lustre/ptlrpc/client.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 2002 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  *
  21  */
  22
  23 #define DEBUG_SUBSYSTEM S_RPC
  24
  25 #include <linux/obd_support.h>
  26 #include <linux/obd_class.h>
  27 #include <linux/lustre_lib.h>
  28 #include <linux/lustre_ha.h>
  29 #include <linux/lustre_import.h>
  30
  31 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
  32                         struct ptlrpc_client *cl)
  33 {
  34         cl->cli_request_portal = req_portal;
  35         cl->cli_reply_portal   = rep_portal;
  36         cl->cli_name           = name;
  37 }
  38
  39 __u8 *ptlrpc_req_to_uuid(struct ptlrpc_request *req)
  40 {
  41         return req->rq_connection->c_remote_uuid;
  42 }
  43
  44 struct ptlrpc_connection *ptlrpc_uuid_to_connection(obd_uuid_t uuid)
  45 {
  46         struct ptlrpc_connection *c;
  47         struct lustre_peer peer;
  48         int err;
  49
  50         err = kportal_uuid_to_peer(uuid, &peer);
  51         if (err != 0) {
  52                 CERROR("cannot find peer %s!\n", uuid);
  53                 return NULL;
  54         }
  55
  56         c = ptlrpc_get_connection(&peer, uuid);
  57         if (c) {
  58                 memcpy(c->c_remote_uuid, uuid, sizeof(c->c_remote_uuid));
  59                 c->c_epoch++;
  60         }
  61
  62         CDEBUG(D_INFO, "%s -> %p\n", uuid, c);
  63
  64         return c;
  65 }
  66
  67 void ptlrpc_readdress_connection(struct ptlrpc_connection *conn,obd_uuid_t uuid)
  68 {
  69         struct lustre_peer peer;
  70         int err;
  71
  72         err = kportal_uuid_to_peer(uuid, &peer);
  73         if (err != 0) {
  74                 CERROR("cannot find peer %s!\n", uuid);
  75                 return;
  76         }
  77
  78         memcpy(&conn->c_peer, &peer, sizeof(peer));
  79         return;
  80 }
  81
  82 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk(struct ptlrpc_connection *conn)
  83 {
  84         struct ptlrpc_bulk_desc *desc;
  85
  86         OBD_ALLOC(desc, sizeof(*desc));
  87         if (desc != NULL) {
  88                 desc->bd_connection = ptlrpc_connection_addref(conn);
  89                 atomic_set(&desc->bd_refcount, 1);
  90                 init_waitqueue_head(&desc->bd_waitq);
  91                 INIT_LIST_HEAD(&desc->bd_page_list);
  92                 INIT_LIST_HEAD(&desc->bd_set_chain);
  93                 ptl_set_inv_handle(&desc->bd_md_h);
  94                 ptl_set_inv_handle(&desc->bd_me_h);
  95         }
  96
  97         return desc;
  98 }
  99
 100 int ptlrpc_bulk_error(struct ptlrpc_bulk_desc *desc)
 101 {
 102         int rc = 0;
 103         if (desc->bd_flags & PTL_RPC_FL_TIMEOUT) {
 104                 rc = (desc->bd_flags & PTL_RPC_FL_INTR ? -ERESTARTSYS :
 105                       -ETIMEDOUT);
 106         }
 107         return rc;
 108 }
 109
 110 struct ptlrpc_bulk_page *ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc)
 111 {
 112         struct ptlrpc_bulk_page *bulk;
 113
 114         OBD_ALLOC(bulk, sizeof(*bulk));
 115         if (bulk != NULL) {
 116                 bulk->bp_desc = desc;
 117                 list_add_tail(&bulk->bp_link, &desc->bd_page_list);
 118                 desc->bd_page_count++;
 119         }
 120         return bulk;
 121 }
 122
 123 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *desc)
 124 {
 125         struct list_head *tmp, *next;
 126         ENTRY;
 127         if (desc == NULL) {
 128                 EXIT;
 129                 return;
 130         }
 131
 132         LASSERT(list_empty(&desc->bd_set_chain));
 133
 134         list_for_each_safe(tmp, next, &desc->bd_page_list) {
 135                 struct ptlrpc_bulk_page *bulk;
