net/rds/af_rds.c

   1 /*
   2  * Copyright (c) 2006, 2018 Oracle and/or its affiliates. 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/ipv6.h>
  39 #include <linux/poll.h>
  40 #include <net/sock.h>
  41
  42 #include "rds.h"
  43
  44 /* this is just used for stats gathering :/ */
  45 static DEFINE_SPINLOCK(rds_sock_lock);
  46 static unsigned long rds_sock_count;
  47 static LIST_HEAD(rds_sock_list);
  48 DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);
  49
  50 /*
  51  * This is called as the final descriptor referencing this socket is closed.
  52  * We have to unbind the socket so that another socket can be bound to the
  53  * address it was using.
  54  *
  55  * We have to be careful about racing with the incoming path.  sock_orphan()
  56  * sets SOCK_DEAD and we use that as an indicator to the rx path that new
  57  * messages shouldn't be queued.
  58  */
  59 static int rds_release(struct socket *sock)
  60 {
  61         struct sock *sk = sock->sk;
  62         struct rds_sock *rs;
  63
  64         if (!sk)
  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
  76         rds_remove_bound(rs);
  77
  78         rds_send_drop_to(rs, NULL);
  79         rds_rdma_drop_keys(rs);
  80         rds_notify_queue_get(rs, NULL);
  81         rds_notify_msg_zcopy_purge(&rs->rs_zcookie_queue);
  82
  83         spin_lock_bh(&rds_sock_lock);
  84         list_del_init(&rs->rs_item);
  85         rds_sock_count--;
  86         spin_unlock_bh(&rds_sock_lock);
  87
  88         rds_trans_put(rs->rs_transport);
  89
  90         sock->sk = NULL;
  91         sock_put(sk);
  92 out:
  93         return 0;
  94 }
  95
  96 /*
  97  * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
  98  * _bh() isn't OK here, we're called from interrupt handlers.  It's probably OK
  99  * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
 100  * this seems more conservative.
 101  * NB - normally, one would use sk_callback_lock for this, but we can
 102  * get here from interrupts, whereas the network code grabs sk_callback_lock
 103  * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
 104  */
 105 void rds_wake_sk_sleep(struct rds_sock *rs)
 106 {
 107         unsigned long flags;
 108
 109         read_lock_irqsave(&rs->rs_recv_lock, flags);
 110         __rds_wake_sk_sleep(rds_rs_to_sk(rs));
 111         read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 112 }
 113
 114 static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
 115                        int peer)
 116 {
 117         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 118         struct sockaddr_in6 *sin6;
 119         struct sockaddr_in *sin;
 120         int uaddr_len;
 121
 122         /* racey, don't care */
 123         if (peer) {
 124                 if (ipv6_addr_any(&rs->rs_conn_addr))
 125                         return -ENOTCONN;
 126
 127                 if (ipv6_addr_v4mapped(&rs->rs_conn_addr)) {
 128                         sin = (struct sockaddr_in *)uaddr;
 129                         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 130                         sin->sin_family = AF_INET;
 131                         sin->sin_port = rs->rs_conn_port;
 132                         sin->sin_addr.s_addr = rs->rs_conn_addr_v4;
 133                         uaddr_len = sizeof(*sin);
 134                 } else {
 135                         sin6 = (struct sockaddr_in6 *)uaddr;
 136                         sin6->sin6_family = AF_INET6;
 137                         sin6->sin6_port = rs->rs_conn_port;
 138                         sin6->sin6_addr = rs->rs_conn_addr;
 139                         sin6->sin6_flowinfo = 0;
 140                         /* scope_id is the same as in the bound address. */
 141                         sin6->sin6_scope_id = rs->rs_bound_scope_id;
 142                         uaddr_len = sizeof(*sin6);
 143                 }
 144         } else {
 145                 /* If socket is not yet bound and the socket is connected,
 146                  * set the return address family to be the same as the
 147                  * connected address, but with 0 address value.  If it is not
 148                  * connected, set the family to be AF_UNSPEC (value 0) and
 149                  * the address size to be that of an IPv4 address.
