4811char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6579e57b 4812{
cf04a4c7
SH
4813 const struct device_driver *driver;
4814 const struct device *parent;
6579e57b
AV
4815
4816 if (len <= 0 || !buffer)
4817 return buffer;
4818 buffer[0] = 0;
4819
4820 parent = dev->dev.parent;
4821
4822 if (!parent)
4823 return buffer;
4824
4825 driver = parent->driver;
4826 if (driver && driver->name)
4827 strlcpy(buffer, driver->name, len);
4828 return buffer;
4829}
4830
4665079c 4831static void __net_exit netdev_exit(struct net *net)
881d966b
EB
4832{
4833 kfree(net->dev_name_head);
4834 kfree(net->dev_index_head);
4835}
4836
022cbae6 4837static struct pernet_operations __net_initdata netdev_net_ops = {
881d966b
EB
4838 .init = netdev_init,
4839 .exit = netdev_exit,
4840};
4841
4665079c 4842static void __net_exit default_device_exit(struct net *net)
ce286d32
EB
4843{
4844 struct net_device *dev, *next;
4845 /*
4846 * Push all migratable of the network devices back to the
4847 * initial network namespace
4848 */
4849 rtnl_lock();
4850 for_each_netdev_safe(net, dev, next) {
4851 int err;
aca51397 4852 char fb_name[IFNAMSIZ];
ce286d32
EB
4853
4854 /* Ignore unmoveable devices (i.e. loopback) */
4855 if (dev->features & NETIF_F_NETNS_LOCAL)
4856 continue;
4857
4858 /* Push remaing network devices to init_net */
aca51397
PE
4859 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4860 err = dev_change_net_namespace(dev, &init_net, fb_name);
ce286d32 4861 if (err) {
aca51397 4862 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
ce286d32 4863 __func__, dev->name, err);
aca51397 4864 BUG();
ce286d32
EB
4865 }
4866 }
4867 rtnl_unlock();
4868}
4869
022cbae6 4870static struct pernet_operations __net_initdata default_device_ops = {
ce286d32
EB
4871 .exit = default_device_exit,
4872};
4873
1da177e4
LT
4874/*
4875 * Initialize the DEV module. At boot time this walks the device list and
4876 * unhooks any devices that fail to initialise (normally hardware not
4877 * present) and leaves us with a valid list of present and active devices.
4878 *
4879 */
4880
4881/*
4882 * This is called single threaded during boot, so no need
4883 * to take the rtnl semaphore.
4884 */
4885static int __init net_dev_init(void)
4886{
4887 int i, rc = -ENOMEM;
4888
4889 BUG_ON(!dev_boot_phase);
4890
1da177e4
LT
4891 if (dev_proc_init())
4892 goto out;
4893
8b41d188 4894 if (netdev_kobject_init())
1da177e4
LT
4895 goto out;
4896
4897 INIT_LIST_HEAD(&ptype_all);
82d8a867 4898 for (i = 0; i < PTYPE_HASH_SIZE; i++)
1da177e4
LT
4899 INIT_LIST_HEAD(&ptype_base[i]);
4900
881d966b
EB
4901 if (register_pernet_subsys(&netdev_net_ops))
4902 goto out;
1da177e4 4903
ce286d32
EB
4904 if (register_pernet_device(&default_device_ops))
4905 goto out;
4906
1da177e4
LT
4907 /*
4908 * Initialise the packet receive queues.
4909 */
4910
6f912042 4911 for_each_possible_cpu(i) {
1da177e4
LT
4912 struct softnet_data *queue;
4913
4914 queue = &per_cpu(softnet_data, i);
4915 skb_queue_head_init(&queue->input_pkt_queue);
1da177e4
LT
4916 queue->completion_queue = NULL;
4917 INIT_LIST_HEAD(&queue->poll_list);
bea3348e
SH
4918
4919 queue->backlog.poll = process_backlog;
4920 queue->backlog.weight = weight_p;
1da177e4
LT
4921 }
4922
db217334
CL
4923 netdev_dma_register();
4924
1da177e4
LT
4925 dev_boot_phase = 0;
4926
962cf36c
CM
4927 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4928 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
1da177e4
LT
4929
4930 hotcpu_notifier(dev_cpu_callback, 0);
4931 dst_init();
4932 dev_mcast_init();
4933 rc = 0;
4934out:
4935 return rc;
4936}
4937
4938subsys_initcall(net_dev_init);
4939
4940EXPORT_SYMBOL(__dev_get_by_index);
4941EXPORT_SYMBOL(__dev_get_by_name);
4942EXPORT_SYMBOL(__dev_remove_pack);
c2373ee9 4943EXPORT_SYMBOL(dev_valid_name);
1da177e4
LT
4944EXPORT_SYMBOL(dev_add_pack);
4945EXPORT_SYMBOL(dev_alloc_name);
4946EXPORT_SYMBOL(dev_close);
4947EXPORT_SYMBOL(dev_get_by_flags);
4948EXPORT_SYMBOL(dev_get_by_index);
4949EXPORT_SYMBOL(dev_get_by_name);
1da177e4
LT
4950EXPORT_SYMBOL(dev_open);
4951EXPORT_SYMBOL(dev_queue_xmit);
4952EXPORT_SYMBOL(dev_remove_pack);
4953EXPORT_SYMBOL(dev_set_allmulti);
4954EXPORT_SYMBOL(dev_set_promiscuity);
4955EXPORT_SYMBOL(dev_change_flags);
4956EXPORT_SYMBOL(dev_set_mtu);
4957EXPORT_SYMBOL(dev_set_mac_address);
4958EXPORT_SYMBOL(free_netdev);
4959EXPORT_SYMBOL(netdev_boot_setup_check);
4960EXPORT_SYMBOL(netdev_set_master);
4961EXPORT_SYMBOL(netdev_state_change);
4962EXPORT_SYMBOL(netif_receive_skb);
4963EXPORT_SYMBOL(netif_rx);
4964EXPORT_SYMBOL(register_gifconf);
4965EXPORT_SYMBOL(register_netdevice);
4966EXPORT_SYMBOL(register_netdevice_notifier);
4967EXPORT_SYMBOL(skb_checksum_help);
4968EXPORT_SYMBOL(synchronize_net);
4969EXPORT_SYMBOL(unregister_netdevice);
4970EXPORT_SYMBOL(unregister_netdevice_notifier);
4971EXPORT_SYMBOL(net_enable_timestamp);
4972EXPORT_SYMBOL(net_disable_timestamp);
4973EXPORT_SYMBOL(dev_get_flags);
4974
4975#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4976EXPORT_SYMBOL(br_handle_frame_hook);
4977EXPORT_SYMBOL(br_fdb_get_hook);
4978EXPORT_SYMBOL(br_fdb_put_hook);
4979#endif
4980
1da177e4 4981EXPORT_SYMBOL(dev_load);
1da177e4
LT
4982
4983EXPORT_PER_CPU_SYMBOL(softnet_data);