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 ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid)
  47 {
  48         struct ptlrpc_connection *c;
  49         struct ptlrpc_peer peer;
  50         int err;
  51
  52         err = ptlrpc_uuid_to_peer(uuid, &peer);
  53         if (err != 0) {
  54                 CERROR("cannot find peer %s!\n", uuid->uuid);
  55                 return NULL;
  56         }
  57
  58         c = ptlrpc_get_connection(&peer, uuid);
  59         if (c) {
  60                 memcpy(c->c_remote_uuid.uuid,
  61                        uuid->uuid, sizeof(c->c_remote_uuid.uuid));
  62         }
  63
  64         CDEBUG(D_INFO, "%s -> %p\n", uuid->uuid, c);
  65
  66         return c;
  67 }
  68
  69 void ptlrpc_readdress_connection(struct ptlrpc_connection *conn,
  70                                  struct obd_uuid *uuid)
  71 {
  72         struct ptlrpc_peer peer;
  73         int err;
  74
  75         err = ptlrpc_uuid_to_peer(uuid, &peer);
  76         if (err != 0) {
  77                 CERROR("cannot find peer %s!\n", uuid->uuid);
  78                 return;
  79         }
  80
  81         memcpy(&conn->c_peer, &peer, sizeof (peer));
  82         return;
  83 }
  84
  85 static inline struct ptlrpc_bulk_desc *new_bulk(int npages, int type, int portal)
  86 {
  87         struct ptlrpc_bulk_desc *desc;
  88
  89         OBD_ALLOC(desc, offsetof (struct ptlrpc_bulk_desc, bd_iov[npages]));
  90         if (!desc)
  91                 return NULL;
  92
  93         spin_lock_init(&desc->bd_lock);
  94         init_waitqueue_head(&desc->bd_waitq);
  95         desc->bd_max_iov = npages;
  96         desc->bd_iov_count = 0;
  97         desc->bd_md_h = PTL_INVALID_HANDLE;
  98         desc->bd_portal = portal;
  99         desc->bd_type = type;
 100
 101         return desc;
 102 }
 103
 104 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
 105                                                int npages, int type, int portal)
 106 {
 107         struct obd_import *imp = req->rq_import;
 108         struct ptlrpc_bulk_desc *desc;
 109
 110         LASSERT(type == BULK_PUT_SINK || type == BULK_GET_SOURCE);
 111         desc = new_bulk(npages, type, portal);
 112         if (desc == NULL)
 113                 RETURN(NULL);
 114
 115         desc->bd_import_generation = req->rq_import_generation;
 116         desc->bd_import = class_import_get(imp);
 117         desc->bd_req = req;
 118
 119         desc->bd_cbid.cbid_fn  = client_bulk_callback;
 120         desc->bd_cbid.cbid_arg = desc;
 121
 122         /* This makes req own desc, and free it when she frees herself */
 123         req->rq_bulk = desc;
 124
 125         return desc;
 126 }
 127
 128 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp (struct ptlrpc_request *req,
 129                                                int npages, int type, int portal)
 130 {
 131         struct obd_export *exp = req->rq_export;
 132         struct ptlrpc_bulk_desc *desc;
 133
 134         LASSERT(type == BULK_PUT_SOURCE || type == BULK_GET_SINK);
 135
 136         desc = new_bulk(npages, type, portal);
 137         if (desc == NULL)
 138                 RETURN(NULL);
 139
 140         desc->bd_export = class_export_get(exp);
 141         desc->bd_req = req;
 142
 143         desc->bd_cbid.cbid_fn  = server_bulk_callback;
 144         desc->bd_cbid.cbid_arg = desc;
 145
 146         /* NB we don't assign rq_bulk here; server-side requests are
 147          * re-used, and the handler frees the bulk desc explicitly. */
 148
 149         return desc;
 150 }
 151
 152 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
 153                            struct page *page, int pageoffset, int len)
 154 {
 155         LASSERT(desc->bd_iov_count < desc->bd_max_iov);
 156         LASSERT(page != NULL);
 157         LASSERT(pageoffset >= 0);
 158         LASSERT(len > 0);
 159         LASSERT(pageoffset + len <= PAGE_SIZE);
 160
 161         desc->bd_nob += len;
 162
 163         ptlrpc_add_bulk_page(desc, page, pageoffset, len);
 164 }
 165
 166 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *desc)
 167 {
 168         ENTRY;
 169
 170         LASSERT(desc != NULL);
 171         LASSERT(desc->bd_iov_count != LI_POISON); /* not freed already */
 172         LASSERT(!desc->bd_network_rw);         /* network hands off or */
 173         LASSERT((desc->bd_export != NULL) ^ (desc->bd_import != NULL));
 174         if (desc->bd_export)
 175                 class_export_put(desc->bd_export);
 176         else
 177                 class_import_put(desc->bd_import);
 178
 179         OBD_FREE(desc, offsetof(struct ptlrpc_bulk_desc,
 180                                 bd_iov[desc->bd_max_iov]));
 181         EXIT;
 182 }
 183
 184 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, int opcode,
 185                                        int count, int *lengths, char **bufs)
 186 {
 187         struct ptlrpc_request *request;
 188         int rc;
 189         ENTRY;
 190
 191         LASSERT((unsigned long)imp > 0x1000);
 192
 193         OBD_ALLOC(request, sizeof(*request));
 194         if (!request) {
 195                 CERROR("request allocation out of memory\n");
 196                 RETURN(NULL);
 197         }
 198
 199         rc = lustre_pack_request(request, count, lengths, bufs);
 200         if (rc) {
 201                 CERROR("cannot pack request %d\n", rc);
 202                 OBD_FREE(request, sizeof(*request));
 203                 RETURN(NULL);
 204         }
 205
 206         request->rq_timeout = obd_timeout;
 207         request->rq_send_state = LUSTRE_IMP_FULL;
 208         request->rq_type = PTL_RPC_MSG_REQUEST;
 209         request->rq_import = class_import_get(imp);
 210
 211         request->rq_req_cbid.cbid_fn  = request_out_callback;
 212         request->rq_req_cbid.cbid_arg = request;
 213
 214         request->rq_reply_cbid.cbid_fn  = reply_in_callback;
 215         request->rq_reply_cbid.cbid_arg = request;
 216
 217         request->rq_phase = RQ_PHASE_NEW;
 218
 219         /* XXX FIXME bug 249 */
 220         request->rq_request_portal = imp->imp_client->cli_request_portal;
 221         request->rq_reply_portal = imp->imp_client->cli_reply_portal;
 222
 223         spin_lock_init(&request->rq_lock);
 224         INIT_LIST_HEAD(&request->rq_list);
 225         INIT_LIST_HEAD(&request->rq_replay_list);
 226         INIT_LIST_HEAD(&request->rq_set_chain);
 227         init_waitqueue_head(&request->rq_reply_waitq);
 228         request->rq_xid = ptlrpc_next_xid();
 229         atomic_set(&request->rq_refcount, 1);
 230
 231         request->rq_reqmsg->opc = opcode;
 232         request->rq_reqmsg->flags = 0;
 233
 234         RETURN(request);
 235 }
 236
 237 struct ptlrpc_request_set *ptlrpc_prep_set(void)
 238 {
 239         struct ptlrpc_request_set *set;
 240
 241         OBD_ALLOC(set, sizeof *set);
 242         if (!set)
 243                 RETURN(NULL);
 244         INIT_LIST_HEAD(&set->set_requests);
 245         init_waitqueue_head(&set->set_waitq);
 246         set->set_remaining = 0;
 247         spin_lock_init(&set->set_new_req_lock);
 248         INIT_LIST_HEAD(&set->set_new_requests);
 249
 250         RETURN(set);
 251 }
 252
 253 /* Finish with this set; opposite of prep_set. */
 254 void ptlrpc_set_destroy(struct ptlrpc_request_set *set)
 255 {
 256         struct list_head *tmp;
 257         struct list_head *next;
 258         int               expected_phase;
 259         int               n = 0;
 260         ENTRY;
 261
 262         /* Requests on the set should either all be completed, or all be new */
 263         expected_phase = (set->set_remaining == 0) ?
