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, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/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_OSD
  45
  46 #include <linux/module.h>
  47 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51
  52 /* LUSTRE_VERSION_CODE */
  53 #include <lustre_ver.h>
  54 /* prerequisite for linux/xattr.h */
  55 #include <linux/types.h>
  56 /* prerequisite for linux/xattr.h */
  57 #include <linux/fs.h>
  58 /* XATTR_{REPLACE,CREATE} */
  59 #include <linux/xattr.h>
  60
  61 #include <ldiskfs/ldiskfs.h>
  62 #include <ldiskfs/xattr.h>
  63 #undef ENTRY
  64 /*
  65  * struct OBD_{ALLOC,FREE}*()
  66  * OBD_FAIL_CHECK
  67  */
  68 #include <obd_support.h>
  69 /* struct ptlrpc_thread */
  70 #include <lustre_net.h>
  71 #include <lustre_fid.h>
  72 /* process_config */
  73 #include <lustre_param.h>
  74
  75 #include "osd_internal.h"
  76 #include "osd_dynlocks.h"
  77
  78 /* llo_* api support */
  79 #include <md_object.h>
  80 #include <lustre_quota.h>
  81
  82 #include <lustre_linkea.h>
  83
  84 int ldiskfs_pdo = 1;
  85 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  86                 "ldiskfs with parallel directory operations");
  87
  88 int ldiskfs_track_declares_assert;
  89 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  90                 "LBUG during tracking of declares");
  91
  92 /* Slab to allocate dynlocks */
  93 struct kmem_cache *dynlock_cachep;
  94
  95 /* Slab to allocate osd_it_ea */
  96 struct kmem_cache *osd_itea_cachep;
  97
  98 static struct lu_kmem_descr ldiskfs_caches[] = {
  99         {
 100                 .ckd_cache = &dynlock_cachep,
 101                 .ckd_name  = "dynlock_cache",
 102                 .ckd_size  = sizeof(struct dynlock_handle)
 103         },
 104         {
 105                 .ckd_cache = &osd_itea_cachep,
 106                 .ckd_name  = "osd_itea_cache",
 107                 .ckd_size  = sizeof(struct osd_it_ea)
 108         },
 109         {
 110                 .ckd_cache = NULL
 111         }
 112 };
 113
 114 static const char dot[] = ".";
 115 static const char dotdot[] = "..";
 116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 117
 118 static const struct lu_object_operations      osd_lu_obj_ops;
 119 static const struct dt_object_operations      osd_obj_ops;
 120 static const struct dt_object_operations      osd_obj_ea_ops;
 121 static const struct dt_object_operations      osd_obj_otable_it_ops;
 122 static const struct dt_index_operations       osd_index_iam_ops;
 123 static const struct dt_index_operations       osd_index_ea_ops;
 124
 125 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 126                           const struct lu_fid *fid);
 127 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 128                                                 struct osd_device *osd);
 129
 130 int osd_trans_declare_op2rb[] = {
 131         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 132         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 133         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 134         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 135         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 136         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 137         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 138         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 139         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 140         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 141         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 142 };
 143
 144 static int osd_has_index(const struct osd_object *obj)
 145 {
 146         return obj->oo_dt.do_index_ops != NULL;
 147 }
 148
 149 static int osd_object_invariant(const struct lu_object *l)
 150 {
 151         return osd_invariant(osd_obj(l));
 152 }
 153
 154 /*
 155  * Concurrency: doesn't matter
 156  */
 157
 158 /*
 159  * Concurrency: doesn't matter
 160  */
 161 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 162 {
 163         struct osd_thread_info *oti = osd_oti_get(env);
 164         return oti->oti_w_locks > 0 && o->oo_owner == env;
 165 }
 166
 167 /*
 168  * Concurrency: doesn't access mutable data
 169  */
 170 static int osd_root_get(const struct lu_env *env,
 171                         struct dt_device *dev, struct lu_fid *f)
 172 {
 173         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 174         return 0;
 175 }
 176
 177 /*
 178  * OSD object methods.
 179  */
 180
 181 /*
 182  * Concurrency: no concurrent access is possible that early in object
 183  * life-cycle.
 184  */
 185 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 186                                           const struct lu_object_header *hdr,
 187                                           struct lu_device *d)
 188 {
 189         struct osd_object *mo;
 190
 191         OBD_ALLOC_PTR(mo);
 192         if (mo != NULL) {
 193                 struct lu_object *l;
 194
 195                 l = &mo->oo_dt.do_lu;
 196                 dt_object_init(&mo->oo_dt, NULL, d);
 197                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 198                 l->lo_ops = &osd_lu_obj_ops;
 199                 init_rwsem(&mo->oo_sem);
 200                 init_rwsem(&mo->oo_ext_idx_sem);
 201                 spin_lock_init(&mo->oo_guard);
 202                 return l;
 203         } else {
 204                 return NULL;
 205         }
 206 }
 207
 208 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 209                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 210 {
 211         int rc;
 212
 213         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 214         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 215                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 216         if (rc > 0) {
 217                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 218                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 219                 rc = 0;
 220                 lustre_lma_swab(lma);
 221                 /* Check LMA compatibility */
 222                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 223                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 224                               "for fid = "DFID", ino = %lu\n",
 225                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 226                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 227                               PFID(&lma->lma_self_fid), inode->i_ino);
 228                         rc = -EOPNOTSUPP;
 229                 }
 230         } else if (rc == 0) {
 231                 rc = -ENODATA;
 232         }
 233
 234         return rc;
 235 }
 236
 237 /*
 238  * retrieve object from backend ext fs.
 239  **/
 240 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 241                        struct osd_inode_id *id)
 242 {
 243         struct inode *inode = NULL;
 244
 245         /* if we look for an inode withing a running
 246          * transaction, then we risk to deadlock */
 247         /* osd_dirent_check_repair() breaks this */
 248         /*LASSERT(current->journal_info == NULL);*/
 249
 250         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 251         if (IS_ERR(inode)) {
 252                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 253                        id->oii_ino, PTR_ERR(inode));
 254         } else if (id->oii_gen != OSD_OII_NOGEN &&
 255                    inode->i_generation != id->oii_gen) {
 256                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 257                        "i_generation = %u\n",
 258                        id->oii_ino, id->oii_gen, inode->i_generation);
 259                 iput(inode);
 260                 inode = ERR_PTR(-ESTALE);
 261         } else if (inode->i_nlink == 0) {
 262                 /* due to parallel readdir and unlink,
 263                 * we can have dead inode here. */
 264                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 265                 iput(inode);
 266                 inode = ERR_PTR(-ESTALE);
 267         } else if (is_bad_inode(inode)) {
 268                 CWARN("%.16s: bad inode: ino = %u\n",
 269                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 270                 iput(inode);
 271                 inode = ERR_PTR(-ENOENT);
 272         } else {
 273                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 274                 if (id->oii_gen == OSD_OII_NOGEN)
 275                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 276
 277                 /* Do not update file c/mtime in ldiskfs.
 278                  * NB: we don't have any lock to protect this because we don't
 279                  * have reference on osd_object now, but contention with
 280                  * another lookup + attr_set can't happen in the tiny window
 281                  * between if (...) and set S_NOCMTIME. */
 282                 if (!(inode->i_flags & S_NOCMTIME))
 283                         inode->i_flags |= S_NOCMTIME;
 284         }
 285         return inode;
 286 }
 287
 288 static struct inode *
 289 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 290              struct osd_inode_id *id, struct lu_fid *fid)
 291 {
 292         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 293         struct inode            *inode;
 294         int                      rc;
 295
 296         inode = osd_iget(info, dev, id);
 297         if (IS_ERR(inode))
 298                 return inode;
 299
 300         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 301         if (rc == 0) {
 302                 *fid = lma->lma_self_fid;
 303         } else if (rc == -ENODATA) {
 304                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 305                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 306                 else
 307                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 308         } else {
 309                 iput(inode);
 310                 inode = ERR_PTR(rc);
 311         }
 312         return inode;
 313 }
 314
 315 static struct inode *osd_iget_check(struct osd_thread_info *info,
 316                                     struct osd_device *dev,
 317                                     const struct lu_fid *fid,
 318                                     struct osd_inode_id *id,
 319                                     bool cached)
 320 {
 321         struct inode    *inode;
 322         int              rc     = 0;
 323         ENTRY;
 324
 325         /* The cached OI mapping is trustable. If we cannot locate the inode
 326          * via the cached OI mapping, then return the failure to the caller
 327          * directly without further OI checking. */
 328
 329         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 330         if (IS_ERR(inode)) {
 331                 rc = PTR_ERR(inode);
 332                 if (cached || (rc != -ENOENT && rc != -ESTALE)) {
 333                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
 334                                id->oii_ino, rc);
 335
 336                         GOTO(put, rc);
 337                 }
 338
 339                 goto check_oi;
 340         }
 341
 342         if (is_bad_inode(inode)) {
 343                 rc = -ENOENT;
 344                 if (cached) {
 345                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
 346
 347                         GOTO(put, rc);
 348                 }
 349
 350                 goto check_oi;
 351         }
 352
 353         if (id->oii_gen != OSD_OII_NOGEN &&
 354             inode->i_generation != id->oii_gen) {
 355                 rc = -ESTALE;
 356                 if (cached) {
 357                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
 358                                "oii_gen = %u, i_generation = %u\n",
 359                                id->oii_ino, id->oii_gen, inode->i_generation);
 360
 361                         GOTO(put, rc);
 362                 }
 363
 364                 goto check_oi;
 365         }
 366
 367         if (inode->i_nlink == 0) {
 368                 rc = -ENOENT;
 369                 if (cached) {
 370                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 371
 372                         GOTO(put, rc);
 373                 }
 374
 375                 goto check_oi;
 376         }
 377
 378         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 379
 380 check_oi:
 381         if (rc != 0) {
 382                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 383
 384                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 385                 /* XXX: There are four possible cases:
 386                  *      1. rc = 0.
 387                  *         Backup/restore caused the OI invalid.
 388                  *      2. rc = 0.
 389                  *         Someone unlinked the object but NOT removed
 390                  *         the OI mapping, such as mount target device
 391                  *         as ldiskfs, and modify something directly.
 392                  *      3. rc = -ENOENT.
 393                  *         Someone just removed the object between the
 394                  *         former oi_lookup and the iget. It is normal.
 395                  *      4. Other failure cases.
 396                  *
 397                  *      Generally, when the device is mounted, it will
 398                  *      auto check whether the system is restored from
 399                  *      file-level backup or not. We trust such detect
 400                  *      to distinguish the 1st case from the 2nd case. */
 401                 if (rc == 0) {
 402                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
 403                             inode->i_generation == id->oii_gen)
 404                                 /* "id->oii_gen != OSD_OII_NOGEN" is for
 405                                  * "@cached == false" case. */
 406                                 rc = -ENOENT;
 407                         else
 408                                 rc = -EREMCHG;
 409                 } else {
 410                         /* If the OI mapping was in OI file before the
 411                          * osd_iget_check(), but now, it is disappear,
 412                          * then it must be removed by race. That is a
 413                          * normal race case. */
 414                 }
 415         } else {
 416                 if (id->oii_gen == OSD_OII_NOGEN)
 417                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 418
 419                 /* Do not update file c/mtime in ldiskfs.
 420                  * NB: we don't have any lock to protect this because we don't
 421                  * have reference on osd_object now, but contention with
 422                  * another lookup + attr_set can't happen in the tiny window
 423                  * between if (...) and set S_NOCMTIME. */
 424                 if (!(inode->i_flags & S_NOCMTIME))
 425                         inode->i_flags |= S_NOCMTIME;
 426         }
 427
 428         GOTO(put, rc);
 429
 430 put:
 431         if (rc != 0) {
 432                 if (!IS_ERR(inode))
 433                         iput(inode);
 434
 435                 inode = ERR_PTR(rc);
 436         }
 437
 438         return inode;
 439 }
 440
 441 /**
 442  * \retval +v: new filter_fid, does not contain self-fid
 443  * \retval 0:  filter_fid_old, contains self-fid
 444  * \retval -v: other failure cases
 445  */
 446 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 447                  struct dentry *dentry, struct lu_fid *fid)
 448 {
 449         struct filter_fid_old   *ff     = &info->oti_ff;
 450         struct ost_id           *ostid  = &info->oti_ostid;
 451         int                      rc;
 452
 453         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 454         if (rc == sizeof(*ff)) {
 455                 rc = 0;
 456                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 457                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 458                 /* XXX: use 0 as the index for compatibility, the caller will
 459                  *      handle index related issues when necessarry. */
 460                 ostid_to_fid(fid, ostid, 0);
 461         } else if (rc == sizeof(struct filter_fid)) {
 462                 rc = 1;
 463         } else if (rc >= 0) {
 464                 rc = -EINVAL;
 465         }
 466
 467         return rc;
 468 }
 469
 470 static int osd_lma_self_repair(struct osd_thread_info *info,
 471                                struct osd_device *osd, struct inode *inode,
 472                                const struct lu_fid *fid, __u32 compat)
 473 {
 474         handle_t *jh;
 475         int       rc;
 476
 477         LASSERT(current->journal_info == NULL);
 478
 479         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 480                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 481         if (IS_ERR(jh)) {
 482                 rc = PTR_ERR(jh);
 483                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 484                       osd_name(osd), rc);
 485                 return rc;
 486         }
 487
 488         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 489         if (rc != 0)
 490                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 491                       osd_name(osd), rc);
 492         ldiskfs_journal_stop(jh);
 493         return rc;
 494 }
 495
 496 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 497 {
 498         struct osd_thread_info  *info   = osd_oti_get(env);
 499         struct osd_device       *osd    = osd_obj2dev(obj);
 500         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 501         struct inode            *inode  = obj->oo_inode;
 502         struct dentry           *dentry = &info->oti_obj_dentry;
 503         struct lu_fid           *fid    = NULL;
 504         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 505         int                      rc;
 506         ENTRY;
 507
 508         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 509         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 510                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 511         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 512                 fid = &lma->lma_self_fid;
 513                 rc = osd_get_idif(info, inode, dentry, fid);
 514                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 515                         /* For the given OST-object, if it has neither LMA nor
 516                          * FID in XATTR_NAME_FID, then the given FID (which is
 517                          * contained in the @obj, from client RPC for locating
 518                          * the OST-object) is trusted. We use it to generate
 519                          * the LMA. */
 520                         osd_lma_self_repair(info, osd, inode, rfid,
 521                                             LMAC_FID_ON_OST);
 522                         RETURN(0);
 523                 }
 524         }
 525
 526         if (rc < 0)
 527                 RETURN(rc);
 528
 529         if (rc > 0) {
 530                 rc = 0;
 531                 lustre_lma_swab(lma);
 532                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 533                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 534                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 535                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 536                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 537                               PFID(rfid), inode->i_ino);
 538                         rc = -EOPNOTSUPP;
 539                 } else {
 540                         fid = &lma->lma_self_fid;
 541                 }
 542         }
 543
 544         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 545                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 546                         struct ost_id   *oi   = &info->oti_ostid;
 547                         struct lu_fid   *fid1 = &info->oti_fid3;
 548                         __u32            idx  = fid_idif_ost_idx(rfid);
 549
 550                         /* For old IDIF, the OST index is not part of the IDIF,
 551                          * Means that different OSTs may have the same IDIFs.
 552                          * Under such case, we need to make some compatible
 553                          * check to make sure to trigger OI scrub properly. */
 554                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 555                                 /* Given @rfid is new, LMA is old. */
 556                                 fid_to_ostid(fid, oi);
 557                                 ostid_to_fid(fid1, oi, idx);
 558                                 if (lu_fid_eq(fid1, rfid)) {
 559                                         if (osd->od_index_in_idif)
 560                                                 osd_lma_self_repair(info, osd,
 561                                                         inode, rfid,
 562                                                         LMAC_FID_ON_OST);
 563                                         RETURN(0);
 564                                 }
 565                         }
 566                 }
 567
 568                 rc = -EREMCHG;
 569         }
 570
 571         RETURN(rc);
 572 }
 573
 574 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 575                           const struct lu_fid *fid,
 576                           const struct lu_object_conf *conf)
 577 {
 578         struct osd_thread_info *info;
 579         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 580         struct osd_device      *dev;
 581         struct osd_idmap_cache *oic;
 582         struct osd_inode_id    *id;
 583         struct inode           *inode;
 584         struct osd_scrub       *scrub;
 585         struct scrub_file      *sf;
 586         int                     result;
 587         int                     saved  = 0;
 588         bool                    cached  = true;
 589         bool                    triggered = false;
 590         ENTRY;
 591
 592         LINVRNT(osd_invariant(obj));
 593         LASSERT(obj->oo_inode == NULL);
 594         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 595
 596         dev = osd_dev(ldev);
 597         scrub = &dev->od_scrub;
 598         sf = &scrub->os_file;
 599         info = osd_oti_get(env);
 600         LASSERT(info);
 601         oic = &info->oti_cache;
 602
 603         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 604                 RETURN(-ENOENT);
 605
 606         /* For the object is created as locking anchor, or for the object to
 607          * be created on disk. No need to osd_oi_lookup() at here because FID
 608          * shouldn't never be re-used, if it's really a duplicate FID from
 609          * unexpected reason, we should be able to detect it later by calling
 610          * do_create->osd_oi_insert(). */
 611         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 612                 GOTO(out, result = 0);
 613
 614         /* Search order: 1. per-thread cache. */
 615         if (lu_fid_eq(fid, &oic->oic_fid) &&
 616             likely(oic->oic_dev == dev)) {
 617                 id = &oic->oic_lid;
 618                 goto iget;
 619         }
 620
 621         id = &info->oti_id;
 622         if (!list_empty(&scrub->os_inconsistent_items)) {
 623                 /* Search order: 2. OI scrub pending list. */
 624                 result = osd_oii_lookup(dev, fid, id);
 625                 if (result == 0)
 626                         goto iget;
 627         }
 628
 629         cached = false;
 630         /* Search order: 3. OI files. */
 631         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 632         if (result == -ENOENT) {
 633                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 634                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
 635                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 636                                       sf->sf_oi_bitmap))
 637                         GOTO(out, result = 0);
 638
 639                 goto trigger;
 640         }
 641
 642         if (result != 0)
 643                 GOTO(out, result);
 644
 645 iget:
 646         inode = osd_iget_check(info, dev, fid, id, cached);
 647         if (IS_ERR(inode)) {
 648                 result = PTR_ERR(inode);
 649                 if (result == -ENOENT || result == -ESTALE)
 650                         GOTO(out, result = -ENOENT);
 651
 652                 if (result == -EREMCHG) {
 653
 654 trigger:
 655                         if (unlikely(triggered))
 656                                 GOTO(out, result = saved);
 657
 658                         triggered = true;
 659                         if (thread_is_running(&scrub->os_thread)) {
 660                                 result = -EINPROGRESS;
 661                         } else if (!dev->od_noscrub) {
 662                                 result = osd_scrub_start(dev, SS_AUTO_FULL |
 663                                         SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 664                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
 665                                               "for "DFID", rc = %d [1]\n",
 666                                               osd_name(dev), PFID(fid), result);
 667                                 if (result == 0 || result == -EALREADY)
 668                                         result = -EINPROGRESS;
 669                                 else
 670                                         result = -EREMCHG;
 671                         }
 672
 673                         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD))
 674                                 GOTO(out, result);
 675
 676                         /* We still have chance to get the valid inode: for the
 677                          * object which is referenced by remote name entry, the
 678                          * object on the local MDT will be linked under the dir
 679                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
 680                          *
 681                          * We do not know whether the object for the given FID
 682                          * is referenced by some remote name entry or not, and
 683                          * especially for DNE II, a multiple-linked object may
 684                          * have many name entries reside on many MDTs.
 685                          *
 686                          * To simplify the operation, OSD will not distinguish
 687                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 688                          * only happened for the RPC from other MDT during the
 689                          * OI scrub, or for the client side RPC with FID only,
 690                          * such as FID to path, or from old connected client. */
 691                         saved = result;
 692                         result = osd_lookup_in_remote_parent(info, dev,
 693                                                              fid, id);
 694                         if (result == 0) {
 695                                 cached = true;
 696                                 goto iget;
 697                         }
 698
 699                         result = saved;
 700                 }
 701
 702                 GOTO(out, result);
 703         }
 704
 705         obj->oo_inode = inode;
 706         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 707
 708         result = osd_check_lma(env, obj);
 709         if (result != 0) {
 710                 if (result == -ENODATA) {
 711                         if (cached) {
 712                                 result = osd_oi_lookup(info, dev, fid, id,
 713                                                        OI_CHECK_FLD);
 714                                 if (result != 0) {
 715                                         /* result == -ENOENT means that the OI
 716                                          * mapping has been removed by race,
 717                                          * the target inode belongs to other
 718                                          * object.
 719                                          *
 720                                          * Others error also can be returned
 721                                          * directly. */
 722                                         iput(inode);
 723                                         obj->oo_inode = NULL;
 724                                         GOTO(out, result);
 725                                 } else {
 726                                         /* result == 0 means the cached OI
 727                                          * mapping is still in the OI file,
 728                                          * the target the inode is valid. */
 729                                 }
 730                         } else {
 731                                 /* The current OI mapping is from the OI file,
 732                                  * since the inode has been found via
 733                                  * osd_iget_check(), no need recheck OI. */
 734                         }
 735
 736                         goto found;
 737                 }
 738
 739                 iput(inode);
 740                 obj->oo_inode = NULL;
 741                 if (result != -EREMCHG)
 742                         GOTO(out, result);
 743
 744                 if (cached) {
 745                         result = osd_oi_lookup(info, dev, fid, id,
 746                                                OI_CHECK_FLD);
 747                         /* result == -ENOENT means the cached OI mapping
 748                          * has been removed from the OI file by race,
 749                          * above target inode belongs to other object.
 750                          *
 751                          * Others error also can be returned directly. */
 752                         if (result != 0)
 753                                 GOTO(out, result);
 754
 755                         /* result == 0, goto trigger */
 756                 } else {
 757                         /* The current OI mapping is from the OI file,
 758                          * since the inode has been found via
 759                          * osd_iget_check(), no need recheck OI. */
 760                 }
 761
 762                 goto trigger;
 763         }
 764
 765 found:
 766         obj->oo_compat_dot_created = 1;
 767         obj->oo_compat_dotdot_created = 1;
 768
 769         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 770                 GOTO(out, result = 0);
 771
 772         LASSERT(obj->oo_hl_head == NULL);
 773         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 774         if (obj->oo_hl_head == NULL) {
 775                 obj->oo_inode = NULL;
 776                 iput(inode);
 777                 GOTO(out, result = -ENOMEM);
 778         }
 779         GOTO(out, result = 0);
 780
 781 out:
 782         if (result != 0 && cached)
 783                 fid_zero(&oic->oic_fid);
 784
 785         LINVRNT(osd_invariant(obj));
 786         return result;
 787 }
 788
 789 /*
 790  * Concurrency: shouldn't matter.
