lustre/mdt/mdt_open.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_open.c
  33  *
  34  * Lustre Metadata Target (mdt) open/close file handling
  35  *
  36  * Author: Huang Hua <huanghua@clusterfs.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include <lustre_acl.h>
  42 #include <lustre_mds.h>
  43 #include <lustre_swab.h>
  44 #include "mdt_internal.h"
  45 #include <lustre_nodemap.h>
  46
  47 /* we do nothing because we do not have refcount now */
  48 static void mdt_mfd_get(void *mfdp)
  49 {
  50 }
  51
  52 static struct portals_handle_ops mfd_open_handle_ops = {
  53         .hop_addref = mdt_mfd_get,
  54         .hop_free   = NULL,
  55 };
  56
  57 /* Create a new mdt_file_data struct, initialize it,
  58  * and insert it to global hash table */
  59 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med)
  60 {
  61         struct mdt_file_data *mfd;
  62         ENTRY;
  63
  64         OBD_ALLOC_PTR(mfd);
  65         if (mfd != NULL) {
  66                 INIT_LIST_HEAD_RCU(&mfd->mfd_open_handle.h_link);
  67                 mfd->mfd_open_handle.h_owner = med;
  68                 INIT_LIST_HEAD(&mfd->mfd_list);
  69                 class_handle_hash(&mfd->mfd_open_handle, &mfd_open_handle_ops);
  70         }
  71
  72         RETURN(mfd);
  73 }
  74
  75 /*
  76  * Find the mfd pointed to by handle in global hash table.
  77  * In case of replay the handle is obsoleted
  78  * but mfd can be found in mfd list by that handle.
  79  * Callers need to be holding med_open_lock.
  80  */
  81 struct mdt_file_data *mdt_open_handle2mfd(struct mdt_export_data *med,
  82                                         const struct lustre_handle *open_handle,
  83                                         bool is_replay_or_resent)
  84 {
  85         struct mdt_file_data   *mfd;
  86         ENTRY;
  87
  88         LASSERT(open_handle != NULL);
  89         mfd = class_handle2object(open_handle->cookie, med);
  90         /* during dw/setattr replay the mfd can be found by old handle */
  91         if (mfd == NULL && is_replay_or_resent) {
  92                 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
  93                         if (mfd->mfd_open_handle_old.cookie ==
  94                             open_handle->cookie)
  95                                 RETURN(mfd);
  96                 }
  97                 mfd = NULL;
  98         }
  99
 100         RETURN(mfd);
 101 }
 102
 103 /* free mfd */
 104 void mdt_mfd_free(struct mdt_file_data *mfd)
 105 {
 106         LASSERT(list_empty(&mfd->mfd_list));
 107         OBD_FREE_RCU(mfd, sizeof *mfd, &mfd->mfd_open_handle);
 108 }
 109
 110 static int mdt_create_data(struct mdt_thread_info *info,
 111                            struct mdt_object *p, struct mdt_object *o)
 112 {
 113         struct md_op_spec     *spec = &info->mti_spec;
 114         struct md_attr        *ma   = &info->mti_attr;
 115         int                    rc   = 0;
 116         ENTRY;
 117
 118         if (!md_should_create(spec->sp_cr_flags))
 119                 RETURN(0);
 120
 121         ma->ma_need = MA_INODE | MA_LOV;
 122         ma->ma_valid = 0;
 123         mutex_lock(&o->mot_lov_mutex);
 124         if (!o->mot_lov_created) {
 125                 rc = mdo_create_data(info->mti_env,
 126                                      p ? mdt_object_child(p) : NULL,
 127                                      mdt_object_child(o), spec, ma);
 128                 if (rc == 0)
 129                         rc = mdt_attr_get_complex(info, o, ma);
 130
 131                 if (rc == 0 && ma->ma_valid & MA_LOV)
 132                         o->mot_lov_created = 1;
 133         }
 134
 135         mutex_unlock(&o->mot_lov_mutex);
 136         RETURN(rc);
 137 }
 138
 139 int mdt_write_read(struct mdt_object *o)
 140 {
 141         int rc = 0;
 142         ENTRY;
 143         spin_lock(&o->mot_write_lock);
 144         rc = o->mot_write_count;
 145         spin_unlock(&o->mot_write_lock);
 146         RETURN(rc);
 147 }
 148
 149 int mdt_write_get(struct mdt_object *o)
 150 {
 151         int rc = 0;
 152         ENTRY;
 153         spin_lock(&o->mot_write_lock);
 154         if (o->mot_write_count < 0)
 155                 rc = -ETXTBSY;
 156         else
 157                 o->mot_write_count++;
 158         spin_unlock(&o->mot_write_lock);
 159
 160         RETURN(rc);
 161 }
 162
 163 void mdt_write_put(struct mdt_object *o)
 164 {
 165         ENTRY;
 166         spin_lock(&o->mot_write_lock);
 167         o->mot_write_count--;
 168         spin_unlock(&o->mot_write_lock);
 169         EXIT;
 170 }
 171
 172 static int mdt_write_deny(struct mdt_object *o)
 173 {
 174         int rc = 0;
 175         ENTRY;
 176         spin_lock(&o->mot_write_lock);
 177         if (o->mot_write_count > 0)
 178                 rc = -ETXTBSY;
 179         else
 180                 o->mot_write_count--;
 181         spin_unlock(&o->mot_write_lock);
 182         RETURN(rc);
 183 }
 184
 185 static void mdt_write_allow(struct mdt_object *o)
 186 {
 187         ENTRY;
 188         spin_lock(&o->mot_write_lock);
 189         o->mot_write_count++;
 190         spin_unlock(&o->mot_write_lock);
 191         EXIT;
 192 }
 193
 194 /* there can be no real transaction so prepare the fake one */
 195 static void mdt_empty_transno(struct mdt_thread_info *info, int rc)
 196 {
 197         struct mdt_device      *mdt = info->mti_mdt;
 198         struct ptlrpc_request  *req = mdt_info_req(info);
 199         struct tg_export_data  *ted;
 200         struct lsd_client_data *lcd;
 201         ENTRY;
 202
 203         if (mdt_rdonly(req->rq_export))
 204                 RETURN_EXIT;
 205
 206         /* transaction has occurred already */
 207         if (lustre_msg_get_transno(req->rq_repmsg) != 0)
 208                 RETURN_EXIT;
 209
 210         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 211                 struct thandle         *th;
 212
 213                 /* generate an empty transaction to get a transno
 214                  * and reply data */
 215                 th = dt_trans_create(info->mti_env, mdt->mdt_bottom);
 216                 if (!IS_ERR(th)) {
 217                         rc = dt_trans_start(info->mti_env, mdt->mdt_bottom, th);
 218                         dt_trans_stop(info->mti_env, mdt->mdt_bottom, th);
 219                 }
 220                 RETURN_EXIT;
 221         }
 222
 223         spin_lock(&mdt->mdt_lut.lut_translock);
 224         if (rc != 0) {
 225                 if (info->mti_transno != 0) {
 226                         struct obd_export *exp = req->rq_export;
 227
 228                         CERROR("%s: replay trans %llu NID %s: rc = %d\n",
 229                                mdt_obd_name(mdt), info->mti_transno,
 230                                obd_export_nid2str(exp), 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 = %llu, last_committed = %llu\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 = %llu < last_transno = %llu\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 open_flags)
 295 {
 296         LASSERT(mfd != NULL);
 297
 298         CDEBUG(D_DENTRY, DFID " Change mfd open_flags %#llo -> %#llo.\n",
 299                PFID(mdt_object_fid(mfd->mfd_object)), mfd->mfd_open_flags,
 300                open_flags);
 301
 302         mfd->mfd_open_flags = open_flags;
 303 }
 304
 305 /**
 306  * prep ma_lmm/ma_lmv for md_attr from reply
 307  */
 308 static void mdt_prep_ma_buf_from_rep(struct mdt_thread_info *info,
 309                                      struct mdt_object *obj,
 310                                      struct md_attr *ma)
 311 {
 312         LASSERT(ma->ma_lmv == NULL && ma->ma_lmm == NULL);
 313         if (S_ISDIR(obj->mot_header.loh_attr)) {
 314                 ma->ma_lmv = req_capsule_server_get(info->mti_pill,
 315                                                     &RMF_MDT_MD);
 316                 ma->ma_lmv_size = req_capsule_get_size(info->mti_pill,
 317                                                        &RMF_MDT_MD,
 318                                                        RCL_SERVER);
 319                 if (ma->ma_lmv_size > 0)
 320                         ma->ma_need |= MA_LMV;
 321         } else {
 322                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
 323                                                     &RMF_MDT_MD);
 324                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 325                                                        &RMF_MDT_MD,
 326                                                        RCL_SERVER);
 327                 if (ma->ma_lmm_size > 0)
 328                         ma->ma_need |= MA_LOV;
 329         }
 330 }
 331
 332 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
 333                         struct mdt_object *o, u64 open_flags, int created,
 334                         struct ldlm_reply *rep)
 335 {
 336         struct ptlrpc_request *req = mdt_info_req(info);
 337         struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
 338         struct mdt_file_data *mfd;
 339         struct md_attr *ma  = &info->mti_attr;
 340         struct lu_attr *la  = &ma->ma_attr;
 341         struct mdt_body *repbody;
 342         bool isdir, isreg;
 343         int rc = 0;
 344
 345         ENTRY;
 346         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 347
 348         isreg = S_ISREG(la->la_mode);
 349         isdir = S_ISDIR(la->la_mode);
 350         if (isreg && !(ma->ma_valid & MA_LOV) &&
 351             !(open_flags & MDS_OPEN_RELEASE)) {
 352                 /*
 353                  * No EA, check whether it is will set regEA and dirEA since in
 354                  * above attr get, these size might be zero, so reset it, to
 355                  * retrieve the MD after create obj.
 356                  */
 357                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 358                                                        &RMF_MDT_MD,
 359                                                        RCL_SERVER);
 360                 /* in replay case, p == NULL */
 361                 rc = mdt_create_data(info, p, o);
 362                 if (rc)
 363                         RETURN(rc);
 364
 365                 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_DISP_STRIPE)
 366                         mdt_set_disposition(info, rep, DISP_OPEN_STRIPE);
 367         }
 368
 369         CDEBUG(D_INODE, "after open, ma_valid bit = %#llx lmm_size = %d\n",
 370                ma->ma_valid, ma->ma_lmm_size);
 371
 372         if (ma->ma_valid & MA_LOV) {
 373                 LASSERT(ma->ma_lmm_size != 0);
 374                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 375                 if (isdir)
 376                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 377                 else
 378                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 379         }
 380
 381         if (ma->ma_valid & MA_LMV) {
 382                 LASSERT(ma->ma_lmv_size != 0);
 383                 repbody->mbo_eadatasize = ma->ma_lmv_size;
 384                 LASSERT(isdir);
 385                 repbody->mbo_valid |= OBD_MD_FLDIREA | OBD_MD_MEA;
 386         }
 387
 388         if (open_flags & MDS_FMODE_WRITE)
 389                 rc = mdt_write_get(o);
 390         else if (open_flags & MDS_FMODE_EXEC)
 391                 rc = mdt_write_deny(o);
 392
 393         if (rc)
 394                 RETURN(rc);
 395
 396         rc = mo_open(info->mti_env, mdt_object_child(o),
 397                      created ? open_flags | MDS_OPEN_CREATED : open_flags);
 398         if (rc != 0) {
 399                 /* If we allow the client to chgrp (CFS_SETGRP_PERM), but the
 400                  * client does not know which suppgid should be sent to the MDS,
 401                  * or some other(s) changed the target file's GID after this RPC
 402                  * sent to the MDS with the suppgid as the original GID, then we
 403                  * give the client another chance to send the right suppgid. */
 404                 if (rc == -EACCES &&
 405                     allow_client_chgrp(info, lu_ucred(info->mti_env)))
 406                         mdt_set_disposition(info, rep, DISP_OPEN_DENY);
 407
 408                 GOTO(err_out, rc);
 409         }
 410
 411         mfd = mdt_mfd_new(med);
 412         if (mfd == NULL)
 413                 GOTO(err_out, rc = -ENOMEM);
 414
 415         /*
 416          * Keep a reference on this object for this open, and is
 417          * released by mdt_mfd_close().
