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 out: 307 return error; 308 309 out_undo: 310 /* Walk through the list backwards calling the exit functions 311 * for the pernet modules whose init functions did not fail. 312 */ 313 list_add(&net->exit_list, &net_exit_list); 314 saved_ops = ops; 315 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 316 ops_exit_list(ops, &net_exit_list); 317 318 ops = saved_ops; 319 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 320 ops_free_list(ops, &net_exit_list); 321 322 rcu_barrier(); 323 goto out; 324 } 325 326 static int __net_init net_defaults_init_net(struct net *net) 327 { 328 net->core.sysctl_somaxconn = SOMAXCONN; 329 return 0; 330 } 331 332 static struct pernet_operations net_defaults_ops = { 333 .init = net_defaults_init_net, 334 }; 335 336 static __init int net_defaults_init(void) 337 { 338 if (register_pernet_subsys(&net_defaults_ops)) 339 panic("Cannot initialize net default settings"); 340 341 return 0; 342 } 343 344 core_initcall(net_defaults_init); 345 346 #ifdef CONFIG_NET_NS 347 static struct ucounts *inc_net_namespaces(struct user_namespace *ns) 348 { 349 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES); 350 } 351 352 static void dec_net_namespaces(struct ucounts *ucounts) 353 { 354 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES); 355 } 356 357 static struct kmem_cache *net_cachep; 358 static struct workqueue_struct *netns_wq; 359 360 static struct net *net_alloc(void) 361 { 362 struct net *net = NULL; 363 struct net_generic *ng; 364 365 ng = net_alloc_generic(); 366 if (!ng) 367 goto out; 368 369 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); 370 if (!net) 371 goto out_free; 372 373 rcu_assign_pointer(net->gen, ng); 374 out: 375 return net; 376 377 out_free: 378 kfree(ng); 379 goto out; 380 } 381 382 static void net_free(struct net *net) 383 { 384 kfree(rcu_access_pointer(net->gen)); 385 kmem_cache_free(net_cachep, net); 386 } 387 388 void net_drop_ns(void *p) 389 { 390 struct net *ns = p; 391 if (ns && refcount_dec_and_test(&ns->passive)) 392 net_free(ns); 393 } 394 395 struct net *copy_net_ns(unsigned long flags, 396 struct user_namespace *user_ns, struct net *old_net) 397 { 398 struct ucounts *ucounts; 399 struct net *net; 400 int rv; 401 402 if (!(flags & CLONE_NEWNET)) 403 return get_net(old_net); 404 405 ucounts = inc_net_namespaces(user_ns); 406 if (!ucounts) 407 return ERR_PTR(-ENOSPC); 408 409 net = net_alloc(); 410 if (!net) { 411 dec_net_namespaces(ucounts); 412 return ERR_PTR(-ENOMEM); 413 } 414 415 get_user_ns(user_ns); 416 417 rv = mutex_lock_killable(&net_mutex); 418 if (rv < 0) { 419 net_free(net); 420 dec_net_namespaces(ucounts); 421 put_user_ns(user_ns); 422 return ERR_PTR(rv); 423 } 424 425 net->ucounts = ucounts; 426 rv = setup_net(net, user_ns); 427 if (rv == 0) { 428 rtnl_lock(); 429 list_add_tail_rcu(&net->list, &net_namespace_list); 430 rtnl_unlock(); 431 } 432 mutex_unlock(&net_mutex); 433 if (rv < 0) { 434 dec_net_namespaces(ucounts); 435 put_user_ns(user_ns); 436 net_drop_ns(net); 437 return ERR_PTR(rv); 438 } 439 return net; 440 } 441 442 static DEFINE_SPINLOCK(cleanup_list_lock); 443 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */ 444 445 static void cleanup_net(struct work_struct *work) 446 { 447 const struct pernet_operations *ops; 448 struct net *net, *tmp; 449 struct list_head net_kill_list; 450 LIST_HEAD(net_exit_list); 451 452 /* Atomically snapshot the list of namespaces to cleanup */ 453 spin_lock_irq(&cleanup_list_lock); 454 list_replace_init(&cleanup_list, &net_kill_list); 455 