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  */
  13
  14 #include <linux/config.h>
  15 #include <linux/module.h>
  16
  17 #include <linux/kernel.h>
  18 #include <linux/init.h>
  19 #include <linux/major.h>
  20 #include <linux/signal.h>
  21 #include <linux/sched.h>
  22 #include <linux/errno.h>
  23 #include <linux/string.h>
  24 #include <linux/stat.h>
  25 #include <linux/socket.h>
  26 #include <linux/un.h>
  27 #include <linux/fcntl.h>
  28 #include <linux/termios.h>
  29 #include <linux/sockios.h>
  30 #include <linux/net.h>
  31 #include <linux/fs.h>
  32 #include <linux/malloc.h>
  33 #include <asm/uaccess.h>
  34 #include <linux/skbuff.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/netlink.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/smp_lock.h>
  39 #include <net/sock.h>
  40 #include <net/scm.h>
  41
  42 #define Nprintk(a...)
  43
  44 #if defined(CONFIG_NETLINK_DEV) || defined(CONFIG_NETLINK_DEV_MODULE)
  45 #define NL_EMULATE_DEV
  46 #endif
  47
  48 #define BUG_TRAP(x) if (!(x)) { printk("Assertion (" #x ") failed at " __FILE__ "(%d):" __FUNCTION__ "\n", __LINE__); }
  49
  50 struct netlink_opt
  51 {
  52         u32                     pid;
  53         unsigned                groups;
  54         u32                     dst_pid;
  55         unsigned                dst_groups;
  56         unsigned long           state;
  57         int                     (*handler)(int unit, struct sk_buff *skb);
  58         wait_queue_head_t       wait;
  59         struct netlink_callback *cb;
  60         spinlock_t              cb_lock;
  61         void                    (*data_ready)(struct sock *sk, int bytes);
  62 };
  63
  64 static struct sock *nl_table[MAX_LINKS];
  65 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
  66
  67 #ifdef NL_EMULATE_DEV
  68 static struct socket *netlink_kernel[MAX_LINKS];
  69 #endif
  70
  71 static int netlink_dump(struct sock *sk);
  72 static void netlink_destroy_callback(struct netlink_callback *cb);
  73
  74 atomic_t netlink_sock_nr;
  75
  76 static rwlock_t nl_table_lock = RW_LOCK_UNLOCKED;
  77 static atomic_t nl_table_users = ATOMIC_INIT(0);
  78
  79 static void netlink_sock_destruct(struct sock *sk)
  80 {
  81         skb_queue_purge(&sk->receive_queue);
  82
  83         if (!sk->dead) {
  84                 printk("Freeing alive netlink socket %p\n", sk);
  85                 return;
  86         }
  87         BUG_TRAP(atomic_read(&sk->rmem_alloc)==0);
  88         BUG_TRAP(atomic_read(&sk->wmem_alloc)==0);
  89         BUG_TRAP(sk->protinfo.af_netlink->cb==NULL);
  90
  91         kfree(sk->protinfo.af_netlink);
  92
  93         atomic_dec(&netlink_sock_nr);
  94 #ifdef NETLINK_REFCNT_DEBUG
  95         printk(KERN_DEBUG "NETLINK %p released, %d are still alive\n", sk, atomic_read(&netlink_sock_nr));
  96 #endif
  97 }
  98
  99 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
 100  * Look, when several writers sleep and reader wakes them up, all but one
 101  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
 102  * this, _but_ remember, it adds useless work on UP machines.
