lustre/target/tgt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2011, 2014, Intel Corporation.
  25  */
  26 /*
  27  * lustre/target/tgt_handler.c
  28  *
  29  * Lustre Unified Target request handler code
  30  *
  31  * Author: Brian Behlendorf <behlendorf1@llnl.gov>
  32  * Author: Mikhail Pershin <mike.pershin@intel.com>
  33  */
  34
  35 #define DEBUG_SUBSYSTEM S_CLASS
  36
  37 #include <linux/user_namespace.h>
  38 #ifdef HAVE_UIDGID_HEADER
  39 # include <linux/uidgid.h>
  40 #endif
  41
  42 #include <obd.h>
  43 #include <obd_class.h>
  44 #include <obd_cksum.h>
  45 #include <md_object.h>
  46 #include <lustre_lfsck.h>
  47 #include <lustre_nodemap.h>
  48
  49 #include "tgt_internal.h"
  50
  51 char *tgt_name(struct lu_target *tgt)
  52 {
  53         LASSERT(tgt->lut_obd != NULL);
  54         return tgt->lut_obd->obd_name;
  55 }
  56 EXPORT_SYMBOL(tgt_name);
  57
  58 /*
  59  * Generic code handling requests that have struct mdt_body passed in:
  60  *
  61  *  - extract mdt_body from request and save it in @tsi, if present;
  62  *
  63  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  64  *  @tsi;
  65  *
  66  *  - if HABEO_CORPUS flag is set for this request type check whether object
  67  *  actually exists on storage (lu_object_exists()).
  68  *
  69  */
  70 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  71 {
  72         const struct mdt_body   *body;
  73         struct lu_object        *obj;
  74         struct req_capsule      *pill = tsi->tsi_pill;
  75         int                      rc;
  76
  77         ENTRY;
  78
  79         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  80         if (body == NULL)
  81                 RETURN(-EFAULT);
  82
  83         tsi->tsi_mdt_body = body;
  84
  85         if (!(body->mbo_valid & OBD_MD_FLID))
  86                 RETURN(0);
  87
  88         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  89          * in any case in pre-2.5 clients. Fix that here if needed */
  90         if (unlikely(fid_is_zero(&body->mbo_fid1)))
  91                 RETURN(0);
  92
  93         if (!fid_is_sane(&body->mbo_fid1)) {
  94                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  95                        PFID(&body->mbo_fid1));
  96                 RETURN(-EINVAL);
  97         }
  98
  99         obj = lu_object_find(tsi->tsi_env,
 100                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
 101                              &body->mbo_fid1, NULL);
 102         if (!IS_ERR(obj)) {
 103                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
 104                         lu_object_put(tsi->tsi_env, obj);
 105                         rc = -ENOENT;
 106                 } else {
 107                         tsi->tsi_corpus = obj;
 108                         rc = 0;
 109                 }
 110         } else {
 111                 rc = PTR_ERR(obj);
 112         }
 113
 114         tsi->tsi_fid = body->mbo_fid1;
 115
 116         RETURN(rc);
 117 }
 118
 119 /**
 120  * Validate oa from client.
 121  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 122  * req are valid.
 123  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 124  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 125  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 126  *       whether this is oi_fid or real ostid. So it will check
 127  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 128  *    c. Old FID-disable osc will send IDIF.
 129  *    d. new FID-enable osc/osp will send normal FID.
 130  *
 131  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 132  * be used for LAST_ID file, and only being accessed inside OST now.
 133  */
 134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 135 {
 136         struct ost_id   *oi     = &oa->o_oi;
 137         u64              seq    = ostid_seq(oi);
 138         u64              id     = ostid_id(oi);
 139         int              rc;
 140         ENTRY;
 141
 142         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
 143                      fid_seq_is_echo(seq))) {
 144                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 145                  * during create, and we will reset this to 1, since this
 146                  * oi_id is basically useless in the following create process,
 147                  * but oi_id == 0 will make it difficult to tell whether it is
 148                  * real FID or ost_id. */
 149                 oi->oi_fid.f_seq = FID_SEQ_ECHO;
 150                 oi->oi_fid.f_oid = id ?: 1;
 151                 oi->oi_fid.f_ver = 0;
 152         } else {
 153                 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
 154
 155                 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
 156                         GOTO(out, rc = -EPROTO);
 157
 158                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 159                  * all of the requests are required to setup FLGROUP */
 160                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 161                         ostid_set_seq_mdt0(oi);
 162                         oa->o_valid |= OBD_MD_FLGROUP;
 163                         seq = ostid_seq(oi);
 164                 }
 165
 166                 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
 167                                fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
 168                         GOTO(out, rc = -EPROTO);
 169
 170                 rc = ostid_to_fid(&tti->tti_fid1, oi,
 171                                   tsi->tsi_tgt->lut_lsd.lsd_osd_index);
 172                 if (unlikely(rc != 0))
 173                         GOTO(out, rc);
 174
 175                 oi->oi_fid = tti->tti_fid1;
 176         }
 177
 178         RETURN(0);
 179
 180 out:
 181         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 182                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
 183                seq, id, rc);
 184         return rc;
 185 }
 186 EXPORT_SYMBOL(tgt_validate_obdo);
 187
 188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
 189 {
 190         unsigned                 max_brw;
 191         struct niobuf_remote    *rnb;
 192         struct obd_ioobj        *ioo;
 193         int                      obj_count;
 194
 195         ENTRY;
 196
 197         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
 198         if (ioo == NULL)
 199                 RETURN(-EPROTO);
 200
 201         rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
 202         if (rnb == NULL)
 203                 RETURN(-EPROTO);
 204
 205         max_brw = ioobj_max_brw_get(ioo);
 206         if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 207                 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 208                        ": rc = %d\n", tgt_name(tsi->tsi_tgt),
 209                        obd_export_nid2str(tsi->tsi_exp), max_brw,
 210                        POSTID(oi), -EPROTO);
 211                 RETURN(-EPROTO);
 212         }
 213         ioo->ioo_oid = *oi;
 214
 215         obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
 216                                         RCL_CLIENT) / sizeof(*ioo);
 217         if (obj_count == 0) {
 218                 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
 219                 RETURN(-EPROTO);
 220         } else if (obj_count > 1) {
 221                 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
 222                        obj_count);
 223                 RETURN(-EPROTO);
 224         }
 225
 226         if (ioo->ioo_bufcnt == 0) {
 227                 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
 228                 RETURN(-EPROTO);
 229         }
 230
 231         if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
 232                 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
 233                           "bulk has too many pages (%d)",
 234                           ioo->ioo_bufcnt);
 235                 RETURN(-EPROTO);
 236         }
 237
 238         RETURN(0);
 239 }
 240
 241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 242 {
 243         struct ost_body         *body;
 244         struct req_capsule      *pill = tsi->tsi_pill;
 245         struct lu_nodemap       *nodemap;
 246         int                      rc;
 247
 248         ENTRY;
 249
 250         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 251         if (body == NULL)
 252                 RETURN(-EFAULT);
 253
 254         rc = tgt_validate_obdo(tsi, &body->oa);
 255         if (rc)
 256                 RETURN(rc);
 257
 258         nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
 259
 260         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 261                                         NODEMAP_CLIENT_TO_FS,
 262                                         body->oa.o_uid);
 263         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 264                                         NODEMAP_CLIENT_TO_FS,
 265                                         body->oa.o_gid);
 266
 267         tsi->tsi_ost_body = body;
 268         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 269
 270         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 271                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 272                 if (rc < 0)
 273                         RETURN(rc);
 274         }
 275
 276         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 277                 if (flags & HABEO_CORPUS) {
 278                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 279                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 280                                tgt_name(tsi->tsi_tgt));
 281                         RETURN(-EPROTO);
 282                 } else {
 283                         RETURN(0);
 284                 }
 285         }
 286
 287         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 288
 289         /*
 290          * OST doesn't get object in advance for further use to prevent
 291          * situations with nested object_find which is potential deadlock.
 292          */
 293         tsi->tsi_corpus = NULL;
 294         RETURN(rc);
 295 }
 296
 297 /*
 298  * Do necessary preprocessing according to handler ->th_flags.
