1 // SPDX-License-Identifier: GPL-2.0-only 2 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 3 4 #include <linux/workqueue.h> 5 #include <linux/rtnetlink.h> 6 #include <linux/cache.h> 7 #include <linux/slab.h> 8 #include <linux/list.h> 9 #include <linux/delay.h> 10 #include <linux/sched.h> 11 #include <linux/idr.h> 12 #include <linux/rculist.h> 13 #include <linux/nsproxy.h> 14 #include <linux/fs.h> 15 #include <linux/proc_ns.h> 16 #include <linux/file.h> 17 #include <linux/export.h> 18 #include <linux/user_namespace.h> 19 #include <linux/net_namespace.h> 20 #include <linux/sched/task.h> 21 #include <linux/uidgid.h> 22 #include <linux/cookie.h> 23 #include <linux/proc_fs.h> 24 25 #include <net/sock.h> 26 #include <net/netlink.h> 27 #include <net/net_namespace.h> 28 #include <net/netns/generic.h> 29 30 /* 31 * Our network namespace constructor/destructor lists 32 */ 33 34 static LIST_HEAD(pernet_list); 35 static struct list_head *first_device = &pernet_list; 36 37 LIST_HEAD(net_namespace_list); 38 EXPORT_SYMBOL_GPL(net_namespace_list); 39 40 /* Protects net_namespace_list. Nests iside rtnl_lock() */ 41 DECLARE_RWSEM(net_rwsem); 42 EXPORT_SYMBOL_GPL(net_rwsem); 43 44 #ifdef CONFIG_KEYS 45 static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) }; 46 #endif 47 48 struct net init_net; 49 EXPORT_SYMBOL(init_net); 50 51 static bool init_net_initialized; 52 /* 53 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids, 54 * init_net_initialized and first_device pointer. 55 * This is internal net namespace object. Please, don't use it 56 * outside. 57 */ 58 DECLARE_RWSEM(pernet_ops_rwsem); 59 EXPORT_SYMBOL_GPL(pernet_ops_rwsem); 60 61 #define MIN_PERNET_OPS_ID \ 62 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *)) 63 64 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ 65 66 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS; 67 68 DEFINE_COOKIE(net_cookie); 69 70 static struct net_generic *net_alloc_generic(void) 71 { 72 unsigned int gen_ptrs = READ_ONCE(max_gen_ptrs); 73 unsigned int generic_size; 74 struct net_generic *ng; 75 76 generic_size = offsetof(struct net_generic, ptr[gen_ptrs]); 77 78 ng = kzalloc(generic_size, GFP_KERNEL); 79 if (ng) 80 ng->s.len = gen_ptrs; 81 82 return ng; 83 } 84 85 static int net_assign_generic(struct net *net, unsigned int id, void *data) 86 { 87 struct net_generic *ng, *old_ng; 88 89 BUG_ON(id < MIN_PERNET_OPS_ID); 90 91 old_ng = rcu_dereference_protected(net->gen, 92 lockdep_is_held(&pernet_ops_rwsem)); 93 if (old_ng->s.len > id) { 94 old_ng->ptr[id] = data; 95 return 0; 96 } 97 98 ng = net_alloc_generic(); 99 if (!ng) 100 return -ENOMEM; 101 102 /* 103 * Some synchronisation notes: 104 * 105 * The net_generic explores the net->gen array inside rcu 106 * read section. Besides once set the net->gen->ptr[x] 107 * pointer never changes (see rules in netns/generic.h). 108 * 109 * That said, we simply duplicate this array and schedule 110 * the old copy for kfree after a grace period. 111 */ 112 113 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID], 114 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *)); 115 ng->ptr[id] = data; 116 117 rcu_assign_pointer(net->gen, ng); 118 kfree_rcu(old_ng, s.rcu); 119 return 0; 120 } 121 122 static int ops_init(const struct pernet_operations *ops, struct net *net) 123 { 124 struct net_generic *ng; 125 int err = -ENOMEM; 126 void *data = NULL; 127 128 if (ops->id) { 129 data = kzalloc(ops->size, GFP_KERNEL); 130 if (!data) 131 goto out; 132 133 err = net_assign_generic(net, *ops->id, data); 134 if (err) 135 goto cleanup; 136 } 137 err = 0; 138 if (ops->init) 139 err = ops->init(net); 140 if (!err) 141 return 0; 142 143 if (ops->id) { 144 ng = rcu_dereference_protected(net->gen, 145 lockdep_is_held(&pernet_ops_rwsem)); 146 ng->ptr[*ops->id] = NULL; 147 } 148 149 cleanup: 150 kfree(data); 151 152 out: 153 return err; 154 } 155 156 static void ops_pre_exit_list(const struct pernet_operations *ops, 157 struct list_head *net_exit_list) 158 { 159 struct net *net; 160 161 if (ops->pre_exit) { 162 list_for_each_entry(net, net_exit_list, exit_list) 163 ops->pre_exit(net); 164 } 165 } 166 167 static void ops_exit_list(const struct pernet_operations *ops, 168 struct list_head *net_exit_list) 169 { 170 struct net *net; 171 if (ops->exit) { 172 list_for_each_entry(net, net_exit_list, exit_list) { 173 ops->exit(net); 174 cond_resched(); 175 } 176 } 177 if (ops->exit_batch) 178 ops->exit_batch(net_exit_list); 179 } 180 181 static void ops_free_list(const struct pernet_operations *ops, 182 struct list_head *net_exit_list) 183 { 184 struct net *net; 185 186 if (ops->id) { 187 list_for_each_entry(net, net_exit_list, exit_list) 188 kfree(net_generic(net, *ops->id)); 189 } 190 } 191 192 /* should be called with nsid_lock held */ 193 static int alloc_netid(struct net *net, struct net *peer, int reqid) 194 { 195 int min = 0, max = 0; 196 197 if (reqid >= 0) { 198 min = reqid; 199 max = reqid + 1; 200 } 201 202 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC); 203 } 204 205 /* This function is used by idr_for_each(). If