 418          */
 419         mdt_object_get(info->mti_env, o);
 420         mfd->mfd_object = o;
 421         mfd->mfd_xid = req->rq_xid;
 422
 423         /*
 424          * @open_flags is always not zero. At least it should be FMODE_READ,
 425          * FMODE_WRITE or MDS_FMODE_EXEC.
 426          */
 427         LASSERT(open_flags != 0);
 428
 429         /* Open handling. */
 430         mdt_mfd_set_mode(mfd, open_flags);
 431
 432         atomic_inc(&o->mot_open_count);
 433         if (open_flags & MDS_OPEN_LEASE)
 434                 atomic_inc(&o->mot_lease_count);
 435
 436         /* replay handle */
 437         if (req_is_replay(req)) {
 438                 struct mdt_file_data *old_mfd;
 439                 /* Check wheather old cookie already exist in
 440                  * the list, becasue when do recovery, client
 441                  * might be disconnected from server, and
 442                  * restart replay, so there maybe some orphan
 443                  * mfd here, we should remove them */
 444                 LASSERT(info->mti_rr.rr_open_handle != NULL);
 445                 spin_lock(&med->med_open_lock);
 446                 old_mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
 447                                               true);
 448                 if (old_mfd != NULL) {
 449                         CDEBUG(D_HA, "delete orphan mfd = %p, fid = "DFID", "
 450                                "cookie = %#llx\n", mfd,
 451                                PFID(mdt_object_fid(mfd->mfd_object)),
 452                                info->mti_rr.rr_open_handle->cookie);
 453                         class_handle_unhash(&old_mfd->mfd_open_handle);
 454                         list_del_init(&old_mfd->mfd_list);
 455                         spin_unlock(&med->med_open_lock);
 456                         /* no attr update for that close */
 457                         la->la_valid = 0;
 458                         ma->ma_valid |= MA_FLAGS;
 459                         ma->ma_attr_flags |= MDS_RECOV_OPEN;
 460                         mdt_mfd_close(info, old_mfd);
 461                         ma->ma_attr_flags &= ~MDS_RECOV_OPEN;
 462                         ma->ma_valid &= ~MA_FLAGS;
 463                 } else {
 464                         spin_unlock(&med->med_open_lock);
 465                         CDEBUG(D_HA, "orphan mfd not found, fid = "DFID", "
 466                                "cookie = %#llx\n",
 467                                PFID(mdt_object_fid(mfd->mfd_object)),
 468                                info->mti_rr.rr_open_handle->cookie);
 469                 }
 470
 471                 CDEBUG(D_HA, "Store old cookie %#llx in new mfd\n",
 472                        info->mti_rr.rr_open_handle->cookie);
 473
 474                 mfd->mfd_open_handle_old = *info->mti_rr.rr_open_handle;
 475         }
 476
 477         repbody->mbo_open_handle.cookie = mfd->mfd_open_handle.h_cookie;
 478
 479         if (req->rq_export->exp_disconnected) {
 480                 spin_lock(&med->med_open_lock);
 481                 class_handle_unhash(&mfd->mfd_open_handle);
 482                 list_del_init(&mfd->mfd_list);
 483                 spin_unlock(&med->med_open_lock);
 484                 mdt_mfd_close(info, mfd);
 485         } else {
 486                 spin_lock(&med->med_open_lock);
 487                 list_add_tail(&mfd->mfd_list, &med->med_open_head);
 488                 spin_unlock(&med->med_open_lock);
 489         }
 490
 491         mdt_empty_transno(info, rc);
 492
 493         RETURN(rc);
 494
 495 err_out:
 496         if (open_flags & MDS_FMODE_WRITE)
 497                 mdt_write_put(o);
 498         else if (open_flags & MDS_FMODE_EXEC)
 499                 mdt_write_allow(o);
 500
 501         return rc;
 502 }
 503
 504 static int mdt_finish_open(struct mdt_thread_info *info,
 505                            struct mdt_object *p, struct mdt_object *o,
 506                            u64 open_flags, int created,
 507                            struct ldlm_reply *rep)
 508 {
 509         struct ptlrpc_request   *req = mdt_info_req(info);
 510         struct obd_export       *exp = req->rq_export;
 511         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 512         struct md_attr          *ma  = &info->mti_attr;
 513         struct lu_attr          *la  = &ma->ma_attr;
 514         struct mdt_file_data    *mfd;
 515         struct mdt_body         *repbody;
 516         int                      rc = 0;
 517         int                      isreg, isdir, islnk;
 518         struct list_head        *t;
 519         ENTRY;
 520
 521         LASSERT(ma->ma_valid & MA_INODE);
 522
 523         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 524
 525         isreg = S_ISREG(la->la_mode);
 526         isdir = S_ISDIR(la->la_mode);
 527         islnk = S_ISLNK(la->la_mode);
 528         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 529
 530         /* compatibility check for 2.10 clients when it tries to open mirrored
 531          * files. 2.10 clients don't verify overlapping components so they
 532          * would read and write mirrored files just as if they were normal
 533          * PFL files, which will cause the problem that sycned mirrors actually
 534          * contain different data.
 535          * Older clients are not a concern here because they don't even
 536          * understand PFL layout. */
 537         if (isreg && !exp_connect_flr(exp) && ma->ma_valid & MA_LOV &&
 538             mdt_lmm_is_flr(ma->ma_lmm)) {
 539                 /* LU-10286: for simplicity clients who don't understand
 540                  * mirrored layout(with connect flag OBD_CONNECT2_FLR) won't
 541                  * be able to open mirrored files */
 542                 RETURN(-EOPNOTSUPP);
 543         }
 544
 545         /* LU-2275, simulate broken behaviour (esp. prevalent in
 546          * pre-2.4 servers where a very strange reply is sent on error
 547          * that looks like it was actually almost successful and a
 548          * failure at the same time.) */
 549         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_NEGATIVE_POSITIVE)) {
 550                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN |
 551                                                DISP_LOOKUP_NEG |
 552                                                DISP_LOOKUP_POS);
 553
 554                 if (open_flags & MDS_OPEN_LOCK)
 555                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
 556
 557                 RETURN(-ENOENT);
 558         }
 559
 560 #ifdef CONFIG_FS_POSIX_ACL
 561         if (exp_connect_flags(exp) & OBD_CONNECT_ACL) {
 562                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
 563                 if (IS_ERR(nodemap))
 564                         RETURN(PTR_ERR(nodemap));
 565
 566                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
 567                 nodemap_putref(nodemap);
 568                 if (rc)
 569                         RETURN(rc);
 570         }
 571 #endif
 572
 573         /*
 574          * If we are following a symlink, don't open; and do not return open
 575          * handle for special nodes as client required.
 576          */
 577         if (islnk || (!isreg && !isdir &&
 578             (exp_connect_flags(req->rq_export) & OBD_CONNECT_NODEVOH))) {
 579                 lustre_msg_set_transno(req->rq_repmsg, 0);
 580                 RETURN(0);
 581         }
 582
 583         /*
 584          * We need to return the existing object's fid back, so it is done here,
 585          * after preparing the reply.
 586          */
 587         if (!created && (open_flags & MDS_OPEN_EXCL) &&
 588             (open_flags & MDS_OPEN_CREAT))
 589                 RETURN(-EEXIST);
 590
 591         /* This can't be done earlier, we need to return reply body */
 592         if (isdir) {
 593                 if (open_flags & (MDS_OPEN_CREAT | MDS_FMODE_WRITE)) {
 594                         /* We are trying to create or write an existing dir. */
 595                         RETURN(-EISDIR);
 596                 }
 597         } else if (open_flags & MDS_OPEN_DIRECTORY)
 598                 RETURN(-ENOTDIR);
 599
 600         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
 601                                  OBD_FAIL_MDS_LDLM_REPLY_NET | OBD_FAIL_ONCE))
 602                 RETURN(-EAGAIN);
 603
 604         mfd = NULL;
 605         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 606                 spin_lock(&med->med_open_lock);
 607                 list_for_each(t, &med->med_open_head) {
 608                         mfd = list_entry(t, struct mdt_file_data, mfd_list);
 609                         if (mfd->mfd_xid == req->rq_xid)
 610                                 break;
 611                         mfd = NULL;
 612                 }
 613                 spin_unlock(&med->med_open_lock);
 614
 615                 if (mfd != NULL) {
 616                         repbody->mbo_open_handle.cookie =
 617                                 mfd->mfd_open_handle.h_cookie;
 618                         /* set repbody->ea_size for resent case */
 619                         if (ma->ma_valid & MA_LOV) {
 620                                 LASSERT(ma->ma_lmm_size != 0);
 621                                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 622                                 if (isdir)
 623                                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 624                                 else
 625                                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 626                         }
 627                         mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 628                         RETURN(0);
 629                 }
 630         }
 631
 632         rc = mdt_mfd_open(info, p, o, open_flags, created, rep);
 633         if (!rc)
 634                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 635
 636         RETURN(rc);
 637 }
 638
 639 void mdt_reconstruct_open(struct mdt_thread_info *info,
 640                           struct mdt_lock_handle *lhc)
 641 {
 642         const struct lu_env *env = info->mti_env;
 643         struct mdt_device *mdt = info->mti_mdt;
 644         struct req_capsule *pill = info->mti_pill;
 645         struct ptlrpc_request *req = mdt_info_req(info);
 646         struct md_attr *ma = &info->mti_attr;
 647         struct mdt_reint_record *rr = &info->mti_rr;
 648         u64 open_flags = info->mti_spec.sp_cr_flags;
 649         struct ldlm_reply *ldlm_rep;
 650         struct mdt_object *parent;
 651         struct mdt_object *child;
 652         struct mdt_body *repbody;
 653         u64 opdata;
 654         int rc;
 655         ENTRY;
 656
 657         LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
 658         ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
 659         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 660
 661         ma->ma_need = MA_INODE | MA_HSM;
 662         ma->ma_valid = 0;
 663
 664         opdata = mdt_req_from_lrd(req, info->mti_reply_data);
 665         mdt_set_disposition(info, ldlm_rep, opdata);
 666
 667         CDEBUG(D_INODE, "This is reconstruct open: disp=%#llx, result=%d\n",
 668                ldlm_rep->lock_policy_res1, req->rq_status);
 669
 670         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
 671             req->rq_status != 0)
 672                 /* We did not create successfully, return error to client. */
 673                 GOTO(out, rc = req->rq_status);
 674
 675         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
 676                 struct obd_export *exp = req->rq_export;
 677                 /*
 678                  * We failed after creation, but we do not know in which step
 679                  * we failed. So try to check the child object.
