net/netlink/af_netlink.c

   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 <net/sock.h>
  48 #include <net/scm.h>
  49
  50 #define Nprintk(a...)
  51
  52 #if defined(CONFIG_NETLINK_DEV) || defined(CONFIG_NETLINK_DEV_MODULE)
  53 #define NL_EMULATE_DEV
  54 #endif
  55
  56 struct netlink_opt
  57 {
  58         u32                     pid;
  59         unsigned                groups;
  60         u32                     dst_pid;
  61         unsigned                dst_groups;
  62         unsigned long           state;
  63         int                     (*handler)(int unit, struct sk_buff *skb);
  64         wait_queue_head_t       wait;
  65         struct netlink_callback *cb;
  66         spinlock_t              cb_lock;
  67         void                    (*data_ready)(struct sock *sk, int bytes);
  68 };
  69
  70 #define nlk_sk(__sk) ((struct netlink_opt *)(__sk)->sk_protinfo)
  71
  72 static struct hlist_head nl_table[MAX_LINKS];
  73 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
  74 static unsigned nl_nonroot[MAX_LINKS];
  75
  76 #ifdef NL_EMULATE_DEV
  77 static struct socket *netlink_kernel[MAX_LINKS];
  78 #endif
  79
  80 static int netlink_dump(struct sock *sk);
  81 static void netlink_destroy_callback(struct netlink_callback *cb);
  82
  83 atomic_t netlink_sock_nr;
  84
  85 static rwlock_t nl_table_lock = RW_LOCK_UNLOCKED;
  86 static atomic_t nl_table_users = ATOMIC_INIT(0);
  87
  88 static struct notifier_block *netlink_chain;
  89
  90 static void netlink_sock_destruct(struct sock *sk)
  91 {
  92         skb_queue_purge(&sk->sk_receive_queue);
  93
  94         if (!sock_flag(sk, SOCK_DEAD)) {
  95                 printk("Freeing alive netlink socket %p\n", sk);
  96                 return;
  97         }
  98         BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
  99         BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
 100         BUG_TRAP(!nlk_sk(sk)->cb);
 101
 102         kfree(nlk_sk(sk));
 103
 104         atomic_dec(&netlink_sock_nr);
 105 #ifdef NETLINK_REFCNT_DEBUG
 106         printk(KERN_DEBUG "NETLINK %p released, %d are still alive\n", sk, atomic_read(&netlink_sock_nr));
 107 #endif
 108 }
 109
 110 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
 111  * Look, when several writers sleep and reader wakes them up, all but one
 112  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
 113  * this, _but_ remember, it adds useless work on UP machines.
 114  */
 115
 116 static void netlink_table_grab(void)
 117 {
 118         write_lock_bh(&nl_table_lock);
 119
 120         if (atomic_read(&nl_table_users)) {
 121                 DECLARE_WAITQUEUE(wait, current);
 122
 123                 add_wait_queue_exclusive(&nl_table_wait, &wait);
 124                 for(;;) {
 125                         set_current_state(TASK_UNINTERRUPTIBLE);
 126                         if (atomic_read(&nl_table_users) == 0)
 127                                 break;
 128                         write_unlock_bh(&nl_table_lock);
 129                         schedule();
 130                         write_lock_bh(&nl_table_lock);
 131                 }
 132
 133                 __set_current_state(TASK_RUNNING);
 134                 remove_wait_queue(&nl_table_wait, &wait);
 135         }
 136 }
 137
 138 static __inline__ void netlink_table_ungrab(void)
 139 {
 140         write_unlock_bh(&nl_table_lock);
 141         wake_up(&nl_table_wait);
 142 }
 143
 144 static __inline__ void
 145 netlink_lock_table(void)
 146 {
 147         /* read_lock() synchronizes us to netlink_table_grab */
 148
 149         read_lock(&nl_table_lock);
 150         atomic_inc(&nl_table_users);
 151         read_unlock(&nl_table_lock);
 152 }
 153
 154 static __inline__ void
 155 netlink_unlock_table(void)
 156 {
 157         if (atomic_dec_and_test(&nl_table_users))
 158                 wake_up(&nl_table_wait);
 159 }
 160
 161 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
 162 {
 163         struct sock *sk;
 164         struct hlist_node *node;
 165
 166         read_lock(&nl_table_lock);
 167         sk_for_each(sk, node, &nl_table[protocol]) {
 168                 if (nlk_sk(sk)->pid == pid) {
 169                         sock_hold(sk);
 170                         goto found;
 171                 }
 172         }
 173         sk = NULL;
 174 found:
 175         read_unlock(&nl_table_lock);
 176         return sk;
 177 }
 178
 179 static struct proto_ops netlink_ops;
 180
 181 static int netlink_insert(struct sock *sk, u32 pid)
 182 {
 183         int err = -EADDRINUSE;
 184         struct sock *osk;
 185         struct hlist_node *node;
 186
