rds/recv.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/kernel.h>
  34 #include <linux/slab.h>
  35 #include <net/sock.h>
  36 #include <linux/in.h>
  37 #include <linux/export.h>
  38 #include <linux/time.h>
  39 #include <linux/rds.h>
  40
  41 #include "rds.h"
  42
  43 void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn,
  44                   __be32 saddr)
  45 {
  46         atomic_set(&inc->i_refcount, 1);
  47         INIT_LIST_HEAD(&inc->i_item);
  48         inc->i_conn = conn;
  49         inc->i_saddr = saddr;
  50         inc->i_rdma_cookie = 0;
  51         inc->i_rx_tstamp.tv_sec = 0;
  52         inc->i_rx_tstamp.tv_usec = 0;
  53 }
  54 EXPORT_SYMBOL_GPL(rds_inc_init);
  55
  56 static void rds_inc_addref(struct rds_incoming *inc)
  57 {
  58         rdsdebug("addref inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
  59         atomic_inc(&inc->i_refcount);
  60 }
  61
  62 void rds_inc_put(struct rds_incoming *inc)
  63 {
  64         rdsdebug("put inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
  65         if (atomic_dec_and_test(&inc->i_refcount)) {
  66                 BUG_ON(!list_empty(&inc->i_item));
  67
  68                 inc->i_conn->c_trans->inc_free(inc);
  69         }
  70 }
  71 EXPORT_SYMBOL_GPL(rds_inc_put);
  72
  73 static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk,
  74                                   struct rds_cong_map *map,
  75                                   int delta, __be16 port)
  76 {
  77         int now_congested;
  78
  79         if (delta == 0)
  80                 return;
  81
  82         rs->rs_rcv_bytes += delta;
  83         now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs);
  84
  85         rdsdebug("rs %p (%pI4:%u) recv bytes %d buf %d "
  86           "now_cong %d delta %d\n",
  87           rs, &rs->rs_bound_addr,
  88           ntohs(rs->rs_bound_port), rs->rs_rcv_bytes,
  89           rds_sk_rcvbuf(rs), now_congested, delta);
  90
  91         /* wasn't -> am congested */
  92         if (!rs->rs_congested && now_congested) {
  93                 rs->rs_congested = 1;
  94                 rds_cong_set_bit(map, port);
  95                 rds_cong_queue_updates(map);
  96         }
  97         /* was -> aren't congested */
  98         /* Require more free space before reporting uncongested to prevent
  99            bouncing cong/uncong state too often */
 100         else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) {
 101                 rs->rs_congested = 0;
 102                 rds_cong_clear_bit(map, port);
 103                 rds_cong_queue_updates(map);
 104         }
 105
 106         /* do nothing if no change in cong state */
 107 }
 108
 109 /*
 110  * Process all extension headers that come with this message.
 111  */
 112 static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs)
 113 {
 114         struct rds_header *hdr = &inc->i_hdr;
 115         unsigned int pos = 0, type, len;
 116         union {
 117                 struct rds_ext_header_version version;
 118                 struct rds_ext_header_rdma rdma;
 119                 struct rds_ext_header_rdma_dest rdma_dest;
 120         } buffer;
 121
 122         while (1) {
 123                 len = sizeof(buffer);
 124                 type = rds_message_next_extension(hdr, &pos, &buffer, &len);
 125                 if (type == RDS_EXTHDR_NONE)
 126                         break;
 127                 /* Process extension header here */
 128                 switch (type) {
 129                 case RDS_EXTHDR_RDMA:
 130                         rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0);
 131                         break;
 132
 133                 case RDS_EXTHDR_RDMA_DEST:
 134                         /* We ignore the size for now. We could stash it
 135                          * somewhere and use it for error checking. */
 136                         inc->i_rdma_cookie = rds_rdma_make_cookie(
 137                                         be32_to_cpu(buffer.rdma_dest.h_rdma_rkey),
 138                                         be32_to_cpu(buffer.rdma_dest.h_rdma_offset));
 139
 140                         break;
 141                 }
 142         }
 143 }
 144
 145 /*
 146  * The transport must make sure that this is serialized against other
 147  * rx and conn reset on this specific conn.
