lustre/target/tgt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2011, 2012, Intel Corporation.
  25  */
  26 /*
  27  * lustre/target/tgt_handler.c
  28  *
  29  * Lustre Unified Target request handler code
  30  *
  31  * Author: Brian Behlendorf <behlendorf1@llnl.gov>
  32  * Author: Mikhail Pershin <mike.pershin@intel.com>
  33  */
  34
  35 #define DEBUG_SUBSYSTEM S_CLASS
  36
  37 #include <obd.h>
  38 #include <obd_class.h>
  39 #include <obd_cksum.h>
  40 #include <md_object.h>
  41
  42 #include "tgt_internal.h"
  43
  44 char *tgt_name(struct lu_target *tgt)
  45 {
  46         LASSERT(tgt->lut_obd != NULL);
  47         return tgt->lut_obd->obd_name;
  48 }
  49 EXPORT_SYMBOL(tgt_name);
  50
  51 /*
  52  * Generic code handling requests that have struct mdt_body passed in:
  53  *
  54  *  - extract mdt_body from request and save it in @tsi, if present;
  55  *
  56  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  57  *  @tsi;
  58  *
  59  *  - if HABEO_CORPUS flag is set for this request type check whether object
  60  *  actually exists on storage (lu_object_exists()).
  61  *
  62  */
  63 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  64 {
  65         const struct mdt_body   *body;
  66         struct lu_object        *obj;
  67         struct req_capsule      *pill = tsi->tsi_pill;
  68         int                      rc;
  69
  70         ENTRY;
  71
  72         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  73         if (body == NULL)
  74                 RETURN(-EFAULT);
  75
  76         tsi->tsi_mdt_body = body;
  77
  78         if (!(body->valid & OBD_MD_FLID))
  79                 RETURN(0);
  80
  81         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  82          * in any case in pre-2.5 clients. Fix that here if needed */
  83         if (unlikely(fid_is_zero(&body->fid1)))
  84                 RETURN(0);
  85
  86         if (!fid_is_sane(&body->fid1)) {
  87                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  88                        PFID(&body->fid1));
  89                 RETURN(-EINVAL);
  90         }
  91
  92         obj = lu_object_find(tsi->tsi_env,
  93                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
  94                              &body->fid1, NULL);
  95         if (!IS_ERR(obj)) {
  96                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
  97                         lu_object_put(tsi->tsi_env, obj);
  98                         /* for capability renew ENOENT will be handled in
  99                          * mdt_renew_capa */
 100                         if (body->valid & OBD_MD_FLOSSCAPA)
 101                                 rc = 0;
 102                         else
 103                                 rc = -ENOENT;
 104                 } else {
 105                         tsi->tsi_corpus = obj;
 106                         rc = 0;
 107                 }
 108         } else {
 109                 rc = PTR_ERR(obj);
 110         }
 111
 112         tsi->tsi_fid = body->fid1;
 113
 114         RETURN(rc);
 115 }
 116
 117 /**
 118  * Validate oa from client.
 119  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 120  * req are valid.
 121  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 122  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 123  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 124  *       whether this is oi_fid or real ostid. So it will check
 125  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 126  *    c. Old FID-disable osc will send IDIF.
 127  *    d. new FID-enable osc/osp will send normal FID.
 128  *
 129  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 130  * be used for LAST_ID file, and only being accessed inside OST now.
 131  */
 132 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 133 {
 134         struct ost_id   *oi     = &oa->o_oi;
 135         obd_seq          seq    = ostid_seq(oi);
 136         obd_id           id     = ostid_id(oi);
 137         int              rc;
 138         ENTRY;
 139
 140         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
 141                      fid_seq_is_echo(seq))) {
 142                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 143                  * during create, and we will reset this to 1, since this
 144                  * oi_id is basically useless in the following create process,
 145                  * but oi_id == 0 will make it difficult to tell whether it is
 146                  * real FID or ost_id. */
 147                 oi->oi_fid.f_seq = FID_SEQ_ECHO;
 148                 oi->oi_fid.f_oid = id ?: 1;
 149                 oi->oi_fid.f_ver = 0;
 150         } else {
 151                 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
 152
 153                 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
 154                         GOTO(out, rc = -EPROTO);
 155
 156                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 157                  * all of the requests are required to setup FLGROUP */
 158                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 159                         ostid_set_seq_mdt0(oi);
 160                         oa->o_valid |= OBD_MD_FLGROUP;
 161                         seq = ostid_seq(oi);
 162                 }
 163
 164                 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
 165                                fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
 166                         GOTO(out, rc = -EPROTO);
 167
 168                 rc = ostid_to_fid(&tti->tti_fid1, oi,
 169                                   tsi->tsi_tgt->lut_lsd.lsd_osd_index);
 170                 if (unlikely(rc != 0))
 171                         GOTO(out, rc);
 172
 173                 oi->oi_fid = tti->tti_fid1;
 174         }
 175
 176         RETURN(0);
 177
 178 out:
 179         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 180                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
 181                seq, id, rc);
 182         return rc;
 183 }
 184 EXPORT_SYMBOL(tgt_validate_obdo);
 185
 186 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
 187 {
 188         unsigned                 max_brw;
 189         struct niobuf_remote    *rnb;
 190         struct obd_ioobj        *ioo;
 191         int                      obj_count;
 192
 193         ENTRY;
 194
 195         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
 196         if (ioo == NULL)
 197                 RETURN(-EPROTO);
 198
 199         rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
 200         if (rnb == NULL)
 201                 RETURN(-EPROTO);
 202
 203         max_brw = ioobj_max_brw_get(ioo);
 204         if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 205                 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 206                        ": rc = %d\n", tgt_name(tsi->tsi_tgt),
 207                        obd_export_nid2str(tsi->tsi_exp), max_brw,
 208                        POSTID(oi), -EPROTO);
 209                 RETURN(-EPROTO);
 210         }
 211         ioo->ioo_oid = *oi;
 212
 213         obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
 214                                         RCL_CLIENT) / sizeof(*ioo);
 215         if (obj_count == 0) {
 216                 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
 217                 RETURN(-EPROTO);
 218         } else if (obj_count > 1) {
 219                 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
 220                        obj_count);
 221                 RETURN(-EPROTO);
 222         }
 223
 224         if (ioo->ioo_bufcnt == 0) {
 225                 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
 226                 RETURN(-EPROTO);
 227         }
 228
 229         if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
 230                 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
 231                           "bulk has too many pages (%d)",
 232                           ioo->ioo_bufcnt);
 233                 RETURN(-EPROTO);
 234         }
 235
 236         RETURN(0);
 237 }
 238
 239 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 240 {
 241         struct ost_body         *body;
 242         struct req_capsule      *pill = tsi->tsi_pill;
 243         struct lustre_capa      *capa;
 244         int                      rc;
 245
 246         ENTRY;
 247
 248         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 249         if (body == NULL)
 250                 RETURN(-EFAULT);
 251
 252         rc = tgt_validate_obdo(tsi, &body->oa);
 253         if (rc)
 254                 RETURN(rc);
 255
 256         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 257                 capa = req_capsule_client_get(pill, &RMF_CAPA1);
 258                 if (capa == NULL) {
 259                         CERROR("%s: OSSCAPA flag is set without capability\n",
 260                                tgt_name(tsi->tsi_tgt));
 261                         RETURN(-EFAULT);
 262                 }
 263         }
 264
 265         tsi->tsi_ost_body = body;
 266         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 267
 268         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 269                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 270                 if (rc < 0)
 271                         RETURN(rc);
 272         }
 273
 274         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 275                 if (flags & HABEO_CORPUS) {
 276                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 277                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 278                                tgt_name(tsi->tsi_tgt));
 279                         RETURN(-EPROTO);
 280                 } else {
 281                         RETURN(0);
 282                 }
 283         }
 284
 285         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 286
 287         /*
 288          * OST doesn't get object in advance for further use to prevent
 289          * situations with nested object_find which is potential deadlock.
