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 #include <linux/lustre_sec.h>
  36
  37 #include "ptlrpc_internal.h"
  38
  39 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
  40                         struct ptlrpc_client *cl)
  41 {
  42         cl->cli_request_portal = req_portal;
  43         cl->cli_reply_portal   = rep_portal;
  44         cl->cli_name           = name;
  45 }
  46
  47 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid)
  48 {
  49         struct ptlrpc_connection *c;
  50         struct ptlrpc_peer peer;
  51         int err;
  52
  53         err = ptlrpc_uuid_to_peer(uuid, &peer);
  54         if (err != 0) {
  55                 CERROR("cannot find peer %s!\n", uuid->uuid);
  56                 return NULL;
  57         }
  58
  59         c = ptlrpc_get_connection(&peer, uuid);
  60         if (c) {
  61                 memcpy(c->c_remote_uuid.uuid,
  62                        uuid->uuid, sizeof(c->c_remote_uuid.uuid));
  63         }
  64
  65         CDEBUG(D_INFO, "%s -> %p\n", uuid->uuid, c);
  66
  67         return c;
  68 }
  69
  70 void ptlrpc_readdress_connection(struct ptlrpc_connection *conn,
  71                                  struct obd_uuid *uuid)
  72 {
  73         struct ptlrpc_peer peer;
  74         int err;
  75
  76         err = ptlrpc_uuid_to_peer(uuid, &peer);
  77         if (err != 0) {
  78                 CERROR("cannot find peer %s!\n", uuid->uuid);
  79                 return;
  80         }
  81
  82         memcpy(&conn->c_peer, &peer, sizeof (peer));
  83         return;
  84 }
  85
  86 static inline struct ptlrpc_bulk_desc *new_bulk(int npages, int type, int portal)
  87 {
  88         struct ptlrpc_bulk_desc *desc;
  89
  90         OBD_ALLOC(desc, offsetof (struct ptlrpc_bulk_desc, bd_iov[npages]));
  91         if (!desc)
  92                 return NULL;
  93
  94         spin_lock_init(&desc->bd_lock);
  95         init_waitqueue_head(&desc->bd_waitq);
  96         desc->bd_max_iov = npages;
  97         desc->bd_iov_count = 0;
  98         desc->bd_md_h = PTL_INVALID_HANDLE;
  99         desc->bd_portal = portal;
 100         desc->bd_type = type;
 101
 102         return desc;
 103 }
 104
 105 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
 106                                                int npages, int type, int portal)
 107 {
 108         struct obd_import *imp = req->rq_import;
 109         struct ptlrpc_bulk_desc *desc;
 110
 111         LASSERT(type == BULK_PUT_SINK || type == BULK_GET_SOURCE);
 112         desc = new_bulk(npages, type, portal);
 113         if (desc == NULL)
 114                 RETURN(NULL);
 115
 116         desc->bd_import_generation = req->rq_import_generation;
 117         desc->bd_import = class_import_get(imp);
 118         desc->bd_req = req;
 119
 120         desc->bd_cbid.cbid_fn  = client_bulk_callback;
 121         desc->bd_cbid.cbid_arg = desc;
 122
 123         /* This makes req own desc, and free it when she frees herself */
 124         req->rq_bulk = desc;
 125
 126         return desc;
 127 }
 128
 129 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp (struct ptlrpc_request *req,
 130                                                int npages, int type, int portal)
 131 {
 132         struct obd_export *exp = req->rq_export;
 133         struct ptlrpc_bulk_desc *desc;
 134
 135         LASSERT(type == BULK_PUT_SOURCE || type == BULK_GET_SINK);
 136
 137         desc = new_bulk(npages, type, portal);
 138         if (desc == NULL)
 139                 RETURN(NULL);
 140
 141         desc->bd_export = class_export_get(exp);
 142         desc->bd_req = req;
 143
 144         desc->bd_cbid.cbid_fn  = server_bulk_callback;
 145         desc->bd_cbid.cbid_arg = desc;
 146
 147         /* NB we don't assign rq_bulk here; server-side requests are
 148          * re-used, and the handler frees the bulk desc explicitly. */
 149
 150         return desc;
 151 }
 152
 153 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
 154                            struct page *page, int pageoffset, int len)
 155 {
 156         LASSERT(desc->bd_iov_count < desc->bd_max_iov);
 157         LASSERT(page != NULL);
 158         LASSERT(pageoffset >= 0);
 159         LASSERT(len > 0);
 160         LASSERT(pageoffset + len <= PAGE_SIZE);
 161
 162         desc->bd_nob += len;
 163
 164         ptlrpc_add_bulk_page(desc, page, pageoffset, len);
 165 }
 166
 167 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *desc)
 168 {
 169         ENTRY;
 170
 171         LASSERT(desc != NULL);
 172         LASSERT(desc->bd_iov_count != LI_POISON); /* not freed already */
 173         LASSERT(!desc->bd_network_rw);         /* network hands off or */
 174         LASSERT((desc->bd_export != NULL) ^ (desc->bd_import != NULL));
 175         if (desc->bd_export)
 176                 class_export_put(desc->bd_export);
 177         else
 178                 class_import_put(desc->bd_import);
 179
 180         OBD_FREE(desc, offsetof(struct ptlrpc_bulk_desc,
 181                                 bd_iov[desc->bd_max_iov]));
 182         EXIT;
 183 }
 184
 185 /* FIXME prep_req now should return error code other than NULL. but
 186  * this is called everywhere :(
 187  */
 188 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
 189                                        int opcode, int count, int *lengths,
 190                                        char **bufs)
 191 {
 192         struct ptlrpc_request *request;
 193         int rc;
 194         ENTRY;
 195
 196         LASSERT((unsigned long)imp > 0x1000);
 197
 198         OBD_ALLOC(request, sizeof(*request));
 199         if (!request) {
 200                 CERROR("request allocation out of memory\n");
 201                 RETURN(NULL);
 202         }
 203
 204         request->rq_import = class_import_get(imp);
 205
 206         rc = ptlrpcs_req_get_cred(request);
 207         if (rc) {
 208                 CDEBUG(D_SEC, "failed to get credential\n");
 209                 GOTO(out_free, rc);
 210         }
 211
 212         /* just a try on refresh, but we proceed even if it failed */
 213         rc = ptlrpcs_cred_refresh(request->rq_cred);
 214         if (!ptlrpcs_cred_is_uptodate(request->rq_cred)) {
 215                 CERROR("req %p: failed to refresh cred %p, rc %d, continue\n",
 216                        request, request->rq_cred, rc);
 217         }
 218
 219         rc = lustre_pack_request(request, count, lengths, bufs);
 220         if (rc) {
 221                 CERROR("cannot pack request %d\n", rc);
 222                 GOTO(out_cred, rc);
 223         }
 224         request->rq_reqmsg->version |= version;
 225
 226         if (imp->imp_server_timeout)
 227                 request->rq_timeout = obd_timeout / 2;
 228         else
 229                 request->rq_timeout = obd_timeout;
 230
 231         request->rq_send_state = LUSTRE_IMP_FULL;
 232         request->rq_type = PTL_RPC_MSG_REQUEST;
 233
 234         request->rq_req_cbid.cbid_fn  = request_out_callback;
 235         request->rq_req_cbid.cbid_arg = request;
 236
 237         request->rq_reply_cbid.cbid_fn  = reply_in_callback;
 238         request->rq_reply_cbid.cbid_arg = request;
 239
 240         request->rq_phase = RQ_PHASE_NEW;
 241
 242         /* XXX FIXME bug 249 */
 243         request->rq_request_portal = imp->imp_client->cli_request_portal;
 244         request->rq_reply_portal = imp->imp_client->cli_reply_portal;
 245
 246         spin_lock_init(&request->rq_lock);
 247         INIT_LIST_HEAD(&request->rq_list);
 248         INIT_LIST_HEAD(&request->rq_replay_list);
 249         INIT_LIST_HEAD(&request->rq_set_chain);
 250         init_waitqueue_head(&request->rq_reply_waitq);
 251         request->rq_xid = ptlrpc_next_xid();
 252         atomic_set(&request->rq_refcount, 1);
 253
 254         request->rq_reqmsg->opc = opcode;
 255         request->rq_reqmsg->flags = 0;
 256         RETURN(request);
 257 out_cred:
 258         ptlrpcs_req_drop_cred(request);
 259 out_free:
 260         class_import_put(imp);
 261         OBD_FREE(request, sizeof(*request));
 262         RETURN(NULL);
 263 }
 264
 265 struct ptlrpc_request_set *ptlrpc_prep_set(void)
 266 {
 267         struct ptlrpc_request_set *set;
 268
 269         OBD_ALLOC(set, sizeof *set);
 270         if (!set)
 271                 RETURN(NULL);
 272         INIT_LIST_HEAD(&set->set_requests);
 273         init_waitqueue_head(&set->set_waitq);
 274         set->set_remaining = 0;
 275         spin_lock_init(&set->set_new_req_lock);
 276         INIT_LIST_HEAD(&set->set_new_requests);
 277
 278         RETURN(set);
 279 }
 280
 281 /* Finish with this set; opposite of prep_set. */
 282 void ptlrpc_set_destroy(struct ptlrpc_request_set *set)
 283 {
 284         struct list_head *tmp;
 285         struct list_head *next;
 286         int               expected_phase;
 287         int               n = 0;
 288         ENTRY;
 289
 290         /* Requests on the set should either all be completed, or all be new */
 291         expected_phase = (set->set_remaining == 0) ?
