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 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 66 struct netlink_sock { 67 /* struct sock has to be the first member of netlink_sock */ 68 struct sock sk; 69 u32 pid; 70 u32 dst_pid; 71 u32 dst_group; 72 u32 flags; 73 u32 subscriptions; 74 u32 ngroups; 75 unsigned long *groups; 76 unsigned long state; 77 wait_queue_head_t wait; 78 struct netlink_callback *cb; 79 spinlock_t cb_lock; 80 void (*data_ready)(struct sock *sk, int bytes); 81 struct module *module; 82 }; 83 84 #define NETLINK_KERNEL_SOCKET 0x1 85 #define NETLINK_RECV_PKTINFO 0x2 86 87 static inline struct netlink_sock *nlk_sk(struct sock *sk) 88 { 89 return (struct netlink_sock *)sk; 90 } 91 92 struct nl_pid_hash { 93 struct hlist_head *table; 94 unsigned long rehash_time; 95 96 unsigned int mask; 97 unsigned int shift; 98 99 unsigned int entries; 100 unsigned int max_shift; 101 102 u32 rnd; 103 }; 104 105 struct netlink_table { 106 struct nl_pid_hash hash; 107 struct hlist_head mc_list; 108 unsigned long *listeners; 109 unsigned int nl_nonroot; 110 unsigned int groups; 111 struct module *module; 112 int registered; 113 }; 114 115 static struct netlink_table *nl_table; 116 117 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait); 118 119 static int netlink_dump(struct sock *sk); 120 static void netlink_destroy_callback(struct netlink_callback *cb); 121 122 static DEFINE_RWLOCK(nl_table_lock); 123 static atomic_t nl_table_users = ATOMIC_INIT(0); 124 125 static ATOMIC_NOTIFIER_HEAD(netlink_chain); 126 127 static u32 netlink_group_mask(u32 group) 128 { 129 return group ? 1 << (group - 1) : 0; 130 } 131 132 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid) 133 { 134 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask]; 135 } 136 137 static void netlink_sock_destruct(struct sock *sk) 138 { 139 skb_queue_purge(&sk->sk_receive_queue); 140 141 if (!sock_flag(sk, SOCK_DEAD)) { 142 printk("Freeing alive netlink socket %p\n", sk); 143 return; 144 } 145 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc)); 146 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc)); 147 BUG_TRAP(!nlk_sk(sk)->cb); 148 BUG_TRAP(!nlk_sk(sk)->groups); 149 } 150 151 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP. 152 * Look, when several writers sleep and reader wakes them up, all but one 153 * immediately hit write lock and grab all the cpus. Exclusive sleep solves 154 * this, _but_ remember, it adds useless work on UP machines. 155 */ 156 157 static void netlink_table_grab(void) 158 { 159 write_lock_irq(&nl_table_lock); 160 161 if (atomic_read(&nl_table_users)) { 162 DECLARE_WAITQUEUE(wait, current); 163 164 add_wait_queue_exclusive(&nl_table_wait, &wait); 165 for(;;) { 166 set_current_state(TASK_UNINTERRUPTIBLE); 167 if (atomic_read(&nl_table_users) == 0) 168 break; 169 write_unlock_irq(&nl_table_lock); 170 schedule(); 171 write_lock_irq(&nl_table_lock); 172 } 173 174 __set_current_state(TASK_RUNNING); 175 remove_wait_queue(&nl_table_wait, &wait); 176 } 177 } 178 179 static __inline__ void netlink_table_ungrab(void) 180 { 181 write_unlock_irq(&nl_table_lock); 182 wake_up(&nl_table_wait); 183 } 184 185 static __inline__ void 186 netlink_lock_table(void) 187 { 188 /* read_lock() synchronizes us to netlink_table_grab */ 189 190 read_lock(&nl_table_lock); 191 atomic_inc(&nl_table_users); 192 read_unlock(&nl_table_lock); 193 } 194 195 static __inline__ void 196 netlink_unlock_table(void) 197 { 198 if (atomic_dec_and_test(&nl_table_users)) 199 wake_up(&nl_table_wait); 200 } 201 202 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid) 203 { 204 struct nl_pid_hash *hash = &nl_table[protocol].hash; 205 struct hlist_head *head; 206 struct sock *sk; 207 struct hlist_node *node; 208 209 read_lock(&nl_table_lock); 210 head = nl_pid_hashfn(hash, pid); 211 sk_for_each(sk, node, head) { 212 if (nlk_sk(sk)->pid == pid) { 213 sock_hold(sk); 214 goto found; 215 } 216 } 217 sk = NULL; 218 found: 219 read_unlock(&nl_table_lock); 220 return sk; 221 } 222 223 static inline struct hlist_head *nl_pid_hash_alloc(size_t size) 224 { 225 if (size <= PAGE_SIZE) 226 return kmalloc(size, GFP_ATOMIC); 227 else 228 return (struct hlist_head *) 229 __get_free_pages(GFP_ATOMIC, get_order(size)); 230 } 231 232 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size) 233 { 234 if (size <= PAGE_SIZE) 235 kfree(table); 236 else 237 free_pages((unsigned long)table, get_order(size)); 238 } 239 240 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow) 241 { 242 unsigned int omask, mask, shift; 243 size_t osize, size; 244 struct hlist_head *otable, *table; 245 int i; 246 247 omask = mask = hash->mask; 248 osize = size = (mask + 1) * sizeof(*table); 249 shift = hash->shift; 250 251 if (grow) { 252 if (++shift > hash->max_shift) 253 return 0; 254 mask = mask * 2 + 1; 255 size *= 2; 256 } 257 258 table = nl_pid_hash_alloc(size); 259 if (!table) 260 return 0; 261 262 memset(table, 0, size); 263 otable = hash->table; 264 hash->table = table; 265 hash->mask = mask; 266 hash->shift = shift; 267 get_random_bytes(&hash->rnd, sizeof(hash->rnd)); 268 269 for (i = 0; i <= omask; i++) { 270 struct sock *sk; 271 struct hlist_node *node, *tmp; 272 273 sk_for_each_safe(sk, node, tmp, &otable[i]) 274 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid)); 275 } 276 277 nl_pid_hash_free(otable, osize); 278 hash->rehash_time = jiffies + 10 * 60 * HZ; 279 return 1; 280 } 281 282 