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 __sock_put(sk); 377 } 378 spin_unlock(&nlk->cb_lock); 379 380 /* OK. Socket is unlinked, and, therefore, 381 no new packets will arrive */ 382 383 sock_orphan(sk); 384 sock->sk = NULL; 385 wake_up_interruptible_all(&nlk->wait); 386 387 skb_queue_purge(&sk->sk_write_queue); 388 389 if (nlk->pid && !nlk->groups) { 390 struct netlink_notify n = { 391 .protocol = sk->sk_protocol, 392 .pid = nlk->pid, 393 }; 394 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n); 395 } 396 397 sock_put(sk); 398 return 0; 399 } 400 401 static int netlink_autobind(struct socket *sock) 402 { 403 struct sock *sk = sock->sk; 404 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 405 struct hlist_head *head; 406 struct sock *osk; 407 struct hlist_node *node; 408 s32 pid = current->pid; 409 int err; 410 static s32 rover = -4097; 411 412 retry: 413 cond_resched(); 414 netlink_table_grab(); 415 head = nl_pid_hashfn(hash, pid); 416 sk_for_each(osk, node, head) { 417 if (nlk_sk(osk)->pid == pid) { 418 /* Bind collision, search negative pid values. */ 419 pid = rover--; 420 if (rover > -4097) 421 rover = -4097; 422 netlink_table_ungrab(); 423 goto retry; 424 } 425 } 426 netlink_table_ungrab(); 427 428 err = netlink_insert(sk, pid); 429 if (err == -EADDRINUSE) 430 goto retry; 431 return 0; 432 } 433 434 static inline int netlink_capable(struct socket *sock, unsigned int flag) 435 { 436 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) || 437 capable(CAP_NET_ADMIN); 438 } 439 440 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 441 { 442 struct sock *sk = sock->sk; 443 struct netlink_sock *nlk = nlk_sk(sk); 444 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 445 int err; 446 447 if (nladdr->nl_family != AF_NETLINK) 448 return -EINVAL; 449 450 /* Only superuser is allowed to listen multicasts */ 451 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV)) 452 return -EPERM; 453 454 if (nlk->pid) { 455 if (nladdr->nl_pid != nlk->pid) 456 return -EINVAL; 457 } else { 458 err = nladdr->nl_pid ? 459 netlink_insert(sk, nladdr->nl_pid) : 460 netlink_autobind(sock); 461 if (err) 462 return err; 463 } 464 465 if (!nladdr->nl_groups && !nlk->groups) 466 return 0; 467 468 netlink_table_grab(); 469 if (nlk->groups && !nladdr->nl_groups) 470 __sk_del_bind_node(sk); 471 else if (!nlk->groups && nladdr->nl_groups) 472 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 473 nlk->groups = nladdr->nl_groups; 474 netlink_table_ungrab(); 475 476 return 0; 477 } 478 479 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 480 int alen, int flags) 481 { 482 int err = 0; 483 struct sock *sk = sock->sk; 484 struct netlink_sock *nlk = nlk_sk(sk); 485 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr; 486 487 if (addr->sa_family == AF_UNSPEC) { 488 sk->sk_state = NETLINK_UNCONNECTED; 489 nlk->dst_pid = 0; 490 nlk->dst_groups = 0; 491 return 0; 492 } 493 if (addr->sa_family != AF_NETLINK) 494 return -EINVAL; 495 496 /* Only superuser is allowed to send multicasts */ 497 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND)) 498 return -EPERM; 499 500 if (!nlk->pid) 501 err = netlink_autobind(sock); 502 503 if (err == 0) { 504 sk->sk_state = NETLINK_CONNECTED; 505 nlk->dst_pid = nladdr->nl_pid; 506 nlk->dst_groups = nladdr->nl_groups; 507 } 508 509 return err; 510 } 511 512 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer) 513 { 514 struct sock *sk = sock->sk; 515 struct netlink_sock *nlk = nlk_sk(sk); 516 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr; 517 518 nladdr->nl_family = AF_NETLINK; 519 nladdr->nl_pad = 0; 520 *addr_len = sizeof(*nladdr); 521 522 if (peer) { 523 nladdr->nl_pid = nlk->dst_pid; 524 nladdr->nl_groups = nlk->dst_groups; 525 } else { 526 nladdr->nl_pid = nlk->pid; 527 nladdr->nl_groups = nlk->groups; 528 } 529 return 0; 530 } 531 532 static void netlink_overrun(struct sock *sk) 533 { 534 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) { 535 sk->sk_err = ENOBUFS; 536 sk->sk_error_report(sk); 537 } 538 } 539 540 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid) 541 { 542 int protocol = ssk->sk_protocol; 543 struct sock *sock; 544 struct netlink_sock *nlk; 545 546 sock = netlink_lookup(protocol, pid); 547 if (!sock) 548 return ERR_PTR(-ECONNREFUSED); 549 550 /* Don't bother queuing skb if kernel socket has no input function */ 551 nlk = nlk_sk(sock); 552 if ((nlk->pid == 0 && !nlk->data_ready) || 553 (sock->sk_state == NETLINK_CONNECTED && 554 nlk->dst_pid != nlk_sk(ssk)->pid)) { 555 sock_put(sock); 556 return ERR_PTR(-ECONNREFUSED); 557 } 558 return sock; 559 } 560 561 struct sock *netlink_getsockbyfilp(struct file *filp) 562 { 563 struct inode *inode = filp->f_dentry->d_inode; 564 struct sock *sock; 565 566 if (!S_ISSOCK(inode->i_mode)) 567 return ERR_PTR(-ENOTSOCK); 568 569 sock = SOCKET_I(inode)->sk; 570 if (sock->sk_family != AF_NETLINK) 571 return ERR_PTR(-EINVAL); 572 573 sock_hold(sock); 574 return sock; 575 } 576 577 /* 578 * Attach a skb to a netlink socket. 579 * The caller must hold a reference to the destination socket. On error, the 580 * reference is dropped. The skb is not send to the destination, just all 581 * all error checks are performed and memory in the queue is reserved. 582 * Return values: 583 * < 0: error. skb freed, reference to sock dropped. 584 * 0: continue 585 * 1: repeat lookup - reference dropped while waiting for socket memory. 586 */ 587 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo) 588 { 589 struct netlink_sock *nlk; 590 591 nlk = nlk_sk(sk); 592 593 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 594 test_bit(0, &nlk->state)) { 595 DECLARE_WAITQUEUE(wait, current); 596 if (!timeo) { 597 if (!nlk->pid) 598 netlink_overrun(sk); 599 sock_put(sk); 600 kfree_skb(skb); 601 return -EAGAIN; 602 } 603 604 __set_current_state(TASK_INTERRUPTIBLE); 605 add_wait_queue(&nlk->wait, &wait); 606 607 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 608 test_bit(0, &nlk->state)) && 609 !sock_flag(sk, SOCK_DEAD)) 610 timeo = schedule_timeout(timeo); 611 612 __set_current_state(TASK_RUNNING); 613 remove_wait_queue(&nlk->wait, &wait); 614 sock_put(sk); 615 616 if (signal_pending(current)) { 617 kfree_skb(skb); 618 return sock_intr_errno(timeo); 619 } 620 return 1; 621 } 622 skb_set_owner_r(skb, sk); 623 return 0; 624 } 625 626 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol) 627 { 628 struct netlink_sock *nlk; 629 int len = skb->len; 630 631 nlk = nlk_sk(sk); 632 633 