1 #include <linux/workqueue.h> 2 #include <linux/rtnetlink.h> 3 #include <linux/cache.h> 4 #include <linux/slab.h> 5 #include <linux/list.h> 6 #include <linux/delay.h> 7 #include <linux/sched.h> 8 #include <linux/idr.h> 9 #include <linux/rculist.h> 10 #include <linux/nsproxy.h> 11 #include <linux/netdevice.h> 12 #include <net/net_namespace.h> 13 #include <net/netns/generic.h> 14 #include <net/rtnetlink.h> 15 16 /* 17 * Our network namespace constructor/destructor lists 18 */ 19 20 static LIST_HEAD(pernet_list); 21 static struct list_head *first_device = &pernet_list; 22 static DEFINE_MUTEX(net_mutex); 23 24 LIST_HEAD(net_namespace_list); 25 EXPORT_SYMBOL_GPL(net_namespace_list); 26 27 struct net init_net; 28 EXPORT_SYMBOL(init_net); 29 30 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ 31 32 static void unregister_netdevices(struct net *net, struct list_head *list) 33 { 34 struct net_device *dev; 35 /* At exit all network devices most be removed from a network 36 * namespace. Do this in the reverse order of registeration. 37 */ 38 for_each_netdev_reverse(net, dev) { 39 if (dev->rtnl_link_ops) 40 dev->rtnl_link_ops->dellink(dev, list); 41 else 42 unregister_netdevice_queue(dev, list); 43 } 44 } 45 46 /* 47 * setup_net runs the initializers for the network namespace object. 48 */ 49 static __net_init int setup_net(struct net *net) 50 { 51 /* Must be called with net_mutex held */ 52 struct pernet_operations *ops; 53 int error = 0; 54 55 atomic_set(&net->count, 1); 56 57 #ifdef NETNS_REFCNT_DEBUG 58 atomic_set(&net->use_count, 0); 59 #endif 60 61 list_for_each_entry(ops, &pernet_list, list) { 62 if (ops->init) { 63 error = ops->init(net); 64 if (error < 0) 65 goto out_undo; 66 } 67 } 68 out: 69 return error; 70 71 out_undo: 72 /* Walk through the list backwards calling the exit functions 73 * for the pernet modules whose init functions did not fail. 74 */ 75 list_for_each_entry_continue_reverse(ops, &pernet_list, list) { 76 if (ops->exit) 77 ops->exit(net); 78 if (&ops->list == first_device) { 79 LIST_HEAD(dev_kill_list); 80 rtnl_lock(); 81 unregister_netdevices(net, &dev_kill_list); 82 unregister_netdevice_many(&dev_kill_list); 83 rtnl_unlock(); 84 } 85 } 86 87 rcu_barrier(); 88 goto out; 89 } 90 91 static struct net_generic *net_alloc_generic(void) 92 { 93 struct net_generic *ng; 94 size_t generic_size = sizeof(struct net_generic) + 95 INITIAL_NET_GEN_PTRS * sizeof(void *); 96 97 ng = kzalloc(generic_size, GFP_KERNEL); 98 if (ng) 99 ng->len = INITIAL_NET_GEN_PTRS; 100 101 return ng; 102 } 103 104 #ifdef CONFIG_NET_NS 105 static struct kmem_cache *net_cachep; 106 static struct workqueue_struct *netns_wq; 107 108 static struct net *net_alloc(void) 109 { 110 struct net *net = NULL; 111 struct net_generic *ng; 112 113 ng = net_alloc_generic(); 114 if (!ng) 115 goto out; 116 117 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); 118 if (!net) 119 goto out_free; 120 121 rcu_assign_pointer(net->gen, ng); 122 out: 123 return net; 124 125 out_free: 126 kfree(ng); 127 goto out; 128 } 129 130 static void net_free(struct net *net) 131 { 132 #ifdef NETNS_REFCNT_DEBUG 133 if (unlikely(atomic_read(&net->use_count) != 0)) { 134 printk(KERN_EMERG "network namespace not free! Usage: %d\n", 135 atomic_read(&net->use_count)); 136 return; 137 } 138 #endif 139 kfree(net->gen); 140 kmem_cache_free(net_cachep, net); 141 } 142 143 static struct net *net_create(void) 144 { 145 struct net *net; 146 int rv; 147 148 net = net_alloc(); 149 if (!net) 150 return ERR_PTR(-ENOMEM); 151 mutex_lock(&net_mutex); 152 rv = setup_net(net); 153 if (rv == 0) { 154 rtnl_lock(); 155 list_add_tail_rcu(&net->list, &net_namespace_list); 156 rtnl_unlock(); 157 } 158 mutex_unlock(&net_mutex); 159 if (rv < 0) { 160 net_free(net); 161 return ERR_PTR(rv); 162 } 163 return net; 164 } 165 166 struct net *copy_net_ns(unsigned long flags, struct net *old_net) 167 { 168 if (!