net/rds/recv.c

   1 /*
   2  * Copyright (c) 2006, 2019 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/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                  struct in6_addr *saddr)
  45 {
  46         int i;
  47
  48         refcount_set(&inc->i_refcount, 1);
  49         INIT_LIST_HEAD(&inc->i_item);
  50         inc->i_conn = conn;
  51         inc->i_saddr = *saddr;
  52         inc->i_rdma_cookie = 0;
  53         inc->i_rx_tstamp.tv_sec = 0;
  54         inc->i_rx_tstamp.tv_usec = 0;
  55
  56         for (i = 0; i < RDS_RX_MAX_TRACES; i++)
  57                 inc->i_rx_lat_trace[i] = 0;
  58 }
  59 EXPORT_SYMBOL_GPL(rds_inc_init);
  60
  61 void rds_inc_path_init(struct rds_incoming *inc, struct rds_conn_path *cp,
  62                        struct in6_addr  *saddr)
  63 {
  64         refcount_set(&inc->i_refcount, 1);
  65         INIT_LIST_HEAD(&inc->i_item);
  66         inc->i_conn = cp->cp_conn;
  67         inc->i_conn_path = cp;
  68         inc->i_saddr = *saddr;
  69         inc->i_rdma_cookie = 0;
  70         inc->i_rx_tstamp.tv_sec = 0;
  71         inc->i_rx_tstamp.tv_usec = 0;
  72 }
  73 EXPORT_SYMBOL_GPL(rds_inc_path_init);
  74
  75 static void rds_inc_addref(struct rds_incoming *inc)
  76 {
  77         rdsdebug("addref inc %p ref %d\n", inc, refcount_read(&inc->i_refcount));
  78         refcount_inc(&inc->i_refcount);
  79 }
  80
  81 void rds_inc_put(struct rds_incoming *inc)
  82 {
  83         rdsdebug("put inc %p ref %d\n", inc, refcount_read(&inc->i_refcount));
  84         if (refcount_dec_and_test(&inc->i_refcount)) {
  85                 BUG_ON(!list_empty(&inc->i_item));
  86
  87                 inc->i_conn->c_trans->inc_free(inc);
  88         }
  89 }
  90 EXPORT_SYMBOL_GPL(rds_inc_put);
  91
  92 static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk,
  93                                   struct rds_cong_map *map,
  94                                   int delta, __be16 port)
  95 {
  96         int now_congested;
  97
  98         if (delta == 0)
  99                 return;
 100
 101         rs->rs_rcv_bytes += delta;
 102         if (delta > 0)
 103                 rds_stats_add(s_recv_bytes_added_to_socket, delta);
 104         else
 105                 rds_stats_add(s_recv_bytes_removed_from_socket, -delta);
 106
 107         /* loop transport doesn't send/recv congestion updates */
 108         if (rs->rs_transport->t_type == RDS_TRANS_LOOP)
 109                 return;
 110
 111         now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs);
 112
 113         rdsdebug("rs %p (%pI6c:%u) recv bytes %d buf %d "
 114           "now_cong %d delta %d\n",
 115           rs, &rs->rs_bound_addr,
 116           ntohs(rs->rs_bound_port), rs->rs_rcv_bytes,
 117           rds_sk_rcvbuf(rs), now_congested, delta);
 118
 119         /* wasn't -> am congested */
 120         if (!rs->rs_congested && now_congested) {
 121                 rs->rs_congested = 1;
 122                 rds_cong_set_bit(map, port);
 123                 rds_cong_queue_updates(map);
 124         }
 125         /* was -> aren't congested */
 126         /* Require more free space before reporting uncongested to prevent
 127            bouncing cong/uncong state too often */
 128         else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) {
 129                 rs->rs_congested = 0;
 130                 rds_cong_clear_bit(map, port);
 131                 rds_cong_queue_updates(map);
 132         }
 133
 134         /* do nothing if no change in cong state */
 135 }
 136
 137 static void rds_conn_peer_gen_update(struct rds_connection *conn,
 138                                      u32 peer_gen_num)
 139 {
 140         int i;
 141         struct rds_message *rm, *tmp;
 142         unsigned long flags;
 143
 144         WARN_ON(conn->c_trans->t_type != RDS_TRANS_TCP);
 145         if (peer_gen_num != 0) {
 146                 if (conn->c_peer_gen_num != 0 &&
 147                     peer_gen_num != conn->c_peer_gen_num) {
 148                         for (i = 0; i < RDS_MPATH_WORKERS; i++) {
