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 #include <lustre_nodemap.h>
  43
  44 #include "tgt_internal.h"
  45
  46 char *tgt_name(struct lu_target *tgt)
  47 {
  48         LASSERT(tgt->lut_obd != NULL);
  49         return tgt->lut_obd->obd_name;
  50 }
  51 EXPORT_SYMBOL(tgt_name);
  52
  53 /*
  54  * Generic code handling requests that have struct mdt_body passed in:
  55  *
  56  *  - extract mdt_body from request and save it in @tsi, if present;
  57  *
  58  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  59  *  @tsi;
  60  *
  61  *  - if HABEO_CORPUS flag is set for this request type check whether object
  62  *  actually exists on storage (lu_object_exists()).
  63  *
  64  */
  65 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  66 {
  67         const struct mdt_body   *body;
  68         struct lu_object        *obj;
  69         struct req_capsule      *pill = tsi->tsi_pill;
  70         int                      rc;
  71
  72         ENTRY;
  73
  74         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  75         if (body == NULL)
  76                 RETURN(-EFAULT);
  77
  78         tsi->tsi_mdt_body = body;
  79
  80         if (!(body->mbo_valid & OBD_MD_FLID))
  81                 RETURN(0);
  82
  83         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  84          * in any case in pre-2.5 clients. Fix that here if needed */
  85         if (unlikely(fid_is_zero(&body->mbo_fid1)))
  86                 RETURN(0);
  87
  88         if (!fid_is_sane(&body->mbo_fid1)) {
  89                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  90                        PFID(&body->mbo_fid1));
  91                 RETURN(-EINVAL);
  92         }
  93
  94         obj = lu_object_find(tsi->tsi_env,
  95                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
  96                              &body->mbo_fid1, NULL);
  97         if (!IS_ERR(obj)) {
  98                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
  99                         lu_object_put(tsi->tsi_env, obj);
 100                         /* for capability renew ENOENT will be handled in
 101                          * mdt_renew_capa */
 102                         if (body->mbo_valid & OBD_MD_FLOSSCAPA)
 103                                 rc = 0;
 104                         else
 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         obd_seq          seq    = ostid_seq(oi);
 138         obd_id           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 lustre_capa      *capa;
 246         struct lu_nodemap       *nodemap;
 247         int                      rc;
 248
 249         ENTRY;
 250
 251         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 252         if (body == NULL)
 253                 RETURN(-EFAULT);
 254
 255         rc = tgt_validate_obdo(tsi, &body->oa);
 256         if (rc)
 257                 RETURN(rc);
 258
 259         nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
 260
 261         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 262                                         NODEMAP_CLIENT_TO_FS,
 263                                         body->oa.o_uid);
 264         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 265                                         NODEMAP_CLIENT_TO_FS,
 266                                         body->oa.o_gid);
 267
 268         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 269                 capa = req_capsule_client_get(pill, &RMF_CAPA1);
 270                 if (capa == NULL) {
 271                         CERROR("%s: OSSCAPA flag is set without capability\n",
 272                                tgt_name(tsi->tsi_tgt));
 273                         RETURN(-EFAULT);
 274                 }
 275         }
 276
 277         tsi->tsi_ost_body = body;
 278         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 279
 280         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 281                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 282                 if (rc < 0)
 283                         RETURN(rc);
 284         }
 285
 286         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 287                 if (flags & HABEO_CORPUS) {
 288                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 289                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 290                                tgt_name(tsi->tsi_tgt));
 291                         RETURN(-EPROTO);
 292                 } else {
 293                         RETURN(0);
 294                 }
 295         }
 296
 297         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 298
 299         /*
 300          * OST doesn't get object in advance for further use to prevent
 301          * situations with nested object_find which is potential deadlock.
 302          */
 303         tsi->tsi_corpus = NULL;
 304         RETURN(rc);
 305 }
 306
 307 /*
 308  * Do necessary preprocessing according to handler ->th_flags.
 309  */
 310 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 311                                   struct tgt_handler *h,
 312                                   struct ptlrpc_request *req)
 313 {
 314         struct req_capsule      *pill = tsi->tsi_pill;
 315         __u32                    flags = h->th_flags;
 316         int                      rc = 0;
 317
 318         ENTRY;
 319
 320         if (tsi->tsi_preprocessed)
 321                 RETURN(0);
 322
 323         LASSERT(h->th_act != NULL);
 324         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 325         LASSERT(current->journal_info == NULL);
 326
 327         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 328                      h->th_fmt != NULL));
 329         if (h->th_fmt != NULL) {
 330                 req_capsule_set(pill, h->th_fmt);
 331                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 332                         rc = tgt_mdt_body_unpack(tsi, flags);
 333                         if (rc < 0)
 334                                 RETURN(rc);
 335                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 336                                                  RCL_CLIENT)) {
 337                         rc = tgt_ost_body_unpack(tsi, flags);
 338                         if (rc < 0)
 339                                 RETURN(rc);
 340                 }
 341         }
 342
 343         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 344                 RETURN(-EROFS);
 345
 346         if (flags & HABEO_CLAVIS) {
 347                 struct ldlm_request *dlm_req;
 348
 349                 LASSERT(h->th_fmt != NULL);
 350
 351                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 352                 if (dlm_req != NULL) {
 353                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 354                                      LDLM_IBITS &&
 355                                      dlm_req->lock_desc.l_policy_data.\
 356                                      l_inodebits.bits == 0)) {
 357                                 /*
 358                                  * Lock without inodebits makes no sense and
 359                                  * will oops later in ldlm. If client miss to
 360                                  * set such bits, do not trigger ASSERTION.
 361                                  *
 362                                  * For liblustre flock case, it maybe zero.
 363                                  */
 364                                 rc = -EPROTO;
 365                         } else {
 366                                 tsi->tsi_dlm_req = dlm_req;
 367                         }
 368                 } else {
 369                         rc = -EFAULT;
 370                 }
 371         }
 372         tsi->tsi_preprocessed = 1;
 373         RETURN(rc);
 374 }
 375
 376 /*
 377  * Invoke handler for this request opc. Also do necessary preprocessing
 378  * (according to handler ->th_flags), and post-processing (setting of
 379  * ->last_{xid,committed}).
 380  */
 381 static int tgt_handle_request0(struct tgt_session_info *tsi,
 382                                struct tgt_handler *h,
 383                                struct ptlrpc_request *req)
 384 {
 385         int      serious = 0;
 386         int      rc;
 387
 388         ENTRY;
 389
 390         /*
 391          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 392          * to put same checks into handlers like mdt_close(), mdt_reint(),
 393          * etc., without talking to mdt authors first. Checking same thing
 394          * there again is useless and returning 0 error without packing reply
 395          * is buggy! Handlers either pack reply or return error.
 396          *
 397          * We return 0 here and do not send any reply in order to emulate
 398          * network failure. Do not send any reply in case any of NET related
 399          * fail_id has occured.