spin_unlock_irq(&cleanup_list_lock); 456 457 mutex_lock(&net_mutex); 458 459 /* Don't let anyone else find us. */ 460 rtnl_lock(); 461 list_for_each_entry(net, &net_kill_list, cleanup_list) { 462 list_del_rcu(&net->list); 463 list_add_tail(&net->exit_list, &net_exit_list); 464 for_each_net(tmp) { 465 int id; 466 467 spin_lock_bh(&tmp->nsid_lock); 468 id = __peernet2id(tmp, net); 469 if (id >= 0) 470 idr_remove(&tmp->netns_ids, id); 471 spin_unlock_bh(&tmp->nsid_lock); 472 if (id >= 0) 473 rtnl_net_notifyid(tmp, RTM_DELNSID, id); 474 } 475 spin_lock_bh(&net->nsid_lock); 476 idr_destroy(&net->netns_ids); 477 spin_unlock_bh(&net->nsid_lock); 478 479 } 480 rtnl_unlock(); 481 482 /* 483 * Another CPU might be rcu-iterating the list, wait for it. 484 * This needs to be before calling the exit() notifiers, so 485 * the rcu_barrier() below isn't sufficient alone. 486 */ 487 synchronize_rcu(); 488 489 /* Run all of the network namespace exit methods */ 490 list_for_each_entry_reverse(ops, &pernet_list, list) 491 ops_exit_list(ops, &net_exit_list); 492 493 /* Free the net generic variables */ 494 list_for_each_entry_reverse(ops, &pernet_list, list) 495 ops_free_list(ops, &net_exit_list); 496 497 mutex_unlock(&net_mutex); 498 499 /* Ensure there are no outstanding rcu callbacks using this 500 * network namespace. 501 */ 502 rcu_barrier(); 503 504 /* Finally it is safe to free my network namespace structure */ 505 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) { 506 list_del_init(&net->exit_list); 507 dec_net_namespaces(net->ucounts); 508 put_user_ns(net->user_ns); 509 net_drop_ns(net); 510 } 511 } 512 513 /** 514 * net_ns_barrier - wait until concurrent net_cleanup_work is done 515 * 516 * cleanup_net runs from work queue and will first remove namespaces 517 * from the global list, then run net exit functions. 518 * 519 * Call this in module exit path to make sure that all netns 520 * ->exit ops have been invoked before the function is removed. 521 */ 522 void net_ns_barrier(void) 523 { 524 mutex_lock(&net_mutex); 525 mutex_unlock(&net_mutex); 526 } 527 EXPORT_SYMBOL(net_ns_barrier); 528 529 static DECLARE_WORK(net_cleanup_work, cleanup_net); 530 531 void __put_net(struct net *net) 532 { 533 /* Cleanup the network namespace in process context */ 534 unsigned long flags; 535 536 spin_lock_irqsave(&cleanup_list_lock, flags); 537 list_add(&net->cleanup_list, &cleanup_list); 538 spin_unlock_irqrestore(&cleanup_list_lock, flags); 539 540 queue_work(netns_wq, &net_cleanup_work); 541 } 542 EXPORT_SYMBOL_GPL(__put_net); 543 544 struct net *get_net_ns_by_fd(int fd) 545 { 546 struct file *file; 547 struct ns_common *ns; 548 struct net *net; 549 550 file = proc_ns_fget(fd); 551 if (IS_ERR(file)) 552 return ERR_CAST(file); 553 554 ns = get_proc_ns(file_inode(file)); 555 if (ns->ops == &netns_operations) 556 net = get_net(container_of(ns, struct net, ns)); 557 else 558 net = ERR_PTR(-EINVAL); 559 560 fput(file); 561 return net; 562 } 563 564 #else 565 struct net *get_net_ns_by_fd(int fd) 566 { 567 return ERR_PTR(-EINVAL); 568 } 569 #endif 570 EXPORT_SYMBOL_GPL(get_net_ns_by_fd); 571 572 struct net *get_net_ns_by_pid(pid_t pid) 573 { 574 struct task_struct *tsk; 575 struct net *net; 576 577 /* Lookup the network namespace */ 578 net = ERR_PTR(-ESRCH); 579 rcu_read_lock(); 580 tsk = find_task_by_vpid(pid); 581 if (tsk) { 582 struct nsproxy *nsproxy; 583 task_lock(tsk); 584 nsproxy = tsk->nsproxy; 585 if (nsproxy) 586 net = get_net(nsproxy->net_ns); 587 task_unlock(tsk); 588 } 589 rcu_read_unlock(); 590 return net; 591 } 592 EXPORT_SYMBOL_GPL(get_net_ns_by_pid); 