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 21 #include <net/sock.h> 22 #include <net/netlink.h> 23 #include <net/net_namespace.h> 24 #include <net/netns/generic.h> 25 26 /* 27 * Our network namespace constructor/destructor lists 28 */ 29 30 static LIST_HEAD(pernet_list); 31 static struct list_head *first_device = &pernet_list; 32 33 LIST_HEAD(net_namespace_list); 34 EXPORT_SYMBOL_GPL(net_namespace_list); 35 36 struct net init_net = { 37 .count = REFCOUNT_INIT(1), 38 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head), 39 }; 40 EXPORT_SYMBOL(init_net); 41 42 static bool init_net_initialized; 43 static unsigned nr_sync_pernet_ops; 44 /* 45 * net_sem: protects: pernet_list, net_generic_ids, nr_sync_pernet_ops, 46 * init_net_initialized and first_device pointer. 47 */ 48 DECLARE_RWSEM(net_sem); 49 50 #define MIN_PERNET_OPS_ID \ 51 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *)) 52 53 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ 54 55 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS; 56 57 static struct net_generic *net_alloc_generic(void) 58 { 59 struct net_generic *ng; 60 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]); 61 62 ng = kzalloc(generic_size, GFP_KERNEL); 63 if (ng) 64 ng->s.len = max_gen_ptrs; 65 66 return ng; 67 } 68 69 static int net_assign_generic(struct net *net, unsigned int id, void *data) 70 { 71 struct net_generic *ng, *old_ng; 72 73 BUG_ON(id < MIN_PERNET_OPS_ID); 74 75 old_ng = rcu_dereference_protected(net->gen, 76 lockdep_is_held(&net_sem)); 77 if (old_ng->s.len > id) { 78 old_ng->ptr[id] = data; 79 return 0; 80 } 81 82 ng = net_alloc_generic(); 83 if (ng == NULL) 84 return -ENOMEM; 85 86 /* 87 * Some synchronisation notes: 88 * 89 * The net_generic explores the net->gen array inside rcu 90 * read section. Besides once set the net->gen->ptr[x] 91 * pointer never changes (see rules in netns/generic.h). 92 * 93 * That said, we simply duplicate this array and schedule 94 * the old copy for kfree after a grace period. 95 */ 96 97 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID], 98 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *)); 99 ng->ptr[id] = data; 100 101 rcu_assign_pointer(net->gen, ng); 102 kfree_rcu(old_ng, s.rcu); 103 return 0; 104 } 105 106 static int ops_init(const struct pernet_operations *ops, struct net *net) 107 { 108 int err = -ENOMEM; 109 void *data = NULL; 110 111 if (ops->id && ops->size) { 112 data = kzalloc(ops->size, GFP_KERNEL); 113 if (!data) 114 goto out; 115 116 err = net_assign_generic(net, *ops->id, data); 117 if (err) 118 goto cleanup; 119 } 120 err = 0; 121 if (ops->init) 122 err = ops->init(net); 123 if (!err) 124 return 0; 125 126 cleanup: 127 kfree(data); 128 129 out: 130 return err; 131 } 132 133 static void ops_free(const struct pernet_operations *ops, struct net *net) 134 { 135 if (ops->id && ops->size) { 136 kfree(net_generic(net, *ops->id)); 137 } 138 } 139 140 static void ops_exit_list(const struct pernet_operations *ops, 141 struct list_head *net_exit_list) 142 { 143 struct net *net; 144 if (ops->exit) { 145 list_for_each_entry(net, net_exit_list, exit_list) 146 ops->exit(net); 147 } 148 if (ops->exit_batch) 149 ops->exit_batch(net_exit_list); 150 } 151 152 static void ops_free_list(const struct pernet_operations *ops, 153 struct list_head *net_exit_list) 154 { 155 struct net *net; 156 if (ops->size && ops->id) { 157 list_for_each_entry(net, net_exit_list, exit_list) 158 ops_free(ops, net); 159 } 160 } 161 162 /* should be called with nsid_lock held */ 163 static int alloc_netid(struct net *net, struct net *peer, int reqid) 164 { 165 int min = 0, max = 0; 166 167 if (reqid >= 0) { 168 min = reqid; 169 max = reqid + 1; 170 } 171 172 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC); 173 } 174 175 /* This function is used by idr_for_each(). If