lustre/mdt/mdt_open.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_open.c
  37  *
  38  * Lustre Metadata Target (mdt) open/close file handling
  39  *
  40  * Author: Huang Hua <huanghua@clusterfs.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_MDS
  44
  45 #include <lustre_acl.h>
  46 #include <lustre_mds.h>
  47 #include "mdt_internal.h"
  48 #include <lustre_nodemap.h>
  49
  50 /* we do nothing because we do not have refcount now */
  51 static void mdt_mfd_get(void *mfdp)
  52 {
  53 }
  54
  55 static struct portals_handle_ops mfd_handle_ops = {
  56         .hop_addref = mdt_mfd_get,
  57         .hop_free   = NULL,
  58 };
  59
  60 /* Create a new mdt_file_data struct, initialize it,
  61  * and insert it to global hash table */
  62 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med)
  63 {
  64         struct mdt_file_data *mfd;
  65         ENTRY;
  66
  67         OBD_ALLOC_PTR(mfd);
  68         if (mfd != NULL) {
  69                 INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
  70                 mfd->mfd_handle.h_owner = med;
  71                 INIT_LIST_HEAD(&mfd->mfd_list);
  72                 class_handle_hash(&mfd->mfd_handle, &mfd_handle_ops);
  73         }
  74
  75         RETURN(mfd);
  76 }
  77
  78 /*
  79  * Find the mfd pointed to by handle in global hash table.
  80  * In case of replay the handle is obsoleted
  81  * but mfd can be found in mfd list by that handle.
  82  * Callers need to be holding med_open_lock.
  83  */
  84 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
  85                                      const struct lustre_handle *handle,
  86                                      bool is_replay_or_resent)
  87 {
  88         struct mdt_file_data   *mfd;
  89         ENTRY;
  90
  91         LASSERT(handle != NULL);
  92         mfd = class_handle2object(handle->cookie, med);
  93         /* during dw/setattr replay the mfd can be found by old handle */
  94         if (mfd == NULL && is_replay_or_resent) {
  95                 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
  96                         if (mfd->mfd_old_handle.cookie == handle->cookie)
  97                                 RETURN(mfd);
  98                 }
  99                 mfd = NULL;
 100         }
 101
 102         RETURN(mfd);
 103 }
 104
 105 /* free mfd */
 106 void mdt_mfd_free(struct mdt_file_data *mfd)
 107 {
 108         LASSERT(list_empty(&mfd->mfd_list));
 109         OBD_FREE_RCU(mfd, sizeof *mfd, &mfd->mfd_handle);
 110 }
 111
 112 static int mdt_create_data(struct mdt_thread_info *info,
 113                            struct mdt_object *p, struct mdt_object *o)
 114 {
 115         struct md_op_spec     *spec = &info->mti_spec;
 116         struct md_attr        *ma   = &info->mti_attr;
 117         int                    rc   = 0;
 118         ENTRY;
 119
 120         if (!md_should_create(spec->sp_cr_flags))
 121                 RETURN(0);
 122
 123         ma->ma_need = MA_INODE | MA_LOV;
 124         ma->ma_valid = 0;
 125         mutex_lock(&o->mot_lov_mutex);
 126         if (!o->mot_lov_created) {
 127                 rc = mdo_create_data(info->mti_env,
 128                                      p ? mdt_object_child(p) : NULL,
 129                                      mdt_object_child(o), spec, ma);
 130                 if (rc == 0)
 131                         rc = mdt_attr_get_complex(info, o, ma);
 132
 133                 if (rc == 0 && ma->ma_valid & MA_LOV)
 134                         o->mot_lov_created = 1;
 135         }
 136
 137         mutex_unlock(&o->mot_lov_mutex);
 138         RETURN(rc);
 139 }
 140
 141 int mdt_write_read(struct mdt_object *o)
 142 {
 143         int rc = 0;
 144         ENTRY;
 145         spin_lock(&o->mot_write_lock);
 146         rc = o->mot_write_count;
 147         spin_unlock(&o->mot_write_lock);
 148         RETURN(rc);
 149 }
 150
 151 int mdt_write_get(struct mdt_object *o)
 152 {
 153         int rc = 0;
 154         ENTRY;
 155         spin_lock(&o->mot_write_lock);
 156         if (o->mot_write_count < 0)
 157                 rc = -ETXTBSY;
 158         else
 159                 o->mot_write_count++;
 160         spin_unlock(&o->mot_write_lock);
 161
 162         RETURN(rc);
 163 }
 164
 165 void mdt_write_put(struct mdt_object *o)
 166 {
 167         ENTRY;
 168         spin_lock(&o->mot_write_lock);
 169         o->mot_write_count--;
 170         spin_unlock(&o->mot_write_lock);
 171         EXIT;
 172 }
 173
 174 static int mdt_write_deny(struct mdt_object *o)
 175 {
 176         int rc = 0;
 177         ENTRY;
 178         spin_lock(&o->mot_write_lock);
 179         if (o->mot_write_count > 0)
 180                 rc = -ETXTBSY;
 181         else
 182                 o->mot_write_count--;
 183         spin_unlock(&o->mot_write_lock);
 184         RETURN(rc);
 185 }
 186
 187 static void mdt_write_allow(struct mdt_object *o)
 188 {
 189         ENTRY;
 190         spin_lock(&o->mot_write_lock);
 191         o->mot_write_count++;
 192         spin_unlock(&o->mot_write_lock);
 193         EXIT;
 194 }
 195
 196 /* there can be no real transaction so prepare the fake one */
 197 static void mdt_empty_transno(struct mdt_thread_info *info, int rc)
 198 {
 199         struct mdt_device      *mdt = info->mti_mdt;
 200         struct ptlrpc_request  *req = mdt_info_req(info);
 201         struct tg_export_data  *ted;
 202         struct lsd_client_data *lcd;
 203
 204         ENTRY;
 205         /* transaction has occurred already */
 206         if (lustre_msg_get_transno(req->rq_repmsg) != 0)
 207                 RETURN_EXIT;
 208
 209         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 210                 struct thandle         *th;
 211
 212                 /* generate an empty transaction to get a transno
 213                  * and reply data */
 214                 th = dt_trans_create(info->mti_env, mdt->mdt_bottom);
 215                 if (!IS_ERR(th)) {
 216                         rc = dt_trans_start(info->mti_env, mdt->mdt_bottom, th);
 217                         dt_trans_stop(info->mti_env, mdt->mdt_bottom, th);
 218                 }
 219                 RETURN_EXIT;
 220         }
 221
 222         spin_lock(&mdt->mdt_lut.lut_translock);
 223         if (rc != 0) {
 224                 if (info->mti_transno != 0) {
 225                         struct obd_export *exp = req->rq_export;
 226
 227                         CERROR("%s: replay trans "LPU64" NID %s: rc = %d\n",
 228                                mdt_obd_name(mdt), info->mti_transno,
 229                                libcfs_nid2str(exp->exp_connection->c_peer.nid),
 230                                rc);
 231                         spin_unlock(&mdt->mdt_lut.lut_translock);
 232                         RETURN_EXIT;
 233                 }
 234         } else if (info->mti_transno == 0) {
 235                 info->mti_transno = ++mdt->mdt_lut.lut_last_transno;
 236         } else {
 237                 /* should be replay */
 238                 if (info->mti_transno > mdt->mdt_lut.lut_last_transno)
 239                         mdt->mdt_lut.lut_last_transno = info->mti_transno;
 240         }
 241         spin_unlock(&mdt->mdt_lut.lut_translock);
 242
 243         CDEBUG(D_INODE, "transno = "LPU64", last_committed = "LPU64"\n",
 244                info->mti_transno,
 245                req->rq_export->exp_obd->obd_last_committed);
 246
 247         req->rq_transno = info->mti_transno;
 248         lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
 249
 250         /* update lcd in memory only for resent cases */
 251         ted = &req->rq_export->exp_target_data;
 252         LASSERT(ted);
 253         mutex_lock(&ted->ted_lcd_lock);
 254         lcd = ted->ted_lcd;
 255         if (info->mti_transno < lcd->lcd_last_transno &&
 256             info->mti_transno != 0) {
 257                 /* This should happen during replay. Do not update
 258                  * last rcvd info if replay req transno < last transno,
 259                  * otherwise the following resend(after replay) can not
 260                  * be checked correctly by xid */
 261                 mutex_unlock(&ted->ted_lcd_lock);
 262                 CDEBUG(D_HA, "%s: transno = "LPU64" < last_transno = "LPU64"\n",
 263                        mdt_obd_name(mdt), info->mti_transno,
 264                        lcd->lcd_last_transno);
 265                 RETURN_EXIT;
 266         }
 267
 268         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
 269                 if (info->mti_transno != 0)
 270                         lcd->lcd_last_close_transno = info->mti_transno;
 271                 lcd->lcd_last_close_xid = req->rq_xid;
 272                 lcd->lcd_last_close_result = rc;
 273         } else {
 274                 /* VBR: save versions in last_rcvd for reconstruct. */
 275                 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
 276                 if (pre_versions) {
 277                         lcd->lcd_pre_versions[0] = pre_versions[0];
 278                         lcd->lcd_pre_versions[1] = pre_versions[1];
 279                         lcd->lcd_pre_versions[2] = pre_versions[2];
 280                         lcd->lcd_pre_versions[3] = pre_versions[3];
 281                 }
 282                 if (info->mti_transno != 0)
 283                         lcd->lcd_last_transno = info->mti_transno;
 284
 285                 lcd->lcd_last_xid = req->rq_xid;
 286                 lcd->lcd_last_result = rc;
 287                 lcd->lcd_last_data = info->mti_opdata;
 288         }
 289         mutex_unlock(&ted->ted_lcd_lock);
 290
 291         EXIT;
 292 }
 293
 294 void mdt_mfd_set_mode(struct mdt_file_data *mfd, __u64 mode)
 295 {
 296         LASSERT(mfd != NULL);
 297
 298         CDEBUG(D_HA, DFID " Change mfd mode "LPO64" -> "LPO64".\n",
 299                PFID(mdt_object_fid(mfd->mfd_object)), mfd->mfd_mode, mode);
 300
 301         mfd->mfd_mode = mode;
 302 }
 303
 304 /**
 305  * prep ma_lmm/ma_lmv for md_attr from reply
 306  */
 307 static void mdt_prep_ma_buf_from_rep(struct mdt_thread_info *info,
 308                                      struct mdt_object *obj,
 309                                      struct md_attr *ma)
 310 {
 311         LASSERT(ma->ma_lmv == NULL && ma->ma_lmm == NULL);
 312         if (S_ISDIR(obj->mot_header.loh_attr)) {
 313                 ma->ma_lmv = req_capsule_server_get(info->mti_pill,
 314                                                     &RMF_MDT_MD);
 315                 ma->ma_lmv_size = req_capsule_get_size(info->mti_pill,
 316                                                        &RMF_MDT_MD,
 317                                                        RCL_SERVER);
 318                 if (ma->ma_lmv_size > 0)
 319                         ma->ma_need |= MA_LMV;
 320         } else {
 321                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
 322                                                     &RMF_MDT_MD);
 323                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 324                                                        &RMF_MDT_MD,
 325                                                        RCL_SERVER);
 326                 if (ma->ma_lmm_size > 0)
 327                         ma->ma_need |= MA_LOV;
 328         }
 329 }
 330
 331 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
 332                         struct mdt_object *o, __u64 flags, int created,
 333                         struct ldlm_reply *rep)
 334 {
 335         struct ptlrpc_request   *req = mdt_info_req(info);
 336         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 337         struct mdt_file_data    *mfd;
 338         struct md_attr          *ma  = &info->mti_attr;
 339         struct lu_attr          *la  = &ma->ma_attr;
 340         struct mdt_body         *repbody;
 341         int                      rc = 0, isdir, isreg;
 342         ENTRY;
 343
 344         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 345
 346         isreg = S_ISREG(la->la_mode);
 347         isdir = S_ISDIR(la->la_mode);
 348         if (isreg && !(ma->ma_valid & MA_LOV) && !(flags & MDS_OPEN_RELEASE)) {
 349                 /*
 350                  * No EA, check whether it is will set regEA and dirEA since in
 351                  * above attr get, these size might be zero, so reset it, to
 352                  * retrieve the MD after create obj.
