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                         req->rq_status = -EPROTO;
 639                         RETURN(ptlrpc_error(req));
 640                 }
 641         }
 642
 643         /* try to release in-memory reply data */
 644         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 645                 tgt_handle_received_xid(req->rq_export,
 646                                 lustre_msg_get_last_xid(req->rq_reqmsg));
 647                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 648                       (MSG_RESENT | MSG_REPLAY)))
 649                         tgt_handle_tag(req->rq_export,
 650                                        lustre_msg_get_tag(req->rq_reqmsg));
 651         }
 652         RETURN(0);
 653 }
 654
 655 int tgt_request_handle(struct ptlrpc_request *req)
 656 {
 657         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 658
 659         struct lustre_msg       *msg = req->rq_reqmsg;
 660         struct tgt_handler      *h;
 661         struct lu_target        *tgt;
 662         int                      request_fail_id = 0;
 663         __u32                    opc = lustre_msg_get_opc(msg);
 664         struct obd_device       *obd;
 665         int                      rc;
 666         bool                     is_connect = false;
 667         ENTRY;
 668
 669         /* Refill the context, to make sure all thread keys are allocated */
 670         lu_env_refill(req->rq_svc_thread->t_env);
 671
 672         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 673         tsi->tsi_pill = &req->rq_pill;
 674         tsi->tsi_env = req->rq_svc_thread->t_env;
 675
 676         /* if request has export then get handlers slice from corresponding
 677          * target, otherwise that should be connect operation */
 678         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 679             opc == MGS_CONNECT) {
 680                 is_connect = true;
 681                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 682                 rc = target_handle_connect(req);
 683                 if (rc != 0) {
 684                         rc = ptlrpc_error(req);
 685                         GOTO(out, rc);
 686                 }
 687                 /* recovery-small test 18c asks to drop connect reply */
 688                 if (unlikely(opc == OST_CONNECT &&
 689                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 690                         GOTO(out, rc = 0);
 691         }
 692
 693         if (unlikely(!class_connected_export(req->rq_export))) {
 694                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
 695                     opc == SEC_CTX_FINI) {
 696                         /* sec context initialization has to be handled
 697                          * by hand in tgt_handle_request0() */
 698                         tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
 699                         h = NULL;
 700                         GOTO(handle_recov, rc = 0);
 701                 }
 702                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 703                        opc, libcfs_id2str(req->rq_peer));
 704                 req->rq_status = -ENOTCONN;
 705                 rc = ptlrpc_error(req);
 706                 GOTO(out, rc);
 707         }
 708
 709         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 710         tsi->tsi_exp = req->rq_export;
 711         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 712                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 713         else
 714                 tsi->tsi_jobid = NULL;
 715
 716         if (tgt == NULL) {
 717                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 718                           class_exp2obd(req->rq_export)->obd_name);
 719                 req->rq_status = -EINVAL;
 720                 rc = ptlrpc_error(req);
 721                 GOTO(out, rc);
 722         }
 723
 724         /* Skip last_xid processing for the recovery thread, otherwise, the
 725          * last_xid on same request could be processed twice: first time when
 726          * processing the incoming request, second time when the request is
 727          * being processed by recovery thread. */
 728         obd = class_exp2obd(req->rq_export);
 729         if (is_connect) {
 730                 /* reset the exp_last_xid on each connection. */
 731                 req->rq_export->exp_last_xid = 0;
 732         } else if (obd->obd_recovery_data.trd_processing_task !=
 733                    current_pid()) {
 734                 rc = process_req_last_xid(req);
 735                 if (rc)
 736                         GOTO(out, rc);
 737         }
 738
 739         request_fail_id = tgt->lut_request_fail_id;
 740         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 741
 742         h = tgt_handler_find_check(req);
 743         if (IS_ERR(h)) {
 744                 req->rq_status = PTR_ERR(h);
 745                 rc = ptlrpc_error(req);
 746                 GOTO(out, rc);
 747         }
 748
 749         LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 750                  h->th_opc, opc);
 751
 752         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 753                 GOTO(out, rc = 0);
 754
 755         rc = lustre_msg_check_version(msg, h->th_version);
 756         if (unlikely(rc)) {
 757                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 758                           " %08x, expecting %08x\n", tgt_name(tgt),
 759                           lustre_msg_get_version(msg), h->th_version);
 760                 req->rq_status = -EINVAL;
 761                 rc = ptlrpc_error(req);
 762                 GOTO(out, rc);
 763         }
 764
 765 handle_recov:
 766         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 767         if (likely(rc == 1)) {
 768                 rc = tgt_handle_request0(tsi, h, req);
 769                 if (rc)
 770                         GOTO(out, rc);
 771         }
 772         EXIT;
 773 out:
 774         req_capsule_fini(tsi->tsi_pill);
 775         if (tsi->tsi_corpus != NULL) {
 776                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 777                 tsi->tsi_corpus = NULL;
 778         }
 779         return rc;
 780 }
 781 EXPORT_SYMBOL(tgt_request_handle);
 782
 783 /** Assign high priority operations to the request if needed. */
 784 int tgt_hpreq_handler(struct ptlrpc_request *req)
 785 {
 786         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 787         struct tgt_handler      *h;
 788         int                      rc;
 789
 790         ENTRY;
 791
 792         if (req->rq_export == NULL)
 793                 RETURN(0);
 794
 795         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 796         tsi->tsi_pill = &req->rq_pill;
 797         tsi->tsi_env = req->rq_svc_thread->t_env;
 798         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 799         tsi->tsi_exp = req->rq_export;
 800
 801         h = tgt_handler_find_check(req);
 802         if (IS_ERR(h)) {
 803                 rc = PTR_ERR(h);
 804                 RETURN(rc);
 805         }
 806
 807         rc = tgt_request_preprocess(tsi, h, req);
 808         if (unlikely(rc != 0))
 809                 RETURN(rc);
 810
 811         if (h->th_hp != NULL)
 812                 h->th_hp(tsi);
 813         RETURN(0);
 814 }
 815 EXPORT_SYMBOL(tgt_hpreq_handler);
 816
 817 void tgt_counter_incr(struct obd_export *exp, int opcode)
 818 {
 819         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 820         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 821                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 822 }
 823 EXPORT_SYMBOL(tgt_counter_incr);
 824
 825 /*
 826  * Unified target generic handlers.
