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 portid; 71 u32 dst_portid; 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 void (*netlink_bind)(int group); 84 struct module *module; 85 }; 86 87 struct listeners { 88 struct rcu_head rcu; 89 unsigned long masks[0]; 90 }; 91 92 #define NETLINK_KERNEL_SOCKET 0x1 93 #define NETLINK_RECV_PKTINFO 0x2 94 #define NETLINK_BROADCAST_SEND_ERROR 0x4 95 #define NETLINK_RECV_NO_ENOBUFS 0x8 96 97 static inline struct netlink_sock *nlk_sk(struct sock *sk) 98 { 99 return container_of(sk, struct netlink_sock, sk); 100 } 101 102 static inline int netlink_is_kernel(struct sock *sk) 103 { 104 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET; 105 } 106 107 struct nl_portid_hash { 108 struct hlist_head *table; 109 unsigned long rehash_time; 110 111 unsigned int mask; 112 unsigned int shift; 113 114 unsigned int entries; 115 unsigned int max_shift; 116 117 u32 rnd; 118 }; 119 120 struct netlink_table { 121 struct nl_portid_hash hash; 122 struct hlist_head mc_list; 123 struct listeners __rcu *listeners; 124 unsigned int flags; 125 unsigned int groups; 126 struct mutex *cb_mutex; 127 struct module *module; 128 void (*bind)(int group); 129 int registered; 130 }; 131 132 static struct netlink_table *nl_table; 133 134 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait); 135 136 static int netlink_dump(struct sock *sk); 137 138 static DEFINE_RWLOCK(nl_table_lock); 139 static atomic_t nl_table_users = ATOMIC_INIT(0); 140 141 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock)); 142 143 static ATOMIC_NOTIFIER_HEAD(netlink_chain); 144 145 static inline u32 netlink_group_mask(u32 group) 146 { 147 return group ? 1 << (group - 1) : 0; 148 } 149 150 static inline struct hlist_head *nl_portid_hashfn(struct nl_portid_hash *hash, u32 portid) 151 { 152 return &hash->table[jhash_1word(portid, hash->rnd) & hash->mask]; 153 } 154 155 static void netlink_destroy_callback(struct netlink_callback *cb) 156 { 157 kfree_skb(cb->skb); 158 kfree(cb); 159 } 160 161 static void netlink_consume_callback(struct netlink_callback *cb) 162 { 163 consume_skb(cb->skb); 164 kfree(cb); 165 } 166 167 static void netlink_sock_destruct(struct sock *sk) 168 { 169 struct netlink_sock *nlk = nlk_sk(sk); 170 171 if (nlk->cb) { 172 if (nlk->cb->done) 173 nlk->cb->done(nlk->cb); 174 175 module_put(nlk->cb->module); 176 netlink_destroy_callback(nlk->cb); 177 } 178 179 skb_queue_purge(&sk->sk_receive_queue); 180 181 if (!sock_flag(sk, SOCK_DEAD)) { 182 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk); 183 return; 184 } 185 186 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 187 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 188 WARN_ON(nlk_sk(sk)->groups); 189 } 190 191 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on 192 * SMP. Look, when several writers sleep and reader wakes them up, all but one 193 * immediately hit write lock and grab all the cpus. Exclusive sleep solves 194 * this, _but_ remember, it adds useless work on UP machines. 195 */ 196 197 void netlink_table_grab(void) 198 __acquires(nl_table_lock) 199 { 200 might_sleep(); 201 202 write_lock_irq(&nl_table_lock); 203 204 if (atomic_read(&nl_table_users)) { 205 DECLARE_WAITQUEUE(wait, current); 206 207 add_wait_queue_exclusive(&nl_table_wait, &wait); 208 for (;;) { 209 set_current_state(TASK_UNINTERRUPTIBLE); 210 if (atomic_read(&nl_table_users) == 0) 211 break; 212 write_unlock_irq(&nl_table_lock); 213 schedule(); 214 write_lock_irq(&nl_table_lock); 215 } 216 217 __set_current_state(TASK_RUNNING); 218 remove_wait_queue(&nl_table_wait, &wait); 219 } 220 } 221 222 void netlink_table_ungrab(void) 223 __releases(nl_table_lock) 224 { 225 write_unlock_irq(&nl_table_lock); 226 wake_up(&nl_table_wait); 227 } 228 229 static inline void 230 netlink_lock_table(void) 231 { 232 /* read_lock() synchronizes us to netlink_table_grab */ 233 234 read_lock(&nl_table_lock); 235 atomic_inc(&nl_table_users); 236 read_unlock(&nl_table_lock); 237 } 238 239 static inline void 240 netlink_unlock_table(void) 241 { 242 if (atomic_dec_and_test(&nl_table_users)) 243 wake_up(&nl_table_wait); 244 } 245 246 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid) 247 { 248 struct nl_portid_hash *hash = &nl_table[protocol].hash; 249 struct hlist_head *head; 250 struct sock *sk; 251 struct hlist_node *node; 252 253 read_lock(&nl_table_lock); 254 head = nl_portid_hashfn(hash, portid); 255 sk_for_each(sk, node, head) { 256 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->portid == portid)) { 257 sock_hold(sk); 258 goto found; 259 } 260 } 261 sk = NULL; 262 found: 263 read_unlock(&nl_table_lock); 264 return sk; 265 } 266 267 static struct hlist_head *nl_portid_hash_zalloc(size_t size) 268 { 269 if (size <= PAGE_SIZE) 270 return kzalloc(size, GFP_ATOMIC); 271 else 272 return (struct hlist_head *) 273 __get_free_pages(GFP_ATOMIC | __GFP_ZERO, 274 get_order(size)); 275 } 276 277 static void nl_portid_hash_free(struct hlist_head *table, size_t size) 278 { 279 if (size <= PAGE_SIZE) 280 kfree(table); 281 else 282 free_pages((unsigned long)table, get_order(size)); 283 } 284 285 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow) 286 { 287 unsigned int omask, mask, shift; 288 size_t osize, size; 289 struct hlist_head *otable, *table; 290 int i; 291 292 omask = mask = hash->mask; 293 osize = size = (mask + 1) * sizeof(*table); 294 shift = hash->shift; 295 296 if (grow) { 297 if (++shift > hash->max_shift) 298 return 0; 299 mask = mask * 2 + 1; 300 size *= 2; 301 } 302 303 table = nl_portid_hash_zalloc(size); 304 if (!table) 305 return 0; 306 307 otable = hash->table; 308 hash->table = table; 309 hash->mask = mask; 310 hash->shift = shift; 311 get_random_bytes(&hash->rnd, sizeof(hash->rnd)); 312 313 for (i = 0; i <= omask; i++) { 314 struct sock *sk; 315 struct hlist_node *node, *tmp; 316 317 sk_for_each_safe(sk, node, tmp, &otable[i]) 318 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid)); 319 } 320 321 nl_portid_hash_free(otable, osize); 322 hash->rehash_time = jiffies + 10 * 60 * HZ; 323 return 1; 324 } 325 326 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len) 327 { 328 int avg = hash->entries >> hash->shift; 329 330 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1)) 331 return 1; 332 333 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) { 334 nl_portid_hash_rehash(hash, 0); 335 return 1; 336 } 337 338 return 0; 339 } 340 341 static const struct proto_ops netlink_ops; 342 343 static void 344 netlink_update_listeners(struct sock *sk) 345 { 346 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 347 struct hlist_node *node; 348 unsigned long mask; 349 unsigned int i; 350 struct listeners *listeners; 351 352 listeners = nl_deref_protected(tbl->listeners); 353 if (!listeners) 354 return; 355 356 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) { 357 mask = 0; 358 sk_for_each_bound(sk, node, &tbl->mc_list) { 359 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups)) 360 mask |= nlk_sk(sk)->groups[i]; 361 } 362 listeners->masks[i] = mask; 363 } 364 /* this function is only called with the netlink table "grabbed", which 365 * makes sure updates are visible before bind or setsockopt return. */ 366 } 367 368 static int netlink_insert(struct sock *sk, struct net *net, u32 portid) 369 { 370 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash; 371 struct hlist_head *head; 372 int err = -EADDRINUSE; 373 struct sock *osk; 374 struct hlist_node *node; 375 int len; 376 377 netlink_table_grab(); 378 head = nl_portid_hashfn(hash, portid); 379 len = 0; 380 sk_for_each(osk, node, head) { 381 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->portid == portid)) 382 break; 383 len++; 384 } 385 if (node) 386 goto err; 387 388 err = -EBUSY; 389 if (nlk_sk(sk)->portid) 390 goto err; 391 392 err = -ENOMEM; 393 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX)) 394 goto err; 395 396 if (len && nl_portid_hash_dilute(hash, len)) 397 head = nl_portid_hashfn(hash, portid); 398 hash->entries++; 399 nlk_sk(sk)->portid = portid; 400 sk_add_node(sk, head); 401 err = 0; 402 403 err: 404 netlink_table_ungrab(); 405 return err; 406 } 407 408 static void netlink_remove(struct sock *sk) 409 { 410 netlink_table_grab(); 411 if (sk_del_node_init(sk)) 412 nl_table[sk->sk_protocol].hash.entries--; 413 if (nlk_sk(sk)->subscriptions) 414 __sk_del_bind_node(sk); 415 netlink_table_ungrab(); 416 } 417 418 static struct proto netlink_proto = { 419 .name = "NETLINK", 420 .owner = THIS_MODULE, 421 .obj_size = sizeof(struct netlink_sock), 422 }; 423 424 static int __netlink_create(struct net *net, struct socket *sock, 425 struct mutex *cb_mutex, int protocol) 426 { 427 struct sock *sk; 428 struct netlink_sock *nlk; 429 430 sock->ops = &netlink_ops; 431 432 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto); 433 if (!sk) 434 return -ENOMEM; 435 436 sock_init_data(sock, sk); 437 438 nlk = nlk_sk(sk); 439 if (cb_mutex) { 440 nlk->cb_mutex = cb_mutex; 441 } else { 442 nlk->cb_mutex = &nlk->cb_def_mutex; 443 mutex_init(nlk->cb_mutex); 444 } 445 init_waitqueue_head(&nlk->wait); 446 447 sk->sk_destruct = netlink_sock_destruct; 448 sk->sk_protocol = protocol; 449 return 0; 450 } 451 452 static int netlink_create(struct net *net, struct socket *sock, int protocol, 453 int kern) 454 { 455 struct module *module = NULL; 456 struct mutex *cb_mutex; 457 struct netlink_sock *nlk; 458 void (*bind)(int group); 459 int err = 0; 460 461 sock->state = SS_UNCONNECTED; 462 463 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 464 return -ESOCKTNOSUPPORT; 465 466 if (protocol < 0 || protocol >= MAX_LINKS) 467 return -EPROTONOSUPPORT; 468 469 netlink_lock_table(); 470 #ifdef CONFIG_MODULES 471 if (!nl_table[protocol].registered) { 472 netlink_unlock_table(); 473 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol); 474 netlink_lock_table(); 475 } 476 #endif 477 if (nl_table[protocol].registered && 478 try_module_get(nl_table[protocol].module)) 479 module = nl_table[protocol].module; 480 else 481 err = -EPROTONOSUPPORT; 482 cb_mutex = nl_table[protocol].cb_mutex; 483 bind = nl_table[protocol].bind; 484 netlink_unlock_table(); 485 486 if (err < 0) 487 goto out; 488 489 err = __netlink_create(net, sock, cb_mutex, protocol); 490 if (err < 0) 491 goto out_module; 492 493 local_bh_disable(); 494 sock_prot_inuse_add(net, &netlink_proto, 1); 495 local_bh_enable(); 496 497 nlk = nlk_sk(sock->sk); 498 nlk->module = module; 499 nlk->netlink_bind = bind; 500 out: 501 return err; 502 503 out_module: 504 module_put(module); 505 goto out; 506 } 507 508 static int netlink_release(struct socket *sock) 509 { 510 struct sock *sk = sock->sk; 511 struct netlink_sock *nlk; 512 513 if (!sk) 514 return 0; 515 516 netlink_remove(sk); 517 sock_orphan(sk); 518 nlk = nlk_sk(sk); 519 520 /* 521 * OK. Socket is unlinked, any packets that arrive now 522 * will be purged. 523 */ 524 525 sock->sk = NULL; 526 wake_up_interruptible_all(&nlk->wait); 527 528 skb_queue_purge(&sk->sk_write_queue); 529 530 if (nlk->portid) { 531 struct netlink_notify n = { 532 .net = sock_net(sk), 533 .protocol = sk->sk_protocol, 534 .portid = nlk->portid, 535 }; 536 atomic_notifier_call_chain(&netlink_chain, 537 NETLINK_URELEASE, &n); 538 } 539 540 module_put(nlk->module); 541 542 netlink_table_grab(); 543 if (netlink_is_kernel(sk)) { 544 BUG_ON(nl_table[sk->sk_protocol].registered == 0); 545 if (--nl_table[sk->sk_protocol].registered == 0) { 546 struct listeners *old; 547 548 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners); 549 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL); 550 kfree_rcu(old, rcu); 551 nl_table[sk->sk_protocol].module = NULL; 552 nl_table[sk->sk_protocol].bind = NULL; 553 nl_table[sk->sk_protocol].flags = 0; 554 nl_table[sk->sk_protocol].registered = 0; 555 } 556 } else if (nlk->subscriptions) { 557 netlink_update_listeners(sk); 558 } 559 netlink_table_ungrab(); 560 561 kfree(nlk->groups); 562 nlk->groups = NULL; 563 564 local_bh_disable(); 565 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1); 566 local_bh_enable(); 567 sock_put(sk); 568 return 0; 569 } 570 571 static int netlink_autobind(struct socket *sock) 572 { 573 struct sock *sk = sock->sk; 574 struct net *net = sock_net(sk); 575 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash; 576 struct hlist_head *head; 577 struct sock *osk; 578 struct hlist_node *node; 579 s32 portid = task_tgid_vnr(current); 580 int err; 581 static s32 rover = -4097; 582 583 retry: 584 cond_resched(); 585 netlink_table_grab(); 586 head = nl_portid_hashfn(hash, portid); 587 sk_for_each(osk, node, head) { 588 if (!net_eq(sock_net(osk), net)) 589 continue; 590 if (nlk_sk(osk)->portid == portid) { 591 /* Bind collision, search negative portid values. */ 592 portid = rover--; 593 if (rover > -4097) 594 rover = -4097; 595 netlink_table_ungrab(); 596 goto retry; 597 } 598 } 599 netlink_table_ungrab(); 600 601 err = netlink_insert(sk, net, portid); 602 if (err == -EADDRINUSE) 603 goto retry; 604 605 /* If 2 threads race to autobind, that is fine. */ 606 if (err == -EBUSY) 607 err = 0; 608 609 return err; 610 } 611 612 static inline int netlink_capable(const struct socket *sock, unsigned int flag) 613 { 614 return (nl_table[sock->sk->sk_protocol].flags & flag) || 615 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN); 616 } 617 618 static void 619 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions) 620 { 621 struct netlink_sock *nlk = nlk_sk(sk); 622 623 if (nlk->subscriptions && !subscriptions) 624 __sk_del_bind_node(sk); 625 else if (!nlk->subscriptions && subscriptions) 626 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 627 nlk->subscriptions = subscriptions; 628 } 629 630 static int netlink_realloc_groups(struct