 264                          RQ_PHASE_COMPLETE : RQ_PHASE_NEW;
 265         list_for_each (tmp, &set->set_requests) {
 266                 struct ptlrpc_request *req =
 267                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 268
 269                 LASSERT(req->rq_phase == expected_phase);
 270                 n++;
 271         }
 272
 273         LASSERT(set->set_remaining == 0 || set->set_remaining == n);
 274
 275         list_for_each_safe(tmp, next, &set->set_requests) {
 276                 struct ptlrpc_request *req =
 277                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 278                 list_del_init(&req->rq_set_chain);
 279
 280                 LASSERT(req->rq_phase == expected_phase);
 281
 282                 if (req->rq_phase == RQ_PHASE_NEW) {
 283
 284                         if (req->rq_interpret_reply != NULL) {
 285                                 int (*interpreter)(struct ptlrpc_request *,
 286                                                    void *, int) =
 287                                         req->rq_interpret_reply;
 288
 289                                 /* higher level (i.e. LOV) failed;
 290                                  * let the sub reqs clean up */
 291                                 req->rq_status = -EBADR;
 292                                 interpreter(req, &req->rq_async_args,
 293                                             req->rq_status);
 294                         }
 295                         set->set_remaining--;
 296                 }
 297
 298                 req->rq_set = NULL;
 299                 ptlrpc_req_finished (req);
 300         }
 301
 302         LASSERT(set->set_remaining == 0);
 303
 304         OBD_FREE(set, sizeof(*set));
 305         EXIT;
 306 }
 307
 308 void ptlrpc_set_add_req(struct ptlrpc_request_set *set,
 309                         struct ptlrpc_request *req)
 310 {
 311         /* The set takes over the caller's request reference */
 312         list_add_tail(&req->rq_set_chain, &set->set_requests);
 313         req->rq_set = set;
 314         set->set_remaining++;
 315         atomic_inc(&req->rq_import->imp_inflight);
 316 }
 317
 318 /* lock so many callers can add things, the context that owns the set
 319  * is supposed to notice these and move them into the set proper. */
 320 void ptlrpc_set_add_new_req(struct ptlrpc_request_set *set,
 321                             struct ptlrpc_request *req)
 322 {
 323         unsigned long flags;
 324         spin_lock_irqsave(&set->set_new_req_lock, flags);
 325         /* The set takes over the caller's request reference */
 326         list_add_tail(&req->rq_set_chain, &set->set_new_requests);
 327         req->rq_set = set;
 328         spin_unlock_irqrestore(&set->set_new_req_lock, flags);
 329 }
 330
 331 /*
 332  * Based on the current state of the import, determine if the request
 333  * can be sent, is an error, or should be delayed.
 334  *
 335  * Returns true if this request should be delayed. If false, and
 336  * *status is set, then the request can not be sent and *status is the
 337  * error code.  If false and status is 0, then request can be sent.
 338  *
 339  * The imp->imp_lock must be held.
 340  */
 341 static int ptlrpc_import_delay_req(struct obd_import *imp,
 342                                    struct ptlrpc_request *req, int *status)
 343 {
 344         int delay = 0;
 345         ENTRY;
 346
 347         LASSERT (status != NULL);
 348         *status = 0;
 349
 350         if (imp->imp_state == LUSTRE_IMP_NEW) {
 351                 DEBUG_REQ(D_ERROR, req, "Uninitialized import.");
 352                 *status = -EIO;
 353                 LBUG();
 354         }
 355         else if (imp->imp_state == LUSTRE_IMP_CLOSED) {
 356                 DEBUG_REQ(D_ERROR, req, "IMP_CLOSED ");
 357                 *status = -EIO;
 358         }
 359         /* allow CONNECT even if import is invalid */
 360         else if (req->rq_send_state == LUSTRE_IMP_CONNECTING &&
 361                  imp->imp_state == LUSTRE_IMP_CONNECTING) {
 362                 ;
 363         }
 364         /*
 365          * If the import has been invalidated (such as by an OST failure), the
 366          * request must fail with -EIO.
 367          */
 368         else if (imp->imp_invalid) {
 369                 DEBUG_REQ(D_ERROR, req, "IMP_INVALID");
 370                 *status = -EIO;
 371         }
 372         else if (req->rq_import_generation != imp->imp_generation) {
 373                 DEBUG_REQ(D_ERROR, req, "req wrong generation:");
 374                 *status = -EIO;
 375         }
 376         else if (req->rq_send_state != imp->imp_state) {
 377                 if (imp->imp_obd->obd_no_recov || imp->imp_dlm_fake
 378                     || req->rq_no_delay)
 379                         *status = -EWOULDBLOCK;
 380                 else
 381                         delay = 1;
 382         }
 383
 384         RETURN(delay);
 385 }
 386
 387 static int ptlrpc_check_reply(struct ptlrpc_request *req)
 388 {
 389         unsigned long flags;
 390         int rc = 0;
 391         ENTRY;
 392
 393         /* serialise with network callback */
 394         spin_lock_irqsave (&req->rq_lock, flags);
 395
 396         if (req->rq_replied) {
 397                 DEBUG_REQ(D_NET, req, "REPLIED:");
 398                 GOTO(out, rc = 1);
 399         }
 400
 401         if (req->rq_net_err && !req->rq_timedout) {
 402                 spin_unlock_irqrestore (&req->rq_lock, flags);
 403                 rc = ptlrpc_expire_one_request(req);
 404                 spin_lock_irqsave (&req->rq_lock, flags);
 405                 GOTO(out, rc);
 406         }
 407
 408         if (req->rq_err) {
 409                 DEBUG_REQ(D_ERROR, req, "ABORTED:");
 410                 GOTO(out, rc = 1);
 411         }
 412
 413         if (req->rq_resend) {
 414                 DEBUG_REQ(D_ERROR, req, "RESEND:");
 415                 GOTO(out, rc = 1);
 416         }
 417
 418         if (req->rq_restart) {
 419                 DEBUG_REQ(D_ERROR, req, "RESTART:");
 420                 GOTO(out, rc = 1);
 421         }
 422         EXIT;
 423  out:
 424         spin_unlock_irqrestore (&req->rq_lock, flags);
 425         DEBUG_REQ(D_NET, req, "rc = %d for", rc);
 426         return rc;
 427 }
 428
 429 static int ptlrpc_check_status(struct ptlrpc_request *req)
 430 {
 431         int err;
 432         ENTRY;
 433
 434         err = req->rq_repmsg->status;
 435         if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 436                 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err == %d",
 437                           err);
 438                 RETURN(err < 0 ? err : -EINVAL);
 439         }
 440
 441         if (err < 0) {
 442                 DEBUG_REQ(D_INFO, req, "status is %d", err);
 443         } else if (err > 0) {
 444                 /* XXX: translate this error from net to host */
 445                 DEBUG_REQ(D_INFO, req, "status is %d", err);
 446         }
 447
 448         RETURN(err);
 449 }
 450
 451 static int after_reply(struct ptlrpc_request *req)
 452 {
 453         unsigned long flags;
 454         struct obd_import *imp = req->rq_import;
 455         int rc;
 456         ENTRY;
 457
 458         LASSERT(!req->rq_receiving_reply);
 459
 460         /* NB Until this point, the whole of the incoming message,
 461          * including buflens, status etc is in the sender's byte order. */
 462
 463 #if SWAB_PARANOIA
 464         /* Clear reply swab mask; this is a new reply in sender's byte order */
 465         req->rq_rep_swab_mask = 0;
 466 #endif
 467         LASSERT (req->rq_nob_received <= req->rq_replen);
 468         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_nob_received);
 469         if (rc) {
 470                 CERROR("unpack_rep failed: %d\n", rc);
 471                 RETURN(-EPROTO);
 472         }
 473
 474         if (req->rq_repmsg->type != PTL_RPC_MSG_REPLY &&
 475             req->rq_repmsg->type != PTL_RPC_MSG_ERR) {
 476                 CERROR("invalid packet type received (type=%u)\n",
 477                        req->rq_repmsg->type);
 478                 RETURN(-EPROTO);
 479         }
 480
 481         /* Store transno in reqmsg for replay. */
 482         req->rq_reqmsg->transno = req->rq_transno = req->rq_repmsg->transno;
 483
 484         rc = ptlrpc_check_status(req);
 485
 486         /* Either we've been evicted, or the server has failed for
 487          * some reason. Try to reconnect, and if that fails, punt to the
 488          * upcall. */
 489         if (rc == -ENOTCONN) {
 490                 if (req->rq_send_state != LUSTRE_IMP_FULL ||
 491                     imp->imp_obd->obd_no_recov || imp->imp_dlm_fake) {
 492                         RETURN(-ENOTCONN);
