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  *
  15  */
  16
  17 #include <linux/config.h>
  18 #include <linux/module.h>
  19
  20 #include <linux/kernel.h>
  21 #include <linux/init.h>
  22 #include <linux/major.h>
  23 #include <linux/signal.h>
  24 #include <linux/sched.h>
  25 #include <linux/errno.h>
  26 #include <linux/string.h>
  27 #include <linux/stat.h>
  28 #include <linux/socket.h>
  29 #include <linux/un.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/termios.h>
  32 #include <linux/sockios.h>
  33 #include <linux/net.h>
  34 #include <linux/fs.h>
  35 #include <linux/slab.h>
  36 #include <asm/uaccess.h>
  37 #include <linux/skbuff.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/rtnetlink.h>
  40 #include <linux/proc_fs.h>
  41 #include <linux/smp_lock.h>
  42 #include <linux/notifier.h>
  43 #include <net/sock.h>
  44 #include <net/scm.h>
  45
  46 #define Nprintk(a...)
  47
  48 #if defined(CONFIG_NETLINK_DEV) || defined(CONFIG_NETLINK_DEV_MODULE)
  49 #define NL_EMULATE_DEV
  50 #endif
  51
  52 struct netlink_opt
  53 {
  54         u32                     pid;
  55         unsigned                groups;
  56         u32                     dst_pid;
  57         unsigned                dst_groups;
  58         unsigned long           state;
  59         int                     (*handler)(int unit, struct sk_buff *skb);
  60         wait_queue_head_t       wait;
  61         struct netlink_callback *cb;
  62         spinlock_t              cb_lock;
  63         void                    (*data_ready)(struct sock *sk, int bytes);
  64 };
  65
  66 static struct sock *nl_table[MAX_LINKS];
  67 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
  68 static unsigned nl_nonroot[MAX_LINKS];
  69
  70 #ifdef NL_EMULATE_DEV
  71 static struct socket *netlink_kernel[MAX_LINKS];
  72 #endif
  73
  74 static int netlink_dump(struct sock *sk);
  75 static void netlink_destroy_callback(struct netlink_callback *cb);
  76
  77 atomic_t netlink_sock_nr;
  78
  79 static rwlock_t nl_table_lock = RW_LOCK_UNLOCKED;
  80 static atomic_t nl_table_users = ATOMIC_INIT(0);
  81
  82 static struct notifier_block *netlink_chain;
  83
  84 static void netlink_sock_destruct(struct sock *sk)
  85 {
  86         skb_queue_purge(&sk->receive_queue);
  87
  88         if (!sk->dead) {
  89                 printk("Freeing alive netlink socket %p\n", sk);
  90                 return;
  91         }
  92         BUG_TRAP(atomic_read(&sk->rmem_alloc)==0);
  93         BUG_TRAP(atomic_read(&sk->wmem_alloc)==0);
  94         BUG_TRAP(sk->protinfo.af_netlink->cb==NULL);
  95
  96         kfree(sk->protinfo.af_netlink);
  97
  98         atomic_dec(&netlink_sock_nr);
  99 #ifdef NETLINK_REFCNT_DEBUG
 100         printk(KERN_DEBUG "NETLINK %p released, %d are still alive\n", sk, atomic_read(&netlink_sock_nr));
 101 #endif
 102 }
 103
 104 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
 105  * Look, when several writers sleep and reader wakes them up, all but one
 106  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
 107  * this, _but_ remember, it adds useless work on UP machines.
