lustre/target/tgt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2011, 2014, Intel Corporation.
  25  */
  26 /*
  27  * lustre/target/tgt_handler.c
  28  *
  29  * Lustre Unified Target request handler code
  30  *
  31  * Author: Brian Behlendorf <behlendorf1@llnl.gov>
  32  * Author: Mikhail Pershin <mike.pershin@intel.com>
  33  */
  34
  35 #define DEBUG_SUBSYSTEM S_CLASS
  36
  37 #include <linux/user_namespace.h>
  38 #ifdef HAVE_UIDGID_HEADER
  39 # include <linux/uidgid.h>
  40 #endif
  41
  42 #include <obd.h>
  43 #include <obd_class.h>
  44 #include <obd_cksum.h>
  45 #include <md_object.h>
  46 #include <lustre_lfsck.h>
  47 #include <lustre_nodemap.h>
  48
  49 #include "tgt_internal.h"
  50
  51 char *tgt_name(struct lu_target *tgt)
  52 {
  53         LASSERT(tgt->lut_obd != NULL);
  54         return tgt->lut_obd->obd_name;
  55 }
  56 EXPORT_SYMBOL(tgt_name);
  57
  58 /*
  59  * Generic code handling requests that have struct mdt_body passed in:
  60  *
  61  *  - extract mdt_body from request and save it in @tsi, if present;
  62  *
  63  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  64  *  @tsi;
  65  *
  66  *  - if HABEO_CORPUS flag is set for this request type check whether object
  67  *  actually exists on storage (lu_object_exists()).
  68  *
  69  */
  70 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  71 {
  72         const struct mdt_body   *body;
  73         struct lu_object        *obj;
  74         struct req_capsule      *pill = tsi->tsi_pill;
  75         int                      rc;
  76
  77         ENTRY;
  78
  79         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  80         if (body == NULL)
  81                 RETURN(-EFAULT);
  82
  83         tsi->tsi_mdt_body = body;
  84
  85         if (!(body->mbo_valid & OBD_MD_FLID))
  86                 RETURN(0);
  87
  88         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  89          * in any case in pre-2.5 clients. Fix that here if needed */
  90         if (unlikely(fid_is_zero(&body->mbo_fid1)))
  91                 RETURN(0);
  92
  93         if (!fid_is_sane(&body->mbo_fid1)) {
  94                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  95                        PFID(&body->mbo_fid1));
  96                 RETURN(-EINVAL);
  97         }
  98
  99         obj = lu_object_find(tsi->tsi_env,
 100                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
 101                              &body->mbo_fid1, NULL);
 102         if (!IS_ERR(obj)) {
 103                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
 104                         lu_object_put(tsi->tsi_env, obj);
 105                         rc = -ENOENT;
 106                 } else {
 107                         tsi->tsi_corpus = obj;
 108                         rc = 0;
 109                 }
 110         } else {
 111                 rc = PTR_ERR(obj);
 112         }
 113
 114         tsi->tsi_fid = body->mbo_fid1;
 115
 116         RETURN(rc);
 117 }
 118
 119 /**
 120  * Validate oa from client.
 121  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 122  * req are valid.
 123  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 124  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 125  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 126  *       whether this is oi_fid or real ostid. So it will check
 127  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 128  *    c. Old FID-disable osc will send IDIF.
 129  *    d. new FID-enable osc/osp will send normal FID.
 130  *
 131  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 132  * be used for LAST_ID file, and only being accessed inside OST now.
 133  */
 134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 135 {
 136         struct ost_id   *oi     = &oa->o_oi;
 137         u64              seq    = ostid_seq(oi);
 138         u64              id     = ostid_id(oi);
 139         int              rc;
 140         ENTRY;
 141
 142         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
 143                      fid_seq_is_echo(seq))) {
 144                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 145                  * during create, and we will reset this to 1, since this
 146                  * oi_id is basically useless in the following create process,
 147                  * but oi_id == 0 will make it difficult to tell whether it is
 148                  * real FID or ost_id. */
 149                 oi->oi_fid.f_seq = FID_SEQ_ECHO;
 150                 oi->oi_fid.f_oid = id ?: 1;
 151                 oi->oi_fid.f_ver = 0;
 152         } else {
 153                 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
 154
 155                 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
 156                         GOTO(out, rc = -EPROTO);
 157
 158                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 159                  * all of the requests are required to setup FLGROUP */
 160                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 161                         ostid_set_seq_mdt0(oi);
 162                         oa->o_valid |= OBD_MD_FLGROUP;
 163                         seq = ostid_seq(oi);
 164                 }
 165
 166                 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
 167                                fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
 168                         GOTO(out, rc = -EPROTO);
 169
 170                 rc = ostid_to_fid(&tti->tti_fid1, oi,
 171                                   tsi->tsi_tgt->lut_lsd.lsd_osd_index);
 172                 if (unlikely(rc != 0))
 173                         GOTO(out, rc);
 174
 175                 oi->oi_fid = tti->tti_fid1;
 176         }
 177
 178         RETURN(0);
 179
 180 out:
 181         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 182                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
 183                seq, id, rc);
 184         return rc;
 185 }
 186 EXPORT_SYMBOL(tgt_validate_obdo);
 187
 188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
 189 {
 190         unsigned                 max_brw;
 191         struct niobuf_remote    *rnb;
 192         struct obd_ioobj        *ioo;
 193         int                      obj_count;
 194
 195         ENTRY;
 196
 197         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
 198         if (ioo == NULL)
 199                 RETURN(-EPROTO);
 200
 201         rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
 202         if (rnb == NULL)
 203                 RETURN(-EPROTO);
 204
 205         max_brw = ioobj_max_brw_get(ioo);
 206         if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 207                 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 208                        ": rc = %d\n", tgt_name(tsi->tsi_tgt),
 209                        obd_export_nid2str(tsi->tsi_exp), max_brw,
 210                        POSTID(oi), -EPROTO);
 211                 RETURN(-EPROTO);
 212         }
 213         ioo->ioo_oid = *oi;
 214
 215         obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
 216                                         RCL_CLIENT) / sizeof(*ioo);
 217         if (obj_count == 0) {
 218                 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
 219                 RETURN(-EPROTO);
 220         } else if (obj_count > 1) {
 221                 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
 222                        obj_count);
 223                 RETURN(-EPROTO);
 224         }
 225
 226         if (ioo->ioo_bufcnt == 0) {
 227                 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
 228                 RETURN(-EPROTO);
 229         }
 230
 231         if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
 232                 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
 233                           "bulk has too many pages (%d)",
 234                           ioo->ioo_bufcnt);
 235                 RETURN(-EPROTO);
 236         }
 237
 238         RETURN(0);
 239 }
 240
 241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 242 {
 243         struct ost_body         *body;
 244         struct req_capsule      *pill = tsi->tsi_pill;
 245         struct lu_nodemap       *nodemap;
 246         int                      rc;
 247
 248         ENTRY;
 249
 250         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 251         if (body == NULL)
 252                 RETURN(-EFAULT);
 253
 254         rc = tgt_validate_obdo(tsi, &body->oa);
 255         if (rc)
 256                 RETURN(rc);
 257
 258         nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
 259
 260         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 261                                         NODEMAP_CLIENT_TO_FS,
 262                                         body->oa.o_uid);
 263         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 264                                         NODEMAP_CLIENT_TO_FS,
 265                                         body->oa.o_gid);
 266
 267         tsi->tsi_ost_body = body;
 268         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 269
 270         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 271                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 272                 if (rc < 0)
 273                         RETURN(rc);
 274         }
 275
 276         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 277                 if (flags & HABEO_CORPUS) {
 278                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 279                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 280                                tgt_name(tsi->tsi_tgt));
 281                         RETURN(-EPROTO);
 282                 } else {
 283                         RETURN(0);
 284                 }
 285         }
 286
 287         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 288
 289         /*
 290          * OST doesn't get object in advance for further use to prevent
 291          * situations with nested object_find which is potential deadlock.
 292          */
 293         tsi->tsi_corpus = NULL;
 294         RETURN(rc);
 295 }
 296
 297 /*
 298  * Do necessary preprocessing according to handler ->th_flags.
