lustre/mdt/mdt_open.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_open.c
  37  *
  38  * Lustre Metadata Target (mdt) open/close file handling
  39  *
  40  * Author: Huang Hua <huanghua@clusterfs.com>
  41  */
  42
  43 #ifndef EXPORT_SYMTAB
  44 # define EXPORT_SYMTAB
  45 #endif
  46 #define DEBUG_SUBSYSTEM S_MDS
  47
  48 #include <lustre_acl.h>
  49 #include <lustre_mds.h>
  50 #include "mdt_internal.h"
  51
  52 /* we do nothing because we do not have refcount now */
  53 static void mdt_mfd_get(void *mfdp)
  54 {
  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(void)
  60 {
  61         struct mdt_file_data *mfd;
  62         ENTRY;
  63
  64         OBD_ALLOC_PTR(mfd);
  65         if (mfd != NULL) {
  66                 CFS_INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
  67                 CFS_INIT_LIST_HEAD(&mfd->mfd_list);
  68                 class_handle_hash(&mfd->mfd_handle, mdt_mfd_get);
  69         }
  70         RETURN(mfd);
  71 }
  72
  73 /*
  74  * Find the mfd pointed to by handle in global hash table.
  75  * In case of replay the handle is obsoleted
  76  * but mfd can be found in mfd list by that handle
  77  */
  78 struct mdt_file_data *mdt_handle2mfd(struct mdt_thread_info *info,
  79                                      const struct lustre_handle *handle)
  80 {
  81         struct ptlrpc_request *req = mdt_info_req(info);
  82         struct mdt_file_data  *mfd;
  83         ENTRY;
  84
  85         LASSERT(handle != NULL);
  86         mfd = class_handle2object(handle->cookie);
  87         /* during dw/setattr replay the mfd can be found by old handle */
  88         if (mfd == NULL &&
  89             lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
  90                 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
  91                 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
  92                         if (mfd->mfd_old_handle.cookie == handle->cookie)
  93                                 RETURN (mfd);
  94                 }
  95                 mfd = NULL;
  96         }
  97         RETURN (mfd);
  98 }
  99
 100 /* free mfd */
 101 void mdt_mfd_free(struct mdt_file_data *mfd)
 102 {
 103         LASSERT(list_empty(&mfd->mfd_list));
 104         OBD_FREE_RCU(mfd, sizeof *mfd, &mfd->mfd_handle);
 105 }
 106
 107 static int mdt_create_data(struct mdt_thread_info *info,
 108                            struct mdt_object *p, struct mdt_object *o)
 109 {
 110         struct md_op_spec     *spec = &info->mti_spec;
 111         struct md_attr        *ma = &info->mti_attr;
 112         int rc;
 113         ENTRY;
 114
 115         if (!md_should_create(spec->sp_cr_flags))
 116                 RETURN(0);
 117
 118         ma->ma_need = MA_INODE | MA_LOV;
 119         ma->ma_valid = 0;
 120         rc = mdo_create_data(info->mti_env,
 121                              p ? mdt_object_child(p) : NULL,
 122                              mdt_object_child(o), spec, ma);
 123         RETURN(rc);
 124 }
 125
 126 static int mdt_epoch_opened(struct mdt_object *mo)
 127 {
 128         return mo->mot_epochcount;
 129 }
 130
 131 int mdt_sizeonmds_enabled(struct mdt_object *mo)
 132 {
 133         return !mo->mot_ioepoch;
 134 }
 135
 136 /* Re-enable Size-on-MDS. */
 137 void mdt_sizeonmds_enable(struct mdt_thread_info *info,
 138                           struct mdt_object *mo)
 139 {
 140        spin_lock(&info->mti_mdt->mdt_ioepoch_lock);
 141        if (info->mti_epoch->ioepoch == mo->mot_ioepoch) {
 142                 LASSERT(!mdt_epoch_opened(mo));
 143                 mo->mot_ioepoch = 0;
 144                 mo->mot_flags = 0;
 145        }
 146        spin_unlock(&info->mti_mdt->mdt_ioepoch_lock);
 147 }
 148
 149 /* Open the epoch. Epoch open is allowed if @writecount is not negative.
 150  * The epoch and writecount handling is performed under the mdt_ioepoch_lock. */
 151 int mdt_epoch_open(struct mdt_thread_info *info, struct mdt_object *o)
 152 {
 153         struct mdt_device *mdt = info->mti_mdt;
 154         int cancel = 0;
 155         int rc = 0;
 156         ENTRY;
 157
 158         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 159             !S_ISREG(lu_object_attr(&o->mot_obj.mo_lu)))
 160                 RETURN(0);
 161
 162         spin_lock(&mdt->mdt_ioepoch_lock);
 163         if (mdt_epoch_opened(o)) {
 164                 /* Epoch continues even if there is no writers yet. */
 165                 CDEBUG(D_INODE, "continue epoch "LPU64" for "DFID"\n",
 166                        o->mot_ioepoch, PFID(mdt_object_fid(o)));
 167         } else {
 168                 if (info->mti_replayepoch > mdt->mdt_ioepoch)
 169                         mdt->mdt_ioepoch = info->mti_replayepoch;
 170                 else
 171                         mdt->mdt_ioepoch++;
 172                 o->mot_ioepoch = info->mti_replayepoch ?
 173                         info->mti_replayepoch : mdt->mdt_ioepoch;
 174                 CDEBUG(D_INODE, "starting epoch "LPU64" for "DFID"\n",
 175                        mdt->mdt_ioepoch, PFID(mdt_object_fid(o)));
 176                 cancel = 1;
 177         }
 178         o->mot_epochcount++;
 179         spin_unlock(&mdt->mdt_ioepoch_lock);
 180
 181         /* Cancel Size-on-MDS attributes on clients if not truncate.
 182          * In the later case, mdt_reint_setattr will do it. */
 183         if (cancel && (info->mti_rr.rr_fid1 != NULL)) {
 184                 struct mdt_lock_handle  *lh = &info->mti_lh[MDT_LH_CHILD];
 185                 mdt_lock_reg_init(lh, LCK_EX);
 186                 rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_UPDATE,
 187                                      MDT_LOCAL_LOCK);
 188                 if (rc == 0)
 189                         mdt_object_unlock(info, o, lh, 1);
 190         }
 191         RETURN(rc);
 192 }
 193
 194 /* Update the on-disk attributes if needed and re-enable Size-on-MDS caching. */
 195 static int mdt_sizeonmds_update(struct mdt_thread_info *info,
 196                                 struct mdt_object *o)
 197 {
 198         ENTRY;
 199
 200         CDEBUG(D_INODE, "Closing epoch "LPU64" on "DFID". Count %d\n",
 201                o->mot_ioepoch, PFID(mdt_object_fid(o)), o->mot_epochcount);
 202
 203         if (info->mti_attr.ma_attr.la_valid & LA_SIZE) {
 204                 /* Do Size-on-MDS attribute update.