 680                  */
 681                 parent = mdt_object_find(env, mdt, rr->rr_fid1);
 682                 if (IS_ERR(parent)) {
 683                         rc = PTR_ERR(parent);
 684                         LCONSOLE_WARN("Parent "DFID" lookup error %d."
 685                                       " Evicting client %s with export %s.\n",
 686                                       PFID(rr->rr_fid1), rc,
 687                                       obd_uuid2str(&exp->exp_client_uuid),
 688                                       obd_export_nid2str(exp));
 689                         mdt_export_evict(exp);
 690                         RETURN_EXIT;
 691                 }
 692
 693                 child = mdt_object_find(env, mdt, rr->rr_fid2);
 694                 if (IS_ERR(child)) {
 695                         rc = PTR_ERR(child);
 696                         LCONSOLE_WARN("cannot lookup child "DFID": rc = %d; "
 697                                       "evicting client %s with export %s\n",
 698                                       PFID(rr->rr_fid2), rc,
 699                                       obd_uuid2str(&exp->exp_client_uuid),
 700                                       obd_export_nid2str(exp));
 701                         mdt_object_put(env, parent);
 702                         mdt_export_evict(exp);
 703                         RETURN_EXIT;
 704                 }
 705
 706                 if (unlikely(mdt_object_remote(child))) {
 707                         /* the child object was created on remote server */
 708                         if (!mdt_is_dne_client(exp)) {
 709                                 /* Return -EIO for old client */
 710                                 mdt_object_put(env, parent);
 711                                 mdt_object_put(env, child);
 712                                 GOTO(out, rc = -EIO);
 713                         }
 714                         repbody->mbo_fid1 = *rr->rr_fid2;
 715                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 716                         rc = 0;
 717                 } else {
 718                         if (mdt_object_exists(child)) {
 719                                 mdt_prep_ma_buf_from_rep(info, child, ma);
 720                                 rc = mdt_attr_get_complex(info, child, ma);
 721                                 if (rc == 0)
 722                                         rc = mdt_finish_open(info, parent,
 723                                                              child, open_flags,
 724                                                              1, ldlm_rep);
 725                         } else {
 726                                 /* the child does not exist, we should do
 727                                  * regular open */
 728                                 mdt_object_put(env, parent);
 729                                 mdt_object_put(env, child);
 730                                 GOTO(regular_open, 0);
 731                         }
 732                 }
 733                 mdt_object_put(env, parent);
 734                 mdt_object_put(env, child);
 735                 GOTO(out, rc);
 736         } else {
 737 regular_open:
 738                 /* We did not try to create, so we are a pure open */
 739                 rc = mdt_reint_open(info, lhc);
 740         }
 741
 742         EXIT;
 743 out:
 744         req->rq_status = rc;
 745         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
 746         LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
 747 }
 748
 749 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep)
 750 {
 751         u64 open_flags = info->mti_spec.sp_cr_flags;
 752         struct mdt_reint_record *rr = &info->mti_rr;
 753         struct md_attr *ma = &info->mti_attr;
 754         struct mdt_object *o;
 755         int rc;
 756
 757         ENTRY;
 758         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
 759         if (IS_ERR(o))
 760                 RETURN(rc = PTR_ERR(o));
 761
 762         if (unlikely(mdt_object_remote(o))) {
 763                 /* the child object was created on remote server */
 764                 struct mdt_body *repbody;
 765
 766                 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 767                                                 DISP_LOOKUP_EXECD |
 768                                                 DISP_LOOKUP_POS));
 769                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 770                 repbody->mbo_fid1 = *rr->rr_fid2;
 771                 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 772                 rc = 0;
 773         } else {
 774                 if (mdt_object_exists(o)) {
 775                         mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 776                                                         DISP_LOOKUP_EXECD |
 777                                                         DISP_LOOKUP_POS));
 778                         mdt_prep_ma_buf_from_rep(info, o, ma);
 779                         rc = mdt_attr_get_complex(info, o, ma);
 780                         if (rc == 0)
 781                                 rc = mdt_finish_open(info, NULL, o, open_flags,
 782                                                      0, rep);
 783                 } else {
 784                         rc = -ENOENT;
 785                 }
 786         }
 787
 788         mdt_object_put(info->mti_env, o);
 789         RETURN(rc);
 790 }
 791
 792 /* lock object for open */
 793 static int mdt_object_open_lock(struct mdt_thread_info *info,
 794                                 struct mdt_object *obj,
 795                                 struct mdt_lock_handle *lhc,
 796                                 __u64 *ibits)
 797 {
 798         struct md_attr *ma = &info->mti_attr;
 799         __u64 open_flags = info->mti_spec.sp_cr_flags;
 800         __u64 trybits = 0;
 801         enum ldlm_mode lm = LCK_CR;
 802         bool acq_lease = !!(open_flags & MDS_OPEN_LEASE);
 803         bool try_layout = false;
 804         bool create_layout = false;
 805         int rc = 0;
 806         int dom_stripes = LMM_NO_DOM;
 807         bool dom_lock = false;
 808
 809         ENTRY;
 810
 811         *ibits = 0;
 812         mdt_lock_handle_init(lhc);
 813
 814         if (req_is_replay(mdt_info_req(info)))
 815                 RETURN(0);
 816
 817         if (S_ISREG(lu_object_attr(&obj->mot_obj))) {
 818                 if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV) &&
 819                     md_should_create(open_flags))
 820                         create_layout = true;
 821                 if (exp_connect_layout(info->mti_exp) && !create_layout &&
 822                     ma->ma_need & MA_LOV)
 823                         try_layout = true;
 824
 825                 /* DoM files can have just MDT stripe or combined MDT + OST
 826                  * stripes.
 827                  * - In the first case the open for read/write will do IO to
 828                  *   the MDT stripe and it makes sense to take IO lock in
 829                  *   advance along with OPEN even if it is blocking lock.
 830                  * - In the second case it is just size of MDT stripe and it
 831                  *   is quite unlikely that client will write into it, though
 832                  *   it may read it. So IO lock will be taken optionally if it
 833                  *   is non-blocking one.
 834                  */
 835                 if (ma->ma_valid & MA_LOV && ma->ma_lmm != NULL)
 836                         dom_stripes = mdt_lmm_dom_entry(ma->ma_lmm);
 837
 838                 if (dom_stripes == LMM_DOM_ONLY &&
 839                     info->mti_mdt->mdt_opts.mo_dom_lock > 0 &&
 840                     !mdt_dom_client_has_lock(info, mdt_object_fid(obj)))
 841                         dom_lock = true;
 842         }
 843
 844         if (acq_lease) {
 845                 /* lease open, acquire write mode of open sem */
 846                 down_write(&obj->mot_open_sem);
 847
 848                 /* Lease exists and ask for new lease */
 849                 if (atomic_read(&obj->mot_lease_count) > 0) {
 850                         /* only exclusive open is supported, so lease
 851                          * are conflicted to each other */
 852                         GOTO(out, rc = -EBUSY);
 853                 }
 854
 855                 /* Lease must be with open lock */
 856                 if (!(open_flags & MDS_OPEN_LOCK)) {
 857                         CERROR("%s: Request lease for file:"DFID ", but open lock "
 858                                "is missed, open_flags = %#llo : rc = %d\n",
 859                                mdt_obd_name(info->mti_mdt),
 860                                PFID(mdt_object_fid(obj)), open_flags, -EPROTO);
 861                         GOTO(out, rc = -EPROTO);
 862                 }
 863
 864                 /* XXX: only exclusive open is supported. */
 865                 lm = LCK_EX;
 866                 *ibits = MDS_INODELOCK_OPEN;
 867
 868                 /* never grant LCK_EX layout lock to client */
 869                 try_layout = false;
 870         } else { /* normal open */
 871                 /* normal open holds read mode of open sem */
 872                 down_read(&obj->mot_open_sem);
 873
 874                 if (open_flags & MDS_OPEN_LOCK) {
 875                         if (open_flags & MDS_FMODE_WRITE)
 876                                 lm = LCK_CW;
 877                         else if (open_flags & MDS_FMODE_EXEC)
 878                                 lm = LCK_PR;
 879                         else
 880                                 lm = LCK_CR;
 881
 882                         *ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN;
 883                 } else if (atomic_read(&obj->mot_lease_count) > 0) {
 884                         if (open_flags & MDS_FMODE_WRITE)
 885                                 lm = LCK_CW;
 886                         else
 887                                 lm = LCK_CR;
 888
 889                         /* revoke lease */
 890                         *ibits = MDS_INODELOCK_OPEN;
 891                         try_layout = false;
 892
 893                         lhc = &info->mti_lh[MDT_LH_LOCAL];
 894                 } else if (dom_lock) {
 895                         lm = (open_flags & MDS_FMODE_WRITE) ? LCK_PW : LCK_PR;
 896                         if (info->mti_mdt->mdt_opts.mo_dom_lock ==
 897                             TRYLOCK_DOM_ON_OPEN) {
 898                                 trybits |= MDS_INODELOCK_DOM |
 899                                            MDS_INODELOCK_LAYOUT;
 900                         } else {
 901                                 /* mo_dom_lock == ALWAYS_DOM_LOCK_ON_OPEN */
 902                                 *ibits = MDS_INODELOCK_DOM;
 903                                 if (info->mti_mdt->mdt_opts.mo_dom_read_open) {
 904                                         trybits |= MDS_INODELOCK_LAYOUT;
 905                                 }
 906                         }
 907                 }
 908
 909                 CDEBUG(D_INODE, "normal open:"DFID" lease count: %d, lm: %d\n",
 910                         PFID(mdt_object_fid(obj)),
 911                         atomic_read(&obj->mot_open_count), lm);
 912         }
 913
 914         mdt_lock_reg_init(lhc, lm);
 915
 916         /* Return lookup lock to validate inode at the client side.
 917          * This is pretty important otherwise MDT will return layout
 918          * lock for each open.
 919          * However this is a double-edged sword because changing
 920          * permission will revoke a huge number of LOOKUP locks.
 921          */
 922         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_OPEN) && try_layout) {
 923                 if (!(*ibits & MDS_INODELOCK_LOOKUP))
 924                         trybits |= MDS_INODELOCK_LOOKUP;
 925                 trybits |= MDS_INODELOCK_LAYOUT;
 926         }
 927
 928         if (*ibits | trybits)
 929                 rc = mdt_object_lock_try(info, obj, lhc, ibits, trybits, false);
 930
 931         CDEBUG(D_INODE, "%s: Requested bits lock:"DFID ", ibits = %#llx/%#llx"
 932                ", open_flags = %#llo, try_layout = %d : rc = %d\n",
 933                mdt_obd_name(info->mti_mdt), PFID(mdt_object_fid(obj)),
 934                *ibits, trybits, open_flags, try_layout, rc);
 935
 936         /* will change layout, revoke layout locks by enqueuing EX lock. */
 937         if (rc == 0 && create_layout) {
 938                 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LAYOUT];
 939
 940                 CDEBUG(D_INODE, "Will create layout, get EX layout lock:"DFID
 941                         ", open_flags = %#llo\n",
 942                         PFID(mdt_object_fid(obj)), open_flags);
 943
 944                 /* We cannot enqueue another lock for the same resource we
 945                  * already have a lock for, due to mechanics of waiting list
 946                  * iterating in ldlm, see LU-3601.
 947                  * As such we'll drop the open lock we just got above here,
 948                  * it's ok not to have this open lock as it's main purpose is to
 949                  * flush unused cached client open handles. */
 950                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
 951                         mdt_object_unlock(info, obj, lhc, 1);
 952
 953                 LASSERT(!try_layout);
 954                 mdt_lock_handle_init(ll);
 955                 mdt_lock_reg_init(ll, LCK_EX);
 956                 rc = mdt_object_lock(info, obj, ll, MDS_INODELOCK_LAYOUT);
 957
 958                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LL_BLOCK, 2);
 959         }
 960
 961         /* Check if there is any other open handles after acquiring
 962          * open lock. At this point, caching open handles have been revoked
 963          * by open lock.
