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 err = -ENOMEM; 578 if (BITS_PER_LONG > 32 && 579 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX)) 580 goto err; 581 582 nlk_sk(sk)->portid = portid; 583 sock_hold(sk); 584 585 err = __netlink_insert(table, sk); 586 if (err) { 587 /* In case the hashtable backend returns with -EBUSY 588 * from here, it must not escape to the caller. 589 */ 590 if (unlikely(err == -EBUSY)) 591 err = -EOVERFLOW; 592 if (err == -EEXIST) 593 err = -EADDRINUSE; 594 sock_put(sk); 595 goto err; 596 } 597 598 /* We need to ensure that the socket is hashed and visible. */ 599 smp_wmb(); 600 nlk_sk(sk)->bound = portid; 601 602 err: 603 release_sock(sk); 604 return err; 605 } 606 607 static void netlink_remove(struct sock *sk) 608 { 609 struct netlink_table *table; 610 611 table = &nl_table[sk->sk_protocol]; 612 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node, 613 netlink_rhashtable_params)) { 614 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 615 __sock_put(sk); 616 } 617 618 netlink_table_grab(); 619 if (nlk_sk(sk)->subscriptions) { 620 __sk_del_bind_node(sk); 621 netlink_update_listeners(sk); 622 } 623 if (sk->sk_protocol == NETLINK_GENERIC) 624 atomic_inc(&genl_sk_destructing_cnt); 625 netlink_table_ungrab(); 626 } 627 628 static struct proto netlink_proto = { 629 .name = "NETLINK", 630 .owner = THIS_MODULE, 631 .obj_size = sizeof(struct netlink_sock), 632 }; 633 634 static int __netlink_create(struct net *net, struct socket *sock, 635 struct mutex *cb_mutex, int protocol, 636 int kern) 637 { 638 struct sock *sk; 639 struct netlink_sock *nlk; 640 641 sock->ops = &netlink_ops; 642 643 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern); 644 if (!sk) 645 return -ENOMEM; 646 647 sock_init_data(sock, sk); 648 649 nlk = nlk_sk(sk); 650 if (cb_mutex) { 651 nlk->cb_mutex = cb_mutex; 652 } else { 653 nlk->cb_mutex = &nlk->cb_def_mutex; 654 mutex_init(nlk->cb_mutex); 655 lockdep_set_class_and_name(nlk->cb_mutex, 656 nlk_cb_mutex_keys + protocol, 657 nlk_cb_mutex_key_strings[protocol]); 658 } 659 init_waitqueue_head(&nlk->wait); 660 661 sk->sk_destruct = netlink_sock_destruct; 662 sk->sk_protocol = protocol; 663 return 0; 664 } 665 666 static int netlink_create(struct net *net, struct socket *sock, int protocol, 667 int kern) 668 { 669 struct module *module = NULL; 670 struct mutex *cb_mutex; 671 struct netlink_sock *nlk; 672 int (*bind)(struct net *net, int group); 673 void (*unbind)(struct net *net, int group); 674 int err = 0; 675 676 sock->state = SS_UNCONNECTED; 677 678 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 679 return -ESOCKTNOSUPPORT; 680 681 if (protocol < 0 || protocol >= MAX_LINKS) 682 return -EPROTONOSUPPORT; 683 protocol = array_index_nospec(protocol, MAX_LINKS); 684 685 netlink_lock_table(); 686 #ifdef CONFIG_MODULES 687 if (!nl_table[protocol].registered) { 688 netlink_unlock_table(); 689 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol); 690 netlink_lock_table(); 691 } 692 #endif 693 if (nl_table[protocol].registered && 694 try_module_get(nl_table[protocol].module)) 695 module = nl_table[protocol].module; 696 else 697 err = -EPROTONOSUPPORT; 698 cb_mutex = nl_table[protocol].cb_mutex; 699 bind = nl_table[protocol].bind; 700 unbind = nl_table[protocol].unbind; 701 netlink_unlock_table(); 702 703 if (err < 0) 704 goto out; 705 706 err = __netlink_create(net, sock, cb_mutex, protocol, kern); 707 if (err < 0) 708 goto out_module; 709 710 local_bh_disable(); 711 sock_prot_inuse_add(net, &netlink_proto, 1); 712 local_bh_enable(); 713 714 nlk = nlk_sk(sock->sk); 715 nlk->module = module; 716 nlk->netlink_bind = bind; 717 nlk->netlink_unbind = unbind; 718 out: 719 return err; 720 721 out_module: 722 module_put(module); 723 goto out; 724 } 725 726 static void deferred_put_nlk_sk(struct rcu_head *head) 727 { 728 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu); 729 struct sock *sk = &nlk->sk; 730 731 kfree(nlk->groups); 732 nlk->groups = NULL; 733 734 if (!refcount_dec_and_test(&sk->sk_refcnt)) 735 return; 736 737 if (nlk->cb_running && nlk->cb.done) { 738 INIT_WORK(&nlk->work, netlink_sock_destruct_work); 739 schedule_work(&nlk->work); 740 return; 741 } 742 743 sk_free(sk); 744 } 745 746 static int netlink_release(struct socket *sock) 747 { 748 struct sock *sk = sock->sk; 749 struct netlink_sock *nlk; 750 751 if (!sk) 752 return 0; 753 754 netlink_remove(sk); 755 sock_orphan(sk); 756 nlk = nlk_sk(sk); 757 758 /* 759 * OK. Socket is unlinked, any packets that arrive now 760 * will be purged. 761 */ 762 763 /* must not acquire netlink_table_lock in any way again before unbind 764 * and notifying genetlink is done as otherwise it might deadlock 765 */ 766 if (nlk->netlink_unbind) { 767 int i; 768 769 for (i = 0; i < nlk->ngroups; i++) 770 if (test_bit(i, nlk->groups)) 771 nlk->netlink_unbind(sock_net(sk), i + 1); 772 } 773 if (sk->sk_protocol == NETLINK_GENERIC && 774 atomic_dec_return(&genl_sk_destructing_cnt) == 0) 775 wake_up(&genl_sk_destructing_waitq); 776 777 sock->sk = NULL; 778 wake_up_interruptible_all(&nlk->wait); 779 780 skb_queue_purge(&sk->sk_write_queue); 781 782 if (nlk->portid && nlk->bound) { 783 struct netlink_notify n = { 784 .net = sock_net(sk), 785 .protocol = sk->sk_protocol, 786 .portid = nlk->portid, 787 }; 788 blocking_notifier_call_chain(&netlink_chain, 789 NETLINK_URELEASE, &n); 790 } 791 792 module_put(nlk->module); 793 794 if (netlink_is_kernel(sk)) { 795 netlink_table_grab(); 796 BUG_ON(nl_table[sk->sk_protocol].registered == 0); 797 if (--nl_table[sk->sk_protocol].registered == 0) { 798 struct listeners *old; 799 800 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners); 801 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL); 802 kfree_rcu(old, rcu); 803 nl_table[sk->sk_protocol].module = NULL; 804 nl_table[sk->sk_protocol].bind = NULL; 805 nl_table[sk->sk_protocol].unbind = NULL; 806 nl_table[sk->sk_protocol].flags = 0; 807 nl_table[sk->sk_protocol].registered = 0; 808 } 809 netlink_table_ungrab(); 810 } 811 812 local_bh_disable(); 813 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1); 814 local_bh_enable(); 815 call_rcu(&nlk->rcu, deferred_put_nlk_sk); 816 return 0; 817 } 818 819 static int netlink_autobind(struct socket *sock) 820 { 821 struct sock *sk = sock->sk; 822 struct net *net = sock_net(sk); 823 struct netlink_table *table = &nl_table[sk->sk_protocol]; 824 s32 portid = task_tgid_vnr(current); 825 int err; 826 s32 rover = -4096; 827 bool ok; 828 829 retry: 830 cond_resched(); 831 rcu_read_lock(); 832 ok = !