net is equal to peer, the 176 * function returns the id so that idr_for_each() stops. Because we cannot 177 * returns the id 0 (idr_for_each() will not stop), we return the magic value 178 * NET_ID_ZERO (-1) for it. 179 */ 180 #define NET_ID_ZERO -1 181 static int net_eq_idr(int id, void *net, void *peer) 182 { 183 if (net_eq(net, peer)) 184 return id ? : NET_ID_ZERO; 185 return 0; 186 } 187 188 /* Should be called with nsid_lock held. If a new id is assigned, the bool alloc 189 * is set to true, thus the caller knows that the new id must be notified via 190 * rtnl. 191 */ 192 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc) 193 { 194 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer); 195 bool alloc_it = *alloc; 196 197 *alloc = false; 198 199 /* Magic value for id 0. */ 200 if (id == NET_ID_ZERO) 201 return 0; 202 if (id > 0) 203 return id; 204 205 if (alloc_it) { 206 id = alloc_netid(net, peer, -1); 207 *alloc = true; 208 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED; 209 } 210 211 return NETNSA_NSID_NOT_ASSIGNED; 212 } 213 214 /* should be called with nsid_lock held */ 215 static int __peernet2id(struct net *net, struct net *peer) 216 { 217 bool no = false; 218 219 return __peernet2id_alloc(net, peer, &no); 220 } 221 222 static void rtnl_net_notifyid(struct net *net, int cmd, int id); 223 /* This function returns the id of a peer netns. If no id is assigned, one will 224 * be allocated and returned. 225 */ 226 int peernet2id_alloc(struct net *net, struct net *peer) 227 { 228 bool alloc = false, alive = false; 229 int id; 230 231 if (refcount_read(&net->count) == 0) 232 return NETNSA_NSID_NOT_ASSIGNED; 233 spin_lock_bh(&net->nsid_lock); 234 /* 235 * When peer is obtained from RCU lists, we may race with 236 * its cleanup. Check whether it's alive, and this guarantees 237 * we never hash a peer back to net->netns_ids, after it has 238 * just been idr_remove()'d from there in cleanup_net(). 239 */ 240 if (maybe_get_net(peer)) 241 alive = alloc = true; 242 id = __peernet2id_alloc(net, peer, &alloc); 243 spin_unlock_bh(&net->nsid_lock); 244 if (alloc && id >= 0) 245 rtnl_net_notifyid(net, RTM_NEWNSID, id); 246 if (alive) 247 put_net(peer); 248 return id; 249 } 250 EXPORT_SYMBOL_GPL(peernet2id_alloc); 251 252 /* This function returns, if assigned, the id of a peer netns. */ 253 int peernet2id(struct net *net, struct net *peer) 254 { 255 int id; 256 257 spin_lock_bh(&net->nsid_lock); 258 id = __peernet2id(net, peer); 259 spin_unlock_bh(&net->nsid_lock); 260 return id; 261 } 262 EXPORT_SYMBOL(peernet2id); 263 264 /* This function returns true is the peer netns has an id assigned into the 265 * current netns. 266 */ 267 bool peernet_has_id(struct net *net, struct net *peer) 268 { 269 return peernet2id(net, peer) >= 0; 270 } 271 272 struct net *get_net_ns_by_id(struct net *net, int id) 273 { 274 struct net *peer; 275 276 if (id < 0) 277 return NULL; 278 279 rcu_read_lock(); 280 peer = idr_find(&net->netns_ids, id); 281 if (peer) 282 peer = maybe_get_net(peer); 283 rcu_read_unlock(); 284 285 return peer; 286 } 287 288 /* 289 * setup_net runs the initializers for the network namespace object. 290 */ 291 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns) 292 { 293 /* Must be called with net_sem held */ 294 const struct pernet_operations *ops, *saved_ops; 295 int error = 0; 296 LIST_HEAD(net_exit_list); 297 298 refcount_set(&net->count, 1); 299 refcount_set(&net->passive, 1); 300 net->dev_base_seq = 1; 301 net->user_ns = user_ns; 302 idr_init(&net->netns_ids); 303 spin_lock_init(&net->nsid_lock); 304 305 list_for_each_entry(ops, &pernet_list, list) { 306 error = ops_init(ops, net); 307 if (error < 0) 308 goto out_undo; 309 } 310 rtnl_lock(); 311 list_add_tail_rcu(&net->list, &net_namespace_list); 312 rtnl_unlock(); 313 out: 314 return error; 315 316 