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 = nladdr->nl_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 1000 /* Only superuser is allowed to listen multicasts */ 1001 if (groups) { 1002 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV)) 1003 return -EPERM; 1004 err = netlink_realloc_groups(sk); 1005 if (err) 1006 return err; 1007 } 1008 1009 if (nlk->ngroups < BITS_PER_LONG) 1010 groups &= (1UL << nlk->ngroups) - 1; 1011 1012 bound = nlk->bound; 1013 if (bound) { 1014 /* Ensure nlk->portid is up-to-date. */ 1015 smp_rmb(); 1016 1017 if (nladdr->nl_pid != nlk->portid) 1018 return -EINVAL; 1019 } 1020 1021 netlink_lock_table(); 1022 if (nlk->netlink_bind && groups) { 1023 int group; 1024 1025 for (group = 0; group < nlk->ngroups; group++) { 1026 if (!test_bit(group, &groups)) 1027 continue; 1028 err = nlk->netlink_bind(net, group + 1); 1029 if (!err) 1030 continue; 1031 netlink_undo_bind(group, groups, sk); 1032 goto unlock; 1033 } 1034 } 1035 1036 /* No need for barriers here as we return to user-space without 1037 * using any of the bound attributes. 1038 */ 1039 if (!bound) { 1040 err = nladdr->nl_pid ? 1041 netlink_insert(sk, nladdr->nl_pid) : 1042 netlink_autobind(sock); 1043 if (err) { 1044 netlink_undo_bind(nlk->ngroups, groups, sk); 1045 goto unlock; 1046 } 1047 } 1048 1049 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0])) 1050 goto unlock; 1051 netlink_unlock_table(); 1052 1053 netlink_table_grab(); 1054 netlink_update_subscriptions(sk, nlk->subscriptions + 1055 hweight32(groups) - 1056 hweight32(nlk->groups[0])); 1057 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups; 1058 netlink_update_listeners(sk); 1059 netlink_table_ungrab(); 1060 1061 return 0; 1062 1063 unlock: 1064 netlink_unlock_table(); 1065 return err; 1066 } 1067 1068 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 1069 int alen, int flags) 1070 { 1071 int err = 0; 1072 struct sock *sk = sock->sk; 1073 struct netlink_sock *nlk = nlk_sk(sk); 1074 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 1075 1076 if (alen < sizeof(addr->sa_family)) 1077 return -EINVAL; 1078 1079 if (addr->sa_family == AF_UNSPEC) { 1080 sk->sk_state = NETLINK_UNCONNECTED; 1081 nlk->dst_portid = 0; 1082 nlk->dst_group = 0; 1083 return 0; 1084 } 1085 if (addr->sa_family != AF_NETLINK) 1086 return -EINVAL; 1087 1088 if (alen < sizeof(struct sockaddr_nl)) 1089 return -EINVAL; 1090 1091 if ((nladdr->nl_groups || nladdr->nl_pid) && 1092 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND)) 1093 return -EPERM; 1094 1095 /* No need for barriers here as we return to user-space without 1096 * using any of the bound attributes. 1097 */ 1098 if (!nlk->bound) 1099 err = netlink_autobind(sock); 1100 1101 if (err == 0) { 1102 sk->sk_state = NETLINK_CONNECTED; 1103 nlk->dst_portid = nladdr->nl_pid; 1104 nlk->dst_group = ffs(nladdr->nl_groups); 1105 } 1106 1107 return err; 1108 } 1109 1110 static int netlink_getname(struct socket *sock, struct sockaddr *addr, 1111 int peer) 1112 { 1113 struct sock *sk = sock->sk; 1114 struct netlink_sock *nlk = nlk_sk(sk); 1115 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr); 1116 1117 nladdr->nl_family = AF_NETLINK; 1118 nladdr->nl_pad = 0; 1119 1120 if (peer) { 1121 nladdr->nl_pid = nlk->dst_portid; 1122 nladdr->nl_groups = netlink_group_mask(nlk->dst_group); 1123 } else { 1124 nladdr->nl_pid = nlk->portid; 1125 netlink_lock_table(); 1126 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0; 1127 netlink_unlock_table(); 1128 } 1129 return sizeof(*nladdr); 1130 } 1131 1132 static int netlink_ioctl(struct socket *sock, unsigned int cmd, 1133 unsigned long arg) 1134 { 1135 /* try to hand this ioctl down to the NIC drivers. 1136 */ 1137 return -ENOIOCTLCMD; 1138 } 1139 1140 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid) 1141 { 1142 struct sock *sock; 1143 struct netlink_sock *nlk; 1144 1145 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid); 1146 if (!sock) 1147 return ERR_PTR(-ECONNREFUSED); 1148 1149 /* Don't bother queuing skb if kernel socket has no input function */ 1150 nlk = nlk_sk(sock); 1151 if (sock->sk_state == NETLINK_CONNECTED && 1152 nlk->dst_portid != nlk_sk(ssk)->portid) { 1153 sock_put(sock); 1154 return ERR_PTR(-ECONNREFUSED); 1155 } 1156 return sock; 1157 } 1158 1159 struct sock *netlink_getsockbyfilp(struct file *filp) 1160 { 1161 struct inode *inode = file_inode(filp); 1162 struct sock *sock; 1163 1164 if (!S_ISSOCK(inode->i_mode)) 1165 return ERR_PTR(-ENOTSOCK); 1166 1167 sock = SOCKET_I(inode)->sk; 1168 if (sock->sk_family != AF_NETLINK) 1169 return ERR_PTR(-EINVAL); 1170 1171 sock_hold(sock); 1172 return sock; 1173 } 1174 1175 static struct sk_buff *netlink_alloc_large_skb(unsigned int size, 1176 int broadcast) 1177 { 1178 struct sk_buff *skb; 1179 void *data; 1180 1181 if (size <= NLMSG_GOODSIZE || broadcast) 1182 return alloc_skb(size, GFP_KERNEL); 1183 1184 size = SKB_DATA_ALIGN(size) + 1185 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 1186 1187 data = vmalloc(size); 1188 if (data == NULL) 1189 return NULL; 1190 1191 skb = __build_skb(data, size); 1192 if (skb == NULL) 1193 vfree(data); 1194 else 1195 skb->destructor = netlink_skb_destructor; 1196 1197 return skb; 1198 } 1199 1200 /* 1201 * Attach a skb to a netlink socket. 1202 * The caller must hold a reference to the destination socket. On error, the 1203 * reference is dropped. The skb is not send to the destination, just all 1204 * all error checks are performed and memory in the queue is reserved. 1205 * Return values: 1206 * < 0: error. skb freed, reference to sock dropped. 1207 * 0: continue 1208 * 1: repeat lookup - reference dropped while waiting for socket memory. 1209 */ 1210 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, 1211 long *timeo, struct sock *ssk) 1212 { 1213 struct netlink_sock *nlk; 1214 1215 nlk = nlk_sk(sk); 1216 1217 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 1218 test_bit(NETLINK_S_CONGESTED, &nlk->state))) { 1219 DECLARE_WAITQUEUE(wait, current); 1220 if (!