(flags & CLONE_NEWNET)) 169 return get_net(old_net); 170 return net_create(); 171 } 172 173 static DEFINE_SPINLOCK(cleanup_list_lock); 174 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */ 175 176 static void cleanup_net(struct work_struct *work) 177 { 178 struct pernet_operations *ops; 179 struct net *net, *tmp; 180 LIST_HEAD(net_kill_list); 181 182 /* Atomically snapshot the list of namespaces to cleanup */ 183 spin_lock_irq(&cleanup_list_lock); 184 list_replace_init(&cleanup_list, &net_kill_list); 185 spin_unlock_irq(&cleanup_list_lock); 186 187 mutex_lock(&net_mutex); 188 189 /* Don't let anyone else find us. */ 190 rtnl_lock(); 191 list_for_each_entry(net, &net_kill_list, cleanup_list) 192 list_del_rcu(&net->list); 193 rtnl_unlock(); 194 195 /* 196 * Another CPU might be rcu-iterating the list, wait for it. 197 * This needs to be before calling the exit() notifiers, so 198 * the rcu_barrier() below isn't sufficient alone. 199 */ 200 synchronize_rcu(); 201 202 /* Run all of the network namespace exit methods */ 203 list_for_each_entry_reverse(ops, &pernet_list, list) { 204 if (ops->exit) { 205 list_for_each_entry(net, &net_kill_list, cleanup_list) 206 ops->exit(net); 207 } 208 if (&ops->list == first_device) { 209 LIST_HEAD(dev_kill_list); 210 rtnl_lock(); 211 list_for_each_entry(net, &net_kill_list, cleanup_list) 212 unregister_netdevices(net, &dev_kill_list); 213 unregister_netdevice_many(&dev_kill_list); 214 rtnl_unlock(); 215 } 216 } 217 218 mutex_unlock(&net_mutex); 219 220 /* Ensure there are no outstanding rcu callbacks using this 221 * network namespace. 222 */ 223 rcu_barrier(); 224 225 /* Finally it is safe to free my network namespace structure */ 226 list_for_each_entry_safe(net, tmp, &net_kill_list, cleanup_list) { 227 list_del_init(&net->cleanup_list); 228 net_free(net); 229 } 230 } 231 static DECLARE_WORK(net_cleanup_work, cleanup_net); 232 233 void __put_net(struct net *net) 234 { 235 /* Cleanup the network namespace in process context */ 236 unsigned long flags; 237 238 spin_lock_irqsave(&cleanup_list_lock, flags); 239 list_add(&net->cleanup_list, &cleanup_list); 240 spin_unlock_irqrestore(&cleanup_list_lock, flags); 241 242 queue_work(netns_wq, &net_cleanup_work); 243 } 244 EXPORT_SYMBOL_GPL(__put_net); 245 246 #else 247 struct net *copy_net_ns(unsigned long flags, struct net *old_net) 248 { 249 if (flags & CLONE_NEWNET) 250 return ERR_PTR(-EINVAL); 251 return old_net; 252 } 253 #endif 254 255 struct net *get_net_ns_by_pid(pid_t pid) 256 { 257 struct task_struct *tsk; 258 struct net *net; 259 260 /* Lookup the network namespace */ 261 net = ERR_PTR(-ESRCH); 262 rcu_read_lock(); 263 tsk = find_task_by_vpid(pid); 264 if (tsk) { 265 struct nsproxy *nsproxy; 266 nsproxy = task_nsproxy(tsk); 267 if (nsproxy) 268 net = get_net(nsproxy->net_ns); 269 } 270 rcu_read_unlock(); 271 return net; 272 } 273 EXPORT_SYMBOL_GPL(get_net_ns_by_pid); 274 275 static int __init net_ns_init(void) 276 { 277 struct net_generic *ng; 278 279 #ifdef CONFIG_NET_NS 280 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), 281 SMP_CACHE_BYTES, 282 SLAB_PANIC, NULL); 283 284 /* Create workqueue for cleanup */ 285 netns_wq = create_singlethread_workqueue("netns"); 286 if (!netns_wq) 287 panic("Could not create netns workq"); 288 #endif 289 290 ng = net_alloc_generic(); 291 if (!ng) 292 panic("Could not allocate generic netns"); 293 294 rcu_assign_pointer(init_net.gen, ng); 295 296 mutex_lock(&net_mutex); 297 if (setup_net(&init_net)) 298 panic("Could not setup the initial network namespace"); 299 300 rtnl_lock(); 301 list_add_tail_rcu(&init_net.list, &net_namespace_list); 302 rtnl_unlock(); 303 304 mutex_unlock(&net_mutex); 305 306 return 0; 307 } 308 309 pure_initcall(net_ns_init); 310 311 #ifdef CONFIG_NET_NS 312 static