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, 2012, 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 <obd.h>
  38 #include <obd_class.h>
  39 #include <obd_cksum.h>
  40 #include <md_object.h>
  41 #include <lustre_lfsck.h>
  42 #include <lustre_nodemap.h>
  43
  44 #include "tgt_internal.h"
  45
  46 char *tgt_name(struct lu_target *tgt)
  47 {
  48         LASSERT(tgt->lut_obd != NULL);
  49         return tgt->lut_obd->obd_name;
  50 }
  51 EXPORT_SYMBOL(tgt_name);
  52
  53 /*
  54  * Generic code handling requests that have struct mdt_body passed in:
  55  *
  56  *  - extract mdt_body from request and save it in @tsi, if present;
  57  *
  58  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  59  *  @tsi;
  60  *
  61  *  - if HABEO_CORPUS flag is set for this request type check whether object
  62  *  actually exists on storage (lu_object_exists()).
  63  *
  64  */
  65 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  66 {
  67         const struct mdt_body   *body;
  68         struct lu_object        *obj;
  69         struct req_capsule      *pill = tsi->tsi_pill;
  70         int                      rc;
  71
  72         ENTRY;
  73
  74         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  75         if (body == NULL)
  76                 RETURN(-EFAULT);
  77
  78         tsi->tsi_mdt_body = body;
  79
  80         if (!(body->mbo_valid & OBD_MD_FLID))
  81                 RETURN(0);
  82
  83         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  84          * in any case in pre-2.5 clients. Fix that here if needed */
  85         if (unlikely(fid_is_zero(&body->mbo_fid1)))
  86                 RETURN(0);
  87
  88         if (!fid_is_sane(&body->mbo_fid1)) {
  89                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  90                        PFID(&body->mbo_fid1));
  91                 RETURN(-EINVAL);
  92         }
  93
  94         obj = lu_object_find(tsi->tsi_env,
  95                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
  96                              &body->mbo_fid1, NULL);
  97         if (!IS_ERR(obj)) {
  98                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
  99                         lu_object_put(tsi->tsi_env, obj);
 100                         /* for capability renew ENOENT will be handled in
 101                          * mdt_renew_capa */
 102                         if (body->mbo_valid & OBD_MD_FLOSSCAPA)
 103                                 rc = 0;
 104                         else
 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         obd_seq          seq    = ostid_seq(oi);
 138         obd_id           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 lustre_capa      *capa;
 246         struct lu_nodemap       *nodemap;
 247         int                      rc;
 248
 249         ENTRY;
 250
 251         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 252         if (body == NULL)
 253                 RETURN(-EFAULT);
 254
 255         rc = tgt_validate_obdo(tsi, &body->oa);
 256         if (rc)
 257                 RETURN(rc);
 258
 259         nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
 260
 261         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 262                                         NODEMAP_CLIENT_TO_FS,
 263                                         body->oa.o_uid);
 264         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 265                                         NODEMAP_CLIENT_TO_FS,
 266                                         body->oa.o_gid);
 267
 268         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
 269                 capa = req_capsule_client_get(pill, &RMF_CAPA1);
 270                 if (capa == NULL) {
 271                         CERROR("%s: OSSCAPA flag is set without capability\n",
 272                                tgt_name(tsi->tsi_tgt));
 273                         RETURN(-EFAULT);
 274                 }
 275         }
 276
 277         tsi->tsi_ost_body = body;
 278         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 279
 280         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 281                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 282                 if (rc < 0)
 283                         RETURN(rc);
 284         }
 285
 286         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 287                 if (flags & HABEO_CORPUS) {
 288                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 289                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 290                                tgt_name(tsi->tsi_tgt));
 291                         RETURN(-EPROTO);
 292                 } else {
 293                         RETURN(0);
 294                 }
 295         }
 296
 297         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 298
 299         /*
 300          * OST doesn't get object in advance for further use to prevent
 301          * situations with nested object_find which is potential deadlock.
 302          */
 303         tsi->tsi_corpus = NULL;
 304         RETURN(rc);
 305 }
 306
 307 /*
 308  * Do necessary preprocessing according to handler ->th_flags.
 309  */
 310 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 311                                   struct tgt_handler *h,
 312                                   struct ptlrpc_request *req)
 313 {
 314         struct req_capsule      *pill = tsi->tsi_pill;
 315         __u32                    flags = h->th_flags;
 316         int                      rc = 0;
 317
 318         ENTRY;
 319
 320         if (tsi->tsi_preprocessed)
 321                 RETURN(0);
 322
 323         LASSERT(h->th_act != NULL);
 324         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 325         LASSERT(current->journal_info == NULL);
 326
 327         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 328                      h->th_fmt != NULL));
 329         if (h->th_fmt != NULL) {
 330                 req_capsule_set(pill, h->th_fmt);
 331                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 332                         rc = tgt_mdt_body_unpack(tsi, flags);
 333                         if (rc < 0)
 334                                 RETURN(rc);
 335                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 336                                                  RCL_CLIENT)) {
 337                         rc = tgt_ost_body_unpack(tsi, flags);
 338                         if (rc < 0)
 339                                 RETURN(rc);
 340                 }
 341         }
 342
 343         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 344                 RETURN(-EROFS);
 345
 346         if (flags & HABEO_CLAVIS) {
 347                 struct ldlm_request *dlm_req;
 348
 349                 LASSERT(h->th_fmt != NULL);
 350
 351                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 352                 if (dlm_req != NULL) {
 353                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 354                                      LDLM_IBITS &&
 355                                      dlm_req->lock_desc.l_policy_data.\
 356                                      l_inodebits.bits == 0)) {
 357                                 /*
 358                                  * Lock without inodebits makes no sense and
 359                                  * will oops later in ldlm. If client miss to
 360                                  * set such bits, do not trigger ASSERTION.
 361                                  *
 362                                  * For liblustre flock case, it maybe zero.
 363                                  */
 364                                 rc = -EPROTO;
 365                         } else {
 366                                 tsi->tsi_dlm_req = dlm_req;
 367                         }
 368                 } else {
 369                         rc = -EFAULT;
 370                 }
 371         }
 372         tsi->tsi_preprocessed = 1;
 373         RETURN(rc);
 374 }
 375
 376 /*
 377  * Invoke handler for this request opc. Also do necessary preprocessing
 378  * (according to handler ->th_flags), and post-processing (setting of
 379  * ->last_{xid,committed}).
 380  */
 381 static int tgt_handle_request0(struct tgt_session_info *tsi,
 382                                struct tgt_handler *h,
 383                                struct ptlrpc_request *req)
 384 {
 385         int      serious = 0;
 386         int      rc;
 387
 388         ENTRY;
 389
 390         /*
 391          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 392          * to put same checks into handlers like mdt_close(), mdt_reint(),
 393          * etc., without talking to mdt authors first. Checking same thing
 394          * there again is useless and returning 0 error without packing reply
 395          * is buggy! Handlers either pack reply or return error.
 396          *
 397          * We return 0 here and do not send any reply in order to emulate
 398          * network failure. Do not send any reply in case any of NET related
 399          * fail_id has occured.
