blob: 262df226b3c9a85852d41af69cc269cb930008fb [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>
Ben Hutchings0187bdf2008-06-19 16:15:47 -070093#include <linux/ethtool.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070094#include <linux/notifier.h>
95#include <linux/skbuff.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020096#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070097#include <net/sock.h>
98#include <linux/rtnetlink.h>
99#include <linux/proc_fs.h>
100#include <linux/seq_file.h>
101#include <linux/stat.h>
102#include <linux/if_bridge.h>
Patrick McHardyb863ceb2007-07-14 18:55:06 -0700103#include <linux/if_macvlan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104#include <net/dst.h>
105#include <net/pkt_sched.h>
106#include <net/checksum.h>
107#include <linux/highmem.h>
108#include <linux/init.h>
109#include <linux/kmod.h>
110#include <linux/module.h>
111#include <linux/kallsyms.h>
112#include <linux/netpoll.h>
113#include <linux/rcupdate.h>
114#include <linux/delay.h>
Johannes Berg295f4a12007-04-26 20:43:56 -0700115#include <net/wext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116#include <net/iw_handler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117#include <asm/current.h>
Steve Grubb5bdb9882005-12-03 08:39:35 -0500118#include <linux/audit.h>
Chris Leechdb217332006-06-17 21:24:58 -0700119#include <linux/dmaengine.h>
Herbert Xuf6a78bf2006-06-22 02:57:17 -0700120#include <linux/err.h>
David S. Millerc7fa9d12006-08-15 16:34:13 -0700121#include <linux/ctype.h>
Jarek Poplawski723e98b2007-05-15 22:46:18 -0700122#include <linux/if_arp.h>
Ben Hutchings6de329e2008-06-16 17:02:28 -0700123#include <linux/if_vlan.h>
David S. Miller8f0f2222008-07-15 03:47:03 -0700124#include <linux/ip.h>
Alexander Duyckad55dca2008-09-20 22:05:50 -0700125#include <net/ip.h>
David S. Miller8f0f2222008-07-15 03:47:03 -0700126#include <linux/ipv6.h>
127#include <linux/in.h>
David S. Millerb6b2fed2008-07-21 09:48:06 -0700128#include <linux/jhash.h>
129#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Pavel Emelyanov342709e2007-10-23 21:14:45 -0700131#include "net-sysfs.h"
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133/*
134 * The list of packet types we will receive (as opposed to discard)
135 * and the routines to invoke.
136 *
137 * Why 16. Because with 16 the only overlap we get on a hash of the
138 * low nibble of the protocol value is RARP/SNAP/X.25.
139 *
140 * NOTE: That is no longer true with the addition of VLAN tags. Not
141 * sure which should go first, but I bet it won't make much
142 * difference if we are running VLANs. The good news is that
143 * this protocol won't be in the list unless compiled in, so
Stephen Hemminger3041a062006-05-26 13:25:24 -0700144 * the average user (w/out VLANs) will not be adversely affected.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 * --BLG
146 *
147 * 0800 IP
148 * 8100 802.1Q VLAN
149 * 0001 802.3
150 * 0002 AX.25
151 * 0004 802.2
152 * 8035 RARP
153 * 0005 SNAP
154 * 0805 X.25
155 * 0806 ARP
156 * 8137 IPX
157 * 0009 Localtalk
158 * 86DD IPv6
159 */
160
Pavel Emelyanov82d8a862007-11-26 20:12:58 +0800161#define PTYPE_HASH_SIZE (16)
162#define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164static DEFINE_SPINLOCK(ptype_lock);
Pavel Emelyanov82d8a862007-11-26 20:12:58 +0800165static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
Stephen Hemminger6b2bedc2007-03-12 14:33:50 -0700166static struct list_head ptype_all __read_mostly; /* Taps */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
Chris Leechdb217332006-06-17 21:24:58 -0700168#ifdef CONFIG_NET_DMA
Dan Williamsd379b012007-07-09 11:56:42 -0700169struct net_dma {
170 struct dma_client client;
171 spinlock_t lock;
172 cpumask_t channel_mask;
Mike Travis0c0b0ac2008-05-02 16:43:08 -0700173 struct dma_chan **channels;
Dan Williamsd379b012007-07-09 11:56:42 -0700174};
175
176static enum dma_state_client
177netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
178 enum dma_state state);
179
180static struct net_dma net_dma = {
181 .client = {
182 .event_callback = netdev_dma_event,
183 },
184};
Chris Leechdb217332006-06-17 21:24:58 -0700185#endif
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187/*
Pavel Emelianov7562f872007-05-03 15:13:45 -0700188 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 * semaphore.
190 *
191 * Pure readers hold dev_base_lock for reading.
192 *
193 * Writers must hold the rtnl semaphore while they loop through the
Pavel Emelianov7562f872007-05-03 15:13:45 -0700194 * dev_base_head list, and hold dev_base_lock for writing when they do the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 * actual updates. This allows pure readers to access the list even
196 * while a writer is preparing to update it.
197 *
198 * To put it another way, dev_base_lock is held for writing only to
199 * protect against pure readers; the rtnl semaphore provides the
200 * protection against other writers.
201 *
202 * See, for example usages, register_netdevice() and
203 * unregister_netdevice(), which must be called with the rtnl
204 * semaphore held.
205 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206DEFINE_RWLOCK(dev_base_lock);
207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208EXPORT_SYMBOL(dev_base_lock);
209
210#define NETDEV_HASHBITS 8
Eric W. Biederman881d9662007-09-17 11:56:21 -0700211#define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Eric W. Biederman881d9662007-09-17 11:56:21 -0700213static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
Eric W. Biederman881d9662007-09-17 11:56:21 -0700216 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
Eric W. Biederman881d9662007-09-17 11:56:21 -0700219static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
Eric W. Biederman881d9662007-09-17 11:56:21 -0700221 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222}
223
Eric W. Biedermance286d32007-09-12 13:53:49 +0200224/* Device list insertion */
225static int list_netdevice(struct net_device *dev)
226{
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +0900227 struct net *net = dev_net(dev);
Eric W. Biedermance286d32007-09-12 13:53:49 +0200228
229 ASSERT_RTNL();
230
231 write_lock_bh(&dev_base_lock);
232 list_add_tail(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
235 write_unlock_bh(&dev_base_lock);
236 return 0;
237}
238
239/* Device list removal */
240static void unlist_netdevice(struct net_device *dev)
241{
242 ASSERT_RTNL();
243
244 /* Unlink dev from the device chain */
245 write_lock_bh(&dev_base_lock);
246 list_del(&dev->dev_list);
247 hlist_del(&dev->name_hlist);
248 hlist_del(&dev->index_hlist);
249 write_unlock_bh(&dev_base_lock);
250}
251
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252/*
253 * Our notifier list
254 */
255
Alan Sternf07d5b92006-05-09 15:23:03 -0700256static RAW_NOTIFIER_HEAD(netdev_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258/*
259 * Device drivers call our routines to queue packets here. We empty the
260 * queue in the local softnet handler.
261 */
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700262
263DEFINE_PER_CPU(struct softnet_data, softnet_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
David S. Millercf508b12008-07-22 14:16:42 -0700265#ifdef CONFIG_LOCKDEP
Jarek Poplawski723e98b2007-05-15 22:46:18 -0700266/*
David S. Millerc773e842008-07-08 23:13:53 -0700267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
Jarek Poplawski723e98b2007-05-15 22:46:18 -0700268 * according to dev->type
269 */
270static const unsigned short netdev_lock_type[] =
271 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
272 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
273 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
274 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
275 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
276 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
277 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
278 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
279 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
280 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
281 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
282 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
283 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
284 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
285 ARPHRD_NONE};
286
287static const char *netdev_lock_name[] =
288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
300 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
301 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
302 "_xmit_NONE"};
303
304static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
David S. Millercf508b12008-07-22 14:16:42 -0700305static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
Jarek Poplawski723e98b2007-05-15 22:46:18 -0700306
307static inline unsigned short netdev_lock_pos(unsigned short dev_type)
308{
309 int i;
310
311 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
312 if (netdev_lock_type[i] == dev_type)
313 return i;
314 /* the last key is used by default */
315 return ARRAY_SIZE(netdev_lock_type) - 1;
316}
317
David S. Millercf508b12008-07-22 14:16:42 -0700318static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
319 unsigned short dev_type)
Jarek Poplawski723e98b2007-05-15 22:46:18 -0700320{
321 int i;
322
323 i = netdev_lock_pos(dev_type);
324 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
325 netdev_lock_name[i]);
326}
David S. Millercf508b12008-07-22 14:16:42 -0700327
328static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
329{
330 int i;
331
332 i = netdev_lock_pos(dev->type);
333 lockdep_set_class_and_name(&dev->addr_list_lock,
334 &netdev_addr_lock_key[i],
335 netdev_lock_name[i]);
336}
Jarek Poplawski723e98b2007-05-15 22:46:18 -0700337#else
David S. Millercf508b12008-07-22 14:16:42 -0700338static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
340{
341}
342static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
Jarek Poplawski723e98b2007-05-15 22:46:18 -0700343{
344}
345#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
347/*******************************************************************************
348
349 Protocol management and registration routines
350
351*******************************************************************************/
352
353/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 * Add a protocol ID to the list. Now that the input handler is
355 * smarter we can dispense with all the messy stuff that used to be
356 * here.
357 *
358 * BEWARE!!! Protocol handlers, mangling input packets,
359 * MUST BE last in hash buckets and checking protocol handlers
360 * MUST start from promiscuous ptype_all chain in net_bh.
361 * It is true now, do not change it.
362 * Explanation follows: if protocol handler, mangling packet, will
363 * be the first on list, it is not able to sense, that packet
364 * is cloned and should be copied-on-write, so that it will
365 * change it and subsequent readers will get broken packet.
366 * --ANK (980803)
367 */
368
369/**
370 * dev_add_pack - add packet handler
371 * @pt: packet type declaration
372 *
373 * Add a protocol handler to the networking stack. The passed &packet_type
374 * is linked into kernel lists and may not be freed until it has been
375 * removed from the kernel lists.
376 *
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900377 * This call does not sleep therefore it can not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 * guarantee all CPU's that are in middle of receiving packets
379 * will see the new packet type (until the next received packet).
380 */
381
382void dev_add_pack(struct packet_type *pt)
383{
384 int hash;
385
386 spin_lock_bh(&ptype_lock);
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -0700387 if (pt->type == htons(ETH_P_ALL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 list_add_rcu(&pt->list, &ptype_all);
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -0700389 else {
Pavel Emelyanov82d8a862007-11-26 20:12:58 +0800390 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 list_add_rcu(&pt->list, &ptype_base[hash]);
392 }
393 spin_unlock_bh(&ptype_lock);
394}
395
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396/**
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
399 *
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900403 * returns.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 *
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
408 */
409void __dev_remove_pack(struct packet_type *pt)
410{
411 struct list_head *head;
412 struct packet_type *pt1;
413
414 spin_lock_bh(&ptype_lock);
415
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -0700416 if (pt->type == htons(ETH_P_ALL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 head = &ptype_all;
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -0700418 else
Pavel Emelyanov82d8a862007-11-26 20:12:58 +0800419 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
421 list_for_each_entry(pt1, head, list) {
422 if (pt == pt1) {
423 list_del_rcu(&pt->list);
424 goto out;
425 }
426 }
427
428 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
429out:
430 spin_unlock_bh(&ptype_lock);
431}
432/**
433 * dev_remove_pack - remove packet handler
434 * @pt: packet type declaration
435 *
436 * Remove a protocol handler that was previously added to the kernel
437 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
438 * from the kernel lists and can be freed or reused once this function
439 * returns.
440 *
441 * This call sleeps to guarantee that no CPU is looking at the packet
442 * type after return.
443 */
444void dev_remove_pack(struct packet_type *pt)
445{
446 __dev_remove_pack(pt);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900447
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 synchronize_net();
449}
450
451/******************************************************************************
452
453 Device Boot-time Settings Routines
454
455*******************************************************************************/
456
457/* Boot time configuration table */
458static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
459
460/**
461 * netdev_boot_setup_add - add new setup entry
462 * @name: name of the device
463 * @map: configured settings for the device
464 *
465 * Adds new setup entry to the dev_boot_setup list. The function
466 * returns 0 on error and 1 on success. This is a generic routine to
467 * all netdevices.
468 */
469static int netdev_boot_setup_add(char *name, struct ifmap *map)
470{
471 struct netdev_boot_setup *s;
472 int i;
473
474 s = dev_boot_setup;
475 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
476 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
477 memset(s[i].name, 0, sizeof(s[i].name));
Wang Chen93b3cff2008-07-01 19:57:19 -0700478 strlcpy(s[i].name, name, IFNAMSIZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 memcpy(&s[i].map, map, sizeof(s[i].map));
480 break;
481 }
482 }
483
484 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
485}
486
487/**
488 * netdev_boot_setup_check - check boot time settings
489 * @dev: the netdevice
490 *
491 * Check boot time settings for the device.
492 * The found settings are set for the device to be used
493 * later in the device probing.
494 * Returns 0 if no settings found, 1 if they are.
495 */
496int netdev_boot_setup_check(struct net_device *dev)
497{
498 struct netdev_boot_setup *s = dev_boot_setup;
499 int i;
500
501 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
502 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
Wang Chen93b3cff2008-07-01 19:57:19 -0700503 !strcmp(dev->name, s[i].name)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 dev->irq = s[i].map.irq;
505 dev->base_addr = s[i].map.base_addr;
506 dev->mem_start = s[i].map.mem_start;
507 dev->mem_end = s[i].map.mem_end;
508 return 1;
509 }
510 }
511 return 0;
512}
513
514
515/**
516 * netdev_boot_base - get address from boot time settings
517 * @prefix: prefix for network device
518 * @unit: id for network device
519 *
520 * Check boot time settings for the base address of device.
521 * The found settings are set for the device to be used
522 * later in the device probing.
523 * Returns 0 if no settings found.
524 */
525unsigned long netdev_boot_base(const char *prefix, int unit)
526{
527 const struct netdev_boot_setup *s = dev_boot_setup;
528 char name[IFNAMSIZ];
529 int i;
530
531 sprintf(name, "%s%d", prefix, unit);
532
533 /*
534 * If device already registered then return base of 1
535 * to indicate not to probe for this interface
536 */
Eric W. Biederman881d9662007-09-17 11:56:21 -0700537 if (__dev_get_by_name(&init_net, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 return 1;
539
540 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
541 if (!strcmp(name, s[i].name))
542 return s[i].map.base_addr;
543 return 0;
544}
545
546/*
547 * Saves at boot time configured settings for any netdevice.
548 */
549int __init netdev_boot_setup(char *str)
550{
551 int ints[5];
552 struct ifmap map;
553
554 str = get_options(str, ARRAY_SIZE(ints), ints);
555 if (!str || !*str)
556 return 0;
557
558 /* Save settings */
559 memset(&map, 0, sizeof(map));
560 if (ints[0] > 0)
561 map.irq = ints[1];
562 if (ints[0] > 1)
563 map.base_addr = ints[2];
564 if (ints[0] > 2)
565 map.mem_start = ints[3];
566 if (ints[0] > 3)
567 map.mem_end = ints[4];
568
569 /* Add new entry to the list */
570 return netdev_boot_setup_add(str, &map);
571}
572
573__setup("netdev=", netdev_boot_setup);
574
575/*******************************************************************************
576
577 Device Interface Subroutines
578
579*******************************************************************************/
580
581/**
582 * __dev_get_by_name - find a device by its name
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700583 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 * @name: name to find
585 *
586 * Find an interface by name. Must be called under RTNL semaphore
587 * or @dev_base_lock. If the name is found a pointer to the device
588 * is returned. If the name is not found then %NULL is returned. The
589 * reference counters are not incremented so the caller must be
590 * careful with locks.
591 */
592
Eric W. Biederman881d9662007-09-17 11:56:21 -0700593struct net_device *__dev_get_by_name(struct net *net, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
595 struct hlist_node *p;
596
Eric W. Biederman881d9662007-09-17 11:56:21 -0700597 hlist_for_each(p, dev_name_hash(net, name)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 struct net_device *dev
599 = hlist_entry(p, struct net_device, name_hlist);
600 if (!strncmp(dev->name, name, IFNAMSIZ))
601 return dev;
602 }
603 return NULL;
604}
605
606/**
607 * dev_get_by_name - find a device by its name
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700608 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 * @name: name to find
610 *
611 * Find an interface by name. This can be called from any
612 * context and does its own locking. The returned handle has
613 * the usage count incremented and the caller must use dev_put() to
614 * release it when it is no longer needed. %NULL is returned if no
615 * matching device is found.
616 */
617
Eric W. Biederman881d9662007-09-17 11:56:21 -0700618struct net_device *dev_get_by_name(struct net *net, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
620 struct net_device *dev;
621
622 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700623 dev = __dev_get_by_name(net, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 if (dev)
625 dev_hold(dev);
626 read_unlock(&dev_base_lock);
627 return dev;
628}
629
630/**
631 * __dev_get_by_index - find a device by its ifindex
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700632 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 * @ifindex: index of device
634 *
635 * Search for an interface by index. Returns %NULL if the device
636 * is not found or a pointer to the device. The device has not
637 * had its reference counter increased so the caller must be careful
638 * about locking. The caller must hold either the RTNL semaphore
639 * or @dev_base_lock.
640 */
641
Eric W. Biederman881d9662007-09-17 11:56:21 -0700642struct net_device *__dev_get_by_index(struct net *net, int ifindex)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
644 struct hlist_node *p;
645
Eric W. Biederman881d9662007-09-17 11:56:21 -0700646 hlist_for_each(p, dev_index_hash(net, ifindex)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 struct net_device *dev
648 = hlist_entry(p, struct net_device, index_hlist);
649 if (dev->ifindex == ifindex)
650 return dev;
651 }
652 return NULL;
653}
654
655
656/**
657 * dev_get_by_index - find a device by its ifindex
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700658 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 * @ifindex: index of device
660 *
661 * Search for an interface by index. Returns NULL if the device
662 * is not found or a pointer to the device. The device returned has
663 * had a reference added and the pointer is safe until the user calls
664 * dev_put to indicate they have finished with it.
665 */
666
Eric W. Biederman881d9662007-09-17 11:56:21 -0700667struct net_device *dev_get_by_index(struct net *net, int ifindex)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668{
669 struct net_device *dev;
670
671 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700672 dev = __dev_get_by_index(net, ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 if (dev)
674 dev_hold(dev);
675 read_unlock(&dev_base_lock);
676 return dev;
677}
678
679/**
680 * dev_getbyhwaddr - find a device by its hardware address
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700681 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 * @type: media type of device
683 * @ha: hardware address
684 *
685 * Search for an interface by MAC address. Returns NULL if the device
686 * is not found or a pointer to the device. The caller must hold the
687 * rtnl semaphore. The returned device has not had its ref count increased
688 * and the caller must therefore be careful about locking
689 *
690 * BUGS:
691 * If the API was consistent this would be __dev_get_by_hwaddr
692 */
693
Eric W. Biederman881d9662007-09-17 11:56:21 -0700694struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
696 struct net_device *dev;
697
698 ASSERT_RTNL();
699
Denis V. Lunev81103a52007-12-12 10:47:38 -0800700 for_each_netdev(net, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 if (dev->type == type &&
702 !memcmp(dev->dev_addr, ha, dev->addr_len))
Pavel Emelianov7562f872007-05-03 15:13:45 -0700703 return dev;
704
705 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706}
707
Jochen Friedrichcf309e32005-09-22 04:44:55 -0300708EXPORT_SYMBOL(dev_getbyhwaddr);
709
Eric W. Biederman881d9662007-09-17 11:56:21 -0700710struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
Patrick McHardy4e9cac22007-05-03 03:28:13 -0700711{
712 struct net_device *dev;
713
714 ASSERT_RTNL();
Eric W. Biederman881d9662007-09-17 11:56:21 -0700715 for_each_netdev(net, dev)
Patrick McHardy4e9cac22007-05-03 03:28:13 -0700716 if (dev->type == type)
Pavel Emelianov7562f872007-05-03 15:13:45 -0700717 return dev;
718
719 return NULL;
Patrick McHardy4e9cac22007-05-03 03:28:13 -0700720}
721
722EXPORT_SYMBOL(__dev_getfirstbyhwtype);
723
Eric W. Biederman881d9662007-09-17 11:56:21 -0700724struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
726 struct net_device *dev;
727
728 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -0700729 dev = __dev_getfirstbyhwtype(net, type);
Patrick McHardy4e9cac22007-05-03 03:28:13 -0700730 if (dev)
731 dev_hold(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 rtnl_unlock();
733 return dev;
734}
735
736EXPORT_SYMBOL(dev_getfirstbyhwtype);
737
738/**
739 * dev_get_by_flags - find any device with given flags
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700740 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 * @if_flags: IFF_* values
742 * @mask: bitmask of bits in if_flags to check
743 *
744 * Search for any interface with the given flags. Returns NULL if a device
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900745 * is not found or a pointer to the device. The device returned has
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 * had a reference added and the pointer is safe until the user calls
747 * dev_put to indicate they have finished with it.
748 */
749
Eric W. Biederman881d9662007-09-17 11:56:21 -0700750struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
Pavel Emelianov7562f872007-05-03 15:13:45 -0700752 struct net_device *dev, *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Pavel Emelianov7562f872007-05-03 15:13:45 -0700754 ret = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700756 for_each_netdev(net, dev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 if (((dev->flags ^ if_flags) & mask) == 0) {
758 dev_hold(dev);
Pavel Emelianov7562f872007-05-03 15:13:45 -0700759 ret = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 break;
761 }
762 }
763 read_unlock(&dev_base_lock);
Pavel Emelianov7562f872007-05-03 15:13:45 -0700764 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765}
766
767/**
768 * dev_valid_name - check if name is okay for network device
769 * @name: name string
770 *
771 * Network device names need to be valid file names to
David S. Millerc7fa9d12006-08-15 16:34:13 -0700772 * to allow sysfs to work. We also disallow any kind of
773 * whitespace.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 */
Mitch Williamsc2373ee2005-11-09 10:34:45 -0800775int dev_valid_name(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
David S. Millerc7fa9d12006-08-15 16:34:13 -0700777 if (*name == '\0')
778 return 0;
Stephen Hemmingerb6fe17d2006-08-29 17:06:13 -0700779 if (strlen(name) >= IFNAMSIZ)
780 return 0;
David S. Millerc7fa9d12006-08-15 16:34:13 -0700781 if (!strcmp(name, ".") || !strcmp(name, ".."))
