lustre/target/tgt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2011, 2012, Intel Corporation.
  25  */
  26 /*
  27  * lustre/target/tgt_handler.c
  28  *
  29  * Lustre Unified Target request handler code
  30  *
  31  * Author: Brian Behlendorf <behlendorf1@llnl.gov>
  32  * Author: Mikhail Pershin <mike.pershin@intel.com>
  33  */
  34
  35 #define DEBUG_SUBSYSTEM S_CLASS
  36
  37 #include <obd.h>
  38 #include <obd_class.h>
  39 #include <obd_cksum.h>
  40 #include <md_object.h>
  41 #include <lustre_lfsck.h>
  42
  43 #include "tgt_internal.h"
  44
  45 char *tgt_name(struct lu_target *tgt)
  46 {
  47         LASSERT(tgt->lut_obd != NULL);
  48         return tgt->lut_obd->obd_name;
  49 }
  50 EXPORT_SYMBOL(tgt_name);
  51
  52 /*
  53  * Generic code handling requests that have struct mdt_body passed in:
  54  *
  55  *  - extract mdt_body from request and save it in @tsi, if present;
  56  *
  57  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  58  *  @tsi;
  59  *
  60  *  - if HABEO_CORPUS flag is set for this request type check whether object
  61  *  actually exists on storage (lu_object_exists()).
  62  *
  63  */
  64 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  65 {
  66         const struct mdt_body   *body;
  67         struct lu_object        *obj;
  68         struct req_capsule      *pill = tsi->tsi_pill;
  69         int                      rc;
  70
  71         ENTRY;
  72
  73         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  74         if (body == NULL)
  75                 RETURN(-EFAULT);
  76
  77         tsi->tsi_mdt_body = body;
  78
  79         if (!(body->valid & OBD_MD_FLID))
  80                 RETURN(0);
  81
  82         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  83          * in any case in pre-2.5 clients. Fix that here if needed */
  84         if (unlikely(fid_is_zero(&body->fid1)))
  85                 RETURN(0);
  86
  87         if (!fid_is_sane(&body->fid1)) {
  88                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  89                        PFID(&body->fid1));
  90                 RETURN(-EINVAL);
  91         }
  92
  93         obj = lu_object_find(tsi->tsi_env,
  94                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
  95                              &body->fid1, NULL);
  96         if (!IS_ERR(obj)) {
  97                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
  98                         lu_object_put(tsi->tsi_env, obj);
  99                         /* for capability renew ENOENT will be handled in
 100                          * mdt_renew_capa */
 101                         if (body->valid & OBD_MD_FLOSSCAPA)
 102                                 rc = 0;
 103                         else
 104                                 rc = -ENOENT;
 105                 } else {
 106                         tsi->tsi_corpus = obj;
 107                         rc = 0;
 108                 }
 109         } else {
 110                 rc = PTR_ERR(obj);
 111         }
 112
 113         tsi->tsi_fid = body->fid1;
 114
 115         RETURN(rc);
 116 }
 117
 118 /**
 119  * Validate oa from client.
 120  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 121  * req are valid.
 122  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 123  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 124  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 125  *       whether this is oi_fid or real ostid. So it will check
 126  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 127  *    c. Old FID-disable osc will send IDIF.
 128  *    d. new FID-enable osc/osp will send normal FID.
 129  *
 130  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 131  * be used for LAST_ID file, and only being accessed inside OST now.
 132  */
 133 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 134 {
 135         struct ost_id   *oi     = &oa->o_oi;
 136         obd_seq          seq    = ostid_seq(oi);
 137         obd_id           id     = ostid_id(oi);
 138         int              rc;
 139         ENTRY;
 140
 141         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
 142                      fid_seq_is_echo(seq))) {
 143                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 144                  * during create, and we will reset this to 1, since this
 145                  * oi_id is basically useless in the following create process,
 146                  * but oi_id == 0 will make it difficult to tell whether it is
 147                  * real FID or ost_id. */
 148                 oi->oi_fid.f_seq = FID_SEQ_ECHO;
 149                 oi->oi_fid.f_oid = id ?: 1;
 150                 oi->oi_fid.f_ver = 0;
 151         } else {
 152                 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
 153
 154                 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
 155                         GOTO(out, rc = -EPROTO);
 156
 157                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 158                  * all of the requests are required to setup FLGROUP */
 159                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 160                         ostid_set_seq_mdt0(oi);
 161                         oa->o_valid |= OBD_MD_FLGROUP;
 162                         seq = ostid_seq(oi);
 163                 }
 164
 165                 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
 166                                fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
 167                         GOTO(out, rc = -EPROTO);
 168
 169                 rc = ostid_to_fid(&tti->tti_fid1, oi,
 170                                   tsi->tsi_tgt->lut_lsd.lsd_osd_index);
 171                 if (unlikely(rc != 0))
 172                         GOTO(out, rc);
 173
 174                 oi->oi_fid = tti->tti_fid1;
 175         }
 176
 177         RETURN(0);
 178
 179 out:
 180         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 181                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
 182                seq, id, rc);
 183         return rc;
 184 }
 185 EXPORT_SYMBOL(tgt_validate_obdo);
 186
 187 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
 188 {
 189         unsigned                 max_brw;
 190         struct niobuf_remote    *rnb;
 191         struct obd_ioobj        *ioo;
 192         int                      obj_count;
 193
 194         ENTRY;
 195
 196         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
 197         if (ioo == NULL)
 198                 RETURN(-EPROTO);
 199
 200         rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
 201         if (rnb == NULL)
 202                 RETURN(-EPROTO);
 203
 204         max_brw = ioobj_max_brw_get(ioo);
 205         if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 206                 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 207                        ": rc = %d\n", tgt_name(tsi->tsi_tgt),
 208                        obd_export_nid2str(tsi->tsi_exp), max_brw,
 209                        POSTID(oi), -EPROTO);
 210                 RETURN(-EPROTO);
 211         }
 212         ioo->ioo_oid = *oi;
 213
 214         obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
 215                                         RCL_CLIENT) / sizeof(*ioo);
 216         if (obj_count == 0) {
 217                 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
 218                 RETURN(-EPROTO);
 219         } else if (obj_count > 1) {
 220                 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
 221                        obj_count);
 222                 RETURN(-EPROTO);
 223         }
 224
 225         if (ioo->ioo_bufcnt == 0) {
 226                 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
 227                 RETURN(-EPROTO);
 228         }
 229
 230         if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
 231                 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
 232                           "bulk has too many pages (%d)",
 233                           ioo->ioo_bufcnt);
 234                 RETURN(-EPROTO);
 235         }
 236
 237         RETURN(0);
 238 }
 239
 240 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 241 {
 242         struct ost_body         *body;
 243         struct req_capsule      *pill = tsi->tsi_pill;
 244         struct lustre_capa      *capa;
 245         int                      rc;
 246
 247         ENTRY;
 248
 249         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 250         if (body == NULL)
 251                 RETURN(-EFAULT);
 252
 253         rc = tgt_validate_obdo(tsi, &body->oa);
 254         if (rc)
 255                 RETURN(rc);
 256
 257         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 258                 capa = req_capsule_client_get(pill, &RMF_CAPA1);
 259                 if (capa == NULL) {
 260                         CERROR("%s: OSSCAPA flag is set without capability\n",
 261                                tgt_name(tsi->tsi_tgt));
 262                         RETURN(-EFAULT);
 263                 }
 264         }
 265
 266         tsi->tsi_ost_body = body;
 267         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 268
 269         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 270                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 271                 if (rc < 0)
 272                         RETURN(rc);
 273         }
 274
 275         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 276                 if (flags & HABEO_CORPUS) {
 277                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 278                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 279                                tgt_name(tsi->tsi_tgt));
 280                         RETURN(-EPROTO);
 281                 } else {
 282                         RETURN(0);
 283                 }
 284         }
 285
 286         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 287
 288         /*
 289          * OST doesn't get object in advance for further use to prevent
 290          * situations with nested object_find which is potential deadlock.
