lustre/target/tgt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2011, 2014, Intel Corporation.
  25  */
  26 /*
  27  * lustre/target/tgt_handler.c
  28  *
  29  * Lustre Unified Target request handler code
  30  *
  31  * Author: Brian Behlendorf <behlendorf1@llnl.gov>
  32  * Author: Mikhail Pershin <mike.pershin@intel.com>
  33  */
  34
  35 #define DEBUG_SUBSYSTEM S_CLASS
  36
  37 #include <obd.h>
  38 #include <obd_class.h>
  39 #include <obd_cksum.h>
  40 #include <md_object.h>
  41 #include <lustre_lfsck.h>
  42 #include <lustre_nodemap.h>
  43
  44 #include "tgt_internal.h"
  45
  46 char *tgt_name(struct lu_target *tgt)
  47 {
  48         LASSERT(tgt->lut_obd != NULL);
  49         return tgt->lut_obd->obd_name;
  50 }
  51 EXPORT_SYMBOL(tgt_name);
  52
  53 /*
  54  * Generic code handling requests that have struct mdt_body passed in:
  55  *
  56  *  - extract mdt_body from request and save it in @tsi, if present;
  57  *
  58  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  59  *  @tsi;
  60  *
  61  *  - if HABEO_CORPUS flag is set for this request type check whether object
  62  *  actually exists on storage (lu_object_exists()).
  63  *
  64  */
  65 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  66 {
  67         const struct mdt_body   *body;
  68         struct lu_object        *obj;
  69         struct req_capsule      *pill = tsi->tsi_pill;
  70         int                      rc;
  71
  72         ENTRY;
  73
  74         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  75         if (body == NULL)
  76                 RETURN(-EFAULT);
  77
  78         tsi->tsi_mdt_body = body;
  79
  80         if (!(body->mbo_valid & OBD_MD_FLID))
  81                 RETURN(0);
  82
  83         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  84          * in any case in pre-2.5 clients. Fix that here if needed */
  85         if (unlikely(fid_is_zero(&body->mbo_fid1)))
  86                 RETURN(0);
  87
  88         if (!fid_is_sane(&body->mbo_fid1)) {
  89                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  90                        PFID(&body->mbo_fid1));
  91                 RETURN(-EINVAL);
  92         }
  93
  94         obj = lu_object_find(tsi->tsi_env,
  95                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
  96                              &body->mbo_fid1, NULL);
  97         if (!IS_ERR(obj)) {
  98                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
  99                         lu_object_put(tsi->tsi_env, obj);
 100                         /* for capability renew ENOENT will be handled in
 101                          * mdt_renew_capa */
 102                         if (body->mbo_valid & OBD_MD_FLOSSCAPA)
 103                                 rc = 0;
 104                         else
 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 lustre_capa      *capa;
 246         struct lu_nodemap       *nodemap;
 247         int                      rc;
 248
 249         ENTRY;
 250
 251         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 252         if (body == NULL)
 253                 RETURN(-EFAULT);
 254
 255         rc = tgt_validate_obdo(tsi, &body->oa);
 256         if (rc)
 257                 RETURN(rc);
 258
 259         nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
 260
 261         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 262                                         NODEMAP_CLIENT_TO_FS,
 263                                         body->oa.o_uid);
 264         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 265                                         NODEMAP_CLIENT_TO_FS,
 266                                         body->oa.o_gid);
 267
 268         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 269                 capa = req_capsule_client_get(pill, &RMF_CAPA1);
 270                 if (capa == NULL) {
 271                         CERROR("%s: OSSCAPA flag is set without capability\n",
 272                                tgt_name(tsi->tsi_tgt));
 273                         RETURN(-EFAULT);
 274                 }
 275         }
 276
 277         tsi->tsi_ost_body = body;
 278         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 279
 280         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 281                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 282                 if (rc < 0)
 283                         RETURN(rc);
 284         }
 285
 286         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 287                 if (flags & HABEO_CORPUS) {
 288                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 289                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 290                                tgt_name(tsi->tsi_tgt));
 291                         RETURN(-EPROTO);
 292                 } else {
 293                         RETURN(0);
 294                 }
 295         }
 296
 297         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 298
 299         /*
 300          * OST doesn't get object in advance for further use to prevent
 301          * situations with nested object_find which is potential deadlock.
 302          */
 303         tsi->tsi_corpus = NULL;
 304         RETURN(rc);
 305 }
 306
 307 /*
 308  * Do necessary preprocessing according to handler ->th_flags.
 309  */
 310 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 311                                   struct tgt_handler *h,
 312                                   struct ptlrpc_request *req)
 313 {
 314         struct req_capsule      *pill = tsi->tsi_pill;
 315         __u32                    flags = h->th_flags;
 316         int                      rc = 0;
 317
 318         ENTRY;
 319
 320         if (tsi->tsi_preprocessed)
 321                 RETURN(0);
 322
 323         LASSERT(h->th_act != NULL);
 324         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 325         LASSERT(current->journal_info == NULL);
 326
 327         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 328                      h->th_fmt != NULL));
 329         if (h->th_fmt != NULL) {
 330                 req_capsule_set(pill, h->th_fmt);
 331                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 332                         rc = tgt_mdt_body_unpack(tsi, flags);
 333                         if (rc < 0)
 334                                 RETURN(rc);
 335                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 336                                                  RCL_CLIENT)) {
 337                         rc = tgt_ost_body_unpack(tsi, flags);
 338                         if (rc < 0)
 339                                 RETURN(rc);
 340                 }
 341         }
 342
 343         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 344                 RETURN(-EROFS);
 345
 346         if (flags & HABEO_CLAVIS) {
 347                 struct ldlm_request *dlm_req;
 348
 349                 LASSERT(h->th_fmt != NULL);
 350
 351                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 352                 if (dlm_req != NULL) {
 353                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 354                                      LDLM_IBITS &&
 355                                      dlm_req->lock_desc.l_policy_data.\
 356                                      l_inodebits.bits == 0)) {
 357                                 /*
 358                                  * Lock without inodebits makes no sense and
 359                                  * will oops later in ldlm. If client miss to
 360                                  * set such bits, do not trigger ASSERTION.
 361                                  *
 362                                  * For liblustre flock case, it maybe zero.
 363                                  */
 364                                 rc = -EPROTO;
 365                         } else {
 366                                 tsi->tsi_dlm_req = dlm_req;
 367                         }
 368                 } else {
 369                         rc = -EFAULT;
 370                 }
 371         }
 372         tsi->tsi_preprocessed = 1;
 373         RETURN(rc);
 374 }
 375
 376 /*
 377  * Invoke handler for this request opc. Also do necessary preprocessing
 378  * (according to handler ->th_flags), and post-processing (setting of
 379  * ->last_{xid,committed}).
 380  */
 381 static int tgt_handle_request0(struct tgt_session_info *tsi,
 382                                struct tgt_handler *h,
 383                                struct ptlrpc_request *req)
 384 {
 385         int      serious = 0;
 386         int      rc;
 387
 388         ENTRY;
 389
 390         /*
 391          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 392          * to put same checks into handlers like mdt_close(), mdt_reint(),
 393          * etc., without talking to mdt authors first. Checking same thing
 394          * there again is useless and returning 0 error without packing reply
 395          * is buggy! Handlers either pack reply or return error.
 396          *
 397          * We return 0 here and do not send any reply in order to emulate
 398          * network failure. Do not send any reply in case any of NET related
 399          * fail_id has occured.
