lustre/target/tgt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2012, 2015, Intel Corporation.
  25  */
  26 /*
  27  * lustre/target/tgt_handler.c
  28  *
  29  * Lustre Unified Target request handler code
  30  *
  31  * Author: Brian Behlendorf <behlendorf1@llnl.gov>
  32  * Author: Mikhail Pershin <mike.pershin@intel.com>
  33  */
  34
  35 #define DEBUG_SUBSYSTEM S_CLASS
  36
  37 #include <linux/user_namespace.h>
  38 #ifdef HAVE_UIDGID_HEADER
  39 # include <linux/uidgid.h>
  40 #endif
  41
  42 #include <obd.h>
  43 #include <obd_class.h>
  44 #include <obd_cksum.h>
  45 #include <md_object.h>
  46 #include <lustre_lfsck.h>
  47 #include <lustre_nodemap.h>
  48
  49 #include "tgt_internal.h"
  50
  51 char *tgt_name(struct lu_target *tgt)
  52 {
  53         LASSERT(tgt->lut_obd != NULL);
  54         return tgt->lut_obd->obd_name;
  55 }
  56 EXPORT_SYMBOL(tgt_name);
  57
  58 /*
  59  * Generic code handling requests that have struct mdt_body passed in:
  60  *
  61  *  - extract mdt_body from request and save it in @tsi, if present;
  62  *
  63  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  64  *  @tsi;
  65  *
  66  *  - if HABEO_CORPUS flag is set for this request type check whether object
  67  *  actually exists on storage (lu_object_exists()).
  68  *
  69  */
  70 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  71 {
  72         const struct mdt_body   *body;
  73         struct lu_object        *obj;
  74         struct req_capsule      *pill = tsi->tsi_pill;
  75         int                      rc;
  76
  77         ENTRY;
  78
  79         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  80         if (body == NULL)
  81                 RETURN(-EFAULT);
  82
  83         tsi->tsi_mdt_body = body;
  84
  85         if (!(body->mbo_valid & OBD_MD_FLID))
  86                 RETURN(0);
  87
  88         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  89          * in any case in pre-2.5 clients. Fix that here if needed */
  90         if (unlikely(fid_is_zero(&body->mbo_fid1)))
  91                 RETURN(0);
  92
  93         if (!fid_is_sane(&body->mbo_fid1)) {
  94                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  95                        PFID(&body->mbo_fid1));
  96                 RETURN(-EINVAL);
  97         }
  98
  99         obj = lu_object_find(tsi->tsi_env,
 100                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
 101                              &body->mbo_fid1, NULL);
 102         if (!IS_ERR(obj)) {
 103                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
 104                         lu_object_put(tsi->tsi_env, obj);
 105                         rc = -ENOENT;
 106                 } else {
 107                         tsi->tsi_corpus = obj;
 108                         rc = 0;
 109                 }
 110         } else {
 111                 rc = PTR_ERR(obj);
 112         }
 113
 114         tsi->tsi_fid = body->mbo_fid1;
 115
 116         RETURN(rc);
 117 }
 118
 119 /**
 120  * Validate oa from client.
 121  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 122  * req are valid.
 123  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 124  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 125  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 126  *       whether this is oi_fid or real ostid. So it will check
 127  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 128  *    c. Old FID-disable osc will send IDIF.
 129  *    d. new FID-enable osc/osp will send normal FID.
 130  *
 131  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 132  * be used for LAST_ID file, and only being accessed inside OST now.
 133  */
 134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 135 {
 136         struct ost_id   *oi     = &oa->o_oi;
 137         u64              seq    = ostid_seq(oi);
 138         u64              id     = ostid_id(oi);
 139         int              rc;
 140         ENTRY;
 141
 142         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
 143                      fid_seq_is_echo(seq))) {
 144                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 145                  * during create, and we will reset this to 1, since this
 146                  * oi_id is basically useless in the following create process,
 147                  * but oi_id == 0 will make it difficult to tell whether it is
 148                  * real FID or ost_id. */
 149                 oi->oi_fid.f_seq = FID_SEQ_ECHO;
 150                 oi->oi_fid.f_oid = id ?: 1;
 151                 oi->oi_fid.f_ver = 0;
 152         } else {
 153                 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
 154
 155                 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
 156                         GOTO(out, rc = -EPROTO);
 157
 158                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 159                  * all of the requests are required to setup FLGROUP */
 160                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 161                         ostid_set_seq_mdt0(oi);
 162                         oa->o_valid |= OBD_MD_FLGROUP;
 163                         seq = ostid_seq(oi);
 164                 }
 165
 166                 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
 167                                fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
 168                         GOTO(out, rc = -EPROTO);
 169
 170                 rc = ostid_to_fid(&tti->tti_fid1, oi,
 171                                   tsi->tsi_tgt->lut_lsd.lsd_osd_index);
 172                 if (unlikely(rc != 0))
 173                         GOTO(out, rc);
 174
 175                 oi->oi_fid = tti->tti_fid1;
 176         }
 177
 178         RETURN(0);
 179
 180 out:
 181         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 182                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
 183                seq, id, rc);
 184         return rc;
 185 }
 186 EXPORT_SYMBOL(tgt_validate_obdo);
 187
 188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
 189 {
 190         unsigned                 max_brw;
 191         struct niobuf_remote    *rnb;
 192         struct obd_ioobj        *ioo;
 193         int                      obj_count;
 194
 195         ENTRY;
 196
 197         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
 198         if (ioo == NULL)
 199                 RETURN(-EPROTO);
 200
 201         rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
 202         if (rnb == NULL)
 203                 RETURN(-EPROTO);
 204
 205         max_brw = ioobj_max_brw_get(ioo);
 206         if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 207                 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 208                        ": rc = %d\n", tgt_name(tsi->tsi_tgt),
 209                        obd_export_nid2str(tsi->tsi_exp), max_brw,
 210                        POSTID(oi), -EPROTO);
 211                 RETURN(-EPROTO);
 212         }
 213         ioo->ioo_oid = *oi;
 214
 215         obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
 216                                         RCL_CLIENT) / sizeof(*ioo);
 217         if (obj_count == 0) {
 218                 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
 219                 RETURN(-EPROTO);
 220         } else if (obj_count > 1) {
 221                 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
 222                        obj_count);
 223                 RETURN(-EPROTO);
 224         }
 225
 226         if (ioo->ioo_bufcnt == 0) {
 227                 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
 228                 RETURN(-EPROTO);
 229         }
 230
 231         if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
 232                 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
 233                           "bulk has too many pages (%d)",
 234                           ioo->ioo_bufcnt);
 235                 RETURN(-EPROTO);
 236         }
 237
 238         RETURN(0);
 239 }
 240
 241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 242 {
 243         struct ost_body         *body;
 244         struct req_capsule      *pill = tsi->tsi_pill;
 245         struct lu_nodemap       *nodemap;
 246         int                      rc;
 247
 248         ENTRY;
 249
 250         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 251         if (body == NULL)
 252                 RETURN(-EFAULT);
 253
 254         rc = tgt_validate_obdo(tsi, &body->oa);
 255         if (rc)
 256                 RETURN(rc);
 257
 258         nodemap = nodemap_get_from_exp(tsi->tsi_exp);
 259         if (IS_ERR(nodemap))
 260                 RETURN(PTR_ERR(nodemap));
 261
 262         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 263                                         NODEMAP_CLIENT_TO_FS,
 264                                         body->oa.o_uid);
 265         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 266                                         NODEMAP_CLIENT_TO_FS,
 267                                         body->oa.o_gid);
 268         nodemap_putref(nodemap);
 269
 270         tsi->tsi_ost_body = body;
 271         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 272
 273         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 274                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 275                 if (rc < 0)
 276                         RETURN(rc);
 277         }
 278
 279         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 280                 if (flags & HABEO_CORPUS) {
 281                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 282                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 283                                tgt_name(tsi->tsi_tgt));
 284                         RETURN(-EPROTO);
 285                 } else {
 286                         RETURN(0);
 287                 }
 288         }
 289
 290         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 291
 292         /*
 293          * OST doesn't get object in advance for further use to prevent
 294          * situations with nested object_find which is potential deadlock.
 295          */
 296         tsi->tsi_corpus = NULL;
 297         RETURN(rc);
 298 }
 299
 300 /*
 301  * Do necessary preprocessing according to handler ->th_flags.
