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                 ptl_set_inv_handle(&desc->bd_md_h);
  93                 ptl_set_inv_handle(&desc->bd_me_h);
  94         }
  95
  96         return desc;
  97 }
  98
  99 struct ptlrpc_bulk_page *ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc)
 100 {
 101         struct ptlrpc_bulk_page *bulk;
 102
 103         OBD_ALLOC(bulk, sizeof(*bulk));
 104         if (bulk != NULL) {
 105                 bulk->bp_desc = desc;
 106                 list_add_tail(&bulk->bp_link, &desc->bd_page_list);
 107                 desc->bd_page_count++;
 108         }
 109         return bulk;
 110 }
 111
 112 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *desc)
 113 {
 114         struct list_head *tmp, *next;
 115         ENTRY;
 116         if (desc == NULL) {
 117                 EXIT;
 118                 return;
 119         }
 120
 121         list_for_each_safe(tmp, next, &desc->bd_page_list) {
 122                 struct ptlrpc_bulk_page *bulk;
 123                 bulk = list_entry(tmp, struct ptlrpc_bulk_page, bp_link);
 124                 ptlrpc_free_bulk_page(bulk);
 125         }
 126
 127         ptlrpc_put_connection(desc->bd_connection);
 128
 129         OBD_FREE(desc, sizeof(*desc));
 130         EXIT;
 131 }
 132
 133 void ptlrpc_free_bulk_page(struct ptlrpc_bulk_page *bulk)
 134 {
 135         ENTRY;
 136         if (bulk == NULL) {
 137                 EXIT;
 138                 return;
 139         }
 140
 141         list_del(&bulk->bp_link);
 142         bulk->bp_desc->bd_page_count--;
 143         OBD_FREE(bulk, sizeof(*bulk));
 144         EXIT;
 145 }
 146
 147 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, int opcode,
 148                                        int count, int *lengths, char **bufs)
 149 {
 150         struct ptlrpc_connection *conn = imp->imp_connection;
 151         struct ptlrpc_request *request;
 152         int rc;
 153         ENTRY;
 154
 155         OBD_ALLOC(request, sizeof(*request));
 156         if (!request) {
 157                 CERROR("request allocation out of memory\n");
 158                 RETURN(NULL);
 159         }
 160
 161         rc = lustre_pack_msg(count, lengths, bufs,
 162                              &request->rq_reqlen, &request->rq_reqmsg);
 163         if (rc) {
 164                 CERROR("cannot pack request %d\n", rc);
 165                 OBD_FREE(request, sizeof(*request));
 166                 RETURN(NULL);
 167         }
 168
 169         request->rq_level = LUSTRE_CONN_FULL;
 170         request->rq_type = PTL_RPC_MSG_REQUEST;
 171         request->rq_import = imp;
 172
 173         /* XXX FIXME bug 625069 */
 174         request->rq_request_portal = imp->imp_client->cli_request_portal;
 175         request->rq_reply_portal = imp->imp_client->cli_reply_portal;
 176
 177         request->rq_connection = ptlrpc_connection_addref(conn);
 178
 179         INIT_LIST_HEAD(&request->rq_list);
 180         /*
 181          * This will be reduced once when the sender is finished (waiting for
 182          * reply, f.e.), and once when the request has been committed and is
 183          * removed from the to-be-committed list.
 184          *
 185          * Also, the refcount will be increased in ptl_send_rpc immediately
 186          * before we hand it off to portals, and there will be a corresponding
 187          * decrease in request_out_cb (which is called to indicate that portals
 188          * is finished with the request, and it can be safely freed).
 189          *
 190          * (Except in the DLM server case, where it will be dropped twice
 191          * by the sender, and then the last time by request_out_callback.)
