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/major.h> 25 #include <linux/signal.h> 26 #include <linux/sched.h> 27 #include <linux/errno.h> 28 #include <linux/string.h> 29 #include <linux/stat.h> 30 #include <linux/socket.h> 31 #include <linux/un.h> 32 #include <linux/fcntl.h> 33 #include <linux/termios.h> 34 #include <linux/sockios.h> 35 #include <linux/net.h> 36 #include <linux/fs.h> 37 #include <linux/slab.h> 38 #include <asm/uaccess.h> 39 #include <linux/skbuff.h> 40 #include <linux/netdevice.h> 41 #include <linux/rtnetlink.h> 42 #include <linux/proc_fs.h> 43 #include <linux/seq_file.h> 44 #include <linux/smp_lock.h> 45 #include <linux/notifier.h> 46 #include <linux/security.h> 47 #include <linux/jhash.h> 48 #include <linux/jiffies.h> 49 #include <linux/random.h> 50 #include <linux/bitops.h> 51 #include <linux/mm.h> 52 #include <linux/types.h> 53 #include <net/sock.h> 54 #include <net/scm.h> 55 56 #define Nprintk(a...) 57 58 struct netlink_sock { 59 /* struct sock has to be the first member of netlink_sock */ 60 struct sock sk; 61 u32 pid; 62 unsigned int groups; 63 u32 dst_pid; 64 unsigned int dst_groups; 65 unsigned long state; 66 wait_queue_head_t wait; 67 struct netlink_callback *cb; 68 spinlock_t cb_lock; 69 void (*data_ready)(struct sock *sk, int bytes); 70 }; 71 72 static inline struct netlink_sock *nlk_sk(struct sock *sk) 73 { 74 return (struct netlink_sock *)sk; 75 } 76 77 struct nl_pid_hash { 78 struct hlist_head *table; 79 unsigned long rehash_time; 80 81 unsigned int mask; 82 unsigned int shift; 83 84 unsigned int entries; 85 unsigned int max_shift; 86 87 u32 rnd; 88 }; 89 90 struct netlink_table { 91 struct nl_pid_hash hash; 92 struct hlist_head mc_list; 93 unsigned int nl_nonroot; 94 }; 95 96 static struct netlink_table *nl_table; 97 98 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait); 99 100 static int netlink_dump(struct sock *sk); 101 static void netlink_destroy_callback(struct netlink_callback *cb); 102 103 static DEFINE_RWLOCK(nl_table_lock); 104 static atomic_t nl_table_users = ATOMIC_INIT(0); 105 106 static struct notifier_block *netlink_chain; 107 108 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid) 109 { 110 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask]; 111 } 112 113 static void netlink_sock_destruct(struct sock *sk) 114 { 115 skb_queue_purge(&sk->sk_receive_queue); 116 117 if (!sock_flag(sk, SOCK_DEAD)) { 118 printk("Freeing alive netlink socket %p\n", sk); 119 return; 120 } 121 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc)); 122 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc)); 123 BUG_TRAP(!nlk_sk(sk)->cb); 124 } 125 126 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP. 127 * Look, when several writers sleep and reader wakes them up, all but one 128 * immediately hit write lock and grab all the cpus. Exclusive sleep solves 129 * this, _but_ remember, it adds useless work on UP machines. 130 */ 131 132 static void netlink_table_grab(void) 133 { 134 write_lock_bh(&nl_table_lock); 135 136 if (atomic_read(&nl_table_users)) { 137 DECLARE_WAITQUEUE(wait, current); 138 139 add_wait_queue_exclusive(&nl_table_wait, &wait); 140 for(;;) { 141 set_current_state(TASK_UNINTERRUPTIBLE); 142 if (atomic_read(&nl_table_users) == 0) 143 break; 144 write_unlock_bh(&nl_table_lock); 145 schedule(); 146 write_lock_bh(&nl_table_lock); 147 } 148 149 __set_current_state(TASK_RUNNING); 150 remove_wait_queue(&nl_table_wait, &wait); 151 } 152 } 153 154 static __inline__ void netlink_table_ungrab(void) 155 { 156 write_unlock_bh(&nl_table_lock); 157 wake_up(&nl_table_wait); 158 } 159 160 static __inline__ void 161 netlink_lock_table(void) 162 { 163 /* read_lock() synchronizes us to netlink_table_grab */ 164 165 read_lock(&nl_table_lock); 166 atomic_inc(&nl_table_users); 167 read_unlock(&nl_table_lock); 168 } 169 170 static __inline__ void 171 netlink_unlock_table(void) 172 { 173 if (atomic_dec_and_test(&nl_table_users)) 174 wake_up(&nl_table_wait); 175 } 176 177 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid) 178 { 179 struct nl_pid_hash *hash = &nl_table[protocol].hash; 180 struct hlist_head *head; 181 struct sock *sk; 182 struct hlist_node *node; 183 184 read_lock(&nl_table_lock); 185 head = nl_pid_hashfn(hash, pid); 186 sk_for_each(sk, node, head) { 187 if (nlk_sk(sk)->pid == pid) { 188 sock_hold(sk); 189 goto found; 190 } 191 } 192 sk = NULL; 193 found: 194 read_unlock(&nl_table_lock); 195 return sk; 196 } 197 198 static inline struct hlist_head *nl_pid_hash_alloc(size_t size) 199 { 200 if (size <= PAGE_SIZE) 201 return kmalloc(size, GFP_ATOMIC); 202 else 203 return (struct hlist_head *) 204 __get_free_pages(GFP_ATOMIC, get_order(size)); 205 } 206 207 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size) 208 { 209 if (size <= PAGE_SIZE) 210 kfree(table); 211 else 212 free_pages((unsigned long)table, get_order(size)); 213 } 214 215 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow) 216 { 217 unsigned int omask, mask, shift; 218 size_t osize, size; 219 struct hlist_head *otable, *table; 220 int i; 221 222 omask = mask = hash->mask; 223 osize = size = (mask + 1) * sizeof(*table); 224 shift = hash->shift; 225 226 if (grow) { 227 if (++shift > hash->max_shift) 228 return 0; 229 mask = mask * 2 + 1; 230 size *= 2; 231 } 232 233 table = nl_pid_hash_alloc(size); 234 if (!table) 235 return 0; 236 237 memset(table, 0, size); 238 otable = hash->table; 239 hash->table = table; 240 hash->mask = mask; 241 hash->shift = shift; 242 get_random_bytes(&hash->rnd, sizeof(hash->rnd)); 243 244 for (i = 0; i <= omask; i++) { 245 struct sock *sk; 246 struct hlist_node *node, *tmp; 247 248 sk_for_each_safe(sk, node, tmp, &otable[i]) 249 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid)); 250 } 251 252 nl_pid_hash_free(otable, osize); 253 hash->rehash_time = jiffies + 10 * 60 * HZ; 254 return 1; 255 } 256 257 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len) 258 { 259 int avg = hash->entries >> hash->shift; 260 261 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1)) 262 return 1; 263 264 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) { 265 nl_pid_hash_rehash(hash, 0); 266 return 1; 267 } 268 269 return 0; 270 } 271 272 static struct proto_ops netlink_ops; 273 274 static int netlink_insert(struct sock *sk, u32 pid) 275 { 276 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 277 struct hlist_head *head; 278 int err = -EADDRINUSE; 279 struct sock *osk; 280 struct hlist_node *node; 281 int len; 282 283 netlink_table_grab(); 284 head = nl_pid_hashfn(hash, pid); 285 len = 0; 286 sk_for_each(osk, node, head) { 287 if (nlk_sk(osk)->pid == pid) 288 break; 289 len++; 290 } 291 if (node) 292 goto err; 293 294 err = -EBUSY; 295 if (nlk_sk(sk)->pid) 296 goto err; 297 298 err = -ENOMEM; 299 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX)) 300 goto err; 301 302 if (len && nl_pid_hash_dilute(hash, len)) 303 head = nl_pid_hashfn(hash, pid); 304 hash->entries++; 305 nlk_sk(sk)->pid = pid; 306 sk_add_node(sk, head); 307 err = 0; 308 309 err: 310 netlink_table_ungrab(); 311 return err; 312 } 313 314 static void netlink_remove(struct sock *sk) 315 { 316 netlink_table_grab(); 317 nl_table[sk->sk_protocol].hash.entries--; 318 sk_del_node_init(sk); 319 if (nlk_sk(sk)->groups) 320 __sk_del_bind_node(sk); 321 netlink_table_ungrab(); 322 } 323 324 static struct proto netlink_proto = { 325 .name = "NETLINK", 326 .owner = THIS_MODULE, 327 .obj_size = sizeof(struct netlink_sock), 328 }; 329 330 static int netlink_create(struct socket *sock, int protocol) 331 { 332 struct sock *sk; 333 struct netlink_sock *nlk; 334 335 sock->state = SS_UNCONNECTED; 336 337 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) 338 return -ESOCKTNOSUPPORT; 339 340 if (protocol<0 || protocol >= MAX_LINKS) 341 return -EPROTONOSUPPORT; 342 343 sock->ops = &netlink_ops; 344 345 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1); 346 if (!sk) 347 return -ENOMEM; 348 349 sock_init_data(sock, sk); 350 351 nlk = nlk_sk(sk); 352 353 spin_lock_init(&nlk->cb_lock); 354 init_waitqueue_head(&nlk->wait); 355 sk->sk_destruct = netlink_sock_destruct; 356 357 sk->sk_protocol = protocol; 358 return 0; 359 } 360 361 static int netlink_release(struct socket *sock) 362 { 363 struct sock *sk = sock->sk; 364 struct netlink_sock *nlk; 365 366 if (!sk) 367 return 0; 368 369 netlink_remove(sk); 370 nlk = nlk_sk(sk); 371 372 spin_lock(&nlk->cb_lock); 373 if (nlk->cb) { 374 nlk->cb->done(nlk->cb); 375 netlink_destroy_callback(nlk->cb); 376 nlk->cb = NULL; 377 __sock_put(sk); 378 } 379 spin_unlock(&nlk->cb_lock); 380 381 /* OK. Socket is unlinked, and, therefore, 