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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_open.c
  33  *
  34  * Lustre Metadata Target (mdt) open/close file handling
  35  *
  36  * Author: Huang Hua <huanghua@clusterfs.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include <lustre_acl.h>
  42 #include <lustre_mds.h>
  43 #include <lustre_swab.h>
  44 #include "mdt_internal.h"
  45 #include <lustre_nodemap.h>
  46
  47 /* we do nothing because we do not have refcount now */
  48 static void mdt_mfd_get(void *mfdp)
  49 {
  50 }
  51
  52 static struct portals_handle_ops mfd_handle_ops = {
  53         .hop_addref = mdt_mfd_get,
  54         .hop_free   = NULL,
  55 };
  56
  57 /* Create a new mdt_file_data struct, initialize it,
  58  * and insert it to global hash table */
  59 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med)
  60 {
  61         struct mdt_file_data *mfd;
  62         ENTRY;
  63
  64         OBD_ALLOC_PTR(mfd);
  65         if (mfd != NULL) {
  66                 INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
  67                 mfd->mfd_handle.h_owner = med;
  68                 INIT_LIST_HEAD(&mfd->mfd_list);
  69                 class_handle_hash(&mfd->mfd_handle, &mfd_handle_ops);
  70         }
  71
  72         RETURN(mfd);
  73 }
  74
  75 /*
  76  * Find the mfd pointed to by handle in global hash table.
  77  * In case of replay the handle is obsoleted
  78  * but mfd can be found in mfd list by that handle.
  79  * Callers need to be holding med_open_lock.
  80  */
  81 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
  82                                      const struct lustre_handle *handle,
  83                                      bool is_replay_or_resent)
  84 {
  85         struct mdt_file_data   *mfd;
  86         ENTRY;
  87
  88         LASSERT(handle != NULL);
  89         mfd = class_handle2object(handle->cookie, med);
  90         /* during dw/setattr replay the mfd can be found by old handle */
  91         if (mfd == NULL && is_replay_or_resent) {
  92                 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
  93                         if (mfd->mfd_old_handle.cookie == handle->cookie)
  94                                 RETURN(mfd);
  95                 }
  96                 mfd = NULL;
  97         }
  98
  99         RETURN(mfd);
 100 }
 101
 102 /* free mfd */
 103 void mdt_mfd_free(struct mdt_file_data *mfd)
 104 {
 105         LASSERT(list_empty(&mfd->mfd_list));
 106         OBD_FREE_RCU(mfd, sizeof *mfd, &mfd->mfd_handle);
 107 }
 108
 109 static int mdt_create_data(struct mdt_thread_info *info,
 110                            struct mdt_object *p, struct mdt_object *o)
 111 {
 112         struct md_op_spec     *spec = &info->mti_spec;
 113         struct md_attr        *ma   = &info->mti_attr;
 114         int                    rc   = 0;
 115         ENTRY;
 116
 117         if (!md_should_create(spec->sp_cr_flags))
 118                 RETURN(0);
 119
 120         ma->ma_need = MA_INODE | MA_LOV;
 121         ma->ma_valid = 0;
 122         mutex_lock(&o->mot_lov_mutex);
 123         if (!o->mot_lov_created) {
 124                 rc = mdo_create_data(info->mti_env,
 125                                      p ? mdt_object_child(p) : NULL,
 126                                      mdt_object_child(o), spec, ma);
 127                 if (rc == 0)
 128                         rc = mdt_attr_get_complex(info, o, ma);
 129
 130                 if (rc == 0 && ma->ma_valid & MA_LOV)
 131                         o->mot_lov_created = 1;
 132         }
 133
 134         mutex_unlock(&o->mot_lov_mutex);
 135         RETURN(rc);
 136 }
 137
 138 int mdt_write_read(struct mdt_object *o)
 139 {
 140         int rc = 0;
 141         ENTRY;
 142         spin_lock(&o->mot_write_lock);
 143         rc = o->mot_write_count;
 144         spin_unlock(&o->mot_write_lock);
 145         RETURN(rc);
 146 }
 147
 148 int mdt_write_get(struct mdt_object *o)
 149 {
 150         int rc = 0;
 151         ENTRY;
 152         spin_lock(&o->mot_write_lock);
 153         if (o->mot_write_count < 0)
 154                 rc = -ETXTBSY;
 155         else
 156                 o->mot_write_count++;
 157         spin_unlock(&o->mot_write_lock);
 158
 159         RETURN(rc);
 160 }
 161
 162 void mdt_write_put(struct mdt_object *o)
 163 {
 164         ENTRY;
 165         spin_lock(&o->mot_write_lock);
 166         o->mot_write_count--;
 167         spin_unlock(&o->mot_write_lock);
 168         EXIT;
 169 }
 170
 171 static int mdt_write_deny(struct mdt_object *o)
 172 {
 173         int rc = 0;
 174         ENTRY;
 175         spin_lock(&o->mot_write_lock);
 176         if (o->mot_write_count > 0)
 177                 rc = -ETXTBSY;
 178         else
 179                 o->mot_write_count--;
 180         spin_unlock(&o->mot_write_lock);
 181         RETURN(rc);
 182 }
 183
 184 static void mdt_write_allow(struct mdt_object *o)
 185 {
 186         ENTRY;
 187         spin_lock(&o->mot_write_lock);
 188         o->mot_write_count++;
 189         spin_unlock(&o->mot_write_lock);
 190         EXIT;
 191 }
 192
 193 /* there can be no real transaction so prepare the fake one */
 194 static void mdt_empty_transno(struct mdt_thread_info *info, int rc)
 195 {
 196         struct mdt_device      *mdt = info->mti_mdt;
 197         struct ptlrpc_request  *req = mdt_info_req(info);
 198         struct tg_export_data  *ted;
 199         struct lsd_client_data *lcd;
 200         ENTRY;
 201
 202         if (mdt_rdonly(req->rq_export))
 203                 RETURN_EXIT;
 204
 205         /* transaction has occurred already */
 206         if (lustre_msg_get_transno(req->rq_repmsg) != 0)
 207                 RETURN_EXIT;
 208
 209         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 210                 struct thandle         *th;
 211
 212                 /* generate an empty transaction to get a transno
 213                  * and reply data */
 214                 th = dt_trans_create(info->mti_env, mdt->mdt_bottom);
 215                 if (!IS_ERR(th)) {
 216                         rc = dt_trans_start(info->mti_env, mdt->mdt_bottom, th);
 217                         dt_trans_stop(info->mti_env, mdt->mdt_bottom, th);
 218                 }
 219                 RETURN_EXIT;
 220         }
 221
 222         spin_lock(&mdt->mdt_lut.lut_translock);
 223         if (rc != 0) {
 224                 if (info->mti_transno != 0) {
 225                         struct obd_export *exp = req->rq_export;
 226
 227                         CERROR("%s: replay trans %llu NID %s: rc = %d\n",
 228                                mdt_obd_name(mdt), info->mti_transno,
 229                                obd_export_nid2str(exp), rc);
 230                         spin_unlock(&mdt->mdt_lut.lut_translock);
 231                         RETURN_EXIT;
 232                 }
 233         } else if (info->mti_transno == 0) {
 234                 info->mti_transno = ++mdt->mdt_lut.lut_last_transno;
 235         } else {
 236                 /* should be replay */
 237                 if (info->mti_transno > mdt->mdt_lut.lut_last_transno)
 238                         mdt->mdt_lut.lut_last_transno = info->mti_transno;
 239         }
 240         spin_unlock(&mdt->mdt_lut.lut_translock);
 241
 242         CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n",
 243                info->mti_transno,
 244                req->rq_export->exp_obd->obd_last_committed);
 245
 246         req->rq_transno = info->mti_transno;
 247         lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
 248
 249         /* update lcd in memory only for resent cases */
 250         ted = &req->rq_export->exp_target_data;
 251         LASSERT(ted);
 252         mutex_lock(&ted->ted_lcd_lock);
 253         lcd = ted->ted_lcd;
 254         if (info->mti_transno < lcd->lcd_last_transno &&
 255             info->mti_transno != 0) {
 256                 /* This should happen during replay. Do not update
 257                  * last rcvd info if replay req transno < last transno,
 258                  * otherwise the following resend(after replay) can not
 259                  * be checked correctly by xid */
 260                 mutex_unlock(&ted->ted_lcd_lock);
 261                 CDEBUG(D_HA, "%s: transno = %llu < last_transno = %llu\n",
 262                        mdt_obd_name(mdt), info->mti_transno,
 263                        lcd->lcd_last_transno);
 264                 RETURN_EXIT;
 265         }
 266
 267         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
 268                 if (info->mti_transno != 0)
 269                         lcd->lcd_last_close_transno = info->mti_transno;
 270                 lcd->lcd_last_close_xid = req->rq_xid;
 271                 lcd->lcd_last_close_result = rc;
 272         } else {
 273                 /* VBR: save versions in last_rcvd for reconstruct. */
 274                 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
 275                 if (pre_versions) {
 276                         lcd->lcd_pre_versions[0] = pre_versions[0];
 277                         lcd->lcd_pre_versions[1] = pre_versions[1];
 278                         lcd->lcd_pre_versions[2] = pre_versions[2];
 279                         lcd->lcd_pre_versions[3] = pre_versions[3];
 280                 }
 281                 if (info->mti_transno != 0)
 282                         lcd->lcd_last_transno = info->mti_transno;
 283
 284                 lcd->lcd_last_xid = req->rq_xid;
 285                 lcd->lcd_last_result = rc;
 286                 lcd->lcd_last_data = info->mti_opdata;
 287         }
 288         mutex_unlock(&ted->ted_lcd_lock);
 289
 290         EXIT;
 291 }
 292
 293 void mdt_mfd_set_mode(struct mdt_file_data *mfd, __u64 mode)
 294 {
 295         LASSERT(mfd != NULL);
 296
 297         CDEBUG(D_DENTRY, DFID " Change mfd mode %#llo -> %#llo.\n",
 298                PFID(mdt_object_fid(mfd->mfd_object)), mfd->mfd_mode, mode);
 299
 300         mfd->mfd_mode = mode;
 301 }
 302
 303 /**
 304  * prep ma_lmm/ma_lmv for md_attr from reply
 305  */
 306 static void mdt_prep_ma_buf_from_rep(struct mdt_thread_info *info,
 307                                      struct mdt_object *obj,
 308                                      struct md_attr *ma)
 309 {
 310         LASSERT(ma->ma_lmv == NULL && ma->ma_lmm == NULL);
 311         if (S_ISDIR(obj->mot_header.loh_attr)) {
 312                 ma->ma_lmv = req_capsule_server_get(info->mti_pill,
 313                                                     &RMF_MDT_MD);
 314                 ma->ma_lmv_size = req_capsule_get_size(info->mti_pill,
 315                                                        &RMF_MDT_MD,
 316                                                        RCL_SERVER);
 317                 if (ma->ma_lmv_size > 0)
 318                         ma->ma_need |= MA_LMV;
 319         } else {
 320                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
 321                                                     &RMF_MDT_MD);
 322                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 323                                                        &RMF_MDT_MD,
 324                                                        RCL_SERVER);
 325                 if (ma->ma_lmm_size > 0)
 326                         ma->ma_need |= MA_LOV;
 327         }
 328 }
 329
 330 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
 331                         struct mdt_object *o, __u64 flags, int created,
 332                         struct ldlm_reply *rep)
 333 {
 334         struct ptlrpc_request   *req = mdt_info_req(info);
 335         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 336         struct mdt_file_data    *mfd;
 337         struct md_attr          *ma  = &info->mti_attr;
 338         struct lu_attr          *la  = &ma->ma_attr;
 339         struct mdt_body         *repbody;
 340         int                      rc = 0, isdir, isreg;
 341         ENTRY;
 342
 343         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 344
 345         isreg = S_ISREG(la->la_mode);
 346         isdir = S_ISDIR(la->la_mode);
 347         if (isreg && !(ma->ma_valid & MA_LOV) && !(flags & MDS_OPEN_RELEASE)) {
 348                 /*
 349                  * No EA, check whether it is will set regEA and dirEA since in
 350                  * above attr get, these size might be zero, so reset it, to
 351                  * retrieve the MD after create obj.