 291          */
 292         tsi->tsi_corpus = NULL;
 293         RETURN(rc);
 294 }
 295
 296 /*
 297  * Do necessary preprocessing according to handler ->th_flags.
 298  */
 299 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 300                                   struct tgt_handler *h,
 301                                   struct ptlrpc_request *req)
 302 {
 303         struct req_capsule      *pill = tsi->tsi_pill;
 304         __u32                    flags = h->th_flags;
 305         int                      rc = 0;
 306
 307         ENTRY;
 308
 309         if (tsi->tsi_preprocessed)
 310                 RETURN(0);
 311
 312         LASSERT(h->th_act != NULL);
 313         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 314         LASSERT(current->journal_info == NULL);
 315
 316         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 317                      h->th_fmt != NULL));
 318         if (h->th_fmt != NULL) {
 319                 req_capsule_set(pill, h->th_fmt);
 320                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 321                         rc = tgt_mdt_body_unpack(tsi, flags);
 322                         if (rc < 0)
 323                                 RETURN(rc);
 324                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 325                                                  RCL_CLIENT)) {
 326                         rc = tgt_ost_body_unpack(tsi, flags);
 327                         if (rc < 0)
 328                                 RETURN(rc);
 329                 }
 330         }
 331
 332         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 333                 RETURN(-EROFS);
 334
 335         if (flags & HABEO_CLAVIS) {
 336                 struct ldlm_request *dlm_req;
 337
 338                 LASSERT(h->th_fmt != NULL);
 339
 340                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 341                 if (dlm_req != NULL) {
 342                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 343                                      LDLM_IBITS &&
 344                                      dlm_req->lock_desc.l_policy_data.\
 345                                      l_inodebits.bits == 0)) {
 346                                 /*
 347                                  * Lock without inodebits makes no sense and
 348                                  * will oops later in ldlm. If client miss to
 349                                  * set such bits, do not trigger ASSERTION.
 350                                  *
 351                                  * For liblustre flock case, it maybe zero.
 352                                  */
 353                                 rc = -EPROTO;
 354                         } else {
 355                                 tsi->tsi_dlm_req = dlm_req;
 356                         }
 357                 } else {
 358                         rc = -EFAULT;
 359                 }
 360         }
 361         tsi->tsi_preprocessed = 1;
 362         RETURN(rc);
 363 }
 364
 365 /*
 366  * Invoke handler for this request opc. Also do necessary preprocessing
 367  * (according to handler ->th_flags), and post-processing (setting of
 368  * ->last_{xid,committed}).
 369  */
 370 static int tgt_handle_request0(struct tgt_session_info *tsi,
 371                                struct tgt_handler *h,
 372                                struct ptlrpc_request *req)
 373 {
 374         int      serious = 0;
 375         int      rc;
 376
 377         ENTRY;
 378
 379         /*
 380          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 381          * to put same checks into handlers like mdt_close(), mdt_reint(),
 382          * etc., without talking to mdt authors first. Checking same thing
 383          * there again is useless and returning 0 error without packing reply
 384          * is buggy! Handlers either pack reply or return error.
 385          *
 386          * We return 0 here and do not send any reply in order to emulate
 387          * network failure. Do not send any reply in case any of NET related
 388          * fail_id has occured.
 389          */
 390         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 391                 RETURN(0);
 392
 393         rc = tgt_request_preprocess(tsi, h, req);
 394         /* pack reply if reply format is fixed */
 395         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 396                 /* Pack reply */
 397                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 398                                           RCL_SERVER))
 399                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 400                                              RCL_SERVER,
 401                                              tsi->tsi_mdt_body->eadatasize);
 402                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 403                                           RCL_SERVER))
 404                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 405                                              RCL_SERVER, 0);
 406
 407                 rc = req_capsule_server_pack(tsi->tsi_pill);
 408         }
 409
 410         if (likely(rc == 0)) {
 411                 /*
 412                  * Process request, there can be two types of rc:
 413                  * 1) errors with msg unpack/pack, other failures outside the
 414                  * operation itself. This is counted as serious errors;
 415                  * 2) errors during fs operation, should be placed in rq_status
 416                  * only
 417                  */
 418                 rc = h->th_act(tsi);
 419                 if (!is_serious(rc) &&
 420                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 421                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 422                                   "pack reply and returned 0 error\n",
 423                                   tgt_name(tsi->tsi_tgt), h->th_name);
 424                         LBUG();
 425                 }
 426                 serious = is_serious(rc);
 427                 rc = clear_serious(rc);
 428         } else {
 429                 serious = 1;
 430         }
 431
 432         req->rq_status = rc;
 433
 434         /*
 435          * ELDLM_* codes which > 0 should be in rq_status only as well as
 436          * all non-serious errors.
 437          */
 438         if (rc > 0 || !serious)
 439                 rc = 0;
 440
 441         LASSERT(current->journal_info == NULL);
 442
 443         if (likely(rc == 0 && req->rq_export))
 444                 target_committed_to_req(req);
 445
 446         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 447         RETURN(0);
 448 }
 449
 450 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 451                                        struct obd_device *obd, int *process)
 452 {
 453         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 454         case MDS_DISCONNECT:
 455         case OST_DISCONNECT:
 456         case OBD_IDX_READ:
 457                 *process = 1;
 458                 RETURN(0);
 459         case MDS_CLOSE:
 460         case MDS_DONE_WRITING:
 461         case MDS_SYNC: /* used in unmounting */
 462         case OBD_PING:
 463         case MDS_REINT:
 464         case UPDATE_OBJ:
 465         case SEQ_QUERY:
 466         case FLD_QUERY:
 467         case LDLM_ENQUEUE:
 468         case OST_CREATE:
 469         case OST_DESTROY:
 470         case OST_PUNCH:
 471         case OST_SETATTR:
 472         case OST_SYNC:
 473         case OST_WRITE:
 474                 *process = target_queue_recovery_request(req, obd);
 475                 RETURN(0);
 476
 477         default:
 478                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 479                 *process = -EAGAIN;
 480                 RETURN(0);
 481         }
 482 }
 483
 484 /*
 485  * Handle recovery. Return:
 486  *        +1: continue request processing;
 487  *       -ve: abort immediately with the given error code;
 488  *         0: send reply with error code in req->rq_status;
 489  */
 490 int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 491 {
 492         ENTRY;
 493
 494         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 495         case MDS_CONNECT:
 496         case OST_CONNECT:
 497         case MGS_CONNECT:
 498         case SEC_CTX_INIT:
 499         case SEC_CTX_INIT_CONT:
 500         case SEC_CTX_FINI:
 501                 RETURN(+1);
 502         }
 503
 504         if (!req->rq_export->exp_obd->obd_replayable)
 505                 RETURN(+1);
 506
 507         /* sanity check: if the xid matches, the request must be marked as a
 508          * resent or replayed */
 509         if (req_xid_is_last(req)) {
 510                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 511                       (MSG_RESENT | MSG_REPLAY))) {
 512                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 513                                   "last_xid, expected REPLAY or RESENT flag "
 514                                   "(%x)", req->rq_xid,
 515                                   lustre_msg_get_flags(req->rq_reqmsg));
 516                         req->rq_status = -ENOTCONN;
 517                         RETURN(-ENOTCONN);
 518                 }
 519         }
 520         /* else: note the opposite is not always true; a RESENT req after a
 521          * failover will usually not match the last_xid, since it was likely