 400          */
 401         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 402                 RETURN(0);
 403
 404         rc = tgt_request_preprocess(tsi, h, req);
 405         /* pack reply if reply format is fixed */
 406         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 407                 /* Pack reply */
 408                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 409                                           RCL_SERVER))
 410                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 411                                              RCL_SERVER,
 412                                              tsi->tsi_mdt_body->mbo_eadatasize);
 413                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 414                                           RCL_SERVER))
 415                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 416                                              RCL_SERVER, 0);
 417
 418                 rc = req_capsule_server_pack(tsi->tsi_pill);
 419         }
 420
 421         if (likely(rc == 0)) {
 422                 /*
 423                  * Process request, there can be two types of rc:
 424                  * 1) errors with msg unpack/pack, other failures outside the
 425                  * operation itself. This is counted as serious errors;
 426                  * 2) errors during fs operation, should be placed in rq_status
 427                  * only
 428                  */
 429                 rc = h->th_act(tsi);
 430                 if (!is_serious(rc) &&
 431                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 432                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 433                                   "pack reply and returned 0 error\n",
 434                                   tgt_name(tsi->tsi_tgt), h->th_name);
 435                         LBUG();
 436                 }
 437                 serious = is_serious(rc);
 438                 rc = clear_serious(rc);
 439         } else {
 440                 serious = 1;
 441         }
 442
 443         req->rq_status = rc;
 444
 445         /*
 446          * ELDLM_* codes which > 0 should be in rq_status only as well as
 447          * all non-serious errors.
 448          */
 449         if (rc > 0 || !serious)
 450                 rc = 0;
 451
 452         LASSERT(current->journal_info == NULL);
 453
 454         if (likely(rc == 0 && req->rq_export))
 455                 target_committed_to_req(req);
 456
 457         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 458         RETURN(0);
 459 }
 460
 461 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 462                                        struct obd_device *obd, int *process)
 463 {
 464         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 465         case MDS_DISCONNECT:
 466         case OST_DISCONNECT:
 467         case OBD_IDX_READ:
 468                 *process = 1;
 469                 RETURN(0);
 470         case MDS_CLOSE:
 471         case MDS_DONE_WRITING:
 472         case MDS_SYNC: /* used in unmounting */
 473         case OBD_PING:
 474         case MDS_REINT:
 475         case OUT_UPDATE:
 476         case SEQ_QUERY:
 477         case FLD_QUERY:
 478         case LDLM_ENQUEUE:
 479         case OST_CREATE:
 480         case OST_DESTROY:
 481         case OST_PUNCH:
 482         case OST_SETATTR:
 483         case OST_SYNC:
 484         case OST_WRITE:
 485                 *process = target_queue_recovery_request(req, obd);
 486                 RETURN(0);
 487
 488         default:
 489                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 490                 *process = -EAGAIN;
 491                 RETURN(0);
 492         }
 493 }
 494
 495 /*
 496  * Handle recovery. Return:
 497  *        +1: continue request processing;
 498  *       -ve: abort immediately with the given error code;
 499  *         0: send reply with error code in req->rq_status;
 500  */
 501 int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 502 {
 503         ENTRY;
 504
 505         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 506         case MDS_CONNECT:
 507         case OST_CONNECT:
 508         case MGS_CONNECT:
 509         case SEC_CTX_INIT:
 510         case SEC_CTX_INIT_CONT:
 511         case SEC_CTX_FINI:
 512                 RETURN(+1);
 513         }
 514
 515         if (!req->rq_export->exp_obd->obd_replayable)
 516                 RETURN(+1);
 517
 518         /* sanity check: if the xid matches, the request must be marked as a
 519          * resent or replayed */
 520         if (req_xid_is_last(req)) {
 521                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 522                       (MSG_RESENT | MSG_REPLAY))) {
 523                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 524                                   "last_xid, expected REPLAY or RESENT flag "
 525                                   "(%x)", req->rq_xid,
 526                                   lustre_msg_get_flags(req->rq_reqmsg));
 527                         req->rq_status = -ENOTCONN;
 528                         RETURN(-ENOTCONN);
 529                 }
 530         }
 531         /* else: note the opposite is not always true; a RESENT req after a
 532          * failover will usually not match the last_xid, since it was likely
 533          * never committed. A REPLAYed request will almost never match the
 534          * last xid, however it could for a committed, but still retained,
 535          * open. */
 536
 537         /* Check for aborted recovery... */
 538         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 539                 int rc;
 540                 int should_process;
 541
 542                 DEBUG_REQ(D_INFO, req, "Got new replay");
 543                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 544                                                  &should_process);
 545                 if (rc != 0 || !should_process)
 546                         RETURN(rc);
 547                 else if (should_process < 0) {
 548                         req->rq_status = should_process;
 549                         rc = ptlrpc_error(req);
 550                         RETURN(rc);
 551                 }
 552         }
 553         RETURN(+1);
 554 }
 555
 556 /* Initial check for request, it is validation mostly */
 557 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 558 {
 559         struct tgt_handler      *h;
 560         struct tgt_opc_slice    *s;
 561         struct lu_target        *tgt;
 562         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 563
 564         ENTRY;
 565
 566         tgt = class_exp2tgt(req->rq_export);
 567
 568         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 569                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 570                         break;
 571
 572         /* opcode was not found in slice */
 573         if (unlikely(s->tos_hs == NULL)) {
 574                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 575                        opc);
 576                 RETURN(ERR_PTR(-ENOTSUPP));
 577         }
 578
 579         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 580         h = s->tos_hs + (opc - s->tos_opc_start);
 581         if (unlikely(h->th_opc == 0)) {
 582                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 583                 RETURN(ERR_PTR(-ENOTSUPP));
 584         }
 585
 586         RETURN(h);
 587 }
 588
 589 int tgt_request_handle(struct ptlrpc_request *req)
 590 {
 591         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 592
 593         struct lustre_msg       *msg = req->rq_reqmsg;
 594         struct tgt_handler      *h;
 595         struct lu_target        *tgt;
 596         int                      request_fail_id = 0;
 597         __u32                    opc = lustre_msg_get_opc(msg);
 598         int                      rc;
 599
 600         ENTRY;
 601
 602         /* Refill the context, to make sure all thread keys are allocated */
 603         lu_env_refill(req->rq_svc_thread->t_env);
 604
 605         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 606         tsi->tsi_pill = &req->rq_pill;
 607         tsi->tsi_env = req->rq_svc_thread->t_env;
 608
 609         /* if request has export then get handlers slice from corresponding
 610          * target, otherwise that should be connect operation */
 611         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 612             opc == MGS_CONNECT) {
 613                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 614                 rc = target_handle_connect(req);
 615                 if (rc != 0) {
 616                         rc = ptlrpc_error(req);
 617                         GOTO(out, rc);
 618                 }
 619                 /* recovery-small test 18c asks to drop connect reply */
 620                 if (unlikely(opc == OST_CONNECT &&
 621                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 622                         GOTO(out, rc = 0);
 623         }
 624
 625         if (unlikely(!class_connected_export(req->rq_export))) {
 626                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 627                        opc, libcfs_id2str(req->rq_peer));
 628                 req->rq_status = -ENOTCONN;
 629                 rc = ptlrpc_error(req);
 630                 GOTO(out, rc);
 631         }
 632
 633         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 634         tsi->tsi_exp = req->rq_export;
 635         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 636                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 637         else
 638                 tsi->tsi_jobid = NULL;
 639
 640         request_fail_id = tgt->lut_request_fail_id;
 641         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 642
 643         h = tgt_handler_find_check(req);
 644         if (IS_ERR(h)) {
 645                 req->rq_status = PTR_ERR(h);
 646                 rc = ptlrpc_error(req);
 647                 GOTO(out, rc);
 648         }
 649
 650         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 651                 GOTO(out, rc = 0);
 652
 653         rc = lustre_msg_check_version(msg, h->th_version);
 654         if (unlikely(rc)) {
 655                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 656                           " %08x, expecting %08x\n", tgt_name(tgt),
 657                           lustre_msg_get_version(msg), h->th_version);
 658                 req->rq_status = -EINVAL;
 659                 rc = ptlrpc_error(req);
 660                 GOTO(out, rc);
 661         }
 662
 663         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 664         if (likely(rc == 1)) {
 665                 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 666                          h->th_opc, opc);
 667                 rc = tgt_handle_request0(tsi, h, req);
 668                 if (rc)
 669                         GOTO(out, rc);