 400          */
 401         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 402                 RETURN(0);
 403
 404         rc = tgt_request_preprocess(tsi, h, req);
 405         /* pack reply if reply format is fixed */
 406         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 407                 /* Pack reply */
 408                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 409                                           RCL_SERVER))
 410                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 411                                              RCL_SERVER,
 412                                              tsi->tsi_mdt_body->mbo_eadatasize);
 413                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 414                                           RCL_SERVER))
 415                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 416                                              RCL_SERVER, 0);
 417
 418                 rc = req_capsule_server_pack(tsi->tsi_pill);
 419         }
 420
 421         if (likely(rc == 0)) {
 422                 /*
 423                  * Process request, there can be two types of rc:
 424                  * 1) errors with msg unpack/pack, other failures outside the
 425                  * operation itself. This is counted as serious errors;
 426                  * 2) errors during fs operation, should be placed in rq_status
 427                  * only
 428                  */
 429                 rc = h->th_act(tsi);
 430                 if (!is_serious(rc) &&
 431                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 432                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 433                                   "pack reply and returned 0 error\n",
 434                                   tgt_name(tsi->tsi_tgt), h->th_name);
 435                         LBUG();
 436                 }
 437                 serious = is_serious(rc);
 438                 rc = clear_serious(rc);
 439         } else {
 440                 serious = 1;
 441         }
 442
 443         req->rq_status = rc;
 444
 445         /*
 446          * ELDLM_* codes which > 0 should be in rq_status only as well as
 447          * all non-serious errors.
 448          */
 449         if (rc > 0 || !serious)
 450                 rc = 0;
 451
 452         LASSERT(current->journal_info == NULL);
 453
 454         if (likely(rc == 0 && req->rq_export))
 455                 target_committed_to_req(req);
 456
 457         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 458         RETURN(0);
 459 }
 460
 461 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 462                                        struct obd_device *obd, int *process)
 463 {
 464         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 465         case MDS_DISCONNECT:
 466         case OST_DISCONNECT:
 467         case OBD_IDX_READ:
 468                 *process = 1;
 469                 RETURN(0);
 470         case MDS_CLOSE:
 471         case MDS_DONE_WRITING:
 472         case MDS_SYNC: /* used in unmounting */
 473         case OBD_PING:
 474         case MDS_REINT:
 475         case OUT_UPDATE:
 476         case SEQ_QUERY:
 477         case FLD_QUERY:
 478         case FLD_READ:
 479         case LDLM_ENQUEUE:
 480         case OST_CREATE:
 481         case OST_DESTROY:
 482         case OST_PUNCH:
 483         case OST_SETATTR:
 484         case OST_SYNC:
 485         case OST_WRITE:
 486                 *process = target_queue_recovery_request(req, obd);
 487                 RETURN(0);
 488
 489         default:
 490                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 491                 *process = -EAGAIN;
 492                 RETURN(0);
 493         }
 494 }
 495
 496 /*
 497  * Handle recovery. Return:
 498  *        +1: continue request processing;
 499  *       -ve: abort immediately with the given error code;
 500  *         0: send reply with error code in req->rq_status;
 501  */
 502 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 503 {
 504         ENTRY;
 505
 506         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 507         case MDS_CONNECT:
 508         case OST_CONNECT:
 509         case MGS_CONNECT:
 510         case SEC_CTX_INIT:
 511         case SEC_CTX_INIT_CONT:
 512         case SEC_CTX_FINI:
 513                 RETURN(+1);
 514         }
 515
 516         if (!req->rq_export->exp_obd->obd_replayable)
 517                 RETURN(+1);
 518
 519         /* sanity check: if the xid matches, the request must be marked as a
 520          * resent or replayed */
 521         if (req_xid_is_last(req)) {
 522                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 523                       (MSG_RESENT | MSG_REPLAY))) {
 524                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 525                                   "last_xid, expected REPLAY or RESENT flag "
 526                                   "(%x)", req->rq_xid,
 527                                   lustre_msg_get_flags(req->rq_reqmsg));
 528                         req->rq_status = -ENOTCONN;
 529                         RETURN(-ENOTCONN);
 530                 }
 531         }
 532         /* else: note the opposite is not always true; a RESENT req after a
 533          * failover will usually not match the last_xid, since it was likely
 534          * never committed. A REPLAYed request will almost never match the
 535          * last xid, however it could for a committed, but still retained,
 536          * open. */
 537
 538         /* Check for aborted recovery... */
 539         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 540                 int rc;
 541                 int should_process;
 542
 543                 DEBUG_REQ(D_INFO, req, "Got new replay");
 544                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 545                                                  &should_process);
 546                 if (rc != 0 || !should_process)
 547                         RETURN(rc);
 548                 else if (should_process < 0) {
 549                         req->rq_status = should_process;
 550                         rc = ptlrpc_error(req);
 551                         RETURN(rc);
 552                 }
 553         }
 554         RETURN(+1);
 555 }
 556
 557 /* Initial check for request, it is validation mostly */
 558 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 559 {
 560         struct tgt_handler      *h;
 561         struct tgt_opc_slice    *s;
 562         struct lu_target        *tgt;
 563         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 564
 565         ENTRY;
 566
 567         tgt = class_exp2tgt(req->rq_export);
 568
 569         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 570                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 571                         break;
 572
 573         /* opcode was not found in slice */
 574         if (unlikely(s->tos_hs == NULL)) {
 575                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 576                        opc);
 577                 RETURN(ERR_PTR(-ENOTSUPP));
 578         }
 579
 580         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 581         h = s->tos_hs + (opc - s->tos_opc_start);
 582         if (unlikely(h->th_opc == 0)) {
 583                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 584                 RETURN(ERR_PTR(-ENOTSUPP));
 585         }
 586
 587         RETURN(h);
 588 }
 589
 590 int tgt_request_handle(struct ptlrpc_request *req)
 591 {
 592         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 593
 594         struct lustre_msg       *msg = req->rq_reqmsg;
 595         struct tgt_handler      *h;
 596         struct lu_target        *tgt;
 597         int                      request_fail_id = 0;
 598         __u32                    opc = lustre_msg_get_opc(msg);
 599         int                      rc;
 600
 601         ENTRY;
 602
 603         /* Refill the context, to make sure all thread keys are allocated */
 604         lu_env_refill(req->rq_svc_thread->t_env);
 605
 606         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 607         tsi->tsi_pill = &req->rq_pill;
 608         tsi->tsi_env = req->rq_svc_thread->t_env;
 609
 610         /* if request has export then get handlers slice from corresponding
 611          * target, otherwise that should be connect operation */
 612         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 613             opc == MGS_CONNECT) {
 614                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 615                 rc = target_handle_connect(req);
 616                 if (rc != 0) {
 617                         rc = ptlrpc_error(req);
 618                         GOTO(out, rc);
 619                 }
 620                 /* recovery-small test 18c asks to drop connect reply */
 621                 if (unlikely(opc == OST_CONNECT &&
 622                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 623                         GOTO(out, rc = 0);
 624         }
 625
 626         if (unlikely(!class_connected_export(req->rq_export))) {
 627                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 628                        opc, libcfs_id2str(req->rq_peer));
 629                 req->rq_status = -ENOTCONN;
 630                 rc = ptlrpc_error(req);
 631                 GOTO(out, rc);
 632         }
 633
 634         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 635         tsi->tsi_exp = req->rq_export;
 636         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 637                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 638         else
 639                 tsi->tsi_jobid = NULL;
 640
 641         request_fail_id = tgt->lut_request_fail_id;
 642         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 643
 644         h = tgt_handler_find_check(req);
 645         if (IS_ERR(h)) {
 646                 req->rq_status = PTR_ERR(h);
 647                 rc = ptlrpc_error(req);
 648                 GOTO(out, rc);
 649         }
 650
 651         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 652                 GOTO(out, rc = 0);
 653
 654         rc = lustre_msg_check_version(msg, h->th_version);
 655         if (unlikely(rc)) {
 656                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 657                           " %08x, expecting %08x\n", tgt_name(tgt),
 658                           lustre_msg_get_version(msg), h->th_version);
 659                 req->rq_status = -EINVAL;
 660                 rc = ptlrpc_error(req);
 661                 GOTO(out, rc);
 662         }
 663
 664         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 665         if (likely(rc == 1)) {
 666                 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 667                          h->th_opc, opc);
 668                 rc = tgt_handle_request0(tsi, h, req);
 669                 if (rc)
 670                         GOTO(out, rc);