 522          * never committed. A REPLAYed request will almost never match the
 523          * last xid, however it could for a committed, but still retained,
 524          * open. */
 525
 526         /* Check for aborted recovery... */
 527         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 528                 int rc;
 529                 int should_process;
 530
 531                 DEBUG_REQ(D_INFO, req, "Got new replay");
 532                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 533                                                  &should_process);
 534                 if (rc != 0 || !should_process)
 535                         RETURN(rc);
 536                 else if (should_process < 0) {
 537                         req->rq_status = should_process;
 538                         rc = ptlrpc_error(req);
 539                         RETURN(rc);
 540                 }
 541         }
 542         RETURN(+1);
 543 }
 544
 545 /* Initial check for request, it is validation mostly */
 546 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 547 {
 548         struct tgt_handler      *h;
 549         struct tgt_opc_slice    *s;
 550         struct lu_target        *tgt;
 551         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 552
 553         ENTRY;
 554
 555         tgt = class_exp2tgt(req->rq_export);
 556
 557         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 558                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 559                         break;
 560
 561         /* opcode was not found in slice */
 562         if (unlikely(s->tos_hs == NULL)) {
 563                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 564                        opc);
 565                 RETURN(ERR_PTR(-ENOTSUPP));
 566         }
 567
 568         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 569         h = s->tos_hs + (opc - s->tos_opc_start);
 570         if (unlikely(h->th_opc == 0)) {
 571                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 572                 RETURN(ERR_PTR(-ENOTSUPP));
 573         }
 574
 575         RETURN(h);
 576 }
 577
 578 int tgt_request_handle(struct ptlrpc_request *req)
 579 {
 580         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 581
 582         struct lustre_msg       *msg = req->rq_reqmsg;
 583         struct tgt_handler      *h;
 584         struct lu_target        *tgt;
 585         int                      request_fail_id = 0;
 586         __u32                    opc = lustre_msg_get_opc(msg);
 587         int                      rc;
 588
 589         ENTRY;
 590
 591         /* Refill the context, to make sure all thread keys are allocated */
 592         lu_env_refill(req->rq_svc_thread->t_env);
 593
 594         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 595         tsi->tsi_pill = &req->rq_pill;
 596         tsi->tsi_env = req->rq_svc_thread->t_env;
 597
 598         /* if request has export then get handlers slice from corresponding
 599          * target, otherwise that should be connect operation */
 600         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 601             opc == MGS_CONNECT) {
 602                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 603                 rc = target_handle_connect(req);
 604                 if (rc != 0) {
 605                         rc = ptlrpc_error(req);
 606                         GOTO(out, rc);
 607                 }
 608                 /* recovery-small test 18c asks to drop connect reply */
 609                 if (unlikely(opc == OST_CONNECT &&
 610                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 611                         GOTO(out, rc = 0);
 612         }
 613
 614         if (unlikely(!class_connected_export(req->rq_export))) {
 615                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 616                        opc, libcfs_id2str(req->rq_peer));
 617                 req->rq_status = -ENOTCONN;
 618                 rc = ptlrpc_error(req);
 619                 GOTO(out, rc);
 620         }
 621
 622         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 623         tsi->tsi_exp = req->rq_export;
 624         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 625                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 626         else
 627                 tsi->tsi_jobid = NULL;
 628
 629         request_fail_id = tgt->lut_request_fail_id;
 630         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 631
 632         h = tgt_handler_find_check(req);
 633         if (IS_ERR(h)) {
 634                 req->rq_status = PTR_ERR(h);
 635                 rc = ptlrpc_error(req);
 636                 GOTO(out, rc);
 637         }
 638
 639         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 640                 GOTO(out, rc = 0);
 641
 642         rc = lustre_msg_check_version(msg, h->th_version);
 643         if (unlikely(rc)) {
 644                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 645                           " %08x, expecting %08x\n", tgt_name(tgt),
 646                           lustre_msg_get_version(msg), h->th_version);
 647                 req->rq_status = -EINVAL;
 648                 rc = ptlrpc_error(req);
 649                 GOTO(out, rc);
 650         }
 651
 652         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 653         if (likely(rc == 1)) {
 654                 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 655                          h->th_opc, opc);
 656                 rc = tgt_handle_request0(tsi, h, req);
 657                 if (rc)
 658                         GOTO(out, rc);
 659         }
 660         EXIT;
 661 out:
 662         req_capsule_fini(tsi->tsi_pill);
 663         if (tsi->tsi_corpus != NULL) {
 664                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 665                 tsi->tsi_corpus = NULL;
 666         }
 667         return rc;
 668 }
 669 EXPORT_SYMBOL(tgt_request_handle);
 670
 671 /** Assign high priority operations to the request if needed. */
 672 int tgt_hpreq_handler(struct ptlrpc_request *req)
 673 {
 674         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 675         struct tgt_handler      *h;
 676         int                      rc;
 677
 678         ENTRY;
 679
 680         if (req->rq_export == NULL)
 681                 RETURN(0);
 682
 683         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 684         tsi->tsi_pill = &req->rq_pill;
 685         tsi->tsi_env = req->rq_svc_thread->t_env;
 686         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 687         tsi->tsi_exp = req->rq_export;
 688
 689         h = tgt_handler_find_check(req);
 690         if (IS_ERR(h)) {
 691                 rc = PTR_ERR(h);
 692                 RETURN(rc);
 693         }
 694
 695         rc = tgt_request_preprocess(tsi, h, req);
 696         if (unlikely(rc != 0))
 697                 RETURN(rc);
 698
 699         if (h->th_hp != NULL)
 700                 h->th_hp(tsi);
 701         RETURN(0);
 702 }
 703 EXPORT_SYMBOL(tgt_hpreq_handler);
 704
 705 void tgt_counter_incr(struct obd_export *exp, int opcode)
 706 {
 707         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 708         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 709                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 710 }
 711 EXPORT_SYMBOL(tgt_counter_incr);
 712
 713 /*
 714  * Unified target generic handlers.
 715  */
 716
 717 /*
 718  * Security functions
 719  */
 720 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 721 {
 722         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 723                                       OBD_CONNECT_RMT_CLIENT_FORCE |
 724                                       OBD_CONNECT_MDS_CAPA |
 725                                       OBD_CONNECT_OSS_CAPA);
 726 }
 727
 728 static int tgt_init_sec_level(struct ptlrpc_request *req)
 729 {
 730         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 731         char                    *client = libcfs_nid2str(req->rq_peer.nid);
 732         struct obd_connect_data *data, *reply;
 733         int                      rc = 0;
 734         bool                     remote;
 735
 736         ENTRY;
 737
 738         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 739         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 740         if (data == NULL || reply == NULL)
 741                 RETURN(-EFAULT);
 742
 743         /* connection from MDT is always trusted */
 744         if (req->rq_auth_usr_mdt) {
 745                 tgt_init_sec_none(reply);
 746                 RETURN(0);
 747         }
 748
 749         /* no GSS support case */
 750         if (!req->rq_auth_gss) {
 751                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 752                         CWARN("client %s -> target %s does not use GSS, "
 753                               "can not run under security level %d.\n",
 754                               client, tgt_name(tgt), tgt->lut_sec_level);
 755                         RETURN(-EACCES);
 756                 } else {
 757                         tgt_init_sec_none(reply);
 758                         RETURN(0);
 759                 }
 760         }
 761
 762         /* old version case */
 763         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 764                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 765                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 766                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 767                         CWARN("client %s -> target %s uses old version, "
