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 <net/sock.h> 20 #include <net/netlink.h> 21 #include <net/net_namespace.h> 22 #include <net/netns/generic.h> 23 24 /* 25 * Our network namespace constructor/destructor lists 26 */ 27 28 static LIST_HEAD(pernet_list); 29 static struct list_head *first_device = &pernet_list; 30 DEFINE_MUTEX(net_mutex); 31 32 LIST_HEAD(net_namespace_list); 33 EXPORT_SYMBOL_GPL(net_namespace_list); 34 35 struct net init_net = { 36 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head), 37 }; 38 EXPORT_SYMBOL(init_net); 39 40 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ 41 42 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS; 43 44 static struct net_generic *net_alloc_generic(void) 45 { 46 struct net_generic *ng; 47 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]); 48 49 ng = kzalloc(generic_size, GFP_KERNEL); 50 if (ng) 51 ng->len = max_gen_ptrs; 52 53 return ng; 54 } 55 56 static int net_assign_generic(struct net *net, int id, void *data) 57 { 58 struct net_generic *ng, *old_ng; 59 60 BUG_ON(!mutex_is_locked(&net_mutex)); 61 BUG_ON(id == 0); 62 63 old_ng = rcu_dereference_protected(net->gen, 64 lockdep_is_held(&net_mutex)); 65 ng = old_ng; 66 if (old_ng->len >= id) 67 goto assign; 68 69 ng = net_alloc_generic(); 70 if (ng == NULL) 71 return -ENOMEM; 72 73 /* 74 * Some synchronisation notes: 75 * 76 * The net_generic explores the net->gen array inside rcu 77 * read section. Besides once set the net->gen->ptr[x] 78 * pointer never changes (see rules in netns/generic.h). 79 * 80 * That said, we simply duplicate this array and schedule 81 * the old copy for kfree after a grace period. 82 */ 83 84 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*)); 85 86 rcu_assign_pointer(net->gen, ng); 87 kfree_rcu(old_ng, rcu); 88 assign: 89 ng->ptr[id - 1] = data; 90 return 0; 91 } 92 93 static int ops_init(const struct pernet_operations *ops, struct net *net) 94 { 95 int err = -ENOMEM; 96 void *data = NULL; 97 98 if (ops->id && ops->size) { 99 data = kzalloc(ops->size, GFP_KERNEL); 100 if (!data) 101 goto out; 102 103 err = net_assign_generic(net, *ops->id, data); 104 if (err) 105 goto cleanup; 106 } 107 err = 0; 108 if (ops->init) 109 err = ops->init(net); 110 if (!err) 111 return 0; 112 113 cleanup: 114 kfree(data); 115 116 out: 117 return err; 118 } 119 120 static void ops_free(const struct pernet_operations *ops, struct net *net) 121 { 122 if (ops->id && ops->size) { 123 int id = *ops->id; 124 kfree(net_generic(net, id)); 125 } 126 } 127 128 static void ops_exit_list(const struct pernet_operations *ops, 129 struct list_head *net_exit_list) 130 { 131 struct net *net; 132 if (ops->exit) { 133 list_for_each_entry(net, net_exit_list, exit_list) 134 ops->exit(net); 135 } 136 if (ops->exit_batch) 137 ops->exit_batch(net_exit_list); 138 } 139 140 static void ops_free_list(const struct pernet_operations *ops, 141 struct list_head *net_exit_list) 142 { 143 struct net *net; 144 if (ops->size && ops->id) { 145 list_for_each_entry(net, net_exit_list, exit_list) 146 ops_free(ops, net); 147 } 148 } 149 150 /* should be called with nsid_lock held */ 151 static int alloc_netid(struct net *net, struct net *peer, int reqid) 152 { 153 int min = 0, max = 0; 154 155 if (reqid >= 0) { 156 min = reqid; 157 max = reqid + 1; 158 } 159 160 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC); 161 } 162 163 /* This function is used by idr_for_each(). If net is equal to peer, the 164 * function returns the id so that idr_for_each() stops. Because we cannot 165 * returns the id 0 (idr_for_each() will not