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_enable_recvtstamp(struct sock *sk, char __user *optval,
 281                                  int optlen)
 282 {
 283         int val, valbool;
 284
 285         if (optlen != sizeof(int))
 286                 return -EFAULT;
 287
 288         if (get_user(val, (int __user *)optval))
 289                 return -EFAULT;
 290
 291         valbool = val ? 1 : 0;
 292
 293         if (valbool)
 294                 sock_set_flag(sk, SOCK_RCVTSTAMP);
 295         else
 296                 sock_reset_flag(sk, SOCK_RCVTSTAMP);
 297
 298         return 0;
 299 }
 300
 301 static int rds_setsockopt(struct socket *sock, int level, int optname,
 302                           char __user *optval, unsigned int optlen)
 303 {
 304         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 305         int ret;
 306
 307         if (level != SOL_RDS) {
 308                 ret = -ENOPROTOOPT;
 309                 goto out;
 310         }
 311
 312         switch (optname) {
 313         case RDS_CANCEL_SENT_TO:
 314                 ret = rds_cancel_sent_to(rs, optval, optlen);
 315                 break;
 316         case RDS_GET_MR:
 317                 ret = rds_get_mr(rs, optval, optlen);
 318                 break;
 319         case RDS_GET_MR_FOR_DEST:
 320                 ret = rds_get_mr_for_dest(rs, optval, optlen);
 321                 break;
 322         case RDS_FREE_MR:
 323                 ret = rds_free_mr(rs, optval, optlen);
 324                 break;
 325         case RDS_RECVERR:
 326                 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
 327                 break;
 328         case RDS_CONG_MONITOR:
 329                 ret = rds_cong_monitor(rs, optval, optlen);
 330                 break;
 331         case SO_RDS_TRANSPORT:
 332                 lock_sock(sock->sk);
 333                 ret = rds_set_transport(rs, optval, optlen);
 334                 release_sock(sock->sk);
 335                 break;
 336         case SO_TIMESTAMP:
 337                 lock_sock(sock->sk);
 338                 ret = rds_enable_recvtstamp(sock->sk, optval, optlen);
 339                 release_sock(sock->sk);
 340                 break;
 341         default:
 342                 ret = -ENOPROTOOPT;
 343         }
 344 out:
 345         return ret;
 346 }
 347
 348 static int rds_getsockopt(struct socket *sock, int level, int optname,
 349                           char __user *optval, int __user *optlen)
 350 {
 351         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 352         int ret = -ENOPROTOOPT, len;
 353         int trans;
 354
 355         if (level != SOL_RDS)
 356                 goto out;
 357
 358         if (get_user(len, optlen)) {
 359                 ret = -EFAULT;
 360                 goto out;
 361         }
 362
 363         switch (optname) {
 364         case RDS_INFO_FIRST ... RDS_INFO_LAST:
 365                 ret = rds_info_getsockopt(sock, optname, optval,
 366                                           optlen);
 367                 break;
 368
 369         case RDS_RECVERR:
 370                 if (len < sizeof(int))
 371                         ret = -EINVAL;
 372                 else
 373                 if (put_user(rs->rs_recverr, (int __user *) optval) ||
 374                     put_user(sizeof(int), optlen))
 375                         ret = -EFAULT;
 376                 else
 377                         ret = 0;
 378                 break;
 379         case SO_RDS_TRANSPORT:
 380                 if (len < sizeof(int)) {
 381                         ret = -EINVAL;
 382                         break;
 383                 }
 384                 trans = (rs->rs_transport ? rs->rs_transport->t_type :
 385                          RDS_TRANS_NONE); /* unbound */
 386                 if (put_user(trans, (int __user *)optval) ||
 387                     put_user(sizeof(int), optlen))
 388                         ret = -EFAULT;
 389                 else
 390                         ret = 0;
 391                 break;
 392         default:
 393                 break;
 394         }
 395
 396 out:
 397         return ret;
 398
 399 }
 400
 401 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
 402                        int addr_len, int flags)
 403 {
 404         struct sock *sk = sock->sk;