 299  */
 300 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 301                                   struct tgt_handler *h,
 302                                   struct ptlrpc_request *req)
 303 {
 304         struct req_capsule      *pill = tsi->tsi_pill;
 305         __u32                    flags = h->th_flags;
 306         int                      rc = 0;
 307
 308         ENTRY;
 309
 310         if (tsi->tsi_preprocessed)
 311                 RETURN(0);
 312
 313         LASSERT(h->th_act != NULL);
 314         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 315         LASSERT(current->journal_info == NULL);
 316
 317         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 318                      h->th_fmt != NULL));
 319         if (h->th_fmt != NULL) {
 320                 req_capsule_set(pill, h->th_fmt);
 321                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 322                         rc = tgt_mdt_body_unpack(tsi, flags);
 323                         if (rc < 0)
 324                                 RETURN(rc);
 325                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 326                                                  RCL_CLIENT)) {
 327                         rc = tgt_ost_body_unpack(tsi, flags);
 328                         if (rc < 0)
 329                                 RETURN(rc);
 330                 }
 331         }
 332
 333         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 334                 RETURN(-EROFS);
 335
 336         if (flags & HABEO_CLAVIS) {
 337                 struct ldlm_request *dlm_req;
 338
 339                 LASSERT(h->th_fmt != NULL);
 340
 341                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 342                 if (dlm_req != NULL) {
 343                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 344                                      LDLM_IBITS &&
 345                                      dlm_req->lock_desc.l_policy_data.\
 346                                      l_inodebits.bits == 0)) {
 347                                 /*
 348                                  * Lock without inodebits makes no sense and
 349                                  * will oops later in ldlm. If client miss to
 350                                  * set such bits, do not trigger ASSERTION.
 351                                  *
 352                                  * For liblustre flock case, it maybe zero.
 353                                  */
 354                                 rc = -EPROTO;
 355                         } else {
 356                                 tsi->tsi_dlm_req = dlm_req;
 357                         }
 358                 } else {
 359                         rc = -EFAULT;
 360                 }
 361         }
 362         tsi->tsi_preprocessed = 1;
 363         RETURN(rc);
 364 }
 365
 366 /*
 367  * Invoke handler for this request opc. Also do necessary preprocessing
 368  * (according to handler ->th_flags), and post-processing (setting of
 369  * ->last_{xid,committed}).
 370  */
 371 static int tgt_handle_request0(struct tgt_session_info *tsi,
 372                                struct tgt_handler *h,
 373                                struct ptlrpc_request *req)
 374 {
 375         int      serious = 0;
 376         int      rc;
 377         __u32    opc = lustre_msg_get_opc(req->rq_reqmsg);
 378
 379         ENTRY;
 380
 381
 382         /* When dealing with sec context requests, no export is associated yet,
 383          * because these requests are sent before *_CONNECT requests.
 384          * A NULL req->rq_export means the normal *_common_slice handlers will
 385          * not be called, because there is no reference to the target.
 386          * So deal with them by hand and jump directly to target_send_reply().
 387          */
 388         switch (opc) {
 389         case SEC_CTX_INIT:
 390         case SEC_CTX_INIT_CONT:
 391         case SEC_CTX_FINI:
 392                 GOTO(out, rc = 0);
 393         }
 394
 395         /*
 396          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 397          * to put same checks into handlers like mdt_close(), mdt_reint(),
 398          * etc., without talking to mdt authors first. Checking same thing
 399          * there again is useless and returning 0 error without packing reply
 400          * is buggy! Handlers either pack reply or return error.
 401          *
 402          * We return 0 here and do not send any reply in order to emulate
 403          * network failure. Do not send any reply in case any of NET related
 404          * fail_id has occured.
 405          */
 406         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 407                 RETURN(0);
 408
 409         rc = tgt_request_preprocess(tsi, h, req);
 410         /* pack reply if reply format is fixed */
 411         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 412                 /* Pack reply */
 413                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 414                                           RCL_SERVER))
 415                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 416                                              RCL_SERVER,
 417                                              tsi->tsi_mdt_body->mbo_eadatasize);
 418                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 419                                           RCL_SERVER))
 420                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 421                                              RCL_SERVER, 0);
 422
 423                 rc = req_capsule_server_pack(tsi->tsi_pill);
 424         }
 425
 426         if (likely(rc == 0)) {
 427                 /*
 428                  * Process request, there can be two types of rc:
 429                  * 1) errors with msg unpack/pack, other failures outside the
 430                  * operation itself. This is counted as serious errors;
 431                  * 2) errors during fs operation, should be placed in rq_status
 432                  * only
 433                  */
 434                 rc = h->th_act(tsi);
 435                 if (!is_serious(rc) &&
 436                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 437                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 438                                   "pack reply and returned 0 error\n",
 439                                   tgt_name(tsi->tsi_tgt), h->th_name);
 440                         LBUG();
 441                 }
 442                 serious = is_serious(rc);
 443                 rc = clear_serious(rc);
 444         } else {
 445                 serious = 1;
 446         }
 447
 448         req->rq_status = rc;
 449
 450         /*
 451          * ELDLM_* codes which > 0 should be in rq_status only as well as
 452          * all non-serious errors.
 453          */
 454         if (rc > 0 || !serious)
 455                 rc = 0;
 456
 457         LASSERT(current->journal_info == NULL);
 458
 459         if (likely(rc == 0 && req->rq_export))
 460                 target_committed_to_req(req);
 461
 462 out:
 463         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 464         RETURN(0);
 465 }
 466
 467 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 468                                        struct obd_device *obd, int *process)
 469 {
 470         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 471         case MDS_DISCONNECT:
 472         case OST_DISCONNECT:
 473         case OBD_IDX_READ:
 474                 *process = 1;
 475                 RETURN(0);
 476         case MDS_CLOSE:
 477         case MDS_SYNC: /* used in unmounting */
 478         case OBD_PING:
 479         case MDS_REINT:
 480         case OUT_UPDATE:
 481         case SEQ_QUERY:
 482         case FLD_QUERY:
 483         case FLD_READ:
 484         case LDLM_ENQUEUE:
 485         case OST_CREATE:
 486         case OST_DESTROY:
 487         case OST_PUNCH:
 488         case OST_SETATTR:
 489         case OST_SYNC:
 490         case OST_WRITE:
 491         case MDS_HSM_PROGRESS:
 492         case MDS_HSM_STATE_SET:
 493         case MDS_HSM_REQUEST:
 494                 *process = target_queue_recovery_request(req, obd);
 495                 RETURN(0);
 496
 497         default:
 498                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 499                 *process = -EAGAIN;
 500                 RETURN(0);
 501         }
 502 }
 503
 504 /*
 505  * Handle recovery. Return:
 506  *        +1: continue request processing;
 507  *       -ve: abort immediately with the given error code;
 508  *         0: send reply with error code in req->rq_status;
 509  */
 510 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 511 {
 512         ENTRY;
 513
 514         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 515         case MDS_CONNECT:
 516         case OST_CONNECT:
 517         case MGS_CONNECT:
 518         case SEC_CTX_INIT:
 519         case SEC_CTX_INIT_CONT:
 520         case SEC_CTX_FINI:
 521                 RETURN(+1);
 522         }
 523
 524         if (!req->rq_export->exp_obd->obd_replayable)
 525                 RETURN(+1);
 526
 527         /* sanity check: if the xid matches, the request must be marked as a
 528          * resent or replayed */
 529         if (req_can_reconstruct(req, NULL)) {
 530                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 531                       (MSG_RESENT | MSG_REPLAY))) {
 532                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 533                                   "saved xid, expected REPLAY or RESENT flag "
 534                                   "(%x)", req->rq_xid,
 535                                   lustre_msg_get_flags(req->rq_reqmsg));
 536                         req->rq_status = -ENOTCONN;
 537                         RETURN(-ENOTCONN);
 538                 }
 539         }
 540         /* else: note the opposite is not always true; a RESENT req after a
 541          * failover will usually not match the last_xid, since it was likely