 290          */
 291         tsi->tsi_corpus = NULL;
 292         RETURN(rc);
 293 }
 294
 295 /*
 296  * Do necessary preprocessing according to handler ->th_flags.
 297  */
 298 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 299                                   struct tgt_handler *h,
 300                                   struct ptlrpc_request *req)
 301 {
 302         struct req_capsule      *pill = tsi->tsi_pill;
 303         __u32                    flags = h->th_flags;
 304         int                      rc = 0;
 305
 306         ENTRY;
 307
 308         if (tsi->tsi_preprocessed)
 309                 RETURN(0);
 310
 311         LASSERT(h->th_act != NULL);
 312         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 313         LASSERT(current->journal_info == NULL);
 314
 315         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 316                      h->th_fmt != NULL));
 317         if (h->th_fmt != NULL) {
 318                 req_capsule_set(pill, h->th_fmt);
 319                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 320                         rc = tgt_mdt_body_unpack(tsi, flags);
 321                         if (rc < 0)
 322                                 RETURN(rc);
 323                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 324                                                  RCL_CLIENT)) {
 325                         rc = tgt_ost_body_unpack(tsi, flags);
 326                         if (rc < 0)
 327                                 RETURN(rc);
 328                 }
 329         }
 330
 331         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 332                 RETURN(-EROFS);
 333
 334         if (flags & HABEO_CLAVIS) {
 335                 struct ldlm_request *dlm_req;
 336
 337                 LASSERT(h->th_fmt != NULL);
 338
 339                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 340                 if (dlm_req != NULL) {
 341                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 342                                      LDLM_IBITS &&
 343                                      dlm_req->lock_desc.l_policy_data.\
 344                                      l_inodebits.bits == 0)) {
 345                                 /*
 346                                  * Lock without inodebits makes no sense and
 347                                  * will oops later in ldlm. If client miss to
 348                                  * set such bits, do not trigger ASSERTION.
 349                                  *
 350                                  * For liblustre flock case, it maybe zero.
 351                                  */
 352                                 rc = -EPROTO;
 353                         } else {
 354                                 tsi->tsi_dlm_req = dlm_req;
 355                         }
 356                 } else {
 357                         rc = -EFAULT;
 358                 }
 359         }
 360         tsi->tsi_preprocessed = 1;
 361         RETURN(rc);
 362 }
 363
 364 /*
 365  * Invoke handler for this request opc. Also do necessary preprocessing
 366  * (according to handler ->th_flags), and post-processing (setting of
 367  * ->last_{xid,committed}).
 368  */
 369 static int tgt_handle_request0(struct tgt_session_info *tsi,
 370                                struct tgt_handler *h,
 371                                struct ptlrpc_request *req)
 372 {
 373         int      serious = 0;
 374         int      rc;
 375
 376         ENTRY;
 377
 378         /*
 379          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 380          * to put same checks into handlers like mdt_close(), mdt_reint(),
 381          * etc., without talking to mdt authors first. Checking same thing
 382          * there again is useless and returning 0 error without packing reply
 383          * is buggy! Handlers either pack reply or return error.
 384          *
 385          * We return 0 here and do not send any reply in order to emulate
 386          * network failure. Do not send any reply in case any of NET related
 387          * fail_id has occured.
 388          */
 389         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 390                 RETURN(0);
 391
 392         rc = tgt_request_preprocess(tsi, h, req);
 393         /* pack reply if reply format is fixed */
 394         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 395                 /* Pack reply */
 396                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 397                                           RCL_SERVER))
 398                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 399                                              RCL_SERVER,
 400                                              tsi->tsi_mdt_body->eadatasize);
 401                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 402                                           RCL_SERVER))
 403                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 404                                              RCL_SERVER, 0);
 405
 406                 rc = req_capsule_server_pack(tsi->tsi_pill);
 407         }
 408
 409         if (likely(rc == 0)) {
 410                 /*
 411                  * Process request, there can be two types of rc:
 412                  * 1) errors with msg unpack/pack, other failures outside the
 413                  * operation itself. This is counted as serious errors;
 414                  * 2) errors during fs operation, should be placed in rq_status
 415                  * only
 416                  */
 417                 rc = h->th_act(tsi);
 418                 if (!is_serious(rc) &&
 419                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 420                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 421                                   "pack reply and returned 0 error\n",
 422                                   tgt_name(tsi->tsi_tgt), h->th_name);
 423                         LBUG();
 424                 }
 425                 serious = is_serious(rc);
 426                 rc = clear_serious(rc);
 427         } else {
 428                 serious = 1;
 429         }
 430
 431         req->rq_status = rc;
 432
 433         /*
 434          * ELDLM_* codes which > 0 should be in rq_status only as well as
 435          * all non-serious errors.
 436          */
 437         if (rc > 0 || !serious)
 438                 rc = 0;
 439
 440         LASSERT(current->journal_info == NULL);
 441
 442         if (likely(rc == 0 && req->rq_export))
 443                 target_committed_to_req(req);
 444
 445         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 446         RETURN(0);
 447 }
 448
 449 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 450                                        struct obd_device *obd, int *process)
 451 {
 452         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 453         case MDS_DISCONNECT:
 454         case OST_DISCONNECT:
 455         case OBD_IDX_READ:
 456                 *process = 1;
 457                 RETURN(0);
 458         case MDS_CLOSE:
 459         case MDS_DONE_WRITING:
 460         case MDS_SYNC: /* used in unmounting */
 461         case OBD_PING:
 462         case MDS_REINT:
 463         case UPDATE_OBJ:
 464         case SEQ_QUERY:
 465         case FLD_QUERY:
 466         case LDLM_ENQUEUE:
 467         case OST_CREATE:
 468         case OST_DESTROY:
 469         case OST_PUNCH:
 470         case OST_SETATTR:
 471         case OST_SYNC:
 472         case OST_WRITE:
 473                 *process = target_queue_recovery_request(req, obd);
 474                 RETURN(0);
 475
 476         default:
 477                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 478                 *process = -EAGAIN;
 479                 RETURN(0);
 480         }
 481 }
 482
 483 /*
 484  * Handle recovery. Return:
 485  *        +1: continue request processing;
 486  *       -ve: abort immediately with the given error code;
 487  *         0: send reply with error code in req->rq_status;
 488  */
 489 int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 490 {
 491         ENTRY;
 492
 493         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 494         case MDS_CONNECT:
 495         case OST_CONNECT:
 496         case MGS_CONNECT:
 497         case SEC_CTX_INIT:
 498         case SEC_CTX_INIT_CONT:
 499         case SEC_CTX_FINI:
 500                 RETURN(+1);
 501         }
