decnet: Use data ready call back, rather than hand coding it
[GitHub/mt8127/android_kernel_alcatel_ttab.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>
1da177e4
LT
111#include <linux/netpoll.h>
112#include <linux/rcupdate.h>
113#include <linux/delay.h>
295f4a1f 114#include <net/wext.h>
1da177e4 115#include <net/iw_handler.h>
1da177e4 116#include <asm/current.h>
5bdb9886 117#include <linux/audit.h>
db217334 118#include <linux/dmaengine.h>
f6a78bfc 119#include <linux/err.h>
c7fa9d18 120#include <linux/ctype.h>
723e98b7 121#include <linux/if_arp.h>
6de329e2 122#include <linux/if_vlan.h>
8f0f2223 123#include <linux/ip.h>
ad55dcaf 124#include <net/ip.h>
8f0f2223
DM
125#include <linux/ipv6.h>
126#include <linux/in.h>
b6b2fed1
DM
127#include <linux/jhash.h>
128#include <linux/random.h>
1da177e4 129
342709ef
PE
130#include "net-sysfs.h"
131
d565b0a1
HX
132/* Instead of increasing this, you should create a hash table. */
133#define MAX_GRO_SKBS 8
134
5d38a079
HX
135/* This should be increased if a protocol with a bigger head is added. */
136#define GRO_MAX_HEAD (MAX_HEADER + 128)
137
1da177e4
LT
138/*
139 * The list of packet types we will receive (as opposed to discard)
140 * and the routines to invoke.
141 *
142 * Why 16. Because with 16 the only overlap we get on a hash of the
143 * low nibble of the protocol value is RARP/SNAP/X.25.
144 *
145 * NOTE: That is no longer true with the addition of VLAN tags. Not
146 * sure which should go first, but I bet it won't make much
147 * difference if we are running VLANs. The good news is that
148 * this protocol won't be in the list unless compiled in, so
3041a069 149 * the average user (w/out VLANs) will not be adversely affected.
1da177e4
LT
150 * --BLG
151 *
152 * 0800 IP
153 * 8100 802.1Q VLAN
154 * 0001 802.3
155 * 0002 AX.25
156 * 0004 802.2
157 * 8035 RARP
158 * 0005 SNAP
159 * 0805 X.25
160 * 0806 ARP
161 * 8137 IPX
162 * 0009 Localtalk
163 * 86DD IPv6
164 */
165
82d8a867
PE
166#define PTYPE_HASH_SIZE (16)
167#define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
168
1da177e4 169static DEFINE_SPINLOCK(ptype_lock);
82d8a867 170static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
6b2bedc3 171static struct list_head ptype_all __read_mostly; /* Taps */
1da177e4 172
1da177e4 173/*
7562f876 174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
1da177e4
LT
175 * semaphore.
176 *
177 * Pure readers hold dev_base_lock for reading.
178 *
179 * Writers must hold the rtnl semaphore while they loop through the
7562f876 180 * dev_base_head list, and hold dev_base_lock for writing when they do the
1da177e4
LT
181 * actual updates. This allows pure readers to access the list even
182 * while a writer is preparing to update it.
183 *
184 * To put it another way, dev_base_lock is held for writing only to
185 * protect against pure readers; the rtnl semaphore provides the
186 * protection against other writers.
187 *
188 * See, for example usages, register_netdevice() and
189 * unregister_netdevice(), which must be called with the rtnl
190 * semaphore held.
191 */
1da177e4
LT
192DEFINE_RWLOCK(dev_base_lock);
193
1da177e4
LT
194EXPORT_SYMBOL(dev_base_lock);
195
196#define NETDEV_HASHBITS 8
881d966b 197#define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
1da177e4 198
881d966b 199static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
1da177e4
LT
200{
201 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
881d966b 202 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
1da177e4
LT
203}
204
881d966b 205static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
1da177e4 206{
881d966b 207 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
1da177e4
LT
208}
209
ce286d32
EB
210/* Device list insertion */
211static int list_netdevice(struct net_device *dev)
212{
c346dca1 213 struct net *net = dev_net(dev);
ce286d32
EB
214
215 ASSERT_RTNL();
216
217 write_lock_bh(&dev_base_lock);
218 list_add_tail(&dev->dev_list, &net->dev_base_head);
219 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
220 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
221 write_unlock_bh(&dev_base_lock);
222 return 0;
223}
224
225/* Device list removal */
226static void unlist_netdevice(struct net_device *dev)
227{
228 ASSERT_RTNL();
229
230 /* Unlink dev from the device chain */
231 write_lock_bh(&dev_base_lock);
232 list_del(&dev->dev_list);
233 hlist_del(&dev->name_hlist);
234 hlist_del(&dev->index_hlist);
235 write_unlock_bh(&dev_base_lock);
236}
237
1da177e4
LT
238/*
239 * Our notifier list
240 */
241
f07d5b94 242static RAW_NOTIFIER_HEAD(netdev_chain);
1da177e4
LT
243
244/*
245 * Device drivers call our routines to queue packets here. We empty the
246 * queue in the local softnet handler.
247 */
bea3348e
SH
248
249DEFINE_PER_CPU(struct softnet_data, softnet_data);
1da177e4 250
cf508b12 251#ifdef CONFIG_LOCKDEP
723e98b7 252/*
c773e847 253 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
723e98b7
JP
254 * according to dev->type
255 */
256static const unsigned short netdev_lock_type[] =
257 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
258 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
259 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
260 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
261 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
262 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
263 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
264 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
265 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
266 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
267 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
268 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
269 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
2d91d78b 270 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
57c81fff 271 ARPHRD_PHONET_PIPE, ARPHRD_VOID, ARPHRD_NONE};
723e98b7
JP
272
273static const char *netdev_lock_name[] =
274 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
275 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
276 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
277 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
278 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
279 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
280 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
281 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
282 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
283 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
284 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
285 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
286 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
2d91d78b 287 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
57c81fff 288 "_xmit_PHONET_PIPE", "_xmit_VOID", "_xmit_NONE"};
723e98b7
JP
289
290static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
cf508b12 291static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
723e98b7
JP
292
293static inline unsigned short netdev_lock_pos(unsigned short dev_type)
294{
295 int i;
296
297 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
298 if (netdev_lock_type[i] == dev_type)
299 return i;
300 /* the last key is used by default */
301 return ARRAY_SIZE(netdev_lock_type) - 1;
302}
303
cf508b12
DM
304static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
305 unsigned short dev_type)
723e98b7
JP
306{
307 int i;
308
309 i = netdev_lock_pos(dev_type);
310 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
311 netdev_lock_name[i]);
312}
cf508b12
DM
313
314static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
315{
316 int i;
317
318 i = netdev_lock_pos(dev->type);
319 lockdep_set_class_and_name(&dev->addr_list_lock,
320 &netdev_addr_lock_key[i],
321 netdev_lock_name[i]);
322}
723e98b7 323#else
cf508b12
DM
324static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
326{
327}
328static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
723e98b7
JP
329{
330}
331#endif
1da177e4
LT
332
333/*******************************************************************************
334
335 Protocol management and registration routines
336
337*******************************************************************************/
338
1da177e4
LT
339/*
340 * Add a protocol ID to the list. Now that the input handler is
341 * smarter we can dispense with all the messy stuff that used to be
342 * here.
343 *
344 * BEWARE!!! Protocol handlers, mangling input packets,
345 * MUST BE last in hash buckets and checking protocol handlers
346 * MUST start from promiscuous ptype_all chain in net_bh.
347 * It is true now, do not change it.
348 * Explanation follows: if protocol handler, mangling packet, will
349 * be the first on list, it is not able to sense, that packet
350 * is cloned and should be copied-on-write, so that it will
351 * change it and subsequent readers will get broken packet.
352 * --ANK (980803)
353 */
354
355/**
356 * dev_add_pack - add packet handler
357 * @pt: packet type declaration
358 *
359 * Add a protocol handler to the networking stack. The passed &packet_type
360 * is linked into kernel lists and may not be freed until it has been
361 * removed from the kernel lists.
362 *
4ec93edb 363 * This call does not sleep therefore it can not
1da177e4
LT
364 * guarantee all CPU's that are in middle of receiving packets
365 * will see the new packet type (until the next received packet).
366 */
367
368void dev_add_pack(struct packet_type *pt)
369{
370 int hash;
371
372 spin_lock_bh(&ptype_lock);
9be9a6b9 373 if (pt->type == htons(ETH_P_ALL))
1da177e4 374 list_add_rcu(&pt->list, &ptype_all);
9be9a6b9 375 else {
82d8a867 376 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
1da177e4
LT
377 list_add_rcu(&pt->list, &ptype_base[hash]);
378 }
379 spin_unlock_bh(&ptype_lock);
380}
381
1da177e4
LT
382/**
383 * __dev_remove_pack - remove packet handler
384 * @pt: packet type declaration
385 *
386 * Remove a protocol handler that was previously added to the kernel
387 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
388 * from the kernel lists and can be freed or reused once this function
4ec93edb 389 * returns.
1da177e4
LT
390 *
391 * The packet type might still be in use by receivers
392 * and must not be freed until after all the CPU's have gone
393 * through a quiescent state.
394 */
395void __dev_remove_pack(struct packet_type *pt)
396{
397 struct list_head *head;
398 struct packet_type *pt1;
399
400 spin_lock_bh(&ptype_lock);
401
9be9a6b9 402 if (pt->type == htons(ETH_P_ALL))
1da177e4 403 head = &ptype_all;
9be9a6b9 404 else
82d8a867 405 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
1da177e4
LT
406
407 list_for_each_entry(pt1, head, list) {
408 if (pt == pt1) {
409 list_del_rcu(&pt->list);
410 goto out;
411 }
412 }
413
414 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
415out:
416 spin_unlock_bh(&ptype_lock);
417}
418/**
419 * dev_remove_pack - remove packet handler
420 * @pt: packet type declaration
421 *
422 * Remove a protocol handler that was previously added to the kernel
423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
424 * from the kernel lists and can be freed or reused once this function
425 * returns.
426 *
427 * This call sleeps to guarantee that no CPU is looking at the packet
428 * type after return.
429 */
430void dev_remove_pack(struct packet_type *pt)
431{
432 __dev_remove_pack(pt);
4ec93edb 433
1da177e4
LT
434 synchronize_net();
435}
436
437/******************************************************************************
438
439 Device Boot-time Settings Routines
440
441*******************************************************************************/
442
443/* Boot time configuration table */
444static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
445
446/**
447 * netdev_boot_setup_add - add new setup entry
448 * @name: name of the device
449 * @map: configured settings for the device
450 *
451 * Adds new setup entry to the dev_boot_setup list. The function
452 * returns 0 on error and 1 on success. This is a generic routine to
453 * all netdevices.
454 */
455static int netdev_boot_setup_add(char *name, struct ifmap *map)
456{
457 struct netdev_boot_setup *s;
458 int i;
459
460 s = dev_boot_setup;
461 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
462 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
463 memset(s[i].name, 0, sizeof(s[i].name));
93b3cff9 464 strlcpy(s[i].name, name, IFNAMSIZ);
1da177e4
LT
465 memcpy(&s[i].map, map, sizeof(s[i].map));
466 break;
467 }
468 }
469
470 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
471}
472
473/**
474 * netdev_boot_setup_check - check boot time settings
475 * @dev: the netdevice
476 *
477 * Check boot time settings for the device.
478 * The found settings are set for the device to be used
479 * later in the device probing.
480 * Returns 0 if no settings found, 1 if they are.
481 */
482int netdev_boot_setup_check(struct net_device *dev)
483{
484 struct netdev_boot_setup *s = dev_boot_setup;
485 int i;
486
487 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
488 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
93b3cff9 489 !strcmp(dev->name, s[i].name)) {
1da177e4
LT
490 dev->irq = s[i].map.irq;
491 dev->base_addr = s[i].map.base_addr;
492 dev->mem_start = s[i].map.mem_start;
493 dev->mem_end = s[i].map.mem_end;
494 return 1;
495 }
496 }
497 return 0;
498}
499
500
501/**
502 * netdev_boot_base - get address from boot time settings
503 * @prefix: prefix for network device
504 * @unit: id for network device
505 *
506 * Check boot time settings for the base address of device.
507 * The found settings are set for the device to be used
508 * later in the device probing.
509 * Returns 0 if no settings found.
510 */
511unsigned long netdev_boot_base(const char *prefix, int unit)
512{
513 const struct netdev_boot_setup *s = dev_boot_setup;
514 char name[IFNAMSIZ];
515 int i;
516
517 sprintf(name, "%s%d", prefix, unit);
518
519 /*
520 * If device already registered then return base of 1
521 * to indicate not to probe for this interface
522 */
881d966b 523 if (__dev_get_by_name(&init_net, name))
1da177e4
LT
524 return 1;
525
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
527 if (!strcmp(name, s[i].name))
528 return s[i].map.base_addr;
529 return 0;
530}
531
532/*
533 * Saves at boot time configured settings for any netdevice.
534 */
535int __init netdev_boot_setup(char *str)
536{
537 int ints[5];
538 struct ifmap map;
539
540 str = get_options(str, ARRAY_SIZE(ints), ints);
541 if (!str || !*str)
542 return 0;
543
544 /* Save settings */
545 memset(&map, 0, sizeof(map));
546 if (ints[0] > 0)
547 map.irq = ints[1];
548 if (ints[0] > 1)
549 map.base_addr = ints[2];
550 if (ints[0] > 2)
551 map.mem_start = ints[3];
552 if (ints[0] > 3)
553 map.mem_end = ints[4];
554
555 /* Add new entry to the list */
556 return netdev_boot_setup_add(str, &map);
557}
558
559__setup("netdev=", netdev_boot_setup);
560
561/*******************************************************************************
562
563 Device Interface Subroutines
564
565*******************************************************************************/
566
567/**
568 * __dev_get_by_name - find a device by its name
c4ea43c5 569 * @net: the applicable net namespace
1da177e4
LT
570 * @name: name to find
571 *
572 * Find an interface by name. Must be called under RTNL semaphore
573 * or @dev_base_lock. If the name is found a pointer to the device
574 * is returned. If the name is not found then %NULL is returned. The
575 * reference counters are not incremented so the caller must be
576 * careful with locks.
577 */
578
881d966b 579struct net_device *__dev_get_by_name(struct net *net, const char *name)
1da177e4
LT
580{
581 struct hlist_node *p;
582
881d966b 583 hlist_for_each(p, dev_name_hash(net, name)) {
1da177e4
LT
584 struct net_device *dev
585 = hlist_entry(p, struct net_device, name_hlist);
586 if (!strncmp(dev->name, name, IFNAMSIZ))
587 return dev;
588 }
589 return NULL;
590}
591
592/**
593 * dev_get_by_name - find a device by its name
c4ea43c5 594 * @net: the applicable net namespace
1da177e4
LT
595 * @name: name to find
596 *
597 * Find an interface by name. This can be called from any
598 * context and does its own locking. The returned handle has
599 * the usage count incremented and the caller must use dev_put() to
600 * release it when it is no longer needed. %NULL is returned if no
601 * matching device is found.
602 */
603
881d966b 604struct net_device *dev_get_by_name(struct net *net, const char *name)
1da177e4
LT
605{
606 struct net_device *dev;
607
608 read_lock(&dev_base_lock);
881d966b 609 dev = __dev_get_by_name(net, name);
1da177e4
LT
610 if (dev)
611 dev_hold(dev);
612 read_unlock(&dev_base_lock);
613 return dev;
614}
615
616/**
617 * __dev_get_by_index - find a device by its ifindex
c4ea43c5 618 * @net: the applicable net namespace
1da177e4
LT
619 * @ifindex: index of device
620 *
621 * Search for an interface by index. Returns %NULL if the device
622 * is not found or a pointer to the device. The device has not
623 * had its reference counter increased so the caller must be careful
624 * about locking. The caller must hold either the RTNL semaphore
625 * or @dev_base_lock.
626 */
627
881d966b 628struct net_device *__dev_get_by_index(struct net *net, int ifindex)
1da177e4
LT
629{
630 struct hlist_node *p;
631
881d966b 632 hlist_for_each(p, dev_index_hash(net, ifindex)) {
1da177e4
LT
633 struct net_device *dev
634 = hlist_entry(p, struct net_device, index_hlist);
635 if (dev->ifindex == ifindex)
636 return dev;
637 }
638 return NULL;
639}
640
641
642/**
643 * dev_get_by_index - find a device by its ifindex
c4ea43c5 644 * @net: the applicable net namespace
1da177e4
LT
645 * @ifindex: index of device
646 *
647 * Search for an interface by index. Returns NULL if the device
648 * is not found or a pointer to the device. The device returned has
649 * had a reference added and the pointer is safe until the user calls
650 * dev_put to indicate they have finished with it.
651 */
652
881d966b 653struct net_device *dev_get_by_index(struct net *net, int ifindex)
1da177e4
LT
654{
655 struct net_device *dev;
656
657 read_lock(&dev_base_lock);
881d966b 658 dev = __dev_get_by_index(net, ifindex);
1da177e4
LT
659 if (dev)
660 dev_hold(dev);
661 read_unlock(&dev_base_lock);
662 return dev;
663}
664
665/**
666 * dev_getbyhwaddr - find a device by its hardware address
c4ea43c5 667 * @net: the applicable net namespace
1da177e4
LT
668 * @type: media type of device
669 * @ha: hardware address
670 *
671 * Search for an interface by MAC address. Returns NULL if the device
672 * is not found or a pointer to the device. The caller must hold the
673 * rtnl semaphore. The returned device has not had its ref count increased
674 * and the caller must therefore be careful about locking
675 *
676 * BUGS:
677 * If the API was consistent this would be __dev_get_by_hwaddr
678 */
679
881d966b 680struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
1da177e4
LT
681{
682 struct net_device *dev;
683
684 ASSERT_RTNL();
685
81103a52 686 for_each_netdev(net, dev)
1da177e4
LT
687 if (dev->type == type &&
688 !memcmp(dev->dev_addr, ha, dev->addr_len))
7562f876
PE
689 return dev;
690
691 return NULL;
1da177e4
LT
692}
693
cf309e3f
JF
694EXPORT_SYMBOL(dev_getbyhwaddr);
695
881d966b 696struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
1da177e4
LT
697{
698 struct net_device *dev;
699
4e9cac2b 700 ASSERT_RTNL();
881d966b 701 for_each_netdev(net, dev)
4e9cac2b 702 if (dev->type == type)
7562f876
PE
703 return dev;
704
705 return NULL;
4e9cac2b
PM
706}
707
708EXPORT_SYMBOL(__dev_getfirstbyhwtype);
709
881d966b 710struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
4e9cac2b
PM
711{
712 struct net_device *dev;
713
714 rtnl_lock();
881d966b 715 dev = __dev_getfirstbyhwtype(net, type);
4e9cac2b
PM
716 if (dev)
717 dev_hold(dev);
1da177e4
LT
718 rtnl_unlock();
719 return dev;
720}
721
722EXPORT_SYMBOL(dev_getfirstbyhwtype);
723
724/**
725 * dev_get_by_flags - find any device with given flags
c4ea43c5 726 * @net: the applicable net namespace
1da177e4
LT
727 * @if_flags: IFF_* values
728 * @mask: bitmask of bits in if_flags to check
729 *
730 * Search for any interface with the given flags. Returns NULL if a device
4ec93edb 731 * is not found or a pointer to the device. The device returned has
1da177e4
LT
732 * had a reference added and the pointer is safe until the user calls
733 * dev_put to indicate they have finished with it.
734 */
735
881d966b 736struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
1da177e4 737{
7562f876 738 struct net_device *dev, *ret;
1da177e4 739
7562f876 740 ret = NULL;
1da177e4 741 read_lock(&dev_base_lock);
881d966b 742 for_each_netdev(net, dev) {
1da177e4
LT
743 if (((dev->flags ^ if_flags) & mask) == 0) {
744 dev_hold(dev);
7562f876 745 ret = dev;
1da177e4
LT
746 break;
747 }
748 }
749 read_unlock(&dev_base_lock);
7562f876 750 return ret;
1da177e4
LT
751}
752
753/**
754 * dev_valid_name - check if name is okay for network device
755 * @name: name string
756 *
757 * Network device names need to be valid file names to
c7fa9d18
DM
758 * to allow sysfs to work. We also disallow any kind of
759 * whitespace.
1da177e4 760 */
c2373ee9 761int dev_valid_name(const char *name)
1da177e4 762{
c7fa9d18
DM
763 if (*name == '\0')
764 return 0;
b6fe17d6
SH
765 if (strlen(name) >= IFNAMSIZ)
766 return 0;
c7fa9d18
DM
767 if (!strcmp(name, ".") || !strcmp(name, ".."))
768 return 0;
769
770 while (*name) {
771 if (*name == '/' || isspace(*name))
772 return 0;
773 name++;
774 }
775 return 1;
1da177e4
LT
776}
777
778/**
b267b179
EB
779 * __dev_alloc_name - allocate a name for a device
780 * @net: network namespace to allocate the device name in
1da177e4 781 * @name: name format string
b267b179 782 * @buf: scratch buffer and result name string
1da177e4
LT
783 *
784 * Passed a format string - eg "lt%d" it will try and find a suitable
3041a069
SH
785 * id. It scans list of devices to build up a free map, then chooses
786 * the first empty slot. The caller must hold the dev_base or rtnl lock
787 * while allocating the name and adding the device in order to avoid
788 * duplicates.
789 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
790 * Returns the number of the unit assigned or a negative errno code.
1da177e4
LT
791 */
792
b267b179 793static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1da177e4
LT
794{
795 int i = 0;
1da177e4
LT
796 const char *p;
797 const int max_netdevices = 8*PAGE_SIZE;
cfcabdcc 798 unsigned long *inuse;
1da177e4
LT
799 struct net_device *d;
800
801 p = strnchr(name, IFNAMSIZ-1, '%');
802 if (p) {
803 /*
804 * Verify the string as this thing may have come from
805 * the user. There must be either one "%d" and no other "%"
806 * characters.
807 */
808 if (p[1] != 'd' || strchr(p + 2, '%'))
809 return -EINVAL;
810
811 /* Use one page as a bit array of possible slots */
cfcabdcc 812 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1da177e4
LT
813 if (!inuse)
814 return -ENOMEM;
815
881d966b 816 for_each_netdev(net, d) {
1da177e4
LT
817 if (!sscanf(d->name, name, &i))
818 continue;
819 if (i < 0 || i >= max_netdevices)
820 continue;
821
822 /* avoid cases where sscanf is not exact inverse of printf */
b267b179 823 snprintf(buf, IFNAMSIZ, name, i);
1da177e4
LT
824 if (!strncmp(buf, d->name, IFNAMSIZ))
825 set_bit(i, inuse);
826 }
827
828 i = find_first_zero_bit(inuse, max_netdevices);
829 free_page((unsigned long) inuse);
830 }
831
b267b179
EB
832 snprintf(buf, IFNAMSIZ, name, i);
833 if (!__dev_get_by_name(net, buf))
1da177e4 834 return i;
1da177e4
LT
835
836 /* It is possible to run out of possible slots
837 * when the name is long and there isn't enough space left
838 * for the digits, or if all bits are used.
839 */
840 return -ENFILE;
841}
842
b267b179
EB
843/**
844 * dev_alloc_name - allocate a name for a device
845 * @dev: device
846 * @name: name format string
847 *
848 * Passed a format string - eg "lt%d" it will try and find a suitable
849 * id. It scans list of devices to build up a free map, then chooses
850 * the first empty slot. The caller must hold the dev_base or rtnl lock
851 * while allocating the name and adding the device in order to avoid
852 * duplicates.