 299  */
 300 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 301                                   struct tgt_handler *h,
 302                                   struct ptlrpc_request *req)
 303 {
 304         struct req_capsule      *pill = tsi->tsi_pill;
 305         __u32                    flags = h->th_flags;
 306         int                      rc = 0;
 307
 308         ENTRY;
 309
 310         if (tsi->tsi_preprocessed)
 311                 RETURN(0);
 312
 313         LASSERT(h->th_act != NULL);
 314         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 315         LASSERT(current->journal_info == NULL);
 316
 317         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 318                      h->th_fmt != NULL));
 319         if (h->th_fmt != NULL) {
 320                 req_capsule_set(pill, h->th_fmt);
 321                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 322                         rc = tgt_mdt_body_unpack(tsi, flags);
 323                         if (rc < 0)
 324                                 RETURN(rc);
 325                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 326                                                  RCL_CLIENT)) {
 327                         rc = tgt_ost_body_unpack(tsi, flags);
 328                         if (rc < 0)
 329                                 RETURN(rc);
 330                 }
 331         }
 332
 333         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 334                 RETURN(-EROFS);
 335
 336         if (flags & HABEO_CLAVIS) {
 337                 struct ldlm_request *dlm_req;
 338
 339                 LASSERT(h->th_fmt != NULL);
 340
 341                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 342                 if (dlm_req != NULL) {
 343                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 344                                      LDLM_IBITS &&
 345                                      dlm_req->lock_desc.l_policy_data.\
 346                                      l_inodebits.bits == 0)) {
 347                                 /*
 348                                  * Lock without inodebits makes no sense and
 349                                  * will oops later in ldlm. If client miss to
 350                                  * set such bits, do not trigger ASSERTION.
 351                                  *
 352                                  * For liblustre flock case, it maybe zero.
 353                                  */
 354                                 rc = -EPROTO;
 355                         } else {
 356                                 tsi->tsi_dlm_req = dlm_req;
 357                         }
 358                 } else {
 359                         rc = -EFAULT;
 360                 }
 361         }
 362         tsi->tsi_preprocessed = 1;
 363         RETURN(rc);
 364 }
 365
 366 /*
 367  * Invoke handler for this request opc. Also do necessary preprocessing
 368  * (according to handler ->th_flags), and post-processing (setting of
 369  * ->last_{xid,committed}).
 370  */
 371 static int tgt_handle_request0(struct tgt_session_info *tsi,
 372                                struct tgt_handler *h,
 373                                struct ptlrpc_request *req)
 374 {
 375         int      serious = 0;
 376         int      rc;
 377         __u32    opc = lustre_msg_get_opc(req->rq_reqmsg);
 378
 379         ENTRY;
 380
 381
 382         /* When dealing with sec context requests, no export is associated yet,
 383          * because these requests are sent before *_CONNECT requests.
 384          * A NULL req->rq_export means the normal *_common_slice handlers will
 385          * not be called, because there is no reference to the target.
 386          * So deal with them by hand and jump directly to target_send_reply().
 387          */
 388         switch (opc) {
 389         case SEC_CTX_INIT:
 390         case SEC_CTX_INIT_CONT:
 391         case SEC_CTX_FINI:
 392                 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
 393                 GOTO(out, rc = 0);
 394         }
 395
 396         /*
 397          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 398          * to put same checks into handlers like mdt_close(), mdt_reint(),
 399          * etc., without talking to mdt authors first. Checking same thing
 400          * there again is useless and returning 0 error without packing reply
 401          * is buggy! Handlers either pack reply or return error.
 402          *
 403          * We return 0 here and do not send any reply in order to emulate
 404          * network failure. Do not send any reply in case any of NET related
 405          * fail_id has occured.
 406          */
 407         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 408                 RETURN(0);
 409         if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
 410                      OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
 411                 RETURN(0);
 412
 413         rc = tgt_request_preprocess(tsi, h, req);
 414         /* pack reply if reply format is fixed */
 415         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 416                 /* Pack reply */
 417                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 418                                           RCL_SERVER))
 419                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 420                                              RCL_SERVER,
 421                                              tsi->tsi_mdt_body->mbo_eadatasize);
 422                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 423                                           RCL_SERVER))
 424                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 425                                              RCL_SERVER, 0);
 426
 427                 rc = req_capsule_server_pack(tsi->tsi_pill);
 428         }
 429
 430         if (likely(rc == 0)) {
 431                 /*
 432                  * Process request, there can be two types of rc:
 433                  * 1) errors with msg unpack/pack, other failures outside the
 434                  * operation itself. This is counted as serious errors;
 435                  * 2) errors during fs operation, should be placed in rq_status
 436                  * only
 437                  */
 438                 rc = h->th_act(tsi);
 439                 if (!is_serious(rc) &&
 440                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 441                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 442                                   "pack reply and returned 0 error\n",
 443                                   tgt_name(tsi->tsi_tgt), h->th_name);
 444                         LBUG();
 445                 }
 446                 serious = is_serious(rc);
 447                 rc = clear_serious(rc);
 448         } else {
 449                 serious = 1;
 450         }
 451
 452         req->rq_status = rc;
 453
 454         /*
 455          * ELDLM_* codes which > 0 should be in rq_status only as well as
 456          * all non-serious errors.
 457          */
 458         if (rc > 0 || !serious)
 459                 rc = 0;
 460
 461         LASSERT(current->journal_info == NULL);
 462
 463         if (likely(rc == 0 && req->rq_export))
 464                 target_committed_to_req(req);
 465
 466 out:
 467         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 468         RETURN(0);
 469 }
 470
 471 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 472                                        struct obd_device *obd, int *process)
 473 {
 474         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 475         case MDS_DISCONNECT:
 476         case OST_DISCONNECT:
 477         case OBD_IDX_READ:
 478                 *process = 1;
 479                 RETURN(0);
 480         case MDS_CLOSE:
 481         case MDS_SYNC: /* used in unmounting */
 482         case OBD_PING:
 483         case MDS_REINT:
 484         case OUT_UPDATE:
 485         case SEQ_QUERY:
 486         case FLD_QUERY:
 487         case FLD_READ:
 488         case LDLM_ENQUEUE:
 489         case OST_CREATE:
 490         case OST_DESTROY:
 491         case OST_PUNCH:
 492         case OST_SETATTR:
 493         case OST_SYNC:
 494         case OST_WRITE:
 495         case MDS_HSM_PROGRESS:
 496         case MDS_HSM_STATE_SET:
 497         case MDS_HSM_REQUEST:
 498                 *process = target_queue_recovery_request(req, obd);
 499                 RETURN(0);
 500
 501         default:
 502                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 503                 *process = -EAGAIN;
 504                 RETURN(0);
 505         }
 506 }
 507
 508 /*
 509  * Handle recovery. Return:
 510  *        +1: continue request processing;
 511  *       -ve: abort immediately with the given error code;
 512  *         0: send reply with error code in req->rq_status;
 513  */
 514 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 515 {
 516         ENTRY;
 517
 518         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 519         case MDS_CONNECT:
 520         case OST_CONNECT:
 521         case MGS_CONNECT:
 522         case SEC_CTX_INIT:
 523         case SEC_CTX_INIT_CONT:
 524         case SEC_CTX_FINI:
 525                 RETURN(+1);
 526         }
 527
 528         if (!req->rq_export->exp_obd->obd_replayable)
 529                 RETURN(+1);
 530
 531         /* sanity check: if the xid matches, the request must be marked as a
 532          * resent or replayed */
 533         if (req_can_reconstruct(req, NULL)) {
 534                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 535                       (MSG_RESENT | MSG_REPLAY))) {
 536                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 537                                   "saved xid, expected REPLAY or RESENT flag "
 538                                   "(%x)", req->rq_xid,
 539                                   lustre_msg_get_flags(req->rq_reqmsg));
 540                         req->rq_status = -ENOTCONN;
 541                         RETURN(-ENOTCONN);
 542                 }
 543         }