 768                               "can not run under security level %d.\n",
 769                               client, tgt_name(tgt), tgt->lut_sec_level);
 770                         RETURN(-EACCES);
 771                 } else {
 772                         CWARN("client %s -> target %s uses old version, "
 773                               "run under security level %d.\n",
 774                               client, tgt_name(tgt), tgt->lut_sec_level);
 775                         tgt_init_sec_none(reply);
 776                         RETURN(0);
 777                 }
 778         }
 779
 780         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 781         if (remote) {
 782                 if (!req->rq_auth_remote)
 783                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 784                                "asked to be remote.\n", client, tgt_name(tgt));
 785         } else if (req->rq_auth_remote) {
 786                 remote = true;
 787                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 788                        "as remote by default.\n", client, tgt_name(tgt));
 789         }
 790
 791         if (remote) {
 792                 if (!tgt->lut_oss_capa) {
 793                         CDEBUG(D_SEC,
 794                                "client %s -> target %s is set as remote,"
 795                                " but OSS capabilities are not enabled: %d.\n",
 796                                client, tgt_name(tgt), tgt->lut_oss_capa);
 797                         RETURN(-EACCES);
 798                 }
 799         } else {
 800                 if (req->rq_auth_uid == INVALID_UID) {
 801                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 802                                "authenticated!\n", client, tgt_name(tgt));
 803                         RETURN(-EACCES);
 804                 }
 805         }
 806
 807
 808         switch (tgt->lut_sec_level) {
 809         case LUSTRE_SEC_NONE:
 810                 if (remote) {
 811                         CDEBUG(D_SEC,
 812                                "client %s -> target %s is set as remote, "
 813                                "can not run under security level %d.\n",
 814                                client, tgt_name(tgt), tgt->lut_sec_level);
 815                         RETURN(-EACCES);
 816                 }
 817                 tgt_init_sec_none(reply);
 818                 break;
 819         case LUSTRE_SEC_REMOTE:
 820                 if (!remote)
 821                         tgt_init_sec_none(reply);
 822                 break;
 823         case LUSTRE_SEC_ALL:
 824                 if (remote)
 825                         break;
 826                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 827                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 828                 if (!tgt->lut_oss_capa)
 829                         reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 830                 if (!tgt->lut_mds_capa)
 831                         reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 832                 break;
 833         default:
 834                 RETURN(-EINVAL);
 835         }
 836
 837         RETURN(rc);
 838 }
 839
 840 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 841 {
 842         struct lu_target        *tgt = class_exp2tgt(exp);
 843         struct sptlrpc_flavor    flvr;
 844         int                      rc = 0;
 845
 846         LASSERT(tgt);
 847         LASSERT(tgt->lut_obd);
 848         LASSERT(tgt->lut_slice);
 849
 850         /* always allow ECHO client */
 851         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 852                             LUSTRE_ECHO_NAME) == 0)) {
 853                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 854                 return 0;
 855         }
 856
 857         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 858                 read_lock(&tgt->lut_sptlrpc_lock);
 859                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 860                                              req->rq_sp_from,
 861                                              req->rq_peer.nid,
 862                                              &flvr);
 863                 read_unlock(&tgt->lut_sptlrpc_lock);
 864
 865                 spin_lock(&exp->exp_lock);
 866                 exp->exp_sp_peer = req->rq_sp_from;
 867                 exp->exp_flvr = flvr;
 868                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 869                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 870                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 871                                "expect %x\n", tgt_name(tgt),
 872                                req->rq_flvr.sf_rpc,
 873                                libcfs_nid2str(req->rq_peer.nid),
 874                                exp->exp_flvr.sf_rpc);
 875                         rc = -EACCES;
 876                 }
 877                 spin_unlock(&exp->exp_lock);
 878         } else {
 879                 if (exp->exp_sp_peer != req->rq_sp_from) {
 880                         CERROR("%s: RPC source %s doesn't match %s\n",
 881                                tgt_name(tgt),
 882                                sptlrpc_part2name(req->rq_sp_from),
 883                                sptlrpc_part2name(exp->exp_sp_peer));
 884                         rc = -EACCES;
 885                 } else {
 886                         rc = sptlrpc_target_export_check(exp, req);
 887                 }
 888         }
 889
 890         return rc;
 891 }
 892
 893 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 894 {
 895         struct sptlrpc_rule_set  tmp_rset;
 896         int                      rc;
 897
 898         sptlrpc_rule_set_init(&tmp_rset);
 899         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 900         if (rc) {
 901                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 902                        tgt_name(tgt), rc);
 903                 return rc;
 904         }
 905
 906         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 907
 908         write_lock(&tgt->lut_sptlrpc_lock);
 909         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 910         tgt->lut_sptlrpc_rset = tmp_rset;
 911         write_unlock(&tgt->lut_sptlrpc_lock);
 912
 913         return 0;
 914 }
 915 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 916
 917 int tgt_connect(struct tgt_session_info *tsi)
 918 {
 919         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 920         struct obd_connect_data *reply;
 921         int                      rc;
 922
 923         ENTRY;
 924
 925         rc = tgt_init_sec_level(req);
 926         if (rc != 0)
 927                 GOTO(out, rc);
 928
 929         /* XXX: better to call this check right after getting new export but
 930          * before last_rcvd slot allocation to avoid server load upon insecure
 931          * connects. This is to be fixed after unifiyng all targets.
 932          */
 933         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
 934         if (rc)
 935                 GOTO(out, rc);
 936
 937         /* To avoid exposing partially initialized connection flags, changes up
 938          * to this point have been staged in reply->ocd_connect_flags. Now that
 939          * connection handling has completed successfully, atomically update
 940          * the connect flags in the shared export data structure. LU-1623 */
 941         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
 942         spin_lock(&tsi->tsi_exp->exp_lock);
 943         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
 944         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
 945         spin_unlock(&tsi->tsi_exp->exp_lock);
 946
 947         RETURN(0);
 948 out:
 949         obd_disconnect(class_export_get(tsi->tsi_exp));
 950         return rc;
 951 }
 952 EXPORT_SYMBOL(tgt_connect);
 953
 954 int tgt_disconnect(struct tgt_session_info *tsi)
 955 {
 956         int rc;
 957
 958         ENTRY;
 959
 960         rc = target_handle_disconnect(tgt_ses_req(tsi));
 961         if (rc)
 962                 RETURN(err_serious(rc));
 963
 964         RETURN(rc);
 965 }
 966 EXPORT_SYMBOL(tgt_disconnect);
 967
 968 /*
 969  * Unified target OBD handlers
 970  */
 971 int tgt_obd_ping(struct tgt_session_info *tsi)
 972 {
 973         int rc;
 974
 975         ENTRY;
 976
 977         rc = target_handle_ping(tgt_ses_req(tsi));
 978         if (rc)
 979                 RETURN(err_serious(rc));
 980
 981         RETURN(rc);
 982 }
 983 EXPORT_SYMBOL(tgt_obd_ping);
 984
 985 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
 986 {
 987         return err_serious(-EOPNOTSUPP);
 988 }
 989 EXPORT_SYMBOL(tgt_obd_log_cancel);
 990
 991 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
 992 {
 993         return err_serious(-EOPNOTSUPP);
 994 }
 995 EXPORT_SYMBOL(tgt_obd_qc_callback);
 996
 997 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
 998 {
 999         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1000         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1001         struct obd_export       *exp = req->rq_export;
1002         struct ptlrpc_bulk_desc *desc;
1003         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1004         int                      tmpcount;
