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, 2003 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 #ifndef __KERNEL__
  25 #include <errno.h>
  26 #include <signal.h>
  27 #include <liblustre.h>
  28 #endif
  29
  30 #include <linux/obd_support.h>
  31 #include <linux/obd_class.h>
  32 #include <linux/lustre_lib.h>
  33 #include <linux/lustre_ha.h>
  34 #include <linux/lustre_import.h>
  35
  36 #include "ptlrpc_internal.h"
  37
  38 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
  39                         struct ptlrpc_client *cl)
  40 {
  41         cl->cli_request_portal = req_portal;
  42         cl->cli_reply_portal   = rep_portal;
  43         cl->cli_name           = name;
  44 }
  45
  46 struct obd_uuid *ptlrpc_req_to_uuid(struct ptlrpc_request *req)
  47 {
  48         return &req->rq_connection->c_remote_uuid;
  49 }
  50
  51 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid)
  52 {
  53         struct ptlrpc_connection *c;
  54         struct ptlrpc_peer peer;
  55         int err;
  56
  57         err = ptlrpc_uuid_to_peer(uuid, &peer);
  58         if (err != 0) {
  59                 CERROR("cannot find peer %s!\n", uuid->uuid);
  60                 return NULL;
  61         }
  62
  63         c = ptlrpc_get_connection(&peer, uuid);
  64         if (c) {
  65                 memcpy(c->c_remote_uuid.uuid,
  66                        uuid->uuid, sizeof(c->c_remote_uuid.uuid));
  67         }
  68
  69         CDEBUG(D_INFO, "%s -> %p\n", uuid->uuid, c);
  70
  71         return c;
  72 }
  73
  74 void ptlrpc_readdress_connection(struct ptlrpc_connection *conn,
  75                                  struct obd_uuid *uuid)
  76 {
  77         struct ptlrpc_peer peer;
  78         int err;
  79
  80         err = ptlrpc_uuid_to_peer(uuid, &peer);
  81         if (err != 0) {
  82                 CERROR("cannot find peer %s!\n", uuid->uuid);
  83                 return;
  84         }
  85
  86         memcpy(&conn->c_peer, &peer, sizeof (peer));
  87         return;
  88 }
  89
  90 static inline struct ptlrpc_bulk_desc *new_bulk(void)
  91 {
  92         struct ptlrpc_bulk_desc *desc;
  93
  94         OBD_ALLOC(desc, sizeof(*desc));
  95         if (!desc)
  96                 return NULL;
  97
  98         spin_lock_init(&desc->bd_lock);
  99         init_waitqueue_head(&desc->bd_waitq);
 100         INIT_LIST_HEAD(&desc->bd_page_list);
 101         desc->bd_md_h = PTL_HANDLE_NONE;
 102         desc->bd_me_h = PTL_HANDLE_NONE;
 103
 104         return desc;
 105 }
 106
 107 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
 108                                                int type, int portal)
 109 {
 110         struct obd_import *imp = req->rq_import;
 111         struct ptlrpc_bulk_desc *desc;
 112
 113         LASSERT(type == BULK_PUT_SINK || type == BULK_GET_SOURCE);
 114
 115         desc = new_bulk();
 116         if (desc == NULL)
 117                 RETURN(NULL);
 118
 119         desc->bd_import_generation = req->rq_import_generation;
 120         desc->bd_import = class_import_get(imp);
 121         desc->bd_req = req;
 122         desc->bd_type = type;
 123         desc->bd_portal = portal;
 124
 125         /* This makes req own desc, and free it when she frees herself */
 126         req->rq_bulk = desc;
 127
 128         return desc;
 129 }
 130
 131 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp (struct ptlrpc_request *req,
 132                                                int type, int portal)
 133 {
 134         struct obd_export *exp = req->rq_export;
 135         struct ptlrpc_bulk_desc *desc;
 136
 137         LASSERT(type == BULK_PUT_SOURCE || type == BULK_GET_SINK);
 138
 139         desc = new_bulk();
 140         if (desc == NULL)
 141                 RETURN(NULL);
 142
 143         desc->bd_export = class_export_get(exp);
 144         desc->bd_req = req;
 145         desc->bd_type = type;
 146         desc->bd_portal = portal;
 147
 148         /* NB we don't assign rq_bulk here; server-side requests are
 149          * re-used, and the handler frees the bulk desc explicitly. */
 150
 151         return desc;
 152 }
 153
 154 int ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
 155                           struct page *page, int pageoffset, int len)
 156 {
 157         struct ptlrpc_bulk_page *bulk;
 158
 159         OBD_ALLOC(bulk, sizeof(*bulk));
 160         if (bulk == NULL)
 161                 return -ENOMEM;
 162
 163         LASSERT(page != NULL);
 164         LASSERT(pageoffset >= 0);
 165         LASSERT(len > 0);
 166         LASSERT(pageoffset + len <= PAGE_SIZE);
 167
 168         bulk->bp_page = page;
 169         bulk->bp_pageoffset = pageoffset;
 170         bulk->bp_buflen = len;
 171
 172         bulk->bp_desc = desc;
 173         list_add_tail(&bulk->bp_link, &desc->bd_page_list);
 174         desc->bd_page_count++;
 175         return 0;
 176 }
 177
 178 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *desc)
 179 {
 180         struct list_head *tmp, *next;
 181         ENTRY;
 182
 183         LASSERT(desc != NULL);
 184         LASSERT(desc->bd_page_count != 0x5a5a5a5a); /* not freed already */
 185         LASSERT(!desc->bd_network_rw);         /* network hands off or */
 186
 187         list_for_each_safe(tmp, next, &desc->bd_page_list) {
 188                 struct ptlrpc_bulk_page *bulk;
 189                 bulk = list_entry(tmp, struct ptlrpc_bulk_page, bp_link);
 190                 ptlrpc_free_bulk_page(bulk);
 191         }
 192
 193         LASSERT(desc->bd_page_count == 0);
 194         LASSERT((desc->bd_export != NULL) ^ (desc->bd_import != NULL));
 195
 196         if (desc->bd_export)
 197                 class_export_put(desc->bd_export);
 198         else
 199                 class_import_put(desc->bd_import);
 200
 201         OBD_FREE(desc, sizeof(*desc));
 202         EXIT;
 203 }
 204
 205 void ptlrpc_free_bulk_page(struct ptlrpc_bulk_page *bulk)
 206 {
 207         LASSERT(bulk != NULL);
 208
 209         list_del(&bulk->bp_link);
 210         bulk->bp_desc->bd_page_count--;
 211         OBD_FREE(bulk, sizeof(*bulk));
 212 }
 213
 214 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, int opcode,
 215                                        int count, int *lengths, char **bufs)
 216 {
 217         struct ptlrpc_request *request;
 218         int rc;
 219         ENTRY;
 220
 221         LASSERT((unsigned long)imp > 0x1000);
 222
 223         OBD_ALLOC(request, sizeof(*request));
 224         if (!request) {
 225                 CERROR("request allocation out of memory\n");
 226                 RETURN(NULL);
 227         }
 228
 229         rc = lustre_pack_request(request, count, lengths, bufs);
 230         if (rc) {
 231                 CERROR("cannot pack request %d\n", rc);
 232                 OBD_FREE(request, sizeof(*request));
 233                 RETURN(NULL);
 234         }
 235
 236         if (imp->imp_server_timeout)
 237                 request->rq_timeout = obd_timeout / 2;
 238         else
 239                 request->rq_timeout = obd_timeout;
 240         request->rq_send_state = LUSTRE_IMP_FULL;
 241         request->rq_type = PTL_RPC_MSG_REQUEST;
 242         request->rq_import = class_import_get(imp);
 243         request->rq_phase = RQ_PHASE_NEW;
 244
 245         /* XXX FIXME bug 249 */
 246         request->rq_request_portal = imp->imp_client->cli_request_portal;
 247         request->rq_reply_portal = imp->imp_client->cli_reply_portal;
 248
 249         request->rq_connection = ptlrpc_connection_addref(imp->imp_connection);
 250
 251         spin_lock_init(&request->rq_lock);
 252         INIT_LIST_HEAD(&request->rq_list);
 253         init_waitqueue_head(&request->rq_wait_for_rep);
 254         request->rq_xid = ptlrpc_next_xid();
 255         atomic_set(&request->rq_refcount, 1);
 256
 257         request->rq_reqmsg->opc = opcode;
 258         request->rq_reqmsg->flags = 0;
 259
 260         RETURN(request);
 261 }
 262
 263 struct ptlrpc_request_set *ptlrpc_prep_set(void)
 264 {
 265         struct ptlrpc_request_set *set;
 266
 267         OBD_ALLOC(set, sizeof *set);
 268         if (!set)
 269                 RETURN(NULL);
 270         INIT_LIST_HEAD(&set->set_requests);
 271         init_waitqueue_head(&set->set_waitq);
 272         set->set_remaining = 0;
 273         spin_lock_init(&set->set_new_req_lock);
 274         INIT_LIST_HEAD(&set->set_new_requests);
 275
 276         RETURN(set);
 277 }
 278
 279 /* Finish with this set; opposite of prep_set. */
 280 void ptlrpc_set_destroy(struct ptlrpc_request_set *set)
 281 {
 282         struct list_head *tmp;
 283         struct list_head *next;
 284         int               expected_phase;
 285         int               n = 0;
 286         ENTRY;
 287
 288         /* Requests on the set should either all be completed, or all be new */
 289         expected_phase = (set->set_remaining == 0) ?