 352                  */
 353                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 354                                                        &RMF_MDT_MD,
 355                                                        RCL_SERVER);
 356                 /* in replay case, p == NULL */
 357                 rc = mdt_create_data(info, p, o);
 358                 if (rc)
 359                         RETURN(rc);
 360
 361                 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_DISP_STRIPE)
 362                         mdt_set_disposition(info, rep, DISP_OPEN_STRIPE);
 363         }
 364
 365         CDEBUG(D_INODE, "after open, ma_valid bit = %#llx lmm_size = %d\n",
 366                ma->ma_valid, ma->ma_lmm_size);
 367
 368         if (ma->ma_valid & MA_LOV) {
 369                 LASSERT(ma->ma_lmm_size != 0);
 370                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 371                 if (isdir)
 372                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 373                 else
 374                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 375         }
 376
 377         if (ma->ma_valid & MA_LMV) {
 378                 LASSERT(ma->ma_lmv_size != 0);
 379                 repbody->mbo_eadatasize = ma->ma_lmv_size;
 380                 LASSERT(isdir);
 381                 repbody->mbo_valid |= OBD_MD_FLDIREA | OBD_MD_MEA;
 382         }
 383
 384         if (flags & MDS_FMODE_WRITE)
 385                 rc = mdt_write_get(o);
 386         else if (flags & MDS_FMODE_EXEC)
 387                 rc = mdt_write_deny(o);
 388
 389         if (rc)
 390                 RETURN(rc);
 391
 392         rc = mo_open(info->mti_env, mdt_object_child(o),
 393                      created ? flags | MDS_OPEN_CREATED : flags);
 394         if (rc != 0) {
 395                 /* If we allow the client to chgrp (CFS_SETGRP_PERM), but the
 396                  * client does not know which suppgid should be sent to the MDS,
 397                  * or some other(s) changed the target file's GID after this RPC
 398                  * sent to the MDS with the suppgid as the original GID, then we
 399                  * give the client another chance to send the right suppgid. */
 400                 if (rc == -EACCES &&
 401                     allow_client_chgrp(info, lu_ucred(info->mti_env)))
 402                         mdt_set_disposition(info, rep, DISP_OPEN_DENY);
 403
 404                 GOTO(err_out, rc);
 405         }
 406
 407         mfd = mdt_mfd_new(med);
 408         if (mfd == NULL)
 409                 GOTO(err_out, rc = -ENOMEM);
 410
 411         /*
 412          * Keep a reference on this object for this open, and is
 413          * released by mdt_mfd_close().
 414          */
 415         mdt_object_get(info->mti_env, o);
 416         mfd->mfd_object = o;
 417         mfd->mfd_xid = req->rq_xid;
 418
 419         /*
 420          * @flags is always not zero. At least it should be FMODE_READ,
 421          * FMODE_WRITE or MDS_FMODE_EXEC.
 422          */
 423         LASSERT(flags != 0);
 424
 425         /* Open handling. */
 426         mdt_mfd_set_mode(mfd, flags);
 427
 428         atomic_inc(&o->mot_open_count);
 429         if (flags & MDS_OPEN_LEASE)
 430                 atomic_inc(&o->mot_lease_count);
 431
 432         /* replay handle */
 433         if (req_is_replay(req)) {
 434                 struct mdt_file_data *old_mfd;
 435                 /* Check wheather old cookie already exist in
 436                  * the list, becasue when do recovery, client
 437                  * might be disconnected from server, and
 438                  * restart replay, so there maybe some orphan
 439                  * mfd here, we should remove them */
 440                 LASSERT(info->mti_rr.rr_handle != NULL);
 441                 spin_lock(&med->med_open_lock);
 442                 old_mfd = mdt_handle2mfd(med, info->mti_rr.rr_handle, true);
 443                 if (old_mfd != NULL) {
 444                         CDEBUG(D_HA, "delete orphan mfd = %p, fid = "DFID", "
 445                                "cookie = %#llx\n", mfd,
 446                                PFID(mdt_object_fid(mfd->mfd_object)),
 447                                info->mti_rr.rr_handle->cookie);
 448                         class_handle_unhash(&old_mfd->mfd_handle);
 449                         list_del_init(&old_mfd->mfd_list);
 450                         spin_unlock(&med->med_open_lock);
 451                         /* no attr update for that close */
 452                         la->la_valid = 0;
 453                         ma->ma_valid |= MA_FLAGS;
 454                         ma->ma_attr_flags |= MDS_RECOV_OPEN;
 455                         mdt_mfd_close(info, old_mfd);
 456                         ma->ma_attr_flags &= ~MDS_RECOV_OPEN;
 457                         ma->ma_valid &= ~MA_FLAGS;
 458                 } else {
 459                         spin_unlock(&med->med_open_lock);
 460                         CDEBUG(D_HA, "orphan mfd not found, fid = "DFID", "
 461                                "cookie = %#llx\n",
 462                                PFID(mdt_object_fid(mfd->mfd_object)),
 463                                info->mti_rr.rr_handle->cookie);
 464                 }
 465
 466                 CDEBUG(D_HA, "Store old cookie %#llx in new mfd\n",
 467                        info->mti_rr.rr_handle->cookie);
 468
 469                 mfd->mfd_old_handle.cookie = info->mti_rr.rr_handle->cookie;
 470         }
 471
 472         repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
 473
 474         if (req->rq_export->exp_disconnected) {
 475                 spin_lock(&med->med_open_lock);
 476                 class_handle_unhash(&mfd->mfd_handle);
 477                 list_del_init(&mfd->mfd_list);
 478                 spin_unlock(&med->med_open_lock);
 479                 mdt_mfd_close(info, mfd);
 480         } else {
 481                 spin_lock(&med->med_open_lock);
 482                 list_add_tail(&mfd->mfd_list, &med->med_open_head);
 483                 spin_unlock(&med->med_open_lock);
 484         }
 485
 486         mdt_empty_transno(info, rc);
 487
 488         RETURN(rc);
 489
 490 err_out:
 491         if (flags & MDS_FMODE_WRITE)
 492                 mdt_write_put(o);
 493         else if (flags & MDS_FMODE_EXEC)
 494                 mdt_write_allow(o);
 495
 496         return rc;
 497 }
 498
 499 static int mdt_finish_open(struct mdt_thread_info *info,
 500                            struct mdt_object *p, struct mdt_object *o,
 501                            __u64 flags, int created, struct ldlm_reply *rep)
 502 {
 503         struct ptlrpc_request   *req = mdt_info_req(info);
 504         struct obd_export       *exp = req->rq_export;
 505         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 506         struct md_attr          *ma  = &info->mti_attr;
 507         struct lu_attr          *la  = &ma->ma_attr;
 508         struct mdt_file_data    *mfd;
 509         struct mdt_body         *repbody;
 510         int                      rc = 0;
 511         int                      isreg, isdir, islnk;
 512         struct list_head        *t;
 513         ENTRY;
 514
 515         LASSERT(ma->ma_valid & MA_INODE);
 516
 517         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 518
 519         isreg = S_ISREG(la->la_mode);
 520         isdir = S_ISDIR(la->la_mode);
 521         islnk = S_ISLNK(la->la_mode);
 522         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 523
 524         /* compatibility check for 2.10 clients when it tries to open mirrored
 525          * files. 2.10 clients don't verify overlapping components so they
 526          * would read and write mirrored files just as if they were normal
 527          * PFL files, which will cause the problem that sycned mirrors actually
 528          * contain different data.
 529          * Older clients are not a concern here because they don't even
 530          * understand PFL layout. */
 531         if (isreg && !exp_connect_flr(exp) && ma->ma_valid & MA_LOV &&
 532             mdt_lmm_is_flr(ma->ma_lmm)) {
 533                 /* LU-10286: for simplicity clients who don't understand
 534                  * mirrored layout(with connect flag OBD_CONNECT2_FLR) won't
 535                  * be able to open mirrored files */
 536                 RETURN(-EOPNOTSUPP);
 537         }
 538
 539         /* LU-2275, simulate broken behaviour (esp. prevalent in
 540          * pre-2.4 servers where a very strange reply is sent on error
 541          * that looks like it was actually almost successful and a
 542          * failure at the same time.) */
 543         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_NEGATIVE_POSITIVE)) {
 544                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN |
 545                                                DISP_LOOKUP_NEG |
 546                                                DISP_LOOKUP_POS);
 547
 548                 if (flags & MDS_OPEN_LOCK)
 549                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
 550
 551                 RETURN(-ENOENT);
 552         }
 553
 554 #ifdef CONFIG_FS_POSIX_ACL
 555         if (exp_connect_flags(exp) & OBD_CONNECT_ACL) {
 556                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
 557                 if (IS_ERR(nodemap))
 558                         RETURN(PTR_ERR(nodemap));
 559
 560                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
 561                 nodemap_putref(nodemap);
 562                 if (rc)
 563                         RETURN(rc);
 564         }
 565 #endif
 566
 567         /*
 568          * If we are following a symlink, don't open; and do not return open
 569          * handle for special nodes as client required.
 570          */
 571         if (islnk || (!isreg && !isdir &&
 572             (exp_connect_flags(req->rq_export) & OBD_CONNECT_NODEVOH))) {
 573                 lustre_msg_set_transno(req->rq_repmsg, 0);
 574                 RETURN(0);
 575         }
 576
 577         /*
 578          * We need to return the existing object's fid back, so it is done here,
 579          * after preparing the reply.
 580          */
 581         if (!created && (flags & MDS_OPEN_EXCL) && (flags & MDS_OPEN_CREAT))
 582                 RETURN(-EEXIST);
 583
 584         /* This can't be done earlier, we need to return reply body */
 585         if (isdir) {
 586                 if (flags & (MDS_OPEN_CREAT | MDS_FMODE_WRITE)) {
 587                         /* We are trying to create or write an existing dir. */
 588                         RETURN(-EISDIR);
 589                 }
 590         } else if (flags & MDS_OPEN_DIRECTORY)
 591                 RETURN(-ENOTDIR);
 592
 593         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
 594                                  OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE)) {
 595                 RETURN(-EAGAIN);
 596         }
 597
 598         mfd = NULL;
 599         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 600                 spin_lock(&med->med_open_lock);
 601                 list_for_each(t, &med->med_open_head) {
 602                         mfd = list_entry(t, struct mdt_file_data, mfd_list);
 603                         if (mfd->mfd_xid == req->rq_xid)
 604                                 break;
 605                         mfd = NULL;
 606                 }
 607                 spin_unlock(&med->med_open_lock);
 608
 609                 if (mfd != NULL) {
 610                         repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
 611                         /* set repbody->ea_size for resent case */
 612                         if (ma->ma_valid & MA_LOV) {
 613                                 LASSERT(ma->ma_lmm_size != 0);
 614                                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 615                                 if (isdir)
 616                                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 617                                 else
 618                                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 619                         }
 620                         mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 621                         RETURN(0);
 622                 }
 623         }
 624
 625         rc = mdt_mfd_open(info, p, o, flags, created, rep);
 626         if (!rc)
 627                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 628
 629         RETURN(rc);
 630 }
 631
 632 void mdt_reconstruct_open(struct mdt_thread_info *info,
 633                           struct mdt_lock_handle *lhc)
 634 {
 635         const struct lu_env *env = info->mti_env;
 636         struct mdt_device       *mdt  = info->mti_mdt;
 637         struct req_capsule      *pill = info->mti_pill;
 638         struct ptlrpc_request   *req  = mdt_info_req(info);
 639         struct md_attr          *ma   = &info->mti_attr;
 640         struct mdt_reint_record *rr   = &info->mti_rr;
 641         __u64                   flags = info->mti_spec.sp_cr_flags;
 642         struct ldlm_reply       *ldlm_rep;
 643         struct mdt_object       *parent;
 644         struct mdt_object       *child;
 645         struct mdt_body         *repbody;
 646         int                      rc;
 647         __u64                    opdata;
 648         ENTRY;
 649
 650         LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
 651         ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
 652         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 653
 654         ma->ma_need = MA_INODE | MA_HSM;
 655         ma->ma_valid = 0;
 656
 657         opdata = mdt_req_from_lrd(req, info->mti_reply_data);
 658         mdt_set_disposition(info, ldlm_rep, opdata);
 659
 660         CDEBUG(D_INODE, "This is reconstruct open: disp=%#llx, result=%d\n",
 661                ldlm_rep->lock_policy_res1, req->rq_status);
 662
 663         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
 664             req->rq_status != 0)
 665                 /* We did not create successfully, return error to client. */
 666                 GOTO(out, rc = req->rq_status);
 667
 668         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
 669                 struct obd_export *exp = req->rq_export;
 670                 /*
 671                  * We failed after creation, but we do not know in which step
 672                  * we failed. So try to check the child object.