 292                          RQ_PHASE_COMPLETE : RQ_PHASE_NEW;
 293         list_for_each (tmp, &set->set_requests) {
 294                 struct ptlrpc_request *req =
 295                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 296
 297                 LASSERT(req->rq_phase == expected_phase);
 298                 n++;
 299         }
 300
 301         LASSERT(set->set_remaining == 0 || set->set_remaining == n);
 302
 303         list_for_each_safe(tmp, next, &set->set_requests) {
 304                 struct ptlrpc_request *req =
 305                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 306                 list_del_init(&req->rq_set_chain);
 307
 308                 LASSERT(req->rq_phase == expected_phase);
 309
 310                 if (req->rq_phase == RQ_PHASE_NEW) {
 311
 312                         if (req->rq_interpret_reply != NULL) {
 313                                 int (*interpreter)(struct ptlrpc_request *,
 314                                                    void *, int) =
 315                                         req->rq_interpret_reply;
 316
 317                                 /* higher level (i.e. LOV) failed;
 318                                  * let the sub reqs clean up */
 319                                 req->rq_status = -EBADR;
 320                                 interpreter(req, &req->rq_async_args,
 321                                             req->rq_status);
 322                         }
 323                         set->set_remaining--;
 324                 }
 325
 326                 req->rq_set = NULL;
 327                 ptlrpc_req_finished (req);
 328         }
 329
 330         LASSERT(set->set_remaining == 0);
 331
 332         OBD_FREE(set, sizeof(*set));
 333         EXIT;
 334 }
 335
 336 void ptlrpc_set_add_req(struct ptlrpc_request_set *set,
 337                         struct ptlrpc_request *req)
 338 {
 339         /* The set takes over the caller's request reference */
 340         list_add_tail(&req->rq_set_chain, &set->set_requests);
 341         req->rq_set = set;
 342         set->set_remaining++;
 343         atomic_inc(&req->rq_import->imp_inflight);
 344 }
 345
 346 /* lock so many callers can add things, the context that owns the set
 347  * is supposed to notice these and move them into the set proper. */
 348 void ptlrpc_set_add_new_req(struct ptlrpc_request_set *set,
 349                             struct ptlrpc_request *req)
 350 {
 351         unsigned long flags;
 352         spin_lock_irqsave(&set->set_new_req_lock, flags);
 353         /* The set takes over the caller's request reference */
 354         list_add_tail(&req->rq_set_chain, &set->set_new_requests);
 355         req->rq_set = set;
 356         spin_unlock_irqrestore(&set->set_new_req_lock, flags);
 357 }
 358
 359 /*
 360  * Based on the current state of the import, determine if the request
 361  * can be sent, is an error, or should be delayed.
 362  *
 363  * Returns true if this request should be delayed. If false, and
 364  * *status is set, then the request can not be sent and *status is the
 365  * error code.  If false and status is 0, then request can be sent.
 366  *
 367  * The imp->imp_lock must be held.
 368  */
 369 static int ptlrpc_import_delay_req(struct obd_import *imp,
 370                                    struct ptlrpc_request *req, int *status)
 371 {
 372         int delay = 0;
 373         ENTRY;
 374
 375         LASSERT (status != NULL);
 376         *status = 0;
 377
 378         if (imp->imp_state == LUSTRE_IMP_NEW) {
 379                 DEBUG_REQ(D_ERROR, req, "Uninitialized import.");
 380                 *status = -EIO;
 381                 LBUG();
 382         }
 383         else if (imp->imp_state == LUSTRE_IMP_CLOSED) {
 384                 DEBUG_REQ(D_ERROR, req, "IMP_CLOSED ");
 385                 *status = -EIO;
 386         }
 387         /* allow CONNECT even if import is invalid */
 388         else if (req->rq_send_state == LUSTRE_IMP_CONNECTING &&
 389                  imp->imp_state == LUSTRE_IMP_CONNECTING) {
 390                 ;
 391         }
 392         /*
 393          * If the import has been invalidated (such as by an OST failure), the
 394          * request must fail with -EIO.
 395          */
 396         else if (imp->imp_invalid) {
 397                 DEBUG_REQ(D_ERROR, req, "IMP_INVALID");
 398                 *status = -EIO;
 399         }
 400         else if (req->rq_import_generation != imp->imp_generation) {
 401                 DEBUG_REQ(D_ERROR, req, "req wrong generation:");
 402                 *status = -EIO;
 403         }
 404         else if (req->rq_send_state != imp->imp_state) {
 405                 if (imp->imp_obd->obd_no_recov || imp->imp_dlm_fake
 406                     || req->rq_no_delay)
 407                         *status = -EWOULDBLOCK;
 408                 else
 409                         delay = 1;
 410         }
 411
 412         RETURN(delay);
 413 }
 414
 415 static int ptlrpc_check_reply(struct ptlrpc_request *req)
 416 {
 417         unsigned long flags;
 418         int rc = 0;
 419         ENTRY;
 420
 421         /* serialise with network callback */
 422         spin_lock_irqsave (&req->rq_lock, flags);
 423
 424         if (req->rq_replied) {
 425                 DEBUG_REQ(D_NET, req, "REPLIED:");
 426                 GOTO(out, rc = 1);
 427         }
 428
 429         if (req->rq_net_err && !req->rq_timedout) {
 430                 spin_unlock_irqrestore (&req->rq_lock, flags);
 431                 rc = ptlrpc_expire_one_request(req);
 432                 spin_lock_irqsave (&req->rq_lock, flags);
 433                 GOTO(out, rc);
 434         }
 435
 436         if (req->rq_err) {
 437                 DEBUG_REQ(D_ERROR, req, "ABORTED:");
 438                 GOTO(out, rc = 1);
 439         }
 440
 441         if (req->rq_resend) {
 442                 DEBUG_REQ(D_ERROR, req, "RESEND:");
 443                 GOTO(out, rc = 1);
 444         }
 445
 446         if (req->rq_restart) {
 447                 DEBUG_REQ(D_ERROR, req, "RESTART:");
 448                 GOTO(out, rc = 1);
 449         }
 450         EXIT;
 451  out:
 452         spin_unlock_irqrestore (&req->rq_lock, flags);
 453         DEBUG_REQ(D_NET, req, "rc = %d for", rc);
 454         return rc;
 455 }
 456
 457 static int ptlrpc_check_status(struct ptlrpc_request *req)
 458 {
 459         int err;
 460         ENTRY;
 461
 462         err = req->rq_repmsg->status;
 463         if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
 464                 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err == %d",
 465                           err);
 466                 RETURN(err < 0 ? err : -EINVAL);
 467         }
 468
 469         if (err < 0) {
 470                 DEBUG_REQ(D_INFO, req, "status is %d", err);
 471         } else if (err > 0) {
 472                 /* XXX: translate this error from net to host */
 473                 DEBUG_REQ(D_INFO, req, "status is %d", err);
 474         }
 475
 476         RETURN(err);
 477 }
 478
 479 static int after_reply(struct ptlrpc_request *req)
 480 {
 481         unsigned long flags;
 482         struct obd_import *imp = req->rq_import;
 483         int rc;
 484         ENTRY;
 485
 486         LASSERT(!req->rq_receiving_reply);
 487
 488         /* NB Until this point, the whole of the incoming message,
 489          * including buflens, status etc is in the sender's byte order. */
 490
 491 #if SWAB_PARANOIA
 492         /* Clear reply swab mask; this is a new reply in sender's byte order */
 493         req->rq_rep_swab_mask = 0;
 494 #endif
 495         LASSERT (req->rq_nob_received <= req->rq_repbuf_len);
 496         rc = ptlrpcs_cli_unwrap_reply(req);
 497         if (rc) {
 498                 CERROR("verify reply error: %d\n", rc);
 499                 RETURN(rc);
 500         }
 501         /* unwrap_reply may request rpc be resend */
 502         if (req->rq_ptlrpcs_restart) {
 503                 req->rq_resend = 1;
 504                 RETURN(0);
 505         }
 506
 507         /* unwrap_reply will set rq_replen as the actual received
 508          * lustre_msg length
 509          */
 510         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
 511         if (rc) {
 512                 CERROR("unpack_rep failed: %d\n", rc);
 513                 RETURN(-EPROTO);
 514         }
 515
 516         if (req->rq_repmsg->type != PTL_RPC_MSG_REPLY &&
 517             req->rq_repmsg->type != PTL_RPC_MSG_ERR) {
 518                 CERROR("invalid packet type received (type=%u)\n",
 519                        req->rq_repmsg->type);
 520                 RETURN(-EPROTO);
 521         }
 522
 523         rc = ptlrpc_check_status(req);
 524
 525         /* Either we've been evicted, or the server has failed for
 526          * some reason. Try to reconnect, and if that fails, punt to the
 527          * upcall. */
 528         if (rc == -ENOTCONN) {
 529                 if (req->rq_send_state != LUSTRE_IMP_FULL ||
 530                     imp->imp_obd->obd_no_recov || imp->imp_dlm_fake) {
 531                         RETURN(-ENOTCONN);
