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) 2012, 2015, 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 = nodemap_get_from_exp(tsi->tsi_exp);
 259         if (IS_ERR(nodemap))
 260                 RETURN(PTR_ERR(nodemap));
 261
 262         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 263                                         NODEMAP_CLIENT_TO_FS,
 264                                         body->oa.o_uid);
 265         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 266                                         NODEMAP_CLIENT_TO_FS,
 267                                         body->oa.o_gid);
 268         nodemap_putref(nodemap);
 269
 270         tsi->tsi_ost_body = body;
 271         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 272
 273         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 274                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 275                 if (rc < 0)
 276                         RETURN(rc);
 277         }
 278
 279         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 280                 if (flags & HABEO_CORPUS) {
 281                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 282                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 283                                tgt_name(tsi->tsi_tgt));
 284                         RETURN(-EPROTO);
 285                 } else {
 286                         RETURN(0);
 287                 }
 288         }
 289
 290         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 291
 292         /*
 293          * OST doesn't get object in advance for further use to prevent
 294          * situations with nested object_find which is potential deadlock.
 295          */
 296         tsi->tsi_corpus = NULL;
 297         RETURN(rc);
 298 }
 299
 300 /*
 301  * Do necessary preprocessing according to handler ->th_flags.
 302  */
 303 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 304                                   struct tgt_handler *h,
 305                                   struct ptlrpc_request *req)
 306 {
 307         struct req_capsule      *pill = tsi->tsi_pill;
 308         __u32                    flags = h->th_flags;
 309         int                      rc = 0;
 310
 311         ENTRY;
 312
 313         if (tsi->tsi_preprocessed)
 314                 RETURN(0);
 315
 316         LASSERT(h->th_act != NULL);
 317         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 318         LASSERT(current->journal_info == NULL);
 319
 320         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 321                      h->th_fmt != NULL));
 322         if (h->th_fmt != NULL) {
 323                 req_capsule_set(pill, h->th_fmt);
 324                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 325                         rc = tgt_mdt_body_unpack(tsi, flags);
 326                         if (rc < 0)
 327                                 RETURN(rc);
 328                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 329                                                  RCL_CLIENT)) {
 330                         rc = tgt_ost_body_unpack(tsi, flags);
 331                         if (rc < 0)
 332                                 RETURN(rc);
 333                 }
 334         }
 335
 336         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 337                 RETURN(-EROFS);
 338
 339         if (flags & HABEO_CLAVIS) {
 340                 struct ldlm_request *dlm_req;
 341
 342                 LASSERT(h->th_fmt != NULL);
 343
 344                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 345                 if (dlm_req != NULL) {
 346                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 347                                      LDLM_IBITS &&
 348                                      dlm_req->lock_desc.l_policy_data.\
 349                                      l_inodebits.bits == 0)) {
 350                                 /*
 351                                  * Lock without inodebits makes no sense and
 352                                  * will oops later in ldlm. If client miss to
 353                                  * set such bits, do not trigger ASSERTION.
 354                                  *
 355                                  * For liblustre flock case, it maybe zero.
 356                                  */
 357                                 rc = -EPROTO;
 358                         } else {
 359                                 tsi->tsi_dlm_req = dlm_req;
 360                         }
 361                 } else {
 362                         rc = -EFAULT;
 363                 }
 364         }
 365         tsi->tsi_preprocessed = 1;
 366         RETURN(rc);
 367 }
 368
 369 /*
 370  * Invoke handler for this request opc. Also do necessary preprocessing
 371  * (according to handler ->th_flags), and post-processing (setting of
 372  * ->last_{xid,committed}).
 373  */
 374 static int tgt_handle_request0(struct tgt_session_info *tsi,
 375                                struct tgt_handler *h,
 376                                struct ptlrpc_request *req)
 377 {
 378         int      serious = 0;
 379         int      rc;
 380         __u32    opc = lustre_msg_get_opc(req->rq_reqmsg);
 381
 382         ENTRY;
 383
 384
 385         /* When dealing with sec context requests, no export is associated yet,
 386          * because these requests are sent before *_CONNECT requests.
 387          * A NULL req->rq_export means the normal *_common_slice handlers will
 388          * not be called, because there is no reference to the target.
 389          * So deal with them by hand and jump directly to target_send_reply().
 390          */
 391         switch (opc) {
 392         case SEC_CTX_INIT:
 393         case SEC_CTX_INIT_CONT:
 394         case SEC_CTX_FINI:
 395                 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
 396                 GOTO(out, rc = 0);
 397         }
 398
 399         /*
 400          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 401          * to put same checks into handlers like mdt_close(), mdt_reint(),
 402          * etc., without talking to mdt authors first. Checking same thing
 403          * there again is useless and returning 0 error without packing reply
 404          * is buggy! Handlers either pack reply or return error.
 405          *
 406          * We return 0 here and do not send any reply in order to emulate
 407          * network failure. Do not send any reply in case any of NET related
 408          * fail_id has occured.
 409          */
 410         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 411                 RETURN(0);
 412         if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
 413                      OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
 414                 RETURN(0);
 415
 416         rc = tgt_request_preprocess(tsi, h, req);
 417         /* pack reply if reply format is fixed */
 418         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 419                 /* Pack reply */
 420                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 421                                           RCL_SERVER))
 422                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 423                                              RCL_SERVER,
 424                                              tsi->tsi_mdt_body->mbo_eadatasize);
 425                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 426                                           RCL_SERVER))
 427                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 428                                              RCL_SERVER, 0);
 429
 430                 rc = req_capsule_server_pack(tsi->tsi_pill);
 431         }
 432
 433         if (likely(rc == 0)) {
 434                 /*
 435                  * Process request, there can be two types of rc:
 436                  * 1) errors with msg unpack/pack, other failures outside the
 437                  * operation itself. This is counted as serious errors;
 438                  * 2) errors during fs operation, should be placed in rq_status
 439                  * only
 440                  */
 441                 rc = h->th_act(tsi);
 442                 if (!is_serious(rc) &&
 443                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 444                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 445                                   "pack reply and returned 0 error\n",
 446                                   tgt_name(tsi->tsi_tgt), h->th_name);
 447                         LBUG();
 448                 }
 449                 serious = is_serious(rc);
 450                 rc = clear_serious(rc);
 451         } else {
 452                 serious = 1;
 453         }
 454
 455         req->rq_status = rc;
 456
 457         /*
 458          * ELDLM_* codes which > 0 should be in rq_status only as well as
 459          * all non-serious errors.