 108  */
 109
 110 static void netlink_table_grab(void)
 111 {
 112         write_lock_bh(&nl_table_lock);
 113
 114         if (atomic_read(&nl_table_users)) {
 115                 DECLARE_WAITQUEUE(wait, current);
 116
 117                 add_wait_queue_exclusive(&nl_table_wait, &wait);
 118                 for(;;) {
 119                         set_current_state(TASK_UNINTERRUPTIBLE);
 120                         if (atomic_read(&nl_table_users) == 0)
 121                                 break;
 122                         write_unlock_bh(&nl_table_lock);
 123                         schedule();
 124                         write_lock_bh(&nl_table_lock);
 125                 }
 126
 127                 __set_current_state(TASK_RUNNING);
 128                 remove_wait_queue(&nl_table_wait, &wait);
 129         }
 130 }
 131
 132 static __inline__ void netlink_table_ungrab(void)
 133 {
 134         write_unlock_bh(&nl_table_lock);
 135         wake_up(&nl_table_wait);
 136 }
 137
 138 static __inline__ void
 139 netlink_lock_table(void)
 140 {
 141         /* read_lock() synchronizes us to netlink_table_grab */
 142
 143         read_lock(&nl_table_lock);
 144         atomic_inc(&nl_table_users);
 145         read_unlock(&nl_table_lock);
 146 }
 147
 148 static __inline__ void
 149 netlink_unlock_table(void)
 150 {
 151         if (atomic_dec_and_test(&nl_table_users))
 152                 wake_up(&nl_table_wait);
 153 }
 154
 155 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
 156 {
 157         struct sock *sk;
 158
 159         read_lock(&nl_table_lock);
 160         for (sk=nl_table[protocol]; sk; sk=sk->next) {
 161                 if (sk->protinfo.af_netlink->pid == pid) {
 162                         sock_hold(sk);
 163                         read_unlock(&nl_table_lock);
 164                         return sk;
 165                 }
 166         }
 167
 168         read_unlock(&nl_table_lock);
 169         return NULL;
 170 }
 171
 172 extern struct proto_ops netlink_ops;
 173
 174 static int netlink_insert(struct sock *sk, u32 pid)
 175 {
 176         int err = -EADDRINUSE;
 177         struct sock *osk;
 178
 179         netlink_table_grab();
 180         for (osk=nl_table[sk->protocol]; osk; osk=osk->next) {
 181                 if (osk->protinfo.af_netlink->pid == pid)
 182                         break;
 183         }
 184         if (osk == NULL) {
 185                 err = -EBUSY;
 186                 if (sk->protinfo.af_netlink->pid == 0) {
 187                         sk->protinfo.af_netlink->pid = pid;
 188                         sk->next = nl_table[sk->protocol];
 189                         nl_table[sk->protocol] = sk;
 190                         sock_hold(sk);
 191                         err = 0;
 192                 }
 193         }
 194         netlink_table_ungrab();
 195         return err;
 196 }
 197
 198 static void netlink_remove(struct sock *sk)
 199 {
 200         struct sock **skp;
 201
 202         netlink_table_grab();
 203         for (skp = &nl_table[sk->protocol]; *skp; skp = &((*skp)->next)) {
 204                 if (*skp == sk) {
 205                         *skp = sk->next;
 206                         __sock_put(sk);
 207                         break;
 208                 }
 209         }
 210         netlink_table_ungrab();
 211 }
 212
 213 static int netlink_create(struct socket *sock, int protocol)
 214 {
 215         struct sock *sk;
 216
 217         sock->state = SS_UNCONNECTED;
 218
 219         if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
 220                 return -ESOCKTNOSUPPORT;
 221
 222         if (protocol<0 || protocol >= MAX_LINKS)
 223                 return -EPROTONOSUPPORT;
 224
 225         sock->ops = &netlink_ops;
 226
 227         sk = sk_alloc(PF_NETLINK, GFP_KERNEL, 1);
 228         if (!sk)
 229                 return -ENOMEM;
 230
 231         sock_init_data(sock,sk);
 232
 233         sk->protinfo.af_netlink = kmalloc(sizeof(struct netlink_opt), GFP_KERNEL);
 234         if (sk->protinfo.af_netlink == NULL) {
 235                 sk_free(sk);
 236                 return -ENOMEM;
 237         }
 238         memset(sk->protinfo.af_netlink, 0, sizeof(struct netlink_opt));
 239
 240         spin_lock_init(&sk->protinfo.af_netlink->cb_lock);
 241         init_waitqueue_head(&sk->protinfo.af_netlink->wait);
 242         sk->destruct = netlink_sock_destruct;