 353                  */
 354                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 355                                                        &RMF_MDT_MD,
 356                                                        RCL_SERVER);
 357                 /* in replay case, p == NULL */
 358                 rc = mdt_create_data(info, p, o);
 359                 if (rc)
 360                         RETURN(rc);
 361
 362                 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_DISP_STRIPE)
 363                         mdt_set_disposition(info, rep, DISP_OPEN_STRIPE);
 364         }
 365
 366         CDEBUG(D_INODE, "after open, ma_valid bit = "LPX64" lmm_size = %d\n",
 367                ma->ma_valid, ma->ma_lmm_size);
 368
 369         if (ma->ma_valid & MA_LOV) {
 370                 LASSERT(ma->ma_lmm_size != 0);
 371                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 372                 if (isdir)
 373                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 374                 else
 375                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 376         }
 377
 378         if (ma->ma_valid & MA_LMV) {
 379                 LASSERT(ma->ma_lmv_size != 0);
 380                 repbody->mbo_eadatasize = ma->ma_lmv_size;
 381                 LASSERT(isdir);
 382                 repbody->mbo_valid |= OBD_MD_FLDIREA | OBD_MD_MEA;
 383         }
 384
 385         if (flags & FMODE_WRITE)
 386                 rc = mdt_write_get(o);
 387         else if (flags & MDS_FMODE_EXEC)
 388                 rc = mdt_write_deny(o);
 389
 390         if (rc)
 391                 RETURN(rc);
 392
 393         rc = mo_open(info->mti_env, mdt_object_child(o),
 394                      created ? flags | MDS_OPEN_CREATED : flags);
 395         if (rc != 0) {
 396                 /* If we allow the client to chgrp (CFS_SETGRP_PERM), but the
 397                  * client does not know which suppgid should be sent to the MDS,
 398                  * or some other(s) changed the target file's GID after this RPC
 399                  * sent to the MDS with the suppgid as the original GID, then we
 400                  * give the client another chance to send the right suppgid. */
 401                 if (rc == -EACCES &&
 402                     allow_client_chgrp(info, lu_ucred(info->mti_env)))
 403                         mdt_set_disposition(info, rep, DISP_OPEN_DENY);
 404
 405                 GOTO(err_out, rc);
 406         }
 407
 408         mfd = mdt_mfd_new(med);
 409         if (mfd == NULL)
 410                 GOTO(err_out, rc = -ENOMEM);
 411
 412         /*
 413          * Keep a reference on this object for this open, and is
 414          * released by mdt_mfd_close().
 415          */
 416         mdt_object_get(info->mti_env, o);
 417         mfd->mfd_object = o;
 418         mfd->mfd_xid = req->rq_xid;
 419
 420         /*
 421          * @flags is always not zero. At least it should be FMODE_READ,
 422          * FMODE_WRITE or MDS_FMODE_EXEC.
 423          */
 424         LASSERT(flags != 0);
 425
 426         /* Open handling. */
 427         mdt_mfd_set_mode(mfd, flags);
 428
 429         atomic_inc(&o->mot_open_count);
 430         if (flags & MDS_OPEN_LEASE)
 431                 atomic_inc(&o->mot_lease_count);
 432
 433         /* replay handle */
 434         if (req_is_replay(req)) {
 435                 struct mdt_file_data *old_mfd;
 436                 /* Check wheather old cookie already exist in
 437                  * the list, becasue when do recovery, client
 438                  * might be disconnected from server, and
 439                  * restart replay, so there maybe some orphan
 440                  * mfd here, we should remove them */
 441                 LASSERT(info->mti_rr.rr_handle != NULL);
 442                 spin_lock(&med->med_open_lock);
 443                 old_mfd = mdt_handle2mfd(med, info->mti_rr.rr_handle, true);
 444                 if (old_mfd != NULL) {
 445                         CDEBUG(D_HA, "delete orphan mfd = %p, fid = "DFID", "
 446                                "cookie = "LPX64"\n", mfd,
 447                                PFID(mdt_object_fid(mfd->mfd_object)),
 448                                info->mti_rr.rr_handle->cookie);
 449                         class_handle_unhash(&old_mfd->mfd_handle);
 450                         list_del_init(&old_mfd->mfd_list);
 451                         spin_unlock(&med->med_open_lock);
 452                         /* no attr update for that close */
 453                         la->la_valid = 0;
 454                         ma->ma_valid |= MA_FLAGS;
 455                         ma->ma_attr_flags |= MDS_RECOV_OPEN;
 456                         mdt_mfd_close(info, old_mfd);
 457                         ma->ma_attr_flags &= ~MDS_RECOV_OPEN;
 458                         ma->ma_valid &= ~MA_FLAGS;
 459                 } else {
 460                         spin_unlock(&med->med_open_lock);
 461                         CDEBUG(D_HA, "orphan mfd not found, fid = "DFID", "
 462                                "cookie = "LPX64"\n",
 463                                PFID(mdt_object_fid(mfd->mfd_object)),
 464                                info->mti_rr.rr_handle->cookie);
 465                 }
 466
 467                 CDEBUG(D_HA, "Store old cookie "LPX64" in new mfd\n",
 468                        info->mti_rr.rr_handle->cookie);
 469
 470                 mfd->mfd_old_handle.cookie = info->mti_rr.rr_handle->cookie;
 471         }
 472
 473         repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
 474
 475         if (req->rq_export->exp_disconnected) {
 476                 spin_lock(&med->med_open_lock);
 477                 class_handle_unhash(&mfd->mfd_handle);
 478                 list_del_init(&mfd->mfd_list);
 479                 spin_unlock(&med->med_open_lock);
 480                 mdt_mfd_close(info, mfd);
 481         } else {
 482                 spin_lock(&med->med_open_lock);
 483                 list_add(&mfd->mfd_list, &med->med_open_head);
 484                 spin_unlock(&med->med_open_lock);
 485         }
 486
 487         mdt_empty_transno(info, rc);
 488
 489         RETURN(rc);
 490
 491 err_out:
 492         if (flags & FMODE_WRITE)
 493                 mdt_write_put(o);
 494         else if (flags & MDS_FMODE_EXEC)
 495                 mdt_write_allow(o);
 496
 497         return rc;
 498 }
 499
 500 static int mdt_finish_open(struct mdt_thread_info *info,
 501                            struct mdt_object *p, struct mdt_object *o,
 502                            __u64 flags, int created, struct ldlm_reply *rep)
 503 {
 504         struct ptlrpc_request   *req = mdt_info_req(info);
 505         struct obd_export       *exp = req->rq_export;
 506         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 507         struct md_attr          *ma  = &info->mti_attr;
 508         struct lu_attr          *la  = &ma->ma_attr;
 509         struct mdt_file_data    *mfd;
 510         struct mdt_body         *repbody;
 511         int                      rc = 0;
 512         int                      isreg, isdir, islnk;
 513         struct list_head        *t;
 514         ENTRY;
 515
 516         LASSERT(ma->ma_valid & MA_INODE);
 517
 518         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 519
 520         isreg = S_ISREG(la->la_mode);
 521         isdir = S_ISDIR(la->la_mode);
 522         islnk = S_ISLNK(la->la_mode);
 523         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 524
 525         /* LU-2275, simulate broken behaviour (esp. prevalent in
 526          * pre-2.4 servers where a very strange reply is sent on error
 527          * that looks like it was actually almost successful and a
 528          * failure at the same time.) */
 529         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_NEGATIVE_POSITIVE)) {
 530                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN |
 531                                                DISP_LOOKUP_NEG |
 532                                                DISP_LOOKUP_POS);
 533
 534                 if (flags & MDS_OPEN_LOCK)
 535                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
 536
 537                 RETURN(-ENOENT);
 538         }
 539
 540 #ifdef CONFIG_FS_POSIX_ACL
 541         if (exp_connect_flags(exp) & OBD_CONNECT_ACL) {
 542                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
 543                 if (IS_ERR(nodemap))
 544                         RETURN(PTR_ERR(nodemap));
 545
 546                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
 547                 nodemap_putref(nodemap);
 548                 if (rc)
 549                         RETURN(rc);
 550         }
 551 #endif
 552
 553         /*
 554          * If we are following a symlink, don't open; and do not return open
 555          * handle for special nodes as client required.
 556          */
 557         if (islnk || (!isreg && !isdir &&
 558             (exp_connect_flags(req->rq_export) & OBD_CONNECT_NODEVOH))) {
 559                 lustre_msg_set_transno(req->rq_repmsg, 0);
 560                 RETURN(0);
 561         }
 562
 563         /*
 564          * We need to return the existing object's fid back, so it is done here,
 565          * after preparing the reply.