 460          */
 461         if (rc > 0 || !serious)
 462                 rc = 0;
 463
 464         LASSERT(current->journal_info == NULL);
 465
 466         if (likely(rc == 0 && req->rq_export))
 467                 target_committed_to_req(req);
 468
 469 out:
 470         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 471         RETURN(0);
 472 }
 473
 474 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 475                                        struct obd_device *obd, int *process)
 476 {
 477         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 478         case MDS_DISCONNECT:
 479         case OST_DISCONNECT:
 480         case OBD_IDX_READ:
 481                 *process = 1;
 482                 RETURN(0);
 483         case MDS_CLOSE:
 484         case MDS_SYNC: /* used in unmounting */
 485         case OBD_PING:
 486         case MDS_REINT:
 487         case OUT_UPDATE:
 488         case SEQ_QUERY:
 489         case FLD_QUERY:
 490         case FLD_READ:
 491         case LDLM_ENQUEUE:
 492         case OST_CREATE:
 493         case OST_DESTROY:
 494         case OST_PUNCH:
 495         case OST_SETATTR:
 496         case OST_SYNC:
 497         case OST_WRITE:
 498         case MDS_HSM_PROGRESS:
 499         case MDS_HSM_STATE_SET:
 500         case MDS_HSM_REQUEST:
 501                 *process = target_queue_recovery_request(req, obd);
 502                 RETURN(0);
 503
 504         default:
 505                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 506                 *process = -EAGAIN;
 507                 RETURN(0);
 508         }
 509 }
 510
 511 /*
 512  * Handle recovery. Return:
 513  *        +1: continue request processing;
 514  *       -ve: abort immediately with the given error code;
 515  *         0: send reply with error code in req->rq_status;
 516  */
 517 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 518 {
 519         ENTRY;
 520
 521         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 522         case MDS_CONNECT:
 523         case OST_CONNECT:
 524         case MGS_CONNECT:
 525         case SEC_CTX_INIT:
 526         case SEC_CTX_INIT_CONT:
 527         case SEC_CTX_FINI:
 528                 RETURN(+1);
 529         }
 530
 531         if (!req->rq_export->exp_obd->obd_replayable)
 532                 RETURN(+1);
 533
 534         /* sanity check: if the xid matches, the request must be marked as a
 535          * resent or replayed */
 536         if (req_can_reconstruct(req, NULL)) {
 537                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 538                       (MSG_RESENT | MSG_REPLAY))) {
 539                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 540                                   "saved xid, expected REPLAY or RESENT flag "
 541                                   "(%x)", req->rq_xid,
 542                                   lustre_msg_get_flags(req->rq_reqmsg));
 543                         req->rq_status = -ENOTCONN;
 544                         RETURN(-ENOTCONN);
 545                 }
 546         }
 547         /* else: note the opposite is not always true; a RESENT req after a
 548          * failover will usually not match the last_xid, since it was likely
 549          * never committed. A REPLAYed request will almost never match the
 550          * last xid, however it could for a committed, but still retained,
 551          * open. */
 552
 553         /* Check for aborted recovery... */
 554         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 555                 int rc;
 556                 int should_process;
 557
 558                 DEBUG_REQ(D_INFO, req, "Got new replay");
 559                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 560                                                  &should_process);
 561                 if (rc != 0 || !should_process)
 562                         RETURN(rc);
 563                 else if (should_process < 0) {
 564                         req->rq_status = should_process;
 565                         rc = ptlrpc_error(req);
 566                         RETURN(rc);
 567                 }
 568         }
 569         RETURN(+1);
 570 }
 571
 572 /* Initial check for request, it is validation mostly */
 573 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 574 {
 575         struct tgt_handler      *h;
 576         struct tgt_opc_slice    *s;
 577         struct lu_target        *tgt;
 578         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 579
 580         ENTRY;
 581
 582         tgt = class_exp2tgt(req->rq_export);
 583         if (unlikely(tgt == NULL)) {
 584                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 585                           class_exp2obd(req->rq_export)->obd_name);
 586                 RETURN(ERR_PTR(-EINVAL));
 587         }
 588
 589         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 590                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 591                         break;
 592
 593         /* opcode was not found in slice */
 594         if (unlikely(s->tos_hs == NULL)) {
 595                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 596                        opc);
 597                 RETURN(ERR_PTR(-ENOTSUPP));
 598         }
 599
 600         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 601         h = s->tos_hs + (opc - s->tos_opc_start);
 602         if (unlikely(h->th_opc == 0)) {
 603                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 604                 RETURN(ERR_PTR(-ENOTSUPP));
 605         }
 606
 607         RETURN(h);
 608 }
 609
 610 static int process_req_last_xid(struct ptlrpc_request *req)
 611 {
 612         __u64   last_xid;
 613         ENTRY;
 614
 615         /* check request's xid is consistent with export's last_xid */
 616         last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
 617         if (last_xid > req->rq_export->exp_last_xid)
 618                 req->rq_export->exp_last_xid = last_xid;
 619
 620         if (req->rq_xid == 0 ||
 621             (req->rq_xid <= req->rq_export->exp_last_xid)) {
 622                 DEBUG_REQ(D_ERROR, req, "Unexpected xid %llx vs. "
 623                           "last_xid %llx\n", req->rq_xid,
 624                           req->rq_export->exp_last_xid);
 625                 /* Some request is allowed to be sent during replay,
 626                  * such as OUT update requests, FLD requests, so it
 627                  * is possible that replay requests has smaller XID
 628                  * than the exp_last_xid.