 243         atomic_inc(&netlink_sock_nr);
 244
 245         sk->protocol=protocol;
 246         return 0;
 247 }
 248
 249 static int netlink_release(struct socket *sock)
 250 {
 251         struct sock *sk = sock->sk;
 252
 253         if (!sk)
 254                 return 0;
 255
 256         netlink_remove(sk);
 257
 258         spin_lock(&sk->protinfo.af_netlink->cb_lock);
 259         if (sk->protinfo.af_netlink->cb) {
 260                 sk->protinfo.af_netlink->cb->done(sk->protinfo.af_netlink->cb);
 261                 netlink_destroy_callback(sk->protinfo.af_netlink->cb);
 262                 sk->protinfo.af_netlink->cb = NULL;
 263                 __sock_put(sk);
 264         }
 265         spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 266
 267         /* OK. Socket is unlinked, and, therefore,
 268            no new packets will arrive */
 269
 270         sock_orphan(sk);
 271         sock->sk = NULL;
 272         wake_up_interruptible_all(&sk->protinfo.af_netlink->wait);
 273
 274         skb_queue_purge(&sk->write_queue);
 275
 276         if (sk->protinfo.af_netlink->pid && !sk->protinfo.af_netlink->groups) {
 277                 struct netlink_notify n = { protocol:sk->protocol,
 278                                             pid:sk->protinfo.af_netlink->pid };
 279                 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
 280         }
 281
 282         sock_put(sk);
 283         return 0;
 284 }
 285
 286 static int netlink_autobind(struct socket *sock)
 287 {
 288         struct sock *sk = sock->sk;
 289         struct sock *osk;
 290         s32 pid = current->pid;
 291         int err;
 292
 293 retry:
 294         netlink_table_grab();
 295         for (osk=nl_table[sk->protocol]; osk; osk=osk->next) {
 296                 if (osk->protinfo.af_netlink->pid == pid) {
 297                         /* Bind collision, search negative pid values. */
 298                         if (pid > 0)
 299                                 pid = -4096;
 300                         pid--;
 301                         netlink_table_ungrab();
 302                         goto retry;
 303                 }
 304         }
 305         netlink_table_ungrab();
 306
 307         err = netlink_insert(sk, pid);
 308         if (err == -EADDRINUSE)
 309                 goto retry;
 310         sk->protinfo.af_netlink->groups = 0;
 311         return 0;
 312 }
 313
 314 static inline int netlink_capable(struct socket *sock, unsigned flag)
 315 {
 316         return (nl_nonroot[sock->sk->protocol] & flag) || capable(CAP_NET_ADMIN);
 317 }
 318
 319 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 320 {
 321         struct sock *sk = sock->sk;
 322         int err;
 323         struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
 324
 325         if (nladdr->nl_family != AF_NETLINK)
 326                 return -EINVAL;
 327
 328         /* Only superuser is allowed to listen multicasts */
 329         if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV))
 330                 return -EPERM;
 331
 332         if (sk->protinfo.af_netlink->pid) {
 333                 if (nladdr->nl_pid != sk->protinfo.af_netlink->pid)
 334                         return -EINVAL;
 335                 sk->protinfo.af_netlink->groups = nladdr->nl_groups;
 336                 return 0;
 337         }
 338
 339         if (nladdr->nl_pid == 0) {
 340                 err = netlink_autobind(sock);
 341                 if (err == 0)
 342                         sk->protinfo.af_netlink->groups = nladdr->nl_groups;
 343                 return err;
 344         }
 345
 346         err = netlink_insert(sk, nladdr->nl_pid);
 347         if (err == 0)
 348                 sk->protinfo.af_netlink->groups = nladdr->nl_groups;
 349         return err;
 350 }
 351
 352 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
 353                            int alen, int flags)
 354 {
 355         int err = 0;
 356         struct sock *sk = sock->sk;
 357         struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
 358
 359         if (addr->sa_family == AF_UNSPEC) {
 360                 sk->protinfo.af_netlink->dst_pid = 0;
 361                 sk->protinfo.af_netlink->dst_groups = 0;
 362                 return 0;
 363         }
 364         if (addr->sa_family != AF_NETLINK)
 365                 return -EINVAL;
 366
 367         /* Only superuser is allowed to send multicasts */