 302  */
 303 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 304                                   struct tgt_handler *h,
 305                                   struct ptlrpc_request *req)
 306 {
 307         struct req_capsule      *pill = tsi->tsi_pill;
 308         __u32                    flags = h->th_flags;
 309         int                      rc = 0;
 310
 311         ENTRY;
 312
 313         if (tsi->tsi_preprocessed)
 314                 RETURN(0);
 315
 316         LASSERT(h->th_act != NULL);
 317         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 318         LASSERT(current->journal_info == NULL);
 319
 320         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 321                      h->th_fmt != NULL));
 322         if (h->th_fmt != NULL) {
 323                 req_capsule_set(pill, h->th_fmt);
 324                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 325                         rc = tgt_mdt_body_unpack(tsi, flags);
 326                         if (rc < 0)
 327                                 RETURN(rc);
 328                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 329                                                  RCL_CLIENT)) {
 330                         rc = tgt_ost_body_unpack(tsi, flags);
 331                         if (rc < 0)
 332                                 RETURN(rc);
 333                 }
 334         }
 335
 336         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 337                 RETURN(-EROFS);
 338
 339         if (flags & HABEO_CLAVIS) {
 340                 struct ldlm_request *dlm_req;
 341
 342                 LASSERT(h->th_fmt != NULL);
 343
 344                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 345                 if (dlm_req != NULL) {
 346                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 347                                      LDLM_IBITS &&
 348                                      dlm_req->lock_desc.l_policy_data.\
 349                                      l_inodebits.bits == 0)) {
 350                                 /*
 351                                  * Lock without inodebits makes no sense and
 352                                  * will oops later in ldlm. If client miss to
 353                                  * set such bits, do not trigger ASSERTION.
 354                                  *
 355                                  * For liblustre flock case, it maybe zero.
 356                                  */
 357                                 rc = -EPROTO;
 358                         } else {
 359                                 tsi->tsi_dlm_req = dlm_req;
 360                         }
 361                 } else {
 362                         rc = -EFAULT;
 363                 }
 364         }
 365         tsi->tsi_preprocessed = 1;
 366         RETURN(rc);
 367 }
 368
 369 /*
 370  * Invoke handler for this request opc. Also do necessary preprocessing
 371  * (according to handler ->th_flags), and post-processing (setting of
 372  * ->last_{xid,committed}).
 373  */
 374 static int tgt_handle_request0(struct tgt_session_info *tsi,
 375                                struct tgt_handler *h,
 376                                struct ptlrpc_request *req)
 377 {
 378         int      serious = 0;
 379         int      rc;
 380         __u32    opc = lustre_msg_get_opc(req->rq_reqmsg);
 381
 382         ENTRY;
 383
 384
 385         /* When dealing with sec context requests, no export is associated yet,
 386          * because these requests are sent before *_CONNECT requests.
 387          * A NULL req->rq_export means the normal *_common_slice handlers will
 388          * not be called, because there is no reference to the target.
 389          * So deal with them by hand and jump directly to target_send_reply().
 390          */
 391         switch (opc) {
 392         case SEC_CTX_INIT:
 393         case SEC_CTX_INIT_CONT:
 394         case SEC_CTX_FINI:
 395                 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
 396                 GOTO(out, rc = 0);
 397         }
 398
 399         /*
 400          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 401          * to put same checks into handlers like mdt_close(), mdt_reint(),
 402          * etc., without talking to mdt authors first. Checking same thing
 403          * there again is useless and returning 0 error without packing reply
 404          * is buggy! Handlers either pack reply or return error.
 405          *
 406          * We return 0 here and do not send any reply in order to emulate
 407          * network failure. Do not send any reply in case any of NET related
 408          * fail_id has occured.
 409          */
 410         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 411                 RETURN(0);
 412         if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
 413                      OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
 414                 RETURN(0);
 415
 416         rc = tgt_request_preprocess(tsi, h, req);
 417         /* pack reply if reply format is fixed */
 418         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 419                 /* Pack reply */
 420                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 421                                           RCL_SERVER))
 422                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 423                                              RCL_SERVER,
 424                                              tsi->tsi_mdt_body->mbo_eadatasize);
 425                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 426                                           RCL_SERVER))
 427                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 428                                              RCL_SERVER, 0);
 429
 430                 rc = req_capsule_server_pack(tsi->tsi_pill);
 431         }
 432
 433         if (likely(rc == 0)) {
 434                 /*
 435                  * Process request, there can be two types of rc:
 436                  * 1) errors with msg unpack/pack, other failures outside the
 437                  * operation itself. This is counted as serious errors;
 438                  * 2) errors during fs operation, should be placed in rq_status
 439                  * only
 440                  */
 441                 rc = h->th_act(tsi);
 442                 if (!is_serious(rc) &&
 443                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 444                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 445                                   "pack reply and returned 0 error\n",
 446                                   tgt_name(tsi->tsi_tgt), h->th_name);
 447                         LBUG();
 448                 }
 449                 serious = is_serious(rc);
 450                 rc = clear_serious(rc);
 451         } else {
 452                 serious = 1;
 453         }
 454
 455         req->rq_status = rc;
 456
 457         /*
 458          * ELDLM_* codes which > 0 should be in rq_status only as well as
 459          * all non-serious errors.
 460          */
 461         if (rc > 0 || !serious)
 462                 rc = 0;
 463
 464         LASSERT(current->journal_info == NULL);
 465
 466         if (likely(rc == 0 && req->rq_export))
 467                 target_committed_to_req(req);
 468
 469 out:
 470         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 471         RETURN(0);
 472 }
 473
 474 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 475                                        struct obd_device *obd, int *process)
 476 {
 477         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 478         case MDS_DISCONNECT:
 479         case OST_DISCONNECT:
 480         case OBD_IDX_READ:
 481                 *process = 1;
 482                 RETURN(0);
 483         case MDS_CLOSE:
 484         case MDS_SYNC: /* used in unmounting */
 485         case OBD_PING:
 486         case MDS_REINT:
 487         case OUT_UPDATE:
 488         case SEQ_QUERY:
 489         case FLD_QUERY:
 490         case FLD_READ:
 491         case LDLM_ENQUEUE:
 492         case OST_CREATE:
 493         case OST_DESTROY:
 494         case OST_PUNCH:
 495         case OST_SETATTR:
 496         case OST_SYNC:
 497         case OST_WRITE:
 498         case MDS_HSM_PROGRESS:
 499         case MDS_HSM_STATE_SET:
 500         case MDS_HSM_REQUEST:
 501                 *process = target_queue_recovery_request(req, obd);
 502                 RETURN(0);
 503
 504         default:
 505                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 506                 *process = -EAGAIN;
 507                 RETURN(0);
 508         }
 509 }
 510
 511 /*
 512  * Handle recovery. Return:
 513  *        +1: continue request processing;
 514  *       -ve: abort immediately with the given error code;
 515  *         0: send reply with error code in req->rq_status;
 516  */
 517 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 518 {
 519         ENTRY;
 520
 521         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 522         case MDS_CONNECT:
 523         case OST_CONNECT:
 524         case MGS_CONNECT:
 525         case SEC_CTX_INIT:
 526         case SEC_CTX_INIT_CONT:
 527         case SEC_CTX_FINI:
 528                 RETURN(+1);
 529         }
 530
 531         if (!req->rq_export->exp_obd->obd_replayable)
 532                 RETURN(+1);
 533
 534         /* sanity check: if the xid matches, the request must be marked as a
 535          * resent or replayed */
 536         if (req_can_reconstruct(req, NULL)) {
 537                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 538                       (MSG_RESENT | MSG_REPLAY))) {
 539                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 540                                   "saved xid, expected REPLAY or RESENT flag "
 541                                   "(%x)", req->rq_xid,
 542                                   lustre_msg_get_flags(req->rq_reqmsg));
 543                         req->rq_status = -ENOTCONN;
 544                         RETURN(-ENOTCONN);
 545                 }
 546         }
 547         /* else: note the opposite is not always true; a RESENT req after a
 548          * failover will usually not match the last_xid, since it was likely
 549          * never committed. A REPLAYed request will almost never match the
 550          * last xid, however it could for a committed, but still retained,
 551          * open. */
 552
 553         /* Check for aborted recovery... */
 554         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 555                 int rc;
 556                 int should_process;
 557
 558                 DEBUG_REQ(D_INFO, req, "Got new replay");
 559                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 560                                                  &should_process);
 561                 if (rc != 0 || !should_process)
 562                         RETURN(rc);
 563                 else if (should_process < 0) {
 564                         req->rq_status = should_process;
 565                         rc = ptlrpc_error(req);
 566                         RETURN(rc);
 567                 }
 568         }
 569         RETURN(+1);
 570 }
 571
 572 /* Initial check for request, it is validation mostly */
 573 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 574 {
 575         struct tgt_handler      *h;
 576         struct tgt_opc_slice    *s;
 577         struct lu_target        *tgt;
 578         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 579
 580         ENTRY;
 581
 582         tgt = class_exp2tgt(req->rq_export);
 583         if (unlikely(tgt == NULL)) {
 584                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 585                           class_exp2obd(req->rq_export)->obd_name);
 586                 RETURN(ERR_PTR(-EINVAL));
 587         }
 588
 589         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 590                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 591                         break;
 592
 593         /* opcode was not found in slice */
 594         if (unlikely(s->tos_hs == NULL)) {
 595                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 596                        opc);
 597                 RETURN(ERR_PTR(-ENOTSUPP));
 598         }
 599
 600         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 601         h = s->tos_hs + (opc - s->tos_opc_start);
 602         if (unlikely(h->th_opc == 0)) {
 603                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 604                 RETURN(ERR_PTR(-ENOTSUPP));
 605         }
 606
 607         RETURN(h);
 608 }
 609
 610 static int process_req_last_xid(struct ptlrpc_request *req)
 611 {
 612         __u64   last_xid;
 613         ENTRY;
 614
 615         /* check request's xid is consistent with export's last_xid */
 616         last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
 617         if (last_xid > req->rq_export->exp_last_xid)
 618                 req->rq_export->exp_last_xid = last_xid;
 619
 620         if (req->rq_xid == 0 ||
 621             (req->rq_xid <= req->rq_export->exp_last_xid)) {
 622                 DEBUG_REQ(D_ERROR, req, "Unexpected xid %llx vs. "
 623                           "last_xid %llx\n", req->rq_xid,
 624                           req->rq_export->exp_last_xid);
 625                 /* Some request is allowed to be sent during replay,
 626                  * such as OUT update requests, FLD requests, so it
 627                  * is possible that replay requests has smaller XID
 628                  * than the exp_last_xid.