 827  */
 828
 829 /*
 830  * Security functions
 831  */
 832 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 833 {
 834         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 835                                       OBD_CONNECT_RMT_CLIENT_FORCE);
 836 }
 837
 838 static int tgt_init_sec_level(struct ptlrpc_request *req)
 839 {
 840         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 841         char                    *client;
 842         struct obd_connect_data *data, *reply;
 843         int                      rc = 0;
 844         bool                     remote;
 845         ENTRY;
 846
 847         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 848         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 849         if (data == NULL || reply == NULL)
 850                 RETURN(-EFAULT);
 851
 852         /* connection from MDT is always trusted */
 853         if (req->rq_auth_usr_mdt) {
 854                 tgt_init_sec_none(reply);
 855                 RETURN(0);
 856         }
 857
 858         if (unlikely(tgt == NULL)) {
 859                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 860                           class_exp2obd(req->rq_export)->obd_name);
 861                 RETURN(-EINVAL);
 862         }
 863
 864         client = libcfs_nid2str(req->rq_peer.nid);
 865         /* no GSS support case */
 866         if (!req->rq_auth_gss) {
 867                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 868                         CWARN("client %s -> target %s does not use GSS, "
 869                               "can not run under security level %d.\n",
 870                               client, tgt_name(tgt), tgt->lut_sec_level);
 871                         RETURN(-EACCES);
 872                 } else {
 873                         tgt_init_sec_none(reply);
 874                         RETURN(0);
 875                 }
 876         }
 877
 878         /* old version case */
 879         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 880                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 881                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 882                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 883                         CWARN("client %s -> target %s uses old version, "
 884                               "can not run under security level %d.\n",
 885                               client, tgt_name(tgt), tgt->lut_sec_level);
 886                         RETURN(-EACCES);
 887                 } else {
 888                         CWARN("client %s -> target %s uses old version, "
 889                               "run under security level %d.\n",
 890                               client, tgt_name(tgt), tgt->lut_sec_level);
 891                         tgt_init_sec_none(reply);
 892                         RETURN(0);
 893                 }
 894         }
 895
 896         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 897         if (remote) {
 898                 if (!req->rq_auth_remote)
 899                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 900                                "asked to be remote.\n", client, tgt_name(tgt));
 901         } else if (req->rq_auth_remote) {
 902                 remote = true;
 903                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 904                        "as remote by default.\n", client, tgt_name(tgt));
 905         }
 906
 907         if (remote == 0) {
 908                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 909                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 910                                "authenticated!\n", client, tgt_name(tgt));
 911                         RETURN(-EACCES);
 912                 }
 913         }
 914
 915
 916         switch (tgt->lut_sec_level) {
 917         case LUSTRE_SEC_NONE:
 918                 if (remote) {
 919                         CDEBUG(D_SEC,
 920                                "client %s -> target %s is set as remote, "
 921                                "can not run under security level %d.\n",
 922                                client, tgt_name(tgt), tgt->lut_sec_level);
 923                         RETURN(-EACCES);
 924                 }
 925                 tgt_init_sec_none(reply);
 926                 break;
 927         case LUSTRE_SEC_REMOTE:
 928                 if (!remote)
 929                         tgt_init_sec_none(reply);
 930                 break;
 931         case LUSTRE_SEC_ALL:
 932                 if (remote)
 933                         break;
 934                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 935                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 936                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 937                 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 938                 break;
 939         default:
 940                 RETURN(-EINVAL);
 941         }
 942
 943         RETURN(rc);
 944 }
 945
 946 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 947 {
 948         struct lu_target        *tgt = class_exp2tgt(exp);
 949         struct sptlrpc_flavor    flvr;
 950         int                      rc = 0;
 951
 952         LASSERT(tgt);
 953         LASSERT(tgt->lut_obd);
 954         LASSERT(tgt->lut_slice);
 955
 956         /* always allow ECHO client */
 957         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 958                             LUSTRE_ECHO_NAME) == 0)) {
 959                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 960                 return 0;
 961         }
 962
 963         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 964                 read_lock(&tgt->lut_sptlrpc_lock);
 965                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 966                                              req->rq_sp_from,
 967                                              req->rq_peer.nid,
 968                                              &flvr);
 969                 read_unlock(&tgt->lut_sptlrpc_lock);
 970
 971                 spin_lock(&exp->exp_lock);
 972                 exp->exp_sp_peer = req->rq_sp_from;
 973                 exp->exp_flvr = flvr;
 974
 975                 /* when on mgs, if no restriction is set, or if client
 976                  * is loopback, allow any flavor */
 977                 if ((strcmp(exp->exp_obd->obd_type->typ_name,
 978                            LUSTRE_MGS_NAME) == 0) &&
 979                      (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
 980                       LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
 981                       == LOLND))
 982                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 983
 984                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 985                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 986                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 987                                "expect %x\n", tgt_name(tgt),
 988                                req->rq_flvr.sf_rpc,
 989                                libcfs_nid2str(req->rq_peer.nid),
 990                                exp->exp_flvr.sf_rpc);
 991                         rc = -EACCES;
 992                 }
 993                 spin_unlock(&exp->exp_lock);
 994         } else {
 995                 if (exp->exp_sp_peer != req->rq_sp_from) {
 996                         CERROR("%s: RPC source %s doesn't match %s\n",
 997                                tgt_name(tgt),
 998                                sptlrpc_part2name(req->rq_sp_from),
 999                                sptlrpc_part2name(exp->exp_sp_peer));
1000                         rc = -EACCES;
1001                 } else {
1002                         rc = sptlrpc_target_export_check(exp, req);
1003                 }
1004         }
1005
1006         return rc;
1007 }
1008
1009 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
1010 {
1011         struct sptlrpc_rule_set  tmp_rset;
1012         int                      rc;
1013
1014         if (unlikely(tgt == NULL)) {
1015                 CERROR("No target passed");
1016                 return -EINVAL;
1017         }
1018
1019         sptlrpc_rule_set_init(&tmp_rset);
1020         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
1021         if (rc) {
1022                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
1023                        tgt_name(tgt), rc);
1024                 return rc;
1025         }
1026
1027         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
1028
1029         write_lock(&tgt->lut_sptlrpc_lock);
1030         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
1031         tgt->lut_sptlrpc_rset = tmp_rset;
1032         write_unlock(&tgt->lut_sptlrpc_lock);
1033
1034         return 0;
1035 }
1036 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
1037
1038 int tgt_connect(struct tgt_session_info *tsi)
1039 {
1040         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1041         struct obd_connect_data *reply;
1042         int                      rc;
1043
1044         ENTRY;
1045
1046         rc = tgt_init_sec_level(req);
1047         if (rc != 0)
1048                 GOTO(out, rc);
1049
1050         /* XXX: better to call this check right after getting new export but
1051          * before last_rcvd slot allocation to avoid server load upon insecure
1052          * connects. This is to be fixed after unifiyng all targets.
