net/rds/send.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/moduleparam.h>
  35 #include <linux/gfp.h>
  36 #include <net/sock.h>
  37 #include <linux/in.h>
  38 #include <linux/list.h>
  39 #include <linux/ratelimit.h>
  40 #include <linux/export.h>
  41 #include <linux/sizes.h>
  42
  43 #include "rds.h"
  44
  45 /* When transmitting messages in rds_send_xmit, we need to emerge from
  46  * time to time and briefly release the CPU. Otherwise the softlock watchdog
  47  * will kick our shin.
  48  * Also, it seems fairer to not let one busy connection stall all the
  49  * others.
  50  *
  51  * send_batch_count is the number of times we'll loop in send_xmit. Setting
  52  * it to 0 will restore the old behavior (where we looped until we had
  53  * drained the queue).
  54  */
  55 static int send_batch_count = SZ_1K;
  56 module_param(send_batch_count, int, 0444);
  57 MODULE_PARM_DESC(send_batch_count, " batch factor when working the send queue");
  58
  59 static void rds_send_remove_from_sock(struct list_head *messages, int status);
  60
  61 /*
  62  * Reset the send state.  Callers must ensure that this doesn't race with
  63  * rds_send_xmit().
  64  */
  65 void rds_send_path_reset(struct rds_conn_path *cp)
  66 {
  67         struct rds_message *rm, *tmp;
  68         unsigned long flags;
  69
  70         if (cp->cp_xmit_rm) {
  71                 rm = cp->cp_xmit_rm;
  72                 cp->cp_xmit_rm = NULL;
  73                 /* Tell the user the RDMA op is no longer mapped by the
  74                  * transport. This isn't entirely true (it's flushed out
  75                  * independently) but as the connection is down, there's
  76                  * no ongoing RDMA to/from that memory */
  77                 rds_message_unmapped(rm);
  78                 rds_message_put(rm);
  79         }
  80
  81         cp->cp_xmit_sg = 0;
  82         cp->cp_xmit_hdr_off = 0;
  83         cp->cp_xmit_data_off = 0;
  84         cp->cp_xmit_atomic_sent = 0;
  85         cp->cp_xmit_rdma_sent = 0;
  86         cp->cp_xmit_data_sent = 0;
  87
  88         cp->cp_conn->c_map_queued = 0;
  89
  90         cp->cp_unacked_packets = rds_sysctl_max_unacked_packets;
  91         cp->cp_unacked_bytes = rds_sysctl_max_unacked_bytes;
  92
  93         /* Mark messages as retransmissions, and move them to the send q */
  94         spin_lock_irqsave(&cp->cp_lock, flags);
  95         list_for_each_entry_safe(rm, tmp, &cp->cp_retrans, m_conn_item) {
  96                 set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags);
  97                 set_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags);
  98         }
  99         list_splice_init(&cp->cp_retrans, &cp->cp_send_queue);
 100         spin_unlock_irqrestore(&cp->cp_lock, flags);
 101 }
 102 EXPORT_SYMBOL_GPL(rds_send_path_reset);
 103
 104 static int acquire_in_xmit(struct rds_conn_path *cp)
 105 {
 106         return test_and_set_bit(RDS_IN_XMIT, &cp->cp_flags) == 0;
 107 }
 108
 109 static void release_in_xmit(struct rds_conn_path *cp)
 110 {
 111         clear_bit(RDS_IN_XMIT, &cp->cp_flags);
 112         smp_mb__after_atomic();
 113         /*
 114          * We don't use wait_on_bit()/wake_up_bit() because our waking is in a
 115          * hot path and finding waiters is very rare.  We don't want to walk
 116          * the system-wide hashed waitqueue buckets in the fast path only to
 117          * almost never find waiters.
 118          */
 119         if (waitqueue_active(&cp->cp_waitq))
 120                 wake_up_all(&cp->cp_waitq);
 121 }
 122
 123 /*
 124  * We're making the conscious trade-off here to only send one message
 125  * down the connection at a time.
 126  *   Pro:
 127  *      - tx queueing is a simple fifo list
 128  *      - reassembly is optional and easily done by transports per conn
 129  *      - no per flow rx lookup at all, straight to the socket
 130  *      - less per-frag memory and wire overhead
 131  *   Con:
 132  *      - queued acks can be delayed behind large messages
 133  *   Depends:
 134  *      - small message latency is higher behind queued large messages
 135  *      - large message latency isn't starved by intervening small sends
 136  */
 137 int rds_send_xmit(struct rds_conn_path *cp)
 138 {
 139         struct rds_connection *conn = cp->cp_conn;
 140         struct rds_message *rm;
 141         unsigned long flags;
 142         unsigned int tmp;
 143         struct scatterlist *sg;
 144         int ret = 0;
 145         LIST_HEAD(to_be_dropped);
 146         int batch_count;
 147         unsigned long send_gen = 0;
 148
 149 restart:
 150         batch_count = 0;
 151
 152         /*
 153          * sendmsg calls here after having queued its message on the send
 154          * queue.  We only have one task feeding the connection at a time.  If
 155          * another thread is already feeding the queue then we back off.  This
 156          * avoids blocking the caller and trading per-connection data between
 157          * caches per message.
 158          */
 159         if (!acquire_in_xmit(cp)) {
 160                 rds_stats_inc(s_send_lock_contention);
 161                 ret = -ENOMEM;
 162                 goto out;
 163         }
 164
 165         /*
 166          * we record the send generation after doing the xmit acquire.
 167          * if someone else manages to jump in and do some work, we'll use
 168          * this to avoid a goto restart farther down.
 169          *
 170          * The acquire_in_xmit() check above ensures that only one
 171          * caller can increment c_send_gen at any time.
 172          */
 173         send_gen = READ_ONCE(cp->cp_send_gen) + 1;
 174         WRITE_ONCE(cp->cp_send_gen, send_gen);
 175
 176         /*
 177          * rds_conn_shutdown() sets the conn state and then tests RDS_IN_XMIT,
 178          * we do the opposite to avoid races.
 179          */
 180         if (!rds_conn_path_up(cp)) {
 181                 release_in_xmit(cp);
 182                 ret = 0;
 183                 goto out;
 184         }
 185
 186         if (conn->c_trans->xmit_path_prepare)
 187                 conn->c_trans->xmit_path_prepare(cp);
 188
 189         /*
 190          * spin trying to push headers and data down the connection until
 191          * the connection doesn't make forward progress.
