Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * NET3 Protocol independent device support routines. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU General Public License |
| 6 | * as published by the Free Software Foundation; either version |
| 7 | * 2 of the License, or (at your option) any later version. |
| 8 | * |
| 9 | * Derived from the non IP parts of dev.c 1.0.19 |
Jesper Juhl | 02c30a8 | 2005-05-05 16:16:16 -0700 | [diff] [blame] | 10 | * Authors: Ross Biro |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
| 12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> |
| 13 | * |
| 14 | * Additional Authors: |
| 15 | * Florian la Roche <rzsfl@rz.uni-sb.de> |
| 16 | * Alan Cox <gw4pts@gw4pts.ampr.org> |
| 17 | * David Hinds <dahinds@users.sourceforge.net> |
| 18 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> |
| 19 | * Adam Sulmicki <adam@cfar.umd.edu> |
| 20 | * Pekka Riikonen <priikone@poesidon.pspt.fi> |
| 21 | * |
| 22 | * Changes: |
| 23 | * D.J. Barrow : Fixed bug where dev->refcnt gets set |
| 24 | * to 2 if register_netdev gets called |
| 25 | * before net_dev_init & also removed a |
| 26 | * few lines of code in the process. |
| 27 | * Alan Cox : device private ioctl copies fields back. |
| 28 | * Alan Cox : Transmit queue code does relevant |
| 29 | * stunts to keep the queue safe. |
| 30 | * Alan Cox : Fixed double lock. |
| 31 | * Alan Cox : Fixed promisc NULL pointer trap |
| 32 | * ???????? : Support the full private ioctl range |
| 33 | * Alan Cox : Moved ioctl permission check into |
| 34 | * drivers |
| 35 | * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI |
| 36 | * Alan Cox : 100 backlog just doesn't cut it when |
| 37 | * you start doing multicast video 8) |
| 38 | * Alan Cox : Rewrote net_bh and list manager. |
| 39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. |
| 40 | * Alan Cox : Took out transmit every packet pass |
| 41 | * Saved a few bytes in the ioctl handler |
| 42 | * Alan Cox : Network driver sets packet type before |
| 43 | * calling netif_rx. Saves a function |
| 44 | * call a packet. |
| 45 | * Alan Cox : Hashed net_bh() |
| 46 | * Richard Kooijman: Timestamp fixes. |
| 47 | * Alan Cox : Wrong field in SIOCGIFDSTADDR |
| 48 | * Alan Cox : Device lock protection. |
| 49 | * Alan Cox : Fixed nasty side effect of device close |
| 50 | * changes. |
| 51 | * Rudi Cilibrasi : Pass the right thing to |
| 52 | * set_mac_address() |
| 53 | * Dave Miller : 32bit quantity for the device lock to |
| 54 | * make it work out on a Sparc. |
| 55 | * Bjorn Ekwall : Added KERNELD hack. |
| 56 | * Alan Cox : Cleaned up the backlog initialise. |
| 57 | * Craig Metz : SIOCGIFCONF fix if space for under |
| 58 | * 1 device. |
| 59 | * Thomas Bogendoerfer : Return ENODEV for dev_open, if there |
| 60 | * is no device open function. |
| 61 | * Andi Kleen : Fix error reporting for SIOCGIFCONF |
| 62 | * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF |
| 63 | * Cyrus Durgin : Cleaned for KMOD |
| 64 | * Adam Sulmicki : Bug Fix : Network Device Unload |
| 65 | * A network device unload needs to purge |
| 66 | * the backlog queue. |
| 67 | * Paul Rusty Russell : SIOCSIFNAME |
| 68 | * Pekka Riikonen : Netdev boot-time settings code |
| 69 | * Andrew Morton : Make unregister_netdevice wait |
| 70 | * indefinitely on dev->refcnt |
| 71 | * J Hadi Salim : - Backlog queue sampling |
| 72 | * - netif_rx() feedback |
| 73 | */ |
| 74 | |
| 75 | #include <asm/uaccess.h> |
| 76 | #include <asm/system.h> |
| 77 | #include <linux/bitops.h> |
Randy Dunlap | 4fc268d | 2006-01-11 12:17:47 -0800 | [diff] [blame] | 78 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 79 | #include <linux/config.h> |
| 80 | #include <linux/cpu.h> |
| 81 | #include <linux/types.h> |
| 82 | #include <linux/kernel.h> |
| 83 | #include <linux/sched.h> |
Arjan van de Ven | 4a3e2f7 | 2006-03-20 22:33:17 -0800 | [diff] [blame] | 84 | #include <linux/mutex.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 85 | #include <linux/string.h> |
| 86 | #include <linux/mm.h> |
| 87 | #include <linux/socket.h> |
| 88 | #include <linux/sockios.h> |
| 89 | #include <linux/errno.h> |
| 90 | #include <linux/interrupt.h> |
| 91 | #include <linux/if_ether.h> |
| 92 | #include <linux/netdevice.h> |
| 93 | #include <linux/etherdevice.h> |
| 94 | #include <linux/notifier.h> |
| 95 | #include <linux/skbuff.h> |
| 96 | #include <net/sock.h> |
| 97 | #include <linux/rtnetlink.h> |
| 98 | #include <linux/proc_fs.h> |
| 99 | #include <linux/seq_file.h> |
| 100 | #include <linux/stat.h> |
| 101 | #include <linux/if_bridge.h> |
| 102 | #include <linux/divert.h> |
| 103 | #include <net/dst.h> |
| 104 | #include <net/pkt_sched.h> |
| 105 | #include <net/checksum.h> |
| 106 | #include <linux/highmem.h> |
| 107 | #include <linux/init.h> |
| 108 | #include <linux/kmod.h> |
| 109 | #include <linux/module.h> |
| 110 | #include <linux/kallsyms.h> |
| 111 | #include <linux/netpoll.h> |
| 112 | #include <linux/rcupdate.h> |
| 113 | #include <linux/delay.h> |
Adrian Bunk | d86b5e0 | 2006-01-21 00:46:55 +0100 | [diff] [blame] | 114 | #include <linux/wireless.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | #include <net/iw_handler.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | #include <asm/current.h> |
Steve Grubb | 5bdb988 | 2005-12-03 08:39:35 -0500 | [diff] [blame] | 117 | #include <linux/audit.h> |
Chris Leech | db21733 | 2006-06-17 21:24:58 -0700 | [diff] [blame] | 118 | #include <linux/dmaengine.h> |
Herbert Xu | f6a78bf | 2006-06-22 02:57:17 -0700 | [diff] [blame] | 119 | #include <linux/err.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | /* |
| 122 | * The list of packet types we will receive (as opposed to discard) |
| 123 | * and the routines to invoke. |
| 124 | * |
| 125 | * Why 16. Because with 16 the only overlap we get on a hash of the |
| 126 | * low nibble of the protocol value is RARP/SNAP/X.25. |
| 127 | * |
| 128 | * NOTE: That is no longer true with the addition of VLAN tags. Not |
| 129 | * sure which should go first, but I bet it won't make much |
| 130 | * difference if we are running VLANs. The good news is that |
| 131 | * this protocol won't be in the list unless compiled in, so |
Stephen Hemminger | 3041a06 | 2006-05-26 13:25:24 -0700 | [diff] [blame] | 132 | * the average user (w/out VLANs) will not be adversely affected. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | * --BLG |
| 134 | * |
| 135 | * 0800 IP |
| 136 | * 8100 802.1Q VLAN |
| 137 | * 0001 802.3 |
| 138 | * 0002 AX.25 |
| 139 | * 0004 802.2 |
| 140 | * 8035 RARP |
| 141 | * 0005 SNAP |
| 142 | * 0805 X.25 |
| 143 | * 0806 ARP |
| 144 | * 8137 IPX |
| 145 | * 0009 Localtalk |
| 146 | * 86DD IPv6 |
| 147 | */ |
| 148 | |
| 149 | static DEFINE_SPINLOCK(ptype_lock); |
| 150 | static struct list_head ptype_base[16]; /* 16 way hashed list */ |
| 151 | static struct list_head ptype_all; /* Taps */ |
| 152 | |
Chris Leech | db21733 | 2006-06-17 21:24:58 -0700 | [diff] [blame] | 153 | #ifdef CONFIG_NET_DMA |
| 154 | static struct dma_client *net_dma_client; |
| 155 | static unsigned int net_dma_count; |
| 156 | static spinlock_t net_dma_event_lock; |
| 157 | #endif |
| 158 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | /* |
Stephen Hemminger | 3041a06 | 2006-05-26 13:25:24 -0700 | [diff] [blame] | 160 | * The @dev_base list is protected by @dev_base_lock and the rtnl |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | * semaphore. |
| 162 | * |
| 163 | * Pure readers hold dev_base_lock for reading. |
| 164 | * |
| 165 | * Writers must hold the rtnl semaphore while they loop through the |
| 166 | * dev_base list, and hold dev_base_lock for writing when they do the |
| 167 | * actual updates. This allows pure readers to access the list even |
| 168 | * while a writer is preparing to update it. |
| 169 | * |
| 170 | * To put it another way, dev_base_lock is held for writing only to |
| 171 | * protect against pure readers; the rtnl semaphore provides the |
| 172 | * protection against other writers. |
| 173 | * |
| 174 | * See, for example usages, register_netdevice() and |
| 175 | * unregister_netdevice(), which must be called with the rtnl |
| 176 | * semaphore held. |
| 177 | */ |
| 178 | struct net_device *dev_base; |
| 179 | static struct net_device **dev_tail = &dev_base; |
| 180 | DEFINE_RWLOCK(dev_base_lock); |
| 181 | |
| 182 | EXPORT_SYMBOL(dev_base); |
| 183 | EXPORT_SYMBOL(dev_base_lock); |
| 184 | |
| 185 | #define NETDEV_HASHBITS 8 |
| 186 | static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS]; |
| 187 | static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS]; |
| 188 | |
| 189 | static inline struct hlist_head *dev_name_hash(const char *name) |
| 190 | { |
| 191 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ)); |
| 192 | return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)]; |
| 193 | } |
| 194 | |
| 195 | static inline struct hlist_head *dev_index_hash(int ifindex) |
| 196 | { |
| 197 | return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)]; |
| 198 | } |
| 199 | |
| 200 | /* |
| 201 | * Our notifier list |
| 202 | */ |
| 203 | |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 204 | static RAW_NOTIFIER_HEAD(netdev_chain); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | |
| 206 | /* |
| 207 | * Device drivers call our routines to queue packets here. We empty the |
| 208 | * queue in the local softnet handler. |
| 209 | */ |
Stephen Hemminger | 31aa02c | 2005-06-23 20:12:48 -0700 | [diff] [blame] | 210 | DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | |
| 212 | #ifdef CONFIG_SYSFS |
| 213 | extern int netdev_sysfs_init(void); |
| 214 | extern int netdev_register_sysfs(struct net_device *); |
| 215 | extern void netdev_unregister_sysfs(struct net_device *); |
| 216 | #else |
| 217 | #define netdev_sysfs_init() (0) |
| 218 | #define netdev_register_sysfs(dev) (0) |
| 219 | #define netdev_unregister_sysfs(dev) do { } while(0) |
| 220 | #endif |
| 221 | |
| 222 | |
| 223 | /******************************************************************************* |
| 224 | |
| 225 | Protocol management and registration routines |
| 226 | |
| 227 | *******************************************************************************/ |
| 228 | |
| 229 | /* |
| 230 | * For efficiency |
| 231 | */ |
| 232 | |
| 233 | int netdev_nit; |
| 234 | |
| 235 | /* |
| 236 | * Add a protocol ID to the list. Now that the input handler is |
| 237 | * smarter we can dispense with all the messy stuff that used to be |
| 238 | * here. |
| 239 | * |
| 240 | * BEWARE!!! Protocol handlers, mangling input packets, |
| 241 | * MUST BE last in hash buckets and checking protocol handlers |
| 242 | * MUST start from promiscuous ptype_all chain in net_bh. |
| 243 | * It is true now, do not change it. |
| 244 | * Explanation follows: if protocol handler, mangling packet, will |
| 245 | * be the first on list, it is not able to sense, that packet |
| 246 | * is cloned and should be copied-on-write, so that it will |
| 247 | * change it and subsequent readers will get broken packet. |
| 248 | * --ANK (980803) |
| 249 | */ |
| 250 | |
| 251 | /** |
| 252 | * dev_add_pack - add packet handler |
| 253 | * @pt: packet type declaration |
| 254 | * |
| 255 | * Add a protocol handler to the networking stack. The passed &packet_type |
| 256 | * is linked into kernel lists and may not be freed until it has been |
| 257 | * removed from the kernel lists. |
| 258 | * |
| 259 | * This call does not sleep therefore it can not |
| 260 | * guarantee all CPU's that are in middle of receiving packets |
| 261 | * will see the new packet type (until the next received packet). |
| 262 | */ |
| 263 | |
| 264 | void dev_add_pack(struct packet_type *pt) |
| 265 | { |
| 266 | int hash; |
| 267 | |
| 268 | spin_lock_bh(&ptype_lock); |
| 269 | if (pt->type == htons(ETH_P_ALL)) { |
| 270 | netdev_nit++; |
| 271 | list_add_rcu(&pt->list, &ptype_all); |
| 272 | } else { |
| 273 | hash = ntohs(pt->type) & 15; |
| 274 | list_add_rcu(&pt->list, &ptype_base[hash]); |
| 275 | } |
| 276 | spin_unlock_bh(&ptype_lock); |
| 277 | } |
| 278 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 279 | /** |
| 280 | * __dev_remove_pack - remove packet handler |
| 281 | * @pt: packet type declaration |
| 282 | * |
| 283 | * Remove a protocol handler that was previously added to the kernel |
| 284 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed |
| 285 | * from the kernel lists and can be freed or reused once this function |
| 286 | * returns. |
| 287 | * |
| 288 | * The packet type might still be in use by receivers |
| 289 | * and must not be freed until after all the CPU's have gone |
| 290 | * through a quiescent state. |
| 291 | */ |
| 292 | void __dev_remove_pack(struct packet_type *pt) |
| 293 | { |
| 294 | struct list_head *head; |
| 295 | struct packet_type *pt1; |
| 296 | |
| 297 | spin_lock_bh(&ptype_lock); |
| 298 | |
| 299 | if (pt->type == htons(ETH_P_ALL)) { |
| 300 | netdev_nit--; |
| 301 | head = &ptype_all; |
| 302 | } else |
| 303 | head = &ptype_base[ntohs(pt->type) & 15]; |
| 304 | |
| 305 | list_for_each_entry(pt1, head, list) { |
| 306 | if (pt == pt1) { |
| 307 | list_del_rcu(&pt->list); |
| 308 | goto out; |
| 309 | } |
| 310 | } |
| 311 | |
| 312 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt); |
| 313 | out: |
| 314 | spin_unlock_bh(&ptype_lock); |
| 315 | } |
| 316 | /** |
| 317 | * dev_remove_pack - remove packet handler |
| 318 | * @pt: packet type declaration |
| 319 | * |
| 320 | * Remove a protocol handler that was previously added to the kernel |
| 321 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed |
| 322 | * from the kernel lists and can be freed or reused once this function |
| 323 | * returns. |
| 324 | * |
| 325 | * This call sleeps to guarantee that no CPU is looking at the packet |
| 326 | * type after return. |
| 327 | */ |
| 328 | void dev_remove_pack(struct packet_type *pt) |
| 329 | { |
| 330 | __dev_remove_pack(pt); |
| 331 | |
| 332 | synchronize_net(); |
| 333 | } |
| 334 | |
| 335 | /****************************************************************************** |
| 336 | |
| 337 | Device Boot-time Settings Routines |
| 338 | |
| 339 | *******************************************************************************/ |
| 340 | |
| 341 | /* Boot time configuration table */ |
| 342 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; |
| 343 | |
| 344 | /** |
| 345 | * netdev_boot_setup_add - add new setup entry |
| 346 | * @name: name of the device |
| 347 | * @map: configured settings for the device |
| 348 | * |
| 349 | * Adds new setup entry to the dev_boot_setup list. The function |
| 350 | * returns 0 on error and 1 on success. This is a generic routine to |
| 351 | * all netdevices. |
| 352 | */ |
| 353 | static int netdev_boot_setup_add(char *name, struct ifmap *map) |
| 354 | { |
| 355 | struct netdev_boot_setup *s; |
| 356 | int i; |
| 357 | |
| 358 | s = dev_boot_setup; |
| 359 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { |
| 360 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { |
| 361 | memset(s[i].name, 0, sizeof(s[i].name)); |
| 362 | strcpy(s[i].name, name); |
| 363 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
| 364 | break; |
| 365 | } |
| 366 | } |
| 367 | |
| 368 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; |
| 369 | } |
| 370 | |
| 371 | /** |
| 372 | * netdev_boot_setup_check - check boot time settings |
| 373 | * @dev: the netdevice |
| 374 | * |
| 375 | * Check boot time settings for the device. |
| 376 | * The found settings are set for the device to be used |
| 377 | * later in the device probing. |
| 378 | * Returns 0 if no settings found, 1 if they are. |
| 379 | */ |
| 380 | int netdev_boot_setup_check(struct net_device *dev) |
| 381 | { |
| 382 | struct netdev_boot_setup *s = dev_boot_setup; |
| 383 | int i; |
| 384 | |
| 385 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { |
| 386 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && |
| 387 | !strncmp(dev->name, s[i].name, strlen(s[i].name))) { |
| 388 | dev->irq = s[i].map.irq; |
| 389 | dev->base_addr = s[i].map.base_addr; |
| 390 | dev->mem_start = s[i].map.mem_start; |
| 391 | dev->mem_end = s[i].map.mem_end; |
| 392 | return 1; |
| 393 | } |
| 394 | } |
| 395 | return 0; |
| 396 | } |
| 397 | |
| 398 | |
| 399 | /** |
| 400 | * netdev_boot_base - get address from boot time settings |
| 401 | * @prefix: prefix for network device |
| 402 | * @unit: id for network device |
| 403 | * |
| 404 | * Check boot time settings for the base address of device. |
| 405 | * The found settings are set for the device to be used |
| 406 | * later in the device probing. |
| 407 | * Returns 0 if no settings found. |
| 408 | */ |
| 409 | unsigned long netdev_boot_base(const char *prefix, int unit) |