sock *sk) 631 { 632 struct netlink_sock *nlk = nlk_sk(sk); 633 unsigned int groups; 634 unsigned long *new_groups; 635 int err = 0; 636 637 netlink_table_grab(); 638 639 groups = nl_table[sk->sk_protocol].groups; 640 if (!nl_table[sk->sk_protocol].registered) { 641 err = -ENOENT; 642 goto out_unlock; 643 } 644 645 if (nlk->ngroups >= groups) 646 goto out_unlock; 647 648 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC); 649 if (new_groups == NULL) { 650 err = -ENOMEM; 651 goto out_unlock; 652 } 653 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0, 654 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups)); 655 656 nlk->groups = new_groups; 657 nlk->ngroups = groups; 658 out_unlock: 659 netlink_table_ungrab(); 660 return err; 661 } 662 663 static int netlink_bind(struct socket *sock, struct sockaddr *addr, 664 int addr_len) 665 { 666 struct sock *sk = sock->sk; 667 struct net *net = sock_net(sk); 668 struct netlink_sock *nlk = nlk_sk(sk); 669 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 670 int err; 671 672 if (addr_len < sizeof(struct sockaddr_nl)) 673 return -EINVAL; 674 675 if (nladdr->nl_family != AF_NETLINK) 676 return -EINVAL; 677 678 /* Only superuser is allowed to listen multicasts */ 679 if (nladdr->nl_groups) { 680 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV)) 681 return -EPERM; 682 err = netlink_realloc_groups(sk); 683 if (err) 684 return err; 685 } 686 687 if (nlk->portid) { 688 if (nladdr->nl_pid != nlk->portid) 689 return -EINVAL; 690 } else { 691 err = nladdr->nl_pid ? 692 netlink_insert(sk, net, nladdr->nl_pid) : 693 netlink_autobind(sock); 694 if (err) 695 return err; 696 } 697 698 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0])) 699 return 0; 700 701 netlink_table_grab(); 702 netlink_update_subscriptions(sk, nlk->subscriptions + 703 hweight32(nladdr->nl_groups) - 704 hweight32(nlk->groups[0])); 705 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups; 706 netlink_update_listeners(sk); 707 netlink_table_ungrab(); 708 709 if (nlk->netlink_bind && nlk->groups[0]) { 710 int i; 711 712 for (i=0; i<nlk->ngroups; i++) { 713 if (test_bit(i, nlk->groups)) 714 nlk->netlink_bind(i); 715 } 716 } 717 718 return 0; 719 } 720 721 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 722 int alen, int flags) 723 { 724 int err = 0; 725 struct sock *sk = sock->sk; 726 struct netlink_sock *nlk = nlk_sk(sk); 727 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 728 729 if (alen < sizeof(addr->sa_family)) 730 return -EINVAL; 731 732 if (addr->sa_family == AF_UNSPEC) { 733 sk->sk_state = NETLINK_UNCONNECTED; 734 nlk->dst_portid = 0; 735 nlk->dst_group = 0; 736 return 0; 737 } 738 if (addr->sa_family != AF_NETLINK) 739 return -EINVAL; 740 741 /* Only superuser is allowed to send multicasts */ 742 if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND)) 743 return -EPERM; 744 745 if (!nlk->portid) 746 err = netlink_autobind(sock); 747 748 if (err == 0) { 749 sk->sk_state = NETLINK_CONNECTED; 750 nlk->dst_portid = nladdr->nl_pid; 751 nlk->dst_group = ffs(nladdr->nl_groups); 752 } 753 754 return err; 755 } 756 757 static int netlink_getname(struct socket *sock, struct sockaddr *addr, 758 int *addr_len, int peer) 759 { 760 struct sock *sk = sock->sk; 761 struct netlink_sock *nlk = nlk_sk(sk); 762 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr); 763 764 nladdr->nl_family = AF_NETLINK; 765 nladdr->nl_pad = 0; 766 *addr_len = sizeof(*nladdr); 767 768 if (peer) { 769 nladdr->nl_pid = nlk->dst_portid; 770 nladdr->nl_groups = netlink_group_mask(nlk->dst_group); 771 } else { 772 nladdr->nl_pid = nlk->portid; 773 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0; 774 } 775 return 0; 776 } 777 778 static void netlink_overrun(struct sock *sk) 779 { 780 struct netlink_sock *nlk = nlk_sk(sk); 781 782 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) { 783 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) { 784 sk->sk_err = ENOBUFS; 785 sk->sk_error_report(sk); 786 } 787 } 788 atomic_inc(&sk->sk_drops); 789 } 790 791 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid) 792 { 793 struct sock *sock; 794 struct netlink_sock *nlk; 795 796 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid); 797 if (!sock) 798 return ERR_PTR(-ECONNREFUSED); 799 800 /* Don't bother queuing skb if kernel socket has no input function */ 801 nlk = nlk_sk(sock); 802 if (sock->sk_state == NETLINK_CONNECTED && 803 nlk->dst_portid != nlk_sk(ssk)->portid) { 804 sock_put(sock); 805 return ERR_PTR(-ECONNREFUSED); 806 } 807 return sock; 808 } 809 810 struct sock *netlink_getsockbyfilp(struct file *filp) 811 { 812 struct inode *inode = filp->f_path.dentry->d_inode; 813 struct sock *sock; 814 815 if (!S_ISSOCK(inode->i_mode)) 816 return ERR_PTR(-ENOTSOCK); 817 818 sock = SOCKET_I(inode)->sk; 819 if (sock->sk_family != AF_NETLINK) 820 return ERR_PTR(-EINVAL); 821 822 sock_hold(sock); 823 return sock; 824 } 825 826 /* 827 * Attach a skb to a netlink socket. 828 * The caller must hold a reference to the destination socket. On error, the 829 * reference is dropped. The skb is not send to the destination, just all 830 * all error checks are performed and memory in the queue is reserved. 831 * Return values: 832 * < 0: error. skb freed, reference to sock dropped. 833 * 0: continue 834 * 1: repeat lookup - reference dropped while waiting for socket memory. 835 */ 836 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, 837 long *timeo, struct sock *ssk) 838 { 839 struct netlink_sock *nlk; 840 841 nlk = nlk_sk(sk); 842 843 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 844 test_bit(0, &nlk->state)) { 845 DECLARE_WAITQUEUE(wait, current); 846 if (!*timeo) { 847 if (!ssk || netlink_is_kernel(ssk)) 848 netlink_overrun(sk); 849 sock_put(sk); 850 kfree_skb(skb); 851 return -EAGAIN; 852 } 853 854 __set_current_state(TASK_INTERRUPTIBLE); 855 add_wait_queue(&nlk->wait, &wait); 856 857 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 858 test_bit(0, &nlk->state)) && 859 !sock_flag(sk, SOCK_DEAD)) 860 *timeo = schedule_timeout(*timeo); 861 862 __set_current_state(TASK_RUNNING); 863 remove_wait_queue(&nlk->wait, &wait); 864 sock_put(sk); 865 866 if (signal_pending(current)) { 867 kfree_skb(skb); 868 return sock_intr_errno(*timeo); 869 } 870 return 1; 871 } 872 skb_set_owner_r(skb, sk); 873 return 0; 874 } 875 876 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb) 877 { 878 int len = skb->len; 879 880 skb_queue_tail(&sk->sk_receive_queue, skb); 881 sk->sk_data_ready(sk, len); 882 return len; 883 } 884 885 int netlink_sendskb(struct sock *sk, struct sk_buff *skb) 886 { 887 int len = __netlink_sendskb(sk, skb); 888 889 sock_put(sk); 890 return len; 891 } 892 893 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 894 { 895 kfree_skb(skb); 896 sock_put(sk); 897 } 898 899 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation) 