853 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
854 * Returns the number of the unit assigned or a negative errno code.
855 */
856
857int dev_alloc_name(struct net_device *dev, const char *name)
858{
859 char buf[IFNAMSIZ];
860 struct net *net;
861 int ret;
862
c346dca1
YH
863 BUG_ON(!dev_net(dev));
864 net = dev_net(dev);
b267b179
EB
865 ret = __dev_alloc_name(net, name, buf);
866 if (ret >= 0)
867 strlcpy(dev->name, buf, IFNAMSIZ);
868 return ret;
869}
870
1da177e4
LT
871
872/**
873 * dev_change_name - change name of a device
874 * @dev: device
875 * @newname: name (or format string) must be at least IFNAMSIZ
876 *
877 * Change name of a device, can pass format strings "eth%d".
878 * for wildcarding.
879 */
cf04a4c7 880int dev_change_name(struct net_device *dev, const char *newname)
1da177e4 881{
fcc5a03a 882 char oldname[IFNAMSIZ];
1da177e4 883 int err = 0;
fcc5a03a 884 int ret;
881d966b 885 struct net *net;
1da177e4
LT
886
887 ASSERT_RTNL();
c346dca1 888 BUG_ON(!dev_net(dev));
1da177e4 889
c346dca1 890 net = dev_net(dev);
1da177e4
LT
891 if (dev->flags & IFF_UP)
892 return -EBUSY;
893
894 if (!dev_valid_name(newname))
895 return -EINVAL;
896
c8d90dca
SH
897 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
898 return 0;
899
fcc5a03a
HX
900 memcpy(oldname, dev->name, IFNAMSIZ);
901
1da177e4
LT
902 if (strchr(newname, '%')) {
903 err = dev_alloc_name(dev, newname);
904 if (err < 0)
905 return err;
1da177e4 906 }
881d966b 907 else if (__dev_get_by_name(net, newname))
1da177e4
LT
908 return -EEXIST;
909 else
910 strlcpy(dev->name, newname, IFNAMSIZ);
911
fcc5a03a 912rollback:
3891845e
EB
913 /* For now only devices in the initial network namespace
914 * are in sysfs.
915 */
916 if (net == &init_net) {
917 ret = device_rename(&dev->dev, dev->name);
918 if (ret) {
919 memcpy(dev->name, oldname, IFNAMSIZ);
920 return ret;
921 }
dcc99773 922 }
7f988eab
HX
923
924 write_lock_bh(&dev_base_lock);
92749821 925 hlist_del(&dev->name_hlist);
881d966b 926 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
7f988eab
HX
927 write_unlock_bh(&dev_base_lock);
928
056925ab 929 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
fcc5a03a
HX
930 ret = notifier_to_errno(ret);
931
932 if (ret) {
933 if (err) {
934 printk(KERN_ERR
935 "%s: name change rollback failed: %d.\n",
936 dev->name, ret);
937 } else {
938 err = ret;
939 memcpy(dev->name, oldname, IFNAMSIZ);
940 goto rollback;
941 }
942 }
1da177e4
LT
943
944 return err;
945}
946
0b815a1a
SH
947/**
948 * dev_set_alias - change ifalias of a device
949 * @dev: device
950 * @alias: name up to IFALIASZ
f0db275a 951 * @len: limit of bytes to copy from info
0b815a1a
SH
952 *
953 * Set ifalias for a device,
954 */
955int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
956{
957 ASSERT_RTNL();
958
959 if (len >= IFALIASZ)
960 return -EINVAL;
961
96ca4a2c
OH
962 if (!len) {
963 if (dev->ifalias) {
964 kfree(dev->ifalias);
965 dev->ifalias = NULL;
966 }
967 return 0;
968 }
969
0b815a1a
SH
970 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
971 if (!dev->ifalias)
972 return -ENOMEM;
973
974 strlcpy(dev->ifalias, alias, len+1);
975 return len;
976}
977
978
d8a33ac4 979/**
3041a069 980 * netdev_features_change - device changes features
d8a33ac4
SH
981 * @dev: device to cause notification
982 *
983 * Called to indicate a device has changed features.
984 */
985void netdev_features_change(struct net_device *dev)
986{
056925ab 987 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
d8a33ac4
SH
988}
989EXPORT_SYMBOL(netdev_features_change);
990
1da177e4
LT
991/**
992 * netdev_state_change - device changes state
993 * @dev: device to cause notification
994 *
995 * Called to indicate a device has changed state. This function calls
996 * the notifier chains for netdev_chain and sends a NEWLINK message
997 * to the routing socket.
998 */
999void netdev_state_change(struct net_device *dev)
1000{
1001 if (dev->flags & IFF_UP) {
056925ab 1002 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1da177e4
LT
1003 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1004 }
1005}
1006
c1da4ac7
OG
1007void netdev_bonding_change(struct net_device *dev)
1008{
1009 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1010}
1011EXPORT_SYMBOL(netdev_bonding_change);
1012
1da177e4
LT
1013/**
1014 * dev_load - load a network module
c4ea43c5 1015 * @net: the applicable net namespace
1da177e4
LT
1016 * @name: name of interface
1017 *
1018 * If a network interface is not present and the process has suitable
1019 * privileges this function loads the module. If module loading is not
1020 * available in this kernel then it becomes a nop.
1021 */
1022
881d966b 1023void dev_load(struct net *net, const char *name)
1da177e4 1024{
4ec93edb 1025 struct net_device *dev;
1da177e4
LT
1026
1027 read_lock(&dev_base_lock);
881d966b 1028 dev = __dev_get_by_name(net, name);
1da177e4
LT
1029 read_unlock(&dev_base_lock);
1030
1031 if (!dev && capable(CAP_SYS_MODULE))
1032 request_module("%s", name);
1033}
1034
1da177e4
LT
1035/**
1036 * dev_open - prepare an interface for use.
1037 * @dev: device to open
1038 *
1039 * Takes a device from down to up state. The device's private open
1040 * function is invoked and then the multicast lists are loaded. Finally
1041 * the device is moved into the up state and a %NETDEV_UP message is
1042 * sent to the netdev notifier chain.
1043 *
1044 * Calling this function on an active interface is a nop. On a failure
1045 * a negative errno code is returned.
1046 */
1047int dev_open(struct net_device *dev)
1048{
d314774c 1049 const struct net_device_ops *ops = dev->netdev_ops;
1da177e4
LT
1050 int ret = 0;
1051
e46b66bc
BH
1052 ASSERT_RTNL();
1053
1da177e4
LT
1054 /*
1055 * Is it already up?
1056 */
1057
1058 if (dev->flags & IFF_UP)
1059 return 0;
1060
1061 /*
1062 * Is it even present?
1063 */
1064 if (!netif_device_present(dev))
1065 return -ENODEV;
1066
1067 /*
1068 * Call device private open method
1069 */
1070 set_bit(__LINK_STATE_START, &dev->state);
bada339b 1071
d314774c
SH
1072 if (ops->ndo_validate_addr)
1073 ret = ops->ndo_validate_addr(dev);
bada339b 1074
d314774c
SH
1075 if (!ret && ops->ndo_open)
1076 ret = ops->ndo_open(dev);
1da177e4 1077
4ec93edb 1078 /*
1da177e4
LT
1079 * If it went open OK then:
1080 */
1081
bada339b
JG
1082 if (ret)
1083 clear_bit(__LINK_STATE_START, &dev->state);
1084 else {
1da177e4
LT
1085 /*
1086 * Set the flags.
1087 */
1088 dev->flags |= IFF_UP;
1089
649274d9
DW
1090 /*
1091 * Enable NET_DMA
1092 */
b4bd07c2 1093 net_dmaengine_get();
649274d9 1094
1da177e4
LT
1095 /*
1096 * Initialize multicasting status
1097 */
4417da66 1098 dev_set_rx_mode(dev);
1da177e4
LT
1099
1100 /*
1101 * Wakeup transmit queue engine
1102 */
1103 dev_activate(dev);
1104
1105 /*
1106 * ... and announce new interface.
1107 */
056925ab 1108 call_netdevice_notifiers(NETDEV_UP, dev);
1da177e4 1109 }
bada339b 1110
1da177e4
LT
1111 return ret;
1112}
1113
1114/**
1115 * dev_close - shutdown an interface.
1116 * @dev: device to shutdown
1117 *
1118 * This function moves an active device into down state. A
1119 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1120 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1121 * chain.
1122 */
1123int dev_close(struct net_device *dev)
1124{
d314774c 1125 const struct net_device_ops *ops = dev->netdev_ops;
e46b66bc
BH
1126 ASSERT_RTNL();
1127
9d5010db
DM
1128 might_sleep();
1129
1da177e4
LT
1130 if (!(dev->flags & IFF_UP))
1131 return 0;
1132
1133 /*
1134 * Tell people we are going down, so that they can
1135 * prepare to death, when device is still operating.
1136 */
056925ab 1137 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1da177e4 1138
1da177e4
LT
1139 clear_bit(__LINK_STATE_START, &dev->state);
1140
1141 /* Synchronize to scheduled poll. We cannot touch poll list,
bea3348e
SH
1142 * it can be even on different cpu. So just clear netif_running().
1143 *
1144 * dev->stop() will invoke napi_disable() on all of it's
1145 * napi_struct instances on this device.
1146 */
1da177e4 1147 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1da177e4 1148
d8b2a4d2
ML
1149 dev_deactivate(dev);
1150
1da177e4
LT
1151 /*
1152 * Call the device specific close. This cannot fail.
1153 * Only if device is UP
1154 *
1155 * We allow it to be called even after a DETACH hot-plug
1156 * event.
1157 */
d314774c
SH
1158 if (ops->ndo_stop)
1159 ops->ndo_stop(dev);
1da177e4
LT
1160
1161 /*
1162 * Device is now down.
1163 */
1164
1165 dev->flags &= ~IFF_UP;
1166
1167 /*
1168 * Tell people we are down
1169 */
056925ab 1170 call_netdevice_notifiers(NETDEV_DOWN, dev);
1da177e4 1171
649274d9
DW
1172 /*
1173 * Shutdown NET_DMA
1174 */
b4bd07c2 1175 net_dmaengine_put();
649274d9 1176
1da177e4
LT
1177 return 0;
1178}
1179
1180
0187bdfb
BH
1181/**
1182 * dev_disable_lro - disable Large Receive Offload on a device
1183 * @dev: device
1184 *
1185 * Disable Large Receive Offload (LRO) on a net device. Must be
1186 * called under RTNL. This is needed if received packets may be
1187 * forwarded to another interface.
1188 */
1189void dev_disable_lro(struct net_device *dev)
1190{
1191 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1192 dev->ethtool_ops->set_flags) {
1193 u32 flags = dev->ethtool_ops->get_flags(dev);
1194 if (flags & ETH_FLAG_LRO) {
1195 flags &= ~ETH_FLAG_LRO;
1196 dev->ethtool_ops->set_flags(dev, flags);
1197 }
1198 }
1199 WARN_ON(dev->features & NETIF_F_LRO);
1200}
1201EXPORT_SYMBOL(dev_disable_lro);
1202
1203
881d966b
EB
1204static int dev_boot_phase = 1;
1205
1da177e4
LT
1206/*
1207 * Device change register/unregister. These are not inline or static
1208 * as we export them to the world.
1209 */
1210
1211/**
1212 * register_netdevice_notifier - register a network notifier block
1213 * @nb: notifier
1214 *
1215 * Register a notifier to be called when network device events occur.
1216 * The notifier passed is linked into the kernel structures and must
1217 * not be reused until it has been unregistered. A negative errno code
1218 * is returned on a failure.
1219 *
1220 * When registered all registration and up events are replayed
4ec93edb 1221 * to the new notifier to allow device to have a race free
1da177e4
LT
1222 * view of the network device list.
1223 */
1224
1225int register_netdevice_notifier(struct notifier_block *nb)
1226{
1227 struct net_device *dev;
fcc5a03a 1228 struct net_device *last;
881d966b 1229 struct net *net;
1da177e4
LT
1230 int err;
1231
1232 rtnl_lock();
f07d5b94 1233 err = raw_notifier_chain_register(&netdev_chain, nb);
fcc5a03a
HX
1234 if (err)
1235 goto unlock;
881d966b
EB
1236 if (dev_boot_phase)
1237 goto unlock;
1238 for_each_net(net) {
1239 for_each_netdev(net, dev) {
1240 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1241 err = notifier_to_errno(err);
1242 if (err)
1243 goto rollback;
1244
1245 if (!(dev->flags & IFF_UP))
1246 continue;
1da177e4 1247
881d966b
EB
1248 nb->notifier_call(nb, NETDEV_UP, dev);
1249 }
1da177e4 1250 }
fcc5a03a
HX
1251
1252unlock:
1da177e4
LT
1253 rtnl_unlock();
1254 return err;
fcc5a03a
HX
1255
1256rollback:
1257 last = dev;
881d966b
EB
1258 for_each_net(net) {
1259 for_each_netdev(net, dev) {
1260 if (dev == last)
1261 break;
fcc5a03a 1262
881d966b
EB
1263 if (dev->flags & IFF_UP) {
1264 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1265 nb->notifier_call(nb, NETDEV_DOWN, dev);
1266 }
1267 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
fcc5a03a 1268 }
fcc5a03a 1269 }
c67625a1
PE
1270
1271 raw_notifier_chain_unregister(&netdev_chain, nb);
fcc5a03a 1272 goto unlock;
1da177e4
LT
1273}
1274
1275/**
1276 * unregister_netdevice_notifier - unregister a network notifier block
1277 * @nb: notifier
1278 *
1279 * Unregister a notifier previously registered by
1280 * register_netdevice_notifier(). The notifier is unlinked into the
1281 * kernel structures and may then be reused. A negative errno code
1282 * is returned on a failure.
1283 */
1284
1285int unregister_netdevice_notifier(struct notifier_block *nb)
1286{
9f514950
HX
1287 int err;
1288
1289 rtnl_lock();
f07d5b94 1290 err = raw_notifier_chain_unregister(&netdev_chain, nb);
9f514950
HX
1291 rtnl_unlock();
1292 return err;
1da177e4
LT
1293}
1294
1295/**
1296 * call_netdevice_notifiers - call all network notifier blocks
1297 * @val: value passed unmodified to notifier function
c4ea43c5 1298 * @dev: net_device pointer passed unmodified to notifier function
1da177e4
LT
1299 *
1300 * Call all network notifier blocks. Parameters and return value
f07d5b94 1301 * are as for raw_notifier_call_chain().
1da177e4
LT
1302 */
1303
ad7379d4 1304int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1da177e4 1305{
ad7379d4 1306 return raw_notifier_call_chain(&netdev_chain, val, dev);
1da177e4
LT
1307}
1308
1309/* When > 0 there are consumers of rx skb time stamps */
1310static atomic_t netstamp_needed = ATOMIC_INIT(0);
1311
1312void net_enable_timestamp(void)
1313{
1314 atomic_inc(&netstamp_needed);
1315}
1316
1317void net_disable_timestamp(void)
1318{
1319 atomic_dec(&netstamp_needed);
1320}
1321
a61bbcf2 1322static inline void net_timestamp(struct sk_buff *skb)
1da177e4
LT
1323{
1324 if (atomic_read(&netstamp_needed))
a61bbcf2 1325 __net_timestamp(skb);
b7aa0bf7
ED
1326 else
1327 skb->tstamp.tv64 = 0;
1da177e4
LT
1328}
1329
1330/*
1331 * Support routine. Sends outgoing frames to any network
1332 * taps currently in use.
1333 */
1334
f6a78bfc 1335static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1da177e4
LT
1336{
1337 struct packet_type *ptype;
a61bbcf2 1338
8caf1539
JP
1339#ifdef CONFIG_NET_CLS_ACT
1340 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1341 net_timestamp(skb);
1342#else
a61bbcf2 1343 net_timestamp(skb);
8caf1539 1344#endif
1da177e4
LT
1345
1346 rcu_read_lock();
1347 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1348 /* Never send packets back to the socket
1349 * they originated from - MvS (miquels@drinkel.ow.org)
1350 */
1351 if ((ptype->dev == dev || !ptype->dev) &&
1352 (ptype->af_packet_priv == NULL ||
1353 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1354 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1355 if (!skb2)
1356 break;
1357
1358 /* skb->nh should be correctly
1359 set by sender, so that the second statement is
1360 just protection against buggy protocols.
1361 */
459a98ed 1362 skb_reset_mac_header(skb2);
1da177e4 1363
d56f90a7 1364 if (skb_network_header(skb2) < skb2->data ||
27a884dc 1365 skb2->network_header > skb2->tail) {
1da177e4
LT
1366 if (net_ratelimit())
1367 printk(KERN_CRIT "protocol %04x is "
1368 "buggy, dev %s\n",
1369 skb2->protocol, dev->name);
c1d2bbe1 1370 skb_reset_network_header(skb2);
1da177e4
LT
1371 }
1372
b0e380b1 1373 skb2->transport_header = skb2->network_header;
1da177e4 1374 skb2->pkt_type = PACKET_OUTGOING;
f2ccd8fa 1375 ptype->func(skb2, skb->dev, ptype, skb->dev);
1da177e4
LT
1376 }
1377 }
1378 rcu_read_unlock();
1379}
1380
56079431 1381
def82a1d 1382static inline void __netif_reschedule(struct Qdisc *q)
56079431 1383{
def82a1d
JP
1384 struct softnet_data *sd;
1385 unsigned long flags;
56079431 1386
def82a1d
JP
1387 local_irq_save(flags);
1388 sd = &__get_cpu_var(softnet_data);
1389 q->next_sched = sd->output_queue;
1390 sd->output_queue = q;
1391 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1392 local_irq_restore(flags);
1393}
1394
1395void __netif_schedule(struct Qdisc *q)
1396{
1397 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1398 __netif_reschedule(q);
56079431
DV
1399}
1400EXPORT_SYMBOL(__netif_schedule);
1401
bea3348e 1402void dev_kfree_skb_irq(struct sk_buff *skb)
56079431 1403{
bea3348e
SH
1404 if (atomic_dec_and_test(&skb->users)) {
1405 struct softnet_data *sd;
1406 unsigned long flags;
56079431 1407
bea3348e
SH
1408 local_irq_save(flags);
1409 sd = &__get_cpu_var(softnet_data);
1410 skb->next = sd->completion_queue;
1411 sd->completion_queue = skb;
1412 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1413 local_irq_restore(flags);
1414 }
56079431 1415}
bea3348e 1416EXPORT_SYMBOL(dev_kfree_skb_irq);
56079431
DV
1417
1418void dev_kfree_skb_any(struct sk_buff *skb)
1419{
1420 if (in_irq() || irqs_disabled())
1421 dev_kfree_skb_irq(skb);
1422 else
1423 dev_kfree_skb(skb);
1424}
1425EXPORT_SYMBOL(dev_kfree_skb_any);
1426
1427
bea3348e
SH
1428/**
1429 * netif_device_detach - mark device as removed
1430 * @dev: network device
1431 *
1432 * Mark device as removed from system and therefore no longer available.
1433 */
56079431
DV
1434void netif_device_detach(struct net_device *dev)
1435{
1436 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1437 netif_running(dev)) {
d543103a 1438 netif_tx_stop_all_queues(dev);
56079431
DV
1439 }
1440}
1441EXPORT_SYMBOL(netif_device_detach);
1442
bea3348e
SH
1443/**
1444 * netif_device_attach - mark device as attached
1445 * @dev: network device
1446 *
1447 * Mark device as attached from system and restart if needed.
1448 */
56079431
DV
1449void netif_device_attach(struct net_device *dev)
1450{
1451 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1452 netif_running(dev)) {
d543103a 1453 netif_tx_wake_all_queues(dev);
4ec93edb 1454 __netdev_watchdog_up(dev);
56079431
DV
1455 }
1456}
1457EXPORT_SYMBOL(netif_device_attach);
1458
6de329e2
BH
1459static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1460{
1461 return ((features & NETIF_F_GEN_CSUM) ||
1462 ((features & NETIF_F_IP_CSUM) &&
1463 protocol == htons(ETH_P_IP)) ||
1464 ((features & NETIF_F_IPV6_CSUM) &&
1c8dbcf6
YZ
1465 protocol == htons(ETH_P_IPV6)) ||
1466 ((features & NETIF_F_FCOE_CRC) &&
1467 protocol == htons(ETH_P_FCOE)));
6de329e2
BH
1468}
1469
1470static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1471{
1472 if (can_checksum_protocol(dev->features, skb->protocol))
1473 return true;
1474
1475 if (skb->protocol == htons(ETH_P_8021Q)) {
1476 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1477 if (can_checksum_protocol(dev->features & dev->vlan_features,
1478 veh->h_vlan_encapsulated_proto))
1479 return true;
1480 }
1481
1482 return false;
1483}
56079431 1484
1da177e4
LT
1485/*
1486 * Invalidate hardware checksum when packet is to be mangled, and
1487 * complete checksum manually on outgoing path.
1488 */
84fa7933 1489int skb_checksum_help(struct sk_buff *skb)
1da177e4 1490{
d3bc23e7 1491 __wsum csum;
663ead3b 1492 int ret = 0, offset;
1da177e4 1493
84fa7933 1494 if (skb->ip_summed == CHECKSUM_COMPLETE)
a430a43d
HX
1495 goto out_set_summed;
1496
1497 if (unlikely(skb_shinfo(skb)->gso_size)) {
a430a43d
HX
1498 /* Let GSO fix up the checksum. */
1499 goto out_set_summed;
1da177e4
LT
1500 }
1501
a030847e
HX
1502 offset = skb->csum_start - skb_headroom(skb);
1503 BUG_ON(offset >= skb_headlen(skb));
1504 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1505
1506 offset += skb->csum_offset;
1507 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1508
1509 if (skb_cloned(skb) &&
1510 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1da177e4
LT
1511 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1512 if (ret)
1513 goto out;
1514 }
1515
a030847e 1516 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
a430a43d 1517out_set_summed:
1da177e4 1518 skb->ip_summed = CHECKSUM_NONE;
4ec93edb 1519out:
1da177e4
LT
1520 return ret;
1521}
1522
f6a78bfc
HX
1523/**
1524 * skb_gso_segment - Perform segmentation on skb.
1525 * @skb: buffer to segment
576a30eb 1526 * @features: features for the output path (see dev->features)
f6a78bfc
HX
1527 *
1528 * This function segments the given skb and returns a list of segments.
