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