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_LDLM_REPLY | OBD_FAIL_ONCE)) {
 602                 RETURN(-EAGAIN);
 603         }
 604
 605         mfd = NULL;
 606         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 607                 spin_lock(&med->med_open_lock);
 608                 list_for_each(t, &med->med_open_head) {
 609                         mfd = list_entry(t, struct mdt_file_data, mfd_list);
 610                         if (mfd->mfd_xid == req->rq_xid)
 611                                 break;
 612                         mfd = NULL;
 613                 }
 614                 spin_unlock(&med->med_open_lock);
 615
 616                 if (mfd != NULL) {
 617                         repbody->mbo_open_handle.cookie =
 618                                 mfd->mfd_open_handle.h_cookie;
 619                         /* set repbody->ea_size for resent case */
 620                         if (ma->ma_valid & MA_LOV) {
 621                                 LASSERT(ma->ma_lmm_size != 0);
 622                                 repbody->mbo_eadatasize = ma->ma_lmm_size;
 623                                 if (isdir)
 624                                         repbody->mbo_valid |= OBD_MD_FLDIREA;
 625                                 else
 626                                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
 627                         }
 628                         mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 629                         RETURN(0);
 630                 }
 631         }
 632
 633         rc = mdt_mfd_open(info, p, o, open_flags, created, rep);
 634         if (!rc)
 635                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 636
 637         RETURN(rc);
 638 }
 639
 640 void mdt_reconstruct_open(struct mdt_thread_info *info,
 641                           struct mdt_lock_handle *lhc)
 642 {
 643         const struct lu_env *env = info->mti_env;
 644         struct mdt_device *mdt = info->mti_mdt;
 645         struct req_capsule *pill = info->mti_pill;
 646         struct ptlrpc_request *req = mdt_info_req(info);
 647         struct md_attr *ma = &info->mti_attr;
 648         struct mdt_reint_record *rr = &info->mti_rr;
 649         u64 open_flags = info->mti_spec.sp_cr_flags;
 650         struct ldlm_reply *ldlm_rep;
 651         struct mdt_object *parent;
 652         struct mdt_object *child;
 653         struct mdt_body *repbody;
 654         u64 opdata;
 655         int rc;
 656         ENTRY;
 657
 658         LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
 659         ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
 660         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 661
 662         ma->ma_need = MA_INODE | MA_HSM;
 663         ma->ma_valid = 0;
 664
 665         opdata = mdt_req_from_lrd(req, info->mti_reply_data);
 666         mdt_set_disposition(info, ldlm_rep, opdata);
 667
 668         CDEBUG(D_INODE, "This is reconstruct open: disp=%#llx, result=%d\n",
 669                ldlm_rep->lock_policy_res1, req->rq_status);
 670
 671         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
 672             req->rq_status != 0)
 673                 /* We did not create successfully, return error to client. */
 674                 GOTO(out, rc = req->rq_status);
 675
 676         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
 677                 struct obd_export *exp = req->rq_export;
 678                 /*
 679                  * We failed after creation, but we do not know in which step
 680                  * we failed. So try to check the child object.
 681                  */
 682                 parent = mdt_object_find(env, mdt, rr->rr_fid1);
 683                 if (IS_ERR(parent)) {
 684                         rc = PTR_ERR(parent);
 685                         LCONSOLE_WARN("Parent "DFID" lookup error %d."
 686                                       " Evicting client %s with export %s.\n",
 687                                       PFID(rr->rr_fid1), rc,
 688                                       obd_uuid2str(&exp->exp_client_uuid),
 689                                       obd_export_nid2str(exp));
 690                         mdt_export_evict(exp);
 691                         RETURN_EXIT;
 692                 }
 693
 694                 child = mdt_object_find(env, mdt, rr->rr_fid2);
 695                 if (IS_ERR(child)) {
 696                         rc = PTR_ERR(child);
 697                         LCONSOLE_WARN("cannot lookup child "DFID": rc = %d; "
 698                                       "evicting client %s with export %s\n",
 699                                       PFID(rr->rr_fid2), rc,
 700                                       obd_uuid2str(&exp->exp_client_uuid),
 701                                       obd_export_nid2str(exp));
 702                         mdt_object_put(env, parent);
 703                         mdt_export_evict(exp);
 704                         RETURN_EXIT;
 705                 }
 706
 707                 if (unlikely(mdt_object_remote(child))) {
 708                         /* the child object was created on remote server */
 709                         if (!mdt_is_dne_client(exp)) {
 710                                 /* Return -EIO for old client */
 711                                 mdt_object_put(env, parent);
 712                                 mdt_object_put(env, child);
 713                                 GOTO(out, rc = -EIO);
 714                         }
 715                         repbody->mbo_fid1 = *rr->rr_fid2;
 716                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 717                         rc = 0;
 718                 } else {
 719                         if (mdt_object_exists(child)) {
 720                                 mdt_prep_ma_buf_from_rep(info, child, ma);
 721                                 rc = mdt_attr_get_complex(info, child, ma);
 722                                 if (rc == 0)
 723                                         rc = mdt_finish_open(info, parent,
 724                                                              child, open_flags,
 725                                                              1, ldlm_rep);
 726                         } else {
 727                                 /* the child does not exist, we should do
 728                                  * regular open */
 729                                 mdt_object_put(env, parent);
 730                                 mdt_object_put(env, child);
 731                                 GOTO(regular_open, 0);
 732                         }
 733                 }
 734                 mdt_object_put(env, parent);
 735                 mdt_object_put(env, child);
 736                 GOTO(out, rc);
 737         } else {
 738 regular_open:
 739                 /* We did not try to create, so we are a pure open */
 740                 rc = mdt_reint_open(info, lhc);
 741         }
 742
 743         EXIT;
 744 out:
 745         req->rq_status = rc;
 746         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
 747         LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
 748 }
 749
 750 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep)
 751 {
 752         u64 open_flags = info->mti_spec.sp_cr_flags;
 753         struct mdt_reint_record *rr = &info->mti_rr;
 754         struct md_attr *ma = &info->mti_attr;
 755         struct mdt_object *o;
 756         int rc;
 757
 758         ENTRY;
 759         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
 760         if (IS_ERR(o))
 761                 RETURN(rc = PTR_ERR(o));
 762
 763         if (unlikely(mdt_object_remote(o))) {
 764                 /* the child object was created on remote server */
 765                 struct mdt_body *repbody;
 766
 767                 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 768                                                 DISP_LOOKUP_EXECD |
 769                                                 DISP_LOOKUP_POS));
 770                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 771                 repbody->mbo_fid1 = *rr->rr_fid2;
 772                 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 773                 rc = 0;
 774         } else {
 775                 if (mdt_object_exists(o)) {
 776                         mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 777                                                         DISP_LOOKUP_EXECD |
 778                                                         DISP_LOOKUP_POS));
 779                         mdt_prep_ma_buf_from_rep(info, o, ma);
 780                         rc = mdt_attr_get_complex(info, o, ma);
 781                         if (rc == 0)
 782                                 rc = mdt_finish_open(info, NULL, o, open_flags,
 783                                                      0, rep);
 784                 } else {
 785                         rc = -ENOENT;
 786                 }
 787         }
 788
 789         mdt_object_put(info->mti_env, o);
 790         RETURN(rc);
 791 }
 792
 793 /* lock object for open */
 794 static int mdt_object_open_lock(struct mdt_thread_info *info,
 795                                 struct mdt_object *obj,
 796                                 struct mdt_lock_handle *lhc,
 797                                 __u64 *ibits)
 798 {
 799         struct md_attr *ma = &info->mti_attr;
 800         __u64 open_flags = info->mti_spec.sp_cr_flags;
 801         __u64 trybits = 0;
 802         enum ldlm_mode lm = LCK_CR;
 803         bool acq_lease = !!(open_flags & MDS_OPEN_LEASE);
 804         bool try_layout = false;
 805         bool create_layout = false;
 806         int rc = 0;
 807         int dom_stripes = LMM_NO_DOM;
 808         bool dom_lock = false;
 809
 810         ENTRY;
 811
 812         *ibits = 0;
 813         mdt_lock_handle_init(lhc);
 814
 815         if (req_is_replay(mdt_info_req(info)))
 816                 RETURN(0);
 817
 818         if (S_ISREG(lu_object_attr(&obj->mot_obj))) {
 819                 if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV) &&
 820                     md_should_create(open_flags))
 821                         create_layout = true;
 822                 if (exp_connect_layout(info->mti_exp) && !create_layout &&
 823                     ma->ma_need & MA_LOV)
 824                         try_layout = true;
 825
 826                 /* DoM files can have just MDT stripe or combined MDT + OST
 827                  * stripes.
 828                  * - In the first case the open for read/write will do IO to
 829                  *   the MDT stripe and it makes sense to take IO lock in
 830                  *   advance along with OPEN even if it is blocking lock.
 831                  * - In the second case it is just size of MDT stripe and it
 832                  *   is quite unlikely that client will write into it, though
 833                  *   it may read it. So IO lock will be taken optionally if it
 834                  *   is non-blocking one.