 149                                 struct rds_conn_path *cp;
 150
 151                                 cp = &conn->c_path[i];
 152                                 spin_lock_irqsave(&cp->cp_lock, flags);
 153                                 cp->cp_next_tx_seq = 1;
 154                                 cp->cp_next_rx_seq = 0;
 155                                 list_for_each_entry_safe(rm, tmp,
 156                                                          &cp->cp_retrans,
 157                                                          m_conn_item) {
 158                                         set_bit(RDS_MSG_FLUSH, &rm->m_flags);
 159                                 }
 160                                 spin_unlock_irqrestore(&cp->cp_lock, flags);
 161                         }
 162                 }
 163                 conn->c_peer_gen_num = peer_gen_num;
 164         }
 165 }
 166
 167 /*
 168  * Process all extension headers that come with this message.
 169  */
 170 static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs)
 171 {
 172         struct rds_header *hdr = &inc->i_hdr;
 173         unsigned int pos = 0, type, len;
 174         union {
 175                 struct rds_ext_header_version version;
 176                 struct rds_ext_header_rdma rdma;
 177                 struct rds_ext_header_rdma_dest rdma_dest;
 178         } buffer;
 179
 180         while (1) {
 181                 len = sizeof(buffer);
 182                 type = rds_message_next_extension(hdr, &pos, &buffer, &len);
 183                 if (type == RDS_EXTHDR_NONE)
 184                         break;
 185                 /* Process extension header here */
 186                 switch (type) {
 187                 case RDS_EXTHDR_RDMA:
 188                         rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0);
 189                         break;
 190
 191                 case RDS_EXTHDR_RDMA_DEST:
 192                         /* We ignore the size for now. We could stash it
 193                          * somewhere and use it for error checking. */
 194                         inc->i_rdma_cookie = rds_rdma_make_cookie(
 195                                         be32_to_cpu(buffer.rdma_dest.h_rdma_rkey),
 196                                         be32_to_cpu(buffer.rdma_dest.h_rdma_offset));
 197
 198                         break;
 199                 }
 200         }
 201 }
 202
 203 static void rds_recv_hs_exthdrs(struct rds_header *hdr,
 204                                 struct rds_connection *conn)
 205 {
 206         unsigned int pos = 0, type, len;
 207         union {
 208                 struct rds_ext_header_version version;
 209                 u16 rds_npaths;
 210                 u32 rds_gen_num;
 211         } buffer;
 212         u32 new_peer_gen_num = 0;
 213
 214         while (1) {
 215                 len = sizeof(buffer);
 216                 type = rds_message_next_extension(hdr, &pos, &buffer, &len);
 217                 if (type == RDS_EXTHDR_NONE)
 218                         break;
 219                 /* Process extension header here */
 220                 switch (type) {
 221                 case RDS_EXTHDR_NPATHS:
 222                         conn->c_npaths = min_t(int, RDS_MPATH_WORKERS,
 223                                                be16_to_cpu(buffer.rds_npaths));
 224                         break;
 225                 case RDS_EXTHDR_GEN_NUM:
 226                         new_peer_gen_num = be32_to_cpu(buffer.rds_gen_num);
 227                         break;
 228                 default:
 229                         pr_warn_ratelimited("ignoring unknown exthdr type "
 230                                              "0x%x\n", type);
 231                 }
 232         }
 233         /* if RDS_EXTHDR_NPATHS was not found, default to a single-path */
 234         conn->c_npaths = max_t(int, conn->c_npaths, 1);
 235         conn->c_ping_triggered = 0;
 236         rds_conn_peer_gen_update(conn, new_peer_gen_num);
 237 }
 238
 239 /* rds_start_mprds() will synchronously start multiple paths when appropriate.