 187         netlink_table_grab();
 188         sk_for_each(osk, node, &nl_table[sk->sk_protocol]) {
 189                 if (nlk_sk(osk)->pid == pid)
 190                         break;
 191         }
 192         if (!node) {
 193                 err = -EBUSY;
 194                 if (nlk_sk(sk)->pid == 0) {
 195                         nlk_sk(sk)->pid = pid;
 196                         sk_add_node(sk, &nl_table[sk->sk_protocol]);
 197                         err = 0;
 198                 }
 199         }
 200         netlink_table_ungrab();
 201         return err;
 202 }
 203
 204 static void netlink_remove(struct sock *sk)
 205 {
 206         netlink_table_grab();
 207         sk_del_node_init(sk);
 208         netlink_table_ungrab();
 209 }
 210
 211 static int netlink_create(struct socket *sock, int protocol)
 212 {
 213         struct sock *sk;
 214         struct netlink_opt *nlk;
 215
 216         sock->state = SS_UNCONNECTED;
 217
 218         if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 219                 return -ESOCKTNOSUPPORT;
 220
 221         if (protocol<0 || protocol >= MAX_LINKS)
 222                 return -EPROTONOSUPPORT;
 223
 224         sock->ops = &netlink_ops;
 225
 226         sk = sk_alloc(PF_NETLINK, GFP_KERNEL, 1, NULL);
 227         if (!sk)
 228                 return -ENOMEM;
 229
 230         sock_init_data(sock,sk);
 231         sk_set_owner(sk, THIS_MODULE);
 232
 233         nlk = sk->sk_protinfo = kmalloc(sizeof(*nlk), GFP_KERNEL);
 234         if (!nlk) {
 235                 sk_free(sk);
 236                 return -ENOMEM;
 237         }
 238         memset(nlk, 0, sizeof(*nlk));
 239
 240         spin_lock_init(&nlk->cb_lock);
 241         init_waitqueue_head(&nlk->wait);
 242         sk->sk_destruct = netlink_sock_destruct;
 243         atomic_inc(&netlink_sock_nr);
 244
 245         sk->sk_protocol = protocol;
 246         return 0;
 247 }
 248
 249 static int netlink_release(struct socket *sock)
 250 {
 251         struct sock *sk = sock->sk;
 252         struct netlink_opt *nlk;
 253
 254         if (!sk)
 255                 return 0;
 256
 257         netlink_remove(sk);
 258         nlk = nlk_sk(sk);
 259
 260         spin_lock(&nlk->cb_lock);
 261         if (nlk->cb) {
 262                 nlk->cb->done(nlk->cb);
 263                 netlink_destroy_callback(nlk->cb);
 264                 nlk->cb = NULL;
 265                 __sock_put(sk);
 266         }
 267         spin_unlock(&nlk->cb_lock);
 268
 269         /* OK. Socket is unlinked, and, therefore,
 270            no new packets will arrive */
 271
 272         sock_orphan(sk);
 273         sock->sk = NULL;
 274         wake_up_interruptible_all(&nlk->wait);
 275
 276         skb_queue_purge(&sk->sk_write_queue);
 277
 278         if (nlk->pid && !nlk->groups) {
 279                 struct netlink_notify n = {
 280                                                 .protocol = sk->sk_protocol,
 281                                                 .pid = nlk->pid,
 282                                           };
 283                 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
 284         }
 285
 286         sock_put(sk);
 287         return 0;
 288 }
 289
 290 static int netlink_autobind(struct socket *sock)
 291 {
 292         struct sock *sk = sock->sk;
 293         struct sock *osk;
 294         struct hlist_node *node;
 295         s32 pid = current->pid;
 296         int err;
 297
 298 retry:
 299         netlink_table_grab();
 300         sk_for_each(osk, node, &nl_table[sk->sk_protocol]) {
 301                 if (nlk_sk(osk)->pid == pid) {
 302                         /* Bind collision, search negative pid values. */
 303                         if (pid > 0)
 304                                 pid = -4096;
 305                         pid--;
 306                         netlink_table_ungrab();
 307                         goto retry;
 308                 }
 309         }
 310         netlink_table_ungrab();
 311
 312         err = netlink_insert(sk, pid);
 313         if (err == -EADDRINUSE)
 314                 goto retry;
 315         nlk_sk(sk)->groups = 0;
 316         return 0;
 317 }
 318
 319 static inline int netlink_capable(struct socket *sock, unsigned flag)
 320 {
 321         return (nl_nonroot[sock->sk->sk_protocol] & flag) ||
 322                capable(CAP_NET_ADMIN);
 323 }
 324
 325 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 326 {
 327         struct sock *sk = sock->sk;
 328         struct netlink_opt *nlk = nlk_sk(sk);
 329         struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
 330         int err;
 331
 332         if (nladdr->nl_family != AF_NETLINK)