net is equal to peer, the 206 * function returns the id so that idr_for_each() stops. Because we cannot 207 * returns the id 0 (idr_for_each() will not stop), we return the magic value 208 * NET_ID_ZERO (-1) for it. 209 */ 210 #define NET_ID_ZERO -1 211 static int net_eq_idr(int id, void *net, void *peer) 212 { 213 if (net_eq(net, peer)) 214 return id ? : NET_ID_ZERO; 215 return 0; 216 } 217 218 /* Must be called from RCU-critical section or with nsid_lock held */ 219 static int __peernet2id(const struct net *net, struct net *peer) 220 { 221 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer); 222 223 /* Magic value for id 0. */ 224 if (id == NET_ID_ZERO) 225 return 0; 226 if (id > 0) 227 return id; 228 229 return NETNSA_NSID_NOT_ASSIGNED; 230 } 231 232 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid, 233 struct nlmsghdr *nlh, gfp_t gfp); 234 /* This function returns the id of a peer netns. If no id is assigned, one will 235 * be allocated and returned. 236 */ 237 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp) 238 { 239 int id; 240 241 if (refcount_read(&net->ns.count) == 0) 242 return NETNSA_NSID_NOT_ASSIGNED; 243 244 spin_lock_bh(&net->nsid_lock); 245 id = __peernet2id(net, peer); 246 if (id >= 0) { 247 spin_unlock_bh(&net->nsid_lock); 248 return id; 249 } 250 251 /* When peer is obtained from RCU lists, we may race with 252 * its cleanup. Check whether it's alive, and this guarantees 253 * we never hash a peer back to net->netns_ids, after it has 254 * just been idr_remove()'d from there in cleanup_net(). 255 */ 256 if (!maybe_get_net(peer)) { 257 spin_unlock_bh(&net->nsid_lock); 258 return NETNSA_NSID_NOT_ASSIGNED; 259 } 260 261 id = alloc_netid(net, peer, -1); 262 spin_unlock_bh(&net->nsid_lock); 263 264 put_net(peer); 265 if (id < 0) 266 return NETNSA_NSID_NOT_ASSIGNED; 267 268 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp); 269 270 return id; 271 } 272 EXPORT_SYMBOL_GPL(peernet2id_alloc); 273 274 /* This function returns, if assigned, the id of a peer netns. */ 275 int peernet2id(const struct net *net, struct net *peer) 276 { 277 int id; 278 279 rcu_read_lock(); 280 id = __peernet2id(net, peer); 281 rcu_read_unlock(); 282 283 return id; 284 } 285 EXPORT_SYMBOL(peernet2id); 286 287 /* This function returns true is the peer netns has an id assigned into the 288 * current netns. 289 */ 290 bool peernet_has_id(const struct net *net, struct net *peer) 291 { 292 return peernet2id(net, peer) >= 0; 293 } 294 295 struct net *get_net_ns_by_id(const struct net *net, int id) 296 { 297 struct net *peer; 298 299 if (id < 0) 300 return NULL; 301 302 rcu_read_lock(); 303 peer = idr_find(&net->netns_ids, id); 304 if (peer) 305 peer = maybe_get_net(peer); 306 rcu_read_unlock(); 307 308 return peer; 309 } 310 EXPORT_SYMBOL_GPL(get_net_ns_by_id); 311 312 /* init code that must occur even if setup_net() is not called. */ 313 static __net_init void preinit_net(struct net *net) 314 { 315 ref_tracker_dir_init(&net->notrefcnt_tracker, 128, "net notrefcnt"); 316 } 317 318 /* 319 * setup_net runs the initializers for the network namespace object. 320 */ 321 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns) 322 { 323 /* Must be called with pernet_ops_rwsem held */ 324 const struct pernet_operations *ops, *saved_ops; 325 LIST_HEAD(net_exit_list); 326 LIST_HEAD(dev_kill_list); 327 int error = 0; 328 329 refcount_set(&net->ns.count, 1); 330 ref_tracker_dir_init(&net->refcnt_tracker, 128, "net refcnt"); 331 332 refcount_set(&net->passive, 1); 333 get_random_bytes(&net->hash_mix, sizeof(u32)); 334 preempt_disable(); 335 net->net_cookie = gen_cookie_next(&net_cookie); 336 preempt_enable(); 337 net->dev_base_seq = 1; 338 net->user_ns = user_ns; 339 idr_init(&net->netns_ids); 340 spin_lock_init(&net->nsid_lock); 341 mutex_init(&net->ipv4.ra_mutex); 342 343 list_for_each_entry(ops, &pernet_list, list) { 344 error = ops_init(ops, net); 345 if (error < 0) 346 goto out_undo; 347 } 348 down_write(&net_rwsem); 349 list_add_tail_rcu(&net->list, &net_namespace_list); 350 up_write(&net_rwsem); 351 out: 352 return error; 353 354 out_undo: 355 /* Walk through the list backwards calling the exit functions 356 * for the pernet modules whose init functions did not fail. 357 */ 358 list_add(&net->exit_list, &net_exit_list); 359 saved_ops = ops; 360 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 361 ops_pre_exit_list(ops, &net_exit_list); 362 363 synchronize_rcu(); 364 365 ops = saved_ops; 366 rtnl_lock(); 367 list_for_each_entry_continue_reverse(ops, &pernet_list, list) { 368 if (ops->exit_batch_rtnl) 369 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list); 370 } 371 unregister_netdevice_many(&dev_kill_list); 372 rtnl_unlock(); 373 374 ops = saved_ops; 375 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 376 ops_exit_list(ops, &net_exit_list); 377 378 ops = saved_ops; 379 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 380 ops_free_list(ops, &net_exit_list); 381 382 rcu_barrier(); 383 goto out; 384 } 385 386 static int __net_init net_defaults_init_net(struct net *net) 387 { 388 net->core.sysctl_somaxconn = SOMAXCONN; 389 /* Limits per socket sk_omem_alloc usage. 