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, 2012, Intel Corporation.
  25  */
  26 /*
  27  * lustre/target/tgt_handler.c
  28  *
  29  * Lustre Unified Target request handler code
  30  *
  31  * Author: Brian Behlendorf <behlendorf1@llnl.gov>
  32  * Author: Mikhail Pershin <mike.pershin@intel.com>
  33  */
  34
  35 #define DEBUG_SUBSYSTEM S_CLASS
  36
  37 #include <obd.h>
  38 #include <obd_class.h>
  39 #include <obd_cksum.h>
  40
  41 #include "tgt_internal.h"
  42
  43 char *tgt_name(struct lu_target *tgt)
  44 {
  45         LASSERT(tgt->lut_obd != NULL);
  46         return tgt->lut_obd->obd_name;
  47 }
  48 EXPORT_SYMBOL(tgt_name);
  49
  50 /*
  51  * Generic code handling requests that have struct mdt_body passed in:
  52  *
  53  *  - extract mdt_body from request and save it in @tsi, if present;
  54  *
  55  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  56  *  @tsi;
  57  *
  58  *  - if HABEO_CORPUS flag is set for this request type check whether object
  59  *  actually exists on storage (lu_object_exists()).
  60  *
  61  */
  62 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  63 {
  64         const struct mdt_body   *body;
  65         struct lu_object        *obj;
  66         struct req_capsule      *pill = tsi->tsi_pill;
  67         int                      rc;
  68
  69         ENTRY;
  70
  71         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  72         if (body == NULL)
  73                 RETURN(-EFAULT);
  74
  75         tsi->tsi_mdt_body = body;
  76
  77         if (!(body->valid & OBD_MD_FLID))
  78                 RETURN(0);
  79
  80         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  81          * in any case in pre-2.5 clients. Fix that here if needed */
  82         if (unlikely(fid_is_zero(&body->fid1)))
  83                 RETURN(0);
  84
  85         if (!fid_is_sane(&body->fid1)) {
  86                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  87                        PFID(&body->fid1));
  88                 RETURN(-EINVAL);
  89         }
  90
  91         obj = lu_object_find(tsi->tsi_env,
  92                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
  93                              &body->fid1, NULL);
  94         if (!IS_ERR(obj)) {
  95                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
  96                         lu_object_put(tsi->tsi_env, obj);
  97                         /* for capability renew ENOENT will be handled in
  98                          * mdt_renew_capa */
  99                         if (body->valid & OBD_MD_FLOSSCAPA)
 100                                 rc = 0;
 101                         else
 102                                 rc = -ENOENT;
 103                 } else {
 104                         tsi->tsi_corpus = obj;
 105                         rc = 0;
 106                 }
 107         } else {
 108                 rc = PTR_ERR(obj);
 109         }
 110         RETURN(rc);
 111 }
 112
 113 /**
 114  * Validate oa from client.
 115  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 116  * req are valid.
 117  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 118  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 119  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 120  *       whether this is oi_fid or real ostid. So it will check
 121  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 122  *    c. Old FID-disable osc will send IDIF.
 123  *    d. new FID-enable osc/osp will send normal FID.
 124  *
 125  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 126  * be used for LAST_ID file, and only being accessed inside OST now.
 127  */
 128 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 129 {
 130         int rc;
 131
 132         ENTRY;
 133
 134         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
 135                      fid_seq_is_echo(oa->o_oi.oi.oi_seq))) {
 136                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 137                  * during create, and we will reset this to 1, since this
 138                  * oi_id is basically useless in the following create process,
 139                  * but oi_id == 0 will make it difficult to tell whether it is
 140                  * real FID or ost_id. */
 141                 oa->o_oi.oi_fid.f_oid = oa->o_oi.oi.oi_id ?: 1;
 142                 oa->o_oi.oi_fid.f_seq = FID_SEQ_ECHO;
 143                 oa->o_oi.oi_fid.f_ver = 0;
 144         } else {
 145                 if (unlikely((oa->o_valid & OBD_MD_FLID &&
 146                               ostid_id(&oa->o_oi) == 0)))
 147                         GOTO(out, rc = -EPROTO);
 148
 149                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 150                  * all of the requests are required to setup FLGROUP */
 151                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 152                         ostid_set_seq_mdt0(&oa->o_oi);
 153                         oa->o_valid |= OBD_MD_FLGROUP;
 154                 }
 155
 156                 if (unlikely(!(fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
 157                                fid_seq_is_mdt0(ostid_seq(&oa->o_oi)) ||
 158                                fid_seq_is_norm(ostid_seq(&oa->o_oi)) ||
 159                                fid_seq_is_echo(ostid_seq(&oa->o_oi)))))
 160                         GOTO(out, rc = -EPROTO);
 161         }
 162         RETURN(0);
 163 out:
 164         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 165                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
 166                ostid_seq(&oa->o_oi), ostid_id(&oa->o_oi), rc);
 167         return rc;
 168 }
 169 EXPORT_SYMBOL(tgt_validate_obdo);
 170
 171 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 172 {
 173         struct ost_body         *body;
 174         struct req_capsule      *pill = tsi->tsi_pill;
 175         struct lustre_capa      *capa;
 176         struct obd_ioobj        *ioo;
 177         int                      rc;
 178
 179         ENTRY;
 180
 181         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 182         if (body == NULL)
 183                 RETURN(-EFAULT);
 184
 185         rc = tgt_validate_obdo(tsi, &body->oa);
 186         if (rc)
 187                 RETURN(rc);
 188
 189         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 190                 capa = req_capsule_client_get(tsi->tsi_pill, &RMF_CAPA1);
 191                 if (capa == NULL) {
 192                         CERROR("%s: OSSCAPA flag is set without capability\n",
 193                                tgt_name(tsi->tsi_tgt));
 194                         RETURN(-EFAULT);
 195                 }
 196         }
 197
 198         tsi->tsi_ost_body = body;
 199
 200         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 201                 unsigned                 max_brw;
 202                 struct niobuf_remote    *rnb;
 203
 204                 ioo = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
 205                 if (ioo == NULL)
 206                         RETURN(-EPROTO);
 207
 208                 rnb = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
 209                 if (rnb == NULL)
 210                         RETURN(-EPROTO);
 211
 212                 max_brw = ioobj_max_brw_get(ioo);
 213                 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 214                         CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 215                                ": rc = %d\n", tgt_name(tsi->tsi_tgt),
 216                                obd_export_nid2str(tsi->tsi_exp), max_brw,
 217                                POSTID(&body->oa.o_oi), -EPROTO);
 218                         RETURN(-EPROTO);
 219                 }
 220                 ioo->ioo_oid = body->oa.o_oi;
 221         }
 222
 223         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 224                 if (flags & HABEO_CORPUS) {
 225                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 226                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 227                                tgt_name(tsi->tsi_tgt));
 228                         RETURN(-EPROTO);
 229                 } else {
 230                         RETURN(0);
 231                 }
 232         }
 233
 234         rc = ostid_to_fid(&tsi->tsi_fid, &body->oa.o_oi, 0);
 235         if (rc != 0)
 236                 RETURN(rc);
 237
 238         if (!fid_is_sane(&tsi->tsi_fid)) {
 239                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
 240                        PFID(&tsi->tsi_fid));
 241                 RETURN(-EINVAL);
 242         }
 243
 244         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 245
 246         /*
 247          * OST doesn't get object in advance for further use to prevent
 248          * situations with nested object_find which is potential deadlock.