1053          */
1054         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1055         if (rc)
1056                 GOTO(out, rc);
1057
1058         /* To avoid exposing partially initialized connection flags, changes up
1059          * to this point have been staged in reply->ocd_connect_flags. Now that
1060          * connection handling has completed successfully, atomically update
1061          * the connect flags in the shared export data structure. LU-1623 */
1062         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1063         spin_lock(&tsi->tsi_exp->exp_lock);
1064         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1065         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1066         spin_unlock(&tsi->tsi_exp->exp_lock);
1067
1068         RETURN(0);
1069 out:
1070         obd_disconnect(class_export_get(tsi->tsi_exp));
1071         return rc;
1072 }
1073 EXPORT_SYMBOL(tgt_connect);
1074
1075 int tgt_disconnect(struct tgt_session_info *tsi)
1076 {
1077         int rc;
1078
1079         ENTRY;
1080
1081         rc = target_handle_disconnect(tgt_ses_req(tsi));
1082         if (rc)
1083                 RETURN(err_serious(rc));
1084
1085         RETURN(rc);
1086 }
1087 EXPORT_SYMBOL(tgt_disconnect);
1088
1089 /*
1090  * Unified target OBD handlers
1091  */
1092 int tgt_obd_ping(struct tgt_session_info *tsi)
1093 {
1094         int rc;
1095
1096         ENTRY;
1097
1098         rc = target_handle_ping(tgt_ses_req(tsi));
1099         if (rc)
1100                 RETURN(err_serious(rc));
1101
1102         RETURN(rc);
1103 }
1104 EXPORT_SYMBOL(tgt_obd_ping);
1105
1106 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1107 {
1108         return err_serious(-EOPNOTSUPP);
1109 }
1110
1111 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
1112 {
1113         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1114         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1115         struct obd_export       *exp = req->rq_export;
1116         struct ptlrpc_bulk_desc *desc;
1117         struct l_wait_info      *lwi = &tti->tti_u.update.tti_wait_info;
1118         int                      i;
1119         int                      rc;
1120
1121         ENTRY;
1122
1123         desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
1124                                   PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
1125                                     MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1126         if (desc == NULL)
1127                 RETURN(-ENOMEM);
1128
1129         for (i = 0; i < rdbuf->rb_nbufs; i++)
1130                 desc->bd_frag_ops->add_iov_frag(desc,
1131                                         rdbuf->rb_bufs[i].lb_buf,
1132                                         rdbuf->rb_bufs[i].lb_len);
1133
1134         rc = target_bulk_io(exp, desc, lwi);
1135         ptlrpc_free_bulk(desc);
1136         RETURN(rc);
1137 }
1138 EXPORT_SYMBOL(tgt_send_buffer);
1139
1140 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1141 {
1142         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1143         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1144         struct obd_export       *exp = req->rq_export;
1145         struct ptlrpc_bulk_desc *desc;
1146         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1147         int                      tmpcount;
1148         int                      tmpsize;
1149         int                      i;
1150         int                      rc;
1151
1152         ENTRY;
1153
1154         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1155                                     PTLRPC_BULK_PUT_SOURCE |
1156                                         PTLRPC_BULK_BUF_KIOV,
1157                                     MDS_BULK_PORTAL,
1158                                     &ptlrpc_bulk_kiov_pin_ops);
1159         if (desc == NULL)
1160                 RETURN(-ENOMEM);
1161
1162         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1163                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1164                  * which is rdpg->rp_count */
1165                 nob = rdpg->rp_count;
1166
1167         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1168              i++, tmpcount -= tmpsize) {
1169                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1170                 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1171                                                  tmpsize);
1172         }
1173
1174         LASSERT(desc->bd_nob == nob);
1175         rc = target_bulk_io(exp, desc, lwi);
1176         ptlrpc_free_bulk(desc);
1177         RETURN(rc);
1178 }
1179 EXPORT_SYMBOL(tgt_sendpage);
1180
1181 /*
1182  * OBD_IDX_READ handler
1183  */
1184 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1185 {
1186         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1187         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1188         struct idx_info         *req_ii, *rep_ii;
1189         int                      rc, i;
1190
1191         ENTRY;
1192
1193         memset(rdpg, 0, sizeof(*rdpg));
1194         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1195
1196         /* extract idx_info buffer from request & reply */
1197         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1198         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1199                 RETURN(err_serious(-EPROTO));
1200
1201         rc = req_capsule_server_pack(tsi->tsi_pill);
1202         if (rc)
1203                 RETURN(err_serious(rc));
1204
1205         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1206         if (rep_ii == NULL)
1207                 RETURN(err_serious(-EFAULT));
1208         rep_ii->ii_magic = IDX_INFO_MAGIC;
1209
1210         /* extract hash to start with */
1211         rdpg->rp_hash = req_ii->ii_hash_start;
1212
1213         /* extract requested attributes */
1214         rdpg->rp_attrs = req_ii->ii_attrs;
1215
1216         /* check that fid packed in request is valid and supported */
1217         if (!fid_is_sane(&req_ii->ii_fid))
1218                 RETURN(-EINVAL);
1219         rep_ii->ii_fid = req_ii->ii_fid;
1220
1221         /* copy flags */
1222         rep_ii->ii_flags = req_ii->ii_flags;
1223
1224         /* compute number of pages to allocate, ii_count is the number of 4KB
1225          * containers */
1226         if (req_ii->ii_count <= 0)
1227                 GOTO(out, rc = -EFAULT);
1228         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1229                                exp_max_brw_size(tsi->tsi_exp));
1230         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1231
1232         /* allocate pages to store the containers */
1233         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1234         if (rdpg->rp_pages == NULL)
1235                 GOTO(out, rc = -ENOMEM);
1236         for (i = 0; i < rdpg->rp_npages; i++) {
1237                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1238                 if (rdpg->rp_pages[i] == NULL)
1239                         GOTO(out, rc = -ENOMEM);
1240         }
1241
1242         /* populate pages with key/record pairs */
1243         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1244         if (rc < 0)
1245                 GOTO(out, rc);
1246
1247         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1248                  "asked %d > %d\n", rc, rdpg->rp_count);
1249
1250         /* send pages to client */
1251         rc = tgt_sendpage(tsi, rdpg, rc);
1252         if (rc)
1253                 GOTO(out, rc);
1254         EXIT;
1255 out:
1256         if (rdpg->rp_pages) {
1257                 for (i = 0; i < rdpg->rp_npages; i++)
1258                         if (rdpg->rp_pages[i])
1259                                 __free_page(rdpg->rp_pages[i]);
1260                 OBD_FREE(rdpg->rp_pages,
1261                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1262         }
1263         return rc;
1264 }
1265
1266 struct tgt_handler tgt_obd_handlers[] = {
1267 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1268 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1269 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1270 };
1271 EXPORT_SYMBOL(tgt_obd_handlers);
1272
1273 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1274              struct dt_object *obj, __u64 start, __u64 end)
1275 {
1276         int rc = 0;
1277
1278         ENTRY;
1279
1280         /* if no objid is specified, it means "sync whole filesystem" */
1281         if (obj == NULL) {
1282                 rc = dt_sync(env, tgt->lut_bottom);
1283         } else if (dt_version_get(env, obj) >
1284                    tgt->lut_obd->obd_last_committed) {
1285                 rc = dt_object_sync(env, obj, start, end);
1286         }
1287         atomic_inc(&tgt->lut_sync_count);
1288
1289         RETURN(rc);
1290 }
1291 EXPORT_SYMBOL(tgt_sync);
1292 /*
1293  * Unified target DLM handlers.
