]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - net/core/dev.c
gro: Name the GRO result enumeration type
[mirror_ubuntu-artful-kernel.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>
9cbc1cb8 129#include <trace/events/napi.h>
1da177e4 130
342709ef
PE
131#include "net-sysfs.h"
132
d565b0a1
HX
133/* Instead of increasing this, you should create a hash table. */
134#define MAX_GRO_SKBS 8
135
5d38a079
HX
136/* This should be increased if a protocol with a bigger head is added. */
137#define GRO_MAX_HEAD (MAX_HEADER + 128)
138
1da177e4
LT
139/*
140 * The list of packet types we will receive (as opposed to discard)
141 * and the routines to invoke.
142 *
143 * Why 16. Because with 16 the only overlap we get on a hash of the
144 * low nibble of the protocol value is RARP/SNAP/X.25.
145 *
146 * NOTE: That is no longer true with the addition of VLAN tags. Not
147 * sure which should go first, but I bet it won't make much
148 * difference if we are running VLANs. The good news is that
149 * this protocol won't be in the list unless compiled in, so
3041a069 150 * the average user (w/out VLANs) will not be adversely affected.
1da177e4
LT
151 * --BLG
152 *
153 * 0800 IP
154 * 8100 802.1Q VLAN
155 * 0001 802.3
156 * 0002 AX.25
157 * 0004 802.2
158 * 8035 RARP
159 * 0005 SNAP
160 * 0805 X.25
161 * 0806 ARP
162 * 8137 IPX
163 * 0009 Localtalk
164 * 86DD IPv6
165 */
166
82d8a867
PE
167#define PTYPE_HASH_SIZE (16)
168#define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
169
1da177e4 170static DEFINE_SPINLOCK(ptype_lock);
82d8a867 171static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
6b2bedc3 172static struct list_head ptype_all __read_mostly; /* Taps */
1da177e4 173
1da177e4 174/*
7562f876 175 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
1da177e4
LT
176 * semaphore.
177 *
178 * Pure readers hold dev_base_lock for reading.
179 *
180 * Writers must hold the rtnl semaphore while they loop through the
7562f876 181 * dev_base_head list, and hold dev_base_lock for writing when they do the
1da177e4
LT
182 * actual updates. This allows pure readers to access the list even
183 * while a writer is preparing to update it.
184 *
185 * To put it another way, dev_base_lock is held for writing only to
186 * protect against pure readers; the rtnl semaphore provides the
187 * protection against other writers.
188 *
189 * See, for example usages, register_netdevice() and
190 * unregister_netdevice(), which must be called with the rtnl
191 * semaphore held.
192 */
1da177e4 193DEFINE_RWLOCK(dev_base_lock);
1da177e4
LT
194EXPORT_SYMBOL(dev_base_lock);
195
881d966b 196static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
1da177e4
LT
197{
198 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
7c28bd0b 199 return &net->dev_name_head[hash & (NETDEV_HASHENTRIES - 1)];
1da177e4
LT
200}
201
881d966b 202static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
1da177e4 203{
7c28bd0b 204 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
1da177e4
LT
205}
206
ce286d32
EB
207/* Device list insertion */
208static int list_netdevice(struct net_device *dev)
209{
c346dca1 210 struct net *net = dev_net(dev);
ce286d32
EB
211
212 ASSERT_RTNL();
213
214 write_lock_bh(&dev_base_lock);
215 list_add_tail(&dev->dev_list, &net->dev_base_head);
216 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
fb699dfd
ED
217 hlist_add_head_rcu(&dev->index_hlist,
218 dev_index_hash(net, dev->ifindex));
ce286d32
EB
219 write_unlock_bh(&dev_base_lock);
220 return 0;
221}
222
fb699dfd
ED
223/* Device list removal
224 * caller must respect a RCU grace period before freeing/reusing dev
225 */
ce286d32
EB
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);
fb699dfd 234 hlist_del_rcu(&dev->index_hlist);
ce286d32
EB
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);
d1b19dff 250EXPORT_PER_CPU_SYMBOL(softnet_data);
1da177e4 251
cf508b12 252#ifdef CONFIG_LOCKDEP
723e98b7 253/*
c773e847 254 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
723e98b7
JP
255 * according to dev->type
256 */
257static const unsigned short netdev_lock_type[] =
258 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
259 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
260 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
261 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
262 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
263 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
264 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
265 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
266 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
267 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
268 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
269 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
270 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
2d91d78b 271 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
929122cd 272 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
fcb94e42 273 ARPHRD_VOID, ARPHRD_NONE};
723e98b7 274
36cbd3dc 275static const char *const netdev_lock_name[] =
723e98b7
JP
276 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
277 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
278 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
279 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
280 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
281 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
282 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
283 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
284 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
285 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
286 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
287 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
288 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
2d91d78b 289 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
929122cd 290 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
fcb94e42 291 "_xmit_VOID", "_xmit_NONE"};
723e98b7
JP
292
293static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
cf508b12 294static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
723e98b7
JP
295
296static inline unsigned short netdev_lock_pos(unsigned short dev_type)
297{
298 int i;
299
300 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
301 if (netdev_lock_type[i] == dev_type)
302 return i;
303 /* the last key is used by default */
304 return ARRAY_SIZE(netdev_lock_type) - 1;
305}
306
cf508b12
DM
307static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
308 unsigned short dev_type)
723e98b7
JP
309{
310 int i;
311
312 i = netdev_lock_pos(dev_type);
313 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
314 netdev_lock_name[i]);
315}
cf508b12
DM
316
317static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
318{
319 int i;
320
321 i = netdev_lock_pos(dev->type);
322 lockdep_set_class_and_name(&dev->addr_list_lock,
323 &netdev_addr_lock_key[i],
324 netdev_lock_name[i]);
325}
723e98b7 326#else
cf508b12
DM
327static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
328 unsigned short dev_type)
329{
330}
331static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
723e98b7
JP
332{
333}
334#endif
1da177e4
LT
335
336/*******************************************************************************
337
338 Protocol management and registration routines
339
340*******************************************************************************/
341
1da177e4
LT
342/*
343 * Add a protocol ID to the list. Now that the input handler is
344 * smarter we can dispense with all the messy stuff that used to be
345 * here.
346 *
347 * BEWARE!!! Protocol handlers, mangling input packets,
348 * MUST BE last in hash buckets and checking protocol handlers
349 * MUST start from promiscuous ptype_all chain in net_bh.
350 * It is true now, do not change it.
351 * Explanation follows: if protocol handler, mangling packet, will
352 * be the first on list, it is not able to sense, that packet
353 * is cloned and should be copied-on-write, so that it will
354 * change it and subsequent readers will get broken packet.
355 * --ANK (980803)
356 */
357
358/**
359 * dev_add_pack - add packet handler
360 * @pt: packet type declaration
361 *
362 * Add a protocol handler to the networking stack. The passed &packet_type
363 * is linked into kernel lists and may not be freed until it has been
364 * removed from the kernel lists.
365 *
4ec93edb 366 * This call does not sleep therefore it can not
1da177e4
LT
367 * guarantee all CPU's that are in middle of receiving packets
368 * will see the new packet type (until the next received packet).
369 */
370
371void dev_add_pack(struct packet_type *pt)
372{
373 int hash;
374
375 spin_lock_bh(&ptype_lock);
9be9a6b9 376 if (pt->type == htons(ETH_P_ALL))
1da177e4 377 list_add_rcu(&pt->list, &ptype_all);
9be9a6b9 378 else {
82d8a867 379 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
1da177e4
LT
380 list_add_rcu(&pt->list, &ptype_base[hash]);
381 }
382 spin_unlock_bh(&ptype_lock);
383}
d1b19dff 384EXPORT_SYMBOL(dev_add_pack);
1da177e4 385
1da177e4
LT
386/**
387 * __dev_remove_pack - remove packet handler
388 * @pt: packet type declaration
389 *
390 * Remove a protocol handler that was previously added to the kernel
391 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
392 * from the kernel lists and can be freed or reused once this function
4ec93edb 393 * returns.
1da177e4
LT
394 *
395 * The packet type might still be in use by receivers
396 * and must not be freed until after all the CPU's have gone
397 * through a quiescent state.
398 */
399void __dev_remove_pack(struct packet_type *pt)
400{
401 struct list_head *head;
402 struct packet_type *pt1;
403
404 spin_lock_bh(&ptype_lock);
405
9be9a6b9 406 if (pt->type == htons(ETH_P_ALL))
1da177e4 407 head = &ptype_all;
9be9a6b9 408 else
82d8a867 409 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
1da177e4
LT
410
411 list_for_each_entry(pt1, head, list) {
412 if (pt == pt1) {
413 list_del_rcu(&pt->list);
414 goto out;
415 }
416 }
417
418 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
419out:
420 spin_unlock_bh(&ptype_lock);
421}
d1b19dff
ED
422EXPORT_SYMBOL(__dev_remove_pack);
423
1da177e4
LT
424/**
425 * dev_remove_pack - remove packet handler
426 * @pt: packet type declaration
427 *
428 * Remove a protocol handler that was previously added to the kernel
429 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
430 * from the kernel lists and can be freed or reused once this function
431 * returns.
432 *
433 * This call sleeps to guarantee that no CPU is looking at the packet
434 * type after return.
435 */
436void dev_remove_pack(struct packet_type *pt)
437{
438 __dev_remove_pack(pt);
4ec93edb 439
1da177e4
LT
440 synchronize_net();
441}
d1b19dff 442EXPORT_SYMBOL(dev_remove_pack);
1da177e4
LT
443
444/******************************************************************************
445
446 Device Boot-time Settings Routines
447
448*******************************************************************************/
449
450/* Boot time configuration table */
451static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
452
453/**
454 * netdev_boot_setup_add - add new setup entry
455 * @name: name of the device
456 * @map: configured settings for the device
457 *
458 * Adds new setup entry to the dev_boot_setup list. The function
459 * returns 0 on error and 1 on success. This is a generic routine to
460 * all netdevices.
461 */
462static int netdev_boot_setup_add(char *name, struct ifmap *map)
463{
464 struct netdev_boot_setup *s;
465 int i;
466
467 s = dev_boot_setup;
468 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
469 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
470 memset(s[i].name, 0, sizeof(s[i].name));
93b3cff9 471 strlcpy(s[i].name, name, IFNAMSIZ);
1da177e4
LT
472 memcpy(&s[i].map, map, sizeof(s[i].map));
473 break;
474 }
475 }
476
477 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
478}
479
480/**
481 * netdev_boot_setup_check - check boot time settings
482 * @dev: the netdevice
483 *
484 * Check boot time settings for the device.
485 * The found settings are set for the device to be used
486 * later in the device probing.
487 * Returns 0 if no settings found, 1 if they are.
488 */
489int netdev_boot_setup_check(struct net_device *dev)
490{
491 struct netdev_boot_setup *s = dev_boot_setup;
492 int i;
493
494 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
495 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
93b3cff9 496 !strcmp(dev->name, s[i].name)) {
1da177e4
LT
497 dev->irq = s[i].map.irq;
498 dev->base_addr = s[i].map.base_addr;
499 dev->mem_start = s[i].map.mem_start;
500 dev->mem_end = s[i].map.mem_end;
501 return 1;
502 }
503 }
504 return 0;
505}
d1b19dff 506EXPORT_SYMBOL(netdev_boot_setup_check);
1da177e4
LT
507
508
509/**
510 * netdev_boot_base - get address from boot time settings
511 * @prefix: prefix for network device
512 * @unit: id for network device
513 *
514 * Check boot time settings for the base address of device.
515 * The found settings are set for the device to be used
516 * later in the device probing.
517 * Returns 0 if no settings found.
518 */
519unsigned long netdev_boot_base(const char *prefix, int unit)
520{
521 const struct netdev_boot_setup *s = dev_boot_setup;
522 char name[IFNAMSIZ];
523 int i;
524
525 sprintf(name, "%s%d", prefix, unit);
526
527 /*
528 * If device already registered then return base of 1
529 * to indicate not to probe for this interface
530 */
881d966b 531 if (__dev_get_by_name(&init_net, name))
1da177e4
LT
532 return 1;
533
534 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
535 if (!strcmp(name, s[i].name))
536 return s[i].map.base_addr;
537 return 0;
538}
539
540/*
541 * Saves at boot time configured settings for any netdevice.
542 */
543int __init netdev_boot_setup(char *str)
544{
545 int ints[5];
546 struct ifmap map;
547
548 str = get_options(str, ARRAY_SIZE(ints), ints);
549 if (!str || !*str)
550 return 0;
551
552 /* Save settings */
553 memset(&map, 0, sizeof(map));
554 if (ints[0] > 0)
555 map.irq = ints[1];
556 if (ints[0] > 1)
557 map.base_addr = ints[2];
558 if (ints[0] > 2)
559 map.mem_start = ints[3];
560 if (ints[0] > 3)
561 map.mem_end = ints[4];
562
563 /* Add new entry to the list */
564 return netdev_boot_setup_add(str, &map);
565}
566
567__setup("netdev=", netdev_boot_setup);
568
569/*******************************************************************************
570
571 Device Interface Subroutines
572
573*******************************************************************************/
574
575/**
576 * __dev_get_by_name - find a device by its name
c4ea43c5 577 * @net: the applicable net namespace
1da177e4
LT
578 * @name: name to find
579 *
580 * Find an interface by name. Must be called under RTNL semaphore
581 * or @dev_base_lock. If the name is found a pointer to the device
582 * is returned. If the name is not found then %NULL is returned. The
583 * reference counters are not incremented so the caller must be
584 * careful with locks.
585 */
586
881d966b 587struct net_device *__dev_get_by_name(struct net *net, const char *name)
1da177e4
LT
588{
589 struct hlist_node *p;
590
881d966b 591 hlist_for_each(p, dev_name_hash(net, name)) {
1da177e4
LT
592 struct net_device *dev
593 = hlist_entry(p, struct net_device, name_hlist);
594 if (!strncmp(dev->name, name, IFNAMSIZ))
595 return dev;
596 }
597 return NULL;
598}
d1b19dff 599EXPORT_SYMBOL(__dev_get_by_name);
1da177e4
LT
600
601/**
602 * dev_get_by_name - find a device by its name
c4ea43c5 603 * @net: the applicable net namespace
1da177e4
LT
604 * @name: name to find
605 *
606 * Find an interface by name. This can be called from any
607 * context and does its own locking. The returned handle has
608 * the usage count incremented and the caller must use dev_put() to
609 * release it when it is no longer needed. %NULL is returned if no
610 * matching device is found.
611 */
612
881d966b 613struct net_device *dev_get_by_name(struct net *net, const char *name)
1da177e4
LT
614{
615 struct net_device *dev;
616
617 read_lock(&dev_base_lock);
881d966b 618 dev = __dev_get_by_name(net, name);
1da177e4
LT
619 if (dev)
620 dev_hold(dev);
621 read_unlock(&dev_base_lock);
622 return dev;
623}
d1b19dff 624EXPORT_SYMBOL(dev_get_by_name);
1da177e4
LT
625
626/**
627 * __dev_get_by_index - find a device by its ifindex
c4ea43c5 628 * @net: the applicable net namespace
1da177e4
LT
629 * @ifindex: index of device
630 *
631 * Search for an interface by index. Returns %NULL if the device
632 * is not found or a pointer to the device. The device has not
633 * had its reference counter increased so the caller must be careful
634 * about locking. The caller must hold either the RTNL semaphore
635 * or @dev_base_lock.
636 */
637
881d966b 638struct net_device *__dev_get_by_index(struct net *net, int ifindex)
1da177e4
LT
639{
640 struct hlist_node *p;
641
881d966b 642 hlist_for_each(p, dev_index_hash(net, ifindex)) {
1da177e4
LT
643 struct net_device *dev
644 = hlist_entry(p, struct net_device, index_hlist);
645 if (dev->ifindex == ifindex)
646 return dev;
647 }
648 return NULL;
649}
d1b19dff 650EXPORT_SYMBOL(__dev_get_by_index);
1da177e4 651
fb699dfd
ED
652/**
653 * dev_get_by_index_rcu - find a device by its ifindex
654 * @net: the applicable net namespace
655 * @ifindex: index of device
656 *
657 * Search for an interface by index. Returns %NULL if the device
658 * is not found or a pointer to the device. The device has not
659 * had its reference counter increased so the caller must be careful
660 * about locking. The caller must hold RCU lock.
661 */
662
663struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
664{
665 struct hlist_node *p;
666 struct net_device *dev;
667 struct hlist_head *head = dev_index_hash(net, ifindex);
668
669 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
670 if (dev->ifindex == ifindex)
671 return dev;
672
673 return NULL;
674}
675EXPORT_SYMBOL(dev_get_by_index_rcu);
676
1da177e4
LT
677
678/**
679 * dev_get_by_index - find a device by its ifindex
c4ea43c5 680 * @net: the applicable net namespace
1da177e4
LT
681 * @ifindex: index of device
682 *
683 * Search for an interface by index. Returns NULL if the device
684 * is not found or a pointer to the device. The device returned has
685 * had a reference added and the pointer is safe until the user calls
686 * dev_put to indicate they have finished with it.
687 */
688
881d966b 689struct net_device *dev_get_by_index(struct net *net, int ifindex)
1da177e4
LT
690{
691 struct net_device *dev;
692
fb699dfd
ED
693 rcu_read_lock();
694 dev = dev_get_by_index_rcu(net, ifindex);
1da177e4
LT
695 if (dev)
696 dev_hold(dev);
fb699dfd 697 rcu_read_unlock();
1da177e4
LT
698 return dev;
699}
d1b19dff 700EXPORT_SYMBOL(dev_get_by_index);
1da177e4
LT
701
702/**
703 * dev_getbyhwaddr - find a device by its hardware address
c4ea43c5 704 * @net: the applicable net namespace
1da177e4
LT
705 * @type: media type of device
706 * @ha: hardware address
707 *
708 * Search for an interface by MAC address. Returns NULL if the device
709 * is not found or a pointer to the device. The caller must hold the
710 * rtnl semaphore. The returned device has not had its ref count increased
711 * and the caller must therefore be careful about locking
712 *
713 * BUGS:
714 * If the API was consistent this would be __dev_get_by_hwaddr
715 */
716
881d966b 717struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
1da177e4
LT
718{
719 struct net_device *dev;
720
721 ASSERT_RTNL();
722
81103a52 723 for_each_netdev(net, dev)
1da177e4
LT
724 if (dev->type == type &&
725 !memcmp(dev->dev_addr, ha, dev->addr_len))
7562f876
PE
726 return dev;
727
728 return NULL;
1da177e4 729}
cf309e3f
JF
730EXPORT_SYMBOL(dev_getbyhwaddr);
731
881d966b 732struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
1da177e4
LT
733{
734 struct net_device *dev;
735
4e9cac2b 736 ASSERT_RTNL();
881d966b 737 for_each_netdev(net, dev)
4e9cac2b 738 if (dev->type == type)
7562f876
PE
739 return dev;
740
741 return NULL;
4e9cac2b 742}
4e9cac2b
PM
743EXPORT_SYMBOL(__dev_getfirstbyhwtype);
744
881d966b 745struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
4e9cac2b
PM
746{
747 struct net_device *dev;
748
749 rtnl_lock();
881d966b 750 dev = __dev_getfirstbyhwtype(net, type);
4e9cac2b
PM
751 if (dev)
752 dev_hold(dev);
1da177e4
LT
753 rtnl_unlock();
754 return dev;
755}
1da177e4
LT
756EXPORT_SYMBOL(dev_getfirstbyhwtype);
757
758/**
759 * dev_get_by_flags - find any device with given flags
c4ea43c5 760 * @net: the applicable net namespace
1da177e4
LT
761 * @if_flags: IFF_* values
762 * @mask: bitmask of bits in if_flags to check
763 *
764 * Search for any interface with the given flags. Returns NULL if a device
4ec93edb 765 * is not found or a pointer to the device. The device returned has
1da177e4
LT
766 * had a reference added and the pointer is safe until the user calls
767 * dev_put to indicate they have finished with it.
768 */
769
d1b19dff
ED
770struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
771 unsigned short mask)
1da177e4 772{
7562f876 773 struct net_device *dev, *ret;
1da177e4 774
7562f876 775 ret = NULL;
1da177e4 776 read_lock(&dev_base_lock);
881d966b 777 for_each_netdev(net, dev) {
1da177e4
LT
778 if (((dev->flags ^ if_flags) & mask) == 0) {
779 dev_hold(dev);
7562f876 780 ret = dev;
1da177e4
LT
781 break;
782 }
783 }
784 read_unlock(&dev_base_lock);
7562f876 785 return ret;
1da177e4 786}
d1b19dff 787EXPORT_SYMBOL(dev_get_by_flags);
1da177e4
LT
788
789/**
790 * dev_valid_name - check if name is okay for network device
791 * @name: name string
792 *
793 * Network device names need to be valid file names to
c7fa9d18
DM
794 * to allow sysfs to work. We also disallow any kind of
795 * whitespace.
1da177e4 796 */
c2373ee9 797int dev_valid_name(const char *name)
1da177e4 798{
c7fa9d18
DM
799 if (*name == '\0')
800 return 0;
b6fe17d6
SH
801 if (strlen(name) >= IFNAMSIZ)
802 return 0;
c7fa9d18
DM
803 if (!strcmp(name, ".") || !strcmp(name, ".."))
804 return 0;
805
806 while (*name) {
807 if (*name == '/' || isspace(*name))
808 return 0;
809 name++;
810 }
811 return 1;
1da177e4 812}
d1b19dff 813EXPORT_SYMBOL(dev_valid_name);
1da177e4
LT
814
815/**
b267b179
EB
816 * __dev_alloc_name - allocate a name for a device
817 * @net: network namespace to allocate the device name in
1da177e4 818 * @name: name format string
b267b179 819 * @buf: scratch buffer and result name string
1da177e4
LT
820 *
821 * Passed a format string - eg "lt%d" it will try and find a suitable
3041a069
SH
822 * id. It scans list of devices to build up a free map, then chooses
823 * the first empty slot. The caller must hold the dev_base or rtnl lock
824 * while allocating the name and adding the device in order to avoid
825 * duplicates.
826 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
827 * Returns the number of the unit assigned or a negative errno code.
1da177e4
LT
828 */
829
b267b179 830static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1da177e4
LT
831{
832 int i = 0;
1da177e4
LT
833 const char *p;
834 const int max_netdevices = 8*PAGE_SIZE;
cfcabdcc 835 unsigned long *inuse;
1da177e4
LT
836 struct net_device *d;
837
838 p = strnchr(name, IFNAMSIZ-1, '%');
839 if (p) {
840 /*
841 * Verify the string as this thing may have come from
842 * the user. There must be either one "%d" and no other "%"
843 * characters.
844 */
845 if (p[1] != 'd' || strchr(p + 2, '%'))
846 return -EINVAL;
847
848 /* Use one page as a bit array of possible slots */
cfcabdcc 849 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1da177e4
LT
850 if (!inuse)
851 return -ENOMEM;
852
881d966b 853 for_each_netdev(net, d) {
1da177e4
LT
854 if (!sscanf(d->name, name, &i))
855 continue;
856 if (i < 0 || i >= max_netdevices)
857 continue;
858
859 /* avoid cases where sscanf is not exact inverse of printf */
b267b179 860 snprintf(buf, IFNAMSIZ, name, i);
1da177e4
LT
861 if (!strncmp(buf, d->name, IFNAMSIZ))
862 set_bit(i, inuse);
863 }
864
865 i = find_first_zero_bit(inuse, max_netdevices);
866 free_page((unsigned long) inuse);
867 }
868
b267b179
EB
869 snprintf(buf, IFNAMSIZ, name, i);
870 if (!__dev_get_by_name(net, buf))
1da177e4 871 return i;
1da177e4
LT
872
873 /* It is possible to run out of possible slots
874 * when the name is long and there isn't enough space left
875 * for the digits, or if all bits are used.
876 */
877 return -ENFILE;
878}
879
b267b179
EB
880/**
881 * dev_alloc_name - allocate a name for a device
882 * @dev: device
883 * @name: name format string
884 *
885 * Passed a format string - eg "lt%d" it will try and find a suitable
886 * id. It scans list of devices to build up a free map, then chooses
887 * the first empty slot. The caller must hold the dev_base or rtnl lock
888 * while allocating the name and adding the device in order to avoid
889 * duplicates.
890 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
891 * Returns the number of the unit assigned or a negative errno code.