 532                 }
 533
 534                 ptlrpc_request_handle_notconn(req);
 535
 536                 RETURN(rc);
 537         }
 538
 539         /* Store transno in reqmsg for replay. */
 540         req->rq_reqmsg->transno = req->rq_transno = req->rq_repmsg->transno;
 541
 542
 543         if (req->rq_import->imp_replayable) {
 544                 spin_lock_irqsave(&imp->imp_lock, flags);
 545                 if (req->rq_transno != 0)
 546                         ptlrpc_retain_replayable_request(req, imp);
 547                 else if (req->rq_commit_cb != NULL) {
 548                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 549                         req->rq_commit_cb(req);
 550                         spin_lock_irqsave(&imp->imp_lock, flags);
 551                 }
 552
 553                 if (req->rq_transno > imp->imp_max_transno)
 554                         imp->imp_max_transno = req->rq_transno;
 555
 556                 /* Replay-enabled imports return commit-status information. */
 557                 if (req->rq_repmsg->last_committed)
 558                         imp->imp_peer_committed_transno =
 559                                 req->rq_repmsg->last_committed;
 560                 ptlrpc_free_committed(imp);
 561                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 562         }
 563
 564         RETURN(rc);
 565 }
 566
 567 static int ptlrpc_send_new_req(struct ptlrpc_request *req)
 568 {
 569         char                   str[PTL_NALFMT_SIZE];
 570         struct obd_import     *imp;
 571         unsigned long          flags;
 572         int rc;
 573         ENTRY;
 574
 575         LASSERT(req->rq_phase == RQ_PHASE_NEW);
 576         req->rq_phase = RQ_PHASE_RPC;
 577
 578         imp = req->rq_import;
 579         spin_lock_irqsave(&imp->imp_lock, flags);
 580
 581         req->rq_import_generation = imp->imp_generation;
 582
 583         if (ptlrpc_import_delay_req(imp, req, &rc)) {
 584                 spin_lock (&req->rq_lock);
 585                 req->rq_waiting = 1;
 586                 spin_unlock (&req->rq_lock);
 587
 588                 DEBUG_REQ(D_HA, req, "req from PID %d waiting for recovery: "
 589                           "(%s != %s)",
 590                           req->rq_reqmsg->status,
 591                           ptlrpc_import_state_name(req->rq_send_state),
 592                           ptlrpc_import_state_name(imp->imp_state));
 593                 LASSERT(list_empty (&req->rq_list));
 594
 595                 list_add_tail(&req->rq_list, &imp->imp_delayed_list);
 596                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 597                 RETURN(0);
 598         }
 599
 600         if (rc != 0) {
 601                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 602                 req->rq_status = rc;
 603                 req->rq_phase = RQ_PHASE_INTERPRET;
 604                 RETURN(rc);
 605         }
 606
 607         /* XXX this is the same as ptlrpc_queue_wait */
 608         LASSERT(list_empty(&req->rq_list));
 609         list_add_tail(&req->rq_list, &imp->imp_sending_list);
 610         spin_unlock_irqrestore(&imp->imp_lock, flags);
 611
 612         req->rq_reqmsg->status = current->pid;
 613         CDEBUG(D_RPCTRACE, "Sending RPC pname:cluuid:pid:xid:ni:nid:opc"
 614                " %s:%s:%d:"LPU64":%s:%s:%d\n", current->comm,
 615                imp->imp_obd->obd_uuid.uuid, req->rq_reqmsg->status,
 616                req->rq_xid,
 617                imp->imp_connection->c_peer.peer_ni->pni_name,
 618                ptlrpc_peernid2str(&imp->imp_connection->c_peer, str),
 619                req->rq_reqmsg->opc);
 620
 621         rc = ptl_send_rpc(req);
 622         if (rc) {
 623                 DEBUG_REQ(D_HA, req, "send failed (%d); expect timeout", rc);
 624                 req->rq_net_err = 1;
 625                 RETURN(rc);
 626         }
 627         RETURN(0);
 628 }
 629
 630 int ptlrpc_check_set(struct ptlrpc_request_set *set)
 631 {
 632         char str[PTL_NALFMT_SIZE];
 633         unsigned long flags;
 634         struct list_head *tmp;
 635         int force_timer_recalc = 0;
 636         ENTRY;
 637
 638         if (set->set_remaining == 0)
 639                 RETURN(1);
 640
 641         list_for_each(tmp, &set->set_requests) {
 642                 struct ptlrpc_request *req =
 643                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 644                 struct obd_import *imp = req->rq_import;
 645                 int rc = 0;
 646
 647                 if (req->rq_phase == RQ_PHASE_NEW &&
 648                     ptlrpc_send_new_req(req)) {
 649                         force_timer_recalc = 1;
 650                 }
 651
 652                 if (!(req->rq_phase == RQ_PHASE_RPC ||
 653                       req->rq_phase == RQ_PHASE_BULK ||
 654                       req->rq_phase == RQ_PHASE_INTERPRET ||
 655                       req->rq_phase == RQ_PHASE_COMPLETE)) {
 656                         DEBUG_REQ(D_ERROR, req, "bad phase %x", req->rq_phase);
 657                         LBUG();
 658                 }
 659
 660                 if (req->rq_phase == RQ_PHASE_COMPLETE)
 661                         continue;
 662
 663                 if (req->rq_phase == RQ_PHASE_INTERPRET)
 664                         GOTO(interpret, req->rq_status);
 665
 666                 if (req->rq_net_err && !req->rq_timedout)
 667                         ptlrpc_expire_one_request(req);
 668
 669                 if (req->rq_err) {
 670                         ptlrpc_unregister_reply(req);
 671                         if (req->rq_status == 0)
 672                                 req->rq_status = -EIO;
 673                         req->rq_phase = RQ_PHASE_INTERPRET;
 674
 675                         spin_lock_irqsave(&imp->imp_lock, flags);
 676                         list_del_init(&req->rq_list);
 677                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 678
 679                         GOTO(interpret, req->rq_status);
 680                 }
 681
 682                 /* ptlrpc_queue_wait->l_wait_event guarantees that rq_intr
 683                  * will only be set after rq_timedout, but the oig waiting
 684                  * path sets rq_intr irrespective of whether ptlrpcd has
 685                  * seen a timeout.  our policy is to only interpret
 686                  * interrupted rpcs after they have timed out */
 687                 if (req->rq_intr && (req->rq_timedout || req->rq_waiting)) {
 688                         /* NB could be on delayed list */
 689                         ptlrpc_unregister_reply(req);
 690                         req->rq_status = -EINTR;
 691                         req->rq_phase = RQ_PHASE_INTERPRET;
 692
 693                         spin_lock_irqsave(&imp->imp_lock, flags);
 694                         list_del_init(&req->rq_list);
 695                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 696
 697                         GOTO(interpret, req->rq_status);
 698                 }
 699
 700                 if (req->rq_phase == RQ_PHASE_RPC) {
 701                         if (req->rq_timedout||req->rq_waiting||req->rq_resend) {
 702                                 int status;
 703
 704                                 ptlrpc_unregister_reply(req);
 705
 706                                 spin_lock_irqsave(&imp->imp_lock, flags);
 707
 708                                 if (ptlrpc_import_delay_req(imp, req, &status)){
 709                                         spin_unlock_irqrestore(&imp->imp_lock,
 710                                                                flags);
 711                                         continue;
 712                                 }
 713
 714                                 list_del_init(&req->rq_list);
 715                                 if (status != 0)  {
 716                                         req->rq_status = status;
 717                                         req->rq_phase = RQ_PHASE_INTERPRET;
 718                                         spin_unlock_irqrestore(&imp->imp_lock,
 719                                                                flags);
 720                                         GOTO(interpret, req->rq_status);
 721                                 }
 722                                 if (req->rq_no_resend) {
 723                                         req->rq_status = -ENOTCONN;
 724                                         req->rq_phase = RQ_PHASE_INTERPRET;
 725                                         spin_unlock_irqrestore(&imp->imp_lock,
 726                                                                flags);
 727                                         GOTO(interpret, req->rq_status);
 728                                 }
 729                                 list_add_tail(&req->rq_list,
 730                                               &imp->imp_sending_list);
 731
 732                                 spin_unlock_irqrestore(&imp->imp_lock, flags);
 733
 734                                 req->rq_waiting = 0;
 735                                 if (req->rq_resend) {