782 return 0;
783
784 while (*name) {
785 if (*name == '/' || isspace(*name))
786 return 0;
787 name++;
788 }
789 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
792/**
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200793 * __dev_alloc_name - allocate a name for a device
794 * @net: network namespace to allocate the device name in
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 * @name: name format string
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200796 * @buf: scratch buffer and result name string
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 *
798 * Passed a format string - eg "lt%d" it will try and find a suitable
Stephen Hemminger3041a062006-05-26 13:25:24 -0700799 * id. It scans list of devices to build up a free map, then chooses
800 * the first empty slot. The caller must hold the dev_base or rtnl lock
801 * while allocating the name and adding the device in order to avoid
802 * duplicates.
803 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
804 * Returns the number of the unit assigned or a negative errno code.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 */
806
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200807static int __dev_alloc_name(struct net *net, const char *name, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808{
809 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 const char *p;
811 const int max_netdevices = 8*PAGE_SIZE;
Stephen Hemmingercfcabdc2007-10-09 01:59:42 -0700812 unsigned long *inuse;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 struct net_device *d;
814
815 p = strnchr(name, IFNAMSIZ-1, '%');
816 if (p) {
817 /*
818 * Verify the string as this thing may have come from
819 * the user. There must be either one "%d" and no other "%"
820 * characters.
821 */
822 if (p[1] != 'd' || strchr(p + 2, '%'))
823 return -EINVAL;
824
825 /* Use one page as a bit array of possible slots */
Stephen Hemmingercfcabdc2007-10-09 01:59:42 -0700826 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 if (!inuse)
828 return -ENOMEM;
829
Eric W. Biederman881d9662007-09-17 11:56:21 -0700830 for_each_netdev(net, d) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 if (!sscanf(d->name, name, &i))
832 continue;
833 if (i < 0 || i >= max_netdevices)
834 continue;
835
836 /* avoid cases where sscanf is not exact inverse of printf */
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200837 snprintf(buf, IFNAMSIZ, name, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 if (!strncmp(buf, d->name, IFNAMSIZ))
839 set_bit(i, inuse);
840 }
841
842 i = find_first_zero_bit(inuse, max_netdevices);
843 free_page((unsigned long) inuse);
844 }
845
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200846 snprintf(buf, IFNAMSIZ, name, i);
847 if (!__dev_get_by_name(net, buf))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
850 /* It is possible to run out of possible slots
851 * when the name is long and there isn't enough space left
852 * for the digits, or if all bits are used.
853 */
854 return -ENFILE;
855}
856
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200857/**
858 * dev_alloc_name - allocate a name for a device
859 * @dev: device
860 * @name: name format string
861 *
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
866 * duplicates.
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
869 */
870
871int dev_alloc_name(struct net_device *dev, const char *name)
872{
873 char buf[IFNAMSIZ];
874 struct net *net;
875 int ret;
876
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +0900877 BUG_ON(!dev_net(dev));
878 net = dev_net(dev);
Eric W. Biedermanb267b172007-09-12 13:48:45 +0200879 ret = __dev_alloc_name(net, name, buf);
880 if (ret >= 0)
881 strlcpy(dev->name, buf, IFNAMSIZ);
882 return ret;
883}
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
886/**
887 * dev_change_name - change name of a device
888 * @dev: device
889 * @newname: name (or format string) must be at least IFNAMSIZ
890 *
891 * Change name of a device, can pass format strings "eth%d".
892 * for wildcarding.
893 */
Stephen Hemmingercf04a4c2008-09-30 02:22:14 -0700894int dev_change_name(struct net_device *dev, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895{
Herbert Xufcc5a032007-07-30 17:03:38 -0700896 char oldname[IFNAMSIZ];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 int err = 0;
Herbert Xufcc5a032007-07-30 17:03:38 -0700898 int ret;
Eric W. Biederman881d9662007-09-17 11:56:21 -0700899 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901 ASSERT_RTNL();
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +0900902 BUG_ON(!dev_net(dev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +0900904 net = dev_net(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if (dev->flags & IFF_UP)
906 return -EBUSY;
907
908 if (!dev_valid_name(newname))
909 return -EINVAL;
910
Stephen Hemmingerc8d90dc2007-10-26 03:53:42 -0700911 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
912 return 0;
913
Herbert Xufcc5a032007-07-30 17:03:38 -0700914 memcpy(oldname, dev->name, IFNAMSIZ);
915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 if (strchr(newname, '%')) {
917 err = dev_alloc_name(dev, newname);
918 if (err < 0)
919 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 }
Eric W. Biederman881d9662007-09-17 11:56:21 -0700921 else if (__dev_get_by_name(net, newname))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 return -EEXIST;
923 else
924 strlcpy(dev->name, newname, IFNAMSIZ);
925
Herbert Xufcc5a032007-07-30 17:03:38 -0700926rollback:
Stephen Hemminger92845ff2008-10-19 23:33:56 -0700927 ret = device_rename(&dev->dev, dev->name);
928 if (ret) {
Stephen Hemmingerdcc99772008-05-14 22:33:38 -0700929 memcpy(dev->name, oldname, IFNAMSIZ);
Stephen Hemminger92845ff2008-10-19 23:33:56 -0700930 return ret;
Stephen Hemmingerdcc99772008-05-14 22:33:38 -0700931 }
Herbert Xu7f988ea2007-07-30 16:35:46 -0700932
933 write_lock_bh(&dev_base_lock);
Eric W. Biederman92749822007-04-03 00:07:30 -0600934 hlist_del(&dev->name_hlist);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700935 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
Herbert Xu7f988ea2007-07-30 16:35:46 -0700936 write_unlock_bh(&dev_base_lock);
937
Pavel Emelyanov056925a2007-09-16 15:42:43 -0700938 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
Herbert Xufcc5a032007-07-30 17:03:38 -0700939 ret = notifier_to_errno(ret);
940
941 if (ret) {
942 if (err) {
943 printk(KERN_ERR
944 "%s: name change rollback failed: %d.\n",
945 dev->name, ret);
946 } else {
947 err = ret;
948 memcpy(dev->name, oldname, IFNAMSIZ);
949 goto rollback;
950 }
951 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
953 return err;
954}
955
956/**
Stephen Hemminger0b815a12008-09-22 21:28:11 -0700957 * dev_set_alias - change ifalias of a device
958 * @dev: device
959 * @alias: name up to IFALIASZ
Stephen Hemmingerf0db2752008-09-30 02:23:58 -0700960 * @len: limit of bytes to copy from info
Stephen Hemminger0b815a12008-09-22 21:28:11 -0700961 *
962 * Set ifalias for a device,
963 */
964int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
965{
966 ASSERT_RTNL();
967
968 if (len >= IFALIASZ)
969 return -EINVAL;
970
Oliver Hartkopp96ca4a22008-09-23 21:23:19 -0700971 if (!len) {
972 if (dev->ifalias) {
973 kfree(dev->ifalias);
974 dev->ifalias = NULL;
975 }
976 return 0;
977 }
978
Stephen Hemminger0b815a12008-09-22 21:28:11 -0700979 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
980 if (!dev->ifalias)
981 return -ENOMEM;
982
983 strlcpy(dev->ifalias, alias, len+1);
984 return len;
985}
986
987
988/**
Stephen Hemminger3041a062006-05-26 13:25:24 -0700989 * netdev_features_change - device changes features
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700990 * @dev: device to cause notification
991 *
992 * Called to indicate a device has changed features.
993 */
994void netdev_features_change(struct net_device *dev)
995{
Pavel Emelyanov056925a2007-09-16 15:42:43 -0700996 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
Stephen Hemmingerd8a33ac2005-05-29 14:13:47 -0700997}
998EXPORT_SYMBOL(netdev_features_change);
999
1000/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 * netdev_state_change - device changes state
1002 * @dev: device to cause notification
1003 *
1004 * Called to indicate a device has changed state. This function calls
1005 * the notifier chains for netdev_chain and sends a NEWLINK message
1006 * to the routing socket.
1007 */
1008void netdev_state_change(struct net_device *dev)
1009{
1010 if (dev->flags & IFF_UP) {
Pavel Emelyanov056925a2007-09-16 15:42:43 -07001011 call_netdevice_notifiers(NETDEV_CHANGE, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1013 }
1014}
1015
Or Gerlitzc1da4ac2008-06-13 18:12:00 -07001016void netdev_bonding_change(struct net_device *dev)
1017{
1018 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1019}
1020EXPORT_SYMBOL(netdev_bonding_change);
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022/**
1023 * dev_load - load a network module
Randy Dunlapc4ea43c2007-10-12 21:17:49 -07001024 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 * @name: name of interface
1026 *
1027 * If a network interface is not present and the process has suitable
1028 * privileges this function loads the module. If module loading is not
1029 * available in this kernel then it becomes a nop.
1030 */
1031
Eric W. Biederman881d9662007-09-17 11:56:21 -07001032void dev_load(struct net *net, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001034 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
1036 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -07001037 dev = __dev_get_by_name(net, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 read_unlock(&dev_base_lock);
1039
1040 if (!dev && capable(CAP_SYS_MODULE))
1041 request_module("%s", name);
1042}
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044/**
1045 * dev_open - prepare an interface for use.
1046 * @dev: device to open
1047 *
1048 * Takes a device from down to up state. The device's private open
1049 * function is invoked and then the multicast lists are loaded. Finally
1050 * the device is moved into the up state and a %NETDEV_UP message is
1051 * sent to the netdev notifier chain.
1052 *
1053 * Calling this function on an active interface is a nop. On a failure
1054 * a negative errno code is returned.
1055 */
1056int dev_open(struct net_device *dev)
1057{
1058 int ret = 0;
1059
Ben Hutchingse46b66b2008-05-08 02:53:17 -07001060 ASSERT_RTNL();
1061
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 /*
1063 * Is it already up?
1064 */
1065
1066 if (dev->flags & IFF_UP)
1067 return 0;
1068
1069 /*
1070 * Is it even present?
1071 */
1072 if (!netif_device_present(dev))
1073 return -ENODEV;
1074
1075 /*
1076 * Call device private open method
1077 */
1078 set_bit(__LINK_STATE_START, &dev->state);
Jeff Garzikbada3392007-10-23 20:19:37 -07001079
1080 if (dev->validate_addr)
1081 ret = dev->validate_addr(dev);
1082
1083 if (!ret && dev->open)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 ret = dev->open(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001086 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 * If it went open OK then:
1088 */
1089
Jeff Garzikbada3392007-10-23 20:19:37 -07001090 if (ret)
1091 clear_bit(__LINK_STATE_START, &dev->state);
1092 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 /*
1094 * Set the flags.
1095 */
1096 dev->flags |= IFF_UP;
1097
1098 /*
1099 * Initialize multicasting status
1100 */
Patrick McHardy4417da62007-06-27 01:28:10 -07001101 dev_set_rx_mode(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 /*
1104 * Wakeup transmit queue engine
1105 */
1106 dev_activate(dev);
1107
1108 /*
1109 * ... and announce new interface.
1110 */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07001111 call_netdevice_notifiers(NETDEV_UP, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 }
Jeff Garzikbada3392007-10-23 20:19:37 -07001113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 return ret;
1115}
1116
1117/**
1118 * dev_close - shutdown an interface.
1119 * @dev: device to shutdown
1120 *
1121 * This function moves an active device into down state. A
1122 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1123 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1124 * chain.
1125 */
1126int dev_close(struct net_device *dev)
1127{
Ben Hutchingse46b66b2008-05-08 02:53:17 -07001128 ASSERT_RTNL();
1129
David S. Miller9d5010d2007-09-12 14:33:25 +02001130 might_sleep();
1131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 if (!(dev->flags & IFF_UP))
1133 return 0;
1134
1135 /*
1136 * Tell people we are going down, so that they can
1137 * prepare to death, when device is still operating.
1138 */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07001139 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 clear_bit(__LINK_STATE_START, &dev->state);
1142
1143 /* Synchronize to scheduled poll. We cannot touch poll list,
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001144 * it can be even on different cpu. So just clear netif_running().
1145 *
1146 * dev->stop() will invoke napi_disable() on all of it's
1147 * napi_struct instances on this device.
1148 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 smp_mb__after_clear_bit(); /* Commit netif_running(). */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
Matti Linnanvuorid8b2a4d2008-02-12 23:10:11 -08001151 dev_deactivate(dev);
1152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 /*
1154 * Call the device specific close. This cannot fail.
1155 * Only if device is UP
1156 *
1157 * We allow it to be called even after a DETACH hot-plug
1158 * event.
1159 */
1160 if (dev->stop)
1161 dev->stop(dev);
1162
1163 /*
1164 * Device is now down.
1165 */
1166
1167 dev->flags &= ~IFF_UP;
1168
1169 /*
1170 * Tell people we are down
1171 */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07001172 call_netdevice_notifiers(NETDEV_DOWN, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
1174 return 0;
1175}
1176
1177
Ben Hutchings0187bdf2008-06-19 16:15:47 -07001178/**
1179 * dev_disable_lro - disable Large Receive Offload on a device
1180 * @dev: device
1181 *
1182 * Disable Large Receive Offload (LRO) on a net device. Must be
1183 * called under RTNL. This is needed if received packets may be
1184 * forwarded to another interface.
1185 */
1186void dev_disable_lro(struct net_device *dev)
1187{
1188 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1189 dev->ethtool_ops->set_flags) {
1190 u32 flags = dev->ethtool_ops->get_flags(dev);
1191 if (flags & ETH_FLAG_LRO) {
1192 flags &= ~ETH_FLAG_LRO;
1193 dev->ethtool_ops->set_flags(dev, flags);
1194 }
1195 }
1196 WARN_ON(dev->features & NETIF_F_LRO);
1197}
1198EXPORT_SYMBOL(dev_disable_lro);
1199
1200
Eric W. Biederman881d9662007-09-17 11:56:21 -07001201static int dev_boot_phase = 1;
1202
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203/*
1204 * Device change register/unregister. These are not inline or static
1205 * as we export them to the world.
1206 */
1207
1208/**
1209 * register_netdevice_notifier - register a network notifier block
1210 * @nb: notifier
1211 *
1212 * Register a notifier to be called when network device events occur.
1213 * The notifier passed is linked into the kernel structures and must
1214 * not be reused until it has been unregistered. A negative errno code
1215 * is returned on a failure.
1216 *
1217 * When registered all registration and up events are replayed
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001218 * to the new notifier to allow device to have a race free
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 * view of the network device list.
1220 */
1221
1222int register_netdevice_notifier(struct notifier_block *nb)
1223{
1224 struct net_device *dev;
Herbert Xufcc5a032007-07-30 17:03:38 -07001225 struct net_device *last;
Eric W. Biederman881d9662007-09-17 11:56:21 -07001226 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 int err;
1228
1229 rtnl_lock();
Alan Sternf07d5b92006-05-09 15:23:03 -07001230 err = raw_notifier_chain_register(&netdev_chain, nb);
Herbert Xufcc5a032007-07-30 17:03:38 -07001231 if (err)
1232 goto unlock;
Eric W. Biederman881d9662007-09-17 11:56:21 -07001233 if (dev_boot_phase)
1234 goto unlock;
1235 for_each_net(net) {
1236 for_each_netdev(net, dev) {
1237 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1238 err = notifier_to_errno(err);
1239 if (err)
1240 goto rollback;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Eric W. Biederman881d9662007-09-17 11:56:21 -07001242 if (!(dev->flags & IFF_UP))
1243 continue;
Herbert Xufcc5a032007-07-30 17:03:38 -07001244
Eric W. Biederman881d9662007-09-17 11:56:21 -07001245 nb->notifier_call(nb, NETDEV_UP, dev);
1246 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
Herbert Xufcc5a032007-07-30 17:03:38 -07001248
1249unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 rtnl_unlock();
1251 return err;
Herbert Xufcc5a032007-07-30 17:03:38 -07001252
1253rollback:
1254 last = dev;
Eric W. Biederman881d9662007-09-17 11:56:21 -07001255 for_each_net(net) {
1256 for_each_netdev(net, dev) {
1257 if (dev == last)
1258 break;
Herbert Xufcc5a032007-07-30 17:03:38 -07001259
Eric W. Biederman881d9662007-09-17 11:56:21 -07001260 if (dev->flags & IFF_UP) {
1261 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1262 nb->notifier_call(nb, NETDEV_DOWN, dev);
1263 }
1264 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
Herbert Xufcc5a032007-07-30 17:03:38 -07001265 }
Herbert Xufcc5a032007-07-30 17:03:38 -07001266 }
Pavel Emelyanovc67625a2007-11-14 15:53:16 -08001267
1268 raw_notifier_chain_unregister(&netdev_chain, nb);
Herbert Xufcc5a032007-07-30 17:03:38 -07001269 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270}
1271
1272/**
1273 * unregister_netdevice_notifier - unregister a network notifier block
1274 * @nb: notifier
1275 *
1276 * Unregister a notifier previously registered by
1277 * register_netdevice_notifier(). The notifier is unlinked into the
1278 * kernel structures and may then be reused. A negative errno code
1279 * is returned on a failure.
1280 */
1281
1282int unregister_netdevice_notifier(struct notifier_block *nb)
1283{
Herbert Xu9f514952006-03-25 01:24:25 -08001284 int err;
1285
1286 rtnl_lock();
Alan Sternf07d5b92006-05-09 15:23:03 -07001287 err = raw_notifier_chain_unregister(&netdev_chain, nb);
Herbert Xu9f514952006-03-25 01:24:25 -08001288 rtnl_unlock();
1289 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290}
1291
1292/**
1293 * call_netdevice_notifiers - call all network notifier blocks
1294 * @val: value passed unmodified to notifier function
Randy Dunlapc4ea43c2007-10-12 21:17:49 -07001295 * @dev: net_device pointer passed unmodified to notifier function
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 *
1297 * Call all network notifier blocks. Parameters and return value
Alan Sternf07d5b92006-05-09 15:23:03 -07001298 * are as for raw_notifier_call_chain().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 */
1300
Eric W. Biedermanad7379d2007-09-16 15:33:32 -07001301int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302{
Eric W. Biedermanad7379d2007-09-16 15:33:32 -07001303 return raw_notifier_call_chain(&netdev_chain, val, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304}
1305
1306/* When > 0 there are consumers of rx skb time stamps */
1307static atomic_t netstamp_needed = ATOMIC_INIT(0);
1308
1309void net_enable_timestamp(void)
1310{
1311 atomic_inc(&netstamp_needed);
1312}
1313
1314void net_disable_timestamp(void)
1315{
1316 atomic_dec(&netstamp_needed);
1317}
1318
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001319static inline void net_timestamp(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320{
1321 if (atomic_read(&netstamp_needed))
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001322 __net_timestamp(skb);
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07001323 else
1324 skb->tstamp.tv64 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325}
1326
1327/*
1328 * Support routine. Sends outgoing frames to any network
1329 * taps currently in use.
1330 */
1331
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001332static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
1334 struct packet_type *ptype;
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001335
1336 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338 rcu_read_lock();
1339 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1340 /* Never send packets back to the socket
1341 * they originated from - MvS (miquels@drinkel.ow.org)
1342 */
1343 if ((ptype->dev == dev || !ptype->dev) &&
1344 (ptype->af_packet_priv == NULL ||
1345 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1346 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1347 if (!skb2)
1348 break;
1349
1350 /* skb->nh should be correctly
1351 set by sender, so that the second statement is
1352 just protection against buggy protocols.
1353 */
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07001354 skb_reset_mac_header(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001356 if (skb_network_header(skb2) < skb2->data ||
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001357 skb2->network_header > skb2->tail) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 if (net_ratelimit())
1359 printk(KERN_CRIT "protocol %04x is "
1360 "buggy, dev %s\n",
1361 skb2->protocol, dev->name);
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -07001362 skb_reset_network_header(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001365 skb2->transport_header = skb2->network_header;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 skb2->pkt_type = PACKET_OUTGOING;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07001367 ptype->func(skb2, skb->dev, ptype, skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 }
1369 }
1370 rcu_read_unlock();
1371}
1372
Denis Vlasenko56079432006-03-29 15:57:29 -08001373
Jarek Poplawskidef82a12008-08-17 21:54:43 -07001374static inline void __netif_reschedule(struct Qdisc *q)
1375{
1376 struct softnet_data *sd;
1377 unsigned long flags;
1378
1379 local_irq_save(flags);
1380 sd = &__get_cpu_var(softnet_data);
1381 q->next_sched = sd->output_queue;
1382 sd->output_queue = q;
1383 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1384 local_irq_restore(flags);
1385}
1386
David S. Miller37437bb2008-07-16 02:15:04 -07001387void __netif_schedule(struct Qdisc *q)
Denis Vlasenko56079432006-03-29 15:57:29 -08001388{
Jarek Poplawskidef82a12008-08-17 21:54:43 -07001389 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1390 __netif_reschedule(q);
Denis Vlasenko56079432006-03-29 15:57:29 -08001391}
1392EXPORT_SYMBOL(__netif_schedule);
1393
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001394void dev_kfree_skb_irq(struct sk_buff *skb)
Denis Vlasenko56079432006-03-29 15:57:29 -08001395{
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001396 if (atomic_dec_and_test(&skb->users)) {
1397 struct softnet_data *sd;
1398 unsigned long flags;
Denis Vlasenko56079432006-03-29 15:57:29 -08001399
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001400 local_irq_save(flags);
1401 sd = &__get_cpu_var(softnet_data);
1402 skb->next = sd->completion_queue;
1403 sd->completion_queue = skb;
1404 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1405 local_irq_restore(flags);
1406 }
Denis Vlasenko56079432006-03-29 15:57:29 -08001407}
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001408EXPORT_SYMBOL(dev_kfree_skb_irq);
Denis Vlasenko56079432006-03-29 15:57:29 -08001409
1410void dev_kfree_skb_any(struct sk_buff *skb)
1411{
1412 if (in_irq() || irqs_disabled())
1413 dev_kfree_skb_irq(skb);
1414 else
1415 dev_kfree_skb(skb);
1416}
1417EXPORT_SYMBOL(dev_kfree_skb_any);
1418
1419
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001420/**
1421 * netif_device_detach - mark device as removed
1422 * @dev: network device
1423 *
1424 * Mark device as removed from system and therefore no longer available.
1425 */
Denis Vlasenko56079432006-03-29 15:57:29 -08001426void netif_device_detach(struct net_device *dev)
1427{
1428 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1429 netif_running(dev)) {
1430 netif_stop_queue(dev);
1431 }
1432}
1433EXPORT_SYMBOL(netif_device_detach);
1434
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001435/**
1436 * netif_device_attach - mark device as attached
1437 * @dev: network device
1438 *
1439 * Mark device as attached from system and restart if needed.