 629                  *
 630                  * Some non-replay requests may have smaller XID as
 631                  * well:
 632                  *
 633                  * - Client send a no_resend RPC, like statfs;
 634                  * - The RPC timedout (or some other error) on client,
 635                  *   then it's removed from the unreplied list;
 636                  * - Client send some other request to bump the
 637                  *   exp_last_xid on server;
 638                  * - The former RPC got chance to be processed;
 639                  */
 640                 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
 641                         RETURN(-EPROTO);
 642         }
 643
 644         /* try to release in-memory reply data */
 645         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 646                 tgt_handle_received_xid(req->rq_export,
 647                                 lustre_msg_get_last_xid(req->rq_reqmsg));
 648                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 649                       (MSG_RESENT | MSG_REPLAY)))
 650                         tgt_handle_tag(req->rq_export,
 651                                        lustre_msg_get_tag(req->rq_reqmsg));
 652         }
 653         RETURN(0);
 654 }
 655
 656 int tgt_request_handle(struct ptlrpc_request *req)
 657 {
 658         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 659
 660         struct lustre_msg       *msg = req->rq_reqmsg;
 661         struct tgt_handler      *h;
 662         struct lu_target        *tgt;
 663         int                      request_fail_id = 0;
 664         __u32                    opc = lustre_msg_get_opc(msg);
 665         struct obd_device       *obd;
 666         int                      rc;
 667         bool                     is_connect = false;
 668         ENTRY;
 669
 670         /* Refill the context, to make sure all thread keys are allocated */
 671         lu_env_refill(req->rq_svc_thread->t_env);
 672
 673         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 674         tsi->tsi_pill = &req->rq_pill;
 675         tsi->tsi_env = req->rq_svc_thread->t_env;
 676
 677         /* if request has export then get handlers slice from corresponding
 678          * target, otherwise that should be connect operation */
 679         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 680             opc == MGS_CONNECT) {
 681                 is_connect = true;
 682                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 683                 rc = target_handle_connect(req);
 684                 if (rc != 0) {
 685                         rc = ptlrpc_error(req);
 686                         GOTO(out, rc);
 687                 }
 688                 /* recovery-small test 18c asks to drop connect reply */
 689                 if (unlikely(opc == OST_CONNECT &&
 690                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 691                         GOTO(out, rc = 0);
 692         }
 693
 694         if (unlikely(!class_connected_export(req->rq_export))) {
 695                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
 696                     opc == SEC_CTX_FINI) {
 697                         /* sec context initialization has to be handled
 698                          * by hand in tgt_handle_request0() */
 699                         tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
 700                         h = NULL;
 701                         GOTO(handle_recov, rc = 0);
 702                 }
 703                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 704                        opc, libcfs_id2str(req->rq_peer));
 705                 req->rq_status = -ENOTCONN;
 706                 rc = ptlrpc_error(req);
 707                 GOTO(out, rc);
 708         }
 709
 710         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 711         tsi->tsi_exp = req->rq_export;
 712         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 713                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 714         else
 715                 tsi->tsi_jobid = NULL;
 716
 717         if (tgt == NULL) {
 718                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 719                           class_exp2obd(req->rq_export)->obd_name);
 720                 req->rq_status = -EINVAL;
 721                 rc = ptlrpc_error(req);
 722                 GOTO(out, rc);
 723         }
 724
 725         /* Skip last_xid processing for the recovery thread, otherwise, the
 726          * last_xid on same request could be processed twice: first time when
 727          * processing the incoming request, second time when the request is
 728          * being processed by recovery thread. */
 729         obd = class_exp2obd(req->rq_export);
 730         if (is_connect) {
 731                 /* reset the exp_last_xid on each connection. */
 732                 req->rq_export->exp_last_xid = 0;
 733         } else if (obd->obd_recovery_data.trd_processing_task !=
 734                    current_pid()) {
 735                 rc = process_req_last_xid(req);
 736                 if (rc) {
 737                         req->rq_status = rc;
 738                         rc = ptlrpc_error(req);
 739                         GOTO(out, rc);
 740                 }
 741         }
 742
 743         request_fail_id = tgt->lut_request_fail_id;
 744         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 745
 746         h = tgt_handler_find_check(req);
 747         if (IS_ERR(h)) {
 748                 req->rq_status = PTR_ERR(h);
 749                 rc = ptlrpc_error(req);
 750                 GOTO(out, rc);
 751         }
 752
 753         LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 754                  h->th_opc, opc);
 755
 756         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 757                 GOTO(out, rc = 0);
 758
 759         rc = lustre_msg_check_version(msg, h->th_version);
 760         if (unlikely(rc)) {
 761                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 762                           " %08x, expecting %08x\n", tgt_name(tgt),
 763                           lustre_msg_get_version(msg), h->th_version);
 764                 req->rq_status = -EINVAL;
 765                 rc = ptlrpc_error(req);
 766                 GOTO(out, rc);
 767         }
 768
 769 handle_recov:
 770         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 771         if (likely(rc == 1)) {
 772                 rc = tgt_handle_request0(tsi, h, req);
 773                 if (rc)
 774                         GOTO(out, rc);
 775         }
 776         EXIT;
 777 out:
 778         req_capsule_fini(tsi->tsi_pill);
 779         if (tsi->tsi_corpus != NULL) {
 780                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 781                 tsi->tsi_corpus = NULL;
 782         }
 783         return rc;
 784 }
 785 EXPORT_SYMBOL(tgt_request_handle);
 786
 787 /** Assign high priority operations to the request if needed. */
 788 int tgt_hpreq_handler(struct ptlrpc_request *req)
 789 {
 790         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 791         struct tgt_handler      *h;
 792         int                      rc;
 793
 794         ENTRY;
 795
 796         if (req->rq_export == NULL)
 797                 RETURN(0);
 798
 799         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 800         tsi->tsi_pill = &req->rq_pill;
 801         tsi->tsi_env = req->rq_svc_thread->t_env;
 802         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 803         tsi->tsi_exp = req->rq_export;
 804
 805         h = tgt_handler_find_check(req);
 806         if (IS_ERR(h)) {
 807                 rc = PTR_ERR(h);
 808                 RETURN(rc);
 809         }
 810
 811         rc = tgt_request_preprocess(tsi, h, req);
 812         if (unlikely(rc != 0))
 813                 RETURN(rc);
 814
 815         if (h->th_hp != NULL)
 816                 h->th_hp(tsi);
 817         RETURN(0);
 818 }
 819 EXPORT_SYMBOL(tgt_hpreq_handler);
 820
 821 void tgt_counter_incr(struct obd_export *exp, int opcode)
 822 {
 823         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 824         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 825                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 826 }
 827 EXPORT_SYMBOL(tgt_counter_incr);
 828
 829 /*
 830  * Unified target generic handlers.