 670         }
 671         EXIT;
 672 out:
 673         req_capsule_fini(tsi->tsi_pill);
 674         if (tsi->tsi_corpus != NULL) {
 675                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 676                 tsi->tsi_corpus = NULL;
 677         }
 678         return rc;
 679 }
 680 EXPORT_SYMBOL(tgt_request_handle);
 681
 682 /** Assign high priority operations to the request if needed. */
 683 int tgt_hpreq_handler(struct ptlrpc_request *req)
 684 {
 685         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 686         struct tgt_handler      *h;
 687         int                      rc;
 688
 689         ENTRY;
 690
 691         if (req->rq_export == NULL)
 692                 RETURN(0);
 693
 694         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 695         tsi->tsi_pill = &req->rq_pill;
 696         tsi->tsi_env = req->rq_svc_thread->t_env;
 697         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 698         tsi->tsi_exp = req->rq_export;
 699
 700         h = tgt_handler_find_check(req);
 701         if (IS_ERR(h)) {
 702                 rc = PTR_ERR(h);
 703                 RETURN(rc);
 704         }
 705
 706         rc = tgt_request_preprocess(tsi, h, req);
 707         if (unlikely(rc != 0))
 708                 RETURN(rc);
 709
 710         if (h->th_hp != NULL)
 711                 h->th_hp(tsi);
 712         RETURN(0);
 713 }
 714 EXPORT_SYMBOL(tgt_hpreq_handler);
 715
 716 void tgt_counter_incr(struct obd_export *exp, int opcode)
 717 {
 718         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 719         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 720                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 721 }
 722 EXPORT_SYMBOL(tgt_counter_incr);
 723
 724 /*
 725  * Unified target generic handlers.
 726  */
 727
 728 /*
 729  * Security functions
 730  */
 731 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 732 {
 733         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 734                                       OBD_CONNECT_RMT_CLIENT_FORCE |
 735                                       OBD_CONNECT_MDS_CAPA |
 736                                       OBD_CONNECT_OSS_CAPA);
 737 }
 738
 739 static int tgt_init_sec_level(struct ptlrpc_request *req)
 740 {
 741         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 742         char                    *client = libcfs_nid2str(req->rq_peer.nid);
 743         struct obd_connect_data *data, *reply;
 744         int                      rc = 0;
 745         bool                     remote;
 746
 747         ENTRY;
 748
 749         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 750         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 751         if (data == NULL || reply == NULL)
 752                 RETURN(-EFAULT);
 753
 754         /* connection from MDT is always trusted */
 755         if (req->rq_auth_usr_mdt) {
 756                 tgt_init_sec_none(reply);
 757                 RETURN(0);
 758         }
 759
 760         /* no GSS support case */
 761         if (!req->rq_auth_gss) {
 762                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 763                         CWARN("client %s -> target %s does not use GSS, "
 764                               "can not run under security level %d.\n",
 765                               client, tgt_name(tgt), tgt->lut_sec_level);
 766                         RETURN(-EACCES);
 767                 } else {
 768                         tgt_init_sec_none(reply);
 769                         RETURN(0);
 770                 }
 771         }
 772
 773         /* old version case */
 774         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 775                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 776                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 777                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 778                         CWARN("client %s -> target %s uses old version, "
 779                               "can not run under security level %d.\n",
 780                               client, tgt_name(tgt), tgt->lut_sec_level);
 781                         RETURN(-EACCES);
 782                 } else {
 783                         CWARN("client %s -> target %s uses old version, "
 784                               "run under security level %d.\n",
 785                               client, tgt_name(tgt), tgt->lut_sec_level);
 786                         tgt_init_sec_none(reply);
 787                         RETURN(0);
 788                 }
 789         }
 790
 791         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 792         if (remote) {
 793                 if (!req->rq_auth_remote)
 794                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 795                                "asked to be remote.\n", client, tgt_name(tgt));
 796         } else if (req->rq_auth_remote) {
 797                 remote = true;
 798                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 799                        "as remote by default.\n", client, tgt_name(tgt));
 800         }
 801
 802         if (remote) {
 803                 if (!tgt->lut_oss_capa) {
 804                         CDEBUG(D_SEC,
 805                                "client %s -> target %s is set as remote,"
 806                                " but OSS capabilities are not enabled: %d.\n",
 807                                client, tgt_name(tgt), tgt->lut_oss_capa);
 808                         RETURN(-EACCES);
 809                 }
 810         } else {
 811                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 812                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 813                                "authenticated!\n", client, tgt_name(tgt));
 814                         RETURN(-EACCES);
 815                 }
 816         }
 817
 818
 819         switch (tgt->lut_sec_level) {
 820         case LUSTRE_SEC_NONE:
 821                 if (remote) {
 822                         CDEBUG(D_SEC,
 823                                "client %s -> target %s is set as remote, "
 824                                "can not run under security level %d.\n",
 825                                client, tgt_name(tgt), tgt->lut_sec_level);
 826                         RETURN(-EACCES);
 827                 }
 828                 tgt_init_sec_none(reply);
 829                 break;
 830         case LUSTRE_SEC_REMOTE:
 831                 if (!remote)
 832                         tgt_init_sec_none(reply);
 833                 break;
 834         case LUSTRE_SEC_ALL:
 835                 if (remote)
 836                         break;
 837                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 838                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 839                 if (!tgt->lut_oss_capa)
 840                         reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 841                 if (!tgt->lut_mds_capa)
 842                         reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 843                 break;
 844         default:
 845                 RETURN(-EINVAL);
 846         }
 847
 848         RETURN(rc);
 849 }
 850
 851 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 852 {
 853         struct lu_target        *tgt = class_exp2tgt(exp);
 854         struct sptlrpc_flavor    flvr;
 855         int                      rc = 0;
 856
 857         LASSERT(tgt);
 858         LASSERT(tgt->lut_obd);
 859         LASSERT(tgt->lut_slice);
 860
 861         /* always allow ECHO client */
 862         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 863                             LUSTRE_ECHO_NAME) == 0)) {
 864                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 865                 return 0;
 866         }
 867
 868         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 869                 read_lock(&tgt->lut_sptlrpc_lock);
 870                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 871                                              req->rq_sp_from,
 872                                              req->rq_peer.nid,
 873                                              &flvr);
 874                 read_unlock(&tgt->lut_sptlrpc_lock);
 875
 876                 spin_lock(&exp->exp_lock);
 877                 exp->exp_sp_peer = req->rq_sp_from;
 878                 exp->exp_flvr = flvr;
 879                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 880                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 881                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 882                                "expect %x\n", tgt_name(tgt),
 883                                req->rq_flvr.sf_rpc,
 884                                libcfs_nid2str(req->rq_peer.nid),
 885                                exp->exp_flvr.sf_rpc);
 886                         rc = -EACCES;
 887                 }
 888                 spin_unlock(&exp->exp_lock);
 889         } else {
 890                 if (exp->exp_sp_peer != req->rq_sp_from) {
 891                         CERROR("%s: RPC source %s doesn't match %s\n",
 892                                tgt_name(tgt),
 893                                sptlrpc_part2name(req->rq_sp_from),
 894                                sptlrpc_part2name(exp->exp_sp_peer));
 895                         rc = -EACCES;
 896                 } else {
 897                         rc = sptlrpc_target_export_check(exp, req);
 898                 }
 899         }
 900
 901         return rc;
 902 }
 903
 904 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 905 {
 906         struct sptlrpc_rule_set  tmp_rset;
 907         int                      rc;
 908
 909         sptlrpc_rule_set_init(&tmp_rset);
 910         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 911         if (rc) {
 912                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 913                        tgt_name(tgt), rc);
 914                 return rc;
 915         }
 916
 917         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 918
 919         write_lock(&tgt->lut_sptlrpc_lock);
 920         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 921         tgt->lut_sptlrpc_rset = tmp_rset;
 922         write_unlock(&tgt->lut_sptlrpc_lock);
 923
 924         return 0;
 925 }
 926 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 927
 928 int tgt_connect(struct tgt_session_info *tsi)
 929 {
 930         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 931         struct obd_connect_data *reply;
 932         int                      rc;
 933
 934         ENTRY;
 935
 936         rc = tgt_init_sec_level(req);
 937         if (rc != 0)
 938                 GOTO(out, rc);
 939
 940         /* XXX: better to call this check right after getting new export but
 941          * before last_rcvd slot allocation to avoid server load upon insecure
 942          * connects. This is to be fixed after unifiyng all targets.
