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_obd_qc_callback(struct tgt_session_info *tsi)
1007 {
1008         return err_serious(-EOPNOTSUPP);
1009 }
1010
1011 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1012 {
1013         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1014         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1015         struct obd_export       *exp = req->rq_export;
1016         struct ptlrpc_bulk_desc *desc;
1017         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1018         int                      tmpcount;
1019         int                      tmpsize;
1020         int                      i;
1021         int                      rc;
1022
1023         ENTRY;
1024
1025         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1026                                     MDS_BULK_PORTAL);
1027         if (desc == NULL)
1028                 RETURN(-ENOMEM);
1029
1030         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1031                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1032                  * which is rdpg->rp_count */
1033                 nob = rdpg->rp_count;
1034
1035         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1036              i++, tmpcount -= tmpsize) {
1037                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1038                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1039         }
1040
1041         LASSERT(desc->bd_nob == nob);
1042         rc = target_bulk_io(exp, desc, lwi);
1043         ptlrpc_free_bulk_pin(desc);
1044         RETURN(rc);
1045 }
1046 EXPORT_SYMBOL(tgt_sendpage);
1047
1048 /*
1049  * OBD_IDX_READ handler
1050  */
1051 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1052 {
1053         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1054         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1055         struct idx_info         *req_ii, *rep_ii;
1056         int                      rc, i;
1057
1058         ENTRY;
1059
1060         memset(rdpg, 0, sizeof(*rdpg));
1061         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1062
1063         /* extract idx_info buffer from request & reply */
1064         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1065         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1066                 RETURN(err_serious(-EPROTO));
1067
1068         rc = req_capsule_server_pack(tsi->tsi_pill);
1069         if (rc)
1070                 RETURN(err_serious(rc));
1071
1072         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1073         if (rep_ii == NULL)
1074                 RETURN(err_serious(-EFAULT));
1075         rep_ii->ii_magic = IDX_INFO_MAGIC;
1076
1077         /* extract hash to start with */
1078         rdpg->rp_hash = req_ii->ii_hash_start;
1079
1080         /* extract requested attributes */
1081         rdpg->rp_attrs = req_ii->ii_attrs;
1082
1083         /* check that fid packed in request is valid and supported */
1084         if (!fid_is_sane(&req_ii->ii_fid))
1085                 RETURN(-EINVAL);
1086         rep_ii->ii_fid = req_ii->ii_fid;
1087
1088         /* copy flags */
1089         rep_ii->ii_flags = req_ii->ii_flags;
1090
1091         /* compute number of pages to allocate, ii_count is the number of 4KB
1092          * containers */
1093         if (req_ii->ii_count <= 0)
1094                 GOTO(out, rc = -EFAULT);
1095         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1096                                exp_max_brw_size(tsi->tsi_exp));
1097         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1098
1099         /* allocate pages to store the containers */
1100         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1101         if (rdpg->rp_pages == NULL)
1102                 GOTO(out, rc = -ENOMEM);
1103         for (i = 0; i < rdpg->rp_npages; i++) {
1104                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1105                 if (rdpg->rp_pages[i] == NULL)
1106                         GOTO(out, rc = -ENOMEM);
1107         }
1108
1109         /* populate pages with key/record pairs */
1110         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1111         if (rc < 0)
1112                 GOTO(out, rc);
1113
1114         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1115                  "asked %d > %d\n", rc, rdpg->rp_count);
1116
1117         /* send pages to client */
1118         rc = tgt_sendpage(tsi, rdpg, rc);
1119         if (rc)
1120                 GOTO(out, rc);
1121         EXIT;
1122 out:
1123         if (rdpg->rp_pages) {
1124                 for (i = 0; i < rdpg->rp_npages; i++)
1125                         if (rdpg->rp_pages[i])
1126                                 __free_page(rdpg->rp_pages[i]);
1127                 OBD_FREE(rdpg->rp_pages,
1128                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1129         }
1130         return rc;
1131 }
1132
1133 struct tgt_handler tgt_obd_handlers[] = {
1134 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1135 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1136 TGT_OBD_HDL_VAR(0,      OBD_QC_CALLBACK,        tgt_obd_qc_callback),
1137 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1138 };
1139 EXPORT_SYMBOL(tgt_obd_handlers);
1140
1141 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1142              struct dt_object *obj, __u64 start, __u64 end)
1143 {
1144         int rc = 0;
1145
1146         ENTRY;
1147
1148         /* if no objid is specified, it means "sync whole filesystem" */
1149         if (obj == NULL) {
1150                 rc = dt_sync(env, tgt->lut_bottom);
1151         } else if (dt_version_get(env, obj) >
1152                    tgt->lut_obd->obd_last_committed) {
1153                 rc = dt_object_sync(env, obj, start, end);
1154         }
1155
1156         RETURN(rc);
1157 }
1158 EXPORT_SYMBOL(tgt_sync);
1159 /*
1160  * Unified target DLM handlers.