 791  */
 792 static void osd_object_init0(struct osd_object *obj)
 793 {
 794         LASSERT(obj->oo_inode != NULL);
 795         obj->oo_dt.do_body_ops = &osd_body_ops;
 796         obj->oo_dt.do_lu.lo_header->loh_attr |=
 797                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 798 }
 799
 800 /*
 801  * Concurrency: no concurrent access is possible that early in object
 802  * life-cycle.
 803  */
 804 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 805                            const struct lu_object_conf *conf)
 806 {
 807         struct osd_object *obj = osd_obj(l);
 808         int result;
 809
 810         LINVRNT(osd_invariant(obj));
 811
 812         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 813                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 814                 l->lo_header->loh_attr |= LOHA_EXISTS;
 815                 return 0;
 816         }
 817
 818         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 819         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 820         if (result == 0 && obj->oo_inode != NULL)
 821                 osd_object_init0(obj);
 822
 823         LINVRNT(osd_invariant(obj));
 824         return result;
 825 }
 826
 827 /*
 828  * Concurrency: no concurrent access is possible that late in object
 829  * life-cycle.
 830  */
 831 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 832 {
 833         struct osd_object *obj = osd_obj(l);
 834
 835         LINVRNT(osd_invariant(obj));
 836
 837         dt_object_fini(&obj->oo_dt);
 838         if (obj->oo_hl_head != NULL)
 839                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 840         OBD_FREE_PTR(obj);
 841 }
 842
 843 /*
 844  * Concurrency: no concurrent access is possible that late in object
 845  * life-cycle.
 846  */
 847 static void osd_index_fini(struct osd_object *o)
 848 {
 849         struct iam_container *bag;
 850
 851         if (o->oo_dir != NULL) {
 852                 bag = &o->oo_dir->od_container;
 853                 if (o->oo_inode != NULL) {
 854                         if (bag->ic_object == o->oo_inode)
 855                                 iam_container_fini(bag);
 856                 }
 857                 OBD_FREE_PTR(o->oo_dir);
 858                 o->oo_dir = NULL;
 859         }
 860 }
 861
 862 /*
 863  * Concurrency: no concurrent access is possible that late in object
 864  * life-cycle (for all existing callers, that is. New callers have to provide
 865  * their own locking.)
 866  */
 867 static int osd_inode_unlinked(const struct inode *inode)
 868 {
 869         return inode->i_nlink == 0;
 870 }
 871
 872 enum {
 873         OSD_TXN_OI_DELETE_CREDITS    = 20,
 874         OSD_TXN_INODE_DELETE_CREDITS = 20
 875 };
 876
 877 /*
 878  * Journal
 879  */
 880
 881 #if OSD_THANDLE_STATS
 882 /**
 883  * Set time when the handle is allocated
 884  */
 885 static void osd_th_alloced(struct osd_thandle *oth)
 886 {
 887         oth->oth_alloced = cfs_time_current();
 888 }
 889
 890 /**
 891  * Set time when the handle started
 892  */
 893 static void osd_th_started(struct osd_thandle *oth)
 894 {
 895         oth->oth_started = cfs_time_current();
 896 }
 897
 898 /**
 899  * Helper function to convert time interval to microseconds packed in
 900  * long int.
 901  */
 902 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 903 {
 904         struct timeval val;
 905
 906         cfs_duration_usec(cfs_time_sub(end, start), &val);
 907         return val.tv_sec * 1000000 + val.tv_usec;
 908 }
 909
 910 /**
 911  * Check whether the we deal with this handle for too long.
 912  */
 913 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 914                                 cfs_time_t alloced, cfs_time_t started,
 915                                 cfs_time_t closed)
 916 {
 917         cfs_time_t now = cfs_time_current();
 918
 919         LASSERT(dev != NULL);
 920
 921         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 922                             interval_to_usec(alloced, started));
 923         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 924                             interval_to_usec(started, closed));
 925         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 926                             interval_to_usec(closed, now));
 927
 928         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 929                 CWARN("transaction handle %p was open for too long: "
 930                       "now "CFS_TIME_T" ,"
 931                       "alloced "CFS_TIME_T" ,"
 932                       "started "CFS_TIME_T" ,"
 933                       "closed "CFS_TIME_T"\n",
 934                       oth, now, alloced, started, closed);
 935                 libcfs_debug_dumpstack(NULL);
 936         }
 937 }
 938
 939 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 940 {                                                                       \
 941         cfs_time_t __closed = cfs_time_current();                       \
 942         cfs_time_t __alloced = oth->oth_alloced;                        \
 943         cfs_time_t __started = oth->oth_started;                        \
 944                                                                         \
 945         expr;                                                           \
 946         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 947 }
 948
 949 #else /* OSD_THANDLE_STATS */
 950
 951 #define osd_th_alloced(h)                  do {} while(0)
 952 #define osd_th_started(h)                  do {} while(0)
 953 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 954
 955 #endif /* OSD_THANDLE_STATS */
 956
 957 /*
 958  * Concurrency: doesn't access mutable data.
 959  */
 960 static int osd_param_is_not_sane(const struct osd_device *dev,
 961                                  const struct thandle *th)
 962 {
 963         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 964
 965         return oh->ot_credits > osd_transaction_size(dev);
 966 }
 967
 968 /*
 969  * Concurrency: shouldn't matter.
 970  */
 971 static void osd_trans_commit_cb(struct super_block *sb,
 972                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 973 {
 974         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 975         struct thandle     *th  = &oh->ot_super;
 976         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 977         struct dt_txn_commit_cb *dcb, *tmp;
 978
 979         LASSERT(oh->ot_handle == NULL);
 980
 981         if (error)
 982                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 983
 984         dt_txn_hook_commit(th);
 985
 986         /* call per-transaction callbacks if any */
 987         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
 988                                  dcb_linkage) {
 989                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 990                          "commit callback entry: magic=%x name='%s'\n",
 991                          dcb->dcb_magic, dcb->dcb_name);
 992                 list_del_init(&dcb->dcb_linkage);
 993                 dcb->dcb_func(NULL, th, dcb, error);
 994         }
 995
 996         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 997         lu_device_put(lud);
 998         th->th_dev = NULL;
 999
1000         lu_context_exit(&th->th_ctx);
1001         lu_context_fini(&th->th_ctx);
1002         OBD_FREE_PTR(oh);
1003 }
1004
1005 static struct thandle *osd_trans_create(const struct lu_env *env,
1006                                         struct dt_device *d)
1007 {
1008         struct osd_thread_info  *oti = osd_oti_get(env);
1009         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1010         struct osd_thandle      *oh;
1011         struct thandle          *th;
1012         ENTRY;
1013
1014         /* on pending IO in this thread should left from prev. request */
1015         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1016
1017         th = ERR_PTR(-ENOMEM);
1018         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1019         if (oh != NULL) {
1020                 oh->ot_quota_trans = &oti->oti_quota_trans;
1021                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1022                 th = &oh->ot_super;
1023                 th->th_dev = d;
1024                 th->th_result = 0;
1025                 th->th_tags = LCT_TX_HANDLE;
1026                 oh->ot_credits = 0;
1027                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1028                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1029                 osd_th_alloced(oh);
1030
1031                 memset(oti->oti_declare_ops, 0,
1032                        sizeof(oti->oti_declare_ops));
1033                 memset(oti->oti_declare_ops_cred, 0,
1034                        sizeof(oti->oti_declare_ops_cred));
1035                 memset(oti->oti_declare_ops_used, 0,
1036                        sizeof(oti->oti_declare_ops_used));
1037         }
1038         RETURN(th);
1039 }
1040
1041 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1042 {
1043         struct osd_thread_info  *oti = osd_oti_get(env);
1044         struct osd_thandle      *oh;
1045
1046         oh = container_of0(th, struct osd_thandle, ot_super);
1047         LASSERT(oh != NULL);
1048
1049         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1050               oti->oti_declare_ops[OSD_OT_CREATE],
1051               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1052               oti->oti_declare_ops_used[OSD_OT_CREATE],
1053               oti->oti_declare_ops[OSD_OT_DESTROY],
1054               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1055               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1056         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1057               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1058               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1059               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1060               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1061               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1062               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1063         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1064               oti->oti_declare_ops[OSD_OT_WRITE],
1065               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1066               oti->oti_declare_ops_used[OSD_OT_WRITE],
1067               oti->oti_declare_ops[OSD_OT_PUNCH],
1068               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1069               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1070               oti->oti_declare_ops[OSD_OT_QUOTA],
1071               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1072               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1073         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1074               oti->oti_declare_ops[OSD_OT_INSERT],
1075               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1076               oti->oti_declare_ops_used[OSD_OT_INSERT],
1077               oti->oti_declare_ops[OSD_OT_DELETE],
1078               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1079               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1080         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1081               oti->oti_declare_ops[OSD_OT_REF_ADD],
1082               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1083               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1084               oti->oti_declare_ops[OSD_OT_REF_DEL],
1085               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1086               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1087 }
1088
1089 /*
1090  * Concurrency: shouldn't matter.
1091  */
1092 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1093                            struct thandle *th)
1094 {
1095         struct osd_thread_info *oti = osd_oti_get(env);
1096         struct osd_device  *dev = osd_dt_dev(d);
1097         handle_t           *jh;
1098         struct osd_thandle *oh;
1099         int rc;
1100
1101         ENTRY;
1102
1103         LASSERT(current->journal_info == NULL);
1104
1105         oh = container_of0(th, struct osd_thandle, ot_super);
1106         LASSERT(oh != NULL);
1107         LASSERT(oh->ot_handle == NULL);
1108
1109         rc = dt_txn_hook_start(env, d, th);
1110         if (rc != 0)
1111                 GOTO(out, rc);
1112
1113         if (unlikely(osd_param_is_not_sane(dev, th))) {
1114                 static unsigned long last_printed;
1115                 static int last_credits;
1116
1117                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1118                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1119                       oh->ot_credits,
1120                       osd_journal(dev)->j_max_transaction_buffers);
1121
1122                 osd_trans_dump_creds(env, th);
1123
1124                 if (last_credits != oh->ot_credits &&
1125                     time_after(jiffies, last_printed +
1126                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1127                         libcfs_debug_dumpstack(NULL);
1128                         last_credits = oh->ot_credits;
1129                         last_printed = jiffies;
1130                 }
1131                 /* XXX Limit the credits to 'max_transaction_buffers', and
1132                  *     let the underlying filesystem to catch the error if
1133                  *     we really need so many credits.
1134                  *
1135                  *     This should be removed when we can calculate the
1136                  *     credits precisely. */
1137                 oh->ot_credits = osd_transaction_size(dev);
1138         }
1139
1140         /*
1141          * XXX temporary stuff. Some abstraction layer should
1142          * be used.
1143          */
1144         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1145         osd_th_started(oh);
1146         if (!IS_ERR(jh)) {
1147                 oh->ot_handle = jh;
1148                 LASSERT(oti->oti_txns == 0);
1149                 lu_context_init(&th->th_ctx, th->th_tags);
1150                 lu_context_enter(&th->th_ctx);
1151
1152                 lu_device_get(&d->dd_lu_dev);
1153                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1154                               "osd-tx", th);
1155                 oti->oti_txns++;
1156                 rc = 0;
1157         } else {
1158                 rc = PTR_ERR(jh);
1159         }
1160 out:
1161         RETURN(rc);
1162 }
1163
1164 static int osd_seq_exists(const struct lu_env *env,
1165                           struct osd_device *osd, u64 seq)
1166 {
1167         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1168         struct seq_server_site  *ss = osd_seq_site(osd);
1169         int                     rc;
1170         ENTRY;
1171
1172         LASSERT(ss != NULL);
1173         LASSERT(ss->ss_server_fld != NULL);
1174
1175         rc = osd_fld_lookup(env, osd, seq, range);
1176         if (rc != 0) {
1177                 if (rc != -ENOENT)
1178                         CERROR("%s: can't lookup FLD sequence "LPX64
1179                                ": rc = %d\n", osd_name(osd), seq, rc);
1180                 RETURN(0);
1181         }
1182
1183         RETURN(ss->ss_node_id == range->lsr_index);
1184 }
1185
1186 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1187 {
1188         struct dt_txn_commit_cb *dcb;
1189         struct dt_txn_commit_cb *tmp;
1190
1191         /* call per-transaction stop callbacks if any */
1192         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1193                                  dcb_linkage) {
1194                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1195                          "commit callback entry: magic=%x name='%s'\n",
1196                          dcb->dcb_magic, dcb->dcb_name);
1197                 list_del_init(&dcb->dcb_linkage);
1198                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1199         }
1200 }
1201
1202 /*
1203  * Concurrency: shouldn't matter.
1204  */
1205 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1206                           struct thandle *th)
1207 {
1208         int                     rc = 0, remove_agents = 0;
1209         struct osd_thandle     *oh;
1210         struct osd_thread_info *oti = osd_oti_get(env);
1211         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1212         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1213         struct qsd_instance    *qsd = osd->od_quota_slave;
1214         struct lquota_trans    *qtrans;
1215         ENTRY;
1216
1217         oh = container_of0(th, struct osd_thandle, ot_super);
1218
1219         remove_agents = oh->ot_remove_agents;
1220
1221         qtrans = oh->ot_quota_trans;
1222         oh->ot_quota_trans = NULL;
1223
1224         if (oh->ot_handle != NULL) {
1225                 handle_t *hdl = oh->ot_handle;
1226
1227                 /*
1228                  * add commit callback
1229                  * notice we don't do this in osd_trans_start()
1230                  * as underlying transaction can change during truncate
1231                  */
1232                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1233                                          &oh->ot_jcb);
1234
1235                 LASSERT(oti->oti_txns == 1);
1236                 oti->oti_txns--;
1237
1238                 rc = dt_txn_hook_stop(env, th);
1239                 if (rc != 0)
1240                         CERROR("%s: failed in transaction hook: rc = %d\n",
1241                                osd_name(osd), rc);
1242
1243                 osd_trans_stop_cb(oh, rc);
1244                 /* hook functions might modify th_sync */
1245                 hdl->h_sync = th->th_sync;
1246
1247                 oh->ot_handle = NULL;
1248                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1249                 if (rc != 0)
1250                         CERROR("%s: failed to stop transaction: rc = %d\n",
1251                                osd_name(osd), rc);
1252         } else {
1253                 osd_trans_stop_cb(oh, th->th_result);
1254                 OBD_FREE_PTR(oh);
1255         }
1256
1257         /* inform the quota slave device that the transaction is stopping */
1258         qsd_op_end(env, qsd, qtrans);
1259
1260         /* as we want IO to journal and data IO be concurrent, we don't block
1261          * awaiting data IO completion in osd_do_bio(), instead we wait here
1262          * once transaction is submitted to the journal. all reqular requests
1263          * don't do direct IO (except read/write), thus this wait_event becomes
1264          * no-op for them.
1265          *
1266          * IMPORTANT: we have to wait till any IO submited by the thread is
1267          * completed otherwise iobuf may be corrupted by different request
1268          */
1269         wait_event(iobuf->dr_wait,
1270                        atomic_read(&iobuf->dr_numreqs) == 0);
1271         osd_fini_iobuf(osd, iobuf);
1272         if (!rc)
1273                 rc = iobuf->dr_error;
1274
1275         if (unlikely(remove_agents != 0))
1276                 osd_process_scheduled_agent_removals(env, osd);
1277
1278         RETURN(rc);
1279 }
1280
1281 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1282 {
1283         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1284                                                ot_super);
1285
1286         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1287         LASSERT(&dcb->dcb_func != NULL);
1288         if (dcb->dcb_flags & DCB_TRANS_STOP)
1289                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1290         else
1291                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1292
1293         return 0;
1294 }
1295
1296 /*
1297  * Called just before object is freed. Releases all resources except for
1298  * object itself (that is released by osd_object_free()).
1299  *
1300  * Concurrency: no concurrent access is possible that late in object
1301  * life-cycle.
1302  */
1303 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1304 {
1305         struct osd_object *obj   = osd_obj(l);
1306         struct inode      *inode = obj->oo_inode;
1307
1308         LINVRNT(osd_invariant(obj));
1309
1310         /*
1311          * If object is unlinked remove fid->ino mapping from object index.
1312          */
1313
1314         osd_index_fini(obj);
1315         if (inode != NULL) {
1316                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1317                 qid_t                    uid = i_uid_read(inode);
1318                 qid_t                    gid = i_gid_read(inode);
1319
1320                 iput(inode);
1321                 obj->oo_inode = NULL;
1322
1323                 if (qsd != NULL) {
1324                         struct osd_thread_info  *info = osd_oti_get(env);
1325                         struct lquota_id_info   *qi = &info->oti_qi;
1326
1327                         /* Release granted quota to master if necessary */
1328                         qi->lqi_id.qid_uid = uid;
1329                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1330
1331                         qi->lqi_id.qid_uid = gid;
1332                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1333                 }
1334         }
1335 }
1336
1337 /*
1338  * Concurrency: ->loo_object_release() is called under site spin-lock.
1339  */
1340 static void osd_object_release(const struct lu_env *env,
1341                                struct lu_object *l)
1342 {
1343 }
1344
1345 /*
1346  * Concurrency: shouldn't matter.
1347  */
1348 static int osd_object_print(const struct lu_env *env, void *cookie,
1349                             lu_printer_t p, const struct lu_object *l)
1350 {
1351         struct osd_object *o = osd_obj(l);
1352         struct iam_descr  *d;
1353
1354         if (o->oo_dir != NULL)
1355                 d = o->oo_dir->od_container.ic_descr;
1356         else
1357                 d = NULL;
1358         return (*p)(env, cookie,
1359                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1360                     o, o->oo_inode,
1361                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1362                     o->oo_inode ? o->oo_inode->i_generation : 0,
1363                     d ? d->id_ops->id_name : "plain");
1364 }
1365
1366 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1367
1368 /*
1369  * Concurrency: shouldn't matter.
1370  */
1371 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1372                struct obd_statfs *sfs)
1373 {
1374         struct osd_device  *osd = osd_dt_dev(d);
1375         struct super_block *sb = osd_sb(osd);
1376         struct kstatfs     *ksfs;
1377         int result = 0;
1378
1379         if (unlikely(osd->od_mnt == NULL))
1380                 return -EINPROGRESS;
1381
1382         /* osd_lproc.c call this without env, allocate ksfs for that case */
1383         if (unlikely(env == NULL)) {
1384                 OBD_ALLOC_PTR(ksfs);
1385                 if (ksfs == NULL)
1386                         return -ENOMEM;
1387         } else {
1388                 ksfs = &osd_oti_get(env)->oti_ksfs;
1389         }
1390
1391         spin_lock(&osd->od_osfs_lock);
1392         result = sb->s_op->statfs(sb->s_root, ksfs);
1393         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1394                 statfs_pack(sfs, ksfs);
1395                 if (unlikely(sb->s_flags & MS_RDONLY))
1396                         sfs->os_state = OS_STATE_READONLY;
1397                 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1398                                               LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1399                         sfs->os_maxbytes = sb->s_maxbytes;
1400                 else
1401                         sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1402         }
1403         spin_unlock(&osd->od_osfs_lock);
1404
1405         if (unlikely(env == NULL))
1406                 OBD_FREE_PTR(ksfs);
1407
1408         /* Reserve a small amount of space for local objects like last_rcvd,
1409          * llog, quota files, ... */
1410         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1411                 sfs->os_bavail = 0;
1412         } else {
1413                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1414                 /** Take out metadata overhead for indirect blocks */
1415                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1416         }
1417
1418         return result;
1419 }
1420
1421 /**
1422  * Estimate space needed for file creations. We assume the largest filename
1423  * which is 2^64 - 1, hence a filename of 20 chars.
1424  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1425  */
1426 #ifdef __LDISKFS_DIR_REC_LEN
1427 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1428 #else
1429 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1430 #endif
1431
1432 /*
1433  * Concurrency: doesn't access mutable data.
1434  */
1435 static void osd_conf_get(const struct lu_env *env,
1436                          const struct dt_device *dev,
1437                          struct dt_device_param *param)
1438 {
1439         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1440         int                ea_overhead;
1441
1442         /*
1443          * XXX should be taken from not-yet-existing fs abstraction layer.
1444          */
1445         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1446         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1447         param->ddp_block_shift  = sb->s_blocksize_bits;
1448         param->ddp_mount_type     = LDD_MT_LDISKFS;
1449         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1450                 param->ddp_maxbytes = sb->s_maxbytes;
1451         else
1452                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1453         /* Overhead estimate should be fairly accurate, so we really take a tiny
1454          * error margin which also avoids fragmenting the filesystem too much */
1455         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1456         /* inode are statically allocated, so per-inode space consumption
1457          * is the space consumed by the directory entry */
1458         param->ddp_inodespace     = PER_OBJ_USAGE;
1459         /* per-fragment overhead to be used by the client code */
1460         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1461         param->ddp_mntopts      = 0;
1462         if (test_opt(sb, XATTR_USER))
1463                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1464         if (test_opt(sb, POSIX_ACL))
1465                 param->ddp_mntopts |= MNTOPT_ACL;
1466
1467         /* LOD might calculate the max stripe count based on max_ea_size,
1468          * so we need take account in the overhead as well,
1469          * xattr_header + magic + xattr_entry_head */
1470         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1471                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1472
1473 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1474         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1475                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1476                                                                 ea_overhead;
1477         else
1478 #endif
1479                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1480 }
1481
1482 /*
1483  * Concurrency: shouldn't matter.
1484  */
1485 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1486 {
1487         int rc;
1488
1489         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1490
1491         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1492
1493         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1494                LUSTRE_OSD_LDISKFS_NAME, rc);
1495
1496         return rc;
1497 }
1498
1499 /**
1500  * Start commit for OSD device.
1501  *
1502  * An implementation of dt_commit_async method for OSD device.
1503  * Asychronously starts underlayng fs sync and thereby a transaction
1504  * commit.
1505  *
1506  * \param env environment
1507  * \param d dt device
1508  *
1509  * \see dt_device_operations
1510  */
1511 static int osd_commit_async(const struct lu_env *env,
1512                             struct dt_device *d)
1513 {
1514         struct super_block *s = osd_sb(osd_dt_dev(d));
1515         ENTRY;
1516
1517         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1518         RETURN(s->s_op->sync_fs(s, 0));
1519 }
1520
1521 /*
1522  * Concurrency: shouldn't matter.
1523  */
1524
1525 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1526 {
1527         struct super_block *sb = osd_sb(osd_dt_dev(d));
1528         struct block_device *dev = sb->s_bdev;
1529 #ifdef HAVE_DEV_SET_RDONLY
1530         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1531         int rc = 0;
1532 #else
1533         int rc = -EOPNOTSUPP;
1534 #endif
1535         ENTRY;
1536
1537 #ifdef HAVE_DEV_SET_RDONLY
1538         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1539
1540         if (jdev && (jdev != dev)) {
1541                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1542                        (long)jdev);
1543                 dev_set_rdonly(jdev);
1544         }
1545         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1546         dev_set_rdonly(dev);
1547 #else
1548         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1549                osd_dt_dev(d)->od_svname, (long)dev, rc);
1550 #endif
1551         RETURN(rc);
1552 }
1553
1554 /**
1555  * Note: we do not count into QUOTA here.