 673                  */
 674                 parent = mdt_object_find(env, mdt, rr->rr_fid1);
 675                 if (IS_ERR(parent)) {
 676                         rc = PTR_ERR(parent);
 677                         LCONSOLE_WARN("Parent "DFID" lookup error %d."
 678                                       " Evicting client %s with export %s.\n",
 679                                       PFID(rr->rr_fid1), rc,
 680                                       obd_uuid2str(&exp->exp_client_uuid),
 681                                       obd_export_nid2str(exp));
 682                         mdt_export_evict(exp);
 683                         RETURN_EXIT;
 684                 }
 685
 686                 child = mdt_object_find(env, mdt, rr->rr_fid2);
 687                 if (IS_ERR(child)) {
 688                         rc = PTR_ERR(child);
 689                         LCONSOLE_WARN("cannot lookup child "DFID": rc = %d; "
 690                                       "evicting client %s with export %s\n",
 691                                       PFID(rr->rr_fid2), rc,
 692                                       obd_uuid2str(&exp->exp_client_uuid),
 693                                       obd_export_nid2str(exp));
 694                         mdt_object_put(env, parent);
 695                         mdt_export_evict(exp);
 696                         RETURN_EXIT;
 697                 }
 698
 699                 if (unlikely(mdt_object_remote(child))) {
 700                         /* the child object was created on remote server */
 701                         if (!mdt_is_dne_client(exp)) {
 702                                 /* Return -EIO for old client */
 703                                 mdt_object_put(env, parent);
 704                                 mdt_object_put(env, child);
 705                                 GOTO(out, rc = -EIO);
 706                         }
 707                         repbody->mbo_fid1 = *rr->rr_fid2;
 708                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 709                         rc = 0;
 710                 } else {
 711                         if (mdt_object_exists(child)) {
 712                                 mdt_prep_ma_buf_from_rep(info, child, ma);
 713                                 rc = mdt_attr_get_complex(info, child, ma);
 714                                 if (rc == 0)
 715                                         rc = mdt_finish_open(info, parent,
 716                                                              child, flags,
 717                                                              1, ldlm_rep);
 718                         } else {
 719                                 /* the child does not exist, we should do
 720                                  * regular open */
 721                                 mdt_object_put(env, parent);
 722                                 mdt_object_put(env, child);
 723                                 GOTO(regular_open, 0);
 724                         }
 725                 }
 726                 mdt_object_put(env, parent);
 727                 mdt_object_put(env, child);
 728                 GOTO(out, rc);
 729         } else {
 730 regular_open:
 731                 /* We did not try to create, so we are a pure open */
 732                 rc = mdt_reint_open(info, lhc);
 733         }
 734
 735         EXIT;
 736 out:
 737         req->rq_status = rc;
 738         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
 739         LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
 740 }
 741
 742 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep)
 743 {
 744         __u64                    flags = info->mti_spec.sp_cr_flags;
 745         struct mdt_reint_record *rr = &info->mti_rr;
 746         struct md_attr          *ma = &info->mti_attr;
 747         struct mdt_object       *o;
 748         int                      rc;
 749         ENTRY;
 750
 751         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
 752         if (IS_ERR(o))
 753                 RETURN(rc = PTR_ERR(o));
 754
 755         if (unlikely(mdt_object_remote(o))) {
 756                 /* the child object was created on remote server */
 757                 struct mdt_body *repbody;
 758
 759                 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 760                                                 DISP_LOOKUP_EXECD |
 761                                                 DISP_LOOKUP_POS));
 762                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 763                 repbody->mbo_fid1 = *rr->rr_fid2;
 764                 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 765                 rc = 0;
 766         } else {
 767                 if (mdt_object_exists(o)) {
 768                         mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 769                                                         DISP_LOOKUP_EXECD |
 770                                                         DISP_LOOKUP_POS));
 771                         mdt_prep_ma_buf_from_rep(info, o, ma);
 772                         rc = mdt_attr_get_complex(info, o, ma);
 773                         if (rc == 0)
 774                                 rc = mdt_finish_open(info, NULL, o, flags, 0,
 775                                                      rep);
 776                 } else {
 777                         rc = -ENOENT;
 778                 }
 779         }
 780
 781         mdt_object_put(info->mti_env, o);
 782         RETURN(rc);
 783 }
 784
 785 /* lock object for open */
 786 static int mdt_object_open_lock(struct mdt_thread_info *info,
 787                                 struct mdt_object *obj,
 788                                 struct mdt_lock_handle *lhc,
 789                                 __u64 *ibits)
 790 {
 791         struct md_attr *ma = &info->mti_attr;
 792         __u64 open_flags = info->mti_spec.sp_cr_flags;
 793         __u64 trybits = 0;
 794         enum ldlm_mode lm = LCK_CR;
 795         bool acq_lease = !!(open_flags & MDS_OPEN_LEASE);
 796         bool try_layout = false;
 797         bool create_layout = false;
 798         int rc = 0;
 799         int dom_stripes = LMM_NO_DOM;
 800         bool dom_lock = false;
 801
 802         ENTRY;
 803
 804         *ibits = 0;
 805         mdt_lock_handle_init(lhc);
 806
 807         if (req_is_replay(mdt_info_req(info)))
 808                 RETURN(0);
 809
 810         if (S_ISREG(lu_object_attr(&obj->mot_obj))) {
 811                 if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV) &&
 812                     md_should_create(open_flags))
 813                         create_layout = true;
 814                 if (exp_connect_layout(info->mti_exp) && !create_layout &&
 815                     ma->ma_need & MA_LOV)
 816                         try_layout = true;
 817
 818                 /* DoM files can have just MDT stripe or combined MDT + OST
 819                  * stripes.
 820                  * - In the first case the open for read/write will do IO to
 821                  *   the MDT stripe and it makes sense to take IO lock in
 822                  *   advance along with OPEN even if it is blocking lock.
 823                  * - In the second case it is just size of MDT stripe and it
 824                  *   is quite unlikely that client will write into it, though
 825                  *   it may read it. So IO lock will be taken optionally if it
 826                  *   is non-blocking one.
 827                  */
 828                 if (ma->ma_valid & MA_LOV && ma->ma_lmm != NULL)
 829                         dom_stripes = mdt_lmm_dom_entry(ma->ma_lmm);
 830
 831                 if (dom_stripes == LMM_DOM_ONLY &&
 832                     info->mti_mdt->mdt_opts.mo_dom_lock != 0 &&
 833                     !mdt_dom_client_has_lock(info, mdt_object_fid(obj)))
 834                         dom_lock = true;
 835         }
 836
 837         if (acq_lease) {
 838                 /* lease open, acquire write mode of open sem */
 839                 down_write(&obj->mot_open_sem);
 840
 841                 /* Lease exists and ask for new lease */
 842                 if (atomic_read(&obj->mot_lease_count) > 0) {
 843                         /* only exclusive open is supported, so lease
 844                          * are conflicted to each other */
 845                         GOTO(out, rc = -EBUSY);
 846                 }
 847
 848                 /* Lease must be with open lock */
 849                 if (!(open_flags & MDS_OPEN_LOCK)) {
 850                         CERROR("%s: Request lease for file:"DFID ", but open lock "
 851                                "is missed, open_flags = %#llo : rc = %d\n",
 852                                mdt_obd_name(info->mti_mdt),
 853                                PFID(mdt_object_fid(obj)), open_flags, -EPROTO);
 854                         GOTO(out, rc = -EPROTO);
 855                 }
 856
 857                 /* XXX: only exclusive open is supported. */
 858                 lm = LCK_EX;
 859                 *ibits = MDS_INODELOCK_OPEN;
 860
 861                 /* never grant LCK_EX layout lock to client */
 862                 try_layout = false;
 863         } else { /* normal open */
 864                 /* normal open holds read mode of open sem */
 865                 down_read(&obj->mot_open_sem);
 866
 867                 if (open_flags & MDS_OPEN_LOCK) {
 868                         if (open_flags & MDS_FMODE_WRITE)
 869                                 lm = LCK_CW;
 870                         else if (open_flags & MDS_FMODE_EXEC)
 871                                 lm = LCK_PR;
 872                         else
 873                                 lm = LCK_CR;
 874
 875                         *ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN;
 876                 } else if (atomic_read(&obj->mot_lease_count) > 0) {
 877                         if (open_flags & MDS_FMODE_WRITE)
 878                                 lm = LCK_CW;
 879                         else
 880                                 lm = LCK_CR;
 881
 882                         /* revoke lease */
 883                         *ibits = MDS_INODELOCK_OPEN;
 884                         try_layout = false;
 885
 886                         lhc = &info->mti_lh[MDT_LH_LOCAL];
 887                 } else if (dom_lock) {
 888                         lm = (open_flags & MDS_FMODE_WRITE) ? LCK_PW : LCK_PR;
 889                         *ibits = MDS_INODELOCK_DOM;
 890                 }
 891
 892                 CDEBUG(D_INODE, "normal open:"DFID" lease count: %d, lm: %d\n",
 893                         PFID(mdt_object_fid(obj)),
 894                         atomic_read(&obj->mot_open_count), lm);
 895         }
 896
 897         mdt_lock_reg_init(lhc, lm);
 898
 899         /* one problem to return layout lock on open is that it may result
 900          * in too many layout locks cached on the client side. */
 901         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_OPEN) && try_layout) {
 902                 trybits |= MDS_INODELOCK_LAYOUT;
 903         }
 904
 905         if (*ibits | trybits)
 906                 rc = mdt_object_lock_try(info, obj, lhc, ibits, trybits, false);
 907
 908         CDEBUG(D_INODE, "%s: Requested bits lock:"DFID ", ibits = %#llx/%#llx"
 909                ", open_flags = %#llo, try_layout = %d : rc = %d\n",
 910                mdt_obd_name(info->mti_mdt), PFID(mdt_object_fid(obj)),
 911                *ibits, trybits, open_flags, try_layout, rc);
 912
 913         /* will change layout, revoke layout locks by enqueuing EX lock. */
 914         if (rc == 0 && create_layout) {
 915                 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LAYOUT];
 916
 917                 CDEBUG(D_INODE, "Will create layout, get EX layout lock:"DFID
 918                         ", open_flags = %#llo\n",
 919                         PFID(mdt_object_fid(obj)), open_flags);
 920
 921                 /* We cannot enqueue another lock for the same resource we
 922                  * already have a lock for, due to mechanics of waiting list
 923                  * iterating in ldlm, see LU-3601.
