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 static const char mfd_open_handle_owner[] = "mdt";
  48
  49 /* Create a new mdt_file_data struct, initialize it,
  50  * and insert it to global hash table */
  51 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med)
  52 {
  53         struct mdt_file_data *mfd;
  54         ENTRY;
  55
  56         OBD_ALLOC_PTR(mfd);
  57         if (mfd != NULL) {
  58                 refcount_set(&mfd->mfd_open_handle.h_ref, 1);
  59                 INIT_HLIST_NODE(&mfd->mfd_open_handle.h_link);
  60                 mfd->mfd_owner = med;
  61                 INIT_LIST_HEAD(&mfd->mfd_list);
  62                 class_handle_hash(&mfd->mfd_open_handle, mfd_open_handle_owner);
  63         }
  64
  65         RETURN(mfd);
  66 }
  67
  68 /*
  69  * Find the mfd pointed to by handle in global hash table.
  70  * In case of replay the handle is obsoleted
  71  * but mfd can be found in mfd list by that handle.
  72  * Callers need to be holding med_open_lock.
  73  */
  74 struct mdt_file_data *mdt_open_handle2mfd(struct mdt_export_data *med,
  75                                         const struct lustre_handle *open_handle,
  76                                         bool is_replay_or_resent)
  77 {
  78         struct mdt_file_data   *mfd;
  79         ENTRY;
  80
  81         LASSERT(open_handle != NULL);
  82         mfd = class_handle2object(open_handle->cookie, mfd_open_handle_owner);
  83         if (mfd)
  84                 refcount_dec(&mfd->mfd_open_handle.h_ref);
  85
  86         /* during dw/setattr replay the mfd can be found by old handle */
  87         if ((!mfd || mfd->mfd_owner != med) && is_replay_or_resent) {
  88                 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
  89                         if (mfd->mfd_open_handle_old.cookie ==
  90                             open_handle->cookie)
  91                                 RETURN(mfd);
  92                 }
  93                 mfd = NULL;
  94         }
  95
  96         RETURN(mfd);
  97 }
  98
  99 /* free mfd */
 100 void mdt_mfd_free(struct mdt_file_data *mfd)
 101 {
 102         LASSERT(refcount_read(&mfd->mfd_open_handle.h_ref) == 1);
 103         LASSERT(list_empty(&mfd->mfd_list));
 104         OBD_FREE_PRE(mfd, sizeof(*mfd), "rcu");
 105         kfree_rcu(mfd, mfd_open_handle.h_rcu);
 106 }
 107
 108 static int mdt_create_data(struct mdt_thread_info *info,
 109                            struct mdt_object *p, struct mdt_object *o)
 110 {
 111         struct md_op_spec     *spec = &info->mti_spec;
 112         struct md_attr        *ma   = &info->mti_attr;
 113         int                    rc   = 0;
 114         ENTRY;
 115
 116         if (!md_should_create(spec->sp_cr_flags))
 117                 RETURN(0);
 118
 119         ma->ma_need = MA_INODE | MA_LOV;
 120         ma->ma_valid = 0;
 121         mutex_lock(&o->mot_lov_mutex);
 122         if (!o->mot_lov_created) {
 123                 rc = mdo_create_data(info->mti_env,
 124                                      p ? mdt_object_child(p) : NULL,
 125                                      mdt_object_child(o), spec, ma);
 126                 if (rc == 0)
 127                         rc = mdt_attr_get_complex(info, o, ma);
 128
 129                 if (rc == 0 && ma->ma_valid & MA_LOV)
 130                         o->mot_lov_created = 1;
 131         }
 132
 133         mutex_unlock(&o->mot_lov_mutex);
 134         RETURN(rc);
 135 }
 136
 137 int mdt_write_read(struct mdt_object *o)
 138 {
 139         int rc = 0;
 140         ENTRY;
 141         spin_lock(&o->mot_write_lock);
 142         rc = o->mot_write_count;
 143         spin_unlock(&o->mot_write_lock);
 144         RETURN(rc);
 145 }
 146
 147 int mdt_write_get(struct mdt_object *o)
 148 {
 149         int rc = 0;
 150         ENTRY;
 151         spin_lock(&o->mot_write_lock);
 152         if (o->mot_write_count < 0)
 153                 rc = -ETXTBSY;
 154         else
 155                 o->mot_write_count++;
 156         spin_unlock(&o->mot_write_lock);
 157
 158         RETURN(rc);
 159 }
 160
 161 void mdt_write_put(struct mdt_object *o)
 162 {
 163         ENTRY;
 164         spin_lock(&o->mot_write_lock);
 165         o->mot_write_count--;
 166         spin_unlock(&o->mot_write_lock);
 167         EXIT;
 168 }
 169
 170 static int mdt_write_deny(struct mdt_object *o)
 171 {
 172         int rc = 0;
 173         ENTRY;
 174         spin_lock(&o->mot_write_lock);
 175         if (o->mot_write_count > 0)
 176                 rc = -ETXTBSY;
 177         else
 178                 o->mot_write_count--;
 179         spin_unlock(&o->mot_write_lock);
 180         RETURN(rc);
 181 }
 182
 183 static void mdt_write_allow(struct mdt_object *o)
 184 {
 185         ENTRY;
 186         spin_lock(&o->mot_write_lock);
 187         o->mot_write_count++;
 188         spin_unlock(&o->mot_write_lock);
 189         EXIT;
 190 }
 191
 192 /* there can be no real transaction so prepare the fake one */
 193 static void mdt_empty_transno(struct mdt_thread_info *info, int rc)
 194 {
 195         struct mdt_device      *mdt = info->mti_mdt;
 196         struct ptlrpc_request  *req = mdt_info_req(info);
 197         struct tg_export_data  *ted;
 198         struct lsd_client_data *lcd;
 199         ENTRY;
 200
 201         if (mdt_rdonly(req->rq_export))
 202                 RETURN_EXIT;
 203
 204         /* transaction has occurred already */
 205         if (lustre_msg_get_transno(req->rq_repmsg) != 0)
 206                 RETURN_EXIT;
 207
 208         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 209                 struct thandle         *th;
 210
 211                 /* generate an empty transaction to get a transno
 212                  * and reply data */
 213                 th = dt_trans_create(info->mti_env, mdt->mdt_bottom);
 214                 if (!IS_ERR(th)) {
 215                         rc = dt_trans_start(info->mti_env, mdt->mdt_bottom, th);
 216                         dt_trans_stop(info->mti_env, mdt->mdt_bottom, th);
 217                 }
 218                 RETURN_EXIT;
 219         }
 220
 221         spin_lock(&mdt->mdt_lut.lut_translock);
 222         if (rc != 0) {
 223                 if (info->mti_transno != 0) {
 224                         struct obd_export *exp = req->rq_export;
 225
 226                         CERROR("%s: replay trans %llu NID %s: rc = %d\n",
 227                                mdt_obd_name(mdt), info->mti_transno,
 228                                obd_export_nid2str(exp), rc);
 229                         spin_unlock(&mdt->mdt_lut.lut_translock);
 230                         RETURN_EXIT;
 231                 }
 232         } else if (info->mti_transno == 0) {
 233                 info->mti_transno = ++mdt->mdt_lut.lut_last_transno;
 234         } else {
 235                 /* should be replay */
 236                 if (info->mti_transno > mdt->mdt_lut.lut_last_transno)
 237                         mdt->mdt_lut.lut_last_transno = info->mti_transno;
 238         }
 239         spin_unlock(&mdt->mdt_lut.lut_translock);
 240
 241         CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n",
 242                info->mti_transno,
 243                req->rq_export->exp_obd->obd_last_committed);
 244
 245         req->rq_transno = info->mti_transno;
 246         lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
 247
 248         /* update lcd in memory only for resent cases */
 249         ted = &req->rq_export->exp_target_data;
 250         LASSERT(ted);
 251         mutex_lock(&ted->ted_lcd_lock);
 252         lcd = ted->ted_lcd;
 253         if (info->mti_transno < lcd->lcd_last_transno &&
 254             info->mti_transno != 0) {
 255                 /* This should happen during replay. Do not update
 256                  * last rcvd info if replay req transno < last transno,
 257                  * otherwise the following resend(after replay) can not
 258                  * be checked correctly by xid */
 259                 mutex_unlock(&ted->ted_lcd_lock);
 260                 CDEBUG(D_HA, "%s: transno = %llu < last_transno = %llu\n",
 261                        mdt_obd_name(mdt), info->mti_transno,
 262                        lcd->lcd_last_transno);
 263                 RETURN_EXIT;
 264         }
 265
 266         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
 267                 if (info->mti_transno != 0)
 268                         lcd->lcd_last_close_transno = info->mti_transno;
 269                 lcd->lcd_last_close_xid = req->rq_xid;
 270                 lcd->lcd_last_close_result = rc;
 271         } else {
 272                 /* VBR: save versions in last_rcvd for reconstruct. */
 273                 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
 274                 if (pre_versions) {
 275                         lcd->lcd_pre_versions[0] = pre_versions[0];
 276                         lcd->lcd_pre_versions[1] = pre_versions[1];
 277                         lcd->lcd_pre_versions[2] = pre_versions[2];
 278                         lcd->lcd_pre_versions[3] = pre_versions[3];
 279                 }
 280                 if (info->mti_transno != 0)
 281                         lcd->lcd_last_transno = info->mti_transno;
 282
 283                 lcd->lcd_last_xid = req->rq_xid;
 284                 lcd->lcd_last_result = rc;
 285                 lcd->lcd_last_data = info->mti_opdata;
 286         }
 287         mutex_unlock(&ted->ted_lcd_lock);
 288
 289         EXIT;
 290 }
 291
 292 void mdt_mfd_set_mode(struct mdt_file_data *mfd, u64 open_flags)
 293 {
 294         LASSERT(mfd != NULL);
 295
 296         CDEBUG(D_DENTRY, DFID " Change mfd open_flags %#llo -> %#llo.\n",
 297                PFID(mdt_object_fid(mfd->mfd_object)), mfd->mfd_open_flags,
 298                open_flags);
 299
 300         mfd->mfd_open_flags = open_flags;
 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         if (ma->ma_lmv || ma->ma_lmm) {
 311                 CDEBUG(D_INFO, DFID " %s already set.\n",
 312                        PFID(mdt_object_fid(obj)),
 313                        ma->ma_lmv ? (ma->ma_lmm ? "ma_lmv and ma_lmm"
 314                                                 : "ma_lmv")
 315                                   : "ma_lmm");
 316                 return;
 317         }
 318
 319         if (S_ISDIR(obj->mot_header.loh_attr)) {
 320                 ma->ma_lmv = req_capsule_server_get(info->mti_pill,
 321                                                     &RMF_MDT_MD);
 322                 ma->ma_lmv_size = req_capsule_get_size(info->mti_pill,
 323                                                        &RMF_MDT_MD,
 324                                                        RCL_SERVER);
 325                 if (ma->ma_lmv_size > 0)
 326                         ma->ma_need |= MA_LMV;
 327         } else {
 328                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
 329                                                     &RMF_MDT_MD);
 330                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 331                                                        &RMF_MDT_MD,
 332                                                        RCL_SERVER);
 333                 if (ma->ma_lmm_size > 0)
 334                         ma->ma_need |= MA_LOV;
 335         }
 336 }
 337
 338 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
 339                         struct mdt_object *o, u64 open_flags, int created,
 340                         struct ldlm_reply *rep)
 341 {
 342         struct ptlrpc_request *req = mdt_info_req(info);
 343         struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
 344         struct mdt_file_data *mfd;
 345         struct md_attr *ma  = &info->mti_attr;
 346         struct lu_attr *la  = &ma->ma_attr;
 347         struct mdt_body *repbody;
 348         bool isdir, isreg;
 349         int rc = 0;
 350
 351         ENTRY;
 352         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 353
 354         isreg = S_ISREG(la->la_mode);
 355         isdir = S_ISDIR(la->la_mode);
 356         if (isreg && !(ma->ma_valid & MA_LOV) &&
 357             !(open_flags & MDS_OPEN_RELEASE)) {
 358                 /*
 359                  * No EA, check whether it is will set regEA and dirEA since in
 360                  * above attr get, these size might be zero, so reset it, to
 361                  * retrieve the MD after create obj.