 368         if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
 369                 return -EPERM;
 370
 371         if (!sk->protinfo.af_netlink->pid)
 372                 err = netlink_autobind(sock);
 373
 374         if (err == 0) {
 375                 sk->protinfo.af_netlink->dst_pid = nladdr->nl_pid;
 376                 sk->protinfo.af_netlink->dst_groups = nladdr->nl_groups;
 377         }
 378
 379         return 0;
 380 }
 381
 382 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
 383 {
 384         struct sock *sk = sock->sk;
 385         struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
 386
 387         nladdr->nl_family = AF_NETLINK;
 388         *addr_len = sizeof(*nladdr);
 389
 390         if (peer) {
 391                 nladdr->nl_pid = sk->protinfo.af_netlink->dst_pid;
 392                 nladdr->nl_groups = sk->protinfo.af_netlink->dst_groups;
 393         } else {
 394                 nladdr->nl_pid = sk->protinfo.af_netlink->pid;
 395                 nladdr->nl_groups = sk->protinfo.af_netlink->groups;
 396         }
 397         return 0;
 398 }
 399
 400 static void netlink_overrun(struct sock *sk)
 401 {
 402         if (!test_and_set_bit(0, &sk->protinfo.af_netlink->state)) {
 403                 sk->err = ENOBUFS;
 404                 sk->error_report(sk);
 405         }
 406 }
 407
 408 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
 409 {
 410         struct sock *sk;
 411         int len = skb->len;
 412         int protocol = ssk->protocol;
 413         long timeo;
 414         DECLARE_WAITQUEUE(wait, current);
 415
 416         timeo = sock_sndtimeo(ssk, nonblock);
 417
 418 retry:
 419         sk = netlink_lookup(protocol, pid);
 420         if (sk == NULL)
 421                 goto no_dst;
 422
 423 #ifdef NL_EMULATE_DEV
 424         if (sk->protinfo.af_netlink->handler) {
 425                 skb_orphan(skb);
 426                 len = sk->protinfo.af_netlink->handler(protocol, skb);
 427                 sock_put(sk);
 428                 return len;
 429         }
 430 #endif
 431
 432         if (atomic_read(&sk->rmem_alloc) > sk->rcvbuf ||
 433             test_bit(0, &sk->protinfo.af_netlink->state)) {
 434                 if (!timeo) {
 435                         if (ssk->protinfo.af_netlink->pid == 0)
 436                                 netlink_overrun(sk);
 437                         sock_put(sk);
 438                         kfree_skb(skb);
 439                         return -EAGAIN;
 440                 }
 441
 442                 __set_current_state(TASK_INTERRUPTIBLE);
 443                 add_wait_queue(&sk->protinfo.af_netlink->wait, &wait);
 444
 445                 if ((atomic_read(&sk->rmem_alloc) > sk->rcvbuf ||
 446                     test_bit(0, &sk->protinfo.af_netlink->state)) &&
 447                     !sk->dead)
 448                         timeo = schedule_timeout(timeo);
 449
 450                 __set_current_state(TASK_RUNNING);
 451                 remove_wait_queue(&sk->protinfo.af_netlink->wait, &wait);
 452                 sock_put(sk);
 453
 454                 if (signal_pending(current)) {
 455                         kfree_skb(skb);
 456                         return sock_intr_errno(timeo);
 457                 }
 458                 goto retry;
 459         }
 460
 461         skb_orphan(skb);
 462         skb_set_owner_r(skb, sk);
 463         skb_queue_tail(&sk->receive_queue, skb);
 464         sk->data_ready(sk, len);
 465         sock_put(sk);
 466         return len;
 467
 468 no_dst:
 469         kfree_skb(skb);
 470         return -ECONNREFUSED;
 471 }
 472
 473 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
 474 {
 475 #ifdef NL_EMULATE_DEV
 476         if (sk->protinfo.af_netlink->handler) {
 477                 skb_orphan(skb);
 478                 sk->protinfo.af_netlink->handler(sk->protocol, skb);
 479                 return 0;
 480         } else
 481 #endif
 482         if (atomic_read(&sk->rmem_alloc) <= sk->rcvbuf &&
 483             !test_bit(0, &sk->protinfo.af_netlink->state)) {
 484                 skb_orphan(skb);
 485                 skb_set_owner_r(skb, sk);
 486                 skb_queue_tail(&sk->receive_queue, skb);
 487                 sk->data_ready(sk, skb->len);
 488                 return 0;
 489         }
 490         return -1;
 491 }
 492
 493 void netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