 290                          RQ_PHASE_COMPLETE : RQ_PHASE_NEW;
 291         list_for_each (tmp, &set->set_requests) {
 292                 struct ptlrpc_request *req =
 293                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 294
 295                 LASSERT(req->rq_phase == expected_phase);
 296                 n++;
 297         }
 298
 299         LASSERT(set->set_remaining == 0 || set->set_remaining == n);
 300
 301         list_for_each_safe(tmp, next, &set->set_requests) {
 302                 struct ptlrpc_request *req =
 303                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 304                 list_del_init(&req->rq_set_chain);
 305
 306                 LASSERT(req->rq_phase == expected_phase);
 307
 308                 if (req->rq_phase == RQ_PHASE_NEW) {
 309
 310                         if (req->rq_interpret_reply != NULL) {
 311                                 int (*interpreter)(struct ptlrpc_request *,
 312                                                    void *, int) =
 313                                         req->rq_interpret_reply;
 314
 315                                 /* higher level (i.e. LOV) failed;
 316                                  * let the sub reqs clean up */
 317                                 req->rq_status = -EBADR;
 318                                 interpreter(req, &req->rq_async_args,
 319                                             req->rq_status);
 320                         }
 321                         set->set_remaining--;
 322                 }
 323
 324                 req->rq_set = NULL;
 325                 ptlrpc_req_finished (req);
 326         }
 327
 328         LASSERT(set->set_remaining == 0);
 329
 330         OBD_FREE(set, sizeof(*set));
 331         EXIT;
 332 }
 333
 334 void ptlrpc_set_add_req(struct ptlrpc_request_set *set,
 335                         struct ptlrpc_request *req)
 336 {
 337         /* The set takes over the caller's request reference */
 338         list_add_tail(&req->rq_set_chain, &set->set_requests);
 339         req->rq_set = set;
 340         set->set_remaining++;
 341 }
 342
 343 /* lock so many callers can add things, the context that owns the set
 344  * is supposed to notice these and move them into the set proper. */
 345 void ptlrpc_set_add_new_req(struct ptlrpc_request_set *set,
 346                             struct ptlrpc_request *req)
 347 {
 348         unsigned long flags;
 349         spin_lock_irqsave(&set->set_new_req_lock, flags);
 350         /* The set takes over the caller's request reference */
 351         list_add_tail(&req->rq_set_chain, &set->set_new_requests);
 352         req->rq_set = set;
 353         spin_unlock_irqrestore(&set->set_new_req_lock, flags);
 354 }
 355
 356 /*
 357  * Based on the current state of the import, determine if the request
 358  * can be sent, is an error, or should be delayed.
 359  *
 360  * Returns true if this request should be delayed. If false, and
 361  * *status is set, then the request can not be sent and *status is the
 362  * error code.  If false and status is 0, then request can be sent.
 363  *
 364  * The imp->imp_lock must be held.
 365  */
 366 static int ptlrpc_import_delay_req(struct obd_import *imp,
 367                                    struct ptlrpc_request *req, int *status)
 368 {
 369         int delay = 0;
 370         ENTRY;
 371
 372         LASSERT (status != NULL);
 373         *status = 0;
 374
 375         /* A new import, or one that has been cleaned up.
 376          */
 377         if (imp->imp_state == LUSTRE_IMP_NEW) {
 378                 DEBUG_REQ(D_ERROR, req, "Uninitialized import.");
 379                 *status = -EIO;
 380         }
 381         /*
 382          * If the import has been invalidated (such as by an OST failure), the
 383          * request must fail with -EIO.
 384          */
 385         else if (imp->imp_invalid) {
 386                 DEBUG_REQ(D_ERROR, req, "IMP_INVALID");
 387                 *status = -EIO;
 388         }
 389         else if (req->rq_import_generation != imp->imp_generation) {
 390                 DEBUG_REQ(D_ERROR, req, "req wrong generation:");
 391                 *status = -EIO;
 392         }
 393         else if (req->rq_send_state != imp->imp_state) {
 394                 if (imp->imp_obd->obd_no_recov || imp->imp_dlm_fake)
 395                         *status = -EWOULDBLOCK;
 396                 else
 397                         delay = 1;
 398         }
 399
 400         RETURN(delay);
 401 }
 402
 403 static int ptlrpc_check_reply(struct ptlrpc_request *req)
 404 {
 405         unsigned long flags;
 406         int rc = 0;
 407         ENTRY;
 408
 409         /* serialise with network callback */
 410         spin_lock_irqsave (&req->rq_lock, flags);
 411
 412         if (req->rq_replied) {
 413                 DEBUG_REQ(D_NET, req, "REPLIED:");
 414                 GOTO(out, rc = 1);
 415         }
 416
 417         if (req->rq_err) {
 418                 DEBUG_REQ(D_ERROR, req, "ABORTED:");
 419                 GOTO(out, rc = 1);
 420         }
 421
 422         if (req->rq_resend) {
 423                 DEBUG_REQ(D_ERROR, req, "RESEND:");
 424                 GOTO(out, rc = 1);
 425         }
 426
 427         if (req->rq_restart) {
 428                 DEBUG_REQ(D_ERROR, req, "RESTART:");
 429                 GOTO(out, rc = 1);
 430         }
 431         EXIT;
 432  out:
 433         spin_unlock_irqrestore (&req->rq_lock, flags);
 434         DEBUG_REQ(D_NET, req, "rc = %d for", rc);
 435         return rc;
 436 }
 437
 438 static int ptlrpc_check_status(struct ptlrpc_request *req)
 439 {
 440         int err;
 441         ENTRY;
 442
 443         err = req->rq_repmsg->status;
 444         if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 445                 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR");
 446                 RETURN(err < 0 ? err : -EINVAL);
 447         }
 448
 449         if (err < 0) {
 450                 DEBUG_REQ(D_INFO, req, "status is %d", err);
 451         } else if (err > 0) {
 452                 /* XXX: translate this error from net to host */
 453                 DEBUG_REQ(D_INFO, req, "status is %d", err);
 454         }
 455
 456         RETURN(err);
 457 }
 458
 459 static int after_reply(struct ptlrpc_request *req, int *restartp)
 460 {
 461         unsigned long flags;
 462         struct obd_import *imp = req->rq_import;
 463         int rc;
 464         ENTRY;
 465
 466         LASSERT(!req->rq_receiving_reply);
 467         LASSERT(req->rq_replied);
 468
 469         if (restartp != NULL)
 470                 *restartp = 0;
 471
 472         /* NB Until this point, the whole of the incoming message,
 473          * including buflens, status etc is in the sender's byte order. */
 474
 475 #if SWAB_PARANOIA
 476         /* Clear reply swab mask; this is a new reply in sender's byte order */
 477         req->rq_rep_swab_mask = 0;
 478 #endif
 479         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
 480         if (rc) {
 481                 CERROR("unpack_rep failed: %d\n", rc);
 482                 RETURN(-EPROTO);
 483         }
 484
 485         if (req->rq_repmsg->type != PTL_RPC_MSG_REPLY &&
 486             req->rq_repmsg->type != PTL_RPC_MSG_ERR) {
 487                 CERROR("invalid packet type received (type=%u)\n",
 488                        req->rq_repmsg->type);
 489                 RETURN(-EPROTO);
 490         }
 491
 492         /* Store transno in reqmsg for replay. */
 493         req->rq_reqmsg->transno = req->rq_transno = req->rq_repmsg->transno;
 494
 495         rc = ptlrpc_check_status(req);
 496
 497         /* Either we've been evicted, or the server has failed for
 498          * some reason. Try to reconnect, and if that fails, punt to the
 499          * upcall. */
 500         if (rc == -ENOTCONN) {
 501                 if (req->rq_send_state != LUSTRE_IMP_FULL ||
 502                     imp->imp_obd->obd_no_recov || imp->imp_dlm_fake) {
 503                         RETURN(-ENOTCONN);
 504                 }
 505