 362                  */
 363                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 364                                                        &RMF_MDT_MD,
 365                                                        RCL_SERVER);
 366                 /* in replay case, p == NULL */
 367                 rc = mdt_create_data(info, p, o);
 368                 if (rc)
 369                         RETURN(rc);
 370
 371                 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_DISP_STRIPE)
 372                         mdt_set_disposition(info, rep, DISP_OPEN_STRIPE);
 373         }
 374
 375         CDEBUG(D_INODE, "after open, ma_valid bit = %#llx lmm_size = %d\n",
 376                ma->ma_valid, ma->ma_lmm_size);
 377
 378         if (ma->ma_valid & MA_LOV) {
 379                 LASSERT(ma->ma_lmm_size != 0);
 380                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 381                 if (isdir)
 382                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 383                 else
 384                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 385         }
 386
 387         if (ma->ma_valid & MA_LMV) {
 388                 LASSERT(ma->ma_lmv_size != 0);
 389                 repbody->mbo_eadatasize = ma->ma_lmv_size;
 390                 LASSERT(isdir);
 391                 repbody->mbo_valid |= OBD_MD_FLDIREA | OBD_MD_MEA;
 392         }
 393
 394         if (open_flags & MDS_FMODE_WRITE)
 395                 rc = mdt_write_get(o);
 396         else if (open_flags & MDS_FMODE_EXEC)
 397                 rc = mdt_write_deny(o);
 398
 399         if (rc)
 400                 RETURN(rc);
 401
 402         rc = mo_open(info->mti_env, mdt_object_child(o),
 403                      created ? open_flags | MDS_OPEN_CREATED : open_flags,
 404                      &info->mti_spec);
 405         if (rc != 0) {
 406                 /* If we allow the client to chgrp (CFS_SETGRP_PERM), but the
 407                  * client does not know which suppgid should be sent to the MDS,
 408                  * or some other(s) changed the target file's GID after this RPC
 409                  * sent to the MDS with the suppgid as the original GID, then we
 410                  * give the client another chance to send the right suppgid. */
 411                 if (rc == -EACCES &&
 412                     allow_client_chgrp(info, lu_ucred(info->mti_env)))
 413                         mdt_set_disposition(info, rep, DISP_OPEN_DENY);
 414
 415                 GOTO(err_out, rc);
 416         }
 417
 418         mfd = mdt_mfd_new(med);
 419         if (mfd == NULL)
 420                 GOTO(err_out, rc = -ENOMEM);
 421
 422         /*
 423          * Keep a reference on this object for this open, and is
 424          * released by mdt_mfd_close().
 425          */
 426         mdt_object_get(info->mti_env, o);
 427         mfd->mfd_object = o;
 428         mfd->mfd_xid = req->rq_xid;
 429
 430         /*
 431          * @open_flags is always not zero. At least it should be FMODE_READ,
 432          * FMODE_WRITE or MDS_FMODE_EXEC.
 433          */
 434         LASSERT(open_flags != 0);
 435
 436         /* Open handling. */
 437         mdt_mfd_set_mode(mfd, open_flags);
 438
 439         atomic_inc(&o->mot_open_count);
 440         if (open_flags & MDS_OPEN_LEASE)
 441                 atomic_inc(&o->mot_lease_count);
 442
 443         /* replay handle */
 444         if (req_is_replay(req)) {
 445                 struct mdt_file_data *old_mfd;
 446                 /* Check wheather old cookie already exist in
 447                  * the list, becasue when do recovery, client
 448                  * might be disconnected from server, and
 449                  * restart replay, so there maybe some orphan
 450                  * mfd here, we should remove them */
 451                 LASSERT(info->mti_rr.rr_open_handle != NULL);
 452                 spin_lock(&med->med_open_lock);
 453                 old_mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
 454                                               true);
 455                 if (old_mfd != NULL) {
 456                         CDEBUG(D_HA, "delete orphan mfd = %p, fid = "DFID", "
 457                                "cookie = %#llx\n", mfd,
 458                                PFID(mdt_object_fid(mfd->mfd_object)),
 459                                info->mti_rr.rr_open_handle->cookie);
 460                         class_handle_unhash(&old_mfd->mfd_open_handle);
 461                         list_del_init(&old_mfd->mfd_list);
 462                         spin_unlock(&med->med_open_lock);
 463                         /* no attr update for that close */
 464                         la->la_valid = 0;
 465                         ma->ma_valid |= MA_FLAGS;
 466                         ma->ma_attr_flags |= MDS_RECOV_OPEN;
 467                         mdt_mfd_close(info, old_mfd);
 468                         ma->ma_attr_flags &= ~MDS_RECOV_OPEN;
 469                         ma->ma_valid &= ~MA_FLAGS;
 470                 } else {
 471                         spin_unlock(&med->med_open_lock);
 472                         CDEBUG(D_HA, "orphan mfd not found, fid = "DFID", "
 473                                "cookie = %#llx\n",
 474                                PFID(mdt_object_fid(mfd->mfd_object)),
 475                                info->mti_rr.rr_open_handle->cookie);
 476                 }
 477
 478                 CDEBUG(D_HA, "Store old cookie %#llx in new mfd\n",
 479                        info->mti_rr.rr_open_handle->cookie);
 480
 481                 mfd->mfd_open_handle_old = *info->mti_rr.rr_open_handle;
 482         }
 483
 484         repbody->mbo_open_handle.cookie = mfd->mfd_open_handle.h_cookie;
 485
 486         if (req->rq_export->exp_disconnected) {
 487                 spin_lock(&med->med_open_lock);
 488                 class_handle_unhash(&mfd->mfd_open_handle);
 489                 list_del_init(&mfd->mfd_list);
 490                 spin_unlock(&med->med_open_lock);
 491                 mdt_mfd_close(info, mfd);
 492         } else {
 493                 spin_lock(&med->med_open_lock);
 494                 list_add_tail(&mfd->mfd_list, &med->med_open_head);
 495                 spin_unlock(&med->med_open_lock);
 496         }
 497
 498         mdt_empty_transno(info, rc);
 499
 500         RETURN(rc);
 501
 502 err_out:
 503         if (open_flags & MDS_FMODE_WRITE)
 504                 mdt_write_put(o);
 505         else if (open_flags & MDS_FMODE_EXEC)
 506                 mdt_write_allow(o);
 507
 508         return rc;
 509 }
 510
 511 static int mdt_finish_open(struct mdt_thread_info *info,
 512                            struct mdt_object *p, struct mdt_object *o,
 513                            u64 open_flags, int created,
 514                            struct ldlm_reply *rep)
 515 {
 516         struct ptlrpc_request   *req = mdt_info_req(info);
 517         struct obd_export       *exp = req->rq_export;
 518         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 519         struct md_attr          *ma  = &info->mti_attr;
 520         struct lu_attr          *la  = &ma->ma_attr;
 521         struct mdt_file_data    *mfd;
 522         struct mdt_body         *repbody;
 523         int                      rc = 0;
 524         int                      isreg, isdir, islnk;
 525         struct list_head        *t;
 526         ENTRY;
 527
 528         LASSERT(ma->ma_valid & MA_INODE);
 529
 530         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 531
 532         isreg = S_ISREG(la->la_mode);
 533         isdir = S_ISDIR(la->la_mode);
 534         islnk = S_ISLNK(la->la_mode);
 535         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 536
 537         /* compatibility check for 2.10 clients when it tries to open mirrored
 538          * files. 2.10 clients don't verify overlapping components so they
 539          * would read and write mirrored files just as if they were normal
 540          * PFL files, which will cause the problem that sycned mirrors actually
 541          * contain different data.
 542          * Older clients are not a concern here because they don't even
 543          * understand PFL layout. */
 544         if (isreg && !exp_connect_flr(exp) && ma->ma_valid & MA_LOV &&
 545             mdt_lmm_is_flr(ma->ma_lmm)) {
 546                 /* LU-10286: for simplicity clients who don't understand
 547                  * mirrored layout(with connect flag OBD_CONNECT2_FLR) won't
 548                  * be able to open mirrored files */
 549                 RETURN(-EOPNOTSUPP);
 550         }
 551
 552         /* Overstriped files can crash older clients */
 553         if (isreg && !exp_connect_overstriping(exp) &&
 554             mdt_lmm_is_overstriping(ma->ma_lmm))
 555                 RETURN(-EOPNOTSUPP);
 556
 557         /* LU-2275, simulate broken behaviour (esp. prevalent in
 558          * pre-2.4 servers where a very strange reply is sent on error
 559          * that looks like it was actually almost successful and a
 560          * failure at the same time.) */
 561         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_NEGATIVE_POSITIVE)) {
 562                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN |
 563                                                DISP_LOOKUP_NEG |
 564                                                DISP_LOOKUP_POS);
 565
 566                 if (open_flags & MDS_OPEN_LOCK)
 567                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
 568
 569                 RETURN(-ENOENT);
 570         }
 571
 572 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 573         if (exp_connect_flags(exp) & OBD_CONNECT_ACL) {
 574                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
 575                 if (IS_ERR(nodemap))
 576                         RETURN(PTR_ERR(nodemap));
 577
 578                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
 579                 nodemap_putref(nodemap);
 580                 if (rc)
 581                         RETURN(rc);
 582         }
 583 #endif
 584
 585         /*
 586          * If we are following a symlink, don't open; and do not return open
 587          * handle for special nodes as client required.
 588          */
 589         if (islnk || (!isreg && !isdir &&
 590             (exp_connect_flags(req->rq_export) & OBD_CONNECT_NODEVOH))) {
 591                 lustre_msg_set_transno(req->rq_repmsg, 0);
 592                 RETURN(0);
 593         }
 594
 595         /*
 596          * We need to return the existing object's fid back, so it is done here,
 597          * after preparing the reply.
 598          */
 599         if (!created && (open_flags & MDS_OPEN_EXCL) &&
 600             (open_flags & MDS_OPEN_CREAT))
 601                 RETURN(-EEXIST);
 602
 603         /* This can't be done earlier, we need to return reply body */
 604         if (isdir) {
 605                 if (open_flags & (MDS_OPEN_CREAT | MDS_FMODE_WRITE)) {
 606                         /* We are trying to create or write an existing dir. */
 607                         RETURN(-EISDIR);
 608                 }
 609         } else if (open_flags & MDS_OPEN_DIRECTORY)
 610                 RETURN(-ENOTDIR);
 611
 612         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
 613                                  OBD_FAIL_MDS_LDLM_REPLY_NET | OBD_FAIL_ONCE))
 614                 RETURN(-EAGAIN);
 615
 616         mfd = NULL;
 617         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 618                 spin_lock(&med->med_open_lock);
 619                 list_for_each(t, &med->med_open_head) {
 620                         mfd = list_entry(t, struct mdt_file_data, mfd_list);
 621                         if (mfd->mfd_xid == req->rq_xid)
 622                                 break;
 623                         mfd = NULL;
 624                 }
 625                 spin_unlock(&med->med_open_lock);
 626
 627                 if (mfd != NULL) {
 628                         repbody->mbo_open_handle.cookie =
 629                                 mfd->mfd_open_handle.h_cookie;
 630                         /* set repbody->ea_size for resent case */
 631                         if (ma->ma_valid & MA_LOV) {
 632                                 LASSERT(ma->ma_lmm_size != 0);
 633                                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 634                                 if (isdir)
 635                                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 636                                 else
 637                                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 638                         }
 639                         mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 640                         RETURN(0);
 641                 }
 642         }
 643
 644         rc = mdt_mfd_open(info, p, o, open_flags, created, rep);
 645         if (!rc)
 646                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 647
 648         RETURN(rc);
 649 }
 650
 651 void mdt_reconstruct_open(struct mdt_thread_info *info,
 652                           struct mdt_lock_handle *lhc)
 653 {
 654         const struct lu_env *env = info->mti_env;
 655         struct mdt_device *mdt = info->mti_mdt;
 656         struct req_capsule *pill = info->mti_pill;
 657         struct ptlrpc_request *req = mdt_info_req(info);
 658         struct md_attr *ma = &info->mti_attr;
 659         struct mdt_reint_record *rr = &info->mti_rr;
 660         u64 open_flags = info->mti_spec.sp_cr_flags;
 661         struct ldlm_reply *ldlm_rep;
 662         struct mdt_object *parent;
 663         struct mdt_object *child;
 664         struct mdt_body *repbody;
 665         u64 opdata;
 666         int rc;
 667         ENTRY;
 668
 669         LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
 670         ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
 671         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 672
 673         ma->ma_need = MA_INODE | MA_HSM;
 674         ma->ma_valid = 0;
 675
 676         opdata = mdt_req_from_lrd(req, info->mti_reply_data);
 677         mdt_set_disposition(info, ldlm_rep, opdata);
 678
 679         CDEBUG(D_INODE, "This is reconstruct open: disp=%#llx, result=%d\n",
 680                ldlm_rep->lock_policy_res1, req->rq_status);
 681
 682         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
 683             req->rq_status != 0)
 684                 /* We did not create successfully, return error to client. */
 685                 GOTO(out, rc = req->rq_status);
 686
 687         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
 688                 struct obd_export *exp = req->rq_export;
 689                 /*
 690                  * We failed after creation, but we do not know in which step
 691                  * we failed. So try to check the child object.
