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 * 17 */ 18 19 #include <linux/config.h> 20 #include <linux/module.h> 21 22 #include <linux/kernel.h> 23 #include <linux/init.h> 24 #include <linux/signal.h> 25 #include <linux/sched.h> 26 #include <linux/errno.h> 27 #include <linux/string.h> 28 #include <linux/stat.h> 29 #include <linux/socket.h> 30 #include <linux/un.h> 31 #include <linux/fcntl.h> 32 #include <linux/termios.h> 33 #include <linux/sockios.h> 34 #include <linux/net.h> 35 #include <linux/fs.h> 36 #include <linux/slab.h> 37 #include <asm/uaccess.h> 38 #include <linux/skbuff.h> 39 #include <linux/netdevice.h> 40 #include <linux/rtnetlink.h> 41 #include <linux/proc_fs.h> 42 #include <linux/seq_file.h> 43 #include <linux/smp_lock.h> 44 #include <linux/notifier.h> 45 #include <linux/security.h> 46 #include <linux/jhash.h> 47 #include <linux/jiffies.h> 48 #include <linux/random.h> 49 #include <linux/bitops.h> 50 #include <linux/mm.h> 51 #include <linux/types.h> 52 #include <net/sock.h> 53 #include <net/scm.h> 54 55 #define Nprintk(a...) 56 57 struct netlink_sock { 58 /* struct sock has to be the first member of netlink_sock */ 59 struct sock sk; 60 u32 pid; 61 unsigned int groups; 62 u32 dst_pid; 63 unsigned int dst_groups; 64 unsigned long state; 65 wait_queue_head_t wait; 66 struct netlink_callback *cb; 67 spinlock_t cb_lock; 68 void (*data_ready)(struct sock *sk, int bytes); 69 }; 70 71 static inline struct netlink_sock *nlk_sk(struct sock *sk) 72 { 73 return (struct netlink_sock *)sk; 74 } 75 76 struct nl_pid_hash { 77 struct hlist_head *table; 78 unsigned long rehash_time; 79 80 unsigned int mask; 81 unsigned int shift; 82 83 unsigned int entries; 84 unsigned int max_shift; 85 86 u32 rnd; 87 }; 88 89 struct netlink_table { 90 struct nl_pid_hash hash; 91 struct hlist_head mc_list; 92 unsigned int nl_nonroot; 93 }; 94 95 static struct netlink_table *nl_table; 96 97 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait); 98 99 static int netlink_dump(struct sock *sk); 100 static void netlink_destroy_callback(struct netlink_callback *cb); 101 102 static DEFINE_RWLOCK(nl_table_lock); 103 static atomic_t nl_table_users = ATOMIC_INIT(0); 104 105 static struct notifier_block *netlink_chain; 106 107 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid) 108 { 109 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask]; 110 } 111 112 static void netlink_sock_destruct(struct sock *sk) 113 { 114 skb_queue_purge(&sk->sk_receive_queue); 115 116 if (!sock_flag(sk, SOCK_DEAD)) { 117 printk("Freeing alive netlink socket %p\n", sk); 118 return; 119 } 120 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc)); 121 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc)); 122 BUG_TRAP(!nlk_sk(sk)->cb); 123 } 124 125 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP. 126 * Look, when several writers sleep and reader wakes them up, all but one 127 * immediately hit write lock and grab all the cpus. Exclusive sleep solves 128 * this, _but_ remember, it adds useless work on UP machines. 129 */ 130 131 static void netlink_table_grab(void) 132 { 133 write_lock_bh(&nl_table_lock); 134 135 if (atomic_read(&nl_table_users)) { 136 DECLARE_WAITQUEUE(wait, current); 137 138 add_wait_queue_exclusive(&nl_table_wait, &wait); 139 for(;;) { 140 set_current_state(TASK_UNINTERRUPTIBLE); 141 if (atomic_read(&nl_table_users) == 0) 142 break; 143 write_unlock_bh(&nl_table_lock); 144 schedule(); 145 write_lock_bh(&nl_table_lock); 146 } 147 148 __set_current_state(TASK_RUNNING); 149 remove_wait_queue(&nl_table_wait, &wait); 150 } 151 } 152 153 static __inline__ void netlink_table_ungrab(void) 154 { 155 write_unlock_bh(&nl_table_lock); 156 wake_up(&nl_table_wait); 157 } 158 159 static __inline__ void 160 netlink_lock_table(void) 161 { 162 /* read_lock() synchronizes us to netlink_table_grab */ 163 164 read_lock(&nl_table_lock); 165 atomic_inc(&nl_table_users); 166 read_unlock(&nl_table_lock); 167 } 168 169 static __inline__ void 170 netlink_unlock_table(void) 171 { 172 if (atomic_dec_and_test(&nl_table_users)) 173 wake_up(&nl_table_wait); 174 } 175 176 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid) 177 { 178 struct nl_pid_hash *hash = &nl_table[protocol].hash; 179 struct hlist_head *head; 180 struct sock *sk; 181 struct hlist_node *node; 182 183 