900 { 901 int delta; 902 903 skb_orphan(skb); 904 905 delta = skb->end - skb->tail; 906 if (delta * 2 < skb->truesize) 907 return skb; 908 909 if (skb_shared(skb)) { 910 struct sk_buff *nskb = skb_clone(skb, allocation); 911 if (!nskb) 912 return skb; 913 consume_skb(skb); 914 skb = nskb; 915 } 916 917 if (!pskb_expand_head(skb, 0, -delta, allocation)) 918 skb->truesize -= delta; 919 920 return skb; 921 } 922 923 static void netlink_rcv_wake(struct sock *sk) 924 { 925 struct netlink_sock *nlk = nlk_sk(sk); 926 927 if (skb_queue_empty(&sk->sk_receive_queue)) 928 clear_bit(0, &nlk->state); 929 if (!test_bit(0, &nlk->state)) 930 wake_up_interruptible(&nlk->wait); 931 } 932 933 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb, 934 struct sock *ssk) 935 { 936 int ret; 937 struct netlink_sock *nlk = nlk_sk(sk); 938 939 ret = -ECONNREFUSED; 940 if (nlk->netlink_rcv != NULL) { 941 ret = skb->len; 942 skb_set_owner_r(skb, sk); 943 NETLINK_CB(skb).ssk = ssk; 944 nlk->netlink_rcv(skb); 945 consume_skb(skb); 946 } else { 947 kfree_skb(skb); 948 } 949 sock_put(sk); 950 return ret; 951 } 952 953 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, 954 u32 portid, int nonblock) 955 { 956 struct sock *sk; 957 int err; 958 long timeo; 959 960 skb = netlink_trim(skb, gfp_any()); 961 962 timeo = sock_sndtimeo(ssk, nonblock); 963 retry: 964 sk = netlink_getsockbyportid(ssk, portid); 965 if (IS_ERR(sk)) { 966 kfree_skb(skb); 967 return PTR_ERR(sk); 968 } 969 if (netlink_is_kernel(sk)) 970 return netlink_unicast_kernel(sk, skb, ssk); 971 972 if (sk_filter(sk, skb)) { 973 err = skb->len; 974 kfree_skb(skb); 975 sock_put(sk); 976 return err; 977 } 978 979 err = netlink_attachskb(sk, skb, &timeo, ssk); 980 if (err == 1) 981 goto retry; 982 if (err) 983 return err; 984 985 return netlink_sendskb(sk, skb); 986 } 987 EXPORT_SYMBOL(netlink_unicast); 988 989 int netlink_has_listeners(struct sock *sk, unsigned int group) 990 { 991 int res = 0; 992 struct listeners *listeners; 993 994 BUG_ON(!netlink_is_kernel(sk)); 995 996 rcu_read_lock(); 997 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners); 998 999 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups) 1000 res = test_bit(group - 1, listeners->masks); 1001 1002 rcu_read_unlock(); 1003 1004 return res; 1005 } 1006 EXPORT_SYMBOL_GPL(netlink_has_listeners); 1007 1008 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 1009 { 1010 struct netlink_sock *nlk = nlk_sk(sk); 1011 1012 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 1013 !test_bit(0, &nlk->state)) { 1014 skb_set_owner_r(skb, sk); 1015 __netlink_sendskb(sk, skb); 1016 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1); 1017 } 1018 return -1; 1019 } 1020 1021 struct netlink_broadcast_data { 1022 struct sock *exclude_sk; 1023 struct net *net; 1024 u32 portid; 1025 u32 group; 1026 int failure; 1027 int delivery_failure; 1028 int congested; 1029 int delivered; 1030 gfp_t allocation; 1031 struct sk_buff *skb, *skb2; 1032 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data); 1033 void *tx_data; 1034 }; 1035 1036 static int do_one_broadcast(struct sock *sk, 1037 struct netlink_broadcast_data *p) 1038 { 1039 struct netlink_sock *nlk = nlk_sk(sk); 1040 int val; 1041 1042 if (p->exclude_sk == sk) 1043 goto out; 1044 1045 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 1046 !test_bit(p->group - 1, nlk->groups)) 1047 goto out; 1048 1049 if (!net_eq(sock_net(sk), p->net)) 1050 goto out; 1051 1052 if (p->failure) { 1053 netlink_overrun(sk); 1054 goto out; 1055 } 1056 1057 sock_hold(sk); 1058 if (p->skb2 == NULL) { 1059 if (skb_shared(p->skb)) { 1060 p->skb2 = skb_clone(p->skb, p->allocation); 1061 } else { 1062 p->skb2 = skb_get(p->skb); 1063 /* 1064 * skb ownership may have been set when 1065 * delivered to a previous socket. 1066 */ 1067 skb_orphan(p->skb2); 1068 } 1069 } 1070 if (p->skb2 == NULL) { 1071 netlink_overrun(sk); 1072 /* Clone failed. Notify ALL listeners. */ 1073 p->failure = 1; 1074 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR) 1075 p->delivery_failure = 1; 1076 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) { 1077 kfree_skb(p->skb2); 1078 p->skb2 = NULL; 1079 } else if (sk_filter(sk, p->skb2)) { 1080 kfree_skb(p->skb2); 1081 p->skb2 = NULL; 1082 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) { 1083 netlink_overrun(sk); 1084 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR) 1085 p->delivery_failure = 1; 1086 } else { 1087 p->congested |= val; 1088 p->delivered = 1; 1089 p->skb2 = NULL; 1090 } 1091 sock_put(sk); 1092 1093 out: 1094 return 0; 1095 } 1096 1097 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid, 1098 u32 group, gfp_t allocation, 1099 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data), 1100 void *filter_data) 1101 { 1102 struct net *net = sock_net(ssk); 1103 struct netlink_broadcast_data info; 1104 struct hlist_node *node; 1105 struct sock *sk; 1106 1107 skb = netlink_trim(skb, allocation); 1108 1109 info.exclude_sk = ssk; 1110 info.net = net; 1111 info.portid = portid; 1112 info.group = group; 1113 info.failure = 0; 1114 info.delivery_failure = 0; 1115 info.congested = 0; 1116 info.delivered = 0; 1117 info.allocation = allocation; 1118 info.skb = skb; 1119 info.skb2 = NULL; 1120 info.tx_filter = filter; 1121 info.tx_data = filter_data; 1122 1123 /* While we sleep in clone, do not allow to change socket list */ 1124 1125 netlink_lock_table(); 1126 1127 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 1128 do_one_broadcast(sk, &info); 1129 1130 consume_skb(skb); 1131 1132 netlink_unlock_table(); 1133 1134 if (info.delivery_failure) { 1135 kfree_skb(info.skb2); 1136 return -ENOBUFS; 1137 } 1138 consume_skb(info.skb2); 1139 1140 if (info.delivered) { 1141 if (info.congested && (allocation & __GFP_WAIT)) 1142 yield(); 1143 return 0; 1144 } 1145 return -ESRCH; 1146 } 1147 EXPORT_SYMBOL(netlink_broadcast_filtered); 1148 1149 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid, 1150 u32 group, gfp_t allocation) 1151 { 1152 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation, 1153 NULL, NULL); 1154 } 1155 EXPORT_SYMBOL(netlink_broadcast); 1156 1157 struct netlink_set_err_data { 1158 struct sock *exclude_sk; 1159 u32 portid; 1160 u32 group; 1161 int code; 1162 }; 1163 1164 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p) 1165 { 1166 struct netlink_sock *nlk = nlk_sk(sk); 1167 int ret = 0; 1168 1169 if (sk == p->exclude_sk) 1170 goto out; 1171 1172 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk))) 1173 goto out; 1174 1175 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || 1176 !test_bit(p->group - 1, nlk->groups)) 1177 goto out; 1178 1179 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) { 1180 ret = 1; 1181 goto out; 1182 } 1183 1184 sk->sk_err = p->code; 1185 sk->sk_error_report(sk); 1186 out: 1187 return ret; 1188 } 1189 1190 /** 1191 * netlink_set_err - report error to broadcast listeners 1192 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create() 1193 * @portid: the PORTID of a process that we want to skip (if any) 1194 * @groups: the broadcast group that will notice the error 1195 * @code: error code, must be negative (as usual in kernelspace) 1196 * 1197 * This function returns the number of broadcast listeners that have set the 1198 * NETLINK_RECV_NO_ENOBUFS socket option. 