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/fs.h> 14 #include <linux/proc_ns.h> 15 #include <linux/file.h> 16 #include <linux/export.h> 17 #include <linux/user_namespace.h> 18 #include <linux/net_namespace.h> 19 #include <linux/rtnetlink.h> 20 #include <net/sock.h> 21 #include <net/netlink.h> 22 #include <net/net_namespace.h> 23 #include <net/netns/generic.h> 24 25 /* 26 * Our network namespace constructor/destructor lists 27 */ 28 29 static LIST_HEAD(pernet_list); 30 static struct list_head *first_device = &pernet_list; 31 DEFINE_MUTEX(net_mutex); 32 33 LIST_HEAD(net_namespace_list); 34 EXPORT_SYMBOL_GPL(net_namespace_list); 35 36 struct net init_net = { 37 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head), 38 }; 39 EXPORT_SYMBOL(init_net); 40 41 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ 42 43 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS; 44 45 static struct net_generic *net_alloc_generic(void) 46 { 47 struct net_generic *ng; 48 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]); 49 50 ng = kzalloc(generic_size, GFP_KERNEL); 51 if (ng) 52 ng->len = max_gen_ptrs; 53 54 return ng; 55 } 56 57 static int net_assign_generic(struct net *net, int id, void *data) 58 { 59 struct net_generic *ng, *old_ng; 60 61 BUG_ON(!mutex_is_locked(&net_mutex)); 62 BUG_ON(id == 0); 63 64 old_ng = rcu_dereference_protected(net->gen, 65 lockdep_is_held(&net_mutex)); 66 ng = old_ng; 67 if (old_ng->len >= id) 68 goto assign; 69 70 ng = net_alloc_generic(); 71 if (ng == NULL) 72 return -ENOMEM; 73 74 /* 75 * Some synchronisation notes: 76 * 77 * The net_generic explores the net->gen array inside rcu 78 * read section. Besides once set the net->gen->ptr[x] 79 * pointer never changes (see rules in netns/generic.h). 80 * 81 * That said, we simply duplicate this array and schedule 82 * the old copy for kfree after a grace period. 83 */ 84 85 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*)); 86 87 rcu_assign_pointer(net->gen, ng); 88 kfree_rcu(old_ng, rcu); 89 assign: 90 ng->ptr[id - 1] = data; 91 return 0; 92 } 93 94 static int ops_init(const struct pernet_operations *ops, struct net *net) 95 { 96 int err = -ENOMEM; 97 void *data = NULL; 98 99 if (ops->id && ops->size) { 100 data = kzalloc(ops->size, GFP_KERNEL); 101 if (!data) 102 goto out; 103 104 err = net_assign_generic(net, *ops->id, data); 105 if (err) 106 goto cleanup; 107 } 108 err = 0; 109 if (ops->init) 110 err = ops->init(net); 111 if (!err) 112 return 0; 113 114 cleanup: 115 kfree(data); 116 117 out: 118 return err; 119 } 120 121 static void ops_free(const struct pernet_operations *ops, struct net *net) 122 { 123 if (ops->id && ops->size) { 124 int id = *ops->id; 125 kfree(net_generic(net, id)); 126 } 127 } 128 129 static void ops_exit_list(const struct pernet_operations *ops, 130 struct list_head *net_exit_list) 131 { 132 struct net *net; 133 if (ops->exit) { 134 list_for_each_entry(net, net_exit_list, exit_list) 135 ops->exit(net); 136 } 137 if (ops->exit_batch) 138 ops->exit_batch(net_exit_list); 139 } 140 141 static void ops_free_list(const struct pernet_operations *ops, 142 struct list_head *net_exit_list) 143 { 144 struct net *net; 145 if (ops->size && ops->id) { 146 list_for_each_entry(net, net_exit_list, exit_list) 147 ops_free(ops, net); 148 } 149 } 150 151 static int alloc_netid(struct net *net, struct net *peer, int reqid) 152 { 153 int min = 0, max = 0; 154 155 ASSERT_RTNL(); 156 157 if (reqid >= 0) { 158 min = reqid; 159 max = reqid + 1; 160 } 161 162 return idr_alloc(&net->netns_ids, peer, min, max, GFP_KERNEL); 163 } 164 165 /* This function is used by idr_for_each(). If net is equal to peer, the 166 * function returns the id so that idr_for_each() stops. Because we cannot 167 * returns the id 0 (idr_for_each() will not stop), we return the magic value 168 * NET_ID_ZERO (-1) for it. 169 */ 170 #define NET_ID_ZERO -1 171 static int net_eq_idr(int id, void *net, void *peer) 172 { 173 if (net_eq(net, peer)) 174 return id ? : NET_ID_ZERO; 175 return 0; 176 } 177 178 static int __peernet2id(struct net *net, struct net *peer, bool alloc) 179 { 180 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer); 181 182 ASSERT_RTNL(); 183 184 /* Magic value for id 0. */ 185 if (id == NET_ID_ZERO) 186 return 0; 187 if (id > 0) 188 return id; 189 190 if (alloc) 191 return alloc_netid(net, peer, -1); 192 193 return -ENOENT; 194 } 195 196 /* This function returns the id of a peer netns. If no id is assigned, one will 197 * be allocated and returned. 198 */ 199 int peernet2id(struct net *net, struct net *peer) 200 { 201 int id = __peernet2id(net, peer, true); 202 203 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED; 204 } 205 EXPORT_SYMBOL(peernet2id); 206 207 struct net *get_net_ns_by_id(struct net *net, int id) 208 { 209 struct net *peer; 210 211 if (id < 0) 212 return NULL; 213 214 rcu_read_lock(); 215 peer = idr_find(&net->netns_ids, id); 216 if (peer) 217 get_net(peer); 218 rcu_read_unlock(); 219 220 return peer; 221 } 222 223 /* 224 * setup_net runs the initializers for the network namespace object. 225 */ 226 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns) 227 { 228 /* Must be called with net_mutex held */ 229 const struct pernet_operations *ops, *saved_ops; 230 int error = 0; 231 LIST_HEAD(net_exit_list); 232 233 atomic_set(&net->count, 1); 234 atomic_set(&net->passive, 1); 235 net->dev_base_seq = 1; 236 net->user_ns = user_ns; 237 idr_init(&net->netns_ids); 238 239 #ifdef NETNS_REFCNT_DEBUG 240 atomic_set(&net->use_count, 0); 241 #endif 242 243 list_for_each_entry(ops, &pernet_list, list) { 244 error = ops_init(ops, net); 245 if (error < 0) 246 goto out_undo; 247 } 248 out: 249 return error; 250 251 out_undo: 252 /* Walk through the list backwards calling the exit functions 253 * for the pernet modules whose init functions did not fail. 254 */ 255 list_add(&net->exit_list, &net_exit_list); 256 saved_ops = ops; 257 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 258 ops_exit_list(ops, &net_exit_list); 259 260 ops = saved_ops; 261 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 262 ops_free_list(ops, &net_exit_list); 263 264 rcu_barrier(); 265 goto out; 266 } 267 268 269 #ifdef CONFIG_NET_NS 270 static struct kmem_cache *net_cachep; 271 static struct workqueue_struct *netns_wq; 272 273 static struct net *net_alloc(void) 274 { 275 struct net *net = NULL; 276 struct net_generic *ng; 277 278 ng = net_alloc_generic(); 279 if (!ng) 280 goto out; 281 282 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); 283 if (!net) 284 goto out_free; 285 286 rcu_assign_pointer(net->gen, ng); 287 out: 288 return net; 289 290 out_free: 291 kfree(ng); 292 goto out; 293 } 294 295 static void net_free(struct net *net) 296 { 297 #ifdef NETNS_REFCNT_DEBUG 298 if (unlikely(atomic_read(&net->use_count) != 0)) { 299 pr_emerg("network namespace not free! Usage: %d\n", 300 atomic_read(&net->use_count)); 301 return; 302 } 303 #endif 304 kfree(rcu_access_pointer(net->gen)); 305 kmem_cache_free(net_cachep, net); 306 } 307 308 void net_drop_ns(void *p) 309 { 310 struct net *ns = p; 311 if (ns && atomic_dec_and_test(&ns->passive)) 312 net_free(ns); 313 } 314 315 struct net *copy_net_ns(unsigned long flags, 316 struct user_namespace *user_ns, struct net *old_net) 317 { 318 struct net *net; 319 int rv; 320 321 if (!