skb_queue_tail(&sk->sk_receive_queue, skb); 634 sk->sk_data_ready(sk, len); 635 sock_put(sk); 636 return len; 637 } 638 639 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 640 { 641 kfree_skb(skb); 642 sock_put(sk); 643 } 644 645 static inline struct sk_buff *netlink_trim(struct sk_buff *skb, int allocation) 646 { 647 int delta; 648 649 skb_orphan(skb); 650 651 delta = skb->end - skb->tail; 652 if (delta * 2 < skb->truesize) 653 return skb; 654 655 if (skb_shared(skb)) { 656 struct sk_buff *nskb = skb_clone(skb, allocation); 657 if (!nskb) 658 return skb; 659 kfree_skb(skb); 660 skb = nskb; 661 } 662 663 if (!pskb_expand_head(skb, 0, -delta, allocation)) 664 skb->truesize -= delta; 665 666 return skb; 667 } 668 669 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock) 670 { 671 struct sock *sk; 672 int err; 673 long timeo; 674 675 skb = netlink_trim(skb, gfp_any()); 676 677 timeo = sock_sndtimeo(ssk, nonblock); 678 retry: 679 sk = netlink_getsockbypid(ssk, pid); 680 if (IS_ERR(sk)) { 681 kfree_skb(skb); 682 return PTR_ERR(sk); 683 } 684 err = netlink_attachskb(sk, skb, nonblock, timeo); 685 if (err == 1) 686 goto retry; 687 if (err) 688 return err; 689 690 return netlink_sendskb(sk, skb, ssk->sk_protocol); 691 } 692 693 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 694 { 695 struct netlink_sock *nlk = nlk_sk(sk); 696 697 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 698 !test_bit(0, &nlk->state)) { 699 skb_set_owner_r(skb, sk); 700 skb_queue_tail(&sk->sk_receive_queue, skb); 701 sk->sk_data_ready(sk, skb->len); 702 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf; 703 } 704 return -1; 705 } 706 707 struct netlink_broadcast_data { 708 struct sock *exclude_sk; 709 u32 pid; 710 u32 group; 711 int failure; 712 int congested; 713 int delivered; 714 int allocation; 715 struct sk_buff *skb, *skb2; 716 }; 717 718 static inline int do_one_broadcast(struct sock *sk, 719 struct netlink_broadcast_data *p) 720 { 721 struct netlink_sock *nlk = nlk_sk(sk); 722 int val; 723 724 if (p->exclude_sk == sk) 725 goto out; 726 727 if (nlk->pid == p->pid || !(nlk->groups & p->group)) 728 goto out; 729 730 if (p->failure) { 731 netlink_overrun(sk); 732 goto out; 733 } 734 735 sock_hold(sk); 736 if (p->skb2 == NULL) { 737 if (atomic_read(&p->skb->users) != 1) { 738 p->skb2 = skb_clone(p->skb, p->allocation); 739 } else { 740 p->skb2 = p->skb; 741 atomic_inc(&p->skb->users); 742 } 743 } 744 if (p->skb2 == NULL) { 745 netlink_overrun(sk); 746 /* Clone failed. Notify ALL listeners. */ 747 p->failure = 1; 748 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) { 749 netlink_overrun(sk); 750 } else { 751 p->congested |= val; 752 p->delivered = 1; 753 p->skb2 = NULL; 754 } 755 sock_put(sk); 756 757 out: 758 return 0; 759 } 760 761 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid, 762 u32 group, int allocation) 763 { 764 struct netlink_broadcast_data info; 765 struct hlist_node *node; 766 struct sock *sk; 767 768 skb = netlink_trim(skb, allocation); 769 770 info.exclude_sk = ssk; 771 info.pid = pid; 772 info.group = group; 773 info.failure = 0; 774 info.congested = 0; 775 info.delivered = 0; 776 info.allocation = allocation; 777 info.skb = skb; 778 info.skb2 = NULL; 779 780 /* While we sleep in clone, do not