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len) 283 { 284 int avg = hash->entries >> hash->shift; 285 286 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1)) 287 return 1; 288 289 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) { 290 nl_pid_hash_rehash(hash, 0); 291 return 1; 292 } 293 294 return 0; 295 } 296 297 static const struct proto_ops netlink_ops; 298 299 static void 300 netlink_update_listeners(struct sock *sk) 301 { 302 struct netlink_table *tbl = &nl_table[sk->sk_protocol]; 303 struct hlist_node *node; 304 unsigned long mask; 305 unsigned int i; 306 307 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) { 308 mask = 0; 309 sk_for_each_bound(sk, node, &tbl->mc_list) 310 mask |= nlk_sk(sk)->groups[i]; 311 tbl->listeners[i] = mask; 312 } 313 /* this function is only called with the netlink table "grabbed", which 314 * makes sure updates are visible before bind or setsockopt return. */ 315 } 316 317 static int netlink_insert(struct sock *sk, u32 pid) 318 { 319 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 320 struct hlist_head *head; 321 int err = -EADDRINUSE; 322 struct sock *osk; 323 struct hlist_node *node; 324 int len; 325 326 netlink_table_grab(); 327 head = nl_pid_hashfn(hash, pid); 328 len = 0; 329 sk_for_each(osk, node, head) { 330 if (nlk_sk(osk)->pid == pid) 331 break; 332 len++; 333 } 334 if (node) 335 goto err; 336 337 err = -EBUSY; 338 if (nlk_sk(sk)->pid) 339 goto err; 340 341 err = -ENOMEM; 342 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX)) 343 goto err; 344 345 if (len && nl_pid_hash_dilute(hash, len)) 346 head = nl_pid_hashfn(hash, pid); 347 hash->entries++; 348 nlk_sk(sk)->pid = pid; 349 sk_add_node(sk, head); 350 err = 0; 351 352 err: 353 netlink_table_ungrab(); 354 return err; 355 } 356 357 static void netlink_remove(struct sock *sk) 358 { 359 netlink_table_grab(); 360 if (sk_del_node_init(sk)) 361 nl_table[sk->sk_protocol].hash.entries--; 362 if (nlk_sk(sk)->subscriptions) 363 __sk_del_bind_node(sk); 364 netlink_table_ungrab(); 365 } 366 367 static struct proto netlink_proto = { 368 .name = "NETLINK", 369 .owner = THIS_MODULE, 370 .obj_size = sizeof(struct netlink_sock), 371 }; 372 373 static int __netlink_create(struct socket *sock, int protocol) 374 { 375 struct sock *sk; 376 struct netlink_sock *nlk; 377 378 sock->ops = &netlink_ops; 379 380 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1); 381 if (!sk) 382 return -ENOMEM; 383 384 sock_init_data(sock, sk); 385 386 nlk = nlk_sk(sk); 387 spin_lock_init(&nlk->cb_lock); 388 init_waitqueue_head(&nlk->wait); 389 390 sk->sk_destruct = netlink_sock_destruct; 391 sk->sk_protocol = protocol; 392 return 0; 393 } 394 395 static int netlink_create(struct socket *sock, int protocol) 396 { 397 struct module *module = NULL; 398 struct netlink_sock *nlk; 399 unsigned int groups; 400 int err = 0; 401 402 sock->state = SS_UNCONNECTED; 403 404 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 405 return -ESOCKTNOSUPPORT; 406 407 if (protocol<0 || protocol >= MAX_LINKS) 408 return -EPROTONOSUPPORT; 409 410 netlink_lock_table(); 411 #ifdef CONFIG_KMOD 412 if (!nl_table[protocol].registered) { 413 netlink_unlock_table(); 414 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol); 415 netlink_lock_table(); 416 } 417 #endif 418 if (nl_table[protocol].registered && 419 try_module_get(nl_table[protocol].module)) 420 module = nl_table[protocol].module; 421 groups = nl_table[protocol].groups; 422 netlink_unlock_table(); 423 424 if ((err = __netlink_create(sock, protocol)) < 0) 425 goto out_module; 426 427 nlk = nlk_sk(sock->sk); 428 nlk->module = module; 429 out: 430 return err; 431 432 out_module: 433 module_put(module); 434 goto out; 435 } 436 437 static int netlink_release(struct socket *sock) 438 { 439 struct sock *sk = sock->sk; 440 struct netlink_sock *nlk; 441 442 if (!sk) 443 return 0; 444 445 netlink_remove(sk); 446 nlk = nlk_sk(sk); 447 448 spin_lock(&nlk->cb_lock); 449 if (nlk->cb) { 450 if (nlk->cb->done) 451 nlk->cb->done(nlk->cb); 452 netlink_destroy_callback(nlk->cb); 453 nlk->cb = NULL; 454 } 455 spin_unlock(&nlk->cb_lock); 456 457 /* OK. Socket is unlinked, and, therefore, 458 no new packets will arrive */ 459 460 sock_orphan(sk); 461 sock->sk = NULL; 462 wake_up_interruptible_all(&nlk->wait); 463 464 skb_queue_purge(&sk->sk_write_queue); 465 466 if (nlk->pid && !nlk->subscriptions) { 467 struct netlink_notify n = { 468 .protocol = sk->sk_protocol, 469 .pid = nlk->pid, 470 }; 471 atomic_notifier_call_chain(&netlink_chain, 472 NETLINK_URELEASE, &n); 473 } 474 475 if (nlk->module) 476 module_put(nlk->module); 477 478 netlink_table_grab(); 479 if (nlk->flags & NETLINK_KERNEL_SOCKET) { 480 kfree(nl_table[sk->sk_protocol].listeners); 481 nl_table[sk->sk_protocol].module = NULL; 482 nl_table[sk->sk_protocol].registered = 0; 483 } else if (nlk->subscriptions) 484 netlink_update_listeners(sk); 485 netlink_table_ungrab(); 486 487 kfree(nlk->groups); 488 nlk->groups = NULL; 489 490 sock_put(sk); 491 return 0; 492 } 493 494 static int netlink_autobind(struct socket *sock) 495 { 496 struct sock *sk = sock->sk; 497 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 498 struct hlist_head *head; 499 struct sock *osk; 500 struct hlist_node *node; 501 s32 pid = current->tgid; 502 int err; 503 static s32 rover = -4097; 504 505 retry: 506 cond_resched(); 507 netlink_table_grab(); 508 head = nl_pid_hashfn(hash, pid); 509 sk_for_each(osk, node, head) { 