 192          */
 193         while (1) {
 194
 195                 rm = cp->cp_xmit_rm;
 196
 197                 /*
 198                  * If between sending messages, we can send a pending congestion
 199                  * map update.
 200                  */
 201                 if (!rm && test_and_clear_bit(0, &conn->c_map_queued)) {
 202                         rm = rds_cong_update_alloc(conn);
 203                         if (IS_ERR(rm)) {
 204                                 ret = PTR_ERR(rm);
 205                                 break;
 206                         }
 207                         rm->data.op_active = 1;
 208                         rm->m_inc.i_conn_path = cp;
 209                         rm->m_inc.i_conn = cp->cp_conn;
 210
 211                         cp->cp_xmit_rm = rm;
 212                 }
 213
 214                 /*
 215                  * If not already working on one, grab the next message.
 216                  *
 217                  * cp_xmit_rm holds a ref while we're sending this message down
 218                  * the connction.  We can use this ref while holding the
 219                  * send_sem.. rds_send_reset() is serialized with it.
 220                  */
 221                 if (!rm) {
 222                         unsigned int len;
 223
 224                         batch_count++;
 225
 226                         /* we want to process as big a batch as we can, but
 227                          * we also want to avoid softlockups.  If we've been
 228                          * through a lot of messages, lets back off and see
 229                          * if anyone else jumps in
 230                          */
 231                         if (batch_count >= send_batch_count)
 232                                 goto over_batch;
 233
 234                         spin_lock_irqsave(&cp->cp_lock, flags);
 235
 236                         if (!list_empty(&cp->cp_send_queue)) {
 237                                 rm = list_entry(cp->cp_send_queue.next,
 238                                                 struct rds_message,
 239                                                 m_conn_item);
 240                                 rds_message_addref(rm);
 241
 242                                 /*
 243                                  * Move the message from the send queue to the retransmit
 244                                  * list right away.
 245                                  */
 246                                 list_move_tail(&rm->m_conn_item,
 247                                                &cp->cp_retrans);
 248                         }
 249
 250                         spin_unlock_irqrestore(&cp->cp_lock, flags);
 251
 252                         if (!rm)
 253                                 break;
 254
 255                         /* Unfortunately, the way Infiniband deals with
 256                          * RDMA to a bad MR key is by moving the entire
 257                          * queue pair to error state. We cold possibly
 258                          * recover from that, but right now we drop the
 259                          * connection.
 260                          * Therefore, we never retransmit messages with RDMA ops.
 261                          */
 262                         if (test_bit(RDS_MSG_FLUSH, &rm->m_flags) ||
 263                             (rm->rdma.op_active &&
 264                             test_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags))) {
 265                                 spin_lock_irqsave(&cp->cp_lock, flags);
 266                                 if (test_and_clear_bit(RDS_MSG_ON_CONN, &rm->m_flags))
 267                                         list_move(&rm->m_conn_item, &to_be_dropped);
 268                                 spin_unlock_irqrestore(&cp->cp_lock, flags);
 269                                 continue;
 270                         }
 271
 272                         /* Require an ACK every once in a while */
 273                         len = ntohl(rm->m_inc.i_hdr.h_len);
 274                         if (cp->cp_unacked_packets == 0 ||
 275                             cp->cp_unacked_bytes < len) {
 276                                 __set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags);
 277
 278                                 cp->cp_unacked_packets =
 279                                         rds_sysctl_max_unacked_packets;
 280                                 cp->cp_unacked_bytes =
 281                                         rds_sysctl_max_unacked_bytes;
 282                                 rds_stats_inc(s_send_ack_required);
 283                         } else {
 284                                 cp->cp_unacked_bytes -= len;
 285                                 cp->cp_unacked_packets--;
 286                         }
 287
 288                         cp->cp_xmit_rm = rm;
 289                 }
 290
 291                 /* The transport either sends the whole rdma or none of it */
 292                 if (rm->rdma.op_active && !cp->cp_xmit_rdma_sent) {
 293                         rm->m_final_op = &rm->rdma;
 294                         /* The transport owns the mapped memory for now.
 295                          * You can't unmap it while it's on the send queue
 296                          */
 297                         set_bit(RDS_MSG_MAPPED, &rm->m_flags);
 298                         ret = conn->c_trans->xmit_rdma(conn, &rm->rdma);
 299                         if (ret) {
 300                                 clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
 301                                 wake_up_interruptible(&rm->m_flush_wait);
 302                                 break;
 303                         }
 304                         cp->cp_xmit_rdma_sent = 1;
 305
 306                 }
 307
 308                 if (rm->atomic.op_active && !cp->cp_xmit_atomic_sent) {
 309                         rm->m_final_op = &rm->atomic;
 310                         /* The transport owns the mapped memory for now.
 311                          * You can't unmap it while it's on the send queue
 312                          */
 313                         set_bit(RDS_MSG_MAPPED, &rm->m_flags);
 314                         ret = conn->c_trans->xmit_atomic(conn, &rm->atomic);
 315                         if (ret) {
 316                                 clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
 317                                 wake_up_interruptible(&rm->m_flush_wait);
 318                                 break;
 319                         }
 320                         cp->cp_xmit_atomic_sent = 1;
 321
 322                 }
 323
 324                 /*
 325                  * A number of cases require an RDS header to be sent
 326                  * even if there is no data.
 327                  * We permit 0-byte sends; rds-ping depends on this.
 328                  * However, if there are exclusively attached silent ops,
 329                  * we skip the hdr/data send, to enable silent operation.