892 */
893
894int dev_alloc_name(struct net_device *dev, const char *name)
895{
896 char buf[IFNAMSIZ];
897 struct net *net;
898 int ret;
899
c346dca1
YH
900 BUG_ON(!dev_net(dev));
901 net = dev_net(dev);
b267b179
EB
902 ret = __dev_alloc_name(net, name, buf);
903 if (ret >= 0)
904 strlcpy(dev->name, buf, IFNAMSIZ);
905 return ret;
906}
d1b19dff 907EXPORT_SYMBOL(dev_alloc_name);
b267b179 908
1da177e4
LT
909
910/**
911 * dev_change_name - change name of a device
912 * @dev: device
913 * @newname: name (or format string) must be at least IFNAMSIZ
914 *
915 * Change name of a device, can pass format strings "eth%d".
916 * for wildcarding.
917 */
cf04a4c7 918int dev_change_name(struct net_device *dev, const char *newname)
1da177e4 919{
fcc5a03a 920 char oldname[IFNAMSIZ];
1da177e4 921 int err = 0;
fcc5a03a 922 int ret;
881d966b 923 struct net *net;
1da177e4
LT
924
925 ASSERT_RTNL();
c346dca1 926 BUG_ON(!dev_net(dev));
1da177e4 927
c346dca1 928 net = dev_net(dev);
1da177e4
LT
929 if (dev->flags & IFF_UP)
930 return -EBUSY;
931
932 if (!dev_valid_name(newname))
933 return -EINVAL;
934
c8d90dca
SH
935 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
936 return 0;
937
fcc5a03a
HX
938 memcpy(oldname, dev->name, IFNAMSIZ);
939
1da177e4
LT
940 if (strchr(newname, '%')) {
941 err = dev_alloc_name(dev, newname);
942 if (err < 0)
943 return err;
d1b19dff 944 } else if (__dev_get_by_name(net, newname))
1da177e4
LT
945 return -EEXIST;
946 else
947 strlcpy(dev->name, newname, IFNAMSIZ);
948
fcc5a03a 949rollback:
3891845e
EB
950 /* For now only devices in the initial network namespace
951 * are in sysfs.
952 */
953 if (net == &init_net) {
954 ret = device_rename(&dev->dev, dev->name);
955 if (ret) {
956 memcpy(dev->name, oldname, IFNAMSIZ);
957 return ret;
958 }
dcc99773 959 }
7f988eab
HX
960
961 write_lock_bh(&dev_base_lock);
92749821 962 hlist_del(&dev->name_hlist);
881d966b 963 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
7f988eab
HX
964 write_unlock_bh(&dev_base_lock);
965
056925ab 966 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
fcc5a03a
HX
967 ret = notifier_to_errno(ret);
968
969 if (ret) {
970 if (err) {
971 printk(KERN_ERR
972 "%s: name change rollback failed: %d.\n",
973 dev->name, ret);
974 } else {
975 err = ret;
976 memcpy(dev->name, oldname, IFNAMSIZ);
977 goto rollback;
978 }
979 }
1da177e4
LT
980
981 return err;
982}
983
0b815a1a
SH
984/**
985 * dev_set_alias - change ifalias of a device
986 * @dev: device
987 * @alias: name up to IFALIASZ
f0db275a 988 * @len: limit of bytes to copy from info
0b815a1a
SH
989 *
990 * Set ifalias for a device,
991 */
992int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
993{
994 ASSERT_RTNL();
995
996 if (len >= IFALIASZ)
997 return -EINVAL;
998
96ca4a2c
OH
999 if (!len) {
1000 if (dev->ifalias) {
1001 kfree(dev->ifalias);
1002 dev->ifalias = NULL;
1003 }
1004 return 0;
1005 }
1006
d1b19dff 1007 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
0b815a1a
SH
1008 if (!dev->ifalias)
1009 return -ENOMEM;
1010
1011 strlcpy(dev->ifalias, alias, len+1);
1012 return len;
1013}
1014
1015
d8a33ac4 1016/**
3041a069 1017 * netdev_features_change - device changes features
d8a33ac4
SH
1018 * @dev: device to cause notification
1019 *
1020 * Called to indicate a device has changed features.
1021 */
1022void netdev_features_change(struct net_device *dev)
1023{
056925ab 1024 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
d8a33ac4
SH
1025}
1026EXPORT_SYMBOL(netdev_features_change);
1027
1da177e4
LT
1028/**
1029 * netdev_state_change - device changes state
1030 * @dev: device to cause notification
1031 *
1032 * Called to indicate a device has changed state. This function calls
1033 * the notifier chains for netdev_chain and sends a NEWLINK message
1034 * to the routing socket.
1035 */
1036void netdev_state_change(struct net_device *dev)
1037{
1038 if (dev->flags & IFF_UP) {
056925ab 1039 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1da177e4
LT
1040 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1041 }
1042}
d1b19dff 1043EXPORT_SYMBOL(netdev_state_change);
1da177e4 1044
75c78500 1045void netdev_bonding_change(struct net_device *dev, unsigned long event)
c1da4ac7 1046{
75c78500 1047 call_netdevice_notifiers(event, dev);
c1da4ac7
OG
1048}
1049EXPORT_SYMBOL(netdev_bonding_change);
1050
1da177e4
LT
1051/**
1052 * dev_load - load a network module
c4ea43c5 1053 * @net: the applicable net namespace
1da177e4
LT
1054 * @name: name of interface
1055 *
1056 * If a network interface is not present and the process has suitable
1057 * privileges this function loads the module. If module loading is not
1058 * available in this kernel then it becomes a nop.
1059 */
1060
881d966b 1061void dev_load(struct net *net, const char *name)
1da177e4 1062{
4ec93edb 1063 struct net_device *dev;
1da177e4
LT
1064
1065 read_lock(&dev_base_lock);
881d966b 1066 dev = __dev_get_by_name(net, name);
1da177e4
LT
1067 read_unlock(&dev_base_lock);
1068
a8f80e8f 1069 if (!dev && capable(CAP_NET_ADMIN))
1da177e4
LT
1070 request_module("%s", name);
1071}
d1b19dff 1072EXPORT_SYMBOL(dev_load);
1da177e4 1073
1da177e4
LT
1074/**
1075 * dev_open - prepare an interface for use.
1076 * @dev: device to open
1077 *
1078 * Takes a device from down to up state. The device's private open
1079 * function is invoked and then the multicast lists are loaded. Finally
1080 * the device is moved into the up state and a %NETDEV_UP message is
1081 * sent to the netdev notifier chain.
1082 *
1083 * Calling this function on an active interface is a nop. On a failure
1084 * a negative errno code is returned.
1085 */
1086int dev_open(struct net_device *dev)
1087{
d314774c 1088 const struct net_device_ops *ops = dev->netdev_ops;
3b8bcfd5 1089 int ret;
1da177e4 1090
e46b66bc
BH
1091 ASSERT_RTNL();
1092
1da177e4
LT
1093 /*
1094 * Is it already up?
1095 */
1096
1097 if (dev->flags & IFF_UP)
1098 return 0;
1099
1100 /*
1101 * Is it even present?
1102 */
1103 if (!netif_device_present(dev))
1104 return -ENODEV;
1105
3b8bcfd5
JB
1106 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1107 ret = notifier_to_errno(ret);
1108 if (ret)
1109 return ret;
1110
1da177e4
LT
1111 /*
1112 * Call device private open method
1113 */
1114 set_bit(__LINK_STATE_START, &dev->state);
bada339b 1115
d314774c
SH
1116 if (ops->ndo_validate_addr)
1117 ret = ops->ndo_validate_addr(dev);
bada339b 1118
d314774c
SH
1119 if (!ret && ops->ndo_open)
1120 ret = ops->ndo_open(dev);
1da177e4 1121
4ec93edb 1122 /*
1da177e4
LT
1123 * If it went open OK then:
1124 */
1125
bada339b
JG
1126 if (ret)
1127 clear_bit(__LINK_STATE_START, &dev->state);
1128 else {
1da177e4
LT
1129 /*
1130 * Set the flags.
1131 */
1132 dev->flags |= IFF_UP;
1133
649274d9
DW
1134 /*
1135 * Enable NET_DMA
1136 */
b4bd07c2 1137 net_dmaengine_get();
649274d9 1138
1da177e4
LT
1139 /*
1140 * Initialize multicasting status
1141 */
4417da66 1142 dev_set_rx_mode(dev);
1da177e4
LT
1143
1144 /*
1145 * Wakeup transmit queue engine
1146 */
1147 dev_activate(dev);
1148
1149 /*
1150 * ... and announce new interface.
1151 */
056925ab 1152 call_netdevice_notifiers(NETDEV_UP, dev);
1da177e4 1153 }
bada339b 1154
1da177e4
LT
1155 return ret;
1156}
d1b19dff 1157EXPORT_SYMBOL(dev_open);
1da177e4
LT
1158
1159/**
1160 * dev_close - shutdown an interface.
1161 * @dev: device to shutdown
1162 *
1163 * This function moves an active device into down state. A
1164 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1165 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1166 * chain.
1167 */
1168int dev_close(struct net_device *dev)
1169{
d314774c 1170 const struct net_device_ops *ops = dev->netdev_ops;
e46b66bc
BH
1171 ASSERT_RTNL();
1172
9d5010db
DM
1173 might_sleep();
1174
1da177e4
LT
1175 if (!(dev->flags & IFF_UP))
1176 return 0;
1177
1178 /*
1179 * Tell people we are going down, so that they can
1180 * prepare to death, when device is still operating.
1181 */
056925ab 1182 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1da177e4 1183
1da177e4
LT
1184 clear_bit(__LINK_STATE_START, &dev->state);
1185
1186 /* Synchronize to scheduled poll. We cannot touch poll list,
bea3348e
SH
1187 * it can be even on different cpu. So just clear netif_running().
1188 *
1189 * dev->stop() will invoke napi_disable() on all of it's
1190 * napi_struct instances on this device.
1191 */
1da177e4 1192 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1da177e4 1193
d8b2a4d2
ML
1194 dev_deactivate(dev);
1195
1da177e4
LT
1196 /*
1197 * Call the device specific close. This cannot fail.
1198 * Only if device is UP
1199 *
1200 * We allow it to be called even after a DETACH hot-plug
1201 * event.
1202 */
d314774c
SH
1203 if (ops->ndo_stop)
1204 ops->ndo_stop(dev);
1da177e4
LT
1205
1206 /*
1207 * Device is now down.
1208 */
1209
1210 dev->flags &= ~IFF_UP;
1211
1212 /*
1213 * Tell people we are down
1214 */
056925ab 1215 call_netdevice_notifiers(NETDEV_DOWN, dev);
1da177e4 1216
649274d9
DW
1217 /*
1218 * Shutdown NET_DMA
1219 */
b4bd07c2 1220 net_dmaengine_put();
649274d9 1221
1da177e4
LT
1222 return 0;
1223}
d1b19dff 1224EXPORT_SYMBOL(dev_close);
1da177e4
LT
1225
1226
0187bdfb
BH
1227/**
1228 * dev_disable_lro - disable Large Receive Offload on a device
1229 * @dev: device
1230 *
1231 * Disable Large Receive Offload (LRO) on a net device. Must be
1232 * called under RTNL. This is needed if received packets may be
1233 * forwarded to another interface.
1234 */
1235void dev_disable_lro(struct net_device *dev)
1236{
1237 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1238 dev->ethtool_ops->set_flags) {
1239 u32 flags = dev->ethtool_ops->get_flags(dev);
1240 if (flags & ETH_FLAG_LRO) {
1241 flags &= ~ETH_FLAG_LRO;
1242 dev->ethtool_ops->set_flags(dev, flags);
1243 }
1244 }
1245 WARN_ON(dev->features & NETIF_F_LRO);
1246}
1247EXPORT_SYMBOL(dev_disable_lro);
1248
1249
881d966b
EB
1250static int dev_boot_phase = 1;
1251
1da177e4
LT
1252/*
1253 * Device change register/unregister. These are not inline or static
1254 * as we export them to the world.
1255 */
1256
1257/**
1258 * register_netdevice_notifier - register a network notifier block
1259 * @nb: notifier
1260 *
1261 * Register a notifier to be called when network device events occur.
1262 * The notifier passed is linked into the kernel structures and must
1263 * not be reused until it has been unregistered. A negative errno code
1264 * is returned on a failure.
1265 *
1266 * When registered all registration and up events are replayed
4ec93edb 1267 * to the new notifier to allow device to have a race free
1da177e4
LT
1268 * view of the network device list.
1269 */
1270
1271int register_netdevice_notifier(struct notifier_block *nb)
1272{
1273 struct net_device *dev;
fcc5a03a 1274 struct net_device *last;
881d966b 1275 struct net *net;
1da177e4
LT
1276 int err;
1277
1278 rtnl_lock();
f07d5b94 1279 err = raw_notifier_chain_register(&netdev_chain, nb);
fcc5a03a
HX
1280 if (err)
1281 goto unlock;
881d966b
EB
1282 if (dev_boot_phase)
1283 goto unlock;
1284 for_each_net(net) {
1285 for_each_netdev(net, dev) {
1286 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1287 err = notifier_to_errno(err);
1288 if (err)
1289 goto rollback;
1290
1291 if (!(dev->flags & IFF_UP))
1292 continue;
1da177e4 1293
881d966b
EB
1294 nb->notifier_call(nb, NETDEV_UP, dev);
1295 }
1da177e4 1296 }
fcc5a03a
HX
1297
1298unlock:
1da177e4
LT
1299 rtnl_unlock();
1300 return err;
fcc5a03a
HX
1301
1302rollback:
1303 last = dev;
881d966b
EB
1304 for_each_net(net) {
1305 for_each_netdev(net, dev) {
1306 if (dev == last)
1307 break;
fcc5a03a 1308
881d966b
EB
1309 if (dev->flags & IFF_UP) {
1310 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1311 nb->notifier_call(nb, NETDEV_DOWN, dev);
1312 }
1313 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
fcc5a03a 1314 }
fcc5a03a 1315 }
c67625a1
PE
1316
1317 raw_notifier_chain_unregister(&netdev_chain, nb);
fcc5a03a 1318 goto unlock;
1da177e4 1319}
d1b19dff 1320EXPORT_SYMBOL(register_netdevice_notifier);
1da177e4
LT
1321
1322/**
1323 * unregister_netdevice_notifier - unregister a network notifier block
1324 * @nb: notifier
1325 *
1326 * Unregister a notifier previously registered by
1327 * register_netdevice_notifier(). The notifier is unlinked into the
1328 * kernel structures and may then be reused. A negative errno code
1329 * is returned on a failure.
1330 */
1331
1332int unregister_netdevice_notifier(struct notifier_block *nb)
1333{
9f514950
HX
1334 int err;
1335
1336 rtnl_lock();
f07d5b94 1337 err = raw_notifier_chain_unregister(&netdev_chain, nb);
9f514950
HX
1338 rtnl_unlock();
1339 return err;
1da177e4 1340}
d1b19dff 1341EXPORT_SYMBOL(unregister_netdevice_notifier);
1da177e4
LT
1342
1343/**
1344 * call_netdevice_notifiers - call all network notifier blocks
1345 * @val: value passed unmodified to notifier function
c4ea43c5 1346 * @dev: net_device pointer passed unmodified to notifier function
1da177e4
LT
1347 *
1348 * Call all network notifier blocks. Parameters and return value
f07d5b94 1349 * are as for raw_notifier_call_chain().
1da177e4
LT
1350 */
1351
ad7379d4 1352int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1da177e4 1353{
ad7379d4 1354 return raw_notifier_call_chain(&netdev_chain, val, dev);
1da177e4
LT
1355}
1356
1357/* When > 0 there are consumers of rx skb time stamps */
1358static atomic_t netstamp_needed = ATOMIC_INIT(0);
1359
1360void net_enable_timestamp(void)
1361{
1362 atomic_inc(&netstamp_needed);
1363}
d1b19dff 1364EXPORT_SYMBOL(net_enable_timestamp);
1da177e4
LT
1365
1366void net_disable_timestamp(void)
1367{
1368 atomic_dec(&netstamp_needed);
1369}
d1b19dff 1370EXPORT_SYMBOL(net_disable_timestamp);
1da177e4 1371
a61bbcf2 1372static inline void net_timestamp(struct sk_buff *skb)
1da177e4
LT
1373{
1374 if (atomic_read(&netstamp_needed))
a61bbcf2 1375 __net_timestamp(skb);
b7aa0bf7
ED
1376 else
1377 skb->tstamp.tv64 = 0;
1da177e4
LT
1378}
1379
1380/*
1381 * Support routine. Sends outgoing frames to any network
1382 * taps currently in use.
1383 */
1384
f6a78bfc 1385static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1da177e4
LT
1386{
1387 struct packet_type *ptype;
a61bbcf2 1388
8caf1539
JP
1389#ifdef CONFIG_NET_CLS_ACT
1390 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1391 net_timestamp(skb);
1392#else
a61bbcf2 1393 net_timestamp(skb);
8caf1539 1394#endif
1da177e4
LT
1395
1396 rcu_read_lock();
1397 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1398 /* Never send packets back to the socket
1399 * they originated from - MvS (miquels@drinkel.ow.org)
1400 */
1401 if ((ptype->dev == dev || !ptype->dev) &&
1402 (ptype->af_packet_priv == NULL ||
1403 (struct sock *)ptype->af_packet_priv != skb->sk)) {
d1b19dff 1404 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1da177e4
LT
1405 if (!skb2)
1406 break;
1407
1408 /* skb->nh should be correctly
1409 set by sender, so that the second statement is
1410 just protection against buggy protocols.
1411 */
459a98ed 1412 skb_reset_mac_header(skb2);
1da177e4 1413
d56f90a7 1414 if (skb_network_header(skb2) < skb2->data ||
27a884dc 1415 skb2->network_header > skb2->tail) {
1da177e4
LT
1416 if (net_ratelimit())
1417 printk(KERN_CRIT "protocol %04x is "
1418 "buggy, dev %s\n",
1419 skb2->protocol, dev->name);
c1d2bbe1 1420 skb_reset_network_header(skb2);
1da177e4
LT
1421 }
1422
b0e380b1 1423 skb2->transport_header = skb2->network_header;
1da177e4 1424 skb2->pkt_type = PACKET_OUTGOING;
f2ccd8fa 1425 ptype->func(skb2, skb->dev, ptype, skb->dev);
1da177e4
LT
1426 }
1427 }
1428 rcu_read_unlock();
1429}
1430
56079431 1431
def82a1d 1432static inline void __netif_reschedule(struct Qdisc *q)
56079431 1433{
def82a1d
JP
1434 struct softnet_data *sd;
1435 unsigned long flags;
56079431 1436
def82a1d
JP
1437 local_irq_save(flags);
1438 sd = &__get_cpu_var(softnet_data);
1439 q->next_sched = sd->output_queue;
1440 sd->output_queue = q;
1441 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1442 local_irq_restore(flags);
1443}
1444
1445void __netif_schedule(struct Qdisc *q)
1446{
1447 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1448 __netif_reschedule(q);
56079431
DV
1449}
1450EXPORT_SYMBOL(__netif_schedule);
1451
bea3348e 1452void dev_kfree_skb_irq(struct sk_buff *skb)
56079431 1453{
bea3348e
SH
1454 if (atomic_dec_and_test(&skb->users)) {
1455 struct softnet_data *sd;
1456 unsigned long flags;
56079431 1457
bea3348e
SH
1458 local_irq_save(flags);
1459 sd = &__get_cpu_var(softnet_data);
1460 skb->next = sd->completion_queue;
1461 sd->completion_queue = skb;
1462 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1463 local_irq_restore(flags);
1464 }
56079431 1465}
bea3348e 1466EXPORT_SYMBOL(dev_kfree_skb_irq);
56079431
DV
1467
1468void dev_kfree_skb_any(struct sk_buff *skb)
1469{
1470 if (in_irq() || irqs_disabled())
1471 dev_kfree_skb_irq(skb);
1472 else
1473 dev_kfree_skb(skb);
1474}
1475EXPORT_SYMBOL(dev_kfree_skb_any);
1476
1477
bea3348e
SH
1478/**
1479 * netif_device_detach - mark device as removed
1480 * @dev: network device
1481 *
1482 * Mark device as removed from system and therefore no longer available.
1483 */
56079431
DV
1484void netif_device_detach(struct net_device *dev)
1485{
1486 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1487 netif_running(dev)) {
d543103a 1488 netif_tx_stop_all_queues(dev);
56079431
DV
1489 }
1490}
1491EXPORT_SYMBOL(netif_device_detach);
1492
bea3348e
SH
1493/**
1494 * netif_device_attach - mark device as attached
1495 * @dev: network device
1496 *
1497 * Mark device as attached from system and restart if needed.
1498 */
56079431
DV
1499void netif_device_attach(struct net_device *dev)
1500{
1501 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1502 netif_running(dev)) {
d543103a 1503 netif_tx_wake_all_queues(dev);
4ec93edb 1504 __netdev_watchdog_up(dev);
56079431
DV
1505 }
1506}
1507EXPORT_SYMBOL(netif_device_attach);
1508
6de329e2
BH
1509static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1510{
1511 return ((features & NETIF_F_GEN_CSUM) ||
1512 ((features & NETIF_F_IP_CSUM) &&
1513 protocol == htons(ETH_P_IP)) ||
1514 ((features & NETIF_F_IPV6_CSUM) &&
1c8dbcf6
YZ
1515 protocol == htons(ETH_P_IPV6)) ||
1516 ((features & NETIF_F_FCOE_CRC) &&
1517 protocol == htons(ETH_P_FCOE)));
6de329e2
BH
1518}
1519
1520static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1521{
1522 if (can_checksum_protocol(dev->features, skb->protocol))
1523 return true;
1524
1525 if (skb->protocol == htons(ETH_P_8021Q)) {
1526 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1527 if (can_checksum_protocol(dev->features & dev->vlan_features,
1528 veh->h_vlan_encapsulated_proto))
1529 return true;
1530 }
1531
1532 return false;
1533}
56079431 1534
1da177e4
LT
1535/*
1536 * Invalidate hardware checksum when packet is to be mangled, and
1537 * complete checksum manually on outgoing path.
1538 */
84fa7933 1539int skb_checksum_help(struct sk_buff *skb)
1da177e4 1540{
d3bc23e7 1541 __wsum csum;
663ead3b 1542 int ret = 0, offset;
1da177e4 1543
84fa7933 1544 if (skb->ip_summed == CHECKSUM_COMPLETE)
a430a43d
HX
1545 goto out_set_summed;
1546
1547 if (unlikely(skb_shinfo(skb)->gso_size)) {
a430a43d
HX
1548 /* Let GSO fix up the checksum. */
1549 goto out_set_summed;
1da177e4
LT
1550 }
1551
a030847e
HX
1552 offset = skb->csum_start - skb_headroom(skb);
1553 BUG_ON(offset >= skb_headlen(skb));
1554 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1555
1556 offset += skb->csum_offset;
1557 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1558
1559 if (skb_cloned(skb) &&
1560 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1da177e4
LT
1561 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1562 if (ret)
1563 goto out;
1564 }
1565
a030847e 1566 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
a430a43d 1567out_set_summed:
1da177e4 1568 skb->ip_summed = CHECKSUM_NONE;
4ec93edb 1569out:
1da177e4
LT
1570 return ret;
1571}
d1b19dff 1572EXPORT_SYMBOL(skb_checksum_help);
1da177e4 1573
f6a78bfc
HX
1574/**
1575 * skb_gso_segment - Perform segmentation on skb.
1576 * @skb: buffer to segment
576a30eb 1577 * @features: features for the output path (see dev->features)
f6a78bfc
HX
1578 *
1579 * This function segments the given skb and returns a list of segments.