 333                 return -EINVAL;
 334
 335         /* Only superuser is allowed to listen multicasts */
 336         if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV))
 337                 return -EPERM;
 338
 339         if (nlk->pid) {
 340                 if (nladdr->nl_pid != nlk->pid)
 341                         return -EINVAL;
 342                 nlk->groups = nladdr->nl_groups;
 343                 return 0;
 344         }
 345
 346         if (nladdr->nl_pid == 0) {
 347                 err = netlink_autobind(sock);
 348                 if (err == 0)
 349                         nlk->groups = nladdr->nl_groups;
 350                 return err;
 351         }
 352
 353         err = netlink_insert(sk, nladdr->nl_pid);
 354         if (err == 0)
 355                 nlk->groups = nladdr->nl_groups;
 356         return err;
 357 }
 358
 359 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
 360                            int alen, int flags)
 361 {
 362         int err = 0;
 363         struct sock *sk = sock->sk;
 364         struct netlink_opt *nlk = nlk_sk(sk);
 365         struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
 366
 367         if (addr->sa_family == AF_UNSPEC) {
 368                 sk->sk_state    = NETLINK_UNCONNECTED;
 369                 nlk->dst_pid    = 0;
 370                 nlk->dst_groups = 0;
 371                 return 0;
 372         }
 373         if (addr->sa_family != AF_NETLINK)
 374                 return -EINVAL;
 375
 376         /* Only superuser is allowed to send multicasts */
 377         if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
 378                 return -EPERM;
 379
 380         if (!nlk->pid)
 381                 err = netlink_autobind(sock);
 382
 383         if (err == 0) {
 384                 sk->sk_state    = NETLINK_CONNECTED;
 385                 nlk->dst_pid    = nladdr->nl_pid;
 386                 nlk->dst_groups = nladdr->nl_groups;
 387         }
 388
 389         return err;
 390 }
 391
 392 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
 393 {
 394         struct sock *sk = sock->sk;
 395         struct netlink_opt *nlk = nlk_sk(sk);
 396         struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
 397
 398         nladdr->nl_family = AF_NETLINK;
 399         nladdr->nl_pad = 0;
 400         *addr_len = sizeof(*nladdr);
 401
 402         if (peer) {
 403                 nladdr->nl_pid = nlk->dst_pid;
 404                 nladdr->nl_groups = nlk->dst_groups;
 405         } else {
 406                 nladdr->nl_pid = nlk->pid;
 407                 nladdr->nl_groups = nlk->groups;
 408         }
 409         return 0;
 410 }
 411
 412 static void netlink_overrun(struct sock *sk)
 413 {
 414         if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
 415                 sk->sk_err = ENOBUFS;
 416                 sk->sk_error_report(sk);
 417         }
 418 }
 419
 420 struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
 421 {
 422         int protocol = ssk->sk_protocol;
 423         struct sock *sock;
 424         struct netlink_opt *nlk;
 425
 426         sock = netlink_lookup(protocol, pid);
 427         if (!sock)
 428                 return ERR_PTR(-ECONNREFUSED);
 429
 430         /* Don't bother queuing skb if kernel socket has no input function */
 431         nlk = nlk_sk(sock);
 432         if ((nlk->pid == 0 && !nlk->data_ready) ||
 433             (sock->sk_state == NETLINK_CONNECTED &&
 434              nlk->dst_pid != nlk_sk(ssk)->pid)) {
 435                 sock_put(sock);
 436                 return ERR_PTR(-ECONNREFUSED);
 437         }
 438         return sock;
 439 }
 440
 441 struct sock *netlink_getsockbyfilp(struct file *filp)
 442 {
 443         struct inode *inode = filp->f_dentry->d_inode;
 444         struct socket *socket;
 445         struct sock *sock;
 446
 447         if (!inode->i_sock || !(socket = SOCKET_I(inode)))
 448                 return ERR_PTR(-ENOTSOCK);
 449
 450         sock = socket->sk;
 451         if (sock->sk_family != AF_NETLINK)
 452                 return ERR_PTR(-EINVAL);
 453
 454         sock_hold(sock);
 455         return sock;
 456 }
 457
 458 /*
 459  * Attach a skb to a netlink socket.
 460  * The caller must hold a reference to the destination socket. On error, the
 461  * reference is dropped. The skb is not send to the destination, just all
 462  * all error checks are performed and memory in the queue is reserved.
 463  * Return values:
 464  * < 0: error. skb freed, reference to sock dropped.
 465  * 0: continue
 466  * 1: repeat lookup - reference dropped while waiting for socket memory.
