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/in.h>
  37 #include <linux/poll.h>
  38 #include <net/sock.h>
  39
  40 #include "rds.h"
  41 #include "rdma.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         unsigned long flags;
  63
  64         if (sk == NULL)
  65                 goto out;
  66
  67         rs = rds_sk_to_rs(sk);
  68
  69         sock_orphan(sk);
  70         /* Note - rds_clear_recv_queue grabs rs_recv_lock, so
  71          * that ensures the recv path has completed messing
  72          * with the socket. */
  73         rds_clear_recv_queue(rs);
  74         rds_cong_remove_socket(rs);
  75         rds_remove_bound(rs);
  76         rds_send_drop_to(rs, NULL);
  77         rds_rdma_drop_keys(rs);
  78         rds_notify_queue_get(rs, NULL);
  79
  80         spin_lock_irqsave(&rds_sock_lock, flags);
  81         list_del_init(&rs->rs_item);
  82         rds_sock_count--;
  83         spin_unlock_irqrestore(&rds_sock_lock, flags);
  84
  85         sock->sk = NULL;
  86         sock_put(sk);
  87 out:
  88         return 0;
  89 }
  90
  91 /*
  92  * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
  93  * _bh() isn't OK here, we're called from interrupt handlers.  It's probably OK
  94  * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
  95  * this seems more conservative.
  96  * NB - normally, one would use sk_callback_lock for this, but we can
  97  * get here from interrupts, whereas the network code grabs sk_callback_lock
  98  * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
  99  */
 100 void rds_wake_sk_sleep(struct rds_sock *rs)
 101 {
 102         unsigned long flags;
 103
 104         read_lock_irqsave(&rs->rs_recv_lock, flags);
 105         __rds_wake_sk_sleep(rds_rs_to_sk(rs));
 106         read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 107 }
 108
 109 static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
 110                        int *uaddr_len, int peer)
 111 {
 112         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 113         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 114
 115         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 116
 117         /* racey, don't care */
 118         if (peer) {
 119                 if (!rs->rs_conn_addr)
 120                         return -ENOTCONN;
 121
 122                 sin->sin_port = rs->rs_conn_port;
 123                 sin->sin_addr.s_addr = rs->rs_conn_addr;
 124         } else {
 125                 sin->sin_port = rs->rs_bound_port;
 126                 sin->sin_addr.s_addr = rs->rs_bound_addr;
 127         }
 128
 129         sin->sin_family = AF_INET;
 130
 131         *uaddr_len = sizeof(*sin);
 132         return 0;
 133 }
 134
 135 /*
 136  * RDS' poll is without a doubt the least intuitive part of the interface,
 137  * as POLLIN and POLLOUT do not behave entirely as you would expect from
 138  * a network protocol.
 139  *
 140  * POLLIN is asserted if
 141  *  -   there is data on the receive queue.
 142  *  -   to signal that a previously congested destination may have become
 143  *      uncongested
 144  *  -   A notification has been queued to the socket (this can be a congestion
 145  *      update, or a RDMA completion).
 146  *
 147  * POLLOUT is asserted if there is room on the send queue. This does not mean
 148  * however, that the next sendmsg() call will succeed. If the application tries
 149  * to send to a congested destination, the system call may still fail (and
 150  * return ENOBUFS).