 192          */
 193         atomic_set(&request->rq_refcount, 2);
 194
 195         spin_lock(&conn->c_lock);
 196         request->rq_xid = HTON__u32(++conn->c_xid_out);
 197         spin_unlock(&conn->c_lock);
 198
 199         request->rq_reqmsg->magic = PTLRPC_MSG_MAGIC;
 200         request->rq_reqmsg->version = PTLRPC_MSG_VERSION;
 201         request->rq_reqmsg->opc = HTON__u32(opcode);
 202
 203         ptlrpc_hdl2req(request, &imp->imp_handle);
 204         RETURN(request);
 205 }
 206
 207 void ptlrpc_req_finished(struct ptlrpc_request *request)
 208 {
 209         if (request == NULL)
 210                 return;
 211
 212         if (atomic_dec_and_test(&request->rq_refcount))
 213                 ptlrpc_free_req(request);
 214 }
 215
 216 void ptlrpc_free_req(struct ptlrpc_request *request)
 217 {
 218         ENTRY;
 219         if (request == NULL) {
 220                 EXIT;
 221                 return;
 222         }
 223
 224         if (atomic_read(&request->rq_refcount) != 0) {
 225                 CERROR("freeing request %p (%d->%s:%d) with refcount %d\n",
 226                        request, request->rq_reqmsg->opc,
 227                        request->rq_connection->c_remote_uuid,
 228                        request->rq_import->imp_client->cli_request_portal,
 229                        request->rq_refcount);
 230                 /* LBUG(); */
 231         }
 232
 233         if (request->rq_repmsg != NULL) {
 234                 OBD_FREE(request->rq_repmsg, request->rq_replen);
 235                 request->rq_repmsg = NULL;
 236                 request->rq_reply_md.start = NULL;
 237         }
 238         if (request->rq_reqmsg != NULL) {
 239                 OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
 240                 request->rq_reqmsg = NULL;
 241         }
 242
 243         if (request->rq_connection) {
 244                 spin_lock(&request->rq_connection->c_lock);
 245                 list_del_init(&request->rq_list);
 246                 spin_unlock(&request->rq_connection->c_lock);
 247         }
 248
 249         ptlrpc_put_connection(request->rq_connection);
 250         OBD_FREE(request, sizeof(*request));
 251         EXIT;
 252 }
 253
 254 static int ptlrpc_check_reply(struct ptlrpc_request *req)
 255 {
 256         int rc = 0;
 257
 258         if (req->rq_repmsg != NULL) {
 259                 struct ptlrpc_connection *conn = req->rq_import->imp_connection;
 260                 spin_lock(&conn->c_lock);
 261                 if (req->rq_level > conn->c_level) {
 262                         CDEBUG(D_HA,
 263                                "rep to xid "LPD64" op %d to %s:%d: "
 264                                "recovery started, ignoring (%d > %d)\n",
 265                                (unsigned long long)req->rq_xid,
 266                                req->rq_reqmsg->opc, conn->c_remote_uuid,
 267                                req->rq_import->imp_client->cli_request_portal,
 268                                req->rq_level, conn->c_level);
 269                         req->rq_repmsg = NULL;
 270                         spin_unlock(&conn->c_lock);
 271                         GOTO(out, rc = 0);
 272                 }
 273                 spin_unlock(&conn->c_lock);
 274                 req->rq_transno = NTOH__u64(req->rq_repmsg->transno);
 275                 req->rq_flags |= PTL_RPC_FL_REPLIED;
 276                 GOTO(out, rc = 1);
 277         }
 278
 279         if (req->rq_flags & PTL_RPC_FL_RESEND) {
 280                 CERROR("-- RESTART --\n");
 281                 GOTO(out, rc = 1);
 282         }
 283
 284         if (req->rq_flags & PTL_RPC_FL_ERR) {
 285                 CERROR("-- ABORTED --\n");
 286                 GOTO(out, rc = 1);
 287         }
 288
 289  out:
 290         CDEBUG(D_NET, "req = %p, rc = %d\n", req, rc);
