1 /* 2 * NETLINK Kernel-user communication protocol. 3 * 4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 6 * Patrick McHardy <kaber@trash.net> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 * 13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith 14 * added netlink_proto_exit 15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br> 16 * use nlk_sk, as sk->protinfo is on a diet 8) 17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org> 18 * - inc module use count of module that owns 19 * the kernel socket in case userspace opens 20 * socket of same protocol 21 * - remove all module support, since netlink is 22 * mandatory if CONFIG_NET=y these days 23 */ 24 25 #include <linux/module.h> 26 27 #include <linux/capability.h> 28 #include <linux/kernel.h> 29 #include <linux/init.h> 30 #include <linux/signal.h> 31 #include <linux/sched.h> 32 #include <linux/errno.h> 33 #include <linux/string.h> 34 #include <linux/stat.h> 35 #include <linux/socket.h> 36 #include <linux/un.h> 37 #include <linux/fcntl.h> 38 #include <linux/termios.h> 39 #include <linux/sockios.h> 40 #include <linux/net.h> 41 #include <linux/fs.h> 42 #include <linux/slab.h> 43 #include <linux/uaccess.h> 44 #include <linux/skbuff.h> 45 #include <linux/netdevice.h> 46 #include <linux/rtnetlink.h> 47 #include <linux/proc_fs.h> 48 #include <linux/seq_file.h> 49 #include <linux/notifier.h> 50 #include <linux/security.h> 51 #include <linux/jhash.h> 52 #include <linux/jiffies.h> 53 #include <linux/random.h> 54 #include <linux/bitops.h> 55 #include <linux/mm.h> 56 #include <linux/types.h> 57 #include <linux/audit.h> 58 #include <linux/mutex.h> 59 #include <linux/vmalloc.h> 60 #include <linux/if_arp.h> 61 #include <linux/rhashtable.h> 62 #include <asm/cacheflush.h> 63 #include <linux/hash.h> 64 #include <linux/genetlink.h> 65 #include <linux/net_namespace.h> 66 #include <linux/nospec.h> 67 68 #include <net/net_namespace.h> 69 #include <net/netns/generic.h> 70 #include <net/sock.h> 71 #include <net/scm.h> 72 #include <net/netlink.h> 73 74 #include "af_netlink.h" 75 76 struct listeners { 77 struct rcu_head rcu; 78 unsigned long masks[0]; 79 }; 80 81 /* state bits */ 82 #define NETLINK_S_CONGESTED 0x0 83 84 static inline int netlink_is_kernel(struct sock *sk) 85 { 86 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET; 87 } 88 89 struct netlink_table *nl_table __read_mostly; 90 EXPORT_SYMBOL_GPL(nl_table); 91 92 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait); 93 94 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS]; 95 96 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = { 97 "nlk_cb_mutex-ROUTE", 98 "nlk_cb_mutex-1", 99 "nlk_cb_mutex-USERSOCK", 100 "nlk_cb_mutex-FIREWALL", 101 "nlk_cb_mutex-SOCK_DIAG", 102 "nlk_cb_mutex-NFLOG", 103 "nlk_cb_mutex-XFRM", 104 "nlk_cb_mutex-SELINUX", 105 "nlk_cb_mutex-ISCSI", 106 "nlk_cb_mutex-AUDIT", 107 "nlk_cb_mutex-FIB_LOOKUP", 108 "nlk_cb_mutex-CONNECTOR", 109 "nlk_cb_mutex-NETFILTER", 110 "nlk_cb_mutex-IP6_FW", 111 "nlk_cb_mutex-DNRTMSG", 112 "nlk_cb_mutex-KOBJECT_UEVENT", 113 "nlk_cb_mutex-GENERIC", 114 "nlk_cb_mutex-17", 115 "nlk_cb_mutex-SCSITRANSPORT", 116 "nlk_cb_mutex-ECRYPTFS", 117 "nlk_cb_mutex-RDMA", 118 "nlk_cb_mutex-CRYPTO", 119 "nlk_cb_mutex-SMC", 120 "nlk_cb_mutex-23", 121 "nlk_cb_mutex-24", 122 "nlk_cb_mutex-25", 123 "nlk_cb_mutex-26", 124 "nlk_cb_mutex-27", 125 "nlk_cb_mutex-28", 126 "nlk_cb_mutex-29", 127 "nlk_cb_mutex-30", 128 "nlk_cb_mutex-31", 129 "nlk_cb_mutex-MAX_LINKS" 130 }; 131 132 static int netlink_dump(struct sock *sk); 133 134 /* nl_table locking explained: 135 * Lookup and traversal are protected with an RCU read-side lock. Insertion 136 * and removal are protected with per bucket lock while using RCU list 137 * modification primitives and may run in parallel to RCU protected lookups. 138 * Destruction of the Netlink socket may only occur *after* nl_table_lock has 139 * been acquired * either during or after the socket has been removed from 140 * the list and after an RCU grace period. 141 */ 142 DEFINE_RWLOCK(nl_table_lock); 143 EXPORT_SYMBOL_GPL(nl_table_lock); 144 static atomic_t nl_table_users = ATOMIC_INIT(0); 145 146 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock)); 147 148 static BLOCKING_NOTIFIER_HEAD(netlink_chain); 149 150 151 static const struct rhashtable_params netlink_rhashtable_params; 152 153 static inline u32 netlink_group_mask(u32 group) 154 { 155 return group ? 1 << (group - 1) : 0; 156 } 157 158 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb, 159 gfp_t gfp_mask) 160 { 161 unsigned int len = skb_end_offset(skb); 162 struct sk_buff *new; 163 164 new = alloc_skb(len, gfp_mask); 165 if (new == NULL) 166 return NULL; 167 168 NETLINK_CB(new).portid = NETLINK_CB(skb).portid; 169 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group; 170 NETLINK_CB(new).creds = NETLINK_CB(skb).creds; 171 172 skb_put_data(new, skb->data, len); 173 return new; 174 } 175 176 static unsigned int netlink_tap_net_id; 177 178 struct netlink_tap_net { 179 struct list_head netlink_tap_all; 180 struct mutex netlink_tap_lock; 181 }; 182 183 int netlink_add_tap(struct netlink_tap *nt) 184 { 185 struct net *net = dev_net(nt->dev); 186 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id); 187 188 if (unlikely(nt->dev->type != ARPHRD_NETLINK)) 189 return -EINVAL; 190 191 mutex_lock(&nn->netlink_tap_lock); 192 list_add_rcu(&nt->list, &nn->netlink_tap_all); 193 mutex_unlock(&nn->netlink_tap_lock); 194 195 __module_get(nt->module); 196 197 return 0; 198 } 199 EXPORT_SYMBOL_GPL(netlink_add_tap); 200 201 static int __netlink_remove_tap(struct netlink_tap *nt) 202 { 203 struct net *net = dev_net(nt->dev); 204 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id); 205 bool found = false; 206 struct netlink_tap *tmp; 207 208 mutex_lock(&nn->netlink_tap_lock); 209 210 list_for_each_entry(tmp, &nn->netlink_tap_all, list) { 211 if (nt == tmp) { 212 list_del_rcu(&nt->list); 213 found = true; 214 goto out; 215 } 216 } 217 218 pr_warn("__netlink_remove_tap: %p not found\n", nt); 219 out: 220 mutex_unlock(&nn->netlink_tap_lock); 221 222 if (found) 223 module_put(nt->module); 224 225 return found ? 0 : -ENODEV; 226 } 227 228 int netlink_remove_tap(struct netlink_tap *nt) 229 { 230 int ret; 231 232 ret = __netlink_remove_tap(nt); 233 synchronize_net(); 234 235 return ret; 236 } 237 EXPORT_SYMBOL_GPL(netlink_remove_tap); 238 239 static __net_init int netlink_tap_init_net(struct net *net) 240 { 241 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id); 242 243 INIT_LIST_HEAD(&nn->netlink_tap_all); 244 mutex_init(&nn->netlink_tap_lock); 245 return 0; 246 } 247 248 static void __net_exit netlink_tap_exit_net(struct net *net) 249 { 250 } 251 252 static struct pernet_operations netlink_tap_net_ops = { 253 .init = netlink_tap_init_net, 254 .exit = netlink_tap_exit_net, 255 .id = &netlink_tap_net_id, 256 .size = sizeof(struct netlink_tap_net), 257 }; 258 259 static bool netlink_filter_tap(const struct sk_buff *skb) 260 { 261 struct sock *sk = skb->sk; 262 263 /* We take the more conservative approach and 264 * whitelist socket protocols that may pass. 265 */ 266 switch (sk->sk_protocol) { 267 case NETLINK_ROUTE: 268 case NETLINK_USERSOCK: 269 case NETLINK_SOCK_DIAG: 270 case NETLINK_NFLOG: 271 case NETLINK_XFRM: 272 case NETLINK_FIB_LOOKUP: 273 case NETLINK_NETFILTER: 274 case NETLINK_GENERIC: 275 return true; 276 } 277 278 return false; 279 } 280 281 static int __netlink_deliver_tap_skb(struct sk_buff *skb, 282 struct net_device *dev) 283 { 284 struct sk_buff *nskb; 285 struct sock *sk = skb->sk; 286 int ret = -ENOMEM; 287 288 if (!net_eq(dev_net(dev), sock_net(sk))) 289 return 0; 290 291 dev_hold(dev); 292 293 if (is_vmalloc_addr(skb->head)) 294 nskb = netlink_to_full_skb(skb, GFP_ATOMIC); 295 else 296 nskb = skb_clone(skb, GFP_ATOMIC); 297 if (nskb) { 298 nskb->dev = dev; 299 nskb->protocol = htons((u16) sk->sk_protocol); 300 nskb->pkt_type = netlink_is_kernel(sk) ? 301 PACKET_KERNEL : PACKET_USER; 302 skb_reset_network_header(nskb); 303 ret = dev_queue_xmit(nskb); 304 if (unlikely(ret > 0)) 305 ret = net_xmit_errno(ret); 306 } 307 308 dev_put(dev); 309 return ret; 310 } 311 312 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn) 313 { 314 int ret; 315 struct netlink_tap *tmp; 316 317 if (!netlink_filter_tap(skb)) 318 return; 319 320 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) { 321 ret = __netlink_deliver_tap_skb(skb, tmp->dev); 322 if (unlikely(ret)) 323 break; 324 } 325 } 326 327 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb) 328 { 329 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id); 330 331 rcu_read_lock(); 332 333 if (unlikely(!list_empty(&nn->netlink_tap_all))) 334 __netlink_deliver_tap(skb, nn); 335 336 rcu_read_unlock(); 337 } 338 339 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src, 340 struct sk_buff *skb) 341 { 342 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src))) 343 netlink_deliver_tap(sock_net(dst), skb); 344 } 345 346 static void netlink_overrun(struct sock *sk) 347 { 348 struct netlink_sock *nlk = nlk_sk(sk); 349 350 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) { 351 if (!test_and_set_bit(NETLINK_S_CONGESTED, 352 &nlk_sk(sk)->state)) { 353 sk->sk_err = ENOBUFS; 354 sk->sk_error_report(sk); 355 } 356 } 357 atomic_inc(&sk->sk_drops); 358 } 359 360 static void netlink_rcv_wake(struct