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 <linux/version.h>
  39 #include <net/sock.h>
  40
  41 #include "rds.h"
  42 #include "rdma.h"
  43 #include "rdma_transport.h"
  44
  45 /* this is just used for stats gathering :/ */
  46 static DEFINE_SPINLOCK(rds_sock_lock);
  47 static unsigned long rds_sock_count;
  48 static LIST_HEAD(rds_sock_list);
  49 DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);
  50
  51 /*
  52  * This is called as the final descriptor referencing this socket is closed.
  53  * We have to unbind the socket so that another socket can be bound to the
  54  * address it was using.
  55  *
  56  * We have to be careful about racing with the incoming path.  sock_orphan()
  57  * sets SOCK_DEAD and we use that as an indicator to the rx path that new
  58  * messages shouldn't be queued.
  59  */
  60 static int rds_release(struct socket *sock)
  61 {
  62         struct sock *sk = sock->sk;
  63         struct rds_sock *rs;
  64         unsigned long flags;
  65
  66         if (sk == NULL)
  67                 goto out;
  68
  69         rs = rds_sk_to_rs(sk);
  70
  71         sock_orphan(sk);
  72         /* Note - rds_clear_recv_queue grabs rs_recv_lock, so
  73          * that ensures the recv path has completed messing
  74          * with the socket. */
  75         rds_clear_recv_queue(rs);
  76         rds_cong_remove_socket(rs);
  77         rds_remove_bound(rs);
  78         rds_send_drop_to(rs, NULL);
  79         rds_rdma_drop_keys(rs);
  80         rds_notify_queue_get(rs, NULL);
  81
  82         spin_lock_irqsave(&rds_sock_lock, flags);
  83         list_del_init(&rs->rs_item);
  84         rds_sock_count--;
  85         spin_unlock_irqrestore(&rds_sock_lock, flags);
  86
  87         sock->sk = NULL;
  88         sock_put(sk);
  89 out:
  90         return 0;
  91 }
  92
  93 /*
  94  * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
  95  * _bh() isn't OK here, we're called from interrupt handlers.  It's probably OK
  96  * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
  97  * this seems more conservative.
  98  * NB - normally, one would use sk_callback_lock for this, but we can
  99  * get here from interrupts, whereas the network code grabs sk_callback_lock
 100  * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
 101  */
 102 void rds_wake_sk_sleep(struct rds_sock *rs)
 103 {
 104         unsigned long flags;
 105
 106         read_lock_irqsave(&rs->rs_recv_lock, flags);
 107         __rds_wake_sk_sleep(rds_rs_to_sk(rs));
 108         read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 109 }
 110
 111 static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
 112                        int *uaddr_len, int peer)
 113 {
 114         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 115         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 116
 117         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 118
 119         /* racey, don't care */
 120         if (peer) {
 121                 if (!rs->rs_conn_addr)
 122                         return -ENOTCONN;
 123
 124                 sin->sin_port = rs->rs_conn_port;
 125                 sin->sin_addr.s_addr = rs->rs_conn_addr;
 126         } else {
 127                 sin->sin_port = rs->rs_bound_port;
 128                 sin->sin_addr.s_addr = rs->rs_bound_addr;
 129         }
 130
 131         sin->sin_family = AF_INET;
 132
 133         *uaddr_len = sizeof(*sin);
 134         return 0;
 135 }
 136
 137 /*
 138  * RDS' poll is without a doubt the least intuitive part of the interface,
 139  * as POLLIN and POLLOUT do not behave entirely as you would expect from
 140  * a network protocol.
 141  *
 142  * POLLIN is asserted if
 143  *  -   there is data on the receive queue.
 144  *  -   to signal that a previously congested destination may have become
 145  *      uncongested
 146  *  -   A notification has been queued to the socket (this can be a congestion
 147  *      update, or a RDMA completion).
 148  *
 149  * POLLOUT is asserted if there is room on the send queue. This does not mean
 150  * however, that the next sendmsg() call will succeed. If the application tries
 151  * to send to a congested destination, the system call may still fail (and
 152  * return ENOBUFS).