 494                        u32 group, int allocation)
 495 {
 496         struct sock *sk;
 497         struct sk_buff *skb2 = NULL;
 498         int protocol = ssk->protocol;
 499         int failure = 0;
 500
 501         /* While we sleep in clone, do not allow to change socket list */
 502
 503         netlink_lock_table();
 504
 505         for (sk = nl_table[protocol]; sk; sk = sk->next) {
 506                 if (ssk == sk)
 507                         continue;
 508
 509                 if (sk->protinfo.af_netlink->pid == pid ||
 510                     !(sk->protinfo.af_netlink->groups&group))
 511                         continue;
 512
 513                 if (failure) {
 514                         netlink_overrun(sk);
 515                         continue;
 516                 }
 517
 518                 sock_hold(sk);
 519                 if (skb2 == NULL) {
 520                         if (atomic_read(&skb->users) != 1) {
 521                                 skb2 = skb_clone(skb, allocation);
 522                         } else {
 523                                 skb2 = skb;
 524                                 atomic_inc(&skb->users);
 525                         }
 526                 }
 527                 if (skb2 == NULL) {
 528                         netlink_overrun(sk);
 529                         /* Clone failed. Notify ALL listeners. */
 530                         failure = 1;
 531                 } else if (netlink_broadcast_deliver(sk, skb2)) {
 532                         netlink_overrun(sk);
 533                 } else
 534                         skb2 = NULL;
 535                 sock_put(sk);
 536         }
 537
 538         netlink_unlock_table();
 539
 540         if (skb2)
 541                 kfree_skb(skb2);
 542         kfree_skb(skb);
 543 }
 544
 545 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
 546 {
 547         struct sock *sk;
 548         int protocol = ssk->protocol;
 549
 550         read_lock(&nl_table_lock);
 551         for (sk = nl_table[protocol]; sk; sk = sk->next) {
 552                 if (ssk == sk)
 553                         continue;
 554
 555                 if (sk->protinfo.af_netlink->pid == pid ||
 556                     !(sk->protinfo.af_netlink->groups&group))
 557                         continue;
 558
 559                 sk->err = code;
 560                 sk->error_report(sk);
 561         }
 562         read_unlock(&nl_table_lock);
 563 }
 564
 565 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, int len,
 566                            struct scm_cookie *scm)
 567 {
 568         struct sock *sk = sock->sk;
 569         struct sockaddr_nl *addr=msg->msg_name;
 570         u32 dst_pid;
 571         u32 dst_groups;
 572         struct sk_buff *skb;
 573         int err;
 574
 575         if (msg->msg_flags&MSG_OOB)
 576                 return -EOPNOTSUPP;
 577
 578         if (msg->msg_namelen) {
 579                 if (addr->nl_family != AF_NETLINK)
 580                         return -EINVAL;
 581                 dst_pid = addr->nl_pid;
 582                 dst_groups = addr->nl_groups;
 583                 if (dst_groups && !netlink_capable(sock, NL_NONROOT_SEND))
 584                         return -EPERM;
 585         } else {
 586                 dst_pid = sk->protinfo.af_netlink->dst_pid;
 587                 dst_groups = sk->protinfo.af_netlink->dst_groups;
 588         }
 589
 590         if (!sk->protinfo.af_netlink->pid) {
 591                 err = netlink_autobind(sock);
 592                 if (err)
 593                         goto out;
 594         }
 595
 596         err = -EMSGSIZE;
 597         if ((unsigned)len > sk->sndbuf-32)
 598                 goto out;
 599         err = -ENOBUFS;
 600         skb = alloc_skb(len, GFP_KERNEL);
 601         if (skb==NULL)
 602                 goto out;
 603
 604         NETLINK_CB(skb).pid = sk->protinfo.af_netlink->pid;
 605         NETLINK_CB(skb).groups = sk->protinfo.af_netlink->groups;
 606         NETLINK_CB(skb).dst_pid = dst_pid;
 607         NETLINK_CB(skb).dst_groups = dst_groups;
 608         memcpy(NETLINK_CREDS(skb), &scm->creds, sizeof(struct ucred));
 609
 610         /* What can I do? Netlink is asynchronous, so that
 611            we will have to save current capabilities to
 612            check them, when this message will be delivered
 613            to corresponding kernel module.   --ANK (980802)