sock *sk) 361 { 362 struct netlink_sock *nlk = nlk_sk(sk); 363 364 if (skb_queue_empty(&sk->sk_receive_queue)) 365 clear_bit(NETLINK_S_CONGESTED, &nlk->state); 366 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state)) 367 wake_up_interruptible(&nlk->wait); 368 } 369 370 static void netlink_skb_destructor(struct sk_buff *skb) 371 { 372 if (is_vmalloc_addr(skb->head)) { 373 if (!skb->cloned || 374 !atomic_dec_return(&(skb_shinfo(skb)->dataref))) 375 vfree(skb->head); 376 377 skb->head = NULL; 378 } 379 if (skb->sk != NULL) 380 sock_rfree(skb); 381 } 382 383 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 384 { 385 WARN_ON(skb->sk != NULL); 386 skb->sk = sk; 387 skb->destructor = netlink_skb_destructor; 388 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 389 sk_mem_charge(sk, skb->truesize); 390 } 391 392 static void netlink_sock_destruct(struct sock *sk) 393 { 394 struct netlink_sock *nlk = nlk_sk(sk); 395 396 if (nlk->cb_running) { 397 if (nlk->cb.done) 398 nlk->cb.done(&nlk->cb); 399 module_put(nlk->cb.module); 400 kfree_skb(nlk->cb.skb); 401 } 402 403 skb_queue_purge(&sk->sk_receive_queue); 404 405 if (!sock_flag(sk, SOCK_DEAD)) { 406 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk); 407 return; 408 } 409 410 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 411 WARN_ON(refcount_read(&sk->sk_wmem_alloc)); 412 WARN_ON(nlk_sk(sk)->groups); 413 } 414 415 static void netlink_sock_destruct_work(struct work_struct *work) 416 { 417 struct netlink_sock *nlk = container_of(work, struct netlink_sock, 418 work); 419 420 sk_free(&nlk->sk); 421 } 422 423 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on 424 * SMP. Look, when several writers sleep and reader wakes them up, all but one 425 * immediately hit write lock and grab all the cpus. Exclusive sleep solves 426 * this, _but_ remember, it adds useless work on UP machines. 427 */ 428 429 void netlink_table_grab(void) 430 __acquires(nl_table_lock) 431 { 432 might_sleep(); 433 434 write_lock_irq(&nl_table_lock); 435 436 if (atomic_read(&nl_table_users)) { 437 DECLARE_WAITQUEUE(wait, current); 438 439 add_wait_queue_exclusive(&nl_table_wait, &wait); 440 for (;;) { 441 set_current_state(TASK_UNINTERRUPTIBLE); 442 if (atomic_read(&nl_table_users) == 0) 443 break; 444 write_unlock_irq(&nl_table_lock); 445 schedule(); 446 write_lock_irq(&nl_table_lock); 447 } 448 449 __set_current_state(TASK_RUNNING); 450 remove_wait_queue(&nl_table_wait, &wait); 451 } 452 } 453 454 void netlink_table_ungrab(void) 455 __releases(nl_table_lock) 456 { 457 write_unlock_irq(&nl_table_lock); 458 wake_up(&nl_table_wait); 459 } 460 461 static inline void 462 netlink_lock_table(void) 463 { 464 /* read_lock() synchronizes us to netlink_table_grab */ 465 466 read_lock(&nl_table_lock); 467 atomic_inc(&nl_table_users); 468 read_unlock(&nl_table_lock); 469 } 470 471 static inline void 472 netlink_unlock_table(void) 473 { 474 if (atomic_dec_and_test(&nl_table_users)) 475 wake_up(&nl_table_wait); 476 } 477 478 struct netlink_compare_arg 479 { 480 possible_net_t pnet; 481 u32 portid; 482 }; 483 484 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */ 485 #define netlink_compare_arg_len \ 486 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32)) 487 488 static inline int netlink_compare(struct rhashtable_compare_arg *arg, 489 const void *ptr) 490 { 491 const struct netlink_compare_arg *x = arg->key; 492 const struct netlink_sock *nlk = ptr; 493 494 return nlk->portid != x->portid || 495 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet)); 496 } 497 498 static void netlink_compare_arg_init(struct netlink_compare_arg *arg, 499 struct net *net, u32 portid) 500 { 501 memset(arg, 0, sizeof(*arg)); 502 write_pnet(&arg->pnet, net); 503 arg->portid = portid; 504 } 505 506 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid, 507 struct net *net) 508 { 509 struct netlink_compare_arg arg; 510 511 netlink_compare_arg_init(&arg, net, portid); 512 return rhashtable_lookup_fast(&table->hash, &arg, 513 netlink_rhashtable_params); 514 } 515 516 static int __netlink_insert(struct netlink_table *table, struct sock *sk) 517 { 518 struct netlink_compare_arg arg; 519 520 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid); 521 return rhashtable_lookup_insert_key(&table->hash, &arg, 522 &nlk_sk(sk)->node, 523 netlink_rhashtable_params); 524 } 525 526 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid) 527 { 528 struct netlink_table *table = &nl_table[protocol]; 529 struct sock *sk; 530 531 rcu_read_lock(); 532 sk = __netlink_lookup(table, portid, net); 533 if (sk) 534 sock_hold(sk); 535 rcu_read_unlock(); 536 537 return sk; 538 } 539 540 static const struct proto_ops netlink_ops; 541 542 static void 543 netlink_update_listeners(struct sock *sk) 544 { 545 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 546 unsigned long mask; 547 unsigned int i; 548 struct listeners *listeners; 549 550 listeners = nl_deref_protected(tbl->listeners); 551 if (!listeners) 552 return; 553 554 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) { 555 mask = 0; 556 sk_for_each_bound(sk, &tbl->mc_list) { 557 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups)) 558 mask |= nlk_sk(sk)->groups[i]; 559 } 560 listeners->masks[i] = mask; 561 } 562 /* this function is only called with the netlink table "grabbed", which 563 * makes sure updates are visible before bind or setsockopt return. */ 564 } 565 566 static int netlink_insert(struct sock *sk, u32 portid) 567 { 568 struct netlink_table *table = &nl_table[sk->sk_protocol]; 569 int err; 570 571 lock_sock(sk); 572 573 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY; 574 if (nlk_sk(sk)->bound) 575 goto err; 576 577 nlk_sk(sk)->portid = portid; 578 sock_hold(sk); 579 580 err = __netlink_insert(table, sk); 581 if (err) { 582 /* In case the hashtable backend returns with -EBUSY 583 * from here, it must not escape to the caller. 584 */ 585 if (unlikely(err == -EBUSY)) 586 err = -EOVERFLOW; 587 if (err == -EEXIST) 588 err = -EADDRINUSE; 589 sock_put(sk); 590 goto err; 591 } 592 593 /* We need to ensure that the socket is hashed and visible. */ 594 smp_wmb(); 595 nlk_sk(sk)->bound = portid; 596 597 err: 598 release_sock(sk); 599 return err; 600 } 601 602 static void netlink_remove(struct sock *sk) 603 { 604 struct netlink_table *table; 605 606 table = &nl_table[sk->sk_protocol]; 607 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node, 608 netlink_rhashtable_params)) { 609 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 610 __sock_put(sk); 611 } 612 613 netlink_table_grab(); 614 if (nlk_sk(sk)->subscriptions) { 615 __sk_del_bind_node(sk); 616 netlink_update_listeners(sk); 617 } 618 if (sk->sk_protocol == NETLINK_GENERIC) 619 atomic_inc(&genl_sk_destructing_cnt); 620 netlink_table_ungrab(); 621 } 622 623 static struct proto netlink_proto = { 624 .name = "NETLINK", 625 .owner = THIS_MODULE, 626 .obj_size = sizeof(struct netlink_sock), 627 }; 628 629 static int __netlink_create(struct net *net, struct socket *sock, 630 struct mutex *cb_mutex, int protocol, 631 int kern) 632 { 633 struct sock *sk; 634 struct netlink_sock *nlk; 635 636 sock->ops = &netlink_ops; 637 638 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern); 639 if (!sk) 640 return -ENOMEM; 641 642 sock_init_data(sock, sk); 643 644 nlk = nlk_sk(sk); 645 if (cb_mutex) { 646 nlk->cb_mutex = cb_mutex; 647 } else { 648 nlk->cb_mutex = &nlk->cb_def_mutex; 649 mutex_init(nlk->cb_mutex); 650 lockdep_set_class_and_name(nlk->cb_mutex, 651 nlk_cb_mutex_keys + protocol, 652 nlk_cb_mutex_key_strings[protocol]); 653 } 654 init_waitqueue_head(&nlk->wait); 655 656 sk->sk_destruct = netlink_sock_destruct; 657 sk->sk_protocol = protocol; 658 return 0; 659 } 660 661 static int netlink_create(struct net *net, struct socket *sock, int protocol, 662 int kern) 663 { 664 struct module *module = NULL; 665 struct mutex *cb_mutex; 666 struct netlink_sock *nlk; 667 int (*bind)(struct net *net, int group); 668 void (*unbind)(struct net *net, int group); 669 int err = 0; 670 671 sock->state = SS_UNCONNECTED; 672 673 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 674 return -ESOCKTNOSUPPORT; 675 676 if (protocol < 0 || protocol >= MAX_LINKS) 677 return -EPROTONOSUPPORT; 678 protocol = array_index_nospec(protocol, MAX_LINKS); 679 680 netlink_lock_table(); 681 #ifdef CONFIG_MODULES 682 if (!nl_table[protocol].registered) { 683 netlink_unlock_table(); 684 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol); 685 netlink_lock_table(); 686 } 687 #endif 688 if (nl_table[protocol].registered && 689 try_module_get(nl_table[protocol].module)) 690 module = nl_table[protocol].module; 691 else 692 err = -EPROTONOSUPPORT; 693 cb_mutex = nl_table[protocol].cb_mutex; 694 bind = nl_table[protocol].bind; 695 unbind = nl_table[protocol].unbind; 696 netlink_unlock_table(); 697 698 if (err < 0) 699 goto out; 700 701 err = __netlink_create(net, sock, cb_mutex, protocol, kern); 702 if (err < 0) 703 goto out_module; 704 705 local_bh_disable(); 706 sock_prot_inuse_add(net, &netlink_proto, 1); 707 local_bh_enable(); 708 709 nlk = nlk_sk(sock->sk); 710 nlk->module = module; 711 nlk->netlink_bind = bind; 712 nlk->netlink_unbind = unbind; 713 out: 714 return err; 715 716 out_module: 717 module_put(module); 718 goto out; 719 } 720 721 static void deferred_put_nlk_sk(struct rcu_head *head) 722 { 723 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu); 724 struct sock *sk = &nlk->sk; 725 726 kfree(nlk->groups); 727 nlk->groups = NULL; 728 729 if (!refcount_dec_and_test(&sk->sk_refcnt)) 730 return; 731 732 if (nlk->cb_running && nlk->cb.done) { 733 INIT_WORK(&nlk->work, netlink_sock_destruct_work); 734 schedule_work(&nlk->work); 735 return; 736 } 737 738 sk_free(sk); 739 } 740 741 static int netlink_release(struct socket *sock) 742 { 743 struct sock *sk = sock->sk; 744 struct netlink_sock *nlk; 745 746 if (!sk) 747 return 0; 748 749 netlink_remove(sk); 750 sock_orphan(sk); 751 nlk = nlk_sk(sk); 752 753 /* 754 * OK. Socket is unlinked, any packets that arrive now 755 * will be purged. 