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
 111         tsi->tsi_fid = body->fid1;
 112
 113         RETURN(rc);
 114 }
 115
 116 /**
 117  * Validate oa from client.
 118  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 119  * req are valid.
 120  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 121  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 122  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 123  *       whether this is oi_fid or real ostid. So it will check
 124  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 125  *    c. Old FID-disable osc will send IDIF.
 126  *    d. new FID-enable osc/osp will send normal FID.
 127  *
 128  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 129  * be used for LAST_ID file, and only being accessed inside OST now.
 130  */
 131 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 132 {
 133         int rc;
 134
 135         ENTRY;
 136
 137         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
 138                      fid_seq_is_echo(oa->o_oi.oi.oi_seq))) {
 139                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 140                  * during create, and we will reset this to 1, since this
 141                  * oi_id is basically useless in the following create process,
 142                  * but oi_id == 0 will make it difficult to tell whether it is
 143                  * real FID or ost_id. */
 144                 oa->o_oi.oi_fid.f_oid = oa->o_oi.oi.oi_id ?: 1;
 145                 oa->o_oi.oi_fid.f_seq = FID_SEQ_ECHO;
 146                 oa->o_oi.oi_fid.f_ver = 0;
 147         } else {
 148                 if (unlikely((oa->o_valid & OBD_MD_FLID &&
 149                               ostid_id(&oa->o_oi) == 0)))
 150                         GOTO(out, rc = -EPROTO);
 151
 152                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 153                  * all of the requests are required to setup FLGROUP */
 154                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 155                         ostid_set_seq_mdt0(&oa->o_oi);
 156                         oa->o_valid |= OBD_MD_FLGROUP;
 157                 }
 158
 159                 if (unlikely(!(fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
 160                                fid_seq_is_mdt0(ostid_seq(&oa->o_oi)) ||
 161                                fid_seq_is_norm(ostid_seq(&oa->o_oi)) ||
 162                                fid_seq_is_echo(ostid_seq(&oa->o_oi)))))
 163                         GOTO(out, rc = -EPROTO);
 164         }
 165         RETURN(0);
 166 out:
 167         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 168                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
 169                ostid_seq(&oa->o_oi), ostid_id(&oa->o_oi), rc);
 170         return rc;
 171 }
 172 EXPORT_SYMBOL(tgt_validate_obdo);
 173
 174 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
 175 {
 176         unsigned                 max_brw;
 177         struct niobuf_remote    *rnb;
 178         struct obd_ioobj        *ioo;
 179         int                      obj_count;
 180
 181         ENTRY;
 182
 183         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
 184         if (ioo == NULL)
 185                 RETURN(-EPROTO);
 186
 187         rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
 188         if (rnb == NULL)
 189                 RETURN(-EPROTO);
 190
 191         max_brw = ioobj_max_brw_get(ioo);
 192         if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 193                 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 194                        ": rc = %d\n", tgt_name(tsi->tsi_tgt),
 195                        obd_export_nid2str(tsi->tsi_exp), max_brw,
 196                        POSTID(oi), -EPROTO);
 197                 RETURN(-EPROTO);
 198         }
 199         ioo->ioo_oid = *oi;
 200
 201         obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
 202                                         RCL_CLIENT) / sizeof(*ioo);
 203         if (obj_count == 0) {
 204                 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
 205                 RETURN(-EPROTO);
 206         } else if (obj_count > 1) {
 207                 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
 208                        obj_count);
 209                 RETURN(-EPROTO);
 210         }
 211
 212         if (ioo->ioo_bufcnt == 0) {
 213                 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
 214                 RETURN(-EPROTO);
 215         }
 216
 217         if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
 218                 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
 219                           "bulk has too many pages (%d)",
 220                           ioo->ioo_bufcnt);
 221                 RETURN(-EPROTO);
 222         }
 223
 224         RETURN(0);
 225 }
 226
 227 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 228 {
 229         struct ost_body         *body;
 230         struct req_capsule      *pill = tsi->tsi_pill;
 231         struct lustre_capa      *capa;
 232         int                      rc;
 233
 234         ENTRY;
 235
 236         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 237         if (body == NULL)
 238                 RETURN(-EFAULT);
 239
 240         rc = tgt_validate_obdo(tsi, &body->oa);
 241         if (rc)
 242                 RETURN(rc);
 243
 244         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 245                 capa = req_capsule_client_get(pill, &RMF_CAPA1);
 246                 if (capa == NULL) {
 247                         CERROR("%s: OSSCAPA flag is set without capability\n",
 248                                tgt_name(tsi->tsi_tgt));
 249                         RETURN(-EFAULT);
 250                 }
 251         }
 252
 253         tsi->tsi_ost_body = body;
 254
 255         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 256                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 257                 if (rc < 0)
 258                         RETURN(rc);
 259         }
 260
 261         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 262                 if (flags & HABEO_CORPUS) {
 263                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 264                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 265                                tgt_name(tsi->tsi_tgt));
 266                         RETURN(-EPROTO);
 267                 } else {
 268                         RETURN(0);
 269                 }
 270         }
 271
 272         rc = ostid_to_fid(&tsi->tsi_fid, &body->oa.o_oi, 0);
 273         if (rc != 0)
 274                 RETURN(rc);
 275
 276         if (!fid_is_sane(&tsi->tsi_fid)) {
 277                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
 278                        PFID(&tsi->tsi_fid));
 279                 RETURN(-EINVAL);
 280         }
 281
 282         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 283
 284         /*
 285          * OST doesn't get object in advance for further use to prevent
 286          * situations with nested object_find which is potential deadlock.
 287          */
 288         tsi->tsi_corpus = NULL;
 289         RETURN(rc);
 290 }
 291
 292 /*
 293  * Do necessary preprocessing according to handler ->th_flags.
 294  */
 295 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 296                                   struct tgt_handler *h,
 297                                   struct ptlrpc_request *req)
 298 {
 299         struct req_capsule      *pill = tsi->tsi_pill;
 300         __u32                    flags = h->th_flags;
 301         int                      rc = 0;
 302
 303         ENTRY;
 304
 305         if (tsi->tsi_preprocessed)
 306                 RETURN(0);
 307
 308         LASSERT(h->th_act != NULL);
 309         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 310         LASSERT(current->journal_info == NULL);
 311
 312         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 313                      h->th_fmt != NULL));
 314         if (h->th_fmt != NULL) {
 315                 req_capsule_set(pill, h->th_fmt);
 316                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 317                         rc = tgt_mdt_body_unpack(tsi, flags);
 318                         if (rc < 0)
 319                                 RETURN(rc);
 320                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 321                                                  RCL_CLIENT)) {
 322                         rc = tgt_ost_body_unpack(tsi, flags);
 323                         if (rc < 0)
 324                                 RETURN(rc);
 325                 }
 326         }
 327
 328         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 329                 RETURN(-EROFS);
 330
 331         if (flags & HABEO_CLAVIS) {
 332                 struct ldlm_request *dlm_req;
 333
 334                 LASSERT(h->th_fmt != NULL);
 335
 336                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 337                 if (dlm_req != NULL) {
 338                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 339                                      LDLM_IBITS &&
 340                                      dlm_req->lock_desc.l_policy_data.\
 341                                      l_inodebits.bits == 0)) {
 342                                 /*
 343                                  * Lock without inodebits makes no sense and
 344                                  * will oops later in ldlm. If client miss to
 345                                  * set such bits, do not trigger ASSERTION.