1199 */ 1200 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code) 1201 { 1202 struct netlink_set_err_data info; 1203 struct hlist_node *node; 1204 struct sock *sk; 1205 int ret = 0; 1206 1207 info.exclude_sk = ssk; 1208 info.portid = portid; 1209 info.group = group; 1210 /* sk->sk_err wants a positive error value */ 1211 info.code = -code; 1212 1213 read_lock(&nl_table_lock); 1214 1215 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 1216 ret += do_one_set_err(sk, &info); 1217 1218 read_unlock(&nl_table_lock); 1219 return ret; 1220 } 1221 EXPORT_SYMBOL(netlink_set_err); 1222 1223 /* must be called with netlink table grabbed */ 1224 static void netlink_update_socket_mc(struct netlink_sock *nlk, 1225 unsigned int group, 1226 int is_new) 1227 { 1228 int old, new = !!is_new, subscriptions; 1229 1230 old = test_bit(group - 1, nlk->groups); 1231 subscriptions = nlk->subscriptions - old + new; 1232 if (new) 1233 __set_bit(group - 1, nlk->groups); 1234 else 1235 __clear_bit(group - 1, nlk->groups); 1236 netlink_update_subscriptions(&nlk->sk, subscriptions); 1237 netlink_update_listeners(&nlk->sk); 1238 } 1239 1240 static int netlink_setsockopt(struct socket *sock, int level, int optname, 1241 char __user *optval, unsigned int optlen) 1242 { 1243 struct sock *sk = sock->sk; 1244 struct netlink_sock *nlk = nlk_sk(sk); 1245 unsigned int val = 0; 1246 int err; 1247 1248 if (level != SOL_NETLINK) 1249 return -ENOPROTOOPT; 1250 1251 if (optlen >= sizeof(int) && 1252 get_user(val, (unsigned int __user *)optval)) 1253 return -EFAULT; 1254 1255 switch (optname) { 1256 case NETLINK_PKTINFO: 1257 if (val) 1258 nlk->flags |= NETLINK_RECV_PKTINFO; 1259 else 1260 nlk->flags &= ~NETLINK_RECV_PKTINFO; 1261 err = 0; 1262 break; 1263 case NETLINK_ADD_MEMBERSHIP: 1264 case NETLINK_DROP_MEMBERSHIP: { 1265 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV)) 1266 return -EPERM; 1267 err = netlink_realloc_groups(sk); 1268 if (err) 1269 return err; 1270 if (!val || val - 1 >= nlk->ngroups) 1271 return -EINVAL; 1272 netlink_table_grab(); 1273 netlink_update_socket_mc(nlk, val, 1274 optname == NETLINK_ADD_MEMBERSHIP); 1275 netlink_table_ungrab(); 1276 1277 if (nlk->netlink_bind) 1278 nlk->netlink_bind(val); 1279 1280 err = 0; 1281 break; 1282 } 1283 case NETLINK_BROADCAST_ERROR: 1284 if (val) 1285 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR; 1286 else 1287 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR; 1288 err = 0; 1289 break; 1290 case NETLINK_NO_ENOBUFS: 1291 if (val) { 1292 nlk->flags |= NETLINK_RECV_NO_ENOBUFS; 1293 clear_bit(0, &nlk->state); 1294 wake_up_interruptible(&nlk->wait); 1295 } else { 1296 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS; 1297 } 1298 err = 0; 1299 break; 1300 default: 1301 err = -ENOPROTOOPT; 1302 } 1303 return err; 1304 } 1305 1306 static int netlink_getsockopt(struct socket *sock, int level, int optname, 1307 char __user *optval, int __user *optlen) 1308 { 1309 struct sock *sk = sock->sk; 1310 struct netlink_sock *nlk = nlk_sk(sk); 1311 int len, val, err; 1312 1313 if (level != SOL_NETLINK) 1314 return -ENOPROTOOPT; 1315 1316 if (get_user(len, optlen)) 1317 return -EFAULT; 1318 if (len < 0) 1319 return -EINVAL; 1320 1321 switch (optname) { 1322 case NETLINK_PKTINFO: 1323 if (len < sizeof(int)) 1324 return -EINVAL; 1325 len = sizeof(int); 1326 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0; 1327 if (put_user(len, optlen) || 1328 put_user(val, optval)) 1329 return -EFAULT; 1330 err = 0; 1331 break; 1332 case NETLINK_BROADCAST_ERROR: 1333 if (len < sizeof(int)) 1334 return -EINVAL; 1335 len = sizeof(int); 1336 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0; 1337 if (put_user(len, optlen) || 1338 put_user(val, optval)) 1339 return -EFAULT; 1340 err = 0; 1341 break; 1342 case NETLINK_NO_ENOBUFS: 1343 if (len < sizeof(int)) 1344 return -EINVAL; 1345 len = sizeof(int); 1346 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0; 1347 if (put_user(len, optlen) || 1348 put_user(val, optval)) 1349 return -EFAULT; 1350 err = 0; 1351 break; 1352 default: 1353 err = -ENOPROTOOPT; 1354 } 1355 return err; 1356 } 1357 1358 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb) 1359 { 1360 struct nl_pktinfo info; 1361 1362 info.group = NETLINK_CB(skb).dst_group; 1363 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info); 1364 } 1365 1366 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock, 1367 struct msghdr *msg, size_t len) 1368 { 1369 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1370 struct sock *sk = sock->sk; 1371 struct netlink_sock *nlk = nlk_sk(sk); 1372 struct sockaddr_nl *addr = msg->msg_name; 1373 u32 dst_portid; 1374 u32 dst_group; 1375 struct sk_buff *skb; 1376 int err; 1377 struct scm_cookie scm; 1378 1379 if (msg->msg_flags&MSG_OOB) 1380 return -EOPNOTSUPP; 1381 1382 if (NULL == siocb->scm) 1383 siocb->scm = &scm; 1384 1385 err = scm_send(sock, msg, siocb->scm, true); 1386 if (err < 0) 1387 return err; 1388 1389 if (msg->msg_namelen) { 1390 err = -EINVAL; 1391 if (addr->nl_family != AF_NETLINK) 1392 goto out; 1393 dst_portid = addr->nl_pid; 1394 dst_group = ffs(addr->nl_groups); 1395 err = -EPERM; 1396 if ((dst_group || dst_portid) && 1397 !netlink_capable(sock, NL_CFG_F_NONROOT_SEND)) 1398 goto out; 1399 } else { 1400 dst_portid = nlk->dst_portid; 1401 dst_group = nlk->dst_group; 1402 } 1403 1404 if (!nlk->portid) { 1405 err = netlink_autobind(sock); 1406 if (err) 1407 goto out; 1408 } 1409 1410 err = -EMSGSIZE; 1411 if (len > sk->sk_sndbuf - 32) 1412 goto out; 1413 err = -ENOBUFS; 1414 skb = alloc_skb(len, GFP_KERNEL); 1415 if (skb == NULL) 1416 goto out; 1417 1418 NETLINK_CB(skb).portid = nlk->portid; 1419 NETLINK_CB(skb).dst_group = dst_group; 1420 NETLINK_CB(skb).creds = siocb->scm->creds; 1421 1422 err = -EFAULT; 1423 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) { 1424 kfree_skb(skb); 1425 goto out; 1426 } 1427 1428 err = security_netlink_send(sk, skb); 1429 if (err) { 1430 kfree_skb(skb); 1431 goto out; 1432 } 1433 1434 if (dst_group) { 1435 atomic_inc(&skb->users); 1436 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL); 1437 } 1438 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT); 1439 1440 out: 1441 scm_destroy(siocb->scm); 1442 return err; 1443 } 1444 1445 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock, 1446 struct msghdr *msg, size_t len, 1447 int flags) 1448 { 1449 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1450 struct scm_cookie scm; 1451 struct sock *sk = sock->sk; 1452 struct netlink_sock *nlk = nlk_sk(sk); 1453 int noblock = flags&MSG_DONTWAIT; 1454 size_t copied; 1455 struct sk_buff *skb, *data_skb; 1456 int err, ret; 1457 1458 if (flags&MSG_OOB) 1459 return -EOPNOTSUPP; 1460 1461 copied = 0; 1462 1463 skb = skb_recv_datagram(sk, flags, noblock, &err); 1464 if (skb == NULL) 1465 goto out; 1466 1467 data_skb = skb; 1468 1469 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES 1470 if (unlikely(skb_shinfo(skb)->frag_list)) { 1471 /* 1472 * If this skb has a frag_list, then here that means that we 1473 * will have to use the frag_list skb's data for compat tasks 1474 * and the regular skb's data for normal (non-compat) tasks. 