1161  */
1162
1163 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1164  * (if requested) */
1165 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1166                             void *data, int flag)
1167 {
1168         struct lu_env            env;
1169         struct lu_target        *tgt;
1170         struct dt_object        *obj;
1171         struct lu_fid            fid;
1172         int                      rc = 0;
1173
1174         ENTRY;
1175
1176         tgt = class_exp2tgt(lock->l_export);
1177
1178         if (unlikely(tgt == NULL)) {
1179                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1180                        class_exp2obd(lock->l_export)->obd_name);
1181                 RETURN(-EINVAL);
1182         }
1183
1184         if (flag == LDLM_CB_CANCELING &&
1185             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1186             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1187              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1188               lock->l_flags & LDLM_FL_CBPENDING))) {
1189                 __u64 start = 0;
1190                 __u64 end = OBD_OBJECT_EOF;
1191
1192                 rc = lu_env_init(&env, LCT_DT_THREAD);
1193                 if (unlikely(rc != 0))
1194                         RETURN(rc);
1195
1196                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1197                                    tgt->lut_lsd.lsd_osd_index);
1198                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1199                 if (IS_ERR(obj))
1200                         GOTO(err_env, rc = PTR_ERR(obj));
1201
1202                 if (!dt_object_exists(obj))
1203                         GOTO(err_put, rc = -ENOENT);
1204
1205                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1206                         start = lock->l_policy_data.l_extent.start;
1207                         end = lock->l_policy_data.l_extent.end;
1208                 }
1209
1210                 rc = tgt_sync(&env, tgt, obj, start, end);
1211                 if (rc < 0) {
1212                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1213                                "cancel: rc = %d\n",
1214                                tgt_name(tgt), PFID(&fid),
1215                                lock->l_policy_data.l_extent.start,
1216                                lock->l_policy_data.l_extent.end, rc);
1217                 }
1218 err_put:
1219                 lu_object_put(&env, &obj->do_lu);
1220 err_env:
1221                 lu_env_fini(&env);
1222         }
1223
1224         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1225         RETURN(rc);
1226 }
1227
1228 static struct ldlm_callback_suite tgt_dlm_cbs = {
1229         .lcs_completion = ldlm_server_completion_ast,
1230         .lcs_blocking   = tgt_blocking_ast,
1231         .lcs_glimpse    = ldlm_server_glimpse_ast
1232 };
1233
1234 int tgt_enqueue(struct tgt_session_info *tsi)
1235 {
1236         struct ptlrpc_request *req = tgt_ses_req(tsi);
1237         int rc;
1238
1239         ENTRY;
1240         /*
1241          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1242          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1243          */
1244         LASSERT(tsi->tsi_dlm_req != NULL);
1245         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1246                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1247         if (rc)
1248                 RETURN(err_serious(rc));
1249
1250         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1251         case LUT_FAIL_MDT:
1252                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1253                 break;
1254         case LUT_FAIL_OST:
1255                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1256                 break;
1257         case LUT_FAIL_MGT:
1258                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1259                 break;
1260         default:
1261                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1262                 break;
1263         }
1264         RETURN(req->rq_status);
1265 }
1266 EXPORT_SYMBOL(tgt_enqueue);
1267
1268 int tgt_convert(struct tgt_session_info *tsi)
1269 {
1270         struct ptlrpc_request *req = tgt_ses_req(tsi);
1271         int rc;
1272
1273         ENTRY;
1274         LASSERT(tsi->tsi_dlm_req);
1275         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1276         if (rc)
1277                 RETURN(err_serious(rc));
1278
1279         RETURN(req->rq_status);
1280 }
1281
1282 int tgt_bl_callback(struct tgt_session_info *tsi)
1283 {
1284         return err_serious(-EOPNOTSUPP);
1285 }
1286
1287 int tgt_cp_callback(struct tgt_session_info *tsi)
1288 {
1289         return err_serious(-EOPNOTSUPP);
1290 }
1291
1292 /* generic LDLM target handler */
1293 struct tgt_handler tgt_dlm_handlers[] = {
1294 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1295 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1296 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1297 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1298 };
1299 EXPORT_SYMBOL(tgt_dlm_handlers);
1300
1301 /*
1302  * Unified target LLOG handlers.