 150                  */
 151                 if (ipv6_addr_any(&rs->rs_bound_addr)) {
 152                         if (ipv6_addr_any(&rs->rs_conn_addr)) {
 153                                 sin = (struct sockaddr_in *)uaddr;
 154                                 memset(sin, 0, sizeof(*sin));
 155                                 sin->sin_family = AF_UNSPEC;
 156                                 return sizeof(*sin);
 157                         }
 158
 159 #if IS_ENABLED(CONFIG_IPV6)
 160                         if (!(ipv6_addr_type(&rs->rs_conn_addr) &
 161                               IPV6_ADDR_MAPPED)) {
 162                                 sin6 = (struct sockaddr_in6 *)uaddr;
 163                                 memset(sin6, 0, sizeof(*sin6));
 164                                 sin6->sin6_family = AF_INET6;
 165                                 return sizeof(*sin6);
 166                         }
 167 #endif
 168
 169                         sin = (struct sockaddr_in *)uaddr;
 170                         memset(sin, 0, sizeof(*sin));
 171                         sin->sin_family = AF_INET;
 172                         return sizeof(*sin);
 173                 }
 174                 if (ipv6_addr_v4mapped(&rs->rs_bound_addr)) {
 175                         sin = (struct sockaddr_in *)uaddr;
 176                         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 177                         sin->sin_family = AF_INET;
 178                         sin->sin_port = rs->rs_bound_port;
 179                         sin->sin_addr.s_addr = rs->rs_bound_addr_v4;
 180                         uaddr_len = sizeof(*sin);
 181                 } else {
 182                         sin6 = (struct sockaddr_in6 *)uaddr;
 183                         sin6->sin6_family = AF_INET6;
 184                         sin6->sin6_port = rs->rs_bound_port;
 185                         sin6->sin6_addr = rs->rs_bound_addr;
 186                         sin6->sin6_flowinfo = 0;
 187                         sin6->sin6_scope_id = rs->rs_bound_scope_id;
 188                         uaddr_len = sizeof(*sin6);
 189                 }
 190         }
 191
 192         return uaddr_len;
 193 }
 194
 195 /*
 196  * RDS' poll is without a doubt the least intuitive part of the interface,
 197  * as EPOLLIN and EPOLLOUT do not behave entirely as you would expect from
 198  * a network protocol.
 199  *
 200  * EPOLLIN is asserted if
 201  *  -   there is data on the receive queue.
 202  *  -   to signal that a previously congested destination may have become
 203  *      uncongested
 204  *  -   A notification has been queued to the socket (this can be a congestion
 205  *      update, or a RDMA completion, or a MSG_ZEROCOPY completion).
 206  *
 207  * EPOLLOUT is asserted if there is room on the send queue. This does not mean
 208  * however, that the next sendmsg() call will succeed. If the application tries
 209  * to send to a congested destination, the system call may still fail (and
 210  * return ENOBUFS).
 211  */
 212 static __poll_t rds_poll(struct file *file, struct socket *sock,
 213                              poll_table *wait)
 214 {
 215         struct sock *sk = sock->sk;
 216         struct rds_sock *rs = rds_sk_to_rs(sk);
 217         __poll_t mask = 0;
 218         unsigned long flags;
 219
 220         poll_wait(file, sk_sleep(sk), wait);
 221
 222         if (rs->rs_seen_congestion)
 223                 poll_wait(file, &rds_poll_waitq, wait);
 224
 225         read_lock_irqsave(&rs->rs_recv_lock, flags);
 226         if (!rs->rs_cong_monitor) {
 227                 /* When a congestion map was updated, we signal EPOLLIN for
 228                  * "historical" reasons. Applications can also poll for
 229                  * WRBAND instead. */
 230                 if (rds_cong_updated_since(&rs->rs_cong_track))
 231                         mask |= (EPOLLIN | EPOLLRDNORM | EPOLLWRBAND);
 232         } else {
 233                 spin_lock(&rs->rs_lock);
 234                 if (rs->rs_cong_notify)
 235                         mask |= (EPOLLIN | EPOLLRDNORM);
 236                 spin_unlock(&rs->rs_lock);
 237         }
 238         if (!list_empty(&rs->rs_recv_queue) ||
 239             !list_empty(&rs->rs_notify_queue) ||
 240             !list_empty(&rs->rs_zcookie_queue.zcookie_head))
 241                 mask |= (EPOLLIN | EPOLLRDNORM);
 242         if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
 243                 mask |= (EPOLLOUT | EPOLLWRNORM);