| 410 | { |
| 411 | const struct netdev_boot_setup *s = dev_boot_setup; |
| 412 | char name[IFNAMSIZ]; |
| 413 | int i; |
| 414 | |
| 415 | sprintf(name, "%s%d", prefix, unit); |
| 416 | |
| 417 | /* |
| 418 | * If device already registered then return base of 1 |
| 419 | * to indicate not to probe for this interface |
| 420 | */ |
| 421 | if (__dev_get_by_name(name)) |
| 422 | return 1; |
| 423 | |
| 424 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) |
| 425 | if (!strcmp(name, s[i].name)) |
| 426 | return s[i].map.base_addr; |
| 427 | return 0; |
| 428 | } |
| 429 | |
| 430 | /* |
| 431 | * Saves at boot time configured settings for any netdevice. |
| 432 | */ |
| 433 | int __init netdev_boot_setup(char *str) |
| 434 | { |
| 435 | int ints[5]; |
| 436 | struct ifmap map; |
| 437 | |
| 438 | str = get_options(str, ARRAY_SIZE(ints), ints); |
| 439 | if (!str || !*str) |
| 440 | return 0; |
| 441 | |
| 442 | /* Save settings */ |
| 443 | memset(&map, 0, sizeof(map)); |
| 444 | if (ints[0] > 0) |
| 445 | map.irq = ints[1]; |
| 446 | if (ints[0] > 1) |
| 447 | map.base_addr = ints[2]; |
| 448 | if (ints[0] > 2) |
| 449 | map.mem_start = ints[3]; |
| 450 | if (ints[0] > 3) |
| 451 | map.mem_end = ints[4]; |
| 452 | |
| 453 | /* Add new entry to the list */ |
| 454 | return netdev_boot_setup_add(str, &map); |
| 455 | } |
| 456 | |
| 457 | __setup("netdev=", netdev_boot_setup); |
| 458 | |
| 459 | /******************************************************************************* |
| 460 | |
| 461 | Device Interface Subroutines |
| 462 | |
| 463 | *******************************************************************************/ |
| 464 | |
| 465 | /** |
| 466 | * __dev_get_by_name - find a device by its name |
| 467 | * @name: name to find |
| 468 | * |
| 469 | * Find an interface by name. Must be called under RTNL semaphore |
| 470 | * or @dev_base_lock. If the name is found a pointer to the device |
| 471 | * is returned. If the name is not found then %NULL is returned. The |
| 472 | * reference counters are not incremented so the caller must be |
| 473 | * careful with locks. |
| 474 | */ |
| 475 | |
| 476 | struct net_device *__dev_get_by_name(const char *name) |
| 477 | { |
| 478 | struct hlist_node *p; |
| 479 | |
| 480 | hlist_for_each(p, dev_name_hash(name)) { |
| 481 | struct net_device *dev |
| 482 | = hlist_entry(p, struct net_device, name_hlist); |
| 483 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
| 484 | return dev; |
| 485 | } |
| 486 | return NULL; |
| 487 | } |
| 488 | |
| 489 | /** |
| 490 | * dev_get_by_name - find a device by its name |
| 491 | * @name: name to find |
| 492 | * |
| 493 | * Find an interface by name. This can be called from any |
| 494 | * context and does its own locking. The returned handle has |
| 495 | * the usage count incremented and the caller must use dev_put() to |
| 496 | * release it when it is no longer needed. %NULL is returned if no |
| 497 | * matching device is found. |
| 498 | */ |
| 499 | |
| 500 | struct net_device *dev_get_by_name(const char *name) |
| 501 | { |
| 502 | struct net_device *dev; |
| 503 | |
| 504 | read_lock(&dev_base_lock); |
| 505 | dev = __dev_get_by_name(name); |
| 506 | if (dev) |
| 507 | dev_hold(dev); |
| 508 | read_unlock(&dev_base_lock); |
| 509 | return dev; |
| 510 | } |
| 511 | |
| 512 | /** |
| 513 | * __dev_get_by_index - find a device by its ifindex |
| 514 | * @ifindex: index of device |
| 515 | * |
| 516 | * Search for an interface by index. Returns %NULL if the device |
| 517 | * is not found or a pointer to the device. The device has not |
| 518 | * had its reference counter increased so the caller must be careful |
| 519 | * about locking. The caller must hold either the RTNL semaphore |
| 520 | * or @dev_base_lock. |
| 521 | */ |
| 522 | |
| 523 | struct net_device *__dev_get_by_index(int ifindex) |
| 524 | { |
| 525 | struct hlist_node *p; |
| 526 | |
| 527 | hlist_for_each(p, dev_index_hash(ifindex)) { |
| 528 | struct net_device *dev |
| 529 | = hlist_entry(p, struct net_device, index_hlist); |
| 530 | if (dev->ifindex == ifindex) |
| 531 | return dev; |
| 532 | } |
| 533 | return NULL; |
| 534 | } |
| 535 | |
| 536 | |
| 537 | /** |
| 538 | * dev_get_by_index - find a device by its ifindex |
| 539 | * @ifindex: index of device |
| 540 | * |
| 541 | * Search for an interface by index. Returns NULL if the device |
| 542 | * is not found or a pointer to the device. The device returned has |
| 543 | * had a reference added and the pointer is safe until the user calls |
| 544 | * dev_put to indicate they have finished with it. |
| 545 | */ |
| 546 | |
| 547 | struct net_device *dev_get_by_index(int ifindex) |
| 548 | { |
| 549 | struct net_device *dev; |
| 550 | |
| 551 | read_lock(&dev_base_lock); |
| 552 | dev = __dev_get_by_index(ifindex); |
| 553 | if (dev) |
| 554 | dev_hold(dev); |
| 555 | read_unlock(&dev_base_lock); |
| 556 | return dev; |
| 557 | } |
| 558 | |
| 559 | /** |
| 560 | * dev_getbyhwaddr - find a device by its hardware address |
| 561 | * @type: media type of device |
| 562 | * @ha: hardware address |
| 563 | * |
| 564 | * Search for an interface by MAC address. Returns NULL if the device |
| 565 | * is not found or a pointer to the device. The caller must hold the |
| 566 | * rtnl semaphore. The returned device has not had its ref count increased |
| 567 | * and the caller must therefore be careful about locking |
| 568 | * |
| 569 | * BUGS: |
| 570 | * If the API was consistent this would be __dev_get_by_hwaddr |
| 571 | */ |
| 572 | |
| 573 | struct net_device *dev_getbyhwaddr(unsigned short type, char *ha) |
| 574 | { |
| 575 | struct net_device *dev; |
| 576 | |
| 577 | ASSERT_RTNL(); |
| 578 | |
| 579 | for (dev = dev_base; dev; dev = dev->next) |
| 580 | if (dev->type == type && |
| 581 | !memcmp(dev->dev_addr, ha, dev->addr_len)) |
| 582 | break; |
| 583 | return dev; |
| 584 | } |
| 585 | |
Jochen Friedrich | cf309e3 | 2005-09-22 04:44:55 -0300 | [diff] [blame] | 586 | EXPORT_SYMBOL(dev_getbyhwaddr); |
| 587 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 588 | struct net_device *dev_getfirstbyhwtype(unsigned short type) |
| 589 | { |
| 590 | struct net_device *dev; |
| 591 | |
| 592 | rtnl_lock(); |
| 593 | for (dev = dev_base; dev; dev = dev->next) { |
| 594 | if (dev->type == type) { |
| 595 | dev_hold(dev); |
| 596 | break; |
| 597 | } |
| 598 | } |
| 599 | rtnl_unlock(); |
| 600 | return dev; |
| 601 | } |
| 602 | |
| 603 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
| 604 | |
| 605 | /** |
| 606 | * dev_get_by_flags - find any device with given flags |
| 607 | * @if_flags: IFF_* values |
| 608 | * @mask: bitmask of bits in if_flags to check |
| 609 | * |
| 610 | * Search for any interface with the given flags. Returns NULL if a device |
| 611 | * is not found or a pointer to the device. The device returned has |
| 612 | * had a reference added and the pointer is safe until the user calls |
| 613 | * dev_put to indicate they have finished with it. |
| 614 | */ |
| 615 | |
| 616 | struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask) |
| 617 | { |
| 618 | struct net_device *dev; |
| 619 | |
| 620 | read_lock(&dev_base_lock); |
| 621 | for (dev = dev_base; dev != NULL; dev = dev->next) { |
| 622 | if (((dev->flags ^ if_flags) & mask) == 0) { |
| 623 | dev_hold(dev); |
| 624 | break; |
| 625 | } |
| 626 | } |
| 627 | read_unlock(&dev_base_lock); |
| 628 | return dev; |
| 629 | } |
| 630 | |
| 631 | /** |
| 632 | * dev_valid_name - check if name is okay for network device |
| 633 | * @name: name string |
| 634 | * |
| 635 | * Network device names need to be valid file names to |
| 636 | * to allow sysfs to work |
| 637 | */ |
Mitch Williams | c2373ee | 2005-11-09 10:34:45 -0800 | [diff] [blame] | 638 | int dev_valid_name(const char *name) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | { |
| 640 | return !(*name == '\0' |
| 641 | || !strcmp(name, ".") |
| 642 | || !strcmp(name, "..") |
| 643 | || strchr(name, '/')); |
| 644 | } |
| 645 | |
| 646 | /** |
| 647 | * dev_alloc_name - allocate a name for a device |
| 648 | * @dev: device |
| 649 | * @name: name format string |
| 650 | * |
| 651 | * Passed a format string - eg "lt%d" it will try and find a suitable |
Stephen Hemminger | 3041a06 | 2006-05-26 13:25:24 -0700 | [diff] [blame] | 652 | * id. It scans list of devices to build up a free map, then chooses |
| 653 | * the first empty slot. The caller must hold the dev_base or rtnl lock |
| 654 | * while allocating the name and adding the device in order to avoid |
| 655 | * duplicates. |
| 656 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). |
| 657 | * Returns the number of the unit assigned or a negative errno code. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | */ |
| 659 | |
| 660 | int dev_alloc_name(struct net_device *dev, const char *name) |
| 661 | { |
| 662 | int i = 0; |
| 663 | char buf[IFNAMSIZ]; |
| 664 | const char *p; |
| 665 | const int max_netdevices = 8*PAGE_SIZE; |
| 666 | long *inuse; |
| 667 | struct net_device *d; |
| 668 | |
| 669 | p = strnchr(name, IFNAMSIZ-1, '%'); |
| 670 | if (p) { |
| 671 | /* |
| 672 | * Verify the string as this thing may have come from |
| 673 | * the user. There must be either one "%d" and no other "%" |
| 674 | * characters. |
| 675 | */ |
| 676 | if (p[1] != 'd' || strchr(p + 2, '%')) |
| 677 | return -EINVAL; |
| 678 | |
| 679 | /* Use one page as a bit array of possible slots */ |
| 680 | inuse = (long *) get_zeroed_page(GFP_ATOMIC); |
| 681 | if (!inuse) |
| 682 | return -ENOMEM; |
| 683 | |
| 684 | for (d = dev_base; d; d = d->next) { |
| 685 | if (!sscanf(d->name, name, &i)) |
| 686 | continue; |
| 687 | if (i < 0 || i >= max_netdevices) |
| 688 | continue; |
| 689 | |
| 690 | /* avoid cases where sscanf is not exact inverse of printf */ |
| 691 | snprintf(buf, sizeof(buf), name, i); |
| 692 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
| 693 | set_bit(i, inuse); |
| 694 | } |
| 695 | |
| 696 | i = find_first_zero_bit(inuse, max_netdevices); |
| 697 | free_page((unsigned long) inuse); |
| 698 | } |
| 699 | |
| 700 | snprintf(buf, sizeof(buf), name, i); |
| 701 | if (!__dev_get_by_name(buf)) { |
| 702 | strlcpy(dev->name, buf, IFNAMSIZ); |
| 703 | return i; |
| 704 | } |
| 705 | |
| 706 | /* It is possible to run out of possible slots |
| 707 | * when the name is long and there isn't enough space left |
| 708 | * for the digits, or if all bits are used. |
| 709 | */ |
| 710 | return -ENFILE; |
| 711 | } |
| 712 | |
| 713 | |
| 714 | /** |
| 715 | * dev_change_name - change name of a device |
| 716 | * @dev: device |
| 717 | * @newname: name (or format string) must be at least IFNAMSIZ |
| 718 | * |
| 719 | * Change name of a device, can pass format strings "eth%d". |
| 720 | * for wildcarding. |
| 721 | */ |
| 722 | int dev_change_name(struct net_device *dev, char *newname) |
| 723 | { |
| 724 | int err = 0; |
| 725 | |
| 726 | ASSERT_RTNL(); |
| 727 | |
| 728 | if (dev->flags & IFF_UP) |
| 729 | return -EBUSY; |
| 730 | |
| 731 | if (!dev_valid_name(newname)) |
| 732 | return -EINVAL; |
| 733 | |
| 734 | if (strchr(newname, '%')) { |
| 735 | err = dev_alloc_name(dev, newname); |
| 736 | if (err < 0) |
| 737 | return err; |
| 738 | strcpy(newname, dev->name); |
| 739 | } |
| 740 | else if (__dev_get_by_name(newname)) |
| 741 | return -EEXIST; |
| 742 | else |
| 743 | strlcpy(dev->name, newname, IFNAMSIZ); |
| 744 | |
| 745 | err = class_device_rename(&dev->class_dev, dev->name); |
| 746 | if (!err) { |
| 747 | hlist_del(&dev->name_hlist); |
| 748 | hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name)); |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 749 | raw_notifier_call_chain(&netdev_chain, |
Alan Stern | e041c68 | 2006-03-27 01:16:30 -0800 | [diff] [blame] | 750 | NETDEV_CHANGENAME, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 751 | } |
| 752 | |
| 753 | return err; |
| 754 | } |
| 755 | |
| 756 | /** |
Stephen Hemminger | 3041a06 | 2006-05-26 13:25:24 -0700 | [diff] [blame] | 757 | * netdev_features_change - device changes features |
Stephen Hemminger | d8a33ac | 2005-05-29 14:13:47 -0700 | [diff] [blame] | 758 | * @dev: device to cause notification |
| 759 | * |
| 760 | * Called to indicate a device has changed features. |
| 761 | */ |
| 762 | void netdev_features_change(struct net_device *dev) |
| 763 | { |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 764 | raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev); |
Stephen Hemminger | d8a33ac | 2005-05-29 14:13:47 -0700 | [diff] [blame] | 765 | } |
| 766 | EXPORT_SYMBOL(netdev_features_change); |
| 767 | |
| 768 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 769 | * netdev_state_change - device changes state |
| 770 | * @dev: device to cause notification |
| 771 | * |
| 772 | * Called to indicate a device has changed state. This function calls |
| 773 | * the notifier chains for netdev_chain and sends a NEWLINK message |
| 774 | * to the routing socket. |
| 775 | */ |
| 776 | void netdev_state_change(struct net_device *dev) |
| 777 | { |
| 778 | if (dev->flags & IFF_UP) { |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 779 | raw_notifier_call_chain(&netdev_chain, |
Alan Stern | e041c68 | 2006-03-27 01:16:30 -0800 | [diff] [blame] | 780 | NETDEV_CHANGE, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 781 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0); |
| 782 | } |
| 783 | } |
| 784 | |
| 785 | /** |
| 786 | * dev_load - load a network module |
| 787 | * @name: name of interface |
| 788 | * |
| 789 | * If a network interface is not present and the process has suitable |
| 790 | * privileges this function loads the module. If module loading is not |
| 791 | * available in this kernel then it becomes a nop. |
| 792 | */ |
| 793 | |
| 794 | void dev_load(const char *name) |
| 795 | { |
| 796 | struct net_device *dev; |
| 797 | |
| 798 | read_lock(&dev_base_lock); |
| 799 | dev = __dev_get_by_name(name); |
| 800 | read_unlock(&dev_base_lock); |
| 801 | |
| 802 | if (!dev && capable(CAP_SYS_MODULE)) |
| 803 | request_module("%s", name); |
| 804 | } |
| 805 | |
| 806 | static int default_rebuild_header(struct sk_buff *skb) |
| 807 | { |
| 808 | printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n", |
| 809 | skb->dev ? skb->dev->name : "NULL!!!"); |
| 810 | kfree_skb(skb); |
| 811 | return 1; |
| 812 | } |
| 813 | |
| 814 | |
| 815 | /** |
| 816 | * dev_open - prepare an interface for use. |
| 817 | * @dev: device to open |
| 818 | * |
| 819 | * Takes a device from down to up state. The device's private open |
| 820 | * function is invoked and then the multicast lists are loaded. Finally |
| 821 | * the device is moved into the up state and a %NETDEV_UP message is |
| 822 | * sent to the netdev notifier chain. |
| 823 | * |
| 824 | * Calling this function on an active interface is a nop. On a failure |
| 825 | * a negative errno code is returned. |
| 826 | */ |
| 827 | int dev_open(struct net_device *dev) |
| 828 | { |
| 829 | int ret = 0; |
| 830 | |
| 831 | /* |
| 832 | * Is it already up? |
| 833 | */ |
| 834 | |
| 835 | if (dev->flags & IFF_UP) |
| 836 | return 0; |
| 837 | |
| 838 | /* |
| 839 | * Is it even present? |
| 840 | */ |
| 841 | if (!netif_device_present(dev)) |
| 842 | return -ENODEV; |
| 843 | |
| 844 | /* |
| 845 | * Call device private open method |
| 846 | */ |
| 847 | set_bit(__LINK_STATE_START, &dev->state); |
| 848 | if (dev->open) { |
| 849 | ret = dev->open(dev); |
| 850 | if (ret) |
| 851 | clear_bit(__LINK_STATE_START, &dev->state); |
| 852 | } |
| 853 | |
| 854 | /* |
| 855 | * If it went open OK then: |
| 856 | */ |
| 857 | |
| 858 | if (!ret) { |
| 859 | /* |
| 860 | * Set the flags. |
| 861 | */ |
| 862 | dev->flags |= IFF_UP; |
| 863 | |
| 864 | /* |
| 865 | * Initialize multicasting status |
| 866 | */ |
| 867 | dev_mc_upload(dev); |
| 868 | |
| 869 | /* |
| 870 | * Wakeup transmit queue engine |
| 871 | */ |
| 872 | dev_activate(dev); |
| 873 | |
| 874 | /* |
| 875 | * ... and announce new interface. |
| 876 | */ |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 877 | raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 878 | } |
| 879 | return ret; |
| 880 | } |
| 881 | |
| 882 | /** |
| 883 | * dev_close - shutdown an interface. |
| 884 | * @dev: device to shutdown |
| 885 | * |
| 886 | * This function moves an active device into down state. A |
| 887 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device |
| 888 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier |
| 889 | * chain. |
| 890 | */ |
| 891 | int dev_close(struct net_device *dev) |
| 892 | { |
| 893 | if (!(dev->flags & IFF_UP)) |
| 894 | return 0; |
| 895 | |
| 896 | /* |
| 897 | * Tell people we are going down, so that they can |