 924                  * As such we'll drop the open lock we just got above here,
 925                  * it's ok not to have this open lock as it's main purpose is to
 926                  * flush unused cached client open handles. */
 927                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
 928                         mdt_object_unlock(info, obj, lhc, 1);
 929
 930                 LASSERT(!try_layout);
 931                 mdt_lock_handle_init(ll);
 932                 mdt_lock_reg_init(ll, LCK_EX);
 933                 rc = mdt_object_lock(info, obj, ll, MDS_INODELOCK_LAYOUT);
 934
 935                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LL_BLOCK, 2);
 936         }
 937
 938         /* Check if there is any other open handles after acquiring
 939          * open lock. At this point, caching open handles have been revoked
 940          * by open lock.
 941          * XXX: Now only exclusive open is supported. Need to check the
 942          * type of open for generic lease support. */
 943         if (rc == 0 && acq_lease) {
 944                 struct ptlrpc_request *req = mdt_info_req(info);
 945                 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
 946                 struct mdt_file_data *mfd;
 947                 bool is_replay_or_resent;
 948                 int open_count = 0;
 949
 950                 /* For lease: application can open a file and then apply lease,
 951                  * @handle contains original open handle in that case.
 952                  * In recovery, open REQ will be replayed and the lease REQ may
 953                  * be resent that means the open handle is already stale, so we
 954                  * need to fix it up here by finding new handle. */
 955                 is_replay_or_resent = req_is_replay(req) ||
 956                         lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
 957
 958                 /* if the request is _not_ a replay request, rr_handle
 959                  * may be used to hold an openhandle which is issuing the
 960                  * lease request, so that this openhandle doesn't count. */
 961                 mfd = mdt_handle2mfd(med, info->mti_rr.rr_handle,
 962                                      is_replay_or_resent);
 963                 if (mfd != NULL)
 964                         ++open_count;
 965
 966                 CDEBUG(D_INODE, "acq_lease "DFID": openers: %d, want: %d\n",
 967                         PFID(mdt_object_fid(obj)),
 968                         atomic_read(&obj->mot_open_count), open_count);
 969
 970                 if (atomic_read(&obj->mot_open_count) > open_count)
 971                         GOTO(out, rc = -EBUSY);
 972         }
 973         GOTO(out, rc);
 974
 975 out:
 976         RETURN(rc);
 977 }
 978
 979 static void mdt_object_open_unlock(struct mdt_thread_info *info,
 980                                    struct mdt_object *obj,
 981                                    struct mdt_lock_handle *lhc,
 982                                    __u64 ibits, int rc)
 983 {
 984         __u64 open_flags = info->mti_spec.sp_cr_flags;
 985         struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LOCAL];
 986         ENTRY;
 987
 988         if (req_is_replay(mdt_info_req(info)))
 989                 RETURN_EXIT;
 990
 991         /* Release local lock - the lock put in MDT_LH_LOCAL will never
 992          * return to client side. */
 993         if (lustre_handle_is_used(&ll->mlh_reg_lh))
 994                 mdt_object_unlock(info, obj, ll, 1);
 995
 996         ll = &info->mti_lh[MDT_LH_LAYOUT];
 997         /* Release local layout lock, layout was created */
 998         if (lustre_handle_is_used(&ll->mlh_reg_lh)) {
 999                 LASSERT(!(ibits & MDS_INODELOCK_LAYOUT));
1000                 mdt_object_unlock(info, obj, ll, 1);
1001         }
1002
1003         if (open_flags & MDS_OPEN_LEASE)
1004                 up_write(&obj->mot_open_sem);
1005         else
1006                 up_read(&obj->mot_open_sem);
1007
1008         /* Cross-ref case, the lock should be returned to the client */
1009         if (ibits == 0 || rc == -MDT_EREMOTE_OPEN)
1010                 RETURN_EXIT;
1011
1012         if (!(open_flags & MDS_OPEN_LOCK) && !(ibits & MDS_INODELOCK_LAYOUT) &&
1013             !(ibits & MDS_INODELOCK_DOM)) {
1014                 /* for the open request, the lock will only return to client
1015                  * if open or layout lock is granted. */
1016                 rc = 1;
1017         }
1018
1019         if (rc != 0 || !lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1020                 struct ldlm_reply       *ldlm_rep;
1021
1022                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1023                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1024                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
1025                         mdt_object_unlock(info, obj, lhc, 1);
1026         }
1027         RETURN_EXIT;
1028 }
1029
1030 /**
1031  * Check release is permitted for the current HSM flags.
1032  */
1033 static bool mdt_hsm_release_allow(const struct md_attr *ma)
1034 {
1035         if (!(ma->ma_valid & MA_HSM))
1036                 return false;
1037
1038         if (ma->ma_hsm.mh_flags & (HS_DIRTY|HS_NORELEASE|HS_LOST))
1039                 return false;
1040
1041         if (!(ma->ma_hsm.mh_flags & HS_ARCHIVED))
1042                 return false;
1043
1044         return true;
1045 }
1046
1047 static int mdt_open_by_fid_lock(struct mdt_thread_info *info,
1048                                 struct ldlm_reply *rep,
1049                                 struct mdt_lock_handle *lhc)
1050 {
1051         const struct lu_env     *env   = info->mti_env;
1052         struct mdt_device       *mdt   = info->mti_mdt;
1053         __u64                    flags = info->mti_spec.sp_cr_flags;
1054         struct mdt_reint_record *rr    = &info->mti_rr;
1055         struct md_attr          *ma    = &info->mti_attr;
1056         struct mdt_object       *parent= NULL;
1057         struct mdt_object       *o;
1058         int                      rc;
1059         bool                     object_locked = false;
1060         __u64                    ibits = 0;
1061         ENTRY;
1062
1063         if (md_should_create(flags)) {
1064                 if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1065                         parent = mdt_object_find(env, mdt, rr->rr_fid1);
1066                         if (IS_ERR(parent)) {
1067                                 CDEBUG(D_INODE, "Fail to find parent "DFID
1068                                        " for anonymous created %ld, try to"
1069                                        " use server-side parent.\n",
1070                                        PFID(rr->rr_fid1), PTR_ERR(parent));
1071                                 parent = NULL;
1072                         }
1073                 }
1074                 if (parent == NULL)
1075                         ma->ma_need |= MA_PFID;
1076         }
1077
1078         o = mdt_object_find(env, mdt, rr->rr_fid2);
1079         if (IS_ERR(o))
1080                 GOTO(out_parent_put, rc = PTR_ERR(o));
1081
1082         if (mdt_object_remote(o)) {
1083                 CDEBUG(D_INFO, "%s: "DFID" is on remote MDT.\n",
1084                        mdt_obd_name(info->mti_mdt),
1085                        PFID(rr->rr_fid2));
1086                 GOTO(out, rc = -EREMOTE);
1087         } else if (!mdt_object_exists(o)) {
1088                 mdt_set_disposition(info, rep,
1089                                     DISP_IT_EXECD |
1090                                     DISP_LOOKUP_EXECD |
1091                                     DISP_LOOKUP_NEG);
1092                 GOTO(out, rc = -ENOENT);
1093         }
1094
1095         mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1096
1097         mdt_prep_ma_buf_from_rep(info, o, ma);
1098         if (flags & MDS_OPEN_RELEASE)
1099                 ma->ma_need |= MA_HSM;
1100         rc = mdt_attr_get_complex(info, o, ma);
1101         if (rc)
1102                 GOTO(out, rc);
1103
1104         /* We should not change file's existing LOV EA */
1105         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
1106             flags & MDS_OPEN_HAS_EA && ma->ma_valid & MA_LOV)
1107                 GOTO(out, rc = -EEXIST);
1108
1109         /* If a release request, check file flags are fine and ask for an
1110          * exclusive open access. */
1111         if (flags & MDS_OPEN_RELEASE && !mdt_hsm_release_allow(ma))
1112                 GOTO(out, rc = -EPERM);
1113
1114         rc = mdt_check_resent_lock(info, o, lhc);
1115         if (rc < 0) {
1116                 GOTO(out, rc);
1117         } else if (rc > 0) {
1118                 rc = mdt_object_open_lock(info, o, lhc, &ibits);
1119                 object_locked = true;
1120                 if (rc)
1121                         GOTO(out_unlock, rc);
1122         }
1123
1124         if (ma->ma_valid & MA_PFID) {
1125                 parent = mdt_object_find(env, mdt, &ma->ma_pfid);
1126                 if (IS_ERR(parent)) {
1127                         CDEBUG(D_INODE, "Fail to find parent "DFID
1128                                " for anonymous created %ld, try to"
1129                                " use system default.\n",
1130                                PFID(&ma->ma_pfid), PTR_ERR(parent));
1131                         parent = NULL;
1132                 }
1133         }
1134
1135         rc = mdt_finish_open(info, parent, o, flags, 0, rep);
1136         if (!rc) {
1137                 mdt_set_disposition(info, rep, DISP_LOOKUP_POS);
1138                 if (flags & MDS_OPEN_LOCK)
1139                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1140                 if (flags & MDS_OPEN_LEASE)
1141                         mdt_set_disposition(info, rep, DISP_OPEN_LEASE);
1142         }
1143         GOTO(out_unlock, rc);
1144
1145 out_unlock:
1146         if (object_locked)
1147                 mdt_object_open_unlock(info, o, lhc, ibits, rc);
1148 out:
1149         mdt_object_put(env, o);
1150         if (rc == 0)
1151                 mdt_pack_size2body(info, rr->rr_fid2,
1152                                    ibits & MDS_INODELOCK_DOM);
1153 out_parent_put:
1154         if (parent != NULL)
1155                 mdt_object_put(env, parent);
1156         return rc;
1157 }
1158
1159 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1160 static int mdt_cross_open(struct mdt_thread_info *info,
1161                           const struct lu_fid *parent_fid,
1162                           const struct lu_fid *fid,
1163                           struct ldlm_reply *rep, __u32 flags)
1164 {
1165         struct md_attr    *ma = &info->mti_attr;
1166         struct mdt_object *o;
1167         int                rc;
1168         ENTRY;
1169
1170         o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1171         if (IS_ERR(o))
1172                 RETURN(rc = PTR_ERR(o));
1173
1174         if (mdt_object_remote(o)) {
1175                 /* Something is wrong here, the object is on another MDS! */
1176                 CERROR("%s: "DFID" isn't on this server!: rc = %d\n",
1177                        mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1178                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1179                                 &o->mot_obj,
1180                                 "Object isn't on this server! FLD error?");
1181                 rc = -EFAULT;
1182         } else {
1183                 if (mdt_object_exists(o)) {
1184                         /* Do permission check for cross-open. */
1185                         rc = mo_permission(info->mti_env, NULL,
1186                                            mdt_object_child(o),
1187                                            NULL, flags | MDS_OPEN_CROSS);
1188                         if (rc)
1189                                 goto out;
1190
1191                         mdt_prep_ma_buf_from_rep(info, o, ma);
1192                         rc = mdt_attr_get_complex(info, o, ma);
1193                         if (rc != 0)
1194                                 GOTO(out, rc);
1195
1196                         rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
1197                 } else {
1198                         /*
1199                          * Something is wrong here. lookup was positive but
1200                          * there is no object!
1201                          */
1202                         CERROR("%s: "DFID" doesn't exist!: rc = %d\n",
1203                               mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1204                         rc = -EFAULT;
1205                 }
1206         }
1207 out:
1208         mdt_object_put(info->mti_env, o);
1209         RETURN(rc);
1210 }
1211
1212 /*
1213  * find root object and take its xattr lock if it's on remote MDT, later create
1214  * may use fs default striping (which is stored in root xattr).
