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