1294  */
1295
1296 /**
1297  * Unified target BAST
1298  *
1299  * Ensure data and metadata are synced to disk when lock is canceled if Sync on
1300  * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
1301  * but if it's cross-MDT lock, because remote object version is not set, a
1302  * filesystem sync is needed.
1303  *
1304  * \param lock server side lock
1305  * \param desc lock desc
1306  * \param data ldlm_cb_set_arg
1307  * \param flag  indicates whether this cancelling or blocking callback
1308  * \retval      0 on success
1309  * \retval      negative number on error
1310  */
1311 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1312                             void *data, int flag)
1313 {
1314         struct lu_env            env;
1315         struct lu_target        *tgt;
1316         struct dt_object        *obj = NULL;
1317         struct lu_fid            fid;
1318         int                      rc = 0;
1319
1320         ENTRY;
1321
1322         tgt = class_exp2tgt(lock->l_export);
1323
1324         if (unlikely(tgt == NULL)) {
1325                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1326                        class_exp2obd(lock->l_export)->obd_name);
1327                 RETURN(-EINVAL);
1328         }
1329
1330         if (flag == LDLM_CB_CANCELING &&
1331             (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
1332             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1333              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1334               ldlm_is_cbpending(lock))) &&
1335             ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
1336              lock->l_resource->lr_type == LDLM_EXTENT)) {
1337                 __u64 start = 0;
1338                 __u64 end = OBD_OBJECT_EOF;
1339
1340                 rc = lu_env_init(&env, LCT_DT_THREAD);
1341                 if (unlikely(rc != 0))
1342                         RETURN(rc);
1343
1344                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1345                                    tgt->lut_lsd.lsd_osd_index);
1346
1347                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1348                         obj = dt_locate(&env, tgt->lut_bottom, &fid);
1349                         if (IS_ERR(obj))
1350                                 GOTO(err_env, rc = PTR_ERR(obj));
1351
1352                         if (!dt_object_exists(obj))
1353                                 GOTO(err_put, rc = -ENOENT);
1354
1355                         start = lock->l_policy_data.l_extent.start;
1356                         end = lock->l_policy_data.l_extent.end;
1357                 }
1358
1359                 rc = tgt_sync(&env, tgt, obj, start, end);
1360                 if (rc < 0) {
1361                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1362                                "cancel: rc = %d\n",
1363                                tgt_name(tgt), PFID(&fid),
1364                                lock->l_policy_data.l_extent.start,
1365                                lock->l_policy_data.l_extent.end, rc);
1366                 }
1367 err_put:
1368                 if (obj != NULL)
1369                         lu_object_put(&env, &obj->do_lu);
1370 err_env:
1371                 lu_env_fini(&env);
1372         }
1373
1374         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1375         RETURN(rc);
1376 }
1377
1378 static struct ldlm_callback_suite tgt_dlm_cbs = {
1379         .lcs_completion = ldlm_server_completion_ast,
1380         .lcs_blocking   = tgt_blocking_ast,
1381         .lcs_glimpse    = ldlm_server_glimpse_ast
1382 };
1383
1384 int tgt_enqueue(struct tgt_session_info *tsi)
1385 {
1386         struct ptlrpc_request *req = tgt_ses_req(tsi);
1387         int rc;
1388
1389         ENTRY;
1390         /*
1391          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1392          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1393          */
1394         LASSERT(tsi->tsi_dlm_req != NULL);
1395         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1396                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1397         if (rc)
1398                 RETURN(err_serious(rc));
1399
1400         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1401         case LUT_FAIL_MDT:
1402                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1403                 break;
1404         case LUT_FAIL_OST:
1405                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1406                 break;
1407         case LUT_FAIL_MGT:
1408                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1409                 break;
1410         default:
1411                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1412                 break;
1413         }
1414         RETURN(req->rq_status);
1415 }
1416 EXPORT_SYMBOL(tgt_enqueue);
1417
1418 int tgt_convert(struct tgt_session_info *tsi)
1419 {
1420         struct ptlrpc_request *req = tgt_ses_req(tsi);
1421         int rc;
1422
1423         ENTRY;
1424         LASSERT(tsi->tsi_dlm_req);
1425         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1426         if (rc)
1427                 RETURN(err_serious(rc));
1428
1429         RETURN(req->rq_status);
1430 }
1431
1432 int tgt_bl_callback(struct tgt_session_info *tsi)
1433 {
1434         return err_serious(-EOPNOTSUPP);
1435 }
1436
1437 int tgt_cp_callback(struct tgt_session_info *tsi)
1438 {
1439         return err_serious(-EOPNOTSUPP);
1440 }
1441
1442 /* generic LDLM target handler */
1443 struct tgt_handler tgt_dlm_handlers[] = {
1444 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1445 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1446 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1447 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1448 };
1449 EXPORT_SYMBOL(tgt_dlm_handlers);
1450
1451 /*
1452  * Unified target LLOG handlers.