 244         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
 245                 mask |= POLLERR;
 246         read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 247
 248         /* clear state any time we wake a seen-congested socket */
 249         if (mask)
 250                 rs->rs_seen_congestion = 0;
 251
 252         return mask;
 253 }
 254
 255 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 256 {
 257         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 258         rds_tos_t utos, tos = 0;
 259
 260         switch (cmd) {
 261         case SIOCRDSSETTOS:
 262                 if (get_user(utos, (rds_tos_t __user *)arg))
 263                         return -EFAULT;
 264
 265                 if (rs->rs_transport &&
 266                     rs->rs_transport->get_tos_map)
 267                         tos = rs->rs_transport->get_tos_map(utos);
 268                 else
 269                         return -ENOIOCTLCMD;
 270
 271                 spin_lock_bh(&rds_sock_lock);
 272                 if (rs->rs_tos || rs->rs_conn) {
 273                         spin_unlock_bh(&rds_sock_lock);
 274                         return -EINVAL;
 275                 }
 276                 rs->rs_tos = tos;
 277                 spin_unlock_bh(&rds_sock_lock);
 278                 break;
 279         case SIOCRDSGETTOS:
 280                 spin_lock_bh(&rds_sock_lock);
 281                 tos = rs->rs_tos;
 282                 spin_unlock_bh(&rds_sock_lock);
 283                 if (put_user(tos, (rds_tos_t __user *)arg))
 284                         return -EFAULT;
 285                 break;
 286         default:
 287                 return -ENOIOCTLCMD;
 288         }
 289
 290         return 0;
 291 }
 292
 293 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
 294                               int len)
 295 {
 296         struct sockaddr_in6 sin6;
 297         struct sockaddr_in sin;
 298         int ret = 0;
 299
 300         /* racing with another thread binding seems ok here */
 301         if (ipv6_addr_any(&rs->rs_bound_addr)) {
 302                 ret = -ENOTCONN; /* XXX not a great errno */
 303                 goto out;
 304         }
 305
 306         if (len < sizeof(struct sockaddr_in)) {
 307                 ret = -EINVAL;
 308                 goto out;
 309         } else if (len < sizeof(struct sockaddr_in6)) {
 310                 /* Assume IPv4 */
 311                 if (copy_from_user(&sin, optval, sizeof(struct sockaddr_in))) {
 312                         ret = -EFAULT;
 313                         goto out;
 314                 }
 315                 ipv6_addr_set_v4mapped(sin.sin_addr.s_addr, &sin6.sin6_addr);
 316                 sin6.sin6_port = sin.sin_port;
 317         } else {
 318                 if (copy_from_user(&sin6, optval,
 319                                    sizeof(struct sockaddr_in6))) {
 320                         ret = -EFAULT;
 321                         goto out;
 322                 }
 323         }
 324
 325         rds_send_drop_to(rs, &sin6);
 326 out:
 327         return ret;
 328 }
 329
 330 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
 331                                int optlen)
 332 {
 333         int value;
 334
 335         if (optlen < sizeof(int))
 336                 return -EINVAL;
 337         if (get_user(value, (int __user *) optval))
 338                 return -EFAULT;
 339         *optvar = !!value;
 340         return 0;
 341 }
 342
 343 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
 344                             int optlen)
 345 {
 346         int ret;
 347
 348         ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
 349         if (ret == 0) {
 350                 if (rs->rs_cong_monitor) {
 351                         rds_cong_add_socket(rs);
 352                 } else {
 353                         rds_cong_remove_socket(rs);
 354                         rs->rs_cong_mask = 0;
 355                         rs->rs_cong_notify = 0;
 356                 }
 357         }
 358         return ret;
 359 }
 360
 361 static int rds_set_transport(struct rds_sock *rs, char __user *optval,
 362                              int optlen)
 363 {
 364         int t_type;
 365
 366         if (rs->rs_transport)
 367                 return -EOPNOTSUPP; /* previously attached to transport */
 368
 369         if (optlen != sizeof(int))
 370                 return -EINVAL;
 371
 372         if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type)))
 373                 return -EFAULT;