 629                  *
 630                  * Some non-replay requests may have smaller XID as
 631                  * well:
 632                  *
 633                  * - Client send a no_resend RPC, like statfs;
 634                  * - The RPC timedout (or some other error) on client,
 635                  *   then it's removed from the unreplied list;
 636                  * - Client send some other request to bump the
 637                  *   exp_last_xid on server;
 638                  * - The former RPC got chance to be processed;
 639                  */
 640                 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
 641                         RETURN(-EPROTO);
 642         }
 643
 644         /* try to release in-memory reply data */
 645         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 646                 tgt_handle_received_xid(req->rq_export,
 647                                 lustre_msg_get_last_xid(req->rq_reqmsg));
 648                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 649                       (MSG_RESENT | MSG_REPLAY)))
 650                         tgt_handle_tag(req->rq_export,
 651                                        lustre_msg_get_tag(req->rq_reqmsg));
 652         }
 653         RETURN(0);
 654 }
 655
 656 int tgt_request_handle(struct ptlrpc_request *req)
 657 {
 658         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 659
 660         struct lustre_msg       *msg = req->rq_reqmsg;
 661         struct tgt_handler      *h;
 662         struct lu_target        *tgt;
 663         int                      request_fail_id = 0;
 664         __u32                    opc = lustre_msg_get_opc(msg);
 665         struct obd_device       *obd;
 666         int                      rc;
 667         bool                     is_connect = false;
 668         ENTRY;
 669
 670         /* Refill the context, to make sure all thread keys are allocated */
 671         lu_env_refill(req->rq_svc_thread->t_env);
 672
 673         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 674         tsi->tsi_pill = &req->rq_pill;
 675         tsi->tsi_env = req->rq_svc_thread->t_env;
 676
 677         /* if request has export then get handlers slice from corresponding
 678          * target, otherwise that should be connect operation */
 679         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 680             opc == MGS_CONNECT) {
 681                 is_connect = true;
 682                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 683                 rc = target_handle_connect(req);
 684                 if (rc != 0) {
 685                         rc = ptlrpc_error(req);
 686                         GOTO(out, rc);
 687                 }
 688                 /* recovery-small test 18c asks to drop connect reply */
 689                 if (unlikely(opc == OST_CONNECT &&
 690                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 691                         GOTO(out, rc = 0);
 692         }
 693
 694         if (unlikely(!class_connected_export(req->rq_export))) {
 695                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
 696                     opc == SEC_CTX_FINI) {
 697                         /* sec context initialization has to be handled
 698                          * by hand in tgt_handle_request0() */
 699                         tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
 700                         h = NULL;
 701                         GOTO(handle_recov, rc = 0);
 702                 }
 703                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 704                        opc, libcfs_id2str(req->rq_peer));
 705                 req->rq_status = -ENOTCONN;
 706                 rc = ptlrpc_error(req);
 707                 GOTO(out, rc);
 708         }
 709
 710         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 711         tsi->tsi_exp = req->rq_export;
 712         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 713                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 714         else
 715                 tsi->tsi_jobid = NULL;
 716
 717         if (tgt == NULL) {
 718                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 719                           class_exp2obd(req->rq_export)->obd_name);
 720                 req->rq_status = -EINVAL;
 721                 rc = ptlrpc_error(req);
 722                 GOTO(out, rc);
 723         }
 724
 725         /* Skip last_xid processing for the recovery thread, otherwise, the
 726          * last_xid on same request could be processed twice: first time when
 727          * processing the incoming request, second time when the request is
 728          * being processed by recovery thread. */
 729         obd = class_exp2obd(req->rq_export);
 730         if (is_connect) {
 731                 /* reset the exp_last_xid on each connection. */
 732                 req->rq_export->exp_last_xid = 0;
 733         } else if (obd->obd_recovery_data.trd_processing_task !=
 734                    current_pid()) {
 735                 rc = process_req_last_xid(req);
 736                 if (rc) {
 737                         req->rq_status = rc;
 738                         rc = ptlrpc_error(req);
 739                         GOTO(out, rc);
 740                 }
 741         }
 742
 743         request_fail_id = tgt->lut_request_fail_id;
 744         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 745
 746         h = tgt_handler_find_check(req);
 747         if (IS_ERR(h)) {
 748                 req->rq_status = PTR_ERR(h);
 749                 rc = ptlrpc_error(req);
 750                 GOTO(out, rc);
 751         }
 752
 753         LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 754                  h->th_opc, opc);
 755
 756         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 757                 GOTO(out, rc = 0);
 758
 759         rc = lustre_msg_check_version(msg, h->th_version);
 760         if (unlikely(rc)) {
 761                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 762                           " %08x, expecting %08x\n", tgt_name(tgt),
 763                           lustre_msg_get_version(msg), h->th_version);
 764                 req->rq_status = -EINVAL;
 765                 rc = ptlrpc_error(req);
 766                 GOTO(out, rc);
 767         }
 768
 769 handle_recov:
 770         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 771         if (likely(rc == 1)) {
 772                 rc = tgt_handle_request0(tsi, h, req);
 773                 if (rc)
 774                         GOTO(out, rc);
 775         }
 776         EXIT;
 777 out:
 778         req_capsule_fini(tsi->tsi_pill);
 779         if (tsi->tsi_corpus != NULL) {
 780                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 781                 tsi->tsi_corpus = NULL;
 782         }
 783         return rc;
 784 }
 785 EXPORT_SYMBOL(tgt_request_handle);
 786
 787 /** Assign high priority operations to the request if needed. */
 788 int tgt_hpreq_handler(struct ptlrpc_request *req)
 789 {
 790         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 791         struct tgt_handler      *h;
 792         int                      rc;
 793
 794         ENTRY;
 795
 796         if (req->rq_export == NULL)
 797                 RETURN(0);
 798
 799         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 800         tsi->tsi_pill = &req->rq_pill;
 801         tsi->tsi_env = req->rq_svc_thread->t_env;
 802         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 803         tsi->tsi_exp = req->rq_export;
 804
 805         h = tgt_handler_find_check(req);
 806         if (IS_ERR(h)) {
 807                 rc = PTR_ERR(h);
 808                 RETURN(rc);
 809         }
 810
 811         rc = tgt_request_preprocess(tsi, h, req);
 812         if (unlikely(rc != 0))
 813                 RETURN(rc);
 814
 815         if (h->th_hp != NULL)
 816                 h->th_hp(tsi);
 817         RETURN(0);
 818 }
 819 EXPORT_SYMBOL(tgt_hpreq_handler);
 820
 821 void tgt_counter_incr(struct obd_export *exp, int opcode)
 822 {
 823         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 824         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 825                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 826 }
 827 EXPORT_SYMBOL(tgt_counter_incr);
 828
 829 /*
 830  * Unified target generic handlers.