 542          * never committed. A REPLAYed request will almost never match the
 543          * last xid, however it could for a committed, but still retained,
 544          * open. */
 545
 546         /* Check for aborted recovery... */
 547         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 548                 int rc;
 549                 int should_process;
 550
 551                 DEBUG_REQ(D_INFO, req, "Got new replay");
 552                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 553                                                  &should_process);
 554                 if (rc != 0 || !should_process)
 555                         RETURN(rc);
 556                 else if (should_process < 0) {
 557                         req->rq_status = should_process;
 558                         rc = ptlrpc_error(req);
 559                         RETURN(rc);
 560                 }
 561         }
 562         RETURN(+1);
 563 }
 564
 565 /* Initial check for request, it is validation mostly */
 566 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 567 {
 568         struct tgt_handler      *h;
 569         struct tgt_opc_slice    *s;
 570         struct lu_target        *tgt;
 571         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 572
 573         ENTRY;
 574
 575         tgt = class_exp2tgt(req->rq_export);
 576         if (unlikely(tgt == NULL)) {
 577                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 578                           class_exp2obd(req->rq_export)->obd_name);
 579                 RETURN(ERR_PTR(-EINVAL));
 580         }
 581
 582         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 583                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 584                         break;
 585
 586         /* opcode was not found in slice */
 587         if (unlikely(s->tos_hs == NULL)) {
 588                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 589                        opc);
 590                 RETURN(ERR_PTR(-ENOTSUPP));
 591         }
 592
 593         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 594         h = s->tos_hs + (opc - s->tos_opc_start);
 595         if (unlikely(h->th_opc == 0)) {
 596                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 597                 RETURN(ERR_PTR(-ENOTSUPP));
 598         }
 599
 600         RETURN(h);
 601 }
 602
 603 int tgt_request_handle(struct ptlrpc_request *req)
 604 {
 605         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 606
 607         struct lustre_msg       *msg = req->rq_reqmsg;
 608         struct tgt_handler      *h;
 609         struct lu_target        *tgt;
 610         int                      request_fail_id = 0;
 611         __u32                    opc = lustre_msg_get_opc(msg);
 612         int                      rc;
 613
 614         ENTRY;
 615
 616         /* Refill the context, to make sure all thread keys are allocated */
 617         lu_env_refill(req->rq_svc_thread->t_env);
 618
 619         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 620         tsi->tsi_pill = &req->rq_pill;
 621         tsi->tsi_env = req->rq_svc_thread->t_env;
 622
 623         /* if request has export then get handlers slice from corresponding
 624          * target, otherwise that should be connect operation */
 625         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 626             opc == MGS_CONNECT) {
 627                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 628                 rc = target_handle_connect(req);
 629                 if (rc != 0) {
 630                         rc = ptlrpc_error(req);
 631                         GOTO(out, rc);
 632                 }
 633                 /* recovery-small test 18c asks to drop connect reply */
 634                 if (unlikely(opc == OST_CONNECT &&
 635                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 636                         GOTO(out, rc = 0);
 637         }
 638
 639         if (unlikely(!class_connected_export(req->rq_export))) {
 640                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
 641                     opc == SEC_CTX_FINI) {
 642                         /* sec context initialization has to be handled
 643                          * by hand in tgt_handle_request0() */
 644                         tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
 645                         h = NULL;
 646                         GOTO(handle_recov, rc = 0);
 647                 }
 648                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 649                        opc, libcfs_id2str(req->rq_peer));
 650                 req->rq_status = -ENOTCONN;
 651                 rc = ptlrpc_error(req);
 652                 GOTO(out, rc);
 653         }
 654
 655         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 656         tsi->tsi_exp = req->rq_export;
 657         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 658                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 659         else
 660                 tsi->tsi_jobid = NULL;
 661
 662         if (tgt == NULL) {
 663                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 664                           class_exp2obd(req->rq_export)->obd_name);
 665                 req->rq_status = -EINVAL;
 666                 rc = ptlrpc_error(req);
 667                 GOTO(out, rc);
 668         }
 669
 670         /* check request's xid is consistent with export's last_xid */
 671         if (req->rq_export != NULL) {
 672                 __u64 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
 673                 if (last_xid != 0)
 674                         req->rq_export->exp_last_xid = last_xid;
 675                 if (req->rq_xid == 0 ||
 676                     req->rq_xid <= req->rq_export->exp_last_xid) {
 677                         DEBUG_REQ(D_ERROR, req,
 678                                   "Unexpected xid %llx vs. last_xid %llx\n",
 679                                   req->rq_xid, req->rq_export->exp_last_xid);
 680 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 93, 0)
 681                         LBUG();
 682 #endif
 683                         req->rq_status = -EPROTO;
 684                         rc = ptlrpc_error(req);
 685                         GOTO(out, rc);
 686                 }
 687         }
 688
 689         request_fail_id = tgt->lut_request_fail_id;
 690         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 691
 692         /* try to release in-memory reply data */
 693         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 694                 tgt_handle_received_xid(req->rq_export,
 695                                 lustre_msg_get_last_xid(req->rq_reqmsg));
 696                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 697                       (MSG_RESENT | MSG_REPLAY)))
 698                         tgt_handle_tag(req->rq_export,
 699                                        lustre_msg_get_tag(req->rq_reqmsg));
 700         }
 701
 702         h = tgt_handler_find_check(req);
 703         if (IS_ERR(h)) {
 704                 req->rq_status = PTR_ERR(h);
 705                 rc = ptlrpc_error(req);
 706                 GOTO(out, rc);
 707         }
 708
 709         LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 710                  h->th_opc, opc);
 711
 712         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 713                 GOTO(out, rc = 0);
 714
 715         rc = lustre_msg_check_version(msg, h->th_version);
 716         if (unlikely(rc)) {
 717                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 718                           " %08x, expecting %08x\n", tgt_name(tgt),
 719                           lustre_msg_get_version(msg), h->th_version);
 720                 req->rq_status = -EINVAL;
 721                 rc = ptlrpc_error(req);
 722                 GOTO(out, rc);
 723         }
 724
 725 handle_recov:
 726         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 727         if (likely(rc == 1)) {
 728                 rc = tgt_handle_request0(tsi, h, req);
 729                 if (rc)
 730                         GOTO(out, rc);
 731         }
 732         EXIT;
 733 out:
 734         req_capsule_fini(tsi->tsi_pill);
 735         if (tsi->tsi_corpus != NULL) {
 736                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 737                 tsi->tsi_corpus = NULL;
 738         }
 739         return rc;
 740 }
 741 EXPORT_SYMBOL(tgt_request_handle);
 742
 743 /** Assign high priority operations to the request if needed. */
 744 int tgt_hpreq_handler(struct ptlrpc_request *req)
 745 {
 746         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 747         struct tgt_handler      *h;
 748         int                      rc;
 749
 750         ENTRY;
 751
 752         if (req->rq_export == NULL)
 753                 RETURN(0);
 754
 755         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 756         tsi->tsi_pill = &req->rq_pill;
 757         tsi->tsi_env = req->rq_svc_thread->t_env;
 758         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 759         tsi->tsi_exp = req->rq_export;
 760
 761         h = tgt_handler_find_check(req);
 762         if (IS_ERR(h)) {
 763                 rc = PTR_ERR(h);
 764                 RETURN(rc);
 765         }
 766
 767         rc = tgt_request_preprocess(tsi, h, req);
 768         if (unlikely(rc != 0))
 769                 RETURN(rc);
 770
 771         if (h->th_hp != NULL)
 772                 h->th_hp(tsi);
 773         RETURN(0);
 774 }
 775 EXPORT_SYMBOL(tgt_hpreq_handler);
 776
 777 void tgt_counter_incr(struct obd_export *exp, int opcode)
 778 {
 779         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 780         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 781                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 782 }
 783 EXPORT_SYMBOL(tgt_counter_incr);
 784
 785 /*
 786  * Unified target generic handlers.