 614          */
 615         NETLINK_CB(skb).eff_cap = current->cap_effective;
 616
 617         err = -EFAULT;
 618         if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
 619                 kfree_skb(skb);
 620                 goto out;
 621         }
 622
 623         if (dst_groups) {
 624                 atomic_inc(&skb->users);
 625                 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL);
 626         }
 627         err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
 628
 629 out:
 630         return err;
 631 }
 632
 633 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, int len,
 634                            int flags, struct scm_cookie *scm)
 635 {
 636         struct sock *sk = sock->sk;
 637         int noblock = flags&MSG_DONTWAIT;
 638         int copied;
 639         struct sk_buff *skb;
 640         int err;
 641
 642         if (flags&MSG_OOB)
 643                 return -EOPNOTSUPP;
 644
 645         copied = 0;
 646
 647         skb = skb_recv_datagram(sk,flags,noblock,&err);
 648         if (skb==NULL)
 649                 goto out;
 650
 651         msg->msg_namelen = 0;
 652
 653         copied = skb->len;
 654         if (len < copied) {
 655                 msg->msg_flags |= MSG_TRUNC;
 656                 copied = len;
 657         }
 658
 659         skb->h.raw = skb->data;
 660         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
 661
 662         if (msg->msg_name) {
 663                 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
 664                 addr->nl_family = AF_NETLINK;
 665                 addr->nl_pid    = NETLINK_CB(skb).pid;
 666                 addr->nl_groups = NETLINK_CB(skb).dst_groups;
 667                 msg->msg_namelen = sizeof(*addr);
 668         }
 669
 670         scm->creds = *NETLINK_CREDS(skb);
 671         skb_free_datagram(sk, skb);
 672
 673         if (sk->protinfo.af_netlink->cb
 674             && atomic_read(&sk->rmem_alloc) <= sk->rcvbuf/2)
 675                 netlink_dump(sk);
 676
 677 out:
 678         if (skb_queue_len(&sk->receive_queue) <= sk->rcvbuf/2) {
 679                 if (skb_queue_len(&sk->receive_queue) == 0)
 680                         clear_bit(0, &sk->protinfo.af_netlink->state);
 681                 if (!test_bit(0, &sk->protinfo.af_netlink->state))
 682                         wake_up_interruptible(&sk->protinfo.af_netlink->wait);
 683         }
 684         return err ? : copied;
 685 }
 686
 687 void netlink_data_ready(struct sock *sk, int len)
 688 {
 689         if (sk->protinfo.af_netlink->data_ready)
 690                 sk->protinfo.af_netlink->data_ready(sk, len);
 691
 692         if (skb_queue_len(&sk->receive_queue) <= sk->rcvbuf/2) {
 693                 if (skb_queue_len(&sk->receive_queue) == 0)
 694                         clear_bit(0, &sk->protinfo.af_netlink->state);
 695                 if (!test_bit(0, &sk->protinfo.af_netlink->state))
 696                         wake_up_interruptible(&sk->protinfo.af_netlink->wait);
 697         }
 698 }
 699
 700 /*
 701  *      We export these functions to other modules. They provide a
 702  *      complete set of kernel non-blocking support for message
 703  *      queueing.
 704  */
 705
 706 struct sock *
 707 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len))
 708 {
 709         struct socket *sock;
 710         struct sock *sk;
 711
 712         if (unit<0 || unit>=MAX_LINKS)
 713                 return NULL;
 714
 715         if (!(sock = sock_alloc()))
 716                 return NULL;
 717
 718         sock->type = SOCK_RAW;
 719
 720         if (netlink_create(sock, unit) < 0) {
 721                 sock_release(sock);
 722                 return NULL;
 723         }
 724         sk = sock->sk;
 725         sk->data_ready = netlink_data_ready;
 726         if (input)
 727                 sk->protinfo.af_netlink->data_ready = input;
 728
 729         netlink_insert(sk, 0);
 730         return sk;
 731 }
 732
 733 void netlink_set_nonroot(int protocol, unsigned flags)
 734 {
 735         if ((unsigned)protocol < MAX_LINKS)
 736                 nl_nonroot[protocol] = flags;
 737 }
 738
 739 static void netlink_destroy_callback(struct netlink_callback *cb)
 740 {
 741         if (cb->skb)
 742                 kfree_skb(cb->skb);
 743         kfree(cb);
 744 }
 745
 746 /*
 747  * It looks a bit ugly.
 748  * It would be better to create kernel thread.
