kernel/drivers/staging/lustre/lustre/ptlrpc/niobuf.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_RPC
  38 #include "../include/obd_support.h"
  39 #include "../include/lustre_net.h"
  40 #include "../include/lustre_lib.h"
  41 #include "../include/obd.h"
  42 #include "../include/obd_class.h"
  43 #include "ptlrpc_internal.h"
  44
  45 /**
  46  * Helper function. Sends \a len bytes from \a base at offset \a offset
  47  * over \a conn connection to portal \a portal.
  48  * Returns 0 on success or error code.
  49  */
  50 static int ptl_send_buf(lnet_handle_md_t *mdh, void *base, int len,
  51                         lnet_ack_req_t ack, struct ptlrpc_cb_id *cbid,
  52                         struct ptlrpc_connection *conn, int portal, __u64 xid,
  53                         unsigned int offset)
  54 {
  55         int           rc;
  56         lnet_md_t        md;
  57
  58         LASSERT(portal != 0);
  59         LASSERT(conn != NULL);
  60         CDEBUG(D_INFO, "conn=%p id %s\n", conn, libcfs_id2str(conn->c_peer));
  61         md.start     = base;
  62         md.length    = len;
  63         md.threshold = (ack == LNET_ACK_REQ) ? 2 : 1;
  64         md.options   = PTLRPC_MD_OPTIONS;
  65         md.user_ptr  = cbid;
  66         md.eq_handle = ptlrpc_eq_h;
  67
  68         if (unlikely(ack == LNET_ACK_REQ &&
  69                      OBD_FAIL_CHECK_ORSET(OBD_FAIL_PTLRPC_ACK,
  70                                           OBD_FAIL_ONCE))) {
  71                 /* don't ask for the ack to simulate failing client */
  72                 ack = LNET_NOACK_REQ;
  73         }
  74
  75         rc = LNetMDBind(md, LNET_UNLINK, mdh);
  76         if (unlikely(rc != 0)) {
  77                 CERROR("LNetMDBind failed: %d\n", rc);
  78                 LASSERT(rc == -ENOMEM);
  79                 return -ENOMEM;
  80         }
  81
  82         CDEBUG(D_NET, "Sending %d bytes to portal %d, xid %lld, offset %u\n",
  83                len, portal, xid, offset);
  84
  85         rc = LNetPut(conn->c_self, *mdh, ack,
  86                      conn->c_peer, portal, xid, offset, 0);
  87         if (unlikely(rc != 0)) {
  88                 int rc2;
  89                 /* We're going to get an UNLINK event when I unlink below,
  90                  * which will complete just like any other failed send, so
  91                  * I fall through and return success here! */
  92                 CERROR("LNetPut(%s, %d, %lld) failed: %d\n",
  93                        libcfs_id2str(conn->c_peer), portal, xid, rc);
  94                 rc2 = LNetMDUnlink(*mdh);
  95                 LASSERTF(rc2 == 0, "rc2 = %d\n", rc2);
  96         }
  97
  98         return 0;
  99 }
 100
 101 static void mdunlink_iterate_helper(lnet_handle_md_t *bd_mds, int count)
 102 {
 103         int i;
 104
 105         for (i = 0; i < count; i++)
 106                 LNetMDUnlink(bd_mds[i]);
 107 }
 108
 109
 110 /**
 111  * Register bulk at the sender for later transfer.
 112  * Returns 0 on success or error code.