 205                  * Size-on-MDS is re-enabled inside. */
 206                 /* XXX: since we have opened the file, it is unnecessary
 207                  * to check permission when close it. Between the "open"
 208                  * and "close", maybe someone has changed the file mode
 209                  * or flags, or the file created mode do not permit wirte,
 210                  * and so on. Just set MDS_PERM_BYPASS for all the cases. */
 211                 info->mti_attr.ma_attr_flags |= MDS_PERM_BYPASS | MDS_SOM;
 212                 info->mti_attr.ma_attr.la_valid &= LA_SIZE | LA_BLOCKS |
 213                                                 LA_ATIME | LA_MTIME | LA_CTIME;
 214                 RETURN(mdt_attr_set(info, o, 0));
 215         } else
 216                 mdt_sizeonmds_enable(info, o);
 217         RETURN(0);
 218 }
 219
 220 /* Epoch closes.
 221  * Returns 1 if epoch does not close.
 222  * Returns 0 if epoch closes.
 223  * Returns -EAGAIN if epoch closes but an Size-on-MDS Update is still needed
 224  * from the client. */
 225 static int mdt_epoch_close(struct mdt_thread_info *info, struct mdt_object *o)
 226 {
 227         int eviction = (mdt_info_req(info) == NULL ? 1 : 0);
 228         struct lu_attr *la = &info->mti_attr.ma_attr;
 229         int achange = 0;
 230         int opened;
 231         int rc = 1;
 232         ENTRY;
 233
 234         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 235             !S_ISREG(lu_object_attr(&o->mot_obj.mo_lu)))
 236                 RETURN(0);
 237
 238         spin_lock(&info->mti_mdt->mdt_ioepoch_lock);
 239
 240         /* Epoch closes only if client tells about it or eviction occures. */
 241         if (eviction || (info->mti_epoch->flags & MF_EPOCH_CLOSE)) {
 242                 LASSERT(o->mot_epochcount);
 243                 o->mot_epochcount--;
 244
 245                 CDEBUG(D_INODE, "Closing epoch "LPU64" on "DFID". Count %d\n",
 246                        o->mot_ioepoch, PFID(mdt_object_fid(o)),
 247                        o->mot_epochcount);
 248
 249                 if (!eviction)
 250                         achange = (info->mti_epoch->flags & MF_SOM_CHANGE);
 251
 252                 rc = 0;
 253                 if (!eviction && !mdt_epoch_opened(o)) {
 254                         /* Epoch ends. Is an Size-on-MDS update needed? */
 255                         if (o->mot_flags & MF_SOM_CHANGE) {
 256                                 /* Some previous writer changed the attribute.
 257                                  * Do not believe to the current Size-on-MDS
 258                                  * update, re-ask client. */
 259                                 rc = -EAGAIN;
 260                         } else if (!(la->la_valid & LA_SIZE) && achange) {
 261                                 /* Attributes were changed by the last writer
 262                                  * only but no Size-on-MDS update is received.*/
 263                                 rc = -EAGAIN;
 264                         }
 265                 }
 266
 267                 if (achange || eviction)
 268                         o->mot_flags |= MF_SOM_CHANGE;
 269         }
 270
 271         opened = mdt_epoch_opened(o);
 272         spin_unlock(&info->mti_mdt->mdt_ioepoch_lock);
 273
 274         /* If eviction occurred, do nothing. */
 275         if ((rc == 0) && !opened && !eviction) {
 276                 /* Epoch ends and wanted Size-on-MDS update is obtained. */
 277                 rc = mdt_sizeonmds_update(info, o);
 278                 /* Avoid the following setattrs of these attributes, e.g.
 279                  * for atime update. */
 280                 info->mti_attr.ma_valid = 0;
 281         }
 282         RETURN(rc);
 283 }
 284
 285 int mdt_write_read(struct mdt_device *mdt, struct mdt_object *o)
 286 {
 287         int rc = 0;
 288         ENTRY;
 289         spin_lock(&mdt->mdt_ioepoch_lock);
 290         rc = o->mot_writecount;
 291         spin_unlock(&mdt->mdt_ioepoch_lock);
 292         RETURN(rc);
 293 }
 294
 295 int mdt_write_get(struct mdt_device *mdt, struct mdt_object *o)
 296 {
 297         int rc = 0;
 298         ENTRY;
 299         spin_lock(&mdt->mdt_ioepoch_lock);
 300         if (o->mot_writecount < 0)
 301                 rc = -ETXTBSY;
 302         else
 303                 o->mot_writecount++;
 304         spin_unlock(&mdt->mdt_ioepoch_lock);
 305         RETURN(rc);
 306 }
 307
 308 static void mdt_write_put(struct mdt_device *mdt, struct mdt_object *o)
 309 {
 310         ENTRY;
 311         spin_lock(&mdt->mdt_ioepoch_lock);
 312         o->mot_writecount--;
 313         spin_unlock(&mdt->mdt_ioepoch_lock);
 314         EXIT;
 315 }
 316
 317 static int mdt_write_deny(struct mdt_device *mdt, struct mdt_object *o)
 318 {
 319         int rc = 0;
 320         ENTRY;
 321         spin_lock(&mdt->mdt_ioepoch_lock);
 322         if (o->mot_writecount > 0)
 323                 rc = -ETXTBSY;
 324         else
 325                 o->mot_writecount--;
 326         spin_unlock(&mdt->mdt_ioepoch_lock);
 327         RETURN(rc);
 328 }
 329
 330 static void mdt_write_allow(struct mdt_device *mdt, struct mdt_object *o)
 331 {
 332         ENTRY;
 333         spin_lock(&mdt->mdt_ioepoch_lock);
 334         o->mot_writecount++;
 335         spin_unlock(&mdt->mdt_ioepoch_lock);
 336         EXIT;
 337 }
 338
 339 /* there can be no real transaction so prepare the fake one */
 340 static void mdt_empty_transno(struct mdt_thread_info* info)
 341 {
 342         struct mdt_device *mdt = info->mti_mdt;
 343         struct ptlrpc_request *req = mdt_info_req(info);
 344
 345         ENTRY;
 346         /* transaction is occured already */
 347         if (lustre_msg_get_transno(req->rq_repmsg) != 0) {
 348                 EXIT;
 349                 return;
 350         }
 351
 352         spin_lock(&mdt->mdt_transno_lock);
 353         if (info->mti_transno == 0) {
 354                 info->mti_transno = ++ mdt->mdt_last_transno;
 355         } else {
 356                 /* should be replay */
 357                 if (info->mti_transno > mdt->mdt_last_transno)
 358                         mdt->mdt_last_transno = info->mti_transno;
 359         }
 360         spin_unlock(&mdt->mdt_transno_lock);
 361
 362         CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n",
 363                         info->mti_transno,
 364                         req->rq_export->exp_obd->obd_last_committed);
 365
 366         req->rq_transno = info->mti_transno;
 367         lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
 368         lustre_msg_set_last_xid(req->rq_repmsg, req->rq_xid);
 369         EXIT;
 370 }
 371
 372 void mdt_mfd_set_mode(struct mdt_file_data *mfd, int mode)
 373 {
 374         LASSERT(mfd != NULL);
 375
 376         CDEBUG(D_HA, "Change mfd %p mode 0x%x->0x%x\n",
 377                mfd, (unsigned int)mfd->mfd_mode, (unsigned int)mode);
 378
 379         mfd->mfd_mode = mode;
 380 }
 381
 382 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
 383                         struct mdt_object *o, int flags, int created)
 384 {
 385         struct ptlrpc_request   *req = mdt_info_req(info);
 386         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 387         struct mdt_file_data    *mfd;
 388         struct md_attr          *ma  = &info->mti_attr;
 389         struct lu_attr          *la  = &ma->ma_attr;
 390         struct mdt_body         *repbody;
 391         int                      rc = 0, isdir, isreg;
 392         ENTRY;
 393
 394         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 395
 396         isreg = S_ISREG(la->la_mode);
 397         isdir = S_ISDIR(la->la_mode);
 398         if (isreg && !(ma->ma_valid & MA_LOV)) {
 399                 /*
 400                  * No EA, check whether it is will set regEA and dirEA since in
 401                  * above attr get, these size might be zero, so reset it, to
 402                  * retrieve the MD after create obj.