 330                  */
 331                 if (rm->data.op_nents == 0) {
 332                         int ops_present;
 333                         int all_ops_are_silent = 1;
 334
 335                         ops_present = (rm->atomic.op_active || rm->rdma.op_active);
 336                         if (rm->atomic.op_active && !rm->atomic.op_silent)
 337                                 all_ops_are_silent = 0;
 338                         if (rm->rdma.op_active && !rm->rdma.op_silent)
 339                                 all_ops_are_silent = 0;
 340
 341                         if (ops_present && all_ops_are_silent
 342                             && !rm->m_rdma_cookie)
 343                                 rm->data.op_active = 0;
 344                 }
 345
 346                 if (rm->data.op_active && !cp->cp_xmit_data_sent) {
 347                         rm->m_final_op = &rm->data;
 348
 349                         ret = conn->c_trans->xmit(conn, rm,
 350                                                   cp->cp_xmit_hdr_off,
 351                                                   cp->cp_xmit_sg,
 352                                                   cp->cp_xmit_data_off);
 353                         if (ret <= 0)
 354                                 break;
 355
 356                         if (cp->cp_xmit_hdr_off < sizeof(struct rds_header)) {
 357                                 tmp = min_t(int, ret,
 358                                             sizeof(struct rds_header) -
 359                                             cp->cp_xmit_hdr_off);
 360                                 cp->cp_xmit_hdr_off += tmp;
 361                                 ret -= tmp;
 362                         }
 363
 364                         sg = &rm->data.op_sg[cp->cp_xmit_sg];
 365                         while (ret) {
 366                                 tmp = min_t(int, ret, sg->length -
 367                                                       cp->cp_xmit_data_off);
 368                                 cp->cp_xmit_data_off += tmp;
 369                                 ret -= tmp;
 370                                 if (cp->cp_xmit_data_off == sg->length) {
 371                                         cp->cp_xmit_data_off = 0;
 372                                         sg++;
 373                                         cp->cp_xmit_sg++;
 374                                         BUG_ON(ret != 0 && cp->cp_xmit_sg ==
 375                                                rm->data.op_nents);
 376                                 }
 377                         }
 378
 379                         if (cp->cp_xmit_hdr_off == sizeof(struct rds_header) &&
 380                             (cp->cp_xmit_sg == rm->data.op_nents))
 381                                 cp->cp_xmit_data_sent = 1;
 382                 }
 383
 384                 /*
 385                  * A rm will only take multiple times through this loop
 386                  * if there is a data op. Thus, if the data is sent (or there was
 387                  * none), then we're done with the rm.
 388                  */
 389                 if (!rm->data.op_active || cp->cp_xmit_data_sent) {
 390                         cp->cp_xmit_rm = NULL;
 391                         cp->cp_xmit_sg = 0;
 392                         cp->cp_xmit_hdr_off = 0;
 393                         cp->cp_xmit_data_off = 0;
 394                         cp->cp_xmit_rdma_sent = 0;
 395                         cp->cp_xmit_atomic_sent = 0;
 396                         cp->cp_xmit_data_sent = 0;
 397
 398                         rds_message_put(rm);
 399                 }
 400         }
 401
 402 over_batch:
 403         if (conn->c_trans->xmit_path_complete)
 404                 conn->c_trans->xmit_path_complete(cp);
 405         release_in_xmit(cp);
 406
 407         /* Nuke any messages we decided not to retransmit. */
 408         if (!list_empty(&to_be_dropped)) {
 409                 /* irqs on here, so we can put(), unlike above */
 410                 list_for_each_entry(rm, &to_be_dropped, m_conn_item)
 411                         rds_message_put(rm);
 412                 rds_send_remove_from_sock(&to_be_dropped, RDS_RDMA_DROPPED);
 413         }
 414
 415         /*
 416          * Other senders can queue a message after we last test the send queue
 417          * but before we clear RDS_IN_XMIT.  In that case they'd back off and
 418          * not try and send their newly queued message.  We need to check the
 419          * send queue after having cleared RDS_IN_XMIT so that their message
 420          * doesn't get stuck on the send queue.
 421          *
 422          * If the transport cannot continue (i.e ret != 0), then it must
 423          * call us when more room is available, such as from the tx
 424          * completion handler.
 425          *
 426          * We have an extra generation check here so that if someone manages
 427          * to jump in after our release_in_xmit, we'll see that they have done
 428          * some work and we will skip our goto
 429          */
 430         if (ret == 0) {
 431                 smp_mb();
 432                 if ((test_bit(0, &conn->c_map_queued) ||
 433                      !list_empty(&cp->cp_send_queue)) &&
 434                         send_gen == READ_ONCE(cp->cp_send_gen)) {
 435                         rds_stats_inc(s_send_lock_queue_raced);
 436                         if (batch_count < send_batch_count)
 437                                 goto restart;
 438                         queue_delayed_work(rds_wq, &cp->cp_send_w, 1);
 439                 }
 440         }
 441 out:
 442         return ret;
 443 }
 444 EXPORT_SYMBOL_GPL(rds_send_xmit);
 445
 446 static void rds_send_sndbuf_remove(struct rds_sock *rs, struct rds_message *rm)
 447 {
 448         u32 len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
 449
 450         assert_spin_locked(&rs->rs_lock);
 451
 452         BUG_ON(rs->rs_snd_bytes < len);
 453         rs->rs_snd_bytes -= len;
 454
 455         if (rs->rs_snd_bytes == 0)
 456                 rds_stats_inc(s_send_queue_empty);
 457 }
 458
 459 static inline int rds_send_is_acked(struct rds_message *rm, u64 ack,
 460                                     is_acked_func is_acked)
 461 {
 462         if (is_acked)
 463                 return is_acked(rm, ack);
 464         return be64_to_cpu(rm->m_inc.i_hdr.h_sequence) <= ack;
 465 }
 466
 467 /*
 468  * This is pretty similar to what happens below in the ACK
 469  * handling code - except that we call here as soon as we get
 470  * the IB send completion on the RDMA op and the accompanying
 471  * message.