 835                  */
 836                 if (ma->ma_valid & MA_LOV && ma->ma_lmm != NULL)
 837                         dom_stripes = mdt_lmm_dom_entry(ma->ma_lmm);
 838
 839                 if (dom_stripes == LMM_DOM_ONLY &&
 840                     info->mti_mdt->mdt_opts.mo_dom_lock > 0 &&
 841                     !mdt_dom_client_has_lock(info, mdt_object_fid(obj)))
 842                         dom_lock = true;
 843         }
 844
 845         if (acq_lease) {
 846                 /* lease open, acquire write mode of open sem */
 847                 down_write(&obj->mot_open_sem);
 848
 849                 /* Lease exists and ask for new lease */
 850                 if (atomic_read(&obj->mot_lease_count) > 0) {
 851                         /* only exclusive open is supported, so lease
 852                          * are conflicted to each other */
 853                         GOTO(out, rc = -EBUSY);
 854                 }
 855
 856                 /* Lease must be with open lock */
 857                 if (!(open_flags & MDS_OPEN_LOCK)) {
 858                         CERROR("%s: Request lease for file:"DFID ", but open lock "
 859                                "is missed, open_flags = %#llo : rc = %d\n",
 860                                mdt_obd_name(info->mti_mdt),
 861                                PFID(mdt_object_fid(obj)), open_flags, -EPROTO);
 862                         GOTO(out, rc = -EPROTO);
 863                 }
 864
 865                 /* XXX: only exclusive open is supported. */
 866                 lm = LCK_EX;
 867                 *ibits = MDS_INODELOCK_OPEN;
 868
 869                 /* never grant LCK_EX layout lock to client */
 870                 try_layout = false;
 871         } else { /* normal open */
 872                 /* normal open holds read mode of open sem */
 873                 down_read(&obj->mot_open_sem);
 874
 875                 if (open_flags & MDS_OPEN_LOCK) {
 876                         if (open_flags & MDS_FMODE_WRITE)
 877                                 lm = LCK_CW;
 878                         else if (open_flags & MDS_FMODE_EXEC)
 879                                 lm = LCK_PR;
 880                         else
 881                                 lm = LCK_CR;
 882
 883                         *ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN;
 884                 } else if (atomic_read(&obj->mot_lease_count) > 0) {
 885                         if (open_flags & MDS_FMODE_WRITE)
 886                                 lm = LCK_CW;
 887                         else
 888                                 lm = LCK_CR;
 889
 890                         /* revoke lease */
 891                         *ibits = MDS_INODELOCK_OPEN;
 892                         try_layout = false;
 893
 894                         lhc = &info->mti_lh[MDT_LH_LOCAL];
 895                 } else if (dom_lock) {
 896                         lm = (open_flags & MDS_FMODE_WRITE) ? LCK_PW : LCK_PR;
 897                         if (info->mti_mdt->mdt_opts.mo_dom_lock ==
 898                             TRYLOCK_DOM_ON_OPEN) {
 899                                 trybits |= MDS_INODELOCK_DOM |
 900                                            MDS_INODELOCK_LAYOUT;
 901                         } else {
 902                                 /* mo_dom_lock == ALWAYS_DOM_LOCK_ON_OPEN */
 903                                 *ibits = MDS_INODELOCK_DOM;
 904                                 if (info->mti_mdt->mdt_opts.mo_dom_read_open) {
 905                                         trybits |= MDS_INODELOCK_LAYOUT;
 906                                 }
 907                         }
 908                 }
 909
 910                 CDEBUG(D_INODE, "normal open:"DFID" lease count: %d, lm: %d\n",
 911                         PFID(mdt_object_fid(obj)),
 912                         atomic_read(&obj->mot_open_count), lm);
 913         }
 914
 915         mdt_lock_reg_init(lhc, lm);
 916
 917         /* Return lookup lock to validate inode at the client side.
 918          * This is pretty important otherwise MDT will return layout
 919          * lock for each open.
 920          * However this is a double-edged sword because changing
 921          * permission will revoke a huge number of LOOKUP locks.
 922          */
 923         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_OPEN) && try_layout) {
 924                 if (!(*ibits & MDS_INODELOCK_LOOKUP))
 925                         trybits |= MDS_INODELOCK_LOOKUP;
 926                 trybits |= MDS_INODELOCK_LAYOUT;
 927         }
 928
 929         if (*ibits | trybits)
 930                 rc = mdt_object_lock_try(info, obj, lhc, ibits, trybits, false);
 931
 932         CDEBUG(D_INODE, "%s: Requested bits lock:"DFID ", ibits = %#llx/%#llx"
 933                ", open_flags = %#llo, try_layout = %d : rc = %d\n",
 934                mdt_obd_name(info->mti_mdt), PFID(mdt_object_fid(obj)),
 935                *ibits, trybits, open_flags, try_layout, rc);
 936
 937         /* will change layout, revoke layout locks by enqueuing EX lock. */
 938         if (rc == 0 && create_layout) {
 939                 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LAYOUT];
 940
 941                 CDEBUG(D_INODE, "Will create layout, get EX layout lock:"DFID
 942                         ", open_flags = %#llo\n",
 943                         PFID(mdt_object_fid(obj)), open_flags);
 944
 945                 /* We cannot enqueue another lock for the same resource we
 946                  * already have a lock for, due to mechanics of waiting list
 947                  * iterating in ldlm, see LU-3601.
 948                  * As such we'll drop the open lock we just got above here,
 949                  * it's ok not to have this open lock as it's main purpose is to
 950                  * flush unused cached client open handles. */
 951                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
 952                         mdt_object_unlock(info, obj, lhc, 1);
 953
 954                 LASSERT(!try_layout);
 955                 mdt_lock_handle_init(ll);
 956                 mdt_lock_reg_init(ll, LCK_EX);
 957                 rc = mdt_object_lock(info, obj, ll, MDS_INODELOCK_LAYOUT);
 958
 959                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LL_BLOCK, 2);
 960         }
 961
 962         /* Check if there is any other open handles after acquiring
 963          * open lock. At this point, caching open handles have been revoked
 964          * by open lock.
 965          * XXX: Now only exclusive open is supported. Need to check the
 966          * type of open for generic lease support. */
 967         if (rc == 0 && acq_lease) {
 968                 struct ptlrpc_request *req = mdt_info_req(info);
 969                 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
 970                 struct mdt_file_data *mfd;
 971                 bool is_replay_or_resent;
 972                 int open_count = 0;
 973
 974                 /* For lease: application can open a file and then apply lease,
 975                  * @handle contains original open handle in that case.
 976                  * In recovery, open REQ will be replayed and the lease REQ may
 977                  * be resent that means the open handle is already stale, so we
 978                  * need to fix it up here by finding new handle. */
 979                 is_replay_or_resent = req_is_replay(req) ||
 980                         lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
 981
 982                 /* if the request is _not_ a replay request, rr_open_handle
 983                  * may be used to hold an open file handle which is issuing the
 984                  * lease request, so that this openhandle doesn't count. */
 985                 mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
 986                                           is_replay_or_resent);
 987                 if (mfd != NULL)
 988                         ++open_count;
 989
 990                 CDEBUG(D_INODE, "acq_lease "DFID": openers: %d, want: %d\n",
 991                         PFID(mdt_object_fid(obj)),
 992                         atomic_read(&obj->mot_open_count), open_count);
 993
 994                 if (atomic_read(&obj->mot_open_count) > open_count)
 995                         GOTO(out, rc = -EBUSY);
 996         }
 997         GOTO(out, rc);
 998
 999 out:
1000         RETURN(rc);
1001 }
1002
1003 static void mdt_object_open_unlock(struct mdt_thread_info *info,
1004                                    struct mdt_object *obj,
1005                                    struct mdt_lock_handle *lhc,
1006                                    __u64 ibits, int rc)
1007 {
1008         __u64 open_flags = info->mti_spec.sp_cr_flags;
1009         struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LOCAL];
1010         ENTRY;
1011
1012         if (req_is_replay(mdt_info_req(info)))
1013                 RETURN_EXIT;
1014
1015         /* Release local lock - the lock put in MDT_LH_LOCAL will never
1016          * return to client side. */
1017         if (lustre_handle_is_used(&ll->mlh_reg_lh))
1018                 mdt_object_unlock(info, obj, ll, 1);
1019
1020         ll = &info->mti_lh[MDT_LH_LAYOUT];
1021         /* Release local layout lock, layout was created */
1022         if (lustre_handle_is_used(&ll->mlh_reg_lh)) {
1023                 LASSERT(!(ibits & MDS_INODELOCK_LAYOUT));
1024                 mdt_object_unlock(info, obj, ll, 1);
1025         }
1026
1027         if (open_flags & MDS_OPEN_LEASE)
1028                 up_write(&obj->mot_open_sem);
1029         else
1030                 up_read(&obj->mot_open_sem);
1031
1032         /* Cross-ref case, the lock should be returned to the client */
1033         if (ibits == 0 || rc == -MDT_EREMOTE_OPEN)
1034                 RETURN_EXIT;
1035
1036         if (!(open_flags & MDS_OPEN_LOCK) && !(ibits & MDS_INODELOCK_LAYOUT) &&
1037             !(ibits & MDS_INODELOCK_DOM)) {
1038                 /* for the open request, the lock will only return to client
1039                  * if open or layout lock is granted. */
1040                 rc = 1;
1041         }
1042
1043         if (rc != 0 || !lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1044                 struct ldlm_reply       *ldlm_rep;
1045
1046                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1047                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1048                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
1049                         mdt_object_unlock(info, obj, lhc, 1);
1050         }
1051         RETURN_EXIT;
1052 }
1053
1054 /**
1055  * Check release is permitted for the current HSM flags.