 506                 ptlrpc_request_handle_notconn(req);
 507
 508                 if (req->rq_err)
 509                         RETURN(-EIO);
 510
 511                 if (req->rq_no_resend)
 512                         RETURN(rc); /* -ENOTCONN */
 513
 514                 if (req->rq_resend) {
 515                         if (restartp == NULL)
 516                                 LBUG(); /* async resend not supported yet */
 517                         spin_lock_irqsave (&req->rq_lock, flags);
 518                         req->rq_resend = 0;
 519                         spin_unlock_irqrestore (&req->rq_lock, flags);
 520                         *restartp = 1;
 521                         lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
 522                         DEBUG_REQ(D_HA, req, "resending: ");
 523                         RETURN(0);
 524                 }
 525
 526                 CERROR("request should be err or resend: %p\n", req);
 527                 LBUG();
 528         }
 529
 530         if (req->rq_import->imp_replayable) {
 531                 spin_lock_irqsave(&imp->imp_lock, flags);
 532                 if (req->rq_replay || req->rq_transno != 0)
 533                         ptlrpc_retain_replayable_request(req, imp);
 534                 else if (req->rq_commit_cb != NULL)
 535                         req->rq_commit_cb(req);
 536
 537                 if (req->rq_transno > imp->imp_max_transno)
 538                         imp->imp_max_transno = req->rq_transno;
 539
 540                 /* Replay-enabled imports return commit-status information. */
 541                 if (req->rq_repmsg->last_committed)
 542                         imp->imp_peer_committed_transno =
 543                                 req->rq_repmsg->last_committed;
 544                 ptlrpc_free_committed(imp);
 545                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 546         }
 547
 548         RETURN(rc);
 549 }
 550
 551 static int ptlrpc_send_new_req(struct ptlrpc_request *req)
 552 {
 553         struct obd_import     *imp;
 554         unsigned long          flags;
 555         int rc;
 556         ENTRY;
 557
 558         LASSERT(req->rq_send_state == LUSTRE_IMP_FULL);
 559         LASSERT(req->rq_phase == RQ_PHASE_NEW);
 560         req->rq_phase = RQ_PHASE_RPC;
 561
 562         imp = req->rq_import;
 563         spin_lock_irqsave(&imp->imp_lock, flags);
 564
 565         if (imp->imp_invalid) {
 566                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 567                 req->rq_status = -EIO;
 568                 req->rq_phase = RQ_PHASE_INTERPRET;
 569                 RETURN(-EIO);
 570         }
 571
 572         req->rq_import_generation = imp->imp_generation;
 573
 574         if (ptlrpc_import_delay_req(imp, req, &rc)) {
 575                 spin_lock (&req->rq_lock);
 576                 req->rq_waiting = 1;
 577                 spin_unlock (&req->rq_lock);
 578
 579                 LASSERT(list_empty (&req->rq_list));
 580
 581                 // list_del(&req->rq_list);
 582                 list_add_tail(&req->rq_list, &imp->imp_delayed_list);
 583                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 584                 RETURN(0);
 585         }
 586
 587         if (rc != 0) {
 588                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 589                 req->rq_status = rc;
 590                 req->rq_phase = RQ_PHASE_INTERPRET;
 591                 RETURN(rc);
 592         }
 593
 594         /* XXX this is the same as ptlrpc_queue_wait */
 595         LASSERT(list_empty(&req->rq_list));
 596         list_add_tail(&req->rq_list, &imp->imp_sending_list);
 597         spin_unlock_irqrestore(&imp->imp_lock, flags);
 598
 599         req->rq_reqmsg->status = current->pid;
 600         CDEBUG(D_RPCTRACE, "Sending RPC pname:cluuid:pid:xid:ni:nid:opc"
 601                " %s:%s:%d:"LPU64":%s:"LPX64":%d\n", current->comm,
 602                imp->imp_obd->obd_uuid.uuid, req->rq_reqmsg->status,
 603                req->rq_xid,
 604                imp->imp_connection->c_peer.peer_ni->pni_name,
 605                imp->imp_connection->c_peer.peer_nid,
 606                req->rq_reqmsg->opc);
 607
 608         rc = ptl_send_rpc(req);
 609         if (rc) {
 610                 DEBUG_REQ(D_HA, req, "send failed (%d); expect timeout", rc);
 611                 req->rq_timeout = 1;
 612                 RETURN(rc);
 613         }
 614         RETURN(0);
 615 }
 616
 617 int ptlrpc_check_set(struct ptlrpc_request_set *set)
 618 {
 619         unsigned long flags;
 620         struct list_head *tmp;
 621         int sending_error = 0;
 622         ENTRY;
 623
 624         if (set->set_remaining == 0)
 625                 RETURN(1);
 626
 627         list_for_each(tmp, &set->set_requests) {
 628                 struct ptlrpc_request *req =
 629                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 630                 struct obd_import *imp = req->rq_import;
 631                 int rc = 0;
 632
 633                 if (req->rq_phase == RQ_PHASE_NEW &&
 634                     ptlrpc_send_new_req(req)) {
 635                         sending_error = 1;
 636                 }
 637
 638                 if (!(req->rq_phase == RQ_PHASE_RPC ||
 639                       req->rq_phase == RQ_PHASE_BULK ||
 640                       req->rq_phase == RQ_PHASE_INTERPRET ||
 641                       req->rq_phase == RQ_PHASE_COMPLETE)) {
 642                         DEBUG_REQ(D_ERROR, req, "bad phase %x", req->rq_phase);
 643                         LBUG();
 644                 }
 645
 646                 if (req->rq_phase == RQ_PHASE_COMPLETE)
 647                         continue;
 648
 649                 if (req->rq_phase == RQ_PHASE_INTERPRET)
 650                         GOTO(interpret, req->rq_status);
 651
 652                 if (req->rq_err) {
 653                         ptlrpc_unregister_reply(req);
 654                         if (req->rq_status == 0)
 655                                 req->rq_status = -EIO;
 656                         req->rq_phase = RQ_PHASE_INTERPRET;
 657
 658                         spin_lock_irqsave(&imp->imp_lock, flags);
 659                         list_del_init(&req->rq_list);
 660                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 661
 662                         GOTO(interpret, req->rq_status);
 663                 }
 664
 665                 if (req->rq_intr) {
 666                         /* NB could be on delayed list */
 667                         ptlrpc_unregister_reply(req);
 668                         req->rq_status = -EINTR;
 669                         req->rq_phase = RQ_PHASE_INTERPRET;
 670
 671                         spin_lock_irqsave(&imp->imp_lock, flags);
 672                         list_del_init(&req->rq_list);
 673                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 674
 675                         GOTO(interpret, req->rq_status);
 676                 }
 677
 678                 if (req->rq_phase == RQ_PHASE_RPC) {
 679                         int do_restart = 0;
 680                         if (req->rq_waiting || req->rq_resend) {
 681                                 int status;
 682                                 spin_lock_irqsave(&imp->imp_lock, flags);
 683
 684                                 if (ptlrpc_import_delay_req(imp, req, &status)) {
 685                                         spin_unlock_irqrestore(&imp->imp_lock,
 686                                                                flags);
 687                                         continue;
 688                                 }
 689
 690                                 list_del(&req->rq_list);
 691                                 list_add_tail(&req->rq_list,
 692                                               &imp->imp_sending_list);
 693
 694                                 if (status != 0)  {
 695                                         req->rq_status = status;
 696                                         req->rq_phase = RQ_PHASE_INTERPRET;
 697                                         spin_unlock_irqrestore(&imp->imp_lock,