read_lock(&nl_table_lock); 184 head = nl_pid_hashfn(hash, pid); 185 sk_for_each(sk, node, head) { 186 if (nlk_sk(sk)->pid == pid) { 187 sock_hold(sk); 188 goto found; 189 } 190 } 191 sk = NULL; 192 found: 193 read_unlock(&nl_table_lock); 194 return sk; 195 } 196 197 static inline struct hlist_head *nl_pid_hash_alloc(size_t size) 198 { 199 if (size <= PAGE_SIZE) 200 return kmalloc(size, GFP_ATOMIC); 201 else 202 return (struct hlist_head *) 203 __get_free_pages(GFP_ATOMIC, get_order(size)); 204 } 205 206 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size) 207 { 208 if (size <= PAGE_SIZE) 209 kfree(table); 210 else 211 free_pages((unsigned long)table, get_order(size)); 212 } 213 214 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow) 215 { 216 unsigned int omask, mask, shift; 217 size_t osize, size; 218 struct hlist_head *otable, *table; 219 int i; 220 221 omask = mask = hash->mask; 222 osize = size = (mask + 1) * sizeof(*table); 223 shift = hash->shift; 224 225 if (grow) { 226 if (++shift > hash->max_shift) 227 return 0; 228 mask = mask * 2 + 1; 229 size *= 2; 230 } 231 232 table = nl_pid_hash_alloc(size); 233 if (!table) 234 return 0; 235 236 memset(table, 0, size); 237 otable = hash->table; 238 hash->table = table; 239 hash->mask = mask; 240 hash->shift = shift; 241 get_random_bytes(&hash->rnd, sizeof(hash->rnd)); 242 243 for (i = 0; i <= omask; i++) { 244 struct sock *sk; 245 struct hlist_node *node, *tmp; 246 247 sk_for_each_safe(sk, node, tmp, &otable[i]) 248 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid)); 249 } 250 251 nl_pid_hash_free(otable, osize); 252 hash->rehash_time = jiffies + 10 * 60 * HZ; 253 return 1; 254 } 255 256 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len) 257 { 258 int avg = hash->entries >> hash->shift; 259 260 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1)) 261 return 1; 262 263 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) { 264 nl_pid_hash_rehash(hash, 0); 265 return 1; 266 } 267 268 return 0; 269 } 270 271 static struct proto_ops netlink_ops; 272 273 static int netlink_insert(struct sock *sk, u32 pid) 274 { 275 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 276 struct hlist_head *head; 277 int err = -EADDRINUSE; 278 struct sock *osk; 279 struct hlist_node *node; 280 int len; 281 282 netlink_table_grab(); 283 head = nl_pid_hashfn(hash, pid); 284 len = 0; 285 sk_for_each(osk, node, head) { 286 if (nlk_sk(osk)->pid == pid) 287 break; 288 len++; 289 } 290 if (node) 291 goto err; 292 293 err = -EBUSY; 294 if (nlk_sk(sk)->pid) 295 goto err; 296 297 err = -ENOMEM; 298 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX)) 299 goto err; 300 301 if (len && nl_pid_hash_dilute(hash, len)) 302 head = nl_pid_hashfn(hash, pid); 303 hash->entries++; 304 nlk_sk(sk)->pid = pid; 305 sk_add_node(sk, head); 306 err = 0; 307 308 err: 309 netlink_table_ungrab(); 310 return err; 311 } 312 313 static void netlink_remove(struct sock *sk) 314 { 315 netlink_table_grab(); 316 nl_table[sk->sk_protocol].hash.entries--; 317 sk_del_node_init(sk); 318 if (nlk_sk(sk)->groups) 319 __sk_del_bind_node(sk); 320 netlink_table_ungrab(); 321 } 322 323 static struct proto netlink_proto = { 324 .name = "NETLINK", 325 .owner = THIS_MODULE, 326 .obj_size = sizeof(struct netlink_sock), 327 }; 328 329 static int netlink_create(struct socket *sock, int protocol) 330 { 331 struct sock *sk; 332 struct netlink_sock *nlk; 333 334 sock->state = SS_UNCONNECTED; 335 336 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 337 return -ESOCKTNOSUPPORT; 338 339 if (protocol<0 || protocol >= MAX_LINKS) 340 return -EPROTONOSUPPORT; 341 342 sock->ops = &netlink_ops; 343 344 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1); 345 if (!sk) 346 return -ENOMEM; 347 348 sock_init_data(sock, sk); 349 350 nlk = nlk_sk(sk); 351 352 spin_lock_init(&nlk->cb_lock); 353 init_waitqueue_head(&nlk->wait); 354 sk->sk_destruct = netlink_sock_destruct; 355 356 sk->sk_protocol = protocol; 357 return 0; 358 } 359 360 static int netlink_release(struct socket *sock) 361 { 362 struct sock *sk = sock->sk; 363 struct netlink_sock *nlk; 364 365 if (!sk) 366 return 0; 367 368 netlink_remove(sk); 369 nlk = nlk_sk(sk); 370 371 spin_lock(&nlk->cb_lock); 372 if (nlk->cb) { 373 nlk->cb->done(nlk->cb); 374 netlink_destroy_callback(nlk->cb); 375 nlk->cb = NULL; 