1475 * 1476 * If we need to send the compat skb, assign it to the 1477 * 'data_skb' variable so that it will be used below for data 1478 * copying. We keep 'skb' for everything else, including 1479 * freeing both later. 1480 */ 1481 if (flags & MSG_CMSG_COMPAT) 1482 data_skb = skb_shinfo(skb)->frag_list; 1483 } 1484 #endif 1485 1486 msg->msg_namelen = 0; 1487 1488 copied = data_skb->len; 1489 if (len < copied) { 1490 msg->msg_flags |= MSG_TRUNC; 1491 copied = len; 1492 } 1493 1494 skb_reset_transport_header(data_skb); 1495 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied); 1496 1497 if (msg->msg_name) { 1498 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name; 1499 addr->nl_family = AF_NETLINK; 1500 addr->nl_pad = 0; 1501 addr->nl_pid = NETLINK_CB(skb).portid; 1502 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group); 1503 msg->msg_namelen = sizeof(*addr); 1504 } 1505 1506 if (nlk->flags & NETLINK_RECV_PKTINFO) 1507 netlink_cmsg_recv_pktinfo(msg, skb); 1508 1509 if (NULL == siocb->scm) { 1510 memset(&scm, 0, sizeof(scm)); 1511 siocb->scm = &scm; 1512 } 1513 siocb->scm->creds = *NETLINK_CREDS(skb); 1514 if (flags & MSG_TRUNC) 1515 copied = data_skb->len; 1516 1517 skb_free_datagram(sk, skb); 1518 1519 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) { 1520 ret = netlink_dump(sk); 1521 if (ret) { 1522 sk->sk_err = ret; 1523 sk->sk_error_report(sk); 1524 } 1525 } 1526 1527 scm_recv(sock, msg, siocb->scm, flags); 1528 out: 1529 netlink_rcv_wake(sk); 1530 return err ? : copied; 1531 } 1532 1533 static void netlink_data_ready(struct sock *sk, int len) 1534 { 1535 BUG(); 1536 } 1537 1538 /* 1539 * We export these functions to other modules. They provide a 1540 * complete set of kernel non-blocking support for message 1541 * queueing. 1542 */ 1543 1544 struct sock * 1545 __netlink_kernel_create(struct net *net, int unit, struct module *module, 1546 struct netlink_kernel_cfg *cfg) 1547 { 1548 struct socket *sock; 1549 struct sock *sk; 1550 struct netlink_sock *nlk; 1551 struct listeners *listeners = NULL; 1552 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL; 1553 unsigned int groups; 1554 1555 BUG_ON(!nl_table); 1556 1557 if (unit < 0 || unit >= MAX_LINKS) 1558 return NULL; 1559 1560 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 1561 return NULL; 1562 1563 /* 1564 * We have to just have a reference on the net from sk, but don't 1565 * get_net it. Besides, we cannot get and then put the net here. 1566 * So we create one inside init_net and the move it to net. 1567 */ 1568 1569 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0) 1570 goto out_sock_release_nosk; 1571 1572 sk = sock->sk; 1573 sk_change_net(sk, net); 1574 1575 if (!cfg || cfg->groups < 32) 1576 groups = 32; 1577 else 1578 groups = cfg->groups; 1579 1580 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 1581 if (!listeners) 1582 goto out_sock_release; 1583 1584 sk->sk_data_ready = netlink_data_ready; 1585 if (cfg && cfg->input) 1586 nlk_sk(sk)->netlink_rcv = cfg->input; 1587 1588 if (netlink_insert(sk, net, 0)) 1589 goto out_sock_release; 1590 1591 nlk = nlk_sk(sk); 1592 nlk->flags |= NETLINK_KERNEL_SOCKET; 1593 1594 netlink_table_grab(); 1595 if (!nl_table[unit].registered) { 1596 nl_table[unit].groups = groups; 1597 rcu_assign_pointer(nl_table[unit].listeners, listeners); 1598 nl_table[unit].cb_mutex = cb_mutex; 1599 nl_table[unit].module = module; 1600 if (cfg) { 1601 nl_table[unit].bind = cfg->bind; 1602 nl_table[unit].flags = cfg->flags; 1603 } 1604 nl_table[unit].registered = 1; 1605 } else { 1606 kfree(listeners); 1607 nl_table[unit].registered++; 1608 } 1609 netlink_table_ungrab(); 1610 return sk; 1611 1612 out_sock_release: 1613 kfree(listeners); 1614 netlink_kernel_release(sk); 1615 return NULL; 1616 1617 out_sock_release_nosk: 1618 sock_release(sock); 1619 return NULL; 1620 } 1621 EXPORT_SYMBOL(__netlink_kernel_create); 1622 1623 void 1624 netlink_kernel_release(struct sock *sk) 1625 { 1626 sk_release_kernel(sk); 1627 } 1628 EXPORT_SYMBOL(netlink_kernel_release); 1629 1630 int __netlink_change_ngroups(struct sock *sk, unsigned int groups) 1631 { 1632 struct listeners *new, *old; 1633 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 1634 1635 if (groups < 32) 1636 groups = 32; 1637 1638 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) { 1639 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC); 1640 if (!new) 1641 return -ENOMEM; 1642 old = nl_deref_protected(tbl->listeners); 1643 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups)); 1644 rcu_assign_pointer(tbl->listeners, new); 1645 1646 kfree_rcu(old, rcu); 1647 } 1648 tbl->groups = groups; 1649 1650 return 0; 1651 } 1652 1653 /** 1654 * netlink_change_ngroups - change number of multicast groups 1655 * 1656 * This changes the number of multicast groups that are available 1657 * on a certain netlink family. Note that it is not possible to 1658 * change the number of groups to below 32. Also note that it does 1659 * not implicitly call netlink_clear_multicast_users() when the 1660 * number of groups is reduced. 1661 * 1662 * @sk: The kernel netlink socket, as returned by netlink_kernel_create(). 1663 * @groups: The new number of groups. 1664 */ 1665 int netlink_change_ngroups(struct sock *sk, unsigned int groups) 1666 { 1667 int err; 1668 1669 netlink_table_grab(); 1670 err = __netlink_change_ngroups(sk, groups); 1671 netlink_table_ungrab(); 1672 1673 return err; 1674 } 1675 1676 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group) 1677 { 1678 struct sock *sk; 1679 struct hlist_node *node; 1680 struct netlink_table *tbl = &nl_table[ksk->sk_protocol]; 1681 1682 sk_for_each_bound(sk, node, &tbl->mc_list) 1683 netlink_update_socket_mc(nlk_sk(sk), group, 0); 1684 } 1685 1686 /** 1687 * netlink_clear_multicast_users - kick off multicast listeners 1688 * 1689 * This function removes all listeners from the given group. 1690 * @ksk: The kernel netlink socket, as returned by 1691 * netlink_kernel_create(). 1692 * @group: The multicast group to clear. 