blob: 7153e94f50ad88c701a39f385ffd71b2ac6b5517 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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>
Randy Dunlap4fc268d2006-01-11 12:17:47 -080078#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#include <linux/cpu.h>
80#include <linux/types.h>
81#include <linux/kernel.h>
82#include <linux/sched.h>
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -080083#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070084#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>
93#include <linux/notifier.h>
94#include <linux/skbuff.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020095#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070096#include <net/sock.h>
97#include <linux/rtnetlink.h>
98#include <linux/proc_fs.h>
99#include <linux/seq_file.h>
100#include <linux/stat.h>
101#include <linux/if_bridge.h>
Patrick McHardyb863ceb2007-07-14 18:55:06 -0700102#include <linux/if_macvlan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103#include <net/dst.h>
104#include <net/pkt_sched.h>
105#include <net/checksum.h>
106#include <linux/highmem.h>
107#include <linux/init.h>
108#include <linux/kmod.h>
109#include <linux/module.h>
110#include <linux/kallsyms.h>
111#include <linux/netpoll.h>
112#include <linux/rcupdate.h>
113#include <linux/delay.h>
Johannes Berg295f4a12007-04-26 20:43:56 -0700114#include <net/wext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115#include <net/iw_handler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116#include <asm/current.h>
Steve Grubb5bdb9882005-12-03 08:39:35 -0500117#include <linux/audit.h>
Chris Leechdb217332006-06-17 21:24:58 -0700118#include <linux/dmaengine.h>
Herbert Xuf6a78bf2006-06-22 02:57:17 -0700119#include <linux/err.h>
David S. Millerc7fa9d12006-08-15 16:34:13 -0700120#include <linux/ctype.h>
Jarek Poplawski723e98b2007-05-15 22:46:18 -0700121#include <linux/if_arp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Pavel Emelyanov342709e2007-10-23 21:14:45 -0700123#include "net-sysfs.h"
124
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125/*
126 * The list of packet types we will receive (as opposed to discard)
127 * and the routines to invoke.
128 *
129 * Why 16. Because with 16 the only overlap we get on a hash of the
130 * low nibble of the protocol value is RARP/SNAP/X.25.
131 *
132 * NOTE: That is no longer true with the addition of VLAN tags. Not
133 * sure which should go first, but I bet it won't make much
134 * difference if we are running VLANs. The good news is that
135 * this protocol won't be in the list unless compiled in, so
Stephen Hemminger3041a062006-05-26 13:25:24 -0700136 * the average user (w/out VLANs) will not be adversely affected.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 * --BLG
138 *
139 * 0800 IP
140 * 8100 802.1Q VLAN
141 * 0001 802.3
142 * 0002 AX.25
143 * 0004 802.2
144 * 8035 RARP
145 * 0005 SNAP
146 * 0805 X.25
147 * 0806 ARP
148 * 8137 IPX
149 * 0009 Localtalk
150 * 86DD IPv6
151 */
152
Pavel Emelyanov82d8a862007-11-26 20:12:58 +0800153#define PTYPE_HASH_SIZE (16)
154#define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
155
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156static DEFINE_SPINLOCK(ptype_lock);
Pavel Emelyanov82d8a862007-11-26 20:12:58 +0800157static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
Stephen Hemminger6b2bedc2007-03-12 14:33:50 -0700158static struct list_head ptype_all __read_mostly; /* Taps */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Chris Leechdb217332006-06-17 21:24:58 -0700160#ifdef CONFIG_NET_DMA
Dan Williamsd379b012007-07-09 11:56:42 -0700161struct net_dma {
162 struct dma_client client;
163 spinlock_t lock;
164 cpumask_t channel_mask;
165 struct dma_chan *channels[NR_CPUS];
166};
167
168static enum dma_state_client
169netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
170 enum dma_state state);
171
172static struct net_dma net_dma = {
173 .client = {
174 .event_callback = netdev_dma_event,
175 },
176};
Chris Leechdb217332006-06-17 21:24:58 -0700177#endif
178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179/*
Pavel Emelianov7562f872007-05-03 15:13:45 -0700180 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 * semaphore.
182 *
183 * Pure readers hold dev_base_lock for reading.
184 *
185 * Writers must hold the rtnl semaphore while they loop through the
Pavel Emelianov7562f872007-05-03 15:13:45 -0700186 * dev_base_head list, and hold dev_base_lock for writing when they do the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 * actual updates. This allows pure readers to access the list even
188 * while a writer is preparing to update it.
189 *
190 * To put it another way, dev_base_lock is held for writing only to
191 * protect against pure readers; the rtnl semaphore provides the
192 * protection against other writers.
193 *
194 * See, for example usages, register_netdevice() and
195 * unregister_netdevice(), which must be called with the rtnl
196 * semaphore held.
197 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198DEFINE_RWLOCK(dev_base_lock);
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200EXPORT_SYMBOL(dev_base_lock);
201
202#define NETDEV_HASHBITS 8
Eric W. Biederman881d9662007-09-17 11:56:21 -0700203#define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Eric W. Biederman881d9662007-09-17 11:56:21 -0700205static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
207 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
Eric W. Biederman881d9662007-09-17 11:56:21 -0700208 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
210
Eric W. Biederman881d9662007-09-17 11:56:21 -0700211static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
Eric W. Biederman881d9662007-09-17 11:56:21 -0700213 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
Eric W. Biedermance286d32007-09-12 13:53:49 +0200216/* Device list insertion */
217static int list_netdevice(struct net_device *dev)
218{
219 struct net *net = dev->nd_net;
220
221 ASSERT_RTNL();
222
223 write_lock_bh(&dev_base_lock);
224 list_add_tail(&dev->dev_list, &net->dev_base_head);
225 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
226 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
227 write_unlock_bh(&dev_base_lock);
228 return 0;
229}
230
231/* Device list removal */
232static void unlist_netdevice(struct net_device *dev)
233{
234 ASSERT_RTNL();
235
236 /* Unlink dev from the device chain */
237 write_lock_bh(&dev_base_lock);
238 list_del(&dev->dev_list);
239 hlist_del(&dev->name_hlist);
240 hlist_del(&dev->index_hlist);
241 write_unlock_bh(&dev_base_lock);
242}
243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244/*
245 * Our notifier list
246 */
247
Alan Sternf07d5b92006-05-09 15:23:03 -0700248static RAW_NOTIFIER_HEAD(netdev_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250/*
251 * Device drivers call our routines to queue packets here. We empty the
252 * queue in the local softnet handler.
253 */
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700254
255DEFINE_PER_CPU(struct softnet_data, softnet_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Jarek Poplawski723e98b2007-05-15 22:46:18 -0700257#ifdef CONFIG_DEBUG_LOCK_ALLOC
258/*
259 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
260 * according to dev->type
261 */
262static const unsigned short netdev_lock_type[] =
263 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
264 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
265 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
266 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
267 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
268 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
269 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
270 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
271 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
272 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
273 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
274 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
275 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
276 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
277 ARPHRD_NONE};
278
279static const char *netdev_lock_name[] =
280 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
281 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
282 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
283 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
284 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
285 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
286 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
287 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
288 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
289 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
290 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
291 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
292 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
293 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
294 "_xmit_NONE"};
295
296static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
297
298static inline unsigned short netdev_lock_pos(unsigned short dev_type)
299{
300 int i;
301
302 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
303 if (netdev_lock_type[i] == dev_type)
304 return i;
305 /* the last key is used by default */
306 return ARRAY_SIZE(netdev_lock_type) - 1;
307}
308
309static inline void netdev_set_lockdep_class(spinlock_t *lock,
310 unsigned short dev_type)
311{
312 int i;
313
314 i = netdev_lock_pos(dev_type);
315 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
316 netdev_lock_name[i]);
317}
318#else
319static inline void netdev_set_lockdep_class(spinlock_t *lock,
320 unsigned short dev_type)
321{
322}
323#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
325/*******************************************************************************
326
327 Protocol management and registration routines
328
329*******************************************************************************/
330
331/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 * Add a protocol ID to the list. Now that the input handler is
333 * smarter we can dispense with all the messy stuff that used to be
334 * here.
335 *
336 * BEWARE!!! Protocol handlers, mangling input packets,
337 * MUST BE last in hash buckets and checking protocol handlers
338 * MUST start from promiscuous ptype_all chain in net_bh.
339 * It is true now, do not change it.
340 * Explanation follows: if protocol handler, mangling packet, will
341 * be the first on list, it is not able to sense, that packet
342 * is cloned and should be copied-on-write, so that it will
343 * change it and subsequent readers will get broken packet.
344 * --ANK (980803)
345 */
346
347/**
348 * dev_add_pack - add packet handler
349 * @pt: packet type declaration
350 *
351 * Add a protocol handler to the networking stack. The passed &packet_type
352 * is linked into kernel lists and may not be freed until it has been
353 * removed from the kernel lists.
354 *
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900355 * This call does not sleep therefore it can not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 * guarantee all CPU's that are in middle of receiving packets
357 * will see the new packet type (until the next received packet).
358 */
359
360void dev_add_pack(struct packet_type *pt)
361{
362 int hash;
363
364 spin_lock_bh(&ptype_lock);
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -0700365 if (pt->type == htons(ETH_P_ALL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 list_add_rcu(&pt->list, &ptype_all);
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -0700367 else {
Pavel Emelyanov82d8a862007-11-26 20:12:58 +0800368 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 list_add_rcu(&pt->list, &ptype_base[hash]);
370 }
371 spin_unlock_bh(&ptype_lock);
372}
373
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374/**
375 * __dev_remove_pack - remove packet handler
376 * @pt: packet type declaration
377 *
378 * Remove a protocol handler that was previously added to the kernel
379 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
380 * from the kernel lists and can be freed or reused once this function
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900381 * returns.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 *
383 * The packet type might still be in use by receivers
384 * and must not be freed until after all the CPU's have gone
385 * through a quiescent state.
386 */
387void __dev_remove_pack(struct packet_type *pt)
388{
389 struct list_head *head;
390 struct packet_type *pt1;
391
392 spin_lock_bh(&ptype_lock);
393
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -0700394 if (pt->type == htons(ETH_P_ALL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 head = &ptype_all;
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -0700396 else
Pavel Emelyanov82d8a862007-11-26 20:12:58 +0800397 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399 list_for_each_entry(pt1, head, list) {
400 if (pt == pt1) {
401 list_del_rcu(&pt->list);
402 goto out;
403 }
404 }
405
406 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
407out:
408 spin_unlock_bh(&ptype_lock);
409}
410/**
411 * dev_remove_pack - remove packet handler
412 * @pt: packet type declaration
413 *
414 * Remove a protocol handler that was previously added to the kernel
415 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
416 * from the kernel lists and can be freed or reused once this function
417 * returns.
418 *
419 * This call sleeps to guarantee that no CPU is looking at the packet
420 * type after return.
421 */
422void dev_remove_pack(struct packet_type *pt)
423{
424 __dev_remove_pack(pt);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900425
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 synchronize_net();
427}
428
429/******************************************************************************
430
431 Device Boot-time Settings Routines
432
433*******************************************************************************/
434
435/* Boot time configuration table */
436static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
437
438/**
439 * netdev_boot_setup_add - add new setup entry
440 * @name: name of the device
441 * @map: configured settings for the device
442 *
443 * Adds new setup entry to the dev_boot_setup list. The function
444 * returns 0 on error and 1 on success. This is a generic routine to
445 * all netdevices.
446 */
447static int netdev_boot_setup_add(char *name, struct ifmap *map)
448{
449 struct netdev_boot_setup *s;
450 int i;
451
452 s = dev_boot_setup;
453 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
454 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
455 memset(s[i].name, 0, sizeof(s[i].name));
456 strcpy(s[i].name, name);
457 memcpy(&s[i].map, map, sizeof(s[i].map));
458 break;
459 }
460 }
461
462 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
463}
464
465/**
466 * netdev_boot_setup_check - check boot time settings
467 * @dev: the netdevice
468 *
469 * Check boot time settings for the device.
470 * The found settings are set for the device to be used
471 * later in the device probing.
472 * Returns 0 if no settings found, 1 if they are.
473 */
474int netdev_boot_setup_check(struct net_device *dev)
475{
476 struct netdev_boot_setup *s = dev_boot_setup;
477 int i;
478
479 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
480 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
481 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
482 dev->irq = s[i].map.irq;
483 dev->base_addr = s[i].map.base_addr;
484 dev->mem_start = s[i].map.mem_start;
485 dev->mem_end = s[i].map.mem_end;
486 return 1;
487 }
488 }
489 return 0;
490}
491
492
493/**
494 * netdev_boot_base - get address from boot time settings
495 * @prefix: prefix for network device
496 * @unit: id for network device
497 *
498 * Check boot time settings for the base address of device.
499 * The found settings are set for the device to be used
500 * later in the device probing.
501 * Returns 0 if no settings found.
502 */
503unsigned long netdev_boot_base(const char *prefix, int unit)
504{
505 const struct netdev_boot_setup *s = dev_boot_setup;
506 char name[IFNAMSIZ];
507 int i;
508
509 sprintf(name, "%s%d", prefix, unit);
510
511 /*
512 * If device already registered then return base of 1
513 * to indicate not to probe for this interface
514 */
Eric W. Biederman881d9662007-09-17 11:56:21 -0700515 if (__dev_get_by_name(&init_net, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 return 1;
517
518 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
519 if (!strcmp(name, s[i].name))
520 return s[i].map.base_addr;
521 return 0;
522}
523
524/*
525 * Saves at boot time configured settings for any netdevice.
526 */
527int __init netdev_boot_setup(char *str)
528{
529 int ints[5];
530 struct ifmap map;
531
532 str = get_options(str, ARRAY_SIZE(ints), ints);
533 if (!str || !*str)
534 return 0;
535
536 /* Save settings */
537 memset(&map, 0, sizeof(map));
538 if (ints[0] > 0)
539 map.irq = ints[1];
540 if (ints[0] > 1)
541 map.base_addr = ints[2];
542 if (ints[0] > 2)
543 map.mem_start = ints[3];
544 if (ints[0] > 3)
545 map.mem_end = ints[4];
546
547 /* Add new entry to the list */
548 return netdev_boot_setup_add(str, &map);
549}
550
551__setup("netdev=", netdev_boot_setup);
552
553/*******************************************************************************
554
555 Device Interface Subroutines
556
557*******************************************************************************/
558
559/**
560 * __dev_get_by_name - find a device by its name
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700561 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 * @name: name to find
563 *
564 * Find an interface by name. Must be called under RTNL semaphore
565 * or @dev_base_lock. If the name is found a pointer to the device
566 * is returned. If the name is not found then %NULL is returned. The
567 * reference counters are not incremented so the caller must be
568 * careful with locks.
569 */
570
Eric W. Biederman881d9662007-09-17 11:56:21 -0700571struct net_device *__dev_get_by_name(struct net *net, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
573 struct hlist_node *p;
574
Eric W. Biederman881d9662007-09-17 11:56:21 -0700575 hlist_for_each(p, dev_name_hash(net, name)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 struct net_device *dev
577 = hlist_entry(p, struct net_device, name_hlist);
578 if (!strncmp(dev->name, name, IFNAMSIZ))
579 return dev;
580 }
581 return NULL;
582}
583
584/**
585 * dev_get_by_name - find a device by its name
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700586 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 * @name: name to find
588 *
589 * Find an interface by name. This can be called from any
590 * context and does its own locking. The returned handle has
591 * the usage count incremented and the caller must use dev_put() to
592 * release it when it is no longer needed. %NULL is returned if no
593 * matching device is found.
594 */
595
Eric W. Biederman881d9662007-09-17 11:56:21 -0700596struct net_device *dev_get_by_name(struct net *net, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
598 struct net_device *dev;
599
600 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700601 dev = __dev_get_by_name(net, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 if (dev)
603 dev_hold(dev);
604 read_unlock(&dev_base_lock);
605 return dev;
606}
607
608/**
609 * __dev_get_by_index - find a device by its ifindex
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700610 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 * @ifindex: index of device
612 *
613 * Search for an interface by index. Returns %NULL if the device
614 * is not found or a pointer to the device. The device has not
615 * had its reference counter increased so the caller must be careful
616 * about locking. The caller must hold either the RTNL semaphore
617 * or @dev_base_lock.
618 */
619
Eric W. Biederman881d9662007-09-17 11:56:21 -0700620struct net_device *__dev_get_by_index(struct net *net, int ifindex)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
622 struct hlist_node *p;
623
Eric W. Biederman881d9662007-09-17 11:56:21 -0700624 hlist_for_each(p, dev_index_hash(net, ifindex)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 struct net_device *dev
626 = hlist_entry(p, struct net_device, index_hlist);
627 if (dev->ifindex == ifindex)
628 return dev;
629 }
630 return NULL;
631}
632
633
634/**
635 * dev_get_by_index - find a device by its ifindex
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700636 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 * @ifindex: index of device
638 *
639 * Search for an interface by index. Returns NULL if the device
640 * is not found or a pointer to the device. The device returned has
641 * had a reference added and the pointer is safe until the user calls
642 * dev_put to indicate they have finished with it.
643 */
644
Eric W. Biederman881d9662007-09-17 11:56:21 -0700645struct net_device *dev_get_by_index(struct net *net, int ifindex)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
647 struct net_device *dev;
648
649 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700650 dev = __dev_get_by_index(net, ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 if (dev)
652 dev_hold(dev);
653 read_unlock(&dev_base_lock);
654 return dev;
655}
656
657/**
658 * dev_getbyhwaddr - find a device by its hardware address
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700659 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 * @type: media type of device
661 * @ha: hardware address
662 *
663 * Search for an interface by MAC address. Returns NULL if the device
664 * is not found or a pointer to the device. The caller must hold the
665 * rtnl semaphore. The returned device has not had its ref count increased
666 * and the caller must therefore be careful about locking
667 *
668 * BUGS:
669 * If the API was consistent this would be __dev_get_by_hwaddr
670 */
671
Eric W. Biederman881d9662007-09-17 11:56:21 -0700672struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
674 struct net_device *dev;
675
676 ASSERT_RTNL();
677
Denis V. Lunev81103a52007-12-12 10:47:38 -0800678 for_each_netdev(net, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 if (dev->type == type &&
680 !memcmp(dev->dev_addr, ha, dev->addr_len))
Pavel Emelianov7562f872007-05-03 15:13:45 -0700681 return dev;
682
683 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685
Jochen Friedrichcf309e32005-09-22 04:44:55 -0300686EXPORT_SYMBOL(dev_getbyhwaddr);
687
Eric W. Biederman881d9662007-09-17 11:56:21 -0700688struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
Patrick McHardy4e9cac22007-05-03 03:28:13 -0700689{
690 struct net_device *dev;
691
692 ASSERT_RTNL();
Eric W. Biederman881d9662007-09-17 11:56:21 -0700693 for_each_netdev(net, dev)
Patrick McHardy4e9cac22007-05-03 03:28:13 -0700694 if (dev->type == type)
Pavel Emelianov7562f872007-05-03 15:13:45 -0700695 return dev;
696
697 return NULL;
Patrick McHardy4e9cac22007-05-03 03:28:13 -0700698}
699
700EXPORT_SYMBOL(__dev_getfirstbyhwtype);
701
Eric W. Biederman881d9662007-09-17 11:56:21 -0700702struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703{
704 struct net_device *dev;
705
706 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -0700707 dev = __dev_getfirstbyhwtype(net, type);
Patrick McHardy4e9cac22007-05-03 03:28:13 -0700708 if (dev)
709 dev_hold(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 rtnl_unlock();
711 return dev;
712}
713
714EXPORT_SYMBOL(dev_getfirstbyhwtype);
715
716/**
717 * dev_get_by_flags - find any device with given flags
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700718 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 * @if_flags: IFF_* values
720 * @mask: bitmask of bits in if_flags to check
721 *
722 * Search for any interface with the given flags. Returns NULL if a device
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900723 * is not found or a pointer to the device. The device returned has
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 * had a reference added and the pointer is safe until the user calls
725 * dev_put to indicate they have finished with it.