 566          */
 567         if (!created && (flags & MDS_OPEN_EXCL) && (flags & MDS_OPEN_CREAT))
 568                 RETURN(-EEXIST);
 569
 570         /* This can't be done earlier, we need to return reply body */
 571         if (isdir) {
 572                 if (flags & (MDS_OPEN_CREAT | FMODE_WRITE)) {
 573                         /* We are trying to create or write an existing dir. */
 574                         RETURN(-EISDIR);
 575                 }
 576         } else if (flags & MDS_OPEN_DIRECTORY)
 577                 RETURN(-ENOTDIR);
 578
 579         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
 580                                  OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE)) {
 581                 RETURN(-EAGAIN);
 582         }
 583
 584         mfd = NULL;
 585         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 586                 spin_lock(&med->med_open_lock);
 587                 list_for_each(t, &med->med_open_head) {
 588                         mfd = list_entry(t, struct mdt_file_data, mfd_list);
 589                         if (mfd->mfd_xid == req->rq_xid)
 590                                 break;
 591                         mfd = NULL;
 592                 }
 593                 spin_unlock(&med->med_open_lock);
 594
 595                 if (mfd != NULL) {
 596                         repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
 597                         /* set repbody->ea_size for resent case */
 598                         if (ma->ma_valid & MA_LOV) {
 599                                 LASSERT(ma->ma_lmm_size != 0);
 600                                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 601                                 if (isdir)
 602                                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 603                                 else
 604                                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 605                         }
 606                         mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 607                         RETURN(0);
 608                 }
 609         }
 610
 611         rc = mdt_mfd_open(info, p, o, flags, created, rep);
 612         if (!rc)
 613                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 614
 615         RETURN(rc);
 616 }
 617
 618 void mdt_reconstruct_open(struct mdt_thread_info *info,
 619                           struct mdt_lock_handle *lhc)
 620 {
 621         const struct lu_env *env = info->mti_env;
 622         struct mdt_device       *mdt  = info->mti_mdt;
 623         struct req_capsule      *pill = info->mti_pill;
 624         struct ptlrpc_request   *req  = mdt_info_req(info);
 625         struct md_attr          *ma   = &info->mti_attr;
 626         struct mdt_reint_record *rr   = &info->mti_rr;
 627         __u64                   flags = info->mti_spec.sp_cr_flags;
 628         struct ldlm_reply       *ldlm_rep;
 629         struct mdt_object       *parent;
 630         struct mdt_object       *child;
 631         struct mdt_body         *repbody;
 632         int                      rc;
 633         __u64                    opdata;
 634         ENTRY;
 635
 636         LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
 637         ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
 638         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 639
 640         ma->ma_need = MA_INODE | MA_HSM;
 641         ma->ma_valid = 0;
 642
 643         opdata = mdt_req_from_lrd(req, info->mti_reply_data);
 644         mdt_set_disposition(info, ldlm_rep, opdata);
 645
 646         CDEBUG(D_INODE, "This is reconstruct open: disp="LPX64", result=%d\n",
 647                ldlm_rep->lock_policy_res1, req->rq_status);
 648
 649         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
 650             req->rq_status != 0)
 651                 /* We did not create successfully, return error to client. */
 652                 GOTO(out, rc = req->rq_status);
 653
 654         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
 655                 struct obd_export *exp = req->rq_export;
 656                 /*
 657                  * We failed after creation, but we do not know in which step
 658                  * we failed. So try to check the child object.
 659                  */
 660                 parent = mdt_object_find(env, mdt, rr->rr_fid1);
 661                 if (IS_ERR(parent)) {
 662                         rc = PTR_ERR(parent);
 663                         LCONSOLE_WARN("Parent "DFID" lookup error %d."
 664                                       " Evicting client %s with export %s.\n",
 665                                       PFID(rr->rr_fid1), rc,
 666                                       obd_uuid2str(&exp->exp_client_uuid),
 667                                       obd_export_nid2str(exp));
 668                         mdt_export_evict(exp);
 669                         RETURN_EXIT;
 670                 }
 671
 672                 child = mdt_object_find(env, mdt, rr->rr_fid2);
 673                 if (IS_ERR(child)) {
 674                         rc = PTR_ERR(child);
 675                         LCONSOLE_WARN("cannot lookup child "DFID": rc = %d; "
 676                                       "evicting client %s with export %s\n",
 677                                       PFID(rr->rr_fid2), rc,
 678                                       obd_uuid2str(&exp->exp_client_uuid),
 679                                       obd_export_nid2str(exp));
 680                         mdt_object_put(env, parent);
 681                         mdt_export_evict(exp);
 682                         RETURN_EXIT;
 683                 }
 684
 685                 if (unlikely(mdt_object_remote(child))) {
 686                         /* the child object was created on remote server */
 687                         if (!mdt_is_dne_client(exp)) {
 688                                 /* Return -EIO for old client */
 689                                 mdt_object_put(env, parent);
 690                                 mdt_object_put(env, child);
 691                                 GOTO(out, rc = -EIO);
 692                         }
 693                         repbody->mbo_fid1 = *rr->rr_fid2;
 694                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 695                         rc = 0;
 696                 } else {
 697                         if (mdt_object_exists(child)) {
 698                                 mdt_prep_ma_buf_from_rep(info, child, ma);
 699                                 rc = mdt_attr_get_complex(info, child, ma);
 700                                 if (rc == 0)
 701                                         rc = mdt_finish_open(info, parent,
 702                                                              child, flags,
 703                                                              1, ldlm_rep);
 704                         } else {
 705                                 /* the child does not exist, we should do
 706                                  * regular open */
 707                                 mdt_object_put(env, parent);
 708                                 mdt_object_put(env, child);
 709                                 GOTO(regular_open, 0);
 710                         }
 711                 }
 712                 mdt_object_put(env, parent);
 713                 mdt_object_put(env, child);
 714                 GOTO(out, rc);
 715         } else {
 716 regular_open:
 717                 /* We did not try to create, so we are a pure open */
 718                 rc = mdt_reint_open(info, lhc);
 719         }
 720
 721         EXIT;
 722 out:
 723         req->rq_status = rc;
 724         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
 725         LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
 726 }
 727
 728 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep)
 729 {
 730         __u64                    flags = info->mti_spec.sp_cr_flags;
 731         struct mdt_reint_record *rr = &info->mti_rr;
 732         struct md_attr          *ma = &info->mti_attr;
 733         struct mdt_object       *o;
 734         int                      rc;
 735         ENTRY;
 736
 737         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
 738         if (IS_ERR(o))
 739                 RETURN(rc = PTR_ERR(o));
 740
 741         if (unlikely(mdt_object_remote(o))) {
 742                 /* the child object was created on remote server */
 743                 struct mdt_body *repbody;
 744
 745                 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 746                                                 DISP_LOOKUP_EXECD |
 747                                                 DISP_LOOKUP_POS));
 748                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 749                 repbody->mbo_fid1 = *rr->rr_fid2;
 750                 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 751                 rc = 0;
 752         } else {
 753                 if (mdt_object_exists(o)) {
 754                         mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 755                                                         DISP_LOOKUP_EXECD |
 756                                                         DISP_LOOKUP_POS));
 757                         mdt_prep_ma_buf_from_rep(info, o, ma);
 758                         rc = mdt_attr_get_complex(info, o, ma);
 759                         if (rc == 0)
 760                                 rc = mdt_finish_open(info, NULL, o, flags, 0,
 761                                                      rep);
 762                 } else {
 763                         rc = -ENOENT;
 764                 }
 765         }
 766
 767         mdt_object_put(info->mti_env, o);
 768         RETURN(rc);
 769 }
 770
 771 /* lock object for open */
 772 static int mdt_object_open_lock(struct mdt_thread_info *info,
 773                                 struct mdt_object *obj,
 774                                 struct mdt_lock_handle *lhc,
 775                                 __u64 *ibits)
 776 {
 777         struct md_attr *ma = &info->mti_attr;
 778         __u64 open_flags = info->mti_spec.sp_cr_flags;
 779         enum ldlm_mode lm = LCK_CR;
 780         bool acq_lease = !!(open_flags & MDS_OPEN_LEASE);
 781         bool try_layout = false;
 782         bool create_layout = false;
 783         int rc = 0;
 784         ENTRY;
 785
 786         *ibits = 0;
 787         mdt_lock_handle_init(lhc);
 788
 789         if (req_is_replay(mdt_info_req(info)))
 790                 RETURN(0);
 791
 792         if (S_ISREG(lu_object_attr(&obj->mot_obj))) {
 793                 if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV) &&
 794                     md_should_create(open_flags))
 795                         create_layout = true;
 796                 if (exp_connect_layout(info->mti_exp) && !create_layout &&
 797                     ma->ma_need & MA_LOV)
 798                         try_layout = true;
 799         }
 800
 801         if (acq_lease) {
 802                 /* lease open, acquire write mode of open sem */
 803                 down_write(&obj->mot_open_sem);
 804
 805                 /* Lease exists and ask for new lease */
 806                 if (atomic_read(&obj->mot_lease_count) > 0) {
 807                         /* only exclusive open is supported, so lease
 808                          * are conflicted to each other */
 809                         GOTO(out, rc = -EBUSY);
 810                 }
 811
 812                 /* Lease must be with open lock */
 813                 if (!(open_flags & MDS_OPEN_LOCK)) {
 814                         CERROR("Request lease for file:"DFID ", but open lock "
 815                                 "is missed, open_flags = "LPO64".\n",
 816                                 PFID(mdt_object_fid(obj)), open_flags);
 817                         GOTO(out, rc = -EPROTO);
 818                 }
 819
 820                 /* XXX: only exclusive open is supported. */
 821                 lm = LCK_EX;
 822                 *ibits = MDS_INODELOCK_OPEN;
 823
 824                 /* never grant LCK_EX layout lock to client */
 825                 try_layout = false;
 826         } else { /* normal open */
 827                 /* normal open holds read mode of open sem */
 828                 down_read(&obj->mot_open_sem);
 829
 830                 if (open_flags & MDS_OPEN_LOCK) {
 831                         if (open_flags & FMODE_WRITE)
 832                                 lm = LCK_CW;
 833                         else if (open_flags & MDS_FMODE_EXEC)
 834                                 lm = LCK_PR;
 835                         else
 836                                 lm = LCK_CR;
 837
 838                         *ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN;
 839                 } else if (atomic_read(&obj->mot_lease_count) > 0) {
 840                         if (open_flags & FMODE_WRITE)
 841                                 lm = LCK_CW;
 842                         else
 843                                 lm = LCK_CR;
 844
 845                         /* revoke lease */
 846                         *ibits = MDS_INODELOCK_OPEN;
 847                         try_layout = false;
 848
 849                         lhc = &info->mti_lh[MDT_LH_LOCAL];
 850                 }
 851                 CDEBUG(D_INODE, "normal open:"DFID" lease count: %d, lm: %d\n",
 852                         PFID(mdt_object_fid(obj)),
 853                         atomic_read(&obj->mot_open_count), lm);
 854         }
 855
 856         mdt_lock_reg_init(lhc, lm);
 857
 858         /* one problem to return layout lock on open is that it may result
 859          * in too many layout locks cached on the client side. */
 860         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_OPEN) && try_layout) {
 861                 /* return lookup lock to validate inode at the client side,
 862                  * this is pretty important otherwise mdt will return layout
 863                  * lock for each open.