out_undo: 317 /* Walk through the list backwards calling the exit functions 318 * for the pernet modules whose init functions did not fail. 319 */ 320 list_add(&net->exit_list, &net_exit_list); 321 saved_ops = ops; 322 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 323 ops_exit_list(ops, &net_exit_list); 324 325 ops = saved_ops; 326 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 327 ops_free_list(ops, &net_exit_list); 328 329 rcu_barrier(); 330 goto out; 331 } 332 333 static int __net_init net_defaults_init_net(struct net *net) 334 { 335 net->core.sysctl_somaxconn = SOMAXCONN; 336 return 0; 337 } 338 339 static struct pernet_operations net_defaults_ops = { 340 .init = net_defaults_init_net, 341 .async = true, 342 }; 343 344 static __init int net_defaults_init(void) 345 { 346 if (register_pernet_subsys(&net_defaults_ops)) 347 panic("Cannot initialize net default settings"); 348 349 return 0; 350 } 351 352 core_initcall(net_defaults_init); 353 354 #ifdef CONFIG_NET_NS 355 static struct ucounts *inc_net_namespaces(struct user_namespace *ns) 356 { 357 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES); 358 } 359 360 static void dec_net_namespaces(struct ucounts *ucounts) 361 { 362 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES); 363 } 364 365 static struct kmem_cache *net_cachep; 366 static struct workqueue_struct *netns_wq; 367 368 static struct net *net_alloc(void) 369 { 370 struct net *net = NULL; 371 struct net_generic *ng; 372 373 ng = net_alloc_generic(); 374 if (!ng) 375 goto out; 376 377 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); 378 if (!net) 379 goto out_free; 380 381 rcu_assign_pointer(net->gen, ng); 382 out: 383 return net; 384 385 out_free: 386 kfree(ng); 387 goto out; 388 } 389 390 static void net_free(struct net *net) 391 { 392 kfree(rcu_access_pointer(net->gen)); 393 kmem_cache_free(net_cachep, net); 394 } 395 396 void net_drop_ns(void *p) 397 { 398 struct net *ns = p; 399 if (ns && refcount_dec_and_test(&ns->passive)) 400 net_free(ns); 401 } 402 403 struct net *copy_net_ns(unsigned long flags, 404 struct user_namespace *user_ns, struct net *old_net) 405 { 406 struct ucounts *ucounts; 407 struct net *net; 408 unsigned write; 409 int rv; 410 411 if (!(flags & CLONE_NEWNET)) 412 return get_net(old_net); 413 414 ucounts = inc_net_namespaces(user_ns); 415 if (!ucounts) 416 return ERR_PTR(-ENOSPC); 417 418 net = net_alloc(); 419 if (!net) { 420 rv = -ENOMEM; 421 goto dec_ucounts; 422 } 423 refcount_set(&net->passive, 1); 424 net->ucounts = ucounts; 425 get_user_ns(user_ns); 426 again: 427 write = READ_ONCE(nr_sync_pernet_ops); 428 if (write) 429 rv = down_write_killable(&net_sem); 430 else 431 rv = down_read_killable(&net_sem); 432 if (rv < 0) 433 goto put_userns; 434 435 if (!write && unlikely(READ_ONCE(nr_sync_pernet_ops))) { 436 up_read(&net_sem); 437 goto again; 438 } 439 rv = setup_net(net, user_ns); 440 441 if (write) 442 up_write(&net_sem); 443 else 444 up_read(&net_sem); 445 446 if (rv < 0) { 447 put_userns: 448 put_user_ns(user_ns); 449 net_drop_ns(net); 450 dec_ucounts: 451 dec_net_namespaces(ucounts); 452 return ERR_PTR(rv); 453 } 454 return net; 455 } 456 457 static void unhash_nsid(struct net *net, struct net *last) 458 { 459 struct net *tmp; 460 /* This function is only called from cleanup_net() work, 461 * and this work is the only process, that may delete 462 * a net from net_namespace_list. So, when the below 463 * is executing, the list may only grow. Thus, we do not 464 * use for_each_net_rcu() or rtnl_lock(). 