 671         }
 672         EXIT;
 673 out:
 674         req_capsule_fini(tsi->tsi_pill);
 675         if (tsi->tsi_corpus != NULL) {
 676                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 677                 tsi->tsi_corpus = NULL;
 678         }
 679         return rc;
 680 }
 681 EXPORT_SYMBOL(tgt_request_handle);
 682
 683 /** Assign high priority operations to the request if needed. */
 684 int tgt_hpreq_handler(struct ptlrpc_request *req)
 685 {
 686         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 687         struct tgt_handler      *h;
 688         int                      rc;
 689
 690         ENTRY;
 691
 692         if (req->rq_export == NULL)
 693                 RETURN(0);
 694
 695         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 696         tsi->tsi_pill = &req->rq_pill;
 697         tsi->tsi_env = req->rq_svc_thread->t_env;
 698         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 699         tsi->tsi_exp = req->rq_export;
 700
 701         h = tgt_handler_find_check(req);
 702         if (IS_ERR(h)) {
 703                 rc = PTR_ERR(h);
 704                 RETURN(rc);
 705         }
 706
 707         rc = tgt_request_preprocess(tsi, h, req);
 708         if (unlikely(rc != 0))
 709                 RETURN(rc);
 710
 711         if (h->th_hp != NULL)
 712                 h->th_hp(tsi);
 713         RETURN(0);
 714 }
 715 EXPORT_SYMBOL(tgt_hpreq_handler);
 716
 717 void tgt_counter_incr(struct obd_export *exp, int opcode)
 718 {
 719         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 720         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 721                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 722 }
 723 EXPORT_SYMBOL(tgt_counter_incr);
 724
 725 /*
 726  * Unified target generic handlers.
 727  */
 728
 729 /*
 730  * Security functions
 731  */
 732 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 733 {
 734         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 735                                       OBD_CONNECT_RMT_CLIENT_FORCE |
 736                                       OBD_CONNECT_MDS_CAPA |
 737                                       OBD_CONNECT_OSS_CAPA);
 738 }
 739
 740 static int tgt_init_sec_level(struct ptlrpc_request *req)
 741 {
 742         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 743         char                    *client = libcfs_nid2str(req->rq_peer.nid);
 744         struct obd_connect_data *data, *reply;
 745         int                      rc = 0;
 746         bool                     remote;
 747
 748         ENTRY;
 749
 750         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 751         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 752         if (data == NULL || reply == NULL)
 753                 RETURN(-EFAULT);
 754
 755         /* connection from MDT is always trusted */
 756         if (req->rq_auth_usr_mdt) {
 757                 tgt_init_sec_none(reply);
 758                 RETURN(0);
 759         }
 760
 761         /* no GSS support case */
 762         if (!req->rq_auth_gss) {
 763                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 764                         CWARN("client %s -> target %s does not use GSS, "
 765                               "can not run under security level %d.\n",
 766                               client, tgt_name(tgt), tgt->lut_sec_level);
 767                         RETURN(-EACCES);
 768                 } else {
 769                         tgt_init_sec_none(reply);
 770                         RETURN(0);
 771                 }
 772         }
 773
 774         /* old version case */
 775         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 776                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 777                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 778                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 779                         CWARN("client %s -> target %s uses old version, "
 780                               "can not run under security level %d.\n",
 781                               client, tgt_name(tgt), tgt->lut_sec_level);
 782                         RETURN(-EACCES);
 783                 } else {
 784                         CWARN("client %s -> target %s uses old version, "
 785                               "run under security level %d.\n",
 786                               client, tgt_name(tgt), tgt->lut_sec_level);
 787                         tgt_init_sec_none(reply);
 788                         RETURN(0);
 789                 }
 790         }
 791
 792         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 793         if (remote) {
 794                 if (!req->rq_auth_remote)
 795                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 796                                "asked to be remote.\n", client, tgt_name(tgt));
 797         } else if (req->rq_auth_remote) {
 798                 remote = true;
 799                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 800                        "as remote by default.\n", client, tgt_name(tgt));
 801         }
 802
 803         if (remote) {
 804                 if (!tgt->lut_oss_capa) {
 805                         CDEBUG(D_SEC,
 806                                "client %s -> target %s is set as remote,"
 807                                " but OSS capabilities are not enabled: %d.\n",
 808                                client, tgt_name(tgt), tgt->lut_oss_capa);
 809                         RETURN(-EACCES);
 810                 }
 811         } else {
 812                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 813                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 814                                "authenticated!\n", client, tgt_name(tgt));
 815                         RETURN(-EACCES);
 816                 }
 817         }
 818
 819
 820         switch (tgt->lut_sec_level) {
 821         case LUSTRE_SEC_NONE:
 822                 if (remote) {
 823                         CDEBUG(D_SEC,
 824                                "client %s -> target %s is set as remote, "
 825                                "can not run under security level %d.\n",
 826                                client, tgt_name(tgt), tgt->lut_sec_level);
 827                         RETURN(-EACCES);
 828                 }
 829                 tgt_init_sec_none(reply);
 830                 break;
 831         case LUSTRE_SEC_REMOTE:
 832                 if (!remote)
 833                         tgt_init_sec_none(reply);
 834                 break;
 835         case LUSTRE_SEC_ALL:
 836                 if (remote)
 837                         break;
 838                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 839                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 840                 if (!tgt->lut_oss_capa)
 841                         reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 842                 if (!tgt->lut_mds_capa)
 843                         reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 844                 break;
 845         default:
 846                 RETURN(-EINVAL);
 847         }
 848
 849         RETURN(rc);
 850 }
 851
 852 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 853 {
 854         struct lu_target        *tgt = class_exp2tgt(exp);
 855         struct sptlrpc_flavor    flvr;
 856         int                      rc = 0;
 857
 858         LASSERT(tgt);
 859         LASSERT(tgt->lut_obd);
 860         LASSERT(tgt->lut_slice);
 861
 862         /* always allow ECHO client */
 863         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 864                             LUSTRE_ECHO_NAME) == 0)) {
 865                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 866                 return 0;
 867         }
 868
 869         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 870                 read_lock(&tgt->lut_sptlrpc_lock);
 871                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 872                                              req->rq_sp_from,
 873                                              req->rq_peer.nid,
 874                                              &flvr);
 875                 read_unlock(&tgt->lut_sptlrpc_lock);
 876
 877                 spin_lock(&exp->exp_lock);
 878                 exp->exp_sp_peer = req->rq_sp_from;
 879                 exp->exp_flvr = flvr;
 880                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 881                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 882                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 883                                "expect %x\n", tgt_name(tgt),
 884                                req->rq_flvr.sf_rpc,
 885                                libcfs_nid2str(req->rq_peer.nid),
 886                                exp->exp_flvr.sf_rpc);
 887                         rc = -EACCES;
 888                 }
 889                 spin_unlock(&exp->exp_lock);
 890         } else {
 891                 if (exp->exp_sp_peer != req->rq_sp_from) {
 892                         CERROR("%s: RPC source %s doesn't match %s\n",
 893                                tgt_name(tgt),
 894                                sptlrpc_part2name(req->rq_sp_from),
 895                                sptlrpc_part2name(exp->exp_sp_peer));
 896                         rc = -EACCES;
 897                 } else {
 898                         rc = sptlrpc_target_export_check(exp, req);
 899                 }
 900         }
 901
 902         return rc;
 903 }
 904
 905 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 906 {
 907         struct sptlrpc_rule_set  tmp_rset;
 908         int                      rc;
 909
 910         sptlrpc_rule_set_init(&tmp_rset);
 911         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 912         if (rc) {
 913                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 914                        tgt_name(tgt), rc);
 915                 return rc;
 916         }
 917
 918         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 919
 920         write_lock(&tgt->lut_sptlrpc_lock);
 921         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 922         tgt->lut_sptlrpc_rset = tmp_rset;
 923         write_unlock(&tgt->lut_sptlrpc_lock);
 924
 925         return 0;
 926 }
 927 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 928
 929 int tgt_connect(struct tgt_session_info *tsi)
 930 {
 931         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 932         struct obd_connect_data *reply;
 933         int                      rc;
 934
 935         ENTRY;
 936
 937         rc = tgt_init_sec_level(req);
 938         if (rc != 0)
 939                 GOTO(out, rc);
 940
 941         /* XXX: better to call this check right after getting new export but
 942          * before last_rcvd slot allocation to avoid server load upon insecure
 943          * connects. This is to be fixed after unifiyng all targets.
