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