 831  */
 832
 833 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 834 {
 835         struct lu_target        *tgt = class_exp2tgt(exp);
 836         struct sptlrpc_flavor    flvr;
 837         int                      rc = 0;
 838
 839         LASSERT(tgt);
 840         LASSERT(tgt->lut_obd);
 841         LASSERT(tgt->lut_slice);
 842
 843         /* always allow ECHO client */
 844         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 845                             LUSTRE_ECHO_NAME) == 0)) {
 846                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 847                 return 0;
 848         }
 849
 850         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 851                 read_lock(&tgt->lut_sptlrpc_lock);
 852                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 853                                              req->rq_sp_from,
 854                                              req->rq_peer.nid,
 855                                              &flvr);
 856                 read_unlock(&tgt->lut_sptlrpc_lock);
 857
 858                 spin_lock(&exp->exp_lock);
 859                 exp->exp_sp_peer = req->rq_sp_from;
 860                 exp->exp_flvr = flvr;
 861
 862                 /* when on mgs, if no restriction is set, or if client
 863                  * is loopback, allow any flavor */
 864                 if ((strcmp(exp->exp_obd->obd_type->typ_name,
 865                            LUSTRE_MGS_NAME) == 0) &&
 866                      (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
 867                       LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
 868                       == LOLND))
 869                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 870
 871                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 872                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 873                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 874                                "expect %x\n", tgt_name(tgt),
 875                                req->rq_flvr.sf_rpc,
 876                                libcfs_nid2str(req->rq_peer.nid),
 877                                exp->exp_flvr.sf_rpc);
 878                         rc = -EACCES;
 879                 }
 880                 spin_unlock(&exp->exp_lock);
 881         } else {
 882                 if (exp->exp_sp_peer != req->rq_sp_from) {
 883                         CERROR("%s: RPC source %s doesn't match %s\n",
 884                                tgt_name(tgt),
 885                                sptlrpc_part2name(req->rq_sp_from),
 886                                sptlrpc_part2name(exp->exp_sp_peer));
 887                         rc = -EACCES;
 888                 } else {
 889                         rc = sptlrpc_target_export_check(exp, req);
 890                 }
 891         }
 892
 893         return rc;
 894 }
 895
 896 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 897 {
 898         struct sptlrpc_rule_set  tmp_rset;
 899         int                      rc;
 900
 901         if (unlikely(tgt == NULL)) {
 902                 CERROR("No target passed");
 903                 return -EINVAL;
 904         }
 905
 906         sptlrpc_rule_set_init(&tmp_rset);
 907         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 908         if (rc) {
 909                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 910                        tgt_name(tgt), rc);
 911                 return rc;
 912         }
 913
 914         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 915
 916         write_lock(&tgt->lut_sptlrpc_lock);
 917         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 918         tgt->lut_sptlrpc_rset = tmp_rset;
 919         write_unlock(&tgt->lut_sptlrpc_lock);
 920
 921         return 0;
 922 }
 923 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 924
 925 int tgt_connect(struct tgt_session_info *tsi)
 926 {
 927         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 928         struct obd_connect_data *reply;
 929         int                      rc;
 930
 931         ENTRY;
 932
 933         /* XXX: better to call this check right after getting new export but
 934          * before last_rcvd slot allocation to avoid server load upon insecure
 935          * connects. This is to be fixed after unifiyng all targets.
 936          */
 937         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
 938         if (rc)
 939                 GOTO(out, rc);
 940
 941         /* To avoid exposing partially initialized connection flags, changes up
 942          * to this point have been staged in reply->ocd_connect_flags. Now that
 943          * connection handling has completed successfully, atomically update
 944          * the connect flags in the shared export data structure. LU-1623 */
 945         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
 946         spin_lock(&tsi->tsi_exp->exp_lock);
 947         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
 948         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
 949         spin_unlock(&tsi->tsi_exp->exp_lock);
 950
 951         RETURN(0);
 952 out:
 953         obd_disconnect(class_export_get(tsi->tsi_exp));
 954         return rc;
 955 }
 956 EXPORT_SYMBOL(tgt_connect);
 957
 958 int tgt_disconnect(struct tgt_session_info *tsi)
 959 {
 960         int rc;
 961
 962         ENTRY;
 963
 964         rc = target_handle_disconnect(tgt_ses_req(tsi));
 965         if (rc)
 966                 RETURN(err_serious(rc));
 967
 968         RETURN(rc);
 969 }
 970 EXPORT_SYMBOL(tgt_disconnect);
 971
 972 /*
 973  * Unified target OBD handlers
 974  */
 975 int tgt_obd_ping(struct tgt_session_info *tsi)
 976 {
 977         int rc;
 978
 979         ENTRY;
 980
 981         rc = target_handle_ping(tgt_ses_req(tsi));
 982         if (rc)
 983                 RETURN(err_serious(rc));
 984
 985         RETURN(rc);
 986 }
 987 EXPORT_SYMBOL(tgt_obd_ping);
 988
 989 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
 990 {
 991         return err_serious(-EOPNOTSUPP);
 992 }
 993
 994 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
 995 {
 996         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
 997         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 998         struct obd_export       *exp = req->rq_export;
 999         struct ptlrpc_bulk_desc *desc;
1000         struct l_wait_info      *lwi = &tti->tti_u.update.tti_wait_info;
1001         int                      i;
1002         int                      rc;
1003
1004         ENTRY;
1005
1006         desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
1007                                   PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