510 if (nlk_sk(osk)->pid == pid) { 511 /* Bind collision, search negative pid values. */ 512 pid = rover--; 513 if (rover > -4097) 514 rover = -4097; 515 netlink_table_ungrab(); 516 goto retry; 517 } 518 } 519 netlink_table_ungrab(); 520 521 err = netlink_insert(sk, pid); 522 if (err == -EADDRINUSE) 523 goto retry; 524 525 /* If 2 threads race to autobind, that is fine. */ 526 if (err == -EBUSY) 527 err = 0; 528 529 return err; 530 } 531 532 static inline int netlink_capable(struct socket *sock, unsigned int flag) 533 { 534 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) || 535 capable(CAP_NET_ADMIN); 536 } 537 538 static void 539 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions) 540 { 541 struct netlink_sock *nlk = nlk_sk(sk); 542 543 if (nlk->subscriptions && !subscriptions) 544 __sk_del_bind_node(sk); 545 else if (!nlk->subscriptions && subscriptions) 546 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 547 nlk->subscriptions = subscriptions; 548 } 549 550 static int netlink_alloc_groups(struct sock *sk) 551 { 552 struct netlink_sock *nlk = nlk_sk(sk); 553 unsigned int groups; 554 int err = 0; 555 556 netlink_lock_table(); 557 groups = nl_table[sk->sk_protocol].groups; 558 if (!nl_table[sk->sk_protocol].registered) 559 err = -ENOENT; 560 netlink_unlock_table(); 561 562 if (err) 563 return err; 564 565 nlk->groups = kmalloc(NLGRPSZ(groups), GFP_KERNEL); 566 if (nlk->groups == NULL) 567 return -ENOMEM; 568 memset(nlk->groups, 0, NLGRPSZ(groups)); 569 nlk->ngroups = groups; 570 return 0; 571 } 572 573 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 574 { 575 struct sock *sk = sock->sk; 576 struct netlink_sock *nlk = nlk_sk(sk); 577 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 578 int err; 579 580 if (nladdr->nl_family != AF_NETLINK) 581 return -EINVAL; 582 583 /* Only superuser is allowed to listen multicasts */ 584 if (nladdr->nl_groups) { 585 if (!netlink_capable(sock, NL_NONROOT_RECV)) 586 return -EPERM; 587 if (nlk->groups == NULL) { 588 err = netlink_alloc_groups(sk); 589 if (err) 590 return err; 591 } 592 } 593 594 if (nlk->pid) { 595 if (nladdr->nl_pid != nlk->pid) 596 return -EINVAL; 597 } else { 598 err = nladdr->nl_pid ? 599 netlink_insert(sk, nladdr->nl_pid) : 600 netlink_autobind(sock); 601 if (err) 602 return err; 603 } 604 605 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0])) 606 return 0; 607 608 netlink_table_grab(); 609 netlink_update_subscriptions(sk, nlk->subscriptions + 610 hweight32(nladdr->nl_groups) - 611 hweight32(nlk->groups[0])); 612 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups; 613 netlink_update_listeners(sk); 614 netlink_table_ungrab(); 615 616 return 0; 617 } 618 619 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 620 int alen, int flags) 621 { 622 int err = 0; 623 struct sock *sk = sock->sk; 624 struct netlink_sock *nlk = nlk_sk(sk); 625 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr; 626 627 if (addr->sa_family == AF_UNSPEC) { 628 sk->sk_state = NETLINK_UNCONNECTED; 629 nlk->dst_pid = 0; 630 nlk->dst_group = 0; 631 return 0; 632 } 633 if (addr->sa_family != AF_NETLINK) 634 return -EINVAL; 635 636 /* Only superuser is allowed to send multicasts */ 637 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND)) 638 return -EPERM; 639 640 if (!nlk->pid) 641 err = netlink_autobind(sock); 642 643 if (err == 0) { 644 sk->sk_state = NETLINK_CONNECTED; 645 nlk->dst_pid = nladdr->nl_pid; 646 nlk->dst_group = ffs(nladdr->nl_groups); 647 } 648 649 return err; 650 } 651 652 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer) 653 { 654 struct sock *sk = sock->sk; 655 struct netlink_sock *nlk = nlk_sk(sk); 656 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr; 657 658 nladdr->nl_family = AF_NETLINK; 659 nladdr->nl_pad = 0; 660 *addr_len = sizeof(*nladdr); 661 662 if (peer) { 663 nladdr->nl_pid = nlk->dst_pid; 664 nladdr->nl_groups = netlink_group_mask(nlk->dst_group); 665 } else { 666 nladdr->nl_pid = nlk->pid; 667 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0; 668 } 669 return 0; 670 } 671 672 static void netlink_overrun(struct sock *sk) 673 { 674 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) { 675 sk->sk_err = ENOBUFS; 676 sk->sk_error_report(sk); 677 } 678 } 679 680 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid) 681 { 682 int protocol = ssk->sk_protocol; 683 struct sock *sock; 684 struct netlink_sock *nlk; 685 686 sock = netlink_lookup(protocol, pid); 687 if (!sock) 688 return ERR_PTR(-ECONNREFUSED); 689 690 /* Don't bother queuing skb if kernel socket has no input function */ 691 nlk = nlk_sk(sock); 692 if ((nlk->pid == 0 && !nlk->data_ready) || 693 (sock->sk_state == NETLINK_CONNECTED && 694 nlk->dst_pid != nlk_sk(ssk)->pid)) { 695 sock_put(sock); 696 return ERR_PTR(-ECONNREFUSED); 697 } 698 return sock; 699 } 700 701 struct sock *netlink_getsockbyfilp(struct file *filp) 702 { 703 struct inode *inode = filp->f_dentry->d_inode; 704 struct sock *sock; 705 706 if (!S_ISSOCK(inode->i_mode)) 707 return ERR_PTR(-ENOTSOCK); 708 709 sock = SOCKET_I(inode)->sk; 710 if (sock->sk_family != AF_NETLINK) 711 return ERR_PTR(-EINVAL); 712 713 sock_hold(sock); 714 return sock; 715 } 716 717 /* 718 * Attach a skb to a netlink socket. 719 * The caller must hold a reference to the destination socket. On error, the 720 * reference is dropped. The skb is not send to the destination, just all 721 * all error checks are performed and memory in the queue is reserved. 