1005         int                      tmpsize;
1006         int                      i;
1007         int                      rc;
1008
1009         ENTRY;
1010
1011         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1012                                     MDS_BULK_PORTAL);
1013         if (desc == NULL)
1014                 RETURN(-ENOMEM);
1015
1016         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1017                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1018                  * which is rdpg->rp_count */
1019                 nob = rdpg->rp_count;
1020
1021         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1022              i++, tmpcount -= tmpsize) {
1023                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1024                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1025         }
1026
1027         LASSERT(desc->bd_nob == nob);
1028         rc = target_bulk_io(exp, desc, lwi);
1029         ptlrpc_free_bulk_pin(desc);
1030         RETURN(rc);
1031 }
1032 EXPORT_SYMBOL(tgt_sendpage);
1033
1034 /*
1035  * OBD_IDX_READ handler
1036  */
1037 int tgt_obd_idx_read(struct tgt_session_info *tsi)
1038 {
1039         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1040         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1041         struct idx_info         *req_ii, *rep_ii;
1042         int                      rc, i;
1043
1044         ENTRY;
1045
1046         memset(rdpg, 0, sizeof(*rdpg));
1047         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1048
1049         /* extract idx_info buffer from request & reply */
1050         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1051         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1052                 RETURN(err_serious(-EPROTO));
1053
1054         rc = req_capsule_server_pack(tsi->tsi_pill);
1055         if (rc)
1056                 RETURN(err_serious(rc));
1057
1058         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1059         if (rep_ii == NULL)
1060                 RETURN(err_serious(-EFAULT));
1061         rep_ii->ii_magic = IDX_INFO_MAGIC;
1062
1063         /* extract hash to start with */
1064         rdpg->rp_hash = req_ii->ii_hash_start;
1065
1066         /* extract requested attributes */
1067         rdpg->rp_attrs = req_ii->ii_attrs;
1068
1069         /* check that fid packed in request is valid and supported */
1070         if (!fid_is_sane(&req_ii->ii_fid))
1071                 RETURN(-EINVAL);
1072         rep_ii->ii_fid = req_ii->ii_fid;
1073
1074         /* copy flags */
1075         rep_ii->ii_flags = req_ii->ii_flags;
1076
1077         /* compute number of pages to allocate, ii_count is the number of 4KB
1078          * containers */
1079         if (req_ii->ii_count <= 0)
1080                 GOTO(out, rc = -EFAULT);
1081         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1082                                exp_max_brw_size(tsi->tsi_exp));
1083         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1084
1085         /* allocate pages to store the containers */
1086         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1087         if (rdpg->rp_pages == NULL)
1088                 GOTO(out, rc = -ENOMEM);
1089         for (i = 0; i < rdpg->rp_npages; i++) {
1090                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1091                 if (rdpg->rp_pages[i] == NULL)
1092                         GOTO(out, rc = -ENOMEM);
1093         }
1094
1095         /* populate pages with key/record pairs */
1096         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1097         if (rc < 0)
1098                 GOTO(out, rc);
1099
1100         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1101                  "asked %d > %d\n", rc, rdpg->rp_count);
1102
1103         /* send pages to client */
1104         rc = tgt_sendpage(tsi, rdpg, rc);
1105         if (rc)
1106                 GOTO(out, rc);
1107         EXIT;
1108 out:
1109         if (rdpg->rp_pages) {
1110                 for (i = 0; i < rdpg->rp_npages; i++)
1111                         if (rdpg->rp_pages[i])
1112                                 __free_page(rdpg->rp_pages[i]);
1113                 OBD_FREE(rdpg->rp_pages,
1114                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1115         }
1116         return rc;
1117 }
1118 EXPORT_SYMBOL(tgt_obd_idx_read);
1119
1120 struct tgt_handler tgt_obd_handlers[] = {
1121 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1122 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1123 TGT_OBD_HDL_VAR(0,      OBD_QC_CALLBACK,        tgt_obd_qc_callback),
1124 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1125 };
1126 EXPORT_SYMBOL(tgt_obd_handlers);
1127
1128 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1129              struct dt_object *obj)
1130 {
1131         int rc = 0;
1132
1133         ENTRY;
1134
1135         /* if no objid is specified, it means "sync whole filesystem" */
1136         if (obj == NULL) {
1137                 rc = dt_sync(env, tgt->lut_bottom);
1138         } else if (dt_version_get(env, obj) >
1139                    tgt->lut_obd->obd_last_committed) {
1140                 rc = dt_object_sync(env, obj);
1141         }
1142
1143         RETURN(rc);
1144 }
1145 EXPORT_SYMBOL(tgt_sync);
1146 /*
1147  * Unified target DLM handlers.
1148  */
1149
1150 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1151  * (if requested) */
1152 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1153                      void *data, int flag)
1154 {
1155         struct lu_env            env;
1156         struct lu_target        *tgt;
1157         struct dt_object        *obj;
1158         struct lu_fid            fid;
1159         int                      rc = 0;
1160
1161         ENTRY;
1162
1163         tgt = class_exp2tgt(lock->l_export);
1164
1165         if (flag == LDLM_CB_CANCELING &&
1166             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1167             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1168              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1169               lock->l_flags & LDLM_FL_CBPENDING))) {
1170                 rc = lu_env_init(&env, LCT_DT_THREAD);
1171                 if (unlikely(rc != 0))
1172                         RETURN(rc);
1173
1174                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1175                                    tgt->lut_lsd.lsd_osd_index);
1176                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1177                 if (IS_ERR(obj))
1178                         GOTO(err_env, rc = PTR_ERR(obj));
1179
1180                 if (!dt_object_exists(obj))
1181                         GOTO(err_put, rc = -ENOENT);
1182
1183                 rc = tgt_sync(&env, tgt, obj);
1184                 if (rc < 0) {
1185                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1186                                "cancel: rc = %d\n",
1187                                tgt_name(tgt), PFID(&fid),
1188                                lock->l_policy_data.l_extent.start,
1189                                lock->l_policy_data.l_extent.end, rc);
1190                 }
1191 err_put:
1192                 lu_object_put(&env, &obj->do_lu);
1193 err_env:
1194                 lu_env_fini(&env);
1195         }
1196
1197         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1198         RETURN(rc);
1199 }
1200
1201 struct ldlm_callback_suite tgt_dlm_cbs = {
1202         .lcs_completion = ldlm_server_completion_ast,
1203         .lcs_blocking   = tgt_blocking_ast,
1204         .lcs_glimpse    = ldlm_server_glimpse_ast
1205 };
1206
1207 int tgt_enqueue(struct tgt_session_info *tsi)
1208 {
1209         struct ptlrpc_request *req = tgt_ses_req(tsi);
1210         int rc;
1211
1212         ENTRY;
1213         /*
1214          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1215          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1216          */
1217         LASSERT(tsi->tsi_dlm_req != NULL);
1218         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1219                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1220         if (rc)
1221                 RETURN(err_serious(rc));
1222
1223         RETURN(req->rq_status);
1224 }
1225 EXPORT_SYMBOL(tgt_enqueue);
1226
1227 int tgt_convert(struct tgt_session_info *tsi)
1228 {
1229         struct ptlrpc_request *req = tgt_ses_req(tsi);
1230         int rc;
1231
1232         ENTRY;
1233         LASSERT(tsi->tsi_dlm_req);
1234         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1235         if (rc)
1236                 RETURN(err_serious(rc));
1237
1238         RETURN(req->rq_status);
1239 }
1240 EXPORT_SYMBOL(tgt_convert);
1241
1242 int tgt_bl_callback(struct tgt_session_info *tsi)
1243 {
1244         return err_serious(-EOPNOTSUPP);
1245 }
1246 EXPORT_SYMBOL(tgt_bl_callback);
1247
1248 int tgt_cp_callback(struct tgt_session_info *tsi)
1249 {
1250         return err_serious(-EOPNOTSUPP);
1251 }
1252 EXPORT_SYMBOL(tgt_cp_callback);
1253
1254 /* generic LDLM target handler */
1255 struct tgt_handler tgt_dlm_handlers[] = {
1256 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1257 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1258 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1259 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1260 };
1261 EXPORT_SYMBOL(tgt_dlm_handlers);
1262
1263 /*
1264  * Unified target LLOG handlers.