__netlink_lookup(table, portid, net); 833 rcu_read_unlock(); 834 if (!ok) { 835 /* Bind collision, search negative portid values. */ 836 if (rover == -4096) 837 /* rover will be in range [S32_MIN, -4097] */ 838 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN); 839 else if (rover >= -4096) 840 rover = -4097; 841 portid = rover--; 842 goto retry; 843 } 844 845 err = netlink_insert(sk, portid); 846 if (err == -EADDRINUSE) 847 goto retry; 848 849 /* If 2 threads race to autobind, that is fine. */ 850 if (err == -EBUSY) 851 err = 0; 852 853 return err; 854 } 855 856 /** 857 * __netlink_ns_capable - General netlink message capability test 858 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace. 859 * @user_ns: The user namespace of the capability to use 860 * @cap: The capability to use 861 * 862 * Test to see if the opener of the socket we received the message 863 * from had when the netlink socket was created and the sender of the 864 * message has has the capability @cap in the user namespace @user_ns. 865 */ 866 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp, 867 struct user_namespace *user_ns, int cap) 868 { 869 return ((nsp->flags & NETLINK_SKB_DST) || 870 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) && 871 ns_capable(user_ns, cap); 872 } 873 EXPORT_SYMBOL(__netlink_ns_capable); 874 875 /** 876 * netlink_ns_capable - General netlink message capability test 877 * @skb: socket buffer holding a netlink command from userspace 878 * @user_ns: The user namespace of the capability to use 879 * @cap: The capability to use 880 * 881 * Test to see if the opener of the socket we received the message 882 * from had when the netlink socket was created and the sender of the 883 * message has has the capability @cap in the user namespace @user_ns. 884 */ 885 bool netlink_ns_capable(const struct sk_buff *skb, 886 struct user_namespace *user_ns, int cap) 887 { 888 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap); 889 } 890 EXPORT_SYMBOL(netlink_ns_capable); 891 892 /** 893 * netlink_capable - Netlink global message capability test 894 * @skb: socket buffer holding a netlink command from userspace 895 * @cap: The capability to use 896 * 897 * Test to see if the opener of the socket we received the message 898 * from had when the netlink socket was created and the sender of the 899 * message has has the capability @cap in all user namespaces. 900 */ 901 bool netlink_capable(const struct sk_buff *skb, int cap) 902 { 903 return netlink_ns_capable(skb, &init_user_ns, cap); 904 } 905 EXPORT_SYMBOL(netlink_capable); 906 907 /** 908 * netlink_net_capable - Netlink network namespace message capability test 909 * @skb: socket buffer holding a netlink command from userspace 910 * @cap: The capability to use 911 * 912 * Test to see if the opener of the socket we received the message 913 * from had when the netlink socket was created and the sender of the 914 * message has has the capability @cap over the network namespace of 915 * the socket we received the message from. 916 */ 917 bool netlink_net_capable(const struct sk_buff *skb, int cap) 918 { 919 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap); 920 } 921 EXPORT_SYMBOL(netlink_net_capable); 922 923 static inline int netlink_allowed(const struct socket *sock, unsigned int flag) 924 { 925 return (nl_table[sock->sk->sk_protocol].flags & flag) || 926 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN); 927 } 928 929 static void 930 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions) 931 { 932 struct netlink_sock *nlk = nlk_sk(sk); 933 934 if (nlk->subscriptions && !subscriptions) 935 __sk_del_bind_node(sk); 936 else if (!nlk->subscriptions && subscriptions) 937 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 938 nlk->subscriptions = subscriptions; 939 } 940 941 static int netlink_realloc_groups(struct sock *sk) 942 { 943 struct netlink_sock *nlk = nlk_sk(sk); 944 unsigned int groups; 945 unsigned long *new_groups; 946 int err = 0; 947 948 netlink_table_grab(); 949 950 groups = nl_table[sk->sk_protocol].groups; 951 if (!nl_table[sk->sk_protocol].registered) { 952 err = -ENOENT; 953 goto out_unlock; 954 } 955 956 if (nlk->ngroups >= groups) 957 goto out_unlock; 958 959 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC); 960 if (new_groups == NULL) { 961 err = -ENOMEM; 962 goto out_unlock; 963 } 964 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0, 965 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups)); 966 967 nlk->groups = new_groups; 968 nlk->ngroups = groups; 969 out_unlock: 970 netlink_table_ungrab(); 971 return err; 972 } 973 974 static void netlink_undo_bind(int group, long unsigned int groups, 975 struct sock *sk) 976 { 977 struct netlink_sock *nlk = nlk_sk(sk); 978 int undo; 979 980 if (!nlk->netlink_unbind) 981 return; 982 983 for (undo = 0; undo < group; undo++) 984 if (test_bit(undo, &groups)) 985 nlk->netlink_unbind(sock_net(sk), undo + 1); 986 } 987 988 static int netlink_bind(struct socket *sock, struct sockaddr *addr, 989 int addr_len) 990 { 991 struct sock *sk = sock->sk; 992 struct net *net = sock_net(sk); 993 struct netlink_sock *nlk = nlk_sk(sk); 994 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 995 int err = 0; 996 unsigned long groups = nladdr->nl_groups; 997 bool bound; 998 999 if (addr_len < sizeof(struct sockaddr_nl)) 1000 return -EINVAL; 1001 1002 if (nladdr->nl_family != AF_NETLINK) 1003 return -EINVAL; 1004 1005 /* Only superuser is allowed to listen multicasts */ 1006 if (groups) { 1007 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV)) 1008 return -EPERM; 1009 err = netlink_realloc_groups(sk); 1010 if (err) 1011 return err; 1012 } 1013 1014 if (nlk->ngroups < BITS_PER_LONG) 1015 groups &= (1UL << nlk->ngroups) - 1; 1016 1017 bound = nlk->bound; 1018 if (bound) { 1019 /* Ensure nlk->portid is up-to-date. */ 1020 smp_rmb(); 1021 1022 if (nladdr->nl_pid != nlk->portid) 1023 return -EINVAL; 1024 } 1025 1026 netlink_lock_table(); 1027 if (nlk->netlink_bind && groups) { 1028 int group; 1029 1030 for (group = 0; group < nlk->ngroups; group++) { 1031 if (!test_bit(group, &groups)) 1032 continue; 1033 err = nlk->netlink_bind(net, group + 1); 1034 if (!err) 1035 continue; 1036 netlink_undo_bind(group, groups, sk); 1037 goto unlock; 1038 } 1039 } 1040 1041 /* No need for barriers here as we return to user-space without 1042 * using any of the bound attributes. 1043 */ 1044 if (!bound) { 1045 err = nladdr->nl_pid ? 1046 netlink_insert(sk, nladdr->nl_pid) : 1047 netlink_autobind(sock); 1048 if (err) { 1049 netlink_undo_bind(nlk->ngroups, groups, sk); 1050 goto unlock; 1051 } 1052 } 1053 1054 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0])) 1055 goto unlock; 1056 netlink_unlock_table(); 1057 1058 netlink_table_grab(); 1059 netlink_update_subscriptions(sk, nlk->subscriptions + 1060 hweight32(groups) - 1061 hweight32(nlk->groups[0])); 1062 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups; 1063 netlink_update_listeners(sk); 1064 netlink_table_ungrab(); 1065 1066 return 0; 1067 1068 unlock: 1069 netlink_unlock_table(); 1070 return err; 1071 } 1072 1073 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 1074 int alen, int flags) 1075 { 1076 int err = 0; 1077 struct sock *sk = sock->sk; 1078 struct netlink_sock *nlk = nlk_sk(sk); 1079 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 1080 1081 if (alen < sizeof(addr->sa_family)) 1082 return -EINVAL; 1083 1084 if (addr->sa_family == AF_UNSPEC) { 1085 sk->sk_state = NETLINK_UNCONNECTED; 1086 nlk->dst_portid = 0; 1087 nlk->dst_group = 0; 1088 return 0; 1089 } 1090 if (addr->sa_family != AF_NETLINK) 1091 return -EINVAL; 1092 1093 if (alen < sizeof(struct sockaddr_nl)) 1094 return -EINVAL; 1095 1096 if ((nladdr->nl_groups || nladdr->nl_pid) && 1097 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND)) 1098 return -EPERM; 1099 1100 /* No need for barriers here as we return to user-space without 1101 * using any of the bound attributes. 