 346                                  *
 347                                  * For liblustre flock case, it maybe zero.
 348                                  */
 349                                 rc = -EPROTO;
 350                         } else {
 351                                 tsi->tsi_dlm_req = dlm_req;
 352                         }
 353                 } else {
 354                         rc = -EFAULT;
 355                 }
 356         }
 357         tsi->tsi_preprocessed = 1;
 358         RETURN(rc);
 359 }
 360
 361 /*
 362  * Invoke handler for this request opc. Also do necessary preprocessing
 363  * (according to handler ->th_flags), and post-processing (setting of
 364  * ->last_{xid,committed}).
 365  */
 366 static int tgt_handle_request0(struct tgt_session_info *tsi,
 367                                struct tgt_handler *h,
 368                                struct ptlrpc_request *req)
 369 {
 370         int      serious = 0;
 371         int      rc;
 372
 373         ENTRY;
 374
 375         /*
 376          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 377          * to put same checks into handlers like mdt_close(), mdt_reint(),
 378          * etc., without talking to mdt authors first. Checking same thing
 379          * there again is useless and returning 0 error without packing reply
 380          * is buggy! Handlers either pack reply or return error.
 381          *
 382          * We return 0 here and do not send any reply in order to emulate
 383          * network failure. Do not send any reply in case any of NET related
 384          * fail_id has occured.
 385          */
 386         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 387                 RETURN(0);
 388
 389         rc = tgt_request_preprocess(tsi, h, req);
 390         /* pack reply if reply format is fixed */
 391         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 392                 /* Pack reply */
 393                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 394                                           RCL_SERVER))
 395                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 396                                              RCL_SERVER,
 397                                              tsi->tsi_mdt_body->eadatasize);
 398                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 399                                           RCL_SERVER))
 400                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 401                                              RCL_SERVER, 0);
 402
 403                 rc = req_capsule_server_pack(tsi->tsi_pill);
 404         }
 405
 406         if (likely(rc == 0)) {
 407                 /*
 408                  * Process request, there can be two types of rc:
 409                  * 1) errors with msg unpack/pack, other failures outside the
 410                  * operation itself. This is counted as serious errors;
 411                  * 2) errors during fs operation, should be placed in rq_status
 412                  * only
 413                  */
 414                 rc = h->th_act(tsi);
 415                 if (!is_serious(rc) &&
 416                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 417                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 418                                   "pack reply and returned 0 error\n",
 419                                   tgt_name(tsi->tsi_tgt), h->th_name);
 420                         LBUG();
 421                 }
 422                 serious = is_serious(rc);
 423                 rc = clear_serious(rc);
 424         } else {
 425                 serious = 1;
 426         }
 427
 428         req->rq_status = rc;
 429
 430         /*
 431          * ELDLM_* codes which > 0 should be in rq_status only as well as
 432          * all non-serious errors.
 433          */
 434         if (rc > 0 || !serious)
 435                 rc = 0;
 436
 437         LASSERT(current->journal_info == NULL);
 438
 439         if (likely(rc == 0 && req->rq_export))
 440                 target_committed_to_req(req);
 441
 442         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 443         RETURN(0);
 444 }
 445
 446 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 447                                        struct obd_device *obd, int *process)
 448 {
 449         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 450         case MDS_DISCONNECT:
 451         case OST_DISCONNECT:
 452         case OBD_IDX_READ:
 453                 *process = 1;
 454                 RETURN(0);
 455         case MDS_CLOSE:
 456         case MDS_DONE_WRITING:
 457         case MDS_SYNC: /* used in unmounting */
 458         case OBD_PING:
 459         case MDS_REINT:
 460         case UPDATE_OBJ:
 461         case SEQ_QUERY:
 462         case FLD_QUERY:
 463         case LDLM_ENQUEUE:
 464         case OST_CREATE:
 465         case OST_DESTROY:
 466         case OST_PUNCH:
 467         case OST_SETATTR:
 468         case OST_SYNC:
 469         case OST_WRITE:
 470                 *process = target_queue_recovery_request(req, obd);
 471                 RETURN(0);
 472
 473         default:
 474                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 475                 *process = -EAGAIN;
 476                 RETURN(0);
 477         }
 478 }
 479
 480 /*
 481  * Handle recovery. Return:
 482  *        +1: continue request processing;
 483  *       -ve: abort immediately with the given error code;
 484  *         0: send reply with error code in req->rq_status;
 485  */
 486 int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 487 {
 488         ENTRY;
 489
 490         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 491         case MDS_CONNECT:
 492         case OST_CONNECT:
 493         case MGS_CONNECT:
 494         case SEC_CTX_INIT:
 495         case SEC_CTX_INIT_CONT:
 496         case SEC_CTX_FINI:
 497                 RETURN(+1);
 498         }
 499
 500         if (!req->rq_export->exp_obd->obd_replayable)
 501                 RETURN(+1);
 502
 503         /* sanity check: if the xid matches, the request must be marked as a
 504          * resent or replayed */
 505         if (req_xid_is_last(req)) {
 506                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 507                       (MSG_RESENT | MSG_REPLAY))) {
 508                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 509                                   "last_xid, expected REPLAY or RESENT flag "
 510                                   "(%x)", req->rq_xid,
 511                                   lustre_msg_get_flags(req->rq_reqmsg));
 512                         req->rq_status = -ENOTCONN;
 513                         RETURN(-ENOTCONN);
 514                 }
 515         }
 516         /* else: note the opposite is not always true; a RESENT req after a
 517          * failover will usually not match the last_xid, since it was likely
 518          * never committed. A REPLAYed request will almost never match the
 519          * last xid, however it could for a committed, but still retained,
 520          * open. */
 521
 522         /* Check for aborted recovery... */
 523         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 524                 int rc;
 525                 int should_process;
 526
 527                 DEBUG_REQ(D_INFO, req, "Got new replay");
 528                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 529                                                  &should_process);
 530                 if (rc != 0 || !should_process)
 531                         RETURN(rc);
 532                 else if (should_process < 0) {
 533                         req->rq_status = should_process;
 534                         rc = ptlrpc_error(req);
 535                         RETURN(rc);
 536                 }
 537         }
 538         RETURN(+1);
 539 }
 540
 541 /* Initial check for request, it is validation mostly */
 542 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 543 {
 544         struct tgt_handler      *h;
 545         struct tgt_opc_slice    *s;
 546         struct lu_target        *tgt;
 547         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 548
 549         ENTRY;
 550
 551         tgt = class_exp2tgt(req->rq_export);
 552
 553         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 554                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 555                         break;
 556
 557         /* opcode was not found in slice */
 558         if (unlikely(s->tos_hs == NULL)) {
 559                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 560                        opc);
 561                 RETURN(ERR_PTR(-ENOTSUPP));
 562         }
 563
 564         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 565         h = s->tos_hs + (opc - s->tos_opc_start);
 566         if (unlikely(h->th_opc == 0)) {
 567                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 568                 RETURN(ERR_PTR(-ENOTSUPP));
 569         }
 570
 571         RETURN(h);
 572 }
 573
 574 int tgt_request_handle(struct ptlrpc_request *req)
 575 {
 576         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 577
 578         struct lustre_msg       *msg = req->rq_reqmsg;
 579         struct tgt_handler      *h;
 580         struct lu_target        *tgt;
 581         int                      request_fail_id = 0;