 103  */
 104
 105 static void netlink_table_grab(void)
 106 {
 107         write_lock_bh(&nl_table_lock);
 108
 109         if (atomic_read(&nl_table_users)) {
 110                 DECLARE_WAITQUEUE(wait, current);
 111
 112                 add_wait_queue_exclusive(&nl_table_wait, &wait);
 113                 for(;;) {
 114                         set_current_state(TASK_UNINTERRUPTIBLE);
 115                         if (atomic_read(&nl_table_users) == 0)
 116                                 break;
 117                         write_unlock_bh(&nl_table_lock);
 118                         schedule();
 119                         write_lock_bh(&nl_table_lock);
 120                 }
 121
 122                 __set_current_state(TASK_RUNNING);
 123                 remove_wait_queue(&nl_table_wait, &wait);
 124         }
 125 }
 126
 127 static __inline__ void netlink_table_ungrab(void)
 128 {
 129         write_unlock_bh(&nl_table_lock);
 130         wake_up(&nl_table_wait);
 131 }
 132
 133 static __inline__ void
 134 netlink_lock_table(void)
 135 {
 136         /* read_lock() synchronizes us to netlink_table_grab */
 137
 138         read_lock(&nl_table_lock);
 139         atomic_inc(&nl_table_users);
 140         read_unlock(&nl_table_lock);
 141 }
 142
 143 static __inline__ void
 144 netlink_unlock_table(void)
 145 {
 146         if (atomic_dec_and_test(&nl_table_users))
 147                 wake_up(&nl_table_wait);
 148 }
 149
 150 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
 151 {
 152         struct sock *sk;
 153
 154         read_lock(&nl_table_lock);
 155         for (sk=nl_table[protocol]; sk; sk=sk->next) {
 156                 if (sk->protinfo.af_netlink->pid == pid) {
 157                         sock_hold(sk);
 158                         read_unlock(&nl_table_lock);
 159                         return sk;
 160                 }
 161         }
 162
 163         read_unlock(&nl_table_lock);
 164         return NULL;
 165 }
 166
 167 extern struct proto_ops netlink_ops;
 168
 169 static int netlink_insert(struct sock *sk, u32 pid)
 170 {
 171         int err = -EADDRINUSE;
 172         struct sock *osk;
 173
 174         netlink_table_grab();
 175         for (osk=nl_table[sk->protocol]; osk; osk=osk->next) {
 176                 if (osk->protinfo.af_netlink->pid == pid)
 177                         break;
 178         }
 179         if (osk == NULL) {
 180                 err = -EBUSY;
 181                 if (sk->protinfo.af_netlink->pid == 0) {
 182                         sk->protinfo.af_netlink->pid = pid;
 183                         sk->next = nl_table[sk->protocol];
 184                         nl_table[sk->protocol] = sk;
 185                         sock_hold(sk);
 186                         err = 0;
 187                 }
 188         }
 189         netlink_table_ungrab();
 190         return err;
 191 }
 192
 193 static void netlink_remove(struct sock *sk)
 194 {
 195         struct sock **skp;
 196
 197         netlink_table_grab();
 198         for (skp = &nl_table[sk->protocol]; *skp; skp = &((*skp)->next)) {
 199                 if (*skp == sk) {
 200                         *skp = sk->next;
 201                         __sock_put(sk);
 202                         break;
 203                 }
 204         }
 205         netlink_table_ungrab();
 206 }
 207
 208 static int netlink_create(struct socket *sock, int protocol)
 209 {
 210         struct sock *sk;
 211
 212         sock->state = SS_UNCONNECTED;
 213
 214         if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 215                 return -ESOCKTNOSUPPORT;
 216
 217         if (protocol<0 || protocol >= MAX_LINKS)
 218                 return -EPROTONOSUPPORT;
 219
 220         sock->ops = &netlink_ops;
 221
 222         sk = sk_alloc(PF_NETLINK, GFP_KERNEL, 1);
 223         if (!sk)
 224                 return -ENOMEM;
 225
 226         sock_init_data(sock,sk);
 227
 228         sk->protinfo.af_netlink = kmalloc(sizeof(struct netlink_opt), GFP_KERNEL);
 229         if (sk->protinfo.af_netlink == NULL) {
 230                 sk_free(sk);
 231                 return -ENOMEM;
 232         }
 233         memset(sk->protinfo.af_netlink, 0, sizeof(struct netlink_opt));
 234
 235         spin_lock_init(&sk->protinfo.af_netlink->cb_lock);