 698                                                                flags);
 699                                         GOTO(interpret, req->rq_status);
 700                                 }
 701                                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 702
 703                                 req->rq_waiting = 0;
 704                                 if (req->rq_resend) {
 705                                         lustre_msg_add_flags(req->rq_reqmsg,
 706                                                              MSG_RESENT);
 707                                         spin_lock_irqsave(&req->rq_lock, flags);
 708                                         req->rq_resend = 0;
 709                                         spin_unlock_irqrestore(&req->rq_lock,
 710                                                                flags);
 711
 712                                         ptlrpc_unregister_reply(req);
 713                                         if (req->rq_bulk) {
 714                                                 __u64 old_xid = req->rq_xid;
 715                                                 ptlrpc_unregister_bulk(req);
 716                                                 /* ensure previous bulk fails */
 717                                                 req->rq_xid = ptlrpc_next_xid();
 718                                                 CDEBUG(D_HA, "resend bulk "
 719                                                        "old x"LPU64
 720                                                        " new x"LPU64"\n",
 721                                                        old_xid, req->rq_xid);
 722                                         }
 723                                 }
 724
 725                                 rc = ptl_send_rpc(req);
 726                                 if (rc) {
 727                                         DEBUG_REQ(D_HA, req, "send failed (%d)",
 728                                                   rc);
 729                                         sending_error = 1;
 730                                         req->rq_timeout = 0;
 731                                 }
 732                         }
 733
 734                         /* Ensure the network callback returned */
 735                         spin_lock_irqsave (&req->rq_lock, flags);
 736                         if (!req->rq_replied) {
 737                                 spin_unlock_irqrestore (&req->rq_lock, flags);
 738                                 continue;
 739                         }
 740                         spin_unlock_irqrestore (&req->rq_lock, flags);
 741
 742                         spin_lock_irqsave(&imp->imp_lock, flags);
 743                         list_del_init(&req->rq_list);
 744                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 745
 746                         req->rq_status = after_reply(req, &do_restart);
 747                         if (do_restart) {
 748                                 spin_lock_irqsave (&req->rq_lock, flags);
 749                                 req->rq_resend = 1; /* ugh */
 750                                 spin_unlock_irqrestore (&req->rq_lock, flags);
 751                                 continue;
 752                         }
 753
 754                         /* If there is no bulk associated with this request,
 755                          * then we're done and should let the interpreter
 756                          * process the reply.  Similarly if the RPC returned
 757                          * an error, and therefore the bulk will never arrive.
 758                          */
 759                         if (req->rq_bulk == NULL || req->rq_status != 0) {
 760                                 req->rq_phase = RQ_PHASE_INTERPRET;
 761                                 GOTO(interpret, req->rq_status);
 762                         }
 763
 764                         req->rq_phase = RQ_PHASE_BULK;
 765                 }
 766
 767                 LASSERT(req->rq_phase == RQ_PHASE_BULK);
 768                 if (!ptlrpc_bulk_complete (req->rq_bulk))
 769                         continue;
 770
 771                 req->rq_phase = RQ_PHASE_INTERPRET;
 772
 773         interpret:
 774                 LASSERT(req->rq_phase == RQ_PHASE_INTERPRET);
 775                 LASSERT(!req->rq_receiving_reply);
 776
 777                 ptlrpc_unregister_reply(req);
 778                 if (req->rq_bulk != NULL)
 779                         ptlrpc_unregister_bulk (req);
 780
 781                 if (req->rq_interpret_reply != NULL) {
 782                         int (*interpreter)(struct ptlrpc_request *,void *,int) =
 783                                 req->rq_interpret_reply;
 784                         req->rq_status = interpreter(req, &req->rq_async_args,
 785                                                      req->rq_status);
 786                 }
 787
 788                 CDEBUG(D_RPCTRACE, "Completed RPC pname:cluuid:pid:xid:ni:nid:"
 789                        "opc %s:%s:%d:"LPU64":%s:"LPX64":%d\n", current->comm,
 790                        imp->imp_obd->obd_uuid.uuid, req->rq_reqmsg->status,
 791                        req->rq_xid,
 792                        imp->imp_connection->c_peer.peer_ni->pni_name,
 793                        imp->imp_connection->c_peer.peer_nid,
 794                        req->rq_reqmsg->opc);
 795
 796                 req->rq_phase = RQ_PHASE_COMPLETE;
 797                 set->set_remaining--;
 798         }
 799
 800         /* If we hit an error, we want to recover promptly. */
 801         RETURN(set->set_remaining == 0 || sending_error);
 802 }
 803
 804 int ptlrpc_expire_one_request(struct ptlrpc_request *req)
 805 {
 806         unsigned long      flags;
 807         struct obd_import *imp = req->rq_import;
 808         ENTRY;
 809
 810         DEBUG_REQ(D_ERROR, req, "timeout");
 811
 812         spin_lock_irqsave (&req->rq_lock, flags);
 813         req->rq_timedout = 1;
 814         spin_unlock_irqrestore (&req->rq_lock, flags);
 815
 816         ptlrpc_unregister_reply (req);
 817
 818         if (imp == NULL) {
 819                 DEBUG_REQ(D_HA, req, "NULL import: already cleaned up?");
 820                 RETURN(1);
 821         }
 822
 823         /* The DLM server doesn't want recovery run on its imports. */
 824         if (imp->imp_dlm_fake)
 825                 RETURN(1);
 826
 827         /* If this request is for recovery or other primordial tasks,
 828          * don't go back to sleep, and don't start recovery again.. */
 829         if (req->rq_send_state != LUSTRE_IMP_FULL || imp->imp_obd->obd_no_recov)
 830                 RETURN(1);
 831
 832         ptlrpc_fail_import(imp, req->rq_import_generation);
 833
 834         RETURN(0);
 835 }
 836
 837 int ptlrpc_expired_set(void *data)
 838 {
 839         struct ptlrpc_request_set *set = data;
 840         struct list_head          *tmp;
 841         time_t                     now = LTIME_S (CURRENT_TIME);
 842         ENTRY;
 843
 844         LASSERT(set != NULL);
 845
 846         /* A timeout expired; see which reqs it applies to... */
 847         list_for_each (tmp, &set->set_requests) {
 848                 struct ptlrpc_request *req =
 849                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 850
 851                 /* request in-flight? */
 852                 if (!((req->rq_phase == RQ_PHASE_RPC && !req->rq_waiting) ||
 853                       (req->rq_phase == RQ_PHASE_BULK)))
 854                         continue;
 855
 856                 if (req->rq_timedout ||           /* already dealt with */
 857                     req->rq_sent + req->rq_timeout > now) /* not expired */
 858                         continue;
 859
 860                 /* deal with this guy */
 861                 ptlrpc_expire_one_request (req);
 862         }
 863
 864         /* When waiting for a whole set, we always to break out of the
 865          * sleep so we can recalculate the timeout, or enable interrupts
 866          * iff everyone's timed out.