*timeo) { 1221 if (!ssk || netlink_is_kernel(ssk)) 1222 netlink_overrun(sk); 1223 sock_put(sk); 1224 kfree_skb(skb); 1225 return -EAGAIN; 1226 } 1227 1228 __set_current_state(TASK_INTERRUPTIBLE); 1229 add_wait_queue(&nlk->wait, &wait); 1230 1231 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 1232 test_bit(NETLINK_S_CONGESTED, &nlk->state)) && 1233 !sock_flag(sk, SOCK_DEAD)) 1234 *timeo = schedule_timeout(*timeo); 1235 1236 __set_current_state(TASK_RUNNING); 1237 remove_wait_queue(&nlk->wait, &wait); 1238 sock_put(sk); 1239 1240 if (signal_pending(current)) { 1241 kfree_skb(skb); 1242 return sock_intr_errno(*timeo); 1243 } 1244 return 1; 1245 } 1246 netlink_skb_set_owner_r(skb, sk); 1247 return 0; 1248 } 1249 1250 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb) 1251 { 1252 int len = skb->len; 1253 1254 netlink_deliver_tap(sock_net(sk), skb); 1255 1256 skb_queue_tail(&sk->sk_receive_queue, skb); 1257 sk->sk_data_ready(sk); 1258 return len; 1259 } 1260 1261 int netlink_sendskb(struct sock *sk, struct sk_buff *skb) 1262 { 1263 int len = __netlink_sendskb(sk, skb); 1264 1265 sock_put(sk); 1266 return len; 1267 } 1268 1269 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 1270 { 1271 kfree_skb(skb); 1272 sock_put(sk); 1273 } 1274 1275 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation) 1276 { 1277 int delta; 1278 1279 WARN_ON(skb->sk != NULL); 1280 delta = skb->end - skb->tail; 1281 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize) 1282 return skb; 1283 1284 if (skb_shared(skb)) { 1285 struct sk_buff *nskb = skb_clone(skb, allocation); 1286 if (!nskb) 1287 return skb; 1288 consume_skb(skb); 1289 skb = nskb; 1290 } 1291 1292 pskb_expand_head(skb, 0, -delta, 1293 (allocation & ~__GFP_DIRECT_RECLAIM) | 1294 __GFP_NOWARN | __GFP_NORETRY); 1295 return skb; 1296 } 1297 1298 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb, 1299 struct sock *ssk) 1300 { 1301 int ret; 1302 struct netlink_sock *nlk = nlk_sk(sk); 1303 1304 ret = -ECONNREFUSED; 1305 if (nlk->netlink_rcv != NULL) { 1306 ret = skb->len; 1307 netlink_skb_set_owner_r(skb, sk); 1308 NETLINK_CB(skb).sk = ssk; 1309 netlink_deliver_tap_kernel(sk, ssk, skb); 1310 nlk->netlink_rcv(skb); 1311 consume_skb(skb); 1312 } else { 1313 kfree_skb(skb); 1314 } 1315 sock_put(sk); 1316 return ret; 1317 } 1318 1319 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, 1320 u32 portid, int nonblock) 1321 { 1322 struct sock *sk; 1323 int err; 1324 long timeo; 1325 1326 skb = netlink_trim(skb, gfp_any()); 1327 1328 timeo = sock_sndtimeo(ssk, nonblock); 1329 retry: 1330 sk = netlink_getsockbyportid(ssk, portid); 1331 if (IS_ERR(sk)) { 1332 kfree_skb(skb); 1333 return PTR_ERR(sk); 1334 } 1335 if (netlink_is_kernel(sk)) 1336 return netlink_unicast_kernel(sk, skb, ssk); 1337 1338 if (sk_filter(sk, skb)) { 1339 err = skb->len; 1340 kfree_skb(skb); 1341 sock_put(sk); 1342 return err; 1343 } 1344 1345 err = netlink_attachskb(sk, skb, &timeo, ssk); 1346 if (err == 1) 1347 goto retry; 1348 if (err) 1349 return err; 1350 1351 return netlink_sendskb(sk, skb); 1352 } 1353 EXPORT_SYMBOL(netlink_unicast); 1354 1355 int netlink_has_listeners(struct sock *sk, unsigned int group) 1356 { 1357 int res = 0; 1358 struct listeners *listeners; 1359 1360 BUG_ON(!netlink_is_kernel(sk)); 1361 1362 rcu_read_lock(); 1363 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners); 1364 1365 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups) 1366 res = test_bit(group - 1, listeners->masks); 1367 1368 rcu_read_unlock(); 1369 1370 return res; 1371 } 1372 EXPORT_SYMBOL_GPL(netlink_has_listeners); 1373 1374 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 1375 { 1376 struct netlink_sock *nlk = nlk_sk(sk); 1377 1378 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 1379 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) { 1380 netlink_skb_set_owner_r(skb, sk); 1381 __netlink_sendskb(sk, skb); 1382 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1); 1383 } 1384 return -1; 1385 } 1386 1387 struct netlink_broadcast_data { 1388 struct sock *exclude_sk; 1389 struct net *net; 1390 u32 portid; 1391 u32 group; 1392 int failure; 1393 int delivery_failure; 1394 int congested; 1395 int delivered; 1396 gfp_t allocation; 1397 struct sk_buff *skb, *skb2; 1398 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data); 1399 void *tx_data; 1400 }; 1401 1402 static void do_one_broadcast(struct sock *sk, 1403 struct netlink_broadcast_data *p) 1404 { 1405 struct netlink_sock *nlk = nlk_sk(sk); 1406 int val; 1407 1408 if (p->exclude_sk == sk) 1409 return; 1410 1411 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 1412 !test_bit(p->group - 1, nlk->groups)) 1413 return; 1414 1415 if (!net_eq(sock_net(sk), p->net)) { 1416 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID)) 1417 return; 1418 1419 if (!peernet_has_id(sock_net(sk), p->net)) 1420 return; 1421 1422 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns, 1423 CAP_NET_BROADCAST)) 1424 return; 1425 } 1426 1427 if (p->failure) { 1428 netlink_overrun(sk); 1429 return; 1430 } 1431 1432 sock_hold(sk); 1433 if (p->skb2 == NULL) { 1434 if (skb_shared(p->skb)) { 1435 p->skb2 = skb_clone(p->skb, p->allocation); 1436 } else { 1437 p->skb2 = skb_get(p->skb); 1438 /* 1439 * skb ownership may have been set when 1440 * delivered to a previous socket. 