| 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
|
| 10 | * Authors: Ross Biro
|
| 11 | * Fred N. van Kempen, <[email protected]>
|
| 12 | * Mark Evans, <[email protected]>
|
| 13 | *
|
| 14 | * Additional Authors:
|
| 15 | * Florian la Roche <[email protected]>
|
| 16 | * Alan Cox <[email protected]>
|
| 17 | * David Hinds <[email protected]>
|
| 18 | * Alexey Kuznetsov <[email protected]>
|
| 19 | * Adam Sulmicki <[email protected]>
|
| 20 | * Pekka Riikonen <[email protected]>
|
| 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>
|
| 78 | #include <linux/config.h>
|
| 79 | #include <linux/cpu.h>
|
| 80 | #include <linux/types.h>
|
| 81 | #include <linux/kernel.h>
|
| 82 | #include <linux/sched.h>
|
| 83 | #include <linux/string.h>
|
| 84 | #include <linux/mm.h>
|
| 85 | #include <linux/socket.h>
|
| 86 | #include <linux/sockios.h>
|
| 87 | #include <linux/errno.h>
|
| 88 | #include <linux/interrupt.h>
|
| 89 | #include <linux/if_ether.h>
|
| 90 | #include <linux/netdevice.h>
|
| 91 | #include <linux/etherdevice.h>
|
| 92 | #include <linux/notifier.h>
|
| 93 | #include <linux/skbuff.h>
|
| 94 | #include <net/sock.h>
|
| 95 | #include <linux/rtnetlink.h>
|
| 96 | #include <linux/proc_fs.h>
|
| 97 | #include <linux/seq_file.h>
|
| 98 | #include <linux/stat.h>
|
| 99 | #include <linux/if_bridge.h>
|
| 100 | #include <linux/divert.h>
|
| 101 | #include <net/dst.h>
|
| 102 | #include <net/pkt_sched.h>
|
| 103 | #include <net/checksum.h>
|
| 104 | #include <linux/highmem.h>
|
| 105 | #include <linux/init.h>
|
| 106 | #include <linux/kmod.h>
|
| 107 | #include <linux/module.h>
|
| 108 | #include <linux/kallsyms.h>
|
| 109 | #include <linux/netpoll.h>
|
| 110 | #include <linux/rcupdate.h>
|
| 111 | #include <linux/delay.h>
|
| 112 | #ifdef CONFIG_NET_RADIO
|
| 113 | #include <linux/wireless.h>
|
| 114 | #include <net/iw_handler.h>
|
| 115 | #endif
|
| 116 | #include <asm/current.h>
|
| 117 |
|
| 118 | |
| 119 | * The list of packet types we will receive (as opposed to discard)
|
| 120 | * and the routines to invoke.
|
| 121 | *
|
| 122 | * Why 16. Because with 16 the only overlap we get on a hash of the
|
| 123 | * low nibble of the protocol value is RARP/SNAP/X.25.
|
| 124 | *
|
| 125 | * NOTE: That is no longer true with the addition of VLAN tags. Not
|
| 126 | * sure which should go first, but I bet it won't make much
|
| 127 | * difference if we are running VLANs. The good news is that
|
| 128 | * this protocol won't be in the list unless compiled in, so
|
| 129 | * the average user (w/out VLANs) will not be adversly affected.
|
| 130 | * --BLG
|
| 131 | *
|
| 132 | * 0800 IP
|
| 133 | * 8100 802.1Q VLAN
|
| 134 | * 0001 802.3
|
| 135 | * 0002 AX.25
|
| 136 | * 0004 802.2
|
| 137 | * 8035 RARP
|
| 138 | * 0005 SNAP
|
| 139 | * 0805 X.25
|
| 140 | * 0806 ARP
|
| 141 | * 8137 IPX
|
| 142 | * 0009 Localtalk
|
| 143 | * 86DD IPv6
|
| 144 |
|
| 145 |
|
| 146 | static DEFINE_SPINLOCK(ptype_lock);
|
| 147 | static struct list_head ptype_base[16];
|
| 148 | static struct list_head ptype_all;
|
| 149 |
|
| 150 | |
| 151 | * The @dev_base list is protected by @dev_base_lock and the rtln
|
| 152 | * semaphore.
|
| 153 | *
|
| 154 | * Pure readers hold dev_base_lock for reading.
|
| 155 | *
|
| 156 | * Writers must hold the rtnl semaphore while they loop through the
|
| 157 | * dev_base list, and hold dev_base_lock for writing when they do the
|
| 158 | * actual updates. This allows pure readers to access the list even
|
| 159 | * while a writer is preparing to update it.
|
| 160 | *
|
| 161 | * To put it another way, dev_base_lock is held for writing only to
|
| 162 | * protect against pure readers; the rtnl semaphore provides the
|
| 163 | * protection against other writers.
|
| 164 | *
|
| 165 | * See, for example usages, register_netdevice() and
|
| 166 | * unregister_netdevice(), which must be called with the rtnl
|
| 167 | * semaphore held.
|
| 168 |
|
| 169 | struct net_device *dev_base;
|
| 170 | static struct net_device **dev_tail = &dev_base;
|
| 171 | DEFINE_RWLOCK(dev_base_lock);
|
| 172 |
|
| 173 | EXPORT_SYMBOL(dev_base);
|
| 174 | EXPORT_SYMBOL(dev_base_lock);
|
| 175 |
|
| 176 | #define NETDEV_HASHBITS 8
|
| 177 | static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
|
| 178 | static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
|
| 179 |
|
| 180 | static inline struct hlist_head *dev_name_hash(const char *name)
|
| 181 | {
|
| 182 | unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
|
| 183 | return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
|
| 184 | }
|
| 185 |
|
| 186 | static inline struct hlist_head *dev_index_hash(int ifindex)
|
| 187 | {
|
| 188 | return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
|
| 189 | }
|
| 190 |
|
| 191 | |
| 192 | * Our notifier list
|
| 193 |
|
| 194 |
|
| 195 | static struct notifier_block *netdev_chain;
|
| 196 |
|
| 197 | |
| 198 | * Device drivers call our routines to queue packets here. We empty the
|
| 199 | * queue in the local softnet handler.
|
| 200 |
|
| 201 | DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
|
| 202 |
|
| 203 | #ifdef CONFIG_SYSFS
|
| 204 | extern int netdev_sysfs_init(void);
|
| 205 | extern int netdev_register_sysfs(struct net_device *);
|
| 206 | extern void netdev_unregister_sysfs(struct net_device *);
|
| 207 | #else
|
| 208 | #define netdev_sysfs_init() (0)
|
| 209 | #define netdev_register_sysfs(dev) (0)
|
| 210 | #define netdev_unregister_sysfs(dev) do { } while(0)
|
| 211 | #endif
|
| 212 |
|
| 213 |
|
| 214 | |
| 215 |
|
| 216 | Protocol management and registration routines
|
| 217 |
|
| 218 | ******************************************************************************
|
| 219 |
|
| 220 | |
| 221 | * For efficiency
|
| 222 |
|
| 223 |
|
| 224 | int netdev_nit;
|
| 225 |
|
| 226 | |
| 227 | * Add a protocol ID to the list. Now that the input handler is
|
| 228 | * smarter we can dispense with all the messy stuff that used to be
|
| 229 | * here.
|
| 230 | *
|
| 231 | * BEWARE!!! Protocol handlers, mangling input packets,
|
| 232 | * MUST BE last in hash buckets and checking protocol handlers
|
| 233 | * MUST start from promiscuous ptype_all chain in net_bh.
|
| 234 | * It is true now, do not change it.
|
| 235 | * Explanation follows: if protocol handler, mangling packet, will
|
| 236 | * be the first on list, it is not able to sense, that packet
|
| 237 | * is cloned and should be copied-on-write, so that it will
|
| 238 | * change it and subsequent readers will get broken packet.
|
| 239 | * --ANK (980803)
|
| 240 |
|
| 241 |
|
| 242 | |
| 243 | * dev_add_pack - add packet handler
|
| 244 | * @pt: packet type declaration
|
| 245 | *
|
| 246 | * Add a protocol handler to the networking stack. The passed &packet_type
|
| 247 | * is linked into kernel lists and may not be freed until it has been
|
| 248 | * removed from the kernel lists.
|
| 249 | *
|
| 250 | * This call does not sleep therefore it can not
|
| 251 | * guarantee all CPU's that are in middle of receiving packets
|
| 252 | * will see the new packet type (until the next received packet).
|
| 253 |
|
| 254 |
|
| 255 | void dev_add_pack(struct packet_type *pt)
|
| 256 | {
|
| 257 | int hash;
|
| 258 |
|
| 259 | spin_lock_bh(&ptype_lock);
|
| 260 | if (pt->type == htons(ETH_P_ALL)) {
|
| 261 | netdev_nit++;
|
| 262 | list_add_rcu(&pt->list, &ptype_all);
|
| 263 | } else {
|
| 264 | hash = ntohs(pt->type) & 15;
|
| 265 | list_add_rcu(&pt->list, &ptype_base[hash]);
|
| 266 | }
|
| 267 | spin_unlock_bh(&ptype_lock);
|
| 268 | }
|
| 269 |
|
| 270 | |
| 271 | * __dev_remove_pack - remove packet handler
|
| 272 | * @pt: packet type declaration
|
| 273 | *
|
| 274 | * Remove a protocol handler that was previously added to the kernel
|
| 275 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed
|
| 276 | * from the kernel lists and can be freed or reused once this function
|
| 277 | * returns.
|
| 278 | *
|
| 279 | * The packet type might still be in use by receivers
|
| 280 | * and must not be freed until after all the CPU's have gone
|
| 281 | * through a quiescent state.
|
| 282 |
|
| 283 | void __dev_remove_pack(struct packet_type *pt)
|
| 284 | {
|
| 285 | struct list_head *head;
|
| 286 | struct packet_type *pt1;
|
| 287 |
|
| 288 | spin_lock_bh(&ptype_lock);
|
| 289 |
|
| 290 | if (pt->type == htons(ETH_P_ALL)) {
|
| 291 | netdev_nit--;
|
| 292 | head = &ptype_all;
|
| 293 | } else
|
| 294 | head = &ptype_base[ntohs(pt->type) & 15];
|
| 295 |
|
| 296 | list_for_each_entry(pt1, head, list) {
|
| 297 | if (pt == pt1) {
|
| 298 | list_del_rcu(&pt->list);
|
| 299 | goto out;
|
| 300 | }
|
| 301 | }
|
| 302 |
|
| 303 | printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
|
| 304 | out:
|
| 305 | spin_unlock_bh(&ptype_lock);
|
| 306 | }
|
| 307 | |
| 308 | * dev_remove_pack - remove packet handler
|
| 309 | * @pt: packet type declaration
|
| 310 | *
|
| 311 | * Remove a protocol handler that was previously added to the kernel
|
| 312 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed
|
| 313 | * from the kernel lists and can be freed or reused once this function
|
| 314 | * returns.
|
| 315 | *
|
| 316 | * This call sleeps to guarantee that no CPU is looking at the packet
|
| 317 | * type after return.
|
| 318 |
|
| 319 | void dev_remove_pack(struct packet_type *pt)
|
| 320 | {
|
| 321 | __dev_remove_pack(pt);
|
| 322 |
|
| 323 | synchronize_net();
|
| 324 | }
|
| 325 |
|
| 326 | |
| 327 |
|
| 328 | Device Boot-time Settings Routines
|
| 329 |
|
| 330 | ******************************************************************************
|
| 331 |
|
| 332 |
|
| 333 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
|
| 334 |
|
| 335 | |
| 336 | * netdev_boot_setup_add - add new setup entry
|
| 337 | * @name: name of the device
|
| 338 | * @map: configured settings for the device
|
| 339 | *
|
| 340 | * Adds new setup entry to the dev_boot_setup list. The function
|
| 341 | * returns 0 on error and 1 on success. This is a generic routine to
|
| 342 | * all netdevices.
|
| 343 |
|
| 344 | static int netdev_boot_setup_add(char *name, struct ifmap *map)
|
| 345 | {
|
| 346 | struct netdev_boot_setup *s;
|
| 347 | int i;
|
| 348 |
|
| 349 | s = dev_boot_setup;
|
| 350 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
|
| 351 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
|
| 352 | memset(s[i].name, 0, sizeof(s[i].name));
|
| 353 | strcpy(s[i].name, name);
|
| 354 | memcpy(&s[i].map, map, sizeof(s[i].map));
|
| 355 | break;
|
| 356 | }
|
| 357 | }
|
| 358 |
|
| 359 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
|
| 360 | }
|
| 361 |
|
| 362 | |
| 363 | * netdev_boot_setup_check - check boot time settings
|
| 364 | * @dev: the netdevice
|
| 365 | *
|
| 366 | * Check boot time settings for the device.
|
| 367 | * The found settings are set for the device to be used
|
| 368 | * later in the device probing.
|
| 369 | * Returns 0 if no settings found, 1 if they are.
|
| 370 |
|
| 371 | int netdev_boot_setup_check(struct net_device *dev)
|
| 372 | {
|
| 373 | struct netdev_boot_setup *s = dev_boot_setup;
|
| 374 | int i;
|
| 375 |
|
| 376 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
|
| 377 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
|
| 378 | !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
|
| 379 | dev->irq = s[i].map.irq;
|
| 380 | dev->base_addr = s[i].map.base_addr;
|
| 381 | dev->mem_start = s[i].map.mem_start;
|
| 382 | dev->mem_end = s[i].map.mem_end;
|
| 383 | return 1;
|
| 384 | }
|
| 385 | }
|
| 386 | return 0;
|
| 387 | }
|
| 388 |
|
| 389 |
|
| 390 | |
| 391 | * netdev_boot_base - get address from boot time settings
|
| 392 | * @prefix: prefix for network device
|
| 393 | * @unit: id for network device
|
| 394 | *
|
| 395 | * Check boot time settings for the base address of device.
|
| 396 | * The found settings are set for the device to be used
|
| 397 | * later in the device probing.
|
| 398 | * Returns 0 if no settings found.
|
| 399 |
|
| 400 | unsigned long netdev_boot_base(const char *prefix, int unit)
|
| 401 | {
|
| 402 | const struct netdev_boot_setup *s = dev_boot_setup;
|
| 403 | char name[IFNAMSIZ];
|
| 404 | int i;
|
| 405 |
|
| 406 | sprintf(name, "%s%d", prefix, unit);
|
| 407 |
|
| 408 | |
| 409 | * If device already registered then return base of 1
|
| 410 | * to indicate not to probe for this interface
|
| 411 |
|
| 412 | if (__dev_get_by_name(name))
|
| 413 | return 1;
|
| 414 |
|
| 415 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
|
| 416 | if (!strcmp(name, s[i].name))
|
| 417 | return s[i].map.base_addr;
|
| 418 | return 0;
|
| 419 | }
|
| 420 |
|
| 421 | |
| 422 | * Saves at boot time configured settings for any netdevice.
|
| 423 |
|
| 424 | int __init netdev_boot_setup(char *str)
|
| 425 | {
|
| 426 | int ints[5];
|
| 427 | struct ifmap map;
|
| 428 |
|
| 429 | str = get_options(str, ARRAY_SIZE(ints), ints);
|
| 430 | if (!str || !*str)
|
| 431 | return 0;
|
| 432 |
|
| 433 |
|
| 434 | memset(&map, 0, sizeof(map));
|
| 435 | if (ints[0] > 0)
|
| 436 | map.irq = ints[1];
|
| 437 | if (ints[0] > 1)
|
| 438 | map.base_addr = ints[2];
|
| 439 | if (ints[0] > 2)
|
| 440 | map.mem_start = ints[3];
|
| 441 | if (ints[0] > 3)
|
| 442 | map.mem_end = ints[4];
|
| 443 |
|
| 444 |
|
| 445 | return netdev_boot_setup_add(str, &map);
|
| 446 | }
|
| 447 |
|
| 448 | __setup("netdev=", netdev_boot_setup);
|
| 449 |
|
| 450 | |
| 451 |
|
| 452 | Device Interface Subroutines
|
| 453 |
|
| 454 | ******************************************************************************
|
| 455 |
|
| 456 | |
| 457 | * __dev_get_by_name - find a device by its name
|
| 458 | * @name: name to find
|
| 459 | *
|
| 460 | * Find an interface by name. Must be called under RTNL semaphore
|
| 461 | * or @dev_base_lock. If the name is found a pointer to the device
|
| 462 | * is returned. If the name is not found then %NULL is returned. The
|
| 463 | * reference counters are not incremented so the caller must be
|
| 464 | * careful with locks.
