lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65 /* process_config */
  66 #include <lustre_param.h>
  67
  68 #include "osd_internal.h"
  69 #include "osd_dynlocks.h"
  70
  71 /* llo_* api support */
  72 #include <md_object.h>
  73 #include <lustre_quota.h>
  74
  75 #include <ldiskfs/xattr.h>
  76
  77 int ldiskfs_pdo = 1;
  78 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  79                 "ldiskfs with parallel directory operations");
  80
  81 int ldiskfs_track_declares_assert;
  82 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  83                 "LBUG during tracking of declares");
  84
  85 /* Slab to allocate dynlocks */
  86 struct kmem_cache *dynlock_cachep;
  87
  88 static struct lu_kmem_descr ldiskfs_caches[] = {
  89         {
  90                 .ckd_cache = &dynlock_cachep,
  91                 .ckd_name  = "dynlock_cache",
  92                 .ckd_size  = sizeof(struct dynlock_handle)
  93         },
  94         {
  95                 .ckd_cache = NULL
  96         }
  97 };
  98
  99 static const char dot[] = ".";
 100 static const char dotdot[] = "..";
 101 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 102
 103 static const struct lu_object_operations      osd_lu_obj_ops;
 104 static const struct dt_object_operations      osd_obj_ops;
 105 static const struct dt_object_operations      osd_obj_ea_ops;
 106 static const struct dt_object_operations      osd_obj_otable_it_ops;
 107 static const struct dt_index_operations       osd_index_iam_ops;
 108 static const struct dt_index_operations       osd_index_ea_ops;
 109
 110 int osd_trans_declare_op2rb[] = {
 111         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 112         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 113         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 114         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 115         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 116         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 117         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 118         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 119         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 120         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 121         [OSD_OT_UPDATE]         = OSD_OT_MAX,
 122         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 123 };
 124
 125 static int osd_has_index(const struct osd_object *obj)
 126 {
 127         return obj->oo_dt.do_index_ops != NULL;
 128 }
 129
 130 static int osd_object_invariant(const struct lu_object *l)
 131 {
 132         return osd_invariant(osd_obj(l));
 133 }
 134
 135 /*
 136  * Concurrency: doesn't matter
 137  */
 138
 139 /*
 140  * Concurrency: doesn't matter
 141  */
 142 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 143 {
 144         struct osd_thread_info *oti = osd_oti_get(env);
 145         return oti->oti_w_locks > 0 && o->oo_owner == env;
 146 }
 147
 148 /*
 149  * Concurrency: doesn't access mutable data
 150  */
 151 static int osd_root_get(const struct lu_env *env,
 152                         struct dt_device *dev, struct lu_fid *f)
 153 {
 154         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 155         return 0;
 156 }
 157
 158 /*
 159  * OSD object methods.
 160  */
 161
 162 /*
 163  * Concurrency: no concurrent access is possible that early in object
 164  * life-cycle.
 165  */
 166 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 167                                           const struct lu_object_header *hdr,
 168                                           struct lu_device *d)
 169 {
 170         struct osd_object *mo;
 171
 172         OBD_ALLOC_PTR(mo);
 173         if (mo != NULL) {
 174                 struct lu_object *l;
 175
 176                 l = &mo->oo_dt.do_lu;
 177                 dt_object_init(&mo->oo_dt, NULL, d);
 178                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 179                 l->lo_ops = &osd_lu_obj_ops;
 180                 init_rwsem(&mo->oo_sem);
 181                 init_rwsem(&mo->oo_ext_idx_sem);
 182                 spin_lock_init(&mo->oo_guard);
 183                 return l;
 184         } else {
 185                 return NULL;
 186         }
 187 }
 188
 189 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 190                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 191 {
 192         int rc;
 193
 194         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 195         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 196                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 197         if (rc > 0) {
 198                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 199                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 200                 rc = 0;
 201                 lustre_lma_swab(lma);
 202                 /* Check LMA compatibility */
 203                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 204                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 205                               "for fid = "DFID", ino = %lu\n",
 206                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 207                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 208                               PFID(&lma->lma_self_fid), inode->i_ino);
 209                         rc = -EOPNOTSUPP;
 210                 }
 211         } else if (rc == 0) {
 212                 rc = -ENODATA;
 213         }
 214
 215         return rc;
 216 }
 217
 218 /*
 219  * retrieve object from backend ext fs.
 220  **/
 221 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 222                        struct osd_inode_id *id)
 223 {
 224         struct inode *inode = NULL;
 225
 226         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 227         if (IS_ERR(inode)) {
 228                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 229                        id->oii_ino, PTR_ERR(inode));
 230         } else if (id->oii_gen != OSD_OII_NOGEN &&
 231                    inode->i_generation != id->oii_gen) {
 232                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 233                        "i_generation = %u\n",
 234                        id->oii_ino, id->oii_gen, inode->i_generation);
 235                 iput(inode);
 236                 inode = ERR_PTR(-ESTALE);
 237         } else if (inode->i_nlink == 0) {
 238                 /* due to parallel readdir and unlink,
 239                 * we can have dead inode here. */
 240                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 241                 iput(inode);
 242                 inode = ERR_PTR(-ESTALE);
 243         } else if (is_bad_inode(inode)) {
 244                 CWARN("%.16s: bad inode: ino = %u\n",
 245                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 246                 iput(inode);
 247                 inode = ERR_PTR(-ENOENT);
 248         } else {
 249                 if (id->oii_gen == OSD_OII_NOGEN)
 250                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 251
 252                 /* Do not update file c/mtime in ldiskfs.
 253                  * NB: we don't have any lock to protect this because we don't
 254                  * have reference on osd_object now, but contention with
 255                  * another lookup + attr_set can't happen in the tiny window
 256                  * between if (...) and set S_NOCMTIME. */
 257                 if (!(inode->i_flags & S_NOCMTIME))
 258                         inode->i_flags |= S_NOCMTIME;
 259         }
 260         return inode;
 261 }
 262
 263 static struct inode *
 264 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 265              struct osd_inode_id *id, struct lu_fid *fid)
 266 {
 267         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 268         struct inode            *inode;
 269         int                      rc;
 270
 271         inode = osd_iget(info, dev, id);
 272         if (IS_ERR(inode))
 273                 return inode;
 274
 275         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 276         if (rc == 0) {
 277                 *fid = lma->lma_self_fid;
 278         } else if (rc == -ENODATA) {
 279                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 280                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 281                 else
 282                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 283         } else {
 284                 iput(inode);
 285                 inode = ERR_PTR(rc);
 286         }
 287         return inode;
 288 }
 289
 290 static struct inode *osd_iget_check(struct osd_thread_info *info,
 291                                     struct osd_device *dev,
 292                                     const struct lu_fid *fid,
 293                                     struct osd_inode_id *id,
 294                                     bool in_oi)
 295 {
 296         struct inode    *inode;
 297         int              rc     = 0;
 298         ENTRY;
 299
 300         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 301         if (IS_ERR(inode)) {
 302                 rc = PTR_ERR(inode);
 303                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
 304                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 305                                id->oii_ino, rc);
 306
 307                         GOTO(put, rc);
 308                 }
 309
 310                 goto check_oi;
 311         }
 312
 313         if (is_bad_inode(inode)) {
 314                 rc = -ENOENT;
 315                 if (!in_oi) {
 316                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 317
 318                         GOTO(put, rc);
 319                 }
 320
 321                 goto check_oi;
 322         }
 323
 324         if (id->oii_gen != OSD_OII_NOGEN &&
 325             inode->i_generation != id->oii_gen) {
 326                 rc = -ESTALE;
 327                 if (!in_oi) {
 328                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 329                                "oii_gen = %u, i_generation = %u\n",
 330                                id->oii_ino, id->oii_gen, inode->i_generation);
 331
 332                         GOTO(put, rc);
 333                 }
 334
 335                 goto check_oi;
 336         }
 337
 338         if (inode->i_nlink == 0) {
 339                 rc = -ENOENT;
 340                 if (!in_oi) {
 341                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 342
 343                         GOTO(put, rc);
 344                 }
 345
 346                 goto check_oi;
 347         }
 348
 349 check_oi:
 350         if (rc != 0) {
 351                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 352
 353                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 354                 /* XXX: There are some possible cases:
 355                  *      1. rc = 0.
 356                  *         Backup/restore caused the OI invalid.
 357                  *      2. rc = 0.
 358                  *         Someone unlinked the object but NOT removed
 359                  *         the OI mapping, such as mount target device
 360                  *         as ldiskfs, and modify something directly.
 361                  *      3. rc = -ENOENT.
 362                  *         Someone just removed the object between the
 363                  *         former oi_lookup and the iget. It is normal.
 364                  *      4. Other failure cases.
 365                  *
 366                  *      Generally, when the device is mounted, it will
 367                  *      auto check whether the system is restored from
 368                  *      file-level backup or not. We trust such detect
 369                  *      to distinguish the 1st case from the 2nd case. */
 370                 if (rc == 0) {
 371                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 372                             inode->i_generation == id->oii_gen)
 373                                 rc = -ENOENT;
 374                         else
 375                                 rc = -EREMCHG;
 376                 }
 377         } else {
 378                 if (id->oii_gen == OSD_OII_NOGEN)
 379                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 380
 381                 /* Do not update file c/mtime in ldiskfs.
 382                  * NB: we don't have any lock to protect this because we don't
 383                  * have reference on osd_object now, but contention with
 384                  * another lookup + attr_set can't happen in the tiny window
 385                  * between if (...) and set S_NOCMTIME. */
 386                 if (!(inode->i_flags & S_NOCMTIME))
 387                         inode->i_flags |= S_NOCMTIME;
 388         }
 389
 390         GOTO(put, rc);
 391
 392 put:
 393         if (rc != 0) {
 394                 if (!IS_ERR(inode))
 395                         iput(inode);
 396
 397                 inode = ERR_PTR(rc);
 398         }
 399
 400         return inode;
 401 }
 402
 403 /**
 404  * \retval +v: new filter_fid, does not contain self-fid
 405  * \retval 0:  filter_fid_old, contains self-fid
 406  * \retval -v: other failure cases
 407  */
 408 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 409                  struct dentry *dentry, struct lu_fid *fid)
 410 {
 411         struct filter_fid_old   *ff     = &info->oti_ff;
 412         struct ost_id           *ostid  = &info->oti_ostid;
 413         int                      rc;
 414
 415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 416         if (rc == sizeof(*ff)) {
 417                 rc = 0;
 418                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 419                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 420                 /* XXX: should use real OST index in the future. LU-3569 */
 421                 ostid_to_fid(fid, ostid, 0);
 422         } else if (rc == sizeof(struct filter_fid)) {
 423                 rc = 1;
 424         } else if (rc >= 0) {
 425                 rc = -EINVAL;
 426         }
 427
 428         return rc;
 429 }
 430
 431 static int osd_lma_self_repair(struct osd_thread_info *info,
 432                                struct osd_device *osd, struct inode *inode,
 433                                const struct lu_fid *fid, __u32 compat)
 434 {
 435         handle_t *jh;
 436         int       rc;
 437
 438         LASSERT(current->journal_info == NULL);
 439
 440         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 441                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 442         if (IS_ERR(jh)) {
 443                 rc = PTR_ERR(jh);
 444                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 445                       osd_name(osd), rc);
 446                 return rc;
 447         }
 448
 449         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 450         if (rc != 0)
 451                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 452                       osd_name(osd), rc);
 453         ldiskfs_journal_stop(jh);
 454         return rc;
 455 }
 456
 457 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 458 {
 459         struct osd_thread_info  *info   = osd_oti_get(env);
 460         struct osd_device       *osd    = osd_obj2dev(obj);
 461         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 462         struct inode            *inode  = obj->oo_inode;
 463         struct dentry           *dentry = &info->oti_obj_dentry;
 464         struct lu_fid           *fid    = NULL;
 465         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 466         int                      rc;
 467         ENTRY;
 468
 469         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 470         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 471                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 472         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 473                 fid = &lma->lma_self_fid;
 474                 rc = osd_get_idif(info, inode, dentry, fid);
 475                 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
 476                         /* For the given OST-object, if it has neither LMA nor
 477                          * FID in XATTR_NAME_FID, then the given FID (which is
 478                          * contained in the @obj, from client RPC for locating
 479                          * the OST-object) is trusted. We use it to generate
 480                          * the LMA. */
 481                         osd_lma_self_repair(info, osd, inode, rfid,
 482                                 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
 483                                 LMAC_FID_ON_OST : 0);
 484                         RETURN(0);
 485                 }
 486         }
 487
 488         if (unlikely(rc == -ENODATA))
 489                 RETURN(0);
 490
 491         if (rc < 0)
 492                 RETURN(rc);
 493
 494         if (rc > 0) {
 495                 rc = 0;
 496                 lustre_lma_swab(lma);
 497                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 498                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 499                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 500                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 501                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 502                               PFID(rfid), inode->i_ino);
 503                         rc = -EOPNOTSUPP;
 504                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
 505                         fid = &lma->lma_self_fid;
 506                 }
 507         }
 508
 509         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 510                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 511                         struct ost_id   *oi   = &info->oti_ostid;
 512                         struct lu_fid   *fid1 = &info->oti_fid3;
 513                         __u32            idx  = fid_idif_ost_idx(rfid);
 514
 515                         /* For old IDIF, the OST index is not part of the IDIF,
 516                          * Means that different OSTs may have the same IDIFs.
 517                          * Under such case, we need to make some compatible
 518                          * check to make sure to trigger OI scrub properly. */
 519                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 520                                 /* Given @rfid is new, LMA is old. */
 521                                 fid_to_ostid(fid, oi);
 522                                 ostid_to_fid(fid1, oi, idx);
 523                                 if (lu_fid_eq(fid1, rfid)) {
 524                                         if (osd->od_lma_self_repair)
 525                                                 osd_lma_self_repair(info, osd,
 526                                                         inode, rfid,
 527                                                         LMAC_FID_ON_OST);
 528                                         RETURN(0);
 529                                 }
 530                         }
 531                 }
 532
 533                 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
 534                        osd_name(osd), PFID(rfid), PFID(fid));
 535                 rc = -EREMCHG;
 536         }
 537
 538         RETURN(rc);
 539 }
 540
 541 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 542                           const struct lu_fid *fid,
 543                           const struct lu_object_conf *conf)
 544 {
 545         struct osd_thread_info *info;
 546         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 547         struct osd_device      *dev;
 548         struct osd_idmap_cache *oic;
 549         struct osd_inode_id    *id;
 550         struct inode           *inode;
 551         struct osd_scrub       *scrub;
 552         struct scrub_file      *sf;
 553         int                     result;
 554         int                     saved  = 0;
 555         bool                    in_oi  = false;
 556         bool                    triggered = false;
 557         ENTRY;
 558
 559         LINVRNT(osd_invariant(obj));
 560         LASSERT(obj->oo_inode == NULL);
 561         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 562
 563         dev = osd_dev(ldev);
 564         scrub = &dev->od_scrub;
 565         sf = &scrub->os_file;
 566         info = osd_oti_get(env);
 567         LASSERT(info);
 568         oic = &info->oti_cache;
 569
 570         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 571                 RETURN(-ENOENT);
 572
 573         /* For the object is created as locking anchor, or for the object to
 574          * be created on disk. No need to osd_oi_lookup() at here because FID
 575          * shouldn't never be re-used, if it's really a duplicate FID from
 576          * unexpected reason, we should be able to detect it later by calling
 577          * do_create->osd_oi_insert(). */
 578         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 579                 GOTO(out, result = 0);
 580
 581         /* Search order: 1. per-thread cache. */
 582         if (lu_fid_eq(fid, &oic->oic_fid) &&
 583             likely(oic->oic_dev == dev)) {
 584                 id = &oic->oic_lid;
 585                 goto iget;
 586         }
 587
 588         id = &info->oti_id;
 589         if (!list_empty(&scrub->os_inconsistent_items)) {
 590                 /* Search order: 2. OI scrub pending list. */
 591                 result = osd_oii_lookup(dev, fid, id);
 592                 if (result == 0)
 593                         goto iget;
 594         }
 595
 596         /* Search order: 3. OI files. */
 597         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 598         if (result == -ENOENT) {
 599                 if (!fid_is_norm(fid) ||
 600                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 601                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 602                                       sf->sf_oi_bitmap))
 603                         GOTO(out, result = 0);
 604
 605                 goto trigger;
 606         }
 607
 608         if (result != 0)
 609                 GOTO(out, result);
 610
 611         in_oi = true;
 612
 613 iget:
 614         inode = osd_iget_check(info, dev, fid, id, in_oi);
 615         if (IS_ERR(inode)) {
 616                 result = PTR_ERR(inode);
 617                 if (result == -ENOENT || result == -ESTALE) {
 618                         if (!in_oi)
 619                                 fid_zero(&oic->oic_fid);
 620
 621                         GOTO(out, result = -ENOENT);
 622                 } else if (result == -EREMCHG) {
 623
 624 trigger:
 625                         if (!in_oi)
 626                                 fid_zero(&oic->oic_fid);
 627
 628                         if (unlikely(triggered))
 629                                 GOTO(out, result = saved);
 630
 631                         triggered = true;
 632                         if (thread_is_running(&scrub->os_thread)) {
 633                                 result = -EINPROGRESS;
 634                         } else if (!dev->od_noscrub) {
 635                                 result = osd_scrub_start(dev);
 636                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
 637                                               "for "DFID", rc = %d [1]\n",
 638                                               osd_name(dev), PFID(fid),result);
 639                                 if (result == 0 || result == -EALREADY)
 640                                         result = -EINPROGRESS;
 641                                 else
 642                                         result = -EREMCHG;
 643                         }
 644
 645                         /* We still have chance to get the valid inode: for the
 646                          * object which is referenced by remote name entry, the
 647                          * object on the local MDT will be linked under the dir
 648                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 649                          *
 650                          * We do not know whether the object for the given FID
 651                          * is referenced by some remote name entry or not, and
 652                          * especially for DNE II, a multiple-linked object may
 653                          * have many name entries reside on many MDTs.
 654                          *
 655                          * To simplify the operation, OSD will not distinguish
 656                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 657                          * only happened for the RPC from other MDT during the
 658                          * OI scrub, or for the client side RPC with FID only,
 659                          * such as FID to path, or from old connected client. */
 660                         saved = result;
 661                         result = osd_lookup_in_remote_parent(info, dev,
 662                                                              fid, id);
 663                         if (result == 0) {
 664                                 in_oi = false;
 665                                 goto iget;
 666                         }
 667
 668                         result = saved;
 669                 }
 670
 671                 GOTO(out, result);
 672         }
 673
 674         obj->oo_inode = inode;
 675         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 676
 677         result = osd_check_lma(env, obj);
 678         if (result != 0) {
 679                 iput(inode);
 680                 obj->oo_inode = NULL;
 681                 if (result == -EREMCHG) {
 682                         if (!in_oi) {
 683                                 result = osd_oi_lookup(info, dev, fid, id,
 684                                                        OI_CHECK_FLD);
 685                                 if (result != 0) {
 686                                         fid_zero(&oic->oic_fid);
 687                                         GOTO(out, result);
 688                                 }
 689                         }
 690
 691                         goto trigger;
 692                 }
 693
 694                 GOTO(out, result);
 695         }
 696
 697         obj->oo_compat_dot_created = 1;
 698         obj->oo_compat_dotdot_created = 1;
 699
 700         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 701                 GOTO(out, result = 0);
 702
 703         LASSERT(obj->oo_hl_head == NULL);
 704         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 705         if (obj->oo_hl_head == NULL) {
 706                 obj->oo_inode = NULL;
 707                 iput(inode);
 708                 GOTO(out, result = -ENOMEM);
 709         }
 710         GOTO(out, result = 0);
 711
 712 out:
 713         LINVRNT(osd_invariant(obj));
 714         return result;
 715 }
 716
 717 /*
 718  * Concurrency: shouldn't matter.
 719  */
 720 static void osd_object_init0(struct osd_object *obj)
 721 {
 722         LASSERT(obj->oo_inode != NULL);
 723         obj->oo_dt.do_body_ops = &osd_body_ops;
 724         obj->oo_dt.do_lu.lo_header->loh_attr |=
 725                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 726 }
 727
 728 /*
 729  * Concurrency: no concurrent access is possible that early in object
 730  * life-cycle.
 731  */
 732 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 733                            const struct lu_object_conf *conf)
 734 {
 735         struct osd_object *obj = osd_obj(l);
 736         int result;
 737
 738         LINVRNT(osd_invariant(obj));
 739
 740         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 741                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 742                 l->lo_header->loh_attr |= LOHA_EXISTS;
 743                 return 0;
 744         }
 745
 746         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 747         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 748         if (result == 0 && obj->oo_inode != NULL)
 749                 osd_object_init0(obj);
 750
 751         LINVRNT(osd_invariant(obj));
 752         return result;
 753 }
 754
 755 /*
 756  * Concurrency: no concurrent access is possible that late in object
 757  * life-cycle.
 758  */
 759 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 760 {
 761         struct osd_object *obj = osd_obj(l);
 762
 763         LINVRNT(osd_invariant(obj));
 764
 765         dt_object_fini(&obj->oo_dt);
 766         if (obj->oo_hl_head != NULL)
 767                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 768         OBD_FREE_PTR(obj);
 769 }
 770
 771 /*
 772  * Concurrency: no concurrent access is possible that late in object
 773  * life-cycle.
 774  */
 775 static void osd_index_fini(struct osd_object *o)
 776 {
 777         struct iam_container *bag;
 778
 779         if (o->oo_dir != NULL) {
 780                 bag = &o->oo_dir->od_container;
 781                 if (o->oo_inode != NULL) {
 782                         if (bag->ic_object == o->oo_inode)
 783                                 iam_container_fini(bag);
 784                 }
 785                 OBD_FREE_PTR(o->oo_dir);
 786                 o->oo_dir = NULL;
 787         }
 788 }
 789
 790 /*
 791  * Concurrency: no concurrent access is possible that late in object
 792  * life-cycle (for all existing callers, that is. New callers have to provide
 793  * their own locking.)
 794  */
 795 static int osd_inode_unlinked(const struct inode *inode)
 796 {
 797         return inode->i_nlink == 0;
 798 }
 799
 800 enum {
 801         OSD_TXN_OI_DELETE_CREDITS    = 20,
 802         OSD_TXN_INODE_DELETE_CREDITS = 20
 803 };
 804
 805 /*
 806  * Journal
 807  */
 808
 809 #if OSD_THANDLE_STATS
 810 /**
 811  * Set time when the handle is allocated
 812  */
 813 static void osd_th_alloced(struct osd_thandle *oth)
 814 {
 815         oth->oth_alloced = cfs_time_current();
 816 }
 817
 818 /**
 819  * Set time when the handle started
 820  */
 821 static void osd_th_started(struct osd_thandle *oth)
 822 {
 823         oth->oth_started = cfs_time_current();
 824 }
 825
 826 /**
 827  * Helper function to convert time interval to microseconds packed in
 828  * long int.
 829  */
 830 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 831 {
 832         struct timeval val;
 833
 834         cfs_duration_usec(cfs_time_sub(end, start), &val);
 835         return val.tv_sec * 1000000 + val.tv_usec;
 836 }
 837
 838 /**
 839  * Check whether the we deal with this handle for too long.