382 no new packets will arrive */ 383 384 sock_orphan(sk); 385 sock->sk = NULL; 386 wake_up_interruptible_all(&nlk->wait); 387 388 skb_queue_purge(&sk->sk_write_queue); 389 390 if (nlk->pid && !nlk->groups) { 391 struct netlink_notify n = { 392 .protocol = sk->sk_protocol, 393 .pid = nlk->pid, 394 }; 395 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n); 396 } 397 398 sock_put(sk); 399 return 0; 400 } 401 402 static int netlink_autobind(struct socket *sock) 403 { 404 struct sock *sk = sock->sk; 405 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash; 406 struct hlist_head *head; 407 struct sock *osk; 408 struct hlist_node *node; 409 s32 pid = current->pid; 410 int err; 411 static s32 rover = -4097; 412 413 retry: 414 cond_resched(); 415 netlink_table_grab(); 416 head = nl_pid_hashfn(hash, pid); 417 sk_for_each(osk, node, head) { 418 if (nlk_sk(osk)->pid == pid) { 419 /* Bind collision, search negative pid values. */ 420 pid = rover--; 421 if (rover > -4097) 422 rover = -4097; 423 netlink_table_ungrab(); 424 goto retry; 425 } 426 } 427 netlink_table_ungrab(); 428 429 err = netlink_insert(sk, pid); 430 if (err == -EADDRINUSE) 431 goto retry; 432 return 0; 433 } 434 435 static inline int netlink_capable(struct socket *sock, unsigned int flag) 436 { 437 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) || 438 capable(CAP_NET_ADMIN); 439 } 440 441 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 442 { 443 struct sock *sk = sock->sk; 444 struct netlink_sock *nlk = nlk_sk(sk); 445 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; 446 int err; 447 448 if (nladdr->nl_family != AF_NETLINK) 449 return -EINVAL; 450 451 /* Only superuser is allowed to listen multicasts */ 452 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV)) 453 return -EPERM; 454 455 if (nlk->pid) { 456 if (nladdr->nl_pid != nlk->pid) 457 return -EINVAL; 458 } else { 459 err = nladdr->nl_pid ? 460 netlink_insert(sk, nladdr->nl_pid) : 461 netlink_autobind(sock); 462 if (err) 463 return err; 464 } 465 466 if (!nladdr->nl_groups && !nlk->groups) 467 return 0; 468 469 netlink_table_grab(); 470 if (nlk->groups && !nladdr->nl_groups) 471 __sk_del_bind_node(sk); 472 else if (!nlk->groups && nladdr->nl_groups) 473 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); 474 nlk->groups = nladdr->nl_groups; 475 netlink_table_ungrab(); 476 477 return 0; 478 } 479 480 static int netlink_connect(struct socket *sock, struct sockaddr *addr, 481 int alen, int flags) 482 { 483 int err = 0; 484 struct sock *sk = sock->sk; 485 struct netlink_sock *nlk = nlk_sk(sk); 486 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr; 487 488 if (addr->sa_family == AF_UNSPEC) { 489 sk->sk_state = NETLINK_UNCONNECTED; 490 nlk->dst_pid = 0; 491 nlk->dst_groups = 0; 492 return 0; 493 } 494 if (addr->sa_family != AF_NETLINK) 495 return -EINVAL; 496 497 /* Only superuser is allowed to send multicasts */ 498 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND)) 499 return -EPERM; 500 501 if (!nlk->pid) 502 err = netlink_autobind(sock); 503 504 if (err == 0) { 505 sk->sk_state = NETLINK_CONNECTED; 506 nlk->dst_pid = nladdr->nl_pid; 507 nlk->dst_groups = nladdr->nl_groups; 508 } 509 510 return err; 511 } 512 513 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer) 514 { 515 struct sock *sk = sock->sk; 516 struct netlink_sock *nlk = nlk_sk(sk); 517 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr; 518 519 nladdr->nl_family = AF_NETLINK; 520 nladdr->nl_pad = 0; 521 *addr_len = sizeof(*nladdr); 522 523 if (peer) { 524 nladdr->nl_pid = nlk->dst_pid; 525 nladdr->nl_groups = nlk->dst_groups; 526 } else { 527 nladdr->nl_pid = nlk->pid; 528 nladdr->nl_groups = nlk->groups; 529 } 530 return 0; 531 } 532 533 static void netlink_overrun(struct sock *sk) 534 { 535 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) { 536 sk->sk_err = ENOBUFS; 537 sk->sk_error_report(sk); 538 } 539 } 540 541 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid) 542 { 543 int protocol = ssk->sk_protocol; 544 struct sock *sock; 545 struct netlink_sock *nlk; 546 547 sock = netlink_lookup(protocol, pid); 548 if (!sock) 549 return ERR_PTR(-ECONNREFUSED); 550 551 /* Don't bother queuing skb if kernel socket has no input function */ 552 nlk = nlk_sk(sock); 553 if ((nlk->pid == 0 && !nlk->data_ready) || 554 (sock->sk_state == NETLINK_CONNECTED && 555 nlk->dst_pid != nlk_sk(ssk)->pid)) { 556 sock_put(sock); 557 return ERR_PTR(-ECONNREFUSED); 558 } 559 return sock; 560 } 561 562 struct sock *netlink_getsockbyfilp(struct file *filp) 563 { 564 struct inode *inode = filp->f_dentry->d_inode; 565 struct sock *sock; 566 567 if (!S_ISSOCK(inode->i_mode)) 568 return ERR_PTR(-ENOTSOCK); 569 570 sock = SOCKET_I(inode)->sk; 571 if (sock->sk_family != AF_NETLINK) 572 return ERR_PTR(-EINVAL); 573 574 sock_hold(sock); 575 return sock; 576 } 577 578 /* 579 * Attach a skb to a netlink socket. 