 736                                         if (!req->rq_ptlrpcs_restart)
 737                                                 lustre_msg_add_flags(
 738                                                         req->rq_reqmsg,
 739                                                         MSG_RESENT);
 740                                         if (req->rq_bulk) {
 741                                                 __u64 old_xid = req->rq_xid;
 742
 743                                                 ptlrpc_unregister_bulk (req);
 744
 745                                                 /* ensure previous bulk fails */
 746                                                 req->rq_xid = ptlrpc_next_xid();
 747                                                 CDEBUG(D_HA, "resend bulk "
 748                                                        "old x"LPU64
 749                                                        " new x"LPU64"\n",
 750                                                        old_xid, req->rq_xid);
 751                                         }
 752                                 }
 753
 754                                 rc = ptl_send_rpc(req);
 755                                 if (rc) {
 756                                         DEBUG_REQ(D_HA, req, "send failed (%d)",
 757                                                   rc);
 758                                         force_timer_recalc = 1;
 759                                         req->rq_net_err = 1;
 760                                 }
 761                                 /* need to reset the timeout */
 762                                 force_timer_recalc = 1;
 763                         }
 764
 765                         /* Still waiting for a reply? */
 766                         if (ptlrpc_client_receiving_reply(req))
 767                                 continue;
 768
 769                         /* Did we actually receive a reply? */
 770                         if (!ptlrpc_client_replied(req))
 771                                 continue;
 772
 773                         spin_lock_irqsave(&imp->imp_lock, flags);
 774                         list_del_init(&req->rq_list);
 775                         spin_unlock_irqrestore(&imp->imp_lock, flags);
 776
 777                         req->rq_status = after_reply(req);
 778                         if (req->rq_resend) {
 779                                 /* Add this req to the delayed list so
 780                                    it can be errored if the import is
 781                                    evicted after recovery. */
 782                                 spin_lock_irqsave (&req->rq_lock, flags);
 783                                 list_add_tail(&req->rq_list,
 784                                               &imp->imp_delayed_list);
 785                                 spin_unlock_irqrestore(&req->rq_lock, flags);
 786                                 continue;
 787                         }
 788
 789                         /* If there is no bulk associated with this request,
 790                          * then we're done and should let the interpreter
 791                          * process the reply.  Similarly if the RPC returned
 792                          * an error, and therefore the bulk will never arrive.
 793                          */
 794                         if (req->rq_bulk == NULL || req->rq_status != 0) {
 795                                 req->rq_phase = RQ_PHASE_INTERPRET;
 796                                 GOTO(interpret, req->rq_status);
 797                         }
 798
 799                         req->rq_phase = RQ_PHASE_BULK;
 800                 }
 801
 802                 LASSERT(req->rq_phase == RQ_PHASE_BULK);
 803                 if (ptlrpc_bulk_active(req->rq_bulk))
 804                         continue;
 805
 806                 if (!req->rq_bulk->bd_success) {
 807                         /* The RPC reply arrived OK, but the bulk screwed
 808                          * up!  Dead wierd since the server told us the RPC
 809                          * was good after getting the REPLY for her GET or
 810                          * the ACK for her PUT. */
 811                         DEBUG_REQ(D_ERROR, req, "bulk transfer failed");
 812                         LBUG();
 813                 }
 814
 815                 req->rq_phase = RQ_PHASE_INTERPRET;
 816
 817         interpret:
 818                 LASSERT(req->rq_phase == RQ_PHASE_INTERPRET);
 819                 LASSERT(!req->rq_receiving_reply);
 820
 821                 ptlrpc_unregister_reply(req);
 822                 if (req->rq_bulk != NULL)
 823                         ptlrpc_unregister_bulk (req);
 824
 825                 req->rq_phase = RQ_PHASE_COMPLETE;
 826
 827                 if (req->rq_interpret_reply != NULL) {
 828                         int (*interpreter)(struct ptlrpc_request *,void *,int) =
 829                                 req->rq_interpret_reply;
 830                         req->rq_status = interpreter(req, &req->rq_async_args,
 831                                                      req->rq_status);
 832                 }
 833
 834                 CDEBUG(D_RPCTRACE, "Completed RPC pname:cluuid:pid:xid:ni:nid:"
 835                        "opc %s:%s:%d:"LPU64":%s:%s:%d\n", current->comm,
 836                        imp->imp_obd->obd_uuid.uuid, req->rq_reqmsg->status,
 837                        req->rq_xid,
 838                        imp->imp_connection->c_peer.peer_ni->pni_name,
 839                        ptlrpc_peernid2str(&imp->imp_connection->c_peer, str),
 840                        req->rq_reqmsg->opc);
 841
 842                 set->set_remaining--;
 843
 844                 atomic_dec(&imp->imp_inflight);
 845                 wake_up(&imp->imp_recovery_waitq);
 846         }
 847
 848         /* If we hit an error, we want to recover promptly. */
 849         RETURN(set->set_remaining == 0 || force_timer_recalc);
 850 }
 851
 852 int ptlrpc_expire_one_request(struct ptlrpc_request *req)
 853 {
 854         unsigned long      flags;
 855         struct obd_import *imp = req->rq_import;
 856         int replied = 0;
 857         ENTRY;
 858
 859         DEBUG_REQ(D_ERROR, req, "timeout (sent at %lu, %lus ago)",
 860                   (long)req->rq_sent, LTIME_S(CURRENT_TIME) - req->rq_sent);
 861
 862         spin_lock_irqsave (&req->rq_lock, flags);
 863         replied = req->rq_replied;
 864         if (!replied)
 865                 req->rq_timedout = 1;
 866         spin_unlock_irqrestore (&req->rq_lock, flags);
 867
 868         if (replied)
 869                 RETURN(0);
 870
 871         ptlrpc_unregister_reply (req);
 872
 873         if (obd_dump_on_timeout)
 874                 portals_debug_dumplog();
 875
 876         if (req->rq_bulk != NULL)
 877                 ptlrpc_unregister_bulk (req);
 878
 879         if (imp == NULL) {
 880                 DEBUG_REQ(D_HA, req, "NULL import: already cleaned up?");
 881                 RETURN(1);
 882         }
 883
 884         /* The DLM server doesn't want recovery run on its imports. */
 885         if (imp->imp_dlm_fake)
 886                 RETURN(1);
 887
 888         /* If this request is for recovery or other primordial tasks,
 889          * then error it out here. */
 890         if (req->rq_send_state != LUSTRE_IMP_FULL ||
 891             imp->imp_obd->obd_no_recov) {
 892                 spin_lock_irqsave (&req->rq_lock, flags);
 893                 req->rq_status = -ETIMEDOUT;
 894                 req->rq_err = 1;
 895                 spin_unlock_irqrestore (&req->rq_lock, flags);
 896                 RETURN(1);
 897         }
 898
 899         ptlrpc_fail_import(imp, req->rq_import_generation);
 900
 901         RETURN(0);
 902 }
 903
 904 int ptlrpc_expired_set(void *data)
 905 {
 906         struct ptlrpc_request_set *set = data;
 907         struct list_head          *tmp;
 908         time_t                     now = LTIME_S(CURRENT_TIME);
 909         ENTRY;
 910
 911         LASSERT(set != NULL);
 912
 913         /* A timeout expired; see which reqs it applies to... */
 914         list_for_each (tmp, &set->set_requests) {
 915                 struct ptlrpc_request *req =
 916                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 917
 918                 /* request in-flight? */
 919                 if (!((req->rq_phase == RQ_PHASE_RPC && !req->rq_waiting
 920                        && !req->rq_resend) ||
 921                       (req->rq_phase == RQ_PHASE_BULK)))
 922                         continue;
 923
 924                 if (req->rq_timedout ||           /* already dealt with */
 925                     req->rq_sent + req->rq_timeout > now) /* not expired */
 926                         continue;
 927
 928                 /* deal with this guy */
 929                 ptlrpc_expire_one_request (req);
 930         }
 931
 932         /* When waiting for a whole set, we always to break out of the
 933          * sleep so we can recalculate the timeout, or enable interrupts
 934          * iff everyone's timed out.
