net/rds/af_rds.c

   1 /*
   2  * Copyright (c) 2006 Oracle.  All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  *
  32  */
  33 #include <linux/module.h>
  34 #include <linux/errno.h>
  35 #include <linux/kernel.h>
  36 #include <linux/gfp.h>
  37 #include <linux/in.h>
  38 #include <linux/poll.h>
  39 #include <net/sock.h>
  40
  41 #include "rds.h"
  42
  43 /* this is just used for stats gathering :/ */
  44 static DEFINE_SPINLOCK(rds_sock_lock);
  45 static unsigned long rds_sock_count;
  46 static LIST_HEAD(rds_sock_list);
  47 DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);
  48
  49 /*
  50  * This is called as the final descriptor referencing this socket is closed.
  51  * We have to unbind the socket so that another socket can be bound to the
  52  * address it was using.
  53  *
  54  * We have to be careful about racing with the incoming path.  sock_orphan()
  55  * sets SOCK_DEAD and we use that as an indicator to the rx path that new
  56  * messages shouldn't be queued.
  57  */
  58 static int rds_release(struct socket *sock)
  59 {
  60         struct sock *sk = sock->sk;
  61         struct rds_sock *rs;
  62
  63         if (!sk)
  64                 goto out;
  65
  66         rs = rds_sk_to_rs(sk);
  67
  68         sock_orphan(sk);
  69         /* Note - rds_clear_recv_queue grabs rs_recv_lock, so
  70          * that ensures the recv path has completed messing
  71          * with the socket. */
  72         rds_clear_recv_queue(rs);
  73         rds_cong_remove_socket(rs);
  74
  75         rds_remove_bound(rs);
  76
  77         rds_send_drop_to(rs, NULL);
  78         rds_rdma_drop_keys(rs);
  79         rds_notify_queue_get(rs, NULL);
  80
  81         spin_lock_bh(&rds_sock_lock);
  82         list_del_init(&rs->rs_item);
  83         rds_sock_count--;
  84         spin_unlock_bh(&rds_sock_lock);
  85
  86         rds_trans_put(rs->rs_transport);
  87
  88         sock->sk = NULL;
  89         sock_put(sk);
  90 out:
  91         return 0;
  92 }
  93
  94 /*
  95  * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
  96  * _bh() isn't OK here, we're called from interrupt handlers.  It's probably OK
  97  * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
  98  * this seems more conservative.
  99  * NB - normally, one would use sk_callback_lock for this, but we can
 100  * get here from interrupts, whereas the network code grabs sk_callback_lock
 101  * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
 102  */
 103 void rds_wake_sk_sleep(struct rds_sock *rs)
 104 {
 105         unsigned long flags;
 106
 107         read_lock_irqsave(&rs->rs_recv_lock, flags);
 108         __rds_wake_sk_sleep(rds_rs_to_sk(rs));
 109         read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 110 }
 111
 112 static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
 113                        int *uaddr_len, int peer)
 114 {
 115         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 116         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 117
 118         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 119
 120         /* racey, don't care */
 121         if (peer) {
 122                 if (!rs->rs_conn_addr)
 123                         return -ENOTCONN;
 124
 125                 sin->sin_port = rs->rs_conn_port;
 126                 sin->sin_addr.s_addr = rs->rs_conn_addr;
 127         } else {
 128                 sin->sin_port = rs->rs_bound_port;
 129                 sin->sin_addr.s_addr = rs->rs_bound_addr;
 130         }
 131
 132         sin->sin_family = AF_INET;
 133
 134         *uaddr_len = sizeof(*sin);
 135         return 0;
 136 }
 137
 138 /*
 139  * RDS' poll is without a doubt the least intuitive part of the interface,
 140  * as POLLIN and POLLOUT do not behave entirely as you would expect from
 141  * a network protocol.
 142  *
 143  * POLLIN is asserted if
 144  *  -   there is data on the receive queue.
 145  *  -   to signal that a previously congested destination may have become
 146  *      uncongested
 147  *  -   A notification has been queued to the socket (this can be a congestion
 148  *      update, or a RDMA completion).
