1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 2 3 #include <linux/workqueue.h> 4 #include <linux/rtnetlink.h> 5 #include <linux/cache.h> 6 #include <linux/slab.h> 7 #include <linux/list.h> 8 #include <linux/delay.h> 9 #include <linux/sched.h> 10 #include <linux/idr.h> 11 #include <linux/rculist.h> 12 #include <linux/nsproxy.h> 13 #include <linux/proc_fs.h> 14 #include <linux/file.h> 15 #include <linux/export.h> 16 #include <linux/user_namespace.h> 17 #include <net/net_namespace.h> 18 #include <net/netns/generic.h> 19 20 /* 21 * Our network namespace constructor/destructor lists 22 */ 23 24 static LIST_HEAD(pernet_list); 25 static struct list_head *first_device = &pernet_list; 26 static DEFINE_MUTEX(net_mutex); 27 28 LIST_HEAD(net_namespace_list); 29 EXPORT_SYMBOL_GPL(net_namespace_list); 30 31 struct net init_net = { 32 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head), 33 }; 34 EXPORT_SYMBOL(init_net); 35 36 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ 37 38 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS; 39 40 static struct net_generic *net_alloc_generic(void) 41 { 42 struct net_generic *ng; 43 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]); 44 45 ng = kzalloc(generic_size, GFP_KERNEL); 46 if (ng) 47 ng->len = max_gen_ptrs; 48 49 return ng; 50 } 51 52 static int net_assign_generic(struct net *net, int id, void *data) 53 { 54 struct net_generic *ng, *old_ng; 55 56 BUG_ON(!mutex_is_locked(&net_mutex)); 57 BUG_ON(id == 0); 58 59 old_ng = rcu_dereference_protected(net->gen, 60 lockdep_is_held(&net_mutex)); 61 ng = old_ng; 62 if (old_ng->len >= id) 63 goto assign; 64 65 ng = net_alloc_generic(); 66 if (ng == NULL) 67 return -ENOMEM; 68 69 /* 70 * Some synchronisation notes: 71 * 72 * The net_generic explores the net->gen array inside rcu 73 * read section. Besides once set the net->gen->ptr[x] 74 * pointer never changes (see rules in netns/generic.h). 75 * 76 * That said, we simply duplicate this array and schedule 77 * the old copy for kfree after a grace period. 78 */ 79 80 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*)); 81 82 rcu_assign_pointer(net->gen, ng); 83 kfree_rcu(old_ng, rcu); 84 assign: 85 ng->ptr[id - 1] = data; 86 return 0; 87 } 88 89 static int ops_init(const struct pernet_operations *ops, struct net *net) 90 { 91 int err = -ENOMEM; 92 void *data = NULL; 93 94 if (ops->id && ops->size) { 95 data = kzalloc(ops->size, GFP_KERNEL); 96 if (!data) 97 goto out; 98 99 err = net_assign_generic(net, *ops->id, data); 100 if (err) 101 goto cleanup; 102 } 103 err = 0; 104 if (ops->init) 105 err = ops->init(net); 106 if (!err) 107 return 0; 108 109 cleanup: 110 kfree(data); 111 112 out: 113 return err; 114 } 115 116 static void ops_free(const struct pernet_operations *ops, struct net *net) 117 { 118 if (ops->id && ops->size) { 119 int id = *ops->id; 120 kfree(net_generic(net, id)); 121 } 122 } 123 124 static void ops_exit_list(const struct pernet_operations *ops, 125 struct list_head *net_exit_list) 126 { 127 struct net *net; 128 if (ops->exit) { 129 list_for_each_entry(net, net_exit_list, exit_list) 130 ops->exit(net); 131 } 132 if (ops->exit_batch) 133 ops->exit_batch(net_exit_list); 134 } 135 136 static void ops_free_list(const struct pernet_operations *ops, 137 struct list_head *net_exit_list) 138 { 139 struct net *net; 140 if (ops->size && ops->id) { 141 list_for_each_entry(net, net_exit_list, exit_list) 142 ops_free(ops, net); 143 } 144 } 145 146 /* 147 * setup_net runs the initializers for the network namespace object. 