1440 */
Denis Vlasenko56079432006-03-29 15:57:29 -08001441void netif_device_attach(struct net_device *dev)
1442{
1443 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1444 netif_running(dev)) {
1445 netif_wake_queue(dev);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001446 __netdev_watchdog_up(dev);
Denis Vlasenko56079432006-03-29 15:57:29 -08001447 }
1448}
1449EXPORT_SYMBOL(netif_device_attach);
1450
Ben Hutchings6de329e2008-06-16 17:02:28 -07001451static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1452{
1453 return ((features & NETIF_F_GEN_CSUM) ||
1454 ((features & NETIF_F_IP_CSUM) &&
1455 protocol == htons(ETH_P_IP)) ||
1456 ((features & NETIF_F_IPV6_CSUM) &&
1457 protocol == htons(ETH_P_IPV6)));
1458}
1459
1460static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1461{
1462 if (can_checksum_protocol(dev->features, skb->protocol))
1463 return true;
1464
1465 if (skb->protocol == htons(ETH_P_8021Q)) {
1466 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1467 if (can_checksum_protocol(dev->features & dev->vlan_features,
1468 veh->h_vlan_encapsulated_proto))
1469 return true;
1470 }
1471
1472 return false;
1473}
Denis Vlasenko56079432006-03-29 15:57:29 -08001474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475/*
1476 * Invalidate hardware checksum when packet is to be mangled, and
1477 * complete checksum manually on outgoing path.
1478 */
Patrick McHardy84fa7932006-08-29 16:44:56 -07001479int skb_checksum_help(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480{
Al Virod3bc23e2006-11-14 21:24:49 -08001481 __wsum csum;
Herbert Xu663ead32007-04-09 11:59:07 -07001482 int ret = 0, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483
Patrick McHardy84fa7932006-08-29 16:44:56 -07001484 if (skb->ip_summed == CHECKSUM_COMPLETE)
Herbert Xua430a432006-07-08 13:34:56 -07001485 goto out_set_summed;
1486
1487 if (unlikely(skb_shinfo(skb)->gso_size)) {
Herbert Xua430a432006-07-08 13:34:56 -07001488 /* Let GSO fix up the checksum. */
1489 goto out_set_summed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 }
1491
Herbert Xua0308472007-10-15 01:47:15 -07001492 offset = skb->csum_start - skb_headroom(skb);
1493 BUG_ON(offset >= skb_headlen(skb));
1494 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1495
1496 offset += skb->csum_offset;
1497 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1498
1499 if (skb_cloned(skb) &&
1500 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1502 if (ret)
1503 goto out;
1504 }
1505
Herbert Xua0308472007-10-15 01:47:15 -07001506 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
Herbert Xua430a432006-07-08 13:34:56 -07001507out_set_summed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 skb->ip_summed = CHECKSUM_NONE;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001509out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 return ret;
1511}
1512
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001513/**
1514 * skb_gso_segment - Perform segmentation on skb.
1515 * @skb: buffer to segment
Herbert Xu576a30e2006-06-27 13:22:38 -07001516 * @features: features for the output path (see dev->features)
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001517 *
1518 * This function segments the given skb and returns a list of segments.
Herbert Xu576a30e2006-06-27 13:22:38 -07001519 *
1520 * It may return NULL if the skb requires no segmentation. This is
1521 * only possible when GSO is used for verifying header integrity.
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001522 */
Herbert Xu576a30e2006-06-27 13:22:38 -07001523struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001524{
1525 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1526 struct packet_type *ptype;
Al Viro252e3342006-11-14 20:48:11 -08001527 __be16 type = skb->protocol;
Herbert Xua430a432006-07-08 13:34:56 -07001528 int err;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001529
1530 BUG_ON(skb_shinfo(skb)->frag_list);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001531
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07001532 skb_reset_mac_header(skb);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001533 skb->mac_len = skb->network_header - skb->mac_header;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001534 __skb_pull(skb, skb->mac_len);
1535
Herbert Xuf9d106a2007-04-23 22:36:13 -07001536 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
Herbert Xua430a432006-07-08 13:34:56 -07001537 if (skb_header_cloned(skb) &&
1538 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1539 return ERR_PTR(err);
1540 }
1541
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001542 rcu_read_lock();
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08001543 list_for_each_entry_rcu(ptype,
1544 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001545 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
Patrick McHardy84fa7932006-08-29 16:44:56 -07001546 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
Herbert Xua430a432006-07-08 13:34:56 -07001547 err = ptype->gso_send_check(skb);
1548 segs = ERR_PTR(err);
1549 if (err || skb_gso_ok(skb, features))
1550 break;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001551 __skb_push(skb, (skb->data -
1552 skb_network_header(skb)));
Herbert Xua430a432006-07-08 13:34:56 -07001553 }
Herbert Xu576a30e2006-06-27 13:22:38 -07001554 segs = ptype->gso_segment(skb, features);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001555 break;
1556 }
1557 }
1558 rcu_read_unlock();
1559
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001560 __skb_push(skb, skb->data - skb_mac_header(skb));
Herbert Xu576a30e2006-06-27 13:22:38 -07001561
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001562 return segs;
1563}
1564
1565EXPORT_SYMBOL(skb_gso_segment);
1566
Herbert Xufb286bb2005-11-10 13:01:24 -08001567/* Take action when hardware reception checksum errors are detected. */
1568#ifdef CONFIG_BUG
1569void netdev_rx_csum_fault(struct net_device *dev)
1570{
1571 if (net_ratelimit()) {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001572 printk(KERN_ERR "%s: hw csum failure.\n",
Stephen Hemminger246a4212005-12-08 15:21:39 -08001573 dev ? dev->name : "<unknown>");
Herbert Xufb286bb2005-11-10 13:01:24 -08001574 dump_stack();
1575 }
1576}
1577EXPORT_SYMBOL(netdev_rx_csum_fault);
1578#endif
1579
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580/* Actually, we should eliminate this check as soon as we know, that:
1581 * 1. IOMMU is present and allows to map all the memory.
1582 * 2. No high memory really exists on this machine.
1583 */
1584
1585static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1586{
Herbert Xu3d3a8532006-06-27 13:33:10 -07001587#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 int i;
1589
1590 if (dev->features & NETIF_F_HIGHDMA)
1591 return 0;
1592
1593 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1594 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1595 return 1;
1596
Herbert Xu3d3a8532006-06-27 13:33:10 -07001597#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 return 0;
1599}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001601struct dev_gso_cb {
1602 void (*destructor)(struct sk_buff *skb);
1603};
1604
1605#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1606
1607static void dev_gso_skb_destructor(struct sk_buff *skb)
1608{
1609 struct dev_gso_cb *cb;
1610
1611 do {
1612 struct sk_buff *nskb = skb->next;
1613
1614 skb->next = nskb->next;
1615 nskb->next = NULL;
1616 kfree_skb(nskb);
1617 } while (skb->next);
1618
1619 cb = DEV_GSO_CB(skb);
1620 if (cb->destructor)
1621 cb->destructor(skb);
1622}
1623
1624/**
1625 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1626 * @skb: buffer to segment
1627 *
1628 * This function segments the given skb and stores the list of segments
1629 * in skb->next.
1630 */
1631static int dev_gso_segment(struct sk_buff *skb)
1632{
1633 struct net_device *dev = skb->dev;
1634 struct sk_buff *segs;
Herbert Xu576a30e2006-06-27 13:22:38 -07001635 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1636 NETIF_F_SG : 0);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001637
Herbert Xu576a30e2006-06-27 13:22:38 -07001638 segs = skb_gso_segment(skb, features);
1639
1640 /* Verifying header integrity only. */
1641 if (!segs)
1642 return 0;
1643
Hirofumi Nakagawa801678c2008-04-29 01:03:09 -07001644 if (IS_ERR(segs))
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001645 return PTR_ERR(segs);
1646
1647 skb->next = segs;
1648 DEV_GSO_CB(skb)->destructor = skb->destructor;
1649 skb->destructor = dev_gso_skb_destructor;
1650
1651 return 0;
1652}
1653
David S. Millerfd2ea0a2008-07-17 01:56:23 -07001654int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1655 struct netdev_queue *txq)
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001656{
1657 if (likely(!skb->next)) {
Stephen Hemminger9be9a6b2007-04-20 17:02:45 -07001658 if (!list_empty(&ptype_all))
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001659 dev_queue_xmit_nit(skb, dev);
1660
Herbert Xu576a30e2006-06-27 13:22:38 -07001661 if (netif_needs_gso(dev, skb)) {
1662 if (unlikely(dev_gso_segment(skb)))
1663 goto out_kfree_skb;
1664 if (skb->next)
1665 goto gso;
1666 }
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001667
Herbert Xu576a30e2006-06-27 13:22:38 -07001668 return dev->hard_start_xmit(skb, dev);
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001669 }
1670
Herbert Xu576a30e2006-06-27 13:22:38 -07001671gso:
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001672 do {
1673 struct sk_buff *nskb = skb->next;
1674 int rc;
1675
1676 skb->next = nskb->next;
1677 nskb->next = NULL;
1678 rc = dev->hard_start_xmit(nskb, dev);
1679 if (unlikely(rc)) {
Michael Chanf54d9e82006-06-25 23:57:04 -07001680 nskb->next = skb->next;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001681 skb->next = nskb;
1682 return rc;
1683 }
David S. Millerfd2ea0a2008-07-17 01:56:23 -07001684 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
Michael Chanf54d9e82006-06-25 23:57:04 -07001685 return NETDEV_TX_BUSY;
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001686 } while (skb->next);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001687
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001688 skb->destructor = DEV_GSO_CB(skb)->destructor;
1689
1690out_kfree_skb:
1691 kfree_skb(skb);
1692 return 0;
1693}
1694
David S. Millerb6b2fed2008-07-21 09:48:06 -07001695static u32 simple_tx_hashrnd;
1696static int simple_tx_hashrnd_initialized = 0;
1697
David S. Miller8f0f2222008-07-15 03:47:03 -07001698static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb)
1699{
David S. Millerb6b2fed2008-07-21 09:48:06 -07001700 u32 addr1, addr2, ports;
1701 u32 hash, ihl;
Alexander Duyckad55dca2008-09-20 22:05:50 -07001702 u8 ip_proto = 0;
David S. Millerb6b2fed2008-07-21 09:48:06 -07001703
1704 if (unlikely(!simple_tx_hashrnd_initialized)) {
1705 get_random_bytes(&simple_tx_hashrnd, 4);
1706 simple_tx_hashrnd_initialized = 1;
1707 }
David S. Miller8f0f2222008-07-15 03:47:03 -07001708
1709 switch (skb->protocol) {
Arnaldo Carvalho de Melo60678042008-09-20 22:20:49 -07001710 case htons(ETH_P_IP):
Alexander Duyckad55dca2008-09-20 22:05:50 -07001711 if (!(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)))
1712 ip_proto = ip_hdr(skb)->protocol;
David S. Millerb6b2fed2008-07-21 09:48:06 -07001713 addr1 = ip_hdr(skb)->saddr;
1714 addr2 = ip_hdr(skb)->daddr;
David S. Miller8f0f2222008-07-15 03:47:03 -07001715 ihl = ip_hdr(skb)->ihl;
David S. Miller8f0f2222008-07-15 03:47:03 -07001716 break;
Arnaldo Carvalho de Melo60678042008-09-20 22:20:49 -07001717 case htons(ETH_P_IPV6):
David S. Miller8f0f2222008-07-15 03:47:03 -07001718 ip_proto = ipv6_hdr(skb)->nexthdr;
David S. Millerb6b2fed2008-07-21 09:48:06 -07001719 addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3];
1720 addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3];
David S. Miller8f0f2222008-07-15 03:47:03 -07001721 ihl = (40 >> 2);
David S. Miller8f0f2222008-07-15 03:47:03 -07001722 break;
1723 default:
1724 return 0;
1725 }
1726
David S. Miller8f0f2222008-07-15 03:47:03 -07001727
1728 switch (ip_proto) {
1729 case IPPROTO_TCP:
1730 case IPPROTO_UDP:
1731 case IPPROTO_DCCP:
1732 case IPPROTO_ESP:
1733 case IPPROTO_AH:
1734 case IPPROTO_SCTP:
1735 case IPPROTO_UDPLITE:
David S. Millerb6b2fed2008-07-21 09:48:06 -07001736 ports = *((u32 *) (skb_network_header(skb) + (ihl * 4)));
David S. Miller8f0f2222008-07-15 03:47:03 -07001737 break;
1738
1739 default:
David S. Millerb6b2fed2008-07-21 09:48:06 -07001740 ports = 0;
David S. Miller8f0f2222008-07-15 03:47:03 -07001741 break;
1742 }
1743
David S. Millerb6b2fed2008-07-21 09:48:06 -07001744 hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd);
1745
1746 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
David S. Miller8f0f2222008-07-15 03:47:03 -07001747}
1748
David S. Millere8a04642008-07-17 00:34:19 -07001749static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1750 struct sk_buff *skb)
1751{
David S. Millerfd2ea0a2008-07-17 01:56:23 -07001752 u16 queue_index = 0;
1753
David S. Millereae792b2008-07-15 03:03:33 -07001754 if (dev->select_queue)
1755 queue_index = dev->select_queue(dev, skb);
David S. Miller8f0f2222008-07-15 03:47:03 -07001756 else if (dev->real_num_tx_queues > 1)
1757 queue_index = simple_tx_hash(dev, skb);
David S. Millereae792b2008-07-15 03:03:33 -07001758
David S. Millerfd2ea0a2008-07-17 01:56:23 -07001759 skb_set_queue_mapping(skb, queue_index);
1760 return netdev_get_tx_queue(dev, queue_index);
David S. Millere8a04642008-07-17 00:34:19 -07001761}
1762
Dave Jonesd29f7492008-07-22 14:09:06 -07001763/**
1764 * dev_queue_xmit - transmit a buffer
1765 * @skb: buffer to transmit
1766 *
1767 * Queue a buffer for transmission to a network device. The caller must
1768 * have set the device and priority and built the buffer before calling
1769 * this function. The function can be called from an interrupt.
1770 *
1771 * A negative errno code is returned on a failure. A success does not
1772 * guarantee the frame will be transmitted as it may be dropped due
1773 * to congestion or traffic shaping.
1774 *
1775 * -----------------------------------------------------------------------------------
1776 * I notice this method can also return errors from the queue disciplines,
1777 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1778 * be positive.
1779 *
1780 * Regardless of the return value, the skb is consumed, so it is currently
1781 * difficult to retry a send to this method. (You can bump the ref count
1782 * before sending to hold a reference for retry if you are careful.)
1783 *
1784 * When calling this method, interrupts MUST be enabled. This is because
1785 * the BH enable code must have IRQs enabled so that it will not deadlock.
1786 * --BLG
1787 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788int dev_queue_xmit(struct sk_buff *skb)
1789{
1790 struct net_device *dev = skb->dev;
David S. Millerdc2b4842008-07-08 17:18:23 -07001791 struct netdev_queue *txq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 struct Qdisc *q;
1793 int rc = -ENOMEM;
1794
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001795 /* GSO will handle the following emulations directly. */
1796 if (netif_needs_gso(dev, skb))
1797 goto gso;
1798
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 if (skb_shinfo(skb)->frag_list &&
1800 !(dev->features & NETIF_F_FRAGLIST) &&
Herbert Xu364c6ba2006-06-09 16:10:40 -07001801 __skb_linearize(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 goto out_kfree_skb;
1803
1804 /* Fragmented skb is linearized if device does not support SG,
1805 * or if at least one of fragments is in highmem and device
1806 * does not support DMA from it.
1807 */
1808 if (skb_shinfo(skb)->nr_frags &&
1809 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
Herbert Xu364c6ba2006-06-09 16:10:40 -07001810 __skb_linearize(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 goto out_kfree_skb;
1812
1813 /* If packet is not checksummed and device does not support
1814 * checksumming for this protocol, complete checksumming here.
1815 */
Herbert Xu663ead32007-04-09 11:59:07 -07001816 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1817 skb_set_transport_header(skb, skb->csum_start -
1818 skb_headroom(skb));
Ben Hutchings6de329e2008-06-16 17:02:28 -07001819 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1820 goto out_kfree_skb;
Herbert Xu663ead32007-04-09 11:59:07 -07001821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Herbert Xuf6a78bf2006-06-22 02:57:17 -07001823gso:
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001824 /* Disable soft irqs for various locks below. Also
1825 * stops preemption for RCU.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001827 rcu_read_lock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
David S. Millereae792b2008-07-15 03:03:33 -07001829 txq = dev_pick_tx(dev, skb);
David S. Millerb0e1e642008-07-08 17:42:10 -07001830 q = rcu_dereference(txq->qdisc);
David S. Miller37437bb2008-07-16 02:15:04 -07001831
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832#ifdef CONFIG_NET_CLS_ACT
1833 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1834#endif
1835 if (q->enqueue) {
David S. Miller5fb66222008-08-02 20:02:43 -07001836 spinlock_t *root_lock = qdisc_lock(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
David S. Miller37437bb2008-07-16 02:15:04 -07001838 spin_lock(root_lock);
1839
David S. Millera9312ae2008-08-17 21:51:03 -07001840 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
David S. Miller96d20312008-08-17 23:37:16 -07001841 kfree_skb(skb);
David S. Millera9312ae2008-08-17 21:51:03 -07001842 rc = NET_XMIT_DROP;
David S. Miller96d20312008-08-17 23:37:16 -07001843 } else {
1844 rc = qdisc_enqueue_root(skb, q);
1845 qdisc_run(q);
David S. Millera9312ae2008-08-17 21:51:03 -07001846 }
David S. Miller37437bb2008-07-16 02:15:04 -07001847 spin_unlock(root_lock);
1848
David S. Miller37437bb2008-07-16 02:15:04 -07001849 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 }
1851
1852 /* The device has no queue. Common case for software devices:
1853 loopback, all the sorts of tunnels...
1854
Herbert Xu932ff272006-06-09 12:20:56 -07001855 Really, it is unlikely that netif_tx_lock protection is necessary
1856 here. (f.e. loopback and IP tunnels are clean ignoring statistics
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 counters.)
1858 However, it is possible, that they rely on protection
1859 made by us here.
1860
1861 Check this and shot the lock. It is not prone from deadlocks.
1862 Either shot noqueue qdisc, it is even simpler 8)
1863 */
1864 if (dev->flags & IFF_UP) {
1865 int cpu = smp_processor_id(); /* ok because BHs are off */
1866
David S. Millerc773e842008-07-08 23:13:53 -07001867 if (txq->xmit_lock_owner != cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
David S. Millerc773e842008-07-08 23:13:53 -07001869 HARD_TX_LOCK(dev, txq, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
David S. Millerfd2ea0a2008-07-17 01:56:23 -07001871 if (!netif_tx_queue_stopped(txq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 rc = 0;
David S. Millerfd2ea0a2008-07-17 01:56:23 -07001873 if (!dev_hard_start_xmit(skb, dev, txq)) {
David S. Millerc773e842008-07-08 23:13:53 -07001874 HARD_TX_UNLOCK(dev, txq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 goto out;
1876 }
1877 }
David S. Millerc773e842008-07-08 23:13:53 -07001878 HARD_TX_UNLOCK(dev, txq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 if (net_ratelimit())
1880 printk(KERN_CRIT "Virtual device %s asks to "
1881 "queue packet!\n", dev->name);
1882 } else {
1883 /* Recursion is detected! It is possible,
1884 * unfortunately */
1885 if (net_ratelimit())
1886 printk(KERN_CRIT "Dead loop on virtual device "
1887 "%s, fix it urgently!\n", dev->name);
1888 }
1889 }
1890
1891 rc = -ENETDOWN;
Herbert Xud4828d82006-06-22 02:28:18 -07001892 rcu_read_unlock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
1894out_kfree_skb:
1895 kfree_skb(skb);
1896 return rc;
1897out:
Herbert Xud4828d82006-06-22 02:28:18 -07001898 rcu_read_unlock_bh();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 return rc;
1900}
1901
1902
1903/*=======================================================================
1904 Receiver routines
1905 =======================================================================*/
1906
Stephen Hemminger6b2bedc2007-03-12 14:33:50 -07001907int netdev_max_backlog __read_mostly = 1000;
1908int netdev_budget __read_mostly = 300;
1909int weight_p __read_mostly = 64; /* old backlog weight */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
1911DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1912
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914/**
1915 * netif_rx - post buffer to the network code
1916 * @skb: buffer to post
1917 *
1918 * This function receives a packet from a device driver and queues it for
1919 * the upper (protocol) levels to process. It always succeeds. The buffer
1920 * may be dropped during processing for congestion control or by the
1921 * protocol layers.
1922 *
1923 * return values:
1924 * NET_RX_SUCCESS (no congestion)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 * NET_RX_DROP (packet was dropped)
1926 *
1927 */
1928
1929int netif_rx(struct sk_buff *skb)
1930{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 struct softnet_data *queue;
1932 unsigned long flags;
1933
1934 /* if netpoll wants it, pretend we never saw it */
1935 if (netpoll_rx(skb))
1936 return NET_RX_DROP;
1937
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07001938 if (!skb->tstamp.tv64)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07001939 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
1941 /*
1942 * The code is rearranged so that the path is the most
1943 * short when CPU is congested, but is still operating.
