lustre/target/tgt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2011, 2014, 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 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 93, 0)
 623                 /* This LBUG() can be triggered in following case:
 624                  *
 625                  * - Client send a no_resend RPC, like statfs;
 626                  * - The RPC timedout (or some other error) on client,
 627                  *   then it's removed from the unreplied list;
 628                  * - Client send some other request to bump the
 629                  *   exp_last_xid on server;
 630                  * - The former RPC got chance to be processed;
 631                  * - LBUG();
 632                  *
 633                  * Let's keep this for debug purpose for now, and it
 634                  * should be removed when release.
 635                  */
 636                 LBUG();
 637 #endif
 638                 req->rq_status = -EPROTO;
 639                 RETURN(ptlrpc_error(req));
 640         }
 641
 642         /* try to release in-memory reply data */
 643         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 644                 tgt_handle_received_xid(req->rq_export,
 645                                 lustre_msg_get_last_xid(req->rq_reqmsg));
 646                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 647                       (MSG_RESENT | MSG_REPLAY)))
 648                         tgt_handle_tag(req->rq_export,
 649                                        lustre_msg_get_tag(req->rq_reqmsg));
 650         }
 651         RETURN(0);
 652 }
 653
 654 int tgt_request_handle(struct ptlrpc_request *req)
 655 {
 656         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 657
 658         struct lustre_msg       *msg = req->rq_reqmsg;
 659         struct tgt_handler      *h;
 660         struct lu_target        *tgt;
 661         int                      request_fail_id = 0;
 662         __u32                    opc = lustre_msg_get_opc(msg);
 663         struct obd_device       *obd;
 664         int                      rc;
 665         bool                     is_connect = false;
 666         ENTRY;
 667
 668         /* Refill the context, to make sure all thread keys are allocated */
 669         lu_env_refill(req->rq_svc_thread->t_env);
 670
 671         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 672         tsi->tsi_pill = &req->rq_pill;
 673         tsi->tsi_env = req->rq_svc_thread->t_env;
 674
 675         /* if request has export then get handlers slice from corresponding
 676          * target, otherwise that should be connect operation */
 677         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 678             opc == MGS_CONNECT) {
 679                 is_connect = true;
 680                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 681                 rc = target_handle_connect(req);
 682                 if (rc != 0) {
 683                         rc = ptlrpc_error(req);
 684                         GOTO(out, rc);
 685                 }
 686                 /* recovery-small test 18c asks to drop connect reply */
 687                 if (unlikely(opc == OST_CONNECT &&
 688                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 689                         GOTO(out, rc = 0);
 690         }
 691
 692         if (unlikely(!class_connected_export(req->rq_export))) {
 693                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
 694                     opc == SEC_CTX_FINI) {
 695                         /* sec context initialization has to be handled
 696                          * by hand in tgt_handle_request0() */
 697                         tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
 698                         h = NULL;
 699                         GOTO(handle_recov, rc = 0);
 700                 }
 701                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 702                        opc, libcfs_id2str(req->rq_peer));
 703                 req->rq_status = -ENOTCONN;
 704                 rc = ptlrpc_error(req);
 705                 GOTO(out, rc);
 706         }
 707
 708         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 709         tsi->tsi_exp = req->rq_export;
 710         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 711                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 712         else
 713                 tsi->tsi_jobid = NULL;
 714
 715         if (tgt == NULL) {
 716                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 717                           class_exp2obd(req->rq_export)->obd_name);
 718                 req->rq_status = -EINVAL;
 719                 rc = ptlrpc_error(req);
 720                 GOTO(out, rc);
 721         }
 722
 723         /* Skip last_xid processing for the recovery thread, otherwise, the
 724          * last_xid on same request could be processed twice: first time when
 725          * processing the incoming request, second time when the request is
 726          * being processed by recovery thread. */
 727         obd = class_exp2obd(req->rq_export);
 728         if (is_connect) {
 729                 /* reset the exp_last_xid on each connection. */
 730                 req->rq_export->exp_last_xid = 0;
 731         } else if (obd->obd_recovery_data.trd_processing_task !=
 732                    current_pid()) {
 733                 rc = process_req_last_xid(req);
 734                 if (rc)
 735                         GOTO(out, rc);
 736         }
 737
 738         request_fail_id = tgt->lut_request_fail_id;
 739         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 740
 741         h = tgt_handler_find_check(req);
 742         if (IS_ERR(h)) {
 743                 req->rq_status = PTR_ERR(h);
 744                 rc = ptlrpc_error(req);
 745                 GOTO(out, rc);
 746         }
 747
 748         LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 749                  h->th_opc, opc);
 750
 751         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 752                 GOTO(out, rc = 0);
 753
 754         rc = lustre_msg_check_version(msg, h->th_version);
 755         if (unlikely(rc)) {
 756                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 757                           " %08x, expecting %08x\n", tgt_name(tgt),
 758                           lustre_msg_get_version(msg), h->th_version);
 759                 req->rq_status = -EINVAL;
 760                 rc = ptlrpc_error(req);
 761                 GOTO(out, rc);
 762         }
 763
 764 handle_recov:
 765         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 766         if (likely(rc == 1)) {
 767                 rc = tgt_handle_request0(tsi, h, req);
 768                 if (rc)
 769                         GOTO(out, rc);
 770         }
 771         EXIT;
 772 out:
 773         req_capsule_fini(tsi->tsi_pill);
 774         if (tsi->tsi_corpus != NULL) {
 775                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 776                 tsi->tsi_corpus = NULL;
 777         }
 778         return rc;
 779 }
 780 EXPORT_SYMBOL(tgt_request_handle);
 781
 782 /** Assign high priority operations to the request if needed. */
 783 int tgt_hpreq_handler(struct ptlrpc_request *req)
 784 {
 785         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 786         struct tgt_handler      *h;
 787         int                      rc;
 788
 789         ENTRY;
 790
 791         if (req->rq_export == NULL)
 792                 RETURN(0);
 793
 794         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 795         tsi->tsi_pill = &req->rq_pill;
 796         tsi->tsi_env = req->rq_svc_thread->t_env;
 797         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 798         tsi->tsi_exp = req->rq_export;
 799
 800         h = tgt_handler_find_check(req);
 801         if (IS_ERR(h)) {
 802                 rc = PTR_ERR(h);
 803                 RETURN(rc);
 804         }
 805
 806         rc = tgt_request_preprocess(tsi, h, req);
 807         if (unlikely(rc != 0))
 808                 RETURN(rc);
 809
 810         if (h->th_hp != NULL)
 811                 h->th_hp(tsi);
 812         RETURN(0);
 813 }
 814 EXPORT_SYMBOL(tgt_hpreq_handler);
 815
 816 void tgt_counter_incr(struct obd_export *exp, int opcode)
 817 {
 818         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 819         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 820                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 821 }
 822 EXPORT_SYMBOL(tgt_counter_incr);
 823
 824 /*
 825  * Unified target generic handlers.
