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) 2012, 2015, 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 <linux/user_namespace.h>
  38 #ifdef HAVE_UIDGID_HEADER
  39 # include <linux/uidgid.h>
  40 #endif
  41
  42 #include <obd.h>
  43 #include <obd_class.h>
  44 #include <obd_cksum.h>
  45 #include <md_object.h>
  46 #include <lustre_lfsck.h>
  47 #include <lustre_nodemap.h>
  48
  49 #include "tgt_internal.h"
  50
  51 char *tgt_name(struct lu_target *tgt)
  52 {
  53         LASSERT(tgt->lut_obd != NULL);
  54         return tgt->lut_obd->obd_name;
  55 }
  56 EXPORT_SYMBOL(tgt_name);
  57
  58 /*
  59  * Generic code handling requests that have struct mdt_body passed in:
  60  *
  61  *  - extract mdt_body from request and save it in @tsi, if present;
  62  *
  63  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  64  *  @tsi;
  65  *
  66  *  - if HABEO_CORPUS flag is set for this request type check whether object
  67  *  actually exists on storage (lu_object_exists()).
  68  *
  69  */
  70 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  71 {
  72         const struct mdt_body   *body;
  73         struct lu_object        *obj;
  74         struct req_capsule      *pill = tsi->tsi_pill;
  75         int                      rc;
  76
  77         ENTRY;
  78
  79         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  80         if (body == NULL)
  81                 RETURN(-EFAULT);
  82
  83         tsi->tsi_mdt_body = body;
  84
  85         if (!(body->mbo_valid & OBD_MD_FLID))
  86                 RETURN(0);
  87
  88         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  89          * in any case in pre-2.5 clients. Fix that here if needed */
  90         if (unlikely(fid_is_zero(&body->mbo_fid1)))
  91                 RETURN(0);
  92
  93         if (!fid_is_sane(&body->mbo_fid1)) {
  94                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  95                        PFID(&body->mbo_fid1));
  96                 RETURN(-EINVAL);
  97         }
  98
  99         obj = lu_object_find(tsi->tsi_env,
 100                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
 101                              &body->mbo_fid1, NULL);
 102         if (!IS_ERR(obj)) {
 103                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
 104                         lu_object_put(tsi->tsi_env, obj);
 105                         rc = -ENOENT;
 106                 } else {
 107                         tsi->tsi_corpus = obj;
 108                         rc = 0;
 109                 }
 110         } else {
 111                 rc = PTR_ERR(obj);
 112         }
 113
 114         tsi->tsi_fid = body->mbo_fid1;
 115
 116         RETURN(rc);
 117 }
 118
 119 /**
 120  * Validate oa from client.
 121  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 122  * req are valid.
 123  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 124  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 125  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 126  *       whether this is oi_fid or real ostid. So it will check
 127  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 128  *    c. Old FID-disable osc will send IDIF.
 129  *    d. new FID-enable osc/osp will send normal FID.
 130  *
 131  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 132  * be used for LAST_ID file, and only being accessed inside OST now.
 133  */
 134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 135 {
 136         struct ost_id   *oi     = &oa->o_oi;
 137         u64              seq    = ostid_seq(oi);
 138         u64              id     = ostid_id(oi);
 139         int              rc;
 140         ENTRY;
 141
 142         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
 143                      fid_seq_is_echo(seq))) {
 144                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 145                  * during create, and we will reset this to 1, since this
 146                  * oi_id is basically useless in the following create process,
 147                  * but oi_id == 0 will make it difficult to tell whether it is
 148                  * real FID or ost_id. */
 149                 oi->oi_fid.f_seq = FID_SEQ_ECHO;
 150                 oi->oi_fid.f_oid = id ?: 1;
 151                 oi->oi_fid.f_ver = 0;
 152         } else {
 153                 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
 154
 155                 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
 156                         GOTO(out, rc = -EPROTO);
 157
 158                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 159                  * all of the requests are required to setup FLGROUP */
 160                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 161                         ostid_set_seq_mdt0(oi);
 162                         oa->o_valid |= OBD_MD_FLGROUP;
 163                         seq = ostid_seq(oi);
 164                 }
 165
 166                 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
 167                                fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
 168                         GOTO(out, rc = -EPROTO);
 169
 170                 rc = ostid_to_fid(&tti->tti_fid1, oi,
 171                                   tsi->tsi_tgt->lut_lsd.lsd_osd_index);
 172                 if (unlikely(rc != 0))
 173                         GOTO(out, rc);
 174
 175                 oi->oi_fid = tti->tti_fid1;
 176         }
 177
 178         RETURN(0);
 179
 180 out:
 181         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 182                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
 183                seq, id, rc);
 184         return rc;
 185 }
 186 EXPORT_SYMBOL(tgt_validate_obdo);
 187
 188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
 189 {
 190         unsigned                 max_brw;
 191         struct niobuf_remote    *rnb;
 192         struct obd_ioobj        *ioo;
 193         int                      obj_count;
 194
 195         ENTRY;
 196
 197         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
 198         if (ioo == NULL)
 199                 RETURN(-EPROTO);
 200
 201         rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
 202         if (rnb == NULL)
 203                 RETURN(-EPROTO);
 204
 205         max_brw = ioobj_max_brw_get(ioo);
 206         if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 207                 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 208                        ": rc = %d\n", tgt_name(tsi->tsi_tgt),
 209                        obd_export_nid2str(tsi->tsi_exp), max_brw,
 210                        POSTID(oi), -EPROTO);
 211                 RETURN(-EPROTO);
 212         }
 213         ioo->ioo_oid = *oi;
 214
 215         obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
 216                                         RCL_CLIENT) / sizeof(*ioo);
 217         if (obj_count == 0) {
 218                 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
 219                 RETURN(-EPROTO);
 220         } else if (obj_count > 1) {
 221                 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
 222                        obj_count);
 223                 RETURN(-EPROTO);
 224         }
 225
 226         if (ioo->ioo_bufcnt == 0) {
 227                 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
 228                 RETURN(-EPROTO);
 229         }
 230
 231         if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
 232                 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
 233                           "bulk has too many pages (%d)",
 234                           ioo->ioo_bufcnt);
 235                 RETURN(-EPROTO);
 236         }
 237
 238         RETURN(0);
 239 }
 240
 241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 242 {
 243         struct ost_body         *body;
 244         struct req_capsule      *pill = tsi->tsi_pill;
 245         struct lu_nodemap       *nodemap;
 246         int                      rc;
 247
 248         ENTRY;
 249
 250         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 251         if (body == NULL)
 252                 RETURN(-EFAULT);
 253
 254         rc = tgt_validate_obdo(tsi, &body->oa);
 255         if (rc)
 256                 RETURN(rc);
 257
 258         nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
 259
 260         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 261                                         NODEMAP_CLIENT_TO_FS,
 262                                         body->oa.o_uid);
 263         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 264                                         NODEMAP_CLIENT_TO_FS,
 265                                         body->oa.o_gid);
 266
 267         tsi->tsi_ost_body = body;
 268         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 269
 270         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 271                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 272                 if (rc < 0)
 273                         RETURN(rc);
 274         }
 275
 276         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 277                 if (flags & HABEO_CORPUS) {
 278                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 279                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 280                                tgt_name(tsi->tsi_tgt));
 281                         RETURN(-EPROTO);
 282                 } else {
 283                         RETURN(0);
 284                 }
 285         }
 286
 287         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 288
 289         /*
 290          * OST doesn't get object in advance for further use to prevent
 291          * situations with nested object_find which is potential deadlock.
 292          */
 293         tsi->tsi_corpus = NULL;
 294         RETURN(rc);
 295 }
 296
 297 /*
 298  * Do necessary preprocessing according to handler ->th_flags.
 299  */
 300 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 301                                   struct tgt_handler *h,
 302                                   struct ptlrpc_request *req)
 303 {
 304         struct req_capsule      *pill = tsi->tsi_pill;
 305         __u32                    flags = h->th_flags;
 306         int                      rc = 0;
 307
 308         ENTRY;
 309
 310         if (tsi->tsi_preprocessed)
 311                 RETURN(0);
 312
 313         LASSERT(h->th_act != NULL);
 314         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 315         LASSERT(current->journal_info == NULL);
 316
 317         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 318                      h->th_fmt != NULL));
 319         if (h->th_fmt != NULL) {
 320                 req_capsule_set(pill, h->th_fmt);
 321                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 322                         rc = tgt_mdt_body_unpack(tsi, flags);
 323                         if (rc < 0)
 324                                 RETURN(rc);
 325                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 326                                                  RCL_CLIENT)) {
 327                         rc = tgt_ost_body_unpack(tsi, flags);
 328                         if (rc < 0)
 329                                 RETURN(rc);
 330                 }
 331         }
 332
 333         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 334                 RETURN(-EROFS);
 335
 336         if (flags & HABEO_CLAVIS) {
 337                 struct ldlm_request *dlm_req;
 338
 339                 LASSERT(h->th_fmt != NULL);
 340
 341                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 342                 if (dlm_req != NULL) {
 343                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 344                                      LDLM_IBITS &&
 345                                      dlm_req->lock_desc.l_policy_data.\
 346                                      l_inodebits.bits == 0)) {
 347                                 /*
 348                                  * Lock without inodebits makes no sense and
 349                                  * will oops later in ldlm. If client miss to
 350                                  * set such bits, do not trigger ASSERTION.