 136                 bulk = list_entry(tmp, struct ptlrpc_bulk_page, bp_link);
 137                 ptlrpc_free_bulk_page(bulk);
 138         }
 139
 140         ptlrpc_put_connection(desc->bd_connection);
 141
 142         OBD_FREE(desc, sizeof(*desc));
 143         EXIT;
 144 }
 145
 146 void ptlrpc_free_bulk_page(struct ptlrpc_bulk_page *bulk)
 147 {
 148         ENTRY;
 149         if (bulk == NULL) {
 150                 EXIT;
 151                 return;
 152         }
 153
 154         list_del(&bulk->bp_link);
 155         bulk->bp_desc->bd_page_count--;
 156         OBD_FREE(bulk, sizeof(*bulk));
 157         EXIT;
 158 }
 159
 160 static int ll_sync_brw_timeout(void *data)
 161 {
 162         struct obd_brw_set *set = data;
 163         struct list_head *tmp;
 164         int failed = 0;
 165         ENTRY;
 166
 167         LASSERT(set);
 168
 169         set->brw_flags |= PTL_RPC_FL_TIMEOUT;
 170
 171         list_for_each(tmp, &set->brw_desc_head) {
 172                 struct ptlrpc_bulk_desc *desc =
 173                         list_entry(tmp, struct ptlrpc_bulk_desc, bd_set_chain);
 174
 175                 /* Skip descriptors that were completed successfully. */
 176                 if (desc->bd_flags & (PTL_BULK_FL_RCVD | PTL_BULK_FL_SENT))
 177                         continue;
 178
 179                 LASSERT(desc->bd_connection);
 180
 181                 /* If PtlMDUnlink succeeds, then it hasn't completed yet.  If it
 182                  * fails, the bulk finished _just_ in time (after the timeout
 183                  * fired but before we got this far) and we'll let it live.
 184                  */
 185                 if (PtlMDUnlink(desc->bd_md_h) != 0) {
 186                         CERROR("Near-miss on OST %s -- need to adjust "
 187                                "obd_timeout?\n",
 188                                desc->bd_connection->c_remote_uuid);
 189                         continue;
 190                 }
 191
 192                 CERROR("IO of %d pages to/from %s:%d (conn %p) timed out\n",
 193                        desc->bd_page_count, desc->bd_connection->c_remote_uuid,
 194                        desc->bd_portal, desc->bd_connection);
 195
 196                 /* This one will "never" arrive, don't wait for it. */
 197                 if (atomic_dec_and_test(&set->brw_refcount))
 198                         wake_up(&set->brw_waitq);
 199
 200                 if (class_signal_connection_failure)
 201                         class_signal_connection_failure(desc->bd_connection);
 202                 else
 203                         failed = 1;
 204         }
 205
 206         /* 0 = We go back to sleep, until we're resumed or interrupted */
 207         /* 1 = We can't be recovered, just abort the syscall with -ETIMEDOUT */
 208         RETURN(failed);
 209 }
 210
 211 static int ll_sync_brw_intr(void *data)
 212 {
 213         struct obd_brw_set *set = data;
 214
 215         ENTRY;
 216         set->brw_flags |= PTL_RPC_FL_INTR;
 217         RETURN(1); /* ignored, as of this writing */
 218 }
 219
 220 int ll_brw_sync_wait(struct obd_brw_set *set, int phase)
 221 {
 222         struct l_wait_info lwi;
 223         struct list_head *tmp, *next;
 224         int rc = 0;
 225         ENTRY;
 226
 227         switch(phase) {
 228         case CB_PHASE_START:
 229                 lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ll_sync_brw_timeout,
 230                                        ll_sync_brw_intr, set);
 231                 rc = l_wait_event(set->brw_waitq,
 232                                   atomic_read(&set->brw_refcount) == 0, &lwi);
 233
 234                 list_for_each_safe(tmp, next, &set->brw_desc_head) {
 235                         struct ptlrpc_bulk_desc *desc =
