lustre/mdt/mdt_open.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, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_open.c
  37  *
  38  * Lustre Metadata Target (mdt) open/close file handling
  39  *
  40  * Author: Huang Hua <huanghua@clusterfs.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_MDS
  44
  45 #include <lustre_acl.h>
  46 #include <lustre_mds.h>
  47 #include "mdt_internal.h"
  48 #include <lustre_nodemap.h>
  49
  50 /* we do nothing because we do not have refcount now */
  51 static void mdt_mfd_get(void *mfdp)
  52 {
  53 }
  54
  55 static struct portals_handle_ops mfd_handle_ops = {
  56         .hop_addref = mdt_mfd_get,
  57         .hop_free   = NULL,
  58 };
  59
  60 /* Create a new mdt_file_data struct, initialize it,
  61  * and insert it to global hash table */
  62 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med)
  63 {
  64         struct mdt_file_data *mfd;
  65         ENTRY;
  66
  67         OBD_ALLOC_PTR(mfd);
  68         if (mfd != NULL) {
  69                 INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
  70                 mfd->mfd_handle.h_owner = med;
  71                 INIT_LIST_HEAD(&mfd->mfd_list);
  72                 class_handle_hash(&mfd->mfd_handle, &mfd_handle_ops);
  73         }
  74
  75         RETURN(mfd);
  76 }
  77
  78 /*
  79  * Find the mfd pointed to by handle in global hash table.
  80  * In case of replay the handle is obsoleted
  81  * but mfd can be found in mfd list by that handle.
  82  * Callers need to be holding med_open_lock.
  83  */
  84 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
  85                                      const struct lustre_handle *handle,
  86                                      bool is_replay_or_resent)
  87 {
  88         struct mdt_file_data   *mfd;
  89         ENTRY;
  90
  91         LASSERT(handle != NULL);
  92         mfd = class_handle2object(handle->cookie, med);
  93         /* during dw/setattr replay the mfd can be found by old handle */
  94         if (mfd == NULL && is_replay_or_resent) {
  95                 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
  96                         if (mfd->mfd_old_handle.cookie == handle->cookie)
  97                                 RETURN(mfd);
  98                 }
  99                 mfd = NULL;
 100         }
 101
 102         RETURN(mfd);
 103 }
 104
 105 /* free mfd */
 106 void mdt_mfd_free(struct mdt_file_data *mfd)
 107 {
 108         LASSERT(list_empty(&mfd->mfd_list));
 109         OBD_FREE_RCU(mfd, sizeof *mfd, &mfd->mfd_handle);
 110 }
 111
 112 static int mdt_create_data(struct mdt_thread_info *info,
 113                            struct mdt_object *p, struct mdt_object *o)
 114 {
 115         struct md_op_spec     *spec = &info->mti_spec;
 116         struct md_attr        *ma   = &info->mti_attr;
 117         int                    rc   = 0;
 118         ENTRY;
 119
 120         if (!md_should_create(spec->sp_cr_flags))
 121                 RETURN(0);
 122
 123         ma->ma_need = MA_INODE | MA_LOV;
 124         ma->ma_valid = 0;
 125         mutex_lock(&o->mot_lov_mutex);
 126         if (!(o->mot_flags & MOF_LOV_CREATED)) {
 127                 rc = mdo_create_data(info->mti_env,
 128                                      p ? mdt_object_child(p) : NULL,
 129                                      mdt_object_child(o), spec, ma);
 130                 if (rc == 0)
 131                         rc = mdt_attr_get_complex(info, o, ma);
 132
 133                 if (rc == 0 && ma->ma_valid & MA_LOV)
 134                         o->mot_flags |= MOF_LOV_CREATED;
 135         }
 136
 137         mutex_unlock(&o->mot_lov_mutex);
 138         RETURN(rc);
 139 }
 140
 141 int mdt_write_read(struct mdt_object *o)
 142 {
 143         int rc = 0;
 144         ENTRY;
 145         spin_lock(&o->mot_write_lock);
 146         rc = o->mot_write_count;
 147         spin_unlock(&o->mot_write_lock);
 148         RETURN(rc);
 149 }
 150
 151 int mdt_write_get(struct mdt_object *o)
 152 {
 153         int rc = 0;
 154         ENTRY;
 155         spin_lock(&o->mot_write_lock);
 156         if (o->mot_write_count < 0)
 157                 rc = -ETXTBSY;
 158         else
 159                 o->mot_write_count++;
 160         spin_unlock(&o->mot_write_lock);
 161
 162         RETURN(rc);
 163 }
 164
 165 void mdt_write_put(struct mdt_object *o)
 166 {
 167         ENTRY;
 168         spin_lock(&o->mot_write_lock);
 169         o->mot_write_count--;
 170         spin_unlock(&o->mot_write_lock);
 171         EXIT;
 172 }
 173
 174 static int mdt_write_deny(struct mdt_object *o)
 175 {
 176         int rc = 0;
 177         ENTRY;
 178         spin_lock(&o->mot_write_lock);
 179         if (o->mot_write_count > 0)
 180                 rc = -ETXTBSY;
 181         else
 182                 o->mot_write_count--;
 183         spin_unlock(&o->mot_write_lock);
 184         RETURN(rc);
 185 }
 186
 187 static void mdt_write_allow(struct mdt_object *o)
 188 {
 189         ENTRY;
 190         spin_lock(&o->mot_write_lock);
 191         o->mot_write_count++;
 192         spin_unlock(&o->mot_write_lock);
 193         EXIT;
 194 }
 195
 196 /* there can be no real transaction so prepare the fake one */
 197 static void mdt_empty_transno(struct mdt_thread_info *info, int rc)
 198 {
 199         struct mdt_device      *mdt = info->mti_mdt;
 200         struct ptlrpc_request  *req = mdt_info_req(info);
 201         struct tg_export_data  *ted;
 202         struct lsd_client_data *lcd;
 203
 204         ENTRY;
 205         /* transaction has occurred already */
 206         if (lustre_msg_get_transno(req->rq_repmsg) != 0)
 207                 RETURN_EXIT;
 208
 209         spin_lock(&mdt->mdt_lut.lut_translock);
 210         if (rc != 0) {
 211                 if (info->mti_transno != 0) {
 212                         struct obd_export *exp = req->rq_export;
 213
 214                         CERROR("%s: replay trans "LPU64" NID %s: rc = %d\n",
 215                                mdt_obd_name(mdt), info->mti_transno,
 216                                libcfs_nid2str(exp->exp_connection->c_peer.nid),
 217                                rc);
 218                         spin_unlock(&mdt->mdt_lut.lut_translock);
 219                         RETURN_EXIT;
 220                 }
 221         } else if (info->mti_transno == 0) {
 222                 info->mti_transno = ++mdt->mdt_lut.lut_last_transno;
 223         } else {
 224                 /* should be replay */
 225                 if (info->mti_transno > mdt->mdt_lut.lut_last_transno)
 226                         mdt->mdt_lut.lut_last_transno = info->mti_transno;
 227         }
 228         spin_unlock(&mdt->mdt_lut.lut_translock);
 229
 230         CDEBUG(D_INODE, "transno = "LPU64", last_committed = "LPU64"\n",
 231                info->mti_transno,
 232                req->rq_export->exp_obd->obd_last_committed);
 233
 234         req->rq_transno = info->mti_transno;
 235         lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
 236
 237         /* update lcd in memory only for resent cases */
 238         ted = &req->rq_export->exp_target_data;
 239         LASSERT(ted);
 240         mutex_lock(&ted->ted_lcd_lock);
 241         lcd = ted->ted_lcd;
 242         if (info->mti_transno < lcd->lcd_last_transno &&
 243             info->mti_transno != 0) {
 244                 /* This should happen during replay. Do not update
 245                  * last rcvd info if replay req transno < last transno,
 246                  * otherwise the following resend(after replay) can not
 247                  * be checked correctly by xid */
 248                 mutex_unlock(&ted->ted_lcd_lock);
 249                 CDEBUG(D_HA, "%s: transno = "LPU64" < last_transno = "LPU64"\n",
 250                        mdt_obd_name(mdt), info->mti_transno,
 251                        lcd->lcd_last_transno);
 252                 RETURN_EXIT;
 253         }
 254
 255         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
 256                 if (info->mti_transno != 0)
 257                         lcd->lcd_last_close_transno = info->mti_transno;
 258                 lcd->lcd_last_close_xid = req->rq_xid;
 259                 lcd->lcd_last_close_result = rc;
 260         } else {
 261                 /* VBR: save versions in last_rcvd for reconstruct. */
 262                 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
 263                 if (pre_versions) {
 264                         lcd->lcd_pre_versions[0] = pre_versions[0];
 265                         lcd->lcd_pre_versions[1] = pre_versions[1];
 266                         lcd->lcd_pre_versions[2] = pre_versions[2];
 267                         lcd->lcd_pre_versions[3] = pre_versions[3];
 268                 }
 269                 if (info->mti_transno != 0)
 270                         lcd->lcd_last_transno = info->mti_transno;
 271
 272                 lcd->lcd_last_xid = req->rq_xid;
 273                 lcd->lcd_last_result = rc;
 274                 lcd->lcd_last_data = info->mti_opdata;
 275         }
 276         mutex_unlock(&ted->ted_lcd_lock);
 277
 278         EXIT;
 279 }
 280
 281 void mdt_mfd_set_mode(struct mdt_file_data *mfd, __u64 mode)
 282 {
 283         LASSERT(mfd != NULL);
 284
 285         CDEBUG(D_HA, DFID " Change mfd mode "LPO64" -> "LPO64".\n",
 286                PFID(mdt_object_fid(mfd->mfd_object)), mfd->mfd_mode, mode);
 287
 288         mfd->mfd_mode = mode;
 289 }
 290
 291 /**
 292  * prep ma_lmm/ma_lmv for md_attr from reply
 293  */
 294 static void mdt_prep_ma_buf_from_rep(struct mdt_thread_info *info,
 295                                      struct mdt_object *obj,
 296                                      struct md_attr *ma)
 297 {
 298         LASSERT(ma->ma_lmv == NULL && ma->ma_lmm == NULL);
 299         if (S_ISDIR(obj->mot_header.loh_attr)) {
 300                 ma->ma_lmv = req_capsule_server_get(info->mti_pill,
 301                                                     &RMF_MDT_MD);
 302                 ma->ma_lmv_size = req_capsule_get_size(info->mti_pill,
 303                                                        &RMF_MDT_MD,
 304                                                        RCL_SERVER);
 305                 if (ma->ma_lmv_size > 0)
 306                         ma->ma_need |= MA_LMV;
 307         } else {
 308                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
 309                                                     &RMF_MDT_MD);
 310                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 311                                                        &RMF_MDT_MD,
 312                                                        RCL_SERVER);
 313                 if (ma->ma_lmm_size > 0)
 314                         ma->ma_need |= MA_LOV;
 315         }
 316 }
 317
 318 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
 319                         struct mdt_object *o, __u64 flags, int created,
 320                         struct ldlm_reply *rep)
 321 {
 322         struct ptlrpc_request   *req = mdt_info_req(info);
 323         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 324         struct mdt_file_data    *mfd;
 325         struct md_attr          *ma  = &info->mti_attr;
 326         struct lu_attr          *la  = &ma->ma_attr;
 327         struct mdt_body         *repbody;
 328         int                      rc = 0, isdir, isreg;
 329         ENTRY;
 330
 331         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 332
 333         isreg = S_ISREG(la->la_mode);
 334         isdir = S_ISDIR(la->la_mode);
 335         if (isreg && !(ma->ma_valid & MA_LOV) && !(flags & MDS_OPEN_RELEASE)) {
 336                 /*
 337                  * No EA, check whether it is will set regEA and dirEA since in
 338                  * above attr get, these size might be zero, so reset it, to
 339                  * retrieve the MD after create obj.