1008                                     MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1009         if (desc == NULL)
1010                 RETURN(-ENOMEM);
1011
1012         for (i = 0; i < rdbuf->rb_nbufs; i++)
1013                 desc->bd_frag_ops->add_iov_frag(desc,
1014                                         rdbuf->rb_bufs[i].lb_buf,
1015                                         rdbuf->rb_bufs[i].lb_len);
1016
1017         rc = target_bulk_io(exp, desc, lwi);
1018         ptlrpc_free_bulk(desc);
1019         RETURN(rc);
1020 }
1021 EXPORT_SYMBOL(tgt_send_buffer);
1022
1023 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1024 {
1025         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1026         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1027         struct obd_export       *exp = req->rq_export;
1028         struct ptlrpc_bulk_desc *desc;
1029         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1030         int                      tmpcount;
1031         int                      tmpsize;
1032         int                      i;
1033         int                      rc;
1034
1035         ENTRY;
1036
1037         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1038                                     PTLRPC_BULK_PUT_SOURCE |
1039                                         PTLRPC_BULK_BUF_KIOV,
1040                                     MDS_BULK_PORTAL,
1041                                     &ptlrpc_bulk_kiov_pin_ops);
1042         if (desc == NULL)
1043                 RETURN(-ENOMEM);
1044
1045         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1046                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1047                  * which is rdpg->rp_count */
1048                 nob = rdpg->rp_count;
1049
1050         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1051              i++, tmpcount -= tmpsize) {
1052                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1053                 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1054                                                  tmpsize);
1055         }
1056
1057         LASSERT(desc->bd_nob == nob);
1058         rc = target_bulk_io(exp, desc, lwi);
1059         ptlrpc_free_bulk(desc);
1060         RETURN(rc);
1061 }
1062 EXPORT_SYMBOL(tgt_sendpage);
1063
1064 /*
1065  * OBD_IDX_READ handler
1066  */
1067 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1068 {
1069         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1070         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1071         struct idx_info         *req_ii, *rep_ii;
1072         int                      rc, i;
1073
1074         ENTRY;
1075
1076         memset(rdpg, 0, sizeof(*rdpg));
1077         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1078
1079         /* extract idx_info buffer from request & reply */
1080         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1081         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1082                 RETURN(err_serious(-EPROTO));
1083
1084         rc = req_capsule_server_pack(tsi->tsi_pill);
1085         if (rc)
1086                 RETURN(err_serious(rc));
1087
1088         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1089         if (rep_ii == NULL)
1090                 RETURN(err_serious(-EFAULT));
1091         rep_ii->ii_magic = IDX_INFO_MAGIC;
1092
1093         /* extract hash to start with */
1094         rdpg->rp_hash = req_ii->ii_hash_start;
1095
1096         /* extract requested attributes */
1097         rdpg->rp_attrs = req_ii->ii_attrs;
1098
1099         /* check that fid packed in request is valid and supported */
1100         if (!fid_is_sane(&req_ii->ii_fid))
1101                 RETURN(-EINVAL);
1102         rep_ii->ii_fid = req_ii->ii_fid;
1103
1104         /* copy flags */
1105         rep_ii->ii_flags = req_ii->ii_flags;
1106
1107         /* compute number of pages to allocate, ii_count is the number of 4KB
1108          * containers */
1109         if (req_ii->ii_count <= 0)
1110                 GOTO(out, rc = -EFAULT);
1111         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1112                                exp_max_brw_size(tsi->tsi_exp));
1113         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1114
1115         /* allocate pages to store the containers */
1116         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1117         if (rdpg->rp_pages == NULL)
1118                 GOTO(out, rc = -ENOMEM);
1119         for (i = 0; i < rdpg->rp_npages; i++) {
1120                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1121                 if (rdpg->rp_pages[i] == NULL)
1122                         GOTO(out, rc = -ENOMEM);
1123         }
1124
1125         /* populate pages with key/record pairs */
1126         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1127         if (rc < 0)
1128                 GOTO(out, rc);
1129
1130         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1131                  "asked %d > %d\n", rc, rdpg->rp_count);
1132
1133         /* send pages to client */
1134         rc = tgt_sendpage(tsi, rdpg, rc);
1135         if (rc)
1136                 GOTO(out, rc);
1137         EXIT;
1138 out:
1139         if (rdpg->rp_pages) {
1140                 for (i = 0; i < rdpg->rp_npages; i++)
1141                         if (rdpg->rp_pages[i])
1142                                 __free_page(rdpg->rp_pages[i]);
1143                 OBD_FREE(rdpg->rp_pages,
1144                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1145         }
1146         return rc;
1147 }
1148
1149 struct tgt_handler tgt_obd_handlers[] = {
1150 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1151 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1152 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1153 };
1154 EXPORT_SYMBOL(tgt_obd_handlers);
1155
1156 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1157              struct dt_object *obj, __u64 start, __u64 end)
1158 {
1159         int rc = 0;
1160
1161         ENTRY;
1162
1163         /* if no objid is specified, it means "sync whole filesystem" */
1164         if (obj == NULL) {
1165                 rc = dt_sync(env, tgt->lut_bottom);
1166         } else if (dt_version_get(env, obj) >
1167                    tgt->lut_obd->obd_last_committed) {
1168                 rc = dt_object_sync(env, obj, start, end);
1169         }
1170         atomic_inc(&tgt->lut_sync_count);
1171
1172         RETURN(rc);
1173 }
1174 EXPORT_SYMBOL(tgt_sync);
1175 /*
1176  * Unified target DLM handlers.
1177  */
1178
1179 /**
1180  * Unified target BAST
1181  *
1182  * Ensure data and metadata are synced to disk when lock is canceled if Sync on
1183  * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
1184  * but if it's cross-MDT lock, because remote object version is not set, a
1185  * filesystem sync is needed.
