lustre/target/tgt_handler.c

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