 249          */
 250         tsi->tsi_corpus = NULL;
 251         RETURN(rc);
 252 }
 253
 254 static int tgt_unpack_req_pack_rep(struct tgt_session_info *tsi, __u32 flags)
 255 {
 256         struct req_capsule      *pill = tsi->tsi_pill;
 257         int                      rc;
 258
 259         ENTRY;
 260
 261         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 262                 rc = tgt_mdt_body_unpack(tsi, flags);
 263         } else if (req_capsule_has_field(pill, &RMF_OST_BODY, RCL_CLIENT)) {
 264                 rc = tgt_ost_body_unpack(tsi, flags);
 265         } else {
 266                 rc = 0;
 267         }
 268
 269         if (rc == 0 && flags & HABEO_REFERO) {
 270                 /* Pack reply */
 271                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 272                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
 273                                              tsi->tsi_mdt_body->eadatasize);
 274                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 275                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
 276                                              RCL_SERVER, 0);
 277
 278                 rc = req_capsule_server_pack(pill);
 279         }
 280         RETURN(rc);
 281 }
 282
 283 /*
 284  * Invoke handler for this request opc. Also do necessary preprocessing
 285  * (according to handler ->th_flags), and post-processing (setting of
 286  * ->last_{xid,committed}).
 287  */
 288 static int tgt_handle_request0(struct tgt_session_info *tsi,
 289                                struct tgt_handler *h,
 290                                struct ptlrpc_request *req)
 291 {
 292         int      serious = 0;
 293         int      rc;
 294         __u32    flags;
 295
 296         ENTRY;
 297
 298         LASSERT(h->th_act != NULL);
 299         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 300         LASSERT(current->journal_info == NULL);
 301
 302         /*
 303          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 304          * to put same checks into handlers like mdt_close(), mdt_reint(),
 305          * etc., without talking to mdt authors first. Checking same thing
 306          * there again is useless and returning 0 error without packing reply
 307          * is buggy! Handlers either pack reply or return error.
 308          *
 309          * We return 0 here and do not send any reply in order to emulate
 310          * network failure. Do not send any reply in case any of NET related
 311          * fail_id has occured.
 312          */
 313         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 314                 RETURN(0);
 315
 316         rc = 0;
 317         flags = h->th_flags;
 318         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 319                      h->th_fmt != NULL));
 320         if (h->th_fmt != NULL) {
 321                 req_capsule_set(tsi->tsi_pill, h->th_fmt);
 322                 rc = tgt_unpack_req_pack_rep(tsi, flags);
 323         }
 324
 325         if (rc == 0 && flags & MUTABOR &&
 326             tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 327                 rc = -EROFS;
 328
 329         if (rc == 0 && flags & HABEO_CLAVIS) {
 330                 struct ldlm_request *dlm_req;
 331
 332                 LASSERT(h->th_fmt != NULL);
 333
 334                 dlm_req = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
 335                 if (dlm_req != NULL) {
 336                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 337                                      LDLM_IBITS &&
 338                                      dlm_req->lock_desc.l_policy_data.\
 339                                      l_inodebits.bits == 0)) {
 340                                 /*
 341                                  * Lock without inodebits makes no sense and
 342                                  * will oops later in ldlm. If client miss to
 343                                  * set such bits, do not trigger ASSERTION.
 344                                  *
 345                                  * For liblustre flock case, it maybe zero.
 346                                  */
 347                                 rc = -EPROTO;
 348                         } else {
 349                                 tsi->tsi_dlm_req = dlm_req;
 350                         }
 351                 } else {
 352                         rc = -EFAULT;
 353                 }
 354         }
 355
 356         if (likely(rc == 0)) {
 357                 /*
 358                  * Process request, there can be two types of rc:
 359                  * 1) errors with msg unpack/pack, other failures outside the
 360                  * operation itself. This is counted as serious errors;
 361                  * 2) errors during fs operation, should be placed in rq_status
 362                  * only
 363                  */
 364                 rc = h->th_act(tsi);
 365                 if (!is_serious(rc) &&
 366                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 367                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 368                                   "pack reply and returned 0 error\n",
 369                                   tgt_name(tsi->tsi_tgt), h->th_name);
 370                         LBUG();
 371                 }
 372                 serious = is_serious(rc);
 373                 rc = clear_serious(rc);
 374         } else {
 375                 serious = 1;
 376         }
 377
 378         req->rq_status = rc;
 379
 380         /*
 381          * ELDLM_* codes which > 0 should be in rq_status only as well as
 382          * all non-serious errors.
 383          */
 384         if (rc > 0 || !serious)
 385                 rc = 0;
 386
 387         LASSERT(current->journal_info == NULL);
 388
 389         if (likely(rc == 0 && req->rq_export))
 390                 target_committed_to_req(req);
 391
 392         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 393         RETURN(0);
 394 }
 395
 396 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 397                                        struct obd_device *obd, int *process)
 398 {
 399         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 400         case MDS_DISCONNECT:
 401         case OST_DISCONNECT:
 402         case OBD_IDX_READ:
 403                 *process = 1;
 404                 RETURN(0);
 405         case MDS_CLOSE:
 406         case MDS_DONE_WRITING:
 407         case MDS_SYNC: /* used in unmounting */
 408         case OBD_PING:
 409         case MDS_REINT:
 410         case UPDATE_OBJ:
 411         case SEQ_QUERY:
 412         case FLD_QUERY:
 413         case LDLM_ENQUEUE:
 414         case OST_CREATE:
 415         case OST_DESTROY:
 416         case OST_PUNCH:
 417         case OST_SETATTR:
 418         case OST_SYNC:
 419         case OST_WRITE:
 420                 *process = target_queue_recovery_request(req, obd);
 421                 RETURN(0);
 422
 423         default:
 424                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 425                 *process = -EAGAIN;
 426                 RETURN(0);
 427         }
 428 }
 429
 430 /*
 431  * Handle recovery. Return:
 432  *        +1: continue request processing;
 433  *       -ve: abort immediately with the given error code;
 434  *         0: send reply with error code in req->rq_status;
 435  */
 436 int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 437 {
 438         ENTRY;
 439
 440         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 441         case MDS_CONNECT:
 442         case OST_CONNECT:
 443         case MGS_CONNECT:
 444         case SEC_CTX_INIT:
 445         case SEC_CTX_INIT_CONT:
 446         case SEC_CTX_FINI:
 447                 RETURN(+1);
 448         }
 449
 450         if (!req->rq_export->exp_obd->obd_replayable)
 451                 RETURN(+1);
 452
 453         /* sanity check: if the xid matches, the request must be marked as a
 454          * resent or replayed */
 455         if (req_xid_is_last(req)) {
 456                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 457                       (MSG_RESENT | MSG_REPLAY))) {
 458                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 459                                   "last_xid, expected REPLAY or RESENT flag "
 460                                   "(%x)", req->rq_xid,
 461                                   lustre_msg_get_flags(req->rq_reqmsg));
 462                         req->rq_status = -ENOTCONN;
 463                         RETURN(-ENOTCONN);
 464                 }
 465         }
 466         /* else: note the opposite is not always true; a RESENT req after a
 467          * failover will usually not match the last_xid, since it was likely
 468          * never committed. A REPLAYed request will almost never match the
 469          * last xid, however it could for a committed, but still retained,