 502
 503         if (!req->rq_export->exp_obd->obd_replayable)
 504                 RETURN(+1);
 505
 506         /* sanity check: if the xid matches, the request must be marked as a
 507          * resent or replayed */
 508         if (req_xid_is_last(req)) {
 509                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 510                       (MSG_RESENT | MSG_REPLAY))) {
 511                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 512                                   "last_xid, expected REPLAY or RESENT flag "
 513                                   "(%x)", req->rq_xid,
 514                                   lustre_msg_get_flags(req->rq_reqmsg));
 515                         req->rq_status = -ENOTCONN;
 516                         RETURN(-ENOTCONN);
 517                 }
 518         }
 519         /* else: note the opposite is not always true; a RESENT req after a
 520          * failover will usually not match the last_xid, since it was likely
 521          * never committed. A REPLAYed request will almost never match the
 522          * last xid, however it could for a committed, but still retained,
 523          * open. */
 524
 525         /* Check for aborted recovery... */
 526         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 527                 int rc;
 528                 int should_process;
 529
 530                 DEBUG_REQ(D_INFO, req, "Got new replay");
 531                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 532                                                  &should_process);
 533                 if (rc != 0 || !should_process)
 534                         RETURN(rc);
 535                 else if (should_process < 0) {
 536                         req->rq_status = should_process;
 537                         rc = ptlrpc_error(req);
 538                         RETURN(rc);
 539                 }
 540         }
 541         RETURN(+1);
 542 }
 543
 544 /* Initial check for request, it is validation mostly */
 545 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 546 {
 547         struct tgt_handler      *h;
 548         struct tgt_opc_slice    *s;
 549         struct lu_target        *tgt;
 550         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 551
 552         ENTRY;
 553
 554         tgt = class_exp2tgt(req->rq_export);
 555
 556         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 557                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 558                         break;
 559
 560         /* opcode was not found in slice */
 561         if (unlikely(s->tos_hs == NULL)) {
 562                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 563                        opc);
 564                 RETURN(ERR_PTR(-ENOTSUPP));
 565         }
 566
 567         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 568         h = s->tos_hs + (opc - s->tos_opc_start);
 569         if (unlikely(h->th_opc == 0)) {
 570                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 571                 RETURN(ERR_PTR(-ENOTSUPP));
 572         }
 573
 574         RETURN(h);
 575 }
 576
 577 int tgt_request_handle(struct ptlrpc_request *req)
 578 {
 579         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 580
 581         struct lustre_msg       *msg = req->rq_reqmsg;
 582         struct tgt_handler      *h;
 583         struct lu_target        *tgt;
 584         int                      request_fail_id = 0;
 585         __u32                    opc = lustre_msg_get_opc(msg);
 586         int                      rc;
 587
 588         ENTRY;
 589
 590         /* Refill the context, to make sure all thread keys are allocated */
 591         lu_env_refill(req->rq_svc_thread->t_env);
 592
 593         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 594         tsi->tsi_pill = &req->rq_pill;
 595         tsi->tsi_env = req->rq_svc_thread->t_env;
 596
 597         /* if request has export then get handlers slice from corresponding
 598          * target, otherwise that should be connect operation */
 599         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 600             opc == MGS_CONNECT) {
 601                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 602                 rc = target_handle_connect(req);
 603                 if (rc != 0) {
 604                         rc = ptlrpc_error(req);
 605                         GOTO(out, rc);
 606                 }
 607                 /* recovery-small test 18c asks to drop connect reply */
 608                 if (unlikely(opc == OST_CONNECT &&
 609                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 610                         GOTO(out, rc = 0);
 611         }
 612
 613         if (unlikely(!class_connected_export(req->rq_export))) {
 614                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 615                        opc, libcfs_id2str(req->rq_peer));
 616                 req->rq_status = -ENOTCONN;
 617                 rc = ptlrpc_error(req);
 618                 GOTO(out, rc);
 619         }
 620
 621         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 622         tsi->tsi_exp = req->rq_export;
 623         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 624                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 625         else
 626                 tsi->tsi_jobid = NULL;
 627
 628         request_fail_id = tgt->lut_request_fail_id;
 629         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 630
 631         h = tgt_handler_find_check(req);
 632         if (IS_ERR(h)) {
 633                 req->rq_status = PTR_ERR(h);
 634                 rc = ptlrpc_error(req);
 635                 GOTO(out, rc);
 636         }
 637
 638         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 639                 GOTO(out, rc = 0);
 640
 641         rc = lustre_msg_check_version(msg, h->th_version);
 642         if (unlikely(rc)) {
 643                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 644                           " %08x, expecting %08x\n", tgt_name(tgt),
 645                           lustre_msg_get_version(msg), h->th_version);
 646                 req->rq_status = -EINVAL;
 647                 rc = ptlrpc_error(req);
 648                 GOTO(out, rc);
 649         }
 650
 651         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 652         if (likely(rc == 1)) {
 653                 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 654                          h->th_opc, opc);
 655                 rc = tgt_handle_request0(tsi, h, req);
 656                 if (rc)
 657                         GOTO(out, rc);
 658         }
 659         EXIT;
 660 out:
 661         req_capsule_fini(tsi->tsi_pill);
 662         if (tsi->tsi_corpus != NULL) {
 663                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 664                 tsi->tsi_corpus = NULL;
 665         }
 666         return rc;
 667 }
 668 EXPORT_SYMBOL(tgt_request_handle);
 669
 670 /** Assign high priority operations to the request if needed. */
 671 int tgt_hpreq_handler(struct ptlrpc_request *req)
 672 {
 673         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 674         struct tgt_handler      *h;
 675         int                      rc;
 676
 677         ENTRY;
 678
 679         if (req->rq_export == NULL)
 680                 RETURN(0);
 681
 682         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 683         tsi->tsi_pill = &req->rq_pill;
 684         tsi->tsi_env = req->rq_svc_thread->t_env;
 685         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 686         tsi->tsi_exp = req->rq_export;
 687
 688         h = tgt_handler_find_check(req);
 689         if (IS_ERR(h)) {
 690                 rc = PTR_ERR(h);
 691                 RETURN(rc);
 692         }
 693
 694         rc = tgt_request_preprocess(tsi, h, req);
 695         if (unlikely(rc != 0))
 696                 RETURN(rc);
 697
 698         if (h->th_hp != NULL)
 699                 h->th_hp(tsi);
 700         RETURN(0);
 701 }
 702 EXPORT_SYMBOL(tgt_hpreq_handler);
 703
 704 void tgt_counter_incr(struct obd_export *exp, int opcode)
 705 {
 706         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 707         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 708                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 709 }
 710 EXPORT_SYMBOL(tgt_counter_incr);
 711
 712 /*
 713  * Unified target generic handlers.