148 */ 149 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns) 150 { 151 /* Must be called with net_mutex held */ 152 const struct pernet_operations *ops, *saved_ops; 153 int error = 0; 154 LIST_HEAD(net_exit_list); 155 156 atomic_set(&net->count, 1); 157 atomic_set(&net->passive, 1); 158 net->dev_base_seq = 1; 159 net->user_ns = user_ns; 160 161 #ifdef NETNS_REFCNT_DEBUG 162 atomic_set(&net->use_count, 0); 163 #endif 164 165 list_for_each_entry(ops, &pernet_list, list) { 166 error = ops_init(ops, net); 167 if (error < 0) 168 goto out_undo; 169 } 170 out: 171 return error; 172 173 out_undo: 174 /* Walk through the list backwards calling the exit functions 175 * for the pernet modules whose init functions did not fail. 176 */ 177 list_add(&net->exit_list, &net_exit_list); 178 saved_ops = ops; 179 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 180 ops_exit_list(ops, &net_exit_list); 181 182 ops = saved_ops; 183 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 184 ops_free_list(ops, &net_exit_list); 185 186 rcu_barrier(); 187 goto out; 188 } 189 190 191 #ifdef CONFIG_NET_NS 192 static struct kmem_cache *net_cachep; 193 static struct workqueue_struct *netns_wq; 194 195 static struct net *net_alloc(void) 196 { 197 struct net *net = NULL; 198 struct net_generic *ng; 199 200 ng = net_alloc_generic(); 201 if (!ng) 202 goto out; 203 204 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); 205 if (!net) 206 goto out_free; 207 208 rcu_assign_pointer(net->gen, ng); 209 out: 210 return net; 211 212 out_free: 213 kfree(ng); 214 goto out; 215 } 216 217 static void net_free(struct net *net) 218 { 219 #ifdef NETNS_REFCNT_DEBUG 220 if (unlikely(atomic_read(&net->use_count) != 0)) { 221 pr_emerg("network namespace not free! Usage: %d\n", 222 atomic_read(&net->use_count)); 223 return; 224 } 225 #endif 226 kfree(net->gen); 227 kmem_cache_free(net_cachep, net); 228 } 229 230 void net_drop_ns(void *p) 231 { 232 struct net *ns = p; 233 if (ns && atomic_dec_and_test(&ns->passive)) 234 net_free(ns); 235 } 236 237 struct net *copy_net_ns(unsigned long flags, 238 struct user_namespace *user_ns, struct net *old_net) 239 { 240 struct net *net; 241 int rv; 242 243 if (!(flags & CLONE_NEWNET)) 244 return get_net(old_net); 245 246 net = net_alloc(); 247 if (!net) 248 return ERR_PTR(-ENOMEM); 249 250 get_user_ns(user_ns); 251 252 mutex_lock(&net_mutex); 253 rv = setup_net(net, user_ns); 254 if (rv == 0) { 255 rtnl_lock(); 256 list_add_tail_rcu(&net->list, &net_namespace_list); 257 rtnl_unlock(); 258 } 259 mutex_unlock(&net_mutex); 260 if (rv < 0) { 261 put_user_ns(user_ns); 262 net_drop_ns(net); 263 return ERR_PTR(rv); 264 } 265 return net; 266 } 267 268 static DEFINE_SPINLOCK(cleanup_list_lock); 269 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */ 270 271 static void cleanup_net(struct work_struct *work) 272 { 273 const struct pernet_operations *ops; 274 struct net *net, *tmp; 275 LIST_HEAD(net_kill_list); 276 LIST_HEAD(net_exit_list); 277 278 /* Atomically snapshot the list of namespaces to cleanup */ 279 spin_lock_irq(&cleanup_list_lock); 280 list_replace_init(&cleanup_list, &net_kill_list); 281 spin_unlock_irq(&cleanup_list_lock); 282 283 mutex_lock(&net_mutex); 284 285 /* Don't let anyone else find us. */ 286 rtnl_lock(); 287 list_for_each_entry(net, &net_kill_list, cleanup_list) { 288 list_del_rcu(&net->list); 289 list_add_tail(&net->exit_list, &net_exit_list); 290 } 291 rtnl_unlock(); 292 293 /* 294 * Another CPU might be rcu-iterating the list, wait for it. 295 * This needs to be before calling the exit() notifiers, so 296 * the rcu_barrier() below isn't sufficient alone. 