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 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License 9 * as published by the Free Software Foundation; either version 10 * 2 of the License, or (at your option) any later version. 11 * 12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith 13 * added netlink_proto_exit 14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br> 15 * use nlk_sk, as sk->protinfo is on a diet 8) 16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org> 17 * - inc module use count of module that owns 18 * the kernel socket in case userspace opens 19 * socket of same protocol 20 * - remove all module support, since netlink is 21 * mandatory if CONFIG_NET=y these days 22 */ 23 24 #include <linux/module.h> 25 26 #include <linux/capability.h> 27 #include <linux/kernel.h> 28 #include <linux/init.h> 29 #include <linux/signal.h> 30 #include <linux/sched.h> 31 #include <linux/errno.h> 32 #include <linux/string.h> 33 #include <linux/stat.h> 34 #include <linux/socket.h> 35 #include <linux/un.h> 36 #include <linux/fcntl.h> 37 #include <linux/termios.h> 38 #include <linux/sockios.h> 39 #include <linux/net.h> 40 #include <linux/fs.h> 41 #include <linux/slab.h> 42 #include <asm/uaccess.h> 43 #include <linux/skbuff.h> 44 #include <linux/netdevice.h> 45 #include <linux/rtnetlink.h> 46 #include <linux/proc_fs.h> 47 #include <linux/seq_file.h> 48 #include <linux/notifier.h> 49 #include <linux/security.h> 50 #include <linux/jhash.h> 51 #include <linux/jiffies.h> 52 #include <linux/random.h> 53 #include <linux/bitops.h> 54 #include <linux/mm.h> 55 #include <linux/types.h> 56 #include <linux/audit.h> 57 #include <linux/mutex.h> 58 59 #include <net/net_namespace.h> 60 #include <net/sock.h> 61 #include <net/scm.h> 62 #include <net/netlink.h> 63 64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8) 65 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long)) 66 67 struct netlink_sock { 68 /* struct sock has to be the first member of netlink_sock */ 69 struct sock sk; 70 u32 pid; 71 u32 dst_pid; 72 u32 dst_group; 73 u32 flags; 74 u32 subscriptions; 75 u32 ngroups; 76 unsigned long *groups; 77 unsigned long state; 78 wait_queue_head_t wait; 79 struct netlink_callback *cb; 80 struct mutex *cb_mutex; 81 struct mutex cb_def_mutex; 82 void (*netlink_rcv)(struct sk_buff *skb); 83 struct module *module; 84 }; 85 86 struct listeners { 87 struct rcu_head rcu; 88 unsigned long masks[0]; 89 }; 90 91 #define NETLINK_KERNEL_SOCKET 0x1 92 #define NETLINK_RECV_PKTINFO 0x2 93 #define NETLINK_BROADCAST_SEND_ERROR 0x4 94 #define NETLINK_RECV_NO_ENOBUFS 0x8 95 96 static inline struct netlink_sock *nlk_sk(struct sock *sk) 97 { 98 return container_of(sk, struct netlink_sock, sk); 99 } 100 101 static inline int netlink_is_kernel(struct sock *sk) 102 { 103 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET; 104 } 105 106 struct nl_pid_hash { 107 struct hlist_head *table; 108 unsigned long rehash_time; 109 110 unsigned int mask; 111 unsigned int shift; 112 113 unsigned int entries; 114 unsigned int max_shift; 115 116 u32 rnd; 117 }; 118 119 struct netlink_table { 120 struct nl_pid_hash hash; 121 struct hlist_head mc_list; 122 struct listeners __rcu *listeners; 123 unsigned int nl_nonroot; 124 unsigned int groups; 125 struct mutex *cb_mutex; 126 struct module *module; 127 int registered; 128 }; 129 130 static struct netlink_table *nl_table; 131 132 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait); 133 134 static int netlink_dump(struct sock *sk); 135 static void netlink_destroy_callback(struct netlink_callback *cb); 136 137 static DEFINE_RWLOCK(nl_table_lock); 138 static atomic_t nl_table_users = ATOMIC_INIT(0); 139 140 static ATOMIC_NOTIFIER_HEAD(netlink_chain); 141 142 static inline u32 netlink_group_mask(u32 group) 143 { 144 return group ? 1 << (group - 1) : 0; 145 } 146 147 static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid) 148 { 149 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask]; 150 } 151 152 static void netlink_sock_destruct(struct sock *sk) 153 { 154 struct netlink_sock *nlk = nlk_sk(sk); 155 156 if (nlk->cb) { 157 if (nlk->cb->done) 158 nlk->cb->done(nlk->cb); 159 netlink_destroy_callback(nlk->cb); 160 } 161 162 skb_queue_purge(&sk->sk_receive_queue); 163 164 if (!sock_flag(sk, SOCK_DEAD)) { 165 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk); 166 return; 167 } 168 169 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 170 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 171 WARN_ON(nlk_sk(sk)->groups); 172 } 173 174 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on 175 * SMP. Look, when several writers sleep and reader wakes them up, all but one 176 * immediately hit write lock and grab all the cpus. Exclusive sleep solves 177 * this, _but_ remember, it adds useless work on UP machines. 178 */ 179 180 void netlink_table_grab(void) 181 __acquires(nl_table_lock) 182 { 183 might_sleep(); 184 185 write_lock_irq(&nl_table_lock); 186 187 if (atomic_read(&nl_table_users)) { 188 DECLARE_WAITQUEUE(wait, current); 189 190 add_wait_queue_exclusive(&nl_table_wait, &wait); 191 for (;;) { 192 set_current_state(TASK_UNINTERRUPTIBLE); 193 if (atomic_read(&nl_table_users) == 0) 194 break; 195 write_unlock_irq(&nl_table_lock); 196 schedule(); 197 write_lock_irq(&nl_table_lock); 198 } 199 200 __set_current_state(TASK_RUNNING); 201 remove_wait_queue(&nl_table_wait, &wait); 202 } 203 } 204 205 void netlink_table_ungrab(void) 206 __releases(nl_table_lock) 207 { 208 write_unlock_irq(&nl_table_lock); 209 wake_up(&nl_table_wait); 210 } 211 212 static inline void 213 netlink_lock_table(void) 214 { 215 /* read_lock() synchronizes us to netlink_table_grab */ 216 217 read_lock(&nl_table_lock); 218 atomic_inc(&nl_table_users); 219 read_unlock(&nl_table_lock); 220 } 221 222 static inline void 223 netlink_unlock_table(void) 224 { 225 if (atomic_dec_and_test(&nl_table_users)) 226 wake_up(&nl_table_wait); 227 } 228 229 static struct sock *netlink_lookup(struct net *net, int protocol, u32 pid) 230 { 231 struct nl_pid_hash *hash = &nl_table[protocol].hash; 232 struct hlist_head *head; 233 struct sock *sk; 234 struct hlist_node *node; 235 236 read_lock(&nl_table_lock); 237 head = nl_pid_hashfn(hash, pid); 238 sk_for_each(sk, node, head) { 239 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) { 240 sock_hold(sk); 241 goto found; 242 } 243 } 244 sk = NULL; 245 found: 246 read_unlock(&nl_table_lock); 247 return sk; 248 } 249 250 static struct hlist_head *nl_pid_hash_zalloc(size_t size) 251 { 252 if (size <= PAGE_SIZE) 253 return kzalloc(size, GFP_ATOMIC); 254 else 255 return (struct hlist_head *) 256 __get_free_pages(GFP_ATOMIC | __GFP_ZERO, 257 get_order(size)); 258 } 259 260 static void nl_pid_hash_free(struct hlist_head *table, size_t size) 261 { 262 if (size <= PAGE_SIZE) 263 kfree(table); 264 else 265 