 374
 375         if (t_type < 0 || t_type >= RDS_TRANS_COUNT)
 376                 return -EINVAL;
 377
 378         rs->rs_transport = rds_trans_get(t_type);
 379
 380         return rs->rs_transport ? 0 : -ENOPROTOOPT;
 381 }
 382
 383 static int rds_enable_recvtstamp(struct sock *sk, char __user *optval,
 384                                  int optlen, int optname)
 385 {
 386         int val, valbool;
 387
 388         if (optlen != sizeof(int))
 389                 return -EFAULT;
 390
 391         if (get_user(val, (int __user *)optval))
 392                 return -EFAULT;
 393
 394         valbool = val ? 1 : 0;
 395
 396         if (optname == SO_TIMESTAMP_NEW)
 397                 sock_set_flag(sk, SOCK_TSTAMP_NEW);
 398
 399         if (valbool)
 400                 sock_set_flag(sk, SOCK_RCVTSTAMP);
 401         else
 402                 sock_reset_flag(sk, SOCK_RCVTSTAMP);
 403
 404         return 0;
 405 }
 406
 407 static int rds_recv_track_latency(struct rds_sock *rs, char __user *optval,
 408                                   int optlen)
 409 {
 410         struct rds_rx_trace_so trace;
 411         int i;
 412
 413         if (optlen != sizeof(struct rds_rx_trace_so))
 414                 return -EFAULT;
 415
 416         if (copy_from_user(&trace, optval, sizeof(trace)))
 417                 return -EFAULT;
 418
 419         if (trace.rx_traces > RDS_MSG_RX_DGRAM_TRACE_MAX)
 420                 return -EFAULT;
 421
 422         rs->rs_rx_traces = trace.rx_traces;
 423         for (i = 0; i < rs->rs_rx_traces; i++) {
 424                 if (trace.rx_trace_pos[i] > RDS_MSG_RX_DGRAM_TRACE_MAX) {
 425                         rs->rs_rx_traces = 0;
 426                         return -EFAULT;
 427                 }
 428                 rs->rs_rx_trace[i] = trace.rx_trace_pos[i];
 429         }
 430
 431         return 0;
 432 }
 433
 434 static int rds_setsockopt(struct socket *sock, int level, int optname,
 435                           char __user *optval, unsigned int optlen)
 436 {
 437         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 438         int ret;
 439
 440         if (level != SOL_RDS) {
 441                 ret = -ENOPROTOOPT;
 442                 goto out;
 443         }
 444
 445         switch (optname) {
 446         case RDS_CANCEL_SENT_TO:
 447                 ret = rds_cancel_sent_to(rs, optval, optlen);
 448                 break;
 449         case RDS_GET_MR:
 450                 ret = rds_get_mr(rs, optval, optlen);
 451                 break;
 452         case RDS_GET_MR_FOR_DEST:
 453                 ret = rds_get_mr_for_dest(rs, optval, optlen);
 454                 break;
 455         case RDS_FREE_MR:
 456                 ret = rds_free_mr(rs, optval, optlen);
 457                 break;
 458         case RDS_RECVERR:
 459                 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
 460                 break;
 461         case RDS_CONG_MONITOR:
 462                 ret = rds_cong_monitor(rs, optval, optlen);
 463                 break;
 464         case SO_RDS_TRANSPORT:
 465                 lock_sock(sock->sk);
 466                 ret = rds_set_transport(rs, optval, optlen);
 467                 release_sock(sock->sk);
 468                 break;
 469         case SO_TIMESTAMP_OLD:
 470         case SO_TIMESTAMP_NEW:
 471                 lock_sock(sock->sk);
 472                 ret = rds_enable_recvtstamp(sock->sk, optval, optlen, optname);
 473                 release_sock(sock->sk);
 474                 break;
 475         case SO_RDS_MSG_RXPATH_LATENCY:
 476                 ret = rds_recv_track_latency(rs, optval, optlen);
 477                 break;
 478         default:
 479                 ret = -ENOPROTOOPT;
 480         }
 481 out:
 482         return ret;
 483 }
 484
 485 static int rds_getsockopt(struct socket *sock, int level, int optname,
 486                           char __user *optval, int __user *optlen)
 487 {
 488         struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 489         int ret = -ENOPROTOOPT, len;
 490         int trans;
 491
 492         if (level != SOL_RDS)
 493                 goto out;
 494
 495         if (get_user(len, optlen)) {
 496                 ret = -EFAULT;
 497                 goto out;
 498         }
 499
 500         switch (optname) {
 501         case RDS_INFO_FIRST ... RDS_INFO_LAST:
 502                 ret = rds_info_getsockopt(sock, optname, optval,
 503                                           optlen);
 504                 break;