 840  */
 841 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 842                                 cfs_time_t alloced, cfs_time_t started,
 843                                 cfs_time_t closed)
 844 {
 845         cfs_time_t now = cfs_time_current();
 846
 847         LASSERT(dev != NULL);
 848
 849         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 850                             interval_to_usec(alloced, started));
 851         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 852                             interval_to_usec(started, closed));
 853         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 854                             interval_to_usec(closed, now));
 855
 856         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 857                 CWARN("transaction handle %p was open for too long: "
 858                       "now "CFS_TIME_T" ,"
 859                       "alloced "CFS_TIME_T" ,"
 860                       "started "CFS_TIME_T" ,"
 861                       "closed "CFS_TIME_T"\n",
 862                       oth, now, alloced, started, closed);
 863                 libcfs_debug_dumpstack(NULL);
 864         }
 865 }
 866
 867 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 868 {                                                                       \
 869         cfs_time_t __closed = cfs_time_current();                       \
 870         cfs_time_t __alloced = oth->oth_alloced;                        \
 871         cfs_time_t __started = oth->oth_started;                        \
 872                                                                         \
 873         expr;                                                           \
 874         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 875 }
 876
 877 #else /* OSD_THANDLE_STATS */
 878
 879 #define osd_th_alloced(h)                  do {} while(0)
 880 #define osd_th_started(h)                  do {} while(0)
 881 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 882
 883 #endif /* OSD_THANDLE_STATS */
 884
 885 /*
 886  * Concurrency: doesn't access mutable data.
 887  */
 888 static int osd_param_is_not_sane(const struct osd_device *dev,
 889                                  const struct thandle *th)
 890 {
 891         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 892
 893         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
 894 }
 895
 896 /*
 897  * Concurrency: shouldn't matter.
 898  */
 899 static void osd_trans_commit_cb(struct super_block *sb,
 900                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 901 {
 902         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 903         struct thandle     *th  = &oh->ot_super;
 904         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 905         struct dt_txn_commit_cb *dcb, *tmp;
 906
 907         LASSERT(oh->ot_handle == NULL);
 908
 909         if (error)
 910                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 911
 912         dt_txn_hook_commit(th);
 913
 914         /* call per-transaction callbacks if any */
 915         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 916                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 917                          "commit callback entry: magic=%x name='%s'\n",
 918                          dcb->dcb_magic, dcb->dcb_name);
 919                 list_del_init(&dcb->dcb_linkage);
 920                 dcb->dcb_func(NULL, th, dcb, error);
 921         }
 922
 923         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 924         lu_device_put(lud);
 925         th->th_dev = NULL;
 926
 927         lu_context_exit(&th->th_ctx);
 928         lu_context_fini(&th->th_ctx);
 929         thandle_put(th);
 930 }
 931
 932 static struct thandle *osd_trans_create(const struct lu_env *env,
 933                                         struct dt_device *d)
 934 {
 935         struct osd_thread_info  *oti = osd_oti_get(env);
 936         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 937         struct osd_thandle      *oh;
 938         struct thandle          *th;
 939         ENTRY;
 940
 941         /* on pending IO in this thread should left from prev. request */
 942         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 943
 944         th = ERR_PTR(-ENOMEM);
 945         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 946         if (oh != NULL) {
 947                 oh->ot_quota_trans = &oti->oti_quota_trans;
 948                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 949                 th = &oh->ot_super;
 950                 th->th_dev = d;
 951                 th->th_result = 0;
 952                 th->th_tags = LCT_TX_HANDLE;
 953                 oh->ot_credits = 0;
 954                 atomic_set(&th->th_refc, 1);
 955                 th->th_alloc_size = sizeof(*oh);
 956                 oti->oti_dev = osd_dt_dev(d);
 957                 INIT_LIST_HEAD(&oh->ot_dcb_list);
 958                 osd_th_alloced(oh);
 959
 960                 memset(oti->oti_declare_ops, 0,
 961                                         sizeof(oti->oti_declare_ops));
 962                 memset(oti->oti_declare_ops_rb, 0,
 963                                         sizeof(oti->oti_declare_ops_rb));
 964                 memset(oti->oti_declare_ops_cred, 0,
 965                                         sizeof(oti->oti_declare_ops_cred));
 966                 oti->oti_rollback = false;
 967         }
 968         RETURN(th);
 969 }
 970
 971 /*
 972  * Concurrency: shouldn't matter.
 973  */
 974 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 975                     struct thandle *th)
 976 {
 977         struct osd_thread_info *oti = osd_oti_get(env);
 978         struct osd_device  *dev = osd_dt_dev(d);
 979         handle_t           *jh;
 980         struct osd_thandle *oh;
 981         int rc;
 982
 983         ENTRY;
 984
 985         LASSERT(current->journal_info == NULL);
 986
 987         oh = container_of0(th, struct osd_thandle, ot_super);
 988         LASSERT(oh != NULL);
 989         LASSERT(oh->ot_handle == NULL);
 990
 991         rc = dt_txn_hook_start(env, d, th);
 992         if (rc != 0)
 993                 GOTO(out, rc);
 994
 995         if (unlikely(osd_param_is_not_sane(dev, th))) {
 996                 static unsigned long last_printed;
 997                 static int last_credits;
 998
 999                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1000                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1001                       oh->ot_credits,
1002                       osd_journal(dev)->j_max_transaction_buffers);
1003                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1004                       oti->oti_declare_ops[OSD_OT_CREATE],
1005                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1006                       oti->oti_declare_ops[OSD_OT_DESTROY],
1007                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1008                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1009                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1010                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1011                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1012                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1013                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1014                       oti->oti_declare_ops[OSD_OT_WRITE],
1015                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1016                       oti->oti_declare_ops[OSD_OT_PUNCH],
1017                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1018                       oti->oti_declare_ops[OSD_OT_QUOTA],
1019                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1020                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1021                       oti->oti_declare_ops[OSD_OT_INSERT],
1022                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1023                       oti->oti_declare_ops[OSD_OT_DELETE],
1024                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1025                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1026                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1027                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1028                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1029                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1030
1031                 if (last_credits != oh->ot_credits &&
1032                     time_after(jiffies, last_printed + 60 * HZ)) {
1033                         libcfs_debug_dumpstack(NULL);
1034                         last_credits = oh->ot_credits;
1035                         last_printed = jiffies;
1036                 }
1037                 /* XXX Limit the credits to 'max_transaction_buffers', and
1038                  *     let the underlying filesystem to catch the error if
1039                  *     we really need so many credits.
1040                  *
1041                  *     This should be removed when we can calculate the
1042                  *     credits precisely. */
1043                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
1044         }
1045
1046         /*
1047          * XXX temporary stuff. Some abstraction layer should
1048          * be used.
1049          */
1050         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1051         osd_th_started(oh);
1052         if (!IS_ERR(jh)) {
1053                 oh->ot_handle = jh;
1054                 LASSERT(oti->oti_txns == 0);
1055                 lu_context_init(&th->th_ctx, th->th_tags);
1056                 lu_context_enter(&th->th_ctx);
1057
1058                 lu_device_get(&d->dd_lu_dev);
1059                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1060                               "osd-tx", th);
1061                 oti->oti_txns++;
1062                 rc = 0;
1063         } else {
1064                 rc = PTR_ERR(jh);
1065         }
1066 out:
1067         RETURN(rc);
1068 }
1069
1070 static int osd_seq_exists(const struct lu_env *env,
1071                               struct osd_device *osd, obd_seq seq)
1072 {
1073         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1074         struct seq_server_site  *ss = osd_seq_site(osd);
1075         int                     rc;
1076         ENTRY;
1077
1078         if (ss == NULL)
1079                 RETURN(1);
1080
1081         rc = osd_fld_lookup(env, osd, seq, range);
1082         if (rc != 0) {
1083                 if (rc != -ENOENT)
1084                         CERROR("%s: can't lookup FLD sequence "LPX64
1085                                ": rc = %d\n", osd_name(osd), seq, rc);
1086                 RETURN(0);
1087         }
1088
1089         RETURN(ss->ss_node_id == range->lsr_index);
1090 }
1091
1092 /*
1093  * Concurrency: shouldn't matter.
1094  */
1095 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1096                           struct thandle *th)
1097 {
1098         int                     rc = 0;
1099         struct osd_thandle     *oh;
1100         struct osd_thread_info *oti = osd_oti_get(env);
1101         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1102         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1103         struct lquota_trans    *qtrans;
1104         ENTRY;
1105
1106         oh = container_of0(th, struct osd_thandle, ot_super);
1107
1108         qtrans = oh->ot_quota_trans;
1109         oh->ot_quota_trans = NULL;
1110
1111         if (oh->ot_handle != NULL) {
1112                 handle_t *hdl = oh->ot_handle;
1113
1114                 /*
1115                  * add commit callback
1116                  * notice we don't do this in osd_trans_start()
1117                  * as underlying transaction can change during truncate
1118                  */
1119                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1120                                          &oh->ot_jcb);
1121
1122                 LASSERT(oti->oti_txns == 1);
1123                 oti->oti_txns--;
1124                 rc = dt_txn_hook_stop(env, th);
1125                 if (rc != 0)
1126                         CERROR("Failure in transaction hook: %d\n", rc);
1127
1128                 /* hook functions might modify th_sync */
1129                 hdl->h_sync = th->th_sync;
1130
1131                 oh->ot_handle = NULL;
1132                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1133                                   rc = ldiskfs_journal_stop(hdl));
1134                 if (rc != 0)
1135                         CERROR("Failure to stop transaction: %d\n", rc);
1136         } else {
1137                 thandle_put(&oh->ot_super);
1138         }
1139
1140         /* inform the quota slave device that the transaction is stopping */
1141         qsd_op_end(env, qsd, qtrans);
1142
1143         /* as we want IO to journal and data IO be concurrent, we don't block
1144          * awaiting data IO completion in osd_do_bio(), instead we wait here
1145          * once transaction is submitted to the journal. all reqular requests
1146          * don't do direct IO (except read/write), thus this wait_event becomes
1147          * no-op for them.
1148          *
1149          * IMPORTANT: we have to wait till any IO submited by the thread is
1150          * completed otherwise iobuf may be corrupted by different request
1151          */
1152         wait_event(iobuf->dr_wait,
1153                        atomic_read(&iobuf->dr_numreqs) == 0);
1154         osd_fini_iobuf(oti->oti_dev, iobuf);
1155         if (!rc)
1156                 rc = iobuf->dr_error;
1157
1158         RETURN(rc);
1159 }
1160
1161 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1162 {
1163         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1164                                                ot_super);
1165
1166         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1167         LASSERT(&dcb->dcb_func != NULL);
1168         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1169
1170         return 0;
1171 }
1172
1173 /*
1174  * Called just before object is freed. Releases all resources except for
1175  * object itself (that is released by osd_object_free()).
1176  *
1177  * Concurrency: no concurrent access is possible that late in object
1178  * life-cycle.
1179  */
1180 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1181 {
1182         struct osd_object *obj   = osd_obj(l);
1183         struct inode      *inode = obj->oo_inode;
1184
1185         LINVRNT(osd_invariant(obj));
1186
1187         /*
1188          * If object is unlinked remove fid->ino mapping from object index.
1189          */
1190
1191         osd_index_fini(obj);
1192         if (inode != NULL) {
1193                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1194                 qid_t                    uid = inode->i_uid;
1195                 qid_t                    gid = inode->i_gid;
1196
1197                 iput(inode);
1198                 obj->oo_inode = NULL;
1199
1200                 if (qsd != NULL) {
1201                         struct osd_thread_info  *info = osd_oti_get(env);
1202                         struct lquota_id_info   *qi = &info->oti_qi;
1203
1204                         /* Release granted quota to master if necessary */
1205                         qi->lqi_id.qid_uid = uid;
1206                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1207
1208                         qi->lqi_id.qid_uid = gid;
1209                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1210                 }
1211         }
1212 }
1213
1214 /*
1215  * Concurrency: ->loo_object_release() is called under site spin-lock.
1216  */
1217 static void osd_object_release(const struct lu_env *env,
1218                                struct lu_object *l)
1219 {
1220 }
1221
1222 /*
1223  * Concurrency: shouldn't matter.
1224  */
1225 static int osd_object_print(const struct lu_env *env, void *cookie,
1226                             lu_printer_t p, const struct lu_object *l)
1227 {
1228         struct osd_object *o = osd_obj(l);
1229         struct iam_descr  *d;
1230
1231         if (o->oo_dir != NULL)
1232                 d = o->oo_dir->od_container.ic_descr;
1233         else
1234                 d = NULL;
1235         return (*p)(env, cookie,
1236                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1237                     o, o->oo_inode,
1238                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1239                     o->oo_inode ? o->oo_inode->i_generation : 0,
1240                     d ? d->id_ops->id_name : "plain");
1241 }
1242
1243 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1244
1245 /*
1246  * Concurrency: shouldn't matter.
1247  */
1248 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1249                struct obd_statfs *sfs)
1250 {
1251         struct osd_device  *osd = osd_dt_dev(d);
1252         struct super_block *sb = osd_sb(osd);
1253         struct kstatfs     *ksfs;
1254         int result = 0;
1255
1256         if (unlikely(osd->od_mnt == NULL))
1257                 return -EINPROGRESS;
1258
1259         /* osd_lproc.c call this without env, allocate ksfs for that case */
1260         if (unlikely(env == NULL)) {
1261                 OBD_ALLOC_PTR(ksfs);
1262                 if (ksfs == NULL)
1263                         return -ENOMEM;
1264         } else {
1265                 ksfs = &osd_oti_get(env)->oti_ksfs;
1266         }
1267
1268         spin_lock(&osd->od_osfs_lock);
1269         result = sb->s_op->statfs(sb->s_root, ksfs);
1270         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1271                 statfs_pack(sfs, ksfs);
1272                 if (sb->s_flags & MS_RDONLY)
1273                         sfs->os_state = OS_STATE_READONLY;
1274         }
1275
1276         spin_unlock(&osd->od_osfs_lock);
1277
1278         if (unlikely(env == NULL))
1279                 OBD_FREE_PTR(ksfs);
1280
1281         /* Reserve a small amount of space for local objects like last_rcvd,
1282          * llog, quota files, ... */
1283         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1284                 sfs->os_bavail = 0;
1285         } else {
1286                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1287                 /** Take out metadata overhead for indirect blocks */
1288                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1289         }
1290
1291         return result;
1292 }
1293
1294 /**
1295  * Estimate space needed for file creations. We assume the largest filename
1296  * which is 2^64 - 1, hence a filename of 20 chars.
1297  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1298  */
1299 #ifdef __LDISKFS_DIR_REC_LEN
1300 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1301 #else
1302 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1303 #endif
1304
1305 /*
1306  * Concurrency: doesn't access mutable data.
1307  */
1308 static void osd_conf_get(const struct lu_env *env,
1309                          const struct dt_device *dev,
1310                          struct dt_device_param *param)
1311 {
1312         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1313         int                ea_overhead;
1314
1315         /*
1316          * XXX should be taken from not-yet-existing fs abstraction layer.
1317          */
1318         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1319         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1320         param->ddp_block_shift  = sb->s_blocksize_bits;
1321         param->ddp_mount_type     = LDD_MT_LDISKFS;
1322         param->ddp_maxbytes       = sb->s_maxbytes;
1323         /* Overhead estimate should be fairly accurate, so we really take a tiny
1324          * error margin which also avoids fragmenting the filesystem too much */
1325         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1326         /* inode are statically allocated, so per-inode space consumption
1327          * is the space consumed by the directory entry */
1328         param->ddp_inodespace     = PER_OBJ_USAGE;
1329         /* per-fragment overhead to be used by the client code */
1330         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1331         param->ddp_mntopts      = 0;
1332         if (test_opt(sb, XATTR_USER))
1333                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1334         if (test_opt(sb, POSIX_ACL))
1335                 param->ddp_mntopts |= MNTOPT_ACL;
1336
1337         /* LOD might calculate the max stripe count based on max_ea_size,
1338          * so we need take account in the overhead as well,
1339          * xattr_header + magic + xattr_entry_head */
1340         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1341                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1342
1343 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1344         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1345                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1346                                                                 ea_overhead;
1347         else
1348 #endif
1349                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1350 }
1351
1352 /*
1353  * Concurrency: shouldn't matter.
1354  */
1355 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1356 {
1357         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1358         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1359 }
1360
1361 /**
1362  * Start commit for OSD device.
1363  *
1364  * An implementation of dt_commit_async method for OSD device.
1365  * Asychronously starts underlayng fs sync and thereby a transaction
1366  * commit.
1367  *
1368  * \param env environment
1369  * \param d dt device
1370  *
1371  * \see dt_device_operations
1372  */
1373 static int osd_commit_async(const struct lu_env *env,
1374                             struct dt_device *d)
1375 {
1376         struct super_block *s = osd_sb(osd_dt_dev(d));
1377         ENTRY;
1378
1379         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1380         RETURN(s->s_op->sync_fs(s, 0));
1381 }
1382
1383 /*
1384  * Concurrency: shouldn't matter.
1385  */
1386
1387 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1388 {
1389         struct super_block *sb = osd_sb(osd_dt_dev(d));
1390         struct block_device *dev = sb->s_bdev;
1391 #ifdef HAVE_DEV_SET_RDONLY
1392         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1393         int rc = 0;
1394 #else
1395         int rc = -EOPNOTSUPP;
1396 #endif
1397         ENTRY;
1398
1399 #ifdef HAVE_DEV_SET_RDONLY
1400         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1401
1402         if (jdev && (jdev != dev)) {
1403                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1404                        (long)jdev);
1405                 dev_set_rdonly(jdev);
1406         }
1407         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1408         dev_set_rdonly(dev);
1409 #else
1410         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1411                osd_dt_dev(d)->od_svname, (long)dev, rc);
1412 #endif
1413         RETURN(rc);
1414 }
1415
1416 /*
1417  * Concurrency: serialization provided by callers.
1418  */
1419 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1420                               int mode, unsigned long timeout, __u32 alg,
1421                               struct lustre_capa_key *keys)
1422 {
1423         struct osd_device *dev = osd_dt_dev(d);
1424         ENTRY;
1425
1426         dev->od_fl_capa = mode;
1427         dev->od_capa_timeout = timeout;
1428         dev->od_capa_alg = alg;
1429         dev->od_capa_keys = keys;
1430         RETURN(0);
1431 }
1432
1433 /**
1434  * Note: we do not count into QUOTA here.
1435  * If we mount with --data_journal we may need more.
1436  */
1437 const int osd_dto_credits_noquota[DTO_NR] = {
1438         /**
1439          * Insert.
1440          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1441          * SINGLEDATA_TRANS_BLOCKS(8)
1442          * XXX Note: maybe iam need more, since iam have more level than
1443          *           EXT3 htree.
1444          */
1445         [DTO_INDEX_INSERT]  = 16,
1446         /**
1447          * Delete
1448          * just modify a single entry, probably merge few within a block
1449          */
1450         [DTO_INDEX_DELETE]  = 1,
1451         /**
1452          * Used for OI scrub
1453          */
1454         [DTO_INDEX_UPDATE]  = 16,
1455         /**
1456          * 4(inode, inode bits, groups, GDT)
1457          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1458          */
1459         [DTO_OBJECT_CREATE] = 4,
1460         /**
1461          * 4(inode, inode bits, groups, GDT)
1462          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1463          */
1464         [DTO_OBJECT_DELETE] = 4,
1465         /**
1466          * Attr set credits (inode)
1467          */
1468         [DTO_ATTR_SET_BASE] = 1,
1469         /**
1470          * Xattr set. The same as xattr of EXT3.
1471          * DATA_TRANS_BLOCKS(14)
1472          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1473          * are also counted in. Do not know why?
1474          */
1475         [DTO_XATTR_SET]     = 14,
1476         /**
1477          * credits for inode change during write.
1478          */
1479         [DTO_WRITE_BASE]    = 3,
1480         /**
1481          * credits for single block write.
1482          */
1483         [DTO_WRITE_BLOCK]   = 14,
1484         /**
1485          * Attr set credits for chown.