465 */ 466 for_each_net(tmp) { 467 int id; 468 469 spin_lock_bh(&tmp->nsid_lock); 470 id = __peernet2id(tmp, net); 471 if (id >= 0) 472 idr_remove(&tmp->netns_ids, id); 473 spin_unlock_bh(&tmp->nsid_lock); 474 if (id >= 0) 475 rtnl_net_notifyid(tmp, RTM_DELNSID, id); 476 if (tmp == last) 477 break; 478 } 479 spin_lock_bh(&net->nsid_lock); 480 idr_destroy(&net->netns_ids); 481 spin_unlock_bh(&net->nsid_lock); 482 } 483 484 static DEFINE_SPINLOCK(cleanup_list_lock); 485 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */ 486 487 static void cleanup_net(struct work_struct *work) 488 { 489 const struct pernet_operations *ops; 490 struct net *net, *tmp, *last; 491 struct list_head net_kill_list; 492 LIST_HEAD(net_exit_list); 493 unsigned write; 494 495 /* Atomically snapshot the list of namespaces to cleanup */ 496 spin_lock_irq(&cleanup_list_lock); 497 list_replace_init(&cleanup_list, &net_kill_list); 498 spin_unlock_irq(&cleanup_list_lock); 499 again: 500 write = READ_ONCE(nr_sync_pernet_ops); 501 if (write) 502 down_write(&net_sem); 503 else 504 down_read(&net_sem); 505 506 if (!write && unlikely(READ_ONCE(nr_sync_pernet_ops))) { 507 up_read(&net_sem); 508 goto again; 509 } 510 511 /* Don't let anyone else find us. */ 512 rtnl_lock(); 513 list_for_each_entry(net, &net_kill_list, cleanup_list) 514 list_del_rcu(&net->list); 515 /* Cache last net. After we unlock rtnl, no one new net 516 * added to net_namespace_list can assign nsid pointer 517 * to a net from net_kill_list (see peernet2id_alloc()). 518 * So, we skip them in unhash_nsid(). 519 * 520 * Note, that unhash_nsid() does not delete nsid links 521 * between net_kill_list's nets, as they've already 522 * deleted from net_namespace_list. But, this would be 523 * useless anyway, as netns_ids are destroyed there. 524 */ 525 last = list_last_entry(&net_namespace_list, struct net, list); 526 rtnl_unlock(); 527 528 list_for_each_entry(net, &net_kill_list, cleanup_list) { 529 unhash_nsid(net, last); 530 list_add_tail(&net->exit_list, &net_exit_list); 531 } 532 533 /* 534 * Another CPU might be rcu-iterating the list, wait for it. 535 * This needs to be before calling the exit() notifiers, so 536 * the rcu_barrier() below isn't sufficient alone. 537 */ 538 synchronize_rcu(); 539 540 /* Run all of the network namespace exit methods */ 541 list_for_each_entry_reverse(ops, &pernet_list, list) 542 ops_exit_list(ops, &net_exit_list); 543 544 /* Free the net generic variables */ 545 list_for_each_entry_reverse(ops, &pernet_list, list) 546 ops_free_list(ops, &net_exit_list); 547 548 if (write) 549 up_write(&net_sem); 550 else 551 up_read(&net_sem); 552 553 /* Ensure there are no outstanding rcu callbacks using this 554 * network namespace. 555 */ 556 rcu_barrier(); 557 558 /* Finally it is safe to free my network namespace structure */ 559 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) { 560 list_del_init(&net->exit_list); 561 dec_net_namespaces(net->ucounts); 562 put_user_ns(net->user_ns); 563 net_drop_ns(net); 564 } 565 } 566 567 /** 568 * net_ns_barrier - wait until concurrent net_cleanup_work is done 569 * 570 * cleanup_net runs from work queue and will first remove namespaces 571 * from the global list, then run net exit functions. 572 * 573 * Call this in module exit path to make sure that all netns 574 * ->exit ops have been invoked before the function is removed. 575 */ 576 void net_ns_barrier(void) 577 { 578 down_write(&net_sem); 579 up_write(&net_sem); 580 } 581 EXPORT_SYMBOL(net_ns_barrier); 582 583 static DECLARE_WORK(net_cleanup_work, cleanup_net); 584 585 void __put_net(struct net *net) 586 { 587 /* Cleanup the network namespace in process context */ 588 unsigned long flags; 589 590 spin_lock_irqsave(&cleanup_list_lock, flags); 591 list_add(&net->cleanup_list, &cleanup_list); 592 spin_unlock_irqrestore(&cleanup_list_lock, flags); 593 594 queue_work(netns_wq, &net_cleanup_work); 595 } 596 EXPORT_SYMBOL_GPL(__put_net); 597 598 struct net *get_net_ns_by_fd(int fd) 599 { 600 struct file *file; 601 struct ns_common *ns; 602 struct net *net; 603 604 file = proc_ns_fget(fd); 605 if (IS_ERR(file)) 606 