| 898 | * prepare to death, when device is still operating. |
| 899 | */ |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 900 | raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 901 | |
| 902 | dev_deactivate(dev); |
| 903 | |
| 904 | clear_bit(__LINK_STATE_START, &dev->state); |
| 905 | |
| 906 | /* Synchronize to scheduled poll. We cannot touch poll list, |
| 907 | * it can be even on different cpu. So just clear netif_running(), |
| 908 | * and wait when poll really will happen. Actually, the best place |
| 909 | * for this is inside dev->stop() after device stopped its irq |
| 910 | * engine, but this requires more changes in devices. */ |
| 911 | |
| 912 | smp_mb__after_clear_bit(); /* Commit netif_running(). */ |
| 913 | while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) { |
| 914 | /* No hurry. */ |
David S. Miller | 6192b54 | 2005-07-28 12:12:58 -0700 | [diff] [blame] | 915 | msleep(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | } |
| 917 | |
| 918 | /* |
| 919 | * Call the device specific close. This cannot fail. |
| 920 | * Only if device is UP |
| 921 | * |
| 922 | * We allow it to be called even after a DETACH hot-plug |
| 923 | * event. |
| 924 | */ |
| 925 | if (dev->stop) |
| 926 | dev->stop(dev); |
| 927 | |
| 928 | /* |
| 929 | * Device is now down. |
| 930 | */ |
| 931 | |
| 932 | dev->flags &= ~IFF_UP; |
| 933 | |
| 934 | /* |
| 935 | * Tell people we are down |
| 936 | */ |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 937 | raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 938 | |
| 939 | return 0; |
| 940 | } |
| 941 | |
| 942 | |
| 943 | /* |
| 944 | * Device change register/unregister. These are not inline or static |
| 945 | * as we export them to the world. |
| 946 | */ |
| 947 | |
| 948 | /** |
| 949 | * register_netdevice_notifier - register a network notifier block |
| 950 | * @nb: notifier |
| 951 | * |
| 952 | * Register a notifier to be called when network device events occur. |
| 953 | * The notifier passed is linked into the kernel structures and must |
| 954 | * not be reused until it has been unregistered. A negative errno code |
| 955 | * is returned on a failure. |
| 956 | * |
| 957 | * When registered all registration and up events are replayed |
| 958 | * to the new notifier to allow device to have a race free |
| 959 | * view of the network device list. |
| 960 | */ |
| 961 | |
| 962 | int register_netdevice_notifier(struct notifier_block *nb) |
| 963 | { |
| 964 | struct net_device *dev; |
| 965 | int err; |
| 966 | |
| 967 | rtnl_lock(); |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 968 | err = raw_notifier_chain_register(&netdev_chain, nb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | if (!err) { |
| 970 | for (dev = dev_base; dev; dev = dev->next) { |
| 971 | nb->notifier_call(nb, NETDEV_REGISTER, dev); |
| 972 | |
| 973 | if (dev->flags & IFF_UP) |
| 974 | nb->notifier_call(nb, NETDEV_UP, dev); |
| 975 | } |
| 976 | } |
| 977 | rtnl_unlock(); |
| 978 | return err; |
| 979 | } |
| 980 | |
| 981 | /** |
| 982 | * unregister_netdevice_notifier - unregister a network notifier block |
| 983 | * @nb: notifier |
| 984 | * |
| 985 | * Unregister a notifier previously registered by |
| 986 | * register_netdevice_notifier(). The notifier is unlinked into the |
| 987 | * kernel structures and may then be reused. A negative errno code |
| 988 | * is returned on a failure. |
| 989 | */ |
| 990 | |
| 991 | int unregister_netdevice_notifier(struct notifier_block *nb) |
| 992 | { |
Herbert Xu | 9f51495 | 2006-03-25 01:24:25 -0800 | [diff] [blame] | 993 | int err; |
| 994 | |
| 995 | rtnl_lock(); |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 996 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
Herbert Xu | 9f51495 | 2006-03-25 01:24:25 -0800 | [diff] [blame] | 997 | rtnl_unlock(); |
| 998 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 999 | } |
| 1000 | |
| 1001 | /** |
| 1002 | * call_netdevice_notifiers - call all network notifier blocks |
| 1003 | * @val: value passed unmodified to notifier function |
| 1004 | * @v: pointer passed unmodified to notifier function |
| 1005 | * |
| 1006 | * Call all network notifier blocks. Parameters and return value |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 1007 | * are as for raw_notifier_call_chain(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1008 | */ |
| 1009 | |
| 1010 | int call_netdevice_notifiers(unsigned long val, void *v) |
| 1011 | { |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 1012 | return raw_notifier_call_chain(&netdev_chain, val, v); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1013 | } |
| 1014 | |
| 1015 | /* When > 0 there are consumers of rx skb time stamps */ |
| 1016 | static atomic_t netstamp_needed = ATOMIC_INIT(0); |
| 1017 | |
| 1018 | void net_enable_timestamp(void) |
| 1019 | { |
| 1020 | atomic_inc(&netstamp_needed); |
| 1021 | } |
| 1022 | |
| 1023 | void net_disable_timestamp(void) |
| 1024 | { |
| 1025 | atomic_dec(&netstamp_needed); |
| 1026 | } |
| 1027 | |
Patrick McHardy | a61bbcf | 2005-08-14 17:24:31 -0700 | [diff] [blame] | 1028 | void __net_timestamp(struct sk_buff *skb) |
| 1029 | { |
| 1030 | struct timeval tv; |
| 1031 | |
| 1032 | do_gettimeofday(&tv); |
| 1033 | skb_set_timestamp(skb, &tv); |
| 1034 | } |
| 1035 | EXPORT_SYMBOL(__net_timestamp); |
| 1036 | |
| 1037 | static inline void net_timestamp(struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1038 | { |
| 1039 | if (atomic_read(&netstamp_needed)) |
Patrick McHardy | a61bbcf | 2005-08-14 17:24:31 -0700 | [diff] [blame] | 1040 | __net_timestamp(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1041 | else { |
Patrick McHardy | a61bbcf | 2005-08-14 17:24:31 -0700 | [diff] [blame] | 1042 | skb->tstamp.off_sec = 0; |
| 1043 | skb->tstamp.off_usec = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1044 | } |
| 1045 | } |
| 1046 | |
| 1047 | /* |
| 1048 | * Support routine. Sends outgoing frames to any network |
| 1049 | * taps currently in use. |
| 1050 | */ |
| 1051 | |
Herbert Xu | f6a78bf | 2006-06-22 02:57:17 -0700 | [diff] [blame] | 1052 | static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1053 | { |
| 1054 | struct packet_type *ptype; |
Patrick McHardy | a61bbcf | 2005-08-14 17:24:31 -0700 | [diff] [blame] | 1055 | |
| 1056 | net_timestamp(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | |
| 1058 | rcu_read_lock(); |
| 1059 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
| 1060 | /* Never send packets back to the socket |
| 1061 | * they originated from - MvS (miquels@drinkel.ow.org) |
| 1062 | */ |
| 1063 | if ((ptype->dev == dev || !ptype->dev) && |
| 1064 | (ptype->af_packet_priv == NULL || |
| 1065 | (struct sock *)ptype->af_packet_priv != skb->sk)) { |
| 1066 | struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC); |
| 1067 | if (!skb2) |
| 1068 | break; |
| 1069 | |
| 1070 | /* skb->nh should be correctly |
| 1071 | set by sender, so that the second statement is |
| 1072 | just protection against buggy protocols. |
| 1073 | */ |
| 1074 | skb2->mac.raw = skb2->data; |
| 1075 | |
| 1076 | if (skb2->nh.raw < skb2->data || |
| 1077 | skb2->nh.raw > skb2->tail) { |
| 1078 | if (net_ratelimit()) |
| 1079 | printk(KERN_CRIT "protocol %04x is " |
| 1080 | "buggy, dev %s\n", |
| 1081 | skb2->protocol, dev->name); |
| 1082 | skb2->nh.raw = skb2->data; |
| 1083 | } |
| 1084 | |
| 1085 | skb2->h.raw = skb2->nh.raw; |
| 1086 | skb2->pkt_type = PACKET_OUTGOING; |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1087 | ptype->func(skb2, skb->dev, ptype, skb->dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1088 | } |
| 1089 | } |
| 1090 | rcu_read_unlock(); |
| 1091 | } |
| 1092 | |
Denis Vlasenko | 5607943 | 2006-03-29 15:57:29 -0800 | [diff] [blame] | 1093 | |
| 1094 | void __netif_schedule(struct net_device *dev) |
| 1095 | { |
| 1096 | if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) { |
| 1097 | unsigned long flags; |
| 1098 | struct softnet_data *sd; |
| 1099 | |
| 1100 | local_irq_save(flags); |
| 1101 | sd = &__get_cpu_var(softnet_data); |
| 1102 | dev->next_sched = sd->output_queue; |
| 1103 | sd->output_queue = dev; |
| 1104 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
| 1105 | local_irq_restore(flags); |
| 1106 | } |
| 1107 | } |
| 1108 | EXPORT_SYMBOL(__netif_schedule); |
| 1109 | |
| 1110 | void __netif_rx_schedule(struct net_device *dev) |
| 1111 | { |
| 1112 | unsigned long flags; |
| 1113 | |
| 1114 | local_irq_save(flags); |
| 1115 | dev_hold(dev); |
| 1116 | list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list); |
| 1117 | if (dev->quota < 0) |
| 1118 | dev->quota += dev->weight; |
| 1119 | else |
| 1120 | dev->quota = dev->weight; |
| 1121 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
| 1122 | local_irq_restore(flags); |
| 1123 | } |
| 1124 | EXPORT_SYMBOL(__netif_rx_schedule); |
| 1125 | |
| 1126 | void dev_kfree_skb_any(struct sk_buff *skb) |
| 1127 | { |
| 1128 | if (in_irq() || irqs_disabled()) |
| 1129 | dev_kfree_skb_irq(skb); |
| 1130 | else |
| 1131 | dev_kfree_skb(skb); |
| 1132 | } |
| 1133 | EXPORT_SYMBOL(dev_kfree_skb_any); |
| 1134 | |
| 1135 | |
| 1136 | /* Hot-plugging. */ |
| 1137 | void netif_device_detach(struct net_device *dev) |
| 1138 | { |
| 1139 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && |
| 1140 | netif_running(dev)) { |
| 1141 | netif_stop_queue(dev); |
| 1142 | } |
| 1143 | } |
| 1144 | EXPORT_SYMBOL(netif_device_detach); |
| 1145 | |
| 1146 | void netif_device_attach(struct net_device *dev) |
| 1147 | { |
| 1148 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && |
| 1149 | netif_running(dev)) { |
| 1150 | netif_wake_queue(dev); |
| 1151 | __netdev_watchdog_up(dev); |
| 1152 | } |
| 1153 | } |
| 1154 | EXPORT_SYMBOL(netif_device_attach); |
| 1155 | |
| 1156 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1157 | /* |
| 1158 | * Invalidate hardware checksum when packet is to be mangled, and |
| 1159 | * complete checksum manually on outgoing path. |
| 1160 | */ |
| 1161 | int skb_checksum_help(struct sk_buff *skb, int inward) |
| 1162 | { |
| 1163 | unsigned int csum; |
| 1164 | int ret = 0, offset = skb->h.raw - skb->data; |
| 1165 | |
| 1166 | if (inward) { |
| 1167 | skb->ip_summed = CHECKSUM_NONE; |
| 1168 | goto out; |
| 1169 | } |
| 1170 | |
| 1171 | if (skb_cloned(skb)) { |
| 1172 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
| 1173 | if (ret) |
| 1174 | goto out; |
| 1175 | } |
| 1176 | |
Kris Katterjohn | 09a6266 | 2006-01-08 22:24:28 -0800 | [diff] [blame] | 1177 | BUG_ON(offset > (int)skb->len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | csum = skb_checksum(skb, offset, skb->len-offset, 0); |
| 1179 | |
| 1180 | offset = skb->tail - skb->h.raw; |
Kris Katterjohn | 09a6266 | 2006-01-08 22:24:28 -0800 | [diff] [blame] | 1181 | BUG_ON(offset <= 0); |
| 1182 | BUG_ON(skb->csum + 2 > offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1183 | |
| 1184 | *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum); |
| 1185 | skb->ip_summed = CHECKSUM_NONE; |
| 1186 | out: |
| 1187 | return ret; |
| 1188 | } |
| 1189 | |
Herbert Xu | f6a78bf | 2006-06-22 02:57:17 -0700 | [diff] [blame] | 1190 | /** |
| 1191 | * skb_gso_segment - Perform segmentation on skb. |
| 1192 | * @skb: buffer to segment |
| 1193 | * @sg: whether scatter-gather is supported on the target. |
| 1194 | * |
| 1195 | * This function segments the given skb and returns a list of segments. |
| 1196 | */ |
| 1197 | struct sk_buff *skb_gso_segment(struct sk_buff *skb, int sg) |
| 1198 | { |
| 1199 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); |
| 1200 | struct packet_type *ptype; |
| 1201 | int type = skb->protocol; |
| 1202 | |
| 1203 | BUG_ON(skb_shinfo(skb)->frag_list); |
| 1204 | BUG_ON(skb->ip_summed != CHECKSUM_HW); |
| 1205 | |
| 1206 | skb->mac.raw = skb->data; |
| 1207 | skb->mac_len = skb->nh.raw - skb->data; |
| 1208 | __skb_pull(skb, skb->mac_len); |
| 1209 | |
| 1210 | rcu_read_lock(); |
| 1211 | list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) { |
| 1212 | if (ptype->type == type && !ptype->dev && ptype->gso_segment) { |
| 1213 | segs = ptype->gso_segment(skb, sg); |
| 1214 | break; |
| 1215 | } |
| 1216 | } |
| 1217 | rcu_read_unlock(); |
| 1218 | |
| 1219 | return segs; |
| 1220 | } |
| 1221 | |
| 1222 | EXPORT_SYMBOL(skb_gso_segment); |
| 1223 | |
Herbert Xu | fb286bb | 2005-11-10 13:01:24 -0800 | [diff] [blame] | 1224 | /* Take action when hardware reception checksum errors are detected. */ |
| 1225 | #ifdef CONFIG_BUG |
| 1226 | void netdev_rx_csum_fault(struct net_device *dev) |
| 1227 | { |
| 1228 | if (net_ratelimit()) { |
Stephen Hemminger | 246a421 | 2005-12-08 15:21:39 -0800 | [diff] [blame] | 1229 | printk(KERN_ERR "%s: hw csum failure.\n", |
| 1230 | dev ? dev->name : "<unknown>"); |
Herbert Xu | fb286bb | 2005-11-10 13:01:24 -0800 | [diff] [blame] | 1231 | dump_stack(); |
| 1232 | } |
| 1233 | } |
| 1234 | EXPORT_SYMBOL(netdev_rx_csum_fault); |
| 1235 | #endif |
| 1236 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1237 | #ifdef CONFIG_HIGHMEM |
| 1238 | /* Actually, we should eliminate this check as soon as we know, that: |
| 1239 | * 1. IOMMU is present and allows to map all the memory. |
| 1240 | * 2. No high memory really exists on this machine. |
| 1241 | */ |
| 1242 | |
| 1243 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
| 1244 | { |
| 1245 | int i; |
| 1246 | |
| 1247 | if (dev->features & NETIF_F_HIGHDMA) |
| 1248 | return 0; |
| 1249 | |
| 1250 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) |
| 1251 | if (PageHighMem(skb_shinfo(skb)->frags[i].page)) |
| 1252 | return 1; |
| 1253 | |
| 1254 | return 0; |
| 1255 | } |
| 1256 | #else |
| 1257 | #define illegal_highdma(dev, skb) (0) |
| 1258 | #endif |
| 1259 | |
Herbert Xu | f6a78bf | 2006-06-22 02:57:17 -0700 | [diff] [blame] | 1260 | struct dev_gso_cb { |
| 1261 | void (*destructor)(struct sk_buff *skb); |
| 1262 | }; |
| 1263 | |
| 1264 | #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb) |
| 1265 | |
| 1266 | static void dev_gso_skb_destructor(struct sk_buff *skb) |
| 1267 | { |
| 1268 | struct dev_gso_cb *cb; |
| 1269 | |
| 1270 | do { |
| 1271 | struct sk_buff *nskb = skb->next; |
| 1272 | |
| 1273 | skb->next = nskb->next; |
| 1274 | nskb->next = NULL; |
| 1275 | kfree_skb(nskb); |
| 1276 | } while (skb->next); |
| 1277 | |
| 1278 | cb = DEV_GSO_CB(skb); |
| 1279 | if (cb->destructor) |
| 1280 | cb->destructor(skb); |
| 1281 | } |
| 1282 | |
| 1283 | /** |
| 1284 | * dev_gso_segment - Perform emulated hardware segmentation on skb. |
| 1285 | * @skb: buffer to segment |
| 1286 | * |
| 1287 | * This function segments the given skb and stores the list of segments |
| 1288 | * in skb->next. |
| 1289 | */ |
| 1290 | static int dev_gso_segment(struct sk_buff *skb) |
| 1291 | { |
| 1292 | struct net_device *dev = skb->dev; |
| 1293 | struct sk_buff *segs; |
| 1294 | |
| 1295 | segs = skb_gso_segment(skb, dev->features & NETIF_F_SG && |
| 1296 | !illegal_highdma(dev, skb)); |
| 1297 | if (unlikely(IS_ERR(segs))) |
| 1298 | return PTR_ERR(segs); |
| 1299 | |
| 1300 | skb->next = segs; |
| 1301 | DEV_GSO_CB(skb)->destructor = skb->destructor; |
| 1302 | skb->destructor = dev_gso_skb_destructor; |
| 1303 | |
| 1304 | return 0; |
| 1305 | } |
| 1306 | |
| 1307 | int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev) |
| 1308 | { |
| 1309 | if (likely(!skb->next)) { |
| 1310 | if (netdev_nit) |
| 1311 | dev_queue_xmit_nit(skb, dev); |
| 1312 | |
| 1313 | if (!netif_needs_gso(dev, skb)) |
| 1314 | return dev->hard_start_xmit(skb, dev); |
| 1315 | |
| 1316 | if (unlikely(dev_gso_segment(skb))) |
| 1317 | goto out_kfree_skb; |
| 1318 | } |
| 1319 | |
| 1320 | do { |
| 1321 | struct sk_buff *nskb = skb->next; |
| 1322 | int rc; |
| 1323 | |
| 1324 | skb->next = nskb->next; |
| 1325 | nskb->next = NULL; |
| 1326 | rc = dev->hard_start_xmit(nskb, dev); |
| 1327 | if (unlikely(rc)) { |
| 1328 | skb->next = nskb; |
| 1329 | return rc; |
| 1330 | } |
| 1331 | } while (skb->next); |
| 1332 | |
| 1333 | skb->destructor = DEV_GSO_CB(skb)->destructor; |
| 1334 | |
| 1335 | out_kfree_skb: |
| 1336 | kfree_skb(skb); |
| 1337 | return 0; |
| 1338 | } |
| 1339 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1340 | #define HARD_TX_LOCK(dev, cpu) { \ |
| 1341 | if ((dev->features & NETIF_F_LLTX) == 0) { \ |
Herbert Xu | 932ff27 | 2006-06-09 12:20:56 -0700 | [diff] [blame] | 1342 | netif_tx_lock(dev); \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1343 | } \ |
| 1344 | } |
| 1345 | |
| 1346 | #define HARD_TX_UNLOCK(dev) { \ |
| 1347 | if ((dev->features & NETIF_F_LLTX) == 0) { \ |
Herbert Xu | 932ff27 | 2006-06-09 12:20:56 -0700 | [diff] [blame] | 1348 | netif_tx_unlock(dev); \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1349 | } \ |