722 * Return values: 723 * < 0: error. skb freed, reference to sock dropped. 724 * 0: continue 725 * 1: repeat lookup - reference dropped while waiting for socket memory. 726 */ 727 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, 728 long timeo, struct sock *ssk) 729 { 730 struct netlink_sock *nlk; 731 732 nlk = nlk_sk(sk); 733 734 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 735 test_bit(0, &nlk->state)) { 736 DECLARE_WAITQUEUE(wait, current); 737 if (!timeo) { 738 if (!ssk || nlk_sk(ssk)->pid == 0) 739 netlink_overrun(sk); 740 sock_put(sk); 741 kfree_skb(skb); 742 return -EAGAIN; 743 } 744 745 __set_current_state(TASK_INTERRUPTIBLE); 746 add_wait_queue(&nlk->wait, &wait); 747 748 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 749 test_bit(0, &nlk->state)) && 750 !sock_flag(sk, SOCK_DEAD)) 751 timeo = schedule_timeout(timeo); 752 753 __set_current_state(TASK_RUNNING); 754 remove_wait_queue(&nlk->wait, &wait); 755 sock_put(sk); 756 757 if (signal_pending(current)) { 758 kfree_skb(skb); 759 return sock_intr_errno(timeo); 760 } 761 return 1; 762 } 763 skb_set_owner_r(skb, sk); 764 return 0; 765 } 766 767 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol) 768 { 769 int len = skb->len; 770 771 skb_queue_tail(&sk->sk_receive_queue, skb); 772 sk->sk_data_ready(sk, len); 773 sock_put(sk); 774 return len; 775 } 776 777 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 778 { 779 kfree_skb(skb); 780 sock_put(sk); 781 } 782 783 static inline struct sk_buff *netlink_trim(struct sk_buff *skb, 784 gfp_t allocation) 785 { 786 int delta; 787 788 skb_orphan(skb); 789 790 delta = skb->end - skb->tail; 791 if (delta * 2 < skb->truesize) 792 return skb; 793 794 if (skb_shared(skb)) { 795 struct sk_buff *nskb = skb_clone(skb, allocation); 796 if (!nskb) 797 return skb; 798 kfree_skb(skb); 799 skb = nskb; 800 } 801 802 if (!pskb_expand_head(skb, 0, -delta, allocation)) 803 skb->truesize -= delta; 804 805 return skb; 806 } 807 808 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock) 809 { 810 struct sock *sk; 811 int err; 812 long timeo; 813 814 skb = netlink_trim(skb, gfp_any()); 815 816 timeo = sock_sndtimeo(ssk, nonblock); 817 retry: 818 sk = netlink_getsockbypid(ssk, pid); 819 if (IS_ERR(sk)) { 820 kfree_skb(skb); 821 return PTR_ERR(sk); 822 } 823 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk); 824 if (err == 1) 825 goto retry; 826 if (err) 827 return err; 828 829 return netlink_sendskb(sk, skb, ssk->sk_protocol); 830 } 831 832 int netlink_has_listeners(struct sock *sk, unsigned int group) 833 { 834 int res = 0; 835 836 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET)); 837 if (group - 1 < nl_table[sk->sk_protocol].groups) 838 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners); 839 return res; 840 } 841 EXPORT_SYMBOL_GPL(netlink_has_listeners); 842 843 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 844 { 845 struct netlink_sock *nlk = nlk_sk(sk); 846 847 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 848 !test_bit(0, &nlk->state)) { 849 skb_set_owner_r(skb, sk); 850 skb_queue_tail(&sk->sk_receive_queue, skb); 851 sk->sk_data_ready(sk, skb->len); 852 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf; 853 } 854 return -1; 855 } 856 857 struct netlink_broadcast_data { 858 struct sock *exclude_sk; 859 u32 pid; 860 u32 group; 861 int failure; 862 int congested; 863 int delivered; 864 gfp_t allocation; 865 struct sk_buff *skb, *skb2; 866 }; 867 868 static inline int do_one_broadcast(struct sock *sk, 869 struct netlink_broadcast_data *p) 870 { 871 struct netlink_sock *nlk = nlk_sk(sk); 872 int val; 873 874 if (p->exclude_sk == sk) 875 goto out; 876 877 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups || 878 !test_bit(p->group - 1, nlk->groups)) 879 goto out; 880 881 if (p->failure) { 882 netlink_overrun(sk); 883 goto out; 884 } 885 886 sock_hold(sk); 887 if (p->skb2 == NULL) { 888 if (skb_shared(p->skb)) { 889 p->skb2 = skb_clone(p->skb, p->allocation); 890 } else { 891 p->skb2 = skb_get(p->skb); 892 /* 893 * skb ownership may have been set when 894 * delivered to a previous socket. 895 */ 896 skb_orphan(p->skb2); 897 } 898 } 899 if (p->skb2 == NULL) { 900 netlink_overrun(sk); 901 /* Clone failed. Notify ALL listeners. */ 902 p->failure = 1; 903 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) { 904 netlink_overrun(sk); 905 } else { 906 p->congested |= val; 907 p->delivered = 1; 908 p->skb2 = NULL; 909 } 910 sock_put(sk); 911 912 out: 913 return 0; 914 } 915 916 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid, 917 u32 group, gfp_t allocation) 918 { 919 struct netlink_broadcast_data info; 920 struct hlist_node *node; 921 struct sock *sk; 922 923 skb = netlink_trim(skb, allocation); 924 925 info.exclude_sk = ssk; 926 info.pid = pid; 927 info.group = group; 928 info.failure = 0; 929 info.congested = 0; 930 info.delivered = 0; 931 info.allocation = allocation; 932 info.skb = skb; 933 info.skb2 = NULL; 934 935 /* While we sleep in clone, do not allow to change socket list */ 936 937 netlink_lock_table(); 938 939 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 940 do_one_broadcast(sk, &info); 941 942 kfree_skb(skb); 943 944 netlink_unlock_table(); 945 946 if (info.skb2) 947 kfree_skb(info.skb2); 948 949 if (info.delivered) { 950 if (info.congested && (allocation & __GFP_WAIT)) 951 yield(); 952 return 0; 953 } 954 if (info.failure) 