390 * TCP zerocopy regular usage needs 128 KB. 391 */ 392 net->core.sysctl_optmem_max = 128 * 1024; 393 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED; 394 395 return 0; 396 } 397 398 static struct pernet_operations net_defaults_ops = { 399 .init = net_defaults_init_net, 400 }; 401 402 static __init int net_defaults_init(void) 403 { 404 if (register_pernet_subsys(&net_defaults_ops)) 405 panic("Cannot initialize net default settings"); 406 407 return 0; 408 } 409 410 core_initcall(net_defaults_init); 411 412 #ifdef CONFIG_NET_NS 413 static struct ucounts *inc_net_namespaces(struct user_namespace *ns) 414 { 415 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES); 416 } 417 418 static void dec_net_namespaces(struct ucounts *ucounts) 419 { 420 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES); 421 } 422 423 static struct kmem_cache *net_cachep __ro_after_init; 424 static struct workqueue_struct *netns_wq; 425 426 static struct net *net_alloc(void) 427 { 428 struct net *net = NULL; 429 struct net_generic *ng; 430 431 ng = net_alloc_generic(); 432 if (!ng) 433 goto out; 434 435 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); 436 if (!net) 437 goto out_free; 438 439 #ifdef CONFIG_KEYS 440 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL); 441 if (!net->key_domain) 442 goto out_free_2; 443 refcount_set(&net->key_domain->usage, 1); 444 #endif 445 446 rcu_assign_pointer(net->gen, ng); 447 out: 448 return net; 449 450 #ifdef CONFIG_KEYS 451 out_free_2: 452 kmem_cache_free(net_cachep, net); 453 net = NULL; 454 #endif 455 out_free: 456 kfree(ng); 457 goto out; 458 } 459 460 static void net_free(struct net *net) 461 { 462 if (refcount_dec_and_test(&net->passive)) { 463 kfree(rcu_access_pointer(net->gen)); 464 465 /* There should not be any trackers left there. */ 466 ref_tracker_dir_exit(&net->notrefcnt_tracker); 467 468 kmem_cache_free(net_cachep, net); 469 } 470 } 471 472 void net_drop_ns(void *p) 473 { 474 struct net *net = (struct net *)p; 475 476 if (net) 477 net_free(net); 478 } 479 480 struct net *copy_net_ns(unsigned long flags, 481 struct user_namespace *user_ns, struct net *old_net) 482 { 483 struct ucounts *ucounts; 484 struct net *net; 485 int rv; 486 487 if (!(flags & CLONE_NEWNET)) 488 return get_net(old_net); 489 490 ucounts = inc_net_namespaces(user_ns); 491 if (!ucounts) 492 return ERR_PTR(-ENOSPC); 493 494 net = net_alloc(); 495 if (!net) { 496 rv = -ENOMEM; 497 goto dec_ucounts; 498 } 499 500 preinit_net(net); 501 refcount_set(&net->passive, 1); 502 net->ucounts = ucounts; 503 get_user_ns(user_ns); 504 505 rv = down_read_killable(&pernet_ops_rwsem); 506 if (rv < 0) 507 goto put_userns; 508 509 rv = setup_net(net, user_ns); 510 511 up_read(&pernet_ops_rwsem); 512 513 if (rv < 0) { 514 put_userns: 515 #ifdef CONFIG_KEYS 516 key_remove_domain(net->key_domain); 517 #endif 518 put_user_ns(user_ns); 519 net_free(net); 520 dec_ucounts: 521 dec_net_namespaces(ucounts); 522 return ERR_PTR(rv); 523 } 524 return net; 525 } 526 527 /** 528 * net_ns_get_ownership - get sysfs ownership data for @net 529 * @net: network namespace in question (can be NULL) 530 * @uid: kernel user ID for sysfs objects 531 * @gid: kernel group ID for sysfs objects 532 * 533 * Returns the uid/gid pair of root in the user namespace associated with the 534 * given network namespace. 535 */ 536 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid) 537 { 538 if (net) { 539 kuid_t ns_root_uid = make_kuid(net->user_ns, 0); 540 kgid_t ns_root_gid = make_kgid(net->user_ns, 0); 541 542 if (uid_valid(ns_root_uid)) 543 *uid = ns_root_uid; 544 545 if (gid_valid(ns_root_gid)) 546 *gid = ns_root_gid; 547 } else { 548 *uid = GLOBAL_ROOT_UID; 549 *gid = GLOBAL_ROOT_GID; 550 } 551 } 552 EXPORT_SYMBOL_GPL(net_ns_get_ownership); 553 554 static void unhash_nsid(struct net *net, struct net *last) 555 { 556 struct net *tmp; 557 /* This function is only called from cleanup_net() work, 558 * and this work is the only process, that may delete 559 * a net from net_namespace_list. So, when the below 560 * is executing, the list may only grow. Thus, we do not 561 * use for_each_net_rcu() or net_rwsem. 562 */ 563 for_each_net(tmp) { 564 int id; 565 566 spin_lock_bh(&tmp->nsid_lock); 567 id = __peernet2id(tmp, net); 568 if (id >= 0) 569 idr_remove(&tmp->netns_ids, id); 570 spin_unlock_bh(&tmp->nsid_lock); 571 if (id >= 0) 572 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL, 573 GFP_KERNEL); 574 if (tmp == last) 575 break; 576 } 577 spin_lock_bh(&net->nsid_lock); 578 idr_destroy(&net->netns_ids); 579 spin_unlock_bh(&net->nsid_lock); 580 } 581 582 static LLIST_HEAD(cleanup_list); 583 584 static void cleanup_net(struct work_struct *work) 585 { 586 const struct pernet_operations *ops; 587 struct net *net, *tmp, *last; 588 struct llist_node *net_kill_list; 589 LIST_HEAD(net_exit_list); 590 LIST_HEAD(dev_kill_list); 591 592 /* Atomically snapshot the list of namespaces to cleanup */ 593 net_kill_list = llist_del_all(&cleanup_list); 594 595 down_read(&pernet_ops_rwsem); 596 597 /* Don't let anyone else find us. */ 598 down_write(&net_rwsem); 599 llist_for_each_entry(net, net_kill_list, cleanup_list) 600 list_del_rcu(&net->list); 601 /* Cache last net. After we unlock rtnl, no one new net 602 * added to net_namespace_list can assign nsid pointer 603 * to a net from net_kill_list (see peernet2id_alloc()). 