580 * The caller must hold a reference to the destination socket. On error, the 581 * reference is dropped. The skb is not send to the destination, just all 582 * all error checks are performed and memory in the queue is reserved. 583 * Return values: 584 * < 0: error. skb freed, reference to sock dropped. 585 * 0: continue 586 * 1: repeat lookup - reference dropped while waiting for socket memory. 587 */ 588 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo) 589 { 590 struct netlink_sock *nlk; 591 592 nlk = nlk_sk(sk); 593 594 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 595 test_bit(0, &nlk->state)) { 596 DECLARE_WAITQUEUE(wait, current); 597 if (!timeo) { 598 if (!nlk->pid) 599 netlink_overrun(sk); 600 sock_put(sk); 601 kfree_skb(skb); 602 return -EAGAIN; 603 } 604 605 __set_current_state(TASK_INTERRUPTIBLE); 606 add_wait_queue(&nlk->wait, &wait); 607 608 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || 609 test_bit(0, &nlk->state)) && 610 !sock_flag(sk, SOCK_DEAD)) 611 timeo = schedule_timeout(timeo); 612 613 __set_current_state(TASK_RUNNING); 614 remove_wait_queue(&nlk->wait, &wait); 615 sock_put(sk); 616 617 if (signal_pending(current)) { 618 kfree_skb(skb); 619 return sock_intr_errno(timeo); 620 } 621 return 1; 622 } 623 skb_set_owner_r(skb, sk); 624 return 0; 625 } 626 627 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol) 628 { 629 struct netlink_sock *nlk; 630 int len = skb->len; 631 632 nlk = nlk_sk(sk); 633 634 skb_queue_tail(&sk->sk_receive_queue, skb); 635 sk->sk_data_ready(sk, len); 636 sock_put(sk); 637 return len; 638 } 639 640 void netlink_detachskb(struct sock *sk, struct sk_buff *skb) 641 { 642 kfree_skb(skb); 643 sock_put(sk); 644 } 645 646 static inline struct sk_buff *netlink_trim(struct sk_buff *skb, int allocation) 647 { 648 int delta; 649 650 skb_orphan(skb); 651 652 delta = skb->end - skb->tail; 653 if (delta * 2 < skb->truesize) 654 return skb; 655 656 if (skb_shared(skb)) { 657 struct sk_buff *nskb = skb_clone(skb, allocation); 658 if (!nskb) 659 return skb; 660 kfree_skb(skb); 661 skb = nskb; 662 } 663 664 if (!pskb_expand_head(skb, 0, -delta, allocation)) 665 skb->truesize -= delta; 666 667 return skb; 668 } 669 670 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock) 671 { 672 struct sock *sk; 673 int err; 674 long timeo; 675 676 skb = netlink_trim(skb, gfp_any()); 677 678 timeo = sock_sndtimeo(ssk, nonblock); 679 retry: 680 sk = netlink_getsockbypid(ssk, pid); 681 if (IS_ERR(sk)) { 682 kfree_skb(skb); 683 return PTR_ERR(sk); 684 } 685 err = netlink_attachskb(sk, skb, nonblock, timeo); 686 if (err == 1) 687 goto retry; 688 if (err) 689 return err; 690 691 return netlink_sendskb(sk, skb, ssk->sk_protocol); 692 } 693 694 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) 695 { 696 struct netlink_sock *nlk = nlk_sk(sk); 697 698 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && 699 !test_bit(0, &nlk->state)) { 700 skb_set_owner_r(skb, sk); 701 skb_queue_tail(&sk->sk_receive_queue, skb); 702 sk->sk_data_ready(sk, skb->len); 703 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf; 704 } 705 return -1; 706 } 707 708 struct netlink_broadcast_data { 709 struct sock *exclude_sk; 710 u32 pid; 711 u32 group; 712 int failure; 713 int congested; 714 int delivered; 715 int allocation; 716 struct sk_buff *skb, *skb2; 717 }; 718 719 static inline int do_one_broadcast(struct sock *sk, 720 struct netlink_broadcast_data *p) 721 { 722 struct netlink_sock *nlk = nlk_sk(sk); 723 int val; 724 725 if (p->exclude_sk == sk) 726 goto out; 727 728 if (nlk->pid == p->pid || !