1215  */
1216 static int mdt_lock_root_xattr(struct mdt_thread_info *info,
1217                                struct mdt_device *mdt)
1218 {
1219         struct mdt_object *md_root = mdt->mdt_md_root;
1220         struct lustre_handle lhroot;
1221         int rc;
1222
1223         if (md_root == NULL) {
1224                 lu_root_fid(&info->mti_tmp_fid1);
1225                 md_root = mdt_object_find(info->mti_env, mdt,
1226                                           &info->mti_tmp_fid1);
1227                 if (IS_ERR(md_root))
1228                         return PTR_ERR(md_root);
1229
1230                 spin_lock(&mdt->mdt_lock);
1231                 if (mdt->mdt_md_root != NULL) {
1232                         spin_unlock(&mdt->mdt_lock);
1233
1234                         LASSERTF(mdt->mdt_md_root == md_root,
1235                                  "Different root object ("
1236                                  DFID") instances, %p, %p\n",
1237                                  PFID(&info->mti_tmp_fid1),
1238                                  mdt->mdt_md_root, md_root);
1239                         LASSERT(atomic_read(
1240                                 &md_root->mot_obj.lo_header->loh_ref) > 1);
1241
1242                         mdt_object_put(info->mti_env, md_root);
1243                 } else {
1244                         mdt->mdt_md_root = md_root;
1245                         spin_unlock(&mdt->mdt_lock);
1246                 }
1247         }
1248
1249         if (md_root->mot_cache_attr || !mdt_object_remote(md_root))
1250                 return 0;
1251
1252         rc = mdt_remote_object_lock(info, md_root, mdt_object_fid(md_root),
1253                                     &lhroot, LCK_PR, MDS_INODELOCK_XATTR,
1254                                     true);
1255         if (rc < 0)
1256                 return rc;
1257
1258         md_root->mot_cache_attr = 1;
1259
1260         /* don't cancel this lock, so that we know the cached xattr is valid. */
1261         ldlm_lock_decref(&lhroot, LCK_PR);
1262
1263         return 0;
1264 }
1265
1266 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1267 {
1268         struct mdt_device       *mdt = info->mti_mdt;
1269         struct ptlrpc_request   *req = mdt_info_req(info);
1270         struct mdt_object       *parent;
1271         struct mdt_object       *child;
1272         struct mdt_lock_handle  *lh;
1273         struct ldlm_reply       *ldlm_rep;
1274         struct mdt_body         *repbody;
1275         struct lu_fid           *child_fid = &info->mti_tmp_fid1;
1276         struct md_attr          *ma = &info->mti_attr;
1277         __u64                    create_flags = info->mti_spec.sp_cr_flags;
1278         __u64                    ibits = 0;
1279         struct mdt_reint_record *rr = &info->mti_rr;
1280         int                      result, rc;
1281         int                      created = 0;
1282         int                      object_locked = 0;
1283         __u32                    msg_flags;
1284         ENTRY;
1285
1286         OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1287                                (obd_timeout + 1) / 4);
1288
1289         mdt_counter_incr(req, LPROC_MDT_OPEN);
1290         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1291
1292         ma->ma_need = MA_INODE;
1293         ma->ma_valid = 0;
1294
1295         LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1296         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1297
1298         if (unlikely(create_flags & MDS_OPEN_JOIN_FILE)) {
1299                 CERROR("file join is not supported anymore.\n");
1300                 GOTO(out, result = err_serious(-EOPNOTSUPP));
1301         }
1302         msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1303
1304         if ((create_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1305             info->mti_spec.u.sp_ea.eadata == NULL)
1306                 GOTO(out, result = err_serious(-EINVAL));
1307
1308         if (create_flags & MDS_FMODE_WRITE &&
1309             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1310                 GOTO(out, result = -EROFS);
1311
1312         CDEBUG(D_INODE, "I am going to open "DFID"/("DNAME"->"DFID") "
1313                "cr_flag=%#llo mode=0%06o msg_flag=0x%x\n",
1314                PFID(rr->rr_fid1), PNAME(&rr->rr_name),
1315                PFID(rr->rr_fid2), create_flags,
1316                ma->ma_attr.la_mode, msg_flags);
1317
1318         if (info->mti_cross_ref) {
1319                 /* This is cross-ref open */
1320                 mdt_set_disposition(info, ldlm_rep,
1321                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD |
1322                              DISP_LOOKUP_POS));
1323                 result = mdt_cross_open(info, rr->rr_fid2, rr->rr_fid1,
1324                                         ldlm_rep, create_flags);
1325                 GOTO(out, result);
1326         } else if (req_is_replay(req)) {
1327                 result = mdt_open_by_fid(info, ldlm_rep);
1328
1329                 if (result != -ENOENT)
1330                         GOTO(out, result);
1331
1332                 /* We didn't find the correct object, so we need to re-create it
1333                  * via a regular replay. */
1334                 if (!(create_flags & MDS_OPEN_CREAT)) {
1335                         DEBUG_REQ(D_ERROR, req,
1336                                   "OPEN & CREAT not in open replay/by_fid.");
1337                         GOTO(out, result = -EFAULT);
1338                 }
1339                 CDEBUG(D_INFO, "No object(1), continue as regular open.\n");
1340         } else if (create_flags & (MDS_OPEN_BY_FID | MDS_OPEN_LOCK)) {
1341                 /*
1342                  * MDS_OPEN_LOCK is checked for backward compatibility with 2.1
1343                  * client.
1344                  */
1345                 result = mdt_open_by_fid_lock(info, ldlm_rep, lhc);
1346                 if (result < 0)
1347                         CDEBUG(D_INFO, "no object for "DFID": %d\n",
1348                                PFID(rr->rr_fid2), result);
1349                 GOTO(out, result);
1350         }
1351
1352         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1353                 GOTO(out, result = err_serious(-ENOMEM));
1354
1355         mdt_set_disposition(info, ldlm_rep,
1356                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1357
1358         if (!lu_name_is_valid(&rr->rr_name))
1359                 GOTO(out, result = -EPROTO);
1360
1361         result = mdt_lock_root_xattr(info, mdt);
1362         if (result < 0)
1363                 GOTO(out, result);
1364
1365 again:
1366         lh = &info->mti_lh[MDT_LH_PARENT];
1367         mdt_lock_pdo_init(lh,
1368                           (create_flags & MDS_OPEN_CREAT) ? LCK_PW : LCK_PR,
1369                           &rr->rr_name);
1370
1371         parent = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
1372         if (IS_ERR(parent))
1373                 GOTO(out, result = PTR_ERR(parent));
1374
1375         result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
1376         if (result != 0) {
1377                 mdt_object_put(info->mti_env, parent);
1378                 GOTO(out, result);
1379         }
1380
1381         /* get and check version of parent */
1382         result = mdt_version_get_check(info, parent, 0);
1383         if (result)
1384                 GOTO(out_parent, result);
1385
1386         fid_zero(child_fid);
1387
1388         result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1389                             &rr->rr_name, child_fid, &info->mti_spec);
1390
1391         LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1392                  "looking for "DFID"/"DNAME", found FID = "DFID"\n",
1393                  PFID(mdt_object_fid(parent)), PNAME(&rr->rr_name),
1394                  PFID(child_fid));
1395
1396         if (result != 0 && result != -ENOENT && result != -ESTALE)
1397                 GOTO(out_parent, result);
1398
1399         if (result == -ENOENT || result == -ESTALE) {
1400                 /* If the object is dead, let's check if the object
1401                  * is being migrated to a new object */
1402                 if (result == -ESTALE) {
1403                         struct lu_buf lmv_buf;
1404
1405                         lmv_buf.lb_buf = info->mti_xattr_buf;
1406                         lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
1407                         rc = mo_xattr_get(info->mti_env,
1408                                           mdt_object_child(parent),
1409                                           &lmv_buf, XATTR_NAME_LMV);
1410                         if (rc > 0) {
1411                                 struct lmv_mds_md_v1 *lmv;
1412
1413                                 lmv = lmv_buf.lb_buf;
1414                                 if (le32_to_cpu(lmv->lmv_hash_type) &
1415                                                 LMV_HASH_FLAG_MIGRATION) {
1416                                         /* Get the new parent FID and retry */
1417                                         mdt_object_unlock_put(info, parent,
1418                                                               lh, 1);
1419                                         mdt_lock_handle_init(lh);
1420                                         fid_le_to_cpu(
1421                                                 (struct lu_fid *)rr->rr_fid1,
1422                                                 &lmv->lmv_stripe_fids[1]);
1423                                         goto again;
1424                                 }
1425                         }
1426                 }
1427
1428                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1429                 if (result == -ESTALE) {
1430                         /*
1431                          * -ESTALE means the parent is a dead(unlinked) dir, so
1432                          * it should return -ENOENT to in accordance with the
1433                          * original mds implementaion.
1434                          */
1435                         GOTO(out_parent, result = -ENOENT);
1436                 }
1437
1438                 if (!(create_flags & MDS_OPEN_CREAT))
1439                         GOTO(out_parent, result);
1440                 if (mdt_rdonly(req->rq_export))
1441                         GOTO(out_parent, result = -EROFS);
1442                 *child_fid = *info->mti_rr.rr_fid2;
1443                 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1444                          PFID(child_fid));
1445                 /* In the function below, .hs_keycmp resolves to
1446                  * lu_obj_hop_keycmp() */
1447                 /* coverity[overrun-buffer-val] */
1448                 child = mdt_object_new(info->mti_env, mdt, child_fid);
1449         } else {
1450                 /*
1451                  * Check for O_EXCL is moved to the mdt_finish_open(), we need to
1452                  * return FID back in that case.
1453                  */
1454                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1455                 child = mdt_object_find(info->mti_env, mdt, child_fid);
1456         }
1457         if (IS_ERR(child))
1458                 GOTO(out_parent, result = PTR_ERR(child));
1459
1460         /** check version of child  */
1461         rc = mdt_version_get_check(info, child, 1);
1462         if (rc)
1463                 GOTO(out_child, result = rc);
1464
1465         if (result == -ENOENT) {
1466                 /* Create under OBF and .lustre is not permitted */
1467                 if (!fid_is_md_operative(rr->rr_fid1))
1468                         GOTO(out_child, result = -EPERM);
1469
1470                 /* save versions in reply */
1471                 mdt_version_get_save(info, parent, 0);
1472                 mdt_version_get_save(info, child, 1);
1473
1474                 /* version of child will be changed */
1475                 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
1476
1477                 /* Not found and with MDS_OPEN_CREAT: let's create it. */
1478                 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1479
1480                 /* Let lower layers know what is lock mode on directory. */
1481                 info->mti_spec.sp_cr_mode =
1482                         mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
1483
1484                 /* Don't do lookup sanity check. We know name doesn't exist. */
1485                 info->mti_spec.sp_cr_lookup = 0;
1486                 info->mti_spec.sp_feat = &dt_directory_features;
1487
1488                 result = mdo_create(info->mti_env, mdt_object_child(parent),
1489                                     &rr->rr_name, mdt_object_child(child),
1490                                     &info->mti_spec, &info->mti_attr);
1491                 if (result == -ERESTART) {
1492                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1493                         GOTO(out_child, result);
1494                 } else {
1495                         mdt_prep_ma_buf_from_rep(info, child, ma);
1496                         /* XXX: we should call this once, see few lines below */
1497                         if (result == 0)
1498                                 result = mdt_attr_get_complex(info, child, ma);
1499
1500                         if (result != 0)
1501                                 GOTO(out_child, result);
1502                 }
1503                 created = 1;
1504                 mdt_counter_incr(req, LPROC_MDT_MKNOD);
1505         } else {
1506                 /*
1507                  * The object is on remote node, return its FID for remote open.
1508                  */
1509                 if (mdt_object_remote(child)) {
1510                         /*
1511                          * Check if this lock already was sent to client and
1512                          * this is resent case. For resent case do not take lock
1513                          * again, use what is already granted.