| 1350 | } |
| 1351 | |
| 1352 | /** |
| 1353 | * dev_queue_xmit - transmit a buffer |
| 1354 | * @skb: buffer to transmit |
| 1355 | * |
| 1356 | * Queue a buffer for transmission to a network device. The caller must |
| 1357 | * have set the device and priority and built the buffer before calling |
| 1358 | * this function. The function can be called from an interrupt. |
| 1359 | * |
| 1360 | * A negative errno code is returned on a failure. A success does not |
| 1361 | * guarantee the frame will be transmitted as it may be dropped due |
| 1362 | * to congestion or traffic shaping. |
Ben Greear | af19136 | 2005-04-24 20:12:36 -0700 | [diff] [blame] | 1363 | * |
| 1364 | * ----------------------------------------------------------------------------------- |
| 1365 | * I notice this method can also return errors from the queue disciplines, |
| 1366 | * including NET_XMIT_DROP, which is a positive value. So, errors can also |
| 1367 | * be positive. |
| 1368 | * |
| 1369 | * Regardless of the return value, the skb is consumed, so it is currently |
| 1370 | * difficult to retry a send to this method. (You can bump the ref count |
| 1371 | * before sending to hold a reference for retry if you are careful.) |
| 1372 | * |
| 1373 | * When calling this method, interrupts MUST be enabled. This is because |
| 1374 | * the BH enable code must have IRQs enabled so that it will not deadlock. |
| 1375 | * --BLG |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1376 | */ |
| 1377 | |
| 1378 | int dev_queue_xmit(struct sk_buff *skb) |
| 1379 | { |
| 1380 | struct net_device *dev = skb->dev; |
| 1381 | struct Qdisc *q; |
| 1382 | int rc = -ENOMEM; |
| 1383 | |
Herbert Xu | f6a78bf | 2006-06-22 02:57:17 -0700 | [diff] [blame] | 1384 | /* GSO will handle the following emulations directly. */ |
| 1385 | if (netif_needs_gso(dev, skb)) |
| 1386 | goto gso; |
| 1387 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1388 | if (skb_shinfo(skb)->frag_list && |
| 1389 | !(dev->features & NETIF_F_FRAGLIST) && |
Herbert Xu | 364c6ba | 2006-06-09 16:10:40 -0700 | [diff] [blame] | 1390 | __skb_linearize(skb)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1391 | goto out_kfree_skb; |
| 1392 | |
| 1393 | /* Fragmented skb is linearized if device does not support SG, |
| 1394 | * or if at least one of fragments is in highmem and device |
| 1395 | * does not support DMA from it. |
| 1396 | */ |
| 1397 | if (skb_shinfo(skb)->nr_frags && |
| 1398 | (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) && |
Herbert Xu | 364c6ba | 2006-06-09 16:10:40 -0700 | [diff] [blame] | 1399 | __skb_linearize(skb)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1400 | goto out_kfree_skb; |
| 1401 | |
| 1402 | /* If packet is not checksummed and device does not support |
| 1403 | * checksumming for this protocol, complete checksumming here. |
| 1404 | */ |
| 1405 | if (skb->ip_summed == CHECKSUM_HW && |
Herbert Xu | 8648b30 | 2006-06-17 22:06:05 -0700 | [diff] [blame] | 1406 | (!(dev->features & NETIF_F_GEN_CSUM) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1407 | (!(dev->features & NETIF_F_IP_CSUM) || |
| 1408 | skb->protocol != htons(ETH_P_IP)))) |
| 1409 | if (skb_checksum_help(skb, 0)) |
| 1410 | goto out_kfree_skb; |
| 1411 | |
Herbert Xu | f6a78bf | 2006-06-22 02:57:17 -0700 | [diff] [blame] | 1412 | gso: |
Eric Dumazet | 2d7ceec | 2005-09-27 15:22:58 -0700 | [diff] [blame] | 1413 | spin_lock_prefetch(&dev->queue_lock); |
| 1414 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1415 | /* Disable soft irqs for various locks below. Also |
| 1416 | * stops preemption for RCU. |
| 1417 | */ |
Herbert Xu | d4828d8 | 2006-06-22 02:28:18 -0700 | [diff] [blame] | 1418 | rcu_read_lock_bh(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1419 | |
| 1420 | /* Updates of qdisc are serialized by queue_lock. |
| 1421 | * The struct Qdisc which is pointed to by qdisc is now a |
| 1422 | * rcu structure - it may be accessed without acquiring |
| 1423 | * a lock (but the structure may be stale.) The freeing of the |
| 1424 | * qdisc will be deferred until it's known that there are no |
| 1425 | * more references to it. |
| 1426 | * |
| 1427 | * If the qdisc has an enqueue function, we still need to |
| 1428 | * hold the queue_lock before calling it, since queue_lock |
| 1429 | * also serializes access to the device queue. |
| 1430 | */ |
| 1431 | |
| 1432 | q = rcu_dereference(dev->qdisc); |
| 1433 | #ifdef CONFIG_NET_CLS_ACT |
| 1434 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS); |
| 1435 | #endif |
| 1436 | if (q->enqueue) { |
| 1437 | /* Grab device queue */ |
| 1438 | spin_lock(&dev->queue_lock); |
| 1439 | |
| 1440 | rc = q->enqueue(skb, q); |
| 1441 | |
| 1442 | qdisc_run(dev); |
| 1443 | |
| 1444 | spin_unlock(&dev->queue_lock); |
| 1445 | rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc; |
| 1446 | goto out; |
| 1447 | } |
| 1448 | |
| 1449 | /* The device has no queue. Common case for software devices: |
| 1450 | loopback, all the sorts of tunnels... |
| 1451 | |
Herbert Xu | 932ff27 | 2006-06-09 12:20:56 -0700 | [diff] [blame] | 1452 | Really, it is unlikely that netif_tx_lock protection is necessary |
| 1453 | here. (f.e. loopback and IP tunnels are clean ignoring statistics |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | counters.) |
| 1455 | However, it is possible, that they rely on protection |
| 1456 | made by us here. |
| 1457 | |
| 1458 | Check this and shot the lock. It is not prone from deadlocks. |
| 1459 | Either shot noqueue qdisc, it is even simpler 8) |
| 1460 | */ |
| 1461 | if (dev->flags & IFF_UP) { |
| 1462 | int cpu = smp_processor_id(); /* ok because BHs are off */ |
| 1463 | |
| 1464 | if (dev->xmit_lock_owner != cpu) { |
| 1465 | |
| 1466 | HARD_TX_LOCK(dev, cpu); |
| 1467 | |
| 1468 | if (!netif_queue_stopped(dev)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1469 | rc = 0; |
Herbert Xu | f6a78bf | 2006-06-22 02:57:17 -0700 | [diff] [blame] | 1470 | if (!dev_hard_start_xmit(skb, dev)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1471 | HARD_TX_UNLOCK(dev); |
| 1472 | goto out; |
| 1473 | } |
| 1474 | } |
| 1475 | HARD_TX_UNLOCK(dev); |
| 1476 | if (net_ratelimit()) |
| 1477 | printk(KERN_CRIT "Virtual device %s asks to " |
| 1478 | "queue packet!\n", dev->name); |
| 1479 | } else { |
| 1480 | /* Recursion is detected! It is possible, |
| 1481 | * unfortunately */ |
| 1482 | if (net_ratelimit()) |
| 1483 | printk(KERN_CRIT "Dead loop on virtual device " |
| 1484 | "%s, fix it urgently!\n", dev->name); |
| 1485 | } |
| 1486 | } |
| 1487 | |
| 1488 | rc = -ENETDOWN; |
Herbert Xu | d4828d8 | 2006-06-22 02:28:18 -0700 | [diff] [blame] | 1489 | rcu_read_unlock_bh(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1490 | |
| 1491 | out_kfree_skb: |
| 1492 | kfree_skb(skb); |
| 1493 | return rc; |
| 1494 | out: |
Herbert Xu | d4828d8 | 2006-06-22 02:28:18 -0700 | [diff] [blame] | 1495 | rcu_read_unlock_bh(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1496 | return rc; |
| 1497 | } |
| 1498 | |
| 1499 | |
| 1500 | /*======================================================================= |
| 1501 | Receiver routines |
| 1502 | =======================================================================*/ |
| 1503 | |
Stephen Hemminger | 51b0bde | 2005-06-23 20:14:40 -0700 | [diff] [blame] | 1504 | int netdev_max_backlog = 1000; |
| 1505 | int netdev_budget = 300; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | int weight_p = 64; /* old backlog weight */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1507 | |
| 1508 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, }; |
| 1509 | |
| 1510 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1511 | /** |
| 1512 | * netif_rx - post buffer to the network code |
| 1513 | * @skb: buffer to post |
| 1514 | * |
| 1515 | * This function receives a packet from a device driver and queues it for |
| 1516 | * the upper (protocol) levels to process. It always succeeds. The buffer |
| 1517 | * may be dropped during processing for congestion control or by the |
| 1518 | * protocol layers. |
| 1519 | * |
| 1520 | * return values: |
| 1521 | * NET_RX_SUCCESS (no congestion) |
| 1522 | * NET_RX_CN_LOW (low congestion) |
| 1523 | * NET_RX_CN_MOD (moderate congestion) |
| 1524 | * NET_RX_CN_HIGH (high congestion) |
| 1525 | * NET_RX_DROP (packet was dropped) |
| 1526 | * |
| 1527 | */ |
| 1528 | |
| 1529 | int netif_rx(struct sk_buff *skb) |
| 1530 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1531 | struct softnet_data *queue; |
| 1532 | unsigned long flags; |
| 1533 | |
| 1534 | /* if netpoll wants it, pretend we never saw it */ |
| 1535 | if (netpoll_rx(skb)) |
| 1536 | return NET_RX_DROP; |
| 1537 | |
Patrick McHardy | a61bbcf | 2005-08-14 17:24:31 -0700 | [diff] [blame] | 1538 | if (!skb->tstamp.off_sec) |
| 1539 | net_timestamp(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1540 | |
| 1541 | /* |
| 1542 | * The code is rearranged so that the path is the most |
| 1543 | * short when CPU is congested, but is still operating. |
| 1544 | */ |
| 1545 | local_irq_save(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1546 | queue = &__get_cpu_var(softnet_data); |
| 1547 | |
| 1548 | __get_cpu_var(netdev_rx_stat).total++; |
| 1549 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) { |
| 1550 | if (queue->input_pkt_queue.qlen) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1551 | enqueue: |
| 1552 | dev_hold(skb->dev); |
| 1553 | __skb_queue_tail(&queue->input_pkt_queue, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1554 | local_irq_restore(flags); |
Stephen Hemminger | 34008d8 | 2005-06-23 20:10:00 -0700 | [diff] [blame] | 1555 | return NET_RX_SUCCESS; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1556 | } |
| 1557 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1558 | netif_rx_schedule(&queue->backlog_dev); |
| 1559 | goto enqueue; |
| 1560 | } |
| 1561 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1562 | __get_cpu_var(netdev_rx_stat).dropped++; |
| 1563 | local_irq_restore(flags); |
| 1564 | |
| 1565 | kfree_skb(skb); |
| 1566 | return NET_RX_DROP; |
| 1567 | } |
| 1568 | |
| 1569 | int netif_rx_ni(struct sk_buff *skb) |
| 1570 | { |
| 1571 | int err; |
| 1572 | |
| 1573 | preempt_disable(); |
| 1574 | err = netif_rx(skb); |
| 1575 | if (local_softirq_pending()) |
| 1576 | do_softirq(); |
| 1577 | preempt_enable(); |
| 1578 | |
| 1579 | return err; |
| 1580 | } |
| 1581 | |
| 1582 | EXPORT_SYMBOL(netif_rx_ni); |
| 1583 | |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1584 | static inline struct net_device *skb_bond(struct sk_buff *skb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1585 | { |
| 1586 | struct net_device *dev = skb->dev; |
| 1587 | |
Jay Vosburgh | 8f903c7 | 2006-02-21 16:36:44 -0800 | [diff] [blame] | 1588 | if (dev->master) { |
| 1589 | /* |
| 1590 | * On bonding slaves other than the currently active |
| 1591 | * slave, suppress duplicates except for 802.3ad |
| 1592 | * ETH_P_SLOW and alb non-mcast/bcast. |
| 1593 | */ |
| 1594 | if (dev->priv_flags & IFF_SLAVE_INACTIVE) { |
| 1595 | if (dev->master->priv_flags & IFF_MASTER_ALB) { |
| 1596 | if (skb->pkt_type != PACKET_BROADCAST && |
| 1597 | skb->pkt_type != PACKET_MULTICAST) |
| 1598 | goto keep; |
| 1599 | } |
| 1600 | |
| 1601 | if (dev->master->priv_flags & IFF_MASTER_8023AD && |
| 1602 | skb->protocol == __constant_htons(ETH_P_SLOW)) |
| 1603 | goto keep; |
| 1604 | |
| 1605 | kfree_skb(skb); |
| 1606 | return NULL; |
| 1607 | } |
| 1608 | keep: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1609 | skb->dev = dev->master; |
Jay Vosburgh | 8f903c7 | 2006-02-21 16:36:44 -0800 | [diff] [blame] | 1610 | } |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1611 | |
| 1612 | return dev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1613 | } |
| 1614 | |
| 1615 | static void net_tx_action(struct softirq_action *h) |
| 1616 | { |
| 1617 | struct softnet_data *sd = &__get_cpu_var(softnet_data); |
| 1618 | |
| 1619 | if (sd->completion_queue) { |
| 1620 | struct sk_buff *clist; |
| 1621 | |
| 1622 | local_irq_disable(); |
| 1623 | clist = sd->completion_queue; |
| 1624 | sd->completion_queue = NULL; |
| 1625 | local_irq_enable(); |
| 1626 | |
| 1627 | while (clist) { |
| 1628 | struct sk_buff *skb = clist; |
| 1629 | clist = clist->next; |
| 1630 | |
| 1631 | BUG_TRAP(!atomic_read(&skb->users)); |
| 1632 | __kfree_skb(skb); |
| 1633 | } |
| 1634 | } |
| 1635 | |
| 1636 | if (sd->output_queue) { |
| 1637 | struct net_device *head; |
| 1638 | |
| 1639 | local_irq_disable(); |
| 1640 | head = sd->output_queue; |
| 1641 | sd->output_queue = NULL; |
| 1642 | local_irq_enable(); |
| 1643 | |
| 1644 | while (head) { |
| 1645 | struct net_device *dev = head; |
| 1646 | head = head->next_sched; |
| 1647 | |
| 1648 | smp_mb__before_clear_bit(); |
| 1649 | clear_bit(__LINK_STATE_SCHED, &dev->state); |
| 1650 | |
| 1651 | if (spin_trylock(&dev->queue_lock)) { |
| 1652 | qdisc_run(dev); |
| 1653 | spin_unlock(&dev->queue_lock); |
| 1654 | } else { |
| 1655 | netif_schedule(dev); |
| 1656 | } |
| 1657 | } |
| 1658 | } |
| 1659 | } |
| 1660 | |
| 1661 | static __inline__ int deliver_skb(struct sk_buff *skb, |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1662 | struct packet_type *pt_prev, |
| 1663 | struct net_device *orig_dev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1664 | { |
| 1665 | atomic_inc(&skb->users); |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1666 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1667 | } |
| 1668 | |
| 1669 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE) |
| 1670 | int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb); |
| 1671 | struct net_bridge; |
| 1672 | struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br, |
| 1673 | unsigned char *addr); |
| 1674 | void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent); |
| 1675 | |
| 1676 | static __inline__ int handle_bridge(struct sk_buff **pskb, |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1677 | struct packet_type **pt_prev, int *ret, |
| 1678 | struct net_device *orig_dev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1679 | { |
| 1680 | struct net_bridge_port *port; |
| 1681 | |
| 1682 | if ((*pskb)->pkt_type == PACKET_LOOPBACK || |
| 1683 | (port = rcu_dereference((*pskb)->dev->br_port)) == NULL) |
| 1684 | return 0; |
| 1685 | |
| 1686 | if (*pt_prev) { |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1687 | *ret = deliver_skb(*pskb, *pt_prev, orig_dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1688 | *pt_prev = NULL; |
| 1689 | } |
| 1690 | |
| 1691 | return br_handle_frame_hook(port, pskb); |
| 1692 | } |
| 1693 | #else |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1694 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1695 | #endif |
| 1696 | |
| 1697 | #ifdef CONFIG_NET_CLS_ACT |
| 1698 | /* TODO: Maybe we should just force sch_ingress to be compiled in |
| 1699 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions |
| 1700 | * a compare and 2 stores extra right now if we dont have it on |
| 1701 | * but have CONFIG_NET_CLS_ACT |
| 1702 | * NOTE: This doesnt stop any functionality; if you dont have |
| 1703 | * the ingress scheduler, you just cant add policies on ingress. |
| 1704 | * |
| 1705 | */ |
| 1706 | static int ing_filter(struct sk_buff *skb) |
| 1707 | { |
| 1708 | struct Qdisc *q; |
| 1709 | struct net_device *dev = skb->dev; |
| 1710 | int result = TC_ACT_OK; |
| 1711 | |
| 1712 | if (dev->qdisc_ingress) { |
| 1713 | __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd); |
| 1714 | if (MAX_RED_LOOP < ttl++) { |
| 1715 | printk("Redir loop detected Dropping packet (%s->%s)\n", |
David S. Miller | 86e65da | 2005-08-09 19:36:29 -0700 | [diff] [blame] | 1716 | skb->input_dev->name, skb->dev->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1717 | return TC_ACT_SHOT; |
| 1718 | } |
| 1719 | |
| 1720 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl); |
| 1721 | |
| 1722 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS); |
David S. Miller | 86e65da | 2005-08-09 19:36:29 -0700 | [diff] [blame] | 1723 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1724 | spin_lock(&dev->ingress_lock); |
| 1725 | if ((q = dev->qdisc_ingress) != NULL) |
| 1726 | result = q->enqueue(skb, q); |
| 1727 | spin_unlock(&dev->ingress_lock); |
| 1728 | |
| 1729 | } |
| 1730 | |
| 1731 | return result; |
| 1732 | } |
| 1733 | #endif |
| 1734 | |
| 1735 | int netif_receive_skb(struct sk_buff *skb) |
| 1736 | { |
| 1737 | struct packet_type *ptype, *pt_prev; |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1738 | struct net_device *orig_dev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1739 | int ret = NET_RX_DROP; |
| 1740 | unsigned short type; |
| 1741 | |
