1 /* 2 * NETLINK Kernel-user communication protocol. 3 * 4 * Authors: Alan Cox <alan@redhat.com> 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/smp_lock.h> 49 #include <linux/notifier.h> 50 #include <linux/security.h> 51 #include <linux/jhash.h> 52 #include <linux/jiffies.h> 53 #include <linux/random.h> 54 #include <linux/bitops.h> 55 #include <linux/mm.h> 56 #include <linux/types.h> 57 #include <linux/audit.h> 58 #include <linux/selinux.h> 59 #include <linux/mutex.h> 60 61 #include <net/sock.h> 62 #include <net/scm.h> 63 #include <net/netlink.h> 64 65 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8) 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 (*data_ready)(struct sock *sk, int bytes); 83 struct module *module; 84 }; 85 86 #define NETLINK_KERNEL_SOCKET 0x1 87 #define NETLINK_RECV_PKTINFO 0x2 88 89 static inline struct netlink_sock *nlk_sk(struct sock *sk) 90 { 91 return (struct netlink_sock *)sk; 92 } 93 94 struct nl_pid_hash { 95 struct hlist_head *table; 96 unsigned long rehash_time; 97 98 unsigned int mask; 99 unsigned int shift; 100 101 unsigned int entries; 102 unsigned int max_shift; 103 104 u32 rnd; 105 }; 106 107 struct netlink_table { 108 struct nl_pid_hash hash; 109 struct hlist_head mc_list; 110 unsigned long *listeners; 111 unsigned int nl_nonroot; 112 unsigned int groups; 113 struct mutex *cb_mutex; 114 struct module *module; 115 int registered; 116 }; 117 118 static struct netlink_table *nl_table; 119 120 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait); 121 122 static int netlink_dump(struct sock *sk); 123 static void netlink_destroy_callback(struct netlink_callback *cb); 124 static void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb); 125 126 static DEFINE_RWLOCK(nl_table_lock); 127 static atomic_t nl_table_users = ATOMIC_INIT(0); 128 129 static ATOMIC_NOTIFIER_HEAD(netlink_chain); 130 131 static u32 netlink_group_mask(u32 group) 132 { 133 return group ? 1 << (group - 1) : 0; 134 } 135 136 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid) 137 { 138 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask]; 139 } 140 141 static void netlink_sock_destruct(struct sock *sk) 142 { 143 struct netlink_sock *nlk = nlk_sk(sk); 144 145 if (nlk->cb) { 146 if (nlk->cb->done) 147 nlk->cb->done(nlk->cb); 148 netlink_destroy_callback(nlk->cb); 149 } 150 151 skb_queue_purge(&sk->sk_receive_queue); 152 153 if (!sock_flag(sk, SOCK_DEAD)) { 154 printk("Freeing alive netlink socket %p\n", sk); 155 return; 156 } 157 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc)); 158 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc)); 159 BUG_TRAP(!nlk_sk(sk)->groups); 160 } 161 162 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP. 163 * Look, when several writers sleep and reader wakes them up, all but one 164 * immediately hit write lock and grab all the cpus. Exclusive sleep solves 165 * this, _but_ remember, it adds useless work on UP machines. 166 */ 167 168 static void netlink_table_grab(void) 169 { 170 write_lock_irq(&nl_table_lock); 171 172 if (atomic_read(&nl_table_users)) { 173 DECLARE_WAITQUEUE(wait, current); 174 175 add_wait_queue_exclusive(&nl_table_wait, &wait); 176 for(;;) { 177 set_current_state(TASK_UNINTERRUPTIBLE); 178 if (atomic_read(&nl_table_users) == 0) 179 break; 180 write_unlock_irq(&nl_table_lock); 181 schedule(); 182 write_lock_irq(&nl_table_lock); 183 } 184 185 __set_current_state(TASK_RUNNING); 186 remove_wait_queue(&nl_table_wait, &wait); 187 } 188 } 189 190 static __inline__ void netlink_table_ungrab(void) 191 { 192 write_unlock_irq(&nl_table_lock); 193 wake_up(&nl_table_wait); 194 } 195 196 static __inline__ void 197 netlink_lock_table(void) 198 { 199 /* read_lock() synchronizes us to netlink_table_grab */ 200 201 read_lock(&nl_table_lock); 202 atomic_inc(&nl_table_users); 203 read_unlock(&nl_table_lock); 204 } 205 206 static __inline__ void 207 netlink_unlock_table(void) 208 { 209 if (atomic_dec_and_test(&nl_table_users)) 210 wake_up(&nl_table_wait); 211 } 212 213 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid) 214 { 215 struct nl_pid_hash *hash = &nl_table[protocol].hash; 216 struct hlist_head *head; 217 struct sock *sk; 218 struct hlist_node *node; 219 220 read_lock(&nl_table_lock); 221 head = nl_pid_hashfn(hash, pid); 222 sk_for_each(sk, node, head) { 223 if (nlk_sk(sk)->pid == pid) { 224 sock_hold(sk); 225 goto found; 226 } 227 } 228 sk = NULL; 229 found: 230 read_unlock(&nl_table_lock); 231 return sk; 232 } 233 234 static inline struct hlist_head *nl_pid_hash_alloc(size_t size) 235 { 236 if (size <= PAGE_SIZE) 237 return kmalloc(size, GFP_ATOMIC); 238 else 239 return (struct hlist_head *) 240 __get_free_pages(GFP_ATOMIC, get_order(size)); 241 } 242 243 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size) 244 { 245 if (size <= PAGE_SIZE) 246 kfree(table); 247 else 248 free_pages((unsigned long)table, get_order(size)); 249 } 250 251 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow) 252 { 253 unsigned int omask, mask, shift; 254 size_t osize, size; 255 struct hlist_head *otable, *table; 256 int i; 257 258 omask = mask = hash->mask; 259 osize = size = (mask + 1) * sizeof(*table); 260 shift = hash->shift; 261 262 if (grow) { 263 if (++shift > hash->max_shift) 264 return 0; 265 mask = mask * 2 + 1; 266 size *= 2; 267 } 268 269 table = nl_pid_hash_alloc(size); 270 if (!table) 271 return 0; 272 273 memset(table, 0, size); 274 otable = hash->table; 275 hash->table = table; 276 hash->mask = mask; 277 hash->shift = shift; 278 get_random_bytes(&hash->rnd, sizeof(hash->rnd)); 279 280 for (i = 0; i <= omask; i++) { 281 struct sock *sk; 282 struct