1514                          */
1515                         LASSERT(lhc != NULL);
1516
1517                         rc = mdt_check_resent_lock(info, child, lhc);
1518                         if (rc < 0) {
1519                                 GOTO(out_child, result = rc);
1520                         } else if (rc > 0) {
1521                                 mdt_lock_handle_init(lhc);
1522                                 mdt_lock_reg_init(lhc, LCK_PR);
1523
1524                                 rc = mdt_object_lock(info, child, lhc,
1525                                                      MDS_INODELOCK_LOOKUP);
1526                         }
1527                         repbody->mbo_fid1 = *mdt_object_fid(child);
1528                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1529                         if (rc != 0)
1530                                 result = rc;
1531                         else
1532                                 result = -MDT_EREMOTE_OPEN;
1533                         GOTO(out_child, result);
1534                 } else if (mdt_object_exists(child)) {
1535                         /* We have to get attr & LOV EA & HSM for this
1536                          * object. */
1537                         mdt_prep_ma_buf_from_rep(info, child, ma);
1538                         ma->ma_need |= MA_HSM;
1539                         result = mdt_attr_get_complex(info, child, ma);
1540                         if (result != 0)
1541                                 GOTO(out_child, result);
1542                 } else {
1543                         /* Object does not exist. Likely FS corruption. */
1544                         CERROR("%s: name '"DNAME"' present, but FID "
1545                                DFID" is invalid\n", mdt_obd_name(info->mti_mdt),
1546                                PNAME(&rr->rr_name), PFID(child_fid));
1547                         GOTO(out_child, result = -EIO);
1548                 }
1549         }
1550
1551         rc = mdt_check_resent_lock(info, child, lhc);
1552         if (rc < 0) {
1553                 GOTO(out_child, result = rc);
1554         } else if (rc == 0) {
1555                 /* the open lock might already be gotten in
1556                  * ldlm_handle_enqueue() */
1557                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1558                 if (create_flags & MDS_OPEN_LOCK)
1559                         mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1560         } else {
1561                 /* get openlock if this isn't replay and client requested it */
1562                 if (!req_is_replay(req)) {
1563                         rc = mdt_object_open_lock(info, child, lhc, &ibits);
1564                         object_locked = 1;
1565                         if (rc != 0)
1566                                 GOTO(out_child_unlock, result = rc);
1567                         else if (create_flags & MDS_OPEN_LOCK)
1568                                 mdt_set_disposition(info, ldlm_rep,
1569                                                     DISP_OPEN_LOCK);
1570                 }
1571         }
1572         /* Try to open it now. */
1573         rc = mdt_finish_open(info, parent, child, create_flags,
1574                              created, ldlm_rep);
1575         if (rc) {
1576                 result = rc;
1577                 /* openlock will be released if mdt_finish_open failed */
1578                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1579
1580                 if (created && create_flags & MDS_OPEN_VOLATILE) {
1581                         CERROR("%s: cannot open volatile file "DFID", orphan "
1582                                "file will be left in PENDING directory until "
1583                                "next reboot, rc = %d\n", mdt_obd_name(mdt),
1584                                PFID(mdt_object_fid(child)), rc);
1585                         GOTO(out_child_unlock, result);
1586                 }
1587
1588                 if (created) {
1589                         ma->ma_need = 0;
1590                         ma->ma_valid = 0;
1591                         rc = mdo_unlink(info->mti_env,
1592                                         mdt_object_child(parent),
1593                                         mdt_object_child(child),
1594                                         &rr->rr_name,
1595                                         &info->mti_attr, 0);
1596                         if (rc != 0)
1597                                 CERROR("%s: "DFID" cleanup of open: rc = %d\n",
1598                                        mdt_obd_name(info->mti_mdt),
1599                                        PFID(mdt_object_fid(child)), rc);
1600                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1601                 }
1602         }
1603         EXIT;
1604 out_child_unlock:
1605         if (object_locked)
1606                 mdt_object_open_unlock(info, child, lhc, ibits, result);
1607 out_child:
1608         mdt_object_put(info->mti_env, child);
1609         if (result == 0)
1610                 mdt_pack_size2body(info, child_fid, ibits & MDS_INODELOCK_DOM);
1611 out_parent:
1612         mdt_object_unlock_put(info, parent, lh, result || !created);
1613 out:
1614         if (result)
1615                 lustre_msg_set_transno(req->rq_repmsg, 0);
1616         return result;
1617 }
1618
1619 /**
1620  * Create an orphan object use local root.
1621  */
1622 static struct mdt_object *mdt_orphan_open(struct mdt_thread_info *info,
1623                                           struct mdt_device *mdt,
1624                                           const struct lu_fid *fid,
1625                                           struct md_attr *attr, fmode_t fmode)
1626 {
1627         const struct lu_env *env = info->mti_env;
1628         struct md_op_spec *spec = &info->mti_spec;
1629         struct lu_fid *local_root_fid = &info->mti_tmp_fid1;
1630         struct mdt_object *obj = NULL;
1631         struct mdt_object *local_root;
1632         static const struct lu_name lname = {
1633                 .ln_name = "i_am_nobody",
1634                 .ln_namelen = sizeof("i_am_nobody") - 1,
1635         };
1636         struct lu_ucred *uc;
1637         cfs_cap_t uc_cap_save;
1638         int rc;
1639         ENTRY;
1640
1641         rc = dt_root_get(env, mdt->mdt_bottom, local_root_fid);
1642         if (rc != 0)
1643                 RETURN(ERR_PTR(rc));
1644
1645         local_root = mdt_object_find(env, mdt, local_root_fid);
1646         if (IS_ERR(local_root))
1647                 RETURN(local_root);
1648
1649         obj = mdt_object_new(env, mdt, fid);
1650         if (IS_ERR(obj))
1651                 GOTO(out, rc = PTR_ERR(obj));
1652
1653         spec->sp_cr_lookup = 0;
1654         spec->sp_feat = &dt_directory_features;
1655         spec->sp_cr_mode = MDL_MINMODE; /* no lock */
1656         spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
1657         if (attr->ma_valid & MA_LOV) {
1658                 spec->u.sp_ea.eadata = attr->ma_lmm;
1659                 spec->u.sp_ea.eadatalen = attr->ma_lmm_size;
1660                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1661         } else {
1662                 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
1663         }
1664
1665         uc = lu_ucred(env);
1666         uc_cap_save = uc->uc_cap;
1667         uc->uc_cap |= 1 << CFS_CAP_DAC_OVERRIDE;
1668         rc = mdo_create(env, mdt_object_child(local_root), &lname,
1669                         mdt_object_child(obj), spec, attr);
1670         uc->uc_cap = uc_cap_save;
1671         if (rc < 0) {
1672                 CERROR("%s: cannot create volatile file "DFID": rc = %d\n",
1673                        mdt_obd_name(mdt), PFID(fid), rc);
1674                 GOTO(out, rc);
1675         }
1676
1677         rc = mo_open(env, mdt_object_child(obj), MDS_OPEN_CREATED);
1678         if (rc < 0)
1679                 CERROR("%s: cannot open volatile file "DFID", orphan "
1680                        "file will be left in PENDING directory until "
1681                        "next reboot, rc = %d\n", mdt_obd_name(mdt),
1682                        PFID(fid), rc);
1683         GOTO(out, rc);
1684
1685 out:
1686         if (rc < 0) {
1687                 if (!IS_ERR(obj))
1688                         mdt_object_put(env, obj);
1689                 obj = ERR_PTR(rc);
1690         }
1691         mdt_object_put(env, local_root);
1692         return obj;
1693 }
1694
1695 /* XXX Look into layout in MDT layer. */
1696 static inline int mdt_hsm_set_released(struct lov_mds_md *lmm)
1697 {
1698         struct lov_comp_md_v1   *comp_v1;
1699         struct lov_mds_md       *v1;
1700         __u32   off;
1701         int     i;
1702
1703         if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1_DEFINED)) {
1704                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1705
1706                 if (comp_v1->lcm_entry_count == 0)
1707                         return -EINVAL;
1708
1709                 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1710                         off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1711                         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
1712                         v1->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1713                 }
1714         } else {
1715                 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1716         }
1717         return 0;
1718 }
1719
1720 static int mdt_hsm_release(struct mdt_thread_info *info, struct mdt_object *o,
1721                            struct md_attr *ma)
1722 {
1723         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LAYOUT];
1724         struct lu_ucred        *uc = mdt_ucred(info);
1725         struct close_data      *data;
1726         struct ldlm_lock       *lease;
1727         struct mdt_object      *orphan;
1728         struct md_attr         *orp_ma;
1729         struct lu_buf          *buf;
1730         cfs_cap_t               cap;
1731         bool                    lease_broken;
1732         int                     rc;
1733         int                     rc2;
1734         ENTRY;
1735
1736         if (mdt_rdonly(info->mti_exp))
1737                 RETURN(-EROFS);
1738
1739         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1740         if (data == NULL)
1741                 RETURN(-EPROTO);
1742
1743         lease = ldlm_handle2lock(&data->cd_handle);
1744         if (lease == NULL)
1745                 RETURN(-ESTALE);
1746
1747         /* try to hold open_sem so that nobody else can open the file */
1748         if (!down_write_trylock(&o->mot_open_sem)) {
1749                 ldlm_lock_cancel(lease);
1750                 GOTO(out_reprocess, rc = -EBUSY);
1751         }
1752
1753         /* Check if the lease open lease has already canceled */
1754         lock_res_and_lock(lease);
1755         lease_broken = ldlm_is_cancel(lease);
1756         unlock_res_and_lock(lease);
1757
1758         LDLM_DEBUG(lease, DFID " lease broken? %d",
1759                    PFID(mdt_object_fid(o)), lease_broken);
1760
1761         /* Cancel server side lease. Client side counterpart should
1762          * have been cancelled. It's okay to cancel it now as we've
1763          * held mot_open_sem. */
1764         ldlm_lock_cancel(lease);
1765
1766         if (lease_broken) /* don't perform release task */
1767                 GOTO(out_unlock, rc = -ESTALE);
1768