 403                  */
 404                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
 405                                                        &RMF_MDT_MD,
 406                                                        RCL_SERVER);
 407                 /* in replay case, p == NULL */
 408                 rc = mdt_create_data(info, p, o);
 409                 if (rc)
 410                         RETURN(rc);
 411         }
 412
 413         CDEBUG(D_INODE, "after open, ma_valid bit = "LPX64" lmm_size = %d\n",
 414                ma->ma_valid, ma->ma_lmm_size);
 415
 416         if (ma->ma_valid & MA_LOV) {
 417                 LASSERT(ma->ma_lmm_size != 0);
 418                 repbody->eadatasize = ma->ma_lmm_size;
 419                 if (isdir)
 420                         repbody->valid |= OBD_MD_FLDIREA;
 421                 else
 422                         repbody->valid |= OBD_MD_FLEASIZE;
 423         }
 424
 425         if (flags & FMODE_WRITE) {
 426                 rc = mdt_write_get(info->mti_mdt, o);
 427                 if (rc == 0) {
 428                         mdt_epoch_open(info, o);
 429                         repbody->ioepoch = o->mot_ioepoch;
 430                 }
 431         } else if (flags & MDS_FMODE_EXEC) {
 432                 rc = mdt_write_deny(info->mti_mdt, o);
 433         }
 434         if (rc)
 435                 RETURN(rc);
 436
 437         rc = mo_open(info->mti_env, mdt_object_child(o),
 438                      created ? flags | MDS_OPEN_CREATED : flags);
 439         if (rc)
 440                 RETURN(rc);
 441
 442         mfd = mdt_mfd_new();
 443         if (mfd != NULL) {
 444                 /*
 445                  * Keep a reference on this object for this open, and is
 446                  * released by mdt_mfd_close().
 447                  */
 448                 mdt_object_get(info->mti_env, o);
 449
 450                 /*
 451                  * @flags is always not zero. At least it should be FMODE_READ,
 452                  * FMODE_WRITE or FMODE_EXEC.
 453                  */
 454                 LASSERT(flags != 0);
 455
 456                 /* Open handling. */
 457                 mdt_mfd_set_mode(mfd, flags);
 458
 459                 mfd->mfd_object = o;
 460                 mfd->mfd_xid = req->rq_xid;
 461
 462                 /* replay handle */
 463                 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
 464                         struct mdt_file_data *old_mfd;
 465                         /* Check wheather old cookie already exist in
 466                          * the list, becasue when do recovery, client
 467                          * might be disconnected from server, and
 468                          * restart replay, so there maybe some orphan
 469                          * mfd here, we should remove them */
 470                         LASSERT(info->mti_rr.rr_handle != NULL);
 471                         old_mfd = mdt_handle2mfd(info, info->mti_rr.rr_handle);
 472                         if (old_mfd) {
 473                                 CDEBUG(D_HA, "del orph mfd %p fid=("DFID") "
 474                                        "cookie=" LPX64"\n",
 475                                        mfd,
 476                                        PFID(mdt_object_fid(mfd->mfd_object)),
 477                                        info->mti_rr.rr_handle->cookie);
 478                                 spin_lock(&med->med_open_lock);
 479                                 class_handle_unhash(&old_mfd->mfd_handle);
 480                                 list_del_init(&old_mfd->mfd_list);
 481                                 spin_unlock(&med->med_open_lock);
 482                                 mdt_mfd_close(info, old_mfd);
 483                         }
 484                         CDEBUG(D_HA, "Store old cookie "LPX64" in new mfd\n",
 485                                info->mti_rr.rr_handle->cookie);
 486                         mfd->mfd_old_handle.cookie =
 487                                                 info->mti_rr.rr_handle->cookie;
 488                 }
 489                 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
 490
 491                 if (req->rq_export->exp_disconnected) {
 492                         spin_lock(&med->med_open_lock);
 493                         class_handle_unhash(&mfd->mfd_handle);
 494                         list_del_init(&mfd->mfd_list);
 495                         spin_unlock(&med->med_open_lock);
 496                         mdt_mfd_close(info, mfd);
 497                 } else {
 498                         spin_lock(&med->med_open_lock);
 499                         list_add(&mfd->mfd_list, &med->med_open_head);
 500                         spin_unlock(&med->med_open_lock);
 501                 }
 502
 503                 mdt_empty_transno(info);
 504         } else
 505                 rc = -ENOMEM;
 506
 507         RETURN(rc);
 508 }
 509
 510
 511 static int mdt_finish_open(struct mdt_thread_info *info,
 512                            struct mdt_object *p, struct mdt_object *o,
 513                            int flags, int created, struct ldlm_reply *rep)
 514 {
 515         struct ptlrpc_request   *req = mdt_info_req(info);
 516         struct obd_export       *exp = req->rq_export;
 517         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 518         struct md_attr          *ma  = &info->mti_attr;
 519         struct lu_attr          *la  = &ma->ma_attr;
 520         struct mdt_file_data    *mfd;
 521         struct mdt_body         *repbody;
 522         int                      rc = 0;
 523         int                      isreg, isdir, islnk;
 524         struct list_head        *t;
 525         ENTRY;
 526
 527         LASSERT(ma->ma_valid & MA_INODE);
 528
 529         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 530
 531         isreg = S_ISREG(la->la_mode);
 532         isdir = S_ISDIR(la->la_mode);
 533         islnk = S_ISLNK(la->la_mode);
 534         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 535
 536         if (exp_connect_rmtclient(exp)) {
 537                 void *buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 538
 539                 rc = mdt_pack_remote_perm(info, o, buf);
 540                 if (rc) {
 541                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 542                         repbody->aclsize = 0;
 543                 } else {
 544                         repbody->valid |= OBD_MD_FLRMTPERM;
 545                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 546                 }
 547         }
 548 #ifdef CONFIG_FS_POSIX_ACL
 549         else if (exp->exp_connect_flags & OBD_CONNECT_ACL) {
 550                 const struct lu_env *env = info->mti_env;
 551                 struct md_object *next = mdt_object_child(o);
 552                 struct lu_buf *buf = &info->mti_buf;
 553
 554                 buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 555                 buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 556                                                    RCL_SERVER);
 557                 if (buf->lb_len > 0) {
 558                         rc = mo_xattr_get(env, next, buf,
 559                                           XATTR_NAME_ACL_ACCESS);
 560                         if (rc < 0) {