 470          * open. */
 471
 472         /* Check for aborted recovery... */
 473         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 474                 int rc;
 475                 int should_process;
 476
 477                 DEBUG_REQ(D_INFO, req, "Got new replay");
 478                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 479                                                  &should_process);
 480                 if (rc != 0 || !should_process)
 481                         RETURN(rc);
 482                 else if (should_process < 0) {
 483                         req->rq_status = should_process;
 484                         rc = ptlrpc_error(req);
 485                         RETURN(rc);
 486                 }
 487         }
 488         RETURN(+1);
 489 }
 490
 491 int tgt_request_handle(struct ptlrpc_request *req)
 492 {
 493         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 494
 495         struct lustre_msg       *msg = req->rq_reqmsg;
 496         struct tgt_handler      *h;
 497         struct tgt_opc_slice    *s;
 498         struct lu_target        *tgt;
 499         int                      request_fail_id = 0;
 500         __u32                    opc = lustre_msg_get_opc(msg);
 501         int                      rc;
 502
 503         ENTRY;
 504
 505         /* Refill the context, to make sure all thread keys are allocated */
 506         lu_env_refill(req->rq_svc_thread->t_env);
 507
 508         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 509         tsi->tsi_pill = &req->rq_pill;
 510         tsi->tsi_env = req->rq_svc_thread->t_env;
 511
 512         /* if request has export then get handlers slice from corresponding
 513          * target, otherwise that should be connect operation */
 514         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 515             opc == MGS_CONNECT) {
 516                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 517                 rc = target_handle_connect(req);
 518                 if (rc != 0) {
 519                         rc = ptlrpc_error(req);
 520                         GOTO(out, rc);
 521                 }
 522                 /* recovery-small test 18c asks to drop connect reply */
 523                 if (unlikely(opc == OST_CONNECT &&
 524                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 525                         GOTO(out, rc = 0);
 526         }
 527
 528         if (unlikely(!class_connected_export(req->rq_export))) {
 529                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 530                        opc, libcfs_id2str(req->rq_peer));
 531                 req->rq_status = -ENOTCONN;
 532                 rc = ptlrpc_error(req);
 533                 GOTO(out, rc);
 534         }
 535
 536         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 537         tsi->tsi_exp = req->rq_export;
 538         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 539                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 540         else
 541                 tsi->tsi_jobid = NULL;
 542
 543         request_fail_id = tgt->lut_request_fail_id;
 544         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 545
 546         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 547                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 548                         break;
 549
 550         /* opcode was not found in slice */
 551         if (unlikely(s->tos_hs == NULL)) {
 552                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt), opc);
 553                 req->rq_status = -ENOTSUPP;
 554                 rc = ptlrpc_error(req);
 555                 GOTO(out, rc);
 556         }
 557
 558         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 559                 GOTO(out, rc = 0);
 560
 561         LASSERT(current->journal_info == NULL);
 562
 563         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 564         h = s->tos_hs + (opc - s->tos_opc_start);
 565         if (unlikely(h->th_opc == 0)) {
 566                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 567                 req->rq_status = -ENOTSUPP;
 568                 rc = ptlrpc_error(req);
 569                 GOTO(out, rc);
 570         }
 571
 572         rc = lustre_msg_check_version(msg, h->th_version);
 573         if (unlikely(rc)) {
 574                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 575                           " %08x, expecting %08x\n", tgt_name(tgt),
 576                           lustre_msg_get_version(msg), h->th_version);
 577                 req->rq_status = -EINVAL;
 578                 rc = ptlrpc_error(req);
 579                 GOTO(out, rc);
 580         }
 581
 582         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 583         if (likely(rc == 1)) {
 584                 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 585                          h->th_opc, opc);
 586                 rc = tgt_handle_request0(tsi, h, req);
 587                 if (rc)
 588                         GOTO(out, rc);
 589         }
 590         EXIT;
 591 out:
 592         req_capsule_fini(tsi->tsi_pill);
 593         tsi->tsi_pill = NULL;
 594         if (tsi->tsi_corpus != NULL) {
 595                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 596                 tsi->tsi_corpus = NULL;
 597         }
 598         tsi->tsi_env = NULL;
 599         tsi->tsi_mdt_body = NULL;
 600         tsi->tsi_dlm_req = NULL;
 601         fid_zero(&tsi->tsi_fid);
 602         memset(&tsi->tsi_resid, 0, sizeof tsi->tsi_resid);
 603         return rc;
 604 }
 605 EXPORT_SYMBOL(tgt_request_handle);
 606
 607 void tgt_counter_incr(struct obd_export *exp, int opcode)
 608 {
 609         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 610         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 611                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 612 }
 613 EXPORT_SYMBOL(tgt_counter_incr);
 614
 615 /*
 616  * Unified target generic handlers.
 617  */
 618
 619 /*
 620  * Security functions
 621  */
 622 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 623 {
 624         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 625                                       OBD_CONNECT_RMT_CLIENT_FORCE |
 626                                       OBD_CONNECT_MDS_CAPA |
 627                                       OBD_CONNECT_OSS_CAPA);
 628 }
 629
 630 static int tgt_init_sec_level(struct ptlrpc_request *req)
 631 {
 632         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 633         char                    *client = libcfs_nid2str(req->rq_peer.nid);
 634         struct obd_connect_data *data, *reply;
 635         int                      rc = 0;
 636         bool                     remote;
 637
 638         ENTRY;
 639
 640         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 641         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 642         if (data == NULL || reply == NULL)
 643                 RETURN(-EFAULT);
 644
 645         /* connection from MDT is always trusted */
 646         if (req->rq_auth_usr_mdt) {
 647                 tgt_init_sec_none(reply);
 648                 RETURN(0);
 649         }
 650
 651         /* no GSS support case */
 652         if (!req->rq_auth_gss) {
 653                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 654                         CWARN("client %s -> target %s does not use GSS, "
 655                               "can not run under security level %d.\n",
 656                               client, tgt_name(tgt), tgt->lut_sec_level);
 657                         RETURN(-EACCES);
 658                 } else {
 659                         tgt_init_sec_none(reply);
 660                         RETURN(0);
 661                 }
 662         }
 663
 664         /* old version case */
 665         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 666                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 667                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 668                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 669                         CWARN("client %s -> target %s uses old version, "
 670                               "can not run under security level %d.\n",
 671                               client, tgt_name(tgt), tgt->lut_sec_level);
 672                         RETURN(-EACCES);
 673                 } else {
 674                         CWARN("client %s -> target %s uses old version, "
 675                               "run under security level %d.\n",
 676                               client, tgt_name(tgt), tgt->lut_sec_level);
 677                         tgt_init_sec_none(reply);
 678                         RETURN(0);
 679                 }
 680         }
 681
 682         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 683         if (remote) {