576a30eb
HX
1529 *
1530 * It may return NULL if the skb requires no segmentation. This is
1531 * only possible when GSO is used for verifying header integrity.
f6a78bfc 1532 */
576a30eb 1533struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
f6a78bfc
HX
1534{
1535 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1536 struct packet_type *ptype;
252e3346 1537 __be16 type = skb->protocol;
a430a43d 1538 int err;
f6a78bfc 1539
459a98ed 1540 skb_reset_mac_header(skb);
b0e380b1 1541 skb->mac_len = skb->network_header - skb->mac_header;
f6a78bfc
HX
1542 __skb_pull(skb, skb->mac_len);
1543
67fd1a73
HX
1544 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1545 struct net_device *dev = skb->dev;
1546 struct ethtool_drvinfo info = {};
1547
1548 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1549 dev->ethtool_ops->get_drvinfo(dev, &info);
1550
1551 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1552 "ip_summed=%d",
1553 info.driver, dev ? dev->features : 0L,
1554 skb->sk ? skb->sk->sk_route_caps : 0L,
1555 skb->len, skb->data_len, skb->ip_summed);
1556
a430a43d
HX
1557 if (skb_header_cloned(skb) &&
1558 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1559 return ERR_PTR(err);
1560 }
1561
f6a78bfc 1562 rcu_read_lock();
82d8a867
PE
1563 list_for_each_entry_rcu(ptype,
1564 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
f6a78bfc 1565 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
84fa7933 1566 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
a430a43d
HX
1567 err = ptype->gso_send_check(skb);
1568 segs = ERR_PTR(err);
1569 if (err || skb_gso_ok(skb, features))
1570 break;
d56f90a7
ACM
1571 __skb_push(skb, (skb->data -
1572 skb_network_header(skb)));
a430a43d 1573 }
576a30eb 1574 segs = ptype->gso_segment(skb, features);
f6a78bfc
HX
1575 break;
1576 }
1577 }
1578 rcu_read_unlock();
1579
98e399f8 1580 __skb_push(skb, skb->data - skb_mac_header(skb));
576a30eb 1581
f6a78bfc
HX
1582 return segs;
1583}
1584
1585EXPORT_SYMBOL(skb_gso_segment);
1586
fb286bb2
HX
1587/* Take action when hardware reception checksum errors are detected. */
1588#ifdef CONFIG_BUG
1589void netdev_rx_csum_fault(struct net_device *dev)
1590{
1591 if (net_ratelimit()) {
4ec93edb 1592 printk(KERN_ERR "%s: hw csum failure.\n",
246a4212 1593 dev ? dev->name : "<unknown>");
fb286bb2
HX
1594 dump_stack();
1595 }
1596}
1597EXPORT_SYMBOL(netdev_rx_csum_fault);
1598#endif
1599
1da177e4
LT
1600/* Actually, we should eliminate this check as soon as we know, that:
1601 * 1. IOMMU is present and allows to map all the memory.
1602 * 2. No high memory really exists on this machine.
1603 */
1604
1605static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1606{
3d3a8533 1607#ifdef CONFIG_HIGHMEM
1da177e4
LT
1608 int i;
1609
1610 if (dev->features & NETIF_F_HIGHDMA)
1611 return 0;
1612
1613 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1614 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1615 return 1;
1616
3d3a8533 1617#endif
1da177e4
LT
1618 return 0;
1619}
1da177e4 1620
f6a78bfc
HX
1621struct dev_gso_cb {
1622 void (*destructor)(struct sk_buff *skb);
1623};
1624
1625#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1626
1627static void dev_gso_skb_destructor(struct sk_buff *skb)
1628{
1629 struct dev_gso_cb *cb;
1630
1631 do {
1632 struct sk_buff *nskb = skb->next;
1633
1634 skb->next = nskb->next;
1635 nskb->next = NULL;
1636 kfree_skb(nskb);
1637 } while (skb->next);
1638
1639 cb = DEV_GSO_CB(skb);
1640 if (cb->destructor)
1641 cb->destructor(skb);
1642}
1643
1644/**
1645 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1646 * @skb: buffer to segment
1647 *
1648 * This function segments the given skb and stores the list of segments
1649 * in skb->next.
1650 */
1651static int dev_gso_segment(struct sk_buff *skb)
1652{
1653 struct net_device *dev = skb->dev;
1654 struct sk_buff *segs;
576a30eb
HX
1655 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1656 NETIF_F_SG : 0);
1657
1658 segs = skb_gso_segment(skb, features);
1659
1660 /* Verifying header integrity only. */
1661 if (!segs)
1662 return 0;
f6a78bfc 1663
801678c5 1664 if (IS_ERR(segs))
f6a78bfc
HX
1665 return PTR_ERR(segs);
1666
1667 skb->next = segs;
1668 DEV_GSO_CB(skb)->destructor = skb->destructor;
1669 skb->destructor = dev_gso_skb_destructor;
1670
1671 return 0;
1672}
1673
fd2ea0a7
DM
1674int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1675 struct netdev_queue *txq)
f6a78bfc 1676{
00829823 1677 const struct net_device_ops *ops = dev->netdev_ops;
ac45f602 1678 int rc;
00829823 1679
f6a78bfc 1680 if (likely(!skb->next)) {
9be9a6b9 1681 if (!list_empty(&ptype_all))
f6a78bfc
HX
1682 dev_queue_xmit_nit(skb, dev);
1683
576a30eb
HX
1684 if (netif_needs_gso(dev, skb)) {
1685 if (unlikely(dev_gso_segment(skb)))
1686 goto out_kfree_skb;
1687 if (skb->next)
1688 goto gso;
1689 }
f6a78bfc 1690
ac45f602
PO
1691 rc = ops->ndo_start_xmit(skb, dev);
1692 /*
1693 * TODO: if skb_orphan() was called by
1694 * dev->hard_start_xmit() (for example, the unmodified
1695 * igb driver does that; bnx2 doesn't), then
1696 * skb_tx_software_timestamp() will be unable to send
1697 * back the time stamp.
1698 *
1699 * How can this be prevented? Always create another
1700 * reference to the socket before calling
1701 * dev->hard_start_xmit()? Prevent that skb_orphan()
1702 * does anything in dev->hard_start_xmit() by clearing
1703 * the skb destructor before the call and restoring it
1704 * afterwards, then doing the skb_orphan() ourselves?
1705 */
ac45f602 1706 return rc;
f6a78bfc
HX
1707 }
1708
576a30eb 1709gso:
f6a78bfc
HX
1710 do {
1711 struct sk_buff *nskb = skb->next;
f6a78bfc
HX
1712
1713 skb->next = nskb->next;
1714 nskb->next = NULL;
00829823 1715 rc = ops->ndo_start_xmit(nskb, dev);
f6a78bfc 1716 if (unlikely(rc)) {
f54d9e8d 1717 nskb->next = skb->next;
f6a78bfc
HX
1718 skb->next = nskb;
1719 return rc;
1720 }
fd2ea0a7 1721 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
f54d9e8d 1722 return NETDEV_TX_BUSY;
f6a78bfc 1723 } while (skb->next);
4ec93edb 1724
f6a78bfc
HX
1725 skb->destructor = DEV_GSO_CB(skb)->destructor;
1726
1727out_kfree_skb:
1728 kfree_skb(skb);
1729 return 0;
1730}
1731
7019298a 1732static u32 skb_tx_hashrnd;
b6b2fed1 1733
9247744e 1734u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
8f0f2223 1735{
7019298a 1736 u32 hash;
b6b2fed1 1737
513de11b
DM
1738 if (skb_rx_queue_recorded(skb)) {
1739 hash = skb_get_rx_queue(skb);
1740 while (unlikely (hash >= dev->real_num_tx_queues))
1741 hash -= dev->real_num_tx_queues;
1742 return hash;
1743 }
ec581f6a
ED
1744
1745 if (skb->sk && skb->sk->sk_hash)
7019298a 1746 hash = skb->sk->sk_hash;
ec581f6a 1747 else
7019298a 1748 hash = skb->protocol;
d5a9e24a 1749
7019298a 1750 hash = jhash_1word(hash, skb_tx_hashrnd);
b6b2fed1
DM
1751
1752 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
8f0f2223 1753}
9247744e 1754EXPORT_SYMBOL(skb_tx_hash);
8f0f2223 1755
e8a0464c
DM
1756static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1757 struct sk_buff *skb)
1758{
00829823 1759 const struct net_device_ops *ops = dev->netdev_ops;
fd2ea0a7
DM
1760 u16 queue_index = 0;
1761
00829823
SH
1762 if (ops->ndo_select_queue)
1763 queue_index = ops->ndo_select_queue(dev, skb);
8f0f2223 1764 else if (dev->real_num_tx_queues > 1)
7019298a 1765 queue_index = skb_tx_hash(dev, skb);
eae792b7 1766
fd2ea0a7
DM
1767 skb_set_queue_mapping(skb, queue_index);
1768 return netdev_get_tx_queue(dev, queue_index);
e8a0464c
DM
1769}
1770
d29f749e
DJ
1771/**
1772 * dev_queue_xmit - transmit a buffer
1773 * @skb: buffer to transmit
1774 *
1775 * Queue a buffer for transmission to a network device. The caller must
1776 * have set the device and priority and built the buffer before calling
1777 * this function. The function can be called from an interrupt.
1778 *
1779 * A negative errno code is returned on a failure. A success does not
1780 * guarantee the frame will be transmitted as it may be dropped due
1781 * to congestion or traffic shaping.
1782 *
1783 * -----------------------------------------------------------------------------------
1784 * I notice this method can also return errors from the queue disciplines,
1785 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1786 * be positive.
1787 *
1788 * Regardless of the return value, the skb is consumed, so it is currently
1789 * difficult to retry a send to this method. (You can bump the ref count
1790 * before sending to hold a reference for retry if you are careful.)
1791 *
1792 * When calling this method, interrupts MUST be enabled. This is because
1793 * the BH enable code must have IRQs enabled so that it will not deadlock.
1794 * --BLG
1795 */
1da177e4
LT
1796int dev_queue_xmit(struct sk_buff *skb)
1797{
1798 struct net_device *dev = skb->dev;
dc2b4847 1799 struct netdev_queue *txq;
1da177e4
LT
1800 struct Qdisc *q;
1801 int rc = -ENOMEM;
1802
f6a78bfc
HX
1803 /* GSO will handle the following emulations directly. */
1804 if (netif_needs_gso(dev, skb))
1805 goto gso;
1806
1da177e4
LT
1807 if (skb_shinfo(skb)->frag_list &&
1808 !(dev->features & NETIF_F_FRAGLIST) &&
364c6bad 1809 __skb_linearize(skb))
1da177e4
LT
1810 goto out_kfree_skb;
1811
1812 /* Fragmented skb is linearized if device does not support SG,
1813 * or if at least one of fragments is in highmem and device
1814 * does not support DMA from it.
1815 */
1816 if (skb_shinfo(skb)->nr_frags &&
1817 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
364c6bad 1818 __skb_linearize(skb))
1da177e4
LT
1819 goto out_kfree_skb;
1820
1821 /* If packet is not checksummed and device does not support
1822 * checksumming for this protocol, complete checksumming here.
1823 */
663ead3b
HX
1824 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1825 skb_set_transport_header(skb, skb->csum_start -
1826 skb_headroom(skb));
6de329e2
BH
1827 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1828 goto out_kfree_skb;
663ead3b 1829 }
1da177e4 1830
f6a78bfc 1831gso:
4ec93edb
YH
1832 /* Disable soft irqs for various locks below. Also
1833 * stops preemption for RCU.
1da177e4 1834 */
4ec93edb 1835 rcu_read_lock_bh();
1da177e4 1836
eae792b7 1837 txq = dev_pick_tx(dev, skb);
b0e1e646 1838 q = rcu_dereference(txq->qdisc);
37437bb2 1839
1da177e4
LT
1840#ifdef CONFIG_NET_CLS_ACT
1841 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1842#endif
1843 if (q->enqueue) {
5fb66229 1844 spinlock_t *root_lock = qdisc_lock(q);
37437bb2
DM
1845
1846 spin_lock(root_lock);
1847
a9312ae8 1848 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
96d20316 1849 kfree_skb(skb);
a9312ae8 1850 rc = NET_XMIT_DROP;
96d20316
DM
1851 } else {
1852 rc = qdisc_enqueue_root(skb, q);
1853 qdisc_run(q);
a9312ae8 1854 }
37437bb2
DM
1855 spin_unlock(root_lock);
1856
37437bb2 1857 goto out;
1da177e4
LT
1858 }
1859
1860 /* The device has no queue. Common case for software devices:
1861 loopback, all the sorts of tunnels...
1862
932ff279
HX
1863 Really, it is unlikely that netif_tx_lock protection is necessary
1864 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1da177e4
LT
1865 counters.)
1866 However, it is possible, that they rely on protection
1867 made by us here.
1868
1869 Check this and shot the lock. It is not prone from deadlocks.
1870 Either shot noqueue qdisc, it is even simpler 8)
1871 */
1872 if (dev->flags & IFF_UP) {
1873 int cpu = smp_processor_id(); /* ok because BHs are off */
1874
c773e847 1875 if (txq->xmit_lock_owner != cpu) {
1da177e4 1876
c773e847 1877 HARD_TX_LOCK(dev, txq, cpu);
1da177e4 1878
fd2ea0a7 1879 if (!netif_tx_queue_stopped(txq)) {
1da177e4 1880 rc = 0;
fd2ea0a7 1881 if (!dev_hard_start_xmit(skb, dev, txq)) {
c773e847 1882 HARD_TX_UNLOCK(dev, txq);
1da177e4
LT
1883 goto out;
1884 }
1885 }
c773e847 1886 HARD_TX_UNLOCK(dev, txq);
1da177e4
LT
1887 if (net_ratelimit())
1888 printk(KERN_CRIT "Virtual device %s asks to "
1889 "queue packet!\n", dev->name);
1890 } else {
1891 /* Recursion is detected! It is possible,
1892 * unfortunately */
1893 if (net_ratelimit())
1894 printk(KERN_CRIT "Dead loop on virtual device "
1895 "%s, fix it urgently!\n", dev->name);
1896 }
1897 }
1898
1899 rc = -ENETDOWN;
d4828d85 1900 rcu_read_unlock_bh();
1da177e4
LT
1901
1902out_kfree_skb:
1903 kfree_skb(skb);
1904 return rc;
1905out:
d4828d85 1906 rcu_read_unlock_bh();
1da177e4
LT
1907 return rc;
1908}
1909
1910
1911/*=======================================================================
1912 Receiver routines
1913 =======================================================================*/
1914
6b2bedc3
SH
1915int netdev_max_backlog __read_mostly = 1000;
1916int netdev_budget __read_mostly = 300;
1917int weight_p __read_mostly = 64; /* old backlog weight */
1da177e4
LT
1918
1919DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1920
1921
1da177e4
LT
1922/**
1923 * netif_rx - post buffer to the network code
1924 * @skb: buffer to post
1925 *
1926 * This function receives a packet from a device driver and queues it for
1927 * the upper (protocol) levels to process. It always succeeds. The buffer
1928 * may be dropped during processing for congestion control or by the
1929 * protocol layers.
1930 *
1931 * return values:
1932 * NET_RX_SUCCESS (no congestion)
1da177e4
LT
1933 * NET_RX_DROP (packet was dropped)
1934 *
1935 */
1936
1937int netif_rx(struct sk_buff *skb)
1938{
1da177e4
LT
1939 struct softnet_data *queue;
1940 unsigned long flags;
1941
1942 /* if netpoll wants it, pretend we never saw it */
1943 if (netpoll_rx(skb))
1944 return NET_RX_DROP;
1945
b7aa0bf7 1946 if (!skb->tstamp.tv64)
a61bbcf2 1947 net_timestamp(skb);
1da177e4
LT
1948
1949 /*
1950 * The code is rearranged so that the path is the most
1951 * short when CPU is congested, but is still operating.
1952 */
1953 local_irq_save(flags);
1da177e4
LT
1954 queue = &__get_cpu_var(softnet_data);
1955
1956 __get_cpu_var(netdev_rx_stat).total++;
1957 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1958 if (queue->input_pkt_queue.qlen) {
1da177e4 1959enqueue:
1da177e4 1960 __skb_queue_tail(&queue->input_pkt_queue, skb);
1da177e4 1961 local_irq_restore(flags);
34008d8c 1962 return NET_RX_SUCCESS;
1da177e4
LT
1963 }
1964
bea3348e 1965 napi_schedule(&queue->backlog);
1da177e4
LT
1966 goto enqueue;
1967 }
1968
1da177e4
LT
1969 __get_cpu_var(netdev_rx_stat).dropped++;
1970 local_irq_restore(flags);
1971
1972 kfree_skb(skb);
1973 return NET_RX_DROP;
1974}
1975
1976int netif_rx_ni(struct sk_buff *skb)
1977{
1978 int err;
1979
1980 preempt_disable();
1981 err = netif_rx(skb);
1982 if (local_softirq_pending())
1983 do_softirq();
1984 preempt_enable();
1985
1986 return err;
1987}
1988
1989EXPORT_SYMBOL(netif_rx_ni);
1990
1da177e4
LT
1991static void net_tx_action(struct softirq_action *h)
1992{
1993 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1994
1995 if (sd->completion_queue) {
1996 struct sk_buff *clist;
1997
1998 local_irq_disable();
1999 clist = sd->completion_queue;
2000 sd->completion_queue = NULL;
2001 local_irq_enable();
2002
2003 while (clist) {
2004 struct sk_buff *skb = clist;
2005 clist = clist->next;
2006
547b792c 2007 WARN_ON(atomic_read(&skb->users));
1da177e4
LT
2008 __kfree_skb(skb);
2009 }
2010 }
2011
2012 if (sd->output_queue) {
37437bb2 2013 struct Qdisc *head;
1da177e4
LT
2014
2015 local_irq_disable();
2016 head = sd->output_queue;
2017 sd->output_queue = NULL;
2018 local_irq_enable();
2019
2020 while (head) {
37437bb2
DM
2021 struct Qdisc *q = head;
2022 spinlock_t *root_lock;
2023
1da177e4
LT
2024 head = head->next_sched;
2025
5fb66229 2026 root_lock = qdisc_lock(q);
37437bb2 2027 if (spin_trylock(root_lock)) {
def82a1d
JP
2028 smp_mb__before_clear_bit();
2029 clear_bit(__QDISC_STATE_SCHED,
2030 &q->state);
37437bb2
DM
2031 qdisc_run(q);
2032 spin_unlock(root_lock);
1da177e4 2033 } else {
195648bb 2034 if (!test_bit(__QDISC_STATE_DEACTIVATED,
e8a83e10 2035 &q->state)) {
195648bb 2036 __netif_reschedule(q);
e8a83e10
JP
2037 } else {
2038 smp_mb__before_clear_bit();
2039 clear_bit(__QDISC_STATE_SCHED,
2040 &q->state);
2041 }
1da177e4
LT
2042 }
2043 }
2044 }
2045}
2046
6f05f629
SH
2047static inline int deliver_skb(struct sk_buff *skb,
2048 struct packet_type *pt_prev,
2049 struct net_device *orig_dev)
1da177e4
LT
2050{
2051 atomic_inc(&skb->users);
f2ccd8fa 2052 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
2053}
2054
2055#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
6229e362 2056/* These hooks defined here for ATM */
1da177e4
LT
2057struct net_bridge;
2058struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2059 unsigned char *addr);
6229e362 2060void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1da177e4 2061
6229e362
SH
2062/*
2063 * If bridge module is loaded call bridging hook.
2064 * returns NULL if packet was consumed.
2065 */
2066struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2067 struct sk_buff *skb) __read_mostly;
2068static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2069 struct packet_type **pt_prev, int *ret,
2070 struct net_device *orig_dev)
1da177e4
LT
2071{
2072 struct net_bridge_port *port;
2073
6229e362
SH
2074 if (skb->pkt_type == PACKET_LOOPBACK ||
2075 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2076 return skb;
1da177e4
LT
2077
2078 if (*pt_prev) {
6229e362 2079 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1da177e4 2080 *pt_prev = NULL;
4ec93edb
YH
2081 }
2082
6229e362 2083 return br_handle_frame_hook(port, skb);
1da177e4
LT
2084}
2085#else
6229e362 2086#define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1da177e4
LT
2087#endif
2088
b863ceb7
PM
2089#if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2090struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2091EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2092
2093static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2094 struct packet_type **pt_prev,
2095 int *ret,
2096 struct net_device *orig_dev)
2097{
2098 if (skb->dev->macvlan_port == NULL)
2099 return skb;
2100
2101 if (*pt_prev) {
2102 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2103 *pt_prev = NULL;
2104 }
2105 return macvlan_handle_frame_hook(skb);
2106}
2107#else
2108#define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2109#endif
2110
1da177e4
LT
2111#ifdef CONFIG_NET_CLS_ACT
2112/* TODO: Maybe we should just force sch_ingress to be compiled in
2113 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2114 * a compare and 2 stores extra right now if we dont have it on
2115 * but have CONFIG_NET_CLS_ACT
4ec93edb 2116 * NOTE: This doesnt stop any functionality; if you dont have
1da177e4
LT
2117 * the ingress scheduler, you just cant add policies on ingress.
2118 *
2119 */
4ec93edb 2120static int ing_filter(struct sk_buff *skb)
1da177e4 2121{
1da177e4 2122 struct net_device *dev = skb->dev;
f697c3e8 2123 u32 ttl = G_TC_RTTL(skb->tc_verd);
555353cf
DM
2124 struct netdev_queue *rxq;
2125 int result = TC_ACT_OK;
2126 struct Qdisc *q;
4ec93edb 2127
f697c3e8
HX
2128 if (MAX_RED_LOOP < ttl++) {
2129 printk(KERN_WARNING
2130 "Redir loop detected Dropping packet (%d->%d)\n",
2131 skb->iif, dev->ifindex);
2132 return TC_ACT_SHOT;
2133 }
1da177e4 2134
f697c3e8
HX
2135 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2136 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
1da177e4 2137
555353cf
DM
2138 rxq = &dev->rx_queue;
2139
83874000 2140 q = rxq->qdisc;
8d50b53d 2141 if (q != &noop_qdisc) {
83874000 2142 spin_lock(qdisc_lock(q));
a9312ae8
DM
2143 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2144 result = qdisc_enqueue_root(skb, q);
83874000
DM
2145 spin_unlock(qdisc_lock(q));
2146 }
f697c3e8
HX
2147
2148 return result;
2149}
86e65da9 2150
f697c3e8
HX
2151static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2152 struct packet_type **pt_prev,
2153 int *ret, struct net_device *orig_dev)
2154{
8d50b53d 2155 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
f697c3e8 2156 goto out;
1da177e4 2157
f697c3e8
HX
2158 if (*pt_prev) {
2159 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2160 *pt_prev = NULL;
2161 } else {
2162 /* Huh? Why does turning on AF_PACKET affect this? */
2163 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1da177e4
LT
2164 }
2165
f697c3e8
HX
2166 switch (ing_filter(skb)) {
2167 case TC_ACT_SHOT:
2168 case TC_ACT_STOLEN:
2169 kfree_skb(skb);
2170 return NULL;
2171 }
2172
2173out:
2174 skb->tc_verd = 0;
2175 return skb;
1da177e4
LT
2176}
2177#endif
2178
bc1d0411
PM
2179/*
2180 * netif_nit_deliver - deliver received packets to network taps
2181 * @skb: buffer
2182 *
2183 * This function is used to deliver incoming packets to network
2184 * taps. It should be used when the normal netif_receive_skb path
2185 * is bypassed, for example because of VLAN acceleration.
2186 */
2187void netif_nit_deliver(struct sk_buff *skb)
2188{
2189 struct packet_type *ptype;
2190
2191 if (list_empty(&ptype_all))
2192 return;
2193
2194 skb_reset_network_header(skb);
2195 skb_reset_transport_header(skb);
2196 skb->mac_len = skb->network_header - skb->mac_header;
2197
2198 rcu_read_lock();
2199 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2200 if (!ptype->dev || ptype->dev == skb->dev)
2201 deliver_skb(skb, ptype, skb->dev);
2202 }
2203 rcu_read_unlock();
2204}
2205
3b582cc1
SH
2206/**
2207 * netif_receive_skb - process receive buffer from network
2208 * @skb: buffer to process
2209 *
2210 * netif_receive_skb() is the main receive data processing function.