 236         init_waitqueue_head(&sk->protinfo.af_netlink->wait);
 237         sk->destruct = netlink_sock_destruct;
 238         atomic_inc(&netlink_sock_nr);
 239
 240         sk->protocol=protocol;
 241         return 0;
 242 }
 243
 244 static int netlink_release(struct socket *sock)
 245 {
 246         struct sock *sk = sock->sk;
 247
 248         if (!sk)
 249                 return 0;
 250
 251         netlink_remove(sk);
 252
 253         spin_lock(&sk->protinfo.af_netlink->cb_lock);
 254         if (sk->protinfo.af_netlink->cb) {
 255                 sk->protinfo.af_netlink->cb->done(sk->protinfo.af_netlink->cb);
 256                 netlink_destroy_callback(sk->protinfo.af_netlink->cb);
 257                 sk->protinfo.af_netlink->cb = NULL;
 258                 __sock_put(sk);
 259         }
 260         spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 261
 262         /* OK. Socket is unlinked, and, therefore,
 263            no new packets will arrive */
 264
 265         sock_orphan(sk);
 266         sock->sk = NULL;
 267         wake_up_interruptible_all(&sk->protinfo.af_netlink->wait);
 268
 269         skb_queue_purge(&sk->write_queue);
 270
 271         sock_put(sk);
 272         return 0;
 273 }
 274
 275 static int netlink_autobind(struct socket *sock)
 276 {
 277         struct sock *sk = sock->sk;
 278         struct sock *osk;
 279         s32 pid = current->pid;
 280         int err;
 281
 282 retry:
 283         netlink_table_grab();
 284         for (osk=nl_table[sk->protocol]; osk; osk=osk->next) {
 285                 if (osk->protinfo.af_netlink->pid == pid) {
 286                         /* Bind collision, search negative pid values. */
 287                         if (pid > 0)
 288                                 pid = -4096;
 289                         pid--;
 290                         netlink_table_ungrab();
 291                         goto retry;
 292                 }
 293         }
 294         netlink_table_ungrab();
 295
 296         err = netlink_insert(sk, pid);
 297         if (err == -EADDRINUSE)
 298                 goto retry;
 299         sk->protinfo.af_netlink->groups = 0;
 300         return 0;
 301 }
 302
 303 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 304 {
 305         struct sock *sk = sock->sk;
 306         int err;
 307         struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
 308
 309         if (nladdr->nl_family != AF_NETLINK)
 310                 return -EINVAL;
 311
 312         /* Only superuser is allowed to listen multicasts */
 313         if (nladdr->nl_groups && !capable(CAP_NET_ADMIN))
 314                 return -EPERM;
 315
 316         if (sk->protinfo.af_netlink->pid) {
 317                 if (nladdr->nl_pid != sk->protinfo.af_netlink->pid)
 318                         return -EINVAL;
 319                 sk->protinfo.af_netlink->groups = nladdr->nl_groups;
 320                 return 0;
 321         }
 322
 323         if (nladdr->nl_pid == 0) {
 324                 err = netlink_autobind(sock);
 325                 if (err == 0)
 326                         sk->protinfo.af_netlink->groups = nladdr->nl_groups;
 327                 return err;
 328         }
 329
 330         err = netlink_insert(sk, nladdr->nl_pid);
 331         if (err == 0)
 332                 sk->protinfo.af_netlink->groups = nladdr->nl_groups;
 333         return err;
 334 }
 335
 336 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
 337                            int alen, int flags)
 338 {
 339         int err = 0;
 340         struct sock *sk = sock->sk;
 341         struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
 342
 343         if (addr->sa_family == AF_UNSPEC) {
 344                 sk->protinfo.af_netlink->dst_pid = 0;
 345                 sk->protinfo.af_netlink->dst_groups = 0;
 346                 return 0;
 347         }
 348         if (addr->sa_family != AF_NETLINK)
 349                 return -EINVAL;
 350
 351         /* Only superuser is allowed to send multicasts */
 352         if (nladdr->nl_groups && !capable(CAP_NET_ADMIN))
 353                 return -EPERM;
 354
 355         if (!sk->protinfo.af_netlink->pid)
 356                 err = netlink_autobind(sock);
 357
 358         if (err == 0) {
