1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 2 3 #include <linux/workqueue.h> 4 #include <linux/rtnetlink.h> 5 #include <linux/cache.h> 6 #include <linux/slab.h> 7 #include <linux/list.h> 8 #include <linux/delay.h> 9 #include <linux/sched.h> 10 #include <linux/idr.h> 11 #include <linux/rculist.h> 12 #include <linux/nsproxy.h> 13 #include <linux/fs.h> 14 #include <linux/proc_ns.h> 15 #include <linux/file.h> 16 #include <linux/export.h> 17 #include <linux/user_namespace.h> 18 #include <linux/net_namespace.h> 19 #include <linux/sched/task.h> 20 #include <linux/uidgid.h> 21 22 #include <net/sock.h> 23 #include <net/netlink.h> 24 #include <net/net_namespace.h> 25 #include <net/netns/generic.h> 26 27 /* 28 * Our network namespace constructor/destructor lists 29 */ 30 31 static LIST_HEAD(pernet_list); 32 static struct list_head *first_device = &pernet_list; 33 34 LIST_HEAD(net_namespace_list); 35 EXPORT_SYMBOL_GPL(net_namespace_list); 36 37 /* Protects net_namespace_list. Nests iside rtnl_lock() */ 38 DECLARE_RWSEM(net_rwsem); 39 EXPORT_SYMBOL_GPL(net_rwsem); 40 41 #ifdef CONFIG_KEYS 42 static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) }; 43 #endif 44 45 struct net init_net = { 46 .count = REFCOUNT_INIT(1), 47 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head), 48 #ifdef CONFIG_KEYS 49 .key_domain = &init_net_key_domain, 50 #endif 51 }; 52 EXPORT_SYMBOL(init_net); 53 54 static bool init_net_initialized; 55 /* 56 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids, 57 * init_net_initialized and first_device pointer. 58 * This is internal net namespace object. Please, don't use it 59 * outside. 60 */ 61 DECLARE_RWSEM(pernet_ops_rwsem); 62 EXPORT_SYMBOL_GPL(pernet_ops_rwsem); 63 64 #define MIN_PERNET_OPS_ID \ 65 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *)) 66 67 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ 68 69 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS; 70 71 static struct net_generic *net_alloc_generic(void) 72 { 73 struct net_generic *ng; 74 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]); 75 76 ng = kzalloc(generic_size, GFP_KERNEL); 77 if (ng) 78 ng->s.len = max_gen_ptrs; 79 80 return ng; 81 } 82 83 static int net_assign_generic(struct net *net, unsigned int id, void *data) 84 { 85 struct net_generic *ng, *old_ng; 86 87 BUG_ON(id < MIN_PERNET_OPS_ID); 88 89 old_ng = rcu_dereference_protected(net->gen, 90 lockdep_is_held(&pernet_ops_rwsem)); 91 if (old_ng->s.len > id) { 92 old_ng->ptr[id] = data; 93 return 0; 94 } 95 96 ng = net_alloc_generic(); 97 if (ng == NULL) 98 return -ENOMEM; 99 100 /* 101 * Some synchronisation notes: 102 * 103 * The net_generic explores the net->gen array inside rcu 104 * read section. Besides once set the net->gen->ptr[x] 105 * pointer never changes (see rules in netns/generic.h). 106 * 107 * That said, we simply duplicate this array and schedule 108 * the old copy for kfree after a grace period. 109 */ 110 111 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID], 112 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *)); 113 ng->ptr[id] = data; 114 115 rcu_assign_pointer(net->gen, ng); 116 kfree_rcu(old_ng, s.rcu); 117 return 0; 118 } 119 120 static int ops_init(const struct pernet_operations *ops, struct net *net) 121 { 122 int err = -ENOMEM; 123 void *data = NULL; 124 125 if (ops->id && ops->size) { 126 data = kzalloc(ops->size, GFP_KERNEL); 127 if (!data) 128 goto out; 129 130 err = net_assign_generic(net, *ops->id, data); 131 if (err) 132 goto cleanup; 133 } 134 err = 0; 135 if (ops->init) 136 err = ops->init(net); 137 if (!err) 138 return 0; 139 140 cleanup: 141 kfree(data); 142 143 out: 144 return err; 145 } 146 147 static void ops_free(const struct pernet_operations *ops, struct net *net) 148 { 149 if (ops->id && ops->size) { 150 kfree(net_generic(net, *ops->id)); 151 } 152 } 153 154 static void ops_exit_list(const struct pernet_operations *ops, 155 struct list_head *net_exit_list) 156 { 157 struct net *net; 158 if (ops->exit) { 159 list_for_each_entry(net, net_exit_list, exit_list) 160 ops->exit(net); 161 } 162 if (ops->exit_batch) 163 ops->exit_batch(net_exit_list); 164 } 165 166 static void ops_free_list(const struct pernet_operations *ops, 167 struct list_head *net_exit_list) 168 { 169 struct net *net; 170 if (ops->size && ops->id) { 171 list_for_each_entry(net, net_exit_list, exit_list) 172 ops_free(ops, net); 173 } 174 } 175 176 /* should be called with nsid_lock held */ 177 static int alloc_netid(struct net *net, struct net *peer, int reqid) 178 { 179 int min = 0, max = 0; 180 181 if (reqid >= 0) { 182 min = reqid; 183 max = reqid + 1; 184 } 185 186 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC); 187 } 188 189 /* This function is used by idr_for_each(). If net is equal to peer, the 190 * function returns the id so that idr_for_each() stops. Because we cannot 191 * returns the id 0 (idr_for_each() will not stop), we return the magic value 192 * NET_ID_ZERO (-1) for it. 193 */ 194 #define NET_ID_ZERO -1 195 static int net_eq_idr(int id, void *net, void *peer) 196 { 197 if (net_eq(net, peer)) 198 return id ? : NET_ID_ZERO; 199 return 0; 200 } 201 202 /* Should be called with nsid_lock held. If a new id is assigned, the bool alloc 203 * is set to true, thus the caller knows that the new id must be notified via 204 * rtnl. 205 */ 206 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc) 207 { 208 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer); 209 bool alloc_it = *alloc; 210 211 *alloc = false; 212 213 /* Magic value for id 0. */ 214 if (id == NET_ID_ZERO) 215 return 0; 216 if (id > 0) 217 return id; 218 219 if (alloc_it) { 220 id = alloc_netid(net, peer, -1); 221 *alloc = true; 222 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED; 223 } 224 225 return NETNSA_NSID_NOT_ASSIGNED; 226 } 227 228 /* should be called with nsid_lock held */ 229 static int __peernet2id(struct net *net, struct net *peer) 230 { 231 bool no = false; 232 233 return __peernet2id_alloc(net, peer, &no); 234 } 235 236 static void rtnl_net_notifyid(struct net *net, int cmd, int id); 237 /* This function returns the id of a peer netns. If no id is assigned, one will 238 * be allocated and returned. 239 */ 240 int peernet2id_alloc(struct net *net, struct net *peer) 241 { 242 bool alloc = false, alive = false; 243 int id; 244 245 if (refcount_read(&net->count) == 0) 246 return NETNSA_NSID_NOT_ASSIGNED; 247 spin_lock_bh(&net->nsid_lock); 248 /* 249 * When peer is obtained from RCU lists, we may race with 250 * its cleanup. Check whether it's alive, and this guarantees 251 * we never hash a peer back to net->netns_ids, after it has 252 * just been idr_remove()'d from there in cleanup_net(). 253 */ 254 if (maybe_get_net(peer)) 255 alive = alloc = true; 256 id = __peernet2id_alloc(net, peer, &alloc); 257 spin_unlock_bh(&net->nsid_lock); 258 if (alloc && id >= 0) 259 rtnl_net_notifyid(net, RTM_NEWNSID, id); 260 if (alive) 261 put_net(peer); 262 return id; 263 } 264 EXPORT_SYMBOL_GPL(peernet2id_alloc); 265 266 /* This function returns, if assigned, the id of a peer netns. */ 267 int peernet2id(struct net *net, struct net *peer) 268 { 269 int id; 270 271 spin_lock_bh(&net->nsid_lock); 272 id = __peernet2id(net, peer); 273 spin_unlock_bh(&net->nsid_lock); 274 return id; 275 } 276 EXPORT_SYMBOL(peernet2id); 277 278 /* This function returns true is the peer netns has an id assigned into the 279 * current netns. 280 */ 281 bool peernet_has_id(struct net *net, struct net *peer) 282 { 283 return peernet2id(net, peer) >= 0; 284 } 285 286 struct net *get_net_ns_by_id(struct net *net, int id) 287 { 288 struct net *peer; 289 290 if (id < 0) 291 return NULL; 292 293 rcu_read_lock(); 294 peer = idr_find(&net->netns_ids, id); 295 if (peer) 296 peer = maybe_get_net(peer); 297 rcu_read_unlock(); 298 299 return peer; 300 } 301 302 /* 303 * setup_net runs the initializers for the network namespace object. 304 */ 305 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns) 306 { 307 /* Must be called with pernet_ops_rwsem held */ 308 const struct pernet_operations *ops, *saved_ops; 309 int error = 0; 310 LIST_HEAD(net_exit_list); 311 312 refcount_set(&net->count, 1); 313 refcount_set(&net->passive, 1); 314 get_random_bytes(&net->hash_mix, sizeof(u32)); 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_exit_list(ops, &net_exit_list); 340 341 ops = saved_ops; 342 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 343 ops_free_list(ops, &net_exit_list); 344 345 rcu_barrier(); 346 goto out; 347 } 348 349 static int __net_init net_defaults_init_net(struct net *net) 350 { 351 net->core.sysctl_somaxconn = SOMAXCONN; 352 return 0; 353 } 354 355 static struct pernet_operations net_defaults_ops = { 356 .init = net_defaults_init_net, 357 }; 358 359 static __init int net_defaults_init(void) 360 { 361 if (register_pernet_subsys(&net_defaults_ops)) 362 panic("Cannot initialize net default settings"); 363 364 return 0; 365 } 366 367 core_initcall(net_defaults_init); 368 369 #ifdef CONFIG_NET_NS 370 static struct ucounts *inc_net_namespaces(struct user_namespace *ns) 371 { 372 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES); 373 } 374 375 static void dec_net_namespaces(struct ucounts *ucounts) 376 { 377 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES); 378 } 379 380 static struct kmem_cache *net_cachep __ro_after_init; 381 static struct workqueue_struct *netns_wq; 382 383 static struct net *net_alloc(void) 384 { 385 struct net *net = NULL; 386 struct net_generic *ng; 387 388 ng = net_alloc_generic(); 389 if (!ng) 390 goto out; 391 392 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); 393 if (!net) 394 goto out_free; 395 396 #ifdef CONFIG_KEYS 397 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL); 398 if (!net->key_domain) 399 goto out_free_2; 400 refcount_set(&net->key_domain->usage, 1); 401 #endif 402 403 rcu_assign_pointer(net->gen, ng); 404 out: 405 return net; 406 407 #ifdef CONFIG_KEYS 408 out_free_2: 409 kmem_cache_free(net_cachep, net); 410 net = NULL; 411 #endif 412 out_free: 413 kfree(ng); 414 goto out; 415 } 416 417 static void net_free(struct net *net) 418 { 419 kfree(rcu_access_pointer(net->gen)); 420 kmem_cache_free(net_cachep, net); 421 } 422 423 void net_drop_ns(void *p) 424 { 425 struct net *ns = p; 426 if (ns && refcount_dec_and_test(&ns->passive)) 427 net_free(ns); 428 } 429 430 struct net *copy_net_ns(unsigned long flags, 431 struct user_namespace *user_ns, struct net *old_net) 432 { 433 struct ucounts *ucounts; 434 struct net *net; 435 int rv; 436 437 if (!(flags & CLONE_NEWNET)) 438 return get_net(old_net); 439 440 ucounts = inc_net_namespaces(user_ns); 441 if (!ucounts) 442 return ERR_PTR(-ENOSPC); 443 444 net = net_alloc(); 445 if (!net) { 446 rv = -ENOMEM; 447 goto dec_ucounts; 448 } 449 refcount_set(&net->passive, 1); 450 net->ucounts = ucounts; 451 get_user_ns(user_ns); 452 453 rv = down_read_killable(&pernet_ops_rwsem); 454 if (rv < 0) 455 goto put_userns; 456 457 rv = setup_net(net, user_ns); 458 459 up_read(&pernet_ops_rwsem); 460 461 if (rv < 0) { 462 put_userns: 463 put_user_ns(user_ns); 464 net_drop_ns(net); 465 dec_ucounts: 466 dec_net_namespaces(ucounts); 467 return ERR_PTR(rv); 468 } 469 return net; 470 } 471 472 /** 473 * net_ns_get_ownership - get sysfs ownership data for @net 474 * @net: network namespace in question (can be NULL) 475 * @uid: kernel user ID for sysfs objects 476 * @gid: kernel group ID for sysfs objects 477 * 478 * Returns the uid/gid pair of root in the user namespace associated with the 479 * given network namespace. 480 */ 481 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid) 482 { 483 if (net) { 484 kuid_t ns_root_uid = make_kuid(net->user_ns, 0); 485 kgid_t ns_root_gid = make_kgid(net->user_ns, 0); 486 487 if (uid_valid(ns_root_uid)) 488 *uid = ns_root_uid; 489 490 if (gid_valid(ns_root_gid)) 491 *gid = ns_root_gid; 492 } else { 493 *uid = GLOBAL_ROOT_UID; 494 *gid = GLOBAL_ROOT_GID; 495 } 496 } 497 EXPORT_SYMBOL_GPL(net_ns_get_ownership); 498 499 static void unhash_nsid(struct net *net, struct net *last) 500 { 501 struct net *tmp; 502 /* This function is only called from cleanup_net() work, 503 * and this work is the only process, that may delete 504 * a net from net_namespace_list. So, when the below 505 * is executing, the list may only grow. Thus, we do not 506 * use for_each_net_rcu() or net_rwsem. 507 */ 508 for_each_net(tmp) { 509 int id; 510 511 spin_lock_bh(&tmp->nsid_lock); 512 id = __peernet2id(tmp, net); 513 if (id >= 0) 514 idr_remove(&tmp->netns_ids, id); 515 spin_unlock_bh(&tmp->nsid_lock); 516 if (id >= 0) 517 rtnl_net_notifyid(tmp, RTM_DELNSID, id); 518 if (tmp == last) 519 break; 520 } 521 spin_lock_bh(&net->nsid_lock); 522 idr_destroy(&net->netns_ids); 523 spin_unlock_bh(&net->nsid_lock); 524 } 525 526 static LLIST_HEAD(cleanup_list); 527 528 static void cleanup_net(struct work_struct *work) 529 { 530 const struct pernet_operations *ops; 531 struct net *net, *tmp, *last; 532 struct llist_node *net_kill_list; 533 LIST_HEAD(net_exit_list); 534 535 /* Atomically snapshot the list of namespaces to cleanup */ 536 net_kill_list = llist_del_all(&cleanup_list); 537 538 down_read(&pernet_ops_rwsem); 539 540 /* Don't let anyone else find us. */ 541 down_write(&net_rwsem); 542 llist_for_each_entry(net, net_kill_list, cleanup_list) 543 list_del_rcu(&net->list); 544 /* Cache last net. After we unlock rtnl, no one new net 545 * added to net_namespace_list can assign nsid pointer 546 * to a net from net_kill_list (see peernet2id_alloc()). 