 291         return rc;
 292 }
 293
 294 int ptlrpc_check_status(struct ptlrpc_request *req, int err)
 295 {
 296         ENTRY;
 297
 298         if (err != 0) {
 299                 CERROR("err is %d\n", err);
 300                 RETURN(err);
 301         }
 302
 303         if (req == NULL) {
 304                 CERROR("req == NULL\n");
 305                 RETURN(-ENOMEM);
 306         }
 307
 308         if (req->rq_repmsg == NULL) {
 309                 CERROR("req->rq_repmsg == NULL\n");
 310                 RETURN(-ENOMEM);
 311         }
 312
 313         err = req->rq_repmsg->status;
 314         if (req->rq_repmsg->type == NTOH__u32(PTL_RPC_MSG_ERR)) {
 315                 CERROR("req->rq_repmsg->type == PTL_RPC_MSG_ERR\n");
 316                 RETURN(err ? err : -EINVAL);
 317         }
 318
 319         if (err != 0) {
 320                 if (err < 0)
 321                         CERROR("req->rq_repmsg->status is %d\n", err);
 322                 else
 323                         CDEBUG(D_INFO, "req->rq_repmsg->status is %d\n", err);
 324                 /* XXX: translate this error from net to host */
 325                 RETURN(err);
 326         }
 327
 328         RETURN(0);
 329 }
 330
 331 static void ptlrpc_cleanup_request_buf(struct ptlrpc_request *request)
 332 {
 333         OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
 334         request->rq_reqmsg = NULL;
 335         request->rq_reqlen = 0;
 336 }
 337
 338 /* Abort this request and cleanup any resources associated with it. */
 339 static int ptlrpc_abort(struct ptlrpc_request *request)
 340 {
 341         /* First remove the ME for the reply; in theory, this means
 342          * that we can tear down the buffer safely. */
 343         PtlMEUnlink(request->rq_reply_me_h);
 344         OBD_FREE(request->rq_reply_md.start, request->rq_replen);
 345         request->rq_repmsg = NULL;
 346         request->rq_replen = 0;
 347         return 0;
 348 }
 349
 350 /* caller must hold conn->c_lock */
 351 void ptlrpc_free_committed(struct ptlrpc_connection *conn)
 352 {
 353         struct list_head *tmp, *saved;
 354         struct ptlrpc_request *req;
 355
 356 restart:
 357         list_for_each_safe(tmp, saved, &conn->c_sending_head) {
 358                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 359
 360                 if (req->rq_flags & PTL_RPC_FL_REPLAY) {
 361                         CDEBUG(D_INFO, "Keeping req %p xid "LPD64" for replay\n",
 362                                req, req->rq_xid);
 363                         continue;
 364                 }
 365
 366                 if (!(req->rq_flags & PTL_RPC_FL_REPLIED)) {
 367                         CDEBUG(D_INFO, "Keeping in-flight req %p xid "LPD64
 368                                " for replay\n", req, req->rq_xid);
 369                         continue;
 370                 }
 371
 372                 /* not yet committed */
 373                 if (req->rq_transno > conn->c_last_committed)
 374                         break;
 375
 376                 CDEBUG(D_INFO, "Marking request %p xid %Ld as committed "
 377                        "transno=%Lu, last_committed=%Lu\n", req,
 378                        (long long)req->rq_xid, (long long)req->rq_transno,
 379                        (long long)conn->c_last_committed);
 380                 if (atomic_dec_and_test(&req->rq_refcount)) {
 381                         /* We do this to prevent free_req deadlock.
 382                          * Restarting after each removal is not so bad, as we are
 383                          * almost always deleting the first item in the list.
 384                          *
 385                          * If we use a recursive lock here, we can skip the
 386                          * unlock/lock/restart sequence.