 714  */
 715
 716 /*
 717  * Security functions
 718  */
 719 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 720 {
 721         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 722                                       OBD_CONNECT_RMT_CLIENT_FORCE |
 723                                       OBD_CONNECT_MDS_CAPA |
 724                                       OBD_CONNECT_OSS_CAPA);
 725 }
 726
 727 static int tgt_init_sec_level(struct ptlrpc_request *req)
 728 {
 729         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 730         char                    *client = libcfs_nid2str(req->rq_peer.nid);
 731         struct obd_connect_data *data, *reply;
 732         int                      rc = 0;
 733         bool                     remote;
 734
 735         ENTRY;
 736
 737         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 738         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 739         if (data == NULL || reply == NULL)
 740                 RETURN(-EFAULT);
 741
 742         /* connection from MDT is always trusted */
 743         if (req->rq_auth_usr_mdt) {
 744                 tgt_init_sec_none(reply);
 745                 RETURN(0);
 746         }
 747
 748         /* no GSS support case */
 749         if (!req->rq_auth_gss) {
 750                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 751                         CWARN("client %s -> target %s does not use GSS, "
 752                               "can not run under security level %d.\n",
 753                               client, tgt_name(tgt), tgt->lut_sec_level);
 754                         RETURN(-EACCES);
 755                 } else {
 756                         tgt_init_sec_none(reply);
 757                         RETURN(0);
 758                 }
 759         }
 760
 761         /* old version case */
 762         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 763                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 764                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 765                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 766                         CWARN("client %s -> target %s uses old version, "
 767                               "can not run under security level %d.\n",
 768                               client, tgt_name(tgt), tgt->lut_sec_level);
 769                         RETURN(-EACCES);
 770                 } else {
 771                         CWARN("client %s -> target %s uses old version, "
 772                               "run under security level %d.\n",
 773                               client, tgt_name(tgt), tgt->lut_sec_level);
 774                         tgt_init_sec_none(reply);
 775                         RETURN(0);
 776                 }
 777         }
 778
 779         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 780         if (remote) {
 781                 if (!req->rq_auth_remote)
 782                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 783                                "asked to be remote.\n", client, tgt_name(tgt));
 784         } else if (req->rq_auth_remote) {
 785                 remote = true;
 786                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 787                        "as remote by default.\n", client, tgt_name(tgt));
 788         }
 789
 790         if (remote) {
 791                 if (!tgt->lut_oss_capa) {
 792                         CDEBUG(D_SEC,
 793                                "client %s -> target %s is set as remote,"
 794                                " but OSS capabilities are not enabled: %d.\n",
 795                                client, tgt_name(tgt), tgt->lut_oss_capa);
 796                         RETURN(-EACCES);
 797                 }
 798         } else {
 799                 if (req->rq_auth_uid == INVALID_UID) {
 800                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 801                                "authenticated!\n", client, tgt_name(tgt));
 802                         RETURN(-EACCES);
 803                 }
 804         }
 805
 806
 807         switch (tgt->lut_sec_level) {
 808         case LUSTRE_SEC_NONE:
 809                 if (remote) {
 810                         CDEBUG(D_SEC,
 811                                "client %s -> target %s is set as remote, "
 812                                "can not run under security level %d.\n",
 813                                client, tgt_name(tgt), tgt->lut_sec_level);
 814                         RETURN(-EACCES);
 815                 }
 816                 tgt_init_sec_none(reply);
 817                 break;
 818         case LUSTRE_SEC_REMOTE:
 819                 if (!remote)
 820                         tgt_init_sec_none(reply);
 821                 break;
 822         case LUSTRE_SEC_ALL:
 823                 if (remote)
 824                         break;
 825                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 826                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 827                 if (!tgt->lut_oss_capa)
 828                         reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 829                 if (!tgt->lut_mds_capa)
 830                         reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 831                 break;
 832         default:
 833                 RETURN(-EINVAL);
 834         }
 835
 836         RETURN(rc);
 837 }
 838
 839 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 840 {
 841         struct lu_target        *tgt = class_exp2tgt(exp);
 842         struct sptlrpc_flavor    flvr;
 843         int                      rc = 0;
 844
 845         LASSERT(tgt);
 846         LASSERT(tgt->lut_obd);
 847         LASSERT(tgt->lut_slice);
 848
 849         /* always allow ECHO client */
 850         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 851                             LUSTRE_ECHO_NAME) == 0)) {
 852                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 853                 return 0;
 854         }
 855
 856         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 857                 read_lock(&tgt->lut_sptlrpc_lock);
 858                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 859                                              req->rq_sp_from,
 860                                              req->rq_peer.nid,
 861                                              &flvr);
 862                 read_unlock(&tgt->lut_sptlrpc_lock);
 863
 864                 spin_lock(&exp->exp_lock);
 865                 exp->exp_sp_peer = req->rq_sp_from;
 866                 exp->exp_flvr = flvr;
 867                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 868                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 869                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 870                                "expect %x\n", tgt_name(tgt),
 871                                req->rq_flvr.sf_rpc,
 872                                libcfs_nid2str(req->rq_peer.nid),
 873                                exp->exp_flvr.sf_rpc);
 874                         rc = -EACCES;
 875                 }
 876                 spin_unlock(&exp->exp_lock);
 877         } else {
 878                 if (exp->exp_sp_peer != req->rq_sp_from) {
 879                         CERROR("%s: RPC source %s doesn't match %s\n",
 880                                tgt_name(tgt),
 881                                sptlrpc_part2name(req->rq_sp_from),
 882                                sptlrpc_part2name(exp->exp_sp_peer));
 883                         rc = -EACCES;
 884                 } else {
 885                         rc = sptlrpc_target_export_check(exp, req);
 886                 }
 887         }
 888
 889         return rc;
 890 }
 891
 892 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 893 {
 894         struct sptlrpc_rule_set  tmp_rset;
 895         int                      rc;
 896
 897         sptlrpc_rule_set_init(&tmp_rset);
 898         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 899         if (rc) {
 900                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 901                        tgt_name(tgt), rc);
 902                 return rc;
 903         }
 904
 905         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 906
 907         write_lock(&tgt->lut_sptlrpc_lock);
 908         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 909         tgt->lut_sptlrpc_rset = tmp_rset;
 910         write_unlock(&tgt->lut_sptlrpc_lock);
 911
 912         return 0;
 913 }
 914 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 915
 916 int tgt_connect(struct tgt_session_info *tsi)
 917 {
 918         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 919         struct obd_connect_data *reply;
 920         int                      rc;
 921
 922         ENTRY;
 923
 924         rc = tgt_init_sec_level(req);
 925         if (rc != 0)
 926                 GOTO(out, rc);
 927
 928         /* XXX: better to call this check right after getting new export but
 929          * before last_rcvd slot allocation to avoid server load upon insecure
 930          * connects. This is to be fixed after unifiyng all targets.
 931          */
 932         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
 933         if (rc)
 934                 GOTO(out, rc);
 935
 936         /* To avoid exposing partially initialized connection flags, changes up
 937          * to this point have been staged in reply->ocd_connect_flags. Now that
 938          * connection handling has completed successfully, atomically update
 939          * the connect flags in the shared export data structure. LU-1623 */
 940         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
 941         spin_lock(&tsi->tsi_exp->exp_lock);
 942         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
 943         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
 944         spin_unlock(&tsi->tsi_exp->exp_lock);
 945
 946         RETURN(0);
 947 out:
 948         obd_disconnect(class_export_get(tsi->tsi_exp));
 949         return rc;
 950 }
 951 EXPORT_SYMBOL(tgt_connect);
 952
 953 int tgt_disconnect(struct tgt_session_info *tsi)
 954 {
 955         int rc;
 956
 957         ENTRY;
 958
 959         rc = target_handle_disconnect(tgt_ses_req(tsi));
 960         if (rc)
 961                 RETURN(err_serious(rc));
 962
 963         RETURN(rc);
 964 }
 965 EXPORT_SYMBOL(tgt_disconnect);
 966
 967 /*
 968  * Unified target OBD handlers