1441 */ 1442 skb_orphan(p->skb2); 1443 } 1444 } 1445 if (p->skb2 == NULL) { 1446 netlink_overrun(sk); 1447 /* Clone failed. Notify ALL listeners. */ 1448 p->failure = 1; 1449 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR) 1450 p->delivery_failure = 1; 1451 goto out; 1452 } 1453 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) { 1454 kfree_skb(p->skb2); 1455 p->skb2 = NULL; 1456 goto out; 1457 } 1458 if (sk_filter(sk, p->skb2)) { 1459 kfree_skb(p->skb2); 1460 p->skb2 = NULL; 1461 goto out; 1462 } 1463 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net); 1464 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED) 1465 NETLINK_CB(p->skb2).nsid_is_set = true; 1466 val = netlink_broadcast_deliver(sk, p->skb2); 1467 if (val < 0) { 1468 netlink_overrun(sk); 1469 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR) 1470 p->delivery_failure = 1; 1471 } else { 1472 p->congested |= val; 1473 p->delivered = 1; 1474 p->skb2 = NULL; 1475 } 1476 out: 1477 sock_put(sk); 1478 } 1479 1480 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid, 1481 u32 group, gfp_t allocation, 1482 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data), 1483 void *filter_data) 1484 { 1485 struct net *net = sock_net(ssk); 1486 struct netlink_broadcast_data info; 1487 struct sock *sk; 1488 1489 skb = netlink_trim(skb, allocation); 1490 1491 info.exclude_sk = ssk; 1492 info.net = net; 1493 info.portid = portid; 1494 info.group = group; 1495 info.failure = 0; 1496 info.delivery_failure = 0; 1497 info.congested = 0; 1498 info.delivered = 0; 1499 info.allocation = allocation; 1500 info.skb = skb; 1501 info.skb2 = NULL; 1502 info.tx_filter = filter; 1503 info.tx_data = filter_data; 1504 1505 /* While we sleep in clone, do not allow to change socket list */ 1506 1507 netlink_lock_table(); 1508 1509 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) 1510 do_one_broadcast(sk, &info); 1511 1512 consume_skb(skb); 1513 1514 netlink_unlock_table(); 1515 1516 if (info.delivery_failure) { 1517 kfree_skb(info.skb2); 1518 return -ENOBUFS; 1519 } 1520 consume_skb(info.skb2); 1521 1522 if (info.delivered) { 1523 if (info.congested && gfpflags_allow_blocking(allocation)) 1524 yield(); 1525 return 0; 1526 } 1527 return -ESRCH; 1528 } 1529 EXPORT_SYMBOL(netlink_broadcast_filtered); 1530 1531 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid, 1532 u32 group, gfp_t allocation) 1533 { 1534 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation, 1535 NULL, NULL); 1536 } 1537 EXPORT_SYMBOL(netlink_broadcast); 1538 1539 struct netlink_set_err_data { 1540 struct sock *exclude_sk; 1541 u32 portid; 1542 u32 group; 1543 int code; 1544 }; 1545 1546 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p) 1547 { 1548 struct netlink_sock *nlk = nlk_sk(sk); 1549 int ret = 0; 1550 1551 if (sk == p->exclude_sk) 1552 goto out; 1553 1554 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk))) 1555 goto out; 1556 1557 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 1558 !test_bit(p->group - 1, nlk->groups)) 1559 goto out; 1560 1561 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) { 1562 ret = 1; 1563 goto out; 1564 } 1565 1566 sk->sk_err = p->code; 1567 sk->sk_error_report(sk); 1568 out: 1569 return ret; 1570 } 1571 1572 /** 1573 * netlink_set_err - report error to broadcast listeners 1574 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create() 1575 * @portid: the PORTID of a process that we want to skip (if any) 1576 * @group: the broadcast group that will notice the error 1577 * @code: error code, must be negative (as usual in kernelspace) 1578 * 1579 * This function returns the number of broadcast listeners that have set the 1580 * NETLINK_NO_ENOBUFS socket option. 1581 */ 1582 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code) 1583 { 1584 struct netlink_set_err_data info; 1585 struct sock *sk; 1586 int ret = 0; 1587 1588 info.exclude_sk = ssk; 1589 info.portid = portid; 1590 info.group = group; 1591 /* sk->sk_err wants a positive error value */ 1592 info.code = -code; 1593 1594 read_lock(&nl_table_lock); 1595 1596 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) 1597 ret += do_one_set_err(sk, &info); 1598 1599 read_unlock(&nl_table_lock); 1600 return ret; 1601 } 1602 EXPORT_SYMBOL(netlink_set_err); 1603 1604 /* must be called with netlink table grabbed */ 1605 static void netlink_update_socket_mc(struct netlink_sock *nlk, 1606 unsigned int group, 1607 int is_new) 1608 { 1609 int old, new = !!is_new, subscriptions; 1610 1611 old = test_bit(group - 1, nlk->groups); 1612 subscriptions = nlk->subscriptions - old + new; 1613 if (new) 1614 __set_bit(group - 1, nlk->groups); 1615 else 1616 __clear_bit(group - 1, nlk->groups); 1617 netlink_update_subscriptions(&nlk->sk, subscriptions); 1618 netlink_update_listeners(&nlk->sk); 1619 } 1620 1621 static int netlink_setsockopt(struct socket *sock, int level, int optname, 1622 char __user *optval, unsigned int optlen) 1623 { 1624 struct sock *sk = sock->sk; 1625 struct netlink_sock *nlk = nlk_sk(sk); 1626 unsigned int val = 0; 1627 int err; 1628 1629 if (level != SOL_NETLINK) 1630 return -ENOPROTOOPT; 1631 1632 if (optlen >= sizeof(int) && 1633 get_user(val, (unsigned int __user *)optval)) 1634 return -EFAULT; 1635 1636 switch (optname) { 1637 case NETLINK_PKTINFO: 1638 if (val) 1639 nlk->flags |= NETLINK_F_RECV_PKTINFO; 1640 else 1641 nlk->flags &= ~NETLINK_F_RECV_PKTINFO; 1642 err = 0; 1643 break; 1644 case NETLINK_ADD_MEMBERSHIP: 1645 case NETLINK_DROP_MEMBERSHIP: { 1646 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV)) 1647 return -EPERM; 1648 err = netlink_realloc_groups(sk); 1649 if (err) 1650 return err; 1651 if (!val || val - 1 >= nlk->ngroups) 1652 return -EINVAL; 1653 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) { 1654 err = nlk->netlink_bind(sock_net(sk), val); 1655 if (err) 1656 return err; 1657 } 1658 netlink_table_grab(); 1659 netlink_update_socket_mc(nlk, val, 1660 optname == NETLINK_ADD_MEMBERSHIP); 1661 netlink_table_ungrab(); 1662 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind) 1663 nlk->netlink_unbind(sock_net(sk), val); 1664 1665 err = 0; 1666 break; 1667 } 1668 case NETLINK_BROADCAST_ERROR: 1669 if (val) 1670 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR; 1671 else 1672 