 505
 506         case RDS_RECVERR:
 507                 if (len < sizeof(int))
 508                         ret = -EINVAL;
 509                 else
 510                 if (put_user(rs->rs_recverr, (int __user *) optval) ||
 511                     put_user(sizeof(int), optlen))
 512                         ret = -EFAULT;
 513                 else
 514                         ret = 0;
 515                 break;
 516         case SO_RDS_TRANSPORT:
 517                 if (len < sizeof(int)) {
 518                         ret = -EINVAL;
 519                         break;
 520                 }
 521                 trans = (rs->rs_transport ? rs->rs_transport->t_type :
 522                          RDS_TRANS_NONE); /* unbound */
 523                 if (put_user(trans, (int __user *)optval) ||
 524                     put_user(sizeof(int), optlen))
 525                         ret = -EFAULT;
 526                 else
 527                         ret = 0;
 528                 break;
 529         default:
 530                 break;
 531         }
 532
 533 out:
 534         return ret;
 535
 536 }
 537
 538 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
 539                        int addr_len, int flags)
 540 {
 541         struct sock *sk = sock->sk;
 542         struct sockaddr_in *sin;
 543         struct rds_sock *rs = rds_sk_to_rs(sk);
 544         int ret = 0;
 545
 546         lock_sock(sk);
 547
 548         switch (uaddr->sa_family) {
 549         case AF_INET:
 550                 sin = (struct sockaddr_in *)uaddr;
 551                 if (addr_len < sizeof(struct sockaddr_in)) {
 552                         ret = -EINVAL;
 553                         break;
 554                 }
 555                 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
 556                         ret = -EDESTADDRREQ;
 557                         break;
 558                 }
 559                 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) ||
 560                     sin->sin_addr.s_addr == htonl(INADDR_BROADCAST)) {
 561                         ret = -EINVAL;
 562                         break;
 563                 }
 564                 ipv6_addr_set_v4mapped(sin->sin_addr.s_addr, &rs->rs_conn_addr);
 565                 rs->rs_conn_port = sin->sin_port;
 566                 break;
 567
 568 #if IS_ENABLED(CONFIG_IPV6)
 569         case AF_INET6: {
 570                 struct sockaddr_in6 *sin6;
 571                 int addr_type;
 572
 573                 sin6 = (struct sockaddr_in6 *)uaddr;
 574                 if (addr_len < sizeof(struct sockaddr_in6)) {
 575                         ret = -EINVAL;
 576                         break;
 577                 }
 578                 addr_type = ipv6_addr_type(&sin6->sin6_addr);
 579                 if (!(addr_type & IPV6_ADDR_UNICAST)) {
 580                         __be32 addr4;
 581
 582                         if (!(addr_type & IPV6_ADDR_MAPPED)) {
 583                                 ret = -EPROTOTYPE;
 584                                 break;
 585                         }
 586
 587                         /* It is a mapped address.  Need to do some sanity
 588                          * checks.
 589                          */
 590                         addr4 = sin6->sin6_addr.s6_addr32[3];
 591                         if (addr4 == htonl(INADDR_ANY) ||
 592                             addr4 == htonl(INADDR_BROADCAST) ||
 593                             IN_MULTICAST(ntohl(addr4))) {
 594                                 ret = -EPROTOTYPE;
 595                                 break;
 596                         }
 597                 }
 598
 599                 if (addr_type & IPV6_ADDR_LINKLOCAL) {
 600                         /* If socket is arleady bound to a link local address,
 601                          * the peer address must be on the same link.
 602                          */
 603                         if (sin6->sin6_scope_id == 0 ||
 604                             (!ipv6_addr_any(&rs->rs_bound_addr) &&
 605                              rs->rs_bound_scope_id &&
 606                              sin6->sin6_scope_id != rs->rs_bound_scope_id)) {
 607                                 ret = -EINVAL;
 608                                 break;
 609                         }
 610                         /* Remember the connected address scope ID.  It will
 611                          * be checked against the binding local address when
 612                          * the socket is bound.