726 */
727
Eric W. Biederman881d9662007-09-17 11:56:21 -0700728struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
Pavel Emelianov7562f872007-05-03 15:13:45 -0700730 struct net_device *dev, *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Pavel Emelianov7562f872007-05-03 15:13:45 -0700732 ret = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700734 for_each_netdev(net, dev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 if (((dev->flags ^ if_flags) & mask) == 0) {
736 dev_hold(dev);
Pavel Emelianov7562f872007-05-03 15:13:45 -0700737 ret = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 break;
739 }
740 }
741 read_unlock(&dev_base_lock);
Pavel Emelianov7562f872007-05-03 15:13:45 -0700742 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
745/**
746 * dev_valid_name - check if name is okay for network device
747 * @name: name string
748 *
749 * Network device names need to be valid file names to
David S. Millerc7fa9d12006-08-15 16:34:13 -0700750 * to allow sysfs to work. We also disallow any kind of
751 * whitespace.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 */
Mitch Williamsc2373ee2005-11-09 10:34:45 -0800753int dev_valid_name(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
David S. Millerc7fa9d12006-08-15 16:34:13 -0700755 if (*name == '\0')
756 return 0;
Stephen Hemmingerb6fe17d2006-08-29 17:06:13 -0700757 if (strlen(name) >= IFNAMSIZ)
758 return 0;
David S. Millerc7fa9d12006-08-15 16:34:13 -0700759 if (!strcmp(name, ".") || !strcmp(name, ".."))
760 return 0;
761
762 while (*name) {
763 if (*name == '/' || isspace(*name))
764 return 0;
765 name++;
766 }
767 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768}
769
770/**
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200771 * __dev_alloc_name - allocate a name for a device
772 * @net: network namespace to allocate the device name in
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 * @name: name format string
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200774 * @buf: scratch buffer and result name string
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 *
776 * Passed a format string - eg "lt%d" it will try and find a suitable
Stephen Hemminger3041a062006-05-26 13:25:24 -0700777 * id. It scans list of devices to build up a free map, then chooses
778 * the first empty slot. The caller must hold the dev_base or rtnl lock
779 * while allocating the name and adding the device in order to avoid
780 * duplicates.
781 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
782 * Returns the number of the unit assigned or a negative errno code.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 */
784
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200785static int __dev_alloc_name(struct net *net, const char *name, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
787 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 const char *p;
789 const int max_netdevices = 8*PAGE_SIZE;
Stephen Hemmingercfcabdc2007-10-09 01:59:42 -0700790 unsigned long *inuse;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 struct net_device *d;
792
793 p = strnchr(name, IFNAMSIZ-1, '%');
794 if (p) {
795 /*
796 * Verify the string as this thing may have come from
797 * the user. There must be either one "%d" and no other "%"
798 * characters.
799 */
800 if (p[1] != 'd' || strchr(p + 2, '%'))
801 return -EINVAL;
802
803 /* Use one page as a bit array of possible slots */
Stephen Hemmingercfcabdc2007-10-09 01:59:42 -0700804 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 if (!inuse)
806 return -ENOMEM;
807
Eric W. Biederman881d9662007-09-17 11:56:21 -0700808 for_each_netdev(net, d) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 if (!sscanf(d->name, name, &i))
810 continue;
811 if (i < 0 || i >= max_netdevices)
812 continue;
813
814 /* avoid cases where sscanf is not exact inverse of printf */
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200815 snprintf(buf, IFNAMSIZ, name, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 if (!strncmp(buf, d->name, IFNAMSIZ))
817 set_bit(i, inuse);
818 }
819
820 i = find_first_zero_bit(inuse, max_netdevices);
821 free_page((unsigned long) inuse);
822 }
823
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200824 snprintf(buf, IFNAMSIZ, name, i);
825 if (!__dev_get_by_name(net, buf))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
828 /* It is possible to run out of possible slots
829 * when the name is long and there isn't enough space left
830 * for the digits, or if all bits are used.
831 */
832 return -ENFILE;
833}
834
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200835/**
836 * dev_alloc_name - allocate a name for a device
837 * @dev: device
838 * @name: name format string
839 *
840 * Passed a format string - eg "lt%d" it will try and find a suitable
841 * id. It scans list of devices to build up a free map, then chooses
842 * the first empty slot. The caller must hold the dev_base or rtnl lock
843 * while allocating the name and adding the device in order to avoid
844 * duplicates.
845 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
846 * Returns the number of the unit assigned or a negative errno code.
847 */
848
849int dev_alloc_name(struct net_device *dev, const char *name)
850{
851 char buf[IFNAMSIZ];
852 struct net *net;
853 int ret;
854
855 BUG_ON(!dev->nd_net);
856 net = dev->nd_net;
857 ret = __dev_alloc_name(net, name, buf);
858 if (ret >= 0)
859 strlcpy(dev->name, buf, IFNAMSIZ);
860 return ret;
861}
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864/**
865 * dev_change_name - change name of a device
866 * @dev: device
867 * @newname: name (or format string) must be at least IFNAMSIZ
868 *
869 * Change name of a device, can pass format strings "eth%d".
870 * for wildcarding.
871 */
872int dev_change_name(struct net_device *dev, char *newname)
873{
Herbert Xufcc5a032007-07-30 17:03:38 -0700874 char oldname[IFNAMSIZ];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 int err = 0;
Herbert Xufcc5a032007-07-30 17:03:38 -0700876 int ret;
Eric W. Biederman881d9662007-09-17 11:56:21 -0700877 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878
879 ASSERT_RTNL();
Eric W. Biederman881d9662007-09-17 11:56:21 -0700880 BUG_ON(!dev->nd_net);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
Eric W. Biederman881d9662007-09-17 11:56:21 -0700882 net = dev->nd_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 if (dev->flags & IFF_UP)
884 return -EBUSY;
885
886 if (!dev_valid_name(newname))
887 return -EINVAL;
888
Stephen Hemmingerc8d90dc2007-10-26 03:53:42 -0700889 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
890 return 0;
891
Herbert Xufcc5a032007-07-30 17:03:38 -0700892 memcpy(oldname, dev->name, IFNAMSIZ);
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 if (strchr(newname, '%')) {
895 err = dev_alloc_name(dev, newname);
896 if (err < 0)
897 return err;
898 strcpy(newname, dev->name);
899 }
Eric W. Biederman881d9662007-09-17 11:56:21 -0700900 else if (__dev_get_by_name(net, newname))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 return -EEXIST;
902 else
903 strlcpy(dev->name, newname, IFNAMSIZ);
904
Herbert Xufcc5a032007-07-30 17:03:38 -0700905rollback:
Eric W. Biederman92749822007-04-03 00:07:30 -0600906 device_rename(&dev->dev, dev->name);
Herbert Xu7f988ea2007-07-30 16:35:46 -0700907
908 write_lock_bh(&dev_base_lock);
Eric W. Biederman92749822007-04-03 00:07:30 -0600909 hlist_del(&dev->name_hlist);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700910 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
Herbert Xu7f988ea2007-07-30 16:35:46 -0700911 write_unlock_bh(&dev_base_lock);
912
Pavel Emelyanov056925a2007-09-16 15:42:43 -0700913 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
Herbert Xufcc5a032007-07-30 17:03:38 -0700914 ret = notifier_to_errno(ret);
915
916 if (ret) {
917 if (err) {
918 printk(KERN_ERR
919 "%s: name change rollback failed: %d.\n",
920 dev->name, ret);
921 } else {
922 err = ret;
923 memcpy(dev->name, oldname, IFNAMSIZ);
924 goto rollback;
925 }
926 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
928 return err;
929}
930
931/**
Stephen Hemminger3041a062006-05-26 13:25:24 -0700932 * netdev_features_change - device changes features
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700933 * @dev: device to cause notification
934 *
935 * Called to indicate a device has changed features.
936 */
937void netdev_features_change(struct net_device *dev)
938{
Pavel Emelyanov056925a2007-09-16 15:42:43 -0700939 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700940}
941EXPORT_SYMBOL(netdev_features_change);
942
943/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 * netdev_state_change - device changes state
945 * @dev: device to cause notification
946 *
947 * Called to indicate a device has changed state. This function calls
948 * the notifier chains for netdev_chain and sends a NEWLINK message
949 * to the routing socket.
950 */
951void netdev_state_change(struct net_device *dev)
952{
953 if (dev->flags & IFF_UP) {
Pavel Emelyanov056925a2007-09-16 15:42:43 -0700954 call_netdevice_notifiers(NETDEV_CHANGE, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
956 }
957}
958
959/**
960 * dev_load - load a network module
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700961 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 * @name: name of interface
963 *
964 * If a network interface is not present and the process has suitable
965 * privileges this function loads the module. If module loading is not
966 * available in this kernel then it becomes a nop.
967 */
968
Eric W. Biederman881d9662007-09-17 11:56:21 -0700969void dev_load(struct net *net, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970{
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900971 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700974 dev = __dev_get_by_name(net, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 read_unlock(&dev_base_lock);
976
977 if (!dev && capable(CAP_SYS_MODULE))
978 request_module("%s", name);
979}
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981/**
982 * dev_open - prepare an interface for use.
983 * @dev: device to open
984 *
985 * Takes a device from down to up state. The device's private open
986 * function is invoked and then the multicast lists are loaded. Finally
987 * the device is moved into the up state and a %NETDEV_UP message is
988 * sent to the netdev notifier chain.
989 *
990 * Calling this function on an active interface is a nop. On a failure
991 * a negative errno code is returned.
992 */
993int dev_open(struct net_device *dev)
994{
995 int ret = 0;
996
997 /*
998 * Is it already up?
999 */
1000
1001 if (dev->flags & IFF_UP)
1002 return 0;
1003
1004 /*
1005 * Is it even present?
1006 */
1007 if (!netif_device_present(dev))
1008 return -ENODEV;
1009
1010 /*
1011 * Call device private open method
1012 */
1013 set_bit(__LINK_STATE_START, &dev->state);
Jeff Garzikbada3392007-10-23 20:19:37 -07001014
1015 if (dev->validate_addr)
1016 ret = dev->validate_addr(dev);
1017
1018 if (!ret && dev->open)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 ret = dev->open(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001021 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 * If it went open OK then:
1023 */
1024
Jeff Garzikbada3392007-10-23 20:19:37 -07001025 if (ret)
1026 clear_bit(__LINK_STATE_START, &dev->state);
1027 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 /*
1029 * Set the flags.
1030 */
1031 dev->flags |= IFF_UP;
1032
1033 /*
1034 * Initialize multicasting status
1035 */
Patrick McHardy4417da62007-06-27 01:28:10 -07001036 dev_set_rx_mode(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
1038 /*
1039 * Wakeup transmit queue engine
1040 */
1041 dev_activate(dev);
1042
1043 /*
1044 * ... and announce new interface.
1045 */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07001046 call_netdevice_notifiers(NETDEV_UP, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
Jeff Garzikbada3392007-10-23 20:19:37 -07001048
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 return ret;
1050}
1051
1052/**
1053 * dev_close - shutdown an interface.
1054 * @dev: device to shutdown
1055 *
1056 * This function moves an active device into down state. A
1057 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1058 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1059 * chain.
1060 */
1061int dev_close(struct net_device *dev)
1062{
David S. Miller9d5010d2007-09-12 14:33:25 +02001063 might_sleep();
1064
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 if (!(dev->flags & IFF_UP))
1066 return 0;
1067
1068 /*
1069 * Tell people we are going down, so that they can
1070 * prepare to death, when device is still operating.
1071 */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07001072 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
1074 dev_deactivate(dev);
1075
1076 clear_bit(__LINK_STATE_START, &dev->state);
1077
1078 /* Synchronize to scheduled poll. We cannot touch poll list,
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001079 * it can be even on different cpu. So just clear netif_running().
1080 *
1081 * dev->stop() will invoke napi_disable() on all of it's
1082 * napi_struct instances on this device.
1083 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 smp_mb__after_clear_bit(); /* Commit netif_running(). */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
1086 /*
1087 * Call the device specific close. This cannot fail.
1088 * Only if device is UP
1089 *
1090 * We allow it to be called even after a DETACH hot-plug
1091 * event.
1092 */
1093 if (dev->stop)
1094 dev->stop(dev);
1095
1096 /*
1097 * Device is now down.
1098 */
1099
1100 dev->flags &= ~IFF_UP;
1101
1102 /*
1103 * Tell people we are down
1104 */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07001105 call_netdevice_notifiers(NETDEV_DOWN, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
1107 return 0;
1108}
1109
1110
Eric W. Biederman881d9662007-09-17 11:56:21 -07001111static int dev_boot_phase = 1;
1112
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113/*
1114 * Device change register/unregister. These are not inline or static
1115 * as we export them to the world.
1116 */
1117
1118/**
1119 * register_netdevice_notifier - register a network notifier block
1120 * @nb: notifier
1121 *
1122 * Register a notifier to be called when network device events occur.
1123 * The notifier passed is linked into the kernel structures and must
1124 * not be reused until it has been unregistered. A negative errno code
1125 * is returned on a failure.
1126 *
1127 * When registered all registration and up events are replayed
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001128 * to the new notifier to allow device to have a race free
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 * view of the network device list.
1130 */
1131
1132int register_netdevice_notifier(struct notifier_block *nb)
1133{
1134 struct net_device *dev;
Herbert Xufcc5a032007-07-30 17:03:38 -07001135 struct net_device *last;
Eric W. Biederman881d9662007-09-17 11:56:21 -07001136 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 int err;
1138
1139 rtnl_lock();
Alan Sternf07d5b92006-05-09 15:23:03 -07001140 err = raw_notifier_chain_register(&netdev_chain, nb);
Herbert Xufcc5a032007-07-30 17:03:38 -07001141 if (err)
1142 goto unlock;
Eric W. Biederman881d9662007-09-17 11:56:21 -07001143 if (dev_boot_phase)
1144 goto unlock;
1145 for_each_net(net) {
1146 for_each_netdev(net, dev) {
1147 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1148 err = notifier_to_errno(err);
1149 if (err)
1150 goto rollback;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
Eric W. Biederman881d9662007-09-17 11:56:21 -07001152 if (!(dev->flags & IFF_UP))
1153 continue;
Herbert Xufcc5a032007-07-30 17:03:38 -07001154
Eric W. Biederman881d9662007-09-17 11:56:21 -07001155 nb->notifier_call(nb, NETDEV_UP, dev);
1156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 }
Herbert Xufcc5a032007-07-30 17:03:38 -07001158
1159unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 rtnl_unlock();
1161 return err;
Herbert Xufcc5a032007-07-30 17:03:38 -07001162
1163rollback:
1164 last = dev;
Eric W. Biederman881d9662007-09-17 11:56:21 -07001165 for_each_net(net) {
1166 for_each_netdev(net, dev) {
1167 if (dev == last)
1168 break;
Herbert Xufcc5a032007-07-30 17:03:38 -07001169
Eric W. Biederman881d9662007-09-17 11:56:21 -07001170 if (dev->flags & IFF_UP) {
1171 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1172 nb->notifier_call(nb, NETDEV_DOWN, dev);
1173 }
1174 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
Herbert Xufcc5a032007-07-30 17:03:38 -07001175 }
Herbert Xufcc5a032007-07-30 17:03:38 -07001176 }
Pavel Emelyanovc67625a2007-11-14 15:53:16 -08001177
1178 raw_notifier_chain_unregister(&netdev_chain, nb);
Herbert Xufcc5a032007-07-30 17:03:38 -07001179 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180}
1181
1182/**
1183 * unregister_netdevice_notifier - unregister a network notifier block
1184 * @nb: notifier
1185 *
1186 * Unregister a notifier previously registered by
1187 * register_netdevice_notifier(). The notifier is unlinked into the
1188 * kernel structures and may then be reused. A negative errno code
1189 * is returned on a failure.
1190 */
1191
1192int unregister_netdevice_notifier(struct notifier_block *nb)
1193{
Herbert Xu9f514952006-03-25 01:24:25 -08001194 int err;
1195
1196 rtnl_lock();
Alan Sternf07d5b92006-05-09 15:23:03 -07001197 err = raw_notifier_chain_unregister(&netdev_chain, nb);
Herbert Xu9f514952006-03-25 01:24:25 -08001198 rtnl_unlock();
1199 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200}
1201
1202/**
1203 * call_netdevice_notifiers - call all network notifier blocks
1204 * @val: value passed unmodified to notifier function
Randy Dunlapc4ea43c2007-10-12 21:17:49 -07001205 * @dev: net_device pointer passed unmodified to notifier function
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 *
1207 * Call all network notifier blocks. Parameters and return value
Alan Sternf07d5b92006-05-09 15:23:03 -07001208 * are as for raw_notifier_call_chain().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 */
1210
Eric W. Biedermanad7379d2007-09-16 15:33:32 -07001211int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212{
Eric W. Biedermanad7379d2007-09-16 15:33:32 -07001213 return raw_notifier_call_chain(&netdev_chain, val, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214}
1215
1216/* When > 0 there are consumers of rx skb time stamps */
1217static atomic_t netstamp_needed = ATOMIC_INIT(0);
1218
1219void net_enable_timestamp(void)
1220{
1221 atomic_inc(&netstamp_needed);
1222}
1223
1224void net_disable_timestamp(void)
1225{
1226 atomic_dec(&netstamp_needed);
1227}
1228
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001229static inline void net_timestamp(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230{
1231 if (atomic_read(&netstamp_needed))
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001232 __net_timestamp(skb);
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07001233 else
1234 skb->tstamp.tv64 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
1236
1237/*
1238 * Support routine. Sends outgoing frames to any network
1239 * taps currently in use.
1240 */
1241
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001242static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
1244 struct packet_type *ptype;
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001245
1246 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
1248 rcu_read_lock();
1249 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1250 /* Never send packets back to the socket
1251 * they originated from - MvS (miquels@drinkel.ow.org)
1252 */
1253 if ((ptype->dev == dev || !ptype->dev) &&
1254 (ptype->af_packet_priv == NULL ||
1255 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1256 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1257 if (!skb2)
1258 break;
1259
1260 /* skb->nh should be correctly
1261 set by sender, so that the second statement is
1262 just protection against buggy protocols.