1303  */
1304 int tgt_llog_open(struct tgt_session_info *tsi)
1305 {
1306         int rc;
1307
1308         ENTRY;
1309
1310         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1311
1312         RETURN(rc);
1313 }
1314 EXPORT_SYMBOL(tgt_llog_open);
1315
1316 int tgt_llog_close(struct tgt_session_info *tsi)
1317 {
1318         int rc;
1319
1320         ENTRY;
1321
1322         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1323
1324         RETURN(rc);
1325 }
1326 EXPORT_SYMBOL(tgt_llog_close);
1327
1328
1329 int tgt_llog_destroy(struct tgt_session_info *tsi)
1330 {
1331         int rc;
1332
1333         ENTRY;
1334
1335         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1336
1337         RETURN(rc);
1338 }
1339
1340 int tgt_llog_read_header(struct tgt_session_info *tsi)
1341 {
1342         int rc;
1343
1344         ENTRY;
1345
1346         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1347
1348         RETURN(rc);
1349 }
1350 EXPORT_SYMBOL(tgt_llog_read_header);
1351
1352 int tgt_llog_next_block(struct tgt_session_info *tsi)
1353 {
1354         int rc;
1355
1356         ENTRY;
1357
1358         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1359
1360         RETURN(rc);
1361 }
1362 EXPORT_SYMBOL(tgt_llog_next_block);
1363
1364 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1365 {
1366         int rc;
1367
1368         ENTRY;
1369
1370         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1371
1372         RETURN(rc);
1373 }
1374 EXPORT_SYMBOL(tgt_llog_prev_block);
1375
1376 /* generic llog target handler */
1377 struct tgt_handler tgt_llog_handlers[] = {
1378 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1379 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1380 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1381 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1382 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1383 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1384 };
1385 EXPORT_SYMBOL(tgt_llog_handlers);
1386
1387 /*
1388  * sec context handlers
1389  */
1390 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1391 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1392 {
1393         return 0;
1394 }
1395
1396 struct tgt_handler tgt_sec_ctx_handlers[] = {
1397 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1398 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1399 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1400 };
1401 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1402
1403 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1404                            struct dt_device *key,
1405                            struct lfsck_request *lr,
1406                            struct thandle *th) = NULL;
1407
1408 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1409                                                 struct dt_device *,
1410                                                 struct lfsck_request *,
1411                                                 struct thandle *))
1412 {
1413         tgt_lfsck_in_notify = notify;
1414 }
1415 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1416
1417 static int (*tgt_lfsck_query)(const struct lu_env *env,
1418                               struct dt_device *key,
1419                               struct lfsck_request *lr) = NULL;
1420
1421 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1422                                            struct dt_device *,
1423                                            struct lfsck_request *))
1424 {
1425         tgt_lfsck_query = query;
1426 }
1427 EXPORT_SYMBOL(tgt_register_lfsck_query);
1428
1429 /* LFSCK request handlers */
1430 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1431 {
1432         const struct lu_env     *env = tsi->tsi_env;
1433         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1434         struct lfsck_request    *lr;
1435         int                      rc;
1436         ENTRY;
1437
1438         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1439         if (lr == NULL)
1440                 RETURN(-EPROTO);
1441
1442         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1443
1444         RETURN(rc);
1445 }
1446
1447 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1448 {
1449         struct lfsck_request    *request;
1450         struct lfsck_reply      *reply;
1451         int                      rc;
1452         ENTRY;
1453
1454         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1455         if (request == NULL)
1456                 RETURN(-EPROTO);
1457
1458         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1459         if (reply == NULL)
1460                 RETURN(-ENOMEM);
1461
1462         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1463         reply->lr_status = rc;
1464
1465         RETURN(rc < 0 ? rc : 0);
1466 }
1467
1468 struct tgt_handler tgt_lfsck_handlers[] = {
1469 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1470 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1471 };
1472 EXPORT_SYMBOL(tgt_lfsck_handlers);
1473
1474 /*
1475  * initialize per-thread page pool (bug 5137).