1556  * If we mount with --data_journal we may need more.
1557  */
1558 const int osd_dto_credits_noquota[DTO_NR] = {
1559         /**
1560          * Insert.
1561          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1562          * SINGLEDATA_TRANS_BLOCKS(8)
1563          * XXX Note: maybe iam need more, since iam have more level than
1564          *           EXT3 htree.
1565          */
1566         [DTO_INDEX_INSERT]  = 16,
1567         /**
1568          * Delete
1569          * just modify a single entry, probably merge few within a block
1570          */
1571         [DTO_INDEX_DELETE]  = 1,
1572         /**
1573          * Used for OI scrub
1574          */
1575         [DTO_INDEX_UPDATE]  = 16,
1576         /**
1577          * 4(inode, inode bits, groups, GDT)
1578          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1579          */
1580         [DTO_OBJECT_CREATE] = 4,
1581         /**
1582          * 4(inode, inode bits, groups, GDT)
1583          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1584          */
1585         [DTO_OBJECT_DELETE] = 4,
1586         /**
1587          * Attr set credits (inode)
1588          */
1589         [DTO_ATTR_SET_BASE] = 1,
1590         /**
1591          * Xattr set. The same as xattr of EXT3.
1592          * DATA_TRANS_BLOCKS(14)
1593          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1594          * are also counted in. Do not know why?
1595          */
1596         [DTO_XATTR_SET]     = 14,
1597         /**
1598          * credits for inode change during write.
1599          */
1600         [DTO_WRITE_BASE]    = 3,
1601         /**
1602          * credits for single block write.
1603          */
1604         [DTO_WRITE_BLOCK]   = 14,
1605         /**
1606          * Attr set credits for chown.
1607          * This is extra credits for setattr, and it is null without quota
1608          */
1609         [DTO_ATTR_SET_CHOWN] = 0
1610 };
1611
1612 static const struct dt_device_operations osd_dt_ops = {
1613         .dt_root_get       = osd_root_get,
1614         .dt_statfs         = osd_statfs,
1615         .dt_trans_create   = osd_trans_create,
1616         .dt_trans_start    = osd_trans_start,
1617         .dt_trans_stop     = osd_trans_stop,
1618         .dt_trans_cb_add   = osd_trans_cb_add,
1619         .dt_conf_get       = osd_conf_get,
1620         .dt_sync           = osd_sync,
1621         .dt_ro             = osd_ro,
1622         .dt_commit_async   = osd_commit_async,
1623 };
1624
1625 static void osd_object_read_lock(const struct lu_env *env,
1626                                  struct dt_object *dt, unsigned role)
1627 {
1628         struct osd_object *obj = osd_dt_obj(dt);
1629         struct osd_thread_info *oti = osd_oti_get(env);
1630
1631         LINVRNT(osd_invariant(obj));
1632
1633         LASSERT(obj->oo_owner != env);
1634         down_read_nested(&obj->oo_sem, role);
1635
1636         LASSERT(obj->oo_owner == NULL);
1637         oti->oti_r_locks++;
1638 }
1639
1640 static void osd_object_write_lock(const struct lu_env *env,
1641                                   struct dt_object *dt, unsigned role)
1642 {
1643         struct osd_object *obj = osd_dt_obj(dt);
1644         struct osd_thread_info *oti = osd_oti_get(env);
1645
1646         LINVRNT(osd_invariant(obj));
1647
1648         LASSERT(obj->oo_owner != env);
1649         down_write_nested(&obj->oo_sem, role);
1650
1651         LASSERT(obj->oo_owner == NULL);
1652         obj->oo_owner = env;
1653         oti->oti_w_locks++;
1654 }
1655
1656 static void osd_object_read_unlock(const struct lu_env *env,
1657                                    struct dt_object *dt)
1658 {
1659         struct osd_object *obj = osd_dt_obj(dt);
1660         struct osd_thread_info *oti = osd_oti_get(env);
1661
1662         LINVRNT(osd_invariant(obj));
1663
1664         LASSERT(oti->oti_r_locks > 0);
1665         oti->oti_r_locks--;
1666         up_read(&obj->oo_sem);
1667 }
1668
1669 static void osd_object_write_unlock(const struct lu_env *env,
1670                                     struct dt_object *dt)
1671 {
1672         struct osd_object *obj = osd_dt_obj(dt);
1673         struct osd_thread_info *oti = osd_oti_get(env);
1674
1675         LINVRNT(osd_invariant(obj));
1676
1677         LASSERT(obj->oo_owner == env);
1678         LASSERT(oti->oti_w_locks > 0);
1679         oti->oti_w_locks--;
1680         obj->oo_owner = NULL;
1681         up_write(&obj->oo_sem);
1682 }
1683
1684 static int osd_object_write_locked(const struct lu_env *env,
1685                                    struct dt_object *dt)
1686 {
1687         struct osd_object *obj = osd_dt_obj(dt);
1688
1689         LINVRNT(osd_invariant(obj));
1690
1691         return obj->oo_owner == env;
1692 }
1693
1694 static struct timespec *osd_inode_time(const struct lu_env *env,
1695                                        struct inode *inode, __u64 seconds)
1696 {
1697         struct osd_thread_info  *oti = osd_oti_get(env);
1698         struct timespec         *t   = &oti->oti_time;
1699
1700         t->tv_sec = seconds;
1701         t->tv_nsec = 0;
1702         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1703         return t;
1704 }
1705
1706
1707 static void osd_inode_getattr(const struct lu_env *env,
1708                               struct inode *inode, struct lu_attr *attr)
1709 {
1710         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1711                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1712                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1713                            LA_TYPE;
1714
1715         attr->la_atime   = LTIME_S(inode->i_atime);
1716         attr->la_mtime   = LTIME_S(inode->i_mtime);
1717         attr->la_ctime   = LTIME_S(inode->i_ctime);
1718         attr->la_mode    = inode->i_mode;
1719         attr->la_size    = i_size_read(inode);
1720         attr->la_blocks  = inode->i_blocks;
1721         attr->la_uid     = i_uid_read(inode);
1722         attr->la_gid     = i_gid_read(inode);
1723         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1724         attr->la_nlink   = inode->i_nlink;
1725         attr->la_rdev    = inode->i_rdev;
1726         attr->la_blksize = 1 << inode->i_blkbits;
1727         attr->la_blkbits = inode->i_blkbits;
1728 }
1729
1730 static int osd_attr_get(const struct lu_env *env,
1731                         struct dt_object *dt,
1732                         struct lu_attr *attr)
1733 {
1734         struct osd_object *obj = osd_dt_obj(dt);
1735
1736         if (!dt_object_exists(dt))
1737                 return -ENOENT;
1738
1739         LASSERT(!dt_object_remote(dt));
1740         LINVRNT(osd_invariant(obj));
1741
1742         spin_lock(&obj->oo_guard);
1743         osd_inode_getattr(env, obj->oo_inode, attr);
1744         spin_unlock(&obj->oo_guard);
1745         return 0;
1746 }
1747
1748 static int osd_declare_attr_set(const struct lu_env *env,
1749                                 struct dt_object *dt,
1750                                 const struct lu_attr *attr,
1751                                 struct thandle *handle)
1752 {
1753         struct osd_thandle     *oh;
1754         struct osd_object      *obj;
1755         struct osd_thread_info *info = osd_oti_get(env);
1756         struct lquota_id_info  *qi = &info->oti_qi;
1757         qid_t                   uid;
1758         qid_t                   gid;
1759         long long               bspace;
1760         int                     rc = 0;
1761         bool                    enforce;
1762         ENTRY;
1763
1764         LASSERT(dt != NULL);
1765         LASSERT(handle != NULL);
1766
1767         obj = osd_dt_obj(dt);
1768         LASSERT(osd_invariant(obj));
1769
1770         oh = container_of0(handle, struct osd_thandle, ot_super);
1771         LASSERT(oh->ot_handle == NULL);
1772
1773         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1774                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1775
1776         if (attr == NULL || obj->oo_inode == NULL)
1777                 RETURN(rc);
1778
1779         bspace   = obj->oo_inode->i_blocks;
1780         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1781         bspace   = toqb(bspace);
1782
1783         /* Changing ownership is always preformed by super user, it should not
1784          * fail with EDQUOT.
1785          *
1786          * We still need to call the osd_declare_qid() to calculate the journal
1787          * credits for updating quota accounting files and to trigger quota
1788          * space adjustment once the operation is completed.*/
1789         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1790                 /* USERQUOTA */
1791                 uid = i_uid_read(obj->oo_inode);
1792                 qi->lqi_type = USRQUOTA;
1793                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1794                 /* inode accounting */
1795                 qi->lqi_is_blk = false;
1796
1797                 /* one more inode for the new uid ... */
1798                 qi->lqi_id.qid_uid = attr->la_uid;
1799                 qi->lqi_space      = 1;
1800                 /* Reserve credits for the new uid */
1801                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1802                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1803                         rc = 0;
1804                 if (rc)
1805                         RETURN(rc);
1806
1807                 /* and one less inode for the current uid */
1808                 qi->lqi_id.qid_uid = uid;
1809                 qi->lqi_space      = -1;
1810                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1811                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1812                         rc = 0;
1813                 if (rc)
1814                         RETURN(rc);
1815
1816                 /* block accounting */
1817                 qi->lqi_is_blk = true;
1818
1819                 /* more blocks for the new uid ... */
1820                 qi->lqi_id.qid_uid = attr->la_uid;
1821                 qi->lqi_space      = bspace;
1822                 /*
1823                  * Credits for the new uid has been reserved, re-use "obj"
1824                  * to save credit reservation.
1825                  */
1826                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1827                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1828                         rc = 0;
1829                 if (rc)
1830                         RETURN(rc);
1831
1832                 /* and finally less blocks for the current uid */
1833                 qi->lqi_id.qid_uid = uid;
1834                 qi->lqi_space      = -bspace;
1835                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1836                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1837                         rc = 0;
1838                 if (rc)
1839                         RETURN(rc);
1840
1841                 /* GROUP QUOTA */
1842                 gid = i_gid_read(obj->oo_inode);
1843                 qi->lqi_type = GRPQUOTA;
1844                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1845
1846                 /* inode accounting */
1847                 qi->lqi_is_blk = false;
1848
1849                 /* one more inode for the new gid ... */
1850                 qi->lqi_id.qid_gid = attr->la_gid;
1851                 qi->lqi_space      = 1;
1852                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1853                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1854                         rc = 0;
1855                 if (rc)
1856                         RETURN(rc);
1857
1858                 /* and one less inode for the current gid */
1859                 qi->lqi_id.qid_gid = gid;
1860                 qi->lqi_space      = -1;
1861                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1862                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1863                         rc = 0;
1864                 if (rc)
1865                         RETURN(rc);
1866
1867                 /* block accounting */
1868                 qi->lqi_is_blk = true;
1869
1870                 /* more blocks for the new gid ... */
1871                 qi->lqi_id.qid_gid = attr->la_gid;
1872                 qi->lqi_space      = bspace;
1873                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1874                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1875                         rc = 0;
1876                 if (rc)
1877                         RETURN(rc);
1878
1879                 /* and finally less blocks for the current gid */
1880                 qi->lqi_id.qid_gid = gid;
1881                 qi->lqi_space      = -bspace;
1882                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1883                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1884                         rc = 0;
1885                 if (rc)
1886                         RETURN(rc);
1887         }
1888
1889         RETURN(rc);
1890 }
1891
1892 static int osd_inode_setattr(const struct lu_env *env,
1893                              struct inode *inode, const struct lu_attr *attr)
1894 {
1895         __u64 bits = attr->la_valid;
1896
1897         /* Only allow set size for regular file */
1898         if (!S_ISREG(inode->i_mode))
1899                 bits &= ~(LA_SIZE | LA_BLOCKS);
1900
1901         if (bits == 0)
1902                 return 0;
1903
1904         if (bits & LA_ATIME)
1905                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1906         if (bits & LA_CTIME)
1907                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1908         if (bits & LA_MTIME)
1909                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1910         if (bits & LA_SIZE) {
1911                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1912                 i_size_write(inode, attr->la_size);
1913         }
1914
1915 #if 0
1916         /* OSD should not change "i_blocks" which is used by quota.
1917          * "i_blocks" should be changed by ldiskfs only. */
1918         if (bits & LA_BLOCKS)
1919                 inode->i_blocks = attr->la_blocks;
1920 #endif
1921         if (bits & LA_MODE)
1922                 inode->i_mode = (inode->i_mode & S_IFMT) |
1923                                 (attr->la_mode & ~S_IFMT);
1924         if (bits & LA_UID)
1925                 i_uid_write(inode, attr->la_uid);
1926         if (bits & LA_GID)
1927                 i_gid_write(inode, attr->la_gid);
1928         if (bits & LA_NLINK)
1929                 set_nlink(inode, attr->la_nlink);
1930         if (bits & LA_RDEV)
1931                 inode->i_rdev = attr->la_rdev;
1932
1933         if (bits & LA_FLAGS) {
1934                 /* always keep S_NOCMTIME */
1935                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1936                                  S_NOCMTIME;
1937         }
1938         return 0;
1939 }
1940
1941 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1942 {
1943         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1944             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1945                 struct iattr    iattr;
1946                 int             rc;
1947
1948                 ll_vfs_dq_init(inode);
1949                 iattr.ia_valid = 0;
1950                 if (attr->la_valid & LA_UID)
1951                         iattr.ia_valid |= ATTR_UID;
1952                 if (attr->la_valid & LA_GID)
1953                         iattr.ia_valid |= ATTR_GID;
1954                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1955                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1956
1957                 rc = ll_vfs_dq_transfer(inode, &iattr);
1958                 if (rc) {
1959                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1960                                "enforcement enabled on the ldiskfs "
1961                                "filesystem?\n", inode->i_sb->s_id, rc);
1962                         return rc;
1963                 }
1964         }
1965         return 0;
1966 }
1967
1968 static int osd_attr_set(const struct lu_env *env,
1969                         struct dt_object *dt,
1970                         const struct lu_attr *attr,
1971                         struct thandle *handle)
1972 {
1973         struct osd_object *obj = osd_dt_obj(dt);
1974         struct inode      *inode;
1975         int rc;
1976
1977         if (!dt_object_exists(dt))
1978                 return -ENOENT;
1979
1980         LASSERT(handle != NULL);
1981         LASSERT(!dt_object_remote(dt));
1982         LASSERT(osd_invariant(obj));
1983
1984         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1985
1986         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1987                 struct osd_thread_info  *oti  = osd_oti_get(env);
1988                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1989                 struct lu_fid           *fid1 = &oti->oti_fid;
1990                 struct osd_inode_id     *id   = &oti->oti_id;
1991                 struct iam_path_descr   *ipd;
1992                 struct iam_container    *bag;
1993                 struct osd_thandle      *oh;
1994                 int                      rc;
1995
1996                 fid_cpu_to_be(fid1, fid0);
1997                 memset(id, 1, sizeof(*id));
1998                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1999                                   fid0)->oi_dir.od_container;
2000                 ipd = osd_idx_ipd_get(env, bag);
2001                 if (unlikely(ipd == NULL))
2002                         RETURN(-ENOMEM);
2003
2004                 oh = container_of0(handle, struct osd_thandle, ot_super);
2005                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2006                                 (const struct iam_rec *)id, ipd);
2007                 osd_ipd_put(env, bag, ipd);
2008                 return(rc > 0 ? 0 : rc);
2009         }
2010
2011         inode = obj->oo_inode;
2012
2013         rc = osd_quota_transfer(inode, attr);
2014         if (rc)
2015                 return rc;
2016
2017         spin_lock(&obj->oo_guard);
2018         rc = osd_inode_setattr(env, inode, attr);
2019         spin_unlock(&obj->oo_guard);
2020
2021         if (!rc)
2022                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2023
2024         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2025
2026         return rc;
2027 }
2028
2029 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2030                                            struct osd_object *obj,
2031                                            const char *name, const int namelen)
2032 {
2033         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2034 }
2035
2036 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2037                       umode_t mode, struct dt_allocation_hint *hint,
2038                       struct thandle *th)
2039 {
2040         int result;
2041         struct osd_device  *osd = osd_obj2dev(obj);
2042         struct osd_thandle *oth;
2043         struct dt_object   *parent = NULL;
2044         struct inode       *inode;
2045
2046         LINVRNT(osd_invariant(obj));
2047         LASSERT(obj->oo_inode == NULL);
2048         LASSERT(obj->oo_hl_head == NULL);
2049
2050         if (S_ISDIR(mode) && ldiskfs_pdo) {
2051                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2052                 if (obj->oo_hl_head == NULL)
2053                         return -ENOMEM;
2054         }
2055
2056         oth = container_of(th, struct osd_thandle, ot_super);
2057         LASSERT(oth->ot_handle->h_transaction != NULL);
2058
2059         if (hint != NULL && hint->dah_parent != NULL &&
2060             !dt_object_remote(hint->dah_parent))
2061                 parent = hint->dah_parent;
2062
2063         inode = ldiskfs_create_inode(oth->ot_handle,
2064                                      parent ? osd_dt_obj(parent)->oo_inode :
2065                                               osd_sb(osd)->s_root->d_inode,
2066                                      mode);
2067         if (!IS_ERR(inode)) {
2068                 /* Do not update file c/mtime in ldiskfs. */
2069                 inode->i_flags |= S_NOCMTIME;
2070
2071                 /* For new created object, it must be consistent,
2072                  * and it is unnecessary to scrub against it. */
2073                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2074
2075                 obj->oo_inode = inode;
2076                 result = 0;
2077         } else {
2078                 if (obj->oo_hl_head != NULL) {
2079                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2080                         obj->oo_hl_head = NULL;
2081                 }
2082                 result = PTR_ERR(inode);
2083         }
2084         LINVRNT(osd_invariant(obj));
2085         return result;
2086 }
2087
2088 enum {
2089         OSD_NAME_LEN = 255
2090 };
2091
2092 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2093                      struct lu_attr *attr,
2094                      struct dt_allocation_hint *hint,
2095                      struct dt_object_format *dof,
2096                      struct thandle *th)
2097 {
2098         int result;
2099         struct osd_thandle *oth;
2100         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2101
2102         LASSERT(S_ISDIR(attr->la_mode));
2103
2104         oth = container_of(th, struct osd_thandle, ot_super);
2105         LASSERT(oth->ot_handle->h_transaction != NULL);
2106         result = osd_mkfile(info, obj, mode, hint, th);
2107
2108         return result;
2109 }
2110
2111 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2112                         struct lu_attr *attr,
2113                         struct dt_allocation_hint *hint,
2114                         struct dt_object_format *dof,
2115                         struct thandle *th)
2116 {
2117         int result;
2118         struct osd_thandle *oth;
2119         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2120
2121         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2122
2123         LASSERT(S_ISREG(attr->la_mode));
2124
2125         oth = container_of(th, struct osd_thandle, ot_super);
2126         LASSERT(oth->ot_handle->h_transaction != NULL);
2127
2128         result = osd_mkfile(info, obj, mode, hint, th);
2129         if (result == 0) {
2130                 LASSERT(obj->oo_inode != NULL);
2131                 if (feat->dif_flags & DT_IND_VARKEY)
2132                         result = iam_lvar_create(obj->oo_inode,
2133                                                  feat->dif_keysize_max,
2134                                                  feat->dif_ptrsize,
2135                                                  feat->dif_recsize_max,
2136                                                  oth->ot_handle);
2137                 else
2138                         result = iam_lfix_create(obj->oo_inode,
2139                                                  feat->dif_keysize_max,
2140                                                  feat->dif_ptrsize,
2141                                                  feat->dif_recsize_max,
2142                                                  oth->ot_handle);
2143
2144         }
2145         return result;
2146 }
2147
2148 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2149                      struct lu_attr *attr,
2150                      struct dt_allocation_hint *hint,
2151                      struct dt_object_format *dof,
2152                      struct thandle *th)
2153 {
2154         LASSERT(S_ISREG(attr->la_mode));
2155         return osd_mkfile(info, obj, (attr->la_mode &
2156                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2157 }
2158
2159 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2160                      struct lu_attr *attr,
2161                      struct dt_allocation_hint *hint,
2162                      struct dt_object_format *dof,
2163                      struct thandle *th)
2164 {
2165         LASSERT(S_ISLNK(attr->la_mode));
2166         return osd_mkfile(info, obj, (attr->la_mode &
2167                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2168 }
2169
2170 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2171                      struct lu_attr *attr,
2172                      struct dt_allocation_hint *hint,
2173                      struct dt_object_format *dof,
2174                      struct thandle *th)
2175 {
2176         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2177         int result;
2178
2179         LINVRNT(osd_invariant(obj));
2180         LASSERT(obj->oo_inode == NULL);
2181         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2182                 S_ISFIFO(mode) || S_ISSOCK(mode));
2183
2184         result = osd_mkfile(info, obj, mode, hint, th);
2185         if (result == 0) {
2186                 LASSERT(obj->oo_inode != NULL);
2187                 /*
2188                  * This inode should be marked dirty for i_rdev.  Currently
2189                  * that is done in the osd_attr_init().
2190                  */
2191                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2192                                    attr->la_rdev);
2193         }
2194         LINVRNT(osd_invariant(obj));
2195         return result;
2196 }
2197
2198 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2199                               struct lu_attr *,
2200                               struct dt_allocation_hint *hint,
2201                               struct dt_object_format *dof,
2202                               struct thandle *);
2203
2204 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2205 {
2206         osd_obj_type_f result;
2207
2208         switch (type) {
2209         case DFT_DIR:
2210                 result = osd_mkdir;
2211                 break;
2212         case DFT_REGULAR:
2213                 result = osd_mkreg;
2214                 break;
2215         case DFT_SYM:
2216                 result = osd_mksym;
2217                 break;
2218         case DFT_NODE:
2219                 result = osd_mknod;
2220                 break;
2221         case DFT_INDEX:
2222                 result = osd_mk_index;
2223                 break;
2224
2225         default:
2226                 LBUG();
2227                 break;
2228         }
2229         return result;
2230 }
2231
2232
2233 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2234                         struct dt_object *parent, struct dt_object *child,
2235                         umode_t child_mode)
2236 {
2237         LASSERT(ah);
2238
2239         ah->dah_parent = parent;
2240         ah->dah_mode = child_mode;
2241 }
2242
2243 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2244                           struct lu_attr *attr, struct dt_object_format *dof)
2245 {
2246         struct inode   *inode = obj->oo_inode;
2247         __u64           valid = attr->la_valid;
2248         int             result;
2249
2250         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2251
2252         if (dof->dof_type != DFT_NODE)
2253                 attr->la_valid &= ~LA_RDEV;
2254         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2255                 attr->la_valid &= ~LA_ATIME;
2256         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2257                 attr->la_valid &= ~LA_CTIME;
2258         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2259                 attr->la_valid &= ~LA_MTIME;
2260
2261         result = osd_quota_transfer(inode, attr);
2262         if (result)
2263                 return;
2264
2265         if (attr->la_valid != 0) {
2266                 result = osd_inode_setattr(info->oti_env, inode, attr);
2267                 /*
2268                  * The osd_inode_setattr() should always succeed here.  The
2269                  * only error that could be returned is EDQUOT when we are
2270                  * trying to change the UID or GID of the inode. However, this
2271                  * should not happen since quota enforcement is no longer
2272                  * enabled on ldiskfs (lquota takes care of it).