 544         /* else: note the opposite is not always true; a RESENT req after a
 545          * failover will usually not match the last_xid, since it was likely
 546          * never committed. A REPLAYed request will almost never match the
 547          * last xid, however it could for a committed, but still retained,
 548          * open. */
 549
 550         /* Check for aborted recovery... */
 551         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 552                 int rc;
 553                 int should_process;
 554
 555                 DEBUG_REQ(D_INFO, req, "Got new replay");
 556                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 557                                                  &should_process);
 558                 if (rc != 0 || !should_process)
 559                         RETURN(rc);
 560                 else if (should_process < 0) {
 561                         req->rq_status = should_process;
 562                         rc = ptlrpc_error(req);
 563                         RETURN(rc);
 564                 }
 565         }
 566         RETURN(+1);
 567 }
 568
 569 /* Initial check for request, it is validation mostly */
 570 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 571 {
 572         struct tgt_handler      *h;
 573         struct tgt_opc_slice    *s;
 574         struct lu_target        *tgt;
 575         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 576
 577         ENTRY;
 578
 579         tgt = class_exp2tgt(req->rq_export);
 580         if (unlikely(tgt == NULL)) {
 581                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 582                           class_exp2obd(req->rq_export)->obd_name);
 583                 RETURN(ERR_PTR(-EINVAL));
 584         }
 585
 586         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 587                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 588                         break;
 589
 590         /* opcode was not found in slice */
 591         if (unlikely(s->tos_hs == NULL)) {
 592                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 593                        opc);
 594                 RETURN(ERR_PTR(-ENOTSUPP));
 595         }
 596
 597         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 598         h = s->tos_hs + (opc - s->tos_opc_start);
 599         if (unlikely(h->th_opc == 0)) {
 600                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 601                 RETURN(ERR_PTR(-ENOTSUPP));
 602         }
 603
 604         RETURN(h);
 605 }
 606
 607 int tgt_request_handle(struct ptlrpc_request *req)
 608 {
 609         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 610
 611         struct lustre_msg       *msg = req->rq_reqmsg;
 612         struct tgt_handler      *h;
 613         struct lu_target        *tgt;
 614         int                      request_fail_id = 0;
 615         __u32                    opc = lustre_msg_get_opc(msg);
 616         int                      rc;
 617
 618         ENTRY;
 619
 620         /* Refill the context, to make sure all thread keys are allocated */
 621         lu_env_refill(req->rq_svc_thread->t_env);
 622
 623         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 624         tsi->tsi_pill = &req->rq_pill;
 625         tsi->tsi_env = req->rq_svc_thread->t_env;
 626
 627         /* if request has export then get handlers slice from corresponding
 628          * target, otherwise that should be connect operation */
 629         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 630             opc == MGS_CONNECT) {
 631                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 632                 rc = target_handle_connect(req);
 633                 if (rc != 0) {
 634                         rc = ptlrpc_error(req);
 635                         GOTO(out, rc);
 636                 }
 637                 /* recovery-small test 18c asks to drop connect reply */
 638                 if (unlikely(opc == OST_CONNECT &&
 639                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 640                         GOTO(out, rc = 0);
 641         }
 642
 643         if (unlikely(!class_connected_export(req->rq_export))) {
 644                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
 645                     opc == SEC_CTX_FINI) {
 646                         /* sec context initialization has to be handled
 647                          * by hand in tgt_handle_request0() */
 648                         tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
 649                         h = NULL;
 650                         GOTO(handle_recov, rc = 0);
 651                 }
 652                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 653                        opc, libcfs_id2str(req->rq_peer));
 654                 req->rq_status = -ENOTCONN;
 655                 rc = ptlrpc_error(req);
 656                 GOTO(out, rc);
 657         }
 658
 659         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 660         tsi->tsi_exp = req->rq_export;
 661         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 662                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 663         else
 664                 tsi->tsi_jobid = NULL;
 665
 666         if (tgt == NULL) {
 667                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 668                           class_exp2obd(req->rq_export)->obd_name);
 669                 req->rq_status = -EINVAL;
 670                 rc = ptlrpc_error(req);
 671                 GOTO(out, rc);
 672         }
 673
 674         /* check request's xid is consistent with export's last_xid */
 675         if (req->rq_export != NULL) {
 676                 __u64 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
 677                 if (last_xid != 0)
 678                         req->rq_export->exp_last_xid = last_xid;
 679                 if (req->rq_xid == 0 ||
 680                     req->rq_xid <= req->rq_export->exp_last_xid) {
 681                         DEBUG_REQ(D_ERROR, req,
 682                                   "Unexpected xid %llx vs. last_xid %llx\n",
 683                                   req->rq_xid, req->rq_export->exp_last_xid);
 684 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 93, 0)
 685                         LBUG();
 686 #endif
 687                         req->rq_status = -EPROTO;
 688                         rc = ptlrpc_error(req);
 689                         GOTO(out, rc);
 690                 }
 691         }
 692
 693         request_fail_id = tgt->lut_request_fail_id;
 694         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 695
 696         /* try to release in-memory reply data */
 697         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 698                 tgt_handle_received_xid(req->rq_export,
 699                                 lustre_msg_get_last_xid(req->rq_reqmsg));
 700                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 701                       (MSG_RESENT | MSG_REPLAY)))
 702                         tgt_handle_tag(req->rq_export,
 703                                        lustre_msg_get_tag(req->rq_reqmsg));
 704         }
 705
 706         h = tgt_handler_find_check(req);
 707         if (IS_ERR(h)) {
 708                 req->rq_status = PTR_ERR(h);
 709                 rc = ptlrpc_error(req);
 710                 GOTO(out, rc);
 711         }
 712
 713         LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 714                  h->th_opc, opc);
 715
 716         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 717                 GOTO(out, rc = 0);
 718
 719         rc = lustre_msg_check_version(msg, h->th_version);
 720         if (unlikely(rc)) {
 721                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 722                           " %08x, expecting %08x\n", tgt_name(tgt),
 723                           lustre_msg_get_version(msg), h->th_version);
 724                 req->rq_status = -EINVAL;
 725                 rc = ptlrpc_error(req);
 726                 GOTO(out, rc);
 727         }
 728
 729 handle_recov:
 730         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 731         if (likely(rc == 1)) {
 732                 rc = tgt_handle_request0(tsi, h, req);
 733                 if (rc)
 734                         GOTO(out, rc);
 735         }
 736         EXIT;
 737 out:
 738         req_capsule_fini(tsi->tsi_pill);
 739         if (tsi->tsi_corpus != NULL) {
 740                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 741                 tsi->tsi_corpus = NULL;
 742         }
 743         return rc;
 744 }
 745 EXPORT_SYMBOL(tgt_request_handle);
 746
 747 /** Assign high priority operations to the request if needed. */
 748 int tgt_hpreq_handler(struct ptlrpc_request *req)
 749 {
 750         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 751         struct tgt_handler      *h;
 752         int                      rc;
 753
 754         ENTRY;
 755
 756         if (req->rq_export == NULL)
 757                 RETURN(0);
 758
 759         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 760         tsi->tsi_pill = &req->rq_pill;
 761         tsi->tsi_env = req->rq_svc_thread->t_env;
 762         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 763         tsi->tsi_exp = req->rq_export;
 764
 765         h = tgt_handler_find_check(req);
 766         if (IS_ERR(h)) {
 767                 rc = PTR_ERR(h);
 768                 RETURN(rc);
 769         }
 770
 771         rc = tgt_request_preprocess(tsi, h, req);
 772         if (unlikely(rc != 0))
 773                 RETURN(rc);
 774
 775         if (h->th_hp != NULL)
 776                 h->th_hp(tsi);
 777         RETURN(0);
 778 }
 779 EXPORT_SYMBOL(tgt_hpreq_handler);
 780
 781 void tgt_counter_incr(struct obd_export *exp, int opcode)
 782 {
 783         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 784         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 785                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 786 }
 787 EXPORT_SYMBOL(tgt_counter_incr);
 788
 789 /*
 790  * Unified target generic handlers.
