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         poll_wait(file, &rds_poll_waitq, wait);
 163
 164         read_lock_irqsave(&rs->rs_recv_lock, flags);
 165         if (!rs->rs_cong_monitor) {
 166                 /* When a congestion map was updated, we signal POLLIN for
 167                  * "historical" reasons. Applications can also poll for
 168                  * WRBAND instead. */
 169                 if (rds_cong_updated_since(&rs->rs_cong_track))
 170                         mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
 171         } else {
 172                 spin_lock(&rs->rs_lock);
 173                 if (rs->rs_cong_notify)
 174                         mask |= (POLLIN | POLLRDNORM);
 175                 spin_unlock(&rs->rs_lock);
 176         }
 177         if (!list_empty(&rs->rs_recv_queue) ||
 178             !list_empty(&rs->rs_notify_queue))
 179                 mask |= (POLLIN | POLLRDNORM);
 180         if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
 181                 mask |= (POLLOUT | POLLWRNORM);
 182         read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 183
 184         return mask;
 185 }
 186
 187 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 188 {
 189         return -ENOIOCTLCMD;
 190 }
 191
 192 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
 193                               int len)
 194 {
 195         struct sockaddr_in sin;
 196         int ret = 0;
 197
 198         /* racing with another thread binding seems ok here */
 199         if (rs->rs_bound_addr == 0) {
 200                 ret = -ENOTCONN; /* XXX not a great errno */
 201                 goto out;
 202         }
 203
 204         if (len < sizeof(struct sockaddr_in)) {
 205                 ret = -EINVAL;
 206                 goto out;
 207         }
 208
 209         if (copy_from_user(&sin, optval, sizeof(sin))) {
 210                 ret = -EFAULT;
 211                 goto out;
 212         }
 213
 214         rds_send_drop_to(rs, &sin);
 215 out:
 216         return ret;
 217 }
 218
 219 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
 220                                int optlen)
 221 {
 222         int value;
 223
 224         if (optlen < sizeof(int))
 225                 return -EINVAL;
 226         if (get_user(value, (int __user *) optval))
 227                 return -EFAULT;
 228         *optvar = !!value;
 229         return 0;
 230 }
 231
 232 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
 233                             int optlen)
 234 {
 235         int ret;
 236
 237         ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
 238         if (ret == 0) {
 239                 if (rs->rs_cong_monitor) {
 240                         rds_cong_add_socket(rs);
 241                 } else {
 242                         rds_cong_remove_socket(rs);
 243                         rs->rs_cong_mask = 0;
 244                         rs->rs_cong_notify = 0;
 245                 }
 246         }
 247         return ret;
 248 }
 249
 250 static int rds_setsockopt(struct socket *sock, int level, int optname,
 251                           char __user *optval, unsigned int optlen)
 252 {
 253         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 254         int ret;
 255
 256         if (level != SOL_RDS) {
 257                 ret = -ENOPROTOOPT;
 258                 goto out;
 259         }
 260
 261         switch (optname) {
 262         case RDS_CANCEL_SENT_TO:
 263                 ret = rds_cancel_sent_to(rs, optval, optlen);
 264                 break;
 265         case RDS_GET_MR:
 266                 ret = rds_get_mr(rs, optval, optlen);
 267                 break;
 268         case RDS_GET_MR_FOR_DEST:
 269                 ret = rds_get_mr_for_dest(rs, optval, optlen);
 270                 break;
 271         case RDS_FREE_MR:
 272                 ret = rds_free_mr(rs, optval, optlen);
 273                 break;
 274         case RDS_RECVERR:
 275                 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
 276                 break;
 277         case RDS_CONG_MONITOR:
 278                 ret = rds_cong_monitor(rs, optval, optlen);
 279                 break;
 280         default:
 281                 ret = -ENOPROTOOPT;
 282         }
 283 out:
 284         return ret;
 285 }
 286
 287 static int rds_getsockopt(struct socket *sock, int level, int optname,
 288                           char __user *optval, int __user *optlen)
 289 {
 290         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 291         int ret = -ENOPROTOOPT, len;
 292
 293         if (level != SOL_RDS)
 294                 goto out;
 295