 964          * XXX: Now only exclusive open is supported. Need to check the
 965          * type of open for generic lease support. */
 966         if (rc == 0 && acq_lease) {
 967                 struct ptlrpc_request *req = mdt_info_req(info);
 968                 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
 969                 struct mdt_file_data *mfd;
 970                 bool is_replay_or_resent;
 971                 int open_count = 0;
 972
 973                 /* For lease: application can open a file and then apply lease,
 974                  * @handle contains original open handle in that case.
 975                  * In recovery, open REQ will be replayed and the lease REQ may
 976                  * be resent that means the open handle is already stale, so we
 977                  * need to fix it up here by finding new handle. */
 978                 is_replay_or_resent = req_is_replay(req) ||
 979                         lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
 980
 981                 /* if the request is _not_ a replay request, rr_open_handle
 982                  * may be used to hold an open file handle which is issuing the
 983                  * lease request, so that this openhandle doesn't count. */
 984                 mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
 985                                           is_replay_or_resent);
 986                 if (mfd != NULL)
 987                         ++open_count;
 988
 989                 CDEBUG(D_INODE, "acq_lease "DFID": openers: %d, want: %d\n",
 990                         PFID(mdt_object_fid(obj)),
 991                         atomic_read(&obj->mot_open_count), open_count);
 992
 993                 if (atomic_read(&obj->mot_open_count) > open_count)
 994                         GOTO(out, rc = -EBUSY);
 995         }
 996         GOTO(out, rc);
 997
 998 out:
 999         RETURN(rc);
1000 }
1001
1002 static void mdt_object_open_unlock(struct mdt_thread_info *info,
1003                                    struct mdt_object *obj,
1004                                    struct mdt_lock_handle *lhc,
1005                                    __u64 ibits, int rc)
1006 {
1007         __u64 open_flags = info->mti_spec.sp_cr_flags;
1008         struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LOCAL];
1009         ENTRY;
1010
1011         if (req_is_replay(mdt_info_req(info)))
1012                 RETURN_EXIT;
1013
1014         /* Release local lock - the lock put in MDT_LH_LOCAL will never
1015          * return to client side. */
1016         if (lustre_handle_is_used(&ll->mlh_reg_lh))
1017                 mdt_object_unlock(info, obj, ll, 1);
1018
1019         ll = &info->mti_lh[MDT_LH_LAYOUT];
1020         /* Release local layout lock, layout was created */
1021         if (lustre_handle_is_used(&ll->mlh_reg_lh)) {
1022                 LASSERT(!(ibits & MDS_INODELOCK_LAYOUT));
1023                 mdt_object_unlock(info, obj, ll, 1);
1024         }
1025
1026         if (open_flags & MDS_OPEN_LEASE)
1027                 up_write(&obj->mot_open_sem);
1028         else
1029                 up_read(&obj->mot_open_sem);
1030
1031         /* Cross-ref case, the lock should be returned to the client */
1032         if (ibits == 0 || rc == -MDT_EREMOTE_OPEN)
1033                 RETURN_EXIT;
1034
1035         if (!(open_flags & MDS_OPEN_LOCK) && !(ibits & MDS_INODELOCK_LAYOUT) &&
1036             !(ibits & MDS_INODELOCK_DOM)) {
1037                 /* for the open request, the lock will only return to client
1038                  * if open or layout lock is granted. */
1039                 rc = 1;
1040         }
1041
1042         if (rc != 0 || !lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1043                 struct ldlm_reply       *ldlm_rep;
1044
1045                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1046                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1047                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
1048                         mdt_object_unlock(info, obj, lhc, 1);
1049         }
1050         RETURN_EXIT;
1051 }
1052
1053 /**
1054  * Check release is permitted for the current HSM flags.
1055  */
1056 static bool mdt_hsm_release_allow(const struct md_attr *ma)
1057 {
1058         if (!(ma->ma_valid & MA_HSM))
1059                 return false;
1060
1061         if (ma->ma_hsm.mh_flags & (HS_DIRTY|HS_NORELEASE|HS_LOST))
1062                 return false;
1063
1064         if (!(ma->ma_hsm.mh_flags & HS_ARCHIVED))
1065                 return false;
1066
1067         return true;
1068 }
1069
1070 static int mdt_open_by_fid_lock(struct mdt_thread_info *info,
1071                                 struct ldlm_reply *rep,
1072                                 struct mdt_lock_handle *lhc)
1073 {
1074         const struct lu_env *env = info->mti_env;
1075         struct mdt_device *mdt = info->mti_mdt;
1076         u64 open_flags = info->mti_spec.sp_cr_flags;
1077         struct mdt_reint_record *rr = &info->mti_rr;
1078         struct md_attr *ma = &info->mti_attr;
1079         struct mdt_object *parent = NULL;
1080         struct mdt_object *o;
1081         bool object_locked = false;
1082         u64 ibits = 0;
1083         int rc;
1084
1085         ENTRY;
1086         if (md_should_create(open_flags)) {
1087                 if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1088                         parent = mdt_object_find(env, mdt, rr->rr_fid1);
1089                         if (IS_ERR(parent)) {
1090                                 CDEBUG(D_INODE, "Fail to find parent "DFID
1091                                        " for anonymous created %ld, try to"
1092                                        " use server-side parent.\n",
1093                                        PFID(rr->rr_fid1), PTR_ERR(parent));
1094                                 parent = NULL;
1095                         }
1096                 }
1097                 if (parent == NULL)
1098                         ma->ma_need |= MA_PFID;
1099         }
1100
1101         o = mdt_object_find(env, mdt, rr->rr_fid2);
1102         if (IS_ERR(o))
1103                 GOTO(out_parent_put, rc = PTR_ERR(o));
1104
1105         if (mdt_object_remote(o)) {
1106                 CDEBUG(D_INFO, "%s: "DFID" is on remote MDT.\n",
1107                        mdt_obd_name(info->mti_mdt),
1108                        PFID(rr->rr_fid2));
1109                 GOTO(out, rc = -EREMOTE);
1110         } else if (!mdt_object_exists(o)) {
1111                 mdt_set_disposition(info, rep,
1112                                     DISP_IT_EXECD |
1113                                     DISP_LOOKUP_EXECD |
1114                                     DISP_LOOKUP_NEG);
1115                 GOTO(out, rc = -ENOENT);
1116         }
1117
1118         mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1119
1120         mdt_prep_ma_buf_from_rep(info, o, ma);
1121         if (open_flags & MDS_OPEN_RELEASE)
1122                 ma->ma_need |= MA_HSM;
1123         rc = mdt_attr_get_complex(info, o, ma);
1124         if (rc)
1125                 GOTO(out, rc);
1126
1127         /* We should not change file's existing LOV EA */
1128         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
1129             open_flags & MDS_OPEN_HAS_EA && ma->ma_valid & MA_LOV)
1130                 GOTO(out, rc = -EEXIST);
1131
1132         /* If a release request, check file open flags are fine and ask for an
1133          * exclusive open access. */
1134         if (open_flags & MDS_OPEN_RELEASE && !mdt_hsm_release_allow(ma))
1135                 GOTO(out, rc = -EPERM);
1136
1137         rc = mdt_check_resent_lock(info, o, lhc);
1138         if (rc < 0) {
1139                 GOTO(out, rc);
1140         } else if (rc > 0) {
1141                 rc = mdt_object_open_lock(info, o, lhc, &ibits);
1142                 object_locked = true;
1143                 if (rc)
1144                         GOTO(out_unlock, rc);
1145         }
1146
1147         if (ma->ma_valid & MA_PFID) {
1148                 parent = mdt_object_find(env, mdt, &ma->ma_pfid);
1149                 if (IS_ERR(parent)) {
1150                         CDEBUG(D_INODE, "Fail to find parent "DFID
1151                                " for anonymous created %ld, try to"
1152                                " use system default.\n",
1153                                PFID(&ma->ma_pfid), PTR_ERR(parent));
1154                         parent = NULL;
1155                 }
1156         }
1157
1158         rc = mdt_finish_open(info, parent, o, open_flags, 0, rep);
1159         if (!rc) {
1160                 mdt_set_disposition(info, rep, DISP_LOOKUP_POS);
1161                 if (open_flags & MDS_OPEN_LOCK)
1162                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1163                 if (open_flags & MDS_OPEN_LEASE)
1164                         mdt_set_disposition(info, rep, DISP_OPEN_LEASE);
1165         }
1166         GOTO(out_unlock, rc);
1167
1168 out_unlock:
1169         if (object_locked)
1170                 mdt_object_open_unlock(info, o, lhc, ibits, rc);
1171 out:
1172         mdt_object_put(env, o);
1173         if (rc == 0)
1174                 mdt_pack_size2body(info, rr->rr_fid2, &lhc->mlh_reg_lh);
1175 out_parent_put:
1176         if (parent != NULL)
1177                 mdt_object_put(env, parent);
1178         return rc;
1179 }
1180
1181 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1182 static int mdt_cross_open(struct mdt_thread_info *info,
1183                           const struct lu_fid *parent_fid,
1184                           const struct lu_fid *fid,
1185                           struct ldlm_reply *rep, u64 open_flags)
1186 {
1187         struct md_attr *ma = &info->mti_attr;
1188         struct mdt_object *o;
1189         int rc;
1190         ENTRY;
1191
1192         o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1193         if (IS_ERR(o))
1194                 RETURN(rc = PTR_ERR(o));
1195
1196         if (mdt_object_remote(o)) {
1197                 /* Something is wrong here, the object is on another MDS! */
1198                 CERROR("%s: "DFID" isn't on this server!: rc = %d\n",
1199                        mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1200                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1201                                 &o->mot_obj,
1202                                 "Object isn't on this server! FLD error?");
1203                 rc = -EFAULT;
1204         } else {
1205                 if (mdt_object_exists(o)) {
1206                         int mask;
1207
1208                         /* Do permission check for cross-open after converting
1209                          * MDS_OPEN_* flags to MAY_* permission mask.
1210                          */
1211                         mask = mds_accmode(open_flags);
1212
1213                         rc = mo_permission(info->mti_env, NULL,
1214                                            mdt_object_child(o), NULL, mask);
1215                         if (rc)
1216                                 goto out;
1217
1218                         mdt_prep_ma_buf_from_rep(info, o, ma);
1219                         rc = mdt_attr_get_complex(info, o, ma);
1220                         if (rc != 0)
1221                                 GOTO(out, rc);
1222
1223                         rc = mdt_finish_open(info, NULL, o, open_flags, 0, rep);
1224                 } else {
1225                         /*
1226                          * Something is wrong here. lookup was positive but
1227                          * there is no object!
1228                          */
1229                         CERROR("%s: "DFID" doesn't exist!: rc = %d\n",
1230                               mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1231                         rc = -EFAULT;
1232                 }
1233         }
1234 out:
1235         mdt_object_put(info->mti_env, o);
1236
1237         RETURN(rc);
1238 }
1239
1240 /*
1241  * find root object and take its xattr lock if it's on remote MDT, later create
1242  * may use fs default striping (which is stored in root xattr).