 493                 }
 494
 495                 ptlrpc_request_handle_notconn(req);
 496
 497                 RETURN(rc);
 498         }
 499
 500         if (req->rq_import->imp_replayable) {
 501                 spin_lock_irqsave(&imp->imp_lock, flags);
 502                 if (req->rq_replay || req->rq_transno != 0)
 503                         ptlrpc_retain_replayable_request(req, imp);
 504                 else if (req->rq_commit_cb != NULL) {
 505                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 506                         req->rq_commit_cb(req);
 507                         spin_lock_irqsave(&imp->imp_lock, flags);
 508                 }
 509
 510                 if (req->rq_transno > imp->imp_max_transno)
 511                         imp->imp_max_transno = req->rq_transno;
 512
 513                 /* Replay-enabled imports return commit-status information. */
 514                 if (req->rq_repmsg->last_committed)
 515                         imp->imp_peer_committed_transno =
 516                                 req->rq_repmsg->last_committed;
 517                 ptlrpc_free_committed(imp);
 518                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 519         }
 520
 521         RETURN(rc);
 522 }
 523
 524 static int ptlrpc_send_new_req(struct ptlrpc_request *req)
 525 {
 526         char                   str[PTL_NALFMT_SIZE];
 527         struct obd_import     *imp;
 528         unsigned long          flags;
 529         int rc;
 530         ENTRY;
 531
 532         LASSERT(req->rq_phase == RQ_PHASE_NEW);
 533         req->rq_phase = RQ_PHASE_RPC;
 534
 535         imp = req->rq_import;
 536         spin_lock_irqsave(&imp->imp_lock, flags);
 537
 538         req->rq_import_generation = imp->imp_generation;
 539
 540         if (ptlrpc_import_delay_req(imp, req, &rc)) {
 541                 spin_lock (&req->rq_lock);
 542                 req->rq_waiting = 1;
 543                 spin_unlock (&req->rq_lock);
 544
 545                 DEBUG_REQ(D_HA, req, "req from PID %d waiting for recovery: "
 546                           "(%s != %s)",
 547                           req->rq_reqmsg->status,
 548                           ptlrpc_import_state_name(req->rq_send_state),
 549                           ptlrpc_import_state_name(imp->imp_state));
 550                 LASSERT(list_empty (&req->rq_list));
 551
 552                 list_add_tail(&req->rq_list, &imp->imp_delayed_list);
 553                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 554                 RETURN(0);
 555         }
 556
 557         if (rc != 0) {
 558                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 559                 req->rq_status = rc;
 560                 req->rq_phase = RQ_PHASE_INTERPRET;
 561                 RETURN(rc);
 562         }
 563
 564         /* XXX this is the same as ptlrpc_queue_wait */
 565         LASSERT(list_empty(&req->rq_list));
 566         list_add_tail(&req->rq_list, &imp->imp_sending_list);
 567         spin_unlock_irqrestore(&imp->imp_lock, flags);
 568
 569         req->rq_reqmsg->status = current->pid;
 570         CDEBUG(D_RPCTRACE, "Sending RPC pname:cluuid:pid:xid:ni:nid:opc"
 571                " %s:%s:%d:"LPU64":%s:%s:%d\n", current->comm,
 572                imp->imp_obd->obd_uuid.uuid, req->rq_reqmsg->status,
 573                req->rq_xid,
 574                imp->imp_connection->c_peer.peer_ni->pni_name,
 575                ptlrpc_peernid2str(&imp->imp_connection->c_peer, str),
 576                req->rq_reqmsg->opc);
 577
 578         rc = ptl_send_rpc(req);
 579         if (rc) {
 580                 DEBUG_REQ(D_HA, req, "send failed (%d); expect timeout", rc);
 581                 req->rq_net_err = 1;
 582                 RETURN(rc);
 583         }
 584         RETURN(0);
 585 }
 586
 587 int ptlrpc_check_set(struct ptlrpc_request_set *set)
 588 {
 589         char str[PTL_NALFMT_SIZE];
 590         unsigned long flags;
 591         struct list_head *tmp;
 592         int force_timer_recalc = 0;
 593         ENTRY;
 594
 595         if (set->set_remaining == 0)
 596                 RETURN(1);
 597
 598         list_for_each(tmp, &set->set_requests) {
 599                 struct ptlrpc_request *req =
 600                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 601                 struct obd_import *imp = req->rq_import;
 602                 int rc = 0;
 603
 604                 if (req->rq_phase == RQ_PHASE_NEW &&
 605                     ptlrpc_send_new_req(req)) {
 606                         force_timer_recalc = 1;
 607                 }
 608
 609                 if (!(req->rq_phase == RQ_PHASE_RPC ||
 610                       req->rq_phase == RQ_PHASE_BULK ||
 611                       req->rq_phase == RQ_PHASE_INTERPRET ||
 612                       req->rq_phase == RQ_PHASE_COMPLETE)) {
 613                         DEBUG_REQ(D_ERROR, req, "bad phase %x", req->rq_phase);
 614                         LBUG();
 615                 }
 616
 617                 if (req->rq_phase == RQ_PHASE_COMPLETE)
 618                         continue;
 619
 620                 if (req->rq_phase == RQ_PHASE_INTERPRET)
 621                         GOTO(interpret, req->rq_status);
 622
 623                 if (req->rq_net_err && !req->rq_timedout)
 624                         ptlrpc_expire_one_request(req);
 625
 626                 if (req->rq_err) {
 627                         ptlrpc_unregister_reply(req);
 628                         if (req->rq_status == 0)
 629                                 req->rq_status = -EIO;
 630                         req->rq_phase = RQ_PHASE_INTERPRET;
 631
 632                         spin_lock_irqsave(&imp->imp_lock, flags);
 633                         list_del_init(&req->rq_list);
 634                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 635
 636                         GOTO(interpret, req->rq_status);
 637                 }
 638
 639                 /* ptlrpc_queue_wait->l_wait_event guarantees that rq_intr
 640                  * will only be set after rq_timedout, but the oig waiting
 641                  * path sets rq_intr irrespective of whether ptlrpcd has
 642                  * seen a timeout.  our policy is to only interpret
 643                  * interrupted rpcs after they have timed out */
 644                 if (req->rq_intr && (req->rq_timedout || req->rq_waiting)) {
 645                         /* NB could be on delayed list */
 646                         ptlrpc_unregister_reply(req);
 647                         req->rq_status = -EINTR;
 648                         req->rq_phase = RQ_PHASE_INTERPRET;
 649
 650                         spin_lock_irqsave(&imp->imp_lock, flags);
 651                         list_del_init(&req->rq_list);
 652                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 653
 654                         GOTO(interpret, req->rq_status);
 655                 }
 656
 657                 if (req->rq_phase == RQ_PHASE_RPC) {
 658                         if (req->rq_waiting || req->rq_resend) {
 659                                 int status;
 660
 661                                 ptlrpc_unregister_reply(req);
 662
 663                                 spin_lock_irqsave(&imp->imp_lock, flags);
 664
 665                                 if (ptlrpc_import_delay_req(imp, req, &status)){
 666                                         spin_unlock_irqrestore(&imp->imp_lock,
 667                                                                flags);
 668                                         continue;
 669                                 }
 670
 671                                 list_del_init(&req->rq_list);
 672                                 if (status != 0)  {
 673                                         req->rq_status = status;
 674                                         req->rq_phase = RQ_PHASE_INTERPRET;
 675                                         spin_unlock_irqrestore(&imp->imp_lock,
 676                                                                flags);
 677                                         GOTO(interpret, req->rq_status);
 678                                 }
 679                                 if (req->rq_no_resend) {
 680                                         req->rq_status = -ENOTCONN;
 681                                         req->rq_phase = RQ_PHASE_INTERPRET;
 682                                         spin_unlock_irqrestore(&imp->imp_lock,
 683                                                                flags);
 684                                         GOTO(interpret, req->rq_status);
 685                                 }
 686                                 list_add_tail(&req->rq_list,
 687                                               &imp->imp_sending_list);
 688