1056  */
1057 static bool mdt_hsm_release_allow(const struct md_attr *ma)
1058 {
1059         if (!(ma->ma_valid & MA_HSM))
1060                 return false;
1061
1062         if (ma->ma_hsm.mh_flags & (HS_DIRTY|HS_NORELEASE|HS_LOST))
1063                 return false;
1064
1065         if (!(ma->ma_hsm.mh_flags & HS_ARCHIVED))
1066                 return false;
1067
1068         return true;
1069 }
1070
1071 static int mdt_open_by_fid_lock(struct mdt_thread_info *info,
1072                                 struct ldlm_reply *rep,
1073                                 struct mdt_lock_handle *lhc)
1074 {
1075         const struct lu_env *env = info->mti_env;
1076         struct mdt_device *mdt = info->mti_mdt;
1077         u64 open_flags = info->mti_spec.sp_cr_flags;
1078         struct mdt_reint_record *rr = &info->mti_rr;
1079         struct md_attr *ma = &info->mti_attr;
1080         struct mdt_object *parent = NULL;
1081         struct mdt_object *o;
1082         bool object_locked = false;
1083         u64 ibits = 0;
1084         int rc;
1085
1086         ENTRY;
1087         if (md_should_create(open_flags)) {
1088                 if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1089                         parent = mdt_object_find(env, mdt, rr->rr_fid1);
1090                         if (IS_ERR(parent)) {
1091                                 CDEBUG(D_INODE, "Fail to find parent "DFID
1092                                        " for anonymous created %ld, try to"
1093                                        " use server-side parent.\n",
1094                                        PFID(rr->rr_fid1), PTR_ERR(parent));
1095                                 parent = NULL;
1096                         }
1097                 }
1098                 if (parent == NULL)
1099                         ma->ma_need |= MA_PFID;
1100         }
1101
1102         o = mdt_object_find(env, mdt, rr->rr_fid2);
1103         if (IS_ERR(o))
1104                 GOTO(out_parent_put, rc = PTR_ERR(o));
1105
1106         if (mdt_object_remote(o)) {
1107                 CDEBUG(D_INFO, "%s: "DFID" is on remote MDT.\n",
1108                        mdt_obd_name(info->mti_mdt),
1109                        PFID(rr->rr_fid2));
1110                 GOTO(out, rc = -EREMOTE);
1111         } else if (!mdt_object_exists(o)) {
1112                 mdt_set_disposition(info, rep,
1113                                     DISP_IT_EXECD |
1114                                     DISP_LOOKUP_EXECD |
1115                                     DISP_LOOKUP_NEG);
1116                 GOTO(out, rc = -ENOENT);
1117         }
1118
1119         mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1120
1121         mdt_prep_ma_buf_from_rep(info, o, ma);
1122         if (open_flags & MDS_OPEN_RELEASE)
1123                 ma->ma_need |= MA_HSM;
1124         rc = mdt_attr_get_complex(info, o, ma);
1125         if (rc)
1126                 GOTO(out, rc);
1127
1128         /* We should not change file's existing LOV EA */
1129         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
1130             open_flags & MDS_OPEN_HAS_EA && ma->ma_valid & MA_LOV)
1131                 GOTO(out, rc = -EEXIST);
1132
1133         /* If a release request, check file open flags are fine and ask for an
1134          * exclusive open access. */
1135         if (open_flags & MDS_OPEN_RELEASE && !mdt_hsm_release_allow(ma))
1136                 GOTO(out, rc = -EPERM);
1137
1138         rc = mdt_check_resent_lock(info, o, lhc);
1139         if (rc < 0) {
1140                 GOTO(out, rc);
1141         } else if (rc > 0) {
1142                 rc = mdt_object_open_lock(info, o, lhc, &ibits);
1143                 object_locked = true;
1144                 if (rc)
1145                         GOTO(out_unlock, rc);
1146         }
1147
1148         if (ma->ma_valid & MA_PFID) {
1149                 parent = mdt_object_find(env, mdt, &ma->ma_pfid);
1150                 if (IS_ERR(parent)) {
1151                         CDEBUG(D_INODE, "Fail to find parent "DFID
1152                                " for anonymous created %ld, try to"
1153                                " use system default.\n",
1154                                PFID(&ma->ma_pfid), PTR_ERR(parent));
1155                         parent = NULL;
1156                 }
1157         }
1158
1159         rc = mdt_finish_open(info, parent, o, open_flags, 0, rep);
1160         if (!rc) {
1161                 mdt_set_disposition(info, rep, DISP_LOOKUP_POS);
1162                 if (open_flags & MDS_OPEN_LOCK)
1163                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1164                 if (open_flags & MDS_OPEN_LEASE)
1165                         mdt_set_disposition(info, rep, DISP_OPEN_LEASE);
1166         }
1167         GOTO(out_unlock, rc);
1168
1169 out_unlock:
1170         if (object_locked)
1171                 mdt_object_open_unlock(info, o, lhc, ibits, rc);
1172 out:
1173         mdt_object_put(env, o);
1174         if (rc == 0)
1175                 mdt_pack_size2body(info, rr->rr_fid2, &lhc->mlh_reg_lh);
1176 out_parent_put:
1177         if (parent != NULL)
1178                 mdt_object_put(env, parent);
1179         return rc;
1180 }
1181
1182 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1183 static int mdt_cross_open(struct mdt_thread_info *info,
1184                           const struct lu_fid *parent_fid,
1185                           const struct lu_fid *fid,
1186                           struct ldlm_reply *rep, u64 open_flags)
1187 {
1188         struct md_attr *ma = &info->mti_attr;
1189         struct mdt_object *o;
1190         int rc;
1191         ENTRY;
1192
1193         o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1194         if (IS_ERR(o))
1195                 RETURN(rc = PTR_ERR(o));
1196
1197         if (mdt_object_remote(o)) {
1198                 /* Something is wrong here, the object is on another MDS! */
1199                 CERROR("%s: "DFID" isn't on this server!: rc = %d\n",
1200                        mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1201                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1202                                 &o->mot_obj,
1203                                 "Object isn't on this server! FLD error?");
1204                 rc = -EFAULT;
1205         } else {
1206                 if (mdt_object_exists(o)) {
1207                         int mask;
1208
1209                         /* Do permission check for cross-open after converting
1210                          * MDS_OPEN_* flags to MAY_* permission mask.
1211                          */
1212                         mask = mds_accmode(open_flags);
1213
1214                         rc = mo_permission(info->mti_env, NULL,
1215                                            mdt_object_child(o), NULL, mask);
1216                         if (rc)
1217                                 goto out;
1218
1219                         mdt_prep_ma_buf_from_rep(info, o, ma);
1220                         rc = mdt_attr_get_complex(info, o, ma);
1221                         if (rc != 0)
1222                                 GOTO(out, rc);
1223
1224                         rc = mdt_finish_open(info, NULL, o, open_flags, 0, rep);
1225                 } else {
1226                         /*
1227                          * Something is wrong here. lookup was positive but
1228                          * there is no object!
1229                          */
1230                         CERROR("%s: "DFID" doesn't exist!: rc = %d\n",
1231                               mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1232                         rc = -EFAULT;
1233                 }
1234         }
1235 out:
1236         mdt_object_put(info->mti_env, o);
1237
1238         RETURN(rc);
1239 }
1240
1241 /*
1242  * find root object and take its xattr lock if it's on remote MDT, later create
1243  * may use fs default striping (which is stored in root xattr).