1265  */
1266 int tgt_llog_open(struct tgt_session_info *tsi)
1267 {
1268         int rc;
1269
1270         ENTRY;
1271
1272         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1273
1274         RETURN(rc);
1275 }
1276 EXPORT_SYMBOL(tgt_llog_open);
1277
1278 int tgt_llog_close(struct tgt_session_info *tsi)
1279 {
1280         int rc;
1281
1282         ENTRY;
1283
1284         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1285
1286         RETURN(rc);
1287 }
1288 EXPORT_SYMBOL(tgt_llog_close);
1289
1290
1291 int tgt_llog_destroy(struct tgt_session_info *tsi)
1292 {
1293         int rc;
1294
1295         ENTRY;
1296
1297         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1298
1299         RETURN(rc);
1300 }
1301 EXPORT_SYMBOL(tgt_llog_destroy);
1302
1303 int tgt_llog_read_header(struct tgt_session_info *tsi)
1304 {
1305         int rc;
1306
1307         ENTRY;
1308
1309         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1310
1311         RETURN(rc);
1312 }
1313 EXPORT_SYMBOL(tgt_llog_read_header);
1314
1315 int tgt_llog_next_block(struct tgt_session_info *tsi)
1316 {
1317         int rc;
1318
1319         ENTRY;
1320
1321         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1322
1323         RETURN(rc);
1324 }
1325 EXPORT_SYMBOL(tgt_llog_next_block);
1326
1327 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1328 {
1329         int rc;
1330
1331         ENTRY;
1332
1333         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1334
1335         RETURN(rc);
1336 }
1337 EXPORT_SYMBOL(tgt_llog_prev_block);
1338
1339 /* generic llog target handler */
1340 struct tgt_handler tgt_llog_handlers[] = {
1341 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1342 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1343 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1344 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1345 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1346 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1347 };
1348 EXPORT_SYMBOL(tgt_llog_handlers);
1349
1350 /*
1351  * sec context handlers
1352  */
1353 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1354 int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1355 {
1356         return 0;
1357 }
1358
1359 struct tgt_handler tgt_sec_ctx_handlers[] = {
1360 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1361 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1362 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1363 };
1364 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1365
1366 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1367                            struct dt_device *key,
1368                            struct lfsck_request *lr) = NULL;
1369
1370 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1371                                                 struct dt_device *,
1372                                                 struct lfsck_request *))
1373 {
1374         tgt_lfsck_in_notify = notify;
1375 }
1376 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1377
1378 static int (*tgt_lfsck_query)(const struct lu_env *env,
1379                               struct dt_device *key,
1380                               struct lfsck_request *lr) = NULL;
1381
1382 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1383                                            struct dt_device *,
1384                                            struct lfsck_request *))
1385 {
1386         tgt_lfsck_query = query;
1387 }
1388 EXPORT_SYMBOL(tgt_register_lfsck_query);
1389
1390 /* LFSCK request handlers */
1391 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1392 {
1393         const struct lu_env     *env = tsi->tsi_env;
1394         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1395         struct lfsck_request    *lr;
1396         int                      rc;
1397         ENTRY;
1398
1399         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1400         if (lr == NULL)
1401                 RETURN(-EPROTO);
1402
1403         rc = tgt_lfsck_in_notify(env, key, lr);
1404
1405         RETURN(rc);
1406 }
1407
1408 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1409 {
1410         struct lfsck_request    *request;
1411         struct lfsck_reply      *reply;
1412         int                      rc      = 0;
1413         ENTRY;
1414
1415         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1416         if (request == NULL)
1417                 RETURN(-EPROTO);
1418
1419         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1420         if (reply == NULL)
1421                 RETURN(-ENOMEM);
1422
1423         reply->lr_status = tgt_lfsck_query(tsi->tsi_env,
1424                                            tsi->tsi_tgt->lut_bottom, request);
1425         if (reply->lr_status < 0)
1426                 rc = reply->lr_status;
1427
1428         RETURN(rc);
1429 }
1430
1431 struct tgt_handler tgt_lfsck_handlers[] = {
1432 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1433 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1434 };
1435 EXPORT_SYMBOL(tgt_lfsck_handlers);
1436
1437 /*
1438  * initialize per-thread page pool (bug 5137).
1439  */
1440 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1441 {
1442         struct tgt_thread_big_cache *tbc;
1443
1444         ENTRY;
1445
1446         LASSERT(thread != NULL);
1447         LASSERT(thread->t_data == NULL);
1448
1449         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1450         if (tbc == NULL)
1451                 RETURN(-ENOMEM);
1452         thread->t_data = tbc;
1453         RETURN(0);
1454 }
1455 EXPORT_SYMBOL(tgt_io_thread_init);
1456
1457 /*
1458  * free per-thread pool created by tgt_thread_init().
1459  */
1460 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1461 {
1462         struct tgt_thread_big_cache *tbc;
1463
1464         ENTRY;
1465
1466         LASSERT(thread != NULL);
1467
1468         /*
1469          * be prepared to handle partially-initialized pools (because this is
1470          * called from ost_io_thread_init() for cleanup.
1471          */
1472         tbc = thread->t_data;
1473         if (tbc != NULL) {
1474                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1475                 thread->t_data = NULL;
1476         }
1477         EXIT;
1478 }
1479 EXPORT_SYMBOL(tgt_io_thread_done);
1480 /**
1481  * Helper function for getting server side [start, start+count] DLM lock
1482  * if asked by client.
1483  */
1484 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1485                     __u64 start, __u64 end, struct lustre_handle *lh,
1486                     int mode, __u64 *flags)
1487 {
1488         ldlm_policy_data_t       policy;
1489         int                      rc;
1490
1491         ENTRY;
1492
1493         LASSERT(lh != NULL);
1494         LASSERT(ns != NULL);
1495         LASSERT(!lustre_handle_is_used(lh));
1496
1497         policy.l_extent.gid = 0;
1498         policy.l_extent.start = start & CFS_PAGE_MASK;
1499
1500         /*
1501          * If ->o_blocks is EOF it means "lock till the end of the file".
1502          * Otherwise, it's size of an extent or hole being punched (in bytes).
1503          */
1504         if (end == OBD_OBJECT_EOF || end < start)
1505                 policy.l_extent.end = OBD_OBJECT_EOF;
1506         else
1507                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1508
1509         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1510                                     flags, ldlm_blocking_ast,
1511                                     ldlm_completion_ast, ldlm_glimpse_ast,
1512                                     NULL, 0, LVB_T_NONE, NULL, lh);
1513         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1514 }
1515 EXPORT_SYMBOL(tgt_extent_lock);
1516
1517 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1518 {
1519         LASSERT(lustre_handle_is_used(lh));
1520         ldlm_lock_decref(lh, mode);
1521 }
1522 EXPORT_SYMBOL(tgt_extent_unlock);
1523
1524 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1525                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1526                  struct lustre_handle *lh, int mode)
1527 {
1528         __u64                    flags = 0;
1529         int                      nrbufs = obj->ioo_bufcnt;
1530         int                      i;
1531
1532         ENTRY;
1533
1534         LASSERT(mode == LCK_PR || mode == LCK_PW);
1535         LASSERT(!lustre_handle_is_used(lh));
1536
1537         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1538                 RETURN(0);
1539
1540         for (i = 1; i < nrbufs; i++)
1541                 if (!(nb[i].flags & OBD_BRW_SRVLOCK))
1542                         RETURN(-EFAULT);
1543
1544         RETURN(tgt_extent_lock(ns, res_id, nb[0].offset,
1545                                nb[nrbufs - 1].offset + nb[nrbufs - 1].len - 1,
1546                                lh, mode, &flags));
1547 }
1548 EXPORT_SYMBOL(tgt_brw_lock);
1549
1550 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1551                     struct lustre_handle *lh, int mode)
1552 {
1553         ENTRY;
1554
1555         LASSERT(mode == LCK_PR || mode == LCK_PW);
1556         LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
1557                 lustre_handle_is_used(lh));
1558         if (lustre_handle_is_used(lh))
1559                 tgt_extent_unlock(lh, mode);
1560         EXIT;
1561 }
1562 EXPORT_SYMBOL(tgt_brw_unlock);
1563
1564 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1565                                struct ptlrpc_bulk_desc *desc, int opc,
1566                                cksum_type_t cksum_type)
1567 {
1568         struct cfs_crypto_hash_desc     *hdesc;
1569         unsigned int                    bufsize;
1570         int                             i, err;
1571         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1572         __u32                           cksum;
1573
1574         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1575         if (IS_ERR(hdesc)) {
1576                 CERROR("%s: unable to initialize checksum hash %s\n",
1577                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1578                 return PTR_ERR(hdesc);
1579         }
1580
1581         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1582         for (i = 0; i < desc->bd_iov_count; i++) {
1583                 /* corrupt the data before we compute the checksum, to
1584                  * simulate a client->OST data error */
1585                 if (i == 0 && opc == OST_WRITE &&
1586                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1587                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1588                         int len = desc->bd_iov[i].kiov_len;
1589                         struct page *np = tgt_page_to_corrupt;
1590                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1591
1592                         if (np) {
1593                                 char *ptr2 = kmap(np) + off;
1594
1595                                 memcpy(ptr2, ptr, len);
1596                                 memcpy(ptr2, "bad3", min(4, len));
1597                                 kunmap(np);
1598                                 desc->bd_iov[i].kiov_page = np;
1599                         } else {
1600                                 CERROR("%s: can't alloc page for corruption\n",