 472  */
 473 void rds_rdma_send_complete(struct rds_message *rm, int status)
 474 {
 475         struct rds_sock *rs = NULL;
 476         struct rm_rdma_op *ro;
 477         struct rds_notifier *notifier;
 478         unsigned long flags;
 479         unsigned int notify = 0;
 480
 481         spin_lock_irqsave(&rm->m_rs_lock, flags);
 482
 483         notify =  rm->rdma.op_notify | rm->data.op_notify;
 484         ro = &rm->rdma;
 485         if (test_bit(RDS_MSG_ON_SOCK, &rm->m_flags) &&
 486             ro->op_active && notify && ro->op_notifier) {
 487                 notifier = ro->op_notifier;
 488                 rs = rm->m_rs;
 489                 sock_hold(rds_rs_to_sk(rs));
 490
 491                 notifier->n_status = status;
 492                 spin_lock(&rs->rs_lock);
 493                 list_add_tail(&notifier->n_list, &rs->rs_notify_queue);
 494                 spin_unlock(&rs->rs_lock);
 495
 496                 ro->op_notifier = NULL;
 497         }
 498
 499         spin_unlock_irqrestore(&rm->m_rs_lock, flags);
 500
 501         if (rs) {
 502                 rds_wake_sk_sleep(rs);
 503                 sock_put(rds_rs_to_sk(rs));
 504         }
 505 }
 506 EXPORT_SYMBOL_GPL(rds_rdma_send_complete);
 507
 508 /*
 509  * Just like above, except looks at atomic op
 510  */
 511 void rds_atomic_send_complete(struct rds_message *rm, int status)
 512 {
 513         struct rds_sock *rs = NULL;
 514         struct rm_atomic_op *ao;
 515         struct rds_notifier *notifier;
 516         unsigned long flags;
 517
 518         spin_lock_irqsave(&rm->m_rs_lock, flags);
 519
 520         ao = &rm->atomic;
 521         if (test_bit(RDS_MSG_ON_SOCK, &rm->m_flags)
 522             && ao->op_active && ao->op_notify && ao->op_notifier) {
 523                 notifier = ao->op_notifier;
 524                 rs = rm->m_rs;
 525                 sock_hold(rds_rs_to_sk(rs));
 526
 527                 notifier->n_status = status;
 528                 spin_lock(&rs->rs_lock);
 529                 list_add_tail(&notifier->n_list, &rs->rs_notify_queue);
 530                 spin_unlock(&rs->rs_lock);
 531
 532                 ao->op_notifier = NULL;
 533         }
 534
 535         spin_unlock_irqrestore(&rm->m_rs_lock, flags);
 536
 537         if (rs) {
 538                 rds_wake_sk_sleep(rs);
 539                 sock_put(rds_rs_to_sk(rs));
 540         }
 541 }
 542 EXPORT_SYMBOL_GPL(rds_atomic_send_complete);
 543
 544 /*
 545  * This is the same as rds_rdma_send_complete except we
 546  * don't do any locking - we have all the ingredients (message,
 547  * socket, socket lock) and can just move the notifier.
 548  */
 549 static inline void
 550 __rds_send_complete(struct rds_sock *rs, struct rds_message *rm, int status)
 551 {
 552         struct rm_rdma_op *ro;
 553         struct rm_atomic_op *ao;
 554
 555         ro = &rm->rdma;
 556         if (ro->op_active && ro->op_notify && ro->op_notifier) {
 557                 ro->op_notifier->n_status = status;
 558                 list_add_tail(&ro->op_notifier->n_list, &rs->rs_notify_queue);
 559                 ro->op_notifier = NULL;
 560         }
 561
 562         ao = &rm->atomic;
 563         if (ao->op_active && ao->op_notify && ao->op_notifier) {
 564                 ao->op_notifier->n_status = status;
 565                 list_add_tail(&ao->op_notifier->n_list, &rs->rs_notify_queue);
 566                 ao->op_notifier = NULL;
 567         }
 568
 569         /* No need to wake the app - caller does this */
 570 }
 571
 572 /*
 573  * This removes messages from the socket's list if they're on it.  The list
 574  * argument must be private to the caller, we must be able to modify it
 575  * without locks.  The messages must have a reference held for their
 576  * position on the list.  This function will drop that reference after
 577  * removing the messages from the 'messages' list regardless of if it found
 578  * the messages on the socket list or not.
 579  */
 580 static void rds_send_remove_from_sock(struct list_head *messages, int status)
 581 {
 582         unsigned long flags;
 583         struct rds_sock *rs = NULL;
 584         struct rds_message *rm;
 585
 586         while (!list_empty(messages)) {
 587                 int was_on_sock = 0;
 588
 589                 rm = list_entry(messages->next, struct rds_message,
 590                                 m_conn_item);
 591                 list_del_init(&rm->m_conn_item);
 592
 593                 /*
 594                  * If we see this flag cleared then we're *sure* that someone
 595                  * else beat us to removing it from the sock.  If we race
 596                  * with their flag update we'll get the lock and then really
 597                  * see that the flag has been cleared.
 598                  *
 599                  * The message spinlock makes sure nobody clears rm->m_rs
 600                  * while we're messing with it. It does not prevent the
 601                  * message from being removed from the socket, though.
 602                  */
 603                 spin_lock_irqsave(&rm->m_rs_lock, flags);
 604                 if (!test_bit(RDS_MSG_ON_SOCK, &rm->m_flags))
 605                         goto unlock_and_drop;
 606
 607                 if (rs != rm->m_rs) {
 608                         if (rs) {
 609                                 rds_wake_sk_sleep(rs);
 610                                 sock_put(rds_rs_to_sk(rs));
 611                         }
 612                         rs = rm->m_rs;
 613                         if (rs)
 614                                 sock_hold(rds_rs_to_sk(rs));
 615                 }
 616                 if (!rs)
 617                         goto unlock_and_drop;
 618                 spin_lock(&rs->rs_lock);
 619
 620                 if (test_and_clear_bit(RDS_MSG_ON_SOCK, &rm->m_flags)) {
 621                         struct rm_rdma_op *ro = &rm->rdma;
 622                         struct rds_notifier *notifier;
 623
 624                         list_del_init(&rm->m_sock_item);
 625                         rds_send_sndbuf_remove(rs, rm);
 626
 627                         if (ro->op_active && ro->op_notifier &&
 628                                (ro->op_notify || (ro->op_recverr && status))) {
 629                                 notifier = ro->op_notifier;
 630                                 list_add_tail(&notifier->n_list,
 631                                                 &rs->rs_notify_queue);
 632                                 if (!notifier->n_status)
 633                                         notifier->n_status = status;
 634                                 rm->rdma.op_notifier = NULL;
 635                         }
 636                         was_on_sock = 1;
 637                         rm->m_rs = NULL;
 638                 }
 639                 spin_unlock(&rs->rs_lock);
 640
 641 unlock_and_drop:
 642                 spin_unlock_irqrestore(&rm->m_rs_lock, flags);
 643                 rds_message_put(rm);
 644                 if (was_on_sock)
 645                         rds_message_put(rm);
 646         }
 647
 648         if (rs) {
 649                 rds_wake_sk_sleep(rs);
 650                 sock_put(rds_rs_to_sk(rs));
 651         }
 652 }
 653
 654 /*
 655  * Transports call here when they've determined that the receiver queued
 656  * messages up to, and including, the given sequence number.  Messages are
 657  * moved to the retrans queue when rds_send_xmit picks them off the send
 658  * queue. This means that in the TCP case, the message may not have been
 659  * assigned the m_ack_seq yet - but that's fine as long as tcp_is_acked
 660  * checks the RDS_MSG_HAS_ACK_SEQ bit.