593 594 static __net_init int net_ns_net_init(struct net *net) 595 { 596 #ifdef CONFIG_NET_NS 597 net->ns.ops = &netns_operations; 598 #endif 599 return ns_alloc_inum(&net->ns); 600 } 601 602 static __net_exit void net_ns_net_exit(struct net *net) 603 { 604 ns_free_inum(&net->ns); 605 } 606 607 static struct pernet_operations __net_initdata net_ns_ops = { 608 .init = net_ns_net_init, 609 .exit = net_ns_net_exit, 610 }; 611 612 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = { 613 [NETNSA_NONE] = { .type = NLA_UNSPEC }, 614 [NETNSA_NSID] = { .type = NLA_S32 }, 615 [NETNSA_PID] = { .type = NLA_U32 }, 616 [NETNSA_FD] = { .type = NLA_U32 }, 617 }; 618 619 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh, 620 struct netlink_ext_ack *extack) 621 { 622 struct net *net = sock_net(skb->sk); 623 struct nlattr *tb[NETNSA_MAX + 1]; 624 struct nlattr *nla; 625 struct net *peer; 626 int nsid, err; 627 628 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX, 629 rtnl_net_policy, extack); 630 if (err < 0) 631 return err; 632 if (!tb[NETNSA_NSID]) { 633 NL_SET_ERR_MSG(extack, "nsid is missing"); 634 return -EINVAL; 635 } 636 nsid = nla_get_s32(tb[NETNSA_NSID]); 637 638 if (tb[NETNSA_PID]) { 639 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 640 nla = tb[NETNSA_PID]; 641 } else if (tb[NETNSA_FD]) { 642 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 643 nla = tb[NETNSA_FD]; 644 } else { 645 NL_SET_ERR_MSG(extack, "Peer netns reference is missing"); 646 return -EINVAL; 647 } 648 if (IS_ERR(peer)) { 649 NL_SET_BAD_ATTR(extack, nla); 650 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid"); 651 return PTR_ERR(peer); 652 } 653 654 spin_lock_bh(&net->nsid_lock); 655 if (__peernet2id(net, peer) >= 0) { 656 spin_unlock_bh(&net->nsid_lock); 657 err = -EEXIST; 658 NL_SET_BAD_ATTR(extack, nla); 659 NL_SET_ERR_MSG(extack, 660 "Peer netns already has a nsid assigned"); 661 goto out; 662 } 663 664 err = alloc_netid(net, peer, nsid); 665 spin_unlock_bh(&net->nsid_lock); 666 if (err >= 0) { 667 rtnl_net_notifyid(net, RTM_NEWNSID, err); 668 err = 0; 669 } else if (err == -ENOSPC && nsid >= 0) { 670 err = -EEXIST; 671 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]); 672 NL_SET_ERR_MSG(extack, "The specified nsid is already used"); 673 } 674 out: 675 put_net(peer); 676 return err; 677 } 678 679 static int rtnl_net_get_size(void) 680 { 681 return NLMSG_ALIGN(sizeof(struct rtgenmsg)) 682 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */ 683 ; 684 } 685 686 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags, 687 int cmd, struct net *net, int nsid) 688 { 689 struct nlmsghdr *nlh; 690 struct rtgenmsg *rth; 691 692 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags); 693 if (!nlh) 694 return -EMSGSIZE; 695 696 rth = nlmsg_data(nlh); 697 rth->rtgen_family = AF_UNSPEC; 698 699 if (nla_put_s32(skb, NETNSA_NSID, nsid)) 700 goto nla_put_failure; 701 702 nlmsg_end(skb, nlh); 703 return 0; 704 705 nla_put_failure: 706 nlmsg_cancel(skb, nlh); 707 return -EMSGSIZE; 708 } 709 710 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh, 711 struct netlink_ext_ack *extack) 712 { 713 struct net *net = sock_net(skb->sk); 714 struct nlattr *tb[NETNSA_MAX + 1]; 715 struct nlattr *nla; 716 struct sk_buff *msg; 717 struct net *peer; 718 int err, id; 719 720 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX, 721 rtnl_net_policy, extack); 722 if (err < 0) 723 return err; 724 if (tb[NETNSA_PID]) { 725 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 726 nla = tb[NETNSA_PID]; 727 } else if (tb[NETNSA_FD]) { 728 