 684                 if (!req->rq_auth_remote)
 685                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 686                                "asked to be remote.\n", client, tgt_name(tgt));
 687         } else if (req->rq_auth_remote) {
 688                 remote = true;
 689                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 690                        "as remote by default.\n", client, tgt_name(tgt));
 691         }
 692
 693         if (remote) {
 694                 if (!tgt->lut_oss_capa) {
 695                         CDEBUG(D_SEC,
 696                                "client %s -> target %s is set as remote,"
 697                                " but OSS capabilities are not enabled: %d.\n",
 698                                client, tgt_name(tgt), tgt->lut_oss_capa);
 699                         RETURN(-EACCES);
 700                 }
 701         } else {
 702                 if (req->rq_auth_uid == INVALID_UID) {
 703                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 704                                "authenticated!\n", client, tgt_name(tgt));
 705                         RETURN(-EACCES);
 706                 }
 707         }
 708
 709
 710         switch (tgt->lut_sec_level) {
 711         case LUSTRE_SEC_NONE:
 712                 if (remote) {
 713                         CDEBUG(D_SEC,
 714                                "client %s -> target %s is set as remote, "
 715                                "can not run under security level %d.\n",
 716                                client, tgt_name(tgt), tgt->lut_sec_level);
 717                         RETURN(-EACCES);
 718                 }
 719                 tgt_init_sec_none(reply);
 720                 break;
 721         case LUSTRE_SEC_REMOTE:
 722                 if (!remote)
 723                         tgt_init_sec_none(reply);
 724                 break;
 725         case LUSTRE_SEC_ALL:
 726                 if (remote)
 727                         break;
 728                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 729                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 730                 if (!tgt->lut_oss_capa)
 731                         reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 732                 if (!tgt->lut_mds_capa)
 733                         reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 734                 break;
 735         default:
 736                 RETURN(-EINVAL);
 737         }
 738
 739         RETURN(rc);
 740 }
 741
 742 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 743 {
 744         struct lu_target        *tgt = class_exp2tgt(exp);
 745         struct sptlrpc_flavor    flvr;
 746         int                      rc = 0;
 747
 748         LASSERT(tgt);
 749         LASSERT(tgt->lut_obd);
 750         LASSERT(tgt->lut_slice);
 751
 752         /* always allow ECHO client */
 753         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 754                             LUSTRE_ECHO_NAME) == 0)) {
 755                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 756                 return 0;
 757         }
 758
 759         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 760                 read_lock(&tgt->lut_sptlrpc_lock);
 761                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 762                                              req->rq_sp_from,
 763                                              req->rq_peer.nid,
 764                                              &flvr);
 765                 read_unlock(&tgt->lut_sptlrpc_lock);
 766
 767                 spin_lock(&exp->exp_lock);
 768                 exp->exp_sp_peer = req->rq_sp_from;
 769                 exp->exp_flvr = flvr;
 770                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 771                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 772                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 773                                "expect %x\n", tgt_name(tgt),
 774                                req->rq_flvr.sf_rpc,
 775                                libcfs_nid2str(req->rq_peer.nid),
 776                                exp->exp_flvr.sf_rpc);
 777                         rc = -EACCES;
 778                 }
 779                 spin_unlock(&exp->exp_lock);
 780         } else {
 781                 if (exp->exp_sp_peer != req->rq_sp_from) {
 782                         CERROR("%s: RPC source %s doesn't match %s\n",
 783                                tgt_name(tgt),
 784                                sptlrpc_part2name(req->rq_sp_from),
 785                                sptlrpc_part2name(exp->exp_sp_peer));
 786                         rc = -EACCES;
 787                 } else {
 788                         rc = sptlrpc_target_export_check(exp, req);
 789                 }
 790         }
 791
 792         return rc;
 793 }
 794
 795 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 796 {
 797         struct sptlrpc_rule_set  tmp_rset;
 798         int                      rc;
 799
 800         sptlrpc_rule_set_init(&tmp_rset);
 801         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 802         if (rc) {
 803                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 804                        tgt_name(tgt), rc);
 805                 return rc;
 806         }
 807
 808         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 809
 810         write_lock(&tgt->lut_sptlrpc_lock);
 811         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 812         tgt->lut_sptlrpc_rset = tmp_rset;
 813         write_unlock(&tgt->lut_sptlrpc_lock);
 814
 815         return 0;
 816 }
 817 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 818
 819 int tgt_connect(struct tgt_session_info *tsi)
 820 {
 821         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 822         struct obd_connect_data *reply;
 823         int                      rc;
 824
 825         ENTRY;
 826
 827         rc = tgt_init_sec_level(req);
 828         if (rc != 0)
 829                 GOTO(out, rc);
 830
 831         /* XXX: better to call this check right after getting new export but
 832          * before last_rcvd slot allocation to avoid server load upon insecure
 833          * connects. This is to be fixed after unifiyng all targets.
 834          */
 835         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
 836         if (rc)
 837                 GOTO(out, rc);
 838
 839         /* To avoid exposing partially initialized connection flags, changes up
 840          * to this point have been staged in reply->ocd_connect_flags. Now that
 841          * connection handling has completed successfully, atomically update
 842          * the connect flags in the shared export data structure. LU-1623 */
 843         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
 844         spin_lock(&tsi->tsi_exp->exp_lock);
 845         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
 846         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
 847         spin_unlock(&tsi->tsi_exp->exp_lock);
 848
 849         RETURN(0);
 850 out:
 851         obd_disconnect(class_export_get(tsi->tsi_exp));
 852         return rc;
 853 }
 854 EXPORT_SYMBOL(tgt_connect);
 855
 856 int tgt_disconnect(struct tgt_session_info *tsi)
 857 {
 858         int rc;
 859
 860         ENTRY;
 861
 862         rc = target_handle_disconnect(tgt_ses_req(tsi));
 863         if (rc)
 864                 RETURN(err_serious(rc));
 865
 866         RETURN(rc);
 867 }
 868 EXPORT_SYMBOL(tgt_disconnect);
 869
 870 /*
 871  * Unified target OBD handlers
 872  */
 873 int tgt_obd_ping(struct tgt_session_info *tsi)
 874 {
 875         int rc;
 876
 877         ENTRY;
 878
 879         rc = target_handle_ping(tgt_ses_req(tsi));
 880         if (rc)
 881                 RETURN(err_serious(rc));
 882
 883         RETURN(rc);
 884 }
 885 EXPORT_SYMBOL(tgt_obd_ping);
 886
 887 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
 888 {
 889         return err_serious(-EOPNOTSUPP);
 890 }
 891 EXPORT_SYMBOL(tgt_obd_log_cancel);
 892
 893 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
 894 {
 895         return err_serious(-EOPNOTSUPP);
 896 }
 897 EXPORT_SYMBOL(tgt_obd_qc_callback);
 898
 899 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
 900 {
 901         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
 902         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 903         struct obd_export       *exp = req->rq_export;
 904         struct ptlrpc_bulk_desc *desc;
 905         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
 906         int                      tmpcount;
 907         int                      tmpsize;
 908         int                      i;
 909         int                      rc;
 910
 911         ENTRY;
 912
 913         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
 914                                     MDS_BULK_PORTAL);