 791  */
 792
 793 /*
 794  * Security functions
 795  */
 796 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 797 {
 798         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 799                                       OBD_CONNECT_RMT_CLIENT_FORCE);
 800 }
 801
 802 static int tgt_init_sec_level(struct ptlrpc_request *req)
 803 {
 804         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 805         char                    *client;
 806         struct obd_connect_data *data, *reply;
 807         int                      rc = 0;
 808         bool                     remote;
 809         ENTRY;
 810
 811         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 812         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 813         if (data == NULL || reply == NULL)
 814                 RETURN(-EFAULT);
 815
 816         /* connection from MDT is always trusted */
 817         if (req->rq_auth_usr_mdt) {
 818                 tgt_init_sec_none(reply);
 819                 RETURN(0);
 820         }
 821
 822         if (unlikely(tgt == NULL)) {
 823                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 824                           class_exp2obd(req->rq_export)->obd_name);
 825                 RETURN(-EINVAL);
 826         }
 827
 828         client = libcfs_nid2str(req->rq_peer.nid);
 829         /* no GSS support case */
 830         if (!req->rq_auth_gss) {
 831                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 832                         CWARN("client %s -> target %s does not use GSS, "
 833                               "can not run under security level %d.\n",
 834                               client, tgt_name(tgt), tgt->lut_sec_level);
 835                         RETURN(-EACCES);
 836                 } else {
 837                         tgt_init_sec_none(reply);
 838                         RETURN(0);
 839                 }
 840         }
 841
 842         /* old version case */
 843         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 844                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 845                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 846                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 847                         CWARN("client %s -> target %s uses old version, "
 848                               "can not run under security level %d.\n",
 849                               client, tgt_name(tgt), tgt->lut_sec_level);
 850                         RETURN(-EACCES);
 851                 } else {
 852                         CWARN("client %s -> target %s uses old version, "
 853                               "run under security level %d.\n",
 854                               client, tgt_name(tgt), tgt->lut_sec_level);
 855                         tgt_init_sec_none(reply);
 856                         RETURN(0);
 857                 }
 858         }
 859
 860         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 861         if (remote) {
 862                 if (!req->rq_auth_remote)
 863                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 864                                "asked to be remote.\n", client, tgt_name(tgt));
 865         } else if (req->rq_auth_remote) {
 866                 remote = true;
 867                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 868                        "as remote by default.\n", client, tgt_name(tgt));
 869         }
 870
 871         if (remote == 0) {
 872                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 873                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 874                                "authenticated!\n", client, tgt_name(tgt));
 875                         RETURN(-EACCES);
 876                 }
 877         }
 878
 879
 880         switch (tgt->lut_sec_level) {
 881         case LUSTRE_SEC_NONE:
 882                 if (remote) {
 883                         CDEBUG(D_SEC,
 884                                "client %s -> target %s is set as remote, "
 885                                "can not run under security level %d.\n",
 886                                client, tgt_name(tgt), tgt->lut_sec_level);
 887                         RETURN(-EACCES);
 888                 }
 889                 tgt_init_sec_none(reply);
 890                 break;
 891         case LUSTRE_SEC_REMOTE:
 892                 if (!remote)
 893                         tgt_init_sec_none(reply);
 894                 break;
 895         case LUSTRE_SEC_ALL:
 896                 if (remote)
 897                         break;
 898                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 899                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 900                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 901                 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 902                 break;
 903         default:
 904                 RETURN(-EINVAL);
 905         }
 906
 907         RETURN(rc);
 908 }
 909
 910 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 911 {
 912         struct lu_target        *tgt = class_exp2tgt(exp);
 913         struct sptlrpc_flavor    flvr;
 914         int                      rc = 0;
 915
 916         LASSERT(tgt);
 917         LASSERT(tgt->lut_obd);
 918         LASSERT(tgt->lut_slice);
 919
 920         /* always allow ECHO client */
 921         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 922                             LUSTRE_ECHO_NAME) == 0)) {
 923                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 924                 return 0;
 925         }
 926
 927         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 928                 read_lock(&tgt->lut_sptlrpc_lock);
 929                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 930                                              req->rq_sp_from,
 931                                              req->rq_peer.nid,
 932                                              &flvr);
 933                 read_unlock(&tgt->lut_sptlrpc_lock);
 934
 935                 spin_lock(&exp->exp_lock);
 936                 exp->exp_sp_peer = req->rq_sp_from;
 937                 exp->exp_flvr = flvr;
 938
 939                 /* when on mgs, if no restriction is set, or if client
 940                  * is loopback, allow any flavor */
 941                 if ((strcmp(exp->exp_obd->obd_type->typ_name,
 942                            LUSTRE_MGS_NAME) == 0) &&
 943                      (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
 944                       LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
 945                       == LOLND))
 946                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 947
 948                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 949                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 950                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 951                                "expect %x\n", tgt_name(tgt),
 952                                req->rq_flvr.sf_rpc,
 953                                libcfs_nid2str(req->rq_peer.nid),
 954                                exp->exp_flvr.sf_rpc);
 955                         rc = -EACCES;
 956                 }
 957                 spin_unlock(&exp->exp_lock);
 958         } else {
 959                 if (exp->exp_sp_peer != req->rq_sp_from) {
 960                         CERROR("%s: RPC source %s doesn't match %s\n",
 961                                tgt_name(tgt),
 962                                sptlrpc_part2name(req->rq_sp_from),
 963                                sptlrpc_part2name(exp->exp_sp_peer));
 964                         rc = -EACCES;
 965                 } else {
 966                         rc = sptlrpc_target_export_check(exp, req);
 967                 }
 968         }
 969
 970         return rc;
 971 }
 972
 973 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 974 {
 975         struct sptlrpc_rule_set  tmp_rset;
 976         int                      rc;
 977
 978         if (unlikely(tgt == NULL)) {
 979                 CERROR("No target passed");
 980                 return -EINVAL;
 981         }
 982
 983         sptlrpc_rule_set_init(&tmp_rset);
 984         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 985         if (rc) {
 986                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 987                        tgt_name(tgt), rc);
 988                 return rc;
 989         }
 990
 991         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 992
 993         write_lock(&tgt->lut_sptlrpc_lock);
 994         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 995         tgt->lut_sptlrpc_rset = tmp_rset;
 996         write_unlock(&tgt->lut_sptlrpc_lock);
 997
 998         return 0;
 999 }
1000 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
1001
1002 int tgt_connect(struct tgt_session_info *tsi)
1003 {
1004         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1005         struct obd_connect_data *reply;
1006         int                      rc;
1007
1008         ENTRY;
1009
1010         rc = tgt_init_sec_level(req);
1011         if (rc != 0)
1012                 GOTO(out, rc);
1013
1014         /* XXX: better to call this check right after getting new export but
1015          * before last_rcvd slot allocation to avoid server load upon insecure
1016          * connects. This is to be fixed after unifiyng all targets.
1017          */
1018         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1019         if (rc)
1020                 GOTO(out, rc);
1021
1022         /* To avoid exposing partially initialized connection flags, changes up
1023          * to this point have been staged in reply->ocd_connect_flags. Now that
1024          * connection handling has completed successfully, atomically update
1025          * the connect flags in the shared export data structure. LU-1623 */
1026         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1027         spin_lock(&tsi->tsi_exp->exp_lock);
1028         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1029         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1030         spin_unlock(&tsi->tsi_exp->exp_lock);
1031
1032         RETURN(0);
1033 out:
1034         obd_disconnect(class_export_get(tsi->tsi_exp));
1035         return rc;
1036 }
1037 EXPORT_SYMBOL(tgt_connect);
1038
1039 int tgt_disconnect(struct tgt_session_info *tsi)
1040 {
1041         int rc;
1042
1043         ENTRY;
1044
1045         rc = target_handle_disconnect(tgt_ses_req(tsi));
1046         if (rc)
1047                 RETURN(err_serious(rc));
1048
1049         RETURN(rc);
1050 }
1051 EXPORT_SYMBOL(tgt_disconnect);
1052
1053 /*
1054  * Unified target OBD handlers
1055  */
1056 int tgt_obd_ping(struct tgt_session_info *tsi)
1057 {
1058         int rc;
1059
1060         ENTRY;
1061
1062         rc = target_handle_ping(tgt_ses_req(tsi));
1063         if (rc)
1064                 RETURN(err_serious(rc));
1065
1066         RETURN(rc);
1067 }
1068 EXPORT_SYMBOL(tgt_obd_ping);
1069
1070 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1071 {