 387                          */
 388                         spin_unlock(&conn->c_lock);
 389                         ptlrpc_free_req(req);
 390                         spin_lock(&conn->c_lock);
 391                         goto restart;
 392                 } else {
 393                         list_del(&req->rq_list);
 394                         list_add(&req->rq_list, &conn->c_dying_head);
 395                 }
 396         }
 397
 398         EXIT;
 399         return;
 400 }
 401
 402 void ptlrpc_cleanup_client(struct obd_import *imp)
 403 {
 404         struct list_head *tmp, *saved;
 405         struct ptlrpc_request *req;
 406         struct ptlrpc_connection *conn = imp->imp_connection;
 407         ENTRY;
 408
 409         LASSERT(conn);
 410
 411 restart1:
 412         spin_lock(&conn->c_lock);
 413         list_for_each_safe(tmp, saved, &conn->c_sending_head) {
 414                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 415                 if (req->rq_import != imp)
 416                         continue;
 417                 /* XXX we should make sure that nobody's sleeping on these! */
 418                 CDEBUG(D_INFO, "Cleaning req %p from sending list.\n", req);
 419                 list_del_init(&req->rq_list);
 420                 req->rq_import = NULL;
 421                 spin_unlock(&conn->c_lock);
 422                 ptlrpc_req_finished(req);
 423                 goto restart1;
 424         }
 425 restart2:
 426         list_for_each_safe(tmp, saved, &conn->c_dying_head) {
 427                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
 428                 if (req->rq_import != imp)
 429                         continue;
 430                 CERROR("Request %p is on the dying list at cleanup!\n", req);
 431                 list_del_init(&req->rq_list);
 432                 req->rq_import = NULL;
 433                 spin_unlock(&conn->c_lock);
 434                 ptlrpc_req_finished(req);
 435                 spin_lock(&conn->c_lock);
 436                 goto restart2;
 437         }
 438         spin_unlock(&conn->c_lock);
 439
 440         EXIT;
 441         return;
 442 }
 443
 444 void ptlrpc_continue_req(struct ptlrpc_request *req)
 445 {
 446         ENTRY;
 447         CDEBUG(D_HA, "continue delayed request "LPD64" opc %d\n",
 448                req->rq_xid, req->rq_reqmsg->opc);
 449         req->rq_reqmsg->addr = req->rq_import->imp_handle.addr;
 450         req->rq_reqmsg->cookie = req->rq_import->imp_handle.cookie;
 451         wake_up(&req->rq_wait_for_rep);
 452         EXIT;
 453 }
 454
 455 void ptlrpc_resend_req(struct ptlrpc_request *req)
 456 {
 457         ENTRY;
 458         CDEBUG(D_HA, "resend request "LPD64", opc %d\n",
 459                req->rq_xid, req->rq_reqmsg->opc);
 460         req->rq_reqmsg->addr = req->rq_import->imp_handle.addr;
 461         req->rq_reqmsg->cookie = req->rq_import->imp_handle.cookie;
 462         req->rq_status = -EAGAIN;
 463         req->rq_level = LUSTRE_CONN_RECOVD;
 464         req->rq_flags |= PTL_RPC_FL_RESEND;
 465         req->rq_flags &= ~PTL_RPC_FL_TIMEOUT;
 466         wake_up(&req->rq_wait_for_rep);
 467         EXIT;
 468 }
 469
 470 void ptlrpc_restart_req(struct ptlrpc_request *req)
 471 {
 472         ENTRY;
 473         CDEBUG(D_HA, "restart completed request "LPD64", opc %d\n",
 474                req->rq_xid, req->rq_reqmsg->opc);
 475         req->rq_status = -ERESTARTSYS;
 476         req->rq_flags |= PTL_RPC_FL_RECOVERY;
 477         req->rq_flags &= ~PTL_RPC_FL_TIMEOUT;
 478         wake_up(&req->rq_wait_for_rep);
 479         EXIT;
 480 }
 481
 482 static int expired_request(void *data)
 483 {
 484         struct ptlrpc_request *req = data;
 485
 486         ENTRY;
 487         CERROR("req xid "LPD64" op %d: timeout on conn to %s:%d\n",
 488                (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
 489                req->rq_connection->c_remote_uuid,
 490                req->rq_import->imp_client->cli_request_portal);
 491         req->rq_flags |= PTL_RPC_FL_TIMEOUT;
 492         if (!req->rq_import->imp_connection->c_recovd_data.rd_recovd)
 493                 RETURN(1);
 494
 495         req->rq_timeout = 0;
 496         req->rq_connection->c_level = LUSTRE_CONN_RECOVD;
 497         recovd_conn_fail(req->rq_import->imp_connection);
 498
 499         /* If this request is for recovery or other primordial tasks,
 500          * don't go back to sleep.