2273                  */
2274                 LASSERTF(result == 0, "%d\n", result);
2275                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2276         }
2277
2278         attr->la_valid = valid;
2279 }
2280
2281 /**
2282  * Helper function for osd_object_create()
2283  *
2284  * \retval 0, on success
2285  */
2286 static int __osd_object_create(struct osd_thread_info *info,
2287                                struct osd_object *obj, struct lu_attr *attr,
2288                                struct dt_allocation_hint *hint,
2289                                struct dt_object_format *dof,
2290                                struct thandle *th)
2291 {
2292         int     result;
2293         __u32   umask;
2294
2295         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2296
2297         /* we drop umask so that permissions we pass are not affected */
2298         umask = current->fs->umask;
2299         current->fs->umask = 0;
2300
2301         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2302                                                   th);
2303         if (likely(obj->oo_inode != NULL)) {
2304                 LASSERT(obj->oo_inode->i_state & I_NEW);
2305
2306                 /* Unlock the inode before attr initialization to avoid
2307                  * unnecessary dqget operations. LU-6378 */
2308                 unlock_new_inode(obj->oo_inode);
2309         }
2310
2311         if (likely(result == 0)) {
2312                 osd_attr_init(info, obj, attr, dof);
2313                 osd_object_init0(obj);
2314         }
2315
2316         /* restore previous umask value */
2317         current->fs->umask = umask;
2318
2319         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2320
2321         return result;
2322 }
2323
2324 /**
2325  * Helper function for osd_object_create()
2326  *
2327  * \retval 0, on success
2328  */
2329 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2330                            const struct lu_fid *fid, struct thandle *th)
2331 {
2332         struct osd_thread_info *info = osd_oti_get(env);
2333         struct osd_inode_id    *id   = &info->oti_id;
2334         struct osd_device      *osd  = osd_obj2dev(obj);
2335         struct osd_thandle     *oh;
2336         int                     rc;
2337
2338         LASSERT(obj->oo_inode != NULL);
2339
2340         oh = container_of0(th, struct osd_thandle, ot_super);
2341         LASSERT(oh->ot_handle);
2342         osd_trans_exec_op(env, th, OSD_OT_INSERT);
2343
2344         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2345         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2346         osd_trans_exec_check(env, th, OSD_OT_INSERT);
2347
2348         return rc;
2349 }
2350
2351 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2352                    u64 seq, struct lu_seq_range *range)
2353 {
2354         struct seq_server_site  *ss = osd_seq_site(osd);
2355
2356         if (fid_seq_is_idif(seq)) {
2357                 fld_range_set_ost(range);
2358                 range->lsr_index = idif_ost_idx(seq);
2359                 return 0;
2360         }
2361
2362         if (!fid_seq_in_fldb(seq)) {
2363                 fld_range_set_mdt(range);
2364                 if (ss != NULL)
2365                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2366                          * at all */
2367                         range->lsr_index = ss->ss_node_id;
2368                 return 0;
2369         }
2370
2371         LASSERT(ss != NULL);
2372         fld_range_set_any(range);
2373         /* OSD will only do local fld lookup */
2374         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2375 }
2376
2377 /*
2378  * Concurrency: no external locking is necessary.
2379  */
2380 static int osd_declare_object_create(const struct lu_env *env,
2381                                      struct dt_object *dt,
2382                                      struct lu_attr *attr,
2383                                      struct dt_allocation_hint *hint,
2384                                      struct dt_object_format *dof,
2385                                      struct thandle *handle)
2386 {
2387         struct osd_thandle      *oh;
2388         int                      rc;
2389         ENTRY;
2390
2391         LASSERT(handle != NULL);
2392
2393         oh = container_of0(handle, struct osd_thandle, ot_super);
2394         LASSERT(oh->ot_handle == NULL);
2395
2396         /* EA object consumes more credits than regular object: osd_mk_index
2397          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
2398          * leaf_node, could involves the block, block bitmap, groups, GDT
2399          * change for each block, so add 4 * 2 credits in that case. */
2400         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2401                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
2402                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
2403         /* Reuse idle OI block may cause additional one OI block
2404          * to be changed. */
2405         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2406                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2407
2408         if (!attr)
2409                 RETURN(0);
2410
2411         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2412                                    osd_dt_obj(dt), false, NULL, false);
2413         if (rc != 0)
2414                 RETURN(rc);
2415
2416         RETURN(rc);
2417 }
2418
2419 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2420                              struct lu_attr *attr,
2421                              struct dt_allocation_hint *hint,
2422                              struct dt_object_format *dof, struct thandle *th)
2423 {
2424         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2425         struct osd_object       *obj    = osd_dt_obj(dt);
2426         struct osd_thread_info  *info   = osd_oti_get(env);
2427         int result;
2428         ENTRY;
2429
2430         if (dt_object_exists(dt))
2431                 return -EEXIST;
2432
2433         LINVRNT(osd_invariant(obj));
2434         LASSERT(!dt_object_remote(dt));
2435         LASSERT(osd_write_locked(env, obj));
2436         LASSERT(th != NULL);
2437
2438         if (unlikely(fid_is_acct(fid)))
2439                 /* Quota files can't be created from the kernel any more,
2440                  * 'tune2fs -O quota' will take care of creating them */
2441                 RETURN(-EPERM);
2442
2443         result = __osd_object_create(info, obj, attr, hint, dof, th);
2444         if (result == 0) {
2445                 result = __osd_oi_insert(env, obj, fid, th);
2446                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2447                         obj->oo_dt.do_body_ops = &osd_body_ops;
2448         }
2449         LASSERT(ergo(result == 0,
2450                 dt_object_exists(dt) && !dt_object_remote(dt)));
2451
2452         LASSERT(osd_invariant(obj));
2453         RETURN(result);
2454 }
2455
2456 /**
2457  * Called to destroy on-disk representation of the object
2458  *
2459  * Concurrency: must be locked
2460  */
2461 static int osd_declare_object_destroy(const struct lu_env *env,
2462                                       struct dt_object *dt,
2463                                       struct thandle *th)
2464 {
2465         struct osd_object  *obj = osd_dt_obj(dt);
2466         struct inode       *inode = obj->oo_inode;
2467         struct osd_thandle *oh;
2468         int                 rc;
2469         ENTRY;
2470
2471         if (inode == NULL)
2472                 RETURN(-ENOENT);
2473
2474         oh = container_of0(th, struct osd_thandle, ot_super);
2475         LASSERT(oh->ot_handle == NULL);
2476
2477         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2478                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2479         /* Recycle idle OI leaf may cause additional three OI blocks
2480          * to be changed. */
2481         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2482                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2483         /* one less inode */
2484         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2485                                    -1, oh, obj, false, NULL, false);
2486         if (rc)
2487                 RETURN(rc);
2488         /* data to be truncated */
2489         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2490                                    0, oh, obj, true, NULL, false);
2491         RETURN(rc);
2492 }
2493
2494 static int osd_object_destroy(const struct lu_env *env,
2495                               struct dt_object *dt,
2496                               struct thandle *th)
2497 {
2498         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2499         struct osd_object      *obj = osd_dt_obj(dt);
2500         struct inode           *inode = obj->oo_inode;
2501         struct osd_device      *osd = osd_obj2dev(obj);
2502         struct osd_thandle     *oh;
2503         int                     result;
2504         ENTRY;
2505
2506         oh = container_of0(th, struct osd_thandle, ot_super);
2507         LASSERT(oh->ot_handle);
2508         LASSERT(inode);
2509         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2510
2511         if (unlikely(fid_is_acct(fid)))
2512                 RETURN(-EPERM);
2513
2514         if (S_ISDIR(inode->i_mode)) {
2515                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2516                         inode->i_nlink == 2);
2517                 /* it will check/delete the inode from remote parent,
2518                  * how to optimize it? unlink performance impaction XXX */
2519                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2520                 if (result != 0 && result != -ENOENT) {
2521                         CERROR("%s: delete inode "DFID": rc = %d\n",
2522                                osd_name(osd), PFID(fid), result);
2523                 }
2524                 spin_lock(&obj->oo_guard);
2525                 clear_nlink(inode);
2526                 spin_unlock(&obj->oo_guard);
2527                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2528         }
2529
2530         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2531
2532         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
2533         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2534                                OI_CHECK_FLD);
2535
2536         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
2537         /* XXX: add to ext3 orphan list */
2538         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2539
2540         /* not needed in the cache anymore */
2541         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2542         obj->oo_destroyed = 1;
2543
2544         RETURN(0);
2545 }
2546
2547 /**
2548  * Put the fid into lustre_mdt_attrs, and then place the structure
2549  * inode's ea. This fid should not be altered during the life time
2550  * of the inode.
2551  *
2552  * \retval +ve, on success
2553  * \retval -ve, on error
2554  *
2555  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2556  */
2557 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2558                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2559 {
2560         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2561         int                      rc;
2562         ENTRY;
2563
2564         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2565                 RETURN(0);
2566
2567         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
2568                 rc = -ENOMEM;
2569
2570         lustre_lma_init(lma, fid, compat, incompat);
2571         lustre_lma_swab(lma);
2572
2573         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2574                              XATTR_CREATE);
2575         /* LMA may already exist, but we need to check that all the
2576          * desired compat/incompat flags have been added. */
2577         if (unlikely(rc == -EEXIST)) {
2578                 if (compat == 0 && incompat == 0)
2579                         RETURN(0);
2580
2581                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2582                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2583                                      LMA_OLD_SIZE);
2584                 if (rc <= 0)
2585                         RETURN(-EINVAL);
2586
2587                 lustre_lma_swab(lma);
2588                 if (!(~lma->lma_compat & compat) &&
2589                     !(~lma->lma_incompat & incompat))
2590                         RETURN(0);
2591
2592                 lma->lma_compat |= compat;
2593                 lma->lma_incompat |= incompat;
2594                 lustre_lma_swab(lma);
2595                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2596                                      sizeof(*lma), XATTR_REPLACE);
2597         }
2598
2599         RETURN(rc);
2600 }
2601
2602 /**
2603  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2604  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2605  * To have compatilibility with 1.8 ldiskfs driver we need to have
2606  * magic number at start of fid data.
2607  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2608  * its inmemory API.
2609  */
2610 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2611                                          const struct lu_fid *fid)
2612 {
2613         if (!fid_is_namespace_visible(fid) ||
2614             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2615                 param->edp_magic = 0;
2616                 return;
2617         }
2618
2619         param->edp_magic = LDISKFS_LUFID_MAGIC;
2620         param->edp_len =  sizeof(struct lu_fid) + 1;
2621         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2622 }
2623
2624 /**
2625  * Try to read the fid from inode ea into dt_rec.
2626  *
2627  * \param fid object fid.
2628  *
2629  * \retval 0 on success
2630  */
2631 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2632                           __u32 ino, struct lu_fid *fid,
2633                           struct osd_inode_id *id)
2634 {
2635         struct osd_thread_info *info  = osd_oti_get(env);
2636         struct inode           *inode;
2637         ENTRY;
2638
2639         osd_id_gen(id, ino, OSD_OII_NOGEN);
2640         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2641         if (IS_ERR(inode))
2642                 RETURN(PTR_ERR(inode));
2643
2644         iput(inode);
2645         RETURN(0);
2646 }
2647
2648 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2649                                         struct inode *dir,
2650                                         struct inode *parent_dir,
2651                                         const struct lu_fid *dot_fid,
2652                                         const struct lu_fid *dot_dot_fid,
2653                                         struct osd_thandle *oth)
2654 {
2655         struct ldiskfs_dentry_param *dot_ldp;
2656         struct ldiskfs_dentry_param *dot_dot_ldp;
2657         __u32 saved_nlink = dir->i_nlink;
2658         int rc;
2659
2660         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2661         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2662
2663         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2664         dot_ldp->edp_magic = 0;
2665
2666         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2667                                     dir, dot_ldp, dot_dot_ldp);
2668         /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
2669          * the subseqent ref_add() will increase the dir->i_nlink
2670          * as 3. That is incorrect for new created directory.
2671          *
2672          * It looks like hack, because we want to make the OSD API
2673          * to be order-independent for new created directory object
2674          * between dt_insert(..) and ref_add() operations.
2675          *
2676          * Here, we only restore the in-RAM dir-inode's nlink attr,
2677          * becuase if the nlink attr is not 2, then there will be
2678          * ref_add() called following the dt_insert(..), such call
2679          * will make both the in-RAM and on-disk dir-inode's nlink
2680          * attr to be set as 2. LU-7447 */
2681         set_nlink(dir, saved_nlink);
2682         return rc;
2683 }
2684
2685 /**
2686  * Create an local agent inode for remote entry
2687  */
2688 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2689                                                   struct osd_device *osd,
2690                                                   struct osd_object *pobj,
2691                                                   const struct lu_fid *fid,
2692                                                   __u32 type,
2693                                                   struct thandle *th)
2694 {
2695         struct osd_thread_info  *info = osd_oti_get(env);
2696         struct inode            *local;
2697         struct osd_thandle      *oh;
2698         int                     rc;
2699         ENTRY;
2700
2701         LASSERT(th);
2702         oh = container_of(th, struct osd_thandle, ot_super);
2703         LASSERT(oh->ot_handle->h_transaction != NULL);
2704
2705         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2706         if (IS_ERR(local)) {
2707                 CERROR("%s: create local error %d\n", osd_name(osd),
2708                        (int)PTR_ERR(local));
2709                 RETURN(local);
2710         }
2711
2712         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2713         unlock_new_inode(local);
2714
2715         /* Set special LMA flag for local agent inode */
2716         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2717         if (rc != 0) {
2718                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2719                        osd_name(osd), PFID(fid), rc);
2720                 RETURN(ERR_PTR(rc));
2721         }
2722
2723         if (!S_ISDIR(type))
2724                 RETURN(local);
2725
2726         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2727                                          lu_object_fid(&pobj->oo_dt.do_lu),
2728                                          fid, oh);
2729         if (rc != 0) {
2730                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2731                         osd_name(osd), PFID(fid), rc);
2732                 RETURN(ERR_PTR(rc));
2733         }
2734
2735         RETURN(local);
2736 }
2737
2738 /**
2739  * when direntry is deleted, we have to take care of possible agent inode
2740  * referenced by that. unfortunately we can't do this at that point:
2741  * iget() within a running transaction leads to deadlock and we better do
2742  * not call that every delete declaration to save performance. so we put
2743  * a potention agent inode on a list and process that once the transaction
2744  * is over. Notice it's not any worse in terms of real orphans as regular
2745  * object destroy doesn't put inodes on the on-disk orphan list. this should
2746  * be addressed separately
2747  */
2748 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
2749                                             struct osd_thandle *oh,
2750                                             __u32 ino)
2751 {
2752         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
2753         struct osd_obj_orphan *oor;
2754
2755         OBD_ALLOC_PTR(oor);
2756         if (oor == NULL)
2757                 return -ENOMEM;
2758
2759         oor->oor_ino = ino;
2760         oor->oor_env = (struct lu_env *)env;
2761         spin_lock(&osd->od_osfs_lock);
2762         list_add(&oor->oor_list, &osd->od_orphan_list);
2763         spin_unlock(&osd->od_osfs_lock);
2764
2765         oh->ot_remove_agents = 1;
2766
2767         return 0;
2768
2769 }
2770
2771 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
2772                                                 struct osd_device *osd)
2773 {
2774         struct osd_thread_info *info = osd_oti_get(env);
2775         struct osd_obj_orphan *oor, *tmp;
2776         struct osd_inode_id id;
2777         struct list_head list;
2778         struct inode *inode;
2779         struct lu_fid fid;
2780         handle_t *jh;
2781         __u32 ino;
2782
2783         INIT_LIST_HEAD(&list);
2784
2785         spin_lock(&osd->od_osfs_lock);
2786         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
2787                 if (oor->oor_env == env) {
2788                         list_del(&oor->oor_list);
2789                         list_add(&oor->oor_list, &list);
2790                 }
2791         }
2792         spin_unlock(&osd->od_osfs_lock);
2793
2794         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
2795
2796                 ino = oor->oor_ino;
2797
2798                 list_del(&oor->oor_list);
2799                 OBD_FREE_PTR(oor);
2800
2801                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
2802                 inode = osd_iget_fid(info, osd, &id, &fid);
2803                 if (IS_ERR(inode))
2804                         continue;
2805
2806                 if (!osd_remote_fid(env, osd, &fid)) {
2807                         iput(inode);
2808                         continue;
2809                 }
2810
2811                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
2812                 clear_nlink(inode);
2813                 mark_inode_dirty(inode);
2814                 ldiskfs_journal_stop(jh);
2815                 iput(inode);
2816         }
2817
2818         return 0;
2819 }
2820
2821 /**
2822  * OSD layer object create function for interoperability mode (b11826).
2823  * This is mostly similar to osd_object_create(). Only difference being, fid is
2824  * inserted into inode ea here.
2825  *
2826  * \retval   0, on success
2827  * \retval -ve, on error
2828  */
2829 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2830                                 struct lu_attr *attr,
2831                                 struct dt_allocation_hint *hint,
2832                                 struct dt_object_format *dof,
2833                                 struct thandle *th)
2834 {
2835         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2836         struct osd_object       *obj    = osd_dt_obj(dt);
2837         struct osd_thread_info  *info   = osd_oti_get(env);
2838         int                      result;
2839
2840         ENTRY;
2841
2842         if (dt_object_exists(dt))
2843                 RETURN(-EEXIST);
2844
2845         LASSERT(osd_invariant(obj));
2846         LASSERT(!dt_object_remote(dt));
2847         LASSERT(osd_write_locked(env, obj));
2848         LASSERT(th != NULL);
2849
2850         if (unlikely(fid_is_acct(fid)))
2851                 /* Quota files can't be created from the kernel any more,
2852                  * 'tune2fs -O quota' will take care of creating them */
2853                 RETURN(-EPERM);
2854
2855         result = __osd_object_create(info, obj, attr, hint, dof, th);
2856         if (result == 0) {
2857                 if (fid_is_idif(fid) &&
2858                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2859                         struct lu_fid *tfid = &info->oti_fid;
2860                         struct ost_id *oi   = &info->oti_ostid;
2861
2862                         fid_to_ostid(fid, oi);
2863                         ostid_to_fid(tfid, oi, 0);
2864                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2865                                                 LMAC_FID_ON_OST, 0);
2866                 } else {
2867                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2868                                 fid_is_on_ost(info, osd_obj2dev(obj),
2869                                               fid, OI_CHECK_FLD) ?
2870                                 LMAC_FID_ON_OST : 0, 0);
2871                 }
2872                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2873                         obj->oo_dt.do_body_ops = &osd_body_ops;
2874         }
2875
2876         if (result == 0)
2877                 result = __osd_oi_insert(env, obj, fid, th);
2878
2879         LASSERT(ergo(result == 0,
2880                      dt_object_exists(dt) && !dt_object_remote(dt)));
2881         LINVRNT(osd_invariant(obj));
2882         RETURN(result);
2883 }
2884
2885 static int osd_declare_object_ref_add(const struct lu_env *env,
2886                                       struct dt_object *dt,
2887                                       struct thandle *handle)
2888 {
2889         struct osd_thandle       *oh;
2890
2891         /* it's possible that object doesn't exist yet */
2892         LASSERT(handle != NULL);
2893
2894         oh = container_of0(handle, struct osd_thandle, ot_super);
2895         LASSERT(oh->ot_handle == NULL);
2896
2897         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2898                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2899
2900         return 0;
2901 }
2902
2903 /*
2904  * Concurrency: @dt is write locked.
2905  */
2906 static int osd_object_ref_add(const struct lu_env *env,
2907                               struct dt_object *dt, struct thandle *th)
2908 {
2909         struct osd_object  *obj = osd_dt_obj(dt);
2910         struct inode       *inode = obj->oo_inode;
2911         struct osd_thandle *oh;
2912         int                 rc = 0;
2913
2914         if (!dt_object_exists(dt))
2915                 return -ENOENT;
2916
2917         LINVRNT(osd_invariant(obj));
2918         LASSERT(!dt_object_remote(dt));
2919         LASSERT(osd_write_locked(env, obj));
2920         LASSERT(th != NULL);
2921
2922         oh = container_of0(th, struct osd_thandle, ot_super);
2923         LASSERT(oh->ot_handle != NULL);
2924
2925         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2926
2927         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2928                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2929         /*
2930          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2931          * (65000) subdirectories by storing "1" in i_nlink if the link count
2932          * would otherwise overflow. Directory tranversal tools understand
2933          * that (st_nlink == 1) indicates that the filesystem dose not track
2934          * hard links count on the directory, and will not abort subdirectory
2935          * scanning early once (st_nlink - 2) subdirs have been found.
2936          *
2937          * This also has to properly handle the case of inodes with nlink == 0
2938          * in case they are being linked into the PENDING directory
2939          */
2940         spin_lock(&obj->oo_guard);
2941         if (unlikely(inode->i_nlink == 0))
2942                 /* inc_nlink from 0 may cause WARN_ON */
2943                 set_nlink(inode, 1);
2944         else {
2945                 ldiskfs_inc_count(oh->ot_handle, inode);
2946                 if (!S_ISDIR(inode->i_mode))
2947                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2948         }
2949         spin_unlock(&obj->oo_guard);
2950
2951         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2952         LINVRNT(osd_invariant(obj));
2953
2954         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
2955
2956         return rc;
2957 }
2958
2959 static int osd_declare_object_ref_del(const struct lu_env *env,
2960                                       struct dt_object *dt,
2961                                       struct thandle *handle)
2962 {
2963         struct osd_thandle *oh;
2964
2965         if (!dt_object_exists(dt))
2966                 return -ENOENT;
2967
2968         LASSERT(!dt_object_remote(dt));
2969         LASSERT(handle != NULL);
2970
2971         oh = container_of0(handle, struct osd_thandle, ot_super);
2972         LASSERT(oh->ot_handle == NULL);
2973
2974         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2975                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2976
2977         return 0;
2978 }
2979
2980 /*
2981  * Concurrency: @dt is write locked.
2982  */
2983 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2984                               struct thandle *th)
2985 {
2986         struct osd_object       *obj = osd_dt_obj(dt);
2987         struct inode            *inode = obj->oo_inode;
2988         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2989         struct osd_thandle      *oh;
2990
2991         if (!dt_object_exists(dt))
2992                 return -ENOENT;
2993
2994         LINVRNT(osd_invariant(obj));
2995         LASSERT(!dt_object_remote(dt));
2996         LASSERT(osd_write_locked(env, obj));
2997         LASSERT(th != NULL);
2998
2999         oh = container_of0(th, struct osd_thandle, ot_super);
3000         LASSERT(oh->ot_handle != NULL);
3001
3002         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3003
3004         spin_lock(&obj->oo_guard);
3005         /* That can be result of upgrade from old Lustre version and
3006          * applied only to local files.  Just skip this ref_del call.