stop), we return the magic value 166 * NET_ID_ZERO (-1) for it. 167 */ 168 #define NET_ID_ZERO -1 169 static int net_eq_idr(int id, void *net, void *peer) 170 { 171 if (net_eq(net, peer)) 172 return id ? : NET_ID_ZERO; 173 return 0; 174 } 175 176 /* Should be called with nsid_lock held. If a new id is assigned, the bool alloc 177 * is set to true, thus the caller knows that the new id must be notified via 178 * rtnl. 179 */ 180 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc) 181 { 182 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer); 183 bool alloc_it = *alloc; 184 185 *alloc = false; 186 187 /* Magic value for id 0. */ 188 if (id == NET_ID_ZERO) 189 return 0; 190 if (id > 0) 191 return id; 192 193 if (alloc_it) { 194 id = alloc_netid(net, peer, -1); 195 *alloc = true; 196 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED; 197 } 198 199 return NETNSA_NSID_NOT_ASSIGNED; 200 } 201 202 /* should be called with nsid_lock held */ 203 static int __peernet2id(struct net *net, struct net *peer) 204 { 205 bool no = false; 206 207 return __peernet2id_alloc(net, peer, &no); 208 } 209 210 static void rtnl_net_notifyid(struct net *net, int cmd, int id); 211 /* This function returns the id of a peer netns. If no id is assigned, one will 212 * be allocated and returned. 213 */ 214 int peernet2id_alloc(struct net *net, struct net *peer) 215 { 216 bool alloc; 217 int id; 218 219 spin_lock_bh(&net->nsid_lock); 220 alloc = atomic_read(&peer->count) == 0 ? false : true; 221 id = __peernet2id_alloc(net, peer, &alloc); 222 spin_unlock_bh(&net->nsid_lock); 223 if (alloc && id >= 0) 224 rtnl_net_notifyid(net, RTM_NEWNSID, id); 225 return id; 226 } 227 EXPORT_SYMBOL(peernet2id_alloc); 228 229 /* This function returns, if assigned, the id of a peer netns. */ 230 int peernet2id(struct net *net, struct net *peer) 231 { 232 int id; 233 234 spin_lock_bh(&net->nsid_lock); 235 id = __peernet2id(net, peer); 236 spin_unlock_bh(&net->nsid_lock); 237 return id; 238 } 239 240 /* This function returns true is the peer netns has an id assigned into the 241 * current netns. 242 */ 243 bool peernet_has_id(struct net *net, struct net *peer) 244 { 245 return peernet2id(net, peer) >= 0; 246 } 247 248 struct net *get_net_ns_by_id(struct net *net, int id) 249 { 250 struct net *peer; 251 252 if (id < 0) 253 return NULL; 254 255 rcu_read_lock(); 256 spin_lock_bh(&net->nsid_lock); 257 peer = idr_find(&net->netns_ids, id); 258 if (peer) 259 get_net(peer); 260 spin_unlock_bh(&net->nsid_lock); 261 rcu_read_unlock(); 262 263 return peer; 264 } 265 266 /* 267 * setup_net runs the initializers for the network namespace object. 268 */ 269 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns) 270 { 271 /* Must be called with net_mutex held */ 272 const struct pernet_operations *ops, *saved_ops; 273 int error = 0; 274 LIST_HEAD(net_exit_list); 275 276 atomic_set(&net->count, 1); 277 atomic_set(&net->passive, 1); 278 net->dev_base_seq = 1; 279 net->user_ns = user_ns; 280 idr_init(&net->netns_ids); 281 spin_lock_init(&net->nsid_lock); 282 283 list_for_each_entry(ops, &pernet_list, list) { 284 error = ops_init(ops, net); 285 if (error < 0) 286 goto out_undo; 287 } 288 out: 289 return error; 290 291 out_undo: 292 /* Walk through the list backwards calling the exit functions 293 * for the pernet modules whose init functions did not fail. 294 */ 295 list_add(&net->exit_list, &net_exit_list); 296 saved_ops = ops; 297 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 298 ops_exit_list(ops, &net_exit_list); 299 300 ops = saved_ops; 301 list_for_each_entry_continue_reverse(ops, &pernet_list, list) 