free_pages((unsigned long)table, get_order(size)); 266 } 267 268 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow) 269 { 270 unsigned int omask, mask, shift; 271 size_t osize, size; 272 struct hlist_head *otable, *table; 273 int i; 274 275 omask = mask = hash->mask; 276 osize = size = (mask + 1) * sizeof(*table); 277 shift = hash->shift; 278 279 if (grow) { 280 if (++shift > hash->max_shift) 281 return 0; 282 mask = mask * 2 + 1; 283 size *= 2; 284 } 285 286 table = nl_pid_hash_zalloc(size); 287 if (!table) 288 return 0; 289 290 otable = hash->table; 291 hash->table = table; 292 hash->mask = mask; 293 hash->shift = shift; 294 get_random_bytes(&hash->rnd, sizeof(hash->rnd)); 295 296 for (i = 0; i <= omask; i++) { 297 struct sock *sk; 298 struct hlist_node *node, *tmp; 299 300 sk_for_each_safe(sk, node, tmp, &otable[i]) 301 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid)); 302 } 303 304 nl_pid_hash_free(otable, osize); 305 hash->rehash_time = jiffies + 10 * 60 * HZ; 306 return 1; 307 } 308 309 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len) 310 { 311 int avg = hash->entries >> hash->shift; 312 313 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1)) 314 return 1; 315 316 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) { 317 nl_pid_hash_rehash(hash, 0); 318 return 1; 319 } 320 321 return 0; 322 } 323 324 static const struct proto_ops netlink_ops; 325 326 static void 327 netlink_update_listeners(struct sock *sk) 328 { 329 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 330 struct hlist_node *node; 331 unsigned long mask; 332 unsigned int i; 333 334 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) { 335 mask = 0; 336 sk_for_each_bound(sk, node, &tbl->mc_list) { 337 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups)) 338 mask |= nlk_sk(sk)->groups[i]; 339 } 340 tbl->listeners->masks[i] = mask; 341 } 342 /* this function is only called with the netlink table "grabbed", which 343 * makes sure updates are visible before bind or setsockopt return. */ 344 } 345 346 static int netlink_insert(struct sock *sk, struct net *net, u32 pid) 347 { 348 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 349 struct hlist_head *head; 350 int err = -EADDRINUSE; 351 struct sock *osk; 352 struct hlist_node *node; 353 int len; 354 355 netlink_table_grab(); 356 head = nl_pid_hashfn(hash, pid); 357 len = 0; 358 sk_for_each(osk, node, head) { 359 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid)) 360 break; 361 len++; 362 } 363 if (node) 364 goto err; 365 366 err = -EBUSY; 367 if (nlk_sk(sk)->pid) 368 goto err; 369 370 err = -ENOMEM; 371 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX)) 372 goto err; 373 374 if (len && nl_pid_hash_dilute(hash, len)) 375 head = nl_pid_hashfn(hash, pid); 376 hash->entries++; 377 nlk_sk(sk)->pid = pid; 378 sk_add_node(sk, head); 379 err = 0; 380 381 err: 382 netlink_table_ungrab(); 383 return err; 384 } 385 386 static void netlink_remove(struct sock *sk) 387 { 388 netlink_table_grab(); 389 if (sk_del_node_init(sk)) 390 nl_table[sk->sk_protocol].hash.entries--; 391 if (nlk_sk(sk)->subscriptions) 392 __sk_del_bind_node(sk); 393 netlink_table_ungrab(); 394 } 395 396 static struct proto netlink_proto = { 397 .name = "NETLINK", 398 .owner = THIS_MODULE, 399 .obj_size = sizeof(struct netlink_sock), 400 }; 401 402 static int __netlink_create(struct net *net, struct socket *sock, 403 struct mutex *cb_mutex, int protocol) 404 { 405 struct sock *sk; 406 struct netlink_sock *nlk; 407 408 sock->ops = &netlink_ops; 409 410 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto); 411 if (!sk) 412 return -ENOMEM; 413 414 sock_init_data(sock, sk); 415 416 nlk = nlk_sk(sk); 417 if (cb_mutex) 418 nlk->cb_mutex = cb_mutex; 419 else { 420 nlk->cb_mutex = &nlk->cb_def_mutex; 421 mutex_init(nlk->cb_mutex); 422 } 423 init_waitqueue_head(&nlk->wait); 424 425 sk->sk_destruct = netlink_sock_destruct; 426 sk->sk_protocol = protocol; 427 return 0; 428 } 429 430 static int netlink_create(struct net *net, struct socket *sock, int protocol, 431 int kern) 432 { 433 struct module *module = NULL; 434 struct mutex *cb_mutex; 435 struct netlink_sock *nlk; 436 int err = 0; 437 438 sock->state = SS_UNCONNECTED; 439 440 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 441 return -ESOCKTNOSUPPORT; 442 443 if (protocol < 0 || protocol >= MAX_LINKS) 444 return -EPROTONOSUPPORT; 445 446 netlink_lock_table(); 447 #ifdef CONFIG_MODULES 448 if (!nl_table[protocol].registered) { 449 netlink_unlock_table(); 450 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol); 451 netlink_lock_table(); 452 } 453 #endif 454 if (nl_table[protocol].registered && 455 try_module_get(nl_table[protocol].module)) 456 module = nl_table[protocol].module; 457 else 458 err = -EPROTONOSUPPORT; 459 cb_mutex = nl_table[protocol].cb_mutex; 460 netlink_unlock_table(); 461 462 if (err < 0) 463 goto out; 464 465 err = __netlink_create(net, sock, cb_mutex, protocol); 466 if (err < 0) 467 goto out_module; 468 469 local_bh_disable(); 470 sock_prot_inuse_add(net, &netlink_proto, 1); 471 local_bh_enable(); 472 473 nlk = nlk_sk(sock->sk); 474 nlk->module = module; 475 out: 476 return err; 477 478 out_module: 479 module_put(module); 480 goto out; 481 } 482 483 static int netlink_release(struct socket *sock) 484 { 485 struct sock *sk = sock->sk; 486 struct netlink_sock *nlk; 487 488 if (!sk) 489 return 0; 490 491 netlink_remove(sk); 492 sock_orphan(sk); 493 nlk = nlk_sk(sk); 494 495 /* 496 * OK. Socket is unlinked, any packets that arrive now 497 * will be purged. 498 */ 499 500 sock->sk = NULL; 501 wake_up_interruptible_all(&nlk->wait); 502 503 skb_queue_purge(&sk->sk_write_queue); 504 505 if (nlk->pid) { 506 struct netlink_notify n = { 507 .net = sock_net(sk), 508 .protocol = sk->sk_protocol, 509 .pid = nlk->pid, 510 }; 511 atomic_notifier_call_chain(&netlink_chain, 512 NETLINK_URELEASE, &n); 513 } 514 515 module_put(nlk->module); 516 517 netlink_table_grab(); 518 if (netlink_is_kernel(sk)) { 519 BUG_ON(nl_table[sk->sk_protocol].registered == 0); 520 if (--nl_table[sk->sk_protocol].registered == 0) { 521 kfree(nl_table[sk->sk_protocol].listeners); 522 nl_table[sk->sk_protocol].module = NULL; 523 nl_table[sk->sk_protocol].registered = 0; 524 } 525 } else if (nlk->subscriptions) 526 netlink_update_listeners(sk); 527 netlink_table_ungrab(); 528 529 kfree(nlk->groups); 530 nlk->groups = NULL; 531 532 local_bh_disable(); 533 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1); 534 local_bh_enable(); 535 sock_put(sk); 536 return 0; 537 } 538 539 static int netlink_autobind(struct socket *sock) 540 { 541 struct sock *sk = sock->sk; 542 struct net *net = sock_net(sk); 543 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 544 struct hlist_head *head; 545 struct sock *osk; 546 struct hlist_node *node; 547 s32 pid = task_tgid_vnr(current); 548 int err; 