 831  */
 832
 833 /*
 834  * Security functions
 835  */
 836 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 837 {
 838         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 839                                       OBD_CONNECT_RMT_CLIENT_FORCE);
 840 }
 841
 842 static int tgt_init_sec_level(struct ptlrpc_request *req)
 843 {
 844         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 845         char                    *client;
 846         struct obd_connect_data *data, *reply;
 847         int                      rc = 0;
 848         bool                     remote;
 849         ENTRY;
 850
 851         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 852         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 853         if (data == NULL || reply == NULL)
 854                 RETURN(-EFAULT);
 855
 856         /* connection from MDT is always trusted */
 857         if (req->rq_auth_usr_mdt) {
 858                 tgt_init_sec_none(reply);
 859                 RETURN(0);
 860         }
 861
 862         if (unlikely(tgt == NULL)) {
 863                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 864                           class_exp2obd(req->rq_export)->obd_name);
 865                 RETURN(-EINVAL);
 866         }
 867
 868         client = libcfs_nid2str(req->rq_peer.nid);
 869         /* no GSS support case */
 870         if (!req->rq_auth_gss) {
 871                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 872                         CWARN("client %s -> target %s does not use GSS, "
 873                               "can not run under security level %d.\n",
 874                               client, tgt_name(tgt), tgt->lut_sec_level);
 875                         RETURN(-EACCES);
 876                 } else {
 877                         tgt_init_sec_none(reply);
 878                         RETURN(0);
 879                 }
 880         }
 881
 882         /* old version case */
 883         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 884                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 885                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 886                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 887                         CWARN("client %s -> target %s uses old version, "
 888                               "can not run under security level %d.\n",
 889                               client, tgt_name(tgt), tgt->lut_sec_level);
 890                         RETURN(-EACCES);
 891                 } else {
 892                         CWARN("client %s -> target %s uses old version, "
 893                               "run under security level %d.\n",
 894                               client, tgt_name(tgt), tgt->lut_sec_level);
 895                         tgt_init_sec_none(reply);
 896                         RETURN(0);
 897                 }
 898         }
 899
 900         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 901         if (remote) {
 902                 if (!req->rq_auth_remote)
 903                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 904                                "asked to be remote.\n", client, tgt_name(tgt));
 905         } else if (req->rq_auth_remote) {
 906                 remote = true;
 907                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 908                        "as remote by default.\n", client, tgt_name(tgt));
 909         }
 910
 911         if (remote == 0) {
 912                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 913                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 914                                "authenticated!\n", client, tgt_name(tgt));
 915                         RETURN(-EACCES);
 916                 }
 917         }
 918
 919
 920         switch (tgt->lut_sec_level) {
 921         case LUSTRE_SEC_NONE:
 922                 if (remote) {
 923                         CDEBUG(D_SEC,
 924                                "client %s -> target %s is set as remote, "
 925                                "can not run under security level %d.\n",
 926                                client, tgt_name(tgt), tgt->lut_sec_level);
 927                         RETURN(-EACCES);
 928                 }
 929                 tgt_init_sec_none(reply);
 930                 break;
 931         case LUSTRE_SEC_REMOTE:
 932                 if (!remote)
 933                         tgt_init_sec_none(reply);
 934                 break;
 935         case LUSTRE_SEC_ALL:
 936                 if (remote)
 937                         break;
 938                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 939                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 940                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 941                 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 942                 break;
 943         default:
 944                 RETURN(-EINVAL);
 945         }
 946
 947         RETURN(rc);
 948 }
 949
 950 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 951 {
 952         struct lu_target        *tgt = class_exp2tgt(exp);
 953         struct sptlrpc_flavor    flvr;
 954         int                      rc = 0;
 955
 956         LASSERT(tgt);
 957         LASSERT(tgt->lut_obd);
 958         LASSERT(tgt->lut_slice);
 959
 960         /* always allow ECHO client */
 961         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 962                             LUSTRE_ECHO_NAME) == 0)) {
 963                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 964                 return 0;
 965         }
 966
 967         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 968                 read_lock(&tgt->lut_sptlrpc_lock);
 969                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 970                                              req->rq_sp_from,
 971                                              req->rq_peer.nid,
 972                                              &flvr);
 973                 read_unlock(&tgt->lut_sptlrpc_lock);
 974
 975                 spin_lock(&exp->exp_lock);
 976                 exp->exp_sp_peer = req->rq_sp_from;
 977                 exp->exp_flvr = flvr;
 978
 979                 /* when on mgs, if no restriction is set, or if client
 980                  * is loopback, allow any flavor */
 981                 if ((strcmp(exp->exp_obd->obd_type->typ_name,
 982                            LUSTRE_MGS_NAME) == 0) &&
 983                      (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
 984                       LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
 985                       == LOLND))
 986                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 987
 988                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 989                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 990                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 991                                "expect %x\n", tgt_name(tgt),
 992                                req->rq_flvr.sf_rpc,
 993                                libcfs_nid2str(req->rq_peer.nid),
 994                                exp->exp_flvr.sf_rpc);
 995                         rc = -EACCES;
 996                 }
 997                 spin_unlock(&exp->exp_lock);
 998         } else {
 999                 if (exp->exp_sp_peer != req->rq_sp_from) {
1000                         CERROR("%s: RPC source %s doesn't match %s\n",
1001                                tgt_name(tgt),
1002                                sptlrpc_part2name(req->rq_sp_from),
1003                                sptlrpc_part2name(exp->exp_sp_peer));
1004                         rc = -EACCES;
1005                 } else {
1006                         rc = sptlrpc_target_export_check(exp, req);
1007                 }
1008         }
1009
1010         return rc;
1011 }
1012
1013 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
1014 {
1015         struct sptlrpc_rule_set  tmp_rset;
1016         int                      rc;
1017
1018         if (unlikely(tgt == NULL)) {
1019                 CERROR("No target passed");
1020                 return -EINVAL;
1021         }
1022
1023         sptlrpc_rule_set_init(&tmp_rset);
1024         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
1025         if (rc) {
1026                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
1027                        tgt_name(tgt), rc);
1028                 return rc;
1029         }
1030
1031         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
1032
1033         write_lock(&tgt->lut_sptlrpc_lock);
1034         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
1035         tgt->lut_sptlrpc_rset = tmp_rset;
1036         write_unlock(&tgt->lut_sptlrpc_lock);
1037
1038         return 0;
1039 }
1040 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
1041
1042 int tgt_connect(struct tgt_session_info *tsi)
1043 {
1044         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1045         struct obd_connect_data *reply;
1046         int                      rc;
1047
1048         ENTRY;
1049
1050         rc = tgt_init_sec_level(req);
1051         if (rc != 0)
1052                 GOTO(out, rc);
1053
1054         /* XXX: better to call this check right after getting new export but
1055          * before last_rcvd slot allocation to avoid server load upon insecure
1056          * connects. This is to be fixed after unifiyng all targets.
1057          */
1058         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1059         if (rc)
1060                 GOTO(out, rc);
1061
1062         /* To avoid exposing partially initialized connection flags, changes up
1063          * to this point have been staged in reply->ocd_connect_flags. Now that
1064          * connection handling has completed successfully, atomically update
1065          * the connect flags in the shared export data structure. LU-1623 */
1066         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1067         spin_lock(&tsi->tsi_exp->exp_lock);
1068         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1069         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1070         spin_unlock(&tsi->tsi_exp->exp_lock);
1071
1072         RETURN(0);
1073 out:
1074         obd_disconnect(class_export_get(tsi->tsi_exp));
1075         return rc;
1076 }
1077 EXPORT_SYMBOL(tgt_connect);
1078
1079 int tgt_disconnect(struct tgt_session_info *tsi)
1080 {
1081         int rc;
1082
1083         ENTRY;
1084
1085         rc = target_handle_disconnect(tgt_ses_req(tsi));
1086         if (rc)
1087                 RETURN(err_serious(rc));
1088
1089         RETURN(rc);
1090 }
1091 EXPORT_SYMBOL(tgt_disconnect);
1092
1093 /*
1094  * Unified target OBD handlers
1095  */
1096 int tgt_obd_ping(struct tgt_session_info *tsi)
1097 {
1098         int rc;
1099
1100         ENTRY;
1101
1102         rc = target_handle_ping(tgt_ses_req(tsi));
1103         if (rc)
1104                 RETURN(err_serious(rc));
1105
1106         RETURN(rc);
1107 }
1108 EXPORT_SYMBOL(tgt_obd_ping);
1109
1110 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1111 {
1112         return err_serious(-EOPNOTSUPP);
1113 }
1114
1115 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
1116 {
1117         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1118         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1119         struct obd_export       *exp = req->rq_export;
1120         struct ptlrpc_bulk_desc *desc;
1121         struct l_wait_info      *lwi = &tti->tti_u.update.tti_wait_info;
1122         int                      i;
1123         int                      rc;
1124
1125         ENTRY;
1126
1127         desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
1128                                   PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
1129                                     MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1130         if (desc == NULL)
1131                 RETURN(-ENOMEM);
1132
1133         for (i = 0; i < rdbuf->rb_nbufs; i++)
1134                 desc->bd_frag_ops->add_iov_frag(desc,