756 */ 757 758 /* must not acquire netlink_table_lock in any way again before unbind 759 * and notifying genetlink is done as otherwise it might deadlock 760 */ 761 if (nlk->netlink_unbind) { 762 int i; 763 764 for (i = 0; i < nlk->ngroups; i++) 765 if (test_bit(i, nlk->groups)) 766 nlk->netlink_unbind(sock_net(sk), i + 1); 767 } 768 if (sk->sk_protocol == NETLINK_GENERIC && 769 atomic_dec_return(&genl_sk_destructing_cnt) == 0) 770 wake_up(&genl_sk_destructing_waitq); 771 772 sock->sk = NULL; 773 wake_up_interruptible_all(&nlk->wait); 774 775 skb_queue_purge(&sk->sk_write_queue); 776 777 if (nlk->portid && nlk->bound) { 778 struct netlink_notify n = { 779 .net = sock_net(sk), 780 .protocol = sk->sk_protocol, 781 .portid = nlk->portid, 782 }; 783 blocking_notifier_call_chain(&netlink_chain, 784 NETLINK_URELEASE, &n); 785 } 786 787 module_put(nlk->module); 788 789 if (netlink_is_kernel(sk)) { 790 netlink_table_grab(); 791 BUG_ON(nl_table[sk->sk_protocol].registered == 0); 792 if (--nl_table[sk->sk_protocol].registered == 0) { 793 struct listeners *old; 794 795 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners); 796 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL); 797 kfree_rcu(old, rcu); 798 nl_table[sk->sk_protocol].module = NULL; 799 nl_table[sk->sk_protocol].bind = NULL; 800 nl_table[sk->sk_protocol].unbind = NULL; 801 nl_table[sk->sk_protocol].flags = 0; 802 nl_table[sk->sk_protocol].registered = 0; 803 } 804 netlink_table_ungrab(); 805 } 806 807 local_bh_disable(); 808 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1); 809 local_bh_enable(); 810 call_rcu(&nlk->rcu, deferred_put_nlk_sk); 811 return 0; 812 } 813 814 static int netlink_autobind(struct socket *sock) 815 { 816 struct sock *sk = sock->sk; 817 struct net *net = sock_net(sk); 818 struct netlink_table *table = &nl_table[sk->sk_protocol]; 819 s32 portid = task_tgid_vnr(current); 820 int err; 821 s32 rover = -4096; 822 bool ok; 823 824 retry: 825 cond_resched(); 826 rcu_read_lock(); 827 ok = !__netlink_lookup(table, portid, net); 828 rcu_read_unlock(); 829 if (!ok) { 830 /* Bind collision, search negative portid values. */ 831 if (rover == -4096) 832 /* rover will be in range [S32_MIN, -4097] */ 833 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN); 834 else if (rover >= -4096) 835 rover = -4097; 836 portid = rover--; 837 goto retry; 838 } 839 840 err = netlink_insert(sk, portid); 841 if (err == -EADDRINUSE) 842 goto retry; 843 844 /* If 2 threads race to autobind, that is fine. */ 845 if (err == -EBUSY) 846 err = 0; 847 848 return err; 849 } 850 851 /** 852 * __netlink_ns_capable - General netlink message capability test 853 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace. 854 * @user_ns: The user namespace of the capability to use 855 * @cap: The capability to use 856 * 857 * Test to see if the opener of the socket we received the message 858 * from had when the netlink socket was created and the sender of the 859 * message has has the capability @cap in the user namespace @user_ns. 860 */ 861 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp, 862 struct user_namespace *user_ns, int cap) 863 { 864 return ((nsp->flags & NETLINK_SKB_DST) || 865 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) && 866 ns_capable(user_ns, cap); 867 } 868 EXPORT_SYMBOL(__netlink_ns_capable); 869 870 /** 871 * netlink_ns_capable - General netlink message capability test 872 * @skb: socket buffer holding a netlink command from userspace 873 * @user_ns: The user namespace of the capability to use 874 * @cap: The capability to use 875 * 876 * Test to see if the opener of the socket we received the message 877 * from had when the netlink socket was created and the sender of the 878 * message has has the capability @cap in the user namespace @user_ns. 879 */ 880 bool netlink_ns_capable(const struct sk_buff *skb, 881 struct user_namespace *user_ns, int cap) 882 { 883 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap); 884 } 885 EXPORT_SYMBOL(netlink_ns_capable); 886 887 /** 888 * netlink_capable - Netlink global message capability test 889 * @skb: socket buffer holding a netlink command from userspace 890 * @cap: The capability to use 891 * 892 * Test to see if the opener of the socket we received the message 893 * from had when the netlink socket was created and the sender of the 894 * message has has the capability @cap in all user namespaces. 895 */ 896 bool netlink_capable(const struct sk_buff *skb, int cap) 897 { 898 return netlink_ns_capable(skb, &init_user_ns, cap); 899 } 900 EXPORT_SYMBOL(netlink_capable); 901 902 /** 903 * netlink_net_capable - Netlink network namespace message capability test 904 * @skb: socket buffer holding a netlink command from userspace 905 * @cap: The capability to use 906 * 907 * Test to see if the opener of the socket we received the message 908 * from had when the netlink socket was created and the sender of the 909 * message has has the capability @cap over the network namespace of 910 * the socket we received the message from. 911 */ 912 bool netlink_net_capable(const struct sk_buff *skb, int cap) 913 { 914 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap); 915 } 916 EXPORT_SYMBOL(netlink_net_capable); 917 918 static inline int netlink_allowed(const struct socket *sock, unsigned int flag) 919 { 920 return (nl_table[sock->sk->sk_protocol].flags & flag) || 921 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN); 922 } 923 924 static void 925 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions) 926 { 927 struct netlink_sock *nlk = nlk_sk(sk); 928 929 if (nlk->subscriptions && !subscriptions) 930 __sk_del_bind_node(sk); 931 else if (!nlk->subscriptions && subscriptions) 932 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 933 nlk->subscriptions = subscriptions; 934 } 935 936 static int netlink_realloc_groups(struct sock *sk) 937 { 938 struct netlink_sock *nlk = nlk_sk(sk); 939 unsigned int groups; 940 unsigned long *new_groups; 941 int err = 0; 942 943 netlink_table_grab(); 944 945 groups = nl_table[sk->sk_protocol].groups; 946 if (!nl_table[sk->sk_protocol].registered) { 947 err = -ENOENT; 948 goto out_unlock; 949 } 950 951 if (nlk->ngroups >= groups) 952 goto out_unlock; 953 954 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC); 955 if (new_groups == NULL) { 956 err = -ENOMEM; 957 goto out_unlock; 958 } 959 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0, 960 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups)); 961 962 nlk->groups = new_groups; 963 nlk->ngroups = groups; 964 out_unlock: 965 netlink_table_ungrab(); 966 return err; 967 } 968 969 static void netlink_undo_bind(int group, long unsigned int groups, 970 struct sock *sk) 971 { 972 struct netlink_sock *nlk = nlk_sk(sk); 973 int undo; 974 975 if (!nlk->netlink_unbind) 976 return; 977 978 for (undo = 0; undo < group; undo++) 979 if (test_bit(undo, &groups)) 980 nlk->netlink_unbind(sock_net(sk), undo + 1); 981 } 982 983 static int netlink_bind(struct socket *sock, struct sockaddr *addr, 984 int addr_len) 985 { 986 struct sock *sk = sock->sk; 987 struct net *net = sock_net(sk); 988 struct netlink_sock *nlk = nlk_sk(sk); 989 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 990 int err = 0; 991 unsigned long groups; 992 bool bound; 993 994 if (addr_len < sizeof(struct sockaddr_nl)) 995 return -EINVAL; 996 997 if (nladdr->nl_family != AF_NETLINK) 998 return -EINVAL; 999 groups = nladdr->nl_groups; 1000 1001 /* Only superuser is allowed to listen multicasts */ 1002 if (groups) { 1003 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV)) 1004 return -EPERM; 1005 err = netlink_realloc_groups(sk); 1006 if (err) 1007 return err; 1008 } 1009 1010 if (nlk->ngroups < BITS_PER_LONG) 1011 groups &= (1UL << nlk->ngroups) - 1; 1012 1013 bound = nlk->bound; 1014 if (bound) { 1015 /* Ensure nlk->portid is up-to-date. */ 1016 smp_rmb(); 1017 1018 if (nladdr->nl_pid != nlk->portid) 1019 return -EINVAL; 1020 } 1021 1022 netlink_lock_table(); 1023 if (nlk->netlink_bind && groups) { 1024 int group; 1025 1026 for (group = 0; group < nlk->ngroups; group++) { 1027 if (!test_bit(group, &groups)) 1028 continue; 1029 err = nlk->netlink_bind(net, group + 1); 1030 if (!err) 1031 continue; 1032 netlink_undo_bind(group, groups, sk); 1033 goto unlock; 1034 } 1035 } 1036 1037 /* No need for barriers here as we return to user-space without 1038 * using any of the bound attributes. 1039 */ 1040 if (!bound) { 1041 err = nladdr->nl_pid ? 