 359                 sk->protinfo.af_netlink->dst_pid = nladdr->nl_pid;
 360                 sk->protinfo.af_netlink->dst_groups = nladdr->nl_groups;
 361         }
 362
 363         return 0;
 364 }
 365
 366 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
 367 {
 368         struct sock *sk = sock->sk;
 369         struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
 370
 371         nladdr->nl_family = AF_NETLINK;
 372         *addr_len = sizeof(*nladdr);
 373
 374         if (peer) {
 375                 nladdr->nl_pid = sk->protinfo.af_netlink->dst_pid;
 376                 nladdr->nl_groups = sk->protinfo.af_netlink->dst_groups;
 377         } else {
 378                 nladdr->nl_pid = sk->protinfo.af_netlink->pid;
 379                 nladdr->nl_groups = sk->protinfo.af_netlink->groups;
 380         }
 381         return 0;
 382 }
 383
 384 static void netlink_overrun(struct sock *sk)
 385 {
 386         if (!test_and_set_bit(0, &sk->protinfo.af_netlink->state)) {
 387                 sk->err = ENOBUFS;
 388                 sk->error_report(sk);
 389         }
 390 }
 391
 392 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
 393 {
 394         struct sock *sk;
 395         int len = skb->len;
 396         int protocol = ssk->protocol;
 397         long timeo;
 398         DECLARE_WAITQUEUE(wait, current);
 399
 400         timeo = sock_sndtimeo(ssk, nonblock);
 401
 402 retry:
 403         sk = netlink_lookup(protocol, pid);
 404         if (sk == NULL)
 405                 goto no_dst;
 406
 407 #ifdef NL_EMULATE_DEV
 408         if (sk->protinfo.af_netlink->handler) {
 409                 skb_orphan(skb);
 410                 len = sk->protinfo.af_netlink->handler(protocol, skb);
 411                 sock_put(sk);
 412                 return len;
 413         }
 414 #endif
 415
 416         if (atomic_read(&sk->rmem_alloc) > sk->rcvbuf ||
 417             test_bit(0, &sk->protinfo.af_netlink->state)) {
 418                 if (!timeo) {
 419                         if (ssk->protinfo.af_netlink->pid == 0)
 420                                 netlink_overrun(sk);
 421                         sock_put(sk);
 422                         kfree_skb(skb);
 423                         return -EAGAIN;
 424                 }
 425
 426                 __set_current_state(TASK_INTERRUPTIBLE);
 427                 add_wait_queue(&sk->protinfo.af_netlink->wait, &wait);
 428
 429                 if ((atomic_read(&sk->rmem_alloc) > sk->rcvbuf ||
 430                     test_bit(0, &sk->protinfo.af_netlink->state)) &&
 431                     !sk->dead)
 432                         timeo = schedule_timeout(timeo);
 433
 434                 __set_current_state(TASK_RUNNING);
 435                 remove_wait_queue(&sk->protinfo.af_netlink->wait, &wait);
 436                 sock_put(sk);
 437
 438                 if (signal_pending(current)) {
 439                         kfree_skb(skb);
 440                         return sock_intr_errno(timeo);
 441                 }
 442                 goto retry;
 443         }
 444
 445         skb_orphan(skb);
 446         skb_set_owner_r(skb, sk);
 447         skb_queue_tail(&sk->receive_queue, skb);
 448         sk->data_ready(sk, len);
 449         sock_put(sk);
 450         return len;
 451
 452 no_dst:
 453         kfree_skb(skb);
 454         return -ECONNREFUSED;
 455 }
 456
 457 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
 458 {
 459 #ifdef NL_EMULATE_DEV
 460         if (sk->protinfo.af_netlink->handler) {
 461                 skb_orphan(skb);
 462                 sk->protinfo.af_netlink->handler(sk->protocol, skb);
 463                 return 0;
 464         } else
 465 #endif
 466         if (atomic_read(&sk->rmem_alloc) <= sk->rcvbuf &&
 467             !test_bit(0, &sk->protinfo.af_netlink->state)) {
 468                 skb_orphan(skb);
 469                 skb_set_owner_r(skb, sk);
 470                 skb_queue_tail(&sk->receive_queue, skb);
 471                 sk->data_ready(sk, skb->len);
 472                 return 0;
 473         }
 474         return -1;
 475 }
 476
 477 void netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
 478                        u32 group, int allocation)
 479 {
 480         struct sock *sk;