 400          */
 401         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 402                 RETURN(0);
 403
 404         rc = tgt_request_preprocess(tsi, h, req);
 405         /* pack reply if reply format is fixed */
 406         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 407                 /* Pack reply */
 408                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 409                                           RCL_SERVER))
 410                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 411                                              RCL_SERVER,
 412                                              tsi->tsi_mdt_body->mbo_eadatasize);
 413                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 414                                           RCL_SERVER))
 415                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 416                                              RCL_SERVER, 0);
 417
 418                 rc = req_capsule_server_pack(tsi->tsi_pill);
 419         }
 420
 421         if (likely(rc == 0)) {
 422                 /*
 423                  * Process request, there can be two types of rc:
 424                  * 1) errors with msg unpack/pack, other failures outside the
 425                  * operation itself. This is counted as serious errors;
 426                  * 2) errors during fs operation, should be placed in rq_status
 427                  * only
 428                  */
 429                 rc = h->th_act(tsi);
 430                 if (!is_serious(rc) &&
 431                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 432                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 433                                   "pack reply and returned 0 error\n",
 434                                   tgt_name(tsi->tsi_tgt), h->th_name);
 435                         LBUG();
 436                 }
 437                 serious = is_serious(rc);
 438                 rc = clear_serious(rc);
 439         } else {
 440                 serious = 1;
 441         }
 442
 443         req->rq_status = rc;
 444
 445         /*
 446          * ELDLM_* codes which > 0 should be in rq_status only as well as
 447          * all non-serious errors.
 448          */
 449         if (rc > 0 || !serious)
 450                 rc = 0;
 451
 452         LASSERT(current->journal_info == NULL);
 453
 454         if (likely(rc == 0 && req->rq_export))
 455                 target_committed_to_req(req);
 456
 457         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 458         RETURN(0);
 459 }
 460
 461 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 462                                        struct obd_device *obd, int *process)
 463 {
 464         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 465         case MDS_DISCONNECT:
 466         case OST_DISCONNECT:
 467         case OBD_IDX_READ:
 468                 *process = 1;
 469                 RETURN(0);
 470         case MDS_CLOSE:
 471         case MDS_DONE_WRITING:
 472         case MDS_SYNC: /* used in unmounting */
 473         case OBD_PING:
 474         case MDS_REINT:
 475         case OUT_UPDATE:
 476         case SEQ_QUERY:
 477         case FLD_QUERY:
 478         case FLD_READ:
 479         case LDLM_ENQUEUE:
 480         case OST_CREATE:
 481         case OST_DESTROY:
 482         case OST_PUNCH:
 483         case OST_SETATTR:
 484         case OST_SYNC:
 485         case OST_WRITE:
 486                 *process = target_queue_recovery_request(req, obd);
 487                 RETURN(0);
 488
 489         default:
 490                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 491                 *process = -EAGAIN;
 492                 RETURN(0);
 493         }
 494 }
 495
 496 /*
 497  * Handle recovery. Return:
 498  *        +1: continue request processing;
 499  *       -ve: abort immediately with the given error code;
 500  *         0: send reply with error code in req->rq_status;
 501  */
 502 int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 503 {
 504         ENTRY;
 505
 506         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 507         case MDS_CONNECT:
 508         case OST_CONNECT:
 509         case MGS_CONNECT:
 510         case SEC_CTX_INIT:
 511         case SEC_CTX_INIT_CONT:
 512         case SEC_CTX_FINI:
 513                 RETURN(+1);
 514         }
 515
 516         if (!req->rq_export->exp_obd->obd_replayable)
 517                 RETURN(+1);
 518
 519         /* sanity check: if the xid matches, the request must be marked as a
 520          * resent or replayed */
 521         if (req_xid_is_last(req)) {
 522                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 523                       (MSG_RESENT | MSG_REPLAY))) {
 524                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 525                                   "last_xid, expected REPLAY or RESENT flag "
 526                                   "(%x)", req->rq_xid,
 527                                   lustre_msg_get_flags(req->rq_reqmsg));
 528                         req->rq_status = -ENOTCONN;
 529                         RETURN(-ENOTCONN);
 530                 }
 531         }
 532         /* else: note the opposite is not always true; a RESENT req after a
 533          * failover will usually not match the last_xid, since it was likely
 534          * never committed. A REPLAYed request will almost never match the
 535          * last xid, however it could for a committed, but still retained,
 536          * open. */
 537
 538         /* Check for aborted recovery... */
 539         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 540                 int rc;
 541                 int should_process;
 542
 543                 DEBUG_REQ(D_INFO, req, "Got new replay");
 544                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 545                                                  &should_process);
 546                 if (rc != 0 || !should_process)
 547                         RETURN(rc);
 548                 else if (should_process < 0) {
 549                         req->rq_status = should_process;
 550                         rc = ptlrpc_error(req);
 551                         RETURN(rc);
 552                 }
 553         }
 554         RETURN(+1);
 555 }
 556
 557 /* Initial check for request, it is validation mostly */
 558 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 559 {
 560         struct tgt_handler      *h;
 561         struct tgt_opc_slice    *s;
 562         struct lu_target        *tgt;
 563         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 564
 565         ENTRY;
 566
 567         tgt = class_exp2tgt(req->rq_export);
 568
 569         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 570                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 571                         break;
 572
 573         /* opcode was not found in slice */
 574         if (unlikely(s->tos_hs == NULL)) {
 575                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 576                        opc);
 577                 RETURN(ERR_PTR(-ENOTSUPP));
 578         }
 579
 580         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 581         h = s->tos_hs + (opc - s->tos_opc_start);
 582         if (unlikely(h->th_opc == 0)) {
 583                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 584                 RETURN(ERR_PTR(-ENOTSUPP));
 585         }
 586
 587         RETURN(h);
 588 }
 589
 590 int tgt_request_handle(struct ptlrpc_request *req)
 591 {
 592         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 593
 594         struct lustre_msg       *msg = req->rq_reqmsg;
 595         struct tgt_handler      *h;
 596         struct lu_target        *tgt;
 597         int                      request_fail_id = 0;
 598         __u32                    opc = lustre_msg_get_opc(msg);
 599         int                      rc;
 600
 601         ENTRY;
 602
 603         /* Refill the context, to make sure all thread keys are allocated */
 604         lu_env_refill(req->rq_svc_thread->t_env);
 605
 606         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 607         tsi->tsi_pill = &req->rq_pill;
 608         tsi->tsi_env = req->rq_svc_thread->t_env;
 609
 610         /* if request has export then get handlers slice from corresponding
 611          * target, otherwise that should be connect operation */
 612         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 613             opc == MGS_CONNECT) {
 614                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 615                 rc = target_handle_connect(req);
 616                 if (rc != 0) {
 617                         rc = ptlrpc_error(req);
 618                         GOTO(out, rc);
 619                 }
 620                 /* recovery-small test 18c asks to drop connect reply */
 621                 if (unlikely(opc == OST_CONNECT &&
 622                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 623                         GOTO(out, rc = 0);
 624         }
 625
 626         if (unlikely(!class_connected_export(req->rq_export))) {
 627                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 628                        opc, libcfs_id2str(req->rq_peer));
 629                 req->rq_status = -ENOTCONN;
 630                 rc = ptlrpc_error(req);
 631                 GOTO(out, rc);
 632         }
 633
 634         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 635         tsi->tsi_exp = req->rq_export;
 636         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 637                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 638         else
 639                 tsi->tsi_jobid = NULL;
 640
 641         request_fail_id = tgt->lut_request_fail_id;
 642         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 643
 644         h = tgt_handler_find_check(req);
 645         if (IS_ERR(h)) {
 646                 req->rq_status = PTR_ERR(h);
 647                 rc = ptlrpc_error(req);
 648                 GOTO(out, rc);
 649         }
 650
 651         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 652                 GOTO(out, rc = 0);
 653
 654         rc = lustre_msg_check_version(msg, h->th_version);
 655         if (unlikely(rc)) {
 656                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 657                           " %08x, expecting %08x\n", tgt_name(tgt),
 658                           lustre_msg_get_version(msg), h->th_version);
 659                 req->rq_status = -EINVAL;
 660                 rc = ptlrpc_error(req);
 661                 GOTO(out, rc);
 662         }
 663
 664         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 665         if (likely(rc == 1)) {
 666                 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 667                          h->th_opc, opc);
 668                 rc = tgt_handle_request0(tsi, h, req);
 669                 if (rc)
 670                         GOTO(out, rc);
 671         }
 672         EXIT;
 673 out:
 674         req_capsule_fini(tsi->tsi_pill);
 675         if (tsi->tsi_corpus != NULL) {
 676                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 677                 tsi->tsi_corpus = NULL;
 678         }
 679         return rc;
 680 }
 681 EXPORT_SYMBOL(tgt_request_handle);
 682
 683 /** Assign high priority operations to the request if needed. */
 684 int tgt_hpreq_handler(struct ptlrpc_request *req)
 685 {
 686         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 687         struct tgt_handler      *h;
 688         int                      rc;
 689
 690         ENTRY;
 691
 692         if (req->rq_export == NULL)
 693                 RETURN(0);
 694
 695         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 696         tsi->tsi_pill = &req->rq_pill;
 697         tsi->tsi_env = req->rq_svc_thread->t_env;
 698         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 699         tsi->tsi_exp = req->rq_export;
 700
 701         h = tgt_handler_find_check(req);
 702         if (IS_ERR(h)) {
 703                 rc = PTR_ERR(h);
 704                 RETURN(rc);
 705         }
 706
 707         rc = tgt_request_preprocess(tsi, h, req);
 708         if (unlikely(rc != 0))
 709                 RETURN(rc);
 710
 711         if (h->th_hp != NULL)
 712                 h->th_hp(tsi);
 713         RETURN(0);
 714 }
 715 EXPORT_SYMBOL(tgt_hpreq_handler);
 716
 717 void tgt_counter_incr(struct obd_export *exp, int opcode)
 718 {
 719         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 720         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 721                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 722 }
 723 EXPORT_SYMBOL(tgt_counter_incr);
 724
 725 /*
 726  * Unified target generic handlers.