 149  *
 150  * POLLOUT is asserted if there is room on the send queue. This does not mean
 151  * however, that the next sendmsg() call will succeed. If the application tries
 152  * to send to a congested destination, the system call may still fail (and
 153  * return ENOBUFS).
 154  */
 155 static unsigned int rds_poll(struct file *file, struct socket *sock,
 156                              poll_table *wait)
 157 {
 158         struct sock *sk = sock->sk;
 159         struct rds_sock *rs = rds_sk_to_rs(sk);
 160         unsigned int mask = 0;
 161         unsigned long flags;
 162
 163         poll_wait(file, sk_sleep(sk), wait);
 164
 165         if (rs->rs_seen_congestion)
 166                 poll_wait(file, &rds_poll_waitq, wait);
 167
 168         read_lock_irqsave(&rs->rs_recv_lock, flags);
 169         if (!rs->rs_cong_monitor) {
 170                 /* When a congestion map was updated, we signal POLLIN for
 171                  * "historical" reasons. Applications can also poll for
 172                  * WRBAND instead. */
 173                 if (rds_cong_updated_since(&rs->rs_cong_track))
 174                         mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
 175         } else {
 176                 spin_lock(&rs->rs_lock);
 177                 if (rs->rs_cong_notify)
 178                         mask |= (POLLIN | POLLRDNORM);
 179                 spin_unlock(&rs->rs_lock);
 180         }
 181         if (!list_empty(&rs->rs_recv_queue) ||
 182             !list_empty(&rs->rs_notify_queue))
 183                 mask |= (POLLIN | POLLRDNORM);
 184         if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
 185                 mask |= (POLLOUT | POLLWRNORM);
 186         read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 187
 188         /* clear state any time we wake a seen-congested socket */
 189         if (mask)
 190                 rs->rs_seen_congestion = 0;
 191
 192         return mask;
 193 }
 194
 195 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 196 {
 197         return -ENOIOCTLCMD;
 198 }
 199
 200 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
 201                               int len)
 202 {
 203         struct sockaddr_in sin;
 204         int ret = 0;
 205
 206         /* racing with another thread binding seems ok here */
 207         if (rs->rs_bound_addr == 0) {
 208                 ret = -ENOTCONN; /* XXX not a great errno */
 209                 goto out;
 210         }
 211
 212         if (len < sizeof(struct sockaddr_in)) {
 213                 ret = -EINVAL;
 214                 goto out;
 215         }
 216
 217         if (copy_from_user(&sin, optval, sizeof(sin))) {
 218                 ret = -EFAULT;
 219                 goto out;
 220         }
 221
 222         rds_send_drop_to(rs, &sin);
 223 out:
 224         return ret;
 225 }
 226
 227 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
 228                                int optlen)
 229 {
 230         int value;
 231
 232         if (optlen < sizeof(int))
 233                 return -EINVAL;
 234         if (get_user(value, (int __user *) optval))
 235                 return -EFAULT;
 236         *optvar = !!value;
 237         return 0;
 238 }
 239
 240 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
 241                             int optlen)
 242 {
 243         int ret;
 244
 245         ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
 246         if (ret == 0) {
 247                 if (rs->rs_cong_monitor) {
 248                         rds_cong_add_socket(rs);
 249                 } else {
 250                         rds_cong_remove_socket(rs);
 251                         rs->rs_cong_mask = 0;
 252                         rs->rs_cong_notify = 0;
 253                 }
 254         }
 255         return ret;
 256 }
 257
 258 static int rds_set_transport(struct rds_sock *rs, char __user *optval,
 259                              int optlen)
 260 {
 261         int t_type;
 262
 263         if (rs->rs_transport)
 264                 return -EOPNOTSUPP; /* previously attached to transport */
 265
 266         if (optlen != sizeof(int))