1476  */
1477 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1478 {
1479         struct tgt_thread_big_cache *tbc;
1480
1481         ENTRY;
1482
1483         LASSERT(thread != NULL);
1484         LASSERT(thread->t_data == NULL);
1485
1486         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1487         if (tbc == NULL)
1488                 RETURN(-ENOMEM);
1489         thread->t_data = tbc;
1490         RETURN(0);
1491 }
1492 EXPORT_SYMBOL(tgt_io_thread_init);
1493
1494 /*
1495  * free per-thread pool created by tgt_thread_init().
1496  */
1497 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1498 {
1499         struct tgt_thread_big_cache *tbc;
1500
1501         ENTRY;
1502
1503         LASSERT(thread != NULL);
1504
1505         /*
1506          * be prepared to handle partially-initialized pools (because this is
1507          * called from ost_io_thread_init() for cleanup.
1508          */
1509         tbc = thread->t_data;
1510         if (tbc != NULL) {
1511                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1512                 thread->t_data = NULL;
1513         }
1514         EXIT;
1515 }
1516 EXPORT_SYMBOL(tgt_io_thread_done);
1517 /**
1518  * Helper function for getting server side [start, start+count] DLM lock
1519  * if asked by client.
1520  */
1521 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1522                     __u64 start, __u64 end, struct lustre_handle *lh,
1523                     int mode, __u64 *flags)
1524 {
1525         ldlm_policy_data_t       policy;
1526         int                      rc;
1527
1528         ENTRY;
1529
1530         LASSERT(lh != NULL);
1531         LASSERT(ns != NULL);
1532         LASSERT(!lustre_handle_is_used(lh));
1533
1534         policy.l_extent.gid = 0;
1535         policy.l_extent.start = start & PAGE_MASK;
1536
1537         /*
1538          * If ->o_blocks is EOF it means "lock till the end of the file".
1539          * Otherwise, it's size of an extent or hole being punched (in bytes).
1540          */
1541         if (end == OBD_OBJECT_EOF || end < start)
1542                 policy.l_extent.end = OBD_OBJECT_EOF;
1543         else
1544                 policy.l_extent.end = end | ~PAGE_MASK;
1545
1546         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1547                                     flags, ldlm_blocking_ast,
1548                                     ldlm_completion_ast, ldlm_glimpse_ast,
1549                                     NULL, 0, LVB_T_NONE, NULL, lh);
1550         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1551 }
1552 EXPORT_SYMBOL(tgt_extent_lock);
1553
1554 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1555 {
1556         LASSERT(lustre_handle_is_used(lh));
1557         ldlm_lock_decref(lh, mode);
1558 }
1559 EXPORT_SYMBOL(tgt_extent_unlock);
1560
1561 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1562                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1563                  struct lustre_handle *lh, int mode)
1564 {
1565         __u64                    flags = 0;
1566         int                      nrbufs = obj->ioo_bufcnt;
1567         int                      i;
1568
1569         ENTRY;
1570
1571         LASSERT(mode == LCK_PR || mode == LCK_PW);
1572         LASSERT(!lustre_handle_is_used(lh));
1573
1574         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1575                 RETURN(0);
1576
1577         for (i = 1; i < nrbufs; i++)
1578                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1579                         RETURN(-EFAULT);
1580
1581         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1582                                nb[nrbufs - 1].rnb_offset +
1583                                nb[nrbufs - 1].rnb_len - 1,
1584                                lh, mode, &flags));
1585 }
1586
1587 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1588                     struct lustre_handle *lh, int mode)
1589 {
1590         ENTRY;
1591
1592         LASSERT(mode == LCK_PR || mode == LCK_PW);
1593         LASSERT((obj->ioo_bufcnt > 0 &&
1594                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1595                 lustre_handle_is_used(lh));
1596
1597         if (lustre_handle_is_used(lh))
1598                 tgt_extent_unlock(lh, mode);
1599         EXIT;
1600 }
1601
1602 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1603                                struct ptlrpc_bulk_desc *desc, int opc,
1604                                cksum_type_t cksum_type)
1605 {
1606         struct cfs_crypto_hash_desc     *hdesc;
1607         unsigned int                    bufsize;
1608         int                             i, err;
1609         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1610         __u32                           cksum;
1611
1612         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1613         if (IS_ERR(hdesc)) {
1614                 CERROR("%s: unable to initialize checksum hash %s\n",
1615                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1616                 return PTR_ERR(hdesc);
1617         }
1618
1619         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1620         for (i = 0; i < desc->bd_iov_count; i++) {
1621                 /* corrupt the data before we compute the checksum, to
1622                  * simulate a client->OST data error */
1623                 if (i == 0 && opc == OST_WRITE &&
1624                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1625                         int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1626                         int len = desc->bd_iov[i].kiov_len;
1627                         struct page *np = tgt_page_to_corrupt;
1628                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1629