2211 * It always succeeds. The buffer may be dropped during processing
2212 * for congestion control or by the protocol layers.
2213 *
2214 * This function may only be called from softirq context and interrupts
2215 * should be enabled.
2216 *
2217 * Return values (usually ignored):
2218 * NET_RX_SUCCESS: no congestion
2219 * NET_RX_DROP: packet was dropped
2220 */
1da177e4
LT
2221int netif_receive_skb(struct sk_buff *skb)
2222{
2223 struct packet_type *ptype, *pt_prev;
f2ccd8fa 2224 struct net_device *orig_dev;
0d7a3681 2225 struct net_device *null_or_orig;
1da177e4 2226 int ret = NET_RX_DROP;
252e3346 2227 __be16 type;
1da177e4 2228
9b22ea56
PM
2229 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2230 return NET_RX_SUCCESS;
2231
1da177e4 2232 /* if we've gotten here through NAPI, check netpoll */
bea3348e 2233 if (netpoll_receive_skb(skb))
1da177e4
LT
2234 return NET_RX_DROP;
2235
b7aa0bf7 2236 if (!skb->tstamp.tv64)
a61bbcf2 2237 net_timestamp(skb);
1da177e4 2238
c01003c2
PM
2239 if (!skb->iif)
2240 skb->iif = skb->dev->ifindex;
86e65da9 2241
0d7a3681 2242 null_or_orig = NULL;
cc9bd5ce
JE
2243 orig_dev = skb->dev;
2244 if (orig_dev->master) {
0d7a3681
JE
2245 if (skb_bond_should_drop(skb))
2246 null_or_orig = orig_dev; /* deliver only exact match */
2247 else
2248 skb->dev = orig_dev->master;
cc9bd5ce 2249 }
8f903c70 2250
1da177e4
LT
2251 __get_cpu_var(netdev_rx_stat).total++;
2252
c1d2bbe1 2253 skb_reset_network_header(skb);
badff6d0 2254 skb_reset_transport_header(skb);
b0e380b1 2255 skb->mac_len = skb->network_header - skb->mac_header;
1da177e4
LT
2256
2257 pt_prev = NULL;
2258
2259 rcu_read_lock();
2260
2261#ifdef CONFIG_NET_CLS_ACT
2262 if (skb->tc_verd & TC_NCLS) {
2263 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2264 goto ncls;
2265 }
2266#endif
2267
2268 list_for_each_entry_rcu(ptype, &ptype_all, list) {
f982307f
JE
2269 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2270 ptype->dev == orig_dev) {
4ec93edb 2271 if (pt_prev)
f2ccd8fa 2272 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
2273 pt_prev = ptype;
2274 }
2275 }
2276
2277#ifdef CONFIG_NET_CLS_ACT
f697c3e8
HX
2278 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2279 if (!skb)
1da177e4 2280 goto out;
1da177e4
LT
2281ncls:
2282#endif
2283
6229e362 2284 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
b863ceb7
PM
2285 if (!skb)
2286 goto out;
2287 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
6229e362 2288 if (!skb)
1da177e4
LT
2289 goto out;
2290
9a279bcb
HX
2291 skb_orphan(skb);
2292
1da177e4 2293 type = skb->protocol;
82d8a867
PE
2294 list_for_each_entry_rcu(ptype,
2295 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1da177e4 2296 if (ptype->type == type &&
f982307f
JE
2297 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2298 ptype->dev == orig_dev)) {
4ec93edb 2299 if (pt_prev)
f2ccd8fa 2300 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
2301 pt_prev = ptype;
2302 }
2303 }
2304
2305 if (pt_prev) {
f2ccd8fa 2306 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
2307 } else {
2308 kfree_skb(skb);
2309 /* Jamal, now you will not able to escape explaining
2310 * me how you were going to use this. :-)
2311 */
2312 ret = NET_RX_DROP;
2313 }
2314
2315out:
2316 rcu_read_unlock();
2317 return ret;
2318}
2319
6e583ce5
SH
2320/* Network device is going away, flush any packets still pending */
2321static void flush_backlog(void *arg)
2322{
2323 struct net_device *dev = arg;
2324 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2325 struct sk_buff *skb, *tmp;
2326
2327 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2328 if (skb->dev == dev) {
2329 __skb_unlink(skb, &queue->input_pkt_queue);
2330 kfree_skb(skb);
2331 }
2332}
2333
d565b0a1
HX
2334static int napi_gro_complete(struct sk_buff *skb)
2335{
2336 struct packet_type *ptype;
2337 __be16 type = skb->protocol;
2338 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2339 int err = -ENOENT;
2340
fc59f9a3
HX
2341 if (NAPI_GRO_CB(skb)->count == 1) {
2342 skb_shinfo(skb)->gso_size = 0;
d565b0a1 2343 goto out;
fc59f9a3 2344 }
d565b0a1
HX
2345
2346 rcu_read_lock();
2347 list_for_each_entry_rcu(ptype, head, list) {
2348 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2349 continue;
2350
2351 err = ptype->gro_complete(skb);
2352 break;
2353 }
2354 rcu_read_unlock();
2355
2356 if (err) {
2357 WARN_ON(&ptype->list == head);
2358 kfree_skb(skb);
2359 return NET_RX_SUCCESS;
2360 }
2361
2362out:
d565b0a1
HX
2363 return netif_receive_skb(skb);
2364}
2365
2366void napi_gro_flush(struct napi_struct *napi)
2367{
2368 struct sk_buff *skb, *next;
2369
2370 for (skb = napi->gro_list; skb; skb = next) {
2371 next = skb->next;
2372 skb->next = NULL;
2373 napi_gro_complete(skb);
2374 }
2375
4ae5544f 2376 napi->gro_count = 0;
d565b0a1
HX
2377 napi->gro_list = NULL;
2378}
2379EXPORT_SYMBOL(napi_gro_flush);
2380
86911732
HX
2381void *skb_gro_header(struct sk_buff *skb, unsigned int hlen)
2382{
2383 unsigned int offset = skb_gro_offset(skb);
2384
2385 hlen += offset;
edbd9e30
HX
2386 if (unlikely(skb_headlen(skb) ||
2387 skb_shinfo(skb)->frags[0].size < hlen ||
86911732
HX
2388 PageHighMem(skb_shinfo(skb)->frags[0].page)))
2389 return pskb_may_pull(skb, hlen) ? skb->data + offset : NULL;
2390
2391 return page_address(skb_shinfo(skb)->frags[0].page) +
edbd9e30 2392 skb_shinfo(skb)->frags[0].page_offset + offset;
86911732
HX
2393}
2394EXPORT_SYMBOL(skb_gro_header);
2395
96e93eab 2396int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
d565b0a1
HX
2397{
2398 struct sk_buff **pp = NULL;
2399 struct packet_type *ptype;
2400 __be16 type = skb->protocol;
2401 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
0da2afd5 2402 int same_flow;
d565b0a1 2403 int mac_len;
5d0d9be8 2404 int ret;
d565b0a1
HX
2405
2406 if (!(skb->dev->features & NETIF_F_GRO))
2407 goto normal;
2408
f17f5c91
HX
2409 if (skb_is_gso(skb) || skb_shinfo(skb)->frag_list)
2410 goto normal;
2411
d565b0a1
HX
2412 rcu_read_lock();
2413 list_for_each_entry_rcu(ptype, head, list) {
d565b0a1
HX
2414 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2415 continue;
2416
86911732 2417 skb_set_network_header(skb, skb_gro_offset(skb));
d565b0a1
HX
2418 mac_len = skb->network_header - skb->mac_header;
2419 skb->mac_len = mac_len;
2420 NAPI_GRO_CB(skb)->same_flow = 0;
2421 NAPI_GRO_CB(skb)->flush = 0;
5d38a079 2422 NAPI_GRO_CB(skb)->free = 0;
d565b0a1 2423
d565b0a1
HX
2424 pp = ptype->gro_receive(&napi->gro_list, skb);
2425 break;
2426 }
2427 rcu_read_unlock();
2428
2429 if (&ptype->list == head)
2430 goto normal;
2431
0da2afd5 2432 same_flow = NAPI_GRO_CB(skb)->same_flow;
5d0d9be8 2433 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
0da2afd5 2434
d565b0a1
HX
2435 if (pp) {
2436 struct sk_buff *nskb = *pp;
2437
2438 *pp = nskb->next;
2439 nskb->next = NULL;
2440 napi_gro_complete(nskb);
4ae5544f 2441 napi->gro_count--;
d565b0a1
HX
2442 }
2443
0da2afd5 2444 if (same_flow)
d565b0a1
HX
2445 goto ok;
2446
4ae5544f 2447 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
d565b0a1 2448 goto normal;
d565b0a1 2449
4ae5544f 2450 napi->gro_count++;
d565b0a1 2451 NAPI_GRO_CB(skb)->count = 1;
86911732 2452 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
d565b0a1
HX
2453 skb->next = napi->gro_list;
2454 napi->gro_list = skb;
5d0d9be8 2455 ret = GRO_HELD;
d565b0a1 2456
ad0f9904
HX
2457pull:
2458 if (unlikely(!pskb_may_pull(skb, skb_gro_offset(skb)))) {
2459 if (napi->gro_list == skb)
2460 napi->gro_list = skb->next;
2461 ret = GRO_DROP;
2462 }
2463
d565b0a1 2464ok:
5d0d9be8 2465 return ret;
d565b0a1
HX
2466
2467normal:
ad0f9904
HX
2468 ret = GRO_NORMAL;
2469 goto pull;
5d38a079 2470}
96e93eab
HX
2471EXPORT_SYMBOL(dev_gro_receive);
2472
2473static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2474{
2475 struct sk_buff *p;
2476
d1c76af9
HX
2477 if (netpoll_rx_on(skb))
2478 return GRO_NORMAL;
2479
96e93eab 2480 for (p = napi->gro_list; p; p = p->next) {
f2bde732
SH
2481 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2482 && !compare_ether_header(skb_mac_header(p),
2483 skb_gro_mac_header(skb));
96e93eab
HX
2484 NAPI_GRO_CB(p)->flush = 0;
2485 }
2486
2487 return dev_gro_receive(napi, skb);
2488}
5d38a079 2489
5d0d9be8 2490int napi_skb_finish(int ret, struct sk_buff *skb)
5d38a079 2491{
5d0d9be8
HX
2492 int err = NET_RX_SUCCESS;
2493
2494 switch (ret) {
2495 case GRO_NORMAL:
5d38a079
HX
2496 return netif_receive_skb(skb);
2497
5d0d9be8
HX
2498 case GRO_DROP:
2499 err = NET_RX_DROP;
2500 /* fall through */
2501
2502 case GRO_MERGED_FREE:
5d38a079
HX
2503 kfree_skb(skb);
2504 break;
2505 }
2506
5d0d9be8
HX
2507 return err;
2508}
2509EXPORT_SYMBOL(napi_skb_finish);
2510
2511int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2512{
86911732
HX
2513 skb_gro_reset_offset(skb);
2514
5d0d9be8 2515 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
d565b0a1
HX
2516}
2517EXPORT_SYMBOL(napi_gro_receive);
2518
96e93eab
HX
2519void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2520{
96e93eab
HX
2521 __skb_pull(skb, skb_headlen(skb));
2522 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2523
2524 napi->skb = skb;
2525}
2526EXPORT_SYMBOL(napi_reuse_skb);
2527
76620aaf 2528struct sk_buff *napi_get_frags(struct napi_struct *napi)
5d38a079
HX
2529{
2530 struct net_device *dev = napi->dev;
2531 struct sk_buff *skb = napi->skb;
5d38a079
HX
2532
2533 if (!skb) {
2534 skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN);
2535 if (!skb)
2536 goto out;
2537
2538 skb_reserve(skb, NET_IP_ALIGN);
80595d59 2539
76620aaf 2540 napi->skb = skb;
80595d59 2541 }
5d38a079 2542
96e93eab
HX
2543out:
2544 return skb;
2545}
76620aaf 2546EXPORT_SYMBOL(napi_get_frags);
96e93eab 2547
5d0d9be8 2548int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
96e93eab 2549{
5d0d9be8 2550 int err = NET_RX_SUCCESS;
96e93eab 2551
5d0d9be8
HX
2552 switch (ret) {
2553 case GRO_NORMAL:
86911732 2554 case GRO_HELD:
86911732
HX
2555 skb->protocol = eth_type_trans(skb, napi->dev);
2556
2557 if (ret == GRO_NORMAL)
2558 return netif_receive_skb(skb);
2559
2560 skb_gro_pull(skb, -ETH_HLEN);
2561 break;
5d38a079 2562
5d0d9be8
HX
2563 case GRO_DROP:
2564 err = NET_RX_DROP;
2565 /* fall through */
5d38a079 2566
5d0d9be8
HX
2567 case GRO_MERGED_FREE:
2568 napi_reuse_skb(napi, skb);
2569 break;
2570 }
5d38a079 2571
5d38a079
HX
2572 return err;
2573}
5d0d9be8
HX
2574EXPORT_SYMBOL(napi_frags_finish);
2575
76620aaf
HX
2576struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2577{
2578 struct sk_buff *skb = napi->skb;
2579 struct ethhdr *eth;
2580
2581 napi->skb = NULL;
2582
2583 skb_reset_mac_header(skb);
2584 skb_gro_reset_offset(skb);
2585
2586 eth = skb_gro_header(skb, sizeof(*eth));
2587 if (!eth) {
2588 napi_reuse_skb(napi, skb);
2589 skb = NULL;
2590 goto out;
2591 }
2592
2593 skb_gro_pull(skb, sizeof(*eth));
2594
2595 /*
2596 * This works because the only protocols we care about don't require
2597 * special handling. We'll fix it up properly at the end.
2598 */
2599 skb->protocol = eth->h_proto;
2600
2601out:
2602 return skb;
2603}
2604EXPORT_SYMBOL(napi_frags_skb);
2605
2606int napi_gro_frags(struct napi_struct *napi)
5d0d9be8 2607{
76620aaf 2608 struct sk_buff *skb = napi_frags_skb(napi);
5d0d9be8
HX
2609
2610 if (!skb)
2611 return NET_RX_DROP;
2612
2613 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2614}
5d38a079
HX
2615EXPORT_SYMBOL(napi_gro_frags);
2616
bea3348e 2617static int process_backlog(struct napi_struct *napi, int quota)
1da177e4
LT
2618{
2619 int work = 0;
1da177e4
LT
2620 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2621 unsigned long start_time = jiffies;
2622
bea3348e
SH
2623 napi->weight = weight_p;
2624 do {
1da177e4 2625 struct sk_buff *skb;
1da177e4
LT
2626
2627 local_irq_disable();
2628 skb = __skb_dequeue(&queue->input_pkt_queue);
bea3348e 2629 if (!skb) {
8f1ead2d 2630 __napi_complete(napi);
bea3348e 2631 local_irq_enable();
8f1ead2d 2632 break;
bea3348e 2633 }
1da177e4
LT
2634 local_irq_enable();
2635
8f1ead2d 2636 netif_receive_skb(skb);
bea3348e 2637 } while (++work < quota && jiffies == start_time);
1da177e4 2638
bea3348e
SH
2639 return work;
2640}
1da177e4 2641
bea3348e
SH
2642/**
2643 * __napi_schedule - schedule for receive
c4ea43c5 2644 * @n: entry to schedule
bea3348e
SH
2645 *
2646 * The entry's receive function will be scheduled to run
2647 */
b5606c2d 2648void __napi_schedule(struct napi_struct *n)
bea3348e
SH
2649{
2650 unsigned long flags;
1da177e4 2651
bea3348e
SH
2652 local_irq_save(flags);
2653 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2654 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2655 local_irq_restore(flags);
1da177e4 2656}
bea3348e
SH
2657EXPORT_SYMBOL(__napi_schedule);
2658
d565b0a1
HX
2659void __napi_complete(struct napi_struct *n)
2660{
2661 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2662 BUG_ON(n->gro_list);
2663
2664 list_del(&n->poll_list);
2665 smp_mb__before_clear_bit();
2666 clear_bit(NAPI_STATE_SCHED, &n->state);
2667}
2668EXPORT_SYMBOL(__napi_complete);
2669
2670void napi_complete(struct napi_struct *n)
2671{
2672 unsigned long flags;
2673
2674 /*
2675 * don't let napi dequeue from the cpu poll list
2676 * just in case its running on a different cpu
2677 */
2678 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2679 return;
2680
2681 napi_gro_flush(n);
2682 local_irq_save(flags);
2683 __napi_complete(n);
2684 local_irq_restore(flags);
2685}
2686EXPORT_SYMBOL(napi_complete);
2687
2688void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2689 int (*poll)(struct napi_struct *, int), int weight)
2690{
2691 INIT_LIST_HEAD(&napi->poll_list);
4ae5544f 2692 napi->gro_count = 0;
d565b0a1 2693 napi->gro_list = NULL;
5d38a079 2694 napi->skb = NULL;
d565b0a1
HX
2695 napi->poll = poll;
2696 napi->weight = weight;
2697 list_add(&napi->dev_list, &dev->napi_list);
d565b0a1 2698 napi->dev = dev;
5d38a079 2699#ifdef CONFIG_NETPOLL
d565b0a1
HX
2700 spin_lock_init(&napi->poll_lock);
2701 napi->poll_owner = -1;
2702#endif
2703 set_bit(NAPI_STATE_SCHED, &napi->state);
2704}
2705EXPORT_SYMBOL(netif_napi_add);
2706
2707void netif_napi_del(struct napi_struct *napi)
2708{
2709 struct sk_buff *skb, *next;
2710
d7b06636 2711 list_del_init(&napi->dev_list);
76620aaf 2712 napi_free_frags(napi);
d565b0a1
HX
2713
2714 for (skb = napi->gro_list; skb; skb = next) {
2715 next = skb->next;
2716 skb->next = NULL;
2717 kfree_skb(skb);
2718 }
2719
2720 napi->gro_list = NULL;
4ae5544f 2721 napi->gro_count = 0;
d565b0a1
HX
2722}
2723EXPORT_SYMBOL(netif_napi_del);
2724
1da177e4
LT
2725
2726static void net_rx_action(struct softirq_action *h)
2727{
bea3348e 2728 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
24f8b238 2729 unsigned long time_limit = jiffies + 2;
51b0bded 2730 int budget = netdev_budget;
53fb95d3
MM
2731 void *have;
2732
1da177e4
LT
2733 local_irq_disable();
2734
bea3348e
SH
2735 while (!list_empty(list)) {
2736 struct napi_struct *n;
2737 int work, weight;
1da177e4 2738
bea3348e 2739 /* If softirq window is exhuasted then punt.
24f8b238
SH
2740 * Allow this to run for 2 jiffies since which will allow
2741 * an average latency of 1.5/HZ.
bea3348e 2742 */
24f8b238 2743 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
1da177e4
LT
2744 goto softnet_break;
2745
2746 local_irq_enable();
2747
bea3348e
SH
2748 /* Even though interrupts have been re-enabled, this
2749 * access is safe because interrupts can only add new
2750 * entries to the tail of this list, and only ->poll()
2751 * calls can remove this head entry from the list.
2752 */
2753 n = list_entry(list->next, struct napi_struct, poll_list);
1da177e4 2754
bea3348e
SH
2755 have = netpoll_poll_lock(n);
2756
2757 weight = n->weight;
2758
0a7606c1
DM
2759 /* This NAPI_STATE_SCHED test is for avoiding a race
2760 * with netpoll's poll_napi(). Only the entity which
2761 * obtains the lock and sees NAPI_STATE_SCHED set will
2762 * actually make the ->poll() call. Therefore we avoid
2763 * accidently calling ->poll() when NAPI is not scheduled.
2764 */
2765 work = 0;
2766 if (test_bit(NAPI_STATE_SCHED, &n->state))
2767 work = n->poll(n, weight);
bea3348e
SH
2768
2769 WARN_ON_ONCE(work > weight);
2770
2771 budget -= work;
2772
2773 local_irq_disable();
2774
2775 /* Drivers must not modify the NAPI state if they
2776 * consume the entire weight. In such cases this code
2777 * still "owns" the NAPI instance and therefore can
2778 * move the instance around on the list at-will.
2779 */
fed17f30
DM
2780 if (unlikely(work == weight)) {
2781 if (unlikely(napi_disable_pending(n)))
2782 __napi_complete(n);
2783 else
2784 list_move_tail(&n->poll_list, list);
2785 }
bea3348e
SH
2786
2787 netpoll_poll_unlock(have);
1da177e4
LT
2788 }
2789out:
515e06c4 2790 local_irq_enable();
bea3348e 2791
db217334
CL
2792#ifdef CONFIG_NET_DMA
2793 /*
2794 * There may not be any more sk_buffs coming right now, so push
2795 * any pending DMA copies to hardware
2796 */
2ba05622 2797 dma_issue_pending_all();
db217334 2798#endif
bea3348e 2799
1da177e4
LT
2800 return;
2801
2802softnet_break:
2803 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2804 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2805 goto out;
2806}
2807
2808static gifconf_func_t * gifconf_list [NPROTO];
2809
2810/**
2811 * register_gifconf - register a SIOCGIF handler
2812 * @family: Address family
2813 * @gifconf: Function handler
2814 *
2815 * Register protocol dependent address dumping routines. The handler
2816 * that is passed must not be freed or reused until it has been replaced
2817 * by another handler.
2818 */
2819int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2820{
2821 if (family >= NPROTO)
2822 return -EINVAL;
2823 gifconf_list[family] = gifconf;
2824 return 0;
2825}
2826
2827
2828/*
2829 * Map an interface index to its name (SIOCGIFNAME)
2830 */
2831
2832/*
2833 * We need this ioctl for efficient implementation of the
2834 * if_indextoname() function required by the IPv6 API. Without
2835 * it, we would have to search all the interfaces to find a
2836 * match. --pb
2837 */
2838
881d966b 2839static int dev_ifname(struct net *net, struct ifreq __user *arg)
1da177e4
LT
2840{
2841 struct net_device *dev;
2842 struct ifreq ifr;
2843
2844 /*
2845 * Fetch the caller's info block.
2846 */
2847
2848 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2849 return -EFAULT;
2850
2851 read_lock(&dev_base_lock);
881d966b 2852 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
1da177e4
LT
2853 if (!dev) {
2854 read_unlock(&dev_base_lock);
2855 return -ENODEV;
2856 }
2857
2858 strcpy(ifr.ifr_name, dev->name);
2859 read_unlock(&dev_base_lock);
2860
2861 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2862 return -EFAULT;
2863 return 0;
2864}
2865
2866/*
2867 * Perform a SIOCGIFCONF call. This structure will change
2868 * size eventually, and there is nothing I can do about it.
2869 * Thus we will need a 'compatibility mode'.
2870 */
2871
881d966b 2872static int dev_ifconf(struct net *net, char __user *arg)
1da177e4
LT
2873{
2874 struct ifconf ifc;
2875 struct net_device *dev;
2876 char __user *pos;
2877 int len;
2878 int total;
2879 int i;
2880
2881 /*
2882 * Fetch the caller's info block.
2883 */
2884
2885 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2886 return -EFAULT;
2887
2888 pos = ifc.ifc_buf;
2889 len = ifc.ifc_len;
2890
2891 /*
2892 * Loop over the interfaces, and write an info block for each.