1072         return err_serious(-EOPNOTSUPP);
1073 }
1074
1075 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1076 {
1077         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1078         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1079         struct obd_export       *exp = req->rq_export;
1080         struct ptlrpc_bulk_desc *desc;
1081         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1082         int                      tmpcount;
1083         int                      tmpsize;
1084         int                      i;
1085         int                      rc;
1086
1087         ENTRY;
1088
1089         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1090                                     PTLRPC_BULK_PUT_SOURCE |
1091                                         PTLRPC_BULK_BUF_KIOV,
1092                                     MDS_BULK_PORTAL,
1093                                     &ptlrpc_bulk_kiov_pin_ops);
1094         if (desc == NULL)
1095                 RETURN(-ENOMEM);
1096
1097         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1098                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1099                  * which is rdpg->rp_count */
1100                 nob = rdpg->rp_count;
1101
1102         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1103              i++, tmpcount -= tmpsize) {
1104                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1105                 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1106                                                  tmpsize);
1107         }
1108
1109         LASSERT(desc->bd_nob == nob);
1110         rc = target_bulk_io(exp, desc, lwi);
1111         ptlrpc_free_bulk(desc);
1112         RETURN(rc);
1113 }
1114 EXPORT_SYMBOL(tgt_sendpage);
1115
1116 /*
1117  * OBD_IDX_READ handler
1118  */
1119 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1120 {
1121         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1122         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1123         struct idx_info         *req_ii, *rep_ii;
1124         int                      rc, i;
1125
1126         ENTRY;
1127
1128         memset(rdpg, 0, sizeof(*rdpg));
1129         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1130
1131         /* extract idx_info buffer from request & reply */
1132         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1133         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1134                 RETURN(err_serious(-EPROTO));
1135
1136         rc = req_capsule_server_pack(tsi->tsi_pill);
1137         if (rc)
1138                 RETURN(err_serious(rc));
1139
1140         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1141         if (rep_ii == NULL)
1142                 RETURN(err_serious(-EFAULT));
1143         rep_ii->ii_magic = IDX_INFO_MAGIC;
1144
1145         /* extract hash to start with */
1146         rdpg->rp_hash = req_ii->ii_hash_start;
1147
1148         /* extract requested attributes */
1149         rdpg->rp_attrs = req_ii->ii_attrs;
1150
1151         /* check that fid packed in request is valid and supported */
1152         if (!fid_is_sane(&req_ii->ii_fid))
1153                 RETURN(-EINVAL);
1154         rep_ii->ii_fid = req_ii->ii_fid;
1155
1156         /* copy flags */
1157         rep_ii->ii_flags = req_ii->ii_flags;
1158
1159         /* compute number of pages to allocate, ii_count is the number of 4KB
1160          * containers */
1161         if (req_ii->ii_count <= 0)
1162                 GOTO(out, rc = -EFAULT);
1163         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1164                                exp_max_brw_size(tsi->tsi_exp));
1165         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1166
1167         /* allocate pages to store the containers */
1168         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1169         if (rdpg->rp_pages == NULL)
1170                 GOTO(out, rc = -ENOMEM);
1171         for (i = 0; i < rdpg->rp_npages; i++) {
1172                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1173                 if (rdpg->rp_pages[i] == NULL)
1174                         GOTO(out, rc = -ENOMEM);
1175         }
1176
1177         /* populate pages with key/record pairs */
1178         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1179         if (rc < 0)
1180                 GOTO(out, rc);
1181
1182         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1183                  "asked %d > %d\n", rc, rdpg->rp_count);
1184
1185         /* send pages to client */
1186         rc = tgt_sendpage(tsi, rdpg, rc);
1187         if (rc)
1188                 GOTO(out, rc);
1189         EXIT;
1190 out:
1191         if (rdpg->rp_pages) {
1192                 for (i = 0; i < rdpg->rp_npages; i++)
1193                         if (rdpg->rp_pages[i])
1194                                 __free_page(rdpg->rp_pages[i]);
1195                 OBD_FREE(rdpg->rp_pages,
1196                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1197         }
1198         return rc;
1199 }
1200
1201 struct tgt_handler tgt_obd_handlers[] = {
1202 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1203 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1204 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1205 };
1206 EXPORT_SYMBOL(tgt_obd_handlers);
1207
1208 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1209              struct dt_object *obj, __u64 start, __u64 end)
1210 {
1211         int rc = 0;
1212
1213         ENTRY;
1214
1215         /* if no objid is specified, it means "sync whole filesystem" */
1216         if (obj == NULL) {
1217                 rc = dt_sync(env, tgt->lut_bottom);
1218         } else if (dt_version_get(env, obj) >
1219                    tgt->lut_obd->obd_last_committed) {
1220                 rc = dt_object_sync(env, obj, start, end);
1221         }
1222
1223         RETURN(rc);
1224 }
1225 EXPORT_SYMBOL(tgt_sync);
1226 /*
1227  * Unified target DLM handlers.
1228  */
1229
1230 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1231  * (if requested) */
1232 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1233                             void *data, int flag)
1234 {
1235         struct lu_env            env;
1236         struct lu_target        *tgt;
1237         struct dt_object        *obj;
1238         struct lu_fid            fid;
1239         int                      rc = 0;
1240
1241         ENTRY;
1242
1243         tgt = class_exp2tgt(lock->l_export);
1244
1245         if (unlikely(tgt == NULL)) {
1246                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1247                        class_exp2obd(lock->l_export)->obd_name);
1248                 RETURN(-EINVAL);
1249         }
1250
1251         if (flag == LDLM_CB_CANCELING &&
1252             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1253             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1254              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1255               lock->l_flags & LDLM_FL_CBPENDING))) {
1256                 __u64 start = 0;
1257                 __u64 end = OBD_OBJECT_EOF;
1258
1259                 rc = lu_env_init(&env, LCT_DT_THREAD);
1260                 if (unlikely(rc != 0))
1261                         RETURN(rc);
1262
1263                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1264                                    tgt->lut_lsd.lsd_osd_index);
1265                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1266                 if (IS_ERR(obj))
1267                         GOTO(err_env, rc = PTR_ERR(obj));
1268
1269                 if (!dt_object_exists(obj))
1270                         GOTO(err_put, rc = -ENOENT);
1271
1272                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1273                         start = lock->l_policy_data.l_extent.start;
1274                         end = lock->l_policy_data.l_extent.end;
1275                 }
1276
1277                 rc = tgt_sync(&env, tgt, obj, start, end);
1278                 if (rc < 0) {
1279                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1280                                "cancel: rc = %d\n",
1281                                tgt_name(tgt), PFID(&fid),
1282                                lock->l_policy_data.l_extent.start,
1283                                lock->l_policy_data.l_extent.end, rc);
1284                 }
1285 err_put:
1286                 lu_object_put(&env, &obj->do_lu);
1287 err_env:
1288                 lu_env_fini(&env);
1289         }
1290
1291         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1292         RETURN(rc);
1293 }
1294
1295 static struct ldlm_callback_suite tgt_dlm_cbs = {
1296         .lcs_completion = ldlm_server_completion_ast,
1297         .lcs_blocking   = tgt_blocking_ast,
1298         .lcs_glimpse    = ldlm_server_glimpse_ast
1299 };
1300
1301 int tgt_enqueue(struct tgt_session_info *tsi)
1302 {
1303         struct ptlrpc_request *req = tgt_ses_req(tsi);
1304         int rc;
1305
1306         ENTRY;
1307         /*
1308          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1309          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1310          */
1311         LASSERT(tsi->tsi_dlm_req != NULL);
1312         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1313                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1314         if (rc)
1315                 RETURN(err_serious(rc));
1316
1317         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1318         case LUT_FAIL_MDT:
1319                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1320                 break;
1321         case LUT_FAIL_OST:
1322                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1323                 break;
1324         case LUT_FAIL_MGT:
1325                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1326                 break;
1327         default:
1328                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1329                 break;
1330         }
1331         RETURN(req->rq_status);
1332 }
1333 EXPORT_SYMBOL(tgt_enqueue);
1334
1335 int tgt_convert(struct tgt_session_info *tsi)
1336 {
1337         struct ptlrpc_request *req = tgt_ses_req(tsi);
1338         int rc;
1339
1340         ENTRY;
1341         LASSERT(tsi->tsi_dlm_req);
1342         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1343         if (rc)
1344                 RETURN(err_serious(rc));
1345
1346         RETURN(req->rq_status);
1347 }
1348
1349 int tgt_bl_callback(struct tgt_session_info *tsi)
1350 {
1351         return err_serious(-EOPNOTSUPP);
1352 }
1353
1354 int tgt_cp_callback(struct tgt_session_info *tsi)
1355 {
1356         return err_serious(-EOPNOTSUPP);
1357 }
1358
1359 /* generic LDLM target handler */
1360 struct tgt_handler tgt_dlm_handlers[] = {
1361 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1362 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1363 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1364 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1365 };
1366 EXPORT_SYMBOL(tgt_dlm_handlers);
1367
1368 /*
1369  * Unified target LLOG handlers.