 240  * The scheme is based on the following rules:
 241  *
 242  * 1. rds_sendmsg on first connect attempt sends the probe ping, with the
 243  *    sender's npaths (s_npaths)
 244  * 2. rcvr of probe-ping knows the mprds_paths = min(s_npaths, r_npaths). It
 245  *    sends back a probe-pong with r_npaths. After that, if rcvr is the
 246  *    smaller ip addr, it starts rds_conn_path_connect_if_down on all
 247  *    mprds_paths.
 248  * 3. sender gets woken up, and can move to rds_conn_path_connect_if_down.
 249  *    If it is the smaller ipaddr, rds_conn_path_connect_if_down can be
 250  *    called after reception of the probe-pong on all mprds_paths.
 251  *    Otherwise (sender of probe-ping is not the smaller ip addr): just call
 252  *    rds_conn_path_connect_if_down on the hashed path. (see rule 4)
 253  * 4. rds_connect_worker must only trigger a connection if laddr < faddr.
 254  * 5. sender may end up queuing the packet on the cp. will get sent out later.
 255  *    when connection is completed.
 256  */
 257 static void rds_start_mprds(struct rds_connection *conn)
 258 {
 259         int i;
 260         struct rds_conn_path *cp;
 261
 262         if (conn->c_npaths > 1 &&
 263             rds_addr_cmp(&conn->c_laddr, &conn->c_faddr) < 0) {
 264                 for (i = 0; i < conn->c_npaths; i++) {
 265                         cp = &conn->c_path[i];
 266                         rds_conn_path_connect_if_down(cp);
 267                 }
 268         }
 269 }
 270
 271 /*
 272  * The transport must make sure that this is serialized against other
 273  * rx and conn reset on this specific conn.
 274  *
 275  * We currently assert that only one fragmented message will be sent
 276  * down a connection at a time.  This lets us reassemble in the conn
 277  * instead of per-flow which means that we don't have to go digging through
 278  * flows to tear down partial reassembly progress on conn failure and
 279  * we save flow lookup and locking for each frag arrival.  It does mean
 280  * that small messages will wait behind large ones.  Fragmenting at all
 281  * is only to reduce the memory consumption of pre-posted buffers.
 282  *
 283  * The caller passes in saddr and daddr instead of us getting it from the
 284  * conn.  This lets loopback, who only has one conn for both directions,
 285  * tell us which roles the addrs in the conn are playing for this message.
 286  */
 287 void rds_recv_incoming(struct rds_connection *conn, struct in6_addr *saddr,
 288                        struct in6_addr *daddr,
 289                        struct rds_incoming *inc, gfp_t gfp)
 290 {
 291         struct rds_sock *rs = NULL;
 292         struct sock *sk;
 293         unsigned long flags;
 294         struct rds_conn_path *cp;
 295
 296         inc->i_conn = conn;
 297         inc->i_rx_jiffies = jiffies;
 298         if (conn->c_trans->t_mp_capable)
 299                 cp = inc->i_conn_path;
 300         else
 301                 cp = &conn->c_path[0];
 302
 303         rdsdebug("conn %p next %llu inc %p seq %llu len %u sport %u dport %u "
 304                  "flags 0x%x rx_jiffies %lu\n", conn,
 305                  (unsigned long long)cp->cp_next_rx_seq,
 306                  inc,
 307                  (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence),
 308                  be32_to_cpu(inc->i_hdr.h_len),
 309                  be16_to_cpu(inc->i_hdr.h_sport),
 310                  be16_to_cpu(inc->i_hdr.h_dport),
 311                  inc->i_hdr.h_flags,
 312                  inc->i_rx_jiffies);
 313
 314         /*
 315          * Sequence numbers should only increase.  Messages get their
 316          * sequence number as they're queued in a sending conn.  They
 317          * can be dropped, though, if the sending socket is closed before
 318          * they hit the wire.  So sequence numbers can skip forward
 319          * under normal operation.  They can also drop back in the conn
 320          * failover case as previously sent messages are resent down the
 321          * new instance of a conn.  We drop those, otherwise we have
 322          * to assume that the next valid seq does not come after a
 323          * hole in the fragment stream.