1244  */
1245 static int mdt_lock_root_xattr(struct mdt_thread_info *info,
1246                                struct mdt_device *mdt)
1247 {
1248         struct mdt_object *md_root = mdt->mdt_md_root;
1249         struct lustre_handle lhroot;
1250         int rc;
1251
1252         if (md_root == NULL) {
1253                 lu_root_fid(&info->mti_tmp_fid1);
1254                 md_root = mdt_object_find(info->mti_env, mdt,
1255                                           &info->mti_tmp_fid1);
1256                 if (IS_ERR(md_root))
1257                         return PTR_ERR(md_root);
1258
1259                 spin_lock(&mdt->mdt_lock);
1260                 if (mdt->mdt_md_root != NULL) {
1261                         spin_unlock(&mdt->mdt_lock);
1262
1263                         LASSERTF(mdt->mdt_md_root == md_root,
1264                                  "Different root object ("
1265                                  DFID") instances, %p, %p\n",
1266                                  PFID(&info->mti_tmp_fid1),
1267                                  mdt->mdt_md_root, md_root);
1268                         LASSERT(atomic_read(
1269                                 &md_root->mot_obj.lo_header->loh_ref) > 1);
1270
1271                         mdt_object_put(info->mti_env, md_root);
1272                 } else {
1273                         mdt->mdt_md_root = md_root;
1274                         spin_unlock(&mdt->mdt_lock);
1275                 }
1276         }
1277
1278         if (md_root->mot_cache_attr || !mdt_object_remote(md_root))
1279                 return 0;
1280
1281         rc = mdt_remote_object_lock(info, md_root, mdt_object_fid(md_root),
1282                                     &lhroot, LCK_PR, MDS_INODELOCK_XATTR,
1283                                     true);
1284         if (rc < 0)
1285                 return rc;
1286
1287         md_root->mot_cache_attr = 1;
1288
1289         /* don't cancel this lock, so that we know the cached xattr is valid. */
1290         ldlm_lock_decref(&lhroot, LCK_PR);
1291
1292         return 0;
1293 }
1294
1295 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1296 {
1297         struct mdt_device *mdt = info->mti_mdt;
1298         struct ptlrpc_request *req = mdt_info_req(info);
1299         struct mdt_object *parent;
1300         struct mdt_object *child;
1301         struct mdt_lock_handle *lh;
1302         struct ldlm_reply *ldlm_rep;
1303         struct mdt_body *repbody;
1304         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1305         struct md_attr *ma = &info->mti_attr;
1306         u64 open_flags = info->mti_spec.sp_cr_flags;
1307         u64 ibits = 0;
1308         struct mdt_reint_record *rr = &info->mti_rr;
1309         int result, rc;
1310         int created = 0;
1311         int object_locked = 0;
1312         u32 msg_flags;
1313
1314         ENTRY;
1315         OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1316                                (obd_timeout + 1) / 4);
1317
1318         mdt_counter_incr(req, LPROC_MDT_OPEN);
1319         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1320
1321         ma->ma_need = MA_INODE;
1322         ma->ma_valid = 0;
1323
1324         LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1325         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1326
1327         if (unlikely(open_flags & MDS_OPEN_JOIN_FILE)) {
1328                 CERROR("file join is not supported anymore.\n");
1329                 GOTO(out, result = err_serious(-EOPNOTSUPP));
1330         }
1331         msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1332
1333         if ((open_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1334             info->mti_spec.u.sp_ea.eadata == NULL)
1335                 GOTO(out, result = err_serious(-EINVAL));
1336
1337         if (open_flags & MDS_FMODE_WRITE &&
1338             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1339                 GOTO(out, result = -EROFS);
1340
1341         CDEBUG(D_INODE, "I am going to open "DFID"/("DNAME"->"DFID") "
1342                "cr_flag=%#llo mode=0%06o msg_flag=0x%x\n",
1343                PFID(rr->rr_fid1), PNAME(&rr->rr_name), PFID(rr->rr_fid2),
1344                open_flags, ma->ma_attr.la_mode, msg_flags);
1345
1346         if (info->mti_cross_ref) {
1347                 /* This is cross-ref open */
1348                 mdt_set_disposition(info, ldlm_rep,
1349                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD |
1350                              DISP_LOOKUP_POS));
1351                 result = mdt_cross_open(info, rr->rr_fid2, rr->rr_fid1,
1352                                         ldlm_rep, open_flags);
1353                 GOTO(out, result);
1354         } else if (req_is_replay(req)) {
1355                 result = mdt_open_by_fid(info, ldlm_rep);
1356
1357                 if (result != -ENOENT)
1358                         GOTO(out, result);
1359
1360                 /* We didn't find the correct object, so we need to re-create it
1361                  * via a regular replay. */
1362                 if (!(open_flags & MDS_OPEN_CREAT)) {
1363                         DEBUG_REQ(D_ERROR, req,
1364                                   "OPEN & CREAT not in open replay/by_fid.");
1365                         GOTO(out, result = -EFAULT);
1366                 }
1367                 CDEBUG(D_INFO, "No object(1), continue as regular open.\n");
1368         } else if (open_flags & (MDS_OPEN_BY_FID | MDS_OPEN_LOCK)) {
1369                 /*
1370                  * MDS_OPEN_LOCK is checked for backward compatibility with 2.1
1371                  * client.
1372                  */
1373                 result = mdt_open_by_fid_lock(info, ldlm_rep, lhc);
1374                 if (result < 0)
1375                         CDEBUG(D_INFO, "no object for "DFID": %d\n",
1376                                PFID(rr->rr_fid2), result);
1377                 GOTO(out, result);
1378         }
1379
1380         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1381                 GOTO(out, result = err_serious(-ENOMEM));
1382
1383         mdt_set_disposition(info, ldlm_rep,
1384                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1385
1386         if (!lu_name_is_valid(&rr->rr_name))
1387                 GOTO(out, result = -EPROTO);
1388
1389         result = mdt_lock_root_xattr(info, mdt);
1390         if (result < 0)
1391                 GOTO(out, result);
1392
1393 again:
1394         lh = &info->mti_lh[MDT_LH_PARENT];
1395         mdt_lock_pdo_init(lh, (open_flags & MDS_OPEN_CREAT) ? LCK_PW : LCK_PR,
1396                           &rr->rr_name);
1397
1398         parent = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
1399         if (IS_ERR(parent))
1400                 GOTO(out, result = PTR_ERR(parent));
1401
1402         result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
1403         if (result != 0) {
1404                 mdt_object_put(info->mti_env, parent);
1405                 GOTO(out, result);
1406         }
1407
1408         /* get and check version of parent */
1409         result = mdt_version_get_check(info, parent, 0);
1410         if (result)
1411                 GOTO(out_parent, result);
1412
1413         fid_zero(child_fid);
1414
1415         result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1416                             &rr->rr_name, child_fid, &info->mti_spec);
1417
1418         LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1419                  "looking for "DFID"/"DNAME", found FID = "DFID"\n",
1420                  PFID(mdt_object_fid(parent)), PNAME(&rr->rr_name),
1421                  PFID(child_fid));
1422
1423         if (result != 0 && result != -ENOENT && result != -ESTALE)
1424                 GOTO(out_parent, result);
1425
1426         if (result == -ENOENT || result == -ESTALE) {
1427                 /* If the object is dead, let's check if the object
1428                  * is being migrated to a new object */
1429                 if (result == -ESTALE) {
1430                         struct lu_buf lmv_buf;
1431
1432                         lmv_buf.lb_buf = info->mti_xattr_buf;
1433                         lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
1434                         rc = mo_xattr_get(info->mti_env,
1435                                           mdt_object_child(parent),
1436                                           &lmv_buf, XATTR_NAME_LMV);
1437                         if (rc > 0) {
1438                                 struct lmv_mds_md_v1 *lmv;
1439
1440                                 lmv = lmv_buf.lb_buf;
1441                                 if (le32_to_cpu(lmv->lmv_hash_type) &
1442                                                 LMV_HASH_FLAG_MIGRATION) {
1443                                         /* Get the new parent FID and retry */
1444                                         mdt_object_unlock_put(info, parent,
1445                                                               lh, 1);
1446                                         mdt_lock_handle_init(lh);
1447                                         fid_le_to_cpu(
1448                                                 (struct lu_fid *)rr->rr_fid1,
1449                                                 &lmv->lmv_stripe_fids[1]);
1450                                         goto again;
1451                                 }
1452                         }
1453                 }
1454
1455                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1456                 if (result == -ESTALE) {
1457                         /*
1458                          * -ESTALE means the parent is a dead(unlinked) dir, so
1459                          * it should return -ENOENT to in accordance with the
1460                          * original mds implementaion.
1461                          */
1462                         GOTO(out_parent, result = -ENOENT);
1463                 }
1464
1465                 if (!(open_flags & MDS_OPEN_CREAT))
1466                         GOTO(out_parent, result);
1467                 if (mdt_rdonly(req->rq_export))
1468                         GOTO(out_parent, result = -EROFS);
1469                 *child_fid = *info->mti_rr.rr_fid2;
1470                 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1471                          PFID(child_fid));
1472                 /* In the function below, .hs_keycmp resolves to
1473                  * lu_obj_hop_keycmp() */
1474                 /* coverity[overrun-buffer-val] */
1475                 child = mdt_object_new(info->mti_env, mdt, child_fid);
1476         } else {
1477                 /*
1478                  * Check for O_EXCL is moved to the mdt_finish_open(),
1479                  * we need to return FID back in that case.
1480                  */
1481                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1482                 child = mdt_object_find(info->mti_env, mdt, child_fid);
1483         }
1484         if (IS_ERR(child))
1485                 GOTO(out_parent, result = PTR_ERR(child));
1486
1487         /** check version of child  */
1488         rc = mdt_version_get_check(info, child, 1);
1489         if (rc)
1490                 GOTO(out_child, result = rc);
1491
1492         if (result == -ENOENT) {
1493                 /* Create under OBF and .lustre is not permitted */
1494                 if (!fid_is_md_operative(rr->rr_fid1))
1495                         GOTO(out_child, result = -EPERM);
1496
1497                 /* save versions in reply */
1498                 mdt_version_get_save(info, parent, 0);
1499                 mdt_version_get_save(info, child, 1);
1500
1501                 /* version of child will be changed */
1502                 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
1503
1504                 /* Not found and with MDS_OPEN_CREAT: let's create it. */
1505                 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1506
1507                 /* Let lower layers know what is lock mode on directory. */
1508                 info->mti_spec.sp_cr_mode =
1509                         mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
1510
1511                 /* Don't do lookup sanity check. We know name doesn't exist. */
1512                 info->mti_spec.sp_cr_lookup = 0;
1513                 info->mti_spec.sp_feat = &dt_directory_features;
1514
1515                 result = mdo_create(info->mti_env, mdt_object_child(parent),
1516                                     &rr->rr_name, mdt_object_child(child),
1517                                     &info->mti_spec, &info->mti_attr);
1518                 if (result == -ERESTART) {
1519                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1520                         GOTO(out_child, result);
1521                 } else {
1522                         mdt_prep_ma_buf_from_rep(info, child, ma);
1523                         /* XXX: we should call this once, see few lines below */
1524                         if (result == 0)
1525                                 result = mdt_attr_get_complex(info, child, ma);
1526
1527                         if (result != 0)
1528                                 GOTO(out_child, result);
1529                 }
1530                 created = 1;
1531                 mdt_counter_incr(req, LPROC_MDT_MKNOD);
1532         } else {
1533                 /*
1534                  * The object is on remote node, return its FID for remote open.