1135                                         rdbuf->rb_bufs[i].lb_buf,
1136                                         rdbuf->rb_bufs[i].lb_len);
1137
1138         rc = target_bulk_io(exp, desc, lwi);
1139         ptlrpc_free_bulk(desc);
1140         RETURN(rc);
1141 }
1142 EXPORT_SYMBOL(tgt_send_buffer);
1143
1144 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1145 {
1146         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1147         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1148         struct obd_export       *exp = req->rq_export;
1149         struct ptlrpc_bulk_desc *desc;
1150         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1151         int                      tmpcount;
1152         int                      tmpsize;
1153         int                      i;
1154         int                      rc;
1155
1156         ENTRY;
1157
1158         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1159                                     PTLRPC_BULK_PUT_SOURCE |
1160                                         PTLRPC_BULK_BUF_KIOV,
1161                                     MDS_BULK_PORTAL,
1162                                     &ptlrpc_bulk_kiov_pin_ops);
1163         if (desc == NULL)
1164                 RETURN(-ENOMEM);
1165
1166         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1167                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1168                  * which is rdpg->rp_count */
1169                 nob = rdpg->rp_count;
1170
1171         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1172              i++, tmpcount -= tmpsize) {
1173                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1174                 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1175                                                  tmpsize);
1176         }
1177
1178         LASSERT(desc->bd_nob == nob);
1179         rc = target_bulk_io(exp, desc, lwi);
1180         ptlrpc_free_bulk(desc);
1181         RETURN(rc);
1182 }
1183 EXPORT_SYMBOL(tgt_sendpage);
1184
1185 /*
1186  * OBD_IDX_READ handler
1187  */
1188 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1189 {
1190         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1191         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1192         struct idx_info         *req_ii, *rep_ii;
1193         int                      rc, i;
1194
1195         ENTRY;
1196
1197         memset(rdpg, 0, sizeof(*rdpg));
1198         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1199
1200         /* extract idx_info buffer from request & reply */
1201         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1202         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1203                 RETURN(err_serious(-EPROTO));
1204
1205         rc = req_capsule_server_pack(tsi->tsi_pill);
1206         if (rc)
1207                 RETURN(err_serious(rc));
1208
1209         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1210         if (rep_ii == NULL)
1211                 RETURN(err_serious(-EFAULT));
1212         rep_ii->ii_magic = IDX_INFO_MAGIC;
1213
1214         /* extract hash to start with */
1215         rdpg->rp_hash = req_ii->ii_hash_start;
1216
1217         /* extract requested attributes */
1218         rdpg->rp_attrs = req_ii->ii_attrs;
1219
1220         /* check that fid packed in request is valid and supported */
1221         if (!fid_is_sane(&req_ii->ii_fid))
1222                 RETURN(-EINVAL);
1223         rep_ii->ii_fid = req_ii->ii_fid;
1224
1225         /* copy flags */
1226         rep_ii->ii_flags = req_ii->ii_flags;
1227
1228         /* compute number of pages to allocate, ii_count is the number of 4KB
1229          * containers */
1230         if (req_ii->ii_count <= 0)
1231                 GOTO(out, rc = -EFAULT);
1232         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1233                                exp_max_brw_size(tsi->tsi_exp));
1234         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1235
1236         /* allocate pages to store the containers */
1237         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1238         if (rdpg->rp_pages == NULL)
1239                 GOTO(out, rc = -ENOMEM);
1240         for (i = 0; i < rdpg->rp_npages; i++) {
1241                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1242                 if (rdpg->rp_pages[i] == NULL)
1243                         GOTO(out, rc = -ENOMEM);
1244         }
1245
1246         /* populate pages with key/record pairs */
1247         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1248         if (rc < 0)
1249                 GOTO(out, rc);
1250
1251         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1252                  "asked %d > %d\n", rc, rdpg->rp_count);
1253
1254         /* send pages to client */
1255         rc = tgt_sendpage(tsi, rdpg, rc);
1256         if (rc)
1257                 GOTO(out, rc);
1258         EXIT;
1259 out:
1260         if (rdpg->rp_pages) {
1261                 for (i = 0; i < rdpg->rp_npages; i++)
1262                         if (rdpg->rp_pages[i])
1263                                 __free_page(rdpg->rp_pages[i]);
1264                 OBD_FREE(rdpg->rp_pages,
1265                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1266         }
1267         return rc;
1268 }
1269
1270 struct tgt_handler tgt_obd_handlers[] = {
1271 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1272 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1273 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1274 };
1275 EXPORT_SYMBOL(tgt_obd_handlers);
1276
1277 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1278              struct dt_object *obj, __u64 start, __u64 end)
1279 {
1280         int rc = 0;
1281
1282         ENTRY;
1283
1284         /* if no objid is specified, it means "sync whole filesystem" */
1285         if (obj == NULL) {
1286                 rc = dt_sync(env, tgt->lut_bottom);
1287         } else if (dt_version_get(env, obj) >
1288                    tgt->lut_obd->obd_last_committed) {
1289                 rc = dt_object_sync(env, obj, start, end);
1290         }
1291         atomic_inc(&tgt->lut_sync_count);
1292
1293         RETURN(rc);
1294 }
1295 EXPORT_SYMBOL(tgt_sync);
1296 /*
1297  * Unified target DLM handlers.
1298  */
1299
1300 /**
1301  * Unified target BAST
1302  *
1303  * Ensure data and metadata are synced to disk when lock is canceled if Sync on
1304  * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
1305  * but if it's cross-MDT lock, because remote object version is not set, a
1306  * filesystem sync is needed.
1307  *
1308  * \param lock server side lock
1309  * \param desc lock desc
1310  * \param data ldlm_cb_set_arg
1311  * \param flag  indicates whether this cancelling or blocking callback
1312  * \retval      0 on success
1313  * \retval      negative number on error
1314  */
1315 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1316                             void *data, int flag)
1317 {
1318         struct lu_env            env;
1319         struct lu_target        *tgt;
1320         struct dt_object        *obj = NULL;
1321         struct lu_fid            fid;
1322         int                      rc = 0;
1323
1324         ENTRY;
1325
1326         tgt = class_exp2tgt(lock->l_export);
1327
1328         if (unlikely(tgt == NULL)) {
1329                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1330                        class_exp2obd(lock->l_export)->obd_name);
1331                 RETURN(-EINVAL);
1332         }
1333
1334         if (flag == LDLM_CB_CANCELING &&
1335             (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
1336             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1337              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1338               ldlm_is_cbpending(lock))) &&
1339             ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
1340              lock->l_resource->lr_type == LDLM_EXTENT)) {
1341                 __u64 start = 0;
1342                 __u64 end = OBD_OBJECT_EOF;
1343
1344                 rc = lu_env_init(&env, LCT_DT_THREAD);
1345                 if (unlikely(rc != 0))
1346                         RETURN(rc);
1347
1348                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1349                                    tgt->lut_lsd.lsd_osd_index);
1350
1351                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1352                         obj = dt_locate(&env, tgt->lut_bottom, &fid);
1353                         if (IS_ERR(obj))
1354                                 GOTO(err_env, rc = PTR_ERR(obj));
1355
1356                         if (!dt_object_exists(obj))
1357                                 GOTO(err_put, rc = -ENOENT);
1358
1359                         start = lock->l_policy_data.l_extent.start;
1360                         end = lock->l_policy_data.l_extent.end;
1361                 }
1362
1363                 rc = tgt_sync(&env, tgt, obj, start, end);
1364                 if (rc < 0) {
1365                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1366                                "cancel: rc = %d\n",
1367                                tgt_name(tgt), PFID(&fid),
1368                                lock->l_policy_data.l_extent.start,
1369                                lock->l_policy_data.l_extent.end, rc);
1370                 }
1371 err_put:
1372                 if (obj != NULL)
1373                         lu_object_put(&env, &obj->do_lu);
1374 err_env:
1375                 lu_env_fini(&env);
1376         }
1377
1378         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1379         RETURN(rc);
1380 }
1381
1382 static struct ldlm_callback_suite tgt_dlm_cbs = {
1383         .lcs_completion = ldlm_server_completion_ast,
1384         .lcs_blocking   = tgt_blocking_ast,
1385         .lcs_glimpse    = ldlm_server_glimpse_ast
1386 };
1387
1388 int tgt_enqueue(struct tgt_session_info *tsi)
1389 {
1390         struct ptlrpc_request *req = tgt_ses_req(tsi);
1391         int rc;
1392
1393         ENTRY;
1394         /*
1395          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1396          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1397          */
1398         LASSERT(tsi->tsi_dlm_req != NULL);
1399         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1400                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1401         if (rc)
1402                 RETURN(err_serious(rc));
1403
1404         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1405         case LUT_FAIL_MDT:
1406                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1407                 break;
1408         case LUT_FAIL_OST:
1409                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1410                 break;
1411         case LUT_FAIL_MGT:
1412                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1413                 break;
1414         default:
1415                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1416                 break;
1417         }
1418         RETURN(req->rq_status);
1419 }
1420 EXPORT_SYMBOL(tgt_enqueue);
1421
1422 int tgt_convert(struct tgt_session_info *tsi)
1423 {
1424         struct ptlrpc_request *req = tgt_ses_req(tsi);
1425         int rc;
1426
1427         ENTRY;
1428         LASSERT(tsi->tsi_dlm_req);
1429         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1430         if (rc)
1431                 RETURN(err_serious(rc));
1432
1433         RETURN(req->rq_status);
1434 }
1435
1436 int tgt_bl_callback(struct tgt_session_info *tsi)
1437 {
1438         return err_serious(-EOPNOTSUPP);
1439 }
1440
1441 int tgt_cp_callback(struct tgt_session_info *tsi)
1442 {
1443         return err_serious(-EOPNOTSUPP);
1444 }
1445
1446 /* generic LDLM target handler */
1447 struct tgt_handler tgt_dlm_handlers[] = {
1448 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1449 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1450 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1451 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1452 };
1453 EXPORT_SYMBOL(tgt_dlm_handlers);
1454
1455 /*
1456  * Unified target LLOG handlers.