 582         __u32                    opc = lustre_msg_get_opc(msg);
 583         int                      rc;
 584
 585         ENTRY;
 586
 587         /* Refill the context, to make sure all thread keys are allocated */
 588         lu_env_refill(req->rq_svc_thread->t_env);
 589
 590         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 591         tsi->tsi_pill = &req->rq_pill;
 592         tsi->tsi_env = req->rq_svc_thread->t_env;
 593
 594         /* if request has export then get handlers slice from corresponding
 595          * target, otherwise that should be connect operation */
 596         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 597             opc == MGS_CONNECT) {
 598                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 599                 rc = target_handle_connect(req);
 600                 if (rc != 0) {
 601                         rc = ptlrpc_error(req);
 602                         GOTO(out, rc);
 603                 }
 604                 /* recovery-small test 18c asks to drop connect reply */
 605                 if (unlikely(opc == OST_CONNECT &&
 606                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 607                         GOTO(out, rc = 0);
 608         }
 609
 610         if (unlikely(!class_connected_export(req->rq_export))) {
 611                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 612                        opc, libcfs_id2str(req->rq_peer));
 613                 req->rq_status = -ENOTCONN;
 614                 rc = ptlrpc_error(req);
 615                 GOTO(out, rc);
 616         }
 617
 618         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 619         tsi->tsi_exp = req->rq_export;
 620         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 621                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 622         else
 623                 tsi->tsi_jobid = NULL;
 624
 625         request_fail_id = tgt->lut_request_fail_id;
 626         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 627
 628         h = tgt_handler_find_check(req);
 629         if (IS_ERR(h)) {
 630                 req->rq_status = PTR_ERR(h);
 631                 rc = ptlrpc_error(req);
 632                 GOTO(out, rc);
 633         }
 634
 635         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 636                 GOTO(out, rc = 0);
 637
 638         rc = lustre_msg_check_version(msg, h->th_version);
 639         if (unlikely(rc)) {
 640                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 641                           " %08x, expecting %08x\n", tgt_name(tgt),
 642                           lustre_msg_get_version(msg), h->th_version);
 643                 req->rq_status = -EINVAL;
 644                 rc = ptlrpc_error(req);
 645                 GOTO(out, rc);
 646         }
 647
 648         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 649         if (likely(rc == 1)) {
 650                 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 651                          h->th_opc, opc);
 652                 rc = tgt_handle_request0(tsi, h, req);
 653                 if (rc)
 654                         GOTO(out, rc);
 655         }
 656         EXIT;
 657 out:
 658         req_capsule_fini(tsi->tsi_pill);
 659         if (tsi->tsi_corpus != NULL) {
 660                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 661                 tsi->tsi_corpus = NULL;
 662         }
 663         return rc;
 664 }
 665 EXPORT_SYMBOL(tgt_request_handle);
 666
 667 /** Assign high priority operations to the request if needed. */
 668 int tgt_hpreq_handler(struct ptlrpc_request *req)
 669 {
 670         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 671         struct tgt_handler      *h;
 672         int                      rc;
 673
 674         ENTRY;
 675
 676         if (req->rq_export == NULL)
 677                 RETURN(0);
 678
 679         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 680         tsi->tsi_pill = &req->rq_pill;
 681         tsi->tsi_env = req->rq_svc_thread->t_env;
 682         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 683         tsi->tsi_exp = req->rq_export;
 684
 685         h = tgt_handler_find_check(req);
 686         if (IS_ERR(h)) {
 687                 rc = PTR_ERR(h);
 688                 RETURN(rc);
 689         }
 690
 691         rc = tgt_request_preprocess(tsi, h, req);
 692         if (unlikely(rc != 0))
 693                 RETURN(rc);
 694
 695         if (h->th_hp != NULL)
 696                 h->th_hp(tsi);
 697         RETURN(0);
 698 }
 699 EXPORT_SYMBOL(tgt_hpreq_handler);
 700
 701 void tgt_counter_incr(struct obd_export *exp, int opcode)
 702 {
 703         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 704         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 705                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 706 }
 707 EXPORT_SYMBOL(tgt_counter_incr);
 708
 709 /*
 710  * Unified target generic handlers.
 711  */
 712
 713 /*
 714  * Security functions
 715  */
 716 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 717 {
 718         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 719                                       OBD_CONNECT_RMT_CLIENT_FORCE |
 720                                       OBD_CONNECT_MDS_CAPA |
 721                                       OBD_CONNECT_OSS_CAPA);
 722 }
 723
 724 static int tgt_init_sec_level(struct ptlrpc_request *req)
 725 {
 726         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 727         char                    *client = libcfs_nid2str(req->rq_peer.nid);
 728         struct obd_connect_data *data, *reply;
 729         int                      rc = 0;
 730         bool                     remote;
 731
 732         ENTRY;
 733
 734         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 735         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 736         if (data == NULL || reply == NULL)
 737                 RETURN(-EFAULT);
 738
 739         /* connection from MDT is always trusted */
 740         if (req->rq_auth_usr_mdt) {
 741                 tgt_init_sec_none(reply);
 742                 RETURN(0);
 743         }
 744
 745         /* no GSS support case */
 746         if (!req->rq_auth_gss) {
 747                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 748                         CWARN("client %s -> target %s does not use GSS, "
 749                               "can not run under security level %d.\n",
 750                               client, tgt_name(tgt), tgt->lut_sec_level);
 751                         RETURN(-EACCES);
 752                 } else {
 753                         tgt_init_sec_none(reply);
 754                         RETURN(0);
 755                 }
 756         }
 757
 758         /* old version case */
 759         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 760                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 761                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 762                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 763                         CWARN("client %s -> target %s uses old version, "
 764                               "can not run under security level %d.\n",
 765                               client, tgt_name(tgt), tgt->lut_sec_level);
 766                         RETURN(-EACCES);
 767                 } else {
 768                         CWARN("client %s -> target %s uses old version, "
 769                               "run under security level %d.\n",
 770                               client, tgt_name(tgt), tgt->lut_sec_level);
 771                         tgt_init_sec_none(reply);
 772                         RETURN(0);
 773                 }
 774         }
 775
 776         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 777         if (remote) {
 778                 if (!req->rq_auth_remote)
 779                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 780                                "asked to be remote.\n", client, tgt_name(tgt));
 781         } else if (req->rq_auth_remote) {
 782                 remote = true;
 783                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 784                        "as remote by default.\n", client, tgt_name(tgt));
 785         }
 786
 787         if (remote) {
 788                 if (!tgt->lut_oss_capa) {
 789                         CDEBUG(D_SEC,
 790                                "client %s -> target %s is set as remote,"
 791                                " but OSS capabilities are not enabled: %d.\n",
 792                                client, tgt_name(tgt), tgt->lut_oss_capa);
 793                         RETURN(-EACCES);
 794                 }
 795         } else {
 796                 if (req->rq_auth_uid == INVALID_UID) {
 797                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 798                                "authenticated!\n", client, tgt_name(tgt));
 799                         RETURN(-EACCES);
 800                 }
 801         }
 802
 803
 804         switch (tgt->lut_sec_level) {
 805         case LUSTRE_SEC_NONE:
 806                 if (remote) {
 807                         CDEBUG(D_SEC,
 808                                "client %s -> target %s is set as remote, "
 809                                "can not run under security level %d.\n",
 810                                client, tgt_name(tgt), tgt->lut_sec_level);