1186  *
1187  * \param lock server side lock
1188  * \param desc lock desc
1189  * \param data ldlm_cb_set_arg
1190  * \param flag  indicates whether this cancelling or blocking callback
1191  * \retval      0 on success
1192  * \retval      negative number on error
1193  */
1194 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1195                             void *data, int flag)
1196 {
1197         struct lu_env            env;
1198         struct lu_target        *tgt;
1199         struct dt_object        *obj = NULL;
1200         struct lu_fid            fid;
1201         int                      rc = 0;
1202
1203         ENTRY;
1204
1205         tgt = class_exp2tgt(lock->l_export);
1206
1207         if (unlikely(tgt == NULL)) {
1208                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1209                        class_exp2obd(lock->l_export)->obd_name);
1210                 RETURN(-EINVAL);
1211         }
1212
1213         if (flag == LDLM_CB_CANCELING &&
1214             (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
1215             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1216              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1217               ldlm_is_cbpending(lock))) &&
1218             ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
1219              lock->l_resource->lr_type == LDLM_EXTENT)) {
1220                 __u64 start = 0;
1221                 __u64 end = OBD_OBJECT_EOF;
1222
1223                 rc = lu_env_init(&env, LCT_DT_THREAD);
1224                 if (unlikely(rc != 0))
1225                         RETURN(rc);
1226
1227                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1228                                    tgt->lut_lsd.lsd_osd_index);
1229
1230                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1231                         obj = dt_locate(&env, tgt->lut_bottom, &fid);
1232                         if (IS_ERR(obj))
1233                                 GOTO(err_env, rc = PTR_ERR(obj));
1234
1235                         if (!dt_object_exists(obj))
1236                                 GOTO(err_put, rc = -ENOENT);
1237
1238                         start = lock->l_policy_data.l_extent.start;
1239                         end = lock->l_policy_data.l_extent.end;
1240                 }
1241
1242                 rc = tgt_sync(&env, tgt, obj, start, end);
1243                 if (rc < 0) {
1244                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1245                                "cancel: rc = %d\n",
1246                                tgt_name(tgt), PFID(&fid),
1247                                lock->l_policy_data.l_extent.start,
1248                                lock->l_policy_data.l_extent.end, rc);
1249                 }
1250 err_put:
1251                 if (obj != NULL)
1252                         lu_object_put(&env, &obj->do_lu);
1253 err_env:
1254                 lu_env_fini(&env);
1255         }
1256
1257         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1258         RETURN(rc);
1259 }
1260
1261 static struct ldlm_callback_suite tgt_dlm_cbs = {
1262         .lcs_completion = ldlm_server_completion_ast,
1263         .lcs_blocking   = tgt_blocking_ast,
1264         .lcs_glimpse    = ldlm_server_glimpse_ast
1265 };
1266
1267 int tgt_enqueue(struct tgt_session_info *tsi)
1268 {
1269         struct ptlrpc_request *req = tgt_ses_req(tsi);
1270         int rc;
1271
1272         ENTRY;
1273         /*
1274          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1275          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1276          */
1277         LASSERT(tsi->tsi_dlm_req != NULL);
1278         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1279                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1280         if (rc)
1281                 RETURN(err_serious(rc));
1282
1283         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1284         case LUT_FAIL_MDT:
1285                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1286                 break;
1287         case LUT_FAIL_OST:
1288                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1289                 break;
1290         case LUT_FAIL_MGT:
1291                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1292                 break;
1293         default:
1294                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1295                 break;
1296         }
1297         RETURN(req->rq_status);
1298 }
1299 EXPORT_SYMBOL(tgt_enqueue);
1300
1301 int tgt_convert(struct tgt_session_info *tsi)
1302 {
1303         struct ptlrpc_request *req = tgt_ses_req(tsi);
1304         int rc;
1305
1306         ENTRY;
1307         LASSERT(tsi->tsi_dlm_req);
1308         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1309         if (rc)
1310                 RETURN(err_serious(rc));
1311
1312         RETURN(req->rq_status);
1313 }
1314
1315 int tgt_bl_callback(struct tgt_session_info *tsi)
1316 {
1317         return err_serious(-EOPNOTSUPP);
1318 }
1319
1320 int tgt_cp_callback(struct tgt_session_info *tsi)
1321 {
1322         return err_serious(-EOPNOTSUPP);
1323 }
1324
1325 /* generic LDLM target handler */
1326 struct tgt_handler tgt_dlm_handlers[] = {
1327 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1328 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1329 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1330 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1331 };
1332 EXPORT_SYMBOL(tgt_dlm_handlers);
1333
1334 /*
1335  * Unified target LLOG handlers.