return ERR_CAST(file); 607 608 ns = get_proc_ns(file_inode(file)); 609 if (ns->ops == &netns_operations) 610 net = get_net(container_of(ns, struct net, ns)); 611 else 612 net = ERR_PTR(-EINVAL); 613 614 fput(file); 615 return net; 616 } 617 618 #else 619 struct net *get_net_ns_by_fd(int fd) 620 { 621 return ERR_PTR(-EINVAL); 622 } 623 #endif 624 EXPORT_SYMBOL_GPL(get_net_ns_by_fd); 625 626 struct net *get_net_ns_by_pid(pid_t pid) 627 { 628 struct task_struct *tsk; 629 struct net *net; 630 631 /* Lookup the network namespace */ 632 net = ERR_PTR(-ESRCH); 633 rcu_read_lock(); 634 tsk = find_task_by_vpid(pid); 635 if (tsk) { 636 struct nsproxy *nsproxy; 637 task_lock(tsk); 638 nsproxy = tsk->nsproxy; 639 if (nsproxy) 640 net = get_net(nsproxy->net_ns); 641 task_unlock(tsk); 642 } 643 rcu_read_unlock(); 644 return net; 645 } 646 EXPORT_SYMBOL_GPL(get_net_ns_by_pid); 647 648 static __net_init int net_ns_net_init(struct net *net) 649 { 650 #ifdef CONFIG_NET_NS 651 net->ns.ops = &netns_operations; 652 #endif 653 return ns_alloc_inum(&net->ns); 654 } 655 656 static __net_exit void net_ns_net_exit(struct net *net) 657 { 658 ns_free_inum(&net->ns); 659 } 660 661 static struct pernet_operations __net_initdata net_ns_ops = { 662 .init = net_ns_net_init, 663 .exit = net_ns_net_exit, 664 .async = true, 665 }; 666 667 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = { 668 [NETNSA_NONE] = { .type = NLA_UNSPEC }, 669 [NETNSA_NSID] = { .type = NLA_S32 }, 670 [NETNSA_PID] = { .type = NLA_U32 }, 671 [NETNSA_FD] = { .type = NLA_U32 }, 672 }; 673 674 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh, 675 struct netlink_ext_ack *extack) 676 { 677 struct net *net = sock_net(skb->sk); 678 struct nlattr *tb[NETNSA_MAX + 1]; 679 struct nlattr *nla; 680 struct net *peer; 681 int nsid, err; 682 683 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX, 684 rtnl_net_policy, extack); 685 if (err < 0) 686 return err; 687 if (!tb[NETNSA_NSID]) { 688 NL_SET_ERR_MSG(extack, "nsid is missing"); 689 return -EINVAL; 690 } 691 nsid = nla_get_s32(tb[NETNSA_NSID]); 692 693 if (tb[NETNSA_PID]) { 694 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 695 nla = tb[NETNSA_PID]; 696 } else if (tb[NETNSA_FD]) { 697 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 698 nla = tb[NETNSA_FD]; 699 } else { 700 NL_SET_ERR_MSG(extack, "Peer netns reference is missing"); 701 return -EINVAL; 702 } 703 if (IS_ERR(peer)) { 704 NL_SET_BAD_ATTR(extack, nla); 705 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid"); 706 return PTR_ERR(peer); 707 } 708 709 spin_lock_bh(&net->nsid_lock); 710 if (__peernet2id(net, peer) >= 0) { 711 spin_unlock_bh(&net->nsid_lock); 712 err = -EEXIST; 713 NL_SET_BAD_ATTR(extack, nla); 714 NL_SET_ERR_MSG(extack, 715 "Peer netns already has a nsid assigned"); 716 goto out; 717 } 718 719 err = alloc_netid(net, peer, nsid); 720 spin_unlock_bh(&net->nsid_lock); 721 if (err >= 0) { 722 rtnl_net_notifyid(net, RTM_NEWNSID, err); 723 err = 0; 724 } else if (err == -ENOSPC && nsid >= 0) { 725 err = -EEXIST; 726 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]); 727 NL_SET_ERR_MSG(extack, "The specified nsid is already used"); 728 } 729 out: 730 put_net(peer); 731 return err; 732 } 733 734 static int rtnl_net_get_size(void) 735 { 736 return NLMSG_ALIGN(sizeof(struct rtgenmsg)) 737 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */ 738 ; 739 } 740 741 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags, 742 int cmd, struct net *net, int nsid) 743 { 744 struct nlmsghdr *nlh; 745 struct rtgenmsg *rth; 746 747 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags); 748 if (!nlh) 749 return -EMSGSIZE; 750 751 rth = nlmsg_data(nlh); 752 rth->rtgen_family = AF_UNSPEC; 753 754 if (nla_put_s32(skb, NETNSA_NSID, nsid)) 755 goto nla_put_failure; 756 757 nlmsg_end(skb, nlh); 758 return 