3007          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3008         if (inode->i_nlink == 0) {
3009                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3010                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3011                              ", maybe an upgraded file? (LU-3915)\n",
3012                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3013                 spin_unlock(&obj->oo_guard);
3014                 return 0;
3015         }
3016
3017         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3018                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3019
3020         ldiskfs_dec_count(oh->ot_handle, inode);
3021         spin_unlock(&obj->oo_guard);
3022
3023         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3024         LINVRNT(osd_invariant(obj));
3025
3026         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3027
3028         return 0;
3029 }
3030
3031 /*
3032  * Get the 64-bit version for an inode.
3033  */
3034 static int osd_object_version_get(const struct lu_env *env,
3035                                   struct dt_object *dt, dt_obj_version_t *ver)
3036 {
3037         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3038
3039         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
3040                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3041         *ver = LDISKFS_I(inode)->i_fs_version;
3042         return 0;
3043 }
3044
3045 /*
3046  * Concurrency: @dt is read locked.
3047  */
3048 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3049                          struct lu_buf *buf, const char *name)
3050 {
3051         struct osd_object      *obj    = osd_dt_obj(dt);
3052         struct inode           *inode  = obj->oo_inode;
3053         struct osd_thread_info *info   = osd_oti_get(env);
3054         struct dentry          *dentry = &info->oti_obj_dentry;
3055
3056         /* version get is not real XATTR but uses xattr API */
3057         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3058                 /* for version we are just using xattr API but change inode
3059                  * field instead */
3060                 if (buf->lb_len == 0)
3061                         return sizeof(dt_obj_version_t);
3062
3063                 if (buf->lb_len < sizeof(dt_obj_version_t))
3064                         return -ERANGE;
3065
3066                 osd_object_version_get(env, dt, buf->lb_buf);
3067
3068                 return sizeof(dt_obj_version_t);
3069         }
3070
3071         if (!dt_object_exists(dt))
3072                 return -ENOENT;
3073
3074         LASSERT(!dt_object_remote(dt));
3075         LASSERT(inode->i_op != NULL);
3076         LASSERT(inode->i_op->getxattr != NULL);
3077
3078         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3079 }
3080
3081
3082 static int osd_declare_xattr_set(const struct lu_env *env,
3083                                  struct dt_object *dt,
3084                                  const struct lu_buf *buf, const char *name,
3085                                  int fl, struct thandle *handle)
3086 {
3087         struct osd_thandle *oh;
3088         int credits;
3089         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3090
3091         LASSERT(handle != NULL);
3092
3093         oh = container_of0(handle, struct osd_thandle, ot_super);
3094         LASSERT(oh->ot_handle == NULL);
3095
3096         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3097                 /* For non-upgrading case, the LMA is set first and
3098                  * usually fit inode. But for upgrade case, the LMA
3099                  * may be in another separated EA block. */
3100                 if (!dt_object_exists(dt))
3101                         credits = 0;
3102                 else if (fl == LU_XATTR_REPLACE)
3103                         credits = 1;
3104                 else
3105                         goto upgrade;
3106         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3107                 credits = 1;
3108         } else {
3109 upgrade:
3110                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3111
3112                 if (buf != NULL) {
3113                         ssize_t buflen;
3114
3115                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3116                                 /* learn xattr size from osd_xattr_get if
3117                                    attribute has not been read yet */
3118                                 buflen = __osd_xattr_get(
3119                                     osd_dt_obj(dt)->oo_inode,
3120                                     &osd_oti_get(env)->oti_obj_dentry,
3121                                     name, NULL, 0);
3122                                 if (buflen < 0)
3123                                         buflen = 0;
3124                         } else {
3125                                 buflen = buf->lb_len;
3126                         }
3127
3128                         if (buflen > sb->s_blocksize) {
3129                                 credits += osd_calc_bkmap_credits(
3130                                     sb, NULL, 0, -1,
3131                                     (buflen + sb->s_blocksize - 1) >>
3132                                     sb->s_blocksize_bits);
3133                         }
3134                 }
3135                 /*
3136                  * xattr set may involve inode quota change, reserve credits for
3137                  * dquot_initialize()
3138                  */
3139                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3140         }
3141
3142         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3143
3144         return 0;
3145 }
3146
3147 /*
3148  * Set the 64-bit version for object
3149  */
3150 static void osd_object_version_set(const struct lu_env *env,
3151                                    struct dt_object *dt,
3152                                    dt_obj_version_t *new_version)
3153 {
3154         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3155
3156         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3157                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3158
3159         LDISKFS_I(inode)->i_fs_version = *new_version;
3160         /** Version is set after all inode operations are finished,
3161          *  so we should mark it dirty here */
3162         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3163 }
3164
3165 /*
3166  * Concurrency: @dt is write locked.
3167  */
3168 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3169                          const struct lu_buf *buf, const char *name, int fl,
3170                          struct thandle *handle)
3171 {
3172         struct osd_object      *obj      = osd_dt_obj(dt);
3173         struct inode           *inode    = obj->oo_inode;
3174         struct osd_thread_info *info     = osd_oti_get(env);
3175         int                     fs_flags = 0;
3176         int                     rc;
3177         ENTRY;
3178
3179         LASSERT(handle != NULL);
3180
3181         /* version set is not real XATTR */
3182         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3183                 /* for version we are just using xattr API but change inode
3184                  * field instead */
3185                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3186                 osd_object_version_set(env, dt, buf->lb_buf);
3187                 return sizeof(dt_obj_version_t);
3188         }
3189
3190         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3191                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3192
3193         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3194         if (fl & LU_XATTR_REPLACE)
3195                 fs_flags |= XATTR_REPLACE;
3196
3197         if (fl & LU_XATTR_CREATE)
3198                 fs_flags |= XATTR_CREATE;
3199
3200         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3201                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3202                 int                      rc;
3203
3204                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3205                 if (rc != 0)
3206                         RETURN(rc);
3207
3208                 lma->lma_incompat |= LMAI_STRIPED;
3209                 lustre_lma_swab(lma);
3210                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3211                                      sizeof(*lma), XATTR_REPLACE);
3212                 if (rc != 0)
3213                         RETURN(rc);
3214         }
3215
3216         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3217             strcmp(name, XATTR_NAME_LINK) == 0)
3218                 return -ENOSPC;
3219
3220         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3221                                fs_flags);
3222         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3223
3224         return rc;
3225 }
3226
3227 /*
3228  * Concurrency: @dt is read locked.
3229  */
3230 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3231                           const struct lu_buf *buf)
3232 {
3233         struct osd_object      *obj    = osd_dt_obj(dt);
3234         struct inode           *inode  = obj->oo_inode;
3235         struct osd_thread_info *info   = osd_oti_get(env);
3236         struct dentry          *dentry = &info->oti_obj_dentry;
3237
3238         if (!dt_object_exists(dt))
3239                 return -ENOENT;
3240
3241         LASSERT(!dt_object_remote(dt));
3242         LASSERT(inode->i_op != NULL);
3243         LASSERT(inode->i_op->listxattr != NULL);
3244
3245         dentry->d_inode = inode;
3246         dentry->d_sb = inode->i_sb;
3247         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3248 }
3249
3250 static int osd_declare_xattr_del(const struct lu_env *env,
3251                                  struct dt_object *dt, const char *name,
3252                                  struct thandle *handle)
3253 {
3254         struct osd_thandle *oh;
3255         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3256
3257         LASSERT(!dt_object_remote(dt));
3258         LASSERT(handle != NULL);
3259
3260         oh = container_of0(handle, struct osd_thandle, ot_super);
3261         LASSERT(oh->ot_handle == NULL);
3262
3263         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3264                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3265         /*
3266          * xattr del may involve inode quota change, reserve credits for
3267          * dquot_initialize()
3268          */
3269         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3270
3271         return 0;
3272 }
3273
3274 /*
3275  * Concurrency: @dt is write locked.
3276  */
3277 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3278                          const char *name, struct thandle *handle)
3279 {
3280         struct osd_object      *obj    = osd_dt_obj(dt);
3281         struct inode           *inode  = obj->oo_inode;
3282         struct osd_thread_info *info   = osd_oti_get(env);
3283         struct dentry          *dentry = &info->oti_obj_dentry;
3284         int                     rc;
3285
3286         if (!dt_object_exists(dt))
3287                 return -ENOENT;
3288
3289         LASSERT(!dt_object_remote(dt));
3290         LASSERT(inode->i_op != NULL);
3291         LASSERT(inode->i_op->removexattr != NULL);
3292         LASSERT(handle != NULL);
3293
3294         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3295
3296         ll_vfs_dq_init(inode);
3297         dentry->d_inode = inode;
3298         dentry->d_sb = inode->i_sb;
3299         rc = inode->i_op->removexattr(dentry, name);
3300         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3301         return rc;
3302 }
3303
3304 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3305                            __u64 start, __u64 end)
3306 {
3307         struct osd_object       *obj    = osd_dt_obj(dt);
3308         struct inode            *inode  = obj->oo_inode;
3309         struct osd_thread_info  *info   = osd_oti_get(env);
3310         struct dentry           *dentry = &info->oti_obj_dentry;
3311         struct file             *file   = &info->oti_file;
3312         int                     rc;
3313
3314         ENTRY;
3315
3316         dentry->d_inode = inode;
3317         dentry->d_sb = inode->i_sb;
3318         file->f_path.dentry = dentry;
3319         file->f_mapping = inode->i_mapping;
3320         file->f_op = inode->i_fop;
3321         set_file_inode(file, inode);
3322
3323         rc = ll_vfs_fsync_range(file, start, end, 0);
3324
3325         RETURN(rc);
3326 }
3327
3328 /*
3329  * Index operations.
3330  */
3331
3332 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3333                            const struct dt_index_features *feat)
3334 {
3335         struct iam_descr *descr;
3336
3337         if (osd_object_is_root(o))
3338                 return feat == &dt_directory_features;
3339
3340         LASSERT(o->oo_dir != NULL);
3341
3342         descr = o->oo_dir->od_container.ic_descr;
3343         if (feat == &dt_directory_features) {
3344                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3345                         return 1;
3346                 else
3347                         return 0;
3348         } else {
3349                 return
3350                         feat->dif_keysize_min <= descr->id_key_size &&
3351                         descr->id_key_size <= feat->dif_keysize_max &&
3352                         feat->dif_recsize_min <= descr->id_rec_size &&
3353                         descr->id_rec_size <= feat->dif_recsize_max &&
3354                         !(feat->dif_flags & (DT_IND_VARKEY |
3355                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3356                         ergo(feat->dif_flags & DT_IND_UPDATE,
3357                              1 /* XXX check that object (and file system) is
3358                                 * writable */);
3359         }
3360 }
3361
3362 static int osd_iam_container_init(const struct lu_env *env,
3363                                   struct osd_object *obj,
3364                                   struct osd_directory *dir)
3365 {
3366         struct iam_container *bag = &dir->od_container;
3367         int result;
3368
3369         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3370         if (result != 0)
3371                 return result;
3372
3373         result = iam_container_setup(bag);
3374         if (result == 0)
3375                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3376         else
3377                 iam_container_fini(bag);
3378
3379         return result;
3380 }
3381
3382
3383 /*
3384  * Concurrency: no external locking is necessary.
3385  */
3386 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3387                          const struct dt_index_features *feat)
3388 {
3389         int                      result;
3390         int                      skip_iam = 0;
3391         struct osd_object       *obj = osd_dt_obj(dt);
3392
3393         LINVRNT(osd_invariant(obj));
3394
3395         if (osd_object_is_root(obj)) {
3396                 dt->do_index_ops = &osd_index_ea_ops;
3397                 result = 0;
3398         } else if (feat == &dt_directory_features) {
3399                 dt->do_index_ops = &osd_index_ea_ops;
3400                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3401                         result = 0;
3402                 else
3403                         result = -ENOTDIR;
3404                 skip_iam = 1;
3405         } else if (unlikely(feat == &dt_otable_features)) {
3406                 dt->do_index_ops = &osd_otable_ops;
3407                 return 0;
3408         } else if (unlikely(feat == &dt_acct_features)) {
3409                 dt->do_index_ops = &osd_acct_index_ops;
3410                 result = 0;
3411                 skip_iam = 1;
3412         } else if (!osd_has_index(obj)) {
3413                 struct osd_directory *dir;
3414
3415                 OBD_ALLOC_PTR(dir);
3416                 if (dir != NULL) {
3417
3418                         spin_lock(&obj->oo_guard);
3419                         if (obj->oo_dir == NULL)
3420                                 obj->oo_dir = dir;
3421                         else
3422                                 /*
3423                                  * Concurrent thread allocated container data.
3424                                  */
3425                                 OBD_FREE_PTR(dir);
3426                         spin_unlock(&obj->oo_guard);
3427                         /*
3428                          * Now, that we have container data, serialize its
3429                          * initialization.
3430                          */
3431                         down_write(&obj->oo_ext_idx_sem);
3432                         /*
3433                          * recheck under lock.
3434                          */
3435                         if (!osd_has_index(obj))
3436                                 result = osd_iam_container_init(env, obj,
3437                                                                 obj->oo_dir);
3438                         else
3439                                 result = 0;
3440                         up_write(&obj->oo_ext_idx_sem);
3441                 } else {
3442                         result = -ENOMEM;
3443                 }
3444         } else {
3445                 result = 0;
3446         }
3447
3448         if (result == 0 && skip_iam == 0) {
3449                 if (!osd_iam_index_probe(env, obj, feat))
3450                         result = -ENOTDIR;
3451         }
3452         LINVRNT(osd_invariant(obj));
3453
3454         if (result == 0 && feat == &dt_quota_glb_features &&
3455             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3456                 result = osd_quota_migration(env, dt);
3457
3458         return result;
3459 }
3460
3461 static int osd_otable_it_attr_get(const struct lu_env *env,
3462                                  struct dt_object *dt,
3463                                  struct lu_attr *attr)
3464 {
3465         attr->la_valid = 0;
3466         return 0;
3467 }
3468
3469 static const struct dt_object_operations osd_obj_ops = {
3470         .do_read_lock         = osd_object_read_lock,
3471         .do_write_lock        = osd_object_write_lock,
3472         .do_read_unlock       = osd_object_read_unlock,
3473         .do_write_unlock      = osd_object_write_unlock,
3474         .do_write_locked      = osd_object_write_locked,
3475         .do_attr_get          = osd_attr_get,
3476         .do_declare_attr_set  = osd_declare_attr_set,
3477         .do_attr_set          = osd_attr_set,
3478         .do_ah_init           = osd_ah_init,
3479         .do_declare_create    = osd_declare_object_create,
3480         .do_create            = osd_object_create,
3481         .do_declare_destroy   = osd_declare_object_destroy,
3482         .do_destroy           = osd_object_destroy,
3483         .do_index_try         = osd_index_try,
3484         .do_declare_ref_add   = osd_declare_object_ref_add,
3485         .do_ref_add           = osd_object_ref_add,
3486         .do_declare_ref_del   = osd_declare_object_ref_del,
3487         .do_ref_del           = osd_object_ref_del,
3488         .do_xattr_get         = osd_xattr_get,
3489         .do_declare_xattr_set = osd_declare_xattr_set,
3490         .do_xattr_set         = osd_xattr_set,
3491         .do_declare_xattr_del = osd_declare_xattr_del,
3492         .do_xattr_del         = osd_xattr_del,
3493         .do_xattr_list        = osd_xattr_list,
3494         .do_object_sync       = osd_object_sync,
3495 };
3496
3497 /**
3498  * dt_object_operations for interoperability mode
3499  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3500  */
3501 static const struct dt_object_operations osd_obj_ea_ops = {
3502         .do_read_lock         = osd_object_read_lock,
3503         .do_write_lock        = osd_object_write_lock,
3504         .do_read_unlock       = osd_object_read_unlock,
3505         .do_write_unlock      = osd_object_write_unlock,
3506         .do_write_locked      = osd_object_write_locked,
3507         .do_attr_get          = osd_attr_get,
3508         .do_declare_attr_set  = osd_declare_attr_set,
3509         .do_attr_set          = osd_attr_set,
3510         .do_ah_init           = osd_ah_init,
3511         .do_declare_create    = osd_declare_object_create,
3512         .do_create            = osd_object_ea_create,
3513         .do_declare_destroy   = osd_declare_object_destroy,
3514         .do_destroy           = osd_object_destroy,
3515         .do_index_try         = osd_index_try,
3516         .do_declare_ref_add   = osd_declare_object_ref_add,
3517         .do_ref_add           = osd_object_ref_add,
3518         .do_declare_ref_del   = osd_declare_object_ref_del,
3519         .do_ref_del           = osd_object_ref_del,
3520         .do_xattr_get         = osd_xattr_get,
3521         .do_declare_xattr_set = osd_declare_xattr_set,
3522         .do_xattr_set         = osd_xattr_set,
3523         .do_declare_xattr_del = osd_declare_xattr_del,
3524         .do_xattr_del         = osd_xattr_del,
3525         .do_xattr_list        = osd_xattr_list,
3526         .do_object_sync       = osd_object_sync,
3527 };
3528
3529 static const struct dt_object_operations osd_obj_otable_it_ops = {
3530         .do_attr_get    = osd_otable_it_attr_get,
3531         .do_index_try   = osd_index_try,
3532 };
3533
3534 static int osd_index_declare_iam_delete(const struct lu_env *env,
3535                                         struct dt_object *dt,
3536                                         const struct dt_key *key,
3537                                         struct thandle *handle)
3538 {
3539         struct osd_thandle    *oh;
3540
3541         oh = container_of0(handle, struct osd_thandle, ot_super);
3542         LASSERT(oh->ot_handle == NULL);
3543
3544         /* Recycle  may cause additional three blocks to be changed. */
3545         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3546                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3547
3548         return 0;
3549 }
3550
3551 /**
3552  *      delete a (key, value) pair from index \a dt specified by \a key
3553  *
3554  *      \param  dt      osd index object
3555  *      \param  key     key for index
3556  *      \param  rec     record reference
3557  *      \param  handle  transaction handler
3558  *
3559  *      \retval  0  success
3560  *      \retval -ve   failure
3561  */
3562 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3563                                 const struct dt_key *key,
3564                                 struct thandle *handle)
3565 {
3566         struct osd_thread_info *oti = osd_oti_get(env);
3567         struct osd_object      *obj = osd_dt_obj(dt);
3568         struct osd_thandle     *oh;
3569         struct iam_path_descr  *ipd;
3570         struct iam_container   *bag = &obj->oo_dir->od_container;
3571         int                     rc;
3572         ENTRY;
3573
3574         if (!dt_object_exists(dt))
3575                 RETURN(-ENOENT);
3576
3577         LINVRNT(osd_invariant(obj));
3578         LASSERT(!dt_object_remote(dt));
3579         LASSERT(bag->ic_object == obj->oo_inode);
3580         LASSERT(handle != NULL);
3581
3582         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3583
3584         ipd = osd_idx_ipd_get(env, bag);
3585         if (unlikely(ipd == NULL))
3586                 RETURN(-ENOMEM);
3587
3588         oh = container_of0(handle, struct osd_thandle, ot_super);
3589         LASSERT(oh->ot_handle != NULL);
3590         LASSERT(oh->ot_handle->h_transaction != NULL);
3591
3592         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3593                 /* swab quota uid/gid provided by caller */
3594                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3595                 key = (const struct dt_key *)&oti->oti_quota_id;
3596         }
3597
3598         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3599         osd_ipd_put(env, bag, ipd);
3600         LINVRNT(osd_invariant(obj));
3601         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3602         RETURN(rc);
3603 }
3604
3605 static int osd_index_declare_ea_delete(const struct lu_env *env,
3606                                        struct dt_object *dt,
3607                                        const struct dt_key *key,
3608                                        struct thandle *handle)
3609 {
3610         struct osd_thandle *oh;
3611         struct inode       *inode;
3612         int                 rc;
3613         ENTRY;
3614
3615         LASSERT(!dt_object_remote(dt));
3616         LASSERT(handle != NULL);
3617
3618         oh = container_of0(handle, struct osd_thandle, ot_super);
3619         LASSERT(oh->ot_handle == NULL);
3620
3621         /* due to DNE we may need to remove an agent inode */
3622         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3623                              osd_dto_credits_noquota[DTO_INDEX_DELETE] +
3624                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3625
3626         inode = osd_dt_obj(dt)->oo_inode;
3627         if (inode == NULL)
3628                 RETURN(-ENOENT);
3629
3630         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3631                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3632         RETURN(rc);
3633 }
3634
3635 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3636                                           struct dt_rec *fid)
3637 {
3638         struct osd_fid_pack *rec;
3639         int                  rc = -ENODATA;
3640
3641         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3642                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3643                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3644                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
3645                         rc = -EINVAL;
3646         }
3647         return rc;
3648 }
3649
3650 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3651                           const struct lu_fid *fid)
3652 {
3653         struct seq_server_site  *ss = osd_seq_site(osd);
3654         ENTRY;
3655
3656         /* FID seqs not in FLDB, must be local seq */
3657         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3658                 RETURN(0);
3659
3660         /* If FLD is not being initialized yet, it only happens during the
3661          * initialization, likely during mgs initialization, and we assume
3662          * this is local FID. */
3663         if (ss == NULL || ss->ss_server_fld == NULL)
3664                 RETURN(0);
3665
3666         /* Only check the local FLDB here */
3667         if (osd_seq_exists(env, osd, fid_seq(fid)))
3668                 RETURN(0);
3669
3670         RETURN(1);
3671 }
3672
3673 /**
3674  * Index delete function for interoperability mode (b11826).
3675  * It will remove the directory entry added by osd_index_ea_insert().
3676  * This entry is needed to maintain name->fid mapping.
3677  *
3678  * \param key,  key i.e. file entry to be deleted
3679  *
3680  * \retval   0, on success
3681  * \retval -ve, on error
3682  */
3683 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3684                                const struct dt_key *key, struct thandle *handle)
3685 {
3686         struct osd_object          *obj = osd_dt_obj(dt);
3687         struct inode               *dir = obj->oo_inode;
3688         struct dentry              *dentry;
3689         struct osd_thandle         *oh;
3690         struct ldiskfs_dir_entry_2 *de = NULL;
3691         struct buffer_head         *bh;
3692         struct htree_lock          *hlock = NULL;
3693         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3694         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3695         int                        rc;
3696         ENTRY;
3697
3698         if (!dt_object_exists(dt))
3699                 RETURN(-ENOENT);
3700
3701         LINVRNT(osd_invariant(obj));
3702         LASSERT(!dt_object_remote(dt));
3703         LASSERT(handle != NULL);
3704
3705         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3706
3707         oh = container_of(handle, struct osd_thandle, ot_super);
3708         LASSERT(oh->ot_handle != NULL);
3709         LASSERT(oh->ot_handle->h_transaction != NULL);
3710
3711         ll_vfs_dq_init(dir);
3712         dentry = osd_child_dentry_get(env, obj,
3713                                       (char *)key, strlen((char *)key));
3714
3715         if (obj->oo_hl_head != NULL) {
3716                 hlock = osd_oti_get(env)->oti_hlock;
3717                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3718                                    dir, LDISKFS_HLOCK_DEL);
3719         } else {
3720                 down_write(&obj->oo_ext_idx_sem);
3721         }
3722
3723         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3724         if (bh) {
3725                 /* If this is not the ".." entry, it might be a remote DNE
3726                  * entry and  we need to check if the FID is for a remote
3727                  * MDT.  If the FID is  not in the directory entry (e.g.