allow to change socket list */ 781 782 netlink_lock_table(); 783 784 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 785 do_one_broadcast(sk, &info); 786 787 netlink_unlock_table(); 788 789 if (info.skb2) 790 kfree_skb(info.skb2); 791 kfree_skb(skb); 792 793 if (info.delivered) { 794 if (info.congested && (allocation & __GFP_WAIT)) 795 yield(); 796 return 0; 797 } 798 if (info.failure) 799 return -ENOBUFS; 800 return -ESRCH; 801 } 802 803 struct netlink_set_err_data { 804 struct sock *exclude_sk; 805 u32 pid; 806 u32 group; 807 int code; 808 }; 809 810 static inline int do_one_set_err(struct sock *sk, 811 struct netlink_set_err_data *p) 812 { 813 struct netlink_sock *nlk = nlk_sk(sk); 814 815 if (sk == p->exclude_sk) 816 goto out; 817 818 if (nlk->pid == p->pid || !(nlk->groups & p->group)) 819 goto out; 820 821 sk->sk_err = p->code; 822 sk->sk_error_report(sk); 823 out: 824 return 0; 825 } 826 827 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code) 828 { 829 struct netlink_set_err_data info; 830 struct hlist_node *node; 831 struct sock *sk; 832 833 info.exclude_sk = ssk; 834 info.pid = pid; 835 info.group = group; 836 info.code = code; 837 838 read_lock(&nl_table_lock); 839 840 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 841 do_one_set_err(sk, &info); 842 843 read_unlock(&nl_table_lock); 844 } 845 846 static inline void netlink_rcv_wake(struct sock *sk) 847 { 848 struct netlink_sock *nlk = nlk_sk(sk); 849 850 if (!skb_queue_len(&sk->sk_receive_queue)) 851 clear_bit(0, &nlk->state); 852 if (!test_bit(0, &nlk->state)) 853 wake_up_interruptible(&nlk->wait); 854 } 855 856 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock, 857 struct msghdr *msg, size_t len) 858 { 859 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 860 struct sock *sk = sock->sk; 861 struct netlink_sock *nlk = nlk_sk(sk); 862 struct sockaddr_nl *addr=msg->msg_name; 863 u32 dst_pid; 864 u32 dst_groups; 865 struct sk_buff *skb; 866 int err; 867 struct scm_cookie scm; 868 869 if (msg->msg_flags&MSG_OOB) 870 return -EOPNOTSUPP; 871 872 if (NULL == siocb->scm) 873 siocb->scm = &scm; 874 err = scm_send(sock, msg, siocb->scm); 875 if (err < 0) 876 return err; 877 878 if (msg->msg_namelen) { 879 if (addr->nl_family != AF_NETLINK) 880 return -EINVAL; 881 dst_pid = addr->nl_pid; 882 dst_groups = addr->nl_groups; 883 if (dst_groups && !netlink_capable(sock, NL_NONROOT_SEND)) 884 return -EPERM; 885 } else { 886 dst_pid = nlk->dst_pid; 887 dst_groups = nlk->dst_groups; 888 } 889 890 if (!nlk->pid) { 891 err = netlink_autobind(sock); 892 if (err) 893 goto out; 894 } 895 896 err = -EMSGSIZE; 897 if (len > sk->sk_sndbuf - 32) 898 goto out; 899 err = -ENOBUFS; 900 skb = alloc_skb(len, GFP_KERNEL); 901 if (skb==NULL) 902 goto out; 903 904 NETLINK_CB(skb).pid = nlk->pid; 905 NETLINK_CB(skb).groups = nlk->groups; 906 NETLINK_CB(skb).dst_pid = dst_pid; 907 NETLINK_CB(skb).dst_groups = dst_groups; 908 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); 909 910 /* What can I do? Netlink is asynchronous, so that 911 we will have to save current capabilities to 912 check them, when this message will be delivered 913 to corresponding kernel module. --ANK (980802) 914 */ 915 916 err = -EFAULT; 917 