1535                  */
1536                 if (mdt_object_remote(child)) {
1537                         /*
1538                          * Check if this lock already was sent to client and
1539                          * this is resent case. For resent case do not take lock
1540                          * again, use what is already granted.
1541                          */
1542                         LASSERT(lhc != NULL);
1543
1544                         rc = mdt_check_resent_lock(info, child, lhc);
1545                         if (rc < 0) {
1546                                 GOTO(out_child, result = rc);
1547                         } else if (rc > 0) {
1548                                 mdt_lock_handle_init(lhc);
1549                                 mdt_lock_reg_init(lhc, LCK_PR);
1550
1551                                 rc = mdt_object_lock(info, child, lhc,
1552                                                      MDS_INODELOCK_LOOKUP);
1553                         }
1554                         repbody->mbo_fid1 = *mdt_object_fid(child);
1555                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1556                         if (rc != 0)
1557                                 result = rc;
1558                         else
1559                                 result = -MDT_EREMOTE_OPEN;
1560                         GOTO(out_child, result);
1561                 } else if (mdt_object_exists(child)) {
1562                         /* We have to get attr & LOV EA & HSM for this
1563                          * object. */
1564                         mdt_prep_ma_buf_from_rep(info, child, ma);
1565                         ma->ma_need |= MA_HSM;
1566                         result = mdt_attr_get_complex(info, child, ma);
1567                         if (result != 0)
1568                                 GOTO(out_child, result);
1569                 } else {
1570                         /* Object does not exist. Likely FS corruption. */
1571                         CERROR("%s: name '"DNAME"' present, but FID "
1572                                DFID" is invalid\n", mdt_obd_name(info->mti_mdt),
1573                                PNAME(&rr->rr_name), PFID(child_fid));
1574                         GOTO(out_child, result = -EIO);
1575                 }
1576         }
1577
1578         rc = mdt_check_resent_lock(info, child, lhc);
1579         if (rc < 0) {
1580                 GOTO(out_child, result = rc);
1581         } else if (rc == 0) {
1582                 /* the open lock might already be gotten in
1583                  * ldlm_handle_enqueue() */
1584                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1585                 if (open_flags & MDS_OPEN_LOCK)
1586                         mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1587         } else {
1588                 /* get openlock if this isn't replay and client requested it */
1589                 if (!req_is_replay(req)) {
1590                         rc = mdt_object_open_lock(info, child, lhc, &ibits);
1591                         object_locked = 1;
1592                         if (rc != 0)
1593                                 GOTO(out_child_unlock, result = rc);
1594                         else if (open_flags & MDS_OPEN_LOCK)
1595                                 mdt_set_disposition(info, ldlm_rep,
1596                                                     DISP_OPEN_LOCK);
1597                 }
1598         }
1599         /* Try to open it now. */
1600         rc = mdt_finish_open(info, parent, child, open_flags,
1601                              created, ldlm_rep);
1602         if (rc) {
1603                 result = rc;
1604                 /* openlock will be released if mdt_finish_open() failed */
1605                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1606
1607                 if (created && (open_flags & MDS_OPEN_VOLATILE)) {
1608                         CERROR("%s: cannot open volatile file "DFID", orphan "
1609                                "file will be left in PENDING directory until "
1610                                "next reboot, rc = %d\n", mdt_obd_name(mdt),
1611                                PFID(mdt_object_fid(child)), rc);
1612                         GOTO(out_child_unlock, result);
1613                 }
1614
1615                 if (created) {
1616                         ma->ma_need = 0;
1617                         ma->ma_valid = 0;
1618                         rc = mdo_unlink(info->mti_env,
1619                                         mdt_object_child(parent),
1620                                         mdt_object_child(child),
1621                                         &rr->rr_name,
1622                                         &info->mti_attr, 0);
1623                         if (rc != 0)
1624                                 CERROR("%s: "DFID" cleanup of open: rc = %d\n",
1625                                        mdt_obd_name(info->mti_mdt),
1626                                        PFID(mdt_object_fid(child)), rc);
1627                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1628                 }
1629         }
1630         EXIT;
1631 out_child_unlock:
1632         if (object_locked)
1633                 mdt_object_open_unlock(info, child, lhc, ibits, result);
1634 out_child:
1635         mdt_object_put(info->mti_env, child);
1636         if (result == 0)
1637                 mdt_pack_size2body(info, child_fid, &lhc->mlh_reg_lh);
1638 out_parent:
1639         mdt_object_unlock_put(info, parent, lh, result || !created);
1640 out:
1641         if (result)
1642                 lustre_msg_set_transno(req->rq_repmsg, 0);
1643         return result;
1644 }
1645
1646 /**
1647  * Create an orphan object use local root.
1648  */
1649 static struct mdt_object *mdt_orphan_open(struct mdt_thread_info *info,
1650                                           struct mdt_device *mdt,
1651                                           const struct lu_fid *fid,
1652                                           struct md_attr *attr, fmode_t fmode)
1653 {
1654         const struct lu_env *env = info->mti_env;
1655         struct md_op_spec *spec = &info->mti_spec;
1656         struct lu_fid *local_root_fid = &info->mti_tmp_fid1;
1657         struct mdt_object *obj = NULL;
1658         struct mdt_object *local_root;
1659         static const struct lu_name lname = {
1660                 .ln_name = "i_am_nobody",
1661                 .ln_namelen = sizeof("i_am_nobody") - 1,
1662         };
1663         struct lu_ucred *uc;
1664         cfs_cap_t uc_cap_save;
1665         int rc;
1666         ENTRY;
1667
1668         rc = dt_root_get(env, mdt->mdt_bottom, local_root_fid);
1669         if (rc != 0)
1670                 RETURN(ERR_PTR(rc));
1671
1672         local_root = mdt_object_find(env, mdt, local_root_fid);
1673         if (IS_ERR(local_root))
1674                 RETURN(local_root);
1675
1676         obj = mdt_object_new(env, mdt, fid);
1677         if (IS_ERR(obj))
1678                 GOTO(out, rc = PTR_ERR(obj));
1679
1680         spec->sp_cr_lookup = 0;
1681         spec->sp_feat = &dt_directory_features;
1682         spec->sp_cr_mode = MDL_MINMODE; /* no lock */
1683         spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
1684         if (attr->ma_valid & MA_LOV) {
1685                 spec->u.sp_ea.eadata = attr->ma_lmm;
1686                 spec->u.sp_ea.eadatalen = attr->ma_lmm_size;
1687                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1688         } else {
1689                 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
1690         }
1691
1692         uc = lu_ucred(env);
1693         uc_cap_save = uc->uc_cap;
1694         uc->uc_cap |= 1 << CFS_CAP_DAC_OVERRIDE;
1695         rc = mdo_create(env, mdt_object_child(local_root), &lname,
1696                         mdt_object_child(obj), spec, attr);
1697         uc->uc_cap = uc_cap_save;
1698         if (rc < 0) {
1699                 CERROR("%s: cannot create volatile file "DFID": rc = %d\n",
1700                        mdt_obd_name(mdt), PFID(fid), rc);
1701                 GOTO(out, rc);
1702         }
1703
1704         rc = mo_open(env, mdt_object_child(obj), MDS_OPEN_CREATED);
1705         if (rc < 0)
1706                 CERROR("%s: cannot open volatile file "DFID", orphan "
1707                        "file will be left in PENDING directory until "
1708                        "next reboot, rc = %d\n", mdt_obd_name(mdt),
1709                        PFID(fid), rc);
1710         GOTO(out, rc);
1711
1712 out:
1713         if (rc < 0) {
1714                 if (!IS_ERR(obj))
1715                         mdt_object_put(env, obj);
1716                 obj = ERR_PTR(rc);
1717         }
1718         mdt_object_put(env, local_root);
1719         return obj;
1720 }
1721
1722 /* XXX Look into layout in MDT layer. */
1723 static inline int mdt_hsm_set_released(struct lov_mds_md *lmm)
1724 {
1725         struct lov_comp_md_v1   *comp_v1;
1726         struct lov_mds_md       *v1;
1727         __u32   off;
1728         int     i;
1729
1730         if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1_DEFINED)) {
1731                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1732
1733                 if (comp_v1->lcm_entry_count == 0)
1734                         return -EINVAL;
1735
1736                 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1737                         off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1738                         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