 351                                  *
 352                                  * For liblustre flock case, it maybe zero.
 353                                  */
 354                                 rc = -EPROTO;
 355                         } else {
 356                                 tsi->tsi_dlm_req = dlm_req;
 357                         }
 358                 } else {
 359                         rc = -EFAULT;
 360                 }
 361         }
 362         tsi->tsi_preprocessed = 1;
 363         RETURN(rc);
 364 }
 365
 366 /*
 367  * Invoke handler for this request opc. Also do necessary preprocessing
 368  * (according to handler ->th_flags), and post-processing (setting of
 369  * ->last_{xid,committed}).
 370  */
 371 static int tgt_handle_request0(struct tgt_session_info *tsi,
 372                                struct tgt_handler *h,
 373                                struct ptlrpc_request *req)
 374 {
 375         int      serious = 0;
 376         int      rc;
 377         __u32    opc = lustre_msg_get_opc(req->rq_reqmsg);
 378
 379         ENTRY;
 380
 381
 382         /* When dealing with sec context requests, no export is associated yet,
 383          * because these requests are sent before *_CONNECT requests.
 384          * A NULL req->rq_export means the normal *_common_slice handlers will
 385          * not be called, because there is no reference to the target.
 386          * So deal with them by hand and jump directly to target_send_reply().
 387          */
 388         switch (opc) {
 389         case SEC_CTX_INIT:
 390         case SEC_CTX_INIT_CONT:
 391         case SEC_CTX_FINI:
 392                 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
 393                 GOTO(out, rc = 0);
 394         }
 395
 396         /*
 397          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 398          * to put same checks into handlers like mdt_close(), mdt_reint(),
 399          * etc., without talking to mdt authors first. Checking same thing
 400          * there again is useless and returning 0 error without packing reply
 401          * is buggy! Handlers either pack reply or return error.
 402          *
 403          * We return 0 here and do not send any reply in order to emulate
 404          * network failure. Do not send any reply in case any of NET related
 405          * fail_id has occured.
 406          */
 407         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 408                 RETURN(0);
 409         if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
 410                      OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
 411                 RETURN(0);
 412
 413         rc = tgt_request_preprocess(tsi, h, req);
 414         /* pack reply if reply format is fixed */
 415         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 416                 /* Pack reply */
 417                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 418                                           RCL_SERVER))
 419                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 420                                              RCL_SERVER,
 421                                              tsi->tsi_mdt_body->mbo_eadatasize);
 422                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 423                                           RCL_SERVER))
 424                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 425                                              RCL_SERVER, 0);
 426
 427                 rc = req_capsule_server_pack(tsi->tsi_pill);
 428         }
 429
 430         if (likely(rc == 0)) {
 431                 /*
 432                  * Process request, there can be two types of rc:
 433                  * 1) errors with msg unpack/pack, other failures outside the
 434                  * operation itself. This is counted as serious errors;
 435                  * 2) errors during fs operation, should be placed in rq_status
 436                  * only
 437                  */
 438                 rc = h->th_act(tsi);
 439                 if (!is_serious(rc) &&
 440                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 441                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 442                                   "pack reply and returned 0 error\n",
 443                                   tgt_name(tsi->tsi_tgt), h->th_name);
 444                         LBUG();
 445                 }
 446                 serious = is_serious(rc);
 447                 rc = clear_serious(rc);
 448         } else {
 449                 serious = 1;
 450         }
 451
 452         req->rq_status = rc;
 453
 454         /*
 455          * ELDLM_* codes which > 0 should be in rq_status only as well as
 456          * all non-serious errors.
 457          */
 458         if (rc > 0 || !serious)
 459                 rc = 0;
 460
 461         LASSERT(current->journal_info == NULL);
 462
 463         if (likely(rc == 0 && req->rq_export))
 464                 target_committed_to_req(req);
 465
 466 out:
 467         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 468         RETURN(0);
 469 }
 470
 471 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 472                                        struct obd_device *obd, int *process)
 473 {
 474         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 475         case MDS_DISCONNECT:
 476         case OST_DISCONNECT:
 477         case OBD_IDX_READ:
 478                 *process = 1;
 479                 RETURN(0);
 480         case MDS_CLOSE:
 481         case MDS_SYNC: /* used in unmounting */
 482         case OBD_PING:
 483         case MDS_REINT:
 484         case OUT_UPDATE:
 485         case SEQ_QUERY:
 486         case FLD_QUERY:
 487         case FLD_READ:
 488         case LDLM_ENQUEUE:
 489         case OST_CREATE:
 490         case OST_DESTROY:
 491         case OST_PUNCH:
 492         case OST_SETATTR:
 493         case OST_SYNC:
 494         case OST_WRITE:
 495         case MDS_HSM_PROGRESS:
 496         case MDS_HSM_STATE_SET:
 497         case MDS_HSM_REQUEST:
 498                 *process = target_queue_recovery_request(req, obd);
 499                 RETURN(0);
 500
 501         default:
 502                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 503                 *process = -EAGAIN;
 504                 RETURN(0);
 505         }
 506 }
 507
 508 /*
 509  * Handle recovery. Return:
 510  *        +1: continue request processing;
 511  *       -ve: abort immediately with the given error code;
 512  *         0: send reply with error code in req->rq_status;
 513  */
 514 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 515 {
 516         ENTRY;
 517
 518         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 519         case MDS_CONNECT:
 520         case OST_CONNECT:
 521         case MGS_CONNECT:
 522         case SEC_CTX_INIT:
 523         case SEC_CTX_INIT_CONT:
 524         case SEC_CTX_FINI:
 525                 RETURN(+1);
 526         }
 527
 528         if (!req->rq_export->exp_obd->obd_replayable)
 529                 RETURN(+1);
 530
 531         /* sanity check: if the xid matches, the request must be marked as a
 532          * resent or replayed */
 533         if (req_can_reconstruct(req, NULL)) {
 534                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 535                       (MSG_RESENT | MSG_REPLAY))) {
 536                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 537                                   "saved xid, expected REPLAY or RESENT flag "
 538                                   "(%x)", req->rq_xid,
 539                                   lustre_msg_get_flags(req->rq_reqmsg));
 540                         req->rq_status = -ENOTCONN;
 541                         RETURN(-ENOTCONN);
 542                 }
 543         }
 544         /* else: note the opposite is not always true; a RESENT req after a
 545          * failover will usually not match the last_xid, since it was likely
 546          * never committed. A REPLAYed request will almost never match the
 547          * last xid, however it could for a committed, but still retained,
 548          * open. */
 549
 550         /* Check for aborted recovery... */
 551         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 552                 int rc;
 553                 int should_process;
 554
 555                 DEBUG_REQ(D_INFO, req, "Got new replay");
 556                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 557                                                  &should_process);
 558                 if (rc != 0 || !should_process)
 559                         RETURN(rc);
 560                 else if (should_process < 0) {
 561                         req->rq_status = should_process;
 562                         rc = ptlrpc_error(req);
 563                         RETURN(rc);
 564                 }
 565         }
 566         RETURN(+1);
 567 }
 568
 569 /* Initial check for request, it is validation mostly */
 570 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 571 {
 572         struct tgt_handler      *h;
 573         struct tgt_opc_slice    *s;
 574         struct lu_target        *tgt;
 575         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 576
 577         ENTRY;
 578
 579         tgt = class_exp2tgt(req->rq_export);
 580         if (unlikely(tgt == NULL)) {
 581                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 582                           class_exp2obd(req->rq_export)->obd_name);
 583                 RETURN(ERR_PTR(-EINVAL));
 584         }
 585
 586         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 587                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 588                         break;
 589
 590         /* opcode was not found in slice */
 591         if (unlikely(s->tos_hs == NULL)) {
 592                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 593                        opc);
 594                 RETURN(ERR_PTR(-ENOTSUPP));
 595         }
 596
 597         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 598         h = s->tos_hs + (opc - s->tos_opc_start);
 599         if (unlikely(h->th_opc == 0)) {
 600                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 601                 RETURN(ERR_PTR(-ENOTSUPP));
 602         }
 603
 604         RETURN(h);
 605 }
 606
 607 static int process_req_last_xid(struct ptlrpc_request *req)
 608 {
 609         __u64   last_xid;
 610         ENTRY;
 611
 612         /* check request's xid is consistent with export's last_xid */
 613         last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
 614         if (last_xid > req->rq_export->exp_last_xid)
 615                 req->rq_export->exp_last_xid = last_xid;
 616
 617         if (req->rq_xid == 0 ||
 618             (req->rq_xid <= req->rq_export->exp_last_xid)) {
 619                 DEBUG_REQ(D_ERROR, req, "Unexpected xid %llx vs. "
 620                           "last_xid %llx\n", req->rq_xid,
 621                           req->rq_export->exp_last_xid);
 622                 /* Some request is allowed to be sent during replay,
 623                  * such as OUT update requests, FLD requests, so it
 624                  * is possible that replay requests has smaller XID
 625                  * than the exp_last_xid.