(flags & CLONE_NEWNET)) 322 return get_net(old_net); 323 324 net = net_alloc(); 325 if (!net) 326 return ERR_PTR(-ENOMEM); 327 328 get_user_ns(user_ns); 329 330 mutex_lock(&net_mutex); 331 rv = setup_net(net, user_ns); 332 if (rv == 0) { 333 rtnl_lock(); 334 list_add_tail_rcu(&net->list, &net_namespace_list); 335 rtnl_unlock(); 336 } 337 mutex_unlock(&net_mutex); 338 if (rv < 0) { 339 put_user_ns(user_ns); 340 net_drop_ns(net); 341 return ERR_PTR(rv); 342 } 343 return net; 344 } 345 346 static DEFINE_SPINLOCK(cleanup_list_lock); 347 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */ 348 349 static void cleanup_net(struct work_struct *work) 350 { 351 const struct pernet_operations *ops; 352 struct net *net, *tmp; 353 struct list_head net_kill_list; 354 LIST_HEAD(net_exit_list); 355 356 /* Atomically snapshot the list of namespaces to cleanup */ 357 spin_lock_irq(&cleanup_list_lock); 358 list_replace_init(&cleanup_list, &net_kill_list); 359 spin_unlock_irq(&cleanup_list_lock); 360 361 mutex_lock(&net_mutex); 362 363 /* Don't let anyone else find us. */ 364 rtnl_lock(); 365 list_for_each_entry(net, &net_kill_list, cleanup_list) { 366 list_del_rcu(&net->list); 367 list_add_tail(&net->exit_list, &net_exit_list); 368 for_each_net(tmp) { 369 int id = __peernet2id(tmp, net, false); 370 371 if (id >= 0) 372 idr_remove(&tmp->netns_ids, id); 373 } 374 idr_destroy(&net->netns_ids); 375 376 } 377 rtnl_unlock(); 378 379 /* 380 * Another CPU might be rcu-iterating the list, wait for it. 381 * This needs to be before calling the exit() notifiers, so 382 * the rcu_barrier() below isn't sufficient alone. 383 */ 384 synchronize_rcu(); 385 386 /* Run all of the network namespace exit methods */ 387 list_for_each_entry_reverse(ops, &pernet_list, list) 388 ops_exit_list(ops, &net_exit_list); 389 390 /* Free the net generic variables */ 391 list_for_each_entry_reverse(ops, &pernet_list, list) 392 ops_free_list(ops, &net_exit_list); 393 394 mutex_unlock(&net_mutex); 395 396 /* Ensure there are no outstanding rcu callbacks using this 397 * network namespace. 398 */ 399 rcu_barrier(); 400 401 /* Finally it is safe to free my network namespace structure */ 402 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) { 403 list_del_init(&net->exit_list); 404 put_user_ns(net->user_ns); 405 net_drop_ns(net); 406 } 407 } 408 static DECLARE_WORK(net_cleanup_work, cleanup_net); 409 410 void __put_net(struct net *net) 411 { 412 /* Cleanup the network namespace in process context */ 413 unsigned long flags; 414 415 spin_lock_irqsave(&cleanup_list_lock, flags); 416 list_add(&net->cleanup_list, &cleanup_list); 417 spin_unlock_irqrestore(&cleanup_list_lock, flags); 418 419 queue_work(netns_wq, &net_cleanup_work); 420 } 421 EXPORT_SYMBOL_GPL(__put_net); 422 423 struct net *get_net_ns_by_fd(int fd) 424 { 425 struct file *file; 426 struct ns_common *ns; 427 struct net *net; 428 429 file = proc_ns_fget(fd); 430 if (IS_ERR(file)) 431 return ERR_CAST(file); 432 433 ns = get_proc_ns(file_inode(file)); 434 if (ns->ops == &netns_operations) 435 net = get_net(container_of(ns, struct net, ns)); 436 else 437 net = ERR_PTR(-EINVAL); 438 439 fput(file); 440 return net; 441 } 442 443 #else 444 struct net *get_net_ns_by_fd(int fd) 445 { 446 return ERR_PTR(-EINVAL); 447 } 448 #endif 449 EXPORT_SYMBOL_GPL(get_net_ns_by_fd); 450 451 struct net *get_net_ns_by_pid(pid_t pid) 452 { 453 struct task_struct *tsk; 454 struct net *net; 455 456 /* Lookup the network namespace */ 457 net = ERR_PTR(-ESRCH); 458 rcu_read_lock(); 459 tsk = find_task_by_vpid(pid); 460 if (tsk) { 461 struct nsproxy *nsproxy; 462 task_lock(tsk); 463 nsproxy = tsk->nsproxy; 464 if (nsproxy) 465 net = get_net(nsproxy->net_ns); 466 task_unlock(tsk); 467 } 468 rcu_read_unlock(); 469 return net; 470 } 471 EXPORT_SYMBOL_GPL(get_net_ns_by_pid); 472 473 static __net_init int net_ns_net_init(struct net *net) 474 { 475 #ifdef CONFIG_NET_NS 476 net->ns.ops = &netns_operations; 477 #endif 478 return ns_alloc_inum(&net->ns); 479 } 480 481 static __net_exit void net_ns_net_exit(struct net *net) 482 { 483 ns_free_inum(&net->ns); 484 } 485 486 static struct pernet_operations __net_initdata net_ns_ops = { 487 .init = net_ns_net_init, 488 .exit = net_ns_net_exit, 489 }; 490 491 static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = { 492 [NETNSA_NONE] = { .type = NLA_UNSPEC }, 493 [NETNSA_NSID] = { .type = NLA_S32 }, 494 [NETNSA_PID] = { .type = NLA_U32 }, 495 [NETNSA_FD] = { .type = NLA_U32 }, 496 }; 497 498 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh) 499 { 500 struct net *net = sock_net(skb->sk); 501 struct nlattr *tb[NETNSA_MAX + 1]; 502 struct net *peer; 503 int nsid, err; 504 505 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX, 506 rtnl_net_policy); 507 if (err < 0) 508 return err; 509 if (!tb[NETNSA_NSID]) 510 return -EINVAL; 511 nsid = nla_get_s32(tb[NETNSA_NSID]); 512 513 if (tb[NETNSA_PID]) 514 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 515 else if (tb[NETNSA_FD]) 516 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 517 else 518 return -EINVAL; 519 if (IS_ERR(peer)) 520 return PTR_ERR(peer); 521 522 if (__peernet2id(net, peer, false) >= 0) { 523 err = -EEXIST; 524 goto out; 525 } 526 527 err = alloc_netid(net, peer, nsid); 528 if (err > 0) 529 err = 0; 530 out: 531 put_net(peer); 532 return err; 533 } 534 535 static int rtnl_net_get_size(void) 536 { 537 return NLMSG_ALIGN(sizeof(struct rtgenmsg)) 538 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */ 539 ; 540 } 541 542 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags, 543 int cmd, struct net *net, struct net *peer) 544 { 545 struct nlmsghdr *nlh; 546 struct rtgenmsg *rth; 547 int id; 548 549 ASSERT_RTNL(); 550 551 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags); 552 if (!nlh) 553 return -EMSGSIZE; 554 555 rth = nlmsg_data(nlh); 556 rth->rtgen_family = AF_UNSPEC; 557 558 id = __peernet2id(net, peer, false); 559 if (id < 0) 560 id = NETNSA_NSID_NOT_ASSIGNED; 561 if (nla_put_s32(skb, NETNSA_NSID, id)) 562 goto nla_put_failure; 563 564 nlmsg_end(skb, nlh); 565 return 0; 566 567 nla_put_failure: 568 nlmsg_cancel(skb, nlh); 569 return -EMSGSIZE; 570 } 571 572 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh) 573 { 574 struct net *net = sock_net(skb->sk); 575 struct nlattr *tb[NETNSA_MAX + 1]; 576 struct sk_buff *msg; 577 int err = -ENOBUFS; 578 struct net *peer; 579 580 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX, 581 rtnl_net_policy); 582 if (err < 0) 583 return err; 584 if (tb[NETNSA_PID]) 585 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 586 else if (tb[NETNSA_FD]) 587 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 588 else 589 return -EINVAL; 590 591 if (IS_ERR(peer)) 592 return PTR_ERR(peer); 593 594 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 595 if (!msg) { 596 err = -ENOMEM; 597 goto out; 598 } 599 600 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0, 601 RTM_GETNSID, net, peer); 602 if (err < 0) 603 goto err_out; 604 605 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid); 606 goto out; 607 608 err_out: 609 nlmsg_free(msg); 610 out: 611 put_net(peer); 612 return err; 613 } 614 615 static int __init net_ns_init(void) 616 { 617 struct net_generic *ng; 618 619 #ifdef CONFIG_NET_NS 620 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), 621 SMP_CACHE_BYTES, 622 SLAB_PANIC, NULL); 623 624 /* Create workqueue for cleanup */ 625 netns_wq = create_singlethread_workqueue("netns"); 626 if (!netns_wq) 627 panic("Could not create netns workq"); 628 #endif 629 630 ng = net_alloc_generic(); 631 if (!ng) 632 panic("Could not