547 * So, we skip them in unhash_nsid(). 548 * 549 * Note, that unhash_nsid() does not delete nsid links 550 * between net_kill_list's nets, as they've already 551 * deleted from net_namespace_list. But, this would be 552 * useless anyway, as netns_ids are destroyed there. 553 */ 554 last = list_last_entry(&net_namespace_list, struct net, list); 555 up_write(&net_rwsem); 556 557 llist_for_each_entry(net, net_kill_list, cleanup_list) { 558 unhash_nsid(net, last); 559 list_add_tail(&net->exit_list, &net_exit_list); 560 } 561 562 /* 563 * Another CPU might be rcu-iterating the list, wait for it. 564 * This needs to be before calling the exit() notifiers, so 565 * the rcu_barrier() below isn't sufficient alone. 566 */ 567 synchronize_rcu(); 568 569 /* Run all of the network namespace exit methods */ 570 list_for_each_entry_reverse(ops, &pernet_list, list) 571 ops_exit_list(ops, &net_exit_list); 572 573 /* Free the net generic variables */ 574 list_for_each_entry_reverse(ops, &pernet_list, list) 575 ops_free_list(ops, &net_exit_list); 576 577 up_read(&pernet_ops_rwsem); 578 579 /* Ensure there are no outstanding rcu callbacks using this 580 * network namespace. 581 */ 582 rcu_barrier(); 583 584 /* Finally it is safe to free my network namespace structure */ 585 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) { 586 list_del_init(&net->exit_list); 587 dec_net_namespaces(net->ucounts); 588 key_remove_domain(net->key_domain); 589 put_user_ns(net->user_ns); 590 net_drop_ns(net); 591 } 592 } 593 594 /** 595 * net_ns_barrier - wait until concurrent net_cleanup_work is done 596 * 597 * cleanup_net runs from work queue and will first remove namespaces 598 * from the global list, then run net exit functions. 599 * 600 * Call this in module exit path to make sure that all netns 601 * ->exit ops have been invoked before the function is removed. 602 */ 603 void net_ns_barrier(void) 604 { 605 down_write(&pernet_ops_rwsem); 606 up_write(&pernet_ops_rwsem); 607 } 608 EXPORT_SYMBOL(net_ns_barrier); 609 610 static DECLARE_WORK(net_cleanup_work, cleanup_net); 611 612 void __put_net(struct net *net) 613 { 614 /* Cleanup the network namespace in process context */ 615 if (llist_add(&net->cleanup_list, &cleanup_list)) 616 queue_work(netns_wq, &net_cleanup_work); 617 } 618 EXPORT_SYMBOL_GPL(__put_net); 619 620 struct net *get_net_ns_by_fd(int fd) 621 { 622 struct file *file; 623 struct ns_common *ns; 624 struct net *net; 625 626 file = proc_ns_fget(fd); 627 if (IS_ERR(file)) 628 return ERR_CAST(file); 629 630 ns = get_proc_ns(file_inode(file)); 631 if (ns->ops == &netns_operations) 632 net = get_net(container_of(ns, struct net, ns)); 633 else 634 net = ERR_PTR(-EINVAL); 635 636 fput(file); 637 return net; 638 } 639 640 #else 641 struct net *get_net_ns_by_fd(int fd) 642 { 643 return ERR_PTR(-EINVAL); 644 } 645 #endif 646 EXPORT_SYMBOL_GPL(get_net_ns_by_fd); 647 648 struct net *get_net_ns_by_pid(pid_t pid) 649 { 650 struct task_struct *tsk; 651 struct net *net; 652 653 /* Lookup the network namespace */ 654 net = ERR_PTR(-ESRCH); 655 rcu_read_lock(); 656 tsk = find_task_by_vpid(pid); 657 if (tsk) { 658 struct nsproxy *nsproxy; 659 task_lock(tsk); 660 nsproxy = tsk->nsproxy; 661 if (nsproxy) 662 net = get_net(nsproxy->net_ns); 663 task_unlock(tsk); 664 } 665 rcu_read_unlock(); 666 return net; 667 } 668 EXPORT_SYMBOL_GPL(get_net_ns_by_pid); 669 670 static __net_init int net_ns_net_init(struct net *net) 671 { 672 #ifdef CONFIG_NET_NS 673 net->ns.ops = &netns_operations; 674 #endif 675 return ns_alloc_inum(&net->ns); 676 } 677 678 static __net_exit void net_ns_net_exit(struct net *net) 679 { 680 ns_free_inum(&net->ns); 681 } 682 683 static struct pernet_operations __net_initdata net_ns_ops = { 684 .init = net_ns_net_init, 685 .exit = net_ns_net_exit, 686 }; 687 688 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = { 689 [NETNSA_NONE] = { .type = NLA_UNSPEC }, 690 [NETNSA_NSID] = { .type = NLA_S32 }, 691 [NETNSA_PID] = { .type = NLA_U32 }, 692 [NETNSA_FD] = { .type = NLA_U32 }, 693 [NETNSA_TARGET_NSID] = { .type = NLA_S32 }, 694 }; 695 696 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh, 697 