576a30eb
HX
1580 *
1581 * It may return NULL if the skb requires no segmentation. This is
1582 * only possible when GSO is used for verifying header integrity.
f6a78bfc 1583 */
576a30eb 1584struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
f6a78bfc
HX
1585{
1586 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1587 struct packet_type *ptype;
252e3346 1588 __be16 type = skb->protocol;
a430a43d 1589 int err;
f6a78bfc 1590
459a98ed 1591 skb_reset_mac_header(skb);
b0e380b1 1592 skb->mac_len = skb->network_header - skb->mac_header;
f6a78bfc
HX
1593 __skb_pull(skb, skb->mac_len);
1594
67fd1a73
HX
1595 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1596 struct net_device *dev = skb->dev;
1597 struct ethtool_drvinfo info = {};
1598
1599 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1600 dev->ethtool_ops->get_drvinfo(dev, &info);
1601
1602 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1603 "ip_summed=%d",
1604 info.driver, dev ? dev->features : 0L,
1605 skb->sk ? skb->sk->sk_route_caps : 0L,
1606 skb->len, skb->data_len, skb->ip_summed);
1607
a430a43d
HX
1608 if (skb_header_cloned(skb) &&
1609 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1610 return ERR_PTR(err);
1611 }
1612
f6a78bfc 1613 rcu_read_lock();
82d8a867
PE
1614 list_for_each_entry_rcu(ptype,
1615 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
f6a78bfc 1616 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
84fa7933 1617 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
a430a43d
HX
1618 err = ptype->gso_send_check(skb);
1619 segs = ERR_PTR(err);
1620 if (err || skb_gso_ok(skb, features))
1621 break;
d56f90a7
ACM
1622 __skb_push(skb, (skb->data -
1623 skb_network_header(skb)));
a430a43d 1624 }
576a30eb 1625 segs = ptype->gso_segment(skb, features);
f6a78bfc
HX
1626 break;
1627 }
1628 }
1629 rcu_read_unlock();
1630
98e399f8 1631 __skb_push(skb, skb->data - skb_mac_header(skb));
576a30eb 1632
f6a78bfc
HX
1633 return segs;
1634}
f6a78bfc
HX
1635EXPORT_SYMBOL(skb_gso_segment);
1636
fb286bb2
HX
1637/* Take action when hardware reception checksum errors are detected. */
1638#ifdef CONFIG_BUG
1639void netdev_rx_csum_fault(struct net_device *dev)
1640{
1641 if (net_ratelimit()) {
4ec93edb 1642 printk(KERN_ERR "%s: hw csum failure.\n",
246a4212 1643 dev ? dev->name : "<unknown>");
fb286bb2
HX
1644 dump_stack();
1645 }
1646}
1647EXPORT_SYMBOL(netdev_rx_csum_fault);
1648#endif
1649
1da177e4
LT
1650/* Actually, we should eliminate this check as soon as we know, that:
1651 * 1. IOMMU is present and allows to map all the memory.
1652 * 2. No high memory really exists on this machine.
1653 */
1654
1655static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1656{
3d3a8533 1657#ifdef CONFIG_HIGHMEM
1da177e4
LT
1658 int i;
1659
1660 if (dev->features & NETIF_F_HIGHDMA)
1661 return 0;
1662
1663 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1664 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1665 return 1;
1666
3d3a8533 1667#endif
1da177e4
LT
1668 return 0;
1669}
1da177e4 1670
f6a78bfc
HX
1671struct dev_gso_cb {
1672 void (*destructor)(struct sk_buff *skb);
1673};
1674
1675#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1676
1677static void dev_gso_skb_destructor(struct sk_buff *skb)
1678{
1679 struct dev_gso_cb *cb;
1680
1681 do {
1682 struct sk_buff *nskb = skb->next;
1683
1684 skb->next = nskb->next;
1685 nskb->next = NULL;
1686 kfree_skb(nskb);
1687 } while (skb->next);
1688
1689 cb = DEV_GSO_CB(skb);
1690 if (cb->destructor)
1691 cb->destructor(skb);
1692}
1693
1694/**
1695 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1696 * @skb: buffer to segment
1697 *
1698 * This function segments the given skb and stores the list of segments
1699 * in skb->next.
1700 */
1701static int dev_gso_segment(struct sk_buff *skb)
1702{
1703 struct net_device *dev = skb->dev;
1704 struct sk_buff *segs;
576a30eb
HX
1705 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1706 NETIF_F_SG : 0);
1707
1708 segs = skb_gso_segment(skb, features);
1709
1710 /* Verifying header integrity only. */
1711 if (!segs)
1712 return 0;
f6a78bfc 1713
801678c5 1714 if (IS_ERR(segs))
f6a78bfc
HX
1715 return PTR_ERR(segs);
1716
1717 skb->next = segs;
1718 DEV_GSO_CB(skb)->destructor = skb->destructor;
1719 skb->destructor = dev_gso_skb_destructor;
1720
1721 return 0;
1722}
1723
fd2ea0a7
DM
1724int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1725 struct netdev_queue *txq)
f6a78bfc 1726{
00829823 1727 const struct net_device_ops *ops = dev->netdev_ops;
ac45f602 1728 int rc;
00829823 1729
f6a78bfc 1730 if (likely(!skb->next)) {
9be9a6b9 1731 if (!list_empty(&ptype_all))
f6a78bfc
HX
1732 dev_queue_xmit_nit(skb, dev);
1733
576a30eb
HX
1734 if (netif_needs_gso(dev, skb)) {
1735 if (unlikely(dev_gso_segment(skb)))
1736 goto out_kfree_skb;
1737 if (skb->next)
1738 goto gso;
1739 }
f6a78bfc 1740
93f154b5
ED
1741 /*
1742 * If device doesnt need skb->dst, release it right now while
1743 * its hot in this cpu cache
1744 */
adf30907
ED
1745 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1746 skb_dst_drop(skb);
1747
ac45f602 1748 rc = ops->ndo_start_xmit(skb, dev);
ec634fe3 1749 if (rc == NETDEV_TX_OK)
08baf561 1750 txq_trans_update(txq);
ac45f602
PO
1751 /*
1752 * TODO: if skb_orphan() was called by
1753 * dev->hard_start_xmit() (for example, the unmodified
1754 * igb driver does that; bnx2 doesn't), then
1755 * skb_tx_software_timestamp() will be unable to send
1756 * back the time stamp.
1757 *
1758 * How can this be prevented? Always create another
1759 * reference to the socket before calling
1760 * dev->hard_start_xmit()? Prevent that skb_orphan()
1761 * does anything in dev->hard_start_xmit() by clearing
1762 * the skb destructor before the call and restoring it
1763 * afterwards, then doing the skb_orphan() ourselves?
1764 */
ac45f602 1765 return rc;
f6a78bfc
HX
1766 }
1767
576a30eb 1768gso:
f6a78bfc
HX
1769 do {
1770 struct sk_buff *nskb = skb->next;
f6a78bfc
HX
1771
1772 skb->next = nskb->next;
1773 nskb->next = NULL;
00829823 1774 rc = ops->ndo_start_xmit(nskb, dev);
ec634fe3 1775 if (unlikely(rc != NETDEV_TX_OK)) {
f54d9e8d 1776 nskb->next = skb->next;
f6a78bfc
HX
1777 skb->next = nskb;
1778 return rc;
1779 }
08baf561 1780 txq_trans_update(txq);
fd2ea0a7 1781 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
f54d9e8d 1782 return NETDEV_TX_BUSY;
f6a78bfc 1783 } while (skb->next);
4ec93edb 1784
f6a78bfc
HX
1785 skb->destructor = DEV_GSO_CB(skb)->destructor;
1786
1787out_kfree_skb:
1788 kfree_skb(skb);
ec634fe3 1789 return NETDEV_TX_OK;
f6a78bfc
HX
1790}
1791
7019298a 1792static u32 skb_tx_hashrnd;
b6b2fed1 1793
9247744e 1794u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
8f0f2223 1795{
7019298a 1796 u32 hash;
b6b2fed1 1797
513de11b
DM
1798 if (skb_rx_queue_recorded(skb)) {
1799 hash = skb_get_rx_queue(skb);
d1b19dff 1800 while (unlikely(hash >= dev->real_num_tx_queues))
513de11b
DM
1801 hash -= dev->real_num_tx_queues;
1802 return hash;
1803 }
ec581f6a
ED
1804
1805 if (skb->sk && skb->sk->sk_hash)
7019298a 1806 hash = skb->sk->sk_hash;
ec581f6a 1807 else
7019298a 1808 hash = skb->protocol;
d5a9e24a 1809
7019298a 1810 hash = jhash_1word(hash, skb_tx_hashrnd);
b6b2fed1
DM
1811
1812 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
8f0f2223 1813}
9247744e 1814EXPORT_SYMBOL(skb_tx_hash);
8f0f2223 1815
e8a0464c
DM
1816static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1817 struct sk_buff *skb)
1818{
a4ee3ce3
KK
1819 u16 queue_index;
1820 struct sock *sk = skb->sk;
1821
1822 if (sk_tx_queue_recorded(sk)) {
1823 queue_index = sk_tx_queue_get(sk);
1824 } else {
1825 const struct net_device_ops *ops = dev->netdev_ops;
1826
1827 if (ops->ndo_select_queue) {
1828 queue_index = ops->ndo_select_queue(dev, skb);
1829 } else {
1830 queue_index = 0;
1831 if (dev->real_num_tx_queues > 1)
1832 queue_index = skb_tx_hash(dev, skb);
fd2ea0a7 1833
a4ee3ce3
KK
1834 if (sk && sk->sk_dst_cache)
1835 sk_tx_queue_set(sk, queue_index);
1836 }
1837 }
eae792b7 1838
fd2ea0a7
DM
1839 skb_set_queue_mapping(skb, queue_index);
1840 return netdev_get_tx_queue(dev, queue_index);
e8a0464c
DM
1841}
1842
bbd8a0d3
KK
1843static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1844 struct net_device *dev,
1845 struct netdev_queue *txq)
1846{
1847 spinlock_t *root_lock = qdisc_lock(q);
1848 int rc;
1849
1850 spin_lock(root_lock);
1851 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1852 kfree_skb(skb);
1853 rc = NET_XMIT_DROP;
1854 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1855 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1856 /*
1857 * This is a work-conserving queue; there are no old skbs
1858 * waiting to be sent out; and the qdisc is not running -
1859 * xmit the skb directly.
1860 */
1861 __qdisc_update_bstats(q, skb->len);
1862 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1863 __qdisc_run(q);
1864 else
1865 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1866
1867 rc = NET_XMIT_SUCCESS;
1868 } else {
1869 rc = qdisc_enqueue_root(skb, q);
1870 qdisc_run(q);
1871 }
1872 spin_unlock(root_lock);
1873
1874 return rc;
1875}
1876
d29f749e
DJ
1877/**
1878 * dev_queue_xmit - transmit a buffer
1879 * @skb: buffer to transmit
1880 *
1881 * Queue a buffer for transmission to a network device. The caller must
1882 * have set the device and priority and built the buffer before calling
1883 * this function. The function can be called from an interrupt.
1884 *
1885 * A negative errno code is returned on a failure. A success does not
1886 * guarantee the frame will be transmitted as it may be dropped due
1887 * to congestion or traffic shaping.
1888 *
1889 * -----------------------------------------------------------------------------------
1890 * I notice this method can also return errors from the queue disciplines,
1891 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1892 * be positive.
1893 *
1894 * Regardless of the return value, the skb is consumed, so it is currently
1895 * difficult to retry a send to this method. (You can bump the ref count
1896 * before sending to hold a reference for retry if you are careful.)
1897 *
1898 * When calling this method, interrupts MUST be enabled. This is because
1899 * the BH enable code must have IRQs enabled so that it will not deadlock.
1900 * --BLG
1901 */
1da177e4
LT
1902int dev_queue_xmit(struct sk_buff *skb)
1903{
1904 struct net_device *dev = skb->dev;
dc2b4847 1905 struct netdev_queue *txq;
1da177e4
LT
1906 struct Qdisc *q;
1907 int rc = -ENOMEM;
1908
f6a78bfc
HX
1909 /* GSO will handle the following emulations directly. */
1910 if (netif_needs_gso(dev, skb))
1911 goto gso;
1912
4cf704fb 1913 if (skb_has_frags(skb) &&
1da177e4 1914 !(dev->features & NETIF_F_FRAGLIST) &&
364c6bad 1915 __skb_linearize(skb))
1da177e4
LT
1916 goto out_kfree_skb;
1917
1918 /* Fragmented skb is linearized if device does not support SG,
1919 * or if at least one of fragments is in highmem and device
1920 * does not support DMA from it.
1921 */
1922 if (skb_shinfo(skb)->nr_frags &&
1923 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
364c6bad 1924 __skb_linearize(skb))
1da177e4
LT
1925 goto out_kfree_skb;
1926
1927 /* If packet is not checksummed and device does not support
1928 * checksumming for this protocol, complete checksumming here.
1929 */
663ead3b
HX
1930 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1931 skb_set_transport_header(skb, skb->csum_start -
1932 skb_headroom(skb));
6de329e2
BH
1933 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1934 goto out_kfree_skb;
663ead3b 1935 }
1da177e4 1936
f6a78bfc 1937gso:
4ec93edb
YH
1938 /* Disable soft irqs for various locks below. Also
1939 * stops preemption for RCU.
1da177e4 1940 */
4ec93edb 1941 rcu_read_lock_bh();
1da177e4 1942
eae792b7 1943 txq = dev_pick_tx(dev, skb);
b0e1e646 1944 q = rcu_dereference(txq->qdisc);
37437bb2 1945
1da177e4 1946#ifdef CONFIG_NET_CLS_ACT
d1b19dff 1947 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
1da177e4
LT
1948#endif
1949 if (q->enqueue) {
bbd8a0d3 1950 rc = __dev_xmit_skb(skb, q, dev, txq);
37437bb2 1951 goto out;
1da177e4
LT
1952 }
1953
1954 /* The device has no queue. Common case for software devices:
1955 loopback, all the sorts of tunnels...
1956
932ff279
HX
1957 Really, it is unlikely that netif_tx_lock protection is necessary
1958 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1da177e4
LT
1959 counters.)
1960 However, it is possible, that they rely on protection
1961 made by us here.
1962
1963 Check this and shot the lock. It is not prone from deadlocks.
1964 Either shot noqueue qdisc, it is even simpler 8)
1965 */
1966 if (dev->flags & IFF_UP) {
1967 int cpu = smp_processor_id(); /* ok because BHs are off */
1968
c773e847 1969 if (txq->xmit_lock_owner != cpu) {
1da177e4 1970
c773e847 1971 HARD_TX_LOCK(dev, txq, cpu);
1da177e4 1972
fd2ea0a7 1973 if (!netif_tx_queue_stopped(txq)) {
03a9a447 1974 rc = NET_XMIT_SUCCESS;
fd2ea0a7 1975 if (!dev_hard_start_xmit(skb, dev, txq)) {
c773e847 1976 HARD_TX_UNLOCK(dev, txq);
1da177e4
LT
1977 goto out;
1978 }
1979 }
c773e847 1980 HARD_TX_UNLOCK(dev, txq);
1da177e4
LT
1981 if (net_ratelimit())
1982 printk(KERN_CRIT "Virtual device %s asks to "
1983 "queue packet!\n", dev->name);
1984 } else {
1985 /* Recursion is detected! It is possible,
1986 * unfortunately */
1987 if (net_ratelimit())
1988 printk(KERN_CRIT "Dead loop on virtual device "
1989 "%s, fix it urgently!\n", dev->name);
1990 }
1991 }
1992
1993 rc = -ENETDOWN;
d4828d85 1994 rcu_read_unlock_bh();
1da177e4
LT
1995
1996out_kfree_skb:
1997 kfree_skb(skb);
1998 return rc;
1999out:
d4828d85 2000 rcu_read_unlock_bh();
1da177e4
LT
2001 return rc;
2002}
d1b19dff 2003EXPORT_SYMBOL(dev_queue_xmit);
1da177e4
LT
2004
2005
2006/*=======================================================================
2007 Receiver routines
2008 =======================================================================*/
2009
6b2bedc3
SH
2010int netdev_max_backlog __read_mostly = 1000;
2011int netdev_budget __read_mostly = 300;
2012int weight_p __read_mostly = 64; /* old backlog weight */
1da177e4
LT
2013
2014DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2015
2016
1da177e4
LT
2017/**
2018 * netif_rx - post buffer to the network code
2019 * @skb: buffer to post
2020 *
2021 * This function receives a packet from a device driver and queues it for
2022 * the upper (protocol) levels to process. It always succeeds. The buffer
2023 * may be dropped during processing for congestion control or by the
2024 * protocol layers.
2025 *
2026 * return values:
2027 * NET_RX_SUCCESS (no congestion)
1da177e4
LT
2028 * NET_RX_DROP (packet was dropped)
2029 *
2030 */
2031
2032int netif_rx(struct sk_buff *skb)
2033{
1da177e4
LT
2034 struct softnet_data *queue;
2035 unsigned long flags;
2036
2037 /* if netpoll wants it, pretend we never saw it */
2038 if (netpoll_rx(skb))
2039 return NET_RX_DROP;
2040
b7aa0bf7 2041 if (!skb->tstamp.tv64)
a61bbcf2 2042 net_timestamp(skb);
1da177e4
LT
2043
2044 /*
2045 * The code is rearranged so that the path is the most
2046 * short when CPU is congested, but is still operating.
2047 */
2048 local_irq_save(flags);
1da177e4
LT
2049 queue = &__get_cpu_var(softnet_data);
2050
2051 __get_cpu_var(netdev_rx_stat).total++;
2052 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2053 if (queue->input_pkt_queue.qlen) {
1da177e4 2054enqueue:
1da177e4 2055 __skb_queue_tail(&queue->input_pkt_queue, skb);
1da177e4 2056 local_irq_restore(flags);
34008d8c 2057 return NET_RX_SUCCESS;
1da177e4
LT
2058 }
2059
bea3348e 2060 napi_schedule(&queue->backlog);
1da177e4
LT
2061 goto enqueue;
2062 }
2063
1da177e4
LT
2064 __get_cpu_var(netdev_rx_stat).dropped++;
2065 local_irq_restore(flags);
2066
2067 kfree_skb(skb);
2068 return NET_RX_DROP;
2069}
d1b19dff 2070EXPORT_SYMBOL(netif_rx);
1da177e4
LT
2071
2072int netif_rx_ni(struct sk_buff *skb)
2073{
2074 int err;
2075
2076 preempt_disable();
2077 err = netif_rx(skb);
2078 if (local_softirq_pending())
2079 do_softirq();
2080 preempt_enable();
2081
2082 return err;
2083}
1da177e4
LT
2084EXPORT_SYMBOL(netif_rx_ni);
2085
1da177e4
LT
2086static void net_tx_action(struct softirq_action *h)
2087{
2088 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2089
2090 if (sd->completion_queue) {
2091 struct sk_buff *clist;
2092
2093 local_irq_disable();
2094 clist = sd->completion_queue;
2095 sd->completion_queue = NULL;
2096 local_irq_enable();
2097
2098 while (clist) {
2099 struct sk_buff *skb = clist;
2100 clist = clist->next;
2101
547b792c 2102 WARN_ON(atomic_read(&skb->users));
1da177e4
LT
2103 __kfree_skb(skb);
2104 }
2105 }
2106
2107 if (sd->output_queue) {
37437bb2 2108 struct Qdisc *head;
1da177e4
LT
2109
2110 local_irq_disable();
2111 head = sd->output_queue;
2112 sd->output_queue = NULL;
2113 local_irq_enable();
2114
2115 while (head) {
37437bb2
DM
2116 struct Qdisc *q = head;
2117 spinlock_t *root_lock;
2118
1da177e4
LT
2119 head = head->next_sched;
2120
5fb66229 2121 root_lock = qdisc_lock(q);
37437bb2 2122 if (spin_trylock(root_lock)) {
def82a1d
JP
2123 smp_mb__before_clear_bit();
2124 clear_bit(__QDISC_STATE_SCHED,
2125 &q->state);
37437bb2
DM
2126 qdisc_run(q);
2127 spin_unlock(root_lock);
1da177e4 2128 } else {
195648bb 2129 if (!test_bit(__QDISC_STATE_DEACTIVATED,
e8a83e10 2130 &q->state)) {
195648bb 2131 __netif_reschedule(q);
e8a83e10
JP
2132 } else {
2133 smp_mb__before_clear_bit();
2134 clear_bit(__QDISC_STATE_SCHED,
2135 &q->state);
2136 }
1da177e4
LT
2137 }
2138 }
2139 }
2140}
2141
6f05f629
SH
2142static inline int deliver_skb(struct sk_buff *skb,
2143 struct packet_type *pt_prev,
2144 struct net_device *orig_dev)
1da177e4
LT
2145{
2146 atomic_inc(&skb->users);
f2ccd8fa 2147 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
2148}
2149
2150#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
da678292
MM
2151
2152#if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2153/* This hook is defined here for ATM LANE */
2154int (*br_fdb_test_addr_hook)(struct net_device *dev,
2155 unsigned char *addr) __read_mostly;
4fb019a0 2156EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
da678292 2157#endif
1da177e4 2158
6229e362
SH
2159/*
2160 * If bridge module is loaded call bridging hook.
2161 * returns NULL if packet was consumed.
2162 */
2163struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2164 struct sk_buff *skb) __read_mostly;
4fb019a0 2165EXPORT_SYMBOL_GPL(br_handle_frame_hook);
da678292 2166
6229e362
SH
2167static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2168 struct packet_type **pt_prev, int *ret,
2169 struct net_device *orig_dev)
1da177e4
LT
2170{
2171 struct net_bridge_port *port;
2172
6229e362
SH
2173 if (skb->pkt_type == PACKET_LOOPBACK ||
2174 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2175 return skb;
1da177e4
LT
2176
2177 if (*pt_prev) {
6229e362 2178 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1da177e4 2179 *pt_prev = NULL;
4ec93edb
YH
2180 }
2181
6229e362 2182 return br_handle_frame_hook(port, skb);
1da177e4
LT
2183}
2184#else
6229e362 2185#define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1da177e4
LT
2186#endif
2187
b863ceb7
PM
2188#if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2189struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2190EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2191
2192static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2193 struct packet_type **pt_prev,
2194 int *ret,
2195 struct net_device *orig_dev)
2196{
2197 if (skb->dev->macvlan_port == NULL)
2198 return skb;
2199
2200 if (*pt_prev) {
2201 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2202 *pt_prev = NULL;
2203 }
2204 return macvlan_handle_frame_hook(skb);
2205}
2206#else
2207#define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2208#endif
2209
1da177e4
LT
2210#ifdef CONFIG_NET_CLS_ACT
2211/* TODO: Maybe we should just force sch_ingress to be compiled in
2212 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2213 * a compare and 2 stores extra right now if we dont have it on
2214 * but have CONFIG_NET_CLS_ACT
4ec93edb 2215 * NOTE: This doesnt stop any functionality; if you dont have
1da177e4
LT
2216 * the ingress scheduler, you just cant add policies on ingress.
2217 *
2218 */
4ec93edb 2219static int ing_filter(struct sk_buff *skb)
1da177e4 2220{
1da177e4 2221 struct net_device *dev = skb->dev;
f697c3e8 2222 u32 ttl = G_TC_RTTL(skb->tc_verd);
555353cf
DM
2223 struct netdev_queue *rxq;
2224 int result = TC_ACT_OK;
2225 struct Qdisc *q;
4ec93edb 2226
f697c3e8
HX
2227 if (MAX_RED_LOOP < ttl++) {
2228 printk(KERN_WARNING
2229 "Redir loop detected Dropping packet (%d->%d)\n",
2230 skb->iif, dev->ifindex);
2231 return TC_ACT_SHOT;
2232 }
1da177e4 2233
f697c3e8
HX
2234 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2235 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
1da177e4 2236
555353cf
DM
2237 rxq = &dev->rx_queue;
2238
83874000 2239 q = rxq->qdisc;
8d50b53d 2240 if (q != &noop_qdisc) {
83874000 2241 spin_lock(qdisc_lock(q));
a9312ae8
DM
2242 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2243 result = qdisc_enqueue_root(skb, q);
83874000
DM
2244 spin_unlock(qdisc_lock(q));
2245 }
f697c3e8
HX
2246
2247 return result;
2248}
86e65da9 2249
f697c3e8
HX
2250static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2251 struct packet_type **pt_prev,
2252 int *ret, struct net_device *orig_dev)
2253{
8d50b53d 2254 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
f697c3e8 2255 goto out;
1da177e4 2256
f697c3e8
HX
2257 if (*pt_prev) {
2258 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2259 *pt_prev = NULL;
2260 } else {
2261 /* Huh? Why does turning on AF_PACKET affect this? */
2262 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1da177e4
LT
2263 }
2264
f697c3e8
HX
2265 switch (ing_filter(skb)) {
2266 case TC_ACT_SHOT:
2267 case TC_ACT_STOLEN:
2268 kfree_skb(skb);
2269 return NULL;
2270 }
2271
2272out:
2273 skb->tc_verd = 0;
2274 return skb;
1da177e4
LT
2275}
2276#endif
2277
bc1d0411
PM
2278/*
2279 * netif_nit_deliver - deliver received packets to network taps
2280 * @skb: buffer
2281 *
2282 * This function is used to deliver incoming packets to network
2283 * taps. It should be used when the normal netif_receive_skb path
2284 * is bypassed, for example because of VLAN acceleration.