2893 */
2894
2895 total = 0;
881d966b 2896 for_each_netdev(net, dev) {
1da177e4
LT
2897 for (i = 0; i < NPROTO; i++) {
2898 if (gifconf_list[i]) {
2899 int done;
2900 if (!pos)
2901 done = gifconf_list[i](dev, NULL, 0);
2902 else
2903 done = gifconf_list[i](dev, pos + total,
2904 len - total);
2905 if (done < 0)
2906 return -EFAULT;
2907 total += done;
2908 }
2909 }
4ec93edb 2910 }
1da177e4
LT
2911
2912 /*
2913 * All done. Write the updated control block back to the caller.
2914 */
2915 ifc.ifc_len = total;
2916
2917 /*
2918 * Both BSD and Solaris return 0 here, so we do too.
2919 */
2920 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2921}
2922
2923#ifdef CONFIG_PROC_FS
2924/*
2925 * This is invoked by the /proc filesystem handler to display a device
2926 * in detail.
2927 */
7562f876 2928void *dev_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2929 __acquires(dev_base_lock)
1da177e4 2930{
e372c414 2931 struct net *net = seq_file_net(seq);
7562f876 2932 loff_t off;
1da177e4 2933 struct net_device *dev;
1da177e4 2934
7562f876
PE
2935 read_lock(&dev_base_lock);
2936 if (!*pos)
2937 return SEQ_START_TOKEN;
1da177e4 2938
7562f876 2939 off = 1;
881d966b 2940 for_each_netdev(net, dev)
7562f876
PE
2941 if (off++ == *pos)
2942 return dev;
1da177e4 2943
7562f876 2944 return NULL;
1da177e4
LT
2945}
2946
2947void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2948{
e372c414 2949 struct net *net = seq_file_net(seq);
1da177e4 2950 ++*pos;
7562f876 2951 return v == SEQ_START_TOKEN ?
881d966b 2952 first_net_device(net) : next_net_device((struct net_device *)v);
1da177e4
LT
2953}
2954
2955void dev_seq_stop(struct seq_file *seq, void *v)
9a429c49 2956 __releases(dev_base_lock)
1da177e4
LT
2957{
2958 read_unlock(&dev_base_lock);
2959}
2960
2961static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2962{
eeda3fd6 2963 const struct net_device_stats *stats = dev_get_stats(dev);
1da177e4 2964
5a1b5898
RR
2965 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2966 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2967 dev->name, stats->rx_bytes, stats->rx_packets,
2968 stats->rx_errors,
2969 stats->rx_dropped + stats->rx_missed_errors,
2970 stats->rx_fifo_errors,
2971 stats->rx_length_errors + stats->rx_over_errors +
2972 stats->rx_crc_errors + stats->rx_frame_errors,
2973 stats->rx_compressed, stats->multicast,
2974 stats->tx_bytes, stats->tx_packets,
2975 stats->tx_errors, stats->tx_dropped,
2976 stats->tx_fifo_errors, stats->collisions,
2977 stats->tx_carrier_errors +
2978 stats->tx_aborted_errors +
2979 stats->tx_window_errors +
2980 stats->tx_heartbeat_errors,
2981 stats->tx_compressed);
1da177e4
LT
2982}
2983
2984/*
2985 * Called from the PROCfs module. This now uses the new arbitrary sized
2986 * /proc/net interface to create /proc/net/dev
2987 */
2988static int dev_seq_show(struct seq_file *seq, void *v)
2989{
2990 if (v == SEQ_START_TOKEN)
2991 seq_puts(seq, "Inter-| Receive "
2992 " | Transmit\n"
2993 " face |bytes packets errs drop fifo frame "
2994 "compressed multicast|bytes packets errs "
2995 "drop fifo colls carrier compressed\n");
2996 else
2997 dev_seq_printf_stats(seq, v);
2998 return 0;
2999}
3000
3001static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3002{
3003 struct netif_rx_stats *rc = NULL;
3004
0c0b0aca 3005 while (*pos < nr_cpu_ids)
4ec93edb 3006 if (cpu_online(*pos)) {
1da177e4
LT
3007 rc = &per_cpu(netdev_rx_stat, *pos);
3008 break;
3009 } else
3010 ++*pos;
3011 return rc;
3012}
3013
3014static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3015{
3016 return softnet_get_online(pos);
3017}
3018
3019static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3020{
3021 ++*pos;
3022 return softnet_get_online(pos);
3023}
3024
3025static void softnet_seq_stop(struct seq_file *seq, void *v)
3026{
3027}
3028
3029static int softnet_seq_show(struct seq_file *seq, void *v)
3030{
3031 struct netif_rx_stats *s = v;
3032
3033 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
31aa02c5 3034 s->total, s->dropped, s->time_squeeze, 0,
c1ebcdb8
SH
3035 0, 0, 0, 0, /* was fastroute */
3036 s->cpu_collision );
1da177e4
LT
3037 return 0;
3038}
3039
f690808e 3040static const struct seq_operations dev_seq_ops = {
1da177e4
LT
3041 .start = dev_seq_start,
3042 .next = dev_seq_next,
3043 .stop = dev_seq_stop,
3044 .show = dev_seq_show,
3045};
3046
3047static int dev_seq_open(struct inode *inode, struct file *file)
3048{
e372c414
DL
3049 return seq_open_net(inode, file, &dev_seq_ops,
3050 sizeof(struct seq_net_private));
1da177e4
LT
3051}
3052
9a32144e 3053static const struct file_operations dev_seq_fops = {
1da177e4
LT
3054 .owner = THIS_MODULE,
3055 .open = dev_seq_open,
3056 .read = seq_read,
3057 .llseek = seq_lseek,
e372c414 3058 .release = seq_release_net,
1da177e4
LT
3059};
3060
f690808e 3061static const struct seq_operations softnet_seq_ops = {
1da177e4
LT
3062 .start = softnet_seq_start,
3063 .next = softnet_seq_next,
3064 .stop = softnet_seq_stop,
3065 .show = softnet_seq_show,
3066};
3067
3068static int softnet_seq_open(struct inode *inode, struct file *file)
3069{
3070 return seq_open(file, &softnet_seq_ops);
3071}
3072
9a32144e 3073static const struct file_operations softnet_seq_fops = {
1da177e4
LT
3074 .owner = THIS_MODULE,
3075 .open = softnet_seq_open,
3076 .read = seq_read,
3077 .llseek = seq_lseek,
3078 .release = seq_release,
3079};
3080
0e1256ff
SH
3081static void *ptype_get_idx(loff_t pos)
3082{
3083 struct packet_type *pt = NULL;
3084 loff_t i = 0;
3085 int t;
3086
3087 list_for_each_entry_rcu(pt, &ptype_all, list) {
3088 if (i == pos)
3089 return pt;
3090 ++i;
3091 }
3092
82d8a867 3093 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
0e1256ff
SH
3094 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3095 if (i == pos)
3096 return pt;
3097 ++i;
3098 }
3099 }
3100 return NULL;
3101}
3102
3103static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
72348a42 3104 __acquires(RCU)
0e1256ff
SH
3105{
3106 rcu_read_lock();
3107 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3108}
3109
3110static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3111{
3112 struct packet_type *pt;
3113 struct list_head *nxt;
3114 int hash;
3115
3116 ++*pos;
3117 if (v == SEQ_START_TOKEN)
3118 return ptype_get_idx(0);
3119
3120 pt = v;
3121 nxt = pt->list.next;
3122 if (pt->type == htons(ETH_P_ALL)) {
3123 if (nxt != &ptype_all)
3124 goto found;
3125 hash = 0;
3126 nxt = ptype_base[0].next;
3127 } else
82d8a867 3128 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
0e1256ff
SH
3129
3130 while (nxt == &ptype_base[hash]) {
82d8a867 3131 if (++hash >= PTYPE_HASH_SIZE)
0e1256ff
SH
3132 return NULL;
3133 nxt = ptype_base[hash].next;
3134 }
3135found:
3136 return list_entry(nxt, struct packet_type, list);
3137}
3138
3139static void ptype_seq_stop(struct seq_file *seq, void *v)
72348a42 3140 __releases(RCU)
0e1256ff
SH
3141{
3142 rcu_read_unlock();
3143}
3144
0e1256ff
SH
3145static int ptype_seq_show(struct seq_file *seq, void *v)
3146{
3147 struct packet_type *pt = v;
3148
3149 if (v == SEQ_START_TOKEN)
3150 seq_puts(seq, "Type Device Function\n");
c346dca1 3151 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
0e1256ff
SH
3152 if (pt->type == htons(ETH_P_ALL))
3153 seq_puts(seq, "ALL ");
3154 else
3155 seq_printf(seq, "%04x", ntohs(pt->type));
3156
908cd2da
AD
3157 seq_printf(seq, " %-8s %pF\n",
3158 pt->dev ? pt->dev->name : "", pt->func);
0e1256ff
SH
3159 }
3160
3161 return 0;
3162}
3163
3164static const struct seq_operations ptype_seq_ops = {
3165 .start = ptype_seq_start,
3166 .next = ptype_seq_next,
3167 .stop = ptype_seq_stop,
3168 .show = ptype_seq_show,
3169};
3170
3171static int ptype_seq_open(struct inode *inode, struct file *file)
3172{
2feb27db
PE
3173 return seq_open_net(inode, file, &ptype_seq_ops,
3174 sizeof(struct seq_net_private));
0e1256ff
SH
3175}
3176
3177static const struct file_operations ptype_seq_fops = {
3178 .owner = THIS_MODULE,
3179 .open = ptype_seq_open,
3180 .read = seq_read,
3181 .llseek = seq_lseek,
2feb27db 3182 .release = seq_release_net,
0e1256ff
SH
3183};
3184
3185
4665079c 3186static int __net_init dev_proc_net_init(struct net *net)
1da177e4
LT
3187{
3188 int rc = -ENOMEM;
3189
881d966b 3190 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
1da177e4 3191 goto out;
881d966b 3192 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
1da177e4 3193 goto out_dev;
881d966b 3194 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
457c4cbc 3195 goto out_softnet;
0e1256ff 3196
881d966b 3197 if (wext_proc_init(net))
457c4cbc 3198 goto out_ptype;
1da177e4
LT
3199 rc = 0;
3200out:
3201 return rc;
457c4cbc 3202out_ptype:
881d966b 3203 proc_net_remove(net, "ptype");
1da177e4 3204out_softnet:
881d966b 3205 proc_net_remove(net, "softnet_stat");
1da177e4 3206out_dev:
881d966b 3207 proc_net_remove(net, "dev");
1da177e4
LT
3208 goto out;
3209}
881d966b 3210
4665079c 3211static void __net_exit dev_proc_net_exit(struct net *net)
881d966b
EB
3212{
3213 wext_proc_exit(net);
3214
3215 proc_net_remove(net, "ptype");
3216 proc_net_remove(net, "softnet_stat");
3217 proc_net_remove(net, "dev");
3218}
3219
022cbae6 3220static struct pernet_operations __net_initdata dev_proc_ops = {
881d966b
EB
3221 .init = dev_proc_net_init,
3222 .exit = dev_proc_net_exit,
3223};
3224
3225static int __init dev_proc_init(void)
3226{
3227 return register_pernet_subsys(&dev_proc_ops);
3228}
1da177e4
LT
3229#else
3230#define dev_proc_init() 0
3231#endif /* CONFIG_PROC_FS */
3232
3233
3234/**
3235 * netdev_set_master - set up master/slave pair
3236 * @slave: slave device
3237 * @master: new master device
3238 *
3239 * Changes the master device of the slave. Pass %NULL to break the
3240 * bonding. The caller must hold the RTNL semaphore. On a failure
3241 * a negative errno code is returned. On success the reference counts
3242 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3243 * function returns zero.
3244 */
3245int netdev_set_master(struct net_device *slave, struct net_device *master)
3246{
3247 struct net_device *old = slave->master;
3248
3249 ASSERT_RTNL();
3250
3251 if (master) {
3252 if (old)
3253 return -EBUSY;
3254 dev_hold(master);
3255 }
3256
3257 slave->master = master;
4ec93edb 3258
1da177e4
LT
3259 synchronize_net();
3260
3261 if (old)
3262 dev_put(old);
3263
3264 if (master)
3265 slave->flags |= IFF_SLAVE;
3266 else
3267 slave->flags &= ~IFF_SLAVE;
3268
3269 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3270 return 0;
3271}
3272
b6c40d68
PM
3273static void dev_change_rx_flags(struct net_device *dev, int flags)
3274{
d314774c
SH
3275 const struct net_device_ops *ops = dev->netdev_ops;
3276
3277 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3278 ops->ndo_change_rx_flags(dev, flags);
b6c40d68
PM
3279}
3280
dad9b335 3281static int __dev_set_promiscuity(struct net_device *dev, int inc)
1da177e4
LT
3282{
3283 unsigned short old_flags = dev->flags;
8192b0c4
DH
3284 uid_t uid;
3285 gid_t gid;
1da177e4 3286
24023451
PM
3287 ASSERT_RTNL();
3288
dad9b335
WC
3289 dev->flags |= IFF_PROMISC;
3290 dev->promiscuity += inc;
3291 if (dev->promiscuity == 0) {
3292 /*
3293 * Avoid overflow.
3294 * If inc causes overflow, untouch promisc and return error.
3295 */
3296 if (inc < 0)
3297 dev->flags &= ~IFF_PROMISC;
3298 else {
3299 dev->promiscuity -= inc;
3300 printk(KERN_WARNING "%s: promiscuity touches roof, "
3301 "set promiscuity failed, promiscuity feature "
3302 "of device might be broken.\n", dev->name);
3303 return -EOVERFLOW;
3304 }
3305 }
52609c0b 3306 if (dev->flags != old_flags) {
1da177e4
LT
3307 printk(KERN_INFO "device %s %s promiscuous mode\n",
3308 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4ec93edb 3309 "left");
8192b0c4
DH
3310 if (audit_enabled) {
3311 current_uid_gid(&uid, &gid);
7759db82
KHK
3312 audit_log(current->audit_context, GFP_ATOMIC,
3313 AUDIT_ANOM_PROMISCUOUS,
3314 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3315 dev->name, (dev->flags & IFF_PROMISC),
3316 (old_flags & IFF_PROMISC),
3317 audit_get_loginuid(current),
8192b0c4 3318 uid, gid,
7759db82 3319 audit_get_sessionid(current));
8192b0c4 3320 }
24023451 3321
b6c40d68 3322 dev_change_rx_flags(dev, IFF_PROMISC);
1da177e4 3323 }
dad9b335 3324 return 0;
1da177e4
LT
3325}
3326
4417da66
PM
3327/**
3328 * dev_set_promiscuity - update promiscuity count on a device
3329 * @dev: device
3330 * @inc: modifier
3331 *
3332 * Add or remove promiscuity from a device. While the count in the device
3333 * remains above zero the interface remains promiscuous. Once it hits zero
3334 * the device reverts back to normal filtering operation. A negative inc
3335 * value is used to drop promiscuity on the device.
dad9b335 3336 * Return 0 if successful or a negative errno code on error.
4417da66 3337 */
dad9b335 3338int dev_set_promiscuity(struct net_device *dev, int inc)
4417da66
PM
3339{
3340 unsigned short old_flags = dev->flags;
dad9b335 3341 int err;
4417da66 3342
dad9b335 3343 err = __dev_set_promiscuity(dev, inc);
4b5a698e 3344 if (err < 0)
dad9b335 3345 return err;
4417da66
PM
3346 if (dev->flags != old_flags)
3347 dev_set_rx_mode(dev);
dad9b335 3348 return err;
4417da66
PM
3349}
3350
1da177e4
LT
3351/**
3352 * dev_set_allmulti - update allmulti count on a device
3353 * @dev: device
3354 * @inc: modifier
3355 *
3356 * Add or remove reception of all multicast frames to a device. While the
3357 * count in the device remains above zero the interface remains listening
3358 * to all interfaces. Once it hits zero the device reverts back to normal
3359 * filtering operation. A negative @inc value is used to drop the counter
3360 * when releasing a resource needing all multicasts.
dad9b335 3361 * Return 0 if successful or a negative errno code on error.
1da177e4
LT
3362 */
3363
dad9b335 3364int dev_set_allmulti(struct net_device *dev, int inc)
1da177e4
LT
3365{
3366 unsigned short old_flags = dev->flags;
3367
24023451
PM
3368 ASSERT_RTNL();
3369
1da177e4 3370 dev->flags |= IFF_ALLMULTI;
dad9b335
WC
3371 dev->allmulti += inc;
3372 if (dev->allmulti == 0) {
3373 /*
3374 * Avoid overflow.
3375 * If inc causes overflow, untouch allmulti and return error.
3376 */
3377 if (inc < 0)
3378 dev->flags &= ~IFF_ALLMULTI;
3379 else {
3380 dev->allmulti -= inc;
3381 printk(KERN_WARNING "%s: allmulti touches roof, "
3382 "set allmulti failed, allmulti feature of "
3383 "device might be broken.\n", dev->name);
3384 return -EOVERFLOW;
3385 }
3386 }
24023451 3387 if (dev->flags ^ old_flags) {
b6c40d68 3388 dev_change_rx_flags(dev, IFF_ALLMULTI);
4417da66 3389 dev_set_rx_mode(dev);
24023451 3390 }
dad9b335 3391 return 0;
4417da66
PM
3392}
3393
3394/*
3395 * Upload unicast and multicast address lists to device and
3396 * configure RX filtering. When the device doesn't support unicast
53ccaae1 3397 * filtering it is put in promiscuous mode while unicast addresses
4417da66
PM
3398 * are present.
3399 */
3400void __dev_set_rx_mode(struct net_device *dev)
3401{
d314774c
SH
3402 const struct net_device_ops *ops = dev->netdev_ops;
3403
4417da66
PM
3404 /* dev_open will call this function so the list will stay sane. */
3405 if (!(dev->flags&IFF_UP))
3406 return;
3407
3408 if (!netif_device_present(dev))
40b77c94 3409 return;
4417da66 3410
d314774c
SH
3411 if (ops->ndo_set_rx_mode)
3412 ops->ndo_set_rx_mode(dev);
4417da66
PM
3413 else {
3414 /* Unicast addresses changes may only happen under the rtnl,
3415 * therefore calling __dev_set_promiscuity here is safe.
3416 */
3417 if (dev->uc_count > 0 && !dev->uc_promisc) {
3418 __dev_set_promiscuity(dev, 1);
3419 dev->uc_promisc = 1;
3420 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3421 __dev_set_promiscuity(dev, -1);
3422 dev->uc_promisc = 0;
3423 }
3424
d314774c
SH
3425 if (ops->ndo_set_multicast_list)
3426 ops->ndo_set_multicast_list(dev);
4417da66
PM
3427 }
3428}
3429
3430void dev_set_rx_mode(struct net_device *dev)
3431{
b9e40857 3432 netif_addr_lock_bh(dev);
4417da66 3433 __dev_set_rx_mode(dev);
b9e40857 3434 netif_addr_unlock_bh(dev);
1da177e4
LT
3435}
3436
f001fde5
JP
3437/* hw addresses list handling functions */
3438
3439static int __hw_addr_add(struct list_head *list, unsigned char *addr,
3440 int addr_len, unsigned char addr_type)
3441{
3442 struct netdev_hw_addr *ha;
3443 int alloc_size;
3444
3445 if (addr_len > MAX_ADDR_LEN)
3446 return -EINVAL;
3447
3448 alloc_size = sizeof(*ha);
3449 if (alloc_size < L1_CACHE_BYTES)
3450 alloc_size = L1_CACHE_BYTES;
3451 ha = kmalloc(alloc_size, GFP_ATOMIC);
3452 if (!ha)
3453 return -ENOMEM;
3454 memcpy(ha->addr, addr, addr_len);
3455 ha->type = addr_type;
3456 list_add_tail_rcu(&ha->list, list);
3457 return 0;
3458}
3459
3460static void ha_rcu_free(struct rcu_head *head)
3461{
3462 struct netdev_hw_addr *ha;
3463
3464 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3465 kfree(ha);
3466}
3467
3468static int __hw_addr_del_ii(struct list_head *list, unsigned char *addr,
3469 int addr_len, unsigned char addr_type,
3470 int ignore_index)
3471{
3472 struct netdev_hw_addr *ha;
3473 int i = 0;
3474
3475 list_for_each_entry(ha, list, list) {
3476 if (i++ != ignore_index &&
3477 !memcmp(ha->addr, addr, addr_len) &&
3478 (ha->type == addr_type || !addr_type)) {
3479 list_del_rcu(&ha->list);
3480 call_rcu(&ha->rcu_head, ha_rcu_free);
3481 return 0;
3482 }
3483 }
3484 return -ENOENT;
3485}
3486
3487static int __hw_addr_add_multiple_ii(struct list_head *to_list,
3488 struct list_head *from_list,
3489 int addr_len, unsigned char addr_type,
3490 int ignore_index)
3491{
3492 int err;
3493 struct netdev_hw_addr *ha, *ha2;
3494 unsigned char type;
3495
3496 list_for_each_entry(ha, from_list, list) {
3497 type = addr_type ? addr_type : ha->type;
3498 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3499 if (err)
3500 goto unroll;
3501 }
3502 return 0;
3503
3504unroll:
3505 list_for_each_entry(ha2, from_list, list) {
3506 if (ha2 == ha)
3507 break;
3508 type = addr_type ? addr_type : ha2->type;
3509 __hw_addr_del_ii(to_list, ha2->addr, addr_len, type,
3510 ignore_index);
3511 }
3512 return err;
3513}
3514
3515static void __hw_addr_del_multiple_ii(struct list_head *to_list,
3516 struct list_head *from_list,
3517 int addr_len, unsigned char addr_type,
3518 int ignore_index)
3519{
3520 struct netdev_hw_addr *ha;
3521 unsigned char type;
3522
3523 list_for_each_entry(ha, from_list, list) {
3524 type = addr_type ? addr_type : ha->type;
3525 __hw_addr_del_ii(to_list, ha->addr, addr_len, addr_type,
3526 ignore_index);
3527 }
3528}
3529
3530static void __hw_addr_flush(struct list_head *list)
3531{
3532 struct netdev_hw_addr *ha, *tmp;
3533
3534 list_for_each_entry_safe(ha, tmp, list, list) {
3535 list_del_rcu(&ha->list);
3536 call_rcu(&ha->rcu_head, ha_rcu_free);
3537 }
3538}
3539
3540/* Device addresses handling functions */
3541
3542static void dev_addr_flush(struct net_device *dev)
3543{
3544 /* rtnl_mutex must be held here */
3545
3546 __hw_addr_flush(&dev->dev_addr_list);
3547 dev->dev_addr = NULL;
3548}
3549
3550static int dev_addr_init(struct net_device *dev)
3551{
3552 unsigned char addr[MAX_ADDR_LEN];
3553 struct netdev_hw_addr *ha;
3554 int err;
3555
3556 /* rtnl_mutex must be held here */
3557
3558 INIT_LIST_HEAD(&dev->dev_addr_list);
3559 memset(addr, 0, sizeof(*addr));
3560 err = __hw_addr_add(&dev->dev_addr_list, addr, sizeof(*addr),
3561 NETDEV_HW_ADDR_T_LAN);
3562 if (!err) {
3563 /*
3564 * Get the first (previously created) address from the list
3565 * and set dev_addr pointer to this location.
3566 */
3567 ha = list_first_entry(&dev->dev_addr_list,
3568 struct netdev_hw_addr, list);
3569 dev->dev_addr = ha->addr;
3570 }
3571 return err;
3572}
3573
3574/**
3575 * dev_addr_add - Add a device address
3576 * @dev: device
3577 * @addr: address to add
3578 * @addr_type: address type
3579 *
3580 * Add a device address to the device or increase the reference count if
3581 * it already exists.