376 } 377 spin_unlock(&nlk->cb_lock); 378 379 /* OK. Socket is unlinked, and, therefore, 380 no new packets will arrive */ 381 382 sock_orphan(sk); 383 sock->sk = NULL; 384 wake_up_interruptible_all(&nlk->wait); 385 386 skb_queue_purge(&sk->sk_write_queue); 387 388 if (nlk->pid && !nlk->groups) { 389 struct netlink_notify n = { 390 .protocol = sk->sk_protocol, 391 .pid = nlk->pid, 392 }; 393 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n); 394 } 395 396 sock_put(sk); 397 return 0; 398 } 399 400 static int netlink_autobind(struct socket *sock) 401 { 402 struct sock *sk = sock->sk; 403 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 404 struct hlist_head *head; 405 struct sock *osk; 406 struct hlist_node *node; 407 s32 pid = current->pid; 408 int err; 409 static s32 rover = -4097; 410 411 retry: 412 cond_resched(); 413 netlink_table_grab(); 414 head = nl_pid_hashfn(hash, pid); 415 sk_for_each(osk, node, head) { 416 if (nlk_sk(osk)->pid == pid) { 417 /* Bind collision, search negative pid values. */ 418 pid = rover--; 419 if (rover > -4097) 420 rover = -4097; 421 netlink_table_ungrab(); 422 goto retry; 423 } 424 } 425 netlink_table_ungrab(); 426 427 err = netlink_insert(sk, pid); 428 if (err == -EADDRINUSE) 429 goto retry; 430 return 0; 431 } 432 433 static inline int netlink_capable(struct socket *sock, unsigned int flag) 434 { 435 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) || 436 capable(CAP_NET_ADMIN); 437 } 438 439 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 440 { 441 struct sock *sk = sock->sk; 442 struct netlink_sock *nlk = nlk_sk(sk); 443 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 444 int err; 445 446 if (nladdr->nl_family != AF_NETLINK) 447 return -EINVAL; 448 449 /* Only superuser is allowed to listen multicasts */ 450 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV)) 451 return -EPERM; 452 453 if (nlk->pid) { 454 if (nladdr->nl_pid != nlk->pid) 455 return -EINVAL; 456 } else { 457 err = nladdr->nl_pid ? 458 netlink_insert(sk, nladdr->nl_pid) : 459 netlink_autobind(sock); 460 if (err) 461 return err; 462 } 463 464 if (!nladdr->nl_groups && !nlk->groups) 465 return 0; 466 467 netlink_table_grab(); 468 if (nlk->groups && !nladdr->nl_groups) 469 __sk_del_bind_node(sk); 470 else if (!nlk->groups && nladdr->nl_groups) 471 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 472 nlk->groups = nladdr->nl_groups; 473 netlink_table_ungrab(); 474 475 return 0; 476 } 477 478 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 479 int alen, int flags) 480 { 481 int err = 0; 482 struct sock *sk = sock->sk; 483 struct netlink_sock *nlk = nlk_sk(sk); 484 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr; 485 486 if (addr->sa_family == AF_UNSPEC) { 487 sk->sk_state = NETLINK_UNCONNECTED; 488 nlk->dst_pid = 0; 489 nlk->dst_groups = 0; 490 return 0; 491 } 492 if (addr->sa_family != AF_NETLINK) 493 return -EINVAL; 494 495 /* Only superuser is allowed to send multicasts */ 496 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND)) 497 return -EPERM; 498 499 if (!nlk->pid) 500 err = netlink_autobind(sock); 501 502 if (err == 0) { 503 sk->sk_state = NETLINK_CONNECTED; 504 nlk->dst_pid = nladdr->nl_pid; 505 nlk->dst_groups = nladdr->nl_groups; 506 } 507 508 return err; 509 } 510 511 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer) 512 { 513 struct sock *sk = sock->sk; 514 struct netlink_sock *nlk = nlk_sk(sk); 515 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr; 516 517 nladdr->nl_family = AF_NETLINK; 518 nladdr->nl_pad = 0; 519 *addr_len = sizeof(*nladdr); 520 521 if (peer) { 522 nladdr->nl_pid = nlk->dst_pid; 523 nladdr->nl_groups = nlk->dst_groups; 524 } else { 525 nladdr->nl_pid = nlk->pid; 526 nladdr->nl_groups = nlk->groups; 527 } 528 return 0; 529 } 530 531 static void netlink_overrun(struct sock *sk) 532 { 533 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) { 534 sk->sk_err = ENOBUFS; 535 sk->sk_error_report(sk); 536 } 537 } 538 539 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid) 540 { 541 int protocol = ssk->sk_protocol; 542 struct sock *sock; 543 struct netlink_sock *nlk; 544 545 sock = netlink_lookup(protocol, pid); 546 if (!sock) 547 return ERR_PTR(-ECONNREFUSED); 548 549 /* Don't bother queuing skb if kernel socket has no input function */ 550 nlk = nlk_sk(sock); 551 if ((nlk->pid == 0 && !nlk->data_ready) || 552 (sock->sk_state == NETLINK_CONNECTED && 553 nlk->dst_pid != nlk_sk(ssk)->pid)) { 554 sock_put(sock); 555 return ERR_PTR(-ECONNREFUSED); 556 } 557 return sock; 558 } 559 560 struct sock *netlink_getsockbyfilp(struct file *filp) 561 { 562 struct inode *inode = filp->f_dentry->d_inode; 563 struct sock *sock; 564 565 if (!S_ISSOCK(inode->i_mode)) 566 return ERR_PTR(-ENOTSOCK); 567 568 sock = SOCKET_I(inode)->sk; 569 if (sock->sk_family != AF_NETLINK) 570 return ERR_PTR(-EINVAL); 571 572 sock_hold(sock); 573 return sock; 574 } 575 576 /* 577 * Attach a skb to a netlink socket. 