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