1630                         if (np) {
1631                                 char *ptr2 = kmap(np) + off;
1632
1633                                 memcpy(ptr2, ptr, len);
1634                                 memcpy(ptr2, "bad3", min(4, len));
1635                                 kunmap(np);
1636                                 desc->bd_iov[i].kiov_page = np;
1637                         } else {
1638                                 CERROR("%s: can't alloc page for corruption\n",
1639                                        tgt_name(tgt));
1640                         }
1641                 }
1642                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1643                                   desc->bd_iov[i].kiov_offset & ~PAGE_MASK,
1644                                   desc->bd_iov[i].kiov_len);
1645
1646                  /* corrupt the data after we compute the checksum, to
1647                  * simulate an OST->client data error */
1648                 if (i == 0 && opc == OST_READ &&
1649                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1650                         int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1651                         int len = desc->bd_iov[i].kiov_len;
1652                         struct page *np = tgt_page_to_corrupt;
1653                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1654
1655                         if (np) {
1656                                 char *ptr2 = kmap(np) + off;
1657
1658                                 memcpy(ptr2, ptr, len);
1659                                 memcpy(ptr2, "bad4", min(4, len));
1660                                 kunmap(np);
1661                                 desc->bd_iov[i].kiov_page = np;
1662                         } else {
1663                                 CERROR("%s: can't alloc page for corruption\n",
1664                                        tgt_name(tgt));
1665                         }
1666                 }
1667         }
1668
1669         bufsize = sizeof(cksum);
1670         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1671
1672         return cksum;
1673 }
1674
1675 int tgt_brw_read(struct tgt_session_info *tsi)
1676 {
1677         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1678         struct ptlrpc_bulk_desc *desc = NULL;
1679         struct obd_export       *exp = tsi->tsi_exp;
1680         struct niobuf_remote    *remote_nb;
1681         struct niobuf_local     *local_nb;
1682         struct obd_ioobj        *ioo;
1683         struct ost_body         *body, *repbody;
1684         struct l_wait_info       lwi;
1685         struct lustre_handle     lockh = { 0 };
1686         int                      npages, nob = 0, rc, i, no_reply = 0;
1687         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1688
1689         ENTRY;
1690
1691         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1692                 CERROR("%s: deny read request from %s to portal %u\n",
1693                        tgt_name(tsi->tsi_tgt),
1694                        obd_export_nid2str(req->rq_export),
1695                        ptlrpc_req2svc(req)->srv_req_portal);
1696                 RETURN(-EPROTO);
1697         }
1698
1699         req->rq_bulk_read = 1;
1700
1701         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1702                 RETURN(-EIO);
1703
1704         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1705                          cfs_fail_val : (obd_timeout + 1) / 4);
1706
1707         /* Check if there is eviction in progress, and if so, wait for it to
1708          * finish */
1709         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1710                 /* We do not care how long it takes */
1711                 lwi = LWI_INTR(NULL, NULL);
1712                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1713                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1714                          &lwi);
1715         }
1716
1717         /* There must be big cache in current thread to process this request
1718          * if it is NULL then something went wrong and it wasn't allocated,
1719          * report -ENOMEM in that case */
1720         if (tbc == NULL)
1721                 RETURN(-ENOMEM);
1722
1723         body = tsi->tsi_ost_body;
1724         LASSERT(body != NULL);
1725
1726         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1727         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1728
1729         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1730         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1731
1732         local_nb = tbc->local;
1733
1734         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1735                           remote_nb, &lockh, LCK_PR);
1736         if (rc != 0)
1737                 RETURN(rc);
1738
1739         /*
1740          * If getting the lock took more time than
1741          * client was willing to wait, drop it. b=11330
1742          */
1743         if (cfs_time_current_sec() > req->rq_deadline ||
1744             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1745                 no_reply = 1;
1746                 CERROR("Dropping timed-out read from %s because locking"
1747                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1748                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1749                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1750                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1751                 GOTO(out_lock, rc = -ETIMEDOUT);
1752         }
1753
1754         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1755         repbody->oa = body->oa;
1756
1757         npages = PTLRPC_MAX_BRW_PAGES;
1758         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1759                         ioo, remote_nb, &npages, local_nb, NULL);