3728                  * upgraded 1.8 filesystem without dirdata enabled) then
3729                  * we need to get the FID from the LMA. For a remote directory
3730                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3731                  * case.
3732                  *
3733                  * Delete the entry before the agent inode in order to
3734                  * simplify error handling.  At worst an error after deleting
3735                  * the entry first might leak the agent inode afterward. The
3736                  * reverse would need filesystem abort in case of error deleting
3737                  * the entry after the agent had been removed, or leave a
3738                  * dangling entry pointing at a random inode. */
3739                 if (strcmp((char *)key, dotdot) != 0) {
3740                         LASSERT(de != NULL);
3741                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3742                         if (rc == -ENODATA) {
3743                                 /* can't get FID, postpone to the end of the
3744                                  * transaction when iget() is safe */
3745                                 osd_schedule_agent_inode_removal(env, oh,
3746                                                 le32_to_cpu(de->inode));
3747                         } else if (rc == 0 &&
3748                                    unlikely(osd_remote_fid(env, osd, fid))) {
3749                                 osd_schedule_agent_inode_removal(env, oh,
3750                                                 le32_to_cpu(de->inode));
3751                         }
3752                 }
3753                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3754                 brelse(bh);
3755         } else {
3756                 rc = -ENOENT;
3757         }
3758         if (hlock != NULL)
3759                 ldiskfs_htree_unlock(hlock);
3760         else
3761                 up_write(&obj->oo_ext_idx_sem);
3762
3763         if (rc != 0)
3764                 GOTO(out, rc);
3765
3766         /* For inode on the remote MDT, .. will point to
3767          * /Agent directory, Check whether it needs to delete
3768          * from agent directory */
3769         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3770                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3771                                                    oh);
3772                 if (rc != 0 && rc != -ENOENT) {
3773                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3774                                osd_name(osd), PFID(fid), rc);
3775                 }
3776
3777                 if (rc == -ENOENT)
3778                         rc = 0;
3779
3780                 GOTO(out, rc);
3781         }
3782 out:
3783
3784         LASSERT(osd_invariant(obj));
3785         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3786         RETURN(rc);
3787 }
3788
3789 /**
3790  *      Lookup index for \a key and copy record to \a rec.
3791  *
3792  *      \param  dt      osd index object
3793  *      \param  key     key for index
3794  *      \param  rec     record reference
3795  *
3796  *      \retval  +ve  success : exact mach
3797  *      \retval  0    return record with key not greater than \a key
3798  *      \retval -ve   failure
3799  */
3800 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3801                                 struct dt_rec *rec, const struct dt_key *key)
3802 {
3803         struct osd_object      *obj = osd_dt_obj(dt);
3804         struct iam_path_descr  *ipd;
3805         struct iam_container   *bag = &obj->oo_dir->od_container;
3806         struct osd_thread_info *oti = osd_oti_get(env);
3807         struct iam_iterator    *it = &oti->oti_idx_it;
3808         struct iam_rec         *iam_rec;
3809         int                     rc;
3810         ENTRY;
3811
3812         if (!dt_object_exists(dt))
3813                 RETURN(-ENOENT);
3814
3815         LASSERT(osd_invariant(obj));
3816         LASSERT(!dt_object_remote(dt));
3817         LASSERT(bag->ic_object == obj->oo_inode);
3818
3819         ipd = osd_idx_ipd_get(env, bag);
3820         if (IS_ERR(ipd))
3821                 RETURN(-ENOMEM);
3822
3823         /* got ipd now we can start iterator. */
3824         iam_it_init(it, bag, 0, ipd);
3825
3826         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3827                 /* swab quota uid/gid provided by caller */
3828                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3829                 key = (const struct dt_key *)&oti->oti_quota_id;
3830         }
3831
3832         rc = iam_it_get(it, (struct iam_key *)key);
3833         if (rc >= 0) {
3834                 if (S_ISDIR(obj->oo_inode->i_mode))
3835                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3836                 else
3837                         iam_rec = (struct iam_rec *) rec;
3838
3839                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3840
3841                 if (S_ISDIR(obj->oo_inode->i_mode))
3842                         osd_fid_unpack((struct lu_fid *) rec,
3843                                        (struct osd_fid_pack *)iam_rec);
3844                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3845                         osd_quota_unpack(obj, rec);
3846         }
3847
3848         iam_it_put(it);
3849         iam_it_fini(it);
3850         osd_ipd_put(env, bag, ipd);
3851
3852         LINVRNT(osd_invariant(obj));
3853
3854         RETURN(rc);
3855 }
3856
3857 static int osd_index_declare_iam_insert(const struct lu_env *env,
3858                                         struct dt_object *dt,
3859                                         const struct dt_rec *rec,
3860                                         const struct dt_key *key,
3861                                         struct thandle *handle)
3862 {
3863         struct osd_thandle *oh;
3864
3865         LASSERT(handle != NULL);
3866
3867         oh = container_of0(handle, struct osd_thandle, ot_super);
3868         LASSERT(oh->ot_handle == NULL);
3869
3870         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3871                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3872
3873         return 0;
3874 }
3875
3876 /**
3877  *      Inserts (key, value) pair in \a dt index object.
3878  *
3879  *      \param  dt      osd index object
3880  *      \param  key     key for index
3881  *      \param  rec     record reference
3882  *      \param  th      transaction handler
3883  *
3884  *      \retval  0  success
3885  *      \retval -ve failure
3886  */
3887 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3888                                 const struct dt_rec *rec,
3889                                 const struct dt_key *key, struct thandle *th,
3890                                 int ignore_quota)
3891 {
3892         struct osd_object     *obj = osd_dt_obj(dt);
3893         struct iam_path_descr *ipd;
3894         struct osd_thandle    *oh;
3895         struct iam_container  *bag;
3896         struct osd_thread_info *oti = osd_oti_get(env);
3897         struct iam_rec         *iam_rec;
3898         int                     rc;
3899         ENTRY;
3900
3901         if (!dt_object_exists(dt))
3902                 RETURN(-ENOENT);
3903
3904         LINVRNT(osd_invariant(obj));
3905         LASSERT(!dt_object_remote(dt));
3906
3907         bag = &obj->oo_dir->od_container;
3908         LASSERT(bag->ic_object == obj->oo_inode);
3909         LASSERT(th != NULL);
3910
3911         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3912
3913         ipd = osd_idx_ipd_get(env, bag);
3914         if (unlikely(ipd == NULL))
3915                 RETURN(-ENOMEM);
3916
3917         oh = container_of0(th, struct osd_thandle, ot_super);
3918         LASSERT(oh->ot_handle != NULL);
3919         LASSERT(oh->ot_handle->h_transaction != NULL);
3920         if (S_ISDIR(obj->oo_inode->i_mode)) {
3921                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3922                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3923         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3924                 /* pack quota uid/gid */
3925                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3926                 key = (const struct dt_key *)&oti->oti_quota_id;
3927                 /* pack quota record */
3928                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3929                 iam_rec = (struct iam_rec *)rec;
3930         } else {
3931                 iam_rec = (struct iam_rec *)rec;
3932         }
3933
3934         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3935                         iam_rec, ipd);
3936         osd_ipd_put(env, bag, ipd);
3937         LINVRNT(osd_invariant(obj));
3938         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3939         RETURN(rc);
3940 }
3941
3942 /**
3943  * Calls ldiskfs_add_entry() to add directory entry
3944  * into the directory. This is required for
3945  * interoperability mode (b11826)
3946  *
3947  * \retval   0, on success
3948  * \retval -ve, on error
3949  */
3950 static int __osd_ea_add_rec(struct osd_thread_info *info,
3951                             struct osd_object *pobj, struct inode  *cinode,
3952                             const char *name, const struct lu_fid *fid,
3953                             struct htree_lock *hlock, struct thandle *th)
3954 {
3955         struct ldiskfs_dentry_param *ldp;
3956         struct dentry               *child;
3957         struct osd_thandle          *oth;
3958         int                          rc;
3959
3960         oth = container_of(th, struct osd_thandle, ot_super);
3961         LASSERT(oth->ot_handle != NULL);
3962         LASSERT(oth->ot_handle->h_transaction != NULL);
3963         LASSERT(pobj->oo_inode);
3964
3965         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3966         if (unlikely(pobj->oo_inode ==
3967                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3968                 ldp->edp_magic = 0;
3969         else
3970                 osd_get_ldiskfs_dirent_param(ldp, fid);
3971         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3972         child->d_fsdata = (void *)ldp;
3973         ll_vfs_dq_init(pobj->oo_inode);
3974         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3975         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3976                 struct ldiskfs_dir_entry_2      *de;
3977                 struct buffer_head              *bh;
3978                 int                              rc1;
3979
3980                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3981                                             NULL, hlock);
3982                 if (bh != NULL) {
3983                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3984                                                                bh);
3985                         if (rc1 == 0) {
3986                                 if (S_ISDIR(cinode->i_mode))
3987                                         de->file_type = LDISKFS_DIRENT_LUFID |
3988                                                         LDISKFS_FT_REG_FILE;
3989                                 else
3990                                         de->file_type = LDISKFS_DIRENT_LUFID |
3991                                                         LDISKFS_FT_DIR;
3992                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
3993                                                               NULL, bh);
3994                                 brelse(bh);
3995                         }
3996                 }
3997         }
3998
3999         RETURN(rc);
4000 }
4001
4002 /**
4003  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4004  * into the directory.Also sets flags into osd object to
4005  * indicate dot and dotdot are created. This is required for
4006  * interoperability mode (b11826)
4007  *
4008  * \param dir   directory for dot and dotdot fixup.
4009  * \param obj   child object for linking
4010  *
4011  * \retval   0, on success
4012  * \retval -ve, on error
4013  */
4014 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4015                               struct osd_object *dir,
4016                               struct inode *parent_dir, const char *name,
4017                               const struct lu_fid *dot_fid,
4018                               const struct lu_fid *dot_dot_fid,
4019                               struct thandle *th)
4020 {
4021         struct inode                *inode = dir->oo_inode;
4022         struct osd_thandle          *oth;
4023         int result = 0;
4024
4025         oth = container_of(th, struct osd_thandle, ot_super);
4026         LASSERT(oth->ot_handle->h_transaction != NULL);
4027         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4028
4029         if (strcmp(name, dot) == 0) {
4030                 if (dir->oo_compat_dot_created) {
4031                         result = -EEXIST;
4032                 } else {
4033                         LASSERT(inode == parent_dir);
4034                         dir->oo_compat_dot_created = 1;
4035                         result = 0;
4036                 }
4037         } else if (strcmp(name, dotdot) == 0) {
4038                 if (!dir->oo_compat_dot_created)
4039                         return -EINVAL;
4040                 /* in case of rename, dotdot is already created */
4041                 if (dir->oo_compat_dotdot_created) {
4042                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4043                                                 dot_dot_fid, NULL, th);
4044                 }
4045
4046                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4047                         struct lu_fid tfid = *dot_dot_fid;
4048
4049                         tfid.f_oid--;
4050                         result = osd_add_dot_dotdot_internal(info,
4051                                         dir->oo_inode, parent_dir, dot_fid,
4052                                         &tfid, oth);
4053                 } else {
4054                         result = osd_add_dot_dotdot_internal(info,
4055                                         dir->oo_inode, parent_dir, dot_fid,
4056                                         dot_dot_fid, oth);
4057                 }
4058
4059                 if (result == 0)
4060                         dir->oo_compat_dotdot_created = 1;
4061         }
4062
4063         return result;
4064 }
4065
4066
4067 /**
4068  * It will call the appropriate osd_add* function and return the
4069  * value, return by respective functions.
4070  */
4071 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4072                           struct inode *cinode, const char *name,
4073                           const struct lu_fid *fid, struct thandle *th)
4074 {
4075         struct osd_thread_info *info   = osd_oti_get(env);
4076         struct htree_lock      *hlock;
4077         int                     rc;
4078
4079         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4080
4081         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4082                                                    name[2] =='\0'))) {
4083                 if (hlock != NULL) {
4084                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4085                                            pobj->oo_inode, 0);
4086                 } else {
4087                         down_write(&pobj->oo_ext_idx_sem);
4088                 }
4089
4090                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4091                                         lu_object_fid(&pobj->oo_dt.do_lu),
4092                                         fid, th);
4093         } else {
4094                 if (hlock != NULL) {
4095                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4096                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4097                 } else {
4098                         down_write(&pobj->oo_ext_idx_sem);
4099                 }
4100
4101                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4102                         struct lu_fid *tfid = &info->oti_fid;
4103
4104                         *tfid = *fid;
4105                         tfid->f_ver = ~0;
4106                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4107                                               tfid, hlock, th);
4108                 } else {
4109                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4110                                               hlock, th);
4111                 }
4112         }
4113         if (hlock != NULL)
4114                 ldiskfs_htree_unlock(hlock);
4115         else
4116                 up_write(&pobj->oo_ext_idx_sem);
4117
4118         return rc;
4119 }
4120
4121 static int
4122 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4123                       struct osd_idmap_cache *oic)
4124 {
4125         struct osd_scrub    *scrub = &dev->od_scrub;
4126         struct lu_fid       *fid   = &oic->oic_fid;
4127         struct osd_inode_id *id    = &oti->oti_id;
4128         int                  once  = 0;
4129         int                  rc;
4130         ENTRY;
4131
4132         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4133                 RETURN(0);
4134
4135         if (scrub->os_pos_current > id->oii_ino)
4136                 RETURN(0);
4137
4138 again:
4139         rc = osd_oi_lookup(oti, dev, fid, id, 0);
4140         if (rc == -ENOENT) {
4141                 struct inode *inode;
4142
4143                 *id = oic->oic_lid;
4144                 inode = osd_iget(oti, dev, &oic->oic_lid);
4145
4146                 /* The inode has been removed (by race maybe). */
4147                 if (IS_ERR(inode)) {
4148                         rc = PTR_ERR(inode);
4149
4150                         RETURN(rc == -ESTALE ? -ENOENT : rc);
4151                 }
4152
4153                 iput(inode);
4154                 /* The OI mapping is lost. */
4155                 if (id->oii_gen != OSD_OII_NOGEN)
4156                         goto trigger;
4157
4158                 /* The inode may has been reused by others, we do not know,
4159                  * leave it to be handled by subsequent osd_fid_lookup(). */
4160                 RETURN(0);
4161         } else if (rc != 0 || osd_id_eq(id, &oic->oic_lid)) {
4162                 RETURN(rc);
4163         }
4164
4165 trigger:
4166         if (thread_is_running(&scrub->os_thread)) {
4167                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4168                 /* There is race condition between osd_oi_lookup and OI scrub.
4169                  * The OI scrub finished just after osd_oi_lookup() failure.
4170                  * Under such case, it is unnecessary to trigger OI scrub again,
4171                  * but try to call osd_oi_lookup() again. */
4172                 if (unlikely(rc == -EAGAIN))
4173                         goto again;
4174
4175                 RETURN(0);
4176         }
4177
4178         if (!dev->od_noscrub && ++once == 1) {
4179                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4180                                      SS_CLEAR_FAILOUT);
4181                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
4182                        "for "DFID", rc = %d [2]\n",
4183                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4184                        PFID(fid), rc);
4185                 if (rc == 0 || rc == -EALREADY)
4186                         goto again;
4187         }
4188
4189         RETURN(0);
4190 }
4191
4192 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4193                                struct osd_device *dev,
4194                                struct osd_idmap_cache *oic,
4195                                struct lu_fid *fid, __u32 ino)
4196 {
4197         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4198         struct inode            *inode;
4199         int                      rc;
4200
4201         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4202         inode = osd_iget(oti, dev, &oic->oic_lid);
4203         if (IS_ERR(inode)) {
4204                 fid_zero(&oic->oic_fid);
4205                 return PTR_ERR(inode);
4206         }
4207
4208         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4209         iput(inode);
4210         if (rc != 0)
4211                 fid_zero(&oic->oic_fid);
4212         else
4213                 *fid = oic->oic_fid = lma->lma_self_fid;
4214         return rc;
4215 }
4216
4217 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4218                       struct osd_inode_id *id, const struct lu_fid *fid)
4219 {
4220         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4221                id->oii_ino, id->oii_gen, info);
4222         info->oti_cache.oic_lid = *id;
4223         info->oti_cache.oic_fid = *fid;
4224         info->oti_cache.oic_dev = osd;
4225 }
4226
4227 /**
4228  * Get parent FID from the linkEA.
4229  *
4230  * For a directory which parent resides on remote MDT, to satisfy the
4231  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4232  * the other hand, to make the lookup(..) on the directory can return
4233  * the real parent FID, we append the real parent FID after its ".."
4234  * name entry in the /REMOTE_PARENT_DIR.
4235  *
4236  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4237  * backup. So after the restore, we cannot get the right parent FID from
4238  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4239  * we have stored the real parent FID in the directory object's linkEA,
4240  * we can parse the linkEA for the real parent FID.
4241  *
4242  * \param[in] env       pointer to the thread context
4243  * \param[in] obj       pointer to the object to be handled
4244  * \param[out]fid       pointer to the buffer to hold the parent FID
4245  *
4246  * \retval              0 for getting the real parent FID successfully
4247  * \retval              negative error number on failure
4248  */
4249 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4250                                     struct osd_object *obj,
4251                                     struct lu_fid *fid)
4252 {
4253         struct osd_thread_info  *oti    = osd_oti_get(env);
4254         struct lu_buf           *buf    = &oti->oti_big_buf;
4255         struct dentry           *dentry = &oti->oti_obj_dentry;
4256         struct inode            *inode  = obj->oo_inode;
4257         struct linkea_data       ldata  = { NULL };
4258         int                      rc;
4259         ENTRY;
4260
4261         fid_zero(fid);
4262         if (!S_ISDIR(inode->i_mode))
4263                 RETURN(-EIO);
4264
4265 again:
4266         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4267                              buf->lb_buf, buf->lb_len);
4268         if (rc == -ERANGE) {
4269                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4270                                      NULL, 0);
4271                 if (rc > 0) {
4272                         lu_buf_realloc(buf, rc);
4273                         if (buf->lb_buf == NULL)
4274                                 RETURN(-ENOMEM);
4275
4276                         goto again;
4277                 }
4278         }
4279
4280         if (unlikely(rc == 0))
4281                 RETURN(-ENODATA);
4282
4283         if (rc < 0)
4284                 RETURN(rc);
4285
4286         if (unlikely(buf->lb_buf == NULL)) {
4287                 lu_buf_realloc(buf, rc);
4288                 if (buf->lb_buf == NULL)
4289                         RETURN(-ENOMEM);
4290
4291                 goto again;
4292         }
4293
4294         ldata.ld_buf = buf;
4295         rc = linkea_init(&ldata);
4296         if (rc == 0) {
4297                 linkea_first_entry(&ldata);
4298                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4299         }
4300
4301         RETURN(rc);
4302 }
4303
4304 /**
4305  * Calls ->lookup() to find dentry. From dentry get inode and
4306  * read inode's ea to get fid. This is required for  interoperability
4307  * mode (b11826)
4308  *
4309  * \retval   0, on success
4310  * \retval -ve, on error
4311  */
4312 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4313                              struct dt_rec *rec, const struct dt_key *key)
4314 {
4315         struct inode                    *dir    = obj->oo_inode;
4316         struct dentry                   *dentry;
4317         struct ldiskfs_dir_entry_2      *de;
4318         struct buffer_head              *bh;
4319         struct lu_fid                   *fid = (struct lu_fid *) rec;
4320         struct htree_lock               *hlock = NULL;
4321         int                             ino;
4322         int                             rc;
4323         ENTRY;
4324
4325         LASSERT(dir->i_op != NULL);
4326         LASSERT(dir->i_op->lookup != NULL);
4327
4328         dentry = osd_child_dentry_get(env, obj,
4329                                       (char *)key, strlen((char *)key));
4330
4331         if (obj->oo_hl_head != NULL) {
4332                 hlock = osd_oti_get(env)->oti_hlock;
4333                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4334                                    dir, LDISKFS_HLOCK_LOOKUP);
4335         } else {
4336                 down_read(&obj->oo_ext_idx_sem);
4337         }
4338
4339         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4340         if (bh) {
4341                 struct osd_thread_info *oti = osd_oti_get(env);
4342                 struct osd_inode_id *id = &oti->oti_id;
4343                 struct osd_idmap_cache *oic = &oti->oti_cache;
4344                 struct osd_device *dev = osd_obj2dev(obj);
4345
4346                 ino = le32_to_cpu(de->inode);
4347                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4348                         brelse(bh);
4349                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4350                         GOTO(out, rc);
4351                 }
4352
4353                 rc = osd_get_fid_from_dentry(de, rec);
4354
4355                 /* done with de, release bh */
4356                 brelse(bh);
4357                 if (rc != 0) {
4358                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
4359                                 const char *name = (const char *)key;
4360
4361                                 /* If the parent is on remote MDT, and there
4362                                  * is no FID-in-dirent, then we have to get
4363                                  * the parent FID from the linkEA.  */
4364                                 if (likely(strlen(name) == 2 &&
4365                                            name[0] == '.' && name[1] == '.'))
4366                                         rc = osd_get_pfid_from_linkea(env, obj,
4367                                                                       fid);
4368                         } else {
4369                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4370                         }
4371                 } else {
4372                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4373                 }
4374
4375                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
4376                         fid_zero(&oic->oic_fid);
4377
4378                         GOTO(out, rc);
4379                 }
4380
4381                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4382                 rc = osd_consistency_check(oti, dev, oic);
4383                 if (rc != 0)
4384                         fid_zero(&oic->oic_fid);
4385         } else {
4386                 rc = -ENOENT;
4387         }
4388
4389         GOTO(out, rc);
4390
4391 out:
4392         if (hlock != NULL)
4393                 ldiskfs_htree_unlock(hlock);
4394         else
4395                 up_read(&obj->oo_ext_idx_sem);
4396         return rc;
4397 }
4398
4399 /**
4400  * Find the osd object for given fid.