|
| 465 |
|
| 466 |
|
| 467 | struct net_device *__dev_get_by_name(const char *name)
|
| 468 | {
|
| 469 | struct hlist_node *p;
|
| 470 |
|
| 471 | hlist_for_each(p, dev_name_hash(name)) {
|
| 472 | struct net_device *dev
|
| 473 | = hlist_entry(p, struct net_device, name_hlist);
|
| 474 | if (!strncmp(dev->name, name, IFNAMSIZ))
|
| 475 | return dev;
|
| 476 | }
|
| 477 | return NULL;
|
| 478 | }
|
| 479 |
|
| 480 | |
| 481 | * dev_get_by_name - find a device by its name
|
| 482 | * @name: name to find
|
| 483 | *
|
| 484 | * Find an interface by name. This can be called from any
|
| 485 | * context and does its own locking. The returned handle has
|
| 486 | * the usage count incremented and the caller must use dev_put() to
|
| 487 | * release it when it is no longer needed. %NULL is returned if no
|
| 488 | * matching device is found.
|
| 489 |
|
| 490 |
|
| 491 | struct net_device *dev_get_by_name(const char *name)
|
| 492 | {
|
| 493 | struct net_device *dev;
|
| 494 |
|
| 495 | read_lock(&dev_base_lock);
|
| 496 | dev = __dev_get_by_name(name);
|
| 497 | if (dev)
|
| 498 | dev_hold(dev);
|
| 499 | read_unlock(&dev_base_lock);
|
| 500 | return dev;
|
| 501 | }
|
| 502 |
|
| 503 | |
| 504 | * __dev_get_by_index - find a device by its ifindex
|
| 505 | * @ifindex: index of device
|
| 506 | *
|
| 507 | * Search for an interface by index. Returns %NULL if the device
|
| 508 | * is not found or a pointer to the device. The device has not
|
| 509 | * had its reference counter increased so the caller must be careful
|
| 510 | * about locking. The caller must hold either the RTNL semaphore
|
| 511 | * or @dev_base_lock.
|
| 512 |
|
| 513 |
|
| 514 | struct net_device *__dev_get_by_index(int ifindex)
|
| 515 | {
|
| 516 | struct hlist_node *p;
|
| 517 |
|
| 518 | hlist_for_each(p, dev_index_hash(ifindex)) {
|
| 519 | struct net_device *dev
|
| 520 | = hlist_entry(p, struct net_device, index_hlist);
|
| 521 | if (dev->ifindex == ifindex)
|
| 522 | return dev;
|
| 523 | }
|
| 524 | return NULL;
|
| 525 | }
|
| 526 |
|
| 527 |
|
| 528 | |
| 529 | * dev_get_by_index - find a device by its ifindex
|
| 530 | * @ifindex: index of device
|
| 531 | *
|
| 532 | * Search for an interface by index. Returns NULL if the device
|
| 533 | * is not found or a pointer to the device. The device returned has
|
| 534 | * had a reference added and the pointer is safe until the user calls
|
| 535 | * dev_put to indicate they have finished with it.
|
| 536 |
|
| 537 |
|
| 538 | struct net_device *dev_get_by_index(int ifindex)
|
| 539 | {
|
| 540 | struct net_device *dev;
|
| 541 |
|
| 542 | read_lock(&dev_base_lock);
|
| 543 | dev = __dev_get_by_index(ifindex);
|
| 544 | if (dev)
|
| 545 | dev_hold(dev);
|
| 546 | read_unlock(&dev_base_lock);
|
| 547 | return dev;
|
| 548 | }
|
| 549 |
|
| 550 | |
| 551 | * dev_getbyhwaddr - find a device by its hardware address
|
| 552 | * @type: media type of device
|
| 553 | * @ha: hardware address
|
| 554 | *
|
| 555 | * Search for an interface by MAC address. Returns NULL if the device
|
| 556 | * is not found or a pointer to the device. The caller must hold the
|
| 557 | * rtnl semaphore. The returned device has not had its ref count increased
|
| 558 | * and the caller must therefore be careful about locking
|
| 559 | *
|
| 560 | * BUGS:
|
| 561 | * If the API was consistent this would be __dev_get_by_hwaddr
|
| 562 |
|
| 563 |
|
| 564 | struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
|
| 565 | {
|
| 566 | struct net_device *dev;
|
| 567 |
|
| 568 | ASSERT_RTNL();
|
| 569 |
|
| 570 | for (dev = dev_base; dev; dev = dev->next)
|
| 571 | if (dev->type == type &&
|
| 572 | !memcmp(dev->dev_addr, ha, dev->addr_len))
|
| 573 | break;
|
| 574 | return dev;
|
| 575 | }
|
| 576 |
|
| 577 | struct net_device *dev_getfirstbyhwtype(unsigned short type)
|
| 578 | {
|
| 579 | struct net_device *dev;
|
| 580 |
|
| 581 | rtnl_lock();
|
| 582 | for (dev = dev_base; dev; dev = dev->next) {
|
| 583 | if (dev->type == type) {
|
| 584 | dev_hold(dev);
|
| 585 | break;
|
| 586 | }
|
| 587 | }
|
| 588 | rtnl_unlock();
|
| 589 | return dev;
|
| 590 | }
|
| 591 |
|
| 592 | EXPORT_SYMBOL(dev_getfirstbyhwtype);
|
| 593 |
|
| 594 | |
| 595 | * dev_get_by_flags - find any device with given flags
|
| 596 | * @if_flags: IFF_* values
|
| 597 | * @mask: bitmask of bits in if_flags to check
|
| 598 | *
|
| 599 | * Search for any interface with the given flags. Returns NULL if a device
|
| 600 | * is not found or a pointer to the device. The device returned has
|
| 601 | * had a reference added and the pointer is safe until the user calls
|
| 602 | * dev_put to indicate they have finished with it.
|
| 603 |
|
| 604 |
|
| 605 | struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
|
| 606 | {
|
| 607 | struct net_device *dev;
|
| 608 |
|
| 609 | read_lock(&dev_base_lock);
|
| 610 | for (dev = dev_base; dev != NULL; dev = dev->next) {
|
| 611 | if (((dev->flags ^ if_flags) & mask) == 0) {
|
| 612 | dev_hold(dev);
|
| 613 | break;
|
| 614 | }
|
| 615 | }
|
| 616 | read_unlock(&dev_base_lock);
|
| 617 | return dev;
|
| 618 | }
|
| 619 |
|
| 620 | |
| 621 | * dev_valid_name - check if name is okay for network device
|
| 622 | * @name: name string
|
| 623 | *
|
| 624 | * Network device names need to be valid file names to
|
| 625 | * to allow sysfs to work
|
| 626 |
|
| 627 | static int dev_valid_name(const char *name)
|
| 628 | {
|
| 629 | return !(*name == '\0'
|
| 630 | || !strcmp(name, ".")
|
| 631 | || !strcmp(name, "..")
|
| 632 | || strchr(name, '/'));
|
| 633 | }
|
| 634 |
|
| 635 | |
| 636 | * dev_alloc_name - allocate a name for a device
|
| 637 | * @dev: device
|
| 638 | * @name: name format string
|
| 639 | *
|
| 640 | * Passed a format string - eg "lt%d" it will try and find a suitable
|
| 641 | * id. Not efficient for many devices, not called a lot. The caller
|
| 642 | * must hold the dev_base or rtnl lock while allocating the name and
|
| 643 | * adding the device in order to avoid duplicates. Returns the number
|
| 644 | * of the unit assigned or a negative errno code.
|
| 645 |
|
| 646 |
|
| 647 | int dev_alloc_name(struct net_device *dev, const char *name)
|
| 648 | {
|
| 649 | int i = 0;
|
| 650 | char buf[IFNAMSIZ];
|
| 651 | const char *p;
|
| 652 | const int max_netdevices = 8*PAGE_SIZE;
|
| 653 | long *inuse;
|
| 654 | struct net_device *d;
|
| 655 |
|
| 656 | p = strnchr(name, IFNAMSIZ-1, '%');
|
| 657 | if (p) {
|
| 658 | |
| 659 | * Verify the string as this thing may have come from
|
| 660 | * the user. There must be either one "%d" and no other "%"
|
| 661 | * characters.
|
| 662 |
|
| 663 | if (p[1] != 'd' || strchr(p + 2, '%'))
|
| 664 | return -EINVAL;
|
| 665 |
|
| 666 |
|
| 667 | inuse = (long *) get_zeroed_page(GFP_ATOMIC);
|
| 668 | if (!inuse)
|
| 669 | return -ENOMEM;
|
| 670 |
|
| 671 | for (d = dev_base; d; d = d->next) {
|
| 672 | if (!sscanf(d->name, name, &i))
|
| 673 | continue;
|
| 674 | if (i < 0 || i >= max_netdevices)
|
| 675 | continue;
|
| 676 |
|
| 677 |
|
| 678 | snprintf(buf, sizeof(buf), name, i);
|
| 679 | if (!strncmp(buf, d->name, IFNAMSIZ))
|
| 680 | set_bit(i, inuse);
|
| 681 | }
|
| 682 |
|
| 683 | i = find_first_zero_bit(inuse, max_netdevices);
|
| 684 | free_page((unsigned long) inuse);
|
| 685 | }
|
| 686 |
|
| 687 | snprintf(buf, sizeof(buf), name, i);
|
| 688 | if (!__dev_get_by_name(buf)) {
|
| 689 | strlcpy(dev->name, buf, IFNAMSIZ);
|
| 690 | return i;
|
| 691 | }
|
| 692 |
|
| 693 | |
| 694 | * when the name is long and there isn't enough space left
|
| 695 | * for the digits, or if all bits are used.
|
| 696 |
|
| 697 | return -ENFILE;
|
| 698 | }
|
| 699 |
|
| 700 |
|
| 701 | |
| 702 | * dev_change_name - change name of a device
|
| 703 | * @dev: device
|
| 704 | * @newname: name (or format string) must be at least IFNAMSIZ
|
| 705 | *
|
| 706 | * Change name of a device, can pass format strings "eth%d".
|
| 707 | * for wildcarding.
|
| 708 |
|
| 709 | int dev_change_name(struct net_device *dev, char *newname)
|
| 710 | {
|
| 711 | int err = 0;
|
| 712 |
|
| 713 | ASSERT_RTNL();
|
| 714 |
|
| 715 | if (dev->flags & IFF_UP)
|
| 716 | return -EBUSY;
|
| 717 |
|
| 718 | if (!dev_valid_name(newname))
|
| 719 | return -EINVAL;
|
| 720 |
|
| 721 | if (strchr(newname, '%')) {
|
| 722 | err = dev_alloc_name(dev, newname);
|
| 723 | if (err < 0)
|
| 724 | return err;
|
| 725 | strcpy(newname, dev->name);
|
| 726 | }
|
| 727 | else if (__dev_get_by_name(newname))
|
| 728 | return -EEXIST;
|
| 729 | else
|
| 730 | strlcpy(dev->name, newname, IFNAMSIZ);
|
| 731 |
|
| 732 | err = class_device_rename(&dev->class_dev, dev->name);
|
| 733 | if (!err) {
|
| 734 | hlist_del(&dev->name_hlist);
|
| 735 | hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
|
| 736 | notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
|
| 737 | }
|
| 738 |
|
| 739 | return err;
|
| 740 | }
|
| 741 |
|
| 742 | |
| 743 | * netdev_features_change - device changes fatures
|
| 744 | * @dev: device to cause notification
|
| 745 | *
|
| 746 | * Called to indicate a device has changed features.
|
| 747 |
|
| 748 | void netdev_features_change(struct net_device *dev)
|
| 749 | {
|
| 750 | notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
|
| 751 | }
|
| 752 | EXPORT_SYMBOL(netdev_features_change);
|
| 753 |
|
| 754 | |
| 755 | * netdev_state_change - device changes state
|
| 756 | * @dev: device to cause notification
|
| 757 | *
|
| 758 | * Called to indicate a device has changed state. This function calls
|
| 759 | * the notifier chains for netdev_chain and sends a NEWLINK message
|
| 760 | * to the routing socket.
|
| 761 |
|
| 762 | void netdev_state_change(struct net_device *dev)
|
| 763 | {
|
| 764 | if (dev->flags & IFF_UP) {
|
| 765 | notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
|
| 766 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
|
| 767 | }
|
| 768 | }
|
| 769 |
|
| 770 | |
| 771 | * dev_load - load a network module
|
| 772 | * @name: name of interface
|
| 773 | *
|
| 774 | * If a network interface is not present and the process has suitable
|
| 775 | * privileges this function loads the module. If module loading is not
|
| 776 | * available in this kernel then it becomes a nop.
|
| 777 |
|
| 778 |
|
| 779 | void dev_load(const char *name)
|
| 780 | {
|
| 781 | struct net_device *dev;
|
| 782 |
|
| 783 | read_lock(&dev_base_lock);
|
| 784 | dev = __dev_get_by_name(name);
|
| 785 | read_unlock(&dev_base_lock);
|
| 786 |
|
| 787 | if (!dev && capable(CAP_SYS_MODULE))
|
| 788 | request_module("%s", name);
|
| 789 | }
|
| 790 |
|
| 791 | static int default_rebuild_header(struct sk_buff *skb)
|
| 792 | {
|
| 793 | printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
|
| 794 | skb->dev ? skb->dev->name : "NULL!!!");
|
| 795 | kfree_skb(skb);
|
| 796 | return 1;
|
| 797 | }
|
| 798 |
|
| 799 |
|
| 800 | |
| 801 | * dev_open - prepare an interface for use.
|
| 802 | * @dev: device to open
|
| 803 | *
|
| 804 | * Takes a device from down to up state. The device's private open
|
| 805 | * function is invoked and then the multicast lists are loaded. Finally
|
| 806 | * the device is moved into the up state and a %NETDEV_UP message is
|
| 807 | * sent to the netdev notifier chain.
|
| 808 | *
|
| 809 | * Calling this function on an active interface is a nop. On a failure
|
| 810 | * a negative errno code is returned.
|
| 811 |
|
| 812 | int dev_open(struct net_device *dev)
|
| 813 | {
|
| 814 | int ret = 0;
|
| 815 |
|
| 816 | |
| 817 | * Is it already up?
|
| 818 |
|
| 819 |
|
| 820 | if (dev->flags & IFF_UP)
|
| 821 | return 0;
|
| 822 |
|
| 823 | |
| 824 | * Is it even present?
|
| 825 |
|
| 826 | if (!netif_device_present(dev))
|
| 827 | return -ENODEV;
|
| 828 |
|
| 829 | |
| 830 | * Call device private open method
|
| 831 |
|
| 832 | set_bit(__LINK_STATE_START, &dev->state);
|
| 833 | if (dev->open) {
|
| 834 | ret = dev->open(dev);
|
| 835 | if (ret)
|
| 836 | clear_bit(__LINK_STATE_START, &dev->state);
|
| 837 | }
|
| 838 |
|
| 839 | |
| 840 | * If it went open OK then:
|
| 841 |
|
| 842 |
|
| 843 | if (!ret) {
|
| 844 | |
| 845 | * Set the flags.
|
| 846 |
|
| 847 | dev->flags |= IFF_UP;
|
| 848 |
|
| 849 | |
| 850 | * Initialize multicasting status
|
| 851 |
|
| 852 | dev_mc_upload(dev);
|
| 853 |
|
| 854 | |
| 855 | * Wakeup transmit queue engine
|
| 856 |
|
| 857 | dev_activate(dev);
|
| 858 |
|
| 859 | |
| 860 | * ... and announce new interface.
|
| 861 |
|
| 862 | notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
|
| 863 | }
|
| 864 | return ret;
|
| 865 | }
|
| 866 |
|
| 867 | |
| 868 | * dev_close - shutdown an interface.
|
| 869 | * @dev: device to shutdown
|
| 870 | *
|
| 871 | * This function moves an active device into down state. A
|
| 872 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
|
| 873 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
|
| 874 | * chain.
|
| 875 |
|
| 876 | int dev_close(struct net_device *dev)
|
| 877 | {
|
| 878 | if (!(dev->flags & IFF_UP))
|
| 879 | return 0;
|
| 880 |
|
| 881 | |
| 882 | * Tell people we are going down, so that they can
|
| 883 | * prepare to death, when device is still operating.
|
| 884 |
|
| 885 | notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
|
| 886 |
|
| 887 | dev_deactivate(dev);
|
| 888 |
|
| 889 | clear_bit(__LINK_STATE_START, &dev->state);
|
| 890 |
|
| 891 | |
| 892 | * it can be even on different cpu. So just clear netif_running(),
|
| 893 | * and wait when poll really will happen. Actually, the best place
|
| 894 | * for this is inside dev->stop() after device stopped its irq
|
| 895 | * engine, but this requires more changes in devices.
|
| 896 |
|
| 897 | smp_mb__after_clear_bit();
|
| 898 | while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
|
| 899 |
|
| 900 | msleep(1);
|
| 901 | }
|
| 902 |
|
| 903 | |
| 904 | * Call the device specific close. This cannot fail.
|
| 905 | * Only if device is UP
|
| 906 | *
|
| 907 | * We allow it to be called even after a DETACH hot-plug
|
| 908 | * event.
|
| 909 |
|
| 910 | if (dev->stop)
|
| 911 | dev->stop(dev);
|
| 912 |
|
| 913 | |
| 914 | * Device is now down.
