lustre/target/tgt_handler.c

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