1486          * This is extra credits for setattr, and it is null without quota
1487          */
1488         [DTO_ATTR_SET_CHOWN] = 0
1489 };
1490
1491 static const struct dt_device_operations osd_dt_ops = {
1492         .dt_root_get       = osd_root_get,
1493         .dt_statfs         = osd_statfs,
1494         .dt_trans_create   = osd_trans_create,
1495         .dt_trans_start    = osd_trans_start,
1496         .dt_trans_stop     = osd_trans_stop,
1497         .dt_trans_cb_add   = osd_trans_cb_add,
1498         .dt_conf_get       = osd_conf_get,
1499         .dt_sync           = osd_sync,
1500         .dt_ro             = osd_ro,
1501         .dt_commit_async   = osd_commit_async,
1502         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1503 };
1504
1505 static void osd_object_read_lock(const struct lu_env *env,
1506                                  struct dt_object *dt, unsigned role)
1507 {
1508         struct osd_object *obj = osd_dt_obj(dt);
1509         struct osd_thread_info *oti = osd_oti_get(env);
1510
1511         LINVRNT(osd_invariant(obj));
1512
1513         LASSERT(obj->oo_owner != env);
1514         down_read_nested(&obj->oo_sem, role);
1515
1516         LASSERT(obj->oo_owner == NULL);
1517         oti->oti_r_locks++;
1518 }
1519
1520 static void osd_object_write_lock(const struct lu_env *env,
1521                                   struct dt_object *dt, unsigned role)
1522 {
1523         struct osd_object *obj = osd_dt_obj(dt);
1524         struct osd_thread_info *oti = osd_oti_get(env);
1525
1526         LINVRNT(osd_invariant(obj));
1527
1528         LASSERT(obj->oo_owner != env);
1529         down_write_nested(&obj->oo_sem, role);
1530
1531         LASSERT(obj->oo_owner == NULL);
1532         obj->oo_owner = env;
1533         oti->oti_w_locks++;
1534 }
1535
1536 static void osd_object_read_unlock(const struct lu_env *env,
1537                                    struct dt_object *dt)
1538 {
1539         struct osd_object *obj = osd_dt_obj(dt);
1540         struct osd_thread_info *oti = osd_oti_get(env);
1541
1542         LINVRNT(osd_invariant(obj));
1543
1544         LASSERT(oti->oti_r_locks > 0);
1545         oti->oti_r_locks--;
1546         up_read(&obj->oo_sem);
1547 }
1548
1549 static void osd_object_write_unlock(const struct lu_env *env,
1550                                     struct dt_object *dt)
1551 {
1552         struct osd_object *obj = osd_dt_obj(dt);
1553         struct osd_thread_info *oti = osd_oti_get(env);
1554
1555         LINVRNT(osd_invariant(obj));
1556
1557         LASSERT(obj->oo_owner == env);
1558         LASSERT(oti->oti_w_locks > 0);
1559         oti->oti_w_locks--;
1560         obj->oo_owner = NULL;
1561         up_write(&obj->oo_sem);
1562 }
1563
1564 static int osd_object_write_locked(const struct lu_env *env,
1565                                    struct dt_object *dt)
1566 {
1567         struct osd_object *obj = osd_dt_obj(dt);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         return obj->oo_owner == env;
1572 }
1573
1574 static int capa_is_sane(const struct lu_env *env,
1575                         struct osd_device *dev,
1576                         struct lustre_capa *capa,
1577                         struct lustre_capa_key *keys)
1578 {
1579         struct osd_thread_info *oti = osd_oti_get(env);
1580         struct lustre_capa *tcapa = &oti->oti_capa;
1581         struct obd_capa *oc;
1582         int i, rc = 0;
1583         ENTRY;
1584
1585         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1586         if (oc) {
1587                 if (capa_is_expired(oc)) {
1588                         DEBUG_CAPA(D_ERROR, capa, "expired");
1589                         rc = -ESTALE;
1590                 }
1591                 capa_put(oc);
1592                 RETURN(rc);
1593         }
1594
1595         if (capa_is_expired_sec(capa)) {
1596                 DEBUG_CAPA(D_ERROR, capa, "expired");
1597                 RETURN(-ESTALE);
1598         }
1599
1600         spin_lock(&capa_lock);
1601         for (i = 0; i < 2; i++) {
1602                 if (keys[i].lk_keyid == capa->lc_keyid) {
1603                         oti->oti_capa_key = keys[i];
1604                         break;
1605                 }
1606         }
1607         spin_unlock(&capa_lock);
1608
1609         if (i == 2) {
1610                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1611                 RETURN(-ESTALE);
1612         }
1613
1614         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1615         if (rc)
1616                 RETURN(rc);
1617
1618         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1619                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1620                 RETURN(-EACCES);
1621         }
1622
1623         oc = capa_add(dev->od_capa_hash, capa);
1624         capa_put(oc);
1625
1626         RETURN(0);
1627 }
1628
1629 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1630                     struct lustre_capa *capa, __u64 opc)
1631 {
1632         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1633         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1634         struct lu_capainfo *lci;
1635         int rc;
1636
1637         if (!osd->od_fl_capa)
1638                 return 0;
1639
1640         if (capa == BYPASS_CAPA)
1641                 return 0;
1642
1643         lci = lu_capainfo_get(env);
1644         if (unlikely(lci == NULL))
1645                 return 0;
1646
1647         if (lci->lci_auth == LC_ID_NONE)
1648                 return 0;
1649
1650         if (capa == NULL) {
1651                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1652                        osd_name(osd), PFID(fid), -EACCES);
1653                 return -EACCES;
1654         }
1655
1656         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1657                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1658                            PFID(fid));
1659                 return -EACCES;
1660         }
1661
1662         if (!capa_opc_supported(capa, opc)) {
1663                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1664                 return -EACCES;
1665         }
1666
1667         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1668         if (rc != 0) {
1669                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1670                 return -EACCES;
1671         }
1672
1673         return 0;
1674 }
1675
1676 static struct timespec *osd_inode_time(const struct lu_env *env,
1677                                        struct inode *inode, __u64 seconds)
1678 {
1679         struct osd_thread_info  *oti = osd_oti_get(env);
1680         struct timespec         *t   = &oti->oti_time;
1681
1682         t->tv_sec = seconds;
1683         t->tv_nsec = 0;
1684         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1685         return t;
1686 }
1687
1688
1689 static void osd_inode_getattr(const struct lu_env *env,
1690                               struct inode *inode, struct lu_attr *attr)
1691 {
1692         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1693                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1694                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1695                                LA_TYPE;
1696
1697         attr->la_atime      = LTIME_S(inode->i_atime);
1698         attr->la_mtime      = LTIME_S(inode->i_mtime);
1699         attr->la_ctime      = LTIME_S(inode->i_ctime);
1700         attr->la_mode       = inode->i_mode;
1701         attr->la_size       = i_size_read(inode);
1702         attr->la_blocks     = inode->i_blocks;
1703         attr->la_uid        = inode->i_uid;
1704         attr->la_gid        = inode->i_gid;
1705         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1706         attr->la_nlink      = inode->i_nlink;
1707         attr->la_rdev       = inode->i_rdev;
1708         attr->la_blksize    = 1 << inode->i_blkbits;
1709         attr->la_blkbits    = inode->i_blkbits;
1710 }
1711
1712 static int osd_attr_get(const struct lu_env *env,
1713                         struct dt_object *dt,
1714                         struct lu_attr *attr,
1715                         struct lustre_capa *capa)
1716 {
1717         struct osd_object *obj = osd_dt_obj(dt);
1718
1719         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1720         LINVRNT(osd_invariant(obj));
1721
1722         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1723                 return -EACCES;
1724
1725         spin_lock(&obj->oo_guard);
1726         osd_inode_getattr(env, obj->oo_inode, attr);
1727         spin_unlock(&obj->oo_guard);
1728         return 0;
1729 }
1730
1731 static int osd_declare_attr_set(const struct lu_env *env,
1732                                 struct dt_object *dt,
1733                                 const struct lu_attr *attr,
1734                                 struct thandle *handle)
1735 {
1736         struct osd_thandle     *oh;
1737         struct osd_object      *obj;
1738         struct osd_thread_info *info = osd_oti_get(env);
1739         struct lquota_id_info  *qi = &info->oti_qi;
1740         long long               bspace;
1741         int                     rc = 0;
1742         bool                    allocated;
1743         ENTRY;
1744
1745         LASSERT(dt != NULL);
1746         LASSERT(handle != NULL);
1747
1748         obj = osd_dt_obj(dt);
1749         LASSERT(osd_invariant(obj));
1750
1751         oh = container_of0(handle, struct osd_thandle, ot_super);
1752         LASSERT(oh->ot_handle == NULL);
1753
1754         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1755                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1756
1757         if (attr == NULL || obj->oo_inode == NULL)
1758                 RETURN(rc);
1759
1760         bspace   = obj->oo_inode->i_blocks;
1761         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1762         bspace   = toqb(bspace);
1763
1764         /* Changing ownership is always preformed by super user, it should not
1765          * fail with EDQUOT.
1766          *
1767          * We still need to call the osd_declare_qid() to calculate the journal
1768          * credits for updating quota accounting files and to trigger quota
1769          * space adjustment once the operation is completed.*/
1770         if ((attr->la_valid & LA_UID) != 0 &&
1771              attr->la_uid != obj->oo_inode->i_uid) {
1772                 qi->lqi_type = USRQUOTA;
1773
1774                 /* inode accounting */
1775                 qi->lqi_is_blk = false;
1776
1777                 /* one more inode for the new owner ... */
1778                 qi->lqi_id.qid_uid = attr->la_uid;
1779                 qi->lqi_space      = 1;
1780                 allocated = (attr->la_uid == 0) ? true : false;
1781                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1782                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1783                         rc = 0;
1784                 if (rc)
1785                         RETURN(rc);
1786
1787                 /* and one less inode for the current uid */
1788                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1789                 qi->lqi_space      = -1;
1790                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1791                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1792                         rc = 0;
1793                 if (rc)
1794                         RETURN(rc);
1795
1796                 /* block accounting */
1797                 qi->lqi_is_blk = true;
1798
1799                 /* more blocks for the new owner ... */
1800                 qi->lqi_id.qid_uid = attr->la_uid;
1801                 qi->lqi_space      = bspace;
1802                 allocated = (attr->la_uid == 0) ? true : false;
1803                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1804                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1805                         rc = 0;
1806                 if (rc)
1807                         RETURN(rc);
1808
1809                 /* and finally less blocks for the current owner */
1810                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1811                 qi->lqi_space      = -bspace;
1812                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1813                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1814                         rc = 0;
1815                 if (rc)
1816                         RETURN(rc);
1817         }
1818
1819         if (attr->la_valid & LA_GID &&
1820             attr->la_gid != obj->oo_inode->i_gid) {
1821                 qi->lqi_type = GRPQUOTA;
1822
1823                 /* inode accounting */
1824                 qi->lqi_is_blk = false;
1825
1826                 /* one more inode for the new group owner ... */
1827                 qi->lqi_id.qid_gid = attr->la_gid;
1828                 qi->lqi_space      = 1;
1829                 allocated = (attr->la_gid == 0) ? true : false;
1830                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1831                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1832                         rc = 0;
1833                 if (rc)
1834                         RETURN(rc);
1835
1836                 /* and one less inode for the current gid */
1837                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1838                 qi->lqi_space      = -1;
1839                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1840                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1841                         rc = 0;
1842                 if (rc)
1843                         RETURN(rc);
1844
1845                 /* block accounting */
1846                 qi->lqi_is_blk = true;
1847
1848                 /* more blocks for the new owner ... */
1849                 qi->lqi_id.qid_gid = attr->la_gid;
1850                 qi->lqi_space      = bspace;
1851                 allocated = (attr->la_gid == 0) ? true : false;
1852                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1853                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1854                         rc = 0;
1855                 if (rc)
1856                         RETURN(rc);
1857
1858                 /* and finally less blocks for the current owner */
1859                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1860                 qi->lqi_space      = -bspace;
1861                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1862                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1863                         rc = 0;
1864                 if (rc)
1865                         RETURN(rc);
1866         }
1867
1868         RETURN(rc);
1869 }
1870
1871 static int osd_inode_setattr(const struct lu_env *env,
1872                              struct inode *inode, const struct lu_attr *attr)
1873 {
1874         __u64 bits;
1875
1876         bits = attr->la_valid;
1877
1878         if (bits & LA_ATIME)
1879                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1880         if (bits & LA_CTIME)
1881                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1882         if (bits & LA_MTIME)
1883                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1884         if (bits & LA_SIZE) {
1885                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1886                 i_size_write(inode, attr->la_size);
1887         }
1888
1889 #if 0
1890         /* OSD should not change "i_blocks" which is used by quota.
1891          * "i_blocks" should be changed by ldiskfs only. */
1892         if (bits & LA_BLOCKS)
1893                 inode->i_blocks = attr->la_blocks;
1894 #endif
1895         if (bits & LA_MODE)
1896                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1897                         (attr->la_mode & ~S_IFMT);
1898         if (bits & LA_UID)
1899                 inode->i_uid    = attr->la_uid;
1900         if (bits & LA_GID)
1901                 inode->i_gid    = attr->la_gid;
1902         if (bits & LA_NLINK)
1903                 set_nlink(inode, attr->la_nlink);
1904         if (bits & LA_RDEV)
1905                 inode->i_rdev   = attr->la_rdev;
1906
1907         if (bits & LA_FLAGS) {
1908                 /* always keep S_NOCMTIME */
1909                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1910                                  S_NOCMTIME;
1911         }
1912         return 0;
1913 }
1914
1915 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1916 {
1917         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1918             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1919                 struct iattr    iattr;
1920                 int             rc;
1921
1922                 iattr.ia_valid = 0;
1923                 if (attr->la_valid & LA_UID)
1924                         iattr.ia_valid |= ATTR_UID;
1925                 if (attr->la_valid & LA_GID)
1926                         iattr.ia_valid |= ATTR_GID;
1927                 iattr.ia_uid = attr->la_uid;
1928                 iattr.ia_gid = attr->la_gid;
1929
1930                 rc = ll_vfs_dq_transfer(inode, &iattr);
1931                 if (rc) {
1932                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1933                                "enforcement enabled on the ldiskfs filesystem?",
1934                                inode->i_sb->s_id, rc);
1935                         return rc;
1936                 }
1937         }
1938         return 0;
1939 }
1940
1941 static int osd_attr_set(const struct lu_env *env,
1942                         struct dt_object *dt,
1943                         const struct lu_attr *attr,
1944                         struct thandle *handle,
1945                         struct lustre_capa *capa)
1946 {
1947         struct osd_object *obj = osd_dt_obj(dt);
1948         struct inode      *inode;
1949         int rc;
1950
1951         LASSERT(handle != NULL);
1952         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1953         LASSERT(osd_invariant(obj));
1954
1955         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1956                 return -EACCES;
1957
1958         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1959
1960         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1961                 struct osd_thread_info  *oti  = osd_oti_get(env);
1962                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1963                 struct lu_fid           *fid1 = &oti->oti_fid;
1964                 struct osd_inode_id     *id   = &oti->oti_id;
1965                 struct iam_path_descr   *ipd;
1966                 struct iam_container    *bag;
1967                 struct osd_thandle      *oh;
1968                 int                      rc;
1969
1970                 fid_cpu_to_be(fid1, fid0);
1971                 memset(id, 1, sizeof(*id));
1972                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1973                                   fid0)->oi_dir.od_container;
1974                 ipd = osd_idx_ipd_get(env, bag);
1975                 if (unlikely(ipd == NULL))
1976                         RETURN(-ENOMEM);
1977
1978                 oh = container_of0(handle, struct osd_thandle, ot_super);
1979                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1980                                 (const struct iam_rec *)id, ipd);
1981                 osd_ipd_put(env, bag, ipd);
1982                 return(rc > 0 ? 0 : rc);
1983         }
1984
1985         inode = obj->oo_inode;
1986         ll_vfs_dq_init(inode);
1987
1988         rc = osd_quota_transfer(inode, attr);
1989         if (rc)
1990                 return rc;
1991
1992         spin_lock(&obj->oo_guard);
1993         rc = osd_inode_setattr(env, inode, attr);
1994         spin_unlock(&obj->oo_guard);
1995
1996         if (!rc)
1997                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1998         return rc;
1999 }
2000
2001 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2002                                     struct osd_object *obj,
2003                                     const char *name, const int namelen)
2004 {
2005         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2006 }
2007
2008 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2009                       umode_t mode, struct dt_allocation_hint *hint,
2010                       struct thandle *th)
2011 {
2012         int result;
2013         struct osd_device  *osd = osd_obj2dev(obj);
2014         struct osd_thandle *oth;
2015         struct dt_object   *parent = NULL;
2016         struct inode       *inode;
2017
2018         LINVRNT(osd_invariant(obj));
2019         LASSERT(obj->oo_inode == NULL);
2020         LASSERT(obj->oo_hl_head == NULL);
2021
2022         if (S_ISDIR(mode) && ldiskfs_pdo) {
2023                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2024                 if (obj->oo_hl_head == NULL)
2025                         return -ENOMEM;
2026         }
2027
2028         oth = container_of(th, struct osd_thandle, ot_super);
2029         LASSERT(oth->ot_handle->h_transaction != NULL);
2030
2031         if (hint && hint->dah_parent)
2032                 parent = hint->dah_parent;
2033
2034         inode = ldiskfs_create_inode(oth->ot_handle,
2035                                      parent ? osd_dt_obj(parent)->oo_inode :
2036                                               osd_sb(osd)->s_root->d_inode,
2037                                      mode);
2038         if (!IS_ERR(inode)) {
2039                 /* Do not update file c/mtime in ldiskfs.
2040                  * NB: don't need any lock because no contention at this
2041                  * early stage */
2042                 inode->i_flags |= S_NOCMTIME;
2043
2044                 /* For new created object, it must be consistent,
2045                  * and it is unnecessary to scrub against it. */
2046                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2047                 obj->oo_inode = inode;
2048                 result = 0;
2049         } else {
2050                 if (obj->oo_hl_head != NULL) {
2051                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2052                         obj->oo_hl_head = NULL;
2053                 }
2054                 result = PTR_ERR(inode);
2055         }
2056         LINVRNT(osd_invariant(obj));
2057         return result;
2058 }
2059
2060 enum {
2061         OSD_NAME_LEN = 255
2062 };
2063
2064 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2065                      struct lu_attr *attr,
2066                      struct dt_allocation_hint *hint,
2067                      struct dt_object_format *dof,
2068                      struct thandle *th)
2069 {
2070         int result;
2071         struct osd_thandle *oth;
2072         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2073
2074         LASSERT(S_ISDIR(attr->la_mode));
2075
2076         oth = container_of(th, struct osd_thandle, ot_super);
2077         LASSERT(oth->ot_handle->h_transaction != NULL);
2078         result = osd_mkfile(info, obj, mode, hint, th);
2079
2080         return result;
2081 }
2082
2083 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2084                         struct lu_attr *attr,
2085                         struct dt_allocation_hint *hint,
2086                         struct dt_object_format *dof,
2087                         struct thandle *th)
2088 {
2089         int result;
2090         struct osd_thandle *oth;
2091         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2092
2093         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2094
2095         LASSERT(S_ISREG(attr->la_mode));
2096
2097         oth = container_of(th, struct osd_thandle, ot_super);
2098         LASSERT(oth->ot_handle->h_transaction != NULL);
2099
2100         result = osd_mkfile(info, obj, mode, hint, th);
2101         if (result == 0) {
2102                 LASSERT(obj->oo_inode != NULL);
2103                 if (feat->dif_flags & DT_IND_VARKEY)
2104                         result = iam_lvar_create(obj->oo_inode,
2105                                                  feat->dif_keysize_max,
2106                                                  feat->dif_ptrsize,
2107                                                  feat->dif_recsize_max,
2108                                                  oth->ot_handle);
2109                 else
2110                         result = iam_lfix_create(obj->oo_inode,
2111                                                  feat->dif_keysize_max,
2112                                                  feat->dif_ptrsize,
2113                                                  feat->dif_recsize_max,
2114                                                  oth->ot_handle);
2115
2116         }
2117         return result;
2118 }
2119
2120 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2121                      struct lu_attr *attr,
2122                      struct dt_allocation_hint *hint,
2123                      struct dt_object_format *dof,
2124                      struct thandle *th)
2125 {
2126         LASSERT(S_ISREG(attr->la_mode));
2127         return osd_mkfile(info, obj, (attr->la_mode &
2128                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2129 }
2130
2131 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2132                      struct lu_attr *attr,
2133                      struct dt_allocation_hint *hint,
2134                      struct dt_object_format *dof,
2135                      struct thandle *th)
2136 {
2137         LASSERT(S_ISLNK(attr->la_mode));
2138         return osd_mkfile(info, obj, (attr->la_mode &
2139                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2140 }
2141
2142 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2143                      struct lu_attr *attr,
2144                      struct dt_allocation_hint *hint,
2145                      struct dt_object_format *dof,
2146                      struct thandle *th)
2147 {
2148         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2149         int result;
2150
2151         LINVRNT(osd_invariant(obj));
2152         LASSERT(obj->oo_inode == NULL);
2153         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2154                 S_ISFIFO(mode) || S_ISSOCK(mode));
2155
2156         result = osd_mkfile(info, obj, mode, hint, th);
2157         if (result == 0) {
2158                 LASSERT(obj->oo_inode != NULL);
2159                 /*
2160                  * This inode should be marked dirty for i_rdev.  Currently
2161                  * that is done in the osd_attr_init().
2162                  */
2163                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2164                                    attr->la_rdev);
2165         }
2166         LINVRNT(osd_invariant(obj));
2167         return result;
2168 }
2169
2170 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2171                               struct lu_attr *,
2172                               struct dt_allocation_hint *hint,
2173                               struct dt_object_format *dof,
2174                               struct thandle *);
2175
2176 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2177 {
2178         osd_obj_type_f result;
2179
2180         switch (type) {
2181         case DFT_DIR:
2182                 result = osd_mkdir;
2183                 break;
2184         case DFT_REGULAR:
2185                 result = osd_mkreg;
2186                 break;
2187         case DFT_SYM:
2188                 result = osd_mksym;
2189                 break;
2190         case DFT_NODE:
2191                 result = osd_mknod;
2192                 break;
2193         case DFT_INDEX:
2194                 result = osd_mk_index;
2195                 break;
2196
2197         default:
2198                 LBUG();
2199                 break;
2200         }
2201         return result;
2202 }
2203
2204
2205 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2206                         struct dt_object *parent, struct dt_object *child,
2207                         umode_t child_mode)
2208 {
2209         LASSERT(ah);
2210
2211         ah->dah_parent = parent;
2212         ah->dah_mode = child_mode;
2213 }
2214
2215 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2216                           struct lu_attr *attr, struct dt_object_format *dof)
2217 {
2218         struct inode   *inode = obj->oo_inode;
2219         __u64           valid = attr->la_valid;
2220         int             result;
2221
2222         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2223
2224         if (dof->dof_type != DFT_NODE)
2225                 attr->la_valid &= ~LA_RDEV;
2226         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2227                 attr->la_valid &= ~LA_ATIME;
2228         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2229                 attr->la_valid &= ~LA_CTIME;
2230         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2231                 attr->la_valid &= ~LA_MTIME;
2232
2233         result = osd_quota_transfer(inode, attr);
2234         if (result)
2235                 return;
2236
2237         if (attr->la_valid != 0) {
2238                 result = osd_inode_setattr(info->oti_env, inode, attr);
2239                 /*
2240                  * The osd_inode_setattr() should always succeed here.  The
2241                  * only error that could be returned is EDQUOT when we are
2242                  * trying to change the UID or GID of the inode. However, this
2243                  * should not happen since quota enforcement is no longer
2244                  * enabled on ldiskfs (lquota takes care of it).
2245                  */
2246                 LASSERTF(result == 0, "%d", result);
2247                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2248         }
2249
2250         attr->la_valid = valid;
2251 }
2252
2253 /**
2254  * Helper function for osd_object_create()
2255  *
2256  * \retval 0, on success
2257  */
2258 static int __osd_object_create(struct osd_thread_info *info,
2259                                struct osd_object *obj, struct lu_attr *attr,
2260                                struct dt_allocation_hint *hint,
2261                                struct dt_object_format *dof,
2262                                struct thandle *th)
2263 {
2264         int     result;
2265         __u32   umask;
2266
2267         /* we drop umask so that permissions we pass are not affected */
2268         umask = current->fs->umask;
2269         current->fs->umask = 0;
2270
2271         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2272                                                   th);
2273         if (result == 0) {
2274                 osd_attr_init(info, obj, attr, dof);
2275                 osd_object_init0(obj);
2276                 /* bz 24037 */
2277                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2278                         unlock_new_inode(obj->oo_inode);
2279         }
2280
2281         /* restore previous umask value */
2282         current->fs->umask = umask;
2283
2284         return result;
2285 }
2286
2287 /**
2288  * Helper function for osd_object_create()
2289  *
2290  * \retval 0, on success
2291  */
2292 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2293                            const struct lu_fid *fid, struct thandle *th)
2294 {
2295         struct osd_thread_info *info = osd_oti_get(env);
2296         struct osd_inode_id    *id   = &info->oti_id;
2297         struct osd_device      *osd  = osd_obj2dev(obj);
2298         struct osd_thandle     *oh;
2299
2300         LASSERT(obj->oo_inode != NULL);
2301
2302         oh = container_of0(th, struct osd_thandle, ot_super);
2303         LASSERT(oh->ot_handle);
2304
2305         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2306         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2307 }
2308
2309 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2310                    obd_seq seq, struct lu_seq_range *range)
2311 {
2312         struct seq_server_site  *ss = osd_seq_site(osd);
2313
2314         if (fid_seq_is_idif(seq)) {
2315                 fld_range_set_ost(range);
2316                 range->lsr_index = idif_ost_idx(seq);
2317                 return 0;
2318         }
2319
2320         if (!fid_seq_in_fldb(seq)) {
2321                 fld_range_set_mdt(range);
2322                 if (ss != NULL)
2323                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2324                          * at all */
2325                         range->lsr_index = ss->ss_node_id;
2326                 return 0;
2327         }
2328
2329         LASSERT(ss != NULL);
2330         fld_range_set_any(range);
2331         /* OSD will only do local fld lookup */
2332         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2333 }
2334
2335 /*
2336  * Concurrency: no external locking is necessary.
2337  */
2338 static int osd_declare_object_create(const struct lu_env *env,
2339                                      struct dt_object *dt,
2340                                      struct lu_attr *attr,
2341                                      struct dt_allocation_hint *hint,
2342                                      struct dt_object_format *dof,
2343                                      struct thandle *handle)
2344 {
2345         struct osd_thandle      *oh;
2346         int                      rc;
2347         ENTRY;
2348
2349         LASSERT(handle != NULL);
2350
2351         oh = container_of0(handle, struct osd_thandle, ot_super);
2352         LASSERT(oh->ot_handle == NULL);
2353
2354         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2355                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2356         /* Reuse idle OI block may cause additional one OI block
2357          * to be changed. */
2358         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2359                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2360
2361         /* If this is directory, then we expect . and .. to be inserted as
2362          * well. The one directory block always needs to be created for the
2363          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2364          * block), there is no danger of needing a tree for the first block.