 148  *
 149  * We currently assert that only one fragmented message will be sent
 150  * down a connection at a time.  This lets us reassemble in the conn
 151  * instead of per-flow which means that we don't have to go digging through
 152  * flows to tear down partial reassembly progress on conn failure and
 153  * we save flow lookup and locking for each frag arrival.  It does mean
 154  * that small messages will wait behind large ones.  Fragmenting at all
 155  * is only to reduce the memory consumption of pre-posted buffers.
 156  *
 157  * The caller passes in saddr and daddr instead of us getting it from the
 158  * conn.  This lets loopback, who only has one conn for both directions,
 159  * tell us which roles the addrs in the conn are playing for this message.
 160  */
 161 void rds_recv_incoming(struct rds_connection *conn, __be32 saddr, __be32 daddr,
 162                        struct rds_incoming *inc, gfp_t gfp)
 163 {
 164         struct rds_sock *rs = NULL;
 165         struct sock *sk;
 166         unsigned long flags;
 167
 168         inc->i_conn = conn;
 169         inc->i_rx_jiffies = jiffies;
 170
 171         rdsdebug("conn %p next %llu inc %p seq %llu len %u sport %u dport %u "
 172                  "flags 0x%x rx_jiffies %lu\n", conn,
 173                  (unsigned long long)conn->c_next_rx_seq,
 174                  inc,
 175                  (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence),
 176                  be32_to_cpu(inc->i_hdr.h_len),
 177                  be16_to_cpu(inc->i_hdr.h_sport),
 178                  be16_to_cpu(inc->i_hdr.h_dport),
 179                  inc->i_hdr.h_flags,
 180                  inc->i_rx_jiffies);
 181
 182         /*
 183          * Sequence numbers should only increase.  Messages get their
 184          * sequence number as they're queued in a sending conn.  They
 185          * can be dropped, though, if the sending socket is closed before
 186          * they hit the wire.  So sequence numbers can skip forward
 187          * under normal operation.  They can also drop back in the conn
 188          * failover case as previously sent messages are resent down the
 189          * new instance of a conn.  We drop those, otherwise we have
 190          * to assume that the next valid seq does not come after a
 191          * hole in the fragment stream.
 192          *
 193          * The headers don't give us a way to realize if fragments of
 194          * a message have been dropped.  We assume that frags that arrive
 195          * to a flow are part of the current message on the flow that is
 196          * being reassembled.  This means that senders can't drop messages
 197          * from the sending conn until all their frags are sent.
 198          *
 199          * XXX we could spend more on the wire to get more robust failure
 200          * detection, arguably worth it to avoid data corruption.
 201          */
 202         if (be64_to_cpu(inc->i_hdr.h_sequence) < conn->c_next_rx_seq &&
 203             (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) {
 204                 rds_stats_inc(s_recv_drop_old_seq);
 205                 goto out;
 206         }
 207         conn->c_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1;
 208
 209         if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) {
 210                 rds_stats_inc(s_recv_ping);
 211                 rds_send_pong(conn, inc->i_hdr.h_sport);
 212                 goto out;
 213         }
 214
 215         rs = rds_find_bound(daddr, inc->i_hdr.h_dport);
 216         if (!rs) {
 217                 rds_stats_inc(s_recv_drop_no_sock);
 218                 goto out;
 219         }
 220
 221         /* Process extension headers */
 222         rds_recv_incoming_exthdrs(inc, rs);
 223
 224         /* We can be racing with rds_release() which marks the socket dead. */
 225         sk = rds_rs_to_sk(rs);
 226
 227         /* serialize with rds_release -> sock_orphan */
 228         write_lock_irqsave(&rs->rs_recv_lock, flags);
 229         if (!sock_flag(sk, SOCK_DEAD)) {
 230                 rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs);
 231                 rds_stats_inc(s_recv_queued);
 232                 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
 233                                       be32_to_cpu(inc->i_hdr.h_len),
 234                                       inc->i_hdr.h_dport);
 235                 if (sock_flag(sk, SOCK_RCVTSTAMP))
 236                         do_gettimeofday(&inc->i_rx_tstamp);
 237                 rds_inc_addref(inc);
 238                 list_add_tail(&inc->i_item, &rs->rs_recv_queue);
 239                 __rds_wake_sk_sleep(sk);
 240         } else {
 241                 rds_stats_inc(s_recv_drop_dead_sock);
 242         }
 243         write_unlock_irqrestore(&rs->rs_recv_lock, flags);
 244
 245 out:
 246         if (rs)
 247                 rds_sock_put(rs);
 248 }
 249 EXPORT_SYMBOL_GPL(rds_recv_incoming);
 250
 251 /*
 252  * be very careful here.  This is being called as the condition in
 253  * wait_event_*() needs to cope with being called many times.