hlist_node *node, *tmp; 283 284 sk_for_each_safe(sk, node, tmp, &otable[i]) 285 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid)); 286 } 287 288 nl_pid_hash_free(otable, osize); 289 hash->rehash_time = jiffies + 10 * 60 * HZ; 290 return 1; 291 } 292 293 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len) 294 { 295 int avg = hash->entries >> hash->shift; 296 297 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1)) 298 return 1; 299 300 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) { 301 nl_pid_hash_rehash(hash, 0); 302 return 1; 303 } 304 305 return 0; 306 } 307 308 static const struct proto_ops netlink_ops; 309 310 static void 311 netlink_update_listeners(struct sock *sk) 312 { 313 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 314 struct hlist_node *node; 315 unsigned long mask; 316 unsigned int i; 317 318 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) { 319 mask = 0; 320 sk_for_each_bound(sk, node, &tbl->mc_list) 321 mask |= nlk_sk(sk)->groups[i]; 322 tbl->listeners[i] = mask; 323 } 324 /* this function is only called with the netlink table "grabbed", which 325 * makes sure updates are visible before bind or setsockopt return. */ 326 } 327 328 static int netlink_insert(struct sock *sk, u32 pid) 329 { 330 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 331 struct hlist_head *head; 332 int err = -EADDRINUSE; 333 struct sock *osk; 334 struct hlist_node *node; 335 int len; 336 337 netlink_table_grab(); 338 head = nl_pid_hashfn(hash, pid); 339 len = 0; 340 sk_for_each(osk, node, head) { 341 if (nlk_sk(osk)->pid == pid) 342 break; 343 len++; 344 } 345 if (node) 346 goto err; 347 348 err = -EBUSY; 349 if (nlk_sk(sk)->pid) 350 goto err; 351 352 err = -ENOMEM; 353 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX)) 354 goto err; 355 356 if (len && nl_pid_hash_dilute(hash, len)) 357 head = nl_pid_hashfn(hash, pid); 358 hash->entries++; 359 nlk_sk(sk)->pid = pid; 360 sk_add_node(sk, head); 361 err = 0; 362 363 err: 364 netlink_table_ungrab(); 365 return err; 366 } 367 368 static void netlink_remove(struct sock *sk) 369 { 370 netlink_table_grab(); 371 if (sk_del_node_init(sk)) 372 nl_table[sk->sk_protocol].hash.entries--; 373 if (nlk_sk(sk)->subscriptions) 374 __sk_del_bind_node(sk); 375 netlink_table_ungrab(); 376 } 377 378 static struct proto netlink_proto = { 379 .name = "NETLINK", 380 .owner = THIS_MODULE, 381 .obj_size = sizeof(struct netlink_sock), 382 }; 383 384 static int __netlink_create(struct socket *sock, struct mutex *cb_mutex, 385 int protocol) 386 { 387 struct sock *sk; 388 struct netlink_sock *nlk; 389 390 sock->ops = &netlink_ops; 391 392 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1); 393 if (!sk) 394 return -ENOMEM; 395 396 sock_init_data(sock, sk); 397 398 nlk = nlk_sk(sk); 399 if (cb_mutex) 400 nlk->cb_mutex = cb_mutex; 401 else { 402 nlk->cb_mutex = &nlk->cb_def_mutex; 403 mutex_init(nlk->cb_mutex); 404 } 405 init_waitqueue_head(&nlk->wait); 406 407 sk->sk_destruct = netlink_sock_destruct; 408 sk->sk_protocol = protocol; 409 return 0; 410 } 411 412 static int netlink_create(struct socket *sock, int protocol) 413 { 414 struct module *module = NULL; 415 struct mutex *cb_mutex; 416 struct netlink_sock *nlk; 417 int err = 0; 418 419 sock->state = SS_UNCONNECTED; 420 421 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 422 return -ESOCKTNOSUPPORT; 423 424 if (protocol<0 || protocol >= MAX_LINKS) 425 return -EPROTONOSUPPORT; 426 427 netlink_lock_table(); 428 #ifdef CONFIG_KMOD 429 if (!nl_table[protocol].registered) { 430 netlink_unlock_table(); 431 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol); 432 netlink_lock_table(); 433 } 434 #endif 435 if (nl_table[protocol].registered && 436 try_module_get(nl_table[protocol].module)) 437 module = nl_table[protocol].module; 438 cb_mutex = nl_table[protocol].cb_mutex; 439 netlink_unlock_table(); 440 441 if ((err = __netlink_create(sock, cb_mutex, protocol)) < 0) 442 goto out_module; 443 444 nlk = nlk_sk(sock->sk); 445 nlk->module = module; 446 out: 447 return err; 448 449 out_module: 450 module_put(module); 451 goto out; 452 } 453 454 static int netlink_release(struct socket *sock) 455 { 456 struct sock *sk = sock->sk; 457 struct netlink_sock *nlk; 458 459 if (!sk) 460 return 0; 461 462 netlink_remove(sk); 463 sock_orphan(sk); 464 nlk = nlk_sk(sk); 465 466 /* 467 * OK. Socket is unlinked, any packets that arrive now 468 * will be purged. 469 */ 470 471 sock->sk = NULL; 472 wake_up_interruptible_all(&nlk->wait); 473 474 skb_queue_purge(&sk->sk_write_queue); 475 476 if (nlk->pid && !nlk->subscriptions) { 477 struct netlink_notify n = { 478 .protocol = sk->sk_protocol, 479 .pid = nlk->pid, 480 }; 481 atomic_notifier_call_chain(&netlink_chain, 482 NETLINK_URELEASE, &n); 483 } 484 485 module_put(nlk->module); 486 487 netlink_table_grab(); 488 if (nlk->flags & NETLINK_KERNEL_SOCKET) { 489 kfree(nl_table[sk->sk_protocol].listeners); 490 nl_table[sk->sk_protocol].module = NULL; 491 nl_table[sk->sk_protocol].registered = 0; 492 } else if (nlk->subscriptions) 493 netlink_update_listeners(sk); 494 netlink_table_ungrab(); 495 496 kfree(nlk->groups); 497 nlk->groups = NULL; 498 499 sock_put(sk); 500 return 0; 501 } 502 503 static int netlink_autobind(struct socket *sock) 504 { 505 struct sock *sk = sock->sk; 506 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 507 struct hlist_head *head; 508 struct sock *osk; 509 struct hlist_node *node; 510 s32 pid = current->tgid; 511 int err; 512 static s32 rover = -4097; 513 514 retry: 515 cond_resched(); 516 netlink_table_grab(); 517 head = nl_pid_hashfn(hash, pid); 518 