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) { 918 kfree_skb(skb); 919 goto out; 920 } 921 922 err = security_netlink_send(sk, skb); 923 if (err) { 924 kfree_skb(skb); 925 goto out; 926 } 927 928 if (dst_groups) { 929 atomic_inc(&skb->users); 930 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL); 931 } 932 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT); 933 934 out: 935 return err; 936 } 937 938 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock, 939 struct msghdr *msg, size_t len, 940 int flags) 941 { 942 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 943 struct scm_cookie scm; 944 struct sock *sk = sock->sk; 945 struct netlink_sock *nlk = nlk_sk(sk); 946 int noblock = flags&MSG_DONTWAIT; 947 size_t copied; 948 struct sk_buff *skb; 949 int err; 950 951 if (flags&MSG_OOB) 952 return -EOPNOTSUPP; 953 954 copied = 0; 955 956 skb = skb_recv_datagram(sk,flags,noblock,&err); 957 if (skb==NULL) 958 goto out; 959 960 msg->msg_namelen = 0; 961 962 copied = skb->len; 963 if (len < copied) { 964 msg->msg_flags |= MSG_TRUNC; 965 copied = len; 966 } 967 968 skb->h.raw = skb->data; 969 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 970 971 if (msg->msg_name) { 972 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name; 973 addr->nl_family = AF_NETLINK; 974 addr->nl_pad = 0; 975 addr->nl_pid = NETLINK_CB(skb).pid; 976 addr->nl_groups = NETLINK_CB(skb).dst_groups; 977 msg->msg_namelen = sizeof(*addr); 978 } 979 980 if (NULL == siocb->scm) { 981 memset(&scm, 0, sizeof(scm)); 982 siocb->scm = &scm; 983 } 984 siocb->scm->creds = *NETLINK_CREDS(skb); 985 skb_free_datagram(sk, skb); 986 987 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) 988 netlink_dump(sk); 989 990 scm_recv(sock, msg, siocb->scm, flags); 991 992 out: 993 netlink_rcv_wake(sk); 994 return err ? : copied; 995 } 996 997 static void netlink_data_ready(struct sock *sk, int len) 998 { 999 struct netlink_sock *nlk = nlk_sk(sk); 1000 1001 if (nlk->data_ready) 1002 nlk->data_ready(sk, len); 1003 netlink_rcv_wake(sk); 1004 } 1005 1006 /* 1007 * We export these functions to other modules. They provide a 1008 * complete set of kernel non-blocking support for message 1009 * queueing. 1010 */ 1011 1012 struct sock * 1013 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len)) 1014 { 1015 struct socket *sock; 1016 struct sock *sk; 1017 1018 if (!nl_table) 1019 return NULL; 1020 1021 if (unit<0 || unit>=MAX_LINKS) 1022 return NULL; 1023 1024 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 1025 return NULL; 1026 1027 if (netlink_create(sock, unit) < 0) { 1028 sock_release(sock); 1029 return NULL; 1030 } 1031 sk = sock->sk; 1032 sk->sk_data_ready = netlink_data_ready; 1033 if (input) 1034 nlk_sk(sk)->data_ready = input; 1035 1036 if (netlink_insert(sk, 0)) { 1037 sock_release(sock); 1038 return NULL; 1039 } 1040 return sk; 1041 } 1042 1043 void netlink_set_nonroot(int protocol, unsigned int flags) 1044 { 1045 if ((unsigned int)protocol < MAX_LINKS) 1046 nl_table[protocol].nl_nonroot = flags; 1047 } 1048 1049 static void netlink_destroy_callback(struct netlink_callback *cb) 1050 { 1051 if (cb->skb) 1052 kfree_skb(cb->skb); 1053 kfree(cb); 1054 } 1055 1056 /* 1057 * It looks a bit ugly. 1058 * It would be better to create kernel thread. 