 113  */
 114 int ptlrpc_register_bulk(struct ptlrpc_request *req)
 115 {
 116         struct ptlrpc_bulk_desc *desc = req->rq_bulk;
 117         lnet_process_id_t peer;
 118         int rc = 0;
 119         int rc2;
 120         int posted_md;
 121         int total_md;
 122         __u64 xid;
 123         lnet_handle_me_t  me_h;
 124         lnet_md_t        md;
 125
 126         if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_BULK_GET_NET))
 127                 return 0;
 128
 129         /* NB no locking required until desc is on the network */
 130         LASSERT(desc->bd_nob > 0);
 131         LASSERT(desc->bd_md_count == 0);
 132         LASSERT(desc->bd_md_max_brw <= PTLRPC_BULK_OPS_COUNT);
 133         LASSERT(desc->bd_iov_count <= PTLRPC_MAX_BRW_PAGES);
 134         LASSERT(desc->bd_req != NULL);
 135         LASSERT(desc->bd_type == BULK_PUT_SINK ||
 136                 desc->bd_type == BULK_GET_SOURCE);
 137
 138         /* cleanup the state of the bulk for it will be reused */
 139         if (req->rq_resend || req->rq_send_state == LUSTRE_IMP_REPLAY)
 140                 desc->bd_nob_transferred = 0;
 141         else
 142                 LASSERT(desc->bd_nob_transferred == 0);
 143
 144         desc->bd_failure = 0;
 145
 146         peer = desc->bd_import->imp_connection->c_peer;
 147
 148         LASSERT(desc->bd_cbid.cbid_fn == client_bulk_callback);
 149         LASSERT(desc->bd_cbid.cbid_arg == desc);
 150
 151         /* An XID is only used for a single request from the client.
 152          * For retried bulk transfers, a new XID will be allocated in
 153          * in ptlrpc_check_set() if it needs to be resent, so it is not
 154          * using the same RDMA match bits after an error.
 155          *
 156          * For multi-bulk RPCs, rq_xid is the last XID needed for bulks. The
 157          * first bulk XID is power-of-two aligned before rq_xid. LU-1431 */
 158         xid = req->rq_xid & ~((__u64)desc->bd_md_max_brw - 1);
 159         LASSERTF(!(desc->bd_registered &&
 160                    req->rq_send_state != LUSTRE_IMP_REPLAY) ||
 161                  xid != desc->bd_last_xid,
 162                  "registered: %d  rq_xid: %llu bd_last_xid: %llu\n",
 163                  desc->bd_registered, xid, desc->bd_last_xid);
 164
 165         total_md = (desc->bd_iov_count + LNET_MAX_IOV - 1) / LNET_MAX_IOV;
 166         desc->bd_registered = 1;
 167         desc->bd_last_xid = xid;
 168         desc->bd_md_count = total_md;
 169         md.user_ptr = &desc->bd_cbid;
 170         md.eq_handle = ptlrpc_eq_h;
 171         md.threshold = 1;                      /* PUT or GET */
 172
 173         for (posted_md = 0; posted_md < total_md; posted_md++, xid++) {
 174                 md.options = PTLRPC_MD_OPTIONS |
 175                              ((desc->bd_type == BULK_GET_SOURCE) ?
 176                               LNET_MD_OP_GET : LNET_MD_OP_PUT);
 177                 ptlrpc_fill_bulk_md(&md, desc, posted_md);
 178
 179                 rc = LNetMEAttach(desc->bd_portal, peer, xid, 0,
 180                                   LNET_UNLINK, LNET_INS_AFTER, &me_h);
 181                 if (rc != 0) {
 182                         CERROR("%s: LNetMEAttach failed x%llu/%d: rc = %d\n",
 183                                desc->bd_import->imp_obd->obd_name, xid,
 184                                posted_md, rc);
 185                         break;
 186                 }
 187
 188                 /* About to let the network at it... */
 189                 rc = LNetMDAttach(me_h, md, LNET_UNLINK,
 190                                   &desc->bd_mds[posted_md]);
 191                 if (rc != 0) {
 192                         CERROR("%s: LNetMDAttach failed x%llu/%d: rc = %d\n",
 193                                desc->bd_import->imp_obd->obd_name, xid,
 194                                posted_md, rc);
 195                         rc2 = LNetMEUnlink(me_h);
 196                         LASSERT(rc2 == 0);
 197                         break;