297 */ 298 synchronize_rcu(); 299 300 /* Run all of the network namespace exit methods */ 301 list_for_each_entry_reverse(ops, &pernet_list, list) 302 ops_exit_list(ops, &net_exit_list); 303 304 /* Free the net generic variables */ 305 list_for_each_entry_reverse(ops, &pernet_list, list) 306 ops_free_list(ops, &net_exit_list); 307 308 mutex_unlock(&net_mutex); 309 310 /* Ensure there are no outstanding rcu callbacks using this 311 * network namespace. 312 */ 313 rcu_barrier(); 314 315 /* Finally it is safe to free my network namespace structure */ 316 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) { 317 list_del_init(&net->exit_list); 318 put_user_ns(net->user_ns); 319 net_drop_ns(net); 320 } 321 } 322 static DECLARE_WORK(net_cleanup_work, cleanup_net); 323 324 void __put_net(struct net *net) 325 { 326 /* Cleanup the network namespace in process context */ 327 unsigned long flags; 328 329 spin_lock_irqsave(&cleanup_list_lock, flags); 330 list_add(&net->cleanup_list, &cleanup_list); 331 spin_unlock_irqrestore(&cleanup_list_lock, flags); 332 333 queue_work(netns_wq, &net_cleanup_work); 334 } 335 EXPORT_SYMBOL_GPL(__put_net); 336 337 struct net *get_net_ns_by_fd(int fd) 338 { 339 struct proc_inode *ei; 340 struct file *file; 341 struct net *net; 342 343 file = proc_ns_fget(fd); 344 if (IS_ERR(file)) 345 return ERR_CAST(file); 346 347 ei = PROC_I(file->f_dentry->d_inode); 348 if (ei->ns_ops == &netns_operations) 349 net = get_net(ei->ns); 350 else 351 net = ERR_PTR(-EINVAL); 352 353 fput(file); 354 return net; 355 } 356 357 #else 358 struct net *get_net_ns_by_fd(int fd) 359 { 360 return ERR_PTR(-EINVAL); 361 } 362 #endif 363 364 struct net *get_net_ns_by_pid(pid_t pid) 365 { 366 struct task_struct *tsk; 367 struct net *net; 368 369 /* Lookup the network namespace */ 370 net = ERR_PTR(-ESRCH); 371 rcu_read_lock(); 372 tsk = find_task_by_vpid(pid); 373 if (tsk) { 374 struct nsproxy *nsproxy; 375 nsproxy = task_nsproxy(tsk); 376 if (nsproxy) 377 net = get_net(nsproxy->net_ns); 378 } 379 rcu_read_unlock(); 380 return net; 381 } 382 EXPORT_SYMBOL_GPL(get_net_ns_by_pid); 383 384 static __net_init int net_ns_net_init(struct net *net) 385 { 386 return proc_alloc_inum(&net->proc_inum); 387 } 388 389 static __net_exit void net_ns_net_exit(struct net *net) 390 { 391 proc_free_inum(net->proc_inum); 392 } 393 394 static struct pernet_operations __net_initdata net_ns_ops = { 395 .init = net_ns_net_init, 396 .exit = net_ns_net_exit, 397 }; 398 399 static int __init net_ns_init(void) 400 { 401 struct net_generic *ng; 402 403 #ifdef CONFIG_NET_NS 404 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), 405 SMP_CACHE_BYTES, 406 SLAB_PANIC, NULL); 407 408 /* Create workqueue for cleanup */ 409 netns_wq = create_singlethread_workqueue("netns"); 410 if (!netns_wq) 411 panic("Could not create netns workq"); 412 #endif 413 414 ng = net_alloc_generic(); 415 if (!ng) 416 panic("Could not allocate generic netns"); 417 418 rcu_assign_pointer(init_net.gen, ng); 419 420 mutex_lock(&net_mutex); 421 if (setup_net(&init_net, &init_user_ns)) 422 panic("Could not setup the initial network namespace"); 423 424 rtnl_lock(); 425 list_add_tail_rcu(&init_net.list, &net_namespace_list); 426 rtnl_unlock(); 427 428 mutex_unlock(&net_mutex); 