 613                          */
 614                         rs->rs_bound_scope_id = sin6->sin6_scope_id;
 615                 }
 616                 rs->rs_conn_addr = sin6->sin6_addr;
 617                 rs->rs_conn_port = sin6->sin6_port;
 618                 break;
 619         }
 620 #endif
 621
 622         default:
 623                 ret = -EAFNOSUPPORT;
 624                 break;
 625         }
 626
 627         release_sock(sk);
 628         return ret;
 629 }
 630
 631 static struct proto rds_proto = {
 632         .name     = "RDS",
 633         .owner    = THIS_MODULE,
 634         .obj_size = sizeof(struct rds_sock),
 635 };
 636
 637 static const struct proto_ops rds_proto_ops = {
 638         .family =       AF_RDS,
 639         .owner =        THIS_MODULE,
 640         .release =      rds_release,
 641         .bind =         rds_bind,
 642         .connect =      rds_connect,
 643         .socketpair =   sock_no_socketpair,
 644         .accept =       sock_no_accept,
 645         .getname =      rds_getname,
 646         .poll =         rds_poll,
 647         .ioctl =        rds_ioctl,
 648         .listen =       sock_no_listen,
 649         .shutdown =     sock_no_shutdown,
 650         .setsockopt =   rds_setsockopt,
 651         .getsockopt =   rds_getsockopt,
 652         .sendmsg =      rds_sendmsg,
 653         .recvmsg =      rds_recvmsg,
 654         .mmap =         sock_no_mmap,
 655         .sendpage =     sock_no_sendpage,
 656 };
 657
 658 static void rds_sock_destruct(struct sock *sk)
 659 {
 660         struct rds_sock *rs = rds_sk_to_rs(sk);
 661
 662         WARN_ON((&rs->rs_item != rs->rs_item.next ||
 663                  &rs->rs_item != rs->rs_item.prev));
 664 }
 665
 666 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
 667 {
 668         struct rds_sock *rs;
 669
 670         sock_init_data(sock, sk);
 671         sock->ops               = &rds_proto_ops;
 672         sk->sk_protocol         = protocol;
 673         sk->sk_destruct         = rds_sock_destruct;
 674
 675         rs = rds_sk_to_rs(sk);
 676         spin_lock_init(&rs->rs_lock);
 677         rwlock_init(&rs->rs_recv_lock);
 678         INIT_LIST_HEAD(&rs->rs_send_queue);
 679         INIT_LIST_HEAD(&rs->rs_recv_queue);
 680         INIT_LIST_HEAD(&rs->rs_notify_queue);
 681         INIT_LIST_HEAD(&rs->rs_cong_list);
 682         rds_message_zcopy_queue_init(&rs->rs_zcookie_queue);
 683         spin_lock_init(&rs->rs_rdma_lock);
 684         rs->rs_rdma_keys = RB_ROOT;
 685         rs->rs_rx_traces = 0;
 686         rs->rs_tos = 0;
 687         rs->rs_conn = NULL;
 688
 689         spin_lock_bh(&rds_sock_lock);
 690         list_add_tail(&rs->rs_item, &rds_sock_list);
 691         rds_sock_count++;
 692         spin_unlock_bh(&rds_sock_lock);
 693
 694         return 0;
 695 }
 696
 697 static int rds_create(struct net *net, struct socket *sock, int protocol,
 698                       int kern)
 699 {
 700         struct sock *sk;
 701
 702         if (sock->type != SOCK_SEQPACKET || protocol)
 703                 return -ESOCKTNOSUPPORT;
 704
 705         sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern);
 706         if (!sk)
 707                 return -ENOMEM;
 708
 709         return __rds_create(sock, sk, protocol);
 710 }
 711
 712 void rds_sock_addref(struct rds_sock *rs)
 713 {
 714         sock_hold(rds_rs_to_sk(rs));
 715 }
 716
 717 void rds_sock_put(struct rds_sock *rs)
 718 {
 719         sock_put(rds_rs_to_sk(rs));
 720 }
 721
 722 static const struct net_proto_family rds_family_ops = {
 723         .family =       AF_RDS,
 724         .create =       rds_create,
 725         .owner  =       THIS_MODULE,
 726 };
 727
 728 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
 729                               struct rds_info_iterator *iter,
 730                               struct rds_info_lengths *lens)
 731 {
 732         struct rds_sock *rs;
 733         struct rds_incoming *inc;
 734         unsigned int total = 0;
 735
 736         len /= sizeof(struct rds_info_message);
 737
 738         spin_lock_bh(&rds_sock_lock);
 739
 740         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 741                 read_lock(&rs->rs_recv_lock);
 742
 743                 /* XXX too lazy to maintain counts.. */