 969  */
 970 int tgt_obd_ping(struct tgt_session_info *tsi)
 971 {
 972         int rc;
 973
 974         ENTRY;
 975
 976         rc = target_handle_ping(tgt_ses_req(tsi));
 977         if (rc)
 978                 RETURN(err_serious(rc));
 979
 980         RETURN(rc);
 981 }
 982 EXPORT_SYMBOL(tgt_obd_ping);
 983
 984 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
 985 {
 986         return err_serious(-EOPNOTSUPP);
 987 }
 988 EXPORT_SYMBOL(tgt_obd_log_cancel);
 989
 990 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
 991 {
 992         return err_serious(-EOPNOTSUPP);
 993 }
 994 EXPORT_SYMBOL(tgt_obd_qc_callback);
 995
 996 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
 997 {
 998         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
 999         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1000         struct obd_export       *exp = req->rq_export;
1001         struct ptlrpc_bulk_desc *desc;
1002         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1003         int                      tmpcount;
1004         int                      tmpsize;
1005         int                      i;
1006         int                      rc;
1007
1008         ENTRY;
1009
1010         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1011                                     MDS_BULK_PORTAL);
1012         if (desc == NULL)
1013                 RETURN(-ENOMEM);
1014
1015         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1016                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1017                  * which is rdpg->rp_count */
1018                 nob = rdpg->rp_count;
1019
1020         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1021              i++, tmpcount -= tmpsize) {
1022                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1023                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1024         }
1025
1026         LASSERT(desc->bd_nob == nob);
1027         rc = target_bulk_io(exp, desc, lwi);
1028         ptlrpc_free_bulk_pin(desc);
1029         RETURN(rc);
1030 }
1031 EXPORT_SYMBOL(tgt_sendpage);
1032
1033 /*
1034  * OBD_IDX_READ handler
1035  */
1036 int tgt_obd_idx_read(struct tgt_session_info *tsi)
1037 {
1038         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1039         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1040         struct idx_info         *req_ii, *rep_ii;
1041         int                      rc, i;
1042
1043         ENTRY;
1044
1045         memset(rdpg, 0, sizeof(*rdpg));
1046         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1047
1048         /* extract idx_info buffer from request & reply */
1049         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1050         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1051                 RETURN(err_serious(-EPROTO));
1052
1053         rc = req_capsule_server_pack(tsi->tsi_pill);
1054         if (rc)
1055                 RETURN(err_serious(rc));
1056
1057         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1058         if (rep_ii == NULL)
1059                 RETURN(err_serious(-EFAULT));
1060         rep_ii->ii_magic = IDX_INFO_MAGIC;
1061
1062         /* extract hash to start with */
1063         rdpg->rp_hash = req_ii->ii_hash_start;
1064
1065         /* extract requested attributes */
1066         rdpg->rp_attrs = req_ii->ii_attrs;
1067
1068         /* check that fid packed in request is valid and supported */
1069         if (!fid_is_sane(&req_ii->ii_fid))
1070                 RETURN(-EINVAL);
1071         rep_ii->ii_fid = req_ii->ii_fid;
1072
1073         /* copy flags */
1074         rep_ii->ii_flags = req_ii->ii_flags;
1075
1076         /* compute number of pages to allocate, ii_count is the number of 4KB
1077          * containers */
1078         if (req_ii->ii_count <= 0)
1079                 GOTO(out, rc = -EFAULT);
1080         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1081                                exp_max_brw_size(tsi->tsi_exp));
1082         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1083
1084         /* allocate pages to store the containers */
1085         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1086         if (rdpg->rp_pages == NULL)
1087                 GOTO(out, rc = -ENOMEM);
1088         for (i = 0; i < rdpg->rp_npages; i++) {
1089                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1090                 if (rdpg->rp_pages[i] == NULL)
1091                         GOTO(out, rc = -ENOMEM);
1092         }
1093
1094         /* populate pages with key/record pairs */
1095         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1096         if (rc < 0)
1097                 GOTO(out, rc);
1098
1099         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1100                  "asked %d > %d\n", rc, rdpg->rp_count);
1101
1102         /* send pages to client */
1103         rc = tgt_sendpage(tsi, rdpg, rc);
1104         if (rc)
1105                 GOTO(out, rc);
1106         EXIT;
1107 out:
1108         if (rdpg->rp_pages) {
1109                 for (i = 0; i < rdpg->rp_npages; i++)
1110                         if (rdpg->rp_pages[i])
1111                                 __free_page(rdpg->rp_pages[i]);
1112                 OBD_FREE(rdpg->rp_pages,
1113                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1114         }
1115         return rc;
1116 }
1117 EXPORT_SYMBOL(tgt_obd_idx_read);
1118
1119 struct tgt_handler tgt_obd_handlers[] = {
1120 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1121 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1122 TGT_OBD_HDL_VAR(0,      OBD_QC_CALLBACK,        tgt_obd_qc_callback),
1123 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1124 };
1125 EXPORT_SYMBOL(tgt_obd_handlers);
1126
1127 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1128              struct dt_object *obj)
1129 {
1130         int rc = 0;
1131
1132         ENTRY;
1133
1134         /* if no objid is specified, it means "sync whole filesystem" */
1135         if (obj == NULL) {
1136                 rc = dt_sync(env, tgt->lut_bottom);
1137         } else if (dt_version_get(env, obj) >
1138                    tgt->lut_obd->obd_last_committed) {
1139                 rc = dt_object_sync(env, obj);
1140         }
1141
1142         RETURN(rc);
1143 }
1144 EXPORT_SYMBOL(tgt_sync);
1145 /*
1146  * Unified target DLM handlers.
1147  */
1148
1149 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1150  * (if requested) */
1151 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1152                      void *data, int flag)
1153 {
1154         struct lu_env            env;
1155         struct lu_target        *tgt;
1156         struct dt_object        *obj;
1157         struct lu_fid            fid;
1158         int                      rc = 0;
1159
1160         ENTRY;
1161
1162         tgt = class_exp2tgt(lock->l_export);
1163
1164         if (flag == LDLM_CB_CANCELING &&
1165             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1166             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1167              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1168               lock->l_flags & LDLM_FL_CBPENDING))) {
1169                 rc = lu_env_init(&env, LCT_DT_THREAD);
1170                 if (unlikely(rc != 0))
1171                         RETURN(rc);
1172
1173                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1174                                    tgt->lut_lsd.lsd_osd_index);
1175                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1176                 if (IS_ERR(obj))
1177                         GOTO(err_env, rc = PTR_ERR(obj));
1178
1179                 if (!dt_object_exists(obj))
1180                         GOTO(err_put, rc = -ENOENT);
1181
1182                 rc = tgt_sync(&env, tgt, obj);
1183                 if (rc < 0) {
1184                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1185                                "cancel: rc = %d\n",
1186                                tgt_name(tgt), PFID(&fid),
1187                                lock->l_policy_data.l_extent.start,
1188                                lock->l_policy_data.l_extent.end, rc);
1189                 }
1190 err_put:
1191                 lu_object_put(&env, &obj->do_lu);
1192 err_env:
1193                 lu_env_fini(&env);
1194         }
1195
1196         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1197         RETURN(rc);
1198 }
1199
1200 struct ldlm_callback_suite tgt_dlm_cbs = {
1201         .lcs_completion = ldlm_server_completion_ast,
1202         .lcs_blocking   = tgt_blocking_ast,
1203         .lcs_glimpse    = ldlm_server_glimpse_ast
1204 };
1205
1206 int tgt_enqueue(struct tgt_session_info *tsi)
1207 {
1208         struct ptlrpc_request *req = tgt_ses_req(tsi);
1209         int rc;
1210
1211         ENTRY;
1212         /*
1213          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1214          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1215          */
1216         LASSERT(tsi->tsi_dlm_req != NULL);
1217         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1218                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1219         if (rc)
1220                 RETURN(err_serious(rc));
1221
1222         RETURN(req->rq_status);
1223 }
1224 EXPORT_SYMBOL(tgt_enqueue);
1225
1226 int tgt_convert(struct tgt_session_info *tsi)
1227 {
1228         struct ptlrpc_request *req = tgt_ses_req(tsi);
1229         int rc;
1230
1231         ENTRY;
1232         LASSERT(tsi->tsi_dlm_req);
1233         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1234         if (rc)
1235                 RETURN(err_serious(rc));
1236
1237         RETURN(req->rq_status);
1238 }
1239 EXPORT_SYMBOL(tgt_convert);
1240
1241 int tgt_bl_callback(struct tgt_session_info *tsi)
1242 {
1243         return err_serious(-EOPNOTSUPP);
1244 }
1245 EXPORT_SYMBOL(tgt_bl_callback);
1246
1247 int tgt_cp_callback(struct tgt_session_info *tsi)
1248 {
1249         return err_serious(-EOPNOTSUPP);
1250 }
1251 EXPORT_SYMBOL(tgt_cp_callback);
1252
1253 /* generic LDLM target handler */
1254 struct tgt_handler tgt_dlm_handlers[] = {
1255 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1256 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1257 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1258 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1259 };
1260 EXPORT_SYMBOL(tgt_dlm_handlers);
1261
1262 /*
1263  * Unified target LLOG handlers.