429 430 register_pernet_subsys(&net_ns_ops); 431 432 return 0; 433 } 434 435 pure_initcall(net_ns_init); 436 437 #ifdef CONFIG_NET_NS 438 static int __register_pernet_operations(struct list_head *list, 439 struct pernet_operations *ops) 440 { 441 struct net *net; 442 int error; 443 LIST_HEAD(net_exit_list); 444 445 list_add_tail(&ops->list, list); 446 if (ops->init || (ops->id && ops->size)) { 447 for_each_net(net) { 448 error = ops_init(ops, net); 449 if (error) 450 goto out_undo; 451 list_add_tail(&net->exit_list, &net_exit_list); 452 } 453 } 454 return 0; 455 456 out_undo: 457 /* If I have an error cleanup all namespaces I initialized */ 458 list_del(&ops->list); 459 ops_exit_list(ops, &net_exit_list); 460 ops_free_list(ops, &net_exit_list); 461 return error; 462 } 463 464 static void __unregister_pernet_operations(struct pernet_operations *ops) 465 { 466 struct net *net; 467 LIST_HEAD(net_exit_list); 468 469 list_del(&ops->list); 470 for_each_net(net) 471 list_add_tail(&net->exit_list, &net_exit_list); 472 ops_exit_list(ops, &net_exit_list); 473 ops_free_list(ops, &net_exit_list); 474 } 475 476 #else 477 478 static int __register_pernet_operations(struct list_head *list, 479 struct pernet_operations *ops) 480 { 481 return ops_init(ops, &init_net); 482 } 483 484 static void __unregister_pernet_operations(struct pernet_operations *ops) 485 { 486 LIST_HEAD(net_exit_list); 487 list_add(&init_net.exit_list, &net_exit_list); 488 ops_exit_list(ops, &net_exit_list); 489 ops_free_list(ops, &net_exit_list); 490 } 491 492 #endif /* CONFIG_NET_NS */ 493 494 static DEFINE_IDA(net_generic_ids); 495 496 static int register_pernet_operations(struct list_head *list, 497 struct pernet_operations *ops) 498 { 499 int error; 500 501 if (ops->id) { 502 again: 503 error = ida_get_new_above(&net_generic_ids, 1, ops->id); 504 if (error < 0) { 505 if (error == -EAGAIN) { 506 ida_pre_get(&net_generic_ids, GFP_KERNEL); 507 goto again; 508 } 509 return error; 510 } 511 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id); 512 } 513 error = __register_pernet_operations(list, ops); 514 if (error) { 515 rcu_barrier(); 516 if (ops->id) 517 ida_remove(&net_generic_ids, *ops->id); 518 } 519 520 return error; 521 } 522 523 static void unregister_pernet_operations(struct pernet_operations *ops) 524 { 525 526 __unregister_pernet_operations(ops); 527 rcu_barrier(); 528 if (ops->id) 529 ida_remove(&net_generic_ids, *ops->id); 530 } 531 532 /** 533 * register_pernet_subsys - register a network namespace subsystem 534 * @ops: pernet operations structure for the subsystem 535 * 536 * Register a subsystem which has init and exit functions 537 * that are called when network namespaces are created and 538 * destroyed respectively. 539 * 540 * When registered all network namespace init functions are 541 * called for every existing network namespace. Allowing kernel 542 * modules to have a race free view of the set of network namespaces. 543 * 544 * When a new network namespace is created all of the init 545 * methods are called in the order in which they were registered. 546 * 547 * When a network namespace is destroyed all of the exit methods 548 * are called in the reverse of the order with which they were 549 * registered. 