 692                  */
 693                 parent = mdt_object_find(env, mdt, rr->rr_fid1);
 694                 if (IS_ERR(parent)) {
 695                         rc = PTR_ERR(parent);
 696                         LCONSOLE_WARN("Parent "DFID" lookup error %d."
 697                                       " Evicting client %s with export %s.\n",
 698                                       PFID(rr->rr_fid1), rc,
 699                                       obd_uuid2str(&exp->exp_client_uuid),
 700                                       obd_export_nid2str(exp));
 701                         mdt_export_evict(exp);
 702                         RETURN_EXIT;
 703                 }
 704
 705                 child = mdt_object_find(env, mdt, rr->rr_fid2);
 706                 if (IS_ERR(child)) {
 707                         rc = PTR_ERR(child);
 708                         LCONSOLE_WARN("cannot lookup child "DFID": rc = %d; "
 709                                       "evicting client %s with export %s\n",
 710                                       PFID(rr->rr_fid2), rc,
 711                                       obd_uuid2str(&exp->exp_client_uuid),
 712                                       obd_export_nid2str(exp));
 713                         mdt_object_put(env, parent);
 714                         mdt_export_evict(exp);
 715                         RETURN_EXIT;
 716                 }
 717
 718                 if (unlikely(mdt_object_remote(child))) {
 719                         mdt_object_put(env, parent);
 720                         mdt_object_put(env, child);
 721                         /* the child object was created on remote server */
 722                         if (!mdt_is_dne_client(exp))
 723                                 /* Return -EIO for old client */
 724                                 GOTO(out, rc = -EIO);
 725                         repbody->mbo_fid1 = *rr->rr_fid2;
 726                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 727                         GOTO(out, rc = 0);
 728                 }
 729                 if (mdt_object_exists(child)) {
 730                         mdt_prep_ma_buf_from_rep(info, child, ma);
 731                         rc = mdt_attr_get_complex(info, child, ma);
 732                         if (!rc)
 733                                 rc = mdt_finish_open(info, parent, child,
 734                                                      open_flags, 1, ldlm_rep);
 735                         mdt_object_put(env, parent);
 736                         mdt_object_put(env, child);
 737                         if (!rc)
 738                                 mdt_pack_size2body(info, rr->rr_fid2,
 739                                                    &lhc->mlh_reg_lh);
 740                         GOTO(out, rc);
 741                 }
 742                 /* the child does not exist, we should do regular open */
 743                 mdt_object_put(env, parent);
 744                 mdt_object_put(env, child);
 745         }
 746         /* We did not try to create, so we are a pure open */
 747         rc = mdt_reint_open(info, lhc);
 748         EXIT;
 749 out:
 750         req->rq_status = rc;
 751         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
 752         LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
 753 }
 754
 755 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep)
 756 {
 757         u64 open_flags = info->mti_spec.sp_cr_flags;
 758         struct mdt_reint_record *rr = &info->mti_rr;
 759         struct md_attr *ma = &info->mti_attr;
 760         struct mdt_object *o;
 761         int rc;
 762
 763         ENTRY;
 764         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
 765         if (IS_ERR(o))
 766                 RETURN(rc = PTR_ERR(o));
 767
 768         if (unlikely(mdt_object_remote(o))) {
 769                 /* the child object was created on remote server */
 770                 struct mdt_body *repbody;
 771
 772                 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 773                                                 DISP_LOOKUP_EXECD |
 774                                                 DISP_LOOKUP_POS));
 775                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 776                 repbody->mbo_fid1 = *rr->rr_fid2;
 777                 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 778                 rc = 0;
 779         } else {
 780                 if (mdt_object_exists(o)) {
 781                         mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 782                                                         DISP_LOOKUP_EXECD |
 783                                                         DISP_LOOKUP_POS));
 784                         mdt_prep_ma_buf_from_rep(info, o, ma);
 785                         rc = mdt_attr_get_complex(info, o, ma);
 786                         if (rc == 0)
 787                                 rc = mdt_finish_open(info, NULL, o, open_flags,
 788                                                      0, rep);
 789                 } else {
 790                         rc = -ENOENT;
 791                 }
 792         }
 793
 794         mdt_object_put(info->mti_env, o);
 795         RETURN(rc);
 796 }
 797
 798 /* lock object for open */
 799 static int mdt_object_open_lock(struct mdt_thread_info *info,
 800                                 struct mdt_object *obj,
 801                                 struct mdt_lock_handle *lhc,
 802                                 __u64 *ibits)
 803 {
 804         struct md_attr *ma = &info->mti_attr;
 805         __u64 open_flags = info->mti_spec.sp_cr_flags;
 806         __u64 trybits = 0;
 807         enum ldlm_mode lm = LCK_CR;
 808         bool acq_lease = !!(open_flags & MDS_OPEN_LEASE);
 809         bool try_layout = false;
 810         bool create_layout = false;
 811         int rc = 0;
 812         __u32 dom_stripe = 0;
 813         unsigned int dom_only = 0;
 814         unsigned int dom_lock = 0;
 815
 816         ENTRY;
 817
 818         *ibits = 0;
 819         mdt_lock_handle_init(lhc);
 820
 821         if (req_is_replay(mdt_info_req(info)))
 822                 RETURN(0);
 823
 824         if (S_ISREG(lu_object_attr(&obj->mot_obj))) {
 825                 if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV) &&
 826                     md_should_create(open_flags))
 827                         create_layout = true;
 828                 if (exp_connect_layout(info->mti_exp) && !create_layout &&
 829                     ma->ma_need & MA_LOV)
 830                         try_layout = true;
 831
 832                 /* DoM files can take IO lock at OPEN when it makes sense,
 833                  * check if file has DoM stripe and ask for lock if client
 834                  * no lock on that resource yet.
 835                  */
 836                 if (ma->ma_valid & MA_LOV && ma->ma_lmm != NULL)
 837                         dom_stripe = mdt_lmm_dom_entry_check(ma->ma_lmm,
 838                                                              &dom_only);
 839                 /* If only DOM stripe is being used then we can expect IO
 840                  * to it after OPEN and will return corresponding DOM ibit
 841                  * using default strategy from mdt_opts.mo_dom_lock.
 842                  * Otherwise trylock mode is used always and DOM ibit will
 843                  * be returned optionally.
 844                  */
 845                 if (dom_stripe &&
 846                     !mdt_dom_client_has_lock(info, mdt_object_fid(obj)))
 847                         dom_lock = !dom_only ? TRYLOCK_DOM_ON_OPEN :
 848                                    info->mti_mdt->mdt_opts.mo_dom_lock;
 849         }
 850
 851         if (acq_lease) {
 852                 /* lease open, acquire write mode of open sem */
 853                 down_write(&obj->mot_open_sem);
 854
 855                 /* Lease exists and ask for new lease */
 856                 if (atomic_read(&obj->mot_lease_count) > 0) {
 857                         /* only exclusive open is supported, so lease
 858                          * are conflicted to each other */
 859                         GOTO(out, rc = -EBUSY);
 860                 }
 861
 862                 /* Lease must be with open lock */
 863                 if (!(open_flags & MDS_OPEN_LOCK)) {
 864                         CERROR("%s: Request lease for file:"DFID ", but open lock "
 865                                "is missed, open_flags = %#llo : rc = %d\n",
 866                                mdt_obd_name(info->mti_mdt),
 867                                PFID(mdt_object_fid(obj)), open_flags, -EPROTO);
 868                         GOTO(out, rc = -EPROTO);
 869                 }
 870
 871                 /* XXX: only exclusive open is supported. */
 872                 lm = LCK_EX;
 873                 *ibits = MDS_INODELOCK_OPEN;
 874
 875                 /* never grant LCK_EX layout lock to client */
 876                 try_layout = false;
 877         } else { /* normal open */
 878                 /* normal open holds read mode of open sem */
 879                 down_read(&obj->mot_open_sem);
 880
 881                 if (open_flags & MDS_OPEN_LOCK) {
 882                         if (open_flags & MDS_FMODE_WRITE)
 883                                 lm = LCK_CW;
 884                         else if (open_flags & MDS_FMODE_EXEC)
 885                                 lm = LCK_PR;
 886                         else
 887                                 lm = LCK_CR;
 888
 889                         *ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN;
 890                 } else if (atomic_read(&obj->mot_lease_count) > 0) {
 891                         if (open_flags & MDS_FMODE_WRITE)
 892                                 lm = LCK_CW;
 893                         else
 894                                 lm = LCK_CR;
 895
 896                         /* revoke lease */
 897                         *ibits = MDS_INODELOCK_OPEN;
 898                         try_layout = false;
 899
 900                         lhc = &info->mti_lh[MDT_LH_LOCAL];
 901                 } else if (dom_lock) {
 902                         lm = (open_flags & MDS_FMODE_WRITE) ? LCK_PW : LCK_PR;
 903                         if (dom_lock == TRYLOCK_DOM_ON_OPEN) {
 904                                 trybits |= MDS_INODELOCK_DOM |
 905                                            MDS_INODELOCK_LAYOUT;
 906                         } else {
 907                                 /* dom_lock == ALWAYS_DOM_LOCK_ON_OPEN */
 908                                 *ibits = MDS_INODELOCK_DOM;
 909                                 if (info->mti_mdt->mdt_opts.mo_dom_read_open)
 910                                         trybits |= MDS_INODELOCK_LAYOUT;
 911                         }
 912                 }
 913
 914                 CDEBUG(D_INODE, "normal open:"DFID" lease count: %d, lm: %d\n",
 915                         PFID(mdt_object_fid(obj)),
 916                         atomic_read(&obj->mot_lease_count), lm);
 917         }
 918
 919         mdt_lock_reg_init(lhc, lm);
 920
 921         /* Return lookup lock to validate inode at the client side.
 922          * This is pretty important otherwise MDT will return layout
 923          * lock for each open.
 924          * However this is a double-edged sword because changing
 925          * permission will revoke a huge number of LOOKUP locks.
 926          */
 927         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_OPEN) && try_layout) {
 928                 if (!(*ibits & MDS_INODELOCK_LOOKUP))
 929                         trybits |= MDS_INODELOCK_LOOKUP;
 930                 trybits |= MDS_INODELOCK_LAYOUT;
 931         }
 932
 933         if (*ibits | trybits)
 934                 rc = mdt_object_lock_try(info, obj, lhc, ibits, trybits, false);
 935
 936         CDEBUG(D_INODE, "%s: Requested bits lock:"DFID ", ibits = %#llx/%#llx"
 937                ", open_flags = %#llo, try_layout = %d : rc = %d\n",
 938                mdt_obd_name(info->mti_mdt), PFID(mdt_object_fid(obj)),
 939                *ibits, trybits, open_flags, try_layout, rc);
 940
 941         /* will change layout, revoke layout locks by enqueuing EX lock. */
 942         if (rc == 0 && create_layout) {
 943                 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LAYOUT];
 944
 945                 CDEBUG(D_INODE, "Will create layout, get EX layout lock:"DFID
 946                         ", open_flags = %#llo\n",
 947                         PFID(mdt_object_fid(obj)), open_flags);
 948
 949                 /* We cannot enqueue another lock for the same resource we
 950                  * already have a lock for, due to mechanics of waiting list
 951                  * iterating in ldlm, see LU-3601.