1601                                        tgt_name(tgt));
1602                         }
1603                 }
1604                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1605                                   desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1606                                   desc->bd_iov[i].kiov_len);
1607
1608                  /* corrupt the data after we compute the checksum, to
1609                  * simulate an OST->client data error */
1610                 if (i == 0 && opc == OST_READ &&
1611                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1612                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1613                         int len = desc->bd_iov[i].kiov_len;
1614                         struct page *np = tgt_page_to_corrupt;
1615                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1616
1617                         if (np) {
1618                                 char *ptr2 = kmap(np) + off;
1619
1620                                 memcpy(ptr2, ptr, len);
1621                                 memcpy(ptr2, "bad4", min(4, len));
1622                                 kunmap(np);
1623                                 desc->bd_iov[i].kiov_page = np;
1624                         } else {
1625                                 CERROR("%s: can't alloc page for corruption\n",
1626                                        tgt_name(tgt));
1627                         }
1628                 }
1629         }
1630
1631         bufsize = 4;
1632         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1633         if (err)
1634                 cfs_crypto_hash_final(hdesc, NULL, NULL);
1635
1636         return cksum;
1637 }
1638
1639 int tgt_brw_read(struct tgt_session_info *tsi)
1640 {
1641         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1642         struct ptlrpc_bulk_desc *desc = NULL;
1643         struct obd_export       *exp = tsi->tsi_exp;
1644         struct niobuf_remote    *remote_nb;
1645         struct niobuf_local     *local_nb;
1646         struct obd_ioobj        *ioo;
1647         struct ost_body         *body, *repbody;
1648         struct l_wait_info       lwi;
1649         struct lustre_handle     lockh = { 0 };
1650         int                      niocount, npages, nob = 0, rc, i;
1651         int                      no_reply = 0;
1652         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1653
1654         ENTRY;
1655
1656         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1657                 CERROR("%s: deny read request from %s to portal %u\n",
1658                        tgt_name(tsi->tsi_tgt),
1659                        obd_export_nid2str(req->rq_export),
1660                        ptlrpc_req2svc(req)->srv_req_portal);
1661                 RETURN(-EPROTO);
1662         }
1663
1664         req->rq_bulk_read = 1;
1665
1666         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1667                 RETURN(-EIO);
1668
1669         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1670                          cfs_fail_val : (obd_timeout + 1) / 4);
1671
1672         /* Check if there is eviction in progress, and if so, wait for it to
1673          * finish */
1674         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1675                 /* We do not care how long it takes */
1676                 lwi = LWI_INTR(NULL, NULL);
1677                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1678                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1679                          &lwi);
1680         }
1681
1682         /* There must be big cache in current thread to process this request
1683          * if it is NULL then something went wrong and it wasn't allocated,
1684          * report -ENOMEM in that case */
1685         if (tbc == NULL)
1686                 RETURN(-ENOMEM);
1687
1688         body = tsi->tsi_ost_body;
1689         LASSERT(body != NULL);
1690
1691         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1692         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1693
1694         niocount = ioo->ioo_bufcnt;
1695         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1696         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1697
1698         local_nb = tbc->local;
1699
1700         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1701                           remote_nb, &lockh, LCK_PR);
1702         if (rc != 0)
1703                 RETURN(rc);
1704
1705         /*
1706          * If getting the lock took more time than
1707          * client was willing to wait, drop it. b=11330
1708          */
1709         if (cfs_time_current_sec() > req->rq_deadline ||
1710             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1711                 no_reply = 1;
1712                 CERROR("Dropping timed-out read from %s because locking"
1713                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1714                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1715                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1716                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1717                 GOTO(out_lock, rc = -ETIMEDOUT);
1718         }
1719
1720         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1721         repbody->oa = body->oa;
1722
1723         npages = PTLRPC_MAX_BRW_PAGES;
1724         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1725                         ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1726         if (rc != 0)
1727                 GOTO(out_lock, rc);
1728
1729         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1730                                     BULK_PUT_SOURCE, OST_BULK_PORTAL);
1731         if (desc == NULL)
1732                 GOTO(out_commitrw, rc = -ENOMEM);
1733
1734         nob = 0;
1735         for (i = 0; i < npages; i++) {
1736                 int page_rc = local_nb[i].rc;
1737
1738                 if (page_rc < 0) {
1739                         rc = page_rc;
1740                         break;
1741                 }
1742
1743                 nob += page_rc;
1744                 if (page_rc != 0) { /* some data! */
1745                         LASSERT(local_nb[i].page != NULL);
1746                         ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1747                                                     local_nb[i].lnb_page_offset,
1748                                                     page_rc);
1749                 }
1750
1751                 if (page_rc != local_nb[i].len) { /* short read */
1752                         /* All subsequent pages should be 0 */
1753                         while (++i < npages)
1754                                 LASSERT(local_nb[i].rc == 0);
1755                         break;
1756                 }
1757         }
1758
1759         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1760                 cksum_type_t cksum_type =
1761                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1762                                           body->oa.o_flags : 0);
1763                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1764                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1765                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1766                                                         OST_READ, cksum_type);
1767                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1768                        repbody->oa.o_cksum);
1769         } else {
1770                 repbody->oa.o_valid = 0;
1771         }
1772         /* We're finishing using body->oa as an input variable */
1773
1774         /* Check if client was evicted while we were doing i/o before touching
1775          * network */
1776         if (likely(rc == 0 &&
1777                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1778                 rc = target_bulk_io(exp, desc, &lwi);
1779                 no_reply = rc != 0;
1780         }
1781
1782 out_commitrw:
1783         /* Must commit after prep above in all cases */
1784         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1785                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1786                           NULL, rc);
1787         if (rc == 0)
1788                 tgt_drop_id(exp, &repbody->oa);
1789 out_lock:
1790         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1791
1792         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1793                 ptlrpc_free_bulk_nopin(desc);
1794
1795         LASSERT(rc <= 0);
1796         if (rc == 0) {
1797                 rc = nob;
1798                 ptlrpc_lprocfs_brw(req, nob);
1799         } else if (no_reply) {
1800                 req->rq_no_reply = 1;
1801                 /* reply out callback would free */
1802                 ptlrpc_req_drop_rs(req);
1803                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1804                               "client will retry: rc %d\n",
1805                               exp->exp_obd->obd_name,
1806                               obd_uuid2str(&exp->exp_client_uuid),
1807                               obd_export_nid2str(exp), rc);
1808         }
1809         /* send a bulk after reply to simulate a network delay or reordering
1810          * by a router */
1811         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1812                 wait_queue_head_t        waitq;
1813                 struct l_wait_info       lwi1;
1814
1815                 CDEBUG(D_INFO, "reorder BULK\n");
1816                 init_waitqueue_head(&waitq);
1817
1818                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1819                 l_wait_event(waitq, 0, &lwi1);
1820                 target_bulk_io(exp, desc, &lwi);
1821                 ptlrpc_free_bulk_nopin(desc);
1822         }
1823
1824         RETURN(rc);
1825 }
1826 EXPORT_SYMBOL(tgt_brw_read);
1827
1828 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1829                               struct ptlrpc_bulk_desc *desc,
1830                               struct niobuf_local *local_nb, int npages,
1831                               obd_count client_cksum, obd_count server_cksum,
1832                               bool mmap)
1833 {
1834         struct obd_export *exp = req->rq_export;
1835         struct ost_body *body;
1836         char *router;
1837         char *via;
1838
1839         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1840         LASSERT(body != NULL);
1841
1842         if (req->rq_peer.nid == desc->bd_sender) {
1843                 via = router = "";
1844         } else {
1845                 via = " via ";
1846                 router = libcfs_nid2str(desc->bd_sender);
1847         }
1848
1849         if (mmap) {
1850                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1851                              client_cksum, server_cksum);
1852                 return;
1853         }
1854
1855         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1856                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1857                            "]: client csum %x, server csum %x\n",
1858                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1859                            via, router,
1860                            body->oa.o_valid & OBD_MD_FLFID ?