1243  */
1244 static int mdt_lock_root_xattr(struct mdt_thread_info *info,
1245                                struct mdt_device *mdt)
1246 {
1247         struct mdt_object *md_root = mdt->mdt_md_root;
1248         struct lustre_handle lhroot;
1249         int rc;
1250
1251         if (md_root == NULL) {
1252                 lu_root_fid(&info->mti_tmp_fid1);
1253                 md_root = mdt_object_find(info->mti_env, mdt,
1254                                           &info->mti_tmp_fid1);
1255                 if (IS_ERR(md_root))
1256                         return PTR_ERR(md_root);
1257
1258                 spin_lock(&mdt->mdt_lock);
1259                 if (mdt->mdt_md_root != NULL) {
1260                         spin_unlock(&mdt->mdt_lock);
1261
1262                         LASSERTF(mdt->mdt_md_root == md_root,
1263                                  "Different root object ("
1264                                  DFID") instances, %p, %p\n",
1265                                  PFID(&info->mti_tmp_fid1),
1266                                  mdt->mdt_md_root, md_root);
1267                         LASSERT(atomic_read(
1268                                 &md_root->mot_obj.lo_header->loh_ref) > 1);
1269
1270                         mdt_object_put(info->mti_env, md_root);
1271                 } else {
1272                         mdt->mdt_md_root = md_root;
1273                         spin_unlock(&mdt->mdt_lock);
1274                 }
1275         }
1276
1277         if (md_root->mot_cache_attr || !mdt_object_remote(md_root))
1278                 return 0;
1279
1280         rc = mdt_remote_object_lock(info, md_root, mdt_object_fid(md_root),
1281                                     &lhroot, LCK_PR, MDS_INODELOCK_XATTR,
1282                                     true);
1283         if (rc < 0)
1284                 return rc;
1285
1286         md_root->mot_cache_attr = 1;
1287
1288         /* don't cancel this lock, so that we know the cached xattr is valid. */
1289         ldlm_lock_decref(&lhroot, LCK_PR);
1290
1291         return 0;
1292 }
1293
1294 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1295 {
1296         struct mdt_device *mdt = info->mti_mdt;
1297         struct ptlrpc_request *req = mdt_info_req(info);
1298         struct mdt_object *parent;
1299         struct mdt_object *child;
1300         struct mdt_lock_handle *lh;
1301         struct ldlm_reply *ldlm_rep;
1302         struct mdt_body *repbody;
1303         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1304         struct md_attr *ma = &info->mti_attr;
1305         u64 open_flags = info->mti_spec.sp_cr_flags;
1306         u64 ibits = 0;
1307         struct mdt_reint_record *rr = &info->mti_rr;
1308         int result, rc;
1309         int created = 0;
1310         int object_locked = 0;
1311         u32 msg_flags;
1312
1313         ENTRY;
1314         OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1315                                (obd_timeout + 1) / 4);
1316
1317         mdt_counter_incr(req, LPROC_MDT_OPEN);
1318         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1319
1320         ma->ma_need = MA_INODE;
1321         ma->ma_valid = 0;
1322
1323         LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1324         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1325
1326         if (unlikely(open_flags & MDS_OPEN_JOIN_FILE)) {
1327                 CERROR("file join is not supported anymore.\n");
1328                 GOTO(out, result = err_serious(-EOPNOTSUPP));
1329         }
1330         msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1331
1332         if ((open_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1333             info->mti_spec.u.sp_ea.eadata == NULL)
1334                 GOTO(out, result = err_serious(-EINVAL));
1335
1336         if (open_flags & MDS_FMODE_WRITE &&
1337             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1338                 GOTO(out, result = -EROFS);
1339
1340         CDEBUG(D_INODE, "I am going to open "DFID"/("DNAME"->"DFID") "
1341                "cr_flag=%#llo mode=0%06o msg_flag=0x%x\n",
1342                PFID(rr->rr_fid1), PNAME(&rr->rr_name), PFID(rr->rr_fid2),
1343                open_flags, ma->ma_attr.la_mode, msg_flags);
1344
1345         if (info->mti_cross_ref) {
1346                 /* This is cross-ref open */
1347                 mdt_set_disposition(info, ldlm_rep,
1348                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD |
1349                              DISP_LOOKUP_POS));
1350                 result = mdt_cross_open(info, rr->rr_fid2, rr->rr_fid1,
1351                                         ldlm_rep, open_flags);
1352                 GOTO(out, result);
1353         } else if (req_is_replay(req)) {
1354                 result = mdt_open_by_fid(info, ldlm_rep);
1355
1356                 if (result != -ENOENT)
1357                         GOTO(out, result);
1358
1359                 /* We didn't find the correct object, so we need to re-create it
1360                  * via a regular replay. */
1361                 if (!(open_flags & MDS_OPEN_CREAT)) {
1362                         DEBUG_REQ(D_ERROR, req,
1363                                   "OPEN & CREAT not in open replay/by_fid.");
1364                         GOTO(out, result = -EFAULT);
1365                 }
1366                 CDEBUG(D_INFO, "No object(1), continue as regular open.\n");
1367         } else if (open_flags & (MDS_OPEN_BY_FID | MDS_OPEN_LOCK)) {
1368                 /*
1369                  * MDS_OPEN_LOCK is checked for backward compatibility with 2.1
1370                  * client.
1371                  */
1372                 result = mdt_open_by_fid_lock(info, ldlm_rep, lhc);
1373                 if (result < 0)
1374                         CDEBUG(D_INFO, "no object for "DFID": %d\n",
1375                                PFID(rr->rr_fid2), result);
1376                 GOTO(out, result);
1377         }
1378
1379         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1380                 GOTO(out, result = err_serious(-ENOMEM));
1381
1382         mdt_set_disposition(info, ldlm_rep,
1383                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1384
1385         if (!lu_name_is_valid(&rr->rr_name))
1386                 GOTO(out, result = -EPROTO);
1387
1388         result = mdt_lock_root_xattr(info, mdt);
1389         if (result < 0)
1390                 GOTO(out, result);
1391
1392 again:
1393         lh = &info->mti_lh[MDT_LH_PARENT];
1394         mdt_lock_pdo_init(lh, (open_flags & MDS_OPEN_CREAT) ? LCK_PW : LCK_PR,
1395                           &rr->rr_name);
1396
1397         parent = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
1398         if (IS_ERR(parent))
1399                 GOTO(out, result = PTR_ERR(parent));
1400
1401         result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
1402         if (result != 0) {
1403                 mdt_object_put(info->mti_env, parent);
1404                 GOTO(out, result);
1405         }
1406
1407         /* get and check version of parent */
1408         result = mdt_version_get_check(info, parent, 0);
1409         if (result)
1410                 GOTO(out_parent, result);
1411
1412         fid_zero(child_fid);
1413
1414         result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1415                             &rr->rr_name, child_fid, &info->mti_spec);
1416
1417         LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1418                  "looking for "DFID"/"DNAME", found FID = "DFID"\n",
1419                  PFID(mdt_object_fid(parent)), PNAME(&rr->rr_name),
1420                  PFID(child_fid));
1421
1422         if (result != 0 && result != -ENOENT && result != -ESTALE)
1423                 GOTO(out_parent, result);
1424
1425         if (result == -ENOENT || result == -ESTALE) {
1426                 /* If the object is dead, let's check if the object
1427                  * is being migrated to a new object */
1428                 if (result == -ESTALE) {
1429                         struct lu_buf lmv_buf;
1430
1431                         lmv_buf.lb_buf = info->mti_xattr_buf;
1432                         lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
1433                         rc = mo_xattr_get(info->mti_env,
1434                                           mdt_object_child(parent),
1435                                           &lmv_buf, XATTR_NAME_LMV);
1436                         if (rc > 0) {
1437                                 struct lmv_mds_md_v1 *lmv;
1438
1439                                 lmv = lmv_buf.lb_buf;
1440                                 if (le32_to_cpu(lmv->lmv_hash_type) &
1441                                                 LMV_HASH_FLAG_MIGRATION) {
1442                                         /* Get the new parent FID and retry */
1443                                         mdt_object_unlock_put(info, parent,
1444                                                               lh, 1);
1445                                         mdt_lock_handle_init(lh);
1446                                         fid_le_to_cpu(
1447                                                 (struct lu_fid *)rr->rr_fid1,
1448                                                 &lmv->lmv_stripe_fids[1]);
1449                                         goto again;
1450                                 }
1451                         }
1452                 }
1453
1454                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1455                 if (result == -ESTALE) {
1456                         /*
1457                          * -ESTALE means the parent is a dead(unlinked) dir, so
1458                          * it should return -ENOENT to in accordance with the
1459                          * original mds implementaion.
1460                          */
1461                         GOTO(out_parent, result = -ENOENT);
1462                 }
1463
1464                 if (!(open_flags & MDS_OPEN_CREAT))
1465                         GOTO(out_parent, result);
1466                 if (mdt_rdonly(req->rq_export))
1467                         GOTO(out_parent, result = -EROFS);
1468                 *child_fid = *info->mti_rr.rr_fid2;
1469                 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1470                          PFID(child_fid));
1471                 /* In the function below, .hs_keycmp resolves to
1472                  * lu_obj_hop_keycmp() */
1473                 /* coverity[overrun-buffer-val] */
1474                 child = mdt_object_new(info->mti_env, mdt, child_fid);
1475         } else {
1476                 /*
1477                  * Check for O_EXCL is moved to the mdt_finish_open(),
1478                  * we need to return FID back in that case.
1479                  */
1480                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1481                 child = mdt_object_find(info->mti_env, mdt, child_fid);
1482         }
1483         if (IS_ERR(child))
1484                 GOTO(out_parent, result = PTR_ERR(child));
1485
1486         /** check version of child  */
1487         rc = mdt_version_get_check(info, child, 1);
1488         if (rc)
1489                 GOTO(out_child, result = rc);
1490
1491         if (result == -ENOENT) {
1492                 /* Create under OBF and .lustre is not permitted */
1493                 if (!fid_is_md_operative(rr->rr_fid1))
1494                         GOTO(out_child, result = -EPERM);
1495
1496                 /* save versions in reply */
1497                 mdt_version_get_save(info, parent, 0);
1498                 mdt_version_get_save(info, child, 1);
1499
1500                 /* version of child will be changed */
1501                 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
1502
1503                 /* Not found and with MDS_OPEN_CREAT: let's create it. */
1504                 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1505
1506                 /* Let lower layers know what is lock mode on directory. */
1507                 info->mti_spec.sp_cr_mode =
1508                         mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
1509
1510                 /* Don't do lookup sanity check. We know name doesn't exist. */
1511                 info->mti_spec.sp_cr_lookup = 0;
1512                 info->mti_spec.sp_feat = &dt_directory_features;
1513
1514                 result = mdo_create(info->mti_env, mdt_object_child(parent),
1515                                     &rr->rr_name, mdt_object_child(child),
1516                                     &info->mti_spec, &info->mti_attr);
1517                 if (result == -ERESTART) {
1518                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1519                         GOTO(out_child, result);
1520                 } else {
1521                         mdt_prep_ma_buf_from_rep(info, child, ma);
1522                         /* XXX: we should call this once, see few lines below */
1523                         if (result == 0)
1524                                 result = mdt_attr_get_complex(info, child, ma);
1525
1526                         if (result != 0)
1527                                 GOTO(out_child, result);
1528                 }
1529                 created = 1;
1530                 mdt_counter_incr(req, LPROC_MDT_MKNOD);
1531         } else {
1532                 /*
1533                  * The object is on remote node, return its FID for remote open.
1534                  */
1535                 if (mdt_object_remote(child)) {
1536                         /*
1537                          * Check if this lock already was sent to client and
1538                          * this is resent case. For resent case do not take lock
1539                          * again, use what is already granted.