 324          *
 325          * The headers don't give us a way to realize if fragments of
 326          * a message have been dropped.  We assume that frags that arrive
 327          * to a flow are part of the current message on the flow that is
 328          * being reassembled.  This means that senders can't drop messages
 329          * from the sending conn until all their frags are sent.
 330          *
 331          * XXX we could spend more on the wire to get more robust failure
 332          * detection, arguably worth it to avoid data corruption.
 333          */
 334         if (be64_to_cpu(inc->i_hdr.h_sequence) < cp->cp_next_rx_seq &&
 335             (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) {
 336                 rds_stats_inc(s_recv_drop_old_seq);
 337                 goto out;
 338         }
 339         cp->cp_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1;
 340
 341         if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) {
 342                 if (inc->i_hdr.h_sport == 0) {
 343                         rdsdebug("ignore ping with 0 sport from %pI6c\n",
 344                                  saddr);
 345                         goto out;
 346                 }
 347                 rds_stats_inc(s_recv_ping);
 348                 rds_send_pong(cp, inc->i_hdr.h_sport);
 349                 /* if this is a handshake ping, start multipath if necessary */
 350                 if (RDS_HS_PROBE(be16_to_cpu(inc->i_hdr.h_sport),
 351                                  be16_to_cpu(inc->i_hdr.h_dport))) {
 352                         rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn);
 353                         rds_start_mprds(cp->cp_conn);
 354                 }
 355                 goto out;
 356         }
 357
 358         if (be16_to_cpu(inc->i_hdr.h_dport) ==  RDS_FLAG_PROBE_PORT &&
 359             inc->i_hdr.h_sport == 0) {
 360                 rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn);
 361                 /* if this is a handshake pong, start multipath if necessary */
 362                 rds_start_mprds(cp->cp_conn);
 363                 wake_up(&cp->cp_conn->c_hs_waitq);
 364                 goto out;
 365         }
 366
 367         rs = rds_find_bound(daddr, inc->i_hdr.h_dport, conn->c_bound_if);
 368         if (!rs) {
 369                 rds_stats_inc(s_recv_drop_no_sock);
 370                 goto out;
 371         }
 372
 373         /* Process extension headers */
 374         rds_recv_incoming_exthdrs(inc, rs);
 375
 376         /* We can be racing with rds_release() which marks the socket dead. */
 377         sk = rds_rs_to_sk(rs);
 378
 379         /* serialize with rds_release -> sock_orphan */
 380         write_lock_irqsave(&rs->rs_recv_lock, flags);
 381         if (!sock_flag(sk, SOCK_DEAD)) {
 382                 rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs);
 383                 rds_stats_inc(s_recv_queued);
 384                 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
 385                                       be32_to_cpu(inc->i_hdr.h_len),
 386                                       inc->i_hdr.h_dport);
 387                 if (sock_flag(sk, SOCK_RCVTSTAMP))
 388                         do_gettimeofday(&inc->i_rx_tstamp);
 389                 rds_inc_addref(inc);
 390                 inc->i_rx_lat_trace[RDS_MSG_RX_END] = local_clock();
 391                 list_add_tail(&inc->i_item, &rs->rs_recv_queue);
 392                 __rds_wake_sk_sleep(sk);
 393         } else {
 394                 rds_stats_inc(s_recv_drop_dead_sock);
 395         }
 396         write_unlock_irqrestore(&rs->rs_recv_lock, flags);
 397
 398 out:
 399         if (rs)
 400                 rds_sock_put(rs);
 401 }
 402 EXPORT_SYMBOL_GPL(rds_recv_incoming);
 403
 404 /*
 405  * be very careful here.  This is being called as the condition in
 406  * wait_event_*() needs to cope with being called many times.