 501          */
 502         if (req->rq_level < LUSTRE_CONN_FULL)
 503                 RETURN(1);
 504         RETURN(0);
 505 }
 506
 507 static int interrupted_request(void *data)
 508 {
 509         struct ptlrpc_request *req = data;
 510         ENTRY;
 511         req->rq_flags |= PTL_RPC_FL_INTR;
 512         RETURN(1); /* ignored, as of this writing */
 513 }
 514
 515 /* If we're being torn down by umount -f, or the import has been
 516  * invalidated (such as by an OST failure), the request must fail with
 517  * -EIO.
 518  *
 519  * Must be called with conn->c_lock held, will drop it if it returns -EIO.
 520  *
 521  * XXX this should just be testing the import, and umount_begin shouldn't touch
 522  * XXX the connection.
 523  */
 524 #define EIO_IF_INVALID(conn, req)                                             \
 525 if ((conn->c_flags & CONN_INVALID) ||                                         \
 526     (req->rq_import->imp_flags & IMP_INVALID)) {                              \
 527         CERROR("req xid "LPD64" op %d to %s:%d: %s_INVALID\n",                \
 528                (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,          \
 529                req->rq_connection->c_remote_uuid,                             \
 530                req->rq_import->imp_client->cli_request_portal,                \
 531                (conn->c_flags & CONN_INVALID) ? "CONN_" : "IMP_");            \
 532         spin_unlock(&conn->c_lock);                                           \
 533         RETURN(-EIO);                                                         \
 534 }
 535
 536 int ptlrpc_queue_wait(struct ptlrpc_request *req)
 537 {
 538         int rc = 0;
 539         struct l_wait_info lwi;
 540         struct ptlrpc_client *cli = req->rq_import->imp_client;
 541         struct ptlrpc_connection *conn = req->rq_import->imp_connection;
 542         ENTRY;
 543
 544         init_waitqueue_head(&req->rq_wait_for_rep);
 545         CDEBUG(D_NET, "subsys: %s req "LPD64" opc %d level %d, conn level %d\n",
 546                cli->cli_name, req->rq_xid, req->rq_reqmsg->opc, req->rq_level,
 547                req->rq_connection->c_level);
 548
 549         /* XXX probably both an import and connection level are needed */
 550         if (req->rq_level > conn->c_level) {
 551                 spin_lock(&conn->c_lock);
 552                 EIO_IF_INVALID(conn, req);
 553                 list_del(&req->rq_list);
 554                 list_add_tail(&req->rq_list, &conn->c_delayed_head);
 555                 spin_unlock(&conn->c_lock);
 556
 557                 CDEBUG(D_HA, "req xid "LPD64" op %d to %s:%d: waiting for "
 558                        "recovery (%d < %d)\n",
 559                        (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
 560                        req->rq_connection->c_remote_uuid,
 561                        req->rq_import->imp_client->cli_request_portal,
 562                        req->rq_level, conn->c_level);
 563
 564                 lwi = LWI_INTR(NULL, NULL);
 565                 rc = l_wait_event(req->rq_wait_for_rep,
 566                                   (req->rq_level <= conn->c_level) ||
 567                                   (req->rq_flags & PTL_RPC_FL_ERR), &lwi);
 568
 569                 spin_lock(&conn->c_lock);
 570                 list_del_init(&req->rq_list);
 571                 spin_unlock(&conn->c_lock);
 572
 573                 if (req->rq_flags & PTL_RPC_FL_ERR)