 787  */
 788
 789 /*
 790  * Security functions
 791  */
 792 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 793 {
 794         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 795                                       OBD_CONNECT_RMT_CLIENT_FORCE);
 796 }
 797
 798 static int tgt_init_sec_level(struct ptlrpc_request *req)
 799 {
 800         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 801         char                    *client;
 802         struct obd_connect_data *data, *reply;
 803         int                      rc = 0;
 804         bool                     remote;
 805         ENTRY;
 806
 807         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 808         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 809         if (data == NULL || reply == NULL)
 810                 RETURN(-EFAULT);
 811
 812         /* connection from MDT is always trusted */
 813         if (req->rq_auth_usr_mdt) {
 814                 tgt_init_sec_none(reply);
 815                 RETURN(0);
 816         }
 817
 818         if (unlikely(tgt == NULL)) {
 819                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 820                           class_exp2obd(req->rq_export)->obd_name);
 821                 RETURN(-EINVAL);
 822         }
 823
 824         client = libcfs_nid2str(req->rq_peer.nid);
 825         /* no GSS support case */
 826         if (!req->rq_auth_gss) {
 827                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 828                         CWARN("client %s -> target %s does not use GSS, "
 829                               "can not run under security level %d.\n",
 830                               client, tgt_name(tgt), tgt->lut_sec_level);
 831                         RETURN(-EACCES);
 832                 } else {
 833                         tgt_init_sec_none(reply);
 834                         RETURN(0);
 835                 }
 836         }
 837
 838         /* old version case */
 839         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 840                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 841                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 842                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 843                         CWARN("client %s -> target %s uses old version, "
 844                               "can not run under security level %d.\n",
 845                               client, tgt_name(tgt), tgt->lut_sec_level);
 846                         RETURN(-EACCES);
 847                 } else {
 848                         CWARN("client %s -> target %s uses old version, "
 849                               "run under security level %d.\n",
 850                               client, tgt_name(tgt), tgt->lut_sec_level);
 851                         tgt_init_sec_none(reply);
 852                         RETURN(0);
 853                 }
 854         }
 855
 856         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 857         if (remote) {
 858                 if (!req->rq_auth_remote)
 859                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 860                                "asked to be remote.\n", client, tgt_name(tgt));
 861         } else if (req->rq_auth_remote) {
 862                 remote = true;
 863                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 864                        "as remote by default.\n", client, tgt_name(tgt));
 865         }
 866
 867         if (remote == 0) {
 868                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 869                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 870                                "authenticated!\n", client, tgt_name(tgt));
 871                         RETURN(-EACCES);
 872                 }
 873         }
 874
 875
 876         switch (tgt->lut_sec_level) {
 877         case LUSTRE_SEC_NONE:
 878                 if (remote) {
 879                         CDEBUG(D_SEC,
 880                                "client %s -> target %s is set as remote, "
 881                                "can not run under security level %d.\n",
 882                                client, tgt_name(tgt), tgt->lut_sec_level);
 883                         RETURN(-EACCES);
 884                 }
 885                 tgt_init_sec_none(reply);
 886                 break;
 887         case LUSTRE_SEC_REMOTE:
 888                 if (!remote)
 889                         tgt_init_sec_none(reply);
 890                 break;
 891         case LUSTRE_SEC_ALL:
 892                 if (remote)
 893                         break;
 894                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 895                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 896                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 897                 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 898                 break;
 899         default:
 900                 RETURN(-EINVAL);
 901         }
 902
 903         RETURN(rc);
 904 }
 905
 906 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 907 {
 908         struct lu_target        *tgt = class_exp2tgt(exp);
 909         struct sptlrpc_flavor    flvr;
 910         int                      rc = 0;
 911
 912         LASSERT(tgt);
 913         LASSERT(tgt->lut_obd);
 914         LASSERT(tgt->lut_slice);
 915
 916         /* always allow ECHO client */
 917         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 918                             LUSTRE_ECHO_NAME) == 0)) {
 919                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 920                 return 0;
 921         }
 922
 923         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 924                 read_lock(&tgt->lut_sptlrpc_lock);
 925                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 926                                              req->rq_sp_from,
 927                                              req->rq_peer.nid,
 928                                              &flvr);
 929                 read_unlock(&tgt->lut_sptlrpc_lock);
 930
 931                 spin_lock(&exp->exp_lock);
 932                 exp->exp_sp_peer = req->rq_sp_from;
 933                 exp->exp_flvr = flvr;
 934                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 935                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 936                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 937                                "expect %x\n", tgt_name(tgt),
 938                                req->rq_flvr.sf_rpc,
 939                                libcfs_nid2str(req->rq_peer.nid),
 940                                exp->exp_flvr.sf_rpc);
 941                         rc = -EACCES;
 942                 }
 943                 spin_unlock(&exp->exp_lock);
 944         } else {
 945                 if (exp->exp_sp_peer != req->rq_sp_from) {
 946                         CERROR("%s: RPC source %s doesn't match %s\n",
 947                                tgt_name(tgt),
 948                                sptlrpc_part2name(req->rq_sp_from),
 949                                sptlrpc_part2name(exp->exp_sp_peer));
 950                         rc = -EACCES;
 951                 } else {
 952                         rc = sptlrpc_target_export_check(exp, req);
 953                 }
 954         }
 955
 956         return rc;
 957 }
 958
 959 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 960 {
 961         struct sptlrpc_rule_set  tmp_rset;
 962         int                      rc;
 963
 964         if (unlikely(tgt == NULL)) {
 965                 CERROR("No target passed");
 966                 return -EINVAL;
 967         }
 968
 969         sptlrpc_rule_set_init(&tmp_rset);
 970         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 971         if (rc) {
 972                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 973                        tgt_name(tgt), rc);
 974                 return rc;
 975         }
 976
 977         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 978
 979         write_lock(&tgt->lut_sptlrpc_lock);
 980         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 981         tgt->lut_sptlrpc_rset = tmp_rset;
 982         write_unlock(&tgt->lut_sptlrpc_lock);
 983
 984         return 0;
 985 }
 986 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 987
 988 int tgt_connect(struct tgt_session_info *tsi)
 989 {
 990         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 991         struct obd_connect_data *reply;
 992         int                      rc;
 993
 994         ENTRY;
 995
 996         rc = tgt_init_sec_level(req);
 997         if (rc != 0)
 998                 GOTO(out, rc);
 999
1000         /* XXX: better to call this check right after getting new export but
1001          * before last_rcvd slot allocation to avoid server load upon insecure
1002          * connects. This is to be fixed after unifiyng all targets.
1003          */
1004         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1005         if (rc)
1006                 GOTO(out, rc);
1007
1008         /* To avoid exposing partially initialized connection flags, changes up
1009          * to this point have been staged in reply->ocd_connect_flags. Now that
1010          * connection handling has completed successfully, atomically update
1011          * the connect flags in the shared export data structure. LU-1623 */
1012         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1013         spin_lock(&tsi->tsi_exp->exp_lock);
1014         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1015         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1016         spin_unlock(&tsi->tsi_exp->exp_lock);
1017
1018         RETURN(0);
1019 out:
1020         obd_disconnect(class_export_get(tsi->tsi_exp));
1021         return rc;
1022 }
1023 EXPORT_SYMBOL(tgt_connect);
1024
1025 int tgt_disconnect(struct tgt_session_info *tsi)
1026 {
1027         int rc;
1028
1029         ENTRY;
1030
1031         rc = target_handle_disconnect(tgt_ses_req(tsi));
1032         if (rc)
1033                 RETURN(err_serious(rc));
1034
1035         RETURN(rc);
1036 }
1037 EXPORT_SYMBOL(tgt_disconnect);
1038
1039 /*
1040  * Unified target OBD handlers
1041  */
1042 int tgt_obd_ping(struct tgt_session_info *tsi)
1043 {
1044         int rc;
1045
1046         ENTRY;
1047
1048         rc = target_handle_ping(tgt_ses_req(tsi));
1049         if (rc)
1050                 RETURN(err_serious(rc));
1051
1052         RETURN(rc);
1053 }
1054 EXPORT_SYMBOL(tgt_obd_ping);
1055
1056 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1057 {
1058         return err_serious(-EOPNOTSUPP);
1059 }
1060
1061 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1062 {
1063         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1064         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1065         struct obd_export       *exp = req->rq_export;