int register_pernet_operations(struct list_head *list, 313 struct pernet_operations *ops) 314 { 315 struct net *net, *undo_net; 316 int error; 317 318 list_add_tail(&ops->list, list); 319 if (ops->init) { 320 for_each_net(net) { 321 error = ops->init(net); 322 if (error) 323 goto out_undo; 324 } 325 } 326 return 0; 327 328 out_undo: 329 /* If I have an error cleanup all namespaces I initialized */ 330 list_del(&ops->list); 331 if (ops->exit) { 332 for_each_net(undo_net) { 333 if (net_eq(undo_net, net)) 334 goto undone; 335 ops->exit(undo_net); 336 } 337 } 338 undone: 339 return error; 340 } 341 342 static void unregister_pernet_operations(struct pernet_operations *ops) 343 { 344 struct net *net; 345 346 list_del(&ops->list); 347 if (ops->exit) 348 for_each_net(net) 349 ops->exit(net); 350 } 351 352 #else 353 354 static int register_pernet_operations(struct list_head *list, 355 struct pernet_operations *ops) 356 { 357 if (ops->init == NULL) 358 return 0; 359 return ops->init(&init_net); 360 } 361 362 static void unregister_pernet_operations(struct pernet_operations *ops) 363 { 364 if (ops->exit) 365 ops->exit(&init_net); 366 } 367 #endif 368 369 static DEFINE_IDA(net_generic_ids); 370 371 /** 372 * register_pernet_subsys - register a network namespace subsystem 373 * @ops: pernet operations structure for the subsystem 374 * 375 * Register a subsystem which has init and exit functions 376 * that are called when network namespaces are created and 377 * destroyed respectively. 378 * 379 * When registered all network namespace init functions are 380 * called for every existing network namespace. Allowing kernel 381 * modules to have a race free view of the set of network namespaces. 382 * 383 * When a new network namespace is created all of the init 384 * methods are called in the order in which they were registered. 385 * 386 * When a network namespace is destroyed all of the exit methods 387 * are called in the reverse of the order with which they were 388 * registered. 389 */ 390 int register_pernet_subsys(struct pernet_operations *ops) 391 { 392 int error; 393 mutex_lock(&net_mutex); 394 error = register_pernet_operations(first_device, ops); 395 mutex_unlock(&net_mutex); 396 return error; 397 } 398 EXPORT_SYMBOL_GPL(register_pernet_subsys); 399 400 /** 401 * unregister_pernet_subsys - unregister a network namespace subsystem 402 * @ops: pernet operations structure to manipulate 403 * 404 * Remove the pernet operations structure from the list to be 405 * used when network namespaces are created or destroyed. In 406 * addition run the exit method for all existing network 407 * namespaces. 408 */ 409 void unregister_pernet_subsys(struct pernet_operations *module) 410 { 411 mutex_lock(&net_mutex); 412 unregister_pernet_operations(module); 413 mutex_unlock(&net_mutex); 414 } 415 EXPORT_SYMBOL_GPL(unregister_pernet_subsys); 416 417 int register_pernet_gen_subsys(int *id, struct pernet_operations *ops) 418 { 419 int rv; 420 421 mutex_lock(&net_mutex); 422 again: 423 rv = ida_get_new_above(&net_generic_ids, 1, id); 424 if (rv < 0) { 425 if (rv == -EAGAIN) { 426 ida_pre_get(&net_generic_ids, GFP_KERNEL); 427 goto again; 428 } 429 goto out; 430 } 431 rv = register_pernet_operations(first_device, ops); 432 if (rv < 0) 433 ida_remove(&net_generic_ids, *id); 434 out: 435 mutex_unlock(&net_mutex); 436 return rv; 437 } 438 EXPORT_SYMBOL_GPL(register_pernet_gen_subsys); 439 440 void unregister_pernet_gen_subsys(int id, struct pernet_operations *ops) 441 { 442 mutex_lock(&net_mutex); 443 unregister_pernet_operations(ops); 444 ida_remove(&net_generic_ids, id); 445 mutex_unlock(&net_mutex); 446 } 447 EXPORT_SYMBOL_GPL(unregister_pernet_gen_subsys); 448 449 /** 450 * register_pernet_device - register a network namespace device 451 * @ops: pernet operations structure for the subsystem 452 * 453 * Register a device which has init and exit functions 454 * that are called when network namespaces are created and 455 * destroyed respectively. 