 661  */
 662 void rds_send_path_drop_acked(struct rds_conn_path *cp, u64 ack,
 663                               is_acked_func is_acked)
 664 {
 665         struct rds_message *rm, *tmp;
 666         unsigned long flags;
 667         LIST_HEAD(list);
 668
 669         spin_lock_irqsave(&cp->cp_lock, flags);
 670
 671         list_for_each_entry_safe(rm, tmp, &cp->cp_retrans, m_conn_item) {
 672                 if (!rds_send_is_acked(rm, ack, is_acked))
 673                         break;
 674
 675                 list_move(&rm->m_conn_item, &list);
 676                 clear_bit(RDS_MSG_ON_CONN, &rm->m_flags);
 677         }
 678
 679         /* order flag updates with spin locks */
 680         if (!list_empty(&list))
 681                 smp_mb__after_atomic();
 682
 683         spin_unlock_irqrestore(&cp->cp_lock, flags);
 684
 685         /* now remove the messages from the sock list as needed */
 686         rds_send_remove_from_sock(&list, RDS_RDMA_SUCCESS);
 687 }
 688 EXPORT_SYMBOL_GPL(rds_send_path_drop_acked);
 689
 690 void rds_send_drop_acked(struct rds_connection *conn, u64 ack,
 691                          is_acked_func is_acked)
 692 {
 693         WARN_ON(conn->c_trans->t_mp_capable);
 694         rds_send_path_drop_acked(&conn->c_path[0], ack, is_acked);
 695 }
 696 EXPORT_SYMBOL_GPL(rds_send_drop_acked);
 697
 698 void rds_send_drop_to(struct rds_sock *rs, struct sockaddr_in *dest)
 699 {
 700         struct rds_message *rm, *tmp;
 701         struct rds_connection *conn;
 702         struct rds_conn_path *cp;
 703         unsigned long flags;
 704         LIST_HEAD(list);
 705
 706         /* get all the messages we're dropping under the rs lock */
 707         spin_lock_irqsave(&rs->rs_lock, flags);
 708
 709         list_for_each_entry_safe(rm, tmp, &rs->rs_send_queue, m_sock_item) {
 710                 if (dest && (dest->sin_addr.s_addr != rm->m_daddr ||
 711                              dest->sin_port != rm->m_inc.i_hdr.h_dport))
 712                         continue;
 713
 714                 list_move(&rm->m_sock_item, &list);
 715                 rds_send_sndbuf_remove(rs, rm);
 716                 clear_bit(RDS_MSG_ON_SOCK, &rm->m_flags);
 717         }
 718
 719         /* order flag updates with the rs lock */
 720         smp_mb__after_atomic();
 721
 722         spin_unlock_irqrestore(&rs->rs_lock, flags);
 723
 724         if (list_empty(&list))
 725                 return;
 726
 727         /* Remove the messages from the conn */
 728         list_for_each_entry(rm, &list, m_sock_item) {
 729
 730                 conn = rm->m_inc.i_conn;
 731                 if (conn->c_trans->t_mp_capable)
 732                         cp = rm->m_inc.i_conn_path;
 733                 else
 734                         cp = &conn->c_path[0];
 735
 736                 spin_lock_irqsave(&cp->cp_lock, flags);
 737                 /*
 738                  * Maybe someone else beat us to removing rm from the conn.
 739                  * If we race with their flag update we'll get the lock and
 740                  * then really see that the flag has been cleared.
 741                  */
 742                 if (!test_and_clear_bit(RDS_MSG_ON_CONN, &rm->m_flags)) {
 743                         spin_unlock_irqrestore(&cp->cp_lock, flags);
 744                         spin_lock_irqsave(&rm->m_rs_lock, flags);
 745                         rm->m_rs = NULL;
 746                         spin_unlock_irqrestore(&rm->m_rs_lock, flags);
 747                         continue;
 748                 }
 749                 list_del_init(&rm->m_conn_item);
 750                 spin_unlock_irqrestore(&cp->cp_lock, flags);
 751
 752                 /*
 753                  * Couldn't grab m_rs_lock in top loop (lock ordering),
 754                  * but we can now.
 755                  */
 756                 spin_lock_irqsave(&rm->m_rs_lock, flags);
 757
 758                 spin_lock(&rs->rs_lock);
 759                 __rds_send_complete(rs, rm, RDS_RDMA_CANCELED);
 760                 spin_unlock(&rs->rs_lock);
 761
 762                 rm->m_rs = NULL;
 763                 spin_unlock_irqrestore(&rm->m_rs_lock, flags);
 764
 765                 rds_message_put(rm);
 766         }
 767
 768         rds_wake_sk_sleep(rs);
 769
 770         while (!list_empty(&list)) {
 771                 rm = list_entry(list.next, struct rds_message, m_sock_item);
 772                 list_del_init(&rm->m_sock_item);
 773                 rds_message_wait(rm);
 774
 775                 /* just in case the code above skipped this message
 776                  * because RDS_MSG_ON_CONN wasn't set, run it again here
 777                  * taking m_rs_lock is the only thing that keeps us
 778                  * from racing with ack processing.
 779                  */
 780                 spin_lock_irqsave(&rm->m_rs_lock, flags);
 781
 782                 spin_lock(&rs->rs_lock);
 783                 __rds_send_complete(rs, rm, RDS_RDMA_CANCELED);
 784                 spin_unlock(&rs->rs_lock);
 785
 786                 rm->m_rs = NULL;
 787                 spin_unlock_irqrestore(&rm->m_rs_lock, flags);
 788
 789                 rds_message_put(rm);
 790         }
 791 }
 792
 793 /*
 794  * we only want this to fire once so we use the callers 'queued'.  It's
 795  * possible that another thread can race with us and remove the
 796  * message from the flow with RDS_CANCEL_SENT_TO.
 797  */
 798 static int rds_send_queue_rm(struct rds_sock *rs, struct rds_connection *conn,
 799                              struct rds_conn_path *cp,
 800                              struct rds_message *rm, __be16 sport,
 801                              __be16 dport, int *queued)
 802 {
 803         unsigned long flags;
 804         u32 len;
 805
 806         if (*queued)
 807                 goto out;
 808
 809         len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
 810
 811         /* this is the only place which holds both the socket's rs_lock
 812          * and the connection's c_lock */
 813         spin_lock_irqsave(&rs->rs_lock, flags);
 814
 815         /*
 816          * If there is a little space in sndbuf, we don't queue anything,
 817          * and userspace gets -EAGAIN. But poll() indicates there's send
 818          * room. This can lead to bad behavior (spinning) if snd_bytes isn't
 819          * freed up by incoming acks. So we check the *old* value of
 820          * rs_snd_bytes here to allow the last msg to exceed the buffer,
 821          * and poll() now knows no more data can be sent.