1760         if (rc != 0)
1761                 GOTO(out_lock, rc);
1762
1763         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1764                                     BULK_PUT_SOURCE, OST_BULK_PORTAL);
1765         if (desc == NULL)
1766                 GOTO(out_commitrw, rc = -ENOMEM);
1767
1768         nob = 0;
1769         for (i = 0; i < npages; i++) {
1770                 int page_rc = local_nb[i].lnb_rc;
1771
1772                 if (page_rc < 0) {
1773                         rc = page_rc;
1774                         break;
1775                 }
1776
1777                 nob += page_rc;
1778                 if (page_rc != 0) { /* some data! */
1779                         LASSERT(local_nb[i].lnb_page != NULL);
1780                         ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
1781                                                     local_nb[i].lnb_page_offset,
1782                                                     page_rc);
1783                 }
1784
1785                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1786                         /* All subsequent pages should be 0 */
1787                         while (++i < npages)
1788                                 LASSERT(local_nb[i].lnb_rc == 0);
1789                         break;
1790                 }
1791         }
1792
1793         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1794                 cksum_type_t cksum_type =
1795                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1796                                           body->oa.o_flags : 0);
1797                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1798                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1799                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1800                                                         OST_READ, cksum_type);
1801                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1802                        repbody->oa.o_cksum);
1803         } else {
1804                 repbody->oa.o_valid = 0;
1805         }
1806         /* We're finishing using body->oa as an input variable */
1807
1808         /* Check if client was evicted while we were doing i/o before touching
1809          * network */
1810         if (likely(rc == 0 &&
1811                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1812                 rc = target_bulk_io(exp, desc, &lwi);
1813                 no_reply = rc != 0;
1814         }
1815
1816 out_commitrw:
1817         /* Must commit after prep above in all cases */
1818         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1819                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1820                           NULL, rc);
1821         if (rc == 0)
1822                 tgt_drop_id(exp, &repbody->oa);
1823 out_lock:
1824         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1825
1826         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1827                 ptlrpc_free_bulk_nopin(desc);
1828
1829         LASSERT(rc <= 0);
1830         if (rc == 0) {
1831                 rc = nob;
1832                 ptlrpc_lprocfs_brw(req, nob);
1833         } else if (no_reply) {
1834                 req->rq_no_reply = 1;
1835                 /* reply out callback would free */
1836                 ptlrpc_req_drop_rs(req);
1837                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1838                               "client will retry: rc %d\n",
1839                               exp->exp_obd->obd_name,
1840                               obd_uuid2str(&exp->exp_client_uuid),
1841                               obd_export_nid2str(exp), rc);
1842         }
1843         /* send a bulk after reply to simulate a network delay or reordering
1844          * by a router */
1845         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1846                 wait_queue_head_t        waitq;
1847                 struct l_wait_info       lwi1;
1848
1849                 CDEBUG(D_INFO, "reorder BULK\n");
1850                 init_waitqueue_head(&waitq);
1851
1852                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1853                 l_wait_event(waitq, 0, &lwi1);
1854                 target_bulk_io(exp, desc, &lwi);
1855                 ptlrpc_free_bulk_nopin(desc);
1856         }
1857
1858         RETURN(rc);
1859 }
1860 EXPORT_SYMBOL(tgt_brw_read);
1861
1862 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1863                               struct ptlrpc_bulk_desc *desc,
1864                               struct niobuf_local *local_nb, int npages,
1865                               u32 client_cksum, u32 server_cksum,
1866                               bool mmap)
1867 {
1868         struct obd_export *exp = req->rq_export;
1869         struct ost_body *body;
1870         char *router = "";
1871         char *via = "";
1872
1873         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1874         LASSERT(body != NULL);
1875
1876         if (req->rq_peer.nid != desc->bd_sender) {
1877                 via = " via ";
1878                 router = libcfs_nid2str(desc->bd_sender);
1879         }
1880
1881         if (mmap) {
1882                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1883                              client_cksum, server_cksum);
1884                 return;
1885         }
1886
1887         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1888                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1889                            "]: client csum %x, server csum %x\n",
1890                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1891                            via, router,
1892                            body->oa.o_valid & OBD_MD_FLFID ?