578 * The caller must hold a reference to the destination socket. On error, the 579 * reference is dropped. The skb is not send to the destination, just all 580 * all error checks are performed and memory in the queue is reserved. 581 * Return values: 582 * < 0: error. skb freed, reference to sock dropped. 583 * 0: continue 584 * 1: repeat lookup - reference dropped while waiting for socket memory. 585 */ 586 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo) 587 { 588 struct netlink_sock *nlk; 589 590 nlk = nlk_sk(sk); 591 592 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 593 test_bit(0, &nlk->state)) { 594 DECLARE_WAITQUEUE(wait, current); 595 if (!timeo) { 596 if (!nlk->pid) 597 netlink_overrun(sk); 598 sock_put(sk); 599 kfree_skb(skb); 600 return -EAGAIN; 601 } 602 603 __set_current_state(TASK_INTERRUPTIBLE); 604 add_wait_queue(&nlk->wait, &wait); 605 606 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 607 test_bit(0, &nlk->state)) && 608 !sock_flag(sk, SOCK_DEAD)) 609 timeo = schedule_timeout(timeo); 610 611 __set_current_state(TASK_RUNNING); 612 remove_wait_queue(&nlk->wait, &wait); 613 sock_put(sk); 614 615 if (signal_pending(current)) { 616 kfree_skb(skb); 617 return sock_intr_errno(timeo); 618 } 619 return 1; 620 } 621 skb_set_owner_r(skb, sk); 622 return 0; 623 } 624 625 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol) 626 { 627 struct netlink_sock *nlk; 628 int len = skb->len; 629 630 nlk = nlk_sk(sk); 631 632 skb_queue_tail(&sk->sk_receive_queue, skb); 633 sk->sk_data_ready(sk, len); 634 sock_put(sk); 635 return len; 636 } 637 638 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 639 { 640 kfree_skb(skb); 641 sock_put(sk); 642 } 643 644 static inline struct sk_buff *netlink_trim(struct sk_buff *skb, int allocation) 645 { 646 int delta; 647 648 skb_orphan(skb); 649 650 delta = skb->end - skb->tail; 651 if (delta * 2 < skb->truesize) 652 return skb; 653 654 if (skb_shared(skb)) { 655 struct sk_buff *nskb = skb_clone(skb, allocation); 656 if (!nskb) 657 return skb; 658 kfree_skb(skb); 659 skb = nskb; 660 } 661 662 if (!pskb_expand_head(skb, 0, -delta, allocation)) 663 skb->truesize -= delta; 664 665 return skb; 666 } 667 668 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock) 669 { 670 struct sock *sk; 671 int err; 672 long timeo; 673 674 skb = netlink_trim(skb, gfp_any()); 675 676 timeo = sock_sndtimeo(ssk, nonblock); 677 retry: 678 sk = netlink_getsockbypid(ssk, pid); 679 if (IS_ERR(sk)) { 680 kfree_skb(skb); 681 return PTR_ERR(sk); 682 } 683 err = netlink_attachskb(sk, skb, nonblock, timeo); 684 if (err == 1) 685 goto retry; 686 if (err) 687 return err; 688 689 return netlink_sendskb(sk, skb, ssk->sk_protocol); 690 } 691 692 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 693 { 694 struct netlink_sock *nlk = nlk_sk(sk); 695 696 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 697 !test_bit(0, &nlk->state)) { 698 skb_set_owner_r(skb, sk); 699 skb_queue_tail(&sk->sk_receive_queue, skb); 700 sk->sk_data_ready(sk, skb->len); 701 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf; 702 } 703 return -1; 704 } 705 706 struct netlink_broadcast_data { 707 struct sock *exclude_sk; 708 u32 pid; 709 u32 group; 710 int failure; 711 int congested; 712 int delivered; 713 int allocation; 714 struct sk_buff *skb, *skb2; 715 }; 716 717 static inline int do_one_broadcast(struct sock *sk, 718 struct netlink_broadcast_data *p) 719 { 720 struct netlink_sock *nlk = nlk_sk(sk); 721 int val; 722 723 if (p->exclude_sk == sk) 724 goto out; 725 726 if (nlk->pid == p->pid || !