 689                                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 690
 691                                 req->rq_waiting = 0;
 692                                 if (req->rq_resend) {
 693                                         lustre_msg_add_flags(req->rq_reqmsg,
 694                                                              MSG_RESENT);
 695                                         if (req->rq_bulk) {
 696                                                 __u64 old_xid = req->rq_xid;
 697
 698                                                 ptlrpc_unregister_bulk (req);
 699
 700                                                 /* ensure previous bulk fails */
 701                                                 req->rq_xid = ptlrpc_next_xid();
 702                                                 CDEBUG(D_HA, "resend bulk "
 703                                                        "old x"LPU64
 704                                                        " new x"LPU64"\n",
 705                                                        old_xid, req->rq_xid);
 706                                         }
 707                                 }
 708
 709                                 rc = ptl_send_rpc(req);
 710                                 if (rc) {
 711                                         DEBUG_REQ(D_HA, req, "send failed (%d)",
 712                                                   rc);
 713                                         force_timer_recalc = 1;
 714                                         req->rq_net_err = 1;
 715                                 }
 716                                 /* need to reset the timeout */
 717                                 force_timer_recalc = 1;
 718                         }
 719
 720                         /* Still waiting for a reply? */
 721                         if (ptlrpc_client_receiving_reply(req))
 722                                 continue;
 723
 724                         /* Did we actually receive a reply? */
 725                         if (!ptlrpc_client_replied(req))
 726                                 continue;
 727
 728                         spin_lock_irqsave(&imp->imp_lock, flags);
 729                         list_del_init(&req->rq_list);
 730                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 731
 732                         req->rq_status = after_reply(req);
 733                         if (req->rq_resend) {
 734                                 /* Add this req to the delayed list so
 735                                    it can be errored if the import is
 736                                    evicted after recovery. */
 737                                 spin_lock_irqsave (&req->rq_lock, flags);
 738                                 list_add_tail(&req->rq_list,
 739                                               &imp->imp_delayed_list);
 740                                 spin_unlock_irqrestore(&req->rq_lock, flags);
 741                                 continue;
 742                         }
 743
 744                         /* If there is no bulk associated with this request,
 745                          * then we're done and should let the interpreter
 746                          * process the reply.  Similarly if the RPC returned
 747                          * an error, and therefore the bulk will never arrive.
 748                          */
 749                         if (req->rq_bulk == NULL || req->rq_status != 0) {
 750                                 req->rq_phase = RQ_PHASE_INTERPRET;
 751                                 GOTO(interpret, req->rq_status);
 752                         }
 753
 754                         req->rq_phase = RQ_PHASE_BULK;
 755                 }
 756
 757                 LASSERT(req->rq_phase == RQ_PHASE_BULK);
 758                 if (ptlrpc_bulk_active(req->rq_bulk))
 759                         continue;
 760
 761                 if (!req->rq_bulk->bd_success) {
 762                         /* The RPC reply arrived OK, but the bulk screwed
 763                          * up!  Dead wierd since the server told us the RPC
 764                          * was good after getting the REPLY for her GET or
 765                          * the ACK for her PUT. */
 766                         DEBUG_REQ(D_ERROR, req, "bulk transfer failed");
 767                         LBUG();
 768                 }
 769
 770                 req->rq_phase = RQ_PHASE_INTERPRET;
 771
 772         interpret:
 773                 LASSERT(req->rq_phase == RQ_PHASE_INTERPRET);
 774                 LASSERT(!req->rq_receiving_reply);
 775
 776                 ptlrpc_unregister_reply(req);
 777                 if (req->rq_bulk != NULL)
 778                         ptlrpc_unregister_bulk (req);
 779
 780                 req->rq_phase = RQ_PHASE_COMPLETE;
 781
 782                 if (req->rq_interpret_reply != NULL) {
 783                         int (*interpreter)(struct ptlrpc_request *,void *,int) =
 784                                 req->rq_interpret_reply;
 785                         req->rq_status = interpreter(req, &req->rq_async_args,
 786                                                      req->rq_status);
 787                 }
 788
 789                 CDEBUG(D_RPCTRACE, "Completed RPC pname:cluuid:pid:xid:ni:nid:"
 790                        "opc %s:%s:%d:"LPU64":%s:%s:%d\n", current->comm,
 791                        imp->imp_obd->obd_uuid.uuid, req->rq_reqmsg->status,
 792                        req->rq_xid,
 793                        imp->imp_connection->c_peer.peer_ni->pni_name,
 794                        ptlrpc_peernid2str(&imp->imp_connection->c_peer, str),
 795                        req->rq_reqmsg->opc);
 796
 797                 set->set_remaining--;
 798
 799                 atomic_dec(&imp->imp_inflight);
 800                 wake_up(&imp->imp_recovery_waitq);
 801         }
 802
 803         /* If we hit an error, we want to recover promptly. */
 804         RETURN(set->set_remaining == 0 || force_timer_recalc);
 805 }
 806
 807 int ptlrpc_expire_one_request(struct ptlrpc_request *req)
 808 {
 809         unsigned long      flags;
 810         struct obd_import *imp = req->rq_import;
 811         ENTRY;
 812
 813         DEBUG_REQ(D_ERROR, req, "timeout");
 814
 815         spin_lock_irqsave (&req->rq_lock, flags);
 816         req->rq_timedout = 1;
 817         spin_unlock_irqrestore (&req->rq_lock, flags);
 818
 819         ptlrpc_unregister_reply (req);
 820
 821         if (req->rq_bulk != NULL)
 822                 ptlrpc_unregister_bulk (req);
 823
 824         if (imp == NULL) {
 825                 DEBUG_REQ(D_HA, req, "NULL import: already cleaned up?");
 826                 RETURN(1);
 827         }
 828
 829         /* The DLM server doesn't want recovery run on its imports. */
 830         if (imp->imp_dlm_fake)
 831                 RETURN(1);
 832
 833         /* If this request is for recovery or other primordial tasks,
 834          * then error it out here. */
 835         if (req->rq_send_state != LUSTRE_IMP_FULL ||
 836             imp->imp_obd->obd_no_recov) {
 837                 spin_lock_irqsave (&req->rq_lock, flags);
 838                 req->rq_status = -ETIMEDOUT;
 839                 req->rq_err = 1;
 840                 spin_unlock_irqrestore (&req->rq_lock, flags);
 841                 RETURN(1);
 842         }
 843
 844         ptlrpc_fail_import(imp, req->rq_import_generation);
 845
 846         RETURN(0);
 847 }
 848
 849 int ptlrpc_expired_set(void *data)
 850 {
 851         struct ptlrpc_request_set *set = data;
 852         struct list_head          *tmp;
 853         time_t                     now = LTIME_S (CURRENT_TIME);
 854         ENTRY;
 855
 856         LASSERT(set != NULL);
 857
 858         /* A timeout expired; see which reqs it applies to... */
 859         list_for_each (tmp, &set->set_requests) {
 860                 struct ptlrpc_request *req =
 861                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 862
 863                 /* request in-flight? */
 864                 if (!((req->rq_phase == RQ_PHASE_RPC && !req->rq_waiting
 865                        && !req->rq_resend) ||
 866                       (req->rq_phase == RQ_PHASE_BULK)))
 867                         continue;
 868
 869                 if (req->rq_timedout ||           /* already dealt with */
 870                     req->rq_sent + req->rq_timeout > now) /* not expired */
 871                         continue;
 872
 873                 /* deal with this guy */
 874                 ptlrpc_expire_one_request (req);
 875         }
 876
 877         /* When waiting for a whole set, we always to break out of the
 878          * sleep so we can recalculate the timeout, or enable interrupts
 879          * iff everyone's timed out.