 340                  */
 341                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 342                                                        &RMF_MDT_MD,
 343                                                        RCL_SERVER);
 344                 /* in replay case, p == NULL */
 345                 rc = mdt_create_data(info, p, o);
 346                 if (rc)
 347                         RETURN(rc);
 348
 349                 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_DISP_STRIPE)
 350                         mdt_set_disposition(info, rep, DISP_OPEN_STRIPE);
 351         }
 352
 353         CDEBUG(D_INODE, "after open, ma_valid bit = "LPX64" lmm_size = %d\n",
 354                ma->ma_valid, ma->ma_lmm_size);
 355
 356         if (ma->ma_valid & MA_LOV) {
 357                 LASSERT(ma->ma_lmm_size != 0);
 358                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 359                 if (isdir)
 360                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 361                 else
 362                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 363         }
 364
 365         if (ma->ma_valid & MA_LMV) {
 366                 LASSERT(ma->ma_lmv_size != 0);
 367                 repbody->mbo_eadatasize = ma->ma_lmv_size;
 368                 LASSERT(isdir);
 369                 repbody->mbo_valid |= OBD_MD_FLDIREA | OBD_MD_MEA;
 370         }
 371
 372         if (flags & FMODE_WRITE)
 373                 rc = mdt_write_get(o);
 374         else if (flags & MDS_FMODE_EXEC)
 375                 rc = mdt_write_deny(o);
 376
 377         if (rc)
 378                 RETURN(rc);
 379
 380         rc = mo_open(info->mti_env, mdt_object_child(o),
 381                      created ? flags | MDS_OPEN_CREATED : flags);
 382         if (rc != 0) {
 383                 /* If we allow the client to chgrp (CFS_SETGRP_PERM), but the
 384                  * client does not know which suppgid should be sent to the MDS,
 385                  * or some other(s) changed the target file's GID after this RPC
 386                  * sent to the MDS with the suppgid as the original GID, then we
 387                  * give the client another chance to send the right suppgid. */
 388                 if (rc == -EACCES &&
 389                     allow_client_chgrp(info, lu_ucred(info->mti_env)))
 390                         mdt_set_disposition(info, rep, DISP_OPEN_DENY);
 391
 392                 GOTO(err_out, rc);
 393         }
 394
 395         mfd = mdt_mfd_new(med);
 396         if (mfd == NULL)
 397                 GOTO(err_out, rc = -ENOMEM);
 398
 399         /*
 400          * Keep a reference on this object for this open, and is
 401          * released by mdt_mfd_close().
 402          */
 403         mdt_object_get(info->mti_env, o);
 404         mfd->mfd_object = o;
 405         mfd->mfd_xid = req->rq_xid;
 406
 407         /*
 408          * @flags is always not zero. At least it should be FMODE_READ,
 409          * FMODE_WRITE or MDS_FMODE_EXEC.
 410          */
 411         LASSERT(flags != 0);
 412
 413         /* Open handling. */
 414         mdt_mfd_set_mode(mfd, flags);
 415
 416         atomic_inc(&o->mot_open_count);
 417         if (flags & MDS_OPEN_LEASE)
 418                 atomic_inc(&o->mot_lease_count);
 419
 420         /* replay handle */
 421         if (req_is_replay(req)) {
 422                 struct mdt_file_data *old_mfd;
 423                 /* Check wheather old cookie already exist in
 424                  * the list, becasue when do recovery, client
 425                  * might be disconnected from server, and
 426                  * restart replay, so there maybe some orphan
 427                  * mfd here, we should remove them */
 428                 LASSERT(info->mti_rr.rr_handle != NULL);
 429                 spin_lock(&med->med_open_lock);
 430                 old_mfd = mdt_handle2mfd(med, info->mti_rr.rr_handle, true);
 431                 if (old_mfd != NULL) {
 432                         CDEBUG(D_HA, "delete orphan mfd = %p, fid = "DFID", "
 433                                "cookie = "LPX64"\n", mfd,
 434                                PFID(mdt_object_fid(mfd->mfd_object)),
 435                                info->mti_rr.rr_handle->cookie);
 436                         class_handle_unhash(&old_mfd->mfd_handle);
 437                         list_del_init(&old_mfd->mfd_list);
 438                         spin_unlock(&med->med_open_lock);
 439                         /* no attr update for that close */
 440                         la->la_valid = 0;
 441                         ma->ma_valid |= MA_FLAGS;
 442                         ma->ma_attr_flags |= MDS_RECOV_OPEN;
 443                         mdt_mfd_close(info, old_mfd);
 444                         ma->ma_attr_flags &= ~MDS_RECOV_OPEN;
 445                         ma->ma_valid &= ~MA_FLAGS;
 446                 } else {
 447                         spin_unlock(&med->med_open_lock);
 448                         CDEBUG(D_HA, "orphan mfd not found, fid = "DFID", "
 449                                "cookie = "LPX64"\n",
 450                                PFID(mdt_object_fid(mfd->mfd_object)),
 451                                info->mti_rr.rr_handle->cookie);
 452                 }
 453
 454                 CDEBUG(D_HA, "Store old cookie "LPX64" in new mfd\n",
 455                        info->mti_rr.rr_handle->cookie);
 456
 457                 mfd->mfd_old_handle.cookie = info->mti_rr.rr_handle->cookie;
 458         }
 459
 460         repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
 461
 462         if (req->rq_export->exp_disconnected) {
 463                 spin_lock(&med->med_open_lock);
 464                 class_handle_unhash(&mfd->mfd_handle);
 465                 list_del_init(&mfd->mfd_list);
 466                 spin_unlock(&med->med_open_lock);
 467                 mdt_mfd_close(info, mfd);
 468         } else {
 469                 spin_lock(&med->med_open_lock);
 470                 list_add(&mfd->mfd_list, &med->med_open_head);
 471                 spin_unlock(&med->med_open_lock);
 472         }
 473
 474         mdt_empty_transno(info, rc);
 475
 476         RETURN(rc);
 477
 478 err_out:
 479         if (flags & FMODE_WRITE)
 480                 mdt_write_put(o);
 481         else if (flags & MDS_FMODE_EXEC)
 482                 mdt_write_allow(o);
 483
 484         return rc;
 485 }
 486
 487 static int mdt_finish_open(struct mdt_thread_info *info,
 488                            struct mdt_object *p, struct mdt_object *o,
 489                            __u64 flags, int created, struct ldlm_reply *rep)
 490 {
 491         struct ptlrpc_request   *req = mdt_info_req(info);
 492         struct obd_export       *exp = req->rq_export;
 493         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 494         struct md_attr          *ma  = &info->mti_attr;
 495         struct lu_attr          *la  = &ma->ma_attr;
 496         struct mdt_file_data    *mfd;
 497         struct mdt_body         *repbody;
 498         int                      rc = 0;
 499         int                      isreg, isdir, islnk;
 500         struct list_head        *t;
 501         ENTRY;
 502
 503         LASSERT(ma->ma_valid & MA_INODE);
 504
 505         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 506
 507         isreg = S_ISREG(la->la_mode);
 508         isdir = S_ISDIR(la->la_mode);
 509         islnk = S_ISLNK(la->la_mode);
 510         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 511
 512         /* LU-2275, simulate broken behaviour (esp. prevalent in
 513          * pre-2.4 servers where a very strange reply is sent on error
 514          * that looks like it was actually almost successful and a
 515          * failure at the same time.) */
 516         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_NEGATIVE_POSITIVE)) {
 517                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN |
 518                                                DISP_LOOKUP_NEG |
 519                                                DISP_LOOKUP_POS);
 520
 521                 if (flags & MDS_OPEN_LOCK)
 522                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
 523
 524                 RETURN(-ENOENT);
 525         }
 526
 527         if (exp_connect_rmtclient(exp)) {
 528                 void *buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 529
 530                 rc = mdt_pack_remote_perm(info, o, buf);
 531                 if (rc) {
 532                         repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
 533                         repbody->mbo_aclsize = 0;
 534                 } else {
 535                         repbody->mbo_valid |= OBD_MD_FLRMTPERM;
 536                         repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
 537                 }
 538         }
 539 #ifdef CONFIG_FS_POSIX_ACL
 540         else if (exp_connect_flags(exp) & OBD_CONNECT_ACL)
 541                 rc = mdt_pack_acl2body(info, repbody, o,
 542                                        exp->exp_target_data.ted_nodemap);
 543 #endif
 544
 545         /*
 546          * If we are following a symlink, don't open; and do not return open
 547          * handle for special nodes as client required.