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 729 nla = tb[NETNSA_FD]; 730 } else { 731 NL_SET_ERR_MSG(extack, "Peer netns reference is missing"); 732 return -EINVAL; 733 } 734 735 if (IS_ERR(peer)) { 736 NL_SET_BAD_ATTR(extack, nla); 737 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid"); 738 return PTR_ERR(peer); 739 } 740 741 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 742 if (!msg) { 743 err = -ENOMEM; 744 goto out; 745 } 746 747 id = peernet2id(net, peer); 748 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0, 749 RTM_NEWNSID, net, id); 750 if (err < 0) 751 goto err_out; 752 753 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid); 754 goto out; 755 756 err_out: 757 nlmsg_free(msg); 758 out: 759 put_net(peer); 760 return err; 761 } 762 763 struct rtnl_net_dump_cb { 764 struct net *net; 765 struct sk_buff *skb; 766 struct netlink_callback *cb; 767 int idx; 768 int s_idx; 769 }; 770 771 static int rtnl_net_dumpid_one(int id, void *peer, void *data) 772 { 773 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data; 774 int ret; 775 776 if (net_cb->idx < net_cb->s_idx) 777 goto cont; 778 779 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid, 780 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI, 781 RTM_NEWNSID, net_cb->net, id); 782 if (ret < 0) 783 return ret; 784 785 cont: 786 net_cb->idx++; 787 return 0; 788 } 789 790 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb) 791 { 792 struct net *net = sock_net(skb->sk); 793 struct rtnl_net_dump_cb net_cb = { 794 .net = net, 795 .skb = skb, 796 .cb = cb, 797 .idx = 0, 798 .s_idx = cb->args[0], 799 }; 800 801 spin_lock_bh(&net->nsid_lock); 802 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb); 803 spin_unlock_bh(&net->nsid_lock); 804 805 cb->args[0] = net_cb.idx; 806 return skb->len; 807 } 808 809 static void rtnl_net_notifyid(struct net *net, int cmd, int id) 810 { 811 struct sk_buff *msg; 812 int err = -ENOMEM; 813 814 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 815 if (!msg) 816 goto out; 817 818 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id); 819 if (err < 0) 820 goto err_out; 821 822 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0); 823 return; 824 825 err_out: 826 nlmsg_free(msg); 827 out: 828 rtnl_set_sk_err(net, RTNLGRP_NSID, err); 829 } 830 831 static int __init net_ns_init(void) 832 { 833 struct net_generic *ng; 834 835 #ifdef CONFIG_NET_NS 836 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), 837 SMP_CACHE_BYTES, 838 SLAB_PANIC, NULL); 839 840 /* Create workqueue for cleanup */ 841 netns_wq = create_singlethread_workqueue("netns"); 842 if (!netns_wq) 843 panic("Could not create netns workq"); 844 #endif 845 846 ng = net_alloc_generic(); 847 if (!ng) 848 panic("Could not allocate generic netns"); 849 850 rcu_assign_pointer(init_net.gen, ng); 851 852 mutex_lock(&net_mutex); 853 if (setup_net(&init_net, &init_user_ns)) 854 panic("Could not setup the initial network namespace"); 855 856 init_net_initialized = true; 857 858 rtnl_lock(); 859 list_add_tail_rcu(&init_net.list, &net_namespace_list); 860 rtnl_unlock(); 861 862 mutex_unlock(&net_mutex); 863 864 register_pernet_subsys(&net_ns_ops); 865 866 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, 867 RTNL_FLAG_DOIT_UNLOCKED); 868 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid, 869 RTNL_FLAG_DOIT_UNLOCKED); 870 871 return 0; 872 } 873 874 pure_initcall(net_ns_init); 875 876 #ifdef CONFIG_NET_NS 877 static int __register_pernet_operations(struct list_head *list, 878 struct pernet_operations *ops) 879 { 880 struct net *net; 881 int error; 