 236                                 list_entry(tmp, struct ptlrpc_bulk_desc,
 237                                            bd_set_chain);
 238                         list_del_init(&desc->bd_set_chain);
 239                         ptlrpc_bulk_decref(desc);
 240                 }
 241                 break;
 242         case CB_PHASE_FINISH:
 243                 if (atomic_dec_and_test(&set->brw_refcount))
 244                         wake_up(&set->brw_waitq);
 245                 break;
 246         default:
 247                 LBUG();
 248         }
 249
 250         RETURN(rc);
 251 }
 252
 253 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, int opcode,
 254                                        int count, int *lengths, char **bufs)
 255 {
 256         struct ptlrpc_connection *conn = imp->imp_connection;
 257         struct ptlrpc_request *request;
 258         int rc;
 259         ENTRY;
 260
 261         OBD_ALLOC(request, sizeof(*request));
 262         if (!request) {
 263                 CERROR("request allocation out of memory\n");
 264                 RETURN(NULL);
 265         }
 266
 267         rc = lustre_pack_msg(count, lengths, bufs,
 268                              &request->rq_reqlen, &request->rq_reqmsg);
 269         if (rc) {
 270                 CERROR("cannot pack request %d\n", rc);
 271                 OBD_FREE(request, sizeof(*request));
 272                 RETURN(NULL);
 273         }
 274
 275         request->rq_level = LUSTRE_CONN_FULL;
 276         request->rq_type = PTL_RPC_MSG_REQUEST;
 277         request->rq_import = imp;
 278
 279         /* XXX FIXME bug 625069 */
 280         request->rq_request_portal = imp->imp_client->cli_request_portal;
 281         request->rq_reply_portal = imp->imp_client->cli_reply_portal;
 282
 283         request->rq_connection = ptlrpc_connection_addref(conn);
 284
 285         INIT_LIST_HEAD(&request->rq_list);
 286         atomic_set(&request->rq_refcount, 1);
 287
 288         spin_lock(&imp->imp_lock);
 289         request->rq_xid = HTON__u32(++imp->imp_last_xid);
 290         spin_unlock(&imp->imp_lock);
 291
 292         request->rq_reqmsg->magic = PTLRPC_MSG_MAGIC;
 293         request->rq_reqmsg->version = PTLRPC_MSG_VERSION;
 294         request->rq_reqmsg->opc = HTON__u32(opcode);
 295         request->rq_reqmsg->flags = 0;
 296
 297         ptlrpc_hdl2req(request, &imp->imp_handle);
 298         RETURN(request);
 299 }
 300
 301 static void __ptlrpc_free_req(struct ptlrpc_request *request, int locked)
 302 {
 303         ENTRY;
 304         if (request == NULL) {
 305                 EXIT;
 306                 return;
 307         }
 308
 309         if (atomic_read(&request->rq_refcount) != 0) {
 310                 CERROR("freeing request %p (%d->%s:%d) with refcount %d\n",
 311                        request, request->rq_reqmsg->opc,
 312                        request->rq_connection->c_remote_uuid,
 313                        request->rq_import->imp_client->cli_request_portal,
 314                        request->rq_refcount);
 315                 /* LBUG(); */
 316         }
 317
 318         if (request->rq_repmsg != NULL) {
 319                 OBD_FREE(request->rq_repmsg, request->rq_replen);
 320                 request->rq_repmsg = NULL;
 321                 request->rq_reply_md.start = NULL;
 322         }
 323         if (request->rq_reqmsg != NULL) {
 324                 OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
 325                 request->rq_reqmsg = NULL;
 326         }
 327
 328         if (request->rq_import) {
 329                 if (!locked)
 330                         spin_lock(&request->rq_import->imp_lock);
 331                 list_del_init(&request->rq_list);
 332                 if (!locked)