 935          */
 936         RETURN(1);
 937 }
 938
 939 void ptlrpc_mark_interrupted(struct ptlrpc_request *req)
 940 {
 941         unsigned long flags;
 942         spin_lock_irqsave(&req->rq_lock, flags);
 943         req->rq_intr = 1;
 944         spin_unlock_irqrestore(&req->rq_lock, flags);
 945 }
 946
 947 void ptlrpc_interrupted_set(void *data)
 948 {
 949         struct ptlrpc_request_set *set = data;
 950         struct list_head *tmp;
 951
 952         LASSERT(set != NULL);
 953         CERROR("INTERRUPTED SET %p\n", set);
 954
 955         list_for_each(tmp, &set->set_requests) {
 956                 struct ptlrpc_request *req =
 957                         list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 958
 959                 if (req->rq_phase != RQ_PHASE_RPC)
 960                         continue;
 961
 962                 ptlrpc_mark_interrupted(req);
 963         }
 964 }
 965
 966 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *set)
 967 {
 968         struct list_head      *tmp;
 969         time_t                 now = LTIME_S(CURRENT_TIME);
 970         time_t                 deadline;
 971         int                    timeout = 0;
 972         struct ptlrpc_request *req;
 973         ENTRY;
 974
 975         SIGNAL_MASK_ASSERT(); /* XXX BUG 1511 */
 976
 977         list_for_each(tmp, &set->set_requests) {
 978                 req = list_entry(tmp, struct ptlrpc_request, rq_set_chain);
 979
 980                 /* request in-flight? */
 981                 if (!((req->rq_phase == RQ_PHASE_RPC && !req->rq_waiting) ||
 982                       (req->rq_phase == RQ_PHASE_BULK)))
 983                         continue;
 984
 985                 if (req->rq_timedout)   /* already timed out */
 986                         continue;
 987
 988                 deadline = req->rq_sent + req->rq_timeout;
 989                 if (deadline <= now)    /* actually expired already */
 990                         timeout = 1;    /* ASAP */
 991                 else if (timeout == 0 || timeout > deadline - now)
 992                         timeout = deadline - now;
 993         }
 994         RETURN(timeout);
 995 }
 996
 997
 998 int ptlrpc_set_wait(struct ptlrpc_request_set *set)
 999 {
1000         struct list_head      *tmp;
1001         struct ptlrpc_request *req;
1002         struct l_wait_info     lwi;
1003         int                    rc, timeout;
1004         ENTRY;
1005
1006         LASSERT(!list_empty(&set->set_requests));
1007         list_for_each(tmp, &set->set_requests) {
1008                 req = list_entry(tmp, struct ptlrpc_request, rq_set_chain);
1009                 if (req->rq_phase == RQ_PHASE_NEW)
1010                         (void)ptlrpc_send_new_req(req);
1011         }
1012
1013         do {
1014                 timeout = ptlrpc_set_next_timeout(set);
1015
1016                 /* wait until all complete, interrupted, or an in-flight
1017                  * req times out */
1018                 CDEBUG(D_HA, "set %p going to sleep for %d seconds\n",
1019                        set, timeout);
1020                 lwi = LWI_TIMEOUT_INTR((timeout ? timeout : 1) * HZ,
1021                                        ptlrpc_expired_set,
1022                                        ptlrpc_interrupted_set, set);
1023                 rc = l_wait_event(set->set_waitq, ptlrpc_check_set(set), &lwi);
1024
1025                 LASSERT(rc == 0 || rc == -EINTR || rc == -ETIMEDOUT);
1026
1027                 /* -EINTR => all requests have been flagged rq_intr so next
1028                  * check completes.
1029                  * -ETIMEOUTD => someone timed out.  When all reqs have
1030                  * timed out, signals are enabled allowing completion with
1031                  * EINTR.
1032                  * I don't really care if we go once more round the loop in
1033                  * the error cases -eeb. */
1034         } while (rc != 0 || set->set_remaining != 0);
1035
1036         LASSERT(set->set_remaining == 0);
1037
1038         rc = 0;
1039         list_for_each(tmp, &set->set_requests) {
1040                 req = list_entry(tmp, struct ptlrpc_request, rq_set_chain);
1041
1042                 LASSERT(req->rq_phase == RQ_PHASE_COMPLETE);
1043                 if (req->rq_status != 0)
1044                         rc = req->rq_status;
1045         }
1046
1047         if (set->set_interpret != NULL) {
1048                 int (*interpreter)(struct ptlrpc_request_set *set,void *,int) =
1049                         set->set_interpret;
1050                 rc = interpreter (set, set->set_arg, rc);
1051         }
1052
1053         RETURN(rc);
1054 }
1055
1056 static void __ptlrpc_free_req(struct ptlrpc_request *request, int locked)
1057 {
1058         ENTRY;
1059         if (request == NULL) {
1060                 EXIT;
1061                 return;
1062         }
1063
1064         LASSERTF(!request->rq_receiving_reply, "req %p\n", request);
1065         LASSERTF(request->rq_rqbd == NULL, "req %p\n",request);/* client-side */
1066         LASSERTF(list_empty(&request->rq_list), "req %p\n", request);
1067         LASSERTF(list_empty(&request->rq_set_chain), "req %p\n", request);
1068         LASSERT(request->rq_cred);
1069
1070         /* We must take it off the imp_replay_list first.  Otherwise, we'll set
1071          * request->rq_reqmsg to NULL while osc_close is dereferencing it. */
1072         if (request->rq_import != NULL) {
1073                 unsigned long flags = 0;
1074                 if (!locked)
1075                         spin_lock_irqsave(&request->rq_import->imp_lock, flags);
1076                 list_del_init(&request->rq_replay_list);
1077                 if (!locked)
1078                         spin_unlock_irqrestore(&request->rq_import->imp_lock,
1079                                                flags);
1080         }
1081         LASSERTF(list_empty(&request->rq_replay_list), "req %p\n", request);
1082
1083         if (atomic_read(&request->rq_refcount) != 0) {
1084                 DEBUG_REQ(D_ERROR, request,
1085                           "freeing request with nonzero refcount");
1086                 LBUG();
1087         }
1088
1089         if (request->rq_repbuf != NULL)
1090                 ptlrpcs_cli_free_repbuf(request);
1091         if (request->rq_reqbuf != NULL)
1092                 ptlrpcs_cli_free_reqbuf(request);
1093
1094         if (request->rq_export != NULL) {
1095                 class_export_put(request->rq_export);
1096                 request->rq_export = NULL;
1097         }
1098         if (request->rq_import != NULL) {
1099                 class_import_put(request->rq_import);
1100                 request->rq_import = NULL;
1101         }
1102         if (request->rq_bulk != NULL)
1103                 ptlrpc_free_bulk(request->rq_bulk);
1104
1105         ptlrpcs_req_drop_cred(request);
1106         OBD_FREE(request, sizeof(*request));
1107         EXIT;
1108 }
1109
1110 void ptlrpc_free_req(struct ptlrpc_request *request)
1111 {
1112         __ptlrpc_free_req(request, 0);
1113 }
1114
1115 static int __ptlrpc_req_finished(struct ptlrpc_request *request, int locked);
1116 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request)
1117 {
1118         LASSERT_SPIN_LOCKED(&request->rq_import->imp_lock);
1119         (void)__ptlrpc_req_finished(request, 1);
1120 }
1121
1122 static int __ptlrpc_req_finished(struct ptlrpc_request *request, int locked)
1123 {
1124         ENTRY;
1125         if (request == NULL)
1126                 RETURN(1);
1127
1128         if (request == LP_POISON ||
1129             request->rq_reqmsg == LP_POISON) {
1130                 CERROR("dereferencing freed request (bug 575)\n");
1131                 LBUG();
1132                 RETURN(1);
1133         }
1134
1135         DEBUG_REQ(D_INFO, request, "refcount now %u",
1136                   atomic_read(&request->rq_refcount) - 1);
1137
1138         if (atomic_dec_and_test(&request->rq_refcount)) {
1139                 __ptlrpc_free_req(request, locked);
1140                 RETURN(1);
1141         }
1142
1143         RETURN(0);
1144 }
1145
1146 void ptlrpc_req_finished(struct ptlrpc_request *request)
1147 {
1148         __ptlrpc_req_finished(request, 0);
1149 }
1150
1151 /* Disengage the client's reply buffer from the network
1152  * NB does _NOT_ unregister any client-side bulk.