2365          */
2366         if (attr && S_ISDIR(attr->la_mode)) {
2367                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2368                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2369                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2370         }
2371
2372         if (!attr)
2373                 RETURN(0);
2374
2375         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2376                                    false, false, NULL, false);
2377         if (rc != 0)
2378                 RETURN(rc);
2379
2380         RETURN(rc);
2381 }
2382
2383 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2384                              struct lu_attr *attr,
2385                              struct dt_allocation_hint *hint,
2386                              struct dt_object_format *dof,
2387                              struct thandle *th)
2388 {
2389         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2390         struct osd_object      *obj    = osd_dt_obj(dt);
2391         struct osd_thread_info *info   = osd_oti_get(env);
2392         int result;
2393
2394         ENTRY;
2395
2396         LINVRNT(osd_invariant(obj));
2397         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2398         LASSERT(osd_write_locked(env, obj));
2399         LASSERT(th != NULL);
2400
2401         if (unlikely(fid_is_acct(fid)))
2402                 /* Quota files can't be created from the kernel any more,
2403                  * 'tune2fs -O quota' will take care of creating them */
2404                 RETURN(-EPERM);
2405
2406         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2407         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2408
2409         result = __osd_object_create(info, obj, attr, hint, dof, th);
2410         if (result == 0)
2411                 result = __osd_oi_insert(env, obj, fid, th);
2412
2413         LASSERT(ergo(result == 0,
2414                      dt_object_exists(dt) && !dt_object_remote(dt)));
2415
2416         LASSERT(osd_invariant(obj));
2417         RETURN(result);
2418 }
2419
2420 /**
2421  * Called to destroy on-disk representation of the object
2422  *
2423  * Concurrency: must be locked
2424  */
2425 static int osd_declare_object_destroy(const struct lu_env *env,
2426                                       struct dt_object *dt,
2427                                       struct thandle *th)
2428 {
2429         struct osd_object  *obj = osd_dt_obj(dt);
2430         struct inode       *inode = obj->oo_inode;
2431         struct osd_thandle *oh;
2432         int                 rc;
2433         ENTRY;
2434
2435         oh = container_of0(th, struct osd_thandle, ot_super);
2436         LASSERT(oh->ot_handle == NULL);
2437         LASSERT(inode);
2438
2439         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2440                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2441         /* Recycle idle OI leaf may cause additional three OI blocks
2442          * to be changed. */
2443         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2444                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2445         /* one less inode */
2446         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2447                                    false, true, NULL, false);
2448         if (rc)
2449                 RETURN(rc);
2450         /* data to be truncated */
2451         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2452                                    true, true, NULL, false);
2453         RETURN(rc);
2454 }
2455
2456 static int osd_object_destroy(const struct lu_env *env,
2457                               struct dt_object *dt,
2458                               struct thandle *th)
2459 {
2460         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2461         struct osd_object      *obj = osd_dt_obj(dt);
2462         struct inode           *inode = obj->oo_inode;
2463         struct osd_device      *osd = osd_obj2dev(obj);
2464         struct osd_thandle     *oh;
2465         int                     result;
2466         ENTRY;
2467
2468         oh = container_of0(th, struct osd_thandle, ot_super);
2469         LASSERT(oh->ot_handle);
2470         LASSERT(inode);
2471         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2472
2473         if (unlikely(fid_is_acct(fid)))
2474                 RETURN(-EPERM);
2475
2476         if (S_ISDIR(inode->i_mode)) {
2477                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2478                         inode->i_nlink == 2);
2479                 /* it will check/delete the inode from remote parent,
2480                  * how to optimize it? unlink performance impaction XXX */
2481                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2482                 if (result != 0 && result != -ENOENT) {
2483                         CERROR("%s: delete inode "DFID": rc = %d\n",
2484                                osd_name(osd), PFID(fid), result);
2485                 }
2486                 spin_lock(&obj->oo_guard);
2487                 clear_nlink(inode);
2488                 spin_unlock(&obj->oo_guard);
2489                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2490         }
2491
2492         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2493
2494         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2495                                OI_CHECK_FLD);
2496
2497         /* XXX: add to ext3 orphan list */
2498         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2499
2500         /* not needed in the cache anymore */
2501         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2502
2503         RETURN(0);
2504 }
2505
2506 /**
2507  * Put the fid into lustre_mdt_attrs, and then place the structure
2508  * inode's ea. This fid should not be altered during the life time
2509  * of the inode.
2510  *
2511  * \retval +ve, on success
2512  * \retval -ve, on error
2513  *
2514  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2515  */
2516 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2517                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2518 {
2519         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2520         int                      rc;
2521         ENTRY;
2522
2523         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2524                 RETURN(0);
2525
2526         lustre_lma_init(lma, fid, compat, incompat);
2527         lustre_lma_swab(lma);
2528
2529         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2530                              XATTR_CREATE);
2531         /* LMA may already exist, but we need to check that all the
2532          * desired compat/incompat flags have been added. */
2533         if (unlikely(rc == -EEXIST)) {
2534                 if (compat == 0 && incompat == 0)
2535                         RETURN(0);
2536
2537                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2538                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2539                                      LMA_OLD_SIZE);
2540                 if (rc <= 0)
2541                         RETURN(-EINVAL);
2542
2543                 lustre_lma_swab(lma);
2544                 if (!(~lma->lma_compat & compat) &&
2545                     !(~lma->lma_incompat & incompat))
2546                         RETURN(0);
2547
2548                 lma->lma_compat |= compat;
2549                 lma->lma_incompat |= incompat;
2550                 lustre_lma_swab(lma);
2551                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2552                                      sizeof(*lma), XATTR_REPLACE);
2553         }
2554
2555         RETURN(rc);
2556 }
2557
2558 /**
2559  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2560  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2561  * To have compatilibility with 1.8 ldiskfs driver we need to have
2562  * magic number at start of fid data.
2563  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2564  * its inmemory API.
2565  */
2566 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2567                                   const struct dt_rec *fid)
2568 {
2569         if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2570             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2571                 param->edp_magic = 0;
2572                 return;
2573         }
2574
2575         param->edp_magic = LDISKFS_LUFID_MAGIC;
2576         param->edp_len =  sizeof(struct lu_fid) + 1;
2577         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2578 }
2579
2580 /**
2581  * Try to read the fid from inode ea into dt_rec.
2582  *
2583  * \param fid object fid.
2584  *
2585  * \retval 0 on success
2586  */
2587 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2588                           __u32 ino, struct lu_fid *fid,
2589                           struct osd_inode_id *id)
2590 {
2591         struct osd_thread_info *info  = osd_oti_get(env);
2592         struct inode           *inode;
2593         ENTRY;
2594
2595         osd_id_gen(id, ino, OSD_OII_NOGEN);
2596         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2597         if (IS_ERR(inode))
2598                 RETURN(PTR_ERR(inode));
2599
2600         iput(inode);
2601         RETURN(0);
2602 }
2603
2604 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2605                                         struct inode *dir,
2606                                         struct inode  *parent_dir,
2607                                         const struct dt_rec *dot_fid,
2608                                         const struct dt_rec *dot_dot_fid,
2609                                         struct osd_thandle *oth)
2610 {
2611         struct ldiskfs_dentry_param *dot_ldp;
2612         struct ldiskfs_dentry_param *dot_dot_ldp;
2613
2614         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2615         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2616
2617         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2618         dot_ldp->edp_magic = 0;
2619         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2620                                         dir, dot_ldp, dot_dot_ldp);
2621 }
2622
2623 /**
2624  * Create an local agent inode for remote entry
2625  */
2626 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2627                                                   struct osd_device *osd,
2628                                                   struct osd_object *pobj,
2629                                                   const struct lu_fid *fid,
2630                                                   struct thandle *th)
2631 {
2632         struct osd_thread_info  *info = osd_oti_get(env);
2633         struct inode            *local;
2634         struct osd_thandle      *oh;
2635         int                     rc;
2636         ENTRY;
2637
2638         LASSERT(th);
2639         oh = container_of(th, struct osd_thandle, ot_super);
2640         LASSERT(oh->ot_handle->h_transaction != NULL);
2641
2642         /* FIXME: Insert index api needs to know the mode of
2643          * the remote object. Just use S_IFDIR for now */
2644         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2645         if (IS_ERR(local)) {
2646                 CERROR("%s: create local error %d\n", osd_name(osd),
2647                        (int)PTR_ERR(local));
2648                 RETURN(local);
2649         }
2650
2651         /* Set special LMA flag for local agent inode */
2652         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2653         if (rc != 0) {
2654                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2655                        osd_name(osd), PFID(fid), rc);
2656                 RETURN(ERR_PTR(rc));
2657         }
2658
2659         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2660                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2661                 (const struct dt_rec *)fid, oh);
2662         if (rc != 0) {
2663                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2664                         osd_name(osd), PFID(fid), rc);
2665                 RETURN(ERR_PTR(rc));
2666         }
2667
2668         RETURN(local);
2669 }
2670
2671 /**
2672  * Delete local agent inode for remote entry
2673  */
2674 static int osd_delete_local_agent_inode(const struct lu_env *env,
2675                                         struct osd_device *osd,
2676                                         const struct lu_fid *fid,
2677                                         __u32 ino, struct osd_thandle *oh)
2678 {
2679         struct osd_thread_info  *oti = osd_oti_get(env);
2680         struct osd_inode_id     *id = &oti->oti_id;
2681         struct inode            *inode;
2682         ENTRY;
2683
2684         id->oii_ino = le32_to_cpu(ino);
2685         id->oii_gen = OSD_OII_NOGEN;
2686         inode = osd_iget(oti, osd, id);
2687         if (IS_ERR(inode)) {
2688                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2689                        PFID(fid), id->oii_ino, id->oii_gen);
2690                 RETURN(PTR_ERR(inode));
2691         }
2692
2693         clear_nlink(inode);
2694         mark_inode_dirty(inode);
2695         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2696                 osd_name(osd), PFID(fid), inode->i_ino);
2697         iput(inode);
2698         RETURN(0);
2699 }
2700
2701 /**
2702  * OSD layer object create function for interoperability mode (b11826).
2703  * This is mostly similar to osd_object_create(). Only difference being, fid is
2704  * inserted into inode ea here.
2705  *
2706  * \retval   0, on success
2707  * \retval -ve, on error
2708  */
2709 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2710                                 struct lu_attr *attr,
2711                                 struct dt_allocation_hint *hint,
2712                                 struct dt_object_format *dof,
2713                                 struct thandle *th)
2714 {
2715         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2716         struct osd_object      *obj    = osd_dt_obj(dt);
2717         struct osd_thread_info *info   = osd_oti_get(env);
2718         int                     result;
2719
2720         ENTRY;
2721
2722         LASSERT(osd_invariant(obj));
2723         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2724         LASSERT(osd_write_locked(env, obj));
2725         LASSERT(th != NULL);
2726
2727         if (unlikely(fid_is_acct(fid)))
2728                 /* Quota files can't be created from the kernel any more,
2729                  * 'tune2fs -O quota' will take care of creating them */
2730                 RETURN(-EPERM);
2731
2732         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2733         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2734
2735         result = __osd_object_create(info, obj, attr, hint, dof, th);
2736         if (result == 0)
2737                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2738                                 fid_is_on_ost(info, osd_obj2dev(obj),
2739                                               fid, OI_CHECK_FLD) ?
2740                                 LMAC_FID_ON_OST : 0, 0);
2741
2742         if (result == 0)
2743                 result = __osd_oi_insert(env, obj, fid, th);
2744
2745         LASSERT(ergo(result == 0,
2746                      dt_object_exists(dt) && !dt_object_remote(dt)));
2747         LINVRNT(osd_invariant(obj));
2748         RETURN(result);
2749 }
2750
2751 static int osd_declare_object_ref_add(const struct lu_env *env,
2752                                       struct dt_object *dt,
2753                                       struct thandle *handle)
2754 {
2755         struct osd_thandle       *oh;
2756
2757         /* it's possible that object doesn't exist yet */
2758         LASSERT(handle != NULL);
2759
2760         oh = container_of0(handle, struct osd_thandle, ot_super);
2761         LASSERT(oh->ot_handle == NULL);
2762
2763         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2764                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2765
2766         return 0;
2767 }
2768
2769 /*
2770  * Concurrency: @dt is write locked.
2771  */
2772 static int osd_object_ref_add(const struct lu_env *env,
2773                               struct dt_object *dt, struct thandle *th)
2774 {
2775         struct osd_object  *obj = osd_dt_obj(dt);
2776         struct inode       *inode = obj->oo_inode;
2777         struct osd_thandle *oh;
2778         int                 rc = 0;
2779
2780         LINVRNT(osd_invariant(obj));
2781         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2782         LASSERT(osd_write_locked(env, obj));
2783         LASSERT(th != NULL);
2784
2785         oh = container_of0(th, struct osd_thandle, ot_super);
2786         LASSERT(oh->ot_handle != NULL);
2787
2788         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2789
2790         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2791                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2792         /*
2793          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2794          * (65000) subdirectories by storing "1" in i_nlink if the link count
2795          * would otherwise overflow. Directory tranversal tools understand
2796          * that (st_nlink == 1) indicates that the filesystem dose not track
2797          * hard links count on the directory, and will not abort subdirectory
2798          * scanning early once (st_nlink - 2) subdirs have been found.
2799          *
2800          * This also has to properly handle the case of inodes with nlink == 0
2801          * in case they are being linked into the PENDING directory
2802          */
2803         spin_lock(&obj->oo_guard);
2804         ldiskfs_inc_count(oh->ot_handle, inode);
2805         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2806         spin_unlock(&obj->oo_guard);
2807
2808         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2809         LINVRNT(osd_invariant(obj));
2810
2811         return rc;
2812 }
2813
2814 static int osd_declare_object_ref_del(const struct lu_env *env,
2815                                       struct dt_object *dt,
2816                                       struct thandle *handle)
2817 {
2818         struct osd_thandle *oh;
2819
2820         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2821         LASSERT(handle != NULL);
2822
2823         oh = container_of0(handle, struct osd_thandle, ot_super);
2824         LASSERT(oh->ot_handle == NULL);
2825
2826         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2827                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2828
2829         return 0;
2830 }
2831
2832 /*
2833  * Concurrency: @dt is write locked.
2834  */
2835 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2836                               struct thandle *th)
2837 {
2838         struct osd_object       *obj = osd_dt_obj(dt);
2839         struct inode            *inode = obj->oo_inode;
2840         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2841         struct osd_thandle      *oh;
2842
2843         LINVRNT(osd_invariant(obj));
2844         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2845         LASSERT(osd_write_locked(env, obj));
2846         LASSERT(th != NULL);
2847
2848         oh = container_of0(th, struct osd_thandle, ot_super);
2849         LASSERT(oh->ot_handle != NULL);
2850
2851         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2852
2853         spin_lock(&obj->oo_guard);
2854         /* That can be result of upgrade from old Lustre version and
2855          * applied only to local files.  Just skip this ref_del call.
2856          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2857         if (inode->i_nlink == 0) {
2858                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2859                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2860                              ", maybe an upgraded file? (LU-3915)\n",
2861                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2862                 spin_unlock(&obj->oo_guard);
2863                 return 0;
2864         }
2865
2866         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2867                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2868
2869         ldiskfs_dec_count(oh->ot_handle, inode);
2870         spin_unlock(&obj->oo_guard);
2871
2872         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2873         LINVRNT(osd_invariant(obj));
2874
2875         return 0;
2876 }
2877
2878 /*
2879  * Get the 64-bit version for an inode.
2880  */
2881 static int osd_object_version_get(const struct lu_env *env,
2882                                   struct dt_object *dt, dt_obj_version_t *ver)
2883 {
2884         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2885
2886         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2887                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2888         *ver = LDISKFS_I(inode)->i_fs_version;
2889         return 0;
2890 }
2891
2892 /*
2893  * Concurrency: @dt is read locked.
2894  */
2895 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2896                          struct lu_buf *buf, const char *name,
2897                          struct lustre_capa *capa)
2898 {
2899         struct osd_object      *obj    = osd_dt_obj(dt);
2900         struct inode           *inode  = obj->oo_inode;
2901         struct osd_thread_info *info   = osd_oti_get(env);
2902         struct dentry          *dentry = &info->oti_obj_dentry;
2903
2904         /* version get is not real XATTR but uses xattr API */
2905         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2906                 /* for version we are just using xattr API but change inode
2907                  * field instead */
2908                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2909                 osd_object_version_get(env, dt, buf->lb_buf);
2910                 return sizeof(dt_obj_version_t);
2911         }
2912
2913         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2914         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2915
2916         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2917                 return -EACCES;
2918
2919         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2920 }
2921
2922
2923 static int osd_declare_xattr_set(const struct lu_env *env,
2924                                  struct dt_object *dt,
2925                                  const struct lu_buf *buf, const char *name,
2926                                  int fl, struct thandle *handle)
2927 {
2928         struct osd_thandle *oh;
2929         int credits;
2930
2931         LASSERT(handle != NULL);
2932
2933         oh = container_of0(handle, struct osd_thandle, ot_super);
2934         LASSERT(oh->ot_handle == NULL);
2935
2936         /* optimistic optimization: LMA is set first and usually fit inode */
2937         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2938                 if (dt_object_exists(dt))
2939                         credits = 0;
2940                 else
2941                         credits = 1;
2942         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2943                 credits = 1;
2944         } else {
2945                 struct osd_device  *osd = osd_dev(dt->do_lu.lo_dev);
2946                 struct super_block *sb = osd_sb(osd);
2947                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2948                 if (buf && buf->lb_len > sb->s_blocksize) {
2949                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2950                                         sb->s_blocksize_bits;
2951                 }
2952         }
2953
2954         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2955
2956         return 0;
2957 }
2958
2959 /*
2960  * Set the 64-bit version for object
2961  */
2962 static void osd_object_version_set(const struct lu_env *env,
2963                                    struct dt_object *dt,
2964                                    dt_obj_version_t *new_version)
2965 {
2966         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2967
2968         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2969                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2970
2971         LDISKFS_I(inode)->i_fs_version = *new_version;
2972         /** Version is set after all inode operations are finished,
2973          *  so we should mark it dirty here */
2974         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2975 }
2976
2977 /*
2978  * Concurrency: @dt is write locked.
2979  */
2980 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2981                          const struct lu_buf *buf, const char *name, int fl,
2982                          struct thandle *handle, struct lustre_capa *capa)
2983 {
2984         struct osd_object      *obj      = osd_dt_obj(dt);
2985         struct inode           *inode    = obj->oo_inode;
2986         struct osd_thread_info *info     = osd_oti_get(env);
2987         int                     fs_flags = 0;
2988         ENTRY;
2989
2990         LASSERT(handle != NULL);
2991
2992         /* version set is not real XATTR */
2993         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2994                 /* for version we are just using xattr API but change inode
2995                  * field instead */
2996                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2997                 osd_object_version_set(env, dt, buf->lb_buf);
2998                 return sizeof(dt_obj_version_t);
2999         }
3000
3001         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3002                 return -EACCES;
3003
3004         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3005                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3006
3007         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3008         if (fl & LU_XATTR_REPLACE)
3009                 fs_flags |= XATTR_REPLACE;
3010
3011         if (fl & LU_XATTR_CREATE)
3012                 fs_flags |= XATTR_CREATE;
3013
3014         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3015                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3016                 int                      rc;
3017
3018                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3019                 if (rc != 0)
3020                         RETURN(rc);
3021
3022                 lma->lma_incompat |= LMAI_STRIPED;
3023                 lustre_lma_swab(lma);
3024                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3025                                      sizeof(*lma), XATTR_REPLACE);
3026                 if (rc != 0)
3027                         RETURN(rc);
3028         }
3029
3030         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3031                                fs_flags);
3032 }
3033
3034 /*
3035  * Concurrency: @dt is read locked.
3036  */
3037 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3038                           struct lu_buf *buf, struct lustre_capa *capa)
3039 {
3040         struct osd_object      *obj    = osd_dt_obj(dt);
3041         struct inode           *inode  = obj->oo_inode;
3042         struct osd_thread_info *info   = osd_oti_get(env);
3043         struct dentry          *dentry = &info->oti_obj_dentry;
3044
3045         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3046         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3047
3048         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3049                 return -EACCES;
3050
3051         dentry->d_inode = inode;
3052         dentry->d_sb = inode->i_sb;
3053         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3054 }
3055
3056 static int osd_declare_xattr_del(const struct lu_env *env,
3057                                  struct dt_object *dt, const char *name,
3058                                  struct thandle *handle)
3059 {
3060         struct osd_thandle *oh;
3061
3062         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3063         LASSERT(handle != NULL);
3064
3065         oh = container_of0(handle, struct osd_thandle, ot_super);
3066         LASSERT(oh->ot_handle == NULL);
3067
3068         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3069                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3070
3071         return 0;
3072 }
3073
3074 /*
3075  * Concurrency: @dt is write locked.
3076  */
3077 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3078                          const char *name, struct thandle *handle,
3079                          struct lustre_capa *capa)
3080 {
3081         struct osd_object      *obj    = osd_dt_obj(dt);
3082         struct inode           *inode  = obj->oo_inode;
3083         struct osd_thread_info *info   = osd_oti_get(env);
3084         struct dentry          *dentry = &info->oti_obj_dentry;
3085         int                     rc;
3086
3087         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3088         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3089         LASSERT(handle != NULL);
3090
3091         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3092                 return -EACCES;
3093
3094         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3095
3096         ll_vfs_dq_init(inode);
3097         dentry->d_inode = inode;
3098         dentry->d_sb = inode->i_sb;
3099         rc = inode->i_op->removexattr(dentry, name);
3100         return rc;
3101 }
3102
3103 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3104                                      struct dt_object *dt,
3105                                      struct lustre_capa *old, __u64 opc)
3106 {
3107         struct osd_thread_info *info = osd_oti_get(env);
3108         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3109         struct osd_object *obj = osd_dt_obj(dt);
3110         struct osd_device *osd = osd_obj2dev(obj);
3111         struct lustre_capa_key *key = &info->oti_capa_key;
3112         struct lustre_capa *capa = &info->oti_capa;
3113         struct obd_capa *oc;
3114         struct lu_capainfo *lci;
3115         int rc;
3116         ENTRY;
3117
3118         if (!osd->od_fl_capa)
3119                 RETURN(ERR_PTR(-ENOENT));
3120
3121         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3122         LINVRNT(osd_invariant(obj));
3123
3124         /* renewal sanity check */
3125         if (old && osd_object_auth(env, dt, old, opc))
3126                 RETURN(ERR_PTR(-EACCES));
3127
3128         lci = lu_capainfo_get(env);
3129         if (unlikely(lci == NULL))
3130                 RETURN(ERR_PTR(-ENOENT));
3131
3132         switch (lci->lci_auth) {
3133         case LC_ID_NONE:
3134                 RETURN(NULL);
3135         case LC_ID_PLAIN:
3136                 capa->lc_uid = obj->oo_inode->i_uid;
3137                 capa->lc_gid = obj->oo_inode->i_gid;
3138                 capa->lc_flags = LC_ID_PLAIN;
3139                 break;
3140         case LC_ID_CONVERT: {
3141                 __u32 d[4], s[4];
3142
3143                 s[0] = obj->oo_inode->i_uid;
3144                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3145                 s[2] = obj->oo_inode->i_gid;
3146                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3147                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3148                 if (unlikely(rc))
3149                         RETURN(ERR_PTR(rc));
3150
3151                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3152                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3153                 capa->lc_flags = LC_ID_CONVERT;
3154                 break;
3155         }
3156         default:
3157                 RETURN(ERR_PTR(-EINVAL));
3158         }
3159
3160         capa->lc_fid = *fid;
3161         capa->lc_opc = opc;
3162         capa->lc_flags |= osd->od_capa_alg << 24;
3163         capa->lc_timeout = osd->od_capa_timeout;
3164         capa->lc_expiry = 0;
3165
3166         oc = capa_lookup(osd->od_capa_hash, capa, 1);
3167         if (oc) {
3168                 LASSERT(!capa_is_expired(oc));
3169                 RETURN(oc);
3170         }
3171
3172         spin_lock(&capa_lock);
3173         *key = osd->od_capa_keys[1];
3174         spin_unlock(&capa_lock);
3175
3176         capa->lc_keyid = key->lk_keyid;
3177         capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3178
3179         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3180         if (rc) {
3181                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3182                 RETURN(ERR_PTR(rc));
3183         }
3184
3185         oc = capa_add(osd->od_capa_hash, capa);
3186         RETURN(oc);
3187 }
3188
3189 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3190                            __u64 start, __u64 end)
3191 {
3192         struct osd_object       *obj    = osd_dt_obj(dt);
3193         struct inode            *inode  = obj->oo_inode;
3194         struct osd_thread_info  *info   = osd_oti_get(env);
3195         struct dentry           *dentry = &info->oti_obj_dentry;
3196         struct file             *file   = &info->oti_file;
3197         int                     rc;
3198
3199         ENTRY;
3200
3201         dentry->d_inode = inode;
3202         dentry->d_sb = inode->i_sb;
3203         file->f_dentry = dentry;
3204         file->f_mapping = inode->i_mapping;
3205         file->f_op = inode->i_fop;
3206         set_file_inode(file, inode);
3207
3208 #ifdef HAVE_FILE_FSYNC_4ARGS
3209         rc = file->f_op->fsync(file, start, end, 0);
3210 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3211         mutex_lock(&inode->i_mutex);
3212         rc = file->f_op->fsync(file, 0);
3213         mutex_unlock(&inode->i_mutex);
3214 #else
3215         mutex_lock(&inode->i_mutex);
3216         rc = file->f_op->fsync(file, dentry, 0);
3217         mutex_unlock(&inode->i_mutex);
3218 #endif
3219
3220         RETURN(rc);
3221 }
3222
3223 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3224                         void **data)
3225 {
3226         struct osd_object *obj = osd_dt_obj(dt);
3227         ENTRY;
3228
3229         *data = (void *)obj->oo_inode;
3230         RETURN(0);
3231 }
3232
3233 /*
3234  * Index operations.