955 return -ENOBUFS; 956 return -ESRCH; 957 } 958 959 struct netlink_set_err_data { 960 struct sock *exclude_sk; 961 u32 pid; 962 u32 group; 963 int code; 964 }; 965 966 static inline int do_one_set_err(struct sock *sk, 967 struct netlink_set_err_data *p) 968 { 969 struct netlink_sock *nlk = nlk_sk(sk); 970 971 if (sk == p->exclude_sk) 972 goto out; 973 974 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups || 975 !test_bit(p->group - 1, nlk->groups)) 976 goto out; 977 978 sk->sk_err = p->code; 979 sk->sk_error_report(sk); 980 out: 981 return 0; 982 } 983 984 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code) 985 { 986 struct netlink_set_err_data info; 987 struct hlist_node *node; 988 struct sock *sk; 989 990 info.exclude_sk = ssk; 991 info.pid = pid; 992 info.group = group; 993 info.code = code; 994 995 read_lock(&nl_table_lock); 996 997 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 998 do_one_set_err(sk, &info); 999 1000 read_unlock(&nl_table_lock); 1001 } 1002 1003 static int netlink_setsockopt(struct socket *sock, int level, int optname, 1004 char __user *optval, int optlen) 1005 { 1006 struct sock *sk = sock->sk; 1007 struct netlink_sock *nlk = nlk_sk(sk); 1008 int val = 0, err; 1009 1010 if (level != SOL_NETLINK) 1011 return -ENOPROTOOPT; 1012 1013 if (optlen >= sizeof(int) && 1014 get_user(val, (int __user *)optval)) 1015 return -EFAULT; 1016 1017 switch (optname) { 1018 case NETLINK_PKTINFO: 1019 if (val) 1020 nlk->flags |= NETLINK_RECV_PKTINFO; 1021 else 1022 nlk->flags &= ~NETLINK_RECV_PKTINFO; 1023 err = 0; 1024 break; 1025 case NETLINK_ADD_MEMBERSHIP: 1026 case NETLINK_DROP_MEMBERSHIP: { 1027 unsigned int subscriptions; 1028 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0; 1029 1030 if (!netlink_capable(sock, NL_NONROOT_RECV)) 1031 return -EPERM; 1032 if (nlk->groups == NULL) { 1033 err = netlink_alloc_groups(sk); 1034 if (err) 1035 return err; 1036 } 1037 if (!val || val - 1 >= nlk->ngroups) 1038 return -EINVAL; 1039 netlink_table_grab(); 1040 old = test_bit(val - 1, nlk->groups); 1041 subscriptions = nlk->subscriptions - old + new; 1042 if (new) 1043 __set_bit(val - 1, nlk->groups); 1044 else 1045 __clear_bit(val - 1, nlk->groups); 1046 netlink_update_subscriptions(sk, subscriptions); 1047 netlink_update_listeners(sk); 1048 netlink_table_ungrab(); 1049 err = 0; 1050 break; 1051 } 1052 default: 1053 err = -ENOPROTOOPT; 1054 } 1055 return err; 1056 } 1057 1058 static int netlink_getsockopt(struct socket *sock, int level, int optname, 1059 char __user *optval, int __user *optlen) 1060 { 1061 struct sock *sk = sock->sk; 1062 struct netlink_sock *nlk = nlk_sk(sk); 1063 int len, val, err; 1064 1065 if (level != SOL_NETLINK) 1066 return -ENOPROTOOPT; 1067 1068 if (get_user(len, optlen)) 1069 return -EFAULT; 1070 if (len < 0) 1071 return -EINVAL; 1072 1073 switch (optname) { 1074 case NETLINK_PKTINFO: 1075 if (len < sizeof(int)) 1076 return -EINVAL; 1077 len = sizeof(int); 1078 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0; 1079 put_user(len, optlen); 1080 put_user(val, optval); 1081 err = 0; 1082 break; 1083 default: 1084 err = -ENOPROTOOPT; 1085 } 1086 return err; 1087 } 1088 1089 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb) 1090 { 1091 struct nl_pktinfo info; 1092 1093 info.group = NETLINK_CB(skb).dst_group; 1094 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info); 1095 } 1096 1097 static inline void netlink_rcv_wake(struct sock *sk) 1098 { 1099 struct netlink_sock *nlk = nlk_sk(sk); 1100 1101 if (skb_queue_empty(&sk->sk_receive_queue)) 1102 clear_bit(0, &nlk->state); 1103 if (!test_bit(0, &nlk->state)) 1104 wake_up_interruptible(&nlk->wait); 1105 } 1106 1107 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock, 1108 struct msghdr *msg, size_t len) 1109 { 1110 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1111 struct sock *sk = sock->sk; 1112 struct netlink_sock *nlk = nlk_sk(sk); 1113 struct sockaddr_nl *addr=msg->msg_name; 1114 u32 dst_pid; 1115 u32 dst_group; 1116 struct sk_buff *skb; 1117 int err; 1118 struct scm_cookie scm; 1119 1120 if (msg->msg_flags&MSG_OOB) 1121 return -EOPNOTSUPP; 1122 1123 if (NULL == siocb->scm) 1124 siocb->scm = &scm; 1125 err = scm_send(sock, msg, siocb->scm); 1126 if (err < 0) 1127 return err; 1128 1129 if (msg->msg_namelen) { 1130 if (addr->nl_family != AF_NETLINK) 1131 return -EINVAL; 1132 dst_pid = addr->nl_pid; 1133 dst_group = ffs(addr->nl_groups); 1134 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND)) 1135 return -EPERM; 1136 } else { 1137 dst_pid = nlk->dst_pid; 1138 dst_group = nlk->dst_group; 1139 } 1140 1141 if (!nlk->pid) { 1142 err = netlink_autobind(sock); 1143 if (err) 1144 goto out; 1145 } 1146 1147 err = -EMSGSIZE; 1148 if (len > sk->sk_sndbuf - 32) 1149 goto out; 1150 err = -ENOBUFS; 1151 skb = alloc_skb(len, GFP_KERNEL); 1152 if (skb==NULL) 1153 goto out; 1154 1155 NETLINK_CB(skb).pid = nlk->pid; 1156 NETLINK_CB(skb).dst_pid = dst_pid; 1157 NETLINK_CB(skb).dst_group = dst_group; 1158 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context); 1159 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid)); 1160 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); 1161 1162 /* What can I do? Netlink is asynchronous, so that 1163 we will have to save current capabilities to 1164 check them, when this message will be delivered 1165 to corresponding kernel module. --ANK (980802) 1166 */ 1167 1168 err = -EFAULT; 