1893                            body->oa.o_parent_seq : (__u64)0,
1894                            body->oa.o_valid & OBD_MD_FLFID ?
1895                            body->oa.o_parent_oid : 0,
1896                            body->oa.o_valid & OBD_MD_FLFID ?
1897                            body->oa.o_parent_ver : 0,
1898                            POSTID(&body->oa.o_oi),
1899                            local_nb[0].lnb_file_offset,
1900                            local_nb[npages-1].lnb_file_offset +
1901                            local_nb[npages - 1].lnb_len - 1,
1902                            client_cksum, server_cksum);
1903 }
1904
1905 int tgt_brw_write(struct tgt_session_info *tsi)
1906 {
1907         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1908         struct ptlrpc_bulk_desc *desc = NULL;
1909         struct obd_export       *exp = req->rq_export;
1910         struct niobuf_remote    *remote_nb;
1911         struct niobuf_local     *local_nb;
1912         struct obd_ioobj        *ioo;
1913         struct ost_body         *body, *repbody;
1914         struct l_wait_info       lwi;
1915         struct lustre_handle     lockh = {0};
1916         __u32                   *rcs;
1917         int                      objcount, niocount, npages;
1918         int                      rc, i, j;
1919         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1920         bool                     no_reply = false, mmap;
1921         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1922
1923         ENTRY;
1924
1925         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1926                 CERROR("%s: deny write request from %s to portal %u\n",
1927                        tgt_name(tsi->tsi_tgt),
1928                        obd_export_nid2str(req->rq_export),
1929                        ptlrpc_req2svc(req)->srv_req_portal);
1930                 RETURN(err_serious(-EPROTO));
1931         }
1932
1933         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1934                 RETURN(err_serious(-ENOSPC));
1935         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1936                 RETURN(err_serious(-EROFS));
1937
1938         req->rq_bulk_write = 1;
1939
1940         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1941                 RETURN(err_serious(-EIO));
1942         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1943                 RETURN(err_serious(-EFAULT));
1944
1945         /* pause before transaction has been started */
1946         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1947                          cfs_fail_val : (obd_timeout + 1) / 4);
1948
1949         /* There must be big cache in current thread to process this request
1950          * if it is NULL then something went wrong and it wasn't allocated,
1951          * report -ENOMEM in that case */
1952         if (tbc == NULL)
1953                 RETURN(-ENOMEM);
1954
1955         body = tsi->tsi_ost_body;
1956         LASSERT(body != NULL);
1957
1958         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1959         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1960
1961         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1962                                         RCL_CLIENT) / sizeof(*ioo);
1963
1964         for (niocount = i = 0; i < objcount; i++)
1965                 niocount += ioo[i].ioo_bufcnt;
1966
1967         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1968         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1969         if (niocount != req_capsule_get_size(&req->rq_pill,
1970                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1971                         sizeof(*remote_nb))
1972                 RETURN(err_serious(-EPROTO));
1973
1974         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1975             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1976                 memory_pressure_set();
1977
1978         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1979                              niocount * sizeof(*rcs));
1980         rc = req_capsule_server_pack(&req->rq_pill);
1981         if (rc != 0)
1982                 GOTO(out, rc = err_serious(rc));
1983
1984         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1985         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1986
1987         local_nb = tbc->local;
1988
1989         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1990                           remote_nb, &lockh, LCK_PW);
1991         if (rc != 0)
1992                 GOTO(out, rc);
1993
1994         /*
1995          * If getting the lock took more time than
1996          * client was willing to wait, drop it. b=11330
1997          */
1998         if (cfs_time_current_sec() > req->rq_deadline ||
1999             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2000                 no_reply = true;
2001                 CERROR("%s: Dropping timed-out write from %s because locking "
2002                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
2003                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2004                        POSTID(&ioo->ioo_oid),
2005                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2006                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2007                 GOTO(out_lock, rc = -ETIMEDOUT);
2008         }
2009
2010         /* Because we already sync grant info with client when reconnect,
2011          * grant info will be cleared for resent req, then fed_grant and
2012          * total_grant will not be modified in following preprw_write */