549 static s32 rover = -4097; 550 551 retry: 552 cond_resched(); 553 netlink_table_grab(); 554 head = nl_pid_hashfn(hash, pid); 555 sk_for_each(osk, node, head) { 556 if (!net_eq(sock_net(osk), net)) 557 continue; 558 if (nlk_sk(osk)->pid == pid) { 559 /* Bind collision, search negative pid values. */ 560 pid = rover--; 561 if (rover > -4097) 562 rover = -4097; 563 netlink_table_ungrab(); 564 goto retry; 565 } 566 } 567 netlink_table_ungrab(); 568 569 err = netlink_insert(sk, net, pid); 570 if (err == -EADDRINUSE) 571 goto retry; 572 573 /* If 2 threads race to autobind, that is fine. */ 574 if (err == -EBUSY) 575 err = 0; 576 577 return err; 578 } 579 580 static inline int netlink_capable(const struct socket *sock, unsigned int flag) 581 { 582 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) || 583 capable(CAP_NET_ADMIN); 584 } 585 586 static void 587 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions) 588 { 589 struct netlink_sock *nlk = nlk_sk(sk); 590 591 if (nlk->subscriptions && !subscriptions) 592 __sk_del_bind_node(sk); 593 else if (!nlk->subscriptions && subscriptions) 594 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 595 nlk->subscriptions = subscriptions; 596 } 597 598 static int netlink_realloc_groups(struct sock *sk) 599 { 600 struct netlink_sock *nlk = nlk_sk(sk); 601 unsigned int groups; 602 unsigned long *new_groups; 603 int err = 0; 604 605 netlink_table_grab(); 606 607 groups = nl_table[sk->sk_protocol].groups; 608 if (!nl_table[sk->sk_protocol].registered) { 609 err = -ENOENT; 610 goto out_unlock; 611 } 612 613 if (nlk->ngroups >= groups) 614 goto out_unlock; 615 616 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC); 617 if (new_groups == NULL) { 618 err = -ENOMEM; 619 goto out_unlock; 620 } 621 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0, 622 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups)); 623 624 nlk->groups = new_groups; 625 nlk->ngroups = groups; 626 out_unlock: 627 netlink_table_ungrab(); 628 return err; 629 } 630 631 static int netlink_bind(struct socket *sock, struct sockaddr *addr, 632 int addr_len) 633 { 634 struct sock *sk = sock->sk; 635 struct net *net = sock_net(sk); 636 struct netlink_sock *nlk = nlk_sk(sk); 637 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 638 int err; 639 640 if (nladdr->nl_family != AF_NETLINK) 641 return -EINVAL; 642 643 /* Only superuser is allowed to listen multicasts */ 644 if (nladdr->nl_groups) { 645 if (!netlink_capable(sock, NL_NONROOT_RECV)) 646 return -EPERM; 647 err = netlink_realloc_groups(sk); 648 if (err) 649 return err; 650 } 651 652 if (nlk->pid) { 653 if (nladdr->nl_pid != nlk->pid) 654 return -EINVAL; 655 } else { 656 err = nladdr->nl_pid ? 657 netlink_insert(sk, net, nladdr->nl_pid) : 658 netlink_autobind(sock); 659 if (err) 660 return err; 661 } 662 663 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0])) 664 return 0; 665 666 netlink_table_grab(); 667 netlink_update_subscriptions(sk, nlk->subscriptions + 668 hweight32(nladdr->nl_groups) - 669 hweight32(nlk->groups[0])); 670 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups; 671 netlink_update_listeners(sk); 672 netlink_table_ungrab(); 673 674 return 0; 675 } 676 677 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 678 int alen, int flags) 679 { 680 int err = 0; 681 struct sock *sk = sock->sk; 682 struct netlink_sock *nlk = nlk_sk(sk); 683 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 684 685 if (alen < sizeof(addr->sa_family)) 686 return -EINVAL; 687 688 if (addr->sa_family == AF_UNSPEC) { 689 sk->sk_state = NETLINK_UNCONNECTED; 690 nlk->dst_pid = 0; 691 nlk->dst_group = 0; 692 return 0; 693 } 694 if (addr->sa_family != AF_NETLINK) 695 return -EINVAL; 696 697 /* Only superuser is allowed to send multicasts */ 698 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND)) 699 return -EPERM; 700 701 if (!nlk->pid) 702 err = netlink_autobind(sock); 703 704 if (err == 0) { 705 sk->sk_state = NETLINK_CONNECTED; 706 nlk->dst_pid = nladdr->nl_pid; 707 nlk->dst_group = ffs(nladdr->nl_groups); 708 } 709 710 return err; 711 } 712 713 static int netlink_getname(struct socket *sock, struct sockaddr *addr, 714 int *addr_len, int peer) 715 { 716 struct sock *sk = sock->sk; 717 struct netlink_sock *nlk = nlk_sk(sk); 718 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr); 719 720 nladdr->nl_family = AF_NETLINK; 721 nladdr->nl_pad = 0; 722 *addr_len = sizeof(*nladdr); 723 724 if (peer) { 725 nladdr->nl_pid = nlk->dst_pid; 726 nladdr->nl_groups = netlink_group_mask(nlk->dst_group); 727 } else { 728 nladdr->nl_pid = nlk->pid; 729 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0; 730 } 731 return 0; 732 } 733 734 static void netlink_overrun(struct sock *sk) 735 { 736 struct netlink_sock *nlk = nlk_sk(sk); 737 738 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) { 739 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) { 740 sk->sk_err = ENOBUFS; 741 sk->sk_error_report(sk); 742 } 743 } 744 atomic_inc(&sk->sk_drops); 745 } 746 747 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid) 748 { 749 struct sock *sock; 750 struct netlink_sock *nlk; 751 752 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid); 753 if (!sock) 754 return ERR_PTR(-ECONNREFUSED); 755 756 /* Don't bother queuing skb if kernel socket has no input function */ 757 nlk = nlk_sk(sock); 758 if (sock->sk_state == NETLINK_CONNECTED && 759 nlk->dst_pid != nlk_sk(ssk)->pid) { 760 sock_put(sock); 761 return ERR_PTR(-ECONNREFUSED); 762 } 763 return sock; 764 } 765 766 struct sock *netlink_getsockbyfilp(struct file *filp) 767 { 768 struct inode *inode = filp->f_path.dentry->d_inode; 769 struct sock *sock; 770 771 if (!S_ISSOCK(inode->i_mode)) 772 return ERR_PTR(-ENOTSOCK); 773 774 sock = SOCKET_I(inode)->sk; 775 if (sock->sk_family != AF_NETLINK) 776 return ERR_PTR(-EINVAL); 777 778 sock_hold(sock); 779 return sock; 780 } 781 782 /* 783 * Attach a skb to a netlink socket. 784 * The caller must hold a reference to the destination socket. On error, the 785 * reference is dropped. The skb is not send to the destination, just all 786 * all error checks are performed and memory in the queue is reserved. 787 * Return values: 788 * < 0: error. skb freed, reference to sock dropped. 789 * 0: continue 790 * 1: repeat lookup - reference dropped while waiting for socket memory. 791 */ 792 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, 793 long *timeo, struct sock *ssk) 794 { 795 struct netlink_sock *nlk; 796 797 nlk = nlk_sk(sk); 798 799 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 800 test_bit(0, &nlk->state)) { 801 DECLARE_WAITQUEUE(wait, current); 802 if (!