1540                          */
1541                         LASSERT(lhc != NULL);
1542
1543                         rc = mdt_check_resent_lock(info, child, lhc);
1544                         if (rc < 0) {
1545                                 GOTO(out_child, result = rc);
1546                         } else if (rc > 0) {
1547                                 mdt_lock_handle_init(lhc);
1548                                 mdt_lock_reg_init(lhc, LCK_PR);
1549
1550                                 rc = mdt_object_lock(info, child, lhc,
1551                                                      MDS_INODELOCK_LOOKUP);
1552                         }
1553                         repbody->mbo_fid1 = *mdt_object_fid(child);
1554                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1555                         if (rc != 0)
1556                                 result = rc;
1557                         else
1558                                 result = -MDT_EREMOTE_OPEN;
1559                         GOTO(out_child, result);
1560                 } else if (mdt_object_exists(child)) {
1561                         /* We have to get attr & LOV EA & HSM for this
1562                          * object. */
1563                         mdt_prep_ma_buf_from_rep(info, child, ma);
1564                         ma->ma_need |= MA_HSM;
1565                         result = mdt_attr_get_complex(info, child, ma);
1566                         if (result != 0)
1567                                 GOTO(out_child, result);
1568                 } else {
1569                         /* Object does not exist. Likely FS corruption. */
1570                         CERROR("%s: name '"DNAME"' present, but FID "
1571                                DFID" is invalid\n", mdt_obd_name(info->mti_mdt),
1572                                PNAME(&rr->rr_name), PFID(child_fid));
1573                         GOTO(out_child, result = -EIO);
1574                 }
1575         }
1576
1577         rc = mdt_check_resent_lock(info, child, lhc);
1578         if (rc < 0) {
1579                 GOTO(out_child, result = rc);
1580         } else if (rc == 0) {
1581                 /* the open lock might already be gotten in
1582                  * ldlm_handle_enqueue() */
1583                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1584                 if (open_flags & MDS_OPEN_LOCK)
1585                         mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1586         } else {
1587                 /* get openlock if this isn't replay and client requested it */
1588                 if (!req_is_replay(req)) {
1589                         rc = mdt_object_open_lock(info, child, lhc, &ibits);
1590                         object_locked = 1;
1591                         if (rc != 0)
1592                                 GOTO(out_child_unlock, result = rc);
1593                         else if (open_flags & MDS_OPEN_LOCK)
1594                                 mdt_set_disposition(info, ldlm_rep,
1595                                                     DISP_OPEN_LOCK);
1596                 }
1597         }
1598         /* Try to open it now. */
1599         rc = mdt_finish_open(info, parent, child, open_flags,
1600                              created, ldlm_rep);
1601         if (rc) {
1602                 result = rc;
1603                 /* openlock will be released if mdt_finish_open() failed */
1604                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1605
1606                 if (created && (open_flags & MDS_OPEN_VOLATILE)) {
1607                         CERROR("%s: cannot open volatile file "DFID", orphan "
1608                                "file will be left in PENDING directory until "
1609                                "next reboot, rc = %d\n", mdt_obd_name(mdt),
1610                                PFID(mdt_object_fid(child)), rc);
1611                         GOTO(out_child_unlock, result);
1612                 }
1613
1614                 if (created) {
1615                         ma->ma_need = 0;
1616                         ma->ma_valid = 0;
1617                         rc = mdo_unlink(info->mti_env,
1618                                         mdt_object_child(parent),
1619                                         mdt_object_child(child),
1620                                         &rr->rr_name,
1621                                         &info->mti_attr, 0);
1622                         if (rc != 0)
1623                                 CERROR("%s: "DFID" cleanup of open: rc = %d\n",
1624                                        mdt_obd_name(info->mti_mdt),
1625                                        PFID(mdt_object_fid(child)), rc);
1626                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1627                 }
1628         }
1629         EXIT;
1630 out_child_unlock:
1631         if (object_locked)
1632                 mdt_object_open_unlock(info, child, lhc, ibits, result);
1633 out_child:
1634         mdt_object_put(info->mti_env, child);
1635         if (result == 0)
1636                 mdt_pack_size2body(info, child_fid, &lhc->mlh_reg_lh);
1637 out_parent:
1638         mdt_object_unlock_put(info, parent, lh, result || !created);
1639 out:
1640         if (result)
1641                 lustre_msg_set_transno(req->rq_repmsg, 0);
1642         return result;
1643 }
1644
1645 /**
1646  * Create an orphan object use local root.
1647  */
1648 static struct mdt_object *mdt_orphan_open(struct mdt_thread_info *info,
1649                                           struct mdt_device *mdt,
1650                                           const struct lu_fid *fid,
1651                                           struct md_attr *attr, fmode_t fmode)
1652 {
1653         const struct lu_env *env = info->mti_env;
1654         struct md_op_spec *spec = &info->mti_spec;
1655         struct lu_fid *local_root_fid = &info->mti_tmp_fid1;
1656         struct mdt_object *obj = NULL;
1657         struct mdt_object *local_root;
1658         static const struct lu_name lname = {
1659                 .ln_name = "i_am_nobody",
1660                 .ln_namelen = sizeof("i_am_nobody") - 1,
1661         };
1662         struct lu_ucred *uc;
1663         cfs_cap_t uc_cap_save;
1664         int rc;
1665         ENTRY;
1666
1667         rc = dt_root_get(env, mdt->mdt_bottom, local_root_fid);
1668         if (rc != 0)
1669                 RETURN(ERR_PTR(rc));
1670
1671         local_root = mdt_object_find(env, mdt, local_root_fid);
1672         if (IS_ERR(local_root))
1673                 RETURN(local_root);
1674
1675         obj = mdt_object_new(env, mdt, fid);
1676         if (IS_ERR(obj))
1677                 GOTO(out, rc = PTR_ERR(obj));
1678
1679         spec->sp_cr_lookup = 0;
1680         spec->sp_feat = &dt_directory_features;
1681         spec->sp_cr_mode = MDL_MINMODE; /* no lock */
1682         spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
1683         if (attr->ma_valid & MA_LOV) {
1684                 spec->u.sp_ea.eadata = attr->ma_lmm;
1685                 spec->u.sp_ea.eadatalen = attr->ma_lmm_size;
1686                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1687         } else {
1688                 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
1689         }
1690
1691         uc = lu_ucred(env);
1692         uc_cap_save = uc->uc_cap;
1693         uc->uc_cap |= 1 << CFS_CAP_DAC_OVERRIDE;
1694         rc = mdo_create(env, mdt_object_child(local_root), &lname,
1695                         mdt_object_child(obj), spec, attr);
1696         uc->uc_cap = uc_cap_save;
1697         if (rc < 0) {
1698                 CERROR("%s: cannot create volatile file "DFID": rc = %d\n",
1699                        mdt_obd_name(mdt), PFID(fid), rc);
1700                 GOTO(out, rc);
1701         }
1702
1703         rc = mo_open(env, mdt_object_child(obj), MDS_OPEN_CREATED);
1704         if (rc < 0)
1705                 CERROR("%s: cannot open volatile file "DFID", orphan "
1706                        "file will be left in PENDING directory until "
1707                        "next reboot, rc = %d\n", mdt_obd_name(mdt),
1708                        PFID(fid), rc);
1709         GOTO(out, rc);
1710
1711 out:
1712         if (rc < 0) {
1713                 if (!IS_ERR(obj))
1714                         mdt_object_put(env, obj);
1715                 obj = ERR_PTR(rc);
1716         }
1717         mdt_object_put(env, local_root);
1718         return obj;
1719 }
1720
1721 /* XXX Look into layout in MDT layer. */
1722 static inline int mdt_hsm_set_released(struct lov_mds_md *lmm)
1723 {
1724         struct lov_comp_md_v1   *comp_v1;
1725         struct lov_mds_md       *v1;
1726         __u32   off;
1727         int     i;
1728
1729         if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1_DEFINED)) {
1730                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1731
1732                 if (comp_v1->lcm_entry_count == 0)
1733                         return -EINVAL;
1734
1735                 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1736                         off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1737                         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
1738                         v1->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1739                 }
1740         } else {
1741                 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1742         }
1743         return 0;
1744 }
1745
1746 static int mdt_hsm_release(struct mdt_thread_info *info, struct mdt_object *o,
1747                            struct md_attr *ma)
1748 {
1749         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LAYOUT];
1750         struct lu_ucred        *uc = mdt_ucred(info);
1751         struct close_data      *data;
1752         struct ldlm_lock       *lease;
1753         struct mdt_object      *orphan;
1754         struct md_attr         *orp_ma;
1755         struct lu_buf          *buf;
1756         cfs_cap_t               cap;
1757         bool                    lease_broken;
1758         int                     rc;
1759         int                     rc2;
1760         ENTRY;
1761
1762         if (mdt_rdonly(info->mti_exp))
1763                 RETURN(-EROFS);
1764
1765         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1766         if (data == NULL)
1767                 RETURN(-EPROTO);
1768
1769         lease = ldlm_handle2lock(&data->cd_handle);
1770         if (lease == NULL)
1771                 RETURN(-ESTALE);
1772
1773         /* try to hold open_sem so that nobody else can open the file */
1774         if (!down_write_trylock(&o->mot_open_sem)) {
1775                 ldlm_lock_cancel(lease);
1776                 GOTO(out_reprocess, rc = -EBUSY);
1777         }
1778
1779         /* Check if the lease open lease has already canceled */
1780         lock_res_and_lock(lease);
1781         lease_broken = ldlm_is_cancel(lease);
1782         unlock_res_and_lock(lease);
1783
1784         LDLM_DEBUG(lease, DFID " lease broken? %d",
1785                    PFID(mdt_object_fid(o)), lease_broken);
1786
1787         /* Cancel server side lease. Client side counterpart should
1788          * have been cancelled. It's okay to cancel it now as we've
1789          * held mot_open_sem. */
1790         ldlm_lock_cancel(lease);
1791
1792         if (lease_broken) /* don't perform release task */
1793                 GOTO(out_unlock, rc = -ESTALE);
1794