 333                         spin_unlock(&request->rq_import->imp_lock);
 334         }
 335
 336         ptlrpc_put_connection(request->rq_connection);
 337         OBD_FREE(request, sizeof(*request));
 338         EXIT;
 339 }
 340
 341 void ptlrpc_free_req(struct ptlrpc_request *request)
 342 {
 343         __ptlrpc_free_req(request, 0);
 344 }
 345
 346 static int __ptlrpc_req_finished(struct ptlrpc_request *request, int locked)
 347 {
 348         ENTRY;
 349         if (request == NULL)
 350                 RETURN(1);
 351
 352         if (atomic_dec_and_test(&request->rq_refcount)) {
 353                 __ptlrpc_free_req(request, locked);
 354                 RETURN(1);
 355         }
 356
 357         DEBUG_REQ(D_INFO, request, "refcount now %u",
 358                   atomic_read(&request->rq_refcount));
 359         RETURN(0);
 360 }
 361
 362 void ptlrpc_req_finished(struct ptlrpc_request *request)
 363 {
 364         __ptlrpc_req_finished(request, 0);
 365 }
 366
 367 static int ptlrpc_check_reply(struct ptlrpc_request *req)
 368 {
 369         int rc = 0;
 370
 371         ENTRY;
 372         if (req->rq_repmsg != NULL) {
 373                 req->rq_transno = NTOH__u64(req->rq_repmsg->transno);
 374                 req->rq_flags |= PTL_RPC_FL_REPLIED;
 375                 GOTO(out, rc = 1);
 376         }
 377
 378         if (req->rq_flags & PTL_RPC_FL_RESEND) {
 379                 DEBUG_REQ(D_ERROR, req, "RESEND:");
 380                 GOTO(out, rc = 1);
 381         }
 382
 383         if (req->rq_flags & PTL_RPC_FL_ERR) {
 384                 DEBUG_REQ(D_ERROR, req, "ABORTED:");
 385                 GOTO(out, rc = 1);
 386         }
 387
 388         if (req->rq_flags & PTL_RPC_FL_RESTART) {
 389                 DEBUG_REQ(D_ERROR, req, "RESTART:");
 390                 GOTO(out, rc = 1);
 391         }
 392  out:
 393         DEBUG_REQ(D_NET, req, "rc = %d for", rc);
 394         return rc;
 395 }
 396
 397 static int ptlrpc_check_status(struct ptlrpc_request *req)
 398 {
 399         int err;
 400         ENTRY;
 401
 402         err = req->rq_repmsg->status;
 403         if (req->rq_repmsg->type == NTOH__u32(PTL_RPC_MSG_ERR)) {
 404                 CERROR("req->rq_repmsg->type == PTL_RPC_MSG_ERR\n");
 405                 RETURN(err ? err : -EINVAL);
 406         }
 407
 408         if (err != 0) {
 409                 if (err < 0)
 410                         CDEBUG(D_INFO, "req->rq_repmsg->status is %d\n", err);
 411                 else
 412                         CDEBUG(D_INFO, "req->rq_repmsg->status is %d\n", err);
 413         }
 414
 415         RETURN(err);
 416 }
 417
 418 static void ptlrpc_cleanup_request_buf(struct ptlrpc_request *request)
 419 {
 420         OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
 421         request->rq_reqmsg = NULL;
 422         request->rq_reqlen = 0;
 423 }
 424
 425 /* Abort this request and cleanup any resources associated with it. */
 426 static int ptlrpc_abort(struct ptlrpc_request *request)
 427 {
 428         /* First remove the ME for the reply; in theory, this means
 429          * that we can tear down the buffer safely. */
 430         PtlMEUnlink(request->rq_reply_me_h);
 431         OBD_FREE(request->rq_reply_md.start, request->rq_replen);
 432         request->rq_repmsg = NULL;
 433         request->rq_replen = 0;
 434         return 0;
 435 }
 436
 437 /* caller must hold imp->imp_lock */
 438 void ptlrpc_free_committed(struct obd_import *imp)
 439 {
 440         struct list_head *tmp, *saved;
 441         struct ptlrpc_request *req;
 442
 443         list_for_each_safe(tmp, saved, &imp->imp_replay_list) {
 444                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 445
 446                 if (req->rq_flags & PTL_RPC_FL_REPLAY) {