1059 */ 1060 1061 static int netlink_dump(struct sock *sk) 1062 { 1063 struct netlink_sock *nlk = nlk_sk(sk); 1064 struct netlink_callback *cb; 1065 struct sk_buff *skb; 1066 struct nlmsghdr *nlh; 1067 int len; 1068 1069 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL); 1070 if (!skb) 1071 return -ENOBUFS; 1072 1073 spin_lock(&nlk->cb_lock); 1074 1075 cb = nlk->cb; 1076 if (cb == NULL) { 1077 spin_unlock(&nlk->cb_lock); 1078 kfree_skb(skb); 1079 return -EINVAL; 1080 } 1081 1082 len = cb->dump(skb, cb); 1083 1084 if (len > 0) { 1085 spin_unlock(&nlk->cb_lock); 1086 skb_queue_tail(&sk->sk_receive_queue, skb); 1087 sk->sk_data_ready(sk, len); 1088 return 0; 1089 } 1090 1091 nlh = __nlmsg_put(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, NLMSG_DONE, sizeof(int)); 1092 nlh->nlmsg_flags |= NLM_F_MULTI; 1093 memcpy(NLMSG_DATA(nlh), &len, sizeof(len)); 1094 skb_queue_tail(&sk->sk_receive_queue, skb); 1095 sk->sk_data_ready(sk, skb->len); 1096 1097 cb->done(cb); 1098 nlk->cb = NULL; 1099 spin_unlock(&nlk->cb_lock); 1100 1101 netlink_destroy_callback(cb); 1102 __sock_put(sk); 1103 return 0; 1104 } 1105 1106 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 1107 struct nlmsghdr *nlh, 1108 int (*dump)(struct sk_buff *skb, struct netlink_callback*), 1109 int (*done)(struct netlink_callback*)) 1110 { 1111 struct netlink_callback *cb; 1112 struct sock *sk; 1113 struct netlink_sock *nlk; 1114 1115 cb = kmalloc(sizeof(*cb), GFP_KERNEL); 1116 if (cb == NULL) 1117 return -ENOBUFS; 1118 1119 memset(cb, 0, sizeof(*cb)); 1120 cb->dump = dump; 1121 cb->done = done; 1122 cb->nlh = nlh; 1123 atomic_inc(&skb->users); 1124 cb->skb = skb; 1125 1126 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid); 1127 if (sk == NULL) { 1128 netlink_destroy_callback(cb); 1129 return -ECONNREFUSED; 1130 } 1131 nlk = nlk_sk(sk); 1132 /* A dump is in progress... */ 1133 spin_lock(&nlk->cb_lock); 1134 if (nlk->cb) { 1135 spin_unlock(&nlk->cb_lock); 1136 netlink_destroy_callback(cb); 1137 sock_put(sk); 1138 return -EBUSY; 1139 } 1140 nlk->cb = cb; 1141 sock_hold(sk); 1142 spin_unlock(&nlk->cb_lock); 1143 1144 netlink_dump(sk); 1145 sock_put(sk); 1146 return 0; 1147 } 1148 1149 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err) 1150 { 1151 struct sk_buff *skb; 1152 struct nlmsghdr *rep; 1153 struct nlmsgerr *errmsg; 1154 int size; 1155 1156 if (err == 0) 1157 size = NLMSG_SPACE(sizeof(struct nlmsgerr)); 1158 else 1159 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len)); 1160 1161 skb = alloc_skb(size, GFP_KERNEL); 1162 if (!skb) { 1163 struct sock *sk; 1164 1165 sk = netlink_lookup(in_skb->sk->sk_protocol, 1166 NETLINK_CB(in_skb).pid); 1167 if (sk) { 1168 sk->sk_err = ENOBUFS; 1169 sk->sk_error_report(sk); 1170 sock_put(sk); 1171 } 1172 return; 1173 } 1174 1175 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, 1176 NLMSG_ERROR, sizeof(struct nlmsgerr)); 1177 errmsg = NLMSG_DATA(rep); 1178 errmsg->error = err; 1179 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr)); 1180 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT); 1181 } 1182 1183 1184 #ifdef CONFIG_PROC_FS 1185 struct nl_seq_iter { 1186 int link; 1187 int hash_idx; 1188 }; 1189 1190 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos) 