1336  */
1337 int tgt_llog_open(struct tgt_session_info *tsi)
1338 {
1339         int rc;
1340
1341         ENTRY;
1342
1343         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1344
1345         RETURN(rc);
1346 }
1347 EXPORT_SYMBOL(tgt_llog_open);
1348
1349 int tgt_llog_close(struct tgt_session_info *tsi)
1350 {
1351         int rc;
1352
1353         ENTRY;
1354
1355         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1356
1357         RETURN(rc);
1358 }
1359 EXPORT_SYMBOL(tgt_llog_close);
1360
1361
1362 int tgt_llog_destroy(struct tgt_session_info *tsi)
1363 {
1364         int rc;
1365
1366         ENTRY;
1367
1368         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1369
1370         RETURN(rc);
1371 }
1372
1373 int tgt_llog_read_header(struct tgt_session_info *tsi)
1374 {
1375         int rc;
1376
1377         ENTRY;
1378
1379         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1380
1381         RETURN(rc);
1382 }
1383 EXPORT_SYMBOL(tgt_llog_read_header);
1384
1385 int tgt_llog_next_block(struct tgt_session_info *tsi)
1386 {
1387         int rc;
1388
1389         ENTRY;
1390
1391         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1392
1393         RETURN(rc);
1394 }
1395 EXPORT_SYMBOL(tgt_llog_next_block);
1396
1397 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1398 {
1399         int rc;
1400
1401         ENTRY;
1402
1403         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1404
1405         RETURN(rc);
1406 }
1407 EXPORT_SYMBOL(tgt_llog_prev_block);
1408
1409 /* generic llog target handler */
1410 struct tgt_handler tgt_llog_handlers[] = {
1411 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1412 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1413 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1414 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1415 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1416 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1417 };
1418 EXPORT_SYMBOL(tgt_llog_handlers);
1419
1420 /*
1421  * sec context handlers
1422  */
1423 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1424 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1425 {
1426         return 0;
1427 }
1428
1429 struct tgt_handler tgt_sec_ctx_handlers[] = {
1430 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1431 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1432 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1433 };
1434 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1435
1436 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1437                            struct dt_device *key,
1438                            struct lfsck_request *lr,
1439                            struct thandle *th) = NULL;
1440
1441 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1442                                                 struct dt_device *,
1443                                                 struct lfsck_request *,
1444                                                 struct thandle *))
1445 {
1446         tgt_lfsck_in_notify = notify;
1447 }
1448 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1449
1450 static int (*tgt_lfsck_query)(const struct lu_env *env,
1451                               struct dt_device *key,
1452                               struct lfsck_request *req,
1453                               struct lfsck_reply *rep,
1454                               struct lfsck_query *que) = NULL;
1455
1456 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1457                                            struct dt_device *,
1458                                            struct lfsck_request *,
1459                                            struct lfsck_reply *,
1460                                            struct lfsck_query *))
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,
1500                              request, reply, NULL);
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         union ldlm_policy_data 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, enum ldlm_mode 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, enum ldlm_mode 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, enum ldlm_mode 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         if (OBD_FAIL_CHECK(OBD_FAIL_OST_READ_SIZE) &&
1841             nob != cfs_fail_val)
1842                 rc = -E2BIG;
1843
1844         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1845                 cksum_type_t cksum_type =
1846                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1847                                           body->oa.o_flags : 0);
1848                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1849                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1850                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1851                                                         OST_READ, cksum_type);
1852                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1853                        repbody->oa.o_cksum);
1854         } else {
1855                 repbody->oa.o_valid = 0;
1856         }
1857         /* We're finishing using body->oa as an input variable */
1858
1859         /* Check if client was evicted while we were doing i/o before touching
1860          * network */
1861         if (likely(rc == 0 &&
1862                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2) &&
1863                    !CFS_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_BULK))) {
1864                 rc = target_bulk_io(exp, desc, &lwi);
1865                 no_reply = rc != 0;
1866         }
1867
1868 out_commitrw:
1869         /* Must commit after prep above in all cases */
1870         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1871                           remote_nb, npages, local_nb, rc);
1872 out_lock:
1873         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1874
1875         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1876                 ptlrpc_free_bulk(desc);
1877
1878         LASSERT(rc <= 0);
1879         if (rc == 0) {
1880                 rc = nob;
1881                 ptlrpc_lprocfs_brw(req, nob);
1882         } else if (no_reply) {
1883                 req->rq_no_reply = 1;
1884                 /* reply out callback would free */
1885                 ptlrpc_req_drop_rs(req);
1886                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1887                               "client will retry: rc %d\n",
1888                               exp->exp_obd->obd_name,
1889                               obd_uuid2str(&exp->exp_client_uuid),
1890                               obd_export_nid2str(exp), rc);
1891         }
1892         /* send a bulk after reply to simulate a network delay or reordering
1893          * by a router */
1894         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1895                 wait_queue_head_t        waitq;
1896                 struct l_wait_info       lwi1;
1897
1898                 CDEBUG(D_INFO, "reorder BULK\n");
1899                 init_waitqueue_head(&waitq);
1900
1901                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1902                 l_wait_event(waitq, 0, &lwi1);
1903                 target_bulk_io(exp, desc, &lwi);
1904                 ptlrpc_free_bulk(desc);
1905         }
1906
1907         RETURN(rc);
1908 }
1909 EXPORT_SYMBOL(tgt_brw_read);
1910
1911 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1912                               struct ptlrpc_bulk_desc *desc,
1913                               struct niobuf_local *local_nb, int npages,
1914                               u32 client_cksum, u32 server_cksum,
1915                               bool mmap)
1916 {
1917         struct obd_export *exp = req->rq_export;
1918         struct ost_body *body;
1919         char *router = "";
1920         char *via = "";
1921
1922         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1923         LASSERT(body != NULL);
1924
1925         if (req->rq_peer.nid != desc->bd_sender) {
1926                 via = " via ";
1927                 router = libcfs_nid2str(desc->bd_sender);
1928         }
1929
1930         if (mmap) {
1931                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1932                              client_cksum, server_cksum);
1933                 return;
1934         }
1935
1936         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1937                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1938                            "]: client csum %x, server csum %x\n",
1939                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1940                            via, router,
1941                            body->oa.o_valid & OBD_MD_FLFID ?
1942                            body->oa.o_parent_seq : (__u64)0,
1943                            body->oa.o_valid & OBD_MD_FLFID ?
1944                            body->oa.o_parent_oid : 0,
1945                            body->oa.o_valid & OBD_MD_FLFID ?