 447                         DEBUG_REQ(D_HA, req, "keeping (FL_REPLAY)");
 448                         continue;
 449                 }
 450
 451                 /* not yet committed */
 452                 if (req->rq_transno > imp->imp_peer_committed_transno)
 453                         break;
 454
 455                 DEBUG_REQ(D_HA, req, "committing (last_committed %Lu)",
 456                           imp->imp_peer_committed_transno);
 457                 __ptlrpc_req_finished(req, 1);
 458         }
 459
 460         EXIT;
 461         return;
 462 }
 463
 464 void ptlrpc_cleanup_client(struct obd_import *imp)
 465 {
 466         struct list_head *tmp, *saved;
 467         struct ptlrpc_request *req;
 468         struct ptlrpc_connection *conn = imp->imp_connection;
 469         ENTRY;
 470
 471         LASSERT(conn);
 472
 473         spin_lock(&imp->imp_lock);
 474         list_for_each_safe(tmp, saved, &imp->imp_replay_list) {
 475                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 476
 477                 /* XXX we should make sure that nobody's sleeping on these! */
 478                 DEBUG_REQ(D_HA, req, "cleaning up from sending list");
 479                 list_del_init(&req->rq_list);
 480                 req->rq_import = NULL;
 481                 __ptlrpc_req_finished(req, 0);
 482         }
 483         spin_unlock(&imp->imp_lock);
 484
 485         EXIT;
 486         return;
 487 }
 488
 489 void ptlrpc_continue_req(struct ptlrpc_request *req)
 490 {
 491         ENTRY;
 492         DEBUG_REQ(D_HA, req, "continuing delayed request");
 493         req->rq_reqmsg->addr = req->rq_import->imp_handle.addr;
 494         req->rq_reqmsg->cookie = req->rq_import->imp_handle.cookie;
 495         wake_up(&req->rq_wait_for_rep);
 496         EXIT;
 497 }
 498
 499 void ptlrpc_resend_req(struct ptlrpc_request *req)
 500 {
 501         ENTRY;
 502         DEBUG_REQ(D_HA, req, "resending");
 503         req->rq_reqmsg->addr = req->rq_import->imp_handle.addr;
 504         req->rq_reqmsg->cookie = req->rq_import->imp_handle.cookie;
 505         req->rq_status = -EAGAIN;
 506         req->rq_level = LUSTRE_CONN_RECOVD;
 507         req->rq_flags |= PTL_RPC_FL_RESEND;
 508         req->rq_flags &= ~PTL_RPC_FL_TIMEOUT;
 509         wake_up(&req->rq_wait_for_rep);
 510         EXIT;
 511 }
 512
 513 void ptlrpc_restart_req(struct ptlrpc_request *req)
 514 {
 515         ENTRY;
 516         DEBUG_REQ(D_HA, req, "restarting (possibly-)completed request");
 517         req->rq_status = -ERESTARTSYS;
 518         req->rq_flags |= PTL_RPC_FL_RESTART;
 519         req->rq_flags &= ~PTL_RPC_FL_TIMEOUT;
 520         wake_up(&req->rq_wait_for_rep);
 521         EXIT;
 522 }
 523
 524 static int expired_request(void *data)
 525 {
 526         struct ptlrpc_request *req = data;
 527
 528         ENTRY;
 529         if (!req) {
 530                 CERROR("NULL req!");
 531                 LBUG();
 532                 RETURN(0);
 533         }
 534
 535         DEBUG_REQ(D_ERROR, req, "timeout");
 536         req->rq_flags |= PTL_RPC_FL_TIMEOUT;
 537
 538         if (!req->rq_import) {
 539                 DEBUG_REQ(D_ERROR, req, "NULL import");
 540                 LBUG();
 541                 RETURN(0);
 542         }
 543
 544         if (!req->rq_import->imp_connection) {
 545                 DEBUG_REQ(D_ERROR, req, "NULL connection");
 546                 LBUG();
 547                 RETURN(0);
 548         }
 549
 550         if (!req->rq_import->imp_connection->c_recovd_data.rd_recovd)
 551                 RETURN(1);
 552
 553         req->rq_timeout = 0;
 554         recovd_conn_fail(req->rq_import->imp_connection);
 555
 556 #if 0
 557         /* If this request is for recovery or other primordial tasks,
 558          * don't go back to sleep.