 864                  * However this is a double-edged sword because changing
 865                  * permission will revoke huge # of LOOKUP locks. */
 866                 *ibits |= MDS_INODELOCK_LAYOUT | MDS_INODELOCK_LOOKUP;
 867                 if (!mdt_object_lock_try(info, obj, lhc, *ibits)) {
 868                         *ibits &= ~(MDS_INODELOCK_LAYOUT|MDS_INODELOCK_LOOKUP);
 869                         if (*ibits != 0)
 870                                 rc = mdt_object_lock(info, obj, lhc, *ibits);
 871                 }
 872         } else if (*ibits != 0) {
 873                 rc = mdt_object_lock(info, obj, lhc, *ibits);
 874         }
 875
 876         CDEBUG(D_INODE, "Requested bits lock:"DFID ", ibits = "LPX64
 877                 ", open_flags = "LPO64", try_layout = %d, rc = %d\n",
 878                 PFID(mdt_object_fid(obj)), *ibits, open_flags, try_layout, rc);
 879
 880         /* will change layout, revoke layout locks by enqueuing EX lock. */
 881         if (rc == 0 && create_layout) {
 882                 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LAYOUT];
 883
 884                 CDEBUG(D_INODE, "Will create layout, get EX layout lock:"DFID
 885                         ", open_flags = "LPO64"\n",
 886                         PFID(mdt_object_fid(obj)), open_flags);
 887
 888                 /* We cannot enqueue another lock for the same resource we
 889                  * already have a lock for, due to mechanics of waiting list
 890                  * iterating in ldlm, see LU-3601.
 891                  * As such we'll drop the open lock we just got above here,
 892                  * it's ok not to have this open lock as it's main purpose is to
 893                  * flush unused cached client open handles. */
 894                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
 895                         mdt_object_unlock(info, obj, lhc, 1);
 896
 897                 LASSERT(!try_layout);
 898                 mdt_lock_handle_init(ll);
 899                 mdt_lock_reg_init(ll, LCK_EX);
 900                 rc = mdt_object_lock(info, obj, ll, MDS_INODELOCK_LAYOUT);
 901
 902                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LL_BLOCK, 2);
 903         }
 904
 905         /* Check if there is any other open handles after acquiring
 906          * open lock. At this point, caching open handles have been revoked
 907          * by open lock.
 908          * XXX: Now only exclusive open is supported. Need to check the
 909          * type of open for generic lease support. */
 910         if (rc == 0 && acq_lease) {
 911                 struct ptlrpc_request *req = mdt_info_req(info);
 912                 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
 913                 struct mdt_file_data *mfd;
 914                 bool is_replay_or_resent;
 915                 int open_count = 0;
 916
 917                 /* For lease: application can open a file and then apply lease,
 918                  * @handle contains original open handle in that case.
 919                  * In recovery, open REQ will be replayed and the lease REQ may
 920                  * be resent that means the open handle is already stale, so we
 921                  * need to fix it up here by finding new handle. */
 922                 is_replay_or_resent = req_is_replay(req) ||
 923                         lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
 924
 925                 /* if the request is _not_ a replay request, rr_handle
 926                  * may be used to hold an openhandle which is issuing the
 927                  * lease request, so that this openhandle doesn't count. */
 928                 mfd = mdt_handle2mfd(med, info->mti_rr.rr_handle,
 929                                      is_replay_or_resent);
 930                 if (mfd != NULL)
 931                         ++open_count;
 932
 933                 CDEBUG(D_INODE, "acq_lease "DFID": openers: %d, want: %d\n",
 934                         PFID(mdt_object_fid(obj)),
 935                         atomic_read(&obj->mot_open_count), open_count);
 936
 937                 if (atomic_read(&obj->mot_open_count) > open_count)
 938                         GOTO(out, rc = -EBUSY);
 939         }
 940         GOTO(out, rc);
 941
 942 out:
 943         RETURN(rc);
 944 }
 945
 946 static void mdt_object_open_unlock(struct mdt_thread_info *info,
 947                                    struct mdt_object *obj,
 948                                    struct mdt_lock_handle *lhc,
 949                                    __u64 ibits, int rc)
 950 {
 951         __u64 open_flags = info->mti_spec.sp_cr_flags;
 952         struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LOCAL];
 953         ENTRY;
 954
 955         if (req_is_replay(mdt_info_req(info)))
 956                 RETURN_EXIT;
 957
 958         /* Release local lock - the lock put in MDT_LH_LOCAL will never
 959          * return to client side. */
 960         if (lustre_handle_is_used(&ll->mlh_reg_lh))
 961                 mdt_object_unlock(info, obj, ll, 1);
 962
 963         ll = &info->mti_lh[MDT_LH_LAYOUT];
 964         /* Release local layout lock, layout was created */
 965         if (lustre_handle_is_used(&ll->mlh_reg_lh)) {
 966                 LASSERT(!(ibits & MDS_INODELOCK_LAYOUT));
 967                 mdt_object_unlock(info, obj, ll, 1);
 968         }
 969
 970         if (open_flags & MDS_OPEN_LEASE)
 971                 up_write(&obj->mot_open_sem);
 972         else
 973                 up_read(&obj->mot_open_sem);
 974
 975         /* Cross-ref case, the lock should be returned to the client */
 976         if (ibits == 0 || rc == -MDT_EREMOTE_OPEN)
 977                 RETURN_EXIT;
 978
 979         if (!(open_flags & MDS_OPEN_LOCK) && !(ibits & MDS_INODELOCK_LAYOUT)) {
 980                 /* for the open request, the lock will only return to client
 981                  * if open or layout lock is granted. */
 982                 rc = 1;
 983         }
 984
 985         if (rc != 0 || !lustre_handle_is_used(&lhc->mlh_reg_lh)) {
 986                 struct ldlm_reply       *ldlm_rep;
 987
 988                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
 989                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
 990                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
 991                         mdt_object_unlock(info, obj, lhc, 1);
 992         }
 993         RETURN_EXIT;
 994 }
 995
 996 /**
 997  * Check release is permitted for the current HSM flags.
 998  */
 999 static bool mdt_hsm_release_allow(const struct md_attr *ma)
1000 {
1001         if (!(ma->ma_valid & MA_HSM))
1002                 return false;
1003
1004         if (ma->ma_hsm.mh_flags & (HS_DIRTY|HS_NORELEASE|HS_LOST))
1005                 return false;
1006
1007         if (!(ma->ma_hsm.mh_flags & HS_ARCHIVED))
1008                 return false;
1009
1010         return true;
1011 }
1012
1013 static int mdt_open_by_fid_lock(struct mdt_thread_info *info,
1014                                 struct ldlm_reply *rep,
1015                                 struct mdt_lock_handle *lhc)
1016 {
1017         const struct lu_env     *env   = info->mti_env;
1018         struct mdt_device       *mdt   = info->mti_mdt;
1019         __u64                    flags = info->mti_spec.sp_cr_flags;
1020         struct mdt_reint_record *rr    = &info->mti_rr;
1021         struct md_attr          *ma    = &info->mti_attr;
1022         struct mdt_object       *parent= NULL;
1023         struct mdt_object       *o;
1024         int                      rc;
1025         bool                     object_locked = false;
1026         __u64                    ibits = 0;
1027         ENTRY;
1028
1029         if (md_should_create(flags) && !(flags & MDS_OPEN_HAS_EA)) {
1030                 if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1031                         parent = mdt_object_find(env, mdt, rr->rr_fid1);
1032                         if (IS_ERR(parent)) {
1033                                 CDEBUG(D_INODE, "Fail to find parent "DFID
1034                                        " for anonymous created %ld, try to"
1035                                        " use server-side parent.\n",
1036                                        PFID(rr->rr_fid1), PTR_ERR(parent));
1037                                 parent = NULL;
1038                         }
1039                 }
1040                 if (parent == NULL)
1041                         ma->ma_need |= MA_PFID;
1042         }
1043
1044         o = mdt_object_find(env, mdt, rr->rr_fid2);
1045         if (IS_ERR(o))
1046                 GOTO(out_parent_put, rc = PTR_ERR(o));
1047
1048         if (mdt_object_remote(o)) {
1049                 CDEBUG(D_INFO, "%s: "DFID" is on remote MDT.\n",
1050                        mdt_obd_name(info->mti_mdt),
1051                        PFID(rr->rr_fid2));
1052                 GOTO(out, rc = -EREMOTE);
1053         } else if (!mdt_object_exists(o)) {
1054                 mdt_set_disposition(info, rep,
1055                                     DISP_IT_EXECD |
1056                                     DISP_LOOKUP_EXECD |
1057                                     DISP_LOOKUP_NEG);
1058                 GOTO(out, rc = -ENOENT);
1059         }
1060
1061         mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1062
1063         mdt_prep_ma_buf_from_rep(info, o, ma);
1064         if (flags & MDS_OPEN_RELEASE)
1065                 ma->ma_need |= MA_HSM;
1066         rc = mdt_attr_get_complex(info, o, ma);
1067         if (rc)
1068                 GOTO(out, rc);
1069
1070         /* We should not change file's existing LOV EA */
1071         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
1072             flags & MDS_OPEN_HAS_EA && ma->ma_valid & MA_LOV)
1073                 GOTO(out, rc = -EEXIST);
1074
1075         /* If a release request, check file flags are fine and ask for an
1076          * exclusive open access. */
1077         if (flags & MDS_OPEN_RELEASE && !mdt_hsm_release_allow(ma))
1078                 GOTO(out, rc = -EPERM);
1079
1080         rc = mdt_check_resent_lock(info, o, lhc);
1081         if (rc < 0) {
1082                 GOTO(out, rc);
1083         } else if (rc > 0) {
1084                 rc = mdt_object_open_lock(info, o, lhc, &ibits);
1085                 object_locked = true;
1086                 if (rc)
1087                         GOTO(out_unlock, rc);
1088         }
1089
1090         if (ma->ma_valid & MA_PFID) {
1091                 parent = mdt_object_find(env, mdt, &ma->ma_pfid);
1092                 if (IS_ERR(parent)) {
1093                         CDEBUG(D_INODE, "Fail to find parent "DFID
1094                                " for anonymous created %ld, try to"
1095                                " use system default.\n",
1096                                PFID(&ma->ma_pfid), PTR_ERR(parent));
1097                         parent = NULL;
1098                 }
1099         }
1100
1101         rc = mdt_finish_open(info, parent, o, flags, 0, rep);
1102         if (!rc) {
1103                 mdt_set_disposition(info, rep, DISP_LOOKUP_POS);
1104                 if (flags & MDS_OPEN_LOCK)
1105                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1106                 if (flags & MDS_OPEN_LEASE)
1107                         mdt_set_disposition(info, rep, DISP_OPEN_LEASE);
1108         }
1109         GOTO(out_unlock, rc);
1110
1111 out_unlock:
1112         if (object_locked)
1113                 mdt_object_open_unlock(info, o, lhc, ibits, rc);
1114 out:
1115         mdt_object_put(env, o);
1116 out_parent_put:
1117         if (parent != NULL)
1118                 mdt_object_put(env, parent);
1119         return rc;
1120 }
1121
1122 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1123 static int mdt_cross_open(struct mdt_thread_info *info,
1124                           const struct lu_fid *parent_fid,
1125                           const struct lu_fid *fid,
1126                           struct ldlm_reply *rep, __u32 flags)
1127 {
1128         struct md_attr    *ma = &info->mti_attr;
1129         struct mdt_object *o;
1130         int                rc;
1131         ENTRY;
1132
1133         o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1134         if (IS_ERR(o))
1135                 RETURN(rc = PTR_ERR(o));
1136
1137         if (mdt_object_remote(o)) {
1138                 /* Something is wrong here, the object is on another MDS! */
1139                 CERROR("%s: "DFID" isn't on this server!: rc = %d\n",
1140                        mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1141                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1142                                 &o->mot_obj,
1143                                 "Object isn't on this server! FLD error?");
1144                 rc = -EFAULT;
1145         } else {
1146                 if (mdt_object_exists(o)) {
1147                         /* Do permission check for cross-open. */
1148                         rc = mo_permission(info->mti_env, NULL,
1149                                            mdt_object_child(o),
1150                                            NULL, flags | MDS_OPEN_CROSS);
1151                         if (rc)
1152                                 goto out;
1153
1154                         mdt_prep_ma_buf_from_rep(info, o, ma);
1155                         rc = mdt_attr_get_complex(info, o, ma);
1156                         if (rc != 0)
1157                                 GOTO(out, rc);
1158
1159                         rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
1160                 } else {
1161                         /*
1162                          * Something is wrong here. lookup was positive but
1163                          * there is no object!