1263 */
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07001264 skb_reset_mac_header(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001266 if (skb_network_header(skb2) < skb2->data ||
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001267 skb2->network_header > skb2->tail) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 if (net_ratelimit())
1269 printk(KERN_CRIT "protocol %04x is "
1270 "buggy, dev %s\n",
1271 skb2->protocol, dev->name);
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -07001272 skb_reset_network_header(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 }
1274
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001275 skb2->transport_header = skb2->network_header;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 skb2->pkt_type = PACKET_OUTGOING;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001277 ptype->func(skb2, skb->dev, ptype, skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 }
1279 }
1280 rcu_read_unlock();
1281}
1282
Denis Vlasenko56079432006-03-29 15:57:29 -08001283
1284void __netif_schedule(struct net_device *dev)
1285{
1286 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1287 unsigned long flags;
1288 struct softnet_data *sd;
1289
1290 local_irq_save(flags);
1291 sd = &__get_cpu_var(softnet_data);
1292 dev->next_sched = sd->output_queue;
1293 sd->output_queue = dev;
1294 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1295 local_irq_restore(flags);
1296 }
1297}
1298EXPORT_SYMBOL(__netif_schedule);
1299
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001300void dev_kfree_skb_irq(struct sk_buff *skb)
Denis Vlasenko56079432006-03-29 15:57:29 -08001301{
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001302 if (atomic_dec_and_test(&skb->users)) {
1303 struct softnet_data *sd;
1304 unsigned long flags;
Denis Vlasenko56079432006-03-29 15:57:29 -08001305
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001306 local_irq_save(flags);
1307 sd = &__get_cpu_var(softnet_data);
1308 skb->next = sd->completion_queue;
1309 sd->completion_queue = skb;
1310 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1311 local_irq_restore(flags);
1312 }
Denis Vlasenko56079432006-03-29 15:57:29 -08001313}
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001314EXPORT_SYMBOL(dev_kfree_skb_irq);
Denis Vlasenko56079432006-03-29 15:57:29 -08001315
1316void dev_kfree_skb_any(struct sk_buff *skb)
1317{
1318 if (in_irq() || irqs_disabled())
1319 dev_kfree_skb_irq(skb);
1320 else
1321 dev_kfree_skb(skb);
1322}
1323EXPORT_SYMBOL(dev_kfree_skb_any);
1324
1325
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001326/**
1327 * netif_device_detach - mark device as removed
1328 * @dev: network device
1329 *
1330 * Mark device as removed from system and therefore no longer available.
1331 */
Denis Vlasenko56079432006-03-29 15:57:29 -08001332void netif_device_detach(struct net_device *dev)
1333{
1334 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1335 netif_running(dev)) {
1336 netif_stop_queue(dev);
1337 }
1338}
1339EXPORT_SYMBOL(netif_device_detach);
1340
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001341/**
1342 * netif_device_attach - mark device as attached
1343 * @dev: network device
1344 *
1345 * Mark device as attached from system and restart if needed.
1346 */
Denis Vlasenko56079432006-03-29 15:57:29 -08001347void netif_device_attach(struct net_device *dev)
1348{
1349 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1350 netif_running(dev)) {
1351 netif_wake_queue(dev);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001352 __netdev_watchdog_up(dev);
Denis Vlasenko56079432006-03-29 15:57:29 -08001353 }
1354}
1355EXPORT_SYMBOL(netif_device_attach);
1356
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358/*
1359 * Invalidate hardware checksum when packet is to be mangled, and
1360 * complete checksum manually on outgoing path.
1361 */
Patrick McHardy84fa7932006-08-29 16:44:56 -07001362int skb_checksum_help(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363{
Al Virod3bc23e2006-11-14 21:24:49 -08001364 __wsum csum;
Herbert Xu663ead32007-04-09 11:59:07 -07001365 int ret = 0, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
Patrick McHardy84fa7932006-08-29 16:44:56 -07001367 if (skb->ip_summed == CHECKSUM_COMPLETE)
Herbert Xua430a432006-07-08 13:34:56 -07001368 goto out_set_summed;
1369
1370 if (unlikely(skb_shinfo(skb)->gso_size)) {
Herbert Xua430a432006-07-08 13:34:56 -07001371 /* Let GSO fix up the checksum. */
1372 goto out_set_summed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 }
1374
Herbert Xua0308472007-10-15 01:47:15 -07001375 offset = skb->csum_start - skb_headroom(skb);
1376 BUG_ON(offset >= skb_headlen(skb));
1377 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1378
1379 offset += skb->csum_offset;
1380 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1381
1382 if (skb_cloned(skb) &&
1383 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1385 if (ret)
1386 goto out;
1387 }
1388
Herbert Xua0308472007-10-15 01:47:15 -07001389 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
Herbert Xua430a432006-07-08 13:34:56 -07001390out_set_summed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 skb->ip_summed = CHECKSUM_NONE;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001392out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 return ret;
1394}
1395
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001396/**
1397 * skb_gso_segment - Perform segmentation on skb.
1398 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07001399 * @features: features for the output path (see dev->features)
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001400 *
1401 * This function segments the given skb and returns a list of segments.
Herbert Xu576a30e2006-06-27 13:22:38 -07001402 *
1403 * It may return NULL if the skb requires no segmentation. This is
1404 * only possible when GSO is used for verifying header integrity.
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001405 */
Herbert Xu576a30e2006-06-27 13:22:38 -07001406struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001407{
1408 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1409 struct packet_type *ptype;
Al Viro252e3342006-11-14 20:48:11 -08001410 __be16 type = skb->protocol;
Herbert Xua430a432006-07-08 13:34:56 -07001411 int err;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001412
1413 BUG_ON(skb_shinfo(skb)->frag_list);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001414
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07001415 skb_reset_mac_header(skb);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001416 skb->mac_len = skb->network_header - skb->mac_header;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001417 __skb_pull(skb, skb->mac_len);
1418
Herbert Xuf9d106a2007-04-23 22:36:13 -07001419 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
Herbert Xua430a432006-07-08 13:34:56 -07001420 if (skb_header_cloned(skb) &&
1421 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1422 return ERR_PTR(err);
1423 }
1424
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001425 rcu_read_lock();
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08001426 list_for_each_entry_rcu(ptype,
1427 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001428 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
Patrick McHardy84fa7932006-08-29 16:44:56 -07001429 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
Herbert Xua430a432006-07-08 13:34:56 -07001430 err = ptype->gso_send_check(skb);
1431 segs = ERR_PTR(err);
1432 if (err || skb_gso_ok(skb, features))
1433 break;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001434 __skb_push(skb, (skb->data -
1435 skb_network_header(skb)));
Herbert Xua430a432006-07-08 13:34:56 -07001436 }
Herbert Xu576a30e2006-06-27 13:22:38 -07001437 segs = ptype->gso_segment(skb, features);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001438 break;
1439 }
1440 }
1441 rcu_read_unlock();
1442
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001443 __skb_push(skb, skb->data - skb_mac_header(skb));
Herbert Xu576a30e2006-06-27 13:22:38 -07001444
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001445 return segs;
1446}
1447
1448EXPORT_SYMBOL(skb_gso_segment);
1449
Herbert Xufb286bb2005-11-10 13:01:24 -08001450/* Take action when hardware reception checksum errors are detected. */
1451#ifdef CONFIG_BUG
1452void netdev_rx_csum_fault(struct net_device *dev)
1453{
1454 if (net_ratelimit()) {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001455 printk(KERN_ERR "%s: hw csum failure.\n",
Stephen Hemminger246a4212005-12-08 15:21:39 -08001456 dev ? dev->name : "<unknown>");
Herbert Xufb286bb2005-11-10 13:01:24 -08001457 dump_stack();
1458 }
1459}
1460EXPORT_SYMBOL(netdev_rx_csum_fault);
1461#endif
1462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463/* Actually, we should eliminate this check as soon as we know, that:
1464 * 1. IOMMU is present and allows to map all the memory.
1465 * 2. No high memory really exists on this machine.
1466 */
1467
1468static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1469{
Herbert Xu3d3a8532006-06-27 13:33:10 -07001470#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 int i;
1472
1473 if (dev->features & NETIF_F_HIGHDMA)
1474 return 0;
1475
1476 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1477 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1478 return 1;
1479
Herbert Xu3d3a8532006-06-27 13:33:10 -07001480#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 return 0;
1482}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001484struct dev_gso_cb {
1485 void (*destructor)(struct sk_buff *skb);
1486};
1487
1488#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1489
1490static void dev_gso_skb_destructor(struct sk_buff *skb)
1491{
1492 struct dev_gso_cb *cb;
1493
1494 do {
1495 struct sk_buff *nskb = skb->next;
1496
1497 skb->next = nskb->next;
1498 nskb->next = NULL;
1499 kfree_skb(nskb);
1500 } while (skb->next);
1501
1502 cb = DEV_GSO_CB(skb);
1503 if (cb->destructor)
1504 cb->destructor(skb);
1505}
1506
1507/**
1508 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1509 * @skb: buffer to segment
1510 *
1511 * This function segments the given skb and stores the list of segments
1512 * in skb->next.
1513 */
1514static int dev_gso_segment(struct sk_buff *skb)
1515{
1516 struct net_device *dev = skb->dev;
1517 struct sk_buff *segs;
Herbert Xu576a30e2006-06-27 13:22:38 -07001518 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1519 NETIF_F_SG : 0);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001520
Herbert Xu576a30e2006-06-27 13:22:38 -07001521 segs = skb_gso_segment(skb, features);
1522
1523 /* Verifying header integrity only. */
1524 if (!segs)
1525 return 0;
1526
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001527 if (unlikely(IS_ERR(segs)))
1528 return PTR_ERR(segs);
1529
1530 skb->next = segs;
1531 DEV_GSO_CB(skb)->destructor = skb->destructor;
1532 skb->destructor = dev_gso_skb_destructor;
1533
1534 return 0;
1535}
1536
1537int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1538{
1539 if (likely(!skb->next)) {
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -07001540 if (!list_empty(&ptype_all))
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001541 dev_queue_xmit_nit(skb, dev);
1542
Herbert Xu576a30e2006-06-27 13:22:38 -07001543 if (netif_needs_gso(dev, skb)) {
1544 if (unlikely(dev_gso_segment(skb)))
1545 goto out_kfree_skb;
1546 if (skb->next)
1547 goto gso;
1548 }
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001549
Herbert Xu576a30e2006-06-27 13:22:38 -07001550 return dev->hard_start_xmit(skb, dev);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001551 }
1552
Herbert Xu576a30e2006-06-27 13:22:38 -07001553gso:
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001554 do {
1555 struct sk_buff *nskb = skb->next;
1556 int rc;
1557
1558 skb->next = nskb->next;
1559 nskb->next = NULL;
1560 rc = dev->hard_start_xmit(nskb, dev);
1561 if (unlikely(rc)) {
Michael Chanf54d9e82006-06-25 23:57:04 -07001562 nskb->next = skb->next;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001563 skb->next = nskb;
1564 return rc;
1565 }
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07001566 if (unlikely((netif_queue_stopped(dev) ||
Pavel Emelyanov668f8952007-10-21 17:01:56 -07001567 netif_subqueue_stopped(dev, skb)) &&
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07001568 skb->next))
Michael Chanf54d9e82006-06-25 23:57:04 -07001569 return NETDEV_TX_BUSY;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001570 } while (skb->next);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001571
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001572 skb->destructor = DEV_GSO_CB(skb)->destructor;
1573
1574out_kfree_skb:
1575 kfree_skb(skb);
1576 return 0;
1577}
1578
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579/**
1580 * dev_queue_xmit - transmit a buffer
1581 * @skb: buffer to transmit
1582 *
1583 * Queue a buffer for transmission to a network device. The caller must
1584 * have set the device and priority and built the buffer before calling
1585 * this function. The function can be called from an interrupt.
1586 *
1587 * A negative errno code is returned on a failure. A success does not
1588 * guarantee the frame will be transmitted as it may be dropped due
1589 * to congestion or traffic shaping.
Ben Greearaf191362005-04-24 20:12:36 -07001590 *
1591 * -----------------------------------------------------------------------------------
1592 * I notice this method can also return errors from the queue disciplines,
1593 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1594 * be positive.
1595 *
1596 * Regardless of the return value, the skb is consumed, so it is currently
1597 * difficult to retry a send to this method. (You can bump the ref count
1598 * before sending to hold a reference for retry if you are careful.)
1599 *
1600 * When calling this method, interrupts MUST be enabled. This is because
1601 * the BH enable code must have IRQs enabled so that it will not deadlock.
1602 * --BLG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 */
1604
1605int dev_queue_xmit(struct sk_buff *skb)
1606{
1607 struct net_device *dev = skb->dev;
1608 struct Qdisc *q;
1609 int rc = -ENOMEM;
1610
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001611 /* GSO will handle the following emulations directly. */
1612 if (netif_needs_gso(dev, skb))
1613 goto gso;
1614
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 if (skb_shinfo(skb)->frag_list &&
1616 !(dev->features & NETIF_F_FRAGLIST) &&
Herbert Xu364c6ba2006-06-09 16:10:40 -07001617 __skb_linearize(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 goto out_kfree_skb;
1619
1620 /* Fragmented skb is linearized if device does not support SG,
1621 * or if at least one of fragments is in highmem and device
1622 * does not support DMA from it.
1623 */
1624 if (skb_shinfo(skb)->nr_frags &&
1625 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
Herbert Xu364c6ba2006-06-09 16:10:40 -07001626 __skb_linearize(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 goto out_kfree_skb;
1628
1629 /* If packet is not checksummed and device does not support
1630 * checksumming for this protocol, complete checksumming here.
1631 */
Herbert Xu663ead32007-04-09 11:59:07 -07001632 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1633 skb_set_transport_header(skb, skb->csum_start -
1634 skb_headroom(skb));
1635
Herbert Xua2988302007-06-28 13:44:37 -07001636 if (!(dev->features & NETIF_F_GEN_CSUM) &&
1637 !((dev->features & NETIF_F_IP_CSUM) &&
1638 skb->protocol == htons(ETH_P_IP)) &&
1639 !((dev->features & NETIF_F_IPV6_CSUM) &&
1640 skb->protocol == htons(ETH_P_IPV6)))
Herbert Xu663ead32007-04-09 11:59:07 -07001641 if (skb_checksum_help(skb))
1642 goto out_kfree_skb;
1643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001645gso:
Eric Dumazet2d7ceec2005-09-27 15:22:58 -07001646 spin_lock_prefetch(&dev->queue_lock);
1647
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001648 /* Disable soft irqs for various locks below. Also
1649 * stops preemption for RCU.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001651 rcu_read_lock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001653 /* Updates of qdisc are serialized by queue_lock.
1654 * The struct Qdisc which is pointed to by qdisc is now a
1655 * rcu structure - it may be accessed without acquiring
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 * a lock (but the structure may be stale.) The freeing of the
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001657 * qdisc will be deferred until it's known that there are no
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 * more references to it.
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001659 *
1660 * If the qdisc has an enqueue function, we still need to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 * hold the queue_lock before calling it, since queue_lock
1662 * also serializes access to the device queue.
1663 */
1664
1665 q = rcu_dereference(dev->qdisc);
1666#ifdef CONFIG_NET_CLS_ACT
1667 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1668#endif
1669 if (q->enqueue) {
1670 /* Grab device queue */
1671 spin_lock(&dev->queue_lock);
Patrick McHardy85670cc2006-09-27 16:45:45 -07001672 q = dev->qdisc;
1673 if (q->enqueue) {
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07001674 /* reset queue_mapping to zero */
Pavel Emelyanovdfa40912007-10-21 16:57:55 -07001675 skb_set_queue_mapping(skb, 0);
Patrick McHardy85670cc2006-09-27 16:45:45 -07001676 rc = q->enqueue(skb, q);
1677 qdisc_run(dev);
1678 spin_unlock(&dev->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Patrick McHardy85670cc2006-09-27 16:45:45 -07001680 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1681 goto out;
1682 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 spin_unlock(&dev->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 }
1685
1686 /* The device has no queue. Common case for software devices:
1687 loopback, all the sorts of tunnels...
1688
Herbert Xu932ff272006-06-09 12:20:56 -07001689 Really, it is unlikely that netif_tx_lock protection is necessary
1690 here. (f.e. loopback and IP tunnels are clean ignoring statistics
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 counters.)
1692 However, it is possible, that they rely on protection
1693 made by us here.
1694
1695 Check this and shot the lock. It is not prone from deadlocks.
1696 Either shot noqueue qdisc, it is even simpler 8)
1697 */
1698 if (dev->flags & IFF_UP) {
1699 int cpu = smp_processor_id(); /* ok because BHs are off */
1700
1701 if (dev->xmit_lock_owner != cpu) {
1702
1703 HARD_TX_LOCK(dev, cpu);
1704
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07001705 if (!netif_queue_stopped(dev) &&
Pavel Emelyanov668f8952007-10-21 17:01:56 -07001706 !netif_subqueue_stopped(dev, skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 rc = 0;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001708 if (!dev_hard_start_xmit(skb, dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 HARD_TX_UNLOCK(dev);
1710 goto out;
1711 }
1712 }
1713 HARD_TX_UNLOCK(dev);
1714 if (net_ratelimit())
1715 printk(KERN_CRIT "Virtual device %s asks to "
1716 "queue packet!\n", dev->name);
1717 } else {
1718 /* Recursion is detected! It is possible,
1719 * unfortunately */
1720 if (net_ratelimit())
1721 printk(KERN_CRIT "Dead loop on virtual device "
1722 "%s, fix it urgently!\n", dev->name);
1723 }
1724 }
1725
1726 rc = -ENETDOWN;
Herbert Xud4828d82006-06-22 02:28:18 -07001727 rcu_read_unlock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
1729out_kfree_skb:
1730 kfree_skb(skb);
1731 return rc;
1732out:
Herbert Xud4828d82006-06-22 02:28:18 -07001733 rcu_read_unlock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 return rc;
1735}
1736
1737
1738/*=======================================================================
1739 Receiver routines
1740 =======================================================================*/
1741
Stephen Hemminger6b2bedc2007-03-12 14:33:50 -07001742int netdev_max_backlog __read_mostly = 1000;
1743int netdev_budget __read_mostly = 300;
1744int weight_p __read_mostly = 64; /* old backlog weight */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
1746DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1747
1748
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749/**
1750 * netif_rx - post buffer to the network code
1751 * @skb: buffer to post
1752 *
1753 * This function receives a packet from a device driver and queues it for
1754 * the upper (protocol) levels to process. It always succeeds. The buffer
1755 * may be dropped during processing for congestion control or by the
1756 * protocol layers.