 574                         RETURN(-EIO);
 575
 576                 if (rc)
 577                         RETURN(rc);
 578
 579                 CERROR("process %d resumed\n", current->pid);
 580         }
 581  resend:
 582         req->rq_timeout = obd_timeout;
 583         spin_lock(&conn->c_lock);
 584         EIO_IF_INVALID(conn, req);
 585
 586         list_del(&req->rq_list);
 587         list_add_tail(&req->rq_list, &conn->c_sending_head);
 588         spin_unlock(&conn->c_lock);
 589         rc = ptl_send_rpc(req);
 590         if (rc) {
 591                 CDEBUG(D_HA, "error %d, opcode %d, need recovery\n", rc,
 592                        req->rq_reqmsg->opc);
 593                 /* the sleep below will time out, triggering recovery */
 594         }
 595
 596         CDEBUG(D_NET, "-- sleeping on req xid "LPD64" op %d to %s:%d\n",
 597                        (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
 598                        req->rq_connection->c_remote_uuid,
 599                        req->rq_import->imp_client->cli_request_portal);
 600         lwi = LWI_TIMEOUT_INTR(req->rq_timeout * HZ, expired_request,
 601                                interrupted_request,req);
 602         l_wait_event(req->rq_wait_for_rep, ptlrpc_check_reply(req), &lwi);
 603         CDEBUG(D_NET, "-- done sleeping on req xid "LPD64" op %d to %s:%d\n",
 604                        (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
 605                        req->rq_connection->c_remote_uuid,
 606                        req->rq_import->imp_client->cli_request_portal);
 607
 608         if (req->rq_flags & PTL_RPC_FL_ERR) {
 609                 ptlrpc_abort(req);
 610                 GOTO(out, rc = -EIO);
 611         }
 612
 613         /* Don't resend if we were interrupted. */
 614         if ((req->rq_flags & (PTL_RPC_FL_RESEND | PTL_RPC_FL_INTR)) ==
 615             PTL_RPC_FL_RESEND) {
 616                 req->rq_flags &= ~PTL_RPC_FL_RESEND;
 617                 CDEBUG(D_HA, "resending req xid "LPD64" op %d to %s:%d\n",
 618                        (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
 619                        req->rq_connection->c_remote_uuid,
 620                        req->rq_import->imp_client->cli_request_portal);
 621                 goto resend;
 622         }
 623
 624         // up(&cli->cli_rpc_sem);
 625         if (req->rq_flags & PTL_RPC_FL_INTR) {
 626                 if (!(req->rq_flags & PTL_RPC_FL_TIMEOUT))
 627                         LBUG(); /* should only be interrupted if we timed out */
 628                 /* Clean up the dangling reply buffers */
 629                 ptlrpc_abort(req);
 630                 GOTO(out, rc = -EINTR);
 631         }
 632
 633         if (req->rq_flags & PTL_RPC_FL_TIMEOUT)
 634                 GOTO(out, rc = -ETIMEDOUT);
 635
 636         if (!(req->rq_flags & PTL_RPC_FL_REPLIED))
 637                 GOTO(out, rc = req->rq_status);
 638
 639         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
 640         if (rc) {
 641                 CERROR("unpack_rep failed: %d\n", rc);
 642                 GOTO(out, rc);
 643         }
 644 #if 0
 645         /* FIXME: Enable when BlueArc makes new release */
 646         if (req->rq_repmsg->type != PTL_RPC_MSG_REPLY &&
 647             req->rq_repmsg->type != PTL_RPC_MSG_ERR) {
 648                 CERROR("invalid packet type received (type=%u)\n",
 649                        req->rq_repmsg->type);
 650                 LBUG();