1946                            body->oa.o_parent_ver : 0,
1947                            POSTID(&body->oa.o_oi),
1948                            local_nb[0].lnb_file_offset,
1949                            local_nb[npages-1].lnb_file_offset +
1950                            local_nb[npages - 1].lnb_len - 1,
1951                            client_cksum, server_cksum);
1952 }
1953
1954 int tgt_brw_write(struct tgt_session_info *tsi)
1955 {
1956         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1957         struct ptlrpc_bulk_desc *desc = NULL;
1958         struct obd_export       *exp = req->rq_export;
1959         struct niobuf_remote    *remote_nb;
1960         struct niobuf_local     *local_nb;
1961         struct obd_ioobj        *ioo;
1962         struct ost_body         *body, *repbody;
1963         struct l_wait_info       lwi;
1964         struct lustre_handle     lockh = {0};
1965         __u32                   *rcs;
1966         int                      objcount, niocount, npages;
1967         int                      rc, i, j;
1968         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1969         bool                     no_reply = false, mmap;
1970         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1971
1972         ENTRY;
1973
1974         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1975                 CERROR("%s: deny write request from %s to portal %u\n",
1976                        tgt_name(tsi->tsi_tgt),
1977                        obd_export_nid2str(req->rq_export),
1978                        ptlrpc_req2svc(req)->srv_req_portal);
1979                 RETURN(err_serious(-EPROTO));
1980         }
1981
1982         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1983                 RETURN(err_serious(-ENOSPC));
1984         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1985                 RETURN(err_serious(-EROFS));
1986
1987         req->rq_bulk_write = 1;
1988
1989         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1990                 RETURN(err_serious(-EIO));
1991         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1992                 RETURN(err_serious(-EFAULT));
1993
1994         /* pause before transaction has been started */
1995         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1996                          cfs_fail_val : (obd_timeout + 1) / 4);
1997
1998         /* There must be big cache in current thread to process this request
1999          * if it is NULL then something went wrong and it wasn't allocated,
2000          * report -ENOMEM in that case */
2001         if (tbc == NULL)
2002                 RETURN(-ENOMEM);
2003
2004         body = tsi->tsi_ost_body;
2005         LASSERT(body != NULL);
2006
2007         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2008         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2009
2010         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2011                                         RCL_CLIENT) / sizeof(*ioo);
2012
2013         for (niocount = i = 0; i < objcount; i++)
2014                 niocount += ioo[i].ioo_bufcnt;
2015
2016         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2017         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2018         if (niocount != req_capsule_get_size(&req->rq_pill,
2019                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2020                         sizeof(*remote_nb))
2021                 RETURN(err_serious(-EPROTO));
2022
2023         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2024             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2025                 memory_pressure_set();
2026
2027         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2028                              niocount * sizeof(*rcs));
2029         rc = req_capsule_server_pack(&req->rq_pill);
2030         if (rc != 0)
2031                 GOTO(out, rc = err_serious(rc));
2032
2033         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2034         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2035
2036         local_nb = tbc->local;
2037
2038         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2039                           remote_nb, &lockh, LCK_PW);
2040         if (rc != 0)
2041                 GOTO(out, rc);
2042
2043         /*
2044          * If getting the lock took more time than
2045          * client was willing to wait, drop it. b=11330
2046          */
2047         if (cfs_time_current_sec() > req->rq_deadline ||
2048             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2049                 no_reply = true;
2050                 CERROR("%s: Dropping timed-out write from %s because locking "
2051                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
2052                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2053                        POSTID(&ioo->ioo_oid),
2054                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2055                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2056                 GOTO(out_lock, rc = -ETIMEDOUT);
2057         }
2058
2059         /* Because we already sync grant info with client when reconnect,
2060          * grant info will be cleared for resent req, then fed_grant and
2061          * total_grant will not be modified in following preprw_write */
2062         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2063                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2064                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2065         }
2066
2067         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2068         if (repbody == NULL)
2069                 GOTO(out_lock, rc = -ENOMEM);
2070         repbody->oa = body->oa;
2071
2072         npages = PTLRPC_MAX_BRW_PAGES;
2073         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2074                         objcount, ioo, remote_nb, &npages, local_nb);
2075         if (rc < 0)
2076                 GOTO(out_lock, rc);
2077
2078         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2079                                     PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2080                                     OST_BULK_PORTAL,
2081                                     &ptlrpc_bulk_kiov_nopin_ops);
2082         if (desc == NULL)
2083                 GOTO(skip_transfer, rc = -ENOMEM);
2084
2085         /* NB Having prepped, we must commit... */
2086         for (i = 0; i < npages; i++)
2087                 desc->bd_frag_ops->add_kiov_frag(desc,
2088                                                  local_nb[i].lnb_page,
2089                                                  local_nb[i].lnb_page_offset,
2090                                                  local_nb[i].lnb_len);
2091
2092         rc = sptlrpc_svc_prep_bulk(req, desc);
2093         if (rc != 0)
2094                 GOTO(skip_transfer, rc);
2095
2096         rc = target_bulk_io(exp, desc, &lwi);
2097         no_reply = rc != 0;
2098
2099 skip_transfer:
2100         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2101                 static int cksum_counter;
2102
2103                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2104                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2105
2106                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2107                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2108                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2109                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2110                                                         OST_WRITE, cksum_type);
2111                 cksum_counter++;
2112
2113                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2114                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2115                                 body->oa.o_flags & OBD_FL_MMAP);
2116
2117                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2118                                           body->oa.o_cksum,
2119                                           repbody->oa.o_cksum, mmap);
2120                         cksum_counter = 0;
2121                 } else if ((cksum_counter & (-cksum_counter)) ==
2122                            cksum_counter) {
2123                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2124                                cksum_counter, libcfs_id2str(req->rq_peer),
2125                                repbody->oa.o_cksum);
2126                 }
2127         }
2128
2129         /* Must commit after prep above in all cases */
2130         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2131                           objcount, ioo, remote_nb, npages, local_nb, rc);
2132         if (rc == -ENOTCONN)
2133                 /* quota acquire process has been given up because
2134                  * either the client has been evicted or the client
2135                  * has timed out the request already */
2136                 no_reply = true;
2137
2138         /*
2139          * Disable sending mtime back to the client. If the client locked the
2140          * whole object, then it has already updated the mtime on its side,
2141          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2142          */
2143         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2144
2145         if (rc == 0) {
2146                 int nob = 0;
2147
2148                 /* set per-requested niobuf return codes */
2149                 for (i = j = 0; i < niocount; i++) {
2150                         int len = remote_nb[i].rnb_len;
2151
2152                         nob += len;
2153                         rcs[i] = 0;
2154                         do {
2155                                 LASSERT(j < npages);
2156                                 if (local_nb[j].lnb_rc < 0)
2157                                         rcs[i] = local_nb[j].lnb_rc;
2158                                 len -= local_nb[j].lnb_len;
2159                                 j++;
2160                         } while (len > 0);
2161                         LASSERT(len == 0);
2162                 }
2163                 LASSERT(j == npages);
2164                 ptlrpc_lprocfs_brw(req, nob);
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