 296         if (get_user(len, optlen)) {
 297                 ret = -EFAULT;
 298                 goto out;
 299         }
 300
 301         switch (optname) {
 302         case RDS_INFO_FIRST ... RDS_INFO_LAST:
 303                 ret = rds_info_getsockopt(sock, optname, optval,
 304                                           optlen);
 305                 break;
 306
 307         case RDS_RECVERR:
 308                 if (len < sizeof(int))
 309                         ret = -EINVAL;
 310                 else
 311                 if (put_user(rs->rs_recverr, (int __user *) optval) ||
 312                     put_user(sizeof(int), optlen))
 313                         ret = -EFAULT;
 314                 else
 315                         ret = 0;
 316                 break;
 317         default:
 318                 break;
 319         }
 320
 321 out:
 322         return ret;
 323
 324 }
 325
 326 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
 327                        int addr_len, int flags)
 328 {
 329         struct sock *sk = sock->sk;
 330         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 331         struct rds_sock *rs = rds_sk_to_rs(sk);
 332         int ret = 0;
 333
 334         lock_sock(sk);
 335
 336         if (addr_len != sizeof(struct sockaddr_in)) {
 337                 ret = -EINVAL;
 338                 goto out;
 339         }
 340
 341         if (sin->sin_family != AF_INET) {
 342                 ret = -EAFNOSUPPORT;
 343                 goto out;
 344         }
 345
 346         if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
 347                 ret = -EDESTADDRREQ;
 348                 goto out;
 349         }
 350
 351         rs->rs_conn_addr = sin->sin_addr.s_addr;
 352         rs->rs_conn_port = sin->sin_port;
 353
 354 out:
 355         release_sock(sk);
 356         return ret;
 357 }
 358
 359 static struct proto rds_proto = {
 360         .name     = "RDS",
 361         .owner    = THIS_MODULE,
 362         .obj_size = sizeof(struct rds_sock),
 363 };
 364
 365 static const struct proto_ops rds_proto_ops = {
 366         .family =       AF_RDS,
 367         .owner =        THIS_MODULE,
 368         .release =      rds_release,
 369         .bind =         rds_bind,
 370         .connect =      rds_connect,
 371         .socketpair =   sock_no_socketpair,
 372         .accept =       sock_no_accept,
 373         .getname =      rds_getname,
 374         .poll =         rds_poll,
 375         .ioctl =        rds_ioctl,
 376         .listen =       sock_no_listen,
 377         .shutdown =     sock_no_shutdown,
 378         .setsockopt =   rds_setsockopt,
 379         .getsockopt =   rds_getsockopt,
 380         .sendmsg =      rds_sendmsg,
 381         .recvmsg =      rds_recvmsg,
 382         .mmap =         sock_no_mmap,
 383         .sendpage =     sock_no_sendpage,
 384 };
 385
 386 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
 387 {
 388         unsigned long flags;
 389         struct rds_sock *rs;
 390
 391         sock_init_data(sock, sk);
 392         sock->ops               = &rds_proto_ops;
 393         sk->sk_protocol         = protocol;
 394
 395         rs = rds_sk_to_rs(sk);
 396         spin_lock_init(&rs->rs_lock);
 397         rwlock_init(&rs->rs_recv_lock);
 398         INIT_LIST_HEAD(&rs->rs_send_queue);
 399         INIT_LIST_HEAD(&rs->rs_recv_queue);
 400         INIT_LIST_HEAD(&rs->rs_notify_queue);
 401         INIT_LIST_HEAD(&rs->rs_cong_list);
 402         spin_lock_init(&rs->rs_rdma_lock);
 403         rs->rs_rdma_keys = RB_ROOT;
 404
 405         spin_lock_irqsave(&rds_sock_lock, flags);
 406         list_add_tail(&rs->rs_item, &rds_sock_list);
 407         rds_sock_count++;
 408         spin_unlock_irqrestore(&rds_sock_lock, flags);
 409
 410         return 0;
 411 }
 412
 413 static int rds_create(struct net *net, struct socket *sock, int protocol,
 414                       int kern)
 415 {
 416         struct sock *sk;
 417
 418         if (sock->type != SOCK_SEQPACKET || protocol)
 419                 return -ESOCKTNOSUPPORT;
 420
 421         sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto);
 422         if (!sk)
 423                 return -ENOMEM;
 424
 425         return __rds_create(sock, sk, protocol);
 426 }
 427
 428 void rds_sock_addref(struct rds_sock *rs)
 429 {
 430         sock_hold(rds_rs_to_sk(rs));
 431 }
 432
 433 void rds_sock_put(struct rds_sock *rs)
 434 {
 435         sock_put(rds_rs_to_sk(rs));
 436 }
 437
 438 static const struct net_proto_family rds_family_ops = {
 439         .family =       AF_RDS,
 440         .create =       rds_create,
 441         .owner  =       THIS_MODULE,
 442 };
 443