3582 *
3583 * The caller must hold the rtnl_mutex.
3584 */
3585int dev_addr_add(struct net_device *dev, unsigned char *addr,
3586 unsigned char addr_type)
3587{
3588 int err;
3589
3590 ASSERT_RTNL();
3591
3592 err = __hw_addr_add(&dev->dev_addr_list, addr, dev->addr_len,
3593 addr_type);
3594 if (!err)
3595 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3596 return err;
3597}
3598EXPORT_SYMBOL(dev_addr_add);
3599
3600/**
3601 * dev_addr_del - Release a device address.
3602 * @dev: device
3603 * @addr: address to delete
3604 * @addr_type: address type
3605 *
3606 * Release reference to a device address and remove it from the device
3607 * if the reference count drops to zero.
3608 *
3609 * The caller must hold the rtnl_mutex.
3610 */
3611int dev_addr_del(struct net_device *dev, unsigned char *addr,
3612 unsigned char addr_type)
3613{
3614 int err;
3615
3616 ASSERT_RTNL();
3617
3618 err = __hw_addr_del_ii(&dev->dev_addr_list, addr, dev->addr_len,
3619 addr_type, 0);
3620 if (!err)
3621 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3622 return err;
3623}
3624EXPORT_SYMBOL(dev_addr_del);
3625
3626/**
3627 * dev_addr_add_multiple - Add device addresses from another device
3628 * @to_dev: device to which addresses will be added
3629 * @from_dev: device from which addresses will be added
3630 * @addr_type: address type - 0 means type will be used from from_dev
3631 *
3632 * Add device addresses of the one device to another.
3633 **
3634 * The caller must hold the rtnl_mutex.
3635 */
3636int dev_addr_add_multiple(struct net_device *to_dev,
3637 struct net_device *from_dev,
3638 unsigned char addr_type)
3639{
3640 int err;
3641
3642 ASSERT_RTNL();
3643
3644 if (from_dev->addr_len != to_dev->addr_len)
3645 return -EINVAL;
3646 err = __hw_addr_add_multiple_ii(&to_dev->dev_addr_list,
3647 &from_dev->dev_addr_list,
3648 to_dev->addr_len, addr_type, 0);
3649 if (!err)
3650 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3651 return err;
3652}
3653EXPORT_SYMBOL(dev_addr_add_multiple);
3654
3655/**
3656 * dev_addr_del_multiple - Delete device addresses by another device
3657 * @to_dev: device where the addresses will be deleted
3658 * @from_dev: device by which addresses the addresses will be deleted
3659 * @addr_type: address type - 0 means type will used from from_dev
3660 *
3661 * Deletes addresses in to device by the list of addresses in from device.
3662 *
3663 * The caller must hold the rtnl_mutex.
3664 */
3665int dev_addr_del_multiple(struct net_device *to_dev,
3666 struct net_device *from_dev,
3667 unsigned char addr_type)
3668{
3669 ASSERT_RTNL();
3670
3671 if (from_dev->addr_len != to_dev->addr_len)
3672 return -EINVAL;
3673 __hw_addr_del_multiple_ii(&to_dev->dev_addr_list,
3674 &from_dev->dev_addr_list,
3675 to_dev->addr_len, addr_type, 0);
3676 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3677 return 0;
3678}
3679EXPORT_SYMBOL(dev_addr_del_multiple);
3680
3681/* unicast and multicast addresses handling functions */
3682
61cbc2fc
PM
3683int __dev_addr_delete(struct dev_addr_list **list, int *count,
3684 void *addr, int alen, int glbl)
bf742482
PM
3685{
3686 struct dev_addr_list *da;
3687
3688 for (; (da = *list) != NULL; list = &da->next) {
3689 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3690 alen == da->da_addrlen) {
3691 if (glbl) {
3692 int old_glbl = da->da_gusers;
3693 da->da_gusers = 0;
3694 if (old_glbl == 0)
3695 break;
3696 }
3697 if (--da->da_users)
3698 return 0;
3699
3700 *list = da->next;
3701 kfree(da);
61cbc2fc 3702 (*count)--;
bf742482
PM
3703 return 0;
3704 }
3705 }
3706 return -ENOENT;
3707}
3708
61cbc2fc
PM
3709int __dev_addr_add(struct dev_addr_list **list, int *count,
3710 void *addr, int alen, int glbl)
bf742482
PM
3711{
3712 struct dev_addr_list *da;
3713
3714 for (da = *list; da != NULL; da = da->next) {
3715 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3716 da->da_addrlen == alen) {
3717 if (glbl) {
3718 int old_glbl = da->da_gusers;
3719 da->da_gusers = 1;
3720 if (old_glbl)
3721 return 0;
3722 }
3723 da->da_users++;
3724 return 0;
3725 }
3726 }
3727
12aa343a 3728 da = kzalloc(sizeof(*da), GFP_ATOMIC);
bf742482
PM
3729 if (da == NULL)
3730 return -ENOMEM;
3731 memcpy(da->da_addr, addr, alen);
3732 da->da_addrlen = alen;
3733 da->da_users = 1;
3734 da->da_gusers = glbl ? 1 : 0;
3735 da->next = *list;
3736 *list = da;
61cbc2fc 3737 (*count)++;
bf742482
PM
3738 return 0;
3739}
3740
4417da66
PM
3741/**
3742 * dev_unicast_delete - Release secondary unicast address.
3743 * @dev: device
0ed72ec4
RD
3744 * @addr: address to delete
3745 * @alen: length of @addr
4417da66
PM
3746 *
3747 * Release reference to a secondary unicast address and remove it
0ed72ec4 3748 * from the device if the reference count drops to zero.
4417da66
PM
3749 *
3750 * The caller must hold the rtnl_mutex.
3751 */
3752int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3753{
3754 int err;
3755
3756 ASSERT_RTNL();
3757
b9e40857 3758 netif_addr_lock_bh(dev);
61cbc2fc
PM
3759 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3760 if (!err)
4417da66 3761 __dev_set_rx_mode(dev);
b9e40857 3762 netif_addr_unlock_bh(dev);
4417da66
PM
3763 return err;
3764}
3765EXPORT_SYMBOL(dev_unicast_delete);
3766
3767/**
3768 * dev_unicast_add - add a secondary unicast address
3769 * @dev: device
5dbaec5d 3770 * @addr: address to add
0ed72ec4 3771 * @alen: length of @addr
4417da66
PM
3772 *
3773 * Add a secondary unicast address to the device or increase
3774 * the reference count if it already exists.
3775 *
3776 * The caller must hold the rtnl_mutex.
3777 */
3778int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3779{
3780 int err;
3781
3782 ASSERT_RTNL();
3783
b9e40857 3784 netif_addr_lock_bh(dev);
61cbc2fc
PM
3785 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3786 if (!err)
4417da66 3787 __dev_set_rx_mode(dev);
b9e40857 3788 netif_addr_unlock_bh(dev);
4417da66
PM
3789 return err;
3790}
3791EXPORT_SYMBOL(dev_unicast_add);
3792
e83a2ea8
CL
3793int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3794 struct dev_addr_list **from, int *from_count)
3795{
3796 struct dev_addr_list *da, *next;
3797 int err = 0;
3798
3799 da = *from;
3800 while (da != NULL) {
3801 next = da->next;
3802 if (!da->da_synced) {
3803 err = __dev_addr_add(to, to_count,
3804 da->da_addr, da->da_addrlen, 0);
3805 if (err < 0)
3806 break;
3807 da->da_synced = 1;
3808 da->da_users++;
3809 } else if (da->da_users == 1) {
3810 __dev_addr_delete(to, to_count,
3811 da->da_addr, da->da_addrlen, 0);
3812 __dev_addr_delete(from, from_count,
3813 da->da_addr, da->da_addrlen, 0);
3814 }
3815 da = next;
3816 }
3817 return err;
3818}
3819
3820void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3821 struct dev_addr_list **from, int *from_count)
3822{
3823 struct dev_addr_list *da, *next;
3824
3825 da = *from;
3826 while (da != NULL) {
3827 next = da->next;
3828 if (da->da_synced) {
3829 __dev_addr_delete(to, to_count,
3830 da->da_addr, da->da_addrlen, 0);
3831 da->da_synced = 0;
3832 __dev_addr_delete(from, from_count,
3833 da->da_addr, da->da_addrlen, 0);
3834 }
3835 da = next;
3836 }
3837}
3838
3839/**
3840 * dev_unicast_sync - Synchronize device's unicast list to another device
3841 * @to: destination device
3842 * @from: source device
3843 *
3844 * Add newly added addresses to the destination device and release
3845 * addresses that have no users left. The source device must be
3846 * locked by netif_tx_lock_bh.
3847 *
3848 * This function is intended to be called from the dev->set_rx_mode
3849 * function of layered software devices.
3850 */
3851int dev_unicast_sync(struct net_device *to, struct net_device *from)
3852{
3853 int err = 0;
3854
b9e40857 3855 netif_addr_lock_bh(to);
e83a2ea8
CL
3856 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3857 &from->uc_list, &from->uc_count);
3858 if (!err)
3859 __dev_set_rx_mode(to);
b9e40857 3860 netif_addr_unlock_bh(to);
e83a2ea8
CL
3861 return err;
3862}
3863EXPORT_SYMBOL(dev_unicast_sync);
3864
3865/**
bc2cda1e 3866 * dev_unicast_unsync - Remove synchronized addresses from the destination device
e83a2ea8
CL
3867 * @to: destination device
3868 * @from: source device
3869 *
3870 * Remove all addresses that were added to the destination device by
3871 * dev_unicast_sync(). This function is intended to be called from the
3872 * dev->stop function of layered software devices.
3873 */
3874void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3875{
b9e40857 3876 netif_addr_lock_bh(from);
e308a5d8 3877 netif_addr_lock(to);
e83a2ea8
CL
3878
3879 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3880 &from->uc_list, &from->uc_count);
3881 __dev_set_rx_mode(to);
3882
e308a5d8 3883 netif_addr_unlock(to);
b9e40857 3884 netif_addr_unlock_bh(from);
e83a2ea8
CL
3885}
3886EXPORT_SYMBOL(dev_unicast_unsync);
3887
12972621
DC
3888static void __dev_addr_discard(struct dev_addr_list **list)
3889{
3890 struct dev_addr_list *tmp;
3891
3892 while (*list != NULL) {
3893 tmp = *list;
3894 *list = tmp->next;
3895 if (tmp->da_users > tmp->da_gusers)
3896 printk("__dev_addr_discard: address leakage! "
3897 "da_users=%d\n", tmp->da_users);
3898 kfree(tmp);
3899 }
3900}
3901
26cc2522 3902static void dev_addr_discard(struct net_device *dev)
4417da66 3903{
b9e40857 3904 netif_addr_lock_bh(dev);
26cc2522 3905
4417da66
PM
3906 __dev_addr_discard(&dev->uc_list);
3907 dev->uc_count = 0;
4417da66 3908
456ad75c
DC
3909 __dev_addr_discard(&dev->mc_list);
3910 dev->mc_count = 0;
26cc2522 3911
b9e40857 3912 netif_addr_unlock_bh(dev);
456ad75c
DC
3913}
3914
f0db275a
SH
3915/**
3916 * dev_get_flags - get flags reported to userspace
3917 * @dev: device
3918 *
3919 * Get the combination of flag bits exported through APIs to userspace.
3920 */
1da177e4
LT
3921unsigned dev_get_flags(const struct net_device *dev)
3922{
3923 unsigned flags;
3924
3925 flags = (dev->flags & ~(IFF_PROMISC |
3926 IFF_ALLMULTI |
b00055aa
SR
3927 IFF_RUNNING |
3928 IFF_LOWER_UP |
3929 IFF_DORMANT)) |
1da177e4
LT
3930 (dev->gflags & (IFF_PROMISC |
3931 IFF_ALLMULTI));
3932
b00055aa
SR
3933 if (netif_running(dev)) {
3934 if (netif_oper_up(dev))
3935 flags |= IFF_RUNNING;
3936 if (netif_carrier_ok(dev))
3937 flags |= IFF_LOWER_UP;
3938 if (netif_dormant(dev))
3939 flags |= IFF_DORMANT;
3940 }
1da177e4
LT
3941
3942 return flags;
3943}
3944
f0db275a
SH
3945/**
3946 * dev_change_flags - change device settings
3947 * @dev: device
3948 * @flags: device state flags
3949 *
3950 * Change settings on device based state flags. The flags are
3951 * in the userspace exported format.
3952 */
1da177e4
LT
3953int dev_change_flags(struct net_device *dev, unsigned flags)
3954{
7c355f53 3955 int ret, changes;
1da177e4
LT
3956 int old_flags = dev->flags;
3957
24023451
PM
3958 ASSERT_RTNL();
3959
1da177e4
LT
3960 /*
3961 * Set the flags on our device.
3962 */
3963
3964 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3965 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3966 IFF_AUTOMEDIA)) |
3967 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3968 IFF_ALLMULTI));
3969
3970 /*
3971 * Load in the correct multicast list now the flags have changed.
3972 */
3973
b6c40d68
PM
3974 if ((old_flags ^ flags) & IFF_MULTICAST)
3975 dev_change_rx_flags(dev, IFF_MULTICAST);
24023451 3976
4417da66 3977 dev_set_rx_mode(dev);
1da177e4
LT
3978
3979 /*
3980 * Have we downed the interface. We handle IFF_UP ourselves
3981 * according to user attempts to set it, rather than blindly
3982 * setting it.
3983 */
3984
3985 ret = 0;
3986 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3987 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3988
3989 if (!ret)
4417da66 3990 dev_set_rx_mode(dev);
1da177e4
LT
3991 }
3992
3993 if (dev->flags & IFF_UP &&
3994 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3995 IFF_VOLATILE)))
056925ab 3996 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1da177e4
LT
3997
3998 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3999 int inc = (flags & IFF_PROMISC) ? +1 : -1;
4000 dev->gflags ^= IFF_PROMISC;
4001 dev_set_promiscuity(dev, inc);
4002 }
4003
4004 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4005 is important. Some (broken) drivers set IFF_PROMISC, when
4006 IFF_ALLMULTI is requested not asking us and not reporting.
4007 */
4008 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4009 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
4010 dev->gflags ^= IFF_ALLMULTI;
4011 dev_set_allmulti(dev, inc);
4012 }
4013
7c355f53
TG
4014 /* Exclude state transition flags, already notified */
4015 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4016 if (changes)
4017 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
1da177e4
LT
4018
4019 return ret;
4020}
4021
f0db275a
SH
4022/**
4023 * dev_set_mtu - Change maximum transfer unit
4024 * @dev: device
4025 * @new_mtu: new transfer unit
4026 *
4027 * Change the maximum transfer size of the network device.
4028 */
1da177e4
LT
4029int dev_set_mtu(struct net_device *dev, int new_mtu)
4030{
d314774c 4031 const struct net_device_ops *ops = dev->netdev_ops;
1da177e4
LT
4032 int err;
4033
4034 if (new_mtu == dev->mtu)
4035 return 0;
4036
4037 /* MTU must be positive. */
4038 if (new_mtu < 0)
4039 return -EINVAL;
4040
4041 if (!netif_device_present(dev))
4042 return -ENODEV;
4043
4044 err = 0;
d314774c
SH
4045 if (ops->ndo_change_mtu)
4046 err = ops->ndo_change_mtu(dev, new_mtu);
1da177e4
LT
4047 else
4048 dev->mtu = new_mtu;
d314774c 4049
1da177e4 4050 if (!err && dev->flags & IFF_UP)
056925ab 4051 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
1da177e4
LT
4052 return err;
4053}
4054
f0db275a
SH
4055/**
4056 * dev_set_mac_address - Change Media Access Control Address
4057 * @dev: device
4058 * @sa: new address
4059 *
4060 * Change the hardware (MAC) address of the device
4061 */
1da177e4
LT
4062int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4063{
d314774c 4064 const struct net_device_ops *ops = dev->netdev_ops;
1da177e4
LT
4065 int err;
4066
d314774c 4067 if (!ops->ndo_set_mac_address)
1da177e4
LT
4068 return -EOPNOTSUPP;
4069 if (sa->sa_family != dev->type)
4070 return -EINVAL;
4071 if (!netif_device_present(dev))
4072 return -ENODEV;
d314774c 4073 err = ops->ndo_set_mac_address(dev, sa);
1da177e4 4074 if (!err)
056925ab 4075 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
1da177e4
LT
4076 return err;
4077}
4078
4079/*
14e3e079 4080 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
1da177e4 4081 */
14e3e079 4082static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
1da177e4
LT
4083{
4084 int err;
881d966b 4085 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
1da177e4
LT
4086
4087 if (!dev)
4088 return -ENODEV;
4089
4090 switch (cmd) {
4091 case SIOCGIFFLAGS: /* Get interface flags */
4092 ifr->ifr_flags = dev_get_flags(dev);
4093 return 0;
4094
1da177e4
LT
4095 case SIOCGIFMETRIC: /* Get the metric on the interface
4096 (currently unused) */
4097 ifr->ifr_metric = 0;
4098 return 0;
4099
1da177e4
LT
4100 case SIOCGIFMTU: /* Get the MTU of a device */
4101 ifr->ifr_mtu = dev->mtu;
4102 return 0;
4103
1da177e4
LT
4104 case SIOCGIFHWADDR:
4105 if (!dev->addr_len)
4106 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4107 else
4108 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4109 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4110 ifr->ifr_hwaddr.sa_family = dev->type;
4111 return 0;
4112
14e3e079
JG
4113 case SIOCGIFSLAVE:
4114 err = -EINVAL;
4115 break;
4116
4117 case SIOCGIFMAP:
4118 ifr->ifr_map.mem_start = dev->mem_start;
4119 ifr->ifr_map.mem_end = dev->mem_end;
4120 ifr->ifr_map.base_addr = dev->base_addr;
4121 ifr->ifr_map.irq = dev->irq;
4122 ifr->ifr_map.dma = dev->dma;
4123 ifr->ifr_map.port = dev->if_port;
4124 return 0;
4125
4126 case SIOCGIFINDEX:
4127 ifr->ifr_ifindex = dev->ifindex;
4128 return 0;
4129
4130 case SIOCGIFTXQLEN:
4131 ifr->ifr_qlen = dev->tx_queue_len;
4132 return 0;
4133
4134 default:
4135 /* dev_ioctl() should ensure this case
4136 * is never reached
4137 */
4138 WARN_ON(1);
4139 err = -EINVAL;
4140 break;
4141
4142 }
4143 return err;
4144}
4145
4146/*
4147 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4148 */
4149static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4150{
4151 int err;
4152 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
5f2f6da7 4153 const struct net_device_ops *ops;
14e3e079
JG
4154
4155 if (!dev)
4156 return -ENODEV;
4157
5f2f6da7
JP
4158 ops = dev->netdev_ops;
4159
14e3e079
JG
4160 switch (cmd) {
4161 case SIOCSIFFLAGS: /* Set interface flags */
4162 return dev_change_flags(dev, ifr->ifr_flags);
4163
4164 case SIOCSIFMETRIC: /* Set the metric on the interface
4165 (currently unused) */
4166 return -EOPNOTSUPP;
4167
4168 case SIOCSIFMTU: /* Set the MTU of a device */
4169 return dev_set_mtu(dev, ifr->ifr_mtu);
4170
1da177e4
LT
4171 case SIOCSIFHWADDR:
4172 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4173
4174 case SIOCSIFHWBROADCAST:
4175 if (ifr->ifr_hwaddr.sa_family != dev->type)
4176 return -EINVAL;
4177 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4178 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
056925ab 4179 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
1da177e4
LT
4180 return 0;
4181
1da177e4 4182 case SIOCSIFMAP:
d314774c 4183 if (ops->ndo_set_config) {
1da177e4
LT
4184 if (!netif_device_present(dev))
4185 return -ENODEV;
d314774c 4186 return ops->ndo_set_config(dev, &ifr->ifr_map);
1da177e4
LT
4187 }
4188 return -EOPNOTSUPP;
4189
4190 case SIOCADDMULTI:
d314774c 4191 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
1da177e4
LT
4192 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4193 return -EINVAL;
4194 if (!netif_device_present(dev))
4195 return -ENODEV;
4196 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4197 dev->addr_len, 1);
4198
4199 case SIOCDELMULTI:
d314774c 4200 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
1da177e4
LT
4201 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4202 return -EINVAL;
4203 if (!netif_device_present(dev))
4204 return -ENODEV;
4205 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4206 dev->addr_len, 1);
4207
1da177e4
LT
4208 case SIOCSIFTXQLEN:
4209 if (ifr->ifr_qlen < 0)
4210 return -EINVAL;
4211 dev->tx_queue_len = ifr->ifr_qlen;
4212 return 0;
4213
4214 case SIOCSIFNAME:
4215 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4216 return dev_change_name(dev, ifr->ifr_newname);
4217
4218 /*
4219 * Unknown or private ioctl
4220 */
4221
4222 default:
4223 if ((cmd >= SIOCDEVPRIVATE &&
4224 cmd <= SIOCDEVPRIVATE + 15) ||
4225 cmd == SIOCBONDENSLAVE ||
4226 cmd == SIOCBONDRELEASE ||
4227 cmd == SIOCBONDSETHWADDR ||
4228 cmd == SIOCBONDSLAVEINFOQUERY ||
4229 cmd == SIOCBONDINFOQUERY ||
4230 cmd == SIOCBONDCHANGEACTIVE ||
4231 cmd == SIOCGMIIPHY ||
4232 cmd == SIOCGMIIREG ||
4233 cmd == SIOCSMIIREG ||
4234 cmd == SIOCBRADDIF ||
4235 cmd == SIOCBRDELIF ||
d24fff22 4236 cmd == SIOCSHWTSTAMP ||
1da177e4
LT
4237 cmd == SIOCWANDEV) {
4238 err = -EOPNOTSUPP;
d314774c 4239 if (ops->ndo_do_ioctl) {
1da177e4 4240 if (netif_device_present(dev))
d314774c 4241 err = ops->ndo_do_ioctl(dev, ifr, cmd);
1da177e4
LT
4242 else
4243 err = -ENODEV;
4244 }
4245 } else
4246 err = -EINVAL;
4247
4248 }
4249 return err;
4250}
4251
4252/*
4253 * This function handles all "interface"-type I/O control requests. The actual
4254 * 'doing' part of this is dev_ifsioc above.
4255 */
4256
4257/**
4258 * dev_ioctl - network device ioctl
c4ea43c5 4259 * @net: the applicable net namespace
1da177e4
LT
4260 * @cmd: command to issue
4261 * @arg: pointer to a struct ifreq in user space
4262 *
4263 * Issue ioctl functions to devices. This is normally called by the
4264 * user space syscall interfaces but can sometimes be useful for
4265 * other purposes. The return value is the return from the syscall if
4266 * positive or a negative errno code on error.
4267 */
4268
881d966b 4269int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
1da177e4
LT
4270{
4271 struct ifreq ifr;
4272 int ret;
4273 char *colon;
4274
4275 /* One special case: SIOCGIFCONF takes ifconf argument
4276 and requires shared lock, because it sleeps writing
4277 to user space.