 749  */
 750
 751 static int netlink_dump(struct sock *sk)
 752 {
 753         struct netlink_callback *cb;
 754         struct sk_buff *skb;
 755         struct nlmsghdr *nlh;
 756         int len;
 757
 758         skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
 759         if (!skb)
 760                 return -ENOBUFS;
 761
 762         spin_lock(&sk->protinfo.af_netlink->cb_lock);
 763
 764         cb = sk->protinfo.af_netlink->cb;
 765         if (cb == NULL) {
 766                 spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 767                 kfree_skb(skb);
 768                 return -EINVAL;
 769         }
 770
 771         len = cb->dump(skb, cb);
 772
 773         if (len > 0) {
 774                 spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 775                 skb_queue_tail(&sk->receive_queue, skb);
 776                 sk->data_ready(sk, len);
 777                 return 0;
 778         }
 779
 780         nlh = __nlmsg_put(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, NLMSG_DONE, sizeof(int));
 781         nlh->nlmsg_flags |= NLM_F_MULTI;
 782         memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
 783         skb_queue_tail(&sk->receive_queue, skb);
 784         sk->data_ready(sk, skb->len);
 785
 786         cb->done(cb);
 787         sk->protinfo.af_netlink->cb = NULL;
 788         spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 789
 790         netlink_destroy_callback(cb);
 791         sock_put(sk);
 792         return 0;
 793 }
 794
 795 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
 796                        struct nlmsghdr *nlh,
 797                        int (*dump)(struct sk_buff *skb, struct netlink_callback*),
 798                        int (*done)(struct netlink_callback*))
 799 {
 800         struct netlink_callback *cb;
 801         struct sock *sk;
 802
 803         cb = kmalloc(sizeof(*cb), GFP_KERNEL);
 804         if (cb == NULL)
 805                 return -ENOBUFS;
 806
 807         memset(cb, 0, sizeof(*cb));
 808         cb->dump = dump;
 809         cb->done = done;
 810         cb->nlh = nlh;
 811         atomic_inc(&skb->users);
 812         cb->skb = skb;
 813
 814         sk = netlink_lookup(ssk->protocol, NETLINK_CB(skb).pid);
 815         if (sk == NULL) {
 816                 netlink_destroy_callback(cb);
 817                 return -ECONNREFUSED;
 818         }
 819         /* A dump is in progress... */
 820         spin_lock(&sk->protinfo.af_netlink->cb_lock);
 821         if (sk->protinfo.af_netlink->cb) {
 822                 spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 823                 netlink_destroy_callback(cb);
 824                 sock_put(sk);
 825                 return -EBUSY;
 826         }
 827         sk->protinfo.af_netlink->cb = cb;
 828         spin_unlock(&sk->protinfo.af_netlink->cb_lock);
 829
 830         netlink_dump(sk);
 831         return 0;
 832 }
 833
 834 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
 835 {
 836         struct sk_buff *skb;
 837         struct nlmsghdr *rep;
 838         struct nlmsgerr *errmsg;
 839         int size;
 840
 841         if (err == 0)
 842                 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
 843         else
 844                 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
 845
 846         skb = alloc_skb(size, GFP_KERNEL);
 847         if (!skb)
 848                 return;
 849
 850         rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
 851                           NLMSG_ERROR, sizeof(struct nlmsgerr));
 852         errmsg = NLMSG_DATA(rep);
 853         errmsg->error = err;
 854         memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
 855         netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
 856 }
 857
 858
 859 #ifdef NL_EMULATE_DEV
 860
 861 static rwlock_t nl_emu_lock = RW_LOCK_UNLOCKED;
 862
 863 /*
 864  *      Backward compatibility.