 561                                 if (rc == -ENODATA) {
 562                                         repbody->aclsize = 0;
 563                                         repbody->valid |= OBD_MD_FLACL;
 564                                         rc = 0;
 565                                 } else if (rc == -EOPNOTSUPP) {
 566                                         rc = 0;
 567                                 } else {
 568                                         CERROR("got acl size: %d\n", rc);
 569                                 }
 570                         } else {
 571                                 repbody->aclsize = rc;
 572                                 repbody->valid |= OBD_MD_FLACL;
 573                                 rc = 0;
 574                         }
 575                 }
 576         }
 577 #endif
 578
 579         if (info->mti_mdt->mdt_opts.mo_mds_capa &&
 580             exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 581                 struct lustre_capa *capa;
 582
 583                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 584                 LASSERT(capa);
 585                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 586                 rc = mo_capa_get(info->mti_env, mdt_object_child(o), capa, 0);
 587                 if (rc)
 588                         RETURN(rc);
 589                 repbody->valid |= OBD_MD_FLMDSCAPA;
 590         }
 591         if (info->mti_mdt->mdt_opts.mo_oss_capa &&
 592             exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
 593             S_ISREG(lu_object_attr(&o->mot_obj.mo_lu))) {
 594                 struct lustre_capa *capa;
 595
 596                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 597                 LASSERT(capa);
 598                 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | capa_open_opc(flags);
 599                 rc = mo_capa_get(info->mti_env, mdt_object_child(o), capa, 0);
 600                 if (rc)
 601                         RETURN(rc);
 602                 repbody->valid |= OBD_MD_FLOSSCAPA;
 603         }
 604
 605         /*
 606          * If we are following a symlink, don't open; and do not return open
 607          * handle for special nodes as client required.
 608          */
 609         if (islnk || (!isreg && !isdir &&
 610             (req->rq_export->exp_connect_flags & OBD_CONNECT_NODEVOH))) {
 611                 lustre_msg_set_transno(req->rq_repmsg, 0);
 612                 RETURN(0);
 613         }
 614
 615         mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
 616
 617         /*
 618          * We need to return the existing object's fid back, so it is done here,
 619          * after preparing the reply.
 620          */
 621         if (!created && (flags & MDS_OPEN_EXCL) && (flags & MDS_OPEN_CREAT))
 622                 RETURN(-EEXIST);
 623
 624         /* This can't be done earlier, we need to return reply body */
 625         if (isdir) {
 626                 if (flags & (MDS_OPEN_CREAT | FMODE_WRITE)) {
 627                         /* We are trying to create or write an existing dir. */
 628                         RETURN(-EISDIR);
 629                 }
 630         } else if (flags & MDS_OPEN_DIRECTORY)
 631                 RETURN(-ENOTDIR);
 632
 633         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
 634                                  OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE)) {
 635                 RETURN(-EAGAIN);
 636         }
 637
 638         mfd = NULL;
 639         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 640                 spin_lock(&med->med_open_lock);
 641                 list_for_each(t, &med->med_open_head) {
 642                         mfd = list_entry(t, struct mdt_file_data, mfd_list);
 643                         if (mfd->mfd_xid == req->rq_xid) {
 644                                 break;
 645                         }
 646                         mfd = NULL;
 647                 }
 648                 spin_unlock(&med->med_open_lock);
 649
 650                 if (mfd != NULL) {
 651                         repbody->handle.cookie = mfd->mfd_handle.h_cookie;
 652                         /*set repbody->ea_size for resent case*/
 653                         if (ma->ma_valid & MA_LOV) {
 654                                 LASSERT(ma->ma_lmm_size != 0);
 655                                 repbody->eadatasize = ma->ma_lmm_size;
 656                                 if (isdir)
 657                                         repbody->valid |= OBD_MD_FLDIREA;
 658                                 else
 659                                         repbody->valid |= OBD_MD_FLEASIZE;
 660                         }
 661                         RETURN(0);
 662                 }
 663         }
 664
 665         rc = mdt_mfd_open(info, p, o, flags, created);
 666         RETURN(rc);
 667 }
 668
 669 extern void mdt_req_from_lcd(struct ptlrpc_request *req,
 670                              struct lsd_client_data *lcd);
 671
 672 void mdt_reconstruct_open(struct mdt_thread_info *info,
 673                           struct mdt_lock_handle *lhc)
 674 {
 675         const struct lu_env *env = info->mti_env;
 676         struct mdt_device       *mdt  = info->mti_mdt;
 677         struct req_capsule      *pill = info->mti_pill;
 678         struct ptlrpc_request   *req  = mdt_info_req(info);
 679         struct mdt_export_data  *med  = &req->rq_export->exp_mdt_data;
 680         struct lsd_client_data  *lcd  = med->med_lcd;
 681         struct md_attr          *ma   = &info->mti_attr;
 682         struct mdt_reint_record *rr   = &info->mti_rr;
 683         __u32                   flags = info->mti_spec.sp_cr_flags;
 684         struct ldlm_reply       *ldlm_rep;
 685         struct mdt_object       *parent;
 686         struct mdt_object       *child;
 687         struct mdt_body         *repbody;
 688         int                      rc;
 689         ENTRY;
 690
 691         LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
 692         ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
 693         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 694
 695         ma->ma_lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 696         ma->ma_lmm_size = req_capsule_get_size(pill, &RMF_MDT_MD,
 697                                                RCL_SERVER);
 698         ma->ma_need = MA_INODE;
 699         if (ma->ma_lmm_size > 0)
 700                 ma->ma_need |= MA_LOV;
 701
 702         ma->ma_valid = 0;
 703
 704         mdt_req_from_lcd(req, med->med_lcd);
 705         mdt_set_disposition(info, ldlm_rep, lcd->lcd_last_data);
 706
 707         CERROR("This is reconstruct open: disp="LPX64", result=%d\n",
 708                 ldlm_rep->lock_policy_res1, req->rq_status);
 709
 710         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
 711             req->rq_status != 0)
 712                 /* We did not create successfully, return error to client. */
 713                 GOTO(out, rc = req->rq_status);
 714
 715         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
 716                 struct obd_export *exp = req->rq_export;
 717                 /*
 718                  * We failed after creation, but we do not know in which step
 719                  * we failed. So try to check the child object.