 254  */
 255 static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc)
 256 {
 257         unsigned long flags;
 258
 259         if (!*inc) {
 260                 read_lock_irqsave(&rs->rs_recv_lock, flags);
 261                 if (!list_empty(&rs->rs_recv_queue)) {
 262                         *inc = list_entry(rs->rs_recv_queue.next,
 263                                           struct rds_incoming,
 264                                           i_item);
 265                         rds_inc_addref(*inc);
 266                 }
 267                 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 268         }
 269
 270         return *inc != NULL;
 271 }
 272
 273 static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc,
 274                             int drop)
 275 {
 276         struct sock *sk = rds_rs_to_sk(rs);
 277         int ret = 0;
 278         unsigned long flags;
 279
 280         write_lock_irqsave(&rs->rs_recv_lock, flags);
 281         if (!list_empty(&inc->i_item)) {
 282                 ret = 1;
 283                 if (drop) {
 284                         /* XXX make sure this i_conn is reliable */
 285                         rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
 286                                               -be32_to_cpu(inc->i_hdr.h_len),
 287                                               inc->i_hdr.h_dport);
 288                         list_del_init(&inc->i_item);
 289                         rds_inc_put(inc);
 290                 }
 291         }
 292         write_unlock_irqrestore(&rs->rs_recv_lock, flags);
 293
 294         rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop);
 295         return ret;
 296 }
 297
 298 /*
 299  * Pull errors off the error queue.
 300  * If msghdr is NULL, we will just purge the error queue.
 301  */
 302 int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr)
 303 {
 304         struct rds_notifier *notifier;
 305         struct rds_rdma_notify cmsg = { 0 }; /* fill holes with zero */
 306         unsigned int count = 0, max_messages = ~0U;
 307         unsigned long flags;
 308         LIST_HEAD(copy);
 309         int err = 0;
 310
 311
 312         /* put_cmsg copies to user space and thus may sleep. We can't do this
 313          * with rs_lock held, so first grab as many notifications as we can stuff
 314          * in the user provided cmsg buffer. We don't try to copy more, to avoid
 315          * losing notifications - except when the buffer is so small that it wouldn't
 316          * even hold a single notification. Then we give him as much of this single
 317          * msg as we can squeeze in, and set MSG_CTRUNC.