 867          */
 868         RETURN(1);
 869 }
 870
 871 void ptlrpc_interrupted_set(void *data)
 872 {
 873         struct ptlrpc_request_set *set = data;
 874         struct list_head *tmp;
 875         unsigned long flags;
 876
 877         LASSERT(set != NULL);
 878         CERROR("INTERRUPTED SET %p\n", set);
 879
 880         list_for_each(tmp, &set->set_requests) {
 881                 struct ptlrpc_request *req =
 882                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 883
 884                 if (req->rq_phase != RQ_PHASE_RPC)
 885                         continue;
 886
 887                 spin_lock_irqsave (&req->rq_lock, flags);
 888                 req->rq_intr = 1;
 889                 spin_unlock_irqrestore (&req->rq_lock, flags);
 890         }
 891 }
 892
 893 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *set)
 894 {
 895         struct list_head      *tmp;
 896         time_t                 now = LTIME_S(CURRENT_TIME);
 897         time_t                 deadline;
 898         int                    timeout = 0;
 899         struct ptlrpc_request *req;
 900         ENTRY;
 901
 902         SIGNAL_MASK_ASSERT(); /* XXX BUG 1511 */
 903
 904         list_for_each(tmp, &set->set_requests) {
 905                 req = list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 906
 907                 /* request in-flight? */
 908                 if (!((req->rq_phase == RQ_PHASE_RPC && !req->rq_waiting) ||
 909                       (req->rq_phase == RQ_PHASE_BULK)))
 910                         continue;
 911
 912                 if (req->rq_timedout)   /* already timed out */
 913                         continue;
 914
 915                 deadline = req->rq_sent + req->rq_timeout;
 916                 if (deadline <= now)    /* actually expired already */
 917                         timeout = 1;    /* ASAP */
 918                 else if (timeout == 0 || timeout > deadline - now)
 919                         timeout = deadline - now;
 920         }
 921         RETURN(timeout);
 922 }
 923
 924
 925 int ptlrpc_set_wait(struct ptlrpc_request_set *set)
 926 {
 927         struct list_head      *tmp;
 928         struct ptlrpc_request *req;
 929         struct l_wait_info     lwi;
 930         int                    rc, timeout;
 931         ENTRY;
 932
 933         LASSERT(!list_empty(&set->set_requests));
 934         list_for_each(tmp, &set->set_requests) {
 935                 req = list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 936                 (void)ptlrpc_send_new_req(req);
 937         }
 938
 939         do {
 940                 timeout = ptlrpc_set_next_timeout(set);
 941
 942                 /* wait until all complete, interrupted, or an in-flight
 943                  * req times out */
 944                 CDEBUG(D_HA, "set %p going to sleep for %d seconds\n",
 945                        set, timeout);
 946                 lwi = LWI_TIMEOUT_INTR((timeout ? timeout : 1) * HZ,
 947                                        ptlrpc_expired_set,
 948                                        ptlrpc_interrupted_set, set);
 949                 rc = l_wait_event(set->set_waitq, ptlrpc_check_set(set), &lwi);
 950
 951                 LASSERT(rc == 0 || rc == -EINTR || rc == -ETIMEDOUT);
 952
 953                 /* -EINTR => all requests have been flagged rq_intr so next
 954                  * check completes.
 955                  * -ETIMEOUTD => someone timed out.  When all reqs have
 956                  * timed out, signals are enabled allowing completion with
 957                  * EINTR.
 958                  * I don't really care if we go once more round the loop in
 959                  * the error cases -eeb. */
 960         } while (rc != 0);
 961
 962         LASSERT(set->set_remaining == 0);
 963
 964         rc = 0;
 965         list_for_each(tmp, &set->set_requests) {
 966                 req = list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 967
 968                 LASSERT(req->rq_phase == RQ_PHASE_COMPLETE);
 969                 if (req->rq_status != 0)
 970                         rc = req->rq_status;
 971         }
 972
 973         if (set->set_interpret != NULL) {
 974                 int (*interpreter)(struct ptlrpc_request_set *set,void *,int) =
 975                         set->set_interpret;
 976                 rc = interpreter (set, &set->set_args, rc);
 977         }
 978
 979         RETURN(rc);
 980 }
 981
 982 static void __ptlrpc_free_req(struct ptlrpc_request *request, int locked)
 983 {
 984         ENTRY;
 985         if (request == NULL) {
 986                 EXIT;
 987                 return;
 988         }
 989
 990         LASSERT(!request->rq_receiving_reply);
 991
 992         /* We must take it off the imp_replay_list first.  Otherwise, we'll set
 993          * request->rq_reqmsg to NULL while osc_close is dereferencing it. */
 994         if (request->rq_import != NULL) {
 995                 unsigned long flags = 0;
 996                 if (!locked)
 997                         spin_lock_irqsave(&request->rq_import->imp_lock, flags);
 998                 list_del_init(&request->rq_list);
 999                 if (!locked)
1000                         spin_unlock_irqrestore(&request->rq_import->imp_lock,
1001                                                flags);
1002         }
1003
1004         if (atomic_read(&request->rq_refcount) != 0) {
1005                 DEBUG_REQ(D_ERROR, request,
1006                           "freeing request with nonzero refcount");
1007                 LBUG();
1008         }
1009
1010         if (request->rq_repmsg != NULL) {
1011                 OBD_FREE(request->rq_repmsg, request->rq_replen);
1012                 request->rq_repmsg = NULL;
1013         }
1014         if (request->rq_reqmsg != NULL) {
1015                 OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
1016                 request->rq_reqmsg = NULL;
1017         }
1018         if (request->rq_export != NULL) {
1019                 class_export_put(request->rq_export);
1020                 request->rq_export = NULL;
1021         }
1022         if (request->rq_import != NULL) {
1023                 class_import_put(request->rq_import);
1024                 request->rq_import = NULL;
1025         }
1026         if (request->rq_bulk != NULL)
1027                 ptlrpc_free_bulk(request->rq_bulk);
1028
1029         ptlrpc_put_connection(request->rq_connection);
1030         OBD_FREE(request, sizeof(*request));
1031         EXIT;
1032 }
1033
1034 void ptlrpc_free_req(struct ptlrpc_request *request)
1035 {
1036         __ptlrpc_free_req(request, 0);
1037 }
1038
1039 static int __ptlrpc_req_finished(struct ptlrpc_request *request, int locked);
1040 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request)
1041 {
1042 #ifdef CONFIG_SMP
1043         LASSERT(spin_is_locked(&request->rq_import->imp_lock));
1044 #endif
1045         (void)__ptlrpc_req_finished(request, 1);
1046 }
1047
1048 static int __ptlrpc_req_finished(struct ptlrpc_request *request, int locked)
1049 {
1050         ENTRY;
1051         if (request == NULL)
1052                 RETURN(1);
1053
1054         if (request == (void *)(unsigned long)(0x5a5a5a5a5a5a5a5a) ||
1055             request->rq_reqmsg == (void *)(unsigned long)(0x5a5a5a5a5a5a5a5a)) {
1056                 CERROR("dereferencing freed request (bug 575)\n");
1057                 LBUG();
1058                 RETURN(1);
1059         }
1060
1061         DEBUG_REQ(D_INFO, request, "refcount now %u",
1062                   atomic_read(&request->rq_refcount) - 1);
1063
1064         if (atomic_dec_and_test(&request->rq_refcount)) {
1065                 __ptlrpc_free_req(request, locked);
1066                 RETURN(1);
1067         }
1068
1069         RETURN(0);
1070 }
1071
1072 void ptlrpc_req_finished(struct ptlrpc_request *request)
1073 {
1074         __ptlrpc_req_finished(request, 0);
1075 }
1076
1077 static void ptlrpc_cleanup_request_buf(struct ptlrpc_request *request)
1078 {
1079         OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
1080         request->rq_reqmsg = NULL;
1081         request->rq_reqlen = 0;
1082 }
1083
1084 /* Disengage the client's reply buffer from the network
1085  * NB does _NOT_ unregister any client-side bulk.
1086  * IDEMPOTENT, but _not_ safe against concurrent callers.
1087  * The request owner (i.e. the thread doing the I/O) must call...