*timeo) { 803 if (!ssk || netlink_is_kernel(ssk)) 804 netlink_overrun(sk); 805 sock_put(sk); 806 kfree_skb(skb); 807 return -EAGAIN; 808 } 809 810 __set_current_state(TASK_INTERRUPTIBLE); 811 add_wait_queue(&nlk->wait, &wait); 812 813 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 814 test_bit(0, &nlk->state)) && 815 !sock_flag(sk, SOCK_DEAD)) 816 *timeo = schedule_timeout(*timeo); 817 818 __set_current_state(TASK_RUNNING); 819 remove_wait_queue(&nlk->wait, &wait); 820 sock_put(sk); 821 822 if (signal_pending(current)) { 823 kfree_skb(skb); 824 return sock_intr_errno(*timeo); 825 } 826 return 1; 827 } 828 skb_set_owner_r(skb, sk); 829 return 0; 830 } 831 832 int netlink_sendskb(struct sock *sk, struct sk_buff *skb) 833 { 834 int len = skb->len; 835 836 skb_queue_tail(&sk->sk_receive_queue, skb); 837 sk->sk_data_ready(sk, len); 838 sock_put(sk); 839 return len; 840 } 841 842 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 843 { 844 kfree_skb(skb); 845 sock_put(sk); 846 } 847 848 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation) 849 { 850 int delta; 851 852 skb_orphan(skb); 853 854 delta = skb->end - skb->tail; 855 if (delta * 2 < skb->truesize) 856 return skb; 857 858 if (skb_shared(skb)) { 859 struct sk_buff *nskb = skb_clone(skb, allocation); 860 if (!nskb) 861 return skb; 862 kfree_skb(skb); 863 skb = nskb; 864 } 865 866 if (!pskb_expand_head(skb, 0, -delta, allocation)) 867 skb->truesize -= delta; 868 869 return skb; 870 } 871 872 static void netlink_rcv_wake(struct sock *sk) 873 { 874 struct netlink_sock *nlk = nlk_sk(sk); 875 876 if (skb_queue_empty(&sk->sk_receive_queue)) 877 clear_bit(0, &nlk->state); 878 if (!test_bit(0, &nlk->state)) 879 wake_up_interruptible(&nlk->wait); 880 } 881 882 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb) 883 { 884 int ret; 885 struct netlink_sock *nlk = nlk_sk(sk); 886 887 ret = -ECONNREFUSED; 888 if (nlk->netlink_rcv != NULL) { 889 ret = skb->len; 890 skb_set_owner_r(skb, sk); 891 nlk->netlink_rcv(skb); 892 } 893 kfree_skb(skb); 894 sock_put(sk); 895 return ret; 896 } 897 898 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, 899 u32 pid, int nonblock) 900 { 901 struct sock *sk; 902 int err; 903 long timeo; 904 905 skb = netlink_trim(skb, gfp_any()); 906 907 timeo = sock_sndtimeo(ssk, nonblock); 908 retry: 909 sk = netlink_getsockbypid(ssk, pid); 910 if (IS_ERR(sk)) { 911 kfree_skb(skb); 912 return PTR_ERR(sk); 913 } 914 if (netlink_is_kernel(sk)) 915 return netlink_unicast_kernel(sk, skb); 916 917 if (sk_filter(sk, skb)) { 918 err = skb->len; 919 kfree_skb(skb); 920 sock_put(sk); 921 return err; 922 } 923 924 err = netlink_attachskb(sk, skb, &timeo, ssk); 925 if (err == 1) 926 goto retry; 927 if (err) 928 return err; 929 930 return netlink_sendskb(sk, skb); 931 } 932 EXPORT_SYMBOL(netlink_unicast); 933 934 int netlink_has_listeners(struct sock *sk, unsigned int group) 935 { 936 int res = 0; 937 struct listeners *listeners; 938 939 BUG_ON(!netlink_is_kernel(sk)); 940 941 rcu_read_lock(); 942 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners); 943 944 if (group - 1 < nl_table[sk->sk_protocol].groups) 945 res = test_bit(group - 1, listeners->masks); 946 947 rcu_read_unlock(); 948 949 return res; 950 } 951 EXPORT_SYMBOL_GPL(netlink_has_listeners); 952 953 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 954 { 955 struct netlink_sock *nlk = nlk_sk(sk); 956 957 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 958 !test_bit(0, &nlk->state)) { 959 skb_set_owner_r(skb, sk); 960 skb_queue_tail(&sk->sk_receive_queue, skb); 961 sk->sk_data_ready(sk, skb->len); 962 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1); 963 } 964 return -1; 965 } 966 967 struct netlink_broadcast_data { 968 struct sock *exclude_sk; 969 struct net *net; 970 u32 pid; 971 u32 group; 972 int failure; 973 int delivery_failure; 974 int congested; 975 int delivered; 976 gfp_t allocation; 977 struct sk_buff *skb, *skb2; 978 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data); 979 void *tx_data; 980 }; 981 982 static int do_one_broadcast(struct sock *sk, 983 struct netlink_broadcast_data *p) 984 { 985 struct netlink_sock *nlk = nlk_sk(sk); 986 int val; 987 988 if (p->exclude_sk == sk) 989 goto out; 990 991 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups || 992 !test_bit(p->group - 1, nlk->groups)) 993 goto out; 994 995 if (!net_eq(sock_net(sk), p->net)) 996 goto out; 997 998 if (p->failure) { 999 netlink_overrun(sk); 1000 goto out; 1001 } 1002 1003 sock_hold(sk); 1004 if (p->skb2 == NULL) { 1005 if (skb_shared(p->skb)) { 1006 p->skb2 = skb_clone(p->skb, p->allocation); 1007 } else { 1008 p->skb2 = skb_get(p->skb); 1009 /* 1010 * skb ownership may have been set when 1011 * delivered to a previous socket. 1012 */ 1013 skb_orphan(p->skb2); 1014 } 1015 } 1016 if (p->skb2 == NULL) { 1017 netlink_overrun(sk); 1018 /* Clone failed. Notify ALL listeners. */ 1019 p->failure = 1; 1020 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR) 1021 p->delivery_failure = 1; 1022 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) { 1023 kfree_skb(p->skb2); 1024 p->skb2 = NULL; 1025 } else if (sk_filter(sk, p->skb2)) { 1026 kfree_skb(p->skb2); 1027 p->skb2 = NULL; 1028 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) { 1029 netlink_overrun(sk); 1030 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR) 1031 p->delivery_failure = 1; 1032 } else { 1033 p->congested |= val; 1034 p->delivered = 1; 1035 p->skb2 = NULL; 1036 } 1037 sock_put(sk); 1038 1039 out: 1040 return 0; 1041 } 1042 1043 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid, 1044 u32 group, gfp_t allocation, 1045 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data), 1046 void *filter_data) 1047 { 1048 struct net *net = sock_net(ssk); 1049 struct netlink_broadcast_data info; 1050 struct hlist_node *node; 1051 struct sock *sk; 1052 1053 skb = netlink_trim(skb, allocation); 1054 1055 info.exclude_sk = ssk; 1056 info.net = net; 1057 info.pid = pid; 1058 info.group = group; 1059 info.failure = 0; 1060 info.delivery_failure = 0; 1061 info.congested = 0; 1062 info.delivered = 0; 1063 info.allocation = allocation; 1064 info.skb = skb; 1065 info.skb2 = NULL; 1066 info.tx_filter = filter; 1067 info.tx_data = filter_data; 1068 1069 /* While we sleep in clone, do not allow to change socket list */ 1070 1071 netlink_lock_table(); 1072 1073 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 1074 do_one_broadcast(sk, &info); 1075 1076 consume_skb(skb); 1077 1078 netlink_unlock_table(); 1079 1080 if (info.delivery_failure) { 1081 kfree_skb(info.skb2); 1082 return -ENOBUFS; 1083 } else 1084 consume_skb(info.skb2); 1085 1086 if (info.delivered) { 1087 if (info.congested && (allocation & __GFP_WAIT)) 1088 yield(); 1089 return 0; 1090 } 1091 return -ESRCH; 1092 } 1093 EXPORT_SYMBOL(netlink_broadcast_filtered); 1094 1095 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid, 1096 u32 group, gfp_t allocation) 1097 { 1098 return netlink_broadcast_filtered(ssk, skb, pid, group, allocation, 1099 NULL, NULL); 1100 } 1101 EXPORT_SYMBOL(netlink_broadcast); 