 318          */
 319         if (msghdr) {
 320                 max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg));
 321                 if (!max_messages)
 322                         max_messages = 1;
 323         }
 324
 325         spin_lock_irqsave(&rs->rs_lock, flags);
 326         while (!list_empty(&rs->rs_notify_queue) && count < max_messages) {
 327                 notifier = list_entry(rs->rs_notify_queue.next,
 328                                 struct rds_notifier, n_list);
 329                 list_move(&notifier->n_list, &copy);
 330                 count++;
 331         }
 332         spin_unlock_irqrestore(&rs->rs_lock, flags);
 333
 334         if (!count)
 335                 return 0;
 336
 337         while (!list_empty(&copy)) {
 338                 notifier = list_entry(copy.next, struct rds_notifier, n_list);
 339
 340                 if (msghdr) {
 341                         cmsg.user_token = notifier->n_user_token;
 342                         cmsg.status = notifier->n_status;
 343
 344                         err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_RDMA_STATUS,
 345                                        sizeof(cmsg), &cmsg);
 346                         if (err)
 347                                 break;
 348                 }
 349
 350                 list_del_init(&notifier->n_list);
 351                 kfree(notifier);
 352         }
 353
 354         /* If we bailed out because of an error in put_cmsg,
 355          * we may be left with one or more notifications that we
 356          * didn't process. Return them to the head of the list. */
 357         if (!list_empty(&copy)) {
 358                 spin_lock_irqsave(&rs->rs_lock, flags);
 359                 list_splice(&copy, &rs->rs_notify_queue);
 360                 spin_unlock_irqrestore(&rs->rs_lock, flags);
 361         }
 362
 363         return err;
 364 }
 365
 366 /*
 367  * Queue a congestion notification
 368  */
 369 static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr)
 370 {
 371         uint64_t notify = rs->rs_cong_notify;
 372         unsigned long flags;
 373         int err;
 374
 375         err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_CONG_UPDATE,
 376                         sizeof(notify), &notify);
 377         if (err)
 378                 return err;
 379
 380         spin_lock_irqsave(&rs->rs_lock, flags);
 381         rs->rs_cong_notify &= ~notify;
 382         spin_unlock_irqrestore(&rs->rs_lock, flags);
 383
 384         return 0;
 385 }
 386
 387 /*
 388  * Receive any control messages.
 389  */
 390 static int rds_cmsg_recv(struct rds_incoming *inc, struct msghdr *msg,
 391                          struct rds_sock *rs)
 392 {
 393         int ret = 0;
 394
 395         if (inc->i_rdma_cookie) {
 396                 ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RDMA_DEST,
 397                                 sizeof(inc->i_rdma_cookie), &inc->i_rdma_cookie);
 398                 if (ret)
 399                         return ret;
 400         }
 401
 402         if ((inc->i_rx_tstamp.tv_sec != 0) &&
 403             sock_flag(rds_rs_to_sk(rs), SOCK_RCVTSTAMP)) {
 404                 ret = put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
 405                                sizeof(struct timeval),
 406                                &inc->i_rx_tstamp);
 407                 if (ret)
 408                         return ret;
 409         }
 410
 411         return 0;
 412 }
 413
 414 int rds_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 415                 int msg_flags)
 416 {
 417         struct sock *sk = sock->sk;
 418         struct rds_sock *rs = rds_sk_to_rs(sk);
 419         long timeo;
 420         int ret = 0, nonblock = msg_flags & MSG_DONTWAIT;
 421         DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
 422         struct rds_incoming *inc = NULL;
 423
 424         /* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */
 425         timeo = sock_rcvtimeo(sk, nonblock);
 426
 427         rdsdebug("size %zu flags 0x%x timeo %ld\n", size, msg_flags, timeo);
 428
 429         if (msg_flags & MSG_OOB)
 430                 goto out;
 431
 432         while (1) {
 433                 struct iov_iter save;
 434                 /* If there are pending notifications, do those - and nothing else */
 435                 if (!list_empty(&rs->rs_notify_queue)) {
 436                         ret = rds_notify_queue_get(rs, msg);
 437                         break;
 438                 }
 439
 440                 if (rs->rs_cong_notify) {
 441                         ret = rds_notify_cong(rs, msg);
 442                         break;
 443                 }
 444
 445                 if (!rds_next_incoming(rs, &inc)) {
 446                         if (nonblock) {
 447                                 ret = -EAGAIN;
 448                                 break;
 449                         }
 450
 451                         timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
 452                                         (!list_empty(&rs->rs_notify_queue) ||
 453                                          rs->rs_cong_notify ||
 454                                          rds_next_incoming(rs, &inc)), timeo);
 455                         rdsdebug("recvmsg woke inc %p timeo %ld\n", inc,
 456                                  timeo);
 457                         if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
 458                                 continue;
 459
 460                         ret = timeo;
 461                         if (ret == 0)
 462                                 ret = -ETIMEDOUT;
 463                         break;
 464                 }
 465
 466                 rdsdebug("copying inc %p from %pI4:%u to user\n", inc,
 467                          &inc->i_conn->c_faddr,
 468                          ntohs(inc->i_hdr.h_sport));
 469                 save = msg->msg_iter;
 470                 ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter);
 471                 if (ret < 0)
 472                         break;
 473
 474                 /*
 475                  * if the message we just copied isn't at the head of the
 476                  * recv queue then someone else raced us to return it, try
 477                  * to get the next message.