 727  */
 728
 729 /*
 730  * Security functions
 731  */
 732 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 733 {
 734         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 735                                       OBD_CONNECT_RMT_CLIENT_FORCE |
 736                                       OBD_CONNECT_MDS_CAPA |
 737                                       OBD_CONNECT_OSS_CAPA);
 738 }
 739
 740 static int tgt_init_sec_level(struct ptlrpc_request *req)
 741 {
 742         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 743         char                    *client = libcfs_nid2str(req->rq_peer.nid);
 744         struct obd_connect_data *data, *reply;
 745         int                      rc = 0;
 746         bool                     remote;
 747
 748         ENTRY;
 749
 750         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 751         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 752         if (data == NULL || reply == NULL)
 753                 RETURN(-EFAULT);
 754
 755         /* connection from MDT is always trusted */
 756         if (req->rq_auth_usr_mdt) {
 757                 tgt_init_sec_none(reply);
 758                 RETURN(0);
 759         }
 760
 761         /* no GSS support case */
 762         if (!req->rq_auth_gss) {
 763                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 764                         CWARN("client %s -> target %s does not use GSS, "
 765                               "can not run under security level %d.\n",
 766                               client, tgt_name(tgt), tgt->lut_sec_level);
 767                         RETURN(-EACCES);
 768                 } else {
 769                         tgt_init_sec_none(reply);
 770                         RETURN(0);
 771                 }
 772         }
 773
 774         /* old version case */
 775         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 776                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 777                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 778                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 779                         CWARN("client %s -> target %s uses old version, "
 780                               "can not run under security level %d.\n",
 781                               client, tgt_name(tgt), tgt->lut_sec_level);
 782                         RETURN(-EACCES);
 783                 } else {
 784                         CWARN("client %s -> target %s uses old version, "
 785                               "run under security level %d.\n",
 786                               client, tgt_name(tgt), tgt->lut_sec_level);
 787                         tgt_init_sec_none(reply);
 788                         RETURN(0);
 789                 }
 790         }
 791
 792         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 793         if (remote) {
 794                 if (!req->rq_auth_remote)
 795                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 796                                "asked to be remote.\n", client, tgt_name(tgt));
 797         } else if (req->rq_auth_remote) {
 798                 remote = true;
 799                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 800                        "as remote by default.\n", client, tgt_name(tgt));
 801         }
 802
 803         if (remote) {
 804                 if (!tgt->lut_oss_capa) {
 805                         CDEBUG(D_SEC,
 806                                "client %s -> target %s is set as remote,"
 807                                " but OSS capabilities are not enabled: %d.\n",
 808                                client, tgt_name(tgt), tgt->lut_oss_capa);
 809                         RETURN(-EACCES);
 810                 }
 811         } else {
 812                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 813                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 814                                "authenticated!\n", client, tgt_name(tgt));
 815                         RETURN(-EACCES);
 816                 }
 817         }
 818
 819
 820         switch (tgt->lut_sec_level) {
 821         case LUSTRE_SEC_NONE:
 822                 if (remote) {
 823                         CDEBUG(D_SEC,
 824                                "client %s -> target %s is set as remote, "
 825                                "can not run under security level %d.\n",
 826                                client, tgt_name(tgt), tgt->lut_sec_level);
 827                         RETURN(-EACCES);
 828                 }
 829                 tgt_init_sec_none(reply);
 830                 break;
 831         case LUSTRE_SEC_REMOTE:
 832                 if (!remote)
 833                         tgt_init_sec_none(reply);
 834                 break;
 835         case LUSTRE_SEC_ALL:
 836                 if (remote)
 837                         break;
 838                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 839                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 840                 if (!tgt->lut_oss_capa)
 841                         reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 842                 if (!tgt->lut_mds_capa)
 843                         reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 844                 break;
 845         default:
 846                 RETURN(-EINVAL);
 847         }
 848
 849         RETURN(rc);
 850 }
 851
 852 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 853 {
 854         struct lu_target        *tgt = class_exp2tgt(exp);
 855         struct sptlrpc_flavor    flvr;
 856         int                      rc = 0;
 857
 858         LASSERT(tgt);
 859         LASSERT(tgt->lut_obd);
 860         LASSERT(tgt->lut_slice);
 861
 862         /* always allow ECHO client */
 863         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 864                             LUSTRE_ECHO_NAME) == 0)) {
 865                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 866                 return 0;
 867         }
 868
 869         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 870                 read_lock(&tgt->lut_sptlrpc_lock);
 871                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 872                                              req->rq_sp_from,
 873                                              req->rq_peer.nid,
 874                                              &flvr);
 875                 read_unlock(&tgt->lut_sptlrpc_lock);
 876
 877                 spin_lock(&exp->exp_lock);
 878                 exp->exp_sp_peer = req->rq_sp_from;
 879                 exp->exp_flvr = flvr;
 880                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 881                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 882                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 883                                "expect %x\n", tgt_name(tgt),
 884                                req->rq_flvr.sf_rpc,
 885                                libcfs_nid2str(req->rq_peer.nid),
 886                                exp->exp_flvr.sf_rpc);
 887                         rc = -EACCES;
 888                 }
 889                 spin_unlock(&exp->exp_lock);
 890         } else {
 891                 if (exp->exp_sp_peer != req->rq_sp_from) {
 892                         CERROR("%s: RPC source %s doesn't match %s\n",
 893                                tgt_name(tgt),
 894                                sptlrpc_part2name(req->rq_sp_from),
 895                                sptlrpc_part2name(exp->exp_sp_peer));
 896                         rc = -EACCES;
 897                 } else {
 898                         rc = sptlrpc_target_export_check(exp, req);
 899                 }
 900         }
 901
 902         return rc;
 903 }
 904
 905 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 906 {
 907         struct sptlrpc_rule_set  tmp_rset;
 908         int                      rc;
 909
 910         sptlrpc_rule_set_init(&tmp_rset);
 911         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 912         if (rc) {
 913                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 914                        tgt_name(tgt), rc);
 915                 return rc;
 916         }
 917
 918         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 919
 920         write_lock(&tgt->lut_sptlrpc_lock);
 921         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 922         tgt->lut_sptlrpc_rset = tmp_rset;
 923         write_unlock(&tgt->lut_sptlrpc_lock);
 924
 925         return 0;
 926 }
 927 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 928
 929 int tgt_connect(struct tgt_session_info *tsi)
 930 {
 931         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 932         struct obd_connect_data *reply;
 933         int                      rc;
 934
 935         ENTRY;
 936
 937         rc = tgt_init_sec_level(req);
 938         if (rc != 0)
 939                 GOTO(out, rc);
 940
 941         /* XXX: better to call this check right after getting new export but
 942          * before last_rcvd slot allocation to avoid server load upon insecure
 943          * connects. This is to be fixed after unifiyng all targets.