 198                 }
 199         }
 200
 201         if (rc != 0) {
 202                 LASSERT(rc == -ENOMEM);
 203                 spin_lock(&desc->bd_lock);
 204                 desc->bd_md_count -= total_md - posted_md;
 205                 spin_unlock(&desc->bd_lock);
 206                 LASSERT(desc->bd_md_count >= 0);
 207                 mdunlink_iterate_helper(desc->bd_mds, desc->bd_md_max_brw);
 208                 req->rq_status = -ENOMEM;
 209                 return -ENOMEM;
 210         }
 211
 212         /* Set rq_xid to matchbits of the final bulk so that server can
 213          * infer the number of bulks that were prepared */
 214         req->rq_xid = --xid;
 215         LASSERTF(desc->bd_last_xid == (req->rq_xid & PTLRPC_BULK_OPS_MASK),
 216                  "bd_last_xid = x%llu, rq_xid = x%llu\n",
 217                  desc->bd_last_xid, req->rq_xid);
 218
 219         spin_lock(&desc->bd_lock);
 220         /* Holler if peer manages to touch buffers before he knows the xid */
 221         if (desc->bd_md_count != total_md)
 222                 CWARN("%s: Peer %s touched %d buffers while I registered\n",
 223                       desc->bd_import->imp_obd->obd_name, libcfs_id2str(peer),
 224                       total_md - desc->bd_md_count);
 225         spin_unlock(&desc->bd_lock);
 226
 227         CDEBUG(D_NET, "Setup %u bulk %s buffers: %u pages %u bytes, xid x%#llx-%#llx, portal %u\n",
 228                desc->bd_md_count,
 229                desc->bd_type == BULK_GET_SOURCE ? "get-source" : "put-sink",
 230                desc->bd_iov_count, desc->bd_nob,
 231                desc->bd_last_xid, req->rq_xid, desc->bd_portal);
 232
 233         return 0;
 234 }
 235 EXPORT_SYMBOL(ptlrpc_register_bulk);
 236
 237 /**
 238  * Disconnect a bulk desc from the network. Idempotent. Not
 239  * thread-safe (i.e. only interlocks with completion callback).
 240  * Returns 1 on success or 0 if network unregistration failed for whatever
 241  * reason.
 242  */
 243 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async)
 244 {
 245         struct ptlrpc_bulk_desc *desc = req->rq_bulk;
 246         wait_queue_head_t            *wq;
 247         struct l_wait_info       lwi;
 248         int                   rc;
 249
 250         LASSERT(!in_interrupt());     /* might sleep */
 251
 252         /* Let's setup deadline for reply unlink. */
 253         if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
 254             async && req->rq_bulk_deadline == 0)
 255                 req->rq_bulk_deadline = get_seconds() + LONG_UNLINK;
 256
 257         if (ptlrpc_client_bulk_active(req) == 0)        /* completed or */
 258                 return 1;                               /* never registered */
 259
 260         LASSERT(desc->bd_req == req);  /* bd_req NULL until registered */
 261
 262         /* the unlink ensures the callback happens ASAP and is the last
 263          * one.  If it fails, it must be because completion just happened,
 264          * but we must still l_wait_event() in this case to give liblustre
 265          * a chance to run client_bulk_callback() */
 266         mdunlink_iterate_helper(desc->bd_mds, desc->bd_md_max_brw);
 267
 268         if (ptlrpc_client_bulk_active(req) == 0)        /* completed or */
 269                 return 1;                               /* never registered */
 270
 271         /* Move to "Unregistering" phase as bulk was not unlinked yet. */
 272         ptlrpc_rqphase_move(req, RQ_PHASE_UNREGISTERING);
 273
 274         /* Do not wait for unlink to finish. */
 275         if (async)
 276                 return 0;
 277
 278         if (req->rq_set != NULL)
 279                 wq = &req->rq_set->set_waitq;
 280         else
 281                 wq = &req->rq_reply_waitq;
 282
 283         for (;;) {
 284                 /* Network access will complete in finite time but the HUGE
 285                  * timeout lets us CWARN for visibility of sluggish NALs */