1944 */
1945 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 queue = &__get_cpu_var(softnet_data);
1947
1948 __get_cpu_var(netdev_rx_stat).total++;
1949 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1950 if (queue->input_pkt_queue.qlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951enqueue:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 __skb_queue_tail(&queue->input_pkt_queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 local_irq_restore(flags);
Stephen Hemminger34008d82005-06-23 20:10:00 -07001954 return NET_RX_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 }
1956
Stephen Hemmingerbea33482007-10-03 16:41:36 -07001957 napi_schedule(&queue->backlog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 goto enqueue;
1959 }
1960
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 __get_cpu_var(netdev_rx_stat).dropped++;
1962 local_irq_restore(flags);
1963
1964 kfree_skb(skb);
1965 return NET_RX_DROP;
1966}
1967
1968int netif_rx_ni(struct sk_buff *skb)
1969{
1970 int err;
1971
1972 preempt_disable();
1973 err = netif_rx(skb);
1974 if (local_softirq_pending())
1975 do_softirq();
1976 preempt_enable();
1977
1978 return err;
1979}
1980
1981EXPORT_SYMBOL(netif_rx_ni);
1982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983static void net_tx_action(struct softirq_action *h)
1984{
1985 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1986
1987 if (sd->completion_queue) {
1988 struct sk_buff *clist;
1989
1990 local_irq_disable();
1991 clist = sd->completion_queue;
1992 sd->completion_queue = NULL;
1993 local_irq_enable();
1994
1995 while (clist) {
1996 struct sk_buff *skb = clist;
1997 clist = clist->next;
1998
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001999 WARN_ON(atomic_read(&skb->users));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 __kfree_skb(skb);
2001 }
2002 }
2003
2004 if (sd->output_queue) {
David S. Miller37437bb2008-07-16 02:15:04 -07002005 struct Qdisc *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
2007 local_irq_disable();
2008 head = sd->output_queue;
2009 sd->output_queue = NULL;
2010 local_irq_enable();
2011
2012 while (head) {
David S. Miller37437bb2008-07-16 02:15:04 -07002013 struct Qdisc *q = head;
2014 spinlock_t *root_lock;
2015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 head = head->next_sched;
2017
David S. Miller5fb66222008-08-02 20:02:43 -07002018 root_lock = qdisc_lock(q);
David S. Miller37437bb2008-07-16 02:15:04 -07002019 if (spin_trylock(root_lock)) {
Jarek Poplawskidef82a12008-08-17 21:54:43 -07002020 smp_mb__before_clear_bit();
2021 clear_bit(__QDISC_STATE_SCHED,
2022 &q->state);
David S. Miller37437bb2008-07-16 02:15:04 -07002023 qdisc_run(q);
2024 spin_unlock(root_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 } else {
David S. Miller195648b2008-08-19 04:00:36 -07002026 if (!test_bit(__QDISC_STATE_DEACTIVATED,
Jarek Poplawskie8a83e12008-09-07 18:41:21 -07002027 &q->state)) {
David S. Miller195648b2008-08-19 04:00:36 -07002028 __netif_reschedule(q);
Jarek Poplawskie8a83e12008-09-07 18:41:21 -07002029 } else {
2030 smp_mb__before_clear_bit();
2031 clear_bit(__QDISC_STATE_SCHED,
2032 &q->state);
2033 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 }
2035 }
2036 }
2037}
2038
Stephen Hemminger6f05f622007-03-08 20:46:03 -08002039static inline int deliver_skb(struct sk_buff *skb,
2040 struct packet_type *pt_prev,
2041 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042{
2043 atomic_inc(&skb->users);
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002044 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045}
2046
2047#if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
Stephen Hemminger6229e362007-03-21 13:38:47 -07002048/* These hooks defined here for ATM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049struct net_bridge;
2050struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2051 unsigned char *addr);
Stephen Hemminger6229e362007-03-21 13:38:47 -07002052void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Stephen Hemminger6229e362007-03-21 13:38:47 -07002054/*
2055 * If bridge module is loaded call bridging hook.
2056 * returns NULL if packet was consumed.
2057 */
2058struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2059 struct sk_buff *skb) __read_mostly;
2060static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2061 struct packet_type **pt_prev, int *ret,
2062 struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063{
2064 struct net_bridge_port *port;
2065
Stephen Hemminger6229e362007-03-21 13:38:47 -07002066 if (skb->pkt_type == PACKET_LOOPBACK ||
2067 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2068 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
2070 if (*pt_prev) {
Stephen Hemminger6229e362007-03-21 13:38:47 -07002071 *ret = deliver_skb(skb, *pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 *pt_prev = NULL;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002073 }
2074
Stephen Hemminger6229e362007-03-21 13:38:47 -07002075 return br_handle_frame_hook(port, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076}
2077#else
Stephen Hemminger6229e362007-03-21 13:38:47 -07002078#define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079#endif
2080
Patrick McHardyb863ceb2007-07-14 18:55:06 -07002081#if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2082struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2083EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2084
2085static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2086 struct packet_type **pt_prev,
2087 int *ret,
2088 struct net_device *orig_dev)
2089{
2090 if (skb->dev->macvlan_port == NULL)
2091 return skb;
2092
2093 if (*pt_prev) {
2094 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2095 *pt_prev = NULL;
2096 }
2097 return macvlan_handle_frame_hook(skb);
2098}
2099#else
2100#define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2101#endif
2102
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103#ifdef CONFIG_NET_CLS_ACT
2104/* TODO: Maybe we should just force sch_ingress to be compiled in
2105 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2106 * a compare and 2 stores extra right now if we dont have it on
2107 * but have CONFIG_NET_CLS_ACT
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002108 * NOTE: This doesnt stop any functionality; if you dont have
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 * the ingress scheduler, you just cant add policies on ingress.
2110 *
2111 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002112static int ing_filter(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 struct net_device *dev = skb->dev;
Herbert Xuf697c3e2007-10-14 00:38:47 -07002115 u32 ttl = G_TC_RTTL(skb->tc_verd);
David S. Miller555353c2008-07-08 17:33:13 -07002116 struct netdev_queue *rxq;
2117 int result = TC_ACT_OK;
2118 struct Qdisc *q;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002119
Herbert Xuf697c3e2007-10-14 00:38:47 -07002120 if (MAX_RED_LOOP < ttl++) {
2121 printk(KERN_WARNING
2122 "Redir loop detected Dropping packet (%d->%d)\n",
2123 skb->iif, dev->ifindex);
2124 return TC_ACT_SHOT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 }
2126
Herbert Xuf697c3e2007-10-14 00:38:47 -07002127 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2128 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2129
David S. Miller555353c2008-07-08 17:33:13 -07002130 rxq = &dev->rx_queue;
2131
David S. Miller83874002008-07-17 00:53:03 -07002132 q = rxq->qdisc;
David S. Miller8d50b532008-07-30 02:37:46 -07002133 if (q != &noop_qdisc) {
David S. Miller83874002008-07-17 00:53:03 -07002134 spin_lock(qdisc_lock(q));
David S. Millera9312ae2008-08-17 21:51:03 -07002135 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2136 result = qdisc_enqueue_root(skb, q);
David S. Miller83874002008-07-17 00:53:03 -07002137 spin_unlock(qdisc_lock(q));
2138 }
Herbert Xuf697c3e2007-10-14 00:38:47 -07002139
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 return result;
2141}
Herbert Xuf697c3e2007-10-14 00:38:47 -07002142
2143static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2144 struct packet_type **pt_prev,
2145 int *ret, struct net_device *orig_dev)
2146{
David S. Miller8d50b532008-07-30 02:37:46 -07002147 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
Herbert Xuf697c3e2007-10-14 00:38:47 -07002148 goto out;
2149
2150 if (*pt_prev) {
2151 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2152 *pt_prev = NULL;
2153 } else {
2154 /* Huh? Why does turning on AF_PACKET affect this? */
2155 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2156 }
2157
2158 switch (ing_filter(skb)) {
2159 case TC_ACT_SHOT:
2160 case TC_ACT_STOLEN:
2161 kfree_skb(skb);
2162 return NULL;
2163 }
2164
2165out:
2166 skb->tc_verd = 0;
2167 return skb;
2168}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169#endif
2170
Patrick McHardybc1d0412008-07-14 22:49:30 -07002171/*
2172 * netif_nit_deliver - deliver received packets to network taps
2173 * @skb: buffer
2174 *
2175 * This function is used to deliver incoming packets to network
2176 * taps. It should be used when the normal netif_receive_skb path
2177 * is bypassed, for example because of VLAN acceleration.
2178 */
2179void netif_nit_deliver(struct sk_buff *skb)
2180{
2181 struct packet_type *ptype;
2182
2183 if (list_empty(&ptype_all))
2184 return;
2185
2186 skb_reset_network_header(skb);
2187 skb_reset_transport_header(skb);
2188 skb->mac_len = skb->network_header - skb->mac_header;
2189
2190 rcu_read_lock();
2191 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2192 if (!ptype->dev || ptype->dev == skb->dev)
2193 deliver_skb(skb, ptype, skb->dev);
2194 }
2195 rcu_read_unlock();
2196}
2197
Stephen Hemminger3b582cc2007-11-01 02:21:47 -07002198/**
2199 * netif_receive_skb - process receive buffer from network
2200 * @skb: buffer to process
2201 *
2202 * netif_receive_skb() is the main receive data processing function.
2203 * It always succeeds. The buffer may be dropped during processing
2204 * for congestion control or by the protocol layers.
2205 *
2206 * This function may only be called from softirq context and interrupts
2207 * should be enabled.
2208 *
2209 * Return values (usually ignored):
2210 * NET_RX_SUCCESS: no congestion
2211 * NET_RX_DROP: packet was dropped
2212 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213int netif_receive_skb(struct sk_buff *skb)
2214{
2215 struct packet_type *ptype, *pt_prev;
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002216 struct net_device *orig_dev;
Joe Eykholt0d7a3682008-07-02 18:22:01 -07002217 struct net_device *null_or_orig;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 int ret = NET_RX_DROP;
Al Viro252e3342006-11-14 20:48:11 -08002219 __be16 type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
2221 /* if we've gotten here through NAPI, check netpoll */
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002222 if (netpoll_receive_skb(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 return NET_RX_DROP;
2224
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002225 if (!skb->tstamp.tv64)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002226 net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
Patrick McHardyc01003c2007-03-29 11:46:52 -07002228 if (!skb->iif)
2229 skb->iif = skb->dev->ifindex;
David S. Miller86e65da2005-08-09 19:36:29 -07002230
Joe Eykholt0d7a3682008-07-02 18:22:01 -07002231 null_or_orig = NULL;
Joe Eykholtcc9bd5c2008-07-02 18:22:00 -07002232 orig_dev = skb->dev;
2233 if (orig_dev->master) {
Joe Eykholt0d7a3682008-07-02 18:22:01 -07002234 if (skb_bond_should_drop(skb))
2235 null_or_orig = orig_dev; /* deliver only exact match */
2236 else
2237 skb->dev = orig_dev->master;
Joe Eykholtcc9bd5c2008-07-02 18:22:00 -07002238 }
Jay Vosburgh8f903c72006-02-21 16:36:44 -08002239
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 __get_cpu_var(netdev_rx_stat).total++;
2241
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -07002242 skb_reset_network_header(skb);
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -03002243 skb_reset_transport_header(skb);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07002244 skb->mac_len = skb->network_header - skb->mac_header;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
2246 pt_prev = NULL;
2247
2248 rcu_read_lock();
2249
Eric W. Biedermanb9f75f42008-06-20 22:16:51 -07002250 /* Don't receive packets in an exiting network namespace */
2251 if (!net_alive(dev_net(skb->dev)))
2252 goto out;
2253
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254#ifdef CONFIG_NET_CLS_ACT
2255 if (skb->tc_verd & TC_NCLS) {
2256 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2257 goto ncls;
2258 }
2259#endif
2260
2261 list_for_each_entry_rcu(ptype, &ptype_all, list) {
Joe Eykholtf9823072008-07-02 18:22:02 -07002262 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2263 ptype->dev == orig_dev) {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002264 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002265 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 pt_prev = ptype;
2267 }
2268 }
2269
2270#ifdef CONFIG_NET_CLS_ACT
Herbert Xuf697c3e2007-10-14 00:38:47 -07002271 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2272 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274ncls:
2275#endif
2276
Stephen Hemminger6229e362007-03-21 13:38:47 -07002277 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2278 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 goto out;
Patrick McHardyb863ceb2007-07-14 18:55:06 -07002280 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2281 if (!skb)
2282 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283
2284 type = skb->protocol;
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08002285 list_for_each_entry_rcu(ptype,
2286 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 if (ptype->type == type &&
Joe Eykholtf9823072008-07-02 18:22:02 -07002288 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2289 ptype->dev == orig_dev)) {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002290 if (pt_prev)
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002291 ret = deliver_skb(skb, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 pt_prev = ptype;
2293 }
2294 }
2295
2296 if (pt_prev) {
David S. Millerf2ccd8f2005-08-09 19:34:12 -07002297 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 } else {
2299 kfree_skb(skb);
2300 /* Jamal, now you will not able to escape explaining
2301 * me how you were going to use this. :-)
2302 */
2303 ret = NET_RX_DROP;
2304 }
2305
2306out:
2307 rcu_read_unlock();
2308 return ret;
2309}
2310
Stephen Hemminger6e583ce2008-08-03 21:29:57 -07002311/* Network device is going away, flush any packets still pending */
2312static void flush_backlog(void *arg)
2313{
2314 struct net_device *dev = arg;
2315 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2316 struct sk_buff *skb, *tmp;
2317
2318 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2319 if (skb->dev == dev) {
2320 __skb_unlink(skb, &queue->input_pkt_queue);
2321 kfree_skb(skb);
2322 }
2323}
2324
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002325static int process_backlog(struct napi_struct *napi, int quota)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326{
2327 int work = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2329 unsigned long start_time = jiffies;
2330
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002331 napi->weight = weight_p;
2332 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 struct sk_buff *skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334
2335 local_irq_disable();
2336 skb = __skb_dequeue(&queue->input_pkt_queue);
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002337 if (!skb) {
2338 __napi_complete(napi);
2339 local_irq_enable();
2340 break;
2341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 local_irq_enable();
2343
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 netif_receive_skb(skb);
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002345 } while (++work < quota && jiffies == start_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002347 return work;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348}
2349
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002350/**
2351 * __napi_schedule - schedule for receive
Randy Dunlapc4ea43c2007-10-12 21:17:49 -07002352 * @n: entry to schedule
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002353 *
2354 * The entry's receive function will be scheduled to run
2355 */
Harvey Harrisonb5606c22008-02-13 15:03:16 -08002356void __napi_schedule(struct napi_struct *n)
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002357{
2358 unsigned long flags;
2359
2360 local_irq_save(flags);
2361 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2362 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2363 local_irq_restore(flags);
2364}
2365EXPORT_SYMBOL(__napi_schedule);
2366
2367
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368static void net_rx_action(struct softirq_action *h)
2369{
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002370 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 unsigned long start_time = jiffies;
Stephen Hemminger51b0bde2005-06-23 20:14:40 -07002372 int budget = netdev_budget;
Matt Mackall53fb95d2005-08-11 19:27:43 -07002373 void *have;
2374
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 local_irq_disable();
2376
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002377 while (!list_empty(list)) {
2378 struct napi_struct *n;
2379 int work, weight;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002381 /* If softirq window is exhuasted then punt.
2382 *
2383 * Note that this is a slight policy change from the
2384 * previous NAPI code, which would allow up to 2
2385 * jiffies to pass before breaking out. The test
2386 * used to be "jiffies - start_time > 1".
2387 */
2388 if (unlikely(budget <= 0 || jiffies != start_time))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 goto softnet_break;
2390
2391 local_irq_enable();
2392
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002393 /* Even though interrupts have been re-enabled, this
2394 * access is safe because interrupts can only add new
2395 * entries to the tail of this list, and only ->poll()
2396 * calls can remove this head entry from the list.
2397 */
2398 n = list_entry(list->next, struct napi_struct, poll_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002400 have = netpoll_poll_lock(n);
2401
2402 weight = n->weight;
2403
David S. Miller0a7606c2007-10-29 21:28:47 -07002404 /* This NAPI_STATE_SCHED test is for avoiding a race
2405 * with netpoll's poll_napi(). Only the entity which
2406 * obtains the lock and sees NAPI_STATE_SCHED set will
2407 * actually make the ->poll() call. Therefore we avoid
2408 * accidently calling ->poll() when NAPI is not scheduled.
2409 */
2410 work = 0;
2411 if (test_bit(NAPI_STATE_SCHED, &n->state))
2412 work = n->poll(n, weight);
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002413
2414 WARN_ON_ONCE(work > weight);
2415
2416 budget -= work;
2417
2418 local_irq_disable();
2419
2420 /* Drivers must not modify the NAPI state if they
2421 * consume the entire weight. In such cases this code
2422 * still "owns" the NAPI instance and therefore can
2423 * move the instance around on the list at-will.
2424 */
David S. Millerfed17f32008-01-07 21:00:40 -08002425 if (unlikely(work == weight)) {
2426 if (unlikely(napi_disable_pending(n)))
2427 __napi_complete(n);
2428 else
2429 list_move_tail(&n->poll_list, list);
2430 }
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002431
2432 netpoll_poll_unlock(have);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 }
2434out:
Shannon Nelson515e06c2007-06-23 23:09:23 -07002435 local_irq_enable();
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002436
Chris Leechdb217332006-06-17 21:24:58 -07002437#ifdef CONFIG_NET_DMA
2438 /*
2439 * There may not be any more sk_buffs coming right now, so push
2440 * any pending DMA copies to hardware
2441 */
Dan Williamsd379b012007-07-09 11:56:42 -07002442 if (!cpus_empty(net_dma.channel_mask)) {
2443 int chan_idx;
Mike Travis0e12f842008-05-12 21:21:13 +02002444 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
Dan Williamsd379b012007-07-09 11:56:42 -07002445 struct dma_chan *chan = net_dma.channels[chan_idx];
2446 if (chan)
2447 dma_async_memcpy_issue_pending(chan);
2448 }
Chris Leechdb217332006-06-17 21:24:58 -07002449 }
2450#endif
Stephen Hemmingerbea33482007-10-03 16:41:36 -07002451
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 return;
2453
2454softnet_break:
2455 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2456 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2457 goto out;
2458}
2459
2460static gifconf_func_t * gifconf_list [NPROTO];
2461
2462/**
2463 * register_gifconf - register a SIOCGIF handler
2464 * @family: Address family
2465 * @gifconf: Function handler
2466 *
2467 * Register protocol dependent address dumping routines. The handler
2468 * that is passed must not be freed or reused until it has been replaced
2469 * by another handler.
2470 */
2471int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2472{
2473 if (family >= NPROTO)
2474 return -EINVAL;
2475 gifconf_list[family] = gifconf;
2476 return 0;
2477}
2478
2479
2480/*
2481 * Map an interface index to its name (SIOCGIFNAME)
2482 */
2483
2484/*
2485 * We need this ioctl for efficient implementation of the
2486 * if_indextoname() function required by the IPv6 API. Without
2487 * it, we would have to search all the interfaces to find a
2488 * match. --pb
2489 */
2490
Eric W. Biederman881d9662007-09-17 11:56:21 -07002491static int dev_ifname(struct net *net, struct ifreq __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492{
2493 struct net_device *dev;
2494 struct ifreq ifr;
2495
2496 /*
2497 * Fetch the caller's info block.
2498 */
2499
2500 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2501 return -EFAULT;
2502
2503 read_lock(&dev_base_lock);
Eric W. Biederman881d9662007-09-17 11:56:21 -07002504 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 if (!dev) {
2506 read_unlock(&dev_base_lock);
2507 return -ENODEV;
2508 }
2509
2510 strcpy(ifr.ifr_name, dev->name);
2511 read_unlock(&dev_base_lock);
2512
2513 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2514 return -EFAULT;
2515 return 0;
2516}
2517
2518/*
2519 * Perform a SIOCGIFCONF call. This structure will change
2520 * size eventually, and there is nothing I can do about it.
2521 * Thus we will need a 'compatibility mode'.
2522 */
2523
Eric W. Biederman881d9662007-09-17 11:56:21 -07002524static int dev_ifconf(struct net *net, char __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525{
2526 struct ifconf ifc;
2527 struct net_device *dev;
2528 char __user *pos;
2529 int len;
2530 int total;
2531 int i;
2532
2533 /*
2534 * Fetch the caller's info block.
2535 */
2536
2537 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2538 return -EFAULT;
2539
2540 pos = ifc.ifc_buf;
2541 len = ifc.ifc_len;
2542
2543 /*
2544 * Loop over the interfaces, and write an info block for each.
2545 */
2546
2547 total = 0;
Eric W. Biederman881d9662007-09-17 11:56:21 -07002548 for_each_netdev(net, dev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 for (i = 0; i < NPROTO; i++) {
2550 if (gifconf_list[i]) {
2551 int done;
2552 if (!pos)
2553 done = gifconf_list[i](dev, NULL, 0);
2554 else
2555 done = gifconf_list[i](dev, pos + total,
2556 len - total);
2557 if (done < 0)
2558 return -EFAULT;
2559 total += done;
2560 }
2561 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
2564 /*
2565 * All done. Write the updated control block back to the caller.
2566 */
2567 ifc.ifc_len = total;
2568
2569 /*
2570 * Both BSD and Solaris return 0 here, so we do too.
2571 */
2572 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2573}
2574
2575#ifdef CONFIG_PROC_FS
2576/*
2577 * This is invoked by the /proc filesystem handler to display a device
2578 * in detail.
2579 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580void *dev_seq_start(struct seq_file *seq, loff_t *pos)
Eric Dumazet9a429c42008-01-01 21:58:02 -08002581 __acquires(dev_base_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582{
Denis V. Luneve372c412007-11-19 22:31:54 -08002583 struct net *net = seq_file_net(seq);
Pavel Emelianov7562f872007-05-03 15:13:45 -07002584 loff_t off;
2585 struct net_device *dev;
2586
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 read_lock(&dev_base_lock);
Pavel Emelianov7562f872007-05-03 15:13:45 -07002588 if (!*pos)
2589 return SEQ_START_TOKEN;
2590
2591 off = 1;
Eric W. Biederman881d9662007-09-17 11:56:21 -07002592 for_each_netdev(net, dev)
Pavel Emelianov7562f872007-05-03 15:13:45 -07002593 if (off++ == *pos)
2594 return dev;
2595
2596 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597}
2598
2599void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2600{
Denis V. Luneve372c412007-11-19 22:31:54 -08002601 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 ++*pos;
Pavel Emelianov7562f872007-05-03 15:13:45 -07002603 return v == SEQ_START_TOKEN ?