1042 netlink_insert(sk, nladdr->nl_pid) : 1043 netlink_autobind(sock); 1044 if (err) { 1045 netlink_undo_bind(nlk->ngroups, groups, sk); 1046 goto unlock; 1047 } 1048 } 1049 1050 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0])) 1051 goto unlock; 1052 netlink_unlock_table(); 1053 1054 netlink_table_grab(); 1055 netlink_update_subscriptions(sk, nlk->subscriptions + 1056 hweight32(groups) - 1057 hweight32(nlk->groups[0])); 1058 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups; 1059 netlink_update_listeners(sk); 1060 netlink_table_ungrab(); 1061 1062 return 0; 1063 1064 unlock: 1065 netlink_unlock_table(); 1066 return err; 1067 } 1068 1069 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 1070 int alen, int flags) 1071 { 1072 int err = 0; 1073 struct sock *sk = sock->sk; 1074 struct netlink_sock *nlk = nlk_sk(sk); 1075 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 1076 1077 if (alen < sizeof(addr->sa_family)) 1078 return -EINVAL; 1079 1080 if (addr->sa_family == AF_UNSPEC) { 1081 sk->sk_state = NETLINK_UNCONNECTED; 1082 nlk->dst_portid = 0; 1083 nlk->dst_group = 0; 1084 return 0; 1085 } 1086 if (addr->sa_family != AF_NETLINK) 1087 return -EINVAL; 1088 1089 if (alen < sizeof(struct sockaddr_nl)) 1090 return -EINVAL; 1091 1092 if ((nladdr->nl_groups || nladdr->nl_pid) && 1093 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND)) 1094 return -EPERM; 1095 1096 /* No need for barriers here as we return to user-space without 1097 * using any of the bound attributes. 1098 */ 1099 if (!nlk->bound) 1100 err = netlink_autobind(sock); 1101 1102 if (err == 0) { 1103 sk->sk_state = NETLINK_CONNECTED; 1104 nlk->dst_portid = nladdr->nl_pid; 1105 nlk->dst_group = ffs(nladdr->nl_groups); 1106 } 1107 1108 return err; 1109 } 1110 1111 static int netlink_getname(struct socket *sock, struct sockaddr *addr, 1112 int peer) 1113 { 1114 struct sock *sk = sock->sk; 1115 struct netlink_sock *nlk = nlk_sk(sk); 1116 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr); 1117 1118 nladdr->nl_family = AF_NETLINK; 1119 nladdr->nl_pad = 0; 1120 1121 if (peer) { 1122 nladdr->nl_pid = nlk->dst_portid; 1123 nladdr->nl_groups = netlink_group_mask(nlk->dst_group); 1124 } else { 1125 nladdr->nl_pid = nlk->portid; 1126 netlink_lock_table(); 1127 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0; 1128 netlink_unlock_table(); 1129 } 1130 return sizeof(*nladdr); 1131 } 1132 1133 static int netlink_ioctl(struct socket *sock, unsigned int cmd, 1134 unsigned long arg) 1135 { 1136 /* try to hand this ioctl down to the NIC drivers. 1137 */ 1138 return -ENOIOCTLCMD; 1139 } 1140 1141 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid) 1142 { 1143 struct sock *sock; 1144 struct netlink_sock *nlk; 1145 1146 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid); 1147 if (!sock) 1148 return ERR_PTR(-ECONNREFUSED); 1149 1150 /* Don't bother queuing skb if kernel socket has no input function */ 1151 nlk = nlk_sk(sock); 1152 if (sock->sk_state == NETLINK_CONNECTED && 1153 nlk->dst_portid != nlk_sk(ssk)->portid) { 1154 sock_put(sock); 1155 return ERR_PTR(-ECONNREFUSED); 1156 } 1157 return sock; 1158 } 1159 1160 struct sock *netlink_getsockbyfilp(struct file *filp) 1161 { 1162 struct inode *inode = file_inode(filp); 1163 struct sock *sock; 1164 1165 if (!S_ISSOCK(inode->i_mode)) 1166 return ERR_PTR(-ENOTSOCK); 1167 1168 sock = SOCKET_I(inode)->sk; 1169 if (sock->sk_family != AF_NETLINK) 1170 return ERR_PTR(-EINVAL); 1171 1172 sock_hold(sock); 1173 return sock; 1174 } 1175 1176 static struct sk_buff *netlink_alloc_large_skb(unsigned int size, 1177 int broadcast) 1178 { 1179 struct sk_buff *skb; 1180 void *data; 1181 1182 if (size <= NLMSG_GOODSIZE || broadcast) 1183 return alloc_skb(size, GFP_KERNEL); 1184 1185 size = SKB_DATA_ALIGN(size) + 1186 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 1187 1188 data = vmalloc(size); 1189 if (data == NULL) 1190 return NULL; 1191 1192 skb = __build_skb(data, size); 1193 if (skb == NULL) 1194 vfree(data); 1195 else 1196 skb->destructor = netlink_skb_destructor; 1197 1198 return skb; 1199 } 1200 1201 /* 1202 * Attach a skb to a netlink socket. 1203 * The caller must hold a reference to the destination socket. On error, the 1204 * reference is dropped. The skb is not send to the destination, just all 1205 * all error checks are performed and memory in the queue is reserved. 1206 * Return values: 1207 * < 0: error. skb freed, reference to sock dropped. 1208 * 0: continue 1209 * 1: repeat lookup - reference dropped while waiting for socket memory. 1210 */ 1211 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, 1212 long *timeo, struct sock *ssk) 1213 { 1214 struct netlink_sock *nlk; 1215 1216 nlk = nlk_sk(sk); 1217 1218 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 1219 test_bit(NETLINK_S_CONGESTED, &nlk->state))) { 1220 DECLARE_WAITQUEUE(wait, current); 1221 if (!*timeo) { 1222 if (!ssk || netlink_is_kernel(ssk)) 1223 netlink_overrun(sk); 1224 sock_put(sk); 1225 kfree_skb(skb); 1226 return -EAGAIN; 1227 } 1228 1229 __set_current_state(TASK_INTERRUPTIBLE); 1230 add_wait_queue(&nlk->wait, &wait); 1231 1232 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 1233 test_bit(NETLINK_S_CONGESTED, &nlk->state)) && 1234 !sock_flag(sk, SOCK_DEAD)) 1235 *timeo = schedule_timeout(*timeo); 1236 1237 __set_current_state(TASK_RUNNING); 1238 remove_wait_queue(&nlk->wait, &wait); 1239 sock_put(sk); 1240 1241 if (signal_pending(current)) { 1242 kfree_skb(skb); 1243 return sock_intr_errno(*timeo); 1244 } 1245 return 1; 1246 } 1247 netlink_skb_set_owner_r(skb, sk); 1248 return 0; 1249 } 1250 1251 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb) 1252 { 1253 int len = skb->len; 1254 1255 netlink_deliver_tap(sock_net(sk), skb); 1256 1257 skb_queue_tail(&sk->sk_receive_queue, skb); 1258 sk->sk_data_ready(sk); 1259 return len; 1260 } 1261 1262 int netlink_sendskb(struct sock *sk, struct sk_buff *skb) 1263 { 1264 int len = __netlink_sendskb(sk, skb); 1265 1266 sock_put(sk); 1267 return len; 1268 } 1269 1270 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 1271 { 1272 kfree_skb(skb); 1273 sock_put(sk); 1274 } 1275 1276 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation) 1277 { 1278 int delta; 1279 1280 WARN_ON(skb->sk != NULL); 1281 delta = skb->end - skb->tail; 1282 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize) 1283 return skb; 1284 1285 if (skb_shared(skb)) { 1286 struct sk_buff *nskb = skb_clone(skb, allocation); 1287 if (!nskb) 1288 return skb; 1289 consume_skb(skb); 1290 skb = nskb; 1291 } 1292 1293 pskb_expand_head(skb, 0, -delta, 1294 (allocation & ~__GFP_DIRECT_RECLAIM) | 1295 __GFP_NOWARN | __GFP_NORETRY); 1296 return skb; 1297 } 1298 1299 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb, 1300 struct sock *ssk) 1301 { 1302 int ret; 1303 struct netlink_sock *nlk = nlk_sk(sk); 1304 1305 ret = -ECONNREFUSED; 1306 if (nlk->netlink_rcv != NULL) { 1307 ret = skb->len; 1308 netlink_skb_set_owner_r(skb, sk); 1309 NETLINK_CB(skb).sk = ssk; 1310 netlink_deliver_tap_kernel(sk, ssk, skb); 1311 nlk->netlink_rcv(skb); 1312 consume_skb(skb); 1313 } else { 1314 kfree_skb(skb); 1315 } 1316 sock_put(sk); 1317 return ret; 1318 } 1319 1320 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, 1321 u32 portid, int nonblock) 1322 { 1323 struct sock *sk; 1324 int err; 1325 long timeo; 1326 1327 skb = netlink_trim(skb, gfp_any()); 1328 1329 timeo = sock_sndtimeo(ssk, nonblock); 1330 retry: 1331 sk = netlink_getsockbyportid(ssk, portid); 1332 if (IS_ERR(sk)) { 1333 kfree_skb(skb); 1334 return PTR_ERR(sk); 1335 } 1336 if (netlink_is_kernel(sk)) 1337 return netlink_unicast_kernel(sk, skb, ssk); 1338 1339 if (sk_filter(sk, skb)) { 1340 err = skb->len; 1341 kfree_skb(skb); 1342 sock_put(sk); 1343 return err; 1344 } 1345 1346 err = netlink_attachskb(sk, skb, &timeo, ssk); 1347 if (err == 1) 1348 goto retry; 1349 if (err) 1350 return err; 1351 1352 return netlink_sendskb(sk, skb); 1353 } 1354 EXPORT_SYMBOL(netlink_unicast); 1355 1356 int netlink_has_listeners(struct sock *sk, unsigned int group) 1357 { 1358 int res = 0; 1359 struct listeners *listeners; 1360 1361 BUG_ON(!netlink_is_kernel(sk)); 1362 1363 rcu_read_lock(); 1364 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners); 1365 1366 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups) 1367 res = test_bit(group - 1, listeners->masks); 1368 1369 rcu_read_unlock(); 1370 1371 return res; 1372 } 1373 EXPORT_SYMBOL_GPL(netlink_has_listeners); 1374 1375 bool netlink_strict_get_check(struct sk_buff *skb) 1376 { 1377 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk); 1378 1379 return nlk->flags & NETLINK_F_STRICT_CHK; 1380 } 1381 EXPORT_SYMBOL_GPL(netlink_strict_get_check); 1382 1383 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 1384 { 1385 struct netlink_sock *nlk = nlk_sk(sk); 1386 1387 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 1388 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) { 1389 netlink_skb_set_owner_r(skb, sk); 1390 __netlink_sendskb(sk, skb); 1391 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1); 1392 } 1393 return -1; 1394 } 1395 1396 struct netlink_broadcast_data { 1397 struct sock *exclude_sk; 1398 struct net *net; 1399 u32 portid; 1400 u32 group; 1401 int failure; 1402 int delivery_failure; 1403 int congested; 1404 int delivered; 1405 gfp_t allocation; 1406 struct sk_buff *skb, *skb2; 1407 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data); 1408 void *tx_data; 1409 }; 1410 1411 static void do_one_broadcast(struct sock *sk, 1412 struct netlink_broadcast_data *p) 1413 { 1414 struct netlink_sock *nlk = nlk_sk(sk); 1415 int val; 1416 1417 if (p->exclude_sk == sk) 1418 return; 1419 1420 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 1421 !test_bit(p->group - 1, nlk->groups)) 1422 return; 1423 1424 if (!net_eq(sock_net(sk), p->net)) { 1425 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID)) 1426 return; 1427 1428 if (!peernet_has_id(sock_net(sk), p->net)) 1429 return; 1430 1431 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns, 1432 CAP_NET_BROADCAST)) 1433 return; 1434 } 1435 1436 if (p->failure) { 1437 netlink_overrun(sk); 1438 return; 1439 } 1440 1441 sock_hold(sk); 1442 if (p->skb2 == NULL) { 1443 if (skb_shared(p->skb)) { 1444 p->skb2 = skb_clone(p->skb, p->allocation); 1445 } else { 1446 p->skb2 = skb_get(p->skb); 1447 /* 1448 * skb ownership may have been set when 1449 * delivered to a previous socket. 