1066         struct ptlrpc_bulk_desc *desc;
1067         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1068         int                      tmpcount;
1069         int                      tmpsize;
1070         int                      i;
1071         int                      rc;
1072
1073         ENTRY;
1074
1075         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1076                                     PTLRPC_BULK_PUT_SOURCE |
1077                                         PTLRPC_BULK_BUF_KIOV,
1078                                     MDS_BULK_PORTAL,
1079                                     &ptlrpc_bulk_kiov_pin_ops);
1080         if (desc == NULL)
1081                 RETURN(-ENOMEM);
1082
1083         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1084                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1085                  * which is rdpg->rp_count */
1086                 nob = rdpg->rp_count;
1087
1088         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1089              i++, tmpcount -= tmpsize) {
1090                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1091                 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1092                                                  tmpsize);
1093         }
1094
1095         LASSERT(desc->bd_nob == nob);
1096         rc = target_bulk_io(exp, desc, lwi);
1097         ptlrpc_free_bulk(desc);
1098         RETURN(rc);
1099 }
1100 EXPORT_SYMBOL(tgt_sendpage);
1101
1102 /*
1103  * OBD_IDX_READ handler
1104  */
1105 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1106 {
1107         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1108         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1109         struct idx_info         *req_ii, *rep_ii;
1110         int                      rc, i;
1111
1112         ENTRY;
1113
1114         memset(rdpg, 0, sizeof(*rdpg));
1115         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1116
1117         /* extract idx_info buffer from request & reply */
1118         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1119         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1120                 RETURN(err_serious(-EPROTO));
1121
1122         rc = req_capsule_server_pack(tsi->tsi_pill);
1123         if (rc)
1124                 RETURN(err_serious(rc));
1125
1126         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1127         if (rep_ii == NULL)
1128                 RETURN(err_serious(-EFAULT));
1129         rep_ii->ii_magic = IDX_INFO_MAGIC;
1130
1131         /* extract hash to start with */
1132         rdpg->rp_hash = req_ii->ii_hash_start;
1133
1134         /* extract requested attributes */
1135         rdpg->rp_attrs = req_ii->ii_attrs;
1136
1137         /* check that fid packed in request is valid and supported */
1138         if (!fid_is_sane(&req_ii->ii_fid))
1139                 RETURN(-EINVAL);
1140         rep_ii->ii_fid = req_ii->ii_fid;
1141
1142         /* copy flags */
1143         rep_ii->ii_flags = req_ii->ii_flags;
1144
1145         /* compute number of pages to allocate, ii_count is the number of 4KB
1146          * containers */
1147         if (req_ii->ii_count <= 0)
1148                 GOTO(out, rc = -EFAULT);
1149         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1150                                exp_max_brw_size(tsi->tsi_exp));
1151         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1152
1153         /* allocate pages to store the containers */
1154         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1155         if (rdpg->rp_pages == NULL)
1156                 GOTO(out, rc = -ENOMEM);
1157         for (i = 0; i < rdpg->rp_npages; i++) {
1158                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1159                 if (rdpg->rp_pages[i] == NULL)
1160                         GOTO(out, rc = -ENOMEM);
1161         }
1162
1163         /* populate pages with key/record pairs */
1164         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1165         if (rc < 0)
1166                 GOTO(out, rc);
1167
1168         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1169                  "asked %d > %d\n", rc, rdpg->rp_count);
1170
1171         /* send pages to client */
1172         rc = tgt_sendpage(tsi, rdpg, rc);
1173         if (rc)
1174                 GOTO(out, rc);
1175         EXIT;
1176 out:
1177         if (rdpg->rp_pages) {
1178                 for (i = 0; i < rdpg->rp_npages; i++)
1179                         if (rdpg->rp_pages[i])
1180                                 __free_page(rdpg->rp_pages[i]);
1181                 OBD_FREE(rdpg->rp_pages,
1182                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1183         }
1184         return rc;
1185 }
1186
1187 struct tgt_handler tgt_obd_handlers[] = {
1188 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1189 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1190 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1191 };
1192 EXPORT_SYMBOL(tgt_obd_handlers);
1193
1194 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1195              struct dt_object *obj, __u64 start, __u64 end)
1196 {
1197         int rc = 0;
1198
1199         ENTRY;
1200
1201         /* if no objid is specified, it means "sync whole filesystem" */
1202         if (obj == NULL) {
1203                 rc = dt_sync(env, tgt->lut_bottom);
1204         } else if (dt_version_get(env, obj) >
1205                    tgt->lut_obd->obd_last_committed) {
1206                 rc = dt_object_sync(env, obj, start, end);
1207         }
1208
1209         RETURN(rc);
1210 }
1211 EXPORT_SYMBOL(tgt_sync);
1212 /*
1213  * Unified target DLM handlers.
1214  */
1215
1216 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1217  * (if requested) */
1218 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1219                             void *data, int flag)
1220 {
1221         struct lu_env            env;
1222         struct lu_target        *tgt;
1223         struct dt_object        *obj;
1224         struct lu_fid            fid;
1225         int                      rc = 0;
1226
1227         ENTRY;
1228
1229         tgt = class_exp2tgt(lock->l_export);
1230
1231         if (unlikely(tgt == NULL)) {
1232                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1233                        class_exp2obd(lock->l_export)->obd_name);
1234                 RETURN(-EINVAL);
1235         }
1236
1237         if (flag == LDLM_CB_CANCELING &&
1238             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1239             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1240              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1241               lock->l_flags & LDLM_FL_CBPENDING))) {
1242                 __u64 start = 0;
1243                 __u64 end = OBD_OBJECT_EOF;
1244
1245                 rc = lu_env_init(&env, LCT_DT_THREAD);
1246                 if (unlikely(rc != 0))
1247                         RETURN(rc);
1248
1249                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1250                                    tgt->lut_lsd.lsd_osd_index);
1251                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1252                 if (IS_ERR(obj))
1253                         GOTO(err_env, rc = PTR_ERR(obj));
1254
1255                 if (!dt_object_exists(obj))
1256                         GOTO(err_put, rc = -ENOENT);
1257
1258                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1259                         start = lock->l_policy_data.l_extent.start;
1260                         end = lock->l_policy_data.l_extent.end;
1261                 }
1262
1263                 rc = tgt_sync(&env, tgt, obj, start, end);
1264                 if (rc < 0) {
1265                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1266                                "cancel: rc = %d\n",
1267                                tgt_name(tgt), PFID(&fid),
1268                                lock->l_policy_data.l_extent.start,
1269                                lock->l_policy_data.l_extent.end, rc);
1270                 }
1271 err_put:
1272                 lu_object_put(&env, &obj->do_lu);
1273 err_env:
1274                 lu_env_fini(&env);
1275         }
1276
1277         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1278         RETURN(rc);
1279 }
1280
1281 static struct ldlm_callback_suite tgt_dlm_cbs = {
1282         .lcs_completion = ldlm_server_completion_ast,
1283         .lcs_blocking   = tgt_blocking_ast,
1284         .lcs_glimpse    = ldlm_server_glimpse_ast
1285 };
1286
1287 int tgt_enqueue(struct tgt_session_info *tsi)
1288 {
1289         struct ptlrpc_request *req = tgt_ses_req(tsi);
1290         int rc;
1291
1292         ENTRY;
1293         /*
1294          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1295          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1296          */
1297         LASSERT(tsi->tsi_dlm_req != NULL);
1298         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1299                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1300         if (rc)
1301                 RETURN(err_serious(rc));
1302
1303         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1304         case LUT_FAIL_MDT:
1305                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1306                 break;
1307         case LUT_FAIL_OST:
1308                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1309                 break;
1310         case LUT_FAIL_MGT:
1311                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1312                 break;
1313         default:
1314                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1315                 break;
1316         }
1317         RETURN(req->rq_status);
1318 }
1319 EXPORT_SYMBOL(tgt_enqueue);
1320
1321 int tgt_convert(struct tgt_session_info *tsi)
1322 {
1323         struct ptlrpc_request *req = tgt_ses_req(tsi);
1324         int rc;
1325
1326         ENTRY;
1327         LASSERT(tsi->tsi_dlm_req);
1328         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1329         if (rc)
1330                 RETURN(err_serious(rc));
1331
1332         RETURN(req->rq_status);
1333 }
1334
1335 int tgt_bl_callback(struct tgt_session_info *tsi)
1336 {
1337         return err_serious(-EOPNOTSUPP);
1338 }
1339
1340 int tgt_cp_callback(struct tgt_session_info *tsi)
1341 {
1342         return err_serious(-EOPNOTSUPP);
1343 }
1344
1345 /* generic LDLM target handler */
1346 struct tgt_handler tgt_dlm_handlers[] = {
1347 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1348 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1349 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1350 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1351 };
1352 EXPORT_SYMBOL(tgt_dlm_handlers);
1353
1354 /*
1355  * Unified target LLOG handlers.