 944          */
 945         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
 946         if (rc)
 947                 GOTO(out, rc);
 948
 949         /* To avoid exposing partially initialized connection flags, changes up
 950          * to this point have been staged in reply->ocd_connect_flags. Now that
 951          * connection handling has completed successfully, atomically update
 952          * the connect flags in the shared export data structure. LU-1623 */
 953         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
 954         spin_lock(&tsi->tsi_exp->exp_lock);
 955         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
 956         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
 957         spin_unlock(&tsi->tsi_exp->exp_lock);
 958
 959         RETURN(0);
 960 out:
 961         obd_disconnect(class_export_get(tsi->tsi_exp));
 962         return rc;
 963 }
 964 EXPORT_SYMBOL(tgt_connect);
 965
 966 int tgt_disconnect(struct tgt_session_info *tsi)
 967 {
 968         int rc;
 969
 970         ENTRY;
 971
 972         rc = target_handle_disconnect(tgt_ses_req(tsi));
 973         if (rc)
 974                 RETURN(err_serious(rc));
 975
 976         RETURN(rc);
 977 }
 978 EXPORT_SYMBOL(tgt_disconnect);
 979
 980 /*
 981  * Unified target OBD handlers
 982  */
 983 int tgt_obd_ping(struct tgt_session_info *tsi)
 984 {
 985         int rc;
 986
 987         ENTRY;
 988
 989         rc = target_handle_ping(tgt_ses_req(tsi));
 990         if (rc)
 991                 RETURN(err_serious(rc));
 992
 993         RETURN(rc);
 994 }
 995 EXPORT_SYMBOL(tgt_obd_ping);
 996
 997 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
 998 {
 999         return err_serious(-EOPNOTSUPP);
1000 }
1001 EXPORT_SYMBOL(tgt_obd_log_cancel);
1002
1003 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
1004 {
1005         return err_serious(-EOPNOTSUPP);
1006 }
1007 EXPORT_SYMBOL(tgt_obd_qc_callback);
1008
1009 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1010 {
1011         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1012         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1013         struct obd_export       *exp = req->rq_export;
1014         struct ptlrpc_bulk_desc *desc;
1015         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1016         int                      tmpcount;
1017         int                      tmpsize;
1018         int                      i;
1019         int                      rc;
1020
1021         ENTRY;
1022
1023         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1024                                     MDS_BULK_PORTAL);
1025         if (desc == NULL)
1026                 RETURN(-ENOMEM);
1027
1028         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1029                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1030                  * which is rdpg->rp_count */
1031                 nob = rdpg->rp_count;
1032
1033         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1034              i++, tmpcount -= tmpsize) {
1035                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1036                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1037         }
1038
1039         LASSERT(desc->bd_nob == nob);
1040         rc = target_bulk_io(exp, desc, lwi);
1041         ptlrpc_free_bulk_pin(desc);
1042         RETURN(rc);
1043 }
1044 EXPORT_SYMBOL(tgt_sendpage);
1045
1046 /*
1047  * OBD_IDX_READ handler
1048  */
1049 int tgt_obd_idx_read(struct tgt_session_info *tsi)
1050 {
1051         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1052         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1053         struct idx_info         *req_ii, *rep_ii;
1054         int                      rc, i;
1055
1056         ENTRY;
1057
1058         memset(rdpg, 0, sizeof(*rdpg));
1059         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1060
1061         /* extract idx_info buffer from request & reply */
1062         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1063         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1064                 RETURN(err_serious(-EPROTO));
1065
1066         rc = req_capsule_server_pack(tsi->tsi_pill);
1067         if (rc)
1068                 RETURN(err_serious(rc));
1069
1070         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1071         if (rep_ii == NULL)
1072                 RETURN(err_serious(-EFAULT));
1073         rep_ii->ii_magic = IDX_INFO_MAGIC;
1074
1075         /* extract hash to start with */
1076         rdpg->rp_hash = req_ii->ii_hash_start;
1077
1078         /* extract requested attributes */
1079         rdpg->rp_attrs = req_ii->ii_attrs;
1080
1081         /* check that fid packed in request is valid and supported */
1082         if (!fid_is_sane(&req_ii->ii_fid))
1083                 RETURN(-EINVAL);
1084         rep_ii->ii_fid = req_ii->ii_fid;
1085
1086         /* copy flags */
1087         rep_ii->ii_flags = req_ii->ii_flags;
1088
1089         /* compute number of pages to allocate, ii_count is the number of 4KB
1090          * containers */
1091         if (req_ii->ii_count <= 0)
1092                 GOTO(out, rc = -EFAULT);
1093         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1094                                exp_max_brw_size(tsi->tsi_exp));
1095         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1096
1097         /* allocate pages to store the containers */
1098         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1099         if (rdpg->rp_pages == NULL)
1100                 GOTO(out, rc = -ENOMEM);
1101         for (i = 0; i < rdpg->rp_npages; i++) {
1102                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1103                 if (rdpg->rp_pages[i] == NULL)
1104                         GOTO(out, rc = -ENOMEM);
1105         }
1106
1107         /* populate pages with key/record pairs */
1108         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1109         if (rc < 0)
1110                 GOTO(out, rc);
1111
1112         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1113                  "asked %d > %d\n", rc, rdpg->rp_count);
1114
1115         /* send pages to client */
1116         rc = tgt_sendpage(tsi, rdpg, rc);
1117         if (rc)
1118                 GOTO(out, rc);
1119         EXIT;
1120 out:
1121         if (rdpg->rp_pages) {
1122                 for (i = 0; i < rdpg->rp_npages; i++)
1123                         if (rdpg->rp_pages[i])
1124                                 __free_page(rdpg->rp_pages[i]);
1125                 OBD_FREE(rdpg->rp_pages,
1126                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1127         }
1128         return rc;
1129 }
1130 EXPORT_SYMBOL(tgt_obd_idx_read);
1131
1132 struct tgt_handler tgt_obd_handlers[] = {
1133 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1134 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1135 TGT_OBD_HDL_VAR(0,      OBD_QC_CALLBACK,        tgt_obd_qc_callback),
1136 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1137 };
1138 EXPORT_SYMBOL(tgt_obd_handlers);
1139
1140 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1141              struct dt_object *obj, __u64 start, __u64 end)
1142 {
1143         int rc = 0;
1144
1145         ENTRY;
1146
1147         /* if no objid is specified, it means "sync whole filesystem" */
1148         if (obj == NULL) {
1149                 rc = dt_sync(env, tgt->lut_bottom);
1150         } else if (dt_version_get(env, obj) >
1151                    tgt->lut_obd->obd_last_committed) {
1152                 rc = dt_object_sync(env, obj, start, end);
1153         }
1154
1155         RETURN(rc);
1156 }
1157 EXPORT_SYMBOL(tgt_sync);
1158 /*
1159  * Unified target DLM handlers.