1739                         v1->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1740                 }
1741         } else {
1742                 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1743         }
1744         return 0;
1745 }
1746
1747 static int mdt_hsm_release(struct mdt_thread_info *info, struct mdt_object *o,
1748                            struct md_attr *ma)
1749 {
1750         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LAYOUT];
1751         struct lu_ucred        *uc = mdt_ucred(info);
1752         struct close_data      *data;
1753         struct ldlm_lock       *lease;
1754         struct mdt_object      *orphan;
1755         struct md_attr         *orp_ma;
1756         struct lu_buf          *buf;
1757         cfs_cap_t               cap;
1758         bool                    lease_broken;
1759         int                     rc;
1760         int                     rc2;
1761         ENTRY;
1762
1763         if (mdt_rdonly(info->mti_exp))
1764                 RETURN(-EROFS);
1765
1766         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1767         if (data == NULL)
1768                 RETURN(-EPROTO);
1769
1770         lease = ldlm_handle2lock(&data->cd_handle);
1771         if (lease == NULL)
1772                 RETURN(-ESTALE);
1773
1774         /* try to hold open_sem so that nobody else can open the file */
1775         if (!down_write_trylock(&o->mot_open_sem)) {
1776                 ldlm_lock_cancel(lease);
1777                 GOTO(out_reprocess, rc = -EBUSY);
1778         }
1779
1780         /* Check if the lease open lease has already canceled */
1781         lock_res_and_lock(lease);
1782         lease_broken = ldlm_is_cancel(lease);
1783         unlock_res_and_lock(lease);
1784
1785         LDLM_DEBUG(lease, DFID " lease broken? %d",
1786                    PFID(mdt_object_fid(o)), lease_broken);
1787
1788         /* Cancel server side lease. Client side counterpart should
1789          * have been cancelled. It's okay to cancel it now as we've
1790          * held mot_open_sem. */
1791         ldlm_lock_cancel(lease);
1792
1793         if (lease_broken) /* don't perform release task */
1794                 GOTO(out_unlock, rc = -ESTALE);
1795
1796         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1797                 GOTO(out_unlock, rc = -EINVAL);
1798
1799         /* ma_need was set before but it seems fine to change it in order to
1800          * avoid modifying the one from RPC */
1801         ma->ma_need = MA_HSM;
1802         rc = mdt_attr_get_complex(info, o, ma);
1803         if (rc != 0)
1804                 GOTO(out_unlock, rc);
1805
1806         if (!mdt_hsm_release_allow(ma))
1807                 GOTO(out_unlock, rc = -EPERM);
1808
1809         /* already released? */
1810         if (ma->ma_hsm.mh_flags & HS_RELEASED)
1811                 GOTO(out_unlock, rc = 0);
1812
1813         /* Compare on-disk and packed data_version */
1814         if (data->cd_data_version != ma->ma_hsm.mh_arch_ver) {
1815                 CDEBUG(D_HSM, DFID" data_version mismatches: packed=%llu"
1816                        " and on-disk=%llu\n", PFID(mdt_object_fid(o)),
1817                        data->cd_data_version, ma->ma_hsm.mh_arch_ver);
1818                 GOTO(out_unlock, rc = -EPERM);
1819         }
1820
1821         ma->ma_valid = MA_INODE;
1822         ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_SIZE;
1823         rc = mo_attr_set(info->mti_env, mdt_object_child(o), ma);
1824         if (rc < 0)
1825                 GOTO(out_unlock, rc);
1826
1827         mutex_lock(&o->mot_som_mutex);
1828         rc2 = mdt_set_som(info, o, SOM_FL_STRICT, ma->ma_attr.la_size,
1829                            ma->ma_attr.la_blocks);
1830         mutex_unlock(&o->mot_som_mutex);
1831         if (rc2 < 0)
1832                 CDEBUG(D_INODE,
1833                        "%s: File "DFID" SOM update failed: rc = %d\n",
1834                        mdt_obd_name(info->mti_mdt),
1835                        PFID(mdt_object_fid(o)), rc2);
1836
1837
1838         ma->ma_need = MA_INODE | MA_LOV;
1839         rc = mdt_attr_get_complex(info, o, ma);
1840         if (rc < 0)
1841                 GOTO(out_unlock, rc);
1842
1843         if (!(ma->ma_valid & MA_LOV)) {
1844                 /* Even empty file are released */
1845                 memset(ma->ma_lmm, 0, sizeof(*ma->ma_lmm));
1846                 ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEFINED);
1847                 ma->ma_lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1848                 ma->ma_lmm->lmm_stripe_size = cpu_to_le32(LOV_MIN_STRIPE_SIZE);
1849                 ma->ma_lmm_size = sizeof(*ma->ma_lmm);
1850         } else {
1851                 /* Magic must be LOV_MAGIC_*_DEFINED or LOD will interpret
1852                  * ma_lmm as lov_user_md, then it will be confused by union of
1853                  * layout_gen and stripe_offset. */
1854                 if ((le32_to_cpu(ma->ma_lmm->lmm_magic) & LOV_MAGIC_MASK) ==
1855                     LOV_MAGIC_MAGIC)
1856                         ma->ma_lmm->lmm_magic |= cpu_to_le32(LOV_MAGIC_DEFINED);
1857                 else
1858                         GOTO(out_unlock, rc = -EINVAL);
1859         }
1860
1861         /* Set file as released. */
1862         rc = mdt_hsm_set_released(ma->ma_lmm);
1863         if (rc)
1864                 GOTO(out_unlock, rc);
1865
1866         orp_ma = &info->mti_u.hsm.attr;
1867         orp_ma->ma_attr.la_mode = S_IFREG | S_IWUSR;
1868         /* We use root ownership to bypass potential quota
1869          * restrictions on the user and group of the file. */
1870         orp_ma->ma_attr.la_uid = 0;
1871         orp_ma->ma_attr.la_gid = 0;
1872         orp_ma->ma_attr.la_valid = LA_MODE | LA_UID | LA_GID;
1873         orp_ma->ma_lmm = ma->ma_lmm;
1874         orp_ma->ma_lmm_size = ma->ma_lmm_size;
1875         orp_ma->ma_valid = MA_INODE | MA_LOV;
1876         orphan = mdt_orphan_open(info, info->mti_mdt, &data->cd_fid, orp_ma,
1877                                  MDS_FMODE_WRITE);
1878         if (IS_ERR(orphan)) {
1879                 CERROR("%s: cannot open orphan file "DFID": rc = %ld\n",
1880                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid),
1881                        PTR_ERR(orphan));
1882                 GOTO(out_unlock, rc = PTR_ERR(orphan));
1883         }
1884
1885         /* Set up HSM attribute for orphan object */
1886         CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
1887         buf = &info->mti_buf;
1888         buf->lb_buf = info->mti_xattr_buf;
1889         buf->lb_len = sizeof(struct hsm_attrs);
1890         ma->ma_hsm.mh_flags |= HS_RELEASED;
1891         lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1892         ma->ma_hsm.mh_flags &= ~HS_RELEASED;
1893
1894         mdt_lock_reg_init(lh, LCK_EX);
1895         rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_LAYOUT |
1896                              MDS_INODELOCK_XATTR);
1897         if (rc != 0)
1898                 GOTO(out_close, rc);
1899
1900         /* The orphan has root ownership so we need to raise
1901          * CAP_FOWNER to set the HSM attributes. */
1902         cap = uc->uc_cap;
1903         uc->uc_cap |= MD_CAP_TO_MASK(CFS_CAP_FOWNER);
1904         rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
1905                           XATTR_NAME_HSM, 0);
1906         uc->uc_cap = cap;
1907         if (rc != 0)
1908                 GOTO(out_layout_lock, rc);
1909
1910         /* Swap layout with orphan objects. */
1911         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o),
1912                              mdt_object_child(orphan),
1913                              SWAP_LAYOUTS_MDS_HSM);
1914         EXIT;
1915
1916 out_layout_lock:
1917         /* Release exclusive LL */
1918         mdt_object_unlock(info, o, lh, 1);
1919 out_close:
1920         /* Close orphan object anyway */
1921         rc2 = mo_close(info->mti_env, mdt_object_child(orphan), orp_ma,
1922                        MDS_FMODE_WRITE);
1923         if (rc2 < 0)
1924                 CERROR("%s: error closing volatile file "DFID": rc = %d\n",
1925                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid), rc2);
1926         LU_OBJECT_DEBUG(D_HSM, info->mti_env, &orphan->mot_obj,
1927                         "object closed");
1928         mdt_object_put(info->mti_env, orphan);
1929
1930 out_unlock:
1931         up_write(&o->mot_open_sem);
1932
1933         /* already released */
1934         if (rc == 0) {
1935                 struct mdt_body *repbody;
1936
1937                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1938                 LASSERT(repbody != NULL);
1939                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
1940         }
1941
1942 out_reprocess:
1943         ldlm_reprocess_all(lease->l_resource);
1944         LDLM_LOCK_PUT(lease);
1945
1946         ma->ma_valid = 0;
1947         ma->ma_need = 0;
1948
1949         return rc;
1950 }
1951
1952 int mdt_close_handle_layouts(struct mdt_thread_info *info,
1953                              struct mdt_object *o, struct md_attr *ma)
1954 {
1955         struct mdt_lock_handle  *lh1 = &info->mti_lh[MDT_LH_NEW];
1956         struct mdt_lock_handle  *lh2 = &info->mti_lh[MDT_LH_OLD];
1957         struct close_data       *data;
1958         struct ldlm_lock        *lease;
1959         struct mdt_object       *o1 = o, *o2;
1960         bool                     lease_broken;
1961         bool                     swap_objects;
1962         int                      rc;
1963         ENTRY;