 943          */
 944         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
 945         if (rc)
 946                 GOTO(out, rc);
 947
 948         /* To avoid exposing partially initialized connection flags, changes up
 949          * to this point have been staged in reply->ocd_connect_flags. Now that
 950          * connection handling has completed successfully, atomically update
 951          * the connect flags in the shared export data structure. LU-1623 */
 952         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
 953         spin_lock(&tsi->tsi_exp->exp_lock);
 954         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
 955         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
 956         spin_unlock(&tsi->tsi_exp->exp_lock);
 957
 958         RETURN(0);
 959 out:
 960         obd_disconnect(class_export_get(tsi->tsi_exp));
 961         return rc;
 962 }
 963 EXPORT_SYMBOL(tgt_connect);
 964
 965 int tgt_disconnect(struct tgt_session_info *tsi)
 966 {
 967         int rc;
 968
 969         ENTRY;
 970
 971         rc = target_handle_disconnect(tgt_ses_req(tsi));
 972         if (rc)
 973                 RETURN(err_serious(rc));
 974
 975         RETURN(rc);
 976 }
 977 EXPORT_SYMBOL(tgt_disconnect);
 978
 979 /*
 980  * Unified target OBD handlers
 981  */
 982 int tgt_obd_ping(struct tgt_session_info *tsi)
 983 {
 984         int rc;
 985
 986         ENTRY;
 987
 988         rc = target_handle_ping(tgt_ses_req(tsi));
 989         if (rc)
 990                 RETURN(err_serious(rc));
 991
 992         RETURN(rc);
 993 }
 994 EXPORT_SYMBOL(tgt_obd_ping);
 995
 996 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
 997 {
 998         return err_serious(-EOPNOTSUPP);
 999 }
1000 EXPORT_SYMBOL(tgt_obd_log_cancel);
1001
1002 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
1003 {
1004         return err_serious(-EOPNOTSUPP);
1005 }
1006 EXPORT_SYMBOL(tgt_obd_qc_callback);
1007
1008 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1009 {
1010         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1011         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1012         struct obd_export       *exp = req->rq_export;
1013         struct ptlrpc_bulk_desc *desc;
1014         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1015         int                      tmpcount;
1016         int                      tmpsize;
1017         int                      i;
1018         int                      rc;
1019
1020         ENTRY;
1021
1022         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1023                                     MDS_BULK_PORTAL);
1024         if (desc == NULL)
1025                 RETURN(-ENOMEM);
1026
1027         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1028                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1029                  * which is rdpg->rp_count */
1030                 nob = rdpg->rp_count;
1031
1032         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1033              i++, tmpcount -= tmpsize) {
1034                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1035                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1036         }
1037
1038         LASSERT(desc->bd_nob == nob);
1039         rc = target_bulk_io(exp, desc, lwi);
1040         ptlrpc_free_bulk_pin(desc);
1041         RETURN(rc);
1042 }
1043 EXPORT_SYMBOL(tgt_sendpage);
1044
1045 /*
1046  * OBD_IDX_READ handler
1047  */
1048 int tgt_obd_idx_read(struct tgt_session_info *tsi)
1049 {
1050         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1051         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1052         struct idx_info         *req_ii, *rep_ii;
1053         int                      rc, i;
1054
1055         ENTRY;
1056
1057         memset(rdpg, 0, sizeof(*rdpg));
1058         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1059
1060         /* extract idx_info buffer from request & reply */
1061         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1062         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1063                 RETURN(err_serious(-EPROTO));
1064
1065         rc = req_capsule_server_pack(tsi->tsi_pill);
1066         if (rc)
1067                 RETURN(err_serious(rc));
1068
1069         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1070         if (rep_ii == NULL)
1071                 RETURN(err_serious(-EFAULT));
1072         rep_ii->ii_magic = IDX_INFO_MAGIC;
1073
1074         /* extract hash to start with */
1075         rdpg->rp_hash = req_ii->ii_hash_start;
1076
1077         /* extract requested attributes */
1078         rdpg->rp_attrs = req_ii->ii_attrs;
1079
1080         /* check that fid packed in request is valid and supported */
1081         if (!fid_is_sane(&req_ii->ii_fid))
1082                 RETURN(-EINVAL);
1083         rep_ii->ii_fid = req_ii->ii_fid;
1084
1085         /* copy flags */
1086         rep_ii->ii_flags = req_ii->ii_flags;
1087
1088         /* compute number of pages to allocate, ii_count is the number of 4KB
1089          * containers */
1090         if (req_ii->ii_count <= 0)
1091                 GOTO(out, rc = -EFAULT);
1092         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1093                                exp_max_brw_size(tsi->tsi_exp));
1094         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1095
1096         /* allocate pages to store the containers */
1097         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1098         if (rdpg->rp_pages == NULL)
1099                 GOTO(out, rc = -ENOMEM);
1100         for (i = 0; i < rdpg->rp_npages; i++) {
1101                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1102                 if (rdpg->rp_pages[i] == NULL)
1103                         GOTO(out, rc = -ENOMEM);
1104         }
1105
1106         /* populate pages with key/record pairs */
1107         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1108         if (rc < 0)
1109                 GOTO(out, rc);
1110
1111         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1112                  "asked %d > %d\n", rc, rdpg->rp_count);
1113
1114         /* send pages to client */
1115         rc = tgt_sendpage(tsi, rdpg, rc);
1116         if (rc)
1117                 GOTO(out, rc);
1118         EXIT;
1119 out:
1120         if (rdpg->rp_pages) {
1121                 for (i = 0; i < rdpg->rp_npages; i++)
1122                         if (rdpg->rp_pages[i])
1123                                 __free_page(rdpg->rp_pages[i]);
1124                 OBD_FREE(rdpg->rp_pages,
1125                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1126         }
1127         return rc;
1128 }
1129 EXPORT_SYMBOL(tgt_obd_idx_read);
1130
1131 struct tgt_handler tgt_obd_handlers[] = {
1132 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1133 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1134 TGT_OBD_HDL_VAR(0,      OBD_QC_CALLBACK,        tgt_obd_qc_callback),
1135 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1136 };
1137 EXPORT_SYMBOL(tgt_obd_handlers);
1138
1139 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1140              struct dt_object *obj, __u64 start, __u64 end)
1141 {
1142         int rc = 0;
1143
1144         ENTRY;
1145
1146         /* if no objid is specified, it means "sync whole filesystem" */
1147         if (obj == NULL) {
1148                 rc = dt_sync(env, tgt->lut_bottom);
1149         } else if (dt_version_get(env, obj) >
1150                    tgt->lut_obd->obd_last_committed) {
1151                 rc = dt_object_sync(env, obj, start, end);
1152         }
1153
1154         RETURN(rc);
1155 }
1156 EXPORT_SYMBOL(tgt_sync);
1157 /*
1158  * Unified target DLM handlers.