 811                         RETURN(-EACCES);
 812                 }
 813                 tgt_init_sec_none(reply);
 814                 break;
 815         case LUSTRE_SEC_REMOTE:
 816                 if (!remote)
 817                         tgt_init_sec_none(reply);
 818                 break;
 819         case LUSTRE_SEC_ALL:
 820                 if (remote)
 821                         break;
 822                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 823                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 824                 if (!tgt->lut_oss_capa)
 825                         reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 826                 if (!tgt->lut_mds_capa)
 827                         reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 828                 break;
 829         default:
 830                 RETURN(-EINVAL);
 831         }
 832
 833         RETURN(rc);
 834 }
 835
 836 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 837 {
 838         struct lu_target        *tgt = class_exp2tgt(exp);
 839         struct sptlrpc_flavor    flvr;
 840         int                      rc = 0;
 841
 842         LASSERT(tgt);
 843         LASSERT(tgt->lut_obd);
 844         LASSERT(tgt->lut_slice);
 845
 846         /* always allow ECHO client */
 847         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 848                             LUSTRE_ECHO_NAME) == 0)) {
 849                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 850                 return 0;
 851         }
 852
 853         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 854                 read_lock(&tgt->lut_sptlrpc_lock);
 855                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 856                                              req->rq_sp_from,
 857                                              req->rq_peer.nid,
 858                                              &flvr);
 859                 read_unlock(&tgt->lut_sptlrpc_lock);
 860
 861                 spin_lock(&exp->exp_lock);
 862                 exp->exp_sp_peer = req->rq_sp_from;
 863                 exp->exp_flvr = flvr;
 864                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 865                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 866                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 867                                "expect %x\n", tgt_name(tgt),
 868                                req->rq_flvr.sf_rpc,
 869                                libcfs_nid2str(req->rq_peer.nid),
 870                                exp->exp_flvr.sf_rpc);
 871                         rc = -EACCES;
 872                 }
 873                 spin_unlock(&exp->exp_lock);
 874         } else {
 875                 if (exp->exp_sp_peer != req->rq_sp_from) {
 876                         CERROR("%s: RPC source %s doesn't match %s\n",
 877                                tgt_name(tgt),
 878                                sptlrpc_part2name(req->rq_sp_from),
 879                                sptlrpc_part2name(exp->exp_sp_peer));
 880                         rc = -EACCES;
 881                 } else {
 882                         rc = sptlrpc_target_export_check(exp, req);
 883                 }
 884         }
 885
 886         return rc;
 887 }
 888
 889 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 890 {
 891         struct sptlrpc_rule_set  tmp_rset;
 892         int                      rc;
 893
 894         sptlrpc_rule_set_init(&tmp_rset);
 895         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 896         if (rc) {
 897                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 898                        tgt_name(tgt), rc);
 899                 return rc;
 900         }
 901
 902         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 903
 904         write_lock(&tgt->lut_sptlrpc_lock);
 905         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 906         tgt->lut_sptlrpc_rset = tmp_rset;
 907         write_unlock(&tgt->lut_sptlrpc_lock);
 908
 909         return 0;
 910 }
 911 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 912
 913 int tgt_connect(struct tgt_session_info *tsi)
 914 {
 915         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 916         struct obd_connect_data *reply;
 917         int                      rc;
 918
 919         ENTRY;
 920
 921         rc = tgt_init_sec_level(req);
 922         if (rc != 0)
 923                 GOTO(out, rc);
 924
 925         /* XXX: better to call this check right after getting new export but
 926          * before last_rcvd slot allocation to avoid server load upon insecure
 927          * connects. This is to be fixed after unifiyng all targets.
 928          */
 929         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
 930         if (rc)
 931                 GOTO(out, rc);
 932
 933         /* To avoid exposing partially initialized connection flags, changes up
 934          * to this point have been staged in reply->ocd_connect_flags. Now that
 935          * connection handling has completed successfully, atomically update
 936          * the connect flags in the shared export data structure. LU-1623 */
 937         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
 938         spin_lock(&tsi->tsi_exp->exp_lock);
 939         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
 940         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
 941         spin_unlock(&tsi->tsi_exp->exp_lock);
 942
 943         RETURN(0);
 944 out:
 945         obd_disconnect(class_export_get(tsi->tsi_exp));
 946         return rc;
 947 }
 948 EXPORT_SYMBOL(tgt_connect);
 949
 950 int tgt_disconnect(struct tgt_session_info *tsi)
 951 {
 952         int rc;
 953
 954         ENTRY;
 955
 956         rc = target_handle_disconnect(tgt_ses_req(tsi));
 957         if (rc)
 958                 RETURN(err_serious(rc));
 959
 960         RETURN(rc);
 961 }
 962 EXPORT_SYMBOL(tgt_disconnect);
 963
 964 /*
 965  * Unified target OBD handlers
 966  */
 967 int tgt_obd_ping(struct tgt_session_info *tsi)
 968 {
 969         int rc;
 970
 971         ENTRY;
 972
 973         rc = target_handle_ping(tgt_ses_req(tsi));
 974         if (rc)
 975                 RETURN(err_serious(rc));
 976
 977         RETURN(rc);
 978 }
 979 EXPORT_SYMBOL(tgt_obd_ping);
 980
 981 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
 982 {
 983         return err_serious(-EOPNOTSUPP);
 984 }
 985 EXPORT_SYMBOL(tgt_obd_log_cancel);
 986
 987 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
 988 {
 989         return err_serious(-EOPNOTSUPP);
 990 }
 991 EXPORT_SYMBOL(tgt_obd_qc_callback);
 992
 993 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
 994 {
 995         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
 996         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 997         struct obd_export       *exp = req->rq_export;
 998         struct ptlrpc_bulk_desc *desc;
 999         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1000         int                      tmpcount;
1001         int                      tmpsize;
1002         int                      i;
1003         int                      rc;
1004
1005         ENTRY;
1006
1007         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1008                                     MDS_BULK_PORTAL);
1009         if (desc == NULL)
1010                 RETURN(-ENOMEM);
1011
1012         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1013                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1014                  * which is rdpg->rp_count */
1015                 nob = rdpg->rp_count;
1016
1017         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1018              i++, tmpcount -= tmpsize) {
1019                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1020                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1021         }
1022
1023         LASSERT(desc->bd_nob == nob);
1024         rc = target_bulk_io(exp, desc, lwi);
1025         ptlrpc_free_bulk_pin(desc);
1026         RETURN(rc);
1027 }
1028 EXPORT_SYMBOL(tgt_sendpage);
1029
1030 /*
1031  * OBD_IDX_READ handler
1032  */
1033 int tgt_obd_idx_read(struct tgt_session_info *tsi)
1034 {
1035         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1036         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1037         struct idx_info         *req_ii, *rep_ii;
1038         int                      rc, i;
1039
1040         ENTRY;
1041
1042         memset(rdpg, 0, sizeof(*rdpg));
1043         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1044
1045         /* extract idx_info buffer from request & reply */
1046         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1047         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1048                 RETURN(err_serious(-EPROTO));
1049
1050         rc = req_capsule_server_pack(tsi->tsi_pill);
1051         if (rc)
1052                 RETURN(err_serious(rc));
1053
1054         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1055         if (rep_ii == NULL)
1056                 RETURN(err_serious(-EFAULT));
1057         rep_ii->ii_magic = IDX_INFO_MAGIC;
1058
1059         /* extract hash to start with */
1060         rdpg->rp_hash = req_ii->ii_hash_start;
1061
1062         /* extract requested attributes */
1063         rdpg->rp_attrs = req_ii->ii_attrs;
1064