882 LIST_HEAD(net_exit_list); 883 884 list_add_tail(&ops->list, list); 885 if (ops->init || (ops->id && ops->size)) { 886 for_each_net(net) { 887 error = ops_init(ops, net); 888 if (error) 889 goto out_undo; 890 list_add_tail(&net->exit_list, &net_exit_list); 891 } 892 } 893 return 0; 894 895 out_undo: 896 /* If I have an error cleanup all namespaces I initialized */ 897 list_del(&ops->list); 898 ops_exit_list(ops, &net_exit_list); 899 ops_free_list(ops, &net_exit_list); 900 return error; 901 } 902 903 static void __unregister_pernet_operations(struct pernet_operations *ops) 904 { 905 struct net *net; 906 LIST_HEAD(net_exit_list); 907 908 list_del(&ops->list); 909 for_each_net(net) 910 list_add_tail(&net->exit_list, &net_exit_list); 911 ops_exit_list(ops, &net_exit_list); 912 ops_free_list(ops, &net_exit_list); 913 } 914 915 #else 916 917 static int __register_pernet_operations(struct list_head *list, 918 struct pernet_operations *ops) 919 { 920 if (!init_net_initialized) { 921 list_add_tail(&ops->list, list); 922 return 0; 923 } 924 925 return ops_init(ops, &init_net); 926 } 927 928 static void __unregister_pernet_operations(struct pernet_operations *ops) 929 { 930 if (!init_net_initialized) { 931 list_del(&ops->list); 932 } else { 933 LIST_HEAD(net_exit_list); 934 list_add(&init_net.exit_list, &net_exit_list); 935 ops_exit_list(ops, &net_exit_list); 936 ops_free_list(ops, &net_exit_list); 937 } 938 } 939 940 #endif /* CONFIG_NET_NS */ 941 942 static DEFINE_IDA(net_generic_ids); 943 944 static int register_pernet_operations(struct list_head *list, 945 struct pernet_operations *ops) 946 { 947 int error; 948 949 if (ops->id) { 950 again: 951 error = ida_get_new_above(&net_generic_ids, MIN_PERNET_OPS_ID, ops->id); 952 if (error < 0) { 953 if (error == -EAGAIN) { 954 ida_pre_get(&net_generic_ids, GFP_KERNEL); 955 goto again; 956 } 957 return error; 958 } 959 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1); 960 } 961 error = __register_pernet_operations(list, ops); 962 if (error) { 963 rcu_barrier(); 964 if (ops->id) 965 ida_remove(&net_generic_ids, *ops->id); 966 } 967 968 return error; 969 } 970 971 static void unregister_pernet_operations(struct pernet_operations *ops) 972 { 973 974 __unregister_pernet_operations(ops); 975 rcu_barrier(); 976 if (ops->id) 977 ida_remove(&net_generic_ids, *ops->id); 978 } 979 980 /** 981 * register_pernet_subsys - register a network namespace subsystem 982 * @ops: pernet operations structure for the subsystem 983 * 984 * Register a subsystem which has init and exit functions 985 * that are called when network namespaces are created and 986 * destroyed respectively. 987 * 988 * When registered all network namespace init functions are 989 * called for every existing network namespace. Allowing kernel 990 * modules to have a race free view of the set of network namespaces. 991 * 992 * When a new network namespace is created all of the init 993 * methods are called in the order in which they were registered. 994 * 995 * When a network namespace is destroyed all of the exit methods 996 * are called in the reverse of the order with which they were 997 * registered. 998 */ 999 int register_pernet_subsys(struct pernet_operations *ops) 1000 { 1001 int error; 1002 mutex_lock(&net_mutex); 1003 error = register_pernet_operations(first_device, ops); 1004 mutex_unlock(&net_mutex); 1005 return error; 1006 } 1007 EXPORT_SYMBOL_GPL(register_pernet_subsys); 1008 1009 /** 1010 * unregister_pernet_subsys - unregister a network namespace subsystem 1011 * @ops: pernet operations structure to manipulate 1012 * 1013 * Remove the pernet operations structure from the list to be 1014 * used when network namespaces are created or destroyed. In 1015 * addition run the exit method for all existing network 1016 * namespaces. 