2285 */
2286void netif_nit_deliver(struct sk_buff *skb)
2287{
2288 struct packet_type *ptype;
2289
2290 if (list_empty(&ptype_all))
2291 return;
2292
2293 skb_reset_network_header(skb);
2294 skb_reset_transport_header(skb);
2295 skb->mac_len = skb->network_header - skb->mac_header;
2296
2297 rcu_read_lock();
2298 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2299 if (!ptype->dev || ptype->dev == skb->dev)
2300 deliver_skb(skb, ptype, skb->dev);
2301 }
2302 rcu_read_unlock();
2303}
2304
3b582cc1
SH
2305/**
2306 * netif_receive_skb - process receive buffer from network
2307 * @skb: buffer to process
2308 *
2309 * netif_receive_skb() is the main receive data processing function.
2310 * It always succeeds. The buffer may be dropped during processing
2311 * for congestion control or by the protocol layers.
2312 *
2313 * This function may only be called from softirq context and interrupts
2314 * should be enabled.
2315 *
2316 * Return values (usually ignored):
2317 * NET_RX_SUCCESS: no congestion
2318 * NET_RX_DROP: packet was dropped
2319 */
1da177e4
LT
2320int netif_receive_skb(struct sk_buff *skb)
2321{
2322 struct packet_type *ptype, *pt_prev;
f2ccd8fa 2323 struct net_device *orig_dev;
0d7a3681 2324 struct net_device *null_or_orig;
1da177e4 2325 int ret = NET_RX_DROP;
252e3346 2326 __be16 type;
1da177e4 2327
81bbb3d4
ED
2328 if (!skb->tstamp.tv64)
2329 net_timestamp(skb);
2330
05423b24 2331 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
9b22ea56
PM
2332 return NET_RX_SUCCESS;
2333
1da177e4 2334 /* if we've gotten here through NAPI, check netpoll */
bea3348e 2335 if (netpoll_receive_skb(skb))
1da177e4
LT
2336 return NET_RX_DROP;
2337
c01003c2
PM
2338 if (!skb->iif)
2339 skb->iif = skb->dev->ifindex;
86e65da9 2340
0d7a3681 2341 null_or_orig = NULL;
cc9bd5ce
JE
2342 orig_dev = skb->dev;
2343 if (orig_dev->master) {
0d7a3681
JE
2344 if (skb_bond_should_drop(skb))
2345 null_or_orig = orig_dev; /* deliver only exact match */
2346 else
2347 skb->dev = orig_dev->master;
cc9bd5ce 2348 }
8f903c70 2349
1da177e4
LT
2350 __get_cpu_var(netdev_rx_stat).total++;
2351
c1d2bbe1 2352 skb_reset_network_header(skb);
badff6d0 2353 skb_reset_transport_header(skb);
b0e380b1 2354 skb->mac_len = skb->network_header - skb->mac_header;
1da177e4
LT
2355
2356 pt_prev = NULL;
2357
2358 rcu_read_lock();
2359
2360#ifdef CONFIG_NET_CLS_ACT
2361 if (skb->tc_verd & TC_NCLS) {
2362 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2363 goto ncls;
2364 }
2365#endif
2366
2367 list_for_each_entry_rcu(ptype, &ptype_all, list) {
f982307f
JE
2368 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2369 ptype->dev == orig_dev) {
4ec93edb 2370 if (pt_prev)
f2ccd8fa 2371 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
2372 pt_prev = ptype;
2373 }
2374 }
2375
2376#ifdef CONFIG_NET_CLS_ACT
f697c3e8
HX
2377 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2378 if (!skb)
1da177e4 2379 goto out;
1da177e4
LT
2380ncls:
2381#endif
2382
6229e362 2383 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
b863ceb7
PM
2384 if (!skb)
2385 goto out;
2386 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
6229e362 2387 if (!skb)
1da177e4
LT
2388 goto out;
2389
2390 type = skb->protocol;
82d8a867
PE
2391 list_for_each_entry_rcu(ptype,
2392 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1da177e4 2393 if (ptype->type == type &&
f982307f
JE
2394 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2395 ptype->dev == orig_dev)) {
4ec93edb 2396 if (pt_prev)
f2ccd8fa 2397 ret = deliver_skb(skb, pt_prev, orig_dev);
1da177e4
LT
2398 pt_prev = ptype;
2399 }
2400 }
2401
2402 if (pt_prev) {
f2ccd8fa 2403 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1da177e4
LT
2404 } else {
2405 kfree_skb(skb);
2406 /* Jamal, now you will not able to escape explaining
2407 * me how you were going to use this. :-)
2408 */
2409 ret = NET_RX_DROP;
2410 }
2411
2412out:
2413 rcu_read_unlock();
2414 return ret;
2415}
d1b19dff 2416EXPORT_SYMBOL(netif_receive_skb);
1da177e4 2417
6e583ce5
SH
2418/* Network device is going away, flush any packets still pending */
2419static void flush_backlog(void *arg)
2420{
2421 struct net_device *dev = arg;
2422 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2423 struct sk_buff *skb, *tmp;
2424
2425 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2426 if (skb->dev == dev) {
2427 __skb_unlink(skb, &queue->input_pkt_queue);
2428 kfree_skb(skb);
2429 }
2430}
2431
d565b0a1
HX
2432static int napi_gro_complete(struct sk_buff *skb)
2433{
2434 struct packet_type *ptype;
2435 __be16 type = skb->protocol;
2436 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2437 int err = -ENOENT;
2438
fc59f9a3
HX
2439 if (NAPI_GRO_CB(skb)->count == 1) {
2440 skb_shinfo(skb)->gso_size = 0;
d565b0a1 2441 goto out;
fc59f9a3 2442 }
d565b0a1
HX
2443
2444 rcu_read_lock();
2445 list_for_each_entry_rcu(ptype, head, list) {
2446 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2447 continue;
2448
2449 err = ptype->gro_complete(skb);
2450 break;
2451 }
2452 rcu_read_unlock();
2453
2454 if (err) {
2455 WARN_ON(&ptype->list == head);
2456 kfree_skb(skb);
2457 return NET_RX_SUCCESS;
2458 }
2459
2460out:
d565b0a1
HX
2461 return netif_receive_skb(skb);
2462}
2463
2464void napi_gro_flush(struct napi_struct *napi)
2465{
2466 struct sk_buff *skb, *next;
2467
2468 for (skb = napi->gro_list; skb; skb = next) {
2469 next = skb->next;
2470 skb->next = NULL;
2471 napi_gro_complete(skb);
2472 }
2473
4ae5544f 2474 napi->gro_count = 0;
d565b0a1
HX
2475 napi->gro_list = NULL;
2476}
2477EXPORT_SYMBOL(napi_gro_flush);
2478
5b252f0c 2479enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
d565b0a1
HX
2480{
2481 struct sk_buff **pp = NULL;
2482 struct packet_type *ptype;
2483 __be16 type = skb->protocol;
2484 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
0da2afd5 2485 int same_flow;
d565b0a1 2486 int mac_len;
5b252f0c 2487 enum gro_result ret;
d565b0a1
HX
2488
2489 if (!(skb->dev->features & NETIF_F_GRO))
2490 goto normal;
2491
4cf704fb 2492 if (skb_is_gso(skb) || skb_has_frags(skb))
f17f5c91
HX
2493 goto normal;
2494
d565b0a1
HX
2495 rcu_read_lock();
2496 list_for_each_entry_rcu(ptype, head, list) {
d565b0a1
HX
2497 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2498 continue;
2499
86911732 2500 skb_set_network_header(skb, skb_gro_offset(skb));
d565b0a1
HX
2501 mac_len = skb->network_header - skb->mac_header;
2502 skb->mac_len = mac_len;
2503 NAPI_GRO_CB(skb)->same_flow = 0;
2504 NAPI_GRO_CB(skb)->flush = 0;
5d38a079 2505 NAPI_GRO_CB(skb)->free = 0;
d565b0a1 2506
d565b0a1
HX
2507 pp = ptype->gro_receive(&napi->gro_list, skb);
2508 break;
2509 }
2510 rcu_read_unlock();
2511
2512 if (&ptype->list == head)
2513 goto normal;
2514
0da2afd5 2515 same_flow = NAPI_GRO_CB(skb)->same_flow;
5d0d9be8 2516 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
0da2afd5 2517
d565b0a1
HX
2518 if (pp) {
2519 struct sk_buff *nskb = *pp;
2520
2521 *pp = nskb->next;
2522 nskb->next = NULL;
2523 napi_gro_complete(nskb);
4ae5544f 2524 napi->gro_count--;
d565b0a1
HX
2525 }
2526
0da2afd5 2527 if (same_flow)
d565b0a1
HX
2528 goto ok;
2529
4ae5544f 2530 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
d565b0a1 2531 goto normal;
d565b0a1 2532
4ae5544f 2533 napi->gro_count++;
d565b0a1 2534 NAPI_GRO_CB(skb)->count = 1;
86911732 2535 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
d565b0a1
HX
2536 skb->next = napi->gro_list;
2537 napi->gro_list = skb;
5d0d9be8 2538 ret = GRO_HELD;
d565b0a1 2539
ad0f9904 2540pull:
cb18978c
HX
2541 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2542 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2543
2544 BUG_ON(skb->end - skb->tail < grow);
2545
2546 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2547
2548 skb->tail += grow;
2549 skb->data_len -= grow;
2550
2551 skb_shinfo(skb)->frags[0].page_offset += grow;
2552 skb_shinfo(skb)->frags[0].size -= grow;
2553
2554 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2555 put_page(skb_shinfo(skb)->frags[0].page);
2556 memmove(skb_shinfo(skb)->frags,
2557 skb_shinfo(skb)->frags + 1,
2558 --skb_shinfo(skb)->nr_frags);
2559 }
ad0f9904
HX
2560 }
2561
d565b0a1 2562ok:
5d0d9be8 2563 return ret;
d565b0a1
HX
2564
2565normal:
ad0f9904
HX
2566 ret = GRO_NORMAL;
2567 goto pull;
5d38a079 2568}
96e93eab
HX
2569EXPORT_SYMBOL(dev_gro_receive);
2570
5b252f0c
BH
2571static gro_result_t
2572__napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
96e93eab
HX
2573{
2574 struct sk_buff *p;
2575
d1c76af9
HX
2576 if (netpoll_rx_on(skb))
2577 return GRO_NORMAL;
2578
96e93eab 2579 for (p = napi->gro_list; p; p = p->next) {
f2bde732
SH
2580 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2581 && !compare_ether_header(skb_mac_header(p),
2582 skb_gro_mac_header(skb));
96e93eab
HX
2583 NAPI_GRO_CB(p)->flush = 0;
2584 }
2585
2586 return dev_gro_receive(napi, skb);
2587}
5d38a079 2588
5b252f0c 2589int napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
5d38a079 2590{
5d0d9be8
HX
2591 int err = NET_RX_SUCCESS;
2592
2593 switch (ret) {
2594 case GRO_NORMAL:
5d38a079
HX
2595 return netif_receive_skb(skb);
2596
5d0d9be8
HX
2597 case GRO_DROP:
2598 err = NET_RX_DROP;
2599 /* fall through */
2600
2601 case GRO_MERGED_FREE:
5d38a079
HX
2602 kfree_skb(skb);
2603 break;
5b252f0c
BH
2604
2605 case GRO_HELD:
2606 case GRO_MERGED:
2607 break;
5d38a079
HX
2608 }
2609
5d0d9be8
HX
2610 return err;
2611}
2612EXPORT_SYMBOL(napi_skb_finish);
2613
78a478d0
HX
2614void skb_gro_reset_offset(struct sk_buff *skb)
2615{
2616 NAPI_GRO_CB(skb)->data_offset = 0;
2617 NAPI_GRO_CB(skb)->frag0 = NULL;
7489594c 2618 NAPI_GRO_CB(skb)->frag0_len = 0;
78a478d0 2619
78d3fd0b 2620 if (skb->mac_header == skb->tail &&
7489594c 2621 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
78a478d0
HX
2622 NAPI_GRO_CB(skb)->frag0 =
2623 page_address(skb_shinfo(skb)->frags[0].page) +
2624 skb_shinfo(skb)->frags[0].page_offset;
7489594c
HX
2625 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2626 }
78a478d0
HX
2627}
2628EXPORT_SYMBOL(skb_gro_reset_offset);
2629
5d0d9be8
HX
2630int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2631{
86911732
HX
2632 skb_gro_reset_offset(skb);
2633
5d0d9be8 2634 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
d565b0a1
HX
2635}
2636EXPORT_SYMBOL(napi_gro_receive);
2637
96e93eab
HX
2638void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2639{
96e93eab
HX
2640 __skb_pull(skb, skb_headlen(skb));
2641 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2642
2643 napi->skb = skb;
2644}
2645EXPORT_SYMBOL(napi_reuse_skb);
2646
76620aaf 2647struct sk_buff *napi_get_frags(struct napi_struct *napi)
5d38a079 2648{
5d38a079 2649 struct sk_buff *skb = napi->skb;
5d38a079
HX
2650
2651 if (!skb) {
89d71a66
ED
2652 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2653 if (skb)
2654 napi->skb = skb;
80595d59 2655 }
96e93eab
HX
2656 return skb;
2657}
76620aaf 2658EXPORT_SYMBOL(napi_get_frags);
96e93eab 2659
5b252f0c
BH
2660int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2661 gro_result_t ret)
96e93eab 2662{
5d0d9be8 2663 int err = NET_RX_SUCCESS;
96e93eab 2664
5d0d9be8
HX
2665 switch (ret) {
2666 case GRO_NORMAL:
86911732 2667 case GRO_HELD:
86911732
HX
2668 skb->protocol = eth_type_trans(skb, napi->dev);
2669
2670 if (ret == GRO_NORMAL)
2671 return netif_receive_skb(skb);
2672
2673 skb_gro_pull(skb, -ETH_HLEN);
2674 break;
5d38a079 2675
5d0d9be8
HX
2676 case GRO_DROP:
2677 err = NET_RX_DROP;
2678 /* fall through */
5d38a079 2679
5d0d9be8
HX
2680 case GRO_MERGED_FREE:
2681 napi_reuse_skb(napi, skb);
2682 break;
5b252f0c
BH
2683
2684 case GRO_MERGED:
2685 break;
5d0d9be8 2686 }
5d38a079 2687
5d38a079
HX
2688 return err;
2689}
5d0d9be8
HX
2690EXPORT_SYMBOL(napi_frags_finish);
2691
76620aaf
HX
2692struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2693{
2694 struct sk_buff *skb = napi->skb;
2695 struct ethhdr *eth;
a5b1cf28
HX
2696 unsigned int hlen;
2697 unsigned int off;
76620aaf
HX
2698
2699 napi->skb = NULL;
2700
2701 skb_reset_mac_header(skb);
2702 skb_gro_reset_offset(skb);
2703
a5b1cf28
HX
2704 off = skb_gro_offset(skb);
2705 hlen = off + sizeof(*eth);
2706 eth = skb_gro_header_fast(skb, off);
2707 if (skb_gro_header_hard(skb, hlen)) {
2708 eth = skb_gro_header_slow(skb, hlen, off);
2709 if (unlikely(!eth)) {
2710 napi_reuse_skb(napi, skb);
2711 skb = NULL;
2712 goto out;
2713 }
76620aaf
HX
2714 }
2715
2716 skb_gro_pull(skb, sizeof(*eth));
2717
2718 /*
2719 * This works because the only protocols we care about don't require
2720 * special handling. We'll fix it up properly at the end.
2721 */
2722 skb->protocol = eth->h_proto;
2723
2724out:
2725 return skb;
2726}
2727EXPORT_SYMBOL(napi_frags_skb);
2728
2729int napi_gro_frags(struct napi_struct *napi)
5d0d9be8 2730{
76620aaf 2731 struct sk_buff *skb = napi_frags_skb(napi);
5d0d9be8
HX
2732
2733 if (!skb)
2734 return NET_RX_DROP;
2735
2736 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2737}
5d38a079
HX
2738EXPORT_SYMBOL(napi_gro_frags);
2739
bea3348e 2740static int process_backlog(struct napi_struct *napi, int quota)
1da177e4
LT
2741{
2742 int work = 0;
1da177e4
LT
2743 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2744 unsigned long start_time = jiffies;
2745
bea3348e
SH
2746 napi->weight = weight_p;
2747 do {
1da177e4 2748 struct sk_buff *skb;
1da177e4
LT
2749
2750 local_irq_disable();
2751 skb = __skb_dequeue(&queue->input_pkt_queue);
bea3348e 2752 if (!skb) {
8f1ead2d 2753 __napi_complete(napi);
bea3348e 2754 local_irq_enable();
8f1ead2d 2755 break;
bea3348e 2756 }
1da177e4
LT
2757 local_irq_enable();
2758
8f1ead2d 2759 netif_receive_skb(skb);
bea3348e 2760 } while (++work < quota && jiffies == start_time);
1da177e4 2761
bea3348e
SH
2762 return work;
2763}
1da177e4 2764
bea3348e
SH
2765/**
2766 * __napi_schedule - schedule for receive
c4ea43c5 2767 * @n: entry to schedule
bea3348e
SH
2768 *
2769 * The entry's receive function will be scheduled to run
2770 */
b5606c2d 2771void __napi_schedule(struct napi_struct *n)
bea3348e
SH
2772{
2773 unsigned long flags;
1da177e4 2774
bea3348e
SH
2775 local_irq_save(flags);
2776 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2777 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2778 local_irq_restore(flags);
1da177e4 2779}
bea3348e
SH
2780EXPORT_SYMBOL(__napi_schedule);
2781
d565b0a1
HX
2782void __napi_complete(struct napi_struct *n)
2783{
2784 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2785 BUG_ON(n->gro_list);
2786
2787 list_del(&n->poll_list);
2788 smp_mb__before_clear_bit();
2789 clear_bit(NAPI_STATE_SCHED, &n->state);
2790}
2791EXPORT_SYMBOL(__napi_complete);
2792
2793void napi_complete(struct napi_struct *n)
2794{
2795 unsigned long flags;
2796
2797 /*
2798 * don't let napi dequeue from the cpu poll list
2799 * just in case its running on a different cpu
2800 */
2801 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2802 return;
2803
2804 napi_gro_flush(n);
2805 local_irq_save(flags);
2806 __napi_complete(n);
2807 local_irq_restore(flags);
2808}
2809EXPORT_SYMBOL(napi_complete);
2810
2811void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2812 int (*poll)(struct napi_struct *, int), int weight)
2813{
2814 INIT_LIST_HEAD(&napi->poll_list);
4ae5544f 2815 napi->gro_count = 0;
d565b0a1 2816 napi->gro_list = NULL;
5d38a079 2817 napi->skb = NULL;
d565b0a1
HX
2818 napi->poll = poll;
2819 napi->weight = weight;
2820 list_add(&napi->dev_list, &dev->napi_list);
d565b0a1 2821 napi->dev = dev;
5d38a079 2822#ifdef CONFIG_NETPOLL
d565b0a1
HX
2823 spin_lock_init(&napi->poll_lock);
2824 napi->poll_owner = -1;
2825#endif
2826 set_bit(NAPI_STATE_SCHED, &napi->state);
2827}
2828EXPORT_SYMBOL(netif_napi_add);
2829
2830void netif_napi_del(struct napi_struct *napi)
2831{
2832 struct sk_buff *skb, *next;
2833
d7b06636 2834 list_del_init(&napi->dev_list);
76620aaf 2835 napi_free_frags(napi);
d565b0a1
HX
2836
2837 for (skb = napi->gro_list; skb; skb = next) {
2838 next = skb->next;
2839 skb->next = NULL;
2840 kfree_skb(skb);
2841 }
2842
2843 napi->gro_list = NULL;
4ae5544f 2844 napi->gro_count = 0;
d565b0a1
HX
2845}
2846EXPORT_SYMBOL(netif_napi_del);
2847
1da177e4
LT
2848
2849static void net_rx_action(struct softirq_action *h)
2850{
bea3348e 2851 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
24f8b238 2852 unsigned long time_limit = jiffies + 2;
51b0bded 2853 int budget = netdev_budget;
53fb95d3
MM
2854 void *have;
2855
1da177e4
LT
2856 local_irq_disable();
2857
bea3348e
SH
2858 while (!list_empty(list)) {
2859 struct napi_struct *n;
2860 int work, weight;
1da177e4 2861
bea3348e 2862 /* If softirq window is exhuasted then punt.
24f8b238
SH
2863 * Allow this to run for 2 jiffies since which will allow
2864 * an average latency of 1.5/HZ.
bea3348e 2865 */
24f8b238 2866 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
1da177e4
LT
2867 goto softnet_break;
2868
2869 local_irq_enable();
2870
bea3348e
SH
2871 /* Even though interrupts have been re-enabled, this
2872 * access is safe because interrupts can only add new
2873 * entries to the tail of this list, and only ->poll()
2874 * calls can remove this head entry from the list.
2875 */
2876 n = list_entry(list->next, struct napi_struct, poll_list);
1da177e4 2877
bea3348e
SH
2878 have = netpoll_poll_lock(n);
2879
2880 weight = n->weight;
2881
0a7606c1
DM
2882 /* This NAPI_STATE_SCHED test is for avoiding a race
2883 * with netpoll's poll_napi(). Only the entity which
2884 * obtains the lock and sees NAPI_STATE_SCHED set will
2885 * actually make the ->poll() call. Therefore we avoid
2886 * accidently calling ->poll() when NAPI is not scheduled.
2887 */
2888 work = 0;
4ea7e386 2889 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
0a7606c1 2890 work = n->poll(n, weight);
4ea7e386
NH
2891 trace_napi_poll(n);
2892 }
bea3348e
SH
2893
2894 WARN_ON_ONCE(work > weight);
2895
2896 budget -= work;
2897
2898 local_irq_disable();
2899
2900 /* Drivers must not modify the NAPI state if they
2901 * consume the entire weight. In such cases this code
2902 * still "owns" the NAPI instance and therefore can
2903 * move the instance around on the list at-will.
2904 */
fed17f30 2905 if (unlikely(work == weight)) {
ff780cd8
HX
2906 if (unlikely(napi_disable_pending(n))) {
2907 local_irq_enable();
2908 napi_complete(n);
2909 local_irq_disable();
2910 } else
fed17f30
DM
2911 list_move_tail(&n->poll_list, list);
2912 }
bea3348e
SH
2913
2914 netpoll_poll_unlock(have);
1da177e4
LT
2915 }
2916out:
515e06c4 2917 local_irq_enable();
bea3348e 2918
db217334
CL
2919#ifdef CONFIG_NET_DMA
2920 /*
2921 * There may not be any more sk_buffs coming right now, so push
2922 * any pending DMA copies to hardware
2923 */
2ba05622 2924 dma_issue_pending_all();
db217334 2925#endif
bea3348e 2926
1da177e4
LT
2927 return;
2928
2929softnet_break:
2930 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2931 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2932 goto out;
2933}
2934
d1b19dff 2935static gifconf_func_t *gifconf_list[NPROTO];
1da177e4
LT
2936
2937/**
2938 * register_gifconf - register a SIOCGIF handler
2939 * @family: Address family
2940 * @gifconf: Function handler
2941 *
2942 * Register protocol dependent address dumping routines. The handler
2943 * that is passed must not be freed or reused until it has been replaced
2944 * by another handler.
2945 */
d1b19dff 2946int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
1da177e4
LT
2947{
2948 if (family >= NPROTO)
2949 return -EINVAL;
2950 gifconf_list[family] = gifconf;
2951 return 0;
2952}
d1b19dff 2953EXPORT_SYMBOL(register_gifconf);
1da177e4
LT
2954
2955
2956/*
2957 * Map an interface index to its name (SIOCGIFNAME)
2958 */
2959
2960/*
2961 * We need this ioctl for efficient implementation of the
2962 * if_indextoname() function required by the IPv6 API. Without
2963 * it, we would have to search all the interfaces to find a
2964 * match. --pb
2965 */
2966
881d966b 2967static int dev_ifname(struct net *net, struct ifreq __user *arg)
1da177e4
LT
2968{
2969 struct net_device *dev;
2970 struct ifreq ifr;
2971
2972 /*
2973 * Fetch the caller's info block.
2974 */
2975
2976 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2977 return -EFAULT;
2978
fb699dfd
ED
2979 rcu_read_lock();
2980 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
1da177e4 2981 if (!dev) {
fb699dfd 2982 rcu_read_unlock();
1da177e4
LT
2983 return -ENODEV;
2984 }
2985
2986 strcpy(ifr.ifr_name, dev->name);
fb699dfd 2987 rcu_read_unlock();
1da177e4
LT
2988
2989 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2990 return -EFAULT;
2991 return 0;
2992}
2993
2994/*
2995 * Perform a SIOCGIFCONF call. This structure will change
2996 * size eventually, and there is nothing I can do about it.