1102 */ 1103 if (!nlk->bound) 1104 err = netlink_autobind(sock); 1105 1106 if (err == 0) { 1107 sk->sk_state = NETLINK_CONNECTED; 1108 nlk->dst_portid = nladdr->nl_pid; 1109 nlk->dst_group = ffs(nladdr->nl_groups); 1110 } 1111 1112 return err; 1113 } 1114 1115 static int netlink_getname(struct socket *sock, struct sockaddr *addr, 1116 int peer) 1117 { 1118 struct sock *sk = sock->sk; 1119 struct netlink_sock *nlk = nlk_sk(sk); 1120 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr); 1121 1122 nladdr->nl_family = AF_NETLINK; 1123 nladdr->nl_pad = 0; 1124 1125 if (peer) { 1126 nladdr->nl_pid = nlk->dst_portid; 1127 nladdr->nl_groups = netlink_group_mask(nlk->dst_group); 1128 } else { 1129 nladdr->nl_pid = nlk->portid; 1130 netlink_lock_table(); 1131 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0; 1132 netlink_unlock_table(); 1133 } 1134 return sizeof(*nladdr); 1135 } 1136 1137 static int netlink_ioctl(struct socket *sock, unsigned int cmd, 1138 unsigned long arg) 1139 { 1140 /* try to hand this ioctl down to the NIC drivers. 1141 */ 1142 return -ENOIOCTLCMD; 1143 } 1144 1145 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid) 1146 { 1147 struct sock *sock; 1148 struct netlink_sock *nlk; 1149 1150 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid); 1151 if (!sock) 1152 return ERR_PTR(-ECONNREFUSED); 1153 1154 /* Don't bother queuing skb if kernel socket has no input function */ 1155 nlk = nlk_sk(sock); 1156 if (sock->sk_state == NETLINK_CONNECTED && 1157 nlk->dst_portid != nlk_sk(ssk)->portid) { 1158 sock_put(sock); 1159 return ERR_PTR(-ECONNREFUSED); 1160 } 1161 return sock; 1162 } 1163 1164 struct sock *netlink_getsockbyfilp(struct file *filp) 1165 { 1166 struct inode *inode = file_inode(filp); 1167 struct sock *sock; 1168 1169 if (!S_ISSOCK(inode->i_mode)) 1170 return ERR_PTR(-ENOTSOCK); 1171 1172 sock = SOCKET_I(inode)->sk; 1173 if (sock->sk_family != AF_NETLINK) 1174 return ERR_PTR(-EINVAL); 1175 1176 sock_hold(sock); 1177 return sock; 1178 } 1179 1180 static struct sk_buff *netlink_alloc_large_skb(unsigned int size, 1181 int broadcast) 1182 { 1183 struct sk_buff *skb; 1184 void *data; 1185 1186 if (size <= NLMSG_GOODSIZE || broadcast) 1187 return alloc_skb(size, GFP_KERNEL); 1188 1189 size = SKB_DATA_ALIGN(size) + 1190 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 1191 1192 data = vmalloc(size); 1193 if (data == NULL) 1194 return NULL; 1195 1196 skb = __build_skb(data, size); 1197 if (skb == NULL) 1198 vfree(data); 1199 else 1200 skb->destructor = netlink_skb_destructor; 1201 1202 return skb; 1203 } 1204 1205 /* 1206 * Attach a skb to a netlink socket. 1207 * The caller must hold a reference to the destination socket. On error, the 1208 * reference is dropped. The skb is not send to the destination, just all 1209 * all error checks are performed and memory in the queue is reserved. 1210 * Return values: 1211 * < 0: error. skb freed, reference to sock dropped. 1212 * 0: continue 1213 * 1: repeat lookup - reference dropped while waiting for socket memory. 1214 */ 1215 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, 1216 long *timeo, struct sock *ssk) 1217 { 1218 struct netlink_sock *nlk; 1219 1220 nlk = nlk_sk(sk); 1221 1222 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 1223 test_bit(NETLINK_S_CONGESTED, &nlk->state))) { 1224 DECLARE_WAITQUEUE(wait, current); 1225 if (!*timeo) { 1226 if (!ssk || netlink_is_kernel(ssk)) 1227 netlink_overrun(sk); 1228 sock_put(sk); 1229 kfree_skb(skb); 1230 return -EAGAIN; 1231 } 1232 1233 __set_current_state(TASK_INTERRUPTIBLE); 1234 add_wait_queue(&nlk->wait, &wait); 1235 1236 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 1237 test_bit(NETLINK_S_CONGESTED, &nlk->state)) && 1238 !sock_flag(sk, SOCK_DEAD)) 1239 *timeo = schedule_timeout(*timeo); 1240 1241 __set_current_state(TASK_RUNNING); 1242 remove_wait_queue(&nlk->wait, &wait); 1243 sock_put(sk); 1244 1245 if (signal_pending(current)) { 1246 kfree_skb(skb); 1247 return sock_intr_errno(*timeo); 1248 } 1249 return 1; 1250 } 1251 netlink_skb_set_owner_r(skb, sk); 1252 return 0; 1253 } 1254 1255 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb) 1256 { 1257 int len = skb->len; 1258 1259 netlink_deliver_tap(sock_net(sk), skb); 1260 1261 skb_queue_tail(&sk->sk_receive_queue, skb); 1262 sk->sk_data_ready(sk); 1263 return len; 1264 } 1265 1266 int netlink_sendskb(struct sock *sk, struct sk_buff *skb) 1267 { 1268 int len = __netlink_sendskb(sk, skb); 1269 1270 sock_put(sk); 1271 return len; 1272 } 1273 1274 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 1275 { 1276 kfree_skb(skb); 1277 sock_put(sk); 1278 } 1279 1280 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation) 1281 { 1282 int delta; 1283 1284 WARN_ON(skb->sk != NULL); 1285 delta = skb->end - skb->tail; 1286 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize) 1287 return skb; 1288 1289 if (skb_shared(skb)) { 1290 struct sk_buff *nskb = skb_clone(skb, allocation); 1291 if (!nskb) 1292 return skb; 1293 consume_skb(skb); 1294 skb = nskb; 1295 } 1296 1297 pskb_expand_head(skb, 0, -delta, 1298 (allocation & ~__GFP_DIRECT_RECLAIM) | 1299 __GFP_NOWARN | __GFP_NORETRY); 1300 return skb; 1301 } 1302 1303 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb, 1304 struct sock *ssk) 1305 { 1306 int ret; 1307 struct netlink_sock *nlk = nlk_sk(sk); 1308 1309 ret = -ECONNREFUSED; 1310 if (nlk->netlink_rcv != NULL) { 1311 ret = skb->len; 1312 netlink_skb_set_owner_r(skb, sk); 1313 NETLINK_CB(skb).sk = ssk; 1314 netlink_deliver_tap_kernel(sk, ssk, skb); 1315 nlk->netlink_rcv(skb); 1316 consume_skb(skb); 1317 } else { 1318 kfree_skb(skb); 1319 } 1320 sock_put(sk); 1321 return ret; 1322 } 1323 1324 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, 1325 u32 portid, int nonblock) 1326 { 1327 struct sock *sk; 1328 int err; 1329 long timeo; 1330 1331 skb = netlink_trim(skb, gfp_any()); 1332 1333 timeo = sock_sndtimeo(ssk, nonblock); 1334 retry: 1335 sk = netlink_getsockbyportid(ssk, portid); 1336 if (IS_ERR(sk)) { 1337 kfree_skb(skb); 1338 return PTR_ERR(sk); 1339 } 1340 if (netlink_is_kernel(sk)) 1341 return netlink_unicast_kernel(sk, skb, ssk); 1342 1343 if (sk_filter(sk, skb)) { 1344 err = skb->len; 1345 kfree_skb(skb); 1346 sock_put(sk); 1347 return err; 1348 } 1349 1350 err = netlink_attachskb(sk, skb, &timeo, ssk); 1351 if (err == 1) 1352 goto retry; 1353 if (err) 1354 return err; 1355 1356 return netlink_sendskb(sk, skb); 1357 } 1358 EXPORT_SYMBOL(netlink_unicast); 1359 1360 int netlink_has_listeners(struct sock *sk, unsigned int group) 1361 { 1362 int res = 0; 1363 struct listeners *listeners; 1364 1365 BUG_ON(!netlink_is_kernel(sk)); 1366 1367 rcu_read_lock(); 1368 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners); 1369 1370 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups) 1371 res = test_bit(group - 1, listeners->masks); 1372 1373 rcu_read_unlock(); 1374 1375 return res; 1376 } 1377 EXPORT_SYMBOL_GPL(netlink_has_listeners); 1378 1379 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 1380 { 1381 struct netlink_sock *nlk = nlk_sk(sk); 1382 1383 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 1384 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) { 1385 netlink_skb_set_owner_r(skb, sk); 1386 __netlink_sendskb(sk, skb); 1387 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1); 1388 } 1389 return -1; 1390 } 1391 1392 struct netlink_broadcast_data { 1393 struct sock *exclude_sk; 1394 struct net *net; 1395 u32 portid; 1396 u32 group; 1397 int failure; 1398 int delivery_failure; 1399 int congested; 1400 int delivered; 