 915         if (desc == NULL)
 916                 RETURN(-ENOMEM);
 917
 918         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
 919                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
 920                  * which is rdpg->rp_count */
 921                 nob = rdpg->rp_count;
 922
 923         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
 924              i++, tmpcount -= tmpsize) {
 925                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
 926                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
 927         }
 928
 929         LASSERT(desc->bd_nob == nob);
 930         rc = target_bulk_io(exp, desc, lwi);
 931         ptlrpc_free_bulk_pin(desc);
 932         RETURN(rc);
 933 }
 934 EXPORT_SYMBOL(tgt_sendpage);
 935
 936 /*
 937  * OBD_IDX_READ handler
 938  */
 939 int tgt_obd_idx_read(struct tgt_session_info *tsi)
 940 {
 941         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
 942         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
 943         struct idx_info         *req_ii, *rep_ii;
 944         int                      rc, i;
 945
 946         ENTRY;
 947
 948         memset(rdpg, 0, sizeof(*rdpg));
 949         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
 950
 951         /* extract idx_info buffer from request & reply */
 952         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
 953         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
 954                 RETURN(err_serious(-EPROTO));
 955
 956         rc = req_capsule_server_pack(tsi->tsi_pill);
 957         if (rc)
 958                 RETURN(err_serious(rc));
 959
 960         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
 961         if (rep_ii == NULL)
 962                 RETURN(err_serious(-EFAULT));
 963         rep_ii->ii_magic = IDX_INFO_MAGIC;
 964
 965         /* extract hash to start with */
 966         rdpg->rp_hash = req_ii->ii_hash_start;
 967
 968         /* extract requested attributes */
 969         rdpg->rp_attrs = req_ii->ii_attrs;
 970
 971         /* check that fid packed in request is valid and supported */
 972         if (!fid_is_sane(&req_ii->ii_fid))
 973                 RETURN(-EINVAL);
 974         rep_ii->ii_fid = req_ii->ii_fid;
 975
 976         /* copy flags */
 977         rep_ii->ii_flags = req_ii->ii_flags;
 978
 979         /* compute number of pages to allocate, ii_count is the number of 4KB
 980          * containers */
 981         if (req_ii->ii_count <= 0)
 982                 GOTO(out, rc = -EFAULT);
 983         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
 984                                exp_max_brw_size(tsi->tsi_exp));
 985         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
 986
 987         /* allocate pages to store the containers */
 988         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
 989         if (rdpg->rp_pages == NULL)
 990                 GOTO(out, rc = -ENOMEM);
 991         for (i = 0; i < rdpg->rp_npages; i++) {
 992                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
 993                 if (rdpg->rp_pages[i] == NULL)
 994                         GOTO(out, rc = -ENOMEM);
 995         }
 996
 997         /* populate pages with key/record pairs */
 998         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
 999         if (rc < 0)
1000                 GOTO(out, rc);
1001
1002         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1003                  "asked %d > %d\n", rc, rdpg->rp_count);
1004
1005         /* send pages to client */
1006         rc = tgt_sendpage(tsi, rdpg, rc);
1007         if (rc)
1008                 GOTO(out, rc);
1009         EXIT;
1010 out:
1011         if (rdpg->rp_pages) {
1012                 for (i = 0; i < rdpg->rp_npages; i++)
1013                         if (rdpg->rp_pages[i])
1014                                 __free_page(rdpg->rp_pages[i]);
1015                 OBD_FREE(rdpg->rp_pages,
1016                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1017         }
1018         return rc;
1019 }
1020 EXPORT_SYMBOL(tgt_obd_idx_read);
1021
1022 struct tgt_handler tgt_obd_handlers[] = {
1023 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1024 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1025 TGT_OBD_HDL_VAR(0,      OBD_QC_CALLBACK,        tgt_obd_qc_callback),
1026 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1027 };
1028 EXPORT_SYMBOL(tgt_obd_handlers);
1029
1030 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1031              struct dt_object *obj)
1032 {
1033         int rc = 0;
1034
1035         ENTRY;
1036
1037         /* if no objid is specified, it means "sync whole filesystem" */
1038         if (obj == NULL) {
1039                 rc = dt_sync(env, tgt->lut_bottom);
1040         } else if (dt_version_get(env, obj) >
1041                    tgt->lut_obd->obd_last_committed) {
1042                 rc = dt_object_sync(env, obj);
1043         }
1044
1045         RETURN(rc);
1046 }
1047 EXPORT_SYMBOL(tgt_sync);
1048 /*
1049  * Unified target DLM handlers.
1050  */
1051
1052 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1053  * (if requested) */
1054 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1055                      void *data, int flag)
1056 {
1057         struct lu_env            env;
1058         struct lu_target        *tgt;
1059         struct dt_object        *obj;
1060         struct lu_fid            fid;
1061         int                      rc = 0;
1062
1063         ENTRY;
1064
1065         tgt = class_exp2tgt(lock->l_export);
1066
1067         if (flag == LDLM_CB_CANCELING &&
1068             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1069             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1070              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1071               lock->l_flags & LDLM_FL_CBPENDING))) {
1072                 rc = lu_env_init(&env, LCT_DT_THREAD);
1073                 if (unlikely(rc != 0))
1074                         RETURN(rc);
1075
1076                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name);
1077                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1078                 if (IS_ERR(obj))
1079                         GOTO(err_env, rc = PTR_ERR(obj));
1080
1081                 if (!dt_object_exists(obj))
1082                         GOTO(err_put, rc = -ENOENT);
1083
1084                 rc = tgt_sync(&env, tgt, obj);
1085                 if (rc < 0) {
1086                         CERROR("%s: sync failed on lock cancel: rc = %d\n",
1087                                tgt_name(tgt), rc);
1088                 }
1089 err_put:
1090                 lu_object_put(&env, &obj->do_lu);
1091 err_env:
1092                 lu_env_fini(&env);
1093         }
1094
1095         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1096         RETURN(rc);
1097 }
1098
1099 struct ldlm_callback_suite tgt_dlm_cbs = {
1100         .lcs_completion = ldlm_server_completion_ast,
1101         .lcs_blocking   = tgt_blocking_ast,
1102         .lcs_glimpse    = ldlm_server_glimpse_ast
1103 };
1104
1105 int tgt_enqueue(struct tgt_session_info *tsi)
1106 {
1107         struct ptlrpc_request *req = tgt_ses_req(tsi);
1108         int rc;
1109
1110         ENTRY;
1111         /*
1112          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1113          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1114          */
1115         LASSERT(tsi->tsi_dlm_req != NULL);
1116         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1117                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1118         if (rc)
1119                 RETURN(err_serious(rc));
1120
1121         RETURN(req->rq_status);
1122 }
1123 EXPORT_SYMBOL(tgt_enqueue);
1124
1125 int tgt_convert(struct tgt_session_info *tsi)
1126 {
1127         struct ptlrpc_request *req = tgt_ses_req(tsi);
1128         int rc;
1129
1130         ENTRY;
1131         LASSERT(tsi->tsi_dlm_req);
1132         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1133         if (rc)
1134                 RETURN(err_serious(rc));
1135
1136         RETURN(req->rq_status);
1137 }
1138 EXPORT_SYMBOL(tgt_convert);
1139
1140 int tgt_bl_callback(struct tgt_session_info *tsi)
1141 {
1142         return err_serious(-EOPNOTSUPP);
1143 }
1144 EXPORT_SYMBOL(tgt_bl_callback);
1145
1146 int tgt_cp_callback(struct tgt_session_info *tsi)
1147 {
1148         return err_serious(-EOPNOTSUPP);
1149 }
1150 EXPORT_SYMBOL(tgt_cp_callback);
1151
1152 /* generic LDLM target handler */
1153 struct tgt_handler tgt_dlm_handlers[] = {
1154 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1155 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1156 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1157 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1158 };
1159 EXPORT_SYMBOL(tgt_dlm_handlers);
1160
1161 /*
1162  * Unified target LLOG handlers.