 481         struct sk_buff *skb2 = NULL;
 482         int protocol = ssk->protocol;
 483         int failure = 0;
 484
 485         /* While we sleep in clone, do not allow to change socket list */
 486
 487         netlink_lock_table();
 488
 489         for (sk = nl_table[protocol]; sk; sk = sk->next) {
 490                 if (ssk == sk)
 491                         continue;
 492
 493                 if (sk->protinfo.af_netlink->pid == pid ||
 494                     !(sk->protinfo.af_netlink->groups&group))
 495                         continue;
 496
 497                 if (failure) {
 498                         netlink_overrun(sk);
 499                         continue;
 500                 }
 501
 502                 sock_hold(sk);
 503                 if (skb2 == NULL) {
 504                         if (atomic_read(&skb->users) != 1) {
 505                                 skb2 = skb_clone(skb, allocation);
 506                         } else {
 507                                 skb2 = skb;
 508                                 atomic_inc(&skb->users);
 509                         }
 510                 }
 511                 if (skb2 == NULL) {
 512                         netlink_overrun(sk);
 513                         /* Clone failed. Notify ALL listeners. */
 514                         failure = 1;
 515                 } else if (netlink_broadcast_deliver(sk, skb2)) {
 516                         netlink_overrun(sk);
 517                 } else
 518                         skb2 = NULL;
 519                 sock_put(sk);
 520         }
 521
 522         netlink_unlock_table();
 523
 524         if (skb2)
 525                 kfree_skb(skb2);
 526         kfree_skb(skb);
 527 }
 528
 529 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
 530 {
 531         struct sock *sk;
 532         int protocol = ssk->protocol;
 533
 534         read_lock(&nl_table_lock);
 535         for (sk = nl_table[protocol]; sk; sk = sk->next) {
 536                 if (ssk == sk)
 537                         continue;
 538
 539                 if (sk->protinfo.af_netlink->pid == pid ||
 540                     !(sk->protinfo.af_netlink->groups&group))
 541                         continue;
 542
 543                 sk->err = code;
 544                 sk->error_report(sk);
 545         }
 546         read_unlock(&nl_table_lock);
 547 }
 548
 549 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, int len,
 550                            struct scm_cookie *scm)
 551 {
 552         struct sock *sk = sock->sk;
 553         struct sockaddr_nl *addr=msg->msg_name;
 554         u32 dst_pid;
 555         u32 dst_groups;
 556         struct sk_buff *skb;
 557         int err;
 558
 559         if (msg->msg_flags&MSG_OOB)
 560                 return -EOPNOTSUPP;
 561
 562         if (msg->msg_namelen) {
 563                 if (addr->nl_family != AF_NETLINK)
 564                         return -EINVAL;
 565                 dst_pid = addr->nl_pid;
 566                 dst_groups = addr->nl_groups;
 567                 if (dst_groups && !capable(CAP_NET_ADMIN))
 568                         return -EPERM;
 569         } else {
 570                 dst_pid = sk->protinfo.af_netlink->dst_pid;
 571                 dst_groups = sk->protinfo.af_netlink->dst_groups;
 572         }
 573
 574         if (!sk->protinfo.af_netlink->pid) {
 575                 err = netlink_autobind(sock);
 576                 if (err)
 577                         goto out;
 578         }
 579
 580         err = -EMSGSIZE;
 581         if ((unsigned)len > sk->sndbuf-32)
 582                 goto out;
 583         err = -ENOBUFS;
 584         skb = alloc_skb(len, GFP_KERNEL);
 585         if (skb==NULL)
 586                 goto out;
 587
 588         NETLINK_CB(skb).pid = sk->protinfo.af_netlink->pid;
 589         NETLINK_CB(skb).groups = sk->protinfo.af_netlink->groups;
 590         NETLINK_CB(skb).dst_pid = dst_pid;
 591         NETLINK_CB(skb).dst_groups = dst_groups;
 592         memcpy(NETLINK_CREDS(skb), &scm->creds, sizeof(struct ucred));
 593
 594         /* What can I do? Netlink is asynchronous, so that
 595            we will have to save current capabilities to
 596            check them, when this message will be delivered
 597            to corresponding kernel module.   --ANK (980802)