 559          */
 560         if (req->rq_level < LUSTRE_CONN_FULL)
 561                 RETURN(1);
 562 #endif
 563         RETURN(0);
 564 }
 565
 566 static int interrupted_request(void *data)
 567 {
 568         struct ptlrpc_request *req = data;
 569         ENTRY;
 570         req->rq_flags |= PTL_RPC_FL_INTR;
 571         RETURN(1); /* ignored, as of this writing */
 572 }
 573
 574 /* If the import has been invalidated (such as by an OST failure), the
 575  * request must fail with -EIO.
 576  *
 577  * Must be called with imp_lock held, will drop it if it returns -EIO.
 578  */
 579 #define EIO_IF_INVALID(req)                                                   \
 580 if (req->rq_import->imp_flags & IMP_INVALID) {                                \
 581         DEBUG_REQ(D_ERROR, req, "IMP_INVALID:");                              \
 582         spin_unlock(&imp->imp_lock);                                          \
 583         RETURN(-EIO);                                                         \
 584 }
 585
 586 int ptlrpc_queue_wait(struct ptlrpc_request *req)
 587 {
 588         int rc = 0;
 589         struct l_wait_info lwi;
 590         struct obd_import *imp = req->rq_import;
 591         struct ptlrpc_connection *conn = imp->imp_connection;
 592         ENTRY;
 593
 594         init_waitqueue_head(&req->rq_wait_for_rep);
 595
 596         /* for distributed debugging */
 597         req->rq_reqmsg->status = HTON__u32(current->pid);
 598         CDEBUG(D_RPCTRACE, "Sending RPC pid:xid:nid:opc %d:"LPU64":%x:%d\n",
 599                NTOH__u32(req->rq_reqmsg->status), req->rq_xid,
 600                conn->c_peer.peer_nid, NTOH__u32(req->rq_reqmsg->opc));
 601
 602         if (req->rq_level > imp->imp_level) {
 603                 spin_lock(&imp->imp_lock);
 604                 EIO_IF_INVALID(req);
 605                 list_del(&req->rq_list);
 606                 list_add_tail(&req->rq_list, &imp->imp_delayed_list);
 607                 spin_unlock(&imp->imp_lock);
 608
 609                 DEBUG_REQ(D_HA, req, "\"%s\" waiting for recovery: (%d < %d)",
 610                           current->comm, req->rq_level, imp->imp_level);
 611                 lwi = LWI_INTR(NULL, NULL);
 612                 rc = l_wait_event(req->rq_wait_for_rep,
 613                                   (req->rq_level <= imp->imp_level) ||
 614                                   (req->rq_flags & PTL_RPC_FL_ERR), &lwi);
 615
 616                 spin_lock(&imp->imp_lock);
 617                 list_del_init(&req->rq_list);
 618                 spin_unlock(&imp->imp_lock);
 619
 620                 if (req->rq_flags & PTL_RPC_FL_ERR)
 621                         RETURN(-EIO);
 622
 623                 if (rc)
 624                         RETURN(rc);
 625
 626                 CERROR("process %d resumed\n", current->pid);
 627         }
 628  resend:
 629         req->rq_timeout = obd_timeout;
 630         spin_lock(&imp->imp_lock);
 631         EIO_IF_INVALID(req);
 632
 633         LASSERT(list_empty(&req->rq_list));
 634         list_add_tail(&req->rq_list, &imp->imp_sending_list);
 635         spin_unlock(&imp->imp_lock);
 636         rc = ptl_send_rpc(req);
 637         if (rc) {
 638                 CDEBUG(D_HA, "error %d, opcode %d, need recovery\n", rc,
 639                        req->rq_reqmsg->opc);