 407  */
 408 static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc)
 409 {
 410         unsigned long flags;
 411
 412         if (!*inc) {
 413                 read_lock_irqsave(&rs->rs_recv_lock, flags);
 414                 if (!list_empty(&rs->rs_recv_queue)) {
 415                         *inc = list_entry(rs->rs_recv_queue.next,
 416                                           struct rds_incoming,
 417                                           i_item);
 418                         rds_inc_addref(*inc);
 419                 }
 420                 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
 421         }
 422
 423         return *inc != NULL;
 424 }
 425
 426 static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc,
 427                             int drop)
 428 {
 429         struct sock *sk = rds_rs_to_sk(rs);
 430         int ret = 0;
 431         unsigned long flags;
 432
 433         write_lock_irqsave(&rs->rs_recv_lock, flags);
 434         if (!list_empty(&inc->i_item)) {
 435                 ret = 1;
 436                 if (drop) {
 437                         /* XXX make sure this i_conn is reliable */
 438                         rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
 439                                               -be32_to_cpu(inc->i_hdr.h_len),
 440                                               inc->i_hdr.h_dport);
 441                         list_del_init(&inc->i_item);
 442                         rds_inc_put(inc);
 443                 }
 444         }
 445         write_unlock_irqrestore(&rs->rs_recv_lock, flags);
 446
 447         rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop);
 448         return ret;
 449 }
 450
 451 /*
 452  * Pull errors off the error queue.
 453  * If msghdr is NULL, we will just purge the error queue.
 454  */
 455 int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr)
 456 {
 457         struct rds_notifier *notifier;
 458         struct rds_rdma_notify cmsg;
 459         unsigned int count = 0, max_messages = ~0U;
 460         unsigned long flags;
 461         LIST_HEAD(copy);
 462         int err = 0;
 463
 464         memset(&cmsg, 0, sizeof(cmsg)); /* fill holes with zero */
 465
 466         /* put_cmsg copies to user space and thus may sleep. We can't do this
 467          * with rs_lock held, so first grab as many notifications as we can stuff
 468          * in the user provided cmsg buffer. We don't try to copy more, to avoid
 469          * losing notifications - except when the buffer is so small that it wouldn't
 470          * even hold a single notification. Then we give him as much of this single
 471          * msg as we can squeeze in, and set MSG_CTRUNC.
 472          */
 473         if (msghdr) {
 474                 max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg));
 475                 if (!max_messages)
 476                         max_messages = 1;
 477         }
 478
 479         spin_lock_irqsave(&rs->rs_lock, flags);
 480         while (!list_empty(&rs->rs_notify_queue) && count < max_messages) {
 481                 notifier = list_entry(rs->rs_notify_queue.next,
 482                                 struct rds_notifier, n_list);
 483                 list_move(&notifier->n_list, &copy);
 484                 count++;
 485         }
 486         spin_unlock_irqrestore(&rs->rs_lock, flags);
 487
 488         if (!count)
 489                 return 0;
 490
 491         while (!list_empty(&copy)) {
 492                 notifier = list_entry(copy.next, struct rds_notifier, n_list);
 493
 494                 if (msghdr) {
 495                         cmsg.user_token = notifier->n_user_token;
 496                         cmsg.status = notifier->n_status;
 497
 498                         err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_RDMA_STATUS,
 499                                        sizeof(cmsg), &cmsg);
 500                         if (err)
 501                                 break;
 502                 }
 503
 504                 list_del_init(&notifier->n_list);
 505                 kfree(notifier);
 506         }
 507
 508         /* If we bailed out because of an error in put_cmsg,
 509          * we may be left with one or more notifications that we
 510          * didn't process. Return them to the head of the list. */
 511         if (!list_empty(&copy)) {
 512                 spin_lock_irqsave(&rs->rs_lock, flags);
 513                 list_splice(&copy, &rs->rs_notify_queue);
 514                 spin_unlock_irqrestore(&rs->rs_lock, flags);
 515         }
 516
 517         return err;
 518 }
 519
 520 /*
 521  * Queue a congestion notification
 522  */
 523 static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr)
 524 {
 525         uint64_t notify = rs->rs_cong_notify;
 526         unsigned long flags;
 527         int err;
 528
 529         err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_CONG_UPDATE,
 530                         sizeof(notify), &notify);
 531         if (err)
 532                 return err;
 533
 534         spin_lock_irqsave(&rs->rs_lock, flags);
 535         rs->rs_cong_notify &= ~notify;
 536         spin_unlock_irqrestore(&rs->rs_lock, flags);
 537
 538         return 0;
 539 }
 540
 541 /*
 542  * Receive any control messages.