1401 gfp_t allocation; 1402 struct sk_buff *skb, *skb2; 1403 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data); 1404 void *tx_data; 1405 }; 1406 1407 static void do_one_broadcast(struct sock *sk, 1408 struct netlink_broadcast_data *p) 1409 { 1410 struct netlink_sock *nlk = nlk_sk(sk); 1411 int val; 1412 1413 if (p->exclude_sk == sk) 1414 return; 1415 1416 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 1417 !test_bit(p->group - 1, nlk->groups)) 1418 return; 1419 1420 if (!net_eq(sock_net(sk), p->net)) { 1421 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID)) 1422 return; 1423 1424 if (!peernet_has_id(sock_net(sk), p->net)) 1425 return; 1426 1427 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns, 1428 CAP_NET_BROADCAST)) 1429 return; 1430 } 1431 1432 if (p->failure) { 1433 netlink_overrun(sk); 1434 return; 1435 } 1436 1437 sock_hold(sk); 1438 if (p->skb2 == NULL) { 1439 if (skb_shared(p->skb)) { 1440 p->skb2 = skb_clone(p->skb, p->allocation); 1441 } else { 1442 p->skb2 = skb_get(p->skb); 1443 /* 1444 * skb ownership may have been set when 1445 * delivered to a previous socket. 1446 */ 1447 skb_orphan(p->skb2); 1448 } 1449 } 1450 if (p->skb2 == NULL) { 1451 netlink_overrun(sk); 1452 /* Clone failed. Notify ALL listeners. */ 1453 p->failure = 1; 1454 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR) 1455 p->delivery_failure = 1; 1456 goto out; 1457 } 1458 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) { 1459 kfree_skb(p->skb2); 1460 p->skb2 = NULL; 1461 goto out; 1462 } 1463 if (sk_filter(sk, p->skb2)) { 1464 kfree_skb(p->skb2); 1465 p->skb2 = NULL; 1466 goto out; 1467 } 1468 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net); 1469 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED) 1470 NETLINK_CB(p->skb2).nsid_is_set = true; 1471 val = netlink_broadcast_deliver(sk, p->skb2); 1472 if (val < 0) { 1473 netlink_overrun(sk); 1474 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR) 1475 p->delivery_failure = 1; 1476 } else { 1477 p->congested |= val; 1478 p->delivered = 1; 1479 p->skb2 = NULL; 1480 } 1481 out: 1482 sock_put(sk); 1483 } 1484 1485 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid, 1486 u32 group, gfp_t allocation, 1487 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data), 1488 void *filter_data) 1489 { 1490 struct net *net = sock_net(ssk); 1491 struct netlink_broadcast_data info; 1492 struct sock *sk; 1493 1494 skb = netlink_trim(skb, allocation); 1495 1496 info.exclude_sk = ssk; 1497 info.net = net; 1498 info.portid = portid; 1499 info.group = group; 1500 info.failure = 0; 1501 info.delivery_failure = 0; 1502 info.congested = 0; 1503 info.delivered = 0; 1504 info.allocation = allocation; 1505 info.skb = skb; 1506 info.skb2 = NULL; 1507 info.tx_filter = filter; 1508 info.tx_data = filter_data; 1509 1510 /* While we sleep in clone, do not allow to change socket list */ 1511 1512 netlink_lock_table(); 1513 1514 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) 1515 do_one_broadcast(sk, &info); 1516 1517 consume_skb(skb); 1518 1519 netlink_unlock_table(); 1520 1521 if (info.delivery_failure) { 1522 kfree_skb(info.skb2); 1523 return -ENOBUFS; 1524 } 1525 consume_skb(info.skb2); 1526 1527 if (info.delivered) { 1528 if (info.congested && gfpflags_allow_blocking(allocation)) 1529 yield(); 1530 return 0; 1531 } 1532 return -ESRCH; 1533 } 1534 EXPORT_SYMBOL(netlink_broadcast_filtered); 1535 1536 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid, 1537 u32 group, gfp_t allocation) 1538 { 1539 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation, 1540 NULL, NULL); 1541 } 1542 EXPORT_SYMBOL(netlink_broadcast); 1543 1544 struct netlink_set_err_data { 1545 struct sock *exclude_sk; 1546 u32 portid; 1547 u32 group; 1548 int code; 1549 }; 1550 1551 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p) 1552 { 1553 struct netlink_sock *nlk = nlk_sk(sk); 1554 int ret = 0; 1555 1556 if (sk == p->exclude_sk) 1557 goto out; 1558 1559 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk))) 1560 goto out; 1561 1562 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 1563 !test_bit(p->group - 1, nlk->groups)) 1564 goto out; 1565 1566 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) { 1567 ret = 1; 1568 goto out; 1569 } 1570 1571 sk->sk_err = p->code; 1572 sk->sk_error_report(sk); 1573 out: 1574 return ret; 1575 } 1576 1577 /** 1578 * netlink_set_err - report error to broadcast listeners 1579 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create() 1580 * @portid: the PORTID of a process that we want to skip (if any) 1581 * @group: the broadcast group that will notice the error 1582 * @code: error code, must be negative (as usual in kernelspace) 1583 * 1584 * This function returns the number of broadcast listeners that have set the 1585 * NETLINK_NO_ENOBUFS socket option. 1586 */ 1587 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code) 1588 { 1589 struct netlink_set_err_data info; 1590 struct sock *sk; 1591 int ret = 0; 1592 1593 info.exclude_sk = ssk; 1594 info.portid = portid; 1595 info.group = group; 1596 /* sk->sk_err wants a positive error value */ 1597 info.code = -code; 1598 1599 read_lock(&nl_table_lock); 1600 1601 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) 1602 ret += do_one_set_err(sk, &info); 1603 1604 read_unlock(&nl_table_lock); 1605 return ret; 1606 } 1607 EXPORT_SYMBOL(netlink_set_err); 1608 1609 /* must be called with netlink table grabbed */ 1610 static void netlink_update_socket_mc(struct netlink_sock *nlk, 1611 unsigned int group, 1612 int is_new) 1613 { 1614 int old, new = !!is_new, subscriptions; 1615 1616 old = test_bit(group - 1, nlk->groups); 1617 subscriptions = nlk->subscriptions - old + new; 1618 if (new) 1619 __set_bit(group - 1, nlk->groups); 1620 else 1621 __clear_bit(group - 1, nlk->groups); 1622 netlink_update_subscriptions(&nlk->sk, subscriptions); 1623 netlink_update_listeners(&nlk->sk); 1624 } 1625 1626 static int netlink_setsockopt(struct socket *sock, int level, int optname, 1627 char __user *optval, unsigned int optlen) 1628 { 1629 struct sock *sk = sock->sk; 1630 struct netlink_sock *nlk = nlk_sk(sk); 1631 unsigned int val = 0; 1632 int err; 1633 1634 if (level != SOL_NETLINK) 1635 return -ENOPROTOOPT; 1636 1637 if (optlen >= sizeof(int) && 1638 get_user(val, (unsigned int __user *)optval)) 1639 return -EFAULT; 1640 1641 switch (optname) { 1642 case NETLINK_PKTINFO: 1643 if (val) 1644 nlk->flags |= NETLINK_F_RECV_PKTINFO; 1645 else 1646 nlk->flags &= ~NETLINK_F_RECV_PKTINFO; 1647 err = 0; 1648 break; 1649 case NETLINK_ADD_MEMBERSHIP: 1650 case NETLINK_DROP_MEMBERSHIP: { 1651 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV)) 1652 return -EPERM; 1653 err = netlink_realloc_groups(sk); 1654 if (err) 1655 return err; 1656 if (!val || val - 1 >= nlk->ngroups) 1657 return -EINVAL; 1658 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) { 1659 err = nlk->netlink_bind(sock_net(sk), val); 1660 if (err) 1661 return err; 1662 } 1663 netlink_table_grab(); 1664 netlink_update_socket_mc(nlk, val, 1665 optname == NETLINK_ADD_MEMBERSHIP); 1666 netlink_table_ungrab(); 1667 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind) 1668 nlk->netlink_unbind(sock_net(sk), val); 1669 1670 err = 0; 1671 break; 1672 } 1673 case NETLINK_BROADCAST_ERROR: 1674 if (val) 1675 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR; 