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR; 1673 err = 0; 1674 break; 1675 case NETLINK_NO_ENOBUFS: 1676 if (val) { 1677 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS; 1678 clear_bit(NETLINK_S_CONGESTED, &nlk->state); 1679 wake_up_interruptible(&nlk->wait); 1680 } else { 1681 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS; 1682 } 1683 err = 0; 1684 break; 1685 case NETLINK_LISTEN_ALL_NSID: 1686 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST)) 1687 return -EPERM; 1688 1689 if (val) 1690 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID; 1691 else 1692 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID; 1693 err = 0; 1694 break; 1695 case NETLINK_CAP_ACK: 1696 if (val) 1697 nlk->flags |= NETLINK_F_CAP_ACK; 1698 else 1699 nlk->flags &= ~NETLINK_F_CAP_ACK; 1700 err = 0; 1701 break; 1702 case NETLINK_EXT_ACK: 1703 if (val) 1704 nlk->flags |= NETLINK_F_EXT_ACK; 1705 else 1706 nlk->flags &= ~NETLINK_F_EXT_ACK; 1707 err = 0; 1708 break; 1709 default: 1710 err = -ENOPROTOOPT; 1711 } 1712 return err; 1713 } 1714 1715 static int netlink_getsockopt(struct socket *sock, int level, int optname, 1716 char __user *optval, int __user *optlen) 1717 { 1718 struct sock *sk = sock->sk; 1719 struct netlink_sock *nlk = nlk_sk(sk); 1720 int len, val, err; 1721 1722 if (level != SOL_NETLINK) 1723 return -ENOPROTOOPT; 1724 1725 if (get_user(len, optlen)) 1726 return -EFAULT; 1727 if (len < 0) 1728 return -EINVAL; 1729 1730 switch (optname) { 1731 case NETLINK_PKTINFO: 1732 if (len < sizeof(int)) 1733 return -EINVAL; 1734 len = sizeof(int); 1735 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0; 1736 if (put_user(len, optlen) || 1737 put_user(val, optval)) 1738 return -EFAULT; 1739 err = 0; 1740 break; 1741 case NETLINK_BROADCAST_ERROR: 1742 if (len < sizeof(int)) 1743 return -EINVAL; 1744 len = sizeof(int); 1745 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0; 1746 if (put_user(len, optlen) || 1747 put_user(val, optval)) 1748 return -EFAULT; 1749 err = 0; 1750 break; 1751 case NETLINK_NO_ENOBUFS: 1752 if (len < sizeof(int)) 1753 return -EINVAL; 1754 len = sizeof(int); 1755 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0; 1756 if (put_user(len, optlen) || 1757 put_user(val, optval)) 1758 return -EFAULT; 1759 err = 0; 1760 break; 1761 case NETLINK_LIST_MEMBERSHIPS: { 1762 int pos, idx, shift; 1763 1764 err = 0; 1765 netlink_lock_table(); 1766 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) { 1767 if (len - pos < sizeof(u32)) 1768 break; 1769 1770 idx = pos / sizeof(unsigned long); 1771 shift = (pos % sizeof(unsigned long)) * 8; 1772 if (put_user((u32)(nlk->groups[idx] >> shift), 1773 (u32 __user *)(optval + pos))) { 1774 err = -EFAULT; 1775 break; 1776 } 1777 } 1778 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen)) 1779 err = -EFAULT; 1780 netlink_unlock_table(); 1781 break; 1782 } 1783 case NETLINK_CAP_ACK: 1784 if (len < sizeof(int)) 1785 return -EINVAL; 1786 len = sizeof(int); 1787 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0; 1788 if (put_user(len, optlen) || 1789 put_user(val, optval)) 1790 return -EFAULT; 1791 err = 0; 1792 break; 1793 case NETLINK_EXT_ACK: 1794 if (len < sizeof(int)) 1795 return -EINVAL; 1796 len = sizeof(int); 1797 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0; 1798 if (put_user(len, optlen) || put_user(val, optval)) 1799 return -EFAULT; 1800 err = 0; 1801 break; 1802 default: 1803 err = -ENOPROTOOPT; 1804 } 1805 return err; 1806 } 1807 1808 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb) 1809 { 1810 struct nl_pktinfo info; 1811 1812 info.group = NETLINK_CB(skb).dst_group; 1813 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info); 1814 } 1815 1816 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg, 1817 struct sk_buff *skb) 1818 { 1819 if (!NETLINK_CB(skb).nsid_is_set) 1820 return; 1821 1822 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int), 1823 &NETLINK_CB(skb).nsid); 1824 } 1825 1826 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1827 { 1828 struct sock *sk = sock->sk; 1829 struct netlink_sock *nlk = nlk_sk(sk); 1830 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name); 1831 u32 dst_portid; 1832 u32 dst_group; 1833 struct sk_buff *skb; 1834 int err; 1835 struct scm_cookie scm; 1836 u32 netlink_skb_flags = 0; 1837 1838 if (msg->msg_flags&MSG_OOB) 1839 return -EOPNOTSUPP; 1840 1841 err = scm_send(sock, msg, &scm, true); 1842 if (err < 0) 1843 return err; 1844 1845 if (msg->msg_namelen) { 1846 err = -EINVAL; 1847 if (msg->msg_namelen < sizeof(struct sockaddr_nl)) 1848 goto out; 1849 if (addr->nl_family != AF_NETLINK) 1850 goto out; 1851 dst_portid = addr->nl_pid; 1852 dst_group = ffs(addr->nl_groups); 1853 err = -EPERM; 1854 if ((dst_group || dst_portid) && 1855 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND)) 1856 goto out; 1857 netlink_skb_flags |= NETLINK_SKB_DST; 1858 } else { 1859 dst_portid = nlk->dst_portid; 1860 dst_group = nlk->dst_group; 1861 } 1862 1863 if (!nlk->bound) { 1864 err = netlink_autobind(sock); 1865 if (err) 1866 goto out; 1867 } else { 1868 /* Ensure nlk is hashed and visible. */ 1869 smp_rmb(); 1870 } 1871 1872 err = -EMSGSIZE; 1873 if (len > sk->sk_sndbuf - 32) 1874 goto out; 1875 err = -ENOBUFS; 1876 skb = netlink_alloc_large_skb(len, dst_group); 1877 if (skb == NULL) 1878 goto out; 1879 1880 NETLINK_CB(skb).portid = nlk->portid; 1881 NETLINK_CB(skb).dst_group = dst_group; 1882 NETLINK_CB(skb).creds = scm.creds; 1883 NETLINK_CB(skb).flags = netlink_skb_flags; 1884 1885 err = -EFAULT; 1886 if (memcpy_from_msg(skb_put(skb, len), msg, len)) { 1887 kfree_skb(skb); 1888 goto out; 1889 } 1890 1891 err = security_netlink_send(sk, skb); 1892 if (err) { 1893 kfree_skb(skb); 1894 goto out; 1895 } 1896 1897 if (dst_group) { 1898 refcount_inc(&skb->users); 1899 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL); 1900 } 1901 