1769         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1770                 GOTO(out_unlock, rc = -EINVAL);
1771
1772         /* ma_need was set before but it seems fine to change it in order to
1773          * avoid modifying the one from RPC */
1774         ma->ma_need = MA_HSM;
1775         rc = mdt_attr_get_complex(info, o, ma);
1776         if (rc != 0)
1777                 GOTO(out_unlock, rc);
1778
1779         if (!mdt_hsm_release_allow(ma))
1780                 GOTO(out_unlock, rc = -EPERM);
1781
1782         /* already released? */
1783         if (ma->ma_hsm.mh_flags & HS_RELEASED)
1784                 GOTO(out_unlock, rc = 0);
1785
1786         /* Compare on-disk and packed data_version */
1787         if (data->cd_data_version != ma->ma_hsm.mh_arch_ver) {
1788                 CDEBUG(D_HSM, DFID" data_version mismatches: packed=%llu"
1789                        " and on-disk=%llu\n", PFID(mdt_object_fid(o)),
1790                        data->cd_data_version, ma->ma_hsm.mh_arch_ver);
1791                 GOTO(out_unlock, rc = -EPERM);
1792         }
1793
1794         ma->ma_valid = MA_INODE;
1795         ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_SIZE;
1796         rc = mo_attr_set(info->mti_env, mdt_object_child(o), ma);
1797         if (rc < 0)
1798                 GOTO(out_unlock, rc);
1799
1800         mutex_lock(&o->mot_som_mutex);
1801         rc2 = mdt_set_som(info, o, SOM_FL_STRICT, ma->ma_attr.la_size,
1802                            ma->ma_attr.la_blocks);
1803         mutex_unlock(&o->mot_som_mutex);
1804         if (rc2 < 0)
1805                 CDEBUG(D_INODE,
1806                        "%s: File "DFID" SOM update failed: rc = %d\n",
1807                        mdt_obd_name(info->mti_mdt),
1808                        PFID(mdt_object_fid(o)), rc2);
1809
1810
1811         ma->ma_need = MA_INODE | MA_LOV;
1812         rc = mdt_attr_get_complex(info, o, ma);
1813         if (rc < 0)
1814                 GOTO(out_unlock, rc);
1815
1816         if (!(ma->ma_valid & MA_LOV)) {
1817                 /* Even empty file are released */
1818                 memset(ma->ma_lmm, 0, sizeof(*ma->ma_lmm));
1819                 ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEFINED);
1820                 ma->ma_lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1821                 ma->ma_lmm->lmm_stripe_size = cpu_to_le32(LOV_MIN_STRIPE_SIZE);
1822                 ma->ma_lmm_size = sizeof(*ma->ma_lmm);
1823         } else {
1824                 /* Magic must be LOV_MAGIC_*_DEFINED or LOD will interpret
1825                  * ma_lmm as lov_user_md, then it will be confused by union of
1826                  * layout_gen and stripe_offset. */
1827                 if ((le32_to_cpu(ma->ma_lmm->lmm_magic) & LOV_MAGIC_MASK) ==
1828                     LOV_MAGIC_MAGIC)
1829                         ma->ma_lmm->lmm_magic |= cpu_to_le32(LOV_MAGIC_DEFINED);
1830                 else
1831                         GOTO(out_unlock, rc = -EINVAL);
1832         }
1833
1834         /* Set file as released. */
1835         rc = mdt_hsm_set_released(ma->ma_lmm);
1836         if (rc)
1837                 GOTO(out_unlock, rc);
1838
1839         orp_ma = &info->mti_u.hsm.attr;
1840         orp_ma->ma_attr.la_mode = S_IFREG | S_IWUSR;
1841         /* We use root ownership to bypass potential quota
1842          * restrictions on the user and group of the file. */
1843         orp_ma->ma_attr.la_uid = 0;
1844         orp_ma->ma_attr.la_gid = 0;
1845         orp_ma->ma_attr.la_valid = LA_MODE | LA_UID | LA_GID;
1846         orp_ma->ma_lmm = ma->ma_lmm;
1847         orp_ma->ma_lmm_size = ma->ma_lmm_size;
1848         orp_ma->ma_valid = MA_INODE | MA_LOV;
1849         orphan = mdt_orphan_open(info, info->mti_mdt, &data->cd_fid, orp_ma,
1850                                  MDS_FMODE_WRITE);
1851         if (IS_ERR(orphan)) {
1852                 CERROR("%s: cannot open orphan file "DFID": rc = %ld\n",
1853                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid),
1854                        PTR_ERR(orphan));
1855                 GOTO(out_unlock, rc = PTR_ERR(orphan));
1856         }
1857
1858         /* Set up HSM attribute for orphan object */
1859         CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
1860         buf = &info->mti_buf;
1861         buf->lb_buf = info->mti_xattr_buf;
1862         buf->lb_len = sizeof(struct hsm_attrs);
1863         ma->ma_hsm.mh_flags |= HS_RELEASED;
1864         lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1865         ma->ma_hsm.mh_flags &= ~HS_RELEASED;
1866
1867         mdt_lock_reg_init(lh, LCK_EX);
1868         rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_LAYOUT |
1869                              MDS_INODELOCK_XATTR);
1870         if (rc != 0)
1871                 GOTO(out_close, rc);
1872
1873         /* The orphan has root ownership so we need to raise
1874          * CAP_FOWNER to set the HSM attributes. */
1875         cap = uc->uc_cap;
1876         uc->uc_cap |= MD_CAP_TO_MASK(CFS_CAP_FOWNER);
1877         rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
1878                           XATTR_NAME_HSM, 0);
1879         uc->uc_cap = cap;
1880         if (rc != 0)
1881                 GOTO(out_layout_lock, rc);
1882
1883         /* Swap layout with orphan objects. */
1884         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o),
1885                              mdt_object_child(orphan),
1886                              SWAP_LAYOUTS_MDS_HSM);
1887         EXIT;
1888
1889 out_layout_lock:
1890         /* Release exclusive LL */
1891         mdt_object_unlock(info, o, lh, 1);
1892 out_close:
1893         /* Close orphan object anyway */
1894         rc2 = mo_close(info->mti_env, mdt_object_child(orphan), orp_ma,
1895                        MDS_FMODE_WRITE);
1896         if (rc2 < 0)
1897                 CERROR("%s: error closing volatile file "DFID": rc = %d\n",
1898                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid), rc2);
1899         LU_OBJECT_DEBUG(D_HSM, info->mti_env, &orphan->mot_obj,
1900                         "object closed");
1901         mdt_object_put(info->mti_env, orphan);
1902
1903 out_unlock:
1904         up_write(&o->mot_open_sem);
1905
1906         /* already released */
1907         if (rc == 0) {
1908                 struct mdt_body *repbody;
1909
1910                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1911                 LASSERT(repbody != NULL);
1912                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
1913         }
1914
1915 out_reprocess:
1916         ldlm_reprocess_all(lease->l_resource);
1917         LDLM_LOCK_PUT(lease);
1918
1919         ma->ma_valid = 0;
1920         ma->ma_need = 0;
1921
1922         return rc;
1923 }
1924
1925 int mdt_close_handle_layouts(struct mdt_thread_info *info,
1926                              struct mdt_object *o, struct md_attr *ma)
1927 {
1928         struct mdt_lock_handle  *lh1 = &info->mti_lh[MDT_LH_NEW];
1929         struct mdt_lock_handle  *lh2 = &info->mti_lh[MDT_LH_OLD];
1930         struct close_data       *data;
1931         struct ldlm_lock        *lease;
1932         struct mdt_object       *o1 = o, *o2;
1933         bool                     lease_broken;
1934         bool                     swap_objects;
1935         int                      rc;
1936         ENTRY;
1937
1938         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
1939                 RETURN(-EROFS);
1940
1941         if (!S_ISREG(lu_object_attr(&o1->mot_obj)))
1942                 RETURN(-EINVAL);
1943
1944         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1945         if (data == NULL)
1946                 RETURN(-EPROTO);
1947
1948         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1949                 RETURN(-EINVAL);
1950
1951         rc = lu_fid_cmp(&data->cd_fid, mdt_object_fid(o));
1952         if (unlikely(rc == 0))
1953                 RETURN(-EINVAL);
1954
1955         /* Exchange o1 and o2, to enforce locking order */
1956         swap_objects = (rc < 0);
1957
1958         lease = ldlm_handle2lock(&data->cd_handle);
1959         if (lease == NULL)
1960                 RETURN(-ESTALE);
1961
1962         o2 = mdt_object_find(info->mti_env, info->mti_mdt, &data->cd_fid);
1963         if (IS_ERR(o2))
1964                 GOTO(out_lease, rc = PTR_ERR(o2));
1965
1966         if (!S_ISREG(lu_object_attr(&o2->mot_obj))) {
1967                 swap_objects = false; /* not swapped yet */
1968                 GOTO(out_obj, rc = -EINVAL);
1969         }
1970
1971         if (swap_objects)
1972                 swap(o1, o2);
1973
1974         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1975                            MAY_WRITE);
1976         if (rc < 0)
1977                 GOTO(out_obj, rc);
1978
1979         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1980                            MAY_WRITE);
1981         if (rc < 0)
1982                 GOTO(out_obj, rc);
1983
1984         /* try to hold open_sem so that nobody else can open the file */
1985         if (!down_write_trylock(&o->mot_open_sem)) {
1986                 ldlm_lock_cancel(lease);
1987                 GOTO(out_obj, rc = -EBUSY);
1988         }
1989
1990         /* Check if the lease open lease has already canceled */
1991         lock_res_and_lock(lease);
1992         lease_broken = ldlm_is_cancel(lease);
1993         unlock_res_and_lock(lease);
1994
1995         LDLM_DEBUG(lease, DFID " lease broken? %d",
1996                    PFID(mdt_object_fid(o)), lease_broken);
1997
1998         /* Cancel server side lease. Client side counterpart should
1999          * have been cancelled. It's okay to cancel it now as we've
2000          * held mot_open_sem. */
2001         ldlm_lock_cancel(lease);
2002
2003         if (lease_broken)
2004                 GOTO(out_unlock_sem, rc = -ESTALE);
2005
2006         mdt_lock_reg_init(lh1, LCK_EX);
2007         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
2008                              MDS_INODELOCK_XATTR);
2009         if (rc < 0)
2010                 GOTO(out_unlock_sem, rc);
2011
2012         mdt_lock_reg_init(lh2, LCK_EX);
2013         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
2014                              MDS_INODELOCK_XATTR);
2015         if (rc < 0)
2016                 GOTO(out_unlock1, rc);
2017
2018         /* Swap layout with orphan object */
2019         if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SWAP) {
2020                 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
2021                                      mdt_object_child(o2), 0);
2022         } else if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_MERGE ||
2023                    ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
2024                 struct lu_buf *buf = &info->mti_buf;
2025                 struct md_rejig_data mrd;
2026
2027                 mrd.mrd_obj = mdt_object_child(o == o1 ? o2 : o1);
2028                 if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT)
2029                         mrd.mrd_mirror_id = data->cd_mirror_id;
2030
2031                 buf->lb_len = sizeof(mrd);
2032                 buf->lb_buf = &mrd;
2033                 rc = mo_xattr_set(info->mti_env, mdt_object_child(o), buf,
2034                                   XATTR_LUSTRE_LOV,
2035                                   ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT ?