 405         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 406         struct rds_sock *rs = rds_sk_to_rs(sk);
 407         int ret = 0;
 408
 409         lock_sock(sk);
 410
 411         if (addr_len != sizeof(struct sockaddr_in)) {
 412                 ret = -EINVAL;
 413                 goto out;
 414         }
 415
 416         if (sin->sin_family != AF_INET) {
 417                 ret = -EAFNOSUPPORT;
 418                 goto out;
 419         }
 420
 421         if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
 422                 ret = -EDESTADDRREQ;
 423                 goto out;
 424         }
 425
 426         rs->rs_conn_addr = sin->sin_addr.s_addr;
 427         rs->rs_conn_port = sin->sin_port;
 428
 429 out:
 430         release_sock(sk);
 431         return ret;
 432 }
 433
 434 static struct proto rds_proto = {
 435         .name     = "RDS",
 436         .owner    = THIS_MODULE,
 437         .obj_size = sizeof(struct rds_sock),
 438 };
 439
 440 static const struct proto_ops rds_proto_ops = {
 441         .family =       AF_RDS,
 442         .owner =        THIS_MODULE,
 443         .release =      rds_release,
 444         .bind =         rds_bind,
 445         .connect =      rds_connect,
 446         .socketpair =   sock_no_socketpair,
 447         .accept =       sock_no_accept,
 448         .getname =      rds_getname,
 449         .poll =         rds_poll,
 450         .ioctl =        rds_ioctl,
 451         .listen =       sock_no_listen,
 452         .shutdown =     sock_no_shutdown,
 453         .setsockopt =   rds_setsockopt,
 454         .getsockopt =   rds_getsockopt,
 455         .sendmsg =      rds_sendmsg,
 456         .recvmsg =      rds_recvmsg,
 457         .mmap =         sock_no_mmap,
 458         .sendpage =     sock_no_sendpage,
 459 };
 460
 461 static void rds_sock_destruct(struct sock *sk)
 462 {
 463         struct rds_sock *rs = rds_sk_to_rs(sk);
 464
 465         WARN_ON((&rs->rs_item != rs->rs_item.next ||
 466                  &rs->rs_item != rs->rs_item.prev));
 467 }
 468
 469 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
 470 {
 471         struct rds_sock *rs;
 472
 473         sock_init_data(sock, sk);
 474         sock->ops               = &rds_proto_ops;
 475         sk->sk_protocol         = protocol;
 476         sk->sk_destruct         = rds_sock_destruct;
 477
 478         rs = rds_sk_to_rs(sk);
 479         spin_lock_init(&rs->rs_lock);
 480         rwlock_init(&rs->rs_recv_lock);
 481         INIT_LIST_HEAD(&rs->rs_send_queue);
 482         INIT_LIST_HEAD(&rs->rs_recv_queue);
 483         INIT_LIST_HEAD(&rs->rs_notify_queue);
 484         INIT_LIST_HEAD(&rs->rs_cong_list);
 485         spin_lock_init(&rs->rs_rdma_lock);
 486         rs->rs_rdma_keys = RB_ROOT;
 487
 488         spin_lock_bh(&rds_sock_lock);
 489         list_add_tail(&rs->rs_item, &rds_sock_list);
 490         rds_sock_count++;
 491         spin_unlock_bh(&rds_sock_lock);
 492
 493         return 0;
 494 }
 495
 496 static int rds_create(struct net *net, struct socket *sock, int protocol,
 497                       int kern)
 498 {
 499         struct sock *sk;
 500
 501         if (sock->type != SOCK_SEQPACKET || protocol)
 502                 return -ESOCKTNOSUPPORT;
 503
 504         sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern);
 505         if (!sk)
 506                 return -ENOMEM;
 507
 508         return __rds_create(sock, sk, protocol);
 509 }
 510
 511 void rds_sock_addref(struct rds_sock *rs)
 512 {
 513         sock_hold(rds_rs_to_sk(rs));
 514 }
 515
 516 void rds_sock_put(struct rds_sock *rs)
 517 {
 518         sock_put(rds_rs_to_sk(rs));
 519 }
 520
 521 static const struct net_proto_family rds_family_ops = {
 522         .family =       AF_RDS,
 523         .create =       rds_create,
 524         .owner  =       THIS_MODULE,
 525 };
 526
 527 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
 528                               struct rds_info_iterator *iter,
 529                               struct rds_info_lengths *lens)
 530 {
 531         struct rds_sock *rs;
 532         struct rds_incoming *inc;
 533         unsigned int total = 0;
 534
 535         len /= sizeof(struct rds_info_message);
 536
 537         spin_lock_bh(&rds_sock_lock);
 538