 548          */
 549         if (islnk || (!isreg && !isdir &&
 550             (exp_connect_flags(req->rq_export) & OBD_CONNECT_NODEVOH))) {
 551                 lustre_msg_set_transno(req->rq_repmsg, 0);
 552                 RETURN(0);
 553         }
 554
 555         /*
 556          * We need to return the existing object's fid back, so it is done here,
 557          * after preparing the reply.
 558          */
 559         if (!created && (flags & MDS_OPEN_EXCL) && (flags & MDS_OPEN_CREAT))
 560                 RETURN(-EEXIST);
 561
 562         /* This can't be done earlier, we need to return reply body */
 563         if (isdir) {
 564                 if (flags & (MDS_OPEN_CREAT | FMODE_WRITE)) {
 565                         /* We are trying to create or write an existing dir. */
 566                         RETURN(-EISDIR);
 567                 }
 568         } else if (flags & MDS_OPEN_DIRECTORY)
 569                 RETURN(-ENOTDIR);
 570
 571         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
 572                                  OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE)) {
 573                 RETURN(-EAGAIN);
 574         }
 575
 576         mfd = NULL;
 577         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 578                 spin_lock(&med->med_open_lock);
 579                 list_for_each(t, &med->med_open_head) {
 580                         mfd = list_entry(t, struct mdt_file_data, mfd_list);
 581                         if (mfd->mfd_xid == req->rq_xid)
 582                                 break;
 583                         mfd = NULL;
 584                 }
 585                 spin_unlock(&med->med_open_lock);
 586
 587                 if (mfd != NULL) {
 588                         repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
 589                         /* set repbody->ea_size for resent case */
 590                         if (ma->ma_valid & MA_LOV) {
 591                                 LASSERT(ma->ma_lmm_size != 0);
 592                                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 593                                 if (isdir)
 594                                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 595                                 else
 596                                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 597                         }
 598                         mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 599                         RETURN(0);
 600                 }
 601         }
 602
 603         rc = mdt_mfd_open(info, p, o, flags, created, rep);
 604         if (!rc)
 605                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 606
 607         RETURN(rc);
 608 }
 609
 610 void mdt_reconstruct_open(struct mdt_thread_info *info,
 611                           struct mdt_lock_handle *lhc)
 612 {
 613         const struct lu_env *env = info->mti_env;
 614         struct mdt_device       *mdt  = info->mti_mdt;
 615         struct req_capsule      *pill = info->mti_pill;
 616         struct ptlrpc_request   *req  = mdt_info_req(info);
 617         struct tg_export_data   *ted  = &req->rq_export->exp_target_data;
 618         struct lsd_client_data  *lcd  = ted->ted_lcd;
 619         struct md_attr          *ma   = &info->mti_attr;
 620         struct mdt_reint_record *rr   = &info->mti_rr;
 621         __u64                   flags = info->mti_spec.sp_cr_flags;
 622         struct ldlm_reply       *ldlm_rep;
 623         struct mdt_object       *parent;
 624         struct mdt_object       *child;
 625         struct mdt_body         *repbody;
 626         int                      rc;
 627         ENTRY;
 628
 629         LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
 630         ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
 631         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 632
 633         ma->ma_need = MA_INODE | MA_HSM;
 634         ma->ma_valid = 0;
 635
 636         mdt_req_from_lcd(req, lcd);
 637         mdt_set_disposition(info, ldlm_rep, lcd->lcd_last_data);
 638
 639         CDEBUG(D_INODE, "This is reconstruct open: disp="LPX64", result=%d\n",
 640                ldlm_rep->lock_policy_res1, req->rq_status);
 641
 642         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
 643             req->rq_status != 0)
 644                 /* We did not create successfully, return error to client. */
 645                 GOTO(out, rc = req->rq_status);
 646
 647         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
 648                 struct obd_export *exp = req->rq_export;
 649                 /*
 650                  * We failed after creation, but we do not know in which step
 651                  * we failed. So try to check the child object.
 652                  */
 653                 parent = mdt_object_find(env, mdt, rr->rr_fid1);
 654                 if (IS_ERR(parent)) {
 655                         rc = PTR_ERR(parent);
 656                         LCONSOLE_WARN("Parent "DFID" lookup error %d."
 657                                       " Evicting client %s with export %s.\n",
 658                                       PFID(rr->rr_fid1), rc,
 659                                       obd_uuid2str(&exp->exp_client_uuid),
 660                                       obd_export_nid2str(exp));
 661                         mdt_export_evict(exp);
 662                         RETURN_EXIT;
 663                 }
 664
 665                 child = mdt_object_find(env, mdt, rr->rr_fid2);
 666                 if (IS_ERR(child)) {
 667                         rc = PTR_ERR(child);
 668                         LCONSOLE_WARN("cannot lookup child "DFID": rc = %d; "
 669                                       "evicting client %s with export %s\n",
 670                                       PFID(rr->rr_fid2), rc,
 671                                       obd_uuid2str(&exp->exp_client_uuid),
 672                                       obd_export_nid2str(exp));
 673                         mdt_object_put(env, parent);
 674                         mdt_export_evict(exp);
 675                         RETURN_EXIT;
 676                 }
 677
 678                 if (unlikely(mdt_object_remote(child))) {
 679                         /* the child object was created on remote server */
 680                         if (!mdt_is_dne_client(exp)) {
 681                                 /* Return -EIO for old client */
 682                                 mdt_object_put(env, parent);
 683                                 mdt_object_put(env, child);
 684                                 GOTO(out, rc = -EIO);
 685                         }
 686                         repbody->mbo_fid1 = *rr->rr_fid2;
 687                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 688                         rc = 0;
 689                 } else {
 690                         if (mdt_object_exists(child)) {
 691                                 mdt_prep_ma_buf_from_rep(info, child, ma);
 692                                 rc = mdt_attr_get_complex(info, child, ma);
 693                                 if (rc == 0)
 694                                         rc = mdt_finish_open(info, parent,
 695                                                              child, flags,
 696                                                              1, ldlm_rep);
 697                         } else {
 698                                 /* the child does not exist, we should do
 699                                  * regular open */
 700                                 mdt_object_put(env, parent);
 701                                 mdt_object_put(env, child);
 702                                 GOTO(regular_open, 0);
 703                         }
 704                 }
 705                 mdt_object_put(env, parent);
 706                 mdt_object_put(env, child);
 707                 GOTO(out, rc);
 708         } else {
 709 regular_open:
 710                 /* We did not try to create, so we are a pure open */
 711                 rc = mdt_reint_open(info, lhc);
 712         }
 713
 714         EXIT;
 715 out:
 716         req->rq_status = rc;
 717         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
 718         LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
 719 }
 720
 721 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep)
 722 {
 723         __u64                    flags = info->mti_spec.sp_cr_flags;
 724         struct mdt_reint_record *rr = &info->mti_rr;
 725         struct md_attr          *ma = &info->mti_attr;
 726         struct mdt_object       *o;
 727         int                      rc;
 728         ENTRY;
 729
 730         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
 731         if (IS_ERR(o))
 732                 RETURN(rc = PTR_ERR(o));
 733
 734         if (unlikely(mdt_object_remote(o))) {
 735                 /* the child object was created on remote server */
 736                 struct mdt_body *repbody;
 737
 738                 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 739                                                 DISP_LOOKUP_EXECD |
 740                                                 DISP_LOOKUP_POS));
 741                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 742                 repbody->mbo_fid1 = *rr->rr_fid2;
 743                 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 744                 rc = 0;
 745         } else {
 746                 if (mdt_object_exists(o)) {
 747                         mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 748                                                         DISP_LOOKUP_EXECD |
 749                                                         DISP_LOOKUP_POS));
 750                         mdt_prep_ma_buf_from_rep(info, o, ma);
 751                         rc = mdt_attr_get_complex(info, o, ma);
 752                         if (rc == 0)
 753                                 rc = mdt_finish_open(info, NULL, o, flags, 0,
 754                                                      rep);
 755                 } else {
 756                         rc = -ENOENT;
 757                 }
 758         }
 759
 760         mdt_object_put(info->mti_env, o);
 761         RETURN(rc);
 762 }
 763
 764 /* lock object for open */
 765 static int mdt_object_open_lock(struct mdt_thread_info *info,
 766                                 struct mdt_object *obj,
 767                                 struct mdt_lock_handle *lhc,
 768                                 __u64 *ibits)
 769 {
 770         struct md_attr  *ma = &info->mti_attr;
 771         __u64            open_flags = info->mti_spec.sp_cr_flags;