struct netlink_ext_ack *extack) 698 { 699 struct net *net = sock_net(skb->sk); 700 struct nlattr *tb[NETNSA_MAX + 1]; 701 struct nlattr *nla; 702 struct net *peer; 703 int nsid, err; 704 705 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb, 706 NETNSA_MAX, rtnl_net_policy, extack); 707 if (err < 0) 708 return err; 709 if (!tb[NETNSA_NSID]) { 710 NL_SET_ERR_MSG(extack, "nsid is missing"); 711 return -EINVAL; 712 } 713 nsid = nla_get_s32(tb[NETNSA_NSID]); 714 715 if (tb[NETNSA_PID]) { 716 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 717 nla = tb[NETNSA_PID]; 718 } else if (tb[NETNSA_FD]) { 719 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 720 nla = tb[NETNSA_FD]; 721 } else { 722 NL_SET_ERR_MSG(extack, "Peer netns reference is missing"); 723 return -EINVAL; 724 } 725 if (IS_ERR(peer)) { 726 NL_SET_BAD_ATTR(extack, nla); 727 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid"); 728 return PTR_ERR(peer); 729 } 730 731 spin_lock_bh(&net->nsid_lock); 732 if (__peernet2id(net, peer) >= 0) { 733 spin_unlock_bh(&net->nsid_lock); 734 err = -EEXIST; 735 NL_SET_BAD_ATTR(extack, nla); 736 NL_SET_ERR_MSG(extack, 737 "Peer netns already has a nsid assigned"); 738 goto out; 739 } 740 741 err = alloc_netid(net, peer, nsid); 742 spin_unlock_bh(&net->nsid_lock); 743 if (err >= 0) { 744 rtnl_net_notifyid(net, RTM_NEWNSID, err); 745 err = 0; 746 } else if (err == -ENOSPC && nsid >= 0) { 747 err = -EEXIST; 748 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]); 749 NL_SET_ERR_MSG(extack, "The specified nsid is already used"); 750 } 751 out: 752 put_net(peer); 753 return err; 754 } 755 756 static int rtnl_net_get_size(void) 757 { 758 return NLMSG_ALIGN(sizeof(struct rtgenmsg)) 759 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */ 760 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */ 761 ; 762 } 763 764 struct net_fill_args { 765 u32 portid; 766 u32 seq; 767 int flags; 768 int cmd; 769 int nsid; 770 bool add_ref; 771 int ref_nsid; 772 }; 773 774 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args) 775 { 776 struct nlmsghdr *nlh; 777 struct rtgenmsg *rth; 778 779 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth), 780 args->flags); 781 if (!nlh) 782 return -EMSGSIZE; 783 784 rth = nlmsg_data(nlh); 785 rth->rtgen_family = AF_UNSPEC; 786 787 if (nla_put_s32(skb, NETNSA_NSID, args->nsid)) 788 goto nla_put_failure; 789 790 if (args->add_ref && 791 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid)) 792 goto nla_put_failure; 793 794 nlmsg_end(skb, nlh); 795 return 0; 796 797 nla_put_failure: 798 nlmsg_cancel(skb, nlh); 799 return -EMSGSIZE; 800 } 801 802 static int rtnl_net_valid_getid_req(struct sk_buff *skb, 803 const struct nlmsghdr *nlh, 804 struct nlattr **tb, 805 struct netlink_ext_ack *extack) 806 { 807 int i, err; 808 809 if (!netlink_strict_get_check(skb)) 810 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), 811 tb, NETNSA_MAX, rtnl_net_policy, 812 extack); 813 814 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb, 815 NETNSA_MAX, rtnl_net_policy, 816 extack); 817 if (err) 818 return err; 819 820 for (i = 0; i <= NETNSA_MAX; i++) { 821 if (!tb[i]) 822 continue; 823 824 switch (i) { 825 case NETNSA_PID: 826 case NETNSA_FD: 827 case NETNSA_NSID: 828 case NETNSA_TARGET_NSID: 829 break; 830 default: 831 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request"); 832 return -EINVAL; 833 } 834 } 835 836 return 0; 837 } 838 839 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh, 840 struct netlink_ext_ack *extack) 841 { 842 struct net *net = sock_net(skb->sk); 843 struct nlattr *tb[NETNSA_MAX + 1]; 844 struct net_fill_args fillargs = { 845 .portid = NETLINK_CB(skb).portid, 846 .seq = nlh->nlmsg_seq, 847 .cmd = RTM_NEWNSID, 848 }; 849 struct net *peer, *target = net; 850 struct nlattr *nla; 851 struct sk_buff *msg; 852 int err; 853 854 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack); 855 if (err < 0) 856 return err; 857 if (tb[NETNSA_PID]) { 858 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 859 nla = tb[NETNSA_PID]; 860 } else if (tb[NETNSA_FD]) { 861 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 862 nla = tb[NETNSA_FD]; 863 } else if (tb[NETNSA_NSID]) { 864 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID])); 865 if (!peer) 866 peer = ERR_PTR(-ENOENT); 867 nla = tb[NETNSA_NSID]; 868 } else { 869 NL_SET_ERR_MSG(extack, "Peer netns reference is missing"); 870 return -EINVAL; 871 } 872 873 if (IS_ERR(peer)) { 874 NL_SET_BAD_ATTR(extack, nla); 875 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid"); 876 return PTR_ERR(peer); 877 } 878 879 if (tb[NETNSA_TARGET_NSID]) { 880 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]); 881 882 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id); 883 if (IS_ERR(target)) { 884 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]); 885 NL_SET_ERR_MSG(extack, 886 "Target netns reference is invalid"); 887 err = PTR_ERR(target); 888 goto out; 889 } 890 fillargs.add_ref = true; 891 fillargs.ref_nsid = peernet2id(net, peer); 892 } 893 894 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 895 if (!msg) { 896 err = -ENOMEM; 897 goto out; 898 } 899 900 fillargs.nsid = peernet2id(target, peer); 901 err = rtnl_net_fill(msg, &fillargs); 902 if (err < 0) 903 goto err_out; 904 905 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid); 906 goto out; 907 908 err_out: 909 nlmsg_free(msg); 910 out: 911 if (fillargs.add_ref) 912 put_net(target); 913 put_net(peer); 914 return err; 915 } 916 917 struct rtnl_net_dump_cb { 918 struct net *tgt_net; 919 struct net *ref_net; 920 struct sk_buff *skb; 921 struct net_fill_args fillargs; 922 int idx; 923 int s_idx; 924 }; 925 926 static int rtnl_net_dumpid_one(int id, void *peer, void *data) 927 { 928 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data; 929 int ret; 930 931 if (net_cb->idx < net_cb->s_idx) 932 goto cont; 933 934 net_cb->fillargs.nsid = id; 935 if (net_cb->fillargs.add_ref) 936 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer); 937 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs); 938 if (ret < 0) 939 return ret; 940 941 cont: 942 net_cb->idx++; 943 return 0; 944 } 945 946 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk, 947 struct rtnl_net_dump_cb *net_cb, 948 struct netlink_callback *cb) 949 { 950 struct netlink_ext_ack *extack = cb->extack; 951 struct nlattr *tb[NETNSA_MAX + 1]; 952 int err, i; 953 954 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb, 955 NETNSA_MAX, rtnl_net_policy, 956 extack); 957 if (err < 0) 958 return err; 959 960 for (i = 0; i <= NETNSA_MAX; i++) { 961 if (!tb[i]) 962 continue; 963 964 if (i == NETNSA_TARGET_NSID) { 965 struct net *net; 966 967 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i])); 968 if (IS_ERR(net)) { 969 NL_SET_BAD_ATTR(extack, tb[i]); 970 NL_SET_ERR_MSG(extack, 971 "Invalid target network namespace id"); 972 return PTR_ERR(net); 973 } 974 net_cb->fillargs.add_ref = true; 975 net_cb->ref_net = net_cb->tgt_net; 976 net_cb->tgt_net = net; 977 } else { 978 NL_SET_BAD_ATTR(extack, tb[i]); 979 NL_SET_ERR_MSG(extack, 980 "Unsupported attribute in dump request"); 981 return -EINVAL; 982 } 983 } 984 985 return 0; 986 } 987 988 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb) 989 { 990 struct rtnl_net_dump_cb net_cb = { 991 .tgt_net = sock_net(skb->sk), 992 .skb = skb, 993 .fillargs = { 994 .portid = NETLINK_CB(cb->skb).portid, 995 .seq = cb->nlh->nlmsg_seq, 996 .flags = NLM_F_MULTI, 997 .cmd = RTM_NEWNSID, 998 }, 999 .idx = 0, 1000 .s_idx = cb->args[0], 1001 }; 1002 int err = 0; 1003 1004 if (cb->strict_check) { 1005 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb); 1006 if (err < 0) 1007 goto end; 1008 } 1009 1010 spin_lock_bh(&net_cb.tgt_net->nsid_lock); 1011 if (net_cb.fillargs.add_ref && 1012 !net_eq(net_cb.ref_net, net_cb.tgt_net) && 1013 !spin_trylock_bh(&net_cb.ref_net->nsid_lock)) { 1014 spin_unlock_bh(&net_cb.tgt_net->nsid_lock); 1015 err = -EAGAIN; 1016 goto end; 1017 } 1018 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb); 1019 if (net_cb.fillargs.add_ref && 1020 !net_eq(net_cb.ref_net, net_cb.tgt_net)) 1021 spin_unlock_bh(&net_cb.ref_net->nsid_lock); 1022 spin_unlock_bh(&net_cb.tgt_net->nsid_lock); 1023 1024 cb->args[0] = net_cb.idx; 1025 end: 1026 if (net_cb.fillargs.add_ref) 1027 put_net(net_cb.tgt_net); 1028 return err < 0 ? err : skb->len; 1029 } 1030 1031 static void rtnl_net_notifyid(struct net *net, int cmd, int id) 1032 { 1033 struct net_fill_args fillargs = { 1034 .cmd = cmd, 1035 .nsid = id, 1036 }; 1037 struct sk_buff *msg; 1038 int err = -ENOMEM; 1039 1040 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 1041 if (!msg) 1042 goto out; 1043 1044 err = rtnl_net_fill(msg, &fillargs); 1045 if (err < 0) 1046 goto err_out; 1047 1048 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0); 1049 