 267                 return -EINVAL;
 268
 269         if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type)))
 270                 return -EFAULT;
 271
 272         if (t_type < 0 || t_type >= RDS_TRANS_COUNT)
 273                 return -EINVAL;
 274
 275         rs->rs_transport = rds_trans_get(t_type);
 276
 277         return rs->rs_transport ? 0 : -ENOPROTOOPT;
 278 }
 279
 280 static int rds_setsockopt(struct socket *sock, int level, int optname,
 281                           char __user *optval, unsigned int optlen)
 282 {
 283         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 284         int ret;
 285
 286         if (level != SOL_RDS) {
 287                 ret = -ENOPROTOOPT;
 288                 goto out;
 289         }
 290
 291         switch (optname) {
 292         case RDS_CANCEL_SENT_TO:
 293                 ret = rds_cancel_sent_to(rs, optval, optlen);
 294                 break;
 295         case RDS_GET_MR:
 296                 ret = rds_get_mr(rs, optval, optlen);
 297                 break;
 298         case RDS_GET_MR_FOR_DEST:
 299                 ret = rds_get_mr_for_dest(rs, optval, optlen);
 300                 break;
 301         case RDS_FREE_MR:
 302                 ret = rds_free_mr(rs, optval, optlen);
 303                 break;
 304         case RDS_RECVERR:
 305                 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
 306                 break;
 307         case RDS_CONG_MONITOR:
 308                 ret = rds_cong_monitor(rs, optval, optlen);
 309                 break;
 310         case SO_RDS_TRANSPORT:
 311                 lock_sock(sock->sk);
 312                 ret = rds_set_transport(rs, optval, optlen);
 313                 release_sock(sock->sk);
 314                 break;
 315         default:
 316                 ret = -ENOPROTOOPT;
 317         }
 318 out:
 319         return ret;
 320 }
 321
 322 static int rds_getsockopt(struct socket *sock, int level, int optname,
 323                           char __user *optval, int __user *optlen)
 324 {
 325         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 326         int ret = -ENOPROTOOPT, len;
 327         int trans;
 328
 329         if (level != SOL_RDS)
 330                 goto out;
 331
 332         if (get_user(len, optlen)) {
 333                 ret = -EFAULT;
 334                 goto out;
 335         }
 336
 337         switch (optname) {
 338         case RDS_INFO_FIRST ... RDS_INFO_LAST:
 339                 ret = rds_info_getsockopt(sock, optname, optval,
 340                                           optlen);
 341                 break;
 342
 343         case RDS_RECVERR:
 344                 if (len < sizeof(int))
 345                         ret = -EINVAL;
 346                 else
 347                 if (put_user(rs->rs_recverr, (int __user *) optval) ||
 348                     put_user(sizeof(int), optlen))
 349                         ret = -EFAULT;
 350                 else
 351                         ret = 0;
 352                 break;
 353         case SO_RDS_TRANSPORT:
 354                 if (len < sizeof(int)) {
 355                         ret = -EINVAL;
 356                         break;
 357                 }
 358                 trans = (rs->rs_transport ? rs->rs_transport->t_type :
 359                          RDS_TRANS_NONE); /* unbound */
 360                 if (put_user(trans, (int __user *)optval) ||
 361                     put_user(sizeof(int), optlen))
 362                         ret = -EFAULT;
 363                 else
 364                         ret = 0;
 365                 break;
 366         default:
 367                 break;
 368         }
 369
 370 out:
 371         return ret;
 372
 373 }
 374
 375 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
 376                        int addr_len, int flags)
 377 {
 378         struct sock *sk = sock->sk;
 379         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 380         struct rds_sock *rs = rds_sk_to_rs(sk);
 381         int ret = 0;
 382
 383         lock_sock(sk);
 384
 385         if (addr_len != sizeof(struct sockaddr_in)) {