1160  */
1161
1162 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1163  * (if requested) */
1164 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1165                      void *data, int flag)
1166 {
1167         struct lu_env            env;
1168         struct lu_target        *tgt;
1169         struct dt_object        *obj;
1170         struct lu_fid            fid;
1171         int                      rc = 0;
1172
1173         ENTRY;
1174
1175         tgt = class_exp2tgt(lock->l_export);
1176
1177         if (flag == LDLM_CB_CANCELING &&
1178             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1179             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1180              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1181               lock->l_flags & LDLM_FL_CBPENDING))) {
1182                 __u64 start = 0;
1183                 __u64 end = OBD_OBJECT_EOF;
1184
1185                 rc = lu_env_init(&env, LCT_DT_THREAD);
1186                 if (unlikely(rc != 0))
1187                         RETURN(rc);
1188
1189                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1190                                    tgt->lut_lsd.lsd_osd_index);
1191                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1192                 if (IS_ERR(obj))
1193                         GOTO(err_env, rc = PTR_ERR(obj));
1194
1195                 if (!dt_object_exists(obj))
1196                         GOTO(err_put, rc = -ENOENT);
1197
1198                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1199                         start = lock->l_policy_data.l_extent.start;
1200                         end = lock->l_policy_data.l_extent.end;
1201                 }
1202
1203                 rc = tgt_sync(&env, tgt, obj, start, end);
1204                 if (rc < 0) {
1205                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1206                                "cancel: rc = %d\n",
1207                                tgt_name(tgt), PFID(&fid),
1208                                lock->l_policy_data.l_extent.start,
1209                                lock->l_policy_data.l_extent.end, rc);
1210                 }
1211 err_put:
1212                 lu_object_put(&env, &obj->do_lu);
1213 err_env:
1214                 lu_env_fini(&env);
1215         }
1216
1217         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1218         RETURN(rc);
1219 }
1220
1221 struct ldlm_callback_suite tgt_dlm_cbs = {
1222         .lcs_completion = ldlm_server_completion_ast,
1223         .lcs_blocking   = tgt_blocking_ast,
1224         .lcs_glimpse    = ldlm_server_glimpse_ast
1225 };
1226
1227 int tgt_enqueue(struct tgt_session_info *tsi)
1228 {
1229         struct ptlrpc_request *req = tgt_ses_req(tsi);
1230         int rc;
1231
1232         ENTRY;
1233         /*
1234          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1235          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1236          */
1237         LASSERT(tsi->tsi_dlm_req != NULL);
1238         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1239                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1240         if (rc)
1241                 RETURN(err_serious(rc));
1242
1243         RETURN(req->rq_status);
1244 }
1245 EXPORT_SYMBOL(tgt_enqueue);
1246
1247 int tgt_convert(struct tgt_session_info *tsi)
1248 {
1249         struct ptlrpc_request *req = tgt_ses_req(tsi);
1250         int rc;
1251
1252         ENTRY;
1253         LASSERT(tsi->tsi_dlm_req);
1254         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1255         if (rc)
1256                 RETURN(err_serious(rc));
1257
1258         RETURN(req->rq_status);
1259 }
1260 EXPORT_SYMBOL(tgt_convert);
1261
1262 int tgt_bl_callback(struct tgt_session_info *tsi)
1263 {
1264         return err_serious(-EOPNOTSUPP);
1265 }
1266 EXPORT_SYMBOL(tgt_bl_callback);
1267
1268 int tgt_cp_callback(struct tgt_session_info *tsi)
1269 {
1270         return err_serious(-EOPNOTSUPP);
1271 }
1272 EXPORT_SYMBOL(tgt_cp_callback);
1273
1274 /* generic LDLM target handler */
1275 struct tgt_handler tgt_dlm_handlers[] = {
1276 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1277 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1278 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1279 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1280 };
1281 EXPORT_SYMBOL(tgt_dlm_handlers);
1282
1283 /*
1284  * Unified target LLOG handlers.
1285  */
1286 int tgt_llog_open(struct tgt_session_info *tsi)
1287 {
1288         int rc;
1289
1290         ENTRY;
1291
1292         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1293
1294         RETURN(rc);
1295 }
1296 EXPORT_SYMBOL(tgt_llog_open);
1297
1298 int tgt_llog_close(struct tgt_session_info *tsi)
1299 {
1300         int rc;
1301
1302         ENTRY;
1303
1304         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1305
1306         RETURN(rc);
1307 }
1308 EXPORT_SYMBOL(tgt_llog_close);
1309
1310
1311 int tgt_llog_destroy(struct tgt_session_info *tsi)
1312 {
1313         int rc;
1314
1315         ENTRY;
1316
1317         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1318
1319         RETURN(rc);
1320 }
1321 EXPORT_SYMBOL(tgt_llog_destroy);
1322
1323 int tgt_llog_read_header(struct tgt_session_info *tsi)
1324 {
1325         int rc;
1326
1327         ENTRY;
1328
1329         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1330
1331         RETURN(rc);
1332 }
1333 EXPORT_SYMBOL(tgt_llog_read_header);
1334
1335 int tgt_llog_next_block(struct tgt_session_info *tsi)
1336 {
1337         int rc;
1338
1339         ENTRY;
1340
1341         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1342
1343         RETURN(rc);
1344 }
1345 EXPORT_SYMBOL(tgt_llog_next_block);
1346
1347 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1348 {
1349         int rc;
1350
1351         ENTRY;
1352
1353         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1354
1355         RETURN(rc);
1356 }
1357 EXPORT_SYMBOL(tgt_llog_prev_block);
1358
1359 /* generic llog target handler */
1360 struct tgt_handler tgt_llog_handlers[] = {
1361 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1362 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1363 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1364 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1365 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1366 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1367 };
1368 EXPORT_SYMBOL(tgt_llog_handlers);
1369
1370 /*
1371  * sec context handlers
1372  */
1373 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1374 int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1375 {
1376         return 0;
1377 }
1378
1379 struct tgt_handler tgt_sec_ctx_handlers[] = {
1380 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1381 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1382 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1383 };
1384 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1385
1386 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1387                            struct dt_device *key,
1388                            struct lfsck_request *lr,
1389                            struct thandle *th) = NULL;
1390
1391 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1392                                                 struct dt_device *,
1393                                                 struct lfsck_request *,
1394                                                 struct thandle *))
1395 {
1396         tgt_lfsck_in_notify = notify;
1397 }
1398 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1399
1400 static int (*tgt_lfsck_query)(const struct lu_env *env,
1401                               struct dt_device *key,
1402                               struct lfsck_request *lr) = NULL;
1403
1404 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1405                                            struct dt_device *,
1406                                            struct lfsck_request *))
1407 {
1408         tgt_lfsck_query = query;
1409 }
1410 EXPORT_SYMBOL(tgt_register_lfsck_query);
1411
1412 /* LFSCK request handlers */
1413 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1414 {
1415         const struct lu_env     *env = tsi->tsi_env;
1416         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1417         struct lfsck_request    *lr;
1418         int                      rc;
1419         ENTRY;
1420
1421         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1422         if (lr == NULL)
1423                 RETURN(-EPROTO);
1424
1425         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1426
1427         RETURN(rc);
1428 }
1429
1430 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1431 {
1432         struct lfsck_request    *request;
1433         struct lfsck_reply      *reply;
1434         int                      rc;
1435         ENTRY;
1436
1437         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1438         if (request == NULL)
1439                 RETURN(-EPROTO);
1440
1441         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1442         if (reply == NULL)
1443                 RETURN(-ENOMEM);
1444
1445         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1446         reply->lr_status = rc;
1447
1448         RETURN(rc < 0 ? rc : 0);
1449 }
1450
1451 struct tgt_handler tgt_lfsck_handlers[] = {
1452 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1453 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1454 };
1455 EXPORT_SYMBOL(tgt_lfsck_handlers);
1456
1457 /*
1458  * initialize per-thread page pool (bug 5137).