 944          */
 945         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
 946         if (rc)
 947                 GOTO(out, rc);
 948
 949         /* To avoid exposing partially initialized connection flags, changes up
 950          * to this point have been staged in reply->ocd_connect_flags. Now that
 951          * connection handling has completed successfully, atomically update
 952          * the connect flags in the shared export data structure. LU-1623 */
 953         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
 954         spin_lock(&tsi->tsi_exp->exp_lock);
 955         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
 956         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
 957         spin_unlock(&tsi->tsi_exp->exp_lock);
 958
 959         RETURN(0);
 960 out:
 961         obd_disconnect(class_export_get(tsi->tsi_exp));
 962         return rc;
 963 }
 964 EXPORT_SYMBOL(tgt_connect);
 965
 966 int tgt_disconnect(struct tgt_session_info *tsi)
 967 {
 968         int rc;
 969
 970         ENTRY;
 971
 972         rc = target_handle_disconnect(tgt_ses_req(tsi));
 973         if (rc)
 974                 RETURN(err_serious(rc));
 975
 976         RETURN(rc);
 977 }
 978 EXPORT_SYMBOL(tgt_disconnect);
 979
 980 /*
 981  * Unified target OBD handlers
 982  */
 983 int tgt_obd_ping(struct tgt_session_info *tsi)
 984 {
 985         int rc;
 986
 987         ENTRY;
 988
 989         rc = target_handle_ping(tgt_ses_req(tsi));
 990         if (rc)
 991                 RETURN(err_serious(rc));
 992
 993         RETURN(rc);
 994 }
 995 EXPORT_SYMBOL(tgt_obd_ping);
 996
 997 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
 998 {
 999         return err_serious(-EOPNOTSUPP);
1000 }
1001
1002 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
1003 {
1004         return err_serious(-EOPNOTSUPP);
1005 }
1006
1007 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1008 {
1009         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1010         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1011         struct obd_export       *exp = req->rq_export;
1012         struct ptlrpc_bulk_desc *desc;
1013         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1014         int                      tmpcount;
1015         int                      tmpsize;
1016         int                      i;
1017         int                      rc;
1018
1019         ENTRY;
1020
1021         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1022                                     MDS_BULK_PORTAL);
1023         if (desc == NULL)
1024                 RETURN(-ENOMEM);
1025
1026         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1027                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1028                  * which is rdpg->rp_count */
1029                 nob = rdpg->rp_count;
1030
1031         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1032              i++, tmpcount -= tmpsize) {
1033                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1034                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1035         }
1036
1037         LASSERT(desc->bd_nob == nob);
1038         rc = target_bulk_io(exp, desc, lwi);
1039         ptlrpc_free_bulk_pin(desc);
1040         RETURN(rc);
1041 }
1042 EXPORT_SYMBOL(tgt_sendpage);
1043
1044 /*
1045  * OBD_IDX_READ handler
1046  */
1047 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1048 {
1049         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1050         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1051         struct idx_info         *req_ii, *rep_ii;
1052         int                      rc, i;
1053
1054         ENTRY;
1055
1056         memset(rdpg, 0, sizeof(*rdpg));
1057         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1058
1059         /* extract idx_info buffer from request & reply */
1060         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1061         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1062                 RETURN(err_serious(-EPROTO));
1063
1064         rc = req_capsule_server_pack(tsi->tsi_pill);
1065         if (rc)
1066                 RETURN(err_serious(rc));
1067
1068         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1069         if (rep_ii == NULL)
1070                 RETURN(err_serious(-EFAULT));
1071         rep_ii->ii_magic = IDX_INFO_MAGIC;
1072
1073         /* extract hash to start with */
1074         rdpg->rp_hash = req_ii->ii_hash_start;
1075
1076         /* extract requested attributes */
1077         rdpg->rp_attrs = req_ii->ii_attrs;
1078
1079         /* check that fid packed in request is valid and supported */
1080         if (!fid_is_sane(&req_ii->ii_fid))
1081                 RETURN(-EINVAL);
1082         rep_ii->ii_fid = req_ii->ii_fid;
1083
1084         /* copy flags */
1085         rep_ii->ii_flags = req_ii->ii_flags;
1086
1087         /* compute number of pages to allocate, ii_count is the number of 4KB
1088          * containers */
1089         if (req_ii->ii_count <= 0)
1090                 GOTO(out, rc = -EFAULT);
1091         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1092                                exp_max_brw_size(tsi->tsi_exp));
1093         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1094
1095         /* allocate pages to store the containers */
1096         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1097         if (rdpg->rp_pages == NULL)
1098                 GOTO(out, rc = -ENOMEM);
1099         for (i = 0; i < rdpg->rp_npages; i++) {
1100                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1101                 if (rdpg->rp_pages[i] == NULL)
1102                         GOTO(out, rc = -ENOMEM);
1103         }
1104
1105         /* populate pages with key/record pairs */
1106         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1107         if (rc < 0)
1108                 GOTO(out, rc);
1109
1110         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1111                  "asked %d > %d\n", rc, rdpg->rp_count);
1112
1113         /* send pages to client */
1114         rc = tgt_sendpage(tsi, rdpg, rc);
1115         if (rc)
1116                 GOTO(out, rc);
1117         EXIT;
1118 out:
1119         if (rdpg->rp_pages) {
1120                 for (i = 0; i < rdpg->rp_npages; i++)
1121                         if (rdpg->rp_pages[i])
1122                                 __free_page(rdpg->rp_pages[i]);
1123                 OBD_FREE(rdpg->rp_pages,
1124                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1125         }
1126         return rc;
1127 }
1128
1129 struct tgt_handler tgt_obd_handlers[] = {
1130 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1131 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1132 TGT_OBD_HDL_VAR(0,      OBD_QC_CALLBACK,        tgt_obd_qc_callback),
1133 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1134 };
1135 EXPORT_SYMBOL(tgt_obd_handlers);
1136
1137 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1138              struct dt_object *obj, __u64 start, __u64 end)
1139 {
1140         int rc = 0;
1141
1142         ENTRY;
1143
1144         /* if no objid is specified, it means "sync whole filesystem" */
1145         if (obj == NULL) {
1146                 rc = dt_sync(env, tgt->lut_bottom);
1147         } else if (dt_version_get(env, obj) >
1148                    tgt->lut_obd->obd_last_committed) {
1149                 rc = dt_object_sync(env, obj, start, end);
1150         }
1151
1152         RETURN(rc);
1153 }
1154 EXPORT_SYMBOL(tgt_sync);
1155 /*
1156  * Unified target DLM handlers.