 826  */
 827
 828 /*
 829  * Security functions
 830  */
 831 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 832 {
 833         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 834                                       OBD_CONNECT_RMT_CLIENT_FORCE);
 835 }
 836
 837 static int tgt_init_sec_level(struct ptlrpc_request *req)
 838 {
 839         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 840         char                    *client;
 841         struct obd_connect_data *data, *reply;
 842         int                      rc = 0;
 843         bool                     remote;
 844         ENTRY;
 845
 846         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 847         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 848         if (data == NULL || reply == NULL)
 849                 RETURN(-EFAULT);
 850
 851         /* connection from MDT is always trusted */
 852         if (req->rq_auth_usr_mdt) {
 853                 tgt_init_sec_none(reply);
 854                 RETURN(0);
 855         }
 856
 857         if (unlikely(tgt == NULL)) {
 858                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 859                           class_exp2obd(req->rq_export)->obd_name);
 860                 RETURN(-EINVAL);
 861         }
 862
 863         client = libcfs_nid2str(req->rq_peer.nid);
 864         /* no GSS support case */
 865         if (!req->rq_auth_gss) {
 866                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 867                         CWARN("client %s -> target %s does not use GSS, "
 868                               "can not run under security level %d.\n",
 869                               client, tgt_name(tgt), tgt->lut_sec_level);
 870                         RETURN(-EACCES);
 871                 } else {
 872                         tgt_init_sec_none(reply);
 873                         RETURN(0);
 874                 }
 875         }
 876
 877         /* old version case */
 878         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 879                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 880                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 881                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 882                         CWARN("client %s -> target %s uses old version, "
 883                               "can not run under security level %d.\n",
 884                               client, tgt_name(tgt), tgt->lut_sec_level);
 885                         RETURN(-EACCES);
 886                 } else {
 887                         CWARN("client %s -> target %s uses old version, "
 888                               "run under security level %d.\n",
 889                               client, tgt_name(tgt), tgt->lut_sec_level);
 890                         tgt_init_sec_none(reply);
 891                         RETURN(0);
 892                 }
 893         }
 894
 895         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 896         if (remote) {
 897                 if (!req->rq_auth_remote)
 898                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 899                                "asked to be remote.\n", client, tgt_name(tgt));
 900         } else if (req->rq_auth_remote) {
 901                 remote = true;
 902                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 903                        "as remote by default.\n", client, tgt_name(tgt));
 904         }
 905
 906         if (remote == 0) {
 907                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 908                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 909                                "authenticated!\n", client, tgt_name(tgt));
 910                         RETURN(-EACCES);
 911                 }
 912         }
 913
 914
 915         switch (tgt->lut_sec_level) {
 916         case LUSTRE_SEC_NONE:
 917                 if (remote) {
 918                         CDEBUG(D_SEC,
 919                                "client %s -> target %s is set as remote, "
 920                                "can not run under security level %d.\n",
 921                                client, tgt_name(tgt), tgt->lut_sec_level);
 922                         RETURN(-EACCES);
 923                 }
 924                 tgt_init_sec_none(reply);
 925                 break;
 926         case LUSTRE_SEC_REMOTE:
 927                 if (!remote)
 928                         tgt_init_sec_none(reply);
 929                 break;
 930         case LUSTRE_SEC_ALL:
 931                 if (remote)
 932                         break;
 933                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 934                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 935                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 936                 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 937                 break;
 938         default:
 939                 RETURN(-EINVAL);
 940         }
 941
 942         RETURN(rc);
 943 }
 944
 945 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 946 {
 947         struct lu_target        *tgt = class_exp2tgt(exp);
 948         struct sptlrpc_flavor    flvr;
 949         int                      rc = 0;
 950
 951         LASSERT(tgt);
 952         LASSERT(tgt->lut_obd);
 953         LASSERT(tgt->lut_slice);
 954
 955         /* always allow ECHO client */
 956         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 957                             LUSTRE_ECHO_NAME) == 0)) {
 958                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 959                 return 0;
 960         }
 961
 962         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 963                 read_lock(&tgt->lut_sptlrpc_lock);
 964                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 965                                              req->rq_sp_from,
 966                                              req->rq_peer.nid,
 967                                              &flvr);
 968                 read_unlock(&tgt->lut_sptlrpc_lock);
 969
 970                 spin_lock(&exp->exp_lock);
 971                 exp->exp_sp_peer = req->rq_sp_from;
 972                 exp->exp_flvr = flvr;
 973
 974                 /* when on mgs, if no restriction is set, or if client
 975                  * is loopback, allow any flavor */
 976                 if ((strcmp(exp->exp_obd->obd_type->typ_name,
 977                            LUSTRE_MGS_NAME) == 0) &&
 978                      (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
 979                       LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
 980                       == LOLND))
 981                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 982
 983                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 984                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 985                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 986                                "expect %x\n", tgt_name(tgt),
 987                                req->rq_flvr.sf_rpc,
 988                                libcfs_nid2str(req->rq_peer.nid),
 989                                exp->exp_flvr.sf_rpc);
 990                         rc = -EACCES;
 991                 }
 992                 spin_unlock(&exp->exp_lock);
 993         } else {
 994                 if (exp->exp_sp_peer != req->rq_sp_from) {
 995                         CERROR("%s: RPC source %s doesn't match %s\n",
 996                                tgt_name(tgt),
 997                                sptlrpc_part2name(req->rq_sp_from),
 998                                sptlrpc_part2name(exp->exp_sp_peer));
 999                         rc = -EACCES;
1000                 } else {
1001                         rc = sptlrpc_target_export_check(exp, req);
1002                 }
1003         }
1004
1005         return rc;
1006 }
1007
1008 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
1009 {
1010         struct sptlrpc_rule_set  tmp_rset;
1011         int                      rc;
1012
1013         if (unlikely(tgt == NULL)) {
1014                 CERROR("No target passed");
1015                 return -EINVAL;
1016         }
1017
1018         sptlrpc_rule_set_init(&tmp_rset);
1019         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
1020         if (rc) {
1021                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
1022                        tgt_name(tgt), rc);
1023                 return rc;
1024         }
1025
1026         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
1027
1028         write_lock(&tgt->lut_sptlrpc_lock);
1029         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
1030         tgt->lut_sptlrpc_rset = tmp_rset;
1031         write_unlock(&tgt->lut_sptlrpc_lock);
1032
1033         return 0;
1034 }
1035 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
1036
1037 int tgt_connect(struct tgt_session_info *tsi)
1038 {
1039         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1040         struct obd_connect_data *reply;
1041         int                      rc;
1042
1043         ENTRY;
1044
1045         rc = tgt_init_sec_level(req);
1046         if (rc != 0)
1047                 GOTO(out, rc);
1048
1049         /* XXX: better to call this check right after getting new export but
1050          * before last_rcvd slot allocation to avoid server load upon insecure
1051          * connects. This is to be fixed after unifiyng all targets.