1450 */ 1451 skb_orphan(p->skb2); 1452 } 1453 } 1454 if (p->skb2 == NULL) { 1455 netlink_overrun(sk); 1456 /* Clone failed. Notify ALL listeners. */ 1457 p->failure = 1; 1458 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR) 1459 p->delivery_failure = 1; 1460 goto out; 1461 } 1462 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) { 1463 kfree_skb(p->skb2); 1464 p->skb2 = NULL; 1465 goto out; 1466 } 1467 if (sk_filter(sk, p->skb2)) { 1468 kfree_skb(p->skb2); 1469 p->skb2 = NULL; 1470 goto out; 1471 } 1472 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net); 1473 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED) 1474 NETLINK_CB(p->skb2).nsid_is_set = true; 1475 val = netlink_broadcast_deliver(sk, p->skb2); 1476 if (val < 0) { 1477 netlink_overrun(sk); 1478 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR) 1479 p->delivery_failure = 1; 1480 } else { 1481 p->congested |= val; 1482 p->delivered = 1; 1483 p->skb2 = NULL; 1484 } 1485 out: 1486 sock_put(sk); 1487 } 1488 1489 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid, 1490 u32 group, gfp_t allocation, 1491 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data), 1492 void *filter_data) 1493 { 1494 struct net *net = sock_net(ssk); 1495 struct netlink_broadcast_data info; 1496 struct sock *sk; 1497 1498 skb = netlink_trim(skb, allocation); 1499 1500 info.exclude_sk = ssk; 1501 info.net = net; 1502 info.portid = portid; 1503 info.group = group; 1504 info.failure = 0; 1505 info.delivery_failure = 0; 1506 info.congested = 0; 1507 info.delivered = 0; 1508 info.allocation = allocation; 1509 info.skb = skb; 1510 info.skb2 = NULL; 1511 info.tx_filter = filter; 1512 info.tx_data = filter_data; 1513 1514 /* While we sleep in clone, do not allow to change socket list */ 1515 1516 netlink_lock_table(); 1517 1518 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) 1519 do_one_broadcast(sk, &info); 1520 1521 consume_skb(skb); 1522 1523 netlink_unlock_table(); 1524 1525 if (info.delivery_failure) { 1526 kfree_skb(info.skb2); 1527 return -ENOBUFS; 1528 } 1529 consume_skb(info.skb2); 1530 1531 if (info.delivered) { 1532 if (info.congested && gfpflags_allow_blocking(allocation)) 1533 yield(); 1534 return 0; 1535 } 1536 return -ESRCH; 1537 } 1538 EXPORT_SYMBOL(netlink_broadcast_filtered); 1539 1540 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid, 1541 u32 group, gfp_t allocation) 1542 { 1543 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation, 1544 NULL, NULL); 1545 } 1546 EXPORT_SYMBOL(netlink_broadcast); 1547 1548 struct netlink_set_err_data { 1549 struct sock *exclude_sk; 1550 u32 portid; 1551 u32 group; 1552 int code; 1553 }; 1554 1555 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p) 1556 { 1557 struct netlink_sock *nlk = nlk_sk(sk); 1558 int ret = 0; 1559 1560 if (sk == p->exclude_sk) 1561 goto out; 1562 1563 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk))) 1564 goto out; 1565 1566 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 1567 !test_bit(p->group - 1, nlk->groups)) 1568 goto out; 1569 1570 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) { 1571 ret = 1; 1572 goto out; 1573 } 1574 1575 sk->sk_err = p->code; 1576 sk->sk_error_report(sk); 1577 out: 1578 return ret; 1579 } 1580 1581 /** 1582 * netlink_set_err - report error to broadcast listeners 1583 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create() 1584 * @portid: the PORTID of a process that we want to skip (if any) 1585 * @group: the broadcast group that will notice the error 1586 * @code: error code, must be negative (as usual in kernelspace) 1587 * 1588 * This function returns the number of broadcast listeners that have set the 1589 * NETLINK_NO_ENOBUFS socket option. 1590 */ 1591 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code) 1592 { 1593 struct netlink_set_err_data info; 1594 struct sock *sk; 1595 int ret = 0; 1596 1597 info.exclude_sk = ssk; 1598 info.portid = portid; 1599 info.group = group; 1600 /* sk->sk_err wants a positive error value */ 1601 info.code = -code; 1602 1603 read_lock(&nl_table_lock); 1604 1605 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) 1606 ret += do_one_set_err(sk, &info); 1607 1608 read_unlock(&nl_table_lock); 1609 return ret; 1610 } 1611 EXPORT_SYMBOL(netlink_set_err); 1612 1613 /* must be called with netlink table grabbed */ 1614 static void netlink_update_socket_mc(struct netlink_sock *nlk, 1615 unsigned int group, 1616 int is_new) 1617 { 1618 int old, new = !!is_new, subscriptions; 1619 1620 old = test_bit(group - 1, nlk->groups); 1621 subscriptions = nlk->subscriptions - old + new; 1622 if (new) 1623 __set_bit(group - 1, nlk->groups); 1624 else 1625 __clear_bit(group - 1, nlk->groups); 1626 netlink_update_subscriptions(&nlk->sk, subscriptions); 1627 netlink_update_listeners(&nlk->sk); 1628 } 1629 1630 static int netlink_setsockopt(struct socket *sock, int level, int optname, 1631 char __user *optval, unsigned int optlen) 1632 { 1633 struct sock *sk = sock->sk; 1634 struct netlink_sock *nlk = nlk_sk(sk); 1635 unsigned int val = 0; 1636 int err; 1637 1638 if (level != SOL_NETLINK) 1639 return -ENOPROTOOPT; 1640 1641 if (optlen >= sizeof(int) && 1642 get_user(val, (unsigned int __user *)optval)) 1643 return -EFAULT; 1644 1645 switch (optname) { 1646 case NETLINK_PKTINFO: 1647 if (val) 1648 nlk->flags |= NETLINK_F_RECV_PKTINFO; 1649 else 1650 nlk->flags &= ~NETLINK_F_RECV_PKTINFO; 1651 err = 0; 1652 break; 1653 case NETLINK_ADD_MEMBERSHIP: 1654 case NETLINK_DROP_MEMBERSHIP: { 1655 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV)) 1656 return -EPERM; 1657 err = netlink_realloc_groups(sk); 1658 if (err) 1659 return err; 1660 if (!val || val - 1 >= nlk->ngroups) 1661 return -EINVAL; 1662 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) { 1663 err = nlk->netlink_bind(sock_net(sk), val); 1664 if (err) 1665 return err; 1666 } 1667 netlink_table_grab(); 1668 netlink_update_socket_mc(nlk, val, 1669 optname == NETLINK_ADD_MEMBERSHIP); 1670 netlink_table_ungrab(); 1671 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind) 1672 nlk->netlink_unbind(sock_net(sk), val); 1673 1674 err = 0; 1675 break; 1676 } 1677 case NETLINK_BROADCAST_ERROR: 1678 if (val) 1679 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR; 1680 else 1681 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR; 1682 err = 0; 1683 break; 1684 case NETLINK_NO_ENOBUFS: 1685 if (val) { 1686 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS; 1687 clear_bit(NETLINK_S_CONGESTED, &nlk->state); 1688 wake_up_interruptible(&nlk->wait); 1689 } else { 1690 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS; 1691 } 1692 err = 0; 1693 break; 1694 case NETLINK_LISTEN_ALL_NSID: 1695 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST)) 1696 return -EPERM; 1697 1698 if (val) 1699 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID; 1700 else 1701 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID; 1702 err = 0; 1703 break; 1704 case NETLINK_CAP_ACK: 1705 if (val) 1706 nlk->flags |= NETLINK_F_CAP_ACK; 1707 else 1708 nlk->flags &= ~NETLINK_F_CAP_ACK; 1709 err = 0; 1710 break; 1711 case NETLINK_EXT_ACK: 1712 if (val) 1713 nlk->flags |= NETLINK_F_EXT_ACK; 1714 else 1715 nlk->flags &= ~NETLINK_F_EXT_ACK; 1716 err = 0; 1717 break; 1718 case NETLINK_GET_STRICT_CHK: 1719 if (val) 1720 nlk->flags |= NETLINK_F_STRICT_CHK; 1721 else 1722 nlk->flags &= ~NETLINK_F_STRICT_CHK; 1723 err = 0; 1724 break; 1725 default: 1726 err = -ENOPROTOOPT; 1727 } 1728 return err; 1729 } 1730 1731 static int netlink_getsockopt(struct socket *sock, int level, int optname, 1732 char __user *optval, int __user *optlen) 1733 { 1734 struct sock *sk = sock->sk; 1735 struct netlink_sock *nlk = nlk_sk(sk); 1736 int len, val, err; 1737 1738 if (level != SOL_NETLINK) 1739 return -ENOPROTOOPT; 1740 1741 if (get_user(len, optlen)) 1742 return -EFAULT; 1743 if (len < 0) 1744 return -EINVAL; 1745 1746 switch (optname) { 1747 case NETLINK_PKTINFO: 1748 if (len < sizeof(int)) 1749 return -EINVAL; 1750 len = sizeof(int); 1751 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0; 1752 if (put_user(len, optlen) || 1753 put_user(val, optval)) 1754 return -EFAULT; 1755 err = 0; 1756 break; 1757 case NETLINK_BROADCAST_ERROR: 1758 if (len < sizeof(int)) 1759 return -EINVAL; 1760 len = sizeof(int); 1761 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0; 1762 if (put_user(len, optlen) || 1763 put_user(val, optval)) 