1065         /* check that fid packed in request is valid and supported */
1066         if (!fid_is_sane(&req_ii->ii_fid))
1067                 RETURN(-EINVAL);
1068         rep_ii->ii_fid = req_ii->ii_fid;
1069
1070         /* copy flags */
1071         rep_ii->ii_flags = req_ii->ii_flags;
1072
1073         /* compute number of pages to allocate, ii_count is the number of 4KB
1074          * containers */
1075         if (req_ii->ii_count <= 0)
1076                 GOTO(out, rc = -EFAULT);
1077         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1078                                exp_max_brw_size(tsi->tsi_exp));
1079         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1080
1081         /* allocate pages to store the containers */
1082         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1083         if (rdpg->rp_pages == NULL)
1084                 GOTO(out, rc = -ENOMEM);
1085         for (i = 0; i < rdpg->rp_npages; i++) {
1086                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1087                 if (rdpg->rp_pages[i] == NULL)
1088                         GOTO(out, rc = -ENOMEM);
1089         }
1090
1091         /* populate pages with key/record pairs */
1092         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1093         if (rc < 0)
1094                 GOTO(out, rc);
1095
1096         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1097                  "asked %d > %d\n", rc, rdpg->rp_count);
1098
1099         /* send pages to client */
1100         rc = tgt_sendpage(tsi, rdpg, rc);
1101         if (rc)
1102                 GOTO(out, rc);
1103         EXIT;
1104 out:
1105         if (rdpg->rp_pages) {
1106                 for (i = 0; i < rdpg->rp_npages; i++)
1107                         if (rdpg->rp_pages[i])
1108                                 __free_page(rdpg->rp_pages[i]);
1109                 OBD_FREE(rdpg->rp_pages,
1110                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1111         }
1112         return rc;
1113 }
1114 EXPORT_SYMBOL(tgt_obd_idx_read);
1115
1116 struct tgt_handler tgt_obd_handlers[] = {
1117 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1118 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1119 TGT_OBD_HDL_VAR(0,      OBD_QC_CALLBACK,        tgt_obd_qc_callback),
1120 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1121 };
1122 EXPORT_SYMBOL(tgt_obd_handlers);
1123
1124 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1125              struct dt_object *obj)
1126 {
1127         int rc = 0;
1128
1129         ENTRY;
1130
1131         /* if no objid is specified, it means "sync whole filesystem" */
1132         if (obj == NULL) {
1133                 rc = dt_sync(env, tgt->lut_bottom);
1134         } else if (dt_version_get(env, obj) >
1135                    tgt->lut_obd->obd_last_committed) {
1136                 rc = dt_object_sync(env, obj);
1137         }
1138
1139         RETURN(rc);
1140 }
1141 EXPORT_SYMBOL(tgt_sync);
1142 /*
1143  * Unified target DLM handlers.
1144  */
1145
1146 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1147  * (if requested) */
1148 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1149                      void *data, int flag)
1150 {
1151         struct lu_env            env;
1152         struct lu_target        *tgt;
1153         struct dt_object        *obj;
1154         struct lu_fid            fid;
1155         int                      rc = 0;
1156
1157         ENTRY;
1158
1159         tgt = class_exp2tgt(lock->l_export);
1160
1161         if (flag == LDLM_CB_CANCELING &&
1162             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1163             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1164              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1165               lock->l_flags & LDLM_FL_CBPENDING))) {
1166                 rc = lu_env_init(&env, LCT_DT_THREAD);
1167                 if (unlikely(rc != 0))
1168                         RETURN(rc);
1169
1170                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name);
1171                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1172                 if (IS_ERR(obj))
1173                         GOTO(err_env, rc = PTR_ERR(obj));
1174
1175                 if (!dt_object_exists(obj))
1176                         GOTO(err_put, rc = -ENOENT);
1177
1178                 rc = tgt_sync(&env, tgt, obj);
1179                 if (rc < 0) {
1180                         CERROR("%s: sync failed on lock cancel: rc = %d\n",
1181                                tgt_name(tgt), rc);
1182                 }
1183 err_put:
1184                 lu_object_put(&env, &obj->do_lu);
1185 err_env:
1186                 lu_env_fini(&env);
1187         }
1188
1189         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1190         RETURN(rc);
1191 }
1192
1193 struct ldlm_callback_suite tgt_dlm_cbs = {
1194         .lcs_completion = ldlm_server_completion_ast,
1195         .lcs_blocking   = tgt_blocking_ast,
1196         .lcs_glimpse    = ldlm_server_glimpse_ast
1197 };
1198
1199 int tgt_enqueue(struct tgt_session_info *tsi)
1200 {
1201         struct ptlrpc_request *req = tgt_ses_req(tsi);
1202         int rc;
1203
1204         ENTRY;
1205         /*
1206          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1207          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1208          */
1209         LASSERT(tsi->tsi_dlm_req != NULL);
1210         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1211                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1212         if (rc)
1213                 RETURN(err_serious(rc));
1214
1215         RETURN(req->rq_status);
1216 }
1217 EXPORT_SYMBOL(tgt_enqueue);
1218
1219 int tgt_convert(struct tgt_session_info *tsi)
1220 {
1221         struct ptlrpc_request *req = tgt_ses_req(tsi);
1222         int rc;
1223
1224         ENTRY;
1225         LASSERT(tsi->tsi_dlm_req);
1226         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1227         if (rc)
1228                 RETURN(err_serious(rc));
1229
1230         RETURN(req->rq_status);
1231 }
1232 EXPORT_SYMBOL(tgt_convert);
1233
1234 int tgt_bl_callback(struct tgt_session_info *tsi)
1235 {
1236         return err_serious(-EOPNOTSUPP);
1237 }
1238 EXPORT_SYMBOL(tgt_bl_callback);
1239
1240 int tgt_cp_callback(struct tgt_session_info *tsi)
1241 {
1242         return err_serious(-EOPNOTSUPP);
1243 }
1244 EXPORT_SYMBOL(tgt_cp_callback);
1245
1246 /* generic LDLM target handler */
1247 struct tgt_handler tgt_dlm_handlers[] = {
1248 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1249 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1250 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1251 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1252 };
1253 EXPORT_SYMBOL(tgt_dlm_handlers);
1254
1255 /*
1256  * Unified target LLOG handlers.
1257  */
1258 int tgt_llog_open(struct tgt_session_info *tsi)
1259 {
1260         int rc;
1261
1262         ENTRY;
1263
1264         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1265
1266         RETURN(rc);
1267 }
1268 EXPORT_SYMBOL(tgt_llog_open);
1269
1270 int tgt_llog_close(struct tgt_session_info *tsi)
1271 {
1272         int rc;
1273
1274         ENTRY;
1275
1276         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1277
1278         RETURN(rc);
1279 }
1280 EXPORT_SYMBOL(tgt_llog_close);
1281
1282
1283 int tgt_llog_destroy(struct tgt_session_info *tsi)
1284 {
1285         int rc;
1286
1287         ENTRY;
1288
1289         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1290
1291         RETURN(rc);
1292 }
1293 EXPORT_SYMBOL(tgt_llog_destroy);
1294
1295 int tgt_llog_read_header(struct tgt_session_info *tsi)
1296 {
1297         int rc;
1298
1299         ENTRY;
1300
1301         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1302
1303         RETURN(rc);
1304 }
1305 EXPORT_SYMBOL(tgt_llog_read_header);
1306
1307 int tgt_llog_next_block(struct tgt_session_info *tsi)
1308 {
1309         int rc;
1310
1311         ENTRY;
1312
1313         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1314
1315         RETURN(rc);
1316 }
1317 EXPORT_SYMBOL(tgt_llog_next_block);
1318
1319 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1320 {
1321         int rc;
1322
1323         ENTRY;
1324
1325         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1326
1327         RETURN(rc);
1328 }
1329 EXPORT_SYMBOL(tgt_llog_prev_block);
1330
1331 /* generic llog target handler */
1332 struct tgt_handler tgt_llog_handlers[] = {
1333 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1334 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1335 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1336 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1337 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1338 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1339 };
1340 EXPORT_SYMBOL(tgt_llog_handlers);
1341
1342 /*
1343  * sec context handlers
1344  */
1345 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1346 int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1347 {
1348         return 0;
1349 }
1350
1351 struct tgt_handler tgt_sec_ctx_handlers[] = {
1352 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1353 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1354 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1355 };
1356 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1357
1358 /*
1359  * initialize per-thread page pool (bug 5137).