 720                  */
 721                 parent = mdt_object_find(env, mdt, rr->rr_fid1);
 722                 if (IS_ERR(parent)) {
 723                         rc = PTR_ERR(parent);
 724                         LCONSOLE_WARN("Parent "DFID" lookup error %d."
 725                                       " Evicting client %s with export %s.\n",
 726                                       PFID(mdt_object_fid(parent)), rc,
 727                                       obd_uuid2str(&exp->exp_client_uuid),
 728                                       obd_export_nid2str(exp));
 729                         mdt_export_evict(exp);
 730                         EXIT;
 731                         return;
 732                 }
 733                 child = mdt_object_find(env, mdt, rr->rr_fid2);
 734                 if (IS_ERR(child)) {
 735                         rc = PTR_ERR(child);
 736                         LCONSOLE_WARN("Child "DFID" lookup error %d."
 737                                       " Evicting client %s with export %s.\n",
 738                                       PFID(mdt_object_fid(child)), rc,
 739                                       obd_uuid2str(&exp->exp_client_uuid),
 740                                       obd_export_nid2str(exp));
 741                         mdt_export_evict(exp);
 742                         EXIT;
 743                         return;
 744                 }
 745                 rc = mdt_object_exists(child);
 746                 if (rc > 0) {
 747                         struct md_object *next;
 748
 749                         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
 750                         next = mdt_object_child(child);
 751                         rc = mo_attr_get(env, next, ma);
 752                         if (rc == 0)
 753                               rc = mdt_finish_open(info, parent, child,
 754                                                    flags, 1, ldlm_rep);
 755                 } else if (rc < 0) {
 756                         /* the child object was created on remote server */
 757                         repbody->fid1 = *rr->rr_fid2;
 758                         repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
 759                         rc = 0;
 760                 } else if (rc == 0) {
 761                         /* the child does not exist, we should do regular open */
 762                         mdt_object_put(env, parent);
 763                         mdt_object_put(env, child);
 764                         GOTO(regular_open, 0);
 765                 }
 766                 mdt_object_put(env, parent);
 767                 mdt_object_put(env, child);
 768                 GOTO(out, rc);
 769         } else {
 770 regular_open:
 771                 /* We did not try to create, so we are a pure open */
 772                 rc = mdt_reint_open(info, lhc);
 773         }
 774
 775         EXIT;
 776 out:
 777         req->rq_status = rc;
 778         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
 779         LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
 780 }
 781
 782 static int mdt_open_by_fid(struct mdt_thread_info* info,
 783                            struct ldlm_reply *rep)
 784 {
 785         const struct lu_env     *env = info->mti_env;
 786         __u32                    flags = info->mti_spec.sp_cr_flags;
 787         struct mdt_reint_record *rr = &info->mti_rr;
 788         struct md_attr          *ma = &info->mti_attr;
 789         struct mdt_object       *o;
 790         int                      rc;
 791         ENTRY;
 792
 793         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
 794         if (IS_ERR(o))
 795                 RETURN(rc = PTR_ERR(o));
 796
 797         rc = mdt_object_exists(o);
 798         if (rc > 0) {
 799                 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 800                                                 DISP_LOOKUP_EXECD |
 801                                                 DISP_LOOKUP_POS));
 802
 803                 rc = mo_attr_get(env, mdt_object_child(o), ma);
 804                 if (rc == 0)
 805                         rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
 806         } else if (rc == 0) {
 807                 rc = -ENOENT;
 808         } else  {
 809                 /* the child object was created on remote server */
 810                 struct mdt_body *repbody;
 811                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 812                 repbody->fid1 = *rr->rr_fid2;
 813                 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
 814                 rc = 0;
 815         }
 816
 817         mdt_object_put(info->mti_env, o);
 818         RETURN(rc);
 819 }
 820
 821 static int mdt_open_anon_by_fid(struct mdt_thread_info* info,
 822                                 struct ldlm_reply *rep,
 823                                 struct mdt_lock_handle *lhc)
 824 {
 825         __u32                    flags = info->mti_spec.sp_cr_flags;
 826         struct mdt_reint_record *rr = &info->mti_rr;
 827         struct md_attr          *ma = &info->mti_attr;
 828         struct mdt_object       *o;
 829         int                      rc;
 830         ldlm_mode_t              lm;
 831         ENTRY;
 832
 833         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
 834         if (IS_ERR(o))
 835                 RETURN(rc = PTR_ERR(o));
 836
 837         rc = mdt_object_exists(o);
 838         if (rc == 0) {
 839                 mdt_set_disposition(info, rep, (DISP_LOOKUP_EXECD |
 840                                     DISP_LOOKUP_NEG));
 841                 GOTO(out, rc = -ENOENT);
 842         } else if (rc < 0) {
 843                 CERROR("NFS remote open shouldn't happen.\n");
 844                 GOTO(out, rc);
 845         }
 846
 847         mdt_set_disposition(info, rep, (DISP_IT_EXECD |
 848                                         DISP_LOOKUP_EXECD |
 849                                         DISP_LOOKUP_POS));
 850
 851         if (flags & FMODE_WRITE)
 852                 lm = LCK_CW;
 853         else if (flags & MDS_FMODE_EXEC)
 854                 lm = LCK_PR;
 855         else
 856                 lm = LCK_CR;
 857
 858         mdt_lock_handle_init(lhc);
 859         mdt_lock_reg_init(lhc, lm);
 860         rc = mdt_object_lock(info, o, lhc,
 861                              MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN,
 862                              MDT_CROSS_LOCK);
 863         if (rc)
 864                 GOTO(out, rc);
 865
 866         rc = mo_attr_get(info->mti_env, mdt_object_child(o), ma);
 867         if (rc)
 868                 GOTO(out, rc);
 869
 870         if (flags & MDS_OPEN_LOCK)
 871                 mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
 872         rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
 873
 874         if (!(flags & MDS_OPEN_LOCK))
 875                 mdt_object_unlock(info, o, lhc, 1);
 876
 877         GOTO(out, rc);
 878 out:
 879         mdt_object_put(info->mti_env, o);
 880         return rc;
 881 }
 882
 883 int mdt_pin(struct mdt_thread_info* info)
 884 {
 885         ENTRY;
 886         RETURN(err_serious(-EOPNOTSUPP));
 887 }
 888
 889 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
 890 static int mdt_cross_open(struct mdt_thread_info* info,
 891                           const struct lu_fid *fid,
 892                           struct ldlm_reply *rep, __u32 flags)
 893 {
 894         struct md_attr    *ma = &info->mti_attr;
 895         struct mdt_object *o;
 896         int                rc;
 897         ENTRY;
 898
 899         o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
 900         if (IS_ERR(o))
 901                 RETURN(rc = PTR_ERR(o));
 902
 903         rc = mdt_object_exists(o);
 904         if (rc > 0) {
 905                 /* Do permission check for cross-open. */
 906                 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o),
 907                                    NULL, flags | MDS_OPEN_CROSS);
 908                 if (rc)
 909                         goto out;
 910
 911                 mdt_set_capainfo(info, 0, fid, BYPASS_CAPA);
 912                 rc = mo_attr_get(info->mti_env, mdt_object_child(o), ma);
 913                 if (rc == 0)
 914                         rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
 915         } else if (rc == 0) {
 916                 /*
 917                  * Something is wrong here. lookup was positive but there is
 918                  * no object!