1861                            body->oa.o_parent_seq : (__u64)0,
1862                            body->oa.o_valid & OBD_MD_FLFID ?
1863                            body->oa.o_parent_oid : 0,
1864                            body->oa.o_valid & OBD_MD_FLFID ?
1865                            body->oa.o_parent_ver : 0,
1866                            POSTID(&body->oa.o_oi),
1867                            local_nb[0].lnb_file_offset,
1868                            local_nb[npages-1].lnb_file_offset +
1869                            local_nb[npages-1].len - 1,
1870                            client_cksum, server_cksum);
1871 }
1872
1873 int tgt_brw_write(struct tgt_session_info *tsi)
1874 {
1875         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1876         struct ptlrpc_bulk_desc *desc = NULL;
1877         struct obd_export       *exp = req->rq_export;
1878         struct niobuf_remote    *remote_nb;
1879         struct niobuf_local     *local_nb;
1880         struct obd_ioobj        *ioo;
1881         struct ost_body         *body, *repbody;
1882         struct l_wait_info       lwi;
1883         struct lustre_handle     lockh = {0};
1884         __u32                   *rcs;
1885         int                      objcount, niocount, npages;
1886         int                      rc, i, j;
1887         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1888         bool                     no_reply = false, mmap;
1889         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1890
1891         ENTRY;
1892
1893         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1894                 CERROR("%s: deny write request from %s to portal %u\n",
1895                        tgt_name(tsi->tsi_tgt),
1896                        obd_export_nid2str(req->rq_export),
1897                        ptlrpc_req2svc(req)->srv_req_portal);
1898                 RETURN(err_serious(-EPROTO));
1899         }
1900
1901         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1902                 RETURN(err_serious(-ENOSPC));
1903         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1904                 RETURN(err_serious(-EROFS));
1905
1906         req->rq_bulk_write = 1;
1907
1908         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1909                 RETURN(err_serious(-EIO));
1910         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1911                 RETURN(err_serious(-EFAULT));
1912
1913         /* pause before transaction has been started */
1914         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1915                          cfs_fail_val : (obd_timeout + 1) / 4);
1916
1917         /* There must be big cache in current thread to process this request
1918          * if it is NULL then something went wrong and it wasn't allocated,
1919          * report -ENOMEM in that case */
1920         if (tbc == NULL)
1921                 RETURN(-ENOMEM);
1922
1923         body = tsi->tsi_ost_body;
1924         LASSERT(body != NULL);
1925
1926         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1927         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1928
1929         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1930                                         RCL_CLIENT) / sizeof(*ioo);
1931
1932         for (niocount = i = 0; i < objcount; i++)
1933                 niocount += ioo[i].ioo_bufcnt;
1934
1935         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1936         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1937         if (niocount != req_capsule_get_size(&req->rq_pill,
1938                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1939                         sizeof(*remote_nb))
1940                 RETURN(err_serious(-EPROTO));
1941
1942         if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
1943             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1944                 memory_pressure_set();
1945
1946         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1947                              niocount * sizeof(*rcs));
1948         rc = req_capsule_server_pack(&req->rq_pill);
1949         if (rc != 0)
1950                 GOTO(out, rc = err_serious(rc));
1951
1952         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1953         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1954
1955         local_nb = tbc->local;
1956
1957         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1958                           remote_nb, &lockh, LCK_PW);
1959         if (rc != 0)
1960                 GOTO(out, rc);
1961
1962         /*
1963          * If getting the lock took more time than
1964          * client was willing to wait, drop it. b=11330
1965          */
1966         if (cfs_time_current_sec() > req->rq_deadline ||
1967             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1968                 no_reply = true;
1969                 CERROR("%s: Dropping timed-out write from %s because locking "
1970                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1971                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1972                        POSTID(&ioo->ioo_oid),
1973                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1974                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1975                 GOTO(out_lock, rc = -ETIMEDOUT);
1976         }
1977
1978         /* Because we already sync grant info with client when reconnect,
1979          * grant info will be cleared for resent req, then fed_grant and
1980          * total_grant will not be modified in following preprw_write */
1981         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1982                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1983                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1984         }
1985
1986         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1987         if (repbody == NULL)
1988                 GOTO(out_lock, rc = -ENOMEM);
1989         repbody->oa = body->oa;
1990
1991         npages = PTLRPC_MAX_BRW_PAGES;
1992         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1993                         objcount, ioo, remote_nb, &npages, local_nb, NULL,
1994                         BYPASS_CAPA);
1995         if (rc < 0)
1996                 GOTO(out_lock, rc);
1997
1998         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1999                                     BULK_GET_SINK, OST_BULK_PORTAL);
2000         if (desc == NULL)
2001                 GOTO(skip_transfer, rc = -ENOMEM);
2002
2003         /* NB Having prepped, we must commit... */
2004         for (i = 0; i < npages; i++)
2005                 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
2006                                             local_nb[i].lnb_page_offset,
2007                                             local_nb[i].len);
2008
2009         rc = sptlrpc_svc_prep_bulk(req, desc);
2010         if (rc != 0)
2011                 GOTO(skip_transfer, rc);
2012
2013         rc = target_bulk_io(exp, desc, &lwi);
2014         no_reply = rc != 0;
2015
2016 skip_transfer:
2017         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2018                 static int cksum_counter;
2019
2020                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2021                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2022
2023                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2024                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2025                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2026                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2027                                                         OST_WRITE, cksum_type);
2028                 cksum_counter++;
2029
2030                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2031                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2032                                 body->oa.o_flags & OBD_FL_MMAP);
2033
2034                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2035                                           body->oa.o_cksum,
2036                                           repbody->oa.o_cksum, mmap);
2037                         cksum_counter = 0;
2038                 } else if ((cksum_counter & (-cksum_counter)) ==
2039                            cksum_counter) {
2040                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2041                                cksum_counter, libcfs_id2str(req->rq_peer),
2042                                repbody->oa.o_cksum);
2043                 }
2044         }
2045
2046         /* Must commit after prep above in all cases */
2047         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2048                           objcount, ioo, remote_nb, npages, local_nb, NULL,
2049                           rc);
2050         if (rc == -ENOTCONN)
2051                 /* quota acquire process has been given up because
2052                  * either the client has been evicted or the client
2053                  * has timed out the request already */
2054                 no_reply = true;
2055
2056         /*
2057          * Disable sending mtime back to the client. If the client locked the
2058          * whole object, then it has already updated the mtime on its side,
2059          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2060          */
2061         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2062
2063         if (rc == 0) {
2064                 int nob = 0;
2065
2066                 /* set per-requested niobuf return codes */
2067                 for (i = j = 0; i < niocount; i++) {
2068                         int len = remote_nb[i].len;
2069
2070                         nob += len;
2071                         rcs[i] = 0;
2072                         do {
2073                                 LASSERT(j < npages);
2074                                 if (local_nb[j].rc < 0)
2075                                         rcs[i] = local_nb[j].rc;
2076                                 len -= local_nb[j].len;
2077                                 j++;
2078                         } while (len > 0);
2079                         LASSERT(len == 0);
2080                 }
2081                 LASSERT(j == npages);
2082                 ptlrpc_lprocfs_brw(req, nob);
2083
2084                 tgt_drop_id(exp, &repbody->oa);
2085         }
2086 out_lock:
2087         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2088         if (desc)
2089                 ptlrpc_free_bulk_nopin(desc);
2090 out:
2091         if (no_reply) {
2092                 req->rq_no_reply = 1;
2093                 /* reply out callback would free */
2094                 ptlrpc_req_drop_rs(req);
2095                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2096                               "client will retry: rc %d\n",
2097                               exp->exp_obd->obd_name,
2098                               obd_uuid2str(&exp->exp_client_uuid),
2099                               obd_export_nid2str(exp), rc);
2100         }
2101         memory_pressure_clr();
2102         RETURN(rc);
2103 }
2104 EXPORT_SYMBOL(tgt_brw_write);