1676 else 1677 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR; 1678 err = 0; 1679 break; 1680 case NETLINK_NO_ENOBUFS: 1681 if (val) { 1682 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS; 1683 clear_bit(NETLINK_S_CONGESTED, &nlk->state); 1684 wake_up_interruptible(&nlk->wait); 1685 } else { 1686 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS; 1687 } 1688 err = 0; 1689 break; 1690 case NETLINK_LISTEN_ALL_NSID: 1691 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST)) 1692 return -EPERM; 1693 1694 if (val) 1695 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID; 1696 else 1697 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID; 1698 err = 0; 1699 break; 1700 case NETLINK_CAP_ACK: 1701 if (val) 1702 nlk->flags |= NETLINK_F_CAP_ACK; 1703 else 1704 nlk->flags &= ~NETLINK_F_CAP_ACK; 1705 err = 0; 1706 break; 1707 case NETLINK_EXT_ACK: 1708 if (val) 1709 nlk->flags |= NETLINK_F_EXT_ACK; 1710 else 1711 nlk->flags &= ~NETLINK_F_EXT_ACK; 1712 err = 0; 1713 break; 1714 default: 1715 err = -ENOPROTOOPT; 1716 } 1717 return err; 1718 } 1719 1720 static int netlink_getsockopt(struct socket *sock, int level, int optname, 1721 char __user *optval, int __user *optlen) 1722 { 1723 struct sock *sk = sock->sk; 1724 struct netlink_sock *nlk = nlk_sk(sk); 1725 int len, val, err; 1726 1727 if (level != SOL_NETLINK) 1728 return -ENOPROTOOPT; 1729 1730 if (get_user(len, optlen)) 1731 return -EFAULT; 1732 if (len < 0) 1733 return -EINVAL; 1734 1735 switch (optname) { 1736 case NETLINK_PKTINFO: 1737 if (len < sizeof(int)) 1738 return -EINVAL; 1739 len = sizeof(int); 1740 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0; 1741 if (put_user(len, optlen) || 1742 put_user(val, optval)) 1743 return -EFAULT; 1744 err = 0; 1745 break; 1746 case NETLINK_BROADCAST_ERROR: 1747 if (len < sizeof(int)) 1748 return -EINVAL; 1749 len = sizeof(int); 1750 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0; 1751 if (put_user(len, optlen) || 1752 put_user(val, optval)) 1753 return -EFAULT; 1754 err = 0; 1755 break; 1756 case NETLINK_NO_ENOBUFS: 1757 if (len < sizeof(int)) 1758 return -EINVAL; 1759 len = sizeof(int); 1760 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0; 1761 if (put_user(len, optlen) || 1762 put_user(val, optval)) 1763 return -EFAULT; 1764 err = 0; 1765 break; 1766 case NETLINK_LIST_MEMBERSHIPS: { 1767 int pos, idx, shift; 1768 1769 err = 0; 1770 netlink_lock_table(); 1771 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) { 1772 if (len - pos < sizeof(u32)) 1773 break; 1774 1775 idx = pos / sizeof(unsigned long); 1776 shift = (pos % sizeof(unsigned long)) * 8; 1777 if (put_user((u32)(nlk->groups[idx] >> shift), 1778 (u32 __user *)(optval + pos))) { 1779 err = -EFAULT; 1780 break; 1781 } 1782 } 1783 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen)) 1784 err = -EFAULT; 1785 netlink_unlock_table(); 1786 break; 1787 } 1788 case NETLINK_CAP_ACK: 1789 if (len < sizeof(int)) 1790 return -EINVAL; 1791 len = sizeof(int); 1792 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0; 1793 if (put_user(len, optlen) || 1794 put_user(val, optval)) 1795 return -EFAULT; 1796 err = 0; 1797 break; 1798 case NETLINK_EXT_ACK: 1799 if (len < sizeof(int)) 1800 return -EINVAL; 1801 len = sizeof(int); 1802 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0; 1803 if (put_user(len, optlen) || put_user(val, optval)) 1804 return -EFAULT; 1805 err = 0; 1806 break; 1807 default: 1808 err = -ENOPROTOOPT; 1809 } 1810 return err; 1811 } 1812 1813 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb) 1814 { 1815 struct nl_pktinfo info; 1816 1817 info.group = NETLINK_CB(skb).dst_group; 1818 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info); 1819 } 1820 1821 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg, 1822 struct sk_buff *skb) 1823 { 1824 if (!NETLINK_CB(skb).nsid_is_set) 1825 return; 1826 1827 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int), 1828 &NETLINK_CB(skb).nsid); 1829 } 1830 1831 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1832 { 1833 struct sock *sk = sock->sk; 1834 struct netlink_sock *nlk = nlk_sk(sk); 1835 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name); 1836 u32 dst_portid; 1837 u32 dst_group; 1838 struct sk_buff *skb; 1839 int err; 1840 struct scm_cookie scm; 1841 u32 netlink_skb_flags = 0; 1842 1843 if (msg->msg_flags&MSG_OOB) 1844 return -EOPNOTSUPP; 1845 1846 err = scm_send(sock, msg, &scm, true); 1847 if (err < 0) 1848 return err; 1849 1850 if (msg->msg_namelen) { 1851 err = -EINVAL; 1852 if (msg->msg_namelen < sizeof(struct sockaddr_nl)) 1853 goto out; 1854 if (addr->nl_family != AF_NETLINK) 1855 goto out; 1856 dst_portid = addr->nl_pid; 1857 dst_group = ffs(addr->nl_groups); 1858 err = -EPERM; 1859 if ((dst_group || dst_portid) && 1860 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND)) 1861 goto out; 1862 netlink_skb_flags |= NETLINK_SKB_DST; 1863 } else { 1864 dst_portid = nlk->dst_portid; 1865 dst_group = nlk->dst_group; 1866 } 1867 1868 if (!nlk->bound) { 1869 err = netlink_autobind(sock); 1870 if (err) 1871 goto out; 1872 } else { 1873 /* Ensure nlk is hashed and visible. */ 1874 smp_rmb(); 1875 } 1876 1877 err = -EMSGSIZE; 1878 if (len > sk->sk_sndbuf - 32) 1879 goto out; 1880 err = -ENOBUFS; 1881 skb = netlink_alloc_large_skb(len, dst_group); 1882 if (skb == NULL) 1883 goto out; 1884 1885 NETLINK_CB(skb).portid = nlk->portid; 1886 NETLINK_CB(skb).dst_group = dst_group; 1887 NETLINK_CB(skb).creds = scm.creds; 1888 NETLINK_CB(skb).flags = netlink_skb_flags; 1889 1890 err = -EFAULT; 1891 if (memcpy_from_msg(skb_put(skb, len), msg, len)) { 1892 kfree_skb(skb); 1893 goto out; 1894 } 1895 1896 err = security_netlink_send(sk, skb); 1897 if (err) { 1898 kfree_skb(skb); 1899 goto out; 1900 } 1901 1902 if (dst_group) { 1903 refcount_inc(&skb->users); 1904 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL); 1905 } 1906 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT); 1907 1908 out: 1909 scm_destroy(&scm); 1910 return err; 1911 } 1912 1913 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1914 int flags) 1915 { 1916 struct scm_cookie scm; 1917 struct sock *sk = sock->sk; 1918 struct netlink_sock *nlk = nlk_sk(sk); 1919 int noblock = flags&MSG_DONTWAIT; 1920 size_t copied; 1921 struct sk_buff *skb, *data_skb; 1922 int err, ret; 1923 1924 if (flags&MSG_OOB) 1925 return -EOPNOTSUPP; 1926 1927 copied = 0; 1928 1929 skb = skb_recv_datagram(sk, flags, noblock, &err); 1930 if (skb == NULL) 1931 goto out; 1932 1933 data_skb = skb; 1934 1935 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES 1936 if (unlikely(skb_shinfo(skb)->frag_list)) { 1937 /* 1938 * If this skb has a frag_list, then here that means that we 1939 * will have to use the frag_list skb's data for compat tasks 1940 * and the regular skb's data for normal (non-compat) tasks. 1941 * 1942 * If we need to send the compat skb, assign it to the 1943 * 'data_skb' variable so that it will be used below for data 1944 * copying. We keep 'skb' for everything else, including 1945 * freeing both later. 