0; 759 760 nla_put_failure: 761 nlmsg_cancel(skb, nlh); 762 return -EMSGSIZE; 763 } 764 765 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh, 766 struct netlink_ext_ack *extack) 767 { 768 struct net *net = sock_net(skb->sk); 769 struct nlattr *tb[NETNSA_MAX + 1]; 770 struct nlattr *nla; 771 struct sk_buff *msg; 772 struct net *peer; 773 int err, id; 774 775 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX, 776 rtnl_net_policy, extack); 777 if (err < 0) 778 return err; 779 if (tb[NETNSA_PID]) { 780 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 781 nla = tb[NETNSA_PID]; 782 } else if (tb[NETNSA_FD]) { 783 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 784 nla = tb[NETNSA_FD]; 785 } else { 786 NL_SET_ERR_MSG(extack, "Peer netns reference is missing"); 787 return -EINVAL; 788 } 789 790 if (IS_ERR(peer)) { 791 NL_SET_BAD_ATTR(extack, nla); 792 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid"); 793 return PTR_ERR(peer); 794 } 795 796 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 797 if (!msg) { 798 err = -ENOMEM; 799 goto out; 800 } 801 802 id = peernet2id(net, peer); 803 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0, 804 RTM_NEWNSID, net, id); 805 if (err < 0) 806 goto err_out; 807 808 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid); 809 goto out; 810 811 err_out: 812 nlmsg_free(msg); 813 out: 814 put_net(peer); 815 return err; 816 } 817 818 struct rtnl_net_dump_cb { 819 struct net *net; 820 struct sk_buff *skb; 821 struct netlink_callback *cb; 822 int idx; 823 int s_idx; 824 }; 825 826 static int rtnl_net_dumpid_one(int id, void *peer, void *data) 827 { 828 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data; 829 int ret; 830 831 if (net_cb->idx < net_cb->s_idx) 832 goto cont; 833 834 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid, 835 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI, 836 RTM_NEWNSID, net_cb->net, id); 837 if (ret < 0) 838 return ret; 839 840 cont: 841 net_cb->idx++; 842 return 0; 843 } 844 845 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb) 846 { 847 struct net *net = sock_net(skb->sk); 848 struct rtnl_net_dump_cb net_cb = { 849 .net = net, 850 .skb = skb, 851 .cb = cb, 852 .idx = 0, 853 .s_idx = cb->args[0], 854 }; 855 856 spin_lock_bh(&net->nsid_lock); 857 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb); 858 spin_unlock_bh(&net->nsid_lock); 859 860 cb->args[0] = net_cb.idx; 861 return skb->len; 862 } 863 864 static void rtnl_net_notifyid(struct net *net, int cmd, int id) 865 { 866 struct sk_buff *msg; 867 int err = -ENOMEM; 868 869 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 870 if (!msg) 871 goto out; 872 873 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id); 874 if (err < 0) 875 goto err_out; 876 877 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0); 878 return; 879 880 err_out: 881 nlmsg_free(msg); 882 out: 883 rtnl_set_sk_err(net, RTNLGRP_NSID, err); 884 } 885 886 static int __init net_ns_init(void) 887 { 888 struct net_generic *ng; 889 890 #ifdef CONFIG_NET_NS 891 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), 892 SMP_CACHE_BYTES, 893 SLAB_PANIC, NULL); 894 895 /* Create workqueue for cleanup */ 896 netns_wq = create_singlethread_workqueue("netns"); 897 if (!netns_wq) 898 panic("Could not create netns workq"); 899 #endif 900 901 ng = net_alloc_generic(); 902 if (!ng) 903 panic("Could not allocate generic netns"); 904 905 rcu_assign_pointer(init_net.gen, ng); 906 907 down_write(&net_sem); 908 if (setup_net(&init_net, &init_user_ns)) 909 panic("Could not setup the initial network namespace"); 910 911 init_net_initialized = true; 912 up_write(&net_sem); 913 914 register_pernet_subsys(&net_ns_ops); 915 916 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, 917 RTNL_FLAG_DOIT_UNLOCKED); 918 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid, 