1964
1965         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
1966                 RETURN(-EROFS);
1967
1968         if (!S_ISREG(lu_object_attr(&o1->mot_obj)))
1969                 RETURN(-EINVAL);
1970
1971         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1972         if (data == NULL)
1973                 RETURN(-EPROTO);
1974
1975         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1976                 RETURN(-EINVAL);
1977
1978         rc = lu_fid_cmp(&data->cd_fid, mdt_object_fid(o));
1979         if (unlikely(rc == 0))
1980                 RETURN(-EINVAL);
1981
1982         /* Exchange o1 and o2, to enforce locking order */
1983         swap_objects = (rc < 0);
1984
1985         lease = ldlm_handle2lock(&data->cd_handle);
1986         if (lease == NULL)
1987                 RETURN(-ESTALE);
1988
1989         o2 = mdt_object_find(info->mti_env, info->mti_mdt, &data->cd_fid);
1990         if (IS_ERR(o2))
1991                 GOTO(out_lease, rc = PTR_ERR(o2));
1992
1993         if (!S_ISREG(lu_object_attr(&o2->mot_obj))) {
1994                 swap_objects = false; /* not swapped yet */
1995                 GOTO(out_obj, rc = -EINVAL);
1996         }
1997
1998         if (swap_objects)
1999                 swap(o1, o2);
2000
2001         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
2002                            MAY_WRITE);
2003         if (rc < 0)
2004                 GOTO(out_obj, rc);
2005
2006         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
2007                            MAY_WRITE);
2008         if (rc < 0)
2009                 GOTO(out_obj, rc);
2010
2011         /* try to hold open_sem so that nobody else can open the file */
2012         if (!down_write_trylock(&o->mot_open_sem)) {
2013                 ldlm_lock_cancel(lease);
2014                 GOTO(out_obj, rc = -EBUSY);
2015         }
2016
2017         /* Check if the lease open lease has already canceled */
2018         lock_res_and_lock(lease);
2019         lease_broken = ldlm_is_cancel(lease);
2020         unlock_res_and_lock(lease);
2021
2022         LDLM_DEBUG(lease, DFID " lease broken? %d",
2023                    PFID(mdt_object_fid(o)), lease_broken);
2024
2025         /* Cancel server side lease. Client side counterpart should
2026          * have been cancelled. It's okay to cancel it now as we've
2027          * held mot_open_sem. */
2028         ldlm_lock_cancel(lease);
2029
2030         if (lease_broken)
2031                 GOTO(out_unlock_sem, rc = -ESTALE);
2032
2033         mdt_lock_reg_init(lh1, LCK_EX);
2034         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
2035                              MDS_INODELOCK_XATTR);
2036         if (rc < 0)
2037                 GOTO(out_unlock_sem, rc);
2038
2039         mdt_lock_reg_init(lh2, LCK_EX);
2040         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
2041                              MDS_INODELOCK_XATTR);
2042         if (rc < 0)
2043                 GOTO(out_unlock1, rc);
2044
2045         /* Swap layout with orphan object */
2046         if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SWAP) {
2047                 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
2048                                      mdt_object_child(o2), 0);
2049         } else if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_MERGE ||
2050                    ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
2051                 struct lu_buf *buf = &info->mti_buf;
2052                 struct md_rejig_data mrd;
2053
2054                 mrd.mrd_obj = mdt_object_child(o == o1 ? o2 : o1);
2055                 if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT)
2056                         mrd.mrd_mirror_id = data->cd_mirror_id;
2057
2058                 buf->lb_len = sizeof(mrd);
2059                 buf->lb_buf = &mrd;
2060                 rc = mo_xattr_set(info->mti_env, mdt_object_child(o), buf,
2061                                   XATTR_LUSTRE_LOV,
2062                                   ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT ?
2063                                   LU_XATTR_SPLIT : LU_XATTR_MERGE);
2064                 if (rc == 0 && ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2065                         int rc2;
2066
2067                         mutex_lock(&o->mot_som_mutex);
2068                         rc2 = mdt_set_som(info, o, SOM_FL_STRICT,
2069                                           ma->ma_attr.la_size,
2070                                           ma->ma_attr.la_blocks);
2071                         mutex_unlock(&o->mot_som_mutex);
2072                         if (rc2 < 0)
2073                                 CERROR(DFID": Setting i_blocks error: %d, "
2074                                        "i_blocks will be reported wrongly and "
2075                                        "can only be fixed in next resync\n",
2076                                        PFID(mdt_object_fid(o)), rc2);
2077                 }
2078         }
2079         if (rc < 0)
2080                 GOTO(out_unlock2, rc);
2081
2082         EXIT;
2083
2084 out_unlock2:
2085         /* Release exclusive LL */
2086         mdt_object_unlock(info, o2, lh2, 1);
2087
2088 out_unlock1:
2089         mdt_object_unlock(info, o1, lh1, 1);
2090
2091 out_unlock_sem:
2092         up_write(&o->mot_open_sem);
2093
2094         /* already swapped */
2095         if (rc == 0) {
2096                 struct mdt_body *repbody;
2097
2098                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2099                 LASSERT(repbody != NULL);
2100                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2101         }
2102
2103 out_obj:
2104         mdt_object_put(info->mti_env, swap_objects ? o1 : o2);
2105
2106         ldlm_reprocess_all(lease->l_resource);
2107
2108 out_lease:
2109         LDLM_LOCK_PUT(lease);
2110
2111         if (ma != NULL) {
2112                 ma->ma_valid = 0;
2113                 ma->ma_need = 0;
2114         }
2115
2116         return rc;
2117 }
2118
2119 static int mdt_close_resync_done(struct mdt_thread_info *info,
2120                                  struct mdt_object *o, struct md_attr *ma)
2121 {
2122         struct close_data       *data;
2123         struct ldlm_lock        *lease;
2124         struct md_layout_change  layout = { 0 };
2125         __u32                   *resync_ids = NULL;
2126         size_t                   resync_count = 0;
2127         bool                     lease_broken;
2128         int                      rc;
2129         ENTRY;
2130
2131         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2132                 RETURN(-EROFS);
2133
2134         if (!S_ISREG(lu_object_attr(&o->mot_obj)))
2135                 RETURN(-EINVAL);
2136
2137         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2138         if (data == NULL)
2139                 RETURN(-EPROTO);
2140
2141         if (ptlrpc_req_need_swab(mdt_info_req(info)))
2142                 lustre_swab_close_data_resync_done(&data->cd_resync);
2143
2144         if (!fid_is_zero(&data->cd_fid))
2145                 RETURN(-EPROTO);
2146
2147         lease = ldlm_handle2lock(&data->cd_handle);
2148         if (lease == NULL)
2149                 RETURN(-ESTALE);
2150
2151         /* try to hold open_sem so that nobody else can open the file */
2152         if (!down_write_trylock(&o->mot_open_sem)) {
2153                 ldlm_lock_cancel(lease);
2154                 GOTO(out_reprocess, rc = -EBUSY);
2155         }
2156
2157         /* Check if the lease open lease has already canceled */
2158         lock_res_and_lock(lease);
2159         lease_broken = ldlm_is_cancel(lease);
2160         unlock_res_and_lock(lease);
2161
2162         LDLM_DEBUG(lease, DFID " lease broken? %d\n",
2163                    PFID(mdt_object_fid(o)), lease_broken);
2164
2165         /* Cancel server side lease. Client side counterpart should
2166          * have been cancelled. It's okay to cancel it now as we've
2167          * held mot_open_sem. */
2168         ldlm_lock_cancel(lease);
2169
2170         if (lease_broken) /* don't perform release task */
2171                 GOTO(out_unlock, rc = -ESTALE);
2172
2173         resync_count = data->cd_resync.resync_count;
2174
2175         if (resync_count > INLINE_RESYNC_ARRAY_SIZE) {
2176                 void *data;
2177
2178                 if (!req_capsule_has_field(info->mti_pill, &RMF_U32,
2179                                            RCL_CLIENT))
2180                         GOTO(out_unlock, rc = -EPROTO);
2181
2182                 OBD_ALLOC(resync_ids, resync_count * sizeof(__u32));
2183                 if (!resync_ids)
2184                         GOTO(out_unlock, rc = -ENOMEM);
2185
2186                 data = req_capsule_client_get(info->mti_pill, &RMF_U32);
2187                 memcpy(resync_ids, data, resync_count * sizeof(__u32));
2188
2189                 layout.mlc_resync_ids = resync_ids;
2190         } else {
2191                 layout.mlc_resync_ids = data->cd_resync.resync_ids_inline;
2192         }
2193
2194         layout.mlc_opc = MD_LAYOUT_RESYNC_DONE;
2195         layout.mlc_resync_count = resync_count;
2196         if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2197                 layout.mlc_som.lsa_valid = SOM_FL_STRICT;
2198                 layout.mlc_som.lsa_size = ma->ma_attr.la_size;
2199                 layout.mlc_som.lsa_blocks = ma->ma_attr.la_blocks;
2200         }
2201         rc = mdt_layout_change(info, o, &layout);
2202         if (rc)
2203                 GOTO(out_unlock, rc);