1370  */
1371 int tgt_llog_open(struct tgt_session_info *tsi)
1372 {
1373         int rc;
1374
1375         ENTRY;
1376
1377         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1378
1379         RETURN(rc);
1380 }
1381 EXPORT_SYMBOL(tgt_llog_open);
1382
1383 int tgt_llog_close(struct tgt_session_info *tsi)
1384 {
1385         int rc;
1386
1387         ENTRY;
1388
1389         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1390
1391         RETURN(rc);
1392 }
1393 EXPORT_SYMBOL(tgt_llog_close);
1394
1395
1396 int tgt_llog_destroy(struct tgt_session_info *tsi)
1397 {
1398         int rc;
1399
1400         ENTRY;
1401
1402         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1403
1404         RETURN(rc);
1405 }
1406
1407 int tgt_llog_read_header(struct tgt_session_info *tsi)
1408 {
1409         int rc;
1410
1411         ENTRY;
1412
1413         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1414
1415         RETURN(rc);
1416 }
1417 EXPORT_SYMBOL(tgt_llog_read_header);
1418
1419 int tgt_llog_next_block(struct tgt_session_info *tsi)
1420 {
1421         int rc;
1422
1423         ENTRY;
1424
1425         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1426
1427         RETURN(rc);
1428 }
1429 EXPORT_SYMBOL(tgt_llog_next_block);
1430
1431 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1432 {
1433         int rc;
1434
1435         ENTRY;
1436
1437         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1438
1439         RETURN(rc);
1440 }
1441 EXPORT_SYMBOL(tgt_llog_prev_block);
1442
1443 /* generic llog target handler */
1444 struct tgt_handler tgt_llog_handlers[] = {
1445 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1446 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1447 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1448 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1449 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1450 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1451 };
1452 EXPORT_SYMBOL(tgt_llog_handlers);
1453
1454 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1455                            struct dt_device *key,
1456                            struct lfsck_request *lr,
1457                            struct thandle *th) = NULL;
1458
1459 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1460                                                 struct dt_device *,
1461                                                 struct lfsck_request *,
1462                                                 struct thandle *))
1463 {
1464         tgt_lfsck_in_notify = notify;
1465 }
1466 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1467
1468 static int (*tgt_lfsck_query)(const struct lu_env *env,
1469                               struct dt_device *key,
1470                               struct lfsck_request *lr) = NULL;
1471
1472 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1473                                            struct dt_device *,
1474                                            struct lfsck_request *))
1475 {
1476         tgt_lfsck_query = query;
1477 }
1478 EXPORT_SYMBOL(tgt_register_lfsck_query);
1479
1480 /* LFSCK request handlers */
1481 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1482 {
1483         const struct lu_env     *env = tsi->tsi_env;
1484         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1485         struct lfsck_request    *lr;
1486         int                      rc;
1487         ENTRY;
1488
1489         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1490         if (lr == NULL)
1491                 RETURN(-EPROTO);
1492
1493         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1494
1495         RETURN(rc);
1496 }
1497
1498 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1499 {
1500         struct lfsck_request    *request;
1501         struct lfsck_reply      *reply;
1502         int                      rc;
1503         ENTRY;
1504
1505         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1506         if (request == NULL)
1507                 RETURN(-EPROTO);
1508
1509         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1510         if (reply == NULL)
1511                 RETURN(-ENOMEM);
1512
1513         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1514         reply->lr_status = rc;
1515
1516         RETURN(rc < 0 ? rc : 0);
1517 }
1518
1519 struct tgt_handler tgt_lfsck_handlers[] = {
1520 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1521 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1522 };
1523 EXPORT_SYMBOL(tgt_lfsck_handlers);
1524
1525 /*
1526  * initialize per-thread page pool (bug 5137).
1527  */
1528 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1529 {
1530         struct tgt_thread_big_cache *tbc;
1531
1532         ENTRY;
1533
1534         LASSERT(thread != NULL);
1535         LASSERT(thread->t_data == NULL);
1536
1537         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1538         if (tbc == NULL)
1539                 RETURN(-ENOMEM);
1540         thread->t_data = tbc;
1541         RETURN(0);
1542 }
1543 EXPORT_SYMBOL(tgt_io_thread_init);
1544
1545 /*
1546  * free per-thread pool created by tgt_thread_init().
1547  */
1548 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1549 {
1550         struct tgt_thread_big_cache *tbc;
1551
1552         ENTRY;
1553
1554         LASSERT(thread != NULL);
1555
1556         /*
1557          * be prepared to handle partially-initialized pools (because this is
1558          * called from ost_io_thread_init() for cleanup.
1559          */
1560         tbc = thread->t_data;
1561         if (tbc != NULL) {
1562                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1563                 thread->t_data = NULL;
1564         }
1565         EXIT;
1566 }
1567 EXPORT_SYMBOL(tgt_io_thread_done);
1568 /**
1569  * Helper function for getting server side [start, start+count] DLM lock
1570  * if asked by client.
1571  */
1572 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1573                     __u64 start, __u64 end, struct lustre_handle *lh,
1574                     int mode, __u64 *flags)
1575 {
1576         union ldlm_policy_data policy;
1577         int rc;
1578
1579         ENTRY;
1580
1581         LASSERT(lh != NULL);
1582         LASSERT(ns != NULL);
1583         LASSERT(!lustre_handle_is_used(lh));
1584
1585         policy.l_extent.gid = 0;
1586         policy.l_extent.start = start & PAGE_MASK;
1587
1588         /*
1589          * If ->o_blocks is EOF it means "lock till the end of the file".
1590          * Otherwise, it's size of an extent or hole being punched (in bytes).
1591          */
1592         if (end == OBD_OBJECT_EOF || end < start)
1593                 policy.l_extent.end = OBD_OBJECT_EOF;
1594         else
1595                 policy.l_extent.end = end | ~PAGE_MASK;
1596
1597         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1598                                     flags, ldlm_blocking_ast,
1599                                     ldlm_completion_ast, ldlm_glimpse_ast,
1600                                     NULL, 0, LVB_T_NONE, NULL, lh);
1601         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1602 }
1603 EXPORT_SYMBOL(tgt_extent_lock);
1604
1605 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1606 {
1607         LASSERT(lustre_handle_is_used(lh));
1608         ldlm_lock_decref(lh, mode);
1609 }
1610 EXPORT_SYMBOL(tgt_extent_unlock);
1611
1612 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1613                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1614                  struct lustre_handle *lh, enum ldlm_mode mode)
1615 {
1616         __u64                    flags = 0;
1617         int                      nrbufs = obj->ioo_bufcnt;
1618         int                      i;
1619
1620         ENTRY;
1621
1622         LASSERT(mode == LCK_PR || mode == LCK_PW);
1623         LASSERT(!lustre_handle_is_used(lh));
1624
1625         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1626                 RETURN(0);
1627
1628         for (i = 1; i < nrbufs; i++)
1629                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1630                         RETURN(-EFAULT);
1631
1632         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1633                                nb[nrbufs - 1].rnb_offset +
1634                                nb[nrbufs - 1].rnb_len - 1,
1635                                lh, mode, &flags));
1636 }
1637
1638 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1639                     struct lustre_handle *lh, enum ldlm_mode mode)
1640 {
1641         ENTRY;
1642
1643         LASSERT(mode == LCK_PR || mode == LCK_PW);
1644         LASSERT((obj->ioo_bufcnt > 0 &&
1645                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1646                 lustre_handle_is_used(lh));
1647
1648         if (lustre_handle_is_used(lh))
1649                 tgt_extent_unlock(lh, mode);
1650         EXIT;
1651 }
1652
1653 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1654                                struct ptlrpc_bulk_desc *desc, int opc,
1655                                cksum_type_t cksum_type)
1656 {
1657         struct cfs_crypto_hash_desc     *hdesc;
1658         unsigned int                    bufsize;
1659         int                             i, err;
1660         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1661         __u32                           cksum;
1662
1663         LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1664
1665         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1666         if (IS_ERR(hdesc)) {
1667                 CERROR("%s: unable to initialize checksum hash %s\n",
1668                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1669                 return PTR_ERR(hdesc);
1670         }
1671
1672         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1673         for (i = 0; i < desc->bd_iov_count; i++) {
1674                 /* corrupt the data before we compute the checksum, to
1675                  * simulate a client->OST data error */
1676                 if (i == 0 && opc == OST_WRITE &&
1677                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1678                         int off = BD_GET_KIOV(desc, i).kiov_offset &
1679                                 ~PAGE_MASK;
1680                         int len = BD_GET_KIOV(desc, i).kiov_len;
1681                         struct page *np = tgt_page_to_corrupt;
1682                         char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1683
1684                         if (np) {
1685                                 char *ptr2 = kmap(np) + off;
1686
1687                                 memcpy(ptr2, ptr, len);
1688                                 memcpy(ptr2, "bad3", min(4, len));
1689                                 kunmap(np);
1690                                 BD_GET_KIOV(desc, i).kiov_page = np;
1691                         } else {
1692                                 CERROR("%s: can't alloc page for corruption\n",
1693                                        tgt_name(tgt));
1694                         }
1695                 }
1696                 cfs_crypto_hash_update_page(hdesc,
1697                                   BD_GET_KIOV(desc, i).kiov_page,
1698                                   BD_GET_KIOV(desc, i).kiov_offset &