 626                  *
 627                  * Some non-replay requests may have smaller XID as
 628                  * well:
 629                  *
 630                  * - Client send a no_resend RPC, like statfs;
 631                  * - The RPC timedout (or some other error) on client,
 632                  *   then it's removed from the unreplied list;
 633                  * - Client send some other request to bump the
 634                  *   exp_last_xid on server;
 635                  * - The former RPC got chance to be processed;
 636                  */
 637                 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
 638                         RETURN(-EPROTO);
 639         }
 640
 641         /* try to release in-memory reply data */
 642         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 643                 tgt_handle_received_xid(req->rq_export,
 644                                 lustre_msg_get_last_xid(req->rq_reqmsg));
 645                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 646                       (MSG_RESENT | MSG_REPLAY)))
 647                         tgt_handle_tag(req->rq_export,
 648                                        lustre_msg_get_tag(req->rq_reqmsg));
 649         }
 650         RETURN(0);
 651 }
 652
 653 int tgt_request_handle(struct ptlrpc_request *req)
 654 {
 655         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 656
 657         struct lustre_msg       *msg = req->rq_reqmsg;
 658         struct tgt_handler      *h;
 659         struct lu_target        *tgt;
 660         int                      request_fail_id = 0;
 661         __u32                    opc = lustre_msg_get_opc(msg);
 662         struct obd_device       *obd;
 663         int                      rc;
 664         bool                     is_connect = false;
 665         ENTRY;
 666
 667         /* Refill the context, to make sure all thread keys are allocated */
 668         lu_env_refill(req->rq_svc_thread->t_env);
 669
 670         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 671         tsi->tsi_pill = &req->rq_pill;
 672         tsi->tsi_env = req->rq_svc_thread->t_env;
 673
 674         /* if request has export then get handlers slice from corresponding
 675          * target, otherwise that should be connect operation */
 676         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 677             opc == MGS_CONNECT) {
 678                 is_connect = true;
 679                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 680                 rc = target_handle_connect(req);
 681                 if (rc != 0) {
 682                         rc = ptlrpc_error(req);
 683                         GOTO(out, rc);
 684                 }
 685                 /* recovery-small test 18c asks to drop connect reply */
 686                 if (unlikely(opc == OST_CONNECT &&
 687                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 688                         GOTO(out, rc = 0);
 689         }
 690
 691         if (unlikely(!class_connected_export(req->rq_export))) {
 692                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
 693                     opc == SEC_CTX_FINI) {
 694                         /* sec context initialization has to be handled
 695                          * by hand in tgt_handle_request0() */
 696                         tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
 697                         h = NULL;
 698                         GOTO(handle_recov, rc = 0);
 699                 }
 700                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 701                        opc, libcfs_id2str(req->rq_peer));
 702                 req->rq_status = -ENOTCONN;
 703                 rc = ptlrpc_error(req);
 704                 GOTO(out, rc);
 705         }
 706
 707         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 708         tsi->tsi_exp = req->rq_export;
 709         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 710                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 711         else
 712                 tsi->tsi_jobid = NULL;
 713
 714         if (tgt == NULL) {
 715                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 716                           class_exp2obd(req->rq_export)->obd_name);
 717                 req->rq_status = -EINVAL;
 718                 rc = ptlrpc_error(req);
 719                 GOTO(out, rc);
 720         }
 721
 722         /* Skip last_xid processing for the recovery thread, otherwise, the
 723          * last_xid on same request could be processed twice: first time when
 724          * processing the incoming request, second time when the request is
 725          * being processed by recovery thread. */
 726         obd = class_exp2obd(req->rq_export);
 727         if (is_connect) {
 728                 /* reset the exp_last_xid on each connection. */
 729                 req->rq_export->exp_last_xid = 0;
 730         } else if (obd->obd_recovery_data.trd_processing_task !=
 731                    current_pid()) {
 732                 rc = process_req_last_xid(req);
 733                 if (rc) {
 734                         req->rq_status = rc;
 735                         rc = ptlrpc_error(req);
 736                         GOTO(out, rc);
 737                 }
 738         }
 739
 740         request_fail_id = tgt->lut_request_fail_id;
 741         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 742
 743         h = tgt_handler_find_check(req);
 744         if (IS_ERR(h)) {
 745                 req->rq_status = PTR_ERR(h);
 746                 rc = ptlrpc_error(req);
 747                 GOTO(out, rc);
 748         }
 749
 750         LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 751                  h->th_opc, opc);
 752
 753         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 754                 GOTO(out, rc = 0);
 755
 756         rc = lustre_msg_check_version(msg, h->th_version);
 757         if (unlikely(rc)) {
 758                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 759                           " %08x, expecting %08x\n", tgt_name(tgt),
 760                           lustre_msg_get_version(msg), h->th_version);
 761                 req->rq_status = -EINVAL;
 762                 rc = ptlrpc_error(req);
 763                 GOTO(out, rc);
 764         }
 765
 766 handle_recov:
 767         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 768         if (likely(rc == 1)) {
 769                 rc = tgt_handle_request0(tsi, h, req);
 770                 if (rc)
 771                         GOTO(out, rc);
 772         }
 773         EXIT;
 774 out:
 775         req_capsule_fini(tsi->tsi_pill);
 776         if (tsi->tsi_corpus != NULL) {
 777                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 778                 tsi->tsi_corpus = NULL;
 779         }
 780         return rc;
 781 }
 782 EXPORT_SYMBOL(tgt_request_handle);
 783
 784 /** Assign high priority operations to the request if needed. */
 785 int tgt_hpreq_handler(struct ptlrpc_request *req)
 786 {
 787         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 788         struct tgt_handler      *h;
 789         int                      rc;
 790
 791         ENTRY;
 792
 793         if (req->rq_export == NULL)
 794                 RETURN(0);
 795
 796         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 797         tsi->tsi_pill = &req->rq_pill;
 798         tsi->tsi_env = req->rq_svc_thread->t_env;
 799         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 800         tsi->tsi_exp = req->rq_export;
 801
 802         h = tgt_handler_find_check(req);
 803         if (IS_ERR(h)) {
 804                 rc = PTR_ERR(h);
 805                 RETURN(rc);
 806         }
 807
 808         rc = tgt_request_preprocess(tsi, h, req);
 809         if (unlikely(rc != 0))
 810                 RETURN(rc);
 811
 812         if (h->th_hp != NULL)
 813                 h->th_hp(tsi);
 814         RETURN(0);
 815 }
 816 EXPORT_SYMBOL(tgt_hpreq_handler);
 817
 818 void tgt_counter_incr(struct obd_export *exp, int opcode)
 819 {
 820         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 821         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 822                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 823 }
 824 EXPORT_SYMBOL(tgt_counter_incr);
 825
 826 /*
 827  * Unified target generic handlers.