302 ops_free_list(ops, &net_exit_list); 303 304 rcu_barrier(); 305 goto out; 306 } 307 308 309 #ifdef CONFIG_NET_NS 310 static struct kmem_cache *net_cachep; 311 static struct workqueue_struct *netns_wq; 312 313 static struct net *net_alloc(void) 314 { 315 struct net *net = NULL; 316 struct net_generic *ng; 317 318 ng = net_alloc_generic(); 319 if (!ng) 320 goto out; 321 322 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); 323 if (!net) 324 goto out_free; 325 326 rcu_assign_pointer(net->gen, ng); 327 out: 328 return net; 329 330 out_free: 331 kfree(ng); 332 goto out; 333 } 334 335 static void net_free(struct net *net) 336 { 337 kfree(rcu_access_pointer(net->gen)); 338 kmem_cache_free(net_cachep, net); 339 } 340 341 void net_drop_ns(void *p) 342 { 343 struct net *ns = p; 344 if (ns && atomic_dec_and_test(&ns->passive)) 345 net_free(ns); 346 } 347 348 struct net *copy_net_ns(unsigned long flags, 349 struct user_namespace *user_ns, struct net *old_net) 350 { 351 struct net *net; 352 int rv; 353 354 if (!(flags & CLONE_NEWNET)) 355 return get_net(old_net); 356 357 net = net_alloc(); 358 if (!net) 359 return ERR_PTR(-ENOMEM); 360 361 get_user_ns(user_ns); 362 363 mutex_lock(&net_mutex); 364 rv = setup_net(net, user_ns); 365 if (rv == 0) { 366 rtnl_lock(); 367 list_add_tail_rcu(&net->list, &net_namespace_list); 368 rtnl_unlock(); 369 } 370 mutex_unlock(&net_mutex); 371 if (rv < 0) { 372 put_user_ns(user_ns); 373 net_drop_ns(net); 374 return ERR_PTR(rv); 375 } 376 return net; 377 } 378 379 static DEFINE_SPINLOCK(cleanup_list_lock); 380 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */ 381 382 static void cleanup_net(struct work_struct *work) 383 { 384 const struct pernet_operations *ops; 385 struct net *net, *tmp; 386 struct list_head net_kill_list; 387 LIST_HEAD(net_exit_list); 388 389 /* Atomically snapshot the list of namespaces to cleanup */ 390 spin_lock_irq(&cleanup_list_lock); 391 list_replace_init(&cleanup_list, &net_kill_list); 392 spin_unlock_irq(&cleanup_list_lock); 393 394 mutex_lock(&net_mutex); 395 396 /* Don't let anyone else find us. */ 397 rtnl_lock(); 398 list_for_each_entry(net, &net_kill_list, cleanup_list) { 399 list_del_rcu(&net->list); 400 list_add_tail(&net->exit_list, &net_exit_list); 401 for_each_net(tmp) { 402 int id; 403 404 spin_lock_bh(&tmp->nsid_lock); 405 id = __peernet2id(tmp, net); 406 if (id >= 0) 407 idr_remove(&tmp->netns_ids, id); 408 spin_unlock_bh(&tmp->nsid_lock); 409 if (id >= 0) 410 rtnl_net_notifyid(tmp, RTM_DELNSID, id); 411 } 412 spin_lock_bh(&net->nsid_lock); 413 idr_destroy(&net->netns_ids); 414 spin_unlock_bh(&net->nsid_lock); 415 416 } 417 rtnl_unlock(); 418 419 /* 420 * Another CPU might be rcu-iterating the list, wait for it. 421 * This needs to be before calling the exit() notifiers, so 422 * the rcu_barrier() below isn't sufficient alone. 423 */ 424 synchronize_rcu(); 425 426 /* Run all of the network namespace exit methods */ 427 list_for_each_entry_reverse(ops, &pernet_list, list) 428 ops_exit_list(ops, &net_exit_list); 429 430 /* Free the net generic variables */ 431 list_for_each_entry_reverse(ops, &pernet_list, list) 432 ops_free_list(ops, &net_exit_list); 433 434 mutex_unlock(&net_mutex); 435 436 /* Ensure there are no outstanding rcu callbacks using this 437 * network namespace. 