|
| 915 |
|
| 916 |
|
| 917 | dev->flags &= ~IFF_UP;
|
| 918 |
|
| 919 | |
| 920 | * Tell people we are down
|
| 921 |
|
| 922 | notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
|
| 923 |
|
| 924 | return 0;
|
| 925 | }
|
| 926 |
|
| 927 |
|
| 928 | |
| 929 | * Device change register/unregister. These are not inline or static
|
| 930 | * as we export them to the world.
|
| 931 |
|
| 932 |
|
| 933 | |
| 934 | * register_netdevice_notifier - register a network notifier block
|
| 935 | * @nb: notifier
|
| 936 | *
|
| 937 | * Register a notifier to be called when network device events occur.
|
| 938 | * The notifier passed is linked into the kernel structures and must
|
| 939 | * not be reused until it has been unregistered. A negative errno code
|
| 940 | * is returned on a failure.
|
| 941 | *
|
| 942 | * When registered all registration and up events are replayed
|
| 943 | * to the new notifier to allow device to have a race free
|
| 944 | * view of the network device list.
|
| 945 |
|
| 946 |
|
| 947 | int register_netdevice_notifier(struct notifier_block *nb)
|
| 948 | {
|
| 949 | struct net_device *dev;
|
| 950 | int err;
|
| 951 |
|
| 952 | rtnl_lock();
|
| 953 | err = notifier_chain_register(&netdev_chain, nb);
|
| 954 | if (!err) {
|
| 955 | for (dev = dev_base; dev; dev = dev->next) {
|
| 956 | nb->notifier_call(nb, NETDEV_REGISTER, dev);
|
| 957 |
|
| 958 | if (dev->flags & IFF_UP)
|
| 959 | nb->notifier_call(nb, NETDEV_UP, dev);
|
| 960 | }
|
| 961 | }
|
| 962 | rtnl_unlock();
|
| 963 | return err;
|
| 964 | }
|
| 965 |
|
| 966 | |
| 967 | * unregister_netdevice_notifier - unregister a network notifier block
|
| 968 | * @nb: notifier
|
| 969 | *
|
| 970 | * Unregister a notifier previously registered by
|
| 971 | * register_netdevice_notifier(). The notifier is unlinked into the
|
| 972 | * kernel structures and may then be reused. A negative errno code
|
| 973 | * is returned on a failure.
|
| 974 |
|
| 975 |
|
| 976 | int unregister_netdevice_notifier(struct notifier_block *nb)
|
| 977 | {
|
| 978 | return notifier_chain_unregister(&netdev_chain, nb);
|
| 979 | }
|
| 980 |
|
| 981 | |
| 982 | * call_netdevice_notifiers - call all network notifier blocks
|
| 983 | * @val: value passed unmodified to notifier function
|
| 984 | * @v: pointer passed unmodified to notifier function
|
| 985 | *
|
| 986 | * Call all network notifier blocks. Parameters and return value
|
| 987 | * are as for notifier_call_chain().
|
| 988 |
|
| 989 |
|
| 990 | int call_netdevice_notifiers(unsigned long val, void *v)
|
| 991 | {
|
| 992 | return notifier_call_chain(&netdev_chain, val, v);
|
| 993 | }
|
| 994 |
|
| 995 |
|
| 996 | static atomic_t netstamp_needed = ATOMIC_INIT(0);
|
| 997 |
|
| 998 | void net_enable_timestamp(void)
|
| 999 | {
|
| 1000 | atomic_inc(&netstamp_needed);
|
| 1001 | }
|
| 1002 |
|
| 1003 | void net_disable_timestamp(void)
|
| 1004 | {
|
| 1005 | atomic_dec(&netstamp_needed);
|
| 1006 | }
|
| 1007 |
|
| 1008 | void __net_timestamp(struct sk_buff *skb)
|
| 1009 | {
|
| 1010 | struct timeval tv;
|
| 1011 |
|
| 1012 | do_gettimeofday(&tv);
|
| 1013 | skb_set_timestamp(skb, &tv);
|
| 1014 | }
|
| 1015 | EXPORT_SYMBOL(__net_timestamp);
|
| 1016 |
|
| 1017 | static inline void net_timestamp(struct sk_buff *skb)
|
| 1018 | {
|
| 1019 | if (atomic_read(&netstamp_needed))
|
| 1020 | __net_timestamp(skb);
|
| 1021 | else {
|
| 1022 | skb->tstamp.off_sec = 0;
|
| 1023 | skb->tstamp.off_usec = 0;
|
| 1024 | }
|
| 1025 | }
|
| 1026 |
|
| 1027 | |
| 1028 | * Support routine. Sends outgoing frames to any network
|
| 1029 | * taps currently in use.
|
| 1030 |
|
| 1031 |
|
| 1032 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
|
| 1033 | {
|
| 1034 | struct packet_type *ptype;
|
| 1035 |
|
| 1036 | net_timestamp(skb);
|
| 1037 |
|
| 1038 | rcu_read_lock();
|
| 1039 | list_for_each_entry_rcu(ptype, &ptype_all, list) {
|
| 1040 | |
| 1041 | * they originated from - MvS ([email protected])
|
| 1042 |
|
| 1043 | if ((ptype->dev == dev || !ptype->dev) &&
|
| 1044 | (ptype->af_packet_priv == NULL ||
|
| 1045 | (struct sock *)ptype->af_packet_priv != skb->sk)) {
|
| 1046 | struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
|
| 1047 | if (!skb2)
|
| 1048 | break;
|
| 1049 |
|
| 1050 | |
| 1051 | set by sender, so that the second statement is
|
| 1052 | just protection against buggy protocols.
|
| 1053 |
|
| 1054 | skb2->mac.raw = skb2->data;
|
| 1055 |
|
| 1056 | if (skb2->nh.raw < skb2->data ||
|
| 1057 | skb2->nh.raw > skb2->tail) {
|
| 1058 | if (net_ratelimit())
|
| 1059 | printk(KERN_CRIT "protocol %04x is "
|
| 1060 | "buggy, dev %s\n",
|
| 1061 | skb2->protocol, dev->name);
|
| 1062 | skb2->nh.raw = skb2->data;
|
| 1063 | }
|
| 1064 |
|
| 1065 | skb2->h.raw = skb2->nh.raw;
|
| 1066 | skb2->pkt_type = PACKET_OUTGOING;
|
| 1067 | ptype->func(skb2, skb->dev, ptype, skb->dev);
|
| 1068 | }
|
| 1069 | }
|
| 1070 | rcu_read_unlock();
|
| 1071 | }
|
| 1072 |
|
| 1073 | |
| 1074 | * Invalidate hardware checksum when packet is to be mangled, and
|
| 1075 | * complete checksum manually on outgoing path.
|
| 1076 |
|
| 1077 | int skb_checksum_help(struct sk_buff *skb, int inward)
|
| 1078 | {
|
| 1079 | unsigned int csum;
|
| 1080 | int ret = 0, offset = skb->h.raw - skb->data;
|
| 1081 |
|
| 1082 | if (inward) {
|
| 1083 | skb->ip_summed = CHECKSUM_NONE;
|
| 1084 | goto out;
|
| 1085 | }
|
| 1086 |
|
| 1087 | if (skb_cloned(skb)) {
|
| 1088 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
|
| 1089 | if (ret)
|
| 1090 | goto out;
|
| 1091 | }
|
| 1092 |
|
| 1093 | if (offset > (int)skb->len)
|
| 1094 | BUG();
|
| 1095 | csum = skb_checksum(skb, offset, skb->len-offset, 0);
|
| 1096 |
|
| 1097 | offset = skb->tail - skb->h.raw;
|
| 1098 | if (offset <= 0)
|
| 1099 | BUG();
|
| 1100 | if (skb->csum + 2 > offset)
|
| 1101 | BUG();
|
| 1102 |
|
| 1103 | *(u16*)(skb->h.raw + skb->csum) = csum_fold(csum);
|
| 1104 | skb->ip_summed = CHECKSUM_NONE;
|
| 1105 | out:
|
| 1106 | return ret;
|
| 1107 | }
|
| 1108 |
|
| 1109 | #ifdef CONFIG_HIGHMEM
|
| 1110 | |
| 1111 | * 1. IOMMU is present and allows to map all the memory.
|
| 1112 | * 2. No high memory really exists on this machine.
|
| 1113 |
|
| 1114 |
|
| 1115 | static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
|
| 1116 | {
|
| 1117 | int i;
|
| 1118 |
|
| 1119 | if (dev->features & NETIF_F_HIGHDMA)
|
| 1120 | return 0;
|
| 1121 |
|
| 1122 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
|
| 1123 | if (PageHighMem(skb_shinfo(skb)->frags[i].page))
|
| 1124 | return 1;
|
| 1125 |
|
| 1126 | return 0;
|
| 1127 | }
|
| 1128 | #else
|
| 1129 | #define illegal_highdma(dev, skb) (0)
|
| 1130 | #endif
|
| 1131 |
|
| 1132 |
|
| 1133 | int __skb_linearize(struct sk_buff *skb, unsigned int __nocast gfp_mask)
|
| 1134 | {
|
| 1135 | unsigned int size;
|
| 1136 | u8 *data;
|
| 1137 | long offset;
|
| 1138 | struct skb_shared_info *ninfo;
|
| 1139 | int headerlen = skb->data - skb->head;
|
| 1140 | int expand = (skb->tail + skb->data_len) - skb->end;
|
| 1141 |
|
| 1142 | if (skb_shared(skb))
|
| 1143 | BUG();
|
| 1144 |
|
| 1145 | if (expand <= 0)
|
| 1146 | expand = 0;
|
| 1147 |
|
| 1148 | size = skb->end - skb->head + expand;
|
| 1149 | size = SKB_DATA_ALIGN(size);
|
| 1150 | data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
|
| 1151 | if (!data)
|
| 1152 | return -ENOMEM;
|
| 1153 |
|
| 1154 |
|
| 1155 | if (skb_copy_bits(skb, -headerlen, data, headerlen + skb->len))
|
| 1156 | BUG();
|
| 1157 |
|
| 1158 |
|
| 1159 | ninfo = (struct skb_shared_info*)(data + size);
|
| 1160 | atomic_set(&ninfo->dataref, 1);
|
| 1161 | ninfo->tso_size = skb_shinfo(skb)->tso_size;
|
| 1162 | ninfo->tso_segs = skb_shinfo(skb)->tso_segs;
|
| 1163 | ninfo->nr_frags = 0;
|
| 1164 | ninfo->frag_list = NULL;
|
| 1165 |
|
| 1166 |
|
| 1167 | offset = data - skb->head;
|
| 1168 |
|
| 1169 |
|
| 1170 | skb_release_data(skb);
|
| 1171 |
|
| 1172 | skb->head = data;
|
| 1173 | skb->end = data + size;
|
| 1174 |
|
| 1175 |
|
| 1176 | skb->h.raw += offset;
|
| 1177 | skb->nh.raw += offset;
|
| 1178 | skb->mac.raw += offset;
|
| 1179 | skb->tail += offset;
|
| 1180 | skb->data += offset;
|
| 1181 |
|
| 1182 |
|
| 1183 | skb->cloned = 0;
|
| 1184 |
|
| 1185 | skb->tail += skb->data_len;
|
| 1186 | skb->data_len = 0;
|
| 1187 | return 0;
|
| 1188 | }
|
| 1189 |
|
| 1190 | #define HARD_TX_LOCK(dev, cpu) { \
|
| 1191 | if ((dev->features & NETIF_F_LLTX) == 0) { \
|
| 1192 | spin_lock(&dev->xmit_lock); \
|
| 1193 | dev->xmit_lock_owner = cpu; \
|
| 1194 | } \
|
| 1195 | }
|
| 1196 |
|
| 1197 | #define HARD_TX_UNLOCK(dev) { \
|
| 1198 | if ((dev->features & NETIF_F_LLTX) == 0) { \
|
| 1199 | dev->xmit_lock_owner = -1; \
|
| 1200 | spin_unlock(&dev->xmit_lock); \
|
| 1201 | } \
|
| 1202 | }
|
| 1203 |
|
| 1204 | |
| 1205 | * dev_queue_xmit - transmit a buffer
|
| 1206 | * @skb: buffer to transmit
|
| 1207 | *
|
| 1208 | * Queue a buffer for transmission to a network device. The caller must
|
| 1209 | * have set the device and priority and built the buffer before calling
|
| 1210 | * this function. The function can be called from an interrupt.
|
| 1211 | *
|
| 1212 | * A negative errno code is returned on a failure. A success does not
|
| 1213 | * guarantee the frame will be transmitted as it may be dropped due
|
| 1214 | * to congestion or traffic shaping.
|
| 1215 | *
|
| 1216 | * -----------------------------------------------------------------------------------
|
| 1217 | * I notice this method can also return errors from the queue disciplines,
|
| 1218 | * including NET_XMIT_DROP, which is a positive value. So, errors can also
|
| 1219 | * be positive.
|
| 1220 | *
|
| 1221 | * Regardless of the return value, the skb is consumed, so it is currently
|
| 1222 | * difficult to retry a send to this method. (You can bump the ref count
|
| 1223 | * before sending to hold a reference for retry if you are careful.)
|
| 1224 | *
|
| 1225 | * When calling this method, interrupts MUST be enabled. This is because
|
| 1226 | * the BH enable code must have IRQs enabled so that it will not deadlock.
|
| 1227 | * --BLG
|
| 1228 |
|
| 1229 |
|
| 1230 | int dev_queue_xmit(struct sk_buff *skb)
|
| 1231 | {
|
| 1232 | struct net_device *dev = skb->dev;
|
| 1233 | struct Qdisc *q;
|
| 1234 | int rc = -ENOMEM;
|
| 1235 |
|
| 1236 | if (skb_shinfo(skb)->frag_list &&
|
| 1237 | !(dev->features & NETIF_F_FRAGLIST) &&
|
| 1238 | __skb_linearize(skb, GFP_ATOMIC))
|
| 1239 | goto out_kfree_skb;
|
| 1240 |
|
| 1241 | |
| 1242 | * or if at least one of fragments is in highmem and device
|
| 1243 | * does not support DMA from it.
|
| 1244 |
|
| 1245 | if (skb_shinfo(skb)->nr_frags &&
|
| 1246 | (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
|
| 1247 | __skb_linearize(skb, GFP_ATOMIC))
|
| 1248 | goto out_kfree_skb;
|
| 1249 |
|
| 1250 | |
| 1251 | * checksumming for this protocol, complete checksumming here.
|
| 1252 |
|
| 1253 | if (skb->ip_summed == CHECKSUM_HW &&
|
| 1254 | (!(dev->features & (NETIF_F_HW_CSUM | NETIF_F_NO_CSUM)) &&
|
| 1255 | (!(dev->features & NETIF_F_IP_CSUM) ||
|
| 1256 | skb->protocol != htons(ETH_P_IP))))
|
| 1257 | if (skb_checksum_help(skb, 0))
|
| 1258 | goto out_kfree_skb;
|
| 1259 |
|
| 1260 | |
| 1261 | * stops preemption for RCU.
|
| 1262 |
|
| 1263 | local_bh_disable();
|
| 1264 |
|
| 1265 | |
| 1266 | * The struct Qdisc which is pointed to by qdisc is now a
|
| 1267 | * rcu structure - it may be accessed without acquiring
|
| 1268 | * a lock (but the structure may be stale.) The freeing of the
|
| 1269 | * qdisc will be deferred until it's known that there are no
|
| 1270 | * more references to it.
|
| 1271 | *
|
| 1272 | * If the qdisc has an enqueue function, we still need to
|
| 1273 | * hold the queue_lock before calling it, since queue_lock
|
| 1274 | * also serializes access to the device queue.
|
| 1275 |
|
| 1276 |
|
| 1277 | q = rcu_dereference(dev->qdisc);
|
| 1278 | #ifdef CONFIG_NET_CLS_ACT
|
| 1279 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
|
| 1280 | #endif
|
| 1281 | if (q->enqueue) {
|
| 1282 |
|
| 1283 | spin_lock(&dev->queue_lock);
|
| 1284 |
|
| 1285 | rc = q->enqueue(skb, q);
|
| 1286 |
|
| 1287 | qdisc_run(dev);
|
| 1288 |
|
| 1289 | spin_unlock(&dev->queue_lock);
|
| 1290 | rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
|
| 1291 | goto out;
|
| 1292 | }
|
| 1293 |
|
| 1294 | |
| 1295 | loopback, all the sorts of tunnels...
|
| 1296 |
|
| 1297 | Really, it is unlikely that xmit_lock protection is necessary here.
|
| 1298 | (f.e. loopback and IP tunnels are clean ignoring statistics
|
| 1299 | counters.)
|
| 1300 | However, it is possible, that they rely on protection
|
| 1301 | made by us here.
|
| 1302 |
|
| 1303 | Check this and shot the lock. It is not prone from deadlocks.
