lustre/target/tgt_handler.c

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