604 * So, we skip them in unhash_nsid(). 605 * 606 * Note, that unhash_nsid() does not delete nsid links 607 * between net_kill_list's nets, as they've already 608 * deleted from net_namespace_list. But, this would be 609 * useless anyway, as netns_ids are destroyed there. 610 */ 611 last = list_last_entry(&net_namespace_list, struct net, list); 612 up_write(&net_rwsem); 613 614 llist_for_each_entry(net, net_kill_list, cleanup_list) { 615 unhash_nsid(net, last); 616 list_add_tail(&net->exit_list, &net_exit_list); 617 } 618 619 /* Run all of the network namespace pre_exit methods */ 620 list_for_each_entry_reverse(ops, &pernet_list, list) 621 ops_pre_exit_list(ops, &net_exit_list); 622 623 /* 624 * Another CPU might be rcu-iterating the list, wait for it. 625 * This needs to be before calling the exit() notifiers, so 626 * the rcu_barrier() below isn't sufficient alone. 627 * Also the pre_exit() and exit() methods need this barrier. 628 */ 629 synchronize_rcu_expedited(); 630 631 rtnl_lock(); 632 list_for_each_entry_reverse(ops, &pernet_list, list) { 633 if (ops->exit_batch_rtnl) 634 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list); 635 } 636 unregister_netdevice_many(&dev_kill_list); 637 rtnl_unlock(); 638 639 /* Run all of the network namespace exit methods */ 640 list_for_each_entry_reverse(ops, &pernet_list, list) 641 ops_exit_list(ops, &net_exit_list); 642 643 /* Free the net generic variables */ 644 list_for_each_entry_reverse(ops, &pernet_list, list) 645 ops_free_list(ops, &net_exit_list); 646 647 up_read(&pernet_ops_rwsem); 648 649 /* Ensure there are no outstanding rcu callbacks using this 650 * network namespace. 651 */ 652 rcu_barrier(); 653 654 /* Finally it is safe to free my network namespace structure */ 655 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) { 656 list_del_init(&net->exit_list); 657 dec_net_namespaces(net->ucounts); 658 #ifdef CONFIG_KEYS 659 key_remove_domain(net->key_domain); 660 #endif 661 put_user_ns(net->user_ns); 662 net_free(net); 663 } 664 } 665 666 /** 667 * net_ns_barrier - wait until concurrent net_cleanup_work is done 668 * 669 * cleanup_net runs from work queue and will first remove namespaces 670 * from the global list, then run net exit functions. 671 * 672 * Call this in module exit path to make sure that all netns 673 * ->exit ops have been invoked before the function is removed. 674 */ 675 void net_ns_barrier(void) 676 { 677 down_write(&pernet_ops_rwsem); 678 up_write(&pernet_ops_rwsem); 679 } 680 EXPORT_SYMBOL(net_ns_barrier); 681 682 static DECLARE_WORK(net_cleanup_work, cleanup_net); 683 684 void __put_net(struct net *net) 685 { 686 ref_tracker_dir_exit(&net->refcnt_tracker); 687 /* Cleanup the network namespace in process context */ 688 if (llist_add(&net->cleanup_list, &cleanup_list)) 689 queue_work(netns_wq, &net_cleanup_work); 690 } 691 EXPORT_SYMBOL_GPL(__put_net); 692 693 /** 694 * get_net_ns - increment the refcount of the network namespace 695 * @ns: common namespace (net) 696 * 697 * Returns the net's common namespace or ERR_PTR() if ref is zero. 698 */ 699 struct ns_common *get_net_ns(struct ns_common *ns) 700 { 701 struct net *net; 702 703 net = maybe_get_net(container_of(ns, struct net, ns)); 704 if (net) 705 return &net->ns; 706 return ERR_PTR(-EINVAL); 707 } 708 EXPORT_SYMBOL_GPL(get_net_ns); 709 710 struct net *get_net_ns_by_fd(int fd) 711 { 712 struct fd f = fdget(fd); 713 struct net *net = ERR_PTR(-EINVAL); 714 715 if (!f.file) 716 return ERR_PTR(-EBADF); 717 718 if (proc_ns_file(f.file)) { 719 struct ns_common *ns = get_proc_ns(file_inode(f.file)); 720 if (ns->ops == &netns_operations) 721 net = get_net(container_of(ns, struct net, ns)); 722 } 723 fdput(f); 724 725 return net; 726 } 727 EXPORT_SYMBOL_GPL(get_net_ns_by_fd); 728 #endif 729 730 struct net *get_net_ns_by_pid(pid_t pid) 731 { 732 struct task_struct *tsk; 733 struct net *net; 734 735 /* Lookup the network namespace */ 736 net = ERR_PTR(-ESRCH); 737 rcu_read_lock(); 738 tsk = find_task_by_vpid(pid); 739 if (tsk) { 740 struct nsproxy *nsproxy; 741 task_lock(tsk); 742 nsproxy = tsk->nsproxy; 743 if (nsproxy) 744 net = get_net(nsproxy->net_ns); 745 task_unlock(tsk); 746 } 747 rcu_read_unlock(); 748 return net; 749 } 750 EXPORT_SYMBOL_GPL(get_net_ns_by_pid); 751 752 static __net_init int net_ns_net_init(struct net *net) 753 { 754 #ifdef CONFIG_NET_NS 755 net->ns.ops = &netns_operations; 756 #endif 757 return ns_alloc_inum(&net->ns); 758 } 759 760 static __net_exit void net_ns_net_exit(struct net *net) 761 { 762 ns_free_inum(&net->ns); 763 } 764 765 static struct pernet_operations __net_initdata net_ns_ops = { 766 .init = net_ns_net_init, 767 .exit = net_ns_net_exit, 768 }; 769 770 