1169 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) { 1170 kfree_skb(skb); 1171 goto out; 1172 } 1173 1174 err = security_netlink_send(sk, skb); 1175 if (err) { 1176 kfree_skb(skb); 1177 goto out; 1178 } 1179 1180 if (dst_group) { 1181 atomic_inc(&skb->users); 1182 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL); 1183 } 1184 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT); 1185 1186 out: 1187 return err; 1188 } 1189 1190 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock, 1191 struct msghdr *msg, size_t len, 1192 int flags) 1193 { 1194 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1195 struct scm_cookie scm; 1196 struct sock *sk = sock->sk; 1197 struct netlink_sock *nlk = nlk_sk(sk); 1198 int noblock = flags&MSG_DONTWAIT; 1199 size_t copied; 1200 struct sk_buff *skb; 1201 int err; 1202 1203 if (flags&MSG_OOB) 1204 return -EOPNOTSUPP; 1205 1206 copied = 0; 1207 1208 skb = skb_recv_datagram(sk,flags,noblock,&err); 1209 if (skb==NULL) 1210 goto out; 1211 1212 msg->msg_namelen = 0; 1213 1214 copied = skb->len; 1215 if (len < copied) { 1216 msg->msg_flags |= MSG_TRUNC; 1217 copied = len; 1218 } 1219 1220 skb->h.raw = skb->data; 1221 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 1222 1223 if (msg->msg_name) { 1224 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name; 1225 addr->nl_family = AF_NETLINK; 1226 addr->nl_pad = 0; 1227 addr->nl_pid = NETLINK_CB(skb).pid; 1228 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group); 1229 msg->msg_namelen = sizeof(*addr); 1230 } 1231 1232 if (nlk->flags & NETLINK_RECV_PKTINFO) 1233 netlink_cmsg_recv_pktinfo(msg, skb); 1234 1235 if (NULL == siocb->scm) { 1236 memset(&scm, 0, sizeof(scm)); 1237 siocb->scm = &scm; 1238 } 1239 siocb->scm->creds = *NETLINK_CREDS(skb); 1240 skb_free_datagram(sk, skb); 1241 1242 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) 1243 netlink_dump(sk); 1244 1245 scm_recv(sock, msg, siocb->scm, flags); 1246 1247 out: 1248 netlink_rcv_wake(sk); 1249 return err ? : copied; 1250 } 1251 1252 static void netlink_data_ready(struct sock *sk, int len) 1253 { 1254 struct netlink_sock *nlk = nlk_sk(sk); 1255 1256 if (nlk->data_ready) 1257 nlk->data_ready(sk, len); 1258 netlink_rcv_wake(sk); 1259 } 1260 1261 /* 1262 * We export these functions to other modules. They provide a 1263 * complete set of kernel non-blocking support for message 1264 * queueing. 1265 */ 1266 1267 struct sock * 1268 netlink_kernel_create(int unit, unsigned int groups, 1269 void (*input)(struct sock *sk, int len), 1270 struct module *module) 1271 { 1272 struct socket *sock; 1273 struct sock *sk; 1274 struct netlink_sock *nlk; 1275 unsigned long *listeners = NULL; 1276 1277 if (!nl_table) 1278 return NULL; 1279 1280 if (unit<0 || unit>=MAX_LINKS) 1281 return NULL; 1282 1283 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 1284 return NULL; 1285 1286 if (__netlink_create(sock, unit) < 0) 1287 goto out_sock_release; 1288 1289 if (groups < 32) 1290 groups = 32; 1291 1292 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL); 1293 if (!listeners) 1294 goto out_sock_release; 1295 1296 sk = sock->sk; 1297 sk->sk_data_ready = netlink_data_ready; 1298 if (input) 1299 nlk_sk(sk)->data_ready = input; 1300 1301 if (netlink_insert(sk, 0)) 1302 goto out_sock_release; 1303 1304 nlk = nlk_sk(sk); 1305 nlk->flags |= NETLINK_KERNEL_SOCKET; 1306 1307 netlink_table_grab(); 1308 nl_table[unit].groups = groups; 1309 nl_table[unit].listeners = listeners; 1310 nl_table[unit].module = module; 1311 nl_table[unit].registered = 1; 1312 netlink_table_ungrab(); 1313 1314 return sk; 1315 1316 out_sock_release: 1317 kfree(listeners); 1318 sock_release(sock); 1319 return NULL; 1320 } 1321 1322 void netlink_set_nonroot(int protocol, unsigned int flags) 1323 { 1324 if ((unsigned int)protocol < MAX_LINKS) 1325 nl_table[protocol].nl_nonroot = flags; 1326 } 1327 1328 static void netlink_destroy_callback(struct netlink_callback *cb) 1329 { 1330 if (cb->skb) 1331 kfree_skb(cb->skb); 1332 kfree(cb); 1333 } 1334 1335 /* 1336 * It looks a bit ugly. 1337 * It would be better to create kernel thread. 1338 */ 1339 1340 static int netlink_dump(struct sock *sk) 1341 { 1342 struct netlink_sock *nlk = nlk_sk(sk); 1343 struct netlink_callback *cb; 1344 struct sk_buff *skb; 1345 struct nlmsghdr *nlh; 1346 int len; 1347 1348 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL); 1349 if (!skb) 1350 return -ENOBUFS; 1351 1352 spin_lock(&nlk->cb_lock); 1353 1354 cb = nlk->cb; 1355 if (cb == NULL) { 1356 spin_unlock(&nlk->cb_lock); 1357 kfree_skb(skb); 1358 return -EINVAL; 1359 } 1360 1361 len = cb->dump(skb, cb); 1362 1363 if (len > 0) { 1364 spin_unlock(&nlk->cb_lock); 1365 skb_queue_tail(&sk->sk_receive_queue, skb); 1366 sk->sk_data_ready(sk, len); 1367 return 0; 1368 } 1369 1370 nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI); 1371 memcpy(NLMSG_DATA(nlh), &len, sizeof(len)); 1372 skb_queue_tail(&sk->sk_receive_queue, skb); 1373 sk->sk_data_ready(sk, skb->len); 1374 1375 if (cb->done) 1376 cb->done(cb); 1377 nlk->cb = NULL; 1378 spin_unlock(&nlk->cb_lock); 1379 1380 netlink_destroy_callback(cb); 1381 return 0; 1382 1383 nlmsg_failure: 1384 return -ENOBUFS; 1385 } 1386 1387 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 1388 struct nlmsghdr *nlh, 1389 int (*dump)(struct sk_buff *skb, struct netlink_callback*), 1390 int (*done)(struct netlink_callback*)) 1391 { 1392 struct netlink_callback *cb; 