 880          */
 881         RETURN(1);
 882 }
 883
 884 void ptlrpc_mark_interrupted(struct ptlrpc_request *req)
 885 {
 886         unsigned long flags;
 887         spin_lock_irqsave(&req->rq_lock, flags);
 888         req->rq_intr = 1;
 889         spin_unlock_irqrestore(&req->rq_lock, flags);
 890 }
 891
 892 void ptlrpc_interrupted_set(void *data)
 893 {
 894         struct ptlrpc_request_set *set = data;
 895         struct list_head *tmp;
 896
 897         LASSERT(set != NULL);
 898         CERROR("INTERRUPTED SET %p\n", set);
 899
 900         list_for_each(tmp, &set->set_requests) {
 901                 struct ptlrpc_request *req =
 902                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 903
 904                 if (req->rq_phase != RQ_PHASE_RPC)
 905                         continue;
 906
 907                 ptlrpc_mark_interrupted(req);
 908         }
 909 }
 910
 911 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *set)
 912 {
 913         struct list_head      *tmp;
 914         time_t                 now = LTIME_S(CURRENT_TIME);
 915         time_t                 deadline;
 916         int                    timeout = 0;
 917         struct ptlrpc_request *req;
 918         ENTRY;
 919
 920         SIGNAL_MASK_ASSERT(); /* XXX BUG 1511 */
 921
 922         list_for_each(tmp, &set->set_requests) {
 923                 req = list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 924
 925                 /* request in-flight? */
 926                 if (!((req->rq_phase == RQ_PHASE_RPC && !req->rq_waiting) ||
 927                       (req->rq_phase == RQ_PHASE_BULK)))
 928                         continue;
 929
 930                 if (req->rq_timedout)   /* already timed out */
 931                         continue;
 932
 933                 deadline = req->rq_sent + req->rq_timeout;
 934                 if (deadline <= now)    /* actually expired already */
 935                         timeout = 1;    /* ASAP */
 936                 else if (timeout == 0 || timeout > deadline - now)
 937                         timeout = deadline - now;
 938         }
 939         RETURN(timeout);
 940 }
 941
 942
 943 int ptlrpc_set_wait(struct ptlrpc_request_set *set)
 944 {
 945         struct list_head      *tmp;
 946         struct ptlrpc_request *req;
 947         struct l_wait_info     lwi;
 948         int                    rc, timeout;
 949         ENTRY;
 950
 951         LASSERT(!list_empty(&set->set_requests));
 952         list_for_each(tmp, &set->set_requests) {
 953                 req = list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 954                 if (req->rq_phase == RQ_PHASE_NEW)
 955                         (void)ptlrpc_send_new_req(req);
 956         }
 957
 958         do {
 959                 timeout = ptlrpc_set_next_timeout(set);
 960
 961                 /* wait until all complete, interrupted, or an in-flight
 962                  * req times out */
 963                 CDEBUG(D_HA, "set %p going to sleep for %d seconds\n",
 964                        set, timeout);
 965                 lwi = LWI_TIMEOUT_INTR((timeout ? timeout : 1) * HZ,
 966                                        ptlrpc_expired_set,
 967                                        ptlrpc_interrupted_set, set);
 968                 rc = l_wait_event(set->set_waitq, ptlrpc_check_set(set), &lwi);
 969
 970                 LASSERT(rc == 0 || rc == -EINTR || rc == -ETIMEDOUT);
 971
 972                 /* -EINTR => all requests have been flagged rq_intr so next
 973                  * check completes.
 974                  * -ETIMEOUTD => someone timed out.  When all reqs have
 975                  * timed out, signals are enabled allowing completion with
 976                  * EINTR.
 977                  * I don't really care if we go once more round the loop in
 978                  * the error cases -eeb. */
 979         } while (rc != 0);
 980
 981         LASSERT(set->set_remaining == 0);
 982
 983         rc = 0;
 984         list_for_each(tmp, &set->set_requests) {
 985                 req = list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 986
 987                 LASSERT(req->rq_phase == RQ_PHASE_COMPLETE);
 988                 if (req->rq_status != 0)
 989                         rc = req->rq_status;
 990         }
 991
 992         if (set->set_interpret != NULL) {
 993                 int (*interpreter)(struct ptlrpc_request_set *set,void *,int) =
 994                         set->set_interpret;
 995                 rc = interpreter (set, &set->set_args, rc);
 996         }
 997
 998         RETURN(rc);
 999 }
1000
1001 static void __ptlrpc_free_req(struct ptlrpc_request *request, int locked)
1002 {
1003         ENTRY;
1004         if (request == NULL) {
1005                 EXIT;
1006                 return;
1007         }
1008
1009         LASSERTF(!request->rq_receiving_reply, "req %p\n", request);
1010         LASSERTF(request->rq_rqbd == NULL, "req %p\n",request);/* client-side */
1011         LASSERTF(list_empty(&request->rq_list), "req %p\n", request);
1012         LASSERTF(list_empty(&request->rq_set_chain), "req %p\n", request);
1013
1014         /* We must take it off the imp_replay_list first.  Otherwise, we'll set
1015          * request->rq_reqmsg to NULL while osc_close is dereferencing it. */
1016         if (request->rq_import != NULL) {
1017                 unsigned long flags = 0;
1018                 if (!locked)
1019                         spin_lock_irqsave(&request->rq_import->imp_lock, flags);
1020                 list_del_init(&request->rq_replay_list);
1021                 if (!locked)
1022                         spin_unlock_irqrestore(&request->rq_import->imp_lock,
1023                                                flags);
1024         }
1025         LASSERTF(list_empty(&request->rq_replay_list), "req %p\n", request);
1026
1027         if (atomic_read(&request->rq_refcount) != 0) {
1028                 DEBUG_REQ(D_ERROR, request,
1029                           "freeing request with nonzero refcount");
1030                 LBUG();
1031         }
1032
1033         if (request->rq_repmsg != NULL) {
1034                 OBD_FREE(request->rq_repmsg, request->rq_replen);
1035                 request->rq_repmsg = NULL;
1036         }
1037         if (request->rq_reqmsg != NULL) {
1038                 OBD_FREE(request->rq_reqmsg, request->rq_reqlen);
1039                 request->rq_reqmsg = NULL;
1040         }
1041         if (request->rq_export != NULL) {
1042                 class_export_put(request->rq_export);
1043                 request->rq_export = NULL;
1044         }
1045         if (request->rq_import != NULL) {
1046                 class_import_put(request->rq_import);
1047                 request->rq_import = NULL;
1048         }
1049         if (request->rq_bulk != NULL)
1050                 ptlrpc_free_bulk(request->rq_bulk);
1051
1052         OBD_FREE(request, sizeof(*request));
1053         EXIT;
1054 }
1055
1056 void ptlrpc_free_req(struct ptlrpc_request *request)
1057 {
1058         __ptlrpc_free_req(request, 0);
1059 }
1060
1061 static int __ptlrpc_req_finished(struct ptlrpc_request *request, int locked);
1062 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request)
1063 {
1064         LASSERT_SPIN_LOCKED(&request->rq_import->imp_lock);
1065         (void)__ptlrpc_req_finished(request, 1);
1066 }
1067
1068 static int __ptlrpc_req_finished(struct ptlrpc_request *request, int locked)
1069 {
1070         ENTRY;
1071         if (request == NULL)
1072                 RETURN(1);
1073
1074         if (request == LP_POISON ||
1075             request->rq_reqmsg == LP_POISON) {
1076                 CERROR("dereferencing freed request (bug 575)\n");
1077                 LBUG();
1078                 RETURN(1);
1079         }
1080
1081         DEBUG_REQ(D_INFO, request, "refcount now %u",
1082                   atomic_read(&request->rq_refcount) - 1);
1083
1084         if (atomic_dec_and_test(&request->rq_refcount)) {
1085                 __ptlrpc_free_req(request, locked);
1086                 RETURN(1);
1087         }
1088
1089         RETURN(0);
1090 }
1091
1092 void ptlrpc_req_finished(struct ptlrpc_request *request)
1093 {
1094         __ptlrpc_req_finished(request, 0);
1095 }
1096
1097 /* Disengage the client's reply buffer from the network
1098  * NB does _NOT_ unregister any client-side bulk.