1946 */ 1947 if (flags & MSG_CMSG_COMPAT) 1948 data_skb = skb_shinfo(skb)->frag_list; 1949 } 1950 #endif 1951 1952 /* Record the max length of recvmsg() calls for future allocations */ 1953 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len); 1954 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len, 1955 SKB_WITH_OVERHEAD(32768)); 1956 1957 copied = data_skb->len; 1958 if (len < copied) { 1959 msg->msg_flags |= MSG_TRUNC; 1960 copied = len; 1961 } 1962 1963 skb_reset_transport_header(data_skb); 1964 err = skb_copy_datagram_msg(data_skb, 0, msg, copied); 1965 1966 if (msg->msg_name) { 1967 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name); 1968 addr->nl_family = AF_NETLINK; 1969 addr->nl_pad = 0; 1970 addr->nl_pid = NETLINK_CB(skb).portid; 1971 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group); 1972 msg->msg_namelen = sizeof(*addr); 1973 } 1974 1975 if (nlk->flags & NETLINK_F_RECV_PKTINFO) 1976 netlink_cmsg_recv_pktinfo(msg, skb); 1977 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID) 1978 netlink_cmsg_listen_all_nsid(sk, msg, skb); 1979 1980 memset(&scm, 0, sizeof(scm)); 1981 scm.creds = *NETLINK_CREDS(skb); 1982 if (flags & MSG_TRUNC) 1983 copied = data_skb->len; 1984 1985 skb_free_datagram(sk, skb); 1986 1987 if (nlk->cb_running && 1988 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) { 1989 ret = netlink_dump(sk); 1990 if (ret) { 1991 sk->sk_err = -ret; 1992 sk->sk_error_report(sk); 1993 } 1994 } 1995 1996 scm_recv(sock, msg, &scm, flags); 1997 out: 1998 netlink_rcv_wake(sk); 1999 return err ? : copied; 2000 } 2001 2002 static void netlink_data_ready(struct sock *sk) 2003 { 2004 BUG(); 2005 } 2006 2007 /* 2008 * We export these functions to other modules. They provide a 2009 * complete set of kernel non-blocking support for message 2010 * queueing. 2011 */ 2012 2013 struct sock * 2014 __netlink_kernel_create(struct net *net, int unit, struct module *module, 2015 struct netlink_kernel_cfg *cfg) 2016 { 2017 struct socket *sock; 2018 struct sock *sk; 2019 struct netlink_sock *nlk; 2020 struct listeners *listeners = NULL; 2021 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL; 2022 unsigned int groups; 2023 2024 BUG_ON(!nl_table); 2025 2026 if (unit < 0 || unit >= MAX_LINKS) 2027 return NULL; 2028 2029 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 2030 return NULL; 2031 2032 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0) 2033 goto out_sock_release_nosk; 2034 2035 sk = sock->sk; 2036 2037 if (!cfg || cfg->groups < 32) 2038 groups = 32; 2039 else 2040 groups = cfg->groups; 2041 2042 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 2043 if (!listeners) 2044 goto out_sock_release; 2045 2046 sk->sk_data_ready = netlink_data_ready; 2047 if (cfg && cfg->input) 2048 nlk_sk(sk)->netlink_rcv = cfg->input; 2049 2050 if (netlink_insert(sk, 0)) 2051 goto out_sock_release; 2052 2053 nlk = nlk_sk(sk); 2054 nlk->flags |= NETLINK_F_KERNEL_SOCKET; 2055 2056 netlink_table_grab(); 2057 if (!nl_table[unit].registered) { 2058 nl_table[unit].groups = groups; 2059 rcu_assign_pointer(nl_table[unit].listeners, listeners); 2060 nl_table[unit].cb_mutex = cb_mutex; 2061 nl_table[unit].module = module; 2062 if (cfg) { 2063 nl_table[unit].bind = cfg->bind; 2064 nl_table[unit].unbind = cfg->unbind; 2065 nl_table[unit].flags = cfg->flags; 2066 if (cfg->compare) 2067 nl_table[unit].compare = cfg->compare; 2068 } 2069 nl_table[unit].registered = 1; 2070 } else { 2071 kfree(listeners); 2072 nl_table[unit].registered++; 2073 } 2074 netlink_table_ungrab(); 2075 return sk; 2076 2077 out_sock_release: 2078 kfree(listeners); 2079 netlink_kernel_release(sk); 2080 return NULL; 2081 2082 out_sock_release_nosk: 2083 sock_release(sock); 2084 return NULL; 2085 } 2086 EXPORT_SYMBOL(__netlink_kernel_create); 2087 2088 void 2089 netlink_kernel_release(struct sock *sk) 2090 { 2091 if (sk == NULL || sk->sk_socket == NULL) 2092 return; 2093 2094 sock_release(sk->sk_socket); 2095 } 2096 EXPORT_SYMBOL(netlink_kernel_release); 2097 2098 int __netlink_change_ngroups(struct sock *sk, unsigned int groups) 2099 { 2100 struct listeners *new, *old; 2101 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 2102 2103 if (groups < 32) 2104 groups = 32; 2105 2106 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) { 2107 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC); 2108 if (!new) 2109 return -ENOMEM; 2110 old = nl_deref_protected(tbl->listeners); 2111 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups)); 2112 rcu_assign_pointer(tbl->listeners, new); 2113 2114 kfree_rcu(old, rcu); 2115 } 2116 tbl->groups = groups; 2117 2118 return 0; 2119 } 2120 2121 /** 2122 * netlink_change_ngroups - change number of multicast groups 2123 * 2124 * This changes the number of multicast groups that are available 2125 * on a certain netlink family. Note that it is not possible to 2126 * change the number of groups to below 32. Also note that it does 2127 * not implicitly call netlink_clear_multicast_users() when the 2128 * number of groups is reduced. 2129 * 2130 * @sk: The kernel netlink socket, as returned by netlink_kernel_create(). 2131 * @groups: The new number of groups. 2132 */ 2133 int netlink_change_ngroups(struct sock *sk, unsigned int groups) 2134 { 2135 int err; 2136 2137 netlink_table_grab(); 2138 err = __netlink_change_ngroups(sk, groups); 2139 netlink_table_ungrab(); 2140 2141 return err; 2142 } 2143 2144 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group) 2145 { 2146 struct sock *sk; 2147 struct netlink_table *tbl = &nl_table[ksk->sk_protocol]; 2148 2149 sk_for_each_bound(sk, &tbl->mc_list) 2150 netlink_update_socket_mc(nlk_sk(sk), group, 0); 2151 } 2152 2153 struct nlmsghdr * 2154 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags) 2155 { 2156 struct nlmsghdr *nlh; 2157 int size = nlmsg_msg_size(len); 2158 2159 nlh = skb_put(skb, NLMSG_ALIGN(size)); 2160 nlh->nlmsg_type = type; 2161 nlh->nlmsg_len = size; 2162 nlh->nlmsg_flags = flags; 2163 nlh->nlmsg_pid = portid; 2164 nlh->nlmsg_seq = seq; 2165 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0) 2166 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size); 2167 return nlh; 2168 } 2169 EXPORT_SYMBOL(__nlmsg_put); 2170 2171 /* 2172 * It looks a bit ugly. 2173 * It would be better to create kernel thread. 2174 */ 2175 2176 static int netlink_dump(struct sock *sk) 2177 { 2178 struct netlink_sock *nlk = nlk_sk(sk); 2179 struct netlink_callback *cb; 2180 struct sk_buff *skb = NULL; 2181 struct nlmsghdr *nlh; 2182 struct module *module; 2183 int err = -ENOBUFS; 2184 int alloc_min_size; 2185 int alloc_size; 2186 2187 mutex_lock(nlk->cb_mutex); 2188 if (!nlk->cb_running) { 2189 err = -EINVAL; 2190 goto errout_skb; 2191 } 2192 2193 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 2194 goto errout_skb; 2195 2196 /* NLMSG_GOODSIZE is small to avoid high order allocations being 2197 * required, but it makes sense to _attempt_ a 16K bytes allocation 2198 * to reduce number of system calls on dump operations, if user 2199 * ever provided a big enough buffer. 2200 */ 2201 cb = &nlk->cb; 2202 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE); 2203 2204 if (alloc_min_size < nlk->max_recvmsg_len) { 2205 alloc_size = nlk->max_recvmsg_len; 2206 skb = alloc_skb(alloc_size, 2207 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) | 2208 __GFP_NOWARN | __GFP_NORETRY); 2209 } 2210 if (!skb) { 2211 alloc_size = alloc_min_size; 2212 skb = alloc_skb(alloc_size, GFP_KERNEL); 2213 } 2214 if (!skb) 2215 goto errout_skb; 2216 2217 /* Trim skb to allocated size. User is expected to provide buffer as 2218 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at 2219 * netlink_recvmsg())). dump will pack as many smaller messages as 2220 * could fit within the allocated skb. skb is typically allocated 2221 * with larger space than required (could be as much as near 2x the 2222 * requested size with align to next power of 2 approach). Allowing 2223 * dump to use the excess space makes it difficult for a user to have a 2224 * reasonable static buffer based on the expected largest dump of a 2225 * single netdev. The outcome is MSG_TRUNC error. 