 543  */
 544 static int rds_cmsg_recv(struct rds_incoming *inc, struct msghdr *msg,
 545                          struct rds_sock *rs)
 546 {
 547         int ret = 0;
 548
 549         if (inc->i_rdma_cookie) {
 550                 ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RDMA_DEST,
 551                                 sizeof(inc->i_rdma_cookie), &inc->i_rdma_cookie);
 552                 if (ret)
 553                         goto out;
 554         }
 555
 556         if ((inc->i_rx_tstamp.tv_sec != 0) &&
 557             sock_flag(rds_rs_to_sk(rs), SOCK_RCVTSTAMP)) {
 558                 ret = put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
 559                                sizeof(struct timeval),
 560                                &inc->i_rx_tstamp);
 561                 if (ret)
 562                         goto out;
 563         }
 564
 565         if (rs->rs_rx_traces) {
 566                 struct rds_cmsg_rx_trace t;
 567                 int i, j;
 568
 569                 memset(&t, 0, sizeof(t));
 570                 inc->i_rx_lat_trace[RDS_MSG_RX_CMSG] = local_clock();
 571                 t.rx_traces =  rs->rs_rx_traces;
 572                 for (i = 0; i < rs->rs_rx_traces; i++) {
 573                         j = rs->rs_rx_trace[i];
 574                         t.rx_trace_pos[i] = j;
 575                         t.rx_trace[i] = inc->i_rx_lat_trace[j + 1] -
 576                                           inc->i_rx_lat_trace[j];
 577                 }
 578
 579                 ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RXPATH_LATENCY,
 580                                sizeof(t), &t);
 581                 if (ret)
 582                         goto out;
 583         }
 584
 585 out:
 586         return ret;
 587 }
 588
 589 static bool rds_recvmsg_zcookie(struct rds_sock *rs, struct msghdr *msg)
 590 {
 591         struct rds_msg_zcopy_queue *q = &rs->rs_zcookie_queue;
 592         struct rds_msg_zcopy_info *info = NULL;
 593         struct rds_zcopy_cookies *done;
 594         unsigned long flags;
 595
 596         if (!msg->msg_control)
 597                 return false;
 598
 599         if (!sock_flag(rds_rs_to_sk(rs), SOCK_ZEROCOPY) ||
 600             msg->msg_controllen < CMSG_SPACE(sizeof(*done)))
 601                 return false;
 602
 603         spin_lock_irqsave(&q->lock, flags);
 604         if (!list_empty(&q->zcookie_head)) {
 605                 info = list_entry(q->zcookie_head.next,
 606                                   struct rds_msg_zcopy_info, rs_zcookie_next);
 607                 list_del(&info->rs_zcookie_next);
 608         }
 609         spin_unlock_irqrestore(&q->lock, flags);
 610         if (!info)
 611                 return false;
 612         done = &info->zcookies;
 613         if (put_cmsg(msg, SOL_RDS, RDS_CMSG_ZCOPY_COMPLETION, sizeof(*done),
 614                      done)) {
 615                 spin_lock_irqsave(&q->lock, flags);
 616                 list_add(&info->rs_zcookie_next, &q->zcookie_head);
 617                 spin_unlock_irqrestore(&q->lock, flags);
 618                 return false;
 619         }
 620         kfree(info);
 621         return true;
 622 }
 623
 624 int rds_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
 625                 int msg_flags)
 626 {
 627         struct sock *sk = sock->sk;
 628         struct rds_sock *rs = rds_sk_to_rs(sk);
 629         long timeo;
 630         int ret = 0, nonblock = msg_flags & MSG_DONTWAIT;
 631         DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
 632         DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
 633         struct rds_incoming *inc = NULL;
 634
 635         /* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */
 636         timeo = sock_rcvtimeo(sk, nonblock);
 637
 638         rdsdebug("size %zu flags 0x%x timeo %ld\n", size, msg_flags, timeo);
 639
 640         if (msg_flags & MSG_OOB)
 641                 goto out;
 642         if (msg_flags & MSG_ERRQUEUE)
 643                 return sock_recv_errqueue(sk, msg, size, SOL_IP, IP_RECVERR);
 644
 645         while (1) {
 646                 /* If there are pending notifications, do those - and nothing else */
 647                 if (!list_empty(&rs->rs_notify_queue)) {
 648                         ret = rds_notify_queue_get(rs, msg);
 649                         break;
 650                 }
 651