allocate generic netns"); 633 634 rcu_assign_pointer(init_net.gen, ng); 635 636 mutex_lock(&net_mutex); 637 if (setup_net(&init_net, &init_user_ns)) 638 panic("Could not setup the initial network namespace"); 639 640 rtnl_lock(); 641 list_add_tail_rcu(&init_net.list, &net_namespace_list); 642 rtnl_unlock(); 643 644 mutex_unlock(&net_mutex); 645 646 register_pernet_subsys(&net_ns_ops); 647 648 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL); 649 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, NULL, NULL); 650 651 return 0; 652 } 653 654 pure_initcall(net_ns_init); 655 656 #ifdef CONFIG_NET_NS 657 static int __register_pernet_operations(struct list_head *list, 658 struct pernet_operations *ops) 659 { 660 struct net *net; 661 int error; 662 LIST_HEAD(net_exit_list); 663 664 list_add_tail(&ops->list, list); 665 if (ops->init || (ops->id && ops->size)) { 666 for_each_net(net) { 667 error = ops_init(ops, net); 668 if (error) 669 goto out_undo; 670 list_add_tail(&net->exit_list, &net_exit_list); 671 } 672 } 673 return 0; 674 675 out_undo: 676 /* If I have an error cleanup all namespaces I initialized */ 677 list_del(&ops->list); 678 ops_exit_list(ops, &net_exit_list); 679 ops_free_list(ops, &net_exit_list); 680 return error; 681 } 682 683 static void __unregister_pernet_operations(struct pernet_operations *ops) 684 { 685 struct net *net; 686 LIST_HEAD(net_exit_list); 687 688 list_del(&ops->list); 689 for_each_net(net) 690 list_add_tail(&net->exit_list, &net_exit_list); 691 ops_exit_list(ops, &net_exit_list); 692 ops_free_list(ops, &net_exit_list); 693 } 694 695 #else 696 697 static int __register_pernet_operations(struct list_head *list, 698 struct pernet_operations *ops) 699 { 700 return ops_init(ops, &init_net); 701 } 702 703 static void __unregister_pernet_operations(struct pernet_operations *ops) 704 { 705 LIST_HEAD(net_exit_list); 706 list_add(&init_net.exit_list, &net_exit_list); 707 ops_exit_list(ops, &net_exit_list); 708 ops_free_list(ops, &net_exit_list); 709 } 710 711 #endif /* CONFIG_NET_NS */ 712 713 static DEFINE_IDA(net_generic_ids); 714 715 static int register_pernet_operations(struct list_head *list, 716 struct pernet_operations *ops) 717 { 718 int error; 719 720 if (ops->id) { 721 again: 722 error = ida_get_new_above(&net_generic_ids, 1, ops->id); 723 if (error < 0) { 724 if (error == -EAGAIN) { 725 ida_pre_get(&net_generic_ids, GFP_KERNEL); 726 goto again; 727 } 728 return error; 729 } 730 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id); 731 } 732 error = __register_pernet_operations(list, ops); 733 if (error) { 734 rcu_barrier(); 735 if (ops->id) 736 ida_remove(&net_generic_ids, *ops->id); 737 } 738 739 return error; 740 } 741 742 static void unregister_pernet_operations(struct pernet_operations *ops) 743 { 744 745 __unregister_pernet_operations(ops); 746 rcu_barrier(); 747 if (ops->id) 748 ida_remove(&net_generic_ids, *ops->id); 749 } 750 751 /** 752 * register_pernet_subsys - register a network namespace subsystem 753 * @ops: pernet operations structure for the subsystem 754 * 755 * Register a subsystem which has init and exit functions 756 * that are called when network namespaces are created and 757 * destroyed respectively. 758 * 759 * When registered all network namespace init functions are 760 * called for every existing network namespace. Allowing kernel 761 * modules to have a race free view of the set of network namespaces. 762 * 763 * When a new network namespace is created all of the init 764 * methods are called in the order in which they were registered. 765 * 766 * When a network namespace is destroyed all of the exit methods 767 * are called in the reverse of the order with which they were 768 * registered. 