1099  * IDEMPOTENT, but _not_ safe against concurrent callers.
1100  * The request owner (i.e. the thread doing the I/O) must call...
1101  */
1102 void ptlrpc_unregister_reply (struct ptlrpc_request *request)
1103 {
1104         int                rc;
1105         wait_queue_head_t *wq;
1106         struct l_wait_info lwi;
1107
1108         LASSERT(!in_interrupt ());             /* might sleep */
1109
1110         if (!ptlrpc_client_receiving_reply(request))
1111                 return;
1112
1113         PtlMDUnlink (request->rq_reply_md_h);
1114
1115         /* We have to l_wait_event() whatever the result, to give liblustre
1116          * a chance to run reply_in_callback() */
1117
1118         if (request->rq_set == NULL)
1119                 wq = &request->rq_set->set_waitq;
1120         else
1121                 wq = &request->rq_reply_waitq;
1122
1123         for (;;) {
1124                 /* Network access will complete in finite time but the HUGE
1125                  * timeout lets us CWARN for visibility of sluggish NALs */
1126                 lwi = LWI_TIMEOUT(300 * HZ, NULL, NULL);
1127                 rc = l_wait_event (*wq, !ptlrpc_client_receiving_reply(request), &lwi);
1128                 if (rc == 0)
1129                         return;
1130
1131                 LASSERT (rc == -ETIMEDOUT);
1132                 DEBUG_REQ(D_WARNING, request, "Unexpectedly long timeout");
1133         }
1134 }
1135
1136 /* caller must hold imp->imp_lock */
1137 void ptlrpc_free_committed(struct obd_import *imp)
1138 {
1139         struct list_head *tmp, *saved;
1140         struct ptlrpc_request *req;
1141         struct ptlrpc_request *last_req = NULL; /* temporary fire escape */
1142         ENTRY;
1143
1144         LASSERT(imp != NULL);
1145
1146         LASSERT_SPIN_LOCKED(&imp->imp_lock);
1147
1148         CDEBUG(D_HA, "%s: committing for last_committed "LPU64"\n",
1149                imp->imp_obd->obd_name, imp->imp_peer_committed_transno);
1150
1151         list_for_each_safe(tmp, saved, &imp->imp_replay_list) {
1152                 req = list_entry(tmp, struct ptlrpc_request, rq_replay_list);
1153
1154                 /* XXX ok to remove when 1357 resolved - rread 05/29/03  */
1155                 LASSERT(req != last_req);
1156                 last_req = req;
1157
1158                 if (req->rq_import_generation < imp->imp_generation) {
1159                         DEBUG_REQ(D_HA, req, "freeing request with old gen");
1160                         GOTO(free_req, 0);
1161                 }
1162
1163                 if (req->rq_replay) {
1164                         DEBUG_REQ(D_HA, req, "keeping (FL_REPLAY)");
1165                         continue;
1166                 }
1167
1168                 /* not yet committed */
1169                 if (req->rq_transno > imp->imp_peer_committed_transno) {
1170                         DEBUG_REQ(D_HA, req, "stopping search");
1171                         break;
1172                 }
1173
1174                 DEBUG_REQ(D_HA, req, "committing (last_committed "LPU64")",
1175                           imp->imp_peer_committed_transno);
1176 free_req:
1177                 if (req->rq_commit_cb != NULL)
1178                         req->rq_commit_cb(req);
1179                 list_del_init(&req->rq_replay_list);
1180                 __ptlrpc_req_finished(req, 1);
1181         }
1182
1183         EXIT;
1184         return;
1185 }
1186
1187 void ptlrpc_cleanup_client(struct obd_import *imp)
1188 {
1189         ENTRY;
1190         EXIT;
1191         return;
1192 }
1193
1194 void ptlrpc_resend_req(struct ptlrpc_request *req)
1195 {
1196         unsigned long flags;
1197
1198         DEBUG_REQ(D_HA, req, "going to resend");
1199         req->rq_reqmsg->handle.cookie = 0;
1200         req->rq_status = -EAGAIN;
1201
1202         spin_lock_irqsave (&req->rq_lock, flags);
1203         req->rq_resend = 1;
1204         req->rq_net_err = 0;
1205         req->rq_timedout = 0;
1206         if (req->rq_bulk) {
1207                 __u64 old_xid = req->rq_xid;
1208
1209                 /* ensure previous bulk fails */
1210                 req->rq_xid = ptlrpc_next_xid();
1211                 CDEBUG(D_HA, "resend bulk old x"LPU64" new x"LPU64"\n",
1212                        old_xid, req->rq_xid);
1213         }
1214         ptlrpc_wake_client_req(req);
1215         spin_unlock_irqrestore (&req->rq_lock, flags);
1216 }
1217
1218 /* XXX: this function and rq_status are currently unused */
1219 void ptlrpc_restart_req(struct ptlrpc_request *req)
1220 {
1221         unsigned long flags;
1222
1223         DEBUG_REQ(D_HA, req, "restarting (possibly-)completed request");
1224         req->rq_status = -ERESTARTSYS;
1225
1226         spin_lock_irqsave (&req->rq_lock, flags);
1227         req->rq_restart = 1;
1228         req->rq_timedout = 0;
1229         ptlrpc_wake_client_req(req);
1230         spin_unlock_irqrestore (&req->rq_lock, flags);
1231 }
1232
1233 static int expired_request(void *data)
1234 {
1235         struct ptlrpc_request *req = data;
1236         ENTRY;
1237
1238         RETURN(ptlrpc_expire_one_request(req));
1239 }
1240
1241 static void interrupted_request(void *data)
1242 {
1243         unsigned long flags;
1244
1245         struct ptlrpc_request *req = data;
1246         DEBUG_REQ(D_HA, req, "request interrupted");
1247         spin_lock_irqsave (&req->rq_lock, flags);
1248         req->rq_intr = 1;
1249         spin_unlock_irqrestore (&req->rq_lock, flags);
1250 }
1251
1252 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req)
1253 {
1254         ENTRY;
1255         atomic_inc(&req->rq_refcount);
1256         RETURN(req);
1257 }
1258
1259 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
1260                                       struct obd_import *imp)
1261 {
1262         struct list_head *tmp;
1263
1264         LASSERT_SPIN_LOCKED(&imp->imp_lock);
1265
1266         /* clear this  for new requests that were resent as well
1267            as resent replayed requests. */
1268         lustre_msg_clear_flags(req->rq_reqmsg,
1269                              MSG_RESENT);
1270
1271         /* don't re-add requests that have been replayed */
1272         if (!list_empty(&req->rq_replay_list))
1273                 return;
1274
1275         lustre_msg_add_flags(req->rq_reqmsg,
1276                              MSG_REPLAY);
1277
1278         LASSERT(imp->imp_replayable);
1279         /* Balanced in ptlrpc_free_committed, usually. */
1280         ptlrpc_request_addref(req);
1281         list_for_each_prev(tmp, &imp->imp_replay_list) {
1282                 struct ptlrpc_request *iter =
1283                         list_entry(tmp, struct ptlrpc_request, rq_replay_list);
1284
1285                 /* We may have duplicate transnos if we create and then
1286                  * open a file, or for closes retained if to match creating
1287                  * opens, so use req->rq_xid as a secondary key.
1288                  * (See bugs 684, 685, and 428.)