 772         ldlm_mode_t      lm = LCK_CR;
 773         bool             acq_lease = !!(open_flags & MDS_OPEN_LEASE);
 774         bool             try_layout = false;
 775         bool             create_layout = false;
 776         int              rc = 0;
 777         ENTRY;
 778
 779         *ibits = 0;
 780         mdt_lock_handle_init(lhc);
 781
 782         if (req_is_replay(mdt_info_req(info)))
 783                 RETURN(0);
 784
 785         if (S_ISREG(lu_object_attr(&obj->mot_obj))) {
 786                 if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV) &&
 787                     md_should_create(open_flags))
 788                         create_layout = true;
 789                 if (exp_connect_layout(info->mti_exp) && !create_layout &&
 790                     ma->ma_need & MA_LOV)
 791                         try_layout = true;
 792         }
 793
 794         if (acq_lease) {
 795                 /* lease open, acquire write mode of open sem */
 796                 down_write(&obj->mot_open_sem);
 797
 798                 /* Lease exists and ask for new lease */
 799                 if (atomic_read(&obj->mot_lease_count) > 0) {
 800                         /* only exclusive open is supported, so lease
 801                          * are conflicted to each other */
 802                         GOTO(out, rc = -EBUSY);
 803                 }
 804
 805                 /* Lease must be with open lock */
 806                 if (!(open_flags & MDS_OPEN_LOCK)) {
 807                         CERROR("Request lease for file:"DFID ", but open lock "
 808                                 "is missed, open_flags = "LPO64".\n",
 809                                 PFID(mdt_object_fid(obj)), open_flags);
 810                         GOTO(out, rc = -EPROTO);
 811                 }
 812
 813                 /* XXX: only exclusive open is supported. */
 814                 lm = LCK_EX;
 815                 *ibits = MDS_INODELOCK_OPEN;
 816
 817                 /* never grant LCK_EX layout lock to client */
 818                 try_layout = false;
 819         } else { /* normal open */
 820                 /* normal open holds read mode of open sem */
 821                 down_read(&obj->mot_open_sem);
 822
 823                 if (open_flags & MDS_OPEN_LOCK) {
 824                         if (open_flags & FMODE_WRITE)
 825                                 lm = LCK_CW;
 826                         else if (open_flags & MDS_FMODE_EXEC)
 827                                 lm = LCK_PR;
 828                         else
 829                                 lm = LCK_CR;
 830
 831                         *ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN;
 832                 } else if (atomic_read(&obj->mot_lease_count) > 0) {
 833                         if (open_flags & FMODE_WRITE)
 834                                 lm = LCK_CW;
 835                         else
 836                                 lm = LCK_CR;
 837
 838                         /* revoke lease */
 839                         *ibits = MDS_INODELOCK_OPEN;
 840                         try_layout = false;
 841
 842                         lhc = &info->mti_lh[MDT_LH_LOCAL];
 843                 }
 844                 CDEBUG(D_INODE, "normal open:"DFID" lease count: %d, lm: %d\n",
 845                         PFID(mdt_object_fid(obj)),
 846                         atomic_read(&obj->mot_open_count), lm);
 847         }
 848
 849         mdt_lock_reg_init(lhc, lm);
 850
 851         /* one problem to return layout lock on open is that it may result
 852          * in too many layout locks cached on the client side. */
 853         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_OPEN) && try_layout) {
 854                 /* return lookup lock to validate inode at the client side,
 855                  * this is pretty important otherwise mdt will return layout
 856                  * lock for each open.
 857                  * However this is a double-edged sword because changing
 858                  * permission will revoke huge # of LOOKUP locks. */
 859                 *ibits |= MDS_INODELOCK_LAYOUT | MDS_INODELOCK_LOOKUP;
 860                 if (!mdt_object_lock_try(info, obj, lhc, *ibits,
 861                                          MDT_CROSS_LOCK)) {
 862                         *ibits &= ~(MDS_INODELOCK_LAYOUT|MDS_INODELOCK_LOOKUP);
 863                         if (*ibits != 0)
 864                                 rc = mdt_object_lock(info, obj, lhc, *ibits,
 865                                                 MDT_CROSS_LOCK);
 866                 }
 867         } else if (*ibits != 0) {
 868                 rc = mdt_object_lock(info, obj, lhc, *ibits, MDT_CROSS_LOCK);
 869         }
 870
 871         CDEBUG(D_INODE, "Requested bits lock:"DFID ", ibits = "LPX64
 872                 ", open_flags = "LPO64", try_layout = %d, rc = %d\n",
 873                 PFID(mdt_object_fid(obj)), *ibits, open_flags, try_layout, rc);
 874
 875         /* will change layout, revoke layout locks by enqueuing EX lock. */
 876         if (rc == 0 && create_layout) {
 877                 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LAYOUT];
 878
 879                 CDEBUG(D_INODE, "Will create layout, get EX layout lock:"DFID
 880                         ", open_flags = "LPO64"\n",
 881                         PFID(mdt_object_fid(obj)), open_flags);
 882
 883                 /* We cannot enqueue another lock for the same resource we
 884                  * already have a lock for, due to mechanics of waiting list
 885                  * iterating in ldlm, see LU-3601.
 886                  * As such we'll drop the open lock we just got above here,
 887                  * it's ok not to have this open lock as it's main purpose is to
 888                  * flush unused cached client open handles. */
 889                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
 890                         mdt_object_unlock(info, obj, lhc, 1);
 891
 892                 LASSERT(!try_layout);
 893                 mdt_lock_handle_init(ll);
 894                 mdt_lock_reg_init(ll, LCK_EX);
 895                 rc = mdt_object_lock(info, obj, ll, MDS_INODELOCK_LAYOUT,
 896                                         MDT_LOCAL_LOCK);
 897
 898                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LL_BLOCK, 2);
 899         }
 900
 901         /* Check if there is any other open handles after acquiring
 902          * open lock. At this point, caching open handles have been revoked
 903          * by open lock.
 904          * XXX: Now only exclusive open is supported. Need to check the
 905          * type of open for generic lease support. */
 906         if (rc == 0 && acq_lease) {
 907                 struct ptlrpc_request *req = mdt_info_req(info);
 908                 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
 909                 struct mdt_file_data *mfd;
 910                 bool is_replay_or_resent;
 911                 int open_count = 0;
 912
 913                 /* For lease: application can open a file and then apply lease,
 914                  * @handle contains original open handle in that case.
 915                  * In recovery, open REQ will be replayed and the lease REQ may
 916                  * be resent that means the open handle is already stale, so we
 917                  * need to fix it up here by finding new handle. */
 918                 is_replay_or_resent = req_is_replay(req) ||
 919                         lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
 920
 921                 /* if the request is _not_ a replay request, rr_handle
 922                  * may be used to hold an openhandle which is issuing the
 923                  * lease request, so that this openhandle doesn't count. */
 924                 mfd = mdt_handle2mfd(med, info->mti_rr.rr_handle,
 925                                      is_replay_or_resent);
 926                 if (mfd != NULL)
 927                         ++open_count;
 928
 929                 CDEBUG(D_INODE, "acq_lease "DFID": openers: %d, want: %d\n",
 930                         PFID(mdt_object_fid(obj)),
 931                         atomic_read(&obj->mot_open_count), open_count);
 932
 933                 if (atomic_read(&obj->mot_open_count) > open_count)
 934                         GOTO(out, rc = -EBUSY);
 935         }
 936         GOTO(out, rc);
 937
 938 out:
 939         RETURN(rc);
 940 }
 941
 942 static void mdt_object_open_unlock(struct mdt_thread_info *info,
 943                                    struct mdt_object *obj,
 944                                    struct mdt_lock_handle *lhc,
 945                                    __u64 ibits, int rc)
 946 {
 947         __u64 open_flags = info->mti_spec.sp_cr_flags;
 948         struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LOCAL];
 949         ENTRY;
 950
 951         if (req_is_replay(mdt_info_req(info)))
 952                 RETURN_EXIT;
 953
 954         /* Release local lock - the lock put in MDT_LH_LOCAL will never
 955          * return to client side. */
 956         if (lustre_handle_is_used(&ll->mlh_reg_lh))
 957                 mdt_object_unlock(info, obj, ll, 1);
 958
 959         ll = &info->mti_lh[MDT_LH_LAYOUT];
 960         /* Release local layout lock, layout was created */
 961         if (lustre_handle_is_used(&ll->mlh_reg_lh)) {
 962                 LASSERT(!(ibits & MDS_INODELOCK_LAYOUT));
 963                 mdt_object_unlock(info, obj, ll, 1);
 964         }
 965
 966         if (open_flags & MDS_OPEN_LEASE)
 967                 up_write(&obj->mot_open_sem);
 968         else
 969                 up_read(&obj->mot_open_sem);
 970
 971         /* Cross-ref case, the lock should be returned to the client */
 972         if (ibits == 0 || rc == -MDT_EREMOTE_OPEN)
 973                 RETURN_EXIT;
 974
 975         if (!(open_flags & MDS_OPEN_LOCK) && !(ibits & MDS_INODELOCK_LAYOUT)) {
 976                 /* for the open request, the lock will only return to client
 977                  * if open or layout lock is granted. */
 978                 rc = 1;
 979         }
 980
 981         if (rc != 0 || !lustre_handle_is_used(&lhc->mlh_reg_lh)) {
 982                 struct ldlm_reply       *ldlm_rep;
 983
 984                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
 985                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
 986                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
 987                         mdt_object_unlock(info, obj, lhc, 1);
 988         }
 989         RETURN_EXIT;
 990 }
 991
 992 /**
 993  * Check release is permitted for the current HSM flags.