1159  */
1160
1161 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1162  * (if requested) */
1163 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1164                      void *data, int flag)
1165 {
1166         struct lu_env            env;
1167         struct lu_target        *tgt;
1168         struct dt_object        *obj;
1169         struct lu_fid            fid;
1170         int                      rc = 0;
1171
1172         ENTRY;
1173
1174         tgt = class_exp2tgt(lock->l_export);
1175
1176         if (flag == LDLM_CB_CANCELING &&
1177             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1178             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1179              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1180               lock->l_flags & LDLM_FL_CBPENDING))) {
1181                 __u64 start = 0;
1182                 __u64 end = OBD_OBJECT_EOF;
1183
1184                 rc = lu_env_init(&env, LCT_DT_THREAD);
1185                 if (unlikely(rc != 0))
1186                         RETURN(rc);
1187
1188                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1189                                    tgt->lut_lsd.lsd_osd_index);
1190                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1191                 if (IS_ERR(obj))
1192                         GOTO(err_env, rc = PTR_ERR(obj));
1193
1194                 if (!dt_object_exists(obj))
1195                         GOTO(err_put, rc = -ENOENT);
1196
1197                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1198                         start = lock->l_policy_data.l_extent.start;
1199                         end = lock->l_policy_data.l_extent.end;
1200                 }
1201
1202                 rc = tgt_sync(&env, tgt, obj, start, end);
1203                 if (rc < 0) {
1204                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1205                                "cancel: rc = %d\n",
1206                                tgt_name(tgt), PFID(&fid),
1207                                lock->l_policy_data.l_extent.start,
1208                                lock->l_policy_data.l_extent.end, rc);
1209                 }
1210 err_put:
1211                 lu_object_put(&env, &obj->do_lu);
1212 err_env:
1213                 lu_env_fini(&env);
1214         }
1215
1216         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1217         RETURN(rc);
1218 }
1219
1220 struct ldlm_callback_suite tgt_dlm_cbs = {
1221         .lcs_completion = ldlm_server_completion_ast,
1222         .lcs_blocking   = tgt_blocking_ast,
1223         .lcs_glimpse    = ldlm_server_glimpse_ast
1224 };
1225
1226 int tgt_enqueue(struct tgt_session_info *tsi)
1227 {
1228         struct ptlrpc_request *req = tgt_ses_req(tsi);
1229         int rc;
1230
1231         ENTRY;
1232         /*
1233          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1234          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1235          */
1236         LASSERT(tsi->tsi_dlm_req != NULL);
1237         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1238                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1239         if (rc)
1240                 RETURN(err_serious(rc));
1241
1242         RETURN(req->rq_status);
1243 }
1244 EXPORT_SYMBOL(tgt_enqueue);
1245
1246 int tgt_convert(struct tgt_session_info *tsi)
1247 {
1248         struct ptlrpc_request *req = tgt_ses_req(tsi);
1249         int rc;
1250
1251         ENTRY;
1252         LASSERT(tsi->tsi_dlm_req);
1253         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1254         if (rc)
1255                 RETURN(err_serious(rc));
1256
1257         RETURN(req->rq_status);
1258 }
1259 EXPORT_SYMBOL(tgt_convert);
1260
1261 int tgt_bl_callback(struct tgt_session_info *tsi)
1262 {
1263         return err_serious(-EOPNOTSUPP);
1264 }
1265 EXPORT_SYMBOL(tgt_bl_callback);
1266
1267 int tgt_cp_callback(struct tgt_session_info *tsi)
1268 {
1269         return err_serious(-EOPNOTSUPP);
1270 }
1271 EXPORT_SYMBOL(tgt_cp_callback);
1272
1273 /* generic LDLM target handler */
1274 struct tgt_handler tgt_dlm_handlers[] = {
1275 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1276 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1277 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1278 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1279 };
1280 EXPORT_SYMBOL(tgt_dlm_handlers);
1281
1282 /*
1283  * Unified target LLOG handlers.
1284  */
1285 int tgt_llog_open(struct tgt_session_info *tsi)
1286 {
1287         int rc;
1288
1289         ENTRY;
1290
1291         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1292
1293         RETURN(rc);
1294 }
1295 EXPORT_SYMBOL(tgt_llog_open);
1296
1297 int tgt_llog_close(struct tgt_session_info *tsi)
1298 {
1299         int rc;
1300
1301         ENTRY;
1302
1303         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1304
1305         RETURN(rc);
1306 }
1307 EXPORT_SYMBOL(tgt_llog_close);
1308
1309
1310 int tgt_llog_destroy(struct tgt_session_info *tsi)
1311 {
1312         int rc;
1313
1314         ENTRY;
1315
1316         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1317
1318         RETURN(rc);
1319 }
1320 EXPORT_SYMBOL(tgt_llog_destroy);
1321
1322 int tgt_llog_read_header(struct tgt_session_info *tsi)
1323 {
1324         int rc;
1325
1326         ENTRY;
1327
1328         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1329
1330         RETURN(rc);
1331 }
1332 EXPORT_SYMBOL(tgt_llog_read_header);
1333
1334 int tgt_llog_next_block(struct tgt_session_info *tsi)
1335 {
1336         int rc;
1337
1338         ENTRY;
1339
1340         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1341
1342         RETURN(rc);
1343 }
1344 EXPORT_SYMBOL(tgt_llog_next_block);
1345
1346 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1347 {
1348         int rc;
1349
1350         ENTRY;
1351
1352         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1353
1354         RETURN(rc);
1355 }
1356 EXPORT_SYMBOL(tgt_llog_prev_block);
1357
1358 /* generic llog target handler */
1359 struct tgt_handler tgt_llog_handlers[] = {
1360 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1361 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1362 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1363 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1364 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1365 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1366 };
1367 EXPORT_SYMBOL(tgt_llog_handlers);
1368
1369 /*
1370  * sec context handlers
1371  */
1372 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1373 int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1374 {
1375         return 0;
1376 }
1377
1378 struct tgt_handler tgt_sec_ctx_handlers[] = {
1379 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1380 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1381 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1382 };
1383 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1384
1385 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1386                            struct dt_device *key,
1387                            struct lfsck_request *lr,
1388                            struct thandle *th) = NULL;
1389
1390 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1391                                                 struct dt_device *,
1392                                                 struct lfsck_request *,
1393                                                 struct thandle *))
1394 {
1395         tgt_lfsck_in_notify = notify;
1396 }
1397 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1398
1399 static int (*tgt_lfsck_query)(const struct lu_env *env,
1400                               struct dt_device *key,
1401                               struct lfsck_request *lr) = NULL;
1402
1403 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1404                                            struct dt_device *,
1405                                            struct lfsck_request *))
1406 {
1407         tgt_lfsck_query = query;
1408 }
1409 EXPORT_SYMBOL(tgt_register_lfsck_query);
1410
1411 /* LFSCK request handlers */
1412 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1413 {
1414         const struct lu_env     *env = tsi->tsi_env;
1415         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1416         struct lfsck_request    *lr;
1417         int                      rc;
1418         ENTRY;
1419
1420         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1421         if (lr == NULL)
1422                 RETURN(-EPROTO);
1423
1424         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1425
1426         RETURN(rc);
1427 }
1428
1429 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1430 {
1431         struct lfsck_request    *request;
1432         struct lfsck_reply      *reply;
1433         int                      rc;
1434         ENTRY;
1435
1436         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1437         if (request == NULL)
1438                 RETURN(-EPROTO);
1439
1440         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1441         if (reply == NULL)
1442                 RETURN(-ENOMEM);
1443
1444         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1445         reply->lr_status = rc;
1446
1447         RETURN(rc < 0 ? rc : 0);
1448 }
1449
1450 struct tgt_handler tgt_lfsck_handlers[] = {
1451 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1452 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1453 };
1454 EXPORT_SYMBOL(tgt_lfsck_handlers);
1455
1456 /*
1457  * initialize per-thread page pool (bug 5137).