 151  */
 152 static unsigned int rds_poll(struct file *file, struct socket *sock,
 153                              poll_table *wait)
 154 {
 155         struct sock *sk = sock->sk;
 156         struct rds_sock *rs = rds_sk_to_rs(sk);
 157         unsigned int mask = 0;
 158         unsigned long flags;
 159
 160         poll_wait(file, sk->sk_sleep, wait);
 161
 162         if (rs->rs_seen_congestion)
 163                 poll_wait(file, &rds_poll_waitq, wait);
 164
 165         read_lock_irqsave(&rs->rs_recv_lock, flags);
 166         if (!rs->rs_cong_monitor) {
 167                 /* When a congestion map was updated, we signal POLLIN for
 168                  * "historical" reasons. Applications can also poll for
 169                  * WRBAND instead. */
 170                 if (rds_cong_updated_since(&rs->rs_cong_track))
 171                         mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
 172         } else {
 173                 spin_lock(&rs->rs_lock);
 174                 if (rs->rs_cong_notify)
 175                         mask |= (POLLIN | POLLRDNORM);
 176                 spin_unlock(&rs->rs_lock);
 177         }
 178         if (!list_empty(&rs->rs_recv_queue) ||
 179             !list_empty(&rs->rs_notify_queue))
 180                 mask |= (POLLIN | POLLRDNORM);
 181         if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
 182                 mask |= (POLLOUT | POLLWRNORM);
 183         read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 184
 185         /* clear state any time we wake a seen-congested socket */
 186         if (mask)
 187                 rs->rs_seen_congestion = 0;
 188
 189         return mask;
 190 }
 191
 192 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 193 {
 194         return -ENOIOCTLCMD;
 195 }
 196
 197 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
 198                               int len)
 199 {
 200         struct sockaddr_in sin;
 201         int ret = 0;
 202
 203         /* racing with another thread binding seems ok here */
 204         if (rs->rs_bound_addr == 0) {
 205                 ret = -ENOTCONN; /* XXX not a great errno */
 206                 goto out;
 207         }
 208
 209         if (len < sizeof(struct sockaddr_in)) {
 210                 ret = -EINVAL;
 211                 goto out;
 212         }
 213
 214         if (copy_from_user(&sin, optval, sizeof(sin))) {
 215                 ret = -EFAULT;
 216                 goto out;
 217         }
 218
 219         rds_send_drop_to(rs, &sin);
 220 out:
 221         return ret;
 222 }
 223
 224 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
 225                                int optlen)
 226 {
 227         int value;
 228
 229         if (optlen < sizeof(int))
 230                 return -EINVAL;
 231         if (get_user(value, (int __user *) optval))
 232                 return -EFAULT;
 233         *optvar = !!value;
 234         return 0;
 235 }
 236
 237 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
 238                             int optlen)
 239 {
 240         int ret;
 241
 242         ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
 243         if (ret == 0) {
 244                 if (rs->rs_cong_monitor) {
 245                         rds_cong_add_socket(rs);
 246                 } else {
 247                         rds_cong_remove_socket(rs);
 248                         rs->rs_cong_mask = 0;
 249                         rs->rs_cong_notify = 0;
 250                 }
 251         }
 252         return ret;
 253 }
 254
 255 static int rds_setsockopt(struct socket *sock, int level, int optname,
 256                           char __user *optval, unsigned int optlen)
 257 {
 258         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 259         int ret;
 260
 261         if (level != SOL_RDS) {
 262                 ret = -ENOPROTOOPT;
 263                 goto out;
 264         }
 265
 266         switch (optname) {
 267         case RDS_CANCEL_SENT_TO:
 268                 ret = rds_cancel_sent_to(rs, optval, optlen);
 269                 break;
 270         case RDS_GET_MR:
 271                 ret = rds_get_mr(rs, optval, optlen);
 272                 break;
 273         case RDS_GET_MR_FOR_DEST:
 274                 ret = rds_get_mr_for_dest(rs, optval, optlen);
 275                 break;
 276         case RDS_FREE_MR:
 277                 ret = rds_free_mr(rs, optval, optlen);
 278                 break;
 279         case RDS_RECVERR:
 280                 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
 281                 break;
 282         case RDS_CONG_MONITOR:
 283                 ret = rds_cong_monitor(rs, optval, optlen);
 284                 break;
 285         default:
 286                 ret = -ENOPROTOOPT;
 287         }
 288 out:
 289         return ret;
 290 }
 291