 286                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(LONG_UNLINK),
 287                                            cfs_time_seconds(1), NULL, NULL);
 288                 rc = l_wait_event(*wq, !ptlrpc_client_bulk_active(req), &lwi);
 289                 if (rc == 0) {
 290                         ptlrpc_rqphase_move(req, req->rq_next_phase);
 291                         return 1;
 292                 }
 293
 294                 LASSERT(rc == -ETIMEDOUT);
 295                 DEBUG_REQ(D_WARNING, req, "Unexpectedly long timeout: desc %p",
 296                           desc);
 297         }
 298         return 0;
 299 }
 300 EXPORT_SYMBOL(ptlrpc_unregister_bulk);
 301
 302 static void ptlrpc_at_set_reply(struct ptlrpc_request *req, int flags)
 303 {
 304         struct ptlrpc_service_part      *svcpt = req->rq_rqbd->rqbd_svcpt;
 305         struct ptlrpc_service           *svc = svcpt->scp_service;
 306         int service_time = max_t(int, get_seconds() -
 307                                  req->rq_arrival_time.tv_sec, 1);
 308
 309         if (!(flags & PTLRPC_REPLY_EARLY) &&
 310             (req->rq_type != PTL_RPC_MSG_ERR) &&
 311             (req->rq_reqmsg != NULL) &&
 312             !(lustre_msg_get_flags(req->rq_reqmsg) &
 313               (MSG_RESENT | MSG_REPLAY |
 314                MSG_REQ_REPLAY_DONE | MSG_LOCK_REPLAY_DONE))) {
 315                 /* early replies, errors and recovery requests don't count
 316                  * toward our service time estimate */
 317                 int oldse = at_measured(&svcpt->scp_at_estimate, service_time);
 318
 319                 if (oldse != 0) {
 320                         DEBUG_REQ(D_ADAPTTO, req,
 321                                   "svc %s changed estimate from %d to %d",
 322                                   svc->srv_name, oldse,
 323                                   at_get(&svcpt->scp_at_estimate));
 324                 }
 325         }
 326         /* Report actual service time for client latency calc */
 327         lustre_msg_set_service_time(req->rq_repmsg, service_time);
 328         /* Report service time estimate for future client reqs, but report 0
 329          * (to be ignored by client) if it's a error reply during recovery.
 330          * (bz15815) */
 331         if (req->rq_type == PTL_RPC_MSG_ERR &&
 332             (req->rq_export == NULL || req->rq_export->exp_obd->obd_recovering))
 333                 lustre_msg_set_timeout(req->rq_repmsg, 0);
 334         else
 335                 lustre_msg_set_timeout(req->rq_repmsg,
 336                                        at_get(&svcpt->scp_at_estimate));
 337
 338         if (req->rq_reqmsg &&
 339             !(lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
 340                 CDEBUG(D_ADAPTTO, "No early reply support: flags=%#x req_flags=%#x magic=%d:%x/%x len=%d\n",
 341                        flags, lustre_msg_get_flags(req->rq_reqmsg),
 342                        lustre_msg_is_v1(req->rq_reqmsg),
 343                        lustre_msg_get_magic(req->rq_reqmsg),
 344                        lustre_msg_get_magic(req->rq_repmsg), req->rq_replen);
 345         }
 346 }
 347
 348 /**
 349  * Send request reply from request \a req reply buffer.
 350  * \a flags defines reply types
 351  * Returns 0 on success or error code
 352  */
 353 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags)
 354 {
 355         struct ptlrpc_reply_state *rs = req->rq_reply_state;
 356         struct ptlrpc_connection  *conn;
 357         int                     rc;
 358
 359         /* We must already have a reply buffer (only ptlrpc_error() may be
 360          * called without one). The reply generated by sptlrpc layer (e.g.
 361          * error notify, etc.) might have NULL rq->reqmsg; Otherwise we must
 362          * have a request buffer which is either the actual (swabbed) incoming
 363          * request, or a saved copy if this is a req saved in
 364          * target_queue_final_reply().