 828  */
 829
 830 /*
 831  * Security functions
 832  */
 833 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 834 {
 835         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 836                                       OBD_CONNECT_RMT_CLIENT_FORCE);
 837 }
 838
 839 static int tgt_init_sec_level(struct ptlrpc_request *req)
 840 {
 841         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 842         char                    *client;
 843         struct obd_connect_data *data, *reply;
 844         int                      rc = 0;
 845         bool                     remote;
 846         ENTRY;
 847
 848         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 849         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 850         if (data == NULL || reply == NULL)
 851                 RETURN(-EFAULT);
 852
 853         /* connection from MDT is always trusted */
 854         if (req->rq_auth_usr_mdt) {
 855                 tgt_init_sec_none(reply);
 856                 RETURN(0);
 857         }
 858
 859         if (unlikely(tgt == NULL)) {
 860                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 861                           class_exp2obd(req->rq_export)->obd_name);
 862                 RETURN(-EINVAL);
 863         }
 864
 865         client = libcfs_nid2str(req->rq_peer.nid);
 866         /* no GSS support case */
 867         if (!req->rq_auth_gss) {
 868                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 869                         CWARN("client %s -> target %s does not use GSS, "
 870                               "can not run under security level %d.\n",
 871                               client, tgt_name(tgt), tgt->lut_sec_level);
 872                         RETURN(-EACCES);
 873                 } else {
 874                         tgt_init_sec_none(reply);
 875                         RETURN(0);
 876                 }
 877         }
 878
 879         /* old version case */
 880         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 881                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 882                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 883                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 884                         CWARN("client %s -> target %s uses old version, "
 885                               "can not run under security level %d.\n",
 886                               client, tgt_name(tgt), tgt->lut_sec_level);
 887                         RETURN(-EACCES);
 888                 } else {
 889                         CWARN("client %s -> target %s uses old version, "
 890                               "run under security level %d.\n",
 891                               client, tgt_name(tgt), tgt->lut_sec_level);
 892                         tgt_init_sec_none(reply);
 893                         RETURN(0);
 894                 }
 895         }
 896
 897         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 898         if (remote) {
 899                 if (!req->rq_auth_remote)
 900                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 901                                "asked to be remote.\n", client, tgt_name(tgt));
 902         } else if (req->rq_auth_remote) {
 903                 remote = true;
 904                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 905                        "as remote by default.\n", client, tgt_name(tgt));
 906         }
 907
 908         if (remote == 0) {
 909                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 910                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 911                                "authenticated!\n", client, tgt_name(tgt));
 912                         RETURN(-EACCES);
 913                 }
 914         }
 915
 916
 917         switch (tgt->lut_sec_level) {
 918         case LUSTRE_SEC_NONE:
 919                 if (remote) {
 920                         CDEBUG(D_SEC,
 921                                "client %s -> target %s is set as remote, "
 922                                "can not run under security level %d.\n",
 923                                client, tgt_name(tgt), tgt->lut_sec_level);
 924                         RETURN(-EACCES);
 925                 }
 926                 tgt_init_sec_none(reply);
 927                 break;
 928         case LUSTRE_SEC_REMOTE:
 929                 if (!remote)
 930                         tgt_init_sec_none(reply);
 931                 break;
 932         case LUSTRE_SEC_ALL:
 933                 if (remote)
 934                         break;
 935                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 936                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 937                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 938                 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 939                 break;
 940         default:
 941                 RETURN(-EINVAL);
 942         }
 943
 944         RETURN(rc);
 945 }
 946
 947 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 948 {
 949         struct lu_target        *tgt = class_exp2tgt(exp);
 950         struct sptlrpc_flavor    flvr;
 951         int                      rc = 0;
 952
 953         LASSERT(tgt);
 954         LASSERT(tgt->lut_obd);
 955         LASSERT(tgt->lut_slice);
 956
 957         /* always allow ECHO client */
 958         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 959                             LUSTRE_ECHO_NAME) == 0)) {
 960                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 961                 return 0;
 962         }
 963
 964         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 965                 read_lock(&tgt->lut_sptlrpc_lock);
 966                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 967                                              req->rq_sp_from,
 968                                              req->rq_peer.nid,
 969                                              &flvr);
 970                 read_unlock(&tgt->lut_sptlrpc_lock);
 971
 972                 spin_lock(&exp->exp_lock);
 973                 exp->exp_sp_peer = req->rq_sp_from;
 974                 exp->exp_flvr = flvr;
 975
 976                 /* when on mgs, if no restriction is set, or if client
 977                  * is loopback, allow any flavor */
 978                 if ((strcmp(exp->exp_obd->obd_type->typ_name,
 979                            LUSTRE_MGS_NAME) == 0) &&
 980                      (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
 981                       LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
 982                       == LOLND))
 983                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 984
 985                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 986                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 987                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 988                                "expect %x\n", tgt_name(tgt),
 989                                req->rq_flvr.sf_rpc,
 990                                libcfs_nid2str(req->rq_peer.nid),
 991                                exp->exp_flvr.sf_rpc);
 992                         rc = -EACCES;
 993                 }
 994                 spin_unlock(&exp->exp_lock);
 995         } else {
 996                 if (exp->exp_sp_peer != req->rq_sp_from) {
 997                         CERROR("%s: RPC source %s doesn't match %s\n",
 998                                tgt_name(tgt),
 999                                sptlrpc_part2name(req->rq_sp_from),
1000                                sptlrpc_part2name(exp->exp_sp_peer));
1001                         rc = -EACCES;
1002                 } else {
1003                         rc = sptlrpc_target_export_check(exp, req);
1004                 }
1005         }
1006
1007         return rc;
1008 }
1009
1010 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
1011 {
1012         struct sptlrpc_rule_set  tmp_rset;
1013         int                      rc;
1014
1015         if (unlikely(tgt == NULL)) {
1016                 CERROR("No target passed");
1017                 return -EINVAL;
1018         }
1019
1020         sptlrpc_rule_set_init(&tmp_rset);
1021         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
1022         if (rc) {
1023                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
1024                        tgt_name(tgt), rc);
1025                 return rc;
1026         }
1027
1028         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
1029
1030         write_lock(&tgt->lut_sptlrpc_lock);
1031         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
1032         tgt->lut_sptlrpc_rset = tmp_rset;
1033         write_unlock(&tgt->lut_sptlrpc_lock);
1034
1035         return 0;
1036 }
1037 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
1038
1039 int tgt_connect(struct tgt_session_info *tsi)
1040 {
1041         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1042         struct obd_connect_data *reply;
1043         int                      rc;
1044
1045         ENTRY;
1046
1047         rc = tgt_init_sec_level(req);
1048         if (rc != 0)
1049                 GOTO(out, rc);
1050
1051         /* XXX: better to call this check right after getting new export but
1052          * before last_rcvd slot allocation to avoid server load upon insecure
1053          * connects. This is to be fixed after unifiyng all targets.
1054          */
1055         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1056         if (rc)
1057                 GOTO(out, rc);
1058
1059         /* To avoid exposing partially initialized connection flags, changes up
1060          * to this point have been staged in reply->ocd_connect_flags. Now that
1061          * connection handling has completed successfully, atomically update
1062          * the connect flags in the shared export data structure. LU-1623 */
1063         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1064         spin_lock(&tsi->tsi_exp->exp_lock);
1065         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1066         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1067         spin_unlock(&tsi->tsi_exp->exp_lock);
1068
1069         RETURN(0);
1070 out:
1071         obd_disconnect(class_export_get(tsi->tsi_exp));
1072         return rc;
1073 }
1074 EXPORT_SYMBOL(tgt_connect);
1075
1076 int tgt_disconnect(struct tgt_session_info *tsi)
1077 {
1078         int rc;
1079
1080         ENTRY;
1081
1082         rc = target_handle_disconnect(tgt_ses_req(tsi));
1083         if (rc)
1084                 RETURN(err_serious(rc));
1085
1086         RETURN(rc);
1087 }
1088 EXPORT_SYMBOL(tgt_disconnect);
1089
1090 /*
1091  * Unified target OBD handlers
1092  */
1093 int tgt_obd_ping(struct tgt_session_info *tsi)
1094 {
1095         int rc;
1096
1097         ENTRY;
1098
1099         rc = target_handle_ping(tgt_ses_req(tsi));
1100         if (rc)
1101                 RETURN(err_serious(rc));
1102
1103         RETURN(rc);
1104 }
1105 EXPORT_SYMBOL(tgt_obd_ping);
1106
1107 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1108 {
1109         return err_serious(-EOPNOTSUPP);
1110 }
1111
1112 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
1113 {
1114         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1115         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1116         struct obd_export       *exp = req->rq_export;
1117         struct ptlrpc_bulk_desc *desc;
1118         struct l_wait_info      *lwi = &tti->tti_u.update.tti_wait_info;
1119         int                      i;
1120         int                      rc;
1121
1122         ENTRY;
1123
1124         desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
1125                                   PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
1126                                     MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1127         if (desc == NULL)
1128                 RETURN(-ENOMEM);
1129
1130         for (i = 0; i < rdbuf->rb_nbufs; i++)
1131                 desc->bd_frag_ops->add_iov_frag(desc,