1360  */
1361 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1362 {
1363         struct tgt_thread_big_cache *tbc;
1364
1365         ENTRY;
1366
1367         LASSERT(thread != NULL);
1368         LASSERT(thread->t_data == NULL);
1369
1370         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1371         if (tbc == NULL)
1372                 RETURN(-ENOMEM);
1373         thread->t_data = tbc;
1374         RETURN(0);
1375 }
1376 EXPORT_SYMBOL(tgt_io_thread_init);
1377
1378 /*
1379  * free per-thread pool created by tgt_thread_init().
1380  */
1381 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1382 {
1383         struct tgt_thread_big_cache *tbc;
1384
1385         ENTRY;
1386
1387         LASSERT(thread != NULL);
1388
1389         /*
1390          * be prepared to handle partially-initialized pools (because this is
1391          * called from ost_io_thread_init() for cleanup.
1392          */
1393         tbc = thread->t_data;
1394         if (tbc != NULL) {
1395                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1396                 thread->t_data = NULL;
1397         }
1398         EXIT;
1399 }
1400 EXPORT_SYMBOL(tgt_io_thread_done);
1401 /**
1402  * Helper function for getting server side [start, start+count] DLM lock
1403  * if asked by client.
1404  */
1405 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1406                     __u64 start, __u64 end, struct lustre_handle *lh,
1407                     int mode, __u64 *flags)
1408 {
1409         ldlm_policy_data_t       policy;
1410         int                      rc;
1411
1412         ENTRY;
1413
1414         LASSERT(lh != NULL);
1415         LASSERT(ns != NULL);
1416         LASSERT(!lustre_handle_is_used(lh));
1417
1418         policy.l_extent.gid = 0;
1419         policy.l_extent.start = start & CFS_PAGE_MASK;
1420
1421         /*
1422          * If ->o_blocks is EOF it means "lock till the end of the file".
1423          * Otherwise, it's size of an extent or hole being punched (in bytes).
1424          */
1425         if (end == OBD_OBJECT_EOF || end < start)
1426                 policy.l_extent.end = OBD_OBJECT_EOF;
1427         else
1428                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1429
1430         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1431                                     flags, ldlm_blocking_ast,
1432                                     ldlm_completion_ast, ldlm_glimpse_ast,
1433                                     NULL, 0, LVB_T_NONE, NULL, lh);
1434         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1435 }
1436 EXPORT_SYMBOL(tgt_extent_lock);
1437
1438 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1439 {
1440         LASSERT(lustre_handle_is_used(lh));
1441         ldlm_lock_decref(lh, mode);
1442 }
1443 EXPORT_SYMBOL(tgt_extent_unlock);
1444
1445 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1446                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1447                  struct lustre_handle *lh, int mode)
1448 {
1449         __u64                    flags = 0;
1450         int                      nrbufs = obj->ioo_bufcnt;
1451         int                      i;
1452
1453         ENTRY;
1454
1455         LASSERT(mode == LCK_PR || mode == LCK_PW);
1456         LASSERT(!lustre_handle_is_used(lh));
1457
1458         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1459                 RETURN(0);
1460
1461         for (i = 1; i < nrbufs; i++)
1462                 if (!(nb[i].flags & OBD_BRW_SRVLOCK))
1463                         RETURN(-EFAULT);
1464
1465         RETURN(tgt_extent_lock(ns, res_id, nb[0].offset,
1466                                nb[nrbufs - 1].offset + nb[nrbufs - 1].len - 1,
1467                                lh, mode, &flags));
1468 }
1469 EXPORT_SYMBOL(tgt_brw_lock);
1470
1471 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1472                     struct lustre_handle *lh, int mode)
1473 {
1474         ENTRY;
1475
1476         LASSERT(mode == LCK_PR || mode == LCK_PW);
1477         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
1478                 lustre_handle_is_used(lh));
1479         if (lustre_handle_is_used(lh))
1480                 tgt_extent_unlock(lh, mode);
1481         EXIT;
1482 }
1483 EXPORT_SYMBOL(tgt_brw_unlock);
1484
1485 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1486                                struct ptlrpc_bulk_desc *desc, int opc,
1487                                cksum_type_t cksum_type)
1488 {
1489         struct cfs_crypto_hash_desc     *hdesc;
1490         unsigned int                    bufsize;
1491         int                             i, err;
1492         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1493         __u32                           cksum;
1494
1495         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1496         if (IS_ERR(hdesc)) {
1497                 CERROR("%s: unable to initialize checksum hash %s\n",
1498                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1499                 return PTR_ERR(hdesc);
1500         }
1501
1502         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1503         for (i = 0; i < desc->bd_iov_count; i++) {
1504                 /* corrupt the data before we compute the checksum, to
1505                  * simulate a client->OST data error */
1506                 if (i == 0 && opc == OST_WRITE &&
1507                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1508                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1509                         int len = desc->bd_iov[i].kiov_len;
1510                         struct page *np = tgt_page_to_corrupt;
1511                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1512
1513                         if (np) {
1514                                 char *ptr2 = kmap(np) + off;
1515
1516                                 memcpy(ptr2, ptr, len);
1517                                 memcpy(ptr2, "bad3", min(4, len));
1518                                 kunmap(np);
1519                                 desc->bd_iov[i].kiov_page = np;
1520                         } else {
1521                                 CERROR("%s: can't alloc page for corruption\n",
1522                                        tgt_name(tgt));
1523                         }
1524                 }
1525                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1526                                   desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1527                                   desc->bd_iov[i].kiov_len);
1528
1529                  /* corrupt the data after we compute the checksum, to
1530                  * simulate an OST->client data error */
1531                 if (i == 0 && opc == OST_READ &&
1532                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1533                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1534                         int len = desc->bd_iov[i].kiov_len;
1535                         struct page *np = tgt_page_to_corrupt;
1536                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1537
1538                         if (np) {
1539                                 char *ptr2 = kmap(np) + off;
1540
1541                                 memcpy(ptr2, ptr, len);
1542                                 memcpy(ptr2, "bad4", min(4, len));
1543                                 kunmap(np);
1544                                 desc->bd_iov[i].kiov_page = np;
1545                         } else {
1546                                 CERROR("%s: can't alloc page for corruption\n",
1547                                        tgt_name(tgt));
1548                         }
1549                 }
1550         }
1551
1552         bufsize = 4;
1553         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1554         if (err)
1555                 cfs_crypto_hash_final(hdesc, NULL, NULL);
1556
1557         return cksum;
1558 }
1559
1560 int tgt_brw_read(struct tgt_session_info *tsi)
1561 {
1562         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1563         struct ptlrpc_bulk_desc *desc = NULL;
1564         struct obd_export       *exp = tsi->tsi_exp;
1565         struct niobuf_remote    *remote_nb;
1566         struct niobuf_local     *local_nb;
1567         struct obd_ioobj        *ioo;
1568         struct ost_body         *body, *repbody;
1569         struct l_wait_info       lwi;
1570         struct lustre_handle     lockh = { 0 };
1571         int                      niocount, npages, nob = 0, rc, i;
1572         int                      no_reply = 0;
1573         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1574
1575         ENTRY;
1576
1577         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1578                 CERROR("%s: deny read request from %s to portal %u\n",
1579                        tgt_name(tsi->tsi_tgt),
1580                        obd_export_nid2str(req->rq_export),
