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