 386                 ret = -EINVAL;
 387                 goto out;
 388         }
 389
 390         if (sin->sin_family != AF_INET) {
 391                 ret = -EAFNOSUPPORT;
 392                 goto out;
 393         }
 394
 395         if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
 396                 ret = -EDESTADDRREQ;
 397                 goto out;
 398         }
 399
 400         rs->rs_conn_addr = sin->sin_addr.s_addr;
 401         rs->rs_conn_port = sin->sin_port;
 402
 403 out:
 404         release_sock(sk);
 405         return ret;
 406 }
 407
 408 static struct proto rds_proto = {
 409         .name     = "RDS",
 410         .owner    = THIS_MODULE,
 411         .obj_size = sizeof(struct rds_sock),
 412 };
 413
 414 static const struct proto_ops rds_proto_ops = {
 415         .family =       AF_RDS,
 416         .owner =        THIS_MODULE,
 417         .release =      rds_release,
 418         .bind =         rds_bind,
 419         .connect =      rds_connect,
 420         .socketpair =   sock_no_socketpair,
 421         .accept =       sock_no_accept,
 422         .getname =      rds_getname,
 423         .poll =         rds_poll,
 424         .ioctl =        rds_ioctl,
 425         .listen =       sock_no_listen,
 426         .shutdown =     sock_no_shutdown,
 427         .setsockopt =   rds_setsockopt,
 428         .getsockopt =   rds_getsockopt,
 429         .sendmsg =      rds_sendmsg,
 430         .recvmsg =      rds_recvmsg,
 431         .mmap =         sock_no_mmap,
 432         .sendpage =     sock_no_sendpage,
 433 };
 434
 435 static void rds_sock_destruct(struct sock *sk)
 436 {
 437         struct rds_sock *rs = rds_sk_to_rs(sk);
 438
 439         WARN_ON((&rs->rs_item != rs->rs_item.next ||
 440                  &rs->rs_item != rs->rs_item.prev));
 441 }
 442
 443 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
 444 {
 445         struct rds_sock *rs;
 446
 447         sock_init_data(sock, sk);
 448         sock->ops               = &rds_proto_ops;
 449         sk->sk_protocol         = protocol;
 450         sk->sk_destruct         = rds_sock_destruct;
 451
 452         rs = rds_sk_to_rs(sk);
 453         spin_lock_init(&rs->rs_lock);
 454         rwlock_init(&rs->rs_recv_lock);
 455         INIT_LIST_HEAD(&rs->rs_send_queue);
 456         INIT_LIST_HEAD(&rs->rs_recv_queue);
 457         INIT_LIST_HEAD(&rs->rs_notify_queue);
 458         INIT_LIST_HEAD(&rs->rs_cong_list);
 459         spin_lock_init(&rs->rs_rdma_lock);
 460         rs->rs_rdma_keys = RB_ROOT;
 461
 462         spin_lock_bh(&rds_sock_lock);
 463         list_add_tail(&rs->rs_item, &rds_sock_list);
 464         rds_sock_count++;
 465         spin_unlock_bh(&rds_sock_lock);
 466
 467         return 0;
 468 }
 469
 470 static int rds_create(struct net *net, struct socket *sock, int protocol,
 471                       int kern)
 472 {
 473         struct sock *sk;
 474
 475         if (sock->type != SOCK_SEQPACKET || protocol)
 476                 return -ESOCKTNOSUPPORT;
 477
 478         sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern);
 479         if (!sk)
 480                 return -ENOMEM;
 481
 482         return __rds_create(sock, sk, protocol);
 483 }
 484
 485 void rds_sock_addref(struct rds_sock *rs)
 486 {
 487         sock_hold(rds_rs_to_sk(rs));
 488 }
 489
 490 void rds_sock_put(struct rds_sock *rs)
 491 {
 492         sock_put(rds_rs_to_sk(rs));
 493 }
 494
 495 static const struct net_proto_family rds_family_ops = {
 496         .family =       AF_RDS,
 497         .create =       rds_create,
 498         .owner  =       THIS_MODULE,
 499 };
 500
 501 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
 502                               struct rds_info_iterator *iter,
 503                               struct rds_info_lengths *lens)
 504 {
 505         struct rds_sock *rs;
 506         struct rds_incoming *inc;
 507         unsigned int total = 0;
 508
 509         len /= sizeof(struct rds_info_message);
 510
 511         spin_lock_bh(&rds_sock_lock);
 512
 513         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 514                 read_lock(&rs->rs_recv_lock);