|
| 1304 | Either shot noqueue qdisc, it is even simpler 8)
|
| 1305 |
|
| 1306 | if (dev->flags & IFF_UP) {
|
| 1307 | int cpu = smp_processor_id();
|
| 1308 |
|
| 1309 | if (dev->xmit_lock_owner != cpu) {
|
| 1310 |
|
| 1311 | HARD_TX_LOCK(dev, cpu);
|
| 1312 |
|
| 1313 | if (!netif_queue_stopped(dev)) {
|
| 1314 | if (netdev_nit)
|
| 1315 | dev_queue_xmit_nit(skb, dev);
|
| 1316 |
|
| 1317 | rc = 0;
|
| 1318 | if (!dev->hard_start_xmit(skb, dev)) {
|
| 1319 | HARD_TX_UNLOCK(dev);
|
| 1320 | goto out;
|
| 1321 | }
|
| 1322 | }
|
| 1323 | HARD_TX_UNLOCK(dev);
|
| 1324 | if (net_ratelimit())
|
| 1325 | printk(KERN_CRIT "Virtual device %s asks to "
|
| 1326 | "queue packet!\n", dev->name);
|
| 1327 | } else {
|
| 1328 | |
| 1329 | * unfortunately
|
| 1330 | if (net_ratelimit())
|
| 1331 | printk(KERN_CRIT "Dead loop on virtual device "
|
| 1332 | "%s, fix it urgently!\n", dev->name);
|
| 1333 | }
|
| 1334 | }
|
| 1335 |
|
| 1336 | rc = -ENETDOWN;
|
| 1337 | local_bh_enable();
|
| 1338 |
|
| 1339 | out_kfree_skb:
|
| 1340 | kfree_skb(skb);
|
| 1341 | return rc;
|
| 1342 | out:
|
| 1343 | local_bh_enable();
|
| 1344 | return rc;
|
| 1345 | }
|
| 1346 |
|
| 1347 |
|
| 1348 | |
| 1349 | Receiver routines
|
| 1350 | =======================================================================
|
| 1351 |
|
| 1352 | int netdev_max_backlog = 1000;
|
| 1353 | int netdev_budget = 300;
|
| 1354 | int weight_p = 64;
|
| 1355 |
|
| 1356 | DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
|
| 1357 |
|
| 1358 |
|
| 1359 | |
| 1360 | * netif_rx - post buffer to the network code
|
| 1361 | * @skb: buffer to post
|
| 1362 | *
|
| 1363 | * This function receives a packet from a device driver and queues it for
|
| 1364 | * the upper (protocol) levels to process. It always succeeds. The buffer
|
| 1365 | * may be dropped during processing for congestion control or by the
|
| 1366 | * protocol layers.
|
| 1367 | *
|
| 1368 | * return values:
|
| 1369 | * NET_RX_SUCCESS (no congestion)
|
| 1370 | * NET_RX_CN_LOW (low congestion)
|
| 1371 | * NET_RX_CN_MOD (moderate congestion)
|
| 1372 | * NET_RX_CN_HIGH (high congestion)
|
| 1373 | * NET_RX_DROP (packet was dropped)
|
| 1374 | *
|
| 1375 |
|
| 1376 |
|
| 1377 | int netif_rx(struct sk_buff *skb)
|
| 1378 | {
|
| 1379 | struct softnet_data *queue;
|
| 1380 | unsigned long flags;
|
| 1381 |
|
| 1382 |
|
| 1383 | if (netpoll_rx(skb))
|
| 1384 | return NET_RX_DROP;
|
| 1385 |
|
| 1386 | if (!skb->tstamp.off_sec)
|
| 1387 | net_timestamp(skb);
|
| 1388 |
|
| 1389 | |
| 1390 | * The code is rearranged so that the path is the most
|
| 1391 | * short when CPU is congested, but is still operating.
|
| 1392 |
|
| 1393 | local_irq_save(flags);
|
| 1394 | queue = &__get_cpu_var(softnet_data);
|
| 1395 |
|
| 1396 | __get_cpu_var(netdev_rx_stat).total++;
|
| 1397 | if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
|
| 1398 | if (queue->input_pkt_queue.qlen) {
|
| 1399 | enqueue:
|
| 1400 | dev_hold(skb->dev);
|
| 1401 | __skb_queue_tail(&queue->input_pkt_queue, skb);
|
| 1402 | local_irq_restore(flags);
|
| 1403 | return NET_RX_SUCCESS;
|
| 1404 | }
|
| 1405 |
|
| 1406 | netif_rx_schedule(&queue->backlog_dev);
|
| 1407 | goto enqueue;
|
| 1408 | }
|
| 1409 |
|
| 1410 | __get_cpu_var(netdev_rx_stat).dropped++;
|
| 1411 | local_irq_restore(flags);
|
| 1412 |
|
| 1413 | kfree_skb(skb);
|
| 1414 | return NET_RX_DROP;
|
| 1415 | }
|
| 1416 |
|
| 1417 | int netif_rx_ni(struct sk_buff *skb)
|
| 1418 | {
|
| 1419 | int err;
|
| 1420 |
|
| 1421 | preempt_disable();
|
| 1422 | err = netif_rx(skb);
|
| 1423 | if (local_softirq_pending())
|
| 1424 | do_softirq();
|
| 1425 | preempt_enable();
|
| 1426 |
|
| 1427 | return err;
|
| 1428 | }
|
| 1429 |
|
| 1430 | EXPORT_SYMBOL(netif_rx_ni);
|
| 1431 |
|
| 1432 | static inline struct net_device *skb_bond(struct sk_buff *skb)
|
| 1433 | {
|
| 1434 | struct net_device *dev = skb->dev;
|
| 1435 |
|
| 1436 | if (dev->master)
|
| 1437 | skb->dev = dev->master;
|
| 1438 |
|
| 1439 | return dev;
|
| 1440 | }
|
| 1441 |
|
| 1442 | static void net_tx_action(struct softirq_action *h)
|
| 1443 | {
|
| 1444 | struct softnet_data *sd = &__get_cpu_var(softnet_data);
|
| 1445 |
|
| 1446 | if (sd->completion_queue) {
|
| 1447 | struct sk_buff *clist;
|
| 1448 |
|
| 1449 | local_irq_disable();
|
| 1450 | clist = sd->completion_queue;
|
| 1451 | sd->completion_queue = NULL;
|
| 1452 | local_irq_enable();
|
| 1453 |
|
| 1454 | while (clist) {
|
| 1455 | struct sk_buff *skb = clist;
|
| 1456 | clist = clist->next;
|
| 1457 |
|
| 1458 | BUG_TRAP(!atomic_read(&skb->users));
|
| 1459 | __kfree_skb(skb);
|
| 1460 | }
|
| 1461 | }
|
| 1462 |
|
| 1463 | if (sd->output_queue) {
|
| 1464 | struct net_device *head;
|
| 1465 |
|
| 1466 | local_irq_disable();
|
| 1467 | head = sd->output_queue;
|
| 1468 | sd->output_queue = NULL;
|
| 1469 | local_irq_enable();
|
| 1470 |
|
| 1471 | while (head) {
|
| 1472 | struct net_device *dev = head;
|
| 1473 | head = head->next_sched;
|
| 1474 |
|
| 1475 | smp_mb__before_clear_bit();
|
| 1476 | clear_bit(__LINK_STATE_SCHED, &dev->state);
|
| 1477 |
|
| 1478 | if (spin_trylock(&dev->queue_lock)) {
|
| 1479 | qdisc_run(dev);
|
| 1480 | spin_unlock(&dev->queue_lock);
|
| 1481 | } else {
|
| 1482 | netif_schedule(dev);
|
| 1483 | }
|
| 1484 | }
|
| 1485 | }
|
| 1486 | }
|
| 1487 |
|
| 1488 | static __inline__ int deliver_skb(struct sk_buff *skb,
|
| 1489 | struct packet_type *pt_prev,
|
| 1490 | struct net_device *orig_dev)
|
| 1491 | {
|
| 1492 | atomic_inc(&skb->users);
|
| 1493 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
|
| 1494 | }
|
| 1495 |
|
| 1496 | #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
|
| 1497 | int (*br_handle_frame_hook)(struct net_bridge_port *p, struct sk_buff **pskb);
|
| 1498 | struct net_bridge;
|
| 1499 | struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
|
| 1500 | unsigned char *addr);
|
| 1501 | void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent);
|
| 1502 |
|
| 1503 | static __inline__ int handle_bridge(struct sk_buff **pskb,
|
| 1504 | struct packet_type **pt_prev, int *ret,
|
| 1505 | struct net_device *orig_dev)
|
| 1506 | {
|
| 1507 | struct net_bridge_port *port;
|
| 1508 |
|
| 1509 | if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
|
| 1510 | (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
|
| 1511 | return 0;
|
| 1512 |
|
| 1513 | if (*pt_prev) {
|
| 1514 | *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
|
| 1515 | *pt_prev = NULL;
|
| 1516 | }
|
| 1517 |
|
| 1518 | return br_handle_frame_hook(port, pskb);
|
| 1519 | }
|
| 1520 | #else
|
| 1521 | #define handle_bridge(skb, pt_prev, ret, orig_dev) (0)
|
| 1522 | #endif
|
| 1523 |
|
| 1524 | #ifdef CONFIG_NET_CLS_ACT
|
| 1525 | |
| 1526 | * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
|
| 1527 | * a compare and 2 stores extra right now if we dont have it on
|
| 1528 | * but have CONFIG_NET_CLS_ACT
|
| 1529 | * NOTE: This doesnt stop any functionality; if you dont have
|
| 1530 | * the ingress scheduler, you just cant add policies on ingress.
|
| 1531 | *
|
| 1532 |
|
| 1533 | static int ing_filter(struct sk_buff *skb)
|
| 1534 | {
|
| 1535 | struct Qdisc *q;
|
| 1536 | struct net_device *dev = skb->dev;
|
| 1537 | int result = TC_ACT_OK;
|
| 1538 |
|
| 1539 | if (dev->qdisc_ingress) {
|
| 1540 | __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
|
| 1541 | if (MAX_RED_LOOP < ttl++) {
|
| 1542 | printk("Redir loop detected Dropping packet (%s->%s)\n",
|
| 1543 | skb->input_dev->name, skb->dev->name);
|
| 1544 | return TC_ACT_SHOT;
|
| 1545 | }
|
| 1546 |
|
| 1547 | skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
|
| 1548 |
|
| 1549 | skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
|
| 1550 |
|
| 1551 | spin_lock(&dev->ingress_lock);
|
| 1552 | if ((q = dev->qdisc_ingress) != NULL)
|
| 1553 | result = q->enqueue(skb, q);
|
| 1554 | spin_unlock(&dev->ingress_lock);
|
| 1555 |
|
| 1556 | }
|
| 1557 |
|
| 1558 | return result;
|
| 1559 | }
|
| 1560 | #endif
|
| 1561 |
|
| 1562 | int netif_receive_skb(struct sk_buff *skb)
|
| 1563 | {
|
| 1564 | struct packet_type *ptype, *pt_prev;
|
| 1565 | struct net_device *orig_dev;
|
| 1566 | int ret = NET_RX_DROP;
|
| 1567 | unsigned short type;
|
| 1568 |
|
| 1569 |
|
| 1570 | if (skb->dev->poll && netpoll_rx(skb))
|
| 1571 | return NET_RX_DROP;
|
| 1572 |
|
| 1573 | if (!skb->tstamp.off_sec)
|
| 1574 | net_timestamp(skb);
|
| 1575 |
|
| 1576 | if (!skb->input_dev)
|
| 1577 | skb->input_dev = skb->dev;
|
| 1578 |
|
| 1579 | orig_dev = skb_bond(skb);
|
| 1580 |
|
| 1581 | __get_cpu_var(netdev_rx_stat).total++;
|
| 1582 |
|
| 1583 | skb->h.raw = skb->nh.raw = skb->data;
|
| 1584 | skb->mac_len = skb->nh.raw - skb->mac.raw;
|
| 1585 |
|
| 1586 | pt_prev = NULL;
|
| 1587 |
|
| 1588 | rcu_read_lock();
|
| 1589 |
|
| 1590 | #ifdef CONFIG_NET_CLS_ACT
|
| 1591 | if (skb->tc_verd & TC_NCLS) {
|
| 1592 | skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
|
| 1593 | goto ncls;
|
| 1594 | }
|
| 1595 | #endif
|
| 1596 |
|
| 1597 | list_for_each_entry_rcu(ptype, &ptype_all, list) {
|
| 1598 | if (!ptype->dev || ptype->dev == skb->dev) {
|
| 1599 | if (pt_prev)
|
| 1600 | ret = deliver_skb(skb, pt_prev, orig_dev);
|
| 1601 | pt_prev = ptype;
|
| 1602 | }
|
| 1603 | }
|
| 1604 |
|
| 1605 | #ifdef CONFIG_NET_CLS_ACT
|
| 1606 | if (pt_prev) {
|
| 1607 | ret = deliver_skb(skb, pt_prev, orig_dev);
|
| 1608 | pt_prev = NULL;
|
| 1609 | } else {
|
| 1610 | skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
|
| 1611 | }
|
| 1612 |
|
| 1613 | ret = ing_filter(skb);
|
| 1614 |
|
| 1615 | if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
|
| 1616 | kfree_skb(skb);
|
| 1617 | goto out;
|
| 1618 | }
|
| 1619 |
|
| 1620 | skb->tc_verd = 0;
|
| 1621 | ncls:
|
| 1622 | #endif
|
| 1623 |
|
| 1624 | handle_diverter(skb);
|
| 1625 |
|
| 1626 | if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
|
| 1627 | goto out;
|
| 1628 |
|
| 1629 | type = skb->protocol;
|
| 1630 | list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
|
| 1631 | if (ptype->type == type &&
|
| 1632 | (!ptype->dev || ptype->dev == skb->dev)) {
|
| 1633 | if (pt_prev)
|
| 1634 | ret = deliver_skb(skb, pt_prev, orig_dev);
|
| 1635 | pt_prev = ptype;
|
| 1636 | }
|
| 1637 | }
|
| 1638 |
|
| 1639 | if (pt_prev) {
|
| 1640 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
|
| 1641 | } else {
|
| 1642 | kfree_skb(skb);
|
| 1643 | |
| 1644 | * me how you were going to use this. :-)
|
| 1645 |
|
| 1646 | ret = NET_RX_DROP;
|
| 1647 | }
|
| 1648 |
|
| 1649 | out:
|
| 1650 | rcu_read_unlock();
|
| 1651 | return ret;
|
| 1652 | }
|
| 1653 |
|
| 1654 | static int process_backlog(struct net_device *backlog_dev, int *budget)
|
| 1655 | {
|
| 1656 | int work = 0;
|
| 1657 | int quota = min(backlog_dev->quota, *budget);
|
| 1658 | struct softnet_data *queue = &__get_cpu_var(softnet_data);
|
| 1659 | unsigned long start_time = jiffies;
|
| 1660 |
|
| 1661 | backlog_dev->weight = weight_p;
|
| 1662 | for (;;) {
|
| 1663 | struct sk_buff *skb;
|
| 1664 | struct net_device *dev;
|
| 1665 |
|
| 1666 | local_irq_disable();
|
| 1667 | skb = __skb_dequeue(&queue->input_pkt_queue);
|
| 1668 | if (!skb)
|
| 1669 | goto job_done;
|
| 1670 | local_irq_enable();
|
| 1671 |
|
| 1672 | dev = skb->dev;
|
| 1673 |
|
| 1674 | netif_receive_skb(skb);
|
| 1675 |
|
| 1676 | dev_put(dev);
|
| 1677 |
|
| 1678 | work++;
|
| 1679 |
|
| 1680 | if (work >= quota || jiffies - start_time > 1)
|
| 1681 | break;
|
| 1682 |
|
| 1683 | }
|
| 1684 |
|
| 1685 | backlog_dev->quota -= work;
|
| 1686 | *budget -= work;
|
| 1687 | return -1;
|
| 1688 |
|
| 1689 | job_done:
|
| 1690 | backlog_dev->quota -= work;
|
| 1691 | *budget -= work;
|
| 1692 |
|
| 1693 | list_del(&backlog_dev->poll_list);
|
| 1694 | smp_mb__before_clear_bit();
|
| 1695 | netif_poll_enable(backlog_dev);
|
| 1696 |
|
| 1697 | local_irq_enable();
|
| 1698 | return 0;
|
| 1699 | }
|
| 1700 |
|
| 1701 | static void net_rx_action(struct softirq_action *h)
|
| 1702 | {
|
| 1703 | struct softnet_data *queue = &__get_cpu_var(softnet_data);
|
| 1704 | unsigned long start_time = jiffies;
|
| 1705 | int budget = netdev_budget;
|
| 1706 | void *have;
|
| 1707 |
|
| 1708 | local_irq_disable();
|
| 1709 |
|
| 1710 | while (!list_empty(&queue->poll_list)) {
|
| 1711 | struct net_device *dev;
|
| 1712 |
|
| 1713 | if (budget <= 0 || jiffies - start_time > 1)
|
| 1714 | goto softnet_break;
|
| 1715 |
|
| 1716 | local_irq_enable();
|
| 1717 |
|
| 1718 | dev = list_entry(queue->poll_list.next,
|
| 1719 | struct net_device, poll_list);
|
| 1720 | have = netpoll_poll_lock(dev);
|
| 1721 |
|
| 1722 | if (dev->quota <= 0 || dev->poll(dev, &budget)) {
|
| 1723 | netpoll_poll_unlock(have);
|
| 1724 | local_irq_disable();
|
| 1725 | list_del(&dev->poll_list);
|
| 1726 | list_add_tail(&dev->poll_list, &queue->poll_list);
|
| 1727 | if (dev->quota < 0)
|
| 1728 | dev->quota += dev->weight;
|
| 1729 | else
|
| 1730 | dev->quota = dev->weight;
|
| 1731 | } else {
|
| 1732 | netpoll_poll_unlock(have);
|
| 1733 | dev_put(dev);
|
| 1734 | local_irq_disable();
|
| 1735 | }
|
| 1736 | }
|
| 1737 | out:
|
| 1738 | local_irq_enable();
|
| 1739 | return;
|
| 1740 |
|
| 1741 | softnet_break:
|
| 1742 | __get_cpu_var(netdev_rx_stat).time_squeeze++;
|
| 1743 | __raise_softirq_irqoff(NET_RX_SOFTIRQ);
|
| 1744 | goto out;
|
| 1745 | }
|
| 1746 |
|
| 1747 | static gifconf_func_t * gifconf_list [NPROTO];
|
| 1748 |
|
| 1749 | |
| 1750 | * register_gifconf - register a SIOCGIF handler
|
| 1751 | * @family: Address family
|
| 1752 | * @gifconf: Function handler
|
| 1753 | *
|
| 1754 | * Register protocol dependent address dumping routines. The handler
|
| 1755 | * that is passed must not be freed or reused until it has been replaced
|
| 1756 | * by another handler.