1393 struct sock *sk; 1394 struct netlink_sock *nlk; 1395 1396 cb = kmalloc(sizeof(*cb), GFP_KERNEL); 1397 if (cb == NULL) 1398 return -ENOBUFS; 1399 1400 memset(cb, 0, sizeof(*cb)); 1401 cb->dump = dump; 1402 cb->done = done; 1403 cb->nlh = nlh; 1404 atomic_inc(&skb->users); 1405 cb->skb = skb; 1406 1407 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid); 1408 if (sk == NULL) { 1409 netlink_destroy_callback(cb); 1410 return -ECONNREFUSED; 1411 } 1412 nlk = nlk_sk(sk); 1413 /* A dump is in progress... */ 1414 spin_lock(&nlk->cb_lock); 1415 if (nlk->cb) { 1416 spin_unlock(&nlk->cb_lock); 1417 netlink_destroy_callback(cb); 1418 sock_put(sk); 1419 return -EBUSY; 1420 } 1421 nlk->cb = cb; 1422 spin_unlock(&nlk->cb_lock); 1423 1424 netlink_dump(sk); 1425 sock_put(sk); 1426 return 0; 1427 } 1428 1429 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err) 1430 { 1431 struct sk_buff *skb; 1432 struct nlmsghdr *rep; 1433 struct nlmsgerr *errmsg; 1434 int size; 1435 1436 if (err == 0) 1437 size = NLMSG_SPACE(sizeof(struct nlmsgerr)); 1438 else 1439 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len)); 1440 1441 skb = alloc_skb(size, GFP_KERNEL); 1442 if (!skb) { 1443 struct sock *sk; 1444 1445 sk = netlink_lookup(in_skb->sk->sk_protocol, 1446 NETLINK_CB(in_skb).pid); 1447 if (sk) { 1448 sk->sk_err = ENOBUFS; 1449 sk->sk_error_report(sk); 1450 sock_put(sk); 1451 } 1452 return; 1453 } 1454 1455 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, 1456 NLMSG_ERROR, sizeof(struct nlmsgerr), 0); 1457 errmsg = NLMSG_DATA(rep); 1458 errmsg->error = err; 1459 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr)); 1460 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT); 1461 } 1462 1463 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *, 1464 struct nlmsghdr *, int *)) 1465 { 1466 unsigned int total_len; 1467 struct nlmsghdr *nlh; 1468 int err; 1469 1470 while (skb->len >= nlmsg_total_size(0)) { 1471 nlh = (struct nlmsghdr *) skb->data; 1472 1473 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len) 1474 return 0; 1475 1476 total_len = min(NLMSG_ALIGN(nlh->nlmsg_len), skb->len); 1477 1478 if (cb(skb, nlh, &err) < 0) { 1479 /* Not an error, but we have to interrupt processing 1480 * here. Note: that in this case we do not pull 1481 * message from skb, it will be processed later. 1482 */ 1483 if (err == 0) 1484 return -1; 1485 netlink_ack(skb, nlh, err); 1486 } else if (nlh->nlmsg_flags & NLM_F_ACK) 1487 netlink_ack(skb, nlh, 0); 1488 1489 skb_pull(skb, total_len); 1490 } 1491 1492 return 0; 1493 } 1494 1495 /** 1496 * nelink_run_queue - Process netlink receive queue. 1497 * @sk: Netlink socket containing the queue 1498 * @qlen: Place to store queue length upon entry 1499 * @cb: Callback function invoked for each netlink message found 1500 * 1501 * Processes as much as there was in the queue upon entry and invokes 1502 * a callback function for each netlink message found. The callback 1503 * function may refuse a message by returning a negative error code 1504 * but setting the error pointer to 0 in which case this function 1505 * returns with a qlen != 0. 1506 * 1507 * qlen must be initialized to 0 before the initial entry, afterwards 1508 * the function may be called repeatedly until qlen reaches 0. 1509 */ 1510 void netlink_run_queue(struct sock *sk, unsigned int *qlen, 1511 int (*cb)(struct sk_buff *, struct nlmsghdr *, int *)) 1512 { 1513 struct sk_buff *skb; 1514 1515 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue)) 1516 *qlen = skb_queue_len(&sk->sk_receive_queue); 1517 1518 for (; *qlen; (*qlen)--) { 1519 skb = skb_dequeue(&sk->sk_receive_queue); 1520 if (netlink_rcv_skb(skb, cb)) { 1521 if (skb->len) 1522 skb_queue_head(&sk->sk_receive_queue, skb); 1523 else { 1524 kfree_skb(skb); 1525 (*qlen)--; 1526 } 1527 break; 1528 } 1529 1530 kfree_skb(skb); 1531 } 1532 } 1533 1534 /** 1535 * netlink_queue_skip - Skip netlink message while processing queue. 1536 * @nlh: Netlink message to be skipped 1537 * @skb: Socket buffer containing the netlink messages. 1538 * 1539 * Pulls the given netlink message off the socket buffer so the next 1540 * call to netlink_queue_run() will not reconsider the message. 1541 */ 1542 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb) 1543 { 1544 int msglen = NLMSG_ALIGN(nlh->nlmsg_len); 1545 1546 if (msglen > skb->len) 1547 msglen = skb->len; 1548 1549 skb_pull(skb, msglen); 1550 } 1551 1552 #ifdef CONFIG_PROC_FS 1553 struct nl_seq_iter { 1554 int link; 1555 int hash_idx; 1556 }; 1557 1558 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos) 1559 { 1560 struct nl_seq_iter *iter = seq->private; 1561 int i, j; 1562 struct sock *s; 1563 struct hlist_node *node; 1564 loff_t off = 0; 1565 1566 for (i=0; i<MAX_LINKS; i++) { 1567 struct nl_pid_hash *hash = &nl_table[i].hash; 1568 1569 for (j = 0; j <= hash->mask; j++) { 1570 sk_for_each(s, node, &hash->table[j]) { 1571 if (off == pos) { 1572 iter->link = i; 1573 iter->hash_idx = j; 1574 return s; 1575 } 1576 ++off; 1577 } 1578 } 1579 } 1580 return NULL; 1581 } 1582 1583 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos) 1584 { 1585 read_lock(&nl_table_lock); 1586 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN; 1587 } 1588 1589 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1590 { 1591 struct sock *s; 1592 struct nl_seq_iter *iter; 1593 int i, j; 1594 1595 ++*pos; 1596 1597 if (v == SEQ_START_TOKEN) 1598 return