1699                                         ~PAGE_MASK,
1700                                   BD_GET_KIOV(desc, i).kiov_len);
1701
1702                  /* corrupt the data after we compute the checksum, to
1703                  * simulate an OST->client data error */
1704                 if (i == 0 && opc == OST_READ &&
1705                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1706                         int off = BD_GET_KIOV(desc, i).kiov_offset
1707                           & ~PAGE_MASK;
1708                         int len = BD_GET_KIOV(desc, i).kiov_len;
1709                         struct page *np = tgt_page_to_corrupt;
1710                         char *ptr =
1711                           kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1712
1713                         if (np) {
1714                                 char *ptr2 = kmap(np) + off;
1715
1716                                 memcpy(ptr2, ptr, len);
1717                                 memcpy(ptr2, "bad4", min(4, len));
1718                                 kunmap(np);
1719                                 BD_GET_KIOV(desc, i).kiov_page = np;
1720                         } else {
1721                                 CERROR("%s: can't alloc page for corruption\n",
1722                                        tgt_name(tgt));
1723                         }
1724                 }
1725         }
1726
1727         bufsize = sizeof(cksum);
1728         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1729
1730         return cksum;
1731 }
1732
1733 int tgt_brw_read(struct tgt_session_info *tsi)
1734 {
1735         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1736         struct ptlrpc_bulk_desc *desc = NULL;
1737         struct obd_export       *exp = tsi->tsi_exp;
1738         struct niobuf_remote    *remote_nb;
1739         struct niobuf_local     *local_nb;
1740         struct obd_ioobj        *ioo;
1741         struct ost_body         *body, *repbody;
1742         struct l_wait_info       lwi;
1743         struct lustre_handle     lockh = { 0 };
1744         int                      npages, nob = 0, rc, i, no_reply = 0;
1745         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1746
1747         ENTRY;
1748
1749         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1750                 CERROR("%s: deny read request from %s to portal %u\n",
1751                        tgt_name(tsi->tsi_tgt),
1752                        obd_export_nid2str(req->rq_export),
1753                        ptlrpc_req2svc(req)->srv_req_portal);
1754                 RETURN(-EPROTO);
1755         }
1756
1757         req->rq_bulk_read = 1;
1758
1759         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1760                 RETURN(-EIO);
1761
1762         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1763                          cfs_fail_val : (obd_timeout + 1) / 4);
1764
1765         /* Check if there is eviction in progress, and if so, wait for it to
1766          * finish */
1767         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1768                 /* We do not care how long it takes */
1769                 lwi = LWI_INTR(NULL, NULL);
1770                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1771                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1772                          &lwi);
1773         }
1774
1775         /* There must be big cache in current thread to process this request
1776          * if it is NULL then something went wrong and it wasn't allocated,
1777          * report -ENOMEM in that case */
1778         if (tbc == NULL)
1779                 RETURN(-ENOMEM);
1780
1781         body = tsi->tsi_ost_body;
1782         LASSERT(body != NULL);
1783
1784         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1785         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1786
1787         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1788         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1789
1790         local_nb = tbc->local;
1791
1792         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1793                           remote_nb, &lockh, LCK_PR);
1794         if (rc != 0)
1795                 RETURN(rc);
1796
1797         /*
1798          * If getting the lock took more time than
1799          * client was willing to wait, drop it. b=11330
1800          */
1801         if (cfs_time_current_sec() > req->rq_deadline ||
1802             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1803                 no_reply = 1;
1804                 CERROR("Dropping timed-out read from %s because locking"
1805                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1806                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1807                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1808                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1809                 GOTO(out_lock, rc = -ETIMEDOUT);
1810         }
1811
1812         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1813         repbody->oa = body->oa;
1814
1815         npages = PTLRPC_MAX_BRW_PAGES;
1816         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1817                         ioo, remote_nb, &npages, local_nb);
1818         if (rc != 0)
1819                 GOTO(out_lock, rc);
1820
1821         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1822                                     PTLRPC_BULK_PUT_SOURCE |
1823                                         PTLRPC_BULK_BUF_KIOV,
1824                                     OST_BULK_PORTAL,
1825                                     &ptlrpc_bulk_kiov_nopin_ops);
1826         if (desc == NULL)
1827                 GOTO(out_commitrw, rc = -ENOMEM);
1828
1829         nob = 0;
1830         for (i = 0; i < npages; i++) {
1831                 int page_rc = local_nb[i].lnb_rc;
1832
1833                 if (page_rc < 0) {
1834                         rc = page_rc;
1835                         break;
1836                 }
1837
1838                 nob += page_rc;
1839                 if (page_rc != 0) { /* some data! */
1840                         LASSERT(local_nb[i].lnb_page != NULL);
1841                         desc->bd_frag_ops->add_kiov_frag
1842                           (desc, local_nb[i].lnb_page,
1843                            local_nb[i].lnb_page_offset,
1844                            page_rc);
1845                 }
1846
1847                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1848                         /* All subsequent pages should be 0 */
1849                         while (++i < npages)
1850                                 LASSERT(local_nb[i].lnb_rc == 0);
1851                         break;
1852                 }
1853         }
1854
1855         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1856                 cksum_type_t cksum_type =
1857                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1858                                           body->oa.o_flags : 0);
1859                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1860                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1861                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1862                                                         OST_READ, cksum_type);
1863                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1864                        repbody->oa.o_cksum);
1865         } else {
1866                 repbody->oa.o_valid = 0;
1867         }
1868         /* We're finishing using body->oa as an input variable */
1869
1870         /* Check if client was evicted while we were doing i/o before touching
1871          * network */
1872         if (likely(rc == 0 &&
1873                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1874                 rc = target_bulk_io(exp, desc, &lwi);
1875                 no_reply = rc != 0;
1876         }
1877
1878 out_commitrw:
1879         /* Must commit after prep above in all cases */
1880         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1881                           remote_nb, npages, local_nb, rc);
1882         if (rc == 0)
1883                 tgt_drop_id(exp, &repbody->oa);
1884 out_lock:
1885         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1886
1887         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1888                 ptlrpc_free_bulk(desc);
1889
1890         LASSERT(rc <= 0);
1891         if (rc == 0) {
1892                 rc = nob;
1893                 ptlrpc_lprocfs_brw(req, nob);
1894         } else if (no_reply) {
1895                 req->rq_no_reply = 1;
1896                 /* reply out callback would free */
1897                 ptlrpc_req_drop_rs(req);
1898                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1899                               "client will retry: rc %d\n",
1900                               exp->exp_obd->obd_name,
1901                               obd_uuid2str(&exp->exp_client_uuid),
1902                               obd_export_nid2str(exp), rc);
1903         }
1904         /* send a bulk after reply to simulate a network delay or reordering
1905          * by a router */
1906         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1907                 wait_queue_head_t        waitq;
1908                 struct l_wait_info       lwi1;
1909
1910                 CDEBUG(D_INFO, "reorder BULK\n");
1911                 init_waitqueue_head(&waitq);
1912
1913                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1914                 l_wait_event(waitq, 0, &lwi1);
1915                 target_bulk_io(exp, desc, &lwi);
1916                 ptlrpc_free_bulk(desc);
1917         }
1918
1919         RETURN(rc);
1920 }
1921 EXPORT_SYMBOL(tgt_brw_read);
1922
1923 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1924                               struct ptlrpc_bulk_desc *desc,
1925                               struct niobuf_local *local_nb, int npages,
1926                               u32 client_cksum, u32 server_cksum,
1927                               bool mmap)
1928 {
1929         struct obd_export *exp = req->rq_export;
1930         struct ost_body *body;
1931         char *router = "";
1932         char *via = "";
1933
1934         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1935         LASSERT(body != NULL);
1936
1937         if (req->rq_peer.nid != desc->bd_sender) {
1938                 via = " via ";
1939                 router = libcfs_nid2str(desc->bd_sender);
1940         }
1941
1942         if (mmap) {
1943                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1944                              client_cksum, server_cksum);
1945                 return;
1946         }
1947
1948         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1949                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1950                            "]: client csum %x, server csum %x\n",
1951                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1952                            via, router,
1953                            body->oa.o_valid & OBD_MD_FLFID ?
1954                            body->oa.o_parent_seq : (__u64)0,
1955                            body->oa.o_valid & OBD_MD_FLFID ?
1956                            body->oa.o_parent_oid : 0,
1957                            body->oa.o_valid & OBD_MD_FLFID ?