2997 * Thus we will need a 'compatibility mode'.
2998 */
2999
881d966b 3000static int dev_ifconf(struct net *net, char __user *arg)
1da177e4
LT
3001{
3002 struct ifconf ifc;
3003 struct net_device *dev;
3004 char __user *pos;
3005 int len;
3006 int total;
3007 int i;
3008
3009 /*
3010 * Fetch the caller's info block.
3011 */
3012
3013 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3014 return -EFAULT;
3015
3016 pos = ifc.ifc_buf;
3017 len = ifc.ifc_len;
3018
3019 /*
3020 * Loop over the interfaces, and write an info block for each.
3021 */
3022
3023 total = 0;
881d966b 3024 for_each_netdev(net, dev) {
1da177e4
LT
3025 for (i = 0; i < NPROTO; i++) {
3026 if (gifconf_list[i]) {
3027 int done;
3028 if (!pos)
3029 done = gifconf_list[i](dev, NULL, 0);
3030 else
3031 done = gifconf_list[i](dev, pos + total,
3032 len - total);
3033 if (done < 0)
3034 return -EFAULT;
3035 total += done;
3036 }
3037 }
4ec93edb 3038 }
1da177e4
LT
3039
3040 /*
3041 * All done. Write the updated control block back to the caller.
3042 */
3043 ifc.ifc_len = total;
3044
3045 /*
3046 * Both BSD and Solaris return 0 here, so we do too.
3047 */
3048 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3049}
3050
3051#ifdef CONFIG_PROC_FS
3052/*
3053 * This is invoked by the /proc filesystem handler to display a device
3054 * in detail.
3055 */
7562f876 3056void *dev_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 3057 __acquires(dev_base_lock)
1da177e4 3058{
e372c414 3059 struct net *net = seq_file_net(seq);
7562f876 3060 loff_t off;
1da177e4 3061 struct net_device *dev;
1da177e4 3062
7562f876
PE
3063 read_lock(&dev_base_lock);
3064 if (!*pos)
3065 return SEQ_START_TOKEN;
1da177e4 3066
7562f876 3067 off = 1;
881d966b 3068 for_each_netdev(net, dev)
7562f876
PE
3069 if (off++ == *pos)
3070 return dev;
1da177e4 3071
7562f876 3072 return NULL;
1da177e4
LT
3073}
3074
3075void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3076{
e372c414 3077 struct net *net = seq_file_net(seq);
1da177e4 3078 ++*pos;
7562f876 3079 return v == SEQ_START_TOKEN ?
881d966b 3080 first_net_device(net) : next_net_device((struct net_device *)v);
1da177e4
LT
3081}
3082
3083void dev_seq_stop(struct seq_file *seq, void *v)
9a429c49 3084 __releases(dev_base_lock)
1da177e4
LT
3085{
3086 read_unlock(&dev_base_lock);
3087}
3088
3089static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3090{
eeda3fd6 3091 const struct net_device_stats *stats = dev_get_stats(dev);
1da177e4 3092
5a1b5898
RR
3093 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3094 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3095 dev->name, stats->rx_bytes, stats->rx_packets,
3096 stats->rx_errors,
3097 stats->rx_dropped + stats->rx_missed_errors,
3098 stats->rx_fifo_errors,
3099 stats->rx_length_errors + stats->rx_over_errors +
3100 stats->rx_crc_errors + stats->rx_frame_errors,
3101 stats->rx_compressed, stats->multicast,
3102 stats->tx_bytes, stats->tx_packets,
3103 stats->tx_errors, stats->tx_dropped,
3104 stats->tx_fifo_errors, stats->collisions,
3105 stats->tx_carrier_errors +
3106 stats->tx_aborted_errors +
3107 stats->tx_window_errors +
3108 stats->tx_heartbeat_errors,
3109 stats->tx_compressed);
1da177e4
LT
3110}
3111
3112/*
3113 * Called from the PROCfs module. This now uses the new arbitrary sized
3114 * /proc/net interface to create /proc/net/dev
3115 */
3116static int dev_seq_show(struct seq_file *seq, void *v)
3117{
3118 if (v == SEQ_START_TOKEN)
3119 seq_puts(seq, "Inter-| Receive "
3120 " | Transmit\n"
3121 " face |bytes packets errs drop fifo frame "
3122 "compressed multicast|bytes packets errs "
3123 "drop fifo colls carrier compressed\n");
3124 else
3125 dev_seq_printf_stats(seq, v);
3126 return 0;
3127}
3128
3129static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3130{
3131 struct netif_rx_stats *rc = NULL;
3132
0c0b0aca 3133 while (*pos < nr_cpu_ids)
4ec93edb 3134 if (cpu_online(*pos)) {
1da177e4
LT
3135 rc = &per_cpu(netdev_rx_stat, *pos);
3136 break;
3137 } else
3138 ++*pos;
3139 return rc;
3140}
3141
3142static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3143{
3144 return softnet_get_online(pos);
3145}
3146
3147static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3148{
3149 ++*pos;
3150 return softnet_get_online(pos);
3151}
3152
3153static void softnet_seq_stop(struct seq_file *seq, void *v)
3154{
3155}
3156
3157static int softnet_seq_show(struct seq_file *seq, void *v)
3158{
3159 struct netif_rx_stats *s = v;
3160
3161 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
31aa02c5 3162 s->total, s->dropped, s->time_squeeze, 0,
c1ebcdb8 3163 0, 0, 0, 0, /* was fastroute */
d1b19dff 3164 s->cpu_collision);
1da177e4
LT
3165 return 0;
3166}
3167
f690808e 3168static const struct seq_operations dev_seq_ops = {
1da177e4
LT
3169 .start = dev_seq_start,
3170 .next = dev_seq_next,
3171 .stop = dev_seq_stop,
3172 .show = dev_seq_show,
3173};
3174
3175static int dev_seq_open(struct inode *inode, struct file *file)
3176{
e372c414
DL
3177 return seq_open_net(inode, file, &dev_seq_ops,
3178 sizeof(struct seq_net_private));
1da177e4
LT
3179}
3180
9a32144e 3181static const struct file_operations dev_seq_fops = {
1da177e4
LT
3182 .owner = THIS_MODULE,
3183 .open = dev_seq_open,
3184 .read = seq_read,
3185 .llseek = seq_lseek,
e372c414 3186 .release = seq_release_net,
1da177e4
LT
3187};
3188
f690808e 3189static const struct seq_operations softnet_seq_ops = {
1da177e4
LT
3190 .start = softnet_seq_start,
3191 .next = softnet_seq_next,
3192 .stop = softnet_seq_stop,
3193 .show = softnet_seq_show,
3194};
3195
3196static int softnet_seq_open(struct inode *inode, struct file *file)
3197{
3198 return seq_open(file, &softnet_seq_ops);
3199}
3200
9a32144e 3201static const struct file_operations softnet_seq_fops = {
1da177e4
LT
3202 .owner = THIS_MODULE,
3203 .open = softnet_seq_open,
3204 .read = seq_read,
3205 .llseek = seq_lseek,
3206 .release = seq_release,
3207};
3208
0e1256ff
SH
3209static void *ptype_get_idx(loff_t pos)
3210{
3211 struct packet_type *pt = NULL;
3212 loff_t i = 0;
3213 int t;
3214
3215 list_for_each_entry_rcu(pt, &ptype_all, list) {
3216 if (i == pos)
3217 return pt;
3218 ++i;
3219 }
3220
82d8a867 3221 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
0e1256ff
SH
3222 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3223 if (i == pos)
3224 return pt;
3225 ++i;
3226 }
3227 }
3228 return NULL;
3229}
3230
3231static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
72348a42 3232 __acquires(RCU)
0e1256ff
SH
3233{
3234 rcu_read_lock();
3235 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3236}
3237
3238static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3239{
3240 struct packet_type *pt;
3241 struct list_head *nxt;
3242 int hash;
3243
3244 ++*pos;
3245 if (v == SEQ_START_TOKEN)
3246 return ptype_get_idx(0);
3247
3248 pt = v;
3249 nxt = pt->list.next;
3250 if (pt->type == htons(ETH_P_ALL)) {
3251 if (nxt != &ptype_all)
3252 goto found;
3253 hash = 0;
3254 nxt = ptype_base[0].next;
3255 } else
82d8a867 3256 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
0e1256ff
SH
3257
3258 while (nxt == &ptype_base[hash]) {
82d8a867 3259 if (++hash >= PTYPE_HASH_SIZE)
0e1256ff
SH
3260 return NULL;
3261 nxt = ptype_base[hash].next;
3262 }
3263found:
3264 return list_entry(nxt, struct packet_type, list);
3265}
3266
3267static void ptype_seq_stop(struct seq_file *seq, void *v)
72348a42 3268 __releases(RCU)
0e1256ff
SH
3269{
3270 rcu_read_unlock();
3271}
3272
0e1256ff
SH
3273static int ptype_seq_show(struct seq_file *seq, void *v)
3274{
3275 struct packet_type *pt = v;
3276
3277 if (v == SEQ_START_TOKEN)
3278 seq_puts(seq, "Type Device Function\n");
c346dca1 3279 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
0e1256ff
SH
3280 if (pt->type == htons(ETH_P_ALL))
3281 seq_puts(seq, "ALL ");
3282 else
3283 seq_printf(seq, "%04x", ntohs(pt->type));
3284
908cd2da
AD
3285 seq_printf(seq, " %-8s %pF\n",
3286 pt->dev ? pt->dev->name : "", pt->func);
0e1256ff
SH
3287 }
3288
3289 return 0;
3290}
3291
3292static const struct seq_operations ptype_seq_ops = {
3293 .start = ptype_seq_start,
3294 .next = ptype_seq_next,
3295 .stop = ptype_seq_stop,
3296 .show = ptype_seq_show,
3297};
3298
3299static int ptype_seq_open(struct inode *inode, struct file *file)
3300{
2feb27db
PE
3301 return seq_open_net(inode, file, &ptype_seq_ops,
3302 sizeof(struct seq_net_private));
0e1256ff
SH
3303}
3304
3305static const struct file_operations ptype_seq_fops = {
3306 .owner = THIS_MODULE,
3307 .open = ptype_seq_open,
3308 .read = seq_read,
3309 .llseek = seq_lseek,
2feb27db 3310 .release = seq_release_net,
0e1256ff
SH
3311};
3312
3313
4665079c 3314static int __net_init dev_proc_net_init(struct net *net)
1da177e4
LT
3315{
3316 int rc = -ENOMEM;
3317
881d966b 3318 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
1da177e4 3319 goto out;
881d966b 3320 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
1da177e4 3321 goto out_dev;
881d966b 3322 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
457c4cbc 3323 goto out_softnet;
0e1256ff 3324
881d966b 3325 if (wext_proc_init(net))
457c4cbc 3326 goto out_ptype;
1da177e4
LT
3327 rc = 0;
3328out:
3329 return rc;
457c4cbc 3330out_ptype:
881d966b 3331 proc_net_remove(net, "ptype");
1da177e4 3332out_softnet:
881d966b 3333 proc_net_remove(net, "softnet_stat");
1da177e4 3334out_dev:
881d966b 3335 proc_net_remove(net, "dev");
1da177e4
LT
3336 goto out;
3337}
881d966b 3338
4665079c 3339static void __net_exit dev_proc_net_exit(struct net *net)
881d966b
EB
3340{
3341 wext_proc_exit(net);
3342
3343 proc_net_remove(net, "ptype");
3344 proc_net_remove(net, "softnet_stat");
3345 proc_net_remove(net, "dev");
3346}
3347
022cbae6 3348static struct pernet_operations __net_initdata dev_proc_ops = {
881d966b
EB
3349 .init = dev_proc_net_init,
3350 .exit = dev_proc_net_exit,
3351};
3352
3353static int __init dev_proc_init(void)
3354{
3355 return register_pernet_subsys(&dev_proc_ops);
3356}
1da177e4
LT
3357#else
3358#define dev_proc_init() 0
3359#endif /* CONFIG_PROC_FS */
3360
3361
3362/**
3363 * netdev_set_master - set up master/slave pair
3364 * @slave: slave device
3365 * @master: new master device
3366 *
3367 * Changes the master device of the slave. Pass %NULL to break the
3368 * bonding. The caller must hold the RTNL semaphore. On a failure
3369 * a negative errno code is returned. On success the reference counts
3370 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3371 * function returns zero.
3372 */
3373int netdev_set_master(struct net_device *slave, struct net_device *master)
3374{
3375 struct net_device *old = slave->master;
3376
3377 ASSERT_RTNL();
3378
3379 if (master) {
3380 if (old)
3381 return -EBUSY;
3382 dev_hold(master);
3383 }
3384
3385 slave->master = master;
4ec93edb 3386
1da177e4
LT
3387 synchronize_net();
3388
3389 if (old)
3390 dev_put(old);
3391
3392 if (master)
3393 slave->flags |= IFF_SLAVE;
3394 else
3395 slave->flags &= ~IFF_SLAVE;
3396
3397 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3398 return 0;
3399}
d1b19dff 3400EXPORT_SYMBOL(netdev_set_master);
1da177e4 3401
b6c40d68
PM
3402static void dev_change_rx_flags(struct net_device *dev, int flags)
3403{
d314774c
SH
3404 const struct net_device_ops *ops = dev->netdev_ops;
3405
3406 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3407 ops->ndo_change_rx_flags(dev, flags);
b6c40d68
PM
3408}
3409
dad9b335 3410static int __dev_set_promiscuity(struct net_device *dev, int inc)
1da177e4
LT
3411{
3412 unsigned short old_flags = dev->flags;
8192b0c4
DH
3413 uid_t uid;
3414 gid_t gid;
1da177e4 3415
24023451
PM
3416 ASSERT_RTNL();
3417
dad9b335
WC
3418 dev->flags |= IFF_PROMISC;
3419 dev->promiscuity += inc;
3420 if (dev->promiscuity == 0) {
3421 /*
3422 * Avoid overflow.
3423 * If inc causes overflow, untouch promisc and return error.
3424 */
3425 if (inc < 0)
3426 dev->flags &= ~IFF_PROMISC;
3427 else {
3428 dev->promiscuity -= inc;
3429 printk(KERN_WARNING "%s: promiscuity touches roof, "
3430 "set promiscuity failed, promiscuity feature "
3431 "of device might be broken.\n", dev->name);
3432 return -EOVERFLOW;
3433 }
3434 }
52609c0b 3435 if (dev->flags != old_flags) {
1da177e4
LT
3436 printk(KERN_INFO "device %s %s promiscuous mode\n",
3437 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4ec93edb 3438 "left");
8192b0c4
DH
3439 if (audit_enabled) {
3440 current_uid_gid(&uid, &gid);
7759db82
KHK
3441 audit_log(current->audit_context, GFP_ATOMIC,
3442 AUDIT_ANOM_PROMISCUOUS,
3443 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3444 dev->name, (dev->flags & IFF_PROMISC),
3445 (old_flags & IFF_PROMISC),
3446 audit_get_loginuid(current),
8192b0c4 3447 uid, gid,
7759db82 3448 audit_get_sessionid(current));
8192b0c4 3449 }
24023451 3450
b6c40d68 3451 dev_change_rx_flags(dev, IFF_PROMISC);
1da177e4 3452 }
dad9b335 3453 return 0;
1da177e4
LT
3454}
3455
4417da66
PM
3456/**
3457 * dev_set_promiscuity - update promiscuity count on a device
3458 * @dev: device
3459 * @inc: modifier
3460 *
3461 * Add or remove promiscuity from a device. While the count in the device
3462 * remains above zero the interface remains promiscuous. Once it hits zero
3463 * the device reverts back to normal filtering operation. A negative inc
3464 * value is used to drop promiscuity on the device.
dad9b335 3465 * Return 0 if successful or a negative errno code on error.
4417da66 3466 */
dad9b335 3467int dev_set_promiscuity(struct net_device *dev, int inc)
4417da66
PM
3468{
3469 unsigned short old_flags = dev->flags;
dad9b335 3470 int err;
4417da66 3471
dad9b335 3472 err = __dev_set_promiscuity(dev, inc);
4b5a698e 3473 if (err < 0)
dad9b335 3474 return err;
4417da66
PM
3475 if (dev->flags != old_flags)
3476 dev_set_rx_mode(dev);
dad9b335 3477 return err;
4417da66 3478}
d1b19dff 3479EXPORT_SYMBOL(dev_set_promiscuity);
4417da66 3480
1da177e4
LT
3481/**
3482 * dev_set_allmulti - update allmulti count on a device
3483 * @dev: device
3484 * @inc: modifier
3485 *
3486 * Add or remove reception of all multicast frames to a device. While the
3487 * count in the device remains above zero the interface remains listening
3488 * to all interfaces. Once it hits zero the device reverts back to normal
3489 * filtering operation. A negative @inc value is used to drop the counter
3490 * when releasing a resource needing all multicasts.
dad9b335 3491 * Return 0 if successful or a negative errno code on error.
1da177e4
LT
3492 */
3493
dad9b335 3494int dev_set_allmulti(struct net_device *dev, int inc)
1da177e4
LT
3495{
3496 unsigned short old_flags = dev->flags;
3497
24023451
PM
3498 ASSERT_RTNL();
3499
1da177e4 3500 dev->flags |= IFF_ALLMULTI;
dad9b335
WC
3501 dev->allmulti += inc;
3502 if (dev->allmulti == 0) {
3503 /*
3504 * Avoid overflow.
3505 * If inc causes overflow, untouch allmulti and return error.
3506 */
3507 if (inc < 0)
3508 dev->flags &= ~IFF_ALLMULTI;
3509 else {
3510 dev->allmulti -= inc;
3511 printk(KERN_WARNING "%s: allmulti touches roof, "
3512 "set allmulti failed, allmulti feature of "
3513 "device might be broken.\n", dev->name);
3514 return -EOVERFLOW;
3515 }
3516 }
24023451 3517 if (dev->flags ^ old_flags) {
b6c40d68 3518 dev_change_rx_flags(dev, IFF_ALLMULTI);
4417da66 3519 dev_set_rx_mode(dev);
24023451 3520 }
dad9b335 3521 return 0;
4417da66 3522}
d1b19dff 3523EXPORT_SYMBOL(dev_set_allmulti);
4417da66
PM
3524
3525/*
3526 * Upload unicast and multicast address lists to device and
3527 * configure RX filtering. When the device doesn't support unicast
53ccaae1 3528 * filtering it is put in promiscuous mode while unicast addresses
4417da66
PM
3529 * are present.
3530 */
3531void __dev_set_rx_mode(struct net_device *dev)
3532{
d314774c
SH
3533 const struct net_device_ops *ops = dev->netdev_ops;
3534
4417da66
PM
3535 /* dev_open will call this function so the list will stay sane. */
3536 if (!(dev->flags&IFF_UP))
3537 return;
3538
3539 if (!netif_device_present(dev))
40b77c94 3540 return;
4417da66 3541
d314774c
SH
3542 if (ops->ndo_set_rx_mode)
3543 ops->ndo_set_rx_mode(dev);
4417da66
PM
3544 else {
3545 /* Unicast addresses changes may only happen under the rtnl,
3546 * therefore calling __dev_set_promiscuity here is safe.
3547 */
31278e71 3548 if (dev->uc.count > 0 && !dev->uc_promisc) {
4417da66
PM
3549 __dev_set_promiscuity(dev, 1);
3550 dev->uc_promisc = 1;
31278e71 3551 } else if (dev->uc.count == 0 && dev->uc_promisc) {
4417da66
PM
3552 __dev_set_promiscuity(dev, -1);
3553 dev->uc_promisc = 0;
3554 }
3555
d314774c
SH
3556 if (ops->ndo_set_multicast_list)
3557 ops->ndo_set_multicast_list(dev);
4417da66
PM
3558 }
3559}
3560
3561void dev_set_rx_mode(struct net_device *dev)
3562{
b9e40857 3563 netif_addr_lock_bh(dev);
4417da66 3564 __dev_set_rx_mode(dev);
b9e40857 3565 netif_addr_unlock_bh(dev);
1da177e4
LT
3566}
3567
f001fde5
JP
3568/* hw addresses list handling functions */
3569
31278e71
JP
3570static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3571 int addr_len, unsigned char addr_type)
f001fde5
JP
3572{
3573 struct netdev_hw_addr *ha;
3574 int alloc_size;
3575
3576 if (addr_len > MAX_ADDR_LEN)
3577 return -EINVAL;
3578
31278e71 3579 list_for_each_entry(ha, &list->list, list) {
ccffad25
JP
3580 if (!memcmp(ha->addr, addr, addr_len) &&
3581 ha->type == addr_type) {
3582 ha->refcount++;
3583 return 0;
3584 }
3585 }
3586
3587
f001fde5
JP
3588 alloc_size = sizeof(*ha);
3589 if (alloc_size < L1_CACHE_BYTES)
3590 alloc_size = L1_CACHE_BYTES;
3591 ha = kmalloc(alloc_size, GFP_ATOMIC);
3592 if (!ha)
3593 return -ENOMEM;
3594 memcpy(ha->addr, addr, addr_len);
3595 ha->type = addr_type;
ccffad25
JP
3596 ha->refcount = 1;
3597 ha->synced = false;
31278e71
JP
3598 list_add_tail_rcu(&ha->list, &list->list);
3599 list->count++;
f001fde5
JP
3600 return 0;
3601}
3602
3603static void ha_rcu_free(struct rcu_head *head)
3604{
3605 struct netdev_hw_addr *ha;
3606
3607 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3608 kfree(ha);
3609}
3610
31278e71
JP
3611static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3612 int addr_len, unsigned char addr_type)
f001fde5
JP
3613{
3614 struct netdev_hw_addr *ha;
f001fde5 3615
31278e71 3616 list_for_each_entry(ha, &list->list, list) {
ccffad25 3617 if (!memcmp(ha->addr, addr, addr_len) &&
f001fde5 3618 (ha->type == addr_type || !addr_type)) {
ccffad25
JP
3619 if (--ha->refcount)
3620 return 0;
f001fde5
JP
3621 list_del_rcu(&ha->list);
3622 call_rcu(&ha->rcu_head, ha_rcu_free);
31278e71 3623 list->count--;
f001fde5
JP
3624 return 0;
3625 }
3626 }
3627 return -ENOENT;
3628}
3629
31278e71
JP
3630static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3631 struct netdev_hw_addr_list *from_list,
3632 int addr_len,
ccffad25 3633 unsigned char addr_type)
f001fde5
JP
3634{
3635 int err;
3636 struct netdev_hw_addr *ha, *ha2;
3637 unsigned char type;
3638
31278e71 3639 list_for_each_entry(ha, &from_list->list, list) {
f001fde5 3640 type = addr_type ? addr_type : ha->type;
31278e71 3641 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
f001fde5
JP
3642 if (err)
3643 goto unroll;
3644 }
3645 return 0;
3646
3647unroll:
31278e71 3648 list_for_each_entry(ha2, &from_list->list, list) {
f001fde5
JP
3649 if (ha2 == ha)
3650 break;
3651 type = addr_type ? addr_type : ha2->type;
31278e71 3652 __hw_addr_del(to_list, ha2->addr, addr_len, type);
f001fde5
JP
3653 }
3654 return err;
3655}
3656
31278e71
JP
3657static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3658 struct netdev_hw_addr_list *from_list,
3659 int addr_len,
ccffad25 3660 unsigned char addr_type)
f001fde5
JP
3661{
3662 struct netdev_hw_addr *ha;
3663 unsigned char type;
3664
31278e71 3665 list_for_each_entry(ha, &from_list->list, list) {
f001fde5 3666 type = addr_type ? addr_type : ha->type;
31278e71 3667 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
ccffad25
JP
3668 }
3669}
3670
31278e71
JP
3671static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3672 struct netdev_hw_addr_list *from_list,
ccffad25
JP
3673 int addr_len)
3674{
3675 int err = 0;
3676 struct netdev_hw_addr *ha, *tmp;
3677
31278e71 3678 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
ccffad25 3679 if (!ha->synced) {
31278e71 3680 err = __hw_addr_add(to_list, ha->addr,
ccffad25
JP
3681 addr_len, ha->type);
3682 if (err)
3683 break;
3684 ha->synced = true;
3685 ha->refcount++;
3686 } else if (ha->refcount == 1) {
31278e71
JP
3687 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3688 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
ccffad25 3689 }
f001fde5 3690 }
ccffad25 3691 return err;
f001fde5
JP
3692}
3693
31278e71
JP
3694static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3695 struct netdev_hw_addr_list *from_list,
ccffad25
JP
3696 int addr_len)
3697{
3698 struct netdev_hw_addr *ha, *tmp;
3699
31278e71 3700 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
ccffad25 3701 if (ha->synced) {
31278e71 3702 __hw_addr_del(to_list, ha->addr,
ccffad25
JP
3703 addr_len, ha->type);
3704 ha->synced = false;
31278e71 3705 __hw_addr_del(from_list, ha->addr,
ccffad25
JP
3706 addr_len, ha->type);
3707 }
3708 }
3709}
3710
31278e71 3711static void __hw_addr_flush(struct netdev_hw_addr_list *list)
f001fde5
JP
3712{
3713 struct netdev_hw_addr *ha, *tmp;
3714
31278e71 3715 list_for_each_entry_safe(ha, tmp, &list->list, list) {
f001fde5
JP
3716 list_del_rcu(&ha->list);
3717 call_rcu(&ha->rcu_head, ha_rcu_free);
3718 }
31278e71
JP
3719 list->count = 0;
3720}
3721
3722static void __hw_addr_init(struct netdev_hw_addr_list *list)
3723{
3724 INIT_LIST_HEAD(&list->list);
3725 list->count = 0;
f001fde5
JP
3726}
3727
3728/* Device addresses handling functions */
3729
3730static void dev_addr_flush(struct net_device *dev)
3731{
3732 /* rtnl_mutex must be held here */
3733
31278e71 3734 __hw_addr_flush(&dev->dev_addrs);
f001fde5
JP
3735 dev->dev_addr = NULL;
3736}
3737
3738static int dev_addr_init(struct net_device *dev)
3739{
3740 unsigned char addr[MAX_ADDR_LEN];
3741 struct netdev_hw_addr *ha;
3742 int err;
3743
3744 /* rtnl_mutex must be held here */
3745
31278e71 3746 __hw_addr_init(&dev->dev_addrs);
0c27922e 3747 memset(addr, 0, sizeof(addr));
31278e71 3748 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
f001fde5
JP
3749 NETDEV_HW_ADDR_T_LAN);
3750 if (!err) {
3751 /*
3752 * Get the first (previously created) address from the list
3753 * and set dev_addr pointer to this location.