2013         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2014                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2015                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2016         }
2017
2018         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2019         if (repbody == NULL)
2020                 GOTO(out_lock, rc = -ENOMEM);
2021         repbody->oa = body->oa;
2022
2023         npages = PTLRPC_MAX_BRW_PAGES;
2024         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2025                         objcount, ioo, remote_nb, &npages, local_nb, NULL);
2026         if (rc < 0)
2027                 GOTO(out_lock, rc);
2028
2029         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2030                                     BULK_GET_SINK, OST_BULK_PORTAL);
2031         if (desc == NULL)
2032                 GOTO(skip_transfer, rc = -ENOMEM);
2033
2034         /* NB Having prepped, we must commit... */
2035         for (i = 0; i < npages; i++)
2036                 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
2037                                             local_nb[i].lnb_page_offset,
2038                                             local_nb[i].lnb_len);
2039
2040         rc = sptlrpc_svc_prep_bulk(req, desc);
2041         if (rc != 0)
2042                 GOTO(skip_transfer, rc);
2043
2044         rc = target_bulk_io(exp, desc, &lwi);
2045         no_reply = rc != 0;
2046
2047 skip_transfer:
2048         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2049                 static int cksum_counter;
2050
2051                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2052                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2053
2054                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2055                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2056                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2057                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2058                                                         OST_WRITE, cksum_type);
2059                 cksum_counter++;
2060
2061                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2062                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2063                                 body->oa.o_flags & OBD_FL_MMAP);
2064
2065                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2066                                           body->oa.o_cksum,
2067                                           repbody->oa.o_cksum, mmap);
2068                         cksum_counter = 0;
2069                 } else if ((cksum_counter & (-cksum_counter)) ==
2070                            cksum_counter) {
2071                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2072                                cksum_counter, libcfs_id2str(req->rq_peer),
2073                                repbody->oa.o_cksum);
2074                 }
2075         }
2076
2077         /* Must commit after prep above in all cases */
2078         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2079                           objcount, ioo, remote_nb, npages, local_nb, NULL,
2080                           rc);
2081         if (rc == -ENOTCONN)
2082                 /* quota acquire process has been given up because
2083                  * either the client has been evicted or the client
2084                  * has timed out the request already */
2085                 no_reply = true;
2086
2087         /*
2088          * Disable sending mtime back to the client. If the client locked the
2089          * whole object, then it has already updated the mtime on its side,
2090          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2091          */
2092         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2093
2094         if (rc == 0) {
2095                 int nob = 0;
2096
2097                 /* set per-requested niobuf return codes */
2098                 for (i = j = 0; i < niocount; i++) {
2099                         int len = remote_nb[i].rnb_len;
2100
2101                         nob += len;
2102                         rcs[i] = 0;
2103                         do {
2104                                 LASSERT(j < npages);
2105                                 if (local_nb[j].lnb_rc < 0)
2106                                         rcs[i] = local_nb[j].lnb_rc;
2107                                 len -= local_nb[j].lnb_len;
2108                                 j++;
2109                         } while (len > 0);
2110                         LASSERT(len == 0);
2111                 }
2112                 LASSERT(j == npages);
2113                 ptlrpc_lprocfs_brw(req, nob);
2114
2115                 tgt_drop_id(exp, &repbody->oa);
2116         }
2117 out_lock:
2118         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2119         if (desc)
2120                 ptlrpc_free_bulk_nopin(desc);
2121 out:
2122         if (no_reply) {
2123                 req->rq_no_reply = 1;
2124                 /* reply out callback would free */
2125                 ptlrpc_req_drop_rs(req);
2126                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2127                               "client will retry: rc %d\n",
2128                               exp->exp_obd->obd_name,
2129                               obd_uuid2str(&exp->exp_client_uuid),
2130                               obd_export_nid2str(exp), rc);
2131         }
2132         memory_pressure_clr();
2133         RETURN(rc);
2134 }
2135 EXPORT_SYMBOL(tgt_brw_write);