 539         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 540                 read_lock(&rs->rs_recv_lock);
 541
 542                 /* XXX too lazy to maintain counts.. */
 543                 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
 544                         total++;
 545                         if (total <= len)
 546                                 rds_inc_info_copy(inc, iter, inc->i_saddr,
 547                                                   rs->rs_bound_addr, 1);
 548                 }
 549
 550                 read_unlock(&rs->rs_recv_lock);
 551         }
 552
 553         spin_unlock_bh(&rds_sock_lock);
 554
 555         lens->nr = total;
 556         lens->each = sizeof(struct rds_info_message);
 557 }
 558
 559 static void rds_sock_info(struct socket *sock, unsigned int len,
 560                           struct rds_info_iterator *iter,
 561                           struct rds_info_lengths *lens)
 562 {
 563         struct rds_info_socket sinfo;
 564         struct rds_sock *rs;
 565
 566         len /= sizeof(struct rds_info_socket);
 567
 568         spin_lock_bh(&rds_sock_lock);
 569
 570         if (len < rds_sock_count)
 571                 goto out;
 572
 573         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 574                 sinfo.sndbuf = rds_sk_sndbuf(rs);
 575                 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
 576                 sinfo.bound_addr = rs->rs_bound_addr;
 577                 sinfo.connected_addr = rs->rs_conn_addr;
 578                 sinfo.bound_port = rs->rs_bound_port;
 579                 sinfo.connected_port = rs->rs_conn_port;
 580                 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
 581
 582                 rds_info_copy(iter, &sinfo, sizeof(sinfo));
 583         }
 584
 585 out:
 586         lens->nr = rds_sock_count;
 587         lens->each = sizeof(struct rds_info_socket);
 588
 589         spin_unlock_bh(&rds_sock_lock);
 590 }
 591
 592 static void rds_exit(void)
 593 {
 594         sock_unregister(rds_family_ops.family);
 595         proto_unregister(&rds_proto);
 596         rds_conn_exit();
 597         rds_cong_exit();
 598         rds_sysctl_exit();
 599         rds_threads_exit();
 600         rds_stats_exit();
 601         rds_page_exit();
 602         rds_bind_lock_destroy();
 603         rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
 604         rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 605 }
 606 module_exit(rds_exit);
 607
 608 u32 rds_gen_num;
 609
 610 static int rds_init(void)
 611 {
 612         int ret;
 613
 614         net_get_random_once(&rds_gen_num, sizeof(rds_gen_num));
 615
 616         ret = rds_bind_lock_init();
 617         if (ret)
 618                 goto out;
 619
 620         ret = rds_conn_init();
 621         if (ret)
 622                 goto out_bind;
 623
 624         ret = rds_threads_init();
 625         if (ret)
 626                 goto out_conn;
 627         ret = rds_sysctl_init();
 628         if (ret)
 629                 goto out_threads;
 630         ret = rds_stats_init();
 631         if (ret)
 632                 goto out_sysctl;
 633         ret = proto_register(&rds_proto, 1);
 634         if (ret)
 635                 goto out_stats;
 636         ret = sock_register(&rds_family_ops);
 637         if (ret)
 638                 goto out_proto;
 639
 640         rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
 641         rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 642
 643         goto out;
 644
 645 out_proto:
 646         proto_unregister(&rds_proto);
 647 out_stats:
 648         rds_stats_exit();
 649 out_sysctl:
 650         rds_sysctl_exit();
 651 out_threads:
 652         rds_threads_exit();
 653 out_conn:
 654         rds_conn_exit();
 655         rds_cong_exit();
 656         rds_page_exit();
 657 out_bind:
 658         rds_bind_lock_destroy();
 659 out:
 660         return ret;
 661 }
 662 module_init(rds_init);
 663
 664 #define DRV_VERSION     "4.0"
 665 #define DRV_RELDATE     "Feb 12, 2009"
 666
 667 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
 668 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
 669                    " v" DRV_VERSION " (" DRV_RELDATE ")");
 670 MODULE_VERSION(DRV_VERSION);
 671 MODULE_LICENSE("Dual BSD/GPL");
 672 MODULE_ALIAS_NETPROTO(PF_RDS);