919 RTNL_FLAG_DOIT_UNLOCKED); 920 921 return 0; 922 } 923 924 pure_initcall(net_ns_init); 925 926 #ifdef CONFIG_NET_NS 927 static int __register_pernet_operations(struct list_head *list, 928 struct pernet_operations *ops) 929 { 930 struct net *net; 931 int error; 932 LIST_HEAD(net_exit_list); 933 934 list_add_tail(&ops->list, list); 935 if (ops->init || (ops->id && ops->size)) { 936 for_each_net(net) { 937 error = ops_init(ops, net); 938 if (error) 939 goto out_undo; 940 list_add_tail(&net->exit_list, &net_exit_list); 941 } 942 } 943 return 0; 944 945 out_undo: 946 /* If I have an error cleanup all namespaces I initialized */ 947 list_del(&ops->list); 948 ops_exit_list(ops, &net_exit_list); 949 ops_free_list(ops, &net_exit_list); 950 return error; 951 } 952 953 static void __unregister_pernet_operations(struct pernet_operations *ops) 954 { 955 struct net *net; 956 LIST_HEAD(net_exit_list); 957 958 list_del(&ops->list); 959 for_each_net(net) 960 list_add_tail(&net->exit_list, &net_exit_list); 961 ops_exit_list(ops, &net_exit_list); 962 ops_free_list(ops, &net_exit_list); 963 } 964 965 #else 966 967 static int __register_pernet_operations(struct list_head *list, 968 struct pernet_operations *ops) 969 { 970 if (!init_net_initialized) { 971 list_add_tail(&ops->list, list); 972 return 0; 973 } 974 975 return ops_init(ops, &init_net); 976 } 977 978 static void __unregister_pernet_operations(struct pernet_operations *ops) 979 { 980 if (!init_net_initialized) { 981 list_del(&ops->list); 982 } else { 983 LIST_HEAD(net_exit_list); 984 list_add(&init_net.exit_list, &net_exit_list); 985 ops_exit_list(ops, &net_exit_list); 986 ops_free_list(ops, &net_exit_list); 987 } 988 } 989 990 #endif /* CONFIG_NET_NS */ 991 992 static DEFINE_IDA(net_generic_ids); 993 994 static int register_pernet_operations(struct list_head *list, 995 struct pernet_operations *ops) 996 { 997 int error; 998 999 if (ops->id) { 1000 again: 1001 error = ida_get_new_above(&net_generic_ids, MIN_PERNET_OPS_ID, ops->id); 1002 if (error < 0) { 1003 if (error == -EAGAIN) { 1004 ida_pre_get(&net_generic_ids, GFP_KERNEL); 1005 goto again; 1006 } 1007 return error; 1008 } 1009 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1); 1010 } 1011 error = __register_pernet_operations(list, ops); 1012 if (error) { 1013 rcu_barrier(); 1014 if (ops->id) 1015 ida_remove(&net_generic_ids, *ops->id); 1016 } else if (!ops->async) { 1017 pr_info_once("Pernet operations %ps are sync.\n", ops); 1018 nr_sync_pernet_ops++; 1019 } 1020 1021 return error; 1022 } 1023 1024 static void unregister_pernet_operations(struct pernet_operations *ops) 1025 { 1026 if (!ops->async) 1027 BUG_ON(nr_sync_pernet_ops-- == 0); 1028 __unregister_pernet_operations(ops); 1029 rcu_barrier(); 1030 if (ops->id) 1031 ida_remove(&net_generic_ids, *ops->id); 1032 } 1033 1034 /** 1035 * register_pernet_subsys - register a network namespace subsystem 1036 * @ops: pernet operations structure for the subsystem 1037 * 1038 * Register a subsystem which has init and exit functions 1039 * that are called when network namespaces are created and 1040 * destroyed respectively. 1041 * 1042 * When registered all network namespace init functions are 1043 * called for every existing network namespace. Allowing kernel 1044 * modules to have a race free view of the set of network namespaces. 1045 * 1046 * When a new network namespace is created all of the init 1047 * methods are called in the order in which they were registered. 1048 * 1049 * When a network namespace is destroyed all of the exit methods 1050 * are called in the reverse of the order with which they were 1051 * registered. 