 952                  * As such we'll drop the open lock we just got above here,
 953                  * it's ok not to have this open lock as it's main purpose is to
 954                  * flush unused cached client open handles. */
 955                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
 956                         mdt_object_unlock(info, obj, lhc, 1);
 957
 958                 LASSERT(!try_layout);
 959                 mdt_lock_handle_init(ll);
 960                 mdt_lock_reg_init(ll, LCK_EX);
 961                 rc = mdt_object_lock(info, obj, ll, MDS_INODELOCK_LAYOUT);
 962
 963                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LL_BLOCK, 2);
 964         }
 965
 966         /* Check if there is any other open handles after acquiring
 967          * open lock. At this point, caching open handles have been revoked
 968          * by open lock.
 969          * XXX: Now only exclusive open is supported. Need to check the
 970          * type of open for generic lease support. */
 971         if (rc == 0 && acq_lease) {
 972                 struct ptlrpc_request *req = mdt_info_req(info);
 973                 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
 974                 struct mdt_file_data *mfd;
 975                 bool is_replay_or_resent;
 976                 int open_count = 0;
 977
 978                 /* For lease: application can open a file and then apply lease,
 979                  * @handle contains original open handle in that case.
 980                  * In recovery, open REQ will be replayed and the lease REQ may
 981                  * be resent that means the open handle is already stale, so we
 982                  * need to fix it up here by finding new handle. */
 983                 is_replay_or_resent = req_is_replay(req) ||
 984                         lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
 985
 986                 /* if the request is _not_ a replay request, rr_open_handle
 987                  * may be used to hold an open file handle which is issuing the
 988                  * lease request, so that this openhandle doesn't count. */
 989                 mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
 990                                           is_replay_or_resent);
 991                 if (mfd != NULL)
 992                         ++open_count;
 993
 994                 CDEBUG(D_INODE, "acq_lease "DFID": openers: %d, want: %d\n",
 995                         PFID(mdt_object_fid(obj)),
 996                         atomic_read(&obj->mot_open_count), open_count);
 997
 998                 if (atomic_read(&obj->mot_open_count) > open_count)
 999                         GOTO(out, rc = -EBUSY);
1000         }
1001         GOTO(out, rc);
1002
1003 out:
1004         RETURN(rc);
1005 }
1006
1007 static void mdt_object_open_unlock(struct mdt_thread_info *info,
1008                                    struct mdt_object *obj,
1009                                    struct mdt_lock_handle *lhc,
1010                                    __u64 ibits, int rc)
1011 {
1012         __u64 open_flags = info->mti_spec.sp_cr_flags;
1013         struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LOCAL];
1014         ENTRY;
1015
1016         if (req_is_replay(mdt_info_req(info)))
1017                 RETURN_EXIT;
1018
1019         /* Release local lock - the lock put in MDT_LH_LOCAL will never
1020          * return to client side. */
1021         if (lustre_handle_is_used(&ll->mlh_reg_lh))
1022                 mdt_object_unlock(info, obj, ll, 1);
1023
1024         ll = &info->mti_lh[MDT_LH_LAYOUT];
1025         /* Release local layout lock, layout was created */
1026         if (lustre_handle_is_used(&ll->mlh_reg_lh)) {
1027                 LASSERT(!(ibits & MDS_INODELOCK_LAYOUT));
1028                 mdt_object_unlock(info, obj, ll, 1);
1029         }
1030
1031         if (open_flags & MDS_OPEN_LEASE)
1032                 up_write(&obj->mot_open_sem);
1033         else
1034                 up_read(&obj->mot_open_sem);
1035
1036         /* Cross-ref case, the lock should be returned to the client */
1037         if (ibits == 0 || rc == -MDT_EREMOTE_OPEN)
1038                 RETURN_EXIT;
1039
1040         if (!(open_flags & MDS_OPEN_LOCK) && !(ibits & MDS_INODELOCK_LAYOUT) &&
1041             !(ibits & MDS_INODELOCK_DOM)) {
1042                 /* for the open request, the lock will only return to client
1043                  * if open or layout lock is granted. */
1044                 rc = 1;
1045         }
1046
1047         if (rc != 0 || !lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1048                 struct ldlm_reply       *ldlm_rep;
1049
1050                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1051                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1052                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
1053                         mdt_object_unlock(info, obj, lhc, 1);
1054         }
1055         RETURN_EXIT;
1056 }
1057
1058 /**
1059  * Check release is permitted for the current HSM flags.
1060  */
1061 static bool mdt_hsm_release_allow(const struct md_attr *ma)
1062 {
1063         if (!(ma->ma_valid & MA_HSM))
1064                 return false;
1065
1066         if (ma->ma_hsm.mh_flags & (HS_DIRTY|HS_NORELEASE|HS_LOST))
1067                 return false;
1068
1069         if (!(ma->ma_hsm.mh_flags & HS_ARCHIVED))
1070                 return false;
1071
1072         return true;
1073 }
1074
1075 static int mdt_open_by_fid_lock(struct mdt_thread_info *info,
1076                                 struct ldlm_reply *rep,
1077                                 struct mdt_lock_handle *lhc)
1078 {
1079         const struct lu_env *env = info->mti_env;
1080         struct mdt_device *mdt = info->mti_mdt;
1081         u64 open_flags = info->mti_spec.sp_cr_flags;
1082         struct mdt_reint_record *rr = &info->mti_rr;
1083         struct md_attr *ma = &info->mti_attr;
1084         struct mdt_object *parent = NULL;
1085         struct mdt_object *o;
1086         bool object_locked = false;
1087         u64 ibits = 0;
1088         int rc;
1089
1090         ENTRY;
1091         if (md_should_create(open_flags)) {
1092                 if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1093                         parent = mdt_object_find(env, mdt, rr->rr_fid1);
1094                         if (IS_ERR(parent)) {
1095                                 CDEBUG(D_INODE, "Fail to find parent "DFID
1096                                        " for anonymous created %ld, try to"
1097                                        " use server-side parent.\n",
1098                                        PFID(rr->rr_fid1), PTR_ERR(parent));
1099                                 parent = NULL;
1100                         }
1101                 }
1102                 if (parent == NULL)
1103                         ma->ma_need |= MA_PFID;
1104         }
1105
1106         o = mdt_object_find(env, mdt, rr->rr_fid2);
1107         if (IS_ERR(o))
1108                 GOTO(out_parent_put, rc = PTR_ERR(o));
1109
1110         if (mdt_object_remote(o)) {
1111                 CDEBUG(D_INFO, "%s: "DFID" is on remote MDT.\n",
1112                        mdt_obd_name(info->mti_mdt),
1113                        PFID(rr->rr_fid2));
1114                 GOTO(out, rc = -EREMOTE);
1115         } else if (!mdt_object_exists(o)) {
1116                 mdt_set_disposition(info, rep,
1117                                     DISP_IT_EXECD |
1118                                     DISP_LOOKUP_EXECD |
1119                                     DISP_LOOKUP_NEG);
1120                 GOTO(out, rc = -ENOENT);
1121         }
1122
1123         mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1124
1125         mdt_prep_ma_buf_from_rep(info, o, ma);
1126         if (open_flags & MDS_OPEN_RELEASE)
1127                 ma->ma_need |= MA_HSM;
1128         rc = mdt_attr_get_complex(info, o, ma);
1129         if (rc)
1130                 GOTO(out, rc);
1131
1132         /* We should not change file's existing LOV EA */
1133         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
1134             open_flags & MDS_OPEN_HAS_EA && ma->ma_valid & MA_LOV)
1135                 GOTO(out, rc = -EEXIST);
1136
1137         /* If a release request, check file open flags are fine and ask for an
1138          * exclusive open access. */
1139         if (open_flags & MDS_OPEN_RELEASE && !mdt_hsm_release_allow(ma))
1140                 GOTO(out, rc = -EPERM);
1141
1142         rc = mdt_check_resent_lock(info, o, lhc);
1143         if (rc < 0) {
1144                 GOTO(out, rc);
1145         } else if (rc > 0) {
1146                 rc = mdt_object_open_lock(info, o, lhc, &ibits);
1147                 object_locked = true;
1148                 if (rc)
1149                         GOTO(out_unlock, rc);
1150         }
1151
1152         if (ma->ma_valid & MA_PFID) {
1153                 parent = mdt_object_find(env, mdt, &ma->ma_pfid);
1154                 if (IS_ERR(parent)) {
1155                         CDEBUG(D_INODE, "Fail to find parent "DFID
1156                                " for anonymous created %ld, try to"
1157                                " use system default.\n",
1158                                PFID(&ma->ma_pfid), PTR_ERR(parent));
1159                         parent = NULL;
1160                 }
1161         }
1162
1163         rc = mdt_finish_open(info, parent, o, open_flags, 0, rep);
1164         if (!rc) {
1165                 mdt_set_disposition(info, rep, DISP_LOOKUP_POS);
1166                 if (open_flags & MDS_OPEN_LOCK)
1167                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1168                 if (open_flags & MDS_OPEN_LEASE)
1169                         mdt_set_disposition(info, rep, DISP_OPEN_LEASE);
1170         }
1171         GOTO(out_unlock, rc);
1172
1173 out_unlock:
1174         if (object_locked)
1175                 mdt_object_open_unlock(info, o, lhc, ibits, rc);
1176 out:
1177         mdt_object_put(env, o);
1178         if (rc == 0)
1179                 mdt_pack_size2body(info, rr->rr_fid2, &lhc->mlh_reg_lh);
1180 out_parent_put:
1181         if (parent != NULL)
1182                 mdt_object_put(env, parent);
1183         return rc;
1184 }
1185
1186 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1187 static int mdt_cross_open(struct mdt_thread_info *info,
1188                           const struct lu_fid *parent_fid,
1189                           const struct lu_fid *fid,
1190                           struct ldlm_reply *rep, u64 open_flags)
1191 {
1192         struct md_attr *ma = &info->mti_attr;
1193         struct mdt_object *o;
1194         int rc;
1195         ENTRY;
1196
1197         o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1198         if (IS_ERR(o))
1199                 RETURN(rc = PTR_ERR(o));
1200
1201         if (mdt_object_remote(o)) {
1202                 /* Something is wrong here, the object is on another MDS! */
1203                 CERROR("%s: "DFID" isn't on this server!: rc = %d\n",
1204                        mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1205                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1206                                 &o->mot_obj,
1207                                 "Object isn't on this server! FLD error?");
1208                 rc = -EFAULT;
1209         } else {
1210                 if (mdt_object_exists(o)) {
1211                         int mask;
1212
1213                         /* Do permission check for cross-open after converting
1214                          * MDS_OPEN_* flags to MAY_* permission mask.
1215                          */
1216                         mask = mds_accmode(open_flags);
1217
1218                         rc = mo_permission(info->mti_env, NULL,
1219                                            mdt_object_child(o), NULL, mask);
1220                         if (rc)
1221                                 goto out;
1222
1223                         mdt_prep_ma_buf_from_rep(info, o, ma);
1224                         rc = mdt_attr_get_complex(info, o, ma);
1225                         if (rc != 0)
1226                                 GOTO(out, rc);
1227
1228                         rc = mdt_pack_secctx_in_reply(info, o);
1229                         if (unlikely(rc))
1230                                 GOTO(out, rc);
1231
1232                         rc = mdt_pack_encctx_in_reply(info, o);
1233                         if (unlikely(rc))
1234                                 GOTO(out, rc);
1235
1236                         rc = mdt_finish_open(info, NULL, o, open_flags, 0, rep);
1237                 } else {
1238                         /*
1239                          * Something is wrong here. lookup was positive but
1240                          * there is no object!
1241                          */
1242                         CERROR("%s: "DFID" doesn't exist!: rc = %d\n",
1243                               mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1244                         rc = -EFAULT;
1245                 }
1246         }
1247 out:
1248         mdt_object_put(info->mti_env, o);
1249
1250         RETURN(rc);
1251 }
1252
1253 /*
1254  * find root object and take its xattr lock if it's on remote MDT, later create
1255  * may use fs default striping (which is stored in root xattr).