1153  * IDEMPOTENT, but _not_ safe against concurrent callers.
1154  * The request owner (i.e. the thread doing the I/O) must call...
1155  */
1156 void ptlrpc_unregister_reply (struct ptlrpc_request *request)
1157 {
1158         int                rc;
1159         wait_queue_head_t *wq;
1160         struct l_wait_info lwi;
1161
1162         LASSERT(!in_interrupt ());             /* might sleep */
1163
1164         if (!ptlrpc_client_receiving_reply(request))
1165                 return;
1166
1167         PtlMDUnlink (request->rq_reply_md_h);
1168
1169         /* We have to l_wait_event() whatever the result, to give liblustre
1170          * a chance to run reply_in_callback() */
1171
1172         if (request->rq_set != NULL)
1173                 wq = &request->rq_set->set_waitq;
1174         else
1175                 wq = &request->rq_reply_waitq;
1176
1177         for (;;) {
1178                 /* Network access will complete in finite time but the HUGE
1179                  * timeout lets us CWARN for visibility of sluggish NALs */
1180                 lwi = LWI_TIMEOUT(300 * HZ, NULL, NULL);
1181                 rc = l_wait_event (*wq, !ptlrpc_client_receiving_reply(request), &lwi);
1182                 if (rc == 0)
1183                         return;
1184
1185                 LASSERT (rc == -ETIMEDOUT);
1186                 DEBUG_REQ(D_WARNING, request, "Unexpectedly long timeout");
1187         }
1188 }
1189
1190 /* caller must hold imp->imp_lock */
1191 void ptlrpc_free_committed(struct obd_import *imp)
1192 {
1193         struct list_head *tmp, *saved;
1194         struct ptlrpc_request *req;
1195         struct ptlrpc_request *last_req = NULL; /* temporary fire escape */
1196         ENTRY;
1197
1198         LASSERT(imp != NULL);
1199
1200         LASSERT_SPIN_LOCKED(&imp->imp_lock);
1201
1202         CDEBUG(D_HA, "%s: committing for last_committed "LPU64"\n",
1203                imp->imp_obd->obd_name, imp->imp_peer_committed_transno);
1204
1205         list_for_each_safe(tmp, saved, &imp->imp_replay_list) {
1206                 req = list_entry(tmp, struct ptlrpc_request, rq_replay_list);
1207
1208                 /* XXX ok to remove when 1357 resolved - rread 05/29/03  */
1209                 LASSERT(req != last_req);
1210                 last_req = req;
1211
1212                 if (req->rq_import_generation < imp->imp_generation) {
1213                         DEBUG_REQ(D_HA, req, "freeing request with old gen");
1214                         GOTO(free_req, 0);
1215                 }
1216
1217                 if (req->rq_replay) {
1218                         DEBUG_REQ(D_HA, req, "keeping (FL_REPLAY)");
1219                         continue;
1220                 }
1221
1222                 /* not yet committed */
1223                 if (req->rq_transno > imp->imp_peer_committed_transno) {
1224                         DEBUG_REQ(D_HA, req, "stopping search");
1225                         break;
1226                 }
1227
1228                 DEBUG_REQ(D_HA, req, "committing (last_committed "LPU64")",
1229                           imp->imp_peer_committed_transno);
1230 free_req:
1231                 if (req->rq_commit_cb != NULL)
1232                         req->rq_commit_cb(req);
1233                 list_del_init(&req->rq_replay_list);
1234                 __ptlrpc_req_finished(req, 1);
1235         }
1236
1237         EXIT;
1238         return;
1239 }
1240
1241 void ptlrpc_cleanup_client(struct obd_import *imp)
1242 {
1243         ENTRY;
1244         EXIT;
1245         return;
1246 }
1247
1248 void ptlrpc_resend_req(struct ptlrpc_request *req)
1249 {
1250         unsigned long flags;
1251
1252         DEBUG_REQ(D_HA, req, "going to resend");
1253         req->rq_reqmsg->handle.cookie = 0;
1254         req->rq_status = -EAGAIN;
1255
1256         spin_lock_irqsave (&req->rq_lock, flags);
1257         req->rq_resend = 1;
1258         req->rq_net_err = 0;
1259         req->rq_timedout = 0;
1260         if (req->rq_bulk) {
1261                 __u64 old_xid = req->rq_xid;
1262
1263                 /* ensure previous bulk fails */
1264                 req->rq_xid = ptlrpc_next_xid();
1265                 CDEBUG(D_HA, "resend bulk old x"LPU64" new x"LPU64"\n",
1266                        old_xid, req->rq_xid);
1267         }
1268         ptlrpc_wake_client_req(req);
1269         spin_unlock_irqrestore (&req->rq_lock, flags);
1270 }
1271
1272 /* XXX: this function and rq_status are currently unused */
1273 void ptlrpc_restart_req(struct ptlrpc_request *req)
1274 {
1275         unsigned long flags;
1276
1277         DEBUG_REQ(D_HA, req, "restarting (possibly-)completed request");
1278         req->rq_status = -ERESTARTSYS;
1279
1280         spin_lock_irqsave (&req->rq_lock, flags);
1281         req->rq_restart = 1;
1282         req->rq_timedout = 0;
1283         ptlrpc_wake_client_req(req);
1284         spin_unlock_irqrestore (&req->rq_lock, flags);
1285 }
1286
1287 static int expired_request(void *data)
1288 {
1289         struct ptlrpc_request *req = data;
1290         ENTRY;
1291
1292         /* some failure can suspend regular timeouts */
1293         if (ptlrpc_check_suspend())
1294                 RETURN(1);
1295
1296         RETURN(ptlrpc_expire_one_request(req));
1297 }
1298
1299 static void interrupted_request(void *data)
1300 {
1301         unsigned long flags;
1302
1303         struct ptlrpc_request *req = data;
1304         DEBUG_REQ(D_HA, req, "request interrupted");
1305         spin_lock_irqsave (&req->rq_lock, flags);
1306         req->rq_intr = 1;
1307         spin_unlock_irqrestore (&req->rq_lock, flags);
1308 }
1309
1310 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req)
1311 {
1312         ENTRY;
1313         atomic_inc(&req->rq_refcount);
1314         RETURN(req);
1315 }
1316
1317 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
1318                                       struct obd_import *imp)
1319 {
1320         struct list_head *tmp;
1321
1322         LASSERT_SPIN_LOCKED(&imp->imp_lock);
1323
1324         /* clear this  for new requests that were resent as well
1325            as resent replayed requests. */
1326         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1327
1328         /* don't re-add requests that have been replayed */
1329         if (!list_empty(&req->rq_replay_list))
1330                 return;
1331
1332         lustre_msg_add_flags(req->rq_reqmsg, MSG_REPLAY);
1333
1334         LASSERT(imp->imp_replayable);
1335         /* Balanced in ptlrpc_free_committed, usually. */
1336         ptlrpc_request_addref(req);
1337         list_for_each_prev(tmp, &imp->imp_replay_list) {
1338                 struct ptlrpc_request *iter =
1339                         list_entry(tmp, struct ptlrpc_request, rq_replay_list);
1340
1341                 /* We may have duplicate transnos if we create and then
1342                  * open a file, or for closes retained if to match creating
1343                  * opens, so use req->rq_xid as a secondary key.
1344                  * (See bugs 684, 685, and 428.)
1345                  * XXX no longer needed, but all opens need transnos!