 467  */
 468 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo)
 469 {
 470         struct netlink_opt *nlk;
 471
 472         nlk = nlk_sk(sk);
 473
 474 #ifdef NL_EMULATE_DEV
 475         if (nlk->handler)
 476                 return 0;
 477 #endif
 478         if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
 479             test_bit(0, &nlk->state)) {
 480                 DECLARE_WAITQUEUE(wait, current);
 481                 if (!timeo) {
 482                         if (!nlk->pid)
 483                                 netlink_overrun(sk);
 484                         sock_put(sk);
 485                         kfree_skb(skb);
 486                         return -EAGAIN;
 487                 }
 488
 489                 __set_current_state(TASK_INTERRUPTIBLE);
 490                 add_wait_queue(&nlk->wait, &wait);
 491
 492                 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
 493                      test_bit(0, &nlk->state)) &&
 494                     !sock_flag(sk, SOCK_DEAD))
 495                         timeo = schedule_timeout(timeo);
 496
 497                 __set_current_state(TASK_RUNNING);
 498                 remove_wait_queue(&nlk->wait, &wait);
 499                 sock_put(sk);
 500
 501                 if (signal_pending(current)) {
 502                         kfree_skb(skb);
 503                         return sock_intr_errno(timeo);
 504                 }
 505                 return 1;
 506         }
 507         skb_orphan(skb);
 508         skb_set_owner_r(skb, sk);
 509         return 0;
 510 }
 511
 512 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
 513 {
 514         struct netlink_opt *nlk;
 515         int len = skb->len;
 516
 517         nlk = nlk_sk(sk);
 518 #ifdef NL_EMULATE_DEV
 519         if (nlk->handler) {
 520                 skb_orphan(skb);
 521                 len = nlk->handler(protocol, skb);
 522                 sock_put(sk);
 523                 return len;
 524         }
 525 #endif
 526
 527         skb_queue_tail(&sk->sk_receive_queue, skb);
 528         sk->sk_data_ready(sk, len);
 529         sock_put(sk);
 530         return len;
 531 }
 532
 533 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
 534 {
 535         kfree_skb(skb);
 536         sock_put(sk);
 537 }
 538
 539 static inline void netlink_trim(struct sk_buff *skb, int allocation)
 540 {
 541         int delta = skb->end - skb->tail;
 542
 543         /* If the packet is charged to a socket, the modification
 544          * of truesize below is illegal and will corrupt socket
 545          * buffer accounting state.
 546          */
 547         BUG_ON(skb->list != NULL);
 548
 549         if (delta * 2 < skb->truesize)
 550                 return;
 551         if (pskb_expand_head(skb, 0, -delta, allocation))
 552                 return;
 553         skb->truesize -= delta;
 554 }
 555
 556 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
 557 {
 558         struct sock *sk;
 559         int err;
 560         long timeo;
 561
 562         netlink_trim(skb, gfp_any());
 563
 564         timeo = sock_sndtimeo(ssk, nonblock);
 565 retry:
 566         sk = netlink_getsockbypid(ssk, pid);
 567         if (IS_ERR(sk)) {
 568                 kfree_skb(skb);
 569                 return PTR_ERR(sk);
 570         }
 571         err = netlink_attachskb(sk, skb, nonblock, timeo);
 572         if (err == 1)
 573                 goto retry;
 574         if (err)
 575                 return err;
 576
 577         return netlink_sendskb(sk, skb, ssk->sk_protocol);
 578 }
 579
 580 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
 581 {
 582         struct netlink_opt *nlk = nlk_sk(sk);
 583 #ifdef NL_EMULATE_DEV
 584         if (nlk->handler) {
 585                 skb_orphan(skb);
 586                 nlk->handler(sk->sk_protocol, skb);
 587                 return 0;
 588         } else
 589 #endif
 590         if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
 591             !test_bit(0, &nlk->state)) {
 592                 skb_orphan(skb);
 593                 skb_set_owner_r(skb, sk);
 594                 skb_queue_tail(&sk->sk_receive_queue, skb);
 595                 sk->sk_data_ready(sk, skb->len);
 596                 return 0;
 597         }
 598         return -1;
 599 }
 600
 601 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
 602                       u32 group, int allocation)
 603 {
 604         struct sock *sk;
 605         struct hlist_node *node;
 606         struct sk_buff *skb2 = NULL;
 607         int protocol = ssk->sk_protocol;
 608         int failure = 0, delivered = 0;
 609
 610         netlink_trim(skb, allocation);
 611
 612         /* While we sleep in clone, do not allow to change socket list */
 613
 614         netlink_lock_table();
 615
 616         sk_for_each(sk, node, &nl_table[protocol]) {
 617                 struct netlink_opt *nlk = nlk_sk(sk);
 618
 619                 if (ssk == sk)
 620                         continue;
 621
 622                 if (nlk->pid == pid || !(nlk->groups & group))
 623                         continue;
 624
 625                 if (failure) {