3754 */
31278e71 3755 ha = list_first_entry(&dev->dev_addrs.list,
f001fde5
JP
3756 struct netdev_hw_addr, list);
3757 dev->dev_addr = ha->addr;
3758 }
3759 return err;
3760}
3761
3762/**
3763 * dev_addr_add - Add a device address
3764 * @dev: device
3765 * @addr: address to add
3766 * @addr_type: address type
3767 *
3768 * Add a device address to the device or increase the reference count if
3769 * it already exists.
3770 *
3771 * The caller must hold the rtnl_mutex.
3772 */
3773int dev_addr_add(struct net_device *dev, unsigned char *addr,
3774 unsigned char addr_type)
3775{
3776 int err;
3777
3778 ASSERT_RTNL();
3779
31278e71 3780 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
f001fde5
JP
3781 if (!err)
3782 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3783 return err;
3784}
3785EXPORT_SYMBOL(dev_addr_add);
3786
3787/**
3788 * dev_addr_del - Release a device address.
3789 * @dev: device
3790 * @addr: address to delete
3791 * @addr_type: address type
3792 *
3793 * Release reference to a device address and remove it from the device
3794 * if the reference count drops to zero.
3795 *
3796 * The caller must hold the rtnl_mutex.
3797 */
3798int dev_addr_del(struct net_device *dev, unsigned char *addr,
3799 unsigned char addr_type)
3800{
3801 int err;
ccffad25 3802 struct netdev_hw_addr *ha;
f001fde5
JP
3803
3804 ASSERT_RTNL();
3805
ccffad25
JP
3806 /*
3807 * We can not remove the first address from the list because
3808 * dev->dev_addr points to that.
3809 */
31278e71
JP
3810 ha = list_first_entry(&dev->dev_addrs.list,
3811 struct netdev_hw_addr, list);
ccffad25
JP
3812 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3813 return -ENOENT;
3814
31278e71 3815 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
ccffad25 3816 addr_type);
f001fde5
JP
3817 if (!err)
3818 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3819 return err;
3820}
3821EXPORT_SYMBOL(dev_addr_del);
3822
3823/**
3824 * dev_addr_add_multiple - Add device addresses from another device
3825 * @to_dev: device to which addresses will be added
3826 * @from_dev: device from which addresses will be added
3827 * @addr_type: address type - 0 means type will be used from from_dev
3828 *
3829 * Add device addresses of the one device to another.
3830 **
3831 * The caller must hold the rtnl_mutex.
3832 */
3833int dev_addr_add_multiple(struct net_device *to_dev,
3834 struct net_device *from_dev,
3835 unsigned char addr_type)
3836{
3837 int err;
3838
3839 ASSERT_RTNL();
3840
3841 if (from_dev->addr_len != to_dev->addr_len)
3842 return -EINVAL;
31278e71 3843 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
ccffad25 3844 to_dev->addr_len, addr_type);
f001fde5
JP
3845 if (!err)
3846 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3847 return err;
3848}
3849EXPORT_SYMBOL(dev_addr_add_multiple);
3850
3851/**
3852 * dev_addr_del_multiple - Delete device addresses by another device
3853 * @to_dev: device where the addresses will be deleted
3854 * @from_dev: device by which addresses the addresses will be deleted
3855 * @addr_type: address type - 0 means type will used from from_dev
3856 *
3857 * Deletes addresses in to device by the list of addresses in from device.
3858 *
3859 * The caller must hold the rtnl_mutex.
3860 */
3861int dev_addr_del_multiple(struct net_device *to_dev,
3862 struct net_device *from_dev,
3863 unsigned char addr_type)
3864{
3865 ASSERT_RTNL();
3866
3867 if (from_dev->addr_len != to_dev->addr_len)
3868 return -EINVAL;
31278e71 3869 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
ccffad25 3870 to_dev->addr_len, addr_type);
f001fde5
JP
3871 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3872 return 0;
3873}
3874EXPORT_SYMBOL(dev_addr_del_multiple);
3875
31278e71 3876/* multicast addresses handling functions */
f001fde5 3877
61cbc2fc
PM
3878int __dev_addr_delete(struct dev_addr_list **list, int *count,
3879 void *addr, int alen, int glbl)
bf742482
PM
3880{
3881 struct dev_addr_list *da;
3882
3883 for (; (da = *list) != NULL; list = &da->next) {
3884 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3885 alen == da->da_addrlen) {
3886 if (glbl) {
3887 int old_glbl = da->da_gusers;
3888 da->da_gusers = 0;
3889 if (old_glbl == 0)
3890 break;
3891 }
3892 if (--da->da_users)
3893 return 0;
3894
3895 *list = da->next;
3896 kfree(da);
61cbc2fc 3897 (*count)--;
bf742482
PM
3898 return 0;
3899 }
3900 }
3901 return -ENOENT;
3902}
3903
61cbc2fc
PM
3904int __dev_addr_add(struct dev_addr_list **list, int *count,
3905 void *addr, int alen, int glbl)
bf742482
PM
3906{
3907 struct dev_addr_list *da;
3908
3909 for (da = *list; da != NULL; da = da->next) {
3910 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3911 da->da_addrlen == alen) {
3912 if (glbl) {
3913 int old_glbl = da->da_gusers;
3914 da->da_gusers = 1;
3915 if (old_glbl)
3916 return 0;
3917 }
3918 da->da_users++;
3919 return 0;
3920 }
3921 }
3922
12aa343a 3923 da = kzalloc(sizeof(*da), GFP_ATOMIC);
bf742482
PM
3924 if (da == NULL)
3925 return -ENOMEM;
3926 memcpy(da->da_addr, addr, alen);
3927 da->da_addrlen = alen;
3928 da->da_users = 1;
3929 da->da_gusers = glbl ? 1 : 0;
3930 da->next = *list;
3931 *list = da;
61cbc2fc 3932 (*count)++;
bf742482
PM
3933 return 0;
3934}
3935
4417da66
PM
3936/**
3937 * dev_unicast_delete - Release secondary unicast address.
3938 * @dev: device
0ed72ec4 3939 * @addr: address to delete
4417da66
PM
3940 *
3941 * Release reference to a secondary unicast address and remove it
0ed72ec4 3942 * from the device if the reference count drops to zero.
4417da66
PM
3943 *
3944 * The caller must hold the rtnl_mutex.
3945 */
ccffad25 3946int dev_unicast_delete(struct net_device *dev, void *addr)
4417da66
PM
3947{
3948 int err;
3949
3950 ASSERT_RTNL();
3951
a6ac65db 3952 netif_addr_lock_bh(dev);
31278e71
JP
3953 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3954 NETDEV_HW_ADDR_T_UNICAST);
61cbc2fc 3955 if (!err)
4417da66 3956 __dev_set_rx_mode(dev);
a6ac65db 3957 netif_addr_unlock_bh(dev);
4417da66
PM
3958 return err;
3959}
3960EXPORT_SYMBOL(dev_unicast_delete);
3961
3962/**
3963 * dev_unicast_add - add a secondary unicast address
3964 * @dev: device
5dbaec5d 3965 * @addr: address to add
4417da66
PM
3966 *
3967 * Add a secondary unicast address to the device or increase
3968 * the reference count if it already exists.
3969 *
3970 * The caller must hold the rtnl_mutex.
3971 */
ccffad25 3972int dev_unicast_add(struct net_device *dev, void *addr)
4417da66
PM
3973{
3974 int err;
3975
3976 ASSERT_RTNL();
3977
a6ac65db 3978 netif_addr_lock_bh(dev);
31278e71
JP
3979 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
3980 NETDEV_HW_ADDR_T_UNICAST);
61cbc2fc 3981 if (!err)
4417da66 3982 __dev_set_rx_mode(dev);
a6ac65db 3983 netif_addr_unlock_bh(dev);
4417da66
PM
3984 return err;
3985}
3986EXPORT_SYMBOL(dev_unicast_add);
3987
e83a2ea8
CL
3988int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3989 struct dev_addr_list **from, int *from_count)
3990{
3991 struct dev_addr_list *da, *next;
3992 int err = 0;
3993
3994 da = *from;
3995 while (da != NULL) {
3996 next = da->next;
3997 if (!da->da_synced) {
3998 err = __dev_addr_add(to, to_count,
3999 da->da_addr, da->da_addrlen, 0);
4000 if (err < 0)
4001 break;
4002 da->da_synced = 1;
4003 da->da_users++;
4004 } else if (da->da_users == 1) {
4005 __dev_addr_delete(to, to_count,
4006 da->da_addr, da->da_addrlen, 0);
4007 __dev_addr_delete(from, from_count,
4008 da->da_addr, da->da_addrlen, 0);
4009 }
4010 da = next;
4011 }
4012 return err;
4013}
c4029083 4014EXPORT_SYMBOL_GPL(__dev_addr_sync);
e83a2ea8
CL
4015
4016void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4017 struct dev_addr_list **from, int *from_count)
4018{
4019 struct dev_addr_list *da, *next;
4020
4021 da = *from;
4022 while (da != NULL) {
4023 next = da->next;
4024 if (da->da_synced) {
4025 __dev_addr_delete(to, to_count,
4026 da->da_addr, da->da_addrlen, 0);
4027 da->da_synced = 0;
4028 __dev_addr_delete(from, from_count,
4029 da->da_addr, da->da_addrlen, 0);
4030 }
4031 da = next;
4032 }
4033}
c4029083 4034EXPORT_SYMBOL_GPL(__dev_addr_unsync);
e83a2ea8
CL
4035
4036/**
4037 * dev_unicast_sync - Synchronize device's unicast list to another device
4038 * @to: destination device
4039 * @from: source device
4040 *
4041 * Add newly added addresses to the destination device and release
a6ac65db
JP
4042 * addresses that have no users left. The source device must be
4043 * locked by netif_tx_lock_bh.
e83a2ea8
CL
4044 *
4045 * This function is intended to be called from the dev->set_rx_mode
4046 * function of layered software devices.
4047 */
4048int dev_unicast_sync(struct net_device *to, struct net_device *from)
4049{
4050 int err = 0;
4051
ccffad25
JP
4052 if (to->addr_len != from->addr_len)
4053 return -EINVAL;
4054
a6ac65db 4055 netif_addr_lock_bh(to);
31278e71 4056 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
e83a2ea8
CL
4057 if (!err)
4058 __dev_set_rx_mode(to);
a6ac65db 4059 netif_addr_unlock_bh(to);
e83a2ea8
CL
4060 return err;
4061}
4062EXPORT_SYMBOL(dev_unicast_sync);
4063
4064/**
bc2cda1e 4065 * dev_unicast_unsync - Remove synchronized addresses from the destination device
e83a2ea8
CL
4066 * @to: destination device
4067 * @from: source device
4068 *
4069 * Remove all addresses that were added to the destination device by
4070 * dev_unicast_sync(). This function is intended to be called from the
4071 * dev->stop function of layered software devices.
4072 */
4073void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4074{
ccffad25
JP
4075 if (to->addr_len != from->addr_len)
4076 return;
e83a2ea8 4077
a6ac65db
JP
4078 netif_addr_lock_bh(from);
4079 netif_addr_lock(to);
31278e71 4080 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
ccffad25 4081 __dev_set_rx_mode(to);
a6ac65db
JP
4082 netif_addr_unlock(to);
4083 netif_addr_unlock_bh(from);
e83a2ea8
CL
4084}
4085EXPORT_SYMBOL(dev_unicast_unsync);
4086
ccffad25
JP
4087static void dev_unicast_flush(struct net_device *dev)
4088{
a6ac65db 4089 netif_addr_lock_bh(dev);
31278e71 4090 __hw_addr_flush(&dev->uc);
a6ac65db 4091 netif_addr_unlock_bh(dev);
ccffad25
JP
4092}
4093
4094static void dev_unicast_init(struct net_device *dev)
4095{
31278e71 4096 __hw_addr_init(&dev->uc);
ccffad25
JP
4097}
4098
4099
12972621
DC
4100static void __dev_addr_discard(struct dev_addr_list **list)
4101{
4102 struct dev_addr_list *tmp;
4103
4104 while (*list != NULL) {
4105 tmp = *list;
4106 *list = tmp->next;
4107 if (tmp->da_users > tmp->da_gusers)
4108 printk("__dev_addr_discard: address leakage! "
4109 "da_users=%d\n", tmp->da_users);
4110 kfree(tmp);
4111 }
4112}
4113
26cc2522 4114static void dev_addr_discard(struct net_device *dev)
4417da66 4115{
b9e40857 4116 netif_addr_lock_bh(dev);
26cc2522 4117
456ad75c
DC
4118 __dev_addr_discard(&dev->mc_list);
4119 dev->mc_count = 0;
26cc2522 4120
b9e40857 4121 netif_addr_unlock_bh(dev);
456ad75c
DC
4122}
4123
f0db275a
SH
4124/**
4125 * dev_get_flags - get flags reported to userspace
4126 * @dev: device
4127 *
4128 * Get the combination of flag bits exported through APIs to userspace.
4129 */
1da177e4
LT
4130unsigned dev_get_flags(const struct net_device *dev)
4131{
4132 unsigned flags;
4133
4134 flags = (dev->flags & ~(IFF_PROMISC |
4135 IFF_ALLMULTI |
b00055aa
SR
4136 IFF_RUNNING |
4137 IFF_LOWER_UP |
4138 IFF_DORMANT)) |
1da177e4
LT
4139 (dev->gflags & (IFF_PROMISC |
4140 IFF_ALLMULTI));
4141
b00055aa
SR
4142 if (netif_running(dev)) {
4143 if (netif_oper_up(dev))
4144 flags |= IFF_RUNNING;
4145 if (netif_carrier_ok(dev))
4146 flags |= IFF_LOWER_UP;
4147 if (netif_dormant(dev))
4148 flags |= IFF_DORMANT;
4149 }
1da177e4
LT
4150
4151 return flags;
4152}
d1b19dff 4153EXPORT_SYMBOL(dev_get_flags);
1da177e4 4154
f0db275a
SH
4155/**
4156 * dev_change_flags - change device settings
4157 * @dev: device
4158 * @flags: device state flags
4159 *
4160 * Change settings on device based state flags. The flags are
4161 * in the userspace exported format.
4162 */
1da177e4
LT
4163int dev_change_flags(struct net_device *dev, unsigned flags)
4164{
7c355f53 4165 int ret, changes;
1da177e4
LT
4166 int old_flags = dev->flags;
4167
24023451
PM
4168 ASSERT_RTNL();
4169
1da177e4
LT
4170 /*
4171 * Set the flags on our device.
4172 */
4173
4174 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4175 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4176 IFF_AUTOMEDIA)) |
4177 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4178 IFF_ALLMULTI));
4179
4180 /*
4181 * Load in the correct multicast list now the flags have changed.
4182 */
4183
b6c40d68
PM
4184 if ((old_flags ^ flags) & IFF_MULTICAST)
4185 dev_change_rx_flags(dev, IFF_MULTICAST);
24023451 4186
4417da66 4187 dev_set_rx_mode(dev);
1da177e4
LT
4188
4189 /*
4190 * Have we downed the interface. We handle IFF_UP ourselves
4191 * according to user attempts to set it, rather than blindly
4192 * setting it.
4193 */
4194
4195 ret = 0;
4196 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4197 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4198
4199 if (!ret)
4417da66 4200 dev_set_rx_mode(dev);
1da177e4
LT
4201 }
4202
4203 if (dev->flags & IFF_UP &&
d1b19dff 4204 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
1da177e4 4205 IFF_VOLATILE)))
056925ab 4206 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1da177e4
LT
4207
4208 if ((flags ^ dev->gflags) & IFF_PROMISC) {
d1b19dff
ED
4209 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4210
1da177e4
LT
4211 dev->gflags ^= IFF_PROMISC;
4212 dev_set_promiscuity(dev, inc);
4213 }
4214
4215 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4216 is important. Some (broken) drivers set IFF_PROMISC, when
4217 IFF_ALLMULTI is requested not asking us and not reporting.
4218 */
4219 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
d1b19dff
ED
4220 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4221
1da177e4
LT
4222 dev->gflags ^= IFF_ALLMULTI;
4223 dev_set_allmulti(dev, inc);
4224 }
4225
7c355f53
TG
4226 /* Exclude state transition flags, already notified */
4227 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4228 if (changes)
4229 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
1da177e4
LT
4230
4231 return ret;
4232}
d1b19dff 4233EXPORT_SYMBOL(dev_change_flags);
1da177e4 4234
f0db275a
SH
4235/**
4236 * dev_set_mtu - Change maximum transfer unit
4237 * @dev: device
4238 * @new_mtu: new transfer unit
4239 *
4240 * Change the maximum transfer size of the network device.
4241 */
1da177e4
LT
4242int dev_set_mtu(struct net_device *dev, int new_mtu)
4243{
d314774c 4244 const struct net_device_ops *ops = dev->netdev_ops;
1da177e4
LT
4245 int err;
4246
4247 if (new_mtu == dev->mtu)
4248 return 0;
4249
4250 /* MTU must be positive. */
4251 if (new_mtu < 0)
4252 return -EINVAL;
4253
4254 if (!netif_device_present(dev))
4255 return -ENODEV;
4256
4257 err = 0;
d314774c
SH
4258 if (ops->ndo_change_mtu)
4259 err = ops->ndo_change_mtu(dev, new_mtu);
1da177e4
LT
4260 else
4261 dev->mtu = new_mtu;
d314774c 4262
1da177e4 4263 if (!err && dev->flags & IFF_UP)
056925ab 4264 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
1da177e4
LT
4265 return err;
4266}
d1b19dff 4267EXPORT_SYMBOL(dev_set_mtu);
1da177e4 4268
f0db275a
SH
4269/**
4270 * dev_set_mac_address - Change Media Access Control Address
4271 * @dev: device
4272 * @sa: new address
4273 *
4274 * Change the hardware (MAC) address of the device
4275 */
1da177e4
LT
4276int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4277{
d314774c 4278 const struct net_device_ops *ops = dev->netdev_ops;
1da177e4
LT
4279 int err;
4280
d314774c 4281 if (!ops->ndo_set_mac_address)
1da177e4
LT
4282 return -EOPNOTSUPP;
4283 if (sa->sa_family != dev->type)
4284 return -EINVAL;
4285 if (!netif_device_present(dev))
4286 return -ENODEV;
d314774c 4287 err = ops->ndo_set_mac_address(dev, sa);
1da177e4 4288 if (!err)
056925ab 4289 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
1da177e4
LT
4290 return err;
4291}
d1b19dff 4292EXPORT_SYMBOL(dev_set_mac_address);
1da177e4
LT
4293
4294/*
14e3e079 4295 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
1da177e4 4296 */
14e3e079 4297static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
1da177e4
LT
4298{
4299 int err;
881d966b 4300 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
1da177e4
LT
4301
4302 if (!dev)
4303 return -ENODEV;
4304
4305 switch (cmd) {
d1b19dff
ED
4306 case SIOCGIFFLAGS: /* Get interface flags */
4307 ifr->ifr_flags = (short) dev_get_flags(dev);
4308 return 0;
1da177e4 4309
d1b19dff
ED
4310 case SIOCGIFMETRIC: /* Get the metric on the interface
4311 (currently unused) */
4312 ifr->ifr_metric = 0;
4313 return 0;
1da177e4 4314
d1b19dff
ED
4315 case SIOCGIFMTU: /* Get the MTU of a device */
4316 ifr->ifr_mtu = dev->mtu;
4317 return 0;
1da177e4 4318
d1b19dff
ED
4319 case SIOCGIFHWADDR:
4320 if (!dev->addr_len)
4321 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4322 else
4323 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4324 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4325 ifr->ifr_hwaddr.sa_family = dev->type;
4326 return 0;
1da177e4 4327
d1b19dff
ED
4328 case SIOCGIFSLAVE:
4329 err = -EINVAL;
4330 break;
14e3e079 4331
d1b19dff
ED
4332 case SIOCGIFMAP:
4333 ifr->ifr_map.mem_start = dev->mem_start;
4334 ifr->ifr_map.mem_end = dev->mem_end;
4335 ifr->ifr_map.base_addr = dev->base_addr;
4336 ifr->ifr_map.irq = dev->irq;
4337 ifr->ifr_map.dma = dev->dma;
4338 ifr->ifr_map.port = dev->if_port;
4339 return 0;
14e3e079 4340
d1b19dff
ED
4341 case SIOCGIFINDEX:
4342 ifr->ifr_ifindex = dev->ifindex;
4343 return 0;
14e3e079 4344
d1b19dff
ED
4345 case SIOCGIFTXQLEN:
4346 ifr->ifr_qlen = dev->tx_queue_len;
4347 return 0;
14e3e079 4348
d1b19dff
ED
4349 default:
4350 /* dev_ioctl() should ensure this case
4351 * is never reached
4352 */
4353 WARN_ON(1);
4354 err = -EINVAL;
4355 break;
14e3e079
JG
4356
4357 }
4358 return err;
4359}
4360
4361/*
4362 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4363 */
4364static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4365{
4366 int err;
4367 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
5f2f6da7 4368 const struct net_device_ops *ops;
14e3e079
JG
4369
4370 if (!dev)
4371 return -ENODEV;
4372
5f2f6da7
JP
4373 ops = dev->netdev_ops;
4374
14e3e079 4375 switch (cmd) {
d1b19dff
ED
4376 case SIOCSIFFLAGS: /* Set interface flags */
4377 return dev_change_flags(dev, ifr->ifr_flags);
14e3e079 4378
d1b19dff
ED
4379 case SIOCSIFMETRIC: /* Set the metric on the interface
4380 (currently unused) */
4381 return -EOPNOTSUPP;
14e3e079 4382
d1b19dff
ED
4383 case SIOCSIFMTU: /* Set the MTU of a device */
4384 return dev_set_mtu(dev, ifr->ifr_mtu);
1da177e4 4385
d1b19dff
ED
4386 case SIOCSIFHWADDR:
4387 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
1da177e4 4388
d1b19dff
ED
4389 case SIOCSIFHWBROADCAST:
4390 if (ifr->ifr_hwaddr.sa_family != dev->type)
4391 return -EINVAL;
4392 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4393 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4394 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4395 return 0;
1da177e4 4396
d1b19dff
ED
4397 case SIOCSIFMAP:
4398 if (ops->ndo_set_config) {
1da177e4
LT
4399 if (!netif_device_present(dev))
4400 return -ENODEV;
d1b19dff
ED
4401 return ops->ndo_set_config(dev, &ifr->ifr_map);
4402 }
4403 return -EOPNOTSUPP;
1da177e4 4404
d1b19dff
ED
4405 case SIOCADDMULTI:
4406 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4407 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4408 return -EINVAL;
4409 if (!netif_device_present(dev))
4410 return -ENODEV;
4411 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4412 dev->addr_len, 1);
4413
4414 case SIOCDELMULTI:
4415 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4416 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4417 return -EINVAL;
4418 if (!netif_device_present(dev))
4419 return -ENODEV;
4420 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4421 dev->addr_len, 1);
1da177e4 4422
d1b19dff
ED
4423 case SIOCSIFTXQLEN:
4424 if (ifr->ifr_qlen < 0)
4425 return -EINVAL;
4426 dev->tx_queue_len = ifr->ifr_qlen;
4427 return 0;
1da177e4 4428
d1b19dff
ED
4429 case SIOCSIFNAME:
4430 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4431 return dev_change_name(dev, ifr->ifr_newname);
1da177e4 4432
d1b19dff
ED
4433 /*
4434 * Unknown or private ioctl
4435 */
4436 default:
4437 if ((cmd >= SIOCDEVPRIVATE &&
4438 cmd <= SIOCDEVPRIVATE + 15) ||
4439 cmd == SIOCBONDENSLAVE ||
4440 cmd == SIOCBONDRELEASE ||
4441 cmd == SIOCBONDSETHWADDR ||
4442 cmd == SIOCBONDSLAVEINFOQUERY ||
4443 cmd == SIOCBONDINFOQUERY ||
4444 cmd == SIOCBONDCHANGEACTIVE ||
4445 cmd == SIOCGMIIPHY ||
4446 cmd == SIOCGMIIREG ||
4447 cmd == SIOCSMIIREG ||
4448 cmd == SIOCBRADDIF ||
4449 cmd == SIOCBRDELIF ||
4450 cmd == SIOCSHWTSTAMP ||
4451 cmd == SIOCWANDEV) {
4452 err = -EOPNOTSUPP;
4453 if (ops->ndo_do_ioctl) {
4454 if (netif_device_present(dev))
4455 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4456 else
4457 err = -ENODEV;
4458 }
4459 } else
4460 err = -EINVAL;
1da177e4
LT
4461
4462 }
4463 return err;
4464}
4465
4466/*
4467 * This function handles all "interface"-type I/O control requests. The actual
4468 * 'doing' part of this is dev_ifsioc above.