 515
 516                 /* XXX too lazy to maintain counts.. */
 517                 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
 518                         total++;
 519                         if (total <= len)
 520                                 rds_inc_info_copy(inc, iter, inc->i_saddr,
 521                                                   rs->rs_bound_addr, 1);
 522                 }
 523
 524                 read_unlock(&rs->rs_recv_lock);
 525         }
 526
 527         spin_unlock_bh(&rds_sock_lock);
 528
 529         lens->nr = total;
 530         lens->each = sizeof(struct rds_info_message);
 531 }
 532
 533 static void rds_sock_info(struct socket *sock, unsigned int len,
 534                           struct rds_info_iterator *iter,
 535                           struct rds_info_lengths *lens)
 536 {
 537         struct rds_info_socket sinfo;
 538         struct rds_sock *rs;
 539
 540         len /= sizeof(struct rds_info_socket);
 541
 542         spin_lock_bh(&rds_sock_lock);
 543
 544         if (len < rds_sock_count)
 545                 goto out;
 546
 547         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 548                 sinfo.sndbuf = rds_sk_sndbuf(rs);
 549                 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
 550                 sinfo.bound_addr = rs->rs_bound_addr;
 551                 sinfo.connected_addr = rs->rs_conn_addr;
 552                 sinfo.bound_port = rs->rs_bound_port;
 553                 sinfo.connected_port = rs->rs_conn_port;
 554                 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
 555
 556                 rds_info_copy(iter, &sinfo, sizeof(sinfo));
 557         }
 558
 559 out:
 560         lens->nr = rds_sock_count;
 561         lens->each = sizeof(struct rds_info_socket);
 562
 563         spin_unlock_bh(&rds_sock_lock);
 564 }
 565
 566 static void rds_exit(void)
 567 {
 568         sock_unregister(rds_family_ops.family);
 569         proto_unregister(&rds_proto);
 570         rds_conn_exit();
 571         rds_cong_exit();
 572         rds_sysctl_exit();
 573         rds_threads_exit();
 574         rds_stats_exit();
 575         rds_page_exit();
 576         rds_bind_lock_destroy();
 577         rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
 578         rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 579 }
 580 module_exit(rds_exit);
 581
 582 static int rds_init(void)
 583 {
 584         int ret;
 585
 586         ret = rds_bind_lock_init();
 587         if (ret)
 588                 goto out;
 589
 590         ret = rds_conn_init();
 591         if (ret)
 592                 goto out_bind;
 593
 594         ret = rds_threads_init();
 595         if (ret)
 596                 goto out_conn;
 597         ret = rds_sysctl_init();
 598         if (ret)
 599                 goto out_threads;
 600         ret = rds_stats_init();
 601         if (ret)
 602                 goto out_sysctl;
 603         ret = proto_register(&rds_proto, 1);
 604         if (ret)
 605                 goto out_stats;
 606         ret = sock_register(&rds_family_ops);
 607         if (ret)
 608                 goto out_proto;
 609
 610         rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
 611         rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 612
 613         goto out;
 614
 615 out_proto:
 616         proto_unregister(&rds_proto);
 617 out_stats:
 618         rds_stats_exit();
 619 out_sysctl:
 620         rds_sysctl_exit();
 621 out_threads:
 622         rds_threads_exit();
 623 out_conn:
 624         rds_conn_exit();
 625         rds_cong_exit();
 626         rds_page_exit();
 627 out_bind:
 628         rds_bind_lock_destroy();
 629 out:
 630         return ret;
 631 }
 632 module_init(rds_init);
 633
 634 #define DRV_VERSION     "4.0"
 635 #define DRV_RELDATE     "Feb 12, 2009"
 636
 637 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
 638 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
 639                    " v" DRV_VERSION " (" DRV_RELDATE ")");
 640 MODULE_VERSION(DRV_VERSION);
 641 MODULE_LICENSE("Dual BSD/GPL");
 642 MODULE_ALIAS_NETPROTO(PF_RDS);