4278 */
4279
4280 if (cmd == SIOCGIFCONF) {
6756ae4b 4281 rtnl_lock();
881d966b 4282 ret = dev_ifconf(net, (char __user *) arg);
6756ae4b 4283 rtnl_unlock();
1da177e4
LT
4284 return ret;
4285 }
4286 if (cmd == SIOCGIFNAME)
881d966b 4287 return dev_ifname(net, (struct ifreq __user *)arg);
1da177e4
LT
4288
4289 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4290 return -EFAULT;
4291
4292 ifr.ifr_name[IFNAMSIZ-1] = 0;
4293
4294 colon = strchr(ifr.ifr_name, ':');
4295 if (colon)
4296 *colon = 0;
4297
4298 /*
4299 * See which interface the caller is talking about.
4300 */
4301
4302 switch (cmd) {
4303 /*
4304 * These ioctl calls:
4305 * - can be done by all.
4306 * - atomic and do not require locking.
4307 * - return a value
4308 */
4309 case SIOCGIFFLAGS:
4310 case SIOCGIFMETRIC:
4311 case SIOCGIFMTU:
4312 case SIOCGIFHWADDR:
4313 case SIOCGIFSLAVE:
4314 case SIOCGIFMAP:
4315 case SIOCGIFINDEX:
4316 case SIOCGIFTXQLEN:
881d966b 4317 dev_load(net, ifr.ifr_name);
1da177e4 4318 read_lock(&dev_base_lock);
14e3e079 4319 ret = dev_ifsioc_locked(net, &ifr, cmd);
1da177e4
LT
4320 read_unlock(&dev_base_lock);
4321 if (!ret) {
4322 if (colon)
4323 *colon = ':';
4324 if (copy_to_user(arg, &ifr,
4325 sizeof(struct ifreq)))
4326 ret = -EFAULT;
4327 }
4328 return ret;
4329
4330 case SIOCETHTOOL:
881d966b 4331 dev_load(net, ifr.ifr_name);
1da177e4 4332 rtnl_lock();
881d966b 4333 ret = dev_ethtool(net, &ifr);
1da177e4
LT
4334 rtnl_unlock();
4335 if (!ret) {
4336 if (colon)
4337 *colon = ':';
4338 if (copy_to_user(arg, &ifr,
4339 sizeof(struct ifreq)))
4340 ret = -EFAULT;
4341 }
4342 return ret;
4343
4344 /*
4345 * These ioctl calls:
4346 * - require superuser power.
4347 * - require strict serialization.
4348 * - return a value
4349 */
4350 case SIOCGMIIPHY:
4351 case SIOCGMIIREG:
4352 case SIOCSIFNAME:
4353 if (!capable(CAP_NET_ADMIN))
4354 return -EPERM;
881d966b 4355 dev_load(net, ifr.ifr_name);
1da177e4 4356 rtnl_lock();
881d966b 4357 ret = dev_ifsioc(net, &ifr, cmd);
1da177e4
LT
4358 rtnl_unlock();
4359 if (!ret) {
4360 if (colon)
4361 *colon = ':';
4362 if (copy_to_user(arg, &ifr,
4363 sizeof(struct ifreq)))
4364 ret = -EFAULT;
4365 }
4366 return ret;
4367
4368 /*
4369 * These ioctl calls:
4370 * - require superuser power.
4371 * - require strict serialization.
4372 * - do not return a value
4373 */
4374 case SIOCSIFFLAGS:
4375 case SIOCSIFMETRIC:
4376 case SIOCSIFMTU:
4377 case SIOCSIFMAP:
4378 case SIOCSIFHWADDR:
4379 case SIOCSIFSLAVE:
4380 case SIOCADDMULTI:
4381 case SIOCDELMULTI:
4382 case SIOCSIFHWBROADCAST:
4383 case SIOCSIFTXQLEN:
4384 case SIOCSMIIREG:
4385 case SIOCBONDENSLAVE:
4386 case SIOCBONDRELEASE:
4387 case SIOCBONDSETHWADDR:
1da177e4
LT
4388 case SIOCBONDCHANGEACTIVE:
4389 case SIOCBRADDIF:
4390 case SIOCBRDELIF:
d24fff22 4391 case SIOCSHWTSTAMP:
1da177e4
LT
4392 if (!capable(CAP_NET_ADMIN))
4393 return -EPERM;
cabcac0b
TG
4394 /* fall through */
4395 case SIOCBONDSLAVEINFOQUERY:
4396 case SIOCBONDINFOQUERY:
881d966b 4397 dev_load(net, ifr.ifr_name);
1da177e4 4398 rtnl_lock();
881d966b 4399 ret = dev_ifsioc(net, &ifr, cmd);
1da177e4
LT
4400 rtnl_unlock();
4401 return ret;
4402
4403 case SIOCGIFMEM:
4404 /* Get the per device memory space. We can add this but
4405 * currently do not support it */
4406 case SIOCSIFMEM:
4407 /* Set the per device memory buffer space.
4408 * Not applicable in our case */
4409 case SIOCSIFLINK:
4410 return -EINVAL;
4411
4412 /*
4413 * Unknown or private ioctl.
4414 */
4415 default:
4416 if (cmd == SIOCWANDEV ||
4417 (cmd >= SIOCDEVPRIVATE &&
4418 cmd <= SIOCDEVPRIVATE + 15)) {
881d966b 4419 dev_load(net, ifr.ifr_name);
1da177e4 4420 rtnl_lock();
881d966b 4421 ret = dev_ifsioc(net, &ifr, cmd);
1da177e4
LT
4422 rtnl_unlock();
4423 if (!ret && copy_to_user(arg, &ifr,
4424 sizeof(struct ifreq)))
4425 ret = -EFAULT;
4426 return ret;
4427 }
1da177e4 4428 /* Take care of Wireless Extensions */
295f4a1f 4429 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
881d966b 4430 return wext_handle_ioctl(net, &ifr, cmd, arg);
1da177e4
LT
4431 return -EINVAL;
4432 }
4433}
4434
4435
4436/**
4437 * dev_new_index - allocate an ifindex
c4ea43c5 4438 * @net: the applicable net namespace
1da177e4
LT
4439 *
4440 * Returns a suitable unique value for a new device interface
4441 * number. The caller must hold the rtnl semaphore or the
4442 * dev_base_lock to be sure it remains unique.
4443 */
881d966b 4444static int dev_new_index(struct net *net)
1da177e4
LT
4445{
4446 static int ifindex;
4447 for (;;) {
4448 if (++ifindex <= 0)
4449 ifindex = 1;
881d966b 4450 if (!__dev_get_by_index(net, ifindex))
1da177e4
LT
4451 return ifindex;
4452 }
4453}
4454
1da177e4 4455/* Delayed registration/unregisteration */
3b5b34fd 4456static LIST_HEAD(net_todo_list);
1da177e4 4457
6f05f629 4458static void net_set_todo(struct net_device *dev)
1da177e4 4459{
1da177e4 4460 list_add_tail(&dev->todo_list, &net_todo_list);
1da177e4
LT
4461}
4462
93ee31f1
DL
4463static void rollback_registered(struct net_device *dev)
4464{
4465 BUG_ON(dev_boot_phase);
4466 ASSERT_RTNL();
4467
4468 /* Some devices call without registering for initialization unwind. */
4469 if (dev->reg_state == NETREG_UNINITIALIZED) {
4470 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4471 "was registered\n", dev->name, dev);
4472
4473 WARN_ON(1);
4474 return;
4475 }
4476
4477 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4478
4479 /* If device is running, close it first. */
4480 dev_close(dev);
4481
4482 /* And unlink it from device chain. */
4483 unlist_netdevice(dev);
4484
4485 dev->reg_state = NETREG_UNREGISTERING;
4486
4487 synchronize_net();
4488
4489 /* Shutdown queueing discipline. */
4490 dev_shutdown(dev);
4491
4492
4493 /* Notify protocols, that we are about to destroy
4494 this device. They should clean all the things.
4495 */
4496 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4497
4498 /*
4499 * Flush the unicast and multicast chains
4500 */
4501 dev_addr_discard(dev);
4502
d314774c
SH
4503 if (dev->netdev_ops->ndo_uninit)
4504 dev->netdev_ops->ndo_uninit(dev);
93ee31f1
DL
4505
4506 /* Notifier chain MUST detach us from master device. */
547b792c 4507 WARN_ON(dev->master);
93ee31f1
DL
4508
4509 /* Remove entries from kobject tree */
4510 netdev_unregister_kobject(dev);
4511
4512 synchronize_net();
4513
4514 dev_put(dev);
4515}
4516
e8a0464c
DM
4517static void __netdev_init_queue_locks_one(struct net_device *dev,
4518 struct netdev_queue *dev_queue,
4519 void *_unused)
c773e847
DM
4520{
4521 spin_lock_init(&dev_queue->_xmit_lock);
cf508b12 4522 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
c773e847
DM
4523 dev_queue->xmit_lock_owner = -1;
4524}
4525
4526static void netdev_init_queue_locks(struct net_device *dev)
4527{
e8a0464c
DM
4528 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4529 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
c773e847
DM
4530}
4531
b63365a2
HX
4532unsigned long netdev_fix_features(unsigned long features, const char *name)
4533{
4534 /* Fix illegal SG+CSUM combinations. */
4535 if ((features & NETIF_F_SG) &&
4536 !(features & NETIF_F_ALL_CSUM)) {
4537 if (name)
4538 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4539 "checksum feature.\n", name);
4540 features &= ~NETIF_F_SG;
4541 }
4542
4543 /* TSO requires that SG is present as well. */
4544 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4545 if (name)
4546 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4547 "SG feature.\n", name);
4548 features &= ~NETIF_F_TSO;
4549 }
4550
4551 if (features & NETIF_F_UFO) {
4552 if (!(features & NETIF_F_GEN_CSUM)) {
4553 if (name)
4554 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4555 "since no NETIF_F_HW_CSUM feature.\n",
4556 name);
4557 features &= ~NETIF_F_UFO;
4558 }
4559
4560 if (!(features & NETIF_F_SG)) {
4561 if (name)
4562 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4563 "since no NETIF_F_SG feature.\n", name);
4564 features &= ~NETIF_F_UFO;
4565 }
4566 }
4567
4568 return features;
4569}
4570EXPORT_SYMBOL(netdev_fix_features);
4571
9d40bbda
DM
4572/* Some devices need to (re-)set their netdev_ops inside
4573 * ->init() or similar. If that happens, we have to setup
4574 * the compat pointers again.
4575 */
4576void netdev_resync_ops(struct net_device *dev)
4577{
4578#ifdef CONFIG_COMPAT_NET_DEV_OPS
4579 const struct net_device_ops *ops = dev->netdev_ops;
4580
4581 dev->init = ops->ndo_init;
4582 dev->uninit = ops->ndo_uninit;
4583 dev->open = ops->ndo_open;
4584 dev->change_rx_flags = ops->ndo_change_rx_flags;
4585 dev->set_rx_mode = ops->ndo_set_rx_mode;
4586 dev->set_multicast_list = ops->ndo_set_multicast_list;
4587 dev->set_mac_address = ops->ndo_set_mac_address;
4588 dev->validate_addr = ops->ndo_validate_addr;
4589 dev->do_ioctl = ops->ndo_do_ioctl;
4590 dev->set_config = ops->ndo_set_config;
4591 dev->change_mtu = ops->ndo_change_mtu;
4592 dev->neigh_setup = ops->ndo_neigh_setup;
4593 dev->tx_timeout = ops->ndo_tx_timeout;
4594 dev->get_stats = ops->ndo_get_stats;
4595 dev->vlan_rx_register = ops->ndo_vlan_rx_register;
4596 dev->vlan_rx_add_vid = ops->ndo_vlan_rx_add_vid;
4597 dev->vlan_rx_kill_vid = ops->ndo_vlan_rx_kill_vid;
4598#ifdef CONFIG_NET_POLL_CONTROLLER
4599 dev->poll_controller = ops->ndo_poll_controller;
4600#endif
4601#endif
4602}
4603EXPORT_SYMBOL(netdev_resync_ops);
4604
1da177e4
LT
4605/**
4606 * register_netdevice - register a network device
4607 * @dev: device to register
4608 *
4609 * Take a completed network device structure and add it to the kernel
4610 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4611 * chain. 0 is returned on success. A negative errno code is returned
4612 * on a failure to set up the device, or if the name is a duplicate.
4613 *
4614 * Callers must hold the rtnl semaphore. You may want
4615 * register_netdev() instead of this.
4616 *
4617 * BUGS:
4618 * The locking appears insufficient to guarantee two parallel registers
4619 * will not get the same name.
4620 */
4621
4622int register_netdevice(struct net_device *dev)
4623{
4624 struct hlist_head *head;
4625 struct hlist_node *p;
4626 int ret;
d314774c 4627 struct net *net = dev_net(dev);
1da177e4
LT
4628
4629 BUG_ON(dev_boot_phase);
4630 ASSERT_RTNL();
4631
b17a7c17
SH
4632 might_sleep();
4633
1da177e4
LT
4634 /* When net_device's are persistent, this will be fatal. */
4635 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
d314774c 4636 BUG_ON(!net);
1da177e4 4637
f1f28aa3 4638 spin_lock_init(&dev->addr_list_lock);
cf508b12 4639 netdev_set_addr_lockdep_class(dev);
c773e847 4640 netdev_init_queue_locks(dev);
1da177e4 4641
1da177e4
LT
4642 dev->iflink = -1;
4643
d314774c 4644#ifdef CONFIG_COMPAT_NET_DEV_OPS
eb39c57f 4645 /* Netdevice_ops API compatibility support.
d314774c
SH
4646 * This is temporary until all network devices are converted.
4647 */
4648 if (dev->netdev_ops) {
9d40bbda 4649 netdev_resync_ops(dev);
d314774c
SH
4650 } else {
4651 char drivername[64];
4652 pr_info("%s (%s): not using net_device_ops yet\n",
4653 dev->name, netdev_drivername(dev, drivername, 64));
4654
4655 /* This works only because net_device_ops and the
eb39c57f 4656 compatibility structure are the same. */
d314774c
SH
4657 dev->netdev_ops = (void *) &(dev->init);
4658 }
4659#endif
4660
1da177e4 4661 /* Init, if this function is available */
d314774c
SH
4662 if (dev->netdev_ops->ndo_init) {
4663 ret = dev->netdev_ops->ndo_init(dev);
1da177e4
LT
4664 if (ret) {
4665 if (ret > 0)
4666 ret = -EIO;
90833aa4 4667 goto out;
1da177e4
LT
4668 }
4669 }
4ec93edb 4670
1da177e4
LT
4671 if (!dev_valid_name(dev->name)) {
4672 ret = -EINVAL;
7ce1b0ed 4673 goto err_uninit;
1da177e4
LT
4674 }
4675
881d966b 4676 dev->ifindex = dev_new_index(net);
1da177e4
LT
4677 if (dev->iflink == -1)
4678 dev->iflink = dev->ifindex;
4679
4680 /* Check for existence of name */
881d966b 4681 head = dev_name_hash(net, dev->name);
1da177e4
LT
4682 hlist_for_each(p, head) {
4683 struct net_device *d
4684 = hlist_entry(p, struct net_device, name_hlist);
4685 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4686 ret = -EEXIST;
7ce1b0ed 4687 goto err_uninit;
1da177e4 4688 }
4ec93edb 4689 }
1da177e4 4690
d212f87b
SH
4691 /* Fix illegal checksum combinations */
4692 if ((dev->features & NETIF_F_HW_CSUM) &&
4693 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4694 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4695 dev->name);
4696 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4697 }
4698
4699 if ((dev->features & NETIF_F_NO_CSUM) &&
4700 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4701 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4702 dev->name);
4703 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4704 }
4705
b63365a2 4706 dev->features = netdev_fix_features(dev->features, dev->name);
1da177e4 4707
e5a4a72d
LB
4708 /* Enable software GSO if SG is supported. */
4709 if (dev->features & NETIF_F_SG)
4710 dev->features |= NETIF_F_GSO;
4711
aaf8cdc3 4712 netdev_initialize_kobject(dev);
8b41d188 4713 ret = netdev_register_kobject(dev);
b17a7c17 4714 if (ret)
7ce1b0ed 4715 goto err_uninit;
b17a7c17
SH
4716 dev->reg_state = NETREG_REGISTERED;
4717
1da177e4
LT
4718 /*
4719 * Default initial state at registry is that the
4720 * device is present.
4721 */
4722
4723 set_bit(__LINK_STATE_PRESENT, &dev->state);
4724
1da177e4 4725 dev_init_scheduler(dev);
1da177e4 4726 dev_hold(dev);
ce286d32 4727 list_netdevice(dev);
1da177e4
LT
4728
4729 /* Notify protocols, that a new device appeared. */
056925ab 4730 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
fcc5a03a 4731 ret = notifier_to_errno(ret);
93ee31f1
DL
4732 if (ret) {
4733 rollback_registered(dev);
4734 dev->reg_state = NETREG_UNREGISTERED;
4735 }
1da177e4
LT
4736
4737out:
4738 return ret;
7ce1b0ed
HX
4739
4740err_uninit:
d314774c
SH
4741 if (dev->netdev_ops->ndo_uninit)
4742 dev->netdev_ops->ndo_uninit(dev);
7ce1b0ed 4743 goto out;
1da177e4
LT
4744}
4745
937f1ba5
BH
4746/**
4747 * init_dummy_netdev - init a dummy network device for NAPI
4748 * @dev: device to init
4749 *
4750 * This takes a network device structure and initialize the minimum
4751 * amount of fields so it can be used to schedule NAPI polls without
4752 * registering a full blown interface. This is to be used by drivers
4753 * that need to tie several hardware interfaces to a single NAPI
4754 * poll scheduler due to HW limitations.
4755 */
4756int init_dummy_netdev(struct net_device *dev)
4757{
4758 /* Clear everything. Note we don't initialize spinlocks
4759 * are they aren't supposed to be taken by any of the
4760 * NAPI code and this dummy netdev is supposed to be
4761 * only ever used for NAPI polls
4762 */
4763 memset(dev, 0, sizeof(struct net_device));
4764
4765 /* make sure we BUG if trying to hit standard
4766 * register/unregister code path
4767 */
4768 dev->reg_state = NETREG_DUMMY;
4769
4770 /* initialize the ref count */
4771 atomic_set(&dev->refcnt, 1);
4772
4773 /* NAPI wants this */
4774 INIT_LIST_HEAD(&dev->napi_list);
4775
4776 /* a dummy interface is started by default */
4777 set_bit(__LINK_STATE_PRESENT, &dev->state);
4778 set_bit(__LINK_STATE_START, &dev->state);
4779
4780 return 0;
4781}
4782EXPORT_SYMBOL_GPL(init_dummy_netdev);
4783
4784
1da177e4
LT
4785/**
4786 * register_netdev - register a network device
4787 * @dev: device to register
4788 *
4789 * Take a completed network device structure and add it to the kernel
4790 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4791 * chain. 0 is returned on success. A negative errno code is returned
4792 * on a failure to set up the device, or if the name is a duplicate.
4793 *
38b4da38 4794 * This is a wrapper around register_netdevice that takes the rtnl semaphore
1da177e4
LT
4795 * and expands the device name if you passed a format string to
4796 * alloc_netdev.
4797 */
4798int register_netdev(struct net_device *dev)
4799{
4800 int err;
4801
4802 rtnl_lock();
4803
4804 /*
4805 * If the name is a format string the caller wants us to do a
4806 * name allocation.
4807 */
4808 if (strchr(dev->name, '%')) {
4809 err = dev_alloc_name(dev, dev->name);
4810 if (err < 0)
4811 goto out;
4812 }
4ec93edb 4813
1da177e4
LT
4814 err = register_netdevice(dev);
4815out:
4816 rtnl_unlock();
4817 return err;
4818}
4819EXPORT_SYMBOL(register_netdev);
4820
4821/*
4822 * netdev_wait_allrefs - wait until all references are gone.
4823 *
4824 * This is called when unregistering network devices.
4825 *
4826 * Any protocol or device that holds a reference should register
4827 * for netdevice notification, and cleanup and put back the
4828 * reference if they receive an UNREGISTER event.
4829 * We can get stuck here if buggy protocols don't correctly
4ec93edb 4830 * call dev_put.
1da177e4
LT
4831 */
4832static void netdev_wait_allrefs(struct net_device *dev)
4833{
4834 unsigned long rebroadcast_time, warning_time;
4835
4836 rebroadcast_time = warning_time = jiffies;
4837 while (atomic_read(&dev->refcnt) != 0) {
4838 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6756ae4b 4839 rtnl_lock();
1da177e4
LT
4840
4841 /* Rebroadcast unregister notification */
056925ab 4842 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
1da177e4
LT
4843
4844 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4845 &dev->state)) {
4846 /* We must not have linkwatch events
4847 * pending on unregister. If this
4848 * happens, we simply run the queue
4849 * unscheduled, resulting in a noop
4850 * for this device.
4851 */
4852 linkwatch_run_queue();
4853 }
4854
6756ae4b 4855 __rtnl_unlock();
1da177e4
LT
4856
4857 rebroadcast_time = jiffies;
4858 }
4859
4860 msleep(250);
4861
4862 if (time_after(jiffies, warning_time + 10 * HZ)) {
4863 printk(KERN_EMERG "unregister_netdevice: "
4864 "waiting for %s to become free. Usage "
4865 "count = %d\n",
4866 dev->name, atomic_read(&dev->refcnt));
4867 warning_time = jiffies;
4868 }
4869 }
4870}
4871
4872/* The sequence is:
4873 *
4874 * rtnl_lock();
4875 * ...
4876 * register_netdevice(x1);
4877 * register_netdevice(x2);
4878 * ...
4879 * unregister_netdevice(y1);
4880 * unregister_netdevice(y2);
4881 * ...
4882 * rtnl_unlock();
4883 * free_netdev(y1);
4884 * free_netdev(y2);
4885 *
58ec3b4d 4886 * We are invoked by rtnl_unlock().
1da177e4 4887 * This allows us to deal with problems:
b17a7c17 4888 * 1) We can delete sysfs objects which invoke hotplug
1da177e4
LT
4889 * without deadlocking with linkwatch via keventd.
4890 * 2) Since we run with the RTNL semaphore not held, we can sleep
4891 * safely in order to wait for the netdev refcnt to drop to zero.
58ec3b4d
HX
4892 *
4893 * We must not return until all unregister events added during
4894 * the interval the lock was held have been completed.
1da177e4 4895 */
1da177e4
LT
4896void netdev_run_todo(void)
4897{
626ab0e6 4898 struct list_head list;
1da177e4 4899
1da177e4 4900 /* Snapshot list, allow later requests */
626ab0e6 4901 list_replace_init(&net_todo_list, &list);
58ec3b4d
HX
4902
4903 __rtnl_unlock();
626ab0e6 4904
1da177e4
LT
4905 while (!list_empty(&list)) {
4906 struct net_device *dev
4907 = list_entry(list.next, struct net_device, todo_list);
4908 list_del(&dev->todo_list);
4909
b17a7c17
SH
4910 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4911 printk(KERN_ERR "network todo '%s' but state %d\n",
4912 dev->name, dev->reg_state);
4913 dump_stack();
4914 continue;
4915 }
1da177e4 4916
b17a7c17 4917 dev->reg_state = NETREG_UNREGISTERED;
1da177e4 4918
6e583ce5
SH
4919 on_each_cpu(flush_backlog, dev, 1);
4920
b17a7c17 4921 netdev_wait_allrefs(dev);
1da177e4 4922
b17a7c17
SH
4923 /* paranoia */
4924 BUG_ON(atomic_read(&dev->refcnt));
547b792c
IJ
4925 WARN_ON(dev->ip_ptr);
4926 WARN_ON(dev->ip6_ptr);
4927 WARN_ON(dev->dn_ptr);
1da177e4 4928
b17a7c17
SH
4929 if (dev->destructor)
4930 dev->destructor(dev);
9093bbb2
SH
4931
4932 /* Free network device */
4933 kobject_put(&dev->dev.kobj);
1da177e4 4934 }
1da177e4
LT
4935}
4936
eeda3fd6
SH
4937/**
4938 * dev_get_stats - get network device statistics
4939 * @dev: device to get statistics from
4940 *
4941 * Get network statistics from device. The device driver may provide
4942 * its own method by setting dev->netdev_ops->get_stats; otherwise
4943 * the internal statistics structure is used.