1346                  */
1347                 if (iter->rq_transno > req->rq_transno)
1348                         continue;
1349
1350                 if (iter->rq_transno == req->rq_transno) {
1351                         LASSERT(iter->rq_xid != req->rq_xid);
1352                         if (iter->rq_xid > req->rq_xid)
1353                                 continue;
1354                 }
1355
1356                 list_add(&req->rq_replay_list, &iter->rq_replay_list);
1357                 return;
1358         }
1359
1360         list_add_tail(&req->rq_replay_list, &imp->imp_replay_list);
1361 }
1362
1363 int ptlrpc_queue_wait(struct ptlrpc_request *req)
1364 {
1365         char str[PTL_NALFMT_SIZE];
1366         int rc = 0;
1367         int brc;
1368         struct l_wait_info lwi;
1369         struct obd_import *imp = req->rq_import;
1370         unsigned long flags;
1371         int timeout = 0;
1372         ENTRY;
1373
1374         LASSERT(req->rq_set == NULL);
1375         LASSERT(!req->rq_receiving_reply);
1376         atomic_inc(&imp->imp_inflight);
1377
1378         if (imp->imp_connection == NULL) {
1379                 CERROR("request on not connected import %s\n",
1380                         imp->imp_obd->obd_name);
1381                 RETURN(-EINVAL);
1382         }
1383
1384         /* for distributed debugging */
1385         req->rq_reqmsg->status = current->pid;
1386         LASSERT(imp->imp_obd != NULL);
1387         CDEBUG(D_RPCTRACE, "Sending RPC pname:cluuid:pid:xid:ni:nid:opc "
1388                "%s:%s:%d:"LPU64":%s:%s:%d\n", current->comm,
1389                imp->imp_obd->obd_uuid.uuid,
1390                req->rq_reqmsg->status, req->rq_xid,
1391                imp->imp_connection->c_peer.peer_ni->pni_name,
1392                ptlrpc_peernid2str(&imp->imp_connection->c_peer, str),
1393                req->rq_reqmsg->opc);
1394
1395         /* Mark phase here for a little debug help */
1396         req->rq_phase = RQ_PHASE_RPC;
1397
1398         spin_lock_irqsave(&imp->imp_lock, flags);
1399         req->rq_import_generation = imp->imp_generation;
1400 restart:
1401         if (ptlrpc_import_delay_req(imp, req, &rc)) {
1402                 list_del(&req->rq_list);
1403
1404                 list_add_tail(&req->rq_list, &imp->imp_delayed_list);
1405                 spin_unlock_irqrestore(&imp->imp_lock, flags);
1406
1407                 DEBUG_REQ(D_HA, req, "\"%s\" waiting for recovery: (%s != %s)",
1408                           current->comm,
1409                           ptlrpc_import_state_name(req->rq_send_state),
1410                           ptlrpc_import_state_name(imp->imp_state));
1411                 lwi = LWI_INTR(interrupted_request, req);
1412                 rc = l_wait_event(req->rq_reply_waitq,
1413                                   (req->rq_send_state == imp->imp_state ||
1414                                    req->rq_err),
1415                                   &lwi);
1416                 DEBUG_REQ(D_HA, req, "\"%s\" awake: (%s == %s or %d == 1)",
1417                           current->comm,
1418                           ptlrpc_import_state_name(imp->imp_state),
1419                           ptlrpc_import_state_name(req->rq_send_state),
1420                           req->rq_err);
1421
1422                 spin_lock_irqsave(&imp->imp_lock, flags);
1423                 list_del_init(&req->rq_list);
1424
1425                 if (req->rq_err) {
1426                         rc = -EIO;
1427                 }
1428                 else if (req->rq_intr) {
1429                         rc = -EINTR;
1430                 }
1431                 else if (req->rq_no_resend) {
1432                         spin_unlock_irqrestore(&imp->imp_lock, flags);
1433                         GOTO(out, rc = -ETIMEDOUT);
1434                 }
1435                 else {
1436                         GOTO(restart, rc);
1437                 }
1438         }
1439
1440         if (rc != 0) {
1441                 list_del_init(&req->rq_list);
1442                 spin_unlock_irqrestore(&imp->imp_lock, flags);
1443                 req->rq_status = rc; // XXX this ok?
1444                 GOTO(out, rc);
1445         }
1446
1447         if (req->rq_resend) {
1448                 if (!req->rq_ptlrpcs_restart)
1449                         lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
1450
1451                 if (req->rq_bulk != NULL)
1452                         ptlrpc_unregister_bulk (req);
1453
1454                 DEBUG_REQ(D_HA, req, "resending: ");
1455         }
1456
1457         /* XXX this is the same as ptlrpc_set_wait */
1458         LASSERT(list_empty(&req->rq_list));
1459         list_add_tail(&req->rq_list, &imp->imp_sending_list);
1460         spin_unlock_irqrestore(&imp->imp_lock, flags);
1461
1462         rc = ptl_send_rpc(req);
1463         if (rc) {
1464                 DEBUG_REQ(D_HA, req, "send failed (%d); recovering", rc);
1465                 timeout = 1;
1466         } else {
1467                 timeout = MAX(req->rq_timeout * HZ, 1);
1468                 DEBUG_REQ(D_NET, req, "-- sleeping for %d jiffies", timeout);
1469         }
1470 repeat:
1471         lwi = LWI_TIMEOUT_INTR(timeout, expired_request, interrupted_request,
1472                                req);
1473         rc = l_wait_event(req->rq_reply_waitq, ptlrpc_check_reply(req), &lwi);
1474         if (rc == -ETIMEDOUT && ptlrpc_check_and_wait_suspend(req))
1475                 goto repeat;
1476         DEBUG_REQ(D_NET, req, "-- done sleeping");
1477
1478         CDEBUG(D_RPCTRACE, "Completed RPC pname:cluuid:pid:xid:ni:nid:opc "
1479                "%s:%s:%d:"LPU64":%s:%s:%d\n", current->comm,
1480                imp->imp_obd->obd_uuid.uuid,
1481                req->rq_reqmsg->status, req->rq_xid,
1482                imp->imp_connection->c_peer.peer_ni->pni_name,
1483                ptlrpc_peernid2str(&imp->imp_connection->c_peer, str),
1484                req->rq_reqmsg->opc);
1485
1486         spin_lock_irqsave(&imp->imp_lock, flags);
1487         list_del_init(&req->rq_list);
1488         spin_unlock_irqrestore(&imp->imp_lock, flags);
1489
1490         /* If the reply was received normally, this just grabs the spinlock
1491          * (ensuring the reply callback has returned), sees that
1492          * req->rq_receiving_reply is clear and returns. */
1493         ptlrpc_unregister_reply (req);
1494
1495         if (req->rq_err)
1496                 GOTO(out, rc = -EIO);
1497
1498         /* Resend if we need to, unless we were interrupted. */
1499         if (req->rq_resend && !req->rq_intr) {
1500                 /* ...unless we were specifically told otherwise. */
1501                 if (req->rq_no_resend)
1502                         GOTO(out, rc = -ETIMEDOUT);
1503                 spin_lock_irqsave(&imp->imp_lock, flags);
1504                 goto restart;
1505         }
1506
1507         if (req->rq_intr) {
1508                 /* Should only be interrupted if we timed out. */
1509                 if (!req->rq_timedout)
1510                         DEBUG_REQ(D_ERROR, req,
1511                                   "rq_intr set but rq_timedout not");
1512                 GOTO(out, rc = -EINTR);
1513         }
1514
1515         if (req->rq_timedout) {                 /* non-recoverable timeout */
1516                 GOTO(out, rc = -ETIMEDOUT);
1517         }
1518
1519         if (!req->rq_replied) {
1520                 /* How can this be? -eeb */
1521                 DEBUG_REQ(D_ERROR, req, "!rq_replied: ");
1522                 LBUG();
1523                 GOTO(out, rc = req->rq_status);
1524         }
1525
1526         rc = after_reply (req);
1527         /* NB may return +ve success rc */
1528         if (req->rq_resend) {
1529                 spin_lock_irqsave(&imp->imp_lock, flags);
1530                 goto restart;
1531         }
1532
1533  out:
1534         if (req->rq_bulk != NULL) {
1535                 if (rc >= 0) {
1536                         /* success so far.  Note that anything going wrong
1537                          * with bulk now, is EXTREMELY strange, since the