Eric W. Biederman881d9662007-09-17 11:56:21 -07002604 first_net_device(net) : next_net_device((struct net_device *)v);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605}
2606
2607void dev_seq_stop(struct seq_file *seq, void *v)
Eric Dumazet9a429c42008-01-01 21:58:02 -08002608 __releases(dev_base_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609{
2610 read_unlock(&dev_base_lock);
2611}
2612
2613static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2614{
Rusty Russellc45d2862007-03-28 14:29:08 -07002615 struct net_device_stats *stats = dev->get_stats(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Rusty Russell5a1b5892007-04-28 21:04:03 -07002617 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2618 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2619 dev->name, stats->rx_bytes, stats->rx_packets,
2620 stats->rx_errors,
2621 stats->rx_dropped + stats->rx_missed_errors,
2622 stats->rx_fifo_errors,
2623 stats->rx_length_errors + stats->rx_over_errors +
2624 stats->rx_crc_errors + stats->rx_frame_errors,
2625 stats->rx_compressed, stats->multicast,
2626 stats->tx_bytes, stats->tx_packets,
2627 stats->tx_errors, stats->tx_dropped,
2628 stats->tx_fifo_errors, stats->collisions,
2629 stats->tx_carrier_errors +
2630 stats->tx_aborted_errors +
2631 stats->tx_window_errors +
2632 stats->tx_heartbeat_errors,
2633 stats->tx_compressed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634}
2635
2636/*
2637 * Called from the PROCfs module. This now uses the new arbitrary sized
2638 * /proc/net interface to create /proc/net/dev
2639 */
2640static int dev_seq_show(struct seq_file *seq, void *v)
2641{
2642 if (v == SEQ_START_TOKEN)
2643 seq_puts(seq, "Inter-| Receive "
2644 " | Transmit\n"
2645 " face |bytes packets errs drop fifo frame "
2646 "compressed multicast|bytes packets errs "
2647 "drop fifo colls carrier compressed\n");
2648 else
2649 dev_seq_printf_stats(seq, v);
2650 return 0;
2651}
2652
2653static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2654{
2655 struct netif_rx_stats *rc = NULL;
2656
Mike Travis0c0b0ac2008-05-02 16:43:08 -07002657 while (*pos < nr_cpu_ids)
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002658 if (cpu_online(*pos)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 rc = &per_cpu(netdev_rx_stat, *pos);
2660 break;
2661 } else
2662 ++*pos;
2663 return rc;
2664}
2665
2666static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2667{
2668 return softnet_get_online(pos);
2669}
2670
2671static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2672{
2673 ++*pos;
2674 return softnet_get_online(pos);
2675}
2676
2677static void softnet_seq_stop(struct seq_file *seq, void *v)
2678{
2679}
2680
2681static int softnet_seq_show(struct seq_file *seq, void *v)
2682{
2683 struct netif_rx_stats *s = v;
2684
2685 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
Stephen Hemminger31aa02c2005-06-23 20:12:48 -07002686 s->total, s->dropped, s->time_squeeze, 0,
Stephen Hemmingerc1ebcdb2005-06-23 20:08:59 -07002687 0, 0, 0, 0, /* was fastroute */
2688 s->cpu_collision );
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 return 0;
2690}
2691
Stephen Hemmingerf6908082007-03-12 14:34:29 -07002692static const struct seq_operations dev_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 .start = dev_seq_start,
2694 .next = dev_seq_next,
2695 .stop = dev_seq_stop,
2696 .show = dev_seq_show,
2697};
2698
2699static int dev_seq_open(struct inode *inode, struct file *file)
2700{
Denis V. Luneve372c412007-11-19 22:31:54 -08002701 return seq_open_net(inode, file, &dev_seq_ops,
2702 sizeof(struct seq_net_private));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703}
2704
Arjan van de Ven9a321442007-02-12 00:55:35 -08002705static const struct file_operations dev_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 .owner = THIS_MODULE,
2707 .open = dev_seq_open,
2708 .read = seq_read,
2709 .llseek = seq_lseek,
Denis V. Luneve372c412007-11-19 22:31:54 -08002710 .release = seq_release_net,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711};
2712
Stephen Hemmingerf6908082007-03-12 14:34:29 -07002713static const struct seq_operations softnet_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 .start = softnet_seq_start,
2715 .next = softnet_seq_next,
2716 .stop = softnet_seq_stop,
2717 .show = softnet_seq_show,
2718};
2719
2720static int softnet_seq_open(struct inode *inode, struct file *file)
2721{
2722 return seq_open(file, &softnet_seq_ops);
2723}
2724
Arjan van de Ven9a321442007-02-12 00:55:35 -08002725static const struct file_operations softnet_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 .owner = THIS_MODULE,
2727 .open = softnet_seq_open,
2728 .read = seq_read,
2729 .llseek = seq_lseek,
2730 .release = seq_release,
2731};
2732
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002733static void *ptype_get_idx(loff_t pos)
2734{
2735 struct packet_type *pt = NULL;
2736 loff_t i = 0;
2737 int t;
2738
2739 list_for_each_entry_rcu(pt, &ptype_all, list) {
2740 if (i == pos)
2741 return pt;
2742 ++i;
2743 }
2744
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08002745 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002746 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2747 if (i == pos)
2748 return pt;
2749 ++i;
2750 }
2751 }
2752 return NULL;
2753}
2754
2755static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
Stephen Hemminger72348a42008-01-21 02:27:29 -08002756 __acquires(RCU)
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002757{
2758 rcu_read_lock();
2759 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2760}
2761
2762static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2763{
2764 struct packet_type *pt;
2765 struct list_head *nxt;
2766 int hash;
2767
2768 ++*pos;
2769 if (v == SEQ_START_TOKEN)
2770 return ptype_get_idx(0);
2771
2772 pt = v;
2773 nxt = pt->list.next;
2774 if (pt->type == htons(ETH_P_ALL)) {
2775 if (nxt != &ptype_all)
2776 goto found;
2777 hash = 0;
2778 nxt = ptype_base[0].next;
2779 } else
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08002780 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002781
2782 while (nxt == &ptype_base[hash]) {
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08002783 if (++hash >= PTYPE_HASH_SIZE)
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002784 return NULL;
2785 nxt = ptype_base[hash].next;
2786 }
2787found:
2788 return list_entry(nxt, struct packet_type, list);
2789}
2790
2791static void ptype_seq_stop(struct seq_file *seq, void *v)
Stephen Hemminger72348a42008-01-21 02:27:29 -08002792 __releases(RCU)
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002793{
2794 rcu_read_unlock();
2795}
2796
2797static void ptype_seq_decode(struct seq_file *seq, void *sym)
2798{
2799#ifdef CONFIG_KALLSYMS
2800 unsigned long offset = 0, symsize;
2801 const char *symname;
2802 char *modname;
2803 char namebuf[128];
2804
2805 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2806 &modname, namebuf);
2807
2808 if (symname) {
2809 char *delim = ":";
2810
2811 if (!modname)
2812 modname = delim = "";
2813 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2814 symname, offset);
2815 return;
2816 }
2817#endif
2818
2819 seq_printf(seq, "[%p]", sym);
2820}
2821
2822static int ptype_seq_show(struct seq_file *seq, void *v)
2823{
2824 struct packet_type *pt = v;
2825
2826 if (v == SEQ_START_TOKEN)
2827 seq_puts(seq, "Type Device Function\n");
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09002828 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002829 if (pt->type == htons(ETH_P_ALL))
2830 seq_puts(seq, "ALL ");
2831 else
2832 seq_printf(seq, "%04x", ntohs(pt->type));
2833
2834 seq_printf(seq, " %-8s ",
2835 pt->dev ? pt->dev->name : "");
2836 ptype_seq_decode(seq, pt->func);
2837 seq_putc(seq, '\n');
2838 }
2839
2840 return 0;
2841}
2842
2843static const struct seq_operations ptype_seq_ops = {
2844 .start = ptype_seq_start,
2845 .next = ptype_seq_next,
2846 .stop = ptype_seq_stop,
2847 .show = ptype_seq_show,
2848};
2849
2850static int ptype_seq_open(struct inode *inode, struct file *file)
2851{
Pavel Emelyanov2feb27d2008-03-24 14:57:45 -07002852 return seq_open_net(inode, file, &ptype_seq_ops,
2853 sizeof(struct seq_net_private));
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002854}
2855
2856static const struct file_operations ptype_seq_fops = {
2857 .owner = THIS_MODULE,
2858 .open = ptype_seq_open,
2859 .read = seq_read,
2860 .llseek = seq_lseek,
Pavel Emelyanov2feb27d2008-03-24 14:57:45 -07002861 .release = seq_release_net,
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002862};
2863
2864
Pavel Emelyanov46650792007-10-08 20:38:39 -07002865static int __net_init dev_proc_net_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866{
2867 int rc = -ENOMEM;
2868
Eric W. Biederman881d9662007-09-17 11:56:21 -07002869 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 goto out;
Eric W. Biederman881d9662007-09-17 11:56:21 -07002871 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 goto out_dev;
Eric W. Biederman881d9662007-09-17 11:56:21 -07002873 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
Eric W. Biederman457c4cb2007-09-12 12:01:34 +02002874 goto out_softnet;
Stephen Hemminger0e1256f2007-03-12 14:35:37 -07002875
Eric W. Biederman881d9662007-09-17 11:56:21 -07002876 if (wext_proc_init(net))
Eric W. Biederman457c4cb2007-09-12 12:01:34 +02002877 goto out_ptype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 rc = 0;
2879out:
2880 return rc;
Eric W. Biederman457c4cb2007-09-12 12:01:34 +02002881out_ptype:
Eric W. Biederman881d9662007-09-17 11:56:21 -07002882 proc_net_remove(net, "ptype");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883out_softnet:
Eric W. Biederman881d9662007-09-17 11:56:21 -07002884 proc_net_remove(net, "softnet_stat");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885out_dev:
Eric W. Biederman881d9662007-09-17 11:56:21 -07002886 proc_net_remove(net, "dev");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 goto out;
2888}
Eric W. Biederman881d9662007-09-17 11:56:21 -07002889
Pavel Emelyanov46650792007-10-08 20:38:39 -07002890static void __net_exit dev_proc_net_exit(struct net *net)
Eric W. Biederman881d9662007-09-17 11:56:21 -07002891{
2892 wext_proc_exit(net);
2893
2894 proc_net_remove(net, "ptype");
2895 proc_net_remove(net, "softnet_stat");
2896 proc_net_remove(net, "dev");
2897}
2898
Denis V. Lunev022cbae2007-11-13 03:23:50 -08002899static struct pernet_operations __net_initdata dev_proc_ops = {
Eric W. Biederman881d9662007-09-17 11:56:21 -07002900 .init = dev_proc_net_init,
2901 .exit = dev_proc_net_exit,
2902};
2903
2904static int __init dev_proc_init(void)
2905{
2906 return register_pernet_subsys(&dev_proc_ops);
2907}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908#else
2909#define dev_proc_init() 0
2910#endif /* CONFIG_PROC_FS */
2911
2912
2913/**
2914 * netdev_set_master - set up master/slave pair
2915 * @slave: slave device
2916 * @master: new master device
2917 *
2918 * Changes the master device of the slave. Pass %NULL to break the
2919 * bonding. The caller must hold the RTNL semaphore. On a failure
2920 * a negative errno code is returned. On success the reference counts
2921 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2922 * function returns zero.
2923 */
2924int netdev_set_master(struct net_device *slave, struct net_device *master)
2925{
2926 struct net_device *old = slave->master;
2927
2928 ASSERT_RTNL();
2929
2930 if (master) {
2931 if (old)
2932 return -EBUSY;
2933 dev_hold(master);
2934 }
2935
2936 slave->master = master;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09002937
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 synchronize_net();
2939
2940 if (old)
2941 dev_put(old);
2942
2943 if (master)
2944 slave->flags |= IFF_SLAVE;
2945 else
2946 slave->flags &= ~IFF_SLAVE;
2947
2948 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2949 return 0;
2950}
2951
Patrick McHardyb6c40d62008-10-07 15:26:48 -07002952static void dev_change_rx_flags(struct net_device *dev, int flags)
2953{
2954 if (dev->flags & IFF_UP && dev->change_rx_flags)
2955 dev->change_rx_flags(dev, flags);
2956}
2957
Wang Chendad9b332008-06-18 01:48:28 -07002958static int __dev_set_promiscuity(struct net_device *dev, int inc)
Patrick McHardy4417da62007-06-27 01:28:10 -07002959{
2960 unsigned short old_flags = dev->flags;
David Howells8192b0c2008-11-14 10:39:10 +11002961 uid_t uid;
2962 gid_t gid;
Patrick McHardy4417da62007-06-27 01:28:10 -07002963
Patrick McHardy24023452007-07-14 18:51:31 -07002964 ASSERT_RTNL();
2965
Wang Chendad9b332008-06-18 01:48:28 -07002966 dev->flags |= IFF_PROMISC;
2967 dev->promiscuity += inc;
2968 if (dev->promiscuity == 0) {
2969 /*
2970 * Avoid overflow.
2971 * If inc causes overflow, untouch promisc and return error.
2972 */
2973 if (inc < 0)
2974 dev->flags &= ~IFF_PROMISC;
2975 else {
2976 dev->promiscuity -= inc;
2977 printk(KERN_WARNING "%s: promiscuity touches roof, "
2978 "set promiscuity failed, promiscuity feature "
2979 "of device might be broken.\n", dev->name);
2980 return -EOVERFLOW;
2981 }
2982 }
Patrick McHardy4417da62007-06-27 01:28:10 -07002983 if (dev->flags != old_flags) {
2984 printk(KERN_INFO "device %s %s promiscuous mode\n",
2985 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2986 "left");
David Howells8192b0c2008-11-14 10:39:10 +11002987 if (audit_enabled) {
2988 current_uid_gid(&uid, &gid);
Klaus Heinrich Kiwi7759db82008-01-23 22:57:45 -05002989 audit_log(current->audit_context, GFP_ATOMIC,
2990 AUDIT_ANOM_PROMISCUOUS,
2991 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2992 dev->name, (dev->flags & IFF_PROMISC),
2993 (old_flags & IFF_PROMISC),
2994 audit_get_loginuid(current),
David Howells8192b0c2008-11-14 10:39:10 +11002995 uid, gid,
Klaus Heinrich Kiwi7759db82008-01-23 22:57:45 -05002996 audit_get_sessionid(current));
David Howells8192b0c2008-11-14 10:39:10 +11002997 }
Patrick McHardy24023452007-07-14 18:51:31 -07002998
Patrick McHardyb6c40d62008-10-07 15:26:48 -07002999 dev_change_rx_flags(dev, IFF_PROMISC);
Patrick McHardy4417da62007-06-27 01:28:10 -07003000 }
Wang Chendad9b332008-06-18 01:48:28 -07003001 return 0;
Patrick McHardy4417da62007-06-27 01:28:10 -07003002}
3003
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004/**
3005 * dev_set_promiscuity - update promiscuity count on a device
3006 * @dev: device
3007 * @inc: modifier
3008 *
Stephen Hemminger3041a062006-05-26 13:25:24 -07003009 * Add or remove promiscuity from a device. While the count in the device
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 * remains above zero the interface remains promiscuous. Once it hits zero
3011 * the device reverts back to normal filtering operation. A negative inc
3012 * value is used to drop promiscuity on the device.
Wang Chendad9b332008-06-18 01:48:28 -07003013 * Return 0 if successful or a negative errno code on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 */
Wang Chendad9b332008-06-18 01:48:28 -07003015int dev_set_promiscuity(struct net_device *dev, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016{
3017 unsigned short old_flags = dev->flags;
Wang Chendad9b332008-06-18 01:48:28 -07003018 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019
Wang Chendad9b332008-06-18 01:48:28 -07003020 err = __dev_set_promiscuity(dev, inc);
Patrick McHardy4b5a6982008-07-06 15:49:08 -07003021 if (err < 0)
Wang Chendad9b332008-06-18 01:48:28 -07003022 return err;
Patrick McHardy4417da62007-06-27 01:28:10 -07003023 if (dev->flags != old_flags)
3024 dev_set_rx_mode(dev);
Wang Chendad9b332008-06-18 01:48:28 -07003025 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026}
3027
3028/**
3029 * dev_set_allmulti - update allmulti count on a device
3030 * @dev: device
3031 * @inc: modifier
3032 *
3033 * Add or remove reception of all multicast frames to a device. While the
3034 * count in the device remains above zero the interface remains listening
3035 * to all interfaces. Once it hits zero the device reverts back to normal
3036 * filtering operation. A negative @inc value is used to drop the counter
3037 * when releasing a resource needing all multicasts.
Wang Chendad9b332008-06-18 01:48:28 -07003038 * Return 0 if successful or a negative errno code on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 */
3040
Wang Chendad9b332008-06-18 01:48:28 -07003041int dev_set_allmulti(struct net_device *dev, int inc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042{
3043 unsigned short old_flags = dev->flags;
3044
Patrick McHardy24023452007-07-14 18:51:31 -07003045 ASSERT_RTNL();
3046
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 dev->flags |= IFF_ALLMULTI;
Wang Chendad9b332008-06-18 01:48:28 -07003048 dev->allmulti += inc;
3049 if (dev->allmulti == 0) {
3050 /*
3051 * Avoid overflow.
3052 * If inc causes overflow, untouch allmulti and return error.
3053 */
3054 if (inc < 0)
3055 dev->flags &= ~IFF_ALLMULTI;
3056 else {
3057 dev->allmulti -= inc;
3058 printk(KERN_WARNING "%s: allmulti touches roof, "
3059 "set allmulti failed, allmulti feature of "
3060 "device might be broken.\n", dev->name);
3061 return -EOVERFLOW;
3062 }
3063 }
Patrick McHardy24023452007-07-14 18:51:31 -07003064 if (dev->flags ^ old_flags) {
Patrick McHardyb6c40d62008-10-07 15:26:48 -07003065 dev_change_rx_flags(dev, IFF_ALLMULTI);
Patrick McHardy4417da62007-06-27 01:28:10 -07003066 dev_set_rx_mode(dev);
Patrick McHardy24023452007-07-14 18:51:31 -07003067 }
Wang Chendad9b332008-06-18 01:48:28 -07003068 return 0;
Patrick McHardy4417da62007-06-27 01:28:10 -07003069}
3070
3071/*
3072 * Upload unicast and multicast address lists to device and
3073 * configure RX filtering. When the device doesn't support unicast
Joe Perches53ccaae2007-12-20 14:02:06 -08003074 * filtering it is put in promiscuous mode while unicast addresses
Patrick McHardy4417da62007-06-27 01:28:10 -07003075 * are present.
3076 */
3077void __dev_set_rx_mode(struct net_device *dev)
3078{
3079 /* dev_open will call this function so the list will stay sane. */
3080 if (!(dev->flags&IFF_UP))
3081 return;
3082
3083 if (!netif_device_present(dev))
YOSHIFUJI Hideaki40b77c92007-07-19 10:43:23 +09003084 return;
Patrick McHardy4417da62007-06-27 01:28:10 -07003085
3086 if (dev->set_rx_mode)
3087 dev->set_rx_mode(dev);
3088 else {
3089 /* Unicast addresses changes may only happen under the rtnl,
3090 * therefore calling __dev_set_promiscuity here is safe.
3091 */
3092 if (dev->uc_count > 0 && !dev->uc_promisc) {
3093 __dev_set_promiscuity(dev, 1);
3094 dev->uc_promisc = 1;
3095 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3096 __dev_set_promiscuity(dev, -1);
3097 dev->uc_promisc = 0;
3098 }
3099
3100 if (dev->set_multicast_list)
3101 dev->set_multicast_list(dev);
3102 }
3103}
3104
3105void dev_set_rx_mode(struct net_device *dev)
3106{
David S. Millerb9e40852008-07-15 00:15:08 -07003107 netif_addr_lock_bh(dev);
Patrick McHardy4417da62007-06-27 01:28:10 -07003108 __dev_set_rx_mode(dev);
David S. Millerb9e40852008-07-15 00:15:08 -07003109 netif_addr_unlock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110}
3111
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07003112int __dev_addr_delete(struct dev_addr_list **list, int *count,
3113 void *addr, int alen, int glbl)
Patrick McHardybf742482007-06-27 01:26:19 -07003114{
3115 struct dev_addr_list *da;
3116
3117 for (; (da = *list) != NULL; list = &da->next) {
3118 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3119 alen == da->da_addrlen) {
3120 if (glbl) {
3121 int old_glbl = da->da_gusers;
3122 da->da_gusers = 0;
3123 if (old_glbl == 0)
3124 break;
3125 }
3126 if (--da->da_users)
3127 return 0;
3128
3129 *list = da->next;
3130 kfree(da);
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07003131 (*count)--;
Patrick McHardybf742482007-06-27 01:26:19 -07003132 return 0;
3133 }
3134 }
3135 return -ENOENT;
3136}
3137
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07003138int __dev_addr_add(struct dev_addr_list **list, int *count,
3139 void *addr, int alen, int glbl)
Patrick McHardybf742482007-06-27 01:26:19 -07003140{
3141 struct dev_addr_list *da;
3142
3143 for (da = *list; da != NULL; da = da->next) {
3144 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3145 da->da_addrlen == alen) {
3146 if (glbl) {
3147 int old_glbl = da->da_gusers;
3148 da->da_gusers = 1;
3149 if (old_glbl)
3150 return 0;
3151 }
3152 da->da_users++;
3153 return 0;
3154 }
3155 }
3156
Jorge Boncompte [DTI2]12aa3432008-02-19 14:17:04 -08003157 da = kzalloc(sizeof(*da), GFP_ATOMIC);
Patrick McHardybf742482007-06-27 01:26:19 -07003158 if (da == NULL)
3159 return -ENOMEM;
3160 memcpy(da->da_addr, addr, alen);
3161 da->da_addrlen = alen;
3162 da->da_users = 1;
3163 da->da_gusers = glbl ? 1 : 0;
3164 da->next = *list;
3165 *list = da;
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07003166 (*count)++;
Patrick McHardybf742482007-06-27 01:26:19 -07003167 return 0;
3168}
3169
Patrick McHardy4417da62007-06-27 01:28:10 -07003170/**
3171 * dev_unicast_delete - Release secondary unicast address.
3172 * @dev: device
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07003173 * @addr: address to delete
3174 * @alen: length of @addr
Patrick McHardy4417da62007-06-27 01:28:10 -07003175 *
3176 * Release reference to a secondary unicast address and remove it
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07003177 * from the device if the reference count drops to zero.
Patrick McHardy4417da62007-06-27 01:28:10 -07003178 *
3179 * The caller must hold the rtnl_mutex.
3180 */
3181int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3182{
3183 int err;
3184
3185 ASSERT_RTNL();
3186
David S. Millerb9e40852008-07-15 00:15:08 -07003187 netif_addr_lock_bh(dev);
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07003188 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3189 if (!err)
Patrick McHardy4417da62007-06-27 01:28:10 -07003190 __dev_set_rx_mode(dev);
David S. Millerb9e40852008-07-15 00:15:08 -07003191 netif_addr_unlock_bh(dev);
Patrick McHardy4417da62007-06-27 01:28:10 -07003192 return err;
3193}
3194EXPORT_SYMBOL(dev_unicast_delete);
3195
3196/**
3197 * dev_unicast_add - add a secondary unicast address
3198 * @dev: device
Wang Chen5dbaec52008-06-27 19:35:16 -07003199 * @addr: address to add
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07003200 * @alen: length of @addr
Patrick McHardy4417da62007-06-27 01:28:10 -07003201 *
3202 * Add a secondary unicast address to the device or increase
3203 * the reference count if it already exists.
3204 *
3205 * The caller must hold the rtnl_mutex.