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = { 771 [NETNSA_NONE] = { .type = NLA_UNSPEC }, 772 [NETNSA_NSID] = { .type = NLA_S32 }, 773 [NETNSA_PID] = { .type = NLA_U32 }, 774 [NETNSA_FD] = { .type = NLA_U32 }, 775 [NETNSA_TARGET_NSID] = { .type = NLA_S32 }, 776 }; 777 778 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh, 779 struct netlink_ext_ack *extack) 780 { 781 struct net *net = sock_net(skb->sk); 782 struct nlattr *tb[NETNSA_MAX + 1]; 783 struct nlattr *nla; 784 struct net *peer; 785 int nsid, err; 786 787 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb, 788 NETNSA_MAX, rtnl_net_policy, extack); 789 if (err < 0) 790 return err; 791 if (!tb[NETNSA_NSID]) { 792 NL_SET_ERR_MSG(extack, "nsid is missing"); 793 return -EINVAL; 794 } 795 nsid = nla_get_s32(tb[NETNSA_NSID]); 796 797 if (tb[NETNSA_PID]) { 798 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 799 nla = tb[NETNSA_PID]; 800 } else if (tb[NETNSA_FD]) { 801 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 802 nla = tb[NETNSA_FD]; 803 } else { 804 NL_SET_ERR_MSG(extack, "Peer netns reference is missing"); 805 return -EINVAL; 806 } 807 if (IS_ERR(peer)) { 808 NL_SET_BAD_ATTR(extack, nla); 809 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid"); 810 return PTR_ERR(peer); 811 } 812 813 spin_lock_bh(&net->nsid_lock); 814 if (__peernet2id(net, peer) >= 0) { 815 spin_unlock_bh(&net->nsid_lock); 816 err = -EEXIST; 817 NL_SET_BAD_ATTR(extack, nla); 818 NL_SET_ERR_MSG(extack, 819 "Peer netns already has a nsid assigned"); 820 goto out; 821 } 822 823 err = alloc_netid(net, peer, nsid); 824 spin_unlock_bh(&net->nsid_lock); 825 if (err >= 0) { 826 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid, 827 nlh, GFP_KERNEL); 828 err = 0; 829 } else if (err == -ENOSPC && nsid >= 0) { 830 err = -EEXIST; 831 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]); 832 NL_SET_ERR_MSG(extack, "The specified nsid is already used"); 833 } 834 out: 835 put_net(peer); 836 return err; 837 } 838 839 static int rtnl_net_get_size(void) 840 { 841 return NLMSG_ALIGN(sizeof(struct rtgenmsg)) 842 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */ 843 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */ 844 ; 845 } 846 847 struct net_fill_args { 848 u32 portid; 849 u32 seq; 850 int flags; 851 int cmd; 852 int nsid; 853 bool add_ref; 854 int ref_nsid; 855 }; 856 857 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args) 858 { 859 struct nlmsghdr *nlh; 860 struct rtgenmsg *rth; 861 862 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth), 863 args->flags); 864 if (!nlh) 865 return -EMSGSIZE; 866 867 rth = nlmsg_data(nlh); 868 rth->rtgen_family = AF_UNSPEC; 869 870 if (nla_put_s32(skb, NETNSA_NSID, args->nsid)) 871 goto nla_put_failure; 872 873 if (args->add_ref && 874 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid)) 875 goto nla_put_failure; 876 877 nlmsg_end(skb, nlh); 878 return 0; 879 880 nla_put_failure: 881 nlmsg_cancel(skb, nlh); 882 return -EMSGSIZE; 883 } 884 885 static int rtnl_net_valid_getid_req(struct sk_buff *skb, 886 const struct nlmsghdr *nlh, 887 struct nlattr **tb, 888 struct netlink_ext_ack *extack) 889 { 890 int i, err; 891 892 if (!netlink_strict_get_check(skb)) 893 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), 894 tb, NETNSA_MAX, rtnl_net_policy, 895 extack); 896 897 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb, 898 NETNSA_MAX, rtnl_net_policy, 899 extack); 900 if (err) 901 return err; 902 903 for (i = 0; i <= NETNSA_MAX; i++) { 904 if (!tb[i]) 905 continue; 906 907 switch (i) { 908 case NETNSA_PID: 909 case NETNSA_FD: 910 case NETNSA_NSID: 911 case NETNSA_TARGET_NSID: 912 break; 913 default: 914 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request"); 915 return -EINVAL; 916 } 917 } 918 919 return 0; 920 } 921 922 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh, 923 struct netlink_ext_ack *extack) 924 { 925 struct net *net = sock_net(skb->sk); 926 struct nlattr *tb[NETNSA_MAX + 1]; 927 struct net_fill_args fillargs = { 928 .portid = NETLINK_CB(skb).portid, 929 .seq = nlh->nlmsg_seq, 930 .cmd = RTM_NEWNSID, 931 }; 932 struct net *peer, *target = net; 933 struct nlattr *nla; 934 struct sk_buff *msg; 935 int err; 936 937 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack); 938 if (err < 0) 939 return err; 940 if (tb[NETNSA_PID]) { 941 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 942 nla = tb[NETNSA_PID]; 943 } else if (tb[NETNSA_FD]) { 944 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 945 nla = tb[NETNSA_FD]; 946 } else if (tb[NETNSA_NSID]) { 947 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID])); 948 if (!peer) 949 peer = ERR_PTR(-ENOENT); 950 nla = tb[NETNSA_NSID]; 951 } else { 952 NL_SET_ERR_MSG(extack, "Peer netns reference is missing"); 953 return -EINVAL; 954 } 955 956 if (IS_ERR(peer)) { 957 NL_SET_BAD_ATTR(extack, nla); 958 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid"); 959 return PTR_ERR(peer); 960 } 961 962 if (tb[NETNSA_TARGET_NSID]) { 963 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]); 964 965 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id); 