1191 { 1192 struct nl_seq_iter *iter = seq->private; 1193 int i, j; 1194 struct sock *s; 1195 struct hlist_node *node; 1196 loff_t off = 0; 1197 1198 for (i=0; i<MAX_LINKS; i++) { 1199 struct nl_pid_hash *hash = &nl_table[i].hash; 1200 1201 for (j = 0; j <= hash->mask; j++) { 1202 sk_for_each(s, node, &hash->table[j]) { 1203 if (off == pos) { 1204 iter->link = i; 1205 iter->hash_idx = j; 1206 return s; 1207 } 1208 ++off; 1209 } 1210 } 1211 } 1212 return NULL; 1213 } 1214 1215 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos) 1216 { 1217 read_lock(&nl_table_lock); 1218 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN; 1219 } 1220 1221 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1222 { 1223 struct sock *s; 1224 struct nl_seq_iter *iter; 1225 int i, j; 1226 1227 ++*pos; 1228 1229 if (v == SEQ_START_TOKEN) 1230 return netlink_seq_socket_idx(seq, 0); 1231 1232 s = sk_next(v); 1233 if (s) 1234 return s; 1235 1236 iter = seq->private; 1237 i = iter->link; 1238 j = iter->hash_idx + 1; 1239 1240 do { 1241 struct nl_pid_hash *hash = &nl_table[i].hash; 1242 1243 for (; j <= hash->mask; j++) { 1244 s = sk_head(&hash->table[j]); 1245 if (s) { 1246 iter->link = i; 1247 iter->hash_idx = j; 1248 return s; 1249 } 1250 } 1251 1252 j = 0; 1253 } while (++i < MAX_LINKS); 1254 1255 return NULL; 1256 } 1257 1258 static void netlink_seq_stop(struct seq_file *seq, void *v) 1259 { 1260 read_unlock(&nl_table_lock); 1261 } 1262 1263 1264 static int netlink_seq_show(struct seq_file *seq, void *v) 1265 { 1266 if (v == SEQ_START_TOKEN) 1267 seq_puts(seq, 1268 "sk Eth Pid Groups " 1269 "Rmem Wmem Dump Locks\n"); 1270 else { 1271 struct sock *s = v; 1272 struct netlink_sock *nlk = nlk_sk(s); 1273 1274 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n", 1275 s, 1276 s->sk_protocol, 1277 nlk->pid, 1278 nlk->groups, 1279 atomic_read(&s->sk_rmem_alloc), 1280 atomic_read(&s->sk_wmem_alloc), 1281 nlk->cb, 1282 atomic_read(&s->sk_refcnt) 1283 ); 1284 1285 } 1286 return 0; 1287 } 1288 1289 static struct seq_operations netlink_seq_ops = { 1290 .start = netlink_seq_start, 1291 .next = netlink_seq_next, 1292 .stop = netlink_seq_stop, 1293 .show = netlink_seq_show, 1294 }; 1295 1296 1297 static int netlink_seq_open(struct inode *inode, struct file *file) 1298 { 1299 struct seq_file *seq; 1300 struct nl_seq_iter *iter; 1301 int err; 1302 1303 iter = kmalloc(sizeof(*iter), GFP_KERNEL); 1304 if (!iter) 1305 return -ENOMEM; 1306 1307 err = seq_open(file, &netlink_seq_ops); 1308 if (err) { 1309 kfree(iter); 1310 return err; 1311 } 1312 1313 memset(iter, 0, sizeof(*iter)); 1314 seq = file->private_data; 1315 seq->private = iter; 1316 return 0; 1317 } 1318 1319 static struct file_operations netlink_seq_fops = { 1320 .owner = THIS_MODULE, 1321 .open = netlink_seq_open, 1322 .read = seq_read, 1323 .llseek = seq_lseek, 1324 .release = seq_release_private, 1325 }; 1326 1327 #endif 1328 1329 int netlink_register_notifier(struct notifier_block *nb) 1330 { 1331 return notifier_chain_register(&netlink_chain, nb); 1332 } 1333 1334 int netlink_unregister_notifier(struct notifier_block *nb) 1335 { 1336 return notifier_chain_unregister(&netlink_chain, nb); 1337 } 1338 1339 static struct proto_ops netlink_ops = { 1340 .family = PF_NETLINK, 