 651                 GOTO(out, rc = -EINVAL);
 652         }
 653 #endif
 654         CDEBUG(D_NET, "got rep "LPD64"\n", req->rq_xid);
 655         if (req->rq_repmsg->status == 0)
 656                 CDEBUG(D_NET, "--> buf %p len %d status %d\n", req->rq_repmsg,
 657                        req->rq_replen, req->rq_repmsg->status);
 658
 659         spin_lock(&conn->c_lock);
 660         conn->c_last_xid = req->rq_repmsg->last_xid;
 661         conn->c_last_committed = req->rq_repmsg->last_committed;
 662         ptlrpc_free_committed(conn);
 663         spin_unlock(&conn->c_lock);
 664
 665         EXIT;
 666  out:
 667         return rc;
 668 }
 669
 670 #undef EIO_IF_INVALID
 671
 672 int ptlrpc_replay_req(struct ptlrpc_request *req)
 673 {
 674         int rc = 0, old_level;
 675         // struct ptlrpc_client *cli = req->rq_import->imp_client;
 676         struct l_wait_info lwi;
 677         ENTRY;
 678
 679         init_waitqueue_head(&req->rq_wait_for_rep);
 680         CDEBUG(D_NET, "req "LPD64" opc %d level %d, conn level %d\n",
 681                req->rq_xid, req->rq_reqmsg->opc, req->rq_level,
 682                req->rq_connection->c_level);
 683
 684         req->rq_timeout = obd_timeout;
 685         req->rq_reqmsg->addr = req->rq_import->imp_handle.addr;
 686         req->rq_reqmsg->cookie = req->rq_import->imp_handle.cookie;
 687
 688         /* temporarily set request to RECOVD level (reset at out:) */
 689         old_level = req->rq_level;
 690         req->rq_level = LUSTRE_CONN_RECOVD;
 691         rc = ptl_send_rpc(req);
 692         if (rc) {
 693                 CERROR("error %d, opcode %d\n", rc, req->rq_reqmsg->opc);
 694                 ptlrpc_cleanup_request_buf(req);
 695                 // up(&cli->cli_rpc_sem);
 696                 GOTO(out, rc = -rc);
 697         }
 698
 699         CDEBUG(D_OTHER, "-- sleeping\n");
 700         lwi = LWI_INTR(NULL, NULL); /* XXX needs timeout, nested recovery */
 701         l_wait_event(req->rq_wait_for_rep, ptlrpc_check_reply(req), &lwi);
 702         CDEBUG(D_OTHER, "-- done\n");
 703
 704         // up(&cli->cli_rpc_sem);
 705
 706         if (!(req->rq_flags & PTL_RPC_FL_REPLIED)) {
 707                 CERROR("Unknown reason for wakeup\n");
 708                 /* XXX Phil - I end up here when I kill obdctl */
 709                 ptlrpc_abort(req);
 710                 GOTO(out, rc = -EINTR);
 711         }
 712
 713         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
 714         if (rc) {
 715                 CERROR("unpack_rep failed: %d\n", rc);
 716                 GOTO(out, rc);
 717         }
 718
 719         CDEBUG(D_NET, "got rep "LPD64"\n", req->rq_xid);
 720
 721         /* let the callback do fixups, possibly including in the request */
 722         if (req->rq_replay_cb)
 723                 req->rq_replay_cb(req, req->rq_replay_cb_data);
 724
 725         if (req->rq_repmsg->status == 0) {
 726                 CDEBUG(D_NET, "--> buf %p len %d status %d\n", req->rq_repmsg,
 727                        req->rq_replen, req->rq_repmsg->status);
 728         } else {
 729                 CERROR("recovery failed: ");
 730                 CERROR("req "LPD64" opc %d level %d, conn level %d\n",
 731                        req->rq_xid, req->rq_reqmsg->opc, req->rq_level,
 732                        req->rq_connection->c_level);
 733                 LBUG();
 734         }
 735
 736  out:
 737         req->rq_level = old_level;
 738         RETURN(rc);
 739 }