 652                 if (rs->rs_cong_notify) {
 653                         ret = rds_notify_cong(rs, msg);
 654                         break;
 655                 }
 656
 657                 if (!rds_next_incoming(rs, &inc)) {
 658                         if (nonblock) {
 659                                 bool reaped = rds_recvmsg_zcookie(rs, msg);
 660
 661                                 ret = reaped ?  0 : -EAGAIN;
 662                                 break;
 663                         }
 664
 665                         timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
 666                                         (!list_empty(&rs->rs_notify_queue) ||
 667                                          rs->rs_cong_notify ||
 668                                          rds_next_incoming(rs, &inc)), timeo);
 669                         rdsdebug("recvmsg woke inc %p timeo %ld\n", inc,
 670                                  timeo);
 671                         if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
 672                                 continue;
 673
 674                         ret = timeo;
 675                         if (ret == 0)
 676                                 ret = -ETIMEDOUT;
 677                         break;
 678                 }
 679
 680                 rdsdebug("copying inc %p from %pI6c:%u to user\n", inc,
 681                          &inc->i_conn->c_faddr,
 682                          ntohs(inc->i_hdr.h_sport));
 683                 ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter);
 684                 if (ret < 0)
 685                         break;
 686
 687                 /*
 688                  * if the message we just copied isn't at the head of the
 689                  * recv queue then someone else raced us to return it, try
 690                  * to get the next message.
 691                  */
 692                 if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) {
 693                         rds_inc_put(inc);
 694                         inc = NULL;
 695                         rds_stats_inc(s_recv_deliver_raced);
 696                         iov_iter_revert(&msg->msg_iter, ret);
 697                         continue;
 698                 }
 699
 700                 if (ret < be32_to_cpu(inc->i_hdr.h_len)) {
 701                         if (msg_flags & MSG_TRUNC)
 702                                 ret = be32_to_cpu(inc->i_hdr.h_len);
 703                         msg->msg_flags |= MSG_TRUNC;
 704                 }
 705
 706                 if (rds_cmsg_recv(inc, msg, rs)) {
 707                         ret = -EFAULT;
 708                         break;
 709                 }
 710                 rds_recvmsg_zcookie(rs, msg);
 711
 712                 rds_stats_inc(s_recv_delivered);
 713
 714                 if (msg->msg_name) {
 715                         if (ipv6_addr_v4mapped(&inc->i_saddr)) {
 716                                 sin = (struct sockaddr_in *)msg->msg_name;
 717
 718                                 sin->sin_family = AF_INET;
 719                                 sin->sin_port = inc->i_hdr.h_sport;
 720                                 sin->sin_addr.s_addr =
 721                                     inc->i_saddr.s6_addr32[3];
 722                                 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 723                                 msg->msg_namelen = sizeof(*sin);
 724                         } else {
 725                                 sin6 = (struct sockaddr_in6 *)msg->msg_name;
 726
 727                                 sin6->sin6_family = AF_INET6;
 728                                 sin6->sin6_port = inc->i_hdr.h_sport;
 729                                 sin6->sin6_addr = inc->i_saddr;
 730                                 sin6->sin6_flowinfo = 0;
 731                                 sin6->sin6_scope_id = rs->rs_bound_scope_id;
 732                                 msg->msg_namelen = sizeof(*sin6);
 733                         }
 734                 }
 735                 break;
 736         }
 737
 738         if (inc)
 739                 rds_inc_put(inc);
 740
 741 out:
 742         return ret;
 743 }
 744
 745 /*
 746  * The socket is being shut down and we're asked to drop messages that were
 747  * queued for recvmsg.  The caller has unbound the socket so the receive path
 748  * won't queue any more incoming fragments or messages on the socket.