 994  */
 995 static bool mdt_hsm_release_allow(const struct md_attr *ma)
 996 {
 997         if (!(ma->ma_valid & MA_HSM))
 998                 return false;
 999
1000         if (ma->ma_hsm.mh_flags & (HS_DIRTY|HS_NORELEASE|HS_LOST))
1001                 return false;
1002
1003         if (!(ma->ma_hsm.mh_flags & HS_ARCHIVED))
1004                 return false;
1005
1006         return true;
1007 }
1008
1009 static int mdt_open_by_fid_lock(struct mdt_thread_info *info,
1010                                 struct ldlm_reply *rep,
1011                                 struct mdt_lock_handle *lhc)
1012 {
1013         const struct lu_env     *env   = info->mti_env;
1014         struct mdt_device       *mdt   = info->mti_mdt;
1015         __u64                    flags = info->mti_spec.sp_cr_flags;
1016         struct mdt_reint_record *rr    = &info->mti_rr;
1017         struct md_attr          *ma    = &info->mti_attr;
1018         struct mdt_object       *parent= NULL;
1019         struct mdt_object       *o;
1020         int                      rc;
1021         int                      object_locked = 0;
1022         __u64                    ibits = 0;
1023         ENTRY;
1024
1025         if (md_should_create(flags) && !(flags & MDS_OPEN_HAS_EA)) {
1026                 if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1027                         parent = mdt_object_find(env, mdt, rr->rr_fid1);
1028                         if (IS_ERR(parent)) {
1029                                 CDEBUG(D_INODE, "Fail to find parent "DFID
1030                                        " for anonymous created %ld, try to"
1031                                        " use server-side parent.\n",
1032                                        PFID(rr->rr_fid1), PTR_ERR(parent));
1033                                 parent = NULL;
1034                         }
1035                 }
1036                 if (parent == NULL)
1037                         ma->ma_need |= MA_PFID;
1038         }
1039
1040         o = mdt_object_find(env, mdt, rr->rr_fid2);
1041         if (IS_ERR(o))
1042                 GOTO(out_parent_put, rc = PTR_ERR(o));
1043
1044         if (mdt_object_remote(o)) {
1045                 CDEBUG(D_INFO, "%s: "DFID" is on remote MDT.\n",
1046                        mdt_obd_name(info->mti_mdt),
1047                        PFID(rr->rr_fid2));
1048                 GOTO(out, rc = -EREMOTE);
1049         } else if (!mdt_object_exists(o)) {
1050                 mdt_set_disposition(info, rep,
1051                                     DISP_IT_EXECD |
1052                                     DISP_LOOKUP_EXECD |
1053                                     DISP_LOOKUP_NEG);
1054                 GOTO(out, rc = -ENOENT);
1055         }
1056
1057         mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1058
1059         mdt_prep_ma_buf_from_rep(info, o, ma);
1060         if (flags & MDS_OPEN_RELEASE)
1061                 ma->ma_need |= MA_HSM;
1062         rc = mdt_attr_get_complex(info, o, ma);
1063         if (rc)
1064                 GOTO(out, rc);
1065
1066         /* We should not change file's existing LOV EA */
1067         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
1068             flags & MDS_OPEN_HAS_EA && ma->ma_valid & MA_LOV)
1069                 GOTO(out, rc = -EEXIST);
1070
1071         /* If a release request, check file flags are fine and ask for an
1072          * exclusive open access. */
1073         if (flags & MDS_OPEN_RELEASE && !mdt_hsm_release_allow(ma))
1074                 GOTO(out, rc = -EPERM);
1075
1076         rc = mdt_check_resent_lock(info, o, lhc);
1077         if (rc < 0) {
1078                 GOTO(out, rc);
1079         } else if (rc > 0) {
1080                 rc = mdt_object_open_lock(info, o, lhc, &ibits);
1081                 object_locked = 1;
1082                 if (rc)
1083                         GOTO(out_unlock, rc);
1084         }
1085
1086         if (ma->ma_valid & MA_PFID) {
1087                 parent = mdt_object_find(env, mdt, &ma->ma_pfid);
1088                 if (IS_ERR(parent)) {
1089                         CDEBUG(D_INODE, "Fail to find parent "DFID
1090                                " for anonymous created %ld, try to"
1091                                " use system default.\n",
1092                                PFID(&ma->ma_pfid), PTR_ERR(parent));
1093                         parent = NULL;
1094                 }
1095         }
1096
1097         rc = mdt_finish_open(info, parent, o, flags, 0, rep);
1098         if (!rc) {
1099                 mdt_set_disposition(info, rep, DISP_LOOKUP_POS);
1100                 if (flags & MDS_OPEN_LOCK)
1101                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1102                 if (flags & MDS_OPEN_LEASE)
1103                         mdt_set_disposition(info, rep, DISP_OPEN_LEASE);
1104         }
1105         GOTO(out_unlock, rc);
1106
1107 out_unlock:
1108         if (object_locked)
1109                 mdt_object_open_unlock(info, o, lhc, ibits, rc);
1110 out:
1111         mdt_object_put(env, o);
1112 out_parent_put:
1113         if (parent != NULL)
1114                 mdt_object_put(env, parent);
1115         return rc;
1116 }
1117
1118 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1119 static int mdt_cross_open(struct mdt_thread_info *info,
1120                           const struct lu_fid *parent_fid,
1121                           const struct lu_fid *fid,
1122                           struct ldlm_reply *rep, __u32 flags)
1123 {
1124         struct md_attr    *ma = &info->mti_attr;
1125         struct mdt_object *o;
1126         int                rc;
1127         ENTRY;
1128
1129         o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1130         if (IS_ERR(o))
1131                 RETURN(rc = PTR_ERR(o));
1132
1133         if (mdt_object_remote(o)) {
1134                 /* Something is wrong here, the object is on another MDS! */
1135                 CERROR("%s: "DFID" isn't on this server!: rc = %d\n",
1136                        mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1137                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1138                                 &o->mot_obj,
1139                                 "Object isn't on this server! FLD error?\n");
1140                 rc = -EFAULT;
1141         } else {
1142                 if (mdt_object_exists(o)) {
1143                         /* Do permission check for cross-open. */
1144                         rc = mo_permission(info->mti_env, NULL,
1145                                            mdt_object_child(o),
1146                                            NULL, flags | MDS_OPEN_CROSS);
1147                         if (rc)
1148                                 goto out;
1149
1150                         mdt_prep_ma_buf_from_rep(info, o, ma);
1151                         rc = mdt_attr_get_complex(info, o, ma);
1152                         if (rc != 0)
1153                                 GOTO(out, rc);
1154
1155                         rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
1156                 } else {
1157                         /*
1158                          * Something is wrong here. lookup was positive but
1159                          * there is no object!
1160                          */
1161                         CERROR("%s: "DFID" doesn't exist!: rc = %d\n",
1162                               mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1163                         rc = -EFAULT;
1164                 }
1165         }
1166 out:
1167         mdt_object_put(info->mti_env, o);
1168         RETURN(rc);
1169 }
1170
1171 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1172 {
1173         struct mdt_device       *mdt = info->mti_mdt;
1174         struct ptlrpc_request   *req = mdt_info_req(info);
1175         struct mdt_object       *parent;
1176         struct mdt_object       *child;
1177         struct mdt_lock_handle  *lh;
1178         struct ldlm_reply       *ldlm_rep;
1179         struct mdt_body         *repbody;
1180         struct lu_fid           *child_fid = &info->mti_tmp_fid1;
1181         struct md_attr          *ma = &info->mti_attr;
1182         __u64                    create_flags = info->mti_spec.sp_cr_flags;
1183         __u64                    ibits = 0;
1184         struct mdt_reint_record *rr = &info->mti_rr;
1185         int                      result, rc;
1186         int                      created = 0;
1187         int                      object_locked = 0;
1188         __u32                    msg_flags;
1189         ENTRY;
1190
1191         OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1192                                (obd_timeout + 1) / 4);
1193
1194         mdt_counter_incr(req, LPROC_MDT_OPEN);
1195         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1196
1197         ma->ma_need = MA_INODE;
1198         ma->ma_valid = 0;
1199
1200         LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1201         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1202
1203         if (unlikely(create_flags & MDS_OPEN_JOIN_FILE)) {
1204                 CERROR("file join is not supported anymore.\n");
1205                 GOTO(out, result = err_serious(-EOPNOTSUPP));
1206         }
1207         msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1208
1209         if ((create_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1210             info->mti_spec.u.sp_ea.eadata == NULL)
1211                 GOTO(out, result = err_serious(-EINVAL));
1212
1213         CDEBUG(D_INODE, "I am going to open "DFID"/("DNAME"->"DFID") "
1214                "cr_flag="LPO64" mode=0%06o msg_flag=0x%x\n",
1215                PFID(rr->rr_fid1), PNAME(&rr->rr_name),
1216                PFID(rr->rr_fid2), create_flags,
1217                ma->ma_attr.la_mode, msg_flags);
1218
1219         if (info->mti_cross_ref) {
1220                 /* This is cross-ref open */
1221                 mdt_set_disposition(info, ldlm_rep,
1222                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD |
1223                              DISP_LOOKUP_POS));
1224                 result = mdt_cross_open(info, rr->rr_fid2, rr->rr_fid1,
1225                                         ldlm_rep, create_flags);
1226                 GOTO(out, result);
1227         } else if (req_is_replay(req) ||
1228             (req->rq_export->exp_libclient && create_flags & MDS_OPEN_HAS_EA)) {
1229                 /* This is a replay request or from liblustre with ea. */
1230                 result = mdt_open_by_fid(info, ldlm_rep);
1231
1232                 if (result != -ENOENT) {
1233                         if (req->rq_export->exp_libclient &&
1234                             create_flags & MDS_OPEN_HAS_EA)
1235                                 GOTO(out, result = 0);
1236                         GOTO(out, result);
1237                 }
1238                 /* We didn't find the correct object, so we need to re-create it
1239                  * via a regular replay. */
1240                 if (!(create_flags & MDS_OPEN_CREAT)) {
1241                         DEBUG_REQ(D_ERROR, req,
1242                                   "OPEN & CREAT not in open replay/by_fid.");
1243                         GOTO(out, result = -EFAULT);
1244                 }
1245                 CDEBUG(D_INFO, "No object(1), continue as regular open.\n");
1246         } else if (create_flags & (MDS_OPEN_BY_FID | MDS_OPEN_LOCK)) {
1247                 /*
1248                  * MDS_OPEN_LOCK is checked for backward compatibility with 2.1
1249                  * client.