966 if (IS_ERR(target)) { 967 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]); 968 NL_SET_ERR_MSG(extack, 969 "Target netns reference is invalid"); 970 err = PTR_ERR(target); 971 goto out; 972 } 973 fillargs.add_ref = true; 974 fillargs.ref_nsid = peernet2id(net, peer); 975 } 976 977 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 978 if (!msg) { 979 err = -ENOMEM; 980 goto out; 981 } 982 983 fillargs.nsid = peernet2id(target, peer); 984 err = rtnl_net_fill(msg, &fillargs); 985 if (err < 0) 986 goto err_out; 987 988 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid); 989 goto out; 990 991 err_out: 992 nlmsg_free(msg); 993 out: 994 if (fillargs.add_ref) 995 put_net(target); 996 put_net(peer); 997 return err; 998 } 999 1000 struct rtnl_net_dump_cb { 1001 struct net *tgt_net; 1002 struct net *ref_net; 1003 struct sk_buff *skb; 1004 struct net_fill_args fillargs; 1005 int idx; 1006 int s_idx; 1007 }; 1008 1009 /* Runs in RCU-critical section. */ 1010 static int rtnl_net_dumpid_one(int id, void *peer, void *data) 1011 { 1012 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data; 1013 int ret; 1014 1015 if (net_cb->idx < net_cb->s_idx) 1016 goto cont; 1017 1018 net_cb->fillargs.nsid = id; 1019 if (net_cb->fillargs.add_ref) 1020 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer); 1021 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs); 1022 if (ret < 0) 1023 return ret; 1024 1025 cont: 1026 net_cb->idx++; 1027 return 0; 1028 } 1029 1030 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk, 1031 struct rtnl_net_dump_cb *net_cb, 1032 struct netlink_callback *cb) 1033 { 1034 struct netlink_ext_ack *extack = cb->extack; 1035 struct nlattr *tb[NETNSA_MAX + 1]; 1036 int err, i; 1037 1038 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb, 1039 NETNSA_MAX, rtnl_net_policy, 1040 extack); 1041 if (err < 0) 1042 return err; 1043 1044 for (i = 0; i <= NETNSA_MAX; i++) { 1045 if (!tb[i]) 1046 continue; 1047 1048 if (i == NETNSA_TARGET_NSID) { 1049 struct net *net; 1050 1051 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i])); 1052 if (IS_ERR(net)) { 1053 NL_SET_BAD_ATTR(extack, tb[i]); 1054 NL_SET_ERR_MSG(extack, 1055 "Invalid target network namespace id"); 1056 return PTR_ERR(net); 1057 } 1058 net_cb->fillargs.add_ref = true; 1059 net_cb->ref_net = net_cb->tgt_net; 1060 net_cb->tgt_net = net; 1061 } else { 1062 NL_SET_BAD_ATTR(extack, tb[i]); 1063 NL_SET_ERR_MSG(extack, 1064 "Unsupported attribute in dump request"); 1065 return -EINVAL; 1066 } 1067 } 1068 1069 return 0; 1070 } 1071 1072 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb) 1073 { 1074 struct rtnl_net_dump_cb net_cb = { 1075 .tgt_net = sock_net(skb->sk), 1076 .skb = skb, 1077 .fillargs = { 1078 .portid = NETLINK_CB(cb->skb).portid, 1079 .seq = cb->nlh->nlmsg_seq, 1080 .flags = NLM_F_MULTI, 1081 .cmd = RTM_NEWNSID, 1082 }, 1083 .idx = 0, 1084 .s_idx = cb->args[0], 1085 }; 1086 int err = 0; 1087 1088 if (cb->strict_check) { 1089 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb); 1090 if (err < 0) 1091 goto end; 1092 } 1093 1094 rcu_read_lock(); 1095 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb); 1096 rcu_read_unlock(); 1097 1098 cb->args[0] = net_cb.idx; 1099 end: 1100 if (net_cb.fillargs.add_ref) 1101 put_net(net_cb.tgt_net); 1102 return err; 1103 } 1104 1105 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid, 1106 struct nlmsghdr *nlh, gfp_t gfp) 1107 { 1108 struct net_fill_args fillargs = { 1109 .portid = portid, 1110 .seq = nlh ? nlh->nlmsg_seq : 0, 1111 .cmd = cmd, 1112 .nsid = id, 1113 }; 1114 struct sk_buff *msg; 1115 int err = -ENOMEM; 1116 1117 msg = nlmsg_new(rtnl_net_get_size(), gfp); 1118 if (!msg) 1119 goto out; 1120 1121 err = rtnl_net_fill(msg, &fillargs); 1122 if (err < 0) 1123 goto err_out; 1124 1125 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp); 1126 return; 1127 1128 err_out: 1129 nlmsg_free(msg); 1130 out: 1131 rtnl_set_sk_err(net, RTNLGRP_NSID, err); 1132 } 1133 1134 #ifdef CONFIG_NET_NS 1135 static void __init netns_ipv4_struct_check(void) 1136 { 1137 /* TX readonly hotpath cache lines */ 1138 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1139 sysctl_tcp_early_retrans); 1140 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1141 sysctl_tcp_tso_win_divisor); 1142 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1143 sysctl_tcp_tso_rtt_log); 1144 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1145 sysctl_tcp_autocorking); 1146 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1147 sysctl_tcp_min_snd_mss); 1148 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1149 sysctl_tcp_notsent_lowat); 1150 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1151 sysctl_tcp_limit_output_bytes); 1152 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1153 sysctl_tcp_min_rtt_wlen); 1154 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1155 sysctl_tcp_wmem); 1156 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx, 1157 sysctl_ip_fwd_use_pmtu); 1158 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_tx, 33); 