3206 */
3207int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3208{
3209 int err;
3210
3211 ASSERT_RTNL();
3212
David S. Millerb9e40852008-07-15 00:15:08 -07003213 netif_addr_lock_bh(dev);
Patrick McHardy61cbc2f2007-06-30 13:35:52 -07003214 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3215 if (!err)
Patrick McHardy4417da62007-06-27 01:28:10 -07003216 __dev_set_rx_mode(dev);
David S. Millerb9e40852008-07-15 00:15:08 -07003217 netif_addr_unlock_bh(dev);
Patrick McHardy4417da62007-06-27 01:28:10 -07003218 return err;
3219}
3220EXPORT_SYMBOL(dev_unicast_add);
3221
Chris Leeche83a2ea2008-01-31 16:53:23 -08003222int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3223 struct dev_addr_list **from, int *from_count)
3224{
3225 struct dev_addr_list *da, *next;
3226 int err = 0;
3227
3228 da = *from;
3229 while (da != NULL) {
3230 next = da->next;
3231 if (!da->da_synced) {
3232 err = __dev_addr_add(to, to_count,
3233 da->da_addr, da->da_addrlen, 0);
3234 if (err < 0)
3235 break;
3236 da->da_synced = 1;
3237 da->da_users++;
3238 } else if (da->da_users == 1) {
3239 __dev_addr_delete(to, to_count,
3240 da->da_addr, da->da_addrlen, 0);
3241 __dev_addr_delete(from, from_count,
3242 da->da_addr, da->da_addrlen, 0);
3243 }
3244 da = next;
3245 }
3246 return err;
3247}
3248
3249void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3250 struct dev_addr_list **from, int *from_count)
3251{
3252 struct dev_addr_list *da, *next;
3253
3254 da = *from;
3255 while (da != NULL) {
3256 next = da->next;
3257 if (da->da_synced) {
3258 __dev_addr_delete(to, to_count,
3259 da->da_addr, da->da_addrlen, 0);
3260 da->da_synced = 0;
3261 __dev_addr_delete(from, from_count,
3262 da->da_addr, da->da_addrlen, 0);
3263 }
3264 da = next;
3265 }
3266}
3267
3268/**
3269 * dev_unicast_sync - Synchronize device's unicast list to another device
3270 * @to: destination device
3271 * @from: source device
3272 *
3273 * Add newly added addresses to the destination device and release
3274 * addresses that have no users left. The source device must be
3275 * locked by netif_tx_lock_bh.
3276 *
3277 * This function is intended to be called from the dev->set_rx_mode
3278 * function of layered software devices.
3279 */
3280int dev_unicast_sync(struct net_device *to, struct net_device *from)
3281{
3282 int err = 0;
3283
David S. Millerb9e40852008-07-15 00:15:08 -07003284 netif_addr_lock_bh(to);
Chris Leeche83a2ea2008-01-31 16:53:23 -08003285 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3286 &from->uc_list, &from->uc_count);
3287 if (!err)
3288 __dev_set_rx_mode(to);
David S. Millerb9e40852008-07-15 00:15:08 -07003289 netif_addr_unlock_bh(to);
Chris Leeche83a2ea2008-01-31 16:53:23 -08003290 return err;
3291}
3292EXPORT_SYMBOL(dev_unicast_sync);
3293
3294/**
Randy Dunlapbc2cda12008-02-13 15:03:25 -08003295 * dev_unicast_unsync - Remove synchronized addresses from the destination device
Chris Leeche83a2ea2008-01-31 16:53:23 -08003296 * @to: destination device
3297 * @from: source device
3298 *
3299 * Remove all addresses that were added to the destination device by
3300 * dev_unicast_sync(). This function is intended to be called from the
3301 * dev->stop function of layered software devices.
3302 */
3303void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3304{
David S. Millerb9e40852008-07-15 00:15:08 -07003305 netif_addr_lock_bh(from);
David S. Millere308a5d2008-07-15 00:13:44 -07003306 netif_addr_lock(to);
Chris Leeche83a2ea2008-01-31 16:53:23 -08003307
3308 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3309 &from->uc_list, &from->uc_count);
3310 __dev_set_rx_mode(to);
3311
David S. Millere308a5d2008-07-15 00:13:44 -07003312 netif_addr_unlock(to);
David S. Millerb9e40852008-07-15 00:15:08 -07003313 netif_addr_unlock_bh(from);
Chris Leeche83a2ea2008-01-31 16:53:23 -08003314}
3315EXPORT_SYMBOL(dev_unicast_unsync);
3316
Denis Cheng12972622007-07-18 02:12:56 -07003317static void __dev_addr_discard(struct dev_addr_list **list)
3318{
3319 struct dev_addr_list *tmp;
3320
3321 while (*list != NULL) {
3322 tmp = *list;
3323 *list = tmp->next;
3324 if (tmp->da_users > tmp->da_gusers)
3325 printk("__dev_addr_discard: address leakage! "
3326 "da_users=%d\n", tmp->da_users);
3327 kfree(tmp);
3328 }
3329}
3330
Denis Cheng26cc2522007-07-18 02:12:03 -07003331static void dev_addr_discard(struct net_device *dev)
Patrick McHardy4417da62007-06-27 01:28:10 -07003332{
David S. Millerb9e40852008-07-15 00:15:08 -07003333 netif_addr_lock_bh(dev);
Denis Cheng26cc2522007-07-18 02:12:03 -07003334
Patrick McHardy4417da62007-06-27 01:28:10 -07003335 __dev_addr_discard(&dev->uc_list);
3336 dev->uc_count = 0;
Patrick McHardy4417da62007-06-27 01:28:10 -07003337
Denis Cheng456ad752007-07-18 02:10:54 -07003338 __dev_addr_discard(&dev->mc_list);
3339 dev->mc_count = 0;
Denis Cheng26cc2522007-07-18 02:12:03 -07003340
David S. Millerb9e40852008-07-15 00:15:08 -07003341 netif_addr_unlock_bh(dev);
Denis Cheng456ad752007-07-18 02:10:54 -07003342}
3343
Stephen Hemmingerf0db2752008-09-30 02:23:58 -07003344/**
3345 * dev_get_flags - get flags reported to userspace
3346 * @dev: device
3347 *
3348 * Get the combination of flag bits exported through APIs to userspace.
3349 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350unsigned dev_get_flags(const struct net_device *dev)
3351{
3352 unsigned flags;
3353
3354 flags = (dev->flags & ~(IFF_PROMISC |
3355 IFF_ALLMULTI |
Stefan Rompfb00055a2006-03-20 17:09:11 -08003356 IFF_RUNNING |
3357 IFF_LOWER_UP |
3358 IFF_DORMANT)) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359 (dev->gflags & (IFF_PROMISC |
3360 IFF_ALLMULTI));
3361
Stefan Rompfb00055a2006-03-20 17:09:11 -08003362 if (netif_running(dev)) {
3363 if (netif_oper_up(dev))
3364 flags |= IFF_RUNNING;
3365 if (netif_carrier_ok(dev))
3366 flags |= IFF_LOWER_UP;
3367 if (netif_dormant(dev))
3368 flags |= IFF_DORMANT;
3369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370
3371 return flags;
3372}
3373
Stephen Hemmingerf0db2752008-09-30 02:23:58 -07003374/**
3375 * dev_change_flags - change device settings
3376 * @dev: device
3377 * @flags: device state flags
3378 *
3379 * Change settings on device based state flags. The flags are
3380 * in the userspace exported format.
3381 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382int dev_change_flags(struct net_device *dev, unsigned flags)
3383{
Thomas Graf7c355f52007-06-05 16:03:03 -07003384 int ret, changes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 int old_flags = dev->flags;
3386
Patrick McHardy24023452007-07-14 18:51:31 -07003387 ASSERT_RTNL();
3388
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389 /*
3390 * Set the flags on our device.
3391 */
3392
3393 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3394 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3395 IFF_AUTOMEDIA)) |
3396 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3397 IFF_ALLMULTI));
3398
3399 /*
3400 * Load in the correct multicast list now the flags have changed.
3401 */
3402
Patrick McHardyb6c40d62008-10-07 15:26:48 -07003403 if ((old_flags ^ flags) & IFF_MULTICAST)
3404 dev_change_rx_flags(dev, IFF_MULTICAST);
Patrick McHardy24023452007-07-14 18:51:31 -07003405
Patrick McHardy4417da62007-06-27 01:28:10 -07003406 dev_set_rx_mode(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407
3408 /*
3409 * Have we downed the interface. We handle IFF_UP ourselves
3410 * according to user attempts to set it, rather than blindly
3411 * setting it.
3412 */
3413
3414 ret = 0;
3415 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3416 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3417
3418 if (!ret)
Patrick McHardy4417da62007-06-27 01:28:10 -07003419 dev_set_rx_mode(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420 }
3421
3422 if (dev->flags & IFF_UP &&
3423 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3424 IFF_VOLATILE)))
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003425 call_netdevice_notifiers(NETDEV_CHANGE, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426
3427 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3428 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3429 dev->gflags ^= IFF_PROMISC;
3430 dev_set_promiscuity(dev, inc);
3431 }
3432
3433 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3434 is important. Some (broken) drivers set IFF_PROMISC, when
3435 IFF_ALLMULTI is requested not asking us and not reporting.
3436 */
3437 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3438 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3439 dev->gflags ^= IFF_ALLMULTI;
3440 dev_set_allmulti(dev, inc);
3441 }
3442
Thomas Graf7c355f52007-06-05 16:03:03 -07003443 /* Exclude state transition flags, already notified */
3444 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3445 if (changes)
3446 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447
3448 return ret;
3449}
3450
Stephen Hemmingerf0db2752008-09-30 02:23:58 -07003451/**
3452 * dev_set_mtu - Change maximum transfer unit
3453 * @dev: device
3454 * @new_mtu: new transfer unit
3455 *
3456 * Change the maximum transfer size of the network device.
3457 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458int dev_set_mtu(struct net_device *dev, int new_mtu)
3459{
3460 int err;
3461
3462 if (new_mtu == dev->mtu)
3463 return 0;
3464
3465 /* MTU must be positive. */
3466 if (new_mtu < 0)
3467 return -EINVAL;
3468
3469 if (!netif_device_present(dev))
3470 return -ENODEV;
3471
3472 err = 0;
3473 if (dev->change_mtu)
3474 err = dev->change_mtu(dev, new_mtu);
3475 else
3476 dev->mtu = new_mtu;
3477 if (!err && dev->flags & IFF_UP)
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003478 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479 return err;
3480}
3481
Stephen Hemmingerf0db2752008-09-30 02:23:58 -07003482/**
3483 * dev_set_mac_address - Change Media Access Control Address
3484 * @dev: device
3485 * @sa: new address
3486 *
3487 * Change the hardware (MAC) address of the device
3488 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3490{
3491 int err;
3492
3493 if (!dev->set_mac_address)
3494 return -EOPNOTSUPP;
3495 if (sa->sa_family != dev->type)
3496 return -EINVAL;
3497 if (!netif_device_present(dev))
3498 return -ENODEV;
3499 err = dev->set_mac_address(dev, sa);
3500 if (!err)
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003501 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 return err;
3503}
3504
3505/*
Jeff Garzik14e3e072007-10-08 00:06:32 -07003506 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003507 */
Jeff Garzik14e3e072007-10-08 00:06:32 -07003508static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509{
3510 int err;
Eric W. Biederman881d9662007-09-17 11:56:21 -07003511 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512
3513 if (!dev)
3514 return -ENODEV;
3515
3516 switch (cmd) {
3517 case SIOCGIFFLAGS: /* Get interface flags */
3518 ifr->ifr_flags = dev_get_flags(dev);
3519 return 0;
3520
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521 case SIOCGIFMETRIC: /* Get the metric on the interface
3522 (currently unused) */
3523 ifr->ifr_metric = 0;
3524 return 0;
3525
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526 case SIOCGIFMTU: /* Get the MTU of a device */
3527 ifr->ifr_mtu = dev->mtu;
3528 return 0;
3529
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530 case SIOCGIFHWADDR:
3531 if (!dev->addr_len)
3532 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3533 else
3534 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3535 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3536 ifr->ifr_hwaddr.sa_family = dev->type;
3537 return 0;
3538
Jeff Garzik14e3e072007-10-08 00:06:32 -07003539 case SIOCGIFSLAVE:
3540 err = -EINVAL;
3541 break;
3542
3543 case SIOCGIFMAP:
3544 ifr->ifr_map.mem_start = dev->mem_start;
3545 ifr->ifr_map.mem_end = dev->mem_end;
3546 ifr->ifr_map.base_addr = dev->base_addr;
3547 ifr->ifr_map.irq = dev->irq;
3548 ifr->ifr_map.dma = dev->dma;
3549 ifr->ifr_map.port = dev->if_port;
3550 return 0;
3551
3552 case SIOCGIFINDEX:
3553 ifr->ifr_ifindex = dev->ifindex;
3554 return 0;
3555
3556 case SIOCGIFTXQLEN:
3557 ifr->ifr_qlen = dev->tx_queue_len;
3558 return 0;
3559
3560 default:
3561 /* dev_ioctl() should ensure this case
3562 * is never reached
3563 */
3564 WARN_ON(1);
3565 err = -EINVAL;
3566 break;
3567
3568 }
3569 return err;
3570}
3571
3572/*
3573 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3574 */
3575static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3576{
3577 int err;
3578 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3579
3580 if (!dev)
3581 return -ENODEV;
3582
3583 switch (cmd) {
3584 case SIOCSIFFLAGS: /* Set interface flags */
3585 return dev_change_flags(dev, ifr->ifr_flags);
3586
3587 case SIOCSIFMETRIC: /* Set the metric on the interface
3588 (currently unused) */
3589 return -EOPNOTSUPP;
3590
3591 case SIOCSIFMTU: /* Set the MTU of a device */
3592 return dev_set_mtu(dev, ifr->ifr_mtu);
3593
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 case SIOCSIFHWADDR:
3595 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3596
3597 case SIOCSIFHWBROADCAST:
3598 if (ifr->ifr_hwaddr.sa_family != dev->type)
3599 return -EINVAL;
3600 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3601 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
Pavel Emelyanov056925a2007-09-16 15:42:43 -07003602 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603 return 0;
3604
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605 case SIOCSIFMAP:
3606 if (dev->set_config) {
3607 if (!netif_device_present(dev))
3608 return -ENODEV;
3609 return dev->set_config(dev, &ifr->ifr_map);
3610 }
3611 return -EOPNOTSUPP;
3612
3613 case SIOCADDMULTI:
Patrick McHardy61ee6bd2008-03-26 02:12:11 -07003614 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3616 return -EINVAL;
3617 if (!netif_device_present(dev))
3618 return -ENODEV;
3619 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3620 dev->addr_len, 1);
3621
3622 case SIOCDELMULTI:
Patrick McHardy61ee6bd2008-03-26 02:12:11 -07003623 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07003624 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3625 return -EINVAL;
3626 if (!netif_device_present(dev))
3627 return -ENODEV;
3628 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3629 dev->addr_len, 1);
3630
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631 case SIOCSIFTXQLEN:
3632 if (ifr->ifr_qlen < 0)
3633 return -EINVAL;
3634 dev->tx_queue_len = ifr->ifr_qlen;
3635 return 0;
3636
3637 case SIOCSIFNAME:
3638 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3639 return dev_change_name(dev, ifr->ifr_newname);
3640
3641 /*
3642 * Unknown or private ioctl
3643 */
3644
3645 default:
3646 if ((cmd >= SIOCDEVPRIVATE &&
3647 cmd <= SIOCDEVPRIVATE + 15) ||
3648 cmd == SIOCBONDENSLAVE ||
3649 cmd == SIOCBONDRELEASE ||
3650 cmd == SIOCBONDSETHWADDR ||
3651 cmd == SIOCBONDSLAVEINFOQUERY ||
3652 cmd == SIOCBONDINFOQUERY ||
3653 cmd == SIOCBONDCHANGEACTIVE ||
3654 cmd == SIOCGMIIPHY ||
3655 cmd == SIOCGMIIREG ||
3656 cmd == SIOCSMIIREG ||
3657 cmd == SIOCBRADDIF ||
3658 cmd == SIOCBRDELIF ||
3659 cmd == SIOCWANDEV) {
3660 err = -EOPNOTSUPP;
3661 if (dev->do_ioctl) {
3662 if (netif_device_present(dev))
3663 err = dev->do_ioctl(dev, ifr,
3664 cmd);
3665 else
3666 err = -ENODEV;
3667 }
3668 } else
3669 err = -EINVAL;
3670
3671 }
3672 return err;
3673}
3674
3675/*
3676 * This function handles all "interface"-type I/O control requests. The actual
3677 * 'doing' part of this is dev_ifsioc above.
3678 */
3679
3680/**
3681 * dev_ioctl - network device ioctl
Randy Dunlapc4ea43c2007-10-12 21:17:49 -07003682 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683 * @cmd: command to issue
3684 * @arg: pointer to a struct ifreq in user space
3685 *
3686 * Issue ioctl functions to devices. This is normally called by the
3687 * user space syscall interfaces but can sometimes be useful for
3688 * other purposes. The return value is the return from the syscall if
3689 * positive or a negative errno code on error.
3690 */
3691
Eric W. Biederman881d9662007-09-17 11:56:21 -07003692int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003693{
3694 struct ifreq ifr;
3695 int ret;
3696 char *colon;
3697
3698 /* One special case: SIOCGIFCONF takes ifconf argument
3699 and requires shared lock, because it sleeps writing
3700 to user space.
3701 */
3702
3703 if (cmd == SIOCGIFCONF) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003704 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003705 ret = dev_ifconf(net, (char __user *) arg);
Stephen Hemminger6756ae42006-03-20 22:23:58 -08003706 rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 return ret;
3708 }
3709 if (cmd == SIOCGIFNAME)
Eric W. Biederman881d9662007-09-17 11:56:21 -07003710 return dev_ifname(net, (struct ifreq __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003711
3712 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3713 return -EFAULT;
3714
3715 ifr.ifr_name[IFNAMSIZ-1] = 0;
3716
3717 colon = strchr(ifr.ifr_name, ':');
3718 if (colon)
3719 *colon = 0;
3720
3721 /*
3722 * See which interface the caller is talking about.
3723 */
3724
3725 switch (cmd) {
3726 /*
3727 * These ioctl calls:
3728 * - can be done by all.
3729 * - atomic and do not require locking.
3730 * - return a value
3731 */
3732 case SIOCGIFFLAGS:
3733 case SIOCGIFMETRIC:
3734 case SIOCGIFMTU:
3735 case SIOCGIFHWADDR:
3736 case SIOCGIFSLAVE:
3737 case SIOCGIFMAP:
3738 case SIOCGIFINDEX:
3739 case SIOCGIFTXQLEN:
Eric W. Biederman881d9662007-09-17 11:56:21 -07003740 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 read_lock(&dev_base_lock);
Jeff Garzik14e3e072007-10-08 00:06:32 -07003742 ret = dev_ifsioc_locked(net, &ifr, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743 read_unlock(&dev_base_lock);
3744 if (!ret) {
3745 if (colon)
3746 *colon = ':';
3747 if (copy_to_user(arg, &ifr,
3748 sizeof(struct ifreq)))
3749 ret = -EFAULT;
3750 }
3751 return ret;
3752
3753 case SIOCETHTOOL:
Eric W. Biederman881d9662007-09-17 11:56:21 -07003754 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003756 ret = dev_ethtool(net, &ifr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757 rtnl_unlock();
3758 if (!ret) {
3759 if (colon)
3760 *colon = ':';
3761 if (copy_to_user(arg, &ifr,
3762 sizeof(struct ifreq)))
3763 ret = -EFAULT;
3764 }
3765 return ret;
3766
3767 /*
3768 * These ioctl calls:
3769 * - require superuser power.
3770 * - require strict serialization.
3771 * - return a value
3772 */
3773 case SIOCGMIIPHY:
3774 case SIOCGMIIREG:
3775 case SIOCSIFNAME:
3776 if (!capable(CAP_NET_ADMIN))
3777 return -EPERM;
Eric W. Biederman881d9662007-09-17 11:56:21 -07003778 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003780 ret = dev_ifsioc(net, &ifr, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 rtnl_unlock();
3782 if (!ret) {
3783 if (colon)
3784 *colon = ':';
3785 if (copy_to_user(arg, &ifr,
3786 sizeof(struct ifreq)))
3787 ret = -EFAULT;
3788 }
3789 return ret;
3790
3791 /*
3792 * These ioctl calls:
3793 * - require superuser power.
3794 * - require strict serialization.
3795 * - do not return a value
3796 */
3797 case SIOCSIFFLAGS:
3798 case SIOCSIFMETRIC:
3799 case SIOCSIFMTU:
3800 case SIOCSIFMAP:
3801 case SIOCSIFHWADDR:
3802 case SIOCSIFSLAVE:
3803 case SIOCADDMULTI:
3804 case SIOCDELMULTI:
3805 case SIOCSIFHWBROADCAST:
3806 case SIOCSIFTXQLEN:
3807 case SIOCSMIIREG:
3808 case SIOCBONDENSLAVE:
3809 case SIOCBONDRELEASE:
3810 case SIOCBONDSETHWADDR:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811 case SIOCBONDCHANGEACTIVE:
3812 case SIOCBRADDIF:
3813 case SIOCBRDELIF:
3814 if (!capable(CAP_NET_ADMIN))
3815 return -EPERM;
Thomas Grafcabcac02006-01-24 12:46:33 -08003816 /* fall through */
3817 case SIOCBONDSLAVEINFOQUERY:
3818 case SIOCBONDINFOQUERY:
Eric W. Biederman881d9662007-09-17 11:56:21 -07003819 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003821 ret = dev_ifsioc(net, &ifr, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822 rtnl_unlock();
3823 return ret;
3824
3825 case SIOCGIFMEM:
3826 /* Get the per device memory space. We can add this but
3827 * currently do not support it */
3828 case SIOCSIFMEM:
3829 /* Set the per device memory buffer space.
3830 * Not applicable in our case */
3831 case SIOCSIFLINK:
3832 return -EINVAL;
3833
3834 /*
3835 * Unknown or private ioctl.
3836 */
3837 default:
3838 if (cmd == SIOCWANDEV ||
3839 (cmd >= SIOCDEVPRIVATE &&
3840 cmd <= SIOCDEVPRIVATE + 15)) {
Eric W. Biederman881d9662007-09-17 11:56:21 -07003841 dev_load(net, ifr.ifr_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003842 rtnl_lock();
Eric W. Biederman881d9662007-09-17 11:56:21 -07003843 ret = dev_ifsioc(net, &ifr, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844 rtnl_unlock();
3845 if (!ret && copy_to_user(arg, &ifr,
3846 sizeof(struct ifreq)))
3847 ret = -EFAULT;
3848 return ret;
3849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850 /* Take care of Wireless Extensions */
Johannes Berg295f4a12007-04-26 20:43:56 -07003851 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
Eric W. Biederman881d9662007-09-17 11:56:21 -07003852 return wext_handle_ioctl(net, &ifr, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853 return -EINVAL;
3854 }
3855}
3856
3857
3858/**
3859 * dev_new_index - allocate an ifindex
Randy Dunlapc4ea43c2007-10-12 21:17:49 -07003860 * @net: the applicable net namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861 *
3862 * Returns a suitable unique value for a new device interface
3863 * number. The caller must hold the rtnl semaphore or the
3864 * dev_base_lock to be sure it remains unique.