sk_for_each(osk, node, head) { 519 if (nlk_sk(osk)->pid == pid) { 520 /* Bind collision, search negative pid values. */ 521 pid = rover--; 522 if (rover > -4097) 523 rover = -4097; 524 netlink_table_ungrab(); 525 goto retry; 526 } 527 } 528 netlink_table_ungrab(); 529 530 err = netlink_insert(sk, pid); 531 if (err == -EADDRINUSE) 532 goto retry; 533 534 /* If 2 threads race to autobind, that is fine. */ 535 if (err == -EBUSY) 536 err = 0; 537 538 return err; 539 } 540 541 static inline int netlink_capable(struct socket *sock, unsigned int flag) 542 { 543 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) || 544 capable(CAP_NET_ADMIN); 545 } 546 547 static void 548 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions) 549 { 550 struct netlink_sock *nlk = nlk_sk(sk); 551 552 if (nlk->subscriptions && !subscriptions) 553 __sk_del_bind_node(sk); 554 else if (!nlk->subscriptions && subscriptions) 555 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 556 nlk->subscriptions = subscriptions; 557 } 558 559 static int netlink_alloc_groups(struct sock *sk) 560 { 561 struct netlink_sock *nlk = nlk_sk(sk); 562 unsigned int groups; 563 int err = 0; 564 565 netlink_lock_table(); 566 groups = nl_table[sk->sk_protocol].groups; 567 if (!nl_table[sk->sk_protocol].registered) 568 err = -ENOENT; 569 netlink_unlock_table(); 570 571 if (err) 572 return err; 573 574 nlk->groups = kzalloc(NLGRPSZ(groups), GFP_KERNEL); 575 if (nlk->groups == NULL) 576 return -ENOMEM; 577 nlk->ngroups = groups; 578 return 0; 579 } 580 581 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 582 { 583 struct sock *sk = sock->sk; 584 struct netlink_sock *nlk = nlk_sk(sk); 585 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 586 int err; 587 588 if (nladdr->nl_family != AF_NETLINK) 589 return -EINVAL; 590 591 /* Only superuser is allowed to listen multicasts */ 592 if (nladdr->nl_groups) { 593 if (!netlink_capable(sock, NL_NONROOT_RECV)) 594 return -EPERM; 595 if (nlk->groups == NULL) { 596 err = netlink_alloc_groups(sk); 597 if (err) 598 return err; 599 } 600 } 601 602 if (nlk->pid) { 603 if (nladdr->nl_pid != nlk->pid) 604 return -EINVAL; 605 } else { 606 err = nladdr->nl_pid ? 607 netlink_insert(sk, nladdr->nl_pid) : 608 netlink_autobind(sock); 609 if (err) 610 return err; 611 } 612 613 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0])) 614 return 0; 615 616 netlink_table_grab(); 617 netlink_update_subscriptions(sk, nlk->subscriptions + 618 hweight32(nladdr->nl_groups) - 619 hweight32(nlk->groups[0])); 620 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups; 621 netlink_update_listeners(sk); 622 netlink_table_ungrab(); 623 624 return 0; 625 } 626 627 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 628 int alen, int flags) 629 { 630 int err = 0; 631 struct sock *sk = sock->sk; 632 struct netlink_sock *nlk = nlk_sk(sk); 633 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr; 634 635 if (addr->sa_family == AF_UNSPEC) { 636 sk->sk_state = NETLINK_UNCONNECTED; 637 nlk->dst_pid = 0; 638 nlk->dst_group = 0; 639 return 0; 640 } 641 if (addr->sa_family != AF_NETLINK) 642 return -EINVAL; 643 644 /* Only superuser is allowed to send multicasts */ 645 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND)) 646 return -EPERM; 647 648 if (!nlk->pid) 649 err = netlink_autobind(sock); 650 651 if (err == 0) { 652 sk->sk_state = NETLINK_CONNECTED; 653 nlk->dst_pid = nladdr->nl_pid; 654 nlk->dst_group = ffs(nladdr->nl_groups); 655 } 656 657 return err; 658 } 659 660 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer) 661 { 662 struct sock *sk = sock->sk; 663 struct netlink_sock *nlk = nlk_sk(sk); 664 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr; 665 666 nladdr->nl_family = AF_NETLINK; 667 nladdr->nl_pad = 0; 668 *addr_len = sizeof(*nladdr); 669 670 if (peer) { 671 nladdr->nl_pid = nlk->dst_pid; 672 nladdr->nl_groups = netlink_group_mask(nlk->dst_group); 673 } else { 674 nladdr->nl_pid = nlk->pid; 675 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0; 676 } 677 return 0; 678 } 679 680 static void netlink_overrun(struct sock *sk) 681 { 682 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) { 683 sk->sk_err = ENOBUFS; 684 sk->sk_error_report(sk); 685 } 686 } 687 688 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid) 689 { 690 int protocol = ssk->sk_protocol; 691 struct sock *sock; 692 struct netlink_sock *nlk; 693 694 sock = netlink_lookup(protocol, pid); 695 if (!sock) 696 return ERR_PTR(-ECONNREFUSED); 697 698 /* Don't bother queuing skb if kernel socket has no input function */ 699 nlk = nlk_sk(sock); 700 if ((nlk->pid == 0 && !nlk->data_ready) || 701 (sock->sk_state == NETLINK_CONNECTED && 702 nlk->dst_pid != nlk_sk(ssk)->pid)) { 703 sock_put(sock); 704 return ERR_PTR(-ECONNREFUSED); 705 } 706 return sock; 707 } 708 709 struct sock *netlink_getsockbyfilp(struct file *filp) 710 { 711 struct inode *inode = filp->f_path.dentry->d_inode; 712 struct sock *sock; 713 714 if (!S_ISSOCK(inode->i_mode)) 715 return ERR_PTR(-ENOTSOCK); 716 717 sock = SOCKET_I(inode)->sk; 718 if (sock->sk_family != AF_NETLINK) 719 return ERR_PTR(-EINVAL); 720 721 sock_hold(sock); 722 return sock; 723 } 724 725 /* 726 * Attach a skb to a netlink socket. 727 * The caller must hold a reference to the destination socket. On error, the 728 * reference is dropped. The skb is not send to the destination, just all 729 * all error checks are performed and memory in the queue is reserved. 730 * Return values: 731 * < 0: error. skb freed, reference to sock dropped. 732 * 0: continue 733 * 1: repeat lookup - reference dropped while waiting for socket memory. 