1459  */
1460 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1461 {
1462         struct tgt_thread_big_cache *tbc;
1463
1464         ENTRY;
1465
1466         LASSERT(thread != NULL);
1467         LASSERT(thread->t_data == NULL);
1468
1469         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1470         if (tbc == NULL)
1471                 RETURN(-ENOMEM);
1472         thread->t_data = tbc;
1473         RETURN(0);
1474 }
1475 EXPORT_SYMBOL(tgt_io_thread_init);
1476
1477 /*
1478  * free per-thread pool created by tgt_thread_init().
1479  */
1480 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1481 {
1482         struct tgt_thread_big_cache *tbc;
1483
1484         ENTRY;
1485
1486         LASSERT(thread != NULL);
1487
1488         /*
1489          * be prepared to handle partially-initialized pools (because this is
1490          * called from ost_io_thread_init() for cleanup.
1491          */
1492         tbc = thread->t_data;
1493         if (tbc != NULL) {
1494                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1495                 thread->t_data = NULL;
1496         }
1497         EXIT;
1498 }
1499 EXPORT_SYMBOL(tgt_io_thread_done);
1500 /**
1501  * Helper function for getting server side [start, start+count] DLM lock
1502  * if asked by client.
1503  */
1504 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1505                     __u64 start, __u64 end, struct lustre_handle *lh,
1506                     int mode, __u64 *flags)
1507 {
1508         ldlm_policy_data_t       policy;
1509         int                      rc;
1510
1511         ENTRY;
1512
1513         LASSERT(lh != NULL);
1514         LASSERT(ns != NULL);
1515         LASSERT(!lustre_handle_is_used(lh));
1516
1517         policy.l_extent.gid = 0;
1518         policy.l_extent.start = start & CFS_PAGE_MASK;
1519
1520         /*
1521          * If ->o_blocks is EOF it means "lock till the end of the file".
1522          * Otherwise, it's size of an extent or hole being punched (in bytes).
1523          */
1524         if (end == OBD_OBJECT_EOF || end < start)
1525                 policy.l_extent.end = OBD_OBJECT_EOF;
1526         else
1527                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1528
1529         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1530                                     flags, ldlm_blocking_ast,
1531                                     ldlm_completion_ast, ldlm_glimpse_ast,
1532                                     NULL, 0, LVB_T_NONE, NULL, lh);
1533         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1534 }
1535 EXPORT_SYMBOL(tgt_extent_lock);
1536
1537 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1538 {
1539         LASSERT(lustre_handle_is_used(lh));
1540         ldlm_lock_decref(lh, mode);
1541 }
1542 EXPORT_SYMBOL(tgt_extent_unlock);
1543
1544 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1545                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1546                  struct lustre_handle *lh, int mode)
1547 {
1548         __u64                    flags = 0;
1549         int                      nrbufs = obj->ioo_bufcnt;
1550         int                      i;
1551
1552         ENTRY;
1553
1554         LASSERT(mode == LCK_PR || mode == LCK_PW);
1555         LASSERT(!lustre_handle_is_used(lh));
1556
1557         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1558                 RETURN(0);
1559
1560         for (i = 1; i < nrbufs; i++)
1561                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1562                         RETURN(-EFAULT);
1563
1564         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1565                                nb[nrbufs - 1].rnb_offset +
1566                                nb[nrbufs - 1].rnb_len - 1,
1567                                lh, mode, &flags));
1568 }
1569 EXPORT_SYMBOL(tgt_brw_lock);
1570
1571 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1572                     struct lustre_handle *lh, int mode)
1573 {
1574         ENTRY;
1575
1576         LASSERT(mode == LCK_PR || mode == LCK_PW);
1577         LASSERT((obj->ioo_bufcnt > 0 &&
1578                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1579                 lustre_handle_is_used(lh));
1580
1581         if (lustre_handle_is_used(lh))
1582                 tgt_extent_unlock(lh, mode);
1583         EXIT;
1584 }
1585 EXPORT_SYMBOL(tgt_brw_unlock);
1586
1587 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1588                                struct ptlrpc_bulk_desc *desc, int opc,
1589                                cksum_type_t cksum_type)
1590 {
1591         struct cfs_crypto_hash_desc     *hdesc;
1592         unsigned int                    bufsize;
1593         int                             i, err;
1594         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1595         __u32                           cksum;
1596
1597         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1598         if (IS_ERR(hdesc)) {
1599                 CERROR("%s: unable to initialize checksum hash %s\n",
1600                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1601                 return PTR_ERR(hdesc);
1602         }
1603
1604         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1605         for (i = 0; i < desc->bd_iov_count; i++) {
1606                 /* corrupt the data before we compute the checksum, to
1607                  * simulate a client->OST data error */
1608                 if (i == 0 && opc == OST_WRITE &&
1609                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1610                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1611                         int len = desc->bd_iov[i].kiov_len;
1612                         struct page *np = tgt_page_to_corrupt;
1613                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1614
1615                         if (np) {
1616                                 char *ptr2 = kmap(np) + off;
1617
1618                                 memcpy(ptr2, ptr, len);
1619                                 memcpy(ptr2, "bad3", min(4, len));
1620                                 kunmap(np);
1621                                 desc->bd_iov[i].kiov_page = np;
1622                         } else {
1623                                 CERROR("%s: can't alloc page for corruption\n",
1624                                        tgt_name(tgt));
1625                         }
1626                 }
1627                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1628                                   desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1629                                   desc->bd_iov[i].kiov_len);
1630
1631                  /* corrupt the data after we compute the checksum, to
1632                  * simulate an OST->client data error */
1633                 if (i == 0 && opc == OST_READ &&
1634                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1635                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1636                         int len = desc->bd_iov[i].kiov_len;
1637                         struct page *np = tgt_page_to_corrupt;
1638                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1639
1640                         if (np) {
1641                                 char *ptr2 = kmap(np) + off;
1642
1643                                 memcpy(ptr2, ptr, len);
1644                                 memcpy(ptr2, "bad4", min(4, len));
1645                                 kunmap(np);
1646                                 desc->bd_iov[i].kiov_page = np;
1647                         } else {
1648                                 CERROR("%s: can't alloc page for corruption\n",
1649                                        tgt_name(tgt));
1650                         }
1651                 }
1652         }
1653
1654         bufsize = sizeof(cksum);
1655         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1656
1657         return cksum;
1658 }
1659
1660 int tgt_brw_read(struct tgt_session_info *tsi)
1661 {
1662         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1663         struct ptlrpc_bulk_desc *desc = NULL;
1664         struct obd_export       *exp = tsi->tsi_exp;
1665         struct niobuf_remote    *remote_nb;
1666         struct niobuf_local     *local_nb;
1667         struct obd_ioobj        *ioo;
1668         struct ost_body         *body, *repbody;
1669         struct l_wait_info       lwi;
1670         struct lustre_handle     lockh = { 0 };
1671         int                      npages, nob = 0, rc, i, no_reply = 0;
1672         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1673
1674         ENTRY;
1675
1676         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1677                 CERROR("%s: deny read request from %s to portal %u\n",
1678                        tgt_name(tsi->tsi_tgt),
1679                        obd_export_nid2str(req->rq_export),