1356  */
1357 int tgt_llog_open(struct tgt_session_info *tsi)
1358 {
1359         int rc;
1360
1361         ENTRY;
1362
1363         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1364
1365         RETURN(rc);
1366 }
1367 EXPORT_SYMBOL(tgt_llog_open);
1368
1369 int tgt_llog_close(struct tgt_session_info *tsi)
1370 {
1371         int rc;
1372
1373         ENTRY;
1374
1375         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1376
1377         RETURN(rc);
1378 }
1379 EXPORT_SYMBOL(tgt_llog_close);
1380
1381
1382 int tgt_llog_destroy(struct tgt_session_info *tsi)
1383 {
1384         int rc;
1385
1386         ENTRY;
1387
1388         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1389
1390         RETURN(rc);
1391 }
1392
1393 int tgt_llog_read_header(struct tgt_session_info *tsi)
1394 {
1395         int rc;
1396
1397         ENTRY;
1398
1399         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1400
1401         RETURN(rc);
1402 }
1403 EXPORT_SYMBOL(tgt_llog_read_header);
1404
1405 int tgt_llog_next_block(struct tgt_session_info *tsi)
1406 {
1407         int rc;
1408
1409         ENTRY;
1410
1411         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1412
1413         RETURN(rc);
1414 }
1415 EXPORT_SYMBOL(tgt_llog_next_block);
1416
1417 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1418 {
1419         int rc;
1420
1421         ENTRY;
1422
1423         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1424
1425         RETURN(rc);
1426 }
1427 EXPORT_SYMBOL(tgt_llog_prev_block);
1428
1429 /* generic llog target handler */
1430 struct tgt_handler tgt_llog_handlers[] = {
1431 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1432 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1433 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1434 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1435 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1436 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1437 };
1438 EXPORT_SYMBOL(tgt_llog_handlers);
1439
1440 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1441                            struct dt_device *key,
1442                            struct lfsck_request *lr,
1443                            struct thandle *th) = NULL;
1444
1445 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1446                                                 struct dt_device *,
1447                                                 struct lfsck_request *,
1448                                                 struct thandle *))
1449 {
1450         tgt_lfsck_in_notify = notify;
1451 }
1452 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1453
1454 static int (*tgt_lfsck_query)(const struct lu_env *env,
1455                               struct dt_device *key,
1456                               struct lfsck_request *lr) = NULL;
1457
1458 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1459                                            struct dt_device *,
1460                                            struct lfsck_request *))
1461 {
1462         tgt_lfsck_query = query;
1463 }
1464 EXPORT_SYMBOL(tgt_register_lfsck_query);
1465
1466 /* LFSCK request handlers */
1467 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1468 {
1469         const struct lu_env     *env = tsi->tsi_env;
1470         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1471         struct lfsck_request    *lr;
1472         int                      rc;
1473         ENTRY;
1474
1475         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1476         if (lr == NULL)
1477                 RETURN(-EPROTO);
1478
1479         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1480
1481         RETURN(rc);
1482 }
1483
1484 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1485 {
1486         struct lfsck_request    *request;
1487         struct lfsck_reply      *reply;
1488         int                      rc;
1489         ENTRY;
1490
1491         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1492         if (request == NULL)
1493                 RETURN(-EPROTO);
1494
1495         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1496         if (reply == NULL)
1497                 RETURN(-ENOMEM);
1498
1499         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1500         reply->lr_status = rc;
1501
1502         RETURN(rc < 0 ? rc : 0);
1503 }
1504
1505 struct tgt_handler tgt_lfsck_handlers[] = {
1506 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1507 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1508 };
1509 EXPORT_SYMBOL(tgt_lfsck_handlers);
1510
1511 /*
1512  * initialize per-thread page pool (bug 5137).
1513  */
1514 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1515 {
1516         struct tgt_thread_big_cache *tbc;
1517
1518         ENTRY;
1519
1520         LASSERT(thread != NULL);
1521         LASSERT(thread->t_data == NULL);
1522
1523         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1524         if (tbc == NULL)
1525                 RETURN(-ENOMEM);
1526         thread->t_data = tbc;
1527         RETURN(0);
1528 }
1529 EXPORT_SYMBOL(tgt_io_thread_init);
1530
1531 /*
1532  * free per-thread pool created by tgt_thread_init().
1533  */
1534 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1535 {
1536         struct tgt_thread_big_cache *tbc;
1537
1538         ENTRY;
1539
1540         LASSERT(thread != NULL);
1541
1542         /*
1543          * be prepared to handle partially-initialized pools (because this is
1544          * called from ost_io_thread_init() for cleanup.
1545          */
1546         tbc = thread->t_data;
1547         if (tbc != NULL) {
1548                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1549                 thread->t_data = NULL;
1550         }
1551         EXIT;
1552 }
1553 EXPORT_SYMBOL(tgt_io_thread_done);
1554 /**
1555  * Helper function for getting server side [start, start+count] DLM lock
1556  * if asked by client.
1557  */
1558 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1559                     __u64 start, __u64 end, struct lustre_handle *lh,
1560                     int mode, __u64 *flags)
1561 {
1562         ldlm_policy_data_t       policy;
1563         int                      rc;
1564
1565         ENTRY;
1566
1567         LASSERT(lh != NULL);
1568         LASSERT(ns != NULL);
1569         LASSERT(!lustre_handle_is_used(lh));
1570
1571         policy.l_extent.gid = 0;
1572         policy.l_extent.start = start & PAGE_MASK;
1573
1574         /*
1575          * If ->o_blocks is EOF it means "lock till the end of the file".
1576          * Otherwise, it's size of an extent or hole being punched (in bytes).
1577          */
1578         if (end == OBD_OBJECT_EOF || end < start)
1579                 policy.l_extent.end = OBD_OBJECT_EOF;
1580         else
1581                 policy.l_extent.end = end | ~PAGE_MASK;
1582
1583         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1584                                     flags, ldlm_blocking_ast,
1585                                     ldlm_completion_ast, ldlm_glimpse_ast,
1586                                     NULL, 0, LVB_T_NONE, NULL, lh);
1587         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1588 }
1589 EXPORT_SYMBOL(tgt_extent_lock);
1590
1591 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1592 {
1593         LASSERT(lustre_handle_is_used(lh));
1594         ldlm_lock_decref(lh, mode);
1595 }
1596 EXPORT_SYMBOL(tgt_extent_unlock);
1597
1598 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1599                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1600                  struct lustre_handle *lh, int mode)
1601 {
1602         __u64                    flags = 0;
1603         int                      nrbufs = obj->ioo_bufcnt;
1604         int                      i;
1605
1606         ENTRY;
1607
1608         LASSERT(mode == LCK_PR || mode == LCK_PW);
1609         LASSERT(!lustre_handle_is_used(lh));
1610
1611         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1612                 RETURN(0);
1613
1614         for (i = 1; i < nrbufs; i++)
1615                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1616                         RETURN(-EFAULT);
1617
1618         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1619                                nb[nrbufs - 1].rnb_offset +
1620                                nb[nrbufs - 1].rnb_len - 1,
1621                                lh, mode, &flags));
1622 }
1623
1624 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1625                     struct lustre_handle *lh, int mode)
1626 {
1627         ENTRY;
1628
1629         LASSERT(mode == LCK_PR || mode == LCK_PW);
1630         LASSERT((obj->ioo_bufcnt > 0 &&
1631                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1632                 lustre_handle_is_used(lh));
1633
1634         if (lustre_handle_is_used(lh))
1635                 tgt_extent_unlock(lh, mode);
1636         EXIT;
1637 }
1638
1639 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1640                                struct ptlrpc_bulk_desc *desc, int opc,
1641                                cksum_type_t cksum_type)
1642 {
1643         struct cfs_crypto_hash_desc     *hdesc;
1644         unsigned int                    bufsize;
1645         int                             i, err;
1646         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1647         __u32                           cksum;
1648
1649         LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1650
1651         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1652         if (IS_ERR(hdesc)) {
1653                 CERROR("%s: unable to initialize checksum hash %s\n",
1654                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1655                 return PTR_ERR(hdesc);
1656         }
1657
1658         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1659         for (i = 0; i < desc->bd_iov_count; i++) {
1660                 /* corrupt the data before we compute the checksum, to
1661                  * simulate a client->OST data error */
1662                 if (i == 0 && opc == OST_WRITE &&
1663                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1664                         int off = BD_GET_KIOV(desc, i).kiov_offset &
1665                                 ~PAGE_MASK;
1666                         int len = BD_GET_KIOV(desc, i).kiov_len;
1667                         struct page *np = tgt_page_to_corrupt;
1668                         char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1669
1670                         if (np) {
1671                                 char *ptr2 = kmap(np) + off;
1672
1673                                 memcpy(ptr2, ptr, len);
1674                                 memcpy(ptr2, "bad3", min(4, len));
1675                                 kunmap(np);
1676                                 BD_GET_KIOV(desc, i).kiov_page = np;
1677                         } else {
1678                                 CERROR("%s: can't alloc page for corruption\n",
1679                                        tgt_name(tgt));
1680                         }
1681                 }
1682                 cfs_crypto_hash_update_page(hdesc,
1683                                   BD_GET_KIOV(desc, i).kiov_page,
1684                                   BD_GET_KIOV(desc, i).kiov_offset &