734 */ 735 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, 736 long timeo, struct sock *ssk) 737 { 738 struct netlink_sock *nlk; 739 740 nlk = nlk_sk(sk); 741 742 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 743 test_bit(0, &nlk->state)) { 744 DECLARE_WAITQUEUE(wait, current); 745 if (!timeo) { 746 if (!ssk || nlk_sk(ssk)->pid == 0) 747 netlink_overrun(sk); 748 sock_put(sk); 749 kfree_skb(skb); 750 return -EAGAIN; 751 } 752 753 __set_current_state(TASK_INTERRUPTIBLE); 754 add_wait_queue(&nlk->wait, &wait); 755 756 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 757 test_bit(0, &nlk->state)) && 758 !sock_flag(sk, SOCK_DEAD)) 759 timeo = schedule_timeout(timeo); 760 761 __set_current_state(TASK_RUNNING); 762 remove_wait_queue(&nlk->wait, &wait); 763 sock_put(sk); 764 765 if (signal_pending(current)) { 766 kfree_skb(skb); 767 return sock_intr_errno(timeo); 768 } 769 return 1; 770 } 771 skb_set_owner_r(skb, sk); 772 return 0; 773 } 774 775 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol) 776 { 777 int len = skb->len; 778 779 skb_queue_tail(&sk->sk_receive_queue, skb); 780 sk->sk_data_ready(sk, len); 781 sock_put(sk); 782 return len; 783 } 784 785 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 786 { 787 kfree_skb(skb); 788 sock_put(sk); 789 } 790 791 static inline struct sk_buff *netlink_trim(struct sk_buff *skb, 792 gfp_t allocation) 793 { 794 int delta; 795 796 skb_orphan(skb); 797 798 delta = skb->end - skb->tail; 799 if (delta * 2 < skb->truesize) 800 return skb; 801 802 if (skb_shared(skb)) { 803 struct sk_buff *nskb = skb_clone(skb, allocation); 804 if (!nskb) 805 return skb; 806 kfree_skb(skb); 807 skb = nskb; 808 } 809 810 if (!pskb_expand_head(skb, 0, -delta, allocation)) 811 skb->truesize -= delta; 812 813 return skb; 814 } 815 816 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock) 817 { 818 struct sock *sk; 819 int err; 820 long timeo; 821 822 skb = netlink_trim(skb, gfp_any()); 823 824 timeo = sock_sndtimeo(ssk, nonblock); 825 retry: 826 sk = netlink_getsockbypid(ssk, pid); 827 if (IS_ERR(sk)) { 828 kfree_skb(skb); 829 return PTR_ERR(sk); 830 } 831 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk); 832 if (err == 1) 833 goto retry; 834 if (err) 835 return err; 836 837 return netlink_sendskb(sk, skb, ssk->sk_protocol); 838 } 839 840 int netlink_has_listeners(struct sock *sk, unsigned int group) 841 { 842 int res = 0; 843 844 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET)); 845 if (group - 1 < nl_table[sk->sk_protocol].groups) 846 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners); 847 return res; 848 } 849 EXPORT_SYMBOL_GPL(netlink_has_listeners); 850 851 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 852 { 853 struct netlink_sock *nlk = nlk_sk(sk); 854 855 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 856 !test_bit(0, &nlk->state)) { 857 skb_set_owner_r(skb, sk); 858 skb_queue_tail(&sk->sk_receive_queue, skb); 859 sk->sk_data_ready(sk, skb->len); 860 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf; 861 } 862 return -1; 863 } 864 865 struct netlink_broadcast_data { 866 struct sock *exclude_sk; 867 u32 pid; 868 u32 group; 869 int failure; 870 int congested; 871 int delivered; 872 gfp_t allocation; 873 struct sk_buff *skb, *skb2; 874 }; 875 876 static inline int do_one_broadcast(struct sock *sk, 877 struct netlink_broadcast_data *p) 878 { 879 struct netlink_sock *nlk = nlk_sk(sk); 880 int val; 881 882 if (p->exclude_sk == sk) 883 goto out; 884 885 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups || 886 !test_bit(p->group - 1, nlk->groups)) 887 goto out; 888 889 if (p->failure) { 890 netlink_overrun(sk); 891 goto out; 892 } 893 894 sock_hold(sk); 895 if (p->skb2 == NULL) { 896 if (skb_shared(p->skb)) { 897 p->skb2 = skb_clone(p->skb, p->allocation); 898 } else { 899 p->skb2 = skb_get(p->skb); 900 /* 901 * skb ownership may have been set when 902 * delivered to a previous socket. 903 */ 904 skb_orphan(p->skb2); 905 } 906 } 907 if (p->skb2 == NULL) { 908 netlink_overrun(sk); 909 /* Clone failed. Notify ALL listeners. */ 910 p->failure = 1; 911 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) { 912 netlink_overrun(sk); 913 } else { 914 p->congested |= val; 915 p->delivered = 1; 916 p->skb2 = NULL; 917 } 918 sock_put(sk); 919 920 out: 921 return 0; 922 } 923 924 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid, 925 u32 group, gfp_t allocation) 926 { 927 struct netlink_broadcast_data info; 928 struct hlist_node *node; 929 struct sock *sk; 930 931 skb = netlink_trim(skb, allocation); 932 933 info.exclude_sk = ssk; 934 info.pid = pid; 935 info.group = group; 936 info.failure = 0; 937 info.congested = 0; 938 info.delivered = 0; 939 info.allocation = allocation; 940 info.skb = skb; 941 info.skb2 = NULL; 942 943 /* While we sleep in clone, do not allow to change socket list */ 944 945 netlink_lock_table(); 946 947 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 948 do_one_broadcast(sk, &info); 949 950 kfree_skb(skb); 951 952 netlink_unlock_table(); 953 954 if (info.skb2) 955 kfree_skb(info.skb2); 956 957 if (info.delivered) { 958 if (info.congested && (allocation & __GFP_WAIT)) 959 yield(); 960 return 0; 961 } 962 if (info.failure) 963 return -ENOBUFS; 964 return -ESRCH; 965 } 966 967 struct netlink_set_err_data { 968 struct sock *exclude_sk; 969 u32 pid; 970 u32 group; 971 int code; 972 }; 973 974 static inline int do_one_set_err(struct sock *sk, 975 struct netlink_set_err_data *p) 976 { 977 struct netlink_sock *nlk = nlk_sk(sk); 978 979 if (sk == p->exclude_sk) 980 goto out; 981 982 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups || 983 !test_bit(p->group - 1, nlk->groups)) 984 goto out; 985 986 sk->sk_err = p->code; 987 sk->sk_error_report(sk); 988 out: 989 return 0; 990 } 991 992 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code) 993 { 994 struct netlink_set_err_data info; 995 struct hlist_node *node; 996 struct sock *sk; 997 998 info.exclude_sk = ssk; 999 info.pid = pid; 1000 info.group = group; 1001 info.code = code; 1002 1003 read_lock(&nl_table_lock); 1004 1005 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 1006 do_one_set_err(sk, &info); 1007 1008 read_unlock(&nl_table_lock); 1009 } 1010 1011 static int netlink_setsockopt(struct socket *sock, int level, int optname, 1012 char __user *optval, int optlen) 1013 { 1014 struct sock *sk = sock->sk; 1015 struct netlink_sock *nlk = nlk_sk(sk); 1016 int val = 0, err; 1017 1018 if (level != SOL_NETLINK) 1019 return -ENOPROTOOPT; 1020 1021 if (optlen >= sizeof(int) && 1022 get_user(val, (int __user *)optval)) 1023 return -EFAULT; 1024 1025 switch (optname) { 1026 case NETLINK_PKTINFO: 1027 if (val) 1028 nlk->flags |= NETLINK_RECV_PKTINFO; 1029 else 1030 nlk->flags &= ~NETLINK_RECV_PKTINFO; 1031 err = 0; 1032 break; 1033 case NETLINK_ADD_MEMBERSHIP: 1034 case NETLINK_DROP_MEMBERSHIP: { 1035 unsigned int subscriptions; 1036 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0; 1037 1038 if (!netlink_capable(sock, NL_NONROOT_RECV)) 1039 return -EPERM; 1040 if (nlk->groups == NULL) { 1041 err = netlink_alloc_groups(sk); 1042 if (err) 1043 return err; 1044 } 1045 if (!val || val - 1 >= nlk->ngroups) 1046 return -EINVAL; 1047 netlink_table_grab(); 1048 old = test_bit(val - 1, nlk->groups); 1049 subscriptions = nlk->subscriptions - old + new; 1050 if (new) 1051 __set_bit(val - 1, nlk->groups); 1052 else 1053 __clear_bit(val - 1, nlk->groups); 1054 netlink_update_subscriptions(sk, subscriptions); 1055 netlink_update_listeners(sk); 1056 netlink_table_ungrab(); 1057 err = 0; 1058 break; 1059 } 1060 default: 1061 err = -ENOPROTOOPT; 1062 } 1063 return err; 1064 } 1065 1066 static int netlink_getsockopt(struct socket *sock, int level, int optname, 1067 char __user *optval, int __user *optlen) 1068 { 1069 struct sock *sk = sock->sk; 1070 struct netlink_sock *nlk = nlk_sk(sk); 1071 int len, val, err; 1072 1073 if (level != SOL_NETLINK) 1074 return -ENOPROTOOPT; 1075 1076 if (get_user(len, optlen)) 1077 return -EFAULT; 1078 if (len < 0) 1079 return -EINVAL; 1080 1081 switch (optname) { 1082 case NETLINK_PKTINFO: 1083 if (len < sizeof(int)) 1084 return -EINVAL; 1085 len = sizeof(int); 1086 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0; 1087 if (put_user(len, optlen) || 1088 put_user(val, optval)) 1089 return -EFAULT; 1090 err = 0; 1091 break; 1092 default: 1093 err = -ENOPROTOOPT; 1094 } 1095 return err; 1096 } 1097 1098 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb) 1099 { 1100 struct nl_pktinfo info; 1101 1102 info.group = NETLINK_CB(skb).dst_group; 1103 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info); 1104 } 1105 1106 static inline void netlink_rcv_wake(struct sock *sk) 1107 { 1108 struct netlink_sock *nlk = nlk_sk(sk); 1109 1110 if (skb_queue_empty(&sk->sk_receive_queue)) 1111 clear_bit(0, &nlk->state); 1112 if (!test_bit(0, &nlk->state)) 1113 wake_up_interruptible(&nlk->wait); 1114 } 1115 1116 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock, 1117 struct msghdr *msg, size_t len) 1118 { 1119 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1120 struct sock *sk = sock->sk; 1121 struct netlink_sock *nlk = nlk_sk(sk); 1122 struct sockaddr_nl *addr=msg->msg_name; 1123 u32 dst_pid; 1124 u32 dst_group; 1125 struct sk_buff *skb; 1126 int err; 1127 struct scm_cookie scm; 1128 1129 if (msg->msg_flags&MSG_OOB) 1130 return -EOPNOTSUPP; 1131 1132 if (NULL == siocb->scm) 1133 siocb->scm = &scm; 1134 err = scm_send(sock, msg, siocb->scm); 1135 if (err < 0) 1136 return err; 1137 1138 if (msg->msg_namelen) { 1139 if (addr->nl_family != AF_NETLINK) 1140 return -EINVAL; 1141 dst_pid = addr->nl_pid; 1142 dst_group = ffs(addr->nl_groups); 1143 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND)) 1144 return -EPERM; 1145 } else { 1146 dst_pid = nlk->dst_pid; 1147 dst_group = nlk->dst_group; 1148 } 1149 1150 if (!nlk->pid) { 1151 err = netlink_autobind(sock); 1152 if (err) 1153 goto out; 1154 } 1155 1156 err = -EMSGSIZE; 1157 if (len > sk->sk_sndbuf - 32) 1158 goto out; 1159 err = -ENOBUFS; 1160 skb = alloc_skb(len, GFP_KERNEL); 1161 if (skb==NULL) 1162 goto out; 1163 1164 NETLINK_CB(skb).pid = nlk->pid; 1165 NETLINK_CB(skb).dst_group = dst_group; 1166 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context); 1167 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid)); 1168 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); 1169 1170 /* What can I do? Netlink is asynchronous, so that 1171 we will have to save current capabilities to 1172 check them, when this message will be delivered 1173 to corresponding kernel module. --ANK (980802) 1174 */ 1175 1176 err = -EFAULT; 1177 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) { 1178 kfree_skb(skb); 1179 goto out; 1180 } 1181 1182 err = security_netlink_send(sk, skb); 1183 if (err) { 1184 kfree_skb(skb); 1185 goto out; 1186 } 1187 1188 if (dst_group) { 1189 atomic_inc(&skb->users); 1190 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL); 1191 } 1192 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT); 1193 1194 out: 1195 return err; 1196 } 1197 1198 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock, 1199 struct msghdr *msg, size_t len, 1200 int flags) 1201 { 1202 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1203 