1102 1103 struct netlink_set_err_data { 1104 struct sock *exclude_sk; 1105 u32 pid; 1106 u32 group; 1107 int code; 1108 }; 1109 1110 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p) 1111 { 1112 struct netlink_sock *nlk = nlk_sk(sk); 1113 int ret = 0; 1114 1115 if (sk == p->exclude_sk) 1116 goto out; 1117 1118 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk))) 1119 goto out; 1120 1121 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups || 1122 !test_bit(p->group - 1, nlk->groups)) 1123 goto out; 1124 1125 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) { 1126 ret = 1; 1127 goto out; 1128 } 1129 1130 sk->sk_err = p->code; 1131 sk->sk_error_report(sk); 1132 out: 1133 return ret; 1134 } 1135 1136 /** 1137 * netlink_set_err - report error to broadcast listeners 1138 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create() 1139 * @pid: the PID of a process that we want to skip (if any) 1140 * @groups: the broadcast group that will notice the error 1141 * @code: error code, must be negative (as usual in kernelspace) 1142 * 1143 * This function returns the number of broadcast listeners that have set the 1144 * NETLINK_RECV_NO_ENOBUFS socket option. 1145 */ 1146 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code) 1147 { 1148 struct netlink_set_err_data info; 1149 struct hlist_node *node; 1150 struct sock *sk; 1151 int ret = 0; 1152 1153 info.exclude_sk = ssk; 1154 info.pid = pid; 1155 info.group = group; 1156 /* sk->sk_err wants a positive error value */ 1157 info.code = -code; 1158 1159 read_lock(&nl_table_lock); 1160 1161 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 1162 ret += do_one_set_err(sk, &info); 1163 1164 read_unlock(&nl_table_lock); 1165 return ret; 1166 } 1167 EXPORT_SYMBOL(netlink_set_err); 1168 1169 /* must be called with netlink table grabbed */ 1170 static void netlink_update_socket_mc(struct netlink_sock *nlk, 1171 unsigned int group, 1172 int is_new) 1173 { 1174 int old, new = !!is_new, subscriptions; 1175 1176 old = test_bit(group - 1, nlk->groups); 1177 subscriptions = nlk->subscriptions - old + new; 1178 if (new) 1179 __set_bit(group - 1, nlk->groups); 1180 else 1181 __clear_bit(group - 1, nlk->groups); 1182 netlink_update_subscriptions(&nlk->sk, subscriptions); 1183 netlink_update_listeners(&nlk->sk); 1184 } 1185 1186 static int netlink_setsockopt(struct socket *sock, int level, int optname, 1187 char __user *optval, unsigned int optlen) 1188 { 1189 struct sock *sk = sock->sk; 1190 struct netlink_sock *nlk = nlk_sk(sk); 1191 unsigned int val = 0; 1192 int err; 1193 1194 if (level != SOL_NETLINK) 1195 return -ENOPROTOOPT; 1196 1197 if (optlen >= sizeof(int) && 1198 get_user(val, (unsigned int __user *)optval)) 1199 return -EFAULT; 1200 1201 switch (optname) { 1202 case NETLINK_PKTINFO: 1203 if (val) 1204 nlk->flags |= NETLINK_RECV_PKTINFO; 1205 else 1206 nlk->flags &= ~NETLINK_RECV_PKTINFO; 1207 err = 0; 1208 break; 1209 case NETLINK_ADD_MEMBERSHIP: 1210 case NETLINK_DROP_MEMBERSHIP: { 1211 if (!netlink_capable(sock, NL_NONROOT_RECV)) 1212 return -EPERM; 1213 err = netlink_realloc_groups(sk); 1214 if (err) 1215 return err; 1216 if (!val || val - 1 >= nlk->ngroups) 1217 return -EINVAL; 1218 netlink_table_grab(); 1219 netlink_update_socket_mc(nlk, val, 1220 optname == NETLINK_ADD_MEMBERSHIP); 1221 netlink_table_ungrab(); 1222 err = 0; 1223 break; 1224 } 1225 case NETLINK_BROADCAST_ERROR: 1226 if (val) 1227 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR; 1228 else 1229 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR; 1230 err = 0; 1231 break; 1232 case NETLINK_NO_ENOBUFS: 1233 if (val) { 1234 nlk->flags |= NETLINK_RECV_NO_ENOBUFS; 1235 clear_bit(0, &nlk->state); 1236 wake_up_interruptible(&nlk->wait); 1237 } else 1238 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS; 1239 err = 0; 1240 break; 1241 default: 1242 err = -ENOPROTOOPT; 1243 } 1244 return err; 1245 } 1246 1247 static int netlink_getsockopt(struct socket *sock, int level, int optname, 1248 char __user *optval, int __user *optlen) 1249 { 1250 struct sock *sk = sock->sk; 1251 struct netlink_sock *nlk = nlk_sk(sk); 1252 int len, val, err; 1253 1254 if (level != SOL_NETLINK) 1255 return -ENOPROTOOPT; 1256 1257 if (get_user(len, optlen)) 1258 return -EFAULT; 1259 if (len < 0) 1260 return -EINVAL; 1261 1262 switch (optname) { 1263 case NETLINK_PKTINFO: 1264 if (len < sizeof(int)) 1265 return -EINVAL; 1266 len = sizeof(int); 1267 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0; 1268 if (put_user(len, optlen) || 1269 put_user(val, optval)) 1270 return -EFAULT; 1271 err = 0; 1272 break; 1273 case NETLINK_BROADCAST_ERROR: 1274 if (len < sizeof(int)) 1275 return -EINVAL; 1276 len = sizeof(int); 1277 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0; 1278 if (put_user(len, optlen) || 1279 put_user(val, optval)) 1280 return -EFAULT; 1281 err = 0; 1282 break; 1283 case NETLINK_NO_ENOBUFS: 1284 if (len < sizeof(int)) 1285 return -EINVAL; 1286 len = sizeof(int); 1287 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0; 1288 if (put_user(len, optlen) || 1289 put_user(val, optval)) 1290 return -EFAULT; 1291 err = 0; 1292 break; 1293 default: 1294 err = -ENOPROTOOPT; 1295 } 1296 return err; 1297 } 1298 1299 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb) 1300 { 1301 struct nl_pktinfo info; 1302 1303 info.group = NETLINK_CB(skb).dst_group; 1304 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info); 1305 } 1306 1307 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock, 1308 struct msghdr *msg, size_t len) 1309 { 1310 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1311 struct sock *sk = sock->sk; 1312 struct netlink_sock *nlk = nlk_sk(sk); 1313 struct sockaddr_nl *addr = msg->msg_name; 1314 u32 dst_pid; 1315 u32 dst_group; 1316 struct sk_buff *skb; 1317 int err; 1318 struct scm_cookie scm; 1319 1320 if (msg->msg_flags&MSG_OOB) 1321 return -EOPNOTSUPP; 1322 1323 if (NULL == siocb->scm) 1324 siocb->scm = &scm; 1325 1326 err = scm_send(sock, msg, siocb->scm); 1327 if (err < 0) 1328 return err; 1329 1330 if (msg->msg_namelen) { 1331 err = -EINVAL; 1332 if (addr->nl_family != AF_NETLINK) 1333 goto out; 1334 dst_pid = addr->nl_pid; 1335 dst_group = ffs(addr->nl_groups); 1336 err = -EPERM; 1337 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND)) 1338 goto out; 1339 } else { 1340 dst_pid = nlk->dst_pid; 1341 dst_group = nlk->dst_group; 1342 } 1343 1344 if (!nlk->pid) { 1345 err = netlink_autobind(sock); 1346 if (err) 1347 goto out; 1348 } 1349 1350 err = -EMSGSIZE; 1351 if (len > sk->sk_sndbuf - 32) 1352 goto out; 1353 err = -ENOBUFS; 1354 skb = alloc_skb(len, GFP_KERNEL); 1355 if (skb == NULL) 1356 goto out; 1357 1358 NETLINK_CB(skb).pid = nlk->pid; 1359 NETLINK_CB(skb).dst_group = dst_group; 1360 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); 1361 1362 err = -EFAULT; 1363 