4944 */
4945const struct net_device_stats *dev_get_stats(struct net_device *dev)
4946 {
4947 const struct net_device_ops *ops = dev->netdev_ops;
4948
4949 if (ops->ndo_get_stats)
4950 return ops->ndo_get_stats(dev);
4951 else
4952 return &dev->stats;
c45d286e 4953}
eeda3fd6 4954EXPORT_SYMBOL(dev_get_stats);
c45d286e 4955
dc2b4847 4956static void netdev_init_one_queue(struct net_device *dev,
e8a0464c
DM
4957 struct netdev_queue *queue,
4958 void *_unused)
dc2b4847 4959{
dc2b4847
DM
4960 queue->dev = dev;
4961}
4962
bb949fbd
DM
4963static void netdev_init_queues(struct net_device *dev)
4964{
e8a0464c
DM
4965 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4966 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
c3f26a26 4967 spin_lock_init(&dev->tx_global_lock);
bb949fbd
DM
4968}
4969
1da177e4 4970/**
f25f4e44 4971 * alloc_netdev_mq - allocate network device
1da177e4
LT
4972 * @sizeof_priv: size of private data to allocate space for
4973 * @name: device name format string
4974 * @setup: callback to initialize device
f25f4e44 4975 * @queue_count: the number of subqueues to allocate
1da177e4
LT
4976 *
4977 * Allocates a struct net_device with private data area for driver use
f25f4e44
PWJ
4978 * and performs basic initialization. Also allocates subquue structs
4979 * for each queue on the device at the end of the netdevice.
1da177e4 4980 */
f25f4e44
PWJ
4981struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4982 void (*setup)(struct net_device *), unsigned int queue_count)
1da177e4 4983{
e8a0464c 4984 struct netdev_queue *tx;
1da177e4 4985 struct net_device *dev;
7943986c 4986 size_t alloc_size;
e8a0464c 4987 void *p;
1da177e4 4988
b6fe17d6
SH
4989 BUG_ON(strlen(name) >= sizeof(dev->name));
4990
fd2ea0a7 4991 alloc_size = sizeof(struct net_device);
d1643d24
AD
4992 if (sizeof_priv) {
4993 /* ensure 32-byte alignment of private area */
4994 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4995 alloc_size += sizeof_priv;
4996 }
4997 /* ensure 32-byte alignment of whole construct */
4998 alloc_size += NETDEV_ALIGN_CONST;
1da177e4 4999
31380de9 5000 p = kzalloc(alloc_size, GFP_KERNEL);
1da177e4 5001 if (!p) {
b6fe17d6 5002 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
1da177e4
LT
5003 return NULL;
5004 }
1da177e4 5005
7943986c 5006 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
e8a0464c
DM
5007 if (!tx) {
5008 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5009 "tx qdiscs.\n");
5010 kfree(p);
5011 return NULL;
5012 }
5013
1da177e4
LT
5014 dev = (struct net_device *)
5015 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
5016 dev->padded = (char *)dev - (char *)p;
c346dca1 5017 dev_net_set(dev, &init_net);
1da177e4 5018
e8a0464c
DM
5019 dev->_tx = tx;
5020 dev->num_tx_queues = queue_count;
fd2ea0a7 5021 dev->real_num_tx_queues = queue_count;
e8a0464c 5022
82cc1a7a 5023 dev->gso_max_size = GSO_MAX_SIZE;
1da177e4 5024
f001fde5 5025 dev_addr_init(dev);
bb949fbd
DM
5026 netdev_init_queues(dev);
5027
d565b0a1 5028 INIT_LIST_HEAD(&dev->napi_list);
1da177e4
LT
5029 setup(dev);
5030 strcpy(dev->name, name);
5031 return dev;
5032}
f25f4e44 5033EXPORT_SYMBOL(alloc_netdev_mq);
1da177e4
LT
5034
5035/**
5036 * free_netdev - free network device
5037 * @dev: device
5038 *
4ec93edb
YH
5039 * This function does the last stage of destroying an allocated device
5040 * interface. The reference to the device object is released.
1da177e4
LT
5041 * If this is the last reference then it will be freed.
5042 */
5043void free_netdev(struct net_device *dev)
5044{
d565b0a1
HX
5045 struct napi_struct *p, *n;
5046
f3005d7f
DL
5047 release_net(dev_net(dev));
5048
e8a0464c
DM
5049 kfree(dev->_tx);
5050
f001fde5
JP
5051 /* Flush device addresses */
5052 dev_addr_flush(dev);
5053
d565b0a1
HX
5054 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5055 netif_napi_del(p);
5056
3041a069 5057 /* Compatibility with error handling in drivers */
1da177e4
LT
5058 if (dev->reg_state == NETREG_UNINITIALIZED) {
5059 kfree((char *)dev - dev->padded);
5060 return;
5061 }
5062
5063 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5064 dev->reg_state = NETREG_RELEASED;
5065
43cb76d9
GKH
5066 /* will free via device release */
5067 put_device(&dev->dev);
1da177e4 5068}
4ec93edb 5069
f0db275a
SH
5070/**
5071 * synchronize_net - Synchronize with packet receive processing
5072 *
5073 * Wait for packets currently being received to be done.
5074 * Does not block later packets from starting.
5075 */
4ec93edb 5076void synchronize_net(void)
1da177e4
LT
5077{
5078 might_sleep();
fbd568a3 5079 synchronize_rcu();
1da177e4
LT
5080}
5081
5082/**
5083 * unregister_netdevice - remove device from the kernel
5084 * @dev: device
5085 *
5086 * This function shuts down a device interface and removes it
d59b54b1 5087 * from the kernel tables.
1da177e4
LT
5088 *
5089 * Callers must hold the rtnl semaphore. You may want
5090 * unregister_netdev() instead of this.
5091 */
5092
22f8cde5 5093void unregister_netdevice(struct net_device *dev)
1da177e4 5094{
a6620712
HX
5095 ASSERT_RTNL();
5096
93ee31f1 5097 rollback_registered(dev);
1da177e4
LT
5098 /* Finish processing unregister after unlock */
5099 net_set_todo(dev);
1da177e4
LT
5100}
5101
5102/**
5103 * unregister_netdev - remove device from the kernel
5104 * @dev: device
5105 *
5106 * This function shuts down a device interface and removes it
d59b54b1 5107 * from the kernel tables.
1da177e4
LT
5108 *
5109 * This is just a wrapper for unregister_netdevice that takes
5110 * the rtnl semaphore. In general you want to use this and not
5111 * unregister_netdevice.
5112 */
5113void unregister_netdev(struct net_device *dev)
5114{
5115 rtnl_lock();
5116 unregister_netdevice(dev);
5117 rtnl_unlock();
5118}
5119
5120EXPORT_SYMBOL(unregister_netdev);
5121
ce286d32
EB
5122/**
5123 * dev_change_net_namespace - move device to different nethost namespace
5124 * @dev: device
5125 * @net: network namespace
5126 * @pat: If not NULL name pattern to try if the current device name
5127 * is already taken in the destination network namespace.
5128 *
5129 * This function shuts down a device interface and moves it
5130 * to a new network namespace. On success 0 is returned, on
5131 * a failure a netagive errno code is returned.
5132 *
5133 * Callers must hold the rtnl semaphore.
5134 */
5135
5136int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5137{
5138 char buf[IFNAMSIZ];
5139 const char *destname;
5140 int err;
5141
5142 ASSERT_RTNL();
5143
5144 /* Don't allow namespace local devices to be moved. */
5145 err = -EINVAL;
5146 if (dev->features & NETIF_F_NETNS_LOCAL)
5147 goto out;
5148
3891845e
EB
5149#ifdef CONFIG_SYSFS
5150 /* Don't allow real devices to be moved when sysfs
5151 * is enabled.
5152 */
5153 err = -EINVAL;
5154 if (dev->dev.parent)
5155 goto out;
5156#endif
5157
ce286d32
EB
5158 /* Ensure the device has been registrered */
5159 err = -EINVAL;
5160 if (dev->reg_state != NETREG_REGISTERED)
5161 goto out;
5162
5163 /* Get out if there is nothing todo */
5164 err = 0;
878628fb 5165 if (net_eq(dev_net(dev), net))
ce286d32
EB
5166 goto out;
5167
5168 /* Pick the destination device name, and ensure
5169 * we can use it in the destination network namespace.
5170 */
5171 err = -EEXIST;
5172 destname = dev->name;
5173 if (__dev_get_by_name(net, destname)) {
5174 /* We get here if we can't use the current device name */
5175 if (!pat)
5176 goto out;
5177 if (!dev_valid_name(pat))
5178 goto out;
5179 if (strchr(pat, '%')) {
5180 if (__dev_alloc_name(net, pat, buf) < 0)
5181 goto out;
5182 destname = buf;
5183 } else
5184 destname = pat;
5185 if (__dev_get_by_name(net, destname))
5186 goto out;
5187 }
5188
5189 /*
5190 * And now a mini version of register_netdevice unregister_netdevice.
5191 */
5192
5193 /* If device is running close it first. */
9b772652 5194 dev_close(dev);
ce286d32
EB
5195
5196 /* And unlink it from device chain */
5197 err = -ENODEV;
5198 unlist_netdevice(dev);
5199
5200 synchronize_net();
5201
5202 /* Shutdown queueing discipline. */
5203 dev_shutdown(dev);
5204
5205 /* Notify protocols, that we are about to destroy
5206 this device. They should clean all the things.
5207 */
5208 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5209
5210 /*
5211 * Flush the unicast and multicast chains
5212 */
5213 dev_addr_discard(dev);
5214
3891845e
EB
5215 netdev_unregister_kobject(dev);
5216
ce286d32 5217 /* Actually switch the network namespace */
c346dca1 5218 dev_net_set(dev, net);
ce286d32
EB
5219
5220 /* Assign the new device name */
5221 if (destname != dev->name)
5222 strcpy(dev->name, destname);
5223
5224 /* If there is an ifindex conflict assign a new one */
5225 if (__dev_get_by_index(net, dev->ifindex)) {
5226 int iflink = (dev->iflink == dev->ifindex);
5227 dev->ifindex = dev_new_index(net);
5228 if (iflink)
5229 dev->iflink = dev->ifindex;
5230 }
5231
8b41d188 5232 /* Fixup kobjects */
aaf8cdc3 5233 err = netdev_register_kobject(dev);
8b41d188 5234 WARN_ON(err);
ce286d32
EB
5235
5236 /* Add the device back in the hashes */
5237 list_netdevice(dev);
5238
5239 /* Notify protocols, that a new device appeared. */
5240 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5241
5242 synchronize_net();
5243 err = 0;
5244out:
5245 return err;
5246}
5247
1da177e4
LT
5248static int dev_cpu_callback(struct notifier_block *nfb,
5249 unsigned long action,
5250 void *ocpu)
5251{
5252 struct sk_buff **list_skb;
37437bb2 5253 struct Qdisc **list_net;
1da177e4
LT
5254 struct sk_buff *skb;
5255 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5256 struct softnet_data *sd, *oldsd;
5257
8bb78442 5258 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
1da177e4
LT
5259 return NOTIFY_OK;
5260
5261 local_irq_disable();
5262 cpu = smp_processor_id();
5263 sd = &per_cpu(softnet_data, cpu);
5264 oldsd = &per_cpu(softnet_data, oldcpu);
5265
5266 /* Find end of our completion_queue. */
5267 list_skb = &sd->completion_queue;
5268 while (*list_skb)
5269 list_skb = &(*list_skb)->next;
5270 /* Append completion queue from offline CPU. */
5271 *list_skb = oldsd->completion_queue;
5272 oldsd->completion_queue = NULL;
5273
5274 /* Find end of our output_queue. */
5275 list_net = &sd->output_queue;
5276 while (*list_net)
5277 list_net = &(*list_net)->next_sched;
5278 /* Append output queue from offline CPU. */
5279 *list_net = oldsd->output_queue;
5280 oldsd->output_queue = NULL;
5281
5282 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5283 local_irq_enable();
5284
5285 /* Process offline CPU's input_pkt_queue */
5286 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5287 netif_rx(skb);
5288
5289 return NOTIFY_OK;
5290}
1da177e4
LT
5291
5292
7f353bf2 5293/**
b63365a2
HX
5294 * netdev_increment_features - increment feature set by one
5295 * @all: current feature set
5296 * @one: new feature set
5297 * @mask: mask feature set
7f353bf2
HX
5298 *
5299 * Computes a new feature set after adding a device with feature set
b63365a2
HX
5300 * @one to the master device with current feature set @all. Will not
5301 * enable anything that is off in @mask. Returns the new feature set.
7f353bf2 5302 */
b63365a2
HX
5303unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5304 unsigned long mask)
5305{
5306 /* If device needs checksumming, downgrade to it. */
5307 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5308 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5309 else if (mask & NETIF_F_ALL_CSUM) {
5310 /* If one device supports v4/v6 checksumming, set for all. */
5311 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5312 !(all & NETIF_F_GEN_CSUM)) {
5313 all &= ~NETIF_F_ALL_CSUM;
5314 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5315 }
e2a6b852 5316
b63365a2
HX
5317 /* If one device supports hw checksumming, set for all. */
5318 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5319 all &= ~NETIF_F_ALL_CSUM;
5320 all |= NETIF_F_HW_CSUM;
5321 }
5322 }
7f353bf2 5323
b63365a2 5324 one |= NETIF_F_ALL_CSUM;
7f353bf2 5325
b63365a2
HX
5326 one |= all & NETIF_F_ONE_FOR_ALL;
5327 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
5328 all |= one & mask & NETIF_F_ONE_FOR_ALL;
7f353bf2
HX
5329
5330 return all;
5331}
b63365a2 5332EXPORT_SYMBOL(netdev_increment_features);
7f353bf2 5333
30d97d35
PE
5334static struct hlist_head *netdev_create_hash(void)
5335{
5336 int i;
5337 struct hlist_head *hash;
5338
5339 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5340 if (hash != NULL)
5341 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5342 INIT_HLIST_HEAD(&hash[i]);
5343
5344 return hash;
5345}
5346
881d966b 5347/* Initialize per network namespace state */
4665079c 5348static int __net_init netdev_init(struct net *net)
881d966b 5349{
881d966b 5350 INIT_LIST_HEAD(&net->dev_base_head);
881d966b 5351
30d97d35
PE
5352 net->dev_name_head = netdev_create_hash();
5353 if (net->dev_name_head == NULL)
5354 goto err_name;
881d966b 5355
30d97d35
PE
5356 net->dev_index_head = netdev_create_hash();
5357 if (net->dev_index_head == NULL)
5358 goto err_idx;
881d966b
EB
5359
5360 return 0;
30d97d35
PE
5361
5362err_idx:
5363 kfree(net->dev_name_head);
5364err_name:
5365 return -ENOMEM;
881d966b
EB
5366}
5367
f0db275a
SH
5368/**
5369 * netdev_drivername - network driver for the device
5370 * @dev: network device
5371 * @buffer: buffer for resulting name
5372 * @len: size of buffer
5373 *
5374 * Determine network driver for device.
5375 */
cf04a4c7 5376char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6579e57b 5377{
cf04a4c7
SH
5378 const struct device_driver *driver;
5379 const struct device *parent;
6579e57b
AV
5380
5381 if (len <= 0 || !buffer)
5382 return buffer;
5383 buffer[0] = 0;
5384
5385 parent = dev->dev.parent;
5386
5387 if (!parent)
5388 return buffer;
5389
5390 driver = parent->driver;
5391 if (driver && driver->name)
5392 strlcpy(buffer, driver->name, len);
5393 return buffer;
5394}
5395
4665079c 5396static void __net_exit netdev_exit(struct net *net)
881d966b
EB
5397{
5398 kfree(net->dev_name_head);
5399 kfree(net->dev_index_head);
5400}
5401
022cbae6 5402static struct pernet_operations __net_initdata netdev_net_ops = {
881d966b
EB
5403 .init = netdev_init,
5404 .exit = netdev_exit,
5405};
5406
4665079c 5407static void __net_exit default_device_exit(struct net *net)
ce286d32 5408{
8eb79863 5409 struct net_device *dev;
ce286d32
EB
5410 /*
5411 * Push all migratable of the network devices back to the
5412 * initial network namespace
5413 */
5414 rtnl_lock();
8eb79863
EB
5415restart:
5416 for_each_netdev(net, dev) {
ce286d32 5417 int err;
aca51397 5418 char fb_name[IFNAMSIZ];
ce286d32
EB
5419
5420 /* Ignore unmoveable devices (i.e. loopback) */
5421 if (dev->features & NETIF_F_NETNS_LOCAL)
5422 continue;
5423
d0c082ce
EB
5424 /* Delete virtual devices */
5425 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5426 dev->rtnl_link_ops->dellink(dev);
8eb79863 5427 goto restart;
d0c082ce
EB
5428 }
5429
ce286d32 5430 /* Push remaing network devices to init_net */
aca51397
PE
5431 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5432 err = dev_change_net_namespace(dev, &init_net, fb_name);
ce286d32 5433 if (err) {
aca51397 5434 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
ce286d32 5435 __func__, dev->name, err);
aca51397 5436 BUG();
ce286d32 5437 }
8eb79863 5438 goto restart;
ce286d32
EB
5439 }
5440 rtnl_unlock();
5441}
5442
022cbae6 5443static struct pernet_operations __net_initdata default_device_ops = {
ce286d32
EB
5444 .exit = default_device_exit,
5445};
5446
1da177e4
LT
5447/*
5448 * Initialize the DEV module. At boot time this walks the device list and
5449 * unhooks any devices that fail to initialise (normally hardware not
5450 * present) and leaves us with a valid list of present and active devices.
5451 *
5452 */
5453
5454/*
5455 * This is called single threaded during boot, so no need
5456 * to take the rtnl semaphore.
5457 */
5458static int __init net_dev_init(void)
5459{
5460 int i, rc = -ENOMEM;
5461
5462 BUG_ON(!dev_boot_phase);
5463
1da177e4
LT
5464 if (dev_proc_init())
5465 goto out;
5466
8b41d188 5467 if (netdev_kobject_init())
1da177e4
LT
5468 goto out;
5469
5470 INIT_LIST_HEAD(&ptype_all);
82d8a867 5471 for (i = 0; i < PTYPE_HASH_SIZE; i++)
1da177e4
LT
5472 INIT_LIST_HEAD(&ptype_base[i]);
5473
881d966b
EB
5474 if (register_pernet_subsys(&netdev_net_ops))
5475 goto out;
1da177e4
LT
5476
5477 /*
5478 * Initialise the packet receive queues.
5479 */
5480
6f912042 5481 for_each_possible_cpu(i) {
1da177e4
LT
5482 struct softnet_data *queue;
5483
5484 queue = &per_cpu(softnet_data, i);
5485 skb_queue_head_init(&queue->input_pkt_queue);
1da177e4
LT
5486 queue->completion_queue = NULL;
5487 INIT_LIST_HEAD(&queue->poll_list);
bea3348e
SH
5488
5489 queue->backlog.poll = process_backlog;
5490 queue->backlog.weight = weight_p;
d565b0a1 5491 queue->backlog.gro_list = NULL;
4ae5544f 5492 queue->backlog.gro_count = 0;
1da177e4
LT
5493 }
5494
1da177e4
LT
5495 dev_boot_phase = 0;
5496
505d4f73
EB
5497 /* The loopback device is special if any other network devices
5498 * is present in a network namespace the loopback device must
5499 * be present. Since we now dynamically allocate and free the
5500 * loopback device ensure this invariant is maintained by
5501 * keeping the loopback device as the first device on the
5502 * list of network devices. Ensuring the loopback devices
5503 * is the first device that appears and the last network device
5504 * that disappears.
5505 */
5506 if (register_pernet_device(&loopback_net_ops))
5507 goto out;
5508
5509 if (register_pernet_device(&default_device_ops))
5510 goto out;
5511
962cf36c
CM
5512 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5513 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
1da177e4
LT
5514
5515 hotcpu_notifier(dev_cpu_callback, 0);
5516 dst_init();
5517 dev_mcast_init();
5518 rc = 0;
5519out:
5520 return rc;
5521}
5522
5523subsys_initcall(net_dev_init);
5524
e88721f8
KK
5525static int __init initialize_hashrnd(void)
5526{
5527 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5528 return 0;
5529}
5530
5531late_initcall_sync(initialize_hashrnd);
5532
1da177e4
LT
5533EXPORT_SYMBOL(__dev_get_by_index);
5534EXPORT_SYMBOL(__dev_get_by_name);
5535EXPORT_SYMBOL(__dev_remove_pack);
c2373ee9 5536EXPORT_SYMBOL(dev_valid_name);
1da177e4
LT
5537EXPORT_SYMBOL(dev_add_pack);
5538EXPORT_SYMBOL(dev_alloc_name);
5539EXPORT_SYMBOL(dev_close);
5540EXPORT_SYMBOL(dev_get_by_flags);
5541EXPORT_SYMBOL(dev_get_by_index);
5542EXPORT_SYMBOL(dev_get_by_name);
1da177e4
LT
5543EXPORT_SYMBOL(dev_open);
5544EXPORT_SYMBOL(dev_queue_xmit);
5545EXPORT_SYMBOL(dev_remove_pack);
5546EXPORT_SYMBOL(dev_set_allmulti);
5547EXPORT_SYMBOL(dev_set_promiscuity);
5548EXPORT_SYMBOL(dev_change_flags);
5549EXPORT_SYMBOL(dev_set_mtu);
5550EXPORT_SYMBOL(dev_set_mac_address);
5551EXPORT_SYMBOL(free_netdev);
5552EXPORT_SYMBOL(netdev_boot_setup_check);
5553EXPORT_SYMBOL(netdev_set_master);
5554EXPORT_SYMBOL(netdev_state_change);
5555EXPORT_SYMBOL(netif_receive_skb);
5556EXPORT_SYMBOL(netif_rx);
5557EXPORT_SYMBOL(register_gifconf);
5558EXPORT_SYMBOL(register_netdevice);
5559EXPORT_SYMBOL(register_netdevice_notifier);
5560EXPORT_SYMBOL(skb_checksum_help);
5561EXPORT_SYMBOL(synchronize_net);
5562EXPORT_SYMBOL(unregister_netdevice);
5563EXPORT_SYMBOL(unregister_netdevice_notifier);
5564EXPORT_SYMBOL(net_enable_timestamp);
5565EXPORT_SYMBOL(net_disable_timestamp);
5566EXPORT_SYMBOL(dev_get_flags);
5567
5568#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
5569EXPORT_SYMBOL(br_handle_frame_hook);
5570EXPORT_SYMBOL(br_fdb_get_hook);
5571EXPORT_SYMBOL(br_fdb_put_hook);
5572#endif
5573
1da177e4 5574EXPORT_SYMBOL(dev_load);
1da177e4
LT
5575
5576EXPORT_PER_CPU_SYMBOL(softnet_data);