|
| 1757 |
|
| 1758 | int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
|
| 1759 | {
|
| 1760 | if (family >= NPROTO)
|
| 1761 | return -EINVAL;
|
| 1762 | gifconf_list[family] = gifconf;
|
| 1763 | return 0;
|
| 1764 | }
|
| 1765 |
|
| 1766 |
|
| 1767 | |
| 1768 | * Map an interface index to its name (SIOCGIFNAME)
|
| 1769 |
|
| 1770 |
|
| 1771 | |
| 1772 | * We need this ioctl for efficient implementation of the
|
| 1773 | * if_indextoname() function required by the IPv6 API. Without
|
| 1774 | * it, we would have to search all the interfaces to find a
|
| 1775 | * match. --pb
|
| 1776 |
|
| 1777 |
|
| 1778 | static int dev_ifname(struct ifreq __user *arg)
|
| 1779 | {
|
| 1780 | struct net_device *dev;
|
| 1781 | struct ifreq ifr;
|
| 1782 |
|
| 1783 | |
| 1784 | * Fetch the caller's info block.
|
| 1785 |
|
| 1786 |
|
| 1787 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
|
| 1788 | return -EFAULT;
|
| 1789 |
|
| 1790 | read_lock(&dev_base_lock);
|
| 1791 | dev = __dev_get_by_index(ifr.ifr_ifindex);
|
| 1792 | if (!dev) {
|
| 1793 | read_unlock(&dev_base_lock);
|
| 1794 | return -ENODEV;
|
| 1795 | }
|
| 1796 |
|
| 1797 | strcpy(ifr.ifr_name, dev->name);
|
| 1798 | read_unlock(&dev_base_lock);
|
| 1799 |
|
| 1800 | if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
|
| 1801 | return -EFAULT;
|
| 1802 | return 0;
|
| 1803 | }
|
| 1804 |
|
| 1805 | |
| 1806 | * Perform a SIOCGIFCONF call. This structure will change
|
| 1807 | * size eventually, and there is nothing I can do about it.
|
| 1808 | * Thus we will need a 'compatibility mode'.
|
| 1809 |
|
| 1810 |
|
| 1811 | static int dev_ifconf(char __user *arg)
|
| 1812 | {
|
| 1813 | struct ifconf ifc;
|
| 1814 | struct net_device *dev;
|
| 1815 | char __user *pos;
|
| 1816 | int len;
|
| 1817 | int total;
|
| 1818 | int i;
|
| 1819 |
|
| 1820 | |
| 1821 | * Fetch the caller's info block.
|
| 1822 |
|
| 1823 |
|
| 1824 | if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
|
| 1825 | return -EFAULT;
|
| 1826 |
|
| 1827 | pos = ifc.ifc_buf;
|
| 1828 | len = ifc.ifc_len;
|
| 1829 |
|
| 1830 | |
| 1831 | * Loop over the interfaces, and write an info block for each.
|
| 1832 |
|
| 1833 |
|
| 1834 | total = 0;
|
| 1835 | for (dev = dev_base; dev; dev = dev->next) {
|
| 1836 | for (i = 0; i < NPROTO; i++) {
|
| 1837 | if (gifconf_list[i]) {
|
| 1838 | int done;
|
| 1839 | if (!pos)
|
| 1840 | done = gifconf_list[i](dev, NULL, 0);
|
| 1841 | else
|
| 1842 | done = gifconf_list[i](dev, pos + total,
|
| 1843 | len - total);
|
| 1844 | if (done < 0)
|
| 1845 | return -EFAULT;
|
| 1846 | total += done;
|
| 1847 | }
|
| 1848 | }
|
| 1849 | }
|
| 1850 |
|
| 1851 | |
| 1852 | * All done. Write the updated control block back to the caller.
|
| 1853 |
|
| 1854 | ifc.ifc_len = total;
|
| 1855 |
|
| 1856 | |
| 1857 | * Both BSD and Solaris return 0 here, so we do too.
|
| 1858 |
|
| 1859 | return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
|
| 1860 | }
|
| 1861 |
|
| 1862 | #ifdef CONFIG_PROC_FS
|
| 1863 | |
| 1864 | * This is invoked by the /proc filesystem handler to display a device
|
| 1865 | * in detail.
|
| 1866 |
|
| 1867 | static __inline__ struct net_device *dev_get_idx(loff_t pos)
|
| 1868 | {
|
| 1869 | struct net_device *dev;
|
| 1870 | loff_t i;
|
| 1871 |
|
| 1872 | for (i = 0, dev = dev_base; dev && i < pos; ++i, dev = dev->next);
|
| 1873 |
|
| 1874 | return i == pos ? dev : NULL;
|
| 1875 | }
|
| 1876 |
|
| 1877 | void *dev_seq_start(struct seq_file *seq, loff_t *pos)
|
| 1878 | {
|
| 1879 | read_lock(&dev_base_lock);
|
| 1880 | return *pos ? dev_get_idx(*pos - 1) : SEQ_START_TOKEN;
|
| 1881 | }
|
| 1882 |
|
| 1883 | void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
|
| 1884 | {
|
| 1885 | ++*pos;
|
| 1886 | return v == SEQ_START_TOKEN ? dev_base : ((struct net_device *)v)->next;
|
| 1887 | }
|
| 1888 |
|
| 1889 | void dev_seq_stop(struct seq_file *seq, void *v)
|
| 1890 | {
|
| 1891 | read_unlock(&dev_base_lock);
|
| 1892 | }
|
| 1893 |
|
| 1894 | static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
|
| 1895 | {
|
| 1896 | if (dev->get_stats) {
|
| 1897 | struct net_device_stats *stats = dev->get_stats(dev);
|
| 1898 |
|
| 1899 | seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
|
| 1900 | "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
|
| 1901 | dev->name, stats->rx_bytes, stats->rx_packets,
|
| 1902 | stats->rx_errors,
|
| 1903 | stats->rx_dropped + stats->rx_missed_errors,
|
| 1904 | stats->rx_fifo_errors,
|
| 1905 | stats->rx_length_errors + stats->rx_over_errors +
|
| 1906 | stats->rx_crc_errors + stats->rx_frame_errors,
|
| 1907 | stats->rx_compressed, stats->multicast,
|
| 1908 | stats->tx_bytes, stats->tx_packets,
|
| 1909 | stats->tx_errors, stats->tx_dropped,
|
| 1910 | stats->tx_fifo_errors, stats->collisions,
|
| 1911 | stats->tx_carrier_errors +
|
| 1912 | stats->tx_aborted_errors +
|
| 1913 | stats->tx_window_errors +
|
| 1914 | stats->tx_heartbeat_errors,
|
| 1915 | stats->tx_compressed);
|
| 1916 | } else
|
| 1917 | seq_printf(seq, "%6s: No statistics available.\n", dev->name);
|
| 1918 | }
|
| 1919 |
|
| 1920 | |
| 1921 | * Called from the PROCfs module. This now uses the new arbitrary sized
|
| 1922 | * /proc/net interface to create /proc/net/dev
|
| 1923 |
|
| 1924 | static int dev_seq_show(struct seq_file *seq, void *v)
|
| 1925 | {
|
| 1926 | if (v == SEQ_START_TOKEN)
|
| 1927 | seq_puts(seq, "Inter-| Receive "
|
| 1928 | " | Transmit\n"
|
| 1929 | " face |bytes packets errs drop fifo frame "
|
| 1930 | "compressed multicast|bytes packets errs "
|
| 1931 | "drop fifo colls carrier compressed\n");
|
| 1932 | else
|
| 1933 | dev_seq_printf_stats(seq, v);
|
| 1934 | return 0;
|
| 1935 | }
|
| 1936 |
|
| 1937 | static struct netif_rx_stats *softnet_get_online(loff_t *pos)
|
| 1938 | {
|
| 1939 | struct netif_rx_stats *rc = NULL;
|
| 1940 |
|
| 1941 | while (*pos < NR_CPUS)
|
| 1942 | if (cpu_online(*pos)) {
|
| 1943 | rc = &per_cpu(netdev_rx_stat, *pos);
|
| 1944 | break;
|
| 1945 | } else
|
| 1946 | ++*pos;
|
| 1947 | return rc;
|
| 1948 | }
|
| 1949 |
|
| 1950 | static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
|
| 1951 | {
|
| 1952 | return softnet_get_online(pos);
|
| 1953 | }
|
| 1954 |
|
| 1955 | static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
|
| 1956 | {
|
| 1957 | ++*pos;
|
| 1958 | return softnet_get_online(pos);
|
| 1959 | }
|
| 1960 |
|
| 1961 | static void softnet_seq_stop(struct seq_file *seq, void *v)
|
| 1962 | {
|
| 1963 | }
|
| 1964 |
|
| 1965 | static int softnet_seq_show(struct seq_file *seq, void *v)
|
| 1966 | {
|
| 1967 | struct netif_rx_stats *s = v;
|
| 1968 |
|
| 1969 | seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
|
| 1970 | s->total, s->dropped, s->time_squeeze, 0,
|
| 1971 | 0, 0, 0, 0,
|
| 1972 | s->cpu_collision );
|
| 1973 | return 0;
|
| 1974 | }
|
| 1975 |
|
| 1976 | static struct seq_operations dev_seq_ops = {
|
| 1977 | .start = dev_seq_start,
|
| 1978 | .next = dev_seq_next,
|
| 1979 | .stop = dev_seq_stop,
|
| 1980 | .show = dev_seq_show,
|
| 1981 | };
|
| 1982 |
|
| 1983 | static int dev_seq_open(struct inode *inode, struct file *file)
|
| 1984 | {
|
| 1985 | return seq_open(file, &dev_seq_ops);
|
| 1986 | }
|
| 1987 |
|
| 1988 | static struct file_operations dev_seq_fops = {
|
| 1989 | .owner = THIS_MODULE,
|
| 1990 | .open = dev_seq_open,
|
| 1991 | .read = seq_read,
|
| 1992 | .llseek = seq_lseek,
|
| 1993 | .release = seq_release,
|
| 1994 | };
|
| 1995 |
|
| 1996 | static struct seq_operations softnet_seq_ops = {
|
| 1997 | .start = softnet_seq_start,
|
| 1998 | .next = softnet_seq_next,
|
| 1999 | .stop = softnet_seq_stop,
|
| 2000 | .show = softnet_seq_show,
|
| 2001 | };
|
| 2002 |
|
| 2003 | static int softnet_seq_open(struct inode *inode, struct file *file)
|
| 2004 | {
|
| 2005 | return seq_open(file, &softnet_seq_ops);
|
| 2006 | }
|
| 2007 |
|
| 2008 | static struct file_operations softnet_seq_fops = {
|
| 2009 | .owner = THIS_MODULE,
|
| 2010 | .open = softnet_seq_open,
|
| 2011 | .read = seq_read,
|
| 2012 | .llseek = seq_lseek,
|
| 2013 | .release = seq_release,
|
| 2014 | };
|
| 2015 |
|
| 2016 | #ifdef WIRELESS_EXT
|
| 2017 | extern int wireless_proc_init(void);
|
| 2018 | #else
|
| 2019 | #define wireless_proc_init() 0
|
| 2020 | #endif
|
| 2021 |
|
| 2022 | static int __init dev_proc_init(void)
|
| 2023 | {
|
| 2024 | int rc = -ENOMEM;
|
| 2025 |
|
| 2026 | if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
|
| 2027 | goto out;
|
| 2028 | if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
|
| 2029 | goto out_dev;
|
| 2030 | if (wireless_proc_init())
|
| 2031 | goto out_softnet;
|
| 2032 | rc = 0;
|
| 2033 | out:
|
| 2034 | return rc;
|
| 2035 | out_softnet:
|
| 2036 | proc_net_remove("softnet_stat");
|
| 2037 | out_dev:
|
| 2038 | proc_net_remove("dev");
|
| 2039 | goto out;
|
| 2040 | }
|
| 2041 | #else
|
| 2042 | #define dev_proc_init() 0
|
| 2043 | #endif
|
| 2044 |
|
| 2045 |
|
| 2046 | |
| 2047 | * netdev_set_master - set up master/slave pair
|
| 2048 | * @slave: slave device
|
| 2049 | * @master: new master device
|
| 2050 | *
|
| 2051 | * Changes the master device of the slave. Pass %NULL to break the
|
| 2052 | * bonding. The caller must hold the RTNL semaphore. On a failure
|
| 2053 | * a negative errno code is returned. On success the reference counts
|
| 2054 | * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
|
| 2055 | * function returns zero.
|
| 2056 |
|
| 2057 | int netdev_set_master(struct net_device *slave, struct net_device *master)
|
| 2058 | {
|
| 2059 | struct net_device *old = slave->master;
|
| 2060 |
|
| 2061 | ASSERT_RTNL();
|
| 2062 |
|
| 2063 | if (master) {
|
| 2064 | if (old)
|
| 2065 | return -EBUSY;
|
| 2066 | dev_hold(master);
|
| 2067 | }
|
| 2068 |
|
| 2069 | slave->master = master;
|
| 2070 |
|
| 2071 | synchronize_net();
|
| 2072 |
|
| 2073 | if (old)
|
| 2074 | dev_put(old);
|
| 2075 |
|
| 2076 | if (master)
|
| 2077 | slave->flags |= IFF_SLAVE;
|
| 2078 | else
|
| 2079 | slave->flags &= ~IFF_SLAVE;
|
| 2080 |
|
| 2081 | rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
|
| 2082 | return 0;
|
| 2083 | }
|
| 2084 |
|
| 2085 | |
| 2086 | * dev_set_promiscuity - update promiscuity count on a device
|
| 2087 | * @dev: device
|
| 2088 | * @inc: modifier
|
| 2089 | *
|
| 2090 | * Add or remove promsicuity from a device. While the count in the device
|
| 2091 | * remains above zero the interface remains promiscuous. Once it hits zero
|
| 2092 | * the device reverts back to normal filtering operation. A negative inc
|
| 2093 | * value is used to drop promiscuity on the device.
|
| 2094 |
|
| 2095 | void dev_set_promiscuity(struct net_device *dev, int inc)
|
| 2096 | {
|
| 2097 | unsigned short old_flags = dev->flags;
|
| 2098 |
|
| 2099 | if ((dev->promiscuity += inc) == 0)
|
| 2100 | dev->flags &= ~IFF_PROMISC;
|
| 2101 | else
|
| 2102 | dev->flags |= IFF_PROMISC;
|
| 2103 | if (dev->flags != old_flags) {
|
| 2104 | dev_mc_upload(dev);
|
| 2105 | printk(KERN_INFO "device %s %s promiscuous mode\n",
|
| 2106 | dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
|
| 2107 | "left");
|
| 2108 | }
|
| 2109 | }
|
| 2110 |
|
| 2111 | |
| 2112 | * dev_set_allmulti - update allmulti count on a device
|
| 2113 | * @dev: device
|
| 2114 | * @inc: modifier
|
| 2115 | *
|
| 2116 | * Add or remove reception of all multicast frames to a device. While the
|
| 2117 | * count in the device remains above zero the interface remains listening
|
| 2118 | * to all interfaces. Once it hits zero the device reverts back to normal
|
| 2119 | * filtering operation. A negative @inc value is used to drop the counter
|
| 2120 | * when releasing a resource needing all multicasts.