| 1742 | /* if we've gotten here through NAPI, check netpoll */ |
| 1743 | if (skb->dev->poll && netpoll_rx(skb)) |
| 1744 | return NET_RX_DROP; |
| 1745 | |
Patrick McHardy | a61bbcf | 2005-08-14 17:24:31 -0700 | [diff] [blame] | 1746 | if (!skb->tstamp.off_sec) |
| 1747 | net_timestamp(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1748 | |
David S. Miller | 86e65da | 2005-08-09 19:36:29 -0700 | [diff] [blame] | 1749 | if (!skb->input_dev) |
| 1750 | skb->input_dev = skb->dev; |
| 1751 | |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1752 | orig_dev = skb_bond(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1753 | |
Jay Vosburgh | 8f903c7 | 2006-02-21 16:36:44 -0800 | [diff] [blame] | 1754 | if (!orig_dev) |
| 1755 | return NET_RX_DROP; |
| 1756 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1757 | __get_cpu_var(netdev_rx_stat).total++; |
| 1758 | |
| 1759 | skb->h.raw = skb->nh.raw = skb->data; |
| 1760 | skb->mac_len = skb->nh.raw - skb->mac.raw; |
| 1761 | |
| 1762 | pt_prev = NULL; |
| 1763 | |
| 1764 | rcu_read_lock(); |
| 1765 | |
| 1766 | #ifdef CONFIG_NET_CLS_ACT |
| 1767 | if (skb->tc_verd & TC_NCLS) { |
| 1768 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd); |
| 1769 | goto ncls; |
| 1770 | } |
| 1771 | #endif |
| 1772 | |
| 1773 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
| 1774 | if (!ptype->dev || ptype->dev == skb->dev) { |
| 1775 | if (pt_prev) |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1776 | ret = deliver_skb(skb, pt_prev, orig_dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1777 | pt_prev = ptype; |
| 1778 | } |
| 1779 | } |
| 1780 | |
| 1781 | #ifdef CONFIG_NET_CLS_ACT |
| 1782 | if (pt_prev) { |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1783 | ret = deliver_skb(skb, pt_prev, orig_dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1784 | pt_prev = NULL; /* noone else should process this after*/ |
| 1785 | } else { |
| 1786 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd); |
| 1787 | } |
| 1788 | |
| 1789 | ret = ing_filter(skb); |
| 1790 | |
| 1791 | if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) { |
| 1792 | kfree_skb(skb); |
| 1793 | goto out; |
| 1794 | } |
| 1795 | |
| 1796 | skb->tc_verd = 0; |
| 1797 | ncls: |
| 1798 | #endif |
| 1799 | |
| 1800 | handle_diverter(skb); |
| 1801 | |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1802 | if (handle_bridge(&skb, &pt_prev, &ret, orig_dev)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1803 | goto out; |
| 1804 | |
| 1805 | type = skb->protocol; |
| 1806 | list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) { |
| 1807 | if (ptype->type == type && |
| 1808 | (!ptype->dev || ptype->dev == skb->dev)) { |
| 1809 | if (pt_prev) |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1810 | ret = deliver_skb(skb, pt_prev, orig_dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1811 | pt_prev = ptype; |
| 1812 | } |
| 1813 | } |
| 1814 | |
| 1815 | if (pt_prev) { |
David S. Miller | f2ccd8f | 2005-08-09 19:34:12 -0700 | [diff] [blame] | 1816 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1817 | } else { |
| 1818 | kfree_skb(skb); |
| 1819 | /* Jamal, now you will not able to escape explaining |
| 1820 | * me how you were going to use this. :-) |
| 1821 | */ |
| 1822 | ret = NET_RX_DROP; |
| 1823 | } |
| 1824 | |
| 1825 | out: |
| 1826 | rcu_read_unlock(); |
| 1827 | return ret; |
| 1828 | } |
| 1829 | |
| 1830 | static int process_backlog(struct net_device *backlog_dev, int *budget) |
| 1831 | { |
| 1832 | int work = 0; |
| 1833 | int quota = min(backlog_dev->quota, *budget); |
| 1834 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
| 1835 | unsigned long start_time = jiffies; |
| 1836 | |
Stephen Hemminger | e387660 | 2005-06-08 14:56:01 -0700 | [diff] [blame] | 1837 | backlog_dev->weight = weight_p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1838 | for (;;) { |
| 1839 | struct sk_buff *skb; |
| 1840 | struct net_device *dev; |
| 1841 | |
| 1842 | local_irq_disable(); |
| 1843 | skb = __skb_dequeue(&queue->input_pkt_queue); |
| 1844 | if (!skb) |
| 1845 | goto job_done; |
| 1846 | local_irq_enable(); |
| 1847 | |
| 1848 | dev = skb->dev; |
| 1849 | |
| 1850 | netif_receive_skb(skb); |
| 1851 | |
| 1852 | dev_put(dev); |
| 1853 | |
| 1854 | work++; |
| 1855 | |
| 1856 | if (work >= quota || jiffies - start_time > 1) |
| 1857 | break; |
| 1858 | |
| 1859 | } |
| 1860 | |
| 1861 | backlog_dev->quota -= work; |
| 1862 | *budget -= work; |
| 1863 | return -1; |
| 1864 | |
| 1865 | job_done: |
| 1866 | backlog_dev->quota -= work; |
| 1867 | *budget -= work; |
| 1868 | |
| 1869 | list_del(&backlog_dev->poll_list); |
| 1870 | smp_mb__before_clear_bit(); |
| 1871 | netif_poll_enable(backlog_dev); |
| 1872 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1873 | local_irq_enable(); |
| 1874 | return 0; |
| 1875 | } |
| 1876 | |
| 1877 | static void net_rx_action(struct softirq_action *h) |
| 1878 | { |
| 1879 | struct softnet_data *queue = &__get_cpu_var(softnet_data); |
| 1880 | unsigned long start_time = jiffies; |
Stephen Hemminger | 51b0bde | 2005-06-23 20:14:40 -0700 | [diff] [blame] | 1881 | int budget = netdev_budget; |
Matt Mackall | 53fb95d | 2005-08-11 19:27:43 -0700 | [diff] [blame] | 1882 | void *have; |
| 1883 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1884 | local_irq_disable(); |
| 1885 | |
| 1886 | while (!list_empty(&queue->poll_list)) { |
| 1887 | struct net_device *dev; |
| 1888 | |
| 1889 | if (budget <= 0 || jiffies - start_time > 1) |
| 1890 | goto softnet_break; |
| 1891 | |
| 1892 | local_irq_enable(); |
| 1893 | |
| 1894 | dev = list_entry(queue->poll_list.next, |
| 1895 | struct net_device, poll_list); |
Matt Mackall | 53fb95d | 2005-08-11 19:27:43 -0700 | [diff] [blame] | 1896 | have = netpoll_poll_lock(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1897 | |
| 1898 | if (dev->quota <= 0 || dev->poll(dev, &budget)) { |
Matt Mackall | 53fb95d | 2005-08-11 19:27:43 -0700 | [diff] [blame] | 1899 | netpoll_poll_unlock(have); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1900 | local_irq_disable(); |
Stephen Hemminger | 8aca8a2 | 2006-03-20 22:26:39 -0800 | [diff] [blame] | 1901 | list_move_tail(&dev->poll_list, &queue->poll_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1902 | if (dev->quota < 0) |
| 1903 | dev->quota += dev->weight; |
| 1904 | else |
| 1905 | dev->quota = dev->weight; |
| 1906 | } else { |
Matt Mackall | 53fb95d | 2005-08-11 19:27:43 -0700 | [diff] [blame] | 1907 | netpoll_poll_unlock(have); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1908 | dev_put(dev); |
| 1909 | local_irq_disable(); |
| 1910 | } |
| 1911 | } |
| 1912 | out: |
Chris Leech | db21733 | 2006-06-17 21:24:58 -0700 | [diff] [blame] | 1913 | #ifdef CONFIG_NET_DMA |
| 1914 | /* |
| 1915 | * There may not be any more sk_buffs coming right now, so push |
| 1916 | * any pending DMA copies to hardware |
| 1917 | */ |
| 1918 | if (net_dma_client) { |
| 1919 | struct dma_chan *chan; |
| 1920 | rcu_read_lock(); |
| 1921 | list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node) |
| 1922 | dma_async_memcpy_issue_pending(chan); |
| 1923 | rcu_read_unlock(); |
| 1924 | } |
| 1925 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1926 | local_irq_enable(); |
| 1927 | return; |
| 1928 | |
| 1929 | softnet_break: |
| 1930 | __get_cpu_var(netdev_rx_stat).time_squeeze++; |
| 1931 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); |
| 1932 | goto out; |
| 1933 | } |
| 1934 | |
| 1935 | static gifconf_func_t * gifconf_list [NPROTO]; |
| 1936 | |
| 1937 | /** |
| 1938 | * register_gifconf - register a SIOCGIF handler |
| 1939 | * @family: Address family |
| 1940 | * @gifconf: Function handler |
| 1941 | * |
| 1942 | * Register protocol dependent address dumping routines. The handler |
| 1943 | * that is passed must not be freed or reused until it has been replaced |
| 1944 | * by another handler. |
| 1945 | */ |
| 1946 | int register_gifconf(unsigned int family, gifconf_func_t * gifconf) |
| 1947 | { |
| 1948 | if (family >= NPROTO) |
| 1949 | return -EINVAL; |
| 1950 | gifconf_list[family] = gifconf; |
| 1951 | return 0; |
| 1952 | } |
| 1953 | |
| 1954 | |
| 1955 | /* |
| 1956 | * Map an interface index to its name (SIOCGIFNAME) |
| 1957 | */ |
| 1958 | |
| 1959 | /* |
| 1960 | * We need this ioctl for efficient implementation of the |
| 1961 | * if_indextoname() function required by the IPv6 API. Without |
| 1962 | * it, we would have to search all the interfaces to find a |
| 1963 | * match. --pb |
| 1964 | */ |
| 1965 | |
| 1966 | static int dev_ifname(struct ifreq __user *arg) |
| 1967 | { |
| 1968 | struct net_device *dev; |
| 1969 | struct ifreq ifr; |
| 1970 | |
| 1971 | /* |
| 1972 | * Fetch the caller's info block. |
| 1973 | */ |
| 1974 | |
| 1975 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) |
| 1976 | return -EFAULT; |
| 1977 | |
| 1978 | read_lock(&dev_base_lock); |
| 1979 | dev = __dev_get_by_index(ifr.ifr_ifindex); |
| 1980 | if (!dev) { |
| 1981 | read_unlock(&dev_base_lock); |
| 1982 | return -ENODEV; |
| 1983 | } |
| 1984 | |
| 1985 | strcpy(ifr.ifr_name, dev->name); |
| 1986 | read_unlock(&dev_base_lock); |
| 1987 | |
| 1988 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq))) |
| 1989 | return -EFAULT; |
| 1990 | return 0; |
| 1991 | } |
| 1992 | |
| 1993 | /* |
| 1994 | * Perform a SIOCGIFCONF call. This structure will change |
| 1995 | * size eventually, and there is nothing I can do about it. |
| 1996 | * Thus we will need a 'compatibility mode'. |
| 1997 | */ |
| 1998 | |
| 1999 | static int dev_ifconf(char __user *arg) |
| 2000 | { |
| 2001 | struct ifconf ifc; |
| 2002 | struct net_device *dev; |
| 2003 | char __user *pos; |
| 2004 | int len; |
| 2005 | int total; |
| 2006 | int i; |
| 2007 | |
| 2008 | /* |
| 2009 | * Fetch the caller's info block. |
| 2010 | */ |
| 2011 | |
| 2012 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf))) |
| 2013 | return -EFAULT; |
| 2014 | |
| 2015 | pos = ifc.ifc_buf; |
| 2016 | len = ifc.ifc_len; |
| 2017 | |
| 2018 | /* |
| 2019 | * Loop over the interfaces, and write an info block for each. |
| 2020 | */ |
| 2021 | |
| 2022 | total = 0; |
| 2023 | for (dev = dev_base; dev; dev = dev->next) { |
| 2024 | for (i = 0; i < NPROTO; i++) { |
| 2025 | if (gifconf_list[i]) { |
| 2026 | int done; |
| 2027 | if (!pos) |
| 2028 | done = gifconf_list[i](dev, NULL, 0); |
| 2029 | else |
| 2030 | done = gifconf_list[i](dev, pos + total, |
| 2031 | len - total); |
| 2032 | if (done < 0) |
| 2033 | return -EFAULT; |
| 2034 | total += done; |
| 2035 | } |
| 2036 | } |
| 2037 | } |
| 2038 | |
| 2039 | /* |
| 2040 | * All done. Write the updated control block back to the caller. |
| 2041 | */ |
| 2042 | ifc.ifc_len = total; |
| 2043 | |
| 2044 | /* |
| 2045 | * Both BSD and Solaris return 0 here, so we do too. |
| 2046 | */ |
| 2047 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0; |
| 2048 | } |
| 2049 | |
| 2050 | #ifdef CONFIG_PROC_FS |
| 2051 | /* |
| 2052 | * This is invoked by the /proc filesystem handler to display a device |
| 2053 | * in detail. |
| 2054 | */ |
| 2055 | static __inline__ struct net_device *dev_get_idx(loff_t pos) |
| 2056 | { |
| 2057 | struct net_device *dev; |
| 2058 | loff_t i; |
| 2059 | |
| 2060 | for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next); |
| 2061 | |
| 2062 | return i == pos ? dev : NULL; |
| 2063 | } |
| 2064 | |
| 2065 | void *dev_seq_start(struct seq_file *seq, loff_t *pos) |
| 2066 | { |
| 2067 | read_lock(&dev_base_lock); |
| 2068 | return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN; |
| 2069 | } |
| 2070 | |
| 2071 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
| 2072 | { |
| 2073 | ++*pos; |
| 2074 | return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next; |
| 2075 | } |
| 2076 | |
| 2077 | void dev_seq_stop(struct seq_file *seq, void *v) |
| 2078 | { |
| 2079 | read_unlock(&dev_base_lock); |
| 2080 | } |
| 2081 | |
| 2082 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev) |
| 2083 | { |
| 2084 | if (dev->get_stats) { |
| 2085 | struct net_device_stats *stats = dev->get_stats(dev); |
| 2086 | |
| 2087 | seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu " |
| 2088 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n", |
| 2089 | dev->name, stats->rx_bytes, stats->rx_packets, |
| 2090 | stats->rx_errors, |
| 2091 | stats->rx_dropped + stats->rx_missed_errors, |
| 2092 | stats->rx_fifo_errors, |
| 2093 | stats->rx_length_errors + stats->rx_over_errors + |
| 2094 | stats->rx_crc_errors + stats->rx_frame_errors, |
| 2095 | stats->rx_compressed, stats->multicast, |
| 2096 | stats->tx_bytes, stats->tx_packets, |
| 2097 | stats->tx_errors, stats->tx_dropped, |
| 2098 | stats->tx_fifo_errors, stats->collisions, |
| 2099 | stats->tx_carrier_errors + |
| 2100 | stats->tx_aborted_errors + |
| 2101 | stats->tx_window_errors + |
| 2102 | stats->tx_heartbeat_errors, |
| 2103 | stats->tx_compressed); |
| 2104 | } else |
| 2105 | seq_printf(seq, "%6s: No statistics available.\n", dev->name); |
| 2106 | } |
| 2107 | |
| 2108 | /* |
| 2109 | * Called from the PROCfs module. This now uses the new arbitrary sized |
| 2110 | * /proc/net interface to create /proc/net/dev |
| 2111 | */ |
| 2112 | static int dev_seq_show(struct seq_file *seq, void *v) |
| 2113 | { |
| 2114 | if (v == SEQ_START_TOKEN) |
| 2115 | seq_puts(seq, "Inter-| Receive " |
| 2116 | " | Transmit\n" |
| 2117 | " face |bytes packets errs drop fifo frame " |
| 2118 | "compressed multicast|bytes packets errs " |
| 2119 | "drop fifo colls carrier compressed\n"); |
| 2120 | else |
| 2121 | dev_seq_printf_stats(seq, v); |
| 2122 | return 0; |
| 2123 | } |
| 2124 | |
| 2125 | static struct netif_rx_stats *softnet_get_online(loff_t *pos) |
| 2126 | { |
| 2127 | struct netif_rx_stats *rc = NULL; |
| 2128 | |
| 2129 | while (*pos < NR_CPUS) |
| 2130 | if (cpu_online(*pos)) { |
| 2131 | rc = &per_cpu(netdev_rx_stat, *pos); |
| 2132 | break; |
| 2133 | } else |
| 2134 | ++*pos; |
| 2135 | return rc; |
| 2136 | } |
| 2137 | |
| 2138 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos) |
| 2139 | { |
| 2140 | return softnet_get_online(pos); |
| 2141 | } |
| 2142 | |
| 2143 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
| 2144 | { |
| 2145 | ++*pos; |
| 2146 | return softnet_get_online(pos); |
| 2147 | } |
| 2148 | |
| 2149 | static void softnet_seq_stop(struct seq_file *seq, void *v) |
| 2150 | { |
| 2151 | } |
| 2152 | |
| 2153 | static int softnet_seq_show(struct seq_file *seq, void *v) |
| 2154 | { |
| 2155 | struct netif_rx_stats *s = v; |
| 2156 | |
| 2157 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n", |
Stephen Hemminger | 31aa02c | 2005-06-23 20:12:48 -0700 | [diff] [blame] | 2158 | s->total, s->dropped, s->time_squeeze, 0, |
Stephen Hemminger | c1ebcdb | 2005-06-23 20:08:59 -0700 | [diff] [blame] | 2159 | 0, 0, 0, 0, /* was fastroute */ |
| 2160 | s->cpu_collision ); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2161 | return 0; |
| 2162 | } |
| 2163 | |
| 2164 | static struct seq_operations dev_seq_ops = { |
| 2165 | .start = dev_seq_start, |
| 2166 | .next = dev_seq_next, |
| 2167 | .stop = dev_seq_stop, |
| 2168 | .show = dev_seq_show, |
| 2169 | }; |
| 2170 | |
| 2171 | static int dev_seq_open(struct inode *inode, struct file *file) |
| 2172 | { |
| 2173 | return seq_open(file, &dev_seq_ops); |
| 2174 | } |
| 2175 | |
| 2176 | static struct file_operations dev_seq_fops = { |
| 2177 | .owner = THIS_MODULE, |
| 2178 | .open = dev_seq_open, |
| 2179 | .read = seq_read, |
| 2180 | .llseek = seq_lseek, |
| 2181 | .release = seq_release, |
| 2182 | }; |
| 2183 | |
| 2184 | static struct seq_operations softnet_seq_ops = { |
| 2185 | .start = softnet_seq_start, |
| 2186 | .next = softnet_seq_next, |
| 2187 | .stop = softnet_seq_stop, |
| 2188 | .show = softnet_seq_show, |
| 2189 | }; |
| 2190 | |
| 2191 | static int softnet_seq_open(struct inode *inode, struct file *file) |
| 2192 | { |
| 2193 | return seq_open(file, &softnet_seq_ops); |
| 2194 | } |
| 2195 | |
| 2196 | static struct file_operations softnet_seq_fops = { |
| 2197 | .owner = THIS_MODULE, |
| 2198 | .open = softnet_seq_open, |
| 2199 | .read = seq_read, |
| 2200 | .llseek = seq_lseek, |
| 2201 | .release = seq_release, |
| 2202 | }; |
| 2203 | |
Adrian Bunk | d86b5e0 | 2006-01-21 00:46:55 +0100 | [diff] [blame] | 2204 | #ifdef CONFIG_WIRELESS_EXT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2205 | extern int wireless_proc_init(void); |
| 2206 | #else |
| 2207 | #define wireless_proc_init() 0 |
| 2208 | #endif |
| 2209 | |
| 2210 | static int __init dev_proc_init(void) |
| 2211 | { |
| 2212 | int rc = -ENOMEM; |
| 2213 | |
| 2214 | if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops)) |
| 2215 | goto out; |