1052 */ 1053 int register_pernet_subsys(struct pernet_operations *ops) 1054 { 1055 int error; 1056 down_write(&net_sem); 1057 error = register_pernet_operations(first_device, ops); 1058 up_write(&net_sem); 1059 return error; 1060 } 1061 EXPORT_SYMBOL_GPL(register_pernet_subsys); 1062 1063 /** 1064 * unregister_pernet_subsys - unregister a network namespace subsystem 1065 * @ops: pernet operations structure to manipulate 1066 * 1067 * Remove the pernet operations structure from the list to be 1068 * used when network namespaces are created or destroyed. In 1069 * addition run the exit method for all existing network 1070 * namespaces. 1071 */ 1072 void unregister_pernet_subsys(struct pernet_operations *ops) 1073 { 1074 down_write(&net_sem); 1075 unregister_pernet_operations(ops); 1076 up_write(&net_sem); 1077 } 1078 EXPORT_SYMBOL_GPL(unregister_pernet_subsys); 1079 1080 /** 1081 * register_pernet_device - register a network namespace device 1082 * @ops: pernet operations structure for the subsystem 1083 * 1084 * Register a device which has init and exit functions 1085 * that are called when network namespaces are created and 1086 * destroyed respectively. 1087 * 1088 * When registered all network namespace init functions are 1089 * called for every existing network namespace. Allowing kernel 1090 * modules to have a race free view of the set of network namespaces. 1091 * 1092 * When a new network namespace is created all of the init 1093 * methods are called in the order in which they were registered. 1094 * 1095 * When a network namespace is destroyed all of the exit methods 1096 * are called in the reverse of the order with which they were 1097 * registered. 1098 */ 1099 int register_pernet_device(struct pernet_operations *ops) 1100 { 1101 int error; 1102 down_write(&net_sem); 1103 error = register_pernet_operations(&pernet_list, ops); 1104 if (!error && (first_device == &pernet_list)) 1105 first_device = &ops->list; 1106 up_write(&net_sem); 1107 return error; 1108 } 1109 EXPORT_SYMBOL_GPL(register_pernet_device); 1110 1111 /** 1112 * unregister_pernet_device - unregister a network namespace netdevice 1113 * @ops: pernet operations structure to manipulate 1114 * 1115 * Remove the pernet operations structure from the list to be 1116 * used when network namespaces are created or destroyed. In 1117 * addition run the exit method for all existing network 1118 * namespaces. 1119 */ 1120 void unregister_pernet_device(struct pernet_operations *ops) 1121 { 1122 down_write(&net_sem); 1123 if (&ops->list == first_device) 1124 first_device = first_device->next; 1125 unregister_pernet_operations(ops); 1126 up_write(&net_sem); 1127 } 1128 EXPORT_SYMBOL_GPL(unregister_pernet_device); 1129 1130 #ifdef CONFIG_NET_NS 1131 static struct ns_common *netns_get(struct task_struct *task) 1132 { 1133 struct net *net = NULL; 1134 struct nsproxy *nsproxy; 1135 1136 task_lock(task); 1137 nsproxy = task->nsproxy; 1138 if (nsproxy) 1139 net = get_net(nsproxy->net_ns); 1140 task_unlock(task); 1141 1142 return net ? &net->ns : NULL; 1143 } 1144 1145 static inline struct net *to_net_ns(struct ns_common *ns) 1146 { 1147 return container_of(ns, struct net, ns); 1148 } 1149 1150 static void netns_put(struct ns_common *ns) 1151 { 1152 put_net(to_net_ns(ns)); 1153 } 1154 1155 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns) 1156 { 1157 struct net *net = to_net_ns(ns); 1158 1159 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) || 1160 !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 1161 return -EPERM; 1162 1163 put_net(nsproxy->net_ns); 1164 nsproxy->net_ns = get_net(net); 1165 return 0; 1166 } 1167 1168 static struct user_namespace *netns_owner(struct ns_common *ns) 1169 { 1170 return to_net_ns(ns)->user_ns; 1171 } 1172 1173 const struct proc_ns_operations netns_operations = { 1174 .name = "net", 1175 .type = CLONE_NEWNET, 1176 .get = netns_get, 1177 .put = netns_put, 1178 .install = netns_install, 1179 .owner = netns_owner, 1180 }; 1181 #endif 1182