1757 *
1758 * return values:
1759 * NET_RX_SUCCESS (no congestion)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 * NET_RX_DROP (packet was dropped)
1761 *
1762 */
1763
1764int netif_rx(struct sk_buff *skb)
1765{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 struct softnet_data *queue;
1767 unsigned long flags;
1768
1769 /* if netpoll wants it, pretend we never saw it */
1770 if (netpoll_rx(skb))
1771 return NET_RX_DROP;
1772
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07001773 if (!skb->tstamp.tv64)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001774 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
1776 /*
1777 * The code is rearranged so that the path is the most
1778 * short when CPU is congested, but is still operating.
1779 */
1780 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 queue = &__get_cpu_var(softnet_data);
1782
1783 __get_cpu_var(netdev_rx_stat).total++;
1784 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1785 if (queue->input_pkt_queue.qlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786enqueue:
1787 dev_hold(skb->dev);
1788 __skb_queue_tail(&queue->input_pkt_queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 local_irq_restore(flags);
Stephen Hemminger34008d82005-06-23 20:10:00 -07001790 return NET_RX_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 }
1792
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001793 napi_schedule(&queue->backlog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 goto enqueue;
1795 }
1796
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 __get_cpu_var(netdev_rx_stat).dropped++;
1798 local_irq_restore(flags);
1799
1800 kfree_skb(skb);
1801 return NET_RX_DROP;
1802}
1803
1804int netif_rx_ni(struct sk_buff *skb)
1805{
1806 int err;
1807
1808 preempt_disable();
1809 err = netif_rx(skb);
1810 if (local_softirq_pending())
1811 do_softirq();
1812 preempt_enable();
1813
1814 return err;
1815}
1816
1817EXPORT_SYMBOL(netif_rx_ni);
1818
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001819static inline struct net_device *skb_bond(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820{
1821 struct net_device *dev = skb->dev;
1822
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001823 if (dev->master) {
David S. Miller7ea49ed2006-08-14 17:08:36 -07001824 if (skb_bond_should_drop(skb)) {
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001825 kfree_skb(skb);
1826 return NULL;
1827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 skb->dev = dev->master;
Jay Vosburgh8f903c72006-02-21 16:36:44 -08001829 }
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001830
1831 return dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832}
1833
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835static void net_tx_action(struct softirq_action *h)
1836{
1837 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1838
1839 if (sd->completion_queue) {
1840 struct sk_buff *clist;
1841
1842 local_irq_disable();
1843 clist = sd->completion_queue;
1844 sd->completion_queue = NULL;
1845 local_irq_enable();
1846
1847 while (clist) {
1848 struct sk_buff *skb = clist;
1849 clist = clist->next;
1850
1851 BUG_TRAP(!atomic_read(&skb->users));
1852 __kfree_skb(skb);
1853 }
1854 }
1855
1856 if (sd->output_queue) {
1857 struct net_device *head;
1858
1859 local_irq_disable();
1860 head = sd->output_queue;
1861 sd->output_queue = NULL;
1862 local_irq_enable();
1863
1864 while (head) {
1865 struct net_device *dev = head;
1866 head = head->next_sched;
1867
1868 smp_mb__before_clear_bit();
1869 clear_bit(__LINK_STATE_SCHED, &dev->state);
1870
1871 if (spin_trylock(&dev->queue_lock)) {
1872 qdisc_run(dev);
1873 spin_unlock(&dev->queue_lock);
1874 } else {
1875 netif_schedule(dev);
1876 }
1877 }
1878 }
1879}
1880
Stephen Hemminger6f05f622007-03-08 20:46:03 -08001881static inline int deliver_skb(struct sk_buff *skb,
1882 struct packet_type *pt_prev,
1883 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884{
1885 atomic_inc(&skb->users);
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001886 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887}
1888
1889#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
Stephen Hemminger6229e362007-03-21 13:38:47 -07001890/* These hooks defined here for ATM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891struct net_bridge;
1892struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1893 unsigned char *addr);
Stephen Hemminger6229e362007-03-21 13:38:47 -07001894void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
Stephen Hemminger6229e362007-03-21 13:38:47 -07001896/*
1897 * If bridge module is loaded call bridging hook.
1898 * returns NULL if packet was consumed.
1899 */
1900struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1901 struct sk_buff *skb) __read_mostly;
1902static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1903 struct packet_type **pt_prev, int *ret,
1904 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905{
1906 struct net_bridge_port *port;
1907
Stephen Hemminger6229e362007-03-21 13:38:47 -07001908 if (skb->pkt_type == PACKET_LOOPBACK ||
1909 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1910 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
1912 if (*pt_prev) {
Stephen Hemminger6229e362007-03-21 13:38:47 -07001913 *ret = deliver_skb(skb, *pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 *pt_prev = NULL;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001915 }
1916
Stephen Hemminger6229e362007-03-21 13:38:47 -07001917 return br_handle_frame_hook(port, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918}
1919#else
Stephen Hemminger6229e362007-03-21 13:38:47 -07001920#define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921#endif
1922
Patrick McHardyb863ceb2007-07-14 18:55:06 -07001923#if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1924struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1925EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1926
1927static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1928 struct packet_type **pt_prev,
1929 int *ret,
1930 struct net_device *orig_dev)
1931{
1932 if (skb->dev->macvlan_port == NULL)
1933 return skb;
1934
1935 if (*pt_prev) {
1936 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1937 *pt_prev = NULL;
1938 }
1939 return macvlan_handle_frame_hook(skb);
1940}
1941#else
1942#define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1943#endif
1944
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945#ifdef CONFIG_NET_CLS_ACT
1946/* TODO: Maybe we should just force sch_ingress to be compiled in
1947 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1948 * a compare and 2 stores extra right now if we dont have it on
1949 * but have CONFIG_NET_CLS_ACT
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001950 * NOTE: This doesnt stop any functionality; if you dont have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 * the ingress scheduler, you just cant add policies on ingress.
1952 *
1953 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001954static int ing_filter(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955{
1956 struct Qdisc *q;
1957 struct net_device *dev = skb->dev;
1958 int result = TC_ACT_OK;
Herbert Xuf697c3e2007-10-14 00:38:47 -07001959 u32 ttl = G_TC_RTTL(skb->tc_verd);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001960
Herbert Xuf697c3e2007-10-14 00:38:47 -07001961 if (MAX_RED_LOOP < ttl++) {
1962 printk(KERN_WARNING
1963 "Redir loop detected Dropping packet (%d->%d)\n",
1964 skb->iif, dev->ifindex);
1965 return TC_ACT_SHOT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 }
1967
Herbert Xuf697c3e2007-10-14 00:38:47 -07001968 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
1969 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
1970
1971 spin_lock(&dev->ingress_lock);
1972 if ((q = dev->qdisc_ingress) != NULL)
1973 result = q->enqueue(skb, q);
1974 spin_unlock(&dev->ingress_lock);
1975
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 return result;
1977}
Herbert Xuf697c3e2007-10-14 00:38:47 -07001978
1979static inline struct sk_buff *handle_ing(struct sk_buff *skb,
1980 struct packet_type **pt_prev,
1981 int *ret, struct net_device *orig_dev)
1982{
1983 if (!skb->dev->qdisc_ingress)
1984 goto out;
1985
1986 if (*pt_prev) {
1987 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1988 *pt_prev = NULL;
1989 } else {
1990 /* Huh? Why does turning on AF_PACKET affect this? */
1991 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1992 }
1993
1994 switch (ing_filter(skb)) {
1995 case TC_ACT_SHOT:
1996 case TC_ACT_STOLEN:
1997 kfree_skb(skb);
1998 return NULL;
1999 }
2000
2001out:
2002 skb->tc_verd = 0;
2003 return skb;
2004}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005#endif
2006
Stephen Hemminger3b582cc2007-11-01 02:21:47 -07002007/**
2008 * netif_receive_skb - process receive buffer from network
2009 * @skb: buffer to process
2010 *
2011 * netif_receive_skb() is the main receive data processing function.
2012 * It always succeeds. The buffer may be dropped during processing
2013 * for congestion control or by the protocol layers.
2014 *
2015 * This function may only be called from softirq context and interrupts
2016 * should be enabled.
2017 *
2018 * Return values (usually ignored):
2019 * NET_RX_SUCCESS: no congestion
2020 * NET_RX_DROP: packet was dropped
2021 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022int netif_receive_skb(struct sk_buff *skb)
2023{
2024 struct packet_type *ptype, *pt_prev;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002025 struct net_device *orig_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 int ret = NET_RX_DROP;
Al Viro252e3342006-11-14 20:48:11 -08002027 __be16 type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
2029 /* if we've gotten here through NAPI, check netpoll */
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002030 if (netpoll_receive_skb(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 return NET_RX_DROP;
2032
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002033 if (!skb->tstamp.tv64)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002034 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
Patrick McHardyc01003c2007-03-29 11:46:52 -07002036 if (!skb->iif)
2037 skb->iif = skb->dev->ifindex;
David S. Miller86e65da2005-08-09 19:36:29 -07002038
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002039 orig_dev = skb_bond(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Jay Vosburgh8f903c72006-02-21 16:36:44 -08002041 if (!orig_dev)
2042 return NET_RX_DROP;
2043
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 __get_cpu_var(netdev_rx_stat).total++;
2045
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -07002046 skb_reset_network_header(skb);
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03002047 skb_reset_transport_header(skb);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07002048 skb->mac_len = skb->network_header - skb->mac_header;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
2050 pt_prev = NULL;
2051
2052 rcu_read_lock();
2053
2054#ifdef CONFIG_NET_CLS_ACT
2055 if (skb->tc_verd & TC_NCLS) {
2056 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2057 goto ncls;
2058 }
2059#endif
2060
2061 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2062 if (!ptype->dev || ptype->dev == skb->dev) {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002063 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002064 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 pt_prev = ptype;
2066 }
2067 }
2068
2069#ifdef CONFIG_NET_CLS_ACT
Herbert Xuf697c3e2007-10-14 00:38:47 -07002070 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2071 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073ncls:
2074#endif
2075
Stephen Hemminger6229e362007-03-21 13:38:47 -07002076 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2077 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 goto out;
Patrick McHardyb863ceb2007-07-14 18:55:06 -07002079 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2080 if (!skb)
2081 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
2083 type = skb->protocol;
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08002084 list_for_each_entry_rcu(ptype,
2085 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 if (ptype->type == type &&
2087 (!ptype->dev || ptype->dev == skb->dev)) {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002088 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002089 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 pt_prev = ptype;
2091 }
2092 }
2093
2094 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002095 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 } else {
2097 kfree_skb(skb);
2098 /* Jamal, now you will not able to escape explaining
2099 * me how you were going to use this. :-)
2100 */
2101 ret = NET_RX_DROP;
2102 }
2103
2104out:
2105 rcu_read_unlock();
2106 return ret;
2107}
2108
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002109static int process_backlog(struct napi_struct *napi, int quota)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110{
2111 int work = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2113 unsigned long start_time = jiffies;
2114
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002115 napi->weight = weight_p;
2116 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 struct sk_buff *skb;
2118 struct net_device *dev;
2119
2120 local_irq_disable();
2121 skb = __skb_dequeue(&queue->input_pkt_queue);
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002122 if (!skb) {
2123 __napi_complete(napi);
2124 local_irq_enable();
2125 break;
2126 }
2127
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 local_irq_enable();
2129
2130 dev = skb->dev;
2131
2132 netif_receive_skb(skb);
2133
2134 dev_put(dev);
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002135 } while (++work < quota && jiffies == start_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002137 return work;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138}
2139
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002140/**
2141 * __napi_schedule - schedule for receive
Randy Dunlapc4ea43c2007-10-12 21:17:49 -07002142 * @n: entry to schedule
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002143 *
2144 * The entry's receive function will be scheduled to run
2145 */
2146void fastcall __napi_schedule(struct napi_struct *n)
2147{
2148 unsigned long flags;
2149
2150 local_irq_save(flags);
2151 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2152 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2153 local_irq_restore(flags);
2154}
2155EXPORT_SYMBOL(__napi_schedule);
2156
2157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158static void net_rx_action(struct softirq_action *h)
2159{
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002160 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 unsigned long start_time = jiffies;
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07002162 int budget = netdev_budget;
Matt Mackall53fb95d2005-08-11 19:27:43 -07002163 void *have;
2164
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 local_irq_disable();
2166
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002167 while (!list_empty(list)) {
2168 struct napi_struct *n;
2169 int work, weight;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002171 /* If softirq window is exhuasted then punt.
2172 *
2173 * Note that this is a slight policy change from the
2174 * previous NAPI code, which would allow up to 2
2175 * jiffies to pass before breaking out. The test
2176 * used to be "jiffies - start_time > 1".
2177 */
2178 if (unlikely(budget <= 0 || jiffies != start_time))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 goto softnet_break;
2180
2181 local_irq_enable();
2182
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002183 /* Even though interrupts have been re-enabled, this
2184 * access is safe because interrupts can only add new
2185 * entries to the tail of this list, and only ->poll()
2186 * calls can remove this head entry from the list.
2187 */
2188 n = list_entry(list->next, struct napi_struct, poll_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002190 have = netpoll_poll_lock(n);
2191
2192 weight = n->weight;
2193
David S. Miller0a7606c2007-10-29 21:28:47 -07002194 /* This NAPI_STATE_SCHED test is for avoiding a race
2195 * with netpoll's poll_napi(). Only the entity which
2196 * obtains the lock and sees NAPI_STATE_SCHED set will
2197 * actually make the ->poll() call. Therefore we avoid
2198 * accidently calling ->poll() when NAPI is not scheduled.
2199 */
2200 work = 0;
2201 if (test_bit(NAPI_STATE_SCHED, &n->state))
2202 work = n->poll(n, weight);
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002203
2204 WARN_ON_ONCE(work > weight);
2205
2206 budget -= work;
2207
2208 local_irq_disable();
2209
2210 /* Drivers must not modify the NAPI state if they
2211 * consume the entire weight. In such cases this code
2212 * still "owns" the NAPI instance and therefore can
2213 * move the instance around on the list at-will.
2214 */
David S. Millerfed17f32008-01-07 21:00:40 -08002215 if (unlikely(work == weight)) {
2216 if (unlikely(napi_disable_pending(n)))
2217 __napi_complete(n);
2218 else
2219 list_move_tail(&n->poll_list, list);
2220 }
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002221
2222 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 }
2224out:
Shannon Nelson515e06c2007-06-23 23:09:23 -07002225 local_irq_enable();
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002226
Chris Leechdb217332006-06-17 21:24:58 -07002227#ifdef CONFIG_NET_DMA
2228 /*
2229 * There may not be any more sk_buffs coming right now, so push
2230 * any pending DMA copies to hardware
2231 */
Dan Williamsd379b012007-07-09 11:56:42 -07002232 if (!cpus_empty(net_dma.channel_mask)) {
2233 int chan_idx;
2234 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2235 struct dma_chan *chan = net_dma.channels[chan_idx];
2236 if (chan)
2237 dma_async_memcpy_issue_pending(chan);
2238 }
Chris Leechdb217332006-06-17 21:24:58 -07002239 }
2240#endif
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002241
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 return;
2243
2244softnet_break:
2245 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2246 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2247 goto out;
2248}
2249
2250static gifconf_func_t * gifconf_list [NPROTO];
2251
2252/**
2253 * register_gifconf - register a SIOCGIF handler
2254 * @family: Address family
2255 * @gifconf: Function handler
2256 *
2257 * Register protocol dependent address dumping routines. The handler
2258 * that is passed must not be freed or reused until it has been replaced
2259 * by another handler.
2260 */
2261int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2262{
2263 if (family >= NPROTO)
2264 return -EINVAL;
2265 gifconf_list[family] = gifconf;
2266 return 0;
2267}
2268
2269
2270/*
2271 * Map an interface index to its name (SIOCGIFNAME)
2272 */
2273
2274/*
2275 * We need this ioctl for efficient implementation of the
2276 * if_indextoname() function required by the IPv6 API. Without
2277 * it, we would have to search all the interfaces to find a
2278 * match. --pb
2279 */
2280
Eric W. Biederman881d9662007-09-17 11:56:21 -07002281static int dev_ifname(struct net *net, struct ifreq __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282{
2283 struct net_device *dev;
2284 struct ifreq ifr;
2285
2286 /*
2287 * Fetch the caller's info block.
2288 */
2289
2290 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2291 return -EFAULT;
2292
2293 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -07002294 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 if (!dev) {
2296 read_unlock(&dev_base_lock);
2297 return -ENODEV;
2298 }
2299
2300 strcpy(ifr.ifr_name, dev->name);
2301 read_unlock(&dev_base_lock);
2302
2303 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2304 return -EFAULT;
2305 return 0;
2306}
2307
2308/*
2309 * Perform a SIOCGIFCONF call. This structure will change
2310 * size eventually, and there is nothing I can do about it.
2311 * Thus we will need a 'compatibility mode'.
2312 */
2313
Eric W. Biederman881d9662007-09-17 11:56:21 -07002314static int dev_ifconf(struct net *net, char __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315{
2316 struct ifconf ifc;
2317 struct net_device *dev;
2318 char __user *pos;
2319 int len;
2320 int total;
2321 int i;
2322
2323 /*
2324 * Fetch the caller's info block.
2325 */
2326
2327 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2328 return -EFAULT;
2329
2330 pos = ifc.ifc_buf;
2331 len = ifc.ifc_len;
2332
2333 /*
2334 * Loop over the interfaces, and write an info block for each.
2335 */
2336
2337 total = 0;
Eric W. Biederman881d9662007-09-17 11:56:21 -07002338 for_each_netdev(net, dev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 for (i = 0; i < NPROTO; i++) {
2340 if (gifconf_list[i]) {
2341 int done;
2342 if (!pos)
2343 done = gifconf_list[i](dev, NULL, 0);
2344 else
2345 done = gifconf_list[i](dev, pos + total,
2346 len - total);
2347 if (done < 0)
2348 return -EFAULT;
2349 total += done;
2350 }
2351 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
2354 /*
2355 * All done. Write the updated control block back to the caller.
2356 */
2357 ifc.ifc_len = total;
2358
2359 /*
2360 * Both BSD and Solaris return 0 here, so we do too.
2361 */
2362 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2363}
2364
2365#ifdef CONFIG_PROC_FS
2366/*
2367 * This is invoked by the /proc filesystem handler to display a device
2368 * in detail.
2369 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2371{
Denis V. Luneve372c412007-11-19 22:31:54 -08002372 struct net *net = seq_file_net(seq);
Pavel Emelianov7562f872007-05-03 15:13:45 -07002373 loff_t off;
2374 struct net_device *dev;
2375
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 read_lock(&dev_base_lock);
Pavel Emelianov7562f872007-05-03 15:13:45 -07002377 if (!*pos)
2378 return SEQ_START_TOKEN;
2379
2380 off = 1;
Eric W. Biederman881d9662007-09-17 11:56:21 -07002381 for_each_netdev(net, dev)
Pavel Emelianov7562f872007-05-03 15:13:45 -07002382 if (off++ == *pos)
2383 return dev;
2384
2385 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386}
2387
2388void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2389{
Denis V. Luneve372c412007-11-19 22:31:54 -08002390 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 ++*pos;
Pavel Emelianov7562f872007-05-03 15:13:45 -07002392 return v == SEQ_START_TOKEN ?