1088  */
1089 void ptlrpc_unregister_reply (struct ptlrpc_request *request)
1090 {
1091         unsigned long flags;
1092         int           rc;
1093         ENTRY;
1094
1095         LASSERT(!in_interrupt ());             /* might sleep */
1096
1097         spin_lock_irqsave (&request->rq_lock, flags);
1098         if (!request->rq_receiving_reply) {     /* not waiting for a reply */
1099                 spin_unlock_irqrestore (&request->rq_lock, flags);
1100                 EXIT;
1101                 /* NB reply buffer not freed here */
1102                 return;
1103         }
1104
1105         LASSERT(!request->rq_replied);         /* callback hasn't completed */
1106         spin_unlock_irqrestore (&request->rq_lock, flags);
1107
1108         rc = PtlMDUnlink (request->rq_reply_md_h);
1109         switch (rc) {
1110         default:
1111                 LBUG ();
1112
1113         case PTL_OK:                            /* unlinked before completion */
1114                 LASSERT(request->rq_receiving_reply);
1115                 LASSERT(!request->rq_replied);
1116                 spin_lock_irqsave (&request->rq_lock, flags);
1117                 request->rq_receiving_reply = 0;
1118                 spin_unlock_irqrestore (&request->rq_lock, flags);
1119                 OBD_FREE(request->rq_repmsg, request->rq_replen);
1120                 request->rq_repmsg = NULL;
1121                 EXIT;
1122                 return;
1123
1124         case PTL_MD_INUSE:                      /* callback in progress */
1125                 for (;;) {
1126                         /* Network access will complete in finite time but
1127                          * the timeout lets us CERROR for visibility */
1128                         struct l_wait_info lwi = LWI_TIMEOUT(10*HZ, NULL, NULL);
1129
1130                         rc = l_wait_event (request->rq_wait_for_rep,
1131                                            request->rq_replied, &lwi);
1132                         LASSERT(rc == 0 || rc == -ETIMEDOUT);
1133                         if (rc == 0) {
1134                                 spin_lock_irqsave (&request->rq_lock, flags);
1135                                 /* Ensure the callback has completed scheduling
1136                                  * me and taken its hands off the request */
1137                                 spin_unlock_irqrestore(&request->rq_lock,flags);
1138                                 break;
1139                         }
1140
1141                         CERROR ("Unexpectedly long timeout: req %p\n", request);
1142                 }
1143                 /* fall through */
1144
1145         case PTL_INV_MD:                        /* callback completed */
1146                 LASSERT(!request->rq_receiving_reply);
1147                 LASSERT(request->rq_replied);
1148                 EXIT;
1149                 return;
1150         }
1151         /* Not Reached */
1152 }
1153
1154 /* caller must hold imp->imp_lock */
1155 void ptlrpc_free_committed(struct obd_import *imp)
1156 {
1157         struct list_head *tmp, *saved;
1158         struct ptlrpc_request *req;
1159         struct ptlrpc_request *last_req = NULL; /* temporary fire escape */
1160         ENTRY;
1161
1162         LASSERT(imp != NULL);
1163
1164 #ifdef CONFIG_SMP
1165         LASSERT(spin_is_locked(&imp->imp_lock));
1166 #endif
1167
1168         CDEBUG(D_HA, "%s: committing for last_committed "LPU64"\n",
1169                imp->imp_obd->obd_name, imp->imp_peer_committed_transno);
1170
1171         list_for_each_safe(tmp, saved, &imp->imp_replay_list) {
1172                 req = list_entry(tmp, struct ptlrpc_request, rq_list);
1173
1174                 /* XXX ok to remove when 1357 resolved - rread 05/29/03  */
1175                 LASSERT(req != last_req);
1176                 last_req = req;
1177
1178                 if (req->rq_import_generation < imp->imp_generation) {
1179                         DEBUG_REQ(D_HA, req, "freeing request with old gen");
1180                         GOTO(free_req, 0);
1181                 }
1182
1183                 if (req->rq_replay) {
1184                         DEBUG_REQ(D_HA, req, "keeping (FL_REPLAY)");
1185                         continue;
1186                 }
1187
1188                 /* not yet committed */
1189                 if (req->rq_transno > imp->imp_peer_committed_transno) {
1190                         DEBUG_REQ(D_HA, req, "stopping search");
1191                         break;
1192                 }
1193
1194                 DEBUG_REQ(D_HA, req, "committing (last_committed "LPU64")",
1195                           imp->imp_peer_committed_transno);
1196 free_req:
1197                 if (req->rq_commit_cb != NULL)
1198                         req->rq_commit_cb(req);
1199                 list_del_init(&req->rq_list);
1200                 __ptlrpc_req_finished(req, 1);
1201         }
1202
1203         EXIT;
1204         return;
1205 }
1206
1207 void ptlrpc_cleanup_client(struct obd_import *imp)
1208 {
1209         ENTRY;
1210         EXIT;
1211         return;
1212 }
1213
1214 void ptlrpc_resend_req(struct ptlrpc_request *req)
1215 {
1216         unsigned long flags;
1217
1218         DEBUG_REQ(D_HA, req, "resending");
1219         req->rq_reqmsg->handle.cookie = 0;
1220         ptlrpc_put_connection(req->rq_connection);
1221         req->rq_connection =
1222                 ptlrpc_connection_addref(req->rq_import->imp_connection);
1223         req->rq_status = -EAGAIN;
1224
1225         spin_lock_irqsave (&req->rq_lock, flags);
1226         req->rq_resend = 1;
1227         req->rq_timedout = 0;
1228         if (req->rq_set != NULL)
1229                 wake_up (&req->rq_set->set_waitq);
1230         else
1231                 wake_up(&req->rq_wait_for_rep);
1232         spin_unlock_irqrestore (&req->rq_lock, flags);
1233 }
1234
1235 /* XXX: this function and rq_status are currently unused */
1236 void ptlrpc_restart_req(struct ptlrpc_request *req)
1237 {
1238         unsigned long flags;
1239
1240         DEBUG_REQ(D_HA, req, "restarting (possibly-)completed request");
1241         req->rq_status = -ERESTARTSYS;
1242
1243         spin_lock_irqsave (&req->rq_lock, flags);
1244         req->rq_restart = 1;
1245         req->rq_timedout = 0;
1246         if (req->rq_set != NULL)
1247                 wake_up (&req->rq_set->set_waitq);
1248         else
1249                 wake_up(&req->rq_wait_for_rep);
1250         spin_unlock_irqrestore (&req->rq_lock, flags);
1251 }
1252
1253 static int expired_request(void *data)
1254 {
1255         struct ptlrpc_request *req = data;
1256         ENTRY;
1257
1258         RETURN(ptlrpc_expire_one_request(req));
1259 }
1260
1261 static void interrupted_request(void *data)
1262 {
1263         unsigned long flags;
1264
1265         struct ptlrpc_request *req = data;
1266         DEBUG_REQ(D_HA, req, "request interrupted");
1267         spin_lock_irqsave (&req->rq_lock, flags);
1268         req->rq_intr = 1;
1269         spin_unlock_irqrestore (&req->rq_lock, flags);
1270 }
1271
1272 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req)
1273 {
1274         ENTRY;
1275         atomic_inc(&req->rq_refcount);
1276         RETURN(req);
1277 }
1278
1279 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
1280                                       struct obd_import *imp)
1281 {
1282         struct list_head *tmp;
1283
1284 #ifdef CONFIG_SMP
1285         LASSERT(spin_is_locked(&imp->imp_lock));
1286 #endif
1287
1288         LASSERT(imp->imp_replayable);
1289         /* Balanced in ptlrpc_free_committed, usually. */
1290         ptlrpc_request_addref(req);
1291         list_for_each_prev(tmp, &imp->imp_replay_list) {
1292                 struct ptlrpc_request *iter =
1293                         list_entry(tmp, struct ptlrpc_request, rq_list);
1294
1295                 /* We may have duplicate transnos if we create and then
1296                  * open a file, or for closes retained if to match creating
1297                  * opens, so use req->rq_xid as a secondary key.
1298                  * (See bugs 684, 685, and 428.)