 292 static int rds_getsockopt(struct socket *sock, int level, int optname,
 293                           char __user *optval, int __user *optlen)
 294 {
 295         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 296         int ret = -ENOPROTOOPT, len;
 297
 298         if (level != SOL_RDS)
 299                 goto out;
 300
 301         if (get_user(len, optlen)) {
 302                 ret = -EFAULT;
 303                 goto out;
 304         }
 305
 306         switch (optname) {
 307         case RDS_INFO_FIRST ... RDS_INFO_LAST:
 308                 ret = rds_info_getsockopt(sock, optname, optval,
 309                                           optlen);
 310                 break;
 311
 312         case RDS_RECVERR:
 313                 if (len < sizeof(int))
 314                         ret = -EINVAL;
 315                 else
 316                 if (put_user(rs->rs_recverr, (int __user *) optval) ||
 317                     put_user(sizeof(int), optlen))
 318                         ret = -EFAULT;
 319                 else
 320                         ret = 0;
 321                 break;
 322         default:
 323                 break;
 324         }
 325
 326 out:
 327         return ret;
 328
 329 }
 330
 331 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
 332                        int addr_len, int flags)
 333 {
 334         struct sock *sk = sock->sk;
 335         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 336         struct rds_sock *rs = rds_sk_to_rs(sk);
 337         int ret = 0;
 338
 339         lock_sock(sk);
 340
 341         if (addr_len != sizeof(struct sockaddr_in)) {
 342                 ret = -EINVAL;
 343                 goto out;
 344         }
 345
 346         if (sin->sin_family != AF_INET) {
 347                 ret = -EAFNOSUPPORT;
 348                 goto out;
 349         }
 350
 351         if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
 352                 ret = -EDESTADDRREQ;
 353                 goto out;
 354         }
 355
 356         rs->rs_conn_addr = sin->sin_addr.s_addr;
 357         rs->rs_conn_port = sin->sin_port;
 358
 359 out:
 360         release_sock(sk);
 361         return ret;
 362 }
 363
 364 static struct proto rds_proto = {
 365         .name     = "RDS",
 366         .owner    = THIS_MODULE,
 367         .obj_size = sizeof(struct rds_sock),
 368 };
 369
 370 static const struct proto_ops rds_proto_ops = {
 371         .family =       AF_RDS,
 372         .owner =        THIS_MODULE,
 373         .release =      rds_release,
 374         .bind =         rds_bind,
 375         .connect =      rds_connect,
 376         .socketpair =   sock_no_socketpair,
 377         .accept =       sock_no_accept,
 378         .getname =      rds_getname,
 379         .poll =         rds_poll,
 380         .ioctl =        rds_ioctl,
 381         .listen =       sock_no_listen,
 382         .shutdown =     sock_no_shutdown,
 383         .setsockopt =   rds_setsockopt,
 384         .getsockopt =   rds_getsockopt,
 385         .sendmsg =      rds_sendmsg,
 386         .recvmsg =      rds_recvmsg,
 387         .mmap =         sock_no_mmap,
 388         .sendpage =     sock_no_sendpage,
 389 };
 390
 391 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
 392 {
 393         unsigned long flags;
 394         struct rds_sock *rs;
 395
 396         sock_init_data(sock, sk);
 397         sock->ops               = &rds_proto_ops;
 398         sk->sk_protocol         = protocol;
 399
 400         rs = rds_sk_to_rs(sk);
 401         spin_lock_init(&rs->rs_lock);
 402         rwlock_init(&rs->rs_recv_lock);
 403         INIT_LIST_HEAD(&rs->rs_send_queue);
 404         INIT_LIST_HEAD(&rs->rs_recv_queue);
 405         INIT_LIST_HEAD(&rs->rs_notify_queue);
 406         INIT_LIST_HEAD(&rs->rs_cong_list);
 407         spin_lock_init(&rs->rs_rdma_lock);
 408         rs->rs_rdma_keys = RB_ROOT;
 409
 410         spin_lock_irqsave(&rds_sock_lock, flags);
 411         list_add_tail(&rs->rs_item, &rds_sock_list);
 412         rds_sock_count++;
 413         spin_unlock_irqrestore(&rds_sock_lock, flags);
 414
 415         return 0;
 416 }
 417
 418 static int rds_create(struct net *net, struct socket *sock, int protocol,
 419                       int kern)
 420 {
 421         struct sock *sk;
 422
 423         if (sock->type != SOCK_SEQPACKET || protocol)
 424                 return -ESOCKTNOSUPPORT;
 425
 426         sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto);
 427         if (!sk)
 428                 return -ENOMEM;
 429
 430         return __rds_create(sock, sk, protocol);
 431 }
 432
 433 void rds_sock_addref(struct rds_sock *rs)
 434 {
 435         sock_hold(rds_rs_to_sk(rs));
 436 }
 437
 438 void rds_sock_put(struct rds_sock *rs)
 439 {
 440         sock_put(rds_rs_to_sk(rs));
 441 }
 442