 822          */
 823         if (rs->rs_snd_bytes < rds_sk_sndbuf(rs)) {
 824                 rs->rs_snd_bytes += len;
 825
 826                 /* let recv side know we are close to send space exhaustion.
 827                  * This is probably not the optimal way to do it, as this
 828                  * means we set the flag on *all* messages as soon as our
 829                  * throughput hits a certain threshold.
 830                  */
 831                 if (rs->rs_snd_bytes >= rds_sk_sndbuf(rs) / 2)
 832                         __set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags);
 833
 834                 list_add_tail(&rm->m_sock_item, &rs->rs_send_queue);
 835                 set_bit(RDS_MSG_ON_SOCK, &rm->m_flags);
 836                 rds_message_addref(rm);
 837                 rm->m_rs = rs;
 838
 839                 /* The code ordering is a little weird, but we're
 840                    trying to minimize the time we hold c_lock */
 841                 rds_message_populate_header(&rm->m_inc.i_hdr, sport, dport, 0);
 842                 rm->m_inc.i_conn = conn;
 843                 rm->m_inc.i_conn_path = cp;
 844                 rds_message_addref(rm);
 845
 846                 spin_lock(&cp->cp_lock);
 847                 rm->m_inc.i_hdr.h_sequence = cpu_to_be64(cp->cp_next_tx_seq++);
 848                 list_add_tail(&rm->m_conn_item, &cp->cp_send_queue);
 849                 set_bit(RDS_MSG_ON_CONN, &rm->m_flags);
 850                 spin_unlock(&cp->cp_lock);
 851
 852                 rdsdebug("queued msg %p len %d, rs %p bytes %d seq %llu\n",
 853                          rm, len, rs, rs->rs_snd_bytes,
 854                          (unsigned long long)be64_to_cpu(rm->m_inc.i_hdr.h_sequence));
 855
 856                 *queued = 1;
 857         }
 858
 859         spin_unlock_irqrestore(&rs->rs_lock, flags);
 860 out:
 861         return *queued;
 862 }
 863
 864 /*
 865  * rds_message is getting to be quite complicated, and we'd like to allocate
 866  * it all in one go. This figures out how big it needs to be up front.
 867  */
 868 static int rds_rm_size(struct msghdr *msg, int data_len)
 869 {
 870         struct cmsghdr *cmsg;
 871         int size = 0;
 872         int cmsg_groups = 0;
 873         int retval;
 874
 875         for_each_cmsghdr(cmsg, msg) {
 876                 if (!CMSG_OK(msg, cmsg))
 877                         return -EINVAL;
 878
 879                 if (cmsg->cmsg_level != SOL_RDS)
 880                         continue;
 881
 882                 switch (cmsg->cmsg_type) {
 883                 case RDS_CMSG_RDMA_ARGS:
 884                         cmsg_groups |= 1;
 885                         retval = rds_rdma_extra_size(CMSG_DATA(cmsg));
 886                         if (retval < 0)
 887                                 return retval;
 888                         size += retval;
 889
 890                         break;
 891
 892                 case RDS_CMSG_RDMA_DEST:
 893                 case RDS_CMSG_RDMA_MAP:
 894                         cmsg_groups |= 2;
 895                         /* these are valid but do no add any size */
 896                         break;
 897
 898                 case RDS_CMSG_ATOMIC_CSWP:
 899                 case RDS_CMSG_ATOMIC_FADD:
 900                 case RDS_CMSG_MASKED_ATOMIC_CSWP:
 901                 case RDS_CMSG_MASKED_ATOMIC_FADD:
 902                         cmsg_groups |= 1;
 903                         size += sizeof(struct scatterlist);
 904                         break;
 905
 906                 default:
 907                         return -EINVAL;
 908                 }
 909
 910         }
 911
 912         size += ceil(data_len, PAGE_SIZE) * sizeof(struct scatterlist);
 913
 914         /* Ensure (DEST, MAP) are never used with (ARGS, ATOMIC) */
 915         if (cmsg_groups == 3)
 916                 return -EINVAL;
 917
 918         return size;
 919 }
 920
 921 static int rds_cmsg_send(struct rds_sock *rs, struct rds_message *rm,
 922                          struct msghdr *msg, int *allocated_mr)
 923 {
 924         struct cmsghdr *cmsg;
 925         int ret = 0;
 926
 927         for_each_cmsghdr(cmsg, msg) {
 928                 if (!CMSG_OK(msg, cmsg))
 929                         return -EINVAL;
 930
 931                 if (cmsg->cmsg_level != SOL_RDS)
 932                         continue;
 933
 934                 /* As a side effect, RDMA_DEST and RDMA_MAP will set
 935                  * rm->rdma.m_rdma_cookie and rm->rdma.m_rdma_mr.
 936                  */
 937                 switch (cmsg->cmsg_type) {
 938                 case RDS_CMSG_RDMA_ARGS:
 939                         ret = rds_cmsg_rdma_args(rs, rm, cmsg);
 940                         break;
 941
 942                 case RDS_CMSG_RDMA_DEST:
 943                         ret = rds_cmsg_rdma_dest(rs, rm, cmsg);
 944                         break;
 945
 946                 case RDS_CMSG_RDMA_MAP:
 947                         ret = rds_cmsg_rdma_map(rs, rm, cmsg);
 948                         if (!ret)
 949                                 *allocated_mr = 1;
 950                         else if (ret == -ENODEV)
 951                                 /* Accommodate the get_mr() case which can fail
 952                                  * if connection isn't established yet.