3235  */
3236
3237 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3238                            const struct dt_index_features *feat)
3239 {
3240         struct iam_descr *descr;
3241
3242         if (osd_object_is_root(o))
3243                 return feat == &dt_directory_features;
3244
3245         LASSERT(o->oo_dir != NULL);
3246
3247         descr = o->oo_dir->od_container.ic_descr;
3248         if (feat == &dt_directory_features) {
3249                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3250                         return 1;
3251                 else
3252                         return 0;
3253         } else {
3254                 return
3255                         feat->dif_keysize_min <= descr->id_key_size &&
3256                         descr->id_key_size <= feat->dif_keysize_max &&
3257                         feat->dif_recsize_min <= descr->id_rec_size &&
3258                         descr->id_rec_size <= feat->dif_recsize_max &&
3259                         !(feat->dif_flags & (DT_IND_VARKEY |
3260                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3261                         ergo(feat->dif_flags & DT_IND_UPDATE,
3262                              1 /* XXX check that object (and file system) is
3263                                 * writable */);
3264         }
3265 }
3266
3267 static int osd_iam_container_init(const struct lu_env *env,
3268                                   struct osd_object *obj,
3269                                   struct osd_directory *dir)
3270 {
3271         struct iam_container *bag = &dir->od_container;
3272         int result;
3273
3274         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3275         if (result != 0)
3276                 return result;
3277
3278         result = iam_container_setup(bag);
3279         if (result == 0)
3280                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3281         else
3282                 iam_container_fini(bag);
3283
3284         return result;
3285 }
3286
3287
3288 /*
3289  * Concurrency: no external locking is necessary.
3290  */
3291 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3292                          const struct dt_index_features *feat)
3293 {
3294         int                      result;
3295         int                      skip_iam = 0;
3296         struct osd_object       *obj = osd_dt_obj(dt);
3297
3298         LINVRNT(osd_invariant(obj));
3299
3300         if (osd_object_is_root(obj)) {
3301                 dt->do_index_ops = &osd_index_ea_ops;
3302                 result = 0;
3303         } else if (feat == &dt_directory_features) {
3304                 dt->do_index_ops = &osd_index_ea_ops;
3305                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3306                         result = 0;
3307                 else
3308                         result = -ENOTDIR;
3309                 skip_iam = 1;
3310         } else if (unlikely(feat == &dt_otable_features)) {
3311                 dt->do_index_ops = &osd_otable_ops;
3312                 return 0;
3313         } else if (unlikely(feat == &dt_acct_features)) {
3314                 dt->do_index_ops = &osd_acct_index_ops;
3315                 result = 0;
3316                 skip_iam = 1;
3317         } else if (!osd_has_index(obj)) {
3318                 struct osd_directory *dir;
3319
3320                 OBD_ALLOC_PTR(dir);
3321                 if (dir != NULL) {
3322
3323                         spin_lock(&obj->oo_guard);
3324                         if (obj->oo_dir == NULL)
3325                                 obj->oo_dir = dir;
3326                         else
3327                                 /*
3328                                  * Concurrent thread allocated container data.
3329                                  */
3330                                 OBD_FREE_PTR(dir);
3331                         spin_unlock(&obj->oo_guard);
3332                         /*
3333                          * Now, that we have container data, serialize its
3334                          * initialization.
3335                          */
3336                         down_write(&obj->oo_ext_idx_sem);
3337                         /*
3338                          * recheck under lock.
3339                          */
3340                         if (!osd_has_index(obj))
3341                                 result = osd_iam_container_init(env, obj, dir);
3342                         else
3343                                 result = 0;
3344                         up_write(&obj->oo_ext_idx_sem);
3345                 } else {
3346                         result = -ENOMEM;
3347                 }
3348         } else {
3349                 result = 0;
3350         }
3351
3352         if (result == 0 && skip_iam == 0) {
3353                 if (!osd_iam_index_probe(env, obj, feat))
3354                         result = -ENOTDIR;
3355         }
3356         LINVRNT(osd_invariant(obj));
3357
3358         if (result == 0 && is_quota_glb_feat(feat) &&
3359             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3360                 result = osd_quota_migration(env, dt, feat);
3361
3362         return result;
3363 }
3364
3365 static int osd_otable_it_attr_get(const struct lu_env *env,
3366                                  struct dt_object *dt,
3367                                  struct lu_attr *attr,
3368                                  struct lustre_capa *capa)
3369 {
3370         attr->la_valid = 0;
3371         return 0;
3372 }
3373
3374 static const struct dt_object_operations osd_obj_ops = {
3375         .do_read_lock         = osd_object_read_lock,
3376         .do_write_lock        = osd_object_write_lock,
3377         .do_read_unlock       = osd_object_read_unlock,
3378         .do_write_unlock      = osd_object_write_unlock,
3379         .do_write_locked      = osd_object_write_locked,
3380         .do_attr_get          = osd_attr_get,
3381         .do_declare_attr_set  = osd_declare_attr_set,
3382         .do_attr_set          = osd_attr_set,
3383         .do_ah_init           = osd_ah_init,
3384         .do_declare_create    = osd_declare_object_create,
3385         .do_create            = osd_object_create,
3386         .do_declare_destroy   = osd_declare_object_destroy,
3387         .do_destroy           = osd_object_destroy,
3388         .do_index_try         = osd_index_try,
3389         .do_declare_ref_add   = osd_declare_object_ref_add,
3390         .do_ref_add           = osd_object_ref_add,
3391         .do_declare_ref_del   = osd_declare_object_ref_del,
3392         .do_ref_del           = osd_object_ref_del,
3393         .do_xattr_get         = osd_xattr_get,
3394         .do_declare_xattr_set = osd_declare_xattr_set,
3395         .do_xattr_set         = osd_xattr_set,
3396         .do_declare_xattr_del = osd_declare_xattr_del,
3397         .do_xattr_del         = osd_xattr_del,
3398         .do_xattr_list        = osd_xattr_list,
3399         .do_capa_get          = osd_capa_get,
3400         .do_object_sync       = osd_object_sync,
3401         .do_data_get          = osd_data_get,
3402 };
3403
3404 /**
3405  * dt_object_operations for interoperability mode
3406  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3407  */
3408 static const struct dt_object_operations osd_obj_ea_ops = {
3409         .do_read_lock         = osd_object_read_lock,
3410         .do_write_lock        = osd_object_write_lock,
3411         .do_read_unlock       = osd_object_read_unlock,
3412         .do_write_unlock      = osd_object_write_unlock,
3413         .do_write_locked      = osd_object_write_locked,
3414         .do_attr_get          = osd_attr_get,
3415         .do_declare_attr_set  = osd_declare_attr_set,
3416         .do_attr_set          = osd_attr_set,
3417         .do_ah_init           = osd_ah_init,
3418         .do_declare_create    = osd_declare_object_create,
3419         .do_create            = osd_object_ea_create,
3420         .do_declare_destroy   = osd_declare_object_destroy,
3421         .do_destroy           = osd_object_destroy,
3422         .do_index_try         = osd_index_try,
3423         .do_declare_ref_add   = osd_declare_object_ref_add,
3424         .do_ref_add           = osd_object_ref_add,
3425         .do_declare_ref_del   = osd_declare_object_ref_del,
3426         .do_ref_del           = osd_object_ref_del,
3427         .do_xattr_get         = osd_xattr_get,
3428         .do_declare_xattr_set = osd_declare_xattr_set,
3429         .do_xattr_set         = osd_xattr_set,
3430         .do_declare_xattr_del = osd_declare_xattr_del,
3431         .do_xattr_del         = osd_xattr_del,
3432         .do_xattr_list        = osd_xattr_list,
3433         .do_capa_get          = osd_capa_get,
3434         .do_object_sync       = osd_object_sync,
3435         .do_data_get          = osd_data_get,
3436 };
3437
3438 static const struct dt_object_operations osd_obj_otable_it_ops = {
3439         .do_attr_get    = osd_otable_it_attr_get,
3440         .do_index_try   = osd_index_try,
3441 };
3442
3443 static int osd_index_declare_iam_delete(const struct lu_env *env,
3444                                         struct dt_object *dt,
3445                                         const struct dt_key *key,
3446                                         struct thandle *handle)
3447 {
3448         struct osd_thandle    *oh;
3449
3450         oh = container_of0(handle, struct osd_thandle, ot_super);
3451         LASSERT(oh->ot_handle == NULL);
3452
3453         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3454                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3455
3456         return 0;
3457 }
3458
3459 /**
3460  *      delete a (key, value) pair from index \a dt specified by \a key
3461  *
3462  *      \param  dt      osd index object
3463  *      \param  key     key for index
3464  *      \param  rec     record reference
3465  *      \param  handle  transaction handler
3466  *
3467  *      \retval  0  success
3468  *      \retval -ve   failure
3469  */
3470
3471 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3472                                 const struct dt_key *key,
3473                                 struct thandle *handle,
3474                                 struct lustre_capa *capa)
3475 {
3476         struct osd_thread_info *oti = osd_oti_get(env);
3477         struct osd_object      *obj = osd_dt_obj(dt);
3478         struct osd_thandle     *oh;
3479         struct iam_path_descr  *ipd;
3480         struct iam_container   *bag = &obj->oo_dir->od_container;
3481         int                     rc;
3482
3483         ENTRY;
3484
3485         LINVRNT(osd_invariant(obj));
3486         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3487         LASSERT(bag->ic_object == obj->oo_inode);
3488         LASSERT(handle != NULL);
3489
3490         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3491                 RETURN(-EACCES);
3492
3493         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3494
3495         ipd = osd_idx_ipd_get(env, bag);
3496         if (unlikely(ipd == NULL))
3497                 RETURN(-ENOMEM);
3498
3499         oh = container_of0(handle, struct osd_thandle, ot_super);
3500         LASSERT(oh->ot_handle != NULL);
3501         LASSERT(oh->ot_handle->h_transaction != NULL);
3502
3503         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3504                 /* swab quota uid/gid provided by caller */
3505                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3506                 key = (const struct dt_key *)&oti->oti_quota_id;
3507         }
3508
3509         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3510         osd_ipd_put(env, bag, ipd);
3511         LINVRNT(osd_invariant(obj));
3512         RETURN(rc);
3513 }
3514
3515 static int osd_index_declare_ea_delete(const struct lu_env *env,
3516                                        struct dt_object *dt,
3517                                        const struct dt_key *key,
3518                                        struct thandle *handle)
3519 {
3520         struct osd_thandle *oh;
3521         struct inode       *inode;
3522         int                 rc;
3523         ENTRY;
3524
3525         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3526         LASSERT(handle != NULL);
3527
3528         oh = container_of0(handle, struct osd_thandle, ot_super);
3529         LASSERT(oh->ot_handle == NULL);
3530
3531         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3532                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3533
3534         inode = osd_dt_obj(dt)->oo_inode;
3535         LASSERT(inode);
3536
3537         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3538                                    true, true, NULL, false);
3539         RETURN(rc);
3540 }
3541
3542 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3543                                           struct dt_rec *fid)
3544 {
3545         struct osd_fid_pack *rec;
3546         int                  rc = -ENODATA;
3547
3548         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3549                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3550                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3551         }
3552         return rc;
3553 }
3554
3555 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3556                           struct lu_fid *fid)
3557 {
3558         ENTRY;
3559
3560         /* FID seqs not in FLDB, must be local seq */
3561         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3562                 RETURN(0);
3563
3564         if (osd_seq_exists(env, osd, fid_seq(fid)))
3565                 RETURN(0);
3566
3567         RETURN(1);
3568 }
3569
3570 /**
3571  * Index delete function for interoperability mode (b11826).
3572  * It will remove the directory entry added by osd_index_ea_insert().
3573  * This entry is needed to maintain name->fid mapping.
3574  *
3575  * \param key,  key i.e. file entry to be deleted
3576  *
3577  * \retval   0, on success
3578  * \retval -ve, on error
3579  */
3580 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3581                                const struct dt_key *key,
3582                                struct thandle *handle,
3583                                struct lustre_capa *capa)
3584 {
3585         struct osd_object          *obj    = osd_dt_obj(dt);
3586         struct inode               *dir    = obj->oo_inode;
3587         struct dentry              *dentry;
3588         struct osd_thandle         *oh;
3589         struct ldiskfs_dir_entry_2 *de = NULL;
3590         struct buffer_head         *bh;
3591         struct htree_lock          *hlock = NULL;
3592         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3593         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3594         int                        rc;
3595         ENTRY;
3596
3597         LINVRNT(osd_invariant(obj));
3598         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3599         LASSERT(handle != NULL);
3600
3601         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3602
3603         oh = container_of(handle, struct osd_thandle, ot_super);
3604         LASSERT(oh->ot_handle != NULL);
3605         LASSERT(oh->ot_handle->h_transaction != NULL);
3606
3607         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3608                 RETURN(-EACCES);
3609
3610         ll_vfs_dq_init(dir);
3611         dentry = osd_child_dentry_get(env, obj,
3612                                       (char *)key, strlen((char *)key));
3613
3614         if (obj->oo_hl_head != NULL) {
3615                 hlock = osd_oti_get(env)->oti_hlock;
3616                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3617                                    dir, LDISKFS_HLOCK_DEL);
3618         } else {
3619                 down_write(&obj->oo_ext_idx_sem);
3620         }
3621
3622         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3623         if (bh) {
3624                 __u32 ino = 0;
3625
3626                 /* If this is not the ".." entry, it might be a remote DNE
3627                  * entry and  we need to check if the FID is for a remote
3628                  * MDT.  If the FID is  not in the directory entry (e.g.
3629                  * upgraded 1.8 filesystem without dirdata enabled) then
3630                  * we need to get the FID from the LMA. For a remote directory
3631                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3632                  * case.
3633                  *
3634                  * Delete the entry before the agent inode in order to
3635                  * simplify error handling.  At worst an error after deleting
3636                  * the entry first might leak the agent inode afterward. The
3637                  * reverse would need filesystem abort in case of error deleting
3638                  * the entry after the agent had been removed, or leave a
3639                  * dangling entry pointing at a random inode. */
3640                 if (strcmp((char *)key, dotdot) != 0) {
3641                         LASSERT(de != NULL);
3642                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3643                         /* If Fid is not in dentry, try to get it from LMA */
3644                         if (rc == -ENODATA) {
3645                                 struct osd_inode_id *id;
3646                                 struct inode *inode;
3647
3648                                 /* Before trying to get fid from the inode,
3649                                  * check whether the inode is valid.
3650                                  *
3651                                  * If the inode has been deleted, do not go
3652                                  * ahead to do osd_ea_fid_get, which will set
3653                                  * the inode to bad inode, which might cause
3654                                  * the inode to be deleted uncorrectly */
3655                                 inode = ldiskfs_iget(osd_sb(osd),
3656                                                      le32_to_cpu(de->inode));
3657                                 if (IS_ERR(inode)) {
3658                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3659                                                "error.\n", osd_name(osd),
3660                                                PFID(fid));
3661                                         rc = PTR_ERR(inode);
3662                                 } else {
3663                                         if (likely(inode->i_nlink != 0)) {
3664                                                 id = &osd_oti_get(env)->oti_id;
3665                                                 rc = osd_ea_fid_get(env, obj,
3666                                                         le32_to_cpu(de->inode),
3667                                                                     fid, id);
3668                                         } else {
3669                                                 CDEBUG(D_INFO, "%s: %u "DFID
3670                                                        "deleted.\n",
3671                                                        osd_name(osd),
3672                                                        le32_to_cpu(de->inode),
3673                                                        PFID(fid));
3674                                                 rc = -ESTALE;
3675                                         }
3676                                         iput(inode);
3677                                 }
3678                         }
3679                         if (rc == 0 &&
3680                             unlikely(osd_remote_fid(env, osd, fid)))
3681                                 /* Need to delete agent inode */
3682                                 ino = le32_to_cpu(de->inode);
3683                 }
3684                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3685                 brelse(bh);
3686                 if (rc == 0 && unlikely(ino != 0)) {
3687                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3688                                                           oh);
3689                         if (rc != 0)
3690                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3691                                        osd_name(osd), PFID(fid), rc);
3692                 }
3693         } else {
3694                 rc = -ENOENT;
3695         }
3696         if (hlock != NULL)
3697                 ldiskfs_htree_unlock(hlock);
3698         else
3699                 up_write(&obj->oo_ext_idx_sem);
3700
3701         if (rc != 0)
3702                 GOTO(out, rc);
3703
3704         /* For inode on the remote MDT, .. will point to
3705          * /Agent directory, Check whether it needs to delete
3706          * from agent directory */
3707         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3708                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3709                                                    oh);
3710                 if (rc != 0 && rc != -ENOENT) {
3711                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3712                                osd_name(osd), PFID(fid), rc);
3713                 }
3714
3715                 if (rc == -ENOENT)
3716                         rc = 0;
3717
3718                 GOTO(out, rc);
3719         }
3720 out:
3721
3722         LASSERT(osd_invariant(obj));
3723         RETURN(rc);
3724 }
3725
3726 /**
3727  *      Lookup index for \a key and copy record to \a rec.
3728  *
3729  *      \param  dt      osd index object
3730  *      \param  key     key for index
3731  *      \param  rec     record reference
3732  *
3733  *      \retval  +ve  success : exact mach
3734  *      \retval  0    return record with key not greater than \a key
3735  *      \retval -ve   failure
3736  */
3737 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3738                                 struct dt_rec *rec, const struct dt_key *key,
3739                                 struct lustre_capa *capa)
3740 {
3741         struct osd_object      *obj = osd_dt_obj(dt);
3742         struct iam_path_descr  *ipd;
3743         struct iam_container   *bag = &obj->oo_dir->od_container;
3744         struct osd_thread_info *oti = osd_oti_get(env);
3745         struct iam_iterator    *it = &oti->oti_idx_it;
3746         struct iam_rec         *iam_rec;
3747         int                     rc;
3748
3749         ENTRY;
3750
3751         LASSERT(osd_invariant(obj));
3752         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3753         LASSERT(bag->ic_object == obj->oo_inode);
3754
3755         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3756                 RETURN(-EACCES);
3757
3758         ipd = osd_idx_ipd_get(env, bag);
3759         if (IS_ERR(ipd))
3760                 RETURN(-ENOMEM);
3761
3762         /* got ipd now we can start iterator. */
3763         iam_it_init(it, bag, 0, ipd);
3764
3765         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3766                 /* swab quota uid/gid provided by caller */
3767                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3768                 key = (const struct dt_key *)&oti->oti_quota_id;
3769         }
3770
3771         rc = iam_it_get(it, (struct iam_key *)key);
3772         if (rc >= 0) {
3773                 if (S_ISDIR(obj->oo_inode->i_mode))
3774                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3775                 else
3776                         iam_rec = (struct iam_rec *) rec;
3777
3778                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3779
3780                 if (S_ISDIR(obj->oo_inode->i_mode))
3781                         osd_fid_unpack((struct lu_fid *) rec,
3782                                        (struct osd_fid_pack *)iam_rec);
3783                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3784                         osd_quota_unpack(obj, rec);
3785         }
3786
3787         iam_it_put(it);
3788         iam_it_fini(it);
3789         osd_ipd_put(env, bag, ipd);
3790
3791         LINVRNT(osd_invariant(obj));
3792
3793         RETURN(rc);
3794 }
3795
3796 static int osd_index_declare_iam_insert(const struct lu_env *env,
3797                                         struct dt_object *dt,
3798                                         const struct dt_rec *rec,
3799                                         const struct dt_key *key,
3800                                         struct thandle *handle)
3801 {
3802         struct osd_thandle *oh;
3803
3804         LASSERT(handle != NULL);
3805
3806         oh = container_of0(handle, struct osd_thandle, ot_super);
3807         LASSERT(oh->ot_handle == NULL);
3808
3809         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3810                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3811
3812         return 0;
3813 }
3814
3815 /**
3816  *      Inserts (key, value) pair in \a dt index object.
3817  *
3818  *      \param  dt      osd index object
3819  *      \param  key     key for index
3820  *      \param  rec     record reference
3821  *      \param  th      transaction handler
3822  *
3823  *      \retval  0  success
3824  *      \retval -ve failure
3825  */
3826 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3827                                 const struct dt_rec *rec,
3828                                 const struct dt_key *key, struct thandle *th,
3829                                 struct lustre_capa *capa, int ignore_quota)
3830 {
3831         struct osd_object     *obj = osd_dt_obj(dt);
3832         struct iam_path_descr *ipd;
3833         struct osd_thandle    *oh;
3834         struct iam_container  *bag = &obj->oo_dir->od_container;
3835         struct osd_thread_info *oti = osd_oti_get(env);
3836         struct iam_rec         *iam_rec;
3837         int                     rc;
3838
3839         ENTRY;
3840
3841         LINVRNT(osd_invariant(obj));
3842         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3843         LASSERT(bag->ic_object == obj->oo_inode);
3844         LASSERT(th != NULL);
3845
3846         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3847                 RETURN(-EACCES);
3848
3849         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3850
3851         ipd = osd_idx_ipd_get(env, bag);
3852         if (unlikely(ipd == NULL))
3853                 RETURN(-ENOMEM);
3854
3855         oh = container_of0(th, struct osd_thandle, ot_super);
3856         LASSERT(oh->ot_handle != NULL);
3857         LASSERT(oh->ot_handle->h_transaction != NULL);
3858         if (S_ISDIR(obj->oo_inode->i_mode)) {
3859                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3860                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3861         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3862                 /* pack quota uid/gid */
3863                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3864                 key = (const struct dt_key *)&oti->oti_quota_id;
3865                 /* pack quota record */
3866                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3867                 iam_rec = (struct iam_rec *)rec;
3868         } else {
3869                 iam_rec = (struct iam_rec *)rec;
3870         }
3871
3872         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3873                         iam_rec, ipd);
3874         osd_ipd_put(env, bag, ipd);
3875         LINVRNT(osd_invariant(obj));
3876         RETURN(rc);
3877 }
3878
3879 /**
3880  * Calls ldiskfs_add_entry() to add directory entry
3881  * into the directory. This is required for
3882  * interoperability mode (b11826)
3883  *
3884  * \retval   0, on success
3885  * \retval -ve, on error
3886  */
3887 static int __osd_ea_add_rec(struct osd_thread_info *info,
3888                             struct osd_object *pobj, struct inode  *cinode,
3889                             const char *name, const struct dt_rec *fid,
3890                             struct htree_lock *hlock, struct thandle *th)
3891 {
3892         struct ldiskfs_dentry_param *ldp;
3893         struct dentry               *child;
3894         struct osd_thandle          *oth;
3895         int                          rc;
3896
3897         oth = container_of(th, struct osd_thandle, ot_super);
3898         LASSERT(oth->ot_handle != NULL);
3899         LASSERT(oth->ot_handle->h_transaction != NULL);
3900         LASSERT(pobj->oo_inode);
3901
3902         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3903         if (unlikely(pobj->oo_inode ==
3904                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3905                 ldp->edp_magic = 0;
3906         else
3907                 osd_get_ldiskfs_dirent_param(ldp, fid);
3908         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3909         child->d_fsdata = (void *)ldp;
3910         ll_vfs_dq_init(pobj->oo_inode);
3911         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3912
3913         RETURN(rc);
3914 }
3915
3916 /**
3917  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3918  * into the directory.Also sets flags into osd object to
3919  * indicate dot and dotdot are created. This is required for
3920  * interoperability mode (b11826)
3921  *
3922  * \param dir   directory for dot and dotdot fixup.