 749  */
 750 void rds_clear_recv_queue(struct rds_sock *rs)
 751 {
 752         struct sock *sk = rds_rs_to_sk(rs);
 753         struct rds_incoming *inc, *tmp;
 754         unsigned long flags;
 755
 756         write_lock_irqsave(&rs->rs_recv_lock, flags);
 757         list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) {
 758                 rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
 759                                       -be32_to_cpu(inc->i_hdr.h_len),
 760                                       inc->i_hdr.h_dport);
 761                 list_del_init(&inc->i_item);
 762                 rds_inc_put(inc);
 763         }
 764         write_unlock_irqrestore(&rs->rs_recv_lock, flags);
 765 }
 766
 767 /*
 768  * inc->i_saddr isn't used here because it is only set in the receive
 769  * path.
 770  */
 771 void rds_inc_info_copy(struct rds_incoming *inc,
 772                        struct rds_info_iterator *iter,
 773                        __be32 saddr, __be32 daddr, int flip)
 774 {
 775         struct rds_info_message minfo;
 776
 777         minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence);
 778         minfo.len = be32_to_cpu(inc->i_hdr.h_len);
 779
 780         if (flip) {
 781                 minfo.laddr = daddr;
 782                 minfo.faddr = saddr;
 783                 minfo.lport = inc->i_hdr.h_dport;
 784                 minfo.fport = inc->i_hdr.h_sport;
 785         } else {
 786                 minfo.laddr = saddr;
 787                 minfo.faddr = daddr;
 788                 minfo.lport = inc->i_hdr.h_sport;
 789                 minfo.fport = inc->i_hdr.h_dport;
 790         }
 791
 792         minfo.flags = 0;
 793
 794         rds_info_copy(iter, &minfo, sizeof(minfo));
 795 }
 796
 797 #if IS_ENABLED(CONFIG_IPV6)
 798 void rds6_inc_info_copy(struct rds_incoming *inc,
 799                         struct rds_info_iterator *iter,
 800                         struct in6_addr *saddr, struct in6_addr *daddr,
 801                         int flip)
 802 {
 803         struct rds6_info_message minfo6;
 804
 805         minfo6.seq = be64_to_cpu(inc->i_hdr.h_sequence);
 806         minfo6.len = be32_to_cpu(inc->i_hdr.h_len);
 807         minfo6.tos = 0;
 808
 809         if (flip) {
 810                 minfo6.laddr = *daddr;
 811                 minfo6.faddr = *saddr;
 812                 minfo6.lport = inc->i_hdr.h_dport;
 813                 minfo6.fport = inc->i_hdr.h_sport;
 814         } else {
 815                 minfo6.laddr = *saddr;
 816                 minfo6.faddr = *daddr;
 817                 minfo6.lport = inc->i_hdr.h_sport;
 818                 minfo6.fport = inc->i_hdr.h_dport;
 819         }
 820
 821         minfo6.flags = 0;
 822
 823         rds_info_copy(iter, &minfo6, sizeof(minfo6));
 824 }
 825 #endif