 919                  */
 920                 CERROR("Cross-ref object doesn't exist!\n");
 921                 rc = -EFAULT;
 922         } else  {
 923                 /* Something is wrong here, the object is on another MDS! */
 924                 CERROR("The object isn't on this server! FLD error?\n");
 925                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 926                                 &o->mot_obj.mo_lu,
 927                                 "Object isn't on this server! FLD error?\n");
 928
 929                 rc = -EFAULT;
 930         }
 931
 932 out:
 933         mdt_object_put(info->mti_env, o);
 934         RETURN(rc);
 935 }
 936
 937 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
 938 {
 939         struct mdt_device       *mdt = info->mti_mdt;
 940         struct ptlrpc_request   *req = mdt_info_req(info);
 941         struct mdt_object       *parent;
 942         struct mdt_object       *child;
 943         struct mdt_lock_handle  *lh;
 944         struct ldlm_reply       *ldlm_rep;
 945         struct mdt_body         *repbody;
 946         struct lu_fid           *child_fid = &info->mti_tmp_fid1;
 947         struct md_attr          *ma = &info->mti_attr;
 948         __u32                    create_flags = info->mti_spec.sp_cr_flags;
 949         struct mdt_reint_record *rr = &info->mti_rr;
 950         struct lu_name          *lname;
 951         int                      result, rc;
 952         int                      created = 0;
 953         __u32                    msg_flags;
 954         ENTRY;
 955
 956         OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
 957                                (obd_timeout + 1) / 4);
 958
 959         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 960
 961         ma->ma_lmm = req_capsule_server_get(info->mti_pill, &RMF_MDT_MD);
 962         ma->ma_lmm_size = req_capsule_get_size(info->mti_pill, &RMF_MDT_MD,
 963                                                RCL_SERVER);
 964         ma->ma_need = MA_INODE;
 965         if (ma->ma_lmm_size > 0)
 966                 ma->ma_need |= MA_LOV;
 967
 968         ma->ma_valid = 0;
 969
 970         LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
 971         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
 972
 973         /* JOIN file was deprecated since 1.6.5, but may be revived one day */
 974         if (create_flags & MDS_OPEN_JOIN_FILE) {
 975                 CERROR("file join is unsupported in this version of Lustre\n");
 976                 GOTO(out, result = err_serious(-EOPNOTSUPP));
 977         }
 978         msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
 979
 980         if ((create_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
 981             info->mti_spec.u.sp_ea.eadata == NULL)
 982                 GOTO(out, result = err_serious(-EINVAL));
 983
 984         CDEBUG(D_INODE, "I am going to open "DFID"/(%s->"DFID") "
 985                "cr_flag=0%o mode=0%06o msg_flag=0x%x\n",
 986                PFID(rr->rr_fid1), rr->rr_name,
 987                PFID(rr->rr_fid2), create_flags,
 988                ma->ma_attr.la_mode, msg_flags);
 989
 990         if (msg_flags & MSG_REPLAY ||
 991             (req->rq_export->exp_libclient && create_flags&MDS_OPEN_HAS_EA)) {
 992                 /* This is a replay request or from liblustre with ea. */
 993                 result = mdt_open_by_fid(info, ldlm_rep);
 994
 995                 if (result != -ENOENT) {
 996                         if (req->rq_export->exp_libclient &&
 997                             create_flags&MDS_OPEN_HAS_EA)
 998                                 GOTO(out, result = 0);
 999                         GOTO(out, result);
1000                 }
1001                 /*
1002                  * We didn't find the correct object, so we need to re-create it
1003                  * via a regular replay.
1004                  */
1005                 if (!(create_flags & MDS_OPEN_CREAT)) {
1006                         DEBUG_REQ(D_ERROR, req,"OPEN & CREAT not in open replay.");
1007                         GOTO(out, result = -EFAULT);
1008                 }
1009                 CDEBUG(D_INFO, "Open replay did find object, continue as "
1010                        "regular open\n");
1011         } else if (rr->rr_namelen == 0 && !info->mti_cross_ref) {
1012                 result = mdt_open_anon_by_fid(info, ldlm_rep, lhc);
1013                 if (result != -ENOENT)
1014                         GOTO(out, result);
1015         }
1016
1017         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1018                 GOTO(out, result = err_serious(-ENOMEM));
1019
1020         mdt_set_disposition(info, ldlm_rep,
1021                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1022
1023         if (info->mti_cross_ref) {
1024                 /* This is cross-ref open */
1025                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1026                 result = mdt_cross_open(info, rr->rr_fid1, ldlm_rep,
1027                                         create_flags);
1028                 GOTO(out, result);
1029         }
1030
1031         lh = &info->mti_lh[MDT_LH_PARENT];
1032         mdt_lock_pdo_init(lh, (create_flags & MDS_OPEN_CREAT) ?
1033                           LCK_PW : LCK_PR, rr->rr_name, rr->rr_namelen);
1034
1035         parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
1036                                       MDS_INODELOCK_UPDATE);
1037         if (IS_ERR(parent))
1038                 GOTO(out, result = PTR_ERR(parent));
1039
1040         fid_zero(child_fid);
1041
1042         lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
1043
1044         result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1045                             lname, child_fid, &info->mti_spec);
1046         LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1047                  "looking for "DFID"/%s, result fid="DFID"\n",
1048                  PFID(mdt_object_fid(parent)), rr->rr_name, PFID(child_fid));
1049
1050         if (result != 0 && result != -ENOENT && result != -ESTALE)
1051                 GOTO(out_parent, result);
1052
1053         if (result == -ENOENT || result == -ESTALE) {
1054                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1055                 if (result == -ESTALE) {
1056                         /*
1057                          * -ESTALE means the parent is a dead(unlinked) dir, so
1058                          * it should return -ENOENT to in accordance with the
1059                          * original mds implementaion.
1060                          */
1061                         GOTO(out_parent, result = -ENOENT);
1062                 }
1063                 if (!(create_flags & MDS_OPEN_CREAT))
1064                         GOTO(out_parent, result);
1065                 *child_fid = *info->mti_rr.rr_fid2;
1066                 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1067                          PFID(child_fid));
1068         } else {
1069                 /*
1070                  * Check for O_EXCL is moved to the mdt_finish_open(), we need to
1071                  * return FID back in that case.
1072                  */
1073                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1074         }
1075
1076         child = mdt_object_find(info->mti_env, mdt, child_fid);
1077         if (IS_ERR(child))
1078                 GOTO(out_parent, result = PTR_ERR(child));
1079
1080         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1081         if (result == -ENOENT) {
1082                 /* Not found and with MDS_OPEN_CREAT: let's create it. */
1083                 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1084
1085                 info->mti_mos[0] = parent;
1086                 info->mti_mos[1] = child;
1087                 result = mdt_version_get_check(info, 0);
1088                 if (result)
1089                         GOTO(out_child, result);
1090
1091                 /* Let lower layers know what is lock mode on directory. */
1092                 info->mti_spec.sp_cr_mode =
1093                         mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
1094
1095                 /*
1096                  * Do not perform lookup sanity check. We know that name does
1097                  * not exist.