1052          */
1053         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1054         if (rc)
1055                 GOTO(out, rc);
1056
1057         /* To avoid exposing partially initialized connection flags, changes up
1058          * to this point have been staged in reply->ocd_connect_flags. Now that
1059          * connection handling has completed successfully, atomically update
1060          * the connect flags in the shared export data structure. LU-1623 */
1061         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1062         spin_lock(&tsi->tsi_exp->exp_lock);
1063         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1064         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1065         spin_unlock(&tsi->tsi_exp->exp_lock);
1066
1067         RETURN(0);
1068 out:
1069         obd_disconnect(class_export_get(tsi->tsi_exp));
1070         return rc;
1071 }
1072 EXPORT_SYMBOL(tgt_connect);
1073
1074 int tgt_disconnect(struct tgt_session_info *tsi)
1075 {
1076         int rc;
1077
1078         ENTRY;
1079
1080         rc = target_handle_disconnect(tgt_ses_req(tsi));
1081         if (rc)
1082                 RETURN(err_serious(rc));
1083
1084         RETURN(rc);
1085 }
1086 EXPORT_SYMBOL(tgt_disconnect);
1087
1088 /*
1089  * Unified target OBD handlers
1090  */
1091 int tgt_obd_ping(struct tgt_session_info *tsi)
1092 {
1093         int rc;
1094
1095         ENTRY;
1096
1097         rc = target_handle_ping(tgt_ses_req(tsi));
1098         if (rc)
1099                 RETURN(err_serious(rc));
1100
1101         RETURN(rc);
1102 }
1103 EXPORT_SYMBOL(tgt_obd_ping);
1104
1105 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1106 {
1107         return err_serious(-EOPNOTSUPP);
1108 }
1109
1110 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
1111 {
1112         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1113         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1114         struct obd_export       *exp = req->rq_export;
1115         struct ptlrpc_bulk_desc *desc;
1116         struct l_wait_info      *lwi = &tti->tti_u.rdbuf.tti_wait_info;
1117         int                      i;
1118         int                      rc;
1119
1120         ENTRY;
1121
1122         desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
1123                                   PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
1124                                     MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1125         if (desc == NULL)
1126                 RETURN(-ENOMEM);
1127
1128         for (i = 0; i < rdbuf->rb_nbufs; i++)
1129                 desc->bd_frag_ops->add_iov_frag(desc,
1130                                         rdbuf->rb_bufs[i]->lb_buf,
1131                                         rdbuf->rb_bufs[i]->lb_len);
1132
1133         rc = target_bulk_io(exp, desc, lwi);
1134         ptlrpc_free_bulk(desc);
1135         RETURN(rc);
1136 }
1137 EXPORT_SYMBOL(tgt_send_buffer);
1138
1139 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1140 {
1141         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1142         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1143         struct obd_export       *exp = req->rq_export;
1144         struct ptlrpc_bulk_desc *desc;
1145         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1146         int                      tmpcount;
1147         int                      tmpsize;
1148         int                      i;
1149         int                      rc;
1150
1151         ENTRY;
1152
1153         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1154                                     PTLRPC_BULK_PUT_SOURCE |
1155                                         PTLRPC_BULK_BUF_KIOV,
1156                                     MDS_BULK_PORTAL,
1157                                     &ptlrpc_bulk_kiov_pin_ops);
1158         if (desc == NULL)
1159                 RETURN(-ENOMEM);
1160
1161         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1162                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1163                  * which is rdpg->rp_count */
1164                 nob = rdpg->rp_count;
1165
1166         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1167              i++, tmpcount -= tmpsize) {
1168                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1169                 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1170                                                  tmpsize);
1171         }
1172
1173         LASSERT(desc->bd_nob == nob);
1174         rc = target_bulk_io(exp, desc, lwi);
1175         ptlrpc_free_bulk(desc);
1176         RETURN(rc);
1177 }
1178 EXPORT_SYMBOL(tgt_sendpage);
1179
1180 /*
1181  * OBD_IDX_READ handler
1182  */
1183 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1184 {
1185         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1186         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1187         struct idx_info         *req_ii, *rep_ii;
1188         int                      rc, i;
1189
1190         ENTRY;
1191
1192         memset(rdpg, 0, sizeof(*rdpg));
1193         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1194
1195         /* extract idx_info buffer from request & reply */
1196         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1197         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1198                 RETURN(err_serious(-EPROTO));
1199
1200         rc = req_capsule_server_pack(tsi->tsi_pill);
1201         if (rc)
1202                 RETURN(err_serious(rc));
1203
1204         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1205         if (rep_ii == NULL)
1206                 RETURN(err_serious(-EFAULT));
1207         rep_ii->ii_magic = IDX_INFO_MAGIC;
1208
1209         /* extract hash to start with */
1210         rdpg->rp_hash = req_ii->ii_hash_start;
1211
1212         /* extract requested attributes */
1213         rdpg->rp_attrs = req_ii->ii_attrs;
1214
1215         /* check that fid packed in request is valid and supported */
1216         if (!fid_is_sane(&req_ii->ii_fid))
1217                 RETURN(-EINVAL);
1218         rep_ii->ii_fid = req_ii->ii_fid;
1219
1220         /* copy flags */
1221         rep_ii->ii_flags = req_ii->ii_flags;
1222
1223         /* compute number of pages to allocate, ii_count is the number of 4KB
1224          * containers */
1225         if (req_ii->ii_count <= 0)
1226                 GOTO(out, rc = -EFAULT);
1227         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1228                                exp_max_brw_size(tsi->tsi_exp));
1229         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1230
1231         /* allocate pages to store the containers */
1232         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1233         if (rdpg->rp_pages == NULL)
1234                 GOTO(out, rc = -ENOMEM);
1235         for (i = 0; i < rdpg->rp_npages; i++) {
1236                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1237                 if (rdpg->rp_pages[i] == NULL)
1238                         GOTO(out, rc = -ENOMEM);
1239         }
1240
1241         /* populate pages with key/record pairs */
1242         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1243         if (rc < 0)
1244                 GOTO(out, rc);
1245
1246         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1247                  "asked %d > %d\n", rc, rdpg->rp_count);
1248
1249         /* send pages to client */
1250         rc = tgt_sendpage(tsi, rdpg, rc);
1251         if (rc)
1252                 GOTO(out, rc);
1253         EXIT;
1254 out:
1255         if (rdpg->rp_pages) {
1256                 for (i = 0; i < rdpg->rp_npages; i++)
1257                         if (rdpg->rp_pages[i])
1258                                 __free_page(rdpg->rp_pages[i]);
1259                 OBD_FREE(rdpg->rp_pages,
1260                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1261         }
1262         return rc;
1263 }
1264
1265 struct tgt_handler tgt_obd_handlers[] = {
1266 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1267 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1268 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1269 };
1270 EXPORT_SYMBOL(tgt_obd_handlers);
1271
1272 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1273              struct dt_object *obj, __u64 start, __u64 end)
1274 {
1275         int rc = 0;
1276
1277         ENTRY;
1278
1279         /* if no objid is specified, it means "sync whole filesystem" */
1280         if (obj == NULL) {
1281                 rc = dt_sync(env, tgt->lut_bottom);
1282         } else if (dt_version_get(env, obj) >
1283                    tgt->lut_obd->obd_last_committed) {
1284                 rc = dt_object_sync(env, obj, start, end);
1285         }
1286
1287         RETURN(rc);
1288 }
1289 EXPORT_SYMBOL(tgt_sync);
1290 /*
1291  * Unified target DLM handlers.