1264  */
1265 int tgt_llog_open(struct tgt_session_info *tsi)
1266 {
1267         int rc;
1268
1269         ENTRY;
1270
1271         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1272
1273         RETURN(rc);
1274 }
1275 EXPORT_SYMBOL(tgt_llog_open);
1276
1277 int tgt_llog_close(struct tgt_session_info *tsi)
1278 {
1279         int rc;
1280
1281         ENTRY;
1282
1283         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1284
1285         RETURN(rc);
1286 }
1287 EXPORT_SYMBOL(tgt_llog_close);
1288
1289
1290 int tgt_llog_destroy(struct tgt_session_info *tsi)
1291 {
1292         int rc;
1293
1294         ENTRY;
1295
1296         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1297
1298         RETURN(rc);
1299 }
1300 EXPORT_SYMBOL(tgt_llog_destroy);
1301
1302 int tgt_llog_read_header(struct tgt_session_info *tsi)
1303 {
1304         int rc;
1305
1306         ENTRY;
1307
1308         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1309
1310         RETURN(rc);
1311 }
1312 EXPORT_SYMBOL(tgt_llog_read_header);
1313
1314 int tgt_llog_next_block(struct tgt_session_info *tsi)
1315 {
1316         int rc;
1317
1318         ENTRY;
1319
1320         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1321
1322         RETURN(rc);
1323 }
1324 EXPORT_SYMBOL(tgt_llog_next_block);
1325
1326 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1327 {
1328         int rc;
1329
1330         ENTRY;
1331
1332         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1333
1334         RETURN(rc);
1335 }
1336 EXPORT_SYMBOL(tgt_llog_prev_block);
1337
1338 /* generic llog target handler */
1339 struct tgt_handler tgt_llog_handlers[] = {
1340 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1341 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1342 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1343 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1344 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1345 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1346 };
1347 EXPORT_SYMBOL(tgt_llog_handlers);
1348
1349 /*
1350  * sec context handlers
1351  */
1352 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1353 int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1354 {
1355         return 0;
1356 }
1357
1358 struct tgt_handler tgt_sec_ctx_handlers[] = {
1359 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1360 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1361 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1362 };
1363 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1364
1365 /*
1366  * initialize per-thread page pool (bug 5137).
1367  */
1368 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1369 {
1370         struct tgt_thread_big_cache *tbc;
1371
1372         ENTRY;
1373
1374         LASSERT(thread != NULL);
1375         LASSERT(thread->t_data == NULL);
1376
1377         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1378         if (tbc == NULL)
1379                 RETURN(-ENOMEM);
1380         thread->t_data = tbc;
1381         RETURN(0);
1382 }
1383 EXPORT_SYMBOL(tgt_io_thread_init);
1384
1385 /*
1386  * free per-thread pool created by tgt_thread_init().
1387  */
1388 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1389 {
1390         struct tgt_thread_big_cache *tbc;
1391
1392         ENTRY;
1393
1394         LASSERT(thread != NULL);
1395
1396         /*
1397          * be prepared to handle partially-initialized pools (because this is
1398          * called from ost_io_thread_init() for cleanup.
1399          */
1400         tbc = thread->t_data;
1401         if (tbc != NULL) {
1402                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1403                 thread->t_data = NULL;
1404         }
1405         EXIT;
1406 }
1407 EXPORT_SYMBOL(tgt_io_thread_done);
1408 /**
1409  * Helper function for getting server side [start, start+count] DLM lock
1410  * if asked by client.
1411  */
1412 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1413                     __u64 start, __u64 end, struct lustre_handle *lh,
1414                     int mode, __u64 *flags)
1415 {
1416         ldlm_policy_data_t       policy;
1417         int                      rc;
1418
1419         ENTRY;
1420
1421         LASSERT(lh != NULL);
1422         LASSERT(ns != NULL);
1423         LASSERT(!lustre_handle_is_used(lh));
1424
1425         policy.l_extent.gid = 0;
1426         policy.l_extent.start = start & CFS_PAGE_MASK;
1427
1428         /*
1429          * If ->o_blocks is EOF it means "lock till the end of the file".
1430          * Otherwise, it's size of an extent or hole being punched (in bytes).
1431          */
1432         if (end == OBD_OBJECT_EOF || end < start)
1433                 policy.l_extent.end = OBD_OBJECT_EOF;
1434         else
1435                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1436
1437         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1438                                     flags, ldlm_blocking_ast,
1439                                     ldlm_completion_ast, ldlm_glimpse_ast,
1440                                     NULL, 0, LVB_T_NONE, NULL, lh);
1441         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1442 }
1443 EXPORT_SYMBOL(tgt_extent_lock);
1444
1445 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1446 {
1447         LASSERT(lustre_handle_is_used(lh));
1448         ldlm_lock_decref(lh, mode);
1449 }
1450 EXPORT_SYMBOL(tgt_extent_unlock);
1451
1452 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1453                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1454                  struct lustre_handle *lh, int mode)
1455 {
1456         __u64                    flags = 0;
1457         int                      nrbufs = obj->ioo_bufcnt;
1458         int                      i;
1459
1460         ENTRY;
1461
1462         LASSERT(mode == LCK_PR || mode == LCK_PW);
1463         LASSERT(!lustre_handle_is_used(lh));
1464
1465         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1466                 RETURN(0);
1467
1468         for (i = 1; i < nrbufs; i++)
1469                 if (!(nb[i].flags & OBD_BRW_SRVLOCK))
1470                         RETURN(-EFAULT);
1471
1472         RETURN(tgt_extent_lock(ns, res_id, nb[0].offset,
1473                                nb[nrbufs - 1].offset + nb[nrbufs - 1].len - 1,
1474                                lh, mode, &flags));
1475 }
1476 EXPORT_SYMBOL(tgt_brw_lock);
1477
1478 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1479                     struct lustre_handle *lh, int mode)
1480 {
1481         ENTRY;
1482
1483         LASSERT(mode == LCK_PR || mode == LCK_PW);
1484         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
1485                 lustre_handle_is_used(lh));
1486         if (lustre_handle_is_used(lh))
1487                 tgt_extent_unlock(lh, mode);
1488         EXIT;
1489 }
1490 EXPORT_SYMBOL(tgt_brw_unlock);
1491
1492 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1493                                struct ptlrpc_bulk_desc *desc, int opc,
1494                                cksum_type_t cksum_type)
1495 {
1496         struct cfs_crypto_hash_desc     *hdesc;
1497         unsigned int                    bufsize;
1498         int                             i, err;
1499         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1500         __u32                           cksum;
1501
1502         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1503         if (IS_ERR(hdesc)) {
1504                 CERROR("%s: unable to initialize checksum hash %s\n",
1505                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1506                 return PTR_ERR(hdesc);
1507         }
1508
1509         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1510         for (i = 0; i < desc->bd_iov_count; i++) {
1511                 /* corrupt the data before we compute the checksum, to
1512                  * simulate a client->OST data error */
1513                 if (i == 0 && opc == OST_WRITE &&
1514                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1515                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1516                         int len = desc->bd_iov[i].kiov_len;
1517                         struct page *np = tgt_page_to_corrupt;
1518                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1519
1520                         if (np) {
1521                                 char *ptr2 = kmap(np) + off;
1522
1523                                 memcpy(ptr2, ptr, len);
1524                                 memcpy(ptr2, "bad3", min(4, len));
1525                                 kunmap(np);
1526                                 desc->bd_iov[i].kiov_page = np;
1527                         } else {
1528                                 CERROR("%s: can't alloc page for corruption\n",
1529                                        tgt_name(tgt));
1530                         }
1531                 }
1532                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1533                                   desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1534                                   desc->bd_iov[i].kiov_len);
1535
1536                  /* corrupt the data after we compute the checksum, to
1537                  * simulate an OST->client data error */
1538                 if (i == 0 && opc == OST_READ &&
1539                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1540                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1541                         int len = desc->bd_iov[i].kiov_len;
1542                         struct page *np = tgt_page_to_corrupt;
1543                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1544
1545                         if (np) {
1546                                 char *ptr2 = kmap(np) + off;
1547
1548                                 memcpy(ptr2, ptr, len);
1549                                 memcpy(ptr2, "bad4", min(4, len));
1550                                 kunmap(np);
1551                                 desc->bd_iov[i].kiov_page = np;
1552                         } else {
1553                                 CERROR("%s: can't alloc page for corruption\n",