1685                                         ~PAGE_MASK,
1686                                   BD_GET_KIOV(desc, i).kiov_len);
1687
1688                  /* corrupt the data after we compute the checksum, to
1689                  * simulate an OST->client data error */
1690                 if (i == 0 && opc == OST_READ &&
1691                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1692                         int off = BD_GET_KIOV(desc, i).kiov_offset
1693                           & ~PAGE_MASK;
1694                         int len = BD_GET_KIOV(desc, i).kiov_len;
1695                         struct page *np = tgt_page_to_corrupt;
1696                         char *ptr =
1697                           kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1698
1699                         if (np) {
1700                                 char *ptr2 = kmap(np) + off;
1701
1702                                 memcpy(ptr2, ptr, len);
1703                                 memcpy(ptr2, "bad4", min(4, len));
1704                                 kunmap(np);
1705                                 BD_GET_KIOV(desc, i).kiov_page = np;
1706                         } else {
1707                                 CERROR("%s: can't alloc page for corruption\n",
1708                                        tgt_name(tgt));
1709                         }
1710                 }
1711         }
1712
1713         bufsize = sizeof(cksum);
1714         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1715
1716         return cksum;
1717 }
1718
1719 int tgt_brw_read(struct tgt_session_info *tsi)
1720 {
1721         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1722         struct ptlrpc_bulk_desc *desc = NULL;
1723         struct obd_export       *exp = tsi->tsi_exp;
1724         struct niobuf_remote    *remote_nb;
1725         struct niobuf_local     *local_nb;
1726         struct obd_ioobj        *ioo;
1727         struct ost_body         *body, *repbody;
1728         struct l_wait_info       lwi;
1729         struct lustre_handle     lockh = { 0 };
1730         int                      npages, nob = 0, rc, i, no_reply = 0;
1731         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1732
1733         ENTRY;
1734
1735         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1736                 CERROR("%s: deny read request from %s to portal %u\n",
1737                        tgt_name(tsi->tsi_tgt),
1738                        obd_export_nid2str(req->rq_export),
1739                        ptlrpc_req2svc(req)->srv_req_portal);
1740                 RETURN(-EPROTO);
1741         }
1742
1743         req->rq_bulk_read = 1;
1744
1745         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1746                 RETURN(-EIO);
1747
1748         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1749                          cfs_fail_val : (obd_timeout + 1) / 4);
1750
1751         /* Check if there is eviction in progress, and if so, wait for it to
1752          * finish */
1753         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1754                 /* We do not care how long it takes */
1755                 lwi = LWI_INTR(NULL, NULL);
1756                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1757                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1758                          &lwi);
1759         }
1760
1761         /* There must be big cache in current thread to process this request
1762          * if it is NULL then something went wrong and it wasn't allocated,
1763          * report -ENOMEM in that case */
1764         if (tbc == NULL)
1765                 RETURN(-ENOMEM);
1766
1767         body = tsi->tsi_ost_body;
1768         LASSERT(body != NULL);
1769
1770         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1771         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1772
1773         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1774         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1775
1776         local_nb = tbc->local;
1777
1778         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1779                           remote_nb, &lockh, LCK_PR);
1780         if (rc != 0)
1781                 RETURN(rc);
1782
1783         /*
1784          * If getting the lock took more time than
1785          * client was willing to wait, drop it. b=11330
1786          */
1787         if (cfs_time_current_sec() > req->rq_deadline ||
1788             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1789                 no_reply = 1;
1790                 CERROR("Dropping timed-out read from %s because locking"
1791                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1792                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1793                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1794                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1795                 GOTO(out_lock, rc = -ETIMEDOUT);
1796         }
1797
1798         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1799         repbody->oa = body->oa;
1800
1801         npages = PTLRPC_MAX_BRW_PAGES;
1802         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1803                         ioo, remote_nb, &npages, local_nb);
1804         if (rc != 0)
1805                 GOTO(out_lock, rc);
1806
1807         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1808                                     PTLRPC_BULK_PUT_SOURCE |
1809                                         PTLRPC_BULK_BUF_KIOV,
1810                                     OST_BULK_PORTAL,
1811                                     &ptlrpc_bulk_kiov_nopin_ops);
1812         if (desc == NULL)
1813                 GOTO(out_commitrw, rc = -ENOMEM);
1814
1815         nob = 0;
1816         for (i = 0; i < npages; i++) {
1817                 int page_rc = local_nb[i].lnb_rc;
1818
1819                 if (page_rc < 0) {
1820                         rc = page_rc;
1821                         break;
1822                 }
1823
1824                 nob += page_rc;
1825                 if (page_rc != 0) { /* some data! */
1826                         LASSERT(local_nb[i].lnb_page != NULL);
1827                         desc->bd_frag_ops->add_kiov_frag
1828                           (desc, local_nb[i].lnb_page,
1829                            local_nb[i].lnb_page_offset,
1830                            page_rc);
1831                 }
1832
1833                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1834                         /* All subsequent pages should be 0 */
1835                         while (++i < npages)
1836                                 LASSERT(local_nb[i].lnb_rc == 0);
1837                         break;
1838                 }
1839         }
1840
1841         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1842                 cksum_type_t cksum_type =
1843                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1844                                           body->oa.o_flags : 0);
1845                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1846                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1847                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1848                                                         OST_READ, cksum_type);
1849                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1850                        repbody->oa.o_cksum);
1851         } else {
1852                 repbody->oa.o_valid = 0;
1853         }
1854         /* We're finishing using body->oa as an input variable */
1855
1856         /* Check if client was evicted while we were doing i/o before touching
1857          * network */
1858         if (likely(rc == 0 &&
1859                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1860                 rc = target_bulk_io(exp, desc, &lwi);
1861                 no_reply = rc != 0;
1862         }
1863
1864 out_commitrw:
1865         /* Must commit after prep above in all cases */
1866         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1867                           remote_nb, npages, local_nb, rc);
1868         if (rc == 0)
1869                 tgt_drop_id(exp, &repbody->oa);
1870 out_lock:
1871         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1872
1873         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1874                 ptlrpc_free_bulk(desc);
1875
1876         LASSERT(rc <= 0);
1877         if (rc == 0) {
1878                 rc = nob;
1879                 ptlrpc_lprocfs_brw(req, nob);
1880         } else if (no_reply) {
1881                 req->rq_no_reply = 1;
1882                 /* reply out callback would free */
1883                 ptlrpc_req_drop_rs(req);
1884                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1885                               "client will retry: rc %d\n",
1886                               exp->exp_obd->obd_name,
1887                               obd_uuid2str(&exp->exp_client_uuid),
1888                               obd_export_nid2str(exp), rc);
1889         }
1890         /* send a bulk after reply to simulate a network delay or reordering
1891          * by a router */
1892         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1893                 wait_queue_head_t        waitq;
1894                 struct l_wait_info       lwi1;
1895
1896                 CDEBUG(D_INFO, "reorder BULK\n");
1897                 init_waitqueue_head(&waitq);
1898
1899                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1900                 l_wait_event(waitq, 0, &lwi1);
1901                 target_bulk_io(exp, desc, &lwi);
1902                 ptlrpc_free_bulk(desc);
1903         }
1904
1905         RETURN(rc);
1906 }
1907 EXPORT_SYMBOL(tgt_brw_read);
1908
1909 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1910                               struct ptlrpc_bulk_desc *desc,
1911                               struct niobuf_local *local_nb, int npages,
1912                               u32 client_cksum, u32 server_cksum,
1913                               bool mmap)
1914 {
1915         struct obd_export *exp = req->rq_export;
1916         struct ost_body *body;
1917         char *router = "";
1918         char *via = "";
1919
1920         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1921         LASSERT(body != NULL);
1922
1923         if (req->rq_peer.nid != desc->bd_sender) {
1924                 via = " via ";
1925                 router = libcfs_nid2str(desc->bd_sender);
1926         }
1927
1928         if (mmap) {
1929                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1930                              client_cksum, server_cksum);
1931                 return;
1932         }
1933
1934         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1935                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1936                            "]: client csum %x, server csum %x\n",
1937                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1938                            via, router,
1939                            body->oa.o_valid & OBD_MD_FLFID ?
1940                            body->oa.o_parent_seq : (__u64)0,
1941                            body->oa.o_valid & OBD_MD_FLFID ?
1942                            body->oa.o_parent_oid : 0,
1943                            body->oa.o_valid & OBD_MD_FLFID ?