1292  */
1293
1294 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1295  * (if requested) */
1296 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1297                             void *data, int flag)
1298 {
1299         struct lu_env            env;
1300         struct lu_target        *tgt;
1301         struct dt_object        *obj;
1302         struct lu_fid            fid;
1303         int                      rc = 0;
1304
1305         ENTRY;
1306
1307         tgt = class_exp2tgt(lock->l_export);
1308
1309         if (unlikely(tgt == NULL)) {
1310                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1311                        class_exp2obd(lock->l_export)->obd_name);
1312                 RETURN(-EINVAL);
1313         }
1314
1315         if (flag == LDLM_CB_CANCELING &&
1316             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1317             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1318              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1319               lock->l_flags & LDLM_FL_CBPENDING))) {
1320                 __u64 start = 0;
1321                 __u64 end = OBD_OBJECT_EOF;
1322
1323                 rc = lu_env_init(&env, LCT_DT_THREAD);
1324                 if (unlikely(rc != 0))
1325                         RETURN(rc);
1326
1327                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1328                                    tgt->lut_lsd.lsd_osd_index);
1329                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1330                 if (IS_ERR(obj))
1331                         GOTO(err_env, rc = PTR_ERR(obj));
1332
1333                 if (!dt_object_exists(obj))
1334                         GOTO(err_put, rc = -ENOENT);
1335
1336                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1337                         start = lock->l_policy_data.l_extent.start;
1338                         end = lock->l_policy_data.l_extent.end;
1339                 }
1340
1341                 rc = tgt_sync(&env, tgt, obj, start, end);
1342                 if (rc < 0) {
1343                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1344                                "cancel: rc = %d\n",
1345                                tgt_name(tgt), PFID(&fid),
1346                                lock->l_policy_data.l_extent.start,
1347                                lock->l_policy_data.l_extent.end, rc);
1348                 }
1349 err_put:
1350                 lu_object_put(&env, &obj->do_lu);
1351 err_env:
1352                 lu_env_fini(&env);
1353         }
1354
1355         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1356         RETURN(rc);
1357 }
1358
1359 static struct ldlm_callback_suite tgt_dlm_cbs = {
1360         .lcs_completion = ldlm_server_completion_ast,
1361         .lcs_blocking   = tgt_blocking_ast,
1362         .lcs_glimpse    = ldlm_server_glimpse_ast
1363 };
1364
1365 int tgt_enqueue(struct tgt_session_info *tsi)
1366 {
1367         struct ptlrpc_request *req = tgt_ses_req(tsi);
1368         int rc;
1369
1370         ENTRY;
1371         /*
1372          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1373          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1374          */
1375         LASSERT(tsi->tsi_dlm_req != NULL);
1376         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1377                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1378         if (rc)
1379                 RETURN(err_serious(rc));
1380
1381         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1382         case LUT_FAIL_MDT:
1383                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1384                 break;
1385         case LUT_FAIL_OST:
1386                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1387                 break;
1388         case LUT_FAIL_MGT:
1389                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1390                 break;
1391         default:
1392                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1393                 break;
1394         }
1395         RETURN(req->rq_status);
1396 }
1397 EXPORT_SYMBOL(tgt_enqueue);
1398
1399 int tgt_convert(struct tgt_session_info *tsi)
1400 {
1401         struct ptlrpc_request *req = tgt_ses_req(tsi);
1402         int rc;
1403
1404         ENTRY;
1405         LASSERT(tsi->tsi_dlm_req);
1406         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1407         if (rc)
1408                 RETURN(err_serious(rc));
1409
1410         RETURN(req->rq_status);
1411 }
1412
1413 int tgt_bl_callback(struct tgt_session_info *tsi)
1414 {
1415         return err_serious(-EOPNOTSUPP);
1416 }
1417
1418 int tgt_cp_callback(struct tgt_session_info *tsi)
1419 {
1420         return err_serious(-EOPNOTSUPP);
1421 }
1422
1423 /* generic LDLM target handler */
1424 struct tgt_handler tgt_dlm_handlers[] = {
1425 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1426 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1427 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1428 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1429 };
1430 EXPORT_SYMBOL(tgt_dlm_handlers);
1431
1432 /*
1433  * Unified target LLOG handlers.