return; 1050 1051 err_out: 1052 nlmsg_free(msg); 1053 out: 1054 rtnl_set_sk_err(net, RTNLGRP_NSID, err); 1055 } 1056 1057 static int __init net_ns_init(void) 1058 { 1059 struct net_generic *ng; 1060 1061 #ifdef CONFIG_NET_NS 1062 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), 1063 SMP_CACHE_BYTES, 1064 SLAB_PANIC|SLAB_ACCOUNT, NULL); 1065 1066 /* Create workqueue for cleanup */ 1067 netns_wq = create_singlethread_workqueue("netns"); 1068 if (!netns_wq) 1069 panic("Could not create netns workq"); 1070 #endif 1071 1072 ng = net_alloc_generic(); 1073 if (!ng) 1074 panic("Could not allocate generic netns"); 1075 1076 rcu_assign_pointer(init_net.gen, ng); 1077 1078 down_write(&pernet_ops_rwsem); 1079 if (setup_net(&init_net, &init_user_ns)) 1080 panic("Could not setup the initial network namespace"); 1081 1082 init_net_initialized = true; 1083 up_write(&pernet_ops_rwsem); 1084 1085 if (register_pernet_subsys(&net_ns_ops)) 1086 panic("Could not register network namespace subsystems"); 1087 1088 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, 1089 RTNL_FLAG_DOIT_UNLOCKED); 1090 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid, 1091 RTNL_FLAG_DOIT_UNLOCKED); 1092 1093 return 0; 1094 } 1095 1096 pure_initcall(net_ns_init); 1097 1098 #ifdef CONFIG_NET_NS 1099 static int __register_pernet_operations(struct list_head *list, 1100 struct pernet_operations *ops) 1101 { 1102 struct net *net; 1103 int error; 1104 LIST_HEAD(net_exit_list); 1105 1106 list_add_tail(&ops->list, list); 1107 if (ops->init || (ops->id && ops->size)) { 1108 /* We held write locked pernet_ops_rwsem, and parallel 1109 * setup_net() and cleanup_net() are not possible. 1110 */ 1111 for_each_net(net) { 1112 error = ops_init(ops, net); 1113 if (error) 1114 goto out_undo; 1115 list_add_tail(&net->exit_list, &net_exit_list); 1116 } 1117 } 1118 return 0; 1119 1120 out_undo: 1121 /* If I have an error cleanup all namespaces I initialized */ 1122 list_del(&ops->list); 1123 ops_exit_list(ops, &net_exit_list); 1124 ops_free_list(ops, &net_exit_list); 1125 return error; 1126 } 1127 1128 static void __unregister_pernet_operations(struct pernet_operations *ops) 1129 { 1130 struct net *net; 1131 LIST_HEAD(net_exit_list); 1132 1133 list_del(&ops->list); 1134 /* See comment in __register_pernet_operations() */ 1135 for_each_net(net) 1136 list_add_tail(&net->exit_list, &net_exit_list); 1137 ops_exit_list(ops, &net_exit_list); 1138 ops_free_list(ops, &net_exit_list); 1139 } 1140 1141 #else 1142 1143 static int __register_pernet_operations(struct list_head *list, 1144 struct pernet_operations *ops) 1145 { 1146 if (!init_net_initialized) { 1147 list_add_tail(&ops->list, list); 1148 return 0; 1149 } 1150 1151 return ops_init(ops, &init_net); 1152 } 1153 1154 static void __unregister_pernet_operations(struct pernet_operations *ops) 1155 { 1156 if (!init_net_initialized) { 1157 list_del(&ops->list); 1158 } else { 1159 LIST_HEAD(net_exit_list); 1160 list_add(&init_net.exit_list, &net_exit_list); 1161 ops_exit_list(ops, &net_exit_list); 1162 ops_free_list(ops, &net_exit_list); 1163 } 1164 } 1165 1166 #endif /* CONFIG_NET_NS */ 1167 1168 static DEFINE_IDA(net_generic_ids); 1169 1170 static int register_pernet_operations(struct list_head *list, 1171 struct pernet_operations *ops) 1172 { 1173 int error; 1174 1175 if (ops->id) { 1176 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID, 1177 GFP_KERNEL); 1178 if (error < 0) 1179 return error; 1180 *ops->id = error; 1181 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1); 1182 } 1183 error = __register_pernet_operations(list, ops); 1184 if (error) { 1185 rcu_barrier(); 1186 if (ops->id) 1187 ida_free(&net_generic_ids, *ops->id); 1188 } 1189 1190 return error; 1191 } 1192 1193 static void unregister_pernet_operations(struct pernet_operations *ops) 1194 { 1195 __unregister_pernet_operations(ops); 1196 rcu_barrier(); 1197 if (ops->id) 1198 ida_free(&net_generic_ids, *ops->id); 1199 } 1200 1201 /** 1202 * register_pernet_subsys - register a network namespace subsystem 1203 * @ops: pernet operations structure for the subsystem 1204 * 1205 * Register a subsystem which has init and exit functions 1206 * that are called when network namespaces are created and 1207 * destroyed respectively. 1208 * 1209 * When registered all network namespace init functions are 1210 * called for every existing network namespace. Allowing kernel 1211 * modules to have a race free view of the set of network namespaces. 