1458  */
1459 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1460 {
1461         struct tgt_thread_big_cache *tbc;
1462
1463         ENTRY;
1464
1465         LASSERT(thread != NULL);
1466         LASSERT(thread->t_data == NULL);
1467
1468         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1469         if (tbc == NULL)
1470                 RETURN(-ENOMEM);
1471         thread->t_data = tbc;
1472         RETURN(0);
1473 }
1474 EXPORT_SYMBOL(tgt_io_thread_init);
1475
1476 /*
1477  * free per-thread pool created by tgt_thread_init().
1478  */
1479 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1480 {
1481         struct tgt_thread_big_cache *tbc;
1482
1483         ENTRY;
1484
1485         LASSERT(thread != NULL);
1486
1487         /*
1488          * be prepared to handle partially-initialized pools (because this is
1489          * called from ost_io_thread_init() for cleanup.
1490          */
1491         tbc = thread->t_data;
1492         if (tbc != NULL) {
1493                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1494                 thread->t_data = NULL;
1495         }
1496         EXIT;
1497 }
1498 EXPORT_SYMBOL(tgt_io_thread_done);
1499 /**
1500  * Helper function for getting server side [start, start+count] DLM lock
1501  * if asked by client.
1502  */
1503 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1504                     __u64 start, __u64 end, struct lustre_handle *lh,
1505                     int mode, __u64 *flags)
1506 {
1507         ldlm_policy_data_t       policy;
1508         int                      rc;
1509
1510         ENTRY;
1511
1512         LASSERT(lh != NULL);
1513         LASSERT(ns != NULL);
1514         LASSERT(!lustre_handle_is_used(lh));
1515
1516         policy.l_extent.gid = 0;
1517         policy.l_extent.start = start & CFS_PAGE_MASK;
1518
1519         /*
1520          * If ->o_blocks is EOF it means "lock till the end of the file".
1521          * Otherwise, it's size of an extent or hole being punched (in bytes).
1522          */
1523         if (end == OBD_OBJECT_EOF || end < start)
1524                 policy.l_extent.end = OBD_OBJECT_EOF;
1525         else
1526                 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1527
1528         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1529                                     flags, ldlm_blocking_ast,
1530                                     ldlm_completion_ast, ldlm_glimpse_ast,
1531                                     NULL, 0, LVB_T_NONE, NULL, lh);
1532         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1533 }
1534 EXPORT_SYMBOL(tgt_extent_lock);
1535
1536 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1537 {
1538         LASSERT(lustre_handle_is_used(lh));
1539         ldlm_lock_decref(lh, mode);
1540 }
1541 EXPORT_SYMBOL(tgt_extent_unlock);
1542
1543 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1544                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1545                  struct lustre_handle *lh, int mode)
1546 {
1547         __u64                    flags = 0;
1548         int                      nrbufs = obj->ioo_bufcnt;
1549         int                      i;
1550
1551         ENTRY;
1552
1553         LASSERT(mode == LCK_PR || mode == LCK_PW);
1554         LASSERT(!lustre_handle_is_used(lh));
1555
1556         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1557                 RETURN(0);
1558
1559         for (i = 1; i < nrbufs; i++)
1560                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1561                         RETURN(-EFAULT);
1562
1563         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1564                                nb[nrbufs - 1].rnb_offset +
1565                                nb[nrbufs - 1].rnb_len - 1,
1566                                lh, mode, &flags));
1567 }
1568 EXPORT_SYMBOL(tgt_brw_lock);
1569
1570 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1571                     struct lustre_handle *lh, int mode)
1572 {
1573         ENTRY;
1574
1575         LASSERT(mode == LCK_PR || mode == LCK_PW);
1576         LASSERT((obj->ioo_bufcnt > 0 &&
1577                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1578                 lustre_handle_is_used(lh));
1579
1580         if (lustre_handle_is_used(lh))
1581                 tgt_extent_unlock(lh, mode);
1582         EXIT;
1583 }
1584 EXPORT_SYMBOL(tgt_brw_unlock);
1585
1586 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1587                                struct ptlrpc_bulk_desc *desc, int opc,
1588                                cksum_type_t cksum_type)
1589 {
1590         struct cfs_crypto_hash_desc     *hdesc;
1591         unsigned int                    bufsize;
1592         int                             i, err;
1593         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1594         __u32                           cksum;
1595
1596         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1597         if (IS_ERR(hdesc)) {
1598                 CERROR("%s: unable to initialize checksum hash %s\n",
1599                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1600                 return PTR_ERR(hdesc);
1601         }
1602
1603         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1604         for (i = 0; i < desc->bd_iov_count; i++) {
1605                 /* corrupt the data before we compute the checksum, to
1606                  * simulate a client->OST data error */
1607                 if (i == 0 && opc == OST_WRITE &&
1608                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1609                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1610                         int len = desc->bd_iov[i].kiov_len;
1611                         struct page *np = tgt_page_to_corrupt;
1612                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1613
1614                         if (np) {
1615                                 char *ptr2 = kmap(np) + off;
1616
1617                                 memcpy(ptr2, ptr, len);
1618                                 memcpy(ptr2, "bad3", min(4, len));
1619                                 kunmap(np);
1620                                 desc->bd_iov[i].kiov_page = np;
1621                         } else {
1622                                 CERROR("%s: can't alloc page for corruption\n",
1623                                        tgt_name(tgt));
1624                         }
1625                 }
1626                 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1627                                   desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1628                                   desc->bd_iov[i].kiov_len);
1629
1630                  /* corrupt the data after we compute the checksum, to
1631                  * simulate an OST->client data error */
1632                 if (i == 0 && opc == OST_READ &&
1633                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1634                         int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1635                         int len = desc->bd_iov[i].kiov_len;
1636                         struct page *np = tgt_page_to_corrupt;
1637                         char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1638
1639                         if (np) {
1640                                 char *ptr2 = kmap(np) + off;
1641
1642                                 memcpy(ptr2, ptr, len);
1643                                 memcpy(ptr2, "bad4", min(4, len));
1644                                 kunmap(np);
1645                                 desc->bd_iov[i].kiov_page = np;
1646                         } else {
1647                                 CERROR("%s: can't alloc page for corruption\n",
1648                                        tgt_name(tgt));
1649                         }
1650                 }
1651         }
1652
1653         bufsize = sizeof(cksum);
1654         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1655
1656         return cksum;
1657 }
1658
1659 int tgt_brw_read(struct tgt_session_info *tsi)
1660 {
1661         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1662         struct ptlrpc_bulk_desc *desc = NULL;
1663         struct obd_export       *exp = tsi->tsi_exp;
1664         struct niobuf_remote    *remote_nb;
1665         struct niobuf_local     *local_nb;
1666         struct obd_ioobj        *ioo;
1667         struct ost_body         *body, *repbody;
1668         struct l_wait_info       lwi;
1669         struct lustre_handle     lockh = { 0 };
1670         int                      npages, nob = 0, rc, i, no_reply = 0;
1671         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1672
1673         ENTRY;
1674
1675         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1676                 CERROR("%s: deny read request from %s to portal %u\n",
1677                        tgt_name(tsi->tsi_tgt),