1157  */
1158
1159 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1160  * (if requested) */
1161 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1162                             void *data, int flag)
1163 {
1164         struct lu_env            env;
1165         struct lu_target        *tgt;
1166         struct dt_object        *obj;
1167         struct lu_fid            fid;
1168         int                      rc = 0;
1169
1170         ENTRY;
1171
1172         tgt = class_exp2tgt(lock->l_export);
1173
1174         if (flag == LDLM_CB_CANCELING &&
1175             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1176             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1177              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1178               lock->l_flags & LDLM_FL_CBPENDING))) {
1179                 __u64 start = 0;
1180                 __u64 end = OBD_OBJECT_EOF;
1181
1182                 rc = lu_env_init(&env, LCT_DT_THREAD);
1183                 if (unlikely(rc != 0))
1184                         RETURN(rc);
1185
1186                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1187                                    tgt->lut_lsd.lsd_osd_index);
1188                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1189                 if (IS_ERR(obj))
1190                         GOTO(err_env, rc = PTR_ERR(obj));
1191
1192                 if (!dt_object_exists(obj))
1193                         GOTO(err_put, rc = -ENOENT);
1194
1195                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1196                         start = lock->l_policy_data.l_extent.start;
1197                         end = lock->l_policy_data.l_extent.end;
1198                 }
1199
1200                 rc = tgt_sync(&env, tgt, obj, start, end);
1201                 if (rc < 0) {
1202                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1203                                "cancel: rc = %d\n",
1204                                tgt_name(tgt), PFID(&fid),
1205                                lock->l_policy_data.l_extent.start,
1206                                lock->l_policy_data.l_extent.end, rc);
1207                 }
1208 err_put:
1209                 lu_object_put(&env, &obj->do_lu);
1210 err_env:
1211                 lu_env_fini(&env);
1212         }
1213
1214         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1215         RETURN(rc);
1216 }
1217
1218 static struct ldlm_callback_suite tgt_dlm_cbs = {
1219         .lcs_completion = ldlm_server_completion_ast,
1220         .lcs_blocking   = tgt_blocking_ast,
1221         .lcs_glimpse    = ldlm_server_glimpse_ast
1222 };
1223
1224 int tgt_enqueue(struct tgt_session_info *tsi)
1225 {
1226         struct ptlrpc_request *req = tgt_ses_req(tsi);
1227         int rc;
1228
1229         ENTRY;
1230         /*
1231          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1232          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1233          */
1234         LASSERT(tsi->tsi_dlm_req != NULL);
1235         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1236                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1237         if (rc)
1238                 RETURN(err_serious(rc));
1239
1240         RETURN(req->rq_status);
1241 }
1242 EXPORT_SYMBOL(tgt_enqueue);
1243
1244 int tgt_convert(struct tgt_session_info *tsi)
1245 {
1246         struct ptlrpc_request *req = tgt_ses_req(tsi);
1247         int rc;
1248
1249         ENTRY;
1250         LASSERT(tsi->tsi_dlm_req);
1251         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1252         if (rc)
1253                 RETURN(err_serious(rc));
1254
1255         RETURN(req->rq_status);
1256 }
1257
1258 int tgt_bl_callback(struct tgt_session_info *tsi)
1259 {
1260         return err_serious(-EOPNOTSUPP);
1261 }
1262
1263 int tgt_cp_callback(struct tgt_session_info *tsi)
1264 {
1265         return err_serious(-EOPNOTSUPP);
1266 }
1267
1268 /* generic LDLM target handler */
1269 struct tgt_handler tgt_dlm_handlers[] = {
1270 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1271 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1272 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1273 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1274 };
1275 EXPORT_SYMBOL(tgt_dlm_handlers);
1276
1277 /*
1278  * Unified target LLOG handlers.
1279  */
1280 int tgt_llog_open(struct tgt_session_info *tsi)
1281 {
1282         int rc;
1283
1284         ENTRY;
1285
1286         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1287
1288         RETURN(rc);
1289 }
1290 EXPORT_SYMBOL(tgt_llog_open);
1291
1292 int tgt_llog_close(struct tgt_session_info *tsi)
1293 {
1294         int rc;
1295
1296         ENTRY;
1297
1298         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1299
1300         RETURN(rc);
1301 }
1302 EXPORT_SYMBOL(tgt_llog_close);
1303
1304
1305 int tgt_llog_destroy(struct tgt_session_info *tsi)
1306 {
1307         int rc;
1308
1309         ENTRY;
1310
1311         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1312
1313         RETURN(rc);
1314 }
1315
1316 int tgt_llog_read_header(struct tgt_session_info *tsi)
1317 {
1318         int rc;
1319
1320         ENTRY;
1321
1322         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1323
1324         RETURN(rc);
1325 }
1326 EXPORT_SYMBOL(tgt_llog_read_header);
1327
1328 int tgt_llog_next_block(struct tgt_session_info *tsi)
1329 {
1330         int rc;
1331
1332         ENTRY;
1333
1334         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1335
1336         RETURN(rc);
1337 }
1338 EXPORT_SYMBOL(tgt_llog_next_block);
1339
1340 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1341 {
1342         int rc;
1343
1344         ENTRY;
1345
1346         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1347
1348         RETURN(rc);
1349 }
1350 EXPORT_SYMBOL(tgt_llog_prev_block);
1351
1352 /* generic llog target handler */
1353 struct tgt_handler tgt_llog_handlers[] = {
1354 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1355 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1356 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1357 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1358 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1359 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1360 };
1361 EXPORT_SYMBOL(tgt_llog_handlers);
1362
1363 /*
1364  * sec context handlers
1365  */
1366 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1367 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1368 {
1369         return 0;
1370 }
1371
1372 struct tgt_handler tgt_sec_ctx_handlers[] = {
1373 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1374 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1375 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1376 };
1377 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1378
1379 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1380                            struct dt_device *key,
1381                            struct lfsck_request *lr,
1382                            struct thandle *th) = NULL;
1383
1384 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1385                                                 struct dt_device *,
1386                                                 struct lfsck_request *,
1387                                                 struct thandle *))
1388 {
1389         tgt_lfsck_in_notify = notify;
1390 }
1391 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1392
1393 static int (*tgt_lfsck_query)(const struct lu_env *env,
1394                               struct dt_device *key,
1395                               struct lfsck_request *lr) = NULL;
1396
1397 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1398                                            struct dt_device *,
1399                                            struct lfsck_request *))
1400 {
1401         tgt_lfsck_query = query;
1402 }
1403 EXPORT_SYMBOL(tgt_register_lfsck_query);
1404
1405 /* LFSCK request handlers */
1406 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1407 {
1408         const struct lu_env     *env = tsi->tsi_env;
1409         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1410         struct lfsck_request    *lr;
1411         int                      rc;
1412         ENTRY;
1413
1414         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1415         if (lr == NULL)
1416                 RETURN(-EPROTO);
1417
1418         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1419
1420         RETURN(rc);
1421 }
1422
1423 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1424 {
1425         struct lfsck_request    *request;
1426         struct lfsck_reply      *reply;
1427         int                      rc;
1428         ENTRY;
1429
1430         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1431         if (request == NULL)
1432                 RETURN(-EPROTO);
1433
1434         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1435         if (reply == NULL)
1436                 RETURN(-ENOMEM);
1437
1438         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1439         reply->lr_status = rc;
1440
1441         RETURN(rc < 0 ? rc : 0);
1442 }
1443
1444 struct tgt_handler tgt_lfsck_handlers[] = {
1445 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1446 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1447 };
1448 EXPORT_SYMBOL(tgt_lfsck_handlers);
1449
1450 /*
1451  * initialize per-thread page pool (bug 5137).