550 */ 551 int register_pernet_subsys(struct pernet_operations *ops) 552 { 553 int error; 554 mutex_lock(&net_mutex); 555 error = register_pernet_operations(first_device, ops); 556 mutex_unlock(&net_mutex); 557 return error; 558 } 559 EXPORT_SYMBOL_GPL(register_pernet_subsys); 560 561 /** 562 * unregister_pernet_subsys - unregister a network namespace subsystem 563 * @ops: pernet operations structure to manipulate 564 * 565 * Remove the pernet operations structure from the list to be 566 * used when network namespaces are created or destroyed. In 567 * addition run the exit method for all existing network 568 * namespaces. 569 */ 570 void unregister_pernet_subsys(struct pernet_operations *ops) 571 { 572 mutex_lock(&net_mutex); 573 unregister_pernet_operations(ops); 574 mutex_unlock(&net_mutex); 575 } 576 EXPORT_SYMBOL_GPL(unregister_pernet_subsys); 577 578 /** 579 * register_pernet_device - register a network namespace device 580 * @ops: pernet operations structure for the subsystem 581 * 582 * Register a device which has init and exit functions 583 * that are called when network namespaces are created and 584 * destroyed respectively. 585 * 586 * When registered all network namespace init functions are 587 * called for every existing network namespace. Allowing kernel 588 * modules to have a race free view of the set of network namespaces. 589 * 590 * When a new network namespace is created all of the init 591 * methods are called in the order in which they were registered. 592 * 593 * When a network namespace is destroyed all of the exit methods 594 * are called in the reverse of the order with which they were 595 * registered. 596 */ 597 int register_pernet_device(struct pernet_operations *ops) 598 { 599 int error; 600 mutex_lock(&net_mutex); 601 error = register_pernet_operations(&pernet_list, ops); 602 if (!error && (first_device == &pernet_list)) 603 first_device = &ops->list; 604 mutex_unlock(&net_mutex); 605 return error; 606 } 607 EXPORT_SYMBOL_GPL(register_pernet_device); 608 609 /** 610 * unregister_pernet_device - unregister a network namespace netdevice 611 * @ops: pernet operations structure to manipulate 612 * 613 * Remove the pernet operations structure from the list to be 614 * used when network namespaces are created or destroyed. In 615 * addition run the exit method for all existing network 616 * namespaces. 617 */ 618 void unregister_pernet_device(struct pernet_operations *ops) 619 { 620 mutex_lock(&net_mutex); 621 if (&ops->list == first_device) 622 first_device = first_device->next; 623 unregister_pernet_operations(ops); 624 mutex_unlock(&net_mutex); 625 } 626 EXPORT_SYMBOL_GPL(unregister_pernet_device); 627 628 #ifdef CONFIG_NET_NS 629 static void *netns_get(struct task_struct *task) 630 { 631 struct net *net = NULL; 632 struct nsproxy *nsproxy; 633 634 rcu_read_lock(); 635 nsproxy = task_nsproxy(task); 636 if (nsproxy) 637 net = get_net(nsproxy->net_ns); 638 rcu_read_unlock(); 639 640 return net; 641 } 642 643 static void netns_put(void *ns) 644 { 645 put_net(ns); 646 } 647 648 static int netns_install(struct nsproxy *nsproxy, void *ns) 649 { 650 struct net *net = ns; 651 652 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) || 653 !nsown_capable(CAP_SYS_ADMIN)) 654 return -EPERM; 655 656 put_net(nsproxy->net_ns); 657 nsproxy->net_ns = get_net(net); 658 return 0; 659 } 660 661 static unsigned int netns_inum(void *ns) 662 { 663 struct net *net = ns; 664 return net->proc_inum; 665 } 666 667 const struct proc_ns_operations netns_operations = { 668 .name = "net", 669 .type = CLONE_NEWNET, 670 .get = netns_get, 671 .put = netns_put, 672 .install = netns_install, 673 .inum = netns_inum, 674 }; 675 #endif 676