1764 return -EFAULT; 1765 err = 0; 1766 break; 1767 case NETLINK_NO_ENOBUFS: 1768 if (len < sizeof(int)) 1769 return -EINVAL; 1770 len = sizeof(int); 1771 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0; 1772 if (put_user(len, optlen) || 1773 put_user(val, optval)) 1774 return -EFAULT; 1775 err = 0; 1776 break; 1777 case NETLINK_LIST_MEMBERSHIPS: { 1778 int pos, idx, shift; 1779 1780 err = 0; 1781 netlink_lock_table(); 1782 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) { 1783 if (len - pos < sizeof(u32)) 1784 break; 1785 1786 idx = pos / sizeof(unsigned long); 1787 shift = (pos % sizeof(unsigned long)) * 8; 1788 if (put_user((u32)(nlk->groups[idx] >> shift), 1789 (u32 __user *)(optval + pos))) { 1790 err = -EFAULT; 1791 break; 1792 } 1793 } 1794 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen)) 1795 err = -EFAULT; 1796 netlink_unlock_table(); 1797 break; 1798 } 1799 case NETLINK_CAP_ACK: 1800 if (len < sizeof(int)) 1801 return -EINVAL; 1802 len = sizeof(int); 1803 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0; 1804 if (put_user(len, optlen) || 1805 put_user(val, optval)) 1806 return -EFAULT; 1807 err = 0; 1808 break; 1809 case NETLINK_EXT_ACK: 1810 if (len < sizeof(int)) 1811 return -EINVAL; 1812 len = sizeof(int); 1813 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0; 1814 if (put_user(len, optlen) || put_user(val, optval)) 1815 return -EFAULT; 1816 err = 0; 1817 break; 1818 case NETLINK_GET_STRICT_CHK: 1819 if (len < sizeof(int)) 1820 return -EINVAL; 1821 len = sizeof(int); 1822 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0; 1823 if (put_user(len, optlen) || put_user(val, optval)) 1824 return -EFAULT; 1825 err = 0; 1826 break; 1827 default: 1828 err = -ENOPROTOOPT; 1829 } 1830 return err; 1831 } 1832 1833 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb) 1834 { 1835 struct nl_pktinfo info; 1836 1837 info.group = NETLINK_CB(skb).dst_group; 1838 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info); 1839 } 1840 1841 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg, 1842 struct sk_buff *skb) 1843 { 1844 if (!NETLINK_CB(skb).nsid_is_set) 1845 return; 1846 1847 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int), 1848 &NETLINK_CB(skb).nsid); 1849 } 1850 1851 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1852 { 1853 struct sock *sk = sock->sk; 1854 struct netlink_sock *nlk = nlk_sk(sk); 1855 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name); 1856 u32 dst_portid; 1857 u32 dst_group; 1858 struct sk_buff *skb; 1859 int err; 1860 struct scm_cookie scm; 1861 u32 netlink_skb_flags = 0; 1862 1863 if (msg->msg_flags&MSG_OOB) 1864 return -EOPNOTSUPP; 1865 1866 err = scm_send(sock, msg, &scm, true); 1867 if (err < 0) 1868 return err; 1869 1870 if (msg->msg_namelen) { 1871 err = -EINVAL; 1872 if (msg->msg_namelen < sizeof(struct sockaddr_nl)) 1873 goto out; 1874 if (addr->nl_family != AF_NETLINK) 1875 goto out; 1876 dst_portid = addr->nl_pid; 1877 dst_group = ffs(addr->nl_groups); 1878 err = -EPERM; 1879 if ((dst_group || dst_portid) && 1880 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND)) 1881 goto out; 1882 netlink_skb_flags |= NETLINK_SKB_DST; 1883 } else { 1884 dst_portid = nlk->dst_portid; 1885 dst_group = nlk->dst_group; 1886 } 1887 1888 if (!nlk->bound) { 1889 err = netlink_autobind(sock); 1890 if (err) 1891 goto out; 1892 } else { 1893 /* Ensure nlk is hashed and visible. */ 1894 smp_rmb(); 1895 } 1896 1897 err = -EMSGSIZE; 1898 if (len > sk->sk_sndbuf - 32) 1899 goto out; 1900 err = -ENOBUFS; 1901 skb = netlink_alloc_large_skb(len, dst_group); 1902 if (skb == NULL) 1903 goto out; 1904 1905 NETLINK_CB(skb).portid = nlk->portid; 1906 NETLINK_CB(skb).dst_group = dst_group; 1907 NETLINK_CB(skb).creds = scm.creds; 1908 NETLINK_CB(skb).flags = netlink_skb_flags; 1909 1910 err = -EFAULT; 1911 if (memcpy_from_msg(skb_put(skb, len), msg, len)) { 1912 kfree_skb(skb); 1913 goto out; 1914 } 1915 1916 err = security_netlink_send(sk, skb); 1917 if (err) { 1918 kfree_skb(skb); 1919 goto out; 1920 } 1921 1922 if (dst_group) { 1923 refcount_inc(&skb->users); 1924 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL); 1925 } 1926 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT); 1927 1928 out: 1929 scm_destroy(&scm); 1930 return err; 1931 } 1932 1933 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1934 int flags) 1935 { 1936 struct scm_cookie scm; 1937 struct sock *sk = sock->sk; 1938 struct netlink_sock *nlk = nlk_sk(sk); 1939 int noblock = flags&MSG_DONTWAIT; 1940 size_t copied; 1941 struct sk_buff *skb, *data_skb; 1942 int err, ret; 1943 1944 if (flags&MSG_OOB) 1945 return -EOPNOTSUPP; 1946 1947 copied = 0; 1948 1949 skb = skb_recv_datagram(sk, flags, noblock, &err); 1950 if (skb == NULL) 1951 goto out; 1952 1953 data_skb = skb; 1954 1955 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES 1956 if (unlikely(skb_shinfo(skb)->frag_list)) { 1957 /* 1958 * If this skb has a frag_list, then here that means that we 1959 * will have to use the frag_list skb's data for compat tasks 1960 * and the regular skb's data for normal (non-compat) tasks. 1961 * 1962 * If we need to send the compat skb, assign it to the 1963 * 'data_skb' variable so that it will be used below for data 1964 * copying. We keep 'skb' for everything else, including 1965 * freeing both later. 1966 */ 1967 if (flags & MSG_CMSG_COMPAT) 1968 data_skb = skb_shinfo(skb)->frag_list; 1969 } 1970 #endif 1971 1972 /* Record the max length of recvmsg() calls for future allocations */ 1973 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len); 1974 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len, 1975 SKB_WITH_OVERHEAD(32768)); 1976 1977 copied = data_skb->len; 1978 if (len < copied) { 1979 msg->msg_flags |= MSG_TRUNC; 1980 copied = len; 1981 } 1982 1983 skb_reset_transport_header(data_skb); 1984 err = skb_copy_datagram_msg(data_skb, 0, msg, copied); 1985 1986 if (msg->msg_name) { 1987 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name); 1988 addr->nl_family = AF_NETLINK; 1989 addr->nl_pad = 0; 1990 addr->nl_pid = NETLINK_CB(skb).portid; 1991 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group); 1992 msg->msg_namelen = sizeof(*addr); 1993 } 1994 1995 if (nlk->flags & NETLINK_F_RECV_PKTINFO) 1996 netlink_cmsg_recv_pktinfo(msg, skb); 1997 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID) 1998 netlink_cmsg_listen_all_nsid(sk, msg, skb); 1999 2000 memset(&scm, 0, sizeof(scm)); 2001 scm.creds = *NETLINK_CREDS(skb); 2002 if (flags & MSG_TRUNC) 2003 copied = data_skb->len; 2004 2005 skb_free_datagram(sk, skb); 2006 2007 if (nlk->cb_running && 2008 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) { 2009 ret = netlink_dump(sk); 2010 if (ret) { 2011 sk->sk_err = -ret; 2012 sk->sk_error_report(sk); 2013 } 2014 } 2015 2016 scm_recv(sock, msg, &scm, flags); 2017 out: 2018 netlink_rcv_wake(sk); 2019 return err ? : copied; 2020 } 2021 2022 static void netlink_data_ready(struct sock *sk) 2023 { 2024 BUG(); 2025 } 2026 2027 /* 2028 * We export these functions to other modules. They provide a 2029 * complete set of kernel non-blocking support for message 2030 * queueing. 2031 */ 2032 2033 struct sock * 2034 __netlink_kernel_create(struct net *net, int unit, struct module *module, 2035 struct netlink_kernel_cfg *cfg) 2036 { 2037 struct socket *sock; 2038 struct sock *sk; 2039 struct netlink_sock *nlk; 2040 struct listeners *listeners = NULL; 2041 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL; 2042 unsigned int groups; 2043 2044 BUG_ON(!nl_table); 2045 2046 if (unit < 0 || unit >= MAX_LINKS) 2047 return NULL; 2048 2049 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 2050 return NULL; 2051 2052 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0) 2053 goto out_sock_release_nosk; 2054 2055 sk = sock->sk; 2056 2057 if (!cfg || cfg->groups < 32) 2058 groups = 32; 2059 else 2060 groups = cfg->groups; 2061 2062 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 2063 if (!listeners) 2064 goto out_sock_release; 2065 2066 sk->sk_data_ready = netlink_data_ready; 2067 if (cfg && cfg->input) 2068 nlk_sk(sk)->netlink_rcv = cfg->input; 2069 2070 if (netlink_insert(sk, 0)) 2071 goto out_sock_release; 2072 2073 nlk = nlk_sk(sk); 2074 nlk->flags |= NETLINK_F_KERNEL_SOCKET; 2075 2076 netlink_table_grab(); 2077 if (!nl_table[unit].registered) { 2078 nl_table[unit].groups = groups; 2079 rcu_assign_pointer(nl_table[unit].listeners, listeners); 2080 nl_table[unit].cb_mutex = cb_mutex; 2081 nl_table[unit].module = module; 2082 if (cfg) { 2083 nl_table[unit].bind = cfg->bind; 2084 nl_table[unit].unbind = cfg->unbind; 2085 nl_table[unit].flags = cfg->flags; 2086 if (cfg->compare) 2087 nl_table[unit].compare = cfg->compare; 2088 } 2089 nl_table[unit].registered = 1; 2090 } else { 2091 kfree(listeners); 2092 nl_table[unit].registered++; 2093 } 2094 netlink_table_ungrab(); 2095 return sk; 2096 2097 out_sock_release: 2098 kfree(listeners); 2099 netlink_kernel_release(sk); 2100 return NULL; 2101 2102 out_sock_release_nosk: 2103 sock_release(sock); 2104 return NULL; 2105 } 2106 EXPORT_SYMBOL(__netlink_kernel_create); 2107 2108 void 2109 netlink_kernel_release(struct sock *sk) 2110 { 2111 if (sk == NULL || sk->sk_socket == NULL) 2112 return; 2113 2114 sock_release(sk->sk_socket); 2115 } 2116 EXPORT_SYMBOL(netlink_kernel_release); 2117 2118 int __netlink_change_ngroups(struct sock *sk, unsigned int groups) 2119 { 2120 struct listeners *new, *old; 2121 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 2122 2123 if (groups < 32) 2124 groups = 32; 2125 2126 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) { 2127 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC); 2128 if (!new) 2129 return -ENOMEM; 2130 old = nl_deref_protected(tbl->listeners); 2131 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups)); 2132 rcu_assign_pointer(tbl->listeners, new); 2133 2134 kfree_rcu(old, rcu); 2135 } 2136 tbl->groups = groups; 2137 2138 return 0; 2139 } 2140 2141 /** 2142 * netlink_change_ngroups - change number of multicast groups 2143 * 2144 * This changes the number of multicast groups that are available 2145 * on a certain netlink family. Note that it is not possible to 2146 * change the number of groups to below 32. Also note that it does 2147 * not implicitly call netlink_clear_multicast_users() when the 2148 * number of groups is reduced. 