1256  */
1257 static int mdt_lock_root_xattr(struct mdt_thread_info *info,
1258                                struct mdt_device *mdt)
1259 {
1260         struct mdt_object *md_root = mdt->mdt_md_root;
1261         struct lustre_handle lhroot;
1262         int rc;
1263
1264         if (md_root == NULL) {
1265                 lu_root_fid(&info->mti_tmp_fid1);
1266                 md_root = mdt_object_find(info->mti_env, mdt,
1267                                           &info->mti_tmp_fid1);
1268                 if (IS_ERR(md_root))
1269                         return PTR_ERR(md_root);
1270
1271                 spin_lock(&mdt->mdt_lock);
1272                 if (mdt->mdt_md_root != NULL) {
1273                         spin_unlock(&mdt->mdt_lock);
1274
1275                         LASSERTF(mdt->mdt_md_root == md_root,
1276                                  "Different root object ("
1277                                  DFID") instances, %p, %p\n",
1278                                  PFID(&info->mti_tmp_fid1),
1279                                  mdt->mdt_md_root, md_root);
1280                         LASSERT(atomic_read(
1281                                 &md_root->mot_obj.lo_header->loh_ref) > 1);
1282
1283                         mdt_object_put(info->mti_env, md_root);
1284                 } else {
1285                         mdt->mdt_md_root = md_root;
1286                         spin_unlock(&mdt->mdt_lock);
1287                 }
1288         }
1289
1290         if (md_root->mot_cache_attr || !mdt_object_remote(md_root))
1291                 return 0;
1292
1293         rc = mdt_remote_object_lock(info, md_root, mdt_object_fid(md_root),
1294                                     &lhroot, LCK_PR, MDS_INODELOCK_XATTR,
1295                                     true);
1296         if (rc < 0)
1297                 return rc;
1298
1299         md_root->mot_cache_attr = 1;
1300
1301         /* don't cancel this lock, so that we know the cached xattr is valid. */
1302         ldlm_lock_decref(&lhroot, LCK_PR);
1303
1304         return 0;
1305 }
1306
1307 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1308 {
1309         struct mdt_device *mdt = info->mti_mdt;
1310         struct ptlrpc_request *req = mdt_info_req(info);
1311         struct mdt_object *parent;
1312         struct mdt_object *child;
1313         struct mdt_lock_handle *lh;
1314         struct ldlm_reply *ldlm_rep;
1315         struct mdt_body *repbody;
1316         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1317         struct md_attr *ma = &info->mti_attr;
1318         u64 open_flags = info->mti_spec.sp_cr_flags;
1319         u64 ibits = 0;
1320         struct mdt_reint_record *rr = &info->mti_rr;
1321         int result, rc;
1322         int created = 0;
1323         int object_locked = 0;
1324         u32 msg_flags;
1325         ktime_t kstart = ktime_get();
1326
1327         ENTRY;
1328         OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1329                                (obd_timeout + 1) / 4);
1330
1331         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1332
1333         ma->ma_need = MA_INODE;
1334         ma->ma_valid = 0;
1335
1336         LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1337         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1338
1339         if (unlikely(open_flags & MDS_OPEN_JOIN_FILE)) {
1340                 CERROR("file join is not supported anymore.\n");
1341                 GOTO(out, result = err_serious(-EOPNOTSUPP));
1342         }
1343         msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1344
1345         if ((open_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1346             info->mti_spec.u.sp_ea.eadata == NULL)
1347                 GOTO(out, result = err_serious(-EINVAL));
1348
1349         if (open_flags & MDS_FMODE_WRITE &&
1350             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1351                 GOTO(out, result = -EROFS);
1352
1353         CDEBUG(D_INODE, "I am going to open "DFID"/("DNAME"->"DFID") "
1354                "cr_flag=%#llo mode=0%06o msg_flag=0x%x\n",
1355                PFID(rr->rr_fid1), PNAME(&rr->rr_name), PFID(rr->rr_fid2),
1356                open_flags, ma->ma_attr.la_mode, msg_flags);
1357
1358         if (info->mti_cross_ref) {
1359                 /* This is cross-ref open */
1360                 mdt_set_disposition(info, ldlm_rep,
1361                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD |
1362                              DISP_LOOKUP_POS));
1363                 result = mdt_cross_open(info, rr->rr_fid2, rr->rr_fid1,
1364                                         ldlm_rep, open_flags);
1365                 GOTO(out, result);
1366         } else if (req_is_replay(req)) {
1367                 result = mdt_open_by_fid(info, ldlm_rep);
1368
1369                 if (result != -ENOENT)
1370                         GOTO(out, result);
1371
1372                 /* We didn't find the correct object, so we need to re-create it
1373                  * via a regular replay. */
1374                 if (!(open_flags & MDS_OPEN_CREAT)) {
1375                         DEBUG_REQ(D_ERROR, req,
1376                                   "OPEN & CREAT not in open replay/by_fid");
1377                         GOTO(out, result = -EFAULT);
1378                 }
1379                 CDEBUG(D_INFO, "No object(1), continue as regular open.\n");
1380         } else if (open_flags & MDS_OPEN_BY_FID) {
1381                 result = mdt_open_by_fid_lock(info, ldlm_rep, lhc);
1382                 if (result < 0)
1383                         CDEBUG(D_INFO, "no object for "DFID": %d\n",
1384                                PFID(rr->rr_fid2), result);
1385                 GOTO(out, result);
1386         }
1387
1388         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1389                 GOTO(out, result = err_serious(-ENOMEM));
1390
1391         mdt_set_disposition(info, ldlm_rep,
1392                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1393
1394         if (!lu_name_is_valid(&rr->rr_name))
1395                 GOTO(out, result = -EPROTO);
1396
1397         result = mdt_lock_root_xattr(info, mdt);
1398         if (result < 0)
1399                 GOTO(out, result);
1400
1401         lh = &info->mti_lh[MDT_LH_PARENT];
1402         mdt_lock_pdo_init(lh, (open_flags & MDS_OPEN_CREAT) ? LCK_PW : LCK_PR,
1403                           &rr->rr_name);
1404
1405         parent = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
1406         if (IS_ERR(parent))
1407                 GOTO(out, result = PTR_ERR(parent));
1408
1409         result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
1410         if (result != 0) {
1411                 mdt_object_put(info->mti_env, parent);
1412                 GOTO(out, result);
1413         }
1414
1415         /* get and check version of parent */
1416         result = mdt_version_get_check(info, parent, 0);
1417         if (result)
1418                 GOTO(out_parent, result);
1419
1420         fid_zero(child_fid);
1421
1422         result = -ENOENT;
1423         if ((open_flags & MDS_OPEN_VOLATILE) == 0)
1424                 result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1425                                     &rr->rr_name, child_fid, &info->mti_spec);
1426
1427         LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1428                  "looking for "DFID"/"DNAME", found FID = "DFID"\n",
1429                  PFID(mdt_object_fid(parent)), PNAME(&rr->rr_name),
1430                  PFID(child_fid));
1431
1432         if (result != 0 && result != -ENOENT)
1433                 GOTO(out_parent, result);
1434
1435         if (result == -ENOENT) {
1436                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1437                 if (!(open_flags & MDS_OPEN_CREAT))
1438                         GOTO(out_parent, result);
1439                 if (mdt_rdonly(req->rq_export))
1440                         GOTO(out_parent, result = -EROFS);
1441                 *child_fid = *info->mti_rr.rr_fid2;
1442                 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1443                          PFID(child_fid));
1444                 /* In the function below, .hs_keycmp resolves to
1445                  * lu_obj_hop_keycmp() */
1446                 /* coverity[overrun-buffer-val] */
1447                 child = mdt_object_new(info->mti_env, mdt, child_fid);
1448         } else {
1449                 /*
1450                  * Check for O_EXCL is moved to the mdt_finish_open(),
1451                  * we need to return FID back in that case.
1452                  */
1453                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1454                 child = mdt_object_find(info->mti_env, mdt, child_fid);
1455         }
1456         if (IS_ERR(child))
1457                 GOTO(out_parent, result = PTR_ERR(child));
1458
1459         /** check version of child  */
1460         rc = mdt_version_get_check(info, child, 1);
1461         if (rc)
1462                 GOTO(out_child, result = rc);
1463
1464         if (result == -ENOENT) {
1465                 /* Create under OBF and .lustre is not permitted */
1466                 if (!fid_is_md_operative(rr->rr_fid1) &&
1467                     (open_flags & MDS_OPEN_VOLATILE) == 0)
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                                  ktime_us_delta(ktime_get(), kstart));
1506         } else {
1507                 /*
1508                  * The object is on remote node, return its FID for remote open.
1509                  */
1510                 if (mdt_object_remote(child)) {
1511                         /*
1512                          * Check if this lock already was sent to client and
1513                          * this is resent case. For resent case do not take lock
1514                          * again, use what is already granted.
1515                          */
1516                         LASSERT(lhc != NULL);
1517
1518                         rc = mdt_check_resent_lock(info, child, lhc);
1519                         if (rc < 0) {
1520                                 GOTO(out_child, result = rc);
1521                         } else if (rc > 0) {
1522                                 mdt_lock_handle_init(lhc);
1523                                 mdt_lock_reg_init(lhc, LCK_PR);
1524
1525                                 rc = mdt_object_lock(info, child, lhc,
1526                                                      MDS_INODELOCK_LOOKUP);
1527                         }
1528                         repbody->mbo_fid1 = *mdt_object_fid(child);
1529                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1530                         if (rc != 0)
1531                                 result = rc;
1532                         else
1533                                 result = -MDT_EREMOTE_OPEN;
1534                         GOTO(out_child, result);
1535                 } else if (mdt_object_exists(child)) {
1536                         /* Check early for MDS_OPEN_DIRECTORY/O_DIRECTORY to
1537                          * avoid opening regular files from lfs getstripe
1538                          * since doing so breaks the leases used by lfs
1539                          * mirror. See LU-13693. */
1540                         if (open_flags & MDS_OPEN_DIRECTORY &&
1541                             S_ISREG(lu_object_attr(&child->mot_obj)))
1542                                 GOTO(out_child, result = -ENOTDIR);
1543
1544                         /* We have to get attr & LOV EA & HSM for this
1545                          * object. */
1546                         mdt_prep_ma_buf_from_rep(info, child, ma);
1547                         ma->ma_need |= MA_HSM;
1548                         result = mdt_attr_get_complex(info, child, ma);
1549                         if (result != 0)
1550                                 GOTO(out_child, result);
1551                 } else {
1552                         /* Object does not exist. Likely FS corruption. */
1553                         CERROR("%s: name '"DNAME"' present, but FID "
1554                                DFID" is invalid\n", mdt_obd_name(info->mti_mdt),
1555                                PNAME(&rr->rr_name), PFID(child_fid));
1556                         GOTO(out_child, result = -EIO);
1557                 }
1558         }
1559
1560         repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1561         repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1562
1563         rc = mdt_pack_secctx_in_reply(info, child);
1564         if (unlikely(rc))
1565                 GOTO(out_child, result = rc);
1566
1567         rc = mdt_pack_encctx_in_reply(info, child);
1568         if (unlikely(rc))
1569                 GOTO(out_child, result = rc);
1570
1571         rc = mdt_check_resent_lock(info, child, lhc);
1572         if (rc < 0) {
1573                 GOTO(out_child, result = rc);
1574         } else if (rc == 0) {
1575                 /* the open lock might already be gotten in
1576                  * ldlm_handle_enqueue() */
1577                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1578                 if (open_flags & MDS_OPEN_LOCK)
1579                         mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1580         } else {
1581                 /* get openlock if this isn't replay and client requested it */
1582                 if (!req_is_replay(req)) {
1583                         rc = mdt_object_open_lock(info, child, lhc, &ibits);
1584                         object_locked = 1;
1585                         if (rc != 0)
1586                                 GOTO(out_child_unlock, result = rc);
1587                         else if (open_flags & MDS_OPEN_LOCK)
1588                                 mdt_set_disposition(info, ldlm_rep,
1589                                                     DISP_OPEN_LOCK);
1590                 }
1591         }
1592         /* Try to open it now. */
1593         rc = mdt_finish_open(info, parent, child, open_flags,
1594                              created, ldlm_rep);
1595         if (rc) {
1596                 result = rc;
1597                 /* openlock will be released if mdt_finish_open() failed */
1598                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1599
1600                 if (created && (open_flags & MDS_OPEN_VOLATILE)) {
1601                         CERROR("%s: cannot open volatile file "DFID", orphan "
1602                                "file will be left in PENDING directory until "
1603                                "next reboot, rc = %d\n", mdt_obd_name(mdt),
1604                                PFID(mdt_object_fid(child)), rc);
1605                         GOTO(out_child_unlock, result);
1606                 }
1607
1608                 if (created) {
1609                         ma->ma_need = 0;
1610                         ma->ma_valid = 0;
1611                         rc = mdo_unlink(info->mti_env,
1612                                         mdt_object_child(parent),
1613                                         mdt_object_child(child),
1614                                         &rr->rr_name,
1615                                         &info->mti_attr, 0);
1616                         if (rc != 0)
1617                                 CERROR("%s: "DFID" cleanup of open: rc = %d\n",
1618                                        mdt_obd_name(info->mti_mdt),
1619                                        PFID(mdt_object_fid(child)), rc);
1620                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1621                 }
1622         }
1623
1624         mdt_counter_incr(req, LPROC_MDT_OPEN,
1625                          ktime_us_delta(ktime_get(), kstart));
1626
1627         EXIT;
1628 out_child_unlock:
1629         if (object_locked)
1630                 mdt_object_open_unlock(info, child, lhc, ibits, result);
1631 out_child:
1632         mdt_object_put(info->mti_env, child);
1633         if (result == 0)
1634                 mdt_pack_size2body(info, child_fid, &lhc->mlh_reg_lh);
1635 out_parent:
1636         mdt_object_unlock_put(info, parent, lh, result || !created);
1637 out:
1638         if (result)
1639                 lustre_msg_set_transno(req->rq_repmsg, 0);
1640         return result;
1641 }
1642
1643 /**
1644  * Create an orphan object use local root.