1453  */
1454 int tgt_llog_open(struct tgt_session_info *tsi)
1455 {
1456         int rc;
1457
1458         ENTRY;
1459
1460         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1461
1462         RETURN(rc);
1463 }
1464 EXPORT_SYMBOL(tgt_llog_open);
1465
1466 int tgt_llog_close(struct tgt_session_info *tsi)
1467 {
1468         int rc;
1469
1470         ENTRY;
1471
1472         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1473
1474         RETURN(rc);
1475 }
1476 EXPORT_SYMBOL(tgt_llog_close);
1477
1478
1479 int tgt_llog_destroy(struct tgt_session_info *tsi)
1480 {
1481         int rc;
1482
1483         ENTRY;
1484
1485         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1486
1487         RETURN(rc);
1488 }
1489
1490 int tgt_llog_read_header(struct tgt_session_info *tsi)
1491 {
1492         int rc;
1493
1494         ENTRY;
1495
1496         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1497
1498         RETURN(rc);
1499 }
1500 EXPORT_SYMBOL(tgt_llog_read_header);
1501
1502 int tgt_llog_next_block(struct tgt_session_info *tsi)
1503 {
1504         int rc;
1505
1506         ENTRY;
1507
1508         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1509
1510         RETURN(rc);
1511 }
1512 EXPORT_SYMBOL(tgt_llog_next_block);
1513
1514 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1515 {
1516         int rc;
1517
1518         ENTRY;
1519
1520         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1521
1522         RETURN(rc);
1523 }
1524 EXPORT_SYMBOL(tgt_llog_prev_block);
1525
1526 /* generic llog target handler */
1527 struct tgt_handler tgt_llog_handlers[] = {
1528 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1529 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1530 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1531 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1532 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1533 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1534 };
1535 EXPORT_SYMBOL(tgt_llog_handlers);
1536
1537 /*
1538  * sec context handlers
1539  */
1540 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1541 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1542 {
1543         return 0;
1544 }
1545
1546 struct tgt_handler tgt_sec_ctx_handlers[] = {
1547 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1548 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1549 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1550 };
1551 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1552
1553 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1554                            struct dt_device *key,
1555                            struct lfsck_request *lr,
1556                            struct thandle *th) = NULL;
1557
1558 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1559                                                 struct dt_device *,
1560                                                 struct lfsck_request *,
1561                                                 struct thandle *))
1562 {
1563         tgt_lfsck_in_notify = notify;
1564 }
1565 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1566
1567 static int (*tgt_lfsck_query)(const struct lu_env *env,
1568                               struct dt_device *key,
1569                               struct lfsck_request *lr) = NULL;
1570
1571 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1572                                            struct dt_device *,
1573                                            struct lfsck_request *))
1574 {
1575         tgt_lfsck_query = query;
1576 }
1577 EXPORT_SYMBOL(tgt_register_lfsck_query);
1578
1579 /* LFSCK request handlers */
1580 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1581 {
1582         const struct lu_env     *env = tsi->tsi_env;
1583         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1584         struct lfsck_request    *lr;
1585         int                      rc;
1586         ENTRY;
1587
1588         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1589         if (lr == NULL)
1590                 RETURN(-EPROTO);
1591
1592         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1593
1594         RETURN(rc);
1595 }
1596
1597 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1598 {
1599         struct lfsck_request    *request;
1600         struct lfsck_reply      *reply;
1601         int                      rc;
1602         ENTRY;
1603
1604         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1605         if (request == NULL)
1606                 RETURN(-EPROTO);
1607
1608         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1609         if (reply == NULL)
1610                 RETURN(-ENOMEM);
1611
1612         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1613         reply->lr_status = rc;
1614
1615         RETURN(rc < 0 ? rc : 0);
1616 }
1617
1618 struct tgt_handler tgt_lfsck_handlers[] = {
1619 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1620 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1621 };
1622 EXPORT_SYMBOL(tgt_lfsck_handlers);
1623
1624 /*
1625  * initialize per-thread page pool (bug 5137).
1626  */
1627 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1628 {
1629         struct tgt_thread_big_cache *tbc;
1630
1631         ENTRY;
1632
1633         LASSERT(thread != NULL);
1634         LASSERT(thread->t_data == NULL);
1635
1636         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1637         if (tbc == NULL)
1638                 RETURN(-ENOMEM);
1639         thread->t_data = tbc;
1640         RETURN(0);
1641 }
1642 EXPORT_SYMBOL(tgt_io_thread_init);
1643
1644 /*
1645  * free per-thread pool created by tgt_thread_init().
1646  */
1647 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1648 {
1649         struct tgt_thread_big_cache *tbc;
1650
1651         ENTRY;
1652
1653         LASSERT(thread != NULL);
1654
1655         /*
1656          * be prepared to handle partially-initialized pools (because this is
1657          * called from ost_io_thread_init() for cleanup.
1658          */
1659         tbc = thread->t_data;
1660         if (tbc != NULL) {
1661                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1662                 thread->t_data = NULL;
1663         }
1664         EXIT;
1665 }
1666 EXPORT_SYMBOL(tgt_io_thread_done);
1667 /**
1668  * Helper function for getting server side [start, start+count] DLM lock
1669  * if asked by client.
1670  */
1671 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1672                     __u64 start, __u64 end, struct lustre_handle *lh,
1673                     int mode, __u64 *flags)
1674 {
1675         union ldlm_policy_data policy;
1676         int rc;
1677
1678         ENTRY;
1679
1680         LASSERT(lh != NULL);
1681         LASSERT(ns != NULL);
1682         LASSERT(!lustre_handle_is_used(lh));
1683
1684         policy.l_extent.gid = 0;
1685         policy.l_extent.start = start & PAGE_MASK;
1686
1687         /*
1688          * If ->o_blocks is EOF it means "lock till the end of the file".
1689          * Otherwise, it's size of an extent or hole being punched (in bytes).