1163  */
1164 int tgt_llog_open(struct tgt_session_info *tsi)
1165 {
1166         int rc;
1167
1168         ENTRY;
1169
1170         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1171
1172         RETURN(rc);
1173 }
1174 EXPORT_SYMBOL(tgt_llog_open);
1175
1176 int tgt_llog_close(struct tgt_session_info *tsi)
1177 {
1178         int rc;
1179
1180         ENTRY;
1181
1182         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1183
1184         RETURN(rc);
1185 }
1186 EXPORT_SYMBOL(tgt_llog_close);
1187
1188
1189 int tgt_llog_destroy(struct tgt_session_info *tsi)
1190 {
1191         int rc;
1192
1193         ENTRY;
1194
1195         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1196
1197         RETURN(rc);
1198 }
1199 EXPORT_SYMBOL(tgt_llog_destroy);
1200
1201 int tgt_llog_read_header(struct tgt_session_info *tsi)
1202 {
1203         int rc;
1204
1205         ENTRY;
1206
1207         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1208
1209         RETURN(rc);
1210 }
1211 EXPORT_SYMBOL(tgt_llog_read_header);
1212
1213 int tgt_llog_next_block(struct tgt_session_info *tsi)
1214 {
1215         int rc;
1216
1217         ENTRY;
1218
1219         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1220
1221         RETURN(rc);
1222 }
1223 EXPORT_SYMBOL(tgt_llog_next_block);
1224
1225 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1226 {
1227         int rc;
1228
1229         ENTRY;
1230
1231         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1232
1233         RETURN(rc);
1234 }
1235 EXPORT_SYMBOL(tgt_llog_prev_block);
1236
1237 /* generic llog target handler */
1238 struct tgt_handler tgt_llog_handlers[] = {
1239 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1240 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1241 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1242 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1243 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1244 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1245 };
1246 EXPORT_SYMBOL(tgt_llog_handlers);
1247
1248 /*
1249  * sec context handlers
1250  */
1251 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1252 int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1253 {
1254         return 0;
1255 }
1256
1257 struct tgt_handler tgt_sec_ctx_handlers[] = {
1258 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1259 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1260 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1261 };
1262 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1263
1264 /*
1265  * initialize per-thread page pool (bug 5137).
1266  */
1267 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1268 {
1269         struct tgt_thread_big_cache *tbc;
1270
1271         ENTRY;
1272
1273         LASSERT(thread != NULL);
1274         LASSERT(thread->t_data == NULL);
1275
1276         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1277         if (tbc == NULL)
1278                 RETURN(-ENOMEM);
1279         thread->t_data = tbc;
1280         RETURN(0);
1281 }
1282 EXPORT_SYMBOL(tgt_io_thread_init);
1283
1284 /*
1285  * free per-thread pool created by tgt_thread_init().
1286  */
1287 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1288 {
1289         struct tgt_thread_big_cache *tbc;
1290
1291         ENTRY;
1292
1293         LASSERT(thread != NULL);
1294
1295         /*
1296          * be prepared to handle partially-initialized pools (because this is
1297          * called from ost_io_thread_init() for cleanup.
1298          */
1299         tbc = thread->t_data;
1300         if (tbc != NULL) {
1301                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1302                 thread->t_data = NULL;
1303         }
1304         EXIT;
1305 }
1306 EXPORT_SYMBOL(tgt_io_thread_done);
1307 /**
1308  * Helper function for getting server side [start, start+count] DLM lock
1309  * if asked by client.
1310  */
1311 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1312                     __u64 start, __u64 end, struct lustre_handle *lh,
1313                     int mode, __u64 *flags)
1314 {
1315         ldlm_policy_data_t       policy;
1316         int                      rc;
1317
1318         ENTRY;
1319
1320         LASSERT(lh != NULL);
1321         LASSERT(ns != NULL);
1322         LASSERT(!lustre_handle_is_used(lh));
1323
1324         policy.l_extent.gid = 0;
1325         policy.l_extent.start = start & CFS_PAGE_MASK;
1326
1327         /*
1328          * If ->o_blocks is EOF it means "lock till the end of the file".
1329          * Otherwise, it's size of an extent or hole being punched (in bytes).
1330          */
1331         if (end == OBD_OBJECT_EOF || end < start)
1332                 policy.l_extent.end = OBD_OBJECT_EOF;
1333         else
1334                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1335
1336         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1337                                     flags, ldlm_blocking_ast,
1338                                     ldlm_completion_ast, ldlm_glimpse_ast,
1339                                     NULL, 0, LVB_T_NONE, NULL, lh);
1340         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1341 }
1342 EXPORT_SYMBOL(tgt_extent_lock);
1343
1344 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1345 {
1346         LASSERT(lustre_handle_is_used(lh));
1347         ldlm_lock_decref(lh, mode);
1348 }
1349 EXPORT_SYMBOL(tgt_extent_unlock);
1350
1351 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1352                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1353                  struct lustre_handle *lh, int mode)
1354 {
1355         __u64                    flags = 0;
1356         int                      nrbufs = obj->ioo_bufcnt;
1357         int                      i;
1358
1359         ENTRY;
1360
1361         LASSERT(mode == LCK_PR || mode == LCK_PW);
1362         LASSERT(!lustre_handle_is_used(lh));
1363
1364         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1365                 RETURN(0);
1366
1367         for (i = 1; i < nrbufs; i++)
1368                 if (!(nb[i].flags & OBD_BRW_SRVLOCK))
1369                         RETURN(-EFAULT);
1370
1371         RETURN(tgt_extent_lock(ns, res_id, nb[0].offset,
1372                                nb[nrbufs - 1].offset + nb[nrbufs - 1].len - 1,
1373                                lh, mode, &flags));
1374 }
1375 EXPORT_SYMBOL(tgt_brw_lock);
1376
1377 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1378                     struct lustre_handle *lh, int mode)
1379 {
1380         ENTRY;
1381
1382         LASSERT(mode == LCK_PR || mode == LCK_PW);
1383         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
1384                 lustre_handle_is_used(lh));
1385         if (lustre_handle_is_used(lh))
1386                 tgt_extent_unlock(lh, mode);
1387         EXIT;
1388 }
1389 EXPORT_SYMBOL(tgt_brw_unlock);
1390
1391 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1392                                struct ptlrpc_bulk_desc *desc, int opc,
1393                                cksum_type_t cksum_type)
1394 {
1395         struct cfs_crypto_hash_desc     *hdesc;
1396         unsigned int                    bufsize;
1397         int                             i, err;
1398         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1399         __u32                           cksum;
1400
1401         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1402         if (IS_ERR(hdesc)) {
1403                 CERROR("%s: unable to initialize checksum hash %s\n",
1404                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1405                 return PTR_ERR(hdesc);
1406         }
1407
1408         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1409         for (i = 0; i < desc->bd_iov_count; i++) {
1410                 /* corrupt the data before we compute the checksum, to
1411                  * simulate a client->OST data error */
1412                 if (i == 0 && opc == OST_WRITE &&
1413                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1414                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1415                         int len = desc->bd_iov[i].kiov_len;
1416                         struct page *np = tgt_page_to_corrupt;
1417                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1418
1419                         if (np) {
1420                                 char *ptr2 = kmap(np) + off;
1421
1422                                 memcpy(ptr2, ptr, len);
1423                                 memcpy(ptr2, "bad3", min(4, len));
1424                                 kunmap(np);
1425                                 desc->bd_iov[i].kiov_page = np;
1426                         } else {
1427                                 CERROR("%s: can't alloc page for corruption\n",
1428                                        tgt_name(tgt));
1429                         }
1430                 }
1431                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1432                                   desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1433                                   desc->bd_iov[i].kiov_len);
1434
1435                  /* corrupt the data after we compute the checksum, to
1436                  * simulate an OST->client data error */
1437                 if (i == 0 && opc == OST_READ &&
1438                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1439                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1440                         int len = desc->bd_iov[i].kiov_len;
1441                         struct page *np = tgt_page_to_corrupt;
1442                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1443
1444                         if (np) {
1445                                 char *ptr2 = kmap(np) + off;
1446
1447                                 memcpy(ptr2, ptr, len);
1448                                 memcpy(ptr2, "bad4", min(4, len));
1449                                 kunmap(np);
1450                                 desc->bd_iov[i].kiov_page = np;
1451                         } else {
1452                                 CERROR("%s: can't alloc page for corruption\n",
1453                                        tgt_name(tgt));