2036                                   LU_XATTR_SPLIT : LU_XATTR_MERGE);
2037                 if (rc == 0 && ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2038                         int rc2;
2039
2040                         mutex_lock(&o->mot_som_mutex);
2041                         rc2 = mdt_set_som(info, o, SOM_FL_STRICT,
2042                                           ma->ma_attr.la_size,
2043                                           ma->ma_attr.la_blocks);
2044                         mutex_unlock(&o->mot_som_mutex);
2045                         if (rc2 < 0)
2046                                 CERROR(DFID": Setting i_blocks error: %d, "
2047                                        "i_blocks will be reported wrongly and "
2048                                        "can only be fixed in next resync\n",
2049                                        PFID(mdt_object_fid(o)), rc2);
2050                 }
2051         }
2052         if (rc < 0)
2053                 GOTO(out_unlock2, rc);
2054
2055         EXIT;
2056
2057 out_unlock2:
2058         /* Release exclusive LL */
2059         mdt_object_unlock(info, o2, lh2, 1);
2060
2061 out_unlock1:
2062         mdt_object_unlock(info, o1, lh1, 1);
2063
2064 out_unlock_sem:
2065         up_write(&o->mot_open_sem);
2066
2067         /* already swapped */
2068         if (rc == 0) {
2069                 struct mdt_body *repbody;
2070
2071                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2072                 LASSERT(repbody != NULL);
2073                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2074         }
2075
2076 out_obj:
2077         mdt_object_put(info->mti_env, swap_objects ? o1 : o2);
2078
2079         ldlm_reprocess_all(lease->l_resource);
2080
2081 out_lease:
2082         LDLM_LOCK_PUT(lease);
2083
2084         if (ma != NULL) {
2085                 ma->ma_valid = 0;
2086                 ma->ma_need = 0;
2087         }
2088
2089         return rc;
2090 }
2091
2092 static int mdt_close_resync_done(struct mdt_thread_info *info,
2093                                  struct mdt_object *o, struct md_attr *ma)
2094 {
2095         struct close_data       *data;
2096         struct ldlm_lock        *lease;
2097         struct md_layout_change  layout = { 0 };
2098         __u32                   *resync_ids = NULL;
2099         size_t                   resync_count = 0;
2100         bool                     lease_broken;
2101         int                      rc;
2102         ENTRY;
2103
2104         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2105                 RETURN(-EROFS);
2106
2107         if (!S_ISREG(lu_object_attr(&o->mot_obj)))
2108                 RETURN(-EINVAL);
2109
2110         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2111         if (data == NULL)
2112                 RETURN(-EPROTO);
2113
2114         if (ptlrpc_req_need_swab(mdt_info_req(info)))
2115                 lustre_swab_close_data_resync_done(&data->cd_resync);
2116
2117         if (!fid_is_zero(&data->cd_fid))
2118                 RETURN(-EPROTO);
2119
2120         lease = ldlm_handle2lock(&data->cd_handle);
2121         if (lease == NULL)
2122                 RETURN(-ESTALE);
2123
2124         /* try to hold open_sem so that nobody else can open the file */
2125         if (!down_write_trylock(&o->mot_open_sem)) {
2126                 ldlm_lock_cancel(lease);
2127                 GOTO(out_reprocess, rc = -EBUSY);
2128         }
2129
2130         /* Check if the lease open lease has already canceled */
2131         lock_res_and_lock(lease);
2132         lease_broken = ldlm_is_cancel(lease);
2133         unlock_res_and_lock(lease);
2134
2135         LDLM_DEBUG(lease, DFID " lease broken? %d\n",
2136                    PFID(mdt_object_fid(o)), lease_broken);
2137
2138         /* Cancel server side lease. Client side counterpart should
2139          * have been cancelled. It's okay to cancel it now as we've
2140          * held mot_open_sem. */
2141         ldlm_lock_cancel(lease);
2142
2143         if (lease_broken) /* don't perform release task */
2144                 GOTO(out_unlock, rc = -ESTALE);
2145
2146         resync_count = data->cd_resync.resync_count;
2147
2148         if (resync_count > INLINE_RESYNC_ARRAY_SIZE) {
2149                 void *data;
2150
2151                 if (!req_capsule_has_field(info->mti_pill, &RMF_U32,
2152                                            RCL_CLIENT))
2153                         GOTO(out_unlock, rc = -EPROTO);
2154
2155                 OBD_ALLOC(resync_ids, resync_count * sizeof(__u32));
2156                 if (!resync_ids)
2157                         GOTO(out_unlock, rc = -ENOMEM);
2158
2159                 data = req_capsule_client_get(info->mti_pill, &RMF_U32);
2160                 memcpy(resync_ids, data, resync_count * sizeof(__u32));
2161
2162                 layout.mlc_resync_ids = resync_ids;
2163         } else {
2164                 layout.mlc_resync_ids = data->cd_resync.resync_ids_inline;
2165         }
2166
2167         layout.mlc_opc = MD_LAYOUT_RESYNC_DONE;
2168         layout.mlc_resync_count = resync_count;
2169         if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2170                 layout.mlc_som.lsa_valid = SOM_FL_STRICT;
2171                 layout.mlc_som.lsa_size = ma->ma_attr.la_size;
2172                 layout.mlc_som.lsa_blocks = ma->ma_attr.la_blocks;
2173         }
2174         rc = mdt_layout_change(info, o, &layout);
2175         if (rc)
2176                 GOTO(out_unlock, rc);
2177
2178         EXIT;
2179
2180 out_unlock:
2181         up_write(&o->mot_open_sem);
2182
2183         /* already released */
2184         if (rc == 0) {
2185                 struct mdt_body *repbody;
2186
2187                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2188                 LASSERT(repbody != NULL);
2189                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2190         }
2191
2192         if (resync_ids)
2193                 OBD_FREE(resync_ids, resync_count * sizeof(__u32));
2194
2195 out_reprocess:
2196         ldlm_reprocess_all(lease->l_resource);
2197         LDLM_LOCK_PUT(lease);
2198
2199         ma->ma_valid = 0;
2200         ma->ma_need = 0;
2201
2202         return rc;
2203 }
2204
2205 #define MFD_CLOSED(mode) ((mode) == MDS_FMODE_CLOSED)
2206 static int mdt_mfd_closed(struct mdt_file_data *mfd)
2207 {
2208         return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_mode));
2209 }
2210
2211 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
2212 {
2213         struct mdt_object *o = mfd->mfd_object;
2214         struct md_object *next = mdt_object_child(o);
2215         struct md_attr *ma = &info->mti_attr;
2216         int rc = 0;
2217         __u64 mode;
2218         __u64 intent;
2219
2220         ENTRY;
2221
2222         mode = mfd->mfd_mode;
2223         intent = ma->ma_attr_flags & MDS_CLOSE_INTENT;
2224
2225         CDEBUG(D_INODE, "%s: close file "DFID" with intent: %llx\n",
2226                mdt_obd_name(info->mti_mdt), PFID(mdt_object_fid(o)), intent);
2227
2228         switch (intent) {
2229         case MDS_HSM_RELEASE: {
2230                 rc = mdt_hsm_release(info, o, ma);
2231                 if (rc < 0) {
2232                         CDEBUG(D_HSM, "%s: File " DFID " release failed: %d\n",
2233                                mdt_obd_name(info->mti_mdt),
2234                                PFID(mdt_object_fid(o)), rc);
2235                         /* continue to close even error occurred. */
2236                 }
2237                 break;
2238         }
2239         case MDS_CLOSE_LAYOUT_MERGE:
2240         case MDS_CLOSE_LAYOUT_SPLIT:
2241         case MDS_CLOSE_LAYOUT_SWAP: {
2242                 rc = mdt_close_handle_layouts(info, o, ma);
2243                 if (rc < 0) {
2244                         CDEBUG(D_INODE,
2245                                "%s: cannot swap layout of "DFID": rc = %d\n",
2246                                mdt_obd_name(info->mti_mdt),
2247                                PFID(mdt_object_fid(o)), rc);
2248                         /* continue to close even if error occurred. */
2249                 }
2250                 break;
2251         }
2252         case MDS_CLOSE_RESYNC_DONE:
2253                 rc = mdt_close_resync_done(info, o, ma);
2254                 break;
2255         default:
2256                 /* nothing */
2257                 break;
2258         }
2259
2260         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
2261             ma->ma_attr.la_valid & (LA_LSIZE | LA_LBLOCKS)) {
2262                 int rc2;
2263
2264                 rc2 = mdt_lsom_update(info, o, false);
2265                 if (rc2 < 0)
2266                         CDEBUG(D_INODE,
2267                                "%s: File " DFID " LSOM failed: rc = %d\n",
2268                                mdt_obd_name(info->mti_mdt),
2269                                PFID(mdt_object_fid(o)), rc2);
2270                         /* continue to close even if error occured. */
2271         }
2272
2273         if (mode & MDS_FMODE_WRITE)
2274                 mdt_write_put(o);
2275         else if (mode & MDS_FMODE_EXEC)
2276                 mdt_write_allow(o);
2277
2278         /* Update atime on close only. */
2279         if ((mode & MDS_FMODE_EXEC || mode & MDS_FMODE_READ ||
2280              mode & MDS_FMODE_WRITE) && (ma->ma_valid & MA_INODE) &&
2281             (ma->ma_attr.la_valid & LA_ATIME)) {
2282                 /* Set the atime only. */
2283                 ma->ma_valid = MA_INODE;
2284                 ma->ma_attr.la_valid = LA_ATIME;
2285                 rc = mo_attr_set(info->mti_env, next, ma);
2286         }
2287
2288         /* If file data is modified, add the dirty flag. */
2289         if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
2290                 rc = mdt_add_dirty_flag(info, o, ma);
2291
2292         ma->ma_need |= MA_INODE;
2293         ma->ma_valid &= ~MA_INODE;
2294
2295         LASSERT(atomic_read(&o->mot_open_count) > 0);
2296         atomic_dec(&o->mot_open_count);
2297         mdt_handle_last_unlink(info, o, ma);
2298
2299         if (!MFD_CLOSED(mode)) {
2300                 rc = mo_close(info->mti_env, next, ma, mode);
2301                 mdt_dom_check_and_discard(info, o);
2302         }
2303
2304         /* adjust open and lease count */
2305         if (mode & MDS_OPEN_LEASE) {
2306                 LASSERT(atomic_read(&o->mot_lease_count) > 0);
2307                 atomic_dec(&o->mot_lease_count);
2308         }
2309
2310         mdt_mfd_free(mfd);
2311         mdt_object_put(info->mti_env, o);
2312
2313         RETURN(rc);
2314 }
2315
2316 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
2317                        struct mdt_body *repbody)
2318 {
2319         struct mdt_export_data *med;
2320         struct mdt_file_data   *mfd;
2321         int                     ret = 0;
2322         int                     rc = 0;
2323         ENTRY;
2324
2325         med = &req->rq_export->exp_mdt_data;
2326         spin_lock(&med->med_open_lock);
2327         mfd = mdt_handle2mfd(med, &info->mti_close_handle, req_is_replay(req));
2328         if (mdt_mfd_closed(mfd)) {
2329                 spin_unlock(&med->med_open_lock);
2330                 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
2331                        ": cookie = %#llx\n", PFID(info->mti_rr.rr_fid1),
2332                        info->mti_close_handle.cookie);
2333                 /** not serious error since bug 3633 */
2334                 rc = -ESTALE;
2335         } else {
2336                 class_handle_unhash(&mfd->mfd_handle);
2337                 list_del_init(&mfd->mfd_list);
2338                 spin_unlock(&med->med_open_lock);
2339                 ret = mdt_mfd_close(info, mfd);
2340         }
2341
2342         RETURN(rc ? rc : ret);
2343 }
2344
2345 int mdt_close(struct tgt_session_info *tsi)
2346 {
2347         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2348         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2349         struct md_attr         *ma = &info->mti_attr;
2350         struct mdt_body        *repbody = NULL;
2351         int rc, ret = 0;
2352         ENTRY;
2353
2354         mdt_counter_incr(req, LPROC_MDT_CLOSE);
2355         /* Close may come with the Size-on-MDS update. Unpack it. */
2356         rc = mdt_close_unpack(info);
2357         if (rc)
2358                 GOTO(out, rc = err_serious(rc));
2359
2360         /* These fields are no longer used and are left for compatibility.
2361          * size is always zero */
2362         req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
2363                              0);
2364         req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
2365                              0);
2366         rc = req_capsule_server_pack(info->mti_pill);
2367         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
2368                 mdt_client_compatibility(info);
2369                 if (rc == 0)
2370                         mdt_fix_reply(info);
2371                 mdt_exit_ucred(info);
2372                 GOTO(out, rc = lustre_msg_get_status(req->rq_repmsg));
2373         }
2374
2375         /* Continue to close handle even if we can not pack reply */
2376         if (rc == 0) {
2377                 repbody = req_capsule_server_get(info->mti_pill,
2378                                                  &RMF_MDT_BODY);
2379                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
2380                                                     &RMF_MDT_MD);
2381                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
2382                                                        &RMF_MDT_MD,
2383                                                        RCL_SERVER);
2384                 ma->ma_need = MA_INODE | MA_LOV;
2385                 repbody->mbo_eadatasize = 0;
2386                 repbody->mbo_aclsize = 0;
2387         } else {
2388                 rc = err_serious(rc);
2389         }
2390
2391         rc = mdt_close_internal(info, req, repbody);
2392         if (rc != -ESTALE)
2393                 mdt_empty_transno(info, rc);
2394
2395         if (repbody != NULL) {
2396                 mdt_client_compatibility(info);
2397                 rc = mdt_fix_reply(info);
2398         }
2399
2400         mdt_exit_ucred(info);
2401         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
2402                 GOTO(out, rc = err_serious(-ENOMEM));
2403
2404         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
2405                                  OBD_FAIL_MDS_CLOSE_NET_REP))
2406                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
2407 out:
2408         mdt_thread_info_fini(info);
2409         RETURN(rc ? rc : ret);
2410 }