Eric W. Biederman881d9662007-09-17 11:56:21 -07002393 first_net_device(net) : next_net_device((struct net_device *)v);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394}
2395
2396void dev_seq_stop(struct seq_file *seq, void *v)
2397{
2398 read_unlock(&dev_base_lock);
2399}
2400
2401static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2402{
Rusty Russellc45d2862007-03-28 14:29:08 -07002403 struct net_device_stats *stats = dev->get_stats(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404
Rusty Russell5a1b5892007-04-28 21:04:03 -07002405 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2406 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2407 dev->name, stats->rx_bytes, stats->rx_packets,
2408 stats->rx_errors,
2409 stats->rx_dropped + stats->rx_missed_errors,
2410 stats->rx_fifo_errors,
2411 stats->rx_length_errors + stats->rx_over_errors +
2412 stats->rx_crc_errors + stats->rx_frame_errors,
2413 stats->rx_compressed, stats->multicast,
2414 stats->tx_bytes, stats->tx_packets,
2415 stats->tx_errors, stats->tx_dropped,
2416 stats->tx_fifo_errors, stats->collisions,
2417 stats->tx_carrier_errors +
2418 stats->tx_aborted_errors +
2419 stats->tx_window_errors +
2420 stats->tx_heartbeat_errors,
2421 stats->tx_compressed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422}
2423
2424/*
2425 * Called from the PROCfs module. This now uses the new arbitrary sized
2426 * /proc/net interface to create /proc/net/dev
2427 */
2428static int dev_seq_show(struct seq_file *seq, void *v)
2429{
2430 if (v == SEQ_START_TOKEN)
2431 seq_puts(seq, "Inter-| Receive "
2432 " | Transmit\n"
2433 " face |bytes packets errs drop fifo frame "
2434 "compressed multicast|bytes packets errs "
2435 "drop fifo colls carrier compressed\n");
2436 else
2437 dev_seq_printf_stats(seq, v);
2438 return 0;
2439}
2440
2441static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2442{
2443 struct netif_rx_stats *rc = NULL;
2444
2445 while (*pos < NR_CPUS)
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002446 if (cpu_online(*pos)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 rc = &per_cpu(netdev_rx_stat, *pos);
2448 break;
2449 } else
2450 ++*pos;
2451 return rc;
2452}
2453
2454static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2455{
2456 return softnet_get_online(pos);
2457}
2458
2459static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2460{
2461 ++*pos;
2462 return softnet_get_online(pos);
2463}
2464
2465static void softnet_seq_stop(struct seq_file *seq, void *v)
2466{
2467}
2468
2469static int softnet_seq_show(struct seq_file *seq, void *v)
2470{
2471 struct netif_rx_stats *s = v;
2472
2473 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
Stephen Hemminger31aa02c2005-06-23 20:12:48 -07002474 s->total, s->dropped, s->time_squeeze, 0,
Stephen Hemmingerc1ebcdb2005-06-23 20:08:59 -07002475 0, 0, 0, 0, /* was fastroute */
2476 s->cpu_collision );
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 return 0;
2478}
2479
Stephen Hemmingerf6908082007-03-12 14:34:29 -07002480static const struct seq_operations dev_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 .start = dev_seq_start,
2482 .next = dev_seq_next,
2483 .stop = dev_seq_stop,
2484 .show = dev_seq_show,
2485};
2486
2487static int dev_seq_open(struct inode *inode, struct file *file)
2488{
Denis V. Luneve372c412007-11-19 22:31:54 -08002489 return seq_open_net(inode, file, &dev_seq_ops,
2490 sizeof(struct seq_net_private));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491}
2492
Arjan van de Ven9a321442007-02-12 00:55:35 -08002493static const struct file_operations dev_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 .owner = THIS_MODULE,
2495 .open = dev_seq_open,
2496 .read = seq_read,
2497 .llseek = seq_lseek,
Denis V. Luneve372c412007-11-19 22:31:54 -08002498 .release = seq_release_net,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499};
2500
Stephen Hemmingerf6908082007-03-12 14:34:29 -07002501static const struct seq_operations softnet_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 .start = softnet_seq_start,
2503 .next = softnet_seq_next,
2504 .stop = softnet_seq_stop,
2505 .show = softnet_seq_show,
2506};
2507
2508static int softnet_seq_open(struct inode *inode, struct file *file)
2509{
2510 return seq_open(file, &softnet_seq_ops);
2511}
2512
Arjan van de Ven9a321442007-02-12 00:55:35 -08002513static const struct file_operations softnet_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 .owner = THIS_MODULE,
2515 .open = softnet_seq_open,
2516 .read = seq_read,
2517 .llseek = seq_lseek,
2518 .release = seq_release,
2519};
2520
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002521static void *ptype_get_idx(loff_t pos)
2522{
2523 struct packet_type *pt = NULL;
2524 loff_t i = 0;
2525 int t;
2526
2527 list_for_each_entry_rcu(pt, &ptype_all, list) {
2528 if (i == pos)
2529 return pt;
2530 ++i;
2531 }
2532
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08002533 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002534 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2535 if (i == pos)
2536 return pt;
2537 ++i;
2538 }
2539 }
2540 return NULL;
2541}
2542
2543static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2544{
2545 rcu_read_lock();
2546 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2547}
2548
2549static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2550{
2551 struct packet_type *pt;
2552 struct list_head *nxt;
2553 int hash;
2554
2555 ++*pos;
2556 if (v == SEQ_START_TOKEN)
2557 return ptype_get_idx(0);
2558
2559 pt = v;
2560 nxt = pt->list.next;
2561 if (pt->type == htons(ETH_P_ALL)) {
2562 if (nxt != &ptype_all)
2563 goto found;
2564 hash = 0;
2565 nxt = ptype_base[0].next;
2566 } else
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08002567 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002568
2569 while (nxt == &ptype_base[hash]) {
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08002570 if (++hash >= PTYPE_HASH_SIZE)
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002571 return NULL;
2572 nxt = ptype_base[hash].next;
2573 }
2574found:
2575 return list_entry(nxt, struct packet_type, list);
2576}
2577
2578static void ptype_seq_stop(struct seq_file *seq, void *v)
2579{
2580 rcu_read_unlock();
2581}
2582
2583static void ptype_seq_decode(struct seq_file *seq, void *sym)
2584{
2585#ifdef CONFIG_KALLSYMS
2586 unsigned long offset = 0, symsize;
2587 const char *symname;
2588 char *modname;
2589 char namebuf[128];
2590
2591 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2592 &modname, namebuf);
2593
2594 if (symname) {
2595 char *delim = ":";
2596
2597 if (!modname)
2598 modname = delim = "";
2599 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2600 symname, offset);
2601 return;
2602 }
2603#endif
2604
2605 seq_printf(seq, "[%p]", sym);
2606}
2607
2608static int ptype_seq_show(struct seq_file *seq, void *v)
2609{
2610 struct packet_type *pt = v;
2611
2612 if (v == SEQ_START_TOKEN)
2613 seq_puts(seq, "Type Device Function\n");
2614 else {
2615 if (pt->type == htons(ETH_P_ALL))
2616 seq_puts(seq, "ALL ");
2617 else
2618 seq_printf(seq, "%04x", ntohs(pt->type));
2619
2620 seq_printf(seq, " %-8s ",
2621 pt->dev ? pt->dev->name : "");
2622 ptype_seq_decode(seq, pt->func);
2623 seq_putc(seq, '\n');
2624 }
2625
2626 return 0;
2627}
2628
2629static const struct seq_operations ptype_seq_ops = {
2630 .start = ptype_seq_start,
2631 .next = ptype_seq_next,
2632 .stop = ptype_seq_stop,
2633 .show = ptype_seq_show,
2634};
2635
2636static int ptype_seq_open(struct inode *inode, struct file *file)
2637{
2638 return seq_open(file, &ptype_seq_ops);
2639}
2640
2641static const struct file_operations ptype_seq_fops = {
2642 .owner = THIS_MODULE,
2643 .open = ptype_seq_open,
2644 .read = seq_read,
2645 .llseek = seq_lseek,
2646 .release = seq_release,
2647};
2648
2649
Pavel Emelyanov46650792007-10-08 20:38:39 -07002650static int __net_init dev_proc_net_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651{
2652 int rc = -ENOMEM;
2653
Eric W. Biederman881d9662007-09-17 11:56:21 -07002654 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 goto out;
Eric W. Biederman881d9662007-09-17 11:56:21 -07002656 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 goto out_dev;
Eric W. Biederman881d9662007-09-17 11:56:21 -07002658 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
Eric W. Biederman457c4cb2007-09-12 12:01:34 +02002659 goto out_softnet;
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002660
Eric W. Biederman881d9662007-09-17 11:56:21 -07002661 if (wext_proc_init(net))
Eric W. Biederman457c4cb2007-09-12 12:01:34 +02002662 goto out_ptype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 rc = 0;
2664out:
2665 return rc;
Eric W. Biederman457c4cb2007-09-12 12:01:34 +02002666out_ptype:
Eric W. Biederman881d9662007-09-17 11:56:21 -07002667 proc_net_remove(net, "ptype");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668out_softnet:
Eric W. Biederman881d9662007-09-17 11:56:21 -07002669 proc_net_remove(net, "softnet_stat");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670out_dev:
Eric W. Biederman881d9662007-09-17 11:56:21 -07002671 proc_net_remove(net, "dev");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 goto out;
2673}
Eric W. Biederman881d9662007-09-17 11:56:21 -07002674
Pavel Emelyanov46650792007-10-08 20:38:39 -07002675static void __net_exit dev_proc_net_exit(struct net *net)
Eric W. Biederman881d9662007-09-17 11:56:21 -07002676{
2677 wext_proc_exit(net);
2678
2679 proc_net_remove(net, "ptype");
2680 proc_net_remove(net, "softnet_stat");
2681 proc_net_remove(net, "dev");
2682}
2683
Denis V. Lunev022cbae2007-11-13 03:23:50 -08002684static struct pernet_operations __net_initdata dev_proc_ops = {
Eric W. Biederman881d9662007-09-17 11:56:21 -07002685 .init = dev_proc_net_init,
2686 .exit = dev_proc_net_exit,
2687};
2688
2689static int __init dev_proc_init(void)
2690{
2691 return register_pernet_subsys(&dev_proc_ops);
2692}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693#else
2694#define dev_proc_init() 0
2695#endif /* CONFIG_PROC_FS */
2696
2697
2698/**
2699 * netdev_set_master - set up master/slave pair
2700 * @slave: slave device
2701 * @master: new master device
2702 *
2703 * Changes the master device of the slave. Pass %NULL to break the
2704 * bonding. The caller must hold the RTNL semaphore. On a failure
2705 * a negative errno code is returned. On success the reference counts
2706 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2707 * function returns zero.
2708 */
2709int netdev_set_master(struct net_device *slave, struct net_device *master)
2710{
2711 struct net_device *old = slave->master;
2712
2713 ASSERT_RTNL();
2714
2715 if (master) {
2716 if (old)
2717 return -EBUSY;
2718 dev_hold(master);
2719 }
2720
2721 slave->master = master;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002722
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 synchronize_net();
2724
2725 if (old)
2726 dev_put(old);
2727
2728 if (master)
2729 slave->flags |= IFF_SLAVE;
2730 else
2731 slave->flags &= ~IFF_SLAVE;
2732
2733 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2734 return 0;
2735}
2736
Patrick McHardy4417da62007-06-27 01:28:10 -07002737static void __dev_set_promiscuity(struct net_device *dev, int inc)
2738{
2739 unsigned short old_flags = dev->flags;
2740
Patrick McHardy24023452007-07-14 18:51:31 -07002741 ASSERT_RTNL();
2742
Patrick McHardy4417da62007-06-27 01:28:10 -07002743 if ((dev->promiscuity += inc) == 0)
2744 dev->flags &= ~IFF_PROMISC;
2745 else
2746 dev->flags |= IFF_PROMISC;
2747 if (dev->flags != old_flags) {
2748 printk(KERN_INFO "device %s %s promiscuous mode\n",
2749 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2750 "left");
2751 audit_log(current->audit_context, GFP_ATOMIC,
2752 AUDIT_ANOM_PROMISCUOUS,
2753 "dev=%s prom=%d old_prom=%d auid=%u",
2754 dev->name, (dev->flags & IFF_PROMISC),
2755 (old_flags & IFF_PROMISC),
2756 audit_get_loginuid(current->audit_context));
Patrick McHardy24023452007-07-14 18:51:31 -07002757
2758 if (dev->change_rx_flags)
2759 dev->change_rx_flags(dev, IFF_PROMISC);
Patrick McHardy4417da62007-06-27 01:28:10 -07002760 }
2761}
2762
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763/**
2764 * dev_set_promiscuity - update promiscuity count on a device
2765 * @dev: device
2766 * @inc: modifier
2767 *
Stephen Hemminger3041a062006-05-26 13:25:24 -07002768 * Add or remove promiscuity from a device. While the count in the device
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 * remains above zero the interface remains promiscuous. Once it hits zero
2770 * the device reverts back to normal filtering operation. A negative inc
2771 * value is used to drop promiscuity on the device.
2772 */
2773void dev_set_promiscuity(struct net_device *dev, int inc)
2774{
2775 unsigned short old_flags = dev->flags;
2776
Patrick McHardy4417da62007-06-27 01:28:10 -07002777 __dev_set_promiscuity(dev, inc);
2778 if (dev->flags != old_flags)
2779 dev_set_rx_mode(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780}
2781
2782/**
2783 * dev_set_allmulti - update allmulti count on a device
2784 * @dev: device
2785 * @inc: modifier
2786 *
2787 * Add or remove reception of all multicast frames to a device. While the
2788 * count in the device remains above zero the interface remains listening
2789 * to all interfaces. Once it hits zero the device reverts back to normal
2790 * filtering operation. A negative @inc value is used to drop the counter
2791 * when releasing a resource needing all multicasts.
2792 */
2793
2794void dev_set_allmulti(struct net_device *dev, int inc)
2795{
2796 unsigned short old_flags = dev->flags;
2797
Patrick McHardy24023452007-07-14 18:51:31 -07002798 ASSERT_RTNL();
2799
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 dev->flags |= IFF_ALLMULTI;
2801 if ((dev->allmulti += inc) == 0)
2802 dev->flags &= ~IFF_ALLMULTI;
Patrick McHardy24023452007-07-14 18:51:31 -07002803 if (dev->flags ^ old_flags) {
2804 if (dev->change_rx_flags)
2805 dev->change_rx_flags(dev, IFF_ALLMULTI);
Patrick McHardy4417da62007-06-27 01:28:10 -07002806 dev_set_rx_mode(dev);
Patrick McHardy24023452007-07-14 18:51:31 -07002807 }
Patrick McHardy4417da62007-06-27 01:28:10 -07002808}
2809
2810/*
2811 * Upload unicast and multicast address lists to device and
2812 * configure RX filtering. When the device doesn't support unicast
Joe Perches53ccaae2007-12-20 14:02:06 -08002813 * filtering it is put in promiscuous mode while unicast addresses
Patrick McHardy4417da62007-06-27 01:28:10 -07002814 * are present.
2815 */
2816void __dev_set_rx_mode(struct net_device *dev)
2817{
2818 /* dev_open will call this function so the list will stay sane. */
2819 if (!(dev->flags&IFF_UP))
2820 return;
2821
2822 if (!netif_device_present(dev))
YOSHIFUJI Hideaki40b77c92007-07-19 10:43:23 +09002823 return;
Patrick McHardy4417da62007-06-27 01:28:10 -07002824
2825 if (dev->set_rx_mode)
2826 dev->set_rx_mode(dev);
2827 else {
2828 /* Unicast addresses changes may only happen under the rtnl,
2829 * therefore calling __dev_set_promiscuity here is safe.
2830 */
2831 if (dev->uc_count > 0 && !dev->uc_promisc) {
2832 __dev_set_promiscuity(dev, 1);
2833 dev->uc_promisc = 1;
2834 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2835 __dev_set_promiscuity(dev, -1);
2836 dev->uc_promisc = 0;
2837 }
2838
2839 if (dev->set_multicast_list)
2840 dev->set_multicast_list(dev);
2841 }
2842}
2843
2844void dev_set_rx_mode(struct net_device *dev)
2845{
2846 netif_tx_lock_bh(dev);
2847 __dev_set_rx_mode(dev);
2848 netif_tx_unlock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849}
2850
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07002851int __dev_addr_delete(struct dev_addr_list **list, int *count,
2852 void *addr, int alen, int glbl)
Patrick McHardybf742482007-06-27 01:26:19 -07002853{
2854 struct dev_addr_list *da;
2855
2856 for (; (da = *list) != NULL; list = &da->next) {
2857 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2858 alen == da->da_addrlen) {
2859 if (glbl) {
2860 int old_glbl = da->da_gusers;
2861 da->da_gusers = 0;
2862 if (old_glbl == 0)
2863 break;
2864 }
2865 if (--da->da_users)
2866 return 0;
2867
2868 *list = da->next;
2869 kfree(da);
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07002870 (*count)--;
Patrick McHardybf742482007-06-27 01:26:19 -07002871 return 0;
2872 }
2873 }
2874 return -ENOENT;
2875}
2876
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07002877int __dev_addr_add(struct dev_addr_list **list, int *count,
2878 void *addr, int alen, int glbl)
Patrick McHardybf742482007-06-27 01:26:19 -07002879{
2880 struct dev_addr_list *da;
2881
2882 for (da = *list; da != NULL; da = da->next) {
2883 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2884 da->da_addrlen == alen) {
2885 if (glbl) {
2886 int old_glbl = da->da_gusers;
2887 da->da_gusers = 1;
2888 if (old_glbl)
2889 return 0;
2890 }
2891 da->da_users++;
2892 return 0;
2893 }
2894 }
2895
2896 da = kmalloc(sizeof(*da), GFP_ATOMIC);
2897 if (da == NULL)
2898 return -ENOMEM;
2899 memcpy(da->da_addr, addr, alen);
2900 da->da_addrlen = alen;
2901 da->da_users = 1;
2902 da->da_gusers = glbl ? 1 : 0;
2903 da->next = *list;
2904 *list = da;
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07002905 (*count)++;
Patrick McHardybf742482007-06-27 01:26:19 -07002906 return 0;
2907}
2908
Patrick McHardy4417da62007-06-27 01:28:10 -07002909/**
2910 * dev_unicast_delete - Release secondary unicast address.
2911 * @dev: device
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07002912 * @addr: address to delete
2913 * @alen: length of @addr
Patrick McHardy4417da62007-06-27 01:28:10 -07002914 *
2915 * Release reference to a secondary unicast address and remove it
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07002916 * from the device if the reference count drops to zero.
Patrick McHardy4417da62007-06-27 01:28:10 -07002917 *
2918 * The caller must hold the rtnl_mutex.