 365          */
 366         LASSERT(req->rq_no_reply == 0);
 367         LASSERT(req->rq_reqbuf != NULL);
 368         LASSERT(rs != NULL);
 369         LASSERT((flags & PTLRPC_REPLY_MAYBE_DIFFICULT) || !rs->rs_difficult);
 370         LASSERT(req->rq_repmsg != NULL);
 371         LASSERT(req->rq_repmsg == rs->rs_msg);
 372         LASSERT(rs->rs_cb_id.cbid_fn == reply_out_callback);
 373         LASSERT(rs->rs_cb_id.cbid_arg == rs);
 374
 375         /* There may be no rq_export during failover */
 376
 377         if (unlikely(req->rq_export && req->rq_export->exp_obd &&
 378                      req->rq_export->exp_obd->obd_fail)) {
 379                 /* Failed obd's only send ENODEV */
 380                 req->rq_type = PTL_RPC_MSG_ERR;
 381                 req->rq_status = -ENODEV;
 382                 CDEBUG(D_HA, "sending ENODEV from failed obd %d\n",
 383                        req->rq_export->exp_obd->obd_minor);
 384         }
 385
 386         /* In order to keep interoperability with the client (< 2.3) which
 387          * doesn't have pb_jobid in ptlrpc_body, We have to shrink the
 388          * ptlrpc_body in reply buffer to ptlrpc_body_v2, otherwise, the
 389          * reply buffer on client will be overflow.
 390          *
 391          * XXX Remove this whenever we drop the interoperability with
 392          * such client.
 393          */
 394         req->rq_replen = lustre_shrink_msg(req->rq_repmsg, 0,
 395                                            sizeof(struct ptlrpc_body_v2), 1);
 396
 397         if (req->rq_type != PTL_RPC_MSG_ERR)
 398                 req->rq_type = PTL_RPC_MSG_REPLY;
 399
 400         lustre_msg_set_type(req->rq_repmsg, req->rq_type);
 401         lustre_msg_set_status(req->rq_repmsg,
 402                               ptlrpc_status_hton(req->rq_status));
 403         lustre_msg_set_opc(req->rq_repmsg,
 404                 req->rq_reqmsg ? lustre_msg_get_opc(req->rq_reqmsg) : 0);
 405
 406         target_pack_pool_reply(req);
 407
 408         ptlrpc_at_set_reply(req, flags);
 409
 410         if (req->rq_export == NULL || req->rq_export->exp_connection == NULL)
 411                 conn = ptlrpc_connection_get(req->rq_peer, req->rq_self, NULL);
 412         else
 413                 conn = ptlrpc_connection_addref(req->rq_export->exp_connection);
 414
 415         if (unlikely(conn == NULL)) {
 416                 CERROR("not replying on NULL connection\n"); /* bug 9635 */
 417                 return -ENOTCONN;
 418         }
 419         ptlrpc_rs_addref(rs);              /* +1 ref for the network */
 420
 421         rc = sptlrpc_svc_wrap_reply(req);
 422         if (unlikely(rc))
 423                 goto out;
 424
 425         req->rq_sent = get_seconds();
 426
 427         rc = ptl_send_buf(&rs->rs_md_h, rs->rs_repbuf, rs->rs_repdata_len,
 428                           (rs->rs_difficult && !rs->rs_no_ack) ?
 429                           LNET_ACK_REQ : LNET_NOACK_REQ,
 430                           &rs->rs_cb_id, conn,
 431                           ptlrpc_req2svc(req)->srv_rep_portal,
 432                           req->rq_xid, req->rq_reply_off);
 433 out:
 434         if (unlikely(rc != 0))
 435                 ptlrpc_req_drop_rs(req);
 436         ptlrpc_connection_put(conn);
 437         return rc;
 438 }
 439 EXPORT_SYMBOL(ptlrpc_send_reply);
 440
 441 int ptlrpc_reply(struct ptlrpc_request *req)
 442 {
 443         if (req->rq_no_reply)
 444                 return 0;
 445         return ptlrpc_send_reply(req, 0);
 446 }
 447 EXPORT_SYMBOL(ptlrpc_reply);
 448
 449 /**
 450  * For request \a req send an error reply back. Create empty
 451  * reply buffers if necessary.