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT); 1902 1903 out: 1904 scm_destroy(&scm); 1905 return err; 1906 } 1907 1908 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1909 int flags) 1910 { 1911 struct scm_cookie scm; 1912 struct sock *sk = sock->sk; 1913 struct netlink_sock *nlk = nlk_sk(sk); 1914 int noblock = flags&MSG_DONTWAIT; 1915 size_t copied; 1916 struct sk_buff *skb, *data_skb; 1917 int err, ret; 1918 1919 if (flags&MSG_OOB) 1920 return -EOPNOTSUPP; 1921 1922 copied = 0; 1923 1924 skb = skb_recv_datagram(sk, flags, noblock, &err); 1925 if (skb == NULL) 1926 goto out; 1927 1928 data_skb = skb; 1929 1930 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES 1931 if (unlikely(skb_shinfo(skb)->frag_list)) { 1932 /* 1933 * If this skb has a frag_list, then here that means that we 1934 * will have to use the frag_list skb's data for compat tasks 1935 * and the regular skb's data for normal (non-compat) tasks. 1936 * 1937 * If we need to send the compat skb, assign it to the 1938 * 'data_skb' variable so that it will be used below for data 1939 * copying. We keep 'skb' for everything else, including 1940 * freeing both later. 1941 */ 1942 if (flags & MSG_CMSG_COMPAT) 1943 data_skb = skb_shinfo(skb)->frag_list; 1944 } 1945 #endif 1946 1947 /* Record the max length of recvmsg() calls for future allocations */ 1948 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len); 1949 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len, 1950 SKB_WITH_OVERHEAD(32768)); 1951 1952 copied = data_skb->len; 1953 if (len < copied) { 1954 msg->msg_flags |= MSG_TRUNC; 1955 copied = len; 1956 } 1957 1958 skb_reset_transport_header(data_skb); 1959 err = skb_copy_datagram_msg(data_skb, 0, msg, copied); 1960 1961 if (msg->msg_name) { 1962 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name); 1963 addr->nl_family = AF_NETLINK; 1964 addr->nl_pad = 0; 1965 addr->nl_pid = NETLINK_CB(skb).portid; 1966 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group); 1967 msg->msg_namelen = sizeof(*addr); 1968 } 1969 1970 if (nlk->flags & NETLINK_F_RECV_PKTINFO) 1971 netlink_cmsg_recv_pktinfo(msg, skb); 1972 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID) 1973 netlink_cmsg_listen_all_nsid(sk, msg, skb); 1974 1975 memset(&scm, 0, sizeof(scm)); 1976 scm.creds = *NETLINK_CREDS(skb); 1977 if (flags & MSG_TRUNC) 1978 copied = data_skb->len; 1979 1980 skb_free_datagram(sk, skb); 1981 1982 if (nlk->cb_running && 1983 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) { 1984 ret = netlink_dump(sk); 1985 if (ret) { 1986 sk->sk_err = -ret; 1987 sk->sk_error_report(sk); 1988 } 1989 } 1990 1991 scm_recv(sock, msg, &scm, flags); 1992 out: 1993 netlink_rcv_wake(sk); 1994 return err ? : copied; 1995 } 1996 1997 static void netlink_data_ready(struct sock *sk) 1998 { 1999 BUG(); 2000 } 2001 2002 /* 2003 * We export these functions to other modules. They provide a 2004 * complete set of kernel non-blocking support for message 2005 * queueing. 2006 */ 2007 2008 struct sock * 2009 __netlink_kernel_create(struct net *net, int unit, struct module *module, 2010 struct netlink_kernel_cfg *cfg) 2011 { 2012 struct socket *sock; 2013 struct sock *sk; 2014 struct netlink_sock *nlk; 2015 struct listeners *listeners = NULL; 2016 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL; 2017 unsigned int groups; 2018 2019 BUG_ON(!nl_table); 2020 2021 if (unit < 0 || unit >= MAX_LINKS) 2022 return NULL; 2023 2024 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 2025 return NULL; 2026 2027 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0) 2028 goto out_sock_release_nosk; 2029 2030 sk = sock->sk; 2031 2032 if (!cfg || cfg->groups < 32) 2033 groups = 32; 2034 else 2035 groups = cfg->groups; 2036 2037 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 2038 if (!listeners) 2039 goto out_sock_release; 2040 2041 sk->sk_data_ready = netlink_data_ready; 2042 if (cfg && cfg->input) 2043 nlk_sk(sk)->netlink_rcv = cfg->input; 2044 2045 if (netlink_insert(sk, 0)) 2046 goto out_sock_release; 2047 2048 nlk = nlk_sk(sk); 2049 nlk->flags |= NETLINK_F_KERNEL_SOCKET; 2050 2051 netlink_table_grab(); 2052 if (!nl_table[unit].registered) { 2053 nl_table[unit].groups = groups; 2054 rcu_assign_pointer(nl_table[unit].listeners, listeners); 2055 nl_table[unit].cb_mutex = cb_mutex; 2056 nl_table[unit].module = module; 2057 if (cfg) { 2058 nl_table[unit].bind = cfg->bind; 2059 nl_table[unit].unbind = cfg->unbind; 2060 nl_table[unit].flags = cfg->flags; 2061 if (cfg->compare) 2062 nl_table[unit].compare = cfg->compare; 2063 } 2064 nl_table[unit].registered = 1; 2065 } else { 2066 kfree(listeners); 2067 nl_table[unit].registered++; 2068 } 2069 netlink_table_ungrab(); 2070 return sk; 2071 2072 out_sock_release: 2073 kfree(listeners); 2074 netlink_kernel_release(sk); 2075 return NULL; 2076 2077 out_sock_release_nosk: 2078 sock_release(sock); 2079 return NULL; 2080 } 2081 EXPORT_SYMBOL(__netlink_kernel_create); 2082 2083 void 2084 netlink_kernel_release(struct sock *sk) 2085 { 2086 if (sk == NULL || sk->sk_socket == NULL) 2087 return; 2088 2089 sock_release(sk->sk_socket); 2090 } 2091 EXPORT_SYMBOL(netlink_kernel_release); 2092 2093 int __netlink_change_ngroups(struct sock *sk, unsigned int groups) 2094 { 2095 struct listeners *new, *old; 2096 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 2097 2098 if (groups < 32) 2099 groups = 32; 2100 2101 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) { 2102 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC); 2103 if (!new) 2104 return -ENOMEM; 2105 old = nl_deref_protected(tbl->listeners); 2106 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups)); 2107 rcu_assign_pointer(tbl->listeners, new); 2108 2109 kfree_rcu(old, rcu); 2110 } 2111 tbl->groups = groups; 2112 2113 return 0; 2114 } 2115 2116 /** 2117 * netlink_change_ngroups - change number of multicast groups 2118 * 2119 * This changes the number of multicast groups that are available 2120 * on a certain netlink family. Note that it is not possible to 2121 * change the number of groups to below 32. Also note that it does 2122 * not implicitly call netlink_clear_multicast_users() when the 2123 * number of groups is reduced. 