2149 * 2150 * @sk: The kernel netlink socket, as returned by netlink_kernel_create(). 2151 * @groups: The new number of groups. 2152 */ 2153 int netlink_change_ngroups(struct sock *sk, unsigned int groups) 2154 { 2155 int err; 2156 2157 netlink_table_grab(); 2158 err = __netlink_change_ngroups(sk, groups); 2159 netlink_table_ungrab(); 2160 2161 return err; 2162 } 2163 2164 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group) 2165 { 2166 struct sock *sk; 2167 struct netlink_table *tbl = &nl_table[ksk->sk_protocol]; 2168 2169 sk_for_each_bound(sk, &tbl->mc_list) 2170 netlink_update_socket_mc(nlk_sk(sk), group, 0); 2171 } 2172 2173 struct nlmsghdr * 2174 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags) 2175 { 2176 struct nlmsghdr *nlh; 2177 int size = nlmsg_msg_size(len); 2178 2179 nlh = skb_put(skb, NLMSG_ALIGN(size)); 2180 nlh->nlmsg_type = type; 2181 nlh->nlmsg_len = size; 2182 nlh->nlmsg_flags = flags; 2183 nlh->nlmsg_pid = portid; 2184 nlh->nlmsg_seq = seq; 2185 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0) 2186 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size); 2187 return nlh; 2188 } 2189 EXPORT_SYMBOL(__nlmsg_put); 2190 2191 /* 2192 * It looks a bit ugly. 2193 * It would be better to create kernel thread. 2194 */ 2195 2196 static int netlink_dump(struct sock *sk) 2197 { 2198 struct netlink_sock *nlk = nlk_sk(sk); 2199 struct netlink_ext_ack extack = {}; 2200 struct netlink_callback *cb; 2201 struct sk_buff *skb = NULL; 2202 struct nlmsghdr *nlh; 2203 struct module *module; 2204 int err = -ENOBUFS; 2205 int alloc_min_size; 2206 int alloc_size; 2207 2208 mutex_lock(nlk->cb_mutex); 2209 if (!nlk->cb_running) { 2210 err = -EINVAL; 2211 goto errout_skb; 2212 } 2213 2214 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2215 goto errout_skb; 2216 2217 /* NLMSG_GOODSIZE is small to avoid high order allocations being 2218 * required, but it makes sense to _attempt_ a 16K bytes allocation 2219 * to reduce number of system calls on dump operations, if user 2220 * ever provided a big enough buffer. 2221 */ 2222 cb = &nlk->cb; 2223 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE); 2224 2225 if (alloc_min_size < nlk->max_recvmsg_len) { 2226 alloc_size = nlk->max_recvmsg_len; 2227 skb = alloc_skb(alloc_size, 2228 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) | 2229 __GFP_NOWARN | __GFP_NORETRY); 2230 } 2231 if (!skb) { 2232 alloc_size = alloc_min_size; 2233 skb = alloc_skb(alloc_size, GFP_KERNEL); 2234 } 2235 if (!skb) 2236 goto errout_skb; 2237 2238 /* Trim skb to allocated size. User is expected to provide buffer as 2239 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at 2240 * netlink_recvmsg())). dump will pack as many smaller messages as 2241 * could fit within the allocated skb. skb is typically allocated 2242 * with larger space than required (could be as much as near 2x the 2243 * requested size with align to next power of 2 approach). Allowing 2244 * dump to use the excess space makes it difficult for a user to have a 2245 * reasonable static buffer based on the expected largest dump of a 2246 * single netdev. The outcome is MSG_TRUNC error. 2247 */ 2248 skb_reserve(skb, skb_tailroom(skb) - alloc_size); 2249 netlink_skb_set_owner_r(skb, sk); 2250 2251 if (nlk->dump_done_errno > 0) { 2252 cb->extack = &extack; 2253 nlk->dump_done_errno = cb->dump(skb, cb); 2254 cb->extack = NULL; 2255 } 2256 2257 if (nlk->dump_done_errno > 0 || 2258 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) { 2259 mutex_unlock(nlk->cb_mutex); 2260 2261 if (sk_filter(sk, skb)) 2262 kfree_skb(skb); 2263 else 2264 __netlink_sendskb(sk, skb); 2265 return 0; 2266 } 2267 2268 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, 2269 sizeof(nlk->dump_done_errno), 2270 NLM_F_MULTI | cb->answer_flags); 2271 if (WARN_ON(!nlh)) 2272 goto errout_skb; 2273 2274 nl_dump_check_consistent(cb, nlh); 2275 2276 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, 2277 sizeof(nlk->dump_done_errno)); 2278 2279 if (extack._msg && nlk->flags & NETLINK_F_EXT_ACK) { 2280 nlh->nlmsg_flags |= NLM_F_ACK_TLVS; 2281 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack._msg)) 2282 nlmsg_end(skb, nlh); 2283 } 2284 2285 if (sk_filter(sk, skb)) 2286 kfree_skb(skb); 2287 else 2288 __netlink_sendskb(sk, skb); 2289 2290 if (cb->done) 2291 cb->done(cb); 2292 2293 nlk->cb_running = false; 2294 module = cb->module; 2295 skb = cb->skb; 2296 mutex_unlock(nlk->cb_mutex); 2297 module_put(module); 2298 consume_skb(skb); 2299 return 0; 2300 2301 errout_skb: 2302 mutex_unlock(nlk->cb_mutex); 2303 kfree_skb(skb); 2304 return err; 2305 } 2306 2307 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 2308 const struct nlmsghdr *nlh, 2309 struct netlink_dump_control *control) 2310 { 2311 struct netlink_sock *nlk, *nlk2; 2312 struct netlink_callback *cb; 2313 struct sock *sk; 2314 int ret; 2315 2316 refcount_inc(&skb->users); 2317 2318 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid); 2319 if (sk == NULL) { 2320 ret = -ECONNREFUSED; 2321 goto error_free; 2322 } 2323 2324 nlk = nlk_sk(sk); 2325 mutex_lock(nlk->cb_mutex); 2326 /* A dump is in progress... */ 2327 if (nlk->cb_running) { 2328 ret = -EBUSY; 2329 goto error_unlock; 2330 } 2331 /* add reference of module which cb->dump belongs to */ 2332 if (!try_module_get(control->module)) { 2333 ret = -EPROTONOSUPPORT; 2334 goto error_unlock; 2335 } 2336 2337 cb = &nlk->cb; 2338 memset(cb, 0, sizeof(*cb)); 2339 cb->dump = control->dump; 2340 cb->done = control->done; 2341 cb->nlh = nlh; 2342 cb->data = control->data; 2343 cb->module = control->module; 2344 cb->min_dump_alloc = control->min_dump_alloc; 2345 cb->skb = skb; 2346 2347 nlk2 = nlk_sk(NETLINK_CB(skb).sk); 2348 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK); 2349 2350 if (control->start) { 2351 ret = control->start(cb); 2352 if (ret) 2353 goto error_put; 2354 } 2355 2356 nlk->cb_running = true; 2357 nlk->dump_done_errno = INT_MAX; 2358 2359 mutex_unlock(nlk->cb_mutex); 2360 2361 ret = netlink_dump(sk); 2362 2363 sock_put(sk); 2364 2365 if (ret) 2366 return ret; 2367 2368 /* We successfully started a dump, by returning -EINTR we 2369 * signal not to send ACK even if it was requested. 2370 */ 2371 return -EINTR; 2372 2373 error_put: 2374 module_put(control->module); 2375 error_unlock: 2376 sock_put(sk); 2377 mutex_unlock(nlk->cb_mutex); 2378 error_free: 2379 kfree_skb(skb); 2380 return ret; 2381 } 2382 EXPORT_SYMBOL(__netlink_dump_start); 2383 2384 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err, 2385 const struct netlink_ext_ack *extack) 2386 { 2387 struct sk_buff *skb; 2388 struct nlmsghdr *rep; 2389 struct nlmsgerr *errmsg; 2390 size_t payload = sizeof(*errmsg); 2391 size_t tlvlen = 0; 2392 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk); 2393 unsigned int flags = 0; 2394 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK; 2395 2396 /* Error messages get the original request appened, unless the user 2397 * requests to cap the error message, and get extra error data if 2398 * requested. 2399 */ 2400 if (nlk_has_extack && extack && extack->_msg) 2401 tlvlen += nla_total_size(strlen(extack->_msg) + 1); 2402 2403 if (err) { 2404 if (!