(nlk->groups & p->group)) 727 goto out; 728 729 if (p->failure) { 730 netlink_overrun(sk); 731 goto out; 732 } 733 734 sock_hold(sk); 735 if (p->skb2 == NULL) { 736 if (atomic_read(&p->skb->users) != 1) { 737 p->skb2 = skb_clone(p->skb, p->allocation); 738 } else { 739 p->skb2 = p->skb; 740 atomic_inc(&p->skb->users); 741 } 742 } 743 if (p->skb2 == NULL) { 744 netlink_overrun(sk); 745 /* Clone failed. Notify ALL listeners. */ 746 p->failure = 1; 747 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) { 748 netlink_overrun(sk); 749 } else { 750 p->congested |= val; 751 p->delivered = 1; 752 p->skb2 = NULL; 753 } 754 sock_put(sk); 755 756 out: 757 return 0; 758 } 759 760 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid, 761 u32 group, int allocation) 762 { 763 struct netlink_broadcast_data info; 764 struct hlist_node *node; 765 struct sock *sk; 766 767 skb = netlink_trim(skb, allocation); 768 769 info.exclude_sk = ssk; 770 info.pid = pid; 771 info.group = group; 772 info.failure = 0; 773 info.congested = 0; 774 info.delivered = 0; 775 info.allocation = allocation; 776 info.skb = skb; 777 info.skb2 = NULL; 778 779 /* While we sleep in clone, do not allow to change socket list */ 780 781 netlink_lock_table(); 782 783 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 784 do_one_broadcast(sk, &info); 785 786 netlink_unlock_table(); 787 788 if (info.skb2) 789 kfree_skb(info.skb2); 790 kfree_skb(skb); 791 792 if (info.delivered) { 793 if (info.congested && (allocation & __GFP_WAIT)) 794 yield(); 795 return 0; 796 } 797 if (info.failure) 798 return -ENOBUFS; 799 return -ESRCH; 800 } 801 802 struct netlink_set_err_data { 803 struct sock *exclude_sk; 804 u32 pid; 805 u32 group; 806 int code; 807 }; 808 809 static inline int do_one_set_err(struct sock *sk, 810 struct netlink_set_err_data *p) 811 { 812 struct netlink_sock *nlk = nlk_sk(sk); 813 814 if (sk == p->exclude_sk) 815 goto out; 816 817 if (nlk->pid == p->pid || !(nlk->groups & p->group)) 818 goto out; 819 820 sk->sk_err = p->code; 821 sk->sk_error_report(sk); 822 out: 823 return 0; 824 } 825 826 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code) 827 { 828 struct netlink_set_err_data info; 829 struct hlist_node *node; 830 struct sock *sk; 831 832 info.exclude_sk = ssk; 833 info.pid = pid; 834 info.group = group; 835 info.code = code; 836 837 read_lock(&nl_table_lock); 838 839 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 840 do_one_set_err(sk, &info); 841 842 read_unlock(&nl_table_lock); 843 } 844 845 static inline void netlink_rcv_wake(struct sock *sk) 846 { 847 struct netlink_sock *nlk = nlk_sk(sk); 848 849 if (!skb_queue_len(&sk->sk_receive_queue)) 850 clear_bit(0, &nlk->state); 851 if (!test_bit(0, &nlk->state)) 852 wake_up_interruptible(&nlk->wait); 853 } 854 855 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock, 856 struct msghdr *msg, size_t len) 857 { 858 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 859 struct sock *sk = sock->sk; 860 struct netlink_sock *nlk = nlk_sk(sk); 861 struct sockaddr_nl *addr=msg->msg_name; 862 u32 dst_pid; 863 u32 dst_groups; 864 struct sk_buff *skb; 865 int err; 866 struct scm_cookie scm; 867 868 if (msg->msg_flags&MSG_OOB) 869 return -EOPNOTSUPP; 870 871 if (NULL == siocb->scm) 872 siocb->scm = &scm; 873 err = scm_send(sock, msg, siocb->scm); 874 if (err < 0) 875 return err; 876 877 if (msg->msg_namelen) { 878 if (addr->nl_family != AF_NETLINK) 879 return -EINVAL; 880 dst_pid = addr->nl_pid; 881 dst_groups = addr->nl_groups; 882 if (dst_groups && !netlink_capable(sock, NL_NONROOT_SEND)) 883 return -EPERM; 884 } else { 885 dst_pid = nlk->dst_pid; 886 dst_groups = nlk->dst_groups; 887 } 888 889 if (!nlk->pid) { 890 err = netlink_autobind(sock); 891 if (err) 892 goto out; 893 } 894 895 err = -EMSGSIZE; 896 if (len > sk->sk_sndbuf - 32) 897 goto out; 898 err = -ENOBUFS; 899 skb = alloc_skb(len, GFP_KERNEL); 900 if (skb==NULL) 901 goto out; 902 903 NETLINK_CB(skb).pid = nlk->pid; 904 NETLINK_CB(skb).groups = nlk->groups; 905 NETLINK_CB(skb).dst_pid = dst_pid; 906 NETLINK_CB(skb).dst_groups = dst_groups; 907 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); 908 909 /* What can I do? Netlink is asynchronous, so that 910 we will have to save current capabilities to 911 check them, when this message will be delivered 912 to corresponding kernel module. --ANK (980802) 913 */ 914 915 err = -EFAULT; 916 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) { 