1680                        ptlrpc_req2svc(req)->srv_req_portal);
1681                 RETURN(-EPROTO);
1682         }
1683
1684         req->rq_bulk_read = 1;
1685
1686         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1687                 RETURN(-EIO);
1688
1689         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1690                          cfs_fail_val : (obd_timeout + 1) / 4);
1691
1692         /* Check if there is eviction in progress, and if so, wait for it to
1693          * finish */
1694         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1695                 /* We do not care how long it takes */
1696                 lwi = LWI_INTR(NULL, NULL);
1697                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1698                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1699                          &lwi);
1700         }
1701
1702         /* There must be big cache in current thread to process this request
1703          * if it is NULL then something went wrong and it wasn't allocated,
1704          * report -ENOMEM in that case */
1705         if (tbc == NULL)
1706                 RETURN(-ENOMEM);
1707
1708         body = tsi->tsi_ost_body;
1709         LASSERT(body != NULL);
1710
1711         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1712         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1713
1714         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1715         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1716
1717         local_nb = tbc->local;
1718
1719         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1720                           remote_nb, &lockh, LCK_PR);
1721         if (rc != 0)
1722                 RETURN(rc);
1723
1724         /*
1725          * If getting the lock took more time than
1726          * client was willing to wait, drop it. b=11330
1727          */
1728         if (cfs_time_current_sec() > req->rq_deadline ||
1729             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1730                 no_reply = 1;
1731                 CERROR("Dropping timed-out read from %s because locking"
1732                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1733                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1734                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1735                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1736                 GOTO(out_lock, rc = -ETIMEDOUT);
1737         }
1738
1739         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1740         repbody->oa = body->oa;
1741
1742         npages = PTLRPC_MAX_BRW_PAGES;
1743         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1744                         ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1745         if (rc != 0)
1746                 GOTO(out_lock, rc);
1747
1748         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1749                                     BULK_PUT_SOURCE, OST_BULK_PORTAL);
1750         if (desc == NULL)
1751                 GOTO(out_commitrw, rc = -ENOMEM);
1752
1753         nob = 0;
1754         for (i = 0; i < npages; i++) {
1755                 int page_rc = local_nb[i].lnb_rc;
1756
1757                 if (page_rc < 0) {
1758                         rc = page_rc;
1759                         break;
1760                 }
1761
1762                 nob += page_rc;
1763                 if (page_rc != 0) { /* some data! */
1764                         LASSERT(local_nb[i].lnb_page != NULL);
1765                         ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
1766                                                     local_nb[i].lnb_page_offset,
1767                                                     page_rc);
1768                 }
1769
1770                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1771                         /* All subsequent pages should be 0 */
1772                         while (++i < npages)
1773                                 LASSERT(local_nb[i].lnb_rc == 0);
1774                         break;
1775                 }
1776         }
1777
1778         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1779                 cksum_type_t cksum_type =
1780                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1781                                           body->oa.o_flags : 0);
1782                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1783                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1784                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1785                                                         OST_READ, cksum_type);
1786                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1787                        repbody->oa.o_cksum);
1788         } else {
1789                 repbody->oa.o_valid = 0;
1790         }
1791         /* We're finishing using body->oa as an input variable */
1792
1793         /* Check if client was evicted while we were doing i/o before touching
1794          * network */
1795         if (likely(rc == 0 &&
1796                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1797                 rc = target_bulk_io(exp, desc, &lwi);
1798                 no_reply = rc != 0;
1799         }
1800
1801 out_commitrw:
1802         /* Must commit after prep above in all cases */
1803         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1804                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1805                           NULL, rc);
1806         if (rc == 0)
1807                 tgt_drop_id(exp, &repbody->oa);
1808 out_lock:
1809         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1810
1811         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1812                 ptlrpc_free_bulk_nopin(desc);
1813
1814         LASSERT(rc <= 0);
1815         if (rc == 0) {
1816                 rc = nob;
1817                 ptlrpc_lprocfs_brw(req, nob);
1818         } else if (no_reply) {
1819                 req->rq_no_reply = 1;
1820                 /* reply out callback would free */
1821                 ptlrpc_req_drop_rs(req);
1822                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1823                               "client will retry: rc %d\n",
1824                               exp->exp_obd->obd_name,
1825                               obd_uuid2str(&exp->exp_client_uuid),
1826                               obd_export_nid2str(exp), rc);
1827         }
1828         /* send a bulk after reply to simulate a network delay or reordering
1829          * by a router */
1830         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1831                 wait_queue_head_t        waitq;
1832                 struct l_wait_info       lwi1;
1833
1834                 CDEBUG(D_INFO, "reorder BULK\n");
1835                 init_waitqueue_head(&waitq);
1836
1837                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1838                 l_wait_event(waitq, 0, &lwi1);
1839                 target_bulk_io(exp, desc, &lwi);
1840                 ptlrpc_free_bulk_nopin(desc);
1841         }
1842
1843         RETURN(rc);
1844 }
1845 EXPORT_SYMBOL(tgt_brw_read);
1846
1847 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1848                               struct ptlrpc_bulk_desc *desc,
1849                               struct niobuf_local *local_nb, int npages,
1850                               obd_count client_cksum, obd_count server_cksum,
1851                               bool mmap)
1852 {
1853         struct obd_export *exp = req->rq_export;
1854         struct ost_body *body;
1855         char *router;
1856         char *via;
1857
1858         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1859         LASSERT(body != NULL);
1860
1861         if (req->rq_peer.nid == desc->bd_sender) {
1862                 via = router = "";
1863         } else {
1864                 via = " via ";
1865                 router = libcfs_nid2str(desc->bd_sender);
1866         }
1867
1868         if (mmap) {
1869                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1870                              client_cksum, server_cksum);
1871                 return;
1872         }
1873
1874         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1875                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1876                            "]: client csum %x, server csum %x\n",
1877                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1878                            via, router,
1879                            body->oa.o_valid & OBD_MD_FLFID ?
1880                            body->oa.o_parent_seq : (__u64)0,
1881                            body->oa.o_valid & OBD_MD_FLFID ?
1882                            body->oa.o_parent_oid : 0,
1883                            body->oa.o_valid & OBD_MD_FLFID ?