1795         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1796                 GOTO(out_unlock, rc = -EINVAL);
1797
1798         /* ma_need was set before but it seems fine to change it in order to
1799          * avoid modifying the one from RPC */
1800         ma->ma_need = MA_HSM;
1801         rc = mdt_attr_get_complex(info, o, ma);
1802         if (rc != 0)
1803                 GOTO(out_unlock, rc);
1804
1805         if (!mdt_hsm_release_allow(ma))
1806                 GOTO(out_unlock, rc = -EPERM);
1807
1808         /* already released? */
1809         if (ma->ma_hsm.mh_flags & HS_RELEASED)
1810                 GOTO(out_unlock, rc = 0);
1811
1812         /* Compare on-disk and packed data_version */
1813         if (data->cd_data_version != ma->ma_hsm.mh_arch_ver) {
1814                 CDEBUG(D_HSM, DFID" data_version mismatches: packed=%llu"
1815                        " and on-disk=%llu\n", PFID(mdt_object_fid(o)),
1816                        data->cd_data_version, ma->ma_hsm.mh_arch_ver);
1817                 GOTO(out_unlock, rc = -EPERM);
1818         }
1819
1820         ma->ma_valid = MA_INODE;
1821         ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_SIZE;
1822         rc = mo_attr_set(info->mti_env, mdt_object_child(o), ma);
1823         if (rc < 0)
1824                 GOTO(out_unlock, rc);
1825
1826         mutex_lock(&o->mot_som_mutex);
1827         rc2 = mdt_set_som(info, o, SOM_FL_STRICT, ma->ma_attr.la_size,
1828                            ma->ma_attr.la_blocks);
1829         mutex_unlock(&o->mot_som_mutex);
1830         if (rc2 < 0)
1831                 CDEBUG(D_INODE,
1832                        "%s: File "DFID" SOM update failed: rc = %d\n",
1833                        mdt_obd_name(info->mti_mdt),
1834                        PFID(mdt_object_fid(o)), rc2);
1835
1836
1837         ma->ma_need = MA_INODE | MA_LOV;
1838         rc = mdt_attr_get_complex(info, o, ma);
1839         if (rc < 0)
1840                 GOTO(out_unlock, rc);
1841
1842         if (!(ma->ma_valid & MA_LOV)) {
1843                 /* Even empty file are released */
1844                 memset(ma->ma_lmm, 0, sizeof(*ma->ma_lmm));
1845                 ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEFINED);
1846                 ma->ma_lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1847                 ma->ma_lmm->lmm_stripe_size = cpu_to_le32(LOV_MIN_STRIPE_SIZE);
1848                 ma->ma_lmm_size = sizeof(*ma->ma_lmm);
1849         } else {
1850                 /* Magic must be LOV_MAGIC_*_DEFINED or LOD will interpret
1851                  * ma_lmm as lov_user_md, then it will be confused by union of
1852                  * layout_gen and stripe_offset. */
1853                 if ((le32_to_cpu(ma->ma_lmm->lmm_magic) & LOV_MAGIC_MASK) ==
1854                     LOV_MAGIC_MAGIC)
1855                         ma->ma_lmm->lmm_magic |= cpu_to_le32(LOV_MAGIC_DEFINED);
1856                 else
1857                         GOTO(out_unlock, rc = -EINVAL);
1858         }
1859
1860         /* Set file as released. */
1861         rc = mdt_hsm_set_released(ma->ma_lmm);
1862         if (rc)
1863                 GOTO(out_unlock, rc);
1864
1865         orp_ma = &info->mti_u.hsm.attr;
1866         orp_ma->ma_attr.la_mode = S_IFREG | S_IWUSR;
1867         /* We use root ownership to bypass potential quota
1868          * restrictions on the user and group of the file. */
1869         orp_ma->ma_attr.la_uid = 0;
1870         orp_ma->ma_attr.la_gid = 0;
1871         orp_ma->ma_attr.la_valid = LA_MODE | LA_UID | LA_GID;
1872         orp_ma->ma_lmm = ma->ma_lmm;
1873         orp_ma->ma_lmm_size = ma->ma_lmm_size;
1874         orp_ma->ma_valid = MA_INODE | MA_LOV;
1875         orphan = mdt_orphan_open(info, info->mti_mdt, &data->cd_fid, orp_ma,
1876                                  MDS_FMODE_WRITE);
1877         if (IS_ERR(orphan)) {
1878                 CERROR("%s: cannot open orphan file "DFID": rc = %ld\n",
1879                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid),
1880                        PTR_ERR(orphan));
1881                 GOTO(out_unlock, rc = PTR_ERR(orphan));
1882         }
1883
1884         /* Set up HSM attribute for orphan object */
1885         CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
1886         buf = &info->mti_buf;
1887         buf->lb_buf = info->mti_xattr_buf;
1888         buf->lb_len = sizeof(struct hsm_attrs);
1889         ma->ma_hsm.mh_flags |= HS_RELEASED;
1890         lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1891         ma->ma_hsm.mh_flags &= ~HS_RELEASED;
1892
1893         mdt_lock_reg_init(lh, LCK_EX);
1894         rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_LAYOUT |
1895                              MDS_INODELOCK_XATTR);
1896         if (rc != 0)
1897                 GOTO(out_close, rc);
1898
1899         /* The orphan has root ownership so we need to raise
1900          * CAP_FOWNER to set the HSM attributes. */
1901         cap = uc->uc_cap;
1902         uc->uc_cap |= MD_CAP_TO_MASK(CFS_CAP_FOWNER);
1903         rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
1904                           XATTR_NAME_HSM, 0);
1905         uc->uc_cap = cap;
1906         if (rc != 0)
1907                 GOTO(out_layout_lock, rc);
1908
1909         /* Swap layout with orphan objects. */
1910         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o),
1911                              mdt_object_child(orphan),
1912                              SWAP_LAYOUTS_MDS_HSM);
1913         EXIT;
1914
1915 out_layout_lock:
1916         /* Release exclusive LL */
1917         mdt_object_unlock(info, o, lh, 1);
1918 out_close:
1919         /* Close orphan object anyway */
1920         rc2 = mo_close(info->mti_env, mdt_object_child(orphan), orp_ma,
1921                        MDS_FMODE_WRITE);
1922         if (rc2 < 0)
1923                 CERROR("%s: error closing volatile file "DFID": rc = %d\n",
1924                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid), rc2);
1925         LU_OBJECT_DEBUG(D_HSM, info->mti_env, &orphan->mot_obj,
1926                         "object closed");
1927         mdt_object_put(info->mti_env, orphan);
1928
1929 out_unlock:
1930         up_write(&o->mot_open_sem);
1931
1932         /* already released */
1933         if (rc == 0) {
1934                 struct mdt_body *repbody;
1935
1936                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1937                 LASSERT(repbody != NULL);
1938                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
1939         }
1940
1941 out_reprocess:
1942         ldlm_reprocess_all(lease->l_resource);
1943         LDLM_LOCK_PUT(lease);
1944
1945         ma->ma_valid = 0;
1946         ma->ma_need = 0;
1947
1948         return rc;
1949 }
1950
1951 int mdt_close_handle_layouts(struct mdt_thread_info *info,
1952                              struct mdt_object *o, struct md_attr *ma)
1953 {
1954         struct mdt_lock_handle  *lh1 = &info->mti_lh[MDT_LH_NEW];
1955         struct mdt_lock_handle  *lh2 = &info->mti_lh[MDT_LH_OLD];
1956         struct close_data       *data;
1957         struct ldlm_lock        *lease;
1958         struct mdt_object       *o1 = o, *o2;
1959         bool                     lease_broken;
1960         bool                     swap_objects;
1961         int                      rc;
1962         ENTRY;
1963
1964         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
1965                 RETURN(-EROFS);
1966
1967         if (!S_ISREG(lu_object_attr(&o1->mot_obj)))
1968                 RETURN(-EINVAL);
1969
1970         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1971         if (data == NULL)
1972                 RETURN(-EPROTO);
1973
1974         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1975                 RETURN(-EINVAL);
1976
1977         rc = lu_fid_cmp(&data->cd_fid, mdt_object_fid(o));
1978         if (unlikely(rc == 0))
1979                 RETURN(-EINVAL);
1980
1981         /* Exchange o1 and o2, to enforce locking order */
1982         swap_objects = (rc < 0);
1983
1984         lease = ldlm_handle2lock(&data->cd_handle);
1985         if (lease == NULL)
1986                 RETURN(-ESTALE);
1987
1988         o2 = mdt_object_find(info->mti_env, info->mti_mdt, &data->cd_fid);
1989         if (IS_ERR(o2))
1990                 GOTO(out_lease, rc = PTR_ERR(o2));
1991
1992         if (!S_ISREG(lu_object_attr(&o2->mot_obj))) {
1993                 swap_objects = false; /* not swapped yet */
1994                 GOTO(out_obj, rc = -EINVAL);
1995         }
1996
1997         if (swap_objects)
1998                 swap(o1, o2);
1999
2000         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
2001                            MAY_WRITE);
2002         if (rc < 0)
2003                 GOTO(out_obj, rc);
2004
2005         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
2006                            MAY_WRITE);
2007         if (rc < 0)
2008                 GOTO(out_obj, rc);
2009
2010         /* try to hold open_sem so that nobody else can open the file */
2011         if (!down_write_trylock(&o->mot_open_sem)) {
2012                 ldlm_lock_cancel(lease);
2013                 GOTO(out_obj, rc = -EBUSY);
2014         }
2015
2016         /* Check if the lease open lease has already canceled */
2017         lock_res_and_lock(lease);
2018         lease_broken = ldlm_is_cancel(lease);
2019         unlock_res_and_lock(lease);
2020
2021         LDLM_DEBUG(lease, DFID " lease broken? %d",
2022                    PFID(mdt_object_fid(o)), lease_broken);
2023
2024         /* Cancel server side lease. Client side counterpart should
2025          * have been cancelled. It's okay to cancel it now as we've
2026          * held mot_open_sem. */
2027         ldlm_lock_cancel(lease);
2028
2029         if (lease_broken)
2030                 GOTO(out_unlock_sem, rc = -ESTALE);
2031
2032         mdt_lock_reg_init(lh1, LCK_EX);
2033         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
2034                              MDS_INODELOCK_XATTR);
2035         if (rc < 0)
2036                 GOTO(out_unlock_sem, rc);
2037
2038         mdt_lock_reg_init(lh2, LCK_EX);
2039         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
2040                              MDS_INODELOCK_XATTR);
2041         if (rc < 0)
2042                 GOTO(out_unlock1, rc);
2043
2044         /* Swap layout with orphan object */
2045         if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SWAP) {
2046                 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
2047                                      mdt_object_child(o2), 0);
2048         } else if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_MERGE ||
2049                    ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
2050                 struct lu_buf *buf = &info->mti_buf;
2051                 struct md_rejig_data mrd;
2052
2053                 mrd.mrd_obj = mdt_object_child(o == o1 ? o2 : o1);
2054                 if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT)
2055                         mrd.mrd_mirror_id = data->cd_mirror_id;
2056
2057                 buf->lb_len = sizeof(mrd);
2058                 buf->lb_buf = &mrd;
2059                 rc = mo_xattr_set(info->mti_env, mdt_object_child(o), buf,
2060                                   XATTR_LUSTRE_LOV,
2061                                   ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT ?