|
| 2121 |
|
| 2122 |
|
| 2123 | void dev_set_allmulti(struct net_device *dev, int inc)
|
| 2124 | {
|
| 2125 | unsigned short old_flags = dev->flags;
|
| 2126 |
|
| 2127 | dev->flags |= IFF_ALLMULTI;
|
| 2128 | if ((dev->allmulti += inc) == 0)
|
| 2129 | dev->flags &= ~IFF_ALLMULTI;
|
| 2130 | if (dev->flags ^ old_flags)
|
| 2131 | dev_mc_upload(dev);
|
| 2132 | }
|
| 2133 |
|
| 2134 | unsigned dev_get_flags(const struct net_device *dev)
|
| 2135 | {
|
| 2136 | unsigned flags;
|
| 2137 |
|
| 2138 | flags = (dev->flags & ~(IFF_PROMISC |
|
| 2139 | IFF_ALLMULTI |
|
| 2140 | IFF_RUNNING)) |
|
| 2141 | (dev->gflags & (IFF_PROMISC |
|
| 2142 | IFF_ALLMULTI));
|
| 2143 |
|
| 2144 | if (netif_running(dev) && netif_carrier_ok(dev))
|
| 2145 | flags |= IFF_RUNNING;
|
| 2146 |
|
| 2147 | return flags;
|
| 2148 | }
|
| 2149 |
|
| 2150 | int dev_change_flags(struct net_device *dev, unsigned flags)
|
| 2151 | {
|
| 2152 | int ret;
|
| 2153 | int old_flags = dev->flags;
|
| 2154 |
|
| 2155 | |
| 2156 | * Set the flags on our device.
|
| 2157 |
|
| 2158 |
|
| 2159 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
|
| 2160 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
|
| 2161 | IFF_AUTOMEDIA)) |
|
| 2162 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
|
| 2163 | IFF_ALLMULTI));
|
| 2164 |
|
| 2165 | |
| 2166 | * Load in the correct multicast list now the flags have changed.
|
| 2167 |
|
| 2168 |
|
| 2169 | dev_mc_upload(dev);
|
| 2170 |
|
| 2171 | |
| 2172 | * Have we downed the interface. We handle IFF_UP ourselves
|
| 2173 | * according to user attempts to set it, rather than blindly
|
| 2174 | * setting it.
|
| 2175 |
|
| 2176 |
|
| 2177 | ret = 0;
|
| 2178 | if ((old_flags ^ flags) & IFF_UP) {
|
| 2179 | ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
|
| 2180 |
|
| 2181 | if (!ret)
|
| 2182 | dev_mc_upload(dev);
|
| 2183 | }
|
| 2184 |
|
| 2185 | if (dev->flags & IFF_UP &&
|
| 2186 | ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
|
| 2187 | IFF_VOLATILE)))
|
| 2188 | notifier_call_chain(&netdev_chain, NETDEV_CHANGE, dev);
|
| 2189 |
|
| 2190 | if ((flags ^ dev->gflags) & IFF_PROMISC) {
|
| 2191 | int inc = (flags & IFF_PROMISC) ? +1 : -1;
|
| 2192 | dev->gflags ^= IFF_PROMISC;
|
| 2193 | dev_set_promiscuity(dev, inc);
|
| 2194 | }
|
| 2195 |
|
| 2196 | |
| 2197 | is important. Some (broken) drivers set IFF_PROMISC, when
|
| 2198 | IFF_ALLMULTI is requested not asking us and not reporting.
|
| 2199 |
|
| 2200 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
|
| 2201 | int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
|
| 2202 | dev->gflags ^= IFF_ALLMULTI;
|
| 2203 | dev_set_allmulti(dev, inc);
|
| 2204 | }
|
| 2205 |
|
| 2206 | if (old_flags ^ dev->flags)
|
| 2207 | rtmsg_ifinfo(RTM_NEWLINK, dev, old_flags ^ dev->flags);
|
| 2208 |
|
| 2209 | return ret;
|
| 2210 | }
|
| 2211 |
|
| 2212 | int dev_set_mtu(struct net_device *dev, int new_mtu)
|
| 2213 | {
|
| 2214 | int err;
|
| 2215 |
|
| 2216 | if (new_mtu == dev->mtu)
|
| 2217 | return 0;
|
| 2218 |
|
| 2219 |
|
| 2220 | if (new_mtu < 0)
|
| 2221 | return -EINVAL;
|
| 2222 |
|
| 2223 | if (!netif_device_present(dev))
|
| 2224 | return -ENODEV;
|
| 2225 |
|
| 2226 | err = 0;
|
| 2227 | if (dev->change_mtu)
|
| 2228 | err = dev->change_mtu(dev, new_mtu);
|
| 2229 | else
|
| 2230 | dev->mtu = new_mtu;
|
| 2231 | if (!err && dev->flags & IFF_UP)
|
| 2232 | notifier_call_chain(&netdev_chain,
|
| 2233 | NETDEV_CHANGEMTU, dev);
|
| 2234 | return err;
|
| 2235 | }
|
| 2236 |
|
| 2237 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
|
| 2238 | {
|
| 2239 | int err;
|
| 2240 |
|
| 2241 | if (!dev->set_mac_address)
|
| 2242 | return -EOPNOTSUPP;
|
| 2243 | if (sa->sa_family != dev->type)
|
| 2244 | return -EINVAL;
|
| 2245 | if (!netif_device_present(dev))
|
| 2246 | return -ENODEV;
|
| 2247 | err = dev->set_mac_address(dev, sa);
|
| 2248 | if (!err)
|
| 2249 | notifier_call_chain(&netdev_chain, NETDEV_CHANGEADDR, dev);
|
| 2250 | return err;
|
| 2251 | }
|
| 2252 |
|
| 2253 | |
| 2254 | * Perform the SIOCxIFxxx calls.
|
| 2255 |
|
| 2256 | static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
|
| 2257 | {
|
| 2258 | int err;
|
| 2259 | struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
|
| 2260 |
|
| 2261 | if (!dev)
|
| 2262 | return -ENODEV;
|
| 2263 |
|
| 2264 | switch (cmd) {
|
| 2265 | case SIOCGIFFLAGS:
|
| 2266 | ifr->ifr_flags = dev_get_flags(dev);
|
| 2267 | return 0;
|
| 2268 |
|
| 2269 | case SIOCSIFFLAGS:
|
| 2270 | return dev_change_flags(dev, ifr->ifr_flags);
|
| 2271 |
|
| 2272 | case SIOCGIFMETRIC: |
| 2273 | (currently unused)
|
| 2274 | ifr->ifr_metric = 0;
|
| 2275 | return 0;
|
| 2276 |
|
| 2277 | case SIOCSIFMETRIC: |
| 2278 | (currently unused)
|
| 2279 | return -EOPNOTSUPP;
|
| 2280 |
|
| 2281 | case SIOCGIFMTU:
|
| 2282 | ifr->ifr_mtu = dev->mtu;
|
| 2283 | return 0;
|
| 2284 |
|
| 2285 | case SIOCSIFMTU:
|
| 2286 | return dev_set_mtu(dev, ifr->ifr_mtu);
|
| 2287 |
|
| 2288 | case SIOCGIFHWADDR:
|
| 2289 | if (!dev->addr_len)
|
| 2290 | memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
|
| 2291 | else
|
| 2292 | memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
|
| 2293 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
|
| 2294 | ifr->ifr_hwaddr.sa_family = dev->type;
|
| 2295 | return 0;
|
| 2296 |
|
| 2297 | case SIOCSIFHWADDR:
|
| 2298 | return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
|
| 2299 |
|
| 2300 | case SIOCSIFHWBROADCAST:
|
| 2301 | if (ifr->ifr_hwaddr.sa_family != dev->type)
|
| 2302 | return -EINVAL;
|
| 2303 | memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
|
| 2304 | min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
|
| 2305 | notifier_call_chain(&netdev_chain,
|
| 2306 | NETDEV_CHANGEADDR, dev);
|
| 2307 | return 0;
|
| 2308 |
|
| 2309 | case SIOCGIFMAP:
|
| 2310 | ifr->ifr_map.mem_start = dev->mem_start;
|
| 2311 | ifr->ifr_map.mem_end = dev->mem_end;
|
| 2312 | ifr->ifr_map.base_addr = dev->base_addr;
|
| 2313 | ifr->ifr_map.irq = dev->irq;
|
| 2314 | ifr->ifr_map.dma = dev->dma;
|
| 2315 | ifr->ifr_map.port = dev->if_port;
|
| 2316 | return 0;
|
| 2317 |
|
| 2318 | case SIOCSIFMAP:
|
| 2319 | if (dev->set_config) {
|
| 2320 | if (!netif_device_present(dev))
|
| 2321 | return -ENODEV;
|
| 2322 | return dev->set_config(dev, &ifr->ifr_map);
|
| 2323 | }
|
| 2324 | return -EOPNOTSUPP;
|
| 2325 |
|
| 2326 | case SIOCADDMULTI:
|
| 2327 | if (!dev->set_multicast_list ||
|
| 2328 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
|
| 2329 | return -EINVAL;
|
| 2330 | if (!netif_device_present(dev))
|
| 2331 | return -ENODEV;
|
| 2332 | return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
|
| 2333 | dev->addr_len, 1);
|
| 2334 |
|
| 2335 | case SIOCDELMULTI:
|
| 2336 | if (!dev->set_multicast_list ||
|
| 2337 | ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
|
| 2338 | return -EINVAL;
|
| 2339 | if (!netif_device_present(dev))
|
| 2340 | return -ENODEV;
|
| 2341 | return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
|
| 2342 | dev->addr_len, 1);
|
| 2343 |
|
| 2344 | case SIOCGIFINDEX:
|
| 2345 | ifr->ifr_ifindex = dev->ifindex;
|
| 2346 | return 0;
|
| 2347 |
|
| 2348 | case SIOCGIFTXQLEN:
|
| 2349 | ifr->ifr_qlen = dev->tx_queue_len;
|
| 2350 | return 0;
|
| 2351 |
|
| 2352 | case SIOCSIFTXQLEN:
|
| 2353 | if (ifr->ifr_qlen < 0)
|
| 2354 | return -EINVAL;
|
| 2355 | dev->tx_queue_len = ifr->ifr_qlen;
|
| 2356 | return 0;
|
| 2357 |
|
| 2358 | case SIOCSIFNAME:
|
| 2359 | ifr->ifr_newname[IFNAMSIZ-1] = '\0';
|
| 2360 | return dev_change_name(dev, ifr->ifr_newname);
|
| 2361 |
|
| 2362 | |
| 2363 | * Unknown or private ioctl
|
| 2364 |
|
| 2365 |
|
| 2366 | default:
|
| 2367 | if ((cmd >= SIOCDEVPRIVATE &&
|
| 2368 | cmd <= SIOCDEVPRIVATE + 15) ||
|
| 2369 | cmd == SIOCBONDENSLAVE ||
|
| 2370 | cmd == SIOCBONDRELEASE ||
|
| 2371 | cmd == SIOCBONDSETHWADDR ||
|
| 2372 | cmd == SIOCBONDSLAVEINFOQUERY ||
|
| 2373 | cmd == SIOCBONDINFOQUERY ||
|
| 2374 | cmd == SIOCBONDCHANGEACTIVE ||
|
| 2375 | cmd == SIOCGMIIPHY ||
|
| 2376 | cmd == SIOCGMIIREG ||
|
| 2377 | cmd == SIOCSMIIREG ||
|
| 2378 | cmd == SIOCBRADDIF ||
|
| 2379 | cmd == SIOCBRDELIF ||
|
| 2380 | cmd == SIOCWANDEV) {
|
| 2381 | err = -EOPNOTSUPP;
|
| 2382 | if (dev->do_ioctl) {
|
| 2383 | if (netif_device_present(dev))
|
| 2384 | err = dev->do_ioctl(dev, ifr,
|
| 2385 | cmd);
|
| 2386 | else
|
| 2387 | err = -ENODEV;
|
| 2388 | }
|
| 2389 | } else
|
| 2390 | err = -EINVAL;
|
| 2391 |
|
| 2392 | }
|
| 2393 | return err;
|
| 2394 | }
|
| 2395 |
|
| 2396 | |
| 2397 | * This function handles all "interface"-type I/O control requests. The actual
|
| 2398 | * 'doing' part of this is dev_ifsioc above.
|
| 2399 |
|
| 2400 |
|
| 2401 | |
| 2402 | * dev_ioctl - network device ioctl
|
| 2403 | * @cmd: command to issue
|
| 2404 | * @arg: pointer to a struct ifreq in user space
|
| 2405 | *
|
| 2406 | * Issue ioctl functions to devices. This is normally called by the
|
| 2407 | * user space syscall interfaces but can sometimes be useful for
|
| 2408 | * other purposes. The return value is the return from the syscall if
|
| 2409 | * positive or a negative errno code on error.
|
| 2410 |
|
| 2411 |
|
| 2412 | int dev_ioctl(unsigned int cmd, void __user *arg)
|
| 2413 | {
|
| 2414 | struct ifreq ifr;
|
| 2415 | int ret;
|
| 2416 | char *colon;
|
| 2417 |
|
| 2418 | |
| 2419 | and requires shared lock, because it sleeps writing
|
| 2420 | to user space.
|
| 2421 |
|
| 2422 |
|
| 2423 | if (cmd == SIOCGIFCONF) {
|
| 2424 | rtnl_shlock();
|
| 2425 | ret = dev_ifconf((char __user *) arg);
|
| 2426 | rtnl_shunlock();
|
| 2427 | return ret;
|
| 2428 | }
|
| 2429 | if (cmd == SIOCGIFNAME)
|
| 2430 | return dev_ifname((struct ifreq __user *)arg);
|
| 2431 |
|
| 2432 | if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
|
| 2433 | return -EFAULT;
|
| 2434 |
|
| 2435 | ifr.ifr_name[IFNAMSIZ-1] = 0;
|
| 2436 |
|
| 2437 | colon = strchr(ifr.ifr_name, ':');
|
| 2438 | if (colon)
|
| 2439 | *colon = 0;
|
| 2440 |
|
| 2441 | |
| 2442 | * See which interface the caller is talking about.
|
| 2443 |
|
| 2444 |
|
| 2445 | switch (cmd) {
|
| 2446 | |
| 2447 | * These ioctl calls:
|
| 2448 | * - can be done by all.
|
| 2449 | * - atomic and do not require locking.
|
| 2450 | * - return a value
|
| 2451 |
|
| 2452 | case SIOCGIFFLAGS:
|
| 2453 | case SIOCGIFMETRIC:
|
| 2454 | case SIOCGIFMTU:
|
| 2455 | case SIOCGIFHWADDR:
|
| 2456 | case SIOCGIFSLAVE:
|
| 2457 | case SIOCGIFMAP:
|
| 2458 | case SIOCGIFINDEX:
|
| 2459 | case SIOCGIFTXQLEN:
|
| 2460 | dev_load(ifr.ifr_name);
|
| 2461 | read_lock(&dev_base_lock);
|
| 2462 | ret = dev_ifsioc(&ifr, cmd);
|
| 2463 | read_unlock(&dev_base_lock);
|
| 2464 | if (!ret) {
|
| 2465 | if (colon)
|
| 2466 | *colon = ':';
|
| 2467 | if (copy_to_user(arg, &ifr,
|
| 2468 | sizeof(struct ifreq)))
|
| 2469 | ret = -EFAULT;
|
| 2470 | }
|
| 2471 | return ret;
|
| 2472 |
|
| 2473 | case SIOCETHTOOL:
|
| 2474 | dev_load(ifr.ifr_name);
|
| 2475 | rtnl_lock();
|
| 2476 | ret = dev_ethtool(&ifr);
|
| 2477 | rtnl_unlock();
|
| 2478 | if (!ret) {
|
| 2479 | if (colon)
|
| 2480 | *colon = ':';
|
| 2481 | if (copy_to_user(arg, &ifr,
|
| 2482 | sizeof(struct ifreq)))
|
| 2483 | ret = -EFAULT;
|
| 2484 | }
|
| 2485 | return ret;
|
| 2486 |
|
| 2487 | |
| 2488 | * These ioctl calls:
|
| 2489 | * - require superuser power.
|
| 2490 | * - require strict serialization.
|
| 2491 | * - return a value
|
| 2492 |
|
| 2493 | case SIOCGMIIPHY:
|
| 2494 | case SIOCGMIIREG:
|
| 2495 | case SIOCSIFNAME:
|
| 2496 | if (!capable(CAP_NET_ADMIN))
|
| 2497 | return -EPERM;
|
| 2498 | dev_load(ifr.ifr_name);
|
| 2499 | rtnl_lock();
|
| 2500 | ret = dev_ifsioc(&ifr, cmd);
|
| 2501 | rtnl_unlock();
|
| 2502 | if (!ret) {
|
| 2503 | if (colon)
|
| 2504 | *colon = ':';
|
| 2505 | if (copy_to_user(arg, &ifr,
|
| 2506 | sizeof(struct ifreq)))
|
| 2507 | ret = -EFAULT;
|
| 2508 | }
|
| 2509 | return ret;
|
| 2510 |
|
| 2511 | |
| 2512 | * These ioctl calls:
|
| 2513 | * - require superuser power.