(nlk->groups & p->group)) 729 goto out; 730 731 if (p->failure) { 732 netlink_overrun(sk); 733 goto out; 734 } 735 736 sock_hold(sk); 737 if (p->skb2 == NULL) { 738 if (atomic_read(&p->skb->users) != 1) { 739 p->skb2 = skb_clone(p->skb, p->allocation); 740 } else { 741 p->skb2 = p->skb; 742 atomic_inc(&p->skb->users); 743 } 744 } 745 if (p->skb2 == NULL) { 746 netlink_overrun(sk); 747 /* Clone failed. Notify ALL listeners. */ 748 p->failure = 1; 749 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) { 750 netlink_overrun(sk); 751 } else { 752 p->congested |= val; 753 p->delivered = 1; 754 p->skb2 = NULL; 755 } 756 sock_put(sk); 757 758 out: 759 return 0; 760 } 761 762 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid, 763 u32 group, int allocation) 764 { 765 struct netlink_broadcast_data info; 766 struct hlist_node *node; 767 struct sock *sk; 768 769 skb = netlink_trim(skb, allocation); 770 771 info.exclude_sk = ssk; 772 info.pid = pid; 773 info.group = group; 774 info.failure = 0; 775 info.congested = 0; 776 info.delivered = 0; 777 info.allocation = allocation; 778 info.skb = skb; 779 info.skb2 = NULL; 780 781 /* While we sleep in clone, do not allow to change socket list */ 782 783 netlink_lock_table(); 784 785 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 786 do_one_broadcast(sk, &info); 787 788 netlink_unlock_table(); 789 790 if (info.skb2) 791 kfree_skb(info.skb2); 792 kfree_skb(skb); 793 794 if (info.delivered) { 795 if (info.congested && (allocation & __GFP_WAIT)) 796 yield(); 797 return 0; 798 } 799 if (info.failure) 800 return -ENOBUFS; 801 return -ESRCH; 802 } 803 804 struct netlink_set_err_data { 805 struct sock *exclude_sk; 806 u32 pid; 807 u32 group; 808 int code; 809 }; 810 811 static inline int do_one_set_err(struct sock *sk, 812 struct netlink_set_err_data *p) 813 { 814 struct netlink_sock *nlk = nlk_sk(sk); 815 816 if (sk == p->exclude_sk) 817 goto out; 818 819 if (nlk->pid == p->pid || !(nlk->groups & p->group)) 820 goto out; 821 822 sk->sk_err = p->code; 823 sk->sk_error_report(sk); 824 out: 825 return 0; 826 } 827 828 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code) 829 { 830 struct netlink_set_err_data info; 831 struct hlist_node *node; 832 struct sock *sk; 833 834 info.exclude_sk = ssk; 835 info.pid = pid; 836 info.group = group; 837 info.code = code; 838 839 read_lock(&nl_table_lock); 840 841 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list) 842 do_one_set_err(sk, &info); 843 844 read_unlock(&nl_table_lock); 845 } 846 847 static inline void netlink_rcv_wake(struct sock *sk) 848 { 849 struct netlink_sock *nlk = nlk_sk(sk); 850 851 if (!skb_queue_len(&sk->sk_receive_queue)) 852 clear_bit(0, &nlk->state); 853 if (!test_bit(0, &nlk->state)) 854 wake_up_interruptible(&nlk->wait); 855 } 856 857 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock, 858 struct msghdr *msg, size_t len) 859 { 860 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 861 struct sock *sk = sock->sk; 862 struct netlink_sock *nlk = nlk_sk(sk); 863 struct sockaddr_nl *addr=msg->msg_name; 864 u32 dst_pid; 865 u32 dst_groups; 866 struct sk_buff *skb; 867 int err; 868 struct scm_cookie scm; 869 870 if (msg->msg_flags&MSG_OOB) 871 return -EOPNOTSUPP; 872 873 if (NULL == siocb->scm) 874 siocb->scm = &scm; 875 err = scm_send(sock, msg, siocb->scm); 876 if (err < 0) 877 return err; 878 879 if (msg->msg_namelen) { 880 if (addr->nl_family != AF_NETLINK) 881 return -EINVAL; 882 dst_pid = addr->nl_pid; 883 dst_groups = addr->nl_groups; 884 if (dst_groups && !netlink_capable(sock, NL_NONROOT_SEND)) 885 return -EPERM; 886 } else { 887 dst_pid = nlk->dst_pid; 888 dst_groups = nlk->dst_groups; 889 } 890 891 if (!nlk->pid) { 892 err = netlink_autobind(sock); 893 if (err) 894 goto out; 895 } 896 897 err = -EMSGSIZE; 898 if (len > sk->sk_sndbuf - 32) 899 goto out; 900 err = -ENOBUFS; 901 skb = alloc_skb(len, GFP_KERNEL); 902 if (skb==NULL) 903 goto out; 904 905 NETLINK_CB(skb).pid = nlk->pid; 906 NETLINK_CB(skb).groups = nlk->groups; 907 NETLINK_CB(skb).dst_pid = dst_pid; 908 NETLINK_CB(skb).dst_groups = dst_groups; 909 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); 910 911 /* What can I do? Netlink is asynchronous, so that 912 we will have to save current capabilities to 913 check them, when this message will be delivered 914 to corresponding kernel module. --ANK (980802) 915 */ 916 917 err = -EFAULT; 918 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) { 919 kfree_skb(skb); 920 goto out; 