1250                  */
1251                 result = mdt_open_by_fid_lock(info, ldlm_rep, lhc);
1252                 if (result < 0)
1253                         CDEBUG(D_INFO, "no object for "DFID": %d\n",
1254                                PFID(rr->rr_fid2), result);
1255                 GOTO(out, result);
1256         }
1257
1258         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1259                 GOTO(out, result = err_serious(-ENOMEM));
1260
1261         mdt_set_disposition(info, ldlm_rep,
1262                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1263
1264         if (!lu_name_is_valid(&rr->rr_name))
1265                 GOTO(out, result = -EPROTO);
1266
1267         lh = &info->mti_lh[MDT_LH_PARENT];
1268         mdt_lock_pdo_init(lh,
1269                           (create_flags & MDS_OPEN_CREAT) ? LCK_PW : LCK_PR,
1270                           &rr->rr_name);
1271
1272         parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
1273                                       MDS_INODELOCK_UPDATE);
1274         if (IS_ERR(parent))
1275                 GOTO(out, result = PTR_ERR(parent));
1276
1277         /* get and check version of parent */
1278         result = mdt_version_get_check(info, parent, 0);
1279         if (result)
1280                 GOTO(out_parent, result);
1281
1282         fid_zero(child_fid);
1283
1284         result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1285                             &rr->rr_name, child_fid, &info->mti_spec);
1286
1287         LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1288                  "looking for "DFID"/"DNAME", found FID = "DFID"\n",
1289                  PFID(mdt_object_fid(parent)), PNAME(&rr->rr_name),
1290                  PFID(child_fid));
1291
1292         if (result != 0 && result != -ENOENT && result != -ESTALE)
1293                 GOTO(out_parent, result);
1294
1295         if (result == -ENOENT || result == -ESTALE) {
1296                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1297                 if (result == -ESTALE) {
1298                         /*
1299                          * -ESTALE means the parent is a dead(unlinked) dir, so
1300                          * it should return -ENOENT to in accordance with the
1301                          * original mds implementaion.
1302                          */
1303                         GOTO(out_parent, result = -ENOENT);
1304                 }
1305                 if (!(create_flags & MDS_OPEN_CREAT))
1306                         GOTO(out_parent, result);
1307                 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1308                         GOTO(out_parent, result = -EROFS);
1309                 *child_fid = *info->mti_rr.rr_fid2;
1310                 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1311                          PFID(child_fid));
1312                 /* In the function below, .hs_keycmp resolves to
1313                  * lu_obj_hop_keycmp() */
1314                 /* coverity[overrun-buffer-val] */
1315                 child = mdt_object_new(info->mti_env, mdt, child_fid);
1316         } else {
1317                 /*
1318                  * Check for O_EXCL is moved to the mdt_finish_open(), we need to
1319                  * return FID back in that case.
1320                  */
1321                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1322                 child = mdt_object_find(info->mti_env, mdt, child_fid);
1323         }
1324         if (IS_ERR(child))
1325                 GOTO(out_parent, result = PTR_ERR(child));
1326
1327         /** check version of child  */
1328         rc = mdt_version_get_check(info, child, 1);
1329         if (rc)
1330                 GOTO(out_child, result = rc);
1331
1332         if (result == -ENOENT) {
1333                 /* Create under OBF and .lustre is not permitted */
1334                 if (!fid_is_md_operative(rr->rr_fid1))
1335                         GOTO(out_child, result = -EPERM);
1336
1337                 /* save versions in reply */
1338                 mdt_version_get_save(info, parent, 0);
1339                 mdt_version_get_save(info, child, 1);
1340
1341                 /* version of child will be changed */
1342                 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
1343
1344                 /* Not found and with MDS_OPEN_CREAT: let's create it. */
1345                 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1346
1347                 /* Let lower layers know what is lock mode on directory. */
1348                 info->mti_spec.sp_cr_mode =
1349                         mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
1350
1351                 /*
1352                  * Do not perform lookup sanity check. We know that name does
1353                  * not exist.
1354                  */
1355                 info->mti_spec.sp_cr_lookup = 0;
1356                 info->mti_spec.sp_feat = &dt_directory_features;
1357
1358                 result = mdo_create(info->mti_env,
1359                                     mdt_object_child(parent),
1360                                     &rr->rr_name,
1361                                     mdt_object_child(child),
1362                                     &info->mti_spec,
1363                                     &info->mti_attr);
1364                 if (result == -ERESTART) {
1365                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1366                         GOTO(out_child, result);
1367                 } else {
1368                         mdt_prep_ma_buf_from_rep(info, child, ma);
1369                         /* XXX: we should call this once, see few lines below */
1370                         if (result == 0)
1371                                 result = mdt_attr_get_complex(info, child, ma);
1372
1373                         if (result != 0)
1374                                 GOTO(out_child, result);
1375                 }
1376                 created = 1;
1377         } else {
1378                 /*
1379                  * The object is on remote node, return its FID for remote open.
1380                  */
1381                 if (mdt_object_remote(child)) {
1382                         /*
1383                          * Check if this lock already was sent to client and
1384                          * this is resent case. For resent case do not take lock
1385                          * again, use what is already granted.
1386                          */
1387                         LASSERT(lhc != NULL);
1388
1389                         rc = mdt_check_resent_lock(info, child, lhc);
1390                         if (rc < 0) {
1391                                 GOTO(out_child, result = rc);
1392                         } else if (rc > 0) {
1393                                 mdt_lock_handle_init(lhc);
1394                                 mdt_lock_reg_init(lhc, LCK_PR);
1395
1396                                 rc = mdt_object_lock(info, child, lhc,
1397                                                      MDS_INODELOCK_LOOKUP,
1398                                                      MDT_CROSS_LOCK);
1399                         }
1400                         repbody->mbo_fid1 = *mdt_object_fid(child);
1401                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1402                         if (rc != 0)
1403                                 result = rc;
1404                         else
1405                                 result = -MDT_EREMOTE_OPEN;
1406                         GOTO(out_child, result);
1407                 } else if (mdt_object_exists(child)) {
1408                         /* We have to get attr & LOV EA & HSM for this
1409                          * object. */
1410                         mdt_prep_ma_buf_from_rep(info, child, ma);
1411                         ma->ma_need |= MA_HSM;
1412                         result = mdt_attr_get_complex(info, child, ma);
1413                         if (result != 0)
1414                                 GOTO(out_child, result);
1415                 } else {
1416                         /* Object does not exist. Likely FS corruption. */
1417                         CERROR("%s: name '"DNAME"' present, but FID "
1418                                DFID" is invalid\n", mdt_obd_name(info->mti_mdt),
1419                                PNAME(&rr->rr_name), PFID(child_fid));
1420                         GOTO(out_child, result = -EIO);
1421                 }
1422         }
1423
1424         rc = mdt_check_resent_lock(info, child, lhc);
1425         if (rc < 0) {
1426                 GOTO(out_child, result = rc);
1427         } else if (rc == 0) {
1428                 /* the open lock might already be gotten in
1429                  * ldlm_handle_enqueue() */
1430                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1431                 if (create_flags & MDS_OPEN_LOCK)
1432                         mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1433         } else {
1434                 /* get openlock if this isn't replay and client requested it */
1435                 if (!req_is_replay(req)) {
1436                         rc = mdt_object_open_lock(info, child, lhc, &ibits);
1437                         object_locked = 1;
1438                         if (rc != 0)
1439                                 GOTO(out_child_unlock, result = rc);
1440                         else if (create_flags & MDS_OPEN_LOCK)
1441                                 mdt_set_disposition(info, ldlm_rep,
1442                                                     DISP_OPEN_LOCK);
1443                 }
1444         }
1445         /* Try to open it now. */
1446         rc = mdt_finish_open(info, parent, child, create_flags,
1447                              created, ldlm_rep);
1448         if (rc) {
1449                 result = rc;
1450                 /* openlock will be released if mdt_finish_open failed */
1451                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1452
1453                 if (created && create_flags & MDS_OPEN_VOLATILE) {
1454                         CERROR("%s: cannot open volatile file "DFID", orphan "
1455                                "file will be left in PENDING directory until "
1456                                "next reboot, rc = %d\n", mdt_obd_name(mdt),
1457                                PFID(mdt_object_fid(child)), rc);
1458                         GOTO(out_child_unlock, result);
1459                 }
1460
1461                 if (created) {
1462                         ma->ma_need = 0;
1463                         ma->ma_valid = 0;
1464                         rc = mdo_unlink(info->mti_env,
1465                                         mdt_object_child(parent),
1466                                         mdt_object_child(child),
1467                                         &rr->rr_name,
1468                                         &info->mti_attr, 0);
1469                         if (rc != 0)
1470                                 CERROR("%s: "DFID" cleanup of open: rc = %d\n",
1471                                        mdt_obd_name(info->mti_mdt),
1472                                        PFID(mdt_object_fid(child)), rc);
1473                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1474                 }
1475         }
1476         EXIT;
1477 out_child_unlock:
1478         if (object_locked)
1479                 mdt_object_open_unlock(info, child, lhc, ibits, result);
1480 out_child:
1481         mdt_object_put(info->mti_env, child);
1482 out_parent:
1483         mdt_object_unlock_put(info, parent, lh, result || !created);
1484 out:
1485         if (result)
1486                 lustre_msg_set_transno(req->rq_repmsg, 0);
1487         return result;
1488 }
1489
1490 /**
1491  * Create an orphan object use local root.