 865  */
 866
 867 int netlink_attach(int unit, int (*function)(int, struct sk_buff *skb))
 868 {
 869         struct sock *sk = netlink_kernel_create(unit, NULL);
 870         if (sk == NULL)
 871                 return -ENOBUFS;
 872         sk->protinfo.af_netlink->handler = function;
 873         write_lock_bh(&nl_emu_lock);
 874         netlink_kernel[unit] = sk->socket;
 875         write_unlock_bh(&nl_emu_lock);
 876         return 0;
 877 }
 878
 879 void netlink_detach(int unit)
 880 {
 881         struct socket *sock;
 882
 883         write_lock_bh(&nl_emu_lock);
 884         sock = netlink_kernel[unit];
 885         netlink_kernel[unit] = NULL;
 886         write_unlock_bh(&nl_emu_lock);
 887
 888         sock_release(sock);
 889 }
 890
 891 int netlink_post(int unit, struct sk_buff *skb)
 892 {
 893         struct socket *sock;
 894
 895         read_lock(&nl_emu_lock);
 896         sock = netlink_kernel[unit];
 897         if (sock) {
 898                 struct sock *sk = sock->sk;
 899                 memset(skb->cb, 0, sizeof(skb->cb));
 900                 sock_hold(sk);
 901                 read_unlock(&nl_emu_lock);
 902
 903                 netlink_broadcast(sk, skb, 0, ~0, GFP_ATOMIC);
 904
 905                 sock_put(sk);
 906                 return 0;
 907         }
 908         read_unlock(&nl_emu_lock);
 909         return -EUNATCH;
 910 }
 911
 912 #endif
 913
 914
 915 #ifdef CONFIG_PROC_FS
 916 static int netlink_read_proc(char *buffer, char **start, off_t offset,
 917                              int length, int *eof, void *data)
 918 {
 919         off_t pos=0;
 920         off_t begin=0;
 921         int len=0;
 922         int i;
 923         struct sock *s;
 924
 925         len+= sprintf(buffer,"sk       Eth Pid    Groups   "
 926                       "Rmem     Wmem     Dump     Locks\n");
 927
 928         for (i=0; i<MAX_LINKS; i++) {
 929                 read_lock(&nl_table_lock);
 930                 for (s = nl_table[i]; s; s = s->next) {
 931                         len+=sprintf(buffer+len,"%p %-3d %-6d %08x %-8d %-8d %p %d",
 932                                      s,
 933                                      s->protocol,
 934                                      s->protinfo.af_netlink->pid,
 935                                      s->protinfo.af_netlink->groups,
 936                                      atomic_read(&s->rmem_alloc),
 937                                      atomic_read(&s->wmem_alloc),
 938                                      s->protinfo.af_netlink->cb,
 939                                      atomic_read(&s->refcnt)
 940                                      );
 941
 942                         buffer[len++]='\n';
 943
 944                         pos=begin+len;
 945                         if(pos<offset) {
 946                                 len=0;
 947                                 begin=pos;
 948                         }
 949                         if(pos>offset+length) {
 950                                 read_unlock(&nl_table_lock);
 951                                 goto done;
 952                         }
 953                 }
 954                 read_unlock(&nl_table_lock);
 955         }
 956         *eof = 1;
 957
 958 done:
 959         *start=buffer+(offset-begin);
 960         len-=(offset-begin);
 961         if(len>length)
 962                 len=length;
 963         if(len<0)
 964                 len=0;
 965         return len;
 966 }
 967 #endif
 968
 969 int netlink_register_notifier(struct notifier_block *nb)
 970 {
 971         return notifier_chain_register(&netlink_chain, nb);
 972 }
 973
 974 int netlink_unregister_notifier(struct notifier_block *nb)
 975 {
 976         return notifier_chain_unregister(&netlink_chain, nb);
 977 }
 978
 979 struct proto_ops netlink_ops = {
 980         family:         PF_NETLINK,
 981
 982         release:        netlink_release,
 983         bind:           netlink_bind,
 984         connect:        netlink_connect,
 985         socketpair:     sock_no_socketpair,
 986         accept:         sock_no_accept,
 987         getname:        netlink_getname,
 988         poll:           datagram_poll,
 989         ioctl:          sock_no_ioctl,
 990         listen:         sock_no_listen,
 991         shutdown:       sock_no_shutdown,
 992         setsockopt:     sock_no_setsockopt,
 993         getsockopt:     sock_no_getsockopt,
 994         sendmsg:        netlink_sendmsg,
 995         recvmsg:        netlink_recvmsg,
 996         mmap:           sock_no_mmap,
 997         sendpage:       sock_no_sendpage,
 998 };
 999
1000 struct net_proto_family netlink_family_ops = {
1001         PF_NETLINK,
1002         netlink_create
1003 };
1004
1005 static int __init netlink_proto_init(void)
1006 {
1007         struct sk_buff *dummy_skb;
1008
1009         if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb)) {
1010                 printk(KERN_CRIT "netlink_init: panic\n");
1011                 return -1;
1012         }
1013         sock_register(&netlink_family_ops);
1014 #ifdef CONFIG_PROC_FS
1015         create_proc_read_entry("net/netlink", 0, 0, netlink_read_proc, NULL);
1016 #endif
1017         return 0;
1018 }
1019
1020 static void __exit netlink_proto_exit(void)
1021 {
1022        sock_unregister(PF_NETLINK);
1023        remove_proc_entry("net/netlink", NULL);
1024 }
1025
1026 module_init(netlink_proto_init);
1027 module_exit(netlink_proto_exit);