(nlk->flags & NETLINK_F_CAP_ACK)) 2405 payload += nlmsg_len(nlh); 2406 else 2407 flags |= NLM_F_CAPPED; 2408 if (nlk_has_extack && extack && extack->bad_attr) 2409 tlvlen += nla_total_size(sizeof(u32)); 2410 } else { 2411 flags |= NLM_F_CAPPED; 2412 2413 if (nlk_has_extack && extack && extack->cookie_len) 2414 tlvlen += nla_total_size(extack->cookie_len); 2415 } 2416 2417 if (tlvlen) 2418 flags |= NLM_F_ACK_TLVS; 2419 2420 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL); 2421 if (!skb) { 2422 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS; 2423 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk); 2424 return; 2425 } 2426 2427 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, 2428 NLMSG_ERROR, payload, flags); 2429 errmsg = nlmsg_data(rep); 2430 errmsg->error = err; 2431 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh)); 2432 2433 if (nlk_has_extack && extack) { 2434 if (extack->_msg) { 2435 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, 2436 extack->_msg)); 2437 } 2438 if (err) { 2439 if (extack->bad_attr && 2440 !WARN_ON((u8 *)extack->bad_attr < in_skb->data || 2441 (u8 *)extack->bad_attr >= in_skb->data + 2442 in_skb->len)) 2443 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS, 2444 (u8 *)extack->bad_attr - 2445 in_skb->data)); 2446 } else { 2447 if (extack->cookie_len) 2448 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE, 2449 extack->cookie_len, 2450 extack->cookie)); 2451 } 2452 } 2453 2454 nlmsg_end(skb, rep); 2455 2456 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT); 2457 } 2458 EXPORT_SYMBOL(netlink_ack); 2459 2460 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *, 2461 struct nlmsghdr *, 2462 struct netlink_ext_ack *)) 2463 { 2464 struct netlink_ext_ack extack; 2465 struct nlmsghdr *nlh; 2466 int err; 2467 2468 while (skb->len >= nlmsg_total_size(0)) { 2469 int msglen; 2470 2471 memset(&extack, 0, sizeof(extack)); 2472 nlh = nlmsg_hdr(skb); 2473 err = 0; 2474 2475 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len) 2476 return 0; 2477 2478 /* Only requests are handled by the kernel */ 2479 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) 2480 goto ack; 2481 2482 /* Skip control messages */ 2483 if (nlh->nlmsg_type < NLMSG_MIN_TYPE) 2484 goto ack; 2485 2486 err = cb(skb, nlh, &extack); 2487 if (err == -EINTR) 2488 goto skip; 2489 2490 ack: 2491 if (nlh->nlmsg_flags & NLM_F_ACK || err) 2492 netlink_ack(skb, nlh, err, &extack); 2493 2494 skip: 2495 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 2496 if (msglen > skb->len) 2497 msglen = skb->len; 2498 skb_pull(skb, msglen); 2499 } 2500 2501 return 0; 2502 } 2503 EXPORT_SYMBOL(netlink_rcv_skb); 2504 2505 /** 2506 * nlmsg_notify - send a notification netlink message 2507 * @sk: netlink socket to use 2508 * @skb: notification message 2509 * @portid: destination netlink portid for reports or 0 2510 * @group: destination multicast group or 0 2511 * @report: 1 to report back, 0 to disable 2512 * @flags: allocation flags 2513 */ 2514 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid, 2515 unsigned int group, int report, gfp_t flags) 2516 { 2517 int err = 0; 2518 2519 if (group) { 2520 int exclude_portid = 0; 2521 2522 if (report) { 2523 refcount_inc(&skb->users); 2524 exclude_portid = portid; 2525 } 2526 2527 /* errors reported via destination sk->sk_err, but propagate 2528 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */ 2529 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags); 2530 } 2531 2532 if (report) { 2533 int err2; 2534 2535 err2 = nlmsg_unicast(sk, skb, portid); 2536 if (!err || err == -ESRCH) 2537 err = err2; 2538 } 2539 2540 return err; 2541 } 2542 EXPORT_SYMBOL(nlmsg_notify); 2543 2544 #ifdef CONFIG_PROC_FS 2545 struct nl_seq_iter { 2546 struct seq_net_private p; 2547 struct rhashtable_iter hti; 2548 int link; 2549 }; 2550 2551 static int netlink_walk_start(struct nl_seq_iter *iter) 2552 { 2553 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti); 2554 rhashtable_walk_start(&iter->hti); 2555 2556 return 0; 2557 } 2558 2559 static void netlink_walk_stop(struct nl_seq_iter *iter) 2560 { 2561 rhashtable_walk_stop(&iter->hti); 2562 rhashtable_walk_exit(&iter->hti); 2563 } 2564 2565 static void *__netlink_seq_next(struct seq_file *seq) 2566 { 2567 struct nl_seq_iter *iter = seq->private; 2568 struct netlink_sock *nlk; 2569 2570 do { 2571 for (;;) { 2572 int err; 2573 2574 nlk = rhashtable_walk_next(&iter->hti); 2575 2576 if (IS_ERR(nlk)) { 2577 if (PTR_ERR(nlk) == -EAGAIN) 2578 continue; 2579 2580 return nlk; 2581 } 2582 2583 if (nlk) 2584 break; 2585 2586 netlink_walk_stop(iter); 2587 if (++iter->link >= MAX_LINKS) 2588 return NULL; 2589 2590 err = netlink_walk_start(iter); 2591 if (err) 2592 return ERR_PTR(err); 2593 } 2594 } while (sock_net(&nlk->sk) != seq_file_net(seq)); 2595 2596 return nlk; 2597 } 2598 2599 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp) 2600 { 2601 struct nl_seq_iter *iter = seq->private; 2602 void *obj = SEQ_START_TOKEN; 2603 loff_t pos; 2604 int err; 2605 2606 iter->link = 0; 2607 2608 err = netlink_walk_start(iter); 2609 if (err) 2610 return ERR_PTR(err); 2611 2612 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--) 2613 obj = __netlink_seq_next(seq); 2614 2615 return obj; 2616 } 2617 2618 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2619 { 2620 ++*pos; 2621 return __netlink_seq_next(seq); 2622 } 2623 2624 static void netlink_seq_stop(struct seq_file *seq, void *v) 2625 { 2626 struct nl_seq_iter *iter = seq->private; 2627 2628 if (iter->link >= MAX_LINKS) 2629 return; 2630 2631 netlink_walk_stop(iter); 2632 } 2633 2634 2635 static int netlink_seq_show(struct seq_file *seq, void *v) 2636 { 2637 if (v == SEQ_START_TOKEN) { 2638 seq_puts(seq, 2639 "sk Eth Pid Groups " 2640 "Rmem Wmem Dump Locks Drops Inode\n"); 2641 } else { 2642 struct sock *s = v; 2643 struct netlink_sock *nlk = nlk_sk(s); 2644 2645 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n", 2646 s, 2647 s->sk_protocol, 2648 nlk->portid, 2649 nlk->groups ? (u32)nlk->groups[0] : 0, 2650 sk_rmem_alloc_get(s), 2651 sk_wmem_alloc_get(s), 2652 nlk->cb_running, 2653 refcount_read(&s->sk_refcnt), 2654 atomic_read(&s->sk_drops), 2655 sock_i_ino(s) 2656 ); 2657 2658 } 2659 return 0; 2660 } 2661 2662 static const struct seq_operations netlink_seq_ops = { 2663 .start = netlink_seq_start, 2664 .next = netlink_seq_next, 2665 .stop = netlink_seq_stop, 2666 .show = netlink_seq_show, 2667 }; 2668 #endif 2669 2670 int netlink_register_notifier(struct notifier_block *nb) 2671 { 2672 return blocking_notifier_chain_register(&netlink_chain, nb); 2673 } 2674 EXPORT_SYMBOL(netlink_register_notifier); 2675 2676 int netlink_unregister_notifier(struct notifier_block *nb) 2677 { 2678 return blocking_notifier_chain_unregister(&netlink_chain, nb); 2679 } 2680 EXPORT_SYMBOL(netlink_unregister_notifier); 2681 2682 static const struct proto_ops netlink_ops = { 2683 .family = PF_NETLINK, 2684 .owner = THIS_MODULE, 2685 .release = netlink_release, 2686 .bind = netlink_bind, 2687 .connect = netlink_connect, 2688 .socketpair = sock_no_socketpair, 2689 .accept = sock_no_accept, 2690 .getname = netlink_getname, 2691 .poll = datagram_poll, 2692 .ioctl = netlink_ioctl, 2693 .listen = sock_no_listen, 2694 .shutdown = sock_no_shutdown, 2695 .setsockopt = netlink_setsockopt, 2696 .getsockopt = netlink_getsockopt, 2697 .sendmsg = netlink_sendmsg, 2698 .recvmsg = netlink_recvmsg, 2699 .mmap = sock_no_mmap, 2700 .sendpage = sock_no_sendpage, 2701 }; 2702 2703 static const struct net_proto_family netlink_family_ops = { 2704 .family = PF_NETLINK, 2705 .create = netlink_create, 2706 .owner = THIS_MODULE, /* for consistency 8) */ 2707 }; 2708 2709 static int __net_init netlink_net_init(struct net *net) 2710 { 2711 #ifdef CONFIG_PROC_FS 2712 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops, 2713 sizeof(struct nl_seq_iter))) 2714 return -ENOMEM; 2715 #endif 2716 return 0; 2717 } 2718 2719 static void __net_exit netlink_net_exit(struct net *net) 2720 { 2721 #ifdef CONFIG_PROC_FS 2722 remove_proc_entry("netlink", net->proc_net); 2723 #endif 2724 } 2725 2726 static void __init netlink_add_usersock_entry(void) 2727 { 2728 struct listeners *listeners; 2729 int groups = 32; 2730 2731 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 2732 if (!listeners) 2733 panic("netlink_add_usersock_entry: Cannot allocate listeners\n"); 2734 2735 netlink_table_grab(); 2736 2737 nl_table[NETLINK_USERSOCK].groups = groups; 2738 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners); 2739 nl_table[NETLINK_USERSOCK].module = THIS_MODULE; 2740 nl_table[NETLINK_USERSOCK].registered = 1; 2741 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND; 2742 2743 netlink_table_ungrab(); 2744 } 2745 2746 static struct pernet_operations __net_initdata netlink_net_ops = { 2747 .init = netlink_net_init, 2748 .exit = netlink_net_exit, 2749 }; 2750 2751 static inline u32 netlink_hash(const void *data, u32 len, u32 seed) 2752 { 2753 const struct netlink_sock *nlk = data; 2754 struct netlink_compare_arg arg; 2755 2756 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid); 2757 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed); 2758 } 2759 2760 static const struct rhashtable_params netlink_rhashtable_params = { 2761 .head_offset = offsetof(struct netlink_sock, node), 2762 .key_len = netlink_compare_arg_len, 2763 .obj_hashfn = netlink_hash, 2764 .obj_cmpfn = netlink_compare, 2765 .automatic_shrinking = true, 2766 }; 2767 2768 static int __init netlink_proto_init(void) 2769 { 2770 int i; 2771 int err = proto_register(&netlink_proto, 0); 2772 2773 if (err != 0) 2774 goto out; 2775 2776 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb)); 2777 2778 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL); 2779 if (!nl_table) 2780 goto panic; 2781 2782 for (i = 0; i < MAX_LINKS; i++) { 2783 if (rhashtable_init(&nl_table[i].hash, 2784 &netlink_rhashtable_params) < 0) { 2785 while (--i > 0) 2786 rhashtable_destroy(&nl_table[i].hash); 2787 kfree(nl_table); 2788 goto panic; 2789 } 2790 } 2791 2792 netlink_add_usersock_entry(); 2793 2794 sock_register(&netlink_family_ops); 2795 register_pernet_subsys(&netlink_net_ops); 2796 register_pernet_subsys(&netlink_tap_net_ops); 2797 /* The netlink device handler may be needed early. */ 2798 rtnetlink_init(); 2799 out: 2800 return err; 2801 panic: 2802 panic("netlink_init: Cannot allocate nl_table\n"); 2803 } 2804 2805 core_initcall(netlink_proto_init); 2806