1164                          */
1165                         CERROR("%s: "DFID" doesn't exist!: rc = %d\n",
1166                               mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1167                         rc = -EFAULT;
1168                 }
1169         }
1170 out:
1171         mdt_object_put(info->mti_env, o);
1172         RETURN(rc);
1173 }
1174
1175 /*
1176  * find root object and take its xattr lock if it's on remote MDT, later create
1177  * may use fs default striping (which is stored in root xattr).
1178  */
1179 static int mdt_lock_root_xattr(struct mdt_thread_info *info,
1180                                struct mdt_device *mdt)
1181 {
1182         struct mdt_object *md_root = mdt->mdt_md_root;
1183         struct lustre_handle lhroot;
1184         int rc;
1185
1186         if (md_root == NULL) {
1187                 lu_root_fid(&info->mti_tmp_fid1);
1188                 md_root = mdt_object_find(info->mti_env, mdt,
1189                                           &info->mti_tmp_fid1);
1190                 if (IS_ERR(md_root))
1191                         return PTR_ERR(md_root);
1192
1193                 spin_lock(&mdt->mdt_lock);
1194                 if (mdt->mdt_md_root != NULL)
1195                         mdt_object_put(info->mti_env, mdt->mdt_md_root);
1196                 mdt->mdt_md_root = md_root;
1197                 spin_unlock(&mdt->mdt_lock);
1198         }
1199
1200         if (md_root->mot_cache_attr || !mdt_object_remote(md_root))
1201                 return 0;
1202
1203         rc = mdt_remote_object_lock(info, md_root, mdt_object_fid(md_root),
1204                                     &lhroot, LCK_PR, MDS_INODELOCK_XATTR,
1205                                     false, true);
1206         if (rc < 0)
1207                 return rc;
1208
1209         md_root->mot_cache_attr = 1;
1210
1211         /* don't cancel this lock, so that we know the cached xattr is valid. */
1212         ldlm_lock_decref(&lhroot, LCK_PR);
1213
1214         return 0;
1215 }
1216
1217 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1218 {
1219         struct mdt_device       *mdt = info->mti_mdt;
1220         struct ptlrpc_request   *req = mdt_info_req(info);
1221         struct mdt_object       *parent;
1222         struct mdt_object       *child;
1223         struct mdt_lock_handle  *lh;
1224         struct ldlm_reply       *ldlm_rep;
1225         struct mdt_body         *repbody;
1226         struct lu_fid           *child_fid = &info->mti_tmp_fid1;
1227         struct md_attr          *ma = &info->mti_attr;
1228         __u64                    create_flags = info->mti_spec.sp_cr_flags;
1229         __u64                    ibits = 0;
1230         struct mdt_reint_record *rr = &info->mti_rr;
1231         int                      result, rc;
1232         int                      created = 0;
1233         int                      object_locked = 0;
1234         __u32                    msg_flags;
1235         ENTRY;
1236
1237         OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1238                                (obd_timeout + 1) / 4);
1239
1240         mdt_counter_incr(req, LPROC_MDT_OPEN);
1241         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1242
1243         ma->ma_need = MA_INODE;
1244         ma->ma_valid = 0;
1245
1246         LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1247         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1248
1249         if (unlikely(create_flags & MDS_OPEN_JOIN_FILE)) {
1250                 CERROR("file join is not supported anymore.\n");
1251                 GOTO(out, result = err_serious(-EOPNOTSUPP));
1252         }
1253         msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1254
1255         if ((create_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1256             info->mti_spec.u.sp_ea.eadata == NULL)
1257                 GOTO(out, result = err_serious(-EINVAL));
1258
1259         if (create_flags & FMODE_WRITE &&
1260             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1261                 GOTO(out, result = -EROFS);
1262
1263         CDEBUG(D_INODE, "I am going to open "DFID"/("DNAME"->"DFID") "
1264                "cr_flag="LPO64" mode=0%06o msg_flag=0x%x\n",
1265                PFID(rr->rr_fid1), PNAME(&rr->rr_name),
1266                PFID(rr->rr_fid2), create_flags,
1267                ma->ma_attr.la_mode, msg_flags);
1268
1269         if (info->mti_cross_ref) {
1270                 /* This is cross-ref open */
1271                 mdt_set_disposition(info, ldlm_rep,
1272                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD |
1273                              DISP_LOOKUP_POS));
1274                 result = mdt_cross_open(info, rr->rr_fid2, rr->rr_fid1,
1275                                         ldlm_rep, create_flags);
1276                 GOTO(out, result);
1277         } else if (req_is_replay(req) ||
1278             (req->rq_export->exp_libclient && create_flags & MDS_OPEN_HAS_EA)) {
1279                 /* This is a replay request or from liblustre with ea. */
1280                 result = mdt_open_by_fid(info, ldlm_rep);
1281
1282                 if (result != -ENOENT) {
1283                         if (req->rq_export->exp_libclient &&
1284                             create_flags & MDS_OPEN_HAS_EA)
1285                                 GOTO(out, result = 0);
1286                         GOTO(out, result);
1287                 }
1288                 /* We didn't find the correct object, so we need to re-create it
1289                  * via a regular replay. */
1290                 if (!(create_flags & MDS_OPEN_CREAT)) {
1291                         DEBUG_REQ(D_ERROR, req,
1292                                   "OPEN & CREAT not in open replay/by_fid.");
1293                         GOTO(out, result = -EFAULT);
1294                 }
1295                 CDEBUG(D_INFO, "No object(1), continue as regular open.\n");
1296         } else if (create_flags & (MDS_OPEN_BY_FID | MDS_OPEN_LOCK)) {
1297                 /*
1298                  * MDS_OPEN_LOCK is checked for backward compatibility with 2.1
1299                  * client.
1300                  */
1301                 result = mdt_open_by_fid_lock(info, ldlm_rep, lhc);
1302                 if (result < 0)
1303                         CDEBUG(D_INFO, "no object for "DFID": %d\n",
1304                                PFID(rr->rr_fid2), result);
1305                 GOTO(out, result);
1306         }
1307
1308         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1309                 GOTO(out, result = err_serious(-ENOMEM));
1310
1311         mdt_set_disposition(info, ldlm_rep,
1312                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1313
1314         if (!lu_name_is_valid(&rr->rr_name))
1315                 GOTO(out, result = -EPROTO);
1316
1317         result = mdt_lock_root_xattr(info, mdt);
1318         if (result < 0)
1319                 GOTO(out, result);
1320
1321 again:
1322         lh = &info->mti_lh[MDT_LH_PARENT];
1323         mdt_lock_pdo_init(lh,
1324                           (create_flags & MDS_OPEN_CREAT) ? LCK_PW : LCK_PR,
1325                           &rr->rr_name);
1326
1327         parent = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
1328         if (IS_ERR(parent))
1329                 GOTO(out, result = PTR_ERR(parent));
1330
1331         result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
1332         if (result != 0) {
1333                 mdt_object_put(info->mti_env, parent);
1334                 GOTO(out, result);
1335         }
1336
1337         /* get and check version of parent */
1338         result = mdt_version_get_check(info, parent, 0);
1339         if (result)
1340                 GOTO(out_parent, result);
1341
1342         fid_zero(child_fid);
1343
1344         result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1345                             &rr->rr_name, child_fid, &info->mti_spec);
1346
1347         LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1348                  "looking for "DFID"/"DNAME", found FID = "DFID"\n",
1349                  PFID(mdt_object_fid(parent)), PNAME(&rr->rr_name),
1350                  PFID(child_fid));
1351
1352         if (result != 0 && result != -ENOENT && result != -ESTALE)
1353                 GOTO(out_parent, result);
1354
1355         if (result == -ENOENT || result == -ESTALE) {
1356                 /* If the object is dead, let's check if the object
1357                  * is being migrated to a new object */
1358                 if (result == -ESTALE) {
1359                         struct lu_buf lmv_buf;
1360
1361                         lmv_buf.lb_buf = info->mti_xattr_buf;
1362                         lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
1363                         rc = mo_xattr_get(info->mti_env,
1364                                           mdt_object_child(parent),
1365                                           &lmv_buf, XATTR_NAME_LMV);
1366                         if (rc > 0) {
1367                                 struct lmv_mds_md_v1 *lmv;
1368
1369                                 lmv = lmv_buf.lb_buf;
1370                                 if (le32_to_cpu(lmv->lmv_hash_type) &
1371                                                 LMV_HASH_FLAG_MIGRATION) {
1372                                         /* Get the new parent FID and retry */
1373                                         mdt_object_unlock_put(info, parent,
1374                                                               lh, 1);
1375                                         mdt_lock_handle_init(lh);
1376                                         fid_le_to_cpu(
1377                                                 (struct lu_fid *)rr->rr_fid1,
1378                                                 &lmv->lmv_stripe_fids[1]);
1379                                         goto again;
1380                                 }
1381                         }
1382                 }
1383
1384                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1385                 if (result == -ESTALE) {
1386                         /*
1387                          * -ESTALE means the parent is a dead(unlinked) dir, so
1388                          * it should return -ENOENT to in accordance with the
1389                          * original mds implementaion.