struct scm_cookie scm; 1204 struct sock *sk = sock->sk; 1205 struct netlink_sock *nlk = nlk_sk(sk); 1206 int noblock = flags&MSG_DONTWAIT; 1207 size_t copied; 1208 struct sk_buff *skb; 1209 int err; 1210 1211 if (flags&MSG_OOB) 1212 return -EOPNOTSUPP; 1213 1214 copied = 0; 1215 1216 skb = skb_recv_datagram(sk,flags,noblock,&err); 1217 if (skb==NULL) 1218 goto out; 1219 1220 msg->msg_namelen = 0; 1221 1222 copied = skb->len; 1223 if (len < copied) { 1224 msg->msg_flags |= MSG_TRUNC; 1225 copied = len; 1226 } 1227 1228 skb_reset_transport_header(skb); 1229 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 1230 1231 if (msg->msg_name) { 1232 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name; 1233 addr->nl_family = AF_NETLINK; 1234 addr->nl_pad = 0; 1235 addr->nl_pid = NETLINK_CB(skb).pid; 1236 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group); 1237 msg->msg_namelen = sizeof(*addr); 1238 } 1239 1240 if (nlk->flags & NETLINK_RECV_PKTINFO) 1241 netlink_cmsg_recv_pktinfo(msg, skb); 1242 1243 if (NULL == siocb->scm) { 1244 memset(&scm, 0, sizeof(scm)); 1245 siocb->scm = &scm; 1246 } 1247 siocb->scm->creds = *NETLINK_CREDS(skb); 1248 if (flags & MSG_TRUNC) 1249 copied = skb->len; 1250 skb_free_datagram(sk, skb); 1251 1252 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) 1253 netlink_dump(sk); 1254 1255 scm_recv(sock, msg, siocb->scm, flags); 1256 out: 1257 netlink_rcv_wake(sk); 1258 return err ? : copied; 1259 } 1260 1261 static void netlink_data_ready(struct sock *sk, int len) 1262 { 1263 struct netlink_sock *nlk = nlk_sk(sk); 1264 1265 if (nlk->data_ready) 1266 nlk->data_ready(sk, len); 1267 netlink_rcv_wake(sk); 1268 } 1269 1270 /* 1271 * We export these functions to other modules. They provide a 1272 * complete set of kernel non-blocking support for message 1273 * queueing. 1274 */ 1275 1276 struct sock * 1277 netlink_kernel_create(int unit, unsigned int groups, 1278 void (*input)(struct sock *sk, int len), 1279 struct mutex *cb_mutex, struct module *module) 1280 { 1281 struct socket *sock; 1282 struct sock *sk; 1283 struct netlink_sock *nlk; 1284 unsigned long *listeners = NULL; 1285 1286 BUG_ON(!nl_table); 1287 1288 if (unit<0 || unit>=MAX_LINKS) 1289 return NULL; 1290 1291 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 1292 return NULL; 1293 1294 if (__netlink_create(sock, cb_mutex, unit) < 0) 1295 goto out_sock_release; 1296 1297 if (groups < 32) 1298 groups = 32; 1299 1300 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL); 1301 if (!listeners) 1302 goto out_sock_release; 1303 1304 sk = sock->sk; 1305 sk->sk_data_ready = netlink_data_ready; 1306 if (input) 1307 nlk_sk(sk)->data_ready = input; 1308 1309 if (netlink_insert(sk, 0)) 1310 goto out_sock_release; 1311 1312 nlk = nlk_sk(sk); 1313 nlk->flags |= NETLINK_KERNEL_SOCKET; 1314 1315 netlink_table_grab(); 1316 nl_table[unit].groups = groups; 1317 nl_table[unit].listeners = listeners; 1318 nl_table[unit].cb_mutex = cb_mutex; 1319 nl_table[unit].module = module; 1320 nl_table[unit].registered = 1; 1321 netlink_table_ungrab(); 1322 1323 return sk; 1324 1325 out_sock_release: 1326 kfree(listeners); 1327 sock_release(sock); 1328 return NULL; 1329 } 1330 1331 void netlink_set_nonroot(int protocol, unsigned int flags) 1332 { 1333 if ((unsigned int)protocol < MAX_LINKS) 1334 nl_table[protocol].nl_nonroot = flags; 1335 } 1336 1337 static void netlink_destroy_callback(struct netlink_callback *cb) 1338 { 1339 if (cb->skb) 1340 kfree_skb(cb->skb); 1341 kfree(cb); 1342 } 1343 1344 /* 1345 * It looks a bit ugly. 1346 * It would be better to create kernel thread. 1347 */ 1348 1349 static int netlink_dump(struct sock *sk) 1350 { 1351 struct netlink_sock *nlk = nlk_sk(sk); 1352 struct netlink_callback *cb; 1353 struct sk_buff *skb; 1354 struct nlmsghdr *nlh; 1355 int len, err = -ENOBUFS; 1356 1357 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL); 1358 if (!skb) 1359 goto errout; 1360 1361 mutex_lock(nlk->cb_mutex); 1362 1363 cb = nlk->cb; 1364 if (cb == NULL) { 1365 err = -EINVAL; 1366 goto errout_skb; 1367 } 1368 1369 len = cb->dump(skb, cb); 1370 1371 if (len > 0) { 1372 mutex_unlock(nlk->cb_mutex); 1373 skb_queue_tail(&sk->sk_receive_queue, skb); 1374 sk->sk_data_ready(sk, len); 1375 return 0; 1376 } 1377 1378 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI); 1379 if (!nlh) 1380 goto errout_skb; 1381 1382 memcpy(nlmsg_data(nlh), &len, sizeof(len)); 1383 1384 skb_queue_tail(&sk->sk_receive_queue, skb); 1385 sk->sk_data_ready(sk, skb->len); 1386 1387 if (cb->done) 1388 cb->done(cb); 1389 nlk->cb = NULL; 1390 mutex_unlock(nlk->cb_mutex); 1391 1392 netlink_destroy_callback(cb); 1393 return 0; 1394 1395 errout_skb: 1396 mutex_unlock(nlk->cb_mutex); 1397 kfree_skb(skb); 1398 errout: 1399 return err; 1400 } 1401 1402 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 1403 struct nlmsghdr *nlh, 1404 int (*dump)(struct sk_buff *skb, struct netlink_callback*), 1405 int (*done)(struct netlink_callback*)) 1406 { 1407 struct netlink_callback *cb; 1408 struct sock *sk; 1409 struct netlink_sock *nlk; 1410 1411 cb = kzalloc(sizeof(*cb), GFP_KERNEL); 1412 if (cb == NULL) 1413 return -ENOBUFS; 1414 1415 cb->dump = dump; 1416 cb->done = done; 1417 cb->nlh = nlh; 1418 atomic_inc(&skb->users); 1419 cb->skb = skb; 1420 1421 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid); 1422 if (sk == NULL) { 1423 netlink_destroy_callback(cb); 1424 return -ECONNREFUSED; 1425 } 1426 nlk = nlk_sk(sk); 1427 /* A dump is in progress... */ 1428 mutex_lock(nlk->cb_mutex); 1429 if (nlk->cb) { 1430 mutex_unlock(nlk->cb_mutex); 1431 netlink_destroy_callback(cb); 1432 sock_put(sk); 