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) { 1364 kfree_skb(skb); 1365 goto out; 1366 } 1367 1368 err = security_netlink_send(sk, skb); 1369 if (err) { 1370 kfree_skb(skb); 1371 goto out; 1372 } 1373 1374 if (dst_group) { 1375 atomic_inc(&skb->users); 1376 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL); 1377 } 1378 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT); 1379 1380 out: 1381 scm_destroy(siocb->scm); 1382 return err; 1383 } 1384 1385 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock, 1386 struct msghdr *msg, size_t len, 1387 int flags) 1388 { 1389 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1390 struct scm_cookie scm; 1391 struct sock *sk = sock->sk; 1392 struct netlink_sock *nlk = nlk_sk(sk); 1393 int noblock = flags&MSG_DONTWAIT; 1394 size_t copied; 1395 struct sk_buff *skb, *data_skb; 1396 int err, ret; 1397 1398 if (flags&MSG_OOB) 1399 return -EOPNOTSUPP; 1400 1401 copied = 0; 1402 1403 skb = skb_recv_datagram(sk, flags, noblock, &err); 1404 if (skb == NULL) 1405 goto out; 1406 1407 data_skb = skb; 1408 1409 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES 1410 if (unlikely(skb_shinfo(skb)->frag_list)) { 1411 /* 1412 * If this skb has a frag_list, then here that means that we 1413 * will have to use the frag_list skb's data for compat tasks 1414 * and the regular skb's data for normal (non-compat) tasks. 1415 * 1416 * If we need to send the compat skb, assign it to the 1417 * 'data_skb' variable so that it will be used below for data 1418 * copying. We keep 'skb' for everything else, including 1419 * freeing both later. 1420 */ 1421 if (flags & MSG_CMSG_COMPAT) 1422 data_skb = skb_shinfo(skb)->frag_list; 1423 } 1424 #endif 1425 1426 msg->msg_namelen = 0; 1427 1428 copied = data_skb->len; 1429 if (len < copied) { 1430 msg->msg_flags |= MSG_TRUNC; 1431 copied = len; 1432 } 1433 1434 skb_reset_transport_header(data_skb); 1435 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied); 1436 1437 if (msg->msg_name) { 1438 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name; 1439 addr->nl_family = AF_NETLINK; 1440 addr->nl_pad = 0; 1441 addr->nl_pid = NETLINK_CB(skb).pid; 1442 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group); 1443 msg->msg_namelen = sizeof(*addr); 1444 } 1445 1446 if (nlk->flags & NETLINK_RECV_PKTINFO) 1447 netlink_cmsg_recv_pktinfo(msg, skb); 1448 1449 if (NULL == siocb->scm) { 1450 memset(&scm, 0, sizeof(scm)); 1451 siocb->scm = &scm; 1452 } 1453 siocb->scm->creds = *NETLINK_CREDS(skb); 1454 if (flags & MSG_TRUNC) 1455 copied = data_skb->len; 1456 1457 skb_free_datagram(sk, skb); 1458 1459 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) { 1460 ret = netlink_dump(sk); 1461 if (ret) { 1462 sk->sk_err = ret; 1463 sk->sk_error_report(sk); 1464 } 1465 } 1466 1467 scm_recv(sock, msg, siocb->scm, flags); 1468 out: 1469 netlink_rcv_wake(sk); 1470 return err ? : copied; 1471 } 1472 1473 static void netlink_data_ready(struct sock *sk, int len) 1474 { 1475 BUG(); 1476 } 1477 1478 /* 1479 * We export these functions to other modules. They provide a 1480 * complete set of kernel non-blocking support for message 1481 * queueing. 1482 */ 1483 1484 struct sock * 1485 netlink_kernel_create(struct net *net, int unit, unsigned int groups, 1486 void (*input)(struct sk_buff *skb), 1487 struct mutex *cb_mutex, struct module *module) 1488 { 1489 struct socket *sock; 1490 struct sock *sk; 1491 struct netlink_sock *nlk; 1492 struct listeners *listeners = NULL; 1493 1494 BUG_ON(!nl_table); 1495 1496 if (unit < 0 || unit >= MAX_LINKS) 1497 return NULL; 1498 1499 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 1500 return NULL; 1501 1502 /* 1503 * We have to just have a reference on the net from sk, but don't 1504 * get_net it. Besides, we cannot get and then put the net here. 1505 * So we create one inside init_net and the move it to net. 1506 */ 1507 1508 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0) 1509 goto out_sock_release_nosk; 1510 1511 sk = sock->sk; 1512 sk_change_net(sk, net); 1513 1514 if (groups < 32) 1515 groups = 32; 1516 1517 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 1518 if (!listeners) 1519 goto out_sock_release; 1520 1521 sk->sk_data_ready = netlink_data_ready; 1522 if (input) 1523 nlk_sk(sk)->netlink_rcv = input; 1524 1525 if (netlink_insert(sk, net, 0)) 1526 goto out_sock_release; 1527 1528 nlk = nlk_sk(sk); 1529 nlk->flags |= NETLINK_KERNEL_SOCKET; 1530 1531 netlink_table_grab(); 1532 if (!nl_table[unit].registered) { 1533 nl_table[unit].groups = groups; 1534 rcu_assign_pointer(nl_table[unit].listeners, listeners); 1535 nl_table[unit].cb_mutex = cb_mutex; 1536 nl_table[unit].module = module; 1537 nl_table[unit].registered = 1; 1538 } else { 1539 kfree(listeners); 1540 nl_table[unit].registered++; 1541 } 1542 netlink_table_ungrab(); 1543 return sk; 1544 1545 out_sock_release: 1546 kfree(listeners); 1547 netlink_kernel_release(sk); 1548 return NULL; 1549 1550 out_sock_release_nosk: 1551 sock_release(sock); 1552 return NULL; 1553 } 1554 EXPORT_SYMBOL(netlink_kernel_create); 1555 1556 1557 void 1558 netlink_kernel_release(struct sock *sk) 1559 { 1560 sk_release_kernel(sk); 1561 } 1562 EXPORT_SYMBOL(netlink_kernel_release); 1563 1564 int __netlink_change_ngroups(struct sock *sk, unsigned int groups) 1565 { 1566 struct listeners *new, *old; 1567 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 1568 1569 if (groups < 32) 1570 groups = 32; 1571 1572 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) { 1573 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC); 1574 if (!new) 1575 return -ENOMEM; 1576 old = rcu_dereference_protected(tbl->listeners, 1); 1577 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups)); 1578 rcu_assign_pointer(tbl->listeners, new); 1579 1580 kfree_rcu(old, rcu); 1581 } 1582 tbl->groups = groups; 1583 1584 return 0; 1585 } 1586 1587 /** 1588 * netlink_change_ngroups - change number of multicast groups 1589 * 1590 * This changes the number of multicast groups that are available 1591 * on a certain netlink family. Note that it is not possible to 1592 * change the number of groups to below 32. Also note that it does 1593 * not implicitly call netlink_clear_multicast_users() when the 1594 * number of groups is reduced. 1595 * 1596 * @sk: The kernel netlink socket, as returned by netlink_kernel_create(). 1597 * @groups: The new number of groups. 1598 */ 1599 int netlink_change_ngroups(struct sock *sk, unsigned int groups) 1600 { 1601 int err; 1602 1603 netlink_table_grab(); 1604 err = __netlink_change_ngroups(sk, groups); 1605 netlink_table_ungrab(); 1606 1607 return err; 1608 } 1609 1610 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group) 1611 { 1612 struct sock *sk; 1613 struct hlist_node *node; 1614 struct netlink_table *tbl = &nl_table[ksk->sk_protocol]; 1615 1616 sk_for_each_bound(sk, node, &tbl->mc_list) 1617 netlink_update_socket_mc(nlk_sk(sk), group, 0); 1618 } 1619 1620 /** 1621 * netlink_clear_multicast_users - kick off multicast listeners 1622 * 1623 * This function removes all listeners from the given group. 