 626                         netlink_overrun(sk);
 627                         continue;
 628                 }
 629
 630                 sock_hold(sk);
 631                 if (skb2 == NULL) {
 632                         if (atomic_read(&skb->users) != 1) {
 633                                 skb2 = skb_clone(skb, allocation);
 634                         } else {
 635                                 skb2 = skb;
 636                                 atomic_inc(&skb->users);
 637                         }
 638                 }
 639                 if (skb2 == NULL) {
 640                         netlink_overrun(sk);
 641                         /* Clone failed. Notify ALL listeners. */
 642                         failure = 1;
 643                 } else if (netlink_broadcast_deliver(sk, skb2)) {
 644                         netlink_overrun(sk);
 645                 } else {
 646                         delivered = 1;
 647                         skb2 = NULL;
 648                 }
 649                 sock_put(sk);
 650         }
 651
 652         netlink_unlock_table();
 653
 654         if (skb2)
 655                 kfree_skb(skb2);
 656         kfree_skb(skb);
 657
 658         if (delivered)
 659                 return 0;
 660         if (failure)
 661                 return -ENOBUFS;
 662         return -ESRCH;
 663 }
 664
 665 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
 666 {
 667         struct sock *sk;
 668         struct hlist_node *node;
 669         int protocol = ssk->sk_protocol;
 670
 671         read_lock(&nl_table_lock);
 672         sk_for_each(sk, node, &nl_table[protocol]) {
 673                 struct netlink_opt *nlk = nlk_sk(sk);
 674                 if (ssk == sk)
 675                         continue;
 676
 677                 if (nlk->pid == pid || !(nlk->groups & group))
 678                         continue;
 679
 680                 sk->sk_err = code;
 681                 sk->sk_error_report(sk);
 682         }
 683         read_unlock(&nl_table_lock);
 684 }
 685
 686 static inline void netlink_rcv_wake(struct sock *sk)
 687 {
 688         struct netlink_opt *nlk = nlk_sk(sk);
 689
 690         if (!skb_queue_len(&sk->sk_receive_queue))
 691                 clear_bit(0, &nlk->state);
 692         if (!test_bit(0, &nlk->state))
 693                 wake_up_interruptible(&nlk->wait);
 694 }
 695
 696 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
 697                            struct msghdr *msg, size_t len)
 698 {
 699         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
 700         struct sock *sk = sock->sk;
 701         struct netlink_opt *nlk = nlk_sk(sk);
 702         struct sockaddr_nl *addr=msg->msg_name;
 703         u32 dst_pid;
 704         u32 dst_groups;
 705         struct sk_buff *skb;
 706         int err;
 707         struct scm_cookie scm;
 708
 709         if (msg->msg_flags&MSG_OOB)
 710                 return -EOPNOTSUPP;
 711
 712         if (NULL == siocb->scm)
 713                 siocb->scm = &scm;
 714         err = scm_send(sock, msg, siocb->scm);
 715         if (err < 0)
 716                 return err;
 717
 718         if (msg->msg_namelen) {
 719                 if (addr->nl_family != AF_NETLINK)
 720                         return -EINVAL;
 721                 dst_pid = addr->nl_pid;
 722                 dst_groups = addr->nl_groups;
 723                 if (dst_groups && !netlink_capable(sock, NL_NONROOT_SEND))
 724                         return -EPERM;
 725         } else {
 726                 dst_pid = nlk->dst_pid;
 727                 dst_groups = nlk->dst_groups;
 728         }
 729
 730         if (!nlk->pid) {
 731                 err = netlink_autobind(sock);
 732                 if (err)
 733                         goto out;
 734         }
 735
 736         err = -EMSGSIZE;
 737         if (len > sk->sk_sndbuf - 32)
 738                 goto out;
 739         err = -ENOBUFS;
 740         skb = alloc_skb(len, GFP_KERNEL);
 741         if (skb==NULL)
 742                 goto out;
 743
 744         NETLINK_CB(skb).pid     = nlk->pid;
 745         NETLINK_CB(skb).groups  = nlk->groups;
 746         NETLINK_CB(skb).dst_pid = dst_pid;
 747         NETLINK_CB(skb).dst_groups = dst_groups;
 748         memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
 749
 750         /* What can I do? Netlink is asynchronous, so that
 751            we will have to save current capabilities to
 752            check them, when this message will be delivered
 753            to corresponding kernel module.   --ANK (980802)
 754          */
 755
 756         err = -EFAULT;
 757         if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
 758                 kfree_skb(skb);
 759                 goto out;
 760         }
 761
 762         err = security_netlink_send(sk, skb);
 763         if (err) {
 764                 kfree_skb(skb);
 765                 goto out;
 766         }
 767
 768         if (dst_groups) {
 769                 atomic_inc(&skb->users);
 770                 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL);
 771         }