1433 return -EBUSY; 1434 } 1435 nlk->cb = cb; 1436 mutex_unlock(nlk->cb_mutex); 1437 1438 netlink_dump(sk); 1439 sock_put(sk); 1440 1441 /* We successfully started a dump, by returning -EINTR we 1442 * signal the queue mangement to interrupt processing of 1443 * any netlink messages so userspace gets a chance to read 1444 * the results. */ 1445 return -EINTR; 1446 } 1447 1448 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err) 1449 { 1450 struct sk_buff *skb; 1451 struct nlmsghdr *rep; 1452 struct nlmsgerr *errmsg; 1453 size_t payload = sizeof(*errmsg); 1454 1455 /* error messages get the original request appened */ 1456 if (err) 1457 payload += nlmsg_len(nlh); 1458 1459 skb = nlmsg_new(payload, GFP_KERNEL); 1460 if (!skb) { 1461 struct sock *sk; 1462 1463 sk = netlink_lookup(in_skb->sk->sk_protocol, 1464 NETLINK_CB(in_skb).pid); 1465 if (sk) { 1466 sk->sk_err = ENOBUFS; 1467 sk->sk_error_report(sk); 1468 sock_put(sk); 1469 } 1470 return; 1471 } 1472 1473 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, 1474 NLMSG_ERROR, sizeof(struct nlmsgerr), 0); 1475 errmsg = nlmsg_data(rep); 1476 errmsg->error = err; 1477 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh)); 1478 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT); 1479 } 1480 1481 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *, 1482 struct nlmsghdr *)) 1483 { 1484 struct nlmsghdr *nlh; 1485 int err; 1486 1487 while (skb->len >= nlmsg_total_size(0)) { 1488 nlh = nlmsg_hdr(skb); 1489 err = 0; 1490 1491 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len) 1492 return 0; 1493 1494 /* Only requests are handled by the kernel */ 1495 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) 1496 goto skip; 1497 1498 /* Skip control messages */ 1499 if (nlh->nlmsg_type < NLMSG_MIN_TYPE) 1500 goto skip; 1501 1502 err = cb(skb, nlh); 1503 if (err == -EINTR) { 1504 /* Not an error, but we interrupt processing */ 1505 netlink_queue_skip(nlh, skb); 1506 return err; 1507 } 1508 skip: 1509 if (nlh->nlmsg_flags & NLM_F_ACK || err) 1510 netlink_ack(skb, nlh, err); 1511 1512 netlink_queue_skip(nlh, skb); 1513 } 1514 1515 return 0; 1516 } 1517 1518 /** 1519 * nelink_run_queue - Process netlink receive queue. 1520 * @sk: Netlink socket containing the queue 1521 * @qlen: Place to store queue length upon entry 1522 * @cb: Callback function invoked for each netlink message found 1523 * 1524 * Processes as much as there was in the queue upon entry and invokes 1525 * a callback function for each netlink message found. The callback 1526 * function may refuse a message by returning a negative error code 1527 * but setting the error pointer to 0 in which case this function 1528 * returns with a qlen != 0. 1529 * 1530 * qlen must be initialized to 0 before the initial entry, afterwards 1531 * the function may be called repeatedly until qlen reaches 0. 1532 * 1533 * The callback function may return -EINTR to signal that processing 1534 * of netlink messages shall be interrupted. In this case the message 1535 * currently being processed will NOT be requeued onto the receive 1536 * queue. 1537 */ 1538 void netlink_run_queue(struct sock *sk, unsigned int *qlen, 1539 int (*cb)(struct sk_buff *, struct nlmsghdr *)) 1540 { 1541 struct sk_buff *skb; 1542 1543 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue)) 1544 *qlen = skb_queue_len(&sk->sk_receive_queue); 1545 1546 for (; *qlen; (*qlen)--) { 1547 skb = skb_dequeue(&sk->sk_receive_queue); 1548 if (netlink_rcv_skb(skb, cb)) { 1549 if (skb->len) 1550 skb_queue_head(&sk->sk_receive_queue, skb); 1551 else { 1552 kfree_skb(skb); 1553 (*qlen)--; 1554 } 1555 break; 1556 } 1557 1558 kfree_skb(skb); 1559 } 1560 } 1561 1562 /** 1563 * netlink_queue_skip - Skip netlink message while processing queue. 1564 * @nlh: Netlink message to be skipped 1565 * @skb: Socket buffer containing the netlink messages. 1566 * 1567 * Pulls the given netlink message off the socket buffer so the next 1568 * call to netlink_queue_run() will not reconsider the message. 1569 */ 1570 static void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb) 1571 { 1572 int msglen = NLMSG_ALIGN(nlh->nlmsg_len); 1573 1574 if (msglen > skb->len) 1575 msglen = skb->len; 1576 1577 skb_pull(skb, msglen); 1578 } 1579 1580 /** 1581 * nlmsg_notify - send a notification netlink message 1582 * @sk: netlink socket to use 1583 * @skb: notification message 1584 * @pid: destination netlink pid for reports or 0 1585 * @group: destination multicast group or 0 1586 * @report: 1 to report back, 0 to disable 1587 * @flags: allocation flags 1588 */ 1589 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid, 1590 unsigned int group, int report, gfp_t flags) 1591 { 1592 int err = 0; 1593 1594 if (group) { 1595 int exclude_pid = 0; 1596 1597 if (report) { 1598 atomic_inc(&skb->users); 1599 exclude_pid = pid; 1600 } 1601 1602 /* errors reported via destination sk->sk_err */ 1603 nlmsg_multicast(sk, skb, exclude_pid, group, flags); 1604 } 1605 1606 if (report) 1607 err = nlmsg_unicast(sk, skb, pid); 1608 1609 return err; 1610 } 1611 1612 #ifdef CONFIG_PROC_FS 1613 struct nl_seq_iter { 1614 int link; 1615 int hash_idx; 1616 }; 1617 1618 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos) 1619 { 1620 struct nl_seq_iter *iter = seq->private; 1621 int i, j; 1622 struct sock *s; 1623 struct hlist_node *node; 1624 loff_t off = 0; 1625 1626 for (i=0; i<MAX_LINKS; i++) { 1627 struct nl_pid_hash *hash = &nl_table[i].hash; 1628 1629 for (j = 0; j <= hash->mask; j++) { 1630 sk_for_each(s, node, &hash->table[j]) { 1631 if (off == pos) { 1632 iter->link = i; 1633 iter->hash_idx = j; 1634 return s; 