1624 * @ksk: The kernel netlink socket, as returned by 1625 * netlink_kernel_create(). 1626 * @group: The multicast group to clear. 1627 */ 1628 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group) 1629 { 1630 netlink_table_grab(); 1631 __netlink_clear_multicast_users(ksk, group); 1632 netlink_table_ungrab(); 1633 } 1634 1635 void netlink_set_nonroot(int protocol, unsigned int flags) 1636 { 1637 if ((unsigned int)protocol < MAX_LINKS) 1638 nl_table[protocol].nl_nonroot = flags; 1639 } 1640 EXPORT_SYMBOL(netlink_set_nonroot); 1641 1642 static void netlink_destroy_callback(struct netlink_callback *cb) 1643 { 1644 kfree_skb(cb->skb); 1645 kfree(cb); 1646 } 1647 1648 /* 1649 * It looks a bit ugly. 1650 * It would be better to create kernel thread. 1651 */ 1652 1653 static int netlink_dump(struct sock *sk) 1654 { 1655 struct netlink_sock *nlk = nlk_sk(sk); 1656 struct netlink_callback *cb; 1657 struct sk_buff *skb = NULL; 1658 struct nlmsghdr *nlh; 1659 int len, err = -ENOBUFS; 1660 int alloc_size; 1661 1662 mutex_lock(nlk->cb_mutex); 1663 1664 cb = nlk->cb; 1665 if (cb == NULL) { 1666 err = -EINVAL; 1667 goto errout_skb; 1668 } 1669 1670 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE); 1671 1672 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL); 1673 if (!skb) 1674 goto errout_skb; 1675 1676 len = cb->dump(skb, cb); 1677 1678 if (len > 0) { 1679 mutex_unlock(nlk->cb_mutex); 1680 1681 if (sk_filter(sk, skb)) 1682 kfree_skb(skb); 1683 else { 1684 skb_queue_tail(&sk->sk_receive_queue, skb); 1685 sk->sk_data_ready(sk, skb->len); 1686 } 1687 return 0; 1688 } 1689 1690 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI); 1691 if (!nlh) 1692 goto errout_skb; 1693 1694 nl_dump_check_consistent(cb, nlh); 1695 1696 memcpy(nlmsg_data(nlh), &len, sizeof(len)); 1697 1698 if (sk_filter(sk, skb)) 1699 kfree_skb(skb); 1700 else { 1701 skb_queue_tail(&sk->sk_receive_queue, skb); 1702 sk->sk_data_ready(sk, skb->len); 1703 } 1704 1705 if (cb->done) 1706 cb->done(cb); 1707 nlk->cb = NULL; 1708 mutex_unlock(nlk->cb_mutex); 1709 1710 netlink_destroy_callback(cb); 1711 return 0; 1712 1713 errout_skb: 1714 mutex_unlock(nlk->cb_mutex); 1715 kfree_skb(skb); 1716 return err; 1717 } 1718 1719 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 1720 const struct nlmsghdr *nlh, 1721 int (*dump)(struct sk_buff *skb, 1722 struct netlink_callback *), 1723 int (*done)(struct netlink_callback *), 1724 u16 min_dump_alloc) 1725 { 1726 struct netlink_callback *cb; 1727 struct sock *sk; 1728 struct netlink_sock *nlk; 1729 int ret; 1730 1731 cb = kzalloc(sizeof(*cb), GFP_KERNEL); 1732 if (cb == NULL) 1733 return -ENOBUFS; 1734 1735 cb->dump = dump; 1736 cb->done = done; 1737 cb->nlh = nlh; 1738 cb->min_dump_alloc = min_dump_alloc; 1739 atomic_inc(&skb->users); 1740 cb->skb = skb; 1741 1742 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid); 1743 if (sk == NULL) { 1744 netlink_destroy_callback(cb); 1745 return -ECONNREFUSED; 1746 } 1747 nlk = nlk_sk(sk); 1748 /* A dump is in progress... */ 1749 mutex_lock(nlk->cb_mutex); 1750 if (nlk->cb) { 1751 mutex_unlock(nlk->cb_mutex); 1752 netlink_destroy_callback(cb); 1753 sock_put(sk); 1754 return -EBUSY; 1755 } 1756 nlk->cb = cb; 1757 mutex_unlock(nlk->cb_mutex); 1758 1759 ret = netlink_dump(sk); 1760 1761 sock_put(sk); 1762 1763 if (ret) 1764 return ret; 1765 1766 /* We successfully started a dump, by returning -EINTR we 1767 * signal not to send ACK even if it was requested. 1768 */ 1769 return -EINTR; 1770 } 1771 EXPORT_SYMBOL(netlink_dump_start); 1772 1773 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err) 1774 { 1775 struct sk_buff *skb; 1776 struct nlmsghdr *rep; 1777 struct nlmsgerr *errmsg; 1778 size_t payload = sizeof(*errmsg); 1779 1780 /* error messages get the original request appened */ 1781 if (err) 1782 payload += nlmsg_len(nlh); 1783 1784 skb = nlmsg_new(payload, GFP_KERNEL); 1785 if (!skb) { 1786 struct sock *sk; 1787 1788 sk = netlink_lookup(sock_net(in_skb->sk), 1789 in_skb->sk->sk_protocol, 1790 NETLINK_CB(in_skb).pid); 1791 if (sk) { 1792 sk->sk_err = ENOBUFS; 1793 sk->sk_error_report(sk); 1794 sock_put(sk); 1795 } 1796 return; 1797 } 1798 1799 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, 1800 NLMSG_ERROR, payload, 0); 1801 errmsg = nlmsg_data(rep); 1802 errmsg->error = err; 1803 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh)); 1804 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT); 1805 } 1806 EXPORT_SYMBOL(netlink_ack); 1807 1808 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *, 1809 struct nlmsghdr *)) 1810 { 1811 struct nlmsghdr *nlh; 1812 int err; 1813 1814 while (skb->len >= nlmsg_total_size(0)) { 1815 int msglen; 1816 1817 nlh = nlmsg_hdr(skb); 1818 err = 0; 1819 1820 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len) 1821 return 0; 1822 1823 /* Only requests are handled by the kernel */ 1824 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) 1825 goto ack; 1826 1827 /* Skip control messages */ 1828 if (nlh->nlmsg_type < NLMSG_MIN_TYPE) 1829 goto ack; 1830 1831 err = cb(skb, nlh); 1832 if (err == -EINTR) 1833 goto skip; 1834 1835 ack: 1836 if (nlh->nlmsg_flags & NLM_F_ACK || err) 1837 netlink_ack(skb, nlh, err); 1838 1839 skip: 1840 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 1841 if (msglen > skb->len) 1842 msglen = skb->len; 1843 skb_pull(skb, msglen); 1844 } 1845 1846 return 0; 1847 } 1848 EXPORT_SYMBOL(netlink_rcv_skb); 1849 1850 /** 1851 * nlmsg_notify - send a notification netlink message 1852 * @sk: netlink socket to use 1853 * @skb: notification message 1854 * @pid: destination netlink pid for reports or 0 1855 * @group: destination multicast group or 0 1856 * @report: 1 to report back, 0 to disable 1857 * @flags: allocation flags 1858 */ 1859 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid, 1860 unsigned int group, int report, gfp_t flags) 1861 { 1862 int err = 0; 1863 1864 if (group) { 1865 int exclude_pid = 0; 1866 1867 if (report) { 1868 atomic_inc(&skb->users); 1869 exclude_pid = pid; 1870 } 1871 1872 /* errors reported via destination sk->sk_err, but propagate 1873 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */ 1874 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags); 1875 } 1876 1877 if (report) { 1878 int err2; 1879 1880 err2 = nlmsg_unicast(sk, skb, pid); 1881 if (!err || err == -ESRCH) 1882 err = err2; 1883 } 1884 1885 return err; 1886 } 1887 EXPORT_SYMBOL(nlmsg_notify); 1888 1889 #ifdef CONFIG_PROC_FS 1890 struct nl_seq_iter { 1891 struct seq_net_private p; 1892 int link; 1893 int hash_idx; 1894 }; 1895 1896 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos) 