1645  */
1646 static struct mdt_object *mdt_orphan_open(struct mdt_thread_info *info,
1647                                           struct mdt_device *mdt,
1648                                           const struct lu_fid *fid,
1649                                           struct md_attr *attr, fmode_t fmode)
1650 {
1651         const struct lu_env *env = info->mti_env;
1652         struct md_op_spec *spec = &info->mti_spec;
1653         struct lu_fid *local_root_fid = &info->mti_tmp_fid1;
1654         struct mdt_object *obj = NULL;
1655         struct mdt_object *local_root;
1656         static const struct lu_name lname = {
1657                 .ln_name = "i_am_nobody",
1658                 .ln_namelen = sizeof("i_am_nobody") - 1,
1659         };
1660         struct lu_ucred *uc;
1661         cfs_cap_t uc_cap_save;
1662         int rc;
1663         ENTRY;
1664
1665         rc = dt_root_get(env, mdt->mdt_bottom, local_root_fid);
1666         if (rc != 0)
1667                 RETURN(ERR_PTR(rc));
1668
1669         local_root = mdt_object_find(env, mdt, local_root_fid);
1670         if (IS_ERR(local_root))
1671                 RETURN(local_root);
1672
1673         obj = mdt_object_new(env, mdt, fid);
1674         if (IS_ERR(obj))
1675                 GOTO(out, rc = PTR_ERR(obj));
1676
1677         spec->sp_cr_lookup = 0;
1678         spec->sp_feat = &dt_directory_features;
1679         spec->sp_cr_mode = MDL_MINMODE; /* no lock */
1680         spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
1681         if (attr->ma_valid & MA_LOV) {
1682                 spec->u.sp_ea.eadata = attr->ma_lmm;
1683                 spec->u.sp_ea.eadatalen = attr->ma_lmm_size;
1684                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1685         } else {
1686                 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
1687         }
1688
1689         uc = lu_ucred(env);
1690         uc_cap_save = uc->uc_cap;
1691         uc->uc_cap |= BIT(CFS_CAP_DAC_OVERRIDE);
1692         rc = mdo_create(env, mdt_object_child(local_root), &lname,
1693                         mdt_object_child(obj), spec, attr);
1694         uc->uc_cap = uc_cap_save;
1695         if (rc < 0) {
1696                 CERROR("%s: cannot create volatile file "DFID": rc = %d\n",
1697                        mdt_obd_name(mdt), PFID(fid), rc);
1698                 GOTO(out, rc);
1699         }
1700
1701         rc = mo_open(env, mdt_object_child(obj), MDS_OPEN_CREATED, spec);
1702         if (rc < 0)
1703                 CERROR("%s: cannot open volatile file "DFID", orphan "
1704                        "file will be left in PENDING directory until "
1705                        "next reboot, rc = %d\n", mdt_obd_name(mdt),
1706                        PFID(fid), rc);
1707         GOTO(out, rc);
1708
1709 out:
1710         if (rc < 0) {
1711                 if (!IS_ERR(obj))
1712                         mdt_object_put(env, obj);
1713                 obj = ERR_PTR(rc);
1714         }
1715         mdt_object_put(env, local_root);
1716         return obj;
1717 }
1718
1719 /* XXX Look into layout in MDT layer. */
1720 static inline int mdt_hsm_set_released(struct lov_mds_md *lmm)
1721 {
1722         struct lov_comp_md_v1   *comp_v1;
1723         struct lov_mds_md       *v1;
1724         __u32   off;
1725         int     i;
1726
1727         if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1_DEFINED)) {
1728                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1729
1730                 if (comp_v1->lcm_entry_count == 0)
1731                         return -EINVAL;
1732
1733                 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1734                         off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1735                         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
1736                         v1->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1737                 }
1738         } else {
1739                 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1740         }
1741         return 0;
1742 }
1743
1744 static inline int mdt_get_lmm_gen(struct lov_mds_md *lmm, __u32 *gen)
1745 {
1746         struct lov_comp_md_v1 *comp_v1;
1747
1748         if (le32_to_cpu(lmm->lmm_magic == LOV_MAGIC_COMP_V1)) {
1749                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1750                 *gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1751         } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1 ||
1752                    le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V3) {
1753                 *gen = le16_to_cpu(lmm->lmm_layout_gen);
1754         } else {
1755                 return -EINVAL;
1756         }
1757         return 0;
1758 }
1759
1760 static int mdt_hsm_release(struct mdt_thread_info *info, struct mdt_object *o,
1761                            struct md_attr *ma)
1762 {
1763         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LAYOUT];
1764         struct lu_ucred        *uc = mdt_ucred(info);
1765         struct close_data      *data;
1766         struct ldlm_lock       *lease;
1767         struct mdt_object      *orphan;
1768         struct md_attr         *orp_ma;
1769         struct lu_buf          *buf;
1770         cfs_cap_t               cap;
1771         bool                    lease_broken;
1772         int                     rc;
1773         int                     rc2;
1774         ENTRY;
1775
1776         if (mdt_rdonly(info->mti_exp))
1777                 RETURN(-EROFS);
1778
1779         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1780         if (data == NULL)
1781                 RETURN(-EPROTO);
1782
1783         lease = ldlm_handle2lock(&data->cd_handle);
1784         if (lease == NULL)
1785                 RETURN(-ESTALE);
1786
1787         /* try to hold open_sem so that nobody else can open the file */
1788         if (!down_write_trylock(&o->mot_open_sem)) {
1789                 ldlm_lock_cancel(lease);
1790                 GOTO(out_reprocess, rc = -EBUSY);
1791         }
1792
1793         /* Check if the lease open lease has already canceled */
1794         lock_res_and_lock(lease);
1795         lease_broken = ldlm_is_cancel(lease);
1796         unlock_res_and_lock(lease);
1797
1798         LDLM_DEBUG(lease, DFID " lease broken? %d",
1799                    PFID(mdt_object_fid(o)), lease_broken);
1800
1801         /* Cancel server side lease. Client side counterpart should
1802          * have been cancelled. It's okay to cancel it now as we've
1803          * held mot_open_sem. */
1804         ldlm_lock_cancel(lease);
1805
1806         if (lease_broken) /* don't perform release task */
1807                 GOTO(out_unlock, rc = -ESTALE);
1808
1809         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1810                 GOTO(out_unlock, rc = -EINVAL);
1811
1812         /* ma_need was set before but it seems fine to change it in order to
1813          * avoid modifying the one from RPC */
1814         ma->ma_need = MA_HSM;
1815         rc = mdt_attr_get_complex(info, o, ma);
1816         if (rc != 0)
1817                 GOTO(out_unlock, rc);
1818
1819         if (ma->ma_attr_flags & MDS_PCC_ATTACH) {
1820                 if (ma->ma_valid & MA_HSM) {
1821                         if (ma->ma_hsm.mh_flags & HS_RELEASED)
1822                                 GOTO(out_unlock, rc = -EALREADY);
1823
1824                         if (ma->ma_hsm.mh_arch_id != data->cd_archive_id)
1825                                 CDEBUG(D_CACHE,
1826                                        DFID" archive id diff: %llu:%u\n",
1827                                        PFID(mdt_object_fid(o)),
1828                                        ma->ma_hsm.mh_arch_id,
1829                                        data->cd_archive_id);
1830
1831                         if (!(ma->ma_hsm.mh_flags & HS_DIRTY) &&
1832                             ma->ma_hsm.mh_arch_ver == data->cd_data_version) {
1833                                 CDEBUG(D_CACHE,
1834                                        DFID" data version matches: packed=%llu "
1835                                        "and on-disk=%llu\n",
1836                                        PFID(mdt_object_fid(o)),
1837                                        data->cd_data_version,
1838                                        ma->ma_hsm.mh_arch_ver);
1839                                 ma->ma_hsm.mh_flags = HS_ARCHIVED | HS_EXISTS;
1840                         }
1841
1842                         if (ma->ma_hsm.mh_flags & HS_DIRTY)
1843                                 ma->ma_hsm.mh_flags = HS_ARCHIVED | HS_EXISTS;
1844                 } else {
1845                         /* Set up HSM attribte for PCC archived object */
1846                         BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1847                                      sizeof(info->mti_xattr_buf));
1848                         buf = &info->mti_buf;
1849                         buf->lb_buf = info->mti_xattr_buf;
1850                         buf->lb_len = sizeof(struct hsm_attrs);
1851                         memset(&ma->ma_hsm, 0, sizeof(ma->ma_hsm));
1852                         ma->ma_hsm.mh_flags = HS_ARCHIVED | HS_EXISTS;
1853                         ma->ma_hsm.mh_arch_id = data->cd_archive_id;
1854                         ma->ma_hsm.mh_arch_ver = data->cd_data_version;
1855                         lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1856
1857                         rc = mo_xattr_set(info->mti_env, mdt_object_child(o),
1858                                           buf, XATTR_NAME_HSM, 0);
1859                         if (rc)
1860                                 GOTO(out_unlock, rc);
1861                 }
1862         } else {
1863                 if (!mdt_hsm_release_allow(ma))
1864                         GOTO(out_unlock, rc = -EPERM);
1865
1866                 /* already released? */
1867                 if (ma->ma_hsm.mh_flags & HS_RELEASED)
1868                         GOTO(out_unlock, rc = 0);
1869
1870                 /* Compare on-disk and packed data_version */
1871                 if (data->cd_data_version != ma->ma_hsm.mh_arch_ver) {
1872                         CDEBUG(D_HSM, DFID" data_version mismatches: "
1873                                "packed=%llu and on-disk=%llu\n",
1874                                PFID(mdt_object_fid(o)),
1875                                data->cd_data_version,
1876                                ma->ma_hsm.mh_arch_ver);
1877                         GOTO(out_unlock, rc = -EPERM);
1878                 }
1879         }
1880
1881         ma->ma_valid = MA_INODE;
1882         ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_SIZE;
1883         rc = mo_attr_set(info->mti_env, mdt_object_child(o), ma);