921 } 922 923 err = security_netlink_send(sk, skb); 924 if (err) { 925 kfree_skb(skb); 926 goto out; 927 } 928 929 if (dst_groups) { 930 atomic_inc(&skb->users); 931 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL); 932 } 933 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT); 934 935 out: 936 return err; 937 } 938 939 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock, 940 struct msghdr *msg, size_t len, 941 int flags) 942 { 943 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 944 struct scm_cookie scm; 945 struct sock *sk = sock->sk; 946 struct netlink_sock *nlk = nlk_sk(sk); 947 int noblock = flags&MSG_DONTWAIT; 948 size_t copied; 949 struct sk_buff *skb; 950 int err; 951 952 if (flags&MSG_OOB) 953 return -EOPNOTSUPP; 954 955 copied = 0; 956 957 skb = skb_recv_datagram(sk,flags,noblock,&err); 958 if (skb==NULL) 959 goto out; 960 961 msg->msg_namelen = 0; 962 963 copied = skb->len; 964 if (len < copied) { 965 msg->msg_flags |= MSG_TRUNC; 966 copied = len; 967 } 968 969 skb->h.raw = skb->data; 970 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 971 972 if (msg->msg_name) { 973 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name; 974 addr->nl_family = AF_NETLINK; 975 addr->nl_pad = 0; 976 addr->nl_pid = NETLINK_CB(skb).pid; 977 addr->nl_groups = NETLINK_CB(skb).dst_groups; 978 msg->msg_namelen = sizeof(*addr); 979 } 980 981 if (NULL == siocb->scm) { 982 memset(&scm, 0, sizeof(scm)); 983 siocb->scm = &scm; 984 } 985 siocb->scm->creds = *NETLINK_CREDS(skb); 986 skb_free_datagram(sk, skb); 987 988 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) 989 netlink_dump(sk); 990 991 scm_recv(sock, msg, siocb->scm, flags); 992 993 out: 994 netlink_rcv_wake(sk); 995 return err ? : copied; 996 } 997 998 static void netlink_data_ready(struct sock *sk, int len) 999 { 1000 struct netlink_sock *nlk = nlk_sk(sk); 1001 1002 if (nlk->data_ready) 1003 nlk->data_ready(sk, len); 1004 netlink_rcv_wake(sk); 1005 } 1006 1007 /* 1008 * We export these functions to other modules. They provide a 1009 * complete set of kernel non-blocking support for message 1010 * queueing. 1011 */ 1012 1013 struct sock * 1014 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len)) 1015 { 1016 struct socket *sock; 1017 struct sock *sk; 1018 1019 if (!nl_table) 1020 return NULL; 1021 1022 if (unit<0 || unit>=MAX_LINKS) 1023 return NULL; 1024 1025 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) 1026 return NULL; 1027 1028 if (netlink_create(sock, unit) < 0) { 1029 sock_release(sock); 1030 return NULL; 1031 } 1032 sk = sock->sk; 1033 sk->sk_data_ready = netlink_data_ready; 1034 if (input) 1035 nlk_sk(sk)->data_ready = input; 1036 1037 if (netlink_insert(sk, 0)) { 1038 sock_release(sock); 1039 return NULL; 1040 } 1041 return sk; 1042 } 1043 1044 void netlink_set_nonroot(int protocol, unsigned int flags) 1045 { 1046 if ((unsigned int)protocol < MAX_LINKS) 1047 nl_table[protocol].nl_nonroot = flags; 1048 } 1049 1050 static void netlink_destroy_callback(struct netlink_callback *cb) 1051 { 1052 if (cb->skb) 1053 kfree_skb(cb->skb); 1054 kfree(cb); 1055 } 1056 1057 /* 1058 * It looks a bit ugly. 1059 * It would be better to create kernel thread. 1060 */ 1061 1062 static int netlink_dump(struct sock *sk) 1063 { 1064 struct netlink_sock *nlk = nlk_sk(sk); 1065 struct netlink_callback *cb; 1066 struct sk_buff *skb; 1067 struct nlmsghdr *nlh; 1068 int len; 1069 1070 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL); 1071 if (!skb) 1072 return -ENOBUFS; 1073 1074 spin_lock(&nlk->cb_lock); 1075 1076 cb = nlk->cb; 1077 if (cb == NULL) { 1078 spin_unlock(&nlk->cb_lock); 1079 kfree_skb(skb); 1080 return -EINVAL; 1081 } 1082 1083 len = cb->dump(skb, cb); 1084 1085 if (len > 0) { 1086 spin_unlock(&nlk->cb_lock); 1087 skb_queue_tail(&sk->sk_receive_queue, skb); 1088 sk->sk_data_ready(sk, len); 1089 return 0; 1090 } 1091 1092 nlh = __nlmsg_put(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, NLMSG_DONE, sizeof(int)); 1093 nlh->nlmsg_flags |= NLM_F_MULTI; 1094 memcpy(NLMSG_DATA(nlh), &len, sizeof(len)); 1095 skb_queue_tail(&sk->sk_receive_queue, skb); 1096 sk->sk_data_ready(sk, skb->len); 1097 1098 cb->done(cb); 1099 nlk->cb = NULL; 1100 spin_unlock(&nlk->cb_lock); 1101 1102 netlink_destroy_callback(cb); 1103 __sock_put(sk); 1104 return 0; 1105 } 1106 