1678                        obd_export_nid2str(req->rq_export),
1679                        ptlrpc_req2svc(req)->srv_req_portal);
1680                 RETURN(-EPROTO);
1681         }
1682
1683         req->rq_bulk_read = 1;
1684
1685         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1686                 RETURN(-EIO);
1687
1688         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1689                          cfs_fail_val : (obd_timeout + 1) / 4);
1690
1691         /* Check if there is eviction in progress, and if so, wait for it to
1692          * finish */
1693         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1694                 /* We do not care how long it takes */
1695                 lwi = LWI_INTR(NULL, NULL);
1696                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1697                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1698                          &lwi);
1699         }
1700
1701         /* There must be big cache in current thread to process this request
1702          * if it is NULL then something went wrong and it wasn't allocated,
1703          * report -ENOMEM in that case */
1704         if (tbc == NULL)
1705                 RETURN(-ENOMEM);
1706
1707         body = tsi->tsi_ost_body;
1708         LASSERT(body != NULL);
1709
1710         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1711         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1712
1713         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1714         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1715
1716         local_nb = tbc->local;
1717
1718         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1719                           remote_nb, &lockh, LCK_PR);
1720         if (rc != 0)
1721                 RETURN(rc);
1722
1723         /*
1724          * If getting the lock took more time than
1725          * client was willing to wait, drop it. b=11330
1726          */
1727         if (cfs_time_current_sec() > req->rq_deadline ||
1728             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1729                 no_reply = 1;
1730                 CERROR("Dropping timed-out read from %s because locking"
1731                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1732                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1733                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1734                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1735                 GOTO(out_lock, rc = -ETIMEDOUT);
1736         }
1737
1738         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1739         repbody->oa = body->oa;
1740
1741         npages = PTLRPC_MAX_BRW_PAGES;
1742         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1743                         ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1744         if (rc != 0)
1745                 GOTO(out_lock, rc);
1746
1747         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1748                                     BULK_PUT_SOURCE, OST_BULK_PORTAL);
1749         if (desc == NULL)
1750                 GOTO(out_commitrw, rc = -ENOMEM);
1751
1752         nob = 0;
1753         for (i = 0; i < npages; i++) {
1754                 int page_rc = local_nb[i].lnb_rc;
1755
1756                 if (page_rc < 0) {
1757                         rc = page_rc;
1758                         break;
1759                 }
1760
1761                 nob += page_rc;
1762                 if (page_rc != 0) { /* some data! */
1763                         LASSERT(local_nb[i].lnb_page != NULL);
1764                         ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
1765                                                     local_nb[i].lnb_page_offset,
1766                                                     page_rc);
1767                 }
1768
1769                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1770                         /* All subsequent pages should be 0 */
1771                         while (++i < npages)
1772                                 LASSERT(local_nb[i].lnb_rc == 0);
1773                         break;
1774                 }
1775         }
1776
1777         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1778                 cksum_type_t cksum_type =
1779                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1780                                           body->oa.o_flags : 0);
1781                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1782                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1783                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1784                                                         OST_READ, cksum_type);
1785                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1786                        repbody->oa.o_cksum);
1787         } else {
1788                 repbody->oa.o_valid = 0;
1789         }
1790         /* We're finishing using body->oa as an input variable */
1791
1792         /* Check if client was evicted while we were doing i/o before touching
1793          * network */
1794         if (likely(rc == 0 &&
1795                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1796                 rc = target_bulk_io(exp, desc, &lwi);
1797                 no_reply = rc != 0;
1798         }
1799
1800 out_commitrw:
1801         /* Must commit after prep above in all cases */
1802         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1803                           &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1804                           NULL, rc);
1805         if (rc == 0)
1806                 tgt_drop_id(exp, &repbody->oa);
1807 out_lock:
1808         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1809
1810         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1811                 ptlrpc_free_bulk_nopin(desc);
1812
1813         LASSERT(rc <= 0);
1814         if (rc == 0) {
1815                 rc = nob;
1816                 ptlrpc_lprocfs_brw(req, nob);
1817         } else if (no_reply) {
1818                 req->rq_no_reply = 1;
1819                 /* reply out callback would free */
1820                 ptlrpc_req_drop_rs(req);
1821                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1822                               "client will retry: rc %d\n",
1823                               exp->exp_obd->obd_name,
1824                               obd_uuid2str(&exp->exp_client_uuid),
1825                               obd_export_nid2str(exp), rc);
1826         }
1827         /* send a bulk after reply to simulate a network delay or reordering
1828          * by a router */
1829         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1830                 wait_queue_head_t        waitq;
1831                 struct l_wait_info       lwi1;
1832
1833                 CDEBUG(D_INFO, "reorder BULK\n");
1834                 init_waitqueue_head(&waitq);
1835
1836                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1837                 l_wait_event(waitq, 0, &lwi1);
1838                 target_bulk_io(exp, desc, &lwi);
1839                 ptlrpc_free_bulk_nopin(desc);
1840         }
1841
1842         RETURN(rc);
1843 }
1844 EXPORT_SYMBOL(tgt_brw_read);
1845
1846 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1847                               struct ptlrpc_bulk_desc *desc,
1848                               struct niobuf_local *local_nb, int npages,
1849                               obd_count client_cksum, obd_count server_cksum,
1850                               bool mmap)
1851 {
1852         struct obd_export *exp = req->rq_export;
1853         struct ost_body *body;
1854         char *router;
1855         char *via;
1856
1857         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1858         LASSERT(body != NULL);
1859
1860         if (req->rq_peer.nid == desc->bd_sender) {
1861                 via = router = "";
1862         } else {
1863                 via = " via ";
1864                 router = libcfs_nid2str(desc->bd_sender);
1865         }
1866
1867         if (mmap) {
1868                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1869                              client_cksum, server_cksum);
1870                 return;
1871         }
1872
1873         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1874                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1875                            "]: client csum %x, server csum %x\n",
1876                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1877                            via, router,
1878                            body->oa.o_valid & OBD_MD_FLFID ?
1879                            body->oa.o_parent_seq : (__u64)0,
1880                            body->oa.o_valid & OBD_MD_FLFID ?
1881                            body->oa.o_parent_oid : 0,
1882                            body->oa.o_valid & OBD_MD_FLFID ?