| 2216 | if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops)) |
| 2217 | goto out_dev; |
| 2218 | if (wireless_proc_init()) |
| 2219 | goto out_softnet; |
| 2220 | rc = 0; |
| 2221 | out: |
| 2222 | return rc; |
| 2223 | out_softnet: |
| 2224 | proc_net_remove("softnet_stat"); |
| 2225 | out_dev: |
| 2226 | proc_net_remove("dev"); |
| 2227 | goto out; |
| 2228 | } |
| 2229 | #else |
| 2230 | #define dev_proc_init() 0 |
| 2231 | #endif /* CONFIG_PROC_FS */ |
| 2232 | |
| 2233 | |
| 2234 | /** |
| 2235 | * netdev_set_master - set up master/slave pair |
| 2236 | * @slave: slave device |
| 2237 | * @master: new master device |
| 2238 | * |
| 2239 | * Changes the master device of the slave. Pass %NULL to break the |
| 2240 | * bonding. The caller must hold the RTNL semaphore. On a failure |
| 2241 | * a negative errno code is returned. On success the reference counts |
| 2242 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the |
| 2243 | * function returns zero. |
| 2244 | */ |
| 2245 | int netdev_set_master(struct net_device *slave, struct net_device *master) |
| 2246 | { |
| 2247 | struct net_device *old = slave->master; |
| 2248 | |
| 2249 | ASSERT_RTNL(); |
| 2250 | |
| 2251 | if (master) { |
| 2252 | if (old) |
| 2253 | return -EBUSY; |
| 2254 | dev_hold(master); |
| 2255 | } |
| 2256 | |
| 2257 | slave->master = master; |
| 2258 | |
| 2259 | synchronize_net(); |
| 2260 | |
| 2261 | if (old) |
| 2262 | dev_put(old); |
| 2263 | |
| 2264 | if (master) |
| 2265 | slave->flags |= IFF_SLAVE; |
| 2266 | else |
| 2267 | slave->flags &= ~IFF_SLAVE; |
| 2268 | |
| 2269 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE); |
| 2270 | return 0; |
| 2271 | } |
| 2272 | |
| 2273 | /** |
| 2274 | * dev_set_promiscuity - update promiscuity count on a device |
| 2275 | * @dev: device |
| 2276 | * @inc: modifier |
| 2277 | * |
Stephen Hemminger | 3041a06 | 2006-05-26 13:25:24 -0700 | [diff] [blame] | 2278 | * Add or remove promiscuity from a device. While the count in the device |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2279 | * remains above zero the interface remains promiscuous. Once it hits zero |
| 2280 | * the device reverts back to normal filtering operation. A negative inc |
| 2281 | * value is used to drop promiscuity on the device. |
| 2282 | */ |
| 2283 | void dev_set_promiscuity(struct net_device *dev, int inc) |
| 2284 | { |
| 2285 | unsigned short old_flags = dev->flags; |
| 2286 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2287 | if ((dev->promiscuity += inc) == 0) |
| 2288 | dev->flags &= ~IFF_PROMISC; |
David Chau | 52609c0 | 2005-07-05 15:11:06 -0700 | [diff] [blame] | 2289 | else |
| 2290 | dev->flags |= IFF_PROMISC; |
| 2291 | if (dev->flags != old_flags) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2292 | dev_mc_upload(dev); |
| 2293 | printk(KERN_INFO "device %s %s promiscuous mode\n", |
| 2294 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" : |
| 2295 | "left"); |
Steve Grubb | 5bdb988 | 2005-12-03 08:39:35 -0500 | [diff] [blame] | 2296 | audit_log(current->audit_context, GFP_ATOMIC, |
| 2297 | AUDIT_ANOM_PROMISCUOUS, |
| 2298 | "dev=%s prom=%d old_prom=%d auid=%u", |
| 2299 | dev->name, (dev->flags & IFF_PROMISC), |
| 2300 | (old_flags & IFF_PROMISC), |
| 2301 | audit_get_loginuid(current->audit_context)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2302 | } |
| 2303 | } |
| 2304 | |
| 2305 | /** |
| 2306 | * dev_set_allmulti - update allmulti count on a device |
| 2307 | * @dev: device |
| 2308 | * @inc: modifier |
| 2309 | * |
| 2310 | * Add or remove reception of all multicast frames to a device. While the |
| 2311 | * count in the device remains above zero the interface remains listening |
| 2312 | * to all interfaces. Once it hits zero the device reverts back to normal |
| 2313 | * filtering operation. A negative @inc value is used to drop the counter |
| 2314 | * when releasing a resource needing all multicasts. |
| 2315 | */ |
| 2316 | |
| 2317 | void dev_set_allmulti(struct net_device *dev, int inc) |
| 2318 | { |
| 2319 | unsigned short old_flags = dev->flags; |
| 2320 | |
| 2321 | dev->flags |= IFF_ALLMULTI; |
| 2322 | if ((dev->allmulti += inc) == 0) |
| 2323 | dev->flags &= ~IFF_ALLMULTI; |
| 2324 | if (dev->flags ^ old_flags) |
| 2325 | dev_mc_upload(dev); |
| 2326 | } |
| 2327 | |
| 2328 | unsigned dev_get_flags(const struct net_device *dev) |
| 2329 | { |
| 2330 | unsigned flags; |
| 2331 | |
| 2332 | flags = (dev->flags & ~(IFF_PROMISC | |
| 2333 | IFF_ALLMULTI | |
Stefan Rompf | b00055a | 2006-03-20 17:09:11 -0800 | [diff] [blame] | 2334 | IFF_RUNNING | |
| 2335 | IFF_LOWER_UP | |
| 2336 | IFF_DORMANT)) | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2337 | (dev->gflags & (IFF_PROMISC | |
| 2338 | IFF_ALLMULTI)); |
| 2339 | |
Stefan Rompf | b00055a | 2006-03-20 17:09:11 -0800 | [diff] [blame] | 2340 | if (netif_running(dev)) { |
| 2341 | if (netif_oper_up(dev)) |
| 2342 | flags |= IFF_RUNNING; |
| 2343 | if (netif_carrier_ok(dev)) |
| 2344 | flags |= IFF_LOWER_UP; |
| 2345 | if (netif_dormant(dev)) |
| 2346 | flags |= IFF_DORMANT; |
| 2347 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2348 | |
| 2349 | return flags; |
| 2350 | } |
| 2351 | |
| 2352 | int dev_change_flags(struct net_device *dev, unsigned flags) |
| 2353 | { |
| 2354 | int ret; |
| 2355 | int old_flags = dev->flags; |
| 2356 | |
| 2357 | /* |
| 2358 | * Set the flags on our device. |
| 2359 | */ |
| 2360 | |
| 2361 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | |
| 2362 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | |
| 2363 | IFF_AUTOMEDIA)) | |
| 2364 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | |
| 2365 | IFF_ALLMULTI)); |
| 2366 | |
| 2367 | /* |
| 2368 | * Load in the correct multicast list now the flags have changed. |
| 2369 | */ |
| 2370 | |
| 2371 | dev_mc_upload(dev); |
| 2372 | |
| 2373 | /* |
| 2374 | * Have we downed the interface. We handle IFF_UP ourselves |
| 2375 | * according to user attempts to set it, rather than blindly |
| 2376 | * setting it. |
| 2377 | */ |
| 2378 | |
| 2379 | ret = 0; |
| 2380 | if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */ |
| 2381 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev); |
| 2382 | |
| 2383 | if (!ret) |
| 2384 | dev_mc_upload(dev); |
| 2385 | } |
| 2386 | |
| 2387 | if (dev->flags & IFF_UP && |
| 2388 | ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI | |
| 2389 | IFF_VOLATILE))) |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 2390 | raw_notifier_call_chain(&netdev_chain, |
Alan Stern | e041c68 | 2006-03-27 01:16:30 -0800 | [diff] [blame] | 2391 | NETDEV_CHANGE, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2392 | |
| 2393 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
| 2394 | int inc = (flags & IFF_PROMISC) ? +1 : -1; |
| 2395 | dev->gflags ^= IFF_PROMISC; |
| 2396 | dev_set_promiscuity(dev, inc); |
| 2397 | } |
| 2398 | |
| 2399 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI |
| 2400 | is important. Some (broken) drivers set IFF_PROMISC, when |
| 2401 | IFF_ALLMULTI is requested not asking us and not reporting. |
| 2402 | */ |
| 2403 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { |
| 2404 | int inc = (flags & IFF_ALLMULTI) ? +1 : -1; |
| 2405 | dev->gflags ^= IFF_ALLMULTI; |
| 2406 | dev_set_allmulti(dev, inc); |
| 2407 | } |
| 2408 | |
| 2409 | if (old_flags ^ dev->flags) |
| 2410 | rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags); |
| 2411 | |
| 2412 | return ret; |
| 2413 | } |
| 2414 | |
| 2415 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
| 2416 | { |
| 2417 | int err; |
| 2418 | |
| 2419 | if (new_mtu == dev->mtu) |
| 2420 | return 0; |
| 2421 | |
| 2422 | /* MTU must be positive. */ |
| 2423 | if (new_mtu < 0) |
| 2424 | return -EINVAL; |
| 2425 | |
| 2426 | if (!netif_device_present(dev)) |
| 2427 | return -ENODEV; |
| 2428 | |
| 2429 | err = 0; |
| 2430 | if (dev->change_mtu) |
| 2431 | err = dev->change_mtu(dev, new_mtu); |
| 2432 | else |
| 2433 | dev->mtu = new_mtu; |
| 2434 | if (!err && dev->flags & IFF_UP) |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 2435 | raw_notifier_call_chain(&netdev_chain, |
Alan Stern | e041c68 | 2006-03-27 01:16:30 -0800 | [diff] [blame] | 2436 | NETDEV_CHANGEMTU, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2437 | return err; |
| 2438 | } |
| 2439 | |
| 2440 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
| 2441 | { |
| 2442 | int err; |
| 2443 | |
| 2444 | if (!dev->set_mac_address) |
| 2445 | return -EOPNOTSUPP; |
| 2446 | if (sa->sa_family != dev->type) |
| 2447 | return -EINVAL; |
| 2448 | if (!netif_device_present(dev)) |
| 2449 | return -ENODEV; |
| 2450 | err = dev->set_mac_address(dev, sa); |
| 2451 | if (!err) |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 2452 | raw_notifier_call_chain(&netdev_chain, |
Alan Stern | e041c68 | 2006-03-27 01:16:30 -0800 | [diff] [blame] | 2453 | NETDEV_CHANGEADDR, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2454 | return err; |
| 2455 | } |
| 2456 | |
| 2457 | /* |
| 2458 | * Perform the SIOCxIFxxx calls. |
| 2459 | */ |
| 2460 | static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd) |
| 2461 | { |
| 2462 | int err; |
| 2463 | struct net_device *dev = __dev_get_by_name(ifr->ifr_name); |
| 2464 | |
| 2465 | if (!dev) |
| 2466 | return -ENODEV; |
| 2467 | |
| 2468 | switch (cmd) { |
| 2469 | case SIOCGIFFLAGS: /* Get interface flags */ |
| 2470 | ifr->ifr_flags = dev_get_flags(dev); |
| 2471 | return 0; |
| 2472 | |
| 2473 | case SIOCSIFFLAGS: /* Set interface flags */ |
| 2474 | return dev_change_flags(dev, ifr->ifr_flags); |
| 2475 | |
| 2476 | case SIOCGIFMETRIC: /* Get the metric on the interface |
| 2477 | (currently unused) */ |
| 2478 | ifr->ifr_metric = 0; |
| 2479 | return 0; |
| 2480 | |
| 2481 | case SIOCSIFMETRIC: /* Set the metric on the interface |
| 2482 | (currently unused) */ |
| 2483 | return -EOPNOTSUPP; |
| 2484 | |
| 2485 | case SIOCGIFMTU: /* Get the MTU of a device */ |
| 2486 | ifr->ifr_mtu = dev->mtu; |
| 2487 | return 0; |
| 2488 | |
| 2489 | case SIOCSIFMTU: /* Set the MTU of a device */ |
| 2490 | return dev_set_mtu(dev, ifr->ifr_mtu); |
| 2491 | |
| 2492 | case SIOCGIFHWADDR: |
| 2493 | if (!dev->addr_len) |
| 2494 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data); |
| 2495 | else |
| 2496 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr, |
| 2497 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); |
| 2498 | ifr->ifr_hwaddr.sa_family = dev->type; |
| 2499 | return 0; |
| 2500 | |
| 2501 | case SIOCSIFHWADDR: |
| 2502 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr); |
| 2503 | |
| 2504 | case SIOCSIFHWBROADCAST: |
| 2505 | if (ifr->ifr_hwaddr.sa_family != dev->type) |
| 2506 | return -EINVAL; |
| 2507 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data, |
| 2508 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len)); |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 2509 | raw_notifier_call_chain(&netdev_chain, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2510 | NETDEV_CHANGEADDR, dev); |
| 2511 | return 0; |
| 2512 | |
| 2513 | case SIOCGIFMAP: |
| 2514 | ifr->ifr_map.mem_start = dev->mem_start; |
| 2515 | ifr->ifr_map.mem_end = dev->mem_end; |
| 2516 | ifr->ifr_map.base_addr = dev->base_addr; |
| 2517 | ifr->ifr_map.irq = dev->irq; |
| 2518 | ifr->ifr_map.dma = dev->dma; |
| 2519 | ifr->ifr_map.port = dev->if_port; |
| 2520 | return 0; |
| 2521 | |
| 2522 | case SIOCSIFMAP: |
| 2523 | if (dev->set_config) { |
| 2524 | if (!netif_device_present(dev)) |
| 2525 | return -ENODEV; |
| 2526 | return dev->set_config(dev, &ifr->ifr_map); |
| 2527 | } |
| 2528 | return -EOPNOTSUPP; |
| 2529 | |
| 2530 | case SIOCADDMULTI: |
| 2531 | if (!dev->set_multicast_list || |
| 2532 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
| 2533 | return -EINVAL; |
| 2534 | if (!netif_device_present(dev)) |
| 2535 | return -ENODEV; |
| 2536 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data, |
| 2537 | dev->addr_len, 1); |
| 2538 | |
| 2539 | case SIOCDELMULTI: |
| 2540 | if (!dev->set_multicast_list || |
| 2541 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC) |
| 2542 | return -EINVAL; |
| 2543 | if (!netif_device_present(dev)) |
| 2544 | return -ENODEV; |
| 2545 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data, |
| 2546 | dev->addr_len, 1); |
| 2547 | |
| 2548 | case SIOCGIFINDEX: |
| 2549 | ifr->ifr_ifindex = dev->ifindex; |
| 2550 | return 0; |
| 2551 | |
| 2552 | case SIOCGIFTXQLEN: |
| 2553 | ifr->ifr_qlen = dev->tx_queue_len; |
| 2554 | return 0; |
| 2555 | |
| 2556 | case SIOCSIFTXQLEN: |
| 2557 | if (ifr->ifr_qlen < 0) |
| 2558 | return -EINVAL; |
| 2559 | dev->tx_queue_len = ifr->ifr_qlen; |
| 2560 | return 0; |
| 2561 | |
| 2562 | case SIOCSIFNAME: |
| 2563 | ifr->ifr_newname[IFNAMSIZ-1] = '\0'; |
| 2564 | return dev_change_name(dev, ifr->ifr_newname); |
| 2565 | |
| 2566 | /* |
| 2567 | * Unknown or private ioctl |
| 2568 | */ |
| 2569 | |
| 2570 | default: |
| 2571 | if ((cmd >= SIOCDEVPRIVATE && |
| 2572 | cmd <= SIOCDEVPRIVATE + 15) || |
| 2573 | cmd == SIOCBONDENSLAVE || |
| 2574 | cmd == SIOCBONDRELEASE || |
| 2575 | cmd == SIOCBONDSETHWADDR || |
| 2576 | cmd == SIOCBONDSLAVEINFOQUERY || |
| 2577 | cmd == SIOCBONDINFOQUERY || |
| 2578 | cmd == SIOCBONDCHANGEACTIVE || |
| 2579 | cmd == SIOCGMIIPHY || |
| 2580 | cmd == SIOCGMIIREG || |
| 2581 | cmd == SIOCSMIIREG || |
| 2582 | cmd == SIOCBRADDIF || |
| 2583 | cmd == SIOCBRDELIF || |
| 2584 | cmd == SIOCWANDEV) { |
| 2585 | err = -EOPNOTSUPP; |
| 2586 | if (dev->do_ioctl) { |
| 2587 | if (netif_device_present(dev)) |
| 2588 | err = dev->do_ioctl(dev, ifr, |
| 2589 | cmd); |
| 2590 | else |
| 2591 | err = -ENODEV; |
| 2592 | } |
| 2593 | } else |
| 2594 | err = -EINVAL; |
| 2595 | |
| 2596 | } |
| 2597 | return err; |
| 2598 | } |
| 2599 | |
| 2600 | /* |
| 2601 | * This function handles all "interface"-type I/O control requests. The actual |
| 2602 | * 'doing' part of this is dev_ifsioc above. |
| 2603 | */ |
| 2604 | |
| 2605 | /** |
| 2606 | * dev_ioctl - network device ioctl |
| 2607 | * @cmd: command to issue |
| 2608 | * @arg: pointer to a struct ifreq in user space |
| 2609 | * |
| 2610 | * Issue ioctl functions to devices. This is normally called by the |
| 2611 | * user space syscall interfaces but can sometimes be useful for |
| 2612 | * other purposes. The return value is the return from the syscall if |
| 2613 | * positive or a negative errno code on error. |
| 2614 | */ |
| 2615 | |
| 2616 | int dev_ioctl(unsigned int cmd, void __user *arg) |
| 2617 | { |
| 2618 | struct ifreq ifr; |
| 2619 | int ret; |
| 2620 | char *colon; |
| 2621 | |
| 2622 | /* One special case: SIOCGIFCONF takes ifconf argument |
| 2623 | and requires shared lock, because it sleeps writing |
| 2624 | to user space. |
| 2625 | */ |
| 2626 | |
| 2627 | if (cmd == SIOCGIFCONF) { |
Stephen Hemminger | 6756ae4 | 2006-03-20 22:23:58 -0800 | [diff] [blame] | 2628 | rtnl_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2629 | ret = dev_ifconf((char __user *) arg); |
Stephen Hemminger | 6756ae4 | 2006-03-20 22:23:58 -0800 | [diff] [blame] | 2630 | rtnl_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2631 | return ret; |
| 2632 | } |
| 2633 | if (cmd == SIOCGIFNAME) |
| 2634 | return dev_ifname((struct ifreq __user *)arg); |
| 2635 | |
| 2636 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq))) |
| 2637 | return -EFAULT; |
| 2638 | |
| 2639 | ifr.ifr_name[IFNAMSIZ-1] = 0; |
| 2640 | |
| 2641 | colon = strchr(ifr.ifr_name, ':'); |
| 2642 | if (colon) |
| 2643 | *colon = 0; |
| 2644 | |
| 2645 | /* |
| 2646 | * See which interface the caller is talking about. |
| 2647 | */ |
| 2648 | |
| 2649 | switch (cmd) { |
| 2650 | /* |
| 2651 | * These ioctl calls: |
| 2652 | * - can be done by all. |
| 2653 | * - atomic and do not require locking. |
| 2654 | * - return a value |
| 2655 | */ |
| 2656 | case SIOCGIFFLAGS: |
| 2657 | case SIOCGIFMETRIC: |
| 2658 | case SIOCGIFMTU: |
| 2659 | case SIOCGIFHWADDR: |
| 2660 | case SIOCGIFSLAVE: |
| 2661 | case SIOCGIFMAP: |
| 2662 | case SIOCGIFINDEX: |
| 2663 | case SIOCGIFTXQLEN: |
| 2664 | dev_load(ifr.ifr_name); |
| 2665 | read_lock(&dev_base_lock); |
| 2666 | ret = dev_ifsioc(&ifr, cmd); |
| 2667 | read_unlock(&dev_base_lock); |
| 2668 | if (!ret) { |
| 2669 | if (colon) |
| 2670 | *colon = ':'; |
| 2671 | if (copy_to_user(arg, &ifr, |
| 2672 | sizeof(struct ifreq))) |
| 2673 | ret = -EFAULT; |
| 2674 | } |
| 2675 | return ret; |
| 2676 | |
| 2677 | case SIOCETHTOOL: |
| 2678 | dev_load(ifr.ifr_name); |
| 2679 | rtnl_lock(); |
| 2680 | ret = dev_ethtool(&ifr); |
| 2681 | rtnl_unlock(); |
| 2682 | if (!ret) { |
| 2683 | if (colon) |
| 2684 | *colon = ':'; |
| 2685 | if (copy_to_user(arg, &ifr, |
| 2686 | sizeof(struct ifreq))) |
| 2687 | ret = -EFAULT; |