2062                                   LU_XATTR_SPLIT : LU_XATTR_MERGE);
2063                 if (rc == 0 && ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2064                         int rc2;
2065
2066                         mutex_lock(&o->mot_som_mutex);
2067                         rc2 = mdt_set_som(info, o, SOM_FL_STRICT,
2068                                           ma->ma_attr.la_size,
2069                                           ma->ma_attr.la_blocks);
2070                         mutex_unlock(&o->mot_som_mutex);
2071                         if (rc2 < 0)
2072                                 CERROR(DFID": Setting i_blocks error: %d, "
2073                                        "i_blocks will be reported wrongly and "
2074                                        "can only be fixed in next resync\n",
2075                                        PFID(mdt_object_fid(o)), rc2);
2076                 }
2077         }
2078         if (rc < 0)
2079                 GOTO(out_unlock2, rc);
2080
2081         EXIT;
2082
2083 out_unlock2:
2084         /* Release exclusive LL */
2085         mdt_object_unlock(info, o2, lh2, 1);
2086
2087 out_unlock1:
2088         mdt_object_unlock(info, o1, lh1, 1);
2089
2090 out_unlock_sem:
2091         up_write(&o->mot_open_sem);
2092
2093         /* already swapped */
2094         if (rc == 0) {
2095                 struct mdt_body *repbody;
2096
2097                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2098                 LASSERT(repbody != NULL);
2099                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2100         }
2101
2102 out_obj:
2103         mdt_object_put(info->mti_env, swap_objects ? o1 : o2);
2104
2105         ldlm_reprocess_all(lease->l_resource);
2106
2107 out_lease:
2108         LDLM_LOCK_PUT(lease);
2109
2110         if (ma != NULL) {
2111                 ma->ma_valid = 0;
2112                 ma->ma_need = 0;
2113         }
2114
2115         return rc;
2116 }
2117
2118 static int mdt_close_resync_done(struct mdt_thread_info *info,
2119                                  struct mdt_object *o, struct md_attr *ma)
2120 {
2121         struct close_data       *data;
2122         struct ldlm_lock        *lease;
2123         struct md_layout_change  layout = { 0 };
2124         __u32                   *resync_ids = NULL;
2125         size_t                   resync_count = 0;
2126         bool                     lease_broken;
2127         int                      rc;
2128         ENTRY;
2129
2130         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2131                 RETURN(-EROFS);
2132
2133         if (!S_ISREG(lu_object_attr(&o->mot_obj)))
2134                 RETURN(-EINVAL);
2135
2136         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2137         if (data == NULL)
2138                 RETURN(-EPROTO);
2139
2140         if (ptlrpc_req_need_swab(mdt_info_req(info)))
2141                 lustre_swab_close_data_resync_done(&data->cd_resync);
2142
2143         if (!fid_is_zero(&data->cd_fid))
2144                 RETURN(-EPROTO);
2145
2146         lease = ldlm_handle2lock(&data->cd_handle);
2147         if (lease == NULL)
2148                 RETURN(-ESTALE);
2149
2150         /* try to hold open_sem so that nobody else can open the file */
2151         if (!down_write_trylock(&o->mot_open_sem)) {
2152                 ldlm_lock_cancel(lease);
2153                 GOTO(out_reprocess, rc = -EBUSY);
2154         }
2155
2156         /* Check if the lease open lease has already canceled */
2157         lock_res_and_lock(lease);
2158         lease_broken = ldlm_is_cancel(lease);
2159         unlock_res_and_lock(lease);
2160
2161         LDLM_DEBUG(lease, DFID " lease broken? %d\n",
2162                    PFID(mdt_object_fid(o)), lease_broken);
2163
2164         /* Cancel server side lease. Client side counterpart should
2165          * have been cancelled. It's okay to cancel it now as we've
2166          * held mot_open_sem. */
2167         ldlm_lock_cancel(lease);
2168
2169         if (lease_broken) /* don't perform release task */
2170                 GOTO(out_unlock, rc = -ESTALE);
2171
2172         resync_count = data->cd_resync.resync_count;
2173
2174         if (resync_count > INLINE_RESYNC_ARRAY_SIZE) {
2175                 void *data;
2176
2177                 if (!req_capsule_has_field(info->mti_pill, &RMF_U32,
2178                                            RCL_CLIENT))
2179                         GOTO(out_unlock, rc = -EPROTO);
2180
2181                 OBD_ALLOC(resync_ids, resync_count * sizeof(__u32));
2182                 if (!resync_ids)
2183                         GOTO(out_unlock, rc = -ENOMEM);
2184
2185                 data = req_capsule_client_get(info->mti_pill, &RMF_U32);
2186                 memcpy(resync_ids, data, resync_count * sizeof(__u32));
2187
2188                 layout.mlc_resync_ids = resync_ids;
2189         } else {
2190                 layout.mlc_resync_ids = data->cd_resync.resync_ids_inline;
2191         }
2192
2193         layout.mlc_opc = MD_LAYOUT_RESYNC_DONE;
2194         layout.mlc_resync_count = resync_count;
2195         if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2196                 layout.mlc_som.lsa_valid = SOM_FL_STRICT;
2197                 layout.mlc_som.lsa_size = ma->ma_attr.la_size;
2198                 layout.mlc_som.lsa_blocks = ma->ma_attr.la_blocks;
2199         }
2200         rc = mdt_layout_change(info, o, &layout);
2201         if (rc)
2202                 GOTO(out_unlock, rc);
2203
2204         EXIT;
2205
2206 out_unlock:
2207         up_write(&o->mot_open_sem);
2208
2209         /* already released */
2210         if (rc == 0) {
2211                 struct mdt_body *repbody;
2212
2213                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2214                 LASSERT(repbody != NULL);
2215                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2216         }
2217
2218         if (resync_ids)
2219                 OBD_FREE(resync_ids, resync_count * sizeof(__u32));
2220
2221 out_reprocess:
2222         ldlm_reprocess_all(lease->l_resource);
2223         LDLM_LOCK_PUT(lease);
2224
2225         ma->ma_valid = 0;
2226         ma->ma_need = 0;
2227
2228         return rc;
2229 }
2230
2231 #define MFD_CLOSED(open_flags) ((open_flags) == MDS_FMODE_CLOSED)
2232
2233 static int mdt_mfd_closed(struct mdt_file_data *mfd)
2234 {
2235         return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_open_flags));
2236 }
2237
2238 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
2239 {
2240         struct mdt_object *o = mfd->mfd_object;
2241         struct md_object *next = mdt_object_child(o);
2242         struct md_attr *ma = &info->mti_attr;
2243         struct lu_fid *ofid = &info->mti_tmp_fid1;
2244         int rc = 0;
2245         u64 open_flags;
2246         u64 intent;
2247         bool discard = false;
2248
2249         ENTRY;
2250
2251         open_flags = mfd->mfd_open_flags;
2252         intent = ma->ma_attr_flags & MDS_CLOSE_INTENT;
2253         *ofid = *mdt_object_fid(o);
2254
2255         CDEBUG(D_INODE, "%s: close file "DFID" with intent: %llx\n",
2256                mdt_obd_name(info->mti_mdt), PFID(ofid), intent);
2257
2258         switch (intent) {
2259         case MDS_HSM_RELEASE: {
2260                 rc = mdt_hsm_release(info, o, ma);
2261                 if (rc < 0) {
2262                         CDEBUG(D_HSM, "%s: File " DFID " release failed: %d\n",
2263                                mdt_obd_name(info->mti_mdt),
2264                                PFID(ofid), rc);
2265                         /* continue to close even error occurred. */
2266                 }
2267                 break;
2268         }
2269         case MDS_CLOSE_LAYOUT_MERGE:
2270         case MDS_CLOSE_LAYOUT_SPLIT:
2271         case MDS_CLOSE_LAYOUT_SWAP: {
2272                 rc = mdt_close_handle_layouts(info, o, ma);
2273                 if (rc < 0) {
2274                         CDEBUG(D_INODE,
2275                                "%s: cannot swap layout of "DFID": rc = %d\n",
2276                                mdt_obd_name(info->mti_mdt),
2277                                PFID(ofid), rc);
2278                         /* continue to close even if error occurred. */
2279                 }
2280                 break;
2281         }
2282         case MDS_CLOSE_RESYNC_DONE:
2283                 rc = mdt_close_resync_done(info, o, ma);
2284                 break;
2285         default:
2286                 /* nothing */
2287                 break;
2288         }
2289
2290         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
2291             ma->ma_attr.la_valid & (LA_LSIZE | LA_LBLOCKS)) {
2292                 int rc2;
2293
2294                 rc2 = mdt_lsom_update(info, o, false);
2295                 if (rc2 < 0)
2296                         CDEBUG(D_INODE,
2297                                "%s: File " DFID " LSOM failed: rc = %d\n",
2298                                mdt_obd_name(info->mti_mdt),
2299                                PFID(ofid), rc2);
2300                         /* continue to close even if error occured. */
2301         }
2302
2303         if (open_flags & MDS_FMODE_WRITE)
2304                 mdt_write_put(o);
2305         else if (open_flags & MDS_FMODE_EXEC)
2306                 mdt_write_allow(o);
2307
2308         /* Update atime on close only. */
2309         if ((open_flags & MDS_FMODE_EXEC || open_flags & MDS_FMODE_READ ||
2310              open_flags & MDS_FMODE_WRITE) && (ma->ma_valid & MA_INODE) &&
2311             (ma->ma_attr.la_valid & LA_ATIME)) {
2312                 /* Set the atime only. */
2313                 ma->ma_valid = MA_INODE;
2314                 ma->ma_attr.la_valid = LA_ATIME;
2315                 rc = mo_attr_set(info->mti_env, next, ma);
2316         }
2317
2318         /* If file data is modified, add the dirty flag. */
2319         if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
2320                 rc = mdt_add_dirty_flag(info, o, ma);
2321
2322         ma->ma_need |= MA_INODE;
2323         ma->ma_valid &= ~MA_INODE;
2324
2325         LASSERT(atomic_read(&o->mot_open_count) > 0);
2326         atomic_dec(&o->mot_open_count);
2327         mdt_handle_last_unlink(info, o, ma);
2328
2329         if (!MFD_CLOSED(open_flags)) {
2330                 rc = mo_close(info->mti_env, next, ma, open_flags);
2331                 discard = mdt_dom_check_for_discard(info, o);
2332         }
2333
2334         /* adjust open and lease count */
2335         if (open_flags & MDS_OPEN_LEASE) {
2336                 LASSERT(atomic_read(&o->mot_lease_count) > 0);
2337                 atomic_dec(&o->mot_lease_count);
2338         }
2339
2340         mdt_mfd_free(mfd);
2341         mdt_object_put(info->mti_env, o);
2342
2343         if (discard)
2344                 mdt_dom_discard_data(info, ofid);
2345
2346         RETURN(rc);
2347 }
2348
2349 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
2350                        struct mdt_body *repbody)
2351 {
2352         struct mdt_export_data *med;
2353         struct mdt_file_data   *mfd;
2354         int                     ret = 0;
2355         int                     rc = 0;
2356         ENTRY;
2357
2358         med = &req->rq_export->exp_mdt_data;
2359         spin_lock(&med->med_open_lock);
2360         mfd = mdt_open_handle2mfd(med, &info->mti_open_handle,
2361                                   req_is_replay(req));
2362         if (mdt_mfd_closed(mfd)) {
2363                 spin_unlock(&med->med_open_lock);
2364                 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
2365                        ": cookie = %#llx\n", PFID(info->mti_rr.rr_fid1),
2366                        info->mti_open_handle.cookie);
2367                 /** not serious error since bug 3633 */
2368                 rc = -ESTALE;
2369         } else {
2370                 class_handle_unhash(&mfd->mfd_open_handle);
2371                 list_del_init(&mfd->mfd_list);
2372                 spin_unlock(&med->med_open_lock);
2373                 ret = mdt_mfd_close(info, mfd);
2374         }
2375
2376         RETURN(rc ? rc : ret);
2377 }
2378
2379 int mdt_close(struct tgt_session_info *tsi)
2380 {
2381         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2382         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2383         struct md_attr         *ma = &info->mti_attr;
2384         struct mdt_body        *repbody = NULL;
2385         int rc, ret = 0;
2386         ENTRY;
2387
2388         mdt_counter_incr(req, LPROC_MDT_CLOSE);
2389         /* Close may come with the Size-on-MDS update. Unpack it. */
2390         rc = mdt_close_unpack(info);
2391         if (rc)
2392                 GOTO(out, rc = err_serious(rc));
2393
2394         /* These fields are no longer used and are left for compatibility.
2395          * size is always zero */
2396         req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
2397                              0);
2398         req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
2399                              0);
2400         rc = req_capsule_server_pack(info->mti_pill);
2401         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
2402                 mdt_client_compatibility(info);
2403                 if (rc == 0)
2404                         mdt_fix_reply(info);
2405                 mdt_exit_ucred(info);
2406                 GOTO(out, rc = lustre_msg_get_status(req->rq_repmsg));
2407         }
2408
2409         /* Continue to close handle even if we can not pack reply */
2410         if (rc == 0) {
2411                 repbody = req_capsule_server_get(info->mti_pill,
2412                                                  &RMF_MDT_BODY);
2413                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
2414                                                     &RMF_MDT_MD);
2415                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
2416                                                        &RMF_MDT_MD,
2417                                                        RCL_SERVER);
2418                 ma->ma_need = MA_INODE | MA_LOV;
2419                 repbody->mbo_eadatasize = 0;
2420                 repbody->mbo_aclsize = 0;
2421         } else {
2422                 rc = err_serious(rc);
2423         }
2424
2425         rc = mdt_close_internal(info, req, repbody);
2426         if (rc != -ESTALE)
2427                 mdt_empty_transno(info, rc);
2428
2429         if (repbody != NULL) {
2430                 mdt_client_compatibility(info);
2431                 rc = mdt_fix_reply(info);
2432         }
2433
2434         mdt_exit_ucred(info);
2435         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
2436                 GOTO(out, rc = err_serious(-ENOMEM));
2437
2438         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
2439                                  OBD_FAIL_MDS_CLOSE_NET_REP))
2440                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
2441 out:
2442         mdt_thread_info_fini(info);
2443         RETURN(rc ? rc : ret);
2444 }