438 */ 439 rcu_barrier(); 440 441 /* Finally it is safe to free my network namespace structure */ 442 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) { 443 list_del_init(&net->exit_list); 444 put_user_ns(net->user_ns); 445 net_drop_ns(net); 446 } 447 } 448 static DECLARE_WORK(net_cleanup_work, cleanup_net); 449 450 void __put_net(struct net *net) 451 { 452 /* Cleanup the network namespace in process context */ 453 unsigned long flags; 454 455 spin_lock_irqsave(&cleanup_list_lock, flags); 456 list_add(&net->cleanup_list, &cleanup_list); 457 spin_unlock_irqrestore(&cleanup_list_lock, flags); 458 459 queue_work(netns_wq, &net_cleanup_work); 460 } 461 EXPORT_SYMBOL_GPL(__put_net); 462 463 struct net *get_net_ns_by_fd(int fd) 464 { 465 struct file *file; 466 struct ns_common *ns; 467 struct net *net; 468 469 file = proc_ns_fget(fd); 470 if (IS_ERR(file)) 471 return ERR_CAST(file); 472 473 ns = get_proc_ns(file_inode(file)); 474 if (ns->ops == &netns_operations) 475 net = get_net(container_of(ns, struct net, ns)); 476 else 477 net = ERR_PTR(-EINVAL); 478 479 fput(file); 480 return net; 481 } 482 483 #else 484 struct net *get_net_ns_by_fd(int fd) 485 { 486 return ERR_PTR(-EINVAL); 487 } 488 #endif 489 EXPORT_SYMBOL_GPL(get_net_ns_by_fd); 490 491 struct net *get_net_ns_by_pid(pid_t pid) 492 { 493 struct task_struct *tsk; 494 struct net *net; 495 496 /* Lookup the network namespace */ 497 net = ERR_PTR(-ESRCH); 498 rcu_read_lock(); 499 tsk = find_task_by_vpid(pid); 500 if (tsk) { 501 struct nsproxy *nsproxy; 502 task_lock(tsk); 503 nsproxy = tsk->nsproxy; 504 if (nsproxy) 505 net = get_net(nsproxy->net_ns); 506 task_unlock(tsk); 507 } 508 rcu_read_unlock(); 509 return net; 510 } 511 EXPORT_SYMBOL_GPL(get_net_ns_by_pid); 512 513 static __net_init int net_ns_net_init(struct net *net) 514 { 515 #ifdef CONFIG_NET_NS 516 net->ns.ops = &netns_operations; 517 #endif 518 return ns_alloc_inum(&net->ns); 519 } 520 521 static __net_exit void net_ns_net_exit(struct net *net) 522 { 523 ns_free_inum(&net->ns); 524 } 525 526 static struct pernet_operations __net_initdata net_ns_ops = { 527 .init = net_ns_net_init, 528 .exit = net_ns_net_exit, 529 }; 530 531 static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = { 532 [NETNSA_NONE] = { .type = NLA_UNSPEC }, 533 [NETNSA_NSID] = { .type = NLA_S32 }, 534 [NETNSA_PID] = { .type = NLA_U32 }, 535 [NETNSA_FD] = { .type = NLA_U32 }, 536 }; 537 538 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh) 539 { 540 struct net *net = sock_net(skb->sk); 541 struct nlattr *tb[NETNSA_MAX + 1]; 542 struct net *peer; 543 int nsid, err; 544 545 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX, 546 rtnl_net_policy); 547 if (err < 0) 548 return err; 549 if (!tb[NETNSA_NSID]) 550 return -EINVAL; 551 nsid = nla_get_s32(tb[NETNSA_NSID]); 552 553 if (tb[NETNSA_PID]) 554 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 555 else if (tb[NETNSA_FD]) 556 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 557 else 558 return -EINVAL; 559 if (IS_ERR(peer)) 560 return PTR_ERR(peer); 561 562 spin_lock_bh(&net->nsid_lock); 563 if (__peernet2id(net, peer) >= 0) { 564 spin_unlock_bh(&net->nsid_lock); 565 err = -EEXIST; 566 goto out; 567 } 568 569 err = alloc_netid(net, peer, nsid); 570 spin_unlock_bh(&net->nsid_lock); 571 if (err >= 0) { 572 rtnl_net_notifyid(net, RTM_NEWNSID, err); 573 err = 0; 574 } 575 out: 576 put_net(peer); 577 return err; 578 } 579 580 static int rtnl_net_get_size(void) 581 { 582 return NLMSG_ALIGN(sizeof(struct rtgenmsg)) 583 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */ 584 ; 585 } 586 587 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags, 588 int cmd, struct net *net, int nsid) 589 { 590 struct nlmsghdr *nlh; 591 struct rtgenmsg *rth; 592 593 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags); 594 if (!nlh) 595 return -EMSGSIZE; 596 597 rth = nlmsg_data(nlh); 598 rth->rtgen_family = AF_UNSPEC; 599 600 if (nla_put_s32(skb, NETNSA_NSID, nsid)) 601 goto nla_put_failure; 602 603 nlmsg_end(skb, nlh); 604 return 0; 605 606 nla_put_failure: 607 nlmsg_cancel(skb, nlh); 608 return -EMSGSIZE; 609 } 610 611 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh) 612 { 613 struct net *net = sock_net(skb->sk); 614 struct nlattr *tb[NETNSA_MAX + 1]; 615 struct sk_buff *msg; 616 struct net *peer; 617 int err, id; 618 619 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX, 620 rtnl_net_policy); 621 if (err < 0) 622 return err; 623 if (tb[NETNSA_PID]) 624 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID])); 625 else if (tb[NETNSA_FD]) 626 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD])); 627 else 628 return -EINVAL; 629 630 if (IS_ERR(peer)) 631 return PTR_ERR(peer); 632 633 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 634 if (!msg) { 635 err = -ENOMEM; 636 goto out; 637 } 638 639 id = peernet2id(net, peer); 640 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0, 641 RTM_NEWNSID, net, id); 642 if (err < 0) 643 goto err_out; 644 645 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid); 646 goto out; 647 648 err_out: 649 nlmsg_free(msg); 650 out: 651 put_net(peer); 652 return err; 653 } 654 655 struct rtnl_net_dump_cb { 656 struct net *net; 657 struct sk_buff *skb; 658 struct netlink_callback *cb; 659 int idx; 660 int s_idx; 661 }; 662 663 static int rtnl_net_dumpid_one(int id, void *peer, void *data) 664 { 665 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data; 666 int ret; 667 668 if (net_cb->idx < net_cb->s_idx) 669 goto cont; 670 671 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid, 672 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI, 673 RTM_NEWNSID, net_cb->net, id); 674 if (ret < 0) 675 return ret; 676 677 cont: 678 net_cb->idx++; 679 return 0; 680 } 681 682 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb) 683 { 684 struct net *net = sock_net(skb->sk); 685 struct rtnl_net_dump_cb net_cb = { 686 .net = net, 687 .skb = skb, 688 .cb = cb, 689 .idx = 0, 690 .s_idx = cb->args[0], 691 }; 692 693 spin_lock_bh(&net->nsid_lock); 694 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb); 695 spin_unlock_bh(&net->nsid_lock); 696 697 cb->args[0] = net_cb.idx; 698 return skb->len; 699 } 700 701 static void rtnl_net_notifyid(struct net *net, int cmd, int id) 702 { 703 struct sk_buff *msg; 704 int err = -ENOMEM; 705 706 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL); 707 if (!msg) 708 goto out; 709 710 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id); 711 if (err < 0) 712 goto err_out; 713 714 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0); 715 return; 716 717 err_out: 718 nlmsg_free(msg); 719 out: 720 rtnl_set_sk_err(net, RTNLGRP_NSID, err); 721 } 722 723 static int __init net_ns_init(void) 724 { 725 struct net_generic *ng; 726 727 #ifdef CONFIG_NET_NS 728 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), 729 SMP_CACHE_BYTES, 730 SLAB_PANIC, NULL); 731 732 /* Create workqueue for cleanup */ 733 netns_wq = create_singlethread_workqueue("netns"); 734 if (!netns_wq) 735 panic("Could not create netns workq"); 736 #endif 737 738 ng = net_alloc_generic(); 739 if (!ng) 740 panic("Could not allocate generic netns"); 741 742 rcu_assign_pointer(init_net.gen, ng); 743 744 mutex_lock(&net_mutex); 745 if (setup_net(&init_net, &init_user_ns)) 746 panic("Could not setup the initial network namespace"); 747 748 rtnl_lock(); 749 list_add_tail_rcu(&init_net.list, &net_namespace_list); 750 