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