1554                                        tgt_name(tgt));
1555                         }
1556                 }
1557         }
1558
1559         bufsize = 4;
1560         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1561         if (err)
1562                 cfs_crypto_hash_final(hdesc, NULL, NULL);
1563
1564         return cksum;
1565 }
1566
1567 int tgt_brw_read(struct tgt_session_info *tsi)
1568 {
1569         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1570         struct ptlrpc_bulk_desc *desc = NULL;
1571         struct obd_export       *exp = tsi->tsi_exp;
1572         struct niobuf_remote    *remote_nb;
1573         struct niobuf_local     *local_nb;
1574         struct obd_ioobj        *ioo;
1575         struct ost_body         *body, *repbody;
1576         struct l_wait_info       lwi;
1577         struct lustre_handle     lockh = { 0 };
1578         int                      niocount, npages, nob = 0, rc, i;
1579         int                      no_reply = 0;
1580         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1581
1582         ENTRY;
1583
1584         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1585                 CERROR("%s: deny read request from %s to portal %u\n",
1586                        tgt_name(tsi->tsi_tgt),
1587                        obd_export_nid2str(req->rq_export),
1588                        ptlrpc_req2svc(req)->srv_req_portal);
1589                 RETURN(-EPROTO);
1590         }
1591
1592         req->rq_bulk_read = 1;
1593
1594         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1595                 RETURN(-EIO);
1596
1597         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1598                          cfs_fail_val : (obd_timeout + 1) / 4);
1599
1600         /* Check if there is eviction in progress, and if so, wait for it to
1601          * finish */
1602         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1603                 /* We do not care how long it takes */
1604                 lwi = LWI_INTR(NULL, NULL);
1605                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1606                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1607                          &lwi);
1608         }
1609
1610         /* There must be big cache in current thread to process this request
1611          * if it is NULL then something went wrong and it wasn't allocated,
1612          * report -ENOMEM in that case */
1613         if (tbc == NULL)
1614                 RETURN(-ENOMEM);
1615
1616         body = tsi->tsi_ost_body;
1617         LASSERT(body != NULL);
1618
1619         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1620         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1621
1622         niocount = ioo->ioo_bufcnt;
1623         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1624         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1625
1626         local_nb = tbc->local;
1627
1628         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1629                           remote_nb, &lockh, LCK_PR);
1630         if (rc != 0)
1631                 RETURN(rc);
1632
1633         /*
1634          * If getting the lock took more time than
1635          * client was willing to wait, drop it. b=11330
1636          */
1637         if (cfs_time_current_sec() > req->rq_deadline ||
1638             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1639                 no_reply = 1;
1640                 CERROR("Dropping timed-out read from %s because locking"
1641                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1642                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1643                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1644                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1645                 GOTO(out_lock, rc = -ETIMEDOUT);
1646         }
1647
1648         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1649         repbody->oa = body->oa;
1650
1651         npages = PTLRPC_MAX_BRW_PAGES;
1652         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1653                         ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1654         if (rc != 0)
1655                 GOTO(out_lock, rc);
1656
1657         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1658                                     BULK_PUT_SOURCE, OST_BULK_PORTAL);
1659         if (desc == NULL)
1660                 GOTO(out_commitrw, rc = -ENOMEM);
1661
1662         nob = 0;
1663         for (i = 0; i < npages; i++) {
1664                 int page_rc = local_nb[i].rc;
1665
1666                 if (page_rc < 0) {
1667                         rc = page_rc;
1668                         break;
1669                 }
1670
1671                 nob += page_rc;
1672                 if (page_rc != 0) { /* some data! */
1673                         LASSERT(local_nb[i].page != NULL);
1674                         ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1675                                                     local_nb[i].lnb_page_offset,
1676                                                     page_rc);
1677                 }
1678
1679                 if (page_rc != local_nb[i].len) { /* short read */
1680                         /* All subsequent pages should be 0 */
1681                         while (++i < npages)
1682                                 LASSERT(local_nb[i].rc == 0);
1683                         break;
1684                 }
1685         }
1686
1687         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1688                 cksum_type_t cksum_type =
1689                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1690                                           body->oa.o_flags : 0);
1691                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1692                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1693                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1694                                                         OST_READ, cksum_type);
1695                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1696                        repbody->oa.o_cksum);
1697         } else {
1698                 repbody->oa.o_valid = 0;
1699         }
1700         /* We're finishing using body->oa as an input variable */
1701
1702         /* Check if client was evicted while we were doing i/o before touching
1703          * network */
1704         if (likely(rc == 0 &&
1705                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1706                 rc = target_bulk_io(exp, desc, &lwi);
1707                 no_reply = rc != 0;
1708         }
1709
1710 out_commitrw:
1711         /* Must commit after prep above in all cases */
1712         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1713                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1714                           NULL, rc);
1715         if (rc == 0)
1716                 tgt_drop_id(exp, &repbody->oa);
1717 out_lock:
1718         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1719
1720         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1721                 ptlrpc_free_bulk_nopin(desc);
1722
1723         LASSERT(rc <= 0);
1724         if (rc == 0) {
1725                 rc = nob;
1726                 ptlrpc_lprocfs_brw(req, nob);
1727         } else if (no_reply) {
1728                 req->rq_no_reply = 1;
1729                 /* reply out callback would free */
1730                 ptlrpc_req_drop_rs(req);
1731                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1732                               "client will retry: rc %d\n",
1733                               exp->exp_obd->obd_name,
1734                               obd_uuid2str(&exp->exp_client_uuid),
1735                               obd_export_nid2str(exp), rc);
1736         }
1737         /* send a bulk after reply to simulate a network delay or reordering
1738          * by a router */
1739         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1740                 wait_queue_head_t        waitq;
1741                 struct l_wait_info       lwi1;
1742
1743                 CDEBUG(D_INFO, "reorder BULK\n");
1744                 init_waitqueue_head(&waitq);
1745
1746                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1747                 l_wait_event(waitq, 0, &lwi1);
1748                 target_bulk_io(exp, desc, &lwi);
1749                 ptlrpc_free_bulk_nopin(desc);
1750         }
1751
1752         RETURN(rc);
1753 }
1754 EXPORT_SYMBOL(tgt_brw_read);
1755
1756 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1757                               struct ptlrpc_bulk_desc *desc,
1758                               struct niobuf_local *local_nb, int npages,
1759                               obd_count client_cksum, obd_count server_cksum,
1760                               bool mmap)
1761 {
1762         struct obd_export *exp = req->rq_export;
1763         struct ost_body *body;
1764         char *router;
1765         char *via;
1766
1767         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1768         LASSERT(body != NULL);
1769
1770         if (req->rq_peer.nid == desc->bd_sender) {
1771                 via = router = "";
1772         } else {
1773                 via = " via ";
1774                 router = libcfs_nid2str(desc->bd_sender);
1775         }
1776
1777         if (mmap) {
1778                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1779                              client_cksum, server_cksum);
1780                 return;
1781         }
1782
1783         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1784                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1785                            "]: client csum %x, server csum %x\n",
1786                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1787                            via, router,
1788                            body->oa.o_valid & OBD_MD_FLFID ?
1789                            body->oa.o_parent_seq : (__u64)0,
1790                            body->oa.o_valid & OBD_MD_FLFID ?
1791                            body->oa.o_parent_oid : 0,
1792                            body->oa.o_valid & OBD_MD_FLFID ?