2919 */
2920int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
2921{
2922 int err;
2923
2924 ASSERT_RTNL();
2925
2926 netif_tx_lock_bh(dev);
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07002927 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2928 if (!err)
Patrick McHardy4417da62007-06-27 01:28:10 -07002929 __dev_set_rx_mode(dev);
Patrick McHardy4417da62007-06-27 01:28:10 -07002930 netif_tx_unlock_bh(dev);
2931 return err;
2932}
2933EXPORT_SYMBOL(dev_unicast_delete);
2934
2935/**
2936 * dev_unicast_add - add a secondary unicast address
2937 * @dev: device
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07002938 * @addr: address to delete
2939 * @alen: length of @addr
Patrick McHardy4417da62007-06-27 01:28:10 -07002940 *
2941 * Add a secondary unicast address to the device or increase
2942 * the reference count if it already exists.
2943 *
2944 * The caller must hold the rtnl_mutex.
2945 */
2946int dev_unicast_add(struct net_device *dev, void *addr, int alen)
2947{
2948 int err;
2949
2950 ASSERT_RTNL();
2951
2952 netif_tx_lock_bh(dev);
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07002953 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2954 if (!err)
Patrick McHardy4417da62007-06-27 01:28:10 -07002955 __dev_set_rx_mode(dev);
Patrick McHardy4417da62007-06-27 01:28:10 -07002956 netif_tx_unlock_bh(dev);
2957 return err;
2958}
2959EXPORT_SYMBOL(dev_unicast_add);
2960
Denis Cheng12972622007-07-18 02:12:56 -07002961static void __dev_addr_discard(struct dev_addr_list **list)
2962{
2963 struct dev_addr_list *tmp;
2964
2965 while (*list != NULL) {
2966 tmp = *list;
2967 *list = tmp->next;
2968 if (tmp->da_users > tmp->da_gusers)
2969 printk("__dev_addr_discard: address leakage! "
2970 "da_users=%d\n", tmp->da_users);
2971 kfree(tmp);
2972 }
2973}
2974
Denis Cheng26cc2522007-07-18 02:12:03 -07002975static void dev_addr_discard(struct net_device *dev)
Patrick McHardy4417da62007-06-27 01:28:10 -07002976{
2977 netif_tx_lock_bh(dev);
Denis Cheng26cc2522007-07-18 02:12:03 -07002978
Patrick McHardy4417da62007-06-27 01:28:10 -07002979 __dev_addr_discard(&dev->uc_list);
2980 dev->uc_count = 0;
Patrick McHardy4417da62007-06-27 01:28:10 -07002981
Denis Cheng456ad752007-07-18 02:10:54 -07002982 __dev_addr_discard(&dev->mc_list);
2983 dev->mc_count = 0;
Denis Cheng26cc2522007-07-18 02:12:03 -07002984
Denis Cheng456ad752007-07-18 02:10:54 -07002985 netif_tx_unlock_bh(dev);
2986}
2987
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988unsigned dev_get_flags(const struct net_device *dev)
2989{
2990 unsigned flags;
2991
2992 flags = (dev->flags & ~(IFF_PROMISC |
2993 IFF_ALLMULTI |
Stefan Rompfb00055a2006-03-20 17:09:11 -08002994 IFF_RUNNING |
2995 IFF_LOWER_UP |
2996 IFF_DORMANT)) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 (dev->gflags & (IFF_PROMISC |
2998 IFF_ALLMULTI));
2999
Stefan Rompfb00055a2006-03-20 17:09:11 -08003000 if (netif_running(dev)) {
3001 if (netif_oper_up(dev))
3002 flags |= IFF_RUNNING;
3003 if (netif_carrier_ok(dev))
3004 flags |= IFF_LOWER_UP;
3005 if (netif_dormant(dev))
3006 flags |= IFF_DORMANT;
3007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008
3009 return flags;
3010}
3011
3012int dev_change_flags(struct net_device *dev, unsigned flags)
3013{
Thomas Graf7c355f52007-06-05 16:03:03 -07003014 int ret, changes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 int old_flags = dev->flags;
3016
Patrick McHardy24023452007-07-14 18:51:31 -07003017 ASSERT_RTNL();
3018
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 /*
3020 * Set the flags on our device.
3021 */
3022
3023 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3024 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3025 IFF_AUTOMEDIA)) |
3026 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3027 IFF_ALLMULTI));
3028
3029 /*
3030 * Load in the correct multicast list now the flags have changed.
3031 */
3032
Patrick McHardy24023452007-07-14 18:51:31 -07003033 if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST)
3034 dev->change_rx_flags(dev, IFF_MULTICAST);
3035
Patrick McHardy4417da62007-06-27 01:28:10 -07003036 dev_set_rx_mode(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037
3038 /*
3039 * Have we downed the interface. We handle IFF_UP ourselves
3040 * according to user attempts to set it, rather than blindly
3041 * setting it.
3042 */
3043
3044 ret = 0;
3045 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3046 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3047
3048 if (!ret)
Patrick McHardy4417da62007-06-27 01:28:10 -07003049 dev_set_rx_mode(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 }
3051
3052 if (dev->flags & IFF_UP &&
3053 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3054 IFF_VOLATILE)))
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003055 call_netdevice_notifiers(NETDEV_CHANGE, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056
3057 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3058 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3059 dev->gflags ^= IFF_PROMISC;
3060 dev_set_promiscuity(dev, inc);
3061 }
3062
3063 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3064 is important. Some (broken) drivers set IFF_PROMISC, when
3065 IFF_ALLMULTI is requested not asking us and not reporting.
3066 */
3067 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3068 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3069 dev->gflags ^= IFF_ALLMULTI;
3070 dev_set_allmulti(dev, inc);
3071 }
3072
Thomas Graf7c355f52007-06-05 16:03:03 -07003073 /* Exclude state transition flags, already notified */
3074 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3075 if (changes)
3076 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077
3078 return ret;
3079}
3080
3081int dev_set_mtu(struct net_device *dev, int new_mtu)
3082{
3083 int err;
3084
3085 if (new_mtu == dev->mtu)
3086 return 0;
3087
3088 /* MTU must be positive. */
3089 if (new_mtu < 0)
3090 return -EINVAL;
3091
3092 if (!netif_device_present(dev))
3093 return -ENODEV;
3094
3095 err = 0;
3096 if (dev->change_mtu)
3097 err = dev->change_mtu(dev, new_mtu);
3098 else
3099 dev->mtu = new_mtu;
3100 if (!err && dev->flags & IFF_UP)
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003101 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 return err;
3103}
3104
3105int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3106{
3107 int err;
3108
3109 if (!dev->set_mac_address)
3110 return -EOPNOTSUPP;
3111 if (sa->sa_family != dev->type)
3112 return -EINVAL;
3113 if (!netif_device_present(dev))
3114 return -ENODEV;
3115 err = dev->set_mac_address(dev, sa);
3116 if (!err)
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003117 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 return err;
3119}
3120
3121/*
Jeff Garzik14e3e072007-10-08 00:06:32 -07003122 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 */
Jeff Garzik14e3e072007-10-08 00:06:32 -07003124static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125{
3126 int err;
Eric W. Biederman881d9662007-09-17 11:56:21 -07003127 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128
3129 if (!dev)
3130 return -ENODEV;
3131
3132 switch (cmd) {
3133 case SIOCGIFFLAGS: /* Get interface flags */
3134 ifr->ifr_flags = dev_get_flags(dev);
3135 return 0;
3136
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137 case SIOCGIFMETRIC: /* Get the metric on the interface
3138 (currently unused) */
3139 ifr->ifr_metric = 0;
3140 return 0;
3141
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 case SIOCGIFMTU: /* Get the MTU of a device */
3143 ifr->ifr_mtu = dev->mtu;
3144 return 0;
3145
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 case SIOCGIFHWADDR:
3147 if (!dev->addr_len)
3148 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3149 else
3150 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3151 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3152 ifr->ifr_hwaddr.sa_family = dev->type;
3153 return 0;
3154
Jeff Garzik14e3e072007-10-08 00:06:32 -07003155 case SIOCGIFSLAVE:
3156 err = -EINVAL;
3157 break;
3158
3159 case SIOCGIFMAP:
3160 ifr->ifr_map.mem_start = dev->mem_start;
3161 ifr->ifr_map.mem_end = dev->mem_end;
3162 ifr->ifr_map.base_addr = dev->base_addr;
3163 ifr->ifr_map.irq = dev->irq;
3164 ifr->ifr_map.dma = dev->dma;
3165 ifr->ifr_map.port = dev->if_port;
3166 return 0;
3167
3168 case SIOCGIFINDEX:
3169 ifr->ifr_ifindex = dev->ifindex;
3170 return 0;
3171
3172 case SIOCGIFTXQLEN:
3173 ifr->ifr_qlen = dev->tx_queue_len;
3174 return 0;
3175
3176 default:
3177 /* dev_ioctl() should ensure this case
3178 * is never reached
3179 */
3180 WARN_ON(1);
3181 err = -EINVAL;
3182 break;
3183
3184 }
3185 return err;
3186}
3187
3188/*
3189 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3190 */
3191static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3192{
3193 int err;
3194 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3195
3196 if (!dev)
3197 return -ENODEV;
3198
3199 switch (cmd) {
3200 case SIOCSIFFLAGS: /* Set interface flags */
3201 return dev_change_flags(dev, ifr->ifr_flags);
3202
3203 case SIOCSIFMETRIC: /* Set the metric on the interface
3204 (currently unused) */
3205 return -EOPNOTSUPP;
3206
3207 case SIOCSIFMTU: /* Set the MTU of a device */
3208 return dev_set_mtu(dev, ifr->ifr_mtu);
3209
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 case SIOCSIFHWADDR:
3211 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3212
3213 case SIOCSIFHWBROADCAST:
3214 if (ifr->ifr_hwaddr.sa_family != dev->type)
3215 return -EINVAL;
3216 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3217 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003218 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219 return 0;
3220
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221 case SIOCSIFMAP:
3222 if (dev->set_config) {
3223 if (!netif_device_present(dev))
3224 return -ENODEV;
3225 return dev->set_config(dev, &ifr->ifr_map);
3226 }
3227 return -EOPNOTSUPP;
3228
3229 case SIOCADDMULTI:
3230 if (!dev->set_multicast_list ||
3231 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3232 return -EINVAL;
3233 if (!netif_device_present(dev))
3234 return -ENODEV;
3235 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3236 dev->addr_len, 1);
3237
3238 case SIOCDELMULTI:
3239 if (!dev->set_multicast_list ||
3240 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3241 return -EINVAL;
3242 if (!netif_device_present(dev))
3243 return -ENODEV;
3244 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3245 dev->addr_len, 1);
3246
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247 case SIOCSIFTXQLEN:
3248 if (ifr->ifr_qlen < 0)
3249 return -EINVAL;
3250 dev->tx_queue_len = ifr->ifr_qlen;
3251 return 0;
3252
3253 case SIOCSIFNAME:
3254 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3255 return dev_change_name(dev, ifr->ifr_newname);
3256
3257 /*
3258 * Unknown or private ioctl
3259 */
3260
3261 default:
3262 if ((cmd >= SIOCDEVPRIVATE &&
3263 cmd <= SIOCDEVPRIVATE + 15) ||
3264 cmd == SIOCBONDENSLAVE ||
3265 cmd == SIOCBONDRELEASE ||
3266 cmd == SIOCBONDSETHWADDR ||
3267 cmd == SIOCBONDSLAVEINFOQUERY ||
3268 cmd == SIOCBONDINFOQUERY ||
3269 cmd == SIOCBONDCHANGEACTIVE ||
3270 cmd == SIOCGMIIPHY ||
3271 cmd == SIOCGMIIREG ||
3272 cmd == SIOCSMIIREG ||
3273 cmd == SIOCBRADDIF ||
3274 cmd == SIOCBRDELIF ||
3275 cmd == SIOCWANDEV) {
3276 err = -EOPNOTSUPP;
3277 if (dev->do_ioctl) {
3278 if (netif_device_present(dev))
3279 err = dev->do_ioctl(dev, ifr,
3280 cmd);
3281 else
3282 err = -ENODEV;
3283 }
3284 } else
3285 err = -EINVAL;
3286
3287 }
3288 return err;
3289}
3290
3291/*
3292 * This function handles all "interface"-type I/O control requests. The actual
3293 * 'doing' part of this is dev_ifsioc above.
3294 */
3295
3296/**
3297 * dev_ioctl - network device ioctl
Randy Dunlapc4ea43c2007-10-12 21:17:49 -07003298 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299 * @cmd: command to issue
3300 * @arg: pointer to a struct ifreq in user space
3301 *
3302 * Issue ioctl functions to devices. This is normally called by the
3303 * user space syscall interfaces but can sometimes be useful for
3304 * other purposes. The return value is the return from the syscall if
3305 * positive or a negative errno code on error.
3306 */
3307
Eric W. Biederman881d9662007-09-17 11:56:21 -07003308int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309{
3310 struct ifreq ifr;
3311 int ret;
3312 char *colon;
3313
3314 /* One special case: SIOCGIFCONF takes ifconf argument
3315 and requires shared lock, because it sleeps writing
3316 to user space.
3317 */
3318
3319 if (cmd == SIOCGIFCONF) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003320 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003321 ret = dev_ifconf(net, (char __user *) arg);
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003322 rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 return ret;
3324 }
3325 if (cmd == SIOCGIFNAME)
Eric W. Biederman881d9662007-09-17 11:56:21 -07003326 return dev_ifname(net, (struct ifreq __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327
3328 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3329 return -EFAULT;
3330
3331 ifr.ifr_name[IFNAMSIZ-1] = 0;
3332
3333 colon = strchr(ifr.ifr_name, ':');
3334 if (colon)
3335 *colon = 0;
3336
3337 /*
3338 * See which interface the caller is talking about.
3339 */
3340
3341 switch (cmd) {
3342 /*
3343 * These ioctl calls:
3344 * - can be done by all.
3345 * - atomic and do not require locking.
3346 * - return a value
3347 */
3348 case SIOCGIFFLAGS:
3349 case SIOCGIFMETRIC:
3350 case SIOCGIFMTU:
3351 case SIOCGIFHWADDR:
3352 case SIOCGIFSLAVE:
3353 case SIOCGIFMAP:
3354 case SIOCGIFINDEX:
3355 case SIOCGIFTXQLEN:
Eric W. Biederman881d9662007-09-17 11:56:21 -07003356 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357 read_lock(&dev_base_lock);
Jeff Garzik14e3e072007-10-08 00:06:32 -07003358 ret = dev_ifsioc_locked(net, &ifr, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359 read_unlock(&dev_base_lock);
3360 if (!ret) {
3361 if (colon)
3362 *colon = ':';
3363 if (copy_to_user(arg, &ifr,
3364 sizeof(struct ifreq)))
3365 ret = -EFAULT;
3366 }
3367 return ret;
3368
3369 case SIOCETHTOOL:
Eric W. Biederman881d9662007-09-17 11:56:21 -07003370 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003372 ret = dev_ethtool(net, &ifr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373 rtnl_unlock();
3374 if (!ret) {
3375 if (colon)
3376 *colon = ':';
3377 if (copy_to_user(arg, &ifr,
3378 sizeof(struct ifreq)))
3379 ret = -EFAULT;
3380 }
3381 return ret;
3382
3383 /*
3384 * These ioctl calls:
3385 * - require superuser power.
3386 * - require strict serialization.
3387 * - return a value
3388 */
3389 case SIOCGMIIPHY:
3390 case SIOCGMIIREG:
3391 case SIOCSIFNAME:
3392 if (!capable(CAP_NET_ADMIN))
3393 return -EPERM;
Eric W. Biederman881d9662007-09-17 11:56:21 -07003394 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003396 ret = dev_ifsioc(net, &ifr, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397 rtnl_unlock();
3398 if (!ret) {
3399 if (colon)
3400 *colon = ':';
3401 if (copy_to_user(arg, &ifr,
3402 sizeof(struct ifreq)))
3403 ret = -EFAULT;
3404 }
3405 return ret;
3406
3407 /*
3408 * These ioctl calls:
3409 * - require superuser power.
3410 * - require strict serialization.
3411 * - do not return a value
3412 */
3413 case SIOCSIFFLAGS:
3414 case SIOCSIFMETRIC:
3415 case SIOCSIFMTU:
3416 case SIOCSIFMAP:
3417 case SIOCSIFHWADDR:
3418 case SIOCSIFSLAVE:
3419 case SIOCADDMULTI:
3420 case SIOCDELMULTI:
3421 case SIOCSIFHWBROADCAST:
3422 case SIOCSIFTXQLEN:
3423 case SIOCSMIIREG:
3424 case SIOCBONDENSLAVE:
3425 case SIOCBONDRELEASE:
3426 case SIOCBONDSETHWADDR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 case SIOCBONDCHANGEACTIVE:
3428 case SIOCBRADDIF:
3429 case SIOCBRDELIF:
3430 if (!capable(CAP_NET_ADMIN))
3431 return -EPERM;
Thomas Grafcabcac02006-01-24 12:46:33 -08003432 /* fall through */
3433 case SIOCBONDSLAVEINFOQUERY:
3434 case SIOCBONDINFOQUERY:
Eric W. Biederman881d9662007-09-17 11:56:21 -07003435 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003437 ret = dev_ifsioc(net, &ifr, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 rtnl_unlock();
3439 return ret;
3440
3441 case SIOCGIFMEM:
3442 /* Get the per device memory space. We can add this but
3443 * currently do not support it */
3444 case SIOCSIFMEM:
3445 /* Set the per device memory buffer space.
3446 * Not applicable in our case */
3447 case SIOCSIFLINK:
3448 return -EINVAL;
3449
3450 /*
3451 * Unknown or private ioctl.
3452 */
3453 default:
3454 if (cmd == SIOCWANDEV ||
3455 (cmd >= SIOCDEVPRIVATE &&
3456 cmd <= SIOCDEVPRIVATE + 15)) {
Eric W. Biederman881d9662007-09-17 11:56:21 -07003457 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003459 ret = dev_ifsioc(net, &ifr, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 rtnl_unlock();
3461 if (!ret && copy_to_user(arg, &ifr,
3462 sizeof(struct ifreq)))
3463 ret = -EFAULT;
3464 return ret;
3465 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466 /* Take care of Wireless Extensions */
Johannes Berg295f4a12007-04-26 20:43:56 -07003467 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
Eric W. Biederman881d9662007-09-17 11:56:21 -07003468 return wext_handle_ioctl(net, &ifr, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469 return -EINVAL;
3470 }
3471}
3472
3473
3474/**
3475 * dev_new_index - allocate an ifindex
Randy Dunlapc4ea43c2007-10-12 21:17:49 -07003476 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477 *
3478 * Returns a suitable unique value for a new device interface
3479 * number. The caller must hold the rtnl semaphore or the
3480 * dev_base_lock to be sure it remains unique.