1098                  */
1099                 info->mti_spec.sp_cr_lookup = 0;
1100                 info->mti_spec.sp_feat = &dt_directory_features;
1101
1102                 result = mdo_create(info->mti_env,
1103                                     mdt_object_child(parent),
1104                                     lname,
1105                                     mdt_object_child(child),
1106                                     &info->mti_spec,
1107                                     &info->mti_attr);
1108                 if (result == -ERESTART) {
1109                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1110                         GOTO(out_child, result);
1111                 } else {
1112                         if (result != 0)
1113                                 GOTO(out_child, result);
1114                 }
1115                 created = 1;
1116         } else {
1117                 /* We have to get attr & lov ea for this object */
1118                 result = mo_attr_get(info->mti_env, mdt_object_child(child),
1119                                      ma);
1120                 /*
1121                  * The object is on remote node, return its FID for remote open.
1122                  */
1123                 if (result == -EREMOTE) {
1124                         /*
1125                          * Check if this lock already was sent to client and
1126                          * this is resent case. For resent case do not take lock
1127                          * again, use what is already granted.
1128                          */
1129                         LASSERT(lhc != NULL);
1130
1131                         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1132                                 struct ldlm_lock *lock;
1133
1134                                 LASSERT(msg_flags & MSG_RESENT);
1135
1136                                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1137                                 if (!lock) {
1138                                         CERROR("Invalid lock handle "LPX64"\n",
1139                                                lhc->mlh_reg_lh.cookie);
1140                                         LBUG();
1141                                 }
1142                                 LASSERT(fid_res_name_eq(mdt_object_fid(child),
1143                                                         &lock->l_resource->lr_name));
1144                                 LDLM_LOCK_PUT(lock);
1145                                 rc = 0;
1146                         } else {
1147                                 mdt_lock_handle_init(lhc);
1148                                 mdt_lock_reg_init(lhc, LCK_PR);
1149
1150                                 rc = mdt_object_lock(info, child, lhc,
1151                                                      MDS_INODELOCK_LOOKUP,
1152                                                      MDT_CROSS_LOCK);
1153                         }
1154                         repbody->fid1 = *mdt_object_fid(child);
1155                         repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
1156                         if (rc != 0)
1157                                 result = rc;
1158                         GOTO(out_child, result);
1159                 }
1160         }
1161
1162         LASSERT(!lustre_handle_is_used(&lhc->mlh_reg_lh));
1163
1164         /* get openlock if this is not replay and if a client requested it */
1165         if (!(msg_flags & MSG_REPLAY) && create_flags & MDS_OPEN_LOCK) {
1166                 ldlm_mode_t lm;
1167
1168                 LASSERT(!created);
1169                 if (create_flags & FMODE_WRITE)
1170                         lm = LCK_CW;
1171                 else if (create_flags & MDS_FMODE_EXEC)
1172                         lm = LCK_PR;
1173                 else
1174                         lm = LCK_CR;
1175                 mdt_lock_handle_init(lhc);
1176                 mdt_lock_reg_init(lhc, lm);
1177                 rc = mdt_object_lock(info, child, lhc,
1178                                      MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN,
1179                                      MDT_CROSS_LOCK);
1180                 if (rc) {
1181                         result = rc;
1182                         GOTO(out_child, result);
1183                 } else {
1184                         result = -EREMOTE;
1185                         mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1186                 }
1187         }
1188
1189         /* Try to open it now. */
1190         rc = mdt_finish_open(info, parent, child, create_flags,
1191                              created, ldlm_rep);
1192         if (rc) {
1193                 result = rc;
1194                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
1195                         /* openlock was acquired and mdt_finish_open failed -
1196                            drop the openlock */
1197                         mdt_object_unlock(info, child, lhc, 1);
1198                 if (created) {
1199                         ma->ma_need = 0;
1200                         ma->ma_valid = 0;
1201                         ma->ma_cookie_size = 0;
1202                         info->mti_no_need_trans = 1;
1203                         rc = mdo_unlink(info->mti_env,
1204                                         mdt_object_child(parent),
1205                                         mdt_object_child(child),
1206                                         lname,
1207                                         &info->mti_attr);
1208                         if (rc != 0)
1209                                 CERROR("Error in cleanup of open\n");
1210                 }
1211         }
1212         EXIT;
1213 out_child:
1214         mdt_object_put(info->mti_env, child);
1215 out_parent:
1216         mdt_object_unlock_put(info, parent, lh, result);
1217 out:
1218         if (result && result != -EREMOTE)
1219                 lustre_msg_set_transno(req->rq_repmsg, 0);
1220         return result;
1221 }
1222
1223 #define MFD_CLOSED(mode) (((mode) & ~(FMODE_EPOCH | FMODE_SOM | \
1224                                       FMODE_EPOCHLCK)) == FMODE_CLOSED)
1225
1226 static int mdt_mfd_closed(struct mdt_file_data *mfd)
1227 {
1228         return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_mode));
1229 }
1230
1231 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
1232 {
1233         struct mdt_object *o = mfd->mfd_object;
1234         struct md_object *next = mdt_object_child(o);
1235         struct md_attr *ma = &info->mti_attr;
1236         int rc = 0, ret = 0;
1237         int mode;
1238         ENTRY;
1239
1240         mode = mfd->mfd_mode;
1241
1242         if ((mode & FMODE_WRITE) || (mode & FMODE_EPOCHLCK)) {
1243                 mdt_write_put(info->mti_mdt, o);
1244                 ret = mdt_epoch_close(info, o);
1245         } else if (mode & MDS_FMODE_EXEC) {
1246                 mdt_write_allow(info->mti_mdt, o);
1247         } else if (mode & FMODE_EPOCH) {
1248                 ret = mdt_epoch_close(info, o);
1249         }
1250
1251         /* Update atime on close only. */
1252         if ((mode & MDS_FMODE_EXEC || mode & FMODE_READ || mode & FMODE_WRITE)
1253             && (ma->ma_valid & MA_INODE) && (ma->ma_attr.la_valid & LA_ATIME)) {
1254                 /* Set the atime only. */
1255                 ma->ma_attr.la_valid = LA_ATIME;
1256                 rc = mo_attr_set(info->mti_env, next, ma);
1257         }
1258
1259         ma->ma_need |= MA_INODE;
1260         ma->ma_valid = 0;
1261
1262         if (!MFD_CLOSED(mode))
1263                 rc = mo_close(info->mti_env, next, ma);
1264         else if (ret == -EAGAIN)
1265                 rc = mo_attr_get(info->mti_env, next, ma);
1266
1267         /* If the object is unlinked, do not try to re-enable SIZEONMDS */
1268         if ((ret == -EAGAIN) && (ma->ma_valid & MA_INODE) &&
1269             (ma->ma_attr.la_nlink == 0)) {
1270                 ret = 0;
1271         }
1272
1273         if ((ret == -EAGAIN) || (ret == 1)) {
1274                 struct mdt_export_data *med;
1275
1276                 /* The epoch has not closed or Size-on-MDS update is needed.