1581                        ptlrpc_req2svc(req)->srv_req_portal);
1582                 RETURN(-EPROTO);
1583         }
1584
1585         req->rq_bulk_read = 1;
1586
1587         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1588                 RETURN(-EIO);
1589
1590         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1591                          cfs_fail_val : (obd_timeout + 1) / 4);
1592
1593         /* Check if there is eviction in progress, and if so, wait for it to
1594          * finish */
1595         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1596                 /* We do not care how long it takes */
1597                 lwi = LWI_INTR(NULL, NULL);
1598                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1599                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1600                          &lwi);
1601         }
1602
1603         /* There must be big cache in current thread to process this request
1604          * if it is NULL then something went wrong and it wasn't allocated,
1605          * report -ENOMEM in that case */
1606         if (tbc == NULL)
1607                 RETURN(-ENOMEM);
1608
1609         body = tsi->tsi_ost_body;
1610         LASSERT(body != NULL);
1611
1612         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1613         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1614
1615         niocount = ioo->ioo_bufcnt;
1616         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1617         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1618
1619         local_nb = tbc->local;
1620
1621         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1622                           remote_nb, &lockh, LCK_PR);
1623         if (rc != 0)
1624                 RETURN(rc);
1625
1626         /*
1627          * If getting the lock took more time than
1628          * client was willing to wait, drop it. b=11330
1629          */
1630         if (cfs_time_current_sec() > req->rq_deadline ||
1631             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1632                 no_reply = 1;
1633                 CERROR("Dropping timed-out read from %s because locking"
1634                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1635                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1636                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1637                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1638                 GOTO(out_lock, rc = -ETIMEDOUT);
1639         }
1640
1641         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1642         repbody->oa = body->oa;
1643
1644         npages = PTLRPC_MAX_BRW_PAGES;
1645         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1646                         ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1647         if (rc != 0)
1648                 GOTO(out_lock, rc);
1649
1650         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1651                                     BULK_PUT_SOURCE, OST_BULK_PORTAL);
1652         if (desc == NULL)
1653                 GOTO(out_commitrw, rc = -ENOMEM);
1654
1655         nob = 0;
1656         for (i = 0; i < npages; i++) {
1657                 int page_rc = local_nb[i].rc;
1658
1659                 if (page_rc < 0) {
1660                         rc = page_rc;
1661                         break;
1662                 }
1663
1664                 nob += page_rc;
1665                 if (page_rc != 0) { /* some data! */
1666                         LASSERT(local_nb[i].page != NULL);
1667                         ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1668                                                     local_nb[i].lnb_page_offset,
1669                                                     page_rc);
1670                 }
1671
1672                 if (page_rc != local_nb[i].len) { /* short read */
1673                         /* All subsequent pages should be 0 */
1674                         while (++i < npages)
1675                                 LASSERT(local_nb[i].rc == 0);
1676                         break;
1677                 }
1678         }
1679
1680         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1681                 cksum_type_t cksum_type =
1682                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1683                                           body->oa.o_flags : 0);
1684                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1685                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1686                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1687                                                         OST_READ, cksum_type);
1688                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1689                        repbody->oa.o_cksum);
1690         } else {
1691                 repbody->oa.o_valid = 0;
1692         }
1693         /* We're finishing using body->oa as an input variable */
1694
1695         /* Check if client was evicted while we were doing i/o before touching
1696          * network */
1697         if (likely(rc == 0 &&
1698                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1699                 rc = target_bulk_io(exp, desc, &lwi);
1700                 no_reply = rc != 0;
1701         }
1702
1703 out_commitrw:
1704         /* Must commit after prep above in all cases */
1705         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1706                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1707                           NULL, rc);
1708         if (rc == 0)
1709                 tgt_drop_id(exp, &repbody->oa);
1710 out_lock:
1711         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1712
1713         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1714                 ptlrpc_free_bulk_nopin(desc);
1715
1716         LASSERT(rc <= 0);
1717         if (rc == 0) {
1718                 rc = nob;
1719                 ptlrpc_lprocfs_brw(req, nob);
1720         } else if (no_reply) {
1721                 req->rq_no_reply = 1;
1722                 /* reply out callback would free */
1723                 ptlrpc_req_drop_rs(req);
1724                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1725                               "client will retry: rc %d\n",
1726                               exp->exp_obd->obd_name,
1727                               obd_uuid2str(&exp->exp_client_uuid),
1728                               obd_export_nid2str(exp), rc);
1729         }
1730         /* send a bulk after reply to simulate a network delay or reordering
1731          * by a router */
1732         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1733                 wait_queue_head_t        waitq;
1734                 struct l_wait_info       lwi1;
1735
1736                 CDEBUG(D_INFO, "reorder BULK\n");
1737                 init_waitqueue_head(&waitq);
1738
1739                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1740                 l_wait_event(waitq, 0, &lwi1);
1741                 target_bulk_io(exp, desc, &lwi);
1742                 ptlrpc_free_bulk_nopin(desc);
1743         }
1744
1745         RETURN(rc);
1746 }
1747 EXPORT_SYMBOL(tgt_brw_read);
1748
1749 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1750                               struct ptlrpc_bulk_desc *desc,
1751                               struct niobuf_local *local_nb, int npages,
1752                               obd_count client_cksum, obd_count server_cksum,
1753                               bool mmap)
1754 {
1755         struct obd_export *exp = req->rq_export;
1756         struct ost_body *body;
1757         char *router;
1758         char *via;
1759
1760         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1761         LASSERT(body != NULL);
1762
1763         if (req->rq_peer.nid == desc->bd_sender) {
1764                 via = router = "";
1765         } else {
1766                 via = " via ";
1767                 router = libcfs_nid2str(desc->bd_sender);
1768         }
1769
1770         if (mmap) {
1771                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1772                              client_cksum, server_cksum);
1773                 return;
1774         }
1775
1776         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1777                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1778                            "]: client csum %x, server csum %x\n",
1779                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1780                            via, router,
1781                            body->oa.o_valid & OBD_MD_FLFID ?
1782                            body->oa.o_parent_seq : (__u64)0,
1783                            body->oa.o_valid & OBD_MD_FLFID ?
1784                            body->oa.o_parent_oid : 0,
1785                            body->oa.o_valid & OBD_MD_FLFID ?