|
| 2514 | * - require strict serialization.
|
| 2515 | * - do not return a value
|
| 2516 |
|
| 2517 | case SIOCSIFFLAGS:
|
| 2518 | case SIOCSIFMETRIC:
|
| 2519 | case SIOCSIFMTU:
|
| 2520 | case SIOCSIFMAP:
|
| 2521 | case SIOCSIFHWADDR:
|
| 2522 | case SIOCSIFSLAVE:
|
| 2523 | case SIOCADDMULTI:
|
| 2524 | case SIOCDELMULTI:
|
| 2525 | case SIOCSIFHWBROADCAST:
|
| 2526 | case SIOCSIFTXQLEN:
|
| 2527 | case SIOCSMIIREG:
|
| 2528 | case SIOCBONDENSLAVE:
|
| 2529 | case SIOCBONDRELEASE:
|
| 2530 | case SIOCBONDSETHWADDR:
|
| 2531 | case SIOCBONDSLAVEINFOQUERY:
|
| 2532 | case SIOCBONDINFOQUERY:
|
| 2533 | case SIOCBONDCHANGEACTIVE:
|
| 2534 | case SIOCBRADDIF:
|
| 2535 | case SIOCBRDELIF:
|
| 2536 | if (!capable(CAP_NET_ADMIN))
|
| 2537 | return -EPERM;
|
| 2538 | dev_load(ifr.ifr_name);
|
| 2539 | rtnl_lock();
|
| 2540 | ret = dev_ifsioc(&ifr, cmd);
|
| 2541 | rtnl_unlock();
|
| 2542 | return ret;
|
| 2543 |
|
| 2544 | case SIOCGIFMEM:
|
| 2545 | |
| 2546 | * currently do not support it
|
| 2547 | case SIOCSIFMEM:
|
| 2548 | |
| 2549 | * Not applicable in our case
|
| 2550 | case SIOCSIFLINK:
|
| 2551 | return -EINVAL;
|
| 2552 |
|
| 2553 | |
| 2554 | * Unknown or private ioctl.
|
| 2555 |
|
| 2556 | default:
|
| 2557 | if (cmd == SIOCWANDEV ||
|
| 2558 | (cmd >= SIOCDEVPRIVATE &&
|
| 2559 | cmd <= SIOCDEVPRIVATE + 15)) {
|
| 2560 | dev_load(ifr.ifr_name);
|
| 2561 | rtnl_lock();
|
| 2562 | ret = dev_ifsioc(&ifr, cmd);
|
| 2563 | rtnl_unlock();
|
| 2564 | if (!ret && copy_to_user(arg, &ifr,
|
| 2565 | sizeof(struct ifreq)))
|
| 2566 | ret = -EFAULT;
|
| 2567 | return ret;
|
| 2568 | }
|
| 2569 | #ifdef WIRELESS_EXT
|
| 2570 |
|
| 2571 | if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
|
| 2572 | |
| 2573 | * `get the encoding parameters', check if
|
| 2574 | * the user has the right to do it
|
| 2575 | if (IW_IS_SET(cmd) || cmd == SIOCGIWENCODE) {
|
| 2576 | if (!capable(CAP_NET_ADMIN))
|
| 2577 | return -EPERM;
|
| 2578 | }
|
| 2579 | dev_load(ifr.ifr_name);
|
| 2580 | rtnl_lock();
|
| 2581 |
|
| 2582 | ret = wireless_process_ioctl(&ifr, cmd);
|
| 2583 | rtnl_unlock();
|
| 2584 | if (IW_IS_GET(cmd) &&
|
| 2585 | copy_to_user(arg, &ifr,
|
| 2586 | sizeof(struct ifreq)))
|
| 2587 | ret = -EFAULT;
|
| 2588 | return ret;
|
| 2589 | }
|
| 2590 | #endif
|
| 2591 | return -EINVAL;
|
| 2592 | }
|
| 2593 | }
|
| 2594 |
|
| 2595 |
|
| 2596 | |
| 2597 | * dev_new_index - allocate an ifindex
|
| 2598 | *
|
| 2599 | * Returns a suitable unique value for a new device interface
|
| 2600 | * number. The caller must hold the rtnl semaphore or the
|
| 2601 | * dev_base_lock to be sure it remains unique.
|
| 2602 |
|
| 2603 | static int dev_new_index(void)
|
| 2604 | {
|
| 2605 | static int ifindex;
|
| 2606 | for (;;) {
|
| 2607 | if (++ifindex <= 0)
|
| 2608 | ifindex = 1;
|
| 2609 | if (!__dev_get_by_index(ifindex))
|
| 2610 | return ifindex;
|
| 2611 | }
|
| 2612 | }
|
| 2613 |
|
| 2614 | static int dev_boot_phase = 1;
|
| 2615 |
|
| 2616 |
|
| 2617 | static DEFINE_SPINLOCK(net_todo_list_lock);
|
| 2618 | static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
|
| 2619 |
|
| 2620 | static inline void net_set_todo(struct net_device *dev)
|
| 2621 | {
|
| 2622 | spin_lock(&net_todo_list_lock);
|
| 2623 | list_add_tail(&dev->todo_list, &net_todo_list);
|
| 2624 | spin_unlock(&net_todo_list_lock);
|
| 2625 | }
|
| 2626 |
|
| 2627 | |
| 2628 | * register_netdevice - register a network device
|
| 2629 | * @dev: device to register
|
| 2630 | *
|
| 2631 | * Take a completed network device structure and add it to the kernel
|
| 2632 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
|
| 2633 | * chain. 0 is returned on success. A negative errno code is returned
|
| 2634 | * on a failure to set up the device, or if the name is a duplicate.
|
| 2635 | *
|
| 2636 | * Callers must hold the rtnl semaphore. You may want
|
| 2637 | * register_netdev() instead of this.
|
| 2638 | *
|
| 2639 | * BUGS:
|
| 2640 | * The locking appears insufficient to guarantee two parallel registers
|
| 2641 | * will not get the same name.
|
| 2642 |
|
| 2643 |
|
| 2644 | int register_netdevice(struct net_device *dev)
|
| 2645 | {
|
| 2646 | struct hlist_head *head;
|
| 2647 | struct hlist_node *p;
|
| 2648 | int ret;
|
| 2649 |
|
| 2650 | BUG_ON(dev_boot_phase);
|
| 2651 | ASSERT_RTNL();
|
| 2652 |
|
| 2653 |
|
| 2654 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
|
| 2655 |
|
| 2656 | spin_lock_init(&dev->queue_lock);
|
| 2657 | spin_lock_init(&dev->xmit_lock);
|
| 2658 | dev->xmit_lock_owner = -1;
|
| 2659 | #ifdef CONFIG_NET_CLS_ACT
|
| 2660 | spin_lock_init(&dev->ingress_lock);
|
| 2661 | #endif
|
| 2662 |
|
| 2663 | ret = alloc_divert_blk(dev);
|
| 2664 | if (ret)
|
| 2665 | goto out;
|
| 2666 |
|
| 2667 | dev->iflink = -1;
|
| 2668 |
|
| 2669 |
|
| 2670 | if (dev->init) {
|
| 2671 | ret = dev->init(dev);
|
| 2672 | if (ret) {
|
| 2673 | if (ret > 0)
|
| 2674 | ret = -EIO;
|
| 2675 | goto out_err;
|
| 2676 | }
|
| 2677 | }
|
| 2678 |
|
| 2679 | if (!dev_valid_name(dev->name)) {
|
| 2680 | ret = -EINVAL;
|
| 2681 | goto out_err;
|
| 2682 | }
|
| 2683 |
|
| 2684 | dev->ifindex = dev_new_index();
|
| 2685 | if (dev->iflink == -1)
|
| 2686 | dev->iflink = dev->ifindex;
|
| 2687 |
|
| 2688 |
|
| 2689 | head = dev_name_hash(dev->name);
|
| 2690 | hlist_for_each(p, head) {
|
| 2691 | struct net_device *d
|
| 2692 | = hlist_entry(p, struct net_device, name_hlist);
|
| 2693 | if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
|
| 2694 | ret = -EEXIST;
|
| 2695 | goto out_err;
|
| 2696 | }
|
| 2697 | }
|
| 2698 |
|
| 2699 |
|
| 2700 | if ((dev->features & NETIF_F_SG) &&
|
| 2701 | !(dev->features & (NETIF_F_IP_CSUM |
|
| 2702 | NETIF_F_NO_CSUM |
|
| 2703 | NETIF_F_HW_CSUM))) {
|
| 2704 | printk("%s: Dropping NETIF_F_SG since no checksum feature.\n",
|
| 2705 | dev->name);
|
| 2706 | dev->features &= ~NETIF_F_SG;
|
| 2707 | }
|
| 2708 |
|
| 2709 |
|
| 2710 | if ((dev->features & NETIF_F_TSO) &&
|
| 2711 | !(dev->features & NETIF_F_SG)) {
|
| 2712 | printk("%s: Dropping NETIF_F_TSO since no SG feature.\n",
|
| 2713 | dev->name);
|
| 2714 | dev->features &= ~NETIF_F_TSO;
|
| 2715 | }
|
| 2716 |
|
| 2717 | |
| 2718 | * nil rebuild_header routine,
|
| 2719 | * that should be never called and used as just bug trap.
|
| 2720 |
|
| 2721 |
|
| 2722 | if (!dev->rebuild_header)
|
| 2723 | dev->rebuild_header = default_rebuild_header;
|
| 2724 |
|
| 2725 | |
| 2726 | * Default initial state at registry is that the
|
| 2727 | * device is present.
|
| 2728 |
|
| 2729 |
|
| 2730 | set_bit(__LINK_STATE_PRESENT, &dev->state);
|
| 2731 |
|
| 2732 | dev->next = NULL;
|
| 2733 | dev_init_scheduler(dev);
|
| 2734 | write_lock_bh(&dev_base_lock);
|
| 2735 | *dev_tail = dev;
|
| 2736 | dev_tail = &dev->next;
|
| 2737 | hlist_add_head(&dev->name_hlist, head);
|
| 2738 | hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
|
| 2739 | dev_hold(dev);
|
| 2740 | dev->reg_state = NETREG_REGISTERING;
|
| 2741 | write_unlock_bh(&dev_base_lock);
|
| 2742 |
|
| 2743 |
|
| 2744 | notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
|
| 2745 |
|
| 2746 |
|
| 2747 | net_set_todo(dev);
|
| 2748 | ret = 0;
|
| 2749 |
|
| 2750 | out:
|
| 2751 | return ret;
|
| 2752 | out_err:
|
| 2753 | free_divert_blk(dev);
|
| 2754 | goto out;
|
| 2755 | }
|
| 2756 |
|
| 2757 | |
| 2758 | * register_netdev - register a network device
|
| 2759 | * @dev: device to register
|
| 2760 | *
|
| 2761 | * Take a completed network device structure and add it to the kernel
|
| 2762 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
|
| 2763 | * chain. 0 is returned on success. A negative errno code is returned
|
| 2764 | * on a failure to set up the device, or if the name is a duplicate.
|
| 2765 | *
|
| 2766 | * This is a wrapper around register_netdev that takes the rtnl semaphore
|
| 2767 | * and expands the device name if you passed a format string to
|
| 2768 | * alloc_netdev.
|
| 2769 |
|
| 2770 | int register_netdev(struct net_device *dev)
|
| 2771 | {
|
| 2772 | int err;
|
| 2773 |
|
| 2774 | rtnl_lock();
|
| 2775 |
|
| 2776 | |
| 2777 | * If the name is a format string the caller wants us to do a
|
| 2778 | * name allocation.
|
| 2779 |
|
| 2780 | if (strchr(dev->name, '%')) {
|
| 2781 | err = dev_alloc_name(dev, dev->name);
|
| 2782 | if (err < 0)
|
| 2783 | goto out;
|
| 2784 | }
|
| 2785 |
|
| 2786 | |
| 2787 | * Back compatibility hook. Kill this one in 2.5
|
| 2788 |
|
| 2789 | if (dev->name[0] == 0 || dev->name[0] == ' ') {
|
| 2790 | err = dev_alloc_name(dev, "eth%d");
|
| 2791 | if (err < 0)
|
| 2792 | goto out;
|
| 2793 | }
|
| 2794 |
|
| 2795 | err = register_netdevice(dev);
|
| 2796 | out:
|
| 2797 | rtnl_unlock();
|
| 2798 | return err;
|
| 2799 | }
|
| 2800 | EXPORT_SYMBOL(register_netdev);
|
| 2801 |
|
| 2802 | |
| 2803 | * netdev_wait_allrefs - wait until all references are gone.
|
| 2804 | *
|
| 2805 | * This is called when unregistering network devices.
|
| 2806 | *
|
| 2807 | * Any protocol or device that holds a reference should register
|
| 2808 | * for netdevice notification, and cleanup and put back the
|
| 2809 | * reference if they receive an UNREGISTER event.
|
| 2810 | * We can get stuck here if buggy protocols don't correctly
|
| 2811 | * call dev_put.
|
| 2812 |
|
| 2813 | static void netdev_wait_allrefs(struct net_device *dev)
|
| 2814 | {
|
| 2815 | unsigned long rebroadcast_time, warning_time;
|
| 2816 |
|
| 2817 | rebroadcast_time = warning_time = jiffies;
|
| 2818 | while (atomic_read(&dev->refcnt) != 0) {
|
| 2819 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
|
| 2820 | rtnl_shlock();
|
| 2821 |
|
| 2822 |
|
| 2823 | notifier_call_chain(&netdev_chain,
|
| 2824 | NETDEV_UNREGISTER, dev);
|
| 2825 |
|
| 2826 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
|
| 2827 | &dev->state)) {
|
| 2828 | |
| 2829 | * pending on unregister. If this
|
| 2830 | * happens, we simply run the queue
|
| 2831 | * unscheduled, resulting in a noop
|
| 2832 | * for this device.
|
| 2833 |
|
| 2834 | linkwatch_run_queue();
|
| 2835 | }
|
| 2836 |
|
| 2837 | rtnl_shunlock();
|
| 2838 |
|
| 2839 | rebroadcast_time = jiffies;
|
| 2840 | }
|
| 2841 |
|
| 2842 | msleep(250);
|
| 2843 |
|
| 2844 | if (time_after(jiffies, warning_time + 10 * HZ)) {
|
| 2845 | printk(KERN_EMERG "unregister_netdevice: "
|
| 2846 | "waiting for %s to become free. Usage "
|
| 2847 | "count = %d\n",
|
| 2848 | dev->name, atomic_read(&dev->refcnt));
|
| 2849 | warning_time = jiffies;
|
| 2850 | }
|
| 2851 | }
|
| 2852 | }
|
| 2853 |
|
| 2854 | |
| 2855 | *
|
| 2856 | * rtnl_lock();
|
| 2857 | * ...
|
| 2858 | * register_netdevice(x1);
|
| 2859 | * register_netdevice(x2);
|
| 2860 | * ...
|
| 2861 | * unregister_netdevice(y1);
|
| 2862 | * unregister_netdevice(y2);
|
| 2863 | * ...
|
| 2864 | * rtnl_unlock();
|
| 2865 | * free_netdev(y1);
|
| 2866 | * free_netdev(y2);
|
| 2867 | *
|
| 2868 | * We are invoked by rtnl_unlock() after it drops the semaphore.
|
| 2869 | * This allows us to deal with problems:
|
| 2870 | * 1) We can create/delete sysfs objects which invoke hotplug
|
| 2871 | * without deadlocking with linkwatch via keventd.
|
| 2872 | * 2) Since we run with the RTNL semaphore not held, we can sleep
|
| 2873 | * safely in order to wait for the netdev refcnt to drop to zero.
|
| 2874 |
|
| 2875 | static DECLARE_MUTEX(net_todo_run_mutex);
|
| 2876 | void netdev_run_todo(void)
|
| 2877 | {
|
| 2878 | struct list_head list = LIST_HEAD_INIT(list);
|
| 2879 | int err;
|
| 2880 |
|
| 2881 |
|
| 2882 |
|
| 2883 | down(&net_todo_run_mutex);
|
| 2884 |
|
| 2885 | |
| 2886 | * until all unregister events invoked by the local processor
|
| 2887 | * have been completed (either by this todo run, or one on
|
| 2888 | * another cpu).