1452  */
1453 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1454 {
1455         struct tgt_thread_big_cache *tbc;
1456
1457         ENTRY;
1458
1459         LASSERT(thread != NULL);
1460         LASSERT(thread->t_data == NULL);
1461
1462         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1463         if (tbc == NULL)
1464                 RETURN(-ENOMEM);
1465         thread->t_data = tbc;
1466         RETURN(0);
1467 }
1468 EXPORT_SYMBOL(tgt_io_thread_init);
1469
1470 /*
1471  * free per-thread pool created by tgt_thread_init().
1472  */
1473 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1474 {
1475         struct tgt_thread_big_cache *tbc;
1476
1477         ENTRY;
1478
1479         LASSERT(thread != NULL);
1480
1481         /*
1482          * be prepared to handle partially-initialized pools (because this is
1483          * called from ost_io_thread_init() for cleanup.
1484          */
1485         tbc = thread->t_data;
1486         if (tbc != NULL) {
1487                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1488                 thread->t_data = NULL;
1489         }
1490         EXIT;
1491 }
1492 EXPORT_SYMBOL(tgt_io_thread_done);
1493 /**
1494  * Helper function for getting server side [start, start+count] DLM lock
1495  * if asked by client.
1496  */
1497 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1498                     __u64 start, __u64 end, struct lustre_handle *lh,
1499                     int mode, __u64 *flags)
1500 {
1501         ldlm_policy_data_t       policy;
1502         int                      rc;
1503
1504         ENTRY;
1505
1506         LASSERT(lh != NULL);
1507         LASSERT(ns != NULL);
1508         LASSERT(!lustre_handle_is_used(lh));
1509
1510         policy.l_extent.gid = 0;
1511         policy.l_extent.start = start & CFS_PAGE_MASK;
1512
1513         /*
1514          * If ->o_blocks is EOF it means "lock till the end of the file".
1515          * Otherwise, it's size of an extent or hole being punched (in bytes).
1516          */
1517         if (end == OBD_OBJECT_EOF || end < start)
1518                 policy.l_extent.end = OBD_OBJECT_EOF;
1519         else
1520                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1521
1522         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1523                                     flags, ldlm_blocking_ast,
1524                                     ldlm_completion_ast, ldlm_glimpse_ast,
1525                                     NULL, 0, LVB_T_NONE, NULL, lh);
1526         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1527 }
1528 EXPORT_SYMBOL(tgt_extent_lock);
1529
1530 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1531 {
1532         LASSERT(lustre_handle_is_used(lh));
1533         ldlm_lock_decref(lh, mode);
1534 }
1535 EXPORT_SYMBOL(tgt_extent_unlock);
1536
1537 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1538                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1539                  struct lustre_handle *lh, int mode)
1540 {
1541         __u64                    flags = 0;
1542         int                      nrbufs = obj->ioo_bufcnt;
1543         int                      i;
1544
1545         ENTRY;
1546
1547         LASSERT(mode == LCK_PR || mode == LCK_PW);
1548         LASSERT(!lustre_handle_is_used(lh));
1549
1550         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1551                 RETURN(0);
1552
1553         for (i = 1; i < nrbufs; i++)
1554                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1555                         RETURN(-EFAULT);
1556
1557         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1558                                nb[nrbufs - 1].rnb_offset +
1559                                nb[nrbufs - 1].rnb_len - 1,
1560                                lh, mode, &flags));
1561 }
1562
1563 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1564                     struct lustre_handle *lh, int mode)
1565 {
1566         ENTRY;
1567
1568         LASSERT(mode == LCK_PR || mode == LCK_PW);
1569         LASSERT((obj->ioo_bufcnt > 0 &&
1570                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1571                 lustre_handle_is_used(lh));
1572
1573         if (lustre_handle_is_used(lh))
1574                 tgt_extent_unlock(lh, mode);
1575         EXIT;
1576 }
1577
1578 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1579                                struct ptlrpc_bulk_desc *desc, int opc,
1580                                cksum_type_t cksum_type)
1581 {
1582         struct cfs_crypto_hash_desc     *hdesc;
1583         unsigned int                    bufsize;
1584         int                             i, err;
1585         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1586         __u32                           cksum;
1587
1588         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1589         if (IS_ERR(hdesc)) {
1590                 CERROR("%s: unable to initialize checksum hash %s\n",
1591                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1592                 return PTR_ERR(hdesc);
1593         }
1594
1595         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1596         for (i = 0; i < desc->bd_iov_count; i++) {
1597                 /* corrupt the data before we compute the checksum, to
1598                  * simulate a client->OST data error */
1599                 if (i == 0 && opc == OST_WRITE &&
1600                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1601                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1602                         int len = desc->bd_iov[i].kiov_len;
1603                         struct page *np = tgt_page_to_corrupt;
1604                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1605
1606                         if (np) {
1607                                 char *ptr2 = kmap(np) + off;
1608
1609                                 memcpy(ptr2, ptr, len);
1610                                 memcpy(ptr2, "bad3", min(4, len));
1611                                 kunmap(np);
1612                                 desc->bd_iov[i].kiov_page = np;
1613                         } else {
1614                                 CERROR("%s: can't alloc page for corruption\n",
1615                                        tgt_name(tgt));
1616                         }
1617                 }
1618                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1619                                   desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1620                                   desc->bd_iov[i].kiov_len);
1621
1622                  /* corrupt the data after we compute the checksum, to
1623                  * simulate an OST->client data error */
1624                 if (i == 0 && opc == OST_READ &&
1625                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1626                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1627                         int len = desc->bd_iov[i].kiov_len;
1628                         struct page *np = tgt_page_to_corrupt;
1629                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1630
1631                         if (np) {
1632                                 char *ptr2 = kmap(np) + off;
1633
1634                                 memcpy(ptr2, ptr, len);
1635                                 memcpy(ptr2, "bad4", min(4, len));
1636                                 kunmap(np);
1637                                 desc->bd_iov[i].kiov_page = np;
1638                         } else {
1639                                 CERROR("%s: can't alloc page for corruption\n",
1640                                        tgt_name(tgt));
1641                         }
1642                 }
1643         }
1644
1645         bufsize = sizeof(cksum);
1646         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1647
1648         return cksum;
1649 }
1650
1651 int tgt_brw_read(struct tgt_session_info *tsi)
1652 {
1653         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1654         struct ptlrpc_bulk_desc *desc = NULL;
1655         struct obd_export       *exp = tsi->tsi_exp;
1656         struct niobuf_remote    *remote_nb;
1657         struct niobuf_local     *local_nb;
1658         struct obd_ioobj        *ioo;
1659         struct ost_body         *body, *repbody;
1660         struct l_wait_info       lwi;
1661         struct lustre_handle     lockh = { 0 };
1662         int                      npages, nob = 0, rc, i, no_reply = 0;
1663         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1664
1665         ENTRY;
1666
1667         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1668                 CERROR("%s: deny read request from %s to portal %u\n",
1669                        tgt_name(tsi->tsi_tgt),
1670                        obd_export_nid2str(req->rq_export),
