lustre/target/tgt_handler.c

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