1492  */
1493 static struct mdt_object *mdt_orphan_open(struct mdt_thread_info *info,
1494                                           struct mdt_device *mdt,
1495                                           const struct lu_fid *fid,
1496                                           struct md_attr *attr, fmode_t fmode)
1497 {
1498         const struct lu_env *env = info->mti_env;
1499         struct md_op_spec *spec = &info->mti_spec;
1500         struct lu_fid *local_root_fid = &info->mti_tmp_fid1;
1501         struct mdt_object *obj = NULL;
1502         struct mdt_object *local_root;
1503         static const struct lu_name lname = {
1504                 .ln_name = "i_am_nobody",
1505                 .ln_namelen = sizeof("i_am_nobody") - 1,
1506         };
1507         struct lu_ucred *uc;
1508         cfs_cap_t uc_cap_save;
1509         int rc;
1510         ENTRY;
1511
1512         rc = dt_root_get(env, mdt->mdt_bottom, local_root_fid);
1513         if (rc != 0)
1514                 RETURN(ERR_PTR(rc));
1515
1516         local_root = mdt_object_find(env, mdt, local_root_fid);
1517         if (IS_ERR(local_root))
1518                 RETURN(local_root);
1519
1520         obj = mdt_object_new(env, mdt, fid);
1521         if (IS_ERR(obj))
1522                 GOTO(out, rc = PTR_ERR(obj));
1523
1524         spec->sp_cr_lookup = 0;
1525         spec->sp_feat = &dt_directory_features;
1526         spec->sp_cr_mode = MDL_MINMODE; /* no lock */
1527         spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
1528         if (attr->ma_valid & MA_LOV) {
1529                 spec->u.sp_ea.eadata = attr->ma_lmm;
1530                 spec->u.sp_ea.eadatalen = attr->ma_lmm_size;
1531                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1532         } else {
1533                 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
1534         }
1535
1536         uc = lu_ucred(env);
1537         uc_cap_save = uc->uc_cap;
1538         uc->uc_cap |= 1 << CFS_CAP_DAC_OVERRIDE;
1539         rc = mdo_create(env, mdt_object_child(local_root), &lname,
1540                         mdt_object_child(obj), spec, attr);
1541         uc->uc_cap = uc_cap_save;
1542         if (rc < 0) {
1543                 CERROR("%s: cannot create volatile file "DFID": rc = %d\n",
1544                        mdt_obd_name(mdt), PFID(fid), rc);
1545                 GOTO(out, rc);
1546         }
1547
1548         rc = mo_open(env, mdt_object_child(obj), MDS_OPEN_CREATED);
1549         if (rc < 0)
1550                 CERROR("%s: cannot open volatile file "DFID", orphan "
1551                        "file will be left in PENDING directory until "
1552                        "next reboot, rc = %d\n", mdt_obd_name(mdt),
1553                        PFID(fid), rc);
1554         GOTO(out, rc);
1555
1556 out:
1557         if (rc < 0) {
1558                 if (!IS_ERR(obj))
1559                         mdt_object_put(env, obj);
1560                 obj = ERR_PTR(rc);
1561         }
1562         mdt_object_put(env, local_root);
1563         return obj;
1564 }
1565
1566 static int mdt_hsm_release(struct mdt_thread_info *info, struct mdt_object *o,
1567                            struct md_attr *ma)
1568 {
1569         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LAYOUT];
1570         struct close_data      *data;
1571         struct ldlm_lock       *lease;
1572         struct mdt_object      *orphan;
1573         struct md_attr         *orp_ma;
1574         struct lu_buf          *buf;
1575         bool                    lease_broken;
1576         int                     rc;
1577         int                     rc2;
1578         ENTRY;
1579
1580         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
1581                 RETURN(-EROFS);
1582
1583         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1584         if (data == NULL)
1585                 RETURN(-EPROTO);
1586
1587         lease = ldlm_handle2lock(&data->cd_handle);
1588         if (lease == NULL)
1589                 RETURN(-ESTALE);
1590
1591         /* try to hold open_sem so that nobody else can open the file */
1592         if (!down_write_trylock(&o->mot_open_sem)) {
1593                 ldlm_lock_cancel(lease);
1594                 GOTO(out_reprocess, rc = -EBUSY);
1595         }
1596
1597         /* Check if the lease open lease has already canceled */
1598         lock_res_and_lock(lease);
1599         lease_broken = ldlm_is_cancel(lease);
1600         unlock_res_and_lock(lease);
1601
1602         LDLM_DEBUG(lease, DFID " lease broken? %d\n",
1603                    PFID(mdt_object_fid(o)), lease_broken);
1604
1605         /* Cancel server side lease. Client side counterpart should
1606          * have been cancelled. It's okay to cancel it now as we've
1607          * held mot_open_sem. */
1608         ldlm_lock_cancel(lease);
1609
1610         if (lease_broken) /* don't perform release task */
1611                 GOTO(out_unlock, rc = -ESTALE);
1612
1613         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1614                 GOTO(out_unlock, rc = -EINVAL);
1615
1616         /* ma_need was set before but it seems fine to change it in order to
1617          * avoid modifying the one from RPC */
1618         ma->ma_need = MA_HSM;
1619         rc = mdt_attr_get_complex(info, o, ma);
1620         if (rc != 0)
1621                 GOTO(out_unlock, rc);
1622
1623         if (!mdt_hsm_release_allow(ma))
1624                 GOTO(out_unlock, rc = -EPERM);
1625
1626         /* already released? */
1627         if (ma->ma_hsm.mh_flags & HS_RELEASED)
1628                 GOTO(out_unlock, rc = 0);
1629
1630         /* Compare on-disk and packed data_version */
1631         if (data->cd_data_version != ma->ma_hsm.mh_arch_ver) {
1632                 CDEBUG(D_HSM, DFID" data_version mismatches: packed="LPU64
1633                        " and on-disk="LPU64"\n", PFID(mdt_object_fid(o)),
1634                        data->cd_data_version, ma->ma_hsm.mh_arch_ver);
1635                 GOTO(out_unlock, rc = -EPERM);
1636         }
1637
1638         ma->ma_valid = MA_INODE;
1639         ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_SIZE;
1640         rc = mo_attr_set(info->mti_env, mdt_object_child(o), ma);
1641         if (rc < 0)
1642                 GOTO(out_unlock, rc);
1643
1644         ma->ma_need = MA_INODE | MA_LOV;
1645         rc = mdt_attr_get_complex(info, o, ma);
1646         if (rc < 0)
1647                 GOTO(out_unlock, rc);
1648
1649         if (!(ma->ma_valid & MA_LOV)) {
1650                 /* Even empty file are released */
1651                 memset(ma->ma_lmm, 0, sizeof(*ma->ma_lmm));
1652                 ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1653                 ma->ma_lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1654                 ma->ma_lmm->lmm_stripe_size = cpu_to_le32(LOV_MIN_STRIPE_SIZE);
1655                 ma->ma_lmm_size = sizeof(*ma->ma_lmm);
1656         } else {
1657                 /* Magic must be LOV_MAGIC_Vx_DEF otherwise LOD will interpret
1658                  * ma_lmm as lov_user_md, then it will be confused by union of
1659                  * layout_gen and stripe_offset. */
1660                 if (le32_to_cpu(ma->ma_lmm->lmm_magic) == LOV_MAGIC_V1)
1661                         ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1662                 else if (le32_to_cpu(ma->ma_lmm->lmm_magic) == LOV_MAGIC_V3)
1663                         ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1664                 else
1665                         GOTO(out_unlock, rc = -EINVAL);
1666         }
1667
1668         /* Set file as released */
1669         ma->ma_lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1670
1671         orp_ma = &info->mti_u.hsm.attr;
1672         orp_ma->ma_attr.la_mode = S_IFREG | S_IWUSR;
1673         orp_ma->ma_attr.la_uid = ma->ma_attr.la_uid;
1674         orp_ma->ma_attr.la_gid = ma->ma_attr.la_gid;
1675         orp_ma->ma_attr.la_valid = LA_MODE | LA_UID | LA_GID;
1676         orp_ma->ma_lmm = ma->ma_lmm;
1677         orp_ma->ma_lmm_size = ma->ma_lmm_size;
1678         orp_ma->ma_valid = MA_INODE | MA_LOV;
1679         orphan = mdt_orphan_open(info, info->mti_mdt, &data->cd_fid, orp_ma,
1680                                  FMODE_WRITE);
1681         if (IS_ERR(orphan)) {
1682                 CERROR("%s: cannot open orphan file "DFID": rc = %ld\n",
1683                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid),
1684                        PTR_ERR(orphan));
1685                 GOTO(out_unlock, rc = PTR_ERR(orphan));
1686         }
1687
1688         /* Set up HSM attribute for orphan object */
1689         CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
1690         buf = &info->mti_buf;