1159 1160 /* TXRX readonly hotpath cache lines */ 1161 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_txrx, 1162 sysctl_tcp_moderate_rcvbuf); 1163 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_txrx, 1); 1164 1165 /* RX readonly hotpath cache line */ 1166 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx, 1167 sysctl_ip_early_demux); 1168 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx, 1169 sysctl_tcp_early_demux); 1170 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx, 1171 sysctl_tcp_reordering); 1172 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx, 1173 sysctl_tcp_rmem); 1174 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_rx, 18); 1175 } 1176 #endif 1177 1178 void __init net_ns_init(void) 1179 { 1180 struct net_generic *ng; 1181 1182 #ifdef CONFIG_NET_NS 1183 netns_ipv4_struct_check(); 1184 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), 1185 SMP_CACHE_BYTES, 1186 SLAB_PANIC|SLAB_ACCOUNT, NULL); 1187 1188 /* Create workqueue for cleanup */ 1189 netns_wq = create_singlethread_workqueue("netns"); 1190 if (!netns_wq) 1191 panic("Could not create netns workq"); 1192 #endif 1193 1194 ng = net_alloc_generic(); 1195 if (!ng) 1196 panic("Could not allocate generic netns"); 1197 1198 rcu_assign_pointer(init_net.gen, ng); 1199 1200 #ifdef CONFIG_KEYS 1201 init_net.key_domain = &init_net_key_domain; 1202 #endif 1203 down_write(&pernet_ops_rwsem); 1204 preinit_net(&init_net); 1205 if (setup_net(&init_net, &init_user_ns)) 1206 panic("Could not setup the initial network namespace"); 1207 1208 init_net_initialized = true; 1209 up_write(&pernet_ops_rwsem); 1210 1211 if (register_pernet_subsys(&net_ns_ops)) 1212 panic("Could not register network namespace subsystems"); 1213 1214 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, 1215 RTNL_FLAG_DOIT_UNLOCKED); 1216 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid, 1217 RTNL_FLAG_DOIT_UNLOCKED | 1218 RTNL_FLAG_DUMP_UNLOCKED); 1219 } 1220 1221 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list) 1222 { 1223 ops_pre_exit_list(ops, net_exit_list); 1224 synchronize_rcu(); 1225 1226 if (ops->exit_batch_rtnl) { 1227 LIST_HEAD(dev_kill_list); 1228 1229 rtnl_lock(); 1230 ops->exit_batch_rtnl(net_exit_list, &dev_kill_list); 1231 unregister_netdevice_many(&dev_kill_list); 1232 rtnl_unlock(); 1233 } 1234 ops_exit_list(ops, net_exit_list); 1235 1236 ops_free_list(ops, net_exit_list); 1237 } 1238 1239 #ifdef CONFIG_NET_NS 1240 static int __register_pernet_operations(struct list_head *list, 1241 struct pernet_operations *ops) 1242 { 1243 struct net *net; 1244 int error; 1245 LIST_HEAD(net_exit_list); 1246 1247 list_add_tail(&ops->list, list); 1248 if (ops->init || ops->id) { 1249 /* We held write locked pernet_ops_rwsem, and parallel 1250 * setup_net() and cleanup_net() are not possible. 1251 */ 1252 for_each_net(net) { 1253 error = ops_init(ops, net); 1254 if (error) 1255 goto out_undo; 1256 list_add_tail(&net->exit_list, &net_exit_list); 1257 } 1258 } 1259 return 0; 1260 1261 out_undo: 1262 /* If I have an error cleanup all namespaces I initialized */ 1263 list_del(&ops->list); 1264 free_exit_list(ops, &net_exit_list); 1265 return error; 1266 } 1267 1268 static void __unregister_pernet_operations(struct pernet_operations *ops) 1269 { 1270 struct net *net; 1271 LIST_HEAD(net_exit_list); 1272 1273 list_del(&ops->list); 1274 /* See comment in __register_pernet_operations() */ 1275 for_each_net(net) 1276 list_add_tail(&net->exit_list, &net_exit_list); 1277 1278 free_exit_list(ops, &net_exit_list); 1279 } 1280 1281 #else 1282 1283 static int __register_pernet_operations(struct list_head *list, 1284 struct pernet_operations *ops) 1285 { 1286 if (!init_net_initialized) { 1287 list_add_tail(&ops->list, list); 1288 return 0; 1289 } 1290 1291 return ops_init(ops, &init_net); 1292 } 1293 1294 static void __unregister_pernet_operations(struct pernet_operations *ops) 1295 { 1296 if (!init_net_initialized) { 1297 list_del(&ops->list); 1298 } else { 1299 LIST_HEAD(net_exit_list); 1300 list_add(&init_net.exit_list, &net_exit_list); 1301 free_exit_list(ops, &net_exit_list); 1302 } 1303 } 1304 1305 #endif /* CONFIG_NET_NS */ 1306 1307 static DEFINE_IDA(net_generic_ids); 1308 1309 static int register_pernet_operations(struct list_head *list, 1310 struct pernet_operations *ops) 1311 { 1312 int error; 1313 1314 if (WARN_ON(!!ops->id ^ !!ops->size)) 1315 return -EINVAL; 1316 1317 if (ops->id) { 1318 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID, 1319 GFP_KERNEL); 1320 if (error < 0) 1321 return error; 1322 *ops->id = error; 1323 /* This does not require READ_ONCE as writers already hold 1324 * pernet_ops_rwsem. But WRITE_ONCE is needed to protect 1325 * net_alloc_generic. 