3923  * \param obj   child object for linking
3924  *
3925  * \retval   0, on success
3926  * \retval -ve, on error
3927  */
3928 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3929                               struct osd_object *dir,
3930                               struct inode  *parent_dir, const char *name,
3931                               const struct dt_rec *dot_fid,
3932                               const struct dt_rec *dot_dot_fid,
3933                               struct thandle *th)
3934 {
3935         struct inode                *inode = dir->oo_inode;
3936         struct osd_thandle          *oth;
3937         int result = 0;
3938
3939         oth = container_of(th, struct osd_thandle, ot_super);
3940         LASSERT(oth->ot_handle->h_transaction != NULL);
3941         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3942
3943         if (strcmp(name, dot) == 0) {
3944                 if (dir->oo_compat_dot_created) {
3945                         result = -EEXIST;
3946                 } else {
3947                         LASSERT(inode == parent_dir);
3948                         dir->oo_compat_dot_created = 1;
3949                         result = 0;
3950                 }
3951         } else if (strcmp(name, dotdot) == 0) {
3952                 if (!dir->oo_compat_dot_created)
3953                         return -EINVAL;
3954                 /* in case of rename, dotdot is already created */
3955                 if (dir->oo_compat_dotdot_created) {
3956                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3957                                                 dot_dot_fid, NULL, th);
3958                 }
3959
3960                 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3961                                                 parent_dir, dot_fid,
3962                                                 dot_dot_fid, oth);
3963                 if (result == 0)
3964                         dir->oo_compat_dotdot_created = 1;
3965         }
3966
3967         return result;
3968 }
3969
3970
3971 /**
3972  * It will call the appropriate osd_add* function and return the
3973  * value, return by respective functions.
3974  */
3975 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3976                           struct inode *cinode, const char *name,
3977                           const struct dt_rec *fid, struct thandle *th)
3978 {
3979         struct osd_thread_info *info   = osd_oti_get(env);
3980         struct htree_lock      *hlock;
3981         int                     rc;
3982
3983         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3984
3985         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3986                                                    name[2] =='\0'))) {
3987                 if (hlock != NULL) {
3988                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3989                                            pobj->oo_inode, 0);
3990                 } else {
3991                         down_write(&pobj->oo_ext_idx_sem);
3992                 }
3993                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3994                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3995                                         fid, th);
3996         } else {
3997                 if (hlock != NULL) {
3998                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3999                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4000                 } else {
4001                         down_write(&pobj->oo_ext_idx_sem);
4002                 }
4003
4004                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4005                         struct lu_fid *tfid = &info->oti_fid;
4006
4007                         *tfid = *(const struct lu_fid *)fid;
4008                         tfid->f_ver = ~0;
4009                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4010                                               (const struct dt_rec *)tfid,
4011                                               hlock, th);
4012                 } else {
4013                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4014                                               hlock, th);
4015                 }
4016         }
4017         if (hlock != NULL)
4018                 ldiskfs_htree_unlock(hlock);
4019         else
4020                 up_write(&pobj->oo_ext_idx_sem);
4021
4022         return rc;
4023 }
4024
4025 static void
4026 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4027                       struct osd_idmap_cache *oic)
4028 {
4029         struct osd_scrub    *scrub = &dev->od_scrub;
4030         struct lu_fid       *fid   = &oic->oic_fid;
4031         struct osd_inode_id *id    = &oti->oti_id;
4032         int                  once  = 0;
4033         int                  rc;
4034         ENTRY;
4035
4036         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4037                 RETURN_EXIT;
4038
4039 again:
4040         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4041         if (rc != 0 && rc != -ENOENT)
4042                 RETURN_EXIT;
4043
4044         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4045                 RETURN_EXIT;
4046
4047         if (thread_is_running(&scrub->os_thread)) {
4048                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4049                 /* There is race condition between osd_oi_lookup and OI scrub.
4050                  * The OI scrub finished just after osd_oi_lookup() failure.
4051                  * Under such case, it is unnecessary to trigger OI scrub again,
4052                  * but try to call osd_oi_lookup() again. */
4053                 if (unlikely(rc == -EAGAIN))
4054                         goto again;
4055
4056                 RETURN_EXIT;
4057         }
4058
4059         if (!dev->od_noscrub && ++once == 1) {
4060                 rc = osd_scrub_start(dev);
4061                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4062                               ", rc = %d [2]\n",
4063                               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4064                               PFID(fid), rc);
4065                 if (rc == 0)
4066                         goto again;
4067         }
4068
4069         EXIT;
4070 }
4071
4072 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4073                                struct osd_device *dev,
4074                                struct osd_idmap_cache *oic,
4075                                struct lu_fid *fid, __u32 ino)
4076 {
4077         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4078         struct inode            *inode;
4079         int                      rc;
4080
4081         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4082         inode = osd_iget(oti, dev, &oic->oic_lid);
4083         if (IS_ERR(inode)) {
4084                 fid_zero(&oic->oic_fid);
4085                 return PTR_ERR(inode);
4086         }
4087
4088         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4089         iput(inode);
4090         if (rc != 0)
4091                 fid_zero(&oic->oic_fid);
4092         else
4093                 *fid = oic->oic_fid = lma->lma_self_fid;
4094         return rc;
4095 }
4096
4097 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4098                      struct osd_inode_id *id, const struct lu_fid *fid)
4099 {
4100         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4101                id->oii_ino, id->oii_gen, info);
4102         info->oti_cache.oic_lid = *id;
4103         info->oti_cache.oic_fid = *fid;
4104         info->oti_cache.oic_dev = osd;
4105
4106         return 0;
4107 }
4108
4109 /**
4110  * Calls ->lookup() to find dentry. From dentry get inode and
4111  * read inode's ea to get fid. This is required for  interoperability
4112  * mode (b11826)
4113  *
4114  * \retval   0, on success
4115  * \retval -ve, on error
4116  */
4117 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4118                              struct dt_rec *rec, const struct dt_key *key)
4119 {
4120         struct inode               *dir    = obj->oo_inode;
4121         struct dentry              *dentry;
4122         struct ldiskfs_dir_entry_2 *de;
4123         struct buffer_head         *bh;
4124         struct lu_fid              *fid = (struct lu_fid *) rec;
4125         struct htree_lock          *hlock = NULL;
4126         int                         ino;
4127         int                         rc;
4128         ENTRY;
4129
4130         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4131
4132         dentry = osd_child_dentry_get(env, obj,
4133                                       (char *)key, strlen((char *)key));
4134
4135         if (obj->oo_hl_head != NULL) {
4136                 hlock = osd_oti_get(env)->oti_hlock;
4137                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4138                                    dir, LDISKFS_HLOCK_LOOKUP);
4139         } else {
4140                 down_read(&obj->oo_ext_idx_sem);
4141         }
4142
4143         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4144         if (bh) {
4145                 struct osd_thread_info *oti = osd_oti_get(env);
4146                 struct osd_inode_id *id = &oti->oti_id;
4147                 struct osd_idmap_cache *oic = &oti->oti_cache;
4148                 struct osd_device *dev = osd_obj2dev(obj);
4149                 struct osd_scrub *scrub = &dev->od_scrub;
4150                 struct scrub_file *sf = &scrub->os_file;
4151
4152                 ino = le32_to_cpu(de->inode);
4153                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4154                         brelse(bh);
4155                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4156                         GOTO(out, rc);
4157                 }
4158
4159                 rc = osd_get_fid_from_dentry(de, rec);
4160
4161                 /* done with de, release bh */
4162                 brelse(bh);
4163                 if (rc != 0)
4164                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
4165                 else
4166                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4167                 if (rc != 0) {
4168                         fid_zero(&oic->oic_fid);
4169                         GOTO(out, rc);
4170                 }
4171
4172                 if (osd_remote_fid(env, dev, fid))
4173                         GOTO(out, rc = 0);
4174
4175                 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4176                                       fid);
4177                 if (rc != 0)
4178                         GOTO(out, rc);
4179                 if ((scrub->os_pos_current <= ino) &&
4180                     ((sf->sf_flags & SF_INCONSISTENT) ||
4181                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4182                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4183                                       sf->sf_oi_bitmap)))
4184                         osd_consistency_check(oti, dev, oic);
4185         } else {
4186                 rc = -ENOENT;
4187         }
4188
4189         GOTO(out, rc);
4190
4191 out:
4192         if (hlock != NULL)
4193                 ldiskfs_htree_unlock(hlock);
4194         else
4195                 up_read(&obj->oo_ext_idx_sem);
4196         return rc;
4197 }
4198
4199 /**
4200  * Find the osd object for given fid.
4201  *
4202  * \param fid need to find the osd object having this fid
4203  *
4204  * \retval osd_object on success
4205  * \retval        -ve on error
4206  */
4207 struct osd_object *osd_object_find(const struct lu_env *env,
4208                                    struct dt_object *dt,
4209                                    const struct lu_fid *fid)
4210 {
4211         struct lu_device  *ludev = dt->do_lu.lo_dev;
4212         struct osd_object *child = NULL;
4213         struct lu_object  *luch;
4214         struct lu_object  *lo;
4215
4216         /*
4217          * at this point topdev might not exist yet
4218          * (i.e. MGS is preparing profiles). so we can
4219          * not rely on topdev and instead lookup with
4220          * our device passed as topdev. this can't work
4221          * if the object isn't cached yet (as osd doesn't
4222          * allocate lu_header). IOW, the object must be
4223          * in the cache, otherwise lu_object_alloc() crashes
4224          * -bzzz
4225          */
4226         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4227                                  ludev : ludev->ld_site->ls_top_dev,
4228                                  fid, NULL);
4229         if (!IS_ERR(luch)) {
4230                 if (lu_object_exists(luch)) {
4231                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4232                         if (lo != NULL)
4233                                 child = osd_obj(lo);
4234                         else
4235                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4236                                                 "lu_object can't be located"
4237                                                 DFID"\n", PFID(fid));
4238
4239                         if (child == NULL) {
4240                                 lu_object_put(env, luch);
4241                                 CERROR("Unable to get osd_object\n");
4242                                 child = ERR_PTR(-ENOENT);
4243                         }
4244                 } else {
4245                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4246                                         "lu_object does not exists "DFID"\n",
4247                                         PFID(fid));
4248                         lu_object_put(env, luch);
4249                         child = ERR_PTR(-ENOENT);
4250                 }
4251         } else {
4252                 child = ERR_CAST(luch);
4253         }
4254
4255         return child;
4256 }
4257
4258 /**
4259  * Put the osd object once done with it.
4260  *
4261  * \param obj osd object that needs to be put
4262  */
4263 static inline void osd_object_put(const struct lu_env *env,
4264                                   struct osd_object *obj)
4265 {
4266         lu_object_put(env, &obj->oo_dt.do_lu);
4267 }
4268
4269 static int osd_index_declare_ea_insert(const struct lu_env *env,
4270                                        struct dt_object *dt,
4271                                        const struct dt_rec *rec,
4272                                        const struct dt_key *key,
4273                                        struct thandle *handle)
4274 {
4275         struct osd_thandle      *oh;
4276         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4277         struct lu_fid           *fid = (struct lu_fid *)rec;
4278         int                     rc;
4279         ENTRY;
4280
4281         LASSERT(!dt_object_remote(dt));
4282         LASSERT(handle != NULL);
4283
4284         oh = container_of0(handle, struct osd_thandle, ot_super);
4285         LASSERT(oh->ot_handle == NULL);
4286
4287         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4288                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4289
4290         if (osd_dt_obj(dt)->oo_inode != NULL) {
4291                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4292
4293                 /* We ignore block quota on meta pool (MDTs), so needn't
4294                  * calculate how many blocks will be consumed by this index
4295                  * insert */
4296                 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
4297                                            oh, true, true, NULL, false);
4298         }
4299
4300         if (fid == NULL)
4301                 RETURN(0);
4302
4303         rc = osd_remote_fid(env, osd, fid);
4304         if (rc <= 0)
4305                 RETURN(rc);
4306
4307         rc = 0;
4308
4309         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4310                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4311         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4312                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4313         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4314                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4315
4316         RETURN(rc);
4317 }
4318
4319 /**
4320  * Index add function for interoperability mode (b11826).
4321  * It will add the directory entry.This entry is needed to
4322  * maintain name->fid mapping.
4323  *
4324  * \param key it is key i.e. file entry to be inserted
4325  * \param rec it is value of given key i.e. fid
4326  *
4327  * \retval   0, on success
4328  * \retval -ve, on error
4329  */
4330 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4331                                const struct dt_rec *rec,
4332                                const struct dt_key *key, struct thandle *th,
4333                                struct lustre_capa *capa, int ignore_quota)
4334 {
4335         struct osd_object       *obj = osd_dt_obj(dt);
4336         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4337         struct lu_fid           *fid = (struct lu_fid *) rec;
4338         const char              *name = (const char *)key;
4339         struct osd_thread_info  *oti   = osd_oti_get(env);
4340         struct osd_inode_id     *id    = &oti->oti_id;
4341         struct inode            *child_inode = NULL;
4342         struct osd_object       *child = NULL;
4343         int                     rc;
4344         ENTRY;
4345
4346         LASSERT(osd_invariant(obj));
4347         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4348         LASSERT(th != NULL);
4349
4350         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4351
4352         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4353                 RETURN(-EACCES);
4354
4355         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4356
4357         rc = osd_remote_fid(env, osd, fid);
4358         if (rc < 0) {
4359                 CERROR("%s: Can not find object "DFID" rc %d\n",
4360                        osd_name(osd), PFID(fid), rc);
4361                 RETURN(rc);
4362         }
4363
4364         if (rc == 1) {
4365                 /* Insert remote entry */
4366                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4367                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4368                         struct osd_thandle      *oh;
4369
4370                         /* If parent on remote MDT, we need put this object
4371                          * under AGENT */
4372                         oh = container_of(th, typeof(*oh), ot_super);
4373                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4374                         if (rc != 0) {
4375                                 CERROR("%s: add "DFID" error: rc = %d\n",
4376                                        osd_name(osd),
4377                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4378                                 RETURN(rc);
4379                         }
4380
4381                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4382                 } else {
4383                         child_inode = osd_create_local_agent_inode(env, osd,
4384                                                                    obj, fid,
4385                                                                    th);
4386                         if (IS_ERR(child_inode))
4387                                 RETURN(PTR_ERR(child_inode));
4388                 }
4389         } else {
4390                 /* Insert local entry */
4391                 child = osd_object_find(env, dt, fid);
4392                 if (IS_ERR(child)) {
4393                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4394                                osd_name(osd), PFID(fid),
4395                                id->oii_ino, id->oii_gen,
4396                                (int)PTR_ERR(child));
4397                         RETURN(PTR_ERR(child));
4398                 }
4399                 child_inode = igrab(child->oo_inode);
4400         }
4401
4402         rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4403
4404         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4405                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4406
4407         iput(child_inode);
4408         if (child != NULL)
4409                 osd_object_put(env, child);
4410         LASSERT(osd_invariant(obj));
4411         RETURN(rc);
4412 }
4413
4414 /**
4415  *  Initialize osd Iterator for given osd index object.
4416  *
4417  *  \param  dt      osd index object
4418  */
4419
4420 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4421                                      struct dt_object *dt,
4422                                      __u32 unused,
4423                                      struct lustre_capa *capa)
4424 {
4425         struct osd_it_iam      *it;
4426         struct osd_thread_info *oti = osd_oti_get(env);
4427         struct osd_object      *obj = osd_dt_obj(dt);
4428         struct lu_object       *lo  = &dt->do_lu;
4429         struct iam_path_descr  *ipd;
4430         struct iam_container   *bag = &obj->oo_dir->od_container;
4431
4432         LASSERT(lu_object_exists(lo));
4433
4434         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4435                 return ERR_PTR(-EACCES);
4436
4437         it = &oti->oti_it;
4438         ipd = osd_it_ipd_get(env, bag);
4439         if (likely(ipd != NULL)) {
4440                 it->oi_obj = obj;
4441                 it->oi_ipd = ipd;
4442                 lu_object_get(lo);
4443                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4444                 return (struct dt_it *)it;
4445         }
4446         return ERR_PTR(-ENOMEM);
4447 }
4448
4449 /**
4450  * free given Iterator.
4451  */
4452
4453 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4454 {
4455         struct osd_it_iam *it = (struct osd_it_iam *)di;
4456         struct osd_object *obj = it->oi_obj;
4457
4458         iam_it_fini(&it->oi_it);
4459         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4460         lu_object_put(env, &obj->oo_dt.do_lu);
4461 }
4462
4463 /**
4464  *  Move Iterator to record specified by \a key
4465  *
4466  *  \param  di      osd iterator
4467  *  \param  key     key for index
4468  *
4469  *  \retval +ve  di points to record with least key not larger than key
4470  *  \retval  0   di points to exact matched key
4471  *  \retval -ve  failure
4472  */
4473
4474 static int osd_it_iam_get(const struct lu_env *env,
4475                           struct dt_it *di, const struct dt_key *key)
4476 {
4477         struct osd_thread_info  *oti = osd_oti_get(env);
4478         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4479
4480         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4481                 /* swab quota uid/gid */
4482                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4483                 key = (struct dt_key *)&oti->oti_quota_id;
4484         }
4485
4486         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4487 }
4488
4489 /**
4490  *  Release Iterator
4491  *
4492  *  \param  di      osd iterator
4493  */
4494 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4495 {
4496         struct osd_it_iam *it = (struct osd_it_iam *)di;
4497
4498         iam_it_put(&it->oi_it);
4499 }
4500
4501 /**
4502  *  Move iterator by one record
4503  *
4504  *  \param  di      osd iterator
4505  *
4506  *  \retval +1   end of container reached
4507  *  \retval  0   success
4508  *  \retval -ve  failure
4509  */
4510
4511 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4512 {
4513         struct osd_it_iam *it = (struct osd_it_iam *)di;
4514
4515         return iam_it_next(&it->oi_it);
4516 }
4517
4518 /**
4519  * Return pointer to the key under iterator.
4520  */
4521
4522 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4523                                  const struct dt_it *di)
4524 {
4525         struct osd_thread_info *oti = osd_oti_get(env);
4526         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4527         struct osd_object      *obj = it->oi_obj;
4528         struct dt_key          *key;
4529
4530         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4531
4532         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4533                 /* swab quota uid/gid */
4534                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4535                 key = (struct dt_key *)&oti->oti_quota_id;
4536         }
4537
4538         return key;
4539 }
4540
4541 /**
4542  * Return size of key under iterator (in bytes)
4543  */
4544
4545 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4546 {
4547         struct osd_it_iam *it = (struct osd_it_iam *)di;
4548
4549         return iam_it_key_size(&it->oi_it);
4550 }
4551
4552 static inline void
4553 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4554 {
4555         /* check if file type is required */
4556         if (ent->lde_attrs & LUDA_TYPE) {
4557                 struct luda_type *lt;
4558                 int align = sizeof(*lt) - 1;
4559
4560                 len = (len + align) & ~align;
4561                 lt = (struct luda_type *)(ent->lde_name + len);
4562                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4563         }
4564
4565         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4566 }
4567
4568 /**
4569  * build lu direct from backend fs dirent.
4570  */
4571
4572 static inline void
4573 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4574                    char *name, __u16 namelen, __u16 type, __u32 attr)
4575 {
4576         ent->lde_attrs = attr | LUDA_FID;
4577         fid_cpu_to_le(&ent->lde_fid, fid);
4578
4579         ent->lde_hash = cpu_to_le64(offset);
4580         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4581
4582         strncpy(ent->lde_name, name, namelen);
4583         ent->lde_name[namelen] = '\0';
4584         ent->lde_namelen = cpu_to_le16(namelen);
4585
4586         /* append lustre attributes */
4587         osd_it_append_attrs(ent, namelen, type);
4588 }
4589
4590 /**
4591  * Return pointer to the record under iterator.
4592  */
4593 static int osd_it_iam_rec(const struct lu_env *env,
4594                           const struct dt_it *di,
4595                           struct dt_rec *dtrec, __u32 attr)
4596 {
4597         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4598         struct osd_thread_info *info = osd_oti_get(env);
4599         ENTRY;
4600
4601         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4602                 const struct osd_fid_pack *rec;
4603                 struct lu_fid             *fid = &info->oti_fid;
4604                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4605                 char                      *name;
4606                 int                        namelen;
4607                 __u64                      hash;
4608                 int                        rc;
4609
4610                 name = (char *)iam_it_key_get(&it->oi_it);
4611                 if (IS_ERR(name))
4612                         RETURN(PTR_ERR(name));
4613
4614                 namelen = iam_it_key_size(&it->oi_it);
4615
4616                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4617                 if (IS_ERR(rec))
4618                         RETURN(PTR_ERR(rec));
4619
4620                 rc = osd_fid_unpack(fid, rec);
4621                 if (rc)
4622                         RETURN(rc);
4623
4624                 hash = iam_it_store(&it->oi_it);
4625
4626                 /* IAM does not store object type in IAM index (dir) */
4627                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4628                                    0, LUDA_FID);
4629         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4630                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4631                            (struct iam_rec *)dtrec);
4632                 osd_quota_unpack(it->oi_obj, dtrec);
4633         } else {
4634                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4635                            (struct iam_rec *)dtrec);
4636         }
4637
4638         RETURN(0);
4639 }
4640
4641 /**
4642  * Returns cookie for current Iterator position.
4643  */
4644 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4645 {
4646         struct osd_it_iam *it = (struct osd_it_iam *)di;
4647
4648         return iam_it_store(&it->oi_it);
4649 }
4650
4651 /**
4652  * Restore iterator from cookie.
4653  *
4654  * \param  di      osd iterator
4655  * \param  hash    Iterator location cookie
4656  *
4657  * \retval +ve  di points to record with least key not larger than key.