1132                                         rdbuf->rb_bufs[i].lb_buf,
1133                                         rdbuf->rb_bufs[i].lb_len);
1134
1135         rc = target_bulk_io(exp, desc, lwi);
1136         ptlrpc_free_bulk(desc);
1137         RETURN(rc);
1138 }
1139 EXPORT_SYMBOL(tgt_send_buffer);
1140
1141 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1142 {
1143         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1144         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1145         struct obd_export       *exp = req->rq_export;
1146         struct ptlrpc_bulk_desc *desc;
1147         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1148         int                      tmpcount;
1149         int                      tmpsize;
1150         int                      i;
1151         int                      rc;
1152
1153         ENTRY;
1154
1155         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1156                                     PTLRPC_BULK_PUT_SOURCE |
1157                                         PTLRPC_BULK_BUF_KIOV,
1158                                     MDS_BULK_PORTAL,
1159                                     &ptlrpc_bulk_kiov_pin_ops);
1160         if (desc == NULL)
1161                 RETURN(-ENOMEM);
1162
1163         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1164                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1165                  * which is rdpg->rp_count */
1166                 nob = rdpg->rp_count;
1167
1168         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1169              i++, tmpcount -= tmpsize) {
1170                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1171                 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1172                                                  tmpsize);
1173         }
1174
1175         LASSERT(desc->bd_nob == nob);
1176         rc = target_bulk_io(exp, desc, lwi);
1177         ptlrpc_free_bulk(desc);
1178         RETURN(rc);
1179 }
1180 EXPORT_SYMBOL(tgt_sendpage);
1181
1182 /*
1183  * OBD_IDX_READ handler
1184  */
1185 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1186 {
1187         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1188         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1189         struct idx_info         *req_ii, *rep_ii;
1190         int                      rc, i;
1191
1192         ENTRY;
1193
1194         memset(rdpg, 0, sizeof(*rdpg));
1195         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1196
1197         /* extract idx_info buffer from request & reply */
1198         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1199         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1200                 RETURN(err_serious(-EPROTO));
1201
1202         rc = req_capsule_server_pack(tsi->tsi_pill);
1203         if (rc)
1204                 RETURN(err_serious(rc));
1205
1206         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1207         if (rep_ii == NULL)
1208                 RETURN(err_serious(-EFAULT));
1209         rep_ii->ii_magic = IDX_INFO_MAGIC;
1210
1211         /* extract hash to start with */
1212         rdpg->rp_hash = req_ii->ii_hash_start;
1213
1214         /* extract requested attributes */
1215         rdpg->rp_attrs = req_ii->ii_attrs;
1216
1217         /* check that fid packed in request is valid and supported */
1218         if (!fid_is_sane(&req_ii->ii_fid))
1219                 RETURN(-EINVAL);
1220         rep_ii->ii_fid = req_ii->ii_fid;
1221
1222         /* copy flags */
1223         rep_ii->ii_flags = req_ii->ii_flags;
1224
1225         /* compute number of pages to allocate, ii_count is the number of 4KB
1226          * containers */
1227         if (req_ii->ii_count <= 0)
1228                 GOTO(out, rc = -EFAULT);
1229         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1230                                exp_max_brw_size(tsi->tsi_exp));
1231         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1232
1233         /* allocate pages to store the containers */
1234         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1235         if (rdpg->rp_pages == NULL)
1236                 GOTO(out, rc = -ENOMEM);
1237         for (i = 0; i < rdpg->rp_npages; i++) {
1238                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1239                 if (rdpg->rp_pages[i] == NULL)
1240                         GOTO(out, rc = -ENOMEM);
1241         }
1242
1243         /* populate pages with key/record pairs */
1244         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1245         if (rc < 0)
1246                 GOTO(out, rc);
1247
1248         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1249                  "asked %d > %d\n", rc, rdpg->rp_count);
1250
1251         /* send pages to client */
1252         rc = tgt_sendpage(tsi, rdpg, rc);
1253         if (rc)
1254                 GOTO(out, rc);
1255         EXIT;
1256 out:
1257         if (rdpg->rp_pages) {
1258                 for (i = 0; i < rdpg->rp_npages; i++)
1259                         if (rdpg->rp_pages[i])
1260                                 __free_page(rdpg->rp_pages[i]);
1261                 OBD_FREE(rdpg->rp_pages,
1262                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1263         }
1264         return rc;
1265 }
1266
1267 struct tgt_handler tgt_obd_handlers[] = {
1268 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1269 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1270 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1271 };
1272 EXPORT_SYMBOL(tgt_obd_handlers);
1273
1274 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1275              struct dt_object *obj, __u64 start, __u64 end)
1276 {
1277         int rc = 0;
1278
1279         ENTRY;
1280
1281         /* if no objid is specified, it means "sync whole filesystem" */
1282         if (obj == NULL) {
1283                 rc = dt_sync(env, tgt->lut_bottom);
1284         } else if (dt_version_get(env, obj) >
1285                    tgt->lut_obd->obd_last_committed) {
1286                 rc = dt_object_sync(env, obj, start, end);
1287         }
1288         atomic_inc(&tgt->lut_sync_count);
1289
1290         RETURN(rc);
1291 }
1292 EXPORT_SYMBOL(tgt_sync);
1293 /*
1294  * Unified target DLM handlers.
1295  */
1296
1297 /**
1298  * Unified target BAST
1299  *
1300  * Ensure data and metadata are synced to disk when lock is canceled if Sync on
1301  * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
1302  * but if it's cross-MDT lock, because remote object version is not set, a
1303  * filesystem sync is needed.
1304  *
1305  * \param lock server side lock
1306  * \param desc lock desc
1307  * \param data ldlm_cb_set_arg
1308  * \param flag  indicates whether this cancelling or blocking callback
1309  * \retval      0 on success
1310  * \retval      negative number on error
1311  */
1312 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1313                             void *data, int flag)
1314 {
1315         struct lu_env            env;
1316         struct lu_target        *tgt;
1317         struct dt_object        *obj = NULL;
1318         struct lu_fid            fid;
1319         int                      rc = 0;
1320
1321         ENTRY;
1322
1323         tgt = class_exp2tgt(lock->l_export);
1324
1325         if (unlikely(tgt == NULL)) {
1326                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1327                        class_exp2obd(lock->l_export)->obd_name);
1328                 RETURN(-EINVAL);
1329         }
1330
1331         if (flag == LDLM_CB_CANCELING &&
1332             (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
1333             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1334              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1335               ldlm_is_cbpending(lock))) &&
1336             ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
1337              lock->l_resource->lr_type == LDLM_EXTENT)) {
1338                 __u64 start = 0;
1339                 __u64 end = OBD_OBJECT_EOF;
1340
1341                 rc = lu_env_init(&env, LCT_DT_THREAD);
1342                 if (unlikely(rc != 0))
1343                         RETURN(rc);
1344
1345                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1346                                    tgt->lut_lsd.lsd_osd_index);
1347
1348                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1349                         obj = dt_locate(&env, tgt->lut_bottom, &fid);
1350                         if (IS_ERR(obj))
1351                                 GOTO(err_env, rc = PTR_ERR(obj));
1352
1353                         if (!dt_object_exists(obj))
1354                                 GOTO(err_put, rc = -ENOENT);
1355
1356                         start = lock->l_policy_data.l_extent.start;
1357                         end = lock->l_policy_data.l_extent.end;
1358                 }
1359
1360                 rc = tgt_sync(&env, tgt, obj, start, end);
1361                 if (rc < 0) {
1362                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1363                                "cancel: rc = %d\n",
1364                                tgt_name(tgt), PFID(&fid),
1365                                lock->l_policy_data.l_extent.start,
1366                                lock->l_policy_data.l_extent.end, rc);
1367                 }
1368 err_put:
1369                 if (obj != NULL)
1370                         lu_object_put(&env, &obj->do_lu);
1371 err_env:
1372                 lu_env_fini(&env);
1373         }
1374
1375         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1376         RETURN(rc);
1377 }
1378
1379 static struct ldlm_callback_suite tgt_dlm_cbs = {
1380         .lcs_completion = ldlm_server_completion_ast,
1381         .lcs_blocking   = tgt_blocking_ast,
1382         .lcs_glimpse    = ldlm_server_glimpse_ast
1383 };
1384
1385 int tgt_enqueue(struct tgt_session_info *tsi)
1386 {
1387         struct ptlrpc_request *req = tgt_ses_req(tsi);
1388         int rc;
1389
1390         ENTRY;
1391         /*
1392          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1393          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1394          */
1395         LASSERT(tsi->tsi_dlm_req != NULL);
1396         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1397                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1398         if (rc)
1399                 RETURN(err_serious(rc));
1400
1401         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1402         case LUT_FAIL_MDT:
1403                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1404                 break;
1405         case LUT_FAIL_OST:
1406                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1407                 break;
1408         case LUT_FAIL_MGT:
1409                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1410                 break;
1411         default:
1412                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1413                 break;
1414         }
1415         RETURN(req->rq_status);
1416 }
1417 EXPORT_SYMBOL(tgt_enqueue);
1418
1419 int tgt_convert(struct tgt_session_info *tsi)
1420 {
1421         struct ptlrpc_request *req = tgt_ses_req(tsi);
1422         int rc;
1423
1424         ENTRY;
1425         LASSERT(tsi->tsi_dlm_req);
1426         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1427         if (rc)
1428                 RETURN(err_serious(rc));
1429
1430         RETURN(req->rq_status);
1431 }
1432
1433 int tgt_bl_callback(struct tgt_session_info *tsi)
1434 {
1435         return err_serious(-EOPNOTSUPP);
1436 }
1437
1438 int tgt_cp_callback(struct tgt_session_info *tsi)
1439 {
1440         return err_serious(-EOPNOTSUPP);
1441 }
1442
1443 /* generic LDLM target handler */
1444 struct tgt_handler tgt_dlm_handlers[] = {
1445 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1446 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1447 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1448 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1449 };
1450 EXPORT_SYMBOL(tgt_dlm_handlers);
1451
1452 /*
1453  * Unified target LLOG handlers.