1690          */
1691         if (end == OBD_OBJECT_EOF || end < start)
1692                 policy.l_extent.end = OBD_OBJECT_EOF;
1693         else
1694                 policy.l_extent.end = end | ~PAGE_MASK;
1695
1696         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1697                                     flags, ldlm_blocking_ast,
1698                                     ldlm_completion_ast, ldlm_glimpse_ast,
1699                                     NULL, 0, LVB_T_NONE, NULL, lh);
1700         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1701 }
1702 EXPORT_SYMBOL(tgt_extent_lock);
1703
1704 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1705 {
1706         LASSERT(lustre_handle_is_used(lh));
1707         ldlm_lock_decref(lh, mode);
1708 }
1709 EXPORT_SYMBOL(tgt_extent_unlock);
1710
1711 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1712                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1713                  struct lustre_handle *lh, enum ldlm_mode mode)
1714 {
1715         __u64                    flags = 0;
1716         int                      nrbufs = obj->ioo_bufcnt;
1717         int                      i;
1718
1719         ENTRY;
1720
1721         LASSERT(mode == LCK_PR || mode == LCK_PW);
1722         LASSERT(!lustre_handle_is_used(lh));
1723
1724         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1725                 RETURN(0);
1726
1727         for (i = 1; i < nrbufs; i++)
1728                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1729                         RETURN(-EFAULT);
1730
1731         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1732                                nb[nrbufs - 1].rnb_offset +
1733                                nb[nrbufs - 1].rnb_len - 1,
1734                                lh, mode, &flags));
1735 }
1736
1737 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1738                     struct lustre_handle *lh, enum ldlm_mode mode)
1739 {
1740         ENTRY;
1741
1742         LASSERT(mode == LCK_PR || mode == LCK_PW);
1743         LASSERT((obj->ioo_bufcnt > 0 &&
1744                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1745                 lustre_handle_is_used(lh));
1746
1747         if (lustre_handle_is_used(lh))
1748                 tgt_extent_unlock(lh, mode);
1749         EXIT;
1750 }
1751
1752 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1753                                struct ptlrpc_bulk_desc *desc, int opc,
1754                                cksum_type_t cksum_type)
1755 {
1756         struct cfs_crypto_hash_desc     *hdesc;
1757         unsigned int                    bufsize;
1758         int                             i, err;
1759         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1760         __u32                           cksum;
1761
1762         LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1763
1764         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1765         if (IS_ERR(hdesc)) {
1766                 CERROR("%s: unable to initialize checksum hash %s\n",
1767                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1768                 return PTR_ERR(hdesc);
1769         }
1770
1771         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1772         for (i = 0; i < desc->bd_iov_count; i++) {
1773                 /* corrupt the data before we compute the checksum, to
1774                  * simulate a client->OST data error */
1775                 if (i == 0 && opc == OST_WRITE &&
1776                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1777                         int off = BD_GET_KIOV(desc, i).kiov_offset &
1778                                 ~PAGE_MASK;
1779                         int len = BD_GET_KIOV(desc, i).kiov_len;
1780                         struct page *np = tgt_page_to_corrupt;
1781                         char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1782
1783                         if (np) {
1784                                 char *ptr2 = kmap(np) + off;
1785
1786                                 memcpy(ptr2, ptr, len);
1787                                 memcpy(ptr2, "bad3", min(4, len));
1788                                 kunmap(np);
1789                                 BD_GET_KIOV(desc, i).kiov_page = np;
1790                         } else {
1791                                 CERROR("%s: can't alloc page for corruption\n",
1792                                        tgt_name(tgt));
1793                         }
1794                 }
1795                 cfs_crypto_hash_update_page(hdesc,
1796                                   BD_GET_KIOV(desc, i).kiov_page,
1797                                   BD_GET_KIOV(desc, i).kiov_offset &
1798                                         ~PAGE_MASK,
1799                                   BD_GET_KIOV(desc, i).kiov_len);
1800
1801                  /* corrupt the data after we compute the checksum, to
1802                  * simulate an OST->client data error */
1803                 if (i == 0 && opc == OST_READ &&
1804                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1805                         int off = BD_GET_KIOV(desc, i).kiov_offset
1806                           & ~PAGE_MASK;
1807                         int len = BD_GET_KIOV(desc, i).kiov_len;
1808                         struct page *np = tgt_page_to_corrupt;
1809                         char *ptr =
1810                           kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1811
1812                         if (np) {
1813                                 char *ptr2 = kmap(np) + off;
1814
1815                                 memcpy(ptr2, ptr, len);
1816                                 memcpy(ptr2, "bad4", min(4, len));
1817                                 kunmap(np);
1818                                 BD_GET_KIOV(desc, i).kiov_page = np;
1819                         } else {
1820                                 CERROR("%s: can't alloc page for corruption\n",
1821                                        tgt_name(tgt));
1822                         }
1823                 }
1824         }
1825
1826         bufsize = sizeof(cksum);
1827         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1828
1829         return cksum;
1830 }
1831
1832 int tgt_brw_read(struct tgt_session_info *tsi)
1833 {
1834         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1835         struct ptlrpc_bulk_desc *desc = NULL;
1836         struct obd_export       *exp = tsi->tsi_exp;
1837         struct niobuf_remote    *remote_nb;
1838         struct niobuf_local     *local_nb;
1839         struct obd_ioobj        *ioo;
1840         struct ost_body         *body, *repbody;
1841         struct l_wait_info       lwi;
1842         struct lustre_handle     lockh = { 0 };
1843         int                      npages, nob = 0, rc, i, no_reply = 0;
1844         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1845
1846         ENTRY;
1847
1848         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1849                 CERROR("%s: deny read request from %s to portal %u\n",
1850                        tgt_name(tsi->tsi_tgt),
1851                        obd_export_nid2str(req->rq_export),
1852                        ptlrpc_req2svc(req)->srv_req_portal);
1853                 RETURN(-EPROTO);
1854         }
1855
1856         req->rq_bulk_read = 1;
1857
1858         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1859                 RETURN(-EIO);
1860
1861         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1862                          cfs_fail_val : (obd_timeout + 1) / 4);
1863
1864         /* Check if there is eviction in progress, and if so, wait for it to
1865          * finish */
1866         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1867                 /* We do not care how long it takes */
1868                 lwi = LWI_INTR(NULL, NULL);
1869                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1870                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1871                          &lwi);
1872         }
1873
1874         /* There must be big cache in current thread to process this request
1875          * if it is NULL then something went wrong and it wasn't allocated,
1876          * report -ENOMEM in that case */
1877         if (tbc == NULL)
1878                 RETURN(-ENOMEM);
1879
1880         body = tsi->tsi_ost_body;
1881         LASSERT(body != NULL);
1882
1883         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1884         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1885
1886         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1887         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1888
1889         local_nb = tbc->local;
1890
1891         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1892                           remote_nb, &lockh, LCK_PR);
1893         if (rc != 0)
1894                 RETURN(rc);
1895
1896         /*
1897          * If getting the lock took more time than
1898          * client was willing to wait, drop it. b=11330
1899          */
1900         if (cfs_time_current_sec() > req->rq_deadline ||
1901             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1902                 no_reply = 1;
1903                 CERROR("Dropping timed-out read from %s because locking"
1904                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1905                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1906                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1907                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1908                 GOTO(out_lock, rc = -ETIMEDOUT);
1909         }
1910
1911         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1912         repbody->oa = body->oa;