 444 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
 445                               struct rds_info_iterator *iter,
 446                               struct rds_info_lengths *lens)
 447 {
 448         struct rds_sock *rs;
 449         struct sock *sk;
 450         struct rds_incoming *inc;
 451         unsigned long flags;
 452         unsigned int total = 0;
 453
 454         len /= sizeof(struct rds_info_message);
 455
 456         spin_lock_irqsave(&rds_sock_lock, flags);
 457
 458         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 459                 sk = rds_rs_to_sk(rs);
 460                 read_lock(&rs->rs_recv_lock);
 461
 462                 /* XXX too lazy to maintain counts.. */
 463                 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
 464                         total++;
 465                         if (total <= len)
 466                                 rds_inc_info_copy(inc, iter, inc->i_saddr,
 467                                                   rs->rs_bound_addr, 1);
 468                 }
 469
 470                 read_unlock(&rs->rs_recv_lock);
 471         }
 472
 473         spin_unlock_irqrestore(&rds_sock_lock, flags);
 474
 475         lens->nr = total;
 476         lens->each = sizeof(struct rds_info_message);
 477 }
 478
 479 static void rds_sock_info(struct socket *sock, unsigned int len,
 480                           struct rds_info_iterator *iter,
 481                           struct rds_info_lengths *lens)
 482 {
 483         struct rds_info_socket sinfo;
 484         struct rds_sock *rs;
 485         unsigned long flags;
 486
 487         len /= sizeof(struct rds_info_socket);
 488
 489         spin_lock_irqsave(&rds_sock_lock, flags);
 490
 491         if (len < rds_sock_count)
 492                 goto out;
 493
 494         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 495                 sinfo.sndbuf = rds_sk_sndbuf(rs);
 496                 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
 497                 sinfo.bound_addr = rs->rs_bound_addr;
 498                 sinfo.connected_addr = rs->rs_conn_addr;
 499                 sinfo.bound_port = rs->rs_bound_port;
 500                 sinfo.connected_port = rs->rs_conn_port;
 501                 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
 502
 503                 rds_info_copy(iter, &sinfo, sizeof(sinfo));
 504         }
 505
 506 out:
 507         lens->nr = rds_sock_count;
 508         lens->each = sizeof(struct rds_info_socket);
 509
 510         spin_unlock_irqrestore(&rds_sock_lock, flags);
 511 }
 512
 513 static void __exit rds_exit(void)
 514 {
 515         sock_unregister(rds_family_ops.family);
 516         proto_unregister(&rds_proto);
 517         rds_conn_exit();
 518         rds_cong_exit();
 519         rds_sysctl_exit();
 520         rds_threads_exit();
 521         rds_stats_exit();
 522         rds_page_exit();
 523         rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
 524         rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 525 }
 526 module_exit(rds_exit);
 527
 528 static int __init rds_init(void)
 529 {
 530         int ret;
 531
 532         ret = rds_conn_init();
 533         if (ret)
 534                 goto out;
 535         ret = rds_threads_init();
 536         if (ret)
 537                 goto out_conn;
 538         ret = rds_sysctl_init();
 539         if (ret)
 540                 goto out_threads;
 541         ret = rds_stats_init();
 542         if (ret)
 543                 goto out_sysctl;
 544         ret = proto_register(&rds_proto, 1);
 545         if (ret)
 546                 goto out_stats;
 547         ret = sock_register(&rds_family_ops);
 548         if (ret)
 549                 goto out_proto;
 550
 551         rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
 552         rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 553
 554         goto out;
 555
 556 out_proto:
 557         proto_unregister(&rds_proto);
 558 out_stats:
 559         rds_stats_exit();
 560 out_sysctl:
 561         rds_sysctl_exit();
 562 out_threads:
 563         rds_threads_exit();
 564 out_conn:
 565         rds_conn_exit();
 566         rds_cong_exit();
 567         rds_page_exit();
 568 out:
 569         return ret;
 570 }
 571 module_init(rds_init);
 572
 573 #define DRV_VERSION     "4.0"
 574 #define DRV_RELDATE     "Feb 12, 2009"
 575
 576 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
 577 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
 578                    " v" DRV_VERSION " (" DRV_RELDATE ")");
 579 MODULE_VERSION(DRV_VERSION);
 580 MODULE_LICENSE("Dual BSD/GPL");
 581 MODULE_ALIAS_NETPROTO(PF_RDS);