1289                  * XXX no longer needed, but all opens need transnos!
1290                  */
1291                 if (iter->rq_transno > req->rq_transno)
1292                         continue;
1293
1294                 if (iter->rq_transno == req->rq_transno) {
1295                         LASSERT(iter->rq_xid != req->rq_xid);
1296                         if (iter->rq_xid > req->rq_xid)
1297                                 continue;
1298                 }
1299
1300                 list_add(&req->rq_replay_list, &iter->rq_replay_list);
1301                 return;
1302         }
1303
1304         list_add_tail(&req->rq_replay_list, &imp->imp_replay_list);
1305 }
1306
1307 int ptlrpc_queue_wait(struct ptlrpc_request *req)
1308 {
1309         char str[PTL_NALFMT_SIZE];
1310         int rc = 0;
1311         int brc;
1312         struct l_wait_info lwi;
1313         struct obd_import *imp = req->rq_import;
1314         unsigned long flags;
1315         int timeout = 0;
1316         ENTRY;
1317
1318         LASSERT(req->rq_set == NULL);
1319         LASSERT(!req->rq_receiving_reply);
1320         atomic_inc(&imp->imp_inflight);
1321
1322         /* for distributed debugging */
1323         req->rq_reqmsg->status = current->pid;
1324         LASSERT(imp->imp_obd != NULL);
1325         CDEBUG(D_RPCTRACE, "Sending RPC pname:cluuid:pid:xid:ni:nid:opc "
1326                "%s:%s:%d:"LPU64":%s:%s:%d\n", current->comm,
1327                imp->imp_obd->obd_uuid.uuid,
1328                req->rq_reqmsg->status, req->rq_xid,
1329                imp->imp_connection->c_peer.peer_ni->pni_name,
1330                ptlrpc_peernid2str(&imp->imp_connection->c_peer, str),
1331                req->rq_reqmsg->opc);
1332
1333         /* Mark phase here for a little debug help */
1334         req->rq_phase = RQ_PHASE_RPC;
1335
1336         spin_lock_irqsave(&imp->imp_lock, flags);
1337         req->rq_import_generation = imp->imp_generation;
1338 restart:
1339         if (ptlrpc_import_delay_req(imp, req, &rc)) {
1340                 list_del(&req->rq_list);
1341
1342                 list_add_tail(&req->rq_list, &imp->imp_delayed_list);
1343                 spin_unlock_irqrestore(&imp->imp_lock, flags);
1344
1345                 DEBUG_REQ(D_HA, req, "\"%s\" waiting for recovery: (%s != %s)",
1346                           current->comm,
1347                           ptlrpc_import_state_name(req->rq_send_state),
1348                           ptlrpc_import_state_name(imp->imp_state));
1349                 lwi = LWI_INTR(interrupted_request, req);
1350                 rc = l_wait_event(req->rq_reply_waitq,
1351                                   (req->rq_send_state == imp->imp_state ||
1352                                    req->rq_err),
1353                                   &lwi);
1354                 DEBUG_REQ(D_HA, req, "\"%s\" awake: (%s == %s or %d == 1)",
1355                           current->comm,
1356                           ptlrpc_import_state_name(imp->imp_state),
1357                           ptlrpc_import_state_name(req->rq_send_state),
1358                           req->rq_err);
1359
1360                 spin_lock_irqsave(&imp->imp_lock, flags);
1361                 list_del_init(&req->rq_list);
1362
1363                 if (req->rq_err) {
1364                         rc = -EIO;
1365                 }
1366                 else if (req->rq_intr) {
1367                         rc = -EINTR;
1368                 }
1369                 else if (req->rq_no_resend) {
1370                         spin_unlock_irqrestore(&imp->imp_lock, flags);
1371                         GOTO(out, rc = -ETIMEDOUT);
1372                 }
1373                 else {
1374                         GOTO(restart, rc);
1375                 }
1376         }
1377
1378         if (rc != 0) {
1379                 list_del_init(&req->rq_list);
1380                 spin_unlock_irqrestore(&imp->imp_lock, flags);
1381                 req->rq_status = rc; // XXX this ok?
1382                 GOTO(out, rc);
1383         }
1384
1385         if (req->rq_resend) {
1386                 lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
1387
1388                 if (req->rq_bulk != NULL)
1389                         ptlrpc_unregister_bulk (req);
1390
1391                 DEBUG_REQ(D_HA, req, "resending: ");
1392         }
1393
1394         /* XXX this is the same as ptlrpc_set_wait */
1395         LASSERT(list_empty(&req->rq_list));
1396         list_add_tail(&req->rq_list, &imp->imp_sending_list);
1397         spin_unlock_irqrestore(&imp->imp_lock, flags);
1398
1399         rc = ptl_send_rpc(req);
1400         if (rc) {
1401                 DEBUG_REQ(D_HA, req, "send failed (%d); recovering", rc);
1402                 timeout = 1;
1403         } else {
1404                 timeout = MAX(req->rq_timeout * HZ, 1);
1405                 DEBUG_REQ(D_NET, req, "-- sleeping");
1406         }
1407         lwi = LWI_TIMEOUT_INTR(timeout, expired_request, interrupted_request,
1408                                req);
1409         l_wait_event(req->rq_reply_waitq, ptlrpc_check_reply(req), &lwi);
1410         DEBUG_REQ(D_NET, req, "-- done sleeping");
1411
1412         CDEBUG(D_RPCTRACE, "Completed RPC pname:cluuid:pid:xid:ni:nid:opc "
1413                "%s:%s:%d:"LPU64":%s:%s:%d\n", current->comm,
1414                imp->imp_obd->obd_uuid.uuid,
1415                req->rq_reqmsg->status, req->rq_xid,
1416                imp->imp_connection->c_peer.peer_ni->pni_name,
1417                ptlrpc_peernid2str(&imp->imp_connection->c_peer, str),
1418                req->rq_reqmsg->opc);
1419
1420         spin_lock_irqsave(&imp->imp_lock, flags);
1421         list_del_init(&req->rq_list);
1422         spin_unlock_irqrestore(&imp->imp_lock, flags);
1423
1424         /* If the reply was received normally, this just grabs the spinlock
1425          * (ensuring the reply callback has returned), sees that
1426          * req->rq_receiving_reply is clear and returns. */
1427         ptlrpc_unregister_reply (req);
1428
1429         if (req->rq_err)
1430                 GOTO(out, rc = -EIO);
1431
1432         /* Resend if we need to, unless we were interrupted. */
1433         if (req->rq_resend && !req->rq_intr) {
1434                 /* ...unless we were specifically told otherwise. */
1435                 if (req->rq_no_resend)
1436                         GOTO(out, rc = -ETIMEDOUT);
1437                 spin_lock_irqsave(&imp->imp_lock, flags);
1438                 goto restart;
1439         }
1440
1441         if (req->rq_intr) {
1442                 /* Should only be interrupted if we timed out. */
1443                 if (!req->rq_timedout)
1444                         DEBUG_REQ(D_ERROR, req,
1445                                   "rq_intr set but rq_timedout not");
1446                 GOTO(out, rc = -EINTR);
1447         }
1448
1449         if (req->rq_timedout) {                 /* non-recoverable timeout */
1450                 GOTO(out, rc = -ETIMEDOUT);
1451         }
1452
1453         if (!req->rq_replied) {
1454                 /* How can this be? -eeb */
1455                 DEBUG_REQ(D_ERROR, req, "!rq_replied: ");
1456                 LBUG();
1457                 GOTO(out, rc = req->rq_status);
1458         }
1459
1460         rc = after_reply (req);
1461         /* NB may return +ve success rc */
1462         if (req->rq_resend) {
1463                 spin_lock_irqsave(&imp->imp_lock, flags);
1464                 goto restart;
1465         }
1466
1467  out:
1468         if (req->rq_bulk != NULL) {
1469                 if (rc >= 0) {
1470                         /* success so far.  Note that anything going wrong