1635 } 1636 ++off; 1637 } 1638 } 1639 } 1640 return NULL; 1641 } 1642 1643 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos) 1644 { 1645 read_lock(&nl_table_lock); 1646 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN; 1647 } 1648 1649 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1650 { 1651 struct sock *s; 1652 struct nl_seq_iter *iter; 1653 int i, j; 1654 1655 ++*pos; 1656 1657 if (v == SEQ_START_TOKEN) 1658 return netlink_seq_socket_idx(seq, 0); 1659 1660 s = sk_next(v); 1661 if (s) 1662 return s; 1663 1664 iter = seq->private; 1665 i = iter->link; 1666 j = iter->hash_idx + 1; 1667 1668 do { 1669 struct nl_pid_hash *hash = &nl_table[i].hash; 1670 1671 for (; j <= hash->mask; j++) { 1672 s = sk_head(&hash->table[j]); 1673 if (s) { 1674 iter->link = i; 1675 iter->hash_idx = j; 1676 return s; 1677 } 1678 } 1679 1680 j = 0; 1681 } while (++i < MAX_LINKS); 1682 1683 return NULL; 1684 } 1685 1686 static void netlink_seq_stop(struct seq_file *seq, void *v) 1687 { 1688 read_unlock(&nl_table_lock); 1689 } 1690 1691 1692 static int netlink_seq_show(struct seq_file *seq, void *v) 1693 { 1694 if (v == SEQ_START_TOKEN) 1695 seq_puts(seq, 1696 "sk Eth Pid Groups " 1697 "Rmem Wmem Dump Locks\n"); 1698 else { 1699 struct sock *s = v; 1700 struct netlink_sock *nlk = nlk_sk(s); 1701 1702 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n", 1703 s, 1704 s->sk_protocol, 1705 nlk->pid, 1706 nlk->groups ? (u32)nlk->groups[0] : 0, 1707 atomic_read(&s->sk_rmem_alloc), 1708 atomic_read(&s->sk_wmem_alloc), 1709 nlk->cb, 1710 atomic_read(&s->sk_refcnt) 1711 ); 1712 1713 } 1714 return 0; 1715 } 1716 1717 static struct seq_operations netlink_seq_ops = { 1718 .start = netlink_seq_start, 1719 .next = netlink_seq_next, 1720 .stop = netlink_seq_stop, 1721 .show = netlink_seq_show, 1722 }; 1723 1724 1725 static int netlink_seq_open(struct inode *inode, struct file *file) 1726 { 1727 struct seq_file *seq; 1728 struct nl_seq_iter *iter; 1729 int err; 1730 1731 iter = kzalloc(sizeof(*iter), GFP_KERNEL); 1732 if (!iter) 1733 return -ENOMEM; 1734 1735 err = seq_open(file, &netlink_seq_ops); 1736 if (err) { 1737 kfree(iter); 1738 return err; 1739 } 1740 1741 seq = file->private_data; 1742 seq->private = iter; 1743 return 0; 1744 } 1745 1746 static const struct file_operations netlink_seq_fops = { 1747 .owner = THIS_MODULE, 1748 .open = netlink_seq_open, 1749 .read = seq_read, 1750 .llseek = seq_lseek, 1751 .release = seq_release_private, 1752 }; 1753 1754 #endif 1755 1756 int netlink_register_notifier(struct notifier_block *nb) 1757 { 1758 return atomic_notifier_chain_register(&netlink_chain, nb); 1759 } 1760 1761 int netlink_unregister_notifier(struct notifier_block *nb) 1762 { 1763 return atomic_notifier_chain_unregister(&netlink_chain, nb); 1764 } 1765 1766 static const struct proto_ops netlink_ops = { 1767 .family = PF_NETLINK, 1768 .owner = THIS_MODULE, 1769 .release = netlink_release, 1770 .bind = netlink_bind, 1771 .connect = netlink_connect, 1772 .socketpair = sock_no_socketpair, 1773 .accept = sock_no_accept, 1774 .getname = netlink_getname, 1775 .poll = datagram_poll, 1776 .ioctl = sock_no_ioctl, 1777 .listen = sock_no_listen, 1778 .shutdown = sock_no_shutdown, 1779 .setsockopt = netlink_setsockopt, 1780 .getsockopt = netlink_getsockopt, 1781 .sendmsg = netlink_sendmsg, 1782 .recvmsg = netlink_recvmsg, 1783 .mmap = sock_no_mmap, 1784 .sendpage = sock_no_sendpage, 1785 }; 1786 1787 static struct net_proto_family netlink_family_ops = { 1788 .family = PF_NETLINK, 1789 .create = netlink_create, 1790 .owner = THIS_MODULE, /* for consistency 8) */ 1791 }; 1792 1793 static int __init netlink_proto_init(void) 1794 { 1795 struct sk_buff *dummy_skb; 1796 int i; 1797 unsigned long max; 1798 unsigned int order; 1799 int err = proto_register(&netlink_proto, 0); 1800 1801 if (err != 0) 1802 goto out; 1803 1804 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb)); 1805 1806 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL); 1807 if (!nl_table) 1808 goto panic; 1809 1810 if (num_physpages >= (128 * 1024)) 1811 max = num_physpages >> (21 - PAGE_SHIFT); 1812 else 1813 max = num_physpages >> (23 - PAGE_SHIFT); 1814 1815 order = get_bitmask_order(max) - 1 + PAGE_SHIFT; 1816 max = (1UL << order) / sizeof(struct hlist_head); 1817 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1; 1818 1819 for (i = 0; i < MAX_LINKS; i++) { 1820 struct nl_pid_hash *hash = &nl_table[i].hash; 1821 1822 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table)); 1823 if (!hash->table) { 1824 while (i-- > 0) 1825 nl_pid_hash_free(nl_table[i].hash.table, 1826 1 * sizeof(*hash->table)); 1827 kfree(nl_table); 1828 goto panic; 1829 } 1830 memset(hash->table, 0, 1 * sizeof(*hash->table)); 1831 hash->max_shift = order; 1832 hash->shift = 0; 1833 hash->mask = 0; 1834 hash->rehash_time = jiffies; 1835 } 1836 1837 sock_register(&netlink_family_ops); 1838 #ifdef CONFIG_PROC_FS 1839 proc_net_fops_create("netlink", 0, &netlink_seq_fops); 1840 #endif 1841 /* The netlink device handler may be needed early. */ 1842 rtnetlink_init(); 1843 out: 1844 return err; 1845 panic: 1846 panic("netlink_init: Cannot allocate nl_table\n"); 1847 } 1848 1849 core_initcall(netlink_proto_init); 1850 1851 EXPORT_SYMBOL(netlink_ack); 1852 EXPORT_SYMBOL(netlink_run_queue); 1853 EXPORT_SYMBOL(netlink_broadcast); 1854 EXPORT_SYMBOL(netlink_dump_start); 1855 EXPORT_SYMBOL(netlink_kernel_create); 1856 EXPORT_SYMBOL(netlink_register_notifier); 1857 EXPORT_SYMBOL(netlink_set_nonroot); 1858 EXPORT_SYMBOL(netlink_unicast); 1859 EXPORT_SYMBOL(netlink_unregister_notifier); 1860 EXPORT_SYMBOL(nlmsg_notify); 1861