 640                 /* sleep for a jiffy, then trigger recovery */
 641                 lwi = LWI_TIMEOUT_INTR(1, expired_request,
 642                                        interrupted_request, req);
 643         } else {
 644                 DEBUG_REQ(D_NET, req, "-- sleeping");
 645                 lwi = LWI_TIMEOUT_INTR(req->rq_timeout * HZ, expired_request,
 646                                        interrupted_request, req);
 647         }
 648         l_wait_event(req->rq_wait_for_rep, ptlrpc_check_reply(req), &lwi);
 649         DEBUG_REQ(D_NET, req, "-- done sleeping");
 650
 651         spin_lock(&imp->imp_lock);
 652         list_del_init(&req->rq_list);
 653         spin_unlock(&imp->imp_lock);
 654
 655         if (req->rq_flags & PTL_RPC_FL_ERR) {
 656                 ptlrpc_abort(req);
 657                 GOTO(out, rc = -EIO);
 658         }
 659
 660         /* Don't resend if we were interrupted. */
 661         if ((req->rq_flags & (PTL_RPC_FL_RESEND | PTL_RPC_FL_INTR)) ==
 662             PTL_RPC_FL_RESEND) {
 663                 req->rq_flags &= ~PTL_RPC_FL_RESEND;
 664                 DEBUG_REQ(D_HA, req, "resending: ");
 665                 goto resend;
 666         }
 667
 668         if (req->rq_flags & PTL_RPC_FL_INTR) {
 669                 if (!(req->rq_flags & PTL_RPC_FL_TIMEOUT))
 670                         LBUG(); /* should only be interrupted if we timed out */
 671                 /* Clean up the dangling reply buffers */
 672                 ptlrpc_abort(req);
 673                 GOTO(out, rc = -EINTR);
 674         }
 675
 676         if (req->rq_flags & PTL_RPC_FL_TIMEOUT)
 677                 GOTO(out, rc = -ETIMEDOUT);
 678
 679         if (!(req->rq_flags & PTL_RPC_FL_REPLIED))
 680                 GOTO(out, rc = req->rq_status);
 681
 682         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
 683         if (rc) {
 684                 CERROR("unpack_rep failed: %d\n", rc);
 685                 GOTO(out, rc);
 686         }
 687 #if 0
 688         /* FIXME: Enable when BlueArc makes new release */
 689         if (req->rq_repmsg->type != PTL_RPC_MSG_REPLY &&
 690             req->rq_repmsg->type != PTL_RPC_MSG_ERR) {
 691                 CERROR("invalid packet type received (type=%u)\n",
 692                        req->rq_repmsg->type);
 693                 LBUG();
 694                 GOTO(out, rc = -EINVAL);
 695         }
 696 #endif
 697         CDEBUG(D_NET, "got rep "LPU64"\n", req->rq_xid);
 698         if (req->rq_repmsg->status == 0)
 699                 CDEBUG(D_NET, "--> buf %p len %d status %d\n", req->rq_repmsg,
 700                        req->rq_replen, req->rq_repmsg->status);
 701
 702
 703         if (req->rq_import->imp_flags & IMP_REPLAYABLE) {
 704                 spin_lock(&imp->imp_lock);
 705                 if (req->rq_flags & PTL_RPC_FL_REPLAY || req->rq_transno != 0) {
 706                         /* Balanced in ptlrpc_free_committed, usually. */
 707                         atomic_inc(&req->rq_refcount);
 708                         list_add_tail(&req->rq_list, &imp->imp_replay_list);
 709                 }
 710
 711                 if (req->rq_transno > imp->imp_max_transno) {
 712                         imp->imp_max_transno = req->rq_transno;
 713                 } else if (req->rq_transno != 0 &&