rtnl_unlock(); 751 752 mutex_unlock(&net_mutex); 753 754 register_pernet_subsys(&net_ns_ops); 755 756 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL); 757 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid, 758 NULL); 759 760 return 0; 761 } 762 763 pure_initcall(net_ns_init); 764 765 #ifdef CONFIG_NET_NS 766 static int __register_pernet_operations(struct list_head *list, 767 struct pernet_operations *ops) 768 { 769 struct net *net; 770 int error; 771 LIST_HEAD(net_exit_list); 772 773 list_add_tail(&ops->list, list); 774 if (ops->init || (ops->id && ops->size)) { 775 for_each_net(net) { 776 error = ops_init(ops, net); 777 if (error) 778 goto out_undo; 779 list_add_tail(&net->exit_list, &net_exit_list); 780 } 781 } 782 return 0; 783 784 out_undo: 785 /* If I have an error cleanup all namespaces I initialized */ 786 list_del(&ops->list); 787 ops_exit_list(ops, &net_exit_list); 788 ops_free_list(ops, &net_exit_list); 789 return error; 790 } 791 792 static void __unregister_pernet_operations(struct pernet_operations *ops) 793 { 794 struct net *net; 795 LIST_HEAD(net_exit_list); 796 797 list_del(&ops->list); 798 for_each_net(net) 799 list_add_tail(&net->exit_list, &net_exit_list); 800 ops_exit_list(ops, &net_exit_list); 801 ops_free_list(ops, &net_exit_list); 802 } 803 804 #else 805 806 static int __register_pernet_operations(struct list_head *list, 807 struct pernet_operations *ops) 808 { 809 return ops_init(ops, &init_net); 810 } 811 812 static void __unregister_pernet_operations(struct pernet_operations *ops) 813 { 814 LIST_HEAD(net_exit_list); 815 list_add(&init_net.exit_list, &net_exit_list); 816 ops_exit_list(ops, &net_exit_list); 817 ops_free_list(ops, &net_exit_list); 818 } 819 820 #endif /* CONFIG_NET_NS */ 821 822 static DEFINE_IDA(net_generic_ids); 823 824 static int register_pernet_operations(struct list_head *list, 825 struct pernet_operations *ops) 826 { 827 int error; 828 829 if (ops->id) { 830 again: 831 error = ida_get_new_above(&net_generic_ids, 1, ops->id); 832 if (error < 0) { 833 if (error == -EAGAIN) { 834 ida_pre_get(&net_generic_ids, GFP_KERNEL); 835 goto again; 836 } 837 return error; 838 } 839 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id); 840 } 841 error = __register_pernet_operations(list, ops); 842 if (error) { 843 rcu_barrier(); 844 if (ops->id) 845 ida_remove(&net_generic_ids, *ops->id); 846 } 847 848 return error; 849 } 850 851 static void unregister_pernet_operations(struct pernet_operations *ops) 852 { 853 854 __unregister_pernet_operations(ops); 855 rcu_barrier(); 856 if (ops->id) 857 ida_remove(&net_generic_ids, *ops->id); 858 } 859 860 /** 861 * register_pernet_subsys - register a network namespace subsystem 862 * @ops: pernet operations structure for the subsystem 863 * 864 * Register a subsystem which has init and exit functions 865 * that are called when network namespaces are created and 866 * destroyed respectively. 867 * 868 * When registered all network namespace init functions are 869 * called for every existing network namespace. Allowing kernel 870 * modules to have a race free view of the set of network namespaces. 871 * 872 * When a new network namespace is created all of the init 873 * methods are called in the order in which they were registered. 874 * 875 * When a network namespace is destroyed all of the exit methods 876 * are called in the reverse of the order with which they were 877 * registered. 