 452  */
 453 int ptlrpc_send_error(struct ptlrpc_request *req, int may_be_difficult)
 454 {
 455         int rc;
 456
 457         if (req->rq_no_reply)
 458                 return 0;
 459
 460         if (!req->rq_repmsg) {
 461                 rc = lustre_pack_reply(req, 1, NULL, NULL);
 462                 if (rc)
 463                         return rc;
 464         }
 465
 466         if (req->rq_status != -ENOSPC && req->rq_status != -EACCES &&
 467             req->rq_status != -EPERM && req->rq_status != -ENOENT &&
 468             req->rq_status != -EINPROGRESS && req->rq_status != -EDQUOT)
 469                 req->rq_type = PTL_RPC_MSG_ERR;
 470
 471         rc = ptlrpc_send_reply(req, may_be_difficult);
 472         return rc;
 473 }
 474 EXPORT_SYMBOL(ptlrpc_send_error);
 475
 476 int ptlrpc_error(struct ptlrpc_request *req)
 477 {
 478         return ptlrpc_send_error(req, 0);
 479 }
 480 EXPORT_SYMBOL(ptlrpc_error);
 481
 482 /**
 483  * Send request \a request.
 484  * if \a noreply is set, don't expect any reply back and don't set up
 485  * reply buffers.
 486  * Returns 0 on success or error code.
 487  */
 488 int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
 489 {
 490         int rc;
 491         int rc2;
 492         int mpflag = 0;
 493         struct ptlrpc_connection *connection;
 494         lnet_handle_me_t  reply_me_h;
 495         lnet_md_t        reply_md;
 496         struct obd_device *obd = request->rq_import->imp_obd;
 497
 498         if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_RPC))
 499                 return 0;
 500
 501         LASSERT(request->rq_type == PTL_RPC_MSG_REQUEST);
 502         LASSERT(request->rq_wait_ctx == 0);
 503
 504         /* If this is a re-transmit, we're required to have disengaged
 505          * cleanly from the previous attempt */
 506         LASSERT(!request->rq_receiving_reply);
 507         LASSERT(!((lustre_msg_get_flags(request->rq_reqmsg) & MSG_REPLAY) &&
 508                 (request->rq_import->imp_state == LUSTRE_IMP_FULL)));
 509
 510         if (unlikely(obd != NULL && obd->obd_fail)) {
 511                 CDEBUG(D_HA, "muting rpc for failed imp obd %s\n",
 512                         obd->obd_name);
 513                 /* this prevents us from waiting in ptlrpc_queue_wait */
 514                 spin_lock(&request->rq_lock);
 515                 request->rq_err = 1;
 516                 spin_unlock(&request->rq_lock);
 517                 request->rq_status = -ENODEV;
 518                 return -ENODEV;
 519         }
 520
 521         connection = request->rq_import->imp_connection;
 522
 523         lustre_msg_set_handle(request->rq_reqmsg,
 524                               &request->rq_import->imp_remote_handle);
 525         lustre_msg_set_type(request->rq_reqmsg, PTL_RPC_MSG_REQUEST);
 526         lustre_msg_set_conn_cnt(request->rq_reqmsg,
 527                                 request->rq_import->imp_conn_cnt);
 528         lustre_msghdr_set_flags(request->rq_reqmsg,
 529                                 request->rq_import->imp_msghdr_flags);
 530
 531         if (request->rq_resend)
 532                 lustre_msg_add_flags(request->rq_reqmsg, MSG_RESENT);
 533
 534         if (request->rq_memalloc)
 535                 mpflag = cfs_memory_pressure_get_and_set();
 536
 537         rc = sptlrpc_cli_wrap_request(request);
 538         if (rc)
 539                 goto out;
 540
 541         /* bulk register should be done after wrap_request() */
 542         if (request->rq_bulk != NULL) {
 543                 rc = ptlrpc_register_bulk(request);
 544                 if (rc != 0)
 545                         goto out;
 546         }
 547
 548         if (!noreply) {
 549                 LASSERT(request->rq_replen != 0);
 550                 if (request->rq_repbuf == NULL) {
 551                         LASSERT(request->rq_repdata == NULL);
 552                         LASSERT(request->rq_repmsg == NULL);
 553                         rc = sptlrpc_cli_alloc_repbuf(request,
 554                                                       request->rq_replen);