 953                                  */
 954                                 ret = -EAGAIN;
 955                         break;
 956                 case RDS_CMSG_ATOMIC_CSWP:
 957                 case RDS_CMSG_ATOMIC_FADD:
 958                 case RDS_CMSG_MASKED_ATOMIC_CSWP:
 959                 case RDS_CMSG_MASKED_ATOMIC_FADD:
 960                         ret = rds_cmsg_atomic(rs, rm, cmsg);
 961                         break;
 962
 963                 default:
 964                         return -EINVAL;
 965                 }
 966
 967                 if (ret)
 968                         break;
 969         }
 970
 971         return ret;
 972 }
 973
 974 static int rds_send_mprds_hash(struct rds_sock *rs, struct rds_connection *conn)
 975 {
 976         int hash;
 977
 978         if (conn->c_npaths == 0)
 979                 hash = RDS_MPATH_HASH(rs, RDS_MPATH_WORKERS);
 980         else
 981                 hash = RDS_MPATH_HASH(rs, conn->c_npaths);
 982         if (conn->c_npaths == 0 && hash != 0) {
 983                 rds_send_ping(conn, 0);
 984
 985                 if (conn->c_npaths == 0) {
 986                         wait_event_interruptible(conn->c_hs_waitq,
 987                                                  (conn->c_npaths != 0));
 988                 }
 989                 if (conn->c_npaths == 1)
 990                         hash = 0;
 991         }
 992         return hash;
 993 }
 994
 995 static int rds_rdma_bytes(struct msghdr *msg, size_t *rdma_bytes)
 996 {
 997         struct rds_rdma_args *args;
 998         struct cmsghdr *cmsg;
 999
1000         for_each_cmsghdr(cmsg, msg) {
1001                 if (!CMSG_OK(msg, cmsg))
1002                         return -EINVAL;
1003
1004                 if (cmsg->cmsg_level != SOL_RDS)
1005                         continue;
1006
1007                 if (cmsg->cmsg_type == RDS_CMSG_RDMA_ARGS) {
1008                         args = CMSG_DATA(cmsg);
1009                         *rdma_bytes += args->remote_vec.bytes;
1010                 }
1011         }
1012         return 0;
1013 }
1014
1015 int rds_sendmsg(struct socket *sock, struct msghdr *msg, size_t payload_len)
1016 {
1017         struct sock *sk = sock->sk;
1018         struct rds_sock *rs = rds_sk_to_rs(sk);
1019         DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1020         __be32 daddr;
1021         __be16 dport;
1022         struct rds_message *rm = NULL;
1023         struct rds_connection *conn;
1024         int ret = 0;
1025         int queued = 0, allocated_mr = 0;
1026         int nonblock = msg->msg_flags & MSG_DONTWAIT;
1027         long timeo = sock_sndtimeo(sk, nonblock);
1028         struct rds_conn_path *cpath;
1029         size_t total_payload_len = payload_len, rdma_payload_len = 0;
1030
1031         /* Mirror Linux UDP mirror of BSD error message compatibility */
1032         /* XXX: Perhaps MSG_MORE someday */
1033         if (msg->msg_flags & ~(MSG_DONTWAIT | MSG_CMSG_COMPAT)) {
1034                 ret = -EOPNOTSUPP;
1035                 goto out;
1036         }
1037
1038         if (msg->msg_namelen) {
1039                 /* XXX fail non-unicast destination IPs? */
1040                 if (msg->msg_namelen < sizeof(*usin) || usin->sin_family != AF_INET) {
1041                         ret = -EINVAL;
1042                         goto out;
1043                 }
1044                 daddr = usin->sin_addr.s_addr;
1045                 dport = usin->sin_port;
1046         } else {
1047                 /* We only care about consistency with ->connect() */
1048                 lock_sock(sk);
1049                 daddr = rs->rs_conn_addr;
1050                 dport = rs->rs_conn_port;
1051                 release_sock(sk);
1052         }
1053
1054         lock_sock(sk);
1055         if (daddr == 0 || rs->rs_bound_addr == 0) {
1056                 release_sock(sk);
1057                 ret = -ENOTCONN; /* XXX not a great errno */
1058                 goto out;
1059         }
1060         release_sock(sk);
1061
1062         ret = rds_rdma_bytes(msg, &rdma_payload_len);
1063         if (ret)
1064                 goto out;
1065
1066         total_payload_len += rdma_payload_len;
1067         if (max_t(size_t, payload_len, rdma_payload_len) > RDS_MAX_MSG_SIZE) {
1068                 ret = -EMSGSIZE;
1069                 goto out;
1070         }
1071
1072         if (payload_len > rds_sk_sndbuf(rs)) {
1073                 ret = -EMSGSIZE;
1074                 goto out;
1075         }
1076
1077         /* size of rm including all sgs */
1078         ret = rds_rm_size(msg, payload_len);
1079         if (ret < 0)
1080                 goto out;
1081
1082         rm = rds_message_alloc(ret, GFP_KERNEL);
1083         if (!rm) {
1084                 ret = -ENOMEM;
1085                 goto out;
1086         }
1087
1088         /* Attach data to the rm */
1089         if (payload_len) {
1090                 rm->data.op_sg = rds_message_alloc_sgs(rm, ceil(payload_len, PAGE_SIZE));
1091                 if (!rm->data.op_sg) {
1092                         ret = -ENOMEM;
1093                         goto out;
1094                 }
1095                 ret = rds_message_copy_from_user(rm, &msg->msg_iter);
1096                 if (ret)
1097                         goto out;
1098         }
1099         rm->data.op_active = 1;
1100
1101         rm->m_daddr = daddr;
1102
1103         /* rds_conn_create has a spinlock that runs with IRQ off.
1104          * Caching the conn in the socket helps a lot. */
1105         if (rs->rs_conn && rs->rs_conn->c_faddr == daddr)
1106                 conn = rs->rs_conn;
1107         else {
1108                 conn = rds_conn_create_outgoing(sock_net(sock->sk),
1109                                                 rs->rs_bound_addr, daddr,
1110                                         rs->rs_transport,
1111                                         sock->sk->sk_allocation);
1112                 if (IS_ERR(conn)) {
1113                         ret = PTR_ERR(conn);
1114                         goto out;
1115                 }
1116                 rs->rs_conn = conn;
1117         }
1118
1119         /* Parse any control messages the user may have included. */
1120         ret = rds_cmsg_send(rs, rm, msg, &allocated_mr);
1121         if (ret) {
1122                 /* Trigger connection so that its ready for the next retry */
1123                 if (ret ==  -EAGAIN)
1124                         rds_conn_connect_if_down(conn);
1125                 goto out;
1126         }
1127
1128         if (rm->rdma.op_active && !conn->c_trans->xmit_rdma) {
1129                 printk_ratelimited(KERN_NOTICE "rdma_op %p conn xmit_rdma %p\n",
1130                                &rm->rdma, conn->c_trans->xmit_rdma);
1131                 ret = -EOPNOTSUPP;
1132                 goto out;
1133         }
1134
1135         if (rm->atomic.op_active && !conn->c_trans->xmit_atomic) {
1136                 printk_ratelimited(KERN_NOTICE "atomic_op %p conn xmit_atomic %p\n",
1137                                &rm->atomic, conn->c_trans->xmit_atomic);
1138                 ret = -EOPNOTSUPP;
1139                 goto out;
1140         }
1141
1142         if (conn->c_trans->t_mp_capable)
1143                 cpath = &conn->c_path[rds_send_mprds_hash(rs, conn)];
1144         else
1145                 cpath = &conn->c_path[0];
1146
1147         rds_conn_path_connect_if_down(cpath);
1148
1149         ret = rds_cong_wait(conn->c_fcong, dport, nonblock, rs);
1150         if (ret) {
1151                 rs->rs_seen_congestion = 1;
1152                 goto out;
1153         }
1154         while (!rds_send_queue_rm(rs, conn, cpath, rm, rs->rs_bound_port,
1155                                   dport, &queued)) {
1156                 rds_stats_inc(s_send_queue_full);
1157
1158                 if (nonblock) {
1159                         ret = -EAGAIN;
1160                         goto out;
1161                 }
1162
1163                 timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
1164                                         rds_send_queue_rm(rs, conn, cpath, rm,
1165                                                           rs->rs_bound_port,
1166                                                           dport,
1167                                                           &queued),
1168                                         timeo);
1169                 rdsdebug("sendmsg woke queued %d timeo %ld\n", queued, timeo);
1170                 if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
1171                         continue;
1172
1173                 ret = timeo;
1174                 if (ret == 0)
1175                         ret = -ETIMEDOUT;
1176                 goto out;
1177         }
1178
1179         /*
1180          * By now we've committed to the send.  We reuse rds_send_worker()
1181          * to retry sends in the rds thread if the transport asks us to.