2226 */ 2227 skb_reserve(skb, skb_tailroom(skb) - alloc_size); 2228 netlink_skb_set_owner_r(skb, sk); 2229 2230 if (nlk->dump_done_errno > 0) 2231 nlk->dump_done_errno = cb->dump(skb, cb); 2232 2233 if (nlk->dump_done_errno > 0 || 2234 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) { 2235 mutex_unlock(nlk->cb_mutex); 2236 2237 if (sk_filter(sk, skb)) 2238 kfree_skb(skb); 2239 else 2240 __netlink_sendskb(sk, skb); 2241 return 0; 2242 } 2243 2244 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, 2245 sizeof(nlk->dump_done_errno), NLM_F_MULTI); 2246 if (WARN_ON(!nlh)) 2247 goto errout_skb; 2248 2249 nl_dump_check_consistent(cb, nlh); 2250 2251 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, 2252 sizeof(nlk->dump_done_errno)); 2253 2254 if (sk_filter(sk, skb)) 2255 kfree_skb(skb); 2256 else 2257 __netlink_sendskb(sk, skb); 2258 2259 if (cb->done) 2260 cb->done(cb); 2261 2262 nlk->cb_running = false; 2263 module = cb->module; 2264 skb = cb->skb; 2265 mutex_unlock(nlk->cb_mutex); 2266 module_put(module); 2267 consume_skb(skb); 2268 return 0; 2269 2270 errout_skb: 2271 mutex_unlock(nlk->cb_mutex); 2272 kfree_skb(skb); 2273 return err; 2274 } 2275 2276 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 2277 const struct nlmsghdr *nlh, 2278 struct netlink_dump_control *control) 2279 { 2280 struct netlink_callback *cb; 2281 struct sock *sk; 2282 struct netlink_sock *nlk; 2283 int ret; 2284 2285 refcount_inc(&skb->users); 2286 2287 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid); 2288 if (sk == NULL) { 2289 ret = -ECONNREFUSED; 2290 goto error_free; 2291 } 2292 2293 nlk = nlk_sk(sk); 2294 mutex_lock(nlk->cb_mutex); 2295 /* A dump is in progress... */ 2296 if (nlk->cb_running) { 2297 ret = -EBUSY; 2298 goto error_unlock; 2299 } 2300 /* add reference of module which cb->dump belongs to */ 2301 if (!try_module_get(control->module)) { 2302 ret = -EPROTONOSUPPORT; 2303 goto error_unlock; 2304 } 2305 2306 cb = &nlk->cb; 2307 memset(cb, 0, sizeof(*cb)); 2308 cb->dump = control->dump; 2309 cb->done = control->done; 2310 cb->nlh = nlh; 2311 cb->data = control->data; 2312 cb->module = control->module; 2313 cb->min_dump_alloc = control->min_dump_alloc; 2314 cb->skb = skb; 2315 2316 if (control->start) { 2317 ret = control->start(cb); 2318 if (ret) 2319 goto error_put; 2320 } 2321 2322 nlk->cb_running = true; 2323 nlk->dump_done_errno = INT_MAX; 2324 2325 mutex_unlock(nlk->cb_mutex); 2326 2327 ret = netlink_dump(sk); 2328 2329 sock_put(sk); 2330 2331 if (ret) 2332 return ret; 2333 2334 /* We successfully started a dump, by returning -EINTR we 2335 * signal not to send ACK even if it was requested. 2336 */ 2337 return -EINTR; 2338 2339 error_put: 2340 module_put(control->module); 2341 error_unlock: 2342 sock_put(sk); 2343 mutex_unlock(nlk->cb_mutex); 2344 error_free: 2345 kfree_skb(skb); 2346 return ret; 2347 } 2348 EXPORT_SYMBOL(__netlink_dump_start); 2349 2350 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err, 2351 const struct netlink_ext_ack *extack) 2352 { 2353 struct sk_buff *skb; 2354 struct nlmsghdr *rep; 2355 struct nlmsgerr *errmsg; 2356 size_t payload = sizeof(*errmsg); 2357 size_t tlvlen = 0; 2358 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk); 2359 unsigned int flags = 0; 2360 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK; 2361 2362 /* Error messages get the original request appened, unless the user 2363 * requests to cap the error message, and get extra error data if 2364 * requested. 2365 */ 2366 if (nlk_has_extack && extack && extack->_msg) 2367 tlvlen += nla_total_size(strlen(extack->_msg) + 1); 2368 2369 if (err) { 2370 if (!(nlk->flags & NETLINK_F_CAP_ACK)) 2371 payload += nlmsg_len(nlh); 2372 else 2373 flags |= NLM_F_CAPPED; 2374 if (nlk_has_extack && extack && extack->bad_attr) 2375 tlvlen += nla_total_size(sizeof(u32)); 2376 } else { 2377 flags |= NLM_F_CAPPED; 2378 2379 if (nlk_has_extack && extack && extack->cookie_len) 2380 tlvlen += nla_total_size(extack->cookie_len); 2381 } 2382 2383 if (tlvlen) 2384 flags |= NLM_F_ACK_TLVS; 2385 2386 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL); 2387 if (!skb) { 2388 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS; 2389 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk); 2390 return; 2391 } 2392 2393 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, 2394 NLMSG_ERROR, payload, flags); 2395 errmsg = nlmsg_data(rep); 2396 errmsg->error = err; 2397 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh)); 2398 2399 if (nlk_has_extack && extack) { 2400 if (extack->_msg) { 2401 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, 2402 extack->_msg)); 2403 } 2404 if (err) { 2405 if (extack->bad_attr && 2406 !WARN_ON((u8 *)extack->bad_attr < in_skb->data || 2407 (u8 *)extack->bad_attr >= in_skb->data + 2408 in_skb->len)) 2409 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS, 2410 (u8 *)extack->bad_attr - 2411 in_skb->data)); 2412 } else { 2413 if (extack->cookie_len) 2414 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE, 2415 extack->cookie_len, 2416 extack->cookie)); 2417 } 2418 } 2419 2420 nlmsg_end(skb, rep); 2421 2422 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT); 2423 } 2424 EXPORT_SYMBOL(netlink_ack); 2425 2426 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *, 2427 struct nlmsghdr *, 2428 struct netlink_ext_ack *)) 2429 { 2430 struct netlink_ext_ack extack; 2431 struct nlmsghdr *nlh; 2432 int err; 2433 2434 while (skb->len >= nlmsg_total_size(0)) { 2435 int msglen; 2436 2437 memset(&extack, 0, sizeof(extack)); 2438 nlh = nlmsg_hdr(skb); 2439 err = 0; 2440 2441 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len) 2442 return 0; 2443 2444 /* Only requests are handled by the kernel */ 2445 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) 2446 goto ack; 2447 2448 /* Skip control messages */ 2449 if (nlh->nlmsg_type < NLMSG_MIN_TYPE) 2450 goto ack; 2451 2452 err = cb(skb, nlh, &extack); 2453 if (err == -EINTR) 2454 goto skip; 2455 2456 ack: 2457 if (nlh->nlmsg_flags & NLM_F_ACK || err) 2458 netlink_ack(skb, nlh, err, &extack); 2459 2460 skip: 2461 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 2462 if (msglen > skb->len) 2463 msglen = skb->len; 2464 skb_pull(skb, msglen); 2465 } 2466 2467 return 0; 2468 } 2469 EXPORT_SYMBOL(netlink_rcv_skb); 2470 2471 /** 2472 * nlmsg_notify - send a notification netlink message 2473 * @sk: netlink socket to use 2474 * @skb: notification message 2475 * @portid: destination netlink portid for reports or 0 2476 * @group: destination multicast group or 0 2477 * @report: 1 to report back, 0 to disable 2478 * @flags: allocation flags 2479 */ 2480 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid, 2481 unsigned int group, int report, gfp_t flags) 2482 { 2483 int err = 0; 2484 2485 if (group) { 2486 int exclude_portid = 0; 2487 2488 if (report) { 2489 