 555                         if (rc) {
 556                                 /* this prevents us from looping in
 557                                  * ptlrpc_queue_wait */
 558                                 spin_lock(&request->rq_lock);
 559                                 request->rq_err = 1;
 560                                 spin_unlock(&request->rq_lock);
 561                                 request->rq_status = rc;
 562                                 goto cleanup_bulk;
 563                         }
 564                 } else {
 565                         request->rq_repdata = NULL;
 566                         request->rq_repmsg = NULL;
 567                 }
 568
 569                 rc = LNetMEAttach(request->rq_reply_portal,/*XXX FIXME bug 249*/
 570                                   connection->c_peer, request->rq_xid, 0,
 571                                   LNET_UNLINK, LNET_INS_AFTER, &reply_me_h);
 572                 if (rc != 0) {
 573                         CERROR("LNetMEAttach failed: %d\n", rc);
 574                         LASSERT(rc == -ENOMEM);
 575                         rc = -ENOMEM;
 576                         goto cleanup_bulk;
 577                 }
 578         }
 579
 580         spin_lock(&request->rq_lock);
 581         /* If the MD attach succeeds, there _will_ be a reply_in callback */
 582         request->rq_receiving_reply = !noreply;
 583         request->rq_req_unlink = 1;
 584         /* We are responsible for unlinking the reply buffer */
 585         request->rq_reply_unlink = !noreply;
 586         /* Clear any flags that may be present from previous sends. */
 587         request->rq_replied = 0;
 588         request->rq_err = 0;
 589         request->rq_timedout = 0;
 590         request->rq_net_err = 0;
 591         request->rq_resend = 0;
 592         request->rq_restart = 0;
 593         request->rq_reply_truncate = 0;
 594         spin_unlock(&request->rq_lock);
 595
 596         if (!noreply) {
 597                 reply_md.start     = request->rq_repbuf;
 598                 reply_md.length    = request->rq_repbuf_len;
 599                 /* Allow multiple early replies */
 600                 reply_md.threshold = LNET_MD_THRESH_INF;
 601                 /* Manage remote for early replies */
 602                 reply_md.options   = PTLRPC_MD_OPTIONS | LNET_MD_OP_PUT |
 603                         LNET_MD_MANAGE_REMOTE |
 604                         LNET_MD_TRUNCATE; /* allow to make EOVERFLOW error */;
 605                 reply_md.user_ptr  = &request->rq_reply_cbid;
 606                 reply_md.eq_handle = ptlrpc_eq_h;
 607
 608                 /* We must see the unlink callback to unset rq_reply_unlink,
 609                    so we can't auto-unlink */
 610                 rc = LNetMDAttach(reply_me_h, reply_md, LNET_RETAIN,
 611                                   &request->rq_reply_md_h);
 612                 if (rc != 0) {
 613                         CERROR("LNetMDAttach failed: %d\n", rc);
 614                         LASSERT(rc == -ENOMEM);
 615                         spin_lock(&request->rq_lock);
 616                         /* ...but the MD attach didn't succeed... */
 617                         request->rq_receiving_reply = 0;
 618                         spin_unlock(&request->rq_lock);
 619                         rc = -ENOMEM;
 620                         goto cleanup_me;
 621                 }
 622
 623                 CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid %llu, portal %u\n",
 624                        request->rq_repbuf_len, request->rq_xid,
 625                        request->rq_reply_portal);
 626         }
 627
 628         /* add references on request for request_out_callback */
 629         ptlrpc_request_addref(request);
 630         if (obd != NULL && obd->obd_svc_stats != NULL)
 631                 lprocfs_counter_add(obd->obd_svc_stats, PTLRPC_REQACTIVE_CNTR,
 632                         atomic_read(&request->rq_import->imp_inflight));
 633
 634         OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_SEND, request->rq_timeout + 5);
