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