1671                        ptlrpc_req2svc(req)->srv_req_portal);
1672                 RETURN(-EPROTO);
1673         }
1674
1675         req->rq_bulk_read = 1;
1676
1677         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1678                 RETURN(-EIO);
1679
1680         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1681                          cfs_fail_val : (obd_timeout + 1) / 4);
1682
1683         /* Check if there is eviction in progress, and if so, wait for it to
1684          * finish */
1685         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1686                 /* We do not care how long it takes */
1687                 lwi = LWI_INTR(NULL, NULL);
1688                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1689                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1690                          &lwi);
1691         }
1692
1693         /* There must be big cache in current thread to process this request
1694          * if it is NULL then something went wrong and it wasn't allocated,
1695          * report -ENOMEM in that case */
1696         if (tbc == NULL)
1697                 RETURN(-ENOMEM);
1698
1699         body = tsi->tsi_ost_body;
1700         LASSERT(body != NULL);
1701
1702         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1703         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1704
1705         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1706         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1707
1708         local_nb = tbc->local;
1709
1710         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1711                           remote_nb, &lockh, LCK_PR);
1712         if (rc != 0)
1713                 RETURN(rc);
1714
1715         /*
1716          * If getting the lock took more time than
1717          * client was willing to wait, drop it. b=11330
1718          */
1719         if (cfs_time_current_sec() > req->rq_deadline ||
1720             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1721                 no_reply = 1;
1722                 CERROR("Dropping timed-out read from %s because locking"
1723                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1724                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1725                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1726                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1727                 GOTO(out_lock, rc = -ETIMEDOUT);
1728         }
1729
1730         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1731         repbody->oa = body->oa;
1732
1733         npages = PTLRPC_MAX_BRW_PAGES;
1734         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1735                         ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1736         if (rc != 0)
1737                 GOTO(out_lock, rc);
1738
1739         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1740                                     BULK_PUT_SOURCE, OST_BULK_PORTAL);
1741         if (desc == NULL)
1742                 GOTO(out_commitrw, rc = -ENOMEM);
1743
1744         nob = 0;
1745         for (i = 0; i < npages; i++) {
1746                 int page_rc = local_nb[i].lnb_rc;
1747
1748                 if (page_rc < 0) {
1749                         rc = page_rc;
1750                         break;
1751                 }
1752
1753                 nob += page_rc;
1754                 if (page_rc != 0) { /* some data! */
1755                         LASSERT(local_nb[i].lnb_page != NULL);
1756                         ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
1757                                                     local_nb[i].lnb_page_offset,
1758                                                     page_rc);
1759                 }
1760
1761                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1762                         /* All subsequent pages should be 0 */
1763                         while (++i < npages)
1764                                 LASSERT(local_nb[i].lnb_rc == 0);
1765                         break;
1766                 }
1767         }
1768
1769         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1770                 cksum_type_t cksum_type =
1771                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1772                                           body->oa.o_flags : 0);
1773                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1774                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1775                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1776                                                         OST_READ, cksum_type);
1777                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1778                        repbody->oa.o_cksum);
1779         } else {
1780                 repbody->oa.o_valid = 0;
1781         }
1782         /* We're finishing using body->oa as an input variable */
1783
1784         /* Check if client was evicted while we were doing i/o before touching
1785          * network */
1786         if (likely(rc == 0 &&
1787                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1788                 rc = target_bulk_io(exp, desc, &lwi);
1789                 no_reply = rc != 0;
1790         }
1791
1792 out_commitrw:
1793         /* Must commit after prep above in all cases */
1794         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1795                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1796                           NULL, rc);
1797         if (rc == 0)
1798                 tgt_drop_id(exp, &repbody->oa);
1799 out_lock:
1800         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1801
1802         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1803                 ptlrpc_free_bulk_nopin(desc);
1804
1805         LASSERT(rc <= 0);
1806         if (rc == 0) {
1807                 rc = nob;
1808                 ptlrpc_lprocfs_brw(req, nob);
1809         } else if (no_reply) {
1810                 req->rq_no_reply = 1;
1811                 /* reply out callback would free */
1812                 ptlrpc_req_drop_rs(req);
1813                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1814                               "client will retry: rc %d\n",
1815                               exp->exp_obd->obd_name,
1816                               obd_uuid2str(&exp->exp_client_uuid),
1817                               obd_export_nid2str(exp), rc);
1818         }
1819         /* send a bulk after reply to simulate a network delay or reordering
1820          * by a router */
1821         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1822                 wait_queue_head_t        waitq;
1823                 struct l_wait_info       lwi1;
1824
1825                 CDEBUG(D_INFO, "reorder BULK\n");
1826                 init_waitqueue_head(&waitq);
1827
1828                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1829                 l_wait_event(waitq, 0, &lwi1);
1830                 target_bulk_io(exp, desc, &lwi);
1831                 ptlrpc_free_bulk_nopin(desc);
1832         }
1833
1834         RETURN(rc);
1835 }
1836 EXPORT_SYMBOL(tgt_brw_read);
1837
1838 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1839                               struct ptlrpc_bulk_desc *desc,
1840                               struct niobuf_local *local_nb, int npages,
1841                               u32 client_cksum, u32 server_cksum,
1842                               bool mmap)
1843 {
1844         struct obd_export *exp = req->rq_export;
1845         struct ost_body *body;
1846         char *router;
1847         char *via;
1848
1849         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1850         LASSERT(body != NULL);
1851
1852         if (req->rq_peer.nid == desc->bd_sender) {
1853                 via = router = "";
1854         } else {
1855                 via = " via ";
1856                 router = libcfs_nid2str(desc->bd_sender);
1857         }
1858
1859         if (mmap) {
1860                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1861                              client_cksum, server_cksum);
1862                 return;
1863         }
1864
1865         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1866                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1867                            "]: client csum %x, server csum %x\n",
1868                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1869                            via, router,
1870                            body->oa.o_valid & OBD_MD_FLFID ?
1871                            body->oa.o_parent_seq : (__u64)0,
1872                            body->oa.o_valid & OBD_MD_FLFID ?
1873                            body->oa.o_parent_oid : 0,
1874                            body->oa.o_valid & OBD_MD_FLFID ?