 772         err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
 773
 774 out:
 775         return err;
 776 }
 777
 778 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
 779                            struct msghdr *msg, size_t len,
 780                            int flags)
 781 {
 782         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
 783         struct scm_cookie scm;
 784         struct sock *sk = sock->sk;
 785         struct netlink_opt *nlk = nlk_sk(sk);
 786         int noblock = flags&MSG_DONTWAIT;
 787         size_t copied;
 788         struct sk_buff *skb;
 789         int err;
 790
 791         if (flags&MSG_OOB)
 792                 return -EOPNOTSUPP;
 793
 794         copied = 0;
 795
 796         skb = skb_recv_datagram(sk,flags,noblock,&err);
 797         if (skb==NULL)
 798                 goto out;
 799
 800         msg->msg_namelen = 0;
 801
 802         copied = skb->len;
 803         if (len < copied) {
 804                 msg->msg_flags |= MSG_TRUNC;
 805                 copied = len;
 806         }
 807
 808         skb->h.raw = skb->data;
 809         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
 810
 811         if (msg->msg_name) {
 812                 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
 813                 addr->nl_family = AF_NETLINK;
 814                 addr->nl_pad    = 0;
 815                 addr->nl_pid    = NETLINK_CB(skb).pid;
 816                 addr->nl_groups = NETLINK_CB(skb).dst_groups;
 817                 msg->msg_namelen = sizeof(*addr);
 818         }
 819
 820         if (NULL == siocb->scm) {
 821                 memset(&scm, 0, sizeof(scm));
 822                 siocb->scm = &scm;
 823         }
 824         siocb->scm->creds = *NETLINK_CREDS(skb);
 825         skb_free_datagram(sk, skb);
 826
 827         if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
 828                 netlink_dump(sk);
 829
 830         scm_recv(sock, msg, siocb->scm, flags);
 831
 832 out:
 833         netlink_rcv_wake(sk);
 834         return err ? : copied;
 835 }
 836
 837 static void netlink_data_ready(struct sock *sk, int len)
 838 {
 839         struct netlink_opt *nlk = nlk_sk(sk);
 840
 841         if (nlk->data_ready)
 842                 nlk->data_ready(sk, len);
 843         netlink_rcv_wake(sk);
 844 }
 845
 846 /*
 847  *      We export these functions to other modules. They provide a
 848  *      complete set of kernel non-blocking support for message
 849  *      queueing.
 850  */
 851
 852 struct sock *
 853 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len))
 854 {
 855         struct socket *sock;
 856         struct sock *sk;
 857
 858         if (unit<0 || unit>=MAX_LINKS)
 859                 return NULL;
 860
 861         if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
 862                 return NULL;
 863
 864         if (netlink_create(sock, unit) < 0) {
 865                 sock_release(sock);
 866                 return NULL;
 867         }
 868         sk = sock->sk;
 869         sk->sk_data_ready = netlink_data_ready;
 870         if (input)
 871                 nlk_sk(sk)->data_ready = input;
 872
 873         netlink_insert(sk, 0);
 874         return sk;
 875 }
 876
 877 void netlink_set_nonroot(int protocol, unsigned flags)
 878 {
 879         if ((unsigned)protocol < MAX_LINKS)
 880                 nl_nonroot[protocol] = flags;
 881 }
 882
 883 static void netlink_destroy_callback(struct netlink_callback *cb)
 884 {
 885         if (cb->skb)
 886                 kfree_skb(cb->skb);
 887         kfree(cb);
 888 }
 889
 890 /*
 891  * It looks a bit ugly.
 892  * It would be better to create kernel thread.
 893  */
 894
 895 static int netlink_dump(struct sock *sk)
 896 {
 897         struct netlink_opt *nlk = nlk_sk(sk);
 898         struct netlink_callback *cb;
 899         struct sk_buff *skb;
 900         struct nlmsghdr *nlh;
 901         int len;
 902
 903         skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
 904         if (!skb)
 905                 return -ENOBUFS;
 906
 907         spin_lock(&nlk->cb_lock);
 908
 909         cb = nlk->cb;
 910         if (cb == NULL) {
 911                 spin_unlock(&nlk->cb_lock);
 912                 kfree_skb(skb);
 913                 return -EINVAL;
 914         }
 915
 916         len = cb->dump(skb, cb);
 917
 918         if (len > 0) {
 919                 spin_unlock(&nlk->cb_lock);
 920                 skb_queue_tail(&sk->sk_receive_queue, skb);
 921                 sk->sk_data_ready(sk, len);
 922                 return 0;
 923         }
 924
 925         nlh = __nlmsg_put(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, NLMSG_DONE, sizeof(int));
 926         nlh->nlmsg_flags |= NLM_F_MULTI;
 927         memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
 928         skb_queue_tail(&sk->sk_receive_queue, skb);
 929         sk->sk_data_ready(sk, skb->len);
 930
 931         cb->done(cb);
 932         nlk->cb = NULL;
 933         spin_unlock(&nlk->cb_lock);
 934
 935         netlink_destroy_callback(cb);
 936         sock_put(sk);
 937         return 0;