|
| 2889 |
|
| 2890 | if (list_empty(&net_todo_list))
|
| 2891 | goto out;
|
| 2892 |
|
| 2893 |
|
| 2894 | spin_lock(&net_todo_list_lock);
|
| 2895 | list_splice_init(&net_todo_list, &list);
|
| 2896 | spin_unlock(&net_todo_list_lock);
|
| 2897 |
|
| 2898 | while (!list_empty(&list)) {
|
| 2899 | struct net_device *dev
|
| 2900 | = list_entry(list.next, struct net_device, todo_list);
|
| 2901 | list_del(&dev->todo_list);
|
| 2902 |
|
| 2903 | switch(dev->reg_state) {
|
| 2904 | case NETREG_REGISTERING:
|
| 2905 | err = netdev_register_sysfs(dev);
|
| 2906 | if (err)
|
| 2907 | printk(KERN_ERR "%s: failed sysfs registration (%d)\n",
|
| 2908 | dev->name, err);
|
| 2909 | dev->reg_state = NETREG_REGISTERED;
|
| 2910 | break;
|
| 2911 |
|
| 2912 | case NETREG_UNREGISTERING:
|
| 2913 | netdev_unregister_sysfs(dev);
|
| 2914 | dev->reg_state = NETREG_UNREGISTERED;
|
| 2915 |
|
| 2916 | netdev_wait_allrefs(dev);
|
| 2917 |
|
| 2918 |
|
| 2919 | BUG_ON(atomic_read(&dev->refcnt));
|
| 2920 | BUG_TRAP(!dev->ip_ptr);
|
| 2921 | BUG_TRAP(!dev->ip6_ptr);
|
| 2922 | BUG_TRAP(!dev->dn_ptr);
|
| 2923 |
|
| 2924 |
|
| 2925 | |
| 2926 | * after this 'dev' may point to freed up memory.
|
| 2927 |
|
| 2928 | if (dev->destructor)
|
| 2929 | dev->destructor(dev);
|
| 2930 | break;
|
| 2931 |
|
| 2932 | default:
|
| 2933 | printk(KERN_ERR "network todo '%s' but state %d\n",
|
| 2934 | dev->name, dev->reg_state);
|
| 2935 | break;
|
| 2936 | }
|
| 2937 | }
|
| 2938 |
|
| 2939 | out:
|
| 2940 | up(&net_todo_run_mutex);
|
| 2941 | }
|
| 2942 |
|
| 2943 | |
| 2944 | * alloc_netdev - allocate network device
|
| 2945 | * @sizeof_priv: size of private data to allocate space for
|
| 2946 | * @name: device name format string
|
| 2947 | * @setup: callback to initialize device
|
| 2948 | *
|
| 2949 | * Allocates a struct net_device with private data area for driver use
|
| 2950 | * and performs basic initialization.
|
| 2951 |
|
| 2952 | struct net_device *alloc_netdev(int sizeof_priv, const char *name,
|
| 2953 | void (*setup)(struct net_device *))
|
| 2954 | {
|
| 2955 | void *p;
|
| 2956 | struct net_device *dev;
|
| 2957 | int alloc_size;
|
| 2958 |
|
| 2959 |
|
| 2960 | alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
|
| 2961 | alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
|
| 2962 |
|
| 2963 | p = kmalloc(alloc_size, GFP_KERNEL);
|
| 2964 | if (!p) {
|
| 2965 | printk(KERN_ERR "alloc_dev: Unable to allocate device.\n");
|
| 2966 | return NULL;
|
| 2967 | }
|
| 2968 | memset(p, 0, alloc_size);
|
| 2969 |
|
| 2970 | dev = (struct net_device *)
|
| 2971 | (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
|
| 2972 | dev->padded = (char *)dev - (char *)p;
|
| 2973 |
|
| 2974 | if (sizeof_priv)
|
| 2975 | dev->priv = netdev_priv(dev);
|
| 2976 |
|
| 2977 | setup(dev);
|
| 2978 | strcpy(dev->name, name);
|
| 2979 | return dev;
|
| 2980 | }
|
| 2981 | EXPORT_SYMBOL(alloc_netdev);
|
| 2982 |
|
| 2983 | |
| 2984 | * free_netdev - free network device
|
| 2985 | * @dev: device
|
| 2986 | *
|
| 2987 | * This function does the last stage of destroying an allocated device
|
| 2988 | * interface. The reference to the device object is released.
|
| 2989 | * If this is the last reference then it will be freed.
|
| 2990 |
|
| 2991 | void free_netdev(struct net_device *dev)
|
| 2992 | {
|
| 2993 | #ifdef CONFIG_SYSFS
|
| 2994 |
|
| 2995 | if (dev->reg_state == NETREG_UNINITIALIZED) {
|
| 2996 | kfree((char *)dev - dev->padded);
|
| 2997 | return;
|
| 2998 | }
|
| 2999 |
|
| 3000 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
|
| 3001 | dev->reg_state = NETREG_RELEASED;
|
| 3002 |
|
| 3003 |
|
| 3004 | class_device_put(&dev->class_dev);
|
| 3005 | #else
|
| 3006 | kfree((char *)dev - dev->padded);
|
| 3007 | #endif
|
| 3008 | }
|
| 3009 |
|
| 3010 |
|
| 3011 | void synchronize_net(void)
|
| 3012 | {
|
| 3013 | might_sleep();
|
| 3014 | synchronize_rcu();
|
| 3015 | }
|
| 3016 |
|
| 3017 | |
| 3018 | * unregister_netdevice - remove device from the kernel
|
| 3019 | * @dev: device
|
| 3020 | *
|
| 3021 | * This function shuts down a device interface and removes it
|
| 3022 | * from the kernel tables. On success 0 is returned, on a failure
|
| 3023 | * a negative errno code is returned.
|
| 3024 | *
|
| 3025 | * Callers must hold the rtnl semaphore. You may want
|
| 3026 | * unregister_netdev() instead of this.
|
| 3027 |
|
| 3028 |
|
| 3029 | int unregister_netdevice(struct net_device *dev)
|
| 3030 | {
|
| 3031 | struct net_device *d, **dp;
|
| 3032 |
|
| 3033 | BUG_ON(dev_boot_phase);
|
| 3034 | ASSERT_RTNL();
|
| 3035 |
|
| 3036 |
|
| 3037 | if (dev->reg_state == NETREG_UNINITIALIZED) {
|
| 3038 | printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
|
| 3039 | "was registered\n", dev->name, dev);
|
| 3040 | return -ENODEV;
|
| 3041 | }
|
| 3042 |
|
| 3043 | BUG_ON(dev->reg_state != NETREG_REGISTERED);
|
| 3044 |
|
| 3045 |
|
| 3046 | if (dev->flags & IFF_UP)
|
| 3047 | dev_close(dev);
|
| 3048 |
|
| 3049 |
|
| 3050 | for (dp = &dev_base; (d = *dp) != NULL; dp = &d->next) {
|
| 3051 | if (d == dev) {
|
| 3052 | write_lock_bh(&dev_base_lock);
|
| 3053 | hlist_del(&dev->name_hlist);
|
| 3054 | hlist_del(&dev->index_hlist);
|
| 3055 | if (dev_tail == &dev->next)
|
| 3056 | dev_tail = dp;
|
| 3057 | *dp = d->next;
|
| 3058 | write_unlock_bh(&dev_base_lock);
|
| 3059 | break;
|
| 3060 | }
|
| 3061 | }
|
| 3062 | if (!d) {
|
| 3063 | printk(KERN_ERR "unregister net_device: '%s' not found\n",
|
| 3064 | dev->name);
|
| 3065 | return -ENODEV;
|
| 3066 | }
|
| 3067 |
|
| 3068 | dev->reg_state = NETREG_UNREGISTERING;
|
| 3069 |
|
| 3070 | synchronize_net();
|
| 3071 |
|
| 3072 |
|
| 3073 | dev_shutdown(dev);
|
| 3074 |
|
| 3075 |
|
| 3076 | |
| 3077 | this device. They should clean all the things.
|
| 3078 |
|
| 3079 | notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
|
| 3080 |
|
| 3081 | |
| 3082 | * Flush the multicast chain
|
| 3083 |
|
| 3084 | dev_mc_discard(dev);
|
| 3085 |
|
| 3086 | if (dev->uninit)
|
| 3087 | dev->uninit(dev);
|
| 3088 |
|
| 3089 |
|
| 3090 | BUG_TRAP(!dev->master);
|
| 3091 |
|
| 3092 | free_divert_blk(dev);
|
| 3093 |
|
| 3094 |
|
| 3095 | net_set_todo(dev);
|
| 3096 |
|
| 3097 | synchronize_net();
|
| 3098 |
|
| 3099 | dev_put(dev);
|
| 3100 | return 0;
|
| 3101 | }
|
| 3102 |
|
| 3103 | |
| 3104 | * unregister_netdev - remove device from the kernel
|
| 3105 | * @dev: device
|
| 3106 | *
|
| 3107 | * This function shuts down a device interface and removes it
|
| 3108 | * from the kernel tables. On success 0 is returned, on a failure
|
| 3109 | * a negative errno code is returned.
|
| 3110 | *
|
| 3111 | * This is just a wrapper for unregister_netdevice that takes
|
| 3112 | * the rtnl semaphore. In general you want to use this and not
|
| 3113 | * unregister_netdevice.
|
| 3114 |
|
| 3115 | void unregister_netdev(struct net_device *dev)
|
| 3116 | {
|
| 3117 | rtnl_lock();
|
| 3118 | unregister_netdevice(dev);
|
| 3119 | rtnl_unlock();
|
| 3120 | }
|
| 3121 |
|
| 3122 | EXPORT_SYMBOL(unregister_netdev);
|
| 3123 |
|
| 3124 | #ifdef CONFIG_HOTPLUG_CPU
|
| 3125 | static int dev_cpu_callback(struct notifier_block *nfb,
|
| 3126 | unsigned long action,
|
| 3127 | void *ocpu)
|
| 3128 | {
|
| 3129 | struct sk_buff **list_skb;
|
| 3130 | struct net_device **list_net;
|
| 3131 | struct sk_buff *skb;
|
| 3132 | unsigned int cpu, oldcpu = (unsigned long)ocpu;
|
| 3133 | struct softnet_data *sd, *oldsd;
|
| 3134 |
|
| 3135 | if (action != CPU_DEAD)
|
| 3136 | return NOTIFY_OK;
|
| 3137 |
|
| 3138 | local_irq_disable();
|
| 3139 | cpu = smp_processor_id();
|
| 3140 | sd = &per_cpu(softnet_data, cpu);
|
| 3141 | oldsd = &per_cpu(softnet_data, oldcpu);
|
| 3142 |
|
| 3143 |
|
| 3144 | list_skb = &sd->completion_queue;
|
| 3145 | while (*list_skb)
|
| 3146 | list_skb = &(*list_skb)->next;
|
| 3147 |
|
| 3148 | *list_skb = oldsd->completion_queue;
|
| 3149 | oldsd->completion_queue = NULL;
|
| 3150 |
|
| 3151 |
|
| 3152 | list_net = &sd->output_queue;
|
| 3153 | while (*list_net)
|
| 3154 | list_net = &(*list_net)->next_sched;
|
| 3155 |
|
| 3156 | *list_net = oldsd->output_queue;
|
| 3157 | oldsd->output_queue = NULL;
|
| 3158 |
|
| 3159 | raise_softirq_irqoff(NET_TX_SOFTIRQ);
|
| 3160 | local_irq_enable();
|
| 3161 |
|
| 3162 |
|
| 3163 | while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
|
| 3164 | netif_rx(skb);
|
| 3165 |
|
| 3166 | return NOTIFY_OK;
|
| 3167 | }
|
| 3168 | #endif
|
| 3169 |
|
| 3170 |
|
| 3171 | |
| 3172 | * Initialize the DEV module. At boot time this walks the device list and
|
| 3173 | * unhooks any devices that fail to initialise (normally hardware not
|
| 3174 | * present) and leaves us with a valid list of present and active devices.
|
| 3175 | *
|
| 3176 |
|
| 3177 |
|
| 3178 | |
| 3179 | * This is called single threaded during boot, so no need
|
| 3180 | * to take the rtnl semaphore.
|
| 3181 |
|
| 3182 | static int __init net_dev_init(void)
|
| 3183 | {
|
| 3184 | int i, rc = -ENOMEM;
|
| 3185 |
|
| 3186 | BUG_ON(!dev_boot_phase);
|
| 3187 |
|
| 3188 | net_random_init();
|
| 3189 |
|
| 3190 | if (dev_proc_init())
|
| 3191 | goto out;
|
| 3192 |
|
| 3193 | if (netdev_sysfs_init())
|
| 3194 | goto out;
|
| 3195 |
|
| 3196 | INIT_LIST_HEAD(&ptype_all);
|
| 3197 | for (i = 0; i < 16; i++)
|
| 3198 | INIT_LIST_HEAD(&ptype_base[i]);
|
| 3199 |
|
| 3200 | for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
|
| 3201 | INIT_HLIST_HEAD(&dev_name_head[i]);
|
| 3202 |
|
| 3203 | for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
|
| 3204 | INIT_HLIST_HEAD(&dev_index_head[i]);
|
| 3205 |
|
| 3206 | |
| 3207 | * Initialise the packet receive queues.
|
| 3208 |
|
| 3209 |
|
| 3210 | for (i = 0; i < NR_CPUS; i++) {
|
| 3211 | struct softnet_data *queue;
|
| 3212 |
|
| 3213 | queue = &per_cpu(softnet_data, i);
|
| 3214 | skb_queue_head_init(&queue->input_pkt_queue);
|
| 3215 | queue->completion_queue = NULL;
|
| 3216 | INIT_LIST_HEAD(&queue->poll_list);
|
| 3217 | set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
|
| 3218 | queue->backlog_dev.weight = weight_p;
|
| 3219 | queue->backlog_dev.poll = process_backlog;
|
| 3220 | atomic_set(&queue->backlog_dev.refcnt, 1);
|
| 3221 | }
|
| 3222 |
|
| 3223 | dev_boot_phase = 0;
|
| 3224 |
|
| 3225 | open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
|
| 3226 | open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
|
| 3227 |
|
| 3228 | hotcpu_notifier(dev_cpu_callback, 0);
|
| 3229 | dst_init();
|
| 3230 | dev_mcast_init();
|
| 3231 | rc = 0;
|
| 3232 | out:
|
| 3233 | return rc;
|
| 3234 | }
|
| 3235 |
|
| 3236 | subsys_initcall(net_dev_init);
|
| 3237 |
|
| 3238 | EXPORT_SYMBOL(__dev_get_by_index);
|
| 3239 | EXPORT_SYMBOL(__dev_get_by_name);
|
| 3240 | EXPORT_SYMBOL(__dev_remove_pack);
|
| 3241 | EXPORT_SYMBOL(__skb_linearize);
|
| 3242 | EXPORT_SYMBOL(dev_add_pack);
|
| 3243 | EXPORT_SYMBOL(dev_alloc_name);
|
| 3244 | EXPORT_SYMBOL(dev_close);
|
| 3245 | EXPORT_SYMBOL(dev_get_by_flags);
|
| 3246 | EXPORT_SYMBOL(dev_get_by_index);
|
| 3247 | EXPORT_SYMBOL(dev_get_by_name);
|
| 3248 | EXPORT_SYMBOL(dev_ioctl);
|
| 3249 | EXPORT_SYMBOL(dev_open);
|
| 3250 | EXPORT_SYMBOL(dev_queue_xmit);
|
| 3251 | EXPORT_SYMBOL(dev_remove_pack);
|
| 3252 | EXPORT_SYMBOL(dev_set_allmulti);
|
| 3253 | EXPORT_SYMBOL(dev_set_promiscuity);
|
| 3254 | EXPORT_SYMBOL(dev_change_flags);
|
| 3255 | EXPORT_SYMBOL(dev_set_mtu);
|
| 3256 | EXPORT_SYMBOL(dev_set_mac_address);
|
| 3257 | EXPORT_SYMBOL(free_netdev);
|
| 3258 | EXPORT_SYMBOL(netdev_boot_setup_check);
|
| 3259 | EXPORT_SYMBOL(netdev_set_master);
|
| 3260 | EXPORT_SYMBOL(netdev_state_change);
|
| 3261 | EXPORT_SYMBOL(netif_receive_skb);
|
| 3262 | EXPORT_SYMBOL(netif_rx);
|
| 3263 | EXPORT_SYMBOL(register_gifconf);
|
| 3264 | EXPORT_SYMBOL(register_netdevice);
|
| 3265 | EXPORT_SYMBOL(register_netdevice_notifier);
|
| 3266 | EXPORT_SYMBOL(skb_checksum_help);
|
| 3267 | EXPORT_SYMBOL(synchronize_net);
|
| 3268 | EXPORT_SYMBOL(unregister_netdevice);
|
| 3269 | EXPORT_SYMBOL(unregister_netdevice_notifier);
|
| 3270 | EXPORT_SYMBOL(net_enable_timestamp);
|
| 3271 | EXPORT_SYMBOL(net_disable_timestamp);
|
| 3272 | EXPORT_SYMBOL(dev_get_flags);
|
| 3273 |
|
| 3274 | #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
|
| 3275 | EXPORT_SYMBOL(br_handle_frame_hook);
|
| 3276 | EXPORT_SYMBOL(br_fdb_get_hook);
|
| 3277 | EXPORT_SYMBOL(br_fdb_put_hook);
|
| 3278 | #endif
|
| 3279 |
|
| 3280 | #ifdef CONFIG_KMOD
|
| 3281 | EXPORT_SYMBOL(dev_load);
|
| 3282 | #endif
|
| 3283 |
|
| 3284 | EXPORT_PER_CPU_SYMBOL(softnet_data);
|