3481 */
Eric W. Biederman881d9662007-09-17 11:56:21 -07003482static int dev_new_index(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483{
3484 static int ifindex;
3485 for (;;) {
3486 if (++ifindex <= 0)
3487 ifindex = 1;
Eric W. Biederman881d9662007-09-17 11:56:21 -07003488 if (!__dev_get_by_index(net, ifindex))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 return ifindex;
3490 }
3491}
3492
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493/* Delayed registration/unregisteration */
3494static DEFINE_SPINLOCK(net_todo_list_lock);
Denis Cheng3b5b34f2007-12-07 00:49:17 -08003495static LIST_HEAD(net_todo_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496
Stephen Hemminger6f05f622007-03-08 20:46:03 -08003497static void net_set_todo(struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498{
3499 spin_lock(&net_todo_list_lock);
3500 list_add_tail(&dev->todo_list, &net_todo_list);
3501 spin_unlock(&net_todo_list_lock);
3502}
3503
Daniel Lezcano93ee31f2007-10-30 15:38:18 -07003504static void rollback_registered(struct net_device *dev)
3505{
3506 BUG_ON(dev_boot_phase);
3507 ASSERT_RTNL();
3508
3509 /* Some devices call without registering for initialization unwind. */
3510 if (dev->reg_state == NETREG_UNINITIALIZED) {
3511 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3512 "was registered\n", dev->name, dev);
3513
3514 WARN_ON(1);
3515 return;
3516 }
3517
3518 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3519
3520 /* If device is running, close it first. */
3521 dev_close(dev);
3522
3523 /* And unlink it from device chain. */
3524 unlist_netdevice(dev);
3525
3526 dev->reg_state = NETREG_UNREGISTERING;
3527
3528 synchronize_net();
3529
3530 /* Shutdown queueing discipline. */
3531 dev_shutdown(dev);
3532
3533
3534 /* Notify protocols, that we are about to destroy
3535 this device. They should clean all the things.
3536 */
3537 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3538
3539 /*
3540 * Flush the unicast and multicast chains
3541 */
3542 dev_addr_discard(dev);
3543
3544 if (dev->uninit)
3545 dev->uninit(dev);
3546
3547 /* Notifier chain MUST detach us from master device. */
3548 BUG_TRAP(!dev->master);
3549
3550 /* Remove entries from kobject tree */
3551 netdev_unregister_kobject(dev);
3552
3553 synchronize_net();
3554
3555 dev_put(dev);
3556}
3557
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558/**
3559 * register_netdevice - register a network device
3560 * @dev: device to register
3561 *
3562 * Take a completed network device structure and add it to the kernel
3563 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3564 * chain. 0 is returned on success. A negative errno code is returned
3565 * on a failure to set up the device, or if the name is a duplicate.
3566 *
3567 * Callers must hold the rtnl semaphore. You may want
3568 * register_netdev() instead of this.
3569 *
3570 * BUGS:
3571 * The locking appears insufficient to guarantee two parallel registers
3572 * will not get the same name.
3573 */
3574
3575int register_netdevice(struct net_device *dev)
3576{
3577 struct hlist_head *head;
3578 struct hlist_node *p;
3579 int ret;
Eric W. Biederman881d9662007-09-17 11:56:21 -07003580 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581
3582 BUG_ON(dev_boot_phase);
3583 ASSERT_RTNL();
3584
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003585 might_sleep();
3586
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587 /* When net_device's are persistent, this will be fatal. */
3588 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
Eric W. Biederman881d9662007-09-17 11:56:21 -07003589 BUG_ON(!dev->nd_net);
3590 net = dev->nd_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591
3592 spin_lock_init(&dev->queue_lock);
Herbert Xu932ff272006-06-09 12:20:56 -07003593 spin_lock_init(&dev->_xmit_lock);
Jarek Poplawski723e98b2007-05-15 22:46:18 -07003594 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595 dev->xmit_lock_owner = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003596 spin_lock_init(&dev->ingress_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598 dev->iflink = -1;
3599
3600 /* Init, if this function is available */
3601 if (dev->init) {
3602 ret = dev->init(dev);
3603 if (ret) {
3604 if (ret > 0)
3605 ret = -EIO;
Adrian Bunk90833aa2006-11-13 16:02:22 -08003606 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607 }
3608 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003609
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610 if (!dev_valid_name(dev->name)) {
3611 ret = -EINVAL;
Herbert Xu7ce1b0e2007-07-30 16:29:40 -07003612 goto err_uninit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613 }
3614
Eric W. Biederman881d9662007-09-17 11:56:21 -07003615 dev->ifindex = dev_new_index(net);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616 if (dev->iflink == -1)
3617 dev->iflink = dev->ifindex;
3618
3619 /* Check for existence of name */
Eric W. Biederman881d9662007-09-17 11:56:21 -07003620 head = dev_name_hash(net, dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 hlist_for_each(p, head) {
3622 struct net_device *d
3623 = hlist_entry(p, struct net_device, name_hlist);
3624 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3625 ret = -EEXIST;
Herbert Xu7ce1b0e2007-07-30 16:29:40 -07003626 goto err_uninit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629
Stephen Hemmingerd212f872007-06-27 00:47:37 -07003630 /* Fix illegal checksum combinations */
3631 if ((dev->features & NETIF_F_HW_CSUM) &&
3632 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3633 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3634 dev->name);
3635 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3636 }
3637
3638 if ((dev->features & NETIF_F_NO_CSUM) &&
3639 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3640 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3641 dev->name);
3642 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3643 }
3644
3645
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 /* Fix illegal SG+CSUM combinations. */
3647 if ((dev->features & NETIF_F_SG) &&
Herbert Xu8648b302006-06-17 22:06:05 -07003648 !(dev->features & NETIF_F_ALL_CSUM)) {
Stephen Hemminger5a8da022006-07-07 16:54:05 -07003649 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650 dev->name);
3651 dev->features &= ~NETIF_F_SG;
3652 }
3653
3654 /* TSO requires that SG is present as well. */
3655 if ((dev->features & NETIF_F_TSO) &&
3656 !(dev->features & NETIF_F_SG)) {
Stephen Hemminger5a8da022006-07-07 16:54:05 -07003657 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658 dev->name);
3659 dev->features &= ~NETIF_F_TSO;
3660 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07003661 if (dev->features & NETIF_F_UFO) {
3662 if (!(dev->features & NETIF_F_HW_CSUM)) {
3663 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3664 "NETIF_F_HW_CSUM feature.\n",
3665 dev->name);
3666 dev->features &= ~NETIF_F_UFO;
3667 }
3668 if (!(dev->features & NETIF_F_SG)) {
3669 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3670 "NETIF_F_SG feature.\n",
3671 dev->name);
3672 dev->features &= ~NETIF_F_UFO;
3673 }
3674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675
Eric W. Biederman8b41d182007-09-26 22:02:53 -07003676 ret = netdev_register_kobject(dev);
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003677 if (ret)
Herbert Xu7ce1b0e2007-07-30 16:29:40 -07003678 goto err_uninit;
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003679 dev->reg_state = NETREG_REGISTERED;
3680
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681 /*
3682 * Default initial state at registry is that the
3683 * device is present.
3684 */
3685
3686 set_bit(__LINK_STATE_PRESENT, &dev->state);
3687
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 dev_init_scheduler(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 dev_hold(dev);
Eric W. Biedermance286d32007-09-12 13:53:49 +02003690 list_netdevice(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691
3692 /* Notify protocols, that a new device appeared. */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003693 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
Herbert Xufcc5a032007-07-30 17:03:38 -07003694 ret = notifier_to_errno(ret);
Daniel Lezcano93ee31f2007-10-30 15:38:18 -07003695 if (ret) {
3696 rollback_registered(dev);
3697 dev->reg_state = NETREG_UNREGISTERED;
3698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699
3700out:
3701 return ret;
Herbert Xu7ce1b0e2007-07-30 16:29:40 -07003702
3703err_uninit:
3704 if (dev->uninit)
3705 dev->uninit(dev);
3706 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707}
3708
3709/**
3710 * register_netdev - register a network device
3711 * @dev: device to register
3712 *
3713 * Take a completed network device structure and add it to the kernel
3714 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3715 * chain. 0 is returned on success. A negative errno code is returned
3716 * on a failure to set up the device, or if the name is a duplicate.
3717 *
Borislav Petkov38b4da32007-04-20 22:14:10 -07003718 * This is a wrapper around register_netdevice that takes the rtnl semaphore
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719 * and expands the device name if you passed a format string to
3720 * alloc_netdev.
3721 */
3722int register_netdev(struct net_device *dev)
3723{
3724 int err;
3725
3726 rtnl_lock();
3727
3728 /*
3729 * If the name is a format string the caller wants us to do a
3730 * name allocation.
3731 */
3732 if (strchr(dev->name, '%')) {
3733 err = dev_alloc_name(dev, dev->name);
3734 if (err < 0)
3735 goto out;
3736 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003737
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738 err = register_netdevice(dev);
3739out:
3740 rtnl_unlock();
3741 return err;
3742}
3743EXPORT_SYMBOL(register_netdev);
3744
3745/*
3746 * netdev_wait_allrefs - wait until all references are gone.
3747 *
3748 * This is called when unregistering network devices.
3749 *
3750 * Any protocol or device that holds a reference should register
3751 * for netdevice notification, and cleanup and put back the
3752 * reference if they receive an UNREGISTER event.
3753 * We can get stuck here if buggy protocols don't correctly
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003754 * call dev_put.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755 */
3756static void netdev_wait_allrefs(struct net_device *dev)
3757{
3758 unsigned long rebroadcast_time, warning_time;
3759
3760 rebroadcast_time = warning_time = jiffies;
3761 while (atomic_read(&dev->refcnt) != 0) {
3762 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003763 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764
3765 /* Rebroadcast unregister notification */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003766 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767
3768 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3769 &dev->state)) {
3770 /* We must not have linkwatch events
3771 * pending on unregister. If this
3772 * happens, we simply run the queue
3773 * unscheduled, resulting in a noop
3774 * for this device.
3775 */
3776 linkwatch_run_queue();
3777 }
3778
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003779 __rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780
3781 rebroadcast_time = jiffies;
3782 }
3783
3784 msleep(250);
3785
3786 if (time_after(jiffies, warning_time + 10 * HZ)) {
3787 printk(KERN_EMERG "unregister_netdevice: "
3788 "waiting for %s to become free. Usage "
3789 "count = %d\n",
3790 dev->name, atomic_read(&dev->refcnt));
3791 warning_time = jiffies;
3792 }
3793 }
3794}
3795
3796/* The sequence is:
3797 *
3798 * rtnl_lock();
3799 * ...
3800 * register_netdevice(x1);
3801 * register_netdevice(x2);
3802 * ...
3803 * unregister_netdevice(y1);
3804 * unregister_netdevice(y2);
3805 * ...
3806 * rtnl_unlock();
3807 * free_netdev(y1);
3808 * free_netdev(y2);
3809 *
3810 * We are invoked by rtnl_unlock() after it drops the semaphore.
3811 * This allows us to deal with problems:
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003812 * 1) We can delete sysfs objects which invoke hotplug
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813 * without deadlocking with linkwatch via keventd.
3814 * 2) Since we run with the RTNL semaphore not held, we can sleep
3815 * safely in order to wait for the netdev refcnt to drop to zero.
3816 */
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003817static DEFINE_MUTEX(net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818void netdev_run_todo(void)
3819{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003820 struct list_head list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821
3822 /* Need to guard against multiple cpu's getting out of order. */
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003823 mutex_lock(&net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003824
3825 /* Not safe to do outside the semaphore. We must not return
3826 * until all unregister events invoked by the local processor
3827 * have been completed (either by this todo run, or one on
3828 * another cpu).
3829 */
3830 if (list_empty(&net_todo_list))
3831 goto out;
3832
3833 /* Snapshot list, allow later requests */
3834 spin_lock(&net_todo_list_lock);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003835 list_replace_init(&net_todo_list, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836 spin_unlock(&net_todo_list_lock);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003837
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838 while (!list_empty(&list)) {
3839 struct net_device *dev
3840 = list_entry(list.next, struct net_device, todo_list);
3841 list_del(&dev->todo_list);
3842
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003843 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844 printk(KERN_ERR "network todo '%s' but state %d\n",
3845 dev->name, dev->reg_state);
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003846 dump_stack();
3847 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848 }
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003849
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003850 dev->reg_state = NETREG_UNREGISTERED;
3851
3852 netdev_wait_allrefs(dev);
3853
3854 /* paranoia */
3855 BUG_ON(atomic_read(&dev->refcnt));
3856 BUG_TRAP(!dev->ip_ptr);
3857 BUG_TRAP(!dev->ip6_ptr);
3858 BUG_TRAP(!dev->dn_ptr);
3859
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07003860 if (dev->destructor)
3861 dev->destructor(dev);
Stephen Hemminger9093bbb2007-05-19 15:39:25 -07003862
3863 /* Free network device */
3864 kobject_put(&dev->dev.kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865 }
3866
3867out:
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -08003868 mutex_unlock(&net_todo_run_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869}
3870
Rusty Russell5a1b5892007-04-28 21:04:03 -07003871static struct net_device_stats *internal_stats(struct net_device *dev)
Rusty Russellc45d2862007-03-28 14:29:08 -07003872{
Rusty Russell5a1b5892007-04-28 21:04:03 -07003873 return &dev->stats;
Rusty Russellc45d2862007-03-28 14:29:08 -07003874}
3875
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876/**
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003877 * alloc_netdev_mq - allocate network device
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 * @sizeof_priv: size of private data to allocate space for
3879 * @name: device name format string
3880 * @setup: callback to initialize device
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003881 * @queue_count: the number of subqueues to allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 *
3883 * Allocates a struct net_device with private data area for driver use
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003884 * and performs basic initialization. Also allocates subquue structs
3885 * for each queue on the device at the end of the netdevice.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003886 */
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003887struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
3888 void (*setup)(struct net_device *), unsigned int queue_count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889{
3890 void *p;
3891 struct net_device *dev;
3892 int alloc_size;
3893
Stephen Hemmingerb6fe17d2006-08-29 17:06:13 -07003894 BUG_ON(strlen(name) >= sizeof(dev->name));
3895
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 /* ensure 32-byte alignment of both the device and private area */
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003897 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST +
Patrick McHardy31ce72a2007-07-20 19:45:45 -07003898 (sizeof(struct net_device_subqueue) * (queue_count - 1))) &
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003899 ~NETDEV_ALIGN_CONST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3901
Paolo 'Blaisorblade' Giarrusso31380de2006-04-06 22:38:28 -07003902 p = kzalloc(alloc_size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903 if (!p) {
Stephen Hemmingerb6fe17d2006-08-29 17:06:13 -07003904 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 return NULL;
3906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907
3908 dev = (struct net_device *)
3909 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3910 dev->padded = (char *)dev - (char *)p;
Eric W. Biederman6d34b1c2007-09-12 12:57:33 +02003911 dev->nd_net = &init_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003913 if (sizeof_priv) {
3914 dev->priv = ((char *)dev +
3915 ((sizeof(struct net_device) +
3916 (sizeof(struct net_device_subqueue) *
Patrick McHardy31ce72a2007-07-20 19:45:45 -07003917 (queue_count - 1)) + NETDEV_ALIGN_CONST)
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003918 & ~NETDEV_ALIGN_CONST));
3919 }
3920
3921 dev->egress_subqueue_count = queue_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922
Rusty Russell5a1b5892007-04-28 21:04:03 -07003923 dev->get_stats = internal_stats;
Stephen Hemmingerbea33482007-10-03 16:41:36 -07003924 netpoll_netdev_init(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925 setup(dev);
3926 strcpy(dev->name, name);
3927 return dev;
3928}
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003929EXPORT_SYMBOL(alloc_netdev_mq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003930
3931/**
3932 * free_netdev - free network device
3933 * @dev: device
3934 *
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003935 * This function does the last stage of destroying an allocated device
3936 * interface. The reference to the device object is released.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937 * If this is the last reference then it will be freed.
3938 */
3939void free_netdev(struct net_device *dev)
3940{
Stephen Hemminger3041a062006-05-26 13:25:24 -07003941 /* Compatibility with error handling in drivers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942 if (dev->reg_state == NETREG_UNINITIALIZED) {
3943 kfree((char *)dev - dev->padded);
3944 return;
3945 }
3946
3947 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3948 dev->reg_state = NETREG_RELEASED;
3949
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07003950 /* will free via device release */
3951 put_device(&dev->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952}
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003953
Linus Torvalds1da177e2005-04-16 15:20:36 -07003954/* Synchronize with packet receive processing. */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09003955void synchronize_net(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956{
3957 might_sleep();
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07003958 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959}
3960
3961/**
3962 * unregister_netdevice - remove device from the kernel
3963 * @dev: device
3964 *
3965 * This function shuts down a device interface and removes it
Wang Chend59b54b2007-12-11 02:28:03 -08003966 * from the kernel tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003967 *
3968 * Callers must hold the rtnl semaphore. You may want
3969 * unregister_netdev() instead of this.
3970 */
3971
Stephen Hemminger22f8cde2007-02-07 00:09:58 -08003972void unregister_netdevice(struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973{
Herbert Xua6620712007-12-12 19:21:56 -08003974 ASSERT_RTNL();
3975
Daniel Lezcano93ee31f2007-10-30 15:38:18 -07003976 rollback_registered(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977 /* Finish processing unregister after unlock */
3978 net_set_todo(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979}
3980
3981/**
3982 * unregister_netdev - remove device from the kernel
3983 * @dev: device
3984 *
3985 * This function shuts down a device interface and removes it
Wang Chend59b54b2007-12-11 02:28:03 -08003986 * from the kernel tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 *
3988 * This is just a wrapper for unregister_netdevice that takes
3989 * the rtnl semaphore. In general you want to use this and not
3990 * unregister_netdevice.
3991 */
3992void unregister_netdev(struct net_device *dev)
3993{
3994 rtnl_lock();
3995 unregister_netdevice(dev);
3996 rtnl_unlock();
3997}
3998
3999EXPORT_SYMBOL(unregister_netdev);
4000
Eric W. Biedermance286d32007-09-12 13:53:49 +02004001/**
4002 * dev_change_net_namespace - move device to different nethost namespace
4003 * @dev: device
4004 * @net: network namespace
4005 * @pat: If not NULL name pattern to try if the current device name
4006 * is already taken in the destination network namespace.
4007 *
4008 * This function shuts down a device interface and moves it
4009 * to a new network namespace. On success 0 is returned, on
4010 * a failure a netagive errno code is returned.
4011 *
4012 * Callers must hold the rtnl semaphore.
4013 */
4014
4015int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4016{
4017 char buf[IFNAMSIZ];
4018 const char *destname;
4019 int err;
4020
4021 ASSERT_RTNL();
4022
4023 /* Don't allow namespace local devices to be moved. */
4024 err = -EINVAL;
4025 if (dev->features & NETIF_F_NETNS_LOCAL)
4026 goto out;
4027
4028 /* Ensure the device has been registrered */
4029 err = -EINVAL;
4030 if (dev->reg_state != NETREG_REGISTERED)
4031 goto out;
4032
4033 /* Get out if there is nothing todo */
4034 err = 0;
4035 if (dev->nd_net == net)
4036 goto out;
4037
4038 /* Pick the destination device name, and ensure
4039 * we can use it in the destination network namespace.