 478                  */
 479                 if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) {
 480                         rds_inc_put(inc);
 481                         inc = NULL;
 482                         rds_stats_inc(s_recv_deliver_raced);
 483                         msg->msg_iter = save;
 484                         continue;
 485                 }
 486
 487                 if (ret < be32_to_cpu(inc->i_hdr.h_len)) {
 488                         if (msg_flags & MSG_TRUNC)
 489                                 ret = be32_to_cpu(inc->i_hdr.h_len);
 490                         msg->msg_flags |= MSG_TRUNC;
 491                 }
 492
 493                 if (rds_cmsg_recv(inc, msg, rs)) {
 494                         ret = -EFAULT;
 495                         goto out;
 496                 }
 497
 498                 rds_stats_inc(s_recv_delivered);
 499
 500                 if (sin) {
 501                         sin->sin_family = AF_INET;
 502                         sin->sin_port = inc->i_hdr.h_sport;
 503                         sin->sin_addr.s_addr = inc->i_saddr;
 504                         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 505                         msg->msg_namelen = sizeof(*sin);
 506                 }
 507                 break;
 508         }
 509
 510         if (inc)
 511                 rds_inc_put(inc);
 512
 513 out:
 514         return ret;
 515 }
 516
 517 /*
 518  * The socket is being shut down and we're asked to drop messages that were
 519  * queued for recvmsg.  The caller has unbound the socket so the receive path
 520  * won't queue any more incoming fragments or messages on the socket.
 521  */
 522 void rds_clear_recv_queue(struct rds_sock *rs)
 523 {
 524         struct sock *sk = rds_rs_to_sk(rs);
 525         struct rds_incoming *inc, *tmp;
 526         unsigned long flags;
 527
 528         write_lock_irqsave(&rs->rs_recv_lock, flags);
 529         list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) {
 530                 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
 531                                       -be32_to_cpu(inc->i_hdr.h_len),
 532                                       inc->i_hdr.h_dport);
 533                 list_del_init(&inc->i_item);
 534                 rds_inc_put(inc);
 535         }
 536         write_unlock_irqrestore(&rs->rs_recv_lock, flags);
 537 }
 538
 539 /*
 540  * inc->i_saddr isn't used here because it is only set in the receive
 541  * path.
 542  */
 543 void rds_inc_info_copy(struct rds_incoming *inc,
 544                        struct rds_info_iterator *iter,
 545                        __be32 saddr, __be32 daddr, int flip)
 546 {
 547         struct rds_info_message minfo;
 548
 549         minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence);
 550         minfo.len = be32_to_cpu(inc->i_hdr.h_len);
 551
 552         if (flip) {
 553                 minfo.laddr = daddr;
 554                 minfo.faddr = saddr;
 555                 minfo.lport = inc->i_hdr.h_dport;
 556                 minfo.fport = inc->i_hdr.h_sport;
 557         } else {
 558                 minfo.laddr = saddr;
 559                 minfo.faddr = daddr;
 560                 minfo.lport = inc->i_hdr.h_sport;
 561                 minfo.fport = inc->i_hdr.h_dport;
 562         }
 563
 564         rds_info_copy(iter, &minfo, sizeof(minfo));
 565 }