1457  */
1458 int tgt_llog_open(struct tgt_session_info *tsi)
1459 {
1460         int rc;
1461
1462         ENTRY;
1463
1464         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1465
1466         RETURN(rc);
1467 }
1468 EXPORT_SYMBOL(tgt_llog_open);
1469
1470 int tgt_llog_close(struct tgt_session_info *tsi)
1471 {
1472         int rc;
1473
1474         ENTRY;
1475
1476         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1477
1478         RETURN(rc);
1479 }
1480 EXPORT_SYMBOL(tgt_llog_close);
1481
1482
1483 int tgt_llog_destroy(struct tgt_session_info *tsi)
1484 {
1485         int rc;
1486
1487         ENTRY;
1488
1489         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1490
1491         RETURN(rc);
1492 }
1493
1494 int tgt_llog_read_header(struct tgt_session_info *tsi)
1495 {
1496         int rc;
1497
1498         ENTRY;
1499
1500         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1501
1502         RETURN(rc);
1503 }
1504 EXPORT_SYMBOL(tgt_llog_read_header);
1505
1506 int tgt_llog_next_block(struct tgt_session_info *tsi)
1507 {
1508         int rc;
1509
1510         ENTRY;
1511
1512         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1513
1514         RETURN(rc);
1515 }
1516 EXPORT_SYMBOL(tgt_llog_next_block);
1517
1518 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1519 {
1520         int rc;
1521
1522         ENTRY;
1523
1524         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1525
1526         RETURN(rc);
1527 }
1528 EXPORT_SYMBOL(tgt_llog_prev_block);
1529
1530 /* generic llog target handler */
1531 struct tgt_handler tgt_llog_handlers[] = {
1532 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1533 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1534 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1535 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1536 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1537 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1538 };
1539 EXPORT_SYMBOL(tgt_llog_handlers);
1540
1541 /*
1542  * sec context handlers
1543  */
1544 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1545 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1546 {
1547         return 0;
1548 }
1549
1550 struct tgt_handler tgt_sec_ctx_handlers[] = {
1551 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1552 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1553 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1554 };
1555 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1556
1557 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1558                            struct dt_device *key,
1559                            struct lfsck_request *lr,
1560                            struct thandle *th) = NULL;
1561
1562 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1563                                                 struct dt_device *,
1564                                                 struct lfsck_request *,
1565                                                 struct thandle *))
1566 {
1567         tgt_lfsck_in_notify = notify;
1568 }
1569 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1570
1571 static int (*tgt_lfsck_query)(const struct lu_env *env,
1572                               struct dt_device *key,
1573                               struct lfsck_request *req,
1574                               struct lfsck_reply *rep,
1575                               struct lfsck_query *que) = NULL;
1576
1577 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1578                                            struct dt_device *,
1579                                            struct lfsck_request *,
1580                                            struct lfsck_reply *,
1581                                            struct lfsck_query *))
1582 {
1583         tgt_lfsck_query = query;
1584 }
1585 EXPORT_SYMBOL(tgt_register_lfsck_query);
1586
1587 /* LFSCK request handlers */
1588 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1589 {
1590         const struct lu_env     *env = tsi->tsi_env;
1591         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1592         struct lfsck_request    *lr;
1593         int                      rc;
1594         ENTRY;
1595
1596         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1597         if (lr == NULL)
1598                 RETURN(-EPROTO);
1599
1600         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1601
1602         RETURN(rc);
1603 }
1604
1605 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1606 {
1607         struct lfsck_request    *request;
1608         struct lfsck_reply      *reply;
1609         int                      rc;
1610         ENTRY;
1611
1612         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1613         if (request == NULL)
1614                 RETURN(-EPROTO);
1615
1616         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1617         if (reply == NULL)
1618                 RETURN(-ENOMEM);
1619
1620         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1621                              request, reply, NULL);
1622
1623         RETURN(rc < 0 ? rc : 0);
1624 }
1625
1626 struct tgt_handler tgt_lfsck_handlers[] = {
1627 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1628 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1629 };
1630 EXPORT_SYMBOL(tgt_lfsck_handlers);
1631
1632 /*
1633  * initialize per-thread page pool (bug 5137).
1634  */
1635 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1636 {
1637         struct tgt_thread_big_cache *tbc;
1638
1639         ENTRY;
1640
1641         LASSERT(thread != NULL);
1642         LASSERT(thread->t_data == NULL);
1643
1644         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1645         if (tbc == NULL)
1646                 RETURN(-ENOMEM);
1647         thread->t_data = tbc;
1648         RETURN(0);
1649 }
1650 EXPORT_SYMBOL(tgt_io_thread_init);
1651
1652 /*
1653  * free per-thread pool created by tgt_thread_init().
1654  */
1655 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1656 {
1657         struct tgt_thread_big_cache *tbc;
1658
1659         ENTRY;
1660
1661         LASSERT(thread != NULL);
1662
1663         /*
1664          * be prepared to handle partially-initialized pools (because this is
1665          * called from ost_io_thread_init() for cleanup.
1666          */
1667         tbc = thread->t_data;
1668         if (tbc != NULL) {
1669                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1670                 thread->t_data = NULL;
1671         }
1672         EXIT;
1673 }
1674 EXPORT_SYMBOL(tgt_io_thread_done);
1675 /**
1676  * Helper function for getting server side [start, start+count] DLM lock
1677  * if asked by client.
1678  */
1679 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1680                     __u64 start, __u64 end, struct lustre_handle *lh,
1681                     int mode, __u64 *flags)
1682 {
1683         union ldlm_policy_data policy;
1684         int rc;
1685
1686         ENTRY;
1687
1688         LASSERT(lh != NULL);
1689         LASSERT(ns != NULL);
1690         LASSERT(!lustre_handle_is_used(lh));
1691
1692         policy.l_extent.gid = 0;
1693         policy.l_extent.start = start & PAGE_MASK;
1694
1695         /*
1696          * If ->o_blocks is EOF it means "lock till the end of the file".
1697          * Otherwise, it's size of an extent or hole being punched (in bytes).
1698          */
1699         if (end == OBD_OBJECT_EOF || end < start)
1700                 policy.l_extent.end = OBD_OBJECT_EOF;
1701         else
1702                 policy.l_extent.end = end | ~PAGE_MASK;
1703
1704         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1705                                     flags, ldlm_blocking_ast,
1706                                     ldlm_completion_ast, ldlm_glimpse_ast,
1707                                     NULL, 0, LVB_T_NONE, NULL, lh);
1708         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1709 }
1710 EXPORT_SYMBOL(tgt_extent_lock);
1711
1712 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1713 {
1714         LASSERT(lustre_handle_is_used(lh));
1715         ldlm_lock_decref(lh, mode);
1716 }
1717 EXPORT_SYMBOL(tgt_extent_unlock);
1718
1719 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1720                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1721                  struct lustre_handle *lh, enum ldlm_mode mode)
1722 {
1723         __u64                    flags = 0;
1724         int                      nrbufs = obj->ioo_bufcnt;
1725         int                      i;
1726
1727         ENTRY;
1728
1729         LASSERT(mode == LCK_PR || mode == LCK_PW);
1730         LASSERT(!lustre_handle_is_used(lh));
1731
1732         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1733                 RETURN(0);
1734
1735         for (i = 1; i < nrbufs; i++)
1736                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1737                         RETURN(-EFAULT);
1738
1739         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1740                                nb[nrbufs - 1].rnb_offset +
1741                                nb[nrbufs - 1].rnb_len - 1,
1742                                lh, mode, &flags));
1743 }
1744
1745 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1746                     struct lustre_handle *lh, enum ldlm_mode mode)
1747 {
1748         ENTRY;
1749
1750         LASSERT(mode == LCK_PR || mode == LCK_PW);
1751         LASSERT((obj->ioo_bufcnt > 0 &&
1752                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1753                 lustre_handle_is_used(lh));
1754
1755         if (lustre_handle_is_used(lh))
1756                 tgt_extent_unlock(lh, mode);
1757         EXIT;
1758 }
1759
1760 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1761                                struct ptlrpc_bulk_desc *desc, int opc,
1762                                cksum_type_t cksum_type)
1763 {
1764         struct cfs_crypto_hash_desc     *hdesc;
1765         unsigned int                    bufsize;
1766         int                             i, err;
1767         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1768         __u32                           cksum;
1769
1770         LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1771
1772         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1773         if (IS_ERR(hdesc)) {
1774                 CERROR("%s: unable to initialize checksum hash %s\n",
1775                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1776                 return PTR_ERR(hdesc);
1777         }
1778
1779         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1780         for (i = 0; i < desc->bd_iov_count; i++) {
1781                 /* corrupt the data before we compute the checksum, to
1782                  * simulate a client->OST data error */
1783                 if (i == 0 && opc == OST_WRITE &&
1784                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1785                         int off = BD_GET_KIOV(desc, i).kiov_offset &
1786                                 ~PAGE_MASK;
1787                         int len = BD_GET_KIOV(desc, i).kiov_len;
1788                         struct page *np = tgt_page_to_corrupt;
1789                         char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1790
1791                         if (np) {
1792                                 char *ptr2 = kmap(np) + off;
1793
1794                                 memcpy(ptr2, ptr, len);
1795                                 memcpy(ptr2, "bad3", min(4, len));
1796                                 kunmap(np);
1797                                 BD_GET_KIOV(desc, i).kiov_page = np;
1798                         } else {
1799                                 CERROR("%s: can't alloc page for corruption\n",