4469 */
4470
4471/**
4472 * dev_ioctl - network device ioctl
c4ea43c5 4473 * @net: the applicable net namespace
1da177e4
LT
4474 * @cmd: command to issue
4475 * @arg: pointer to a struct ifreq in user space
4476 *
4477 * Issue ioctl functions to devices. This is normally called by the
4478 * user space syscall interfaces but can sometimes be useful for
4479 * other purposes. The return value is the return from the syscall if
4480 * positive or a negative errno code on error.
4481 */
4482
881d966b 4483int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
1da177e4
LT
4484{
4485 struct ifreq ifr;
4486 int ret;
4487 char *colon;
4488
4489 /* One special case: SIOCGIFCONF takes ifconf argument
4490 and requires shared lock, because it sleeps writing
4491 to user space.
4492 */
4493
4494 if (cmd == SIOCGIFCONF) {
6756ae4b 4495 rtnl_lock();
881d966b 4496 ret = dev_ifconf(net, (char __user *) arg);
6756ae4b 4497 rtnl_unlock();
1da177e4
LT
4498 return ret;
4499 }
4500 if (cmd == SIOCGIFNAME)
881d966b 4501 return dev_ifname(net, (struct ifreq __user *)arg);
1da177e4
LT
4502
4503 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4504 return -EFAULT;
4505
4506 ifr.ifr_name[IFNAMSIZ-1] = 0;
4507
4508 colon = strchr(ifr.ifr_name, ':');
4509 if (colon)
4510 *colon = 0;
4511
4512 /*
4513 * See which interface the caller is talking about.
4514 */
4515
4516 switch (cmd) {
d1b19dff
ED
4517 /*
4518 * These ioctl calls:
4519 * - can be done by all.
4520 * - atomic and do not require locking.
4521 * - return a value
4522 */
4523 case SIOCGIFFLAGS:
4524 case SIOCGIFMETRIC:
4525 case SIOCGIFMTU:
4526 case SIOCGIFHWADDR:
4527 case SIOCGIFSLAVE:
4528 case SIOCGIFMAP:
4529 case SIOCGIFINDEX:
4530 case SIOCGIFTXQLEN:
4531 dev_load(net, ifr.ifr_name);
4532 read_lock(&dev_base_lock);
4533 ret = dev_ifsioc_locked(net, &ifr, cmd);
4534 read_unlock(&dev_base_lock);
4535 if (!ret) {
4536 if (colon)
4537 *colon = ':';
4538 if (copy_to_user(arg, &ifr,
4539 sizeof(struct ifreq)))
4540 ret = -EFAULT;
4541 }
4542 return ret;
1da177e4 4543
d1b19dff
ED
4544 case SIOCETHTOOL:
4545 dev_load(net, ifr.ifr_name);
4546 rtnl_lock();
4547 ret = dev_ethtool(net, &ifr);
4548 rtnl_unlock();
4549 if (!ret) {
4550 if (colon)
4551 *colon = ':';
4552 if (copy_to_user(arg, &ifr,
4553 sizeof(struct ifreq)))
4554 ret = -EFAULT;
4555 }
4556 return ret;
1da177e4 4557
d1b19dff
ED
4558 /*
4559 * These ioctl calls:
4560 * - require superuser power.
4561 * - require strict serialization.
4562 * - return a value
4563 */
4564 case SIOCGMIIPHY:
4565 case SIOCGMIIREG:
4566 case SIOCSIFNAME:
4567 if (!capable(CAP_NET_ADMIN))
4568 return -EPERM;
4569 dev_load(net, ifr.ifr_name);
4570 rtnl_lock();
4571 ret = dev_ifsioc(net, &ifr, cmd);
4572 rtnl_unlock();
4573 if (!ret) {
4574 if (colon)
4575 *colon = ':';
4576 if (copy_to_user(arg, &ifr,
4577 sizeof(struct ifreq)))
4578 ret = -EFAULT;
4579 }
4580 return ret;
1da177e4 4581
d1b19dff
ED
4582 /*
4583 * These ioctl calls:
4584 * - require superuser power.
4585 * - require strict serialization.
4586 * - do not return a value
4587 */
4588 case SIOCSIFFLAGS:
4589 case SIOCSIFMETRIC:
4590 case SIOCSIFMTU:
4591 case SIOCSIFMAP:
4592 case SIOCSIFHWADDR:
4593 case SIOCSIFSLAVE:
4594 case SIOCADDMULTI:
4595 case SIOCDELMULTI:
4596 case SIOCSIFHWBROADCAST:
4597 case SIOCSIFTXQLEN:
4598 case SIOCSMIIREG:
4599 case SIOCBONDENSLAVE:
4600 case SIOCBONDRELEASE:
4601 case SIOCBONDSETHWADDR:
4602 case SIOCBONDCHANGEACTIVE:
4603 case SIOCBRADDIF:
4604 case SIOCBRDELIF:
4605 case SIOCSHWTSTAMP:
4606 if (!capable(CAP_NET_ADMIN))
4607 return -EPERM;
4608 /* fall through */
4609 case SIOCBONDSLAVEINFOQUERY:
4610 case SIOCBONDINFOQUERY:
4611 dev_load(net, ifr.ifr_name);
4612 rtnl_lock();
4613 ret = dev_ifsioc(net, &ifr, cmd);
4614 rtnl_unlock();
4615 return ret;
4616
4617 case SIOCGIFMEM:
4618 /* Get the per device memory space. We can add this but
4619 * currently do not support it */
4620 case SIOCSIFMEM:
4621 /* Set the per device memory buffer space.
4622 * Not applicable in our case */
4623 case SIOCSIFLINK:
4624 return -EINVAL;
4625
4626 /*
4627 * Unknown or private ioctl.
4628 */
4629 default:
4630 if (cmd == SIOCWANDEV ||
4631 (cmd >= SIOCDEVPRIVATE &&
4632 cmd <= SIOCDEVPRIVATE + 15)) {
881d966b 4633 dev_load(net, ifr.ifr_name);
1da177e4 4634 rtnl_lock();
881d966b 4635 ret = dev_ifsioc(net, &ifr, cmd);
1da177e4 4636 rtnl_unlock();
d1b19dff
ED
4637 if (!ret && copy_to_user(arg, &ifr,
4638 sizeof(struct ifreq)))
4639 ret = -EFAULT;
1da177e4 4640 return ret;
d1b19dff
ED
4641 }
4642 /* Take care of Wireless Extensions */
4643 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4644 return wext_handle_ioctl(net, &ifr, cmd, arg);
4645 return -EINVAL;
1da177e4
LT
4646 }
4647}
4648
4649
4650/**
4651 * dev_new_index - allocate an ifindex
c4ea43c5 4652 * @net: the applicable net namespace
1da177e4
LT
4653 *
4654 * Returns a suitable unique value for a new device interface
4655 * number. The caller must hold the rtnl semaphore or the
4656 * dev_base_lock to be sure it remains unique.
4657 */
881d966b 4658static int dev_new_index(struct net *net)
1da177e4
LT
4659{
4660 static int ifindex;
4661 for (;;) {
4662 if (++ifindex <= 0)
4663 ifindex = 1;
881d966b 4664 if (!__dev_get_by_index(net, ifindex))
1da177e4
LT
4665 return ifindex;
4666 }
4667}
4668
1da177e4 4669/* Delayed registration/unregisteration */
3b5b34fd 4670static LIST_HEAD(net_todo_list);
1da177e4 4671
6f05f629 4672static void net_set_todo(struct net_device *dev)
1da177e4 4673{
1da177e4 4674 list_add_tail(&dev->todo_list, &net_todo_list);
1da177e4
LT
4675}
4676
9b5e383c 4677static void rollback_registered_many(struct list_head *head)
93ee31f1 4678{
9b5e383c
ED
4679 struct net_device *dev;
4680
93ee31f1
DL
4681 BUG_ON(dev_boot_phase);
4682 ASSERT_RTNL();
4683
9b5e383c
ED
4684 list_for_each_entry(dev, head, unreg_list) {
4685 /* Some devices call without registering
4686 * for initialization unwind.
4687 */
4688 if (dev->reg_state == NETREG_UNINITIALIZED) {
4689 pr_debug("unregister_netdevice: device %s/%p never "
4690 "was registered\n", dev->name, dev);
93ee31f1 4691
9b5e383c
ED
4692 WARN_ON(1);
4693 return;
4694 }
93ee31f1 4695
9b5e383c 4696 BUG_ON(dev->reg_state != NETREG_REGISTERED);
93ee31f1 4697
9b5e383c
ED
4698 /* If device is running, close it first. */
4699 dev_close(dev);
93ee31f1 4700
9b5e383c
ED
4701 /* And unlink it from device chain. */
4702 unlist_netdevice(dev);
93ee31f1 4703
9b5e383c
ED
4704 dev->reg_state = NETREG_UNREGISTERING;
4705 }
93ee31f1
DL
4706
4707 synchronize_net();
4708
9b5e383c
ED
4709 list_for_each_entry(dev, head, unreg_list) {
4710 /* Shutdown queueing discipline. */
4711 dev_shutdown(dev);
93ee31f1
DL
4712
4713
9b5e383c
ED
4714 /* Notify protocols, that we are about to destroy
4715 this device. They should clean all the things.
4716 */
4717 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
93ee31f1 4718
9b5e383c
ED
4719 /*
4720 * Flush the unicast and multicast chains
4721 */
4722 dev_unicast_flush(dev);
4723 dev_addr_discard(dev);
93ee31f1 4724
9b5e383c
ED
4725 if (dev->netdev_ops->ndo_uninit)
4726 dev->netdev_ops->ndo_uninit(dev);
93ee31f1 4727
9b5e383c
ED
4728 /* Notifier chain MUST detach us from master device. */
4729 WARN_ON(dev->master);
93ee31f1 4730
9b5e383c
ED
4731 /* Remove entries from kobject tree */
4732 netdev_unregister_kobject(dev);
4733 }
93ee31f1
DL
4734
4735 synchronize_net();
4736
9b5e383c
ED
4737 list_for_each_entry(dev, head, unreg_list)
4738 dev_put(dev);
4739}
4740
4741static void rollback_registered(struct net_device *dev)
4742{
4743 LIST_HEAD(single);
4744
4745 list_add(&dev->unreg_list, &single);
4746 rollback_registered_many(&single);
93ee31f1
DL
4747}
4748
e8a0464c
DM
4749static void __netdev_init_queue_locks_one(struct net_device *dev,
4750 struct netdev_queue *dev_queue,
4751 void *_unused)
c773e847
DM
4752{
4753 spin_lock_init(&dev_queue->_xmit_lock);
cf508b12 4754 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
c773e847
DM
4755 dev_queue->xmit_lock_owner = -1;
4756}
4757
4758static void netdev_init_queue_locks(struct net_device *dev)
4759{
e8a0464c
DM
4760 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4761 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
c773e847
DM
4762}
4763
b63365a2
HX
4764unsigned long netdev_fix_features(unsigned long features, const char *name)
4765{
4766 /* Fix illegal SG+CSUM combinations. */
4767 if ((features & NETIF_F_SG) &&
4768 !(features & NETIF_F_ALL_CSUM)) {
4769 if (name)
4770 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4771 "checksum feature.\n", name);
4772 features &= ~NETIF_F_SG;
4773 }
4774
4775 /* TSO requires that SG is present as well. */
4776 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4777 if (name)
4778 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4779 "SG feature.\n", name);
4780 features &= ~NETIF_F_TSO;
4781 }
4782
4783 if (features & NETIF_F_UFO) {
4784 if (!(features & NETIF_F_GEN_CSUM)) {
4785 if (name)
4786 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4787 "since no NETIF_F_HW_CSUM feature.\n",
4788 name);
4789 features &= ~NETIF_F_UFO;
4790 }
4791
4792 if (!(features & NETIF_F_SG)) {
4793 if (name)
4794 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4795 "since no NETIF_F_SG feature.\n", name);
4796 features &= ~NETIF_F_UFO;
4797 }
4798 }
4799
4800 return features;
4801}
4802EXPORT_SYMBOL(netdev_fix_features);
4803
1da177e4
LT
4804/**
4805 * register_netdevice - register a network device
4806 * @dev: device to register
4807 *
4808 * Take a completed network device structure and add it to the kernel
4809 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4810 * chain. 0 is returned on success. A negative errno code is returned
4811 * on a failure to set up the device, or if the name is a duplicate.
4812 *
4813 * Callers must hold the rtnl semaphore. You may want
4814 * register_netdev() instead of this.
4815 *
4816 * BUGS:
4817 * The locking appears insufficient to guarantee two parallel registers
4818 * will not get the same name.
4819 */
4820
4821int register_netdevice(struct net_device *dev)
4822{
4823 struct hlist_head *head;
4824 struct hlist_node *p;
4825 int ret;
d314774c 4826 struct net *net = dev_net(dev);
1da177e4
LT
4827
4828 BUG_ON(dev_boot_phase);
4829 ASSERT_RTNL();
4830
b17a7c17
SH
4831 might_sleep();
4832
1da177e4
LT
4833 /* When net_device's are persistent, this will be fatal. */
4834 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
d314774c 4835 BUG_ON(!net);
1da177e4 4836
f1f28aa3 4837 spin_lock_init(&dev->addr_list_lock);
cf508b12 4838 netdev_set_addr_lockdep_class(dev);
c773e847 4839 netdev_init_queue_locks(dev);
1da177e4 4840
1da177e4
LT
4841 dev->iflink = -1;
4842
4843 /* Init, if this function is available */
d314774c
SH
4844 if (dev->netdev_ops->ndo_init) {
4845 ret = dev->netdev_ops->ndo_init(dev);
1da177e4
LT
4846 if (ret) {
4847 if (ret > 0)
4848 ret = -EIO;
90833aa4 4849 goto out;
1da177e4
LT
4850 }
4851 }
4ec93edb 4852
1da177e4
LT
4853 if (!dev_valid_name(dev->name)) {
4854 ret = -EINVAL;
7ce1b0ed 4855 goto err_uninit;
1da177e4
LT
4856 }
4857
881d966b 4858 dev->ifindex = dev_new_index(net);
1da177e4
LT
4859 if (dev->iflink == -1)
4860 dev->iflink = dev->ifindex;
4861
4862 /* Check for existence of name */
881d966b 4863 head = dev_name_hash(net, dev->name);
1da177e4
LT
4864 hlist_for_each(p, head) {
4865 struct net_device *d
4866 = hlist_entry(p, struct net_device, name_hlist);
4867 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4868 ret = -EEXIST;
7ce1b0ed 4869 goto err_uninit;
1da177e4 4870 }
4ec93edb 4871 }
1da177e4 4872
d212f87b
SH
4873 /* Fix illegal checksum combinations */
4874 if ((dev->features & NETIF_F_HW_CSUM) &&
4875 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4876 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4877 dev->name);
4878 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4879 }
4880
4881 if ((dev->features & NETIF_F_NO_CSUM) &&
4882 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4883 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4884 dev->name);
4885 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4886 }
4887
b63365a2 4888 dev->features = netdev_fix_features(dev->features, dev->name);
1da177e4 4889
e5a4a72d
LB
4890 /* Enable software GSO if SG is supported. */
4891 if (dev->features & NETIF_F_SG)
4892 dev->features |= NETIF_F_GSO;
4893
aaf8cdc3 4894 netdev_initialize_kobject(dev);
7ffbe3fd
JB
4895
4896 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4897 ret = notifier_to_errno(ret);
4898 if (ret)
4899 goto err_uninit;
4900
8b41d188 4901 ret = netdev_register_kobject(dev);
b17a7c17 4902 if (ret)
7ce1b0ed 4903 goto err_uninit;
b17a7c17
SH
4904 dev->reg_state = NETREG_REGISTERED;
4905
1da177e4
LT
4906 /*
4907 * Default initial state at registry is that the
4908 * device is present.
4909 */
4910
4911 set_bit(__LINK_STATE_PRESENT, &dev->state);
4912
1da177e4 4913 dev_init_scheduler(dev);
1da177e4 4914 dev_hold(dev);
ce286d32 4915 list_netdevice(dev);
1da177e4
LT
4916
4917 /* Notify protocols, that a new device appeared. */
056925ab 4918 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
fcc5a03a 4919 ret = notifier_to_errno(ret);
93ee31f1
DL
4920 if (ret) {
4921 rollback_registered(dev);
4922 dev->reg_state = NETREG_UNREGISTERED;
4923 }
1da177e4
LT
4924
4925out:
4926 return ret;
7ce1b0ed
HX
4927
4928err_uninit:
d314774c
SH
4929 if (dev->netdev_ops->ndo_uninit)
4930 dev->netdev_ops->ndo_uninit(dev);
7ce1b0ed 4931 goto out;
1da177e4 4932}
d1b19dff 4933EXPORT_SYMBOL(register_netdevice);
1da177e4 4934
937f1ba5
BH
4935/**
4936 * init_dummy_netdev - init a dummy network device for NAPI
4937 * @dev: device to init
4938 *
4939 * This takes a network device structure and initialize the minimum
4940 * amount of fields so it can be used to schedule NAPI polls without
4941 * registering a full blown interface. This is to be used by drivers
4942 * that need to tie several hardware interfaces to a single NAPI
4943 * poll scheduler due to HW limitations.
4944 */
4945int init_dummy_netdev(struct net_device *dev)
4946{
4947 /* Clear everything. Note we don't initialize spinlocks
4948 * are they aren't supposed to be taken by any of the
4949 * NAPI code and this dummy netdev is supposed to be
4950 * only ever used for NAPI polls
4951 */
4952 memset(dev, 0, sizeof(struct net_device));
4953
4954 /* make sure we BUG if trying to hit standard
4955 * register/unregister code path
4956 */
4957 dev->reg_state = NETREG_DUMMY;
4958
4959 /* initialize the ref count */
4960 atomic_set(&dev->refcnt, 1);
4961
4962 /* NAPI wants this */
4963 INIT_LIST_HEAD(&dev->napi_list);
4964
4965 /* a dummy interface is started by default */
4966 set_bit(__LINK_STATE_PRESENT, &dev->state);
4967 set_bit(__LINK_STATE_START, &dev->state);
4968
4969 return 0;
4970}
4971EXPORT_SYMBOL_GPL(init_dummy_netdev);
4972
4973
1da177e4
LT
4974/**
4975 * register_netdev - register a network device
4976 * @dev: device to register
4977 *
4978 * Take a completed network device structure and add it to the kernel
4979 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4980 * chain. 0 is returned on success. A negative errno code is returned
4981 * on a failure to set up the device, or if the name is a duplicate.
4982 *
38b4da38 4983 * This is a wrapper around register_netdevice that takes the rtnl semaphore
1da177e4
LT
4984 * and expands the device name if you passed a format string to
4985 * alloc_netdev.
4986 */
4987int register_netdev(struct net_device *dev)
4988{
4989 int err;
4990
4991 rtnl_lock();
4992
4993 /*
4994 * If the name is a format string the caller wants us to do a
4995 * name allocation.
4996 */
4997 if (strchr(dev->name, '%')) {
4998 err = dev_alloc_name(dev, dev->name);
4999 if (err < 0)
5000 goto out;
5001 }
4ec93edb 5002
1da177e4
LT
5003 err = register_netdevice(dev);
5004out:
5005 rtnl_unlock();
5006 return err;
5007}
5008EXPORT_SYMBOL(register_netdev);
5009
5010/*
5011 * netdev_wait_allrefs - wait until all references are gone.
5012 *
5013 * This is called when unregistering network devices.
5014 *
5015 * Any protocol or device that holds a reference should register
5016 * for netdevice notification, and cleanup and put back the
5017 * reference if they receive an UNREGISTER event.
5018 * We can get stuck here if buggy protocols don't correctly
4ec93edb 5019 * call dev_put.
1da177e4
LT
5020 */
5021static void netdev_wait_allrefs(struct net_device *dev)
5022{
5023 unsigned long rebroadcast_time, warning_time;
5024
5025 rebroadcast_time = warning_time = jiffies;
5026 while (atomic_read(&dev->refcnt) != 0) {
5027 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6756ae4b 5028 rtnl_lock();
1da177e4
LT
5029
5030 /* Rebroadcast unregister notification */
056925ab 5031 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
1da177e4
LT
5032
5033 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5034 &dev->state)) {
5035 /* We must not have linkwatch events
5036 * pending on unregister. If this
5037 * happens, we simply run the queue
5038 * unscheduled, resulting in a noop
5039 * for this device.
5040 */
5041 linkwatch_run_queue();
5042 }
5043
6756ae4b 5044 __rtnl_unlock();
1da177e4
LT
5045
5046 rebroadcast_time = jiffies;
5047 }
5048
5049 msleep(250);
5050
5051 if (time_after(jiffies, warning_time + 10 * HZ)) {
5052 printk(KERN_EMERG "unregister_netdevice: "
5053 "waiting for %s to become free. Usage "
5054 "count = %d\n",
5055 dev->name, atomic_read(&dev->refcnt));
5056 warning_time = jiffies;
5057 }
5058 }
5059}
5060
5061/* The sequence is:
5062 *
5063 * rtnl_lock();
5064 * ...
5065 * register_netdevice(x1);
5066 * register_netdevice(x2);
5067 * ...
5068 * unregister_netdevice(y1);
5069 * unregister_netdevice(y2);
5070 * ...
5071 * rtnl_unlock();
5072 * free_netdev(y1);
5073 * free_netdev(y2);
5074 *
58ec3b4d 5075 * We are invoked by rtnl_unlock().
1da177e4 5076 * This allows us to deal with problems:
b17a7c17 5077 * 1) We can delete sysfs objects which invoke hotplug
1da177e4
LT
5078 * without deadlocking with linkwatch via keventd.
5079 * 2) Since we run with the RTNL semaphore not held, we can sleep
5080 * safely in order to wait for the netdev refcnt to drop to zero.
58ec3b4d
HX
5081 *
5082 * We must not return until all unregister events added during
5083 * the interval the lock was held have been completed.