1454                         }
1455                 }
1456         }
1457
1458         bufsize = 4;
1459         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1460         if (err)
1461                 cfs_crypto_hash_final(hdesc, NULL, NULL);
1462
1463         return cksum;
1464 }
1465
1466 int tgt_brw_read(struct tgt_session_info *tsi)
1467 {
1468         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1469         struct ptlrpc_bulk_desc *desc = NULL;
1470         struct obd_export       *exp = tsi->tsi_exp;
1471         struct niobuf_remote    *remote_nb;
1472         struct niobuf_local     *local_nb;
1473         struct obd_ioobj        *ioo;
1474         struct ost_body         *body, *repbody;
1475         struct l_wait_info       lwi;
1476         struct lustre_handle     lockh = { 0 };
1477         int                      niocount, npages, nob = 0, rc, i;
1478         int                      no_reply = 0;
1479         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1480
1481         ENTRY;
1482
1483         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1484                 CERROR("%s: deny read request from %s to portal %u\n",
1485                        tgt_name(tsi->tsi_tgt),
1486                        obd_export_nid2str(req->rq_export),
1487                        ptlrpc_req2svc(req)->srv_req_portal);
1488                 RETURN(-EPROTO);
1489         }
1490
1491         req->rq_bulk_read = 1;
1492
1493         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1494                 RETURN(-EIO);
1495
1496         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1497                          cfs_fail_val : (obd_timeout + 1) / 4);
1498
1499         /* Check if there is eviction in progress, and if so, wait for it to
1500          * finish */
1501         if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1502                 /* We do not care how long it takes */
1503                 lwi = LWI_INTR(NULL, NULL);
1504                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1505                          !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
1506                          &lwi);
1507         }
1508
1509         /* There must be big cache in current thread to process this request
1510          * if it is NULL then something went wrong and it wasn't allocated,
1511          * report -ENOMEM in that case */
1512         if (tbc == NULL)
1513                 RETURN(-ENOMEM);
1514
1515         body = tsi->tsi_ost_body;
1516         LASSERT(body != NULL);
1517
1518         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1519         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1520
1521         niocount = ioo->ioo_bufcnt;
1522         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1523         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1524
1525         local_nb = tbc->local;
1526
1527         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1528                           remote_nb, &lockh, LCK_PR);
1529         if (rc != 0)
1530                 RETURN(rc);
1531
1532         /*
1533          * If getting the lock took more time than
1534          * client was willing to wait, drop it. b=11330
1535          */
1536         if (cfs_time_current_sec() > req->rq_deadline ||
1537             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1538                 no_reply = 1;
1539                 CERROR("Dropping timed-out read from %s because locking"
1540                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1541                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1542                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1543                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1544                 GOTO(out_lock, rc = -ETIMEDOUT);
1545         }
1546
1547         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1548         repbody->oa = body->oa;
1549
1550         npages = PTLRPC_MAX_BRW_PAGES;
1551         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1552                         ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1553         if (rc != 0)
1554                 GOTO(out_lock, rc);
1555
1556         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1557                                     BULK_PUT_SOURCE, OST_BULK_PORTAL);
1558         if (desc == NULL)
1559                 GOTO(out_commitrw, rc = -ENOMEM);
1560
1561         nob = 0;
1562         for (i = 0; i < npages; i++) {
1563                 int page_rc = local_nb[i].rc;
1564
1565                 if (page_rc < 0) {
1566                         rc = page_rc;
1567                         break;
1568                 }
1569
1570                 nob += page_rc;
1571                 if (page_rc != 0) { /* some data! */
1572                         LASSERT(local_nb[i].page != NULL);
1573                         ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1574                                                     local_nb[i].lnb_page_offset,
1575                                                     page_rc);
1576                 }
1577
1578                 if (page_rc != local_nb[i].len) { /* short read */
1579                         /* All subsequent pages should be 0 */
1580                         while (++i < npages)
1581                                 LASSERT(local_nb[i].rc == 0);
1582                         break;
1583                 }
1584         }
1585
1586         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1587                 cksum_type_t cksum_type =
1588                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1589                                           body->oa.o_flags : 0);
1590                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1591                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1592                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1593                                                         OST_READ, cksum_type);
1594                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1595                        repbody->oa.o_cksum);
1596         } else {
1597                 repbody->oa.o_valid = 0;
1598         }
1599         /* We're finishing using body->oa as an input variable */
1600
1601         /* Check if client was evicted while we were doing i/o before touching
1602          * network */
1603         if (likely(rc == 0 &&
1604                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1605                 rc = target_bulk_io(exp, desc, &lwi);
1606                 no_reply = rc != 0;
1607         }
1608
1609 out_commitrw:
1610         /* Must commit after prep above in all cases */
1611         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1612                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1613                           NULL, rc);
1614         if (rc == 0)
1615                 tgt_drop_id(exp, &repbody->oa);
1616 out_lock:
1617         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1618
1619         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1620                 ptlrpc_free_bulk_nopin(desc);
1621
1622         LASSERT(rc <= 0);
1623         if (rc == 0) {
1624                 rc = nob;
1625                 ptlrpc_lprocfs_brw(req, nob);
1626         } else if (no_reply) {
1627                 req->rq_no_reply = 1;
1628                 /* reply out callback would free */
1629                 ptlrpc_req_drop_rs(req);
1630                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1631                               "client will retry: rc %d\n",
1632                               exp->exp_obd->obd_name,
1633                               obd_uuid2str(&exp->exp_client_uuid),
1634                               obd_export_nid2str(exp), rc);
1635         }
1636         /* send a bulk after reply to simulate a network delay or reordering
1637          * by a router */
1638         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1639                 wait_queue_head_t        waitq;
1640                 struct l_wait_info       lwi1;
1641
1642                 CDEBUG(D_INFO, "reorder BULK\n");
1643                 init_waitqueue_head(&waitq);
1644
1645                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1646                 l_wait_event(waitq, 0, &lwi1);
1647                 target_bulk_io(exp, desc, &lwi);
1648                 ptlrpc_free_bulk_nopin(desc);
1649         }
1650
1651         RETURN(rc);
1652 }
1653 EXPORT_SYMBOL(tgt_brw_read);
1654
1655 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1656                               struct ptlrpc_bulk_desc *desc,
1657                               struct niobuf_local *local_nb, int npages,
1658                               obd_count client_cksum, obd_count server_cksum,
1659                               bool mmap)
1660 {
1661         struct obd_export *exp = req->rq_export;
1662         struct ost_body *body;
1663         char *router;
1664         char *via;
1665
1666         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1667         LASSERT(body != NULL);
1668
1669         if (req->rq_peer.nid == desc->bd_sender) {
1670                 via = router = "";
1671         } else {
1672                 via = " via ";
1673                 router = libcfs_nid2str(desc->bd_sender);
1674         }
1675
1676         if (mmap) {
1677                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1678                              client_cksum, server_cksum);
1679                 return;
1680         }
1681
1682         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1683                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1684                            "]: client csum %x, server csum %x\n",
1685                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1686                            via, router,
1687                            body->oa.o_valid & OBD_MD_FLFID ?
1688                            body->oa.o_parent_seq : (__u64)0,
1689                            body->oa.o_valid & OBD_MD_FLFID ?
1690                            body->oa.o_parent_oid : 0,
1691                            body->oa.o_valid & OBD_MD_FLFID ?