1341 .owner = THIS_MODULE, 1342 .release = netlink_release, 1343 .bind = netlink_bind, 1344 .connect = netlink_connect, 1345 .socketpair = sock_no_socketpair, 1346 .accept = sock_no_accept, 1347 .getname = netlink_getname, 1348 .poll = datagram_poll, 1349 .ioctl = sock_no_ioctl, 1350 .listen = sock_no_listen, 1351 .shutdown = sock_no_shutdown, 1352 .setsockopt = sock_no_setsockopt, 1353 .getsockopt = sock_no_getsockopt, 1354 .sendmsg = netlink_sendmsg, 1355 .recvmsg = netlink_recvmsg, 1356 .mmap = sock_no_mmap, 1357 .sendpage = sock_no_sendpage, 1358 }; 1359 1360 static struct net_proto_family netlink_family_ops = { 1361 .family = PF_NETLINK, 1362 .create = netlink_create, 1363 .owner = THIS_MODULE, /* for consistency 8) */ 1364 }; 1365 1366 extern void netlink_skb_parms_too_large(void); 1367 1368 static int __init netlink_proto_init(void) 1369 { 1370 struct sk_buff *dummy_skb; 1371 int i; 1372 unsigned long max; 1373 unsigned int order; 1374 int err = proto_register(&netlink_proto, 0); 1375 1376 if (err != 0) 1377 goto out; 1378 1379 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb)) 1380 netlink_skb_parms_too_large(); 1381 1382 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL); 1383 if (!nl_table) { 1384 enomem: 1385 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n"); 1386 return -ENOMEM; 1387 } 1388 1389 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS); 1390 1391 if (num_physpages >= (128 * 1024)) 1392 max = num_physpages >> (21 - PAGE_SHIFT); 1393 else 1394 max = num_physpages >> (23 - PAGE_SHIFT); 1395 1396 order = get_bitmask_order(max) - 1 + PAGE_SHIFT; 1397 max = (1UL << order) / sizeof(struct hlist_head); 1398 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1; 1399 1400 for (i = 0; i < MAX_LINKS; i++) { 1401 struct nl_pid_hash *hash = &nl_table[i].hash; 1402 1403 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table)); 1404 if (!hash->table) { 1405 while (i-- > 0) 1406 nl_pid_hash_free(nl_table[i].hash.table, 1407 1 * sizeof(*hash->table)); 1408 kfree(nl_table); 1409 goto enomem; 1410 } 1411 memset(hash->table, 0, 1 * sizeof(*hash->table)); 1412 hash->max_shift = order; 1413 hash->shift = 0; 1414 hash->mask = 0; 1415 hash->rehash_time = jiffies; 1416 } 1417 1418 sock_register(&netlink_family_ops); 1419 #ifdef CONFIG_PROC_FS 1420 proc_net_fops_create("netlink", 0, &netlink_seq_fops); 1421 #endif 1422 /* The netlink device handler may be needed early. */ 1423 rtnetlink_init(); 1424 out: 1425 return err; 1426 } 1427 1428 static void __exit netlink_proto_exit(void) 1429 { 1430 sock_unregister(PF_NETLINK); 1431 proc_net_remove("netlink"); 1432 kfree(nl_table); 1433 nl_table = NULL; 1434 proto_unregister(&netlink_proto); 1435 } 1436 1437 core_initcall(netlink_proto_init); 1438 module_exit(netlink_proto_exit); 1439 1440 MODULE_LICENSE("GPL"); 1441 1442 MODULE_ALIAS_NETPROTO(PF_NETLINK); 1443 1444 EXPORT_SYMBOL(netlink_ack); 1445 EXPORT_SYMBOL(netlink_broadcast); 1446 EXPORT_SYMBOL(netlink_dump_start); 1447 EXPORT_SYMBOL(netlink_kernel_create); 1448 EXPORT_SYMBOL(netlink_register_notifier); 1449 EXPORT_SYMBOL(netlink_set_err); 1450 EXPORT_SYMBOL(netlink_set_nonroot); 1451 EXPORT_SYMBOL(netlink_unicast); 1452 EXPORT_SYMBOL(netlink_unregister_notifier); 1453 1454