769 */ 770 int register_pernet_subsys(struct pernet_operations *ops) 771 { 772 int error; 773 mutex_lock(&net_mutex); 774 error = register_pernet_operations(first_device, ops); 775 mutex_unlock(&net_mutex); 776 return error; 777 } 778 EXPORT_SYMBOL_GPL(register_pernet_subsys); 779 780 /** 781 * unregister_pernet_subsys - unregister a network namespace subsystem 782 * @ops: pernet operations structure to manipulate 783 * 784 * Remove the pernet operations structure from the list to be 785 * used when network namespaces are created or destroyed. In 786 * addition run the exit method for all existing network 787 * namespaces. 788 */ 789 void unregister_pernet_subsys(struct pernet_operations *ops) 790 { 791 mutex_lock(&net_mutex); 792 unregister_pernet_operations(ops); 793 mutex_unlock(&net_mutex); 794 } 795 EXPORT_SYMBOL_GPL(unregister_pernet_subsys); 796 797 /** 798 * register_pernet_device - register a network namespace device 799 * @ops: pernet operations structure for the subsystem 800 * 801 * Register a device which has init and exit functions 802 * that are called when network namespaces are created and 803 * destroyed respectively. 804 * 805 * When registered all network namespace init functions are 806 * called for every existing network namespace. Allowing kernel 807 * modules to have a race free view of the set of network namespaces. 808 * 809 * When a new network namespace is created all of the init 810 * methods are called in the order in which they were registered. 811 * 812 * When a network namespace is destroyed all of the exit methods 813 * are called in the reverse of the order with which they were 814 * registered. 815 */ 816 int register_pernet_device(struct pernet_operations *ops) 817 { 818 int error; 819 mutex_lock(&net_mutex); 820 error = register_pernet_operations(&pernet_list, ops); 821 if (!error && (first_device == &pernet_list)) 822 first_device = &ops->list; 823 mutex_unlock(&net_mutex); 824 return error; 825 } 826 EXPORT_SYMBOL_GPL(register_pernet_device); 827 828 /** 829 * unregister_pernet_device - unregister a network namespace netdevice 830 * @ops: pernet operations structure to manipulate 831 * 832 * Remove the pernet operations structure from the list to be 833 * used when network namespaces are created or destroyed. In 834 * addition run the exit method for all existing network 835 * namespaces. 836 */ 837 void unregister_pernet_device(struct pernet_operations *ops) 838 { 839 mutex_lock(&net_mutex); 840 if (&ops->list == first_device) 841 first_device = first_device->next; 842 unregister_pernet_operations(ops); 843 mutex_unlock(&net_mutex); 844 } 845 EXPORT_SYMBOL_GPL(unregister_pernet_device); 846 847 #ifdef CONFIG_NET_NS 848 static struct ns_common *netns_get(struct task_struct *task) 849 { 850 struct net *net = NULL; 851 struct nsproxy *nsproxy; 852 853 task_lock(task); 854 nsproxy = task->nsproxy; 855 if (nsproxy) 856 net = get_net(nsproxy->net_ns); 857 task_unlock(task); 858 859 return net ? &net->ns : NULL; 860 } 861 862 static inline struct net *to_net_ns(struct ns_common *ns) 863 { 864 return container_of(ns, struct net, ns); 865 } 866 867 static void netns_put(struct ns_common *ns) 868 { 869 put_net(to_net_ns(ns)); 870 } 871 872 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns) 873 { 874 struct net *net = to_net_ns(ns); 875 876 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) || 877 !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 878 return -EPERM; 879 880 put_net(nsproxy->net_ns); 881 nsproxy->net_ns = get_net(net); 882 return 0; 883 } 884 885 const struct proc_ns_operations netns_operations = { 886 .name = "net", 887 .type = CLONE_NEWNET, 888 .get = netns_get, 889 .put = netns_put, 890 .install = netns_install, 891 }; 892 #endif 893