4401  *
4402  * \param fid need to find the osd object having this fid
4403  *
4404  * \retval osd_object on success
4405  * \retval        -ve on error
4406  */
4407 static struct osd_object *osd_object_find(const struct lu_env *env,
4408                                           struct dt_object *dt,
4409                                           const struct lu_fid *fid)
4410 {
4411         struct lu_device  *ludev = dt->do_lu.lo_dev;
4412         struct osd_object *child = NULL;
4413         struct lu_object  *luch;
4414         struct lu_object  *lo;
4415
4416         /*
4417          * at this point topdev might not exist yet
4418          * (i.e. MGS is preparing profiles). so we can
4419          * not rely on topdev and instead lookup with
4420          * our device passed as topdev. this can't work
4421          * if the object isn't cached yet (as osd doesn't
4422          * allocate lu_header). IOW, the object must be
4423          * in the cache, otherwise lu_object_alloc() crashes
4424          * -bzzz
4425          */
4426         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4427                                  ludev : ludev->ld_site->ls_top_dev,
4428                                  fid, NULL);
4429         if (!IS_ERR(luch)) {
4430                 if (lu_object_exists(luch)) {
4431                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4432                         if (lo != NULL)
4433                                 child = osd_obj(lo);
4434                         else
4435                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4436                                                 "lu_object can't be located"
4437                                                 DFID"\n", PFID(fid));
4438
4439                         if (child == NULL) {
4440                                 lu_object_put(env, luch);
4441                                 CERROR("Unable to get osd_object\n");
4442                                 child = ERR_PTR(-ENOENT);
4443                         }
4444                 } else {
4445                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4446                                         "lu_object does not exists "DFID"\n",
4447                                         PFID(fid));
4448                         lu_object_put(env, luch);
4449                         child = ERR_PTR(-ENOENT);
4450                 }
4451         } else {
4452                 child = ERR_CAST(luch);
4453         }
4454
4455         return child;
4456 }
4457
4458 /**
4459  * Put the osd object once done with it.
4460  *
4461  * \param obj osd object that needs to be put
4462  */
4463 static inline void osd_object_put(const struct lu_env *env,
4464                                   struct osd_object *obj)
4465 {
4466         lu_object_put(env, &obj->oo_dt.do_lu);
4467 }
4468
4469 static int osd_index_declare_ea_insert(const struct lu_env *env,
4470                                        struct dt_object *dt,
4471                                        const struct dt_rec *rec,
4472                                        const struct dt_key *key,
4473                                        struct thandle *handle)
4474 {
4475         struct osd_thandle      *oh;
4476         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4477         struct lu_fid           *fid = (struct lu_fid *)rec;
4478         int                      credits, rc = 0;
4479         ENTRY;
4480
4481         LASSERT(!dt_object_remote(dt));
4482         LASSERT(handle != NULL);
4483
4484         oh = container_of0(handle, struct osd_thandle, ot_super);
4485         LASSERT(oh->ot_handle == NULL);
4486
4487         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
4488         if (fid != NULL) {
4489                 rc = osd_remote_fid(env, osd, fid);
4490                 if (unlikely(rc < 0))
4491                         RETURN(rc);
4492                 if (rc > 0) {
4493                         /* a reference to remote inode is represented by an
4494                          * agent inode which we have to create */
4495                         credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
4496                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4497                 }
4498                 rc = 0;
4499         }
4500
4501         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
4502
4503         if (osd_dt_obj(dt)->oo_inode != NULL) {
4504                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4505
4506                 /* We ignore block quota on meta pool (MDTs), so needn't
4507                  * calculate how many blocks will be consumed by this index
4508                  * insert */
4509                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4510                                            i_gid_read(inode), 0, oh,
4511                                            osd_dt_obj(dt), true, NULL, false);
4512         }
4513
4514         RETURN(rc);
4515 }
4516
4517 /**
4518  * Index add function for interoperability mode (b11826).
4519  * It will add the directory entry.This entry is needed to
4520  * maintain name->fid mapping.
4521  *
4522  * \param key it is key i.e. file entry to be inserted
4523  * \param rec it is value of given key i.e. fid
4524  *
4525  * \retval   0, on success
4526  * \retval -ve, on error
4527  */
4528 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4529                                const struct dt_rec *rec,
4530                                const struct dt_key *key, struct thandle *th,
4531                                int ignore_quota)
4532 {
4533         struct osd_object       *obj = osd_dt_obj(dt);
4534         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4535         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4536         const struct lu_fid     *fid    = rec1->rec_fid;
4537         const char              *name = (const char *)key;
4538         struct osd_thread_info  *oti   = osd_oti_get(env);
4539         struct osd_inode_id     *id    = &oti->oti_id;
4540         struct inode            *child_inode = NULL;
4541         struct osd_object       *child = NULL;
4542         int                     rc;
4543         ENTRY;
4544
4545         if (!dt_object_exists(dt))
4546                 RETURN(-ENOENT);
4547
4548         LASSERT(osd_invariant(obj));
4549         LASSERT(!dt_object_remote(dt));
4550         LASSERT(th != NULL);
4551
4552         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4553
4554         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4555
4556         rc = osd_remote_fid(env, osd, fid);
4557         if (rc < 0) {
4558                 CERROR("%s: Can not find object "DFID" rc %d\n",
4559                        osd_name(osd), PFID(fid), rc);
4560                 RETURN(rc);
4561         }
4562
4563         if (rc == 1) {
4564                 /* Insert remote entry */
4565                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4566                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4567                         struct osd_thandle      *oh;
4568
4569                         /* If parent on remote MDT, we need put this object
4570                          * under AGENT */
4571                         oh = container_of(th, typeof(*oh), ot_super);
4572                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4573                         if (rc != 0) {
4574                                 CERROR("%s: add "DFID" error: rc = %d\n",
4575                                        osd_name(osd),
4576                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4577                                 RETURN(rc);
4578                         }
4579
4580                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4581                 } else {
4582                         child_inode = osd_create_local_agent_inode(env, osd,
4583                                         obj, fid, rec1->rec_type & S_IFMT, th);
4584                         if (IS_ERR(child_inode))
4585                                 RETURN(PTR_ERR(child_inode));
4586                 }
4587         } else {
4588                 /* Insert local entry */
4589                 child = osd_object_find(env, dt, fid);
4590                 if (IS_ERR(child)) {
4591                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4592                                osd_name(osd), PFID(fid),
4593                                id->oii_ino, id->oii_gen,
4594                                (int)PTR_ERR(child));
4595                         RETURN(PTR_ERR(child));
4596                 }
4597                 child_inode = igrab(child->oo_inode);
4598         }
4599
4600         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4601
4602         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4603                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4604
4605         iput(child_inode);
4606         if (child != NULL)
4607                 osd_object_put(env, child);
4608         LASSERT(osd_invariant(obj));
4609         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4610         RETURN(rc);
4611 }
4612
4613 /**
4614  *  Initialize osd Iterator for given osd index object.
4615  *
4616  *  \param  dt      osd index object
4617  */
4618
4619 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4620                                      struct dt_object *dt,
4621                                      __u32 unused)
4622 {
4623         struct osd_it_iam      *it;
4624         struct osd_object      *obj = osd_dt_obj(dt);
4625         struct lu_object       *lo  = &dt->do_lu;
4626         struct iam_path_descr  *ipd;
4627         struct iam_container   *bag = &obj->oo_dir->od_container;
4628
4629         if (!dt_object_exists(dt))
4630                 return ERR_PTR(-ENOENT);
4631
4632         OBD_ALLOC_PTR(it);
4633         if (it == NULL)
4634                 return ERR_PTR(-ENOMEM);
4635
4636         ipd = osd_it_ipd_get(env, bag);
4637         if (likely(ipd != NULL)) {
4638                 it->oi_obj = obj;
4639                 it->oi_ipd = ipd;
4640                 lu_object_get(lo);
4641                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4642                 return (struct dt_it *)it;
4643         } else {
4644                 OBD_FREE_PTR(it);
4645                 return ERR_PTR(-ENOMEM);
4646         }
4647 }
4648
4649 /**
4650  * free given Iterator.
4651  */
4652
4653 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4654 {
4655         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4656         struct osd_object       *obj = it->oi_obj;
4657
4658         iam_it_fini(&it->oi_it);
4659         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4660         lu_object_put(env, &obj->oo_dt.do_lu);
4661         OBD_FREE_PTR(it);
4662 }
4663
4664 /**
4665  *  Move Iterator to record specified by \a key
4666  *
4667  *  \param  di      osd iterator
4668  *  \param  key     key for index
4669  *
4670  *  \retval +ve  di points to record with least key not larger than key
4671  *  \retval  0   di points to exact matched key
4672  *  \retval -ve  failure
4673  */
4674
4675 static int osd_it_iam_get(const struct lu_env *env,
4676                           struct dt_it *di, const struct dt_key *key)
4677 {
4678         struct osd_thread_info  *oti = osd_oti_get(env);
4679         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4680
4681         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4682                 /* swab quota uid/gid */
4683                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4684                 key = (struct dt_key *)&oti->oti_quota_id;
4685         }
4686
4687         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4688 }
4689
4690 /**
4691  *  Release Iterator
4692  *
4693  *  \param  di      osd iterator
4694  */
4695 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4696 {
4697         struct osd_it_iam *it = (struct osd_it_iam *)di;
4698
4699         iam_it_put(&it->oi_it);
4700 }
4701
4702 /**
4703  *  Move iterator by one record
4704  *
4705  *  \param  di      osd iterator
4706  *
4707  *  \retval +1   end of container reached
4708  *  \retval  0   success
4709  *  \retval -ve  failure
4710  */
4711
4712 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4713 {
4714         struct osd_it_iam *it = (struct osd_it_iam *)di;
4715
4716         return iam_it_next(&it->oi_it);
4717 }
4718
4719 /**
4720  * Return pointer to the key under iterator.
4721  */
4722
4723 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4724                                  const struct dt_it *di)
4725 {
4726         struct osd_thread_info *oti = osd_oti_get(env);
4727         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4728         struct osd_object      *obj = it->oi_obj;
4729         struct dt_key          *key;
4730
4731         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4732
4733         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4734                 /* swab quota uid/gid */
4735                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4736                 key = (struct dt_key *)&oti->oti_quota_id;
4737         }
4738
4739         return key;
4740 }
4741
4742 /**
4743  * Return size of key under iterator (in bytes)
4744  */
4745
4746 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4747 {
4748         struct osd_it_iam *it = (struct osd_it_iam *)di;
4749
4750         return iam_it_key_size(&it->oi_it);
4751 }
4752
4753 static inline void
4754 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4755 {
4756         /* check if file type is required */
4757         if (ent->lde_attrs & LUDA_TYPE) {
4758                 struct luda_type *lt;
4759                 int align = sizeof(*lt) - 1;
4760
4761                 len = (len + align) & ~align;
4762                 lt = (struct luda_type *)(ent->lde_name + len);
4763                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4764         }
4765
4766         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4767 }
4768
4769 /**
4770  * build lu direct from backend fs dirent.
4771  */
4772
4773 static inline void
4774 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4775                    char *name, __u16 namelen, __u16 type, __u32 attr)
4776 {
4777         ent->lde_attrs = attr | LUDA_FID;
4778         fid_cpu_to_le(&ent->lde_fid, fid);
4779
4780         ent->lde_hash = cpu_to_le64(offset);
4781         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4782
4783         strncpy(ent->lde_name, name, namelen);
4784         ent->lde_name[namelen] = '\0';
4785         ent->lde_namelen = cpu_to_le16(namelen);
4786
4787         /* append lustre attributes */
4788         osd_it_append_attrs(ent, namelen, type);
4789 }
4790
4791 /**
4792  * Return pointer to the record under iterator.
4793  */
4794 static int osd_it_iam_rec(const struct lu_env *env,
4795                           const struct dt_it *di,
4796                           struct dt_rec *dtrec, __u32 attr)
4797 {
4798         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4799         struct osd_thread_info *info = osd_oti_get(env);
4800         ENTRY;
4801
4802         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4803                 const struct osd_fid_pack *rec;
4804                 struct lu_fid             *fid = &info->oti_fid;
4805                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4806                 char                      *name;
4807                 int                        namelen;
4808                 __u64                      hash;
4809                 int                        rc;
4810
4811                 name = (char *)iam_it_key_get(&it->oi_it);
4812                 if (IS_ERR(name))
4813                         RETURN(PTR_ERR(name));
4814
4815                 namelen = iam_it_key_size(&it->oi_it);
4816
4817                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4818                 if (IS_ERR(rec))
4819                         RETURN(PTR_ERR(rec));
4820
4821                 rc = osd_fid_unpack(fid, rec);
4822                 if (rc)
4823                         RETURN(rc);
4824
4825                 hash = iam_it_store(&it->oi_it);
4826
4827                 /* IAM does not store object type in IAM index (dir) */
4828                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4829                                    0, LUDA_FID);
4830         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4831                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4832                            (struct iam_rec *)dtrec);
4833                 osd_quota_unpack(it->oi_obj, dtrec);
4834         } else {
4835                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4836                            (struct iam_rec *)dtrec);
4837         }
4838
4839         RETURN(0);
4840 }
4841
4842 /**
4843  * Returns cookie for current Iterator position.
4844  */
4845 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4846 {
4847         struct osd_it_iam *it = (struct osd_it_iam *)di;
4848
4849         return iam_it_store(&it->oi_it);
4850 }
4851
4852 /**
4853  * Restore iterator from cookie.
4854  *
4855  * \param  di      osd iterator
4856  * \param  hash    Iterator location cookie
4857  *
4858  * \retval +ve  di points to record with least key not larger than key.
4859  * \retval  0   di points to exact matched key
4860  * \retval -ve  failure
4861  */
4862
4863 static int osd_it_iam_load(const struct lu_env *env,
4864                            const struct dt_it *di, __u64 hash)
4865 {
4866         struct osd_it_iam *it = (struct osd_it_iam *)di;
4867
4868         return iam_it_load(&it->oi_it, hash);
4869 }
4870
4871 static const struct dt_index_operations osd_index_iam_ops = {
4872         .dio_lookup         = osd_index_iam_lookup,
4873         .dio_declare_insert = osd_index_declare_iam_insert,
4874         .dio_insert         = osd_index_iam_insert,
4875         .dio_declare_delete = osd_index_declare_iam_delete,
4876         .dio_delete         = osd_index_iam_delete,
4877         .dio_it     = {
4878                 .init     = osd_it_iam_init,
4879                 .fini     = osd_it_iam_fini,
4880                 .get      = osd_it_iam_get,
4881                 .put      = osd_it_iam_put,
4882                 .next     = osd_it_iam_next,
4883                 .key      = osd_it_iam_key,
4884                 .key_size = osd_it_iam_key_size,
4885                 .rec      = osd_it_iam_rec,
4886                 .store    = osd_it_iam_store,
4887                 .load     = osd_it_iam_load
4888         }
4889 };
4890
4891
4892 /**
4893  * Creates or initializes iterator context.
4894  *
4895  * \retval struct osd_it_ea, iterator structure on success
4896  *
4897  */
4898 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4899                                     struct dt_object *dt,
4900                                     __u32 attr)
4901 {
4902         struct osd_object       *obj  = osd_dt_obj(dt);
4903         struct osd_thread_info  *info = osd_oti_get(env);
4904         struct osd_it_ea        *oie;
4905         struct file             *file;
4906         struct lu_object        *lo   = &dt->do_lu;
4907         struct dentry           *obj_dentry;
4908         ENTRY;
4909
4910         if (!dt_object_exists(dt) || obj->oo_destroyed)
4911                 RETURN(ERR_PTR(-ENOENT));
4912
4913         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
4914         if (oie == NULL)
4915                 RETURN(ERR_PTR(-ENOMEM));
4916         obj_dentry = &oie->oie_dentry;
4917
4918         obj_dentry->d_inode = obj->oo_inode;
4919         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4920         obj_dentry->d_name.hash = 0;
4921
4922         oie->oie_rd_dirent       = 0;
4923         oie->oie_it_dirent       = 0;
4924         oie->oie_dirent          = NULL;
4925         if (unlikely(!info->oti_it_ea_buf_used)) {
4926                 oie->oie_buf = info->oti_it_ea_buf;
4927                 info->oti_it_ea_buf_used = 1;
4928         } else {
4929                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4930                 if (oie->oie_buf == NULL)
4931                         RETURN(ERR_PTR(-ENOMEM));
4932         }
4933         oie->oie_obj             = obj;
4934
4935         file = &oie->oie_file;
4936
4937         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4938         if (attr & LUDA_64BITHASH)
4939                 file->f_mode    = FMODE_64BITHASH;
4940         else
4941                 file->f_mode    = FMODE_32BITHASH;
4942         file->f_path.dentry     = obj_dentry;
4943         file->f_mapping         = obj->oo_inode->i_mapping;
4944         file->f_op              = obj->oo_inode->i_fop;
4945         set_file_inode(file, obj->oo_inode);
4946
4947         lu_object_get(lo);
4948         RETURN((struct dt_it *) oie);
4949 }
4950
4951 /**
4952  * Destroy or finishes iterator context.
4953  *
4954  * \param di iterator structure to be destroyed
4955  */
4956 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4957 {
4958         struct osd_thread_info  *info = osd_oti_get(env);
4959         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
4960         struct osd_object       *obj    = oie->oie_obj;
4961         struct inode            *inode  = obj->oo_inode;
4962
4963         ENTRY;
4964         oie->oie_file.f_op->release(inode, &oie->oie_file);
4965         lu_object_put(env, &obj->oo_dt.do_lu);
4966         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
4967                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4968         else
4969                 info->oti_it_ea_buf_used = 0;
4970         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
4971         EXIT;
4972 }
4973
4974 /**
4975  * It position the iterator at given key, so that next lookup continues from
4976  * that key Or it is similar to dio_it->load() but based on a key,
4977  * rather than file position.
4978  *
4979  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4980  * to the beginning.
4981  *
4982  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4983  */
4984 static int osd_it_ea_get(const struct lu_env *env,
4985                          struct dt_it *di, const struct dt_key *key)
4986 {
4987         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4988
4989         ENTRY;
4990         LASSERT(((const char *)key)[0] == '\0');
4991         it->oie_file.f_pos      = 0;
4992         it->oie_rd_dirent       = 0;
4993         it->oie_it_dirent       = 0;
4994         it->oie_dirent          = NULL;
4995
4996         RETURN(+1);
4997 }
4998
4999 /**
5000  * Does nothing
5001  */
5002 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5003 {
5004 }
5005
5006 struct osd_filldir_cbs {
5007 #ifdef HAVE_DIR_CONTEXT
5008         struct dir_context ctx;
5009 #endif
5010         struct osd_it_ea  *it;
5011 };
5012 /**
5013  * It is called internally by ->readdir(). It fills the
5014  * iterator's in-memory data structure with required
5015  * information i.e. name, namelen, rec_size etc.
5016  *
5017  * \param buf in which information to be filled in.
5018  * \param name name of the file in given dir
5019  *
5020  * \retval 0 on success
5021  * \retval 1 on buffer full
5022  */
5023 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5024                                loff_t offset, __u64 ino,
5025                                unsigned d_type)
5026 {
5027         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
5028         struct osd_object       *obj  = it->oie_obj;
5029         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5030         struct lu_fid           *fid  = &ent->oied_fid;
5031         struct osd_fid_pack     *rec;
5032         ENTRY;
5033
5034         /* this should never happen */
5035         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5036                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5037                 RETURN(-EIO);
5038         }
5039
5040         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5041             OSD_IT_EA_BUFSIZE)
5042                 RETURN(1);
5043
5044         /* "." is just the object itself. */
5045         if (namelen == 1 && name[0] == '.') {
5046                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5047         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5048                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5049                 if (osd_fid_unpack(fid, rec) != 0)
5050                         fid_zero(fid);
5051         } else {
5052                 fid_zero(fid);
5053         }
5054         d_type &= ~LDISKFS_DIRENT_LUFID;
5055
5056         /* NOT export local root. */
5057         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5058                 ino = obj->oo_inode->i_ino;
5059                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5060         }
5061
5062         ent->oied_ino     = ino;
5063         ent->oied_off     = offset;
5064         ent->oied_namelen = namelen;
5065         ent->oied_type    = d_type;
5066
5067         memcpy(ent->oied_name, name, namelen);
5068
5069         it->oie_rd_dirent++;
5070         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5071         RETURN(0);
5072 }
5073
5074 /**
5075  * Calls ->readdir() to load a directory entry at a time
5076  * and stored it in iterator's in-memory data structure.
5077  *
5078  * \param di iterator's in memory structure
5079  *
5080  * \retval   0 on success
5081  * \retval -ve on error
5082  * \retval +1 reach the end of entry
5083  */
5084 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5085                                const struct dt_it *di)
5086 {
5087         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5088         struct osd_object  *obj   = it->oie_obj;
5089         struct inode       *inode = obj->oo_inode;
5090         struct htree_lock  *hlock = NULL;
5091         struct file        *filp  = &it->oie_file;
5092         int                 rc = 0;
5093         struct osd_filldir_cbs buf = {
5094 #ifdef HAVE_DIR_CONTEXT
5095                 .ctx.actor = osd_ldiskfs_filldir,
5096 #endif
5097                 .it = it
5098         };
5099
5100         ENTRY;
5101         it->oie_dirent = it->oie_buf;
5102         it->oie_rd_dirent = 0;
5103
5104         if (obj->oo_hl_head != NULL) {
5105                 hlock = osd_oti_get(env)->oti_hlock;
5106                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5107                                    inode, LDISKFS_HLOCK_READDIR);
5108         } else {
5109                 down_read(&obj->oo_ext_idx_sem);
5110         }
5111
5112 #ifdef HAVE_DIR_CONTEXT
5113         buf.ctx.pos = filp->f_pos;
5114         rc = inode->i_fop->iterate(filp, &buf.ctx);
5115         filp->f_pos = buf.ctx.pos;
5116 #else
5117         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5118 #endif
5119
5120         if (hlock != NULL)
5121                 ldiskfs_htree_unlock(hlock);
5122         else
5123                 up_read(&obj->oo_ext_idx_sem);
5124
5125         if (it->oie_rd_dirent == 0) {
5126                 /*If it does not get any dirent, it means it has been reached
5127                  *to the end of the dir */
5128                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5129                 if (rc == 0)
5130                         rc = 1;
5131         } else {
5132                 it->oie_dirent = it->oie_buf;
5133                 it->oie_it_dirent = 1;
5134         }
5135
5136         RETURN(rc);
5137 }
5138
5139 /**
5140  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5141  * to load a directory entry at a time and stored it in
5142  * iterator's in-memory data structure.
5143  *
5144  * \param di iterator's in memory structure
5145  *
5146  * \retval +ve iterator reached to end
5147  * \retval   0 iterator not reached to end
5148  * \retval -ve on error
5149  */
5150 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5151 {
5152         struct osd_it_ea *it = (struct osd_it_ea *)di;
5153         int rc;
5154
5155         ENTRY;
5156
5157         if (it->oie_it_dirent < it->oie_rd_dirent) {
5158                 it->oie_dirent =
5159                         (void *) it->oie_dirent +
5160                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5161                                        it->oie_dirent->oied_namelen);
5162                 it->oie_it_dirent++;
5163                 RETURN(0);
5164         } else {
5165                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5166                         rc = +1;
5167                 else
5168                         rc = osd_ldiskfs_it_fill(env, di);
5169         }
5170
5171         RETURN(rc);
5172 }
5173
5174 /**
5175  * Returns the key at current position from iterator's in memory structure.