3865 */
Eric W. Biederman881d9662007-09-17 11:56:21 -07003866static int dev_new_index(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867{
3868 static int ifindex;
3869 for (;;) {
3870 if (++ifindex <= 0)
3871 ifindex = 1;
Eric W. Biederman881d9662007-09-17 11:56:21 -07003872 if (!__dev_get_by_index(net, ifindex))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873 return ifindex;
3874 }
3875}
3876
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877/* Delayed registration/unregisteration */
Denis Cheng3b5b34f2007-12-07 00:49:17 -08003878static LIST_HEAD(net_todo_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879
Stephen Hemminger6f05f622007-03-08 20:46:03 -08003880static void net_set_todo(struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003881{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 list_add_tail(&dev->todo_list, &net_todo_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003883}
3884
Daniel Lezcano93ee31f2007-10-30 15:38:18 -07003885static void rollback_registered(struct net_device *dev)
3886{
3887 BUG_ON(dev_boot_phase);
3888 ASSERT_RTNL();
3889
3890 /* Some devices call without registering for initialization unwind. */
3891 if (dev->reg_state == NETREG_UNINITIALIZED) {
3892 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3893 "was registered\n", dev->name, dev);
3894
3895 WARN_ON(1);
3896 return;
3897 }
3898
3899 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3900
3901 /* If device is running, close it first. */
3902 dev_close(dev);
3903
3904 /* And unlink it from device chain. */
3905 unlist_netdevice(dev);
3906
3907 dev->reg_state = NETREG_UNREGISTERING;
3908
3909 synchronize_net();
3910
3911 /* Shutdown queueing discipline. */
3912 dev_shutdown(dev);
3913
3914
3915 /* Notify protocols, that we are about to destroy
3916 this device. They should clean all the things.
3917 */
3918 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3919
3920 /*
3921 * Flush the unicast and multicast chains
3922 */
3923 dev_addr_discard(dev);
3924
3925 if (dev->uninit)
3926 dev->uninit(dev);
3927
3928 /* Notifier chain MUST detach us from master device. */
Ilpo Järvinen547b7922008-07-25 21:43:18 -07003929 WARN_ON(dev->master);
Daniel Lezcano93ee31f2007-10-30 15:38:18 -07003930
3931 /* Remove entries from kobject tree */
3932 netdev_unregister_kobject(dev);
3933
3934 synchronize_net();
3935
3936 dev_put(dev);
3937}
3938
David S. Millere8a04642008-07-17 00:34:19 -07003939static void __netdev_init_queue_locks_one(struct net_device *dev,
3940 struct netdev_queue *dev_queue,
3941 void *_unused)
David S. Millerc773e842008-07-08 23:13:53 -07003942{
3943 spin_lock_init(&dev_queue->_xmit_lock);
David S. Millercf508b12008-07-22 14:16:42 -07003944 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
David S. Millerc773e842008-07-08 23:13:53 -07003945 dev_queue->xmit_lock_owner = -1;
3946}
3947
3948static void netdev_init_queue_locks(struct net_device *dev)
3949{
David S. Millere8a04642008-07-17 00:34:19 -07003950 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3951 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
David S. Millerc773e842008-07-08 23:13:53 -07003952}
3953
Herbert Xub63365a2008-10-23 01:11:29 -07003954unsigned long netdev_fix_features(unsigned long features, const char *name)
3955{
3956 /* Fix illegal SG+CSUM combinations. */
3957 if ((features & NETIF_F_SG) &&
3958 !(features & NETIF_F_ALL_CSUM)) {
3959 if (name)
3960 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
3961 "checksum feature.\n", name);
3962 features &= ~NETIF_F_SG;
3963 }
3964
3965 /* TSO requires that SG is present as well. */
3966 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
3967 if (name)
3968 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
3969 "SG feature.\n", name);
3970 features &= ~NETIF_F_TSO;
3971 }
3972
3973 if (features & NETIF_F_UFO) {
3974 if (!(features & NETIF_F_GEN_CSUM)) {
3975 if (name)
3976 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3977 "since no NETIF_F_HW_CSUM feature.\n",
3978 name);
3979 features &= ~NETIF_F_UFO;
3980 }
3981
3982 if (!(features & NETIF_F_SG)) {
3983 if (name)
3984 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3985 "since no NETIF_F_SG feature.\n", name);
3986 features &= ~NETIF_F_UFO;
3987 }
3988 }
3989
3990 return features;
3991}
3992EXPORT_SYMBOL(netdev_fix_features);
3993
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994/**
3995 * register_netdevice - register a network device
3996 * @dev: device to register
3997 *
3998 * Take a completed network device structure and add it to the kernel
3999 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4000 * chain. 0 is returned on success. A negative errno code is returned
4001 * on a failure to set up the device, or if the name is a duplicate.
4002 *
4003 * Callers must hold the rtnl semaphore. You may want
4004 * register_netdev() instead of this.
4005 *
4006 * BUGS:
4007 * The locking appears insufficient to guarantee two parallel registers
4008 * will not get the same name.
4009 */
4010
4011int register_netdevice(struct net_device *dev)
4012{
4013 struct hlist_head *head;
4014 struct hlist_node *p;
4015 int ret;
Eric W. Biederman881d9662007-09-17 11:56:21 -07004016 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017
4018 BUG_ON(dev_boot_phase);
4019 ASSERT_RTNL();
4020
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004021 might_sleep();
4022
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 /* When net_device's are persistent, this will be fatal. */
4024 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09004025 BUG_ON(!dev_net(dev));
4026 net = dev_net(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004027
David S. Millerf1f28aa2008-07-15 00:08:33 -07004028 spin_lock_init(&dev->addr_list_lock);
David S. Millercf508b12008-07-22 14:16:42 -07004029 netdev_set_addr_lockdep_class(dev);
David S. Millerc773e842008-07-08 23:13:53 -07004030 netdev_init_queue_locks(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032 dev->iflink = -1;
4033
4034 /* Init, if this function is available */
4035 if (dev->init) {
4036 ret = dev->init(dev);
4037 if (ret) {
4038 if (ret > 0)
4039 ret = -EIO;
Adrian Bunk90833aa2006-11-13 16:02:22 -08004040 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041 }
4042 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09004043
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044 if (!dev_valid_name(dev->name)) {
4045 ret = -EINVAL;
Herbert Xu7ce1b0e2007-07-30 16:29:40 -07004046 goto err_uninit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047 }
4048
Eric W. Biederman881d9662007-09-17 11:56:21 -07004049 dev->ifindex = dev_new_index(net);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050 if (dev->iflink == -1)
4051 dev->iflink = dev->ifindex;
4052
4053 /* Check for existence of name */
Eric W. Biederman881d9662007-09-17 11:56:21 -07004054 head = dev_name_hash(net, dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055 hlist_for_each(p, head) {
4056 struct net_device *d
4057 = hlist_entry(p, struct net_device, name_hlist);
4058 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4059 ret = -EEXIST;
Herbert Xu7ce1b0e2007-07-30 16:29:40 -07004060 goto err_uninit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09004062 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063
Stephen Hemmingerd212f872007-06-27 00:47:37 -07004064 /* Fix illegal checksum combinations */
4065 if ((dev->features & NETIF_F_HW_CSUM) &&
4066 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4067 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4068 dev->name);
4069 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4070 }
4071
4072 if ((dev->features & NETIF_F_NO_CSUM) &&
4073 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4074 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4075 dev->name);
4076 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4077 }
4078
Herbert Xub63365a2008-10-23 01:11:29 -07004079 dev->features = netdev_fix_features(dev->features, dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080
Lennert Buytenheke5a4a722008-08-03 01:23:10 -07004081 /* Enable software GSO if SG is supported. */
4082 if (dev->features & NETIF_F_SG)
4083 dev->features |= NETIF_F_GSO;
4084
Daniel Lezcanoaaf8cdc2008-05-02 17:00:58 -07004085 netdev_initialize_kobject(dev);
Eric W. Biederman8b41d182007-09-26 22:02:53 -07004086 ret = netdev_register_kobject(dev);
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004087 if (ret)
Herbert Xu7ce1b0e2007-07-30 16:29:40 -07004088 goto err_uninit;
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004089 dev->reg_state = NETREG_REGISTERED;
4090
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 /*
4092 * Default initial state at registry is that the
4093 * device is present.
4094 */
4095
4096 set_bit(__LINK_STATE_PRESENT, &dev->state);
4097
Linus Torvalds1da177e2005-04-16 15:20:36 -07004098 dev_init_scheduler(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004099 dev_hold(dev);
Eric W. Biedermance286d32007-09-12 13:53:49 +02004100 list_netdevice(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004101
4102 /* Notify protocols, that a new device appeared. */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07004103 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
Herbert Xufcc5a032007-07-30 17:03:38 -07004104 ret = notifier_to_errno(ret);
Daniel Lezcano93ee31f2007-10-30 15:38:18 -07004105 if (ret) {
4106 rollback_registered(dev);
4107 dev->reg_state = NETREG_UNREGISTERED;
4108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109
4110out:
4111 return ret;
Herbert Xu7ce1b0e2007-07-30 16:29:40 -07004112
4113err_uninit:
4114 if (dev->uninit)
4115 dev->uninit(dev);
4116 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117}
4118
4119/**
4120 * register_netdev - register a network device
4121 * @dev: device to register
4122 *
4123 * Take a completed network device structure and add it to the kernel
4124 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4125 * chain. 0 is returned on success. A negative errno code is returned
4126 * on a failure to set up the device, or if the name is a duplicate.
4127 *
Borislav Petkov38b4da32007-04-20 22:14:10 -07004128 * This is a wrapper around register_netdevice that takes the rtnl semaphore
Linus Torvalds1da177e2005-04-16 15:20:36 -07004129 * and expands the device name if you passed a format string to
4130 * alloc_netdev.
4131 */
4132int register_netdev(struct net_device *dev)
4133{
4134 int err;
4135
4136 rtnl_lock();
4137
4138 /*
4139 * If the name is a format string the caller wants us to do a
4140 * name allocation.
4141 */
4142 if (strchr(dev->name, '%')) {
4143 err = dev_alloc_name(dev, dev->name);
4144 if (err < 0)
4145 goto out;
4146 }
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09004147
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148 err = register_netdevice(dev);
4149out:
4150 rtnl_unlock();
4151 return err;
4152}
4153EXPORT_SYMBOL(register_netdev);
4154
4155/*
4156 * netdev_wait_allrefs - wait until all references are gone.
4157 *
4158 * This is called when unregistering network devices.
4159 *
4160 * Any protocol or device that holds a reference should register
4161 * for netdevice notification, and cleanup and put back the
4162 * reference if they receive an UNREGISTER event.
4163 * We can get stuck here if buggy protocols don't correctly
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09004164 * call dev_put.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004165 */
4166static void netdev_wait_allrefs(struct net_device *dev)
4167{
4168 unsigned long rebroadcast_time, warning_time;
4169
4170 rebroadcast_time = warning_time = jiffies;
4171 while (atomic_read(&dev->refcnt) != 0) {
4172 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
Stephen Hemminger6756ae42006-03-20 22:23:58 -08004173 rtnl_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004174
4175 /* Rebroadcast unregister notification */
Pavel Emelyanov056925a2007-09-16 15:42:43 -07004176 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177
4178 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4179 &dev->state)) {
4180 /* We must not have linkwatch events
4181 * pending on unregister. If this
4182 * happens, we simply run the queue
4183 * unscheduled, resulting in a noop
4184 * for this device.
4185 */
4186 linkwatch_run_queue();
4187 }
4188
Stephen Hemminger6756ae42006-03-20 22:23:58 -08004189 __rtnl_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004190
4191 rebroadcast_time = jiffies;
4192 }
4193
4194 msleep(250);
4195
4196 if (time_after(jiffies, warning_time + 10 * HZ)) {
4197 printk(KERN_EMERG "unregister_netdevice: "
4198 "waiting for %s to become free. Usage "
4199 "count = %d\n",
4200 dev->name, atomic_read(&dev->refcnt));
4201 warning_time = jiffies;
4202 }
4203 }
4204}
4205
4206/* The sequence is:
4207 *
4208 * rtnl_lock();
4209 * ...
4210 * register_netdevice(x1);
4211 * register_netdevice(x2);
4212 * ...
4213 * unregister_netdevice(y1);
4214 * unregister_netdevice(y2);
4215 * ...
4216 * rtnl_unlock();
4217 * free_netdev(y1);
4218 * free_netdev(y2);
4219 *
Herbert Xu58ec3b42008-10-07 15:50:03 -07004220 * We are invoked by rtnl_unlock().
Linus Torvalds1da177e2005-04-16 15:20:36 -07004221 * This allows us to deal with problems:
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004222 * 1) We can delete sysfs objects which invoke hotplug
Linus Torvalds1da177e2005-04-16 15:20:36 -07004223 * without deadlocking with linkwatch via keventd.
4224 * 2) Since we run with the RTNL semaphore not held, we can sleep
4225 * safely in order to wait for the netdev refcnt to drop to zero.
Herbert Xu58ec3b42008-10-07 15:50:03 -07004226 *
4227 * We must not return until all unregister events added during
4228 * the interval the lock was held have been completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004229 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004230void netdev_run_todo(void)
4231{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07004232 struct list_head list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004233
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234 /* Snapshot list, allow later requests */
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07004235 list_replace_init(&net_todo_list, &list);
Herbert Xu58ec3b42008-10-07 15:50:03 -07004236
4237 __rtnl_unlock();
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07004238
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239 while (!list_empty(&list)) {
4240 struct net_device *dev
4241 = list_entry(list.next, struct net_device, todo_list);
4242 list_del(&dev->todo_list);
4243
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004244 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245 printk(KERN_ERR "network todo '%s' but state %d\n",
4246 dev->name, dev->reg_state);
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004247 dump_stack();
4248 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004249 }
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004250
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004251 dev->reg_state = NETREG_UNREGISTERED;
4252
Stephen Hemminger6e583ce2008-08-03 21:29:57 -07004253 on_each_cpu(flush_backlog, dev, 1);
4254
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004255 netdev_wait_allrefs(dev);
4256
4257 /* paranoia */
4258 BUG_ON(atomic_read(&dev->refcnt));
Ilpo Järvinen547b7922008-07-25 21:43:18 -07004259 WARN_ON(dev->ip_ptr);
4260 WARN_ON(dev->ip6_ptr);
4261 WARN_ON(dev->dn_ptr);
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004262
Stephen Hemmingerb17a7c12006-05-10 13:21:17 -07004263 if (dev->destructor)
4264 dev->destructor(dev);
Stephen Hemminger9093bbb2007-05-19 15:39:25 -07004265
4266 /* Free network device */
4267 kobject_put(&dev->dev.kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004268 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004269}
4270
Rusty Russell5a1b5892007-04-28 21:04:03 -07004271static struct net_device_stats *internal_stats(struct net_device *dev)
Rusty Russellc45d2862007-03-28 14:29:08 -07004272{
Rusty Russell5a1b5892007-04-28 21:04:03 -07004273 return &dev->stats;
Rusty Russellc45d2862007-03-28 14:29:08 -07004274}
4275
David S. Millerdc2b4842008-07-08 17:18:23 -07004276static void netdev_init_one_queue(struct net_device *dev,
David S. Millere8a04642008-07-17 00:34:19 -07004277 struct netdev_queue *queue,
4278 void *_unused)
David S. Millerdc2b4842008-07-08 17:18:23 -07004279{
David S. Millerdc2b4842008-07-08 17:18:23 -07004280 queue->dev = dev;
4281}
4282
David S. Millerbb949fb2008-07-08 16:55:56 -07004283static void netdev_init_queues(struct net_device *dev)
4284{
David S. Millere8a04642008-07-17 00:34:19 -07004285 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4286 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
David S. Millerc3f26a22008-07-31 16:58:50 -07004287 spin_lock_init(&dev->tx_global_lock);
David S. Millerbb949fb2008-07-08 16:55:56 -07004288}
4289
Linus Torvalds1da177e2005-04-16 15:20:36 -07004290/**
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07004291 * alloc_netdev_mq - allocate network device
Linus Torvalds1da177e2005-04-16 15:20:36 -07004292 * @sizeof_priv: size of private data to allocate space for
4293 * @name: device name format string
4294 * @setup: callback to initialize device
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07004295 * @queue_count: the number of subqueues to allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07004296 *
4297 * Allocates a struct net_device with private data area for driver use
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07004298 * and performs basic initialization. Also allocates subquue structs
4299 * for each queue on the device at the end of the netdevice.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004300 */
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07004301struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4302 void (*setup)(struct net_device *), unsigned int queue_count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004303{
David S. Millere8a04642008-07-17 00:34:19 -07004304 struct netdev_queue *tx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004305 struct net_device *dev;
Stephen Hemminger79439862008-07-21 13:28:44 -07004306 size_t alloc_size;
David S. Millere8a04642008-07-17 00:34:19 -07004307 void *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308
Stephen Hemmingerb6fe17d2006-08-29 17:06:13 -07004309 BUG_ON(strlen(name) >= sizeof(dev->name));
4310
David S. Millerfd2ea0a2008-07-17 01:56:23 -07004311 alloc_size = sizeof(struct net_device);
Alexey Dobriyand1643d22008-04-18 15:43:32 -07004312 if (sizeof_priv) {
4313 /* ensure 32-byte alignment of private area */
4314 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4315 alloc_size += sizeof_priv;
4316 }
4317 /* ensure 32-byte alignment of whole construct */
4318 alloc_size += NETDEV_ALIGN_CONST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004319
Paolo 'Blaisorblade' Giarrusso31380de2006-04-06 22:38:28 -07004320 p = kzalloc(alloc_size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004321 if (!p) {
Stephen Hemmingerb6fe17d2006-08-29 17:06:13 -07004322 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004323 return NULL;
4324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004325
Stephen Hemminger79439862008-07-21 13:28:44 -07004326 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
David S. Millere8a04642008-07-17 00:34:19 -07004327 if (!tx) {
4328 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4329 "tx qdiscs.\n");
4330 kfree(p);
4331 return NULL;
4332 }
4333
Linus Torvalds1da177e2005-04-16 15:20:36 -07004334 dev = (struct net_device *)
4335 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4336 dev->padded = (char *)dev - (char *)p;
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09004337 dev_net_set(dev, &init_net);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004338
David S. Millere8a04642008-07-17 00:34:19 -07004339 dev->_tx = tx;
4340 dev->num_tx_queues = queue_count;
David S. Millerfd2ea0a2008-07-17 01:56:23 -07004341 dev->real_num_tx_queues = queue_count;
David S. Millere8a04642008-07-17 00:34:19 -07004342
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07004343 if (sizeof_priv) {
4344 dev->priv = ((char *)dev +
David S. Millerfd2ea0a2008-07-17 01:56:23 -07004345 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07004346 & ~NETDEV_ALIGN_CONST));
4347 }
4348
Peter P Waskiewicz Jr82cc1a72008-03-21 03:43:19 -07004349 dev->gso_max_size = GSO_MAX_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004350
David S. Millerbb949fb2008-07-08 16:55:56 -07004351 netdev_init_queues(dev);
4352
Rusty Russell5a1b5892007-04-28 21:04:03 -07004353 dev->get_stats = internal_stats;
Stephen Hemmingerbea33482007-10-03 16:41:36 -07004354 netpoll_netdev_init(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355 setup(dev);
4356 strcpy(dev->name, name);
4357 return dev;
4358}
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07004359EXPORT_SYMBOL(alloc_netdev_mq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004360
4361/**
4362 * free_netdev - free network device
4363 * @dev: device
4364 *
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09004365 * This function does the last stage of destroying an allocated device
4366 * interface. The reference to the device object is released.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367 * If this is the last reference then it will be freed.
4368 */
4369void free_netdev(struct net_device *dev)
4370{
Denis V. Lunevf3005d72008-04-16 02:02:18 -07004371 release_net(dev_net(dev));
4372
David S. Millere8a04642008-07-17 00:34:19 -07004373 kfree(dev->_tx);
4374
Stephen Hemminger3041a062006-05-26 13:25:24 -07004375 /* Compatibility with error handling in drivers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004376 if (dev->reg_state == NETREG_UNINITIALIZED) {
4377 kfree((char *)dev - dev->padded);
4378 return;
4379 }
4380
4381 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4382 dev->reg_state = NETREG_RELEASED;
4383
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07004384 /* will free via device release */
4385 put_device(&dev->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004386}
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09004387
Stephen Hemmingerf0db2752008-09-30 02:23:58 -07004388/**
4389 * synchronize_net - Synchronize with packet receive processing
4390 *
4391 * Wait for packets currently being received to be done.
4392 * Does not block later packets from starting.
4393 */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09004394void synchronize_net(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004395{
4396 might_sleep();
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07004397 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004398}
4399
4400/**
4401 * unregister_netdevice - remove device from the kernel
4402 * @dev: device
4403 *
4404 * This function shuts down a device interface and removes it
Wang Chend59b54b2007-12-11 02:28:03 -08004405 * from the kernel tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004406 *
4407 * Callers must hold the rtnl semaphore. You may want
4408 * unregister_netdev() instead of this.
4409 */
4410
Stephen Hemminger22f8cde2007-02-07 00:09:58 -08004411void unregister_netdevice(struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412{
Herbert Xua6620712007-12-12 19:21:56 -08004413 ASSERT_RTNL();
4414
Daniel Lezcano93ee31f2007-10-30 15:38:18 -07004415 rollback_registered(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416 /* Finish processing unregister after unlock */
4417 net_set_todo(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004418}
4419
4420/**
4421 * unregister_netdev - remove device from the kernel
4422 * @dev: device
4423 *
4424 * This function shuts down a device interface and removes it
Wang Chend59b54b2007-12-11 02:28:03 -08004425 * from the kernel tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426 *
4427 * This is just a wrapper for unregister_netdevice that takes
4428 * the rtnl semaphore. In general you want to use this and not
4429 * unregister_netdevice.
4430 */
4431void unregister_netdev(struct net_device *dev)
4432{
4433 rtnl_lock();
4434 unregister_netdevice(dev);
4435 rtnl_unlock();
4436}
4437
4438EXPORT_SYMBOL(unregister_netdev);
4439
Eric W. Biedermance286d32007-09-12 13:53:49 +02004440/**
4441 * dev_change_net_namespace - move device to different nethost namespace
4442 * @dev: device
4443 * @net: network namespace
4444 * @pat: If not NULL name pattern to try if the current device name
4445 * is already taken in the destination network namespace.
4446 *
4447 * This function shuts down a device interface and moves it
4448 * to a new network namespace. On success 0 is returned, on
4449 * a failure a netagive errno code is returned.
4450 *
4451 * Callers must hold the rtnl semaphore.
4452 */
4453
4454int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4455{
4456 char buf[IFNAMSIZ];
4457 const char *destname;
4458 int err;
4459
4460 ASSERT_RTNL();
4461
4462 /* Don't allow namespace local devices to be moved. */
4463 err = -EINVAL;
4464 if (dev->features & NETIF_F_NETNS_LOCAL)
4465 goto out;
4466
4467 /* Ensure the device has been registrered */
4468 err = -EINVAL;
4469 if (dev->reg_state != NETREG_REGISTERED)
4470 goto out;
4471
4472 /* Get out if there is nothing todo */
4473 err = 0;
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09004474 if (net_eq(dev_net(dev), net))
Eric W. Biedermance286d32007-09-12 13:53:49 +02004475 goto out;
4476
4477 /* Pick the destination device name, and ensure
4478 * we can use it in the destination network namespace.