netlink_seq_socket_idx(seq, 0); 1599 1600 s = sk_next(v); 1601 if (s) 1602 return s; 1603 1604 iter = seq->private; 1605 i = iter->link; 1606 j = iter->hash_idx + 1; 1607 1608 do { 1609 struct nl_pid_hash *hash = &nl_table[i].hash; 1610 1611 for (; j <= hash->mask; j++) { 1612 s = sk_head(&hash->table[j]); 1613 if (s) { 1614 iter->link = i; 1615 iter->hash_idx = j; 1616 return s; 1617 } 1618 } 1619 1620 j = 0; 1621 } while (++i < MAX_LINKS); 1622 1623 return NULL; 1624 } 1625 1626 static void netlink_seq_stop(struct seq_file *seq, void *v) 1627 { 1628 read_unlock(&nl_table_lock); 1629 } 1630 1631 1632 static int netlink_seq_show(struct seq_file *seq, void *v) 1633 { 1634 if (v == SEQ_START_TOKEN) 1635 seq_puts(seq, 1636 "sk Eth Pid Groups " 1637 "Rmem Wmem Dump Locks\n"); 1638 else { 1639 struct sock *s = v; 1640 struct netlink_sock *nlk = nlk_sk(s); 1641 1642 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n", 1643 s, 1644 s->sk_protocol, 1645 nlk->pid, 1646 nlk->groups ? (u32)nlk->groups[0] : 0, 1647 atomic_read(&s->sk_rmem_alloc), 1648 atomic_read(&s->sk_wmem_alloc), 1649 nlk->cb, 1650 atomic_read(&s->sk_refcnt) 1651 ); 1652 1653 } 1654 return 0; 1655 } 1656 1657 static struct seq_operations netlink_seq_ops = { 1658 .start = netlink_seq_start, 1659 .next = netlink_seq_next, 1660 .stop = netlink_seq_stop, 1661 .show = netlink_seq_show, 1662 }; 1663 1664 1665 static int netlink_seq_open(struct inode *inode, struct file *file) 1666 { 1667 struct seq_file *seq; 1668 struct nl_seq_iter *iter; 1669 int err; 1670 1671 iter = kmalloc(sizeof(*iter), GFP_KERNEL); 1672 if (!iter) 1673 return -ENOMEM; 1674 1675 err = seq_open(file, &netlink_seq_ops); 1676 if (err) { 1677 kfree(iter); 1678 return err; 1679 } 1680 1681 memset(iter, 0, sizeof(*iter)); 1682 seq = file->private_data; 1683 seq->private = iter; 1684 return 0; 1685 } 1686 1687 static struct file_operations netlink_seq_fops = { 1688 .owner = THIS_MODULE, 1689 .open = netlink_seq_open, 1690 .read = seq_read, 1691 .llseek = seq_lseek, 1692 .release = seq_release_private, 1693 }; 1694 1695 #endif 1696 1697 int netlink_register_notifier(struct notifier_block *nb) 1698 { 1699 return atomic_notifier_chain_register(&netlink_chain, nb); 1700 } 1701 1702 int netlink_unregister_notifier(struct notifier_block *nb) 1703 { 1704 return atomic_notifier_chain_unregister(&netlink_chain, nb); 1705 } 1706 1707 static const struct proto_ops netlink_ops = { 1708 .family = PF_NETLINK, 1709 .owner = THIS_MODULE, 1710 .release = netlink_release, 1711 .bind = netlink_bind, 1712 .connect = netlink_connect, 1713 .socketpair = sock_no_socketpair, 1714 .accept = sock_no_accept, 1715 .getname = netlink_getname, 1716 .poll = datagram_poll, 1717 .ioctl = sock_no_ioctl, 1718 .listen = sock_no_listen, 1719 .shutdown = sock_no_shutdown, 1720 .setsockopt = netlink_setsockopt, 1721 .getsockopt = netlink_getsockopt, 1722 .sendmsg = netlink_sendmsg, 1723 .recvmsg = netlink_recvmsg, 1724 .mmap = sock_no_mmap, 1725 .sendpage = sock_no_sendpage, 1726 }; 1727 1728 static struct net_proto_family netlink_family_ops = { 1729 .family = PF_NETLINK, 1730 .create = netlink_create, 1731 .owner = THIS_MODULE, /* for consistency 8) */ 1732 }; 1733 1734 extern void netlink_skb_parms_too_large(void); 1735 1736 static int __init netlink_proto_init(void) 1737 { 1738 struct sk_buff *dummy_skb; 1739 int i; 1740 unsigned long max; 1741 unsigned int order; 1742 int err = proto_register(&netlink_proto, 0); 1743 1744 if (err != 0) 1745 goto out; 1746 1747 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb)) 1748 netlink_skb_parms_too_large(); 1749 1750 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL); 1751 if (!nl_table) { 1752 enomem: 1753 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n"); 1754 return -ENOMEM; 1755 } 1756 1757 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS); 1758 1759 if (num_physpages >= (128 * 1024)) 1760 max = num_physpages >> (21 - PAGE_SHIFT); 1761 else 1762 max = num_physpages >> (23 - PAGE_SHIFT); 1763 1764 order = get_bitmask_order(max) - 1 + PAGE_SHIFT; 1765 max = (1UL << order) / sizeof(struct hlist_head); 1766 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1; 1767 1768 for (i = 0; i < MAX_LINKS; i++) { 1769 struct nl_pid_hash *hash = &nl_table[i].hash; 1770 1771 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table)); 1772 if (!hash->table) { 1773 while (i-- > 0) 1774 nl_pid_hash_free(nl_table[i].hash.table, 1775 1 * sizeof(*hash->table)); 1776 kfree(nl_table); 1777 goto enomem; 1778 } 1779 memset(hash->table, 0, 1 * sizeof(*hash->table)); 1780 hash->max_shift = order; 1781 hash->shift = 0; 1782 hash->mask = 0; 1783 hash->rehash_time = jiffies; 1784 } 1785 1786 sock_register(&netlink_family_ops); 1787 #ifdef CONFIG_PROC_FS 1788 proc_net_fops_create("netlink", 0, &netlink_seq_fops); 1789 #endif 1790 /* The netlink device handler may be needed early. */ 1791 rtnetlink_init(); 1792 out: 1793 return err; 1794 } 1795 1796 core_initcall(netlink_proto_init); 1797 1798 EXPORT_SYMBOL(netlink_ack); 1799 EXPORT_SYMBOL(netlink_run_queue); 1800 EXPORT_SYMBOL(netlink_queue_skip); 1801 EXPORT_SYMBOL(netlink_broadcast); 1802 EXPORT_SYMBOL(netlink_dump_start); 1803 EXPORT_SYMBOL(netlink_kernel_create); 1804 EXPORT_SYMBOL(netlink_register_notifier); 1805 EXPORT_SYMBOL(netlink_set_err); 1806 EXPORT_SYMBOL(netlink_set_nonroot); 1807 EXPORT_SYMBOL(netlink_unicast); 1808 EXPORT_SYMBOL(netlink_unregister_notifier); 1809 1810