917 kfree_skb(skb); 918 goto out; 919 } 920 921 err = security_netlink_send(sk, skb); 922 if (err) { 923 kfree_skb(skb); 924 goto out; 925 } 926 927 if (dst_groups) { 928 atomic_inc(&skb->users); 929 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL); 930 } 931 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT); 932 933 out: 934 return err; 935 } 936 937 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock, 938 struct msghdr *msg, size_t len, 939 int flags) 940 { 941 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 942 struct scm_cookie scm; 943 struct sock *sk = sock->sk; 944 struct netlink_sock *nlk = nlk_sk(sk); 945 int noblock = flags&MSG_DONTWAIT; 946 size_t copied; 947 struct sk_buff *skb; 948 int err; 949 950 if (flags&MSG_OOB) 951 return -EOPNOTSUPP; 952 953 copied = 0; 954 955 skb = skb_recv_datagram(sk,flags,noblock,&err); 956 if (skb==NULL) 957 goto out; 958 959 msg->msg_namelen = 0; 960 961 copied = skb->len; 962 if (len < copied) { 963 msg->msg_flags |= MSG_TRUNC; 964 copied = len; 965 } 966 967 skb->h.raw = skb->data; 968 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 969 970 if (msg->msg_name) { 971 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name; 972 addr->nl_family = AF_NETLINK; 973 addr->nl_pad = 0; 974 addr->nl_pid = NETLINK_CB(skb).pid; 975 addr->nl_groups = NETLINK_CB(skb).dst_groups; 976 msg->msg_namelen = sizeof(*addr); 977 } 978 979 if (NULL == siocb->scm) { 980 memset(&scm, 0, sizeof(scm)); 981 siocb->scm = &scm; 982 } 983 siocb->scm->creds = *NETLINK_CREDS(skb); 984 skb_free_datagram(sk, skb); 985 986 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) 987 netlink_dump(sk); 988 989 scm_recv(sock, msg, siocb->scm, flags); 990 991 out: 992 netlink_rcv_wake(sk); 993 return err ? : copied; 994 } 995 996 static void netlink_data_ready(struct sock *sk, int len) 997 { 998 struct netlink_sock *nlk = nlk_sk(sk); 999 1000 if (nlk->data_ready) 1001 nlk->data_ready(sk, len); 1002 netlink_rcv_wake(sk); 1003 } 1004 1005 /* 1006 * We export these functions to other modules. They provide a 1007 * complete set of kernel non-blocking support for message 1008 * queueing. 1009 */ 1010 1011 struct sock * 1012 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len)) 1013 { 1014 struct socket *sock; 1015 struct sock *sk; 1016 1017 if (!nl_table) 1018 return NULL; 1019 1020 if (unit<0 || unit>=MAX_LINKS) 1021 return NULL; 1022 1023 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 1024 return NULL; 1025 1026 if (netlink_create(sock, unit) < 0) { 1027 sock_release(sock); 1028 return NULL; 1029 } 1030 sk = sock->sk; 1031 sk->sk_data_ready = netlink_data_ready; 1032 if (input) 1033 nlk_sk(sk)->data_ready = input; 1034 1035 if (netlink_insert(sk, 0)) { 1036 sock_release(sock); 1037 return NULL; 1038 } 1039 return sk; 1040 } 1041 1042 void netlink_set_nonroot(int protocol, unsigned int flags) 1043 { 1044 if ((unsigned int)protocol < MAX_LINKS) 1045 nl_table[protocol].nl_nonroot = flags; 1046 } 1047 1048 static void netlink_destroy_callback(struct netlink_callback *cb) 1049 { 1050 if (cb->skb) 1051 kfree_skb(cb->skb); 1052 kfree(cb); 1053 } 1054 1055 /* 1056 * It looks a bit ugly. 1057 * It would be better to create kernel thread. 1058 */ 1059 1060 static int netlink_dump(struct sock *sk) 1061 { 1062 struct netlink_sock *nlk = nlk_sk(sk); 1063 struct netlink_callback *cb; 1064 struct sk_buff *skb; 1065 struct nlmsghdr *nlh; 1066 int len; 1067 1068 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL); 1069 if (!skb) 1070 return -ENOBUFS; 1071 1072 spin_lock(&nlk->cb_lock); 1073 1074 cb = nlk->cb; 1075 if (cb == NULL) { 1076 spin_unlock(&nlk->cb_lock); 1077 kfree_skb(skb); 1078 return -EINVAL; 1079 } 1080 1081 len = cb->dump(skb, cb); 1082 1083 if (len > 0) { 1084 spin_unlock(&nlk->cb_lock); 1085 skb_queue_tail(&sk->sk_receive_queue, skb); 1086 sk->sk_data_ready(sk, len); 1087 return 0; 1088 } 1089 1090 nlh = __nlmsg_put(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, NLMSG_DONE, sizeof(int)); 1091 nlh->nlmsg_flags |= NLM_F_MULTI; 1092 memcpy(NLMSG_DATA(nlh), &len, sizeof(len)); 1093 skb_queue_tail(&sk->sk_receive_queue, skb); 1094 sk->sk_data_ready(sk, skb->len); 1095 1096 cb->done(cb); 1097 nlk->cb = NULL; 1098 spin_unlock(&nlk->cb_lock); 1099 1100 netlink_destroy_callback(cb); 1101 