2124 * 2125 * @sk: The kernel netlink socket, as returned by netlink_kernel_create(). 2126 * @groups: The new number of groups. 2127 */ 2128 int netlink_change_ngroups(struct sock *sk, unsigned int groups) 2129 { 2130 int err; 2131 2132 netlink_table_grab(); 2133 err = __netlink_change_ngroups(sk, groups); 2134 netlink_table_ungrab(); 2135 2136 return err; 2137 } 2138 2139 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group) 2140 { 2141 struct sock *sk; 2142 struct netlink_table *tbl = &nl_table[ksk->sk_protocol]; 2143 2144 sk_for_each_bound(sk, &tbl->mc_list) 2145 netlink_update_socket_mc(nlk_sk(sk), group, 0); 2146 } 2147 2148 struct nlmsghdr * 2149 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags) 2150 { 2151 struct nlmsghdr *nlh; 2152 int size = nlmsg_msg_size(len); 2153 2154 nlh = skb_put(skb, NLMSG_ALIGN(size)); 2155 nlh->nlmsg_type = type; 2156 nlh->nlmsg_len = size; 2157 nlh->nlmsg_flags = flags; 2158 nlh->nlmsg_pid = portid; 2159 nlh->nlmsg_seq = seq; 2160 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0) 2161 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size); 2162 return nlh; 2163 } 2164 EXPORT_SYMBOL(__nlmsg_put); 2165 2166 /* 2167 * It looks a bit ugly. 2168 * It would be better to create kernel thread. 2169 */ 2170 2171 static int netlink_dump(struct sock *sk) 2172 { 2173 struct netlink_sock *nlk = nlk_sk(sk); 2174 struct netlink_callback *cb; 2175 struct sk_buff *skb = NULL; 2176 struct nlmsghdr *nlh; 2177 struct module *module; 2178 int err = -ENOBUFS; 2179 int alloc_min_size; 2180 int alloc_size; 2181 2182 mutex_lock(nlk->cb_mutex); 2183 if (!nlk->cb_running) { 2184 err = -EINVAL; 2185 goto errout_skb; 2186 } 2187 2188 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2189 goto errout_skb; 2190 2191 /* NLMSG_GOODSIZE is small to avoid high order allocations being 2192 * required, but it makes sense to _attempt_ a 16K bytes allocation 2193 * to reduce number of system calls on dump operations, if user 2194 * ever provided a big enough buffer. 2195 */ 2196 cb = &nlk->cb; 2197 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE); 2198 2199 if (alloc_min_size < nlk->max_recvmsg_len) { 2200 alloc_size = nlk->max_recvmsg_len; 2201 skb = alloc_skb(alloc_size, 2202 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) | 2203 __GFP_NOWARN | __GFP_NORETRY); 2204 } 2205 if (!skb) { 2206 alloc_size = alloc_min_size; 2207 skb = alloc_skb(alloc_size, GFP_KERNEL); 2208 } 2209 if (!skb) 2210 goto errout_skb; 2211 2212 /* Trim skb to allocated size. User is expected to provide buffer as 2213 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at 2214 * netlink_recvmsg())). dump will pack as many smaller messages as 2215 * could fit within the allocated skb. skb is typically allocated 2216 * with larger space than required (could be as much as near 2x the 2217 * requested size with align to next power of 2 approach). Allowing 2218 * dump to use the excess space makes it difficult for a user to have a 2219 * reasonable static buffer based on the expected largest dump of a 2220 * single netdev. The outcome is MSG_TRUNC error. 2221 */ 2222 skb_reserve(skb, skb_tailroom(skb) - alloc_size); 2223 netlink_skb_set_owner_r(skb, sk); 2224 2225 if (nlk->dump_done_errno > 0) 2226 nlk->dump_done_errno = cb->dump(skb, cb); 2227 2228 if (nlk->dump_done_errno > 0 || 2229 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) { 2230 mutex_unlock(nlk->cb_mutex); 2231 2232 if (sk_filter(sk, skb)) 2233 kfree_skb(skb); 2234 else 2235 __netlink_sendskb(sk, skb); 2236 return 0; 2237 } 2238 2239 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, 2240 sizeof(nlk->dump_done_errno), NLM_F_MULTI); 2241 if (WARN_ON(!nlh)) 2242 goto errout_skb; 2243 2244 nl_dump_check_consistent(cb, nlh); 2245 2246 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, 2247 sizeof(nlk->dump_done_errno)); 2248 2249 if (sk_filter(sk, skb)) 2250 kfree_skb(skb); 2251 else 2252 __netlink_sendskb(sk, skb); 2253 2254 if (cb->done) 2255 cb->done(cb); 2256 2257 nlk->cb_running = false; 2258 module = cb->module; 2259 skb = cb->skb; 2260 mutex_unlock(nlk->cb_mutex); 2261 module_put(module); 2262 consume_skb(skb); 2263 return 0; 2264 2265 errout_skb: 2266 mutex_unlock(nlk->cb_mutex); 2267 kfree_skb(skb); 2268 return err; 2269 } 2270 2271 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 2272 const struct nlmsghdr *nlh, 2273 struct netlink_dump_control *control) 2274 { 2275 struct netlink_callback *cb; 2276 struct sock *sk; 2277 struct netlink_sock *nlk; 2278 int ret; 2279 2280 refcount_inc(&skb->users); 2281 2282 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid); 2283 if (sk == NULL) { 2284 ret = -ECONNREFUSED; 2285 goto error_free; 2286 } 2287 2288 nlk = nlk_sk(sk); 2289 mutex_lock(nlk->cb_mutex); 2290 /* A dump is in progress... */ 2291 if (nlk->cb_running) { 2292 ret = -EBUSY; 2293 goto error_unlock; 2294 } 2295 /* add reference of module which cb->dump belongs to */ 2296 if (!try_module_get(control->module)) { 2297 ret = -EPROTONOSUPPORT; 2298 goto error_unlock; 2299 } 2300 2301 cb = &nlk->cb; 2302 memset(cb, 0, sizeof(*cb)); 2303 cb->dump = control->dump; 2304 cb->done = control->done; 2305 cb->nlh = nlh; 2306 cb->data = control->data; 2307 cb->module = control->module; 2308 cb->min_dump_alloc = control->min_dump_alloc; 2309 cb->skb = skb; 2310 2311 if (control->start) { 2312 ret = control->start(cb); 2313 if (ret) 2314 goto error_put; 2315 } 2316 2317 nlk->cb_running = true; 2318 nlk->dump_done_errno = INT_MAX; 2319 2320 mutex_unlock(nlk->cb_mutex); 2321 2322 ret = netlink_dump(sk); 2323 2324 sock_put(sk); 2325 2326 if (ret) 2327 return ret; 2328 2329 /* We successfully started a dump, by returning -EINTR we 2330 * signal not to send ACK even if it was requested. 