1884                            body->oa.o_parent_ver : 0,
1885                            POSTID(&body->oa.o_oi),
1886                            local_nb[0].lnb_file_offset,
1887                            local_nb[npages-1].lnb_file_offset +
1888                            local_nb[npages - 1].lnb_len - 1,
1889                            client_cksum, server_cksum);
1890 }
1891
1892 int tgt_brw_write(struct tgt_session_info *tsi)
1893 {
1894         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1895         struct ptlrpc_bulk_desc *desc = NULL;
1896         struct obd_export       *exp = req->rq_export;
1897         struct niobuf_remote    *remote_nb;
1898         struct niobuf_local     *local_nb;
1899         struct obd_ioobj        *ioo;
1900         struct ost_body         *body, *repbody;
1901         struct l_wait_info       lwi;
1902         struct lustre_handle     lockh = {0};
1903         __u32                   *rcs;
1904         int                      objcount, niocount, npages;
1905         int                      rc, i, j;
1906         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1907         bool                     no_reply = false, mmap;
1908         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1909
1910         ENTRY;
1911
1912         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1913                 CERROR("%s: deny write request from %s to portal %u\n",
1914                        tgt_name(tsi->tsi_tgt),
1915                        obd_export_nid2str(req->rq_export),
1916                        ptlrpc_req2svc(req)->srv_req_portal);
1917                 RETURN(err_serious(-EPROTO));
1918         }
1919
1920         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1921                 RETURN(err_serious(-ENOSPC));
1922         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1923                 RETURN(err_serious(-EROFS));
1924
1925         req->rq_bulk_write = 1;
1926
1927         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1928                 RETURN(err_serious(-EIO));
1929         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1930                 RETURN(err_serious(-EFAULT));
1931
1932         /* pause before transaction has been started */
1933         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1934                          cfs_fail_val : (obd_timeout + 1) / 4);
1935
1936         /* There must be big cache in current thread to process this request
1937          * if it is NULL then something went wrong and it wasn't allocated,
1938          * report -ENOMEM in that case */
1939         if (tbc == NULL)
1940                 RETURN(-ENOMEM);
1941
1942         body = tsi->tsi_ost_body;
1943         LASSERT(body != NULL);
1944
1945         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1946         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1947
1948         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1949                                         RCL_CLIENT) / sizeof(*ioo);
1950
1951         for (niocount = i = 0; i < objcount; i++)
1952                 niocount += ioo[i].ioo_bufcnt;
1953
1954         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1955         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1956         if (niocount != req_capsule_get_size(&req->rq_pill,
1957                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1958                         sizeof(*remote_nb))
1959                 RETURN(err_serious(-EPROTO));
1960
1961         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1962             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1963                 memory_pressure_set();
1964
1965         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1966                              niocount * sizeof(*rcs));
1967         rc = req_capsule_server_pack(&req->rq_pill);
1968         if (rc != 0)
1969                 GOTO(out, rc = err_serious(rc));
1970
1971         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1972         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1973
1974         local_nb = tbc->local;
1975
1976         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1977                           remote_nb, &lockh, LCK_PW);
1978         if (rc != 0)
1979                 GOTO(out, rc);
1980
1981         /*
1982          * If getting the lock took more time than
1983          * client was willing to wait, drop it. b=11330
1984          */
1985         if (cfs_time_current_sec() > req->rq_deadline ||
1986             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1987                 no_reply = true;
1988                 CERROR("%s: Dropping timed-out write from %s because locking "
1989                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1990                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1991                        POSTID(&ioo->ioo_oid),
1992                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1993                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1994                 GOTO(out_lock, rc = -ETIMEDOUT);
1995         }
1996
1997         /* Because we already sync grant info with client when reconnect,
1998          * grant info will be cleared for resent req, then fed_grant and
1999          * total_grant will not be modified in following preprw_write */
2000         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2001                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2002                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2003         }
2004
2005         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2006         if (repbody == NULL)
2007                 GOTO(out_lock, rc = -ENOMEM);
2008         repbody->oa = body->oa;
2009
2010         npages = PTLRPC_MAX_BRW_PAGES;
2011         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2012                         objcount, ioo, remote_nb, &npages, local_nb, NULL,
2013                         BYPASS_CAPA);
2014         if (rc < 0)
2015                 GOTO(out_lock, rc);
2016
2017         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2018                                     BULK_GET_SINK, OST_BULK_PORTAL);
2019         if (desc == NULL)
2020                 GOTO(skip_transfer, rc = -ENOMEM);
2021
2022         /* NB Having prepped, we must commit... */
2023         for (i = 0; i < npages; i++)
2024                 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
2025                                             local_nb[i].lnb_page_offset,
2026                                             local_nb[i].lnb_len);
2027
2028         rc = sptlrpc_svc_prep_bulk(req, desc);
2029         if (rc != 0)
2030                 GOTO(skip_transfer, rc);
2031
2032         rc = target_bulk_io(exp, desc, &lwi);
2033         no_reply = rc != 0;
2034
2035 skip_transfer:
2036         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2037                 static int cksum_counter;
2038
2039                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2040                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2041
2042                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2043                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2044                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2045                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2046                                                         OST_WRITE, cksum_type);
2047                 cksum_counter++;
2048
2049                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2050                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2051                                 body->oa.o_flags & OBD_FL_MMAP);
2052
2053                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2054                                           body->oa.o_cksum,
2055                                           repbody->oa.o_cksum, mmap);
2056                         cksum_counter = 0;
2057                 } else if ((cksum_counter & (-cksum_counter)) ==
2058                            cksum_counter) {
2059                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2060                                cksum_counter, libcfs_id2str(req->rq_peer),
2061                                repbody->oa.o_cksum);
2062                 }
2063         }
2064
2065         /* Must commit after prep above in all cases */
2066         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2067                           objcount, ioo, remote_nb, npages, local_nb, NULL,
2068                           rc);
2069         if (rc == -ENOTCONN)
2070                 /* quota acquire process has been given up because
2071                  * either the client has been evicted or the client
2072                  * has timed out the request already */
2073                 no_reply = true;
2074
2075         /*
2076          * Disable sending mtime back to the client. If the client locked the
2077          * whole object, then it has already updated the mtime on its side,
2078          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2079          */
2080         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2081
2082         if (rc == 0) {
2083                 int nob = 0;
2084
2085                 /* set per-requested niobuf return codes */
2086                 for (i = j = 0; i < niocount; i++) {
2087                         int len = remote_nb[i].rnb_len;
2088
2089                         nob += len;
2090                         rcs[i] = 0;
2091                         do {
2092                                 LASSERT(j < npages);
2093                                 if (local_nb[j].lnb_rc < 0)
2094                                         rcs[i] = local_nb[j].lnb_rc;
2095                                 len -= local_nb[j].lnb_len;
2096                                 j++;
2097                         } while (len > 0);
2098                         LASSERT(len == 0);
2099                 }
2100                 LASSERT(j == npages);
2101                 ptlrpc_lprocfs_brw(req, nob);
2102
2103                 tgt_drop_id(exp, &repbody->oa);
2104         }
2105 out_lock:
2106         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2107         if (desc)
2108                 ptlrpc_free_bulk_nopin(desc);
2109 out:
2110         if (no_reply) {
2111                 req->rq_no_reply = 1;
2112                 /* reply out callback would free */
2113                 ptlrpc_req_drop_rs(req);
2114                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2115                               "client will retry: rc %d\n",
2116                               exp->exp_obd->obd_name,
2117                               obd_uuid2str(&exp->exp_client_uuid),
2118                               obd_export_nid2str(exp), rc);
2119         }
2120         memory_pressure_clr();
2121         RETURN(rc);
2122 }
2123 EXPORT_SYMBOL(tgt_brw_write);