1390                          */
1391                         GOTO(out_parent, result = -ENOENT);
1392                 }
1393
1394                 if (!(create_flags & MDS_OPEN_CREAT))
1395                         GOTO(out_parent, result);
1396                 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1397                         GOTO(out_parent, result = -EROFS);
1398                 *child_fid = *info->mti_rr.rr_fid2;
1399                 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1400                          PFID(child_fid));
1401                 /* In the function below, .hs_keycmp resolves to
1402                  * lu_obj_hop_keycmp() */
1403                 /* coverity[overrun-buffer-val] */
1404                 child = mdt_object_new(info->mti_env, mdt, child_fid);
1405         } else {
1406                 /*
1407                  * Check for O_EXCL is moved to the mdt_finish_open(), we need to
1408                  * return FID back in that case.
1409                  */
1410                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1411                 child = mdt_object_find(info->mti_env, mdt, child_fid);
1412         }
1413         if (IS_ERR(child))
1414                 GOTO(out_parent, result = PTR_ERR(child));
1415
1416         /** check version of child  */
1417         rc = mdt_version_get_check(info, child, 1);
1418         if (rc)
1419                 GOTO(out_child, result = rc);
1420
1421         if (result == -ENOENT) {
1422                 /* Create under OBF and .lustre is not permitted */
1423                 if (!fid_is_md_operative(rr->rr_fid1))
1424                         GOTO(out_child, result = -EPERM);
1425
1426                 /* save versions in reply */
1427                 mdt_version_get_save(info, parent, 0);
1428                 mdt_version_get_save(info, child, 1);
1429
1430                 /* version of child will be changed */
1431                 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
1432
1433                 /* Not found and with MDS_OPEN_CREAT: let's create it. */
1434                 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1435
1436                 /* Let lower layers know what is lock mode on directory. */
1437                 info->mti_spec.sp_cr_mode =
1438                         mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
1439
1440                 /*
1441                  * Do not perform lookup sanity check. We know that name does
1442                  * not exist.
1443                  */
1444                 info->mti_spec.sp_cr_lookup = 0;
1445                 info->mti_spec.sp_feat = &dt_directory_features;
1446
1447                 result = mdo_create(info->mti_env,
1448                                     mdt_object_child(parent),
1449                                     &rr->rr_name,
1450                                     mdt_object_child(child),
1451                                     &info->mti_spec,
1452                                     &info->mti_attr);
1453                 if (result == -ERESTART) {
1454                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1455                         GOTO(out_child, result);
1456                 } else {
1457                         mdt_prep_ma_buf_from_rep(info, child, ma);
1458                         /* XXX: we should call this once, see few lines below */
1459                         if (result == 0)
1460                                 result = mdt_attr_get_complex(info, child, ma);
1461
1462                         if (result != 0)
1463                                 GOTO(out_child, result);
1464                 }
1465                 created = 1;
1466                 mdt_counter_incr(req, LPROC_MDT_MKNOD);
1467         } else {
1468                 /*
1469                  * The object is on remote node, return its FID for remote open.
1470                  */
1471                 if (mdt_object_remote(child)) {
1472                         /*
1473                          * Check if this lock already was sent to client and
1474                          * this is resent case. For resent case do not take lock
1475                          * again, use what is already granted.
1476                          */
1477                         LASSERT(lhc != NULL);
1478
1479                         rc = mdt_check_resent_lock(info, child, lhc);
1480                         if (rc < 0) {
1481                                 GOTO(out_child, result = rc);
1482                         } else if (rc > 0) {
1483                                 mdt_lock_handle_init(lhc);
1484                                 mdt_lock_reg_init(lhc, LCK_PR);
1485
1486                                 rc = mdt_object_lock(info, child, lhc,
1487                                                      MDS_INODELOCK_LOOKUP);
1488                         }
1489                         repbody->mbo_fid1 = *mdt_object_fid(child);
1490                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1491                         if (rc != 0)
1492                                 result = rc;
1493                         else
1494                                 result = -MDT_EREMOTE_OPEN;
1495                         GOTO(out_child, result);
1496                 } else if (mdt_object_exists(child)) {
1497                         /* We have to get attr & LOV EA & HSM for this
1498                          * object. */
1499                         mdt_prep_ma_buf_from_rep(info, child, ma);
1500                         ma->ma_need |= MA_HSM;
1501                         result = mdt_attr_get_complex(info, child, ma);
1502                         if (result != 0)
1503                                 GOTO(out_child, result);
1504                 } else {
1505                         /* Object does not exist. Likely FS corruption. */
1506                         CERROR("%s: name '"DNAME"' present, but FID "
1507                                DFID" is invalid\n", mdt_obd_name(info->mti_mdt),
1508                                PNAME(&rr->rr_name), PFID(child_fid));
1509                         GOTO(out_child, result = -EIO);
1510                 }
1511         }
1512
1513         rc = mdt_check_resent_lock(info, child, lhc);
1514         if (rc < 0) {
1515                 GOTO(out_child, result = rc);
1516         } else if (rc == 0) {
1517                 /* the open lock might already be gotten in
1518                  * ldlm_handle_enqueue() */
1519                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1520                 if (create_flags & MDS_OPEN_LOCK)
1521                         mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1522         } else {
1523                 /* get openlock if this isn't replay and client requested it */
1524                 if (!req_is_replay(req)) {
1525                         rc = mdt_object_open_lock(info, child, lhc, &ibits);
1526                         object_locked = 1;
1527                         if (rc != 0)
1528                                 GOTO(out_child_unlock, result = rc);
1529                         else if (create_flags & MDS_OPEN_LOCK)
1530                                 mdt_set_disposition(info, ldlm_rep,
1531                                                     DISP_OPEN_LOCK);
1532                 }
1533         }
1534         /* Try to open it now. */
1535         rc = mdt_finish_open(info, parent, child, create_flags,
1536                              created, ldlm_rep);
1537         if (rc) {
1538                 result = rc;
1539                 /* openlock will be released if mdt_finish_open failed */
1540                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1541
1542                 if (created && create_flags & MDS_OPEN_VOLATILE) {
1543                         CERROR("%s: cannot open volatile file "DFID", orphan "
1544                                "file will be left in PENDING directory until "
1545                                "next reboot, rc = %d\n", mdt_obd_name(mdt),
1546                                PFID(mdt_object_fid(child)), rc);
1547                         GOTO(out_child_unlock, result);
1548                 }
1549
1550                 if (created) {
1551                         ma->ma_need = 0;
1552                         ma->ma_valid = 0;
1553                         rc = mdo_unlink(info->mti_env,
1554                                         mdt_object_child(parent),
1555                                         mdt_object_child(child),
1556                                         &rr->rr_name,
1557                                         &info->mti_attr, 0);
1558                         if (rc != 0)
1559                                 CERROR("%s: "DFID" cleanup of open: rc = %d\n",
1560                                        mdt_obd_name(info->mti_mdt),
1561                                        PFID(mdt_object_fid(child)), rc);
1562                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1563                 }
1564         }
1565         EXIT;
1566 out_child_unlock:
1567         if (object_locked)
1568                 mdt_object_open_unlock(info, child, lhc, ibits, result);
1569 out_child:
1570         mdt_object_put(info->mti_env, child);
1571 out_parent:
1572         mdt_object_unlock_put(info, parent, lh, result || !created);
1573 out:
1574         if (result)
1575                 lustre_msg_set_transno(req->rq_repmsg, 0);
1576         return result;
1577 }
1578
1579 /**
1580  * Create an orphan object use local root.