1944                            body->oa.o_parent_ver : 0,
1945                            POSTID(&body->oa.o_oi),
1946                            local_nb[0].lnb_file_offset,
1947                            local_nb[npages-1].lnb_file_offset +
1948                            local_nb[npages - 1].lnb_len - 1,
1949                            client_cksum, server_cksum);
1950 }
1951
1952 int tgt_brw_write(struct tgt_session_info *tsi)
1953 {
1954         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1955         struct ptlrpc_bulk_desc *desc = NULL;
1956         struct obd_export       *exp = req->rq_export;
1957         struct niobuf_remote    *remote_nb;
1958         struct niobuf_local     *local_nb;
1959         struct obd_ioobj        *ioo;
1960         struct ost_body         *body, *repbody;
1961         struct l_wait_info       lwi;
1962         struct lustre_handle     lockh = {0};
1963         __u32                   *rcs;
1964         int                      objcount, niocount, npages;
1965         int                      rc, i, j;
1966         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1967         bool                     no_reply = false, mmap;
1968         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1969
1970         ENTRY;
1971
1972         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1973                 CERROR("%s: deny write request from %s to portal %u\n",
1974                        tgt_name(tsi->tsi_tgt),
1975                        obd_export_nid2str(req->rq_export),
1976                        ptlrpc_req2svc(req)->srv_req_portal);
1977                 RETURN(err_serious(-EPROTO));
1978         }
1979
1980         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1981                 RETURN(err_serious(-ENOSPC));
1982         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1983                 RETURN(err_serious(-EROFS));
1984
1985         req->rq_bulk_write = 1;
1986
1987         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1988                 RETURN(err_serious(-EIO));
1989         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1990                 RETURN(err_serious(-EFAULT));
1991
1992         /* pause before transaction has been started */
1993         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1994                          cfs_fail_val : (obd_timeout + 1) / 4);
1995
1996         /* There must be big cache in current thread to process this request
1997          * if it is NULL then something went wrong and it wasn't allocated,
1998          * report -ENOMEM in that case */
1999         if (tbc == NULL)
2000                 RETURN(-ENOMEM);
2001
2002         body = tsi->tsi_ost_body;
2003         LASSERT(body != NULL);
2004
2005         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2006         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2007
2008         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2009                                         RCL_CLIENT) / sizeof(*ioo);
2010
2011         for (niocount = i = 0; i < objcount; i++)
2012                 niocount += ioo[i].ioo_bufcnt;
2013
2014         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2015         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2016         if (niocount != req_capsule_get_size(&req->rq_pill,
2017                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2018                         sizeof(*remote_nb))
2019                 RETURN(err_serious(-EPROTO));
2020
2021         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2022             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2023                 memory_pressure_set();
2024
2025         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2026                              niocount * sizeof(*rcs));
2027         rc = req_capsule_server_pack(&req->rq_pill);
2028         if (rc != 0)
2029                 GOTO(out, rc = err_serious(rc));
2030
2031         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2032         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2033
2034         local_nb = tbc->local;
2035
2036         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2037                           remote_nb, &lockh, LCK_PW);
2038         if (rc != 0)
2039                 GOTO(out, rc);
2040
2041         /*
2042          * If getting the lock took more time than
2043          * client was willing to wait, drop it. b=11330
2044          */
2045         if (cfs_time_current_sec() > req->rq_deadline ||
2046             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2047                 no_reply = true;
2048                 CERROR("%s: Dropping timed-out write from %s because locking "
2049                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
2050                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2051                        POSTID(&ioo->ioo_oid),
2052                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2053                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2054                 GOTO(out_lock, rc = -ETIMEDOUT);
2055         }
2056
2057         /* Because we already sync grant info with client when reconnect,
2058          * grant info will be cleared for resent req, then fed_grant and
2059          * total_grant will not be modified in following preprw_write */
2060         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2061                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2062                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2063         }
2064
2065         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2066         if (repbody == NULL)
2067                 GOTO(out_lock, rc = -ENOMEM);
2068         repbody->oa = body->oa;
2069
2070         npages = PTLRPC_MAX_BRW_PAGES;
2071         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2072                         objcount, ioo, remote_nb, &npages, local_nb);
2073         if (rc < 0)
2074                 GOTO(out_lock, rc);
2075
2076         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2077                                     PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2078                                     OST_BULK_PORTAL,
2079                                     &ptlrpc_bulk_kiov_nopin_ops);
2080         if (desc == NULL)
2081                 GOTO(skip_transfer, rc = -ENOMEM);
2082
2083         /* NB Having prepped, we must commit... */
2084         for (i = 0; i < npages; i++)
2085                 desc->bd_frag_ops->add_kiov_frag(desc,
2086                                                  local_nb[i].lnb_page,
2087                                                  local_nb[i].lnb_page_offset,
2088                                                  local_nb[i].lnb_len);
2089
2090         rc = sptlrpc_svc_prep_bulk(req, desc);
2091         if (rc != 0)
2092                 GOTO(skip_transfer, rc);
2093
2094         rc = target_bulk_io(exp, desc, &lwi);
2095         no_reply = rc != 0;
2096
2097 skip_transfer:
2098         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2099                 static int cksum_counter;
2100
2101                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2102                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2103
2104                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2105                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2106                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2107                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2108                                                         OST_WRITE, cksum_type);
2109                 cksum_counter++;
2110
2111                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2112                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2113                                 body->oa.o_flags & OBD_FL_MMAP);
2114
2115                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2116                                           body->oa.o_cksum,
2117                                           repbody->oa.o_cksum, mmap);
2118                         cksum_counter = 0;
2119                 } else if ((cksum_counter & (-cksum_counter)) ==
2120                            cksum_counter) {
2121                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2122                                cksum_counter, libcfs_id2str(req->rq_peer),
2123                                repbody->oa.o_cksum);
2124                 }
2125         }
2126
2127         /* Must commit after prep above in all cases */
2128         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2129                           objcount, ioo, remote_nb, npages, local_nb, rc);
2130         if (rc == -ENOTCONN)
2131                 /* quota acquire process has been given up because
2132                  * either the client has been evicted or the client
2133                  * has timed out the request already */
2134                 no_reply = true;
2135
2136         /*
2137          * Disable sending mtime back to the client. If the client locked the
2138          * whole object, then it has already updated the mtime on its side,
2139          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2140          */
2141         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2142
2143         if (rc == 0) {
2144                 int nob = 0;
2145
2146                 /* set per-requested niobuf return codes */
2147                 for (i = j = 0; i < niocount; i++) {
2148                         int len = remote_nb[i].rnb_len;
2149
2150                         nob += len;
2151                         rcs[i] = 0;
2152                         do {
2153                                 LASSERT(j < npages);
2154                                 if (local_nb[j].lnb_rc < 0)
2155                                         rcs[i] = local_nb[j].lnb_rc;
2156                                 len -= local_nb[j].lnb_len;
2157                                 j++;
2158                         } while (len > 0);
2159                         LASSERT(len == 0);
2160                 }
2161                 LASSERT(j == npages);
2162                 ptlrpc_lprocfs_brw(req, nob);
2163
2164                 tgt_drop_id(exp, &repbody->oa);
2165         }
2166 out_lock:
2167         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2168         if (desc)
2169                 ptlrpc_free_bulk(desc);
2170 out:
2171         if (no_reply) {
2172                 req->rq_no_reply = 1;
2173                 /* reply out callback would free */
2174                 ptlrpc_req_drop_rs(req);
2175                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2176                               "client will retry: rc %d\n",
2177                               exp->exp_obd->obd_name,
2178                               obd_uuid2str(&exp->exp_client_uuid),
2179                               obd_export_nid2str(exp), rc);
2180         }
2181         memory_pressure_clr();
2182         RETURN(rc);
2183 }
2184 EXPORT_SYMBOL(tgt_brw_write);
2185
2186 /* Check if request can be reconstructed from saved reply data
2187  * A copy of the reply data is returned in @trd if the pointer is not NULL
2188  */
2189 bool req_can_reconstruct(struct ptlrpc_request *req,
2190                          struct tg_reply_data *trd)
2191 {
2192         struct tg_export_data *ted = &req->rq_export->exp_target_data;
2193         struct lsd_client_data *lcd = ted->ted_lcd;
2194         bool found;
2195
2196         if (tgt_is_multimodrpcs_client(req->rq_export))
2197                 return tgt_lookup_reply(req, trd);
2198
2199         mutex_lock(&ted->ted_lcd_lock);
2200         found = req->rq_xid == lcd->lcd_last_xid ||
2201                 req->rq_xid == lcd->lcd_last_close_xid;
2202
2203         if (found && trd != NULL) {
2204                 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2205                         trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2206                         trd->trd_reply.lrd_transno =
2207                                                 lcd->lcd_last_close_transno;
2208                         trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2209                 } else {
2210                         trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2211                         trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2212                         trd->trd_reply.lrd_result = lcd->lcd_last_result;
2213                         trd->trd_reply.lrd_data = lcd->lcd_last_data;
2214                         trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2215                         trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2216                         trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2217                         trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2218                 }
2219         }
2220         mutex_unlock(&ted->ted_lcd_lock);
2221
2222         return found;
2223 }
2224 EXPORT_SYMBOL(req_can_reconstruct);
2225