1883                            body->oa.o_parent_ver : 0,
1884                            POSTID(&body->oa.o_oi),
1885                            local_nb[0].lnb_file_offset,
1886                            local_nb[npages-1].lnb_file_offset +
1887                            local_nb[npages - 1].lnb_len - 1,
1888                            client_cksum, server_cksum);
1889 }
1890
1891 int tgt_brw_write(struct tgt_session_info *tsi)
1892 {
1893         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1894         struct ptlrpc_bulk_desc *desc = NULL;
1895         struct obd_export       *exp = req->rq_export;
1896         struct niobuf_remote    *remote_nb;
1897         struct niobuf_local     *local_nb;
1898         struct obd_ioobj        *ioo;
1899         struct ost_body         *body, *repbody;
1900         struct l_wait_info       lwi;
1901         struct lustre_handle     lockh = {0};
1902         __u32                   *rcs;
1903         int                      objcount, niocount, npages;
1904         int                      rc, i, j;
1905         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1906         bool                     no_reply = false, mmap;
1907         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1908
1909         ENTRY;
1910
1911         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1912                 CERROR("%s: deny write request from %s to portal %u\n",
1913                        tgt_name(tsi->tsi_tgt),
1914                        obd_export_nid2str(req->rq_export),
1915                        ptlrpc_req2svc(req)->srv_req_portal);
1916                 RETURN(err_serious(-EPROTO));
1917         }
1918
1919         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1920                 RETURN(err_serious(-ENOSPC));
1921         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1922                 RETURN(err_serious(-EROFS));
1923
1924         req->rq_bulk_write = 1;
1925
1926         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1927                 RETURN(err_serious(-EIO));
1928         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1929                 RETURN(err_serious(-EFAULT));
1930
1931         /* pause before transaction has been started */
1932         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1933                          cfs_fail_val : (obd_timeout + 1) / 4);
1934
1935         /* There must be big cache in current thread to process this request
1936          * if it is NULL then something went wrong and it wasn't allocated,
1937          * report -ENOMEM in that case */
1938         if (tbc == NULL)
1939                 RETURN(-ENOMEM);
1940
1941         body = tsi->tsi_ost_body;
1942         LASSERT(body != NULL);
1943
1944         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1945         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1946
1947         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1948                                         RCL_CLIENT) / sizeof(*ioo);
1949
1950         for (niocount = i = 0; i < objcount; i++)
1951                 niocount += ioo[i].ioo_bufcnt;
1952
1953         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1954         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1955         if (niocount != req_capsule_get_size(&req->rq_pill,
1956                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1957                         sizeof(*remote_nb))
1958                 RETURN(err_serious(-EPROTO));
1959
1960         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1961             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1962                 memory_pressure_set();
1963
1964         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1965                              niocount * sizeof(*rcs));
1966         rc = req_capsule_server_pack(&req->rq_pill);
1967         if (rc != 0)
1968                 GOTO(out, rc = err_serious(rc));
1969
1970         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1971         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1972
1973         local_nb = tbc->local;
1974
1975         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1976                           remote_nb, &lockh, LCK_PW);
1977         if (rc != 0)
1978                 GOTO(out, rc);
1979
1980         /*
1981          * If getting the lock took more time than
1982          * client was willing to wait, drop it. b=11330
1983          */
1984         if (cfs_time_current_sec() > req->rq_deadline ||
1985             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1986                 no_reply = true;
1987                 CERROR("%s: Dropping timed-out write from %s because locking "
1988                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1989                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1990                        POSTID(&ioo->ioo_oid),
1991                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1992                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1993                 GOTO(out_lock, rc = -ETIMEDOUT);
1994         }
1995
1996         /* Because we already sync grant info with client when reconnect,
1997          * grant info will be cleared for resent req, then fed_grant and
1998          * total_grant will not be modified in following preprw_write */
1999         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2000                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2001                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2002         }
2003
2004         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2005         if (repbody == NULL)
2006                 GOTO(out_lock, rc = -ENOMEM);
2007         repbody->oa = body->oa;
2008
2009         npages = PTLRPC_MAX_BRW_PAGES;
2010         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2011                         objcount, ioo, remote_nb, &npages, local_nb, NULL,
2012                         BYPASS_CAPA);
2013         if (rc < 0)
2014                 GOTO(out_lock, rc);
2015
2016         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2017                                     BULK_GET_SINK, OST_BULK_PORTAL);
2018         if (desc == NULL)
2019                 GOTO(skip_transfer, rc = -ENOMEM);
2020
2021         /* NB Having prepped, we must commit... */
2022         for (i = 0; i < npages; i++)
2023                 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
2024                                             local_nb[i].lnb_page_offset,
2025                                             local_nb[i].lnb_len);
2026
2027         rc = sptlrpc_svc_prep_bulk(req, desc);
2028         if (rc != 0)
2029                 GOTO(skip_transfer, rc);
2030
2031         rc = target_bulk_io(exp, desc, &lwi);
2032         no_reply = rc != 0;
2033
2034 skip_transfer:
2035         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2036                 static int cksum_counter;
2037
2038                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2039                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2040
2041                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2042                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2043                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2044                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2045                                                         OST_WRITE, cksum_type);
2046                 cksum_counter++;
2047
2048                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2049                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2050                                 body->oa.o_flags & OBD_FL_MMAP);
2051
2052                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2053                                           body->oa.o_cksum,
2054                                           repbody->oa.o_cksum, mmap);
2055                         cksum_counter = 0;
2056                 } else if ((cksum_counter & (-cksum_counter)) ==
2057                            cksum_counter) {
2058                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2059                                cksum_counter, libcfs_id2str(req->rq_peer),
2060                                repbody->oa.o_cksum);
2061                 }
2062         }
2063
2064         /* Must commit after prep above in all cases */
2065         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2066                           objcount, ioo, remote_nb, npages, local_nb, NULL,
2067                           rc);
2068         if (rc == -ENOTCONN)
2069                 /* quota acquire process has been given up because
2070                  * either the client has been evicted or the client
2071                  * has timed out the request already */
2072                 no_reply = true;
2073
2074         /*
2075          * Disable sending mtime back to the client. If the client locked the
2076          * whole object, then it has already updated the mtime on its side,
2077          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2078          */
2079         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2080
2081         if (rc == 0) {
2082                 int nob = 0;
2083
2084                 /* set per-requested niobuf return codes */
2085                 for (i = j = 0; i < niocount; i++) {
2086                         int len = remote_nb[i].rnb_len;
2087
2088                         nob += len;
2089                         rcs[i] = 0;
2090                         do {
2091                                 LASSERT(j < npages);
2092                                 if (local_nb[j].lnb_rc < 0)
2093                                         rcs[i] = local_nb[j].lnb_rc;
2094                                 len -= local_nb[j].lnb_len;
2095                                 j++;
2096                         } while (len > 0);
2097                         LASSERT(len == 0);
2098                 }
2099                 LASSERT(j == npages);
2100                 ptlrpc_lprocfs_brw(req, nob);
2101
2102                 tgt_drop_id(exp, &repbody->oa);
2103         }
2104 out_lock:
2105         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2106         if (desc)
2107                 ptlrpc_free_bulk_nopin(desc);
2108 out:
2109         if (no_reply) {
2110                 req->rq_no_reply = 1;
2111                 /* reply out callback would free */
2112                 ptlrpc_req_drop_rs(req);
2113                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2114                               "client will retry: rc %d\n",
2115                               exp->exp_obd->obd_name,
2116                               obd_uuid2str(&exp->exp_client_uuid),
2117                               obd_export_nid2str(exp), rc);
2118         }
2119         memory_pressure_clr();
2120         RETURN(rc);
2121 }
2122 EXPORT_SYMBOL(tgt_brw_write);