 153  */
 154 static unsigned int rds_poll(struct file *file, struct socket *sock,
 155                              poll_table *wait)
 156 {
 157         struct sock *sk = sock->sk;
 158         struct rds_sock *rs = rds_sk_to_rs(sk);
 159         unsigned int mask = 0;
 160         unsigned long flags;
 161
 162         poll_wait(file, sk->sk_sleep, wait);
 163
 164         poll_wait(file, &rds_poll_waitq, wait);
 165
 166         read_lock_irqsave(&rs->rs_recv_lock, flags);
 167         if (!rs->rs_cong_monitor) {
 168                 /* When a congestion map was updated, we signal POLLIN for
 169                  * "historical" reasons. Applications can also poll for
 170                  * WRBAND instead. */
 171                 if (rds_cong_updated_since(&rs->rs_cong_track))
 172                         mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
 173         } else {
 174                 spin_lock(&rs->rs_lock);
 175                 if (rs->rs_cong_notify)
 176                         mask |= (POLLIN | POLLRDNORM);
 177                 spin_unlock(&rs->rs_lock);
 178         }
 179         if (!list_empty(&rs->rs_recv_queue)
 180          || !list_empty(&rs->rs_notify_queue))
 181                 mask |= (POLLIN | POLLRDNORM);
 182         if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
 183                 mask |= (POLLOUT | POLLWRNORM);
 184         read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 185
 186         return mask;
 187 }
 188
 189 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 190 {
 191         return -ENOIOCTLCMD;
 192 }
 193
 194 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
 195                               int len)
 196 {
 197         struct sockaddr_in sin;
 198         int ret = 0;
 199
 200         /* racing with another thread binding seems ok here */
 201         if (rs->rs_bound_addr == 0) {
 202                 ret = -ENOTCONN; /* XXX not a great errno */
 203                 goto out;
 204         }
 205
 206         if (len < sizeof(struct sockaddr_in)) {
 207                 ret = -EINVAL;
 208                 goto out;
 209         }
 210
 211         if (copy_from_user(&sin, optval, sizeof(sin))) {
 212                 ret = -EFAULT;
 213                 goto out;
 214         }
 215
 216         rds_send_drop_to(rs, &sin);
 217 out:
 218         return ret;
 219 }
 220
 221 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
 222                                int optlen)
 223 {
 224         int value;
 225
 226         if (optlen < sizeof(int))
 227                 return -EINVAL;
 228         if (get_user(value, (int __user *) optval))
 229                 return -EFAULT;
 230         *optvar = !!value;
 231         return 0;
 232 }
 233
 234 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
 235                             int optlen)
 236 {
 237         int ret;
 238
 239         ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
 240         if (ret == 0) {
 241                 if (rs->rs_cong_monitor) {
 242                         rds_cong_add_socket(rs);
 243                 } else {
 244                         rds_cong_remove_socket(rs);
 245                         rs->rs_cong_mask = 0;
 246                         rs->rs_cong_notify = 0;
 247                 }
 248         }
 249         return ret;
 250 }
 251
 252 static int rds_setsockopt(struct socket *sock, int level, int optname,
 253                           char __user *optval, int optlen)
 254 {
 255         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 256         int ret;
 257
 258         if (level != SOL_RDS) {
 259                 ret = -ENOPROTOOPT;
 260                 goto out;
 261         }
 262
 263         switch (optname) {
 264         case RDS_CANCEL_SENT_TO:
 265                 ret = rds_cancel_sent_to(rs, optval, optlen);
 266                 break;
 267         case RDS_GET_MR:
 268                 ret = rds_get_mr(rs, optval, optlen);
 269                 break;
 270         case RDS_FREE_MR:
 271                 ret = rds_free_mr(rs, optval, optlen);
 272                 break;
 273         case RDS_RECVERR:
 274                 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
 275                 break;
 276         case RDS_CONG_MONITOR:
 277                 ret = rds_cong_monitor(rs, optval, optlen);
 278                 break;
 279         default:
 280                 ret = -ENOPROTOOPT;
 281         }
 282 out:
 283         return ret;
 284 }
 285
 286 static int rds_getsockopt(struct socket *sock, int level, int optname,