return 0; 1102 } 1103 1104 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 1105 struct nlmsghdr *nlh, 1106 int (*dump)(struct sk_buff *skb, struct netlink_callback*), 1107 int (*done)(struct netlink_callback*)) 1108 { 1109 struct netlink_callback *cb; 1110 struct sock *sk; 1111 struct netlink_sock *nlk; 1112 1113 cb = kmalloc(sizeof(*cb), GFP_KERNEL); 1114 if (cb == NULL) 1115 return -ENOBUFS; 1116 1117 memset(cb, 0, sizeof(*cb)); 1118 cb->dump = dump; 1119 cb->done = done; 1120 cb->nlh = nlh; 1121 atomic_inc(&skb->users); 1122 cb->skb = skb; 1123 1124 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid); 1125 if (sk == NULL) { 1126 netlink_destroy_callback(cb); 1127 return -ECONNREFUSED; 1128 } 1129 nlk = nlk_sk(sk); 1130 /* A dump is in progress... */ 1131 spin_lock(&nlk->cb_lock); 1132 if (nlk->cb) { 1133 spin_unlock(&nlk->cb_lock); 1134 netlink_destroy_callback(cb); 1135 sock_put(sk); 1136 return -EBUSY; 1137 } 1138 nlk->cb = cb; 1139 spin_unlock(&nlk->cb_lock); 1140 1141 netlink_dump(sk); 1142 sock_put(sk); 1143 return 0; 1144 } 1145 1146 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err) 1147 { 1148 struct sk_buff *skb; 1149 struct nlmsghdr *rep; 1150 struct nlmsgerr *errmsg; 1151 int size; 1152 1153 if (err == 0) 1154 size = NLMSG_SPACE(sizeof(struct nlmsgerr)); 1155 else 1156 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len)); 1157 1158 skb = alloc_skb(size, GFP_KERNEL); 1159 if (!skb) { 1160 struct sock *sk; 1161 1162 sk = netlink_lookup(in_skb->sk->sk_protocol, 1163 NETLINK_CB(in_skb).pid); 1164 if (sk) { 1165 sk->sk_err = ENOBUFS; 1166 sk->sk_error_report(sk); 1167 sock_put(sk); 1168 } 1169 return; 1170 } 1171 1172 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, 1173 NLMSG_ERROR, sizeof(struct nlmsgerr)); 1174 errmsg = NLMSG_DATA(rep); 1175 errmsg->error = err; 1176 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr)); 1177 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT); 1178 } 1179 1180 1181 #ifdef CONFIG_PROC_FS 1182 struct nl_seq_iter { 1183 int link; 1184 int hash_idx; 1185 }; 1186 1187 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos) 1188 { 1189 struct nl_seq_iter *iter = seq->private; 1190 int i, j; 1191 struct sock *s; 1192 struct hlist_node *node; 1193 loff_t off = 0; 1194 1195 for (i=0; i<MAX_LINKS; i++) { 1196 struct nl_pid_hash *hash = &nl_table[i].hash; 1197 1198 for (j = 0; j <= hash->mask; j++) { 1199 sk_for_each(s, node, &hash->table[j]) { 1200 if (off == pos) { 1201 iter->link = i; 1202 iter->hash_idx = j; 1203 return s; 1204 } 1205 ++off; 1206 } 1207 } 1208 } 1209 return NULL; 1210 } 1211 1212 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos) 1213 { 1214 read_lock(&nl_table_lock); 1215 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN; 1216 } 1217 1218 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1219 { 1220 struct sock *s; 1221 struct nl_seq_iter *iter; 1222 int i, j; 1223 1224 ++*pos; 1225 1226 if (v == SEQ_START_TOKEN) 1227 return netlink_seq_socket_idx(seq, 0); 1228 1229 s = sk_next(v); 1230 if (s) 1231 return s; 1232 1233 iter = seq->private; 1234 i = iter->link; 1235 j = iter->hash_idx + 1; 1236 1237 do { 1238 struct nl_pid_hash *hash = &nl_table[i].hash; 1239 1240 for (; j <= hash->mask; j++) { 1241 s = sk_head(&hash->table[j]); 1242 if (s) { 1243 iter->link = i; 1244 iter->hash_idx = j; 1245 return s; 1246 } 1247 } 1248 1249 j = 0; 1250 } while (++i < MAX_LINKS); 1251 1252 return NULL; 1253 } 1254 1255 static void netlink_seq_stop(struct seq_file *seq, void *v) 1256 { 1257 read_unlock(&nl_table_lock); 1258 } 1259 1260 1261 static int netlink_seq_show(struct seq_file *seq, void *v) 1262 { 1263 if (v == SEQ_START_TOKEN) 1264 seq_puts(seq, 1265 "sk Eth Pid Groups " 1266 "Rmem Wmem Dump Locks\n"); 1267 else { 1268 struct sock *s = v; 1269 struct netlink_sock *nlk = nlk_sk(s); 1270 1271 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n", 1272 s, 1273 s->sk_protocol, 1274 nlk->pid, 1275 nlk->groups, 1276 atomic_read(&s->sk_rmem_alloc), 1277 atomic_read(&s->sk_wmem_alloc), 1278 nlk->cb, 1279 atomic_read(&s->sk_refcnt) 1280 ); 1281 1282 } 1283 return 0; 1284 } 1285 1286 static struct seq_operations netlink_seq_ops = { 1287 .start = netlink_seq_start, 1288 .next = netlink_seq_next, 1289 .stop = netlink_seq_stop, 1290 .show = netlink_seq_show, 