 635
 636         do_gettimeofday(&request->rq_arrival_time);
 637         request->rq_sent = get_seconds();
 638         /* We give the server rq_timeout secs to process the req, and
 639            add the network latency for our local timeout. */
 640         request->rq_deadline = request->rq_sent + request->rq_timeout +
 641                 ptlrpc_at_get_net_latency(request);
 642
 643         ptlrpc_pinger_sending_on_import(request->rq_import);
 644
 645         DEBUG_REQ(D_INFO, request, "send flg=%x",
 646                   lustre_msg_get_flags(request->rq_reqmsg));
 647         rc = ptl_send_buf(&request->rq_req_md_h,
 648                           request->rq_reqbuf, request->rq_reqdata_len,
 649                           LNET_NOACK_REQ, &request->rq_req_cbid,
 650                           connection,
 651                           request->rq_request_portal,
 652                           request->rq_xid, 0);
 653         if (rc == 0)
 654                 goto out;
 655
 656         ptlrpc_req_finished(request);
 657         if (noreply)
 658                 goto out;
 659
 660  cleanup_me:
 661         /* MEUnlink is safe; the PUT didn't even get off the ground, and
 662          * nobody apart from the PUT's target has the right nid+XID to
 663          * access the reply buffer. */
 664         rc2 = LNetMEUnlink(reply_me_h);
 665         LASSERT(rc2 == 0);
 666         /* UNLINKED callback called synchronously */
 667         LASSERT(!request->rq_receiving_reply);
 668
 669  cleanup_bulk:
 670         /* We do sync unlink here as there was no real transfer here so
 671          * the chance to have long unlink to sluggish net is smaller here. */
 672         ptlrpc_unregister_bulk(request, 0);
 673  out:
 674         if (request->rq_memalloc)
 675                 cfs_memory_pressure_restore(mpflag);
 676         return rc;
 677 }
 678 EXPORT_SYMBOL(ptl_send_rpc);
 679
 680 /**
 681  * Register request buffer descriptor for request receiving.
 682  */
 683 int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd)
 684 {
 685         struct ptlrpc_service     *service = rqbd->rqbd_svcpt->scp_service;
 686         static lnet_process_id_t  match_id = {LNET_NID_ANY, LNET_PID_ANY};
 687         int                       rc;
 688         lnet_md_t                md;
 689         lnet_handle_me_t          me_h;
 690
 691         CDEBUG(D_NET, "LNetMEAttach: portal %d\n",
 692                service->srv_req_portal);
 693
 694         if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_RQBD))
 695                 return -ENOMEM;
 696
 697         /* NB: CPT affinity service should use new LNet flag LNET_INS_LOCAL,
 698          * which means buffer can only be attached on local CPT, and LND
 699          * threads can find it by grabbing a local lock */
 700         rc = LNetMEAttach(service->srv_req_portal,
 701                           match_id, 0, ~0, LNET_UNLINK,
 702                           rqbd->rqbd_svcpt->scp_cpt >= 0 ?
 703                           LNET_INS_LOCAL : LNET_INS_AFTER, &me_h);
 704         if (rc != 0) {
 705                 CERROR("LNetMEAttach failed: %d\n", rc);
 706                 return -ENOMEM;
 707         }
 708
 709         LASSERT(rqbd->rqbd_refcount == 0);
 710         rqbd->rqbd_refcount = 1;
 711
 712         md.start     = rqbd->rqbd_buffer;
 713         md.length    = service->srv_buf_size;
 714         md.max_size  = service->srv_max_req_size;
 715         md.threshold = LNET_MD_THRESH_INF;
 716         md.options   = PTLRPC_MD_OPTIONS | LNET_MD_OP_PUT | LNET_MD_MAX_SIZE;
 717         md.user_ptr  = &rqbd->rqbd_cbid;
 718         md.eq_handle = ptlrpc_eq_h;
 719
 720         rc = LNetMDAttach(me_h, md, LNET_UNLINK, &rqbd->rqbd_md_h);
 721         if (rc == 0)
 722                 return 0;
 723
 724         CERROR("LNetMDAttach failed: %d;\n", rc);
 725         LASSERT(rc == -ENOMEM);
 726         rc = LNetMEUnlink(me_h);
 727         LASSERT(rc == 0);
 728         rqbd->rqbd_refcount = 0;
 729
 730         return -ENOMEM;
 731 }