1800                                        tgt_name(tgt));
1801                         }
1802                 }
1803                 cfs_crypto_hash_update_page(hdesc,
1804                                   BD_GET_KIOV(desc, i).kiov_page,
1805                                   BD_GET_KIOV(desc, i).kiov_offset &
1806                                         ~PAGE_MASK,
1807                                   BD_GET_KIOV(desc, i).kiov_len);
1808
1809                  /* corrupt the data after we compute the checksum, to
1810                  * simulate an OST->client data error */
1811                 if (i == 0 && opc == OST_READ &&
1812                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1813                         int off = BD_GET_KIOV(desc, i).kiov_offset
1814                           & ~PAGE_MASK;
1815                         int len = BD_GET_KIOV(desc, i).kiov_len;
1816                         struct page *np = tgt_page_to_corrupt;
1817                         char *ptr =
1818                           kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1819
1820                         if (np) {
1821                                 char *ptr2 = kmap(np) + off;
1822
1823                                 memcpy(ptr2, ptr, len);
1824                                 memcpy(ptr2, "bad4", min(4, len));
1825                                 kunmap(np);
1826                                 BD_GET_KIOV(desc, i).kiov_page = np;
1827                         } else {
1828                                 CERROR("%s: can't alloc page for corruption\n",
1829                                        tgt_name(tgt));
1830                         }
1831                 }
1832         }
1833
1834         bufsize = sizeof(cksum);
1835         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1836
1837         return cksum;
1838 }
1839
1840 int tgt_brw_read(struct tgt_session_info *tsi)
1841 {
1842         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1843         struct ptlrpc_bulk_desc *desc = NULL;
1844         struct obd_export       *exp = tsi->tsi_exp;
1845         struct niobuf_remote    *remote_nb;
1846         struct niobuf_local     *local_nb;
1847         struct obd_ioobj        *ioo;
1848         struct ost_body         *body, *repbody;
1849         struct l_wait_info       lwi;
1850         struct lustre_handle     lockh = { 0 };
1851         int                      npages, nob = 0, rc, i, no_reply = 0;
1852         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1853
1854         ENTRY;
1855
1856         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1857                 CERROR("%s: deny read request from %s to portal %u\n",
1858                        tgt_name(tsi->tsi_tgt),
1859                        obd_export_nid2str(req->rq_export),
1860                        ptlrpc_req2svc(req)->srv_req_portal);
1861                 RETURN(-EPROTO);
1862         }
1863
1864         req->rq_bulk_read = 1;
1865
1866         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1867                 RETURN(-EIO);
1868
1869         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1870                          cfs_fail_val : (obd_timeout + 1) / 4);
1871
1872         /* Check if there is eviction in progress, and if so, wait for it to
1873          * finish */
1874         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1875                 /* We do not care how long it takes */
1876                 lwi = LWI_INTR(NULL, NULL);
1877                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1878                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1879                          &lwi);
1880         }
1881
1882         /* There must be big cache in current thread to process this request
1883          * if it is NULL then something went wrong and it wasn't allocated,
1884          * report -ENOMEM in that case */
1885         if (tbc == NULL)
1886                 RETURN(-ENOMEM);
1887
1888         body = tsi->tsi_ost_body;
1889         LASSERT(body != NULL);
1890
1891         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1892         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1893
1894         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1895         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1896
1897         local_nb = tbc->local;
1898
1899         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1900                           remote_nb, &lockh, LCK_PR);
1901         if (rc != 0)
1902                 RETURN(rc);
1903
1904         /*
1905          * If getting the lock took more time than
1906          * client was willing to wait, drop it. b=11330
1907          */
1908         if (cfs_time_current_sec() > req->rq_deadline ||
1909             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1910                 no_reply = 1;
1911                 CERROR("Dropping timed-out read from %s because locking"
1912                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1913                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1914                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1915                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1916                 GOTO(out_lock, rc = -ETIMEDOUT);
1917         }
1918
1919         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1920         repbody->oa = body->oa;
1921
1922         npages = PTLRPC_MAX_BRW_PAGES;
1923         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1924                         ioo, remote_nb, &npages, local_nb);
1925         if (rc != 0)
1926                 GOTO(out_lock, rc);
1927
1928         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1929                                     PTLRPC_BULK_PUT_SOURCE |
1930                                         PTLRPC_BULK_BUF_KIOV,
1931                                     OST_BULK_PORTAL,
1932                                     &ptlrpc_bulk_kiov_nopin_ops);
1933         if (desc == NULL)
1934                 GOTO(out_commitrw, rc = -ENOMEM);
1935
1936         nob = 0;
1937         for (i = 0; i < npages; i++) {
1938                 int page_rc = local_nb[i].lnb_rc;
1939
1940                 if (page_rc < 0) {
1941                         rc = page_rc;
1942                         break;
1943                 }
1944
1945                 nob += page_rc;
1946                 if (page_rc != 0) { /* some data! */
1947                         LASSERT(local_nb[i].lnb_page != NULL);
1948                         desc->bd_frag_ops->add_kiov_frag
1949                           (desc, local_nb[i].lnb_page,
1950                            local_nb[i].lnb_page_offset,
1951                            page_rc);
1952                 }
1953
1954                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1955                         /* All subsequent pages should be 0 */
1956                         while (++i < npages)
1957                                 LASSERT(local_nb[i].lnb_rc == 0);
1958                         break;
1959                 }
1960         }
1961         if (OBD_FAIL_CHECK(OBD_FAIL_OST_READ_SIZE) &&
1962             nob != cfs_fail_val)
1963                 rc = -E2BIG;
1964
1965         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1966                 cksum_type_t cksum_type =
1967                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1968                                           body->oa.o_flags : 0);
1969                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1970                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1971                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1972                                                         OST_READ, cksum_type);
1973                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1974                        repbody->oa.o_cksum);
1975         } else {
1976                 repbody->oa.o_valid = 0;
1977         }
1978         /* We're finishing using body->oa as an input variable */
1979
1980         /* Check if client was evicted while we were doing i/o before touching
1981          * network */
1982         if (likely(rc == 0 &&
1983                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1984                 rc = target_bulk_io(exp, desc, &lwi);
1985                 no_reply = rc != 0;
1986         }
1987
1988 out_commitrw:
1989         /* Must commit after prep above in all cases */
1990         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1991                           remote_nb, npages, local_nb, rc);
1992         if (rc == 0)
1993                 tgt_drop_id(exp, &repbody->oa);
1994 out_lock:
1995         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1996
1997         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1998                 ptlrpc_free_bulk(desc);
1999
2000         LASSERT(rc <= 0);
2001         if (rc == 0) {
2002                 rc = nob;
2003                 ptlrpc_lprocfs_brw(req, nob);
2004         } else if (no_reply) {
2005                 req->rq_no_reply = 1;
2006                 /* reply out callback would free */
2007                 ptlrpc_req_drop_rs(req);
2008                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
2009                               "client will retry: rc %d\n",
2010                               exp->exp_obd->obd_name,
2011                               obd_uuid2str(&exp->exp_client_uuid),
2012                               obd_export_nid2str(exp), rc);
2013         }
2014         /* send a bulk after reply to simulate a network delay or reordering
2015          * by a router */
2016         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
2017                 wait_queue_head_t        waitq;
2018                 struct l_wait_info       lwi1;
2019
2020                 CDEBUG(D_INFO, "reorder BULK\n");
2021                 init_waitqueue_head(&waitq);
2022
2023                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
2024                 l_wait_event(waitq, 0, &lwi1);
2025                 target_bulk_io(exp, desc, &lwi);
2026                 ptlrpc_free_bulk(desc);
2027         }
2028
2029         RETURN(rc);
2030 }
2031 EXPORT_SYMBOL(tgt_brw_read);
2032
2033 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
2034                               struct ptlrpc_bulk_desc *desc,
2035                               struct niobuf_local *local_nb, int npages,
2036                               u32 client_cksum, u32 server_cksum,
2037                               bool mmap)
2038 {
2039         struct obd_export *exp = req->rq_export;
2040         struct ost_body *body;
2041         char *router = "";
2042         char *via = "";
2043
2044         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2045         LASSERT(body != NULL);
2046
2047         if (req->rq_peer.nid != desc->bd_sender) {
2048                 via = " via ";
2049                 router = libcfs_nid2str(desc->bd_sender);
2050         }
2051
2052         if (mmap) {
2053                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
2054                              client_cksum, server_cksum);
2055                 return;
2056         }
2057
2058         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
2059                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
2060                            "]: client csum %x, server csum %x\n",
2061                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
2062                            via, router,
2063                            body->oa.o_valid & OBD_MD_FLFID ?
2064                            body->oa.o_parent_seq : (__u64)0,
2065                            body->oa.o_valid & OBD_MD_FLFID ?
2066                            body->oa.o_parent_oid : 0,
2067                            body->oa.o_valid & OBD_MD_FLFID ?