1299                  * XXX no longer needed, but all opens need transnos!
1300                  */
1301                 if (iter->rq_transno > req->rq_transno)
1302                         continue;
1303
1304                 if (iter->rq_transno == req->rq_transno) {
1305                         LASSERT(iter->rq_xid != req->rq_xid);
1306                         if (iter->rq_xid > req->rq_xid)
1307                                 continue;
1308                 }
1309
1310                 list_add(&req->rq_list, &iter->rq_list);
1311                 return;
1312         }
1313
1314         list_add_tail(&req->rq_list, &imp->imp_replay_list);
1315 }
1316
1317 int ptlrpc_queue_wait(struct ptlrpc_request *req)
1318 {
1319         int rc = 0;
1320         int brc;
1321         struct l_wait_info lwi;
1322         struct obd_import *imp = req->rq_import;
1323         unsigned long flags;
1324         int do_restart = 0;
1325         int timeout = 0;
1326         ENTRY;
1327
1328         LASSERT(req->rq_set == NULL);
1329         LASSERT(!req->rq_receiving_reply);
1330
1331         /* for distributed debugging */
1332         req->rq_reqmsg->status = current->pid;
1333         LASSERT(imp->imp_obd != NULL);
1334         CDEBUG(D_RPCTRACE, "Sending RPC pname:cluuid:pid:xid:ni:nid:opc "
1335                "%s:%s:%d:"LPU64":%s:"LPX64":%d\n", current->comm,
1336                imp->imp_obd->obd_uuid.uuid,
1337                req->rq_reqmsg->status, req->rq_xid,
1338                imp->imp_connection->c_peer.peer_ni->pni_name,
1339                imp->imp_connection->c_peer.peer_nid,
1340                req->rq_reqmsg->opc);
1341
1342         /* Mark phase here for a little debug help */
1343         req->rq_phase = RQ_PHASE_RPC;
1344
1345         spin_lock_irqsave(&imp->imp_lock, flags);
1346         req->rq_import_generation = imp->imp_generation;
1347 restart:
1348         if (ptlrpc_import_delay_req(imp, req, &rc)) {
1349                 list_del(&req->rq_list);
1350
1351                 list_add_tail(&req->rq_list, &imp->imp_delayed_list);
1352                 spin_unlock_irqrestore(&imp->imp_lock, flags);
1353
1354                 DEBUG_REQ(D_HA, req, "\"%s\" waiting for recovery: (%d > %d)",
1355                           current->comm, req->rq_send_state, imp->imp_state);
1356                 lwi = LWI_INTR(interrupted_request, req);
1357                 rc = l_wait_event(req->rq_wait_for_rep,
1358                                   (req->rq_send_state == imp->imp_state ||
1359                                    req->rq_err),
1360                                   &lwi);
1361                 DEBUG_REQ(D_HA, req, "\"%s\" awake: (%d > %d or %d == 1)",
1362                           current->comm, imp->imp_state, req->rq_send_state,
1363                           req->rq_err);
1364
1365                 spin_lock_irqsave(&imp->imp_lock, flags);
1366                 list_del_init(&req->rq_list);
1367
1368                 if (req->rq_err) {
1369                         rc = -EIO;
1370                 }
1371                 else if (req->rq_intr) {
1372                         rc = -EINTR;
1373                 }
1374                 else {
1375                         GOTO(restart, rc);
1376                 }
1377         }
1378
1379         if (rc != 0) {
1380                 list_del_init(&req->rq_list);
1381                 spin_unlock_irqrestore(&imp->imp_lock, flags);
1382                 req->rq_status = rc; // XXX this ok?
1383                 GOTO(out, rc);
1384         }
1385
1386         /* XXX this is the same as ptlrpc_set_wait */
1387         LASSERT(list_empty(&req->rq_list));
1388         list_add_tail(&req->rq_list, &imp->imp_sending_list);
1389         spin_unlock_irqrestore(&imp->imp_lock, flags);
1390
1391         rc = ptl_send_rpc(req);
1392         if (rc) {
1393                 DEBUG_REQ(D_HA, req, "send failed (%d); recovering", rc);
1394                 timeout = 1;
1395         } else {
1396                 timeout = MAX(req->rq_timeout * HZ, 1);
1397                 DEBUG_REQ(D_NET, req, "-- sleeping");
1398         }
1399         lwi = LWI_TIMEOUT_INTR(timeout, expired_request, interrupted_request,
1400                                req);
1401         l_wait_event(req->rq_wait_for_rep, ptlrpc_check_reply(req), &lwi);
1402         DEBUG_REQ(D_NET, req, "-- done sleeping");
1403
1404         CDEBUG(D_RPCTRACE, "Completed RPC pname:cluuid:pid:xid:ni:nid:opc "
1405                "%s:%s:%d:"LPU64":%s:"LPX64":%d\n", current->comm,
1406                imp->imp_obd->obd_uuid.uuid,
1407                req->rq_reqmsg->status, req->rq_xid,
1408                imp->imp_connection->c_peer.peer_ni->pni_name,
1409                imp->imp_connection->c_peer.peer_nid,
1410                req->rq_reqmsg->opc);
1411
1412         spin_lock_irqsave(&imp->imp_lock, flags);
1413         list_del_init(&req->rq_list);
1414         spin_unlock_irqrestore(&imp->imp_lock, flags);
1415
1416         /* If the reply was received normally, this just grabs the spinlock
1417          * (ensuring the reply callback has returned), sees that
1418          * req->rq_receiving_reply is clear and returns. */
1419         ptlrpc_unregister_reply (req);
1420
1421         if (req->rq_err)
1422                 GOTO(out, rc = -EIO);
1423
1424         /* Resend if we need to, unless we were interrupted. */
1425         if (req->rq_resend && !req->rq_intr) {
1426                 /* ...unless we were specifically told otherwise. */
1427                 if (req->rq_no_resend)
1428                         GOTO(out, rc = -ETIMEDOUT);
1429                 spin_lock_irqsave (&req->rq_lock, flags);
1430                 req->rq_resend = 0;
1431                 spin_unlock_irqrestore (&req->rq_lock, flags);
1432                 lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
1433
1434                 if (req->rq_bulk != NULL)
1435                         ptlrpc_unregister_bulk (req);
1436
1437                 DEBUG_REQ(D_HA, req, "resending: ");
1438                 spin_lock_irqsave(&imp->imp_lock, flags);
1439                 goto restart;
1440         }
1441
1442         if (req->rq_intr) {
1443                 /* Should only be interrupted if we timed out. */
1444                 if (!req->rq_timedout)
1445                         DEBUG_REQ(D_ERROR, req,
1446                                   "rq_intr set but rq_timedout not");
1447                 GOTO(out, rc = -EINTR);
1448         }
1449
1450         if (req->rq_timedout) {                 /* non-recoverable timeout */
1451                 GOTO(out, rc = -ETIMEDOUT);
1452         }
1453
1454         if (!req->rq_replied) {
1455                 /* How can this be? -eeb */
1456                 DEBUG_REQ(D_ERROR, req, "!rq_replied: ");
1457                 LBUG();
1458                 GOTO(out, rc = req->rq_status);
1459         }
1460
1461         rc = after_reply (req, &do_restart);
1462         /* NB may return +ve success rc */
1463         if (do_restart) {
1464                 if (req->rq_bulk != NULL)
1465                         ptlrpc_unregister_bulk (req);
1466                 DEBUG_REQ(D_HA, req, "resending: ");
1467                 spin_lock_irqsave(&imp->imp_lock, flags);
1468                 goto restart;
1469         }
1470
1471  out:
1472         if (req->rq_bulk != NULL) {
1473                 if (rc >= 0) {                  /* success so far */
1474                         lwi = LWI_TIMEOUT(timeout, NULL, NULL);
1475                         brc = l_wait_event(req->rq_wait_for_rep,
1476                                            ptlrpc_bulk_complete(req->rq_bulk),
1477                                            &lwi);
1478                         if (brc != 0) {
1479                                 LASSERT(brc == -ETIMEDOUT);
1480                                 CERROR ("Timed out waiting for bulk\n");
1481                                 rc = brc;
1482                         }
1483                 }
1484                 if (rc < 0)
1485                         ptlrpc_unregister_bulk (req);
1486         }
1487
1488         LASSERT(!req->rq_receiving_reply);
1489         req->rq_phase = RQ_PHASE_INTERPRET;
1490         RETURN(rc);
1491 }
1492
1493 int ptlrpc_replay_req(struct ptlrpc_request *req)
1494 {
1495         int rc = 0, old_state, old_status = 0;
1496         // struct ptlrpc_client *cli = req->rq_import->imp_client;
1497         struct l_wait_info lwi;
1498         ENTRY;
1499
1500         LASSERT(req->rq_import->imp_state == LUSTRE_IMP_REPLAY);
1501
1502         /* I don't touch rq_phase here, so the debug log can show what
1503          * state it was left in */
1504
1505         /* Not handling automatic bulk replay yet (or ever?) */
1506         LASSERT(req->rq_bulk == NULL);
1507
1508         DEBUG_REQ(D_NET, req, "about to replay");
1509
1510         /* Update request's state, since we might have a new connection. */
1511         ptlrpc_put_connection(req->rq_connection);
1512         req->rq_connection =
1513                 ptlrpc_connection_addref(req->rq_import->imp_connection);
1514
1515         /* temporarily set request to REPLAY level---not strictly
1516          * necessary since ptl_send_rpc doesn't check state, but let's
1517          * be consistent.*/
1518         old_state = req->rq_send_state;
1519
1520         /*
1521          * Q: "How can a req get on the replay list if it wasn't replied?"