1434  */
1435 int tgt_llog_open(struct tgt_session_info *tsi)
1436 {
1437         int rc;
1438
1439         ENTRY;
1440
1441         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1442
1443         RETURN(rc);
1444 }
1445 EXPORT_SYMBOL(tgt_llog_open);
1446
1447 int tgt_llog_close(struct tgt_session_info *tsi)
1448 {
1449         int rc;
1450
1451         ENTRY;
1452
1453         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1454
1455         RETURN(rc);
1456 }
1457 EXPORT_SYMBOL(tgt_llog_close);
1458
1459
1460 int tgt_llog_destroy(struct tgt_session_info *tsi)
1461 {
1462         int rc;
1463
1464         ENTRY;
1465
1466         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1467
1468         RETURN(rc);
1469 }
1470
1471 int tgt_llog_read_header(struct tgt_session_info *tsi)
1472 {
1473         int rc;
1474
1475         ENTRY;
1476
1477         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1478
1479         RETURN(rc);
1480 }
1481 EXPORT_SYMBOL(tgt_llog_read_header);
1482
1483 int tgt_llog_next_block(struct tgt_session_info *tsi)
1484 {
1485         int rc;
1486
1487         ENTRY;
1488
1489         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1490
1491         RETURN(rc);
1492 }
1493 EXPORT_SYMBOL(tgt_llog_next_block);
1494
1495 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1496 {
1497         int rc;
1498
1499         ENTRY;
1500
1501         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1502
1503         RETURN(rc);
1504 }
1505 EXPORT_SYMBOL(tgt_llog_prev_block);
1506
1507 /* generic llog target handler */
1508 struct tgt_handler tgt_llog_handlers[] = {
1509 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1510 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1511 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1512 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1513 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1514 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1515 };
1516 EXPORT_SYMBOL(tgt_llog_handlers);
1517
1518 /*
1519  * sec context handlers
1520  */
1521 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1522 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1523 {
1524         return 0;
1525 }
1526
1527 struct tgt_handler tgt_sec_ctx_handlers[] = {
1528 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1529 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1530 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1531 };
1532 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1533
1534 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1535                            struct dt_device *key,
1536                            struct lfsck_request *lr,
1537                            struct thandle *th) = NULL;
1538
1539 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1540                                                 struct dt_device *,
1541                                                 struct lfsck_request *,
1542                                                 struct thandle *))
1543 {
1544         tgt_lfsck_in_notify = notify;
1545 }
1546 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1547
1548 static int (*tgt_lfsck_query)(const struct lu_env *env,
1549                               struct dt_device *key,
1550                               struct lfsck_request *lr) = NULL;
1551
1552 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1553                                            struct dt_device *,
1554                                            struct lfsck_request *))
1555 {
1556         tgt_lfsck_query = query;
1557 }
1558 EXPORT_SYMBOL(tgt_register_lfsck_query);
1559
1560 /* LFSCK request handlers */
1561 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1562 {
1563         const struct lu_env     *env = tsi->tsi_env;
1564         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1565         struct lfsck_request    *lr;
1566         int                      rc;
1567         ENTRY;
1568
1569         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1570         if (lr == NULL)
1571                 RETURN(-EPROTO);
1572
1573         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1574
1575         RETURN(rc);
1576 }
1577
1578 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1579 {
1580         struct lfsck_request    *request;
1581         struct lfsck_reply      *reply;
1582         int                      rc;
1583         ENTRY;
1584
1585         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1586         if (request == NULL)
1587                 RETURN(-EPROTO);
1588
1589         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1590         if (reply == NULL)
1591                 RETURN(-ENOMEM);
1592
1593         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1594         reply->lr_status = rc;
1595
1596         RETURN(rc < 0 ? rc : 0);
1597 }
1598
1599 struct tgt_handler tgt_lfsck_handlers[] = {
1600 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1601 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1602 };
1603 EXPORT_SYMBOL(tgt_lfsck_handlers);
1604
1605 /*
1606  * initialize per-thread page pool (bug 5137).
1607  */
1608 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1609 {
1610         struct tgt_thread_big_cache *tbc;
1611
1612         ENTRY;
1613
1614         LASSERT(thread != NULL);
1615         LASSERT(thread->t_data == NULL);
1616
1617         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1618         if (tbc == NULL)
1619                 RETURN(-ENOMEM);
1620         thread->t_data = tbc;
1621         RETURN(0);
1622 }
1623 EXPORT_SYMBOL(tgt_io_thread_init);
1624
1625 /*
1626  * free per-thread pool created by tgt_thread_init().
1627  */
1628 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1629 {
1630         struct tgt_thread_big_cache *tbc;
1631
1632         ENTRY;
1633
1634         LASSERT(thread != NULL);
1635
1636         /*
1637          * be prepared to handle partially-initialized pools (because this is
1638          * called from ost_io_thread_init() for cleanup.
1639          */
1640         tbc = thread->t_data;
1641         if (tbc != NULL) {
1642                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1643                 thread->t_data = NULL;
1644         }
1645         EXIT;
1646 }
1647 EXPORT_SYMBOL(tgt_io_thread_done);
1648 /**
1649  * Helper function for getting server side [start, start+count] DLM lock
1650  * if asked by client.
1651  */
1652 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1653                     __u64 start, __u64 end, struct lustre_handle *lh,
1654                     int mode, __u64 *flags)
1655 {
1656         union ldlm_policy_data policy;
1657         int rc;
1658
1659         ENTRY;
1660
1661         LASSERT(lh != NULL);
1662         LASSERT(ns != NULL);
1663         LASSERT(!lustre_handle_is_used(lh));
1664
1665         policy.l_extent.gid = 0;
1666         policy.l_extent.start = start & PAGE_MASK;
1667
1668         /*
1669          * If ->o_blocks is EOF it means "lock till the end of the file".
1670          * Otherwise, it's size of an extent or hole being punched (in bytes).