456 * 457 * When registered all network namespace init functions are 458 * called for every existing network namespace. Allowing kernel 459 * modules to have a race free view of the set of network namespaces. 460 * 461 * When a new network namespace is created all of the init 462 * methods are called in the order in which they were registered. 463 * 464 * When a network namespace is destroyed all of the exit methods 465 * are called in the reverse of the order with which they were 466 * registered. 467 */ 468 int register_pernet_device(struct pernet_operations *ops) 469 { 470 int error; 471 mutex_lock(&net_mutex); 472 error = register_pernet_operations(&pernet_list, ops); 473 if (!error && (first_device == &pernet_list)) 474 first_device = &ops->list; 475 mutex_unlock(&net_mutex); 476 return error; 477 } 478 EXPORT_SYMBOL_GPL(register_pernet_device); 479 480 int register_pernet_gen_device(int *id, struct pernet_operations *ops) 481 { 482 int error; 483 mutex_lock(&net_mutex); 484 again: 485 error = ida_get_new_above(&net_generic_ids, 1, id); 486 if (error) { 487 if (error == -EAGAIN) { 488 ida_pre_get(&net_generic_ids, GFP_KERNEL); 489 goto again; 490 } 491 goto out; 492 } 493 error = register_pernet_operations(&pernet_list, ops); 494 if (error) 495 ida_remove(&net_generic_ids, *id); 496 else if (first_device == &pernet_list) 497 first_device = &ops->list; 498 out: 499 mutex_unlock(&net_mutex); 500 return error; 501 } 502 EXPORT_SYMBOL_GPL(register_pernet_gen_device); 503 504 /** 505 * unregister_pernet_device - unregister a network namespace netdevice 506 * @ops: pernet operations structure to manipulate 507 * 508 * Remove the pernet operations structure from the list to be 509 * used when network namespaces are created or destroyed. In 510 * addition run the exit method for all existing network 511 * namespaces. 512 */ 513 void unregister_pernet_device(struct pernet_operations *ops) 514 { 515 mutex_lock(&net_mutex); 516 if (&ops->list == first_device) 517 first_device = first_device->next; 518 unregister_pernet_operations(ops); 519 mutex_unlock(&net_mutex); 520 } 521 EXPORT_SYMBOL_GPL(unregister_pernet_device); 522 523 void unregister_pernet_gen_device(int id, struct pernet_operations *ops) 524 { 525 mutex_lock(&net_mutex); 526 if (&ops->list == first_device) 527 first_device = first_device->next; 528 unregister_pernet_operations(ops); 529 ida_remove(&net_generic_ids, id); 530 mutex_unlock(&net_mutex); 531 } 532 EXPORT_SYMBOL_GPL(unregister_pernet_gen_device); 533 534 static void net_generic_release(struct rcu_head *rcu) 535 { 536 struct net_generic *ng; 537 538 ng = container_of(rcu, struct net_generic, rcu); 539 kfree(ng); 540 } 541 542 int net_assign_generic(struct net *net, int id, void *data) 543 { 544 struct net_generic *ng, *old_ng; 545 546 BUG_ON(!mutex_is_locked(&net_mutex)); 547 BUG_ON(id == 0); 548 549 ng = old_ng = net->gen; 550 if (old_ng->len >= id) 551 goto assign; 552 553 ng = kzalloc(sizeof(struct net_generic) + 554 id * sizeof(void *), GFP_KERNEL); 555 if (ng == NULL) 556 return -ENOMEM; 557 558 /* 559 * Some synchronisation notes: 560 * 561 * The net_generic explores the net->gen array inside rcu 562 * read section. Besides once set the net->gen->ptr[x] 563 * pointer never changes (see rules in netns/generic.h). 564 * 565 * That said, we simply duplicate this array and schedule 566 * the old copy for kfree after a grace period. 567 */ 568 569 ng->len = id; 570 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*)); 571 572 rcu_assign_pointer(net->gen, ng); 573 call_rcu(&old_ng->rcu, net_generic_release); 574 assign: 575 ng->ptr[id - 1] = data; 576 return 0; 577 } 578 EXPORT_SYMBOL_GPL(net_assign_generic); 579