1897 { 1898 struct nl_seq_iter *iter = seq->private; 1899 int i, j; 1900 struct sock *s; 1901 struct hlist_node *node; 1902 loff_t off = 0; 1903 1904 for (i = 0; i < MAX_LINKS; i++) { 1905 struct nl_pid_hash *hash = &nl_table[i].hash; 1906 1907 for (j = 0; j <= hash->mask; j++) { 1908 sk_for_each(s, node, &hash->table[j]) { 1909 if (sock_net(s) != seq_file_net(seq)) 1910 continue; 1911 if (off == pos) { 1912 iter->link = i; 1913 iter->hash_idx = j; 1914 return s; 1915 } 1916 ++off; 1917 } 1918 } 1919 } 1920 return NULL; 1921 } 1922 1923 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos) 1924 __acquires(nl_table_lock) 1925 { 1926 read_lock(&nl_table_lock); 1927 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN; 1928 } 1929 1930 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1931 { 1932 struct sock *s; 1933 struct nl_seq_iter *iter; 1934 int i, j; 1935 1936 ++*pos; 1937 1938 if (v == SEQ_START_TOKEN) 1939 return netlink_seq_socket_idx(seq, 0); 1940 1941 iter = seq->private; 1942 s = v; 1943 do { 1944 s = sk_next(s); 1945 } while (s && sock_net(s) != seq_file_net(seq)); 1946 if (s) 1947 return s; 1948 1949 i = iter->link; 1950 j = iter->hash_idx + 1; 1951 1952 do { 1953 struct nl_pid_hash *hash = &nl_table[i].hash; 1954 1955 for (; j <= hash->mask; j++) { 1956 s = sk_head(&hash->table[j]); 1957 while (s && sock_net(s) != seq_file_net(seq)) 1958 s = sk_next(s); 1959 if (s) { 1960 iter->link = i; 1961 iter->hash_idx = j; 1962 return s; 1963 } 1964 } 1965 1966 j = 0; 1967 } while (++i < MAX_LINKS); 1968 1969 return NULL; 1970 } 1971 1972 static void netlink_seq_stop(struct seq_file *seq, void *v) 1973 __releases(nl_table_lock) 1974 { 1975 read_unlock(&nl_table_lock); 1976 } 1977 1978 1979 static int netlink_seq_show(struct seq_file *seq, void *v) 1980 { 1981 if (v == SEQ_START_TOKEN) 1982 seq_puts(seq, 1983 "sk Eth Pid Groups " 1984 "Rmem Wmem Dump Locks Drops Inode\n"); 1985 else { 1986 struct sock *s = v; 1987 struct netlink_sock *nlk = nlk_sk(s); 1988 1989 seq_printf(seq, "%pK %-3d %-6d %08x %-8d %-8d %pK %-8d %-8d %-8lu\n", 1990 s, 1991 s->sk_protocol, 1992 nlk->pid, 1993 nlk->groups ? (u32)nlk->groups[0] : 0, 1994 sk_rmem_alloc_get(s), 1995 sk_wmem_alloc_get(s), 1996 nlk->cb, 1997 atomic_read(&s->sk_refcnt), 1998 atomic_read(&s->sk_drops), 1999 sock_i_ino(s) 2000 ); 2001 2002 } 2003 return 0; 2004 } 2005 2006 static const struct seq_operations netlink_seq_ops = { 2007 .start = netlink_seq_start, 2008 .next = netlink_seq_next, 2009 .stop = netlink_seq_stop, 2010 .show = netlink_seq_show, 2011 }; 2012 2013 2014 static int netlink_seq_open(struct inode *inode, struct file *file) 2015 { 2016 return seq_open_net(inode, file, &netlink_seq_ops, 2017 sizeof(struct nl_seq_iter)); 2018 } 2019 2020 static const struct file_operations netlink_seq_fops = { 2021 .owner = THIS_MODULE, 2022 .open = netlink_seq_open, 2023 .read = seq_read, 2024 .llseek = seq_lseek, 2025 .release = seq_release_net, 2026 }; 2027 2028 #endif 2029 2030 int netlink_register_notifier(struct notifier_block *nb) 2031 { 2032 return atomic_notifier_chain_register(&netlink_chain, nb); 2033 } 2034 EXPORT_SYMBOL(netlink_register_notifier); 2035 2036 int netlink_unregister_notifier(struct notifier_block *nb) 2037 { 2038 return atomic_notifier_chain_unregister(&netlink_chain, nb); 2039 } 2040 EXPORT_SYMBOL(netlink_unregister_notifier); 2041 2042 static const struct proto_ops netlink_ops = { 2043 .family = PF_NETLINK, 2044 .owner = THIS_MODULE, 2045 .release = netlink_release, 2046 .bind = netlink_bind, 2047 .connect = netlink_connect, 2048 .socketpair = sock_no_socketpair, 2049 .accept = sock_no_accept, 2050 .getname = netlink_getname, 2051 .poll = datagram_poll, 2052 .ioctl = sock_no_ioctl, 2053 .listen = sock_no_listen, 2054 .shutdown = sock_no_shutdown, 2055 .setsockopt = netlink_setsockopt, 2056 .getsockopt = netlink_getsockopt, 2057 .sendmsg = netlink_sendmsg, 2058 .recvmsg = netlink_recvmsg, 2059 .mmap = sock_no_mmap, 2060 .sendpage = sock_no_sendpage, 2061 }; 2062 2063 static const struct net_proto_family netlink_family_ops = { 2064 .family = PF_NETLINK, 2065 .create = netlink_create, 2066 .owner = THIS_MODULE, /* for consistency 8) */ 2067 }; 2068 2069 static int __net_init netlink_net_init(struct net *net) 2070 { 2071 #ifdef CONFIG_PROC_FS 2072 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops)) 2073 return -ENOMEM; 2074 #endif 2075 return 0; 2076 } 2077 2078 static void __net_exit netlink_net_exit(struct net *net) 2079 { 2080 #ifdef CONFIG_PROC_FS 2081 proc_net_remove(net, "netlink"); 2082 #endif 2083 } 2084 2085 static void __init netlink_add_usersock_entry(void) 2086 { 2087 struct listeners *listeners; 2088 int groups = 32; 2089 2090 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 2091 if (!listeners) 2092 panic("netlink_add_usersock_entry: Cannot allocate listeners\n"); 2093 2094 netlink_table_grab(); 2095 2096 nl_table[NETLINK_USERSOCK].groups = groups; 2097 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners); 2098 nl_table[NETLINK_USERSOCK].module = THIS_MODULE; 2099 nl_table[NETLINK_USERSOCK].registered = 1; 2100 2101 netlink_table_ungrab(); 2102 } 2103 2104 static struct pernet_operations __net_initdata netlink_net_ops = { 2105 .init = netlink_net_init, 2106 .exit = netlink_net_exit, 2107 }; 2108 2109 static int __init netlink_proto_init(void) 2110 { 2111 struct sk_buff *dummy_skb; 2112 int i; 2113 unsigned long limit; 2114 unsigned int order; 2115 int err = proto_register(&netlink_proto, 0); 2116 2117 if (err != 0) 2118 goto out; 2119 2120 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb)); 2121 2122 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL); 2123 if (!nl_table) 2124 goto panic; 2125 2126 if (totalram_pages >= (128 * 1024)) 2127 limit = totalram_pages >> (21 - PAGE_SHIFT); 2128 else 2129 limit = totalram_pages >> (23 - PAGE_SHIFT); 2130 2131 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT; 2132 limit = (1UL << order) / sizeof(struct hlist_head); 2133 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1; 2134 2135 for (i = 0; i < MAX_LINKS; i++) { 2136 struct nl_pid_hash *hash = &nl_table[i].hash; 2137 2138 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table)); 2139 if (!hash->table) { 2140 while (i-- > 0) 2141 nl_pid_hash_free(nl_table[i].hash.table, 2142 1 * sizeof(*hash->table)); 2143 kfree(nl_table); 2144 goto panic; 2145 } 2146 hash->max_shift = order; 2147 hash->shift = 0; 2148 hash->mask = 0; 2149 hash->rehash_time = jiffies; 2150 } 2151 2152 netlink_add_usersock_entry(); 2153 2154 sock_register(&netlink_family_ops); 2155 register_pernet_subsys(&netlink_net_ops); 2156 /* The netlink device handler may be needed early. */ 2157 rtnetlink_init(); 2158 out: 2159 return err; 2160 panic: 2161 panic("netlink_init: Cannot allocate nl_table\n"); 2162 } 2163 2164 core_initcall(netlink_proto_init); 2165