 598          */
 599         NETLINK_CB(skb).eff_cap = current->cap_effective;
 600
 601         err = -EFAULT;
 602         if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
 603                 kfree_skb(skb);
 604                 goto out;
 605         }
 606
 607         if (dst_groups) {
 608                 atomic_inc(&skb->users);
 609                 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL);
 610         }
 611         err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
 612
 613 out:
 614         return err;
 615 }
 616
 617 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, int len,
 618                            int flags, struct scm_cookie *scm)
 619 {
 620         struct sock *sk = sock->sk;
 621         int noblock = flags&MSG_DONTWAIT;
 622         int copied;
 623         struct sk_buff *skb;
 624         int err;
 625
 626         if (flags&MSG_OOB)
 627                 return -EOPNOTSUPP;
 628
 629         copied = 0;
 630
 631         skb = skb_recv_datagram(sk,flags,noblock,&err);
 632         if (skb==NULL)
 633                 goto out;
 634
 635         msg->msg_namelen = 0;
 636
 637         copied = skb->len;
 638         if (len < copied) {
 639                 msg->msg_flags |= MSG_TRUNC;
 640                 copied = len;
 641         }
 642
 643         skb->h.raw = skb->data;
 644         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
 645
 646         if (msg->msg_name) {
 647                 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
 648                 addr->nl_family = AF_NETLINK;
 649                 addr->nl_pid    = NETLINK_CB(skb).pid;
 650                 addr->nl_groups = NETLINK_CB(skb).dst_groups;
 651                 msg->msg_namelen = sizeof(*addr);
 652         }
 653
 654         scm->creds = *NETLINK_CREDS(skb);
 655         skb_free_datagram(sk, skb);
 656
 657         if (sk->protinfo.af_netlink->cb
 658             && atomic_read(&sk->rmem_alloc) <= sk->rcvbuf/2)
 659                 netlink_dump(sk);
 660
 661 out:
 662         if (skb_queue_len(&sk->receive_queue) <= sk->rcvbuf/2) {
 663                 if (skb_queue_len(&sk->receive_queue) == 0)
 664                         clear_bit(0, &sk->protinfo.af_netlink->state);
 665                 if (!test_bit(0, &sk->protinfo.af_netlink->state))
 666                         wake_up_interruptible(&sk->protinfo.af_netlink->wait);
 667         }
 668         return err ? : copied;
 669 }
 670
 671 void netlink_data_ready(struct sock *sk, int len)
 672 {
 673         if (sk->protinfo.af_netlink->data_ready)
 674                 sk->protinfo.af_netlink->data_ready(sk, len);
 675
 676         if (skb_queue_len(&sk->receive_queue) <= sk->rcvbuf/2) {
 677                 if (skb_queue_len(&sk->receive_queue) == 0)
 678                         clear_bit(0, &sk->protinfo.af_netlink->state);
 679                 if (!test_bit(0, &sk->protinfo.af_netlink->state))
 680                         wake_up_interruptible(&sk->protinfo.af_netlink->wait);
 681         }
 682 }
 683
 684 /*
 685  *      We export these functions to other modules. They provide a
 686  *      complete set of kernel non-blocking support for message
 687  *      queueing.
 688  */
 689
 690 struct sock *
 691 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len))
 692 {
 693         struct socket *sock;
 694         struct sock *sk;
 695
 696         if (unit<0 || unit>=MAX_LINKS)
 697                 return NULL;
 698
 699         if (!(sock = sock_alloc()))
 700                 return NULL;
 701
 702         sock->type = SOCK_RAW;
 703
 704         if (netlink_create(sock, unit) < 0) {
 705                 sock_release(sock);
 706                 return NULL;
 707         }
 708         sk = sock->sk;
 709         sk->data_ready = netlink_data_ready;
 710         if (input)
 711                 sk->protinfo.af_netlink->data_ready = input;
 712
 713         netlink_insert(sk, 0);
 714         return sk;
 715 }
 716
 717 static void netlink_destroy_callback(struct netlink_callback *cb)
 718 {
 719         if (cb->skb)
 720                 kfree_skb(cb->skb);
 721         kfree(cb);
 722 }
 723
 724 /*
 725  * It looks a bit ugly.
 726  * It would be better to create kernel thread.