1182          */
1183         rds_stats_inc(s_send_queued);
1184
1185         ret = rds_send_xmit(cpath);
1186         if (ret == -ENOMEM || ret == -EAGAIN)
1187                 queue_delayed_work(rds_wq, &cpath->cp_send_w, 1);
1188
1189         rds_message_put(rm);
1190         return payload_len;
1191
1192 out:
1193         /* If the user included a RDMA_MAP cmsg, we allocated a MR on the fly.
1194          * If the sendmsg goes through, we keep the MR. If it fails with EAGAIN
1195          * or in any other way, we need to destroy the MR again */
1196         if (allocated_mr)
1197                 rds_rdma_unuse(rs, rds_rdma_cookie_key(rm->m_rdma_cookie), 1);
1198
1199         if (rm)
1200                 rds_message_put(rm);
1201         return ret;
1202 }
1203
1204 /*
1205  * send out a probe. Can be shared by rds_send_ping,
1206  * rds_send_pong, rds_send_hb.
1207  * rds_send_hb should use h_flags
1208  *   RDS_FLAG_HB_PING|RDS_FLAG_ACK_REQUIRED
1209  * or
1210  *   RDS_FLAG_HB_PONG|RDS_FLAG_ACK_REQUIRED
1211  */
1212 static int
1213 rds_send_probe(struct rds_conn_path *cp, __be16 sport,
1214                __be16 dport, u8 h_flags)
1215 {
1216         struct rds_message *rm;
1217         unsigned long flags;
1218         int ret = 0;
1219
1220         rm = rds_message_alloc(0, GFP_ATOMIC);
1221         if (!rm) {
1222                 ret = -ENOMEM;
1223                 goto out;
1224         }
1225
1226         rm->m_daddr = cp->cp_conn->c_faddr;
1227         rm->data.op_active = 1;
1228
1229         rds_conn_path_connect_if_down(cp);
1230
1231         ret = rds_cong_wait(cp->cp_conn->c_fcong, dport, 1, NULL);
1232         if (ret)
1233                 goto out;
1234
1235         spin_lock_irqsave(&cp->cp_lock, flags);
1236         list_add_tail(&rm->m_conn_item, &cp->cp_send_queue);
1237         set_bit(RDS_MSG_ON_CONN, &rm->m_flags);
1238         rds_message_addref(rm);
1239         rm->m_inc.i_conn = cp->cp_conn;
1240         rm->m_inc.i_conn_path = cp;
1241
1242         rds_message_populate_header(&rm->m_inc.i_hdr, sport, dport,
1243                                     cp->cp_next_tx_seq);
1244         rm->m_inc.i_hdr.h_flags |= h_flags;
1245         cp->cp_next_tx_seq++;
1246
1247         if (RDS_HS_PROBE(be16_to_cpu(sport), be16_to_cpu(dport)) &&
1248             cp->cp_conn->c_trans->t_mp_capable) {
1249                 u16 npaths = cpu_to_be16(RDS_MPATH_WORKERS);
1250                 u32 my_gen_num = cpu_to_be32(cp->cp_conn->c_my_gen_num);
1251
1252                 rds_message_add_extension(&rm->m_inc.i_hdr,
1253                                           RDS_EXTHDR_NPATHS, &npaths,
1254                                           sizeof(npaths));
1255                 rds_message_add_extension(&rm->m_inc.i_hdr,
1256                                           RDS_EXTHDR_GEN_NUM,
1257                                           &my_gen_num,
1258                                           sizeof(u32));
1259         }
1260         spin_unlock_irqrestore(&cp->cp_lock, flags);
1261
1262         rds_stats_inc(s_send_queued);
1263         rds_stats_inc(s_send_pong);
1264
1265         /* schedule the send work on rds_wq */
1266         queue_delayed_work(rds_wq, &cp->cp_send_w, 1);
1267
1268         rds_message_put(rm);
1269         return 0;
1270
1271 out:
1272         if (rm)
1273                 rds_message_put(rm);
1274         return ret;
1275 }
1276
1277 int
1278 rds_send_pong(struct rds_conn_path *cp, __be16 dport)
1279 {
1280         return rds_send_probe(cp, 0, dport, 0);
1281 }
1282
1283 void
1284 rds_send_ping(struct rds_connection *conn, int cp_index)
1285 {
1286         unsigned long flags;
1287         struct rds_conn_path *cp = &conn->c_path[cp_index];
1288
1289         spin_lock_irqsave(&cp->cp_lock, flags);
1290         if (conn->c_ping_triggered) {
1291                 spin_unlock_irqrestore(&cp->cp_lock, flags);
1292                 return;
1293         }
1294         conn->c_ping_triggered = 1;
1295         spin_unlock_irqrestore(&cp->cp_lock, flags);
1296         rds_send_probe(cp, cpu_to_be16(RDS_FLAG_PROBE_PORT), 0, 0);
1297 }
1298 EXPORT_SYMBOL_GPL(rds_send_ping);