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