1693 */ 1694 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group) 1695 { 1696 netlink_table_grab(); 1697 __netlink_clear_multicast_users(ksk, group); 1698 netlink_table_ungrab(); 1699 } 1700 1701 struct nlmsghdr * 1702 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags) 1703 { 1704 struct nlmsghdr *nlh; 1705 int size = NLMSG_LENGTH(len); 1706 1707 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size)); 1708 nlh->nlmsg_type = type; 1709 nlh->nlmsg_len = size; 1710 nlh->nlmsg_flags = flags; 1711 nlh->nlmsg_pid = portid; 1712 nlh->nlmsg_seq = seq; 1713 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0) 1714 memset(NLMSG_DATA(nlh) + len, 0, NLMSG_ALIGN(size) - size); 1715 return nlh; 1716 } 1717 EXPORT_SYMBOL(__nlmsg_put); 1718 1719 /* 1720 * It looks a bit ugly. 1721 * It would be better to create kernel thread. 1722 */ 1723 1724 static int netlink_dump(struct sock *sk) 1725 { 1726 struct netlink_sock *nlk = nlk_sk(sk); 1727 struct netlink_callback *cb; 1728 struct sk_buff *skb = NULL; 1729 struct nlmsghdr *nlh; 1730 int len, err = -ENOBUFS; 1731 int alloc_size; 1732 1733 mutex_lock(nlk->cb_mutex); 1734 1735 cb = nlk->cb; 1736 if (cb == NULL) { 1737 err = -EINVAL; 1738 goto errout_skb; 1739 } 1740 1741 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE); 1742 1743 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL); 1744 if (!skb) 1745 goto errout_skb; 1746 1747 len = cb->dump(skb, cb); 1748 1749 if (len > 0) { 1750 mutex_unlock(nlk->cb_mutex); 1751 1752 if (sk_filter(sk, skb)) 1753 kfree_skb(skb); 1754 else 1755 __netlink_sendskb(sk, skb); 1756 return 0; 1757 } 1758 1759 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI); 1760 if (!nlh) 1761 goto errout_skb; 1762 1763 nl_dump_check_consistent(cb, nlh); 1764 1765 memcpy(nlmsg_data(nlh), &len, sizeof(len)); 1766 1767 if (sk_filter(sk, skb)) 1768 kfree_skb(skb); 1769 else 1770 __netlink_sendskb(sk, skb); 1771 1772 if (cb->done) 1773 cb->done(cb); 1774 nlk->cb = NULL; 1775 mutex_unlock(nlk->cb_mutex); 1776 1777 module_put(cb->module); 1778 netlink_consume_callback(cb); 1779 return 0; 1780 1781 errout_skb: 1782 mutex_unlock(nlk->cb_mutex); 1783 kfree_skb(skb); 1784 return err; 1785 } 1786 1787 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 1788 const struct nlmsghdr *nlh, 1789 struct netlink_dump_control *control) 1790 { 1791 struct netlink_callback *cb; 1792 struct sock *sk; 1793 struct netlink_sock *nlk; 1794 int ret; 1795 1796 cb = kzalloc(sizeof(*cb), GFP_KERNEL); 1797 if (cb == NULL) 1798 return -ENOBUFS; 1799 1800 cb->dump = control->dump; 1801 cb->done = control->done; 1802 cb->nlh = nlh; 1803 cb->data = control->data; 1804 cb->module = control->module; 1805 cb->min_dump_alloc = control->min_dump_alloc; 1806 atomic_inc(&skb->users); 1807 cb->skb = skb; 1808 1809 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid); 1810 if (sk == NULL) { 1811 netlink_destroy_callback(cb); 1812 return -ECONNREFUSED; 1813 } 1814 nlk = nlk_sk(sk); 1815 1816 mutex_lock(nlk->cb_mutex); 1817 /* A dump is in progress... */ 1818 if (nlk->cb) { 1819 mutex_unlock(nlk->cb_mutex); 1820 netlink_destroy_callback(cb); 1821 ret = -EBUSY; 1822 goto out; 1823 } 1824 /* add reference of module which cb->dump belongs to */ 1825 if (!try_module_get(cb->module)) { 1826 mutex_unlock(nlk->cb_mutex); 1827 netlink_destroy_callback(cb); 1828 ret = -EPROTONOSUPPORT; 1829 goto out; 1830 } 1831 1832 nlk->cb = cb; 1833 mutex_unlock(nlk->cb_mutex); 1834 1835 ret = netlink_dump(sk); 1836 out: 1837 sock_put(sk); 1838 1839 if (ret) 1840 return ret; 1841 1842 /* We successfully started a dump, by returning -EINTR we 1843 * signal not to send ACK even if it was requested. 1844 */ 1845 return -EINTR; 1846 } 1847 EXPORT_SYMBOL(__netlink_dump_start); 1848 1849 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err) 1850 { 1851 struct sk_buff *skb; 1852 struct nlmsghdr *rep; 1853 struct nlmsgerr *errmsg; 1854 size_t payload = sizeof(*errmsg); 1855 1856 /* error messages get the original request appened */ 1857 if (err) 1858 payload += nlmsg_len(nlh); 1859 1860 skb = nlmsg_new(payload, GFP_KERNEL); 1861 if (!skb) { 1862 struct sock *sk; 1863 1864 sk = netlink_lookup(sock_net(in_skb->sk), 1865 in_skb->sk->sk_protocol, 1866 NETLINK_CB(in_skb).portid); 1867 if (sk) { 1868 sk->sk_err = ENOBUFS; 1869 sk->sk_error_report(sk); 1870 sock_put(sk); 1871 } 1872 return; 1873 } 1874 1875 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, 1876 NLMSG_ERROR, payload, 0); 1877 errmsg = nlmsg_data(rep); 1878 errmsg->error = err; 1879 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh)); 1880 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT); 1881 } 1882 EXPORT_SYMBOL(netlink_ack); 1883 1884 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *, 1885 struct nlmsghdr *)) 1886 { 1887 struct nlmsghdr *nlh; 1888 int err; 1889 1890 while (skb->len >= nlmsg_total_size(0)) { 1891 int msglen; 1892 1893 nlh = nlmsg_hdr(skb); 1894 err = 0; 1895 1896 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len) 1897 return 0; 1898 1899 /* Only requests are handled by the kernel */ 1900 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) 1901 goto ack; 1902 1903 /* Skip control messages */ 1904 if (nlh->nlmsg_type < NLMSG_MIN_TYPE) 1905 goto ack; 1906 1907 err = cb(skb, nlh); 1908 if (err == -EINTR) 1909 goto skip; 1910 1911 ack: 1912 if (nlh->nlmsg_flags & NLM_F_ACK || err) 1913 netlink_ack(skb, nlh, err); 1914 1915 skip: 1916 msglen = NLMSG_ALIGN(nlh->nlmsg_len); 1917 if (msglen > skb->len) 1918 msglen = skb->len; 1919 skb_pull(skb, msglen); 1920 } 1921 1922 return 0; 1923 } 1924 EXPORT_SYMBOL(netlink_rcv_skb); 1925 1926 /** 1927 * nlmsg_notify - send a notification netlink message 1928 * @sk: netlink socket to use 1929 * @skb: notification message 1930 * @portid: destination netlink portid for reports or 0 1931 * @group: destination multicast group or 0 1932 * @report: 1 to report back, 0 to disable 1933 * @flags: allocation flags 1934 */ 1935 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid, 1936 unsigned int group, int report, gfp_t flags) 1937 { 1938 int err = 0; 1939 1940 if (group) { 1941 int exclude_portid = 0; 1942 1943 if (report) { 1944 atomic_inc(&skb->users); 1945 exclude_portid = portid; 1946 } 1947 1948 /* errors reported via destination sk->sk_err, but propagate 1949 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */ 1950 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags); 1951 } 1952 1953 if (report) { 1954 int err2; 1955 1956 err2 = nlmsg_unicast(sk, skb, portid); 1957 if (!err || err == -ESRCH) 1958 err = err2; 1959 } 1960 1961 return err; 1962 } 1963 EXPORT_SYMBOL(nlmsg_notify); 1964 1965 #ifdef CONFIG_PROC_FS 1966 struct nl_seq_iter { 1967 struct seq_net_private p; 1968 int link; 1969 int hash_idx; 1970 }; 1971 1972 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos) 1973 { 1974 struct nl_seq_iter *iter = seq->private; 1975 