1671          */
1672         if (end == OBD_OBJECT_EOF || end < start)
1673                 policy.l_extent.end = OBD_OBJECT_EOF;
1674         else
1675                 policy.l_extent.end = end | ~PAGE_MASK;
1676
1677         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1678                                     flags, ldlm_blocking_ast,
1679                                     ldlm_completion_ast, ldlm_glimpse_ast,
1680                                     NULL, 0, LVB_T_NONE, NULL, lh);
1681         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1682 }
1683 EXPORT_SYMBOL(tgt_extent_lock);
1684
1685 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1686 {
1687         LASSERT(lustre_handle_is_used(lh));
1688         ldlm_lock_decref(lh, mode);
1689 }
1690 EXPORT_SYMBOL(tgt_extent_unlock);
1691
1692 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1693                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1694                  struct lustre_handle *lh, enum ldlm_mode mode)
1695 {
1696         __u64                    flags = 0;
1697         int                      nrbufs = obj->ioo_bufcnt;
1698         int                      i;
1699
1700         ENTRY;
1701
1702         LASSERT(mode == LCK_PR || mode == LCK_PW);
1703         LASSERT(!lustre_handle_is_used(lh));
1704
1705         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1706                 RETURN(0);
1707
1708         for (i = 1; i < nrbufs; i++)
1709                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1710                         RETURN(-EFAULT);
1711
1712         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1713                                nb[nrbufs - 1].rnb_offset +
1714                                nb[nrbufs - 1].rnb_len - 1,
1715                                lh, mode, &flags));
1716 }
1717
1718 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1719                     struct lustre_handle *lh, enum ldlm_mode mode)
1720 {
1721         ENTRY;
1722
1723         LASSERT(mode == LCK_PR || mode == LCK_PW);
1724         LASSERT((obj->ioo_bufcnt > 0 &&
1725                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1726                 lustre_handle_is_used(lh));
1727
1728         if (lustre_handle_is_used(lh))
1729                 tgt_extent_unlock(lh, mode);
1730         EXIT;
1731 }
1732
1733 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1734                                struct ptlrpc_bulk_desc *desc, int opc,
1735                                cksum_type_t cksum_type)
1736 {
1737         struct cfs_crypto_hash_desc     *hdesc;
1738         unsigned int                    bufsize;
1739         int                             i, err;
1740         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1741         __u32                           cksum;
1742
1743         LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1744
1745         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1746         if (IS_ERR(hdesc)) {
1747                 CERROR("%s: unable to initialize checksum hash %s\n",
1748                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1749                 return PTR_ERR(hdesc);
1750         }
1751
1752         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1753         for (i = 0; i < desc->bd_iov_count; i++) {
1754                 /* corrupt the data before we compute the checksum, to
1755                  * simulate a client->OST data error */
1756                 if (i == 0 && opc == OST_WRITE &&
1757                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1758                         int off = BD_GET_KIOV(desc, i).kiov_offset &
1759                                 ~PAGE_MASK;
1760                         int len = BD_GET_KIOV(desc, i).kiov_len;
1761                         struct page *np = tgt_page_to_corrupt;
1762                         char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1763
1764                         if (np) {
1765                                 char *ptr2 = kmap(np) + off;
1766
1767                                 memcpy(ptr2, ptr, len);
1768                                 memcpy(ptr2, "bad3", min(4, len));
1769                                 kunmap(np);
1770                                 BD_GET_KIOV(desc, i).kiov_page = np;
1771                         } else {
1772                                 CERROR("%s: can't alloc page for corruption\n",
1773                                        tgt_name(tgt));
1774                         }
1775                 }
1776                 cfs_crypto_hash_update_page(hdesc,
1777                                   BD_GET_KIOV(desc, i).kiov_page,
1778                                   BD_GET_KIOV(desc, i).kiov_offset &
1779                                         ~PAGE_MASK,
1780                                   BD_GET_KIOV(desc, i).kiov_len);
1781
1782                  /* corrupt the data after we compute the checksum, to
1783                  * simulate an OST->client data error */
1784                 if (i == 0 && opc == OST_READ &&
1785                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1786                         int off = BD_GET_KIOV(desc, i).kiov_offset
1787                           & ~PAGE_MASK;
1788                         int len = BD_GET_KIOV(desc, i).kiov_len;
1789                         struct page *np = tgt_page_to_corrupt;
1790                         char *ptr =
1791                           kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1792
1793                         if (np) {
1794                                 char *ptr2 = kmap(np) + off;
1795
1796                                 memcpy(ptr2, ptr, len);
1797                                 memcpy(ptr2, "bad4", min(4, len));
1798                                 kunmap(np);
1799                                 BD_GET_KIOV(desc, i).kiov_page = np;
1800                         } else {
1801                                 CERROR("%s: can't alloc page for corruption\n",
1802                                        tgt_name(tgt));
1803                         }
1804                 }
1805         }
1806
1807         bufsize = sizeof(cksum);
1808         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1809
1810         return cksum;
1811 }
1812
1813 int tgt_brw_read(struct tgt_session_info *tsi)
1814 {
1815         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1816         struct ptlrpc_bulk_desc *desc = NULL;
1817         struct obd_export       *exp = tsi->tsi_exp;
1818         struct niobuf_remote    *remote_nb;
1819         struct niobuf_local     *local_nb;
1820         struct obd_ioobj        *ioo;
1821         struct ost_body         *body, *repbody;
1822         struct l_wait_info       lwi;
1823         struct lustre_handle     lockh = { 0 };
1824         int                      npages, nob = 0, rc, i, no_reply = 0;
1825         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1826
1827         ENTRY;
1828
1829         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1830                 CERROR("%s: deny read request from %s to portal %u\n",
1831                        tgt_name(tsi->tsi_tgt),
1832                        obd_export_nid2str(req->rq_export),
1833                        ptlrpc_req2svc(req)->srv_req_portal);
1834                 RETURN(-EPROTO);
1835         }
1836
1837         req->rq_bulk_read = 1;
1838
1839         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1840                 RETURN(-EIO);
1841
1842         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1843                          cfs_fail_val : (obd_timeout + 1) / 4);
1844
1845         /* Check if there is eviction in progress, and if so, wait for it to
1846          * finish */
1847         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1848                 /* We do not care how long it takes */
1849                 lwi = LWI_INTR(NULL, NULL);
1850                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1851                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1852                          &lwi);
1853         }
1854
1855         /* There must be big cache in current thread to process this request
1856          * if it is NULL then something went wrong and it wasn't allocated,
1857          * report -ENOMEM in that case */
1858         if (tbc == NULL)
1859                 RETURN(-ENOMEM);
1860
1861         body = tsi->tsi_ost_body;
1862         LASSERT(body != NULL);
1863
1864         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1865         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1866
1867         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1868         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1869
1870         local_nb = tbc->local;
1871
1872         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1873                           remote_nb, &lockh, LCK_PR);
1874         if (rc != 0)
1875                 RETURN(rc);
1876
1877         /*
1878          * If getting the lock took more time than
1879          * client was willing to wait, drop it. b=11330
1880          */
1881         if (cfs_time_current_sec() > req->rq_deadline ||
1882             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1883                 no_reply = 1;
1884                 CERROR("Dropping timed-out read from %s because locking"
1885                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1886                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1887                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1888                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1889                 GOTO(out_lock, rc = -ETIMEDOUT);
1890         }
1891