1da177e4 5084 */
1da177e4
LT
5085void netdev_run_todo(void)
5086{
626ab0e6 5087 struct list_head list;
1da177e4 5088
1da177e4 5089 /* Snapshot list, allow later requests */
626ab0e6 5090 list_replace_init(&net_todo_list, &list);
58ec3b4d
HX
5091
5092 __rtnl_unlock();
626ab0e6 5093
1da177e4
LT
5094 while (!list_empty(&list)) {
5095 struct net_device *dev
5096 = list_entry(list.next, struct net_device, todo_list);
5097 list_del(&dev->todo_list);
5098
b17a7c17
SH
5099 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5100 printk(KERN_ERR "network todo '%s' but state %d\n",
5101 dev->name, dev->reg_state);
5102 dump_stack();
5103 continue;
5104 }
1da177e4 5105
b17a7c17 5106 dev->reg_state = NETREG_UNREGISTERED;
1da177e4 5107
6e583ce5
SH
5108 on_each_cpu(flush_backlog, dev, 1);
5109
b17a7c17 5110 netdev_wait_allrefs(dev);
1da177e4 5111
b17a7c17
SH
5112 /* paranoia */
5113 BUG_ON(atomic_read(&dev->refcnt));
547b792c
IJ
5114 WARN_ON(dev->ip_ptr);
5115 WARN_ON(dev->ip6_ptr);
5116 WARN_ON(dev->dn_ptr);
1da177e4 5117
b17a7c17
SH
5118 if (dev->destructor)
5119 dev->destructor(dev);
9093bbb2
SH
5120
5121 /* Free network device */
5122 kobject_put(&dev->dev.kobj);
1da177e4 5123 }
1da177e4
LT
5124}
5125
eeda3fd6
SH
5126/**
5127 * dev_get_stats - get network device statistics
5128 * @dev: device to get statistics from
5129 *
5130 * Get network statistics from device. The device driver may provide
5131 * its own method by setting dev->netdev_ops->get_stats; otherwise
5132 * the internal statistics structure is used.
5133 */
5134const struct net_device_stats *dev_get_stats(struct net_device *dev)
7004bf25 5135{
eeda3fd6
SH
5136 const struct net_device_ops *ops = dev->netdev_ops;
5137
5138 if (ops->ndo_get_stats)
5139 return ops->ndo_get_stats(dev);
7004bf25
ED
5140 else {
5141 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5142 struct net_device_stats *stats = &dev->stats;
5143 unsigned int i;
5144 struct netdev_queue *txq;
5145
5146 for (i = 0; i < dev->num_tx_queues; i++) {
5147 txq = netdev_get_tx_queue(dev, i);
5148 tx_bytes += txq->tx_bytes;
5149 tx_packets += txq->tx_packets;
5150 tx_dropped += txq->tx_dropped;
5151 }
5152 if (tx_bytes || tx_packets || tx_dropped) {
5153 stats->tx_bytes = tx_bytes;
5154 stats->tx_packets = tx_packets;
5155 stats->tx_dropped = tx_dropped;
5156 }
5157 return stats;
5158 }
c45d286e 5159}
eeda3fd6 5160EXPORT_SYMBOL(dev_get_stats);
c45d286e 5161
dc2b4847 5162static void netdev_init_one_queue(struct net_device *dev,
e8a0464c
DM
5163 struct netdev_queue *queue,
5164 void *_unused)
dc2b4847 5165{
dc2b4847
DM
5166 queue->dev = dev;
5167}
5168
bb949fbd
DM
5169static void netdev_init_queues(struct net_device *dev)
5170{
e8a0464c
DM
5171 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5172 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
c3f26a26 5173 spin_lock_init(&dev->tx_global_lock);
bb949fbd
DM
5174}
5175
1da177e4 5176/**
f25f4e44 5177 * alloc_netdev_mq - allocate network device
1da177e4
LT
5178 * @sizeof_priv: size of private data to allocate space for
5179 * @name: device name format string
5180 * @setup: callback to initialize device
f25f4e44 5181 * @queue_count: the number of subqueues to allocate
1da177e4
LT
5182 *
5183 * Allocates a struct net_device with private data area for driver use
f25f4e44
PWJ
5184 * and performs basic initialization. Also allocates subquue structs
5185 * for each queue on the device at the end of the netdevice.
1da177e4 5186 */
f25f4e44
PWJ
5187struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5188 void (*setup)(struct net_device *), unsigned int queue_count)
1da177e4 5189{
e8a0464c 5190 struct netdev_queue *tx;
1da177e4 5191 struct net_device *dev;
7943986c 5192 size_t alloc_size;
1ce8e7b5 5193 struct net_device *p;
1da177e4 5194
b6fe17d6
SH
5195 BUG_ON(strlen(name) >= sizeof(dev->name));
5196
fd2ea0a7 5197 alloc_size = sizeof(struct net_device);
d1643d24
AD
5198 if (sizeof_priv) {
5199 /* ensure 32-byte alignment of private area */
1ce8e7b5 5200 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
d1643d24
AD
5201 alloc_size += sizeof_priv;
5202 }
5203 /* ensure 32-byte alignment of whole construct */
1ce8e7b5 5204 alloc_size += NETDEV_ALIGN - 1;
1da177e4 5205
31380de9 5206 p = kzalloc(alloc_size, GFP_KERNEL);
1da177e4 5207 if (!p) {
b6fe17d6 5208 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
1da177e4
LT
5209 return NULL;
5210 }
1da177e4 5211
7943986c 5212 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
e8a0464c
DM
5213 if (!tx) {
5214 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5215 "tx qdiscs.\n");
ab9c73cc 5216 goto free_p;
e8a0464c
DM
5217 }
5218
1ce8e7b5 5219 dev = PTR_ALIGN(p, NETDEV_ALIGN);
1da177e4 5220 dev->padded = (char *)dev - (char *)p;
ab9c73cc
JP
5221
5222 if (dev_addr_init(dev))
5223 goto free_tx;
5224
ccffad25
JP
5225 dev_unicast_init(dev);
5226
c346dca1 5227 dev_net_set(dev, &init_net);
1da177e4 5228
e8a0464c
DM
5229 dev->_tx = tx;
5230 dev->num_tx_queues = queue_count;
fd2ea0a7 5231 dev->real_num_tx_queues = queue_count;
e8a0464c 5232
82cc1a7a 5233 dev->gso_max_size = GSO_MAX_SIZE;
1da177e4 5234
bb949fbd
DM
5235 netdev_init_queues(dev);
5236
d565b0a1 5237 INIT_LIST_HEAD(&dev->napi_list);
93f154b5 5238 dev->priv_flags = IFF_XMIT_DST_RELEASE;
1da177e4
LT
5239 setup(dev);
5240 strcpy(dev->name, name);
5241 return dev;
ab9c73cc
JP
5242
5243free_tx:
5244 kfree(tx);
5245
5246free_p:
5247 kfree(p);
5248 return NULL;
1da177e4 5249}
f25f4e44 5250EXPORT_SYMBOL(alloc_netdev_mq);
1da177e4
LT
5251
5252/**
5253 * free_netdev - free network device
5254 * @dev: device
5255 *
4ec93edb
YH
5256 * This function does the last stage of destroying an allocated device
5257 * interface. The reference to the device object is released.
1da177e4
LT
5258 * If this is the last reference then it will be freed.
5259 */
5260void free_netdev(struct net_device *dev)
5261{
d565b0a1
HX
5262 struct napi_struct *p, *n;
5263
f3005d7f
DL
5264 release_net(dev_net(dev));
5265
e8a0464c
DM
5266 kfree(dev->_tx);
5267
f001fde5
JP
5268 /* Flush device addresses */
5269 dev_addr_flush(dev);
5270
d565b0a1
HX
5271 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5272 netif_napi_del(p);
5273
3041a069 5274 /* Compatibility with error handling in drivers */
1da177e4
LT
5275 if (dev->reg_state == NETREG_UNINITIALIZED) {
5276 kfree((char *)dev - dev->padded);
5277 return;
5278 }
5279
5280 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5281 dev->reg_state = NETREG_RELEASED;
5282
43cb76d9
GKH
5283 /* will free via device release */
5284 put_device(&dev->dev);
1da177e4 5285}
d1b19dff 5286EXPORT_SYMBOL(free_netdev);
4ec93edb 5287
f0db275a
SH
5288/**
5289 * synchronize_net - Synchronize with packet receive processing
5290 *
5291 * Wait for packets currently being received to be done.
5292 * Does not block later packets from starting.
5293 */
4ec93edb 5294void synchronize_net(void)
1da177e4
LT
5295{
5296 might_sleep();
fbd568a3 5297 synchronize_rcu();
1da177e4 5298}
d1b19dff 5299EXPORT_SYMBOL(synchronize_net);
1da177e4
LT
5300
5301/**
44a0873d 5302 * unregister_netdevice_queue - remove device from the kernel
1da177e4 5303 * @dev: device
44a0873d
ED
5304 * @head: list
5305
1da177e4 5306 * This function shuts down a device interface and removes it
d59b54b1 5307 * from the kernel tables.
44a0873d 5308 * If head not NULL, device is queued to be unregistered later.
1da177e4
LT
5309 *
5310 * Callers must hold the rtnl semaphore. You may want
5311 * unregister_netdev() instead of this.
5312 */
5313
44a0873d 5314void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
1da177e4 5315{
a6620712
HX
5316 ASSERT_RTNL();
5317
44a0873d
ED
5318 if (head) {
5319 list_add_tail(&dev->unreg_list, head);
5320 } else {
5321 rollback_registered(dev);
5322 /* Finish processing unregister after unlock */
5323 net_set_todo(dev);
5324 }
1da177e4 5325}
44a0873d 5326EXPORT_SYMBOL(unregister_netdevice_queue);
1da177e4 5327
9b5e383c
ED
5328/**
5329 * unregister_netdevice_many - unregister many devices
5330 * @head: list of devices
5331 *
5332 */
5333void unregister_netdevice_many(struct list_head *head)
5334{
5335 struct net_device *dev;
5336
5337 if (!list_empty(head)) {
5338 rollback_registered_many(head);
5339 list_for_each_entry(dev, head, unreg_list)
5340 net_set_todo(dev);
5341 }
5342}
63c8099d 5343EXPORT_SYMBOL(unregister_netdevice_many);
9b5e383c 5344
1da177e4
LT
5345/**
5346 * unregister_netdev - remove device from the kernel
5347 * @dev: device
5348 *
5349 * This function shuts down a device interface and removes it
d59b54b1 5350 * from the kernel tables.
1da177e4
LT
5351 *
5352 * This is just a wrapper for unregister_netdevice that takes
5353 * the rtnl semaphore. In general you want to use this and not
5354 * unregister_netdevice.
5355 */
5356void unregister_netdev(struct net_device *dev)
5357{
5358 rtnl_lock();
5359 unregister_netdevice(dev);
5360 rtnl_unlock();
5361}
1da177e4
LT
5362EXPORT_SYMBOL(unregister_netdev);
5363
ce286d32
EB
5364/**
5365 * dev_change_net_namespace - move device to different nethost namespace
5366 * @dev: device
5367 * @net: network namespace
5368 * @pat: If not NULL name pattern to try if the current device name
5369 * is already taken in the destination network namespace.
5370 *
5371 * This function shuts down a device interface and moves it
5372 * to a new network namespace. On success 0 is returned, on
5373 * a failure a netagive errno code is returned.
5374 *
5375 * Callers must hold the rtnl semaphore.
5376 */
5377
5378int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5379{
5380 char buf[IFNAMSIZ];
5381 const char *destname;
5382 int err;
5383
5384 ASSERT_RTNL();
5385
5386 /* Don't allow namespace local devices to be moved. */
5387 err = -EINVAL;
5388 if (dev->features & NETIF_F_NETNS_LOCAL)
5389 goto out;
5390
3891845e
EB
5391#ifdef CONFIG_SYSFS
5392 /* Don't allow real devices to be moved when sysfs
5393 * is enabled.
5394 */
5395 err = -EINVAL;
5396 if (dev->dev.parent)
5397 goto out;
5398#endif
5399
ce286d32
EB
5400 /* Ensure the device has been registrered */
5401 err = -EINVAL;
5402 if (dev->reg_state != NETREG_REGISTERED)
5403 goto out;
5404
5405 /* Get out if there is nothing todo */
5406 err = 0;
878628fb 5407 if (net_eq(dev_net(dev), net))
ce286d32
EB
5408 goto out;
5409
5410 /* Pick the destination device name, and ensure
5411 * we can use it in the destination network namespace.
5412 */
5413 err = -EEXIST;
5414 destname = dev->name;
5415 if (__dev_get_by_name(net, destname)) {
5416 /* We get here if we can't use the current device name */
5417 if (!pat)
5418 goto out;
5419 if (!dev_valid_name(pat))
5420 goto out;
5421 if (strchr(pat, '%')) {
5422 if (__dev_alloc_name(net, pat, buf) < 0)
5423 goto out;
5424 destname = buf;
5425 } else
5426 destname = pat;
5427 if (__dev_get_by_name(net, destname))
5428 goto out;
5429 }
5430
5431 /*
5432 * And now a mini version of register_netdevice unregister_netdevice.
5433 */
5434
5435 /* If device is running close it first. */
9b772652 5436 dev_close(dev);
ce286d32
EB
5437
5438 /* And unlink it from device chain */
5439 err = -ENODEV;
5440 unlist_netdevice(dev);
5441
5442 synchronize_net();
5443
5444 /* Shutdown queueing discipline. */
5445 dev_shutdown(dev);
5446
5447 /* Notify protocols, that we are about to destroy
5448 this device. They should clean all the things.
5449 */
5450 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5451
5452 /*
5453 * Flush the unicast and multicast chains
5454 */
ccffad25 5455 dev_unicast_flush(dev);
ce286d32
EB
5456 dev_addr_discard(dev);
5457
3891845e
EB
5458 netdev_unregister_kobject(dev);
5459
ce286d32 5460 /* Actually switch the network namespace */
c346dca1 5461 dev_net_set(dev, net);
ce286d32
EB
5462
5463 /* Assign the new device name */
5464 if (destname != dev->name)
5465 strcpy(dev->name, destname);
5466
5467 /* If there is an ifindex conflict assign a new one */
5468 if (__dev_get_by_index(net, dev->ifindex)) {
5469 int iflink = (dev->iflink == dev->ifindex);
5470 dev->ifindex = dev_new_index(net);
5471 if (iflink)
5472 dev->iflink = dev->ifindex;
5473 }
5474
8b41d188 5475 /* Fixup kobjects */
aaf8cdc3 5476 err = netdev_register_kobject(dev);
8b41d188 5477 WARN_ON(err);
ce286d32
EB
5478
5479 /* Add the device back in the hashes */
5480 list_netdevice(dev);
5481
5482 /* Notify protocols, that a new device appeared. */
5483 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5484
5485 synchronize_net();
5486 err = 0;
5487out:
5488 return err;
5489}
463d0183 5490EXPORT_SYMBOL_GPL(dev_change_net_namespace);
ce286d32 5491
1da177e4
LT
5492static int dev_cpu_callback(struct notifier_block *nfb,
5493 unsigned long action,
5494 void *ocpu)
5495{
5496 struct sk_buff **list_skb;
37437bb2 5497 struct Qdisc **list_net;
1da177e4
LT
5498 struct sk_buff *skb;
5499 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5500 struct softnet_data *sd, *oldsd;
5501
8bb78442 5502 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
1da177e4
LT
5503 return NOTIFY_OK;
5504
5505 local_irq_disable();
5506 cpu = smp_processor_id();
5507 sd = &per_cpu(softnet_data, cpu);
5508 oldsd = &per_cpu(softnet_data, oldcpu);
5509
5510 /* Find end of our completion_queue. */
5511 list_skb = &sd->completion_queue;
5512 while (*list_skb)
5513 list_skb = &(*list_skb)->next;
5514 /* Append completion queue from offline CPU. */
5515 *list_skb = oldsd->completion_queue;
5516 oldsd->completion_queue = NULL;
5517
5518 /* Find end of our output_queue. */
5519 list_net = &sd->output_queue;
5520 while (*list_net)
5521 list_net = &(*list_net)->next_sched;
5522 /* Append output queue from offline CPU. */
5523 *list_net = oldsd->output_queue;
5524 oldsd->output_queue = NULL;
5525
5526 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5527 local_irq_enable();
5528
5529 /* Process offline CPU's input_pkt_queue */
5530 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5531 netif_rx(skb);
5532
5533 return NOTIFY_OK;
5534}
1da177e4
LT
5535
5536
7f353bf2 5537/**
b63365a2
HX
5538 * netdev_increment_features - increment feature set by one
5539 * @all: current feature set
5540 * @one: new feature set
5541 * @mask: mask feature set
7f353bf2
HX
5542 *
5543 * Computes a new feature set after adding a device with feature set
b63365a2
HX
5544 * @one to the master device with current feature set @all. Will not
5545 * enable anything that is off in @mask. Returns the new feature set.
7f353bf2 5546 */
b63365a2
HX
5547unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5548 unsigned long mask)
5549{
5550 /* If device needs checksumming, downgrade to it. */
d1b19dff 5551 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
b63365a2
HX
5552 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5553 else if (mask & NETIF_F_ALL_CSUM) {
5554 /* If one device supports v4/v6 checksumming, set for all. */
5555 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5556 !(all & NETIF_F_GEN_CSUM)) {
5557 all &= ~NETIF_F_ALL_CSUM;
5558 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5559 }
e2a6b852 5560
b63365a2
HX
5561 /* If one device supports hw checksumming, set for all. */
5562 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5563 all &= ~NETIF_F_ALL_CSUM;
5564 all |= NETIF_F_HW_CSUM;
5565 }
5566 }
7f353bf2 5567
b63365a2 5568 one |= NETIF_F_ALL_CSUM;
7f353bf2 5569
b63365a2 5570 one |= all & NETIF_F_ONE_FOR_ALL;
d9f5950f 5571 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
b63365a2 5572 all |= one & mask & NETIF_F_ONE_FOR_ALL;
7f353bf2
HX
5573
5574 return all;
5575}
b63365a2 5576EXPORT_SYMBOL(netdev_increment_features);
7f353bf2 5577
30d97d35
PE
5578static struct hlist_head *netdev_create_hash(void)
5579{
5580 int i;
5581 struct hlist_head *hash;
5582
5583 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5584 if (hash != NULL)
5585 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5586 INIT_HLIST_HEAD(&hash[i]);
5587
5588 return hash;
5589}
5590
881d966b 5591/* Initialize per network namespace state */
4665079c 5592static int __net_init netdev_init(struct net *net)
881d966b 5593{
881d966b 5594 INIT_LIST_HEAD(&net->dev_base_head);
881d966b 5595
30d97d35
PE
5596 net->dev_name_head = netdev_create_hash();
5597 if (net->dev_name_head == NULL)
5598 goto err_name;
881d966b 5599
30d97d35
PE
5600 net->dev_index_head = netdev_create_hash();
5601 if (net->dev_index_head == NULL)
5602 goto err_idx;
881d966b
EB
5603
5604 return 0;
30d97d35
PE
5605
5606err_idx:
5607 kfree(net->dev_name_head);
5608err_name:
5609 return -ENOMEM;
881d966b
EB
5610}
5611
f0db275a
SH
5612/**
5613 * netdev_drivername - network driver for the device
5614 * @dev: network device
5615 * @buffer: buffer for resulting name
5616 * @len: size of buffer
5617 *
5618 * Determine network driver for device.
5619 */
cf04a4c7 5620char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
6579e57b 5621{
cf04a4c7
SH
5622 const struct device_driver *driver;
5623 const struct device *parent;
6579e57b
AV
5624
5625 if (len <= 0 || !buffer)
5626 return buffer;
5627 buffer[0] = 0;
5628
5629 parent = dev->dev.parent;
5630
5631 if (!parent)
5632 return buffer;
5633
5634 driver = parent->driver;
5635 if (driver && driver->name)
5636 strlcpy(buffer, driver->name, len);
5637 return buffer;
5638}
5639
4665079c 5640static void __net_exit netdev_exit(struct net *net)
881d966b
EB
5641{
5642 kfree(net->dev_name_head);
5643 kfree(net->dev_index_head);
5644}
5645
022cbae6 5646static struct pernet_operations __net_initdata netdev_net_ops = {
881d966b
EB
5647 .init = netdev_init,
5648 .exit = netdev_exit,
5649};
5650
4665079c 5651static void __net_exit default_device_exit(struct net *net)
ce286d32 5652{
8eb79863 5653 struct net_device *dev;
ce286d32
EB
5654 /*
5655 * Push all migratable of the network devices back to the
5656 * initial network namespace
5657 */
5658 rtnl_lock();
8eb79863
EB
5659restart:
5660 for_each_netdev(net, dev) {
ce286d32 5661 int err;
aca51397 5662 char fb_name[IFNAMSIZ];
ce286d32
EB
5663
5664 /* Ignore unmoveable devices (i.e. loopback) */
5665 if (dev->features & NETIF_F_NETNS_LOCAL)
5666 continue;
5667
d0c082ce
EB
5668 /* Delete virtual devices */
5669 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
23289a37 5670 dev->rtnl_link_ops->dellink(dev, NULL);
8eb79863 5671 goto restart;
d0c082ce
EB
5672 }
5673
ce286d32 5674 /* Push remaing network devices to init_net */
aca51397
PE
5675 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5676 err = dev_change_net_namespace(dev, &init_net, fb_name);
ce286d32 5677 if (err) {
aca51397 5678 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
ce286d32 5679 __func__, dev->name, err);
aca51397 5680 BUG();
ce286d32 5681 }
8eb79863 5682 goto restart;
ce286d32
EB
5683 }
5684 rtnl_unlock();
5685}
5686
022cbae6 5687static struct pernet_operations __net_initdata default_device_ops = {
ce286d32
EB
5688 .exit = default_device_exit,
5689};
5690
1da177e4
LT
5691/*
5692 * Initialize the DEV module. At boot time this walks the device list and
5693 * unhooks any devices that fail to initialise (normally hardware not
5694 * present) and leaves us with a valid list of present and active devices.
5695 *
5696 */
5697
5698/*
5699 * This is called single threaded during boot, so no need
5700 * to take the rtnl semaphore.
5701 */
5702static int __init net_dev_init(void)
5703{
5704 int i, rc = -ENOMEM;
5705
5706 BUG_ON(!dev_boot_phase);
5707
1da177e4
LT
5708 if (dev_proc_init())
5709 goto out;
5710
8b41d188 5711 if (netdev_kobject_init())
1da177e4
LT
5712 goto out;
5713
5714 INIT_LIST_HEAD(&ptype_all);
82d8a867 5715 for (i = 0; i < PTYPE_HASH_SIZE; i++)
1da177e4
LT
5716 INIT_LIST_HEAD(&ptype_base[i]);
5717
881d966b
EB
5718 if (register_pernet_subsys(&netdev_net_ops))
5719 goto out;
1da177e4
LT
5720
5721 /*
5722 * Initialise the packet receive queues.
5723 */
5724
6f912042 5725 for_each_possible_cpu(i) {
1da177e4
LT
5726 struct softnet_data *queue;
5727
5728 queue = &per_cpu(softnet_data, i);
5729 skb_queue_head_init(&queue->input_pkt_queue);
1da177e4
LT
5730 queue->completion_queue = NULL;
5731 INIT_LIST_HEAD(&queue->poll_list);
bea3348e
SH
5732
5733 queue->backlog.poll = process_backlog;
5734 queue->backlog.weight = weight_p;
d565b0a1 5735 queue->backlog.gro_list = NULL;
4ae5544f 5736 queue->backlog.gro_count = 0;
1da177e4
LT
5737 }
5738
1da177e4
LT
5739 dev_boot_phase = 0;
5740
505d4f73
EB
5741 /* The loopback device is special if any other network devices
5742 * is present in a network namespace the loopback device must
5743 * be present. Since we now dynamically allocate and free the
5744 * loopback device ensure this invariant is maintained by
5745 * keeping the loopback device as the first device on the
5746 * list of network devices. Ensuring the loopback devices
5747 * is the first device that appears and the last network device
5748 * that disappears.
5749 */
5750 if (register_pernet_device(&loopback_net_ops))
5751 goto out;
5752
5753 if (register_pernet_device(&default_device_ops))
5754 goto out;
5755
962cf36c
CM
5756 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5757 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
1da177e4
LT
5758
5759 hotcpu_notifier(dev_cpu_callback, 0);
5760 dst_init();
5761 dev_mcast_init();
5762 rc = 0;
5763out:
5764 return rc;
5765}
5766
5767subsys_initcall(net_dev_init);
5768
e88721f8
KK
5769static int __init initialize_hashrnd(void)
5770{
5771 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5772 return 0;
5773}
5774
5775late_initcall_sync(initialize_hashrnd);
5776