 287                           char __user *optval, int __user *optlen)
 288 {
 289         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 290         int ret = -ENOPROTOOPT, len;
 291
 292         if (level != SOL_RDS)
 293                 goto out;
 294
 295         if (get_user(len, optlen)) {
 296                 ret = -EFAULT;
 297                 goto out;
 298         }
 299
 300         switch (optname) {
 301         case RDS_INFO_FIRST ... RDS_INFO_LAST:
 302                 ret = rds_info_getsockopt(sock, optname, optval,
 303                                           optlen);
 304                 break;
 305
 306         case RDS_RECVERR:
 307                 if (len < sizeof(int))
 308                         ret = -EINVAL;
 309                 else
 310                 if (put_user(rs->rs_recverr, (int __user *) optval)
 311                  || put_user(sizeof(int), optlen))
 312                         ret = -EFAULT;
 313                 else
 314                         ret = 0;
 315                 break;
 316         default:
 317                 break;
 318         }
 319
 320 out:
 321         return ret;
 322
 323 }
 324
 325 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
 326                        int addr_len, int flags)
 327 {
 328         struct sock *sk = sock->sk;
 329         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 330         struct rds_sock *rs = rds_sk_to_rs(sk);
 331         int ret = 0;
 332
 333         lock_sock(sk);
 334
 335         if (addr_len != sizeof(struct sockaddr_in)) {
 336                 ret = -EINVAL;
 337                 goto out;
 338         }
 339
 340         if (sin->sin_family != AF_INET) {
 341                 ret = -EAFNOSUPPORT;
 342                 goto out;
 343         }
 344
 345         if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
 346                 ret = -EDESTADDRREQ;
 347                 goto out;
 348         }
 349
 350         rs->rs_conn_addr = sin->sin_addr.s_addr;
 351         rs->rs_conn_port = sin->sin_port;
 352
 353 out:
 354         release_sock(sk);
 355         return ret;
 356 }
 357
 358 static struct proto rds_proto = {
 359         .name     = "RDS",
 360         .owner    = THIS_MODULE,
 361         .obj_size = sizeof(struct rds_sock),
 362 };
 363
 364 static struct proto_ops rds_proto_ops = {
 365         .family =       AF_RDS,
 366         .owner =        THIS_MODULE,
 367         .release =      rds_release,
 368         .bind =         rds_bind,
 369         .connect =      rds_connect,
 370         .socketpair =   sock_no_socketpair,
 371         .accept =       sock_no_accept,
 372         .getname =      rds_getname,
 373         .poll =         rds_poll,
 374         .ioctl =        rds_ioctl,
 375         .listen =       sock_no_listen,
 376         .shutdown =     sock_no_shutdown,
 377         .setsockopt =   rds_setsockopt,
 378         .getsockopt =   rds_getsockopt,
 379         .sendmsg =      rds_sendmsg,
 380         .recvmsg =      rds_recvmsg,
 381         .mmap =         sock_no_mmap,
 382         .sendpage =     sock_no_sendpage,
 383 };
 384
 385 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
 386 {
 387         unsigned long flags;
 388         struct rds_sock *rs;
 389
 390         sock_init_data(sock, sk);
 391         sock->ops               = &rds_proto_ops;
 392         sk->sk_protocol         = protocol;
 393
 394         rs = rds_sk_to_rs(sk);
 395         spin_lock_init(&rs->rs_lock);
 396         rwlock_init(&rs->rs_recv_lock);
 397         INIT_LIST_HEAD(&rs->rs_send_queue);
 398         INIT_LIST_HEAD(&rs->rs_recv_queue);
 399         INIT_LIST_HEAD(&rs->rs_notify_queue);
 400         INIT_LIST_HEAD(&rs->rs_cong_list);
 401         spin_lock_init(&rs->rs_rdma_lock);
 402         rs->rs_rdma_keys = RB_ROOT;
 403
 404         spin_lock_irqsave(&rds_sock_lock, flags);
 405         list_add_tail(&rs->rs_item, &rds_sock_list);
 406         rds_sock_count++;
 407         spin_unlock_irqrestore(&rds_sock_lock, flags);
 408
 409         return 0;
 410 }
 411
 412 static int rds_create(struct net *net, struct socket *sock, int protocol)
 413 {
 414         struct sock *sk;
 415
 416         if (sock->type != SOCK_SEQPACKET || protocol)
 417                 return -ESOCKTNOSUPPORT;
 418
 419         sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto);
 420         if (!sk)
 421                 return -ENOMEM;
 422
 423         return __rds_create(sock, sk, protocol);
 424 }
 425
 426 void rds_sock_addref(struct rds_sock *rs)
 427 {