1793                            body->oa.o_parent_ver : 0,
1794                            POSTID(&body->oa.o_oi),
1795                            local_nb[0].lnb_file_offset,
1796                            local_nb[npages-1].lnb_file_offset +
1797                            local_nb[npages-1].len - 1,
1798                            client_cksum, server_cksum);
1799 }
1800
1801 int tgt_brw_write(struct tgt_session_info *tsi)
1802 {
1803         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1804         struct ptlrpc_bulk_desc *desc = NULL;
1805         struct obd_export       *exp = req->rq_export;
1806         struct niobuf_remote    *remote_nb;
1807         struct niobuf_local     *local_nb;
1808         struct obd_ioobj        *ioo;
1809         struct ost_body         *body, *repbody;
1810         struct l_wait_info       lwi;
1811         struct lustre_handle     lockh = {0};
1812         __u32                   *rcs;
1813         int                      objcount, niocount, npages;
1814         int                      rc, i, j;
1815         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1816         bool                     no_reply = false, mmap;
1817         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1818
1819         ENTRY;
1820
1821         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1822                 CERROR("%s: deny write request from %s to portal %u\n",
1823                        tgt_name(tsi->tsi_tgt),
1824                        obd_export_nid2str(req->rq_export),
1825                        ptlrpc_req2svc(req)->srv_req_portal);
1826                 RETURN(err_serious(-EPROTO));
1827         }
1828
1829         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1830                 RETURN(err_serious(-ENOSPC));
1831         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1832                 RETURN(err_serious(-EROFS));
1833
1834         req->rq_bulk_write = 1;
1835
1836         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1837                 RETURN(err_serious(-EIO));
1838         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1839                 RETURN(err_serious(-EFAULT));
1840
1841         /* pause before transaction has been started */
1842         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1843                          cfs_fail_val : (obd_timeout + 1) / 4);
1844
1845         /* There must be big cache in current thread to process this request
1846          * if it is NULL then something went wrong and it wasn't allocated,
1847          * report -ENOMEM in that case */
1848         if (tbc == NULL)
1849                 RETURN(-ENOMEM);
1850
1851         body = tsi->tsi_ost_body;
1852         LASSERT(body != NULL);
1853
1854         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1855         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1856
1857         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1858                                         RCL_CLIENT) / sizeof(*ioo);
1859
1860         for (niocount = i = 0; i < objcount; i++)
1861                 niocount += ioo[i].ioo_bufcnt;
1862
1863         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1864         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1865         if (niocount != req_capsule_get_size(&req->rq_pill,
1866                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1867                         sizeof(*remote_nb))
1868                 RETURN(err_serious(-EPROTO));
1869
1870         if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
1871             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1872                 memory_pressure_set();
1873
1874         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1875                              niocount * sizeof(*rcs));
1876         rc = req_capsule_server_pack(&req->rq_pill);
1877         if (rc != 0)
1878                 GOTO(out, rc = err_serious(rc));
1879
1880         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1881         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1882
1883         local_nb = tbc->local;
1884
1885         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1886                           remote_nb, &lockh, LCK_PW);
1887         if (rc != 0)
1888                 GOTO(out, rc);
1889
1890         /*
1891          * If getting the lock took more time than
1892          * client was willing to wait, drop it. b=11330
1893          */
1894         if (cfs_time_current_sec() > req->rq_deadline ||
1895             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1896                 no_reply = true;
1897                 CERROR("%s: Dropping timed-out write from %s because locking "
1898                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1899                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1900                        POSTID(&ioo->ioo_oid),
1901                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1902                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1903                 GOTO(out_lock, rc = -ETIMEDOUT);
1904         }
1905
1906         /* Because we already sync grant info with client when reconnect,
1907          * grant info will be cleared for resent req, then fed_grant and
1908          * total_grant will not be modified in following preprw_write */
1909         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1910                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1911                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1912         }
1913
1914         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1915         if (repbody == NULL)
1916                 GOTO(out_lock, rc = -ENOMEM);
1917         repbody->oa = body->oa;
1918
1919         npages = PTLRPC_MAX_BRW_PAGES;
1920         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1921                         objcount, ioo, remote_nb, &npages, local_nb, NULL,
1922                         BYPASS_CAPA);
1923         if (rc < 0)
1924                 GOTO(out_lock, rc);
1925
1926         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1927                                     BULK_GET_SINK, OST_BULK_PORTAL);
1928         if (desc == NULL)
1929                 GOTO(skip_transfer, rc = -ENOMEM);
1930
1931         /* NB Having prepped, we must commit... */
1932         for (i = 0; i < npages; i++)
1933                 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1934                                             local_nb[i].lnb_page_offset,
1935                                             local_nb[i].len);
1936
1937         rc = sptlrpc_svc_prep_bulk(req, desc);
1938         if (rc != 0)
1939                 GOTO(skip_transfer, rc);
1940
1941         rc = target_bulk_io(exp, desc, &lwi);
1942         no_reply = rc != 0;
1943
1944 skip_transfer:
1945         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
1946                 static int cksum_counter;
1947
1948                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1949                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1950
1951                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1952                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1953                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1954                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1955                                                         OST_WRITE, cksum_type);
1956                 cksum_counter++;
1957
1958                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
1959                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
1960                                 body->oa.o_flags & OBD_FL_MMAP);
1961
1962                         tgt_warn_on_cksum(req, desc, local_nb, npages,
1963                                           body->oa.o_cksum,
1964                                           repbody->oa.o_cksum, mmap);
1965                         cksum_counter = 0;
1966                 } else if ((cksum_counter & (-cksum_counter)) ==
1967                            cksum_counter) {
1968                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1969                                cksum_counter, libcfs_id2str(req->rq_peer),
1970                                repbody->oa.o_cksum);
1971                 }
1972         }
1973
1974         /* Must commit after prep above in all cases */
1975         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1976                           objcount, ioo, remote_nb, npages, local_nb, NULL,
1977                           rc);
1978         if (rc == -ENOTCONN)
1979                 /* quota acquire process has been given up because
1980                  * either the client has been evicted or the client
1981                  * has timed out the request already */
1982                 no_reply = true;
1983
1984         /*
1985          * Disable sending mtime back to the client. If the client locked the
1986          * whole object, then it has already updated the mtime on its side,
1987          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1988          */
1989         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1990
1991         if (rc == 0) {
1992                 int nob = 0;
1993
1994                 /* set per-requested niobuf return codes */
1995                 for (i = j = 0; i < niocount; i++) {
1996                         int len = remote_nb[i].len;
1997
1998                         nob += len;
1999                         rcs[i] = 0;
2000                         do {
2001                                 LASSERT(j < npages);
2002                                 if (local_nb[j].rc < 0)
2003                                         rcs[i] = local_nb[j].rc;
2004                                 len -= local_nb[j].len;
2005                                 j++;
2006                         } while (len > 0);
2007                         LASSERT(len == 0);
2008                 }
2009                 LASSERT(j == npages);
2010                 ptlrpc_lprocfs_brw(req, nob);
2011
2012                 tgt_drop_id(exp, &repbody->oa);
2013         }
2014 out_lock:
2015         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2016         if (desc)
2017                 ptlrpc_free_bulk_nopin(desc);
2018 out:
2019         if (no_reply) {
2020                 req->rq_no_reply = 1;
2021                 /* reply out callback would free */
2022                 ptlrpc_req_drop_rs(req);
2023                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2024                               "client will retry: rc %d\n",
2025                               exp->exp_obd->obd_name,
2026                               obd_uuid2str(&exp->exp_client_uuid),
2027                               obd_export_nid2str(exp), rc);
2028         }
2029         memory_pressure_clr();
2030         RETURN(rc);
2031 }
2032 EXPORT_SYMBOL(tgt_brw_write);