1892         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1893         repbody->oa = body->oa;
1894
1895         npages = PTLRPC_MAX_BRW_PAGES;
1896         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1897                         ioo, remote_nb, &npages, local_nb);
1898         if (rc != 0)
1899                 GOTO(out_lock, rc);
1900
1901         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1902                                     PTLRPC_BULK_PUT_SOURCE |
1903                                         PTLRPC_BULK_BUF_KIOV,
1904                                     OST_BULK_PORTAL,
1905                                     &ptlrpc_bulk_kiov_nopin_ops);
1906         if (desc == NULL)
1907                 GOTO(out_commitrw, rc = -ENOMEM);
1908
1909         nob = 0;
1910         for (i = 0; i < npages; i++) {
1911                 int page_rc = local_nb[i].lnb_rc;
1912
1913                 if (page_rc < 0) {
1914                         rc = page_rc;
1915                         break;
1916                 }
1917
1918                 nob += page_rc;
1919                 if (page_rc != 0) { /* some data! */
1920                         LASSERT(local_nb[i].lnb_page != NULL);
1921                         desc->bd_frag_ops->add_kiov_frag
1922                           (desc, local_nb[i].lnb_page,
1923                            local_nb[i].lnb_page_offset,
1924                            page_rc);
1925                 }
1926
1927                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1928                         /* All subsequent pages should be 0 */
1929                         while (++i < npages)
1930                                 LASSERT(local_nb[i].lnb_rc == 0);
1931                         break;
1932                 }
1933         }
1934
1935         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1936                 cksum_type_t cksum_type =
1937                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1938                                           body->oa.o_flags : 0);
1939                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1940                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1941                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1942                                                         OST_READ, cksum_type);
1943                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1944                        repbody->oa.o_cksum);
1945         } else {
1946                 repbody->oa.o_valid = 0;
1947         }
1948         /* We're finishing using body->oa as an input variable */
1949
1950         /* Check if client was evicted while we were doing i/o before touching
1951          * network */
1952         if (likely(rc == 0 &&
1953                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1954                 rc = target_bulk_io(exp, desc, &lwi);
1955                 no_reply = rc != 0;
1956         }
1957
1958 out_commitrw:
1959         /* Must commit after prep above in all cases */
1960         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1961                           remote_nb, npages, local_nb, rc);
1962         if (rc == 0)
1963                 tgt_drop_id(exp, &repbody->oa);
1964 out_lock:
1965         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1966
1967         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1968                 ptlrpc_free_bulk(desc);
1969
1970         LASSERT(rc <= 0);
1971         if (rc == 0) {
1972                 rc = nob;
1973                 ptlrpc_lprocfs_brw(req, nob);
1974         } else if (no_reply) {
1975                 req->rq_no_reply = 1;
1976                 /* reply out callback would free */
1977                 ptlrpc_req_drop_rs(req);
1978                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1979                               "client will retry: rc %d\n",
1980                               exp->exp_obd->obd_name,
1981                               obd_uuid2str(&exp->exp_client_uuid),
1982                               obd_export_nid2str(exp), rc);
1983         }
1984         /* send a bulk after reply to simulate a network delay or reordering
1985          * by a router */
1986         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1987                 wait_queue_head_t        waitq;
1988                 struct l_wait_info       lwi1;
1989
1990                 CDEBUG(D_INFO, "reorder BULK\n");
1991                 init_waitqueue_head(&waitq);
1992
1993                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1994                 l_wait_event(waitq, 0, &lwi1);
1995                 target_bulk_io(exp, desc, &lwi);
1996                 ptlrpc_free_bulk(desc);
1997         }
1998
1999         RETURN(rc);
2000 }
2001 EXPORT_SYMBOL(tgt_brw_read);
2002
2003 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
2004                               struct ptlrpc_bulk_desc *desc,
2005                               struct niobuf_local *local_nb, int npages,
2006                               u32 client_cksum, u32 server_cksum,
2007                               bool mmap)
2008 {
2009         struct obd_export *exp = req->rq_export;
2010         struct ost_body *body;
2011         char *router = "";
2012         char *via = "";
2013
2014         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2015         LASSERT(body != NULL);
2016
2017         if (req->rq_peer.nid != desc->bd_sender) {
2018                 via = " via ";
2019                 router = libcfs_nid2str(desc->bd_sender);
2020         }
2021
2022         if (mmap) {
2023                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
2024                              client_cksum, server_cksum);
2025                 return;
2026         }
2027
2028         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
2029                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
2030                            "]: client csum %x, server csum %x\n",
2031                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
2032                            via, router,
2033                            body->oa.o_valid & OBD_MD_FLFID ?
2034                            body->oa.o_parent_seq : (__u64)0,
2035                            body->oa.o_valid & OBD_MD_FLFID ?
2036                            body->oa.o_parent_oid : 0,
2037                            body->oa.o_valid & OBD_MD_FLFID ?
2038                            body->oa.o_parent_ver : 0,
2039                            POSTID(&body->oa.o_oi),
2040                            local_nb[0].lnb_file_offset,
2041                            local_nb[npages-1].lnb_file_offset +
2042                            local_nb[npages - 1].lnb_len - 1,
2043                            client_cksum, server_cksum);
2044 }
2045
2046 int tgt_brw_write(struct tgt_session_info *tsi)
2047 {
2048         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2049         struct ptlrpc_bulk_desc *desc = NULL;
2050         struct obd_export       *exp = req->rq_export;
2051         struct niobuf_remote    *remote_nb;
2052         struct niobuf_local     *local_nb;
2053         struct obd_ioobj        *ioo;
2054         struct ost_body         *body, *repbody;
2055         struct l_wait_info       lwi;
2056         struct lustre_handle     lockh = {0};
2057         __u32                   *rcs;
2058         int                      objcount, niocount, npages;
2059         int                      rc, i, j;
2060         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
2061         bool                     no_reply = false, mmap;
2062         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2063
2064         ENTRY;
2065
2066         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2067                 CERROR("%s: deny write request from %s to portal %u\n",
2068                        tgt_name(tsi->tsi_tgt),
2069                        obd_export_nid2str(req->rq_export),
2070                        ptlrpc_req2svc(req)->srv_req_portal);
2071                 RETURN(err_serious(-EPROTO));
2072         }
2073
2074         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2075                 RETURN(err_serious(-ENOSPC));
2076         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2077                 RETURN(err_serious(-EROFS));
2078
2079         req->rq_bulk_write = 1;
2080
2081         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2082                 RETURN(err_serious(-EIO));
2083         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2084                 RETURN(err_serious(-EFAULT));
2085
2086         /* pause before transaction has been started */
2087         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2088                          cfs_fail_val : (obd_timeout + 1) / 4);
2089
2090         /* There must be big cache in current thread to process this request
2091          * if it is NULL then something went wrong and it wasn't allocated,
2092          * report -ENOMEM in that case */
2093         if (tbc == NULL)
2094                 RETURN(-ENOMEM);
2095
2096         body = tsi->tsi_ost_body;
2097         LASSERT(body != NULL);
2098
2099         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2100         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2101
2102         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2103                                         RCL_CLIENT) / sizeof(*ioo);
2104
2105         for (niocount = i = 0; i < objcount; i++)
2106                 niocount += ioo[i].ioo_bufcnt;
2107
2108         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2109         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2110         if (niocount != req_capsule_get_size(&req->rq_pill,