4658  * \retval  0   di points to exact matched key
4659  * \retval -ve  failure
4660  */
4661
4662 static int osd_it_iam_load(const struct lu_env *env,
4663                            const struct dt_it *di, __u64 hash)
4664 {
4665         struct osd_it_iam *it = (struct osd_it_iam *)di;
4666
4667         return iam_it_load(&it->oi_it, hash);
4668 }
4669
4670 static const struct dt_index_operations osd_index_iam_ops = {
4671         .dio_lookup         = osd_index_iam_lookup,
4672         .dio_declare_insert = osd_index_declare_iam_insert,
4673         .dio_insert         = osd_index_iam_insert,
4674         .dio_declare_delete = osd_index_declare_iam_delete,
4675         .dio_delete         = osd_index_iam_delete,
4676         .dio_it     = {
4677                 .init     = osd_it_iam_init,
4678                 .fini     = osd_it_iam_fini,
4679                 .get      = osd_it_iam_get,
4680                 .put      = osd_it_iam_put,
4681                 .next     = osd_it_iam_next,
4682                 .key      = osd_it_iam_key,
4683                 .key_size = osd_it_iam_key_size,
4684                 .rec      = osd_it_iam_rec,
4685                 .store    = osd_it_iam_store,
4686                 .load     = osd_it_iam_load
4687         }
4688 };
4689
4690
4691 /**
4692  * Creates or initializes iterator context.
4693  *
4694  * \retval struct osd_it_ea, iterator structure on success
4695  *
4696  */
4697 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4698                                     struct dt_object *dt,
4699                                     __u32 attr,
4700                                     struct lustre_capa *capa)
4701 {
4702         struct osd_object       *obj  = osd_dt_obj(dt);
4703         struct osd_thread_info  *info = osd_oti_get(env);
4704         struct osd_it_ea        *it   = &info->oti_it_ea;
4705         struct file             *file = &it->oie_file;
4706         struct lu_object        *lo   = &dt->do_lu;
4707         struct dentry           *obj_dentry = &info->oti_it_dentry;
4708         ENTRY;
4709         LASSERT(lu_object_exists(lo));
4710
4711         obj_dentry->d_inode = obj->oo_inode;
4712         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4713         obj_dentry->d_name.hash = 0;
4714
4715         it->oie_rd_dirent       = 0;
4716         it->oie_it_dirent       = 0;
4717         it->oie_dirent          = NULL;
4718         it->oie_buf             = info->oti_it_ea_buf;
4719         it->oie_obj             = obj;
4720
4721         /* Reset the "file" totally to avoid to reuse any old value from
4722          * former readdir handling, the "file->f_pos" should be zero. */
4723         memset(file, 0, sizeof(*file));
4724         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4725         if (attr & LUDA_64BITHASH)
4726                 file->f_mode    = FMODE_64BITHASH;
4727         else
4728                 file->f_mode    = FMODE_32BITHASH;
4729         file->f_dentry          = obj_dentry;
4730         file->f_mapping         = obj->oo_inode->i_mapping;
4731         file->f_op              = obj->oo_inode->i_fop;
4732         set_file_inode(file, obj->oo_inode);
4733
4734         lu_object_get(lo);
4735         RETURN((struct dt_it *) it);
4736 }
4737
4738 /**
4739  * Destroy or finishes iterator context.
4740  *
4741  * \param di iterator structure to be destroyed
4742  */
4743 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4744 {
4745         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4746         struct osd_object    *obj  = it->oie_obj;
4747         struct inode       *inode  = obj->oo_inode;
4748
4749         ENTRY;
4750         it->oie_file.f_op->release(inode, &it->oie_file);
4751         lu_object_put(env, &obj->oo_dt.do_lu);
4752         EXIT;
4753 }
4754
4755 /**
4756  * It position the iterator at given key, so that next lookup continues from
4757  * that key Or it is similar to dio_it->load() but based on a key,
4758  * rather than file position.
4759  *
4760  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4761  * to the beginning.
4762  *
4763  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4764  */
4765 static int osd_it_ea_get(const struct lu_env *env,
4766                          struct dt_it *di, const struct dt_key *key)
4767 {
4768         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4769
4770         ENTRY;
4771         LASSERT(((const char *)key)[0] == '\0');
4772         it->oie_file.f_pos      = 0;
4773         it->oie_rd_dirent       = 0;
4774         it->oie_it_dirent       = 0;
4775         it->oie_dirent          = NULL;
4776
4777         RETURN(+1);
4778 }
4779
4780 /**
4781  * Does nothing
4782  */
4783 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4784 {
4785 }
4786
4787 struct osd_filldir_cbs {
4788 #ifdef HAVE_DIR_CONTEXT
4789         struct dir_context ctx;
4790 #endif
4791         struct osd_it_ea  *it;
4792 };
4793 /**
4794  * It is called internally by ->readdir(). It fills the
4795  * iterator's in-memory data structure with required
4796  * information i.e. name, namelen, rec_size etc.
4797  *
4798  * \param buf in which information to be filled in.
4799  * \param name name of the file in given dir
4800  *
4801  * \retval 0 on success
4802  * \retval 1 on buffer full
4803  */
4804 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4805                                loff_t offset, __u64 ino,
4806                                unsigned d_type)
4807 {
4808         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
4809         struct osd_object       *obj  = it->oie_obj;
4810         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4811         struct lu_fid           *fid  = &ent->oied_fid;
4812         struct osd_fid_pack     *rec;
4813         ENTRY;
4814
4815         /* this should never happen */
4816         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4817                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4818                 RETURN(-EIO);
4819         }
4820
4821         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4822             OSD_IT_EA_BUFSIZE)
4823                 RETURN(1);
4824
4825         /* "." is just the object itself. */
4826         if (namelen == 1 && name[0] == '.') {
4827                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4828         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4829                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4830                 if (osd_fid_unpack(fid, rec) != 0)
4831                         fid_zero(fid);
4832         } else {
4833                 fid_zero(fid);
4834         }
4835         d_type &= ~LDISKFS_DIRENT_LUFID;
4836
4837         /* NOT export local root. */
4838         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4839                 ino = obj->oo_inode->i_ino;
4840                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4841         }
4842
4843         ent->oied_ino     = ino;
4844         ent->oied_off     = offset;
4845         ent->oied_namelen = namelen;
4846         ent->oied_type    = d_type;
4847
4848         memcpy(ent->oied_name, name, namelen);
4849
4850         it->oie_rd_dirent++;
4851         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4852         RETURN(0);
4853 }
4854
4855 /**
4856  * Calls ->readdir() to load a directory entry at a time
4857  * and stored it in iterator's in-memory data structure.
4858  *
4859  * \param di iterator's in memory structure
4860  *
4861  * \retval   0 on success
4862  * \retval -ve on error
4863  * \retval +1 reach the end of entry
4864  */
4865 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4866                                const struct dt_it *di)
4867 {
4868         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4869         struct osd_object  *obj   = it->oie_obj;
4870         struct inode       *inode = obj->oo_inode;
4871         struct htree_lock  *hlock = NULL;
4872         struct file        *filp  = &it->oie_file;
4873         int                 rc = 0;
4874         struct osd_filldir_cbs buf = {
4875 #ifdef HAVE_DIR_CONTEXT
4876                 .ctx.actor = osd_ldiskfs_filldir,
4877 #endif
4878                 .it = it
4879         };
4880
4881         ENTRY;
4882         it->oie_dirent = it->oie_buf;
4883         it->oie_rd_dirent = 0;
4884
4885         if (obj->oo_hl_head != NULL) {
4886                 hlock = osd_oti_get(env)->oti_hlock;
4887                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4888                                    inode, LDISKFS_HLOCK_READDIR);
4889         } else {
4890                 down_read(&obj->oo_ext_idx_sem);
4891         }
4892
4893 #ifdef HAVE_DIR_CONTEXT
4894         buf.ctx.pos = filp->f_pos;
4895         rc = inode->i_fop->iterate(filp, &buf.ctx);
4896         filp->f_pos = buf.ctx.pos;
4897 #else
4898         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4899 #endif
4900
4901         if (hlock != NULL)
4902                 ldiskfs_htree_unlock(hlock);
4903         else
4904                 up_read(&obj->oo_ext_idx_sem);
4905
4906         if (it->oie_rd_dirent == 0) {
4907                 /*If it does not get any dirent, it means it has been reached
4908                  *to the end of the dir */
4909                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4910                 if (rc == 0)
4911                         rc = 1;
4912         } else {
4913                 it->oie_dirent = it->oie_buf;
4914                 it->oie_it_dirent = 1;
4915         }
4916
4917         RETURN(rc);
4918 }
4919
4920 /**
4921  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4922  * to load a directory entry at a time and stored it in
4923  * iterator's in-memory data structure.
4924  *
4925  * \param di iterator's in memory structure
4926  *
4927  * \retval +ve iterator reached to end
4928  * \retval   0 iterator not reached to end
4929  * \retval -ve on error
4930  */
4931 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4932 {
4933         struct osd_it_ea *it = (struct osd_it_ea *)di;
4934         int rc;
4935
4936         ENTRY;
4937
4938         if (it->oie_it_dirent < it->oie_rd_dirent) {
4939                 it->oie_dirent =
4940                         (void *) it->oie_dirent +
4941                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4942                                        it->oie_dirent->oied_namelen);
4943                 it->oie_it_dirent++;
4944                 RETURN(0);
4945         } else {
4946                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4947                         rc = +1;
4948                 else
4949                         rc = osd_ldiskfs_it_fill(env, di);
4950         }
4951
4952         RETURN(rc);
4953 }
4954
4955 /**
4956  * Returns the key at current position from iterator's in memory structure.
4957  *
4958  * \param di iterator's in memory structure
4959  *
4960  * \retval key i.e. struct dt_key on success
4961  */
4962 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4963                                     const struct dt_it *di)
4964 {
4965         struct osd_it_ea *it = (struct osd_it_ea *)di;
4966
4967         return (struct dt_key *)it->oie_dirent->oied_name;
4968 }
4969
4970 /**
4971  * Returns the key's size at current position from iterator's in memory structure.
4972  *
4973  * \param di iterator's in memory structure
4974  *
4975  * \retval key_size i.e. struct dt_key on success
4976  */
4977 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4978 {
4979         struct osd_it_ea *it = (struct osd_it_ea *)di;
4980
4981         return it->oie_dirent->oied_namelen;
4982 }
4983
4984 static int
4985 osd_dirent_update(handle_t *jh, struct super_block *sb,
4986                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4987                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4988 {
4989         struct osd_fid_pack *rec;
4990         int                  rc;
4991         ENTRY;
4992
4993         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4994         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4995
4996         rc = ldiskfs_journal_get_write_access(jh, bh);
4997         if (rc != 0) {
4998                 CERROR("%.16s: fail to write access for update dirent: "
4999                        "name = %.*s, rc = %d\n",
5000                        LDISKFS_SB(sb)->s_es->s_volume_name,
5001                        ent->oied_namelen, ent->oied_name, rc);
5002                 RETURN(rc);
5003         }
5004
5005         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5006         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5007         rc = ldiskfs_journal_dirty_metadata(jh, bh);
5008         if (rc != 0)
5009                 CERROR("%.16s: fail to dirty metadata for update dirent: "
5010                        "name = %.*s, rc = %d\n",
5011                        LDISKFS_SB(sb)->s_es->s_volume_name,
5012                        ent->oied_namelen, ent->oied_name, rc);
5013
5014         RETURN(rc);
5015 }
5016
5017 static inline int
5018 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5019 {
5020         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5021             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5022                 return 1;
5023         else
5024                 return 0;
5025 }
5026
5027 static inline int
5028 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5029 {
5030         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5031                  "dot_dotdot = %d\n", dot_dotdot);
5032
5033         if (LDISKFS_DIR_REC_LEN(de) >=
5034             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5035                 return 1;
5036         else
5037                 return 0;
5038 }
5039
5040 static int
5041 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5042                     struct inode *dir, struct inode *inode,
5043                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5044                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5045                     struct htree_lock *hlock)
5046 {
5047         struct dentry               *dentry;
5048         struct osd_fid_pack         *rec;
5049         struct ldiskfs_dentry_param *ldp;
5050         int                          rc;
5051         ENTRY;
5052
5053         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5054                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5055                 RETURN(0);
5056
5057         /* There is enough space to hold the FID-in-dirent. */
5058         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5059                                  dir->i_sb->s_blocksize)) {
5060                 rc = ldiskfs_journal_get_write_access(jh, bh);
5061                 if (rc != 0) {
5062                         CERROR("%.16s: fail to write access for reinsert "
5063                                "dirent: name = %.*s, rc = %d\n",
5064                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5065                                ent->oied_namelen, ent->oied_name, rc);
5066                         RETURN(rc);
5067                 }
5068
5069                 de->name[de->name_len] = 0;
5070                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5071                 rec->fp_len = sizeof(struct lu_fid) + 1;
5072                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5073                 de->file_type |= LDISKFS_DIRENT_LUFID;
5074
5075                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5076                 if (rc != 0)
5077                         CERROR("%.16s: fail to dirty metadata for reinsert "
5078                                "dirent: name = %.*s, rc = %d\n",
5079                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5080                                ent->oied_namelen, ent->oied_name, rc);
5081
5082                 RETURN(rc);
5083         }
5084
5085         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5086         if (rc != 0) {
5087                 CERROR("%.16s: fail to delete entry for reinsert dirent: "
5088                        "name = %.*s, rc = %d\n",
5089                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5090                        ent->oied_namelen, ent->oied_name, rc);
5091                 RETURN(rc);
5092         }
5093
5094         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5095                                            ent->oied_namelen);
5096         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5097         osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
5098         dentry->d_fsdata = (void *)ldp;
5099         ll_vfs_dq_init(dir);
5100         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5101         /* It is too bad, we cannot reinsert the name entry back.
5102          * That means we lose it! */
5103         if (rc != 0)
5104                 CERROR("%.16s: fail to insert entry for reinsert dirent: "
5105                        "name = %.*s, rc = %d\n",
5106                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5107                        ent->oied_namelen, ent->oied_name, rc);
5108
5109         RETURN(rc);
5110 }
5111
5112 static int
5113 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5114                         struct osd_it_ea *it, struct lu_fid *fid,
5115                         struct osd_inode_id *id, __u32 *attr)
5116 {
5117         struct osd_thread_info     *info        = osd_oti_get(env);
5118         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5119         struct osd_device          *dev         = osd_obj2dev(obj);
5120         struct super_block         *sb          = osd_sb(dev);
5121         const char                 *devname     =
5122                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5123         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5124         struct inode               *dir         = obj->oo_inode;
5125         struct htree_lock          *hlock       = NULL;
5126         struct buffer_head         *bh          = NULL;
5127         handle_t                   *jh          = NULL;
5128         struct ldiskfs_dir_entry_2 *de;
5129         struct dentry              *dentry;
5130         struct inode               *inode;
5131         int                         credits;
5132         int                         rc;
5133         int                         dot_dotdot  = 0;
5134         bool                        dirty       = false;
5135         ENTRY;
5136
5137         if (ent->oied_name[0] == '.') {
5138                 if (ent->oied_namelen == 1)
5139                         dot_dotdot = 1;
5140                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5141                         dot_dotdot = 2;
5142         }
5143
5144         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5145                                       ent->oied_namelen);
5146
5147         /* We need to ensure that the name entry is still valid.
5148          * Because it may be removed or renamed by other already.
5149          *
5150          * The unlink or rename operation will start journal before PDO lock,
5151          * so to avoid deadlock, here we need to start journal handle before
5152          * related PDO lock also. But because we do not know whether there
5153          * will be something to be repaired before PDO lock, we just start
5154          * journal without conditions.
5155          *
5156          * We may need to remove the name entry firstly, then insert back.
5157          * One credit is for user quota file update.
5158          * One credit is for group quota file update.
5159          * Two credits are for dirty inode. */
5160         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5161                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5162
5163 again:
5164         if (dev->od_dirent_journal) {
5165                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5166                 if (IS_ERR(jh)) {
5167                         rc = PTR_ERR(jh);
5168                         CERROR("%.16s: fail to start trans for dirent "
5169                                "check_repair: credits %d, name %.*s, rc %d\n",
5170                                devname, credits, ent->oied_namelen,
5171                                ent->oied_name, rc);
5172                         RETURN(rc);
5173                 }
5174
5175                 if (obj->oo_hl_head != NULL) {
5176                         hlock = osd_oti_get(env)->oti_hlock;
5177                         /* "0" means exclusive lock for the whole directory.
5178                          * We need to prevent others access such name entry
5179                          * during the delete + insert. Neither HLOCK_ADD nor
5180                          * HLOCK_DEL cannot guarantee the atomicity. */
5181                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5182                 } else {
5183                         down_write(&obj->oo_ext_idx_sem);
5184                 }
5185         } else {
5186                 if (obj->oo_hl_head != NULL) {
5187                         hlock = osd_oti_get(env)->oti_hlock;
5188                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5189                                            LDISKFS_HLOCK_LOOKUP);
5190                 } else {
5191                         down_read(&obj->oo_ext_idx_sem);
5192                 }
5193         }
5194
5195         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5196         /* For dot/dotdot entry, if there is not enough space to hold the
5197          * FID-in-dirent, just keep them there. It only happens when the
5198          * device upgraded from 1.8 or restored from MDT file-level backup.
5199          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5200          * and needs to get FID from LMA when readdir, it will not affect the
5201          * performance much. */
5202         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5203             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5204                 *attr |= LUDA_IGNORE;
5205                 GOTO(out_journal, rc = 0);
5206         }
5207
5208         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5209         inode = osd_iget(info, dev, id);
5210         if (IS_ERR(inode)) {
5211                 rc = PTR_ERR(inode);
5212                 if (rc == -ENOENT || rc == -ESTALE) {
5213                         *attr |= LUDA_IGNORE;
5214                         rc = 0;
5215                 }
5216
5217                 GOTO(out_journal, rc);
5218         }
5219
5220         /* skip the REMOTE_PARENT_DIR. */
5221         if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
5222                 GOTO(out_inode, rc = 0);
5223
5224         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5225         if (rc == 0) {
5226                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5227
5228                 if (fid_is_sane(fid)) {
5229                         /* FID-in-dirent is valid. */
5230                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5231                                 GOTO(out_inode, rc = 0);
5232
5233                         /* Do not repair under dryrun mode. */
5234                         if (*attr & LUDA_VERIFY_DRYRUN) {
5235                                 *attr |= LUDA_REPAIR;
5236                                 GOTO(out_inode, rc = 0);
5237                         }
5238
5239                         if (!dev->od_dirent_journal) {
5240                                 iput(inode);
5241                                 brelse(bh);
5242                                 if (hlock != NULL)
5243                                         ldiskfs_htree_unlock(hlock);
5244                                 else
5245                                         up_read(&obj->oo_ext_idx_sem);
5246                                 dev->od_dirent_journal = 1;
5247                                 goto again;
5248                         }
5249
5250                         *fid = lma->lma_self_fid;
5251                         dirty = true;
5252                         /* Update the FID-in-dirent. */
5253                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5254                         if (rc == 0)
5255                                 *attr |= LUDA_REPAIR;
5256                 } else {
5257                         /* Do not repair under dryrun mode. */
5258                         if (*attr & LUDA_VERIFY_DRYRUN) {
5259                                 *fid = lma->lma_self_fid;
5260                                 *attr |= LUDA_REPAIR;
5261                                 GOTO(out_inode, rc = 0);
5262                         }
5263
5264                         if (!dev->od_dirent_journal) {
5265                                 iput(inode);
5266                                 brelse(bh);
5267                                 if (hlock != NULL)
5268                                         ldiskfs_htree_unlock(hlock);
5269                                 else
5270                                         up_read(&obj->oo_ext_idx_sem);
5271                                 dev->od_dirent_journal = 1;
5272                                 goto again;
5273                         }
5274
5275                         *fid = lma->lma_self_fid;
5276                         dirty = true;
5277                         /* Append the FID-in-dirent. */
5278                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5279                                                  fid, bh, de, hlock);
5280                         if (rc == 0)
5281                                 *attr |= LUDA_REPAIR;
5282                 }
5283         } else if (rc == -ENODATA) {
5284                 /* Do not repair under dryrun mode. */
5285                 if (*attr & LUDA_VERIFY_DRYRUN) {
5286                         if (fid_is_sane(fid)) {
5287                                 *attr |= LUDA_REPAIR;
5288                         } else {
5289                                 lu_igif_build(fid, inode->i_ino,
5290                                               inode->i_generation);
5291                                 *attr |= LUDA_UPGRADE;
5292                         }
5293                         GOTO(out_inode, rc = 0);
5294                 }
5295
5296                 if (!dev->od_dirent_journal) {
5297                         iput(inode);
5298                         brelse(bh);
5299                         if (hlock != NULL)
5300                                 ldiskfs_htree_unlock(hlock);
5301                         else
5302                                 up_read(&obj->oo_ext_idx_sem);
5303                         dev->od_dirent_journal = 1;
5304                         goto again;
5305                 }
5306
5307                 dirty = true;
5308                 if (unlikely(fid_is_sane(fid))) {
5309                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5310                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5311                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5312                         if (rc == 0)
5313                                 *attr |= LUDA_REPAIR;
5314                 } else {
5315                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5316                         /* It is probably IGIF object. Only aappend the
5317                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5318                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5319                                                  fid, bh, de, hlock);
5320                         if (rc == 0)
5321                                 *attr |= LUDA_UPGRADE;
5322                 }
5323         }
5324
5325         GOTO(out_inode, rc);
5326
5327 out_inode:
5328         iput(inode);
5329
5330 out_journal:
5331         brelse(bh);
5332         if (hlock != NULL) {
5333                 ldiskfs_htree_unlock(hlock);
5334         } else {
5335                 if (dev->od_dirent_journal)
5336                         up_write(&obj->oo_ext_idx_sem);
5337                 else
5338                         up_read(&obj->oo_ext_idx_sem);
5339         }
5340         if (jh != NULL)
5341                 ldiskfs_journal_stop(jh);
5342         if (rc >= 0 && !dirty)
5343                 dev->od_dirent_journal = 0;
5344         return rc;
5345 }
5346
5347 /**
5348  * Returns the value at current position from iterator's in memory structure.
5349  *
5350  * \param di struct osd_it_ea, iterator's in memory structure
5351  * \param attr attr requested for dirent.
5352  * \param lde lustre dirent
5353  *
5354  * \retval   0 no error and \param lde has correct lustre dirent.
5355  * \retval -ve on error
5356  */
5357 static inline int osd_it_ea_rec(const struct lu_env *env,
5358                                 const struct dt_it *di,
5359                                 struct dt_rec *dtrec, __u32 attr)
5360 {
5361         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5362         struct osd_object      *obj   = it->oie_obj;
5363         struct osd_device      *dev   = osd_obj2dev(obj);
5364         struct osd_scrub       *scrub = &dev->od_scrub;
5365         struct scrub_file      *sf    = &scrub->os_file;
5366         struct osd_thread_info *oti   = osd_oti_get(env);
5367         struct osd_inode_id    *id    = &oti->oti_id;
5368         struct osd_idmap_cache *oic   = &oti->oti_cache;
5369         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5370         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5371         __u32                   ino   = it->oie_dirent->oied_ino;
5372         int                     rc    = 0;
5373         ENTRY;
5374
5375         if (attr & LUDA_VERIFY) {
5376                 attr |= LUDA_TYPE;
5377                 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5378                         attr |= LUDA_IGNORE;
5379                         rc = 0;
5380                 } else {
5381                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5382                                                      &attr);
5383                 }
5384         } else {
5385                 attr &= ~LU_DIRENT_ATTRS_MASK;
5386                 if (!fid_is_sane(fid)) {
5387                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5388                             likely(it->oie_dirent->oied_namelen != 2 ||
5389                                    it->oie_dirent->oied_name[0] != '.' ||
5390                                    it->oie_dirent->oied_name[1] != '.'))