1692                            body->oa.o_parent_ver : 0,
1693                            POSTID(&body->oa.o_oi),
1694                            local_nb[0].lnb_file_offset,
1695                            local_nb[npages-1].lnb_file_offset +
1696                            local_nb[npages-1].len - 1,
1697                            client_cksum, server_cksum);
1698 }
1699
1700 int tgt_brw_write(struct tgt_session_info *tsi)
1701 {
1702         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1703         struct ptlrpc_bulk_desc *desc = NULL;
1704         struct obd_export       *exp = req->rq_export;
1705         struct niobuf_remote    *remote_nb;
1706         struct niobuf_local     *local_nb;
1707         struct obd_ioobj        *ioo;
1708         struct ost_body         *body, *repbody;
1709         struct l_wait_info       lwi;
1710         struct lustre_handle     lockh = {0};
1711         __u32                   *rcs;
1712         int                      objcount, niocount, npages;
1713         int                      rc, i, j;
1714         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1715         bool                     no_reply = false, mmap;
1716         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1717
1718         ENTRY;
1719
1720         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1721                 CERROR("%s: deny write request from %s to portal %u\n",
1722                        tgt_name(tsi->tsi_tgt),
1723                        obd_export_nid2str(req->rq_export),
1724                        ptlrpc_req2svc(req)->srv_req_portal);
1725                 RETURN(err_serious(-EPROTO));
1726         }
1727
1728         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1729                 RETURN(err_serious(-ENOSPC));
1730         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1731                 RETURN(err_serious(-EROFS));
1732
1733         req->rq_bulk_write = 1;
1734
1735         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1736                 RETURN(err_serious(-EIO));
1737         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1738                 RETURN(err_serious(-EFAULT));
1739
1740         /* pause before transaction has been started */
1741         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1742                          cfs_fail_val : (obd_timeout + 1) / 4);
1743
1744         /* There must be big cache in current thread to process this request
1745          * if it is NULL then something went wrong and it wasn't allocated,
1746          * report -ENOMEM in that case */
1747         if (tbc == NULL)
1748                 RETURN(-ENOMEM);
1749
1750         body = tsi->tsi_ost_body;
1751         LASSERT(body != NULL);
1752
1753         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1754         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1755
1756         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1757                                         RCL_CLIENT) / sizeof(*ioo);
1758
1759         for (niocount = i = 0; i < objcount; i++)
1760                 niocount += ioo[i].ioo_bufcnt;
1761
1762         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1763         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1764         if (niocount != req_capsule_get_size(&req->rq_pill,
1765                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1766                         sizeof(*remote_nb))
1767                 RETURN(err_serious(-EPROTO));
1768
1769         if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
1770             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1771                 memory_pressure_set();
1772
1773         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1774                              niocount * sizeof(*rcs));
1775         rc = req_capsule_server_pack(&req->rq_pill);
1776         if (rc != 0)
1777                 GOTO(out, rc = err_serious(rc));
1778
1779         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1780         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1781
1782         local_nb = tbc->local;
1783
1784         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1785                           remote_nb, &lockh, LCK_PW);
1786         if (rc != 0)
1787                 GOTO(out, rc);
1788
1789         /*
1790          * If getting the lock took more time than
1791          * client was willing to wait, drop it. b=11330
1792          */
1793         if (cfs_time_current_sec() > req->rq_deadline ||
1794             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1795                 no_reply = true;
1796                 CERROR("%s: Dropping timed-out write from %s because locking "
1797                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1798                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1799                        POSTID(&ioo->ioo_oid),
1800                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1801                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1802                 GOTO(out_lock, rc = -ETIMEDOUT);
1803         }
1804
1805         /* Because we already sync grant info with client when reconnect,
1806          * grant info will be cleared for resent req, then fed_grant and
1807          * total_grant will not be modified in following preprw_write */
1808         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1809                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1810                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1811         }
1812
1813         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1814         if (repbody == NULL)
1815                 GOTO(out_lock, rc = -ENOMEM);
1816         repbody->oa = body->oa;
1817
1818         npages = PTLRPC_MAX_BRW_PAGES;
1819         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1820                         objcount, ioo, remote_nb, &npages, local_nb, NULL,
1821                         BYPASS_CAPA);
1822         if (rc < 0)
1823                 GOTO(out_lock, rc);
1824
1825         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1826                                     BULK_GET_SINK, OST_BULK_PORTAL);
1827         if (desc == NULL)
1828                 GOTO(skip_transfer, rc = -ENOMEM);
1829
1830         /* NB Having prepped, we must commit... */
1831         for (i = 0; i < npages; i++)
1832                 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1833                                             local_nb[i].lnb_page_offset,
1834                                             local_nb[i].len);
1835
1836         rc = sptlrpc_svc_prep_bulk(req, desc);
1837         if (rc != 0)
1838                 GOTO(skip_transfer, rc);
1839
1840         rc = target_bulk_io(exp, desc, &lwi);
1841         no_reply = rc != 0;
1842
1843 skip_transfer:
1844         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
1845                 static int cksum_counter;
1846
1847                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1848                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1849
1850                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1851                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1852                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1853                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1854                                                         OST_WRITE, cksum_type);
1855                 cksum_counter++;
1856
1857                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
1858                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
1859                                 body->oa.o_flags & OBD_FL_MMAP);
1860
1861                         tgt_warn_on_cksum(req, desc, local_nb, npages,
1862                                           body->oa.o_cksum,
1863                                           repbody->oa.o_cksum, mmap);
1864                         cksum_counter = 0;
1865                 } else if ((cksum_counter & (-cksum_counter)) ==
1866                            cksum_counter) {
1867                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1868                                cksum_counter, libcfs_id2str(req->rq_peer),
1869                                repbody->oa.o_cksum);
1870                 }
1871         }
1872
1873         /* Must commit after prep above in all cases */
1874         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1875                           objcount, ioo, remote_nb, npages, local_nb, NULL,
1876                           rc);
1877         if (rc == -ENOTCONN)
1878                 /* quota acquire process has been given up because
1879                  * either the client has been evicted or the client
1880                  * has timed out the request already */
1881                 no_reply = true;
1882
1883         /*
1884          * Disable sending mtime back to the client. If the client locked the
1885          * whole object, then it has already updated the mtime on its side,
1886          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1887          */
1888         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1889
1890         if (rc == 0) {
1891                 int nob = 0;
1892
1893                 /* set per-requested niobuf return codes */
1894                 for (i = j = 0; i < niocount; i++) {
1895                         int len = remote_nb[i].len;
1896
1897                         nob += len;
1898                         rcs[i] = 0;
1899                         do {
1900                                 LASSERT(j < npages);
1901                                 if (local_nb[j].rc < 0)
1902                                         rcs[i] = local_nb[j].rc;
1903                                 len -= local_nb[j].len;
1904                                 j++;
1905                         } while (len > 0);
1906                         LASSERT(len == 0);
1907                 }
1908                 LASSERT(j == npages);
1909                 ptlrpc_lprocfs_brw(req, nob);
1910
1911                 tgt_drop_id(exp, &repbody->oa);
1912         }
1913 out_lock:
1914         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
1915         if (desc)
1916                 ptlrpc_free_bulk_nopin(desc);
1917 out:
1918         if (no_reply) {
1919                 req->rq_no_reply = 1;
1920                 /* reply out callback would free */
1921                 ptlrpc_req_drop_rs(req);
1922                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
1923                               "client will retry: rc %d\n",
1924                               exp->exp_obd->obd_name,
1925                               obd_uuid2str(&exp->exp_client_uuid),
1926                               obd_export_nid2str(exp), rc);
1927         }
1928         memory_pressure_clr();
1929         RETURN(rc);
1930 }
1931 EXPORT_SYMBOL(tgt_brw_write);