1107 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb, 1108 struct nlmsghdr *nlh, 1109 int (*dump)(struct sk_buff *skb, struct netlink_callback*), 1110 int (*done)(struct netlink_callback*)) 1111 { 1112 struct netlink_callback *cb; 1113 struct sock *sk; 1114 struct netlink_sock *nlk; 1115 1116 cb = kmalloc(sizeof(*cb), GFP_KERNEL); 1117 if (cb == NULL) 1118 return -ENOBUFS; 1119 1120 memset(cb, 0, sizeof(*cb)); 1121 cb->dump = dump; 1122 cb->done = done; 1123 cb->nlh = nlh; 1124 atomic_inc(&skb->users); 1125 cb->skb = skb; 1126 1127 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid); 1128 if (sk == NULL) { 1129 netlink_destroy_callback(cb); 1130 return -ECONNREFUSED; 1131 } 1132 nlk = nlk_sk(sk); 1133 /* A dump is in progress... */ 1134 spin_lock(&nlk->cb_lock); 1135 if (nlk->cb) { 1136 spin_unlock(&nlk->cb_lock); 1137 netlink_destroy_callback(cb); 1138 sock_put(sk); 1139 return -EBUSY; 1140 } 1141 nlk->cb = cb; 1142 sock_hold(sk); 1143 spin_unlock(&nlk->cb_lock); 1144 1145 netlink_dump(sk); 1146 sock_put(sk); 1147 return 0; 1148 } 1149 1150 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err) 1151 { 1152 struct sk_buff *skb; 1153 struct nlmsghdr *rep; 1154 struct nlmsgerr *errmsg; 1155 int size; 1156 1157 if (err == 0) 1158 size = NLMSG_SPACE(sizeof(struct nlmsgerr)); 1159 else 1160 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len)); 1161 1162 skb = alloc_skb(size, GFP_KERNEL); 1163 if (!skb) { 1164 struct sock *sk; 1165 1166 sk = netlink_lookup(in_skb->sk->sk_protocol, 1167 NETLINK_CB(in_skb).pid); 1168 if (sk) { 1169 sk->sk_err = ENOBUFS; 1170 sk->sk_error_report(sk); 1171 sock_put(sk); 1172 } 1173 return; 1174 } 1175 1176 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, 1177 NLMSG_ERROR, sizeof(struct nlmsgerr)); 1178 errmsg = NLMSG_DATA(rep); 1179 errmsg->error = err; 1180 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr)); 1181 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT); 1182 } 1183 1184 1185 #ifdef CONFIG_PROC_FS 1186 struct nl_seq_iter { 1187 int link; 1188 int hash_idx; 1189 }; 1190 1191 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos) 1192 { 1193 struct nl_seq_iter *iter = seq->private; 1194 int i, j; 1195 struct sock *s; 1196 struct hlist_node *node; 1197 loff_t off = 0; 1198 1199 for (i=0; i<MAX_LINKS; i++) { 1200 struct nl_pid_hash *hash = &nl_table[i].hash; 1201 1202 for (j = 0; j <= hash->mask; j++) { 1203 sk_for_each(s, node, &hash->table[j]) { 1204 if (off == pos) { 1205 iter->link = i; 1206 iter->hash_idx = j; 1207 return s; 1208 } 1209 ++off; 1210 } 1211 } 1212 } 1213 return NULL; 1214 } 1215 1216 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos) 1217 { 1218 read_lock(&nl_table_lock); 1219 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN; 1220 } 1221 1222 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1223 { 1224 struct sock *s; 1225 struct nl_seq_iter *iter; 1226 int i, j; 1227 1228 ++*pos; 1229 1230 if (v == SEQ_START_TOKEN) 1231 return netlink_seq_socket_idx(seq, 0); 1232 1233 s = sk_next(v); 1234 if (s) 1235 return s; 1236 1237 iter = seq->private; 1238 i = iter->link; 1239 j = iter->hash_idx + 1; 1240 1241 do { 1242 struct nl_pid_hash *hash = &nl_table[i].hash; 1243 1244 for (; j <= hash->mask; j++) { 1245 s = sk_head(&hash->table[j]); 1246 if (s) { 1247 iter->link = i; 1248 iter->hash_idx = j; 1249 return s; 1250 } 1251 } 1252 1253 j = 0; 1254 } while (++i < MAX_LINKS); 1255 1256 return NULL; 1257 } 1258 1259 static void netlink_seq_stop(struct seq_file *seq, void *v) 1260 { 1261 read_unlock(&nl_table_lock); 1262 } 1263 1264 1265 static int netlink_seq_show(struct seq_file *seq, void *v) 1266 { 1267 if (v == SEQ_START_TOKEN) 1268 seq_puts(seq, 1269 "sk Eth Pid Groups " 1270 "Rmem Wmem Dump Locks\n"); 1271 else { 1272 struct sock *s = v; 1273 struct netlink_sock *nlk = nlk_sk(s); 1274 1275 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n", 1276 s, 1277 s->sk_protocol, 1278 nlk->pid, 1279 nlk->groups, 1280 atomic_read(&s->sk_rmem_alloc), 1281 atomic_read(&s->sk_wmem_alloc), 1282 nlk->cb, 1283 atomic_read(&s->sk_refcnt) 1284 ); 1285 1286 } 1287 return 0; 1288 } 1289 1290 static struct seq_operations netlink_seq_ops = { 1291 .start = netlink_seq_start, 1292 .next = netlink_seq_next, 1293 .stop = netlink_seq_stop, 1294 .show = netlink_seq_show, 1295 }; 1296 1297 1298 static int netlink_seq_open(struct inode *inode, struct file *file) 1299 { 1300 struct seq_file *seq; 1301 struct nl_seq_iter *iter; 1302 int err; 1303 1304 iter = kmalloc(sizeof(*iter), GFP_KERNEL); 1305 if (!iter) 1306 return -ENOMEM; 1307 1308 err = seq_open(file, &netlink_seq_ops); 1309 if (err) { 1310 kfree(iter); 1311 return err; 1312 } 1313 1314 memset(iter, 0, sizeof(*iter)); 1315 seq = file->private_data; 1316 seq->private = iter; 1317 return 0; 1318 } 1319 1320 static struct file_operations netlink_seq_fops = { 1321 .owner = THIS_MODULE, 1322 .open = netlink_seq_open, 1323 .read = seq_read, 1324 .llseek = seq_lseek, 1325 .release = seq_release_private, 1326 }; 1327 1328 #endif 1329 1330 int netlink_register_notifier(struct notifier_block *nb) 1331 { 1332 return notifier_chain_register(&netlink_chain, nb); 1333 } 1334 1335 int netlink_unregister_notifier(struct notifier_block *nb) 1336 { 1337 return notifier_chain_unregister(&netlink_chain, nb); 1338 } 1339 1340 static struct proto_ops netlink_ops = { 1341 .family = PF_NETLINK, 1342 .owner = THIS_MODULE, 1343 .release = netlink_release, 1344 .bind = netlink_bind, 1345 .connect = netlink_connect, 1346 .socketpair = sock_no_socketpair, 1347 .accept = sock_no_accept, 1348 .getname = netlink_getname, 1349 .poll = datagram_poll, 1350 .ioctl = sock_no_ioctl, 1351 .listen = sock_no_listen, 1352 .shutdown = sock_no_shutdown, 1353 .setsockopt = sock_no_setsockopt, 1354 .getsockopt = sock_no_getsockopt, 1355 .sendmsg = netlink_sendmsg, 1356 .recvmsg = netlink_recvmsg, 1357 .mmap = sock_no_mmap, 1358 .sendpage = sock_no_sendpage, 1359 }; 1360 1361 static struct net_proto_family netlink_family_ops = { 1362 .family = PF_NETLINK, 1363 .create = netlink_create, 1364 .owner = THIS_MODULE, /* for consistency 8) */ 1365 }; 1366 1367 extern void netlink_skb_parms_too_large(void); 1368 1369 static int __init netlink_proto_init(void) 1370 { 1371 struct sk_buff *dummy_skb; 1372 int i; 1373 unsigned long max; 1374 unsigned int order; 1375 int err = proto_register(&netlink_proto, 0); 1376 1377 if (err != 0) 1378 goto out; 1379 1380 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb)) 1381 netlink_skb_parms_too_large(); 1382 1383 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL); 1384 if (!nl_table) { 1385 enomem: 1386 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n"); 1387 return -ENOMEM; 1388 } 1389 1390 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS); 1391 1392 if (num_physpages >= (128 * 1024)) 1393 max = num_physpages >> (21 - PAGE_SHIFT); 1394 else 1395 max = num_physpages >> (23 - PAGE_SHIFT); 1396 1397 order = get_bitmask_order(max) - 1 + PAGE_SHIFT; 1398 max = (1UL << order) / sizeof(struct hlist_head); 1399 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1; 1400 1401 for (i = 0; i < MAX_LINKS; i++) { 1402 struct nl_pid_hash *hash = &nl_table[i].hash; 1403 1404 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table)); 1405 if (!hash->table) { 1406 while (i-- > 0) 1407 nl_pid_hash_free(nl_table[i].hash.table, 1408 1 * sizeof(*hash->table)); 1409 kfree(nl_table); 1410 goto enomem; 1411 } 1412 memset(hash->table, 0, 1 * sizeof(*hash->table)); 1413 hash->max_shift = order; 1414 hash->shift = 0; 1415 hash->mask = 0; 1416 hash->rehash_time = jiffies; 1417 } 1418 1419 sock_register(&netlink_family_ops); 1420 #ifdef CONFIG_PROC_FS 1421 proc_net_fops_create("netlink", 0, &netlink_seq_fops); 1422 #endif 1423 /* The netlink device handler may be needed early. */ 1424 rtnetlink_init(); 1425 out: 1426 return err; 1427 } 1428 1429 static void __exit netlink_proto_exit(void) 1430 { 1431 sock_unregister(PF_NETLINK); 1432 proc_net_remove("netlink"); 1433 kfree(nl_table); 1434 nl_table = NULL; 1435 proto_unregister(&netlink_proto); 1436 } 1437 1438 core_initcall(netlink_proto_init); 1439 module_exit(netlink_proto_exit); 1440 1441 MODULE_LICENSE("GPL"); 1442 1443 MODULE_ALIAS_NETPROTO(PF_NETLINK); 1444 1445 EXPORT_SYMBOL(netlink_ack); 1446 EXPORT_SYMBOL(netlink_broadcast); 1447 EXPORT_SYMBOL(netlink_dump_start); 1448 EXPORT_SYMBOL(netlink_kernel_create); 1449 EXPORT_SYMBOL(netlink_register_notifier); 1450 EXPORT_SYMBOL(netlink_set_err); 1451 EXPORT_SYMBOL(netlink_set_nonroot); 1452 EXPORT_SYMBOL(netlink_unicast); 1453 EXPORT_SYMBOL(netlink_unregister_notifier); 1454 1455