2331 */ 2332 return -EINTR; 2333 2334 error_put: 2335 module_put(control->module); 2336 error_unlock: 2337 sock_put(sk); 2338 mutex_unlock(nlk->cb_mutex); 2339 error_free: 2340 kfree_skb(skb); 2341 return ret; 2342 } 2343 EXPORT_SYMBOL(__netlink_dump_start); 2344 2345 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err, 2346 const struct netlink_ext_ack *extack) 2347 { 2348 struct sk_buff *skb; 2349 struct nlmsghdr *rep; 2350 struct nlmsgerr *errmsg; 2351 size_t payload = sizeof(*errmsg); 2352 size_t tlvlen = 0; 2353 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk); 2354 unsigned int flags = 0; 2355 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK; 2356 2357 /* Error messages get the original request appened, unless the user 2358 * requests to cap the error message, and get extra error data if 2359 * requested. 2360 */ 2361 if (nlk_has_extack && extack && extack->_msg) 2362 tlvlen += nla_total_size(strlen(extack->_msg) + 1); 2363 2364 if (err) { 2365 if (!(nlk->flags & NETLINK_F_CAP_ACK)) 2366 payload += nlmsg_len(nlh); 2367 else 2368 flags |= NLM_F_CAPPED; 2369 if (nlk_has_extack && extack && extack->bad_attr) 2370 tlvlen += nla_total_size(sizeof(u32)); 2371 } else { 2372 flags |= NLM_F_CAPPED; 2373 2374 if (nlk_has_extack && extack && extack->cookie_len) 2375 tlvlen += nla_total_size(extack->cookie_len); 2376 } 2377 2378 if (tlvlen) 2379 flags |= NLM_F_ACK_TLVS; 2380 2381 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL); 2382 if (!skb) { 2383 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS; 2384 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk); 2385 return; 2386 } 2387 2388 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, 2389 NLMSG_ERROR, payload, flags); 2390 errmsg = nlmsg_data(rep); 2391 errmsg->error = err; 2392 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh)); 2393 2394 if (nlk_has_extack && extack) { 2395 if (extack->_msg) { 2396 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, 2397 extack->_msg)); 2398 } 2399 if (err) { 2400 if (extack->bad_attr && 2401 !WARN_ON((u8 *)extack->bad_attr < in_skb->data || 2402 (u8 *)extack->bad_attr >= in_skb->data + 2403 in_skb->len)) 2404 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS, 2405 (u8 *)extack->bad_attr - 2406 in_skb->data)); 2407 } else { 2408 if (extack->cookie_len) 2409 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE, 2410 extack->cookie_len, 2411 extack->cookie)); 2412 } 2413 } 2414 2415 nlmsg_end(skb, rep); 2416 2417 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT); 2418 } 2419 EXPORT_SYMBOL(netlink_ack); 2420 2421 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *, 2422 struct nlmsghdr *, 2423 struct netlink_ext_ack *)) 2424 { 2425 struct netlink_ext_ack extack; 2426 struct nlmsghdr *nlh; 2427 int err; 2428 2429 while (skb->len >= nlmsg_total_size(0)) { 2430 int msglen; 2431 2432 memset(&extack, 0, sizeof(extack)); 2433 nlh = nlmsg_hdr(skb); 2434 err = 0; 2435 2436 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len) 2437 return 0; 2438 2439 /* Only requests are handled by the kernel */ 2440 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) 2441 goto ack; 2442 2443 /* Skip control messages */ 2444 if (nlh->nlmsg_type < NLMSG_MIN_TYPE) 2445 goto ack; 2446 2447 err = cb(skb, nlh, &extack); 2448 if (err == -EINTR) 2449 goto skip; 2450 2451 ack: 2452 if (nlh->nlmsg_flags & NLM_F_ACK || err) 2453 netlink_ack(skb, nlh, err, &extack); 2454 2455 skip: 2456 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 2457 if (msglen > skb->len) 2458 msglen = skb->len; 2459 skb_pull(skb, msglen); 2460 } 2461 2462 return 0; 2463 } 2464 EXPORT_SYMBOL(netlink_rcv_skb); 2465 2466 /** 2467 * nlmsg_notify - send a notification netlink message 2468 * @sk: netlink socket to use 2469 * @skb: notification message 2470 * @portid: destination netlink portid for reports or 0 2471 * @group: destination multicast group or 0 2472 * @report: 1 to report back, 0 to disable 2473 * @flags: allocation flags 2474 */ 2475 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid, 2476 unsigned int group, int report, gfp_t flags) 2477 { 2478 int err = 0; 2479 2480 if (group) { 2481 int exclude_portid = 0; 2482 2483 if (report) { 2484 refcount_inc(&skb->users); 2485 exclude_portid = portid; 2486 } 2487 2488 /* errors reported via destination sk->sk_err, but propagate 2489 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */ 2490 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags); 2491 } 2492 2493 if (report) { 2494 int err2; 2495 2496 err2 = nlmsg_unicast(sk, skb, portid); 2497 if (!err || err == -ESRCH) 2498 err = err2; 2499 } 2500 2501 return err; 2502 } 2503 EXPORT_SYMBOL(nlmsg_notify); 2504 2505 #ifdef CONFIG_PROC_FS 2506 struct nl_seq_iter { 2507 struct seq_net_private p; 2508 struct rhashtable_iter hti; 2509 int link; 2510 }; 2511 2512 static int netlink_walk_start(struct nl_seq_iter *iter) 2513 { 2514 int err; 2515 2516 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti, 2517 GFP_KERNEL); 2518 if (err) { 2519 iter->link = MAX_LINKS; 2520 return err; 2521 } 2522 2523 rhashtable_walk_start(&iter->hti); 2524 2525 return 0; 2526 } 2527 2528 static void netlink_walk_stop(struct nl_seq_iter *iter) 2529 { 2530 rhashtable_walk_stop(&iter->hti); 2531 rhashtable_walk_exit(&iter->hti); 2532 } 2533 2534 static void *__netlink_seq_next(struct seq_file *seq) 2535 { 2536 struct nl_seq_iter *iter = seq->private; 2537 struct netlink_sock *nlk; 2538 2539 do { 2540 for (;;) { 2541 int err; 2542 2543 nlk = rhashtable_walk_next(&iter->hti); 2544 2545 if (IS_ERR(nlk)) { 2546 if (PTR_ERR(nlk) == -EAGAIN) 2547 continue; 2548 2549 return nlk; 2550 } 2551 2552 if (nlk) 2553 break; 2554 2555 