4040 */
4041 err = -EEXIST;
4042 destname = dev->name;
4043 if (__dev_get_by_name(net, destname)) {
4044 /* We get here if we can't use the current device name */
4045 if (!pat)
4046 goto out;
4047 if (!dev_valid_name(pat))
4048 goto out;
4049 if (strchr(pat, '%')) {
4050 if (__dev_alloc_name(net, pat, buf) < 0)
4051 goto out;
4052 destname = buf;
4053 } else
4054 destname = pat;
4055 if (__dev_get_by_name(net, destname))
4056 goto out;
4057 }
4058
4059 /*
4060 * And now a mini version of register_netdevice unregister_netdevice.
4061 */
4062
4063 /* If device is running close it first. */
Pavel Emelyanov9b772652007-10-10 02:49:09 -07004064 dev_close(dev);
Eric W. Biedermance286d32007-09-12 13:53:49 +02004065
4066 /* And unlink it from device chain */
4067 err = -ENODEV;
4068 unlist_netdevice(dev);
4069
4070 synchronize_net();
4071
4072 /* Shutdown queueing discipline. */
4073 dev_shutdown(dev);
4074
4075 /* Notify protocols, that we are about to destroy
4076 this device. They should clean all the things.
4077 */
4078 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4079
4080 /*
4081 * Flush the unicast and multicast chains
4082 */
4083 dev_addr_discard(dev);
4084
4085 /* Actually switch the network namespace */
4086 dev->nd_net = net;
4087
4088 /* Assign the new device name */
4089 if (destname != dev->name)
4090 strcpy(dev->name, destname);
4091
4092 /* If there is an ifindex conflict assign a new one */
4093 if (__dev_get_by_index(net, dev->ifindex)) {
4094 int iflink = (dev->iflink == dev->ifindex);
4095 dev->ifindex = dev_new_index(net);
4096 if (iflink)
4097 dev->iflink = dev->ifindex;
4098 }
4099
Eric W. Biederman8b41d182007-09-26 22:02:53 -07004100 /* Fixup kobjects */
Eric W. Biedermance286d32007-09-12 13:53:49 +02004101 err = device_rename(&dev->dev, dev->name);
Eric W. Biederman8b41d182007-09-26 22:02:53 -07004102 WARN_ON(err);
Eric W. Biedermance286d32007-09-12 13:53:49 +02004103
4104 /* Add the device back in the hashes */
4105 list_netdevice(dev);
4106
4107 /* Notify protocols, that a new device appeared. */
4108 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4109
4110 synchronize_net();
4111 err = 0;
4112out:
4113 return err;
4114}
4115
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116static int dev_cpu_callback(struct notifier_block *nfb,
4117 unsigned long action,
4118 void *ocpu)
4119{
4120 struct sk_buff **list_skb;
4121 struct net_device **list_net;
4122 struct sk_buff *skb;
4123 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4124 struct softnet_data *sd, *oldsd;
4125
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004126 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004127 return NOTIFY_OK;
4128
4129 local_irq_disable();
4130 cpu = smp_processor_id();
4131 sd = &per_cpu(softnet_data, cpu);
4132 oldsd = &per_cpu(softnet_data, oldcpu);
4133
4134 /* Find end of our completion_queue. */
4135 list_skb = &sd->completion_queue;
4136 while (*list_skb)
4137 list_skb = &(*list_skb)->next;
4138 /* Append completion queue from offline CPU. */
4139 *list_skb = oldsd->completion_queue;
4140 oldsd->completion_queue = NULL;
4141
4142 /* Find end of our output_queue. */
4143 list_net = &sd->output_queue;
4144 while (*list_net)
4145 list_net = &(*list_net)->next_sched;
4146 /* Append output queue from offline CPU. */
4147 *list_net = oldsd->output_queue;
4148 oldsd->output_queue = NULL;
4149
4150 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4151 local_irq_enable();
4152
4153 /* Process offline CPU's input_pkt_queue */
4154 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4155 netif_rx(skb);
4156
4157 return NOTIFY_OK;
4158}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004159
Chris Leechdb217332006-06-17 21:24:58 -07004160#ifdef CONFIG_NET_DMA
4161/**
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07004162 * net_dma_rebalance - try to maintain one DMA channel per CPU
4163 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4164 *
4165 * This is called when the number of channels allocated to the net_dma client
4166 * changes. The net_dma client tries to have one DMA channel per CPU.
Chris Leechdb217332006-06-17 21:24:58 -07004167 */
Dan Williamsd379b012007-07-09 11:56:42 -07004168
4169static void net_dma_rebalance(struct net_dma *net_dma)
Chris Leechdb217332006-06-17 21:24:58 -07004170{
Dan Williamsd379b012007-07-09 11:56:42 -07004171 unsigned int cpu, i, n, chan_idx;
Chris Leechdb217332006-06-17 21:24:58 -07004172 struct dma_chan *chan;
4173
Dan Williamsd379b012007-07-09 11:56:42 -07004174 if (cpus_empty(net_dma->channel_mask)) {
Chris Leechdb217332006-06-17 21:24:58 -07004175 for_each_online_cpu(cpu)
Alexey Dobriyan29bbd722006-08-02 15:02:31 -07004176 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
Chris Leechdb217332006-06-17 21:24:58 -07004177 return;
4178 }
4179
4180 i = 0;
4181 cpu = first_cpu(cpu_online_map);
4182
Dan Williamsd379b012007-07-09 11:56:42 -07004183 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
4184 chan = net_dma->channels[chan_idx];
4185
4186 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4187 + (i < (num_online_cpus() %
4188 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
Chris Leechdb217332006-06-17 21:24:58 -07004189
4190 while(n) {
Alexey Dobriyan29bbd722006-08-02 15:02:31 -07004191 per_cpu(softnet_data, cpu).net_dma = chan;
Chris Leechdb217332006-06-17 21:24:58 -07004192 cpu = next_cpu(cpu, cpu_online_map);
4193 n--;
4194 }
4195 i++;
4196 }
Chris Leechdb217332006-06-17 21:24:58 -07004197}
4198
4199/**
4200 * netdev_dma_event - event callback for the net_dma_client
4201 * @client: should always be net_dma_client
Randy Dunlapf4b8ea72006-06-22 16:00:11 -07004202 * @chan: DMA channel for the event
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07004203 * @state: DMA state to be handled
Chris Leechdb217332006-06-17 21:24:58 -07004204 */
Dan Williamsd379b012007-07-09 11:56:42 -07004205static enum dma_state_client
4206netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4207 enum dma_state state)
Chris Leechdb217332006-06-17 21:24:58 -07004208{
Dan Williamsd379b012007-07-09 11:56:42 -07004209 int i, found = 0, pos = -1;
4210 struct net_dma *net_dma =
4211 container_of(client, struct net_dma, client);
4212 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4213
4214 spin_lock(&net_dma->lock);
4215 switch (state) {
4216 case DMA_RESOURCE_AVAILABLE:
4217 for (i = 0; i < NR_CPUS; i++)
4218 if (net_dma->channels[i] == chan) {
4219 found = 1;
4220 break;
4221 } else if (net_dma->channels[i] == NULL && pos < 0)
4222 pos = i;
4223
4224 if (!found && pos >= 0) {
4225 ack = DMA_ACK;
4226 net_dma->channels[pos] = chan;
4227 cpu_set(pos, net_dma->channel_mask);
4228 net_dma_rebalance(net_dma);
4229 }
Chris Leechdb217332006-06-17 21:24:58 -07004230 break;
4231 case DMA_RESOURCE_REMOVED:
Dan Williamsd379b012007-07-09 11:56:42 -07004232 for (i = 0; i < NR_CPUS; i++)
4233 if (net_dma->channels[i] == chan) {
4234 found = 1;
4235 pos = i;
4236 break;
4237 }
4238
4239 if (found) {
4240 ack = DMA_ACK;
4241 cpu_clear(pos, net_dma->channel_mask);
4242 net_dma->channels[i] = NULL;
4243 net_dma_rebalance(net_dma);
4244 }
Chris Leechdb217332006-06-17 21:24:58 -07004245 break;
4246 default:
4247 break;
4248 }
Dan Williamsd379b012007-07-09 11:56:42 -07004249 spin_unlock(&net_dma->lock);
4250
4251 return ack;
Chris Leechdb217332006-06-17 21:24:58 -07004252}
4253
4254/**
4255 * netdev_dma_regiser - register the networking subsystem as a DMA client
4256 */
4257static int __init netdev_dma_register(void)
4258{
Dan Williamsd379b012007-07-09 11:56:42 -07004259 spin_lock_init(&net_dma.lock);
4260 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4261 dma_async_client_register(&net_dma.client);
4262 dma_async_client_chan_request(&net_dma.client);
Chris Leechdb217332006-06-17 21:24:58 -07004263 return 0;
4264}
4265
4266#else
4267static int __init netdev_dma_register(void) { return -ENODEV; }
4268#endif /* CONFIG_NET_DMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004269
Herbert Xu7f353bf2007-08-10 15:47:58 -07004270/**
4271 * netdev_compute_feature - compute conjunction of two feature sets
4272 * @all: first feature set
4273 * @one: second feature set
4274 *
4275 * Computes a new feature set after adding a device with feature set
4276 * @one to the master device with current feature set @all. Returns
4277 * the new feature set.
4278 */
4279int netdev_compute_features(unsigned long all, unsigned long one)
4280{
4281 /* if device needs checksumming, downgrade to hw checksumming */
4282 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4283 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4284
4285 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4286 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4287 all ^= NETIF_F_HW_CSUM
4288 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4289
4290 if (one & NETIF_F_GSO)
4291 one |= NETIF_F_GSO_SOFTWARE;
4292 one |= NETIF_F_GSO;
4293
4294 /* If even one device supports robust GSO, enable it for all. */
4295 if (one & NETIF_F_GSO_ROBUST)
4296 all |= NETIF_F_GSO_ROBUST;
4297
4298 all &= one | NETIF_F_LLTX;
4299
4300 if (!(all & NETIF_F_ALL_CSUM))
4301 all &= ~NETIF_F_SG;
4302 if (!(all & NETIF_F_SG))
4303 all &= ~NETIF_F_GSO_MASK;
4304
4305 return all;
4306}
4307EXPORT_SYMBOL(netdev_compute_features);
4308
Pavel Emelyanov30d97d32007-09-16 15:40:33 -07004309static struct hlist_head *netdev_create_hash(void)
4310{
4311 int i;
4312 struct hlist_head *hash;
4313
4314 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4315 if (hash != NULL)
4316 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4317 INIT_HLIST_HEAD(&hash[i]);
4318
4319 return hash;
4320}
4321
Eric W. Biederman881d9662007-09-17 11:56:21 -07004322/* Initialize per network namespace state */
Pavel Emelyanov46650792007-10-08 20:38:39 -07004323static int __net_init netdev_init(struct net *net)
Eric W. Biederman881d9662007-09-17 11:56:21 -07004324{
Eric W. Biederman881d9662007-09-17 11:56:21 -07004325 INIT_LIST_HEAD(&net->dev_base_head);
Eric W. Biederman881d9662007-09-17 11:56:21 -07004326
Pavel Emelyanov30d97d32007-09-16 15:40:33 -07004327 net->dev_name_head = netdev_create_hash();
4328 if (net->dev_name_head == NULL)
4329 goto err_name;
Eric W. Biederman881d9662007-09-17 11:56:21 -07004330
Pavel Emelyanov30d97d32007-09-16 15:40:33 -07004331 net->dev_index_head = netdev_create_hash();
4332 if (net->dev_index_head == NULL)
4333 goto err_idx;
Eric W. Biederman881d9662007-09-17 11:56:21 -07004334
4335 return 0;
Pavel Emelyanov30d97d32007-09-16 15:40:33 -07004336
4337err_idx:
4338 kfree(net->dev_name_head);
4339err_name:
4340 return -ENOMEM;
Eric W. Biederman881d9662007-09-17 11:56:21 -07004341}
4342
Pavel Emelyanov46650792007-10-08 20:38:39 -07004343static void __net_exit netdev_exit(struct net *net)
Eric W. Biederman881d9662007-09-17 11:56:21 -07004344{
4345 kfree(net->dev_name_head);
4346 kfree(net->dev_index_head);
4347}
4348
Denis V. Lunev022cbae2007-11-13 03:23:50 -08004349static struct pernet_operations __net_initdata netdev_net_ops = {
Eric W. Biederman881d9662007-09-17 11:56:21 -07004350 .init = netdev_init,
4351 .exit = netdev_exit,
4352};
4353
Pavel Emelyanov46650792007-10-08 20:38:39 -07004354static void __net_exit default_device_exit(struct net *net)
Eric W. Biedermance286d32007-09-12 13:53:49 +02004355{
4356 struct net_device *dev, *next;
4357 /*
4358 * Push all migratable of the network devices back to the
4359 * initial network namespace
4360 */
4361 rtnl_lock();
4362 for_each_netdev_safe(net, dev, next) {
4363 int err;
4364
4365 /* Ignore unmoveable devices (i.e. loopback) */
4366 if (dev->features & NETIF_F_NETNS_LOCAL)
4367 continue;
4368
4369 /* Push remaing network devices to init_net */
4370 err = dev_change_net_namespace(dev, &init_net, "dev%d");
4371 if (err) {
4372 printk(KERN_WARNING "%s: failed to move %s to init_net: %d\n",
4373 __func__, dev->name, err);
4374 unregister_netdevice(dev);
4375 }
4376 }
4377 rtnl_unlock();
4378}
4379
Denis V. Lunev022cbae2007-11-13 03:23:50 -08004380static struct pernet_operations __net_initdata default_device_ops = {
Eric W. Biedermance286d32007-09-12 13:53:49 +02004381 .exit = default_device_exit,
4382};
4383
Linus Torvalds1da177e2005-04-16 15:20:36 -07004384/*
4385 * Initialize the DEV module. At boot time this walks the device list and
4386 * unhooks any devices that fail to initialise (normally hardware not
4387 * present) and leaves us with a valid list of present and active devices.
4388 *
4389 */
4390
4391/*
4392 * This is called single threaded during boot, so no need
4393 * to take the rtnl semaphore.
4394 */
4395static int __init net_dev_init(void)
4396{
4397 int i, rc = -ENOMEM;
4398
4399 BUG_ON(!dev_boot_phase);
4400
Linus Torvalds1da177e2005-04-16 15:20:36 -07004401 if (dev_proc_init())
4402 goto out;
4403
Eric W. Biederman8b41d182007-09-26 22:02:53 -07004404 if (netdev_kobject_init())
Linus Torvalds1da177e2005-04-16 15:20:36 -07004405 goto out;
4406
4407 INIT_LIST_HEAD(&ptype_all);
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08004408 for (i = 0; i < PTYPE_HASH_SIZE; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004409 INIT_LIST_HEAD(&ptype_base[i]);
4410
Eric W. Biederman881d9662007-09-17 11:56:21 -07004411 if (register_pernet_subsys(&netdev_net_ops))
4412 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004413
Eric W. Biedermance286d32007-09-12 13:53:49 +02004414 if (register_pernet_device(&default_device_ops))
4415 goto out;
4416
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417 /*
4418 * Initialise the packet receive queues.
4419 */
4420
KAMEZAWA Hiroyuki6f912042006-04-10 22:52:50 -07004421 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004422 struct softnet_data *queue;
4423
4424 queue = &per_cpu(softnet_data, i);
4425 skb_queue_head_init(&queue->input_pkt_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426 queue->completion_queue = NULL;
4427 INIT_LIST_HEAD(&queue->poll_list);
Stephen Hemmingerbea33482007-10-03 16:41:36 -07004428
4429 queue->backlog.poll = process_backlog;
4430 queue->backlog.weight = weight_p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004431 }
4432
Chris Leechdb217332006-06-17 21:24:58 -07004433 netdev_dma_register();
4434
Linus Torvalds1da177e2005-04-16 15:20:36 -07004435 dev_boot_phase = 0;
4436
4437 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4438 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4439
4440 hotcpu_notifier(dev_cpu_callback, 0);
4441 dst_init();
4442 dev_mcast_init();
4443 rc = 0;
4444out:
4445 return rc;
4446}
4447
4448subsys_initcall(net_dev_init);
4449
4450EXPORT_SYMBOL(__dev_get_by_index);
4451EXPORT_SYMBOL(__dev_get_by_name);
4452EXPORT_SYMBOL(__dev_remove_pack);
Mitch Williamsc2373ee2005-11-09 10:34:45 -08004453EXPORT_SYMBOL(dev_valid_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004454EXPORT_SYMBOL(dev_add_pack);
4455EXPORT_SYMBOL(dev_alloc_name);
4456EXPORT_SYMBOL(dev_close);
4457EXPORT_SYMBOL(dev_get_by_flags);
4458EXPORT_SYMBOL(dev_get_by_index);
4459EXPORT_SYMBOL(dev_get_by_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004460EXPORT_SYMBOL(dev_open);
4461EXPORT_SYMBOL(dev_queue_xmit);
4462EXPORT_SYMBOL(dev_remove_pack);
4463EXPORT_SYMBOL(dev_set_allmulti);
4464EXPORT_SYMBOL(dev_set_promiscuity);
4465EXPORT_SYMBOL(dev_change_flags);
4466EXPORT_SYMBOL(dev_set_mtu);
4467EXPORT_SYMBOL(dev_set_mac_address);
4468EXPORT_SYMBOL(free_netdev);
4469EXPORT_SYMBOL(netdev_boot_setup_check);
4470EXPORT_SYMBOL(netdev_set_master);
4471EXPORT_SYMBOL(netdev_state_change);
4472EXPORT_SYMBOL(netif_receive_skb);
4473EXPORT_SYMBOL(netif_rx);
4474EXPORT_SYMBOL(register_gifconf);
4475EXPORT_SYMBOL(register_netdevice);
4476EXPORT_SYMBOL(register_netdevice_notifier);
4477EXPORT_SYMBOL(skb_checksum_help);
4478EXPORT_SYMBOL(synchronize_net);
4479EXPORT_SYMBOL(unregister_netdevice);
4480EXPORT_SYMBOL(unregister_netdevice_notifier);
4481EXPORT_SYMBOL(net_enable_timestamp);
4482EXPORT_SYMBOL(net_disable_timestamp);
4483EXPORT_SYMBOL(dev_get_flags);
4484
4485#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4486EXPORT_SYMBOL(br_handle_frame_hook);
4487EXPORT_SYMBOL(br_fdb_get_hook);
4488EXPORT_SYMBOL(br_fdb_put_hook);
4489#endif
4490
4491#ifdef CONFIG_KMOD
4492EXPORT_SYMBOL(dev_load);
4493#endif
4494
4495EXPORT_PER_CPU_SYMBOL(softnet_data);