2068                            body->oa.o_parent_ver : 0,
2069                            POSTID(&body->oa.o_oi),
2070                            local_nb[0].lnb_file_offset,
2071                            local_nb[npages-1].lnb_file_offset +
2072                            local_nb[npages - 1].lnb_len - 1,
2073                            client_cksum, server_cksum);
2074 }
2075
2076 int tgt_brw_write(struct tgt_session_info *tsi)
2077 {
2078         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2079         struct ptlrpc_bulk_desc *desc = NULL;
2080         struct obd_export       *exp = req->rq_export;
2081         struct niobuf_remote    *remote_nb;
2082         struct niobuf_local     *local_nb;
2083         struct obd_ioobj        *ioo;
2084         struct ost_body         *body, *repbody;
2085         struct l_wait_info       lwi;
2086         struct lustre_handle     lockh = {0};
2087         __u32                   *rcs;
2088         int                      objcount, niocount, npages;
2089         int                      rc, i, j;
2090         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
2091         bool                     no_reply = false, mmap;
2092         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2093
2094         ENTRY;
2095
2096         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2097                 CERROR("%s: deny write request from %s to portal %u\n",
2098                        tgt_name(tsi->tsi_tgt),
2099                        obd_export_nid2str(req->rq_export),
2100                        ptlrpc_req2svc(req)->srv_req_portal);
2101                 RETURN(err_serious(-EPROTO));
2102         }
2103
2104         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2105                 RETURN(err_serious(-ENOSPC));
2106         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2107                 RETURN(err_serious(-EROFS));
2108
2109         req->rq_bulk_write = 1;
2110
2111         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2112                 RETURN(err_serious(-EIO));
2113         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2114                 RETURN(err_serious(-EFAULT));
2115
2116         /* pause before transaction has been started */
2117         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2118                          cfs_fail_val : (obd_timeout + 1) / 4);
2119
2120         /* There must be big cache in current thread to process this request
2121          * if it is NULL then something went wrong and it wasn't allocated,
2122          * report -ENOMEM in that case */
2123         if (tbc == NULL)
2124                 RETURN(-ENOMEM);
2125
2126         body = tsi->tsi_ost_body;
2127         LASSERT(body != NULL);
2128
2129         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2130         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2131
2132         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2133                                         RCL_CLIENT) / sizeof(*ioo);
2134
2135         for (niocount = i = 0; i < objcount; i++)
2136                 niocount += ioo[i].ioo_bufcnt;
2137
2138         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2139         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2140         if (niocount != req_capsule_get_size(&req->rq_pill,
2141                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2142                         sizeof(*remote_nb))
2143                 RETURN(err_serious(-EPROTO));
2144
2145         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2146             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2147                 memory_pressure_set();
2148
2149         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2150                              niocount * sizeof(*rcs));
2151         rc = req_capsule_server_pack(&req->rq_pill);
2152         if (rc != 0)
2153                 GOTO(out, rc = err_serious(rc));
2154
2155         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2156         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2157
2158         local_nb = tbc->local;
2159
2160         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2161                           remote_nb, &lockh, LCK_PW);
2162         if (rc != 0)
2163                 GOTO(out, rc);
2164
2165         /*
2166          * If getting the lock took more time than
2167          * client was willing to wait, drop it. b=11330
2168          */
2169         if (cfs_time_current_sec() > req->rq_deadline ||
2170             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2171                 no_reply = true;
2172                 CERROR("%s: Dropping timed-out write from %s because locking "
2173                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
2174                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2175                        POSTID(&ioo->ioo_oid),
2176                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2177                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2178                 GOTO(out_lock, rc = -ETIMEDOUT);
2179         }
2180
2181         /* Because we already sync grant info with client when reconnect,
2182          * grant info will be cleared for resent req, then fed_grant and
2183          * total_grant will not be modified in following preprw_write */
2184         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2185                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2186                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2187         }
2188
2189         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2190         if (repbody == NULL)
2191                 GOTO(out_lock, rc = -ENOMEM);
2192         repbody->oa = body->oa;
2193
2194         npages = PTLRPC_MAX_BRW_PAGES;
2195         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2196                         objcount, ioo, remote_nb, &npages, local_nb);
2197         if (rc < 0)
2198                 GOTO(out_lock, rc);
2199
2200         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2201                                     PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2202                                     OST_BULK_PORTAL,
2203                                     &ptlrpc_bulk_kiov_nopin_ops);
2204         if (desc == NULL)
2205                 GOTO(skip_transfer, rc = -ENOMEM);
2206
2207         /* NB Having prepped, we must commit... */
2208         for (i = 0; i < npages; i++)
2209                 desc->bd_frag_ops->add_kiov_frag(desc,
2210                                                  local_nb[i].lnb_page,
2211                                                  local_nb[i].lnb_page_offset,
2212                                                  local_nb[i].lnb_len);
2213
2214         rc = sptlrpc_svc_prep_bulk(req, desc);
2215         if (rc != 0)
2216                 GOTO(skip_transfer, rc);
2217
2218         rc = target_bulk_io(exp, desc, &lwi);
2219         no_reply = rc != 0;
2220
2221 skip_transfer:
2222         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2223                 static int cksum_counter;
2224
2225                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2226                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2227
2228                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2229                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2230                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2231                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2232                                                         OST_WRITE, cksum_type);
2233                 cksum_counter++;
2234
2235                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2236                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2237                                 body->oa.o_flags & OBD_FL_MMAP);
2238
2239                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2240                                           body->oa.o_cksum,
2241                                           repbody->oa.o_cksum, mmap);
2242                         cksum_counter = 0;
2243                 } else if ((cksum_counter & (-cksum_counter)) ==
2244                            cksum_counter) {
2245                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2246                                cksum_counter, libcfs_id2str(req->rq_peer),
2247                                repbody->oa.o_cksum);
2248                 }
2249         }
2250
2251         /* Must commit after prep above in all cases */
2252         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2253                           objcount, ioo, remote_nb, npages, local_nb, rc);
2254         if (rc == -ENOTCONN)
2255                 /* quota acquire process has been given up because
2256                  * either the client has been evicted or the client
2257                  * has timed out the request already */
2258                 no_reply = true;
2259
2260         /*
2261          * Disable sending mtime back to the client. If the client locked the
2262          * whole object, then it has already updated the mtime on its side,
2263          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2264          */
2265         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2266
2267         if (rc == 0) {
2268                 int nob = 0;
2269
2270                 /* set per-requested niobuf return codes */
2271                 for (i = j = 0; i < niocount; i++) {
2272                         int len = remote_nb[i].rnb_len;
2273
2274                         nob += len;
2275                         rcs[i] = 0;
2276                         do {
2277                                 LASSERT(j < npages);
2278                                 if (local_nb[j].lnb_rc < 0)
2279                                         rcs[i] = local_nb[j].lnb_rc;
2280                                 len -= local_nb[j].lnb_len;
2281                                 j++;
2282                         } while (len > 0);
2283                         LASSERT(len == 0);
2284                 }
2285                 LASSERT(j == npages);
2286                 ptlrpc_lprocfs_brw(req, nob);
2287
2288                 tgt_drop_id(exp, &repbody->oa);
2289         }
2290 out_lock:
2291         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2292         if (desc)
2293                 ptlrpc_free_bulk(desc);
2294 out:
2295         if (no_reply) {
2296                 req->rq_no_reply = 1;
2297                 /* reply out callback would free */
2298                 ptlrpc_req_drop_rs(req);
2299                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2300                               "client will retry: rc %d\n",
2301                               exp->exp_obd->obd_name,
2302                               obd_uuid2str(&exp->exp_client_uuid),
2303                               obd_export_nid2str(exp), rc);
2304         }
2305         memory_pressure_clr();
2306         RETURN(rc);
2307 }
2308 EXPORT_SYMBOL(tgt_brw_write);
2309
2310 /* Check if request can be reconstructed from saved reply data
2311  * A copy of the reply data is returned in @trd if the pointer is not NULL
2312  */
2313 bool req_can_reconstruct(struct ptlrpc_request *req,
2314                          struct tg_reply_data *trd)
2315 {
2316         struct tg_export_data *ted = &req->rq_export->exp_target_data;
2317         struct lsd_client_data *lcd = ted->ted_lcd;
2318         bool found;
2319
2320         if (tgt_is_multimodrpcs_client(req->rq_export))
2321                 return tgt_lookup_reply(req, trd);
2322
2323         mutex_lock(&ted->ted_lcd_lock);
2324         found = req->rq_xid == lcd->lcd_last_xid ||
2325                 req->rq_xid == lcd->lcd_last_close_xid;
2326
2327         if (found && trd != NULL) {
2328                 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2329                         trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2330                         trd->trd_reply.lrd_transno =
2331                                                 lcd->lcd_last_close_transno;
2332                         trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2333                 } else {
2334                         trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2335                         trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2336                         trd->trd_reply.lrd_result = lcd->lcd_last_result;
2337                         trd->trd_reply.lrd_data = lcd->lcd_last_data;
2338                         trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2339                         trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2340                         trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2341                         trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2342                 }
2343         }
2344         mutex_unlock(&ted->ted_lcd_lock);
2345
2346         return found;
2347 }
2348 EXPORT_SYMBOL(req_can_reconstruct);
2349