2111                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2112                         sizeof(*remote_nb))
2113                 RETURN(err_serious(-EPROTO));
2114
2115         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2116             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2117                 memory_pressure_set();
2118
2119         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2120                              niocount * sizeof(*rcs));
2121         rc = req_capsule_server_pack(&req->rq_pill);
2122         if (rc != 0)
2123                 GOTO(out, rc = err_serious(rc));
2124
2125         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2126         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2127
2128         local_nb = tbc->local;
2129
2130         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2131                           remote_nb, &lockh, LCK_PW);
2132         if (rc != 0)
2133                 GOTO(out, rc);
2134
2135         /*
2136          * If getting the lock took more time than
2137          * client was willing to wait, drop it. b=11330
2138          */
2139         if (cfs_time_current_sec() > req->rq_deadline ||
2140             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2141                 no_reply = true;
2142                 CERROR("%s: Dropping timed-out write from %s because locking "
2143                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
2144                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2145                        POSTID(&ioo->ioo_oid),
2146                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2147                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2148                 GOTO(out_lock, rc = -ETIMEDOUT);
2149         }
2150
2151         /* Because we already sync grant info with client when reconnect,
2152          * grant info will be cleared for resent req, then fed_grant and
2153          * total_grant will not be modified in following preprw_write */
2154         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2155                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2156                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2157         }
2158
2159         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2160         if (repbody == NULL)
2161                 GOTO(out_lock, rc = -ENOMEM);
2162         repbody->oa = body->oa;
2163
2164         npages = PTLRPC_MAX_BRW_PAGES;
2165         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2166                         objcount, ioo, remote_nb, &npages, local_nb);
2167         if (rc < 0)
2168                 GOTO(out_lock, rc);
2169
2170         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2171                                     PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2172                                     OST_BULK_PORTAL,
2173                                     &ptlrpc_bulk_kiov_nopin_ops);
2174         if (desc == NULL)
2175                 GOTO(skip_transfer, rc = -ENOMEM);
2176
2177         /* NB Having prepped, we must commit... */
2178         for (i = 0; i < npages; i++)
2179                 desc->bd_frag_ops->add_kiov_frag(desc,
2180                                                  local_nb[i].lnb_page,
2181                                                  local_nb[i].lnb_page_offset,
2182                                                  local_nb[i].lnb_len);
2183
2184         rc = sptlrpc_svc_prep_bulk(req, desc);
2185         if (rc != 0)
2186                 GOTO(skip_transfer, rc);
2187
2188         rc = target_bulk_io(exp, desc, &lwi);
2189         no_reply = rc != 0;
2190
2191 skip_transfer:
2192         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2193                 static int cksum_counter;
2194
2195                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2196                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2197
2198                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2199                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2200                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2201                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2202                                                         OST_WRITE, cksum_type);
2203                 cksum_counter++;
2204
2205                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2206                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2207                                 body->oa.o_flags & OBD_FL_MMAP);
2208
2209                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2210                                           body->oa.o_cksum,
2211                                           repbody->oa.o_cksum, mmap);
2212                         cksum_counter = 0;
2213                 } else if ((cksum_counter & (-cksum_counter)) ==
2214                            cksum_counter) {
2215                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2216                                cksum_counter, libcfs_id2str(req->rq_peer),
2217                                repbody->oa.o_cksum);
2218                 }
2219         }
2220
2221         /* Must commit after prep above in all cases */
2222         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2223                           objcount, ioo, remote_nb, npages, local_nb, rc);
2224         if (rc == -ENOTCONN)
2225                 /* quota acquire process has been given up because
2226                  * either the client has been evicted or the client
2227                  * has timed out the request already */
2228                 no_reply = true;
2229
2230         /*
2231          * Disable sending mtime back to the client. If the client locked the
2232          * whole object, then it has already updated the mtime on its side,
2233          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2234          */
2235         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2236
2237         if (rc == 0) {
2238                 int nob = 0;
2239
2240                 /* set per-requested niobuf return codes */
2241                 for (i = j = 0; i < niocount; i++) {
2242                         int len = remote_nb[i].rnb_len;
2243
2244                         nob += len;
2245                         rcs[i] = 0;
2246                         do {
2247                                 LASSERT(j < npages);
2248                                 if (local_nb[j].lnb_rc < 0)
2249                                         rcs[i] = local_nb[j].lnb_rc;
2250                                 len -= local_nb[j].lnb_len;
2251                                 j++;
2252                         } while (len > 0);
2253                         LASSERT(len == 0);
2254                 }
2255                 LASSERT(j == npages);
2256                 ptlrpc_lprocfs_brw(req, nob);
2257
2258                 tgt_drop_id(exp, &repbody->oa);
2259         }
2260 out_lock:
2261         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2262         if (desc)
2263                 ptlrpc_free_bulk(desc);
2264 out:
2265         if (no_reply) {
2266                 req->rq_no_reply = 1;
2267                 /* reply out callback would free */
2268                 ptlrpc_req_drop_rs(req);
2269                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2270                               "client will retry: rc %d\n",
2271                               exp->exp_obd->obd_name,
2272                               obd_uuid2str(&exp->exp_client_uuid),
2273                               obd_export_nid2str(exp), rc);
2274         }
2275         memory_pressure_clr();
2276         RETURN(rc);
2277 }
2278 EXPORT_SYMBOL(tgt_brw_write);
2279
2280 /* Check if request can be reconstructed from saved reply data
2281  * A copy of the reply data is returned in @trd if the pointer is not NULL
2282  */
2283 bool req_can_reconstruct(struct ptlrpc_request *req,
2284                          struct tg_reply_data *trd)
2285 {
2286         struct tg_export_data *ted = &req->rq_export->exp_target_data;
2287         struct lsd_client_data *lcd = ted->ted_lcd;
2288         bool found;
2289
2290         if (tgt_is_multimodrpcs_client(req->rq_export))
2291                 return tgt_lookup_reply(req, trd);
2292
2293         mutex_lock(&ted->ted_lcd_lock);
2294         found = req->rq_xid == lcd->lcd_last_xid ||
2295                 req->rq_xid == lcd->lcd_last_close_xid;
2296
2297         if (found && trd != NULL) {
2298                 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2299                         trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2300                         trd->trd_reply.lrd_transno =
2301                                                 lcd->lcd_last_close_transno;
2302                         trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2303                 } else {
2304                         trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2305                         trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2306                         trd->trd_reply.lrd_result = lcd->lcd_last_result;
2307                         trd->trd_reply.lrd_data = lcd->lcd_last_data;
2308                         trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2309                         trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2310                         trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2311                         trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2312                 }
2313         }
2314         mutex_unlock(&ted->ted_lcd_lock);
2315
2316         return found;
2317 }
2318 EXPORT_SYMBOL(req_can_reconstruct);
2319