1884         if (rc < 0)
1885                 GOTO(out_unlock, rc);
1886
1887         mutex_lock(&o->mot_som_mutex);
1888         rc2 = mdt_set_som(info, o, SOM_FL_STRICT, ma->ma_attr.la_size,
1889                            ma->ma_attr.la_blocks);
1890         mutex_unlock(&o->mot_som_mutex);
1891         if (rc2 < 0)
1892                 CDEBUG(D_INODE,
1893                        "%s: File "DFID" SOM update failed: rc = %d\n",
1894                        mdt_obd_name(info->mti_mdt),
1895                        PFID(mdt_object_fid(o)), rc2);
1896
1897
1898         ma->ma_need = MA_INODE | MA_LOV;
1899         rc = mdt_attr_get_complex(info, o, ma);
1900         if (rc < 0)
1901                 GOTO(out_unlock, rc);
1902
1903         if (!(ma->ma_valid & MA_LOV)) {
1904                 /* Even empty file are released */
1905                 memset(ma->ma_lmm, 0, sizeof(*ma->ma_lmm));
1906                 ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEFINED);
1907                 ma->ma_lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1908                 ma->ma_lmm->lmm_stripe_size = cpu_to_le32(LOV_MIN_STRIPE_SIZE);
1909                 ma->ma_lmm_size = sizeof(*ma->ma_lmm);
1910         } else {
1911                 /* Magic must be LOV_MAGIC_*_DEFINED or LOD will interpret
1912                  * ma_lmm as lov_user_md, then it will be confused by union of
1913                  * layout_gen and stripe_offset. */
1914                 if ((le32_to_cpu(ma->ma_lmm->lmm_magic) & LOV_MAGIC_MASK) ==
1915                     LOV_MAGIC_MAGIC)
1916                         ma->ma_lmm->lmm_magic |= cpu_to_le32(LOV_MAGIC_DEFINED);
1917                 else
1918                         GOTO(out_unlock, rc = -EINVAL);
1919         }
1920
1921         /* Set file as released. */
1922         rc = mdt_hsm_set_released(ma->ma_lmm);
1923         if (rc)
1924                 GOTO(out_unlock, rc);
1925
1926         orp_ma = &info->mti_u.hsm.attr;
1927         orp_ma->ma_attr.la_mode = S_IFREG | S_IWUSR;
1928         /* We use root ownership to bypass potential quota
1929          * restrictions on the user and group of the file. */
1930         orp_ma->ma_attr.la_uid = 0;
1931         orp_ma->ma_attr.la_gid = 0;
1932         orp_ma->ma_attr.la_valid = LA_MODE | LA_UID | LA_GID;
1933         orp_ma->ma_lmm = ma->ma_lmm;
1934         orp_ma->ma_lmm_size = ma->ma_lmm_size;
1935         orp_ma->ma_valid = MA_INODE | MA_LOV;
1936         orphan = mdt_orphan_open(info, info->mti_mdt, &data->cd_fid, orp_ma,
1937                                  MDS_FMODE_WRITE);
1938         if (IS_ERR(orphan)) {
1939                 CERROR("%s: cannot open orphan file "DFID": rc = %ld\n",
1940                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid),
1941                        PTR_ERR(orphan));
1942                 GOTO(out_unlock, rc = PTR_ERR(orphan));
1943         }
1944
1945         /* Set up HSM attribute for orphan object */
1946         BUILD_BUG_ON(sizeof(struct hsm_attrs) > sizeof(info->mti_xattr_buf));
1947         buf = &info->mti_buf;
1948         buf->lb_buf = info->mti_xattr_buf;
1949         buf->lb_len = sizeof(struct hsm_attrs);
1950         ma->ma_hsm.mh_flags |= HS_RELEASED;
1951         lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1952         ma->ma_hsm.mh_flags &= ~HS_RELEASED;
1953
1954         mdt_lock_reg_init(lh, LCK_EX);
1955         rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_LAYOUT |
1956                              MDS_INODELOCK_XATTR);
1957         if (rc != 0)
1958                 GOTO(out_close, rc);
1959
1960         /* The orphan has root ownership so we need to raise
1961          * CAP_FOWNER to set the HSM attributes. */
1962         cap = uc->uc_cap;
1963         uc->uc_cap |= MD_CAP_TO_MASK(CFS_CAP_FOWNER);
1964         rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
1965                           XATTR_NAME_HSM, 0);
1966         uc->uc_cap = cap;
1967         if (rc != 0)
1968                 GOTO(out_layout_lock, rc);
1969
1970         /* Swap layout with orphan objects. */
1971         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o),
1972                              mdt_object_child(orphan),
1973                              SWAP_LAYOUTS_MDS_HSM);
1974
1975         if (!rc && ma->ma_attr_flags & MDS_PCC_ATTACH) {
1976                 ma->ma_need = MA_LOV;
1977                 rc = mdt_attr_get_complex(info, o, ma);
1978         }
1979
1980         EXIT;
1981
1982 out_layout_lock:
1983         /* Release exclusive LL */
1984         mdt_object_unlock(info, o, lh, 1);
1985 out_close:
1986         /* Close orphan object anyway */
1987         rc2 = mo_close(info->mti_env, mdt_object_child(orphan), orp_ma,
1988                        MDS_FMODE_WRITE);
1989         if (rc2 < 0)
1990                 CERROR("%s: error closing volatile file "DFID": rc = %d\n",
1991                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid), rc2);
1992         LU_OBJECT_DEBUG(D_HSM, info->mti_env, &orphan->mot_obj,
1993                         "object closed");
1994         mdt_object_put(info->mti_env, orphan);
1995
1996 out_unlock:
1997         up_write(&o->mot_open_sem);
1998
1999         /* already released */
2000         if (rc == 0) {
2001                 struct mdt_body *repbody;
2002
2003                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2004                 LASSERT(repbody != NULL);
2005                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2006                 if (ma->ma_attr_flags & MDS_PCC_ATTACH) {
2007                         LASSERT(ma->ma_valid & MA_LOV);
2008                         rc = mdt_get_lmm_gen(ma->ma_lmm,
2009                                              &repbody->mbo_layout_gen);
2010                         if (!rc)
2011                                 repbody->mbo_valid |= OBD_MD_LAYOUT_VERSION;
2012                 }
2013         }
2014
2015 out_reprocess:
2016         ldlm_reprocess_all(lease->l_resource, lease);
2017         LDLM_LOCK_PUT(lease);
2018
2019         ma->ma_valid = 0;
2020         ma->ma_need = 0;
2021
2022         return rc;
2023 }
2024
2025 int mdt_close_handle_layouts(struct mdt_thread_info *info,
2026                              struct mdt_object *o, struct md_attr *ma)
2027 {
2028         struct mdt_lock_handle  *lh1 = &info->mti_lh[MDT_LH_NEW];
2029         struct mdt_lock_handle  *lh2 = &info->mti_lh[MDT_LH_OLD];
2030         struct close_data       *data;
2031         struct ldlm_lock        *lease;
2032         struct mdt_object       *o1 = o, *o2;
2033         bool                     lease_broken;
2034         bool                     swap_objects;
2035         int                      rc;
2036         ENTRY;
2037
2038         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2039                 RETURN(-EROFS);
2040
2041         if (!S_ISREG(lu_object_attr(&o1->mot_obj)))
2042                 RETURN(-EINVAL);
2043
2044         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2045         if (data == NULL)
2046                 RETURN(-EPROTO);
2047
2048         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
2049                 RETURN(-EINVAL);
2050
2051         rc = lu_fid_cmp(&data->cd_fid, mdt_object_fid(o));
2052         if (unlikely(rc == 0))
2053                 RETURN(-EINVAL);
2054
2055         /* Exchange o1 and o2, to enforce locking order */
2056         swap_objects = (rc < 0);
2057
2058         lease = ldlm_handle2lock(&data->cd_handle);
2059         if (lease == NULL)
2060                 RETURN(-ESTALE);
2061
2062         o2 = mdt_object_find(info->mti_env, info->mti_mdt, &data->cd_fid);
2063         if (IS_ERR(o2))
2064                 GOTO(out_lease, rc = PTR_ERR(o2));
2065
2066         if (!S_ISREG(lu_object_attr(&o2->mot_obj))) {
2067                 swap_objects = false; /* not swapped yet */
2068                 GOTO(out_obj, rc = -EINVAL);
2069         }
2070
2071         if (swap_objects)
2072                 swap(o1, o2);
2073
2074         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
2075                            MAY_WRITE);
2076         if (rc < 0)
2077                 GOTO(out_obj, rc);
2078
2079         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
2080                            MAY_WRITE);
2081         if (rc < 0)
2082                 GOTO(out_obj, rc);
2083
2084         /* try to hold open_sem so that nobody else can open the file */
2085         if (!down_write_trylock(&o->mot_open_sem)) {
2086                 ldlm_lock_cancel(lease);
2087                 GOTO(out_obj, rc = -EBUSY);
2088         }
2089
2090         /* Check if the lease open lease has already canceled */
2091         lock_res_and_lock(lease);
2092         lease_broken = ldlm_is_cancel(lease);
2093         unlock_res_and_lock(lease);
2094
2095         LDLM_DEBUG(lease, DFID " lease broken? %d",
2096                    PFID(mdt_object_fid(o)), lease_broken);
2097
2098         /* Cancel server side lease. Client side counterpart should
2099          * have been cancelled. It's okay to cancel it now as we've
2100          * held mot_open_sem. */
2101         ldlm_lock_cancel(lease);
2102
2103         if (lease_broken)
2104                 GOTO(out_unlock_sem, rc = -ESTALE);
2105
2106         mdt_lock_reg_init(lh1, LCK_EX);
2107         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
2108                              MDS_INODELOCK_XATTR);
2109         if (rc < 0)
2110                 GOTO(out_unlock_sem, rc);
2111
2112         mdt_lock_reg_init(lh2, LCK_EX);
2113         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
2114                              MDS_INODELOCK_XATTR);
2115         if (rc < 0)
2116                 GOTO(out_unlock1, rc);
2117
2118         /* Swap layout with orphan object */
2119         if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SWAP) {
2120                 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
2121                                      mdt_object_child(o2), 0);
2122         } else if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_MERGE ||
2123                    ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
2124                 struct lu_buf *buf = &info->mti_buf;
2125                 struct md_rejig_data mrd;
2126
2127                 mrd.mrd_obj = mdt_object_child(o == o1 ? o2 : o1);
2128                 if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT)
2129                         mrd.mrd_mirror_id = data->cd_mirror_id;
2130
2131                 buf->lb_len = sizeof(mrd);
2132                 buf->lb_buf = &mrd;
2133                 rc = mo_xattr_set(info->mti_env, mdt_object_child(o), buf,
2134                                   XATTR_LUSTRE_LOV,
2135                                   ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT ?