netlink_walk_stop(iter); 2556 if (++iter->link >= MAX_LINKS) 2557 return NULL; 2558 2559 err = netlink_walk_start(iter); 2560 if (err) 2561 return ERR_PTR(err); 2562 } 2563 } while (sock_net(&nlk->sk) != seq_file_net(seq)); 2564 2565 return nlk; 2566 } 2567 2568 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp) 2569 { 2570 struct nl_seq_iter *iter = seq->private; 2571 void *obj = SEQ_START_TOKEN; 2572 loff_t pos; 2573 int err; 2574 2575 iter->link = 0; 2576 2577 err = netlink_walk_start(iter); 2578 if (err) 2579 return ERR_PTR(err); 2580 2581 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--) 2582 obj = __netlink_seq_next(seq); 2583 2584 return obj; 2585 } 2586 2587 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2588 { 2589 ++*pos; 2590 return __netlink_seq_next(seq); 2591 } 2592 2593 static void netlink_seq_stop(struct seq_file *seq, void *v) 2594 { 2595 struct nl_seq_iter *iter = seq->private; 2596 2597 if (iter->link >= MAX_LINKS) 2598 return; 2599 2600 netlink_walk_stop(iter); 2601 } 2602 2603 2604 static int netlink_seq_show(struct seq_file *seq, void *v) 2605 { 2606 if (v == SEQ_START_TOKEN) { 2607 seq_puts(seq, 2608 "sk Eth Pid Groups " 2609 "Rmem Wmem Dump Locks Drops Inode\n"); 2610 } else { 2611 struct sock *s = v; 2612 struct netlink_sock *nlk = nlk_sk(s); 2613 2614 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8d %-8lu\n", 2615 s, 2616 s->sk_protocol, 2617 nlk->portid, 2618 nlk->groups ? (u32)nlk->groups[0] : 0, 2619 sk_rmem_alloc_get(s), 2620 sk_wmem_alloc_get(s), 2621 nlk->cb_running, 2622 refcount_read(&s->sk_refcnt), 2623 atomic_read(&s->sk_drops), 2624 sock_i_ino(s) 2625 ); 2626 2627 } 2628 return 0; 2629 } 2630 2631 static const struct seq_operations netlink_seq_ops = { 2632 .start = netlink_seq_start, 2633 .next = netlink_seq_next, 2634 .stop = netlink_seq_stop, 2635 .show = netlink_seq_show, 2636 }; 2637 #endif 2638 2639 int netlink_register_notifier(struct notifier_block *nb) 2640 { 2641 return blocking_notifier_chain_register(&netlink_chain, nb); 2642 } 2643 EXPORT_SYMBOL(netlink_register_notifier); 2644 2645 int netlink_unregister_notifier(struct notifier_block *nb) 2646 { 2647 return blocking_notifier_chain_unregister(&netlink_chain, nb); 2648 } 2649 EXPORT_SYMBOL(netlink_unregister_notifier); 2650 2651 static const struct proto_ops netlink_ops = { 2652 .family = PF_NETLINK, 2653 .owner = THIS_MODULE, 2654 .release = netlink_release, 2655 .bind = netlink_bind, 2656 .connect = netlink_connect, 2657 .socketpair = sock_no_socketpair, 2658 .accept = sock_no_accept, 2659 .getname = netlink_getname, 2660 .poll = datagram_poll, 2661 .ioctl = netlink_ioctl, 2662 .listen = sock_no_listen, 2663 .shutdown = sock_no_shutdown, 2664 .setsockopt = netlink_setsockopt, 2665 .getsockopt = netlink_getsockopt, 2666 .sendmsg = netlink_sendmsg, 2667 .recvmsg = netlink_recvmsg, 2668 .mmap = sock_no_mmap, 2669 .sendpage = sock_no_sendpage, 2670 }; 2671 2672 static const struct net_proto_family netlink_family_ops = { 2673 .family = PF_NETLINK, 2674 .create = netlink_create, 2675 .owner = THIS_MODULE, /* for consistency 8) */ 2676 }; 2677 2678 static int __net_init netlink_net_init(struct net *net) 2679 { 2680 #ifdef CONFIG_PROC_FS 2681 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops, 2682 sizeof(struct nl_seq_iter))) 2683 return -ENOMEM; 2684 #endif 2685 return 0; 2686 } 2687 2688 static void __net_exit netlink_net_exit(struct net *net) 2689 { 2690 #ifdef CONFIG_PROC_FS 2691 remove_proc_entry("netlink", net->proc_net); 2692 #endif 2693 } 2694 2695 static void __init netlink_add_usersock_entry(void) 2696 { 2697 struct listeners *listeners; 2698 int groups = 32; 2699 2700 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 2701 if (!listeners) 2702 panic("netlink_add_usersock_entry: Cannot allocate listeners\n"); 2703 2704 netlink_table_grab(); 2705 2706 nl_table[NETLINK_USERSOCK].groups = groups; 2707 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners); 2708 nl_table[NETLINK_USERSOCK].module = THIS_MODULE; 2709 nl_table[NETLINK_USERSOCK].registered = 1; 2710 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND; 2711 2712 netlink_table_ungrab(); 2713 } 2714 2715 static struct pernet_operations __net_initdata netlink_net_ops = { 2716 .init = netlink_net_init, 2717 .exit = netlink_net_exit, 2718 }; 2719 2720 static inline u32 netlink_hash(const void *data, u32 len, u32 seed) 2721 { 2722 const struct netlink_sock *nlk = data; 2723 struct netlink_compare_arg arg; 2724 2725 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid); 2726 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed); 2727 } 2728 2729 static const struct rhashtable_params netlink_rhashtable_params = { 2730 .head_offset = offsetof(struct netlink_sock, node), 2731 .key_len = netlink_compare_arg_len, 2732 .obj_hashfn = netlink_hash, 2733 .obj_cmpfn = netlink_compare, 2734 .automatic_shrinking = true, 2735 }; 2736 2737 static int __init netlink_proto_init(void) 2738 { 2739 int i; 2740 int err = proto_register(&netlink_proto, 0); 2741 2742 if (err != 0) 2743 goto out; 2744 2745 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb)); 2746 2747 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL); 2748 if (!nl_table) 2749 goto panic; 2750 2751 for (i = 0; i < MAX_LINKS; i++) { 2752 if (rhashtable_init(&nl_table[i].hash, 2753 &netlink_rhashtable_params) < 0) { 2754 while (--i > 0) 2755 rhashtable_destroy(&nl_table[i].hash); 2756 kfree(nl_table); 2757 goto panic; 2758 } 2759 } 2760 2761 netlink_add_usersock_entry(); 2762 2763 sock_register(&netlink_family_ops); 2764 register_pernet_subsys(&netlink_net_ops); 2765 register_pernet_subsys(&netlink_tap_net_ops); 2766 /* The netlink device handler may be needed early. */ 2767 rtnetlink_init(); 2768 out: 2769 return err; 2770 panic: 2771 panic("netlink_init: Cannot allocate nl_table\n"); 2772 } 2773 2774 core_initcall(netlink_proto_init); 2775