1017 */ 1018 void unregister_pernet_subsys(struct pernet_operations *ops) 1019 { 1020 mutex_lock(&net_mutex); 1021 unregister_pernet_operations(ops); 1022 mutex_unlock(&net_mutex); 1023 } 1024 EXPORT_SYMBOL_GPL(unregister_pernet_subsys); 1025 1026 /** 1027 * register_pernet_device - register a network namespace device 1028 * @ops: pernet operations structure for the subsystem 1029 * 1030 * Register a device which has init and exit functions 1031 * that are called when network namespaces are created and 1032 * destroyed respectively. 1033 * 1034 * When registered all network namespace init functions are 1035 * called for every existing network namespace. Allowing kernel 1036 * modules to have a race free view of the set of network namespaces. 1037 * 1038 * When a new network namespace is created all of the init 1039 * methods are called in the order in which they were registered. 1040 * 1041 * When a network namespace is destroyed all of the exit methods 1042 * are called in the reverse of the order with which they were 1043 * registered. 1044 */ 1045 int register_pernet_device(struct pernet_operations *ops) 1046 { 1047 int error; 1048 mutex_lock(&net_mutex); 1049 error = register_pernet_operations(&pernet_list, ops); 1050 if (!error && (first_device == &pernet_list)) 1051 first_device = &ops->list; 1052 mutex_unlock(&net_mutex); 1053 return error; 1054 } 1055 EXPORT_SYMBOL_GPL(register_pernet_device); 1056 1057 /** 1058 * unregister_pernet_device - unregister a network namespace netdevice 1059 * @ops: pernet operations structure to manipulate 1060 * 1061 * Remove the pernet operations structure from the list to be 1062 * used when network namespaces are created or destroyed. In 1063 * addition run the exit method for all existing network 1064 * namespaces. 1065 */ 1066 void unregister_pernet_device(struct pernet_operations *ops) 1067 { 1068 mutex_lock(&net_mutex); 1069 if (&ops->list == first_device) 1070 first_device = first_device->next; 1071 unregister_pernet_operations(ops); 1072 mutex_unlock(&net_mutex); 1073 } 1074 EXPORT_SYMBOL_GPL(unregister_pernet_device); 1075 1076 #ifdef CONFIG_NET_NS 1077 static struct ns_common *netns_get(struct task_struct *task) 1078 { 1079 struct net *net = NULL; 1080 struct nsproxy *nsproxy; 1081 1082 task_lock(task); 1083 nsproxy = task->nsproxy; 1084 if (nsproxy) 1085 net = get_net(nsproxy->net_ns); 1086 task_unlock(task); 1087 1088 return net ? &net->ns : NULL; 1089 } 1090 1091 static inline struct net *to_net_ns(struct ns_common *ns) 1092 { 1093 return container_of(ns, struct net, ns); 1094 } 1095 1096 static void netns_put(struct ns_common *ns) 1097 { 1098 put_net(to_net_ns(ns)); 1099 } 1100 1101 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns) 1102 { 1103 struct net *net = to_net_ns(ns); 1104 1105 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) || 1106 !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 1107 return -EPERM; 1108 1109 put_net(nsproxy->net_ns); 1110 nsproxy->net_ns = get_net(net); 1111 return 0; 1112 } 1113 1114 static struct user_namespace *netns_owner(struct ns_common *ns) 1115 { 1116 return to_net_ns(ns)->user_ns; 1117 } 1118 1119 const struct proc_ns_operations netns_operations = { 1120 .name = "net", 1121 .type = CLONE_NEWNET, 1122 .get = netns_get, 1123 .put = netns_put, 1124 .install = netns_install, 1125 .owner = netns_owner, 1126 }; 1127 #endif 1128