1691         buf->lb_buf = info->mti_xattr_buf;
1692         buf->lb_len = sizeof(struct hsm_attrs);
1693         ma->ma_hsm.mh_flags |= HS_RELEASED;
1694         lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1695         ma->ma_hsm.mh_flags &= ~HS_RELEASED;
1696
1697         mdt_lock_reg_init(lh, LCK_EX);
1698         rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_LAYOUT |
1699                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1700         if (rc != 0)
1701                 GOTO(out_close, rc);
1702
1703         rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
1704                           XATTR_NAME_HSM, 0);
1705
1706         if (rc == 0)
1707                 /* Swap layout with orphan object */
1708                 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o),
1709                                      mdt_object_child(orphan),
1710                                      SWAP_LAYOUTS_MDS_HSM);
1711
1712         /* Release exclusive LL */
1713         mdt_object_unlock(info, o, lh, 1);
1714
1715         EXIT;
1716
1717 out_close:
1718         /* Close orphan object anyway */
1719         rc2 = mo_close(info->mti_env, mdt_object_child(orphan), orp_ma,
1720                        FMODE_WRITE);
1721         if (rc2 < 0)
1722                 CERROR("%s: error closing volatile file "DFID": rc = %d\n",
1723                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid), rc2);
1724         LU_OBJECT_DEBUG(D_HSM, info->mti_env, &orphan->mot_obj,
1725                         "object closed\n");
1726         mdt_object_put(info->mti_env, orphan);
1727
1728 out_unlock:
1729         up_write(&o->mot_open_sem);
1730
1731         if (rc == 0) { /* already released */
1732                 struct mdt_body *repbody;
1733                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1734                 LASSERT(repbody != NULL);
1735                 repbody->mbo_valid |= OBD_MD_FLRELEASED;
1736         }
1737
1738 out_reprocess:
1739         ldlm_reprocess_all(lease->l_resource);
1740         LDLM_LOCK_PUT(lease);
1741
1742         ma->ma_valid = 0;
1743         ma->ma_need = 0;
1744
1745         return rc;
1746 }
1747
1748 #define MFD_CLOSED(mode) ((mode) == MDS_FMODE_CLOSED)
1749
1750 static int mdt_mfd_closed(struct mdt_file_data *mfd)
1751 {
1752         return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_mode));
1753 }
1754
1755 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
1756 {
1757         struct mdt_object *o = mfd->mfd_object;
1758         struct md_object *next = mdt_object_child(o);
1759         struct md_attr *ma = &info->mti_attr;
1760         int rc = 0;
1761         __u64 mode;
1762         ENTRY;
1763
1764         mode = mfd->mfd_mode;
1765
1766         if (ma->ma_attr_flags & MDS_HSM_RELEASE) {
1767                 rc = mdt_hsm_release(info, o, ma);
1768                 if (rc < 0) {
1769                         CDEBUG(D_HSM, "%s: File " DFID " release failed: %d\n",
1770                                 mdt_obd_name(info->mti_mdt),
1771                                 PFID(mdt_object_fid(o)), rc);
1772                         /* continue to close even error occurred. */
1773                 }
1774         }
1775
1776         if (mode & FMODE_WRITE)
1777                 mdt_write_put(o);
1778         else if (mode & MDS_FMODE_EXEC)
1779                 mdt_write_allow(o);
1780
1781         /* Update atime on close only. */
1782         if ((mode & MDS_FMODE_EXEC || mode & FMODE_READ || mode & FMODE_WRITE)
1783             && (ma->ma_valid & MA_INODE) && (ma->ma_attr.la_valid & LA_ATIME)) {
1784                 /* Set the atime only. */
1785                 ma->ma_valid = MA_INODE;
1786                 ma->ma_attr.la_valid = LA_ATIME;
1787                 rc = mo_attr_set(info->mti_env, next, ma);
1788         }
1789
1790         /* If file data is modified, add the dirty flag. */
1791         if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
1792                 rc = mdt_add_dirty_flag(info, o, ma);
1793
1794         ma->ma_need |= MA_INODE;
1795         ma->ma_valid &= ~MA_INODE;
1796
1797         if (!MFD_CLOSED(mode))
1798                 rc = mo_close(info->mti_env, next, ma, mode);
1799
1800         /* adjust open and lease count */
1801         if (mode & MDS_OPEN_LEASE) {
1802                 LASSERT(atomic_read(&o->mot_lease_count) > 0);
1803                 atomic_dec(&o->mot_lease_count);
1804         }
1805
1806         LASSERT(atomic_read(&o->mot_open_count) > 0);
1807         atomic_dec(&o->mot_open_count);
1808         mdt_mfd_free(mfd);
1809         mdt_object_put(info->mti_env, o);
1810
1811         RETURN(rc);
1812 }
1813
1814 int mdt_close(struct tgt_session_info *tsi)
1815 {
1816         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1817         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1818         struct mdt_export_data *med;
1819         struct mdt_file_data   *mfd;
1820         struct mdt_object      *o;
1821         struct md_attr         *ma = &info->mti_attr;
1822         struct mdt_body        *repbody = NULL;
1823         int rc, ret = 0;
1824         ENTRY;
1825
1826         mdt_counter_incr(req, LPROC_MDT_CLOSE);
1827         /* Close may come with the Size-on-MDS update. Unpack it. */
1828         rc = mdt_close_unpack(info);
1829         if (rc)
1830                 GOTO(out, rc = err_serious(rc));
1831
1832         /* These fields are no longer used and are left for compatibility.
1833          * size is always zero */
1834         req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
1835                              0);
1836         req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
1837                              0);
1838         rc = req_capsule_server_pack(info->mti_pill);
1839         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
1840                 mdt_client_compatibility(info);
1841                 if (rc == 0)
1842                         mdt_fix_reply(info);
1843                 mdt_exit_ucred(info);
1844                 GOTO(out, rc = lustre_msg_get_status(req->rq_repmsg));
1845         }
1846
1847         /* Continue to close handle even if we can not pack reply */
1848         if (rc == 0) {
1849                 repbody = req_capsule_server_get(info->mti_pill,
1850                                                  &RMF_MDT_BODY);
1851                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
1852                                                     &RMF_MDT_MD);
1853                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
1854                                                        &RMF_MDT_MD,
1855                                                        RCL_SERVER);
1856                 ma->ma_need = MA_INODE | MA_LOV;
1857                 repbody->mbo_eadatasize = 0;
1858                 repbody->mbo_aclsize = 0;
1859         } else {
1860                 rc = err_serious(rc);
1861         }
1862
1863         med = &req->rq_export->exp_mdt_data;
1864         spin_lock(&med->med_open_lock);
1865         mfd = mdt_handle2mfd(med, &info->mti_close_handle, req_is_replay(req));
1866         if (mdt_mfd_closed(mfd)) {
1867                 spin_unlock(&med->med_open_lock);
1868                 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
1869                        ": cookie = "LPX64"\n", PFID(info->mti_rr.rr_fid1),
1870                        info->mti_close_handle.cookie);
1871                 /** not serious error since bug 3633 */
1872                 rc = -ESTALE;
1873         } else {
1874                 class_handle_unhash(&mfd->mfd_handle);
1875                 list_del_init(&mfd->mfd_list);
1876                 spin_unlock(&med->med_open_lock);
1877
1878                 /* Do not lose object before last unlink. */
1879                 o = mfd->mfd_object;
1880                 mdt_object_get(info->mti_env, o);
1881                 ret = mdt_mfd_close(info, mfd);
1882                 if (repbody != NULL)
1883                         rc = mdt_handle_last_unlink(info, o, ma);
1884                 mdt_empty_transno(info, rc);
1885                 mdt_object_put(info->mti_env, o);
1886         }
1887         if (repbody != NULL) {
1888                 mdt_client_compatibility(info);
1889                 rc = mdt_fix_reply(info);
1890         }
1891
1892         mdt_exit_ucred(info);
1893         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
1894                 GOTO(out, rc = err_serious(-ENOMEM));
1895
1896         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
1897                                  OBD_FAIL_MDS_CLOSE_NET_REP))
1898                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
1899 out:
1900         mdt_thread_info_fini(info);
1901         RETURN(rc ? rc : ret);
1902 }