1538                          * server must have believed that the bulk
1539                          * tranferred OK before she replied with success to
1540                          * me. */
1541                         lwi = LWI_TIMEOUT(timeout, NULL, NULL);
1542                         brc = l_wait_event(req->rq_reply_waitq,
1543                                            !ptlrpc_bulk_active(req->rq_bulk),
1544                                            &lwi);
1545                         LASSERT(brc == 0 || brc == -ETIMEDOUT);
1546                         if (brc != 0) {
1547                                 LASSERT(brc == -ETIMEDOUT);
1548                                 DEBUG_REQ(D_ERROR, req, "bulk timed out");
1549                                 rc = brc;
1550                         } else if (!req->rq_bulk->bd_success) {
1551                                 DEBUG_REQ(D_ERROR, req, "bulk transfer failed");
1552                                 rc = -EIO;
1553                         }
1554                 }
1555                 if (rc < 0)
1556                         ptlrpc_unregister_bulk (req);
1557         }
1558
1559         LASSERT(!req->rq_receiving_reply);
1560         req->rq_phase = RQ_PHASE_INTERPRET;
1561
1562         atomic_dec(&imp->imp_inflight);
1563         wake_up(&imp->imp_recovery_waitq);
1564         RETURN(rc);
1565 }
1566
1567 struct ptlrpc_replay_async_args {
1568         int praa_old_state;
1569         int praa_old_status;
1570 };
1571
1572 static int ptlrpc_replay_interpret(struct ptlrpc_request *req,
1573                                     void * data, int rc)
1574 {
1575         struct ptlrpc_replay_async_args *aa = data;
1576         struct obd_import *imp = req->rq_import;
1577         unsigned long flags;
1578
1579         atomic_dec(&imp->imp_replay_inflight);
1580
1581         if (!req->rq_replied) {
1582                 CERROR("request replay timed out, restarting recovery\n");
1583                 GOTO(out, rc = -ETIMEDOUT);
1584         }
1585
1586 #if SWAB_PARANOIA
1587         /* Clear reply swab mask; this is a new reply in sender's byte order */
1588         req->rq_rep_swab_mask = 0;
1589 #endif
1590         LASSERT (req->rq_nob_received <= req->rq_repbuf_len);
1591         rc = lustre_unpack_msg(req->rq_repmsg, req->rq_replen);
1592         if (rc) {
1593                 CERROR("unpack_rep failed: %d\n", rc);
1594                 GOTO(out, rc = -EPROTO);
1595         }
1596
1597         if (req->rq_repmsg->type == PTL_RPC_MSG_ERR &&
1598             req->rq_repmsg->status == -ENOTCONN)
1599                 GOTO(out, rc = req->rq_repmsg->status);
1600
1601         /* The transno had better not change over replay. */
1602         LASSERT(req->rq_reqmsg->transno == req->rq_repmsg->transno);
1603
1604         DEBUG_REQ(D_HA, req, "got rep");
1605
1606         /* let the callback do fixups, possibly including in the request */
1607         if (req->rq_replay_cb)
1608                 req->rq_replay_cb(req);
1609
1610         if (req->rq_replied && req->rq_repmsg->status != aa->praa_old_status) {
1611                 DEBUG_REQ(D_ERROR, req, "status %d, old was %d",
1612                           req->rq_repmsg->status, aa->praa_old_status);
1613         } else {
1614                 /* Put it back for re-replay. */
1615                 req->rq_repmsg->status = aa->praa_old_status;
1616         }
1617
1618         spin_lock_irqsave(&imp->imp_lock, flags);
1619         imp->imp_last_replay_transno = req->rq_transno;
1620         spin_unlock_irqrestore(&imp->imp_lock, flags);
1621
1622         /* continue with recovery */
1623         rc = ptlrpc_import_recovery_state_machine(imp);
1624  out:
1625         req->rq_send_state = aa->praa_old_state;
1626
1627         if (rc != 0)
1628                 /* this replay failed, so restart recovery */
1629                 ptlrpc_connect_import(imp, NULL);
1630
1631         RETURN(rc);
1632 }
1633
1634
1635 int ptlrpc_replay_req(struct ptlrpc_request *req)
1636 {
1637         struct ptlrpc_replay_async_args *aa;
1638         ENTRY;
1639
1640         LASSERT(req->rq_import->imp_state == LUSTRE_IMP_REPLAY);
1641
1642         /* Not handling automatic bulk replay yet (or ever?) */
1643         LASSERT(req->rq_bulk == NULL);
1644
1645         DEBUG_REQ(D_HA, req, "REPLAY");
1646
1647         LASSERT (sizeof (*aa) <= sizeof (req->rq_async_args));
1648         aa = (struct ptlrpc_replay_async_args *)&req->rq_async_args;
1649         memset(aa, 0, sizeof *aa);
1650
1651         /* Prepare request to be resent with ptlrpcd */
1652         aa->praa_old_state = req->rq_send_state;
1653         req->rq_send_state = LUSTRE_IMP_REPLAY;
1654         req->rq_phase = RQ_PHASE_NEW;
1655         aa->praa_old_status = req->rq_repmsg->status;
1656         req->rq_status = 0;
1657
1658         req->rq_interpret_reply = ptlrpc_replay_interpret;
1659         atomic_inc(&req->rq_import->imp_replay_inflight);
1660         ptlrpc_request_addref(req); /* ptlrpcd needs a ref */
1661
1662         ptlrpcd_add_req(req);
1663         RETURN(0);
1664 }
1665
1666 void ptlrpc_abort_inflight(struct obd_import *imp)
1667 {
1668         unsigned long flags;
1669         struct list_head *tmp, *n;
1670         ENTRY;
1671
1672         /* Make sure that no new requests get processed for this import.
1673          * ptlrpc_{queue,set}_wait must (and does) hold imp_lock while testing
1674          * this flag and then putting requests on sending_list or delayed_list.
1675          */
1676         spin_lock_irqsave(&imp->imp_lock, flags);
1677
1678         /* XXX locking?  Maybe we should remove each request with the list
1679          * locked?  Also, how do we know if the requests on the list are
1680          * being freed at this time?
1681          */
1682         list_for_each_safe(tmp, n, &imp->imp_sending_list) {
1683                 struct ptlrpc_request *req =
1684                         list_entry(tmp, struct ptlrpc_request, rq_list);
1685
1686                 DEBUG_REQ(D_HA, req, "inflight");
1687
1688                 spin_lock (&req->rq_lock);
1689                 if (req->rq_import_generation < imp->imp_generation) {
1690                         req->rq_err = 1;
1691                         ptlrpc_wake_client_req(req);
1692                 }
1693                 spin_unlock (&req->rq_lock);
1694         }
1695
1696         list_for_each_safe(tmp, n, &imp->imp_delayed_list) {
1697                 struct ptlrpc_request *req =
1698                         list_entry(tmp, struct ptlrpc_request, rq_list);
1699
1700                 DEBUG_REQ(D_HA, req, "aborting waiting req");
1701
1702                 spin_lock (&req->rq_lock);
1703                 if (req->rq_import_generation < imp->imp_generation) {
1704                         req->rq_err = 1;
1705                         ptlrpc_wake_client_req(req);
1706                 }
1707                 spin_unlock (&req->rq_lock);
1708         }
1709
1710         list_for_each_safe(tmp, n, &imp->imp_rawrpc_list) {
1711                 struct ptlrpc_request *req =
1712                         list_entry(tmp, struct ptlrpc_request, rq_list);
1713
1714                 DEBUG_REQ(D_HA, req, "aborting raw rpc");
1715
1716                 spin_lock(&req->rq_lock);
1717                 req->rq_err = 1;
1718                 ptlrpc_wake_client_req(req);
1719                 spin_unlock(&req->rq_lock);
1720         }
1721
1722         /* Last chance to free reqs left on the replay list, but we
1723          * will still leak reqs that haven't comitted.  */
1724         if (imp->imp_replayable)
1725                 ptlrpc_free_committed(imp);
1726
1727         spin_unlock_irqrestore(&imp->imp_lock, flags);
1728
1729         EXIT;
1730 }
1731
1732 static __u64 ptlrpc_last_xid = 0;
1733 static spinlock_t ptlrpc_last_xid_lock = SPIN_LOCK_UNLOCKED;
1734
1735 __u64 ptlrpc_next_xid(void)
1736 {
1737         __u64 tmp;
1738         spin_lock(&ptlrpc_last_xid_lock);
1739         tmp = ++ptlrpc_last_xid;
1740         spin_unlock(&ptlrpc_last_xid_lock);
1741         return tmp;
1742 }
1743
1744