 714                            imp->imp_level == LUSTRE_CONN_FULL) {
 715                         CERROR("got transno "LPD64" after "LPD64": recovery "
 716                                "may not work\n", req->rq_transno,
 717                                imp->imp_max_transno);
 718                 }
 719
 720                 /* Replay-enabled imports return commit-status information. */
 721                 imp->imp_peer_last_xid = req->rq_repmsg->last_xid;
 722                 imp->imp_peer_committed_transno =
 723                         req->rq_repmsg->last_committed;
 724                 spin_unlock(&imp->imp_lock);
 725                 ptlrpc_free_committed(imp);
 726         }
 727
 728         rc = ptlrpc_check_status(req);
 729
 730         EXIT;
 731  out:
 732         return rc;
 733 }
 734
 735 #undef EIO_IF_INVALID
 736
 737 int ptlrpc_replay_req(struct ptlrpc_request *req)
 738 {
 739         int rc = 0, old_level, old_status = 0;
 740         // struct ptlrpc_client *cli = req->rq_import->imp_client;
 741         struct l_wait_info lwi;
 742         ENTRY;
 743
 744         init_waitqueue_head(&req->rq_wait_for_rep);
 745         DEBUG_REQ(D_NET, req, "");
 746
 747         req->rq_timeout = obd_timeout;
 748         req->rq_reqmsg->addr = req->rq_import->imp_handle.addr;
 749         req->rq_reqmsg->cookie = req->rq_import->imp_handle.cookie;
 750
 751         /* temporarily set request to RECOVD level (reset at out:) */
 752         old_level = req->rq_level;
 753         if (req->rq_flags & PTL_RPC_FL_REPLIED)
 754                 old_status = req->rq_repmsg->status;
 755         req->rq_level = LUSTRE_CONN_RECOVD;
 756         rc = ptl_send_rpc(req);
 757         if (rc) {
 758                 CERROR("error %d, opcode %d\n", rc, req->rq_reqmsg->opc);
 759                 ptlrpc_cleanup_request_buf(req);
 760                 // up(&cli->cli_rpc_sem);
 761                 GOTO(out, rc = -rc);
 762         }
 763
 764         CDEBUG(D_OTHER, "-- sleeping\n");
 765         lwi = LWI_INTR(NULL, NULL); /* XXX needs timeout, nested recovery */
 766         l_wait_event(req->rq_wait_for_rep, ptlrpc_check_reply(req), &lwi);
 767         CDEBUG(D_OTHER, "-- done\n");
 768
 769         // up(&cli->cli_rpc_sem);
 770
 771         if (!(req->rq_flags & PTL_RPC_FL_REPLIED)) {
 772                 CERROR("Unknown reason for wakeup\n");
 773                 /* XXX Phil - I end up here when I kill obdctl */
 774                 ptlrpc_abort(req);
 775                 GOTO(out, rc = -EINTR);
 776         }
 777
 778         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
 779         if (rc) {
 780                 CERROR("unpack_rep failed: %d\n", rc);
 781                 GOTO(out, rc);
 782         }
 783
 784         CDEBUG(D_NET, "got rep "LPD64"\n", req->rq_xid);
 785
 786         /* let the callback do fixups, possibly including in the request */
 787         if (req->rq_replay_cb)
 788                 req->rq_replay_cb(req);
 789
 790         if ((req->rq_flags & PTL_RPC_FL_REPLIED) &&
 791             req->rq_repmsg->status != old_status) {
 792                 DEBUG_REQ(D_HA, req, "status %d, old was %d",
 793                           req->rq_repmsg->status, old_status);
 794         }
 795
 796  out:
 797         req->rq_level = old_level;
 798         RETURN(rc);
 799 }