1454  */
1455 int tgt_llog_open(struct tgt_session_info *tsi)
1456 {
1457         int rc;
1458
1459         ENTRY;
1460
1461         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1462
1463         RETURN(rc);
1464 }
1465 EXPORT_SYMBOL(tgt_llog_open);
1466
1467 int tgt_llog_close(struct tgt_session_info *tsi)
1468 {
1469         int rc;
1470
1471         ENTRY;
1472
1473         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1474
1475         RETURN(rc);
1476 }
1477 EXPORT_SYMBOL(tgt_llog_close);
1478
1479
1480 int tgt_llog_destroy(struct tgt_session_info *tsi)
1481 {
1482         int rc;
1483
1484         ENTRY;
1485
1486         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1487
1488         RETURN(rc);
1489 }
1490
1491 int tgt_llog_read_header(struct tgt_session_info *tsi)
1492 {
1493         int rc;
1494
1495         ENTRY;
1496
1497         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1498
1499         RETURN(rc);
1500 }
1501 EXPORT_SYMBOL(tgt_llog_read_header);
1502
1503 int tgt_llog_next_block(struct tgt_session_info *tsi)
1504 {
1505         int rc;
1506
1507         ENTRY;
1508
1509         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1510
1511         RETURN(rc);
1512 }
1513 EXPORT_SYMBOL(tgt_llog_next_block);
1514
1515 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1516 {
1517         int rc;
1518
1519         ENTRY;
1520
1521         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1522
1523         RETURN(rc);
1524 }
1525 EXPORT_SYMBOL(tgt_llog_prev_block);
1526
1527 /* generic llog target handler */
1528 struct tgt_handler tgt_llog_handlers[] = {
1529 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1530 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1531 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1532 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1533 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1534 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1535 };
1536 EXPORT_SYMBOL(tgt_llog_handlers);
1537
1538 /*
1539  * sec context handlers
1540  */
1541 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1542 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1543 {
1544         return 0;
1545 }
1546
1547 struct tgt_handler tgt_sec_ctx_handlers[] = {
1548 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1549 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1550 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1551 };
1552 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1553
1554 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1555                            struct dt_device *key,
1556                            struct lfsck_request *lr,
1557                            struct thandle *th) = NULL;
1558
1559 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1560                                                 struct dt_device *,
1561                                                 struct lfsck_request *,
1562                                                 struct thandle *))
1563 {
1564         tgt_lfsck_in_notify = notify;
1565 }
1566 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1567
1568 static int (*tgt_lfsck_query)(const struct lu_env *env,
1569                               struct dt_device *key,
1570                               struct lfsck_request *req,
1571                               struct lfsck_reply *rep,
1572                               struct lfsck_query *que) = NULL;
1573
1574 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1575                                            struct dt_device *,
1576                                            struct lfsck_request *,
1577                                            struct lfsck_reply *,
1578                                            struct lfsck_query *))
1579 {
1580         tgt_lfsck_query = query;
1581 }
1582 EXPORT_SYMBOL(tgt_register_lfsck_query);
1583
1584 /* LFSCK request handlers */
1585 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1586 {
1587         const struct lu_env     *env = tsi->tsi_env;
1588         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1589         struct lfsck_request    *lr;
1590         int                      rc;
1591         ENTRY;
1592
1593         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1594         if (lr == NULL)
1595                 RETURN(-EPROTO);
1596
1597         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1598
1599         RETURN(rc);
1600 }
1601
1602 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1603 {
1604         struct lfsck_request    *request;
1605         struct lfsck_reply      *reply;
1606         int                      rc;
1607         ENTRY;
1608
1609         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1610         if (request == NULL)
1611                 RETURN(-EPROTO);
1612
1613         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1614         if (reply == NULL)
1615                 RETURN(-ENOMEM);
1616
1617         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1618                              request, reply, NULL);
1619
1620         RETURN(rc < 0 ? rc : 0);
1621 }
1622
1623 struct tgt_handler tgt_lfsck_handlers[] = {
1624 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1625 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1626 };
1627 EXPORT_SYMBOL(tgt_lfsck_handlers);
1628
1629 /*
1630  * initialize per-thread page pool (bug 5137).
1631  */
1632 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1633 {
1634         struct tgt_thread_big_cache *tbc;
1635
1636         ENTRY;
1637
1638         LASSERT(thread != NULL);
1639         LASSERT(thread->t_data == NULL);
1640
1641         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1642         if (tbc == NULL)
1643                 RETURN(-ENOMEM);
1644         thread->t_data = tbc;
1645         RETURN(0);
1646 }
1647 EXPORT_SYMBOL(tgt_io_thread_init);
1648
1649 /*
1650  * free per-thread pool created by tgt_thread_init().
1651  */
1652 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1653 {
1654         struct tgt_thread_big_cache *tbc;
1655
1656         ENTRY;
1657
1658         LASSERT(thread != NULL);
1659
1660         /*
1661          * be prepared to handle partially-initialized pools (because this is
1662          * called from ost_io_thread_init() for cleanup.
1663          */
1664         tbc = thread->t_data;
1665         if (tbc != NULL) {
1666                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1667                 thread->t_data = NULL;
1668         }
1669         EXIT;
1670 }
1671 EXPORT_SYMBOL(tgt_io_thread_done);
1672 /**
1673  * Helper function for getting server side [start, start+count] DLM lock
1674  * if asked by client.
1675  */
1676 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1677                     __u64 start, __u64 end, struct lustre_handle *lh,
1678                     int mode, __u64 *flags)
1679 {
1680         union ldlm_policy_data policy;
1681         int rc;
1682
1683         ENTRY;
1684
1685         LASSERT(lh != NULL);
1686         LASSERT(ns != NULL);
1687         LASSERT(!lustre_handle_is_used(lh));
1688
1689         policy.l_extent.gid = 0;
1690         policy.l_extent.start = start & PAGE_MASK;
1691
1692         /*
1693          * If ->o_blocks is EOF it means "lock till the end of the file".
1694          * Otherwise, it's size of an extent or hole being punched (in bytes).
1695          */
1696         if (end == OBD_OBJECT_EOF || end < start)
1697                 policy.l_extent.end = OBD_OBJECT_EOF;
1698         else
1699                 policy.l_extent.end = end | ~PAGE_MASK;
1700
1701         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1702                                     flags, ldlm_blocking_ast,
1703                                     ldlm_completion_ast, ldlm_glimpse_ast,
1704                                     NULL, 0, LVB_T_NONE, NULL, lh);
1705         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1706 }
1707 EXPORT_SYMBOL(tgt_extent_lock);
1708
1709 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1710 {
1711         LASSERT(lustre_handle_is_used(lh));
1712         ldlm_lock_decref(lh, mode);
1713 }
1714 EXPORT_SYMBOL(tgt_extent_unlock);
1715
1716 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1717                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1718                  struct lustre_handle *lh, enum ldlm_mode mode)
1719 {
1720         __u64                    flags = 0;
1721         int                      nrbufs = obj->ioo_bufcnt;
1722         int                      i;
1723
1724         ENTRY;
1725
1726         LASSERT(mode == LCK_PR || mode == LCK_PW);
1727         LASSERT(!lustre_handle_is_used(lh));
1728
1729         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1730                 RETURN(0);
1731
1732         for (i = 1; i < nrbufs; i++)
1733                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1734                         RETURN(-EFAULT);
1735
1736         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1737                                nb[nrbufs - 1].rnb_offset +
1738                                nb[nrbufs - 1].rnb_len - 1,
1739                                lh, mode, &flags));
1740 }
1741
1742 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1743                     struct lustre_handle *lh, enum ldlm_mode mode)
1744 {
1745         ENTRY;
1746
1747         LASSERT(mode == LCK_PR || mode == LCK_PW);
1748         LASSERT((obj->ioo_bufcnt > 0 &&
1749                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1750                 lustre_handle_is_used(lh));
1751
1752         if (lustre_handle_is_used(lh))
1753                 tgt_extent_unlock(lh, mode);
1754         EXIT;
1755 }
1756
1757 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1758                                struct ptlrpc_bulk_desc *desc, int opc,
1759                                cksum_type_t cksum_type)
1760 {
1761         struct cfs_crypto_hash_desc     *hdesc;
1762         unsigned int                    bufsize;
1763         int                             i, err;
1764         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1765         __u32                           cksum;
1766
1767         LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1768
1769         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1770         if (IS_ERR(hdesc)) {
1771                 CERROR("%s: unable to initialize checksum hash %s\n",
1772                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1773                 return PTR_ERR(hdesc);
1774         }
1775
1776         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1777         for (i = 0; i < desc->bd_iov_count; i++) {
1778                 /* corrupt the data before we compute the checksum, to
1779                  * simulate a client->OST data error */
1780                 if (i == 0 && opc == OST_WRITE &&
1781                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1782                         int off = BD_GET_KIOV(desc, i).kiov_offset &
1783                                 ~PAGE_MASK;
1784                         int len = BD_GET_KIOV(desc, i).kiov_len;
1785                         struct page *np = tgt_page_to_corrupt;
1786                         char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1787
1788                         if (np) {
1789                                 char *ptr2 = kmap(np) + off;
1790
1791                                 memcpy(ptr2, ptr, len);
1792                                 memcpy(ptr2, "bad3", min(4, len));
1793                                 kunmap(np);
1794                                 BD_GET_KIOV(desc, i).kiov_page = np;
1795                         } else {
1796                                 CERROR("%s: can't alloc page for corruption\n",