| 2688 | } |
| 2689 | return ret; |
| 2690 | |
| 2691 | /* |
| 2692 | * These ioctl calls: |
| 2693 | * - require superuser power. |
| 2694 | * - require strict serialization. |
| 2695 | * - return a value |
| 2696 | */ |
| 2697 | case SIOCGMIIPHY: |
| 2698 | case SIOCGMIIREG: |
| 2699 | case SIOCSIFNAME: |
| 2700 | if (!capable(CAP_NET_ADMIN)) |
| 2701 | return -EPERM; |
| 2702 | dev_load(ifr.ifr_name); |
| 2703 | rtnl_lock(); |
| 2704 | ret = dev_ifsioc(&ifr, cmd); |
| 2705 | rtnl_unlock(); |
| 2706 | if (!ret) { |
| 2707 | if (colon) |
| 2708 | *colon = ':'; |
| 2709 | if (copy_to_user(arg, &ifr, |
| 2710 | sizeof(struct ifreq))) |
| 2711 | ret = -EFAULT; |
| 2712 | } |
| 2713 | return ret; |
| 2714 | |
| 2715 | /* |
| 2716 | * These ioctl calls: |
| 2717 | * - require superuser power. |
| 2718 | * - require strict serialization. |
| 2719 | * - do not return a value |
| 2720 | */ |
| 2721 | case SIOCSIFFLAGS: |
| 2722 | case SIOCSIFMETRIC: |
| 2723 | case SIOCSIFMTU: |
| 2724 | case SIOCSIFMAP: |
| 2725 | case SIOCSIFHWADDR: |
| 2726 | case SIOCSIFSLAVE: |
| 2727 | case SIOCADDMULTI: |
| 2728 | case SIOCDELMULTI: |
| 2729 | case SIOCSIFHWBROADCAST: |
| 2730 | case SIOCSIFTXQLEN: |
| 2731 | case SIOCSMIIREG: |
| 2732 | case SIOCBONDENSLAVE: |
| 2733 | case SIOCBONDRELEASE: |
| 2734 | case SIOCBONDSETHWADDR: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2735 | case SIOCBONDCHANGEACTIVE: |
| 2736 | case SIOCBRADDIF: |
| 2737 | case SIOCBRDELIF: |
| 2738 | if (!capable(CAP_NET_ADMIN)) |
| 2739 | return -EPERM; |
Thomas Graf | cabcac0 | 2006-01-24 12:46:33 -0800 | [diff] [blame] | 2740 | /* fall through */ |
| 2741 | case SIOCBONDSLAVEINFOQUERY: |
| 2742 | case SIOCBONDINFOQUERY: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2743 | dev_load(ifr.ifr_name); |
| 2744 | rtnl_lock(); |
| 2745 | ret = dev_ifsioc(&ifr, cmd); |
| 2746 | rtnl_unlock(); |
| 2747 | return ret; |
| 2748 | |
| 2749 | case SIOCGIFMEM: |
| 2750 | /* Get the per device memory space. We can add this but |
| 2751 | * currently do not support it */ |
| 2752 | case SIOCSIFMEM: |
| 2753 | /* Set the per device memory buffer space. |
| 2754 | * Not applicable in our case */ |
| 2755 | case SIOCSIFLINK: |
| 2756 | return -EINVAL; |
| 2757 | |
| 2758 | /* |
| 2759 | * Unknown or private ioctl. |
| 2760 | */ |
| 2761 | default: |
| 2762 | if (cmd == SIOCWANDEV || |
| 2763 | (cmd >= SIOCDEVPRIVATE && |
| 2764 | cmd <= SIOCDEVPRIVATE + 15)) { |
| 2765 | dev_load(ifr.ifr_name); |
| 2766 | rtnl_lock(); |
| 2767 | ret = dev_ifsioc(&ifr, cmd); |
| 2768 | rtnl_unlock(); |
| 2769 | if (!ret && copy_to_user(arg, &ifr, |
| 2770 | sizeof(struct ifreq))) |
| 2771 | ret = -EFAULT; |
| 2772 | return ret; |
| 2773 | } |
Adrian Bunk | d86b5e0 | 2006-01-21 00:46:55 +0100 | [diff] [blame] | 2774 | #ifdef CONFIG_WIRELESS_EXT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2775 | /* Take care of Wireless Extensions */ |
| 2776 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) { |
| 2777 | /* If command is `set a parameter', or |
| 2778 | * `get the encoding parameters', check if |
| 2779 | * the user has the right to do it */ |
Jean Tourrilhes | a417016 | 2006-04-04 15:53:43 -0700 | [diff] [blame] | 2780 | if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE |
| 2781 | || cmd == SIOCGIWENCODEEXT) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2782 | if (!capable(CAP_NET_ADMIN)) |
| 2783 | return -EPERM; |
| 2784 | } |
| 2785 | dev_load(ifr.ifr_name); |
| 2786 | rtnl_lock(); |
| 2787 | /* Follow me in net/core/wireless.c */ |
| 2788 | ret = wireless_process_ioctl(&ifr, cmd); |
| 2789 | rtnl_unlock(); |
| 2790 | if (IW_IS_GET(cmd) && |
| 2791 | copy_to_user(arg, &ifr, |
| 2792 | sizeof(struct ifreq))) |
| 2793 | ret = -EFAULT; |
| 2794 | return ret; |
| 2795 | } |
Adrian Bunk | d86b5e0 | 2006-01-21 00:46:55 +0100 | [diff] [blame] | 2796 | #endif /* CONFIG_WIRELESS_EXT */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2797 | return -EINVAL; |
| 2798 | } |
| 2799 | } |
| 2800 | |
| 2801 | |
| 2802 | /** |
| 2803 | * dev_new_index - allocate an ifindex |
| 2804 | * |
| 2805 | * Returns a suitable unique value for a new device interface |
| 2806 | * number. The caller must hold the rtnl semaphore or the |
| 2807 | * dev_base_lock to be sure it remains unique. |
| 2808 | */ |
| 2809 | static int dev_new_index(void) |
| 2810 | { |
| 2811 | static int ifindex; |
| 2812 | for (;;) { |
| 2813 | if (++ifindex <= 0) |
| 2814 | ifindex = 1; |
| 2815 | if (!__dev_get_by_index(ifindex)) |
| 2816 | return ifindex; |
| 2817 | } |
| 2818 | } |
| 2819 | |
| 2820 | static int dev_boot_phase = 1; |
| 2821 | |
| 2822 | /* Delayed registration/unregisteration */ |
| 2823 | static DEFINE_SPINLOCK(net_todo_list_lock); |
| 2824 | static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list); |
| 2825 | |
| 2826 | static inline void net_set_todo(struct net_device *dev) |
| 2827 | { |
| 2828 | spin_lock(&net_todo_list_lock); |
| 2829 | list_add_tail(&dev->todo_list, &net_todo_list); |
| 2830 | spin_unlock(&net_todo_list_lock); |
| 2831 | } |
| 2832 | |
| 2833 | /** |
| 2834 | * register_netdevice - register a network device |
| 2835 | * @dev: device to register |
| 2836 | * |
| 2837 | * Take a completed network device structure and add it to the kernel |
| 2838 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier |
| 2839 | * chain. 0 is returned on success. A negative errno code is returned |
| 2840 | * on a failure to set up the device, or if the name is a duplicate. |
| 2841 | * |
| 2842 | * Callers must hold the rtnl semaphore. You may want |
| 2843 | * register_netdev() instead of this. |
| 2844 | * |
| 2845 | * BUGS: |
| 2846 | * The locking appears insufficient to guarantee two parallel registers |
| 2847 | * will not get the same name. |
| 2848 | */ |
| 2849 | |
| 2850 | int register_netdevice(struct net_device *dev) |
| 2851 | { |
| 2852 | struct hlist_head *head; |
| 2853 | struct hlist_node *p; |
| 2854 | int ret; |
| 2855 | |
| 2856 | BUG_ON(dev_boot_phase); |
| 2857 | ASSERT_RTNL(); |
| 2858 | |
Stephen Hemminger | b17a7c1 | 2006-05-10 13:21:17 -0700 | [diff] [blame] | 2859 | might_sleep(); |
| 2860 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2861 | /* When net_device's are persistent, this will be fatal. */ |
| 2862 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); |
| 2863 | |
| 2864 | spin_lock_init(&dev->queue_lock); |
Herbert Xu | 932ff27 | 2006-06-09 12:20:56 -0700 | [diff] [blame] | 2865 | spin_lock_init(&dev->_xmit_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2866 | dev->xmit_lock_owner = -1; |
| 2867 | #ifdef CONFIG_NET_CLS_ACT |
| 2868 | spin_lock_init(&dev->ingress_lock); |
| 2869 | #endif |
| 2870 | |
| 2871 | ret = alloc_divert_blk(dev); |
| 2872 | if (ret) |
| 2873 | goto out; |
| 2874 | |
| 2875 | dev->iflink = -1; |
| 2876 | |
| 2877 | /* Init, if this function is available */ |
| 2878 | if (dev->init) { |
| 2879 | ret = dev->init(dev); |
| 2880 | if (ret) { |
| 2881 | if (ret > 0) |
| 2882 | ret = -EIO; |
| 2883 | goto out_err; |
| 2884 | } |
| 2885 | } |
| 2886 | |
| 2887 | if (!dev_valid_name(dev->name)) { |
| 2888 | ret = -EINVAL; |
| 2889 | goto out_err; |
| 2890 | } |
| 2891 | |
| 2892 | dev->ifindex = dev_new_index(); |
| 2893 | if (dev->iflink == -1) |
| 2894 | dev->iflink = dev->ifindex; |
| 2895 | |
| 2896 | /* Check for existence of name */ |
| 2897 | head = dev_name_hash(dev->name); |
| 2898 | hlist_for_each(p, head) { |
| 2899 | struct net_device *d |
| 2900 | = hlist_entry(p, struct net_device, name_hlist); |
| 2901 | if (!strncmp(d->name, dev->name, IFNAMSIZ)) { |
| 2902 | ret = -EEXIST; |
| 2903 | goto out_err; |
| 2904 | } |
| 2905 | } |
| 2906 | |
| 2907 | /* Fix illegal SG+CSUM combinations. */ |
| 2908 | if ((dev->features & NETIF_F_SG) && |
Herbert Xu | 8648b30 | 2006-06-17 22:06:05 -0700 | [diff] [blame] | 2909 | !(dev->features & NETIF_F_ALL_CSUM)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2910 | printk("%s: Dropping NETIF_F_SG since no checksum feature.\n", |
| 2911 | dev->name); |
| 2912 | dev->features &= ~NETIF_F_SG; |
| 2913 | } |
| 2914 | |
| 2915 | /* TSO requires that SG is present as well. */ |
| 2916 | if ((dev->features & NETIF_F_TSO) && |
| 2917 | !(dev->features & NETIF_F_SG)) { |
| 2918 | printk("%s: Dropping NETIF_F_TSO since no SG feature.\n", |
| 2919 | dev->name); |
| 2920 | dev->features &= ~NETIF_F_TSO; |
| 2921 | } |
Ananda Raju | e89e9cf | 2005-10-18 15:46:41 -0700 | [diff] [blame] | 2922 | if (dev->features & NETIF_F_UFO) { |
| 2923 | if (!(dev->features & NETIF_F_HW_CSUM)) { |
| 2924 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " |
| 2925 | "NETIF_F_HW_CSUM feature.\n", |
| 2926 | dev->name); |
| 2927 | dev->features &= ~NETIF_F_UFO; |
| 2928 | } |
| 2929 | if (!(dev->features & NETIF_F_SG)) { |
| 2930 | printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no " |
| 2931 | "NETIF_F_SG feature.\n", |
| 2932 | dev->name); |
| 2933 | dev->features &= ~NETIF_F_UFO; |
| 2934 | } |
| 2935 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2936 | |
| 2937 | /* |
| 2938 | * nil rebuild_header routine, |
| 2939 | * that should be never called and used as just bug trap. |
| 2940 | */ |
| 2941 | |
| 2942 | if (!dev->rebuild_header) |
| 2943 | dev->rebuild_header = default_rebuild_header; |
| 2944 | |
Stephen Hemminger | b17a7c1 | 2006-05-10 13:21:17 -0700 | [diff] [blame] | 2945 | ret = netdev_register_sysfs(dev); |
| 2946 | if (ret) |
| 2947 | goto out_err; |
| 2948 | dev->reg_state = NETREG_REGISTERED; |
| 2949 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2950 | /* |
| 2951 | * Default initial state at registry is that the |
| 2952 | * device is present. |
| 2953 | */ |
| 2954 | |
| 2955 | set_bit(__LINK_STATE_PRESENT, &dev->state); |
| 2956 | |
| 2957 | dev->next = NULL; |
| 2958 | dev_init_scheduler(dev); |
| 2959 | write_lock_bh(&dev_base_lock); |
| 2960 | *dev_tail = dev; |
| 2961 | dev_tail = &dev->next; |
| 2962 | hlist_add_head(&dev->name_hlist, head); |
| 2963 | hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex)); |
| 2964 | dev_hold(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2965 | write_unlock_bh(&dev_base_lock); |
| 2966 | |
| 2967 | /* Notify protocols, that a new device appeared. */ |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 2968 | raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2969 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2970 | ret = 0; |
| 2971 | |
| 2972 | out: |
| 2973 | return ret; |
| 2974 | out_err: |
| 2975 | free_divert_blk(dev); |
| 2976 | goto out; |
| 2977 | } |
| 2978 | |
| 2979 | /** |
| 2980 | * register_netdev - register a network device |
| 2981 | * @dev: device to register |
| 2982 | * |
| 2983 | * Take a completed network device structure and add it to the kernel |
| 2984 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier |
| 2985 | * chain. 0 is returned on success. A negative errno code is returned |
| 2986 | * on a failure to set up the device, or if the name is a duplicate. |
| 2987 | * |
| 2988 | * This is a wrapper around register_netdev that takes the rtnl semaphore |
| 2989 | * and expands the device name if you passed a format string to |
| 2990 | * alloc_netdev. |
| 2991 | */ |
| 2992 | int register_netdev(struct net_device *dev) |
| 2993 | { |
| 2994 | int err; |
| 2995 | |
| 2996 | rtnl_lock(); |
| 2997 | |
| 2998 | /* |
| 2999 | * If the name is a format string the caller wants us to do a |
| 3000 | * name allocation. |
| 3001 | */ |
| 3002 | if (strchr(dev->name, '%')) { |
| 3003 | err = dev_alloc_name(dev, dev->name); |
| 3004 | if (err < 0) |
| 3005 | goto out; |
| 3006 | } |
| 3007 | |
| 3008 | /* |
| 3009 | * Back compatibility hook. Kill this one in 2.5 |
| 3010 | */ |
| 3011 | if (dev->name[0] == 0 || dev->name[0] == ' ') { |
| 3012 | err = dev_alloc_name(dev, "eth%d"); |
| 3013 | if (err < 0) |
| 3014 | goto out; |
| 3015 | } |
| 3016 | |
| 3017 | err = register_netdevice(dev); |
| 3018 | out: |
| 3019 | rtnl_unlock(); |
| 3020 | return err; |
| 3021 | } |
| 3022 | EXPORT_SYMBOL(register_netdev); |
| 3023 | |
| 3024 | /* |
| 3025 | * netdev_wait_allrefs - wait until all references are gone. |
| 3026 | * |
| 3027 | * This is called when unregistering network devices. |
| 3028 | * |
| 3029 | * Any protocol or device that holds a reference should register |
| 3030 | * for netdevice notification, and cleanup and put back the |
| 3031 | * reference if they receive an UNREGISTER event. |
| 3032 | * We can get stuck here if buggy protocols don't correctly |
| 3033 | * call dev_put. |
| 3034 | */ |
| 3035 | static void netdev_wait_allrefs(struct net_device *dev) |
| 3036 | { |
| 3037 | unsigned long rebroadcast_time, warning_time; |
| 3038 | |
| 3039 | rebroadcast_time = warning_time = jiffies; |
| 3040 | while (atomic_read(&dev->refcnt) != 0) { |
| 3041 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
Stephen Hemminger | 6756ae4 | 2006-03-20 22:23:58 -0800 | [diff] [blame] | 3042 | rtnl_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3043 | |
| 3044 | /* Rebroadcast unregister notification */ |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 3045 | raw_notifier_call_chain(&netdev_chain, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3046 | NETDEV_UNREGISTER, dev); |
| 3047 | |
| 3048 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
| 3049 | &dev->state)) { |
| 3050 | /* We must not have linkwatch events |
| 3051 | * pending on unregister. If this |
| 3052 | * happens, we simply run the queue |
| 3053 | * unscheduled, resulting in a noop |
| 3054 | * for this device. |
| 3055 | */ |
| 3056 | linkwatch_run_queue(); |
| 3057 | } |
| 3058 | |
Stephen Hemminger | 6756ae4 | 2006-03-20 22:23:58 -0800 | [diff] [blame] | 3059 | __rtnl_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3060 | |
| 3061 | rebroadcast_time = jiffies; |
| 3062 | } |
| 3063 | |
| 3064 | msleep(250); |
| 3065 | |
| 3066 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
| 3067 | printk(KERN_EMERG "unregister_netdevice: " |
| 3068 | "waiting for %s to become free. Usage " |
| 3069 | "count = %d\n", |
| 3070 | dev->name, atomic_read(&dev->refcnt)); |
| 3071 | warning_time = jiffies; |
| 3072 | } |
| 3073 | } |
| 3074 | } |
| 3075 | |
| 3076 | /* The sequence is: |
| 3077 | * |
| 3078 | * rtnl_lock(); |
| 3079 | * ... |
| 3080 | * register_netdevice(x1); |
| 3081 | * register_netdevice(x2); |
| 3082 | * ... |
| 3083 | * unregister_netdevice(y1); |
| 3084 | * unregister_netdevice(y2); |
| 3085 | * ... |
| 3086 | * rtnl_unlock(); |
| 3087 | * free_netdev(y1); |
| 3088 | * free_netdev(y2); |
| 3089 | * |
| 3090 | * We are invoked by rtnl_unlock() after it drops the semaphore. |
| 3091 | * This allows us to deal with problems: |
Stephen Hemminger | b17a7c1 | 2006-05-10 13:21:17 -0700 | [diff] [blame] | 3092 | * 1) We can delete sysfs objects which invoke hotplug |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3093 | * without deadlocking with linkwatch via keventd. |
| 3094 | * 2) Since we run with the RTNL semaphore not held, we can sleep |
| 3095 | * safely in order to wait for the netdev refcnt to drop to zero. |
| 3096 | */ |
Arjan van de Ven | 4a3e2f7 | 2006-03-20 22:33:17 -0800 | [diff] [blame] | 3097 | static DEFINE_MUTEX(net_todo_run_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3098 | void netdev_run_todo(void) |
| 3099 | { |
Oleg Nesterov | 626ab0e | 2006-06-23 02:05:55 -0700 | [diff] [blame] | 3100 | struct list_head list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3101 | |
| 3102 | /* Need to guard against multiple cpu's getting out of order. */ |
Arjan van de Ven | 4a3e2f7 | 2006-03-20 22:33:17 -0800 | [diff] [blame] | 3103 | mutex_lock(&net_todo_run_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3104 | |
| 3105 | /* Not safe to do outside the semaphore. We must not return |
| 3106 | * until all unregister events invoked by the local processor |
| 3107 | * have been completed (either by this todo run, or one on |
| 3108 | * another cpu). |
| 3109 | */ |
| 3110 | if (list_empty(&net_todo_list)) |
| 3111 | goto out; |
| 3112 | |
| 3113 | /* Snapshot list, allow later requests */ |
| 3114 | spin_lock(&net_todo_list_lock); |
Oleg Nesterov | 626ab0e | 2006-06-23 02:05:55 -0700 | [diff] [blame] | 3115 | list_replace_init(&net_todo_list, &list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3116 | spin_unlock(&net_todo_list_lock); |
Oleg Nesterov | 626ab0e | 2006-06-23 02:05:55 -0700 | [diff] [blame] | 3117 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3118 | while (!list_empty(&list)) { |
| 3119 | struct net_device *dev |
| 3120 | = list_entry(list.next, struct net_device, todo_list); |
| 3121 | list_del(&dev->todo_list); |
| 3122 | |