 727  */
 728
 729 static int netlink_dump(struct sock *sk)
 730 {
 731         struct netlink_callback *cb;
 732         struct sk_buff *skb;
 733         struct nlmsghdr *nlh;
 734         int len;
 735
 736         skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
 737         if (!skb)
 738                 return -ENOBUFS;
 739
 740         spin_lock(&sk->protinfo.af_netlink->cb_lock);
 741
 742         cb = sk->protinfo.af_netlink->cb;
 743         if (cb == NULL) {
 744                 spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 745                 kfree_skb(skb);
 746                 return -EINVAL;
 747         }
 748
 749         len = cb->dump(skb, cb);
 750
 751         if (len > 0) {
 752                 spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 753                 skb_queue_tail(&sk->receive_queue, skb);
 754                 sk->data_ready(sk, len);
 755                 return 0;
 756         }
 757
 758         nlh = __nlmsg_put(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, NLMSG_DONE, sizeof(int));
 759         nlh->nlmsg_flags |= NLM_F_MULTI;
 760         memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
 761         skb_queue_tail(&sk->receive_queue, skb);
 762         sk->data_ready(sk, skb->len);
 763
 764         cb->done(cb);
 765         sk->protinfo.af_netlink->cb = NULL;
 766         spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 767
 768         netlink_destroy_callback(cb);
 769         sock_put(sk);
 770         return 0;
 771 }
 772
 773 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
 774                        struct nlmsghdr *nlh,
 775                        int (*dump)(struct sk_buff *skb, struct netlink_callback*),
 776                        int (*done)(struct netlink_callback*))
 777 {
 778         struct netlink_callback *cb;
 779         struct sock *sk;
 780
 781         cb = kmalloc(sizeof(*cb), GFP_KERNEL);
 782         if (cb == NULL)
 783                 return -ENOBUFS;
 784
 785         memset(cb, 0, sizeof(*cb));
 786         cb->dump = dump;
 787         cb->done = done;
 788         cb->nlh = nlh;
 789         atomic_inc(&skb->users);
 790         cb->skb = skb;
 791
 792         sk = netlink_lookup(ssk->protocol, NETLINK_CB(skb).pid);
 793         if (sk == NULL) {
 794                 netlink_destroy_callback(cb);
 795                 return -ECONNREFUSED;
 796         }
 797         /* A dump is in progress... */
 798         spin_lock(&sk->protinfo.af_netlink->cb_lock);
 799         if (sk->protinfo.af_netlink->cb) {
 800                 spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 801                 netlink_destroy_callback(cb);
 802                 sock_put(sk);
 803                 return -EBUSY;
 804         }
 805         sk->protinfo.af_netlink->cb = cb;
 806         spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 807
 808         netlink_dump(sk);
 809         return 0;
 810 }
 811
 812 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
 813 {
 814         struct sk_buff *skb;
 815         struct nlmsghdr *rep;
 816         struct nlmsgerr *errmsg;
 817         int size;
 818
 819         if (err == 0)
 820                 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
 821         else
 822                 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
 823
 824         skb = alloc_skb(size, GFP_KERNEL);
 825         if (!skb)
 826                 return;
 827
 828         rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
 829                           NLMSG_ERROR, sizeof(struct nlmsgerr));
 830         errmsg = NLMSG_DATA(rep);
 831         errmsg->error = err;
 832         memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
 833         netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
 834 }
 835
 836
 837 #ifdef NL_EMULATE_DEV
 838
 839 static rwlock_t nl_emu_lock = RW_LOCK_UNLOCKED;
 840
 841 /*
 842  *      Backward compatibility.