1581  */
1582 static struct mdt_object *mdt_orphan_open(struct mdt_thread_info *info,
1583                                           struct mdt_device *mdt,
1584                                           const struct lu_fid *fid,
1585                                           struct md_attr *attr, fmode_t fmode)
1586 {
1587         const struct lu_env *env = info->mti_env;
1588         struct md_op_spec *spec = &info->mti_spec;
1589         struct lu_fid *local_root_fid = &info->mti_tmp_fid1;
1590         struct mdt_object *obj = NULL;
1591         struct mdt_object *local_root;
1592         static const struct lu_name lname = {
1593                 .ln_name = "i_am_nobody",
1594                 .ln_namelen = sizeof("i_am_nobody") - 1,
1595         };
1596         struct lu_ucred *uc;
1597         cfs_cap_t uc_cap_save;
1598         int rc;
1599         ENTRY;
1600
1601         rc = dt_root_get(env, mdt->mdt_bottom, local_root_fid);
1602         if (rc != 0)
1603                 RETURN(ERR_PTR(rc));
1604
1605         local_root = mdt_object_find(env, mdt, local_root_fid);
1606         if (IS_ERR(local_root))
1607                 RETURN(local_root);
1608
1609         obj = mdt_object_new(env, mdt, fid);
1610         if (IS_ERR(obj))
1611                 GOTO(out, rc = PTR_ERR(obj));
1612
1613         spec->sp_cr_lookup = 0;
1614         spec->sp_feat = &dt_directory_features;
1615         spec->sp_cr_mode = MDL_MINMODE; /* no lock */
1616         spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
1617         if (attr->ma_valid & MA_LOV) {
1618                 spec->u.sp_ea.eadata = attr->ma_lmm;
1619                 spec->u.sp_ea.eadatalen = attr->ma_lmm_size;
1620                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1621         } else {
1622                 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
1623         }
1624
1625         uc = lu_ucred(env);
1626         uc_cap_save = uc->uc_cap;
1627         uc->uc_cap |= 1 << CFS_CAP_DAC_OVERRIDE;
1628         rc = mdo_create(env, mdt_object_child(local_root), &lname,
1629                         mdt_object_child(obj), spec, attr);
1630         uc->uc_cap = uc_cap_save;
1631         if (rc < 0) {
1632                 CERROR("%s: cannot create volatile file "DFID": rc = %d\n",
1633                        mdt_obd_name(mdt), PFID(fid), rc);
1634                 GOTO(out, rc);
1635         }
1636
1637         rc = mo_open(env, mdt_object_child(obj), MDS_OPEN_CREATED);
1638         if (rc < 0)
1639                 CERROR("%s: cannot open volatile file "DFID", orphan "
1640                        "file will be left in PENDING directory until "
1641                        "next reboot, rc = %d\n", mdt_obd_name(mdt),
1642                        PFID(fid), rc);
1643         GOTO(out, rc);
1644
1645 out:
1646         if (rc < 0) {
1647                 if (!IS_ERR(obj))
1648                         mdt_object_put(env, obj);
1649                 obj = ERR_PTR(rc);
1650         }
1651         mdt_object_put(env, local_root);
1652         return obj;
1653 }
1654
1655 static int mdt_hsm_release(struct mdt_thread_info *info, struct mdt_object *o,
1656                            struct md_attr *ma)
1657 {
1658         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LAYOUT];
1659         struct close_data      *data;
1660         struct ldlm_lock       *lease;
1661         struct mdt_object      *orphan;
1662         struct md_attr         *orp_ma;
1663         struct lu_buf          *buf;
1664         bool                    lease_broken;
1665         int                     rc;
1666         int                     rc2;
1667         ENTRY;
1668
1669         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
1670                 RETURN(-EROFS);
1671
1672         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1673         if (data == NULL)
1674                 RETURN(-EPROTO);
1675
1676         lease = ldlm_handle2lock(&data->cd_handle);
1677         if (lease == NULL)
1678                 RETURN(-ESTALE);
1679
1680         /* try to hold open_sem so that nobody else can open the file */
1681         if (!down_write_trylock(&o->mot_open_sem)) {
1682                 ldlm_lock_cancel(lease);
1683                 GOTO(out_reprocess, rc = -EBUSY);
1684         }
1685
1686         /* Check if the lease open lease has already canceled */
1687         lock_res_and_lock(lease);
1688         lease_broken = ldlm_is_cancel(lease);
1689         unlock_res_and_lock(lease);
1690
1691         LDLM_DEBUG(lease, DFID " lease broken? %d",
1692                    PFID(mdt_object_fid(o)), lease_broken);
1693
1694         /* Cancel server side lease. Client side counterpart should
1695          * have been cancelled. It's okay to cancel it now as we've
1696          * held mot_open_sem. */
1697         ldlm_lock_cancel(lease);
1698
1699         if (lease_broken) /* don't perform release task */
1700                 GOTO(out_unlock, rc = -ESTALE);
1701
1702         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1703                 GOTO(out_unlock, rc = -EINVAL);
1704
1705         /* ma_need was set before but it seems fine to change it in order to
1706          * avoid modifying the one from RPC */
1707         ma->ma_need = MA_HSM;
1708         rc = mdt_attr_get_complex(info, o, ma);
1709         if (rc != 0)
1710                 GOTO(out_unlock, rc);
1711
1712         if (!mdt_hsm_release_allow(ma))
1713                 GOTO(out_unlock, rc = -EPERM);
1714
1715         /* already released? */
1716         if (ma->ma_hsm.mh_flags & HS_RELEASED)
1717                 GOTO(out_unlock, rc = 0);
1718
1719         /* Compare on-disk and packed data_version */
1720         if (data->cd_data_version != ma->ma_hsm.mh_arch_ver) {
1721                 CDEBUG(D_HSM, DFID" data_version mismatches: packed="LPU64
1722                        " and on-disk="LPU64"\n", PFID(mdt_object_fid(o)),
1723                        data->cd_data_version, ma->ma_hsm.mh_arch_ver);
1724                 GOTO(out_unlock, rc = -EPERM);
1725         }
1726
1727         ma->ma_valid = MA_INODE;
1728         ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_SIZE;
1729         rc = mo_attr_set(info->mti_env, mdt_object_child(o), ma);
1730         if (rc < 0)
1731                 GOTO(out_unlock, rc);
1732
1733         ma->ma_need = MA_INODE | MA_LOV;
1734         rc = mdt_attr_get_complex(info, o, ma);
1735         if (rc < 0)
1736                 GOTO(out_unlock, rc);
1737
1738         if (!(ma->ma_valid & MA_LOV)) {
1739                 /* Even empty file are released */
1740                 memset(ma->ma_lmm, 0, sizeof(*ma->ma_lmm));
1741                 ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1742                 ma->ma_lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1743                 ma->ma_lmm->lmm_stripe_size = cpu_to_le32(LOV_MIN_STRIPE_SIZE);
1744                 ma->ma_lmm_size = sizeof(*ma->ma_lmm);
1745         } else {
1746                 /* Magic must be LOV_MAGIC_Vx_DEF otherwise LOD will interpret
1747                  * ma_lmm as lov_user_md, then it will be confused by union of
1748                  * layout_gen and stripe_offset. */
1749                 if (le32_to_cpu(ma->ma_lmm->lmm_magic) == LOV_MAGIC_V1)
1750                         ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1751                 else if (le32_to_cpu(ma->ma_lmm->lmm_magic) == LOV_MAGIC_V3)
1752                         ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1753                 else
1754                         GOTO(out_unlock, rc = -EINVAL);
1755         }
1756
1757         /* Set file as released */
1758         ma->ma_lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1759
1760         orp_ma = &info->mti_u.hsm.attr;
1761         orp_ma->ma_attr.la_mode = S_IFREG | S_IWUSR;
1762         orp_ma->ma_attr.la_uid = ma->ma_attr.la_uid;
1763         orp_ma->ma_attr.la_gid = ma->ma_attr.la_gid;
1764         orp_ma->ma_attr.la_valid = LA_MODE | LA_UID | LA_GID;
1765         orp_ma->ma_lmm = ma->ma_lmm;
1766         orp_ma->ma_lmm_size = ma->ma_lmm_size;
1767         orp_ma->ma_valid = MA_INODE | MA_LOV;
1768         orphan = mdt_orphan_open(info, info->mti_mdt, &data->cd_fid, orp_ma,
1769                                  FMODE_WRITE);
1770         if (IS_ERR(orphan)) {
1771                 CERROR("%s: cannot open orphan file "DFID": rc = %ld\n",
1772                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid),
1773                        PTR_ERR(orphan));
1774                 GOTO(out_unlock, rc = PTR_ERR(orphan));
1775         }
1776
1777         /* Set up HSM attribute for orphan object */
1778         CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
1779         buf = &info->mti_buf;
1780         buf->lb_buf = info->mti_xattr_buf;
1781         buf->lb_len = sizeof(struct hsm_attrs);
1782         ma->ma_hsm.mh_flags |= HS_RELEASED;
1783         lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1784         ma->ma_hsm.mh_flags &= ~HS_RELEASED;
1785
1786         mdt_lock_reg_init(lh, LCK_EX);
1787         rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_LAYOUT |
1788                              MDS_INODELOCK_XATTR);
1789         if (rc != 0)
1790                 GOTO(out_close, rc);
1791
1792         rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
1793                           XATTR_NAME_HSM, 0);
1794
1795         if (rc == 0)
1796                 /* Swap layout with orphan object */
1797                 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o),
1798                                      mdt_object_child(orphan),
1799                                      SWAP_LAYOUTS_MDS_HSM);
1800
1801         /* Release exclusive LL */
1802         mdt_object_unlock(info, o, lh, 1);
1803
1804         EXIT;
1805
1806 out_close:
1807         /* Close orphan object anyway */
1808         rc2 = mo_close(info->mti_env, mdt_object_child(orphan), orp_ma,
1809                        FMODE_WRITE);
1810         if (rc2 < 0)
1811                 CERROR("%s: error closing volatile file "DFID": rc = %d\n",
1812                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid), rc2);
1813         LU_OBJECT_DEBUG(D_HSM, info->mti_env, &orphan->mot_obj,
1814                         "object closed");
1815         mdt_object_put(info->mti_env, orphan);
1816
1817 out_unlock:
1818         up_write(&o->mot_open_sem);
1819
1820         /* already released */
1821         if (rc == 0) {
1822                 struct mdt_body *repbody;
1823
1824                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1825                 LASSERT(repbody != NULL);
1826                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
1827         }
1828
1829 out_reprocess:
1830         ldlm_reprocess_all(lease->l_resource);
1831         LDLM_LOCK_PUT(lease);
1832
1833         ma->ma_valid = 0;
1834         ma->ma_need = 0;
1835
1836         return rc;
1837 }
1838
1839 int mdt_close_swap_layouts(struct mdt_thread_info *info,
1840                            struct mdt_object *o, struct md_attr *ma)
1841 {
1842         struct mdt_lock_handle  *lh1 = &info->mti_lh[MDT_LH_NEW];
1843         struct mdt_lock_handle  *lh2 = &info->mti_lh[MDT_LH_OLD];
1844         struct close_data       *data;
1845         struct ldlm_lock        *lease;
1846         struct mdt_object       *o1 = o, *o2;
1847         bool                     lease_broken;
1848         bool                     swap_objects;
1849         int                      rc;
1850         ENTRY;
1851
1852         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
1853                 RETURN(-EROFS);
1854