1326 */ 1327 WRITE_ONCE(max_gen_ptrs, max(max_gen_ptrs, *ops->id + 1)); 1328 } 1329 error = __register_pernet_operations(list, ops); 1330 if (error) { 1331 rcu_barrier(); 1332 if (ops->id) 1333 ida_free(&net_generic_ids, *ops->id); 1334 } 1335 1336 return error; 1337 } 1338 1339 static void unregister_pernet_operations(struct pernet_operations *ops) 1340 { 1341 __unregister_pernet_operations(ops); 1342 rcu_barrier(); 1343 if (ops->id) 1344 ida_free(&net_generic_ids, *ops->id); 1345 } 1346 1347 /** 1348 * register_pernet_subsys - register a network namespace subsystem 1349 * @ops: pernet operations structure for the subsystem 1350 * 1351 * Register a subsystem which has init and exit functions 1352 * that are called when network namespaces are created and 1353 * destroyed respectively. 1354 * 1355 * When registered all network namespace init functions are 1356 * called for every existing network namespace. Allowing kernel 1357 * modules to have a race free view of the set of network namespaces. 1358 * 1359 * When a new network namespace is created all of the init 1360 * methods are called in the order in which they were registered. 1361 * 1362 * When a network namespace is destroyed all of the exit methods 1363 * are called in the reverse of the order with which they were 1364 * registered. 1365 */ 1366 int register_pernet_subsys(struct pernet_operations *ops) 1367 { 1368 int error; 1369 down_write(&pernet_ops_rwsem); 1370 error = register_pernet_operations(first_device, ops); 1371 up_write(&pernet_ops_rwsem); 1372 return error; 1373 } 1374 EXPORT_SYMBOL_GPL(register_pernet_subsys); 1375 1376 /** 1377 * unregister_pernet_subsys - unregister a network namespace subsystem 1378 * @ops: pernet operations structure to manipulate 1379 * 1380 * Remove the pernet operations structure from the list to be 1381 * used when network namespaces are created or destroyed. In 1382 * addition run the exit method for all existing network 1383 * namespaces. 1384 */ 1385 void unregister_pernet_subsys(struct pernet_operations *ops) 1386 { 1387 down_write(&pernet_ops_rwsem); 1388 unregister_pernet_operations(ops); 1389 up_write(&pernet_ops_rwsem); 1390 } 1391 EXPORT_SYMBOL_GPL(unregister_pernet_subsys); 1392 1393 /** 1394 * register_pernet_device - register a network namespace device 1395 * @ops: pernet operations structure for the subsystem 1396 * 1397 * Register a device which has init and exit functions 1398 * that are called when network namespaces are created and 1399 * destroyed respectively. 1400 * 1401 * When registered all network namespace init functions are 1402 * called for every existing network namespace. Allowing kernel 1403 * modules to have a race free view of the set of network namespaces. 1404 * 1405 * When a new network namespace is created all of the init 1406 * methods are called in the order in which they were registered. 1407 * 1408 * When a network namespace is destroyed all of the exit methods 1409 * are called in the reverse of the order with which they were 1410 * registered. 1411 */ 1412 int register_pernet_device(struct pernet_operations *ops) 1413 { 1414 int error; 1415 down_write(&pernet_ops_rwsem); 1416 error = register_pernet_operations(&pernet_list, ops); 1417 if (!error && (first_device == &pernet_list)) 1418 first_device = &ops->list; 1419 up_write(&pernet_ops_rwsem); 1420 return error; 1421 } 1422 EXPORT_SYMBOL_GPL(register_pernet_device); 1423 1424 /** 1425 * unregister_pernet_device - unregister a network namespace netdevice 1426 * @ops: pernet operations structure to manipulate 1427 * 1428 * Remove the pernet operations structure from the list to be 1429 * used when network namespaces are created or destroyed. In 1430 * addition run the exit method for all existing network 1431 * namespaces. 1432 */ 1433 void unregister_pernet_device(struct pernet_operations *ops) 1434 { 1435 down_write(&pernet_ops_rwsem); 1436 if (&ops->list == first_device) 1437 first_device = first_device->next; 1438 unregister_pernet_operations(ops); 1439 up_write(&pernet_ops_rwsem); 1440 } 1441 EXPORT_SYMBOL_GPL(unregister_pernet_device); 1442 1443 #ifdef CONFIG_NET_NS 1444 static struct ns_common *netns_get(struct task_struct *task) 1445 { 1446 struct net *net = NULL; 1447 struct nsproxy *nsproxy; 1448 1449 task_lock(task); 1450 nsproxy = task->nsproxy; 1451 if (nsproxy) 1452 net = get_net(nsproxy->net_ns); 1453 task_unlock(task); 1454 1455 return net ? &net->ns : NULL; 1456 } 1457 1458 static inline struct net *to_net_ns(struct ns_common *ns) 1459 { 1460 return container_of(ns, struct net, ns); 1461 } 1462 1463 static void netns_put(struct ns_common *ns) 1464 { 1465 put_net(to_net_ns(ns)); 1466 } 1467 1468 static int netns_install(struct nsset *nsset, struct ns_common *ns) 1469 { 1470 struct nsproxy *nsproxy = nsset->nsproxy; 1471 struct net *net = to_net_ns(ns); 1472 1473 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) || 1474 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN)) 1475 return -EPERM; 1476 1477 put_net(nsproxy->net_ns); 1478 nsproxy->net_ns = get_net(net); 1479 return 0; 1480 } 1481 1482 static struct user_namespace *netns_owner(struct ns_common *ns) 1483 { 1484 return to_net_ns(ns)->user_ns; 1485 } 1486 1487 const struct proc_ns_operations netns_operations = { 1488 .name = "net", 1489 .type = CLONE_NEWNET, 1490 .get = netns_get, 1491 .put = netns_put, 1492 .install = netns_install, 1493 .owner = netns_owner, 1494 }; 1495 #endif 1496