 938 }
 939
 940 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
 941                        struct nlmsghdr *nlh,
 942                        int (*dump)(struct sk_buff *skb, struct netlink_callback*),
 943                        int (*done)(struct netlink_callback*))
 944 {
 945         struct netlink_callback *cb;
 946         struct sock *sk;
 947         struct netlink_opt *nlk;
 948
 949         cb = kmalloc(sizeof(*cb), GFP_KERNEL);
 950         if (cb == NULL)
 951                 return -ENOBUFS;
 952
 953         memset(cb, 0, sizeof(*cb));
 954         cb->dump = dump;
 955         cb->done = done;
 956         cb->nlh = nlh;
 957         atomic_inc(&skb->users);
 958         cb->skb = skb;
 959
 960         sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
 961         if (sk == NULL) {
 962                 netlink_destroy_callback(cb);
 963                 return -ECONNREFUSED;
 964         }
 965         nlk = nlk_sk(sk);
 966         /* A dump is in progress... */
 967         spin_lock(&nlk->cb_lock);
 968         if (nlk->cb) {
 969                 spin_unlock(&nlk->cb_lock);
 970                 netlink_destroy_callback(cb);
 971                 sock_put(sk);
 972                 return -EBUSY;
 973         }
 974         nlk->cb = cb;
 975         spin_unlock(&nlk->cb_lock);
 976
 977         netlink_dump(sk);
 978         return 0;
 979 }
 980
 981 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
 982 {
 983         struct sk_buff *skb;
 984         struct nlmsghdr *rep;
 985         struct nlmsgerr *errmsg;
 986         int size;
 987
 988         if (err == 0)
 989                 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
 990         else
 991                 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
 992
 993         skb = alloc_skb(size, GFP_KERNEL);
 994         if (!skb) {
 995                 struct sock *sk;
 996
 997                 sk = netlink_lookup(in_skb->sk->sk_protocol,
 998                                     NETLINK_CB(in_skb).pid);
 999                 if (sk) {
1000                         sk->sk_err = ENOBUFS;
1001                         sk->sk_error_report(sk);
1002                         sock_put(sk);
1003                 }
1004                 return;
1005         }
1006
1007         rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1008                           NLMSG_ERROR, sizeof(struct nlmsgerr));
1009         errmsg = NLMSG_DATA(rep);
1010         errmsg->error = err;
1011         memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1012         netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1013 }
1014
1015
1016 #ifdef NL_EMULATE_DEV
1017
1018 static rwlock_t nl_emu_lock = RW_LOCK_UNLOCKED;
1019
1020 /*
1021  *      Backward compatibility.
1022  */
1023
1024 int netlink_attach(int unit, int (*function)(int, struct sk_buff *skb))
1025 {
1026         struct sock *sk = netlink_kernel_create(unit, NULL);
1027         if (sk == NULL)
1028                 return -ENOBUFS;
1029         nlk_sk(sk)->handler = function;
1030         write_lock_bh(&nl_emu_lock);
1031         netlink_kernel[unit] = sk->sk_socket;
1032         write_unlock_bh(&nl_emu_lock);
1033         return 0;
1034 }
1035
1036 void netlink_detach(int unit)
1037 {
1038         struct socket *sock;
1039
1040         write_lock_bh(&nl_emu_lock);
1041         sock = netlink_kernel[unit];
1042         netlink_kernel[unit] = NULL;
1043         write_unlock_bh(&nl_emu_lock);
1044
1045         sock_release(sock);
1046 }
1047
1048 int netlink_post(int unit, struct sk_buff *skb)
1049 {
1050         struct socket *sock;
1051
1052         read_lock(&nl_emu_lock);
1053         sock = netlink_kernel[unit];
1054         if (sock) {
1055                 struct sock *sk = sock->sk;
1056                 memset(skb->cb, 0, sizeof(skb->cb));
1057                 sock_hold(sk);
1058                 read_unlock(&nl_emu_lock);
1059
1060                 netlink_broadcast(sk, skb, 0, ~0, GFP_ATOMIC);
1061
1062                 sock_put(sk);
1063                 return 0;
1064         }
1065         read_unlock(&nl_emu_lock);
1066         return -EUNATCH;
1067 }
1068
1069 #endif
1070
1071 #ifdef CONFIG_PROC_FS
1072 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1073 {
1074         long i;
1075         struct sock *s;
1076         struct hlist_node *node;
1077         loff_t off = 0;
1078
1079         for (i=0; i<MAX_LINKS; i++) {
1080                 sk_for_each(s, node, &nl_table[i]) {
1081                         if (off == pos) {
1082                                 seq->private = (void *) i;
1083                                 return s;
1084                         }
1085                         ++off;
1086                 }
1087         }
1088         return NULL;
1089 }
1090
1091 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1092 {
1093         read_lock(&nl_table_lock);
1094         return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1095 }
1096
1097 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1098 {
1099         struct sock *s;
1100
1101         ++*pos;
1102
1103         if (v == SEQ_START_TOKEN)