4479 */
4480 err = -EEXIST;
4481 destname = dev->name;
4482 if (__dev_get_by_name(net, destname)) {
4483 /* We get here if we can't use the current device name */
4484 if (!pat)
4485 goto out;
4486 if (!dev_valid_name(pat))
4487 goto out;
4488 if (strchr(pat, '%')) {
4489 if (__dev_alloc_name(net, pat, buf) < 0)
4490 goto out;
4491 destname = buf;
4492 } else
4493 destname = pat;
4494 if (__dev_get_by_name(net, destname))
4495 goto out;
4496 }
4497
4498 /*
4499 * And now a mini version of register_netdevice unregister_netdevice.
4500 */
4501
4502 /* If device is running close it first. */
Pavel Emelyanov9b772652007-10-10 02:49:09 -07004503 dev_close(dev);
Eric W. Biedermance286d32007-09-12 13:53:49 +02004504
4505 /* And unlink it from device chain */
4506 err = -ENODEV;
4507 unlist_netdevice(dev);
4508
4509 synchronize_net();
4510
4511 /* Shutdown queueing discipline. */
4512 dev_shutdown(dev);
4513
4514 /* Notify protocols, that we are about to destroy
4515 this device. They should clean all the things.
4516 */
4517 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4518
4519 /*
4520 * Flush the unicast and multicast chains
4521 */
4522 dev_addr_discard(dev);
4523
4524 /* Actually switch the network namespace */
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09004525 dev_net_set(dev, net);
Eric W. Biedermance286d32007-09-12 13:53:49 +02004526
4527 /* Assign the new device name */
4528 if (destname != dev->name)
4529 strcpy(dev->name, destname);
4530
4531 /* If there is an ifindex conflict assign a new one */
4532 if (__dev_get_by_index(net, dev->ifindex)) {
4533 int iflink = (dev->iflink == dev->ifindex);
4534 dev->ifindex = dev_new_index(net);
4535 if (iflink)
4536 dev->iflink = dev->ifindex;
4537 }
4538
Eric W. Biederman8b41d182007-09-26 22:02:53 -07004539 /* Fixup kobjects */
Daniel Lezcanoaaf8cdc2008-05-02 17:00:58 -07004540 netdev_unregister_kobject(dev);
4541 err = netdev_register_kobject(dev);
Eric W. Biederman8b41d182007-09-26 22:02:53 -07004542 WARN_ON(err);
Eric W. Biedermance286d32007-09-12 13:53:49 +02004543
4544 /* Add the device back in the hashes */
4545 list_netdevice(dev);
4546
4547 /* Notify protocols, that a new device appeared. */
4548 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4549
4550 synchronize_net();
4551 err = 0;
4552out:
4553 return err;
4554}
4555
Linus Torvalds1da177e2005-04-16 15:20:36 -07004556static int dev_cpu_callback(struct notifier_block *nfb,
4557 unsigned long action,
4558 void *ocpu)
4559{
4560 struct sk_buff **list_skb;
David S. Miller37437bb2008-07-16 02:15:04 -07004561 struct Qdisc **list_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004562 struct sk_buff *skb;
4563 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4564 struct softnet_data *sd, *oldsd;
4565
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004566 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004567 return NOTIFY_OK;
4568
4569 local_irq_disable();
4570 cpu = smp_processor_id();
4571 sd = &per_cpu(softnet_data, cpu);
4572 oldsd = &per_cpu(softnet_data, oldcpu);
4573
4574 /* Find end of our completion_queue. */
4575 list_skb = &sd->completion_queue;
4576 while (*list_skb)
4577 list_skb = &(*list_skb)->next;
4578 /* Append completion queue from offline CPU. */
4579 *list_skb = oldsd->completion_queue;
4580 oldsd->completion_queue = NULL;
4581
4582 /* Find end of our output_queue. */
4583 list_net = &sd->output_queue;
4584 while (*list_net)
4585 list_net = &(*list_net)->next_sched;
4586 /* Append output queue from offline CPU. */
4587 *list_net = oldsd->output_queue;
4588 oldsd->output_queue = NULL;
4589
4590 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4591 local_irq_enable();
4592
4593 /* Process offline CPU's input_pkt_queue */
4594 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4595 netif_rx(skb);
4596
4597 return NOTIFY_OK;
4598}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004599
Chris Leechdb217332006-06-17 21:24:58 -07004600#ifdef CONFIG_NET_DMA
4601/**
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07004602 * net_dma_rebalance - try to maintain one DMA channel per CPU
4603 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4604 *
4605 * This is called when the number of channels allocated to the net_dma client
4606 * changes. The net_dma client tries to have one DMA channel per CPU.
Chris Leechdb217332006-06-17 21:24:58 -07004607 */
Dan Williamsd379b012007-07-09 11:56:42 -07004608
4609static void net_dma_rebalance(struct net_dma *net_dma)
Chris Leechdb217332006-06-17 21:24:58 -07004610{
Dan Williamsd379b012007-07-09 11:56:42 -07004611 unsigned int cpu, i, n, chan_idx;
Chris Leechdb217332006-06-17 21:24:58 -07004612 struct dma_chan *chan;
4613
Dan Williamsd379b012007-07-09 11:56:42 -07004614 if (cpus_empty(net_dma->channel_mask)) {
Chris Leechdb217332006-06-17 21:24:58 -07004615 for_each_online_cpu(cpu)
Alexey Dobriyan29bbd722006-08-02 15:02:31 -07004616 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
Chris Leechdb217332006-06-17 21:24:58 -07004617 return;
4618 }
4619
4620 i = 0;
4621 cpu = first_cpu(cpu_online_map);
4622
Mike Travis0e12f842008-05-12 21:21:13 +02004623 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
Dan Williamsd379b012007-07-09 11:56:42 -07004624 chan = net_dma->channels[chan_idx];
4625
4626 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4627 + (i < (num_online_cpus() %
4628 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
Chris Leechdb217332006-06-17 21:24:58 -07004629
4630 while(n) {
Alexey Dobriyan29bbd722006-08-02 15:02:31 -07004631 per_cpu(softnet_data, cpu).net_dma = chan;
Chris Leechdb217332006-06-17 21:24:58 -07004632 cpu = next_cpu(cpu, cpu_online_map);
4633 n--;
4634 }
4635 i++;
4636 }
Chris Leechdb217332006-06-17 21:24:58 -07004637}
4638
4639/**
4640 * netdev_dma_event - event callback for the net_dma_client
4641 * @client: should always be net_dma_client
Randy Dunlapf4b8ea72006-06-22 16:00:11 -07004642 * @chan: DMA channel for the event
Randy Dunlap0ed72ec2007-07-26 00:03:29 -07004643 * @state: DMA state to be handled
Chris Leechdb217332006-06-17 21:24:58 -07004644 */
Dan Williamsd379b012007-07-09 11:56:42 -07004645static enum dma_state_client
4646netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4647 enum dma_state state)
Chris Leechdb217332006-06-17 21:24:58 -07004648{
Dan Williamsd379b012007-07-09 11:56:42 -07004649 int i, found = 0, pos = -1;
4650 struct net_dma *net_dma =
4651 container_of(client, struct net_dma, client);
4652 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4653
4654 spin_lock(&net_dma->lock);
4655 switch (state) {
4656 case DMA_RESOURCE_AVAILABLE:
Mike Travis0c0b0ac2008-05-02 16:43:08 -07004657 for (i = 0; i < nr_cpu_ids; i++)
Dan Williamsd379b012007-07-09 11:56:42 -07004658 if (net_dma->channels[i] == chan) {
4659 found = 1;
4660 break;
4661 } else if (net_dma->channels[i] == NULL && pos < 0)
4662 pos = i;
4663
4664 if (!found && pos >= 0) {
4665 ack = DMA_ACK;
4666 net_dma->channels[pos] = chan;
4667 cpu_set(pos, net_dma->channel_mask);
4668 net_dma_rebalance(net_dma);
4669 }
Chris Leechdb217332006-06-17 21:24:58 -07004670 break;
4671 case DMA_RESOURCE_REMOVED:
Mike Travis0c0b0ac2008-05-02 16:43:08 -07004672 for (i = 0; i < nr_cpu_ids; i++)
Dan Williamsd379b012007-07-09 11:56:42 -07004673 if (net_dma->channels[i] == chan) {
4674 found = 1;
4675 pos = i;
4676 break;
4677 }
4678
4679 if (found) {
4680 ack = DMA_ACK;
4681 cpu_clear(pos, net_dma->channel_mask);
4682 net_dma->channels[i] = NULL;
4683 net_dma_rebalance(net_dma);
4684 }
Chris Leechdb217332006-06-17 21:24:58 -07004685 break;
4686 default:
4687 break;
4688 }
Dan Williamsd379b012007-07-09 11:56:42 -07004689 spin_unlock(&net_dma->lock);
4690
4691 return ack;
Chris Leechdb217332006-06-17 21:24:58 -07004692}
4693
4694/**
Stephen Hemmingerf0db2752008-09-30 02:23:58 -07004695 * netdev_dma_register - register the networking subsystem as a DMA client
Chris Leechdb217332006-06-17 21:24:58 -07004696 */
4697static int __init netdev_dma_register(void)
4698{
Mike Travis0c0b0ac2008-05-02 16:43:08 -07004699 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4700 GFP_KERNEL);
4701 if (unlikely(!net_dma.channels)) {
4702 printk(KERN_NOTICE
4703 "netdev_dma: no memory for net_dma.channels\n");
4704 return -ENOMEM;
4705 }
Dan Williamsd379b012007-07-09 11:56:42 -07004706 spin_lock_init(&net_dma.lock);
4707 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4708 dma_async_client_register(&net_dma.client);
4709 dma_async_client_chan_request(&net_dma.client);
Chris Leechdb217332006-06-17 21:24:58 -07004710 return 0;
4711}
4712
4713#else
4714static int __init netdev_dma_register(void) { return -ENODEV; }
4715#endif /* CONFIG_NET_DMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004716
Herbert Xu7f353bf2007-08-10 15:47:58 -07004717/**
Herbert Xub63365a2008-10-23 01:11:29 -07004718 * netdev_increment_features - increment feature set by one
4719 * @all: current feature set
4720 * @one: new feature set
4721 * @mask: mask feature set
Herbert Xu7f353bf2007-08-10 15:47:58 -07004722 *
4723 * Computes a new feature set after adding a device with feature set
Herbert Xub63365a2008-10-23 01:11:29 -07004724 * @one to the master device with current feature set @all. Will not
4725 * enable anything that is off in @mask. Returns the new feature set.
Herbert Xu7f353bf2007-08-10 15:47:58 -07004726 */
Herbert Xub63365a2008-10-23 01:11:29 -07004727unsigned long netdev_increment_features(unsigned long all, unsigned long one,
4728 unsigned long mask)
Herbert Xu7f353bf2007-08-10 15:47:58 -07004729{
Herbert Xub63365a2008-10-23 01:11:29 -07004730 /* If device needs checksumming, downgrade to it. */
4731 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4732 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
4733 else if (mask & NETIF_F_ALL_CSUM) {
4734 /* If one device supports v4/v6 checksumming, set for all. */
4735 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
4736 !(all & NETIF_F_GEN_CSUM)) {
4737 all &= ~NETIF_F_ALL_CSUM;
4738 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
4739 }
Herbert Xu7f353bf2007-08-10 15:47:58 -07004740
Herbert Xub63365a2008-10-23 01:11:29 -07004741 /* If one device supports hw checksumming, set for all. */
4742 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
4743 all &= ~NETIF_F_ALL_CSUM;
4744 all |= NETIF_F_HW_CSUM;
4745 }
4746 }
Herbert Xu7f353bf2007-08-10 15:47:58 -07004747
Herbert Xub63365a2008-10-23 01:11:29 -07004748 one |= NETIF_F_ALL_CSUM;
Herbert Xu7f353bf2007-08-10 15:47:58 -07004749
Herbert Xub63365a2008-10-23 01:11:29 -07004750 one |= all & NETIF_F_ONE_FOR_ALL;
4751 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
4752 all |= one & mask & NETIF_F_ONE_FOR_ALL;
Herbert Xu7f353bf2007-08-10 15:47:58 -07004753
4754 return all;
4755}
Herbert Xub63365a2008-10-23 01:11:29 -07004756EXPORT_SYMBOL(netdev_increment_features);
Herbert Xu7f353bf2007-08-10 15:47:58 -07004757
Pavel Emelyanov30d97d32007-09-16 15:40:33 -07004758static struct hlist_head *netdev_create_hash(void)
4759{
4760 int i;
4761 struct hlist_head *hash;
4762
4763 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4764 if (hash != NULL)
4765 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4766 INIT_HLIST_HEAD(&hash[i]);
4767
4768 return hash;
4769}
4770
Eric W. Biederman881d9662007-09-17 11:56:21 -07004771/* Initialize per network namespace state */
Pavel Emelyanov46650792007-10-08 20:38:39 -07004772static int __net_init netdev_init(struct net *net)
Eric W. Biederman881d9662007-09-17 11:56:21 -07004773{
Eric W. Biederman881d9662007-09-17 11:56:21 -07004774 INIT_LIST_HEAD(&net->dev_base_head);
Eric W. Biederman881d9662007-09-17 11:56:21 -07004775
Pavel Emelyanov30d97d32007-09-16 15:40:33 -07004776 net->dev_name_head = netdev_create_hash();
4777 if (net->dev_name_head == NULL)
4778 goto err_name;
Eric W. Biederman881d9662007-09-17 11:56:21 -07004779
Pavel Emelyanov30d97d32007-09-16 15:40:33 -07004780 net->dev_index_head = netdev_create_hash();
4781 if (net->dev_index_head == NULL)
4782 goto err_idx;
Eric W. Biederman881d9662007-09-17 11:56:21 -07004783
4784 return 0;
Pavel Emelyanov30d97d32007-09-16 15:40:33 -07004785
4786err_idx:
4787 kfree(net->dev_name_head);
4788err_name:
4789 return -ENOMEM;
Eric W. Biederman881d9662007-09-17 11:56:21 -07004790}
4791
Stephen Hemmingerf0db2752008-09-30 02:23:58 -07004792/**
4793 * netdev_drivername - network driver for the device
4794 * @dev: network device
4795 * @buffer: buffer for resulting name
4796 * @len: size of buffer
4797 *
4798 * Determine network driver for device.
4799 */
Stephen Hemmingercf04a4c2008-09-30 02:22:14 -07004800char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
Arjan van de Ven6579e572008-07-21 13:31:48 -07004801{
Stephen Hemmingercf04a4c2008-09-30 02:22:14 -07004802 const struct device_driver *driver;
4803 const struct device *parent;
Arjan van de Ven6579e572008-07-21 13:31:48 -07004804
4805 if (len <= 0 || !buffer)
4806 return buffer;
4807 buffer[0] = 0;
4808
4809 parent = dev->dev.parent;
4810
4811 if (!parent)
4812 return buffer;
4813
4814 driver = parent->driver;
4815 if (driver && driver->name)
4816 strlcpy(buffer, driver->name, len);
4817 return buffer;
4818}
4819
Pavel Emelyanov46650792007-10-08 20:38:39 -07004820static void __net_exit netdev_exit(struct net *net)
Eric W. Biederman881d9662007-09-17 11:56:21 -07004821{
4822 kfree(net->dev_name_head);
4823 kfree(net->dev_index_head);
4824}
4825
Denis V. Lunev022cbae2007-11-13 03:23:50 -08004826static struct pernet_operations __net_initdata netdev_net_ops = {
Eric W. Biederman881d9662007-09-17 11:56:21 -07004827 .init = netdev_init,
4828 .exit = netdev_exit,
4829};
4830
Pavel Emelyanov46650792007-10-08 20:38:39 -07004831static void __net_exit default_device_exit(struct net *net)
Eric W. Biedermance286d32007-09-12 13:53:49 +02004832{
4833 struct net_device *dev, *next;
4834 /*
4835 * Push all migratable of the network devices back to the
4836 * initial network namespace
4837 */
4838 rtnl_lock();
4839 for_each_netdev_safe(net, dev, next) {
4840 int err;
Pavel Emelyanovaca51392008-05-08 01:24:25 -07004841 char fb_name[IFNAMSIZ];
Eric W. Biedermance286d32007-09-12 13:53:49 +02004842
4843 /* Ignore unmoveable devices (i.e. loopback) */
4844 if (dev->features & NETIF_F_NETNS_LOCAL)
4845 continue;
4846
4847 /* Push remaing network devices to init_net */
Pavel Emelyanovaca51392008-05-08 01:24:25 -07004848 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4849 err = dev_change_net_namespace(dev, &init_net, fb_name);
Eric W. Biedermance286d32007-09-12 13:53:49 +02004850 if (err) {
Pavel Emelyanovaca51392008-05-08 01:24:25 -07004851 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
Eric W. Biedermance286d32007-09-12 13:53:49 +02004852 __func__, dev->name, err);
Pavel Emelyanovaca51392008-05-08 01:24:25 -07004853 BUG();
Eric W. Biedermance286d32007-09-12 13:53:49 +02004854 }
4855 }
4856 rtnl_unlock();
4857}
4858
Denis V. Lunev022cbae2007-11-13 03:23:50 -08004859static struct pernet_operations __net_initdata default_device_ops = {
Eric W. Biedermance286d32007-09-12 13:53:49 +02004860 .exit = default_device_exit,
4861};
4862
Linus Torvalds1da177e2005-04-16 15:20:36 -07004863/*
4864 * Initialize the DEV module. At boot time this walks the device list and
4865 * unhooks any devices that fail to initialise (normally hardware not
4866 * present) and leaves us with a valid list of present and active devices.
4867 *
4868 */
4869
4870/*
4871 * This is called single threaded during boot, so no need
4872 * to take the rtnl semaphore.
4873 */
4874static int __init net_dev_init(void)
4875{
4876 int i, rc = -ENOMEM;
4877
4878 BUG_ON(!dev_boot_phase);
4879
Linus Torvalds1da177e2005-04-16 15:20:36 -07004880 if (dev_proc_init())
4881 goto out;
4882
Eric W. Biederman8b41d182007-09-26 22:02:53 -07004883 if (netdev_kobject_init())
Linus Torvalds1da177e2005-04-16 15:20:36 -07004884 goto out;
4885
4886 INIT_LIST_HEAD(&ptype_all);
Pavel Emelyanov82d8a862007-11-26 20:12:58 +08004887 for (i = 0; i < PTYPE_HASH_SIZE; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004888 INIT_LIST_HEAD(&ptype_base[i]);
4889
Eric W. Biederman881d9662007-09-17 11:56:21 -07004890 if (register_pernet_subsys(&netdev_net_ops))
4891 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004892
Eric W. Biedermance286d32007-09-12 13:53:49 +02004893 if (register_pernet_device(&default_device_ops))
4894 goto out;
4895
Linus Torvalds1da177e2005-04-16 15:20:36 -07004896 /*
4897 * Initialise the packet receive queues.
4898 */
4899
KAMEZAWA Hiroyuki6f912042006-04-10 22:52:50 -07004900 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004901 struct softnet_data *queue;
4902
4903 queue = &per_cpu(softnet_data, i);
4904 skb_queue_head_init(&queue->input_pkt_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004905 queue->completion_queue = NULL;
4906 INIT_LIST_HEAD(&queue->poll_list);
Stephen Hemmingerbea33482007-10-03 16:41:36 -07004907
4908 queue->backlog.poll = process_backlog;
4909 queue->backlog.weight = weight_p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004910 }
4911
Chris Leechdb217332006-06-17 21:24:58 -07004912 netdev_dma_register();
4913
Linus Torvalds1da177e2005-04-16 15:20:36 -07004914 dev_boot_phase = 0;
4915
Carlos R. Mafra962cf362008-05-15 11:15:37 -03004916 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4917 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918
4919 hotcpu_notifier(dev_cpu_callback, 0);
4920 dst_init();
4921 dev_mcast_init();
4922 rc = 0;
4923out:
4924 return rc;
4925}
4926
4927subsys_initcall(net_dev_init);
4928
4929EXPORT_SYMBOL(__dev_get_by_index);
4930EXPORT_SYMBOL(__dev_get_by_name);
4931EXPORT_SYMBOL(__dev_remove_pack);
Mitch Williamsc2373ee2005-11-09 10:34:45 -08004932EXPORT_SYMBOL(dev_valid_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004933EXPORT_SYMBOL(dev_add_pack);
4934EXPORT_SYMBOL(dev_alloc_name);
4935EXPORT_SYMBOL(dev_close);
4936EXPORT_SYMBOL(dev_get_by_flags);
4937EXPORT_SYMBOL(dev_get_by_index);
4938EXPORT_SYMBOL(dev_get_by_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004939EXPORT_SYMBOL(dev_open);
4940EXPORT_SYMBOL(dev_queue_xmit);
4941EXPORT_SYMBOL(dev_remove_pack);
4942EXPORT_SYMBOL(dev_set_allmulti);
4943EXPORT_SYMBOL(dev_set_promiscuity);
4944EXPORT_SYMBOL(dev_change_flags);
4945EXPORT_SYMBOL(dev_set_mtu);
4946EXPORT_SYMBOL(dev_set_mac_address);
4947EXPORT_SYMBOL(free_netdev);
4948EXPORT_SYMBOL(netdev_boot_setup_check);
4949EXPORT_SYMBOL(netdev_set_master);
4950EXPORT_SYMBOL(netdev_state_change);
4951EXPORT_SYMBOL(netif_receive_skb);
4952EXPORT_SYMBOL(netif_rx);
4953EXPORT_SYMBOL(register_gifconf);
4954EXPORT_SYMBOL(register_netdevice);
4955EXPORT_SYMBOL(register_netdevice_notifier);
4956EXPORT_SYMBOL(skb_checksum_help);
4957EXPORT_SYMBOL(synchronize_net);
4958EXPORT_SYMBOL(unregister_netdevice);
4959EXPORT_SYMBOL(unregister_netdevice_notifier);
4960EXPORT_SYMBOL(net_enable_timestamp);
4961EXPORT_SYMBOL(net_disable_timestamp);
4962EXPORT_SYMBOL(dev_get_flags);
4963
4964#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4965EXPORT_SYMBOL(br_handle_frame_hook);
4966EXPORT_SYMBOL(br_fdb_get_hook);
4967EXPORT_SYMBOL(br_fdb_put_hook);
4968#endif
4969
Linus Torvalds1da177e2005-04-16 15:20:36 -07004970EXPORT_SYMBOL(dev_load);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004971
4972EXPORT_PER_CPU_SYMBOL(softnet_data);