1797                                        tgt_name(tgt));
1798                         }
1799                 }
1800                 cfs_crypto_hash_update_page(hdesc,
1801                                   BD_GET_KIOV(desc, i).kiov_page,
1802                                   BD_GET_KIOV(desc, i).kiov_offset &
1803                                         ~PAGE_MASK,
1804                                   BD_GET_KIOV(desc, i).kiov_len);
1805
1806                  /* corrupt the data after we compute the checksum, to
1807                  * simulate an OST->client data error */
1808                 if (i == 0 && opc == OST_READ &&
1809                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1810                         int off = BD_GET_KIOV(desc, i).kiov_offset
1811                           & ~PAGE_MASK;
1812                         int len = BD_GET_KIOV(desc, i).kiov_len;
1813                         struct page *np = tgt_page_to_corrupt;
1814                         char *ptr =
1815                           kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1816
1817                         if (np) {
1818                                 char *ptr2 = kmap(np) + off;
1819
1820                                 memcpy(ptr2, ptr, len);
1821                                 memcpy(ptr2, "bad4", min(4, len));
1822                                 kunmap(np);
1823                                 BD_GET_KIOV(desc, i).kiov_page = np;
1824                         } else {
1825                                 CERROR("%s: can't alloc page for corruption\n",
1826                                        tgt_name(tgt));
1827                         }
1828                 }
1829         }
1830
1831         bufsize = sizeof(cksum);
1832         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1833
1834         return cksum;
1835 }
1836
1837 int tgt_brw_read(struct tgt_session_info *tsi)
1838 {
1839         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1840         struct ptlrpc_bulk_desc *desc = NULL;
1841         struct obd_export       *exp = tsi->tsi_exp;
1842         struct niobuf_remote    *remote_nb;
1843         struct niobuf_local     *local_nb;
1844         struct obd_ioobj        *ioo;
1845         struct ost_body         *body, *repbody;
1846         struct l_wait_info       lwi;
1847         struct lustre_handle     lockh = { 0 };
1848         int                      npages, nob = 0, rc, i, no_reply = 0;
1849         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1850
1851         ENTRY;
1852
1853         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1854                 CERROR("%s: deny read request from %s to portal %u\n",
1855                        tgt_name(tsi->tsi_tgt),
1856                        obd_export_nid2str(req->rq_export),
1857                        ptlrpc_req2svc(req)->srv_req_portal);
1858                 RETURN(-EPROTO);
1859         }
1860
1861         req->rq_bulk_read = 1;
1862
1863         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1864                 RETURN(-EIO);
1865
1866         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1867                          cfs_fail_val : (obd_timeout + 1) / 4);
1868
1869         /* Check if there is eviction in progress, and if so, wait for it to
1870          * finish */
1871         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1872                 /* We do not care how long it takes */
1873                 lwi = LWI_INTR(NULL, NULL);
1874                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1875                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1876                          &lwi);
1877         }
1878
1879         /* There must be big cache in current thread to process this request
1880          * if it is NULL then something went wrong and it wasn't allocated,
1881          * report -ENOMEM in that case */
1882         if (tbc == NULL)
1883                 RETURN(-ENOMEM);
1884
1885         body = tsi->tsi_ost_body;
1886         LASSERT(body != NULL);
1887
1888         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1889         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1890
1891         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1892         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1893
1894         local_nb = tbc->local;
1895
1896         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1897                           remote_nb, &lockh, LCK_PR);
1898         if (rc != 0)
1899                 RETURN(rc);
1900
1901         /*
1902          * If getting the lock took more time than
1903          * client was willing to wait, drop it. b=11330
1904          */
1905         if (cfs_time_current_sec() > req->rq_deadline ||
1906             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1907                 no_reply = 1;
1908                 CERROR("Dropping timed-out read from %s because locking"
1909                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1910                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1911                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1912                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1913                 GOTO(out_lock, rc = -ETIMEDOUT);
1914         }
1915
1916         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1917         repbody->oa = body->oa;
1918
1919         npages = PTLRPC_MAX_BRW_PAGES;
1920         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1921                         ioo, remote_nb, &npages, local_nb);
1922         if (rc != 0)
1923                 GOTO(out_lock, rc);
1924
1925         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1926                                     PTLRPC_BULK_PUT_SOURCE |
1927                                         PTLRPC_BULK_BUF_KIOV,
1928                                     OST_BULK_PORTAL,
1929                                     &ptlrpc_bulk_kiov_nopin_ops);
1930         if (desc == NULL)
1931                 GOTO(out_commitrw, rc = -ENOMEM);
1932
1933         nob = 0;
1934         for (i = 0; i < npages; i++) {
1935                 int page_rc = local_nb[i].lnb_rc;
1936
1937                 if (page_rc < 0) {
1938                         rc = page_rc;
1939                         break;
1940                 }
1941
1942                 nob += page_rc;
1943                 if (page_rc != 0) { /* some data! */
1944                         LASSERT(local_nb[i].lnb_page != NULL);
1945                         desc->bd_frag_ops->add_kiov_frag
1946                           (desc, local_nb[i].lnb_page,
1947                            local_nb[i].lnb_page_offset,
1948                            page_rc);
1949                 }
1950
1951                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1952                         /* All subsequent pages should be 0 */
1953                         while (++i < npages)
1954                                 LASSERT(local_nb[i].lnb_rc == 0);
1955                         break;
1956                 }
1957         }
1958         if (OBD_FAIL_CHECK(OBD_FAIL_OST_READ_SIZE) &&
1959             nob != cfs_fail_val)
1960                 rc = -E2BIG;
1961
1962         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1963                 cksum_type_t cksum_type =
1964                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1965                                           body->oa.o_flags : 0);
1966                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1967                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1968                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1969                                                         OST_READ, cksum_type);
1970                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1971                        repbody->oa.o_cksum);
1972         } else {
1973                 repbody->oa.o_valid = 0;
1974         }
1975         /* We're finishing using body->oa as an input variable */
1976
1977         /* Check if client was evicted while we were doing i/o before touching
1978          * network */
1979         if (likely(rc == 0 &&
1980                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1981                 rc = target_bulk_io(exp, desc, &lwi);
1982                 no_reply = rc != 0;
1983         }
1984
1985 out_commitrw:
1986         /* Must commit after prep above in all cases */
1987         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1988                           remote_nb, npages, local_nb, rc);
1989         if (rc == 0)
1990                 tgt_drop_id(exp, &repbody->oa);
1991 out_lock:
1992         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1993
1994         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1995                 ptlrpc_free_bulk(desc);
1996
1997         LASSERT(rc <= 0);
1998         if (rc == 0) {
1999                 rc = nob;
2000                 ptlrpc_lprocfs_brw(req, nob);
2001         } else if (no_reply) {
2002                 req->rq_no_reply = 1;
2003                 /* reply out callback would free */
2004                 ptlrpc_req_drop_rs(req);
2005                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
2006                               "client will retry: rc %d\n",
2007                               exp->exp_obd->obd_name,
2008                               obd_uuid2str(&exp->exp_client_uuid),
2009                               obd_export_nid2str(exp), rc);
2010         }
2011         /* send a bulk after reply to simulate a network delay or reordering
2012          * by a router */
2013         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
2014                 wait_queue_head_t        waitq;
2015                 struct l_wait_info       lwi1;
2016
2017                 CDEBUG(D_INFO, "reorder BULK\n");
2018                 init_waitqueue_head(&waitq);
2019
2020                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
2021                 l_wait_event(waitq, 0, &lwi1);
2022                 target_bulk_io(exp, desc, &lwi);
2023                 ptlrpc_free_bulk(desc);
2024         }
2025
2026         RETURN(rc);
2027 }
2028 EXPORT_SYMBOL(tgt_brw_read);
2029
2030 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
2031                               struct ptlrpc_bulk_desc *desc,
2032                               struct niobuf_local *local_nb, int npages,
2033                               u32 client_cksum, u32 server_cksum,
2034                               bool mmap)
2035 {
2036         struct obd_export *exp = req->rq_export;
2037         struct ost_body *body;
2038         char *router = "";
2039         char *via = "";
2040
2041         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2042         LASSERT(body != NULL);
2043
2044         if (req->rq_peer.nid != desc->bd_sender) {
2045                 via = " via ";
2046                 router = libcfs_nid2str(desc->bd_sender);
2047         }
2048
2049         if (mmap) {
2050                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
2051                              client_cksum, server_cksum);
2052                 return;
2053         }
2054
2055         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
2056                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
2057                            "]: client csum %x, server csum %x\n",
2058                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
2059                            via, router,
2060                            body->oa.o_valid & OBD_MD_FLFID ?
2061                            body->oa.o_parent_seq : (__u64)0,
2062                            body->oa.o_valid & OBD_MD_FLFID ?
2063                            body->oa.o_parent_oid : 0,
2064                            body->oa.o_valid & OBD_MD_FLFID ?