1958                            body->oa.o_parent_ver : 0,
1959                            POSTID(&body->oa.o_oi),
1960                            local_nb[0].lnb_file_offset,
1961                            local_nb[npages-1].lnb_file_offset +
1962                            local_nb[npages - 1].lnb_len - 1,
1963                            client_cksum, server_cksum);
1964 }
1965
1966 int tgt_brw_write(struct tgt_session_info *tsi)
1967 {
1968         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1969         struct ptlrpc_bulk_desc *desc = NULL;
1970         struct obd_export       *exp = req->rq_export;
1971         struct niobuf_remote    *remote_nb;
1972         struct niobuf_local     *local_nb;
1973         struct obd_ioobj        *ioo;
1974         struct ost_body         *body, *repbody;
1975         struct l_wait_info       lwi;
1976         struct lustre_handle     lockh = {0};
1977         __u32                   *rcs;
1978         int                      objcount, niocount, npages;
1979         int                      rc, i, j;
1980         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1981         bool                     no_reply = false, mmap;
1982         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1983
1984         ENTRY;
1985
1986         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1987                 CERROR("%s: deny write request from %s to portal %u\n",
1988                        tgt_name(tsi->tsi_tgt),
1989                        obd_export_nid2str(req->rq_export),
1990                        ptlrpc_req2svc(req)->srv_req_portal);
1991                 RETURN(err_serious(-EPROTO));
1992         }
1993
1994         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1995                 RETURN(err_serious(-ENOSPC));
1996         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1997                 RETURN(err_serious(-EROFS));
1998
1999         req->rq_bulk_write = 1;
2000
2001         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2002                 RETURN(err_serious(-EIO));
2003         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2004                 RETURN(err_serious(-EFAULT));
2005
2006         /* pause before transaction has been started */
2007         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2008                          cfs_fail_val : (obd_timeout + 1) / 4);
2009
2010         /* There must be big cache in current thread to process this request
2011          * if it is NULL then something went wrong and it wasn't allocated,
2012          * report -ENOMEM in that case */
2013         if (tbc == NULL)
2014                 RETURN(-ENOMEM);
2015
2016         body = tsi->tsi_ost_body;
2017         LASSERT(body != NULL);
2018
2019         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2020         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2021
2022         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2023                                         RCL_CLIENT) / sizeof(*ioo);
2024
2025         for (niocount = i = 0; i < objcount; i++)
2026                 niocount += ioo[i].ioo_bufcnt;
2027
2028         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2029         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2030         if (niocount != req_capsule_get_size(&req->rq_pill,
2031                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2032                         sizeof(*remote_nb))
2033                 RETURN(err_serious(-EPROTO));
2034
2035         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2036             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2037                 memory_pressure_set();
2038
2039         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2040                              niocount * sizeof(*rcs));
2041         rc = req_capsule_server_pack(&req->rq_pill);
2042         if (rc != 0)
2043                 GOTO(out, rc = err_serious(rc));
2044
2045         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2046         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2047
2048         local_nb = tbc->local;
2049
2050         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2051                           remote_nb, &lockh, LCK_PW);
2052         if (rc != 0)
2053                 GOTO(out, rc);
2054
2055         /*
2056          * If getting the lock took more time than
2057          * client was willing to wait, drop it. b=11330
2058          */
2059         if (cfs_time_current_sec() > req->rq_deadline ||
2060             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2061                 no_reply = true;
2062                 CERROR("%s: Dropping timed-out write from %s because locking "
2063                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
2064                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2065                        POSTID(&ioo->ioo_oid),
2066                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2067                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2068                 GOTO(out_lock, rc = -ETIMEDOUT);
2069         }
2070
2071         /* Because we already sync grant info with client when reconnect,
2072          * grant info will be cleared for resent req, then fed_grant and
2073          * total_grant will not be modified in following preprw_write */
2074         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2075                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2076                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2077         }
2078
2079         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2080         if (repbody == NULL)
2081                 GOTO(out_lock, rc = -ENOMEM);
2082         repbody->oa = body->oa;
2083
2084         npages = PTLRPC_MAX_BRW_PAGES;
2085         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2086                         objcount, ioo, remote_nb, &npages, local_nb);
2087         if (rc < 0)
2088                 GOTO(out_lock, rc);
2089
2090         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2091                                     PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2092                                     OST_BULK_PORTAL,
2093                                     &ptlrpc_bulk_kiov_nopin_ops);
2094         if (desc == NULL)
2095                 GOTO(skip_transfer, rc = -ENOMEM);
2096
2097         /* NB Having prepped, we must commit... */
2098         for (i = 0; i < npages; i++)
2099                 desc->bd_frag_ops->add_kiov_frag(desc,
2100                                                  local_nb[i].lnb_page,
2101                                                  local_nb[i].lnb_page_offset,
2102                                                  local_nb[i].lnb_len);
2103
2104         rc = sptlrpc_svc_prep_bulk(req, desc);
2105         if (rc != 0)
2106                 GOTO(skip_transfer, rc);
2107
2108         rc = target_bulk_io(exp, desc, &lwi);
2109         no_reply = rc != 0;
2110
2111 skip_transfer:
2112         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2113                 static int cksum_counter;
2114
2115                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2116                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2117
2118                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2119                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2120                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2121                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2122                                                         OST_WRITE, cksum_type);
2123                 cksum_counter++;
2124
2125                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2126                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2127                                 body->oa.o_flags & OBD_FL_MMAP);
2128
2129                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2130                                           body->oa.o_cksum,
2131                                           repbody->oa.o_cksum, mmap);
2132                         cksum_counter = 0;
2133                 } else if ((cksum_counter & (-cksum_counter)) ==
2134                            cksum_counter) {
2135                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2136                                cksum_counter, libcfs_id2str(req->rq_peer),
2137                                repbody->oa.o_cksum);
2138                 }
2139         }
2140
2141         /* Must commit after prep above in all cases */
2142         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2143                           objcount, ioo, remote_nb, npages, local_nb, rc);
2144         if (rc == -ENOTCONN)
2145                 /* quota acquire process has been given up because
2146                  * either the client has been evicted or the client
2147                  * has timed out the request already */
2148                 no_reply = true;
2149
2150         /*
2151          * Disable sending mtime back to the client. If the client locked the
2152          * whole object, then it has already updated the mtime on its side,
2153          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2154          */
2155         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2156
2157         if (rc == 0) {
2158                 int nob = 0;
2159
2160                 /* set per-requested niobuf return codes */
2161                 for (i = j = 0; i < niocount; i++) {
2162                         int len = remote_nb[i].rnb_len;
2163
2164                         nob += len;
2165                         rcs[i] = 0;
2166                         do {
2167                                 LASSERT(j < npages);
2168                                 if (local_nb[j].lnb_rc < 0)
2169                                         rcs[i] = local_nb[j].lnb_rc;
2170                                 len -= local_nb[j].lnb_len;
2171                                 j++;
2172                         } while (len > 0);
2173                         LASSERT(len == 0);
2174                 }
2175                 LASSERT(j == npages);
2176                 ptlrpc_lprocfs_brw(req, nob);
2177
2178                 tgt_drop_id(exp, &repbody->oa);
2179         }
2180 out_lock:
2181         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2182         if (desc)
2183                 ptlrpc_free_bulk(desc);
2184 out:
2185         if (no_reply) {
2186                 req->rq_no_reply = 1;
2187                 /* reply out callback would free */
2188                 ptlrpc_req_drop_rs(req);
2189                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2190                               "client will retry: rc %d\n",
2191                               exp->exp_obd->obd_name,
2192                               obd_uuid2str(&exp->exp_client_uuid),
2193                               obd_export_nid2str(exp), rc);
2194         }
2195         memory_pressure_clr();
2196         RETURN(rc);
2197 }
2198 EXPORT_SYMBOL(tgt_brw_write);
2199
2200 /* Check if request can be reconstructed from saved reply data
2201  * A copy of the reply data is returned in @trd if the pointer is not NULL
2202  */
2203 bool req_can_reconstruct(struct ptlrpc_request *req,
2204                          struct tg_reply_data *trd)
2205 {
2206         struct tg_export_data *ted = &req->rq_export->exp_target_data;
2207         struct lsd_client_data *lcd = ted->ted_lcd;
2208         bool found;
2209
2210         if (tgt_is_multimodrpcs_client(req->rq_export))
2211                 return tgt_lookup_reply(req, trd);
2212
2213         mutex_lock(&ted->ted_lcd_lock);
2214         found = req->rq_xid == lcd->lcd_last_xid ||
2215                 req->rq_xid == lcd->lcd_last_close_xid;
2216
2217         if (found && trd != NULL) {
2218                 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2219                         trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2220                         trd->trd_reply.lrd_transno =
2221                                                 lcd->lcd_last_close_transno;
2222                         trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2223                 } else {
2224                         trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2225                         trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2226                         trd->trd_reply.lrd_result = lcd->lcd_last_result;
2227                         trd->trd_reply.lrd_data = lcd->lcd_last_data;
2228                         trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2229                         trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2230                         trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2231                         trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2232                 }
2233         }
2234         mutex_unlock(&ted->ted_lcd_lock);
2235
2236         return found;
2237 }
2238 EXPORT_SYMBOL(req_can_reconstruct);
2239