int i, j; 1976 struct sock *s; 1977 struct hlist_node *node; 1978 loff_t off = 0; 1979 1980 for (i = 0; i < MAX_LINKS; i++) { 1981 struct nl_portid_hash *hash = &nl_table[i].hash; 1982 1983 for (j = 0; j <= hash->mask; j++) { 1984 sk_for_each(s, node, &hash->table[j]) { 1985 if (sock_net(s) != seq_file_net(seq)) 1986 continue; 1987 if (off == pos) { 1988 iter->link = i; 1989 iter->hash_idx = j; 1990 return s; 1991 } 1992 ++off; 1993 } 1994 } 1995 } 1996 return NULL; 1997 } 1998 1999 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos) 2000 __acquires(nl_table_lock) 2001 { 2002 read_lock(&nl_table_lock); 2003 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN; 2004 } 2005 2006 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2007 { 2008 struct sock *s; 2009 struct nl_seq_iter *iter; 2010 int i, j; 2011 2012 ++*pos; 2013 2014 if (v == SEQ_START_TOKEN) 2015 return netlink_seq_socket_idx(seq, 0); 2016 2017 iter = seq->private; 2018 s = v; 2019 do { 2020 s = sk_next(s); 2021 } while (s && sock_net(s) != seq_file_net(seq)); 2022 if (s) 2023 return s; 2024 2025 i = iter->link; 2026 j = iter->hash_idx + 1; 2027 2028 do { 2029 struct nl_portid_hash *hash = &nl_table[i].hash; 2030 2031 for (; j <= hash->mask; j++) { 2032 s = sk_head(&hash->table[j]); 2033 while (s && sock_net(s) != seq_file_net(seq)) 2034 s = sk_next(s); 2035 if (s) { 2036 iter->link = i; 2037 iter->hash_idx = j; 2038 return s; 2039 } 2040 } 2041 2042 j = 0; 2043 } while (++i < MAX_LINKS); 2044 2045 return NULL; 2046 } 2047 2048 static void netlink_seq_stop(struct seq_file *seq, void *v) 2049 __releases(nl_table_lock) 2050 { 2051 read_unlock(&nl_table_lock); 2052 } 2053 2054 2055 static int netlink_seq_show(struct seq_file *seq, void *v) 2056 { 2057 if (v == SEQ_START_TOKEN) { 2058 seq_puts(seq, 2059 "sk Eth Pid Groups " 2060 "Rmem Wmem Dump Locks Drops Inode\n"); 2061 } else { 2062 struct sock *s = v; 2063 struct netlink_sock *nlk = nlk_sk(s); 2064 2065 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %pK %-8d %-8d %-8lu\n", 2066 s, 2067 s->sk_protocol, 2068 nlk->portid, 2069 nlk->groups ? (u32)nlk->groups[0] : 0, 2070 sk_rmem_alloc_get(s), 2071 sk_wmem_alloc_get(s), 2072 nlk->cb, 2073 atomic_read(&s->sk_refcnt), 2074 atomic_read(&s->sk_drops), 2075 sock_i_ino(s) 2076 ); 2077 2078 } 2079 return 0; 2080 } 2081 2082 static const struct seq_operations netlink_seq_ops = { 2083 .start = netlink_seq_start, 2084 .next = netlink_seq_next, 2085 .stop = netlink_seq_stop, 2086 .show = netlink_seq_show, 2087 }; 2088 2089 2090 static int netlink_seq_open(struct inode *inode, struct file *file) 2091 { 2092 return seq_open_net(inode, file, &netlink_seq_ops, 2093 sizeof(struct nl_seq_iter)); 2094 } 2095 2096 static const struct file_operations netlink_seq_fops = { 2097 .owner = THIS_MODULE, 2098 .open = netlink_seq_open, 2099 .read = seq_read, 2100 .llseek = seq_lseek, 2101 .release = seq_release_net, 2102 }; 2103 2104 #endif 2105 2106 int netlink_register_notifier(struct notifier_block *nb) 2107 { 2108 return atomic_notifier_chain_register(&netlink_chain, nb); 2109 } 2110 EXPORT_SYMBOL(netlink_register_notifier); 2111 2112 int netlink_unregister_notifier(struct notifier_block *nb) 2113 { 2114 return atomic_notifier_chain_unregister(&netlink_chain, nb); 2115 } 2116 EXPORT_SYMBOL(netlink_unregister_notifier); 2117 2118 static const struct proto_ops netlink_ops = { 2119 .family = PF_NETLINK, 2120 .owner = THIS_MODULE, 2121 .release = netlink_release, 2122 .bind = netlink_bind, 2123 .connect = netlink_connect, 2124 .socketpair = sock_no_socketpair, 2125 .accept = sock_no_accept, 2126 .getname = netlink_getname, 2127 .poll = datagram_poll, 2128 .ioctl = sock_no_ioctl, 2129 .listen = sock_no_listen, 2130 .shutdown = sock_no_shutdown, 2131 .setsockopt = netlink_setsockopt, 2132 .getsockopt = netlink_getsockopt, 2133 .sendmsg = netlink_sendmsg, 2134 .recvmsg = netlink_recvmsg, 2135 .mmap = sock_no_mmap, 2136 .sendpage = sock_no_sendpage, 2137 }; 2138 2139 static const struct net_proto_family netlink_family_ops = { 2140 .family = PF_NETLINK, 2141 .create = netlink_create, 2142 .owner = THIS_MODULE, /* for consistency 8) */ 2143 }; 2144 2145 static int __net_init netlink_net_init(struct net *net) 2146 { 2147 #ifdef CONFIG_PROC_FS 2148 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops)) 2149 return -ENOMEM; 2150 #endif 2151 return 0; 2152 } 2153 2154 static void __net_exit netlink_net_exit(struct net *net) 2155 { 2156 #ifdef CONFIG_PROC_FS 2157 remove_proc_entry("netlink", net->proc_net); 2158 #endif 2159 } 2160 2161 static void __init netlink_add_usersock_entry(void) 2162 { 2163 struct listeners *listeners; 2164 int groups = 32; 2165 2166 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); 2167 if (!listeners) 2168 panic("netlink_add_usersock_entry: Cannot allocate listeners\n"); 2169 2170 netlink_table_grab(); 2171 2172 nl_table[NETLINK_USERSOCK].groups = groups; 2173 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners); 2174 nl_table[NETLINK_USERSOCK].module = THIS_MODULE; 2175 nl_table[NETLINK_USERSOCK].registered = 1; 2176 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND; 2177 2178 netlink_table_ungrab(); 2179 } 2180 2181 static struct pernet_operations __net_initdata netlink_net_ops = { 2182 .init = netlink_net_init, 2183 .exit = netlink_net_exit, 2184 }; 2185 2186 static int __init netlink_proto_init(void) 2187 { 2188 int i; 2189 unsigned long limit; 2190 unsigned int order; 2191 int err = proto_register(&netlink_proto, 0); 2192 2193 if (err != 0) 2194 goto out; 2195 2196 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb)); 2197 2198 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL); 2199 if (!nl_table) 2200 goto panic; 2201 2202 if (totalram_pages >= (128 * 1024)) 2203 limit = totalram_pages >> (21 - PAGE_SHIFT); 2204 else 2205 limit = totalram_pages >> (23 - PAGE_SHIFT); 2206 2207 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT; 2208 limit = (1UL << order) / sizeof(struct hlist_head); 2209 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1; 2210 2211 for (i = 0; i < MAX_LINKS; i++) { 2212 struct nl_portid_hash *hash = &nl_table[i].hash; 2213 2214 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table)); 2215 if (!hash->table) { 2216 while (i-- > 0) 2217 nl_portid_hash_free(nl_table[i].hash.table, 2218 1 * sizeof(*hash->table)); 2219 kfree(nl_table); 2220 goto panic; 2221 } 2222 hash->max_shift = order; 2223 hash->shift = 0; 2224 hash->mask = 0; 2225 hash->rehash_time = jiffies; 2226 } 2227 2228 netlink_add_usersock_entry(); 2229 2230 sock_register(&netlink_family_ops); 2231 register_pernet_subsys(&netlink_net_ops); 2232 /* The netlink device handler may be needed early. */ 2233 rtnetlink_init(); 2234 out: 2235 return err; 2236 panic: 2237 panic("netlink_init: Cannot allocate nl_table\n"); 2238 } 2239 2240 core_initcall(netlink_proto_init); 2241