5391                                 RETURN(-ENOENT);
5392
5393                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
5394                 } else {
5395                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5396                 }
5397         }
5398
5399         /* Pack the entry anyway, at least the offset is right. */
5400         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5401                            it->oie_dirent->oied_name,
5402                            it->oie_dirent->oied_namelen,
5403                            it->oie_dirent->oied_type, attr);
5404
5405         if (rc < 0)
5406                 RETURN(rc);
5407
5408         if (osd_remote_fid(env, dev, fid))
5409                 RETURN(0);
5410
5411         if (likely(!(attr & LUDA_IGNORE)))
5412                 rc = osd_add_oi_cache(oti, dev, id, fid);
5413
5414         if (!(attr & LUDA_VERIFY) &&
5415             (scrub->os_pos_current <= ino) &&
5416             ((sf->sf_flags & SF_INCONSISTENT) ||
5417              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5418              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5419                 osd_consistency_check(oti, dev, oic);
5420
5421         RETURN(rc);
5422 }
5423
5424 /**
5425  * Returns the record size size at current position.
5426  *
5427  * This function will return record(lu_dirent) size in bytes.
5428  *
5429  * \param[in] env       execution environment
5430  * \param[in] di        iterator's in memory structure
5431  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5432  *                      calculate the lu_dirent size.
5433  *
5434  * \retval      record size(in bytes & in memory) of the current lu_dirent
5435  *              entry.
5436  */
5437 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5438                               __u32 attr)
5439 {
5440         struct osd_it_ea *it = (struct osd_it_ea *)di;
5441
5442         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5443 }
5444
5445 /**
5446  * Returns a cookie for current position of the iterator head, so that
5447  * user can use this cookie to load/start the iterator next time.
5448  *
5449  * \param di iterator's in memory structure
5450  *
5451  * \retval cookie for current position, on success
5452  */
5453 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5454 {
5455         struct osd_it_ea *it = (struct osd_it_ea *)di;
5456
5457         return it->oie_dirent->oied_off;
5458 }
5459
5460 /**
5461  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5462  * to load a directory entry at a time and stored it i inn,
5463  * in iterator's in-memory data structure.
5464  *
5465  * \param di struct osd_it_ea, iterator's in memory structure
5466  *
5467  * \retval +ve on success
5468  * \retval -ve on error
5469  */
5470 static int osd_it_ea_load(const struct lu_env *env,
5471                           const struct dt_it *di, __u64 hash)
5472 {
5473         struct osd_it_ea *it = (struct osd_it_ea *)di;
5474         int rc;
5475
5476         ENTRY;
5477         it->oie_file.f_pos = hash;
5478
5479         rc =  osd_ldiskfs_it_fill(env, di);
5480         if (rc > 0)
5481                 rc = -ENODATA;
5482
5483         if (rc == 0)
5484                 rc = +1;
5485
5486         RETURN(rc);
5487 }
5488
5489 /**
5490  * Index lookup function for interoperability mode (b11826).
5491  *
5492  * \param key,  key i.e. file name to be searched
5493  *
5494  * \retval +ve, on success
5495  * \retval -ve, on error
5496  */
5497 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5498                                struct dt_rec *rec, const struct dt_key *key,
5499                                struct lustre_capa *capa)
5500 {
5501         struct osd_object *obj = osd_dt_obj(dt);
5502         int rc = 0;
5503
5504         ENTRY;
5505
5506         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5507         LINVRNT(osd_invariant(obj));
5508
5509         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5510                 return -EACCES;
5511
5512         rc = osd_ea_lookup_rec(env, obj, rec, key);
5513         if (rc == 0)
5514                 rc = +1;
5515         RETURN(rc);
5516 }
5517
5518 /**
5519  * Index and Iterator operations for interoperability
5520  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5521  */
5522 static const struct dt_index_operations osd_index_ea_ops = {
5523         .dio_lookup         = osd_index_ea_lookup,
5524         .dio_declare_insert = osd_index_declare_ea_insert,
5525         .dio_insert         = osd_index_ea_insert,
5526         .dio_declare_delete = osd_index_declare_ea_delete,
5527         .dio_delete         = osd_index_ea_delete,
5528         .dio_it     = {
5529                 .init     = osd_it_ea_init,
5530                 .fini     = osd_it_ea_fini,
5531                 .get      = osd_it_ea_get,
5532                 .put      = osd_it_ea_put,
5533                 .next     = osd_it_ea_next,
5534                 .key      = osd_it_ea_key,
5535                 .key_size = osd_it_ea_key_size,
5536                 .rec      = osd_it_ea_rec,
5537                 .rec_size = osd_it_ea_rec_size,
5538                 .store    = osd_it_ea_store,
5539                 .load     = osd_it_ea_load
5540         }
5541 };
5542
5543 static void *osd_key_init(const struct lu_context *ctx,
5544                           struct lu_context_key *key)
5545 {
5546         struct osd_thread_info *info;
5547
5548         OBD_ALLOC_PTR(info);
5549         if (info == NULL)
5550                 return ERR_PTR(-ENOMEM);
5551
5552         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5553         if (info->oti_it_ea_buf == NULL)
5554                 goto out_free_info;
5555
5556         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5557
5558         info->oti_hlock = ldiskfs_htree_lock_alloc();
5559         if (info->oti_hlock == NULL)
5560                 goto out_free_ea;
5561
5562         return info;
5563
5564  out_free_ea:
5565         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5566  out_free_info:
5567         OBD_FREE_PTR(info);
5568         return ERR_PTR(-ENOMEM);
5569 }
5570
5571 static void osd_key_fini(const struct lu_context *ctx,
5572                          struct lu_context_key *key, void* data)
5573 {
5574         struct osd_thread_info *info = data;
5575
5576         if (info->oti_hlock != NULL)
5577                 ldiskfs_htree_lock_free(info->oti_hlock);
5578         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5579         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5580         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5581         OBD_FREE_PTR(info);
5582 }
5583
5584 static void osd_key_exit(const struct lu_context *ctx,
5585                          struct lu_context_key *key, void *data)
5586 {
5587         struct osd_thread_info *info = data;
5588
5589         LASSERT(info->oti_r_locks == 0);
5590         LASSERT(info->oti_w_locks == 0);
5591         LASSERT(info->oti_txns    == 0);
5592 }
5593
5594 /* type constructor/destructor: osd_type_init, osd_type_fini */
5595 LU_TYPE_INIT_FINI(osd, &osd_key);
5596
5597 struct lu_context_key osd_key = {
5598         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5599         .lct_init = osd_key_init,
5600         .lct_fini = osd_key_fini,
5601         .lct_exit = osd_key_exit
5602 };
5603
5604
5605 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5606                            const char *name, struct lu_device *next)
5607 {
5608         struct osd_device *osd = osd_dev(d);
5609
5610         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5611             >= sizeof(osd->od_svname))
5612                 return -E2BIG;
5613         return osd_procfs_init(osd, name);
5614 }
5615
5616 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5617 {
5618         struct seq_server_site  *ss = osd_seq_site(osd);
5619         int                     rc;
5620         ENTRY;
5621
5622         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5623                 RETURN(0);
5624
5625         if (unlikely(ss == NULL))
5626                 RETURN(-ENODEV);
5627
5628         OBD_ALLOC_PTR(osd->od_cl_seq);
5629         if (osd->od_cl_seq == NULL)
5630                 RETURN(-ENOMEM);
5631
5632         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5633                              osd->od_svname, ss->ss_server_seq);
5634
5635         if (rc != 0) {
5636                 OBD_FREE_PTR(osd->od_cl_seq);
5637                 osd->od_cl_seq = NULL;
5638         }
5639
5640         RETURN(rc);
5641 }
5642
5643 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5644 {
5645         if (osd->od_cl_seq == NULL)
5646                 return;
5647
5648         seq_client_fini(osd->od_cl_seq);
5649         OBD_FREE_PTR(osd->od_cl_seq);
5650         osd->od_cl_seq = NULL;
5651 }
5652
5653 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5654 {
5655         ENTRY;
5656
5657         /* shutdown quota slave instance associated with the device */
5658         if (o->od_quota_slave != NULL) {
5659                 qsd_fini(env, o->od_quota_slave);
5660                 o->od_quota_slave = NULL;
5661         }
5662
5663         osd_fid_fini(env, o);
5664
5665         RETURN(0);
5666 }
5667
5668 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5669 {
5670         ENTRY;
5671
5672         if (o->od_mnt != NULL) {
5673                 shrink_dcache_sb(osd_sb(o));
5674                 osd_sync(env, &o->od_dt_dev);
5675
5676                 mntput(o->od_mnt);
5677                 o->od_mnt = NULL;
5678         }
5679
5680         EXIT;
5681 }
5682
5683 static int osd_mount(const struct lu_env *env,
5684                      struct osd_device *o, struct lustre_cfg *cfg)
5685 {
5686         const char              *name  = lustre_cfg_string(cfg, 0);
5687         const char              *dev  = lustre_cfg_string(cfg, 1);
5688         const char              *opts;
5689         unsigned long            page, s_flags, lmd_flags = 0;
5690         struct page             *__page;
5691         struct file_system_type *type;
5692         char                    *options = NULL;
5693         char                    *str;
5694         struct osd_thread_info  *info = osd_oti_get(env);
5695         struct lu_fid           *fid = &info->oti_fid;
5696         struct inode            *inode;
5697         int                      rc = 0;
5698         ENTRY;
5699
5700         if (o->od_mnt != NULL)
5701                 RETURN(0);
5702
5703         if (strlen(dev) >= sizeof(o->od_mntdev))
5704                 RETURN(-E2BIG);
5705         strcpy(o->od_mntdev, dev);
5706
5707         OBD_PAGE_ALLOC(__page, GFP_IOFS);
5708         if (__page == NULL)
5709                 GOTO(out, rc = -ENOMEM);
5710
5711         str = lustre_cfg_string(cfg, 2);
5712         s_flags = simple_strtoul(str, NULL, 0);
5713         str = strstr(str, ":");
5714         if (str)
5715                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5716         opts = lustre_cfg_string(cfg, 3);
5717         page = (unsigned long)page_address(__page);
5718         options = (char *)page;
5719         *options = '\0';
5720         if (opts == NULL)
5721                 strcat(options, "user_xattr,acl");
5722         else
5723                 strcat(options, opts);
5724
5725         /* Glom up mount options */
5726         if (*options != '\0')
5727                 strcat(options, ",");
5728         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5729
5730         type = get_fs_type("ldiskfs");
5731         if (!type) {
5732                 CERROR("%s: cannot find ldiskfs module\n", name);
5733                 GOTO(out, rc = -ENODEV);
5734         }
5735
5736         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5737         module_put(type->owner);
5738
5739         if (IS_ERR(o->od_mnt)) {
5740                 rc = PTR_ERR(o->od_mnt);
5741                 o->od_mnt = NULL;
5742                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5743                 GOTO(out, rc);
5744         }
5745
5746 #ifdef HAVE_DEV_SET_RDONLY
5747         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5748                 CERROR("%s: underlying device %s is marked as read-only. "
5749                        "Setup failed\n", name, dev);
5750                 GOTO(out_mnt, rc = -EROFS);
5751         }
5752 #endif
5753
5754         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5755                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5756                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5757                 GOTO(out_mnt, rc = -EINVAL);
5758         }
5759
5760 #ifdef LDISKFS_MOUNT_DIRDATA
5761         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5762                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5763                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5764         else if (!o->od_is_ost)
5765                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5766                       "enabling the dirdata feature. If you do not want to "
5767                       "downgrade to Lustre-1.x again, you can enable it via "
5768                       "'tune2fs -O dirdata device'\n", name, dev);
5769 #endif
5770         inode = osd_sb(o)->s_root->d_inode;
5771         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5772         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5773         if (rc != 0) {
5774                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5775                 GOTO(out_mnt, rc);
5776         }
5777
5778         if (lmd_flags & LMD_FLG_NOSCRUB)
5779                 o->od_noscrub = 1;
5780
5781         GOTO(out, rc = 0);
5782
5783 out_mnt:
5784         mntput(o->od_mnt);
5785         o->od_mnt = NULL;
5786
5787 out:
5788         if (__page)
5789                 OBD_PAGE_FREE(__page);
5790
5791         return rc;
5792 }
5793
5794 static struct lu_device *osd_device_fini(const struct lu_env *env,
5795                                          struct lu_device *d)
5796 {
5797         struct osd_device *o = osd_dev(d);
5798         ENTRY;
5799
5800         osd_shutdown(env, o);
5801         osd_procfs_fini(o);
5802         osd_scrub_cleanup(env, o);
5803         osd_obj_map_fini(o);
5804         osd_umount(env, o);
5805
5806         RETURN(NULL);
5807 }
5808
5809 static int osd_device_init0(const struct lu_env *env,
5810                             struct osd_device *o,
5811                             struct lustre_cfg *cfg)
5812 {
5813         struct lu_device        *l = osd2lu_dev(o);
5814         struct osd_thread_info *info;
5815         int                     rc;
5816         int                     cplen = 0;
5817
5818         /* if the module was re-loaded, env can loose its keys */
5819         rc = lu_env_refill((struct lu_env *) env);
5820         if (rc)
5821                 GOTO(out, rc);
5822         info = osd_oti_get(env);
5823         LASSERT(info);
5824
5825         l->ld_ops = &osd_lu_ops;
5826         o->od_dt_dev.dd_ops = &osd_dt_ops;
5827
5828         spin_lock_init(&o->od_osfs_lock);
5829         mutex_init(&o->od_otable_mutex);
5830
5831         o->od_capa_hash = init_capa_hash();
5832         if (o->od_capa_hash == NULL)
5833                 GOTO(out, rc = -ENOMEM);
5834
5835         o->od_read_cache = 1;
5836         o->od_writethrough_cache = 1;
5837         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5838
5839         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5840                         sizeof(o->od_svname));
5841         if (cplen >= sizeof(o->od_svname)) {
5842                 rc = -E2BIG;
5843                 GOTO(out_capa, rc);
5844         }
5845
5846         if (server_name_is_ost(o->od_svname))
5847                 o->od_is_ost = 1;
5848
5849         rc = osd_mount(env, o, cfg);
5850         if (rc != 0)
5851                 GOTO(out_capa, rc);
5852
5853         rc = osd_obj_map_init(env, o);
5854         if (rc != 0)
5855                 GOTO(out_mnt, rc);
5856
5857         rc = lu_site_init(&o->od_site, l);
5858         if (rc != 0)
5859                 GOTO(out_compat, rc);
5860         o->od_site.ls_bottom_dev = l;
5861
5862         rc = lu_site_init_finish(&o->od_site);
5863         if (rc != 0)
5864                 GOTO(out_site, rc);
5865
5866         /* self-repair LMA by default */
5867         o->od_lma_self_repair = 1;
5868
5869         INIT_LIST_HEAD(&o->od_ios_list);
5870         /* setup scrub, including OI files initialization */
5871         rc = osd_scrub_setup(env, o);
5872         if (rc < 0)
5873                 GOTO(out_site, rc);
5874
5875         rc = osd_procfs_init(o, o->od_svname);
5876         if (rc != 0) {
5877                 CERROR("%s: can't initialize procfs: rc = %d\n",
5878                        o->od_svname, rc);
5879                 GOTO(out_scrub, rc);
5880         }
5881
5882         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5883
5884         /* initialize quota slave instance */
5885         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5886                                      o->od_proc_entry);
5887         if (IS_ERR(o->od_quota_slave)) {
5888                 rc = PTR_ERR(o->od_quota_slave);
5889                 o->od_quota_slave = NULL;
5890                 GOTO(out_procfs, rc);
5891         }
5892
5893         RETURN(0);
5894
5895 out_procfs:
5896         osd_procfs_fini(o);
5897 out_scrub:
5898         osd_scrub_cleanup(env, o);
5899 out_site:
5900         lu_site_fini(&o->od_site);
5901 out_compat:
5902         osd_obj_map_fini(o);
5903 out_mnt:
5904         osd_umount(env, o);
5905 out_capa:
5906         cleanup_capa_hash(o->od_capa_hash);
5907 out:
5908         return rc;
5909 }
5910
5911 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5912                                           struct lu_device_type *t,
5913                                           struct lustre_cfg *cfg)
5914 {
5915         struct osd_device *o;
5916         int                rc;
5917
5918         OBD_ALLOC_PTR(o);
5919         if (o == NULL)
5920                 return ERR_PTR(-ENOMEM);
5921
5922         rc = dt_device_init(&o->od_dt_dev, t);
5923         if (rc == 0) {
5924                 /* Because the ctx might be revived in dt_device_init,
5925                  * refill the env here */
5926                 lu_env_refill((struct lu_env *)env);
5927                 rc = osd_device_init0(env, o, cfg);
5928                 if (rc)
5929                         dt_device_fini(&o->od_dt_dev);
5930         }
5931
5932         if (unlikely(rc != 0))
5933                 OBD_FREE_PTR(o);
5934
5935         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5936 }
5937
5938 static struct lu_device *osd_device_free(const struct lu_env *env,
5939                                          struct lu_device *d)
5940 {
5941         struct osd_device *o = osd_dev(d);
5942         ENTRY;
5943
5944         cleanup_capa_hash(o->od_capa_hash);
5945         /* XXX: make osd top device in order to release reference */
5946         d->ld_site->ls_top_dev = d;
5947         lu_site_purge(env, d->ld_site, -1);
5948         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5949                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5950                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5951         }
5952         lu_site_fini(&o->od_site);
5953         dt_device_fini(&o->od_dt_dev);
5954         OBD_FREE_PTR(o);
5955         RETURN(NULL);
5956 }
5957
5958 static int osd_process_config(const struct lu_env *env,
5959                               struct lu_device *d, struct lustre_cfg *cfg)
5960 {
5961         struct osd_device               *o = osd_dev(d);
5962         int                             rc;
5963         ENTRY;
5964
5965         switch (cfg->lcfg_command) {
5966         case LCFG_SETUP:
5967                 rc = osd_mount(env, o, cfg);
5968                 break;
5969         case LCFG_CLEANUP:
5970                 lu_dev_del_linkage(d->ld_site, d);
5971                 rc = osd_shutdown(env, o);
5972                 break;
5973         case LCFG_PARAM:
5974                 LASSERT(&o->od_dt_dev);
5975                 rc = class_process_proc_seq_param(PARAM_OSD,
5976                                                   lprocfs_osd_obd_vars,
5977                                                   cfg, &o->od_dt_dev);
5978                 if (rc > 0 || rc == -ENOSYS)
5979                         rc = class_process_proc_seq_param(PARAM_OST,
5980                                                           lprocfs_osd_obd_vars,
5981                                                           cfg, &o->od_dt_dev);
5982                 break;
5983         default:
5984                 rc = -ENOSYS;
5985         }
5986
5987         RETURN(rc);
5988 }
5989
5990 static int osd_recovery_complete(const struct lu_env *env,
5991                                  struct lu_device *d)
5992 {
5993         struct osd_device       *osd = osd_dev(d);
5994         int                      rc = 0;
5995         ENTRY;
5996
5997         if (osd->od_quota_slave == NULL)
5998                 RETURN(0);
5999
6000         /* start qsd instance on recovery completion, this notifies the quota
6001          * slave code that we are about to process new requests now */
6002         rc = qsd_start(env, osd->od_quota_slave);
6003         RETURN(rc);
6004 }
6005
6006 /*
6007  * we use exports to track all osd users
6008  */
6009 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6010                            struct obd_device *obd, struct obd_uuid *cluuid,
6011                            struct obd_connect_data *data, void *localdata)
6012 {
6013         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6014         struct lustre_handle  conn;
6015         int                   rc;
6016         ENTRY;
6017
6018         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6019
6020         rc = class_connect(&conn, obd, cluuid);
6021         if (rc)
6022                 RETURN(rc);
6023
6024         *exp = class_conn2export(&conn);
6025
6026         spin_lock(&osd->od_osfs_lock);
6027         osd->od_connects++;
6028         spin_unlock(&osd->od_osfs_lock);
6029
6030         RETURN(0);
6031 }
6032
6033 /*
6034  * once last export (we don't count self-export) disappeared
6035  * osd can be released
6036  */
6037 static int osd_obd_disconnect(struct obd_export *exp)
6038 {
6039         struct obd_device *obd = exp->exp_obd;
6040         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6041         int                rc, release = 0;
6042         ENTRY;
6043
6044         /* Only disconnect the underlying layers on the final disconnect. */
6045         spin_lock(&osd->od_osfs_lock);
6046         osd->od_connects--;
6047         if (osd->od_connects == 0)
6048                 release = 1;
6049         spin_unlock(&osd->od_osfs_lock);
6050
6051         rc = class_disconnect(exp); /* bz 9811 */
6052
6053         if (rc == 0 && release)
6054                 class_manual_cleanup(obd);
6055         RETURN(rc);
6056 }
6057
6058 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6059                        struct lu_device *dev)
6060 {
6061         struct osd_device *osd = osd_dev(dev);
6062         int                result = 0;
6063         ENTRY;
6064
6065         if (osd->od_quota_slave != NULL) {
6066                 /* set up quota slave objects */
6067                 result = qsd_prepare(env, osd->od_quota_slave);
6068                 if (result != 0)
6069                         RETURN(result);
6070         }
6071
6072         result = osd_fid_init(env, osd);
6073
6074         RETURN(result);
6075 }
6076
6077 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6078                   struct lu_fid *fid, struct md_op_data *op_data)
6079 {
6080         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6081
6082         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6083 }
6084
6085 static const struct lu_object_operations osd_lu_obj_ops = {
6086         .loo_object_init      = osd_object_init,
6087         .loo_object_delete    = osd_object_delete,
6088         .loo_object_release   = osd_object_release,
6089         .loo_object_free      = osd_object_free,
6090         .loo_object_print     = osd_object_print,
6091         .loo_object_invariant = osd_object_invariant
6092 };
6093
6094 const struct lu_device_operations osd_lu_ops = {
6095         .ldo_object_alloc      = osd_object_alloc,
6096         .ldo_process_config    = osd_process_config,
6097         .ldo_recovery_complete = osd_recovery_complete,
6098         .ldo_prepare           = osd_prepare,
6099 };
6100
6101 static const struct lu_device_type_operations osd_device_type_ops = {
6102         .ldto_init = osd_type_init,
6103         .ldto_fini = osd_type_fini,
6104
6105         .ldto_start = osd_type_start,
6106         .ldto_stop  = osd_type_stop,
6107
6108         .ldto_device_alloc = osd_device_alloc,
6109         .ldto_device_free  = osd_device_free,
6110
6111         .ldto_device_init    = osd_device_init,
6112         .ldto_device_fini    = osd_device_fini
6113 };
6114
6115 struct lu_device_type osd_device_type = {
6116         .ldt_tags     = LU_DEVICE_DT,
6117         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6118         .ldt_ops      = &osd_device_type_ops,
6119         .ldt_ctx_tags = LCT_LOCAL,
6120 };
6121
6122 /*
6123  * lprocfs legacy support.
6124  */
6125 static struct obd_ops osd_obd_device_ops = {
6126         .o_owner = THIS_MODULE,
6127         .o_connect      = osd_obd_connect,
6128         .o_disconnect   = osd_obd_disconnect,
6129         .o_fid_alloc    = osd_fid_alloc,
6130 };
6131
6132 static int __init osd_mod_init(void)
6133 {
6134         int rc;
6135
6136         osd_oi_mod_init();
6137
6138         rc = lu_kmem_init(ldiskfs_caches);
6139         if (rc)
6140                 return rc;
6141
6142         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6143                                  lprocfs_osd_module_vars,
6144 #ifndef HAVE_ONLY_PROCFS_SEQ
6145                                  NULL,
6146 #endif
6147                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6148         if (rc)
6149                 lu_kmem_fini(ldiskfs_caches);
6150         return rc;
6151 }
6152
6153 static void __exit osd_mod_exit(void)
6154 {
6155         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6156         lu_kmem_fini(ldiskfs_caches);
6157 }
6158
6159 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6160 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6161 MODULE_LICENSE("GPL");
6162
6163 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);