 428         sock_hold(rds_rs_to_sk(rs));
 429 }
 430
 431 void rds_sock_put(struct rds_sock *rs)
 432 {
 433         sock_put(rds_rs_to_sk(rs));
 434 }
 435
 436 static struct net_proto_family rds_family_ops = {
 437         .family =       AF_RDS,
 438         .create =       rds_create,
 439         .owner  =       THIS_MODULE,
 440 };
 441
 442 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
 443                               struct rds_info_iterator *iter,
 444                               struct rds_info_lengths *lens)
 445 {
 446         struct rds_sock *rs;
 447         struct sock *sk;
 448         struct rds_incoming *inc;
 449         unsigned long flags;
 450         unsigned int total = 0;
 451
 452         len /= sizeof(struct rds_info_message);
 453
 454         spin_lock_irqsave(&rds_sock_lock, flags);
 455
 456         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 457                 sk = rds_rs_to_sk(rs);
 458                 read_lock(&rs->rs_recv_lock);
 459
 460                 /* XXX too lazy to maintain counts.. */
 461                 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
 462                         total++;
 463                         if (total <= len)
 464                                 rds_inc_info_copy(inc, iter, inc->i_saddr,
 465                                                   rs->rs_bound_addr, 1);
 466                 }
 467
 468                 read_unlock(&rs->rs_recv_lock);
 469         }
 470
 471         spin_unlock_irqrestore(&rds_sock_lock, flags);
 472
 473         lens->nr = total;
 474         lens->each = sizeof(struct rds_info_message);
 475 }
 476
 477 static void rds_sock_info(struct socket *sock, unsigned int len,
 478                           struct rds_info_iterator *iter,
 479                           struct rds_info_lengths *lens)
 480 {
 481         struct rds_info_socket sinfo;
 482         struct rds_sock *rs;
 483         unsigned long flags;
 484
 485         len /= sizeof(struct rds_info_socket);
 486
 487         spin_lock_irqsave(&rds_sock_lock, flags);
 488
 489         if (len < rds_sock_count)
 490                 goto out;
 491
 492         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 493                 sinfo.sndbuf = rds_sk_sndbuf(rs);
 494                 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
 495                 sinfo.bound_addr = rs->rs_bound_addr;
 496                 sinfo.connected_addr = rs->rs_conn_addr;
 497                 sinfo.bound_port = rs->rs_bound_port;
 498                 sinfo.connected_port = rs->rs_conn_port;
 499                 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
 500
 501                 rds_info_copy(iter, &sinfo, sizeof(sinfo));
 502         }
 503
 504 out:
 505         lens->nr = rds_sock_count;
 506         lens->each = sizeof(struct rds_info_socket);
 507
 508         spin_unlock_irqrestore(&rds_sock_lock, flags);
 509 }
 510
 511 static void __exit rds_exit(void)
 512 {
 513         rds_rdma_exit();
 514         sock_unregister(rds_family_ops.family);
 515         proto_unregister(&rds_proto);
 516         rds_conn_exit();
 517         rds_cong_exit();
 518         rds_sysctl_exit();
 519         rds_threads_exit();
 520         rds_stats_exit();
 521         rds_page_exit();
 522         rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
 523         rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 524 }
 525 module_exit(rds_exit);
 526
 527 static int __init rds_init(void)
 528 {
 529         int ret;
 530
 531         ret = rds_conn_init();
 532         if (ret)
 533                 goto out;
 534         ret = rds_threads_init();
 535         if (ret)
 536                 goto out_conn;
 537         ret = rds_sysctl_init();
 538         if (ret)
 539                 goto out_threads;
 540         ret = rds_stats_init();
 541         if (ret)
 542                 goto out_sysctl;
 543         ret = proto_register(&rds_proto, 1);
 544         if (ret)
 545                 goto out_stats;
 546         ret = sock_register(&rds_family_ops);
 547         if (ret)
 548                 goto out_proto;
 549
 550         rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
 551         rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 552
 553         /* ib/iwarp transports currently compiled-in */
 554         ret = rds_rdma_init();
 555         if (ret)
 556                 goto out_sock;
 557         goto out;
 558
 559 out_sock:
 560         sock_unregister(rds_family_ops.family);
 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);