2065                            body->oa.o_parent_ver : 0,
2066                            POSTID(&body->oa.o_oi),
2067                            local_nb[0].lnb_file_offset,
2068                            local_nb[npages-1].lnb_file_offset +
2069                            local_nb[npages - 1].lnb_len - 1,
2070                            client_cksum, server_cksum);
2071 }
2072
2073 int tgt_brw_write(struct tgt_session_info *tsi)
2074 {
2075         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2076         struct ptlrpc_bulk_desc *desc = NULL;
2077         struct obd_export       *exp = req->rq_export;
2078         struct niobuf_remote    *remote_nb;
2079         struct niobuf_local     *local_nb;
2080         struct obd_ioobj        *ioo;
2081         struct ost_body         *body, *repbody;
2082         struct l_wait_info       lwi;
2083         struct lustre_handle     lockh = {0};
2084         __u32                   *rcs;
2085         int                      objcount, niocount, npages;
2086         int                      rc, i, j;
2087         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
2088         bool                     no_reply = false, mmap;
2089         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2090
2091         ENTRY;
2092
2093         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2094                 CERROR("%s: deny write request from %s to portal %u\n",
2095                        tgt_name(tsi->tsi_tgt),
2096                        obd_export_nid2str(req->rq_export),
2097                        ptlrpc_req2svc(req)->srv_req_portal);
2098                 RETURN(err_serious(-EPROTO));
2099         }
2100
2101         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2102                 RETURN(err_serious(-ENOSPC));
2103         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2104                 RETURN(err_serious(-EROFS));
2105
2106         req->rq_bulk_write = 1;
2107
2108         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2109                 RETURN(err_serious(-EIO));
2110         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2111                 RETURN(err_serious(-EFAULT));
2112
2113         /* pause before transaction has been started */
2114         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2115                          cfs_fail_val : (obd_timeout + 1) / 4);
2116
2117         /* There must be big cache in current thread to process this request
2118          * if it is NULL then something went wrong and it wasn't allocated,
2119          * report -ENOMEM in that case */
2120         if (tbc == NULL)
2121                 RETURN(-ENOMEM);
2122
2123         body = tsi->tsi_ost_body;
2124         LASSERT(body != NULL);
2125
2126         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2127         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2128
2129         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2130                                         RCL_CLIENT) / sizeof(*ioo);
2131
2132         for (niocount = i = 0; i < objcount; i++)
2133                 niocount += ioo[i].ioo_bufcnt;
2134
2135         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2136         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2137         if (niocount != req_capsule_get_size(&req->rq_pill,
2138                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2139                         sizeof(*remote_nb))
2140                 RETURN(err_serious(-EPROTO));
2141
2142         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2143             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2144                 memory_pressure_set();
2145
2146         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2147                              niocount * sizeof(*rcs));
2148         rc = req_capsule_server_pack(&req->rq_pill);
2149         if (rc != 0)
2150                 GOTO(out, rc = err_serious(rc));
2151
2152         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2153         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2154
2155         local_nb = tbc->local;
2156
2157         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2158                           remote_nb, &lockh, LCK_PW);
2159         if (rc != 0)
2160                 GOTO(out, rc);
2161
2162         /*
2163          * If getting the lock took more time than
2164          * client was willing to wait, drop it. b=11330
2165          */
2166         if (cfs_time_current_sec() > req->rq_deadline ||
2167             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2168                 no_reply = true;
2169                 CERROR("%s: Dropping timed-out write from %s because locking "
2170                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
2171                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2172                        POSTID(&ioo->ioo_oid),
2173                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2174                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2175                 GOTO(out_lock, rc = -ETIMEDOUT);
2176         }
2177
2178         /* Because we already sync grant info with client when reconnect,
2179          * grant info will be cleared for resent req, then fed_grant and
2180          * total_grant will not be modified in following preprw_write */
2181         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2182                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2183                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2184         }
2185
2186         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2187         if (repbody == NULL)
2188                 GOTO(out_lock, rc = -ENOMEM);
2189         repbody->oa = body->oa;
2190
2191         npages = PTLRPC_MAX_BRW_PAGES;
2192         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2193                         objcount, ioo, remote_nb, &npages, local_nb);
2194         if (rc < 0)
2195                 GOTO(out_lock, rc);
2196
2197         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2198                                     PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2199                                     OST_BULK_PORTAL,
2200                                     &ptlrpc_bulk_kiov_nopin_ops);
2201         if (desc == NULL)
2202                 GOTO(skip_transfer, rc = -ENOMEM);
2203
2204         /* NB Having prepped, we must commit... */
2205         for (i = 0; i < npages; i++)
2206                 desc->bd_frag_ops->add_kiov_frag(desc,
2207                                                  local_nb[i].lnb_page,
2208                                                  local_nb[i].lnb_page_offset,
2209                                                  local_nb[i].lnb_len);
2210
2211         rc = sptlrpc_svc_prep_bulk(req, desc);
2212         if (rc != 0)
2213                 GOTO(skip_transfer, rc);
2214
2215         rc = target_bulk_io(exp, desc, &lwi);
2216         no_reply = rc != 0;
2217
2218 skip_transfer:
2219         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2220                 static int cksum_counter;
2221
2222                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2223                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2224
2225                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2226                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2227                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2228                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2229                                                         OST_WRITE, cksum_type);
2230                 cksum_counter++;
2231
2232                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2233                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2234                                 body->oa.o_flags & OBD_FL_MMAP);
2235
2236                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2237                                           body->oa.o_cksum,
2238                                           repbody->oa.o_cksum, mmap);
2239                         cksum_counter = 0;
2240                 } else if ((cksum_counter & (-cksum_counter)) ==
2241                            cksum_counter) {
2242                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2243                                cksum_counter, libcfs_id2str(req->rq_peer),
2244                                repbody->oa.o_cksum);
2245                 }
2246         }
2247
2248         /* Must commit after prep above in all cases */
2249         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2250                           objcount, ioo, remote_nb, npages, local_nb, rc);
2251         if (rc == -ENOTCONN)
2252                 /* quota acquire process has been given up because
2253                  * either the client has been evicted or the client
2254                  * has timed out the request already */
2255                 no_reply = true;
2256
2257         /*
2258          * Disable sending mtime back to the client. If the client locked the
2259          * whole object, then it has already updated the mtime on its side,
2260          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2261          */
2262         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2263
2264         if (rc == 0) {
2265                 int nob = 0;
2266
2267                 /* set per-requested niobuf return codes */
2268                 for (i = j = 0; i < niocount; i++) {
2269                         int len = remote_nb[i].rnb_len;
2270
2271                         nob += len;
2272                         rcs[i] = 0;
2273                         do {
2274                                 LASSERT(j < npages);
2275                                 if (local_nb[j].lnb_rc < 0)
2276                                         rcs[i] = local_nb[j].lnb_rc;
2277                                 len -= local_nb[j].lnb_len;
2278                                 j++;
2279                         } while (len > 0);
2280                         LASSERT(len == 0);
2281                 }
2282                 LASSERT(j == npages);
2283                 ptlrpc_lprocfs_brw(req, nob);
2284
2285                 tgt_drop_id(exp, &repbody->oa);
2286         }
2287 out_lock:
2288         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2289         if (desc)
2290                 ptlrpc_free_bulk(desc);
2291 out:
2292         if (no_reply) {
2293                 req->rq_no_reply = 1;
2294                 /* reply out callback would free */
2295                 ptlrpc_req_drop_rs(req);
2296                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2297                               "client will retry: rc %d\n",
2298                               exp->exp_obd->obd_name,
2299                               obd_uuid2str(&exp->exp_client_uuid),
2300                               obd_export_nid2str(exp), rc);
2301         }
2302         memory_pressure_clr();
2303         RETURN(rc);
2304 }
2305 EXPORT_SYMBOL(tgt_brw_write);
2306
2307 /* Check if request can be reconstructed from saved reply data
2308  * A copy of the reply data is returned in @trd if the pointer is not NULL
2309  */
2310 bool req_can_reconstruct(struct ptlrpc_request *req,
2311                          struct tg_reply_data *trd)
2312 {
2313         struct tg_export_data *ted = &req->rq_export->exp_target_data;
2314         struct lsd_client_data *lcd = ted->ted_lcd;
2315         bool found;
2316
2317         if (tgt_is_multimodrpcs_client(req->rq_export))
2318                 return tgt_lookup_reply(req, trd);
2319
2320         mutex_lock(&ted->ted_lcd_lock);
2321         found = req->rq_xid == lcd->lcd_last_xid ||
2322                 req->rq_xid == lcd->lcd_last_close_xid;
2323
2324         if (found && trd != NULL) {
2325                 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2326                         trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2327                         trd->trd_reply.lrd_transno =
2328                                                 lcd->lcd_last_close_transno;
2329                         trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2330                 } else {
2331                         trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2332                         trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2333                         trd->trd_reply.lrd_result = lcd->lcd_last_result;
2334                         trd->trd_reply.lrd_data = lcd->lcd_last_data;
2335                         trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2336                         trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2337                         trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2338                         trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2339                 }
2340         }
2341         mutex_unlock(&ted->ted_lcd_lock);
2342
2343         return found;
2344 }
2345 EXPORT_SYMBOL(req_can_reconstruct);
2346