1875                            body->oa.o_parent_ver : 0,
1876                            POSTID(&body->oa.o_oi),
1877                            local_nb[0].lnb_file_offset,
1878                            local_nb[npages-1].lnb_file_offset +
1879                            local_nb[npages - 1].lnb_len - 1,
1880                            client_cksum, server_cksum);
1881 }
1882
1883 int tgt_brw_write(struct tgt_session_info *tsi)
1884 {
1885         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1886         struct ptlrpc_bulk_desc *desc = NULL;
1887         struct obd_export       *exp = req->rq_export;
1888         struct niobuf_remote    *remote_nb;
1889         struct niobuf_local     *local_nb;
1890         struct obd_ioobj        *ioo;
1891         struct ost_body         *body, *repbody;
1892         struct l_wait_info       lwi;
1893         struct lustre_handle     lockh = {0};
1894         __u32                   *rcs;
1895         int                      objcount, niocount, npages;
1896         int                      rc, i, j;
1897         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1898         bool                     no_reply = false, mmap;
1899         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1900
1901         ENTRY;
1902
1903         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1904                 CERROR("%s: deny write request from %s to portal %u\n",
1905                        tgt_name(tsi->tsi_tgt),
1906                        obd_export_nid2str(req->rq_export),
1907                        ptlrpc_req2svc(req)->srv_req_portal);
1908                 RETURN(err_serious(-EPROTO));
1909         }
1910
1911         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1912                 RETURN(err_serious(-ENOSPC));
1913         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1914                 RETURN(err_serious(-EROFS));
1915
1916         req->rq_bulk_write = 1;
1917
1918         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1919                 RETURN(err_serious(-EIO));
1920         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1921                 RETURN(err_serious(-EFAULT));
1922
1923         /* pause before transaction has been started */
1924         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1925                          cfs_fail_val : (obd_timeout + 1) / 4);
1926
1927         /* There must be big cache in current thread to process this request
1928          * if it is NULL then something went wrong and it wasn't allocated,
1929          * report -ENOMEM in that case */
1930         if (tbc == NULL)
1931                 RETURN(-ENOMEM);
1932
1933         body = tsi->tsi_ost_body;
1934         LASSERT(body != NULL);
1935
1936         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1937         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1938
1939         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1940                                         RCL_CLIENT) / sizeof(*ioo);
1941
1942         for (niocount = i = 0; i < objcount; i++)
1943                 niocount += ioo[i].ioo_bufcnt;
1944
1945         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1946         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1947         if (niocount != req_capsule_get_size(&req->rq_pill,
1948                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1949                         sizeof(*remote_nb))
1950                 RETURN(err_serious(-EPROTO));
1951
1952         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1953             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1954                 memory_pressure_set();
1955
1956         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1957                              niocount * sizeof(*rcs));
1958         rc = req_capsule_server_pack(&req->rq_pill);
1959         if (rc != 0)
1960                 GOTO(out, rc = err_serious(rc));
1961
1962         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1963         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1964
1965         local_nb = tbc->local;
1966
1967         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1968                           remote_nb, &lockh, LCK_PW);
1969         if (rc != 0)
1970                 GOTO(out, rc);
1971
1972         /*
1973          * If getting the lock took more time than
1974          * client was willing to wait, drop it. b=11330
1975          */
1976         if (cfs_time_current_sec() > req->rq_deadline ||
1977             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1978                 no_reply = true;
1979                 CERROR("%s: Dropping timed-out write from %s because locking "
1980                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1981                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1982                        POSTID(&ioo->ioo_oid),
1983                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1984                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1985                 GOTO(out_lock, rc = -ETIMEDOUT);
1986         }
1987
1988         /* Because we already sync grant info with client when reconnect,
1989          * grant info will be cleared for resent req, then fed_grant and
1990          * total_grant will not be modified in following preprw_write */
1991         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1992                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1993                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1994         }
1995
1996         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1997         if (repbody == NULL)
1998                 GOTO(out_lock, rc = -ENOMEM);
1999         repbody->oa = body->oa;
2000
2001         npages = PTLRPC_MAX_BRW_PAGES;
2002         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2003                         objcount, ioo, remote_nb, &npages, local_nb, NULL,
2004                         BYPASS_CAPA);
2005         if (rc < 0)
2006                 GOTO(out_lock, rc);
2007
2008         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2009                                     BULK_GET_SINK, OST_BULK_PORTAL);
2010         if (desc == NULL)
2011                 GOTO(skip_transfer, rc = -ENOMEM);
2012
2013         /* NB Having prepped, we must commit... */
2014         for (i = 0; i < npages; i++)
2015                 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
2016                                             local_nb[i].lnb_page_offset,
2017                                             local_nb[i].lnb_len);
2018
2019         rc = sptlrpc_svc_prep_bulk(req, desc);
2020         if (rc != 0)
2021                 GOTO(skip_transfer, rc);
2022
2023         rc = target_bulk_io(exp, desc, &lwi);
2024         no_reply = rc != 0;
2025
2026 skip_transfer:
2027         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2028                 static int cksum_counter;
2029
2030                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2031                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2032
2033                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2034                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2035                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2036                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2037                                                         OST_WRITE, cksum_type);
2038                 cksum_counter++;
2039
2040                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2041                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2042                                 body->oa.o_flags & OBD_FL_MMAP);
2043
2044                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2045                                           body->oa.o_cksum,
2046                                           repbody->oa.o_cksum, mmap);
2047                         cksum_counter = 0;
2048                 } else if ((cksum_counter & (-cksum_counter)) ==
2049                            cksum_counter) {
2050                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2051                                cksum_counter, libcfs_id2str(req->rq_peer),
2052                                repbody->oa.o_cksum);
2053                 }
2054         }
2055
2056         /* Must commit after prep above in all cases */
2057         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2058                           objcount, ioo, remote_nb, npages, local_nb, NULL,
2059                           rc);
2060         if (rc == -ENOTCONN)
2061                 /* quota acquire process has been given up because
2062                  * either the client has been evicted or the client
2063                  * has timed out the request already */
2064                 no_reply = true;
2065
2066         /*
2067          * Disable sending mtime back to the client. If the client locked the
2068          * whole object, then it has already updated the mtime on its side,
2069          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2070          */
2071         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2072
2073         if (rc == 0) {
2074                 int nob = 0;
2075
2076                 /* set per-requested niobuf return codes */
2077                 for (i = j = 0; i < niocount; i++) {
2078                         int len = remote_nb[i].rnb_len;
2079
2080                         nob += len;
2081                         rcs[i] = 0;
2082                         do {
2083                                 LASSERT(j < npages);
2084                                 if (local_nb[j].lnb_rc < 0)
2085                                         rcs[i] = local_nb[j].lnb_rc;
2086                                 len -= local_nb[j].lnb_len;
2087                                 j++;
2088                         } while (len > 0);
2089                         LASSERT(len == 0);
2090                 }
2091                 LASSERT(j == npages);
2092                 ptlrpc_lprocfs_brw(req, nob);
2093
2094                 tgt_drop_id(exp, &repbody->oa);
2095         }
2096 out_lock:
2097         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2098         if (desc)
2099                 ptlrpc_free_bulk_nopin(desc);
2100 out:
2101         if (no_reply) {
2102                 req->rq_no_reply = 1;
2103                 /* reply out callback would free */
2104                 ptlrpc_req_drop_rs(req);
2105                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2106                               "client will retry: rc %d\n",
2107                               exp->exp_obd->obd_name,
2108                               obd_uuid2str(&exp->exp_client_uuid),
2109                               obd_export_nid2str(exp), rc);
2110         }
2111         memory_pressure_clr();
2112         RETURN(rc);
2113 }
2114 EXPORT_SYMBOL(tgt_brw_write);