1522          * A: "If we failed during the replay of this request, it will still
1523          *     be on the list, but rq_replied will have been reset to 0."
1524          */
1525         if (req->rq_replied)
1526                 old_status = req->rq_repmsg->status;
1527         req->rq_send_state = LUSTRE_IMP_REPLAY;
1528         rc = ptl_send_rpc(req);
1529         if (rc) {
1530                 CERROR("error %d, opcode %d\n", rc, req->rq_reqmsg->opc);
1531                 ptlrpc_cleanup_request_buf(req);
1532                 // up(&cli->cli_rpc_sem);
1533                 GOTO(out, rc = -rc);
1534         }
1535
1536         CDEBUG(D_OTHER, "-- sleeping\n");
1537         lwi = LWI_INTR(NULL, NULL); /* XXX needs timeout, nested recovery */
1538         l_wait_event(req->rq_wait_for_rep, ptlrpc_check_reply(req), &lwi);
1539         CDEBUG(D_OTHER, "-- done\n");
1540
1541         // up(&cli->cli_rpc_sem);
1542
1543         /* If the reply was received normally, this just grabs the spinlock
1544          * (ensuring the reply callback has returned), sees that
1545          * req->rq_receiving_reply is clear and returns. */
1546         ptlrpc_unregister_reply (req);
1547
1548         if (!req->rq_replied) {
1549                 CERROR("Unknown reason for wakeup\n");
1550                 /* XXX Phil - I end up here when I kill obdctl */
1551                 /* ...that's because signals aren't all masked in
1552                  * l_wait_event() -eeb */
1553                 GOTO(out, rc = -EINTR);
1554         }
1555
1556 #if SWAB_PARANOIA
1557         /* Clear reply swab mask; this is a new reply in sender's byte order */
1558         req->rq_rep_swab_mask = 0;
1559 #endif
1560         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
1561         if (rc) {
1562                 CERROR("unpack_rep failed: %d\n", rc);
1563                 GOTO(out, rc = -EPROTO);
1564         }
1565 #if 0
1566         /* FIXME: Enable when BlueArc makes new release */
1567         if (req->rq_repmsg->type != PTL_RPC_MSG_REPLY &&
1568             req->rq_repmsg->type != PTL_RPC_MSG_ERR) {
1569                 CERROR("invalid packet type received (type=%u)\n",
1570                        req->rq_repmsg->type);
1571                 GOTO(out, rc = -EPROTO);
1572         }
1573 #endif
1574
1575         if (req->rq_repmsg->type == PTL_RPC_MSG_ERR &&
1576             req->rq_repmsg->status == -ENOTCONN)
1577                 GOTO(out, rc = req->rq_repmsg->status);
1578
1579         /* The transno had better not change over replay. */
1580         LASSERT(req->rq_reqmsg->transno == req->rq_repmsg->transno);
1581
1582         CDEBUG(D_NET, "got rep "LPD64"\n", req->rq_xid);
1583
1584         /* let the callback do fixups, possibly including in the request */
1585         if (req->rq_replay_cb)
1586                 req->rq_replay_cb(req);
1587
1588         if (req->rq_replied && req->rq_repmsg->status != old_status) {
1589                 DEBUG_REQ(D_ERROR, req, "status %d, old was %d",
1590                           req->rq_repmsg->status, old_status);
1591         } else {
1592                 /* Put it back for re-replay. */
1593                 req->rq_status = old_status;
1594         }
1595
1596  out:
1597         req->rq_send_state = old_state;
1598         RETURN(rc);
1599 }
1600
1601 void ptlrpc_abort_inflight(struct obd_import *imp)
1602 {
1603         unsigned long flags;
1604         struct list_head *tmp, *n;
1605         ENTRY;
1606
1607         /* Make sure that no new requests get processed for this import.
1608          * ptlrpc_{queue,set}_wait must (and does) hold imp_lock while testing
1609          * this flag and then putting requests on sending_list or delayed_list.
1610          */
1611         spin_lock_irqsave(&imp->imp_lock, flags);
1612
1613         /* XXX locking?  Maybe we should remove each request with the list
1614          * locked?  Also, how do we know if the requests on the list are
1615          * being freed at this time?
1616          */
1617         list_for_each_safe(tmp, n, &imp->imp_sending_list) {
1618                 struct ptlrpc_request *req =
1619                         list_entry(tmp, struct ptlrpc_request, rq_list);
1620
1621                 DEBUG_REQ(D_HA, req, "inflight");
1622
1623                 spin_lock (&req->rq_lock);
1624                 if (req->rq_import_generation < imp->imp_generation) {
1625                         req->rq_err = 1;
1626                         if (req->rq_set != NULL)
1627                                 wake_up(&req->rq_set->set_waitq);
1628                         else
1629                                 wake_up(&req->rq_wait_for_rep);
1630                 }
1631                 spin_unlock (&req->rq_lock);
1632         }
1633
1634         list_for_each_safe(tmp, n, &imp->imp_delayed_list) {
1635                 struct ptlrpc_request *req =
1636                         list_entry(tmp, struct ptlrpc_request, rq_list);
1637
1638                 DEBUG_REQ(D_HA, req, "aborting waiting req");
1639
1640                 spin_lock (&req->rq_lock);
1641                 if (req->rq_import_generation < imp->imp_generation) {
1642                         req->rq_err = 1;
1643                         if (req->rq_set != NULL)
1644                                 wake_up(&req->rq_set->set_waitq);
1645                         else
1646                                 wake_up(&req->rq_wait_for_rep);
1647                 }
1648                 spin_unlock (&req->rq_lock);
1649         }
1650
1651         /* Last chance to free reqs left on the replay list, but we
1652          * will still leak reqs that haven't comitted.  */
1653         if (imp->imp_replayable)
1654                 ptlrpc_free_committed(imp);
1655
1656         spin_unlock_irqrestore(&imp->imp_lock, flags);
1657
1658         EXIT;
1659 }
1660
1661 static __u64 ptlrpc_last_xid = 0;
1662 static spinlock_t ptlrpc_last_xid_lock = SPIN_LOCK_UNLOCKED;
1663
1664 __u64 ptlrpc_next_xid(void)
1665 {
1666         __u64 tmp;
1667         spin_lock(&ptlrpc_last_xid_lock);
1668         tmp = ++ptlrpc_last_xid;
1669         spin_unlock(&ptlrpc_last_xid_lock);
1670         return tmp;
1671 }
1672
1673