1104                 return netlink_seq_socket_idx(seq, 0);
1105
1106         s = sk_next(v);
1107         if (!s) {
1108                 long i = (long)seq->private;
1109
1110                 while (++i < MAX_LINKS) {
1111                         s = sk_head(&nl_table[i]);
1112                         if (s) {
1113                                 seq->private = (void *) i;
1114                                 break;
1115                         }
1116                 }
1117         }
1118         return s;
1119 }
1120
1121 static void netlink_seq_stop(struct seq_file *seq, void *v)
1122 {
1123         read_unlock(&nl_table_lock);
1124 }
1125
1126
1127 static int netlink_seq_show(struct seq_file *seq, void *v)
1128 {
1129         if (v == SEQ_START_TOKEN)
1130                 seq_puts(seq,
1131                          "sk       Eth Pid    Groups   "
1132                          "Rmem     Wmem     Dump     Locks\n");
1133         else {
1134                 struct sock *s = v;
1135                 struct netlink_opt *nlk = nlk_sk(s);
1136
1137                 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1138                            s,
1139                            s->sk_protocol,
1140                            nlk->pid,
1141                            nlk->groups,
1142                            atomic_read(&s->sk_rmem_alloc),
1143                            atomic_read(&s->sk_wmem_alloc),
1144                            nlk->cb,
1145                            atomic_read(&s->sk_refcnt)
1146                         );
1147
1148         }
1149         return 0;
1150 }
1151
1152 static struct seq_operations netlink_seq_ops = {
1153         .start  = netlink_seq_start,
1154         .next   = netlink_seq_next,
1155         .stop   = netlink_seq_stop,
1156         .show   = netlink_seq_show,
1157 };
1158
1159
1160 static int netlink_seq_open(struct inode *inode, struct file *file)
1161 {
1162         return seq_open(file, &netlink_seq_ops);
1163 }
1164
1165 static struct file_operations netlink_seq_fops = {
1166         .owner          = THIS_MODULE,
1167         .open           = netlink_seq_open,
1168         .read           = seq_read,
1169         .llseek         = seq_lseek,
1170         .release        = seq_release,
1171 };
1172
1173 #endif
1174
1175 int netlink_register_notifier(struct notifier_block *nb)
1176 {
1177         return notifier_chain_register(&netlink_chain, nb);
1178 }
1179
1180 int netlink_unregister_notifier(struct notifier_block *nb)
1181 {
1182         return notifier_chain_unregister(&netlink_chain, nb);
1183 }
1184
1185 static struct proto_ops netlink_ops = {
1186         .family =       PF_NETLINK,
1187         .owner =        THIS_MODULE,
1188         .release =      netlink_release,
1189         .bind =         netlink_bind,
1190         .connect =      netlink_connect,
1191         .socketpair =   sock_no_socketpair,
1192         .accept =       sock_no_accept,
1193         .getname =      netlink_getname,
1194         .poll =         datagram_poll,
1195         .ioctl =        sock_no_ioctl,
1196         .listen =       sock_no_listen,
1197         .shutdown =     sock_no_shutdown,
1198         .setsockopt =   sock_no_setsockopt,
1199         .getsockopt =   sock_no_getsockopt,
1200         .sendmsg =      netlink_sendmsg,
1201         .recvmsg =      netlink_recvmsg,
1202         .mmap =         sock_no_mmap,
1203         .sendpage =     sock_no_sendpage,
1204 };
1205
1206 static struct net_proto_family netlink_family_ops = {
1207         .family = PF_NETLINK,
1208         .create = netlink_create,
1209         .owner  = THIS_MODULE,  /* for consistency 8) */
1210 };
1211
1212 static int __init netlink_proto_init(void)
1213 {
1214         struct sk_buff *dummy_skb;
1215
1216         if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb)) {
1217                 printk(KERN_CRIT "netlink_init: panic\n");
1218                 return -1;
1219         }
1220         sock_register(&netlink_family_ops);
1221
1222 #ifdef CONFIG_PROC_FS
1223         proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1224 #endif
1225
1226         /* The netlink device handler may be needed early. */
1227         rtnetlink_init();
1228
1229         return 0;
1230 }
1231
1232 static void __exit netlink_proto_exit(void)
1233 {
1234        sock_unregister(PF_NETLINK);
1235        proc_net_remove("netlink");
1236 }
1237
1238 core_initcall(netlink_proto_init);
1239 module_exit(netlink_proto_exit);
1240
1241 MODULE_LICENSE("GPL");
1242
1243 MODULE_ALIAS_NETPROTO(PF_NETLINK);
1244
1245 EXPORT_SYMBOL(netlink_ack);
1246 EXPORT_SYMBOL(netlink_broadcast);
1247 EXPORT_SYMBOL(netlink_dump_start);
1248 EXPORT_SYMBOL(netlink_kernel_create);
1249 EXPORT_SYMBOL(netlink_register_notifier);
1250 EXPORT_SYMBOL(netlink_set_err);
1251 EXPORT_SYMBOL(netlink_set_nonroot);
1252 EXPORT_SYMBOL(netlink_unicast);
1253 EXPORT_SYMBOL(netlink_unregister_notifier);
1254
1255 #if defined(CONFIG_NETLINK_DEV) || defined(CONFIG_NETLINK_DEV_MODULE)
1256 EXPORT_SYMBOL(netlink_attach);
1257 EXPORT_SYMBOL(netlink_detach);
1258 EXPORT_SYMBOL(netlink_post);
1259 #endif