2204
2205         EXIT;
2206
2207 out_unlock:
2208         up_write(&o->mot_open_sem);
2209
2210         /* already released */
2211         if (rc == 0) {
2212                 struct mdt_body *repbody;
2213
2214                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2215                 LASSERT(repbody != NULL);
2216                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2217         }
2218
2219         if (resync_ids)
2220                 OBD_FREE(resync_ids, resync_count * sizeof(__u32));
2221
2222 out_reprocess:
2223         ldlm_reprocess_all(lease->l_resource);
2224         LDLM_LOCK_PUT(lease);
2225
2226         ma->ma_valid = 0;
2227         ma->ma_need = 0;
2228
2229         return rc;
2230 }
2231
2232 #define MFD_CLOSED(open_flags) ((open_flags) == MDS_FMODE_CLOSED)
2233
2234 static int mdt_mfd_closed(struct mdt_file_data *mfd)
2235 {
2236         return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_open_flags));
2237 }
2238
2239 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
2240 {
2241         struct mdt_object *o = mfd->mfd_object;
2242         struct md_object *next = mdt_object_child(o);
2243         struct md_attr *ma = &info->mti_attr;
2244         struct lu_fid *ofid = &info->mti_tmp_fid1;
2245         int rc = 0;
2246         u64 open_flags;
2247         u64 intent;
2248         bool discard = false;
2249
2250         ENTRY;
2251
2252         open_flags = mfd->mfd_open_flags;
2253         intent = ma->ma_attr_flags & MDS_CLOSE_INTENT;
2254         *ofid = *mdt_object_fid(o);
2255
2256         CDEBUG(D_INODE, "%s: close file "DFID" with intent: %llx\n",
2257                mdt_obd_name(info->mti_mdt), PFID(ofid), intent);
2258
2259         switch (intent) {
2260         case MDS_HSM_RELEASE: {
2261                 rc = mdt_hsm_release(info, o, ma);
2262                 if (rc < 0) {
2263                         CDEBUG(D_HSM, "%s: File " DFID " release failed: %d\n",
2264                                mdt_obd_name(info->mti_mdt),
2265                                PFID(ofid), rc);
2266                         /* continue to close even error occurred. */
2267                 }
2268                 break;
2269         }
2270         case MDS_CLOSE_LAYOUT_MERGE:
2271         case MDS_CLOSE_LAYOUT_SPLIT:
2272         case MDS_CLOSE_LAYOUT_SWAP: {
2273                 rc = mdt_close_handle_layouts(info, o, ma);
2274                 if (rc < 0) {
2275                         CDEBUG(D_INODE,
2276                                "%s: cannot swap layout of "DFID": rc = %d\n",
2277                                mdt_obd_name(info->mti_mdt),
2278                                PFID(ofid), rc);
2279                         /* continue to close even if error occurred. */
2280                 }
2281                 break;
2282         }
2283         case MDS_CLOSE_RESYNC_DONE:
2284                 rc = mdt_close_resync_done(info, o, ma);
2285                 break;
2286         default:
2287                 /* nothing */
2288                 break;
2289         }
2290
2291         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
2292             ma->ma_attr.la_valid & (LA_LSIZE | LA_LBLOCKS)) {
2293                 int rc2;
2294
2295                 rc2 = mdt_lsom_update(info, o, false);
2296                 if (rc2 < 0)
2297                         CDEBUG(D_INODE,
2298                                "%s: File " DFID " LSOM failed: rc = %d\n",
2299                                mdt_obd_name(info->mti_mdt),
2300                                PFID(ofid), rc2);
2301                         /* continue to close even if error occured. */
2302         }
2303
2304         if (open_flags & MDS_FMODE_WRITE)
2305                 mdt_write_put(o);
2306         else if (open_flags & MDS_FMODE_EXEC)
2307                 mdt_write_allow(o);
2308
2309         /* Update atime on close only. */
2310         if ((open_flags & MDS_FMODE_EXEC || open_flags & MDS_FMODE_READ ||
2311              open_flags & MDS_FMODE_WRITE) && (ma->ma_valid & MA_INODE) &&
2312             (ma->ma_attr.la_valid & LA_ATIME)) {
2313                 /* Set the atime only. */
2314                 ma->ma_valid = MA_INODE;
2315                 ma->ma_attr.la_valid = LA_ATIME;
2316                 rc = mo_attr_set(info->mti_env, next, ma);
2317         }
2318
2319         /* If file data is modified, add the dirty flag. */
2320         if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
2321                 rc = mdt_add_dirty_flag(info, o, ma);
2322
2323         ma->ma_need |= MA_INODE;
2324         ma->ma_valid &= ~MA_INODE;
2325
2326         LASSERT(atomic_read(&o->mot_open_count) > 0);
2327         atomic_dec(&o->mot_open_count);
2328         mdt_handle_last_unlink(info, o, ma);
2329
2330         if (!MFD_CLOSED(open_flags)) {
2331                 rc = mo_close(info->mti_env, next, ma, open_flags);
2332                 discard = mdt_dom_check_for_discard(info, o);
2333         }
2334
2335         /* adjust open and lease count */
2336         if (open_flags & MDS_OPEN_LEASE) {
2337                 LASSERT(atomic_read(&o->mot_lease_count) > 0);
2338                 atomic_dec(&o->mot_lease_count);
2339         }
2340
2341         mdt_mfd_free(mfd);
2342         mdt_object_put(info->mti_env, o);
2343
2344         if (discard)
2345                 mdt_dom_discard_data(info, ofid);
2346
2347         RETURN(rc);
2348 }
2349
2350 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
2351                        struct mdt_body *repbody)
2352 {
2353         struct mdt_export_data *med;
2354         struct mdt_file_data   *mfd;
2355         int                     ret = 0;
2356         int                     rc = 0;
2357         ENTRY;
2358
2359         med = &req->rq_export->exp_mdt_data;
2360         spin_lock(&med->med_open_lock);
2361         mfd = mdt_open_handle2mfd(med, &info->mti_open_handle,
2362                                   req_is_replay(req));
2363         if (mdt_mfd_closed(mfd)) {
2364                 spin_unlock(&med->med_open_lock);
2365                 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
2366                        ": cookie = %#llx\n", PFID(info->mti_rr.rr_fid1),
2367                        info->mti_open_handle.cookie);
2368                 /** not serious error since bug 3633 */
2369                 rc = -ESTALE;
2370         } else {
2371                 class_handle_unhash(&mfd->mfd_open_handle);
2372                 list_del_init(&mfd->mfd_list);
2373                 spin_unlock(&med->med_open_lock);
2374                 ret = mdt_mfd_close(info, mfd);
2375         }
2376
2377         RETURN(rc ? rc : ret);
2378 }
2379
2380 int mdt_close(struct tgt_session_info *tsi)
2381 {
2382         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2383         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2384         struct md_attr         *ma = &info->mti_attr;
2385         struct mdt_body        *repbody = NULL;
2386         int rc, ret = 0;
2387         ENTRY;
2388
2389         mdt_counter_incr(req, LPROC_MDT_CLOSE);
2390         /* Close may come with the Size-on-MDS update. Unpack it. */
2391         rc = mdt_close_unpack(info);
2392         if (rc)
2393                 GOTO(out, rc = err_serious(rc));
2394
2395         /* These fields are no longer used and are left for compatibility.
2396          * size is always zero */
2397         req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
2398                              0);
2399         req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
2400                              0);
2401         rc = req_capsule_server_pack(info->mti_pill);
2402         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
2403                 mdt_client_compatibility(info);
2404                 if (rc == 0)
2405                         mdt_fix_reply(info);
2406                 mdt_exit_ucred(info);
2407                 GOTO(out, rc = lustre_msg_get_status(req->rq_repmsg));
2408         }
2409
2410         /* Continue to close handle even if we can not pack reply */
2411         if (rc == 0) {
2412                 repbody = req_capsule_server_get(info->mti_pill,
2413                                                  &RMF_MDT_BODY);
2414                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
2415                                                     &RMF_MDT_MD);
2416                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
2417                                                        &RMF_MDT_MD,
2418                                                        RCL_SERVER);
2419                 ma->ma_need = MA_INODE | MA_LOV;
2420                 repbody->mbo_eadatasize = 0;
2421                 repbody->mbo_aclsize = 0;
2422         } else {
2423                 rc = err_serious(rc);
2424         }
2425
2426         rc = mdt_close_internal(info, req, repbody);
2427         if (rc != -ESTALE)
2428                 mdt_empty_transno(info, rc);
2429
2430         if (repbody != NULL) {
2431                 mdt_client_compatibility(info);
2432                 rc = mdt_fix_reply(info);
2433         }
2434
2435         mdt_exit_ucred(info);
2436         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
2437                 GOTO(out, rc = err_serious(-ENOMEM));
2438
2439         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
2440                                  OBD_FAIL_MDS_CLOSE_NET_REP))
2441                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
2442 out:
2443         mdt_thread_info_fini(info);
2444         RETURN(rc ? rc : ret);
2445 }