1913
1914         npages = PTLRPC_MAX_BRW_PAGES;
1915         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1916                         ioo, remote_nb, &npages, local_nb);
1917         if (rc != 0)
1918                 GOTO(out_lock, rc);
1919
1920         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1921                                     PTLRPC_BULK_PUT_SOURCE |
1922                                         PTLRPC_BULK_BUF_KIOV,
1923                                     OST_BULK_PORTAL,
1924                                     &ptlrpc_bulk_kiov_nopin_ops);
1925         if (desc == NULL)
1926                 GOTO(out_commitrw, rc = -ENOMEM);
1927
1928         nob = 0;
1929         for (i = 0; i < npages; i++) {
1930                 int page_rc = local_nb[i].lnb_rc;
1931
1932                 if (page_rc < 0) {
1933                         rc = page_rc;
1934                         break;
1935                 }
1936
1937                 nob += page_rc;
1938                 if (page_rc != 0) { /* some data! */
1939                         LASSERT(local_nb[i].lnb_page != NULL);
1940                         desc->bd_frag_ops->add_kiov_frag
1941                           (desc, local_nb[i].lnb_page,
1942                            local_nb[i].lnb_page_offset,
1943                            page_rc);
1944                 }
1945
1946                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1947                         /* All subsequent pages should be 0 */
1948                         while (++i < npages)
1949                                 LASSERT(local_nb[i].lnb_rc == 0);
1950                         break;
1951                 }
1952         }
1953         if (OBD_FAIL_CHECK(OBD_FAIL_OST_READ_SIZE) &&
1954             nob != cfs_fail_val)
1955                 rc = -E2BIG;
1956
1957         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1958                 cksum_type_t cksum_type =
1959                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1960                                           body->oa.o_flags : 0);
1961                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1962                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1963                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1964                                                         OST_READ, cksum_type);
1965                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1966                        repbody->oa.o_cksum);
1967         } else {
1968                 repbody->oa.o_valid = 0;
1969         }
1970         /* We're finishing using body->oa as an input variable */
1971
1972         /* Check if client was evicted while we were doing i/o before touching
1973          * network */
1974         if (likely(rc == 0 &&
1975                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1976                 rc = target_bulk_io(exp, desc, &lwi);
1977                 no_reply = rc != 0;
1978         }
1979
1980 out_commitrw:
1981         /* Must commit after prep above in all cases */
1982         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1983                           remote_nb, npages, local_nb, rc);
1984         if (rc == 0)
1985                 tgt_drop_id(exp, &repbody->oa);
1986 out_lock:
1987         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1988
1989         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1990                 ptlrpc_free_bulk(desc);
1991
1992         LASSERT(rc <= 0);
1993         if (rc == 0) {
1994                 rc = nob;
1995                 ptlrpc_lprocfs_brw(req, nob);
1996         } else if (no_reply) {
1997                 req->rq_no_reply = 1;
1998                 /* reply out callback would free */
1999                 ptlrpc_req_drop_rs(req);
2000                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
2001                               "client will retry: rc %d\n",
2002                               exp->exp_obd->obd_name,
2003                               obd_uuid2str(&exp->exp_client_uuid),
2004                               obd_export_nid2str(exp), rc);
2005         }
2006         /* send a bulk after reply to simulate a network delay or reordering
2007          * by a router */
2008         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
2009                 wait_queue_head_t        waitq;
2010                 struct l_wait_info       lwi1;
2011
2012                 CDEBUG(D_INFO, "reorder BULK\n");
2013                 init_waitqueue_head(&waitq);
2014
2015                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
2016                 l_wait_event(waitq, 0, &lwi1);
2017                 target_bulk_io(exp, desc, &lwi);
2018                 ptlrpc_free_bulk(desc);
2019         }
2020
2021         RETURN(rc);
2022 }
2023 EXPORT_SYMBOL(tgt_brw_read);
2024
2025 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
2026                               struct ptlrpc_bulk_desc *desc,
2027                               struct niobuf_local *local_nb, int npages,
2028                               u32 client_cksum, u32 server_cksum,
2029                               bool mmap)
2030 {
2031         struct obd_export *exp = req->rq_export;
2032         struct ost_body *body;
2033         char *router = "";
2034         char *via = "";
2035
2036         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2037         LASSERT(body != NULL);
2038
2039         if (req->rq_peer.nid != desc->bd_sender) {
2040                 via = " via ";
2041                 router = libcfs_nid2str(desc->bd_sender);
2042         }
2043
2044         if (mmap) {
2045                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
2046                              client_cksum, server_cksum);
2047                 return;
2048         }
2049
2050         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
2051                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
2052                            "]: client csum %x, server csum %x\n",
2053                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
2054                            via, router,
2055                            body->oa.o_valid & OBD_MD_FLFID ?
2056                            body->oa.o_parent_seq : (__u64)0,
2057                            body->oa.o_valid & OBD_MD_FLFID ?
2058                            body->oa.o_parent_oid : 0,
2059                            body->oa.o_valid & OBD_MD_FLFID ?
2060                            body->oa.o_parent_ver : 0,
2061                            POSTID(&body->oa.o_oi),
2062                            local_nb[0].lnb_file_offset,
2063                            local_nb[npages-1].lnb_file_offset +
2064                            local_nb[npages - 1].lnb_len - 1,
2065                            client_cksum, server_cksum);
2066 }
2067
2068 int tgt_brw_write(struct tgt_session_info *tsi)
2069 {
2070         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2071         struct ptlrpc_bulk_desc *desc = NULL;
2072         struct obd_export       *exp = req->rq_export;
2073         struct niobuf_remote    *remote_nb;
2074         struct niobuf_local     *local_nb;
2075         struct obd_ioobj        *ioo;
2076         struct ost_body         *body, *repbody;
2077         struct l_wait_info       lwi;
2078         struct lustre_handle     lockh = {0};
2079         __u32                   *rcs;
2080         int                      objcount, niocount, npages;
2081         int                      rc, i, j;
2082         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
2083         bool                     no_reply = false, mmap;
2084         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2085
2086         ENTRY;
2087
2088         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2089                 CERROR("%s: deny write request from %s to portal %u\n",
2090                        tgt_name(tsi->tsi_tgt),
2091                        obd_export_nid2str(req->rq_export),
2092                        ptlrpc_req2svc(req)->srv_req_portal);
2093                 RETURN(err_serious(-EPROTO));
2094         }
2095
2096         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2097                 RETURN(err_serious(-ENOSPC));
2098         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2099                 RETURN(err_serious(-EROFS));
2100
2101         req->rq_bulk_write = 1;
2102
2103         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2104                 RETURN(err_serious(-EIO));
2105         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2106                 RETURN(err_serious(-EFAULT));
2107
2108         /* pause before transaction has been started */
2109         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2110                          cfs_fail_val : (obd_timeout + 1) / 4);
2111
2112         /* There must be big cache in current thread to process this request
2113          * if it is NULL then something went wrong and it wasn't allocated,
2114          * report -ENOMEM in that case */
2115         if (tbc == NULL)
2116                 RETURN(-ENOMEM);
2117
2118         body = tsi->tsi_ost_body;
2119         LASSERT(body != NULL);
2120
2121         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2122         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2123
2124         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2125                                         RCL_CLIENT) / sizeof(*ioo);
2126
2127         for (niocount = i = 0; i < objcount; i++)
2128                 niocount += ioo[i].ioo_bufcnt;
2129
2130         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2131         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2132         if (niocount != req_capsule_get_size(&req->rq_pill,