 443 static const struct net_proto_family rds_family_ops = {
 444         .family =       AF_RDS,
 445         .create =       rds_create,
 446         .owner  =       THIS_MODULE,
 447 };
 448
 449 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
 450                               struct rds_info_iterator *iter,
 451                               struct rds_info_lengths *lens)
 452 {
 453         struct rds_sock *rs;
 454         struct sock *sk;
 455         struct rds_incoming *inc;
 456         unsigned long flags;
 457         unsigned int total = 0;
 458
 459         len /= sizeof(struct rds_info_message);
 460
 461         spin_lock_irqsave(&rds_sock_lock, flags);
 462
 463         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 464                 sk = rds_rs_to_sk(rs);
 465                 read_lock(&rs->rs_recv_lock);
 466
 467                 /* XXX too lazy to maintain counts.. */
 468                 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
 469                         total++;
 470                         if (total <= len)
 471                                 rds_inc_info_copy(inc, iter, inc->i_saddr,
 472                                                   rs->rs_bound_addr, 1);
 473                 }
 474
 475                 read_unlock(&rs->rs_recv_lock);
 476         }
 477
 478         spin_unlock_irqrestore(&rds_sock_lock, flags);
 479
 480         lens->nr = total;
 481         lens->each = sizeof(struct rds_info_message);
 482 }
 483
 484 static void rds_sock_info(struct socket *sock, unsigned int len,
 485                           struct rds_info_iterator *iter,
 486                           struct rds_info_lengths *lens)
 487 {
 488         struct rds_info_socket sinfo;
 489         struct rds_sock *rs;
 490         unsigned long flags;
 491
 492         len /= sizeof(struct rds_info_socket);
 493
 494         spin_lock_irqsave(&rds_sock_lock, flags);
 495
 496         if (len < rds_sock_count)
 497                 goto out;
 498
 499         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 500                 sinfo.sndbuf = rds_sk_sndbuf(rs);
 501                 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
 502                 sinfo.bound_addr = rs->rs_bound_addr;
 503                 sinfo.connected_addr = rs->rs_conn_addr;
 504                 sinfo.bound_port = rs->rs_bound_port;
 505                 sinfo.connected_port = rs->rs_conn_port;
 506                 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
 507
 508                 rds_info_copy(iter, &sinfo, sizeof(sinfo));
 509         }
 510
 511 out:
 512         lens->nr = rds_sock_count;
 513         lens->each = sizeof(struct rds_info_socket);
 514
 515         spin_unlock_irqrestore(&rds_sock_lock, flags);
 516 }
 517
 518 static void __exit rds_exit(void)
 519 {
 520         sock_unregister(rds_family_ops.family);
 521         proto_unregister(&rds_proto);
 522         rds_conn_exit();
 523         rds_cong_exit();
 524         rds_sysctl_exit();
 525         rds_threads_exit();
 526         rds_stats_exit();
 527         rds_page_exit();
 528         rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
 529         rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 530 }
 531 module_exit(rds_exit);
 532
 533 static int __init rds_init(void)
 534 {
 535         int ret;
 536
 537         ret = rds_conn_init();
 538         if (ret)
 539                 goto out;
 540         ret = rds_threads_init();
 541         if (ret)
 542                 goto out_conn;
 543         ret = rds_sysctl_init();
 544         if (ret)
 545                 goto out_threads;
 546         ret = rds_stats_init();
 547         if (ret)
 548                 goto out_sysctl;
 549         ret = proto_register(&rds_proto, 1);
 550         if (ret)
 551                 goto out_stats;
 552         ret = sock_register(&rds_family_ops);
 553         if (ret)
 554                 goto out_proto;
 555
 556         rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
 557         rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 558
 559         goto out;
 560
 561 out_proto:
 562         proto_unregister(&rds_proto);
 563 out_stats:
 564         rds_stats_exit();
 565 out_sysctl:
 566         rds_sysctl_exit();
 567 out_threads:
 568         rds_threads_exit();
 569 out_conn:
 570         rds_conn_exit();
 571         rds_cong_exit();
 572         rds_page_exit();
 573 out:
 574         return ret;
 575 }
 576 module_init(rds_init);
 577
 578 #define DRV_VERSION     "4.0"
 579 #define DRV_RELDATE     "Feb 12, 2009"
 580
 581 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
 582 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
 583                    " v" DRV_VERSION " (" DRV_RELDATE ")");
 584 MODULE_VERSION(DRV_VERSION);
 585 MODULE_LICENSE("Dual BSD/GPL");
 586 MODULE_ALIAS_NETPROTO(PF_RDS);