1786                            body->oa.o_parent_ver : 0,
1787                            POSTID(&body->oa.o_oi),
1788                            local_nb[0].lnb_file_offset,
1789                            local_nb[npages-1].lnb_file_offset +
1790                            local_nb[npages-1].len - 1,
1791                            client_cksum, server_cksum);
1792 }
1793
1794 int tgt_brw_write(struct tgt_session_info *tsi)
1795 {
1796         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1797         struct ptlrpc_bulk_desc *desc = NULL;
1798         struct obd_export       *exp = req->rq_export;
1799         struct niobuf_remote    *remote_nb;
1800         struct niobuf_local     *local_nb;
1801         struct obd_ioobj        *ioo;
1802         struct ost_body         *body, *repbody;
1803         struct l_wait_info       lwi;
1804         struct lustre_handle     lockh = {0};
1805         __u32                   *rcs;
1806         int                      objcount, niocount, npages;
1807         int                      rc, i, j;
1808         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1809         bool                     no_reply = false, mmap;
1810         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1811
1812         ENTRY;
1813
1814         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1815                 CERROR("%s: deny write request from %s to portal %u\n",
1816                        tgt_name(tsi->tsi_tgt),
1817                        obd_export_nid2str(req->rq_export),
1818                        ptlrpc_req2svc(req)->srv_req_portal);
1819                 RETURN(err_serious(-EPROTO));
1820         }
1821
1822         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1823                 RETURN(err_serious(-ENOSPC));
1824         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1825                 RETURN(err_serious(-EROFS));
1826
1827         req->rq_bulk_write = 1;
1828
1829         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1830                 RETURN(err_serious(-EIO));
1831         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1832                 RETURN(err_serious(-EFAULT));
1833
1834         /* pause before transaction has been started */
1835         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1836                          cfs_fail_val : (obd_timeout + 1) / 4);
1837
1838         /* There must be big cache in current thread to process this request
1839          * if it is NULL then something went wrong and it wasn't allocated,
1840          * report -ENOMEM in that case */
1841         if (tbc == NULL)
1842                 RETURN(-ENOMEM);
1843
1844         body = tsi->tsi_ost_body;
1845         LASSERT(body != NULL);
1846
1847         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1848         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1849
1850         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1851                                         RCL_CLIENT) / sizeof(*ioo);
1852
1853         for (niocount = i = 0; i < objcount; i++)
1854                 niocount += ioo[i].ioo_bufcnt;
1855
1856         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1857         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1858         if (niocount != req_capsule_get_size(&req->rq_pill,
1859                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1860                         sizeof(*remote_nb))
1861                 RETURN(err_serious(-EPROTO));
1862
1863         if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
1864             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1865                 memory_pressure_set();
1866
1867         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1868                              niocount * sizeof(*rcs));
1869         rc = req_capsule_server_pack(&req->rq_pill);
1870         if (rc != 0)
1871                 GOTO(out, rc = err_serious(rc));
1872
1873         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1874         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1875
1876         local_nb = tbc->local;
1877
1878         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1879                           remote_nb, &lockh, LCK_PW);
1880         if (rc != 0)
1881                 GOTO(out, rc);
1882
1883         /*
1884          * If getting the lock took more time than
1885          * client was willing to wait, drop it. b=11330
1886          */
1887         if (cfs_time_current_sec() > req->rq_deadline ||
1888             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1889                 no_reply = true;
1890                 CERROR("%s: Dropping timed-out write from %s because locking "
1891                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1892                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1893                        POSTID(&ioo->ioo_oid),
1894                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1895                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1896                 GOTO(out_lock, rc = -ETIMEDOUT);
1897         }
1898
1899         /* Because we already sync grant info with client when reconnect,
1900          * grant info will be cleared for resent req, then fed_grant and
1901          * total_grant will not be modified in following preprw_write */
1902         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1903                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1904                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1905         }
1906
1907         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1908         if (repbody == NULL)
1909                 GOTO(out_lock, rc = -ENOMEM);
1910         repbody->oa = body->oa;
1911
1912         npages = PTLRPC_MAX_BRW_PAGES;
1913         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1914                         objcount, ioo, remote_nb, &npages, local_nb, NULL,
1915                         BYPASS_CAPA);
1916         if (rc < 0)
1917                 GOTO(out_lock, rc);
1918
1919         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1920                                     BULK_GET_SINK, OST_BULK_PORTAL);
1921         if (desc == NULL)
1922                 GOTO(skip_transfer, rc = -ENOMEM);
1923
1924         /* NB Having prepped, we must commit... */
1925         for (i = 0; i < npages; i++)
1926                 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1927                                             local_nb[i].lnb_page_offset,
1928                                             local_nb[i].len);
1929
1930         rc = sptlrpc_svc_prep_bulk(req, desc);
1931         if (rc != 0)
1932                 GOTO(skip_transfer, rc);
1933
1934         rc = target_bulk_io(exp, desc, &lwi);
1935         no_reply = rc != 0;
1936
1937 skip_transfer:
1938         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
1939                 static int cksum_counter;
1940
1941                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1942                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1943
1944                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1945                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1946                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1947                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1948                                                         OST_WRITE, cksum_type);
1949                 cksum_counter++;
1950
1951                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
1952                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
1953                                 body->oa.o_flags & OBD_FL_MMAP);
1954
1955                         tgt_warn_on_cksum(req, desc, local_nb, npages,
1956                                           body->oa.o_cksum,
1957                                           repbody->oa.o_cksum, mmap);
1958                         cksum_counter = 0;
1959                 } else if ((cksum_counter & (-cksum_counter)) ==
1960                            cksum_counter) {
1961                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1962                                cksum_counter, libcfs_id2str(req->rq_peer),
1963                                repbody->oa.o_cksum);
1964                 }
1965         }
1966
1967         /* Must commit after prep above in all cases */
1968         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1969                           objcount, ioo, remote_nb, npages, local_nb, NULL,
1970                           rc);
1971         if (rc == -ENOTCONN)
1972                 /* quota acquire process has been given up because
1973                  * either the client has been evicted or the client
1974                  * has timed out the request already */
1975                 no_reply = true;
1976
1977         /*
1978          * Disable sending mtime back to the client. If the client locked the
1979          * whole object, then it has already updated the mtime on its side,
1980          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1981          */
1982         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1983
1984         if (rc == 0) {
1985                 int nob = 0;
1986
1987                 /* set per-requested niobuf return codes */
1988                 for (i = j = 0; i < niocount; i++) {
1989                         int len = remote_nb[i].len;
1990
1991                         nob += len;
1992                         rcs[i] = 0;
1993                         do {
1994                                 LASSERT(j < npages);
1995                                 if (local_nb[j].rc < 0)
1996                                         rcs[i] = local_nb[j].rc;
1997                                 len -= local_nb[j].len;
1998                                 j++;
1999                         } while (len > 0);
2000                         LASSERT(len == 0);
2001                 }
2002                 LASSERT(j == npages);
2003                 ptlrpc_lprocfs_brw(req, nob);
2004
2005                 tgt_drop_id(exp, &repbody->oa);
2006         }
2007 out_lock:
2008         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2009         if (desc)
2010                 ptlrpc_free_bulk_nopin(desc);
2011 out:
2012         if (no_reply) {
2013                 req->rq_no_reply = 1;
2014                 /* reply out callback would free */
2015                 ptlrpc_req_drop_rs(req);
2016                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2017                               "client will retry: rc %d\n",
2018                               exp->exp_obd->obd_name,
2019                               obd_uuid2str(&exp->exp_client_uuid),
2020                               obd_export_nid2str(exp), rc);
2021         }
2022         memory_pressure_clr();
2023         RETURN(rc);
2024 }
2025 EXPORT_SYMBOL(tgt_brw_write);