1212 * 1213 * When a new network namespace is created all of the init 1214 * methods are called in the order in which they were registered. 1215 * 1216 * When a network namespace is destroyed all of the exit methods 1217 * are called in the reverse of the order with which they were 1218 * registered. 1219 */ 1220 int register_pernet_subsys(struct pernet_operations *ops) 1221 { 1222 int error; 1223 down_write(&pernet_ops_rwsem); 1224 error = register_pernet_operations(first_device, ops); 1225 up_write(&pernet_ops_rwsem); 1226 return error; 1227 } 1228 EXPORT_SYMBOL_GPL(register_pernet_subsys); 1229 1230 /** 1231 * unregister_pernet_subsys - unregister a network namespace subsystem 1232 * @ops: pernet operations structure to manipulate 1233 * 1234 * Remove the pernet operations structure from the list to be 1235 * used when network namespaces are created or destroyed. In 1236 * addition run the exit method for all existing network 1237 * namespaces. 1238 */ 1239 void unregister_pernet_subsys(struct pernet_operations *ops) 1240 { 1241 down_write(&pernet_ops_rwsem); 1242 unregister_pernet_operations(ops); 1243 up_write(&pernet_ops_rwsem); 1244 } 1245 EXPORT_SYMBOL_GPL(unregister_pernet_subsys); 1246 1247 /** 1248 * register_pernet_device - register a network namespace device 1249 * @ops: pernet operations structure for the subsystem 1250 * 1251 * Register a device which has init and exit functions 1252 * that are called when network namespaces are created and 1253 * destroyed respectively. 1254 * 1255 * When registered all network namespace init functions are 1256 * called for every existing network namespace. Allowing kernel 1257 * modules to have a race free view of the set of network namespaces. 1258 * 1259 * When a new network namespace is created all of the init 1260 * methods are called in the order in which they were registered. 1261 * 1262 * When a network namespace is destroyed all of the exit methods 1263 * are called in the reverse of the order with which they were 1264 * registered. 1265 */ 1266 int register_pernet_device(struct pernet_operations *ops) 1267 { 1268 int error; 1269 down_write(&pernet_ops_rwsem); 1270 error = register_pernet_operations(&pernet_list, ops); 1271 if (!error && (first_device == &pernet_list)) 1272 first_device = &ops->list; 1273 up_write(&pernet_ops_rwsem); 1274 return error; 1275 } 1276 EXPORT_SYMBOL_GPL(register_pernet_device); 1277 1278 /** 1279 * unregister_pernet_device - unregister a network namespace netdevice 1280 * @ops: pernet operations structure to manipulate 1281 * 1282 * Remove the pernet operations structure from the list to be 1283 * used when network namespaces are created or destroyed. In 1284 * addition run the exit method for all existing network 1285 * namespaces. 1286 */ 1287 void unregister_pernet_device(struct pernet_operations *ops) 1288 { 1289 down_write(&pernet_ops_rwsem); 1290 if (&ops->list == first_device) 1291 first_device = first_device->next; 1292 unregister_pernet_operations(ops); 1293 up_write(&pernet_ops_rwsem); 1294 } 1295 EXPORT_SYMBOL_GPL(unregister_pernet_device); 1296 1297 #ifdef CONFIG_NET_NS 1298 static struct ns_common *netns_get(struct task_struct *task) 1299 { 1300 struct net *net = NULL; 1301 struct nsproxy *nsproxy; 1302 1303 task_lock(task); 1304 nsproxy = task->nsproxy; 1305 if (nsproxy) 1306 net = get_net(nsproxy->net_ns); 1307 task_unlock(task); 1308 1309 return net ? &net->ns : NULL; 1310 } 1311 1312 static inline struct net *to_net_ns(struct ns_common *ns) 1313 { 1314 return container_of(ns, struct net, ns); 1315 } 1316 1317 static void netns_put(struct ns_common *ns) 1318 { 1319 put_net(to_net_ns(ns)); 1320 } 1321 1322 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns) 1323 { 1324 struct net *net = to_net_ns(ns); 1325 1326 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) || 1327 !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 1328 return -EPERM; 1329 1330 put_net(nsproxy->net_ns); 1331 nsproxy->net_ns = get_net(net); 1332 return 0; 1333 } 1334 1335 static struct user_namespace *netns_owner(struct ns_common *ns) 1336 { 1337 return to_net_ns(ns)->user_ns; 1338 } 1339 1340 const struct proc_ns_operations netns_operations = { 1341 .name = "net", 1342 .type = CLONE_NEWNET, 1343 .get = netns_get, 1344 .put = netns_put, 1345 .install = netns_install, 1346 .owner = netns_owner, 1347 }; 1348 #endif 1349