2133                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2134                         sizeof(*remote_nb))
2135                 RETURN(err_serious(-EPROTO));
2136
2137         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2138             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2139                 memory_pressure_set();
2140
2141         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2142                              niocount * sizeof(*rcs));
2143         rc = req_capsule_server_pack(&req->rq_pill);
2144         if (rc != 0)
2145                 GOTO(out, rc = err_serious(rc));
2146
2147         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2148         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2149
2150         local_nb = tbc->local;
2151
2152         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2153                           remote_nb, &lockh, LCK_PW);
2154         if (rc != 0)
2155                 GOTO(out, rc);
2156
2157         /*
2158          * If getting the lock took more time than
2159          * client was willing to wait, drop it. b=11330
2160          */
2161         if (cfs_time_current_sec() > req->rq_deadline ||
2162             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2163                 no_reply = true;
2164                 CERROR("%s: Dropping timed-out write from %s because locking "
2165                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
2166                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2167                        POSTID(&ioo->ioo_oid),
2168                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2169                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2170                 GOTO(out_lock, rc = -ETIMEDOUT);
2171         }
2172
2173         /* Because we already sync grant info with client when reconnect,
2174          * grant info will be cleared for resent req, then fed_grant and
2175          * total_grant will not be modified in following preprw_write */
2176         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2177                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2178                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2179         }
2180
2181         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2182         if (repbody == NULL)
2183                 GOTO(out_lock, rc = -ENOMEM);
2184         repbody->oa = body->oa;
2185
2186         npages = PTLRPC_MAX_BRW_PAGES;
2187         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2188                         objcount, ioo, remote_nb, &npages, local_nb);
2189         if (rc < 0)
2190                 GOTO(out_lock, rc);
2191
2192         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2193                                     PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2194                                     OST_BULK_PORTAL,
2195                                     &ptlrpc_bulk_kiov_nopin_ops);
2196         if (desc == NULL)
2197                 GOTO(skip_transfer, rc = -ENOMEM);
2198
2199         /* NB Having prepped, we must commit... */
2200         for (i = 0; i < npages; i++)
2201                 desc->bd_frag_ops->add_kiov_frag(desc,
2202                                                  local_nb[i].lnb_page,
2203                                                  local_nb[i].lnb_page_offset,
2204                                                  local_nb[i].lnb_len);
2205
2206         rc = sptlrpc_svc_prep_bulk(req, desc);
2207         if (rc != 0)
2208                 GOTO(skip_transfer, rc);
2209
2210         rc = target_bulk_io(exp, desc, &lwi);
2211         no_reply = rc != 0;
2212
2213 skip_transfer:
2214         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2215                 static int cksum_counter;
2216
2217                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2218                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2219
2220                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2221                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2222                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2223                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2224                                                         OST_WRITE, cksum_type);
2225                 cksum_counter++;
2226
2227                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2228                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2229                                 body->oa.o_flags & OBD_FL_MMAP);
2230
2231                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2232                                           body->oa.o_cksum,
2233                                           repbody->oa.o_cksum, mmap);
2234                         cksum_counter = 0;
2235                 } else if ((cksum_counter & (-cksum_counter)) ==
2236                            cksum_counter) {
2237                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2238                                cksum_counter, libcfs_id2str(req->rq_peer),
2239                                repbody->oa.o_cksum);
2240                 }
2241         }
2242
2243         /* Must commit after prep above in all cases */
2244         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2245                           objcount, ioo, remote_nb, npages, local_nb, rc);
2246         if (rc == -ENOTCONN)
2247                 /* quota acquire process has been given up because
2248                  * either the client has been evicted or the client
2249                  * has timed out the request already */
2250                 no_reply = true;
2251
2252         /*
2253          * Disable sending mtime back to the client. If the client locked the
2254          * whole object, then it has already updated the mtime on its side,
2255          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2256          */
2257         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2258
2259         if (rc == 0) {
2260                 int nob = 0;
2261
2262                 /* set per-requested niobuf return codes */
2263                 for (i = j = 0; i < niocount; i++) {
2264                         int len = remote_nb[i].rnb_len;
2265
2266                         nob += len;
2267                         rcs[i] = 0;
2268                         do {
2269                                 LASSERT(j < npages);
2270                                 if (local_nb[j].lnb_rc < 0)
2271                                         rcs[i] = local_nb[j].lnb_rc;
2272                                 len -= local_nb[j].lnb_len;
2273                                 j++;
2274                         } while (len > 0);
2275                         LASSERT(len == 0);
2276                 }
2277                 LASSERT(j == npages);
2278                 ptlrpc_lprocfs_brw(req, nob);
2279
2280                 tgt_drop_id(exp, &repbody->oa);
2281         }
2282 out_lock:
2283         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2284         if (desc)
2285                 ptlrpc_free_bulk(desc);
2286 out:
2287         if (no_reply) {
2288                 req->rq_no_reply = 1;
2289                 /* reply out callback would free */
2290                 ptlrpc_req_drop_rs(req);
2291                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2292                               "client will retry: rc %d\n",
2293                               exp->exp_obd->obd_name,
2294                               obd_uuid2str(&exp->exp_client_uuid),
2295                               obd_export_nid2str(exp), rc);
2296         }
2297         memory_pressure_clr();
2298         RETURN(rc);
2299 }
2300 EXPORT_SYMBOL(tgt_brw_write);
2301
2302 /* Check if request can be reconstructed from saved reply data
2303  * A copy of the reply data is returned in @trd if the pointer is not NULL
2304  */
2305 bool req_can_reconstruct(struct ptlrpc_request *req,
2306                          struct tg_reply_data *trd)
2307 {
2308         struct tg_export_data *ted = &req->rq_export->exp_target_data;
2309         struct lsd_client_data *lcd = ted->ted_lcd;
2310         bool found;
2311
2312         if (tgt_is_multimodrpcs_client(req->rq_export))
2313                 return tgt_lookup_reply(req, trd);
2314
2315         mutex_lock(&ted->ted_lcd_lock);
2316         found = req->rq_xid == lcd->lcd_last_xid ||
2317                 req->rq_xid == lcd->lcd_last_close_xid;
2318
2319         if (found && trd != NULL) {
2320                 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2321                         trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2322                         trd->trd_reply.lrd_transno =
2323                                                 lcd->lcd_last_close_transno;
2324                         trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2325                 } else {
2326                         trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2327                         trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2328                         trd->trd_reply.lrd_result = lcd->lcd_last_result;
2329                         trd->trd_reply.lrd_data = lcd->lcd_last_data;
2330                         trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2331                         trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2332                         trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2333                         trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2334                 }
2335         }
2336         mutex_unlock(&ted->ted_lcd_lock);
2337
2338         return found;
2339 }
2340 EXPORT_SYMBOL(req_can_reconstruct);
2341