Stephen Hemminger | b17a7c1 | 2006-05-10 13:21:17 -0700 | [diff] [blame] | 3123 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3124 | printk(KERN_ERR "network todo '%s' but state %d\n", |
| 3125 | dev->name, dev->reg_state); |
Stephen Hemminger | b17a7c1 | 2006-05-10 13:21:17 -0700 | [diff] [blame] | 3126 | dump_stack(); |
| 3127 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3128 | } |
Stephen Hemminger | b17a7c1 | 2006-05-10 13:21:17 -0700 | [diff] [blame] | 3129 | |
| 3130 | netdev_unregister_sysfs(dev); |
| 3131 | dev->reg_state = NETREG_UNREGISTERED; |
| 3132 | |
| 3133 | netdev_wait_allrefs(dev); |
| 3134 | |
| 3135 | /* paranoia */ |
| 3136 | BUG_ON(atomic_read(&dev->refcnt)); |
| 3137 | BUG_TRAP(!dev->ip_ptr); |
| 3138 | BUG_TRAP(!dev->ip6_ptr); |
| 3139 | BUG_TRAP(!dev->dn_ptr); |
| 3140 | |
| 3141 | /* It must be the very last action, |
| 3142 | * after this 'dev' may point to freed up memory. |
| 3143 | */ |
| 3144 | if (dev->destructor) |
| 3145 | dev->destructor(dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3146 | } |
| 3147 | |
| 3148 | out: |
Arjan van de Ven | 4a3e2f7 | 2006-03-20 22:33:17 -0800 | [diff] [blame] | 3149 | mutex_unlock(&net_todo_run_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3150 | } |
| 3151 | |
| 3152 | /** |
| 3153 | * alloc_netdev - allocate network device |
| 3154 | * @sizeof_priv: size of private data to allocate space for |
| 3155 | * @name: device name format string |
| 3156 | * @setup: callback to initialize device |
| 3157 | * |
| 3158 | * Allocates a struct net_device with private data area for driver use |
| 3159 | * and performs basic initialization. |
| 3160 | */ |
| 3161 | struct net_device *alloc_netdev(int sizeof_priv, const char *name, |
| 3162 | void (*setup)(struct net_device *)) |
| 3163 | { |
| 3164 | void *p; |
| 3165 | struct net_device *dev; |
| 3166 | int alloc_size; |
| 3167 | |
| 3168 | /* ensure 32-byte alignment of both the device and private area */ |
| 3169 | alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST; |
| 3170 | alloc_size += sizeof_priv + NETDEV_ALIGN_CONST; |
| 3171 | |
Paolo 'Blaisorblade' Giarrusso | 31380de | 2006-04-06 22:38:28 -0700 | [diff] [blame] | 3172 | p = kzalloc(alloc_size, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3173 | if (!p) { |
| 3174 | printk(KERN_ERR "alloc_dev: Unable to allocate device.\n"); |
| 3175 | return NULL; |
| 3176 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3177 | |
| 3178 | dev = (struct net_device *) |
| 3179 | (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST); |
| 3180 | dev->padded = (char *)dev - (char *)p; |
| 3181 | |
| 3182 | if (sizeof_priv) |
| 3183 | dev->priv = netdev_priv(dev); |
| 3184 | |
| 3185 | setup(dev); |
| 3186 | strcpy(dev->name, name); |
| 3187 | return dev; |
| 3188 | } |
| 3189 | EXPORT_SYMBOL(alloc_netdev); |
| 3190 | |
| 3191 | /** |
| 3192 | * free_netdev - free network device |
| 3193 | * @dev: device |
| 3194 | * |
| 3195 | * This function does the last stage of destroying an allocated device |
| 3196 | * interface. The reference to the device object is released. |
| 3197 | * If this is the last reference then it will be freed. |
| 3198 | */ |
| 3199 | void free_netdev(struct net_device *dev) |
| 3200 | { |
| 3201 | #ifdef CONFIG_SYSFS |
Stephen Hemminger | 3041a06 | 2006-05-26 13:25:24 -0700 | [diff] [blame] | 3202 | /* Compatibility with error handling in drivers */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3203 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
| 3204 | kfree((char *)dev - dev->padded); |
| 3205 | return; |
| 3206 | } |
| 3207 | |
| 3208 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); |
| 3209 | dev->reg_state = NETREG_RELEASED; |
| 3210 | |
| 3211 | /* will free via class release */ |
| 3212 | class_device_put(&dev->class_dev); |
| 3213 | #else |
| 3214 | kfree((char *)dev - dev->padded); |
| 3215 | #endif |
| 3216 | } |
| 3217 | |
| 3218 | /* Synchronize with packet receive processing. */ |
| 3219 | void synchronize_net(void) |
| 3220 | { |
| 3221 | might_sleep(); |
Paul E. McKenney | fbd568a3e | 2005-05-01 08:59:04 -0700 | [diff] [blame] | 3222 | synchronize_rcu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3223 | } |
| 3224 | |
| 3225 | /** |
| 3226 | * unregister_netdevice - remove device from the kernel |
| 3227 | * @dev: device |
| 3228 | * |
| 3229 | * This function shuts down a device interface and removes it |
| 3230 | * from the kernel tables. On success 0 is returned, on a failure |
| 3231 | * a negative errno code is returned. |
| 3232 | * |
| 3233 | * Callers must hold the rtnl semaphore. You may want |
| 3234 | * unregister_netdev() instead of this. |
| 3235 | */ |
| 3236 | |
| 3237 | int unregister_netdevice(struct net_device *dev) |
| 3238 | { |
| 3239 | struct net_device *d, **dp; |
| 3240 | |
| 3241 | BUG_ON(dev_boot_phase); |
| 3242 | ASSERT_RTNL(); |
| 3243 | |
| 3244 | /* Some devices call without registering for initialization unwind. */ |
| 3245 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
| 3246 | printk(KERN_DEBUG "unregister_netdevice: device %s/%p never " |
| 3247 | "was registered\n", dev->name, dev); |
| 3248 | return -ENODEV; |
| 3249 | } |
| 3250 | |
| 3251 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
| 3252 | |
| 3253 | /* If device is running, close it first. */ |
| 3254 | if (dev->flags & IFF_UP) |
| 3255 | dev_close(dev); |
| 3256 | |
| 3257 | /* And unlink it from device chain. */ |
| 3258 | for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) { |
| 3259 | if (d == dev) { |
| 3260 | write_lock_bh(&dev_base_lock); |
| 3261 | hlist_del(&dev->name_hlist); |
| 3262 | hlist_del(&dev->index_hlist); |
| 3263 | if (dev_tail == &dev->next) |
| 3264 | dev_tail = dp; |
| 3265 | *dp = d->next; |
| 3266 | write_unlock_bh(&dev_base_lock); |
| 3267 | break; |
| 3268 | } |
| 3269 | } |
| 3270 | if (!d) { |
| 3271 | printk(KERN_ERR "unregister net_device: '%s' not found\n", |
| 3272 | dev->name); |
| 3273 | return -ENODEV; |
| 3274 | } |
| 3275 | |
| 3276 | dev->reg_state = NETREG_UNREGISTERING; |
| 3277 | |
| 3278 | synchronize_net(); |
| 3279 | |
| 3280 | /* Shutdown queueing discipline. */ |
| 3281 | dev_shutdown(dev); |
| 3282 | |
| 3283 | |
| 3284 | /* Notify protocols, that we are about to destroy |
| 3285 | this device. They should clean all the things. |
| 3286 | */ |
Alan Stern | f07d5b9 | 2006-05-09 15:23:03 -0700 | [diff] [blame] | 3287 | raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3288 | |
| 3289 | /* |
| 3290 | * Flush the multicast chain |
| 3291 | */ |
| 3292 | dev_mc_discard(dev); |
| 3293 | |
| 3294 | if (dev->uninit) |
| 3295 | dev->uninit(dev); |
| 3296 | |
| 3297 | /* Notifier chain MUST detach us from master device. */ |
| 3298 | BUG_TRAP(!dev->master); |
| 3299 | |
| 3300 | free_divert_blk(dev); |
| 3301 | |
| 3302 | /* Finish processing unregister after unlock */ |
| 3303 | net_set_todo(dev); |
| 3304 | |
| 3305 | synchronize_net(); |
| 3306 | |
| 3307 | dev_put(dev); |
| 3308 | return 0; |
| 3309 | } |
| 3310 | |
| 3311 | /** |
| 3312 | * unregister_netdev - remove device from the kernel |
| 3313 | * @dev: device |
| 3314 | * |
| 3315 | * This function shuts down a device interface and removes it |
| 3316 | * from the kernel tables. On success 0 is returned, on a failure |
| 3317 | * a negative errno code is returned. |
| 3318 | * |
| 3319 | * This is just a wrapper for unregister_netdevice that takes |
| 3320 | * the rtnl semaphore. In general you want to use this and not |
| 3321 | * unregister_netdevice. |
| 3322 | */ |
| 3323 | void unregister_netdev(struct net_device *dev) |
| 3324 | { |
| 3325 | rtnl_lock(); |
| 3326 | unregister_netdevice(dev); |
| 3327 | rtnl_unlock(); |
| 3328 | } |
| 3329 | |
| 3330 | EXPORT_SYMBOL(unregister_netdev); |
| 3331 | |
| 3332 | #ifdef CONFIG_HOTPLUG_CPU |
| 3333 | static int dev_cpu_callback(struct notifier_block *nfb, |
| 3334 | unsigned long action, |
| 3335 | void *ocpu) |
| 3336 | { |
| 3337 | struct sk_buff **list_skb; |
| 3338 | struct net_device **list_net; |
| 3339 | struct sk_buff *skb; |
| 3340 | unsigned int cpu, oldcpu = (unsigned long)ocpu; |
| 3341 | struct softnet_data *sd, *oldsd; |
| 3342 | |
| 3343 | if (action != CPU_DEAD) |
| 3344 | return NOTIFY_OK; |
| 3345 | |
| 3346 | local_irq_disable(); |
| 3347 | cpu = smp_processor_id(); |
| 3348 | sd = &per_cpu(softnet_data, cpu); |
| 3349 | oldsd = &per_cpu(softnet_data, oldcpu); |
| 3350 | |
| 3351 | /* Find end of our completion_queue. */ |
| 3352 | list_skb = &sd->completion_queue; |
| 3353 | while (*list_skb) |
| 3354 | list_skb = &(*list_skb)->next; |
| 3355 | /* Append completion queue from offline CPU. */ |
| 3356 | *list_skb = oldsd->completion_queue; |
| 3357 | oldsd->completion_queue = NULL; |
| 3358 | |
| 3359 | /* Find end of our output_queue. */ |
| 3360 | list_net = &sd->output_queue; |
| 3361 | while (*list_net) |
| 3362 | list_net = &(*list_net)->next_sched; |
| 3363 | /* Append output queue from offline CPU. */ |
| 3364 | *list_net = oldsd->output_queue; |
| 3365 | oldsd->output_queue = NULL; |
| 3366 | |
| 3367 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
| 3368 | local_irq_enable(); |
| 3369 | |
| 3370 | /* Process offline CPU's input_pkt_queue */ |
| 3371 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) |
| 3372 | netif_rx(skb); |
| 3373 | |
| 3374 | return NOTIFY_OK; |
| 3375 | } |
| 3376 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 3377 | |
Chris Leech | db21733 | 2006-06-17 21:24:58 -0700 | [diff] [blame] | 3378 | #ifdef CONFIG_NET_DMA |
| 3379 | /** |
| 3380 | * net_dma_rebalance - |
| 3381 | * This is called when the number of channels allocated to the net_dma_client |
| 3382 | * changes. The net_dma_client tries to have one DMA channel per CPU. |
| 3383 | */ |
| 3384 | static void net_dma_rebalance(void) |
| 3385 | { |
| 3386 | unsigned int cpu, i, n; |
| 3387 | struct dma_chan *chan; |
| 3388 | |
| 3389 | lock_cpu_hotplug(); |
| 3390 | |
| 3391 | if (net_dma_count == 0) { |
| 3392 | for_each_online_cpu(cpu) |
| 3393 | rcu_assign_pointer(per_cpu(softnet_data.net_dma, cpu), NULL); |
| 3394 | unlock_cpu_hotplug(); |
| 3395 | return; |
| 3396 | } |
| 3397 | |
| 3398 | i = 0; |
| 3399 | cpu = first_cpu(cpu_online_map); |
| 3400 | |
| 3401 | rcu_read_lock(); |
| 3402 | list_for_each_entry(chan, &net_dma_client->channels, client_node) { |
| 3403 | n = ((num_online_cpus() / net_dma_count) |
| 3404 | + (i < (num_online_cpus() % net_dma_count) ? 1 : 0)); |
| 3405 | |
| 3406 | while(n) { |
| 3407 | per_cpu(softnet_data.net_dma, cpu) = chan; |
| 3408 | cpu = next_cpu(cpu, cpu_online_map); |
| 3409 | n--; |
| 3410 | } |
| 3411 | i++; |
| 3412 | } |
| 3413 | rcu_read_unlock(); |
| 3414 | |
| 3415 | unlock_cpu_hotplug(); |
| 3416 | } |
| 3417 | |
| 3418 | /** |
| 3419 | * netdev_dma_event - event callback for the net_dma_client |
| 3420 | * @client: should always be net_dma_client |
Randy Dunlap | f4b8ea7 | 2006-06-22 16:00:11 -0700 | [diff] [blame] | 3421 | * @chan: DMA channel for the event |
| 3422 | * @event: event type |
Chris Leech | db21733 | 2006-06-17 21:24:58 -0700 | [diff] [blame] | 3423 | */ |
| 3424 | static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan, |
| 3425 | enum dma_event event) |
| 3426 | { |
| 3427 | spin_lock(&net_dma_event_lock); |
| 3428 | switch (event) { |
| 3429 | case DMA_RESOURCE_ADDED: |
| 3430 | net_dma_count++; |
| 3431 | net_dma_rebalance(); |
| 3432 | break; |
| 3433 | case DMA_RESOURCE_REMOVED: |
| 3434 | net_dma_count--; |
| 3435 | net_dma_rebalance(); |
| 3436 | break; |
| 3437 | default: |
| 3438 | break; |
| 3439 | } |
| 3440 | spin_unlock(&net_dma_event_lock); |
| 3441 | } |
| 3442 | |
| 3443 | /** |
| 3444 | * netdev_dma_regiser - register the networking subsystem as a DMA client |
| 3445 | */ |
| 3446 | static int __init netdev_dma_register(void) |
| 3447 | { |
| 3448 | spin_lock_init(&net_dma_event_lock); |
| 3449 | net_dma_client = dma_async_client_register(netdev_dma_event); |
| 3450 | if (net_dma_client == NULL) |
| 3451 | return -ENOMEM; |
| 3452 | |
| 3453 | dma_async_client_chan_request(net_dma_client, num_online_cpus()); |
| 3454 | return 0; |
| 3455 | } |
| 3456 | |
| 3457 | #else |
| 3458 | static int __init netdev_dma_register(void) { return -ENODEV; } |
| 3459 | #endif /* CONFIG_NET_DMA */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3460 | |
| 3461 | /* |
| 3462 | * Initialize the DEV module. At boot time this walks the device list and |
| 3463 | * unhooks any devices that fail to initialise (normally hardware not |
| 3464 | * present) and leaves us with a valid list of present and active devices. |
| 3465 | * |
| 3466 | */ |
| 3467 | |
| 3468 | /* |
| 3469 | * This is called single threaded during boot, so no need |
| 3470 | * to take the rtnl semaphore. |
| 3471 | */ |
| 3472 | static int __init net_dev_init(void) |
| 3473 | { |
| 3474 | int i, rc = -ENOMEM; |
| 3475 | |
| 3476 | BUG_ON(!dev_boot_phase); |
| 3477 | |
| 3478 | net_random_init(); |
| 3479 | |
| 3480 | if (dev_proc_init()) |
| 3481 | goto out; |
| 3482 | |
| 3483 | if (netdev_sysfs_init()) |
| 3484 | goto out; |
| 3485 | |
| 3486 | INIT_LIST_HEAD(&ptype_all); |
| 3487 | for (i = 0; i < 16; i++) |
| 3488 | INIT_LIST_HEAD(&ptype_base[i]); |
| 3489 | |
| 3490 | for (i = 0; i < ARRAY_SIZE(dev_name_head); i++) |
| 3491 | INIT_HLIST_HEAD(&dev_name_head[i]); |
| 3492 | |
| 3493 | for (i = 0; i < ARRAY_SIZE(dev_index_head); i++) |
| 3494 | INIT_HLIST_HEAD(&dev_index_head[i]); |
| 3495 | |
| 3496 | /* |
| 3497 | * Initialise the packet receive queues. |
| 3498 | */ |
| 3499 | |
KAMEZAWA Hiroyuki | 6f91204 | 2006-04-10 22:52:50 -0700 | [diff] [blame] | 3500 | for_each_possible_cpu(i) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3501 | struct softnet_data *queue; |
| 3502 | |
| 3503 | queue = &per_cpu(softnet_data, i); |
| 3504 | skb_queue_head_init(&queue->input_pkt_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3505 | queue->completion_queue = NULL; |
| 3506 | INIT_LIST_HEAD(&queue->poll_list); |
| 3507 | set_bit(__LINK_STATE_START, &queue->backlog_dev.state); |
| 3508 | queue->backlog_dev.weight = weight_p; |
| 3509 | queue->backlog_dev.poll = process_backlog; |
| 3510 | atomic_set(&queue->backlog_dev.refcnt, 1); |
| 3511 | } |
| 3512 | |
Chris Leech | db21733 | 2006-06-17 21:24:58 -0700 | [diff] [blame] | 3513 | netdev_dma_register(); |
| 3514 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3515 | dev_boot_phase = 0; |
| 3516 | |
| 3517 | open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL); |
| 3518 | open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL); |
| 3519 | |
| 3520 | hotcpu_notifier(dev_cpu_callback, 0); |
| 3521 | dst_init(); |
| 3522 | dev_mcast_init(); |
| 3523 | rc = 0; |
| 3524 | out: |
| 3525 | return rc; |
| 3526 | } |
| 3527 | |
| 3528 | subsys_initcall(net_dev_init); |
| 3529 | |
| 3530 | EXPORT_SYMBOL(__dev_get_by_index); |
| 3531 | EXPORT_SYMBOL(__dev_get_by_name); |
| 3532 | EXPORT_SYMBOL(__dev_remove_pack); |
Mitch Williams | c2373ee | 2005-11-09 10:34:45 -0800 | [diff] [blame] | 3533 | EXPORT_SYMBOL(dev_valid_name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3534 | EXPORT_SYMBOL(dev_add_pack); |
| 3535 | EXPORT_SYMBOL(dev_alloc_name); |
| 3536 | EXPORT_SYMBOL(dev_close); |
| 3537 | EXPORT_SYMBOL(dev_get_by_flags); |
| 3538 | EXPORT_SYMBOL(dev_get_by_index); |
| 3539 | EXPORT_SYMBOL(dev_get_by_name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3540 | EXPORT_SYMBOL(dev_open); |
| 3541 | EXPORT_SYMBOL(dev_queue_xmit); |
| 3542 | EXPORT_SYMBOL(dev_remove_pack); |
| 3543 | EXPORT_SYMBOL(dev_set_allmulti); |
| 3544 | EXPORT_SYMBOL(dev_set_promiscuity); |
| 3545 | EXPORT_SYMBOL(dev_change_flags); |
| 3546 | EXPORT_SYMBOL(dev_set_mtu); |
| 3547 | EXPORT_SYMBOL(dev_set_mac_address); |
| 3548 | EXPORT_SYMBOL(free_netdev); |
| 3549 | EXPORT_SYMBOL(netdev_boot_setup_check); |
| 3550 | EXPORT_SYMBOL(netdev_set_master); |
| 3551 | EXPORT_SYMBOL(netdev_state_change); |
| 3552 | EXPORT_SYMBOL(netif_receive_skb); |
| 3553 | EXPORT_SYMBOL(netif_rx); |
| 3554 | EXPORT_SYMBOL(register_gifconf); |
| 3555 | EXPORT_SYMBOL(register_netdevice); |
| 3556 | EXPORT_SYMBOL(register_netdevice_notifier); |
| 3557 | EXPORT_SYMBOL(skb_checksum_help); |
| 3558 | EXPORT_SYMBOL(synchronize_net); |
| 3559 | EXPORT_SYMBOL(unregister_netdevice); |
| 3560 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
| 3561 | EXPORT_SYMBOL(net_enable_timestamp); |
| 3562 | EXPORT_SYMBOL(net_disable_timestamp); |
| 3563 | EXPORT_SYMBOL(dev_get_flags); |
| 3564 | |
| 3565 | #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE) |
| 3566 | EXPORT_SYMBOL(br_handle_frame_hook); |
| 3567 | EXPORT_SYMBOL(br_fdb_get_hook); |
| 3568 | EXPORT_SYMBOL(br_fdb_put_hook); |
| 3569 | #endif |
| 3570 | |
| 3571 | #ifdef CONFIG_KMOD |
| 3572 | EXPORT_SYMBOL(dev_load); |
| 3573 | #endif |
| 3574 | |
| 3575 | EXPORT_PER_CPU_SYMBOL(softnet_data); |