878 */ 879 int register_pernet_subsys(struct pernet_operations *ops) 880 { 881 int error; 882 mutex_lock(&net_mutex); 883 error = register_pernet_operations(first_device, ops); 884 mutex_unlock(&net_mutex); 885 return error; 886 } 887 EXPORT_SYMBOL_GPL(register_pernet_subsys); 888 889 /** 890 * unregister_pernet_subsys - unregister a network namespace subsystem 891 * @ops: pernet operations structure to manipulate 892 * 893 * Remove the pernet operations structure from the list to be 894 * used when network namespaces are created or destroyed. In 895 * addition run the exit method for all existing network 896 * namespaces. 897 */ 898 void unregister_pernet_subsys(struct pernet_operations *ops) 899 { 900 mutex_lock(&net_mutex); 901 unregister_pernet_operations(ops); 902 mutex_unlock(&net_mutex); 903 } 904 EXPORT_SYMBOL_GPL(unregister_pernet_subsys); 905 906 /** 907 * register_pernet_device - register a network namespace device 908 * @ops: pernet operations structure for the subsystem 909 * 910 * Register a device which has init and exit functions 911 * that are called when network namespaces are created and 912 * destroyed respectively. 913 * 914 * When registered all network namespace init functions are 915 * called for every existing network namespace. Allowing kernel 916 * modules to have a race free view of the set of network namespaces. 917 * 918 * When a new network namespace is created all of the init 919 * methods are called in the order in which they were registered. 920 * 921 * When a network namespace is destroyed all of the exit methods 922 * are called in the reverse of the order with which they were 923 * registered. 924 */ 925 int register_pernet_device(struct pernet_operations *ops) 926 { 927 int error; 928 mutex_lock(&net_mutex); 929 error = register_pernet_operations(&pernet_list, ops); 930 if (!error && (first_device == &pernet_list)) 931 first_device = &ops->list; 932 mutex_unlock(&net_mutex); 933 return error; 934 } 935 EXPORT_SYMBOL_GPL(register_pernet_device); 936 937 /** 938 * unregister_pernet_device - unregister a network namespace netdevice 939 * @ops: pernet operations structure to manipulate 940 * 941 * Remove the pernet operations structure from the list to be 942 * used when network namespaces are created or destroyed. In 943 * addition run the exit method for all existing network 944 * namespaces. 945 */ 946 void unregister_pernet_device(struct pernet_operations *ops) 947 { 948 mutex_lock(&net_mutex); 949 if (&ops->list == first_device) 950 first_device = first_device->next; 951 unregister_pernet_operations(ops); 952 mutex_unlock(&net_mutex); 953 } 954 EXPORT_SYMBOL_GPL(unregister_pernet_device); 955 956 #ifdef CONFIG_NET_NS 957 static struct ns_common *netns_get(struct task_struct *task) 958 { 959 struct net *net = NULL; 960 struct nsproxy *nsproxy; 961 962 task_lock(task); 963 nsproxy = task->nsproxy; 964 if (nsproxy) 965 net = get_net(nsproxy->net_ns); 966 task_unlock(task); 967 968 return net ? &net->ns : NULL; 969 } 970 971 static inline struct net *to_net_ns(struct ns_common *ns) 972 { 973 return container_of(ns, struct net, ns); 974 } 975 976 static void netns_put(struct ns_common *ns) 977 { 978 put_net(to_net_ns(ns)); 979 } 980 981 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns) 982 { 983 struct net *net = to_net_ns(ns); 984 985 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) || 986 !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 987 return -EPERM; 988 989 put_net(nsproxy->net_ns); 990 nsproxy->net_ns = get_net(net); 991 return 0; 992 } 993 994 const struct proc_ns_operations netns_operations = { 995 .name = "net", 996 .type = CLONE_NEWNET, 997 .get = netns_get, 998 .put = netns_put, 999 .install = netns_install, 1000 }; 1001 #endif 1002