 744                 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
 745                         total++;
 746                         if (total <= len)
 747                                 rds_inc_info_copy(inc, iter,
 748                                                   inc->i_saddr.s6_addr32[3],
 749                                                   rs->rs_bound_addr_v4,
 750                                                   1);
 751                 }
 752
 753                 read_unlock(&rs->rs_recv_lock);
 754         }
 755
 756         spin_unlock_bh(&rds_sock_lock);
 757
 758         lens->nr = total;
 759         lens->each = sizeof(struct rds_info_message);
 760 }
 761
 762 static void rds_sock_info(struct socket *sock, unsigned int len,
 763                           struct rds_info_iterator *iter,
 764                           struct rds_info_lengths *lens)
 765 {
 766         struct rds_info_socket sinfo;
 767         struct rds_sock *rs;
 768
 769         len /= sizeof(struct rds_info_socket);
 770
 771         spin_lock_bh(&rds_sock_lock);
 772
 773         if (len < rds_sock_count)
 774                 goto out;
 775
 776         list_for_each_entry(rs, &rds_sock_list, rs_item) {
 777                 sinfo.sndbuf = rds_sk_sndbuf(rs);
 778                 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
 779                 sinfo.bound_addr = rs->rs_bound_addr_v4;
 780                 sinfo.connected_addr = rs->rs_conn_addr_v4;
 781                 sinfo.bound_port = rs->rs_bound_port;
 782                 sinfo.connected_port = rs->rs_conn_port;
 783                 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
 784
 785                 rds_info_copy(iter, &sinfo, sizeof(sinfo));
 786         }
 787
 788 out:
 789         lens->nr = rds_sock_count;
 790         lens->each = sizeof(struct rds_info_socket);
 791
 792         spin_unlock_bh(&rds_sock_lock);
 793 }
 794
 795 static void rds_exit(void)
 796 {
 797         sock_unregister(rds_family_ops.family);
 798         proto_unregister(&rds_proto);
 799         rds_conn_exit();
 800         rds_cong_exit();
 801         rds_sysctl_exit();
 802         rds_threads_exit();
 803         rds_stats_exit();
 804         rds_page_exit();
 805         rds_bind_lock_destroy();
 806         rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
 807         rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 808 }
 809 module_exit(rds_exit);
 810
 811 u32 rds_gen_num;
 812
 813 static int rds_init(void)
 814 {
 815         int ret;
 816
 817         net_get_random_once(&rds_gen_num, sizeof(rds_gen_num));
 818
 819         ret = rds_bind_lock_init();
 820         if (ret)
 821                 goto out;
 822
 823         ret = rds_conn_init();
 824         if (ret)
 825                 goto out_bind;
 826
 827         ret = rds_threads_init();
 828         if (ret)
 829                 goto out_conn;
 830         ret = rds_sysctl_init();
 831         if (ret)
 832                 goto out_threads;
 833         ret = rds_stats_init();
 834         if (ret)
 835                 goto out_sysctl;
 836         ret = proto_register(&rds_proto, 1);
 837         if (ret)
 838                 goto out_stats;
 839         ret = sock_register(&rds_family_ops);
 840         if (ret)
 841                 goto out_proto;
 842
 843         rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
 844         rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
 845
 846         goto out;
 847
 848 out_proto:
 849         proto_unregister(&rds_proto);
 850 out_stats:
 851         rds_stats_exit();
 852 out_sysctl:
 853         rds_sysctl_exit();
 854 out_threads:
 855         rds_threads_exit();
 856 out_conn:
 857         rds_conn_exit();
 858         rds_cong_exit();
 859         rds_page_exit();
 860 out_bind:
 861         rds_bind_lock_destroy();
 862 out:
 863         return ret;
 864 }
 865 module_init(rds_init);
 866
 867 #define DRV_VERSION     "4.0"
 868 #define DRV_RELDATE     "Feb 12, 2009"
 869
 870 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
 871 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
 872                    " v" DRV_VERSION " (" DRV_RELDATE ")");
 873 MODULE_VERSION(DRV_VERSION);
 874 MODULE_LICENSE("Dual BSD/GPL");
 875 MODULE_ALIAS_NETPROTO(PF_RDS);