1855         if (!S_ISREG(lu_object_attr(&o1->mot_obj)))
1856                 RETURN(-EINVAL);
1857
1858         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1859         if (data == NULL)
1860                 RETURN(-EPROTO);
1861
1862         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1863                 RETURN(-EINVAL);
1864
1865         rc = lu_fid_cmp(&data->cd_fid, mdt_object_fid(o));
1866         if (unlikely(rc == 0))
1867                 RETURN(-EINVAL);
1868
1869         /* Exchange o1 and o2, to enforce locking order */
1870         swap_objects = (rc < 0);
1871
1872         lease = ldlm_handle2lock(&data->cd_handle);
1873         if (lease == NULL)
1874                 RETURN(-ESTALE);
1875
1876         o2 = mdt_object_find(info->mti_env, info->mti_mdt, &data->cd_fid);
1877         if (IS_ERR(o2))
1878                 GOTO(out_lease, rc = PTR_ERR(o2));
1879
1880         if (!S_ISREG(lu_object_attr(&o2->mot_obj))) {
1881                 swap_objects = false; /* not swapped yet */
1882                 GOTO(out_obj, rc = -EINVAL);
1883         }
1884
1885         if (swap_objects)
1886                 swap(o1, o2);
1887
1888         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1889                            MAY_WRITE);
1890         if (rc < 0)
1891                 GOTO(out_obj, rc);
1892
1893         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1894                            MAY_WRITE);
1895         if (rc < 0)
1896                 GOTO(out_obj, rc);
1897
1898         /* try to hold open_sem so that nobody else can open the file */
1899         if (!down_write_trylock(&o->mot_open_sem)) {
1900                 ldlm_lock_cancel(lease);
1901                 GOTO(out_obj, rc = -EBUSY);
1902         }
1903
1904         /* Check if the lease open lease has already canceled */
1905         lock_res_and_lock(lease);
1906         lease_broken = ldlm_is_cancel(lease);
1907         unlock_res_and_lock(lease);
1908
1909         LDLM_DEBUG(lease, DFID " lease broken? %d",
1910                    PFID(mdt_object_fid(o)), lease_broken);
1911
1912         /* Cancel server side lease. Client side counterpart should
1913          * have been cancelled. It's okay to cancel it now as we've
1914          * held mot_open_sem. */
1915         ldlm_lock_cancel(lease);
1916
1917         if (lease_broken)
1918                 GOTO(out_unlock_sem, rc = -ESTALE);
1919
1920         mdt_lock_reg_init(lh1, LCK_EX);
1921         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1922                              MDS_INODELOCK_XATTR);
1923         if (rc < 0)
1924                 GOTO(out_unlock_sem, rc);
1925
1926         mdt_lock_reg_init(lh2, LCK_EX);
1927         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1928                              MDS_INODELOCK_XATTR);
1929         if (rc < 0)
1930                 GOTO(out_unlock1, rc);
1931
1932         /* Swap layout with orphan object */
1933         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1934                              mdt_object_child(o2), 0);
1935         if (rc < 0)
1936                 GOTO(out_unlock2, rc);
1937
1938         EXIT;
1939
1940 out_unlock2:
1941         /* Release exclusive LL */
1942         mdt_object_unlock(info, o2, lh2, 1);
1943
1944 out_unlock1:
1945         mdt_object_unlock(info, o1, lh1, 1);
1946
1947 out_unlock_sem:
1948         up_write(&o->mot_open_sem);
1949
1950         /* already swapped */
1951         if (rc == 0) {
1952                 struct mdt_body *repbody;
1953
1954                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1955                 LASSERT(repbody != NULL);
1956                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
1957         }
1958
1959 out_obj:
1960         mdt_object_put(info->mti_env, swap_objects ? o1 : o2);
1961
1962         ldlm_reprocess_all(lease->l_resource);
1963
1964 out_lease:
1965         LDLM_LOCK_PUT(lease);
1966
1967         if (ma != NULL) {
1968                 ma->ma_valid = 0;
1969                 ma->ma_need = 0;
1970         }
1971
1972         return rc;
1973 }
1974
1975 #define MFD_CLOSED(mode) ((mode) == MDS_FMODE_CLOSED)
1976 static int mdt_mfd_closed(struct mdt_file_data *mfd)
1977 {
1978         return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_mode));
1979 }
1980
1981 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
1982 {
1983         struct mdt_object *o = mfd->mfd_object;
1984         struct md_object *next = mdt_object_child(o);
1985         struct md_attr *ma = &info->mti_attr;
1986         int rc = 0;
1987         __u64 mode;
1988         ENTRY;
1989
1990         mode = mfd->mfd_mode;
1991
1992         if (ma->ma_attr_flags & MDS_HSM_RELEASE) {
1993                 rc = mdt_hsm_release(info, o, ma);
1994                 if (rc < 0) {
1995                         CDEBUG(D_HSM, "%s: File " DFID " release failed: %d\n",
1996                                mdt_obd_name(info->mti_mdt),
1997                                PFID(mdt_object_fid(o)), rc);
1998                         /* continue to close even error occurred. */
1999                 }
2000         }
2001
2002         if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SWAP) {
2003                 rc = mdt_close_swap_layouts(info, o, ma);
2004                 if (rc < 0) {
2005                         CDEBUG(D_INODE,
2006                                "%s: cannot swap layout of "DFID": rc=%d\n",
2007                                mdt_obd_name(info->mti_mdt),
2008                                PFID(mdt_object_fid(o)), rc);
2009                         /* continue to close even if error occurred. */
2010                 }
2011         }
2012
2013         if (mode & FMODE_WRITE)
2014                 mdt_write_put(o);
2015         else if (mode & MDS_FMODE_EXEC)
2016                 mdt_write_allow(o);
2017
2018         /* Update atime on close only. */
2019         if ((mode & MDS_FMODE_EXEC || mode & FMODE_READ || mode & FMODE_WRITE)
2020             && (ma->ma_valid & MA_INODE) && (ma->ma_attr.la_valid & LA_ATIME)) {
2021                 /* Set the atime only. */
2022                 ma->ma_valid = MA_INODE;
2023                 ma->ma_attr.la_valid = LA_ATIME;
2024                 rc = mo_attr_set(info->mti_env, next, ma);
2025         }
2026
2027         /* If file data is modified, add the dirty flag. */
2028         if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
2029                 rc = mdt_add_dirty_flag(info, o, ma);
2030
2031         ma->ma_need |= MA_INODE;
2032         ma->ma_valid &= ~MA_INODE;
2033
2034         if (!MFD_CLOSED(mode))
2035                 rc = mo_close(info->mti_env, next, ma, mode);
2036
2037         /* adjust open and lease count */
2038         if (mode & MDS_OPEN_LEASE) {
2039                 LASSERT(atomic_read(&o->mot_lease_count) > 0);
2040                 atomic_dec(&o->mot_lease_count);
2041         }
2042
2043         LASSERT(atomic_read(&o->mot_open_count) > 0);
2044         atomic_dec(&o->mot_open_count);
2045         mdt_mfd_free(mfd);
2046         mdt_object_put(info->mti_env, o);
2047
2048         RETURN(rc);
2049 }
2050
2051 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
2052                        struct mdt_body *repbody)
2053 {
2054         struct mdt_export_data *med;
2055         struct mdt_file_data   *mfd;
2056         struct mdt_object      *o;
2057         struct md_attr         *ma = &info->mti_attr;
2058         int                     ret = 0;
2059         int                     rc = 0;
2060         ENTRY;
2061
2062         med = &req->rq_export->exp_mdt_data;
2063         spin_lock(&med->med_open_lock);
2064         mfd = mdt_handle2mfd(med, &info->mti_close_handle, req_is_replay(req));
2065         if (mdt_mfd_closed(mfd)) {
2066                 spin_unlock(&med->med_open_lock);
2067                 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
2068                        ": cookie = "LPX64"\n", PFID(info->mti_rr.rr_fid1),
2069                        info->mti_close_handle.cookie);
2070                 /** not serious error since bug 3633 */
2071                 rc = -ESTALE;
2072         } else {
2073                 class_handle_unhash(&mfd->mfd_handle);
2074                 list_del_init(&mfd->mfd_list);
2075                 spin_unlock(&med->med_open_lock);
2076
2077                 /* Do not lose object before last unlink. */
2078                 o = mfd->mfd_object;
2079                 mdt_object_get(info->mti_env, o);
2080                 ret = mdt_mfd_close(info, mfd);
2081                 if (repbody != NULL)
2082                         rc = mdt_handle_last_unlink(info, o, ma);
2083                 mdt_object_put(info->mti_env, o);
2084         }
2085
2086         RETURN(rc ? rc : ret);
2087 }
2088
2089 int mdt_close(struct tgt_session_info *tsi)
2090 {
2091         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2092         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2093         struct md_attr         *ma = &info->mti_attr;
2094         struct mdt_body        *repbody = NULL;
2095         int rc, ret = 0;
2096         ENTRY;
2097
2098         mdt_counter_incr(req, LPROC_MDT_CLOSE);
2099         /* Close may come with the Size-on-MDS update. Unpack it. */
2100         rc = mdt_close_unpack(info);
2101         if (rc)
2102                 GOTO(out, rc = err_serious(rc));
2103
2104         /* These fields are no longer used and are left for compatibility.
2105          * size is always zero */
2106         req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
2107                              0);
2108         req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
2109                              0);
2110         rc = req_capsule_server_pack(info->mti_pill);
2111         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
2112                 mdt_client_compatibility(info);
2113                 if (rc == 0)
2114                         mdt_fix_reply(info);
2115                 mdt_exit_ucred(info);
2116                 GOTO(out, rc = lustre_msg_get_status(req->rq_repmsg));
2117         }
2118
2119         /* Continue to close handle even if we can not pack reply */
2120         if (rc == 0) {
2121                 repbody = req_capsule_server_get(info->mti_pill,
2122                                                  &RMF_MDT_BODY);
2123                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
2124                                                     &RMF_MDT_MD);
2125                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
2126                                                        &RMF_MDT_MD,
2127                                                        RCL_SERVER);
2128                 ma->ma_need = MA_INODE | MA_LOV;
2129                 repbody->mbo_eadatasize = 0;
2130                 repbody->mbo_aclsize = 0;
2131         } else {
2132                 rc = err_serious(rc);
2133         }
2134
2135         rc = mdt_close_internal(info, req, repbody);
2136         if (rc != -ESTALE)
2137                 mdt_empty_transno(info, rc);
2138
2139         if (repbody != NULL) {
2140                 mdt_client_compatibility(info);
2141                 rc = mdt_fix_reply(info);
2142         }
2143
2144         mdt_exit_ucred(info);
2145         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
2146                 GOTO(out, rc = err_serious(-ENOMEM));
2147
2148         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
2149                                  OBD_FAIL_MDS_CLOSE_NET_REP))
2150                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
2151 out:
2152         mdt_thread_info_fini(info);
2153         RETURN(rc ? rc : ret);
2154 }