refcount_inc(&skb->users); 2490 exclude_portid = portid; 2491 } 2492 2493 /* errors reported via destination sk->sk_err, but propagate 2494 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */ 2495 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags); 2496 } 2497 2498 if (report) { 2499 int err2; 2500 2501 err2 = nlmsg_unicast(sk, skb, portid); 2502 if (!err || err == -ESRCH) 2503 err = err2; 2504 } 2505 2506 return err; 2507 } 2508 EXPORT_SYMBOL(nlmsg_notify); 2509 2510 #ifdef CONFIG_PROC_FS 2511 struct nl_seq_iter { 2512 struct seq_net_private p; 2513 struct rhashtable_iter hti; 2514 int link; 2515 }; 2516 2517 static int netlink_walk_start(struct nl_seq_iter *iter) 2518 { 2519 int err; 2520 2521 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti, 2522 GFP_KERNEL); 2523 if (err) { 2524 iter->link = MAX_LINKS; 2525 return err; 2526 } 2527 2528 rhashtable_walk_start(&iter->hti); 2529 2530 return 0; 2531 } 2532 2533 static void netlink_walk_stop(struct nl_seq_iter *iter) 2534 { 2535 rhashtable_walk_stop(&iter->hti); 2536 rhashtable_walk_exit(&iter->hti); 2537 } 2538 2539 static void *__netlink_seq_next(struct seq_file *seq) 2540 { 2541 struct nl_seq_iter *iter = seq->private; 2542 struct netlink_sock *nlk; 2543 2544 do { 2545 for (;;) { 2546 int err; 2547 2548 nlk = rhashtable_walk_next(&iter->hti); 2549 2550 if (IS_ERR(nlk)) { 2551 if (PTR_ERR(nlk) == -EAGAIN) 2552 continue; 2553 2554 return nlk; 2555 } 2556 2557 if (nlk) 2558 break; 2559 2560 netlink_walk_stop(iter); 2561 if (++iter->link >= MAX_LINKS) 2562 return NULL; 2563 2564 err = netlink_walk_start(iter); 2565 if (err) 2566 return ERR_PTR(err); 2567 } 2568 } while (sock_net(&nlk->sk) != seq_file_net(seq)); 2569 2570 return nlk; 2571 } 2572 2573 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp) 2574 { 2575 struct nl_seq_iter *iter = seq->private; 2576 void *obj = SEQ_START_TOKEN; 2577 loff_t pos; 2578 int err; 2579 2580 iter->link = 0; 2581 2582 err = netlink_walk_start(iter); 2583 if (err) 2584 return ERR_PTR(err); 2585 2586 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--) 2587 obj = __netlink_seq_next(seq); 2588 2589 return obj; 2590 } 2591 2592 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2593 { 2594 ++*pos; 2595 return __netlink_seq_next(seq); 2596 } 2597 2598 static void netlink_seq_stop(struct seq_file *seq, void *v) 2599 { 2600 struct nl_seq_iter *iter = seq->private; 2601 2602 if (iter->link >= MAX_LINKS) 2603 return; 2604 2605 netlink_walk_stop(iter); 2606 } 2607 2608 2609 static int netlink_seq_show(struct seq_file *seq, void *v) 2610 { 2611 if (v == SEQ_START_TOKEN) { 2612 seq_puts(seq, 2613 "sk Eth Pid Groups " 2614 "Rmem Wmem Dump Locks Drops Inode\n"); 2615 } else { 2616 struct sock *s = v; 2617 struct netlink_sock *nlk = nlk_sk(s); 2618 2619 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8d %-8lu\n", 2620 s, 2621 s->sk_protocol, 2622 nlk->portid, 2623 nlk->groups ? (u32)nlk->groups[0] : 0, 2624 sk_rmem_alloc_get(s), 2625 sk_wmem_alloc_get(s), 2626 nlk->cb_running, 2627 refcount_read(&s->sk_refcnt), 2628 atomic_read(&s->sk_drops), 2629 sock_i_ino(s) 2630 ); 2631 2632 } 2633 return 0; 2634 } 2635 2636 static const struct seq_operations netlink_seq_ops = { 2637 .start = netlink_seq_start, 2638 .next = netlink_seq_next, 2639 .stop = netlink_seq_stop, 2640 .show = netlink_seq_show, 2641 }; 2642 #endif 2643 2644 int netlink_register_notifier(struct notifier_block *nb) 2645 { 2646 return blocking_notifier_chain_register(&netlink_chain, nb); 2647 } 2648 EXPORT_SYMBOL(netlink_register_notifier); 2649 2650 int netlink_unregister_notifier(struct notifier_block *nb) 2651 { 2652 return blocking_notifier_chain_unregister(&netlink_chain, nb); 2653 } 2654 EXPORT_SYMBOL(netlink_unregister_notifier); 2655 2656 static const struct proto_ops netlink_ops = { 2657 .family = PF_NETLINK, 2658 .owner = THIS_MODULE, 2659 .release = netlink_release, 2660 .bind = netlink_bind, 2661 .connect = netlink_connect, 2662 .socketpair = sock_no_socketpair, 2663 .accept = sock_no_accept, 2664 .getname = netlink_getname, 2665 .poll = datagram_poll, 2666 .ioctl = netlink_ioctl, 2667 .listen = sock_no_listen, 2668 .shutdown = sock_no_shutdown, 2669 .setsockopt = netlink_setsockopt, 2670 .getsockopt = netlink_getsockopt, 2671 .sendmsg = netlink_sendmsg, 2672 .recvmsg = netlink_recvmsg, 2673 .mmap = sock_no_mmap, 2674 .sendpage = sock_no_sendpage, 2675 }; 2676 2677 static const struct net_proto_family netlink_family_ops = { 2678 .family = PF_NETLINK, 2679 .create = netlink_create, 2680 .owner = THIS_MODULE, /* for consistency 8) */ 2681 }; 2682 2683 static int __net_init netlink_net_init(struct net *net) 2684 { 2685 #ifdef CONFIG_PROC_FS 2686 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops, 2687 sizeof(struct nl_seq_iter))) 2688 return -ENOMEM; 2689 #endif 2690 return 0; 2691 } 2692 2693 static void __net_exit netlink_net_exit(struct net *net) 2694 { 2695 #ifdef CONFIG_PROC_FS 2696 remove_proc_entry("netlink", net->proc_net); 2697 #endif 2698 } 2699 2700 static void __init netlink_add_usersock_entry(void) 2701 { 2702 struct listeners *listeners; 2703 int groups = 32; 2704 2705 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 2706 if (!listeners) 2707 panic("netlink_add_usersock_entry: Cannot allocate listeners\n"); 2708 2709 netlink_table_grab(); 2710 2711 nl_table[NETLINK_USERSOCK].groups = groups; 2712 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners); 2713 nl_table[NETLINK_USERSOCK].module = THIS_MODULE; 2714 nl_table[NETLINK_USERSOCK].registered = 1; 2715 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND; 2716 2717 netlink_table_ungrab(); 2718 } 2719 2720 static struct pernet_operations __net_initdata netlink_net_ops = { 2721 .init = netlink_net_init, 2722 .exit = netlink_net_exit, 2723 }; 2724 2725 static inline u32 netlink_hash(const void *data, u32 len, u32 seed) 2726 { 2727 const struct netlink_sock *nlk = data; 2728 struct netlink_compare_arg arg; 2729 2730 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid); 2731 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed); 2732 } 2733 2734 static const struct rhashtable_params netlink_rhashtable_params = { 2735 .head_offset = offsetof(struct netlink_sock, node), 2736 .key_len = netlink_compare_arg_len, 2737 .obj_hashfn = netlink_hash, 2738 .obj_cmpfn = netlink_compare, 2739 .automatic_shrinking = true, 2740 }; 2741 2742 static int __init netlink_proto_init(void) 2743 { 2744 int i; 2745 int err = proto_register(&netlink_proto, 0); 2746 2747 if (err != 0) 2748 goto out; 2749 2750 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb)); 2751 2752 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL); 2753 if (!nl_table) 2754 goto panic; 2755 2756 for (i = 0; i < MAX_LINKS; i++) { 2757 if (rhashtable_init(&nl_table[i].hash, 2758 &netlink_rhashtable_params) < 0) { 2759 while (--i > 0) 2760 rhashtable_destroy(&nl_table[i].hash); 2761 kfree(nl_table); 2762 goto panic; 2763 } 2764 } 2765 2766 netlink_add_usersock_entry(); 2767 2768 sock_register(&netlink_family_ops); 2769 register_pernet_subsys(&netlink_net_ops); 2770 register_pernet_subsys(&netlink_tap_net_ops); 2771 /* The netlink device handler may be needed early. */ 2772 rtnetlink_init(); 2773 out: 2774 return err; 2775 panic: 2776 panic("netlink_init: Cannot allocate nl_table\n"); 2777 } 2778 2779 core_initcall(netlink_proto_init); 2780