 843  */
 844
 845 int netlink_attach(int unit, int (*function)(int, struct sk_buff *skb))
 846 {
 847         struct sock *sk = netlink_kernel_create(unit, NULL);
 848         if (sk == NULL)
 849                 return -ENOBUFS;
 850         sk->protinfo.af_netlink->handler = function;
 851         write_lock_bh(&nl_emu_lock);
 852         netlink_kernel[unit] = sk->socket;
 853         write_unlock_bh(&nl_emu_lock);
 854         return 0;
 855 }
 856
 857 void netlink_detach(int unit)
 858 {
 859         struct socket *sock;
 860
 861         write_lock_bh(&nl_emu_lock);
 862         sock = netlink_kernel[unit];
 863         netlink_kernel[unit] = NULL;
 864         write_unlock_bh(&nl_emu_lock);
 865
 866         sock_release(sock);
 867 }
 868
 869 int netlink_post(int unit, struct sk_buff *skb)
 870 {
 871         struct socket *sock;
 872
 873         read_lock(&nl_emu_lock);
 874         sock = netlink_kernel[unit];
 875         if (sock) {
 876                 struct sock *sk = sock->sk;
 877                 memset(skb->cb, 0, sizeof(skb->cb));
 878                 sock_hold(sk);
 879                 read_unlock(&nl_emu_lock);
 880
 881                 netlink_broadcast(sk, skb, 0, ~0, GFP_ATOMIC);
 882
 883                 sock_put(sk);
 884                 return 0;
 885         }
 886         read_unlock(&nl_emu_lock);
 887         return -EUNATCH;
 888 }
 889
 890 #endif
 891
 892
 893 #ifdef CONFIG_PROC_FS
 894 static int netlink_read_proc(char *buffer, char **start, off_t offset,
 895                              int length, int *eof, void *data)
 896 {
 897         off_t pos=0;
 898         off_t begin=0;
 899         int len=0;
 900         int i;
 901         struct sock *s;
 902
 903         len+= sprintf(buffer,"sk       Eth Pid    Groups   "
 904                       "Rmem     Wmem     Dump     Locks\n");
 905
 906         for (i=0; i<MAX_LINKS; i++) {
 907                 read_lock(&nl_table_lock);
 908                 for (s = nl_table[i]; s; s = s->next) {
 909                         len+=sprintf(buffer+len,"%p %-3d %-6d %08x %-8d %-8d %p %d",
 910                                      s,
 911                                      s->protocol,
 912                                      s->protinfo.af_netlink->pid,
 913                                      s->protinfo.af_netlink->groups,
 914                                      atomic_read(&s->rmem_alloc),
 915                                      atomic_read(&s->wmem_alloc),
 916                                      s->protinfo.af_netlink->cb,
 917                                      atomic_read(&s->refcnt)
 918                                      );
 919
 920                         buffer[len++]='\n';
 921
 922                         pos=begin+len;
 923                         if(pos<offset) {
 924                                 len=0;
 925                                 begin=pos;
 926                         }
 927                         if(pos>offset+length) {
 928                                 read_unlock(&nl_table_lock);
 929                                 goto done;
 930                         }
 931                 }
 932                 read_unlock(&nl_table_lock);
 933         }
 934         *eof = 1;
 935
 936 done:
 937         *start=buffer+(offset-begin);
 938         len-=(offset-begin);
 939         if(len>length)
 940                 len=length;
 941         if(len<0)
 942                 len=0;
 943         return len;
 944 }
 945 #endif
 946
 947 struct proto_ops netlink_ops = {
 948         family:         PF_NETLINK,
 949
 950         release:        netlink_release,
 951         bind:           netlink_bind,
 952         connect:        netlink_connect,
 953         socketpair:     sock_no_socketpair,
 954         accept:         sock_no_accept,
 955         getname:        netlink_getname,
 956         poll:           datagram_poll,
 957         ioctl:          sock_no_ioctl,
 958         listen:         sock_no_listen,
 959         shutdown:       sock_no_shutdown,
 960         setsockopt:     sock_no_setsockopt,
 961         getsockopt:     sock_no_getsockopt,
 962         sendmsg:        netlink_sendmsg,
 963         recvmsg:        netlink_recvmsg,
 964         mmap:           sock_no_mmap,
 965 };
 966
 967 struct net_proto_family netlink_family_ops = {
 968         PF_NETLINK,
 969         netlink_create
 970 };
 971
 972 static int __init netlink_proto_init(void)
 973 {
 974         struct sk_buff *dummy_skb;
 975
 976         if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb)) {
 977                 printk(KERN_CRIT "netlink_init: panic\n");
 978                 return -1;
 979         }
 980         sock_register(&netlink_family_ops);
 981 #ifdef CONFIG_PROC_FS
 982         create_proc_read_entry("net/netlink", 0, 0, netlink_read_proc, NULL);
 983 #endif
 984         return 0;
 985 }
 986
 987 module_init(netlink_proto_init);