1277                  * Put mfd back into the list. */
1278                 LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
1279                 mdt_mfd_set_mode(mfd, (ret == 1 ? FMODE_EPOCH : FMODE_SOM));
1280
1281                 LASSERT(mdt_info_req(info));
1282                 med = &mdt_info_req(info)->rq_export->exp_mdt_data;
1283                 spin_lock(&med->med_open_lock);
1284                 list_add(&mfd->mfd_list, &med->med_open_head);
1285                 class_handle_hash_back(&mfd->mfd_handle);
1286                 spin_unlock(&med->med_open_lock);
1287
1288                 if (ret == 1) {
1289                         ret = 0;
1290                 } else {
1291                         CDEBUG(D_INODE, "Size-on-MDS attribute update is "
1292                                "needed on "DFID"\n", PFID(mdt_object_fid(o)));
1293                 }
1294         } else {
1295                 mdt_mfd_free(mfd);
1296                 mdt_object_put(info->mti_env, o);
1297         }
1298
1299         RETURN(rc ? rc : ret);
1300 }
1301
1302 int mdt_close(struct mdt_thread_info *info)
1303 {
1304         struct mdt_export_data *med;
1305         struct mdt_file_data   *mfd;
1306         struct mdt_object      *o;
1307         struct md_attr         *ma = &info->mti_attr;
1308         struct mdt_body        *repbody = NULL;
1309         struct ptlrpc_request  *req = mdt_info_req(info);
1310         int rc, ret = 0;
1311         ENTRY;
1312
1313         /* Close may come with the Size-on-MDS update. Unpack it. */
1314         rc = mdt_close_unpack(info);
1315         if (rc)
1316                 RETURN(err_serious(rc));
1317
1318         LASSERT(info->mti_epoch);
1319
1320         req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
1321                              info->mti_mdt->mdt_max_mdsize);
1322         req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
1323                              info->mti_mdt->mdt_max_cookiesize);
1324         rc = req_capsule_server_pack(info->mti_pill);
1325         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
1326                 if (rc == 0)
1327                         mdt_shrink_reply(info);
1328                 RETURN(lustre_msg_get_status(req->rq_repmsg));
1329         }
1330
1331         /* Continue to close handle even if we can not pack reply */
1332         if (rc == 0) {
1333                 repbody = req_capsule_server_get(info->mti_pill,
1334                                                  &RMF_MDT_BODY);
1335                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
1336                                                     &RMF_MDT_MD);
1337                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
1338                                                        &RMF_MDT_MD,
1339                                                        RCL_SERVER);
1340                 ma->ma_cookie = req_capsule_server_get(info->mti_pill,
1341                                                        &RMF_LOGCOOKIES);
1342                 ma->ma_cookie_size = req_capsule_get_size(info->mti_pill,
1343                                                           &RMF_LOGCOOKIES,
1344                                                           RCL_SERVER);
1345                 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
1346                 repbody->eadatasize = 0;
1347                 repbody->aclsize = 0;
1348         } else {
1349                 rc = err_serious(rc);
1350         }
1351
1352         med = &req->rq_export->exp_mdt_data;
1353         spin_lock(&med->med_open_lock);
1354         mfd = mdt_handle2mfd(info, &info->mti_epoch->handle);
1355         if (mdt_mfd_closed(mfd)) {
1356                 spin_unlock(&med->med_open_lock);
1357                 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
1358                        ": cookie = "LPX64"\n", PFID(info->mti_rr.rr_fid1),
1359                        info->mti_epoch->handle.cookie);
1360                 rc = err_serious(-ESTALE);
1361         } else {
1362                 class_handle_unhash(&mfd->mfd_handle);
1363                 list_del_init(&mfd->mfd_list);
1364                 spin_unlock(&med->med_open_lock);
1365
1366                 /* Do not lose object before last unlink. */
1367                 o = mfd->mfd_object;
1368                 mdt_object_get(info->mti_env, o);
1369                 ret = mdt_mfd_close(info, mfd);
1370                 if (repbody != NULL)
1371                         rc = mdt_handle_last_unlink(info, o, ma);
1372                 mdt_empty_transno(info);
1373                 mdt_object_put(info->mti_env, o);
1374         }
1375         if (repbody != NULL)
1376                 mdt_shrink_reply(info);
1377
1378         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
1379                 RETURN(err_serious(-ENOMEM));
1380
1381         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
1382                                  OBD_FAIL_MDS_CLOSE_NET_REP))
1383                 info->mti_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
1384         RETURN(rc ? rc : ret);
1385 }
1386
1387 int mdt_done_writing(struct mdt_thread_info *info)
1388 {
1389         struct mdt_body         *repbody = NULL;
1390         struct mdt_export_data  *med;
1391         struct mdt_file_data    *mfd;
1392         int rc;
1393         ENTRY;
1394
1395         rc = req_capsule_server_pack(info->mti_pill);
1396         if (rc)
1397                 RETURN(err_serious(rc));
1398
1399         repbody = req_capsule_server_get(info->mti_pill,
1400                                          &RMF_MDT_BODY);
1401         repbody->eadatasize = 0;
1402         repbody->aclsize = 0;
1403
1404         /* Done Writing may come with the Size-on-MDS update. Unpack it. */
1405         rc = mdt_close_unpack(info);
1406         if (rc)
1407                 RETURN(err_serious(rc));
1408
1409         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL))
1410                 RETURN(lustre_msg_get_status(mdt_info_req(info)->rq_repmsg));
1411
1412         med = &info->mti_exp->exp_mdt_data;
1413         spin_lock(&med->med_open_lock);
1414         mfd = mdt_handle2mfd(info, &info->mti_epoch->handle);
1415         if (mfd == NULL) {
1416                 spin_unlock(&med->med_open_lock);
1417                 CDEBUG(D_INODE, "no handle for done write: fid = "DFID
1418                        ": cookie = "LPX64" ioepoch = "LPU64"\n",
1419                        PFID(info->mti_rr.rr_fid1),
1420                        info->mti_epoch->handle.cookie,
1421                        info->mti_epoch->ioepoch);
1422                 RETURN(-ESTALE);
1423         }
1424
1425         LASSERT(mfd->mfd_mode == FMODE_EPOCH ||
1426                 mfd->mfd_mode == FMODE_EPOCHLCK);
1427         class_handle_unhash(&mfd->mfd_handle);
1428         list_del_init(&mfd->mfd_list);
1429         spin_unlock(&med->med_open_lock);
1430
1431         /* Set EPOCH CLOSE flag if not set by client. */
1432         info->mti_epoch->flags |= MF_EPOCH_CLOSE;
1433         info->mti_attr.ma_valid = 0;
1434         rc = mdt_mfd_close(info, mfd);
1435         mdt_empty_transno(info);
1436         RETURN(rc);
1437 }