2136                                   LU_XATTR_SPLIT : LU_XATTR_MERGE);
2137                 if (rc == 0 && ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2138                         int rc2;
2139
2140                         mutex_lock(&o->mot_som_mutex);
2141                         rc2 = mdt_set_som(info, o, SOM_FL_STRICT,
2142                                           ma->ma_attr.la_size,
2143                                           ma->ma_attr.la_blocks);
2144                         mutex_unlock(&o->mot_som_mutex);
2145                         if (rc2 < 0)
2146                                 CERROR(DFID": Setting i_blocks error: %d, "
2147                                        "i_blocks will be reported wrongly and "
2148                                        "can only be fixed in next resync\n",
2149                                        PFID(mdt_object_fid(o)), rc2);
2150                 }
2151         }
2152         if (rc < 0)
2153                 GOTO(out_unlock2, rc);
2154
2155         EXIT;
2156
2157 out_unlock2:
2158         /* Release exclusive LL */
2159         mdt_object_unlock(info, o2, lh2, 1);
2160
2161 out_unlock1:
2162         mdt_object_unlock(info, o1, lh1, 1);
2163
2164 out_unlock_sem:
2165         up_write(&o->mot_open_sem);
2166
2167         /* already swapped */
2168         if (rc == 0) {
2169                 struct mdt_body *repbody;
2170
2171                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2172                 LASSERT(repbody != NULL);
2173                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2174         }
2175
2176 out_obj:
2177         mdt_object_put(info->mti_env, swap_objects ? o1 : o2);
2178
2179         ldlm_reprocess_all(lease->l_resource, lease);
2180
2181 out_lease:
2182         LDLM_LOCK_PUT(lease);
2183
2184         if (ma != NULL) {
2185                 ma->ma_valid = 0;
2186                 ma->ma_need = 0;
2187         }
2188
2189         return rc;
2190 }
2191
2192 static int mdt_close_resync_done(struct mdt_thread_info *info,
2193                                  struct mdt_object *o, struct md_attr *ma)
2194 {
2195         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LOCAL];
2196         struct close_data *data;
2197         struct ldlm_lock *lease;
2198         struct md_layout_change layout = { 0 };
2199         __u32 *resync_ids = NULL;
2200         size_t resync_count = 0;
2201         bool lease_broken;
2202         int rc;
2203
2204         ENTRY;
2205
2206         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2207                 RETURN(-EROFS);
2208
2209         if (!S_ISREG(lu_object_attr(&o->mot_obj)))
2210                 RETURN(-EINVAL);
2211
2212         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2213         if (data == NULL)
2214                 RETURN(-EPROTO);
2215
2216         if (ptlrpc_req_need_swab(mdt_info_req(info)))
2217                 lustre_swab_close_data_resync_done(&data->cd_resync);
2218
2219         if (!fid_is_zero(&data->cd_fid))
2220                 RETURN(-EPROTO);
2221
2222         lease = ldlm_handle2lock(&data->cd_handle);
2223         if (lease == NULL)
2224                 RETURN(-ESTALE);
2225
2226         /* try to hold open_sem so that nobody else can open the file */
2227         if (!down_write_trylock(&o->mot_open_sem)) {
2228                 ldlm_lock_cancel(lease);
2229                 GOTO(out_reprocess, rc = -EBUSY);
2230         }
2231
2232         /* Check if the lease open lease has already canceled */
2233         lock_res_and_lock(lease);
2234         lease_broken = ldlm_is_cancel(lease);
2235         unlock_res_and_lock(lease);
2236
2237         LDLM_DEBUG(lease, DFID " lease broken? %d\n",
2238                    PFID(mdt_object_fid(o)), lease_broken);
2239
2240         /* Cancel server side lease. Client side counterpart should
2241          * have been cancelled. It's okay to cancel it now as we've
2242          * held mot_open_sem. */
2243         ldlm_lock_cancel(lease);
2244
2245         if (lease_broken) /* don't perform release task */
2246                 GOTO(out_unlock, rc = -ESTALE);
2247
2248         resync_count = data->cd_resync.resync_count;
2249
2250         if (resync_count > INLINE_RESYNC_ARRAY_SIZE) {
2251                 void *data;
2252
2253                 if (!req_capsule_has_field(info->mti_pill, &RMF_U32,
2254                                            RCL_CLIENT))
2255                         GOTO(out_unlock, rc = -EPROTO);
2256
2257                 OBD_ALLOC_PTR_ARRAY(resync_ids, resync_count);
2258                 if (!resync_ids)
2259                         GOTO(out_unlock, rc = -ENOMEM);
2260
2261                 data = req_capsule_client_get(info->mti_pill, &RMF_U32);
2262                 memcpy(resync_ids, data, resync_count * sizeof(__u32));
2263
2264                 layout.mlc_resync_ids = resync_ids;
2265         } else {
2266                 layout.mlc_resync_ids = data->cd_resync.resync_ids_inline;
2267         }
2268
2269         layout.mlc_opc = MD_LAYOUT_RESYNC_DONE;
2270         layout.mlc_resync_count = resync_count;
2271         if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2272                 layout.mlc_som.lsa_valid = SOM_FL_STRICT;
2273                 layout.mlc_som.lsa_size = ma->ma_attr.la_size;
2274                 layout.mlc_som.lsa_blocks = ma->ma_attr.la_blocks;
2275         }
2276         rc = mdt_layout_change(info, o, lhc, &layout);
2277         if (rc)
2278                 GOTO(out_unlock, rc);
2279
2280         mdt_object_unlock(info, o, lhc, 0);
2281
2282         EXIT;
2283
2284 out_unlock:
2285         up_write(&o->mot_open_sem);
2286
2287         /* already released */
2288         if (rc == 0) {
2289                 struct mdt_body *repbody;
2290
2291                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2292                 LASSERT(repbody != NULL);
2293                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2294         }
2295
2296         if (resync_ids)
2297                 OBD_FREE_PTR_ARRAY(resync_ids, resync_count);
2298
2299 out_reprocess:
2300         ldlm_reprocess_all(lease->l_resource, lease);
2301         LDLM_LOCK_PUT(lease);
2302
2303         ma->ma_valid = 0;
2304         ma->ma_need = 0;
2305
2306         return rc;
2307 }
2308
2309 #define MFD_CLOSED(open_flags) ((open_flags) == MDS_FMODE_CLOSED)
2310
2311 static int mdt_mfd_closed(struct mdt_file_data *mfd)
2312 {
2313         return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_open_flags));
2314 }
2315
2316 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
2317 {
2318         struct mdt_object *o = mfd->mfd_object;
2319         struct md_object *next = mdt_object_child(o);
2320         struct md_attr *ma = &info->mti_attr;
2321         struct lu_fid *ofid = &info->mti_tmp_fid1;
2322         int rc = 0;
2323         u64 open_flags;
2324         u64 intent;
2325
2326         ENTRY;
2327
2328         open_flags = mfd->mfd_open_flags;
2329         intent = ma->ma_attr_flags & MDS_CLOSE_INTENT;
2330         *ofid = *mdt_object_fid(o);
2331
2332         /* the below message is checked in replay-single.sh test_46 */
2333         CDEBUG(D_INODE, "%s: %sclosing file handle "DFID" with intent: %llx\n",
2334                mdt_obd_name(info->mti_mdt),
2335                ma->ma_valid & MA_FORCE_LOG ? "force " : "", PFID(ofid), intent);
2336
2337         switch (intent) {
2338         case MDS_HSM_RELEASE: {
2339                 rc = mdt_hsm_release(info, o, ma);
2340                 if (rc < 0) {
2341                         CDEBUG(D_HSM, "%s: File " DFID " release failed: %d\n",
2342                                mdt_obd_name(info->mti_mdt),
2343                                PFID(ofid), rc);
2344                         /* continue to close even error occurred. */
2345                 }
2346                 break;
2347         }
2348         case MDS_CLOSE_LAYOUT_MERGE:
2349         case MDS_CLOSE_LAYOUT_SPLIT:
2350         case MDS_CLOSE_LAYOUT_SWAP: {
2351                 rc = mdt_close_handle_layouts(info, o, ma);
2352                 if (rc < 0) {
2353                         CDEBUG(D_INODE,
2354                                "%s: cannot swap layout of "DFID": rc = %d\n",
2355                                mdt_obd_name(info->mti_mdt),
2356                                PFID(ofid), rc);
2357                         /* continue to close even if error occurred. */
2358                 }
2359                 break;
2360         }
2361         case MDS_CLOSE_RESYNC_DONE:
2362                 rc = mdt_close_resync_done(info, o, ma);
2363                 break;
2364         default:
2365                 /* nothing */
2366                 break;
2367         }
2368
2369         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
2370             ma->ma_attr.la_valid & (LA_LSIZE | LA_LBLOCKS)) {
2371                 int rc2;
2372
2373                 rc2 = mdt_lsom_update(info, o, false);
2374                 if (rc2 < 0)
2375                         CDEBUG(D_INODE,
2376                                "%s: File " DFID " LSOM failed: rc = %d\n",
2377                                mdt_obd_name(info->mti_mdt),
2378                                PFID(ofid), rc2);
2379                         /* continue to close even if error occured. */
2380         }
2381
2382         if (open_flags & MDS_FMODE_WRITE)
2383                 mdt_write_put(o);
2384         else if (open_flags & MDS_FMODE_EXEC)
2385                 mdt_write_allow(o);
2386
2387         /* Update atime|mtime|ctime on close. */
2388         if ((open_flags & MDS_FMODE_EXEC || open_flags & MDS_FMODE_READ ||
2389              open_flags & MDS_FMODE_WRITE) && (ma->ma_valid & MA_INODE) &&
2390             (ma->ma_attr.la_valid & LA_ATIME ||
2391              ma->ma_attr.la_valid & LA_MTIME ||
2392              ma->ma_attr.la_valid & LA_CTIME)) {
2393                 ma->ma_valid = MA_INODE;
2394                 ma->ma_attr_flags |= MDS_CLOSE_UPDATE_TIMES;
2395                 ma->ma_attr.la_valid &= (LA_ATIME | LA_MTIME | LA_CTIME);
2396                 rc = mo_attr_set(info->mti_env, next, ma);
2397         }
2398
2399         /* If file data is modified, add the dirty flag. */
2400         if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
2401                 rc = mdt_add_dirty_flag(info, o, ma);
2402
2403         ma->ma_need |= MA_INODE;
2404         ma->ma_valid &= ~MA_INODE;
2405
2406         LASSERT(atomic_read(&o->mot_open_count) > 0);
2407         atomic_dec(&o->mot_open_count);
2408         mdt_handle_last_unlink(info, o, ma);
2409
2410         if (!MFD_CLOSED(open_flags)) {
2411                 rc = mo_close(info->mti_env, next, ma, open_flags);
2412                 if (mdt_dom_check_for_discard(info, o))
2413                         mdt_dom_discard_data(info, o);
2414         }
2415
2416         /* adjust open and lease count */
2417         if (open_flags & MDS_OPEN_LEASE) {
2418                 LASSERT(atomic_read(&o->mot_lease_count) > 0);
2419                 atomic_dec(&o->mot_lease_count);
2420         }
2421
2422         mdt_mfd_free(mfd);
2423         mdt_object_put(info->mti_env, o);
2424
2425         RETURN(rc);
2426 }
2427
2428 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
2429                        struct mdt_body *repbody)
2430 {
2431         struct mdt_export_data *med;
2432         struct mdt_file_data   *mfd;
2433         int                     ret = 0;
2434         int                     rc = 0;
2435         ENTRY;
2436
2437         med = &req->rq_export->exp_mdt_data;
2438         spin_lock(&med->med_open_lock);
2439         mfd = mdt_open_handle2mfd(med, &info->mti_open_handle,
2440                                   req_is_replay(req));
2441         if (mdt_mfd_closed(mfd)) {
2442                 spin_unlock(&med->med_open_lock);
2443                 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
2444                        ": cookie = %#llx\n", PFID(info->mti_rr.rr_fid1),
2445                        info->mti_open_handle.cookie);
2446                 /** not serious error since bug 3633 */
2447                 rc = -ESTALE;
2448         } else {
2449                 class_handle_unhash(&mfd->mfd_open_handle);
2450                 list_del_init(&mfd->mfd_list);
2451                 spin_unlock(&med->med_open_lock);
2452                 ret = mdt_mfd_close(info, mfd);
2453         }
2454
2455         RETURN(rc ? rc : ret);
2456 }
2457
2458 int mdt_close(struct tgt_session_info *tsi)
2459 {
2460         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2461         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2462         struct md_attr         *ma = &info->mti_attr;
2463         struct mdt_body        *repbody = NULL;
2464         ktime_t                 kstart = ktime_get();
2465         int rc, ret = 0;
2466         ENTRY;
2467
2468         /* Close may come with the Size-on-MDS update. Unpack it. */
2469         rc = mdt_close_unpack(info);
2470         if (rc)
2471                 GOTO(out, rc = err_serious(rc));
2472
2473         /* These fields are no longer used and are left for compatibility.
2474          * size is always zero */
2475         req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
2476                              0);
2477         req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
2478                              0);
2479         rc = req_capsule_server_pack(info->mti_pill);
2480         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
2481                 mdt_client_compatibility(info);
2482                 if (rc == 0)
2483                         mdt_fix_reply(info);
2484                 mdt_exit_ucred(info);
2485                 GOTO(out, rc = lustre_msg_get_status(req->rq_repmsg));
2486         }
2487
2488         /* Continue to close handle even if we can not pack reply */
2489         if (rc == 0) {
2490                 repbody = req_capsule_server_get(info->mti_pill,
2491                                                  &RMF_MDT_BODY);
2492                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
2493                                                     &RMF_MDT_MD);
2494                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
2495                                                        &RMF_MDT_MD,
2496                                                        RCL_SERVER);
2497                 ma->ma_need = MA_INODE | MA_LOV;
2498                 repbody->mbo_eadatasize = 0;
2499                 repbody->mbo_aclsize = 0;
2500         } else {
2501                 rc = err_serious(rc);
2502         }
2503
2504         rc = mdt_close_internal(info, req, repbody);
2505         if (rc != -ESTALE)
2506                 mdt_empty_transno(info, rc);
2507
2508         if (repbody != NULL) {
2509                 mdt_client_compatibility(info);
2510                 rc = mdt_fix_reply(info);
2511         }
2512
2513         mdt_exit_ucred(info);
2514         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
2515                 GOTO(out, rc = err_serious(-ENOMEM));
2516
2517         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
2518                                  OBD_FAIL_MDS_CLOSE_NET_REP))
2519                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
2520 out:
2521         mdt_thread_info_fini(info);
2522         if (rc == 0)
2523                 mdt_counter_incr(req, LPROC_MDT_CLOSE,
2524                                  ktime_us_delta(ktime_get(), kstart));
2525         RETURN(rc ? rc : ret);
2526 }