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