1471                          * with bulk now, is EXTREMELY strange, since the
1472                          * server must have believed that the bulk
1473                          * tranferred OK before she replied with success to
1474                          * me. */
1475                         lwi = LWI_TIMEOUT(timeout, NULL, NULL);
1476                         brc = l_wait_event(req->rq_reply_waitq,
1477                                            !ptlrpc_bulk_active(req->rq_bulk),
1478                                            &lwi);
1479                         LASSERT(brc == 0 || brc == -ETIMEDOUT);
1480                         if (brc != 0) {
1481                                 LASSERT(brc == -ETIMEDOUT);
1482                                 DEBUG_REQ(D_ERROR, req, "bulk timed out");
1483                                 rc = brc;
1484                         } else if (!req->rq_bulk->bd_success) {
1485                                 DEBUG_REQ(D_ERROR, req, "bulk transfer failed");
1486                                 rc = -EIO;
1487                         }
1488                 }
1489                 if (rc < 0)
1490                         ptlrpc_unregister_bulk (req);
1491         }
1492
1493         LASSERT(!req->rq_receiving_reply);
1494         req->rq_phase = RQ_PHASE_INTERPRET;
1495
1496         atomic_dec(&imp->imp_inflight);
1497         wake_up(&imp->imp_recovery_waitq);
1498         RETURN(rc);
1499 }
1500
1501 struct ptlrpc_replay_async_args {
1502         int praa_old_state;
1503         int praa_old_status;
1504 };
1505
1506 static int ptlrpc_replay_interpret(struct ptlrpc_request *req,
1507                                     void * data, int rc)
1508 {
1509         struct ptlrpc_replay_async_args *aa = data;
1510         struct obd_import *imp = req->rq_import;
1511         unsigned long flags;
1512
1513         atomic_dec(&imp->imp_replay_inflight);
1514
1515         if (!req->rq_replied) {
1516                 CERROR("request replay timed out, restarting recovery\n");
1517                 GOTO(out, rc = -ETIMEDOUT);
1518         }
1519
1520 #if SWAB_PARANOIA
1521         /* Clear reply swab mask; this is a new reply in sender's byte order */
1522         req->rq_rep_swab_mask = 0;
1523 #endif
1524         LASSERT (req->rq_nob_received <= req->rq_replen);
1525         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_nob_received);
1526         if (rc) {
1527                 CERROR("unpack_rep failed: %d\n", rc);
1528                 GOTO(out, rc = -EPROTO);
1529         }
1530
1531         if (req->rq_repmsg->type == PTL_RPC_MSG_ERR &&
1532             req->rq_repmsg->status == -ENOTCONN)
1533                 GOTO(out, rc = req->rq_repmsg->status);
1534
1535         /* The transno had better not change over replay. */
1536         LASSERT(req->rq_reqmsg->transno == req->rq_repmsg->transno);
1537
1538         DEBUG_REQ(D_HA, req, "got rep");
1539
1540         /* let the callback do fixups, possibly including in the request */
1541         if (req->rq_replay_cb)
1542                 req->rq_replay_cb(req);
1543
1544         if (req->rq_replied && req->rq_repmsg->status != aa->praa_old_status) {
1545                 DEBUG_REQ(D_ERROR, req, "status %d, old was %d",
1546                           req->rq_repmsg->status, aa->praa_old_status);
1547         } else {
1548                 /* Put it back for re-replay. */
1549                 req->rq_repmsg->status = aa->praa_old_status;
1550         }
1551
1552         spin_lock_irqsave(&imp->imp_lock, flags);
1553         imp->imp_last_replay_transno = req->rq_transno;
1554         spin_unlock_irqrestore(&imp->imp_lock, flags);
1555
1556         /* continue with recovery */
1557         rc = ptlrpc_import_recovery_state_machine(imp);
1558  out:
1559         req->rq_send_state = aa->praa_old_state;
1560
1561         if (rc != 0)
1562                 /* this replay failed, so restart recovery */
1563                 ptlrpc_connect_import(imp, NULL);
1564
1565         RETURN(rc);
1566 }
1567
1568
1569 int ptlrpc_replay_req(struct ptlrpc_request *req)
1570 {
1571         struct ptlrpc_replay_async_args *aa;
1572         ENTRY;
1573
1574         LASSERT(req->rq_import->imp_state == LUSTRE_IMP_REPLAY);
1575
1576         /* Not handling automatic bulk replay yet (or ever?) */
1577         LASSERT(req->rq_bulk == NULL);
1578
1579         DEBUG_REQ(D_HA, req, "REPLAY");
1580
1581         LASSERT (sizeof (*aa) <= sizeof (req->rq_async_args));
1582         aa = (struct ptlrpc_replay_async_args *)&req->rq_async_args;
1583         memset(aa, 0, sizeof *aa);
1584
1585         /* Prepare request to be resent with ptlrpcd */
1586         aa->praa_old_state = req->rq_send_state;
1587         req->rq_send_state = LUSTRE_IMP_REPLAY;
1588         req->rq_phase = RQ_PHASE_NEW;
1589         aa->praa_old_status = req->rq_repmsg->status;
1590         req->rq_status = 0;
1591
1592         req->rq_interpret_reply = ptlrpc_replay_interpret;
1593         atomic_inc(&req->rq_import->imp_replay_inflight);
1594         ptlrpc_request_addref(req); /* ptlrpcd needs a ref */
1595
1596         ptlrpcd_add_req(req);
1597         RETURN(0);
1598 }
1599
1600 void ptlrpc_abort_inflight(struct obd_import *imp)
1601 {
1602         unsigned long flags;
1603         struct list_head *tmp, *n;
1604         ENTRY;
1605
1606         /* Make sure that no new requests get processed for this import.
1607          * ptlrpc_{queue,set}_wait must (and does) hold imp_lock while testing
1608          * this flag and then putting requests on sending_list or delayed_list.
1609          */
1610         spin_lock_irqsave(&imp->imp_lock, flags);
1611
1612         /* XXX locking?  Maybe we should remove each request with the list
1613          * locked?  Also, how do we know if the requests on the list are
1614          * being freed at this time?
1615          */
1616         list_for_each_safe(tmp, n, &imp->imp_sending_list) {
1617                 struct ptlrpc_request *req =
1618                         list_entry(tmp, struct ptlrpc_request, rq_list);
1619
1620                 DEBUG_REQ(D_HA, req, "inflight");
1621
1622                 spin_lock (&req->rq_lock);
1623                 if (req->rq_import_generation < imp->imp_generation) {
1624                         req->rq_err = 1;
1625                         ptlrpc_wake_client_req(req);
1626                 }
1627                 spin_unlock (&req->rq_lock);
1628         }
1629
1630         list_for_each_safe(tmp, n, &imp->imp_delayed_list) {
1631                 struct ptlrpc_request *req =
1632                         list_entry(tmp, struct ptlrpc_request, rq_list);
1633
1634                 DEBUG_REQ(D_HA, req, "aborting waiting req");
1635
1636                 spin_lock (&req->rq_lock);
1637                 if (req->rq_import_generation < imp->imp_generation) {
1638                         req->rq_err = 1;
1639                         ptlrpc_wake_client_req(req);
1640                 }
1641                 spin_unlock (&req->rq_lock);
1642         }
1643
1644         /* Last chance to free reqs left on the replay list, but we
1645          * will still leak reqs that haven't comitted.  */
1646         if (imp->imp_replayable)
1647                 ptlrpc_free_committed(imp);
1648
1649         spin_unlock_irqrestore(&imp->imp_lock, flags);
1650
1651         EXIT;
1652 }
1653
1654 static __u64 ptlrpc_last_xid = 0;
1655 static spinlock_t ptlrpc_last_xid_lock = SPIN_LOCK_UNLOCKED;
1656
1657 __u64 ptlrpc_next_xid(void)
1658 {
1659         __u64 tmp;
1660         spin_lock(&ptlrpc_last_xid_lock);
1661         tmp = ++ptlrpc_last_xid;
1662         spin_unlock(&ptlrpc_last_xid_lock);
1663         return tmp;
1664 }
1665
1666