1291 }; 1292 1293 1294 static int netlink_seq_open(struct inode *inode, struct file *file) 1295 { 1296 struct seq_file *seq; 1297 struct nl_seq_iter *iter; 1298 int err; 1299 1300 iter = kmalloc(sizeof(*iter), GFP_KERNEL); 1301 if (!iter) 1302 return -ENOMEM; 1303 1304 err = seq_open(file, &netlink_seq_ops); 1305 if (err) { 1306 kfree(iter); 1307 return err; 1308 } 1309 1310 memset(iter, 0, sizeof(*iter)); 1311 seq = file->private_data; 1312 seq->private = iter; 1313 return 0; 1314 } 1315 1316 static struct file_operations netlink_seq_fops = { 1317 .owner = THIS_MODULE, 1318 .open = netlink_seq_open, 1319 .read = seq_read, 1320 .llseek = seq_lseek, 1321 .release = seq_release_private, 1322 }; 1323 1324 #endif 1325 1326 int netlink_register_notifier(struct notifier_block *nb) 1327 { 1328 return notifier_chain_register(&netlink_chain, nb); 1329 } 1330 1331 int netlink_unregister_notifier(struct notifier_block *nb) 1332 { 1333 return notifier_chain_unregister(&netlink_chain, nb); 1334 } 1335 1336 static struct proto_ops netlink_ops = { 1337 .family = PF_NETLINK, 1338 .owner = THIS_MODULE, 1339 .release = netlink_release, 1340 .bind = netlink_bind, 1341 .connect = netlink_connect, 1342 .socketpair = sock_no_socketpair, 1343 .accept = sock_no_accept, 1344 .getname = netlink_getname, 1345 .poll = datagram_poll, 1346 .ioctl = sock_no_ioctl, 1347 .listen = sock_no_listen, 1348 .shutdown = sock_no_shutdown, 1349 .setsockopt = sock_no_setsockopt, 1350 .getsockopt = sock_no_getsockopt, 1351 .sendmsg = netlink_sendmsg, 1352 .recvmsg = netlink_recvmsg, 1353 .mmap = sock_no_mmap, 1354 .sendpage = sock_no_sendpage, 1355 }; 1356 1357 static struct net_proto_family netlink_family_ops = { 1358 .family = PF_NETLINK, 1359 .create = netlink_create, 1360 .owner = THIS_MODULE, /* for consistency 8) */ 1361 }; 1362 1363 extern void netlink_skb_parms_too_large(void); 1364 1365 static int __init netlink_proto_init(void) 1366 { 1367 struct sk_buff *dummy_skb; 1368 int i; 1369 unsigned long max; 1370 unsigned int order; 1371 int err = proto_register(&netlink_proto, 0); 1372 1373 if (err != 0) 1374 goto out; 1375 1376 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb)) 1377 netlink_skb_parms_too_large(); 1378 1379 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL); 1380 if (!nl_table) { 1381 enomem: 1382 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n"); 1383 return -ENOMEM; 1384 } 1385 1386 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS); 1387 1388 if (num_physpages >= (128 * 1024)) 1389 max = num_physpages >> (21 - PAGE_SHIFT); 1390 else 1391 max = num_physpages >> (23 - PAGE_SHIFT); 1392 1393 order = get_bitmask_order(max) - 1 + PAGE_SHIFT; 1394 max = (1UL << order) / sizeof(struct hlist_head); 1395 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1; 1396 1397 for (i = 0; i < MAX_LINKS; i++) { 1398 struct nl_pid_hash *hash = &nl_table[i].hash; 1399 1400 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table)); 1401 if (!hash->table) { 1402 while (i-- > 0) 1403 nl_pid_hash_free(nl_table[i].hash.table, 1404 1 * sizeof(*hash->table)); 1405 kfree(nl_table); 1406 goto enomem; 1407 } 1408 memset(hash->table, 0, 1 * sizeof(*hash->table)); 1409 hash->max_shift = order; 1410 hash->shift = 0; 1411 hash->mask = 0; 1412 hash->rehash_time = jiffies; 1413 } 1414 1415 sock_register(&netlink_family_ops); 1416 #ifdef CONFIG_PROC_FS 1417 proc_net_fops_create("netlink", 0, &netlink_seq_fops); 1418 #endif 1419 /* The netlink device handler may be needed early. */ 1420 rtnetlink_init(); 1421 out: 1422 return err; 1423 } 1424 1425 static void __exit netlink_proto_exit(void) 1426 { 1427 sock_unregister(PF_NETLINK); 1428 proc_net_remove("netlink"); 1429 kfree(nl_table); 1430 nl_table = NULL; 1431 proto_unregister(&netlink_proto); 1432 } 1433 1434 core_initcall(netlink_proto_init); 1435 module_exit(netlink_proto_exit); 1436 1437 MODULE_LICENSE("GPL"); 1438 1439 MODULE_ALIAS_NETPROTO(PF_NETLINK); 1440 1441 EXPORT_SYMBOL(netlink_ack); 1442 EXPORT_SYMBOL(netlink_broadcast); 1443 EXPORT_SYMBOL(netlink_dump_start); 1444 EXPORT_SYMBOL(netlink_kernel_create); 1445 EXPORT_SYMBOL(netlink_register_notifier); 1446 EXPORT_SYMBOL(netlink_set_err); 1447 EXPORT_SYMBOL(netlink_set_nonroot); 1448 EXPORT_SYMBOL(netlink_unicast); 1449 EXPORT_SYMBOL(netlink_unregister_notifier); 1450 1451