5176  *
5177  * \param di iterator's in memory structure
5178  *
5179  * \retval key i.e. struct dt_key on success
5180  */
5181 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5182                                     const struct dt_it *di)
5183 {
5184         struct osd_it_ea *it = (struct osd_it_ea *)di;
5185
5186         return (struct dt_key *)it->oie_dirent->oied_name;
5187 }
5188
5189 /**
5190  * Returns the key's size at current position from iterator's in memory structure.
5191  *
5192  * \param di iterator's in memory structure
5193  *
5194  * \retval key_size i.e. struct dt_key on success
5195  */
5196 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5197 {
5198         struct osd_it_ea *it = (struct osd_it_ea *)di;
5199
5200         return it->oie_dirent->oied_namelen;
5201 }
5202
5203 static inline bool
5204 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5205 {
5206         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5207                  "dot_dotdot = %d\n", dot_dotdot);
5208
5209         if (LDISKFS_DIR_REC_LEN(de) >=
5210             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5211                 return true;
5212
5213         return false;
5214 }
5215
5216 static inline bool
5217 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
5218                      unsigned blocksize, int dot_dotdot)
5219 {
5220         if (dot_dotdot > 0)
5221                 return osd_dot_dotdot_has_space(de, dot_dotdot);
5222
5223         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
5224             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5225                 return true;
5226
5227         return false;
5228 }
5229
5230 static int
5231 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5232                     struct dentry *dentry, const struct lu_fid *fid,
5233                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5234                     struct htree_lock *hlock, int dot_dotdot)
5235 {
5236         struct inode                *dir        = dentry->d_parent->d_inode;
5237         struct inode                *inode      = dentry->d_inode;
5238         struct osd_fid_pack         *rec;
5239         struct ldiskfs_dentry_param *ldp;
5240         int                          namelen    = dentry->d_name.len;
5241         int                          rc;
5242         ENTRY;
5243
5244         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5245                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5246                 RETURN(0);
5247
5248         /* There is enough space to hold the FID-in-dirent. */
5249         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize,
5250                                  dot_dotdot)) {
5251                 rc = ldiskfs_journal_get_write_access(jh, bh);
5252                 if (rc != 0)
5253                         RETURN(rc);
5254
5255                 de->name[namelen] = 0;
5256                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
5257                 rec->fp_len = sizeof(struct lu_fid) + 1;
5258                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5259                 de->file_type |= LDISKFS_DIRENT_LUFID;
5260                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5261
5262                 RETURN(rc);
5263         }
5264
5265         LASSERTF(dot_dotdot == 0, "dot_dotdot = %d\n", dot_dotdot);
5266
5267         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5268         if (rc != 0)
5269                 RETURN(rc);
5270
5271         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5272         osd_get_ldiskfs_dirent_param(ldp, fid);
5273         dentry->d_fsdata = (void *)ldp;
5274         ll_vfs_dq_init(dir);
5275         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5276         /* It is too bad, we cannot reinsert the name entry back.
5277          * That means we lose it! */
5278         if (rc != 0)
5279                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5280                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5281                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5282                        dir->i_ino, dir->i_generation, namelen,
5283                        dentry->d_name.name, PFID(fid), rc);
5284
5285         RETURN(rc);
5286 }
5287
5288 static int
5289 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5290                         struct osd_it_ea *it, struct lu_fid *fid,
5291                         struct osd_inode_id *id, __u32 *attr)
5292 {
5293         struct osd_thread_info     *info        = osd_oti_get(env);
5294         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5295         struct osd_device          *dev         = osd_obj2dev(obj);
5296         struct super_block         *sb          = osd_sb(dev);
5297         const char                 *devname     =
5298                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5299         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5300         struct inode               *dir         = obj->oo_inode;
5301         struct htree_lock          *hlock       = NULL;
5302         struct buffer_head         *bh          = NULL;
5303         handle_t                   *jh          = NULL;
5304         struct ldiskfs_dir_entry_2 *de;
5305         struct dentry              *dentry;
5306         struct inode               *inode;
5307         int                         credits;
5308         int                         rc;
5309         int                         dot_dotdot  = 0;
5310         bool                        dirty       = false;
5311         ENTRY;
5312
5313         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5314         inode = osd_iget(info, dev, id);
5315         if (IS_ERR(inode)) {
5316                 rc = PTR_ERR(inode);
5317                 if (rc == -ENOENT || rc == -ESTALE) {
5318                         *attr |= LUDA_UNKNOWN;
5319                         rc = 0;
5320                 } else {
5321                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5322                                "check_repair, dir = %lu/%u, name = %.*s: "
5323                                "rc = %d\n",
5324                                devname, dir->i_ino, dir->i_generation,
5325                                ent->oied_namelen, ent->oied_name, rc);
5326                 }
5327
5328                 RETURN(rc);
5329         }
5330
5331         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5332                                            ent->oied_namelen);
5333         rc = osd_get_lma(info, inode, dentry, lma);
5334         if (rc == -ENODATA)
5335                 lma = NULL;
5336         else if (rc != 0)
5337                 GOTO(out, rc);
5338
5339         if (ent->oied_name[0] == '.') {
5340                 if (ent->oied_namelen == 1)
5341                         dot_dotdot = 1;
5342                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5343                         dot_dotdot = 2;
5344         }
5345
5346         /* We need to ensure that the name entry is still valid.
5347          * Because it may be removed or renamed by other already.
5348          *
5349          * The unlink or rename operation will start journal before PDO lock,
5350          * so to avoid deadlock, here we need to start journal handle before
5351          * related PDO lock also. But because we do not know whether there
5352          * will be something to be repaired before PDO lock, we just start
5353          * journal without conditions.
5354          *
5355          * We may need to remove the name entry firstly, then insert back.
5356          * One credit is for user quota file update.
5357          * One credit is for group quota file update.
5358          * Two credits are for dirty inode. */
5359         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5360                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5361
5362         if (dev->od_dirent_journal != 0) {
5363
5364 again:
5365                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5366                 if (IS_ERR(jh)) {
5367                         rc = PTR_ERR(jh);
5368                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5369                                "check_repair, dir = %lu/%u, credits = %d, "
5370                                "name = %.*s: rc = %d\n",
5371                                devname, dir->i_ino, dir->i_generation, credits,
5372                                ent->oied_namelen, ent->oied_name, rc);
5373
5374                         GOTO(out_inode, rc);
5375                 }
5376
5377                 if (obj->oo_hl_head != NULL) {
5378                         hlock = osd_oti_get(env)->oti_hlock;
5379                         /* "0" means exclusive lock for the whole directory.
5380                          * We need to prevent others access such name entry
5381                          * during the delete + insert. Neither HLOCK_ADD nor
5382                          * HLOCK_DEL cannot guarantee the atomicity. */
5383                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5384                 } else {
5385                         down_write(&obj->oo_ext_idx_sem);
5386                 }
5387         } else {
5388                 if (obj->oo_hl_head != NULL) {
5389                         hlock = osd_oti_get(env)->oti_hlock;
5390                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5391                                            LDISKFS_HLOCK_LOOKUP);
5392                 } else {
5393                         down_read(&obj->oo_ext_idx_sem);
5394                 }
5395         }
5396
5397         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5398         /* For dot/dotdot entry, if there is not enough space to hold the
5399          * FID-in-dirent, just keep them there. It only happens when the
5400          * device upgraded from 1.8 or restored from MDT file-level backup.
5401          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5402          * and needs to get FID from LMA when readdir, it will not affect the
5403          * performance much. */
5404         if ((bh == NULL) || (le32_to_cpu(de->inode) != inode->i_ino) ||
5405             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5406                 *attr |= LUDA_IGNORE;
5407
5408                 GOTO(out, rc = 0);
5409         }
5410
5411         if (lma != NULL) {
5412                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
5413                         struct lu_fid *tfid = &lma->lma_self_fid;
5414
5415                         *attr |= LUDA_IGNORE;
5416                         /* It must be REMOTE_PARENT_DIR and as the
5417                          * dotdot entry of remote directory */
5418                         if (unlikely(dot_dotdot != 2 ||
5419                                      fid_seq(tfid) != FID_SEQ_LOCAL_FILE ||
5420                                      fid_oid(tfid) != REMOTE_PARENT_DIR_OID)) {
5421                                 CDEBUG(D_LFSCK, "%.16s: expect remote agent "
5422                                        "parent directory, but got %.*s under "
5423                                        "dir = %lu/%u with the FID "DFID"\n",
5424                                        devname, ent->oied_namelen,
5425                                        ent->oied_name, dir->i_ino,
5426                                        dir->i_generation, PFID(tfid));
5427
5428                                 GOTO(out, rc = -EIO);
5429                         }
5430
5431                         GOTO(out, rc = 0);
5432                 }
5433
5434                 if (fid_is_sane(fid)) {
5435                         /* FID-in-dirent is valid. */
5436                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5437                                 GOTO(out, rc = 0);
5438
5439                         /* Do not repair under dryrun mode. */
5440                         if (*attr & LUDA_VERIFY_DRYRUN) {
5441                                 *attr |= LUDA_REPAIR;
5442
5443                                 GOTO(out, rc = 0);
5444                         }
5445
5446                         if (jh == NULL) {
5447                                 brelse(bh);
5448                                 dev->od_dirent_journal = 1;
5449                                 if (hlock != NULL) {
5450                                         ldiskfs_htree_unlock(hlock);
5451                                         hlock = NULL;
5452                                 } else {
5453                                         up_read(&obj->oo_ext_idx_sem);
5454                                 }
5455
5456                                 goto again;
5457                         }
5458
5459                         *fid = lma->lma_self_fid;
5460                         dirty = true;
5461                         /* Update the FID-in-dirent. */
5462                         rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5463                                                  hlock, dot_dotdot);
5464                         if (rc == 0)
5465                                 *attr |= LUDA_REPAIR;
5466                         else
5467                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5468                                        "in the dirent, dir = %lu/%u, "
5469                                        "name = %.*s, "DFID": rc = %d\n",
5470                                        devname, dir->i_ino, dir->i_generation,
5471                                        ent->oied_namelen, ent->oied_name,
5472                                        PFID(fid), rc);
5473                 } else {
5474                         /* Do not repair under dryrun mode. */
5475                         if (*attr & LUDA_VERIFY_DRYRUN) {
5476                                 *fid = lma->lma_self_fid;
5477                                 *attr |= LUDA_REPAIR;
5478
5479                                 GOTO(out, rc = 0);
5480                         }
5481
5482                         if (jh == NULL) {
5483                                 brelse(bh);
5484                                 dev->od_dirent_journal = 1;
5485                                 if (hlock != NULL) {
5486                                         ldiskfs_htree_unlock(hlock);
5487                                         hlock = NULL;
5488                                 } else {
5489                                         up_read(&obj->oo_ext_idx_sem);
5490                                 }
5491
5492                                 goto again;
5493                         }
5494
5495                         *fid = lma->lma_self_fid;
5496                         dirty = true;
5497                         /* Append the FID-in-dirent. */
5498                         rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5499                                                  hlock, dot_dotdot);
5500                         if (rc == 0)
5501                                 *attr |= LUDA_REPAIR;
5502                         else
5503                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5504                                        "after the dirent, dir = %lu/%u, "
5505                                        "name = %.*s, "DFID": rc = %d\n",
5506                                        devname, dir->i_ino, dir->i_generation,
5507                                        ent->oied_namelen, ent->oied_name,
5508                                        PFID(fid), rc);
5509                 }
5510         } else {
5511                 /* Do not repair under dryrun mode. */
5512                 if (*attr & LUDA_VERIFY_DRYRUN) {
5513                         if (fid_is_sane(fid)) {
5514                                 *attr |= LUDA_REPAIR;
5515                         } else {
5516                                 lu_igif_build(fid, inode->i_ino,
5517                                               inode->i_generation);
5518                                 *attr |= LUDA_UPGRADE;
5519                         }
5520
5521                         GOTO(out, rc = 0);
5522                 }
5523
5524                 if (jh == NULL) {
5525                         brelse(bh);
5526                         dev->od_dirent_journal = 1;
5527                         if (hlock != NULL) {
5528                                 ldiskfs_htree_unlock(hlock);
5529                                 hlock = NULL;
5530                         } else {
5531                                 up_read(&obj->oo_ext_idx_sem);
5532                         }
5533
5534                         goto again;
5535                 }
5536
5537                 dirty = true;
5538                 if (unlikely(fid_is_sane(fid))) {
5539                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5540                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5541                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5542                         if (rc == 0)
5543                                 *attr |= LUDA_REPAIR;
5544                         else
5545                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5546                                        "update dirent, dir = %lu/%u, "
5547                                        "name = %.*s, "DFID": rc = %d\n",
5548                                        devname, dir->i_ino, dir->i_generation,
5549                                        ent->oied_namelen, ent->oied_name,
5550                                        PFID(fid), rc);
5551                 } else {
5552                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5553                         /* It is probably IGIF object. Only aappend the
5554                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5555                         rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5556                                                  hlock, dot_dotdot);
5557                         if (rc == 0)
5558                                 *attr |= LUDA_UPGRADE;
5559                         else
5560                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5561                                        "after the dirent, dir = %lu/%u, "
5562                                        "name = %.*s, "DFID": rc = %d\n",
5563                                        devname, dir->i_ino, dir->i_generation,
5564                                        ent->oied_namelen, ent->oied_name,
5565                                        PFID(fid), rc);
5566                 }
5567         }
5568
5569         GOTO(out, rc);
5570
5571 out:
5572         brelse(bh);
5573         if (hlock != NULL) {
5574                 ldiskfs_htree_unlock(hlock);
5575         } else {
5576                 if (dev->od_dirent_journal != 0)
5577                         up_write(&obj->oo_ext_idx_sem);
5578                 else
5579                         up_read(&obj->oo_ext_idx_sem);
5580         }
5581
5582         if (jh != NULL)
5583                 ldiskfs_journal_stop(jh);
5584
5585 out_inode:
5586         iput(inode);
5587         if (rc >= 0 && !dirty)
5588                 dev->od_dirent_journal = 0;
5589
5590         return rc;
5591 }
5592
5593 /**
5594  * Returns the value at current position from iterator's in memory structure.
5595  *
5596  * \param di struct osd_it_ea, iterator's in memory structure
5597  * \param attr attr requested for dirent.
5598  * \param lde lustre dirent
5599  *
5600  * \retval   0 no error and \param lde has correct lustre dirent.
5601  * \retval -ve on error
5602  */
5603 static inline int osd_it_ea_rec(const struct lu_env *env,
5604                                 const struct dt_it *di,
5605                                 struct dt_rec *dtrec, __u32 attr)
5606 {
5607         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5608         struct osd_object      *obj   = it->oie_obj;
5609         struct osd_device      *dev   = osd_obj2dev(obj);
5610         struct osd_thread_info *oti   = osd_oti_get(env);
5611         struct osd_inode_id    *id    = &oti->oti_id;
5612         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5613         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5614         __u32                   ino   = it->oie_dirent->oied_ino;
5615         int                     rc    = 0;
5616         ENTRY;
5617
5618         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5619
5620         if (attr & LUDA_VERIFY) {
5621                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5622                         attr |= LUDA_IGNORE;
5623                         /* If the parent is on remote MDT, and there
5624                          * is no FID-in-dirent, then we have to get
5625                          * the parent FID from the linkEA.  */
5626                         if (!fid_is_sane(fid) &&
5627                             it->oie_dirent->oied_namelen == 2 &&
5628                             it->oie_dirent->oied_name[0] == '.' &&
5629                             it->oie_dirent->oied_name[1] == '.')
5630                                 osd_get_pfid_from_linkea(env, obj, fid);
5631                 } else {
5632                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5633                                                      &attr);
5634                 }
5635
5636                 if (!fid_is_sane(fid)) {
5637                         attr &= ~LUDA_IGNORE;
5638                         attr |= LUDA_UNKNOWN;
5639                 }
5640         } else {
5641                 attr &= ~LU_DIRENT_ATTRS_MASK;
5642                 if (!fid_is_sane(fid)) {
5643                         bool is_dotdot = false;
5644                         if (it->oie_dirent->oied_namelen == 2 &&
5645                             it->oie_dirent->oied_name[0] == '.' &&
5646                             it->oie_dirent->oied_name[1] == '.')
5647                                 is_dotdot = true;
5648                         /* If the parent is on remote MDT, and there
5649                          * is no FID-in-dirent, then we have to get
5650                          * the parent FID from the linkEA.  */
5651                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
5652                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5653                         } else {
5654                                 if (is_dotdot == false &&
5655                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5656                                         RETURN(-ENOENT);
5657
5658                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5659                         }
5660                 } else {
5661                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5662                 }
5663         }
5664
5665         /* Pack the entry anyway, at least the offset is right. */
5666         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5667                            it->oie_dirent->oied_name,
5668                            it->oie_dirent->oied_namelen,
5669                            it->oie_dirent->oied_type, attr);
5670
5671         if (rc < 0)
5672                 RETURN(rc);
5673
5674         if (osd_remote_fid(env, dev, fid))
5675                 RETURN(0);
5676
5677         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
5678                 osd_add_oi_cache(oti, dev, id, fid);
5679
5680         RETURN(rc > 0 ? 0 : rc);
5681 }
5682
5683 /**
5684  * Returns the record size size at current position.
5685  *
5686  * This function will return record(lu_dirent) size in bytes.
5687  *
5688  * \param[in] env       execution environment
5689  * \param[in] di        iterator's in memory structure
5690  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5691  *                      calculate the lu_dirent size.
5692  *
5693  * \retval      record size(in bytes & in memory) of the current lu_dirent
5694  *              entry.
5695  */
5696 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5697                               __u32 attr)
5698 {
5699         struct osd_it_ea *it = (struct osd_it_ea *)di;
5700
5701         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5702 }
5703
5704 /**
5705  * Returns a cookie for current position of the iterator head, so that
5706  * user can use this cookie to load/start the iterator next time.
5707  *
5708  * \param di iterator's in memory structure
5709  *
5710  * \retval cookie for current position, on success
5711  */
5712 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5713 {
5714         struct osd_it_ea *it = (struct osd_it_ea *)di;
5715
5716         return it->oie_dirent->oied_off;
5717 }
5718
5719 /**
5720  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5721  * to load a directory entry at a time and stored it i inn,
5722  * in iterator's in-memory data structure.
5723  *
5724  * \param di struct osd_it_ea, iterator's in memory structure
5725  *
5726  * \retval +ve on success
5727  * \retval -ve on error
5728  */
5729 static int osd_it_ea_load(const struct lu_env *env,
5730                           const struct dt_it *di, __u64 hash)
5731 {
5732         struct osd_it_ea *it = (struct osd_it_ea *)di;
5733         int rc;
5734
5735         ENTRY;
5736         it->oie_file.f_pos = hash;
5737
5738         rc =  osd_ldiskfs_it_fill(env, di);
5739         if (rc > 0)
5740                 rc = -ENODATA;
5741
5742         if (rc == 0)
5743                 rc = +1;
5744
5745         RETURN(rc);
5746 }
5747
5748 /**
5749  * Index lookup function for interoperability mode (b11826).
5750  *
5751  * \param key,  key i.e. file name to be searched
5752  *
5753  * \retval +ve, on success
5754  * \retval -ve, on error
5755  */
5756 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5757                                struct dt_rec *rec, const struct dt_key *key)
5758 {
5759         struct osd_object *obj = osd_dt_obj(dt);
5760         int rc = 0;
5761
5762         ENTRY;
5763
5764         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5765         LINVRNT(osd_invariant(obj));
5766
5767         rc = osd_ea_lookup_rec(env, obj, rec, key);
5768         if (rc == 0)
5769                 rc = +1;
5770         RETURN(rc);
5771 }
5772
5773 /**
5774  * Index and Iterator operations for interoperability
5775  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5776  */
5777 static const struct dt_index_operations osd_index_ea_ops = {
5778         .dio_lookup         = osd_index_ea_lookup,
5779         .dio_declare_insert = osd_index_declare_ea_insert,
5780         .dio_insert         = osd_index_ea_insert,
5781         .dio_declare_delete = osd_index_declare_ea_delete,
5782         .dio_delete         = osd_index_ea_delete,
5783         .dio_it     = {
5784                 .init     = osd_it_ea_init,
5785                 .fini     = osd_it_ea_fini,
5786                 .get      = osd_it_ea_get,
5787                 .put      = osd_it_ea_put,
5788                 .next     = osd_it_ea_next,
5789                 .key      = osd_it_ea_key,
5790                 .key_size = osd_it_ea_key_size,
5791                 .rec      = osd_it_ea_rec,
5792                 .rec_size = osd_it_ea_rec_size,
5793                 .store    = osd_it_ea_store,
5794                 .load     = osd_it_ea_load
5795         }
5796 };
5797
5798 static void *osd_key_init(const struct lu_context *ctx,
5799                           struct lu_context_key *key)
5800 {
5801         struct osd_thread_info *info;
5802
5803         OBD_ALLOC_PTR(info);
5804         if (info == NULL)
5805                 return ERR_PTR(-ENOMEM);
5806
5807         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5808         if (info->oti_it_ea_buf == NULL)
5809                 goto out_free_info;
5810
5811         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5812
5813         info->oti_hlock = ldiskfs_htree_lock_alloc();
5814         if (info->oti_hlock == NULL)
5815                 goto out_free_ea;
5816
5817         return info;
5818
5819  out_free_ea:
5820         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5821  out_free_info:
5822         OBD_FREE_PTR(info);
5823         return ERR_PTR(-ENOMEM);
5824 }
5825
5826 static void osd_key_fini(const struct lu_context *ctx,
5827                          struct lu_context_key *key, void* data)
5828 {
5829         struct osd_thread_info *info = data;
5830         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
5831
5832         if (info->oti_inode != NULL)
5833                 OBD_FREE_PTR(lli);
5834         if (info->oti_hlock != NULL)
5835                 ldiskfs_htree_lock_free(info->oti_hlock);
5836         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5837         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5838         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5839         lu_buf_free(&info->oti_big_buf);
5840         OBD_FREE_PTR(info);
5841 }
5842
5843 static void osd_key_exit(const struct lu_context *ctx,
5844                          struct lu_context_key *key, void *data)
5845 {
5846         struct osd_thread_info *info = data;
5847
5848         LASSERT(info->oti_r_locks == 0);
5849         LASSERT(info->oti_w_locks == 0);
5850         LASSERT(info->oti_txns    == 0);
5851 }
5852
5853 /* type constructor/destructor: osd_type_init, osd_type_fini */
5854 LU_TYPE_INIT_FINI(osd, &osd_key);
5855
5856 struct lu_context_key osd_key = {
5857         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5858         .lct_init = osd_key_init,
5859         .lct_fini = osd_key_fini,
5860         .lct_exit = osd_key_exit
5861 };
5862
5863
5864 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5865                            const char *name, struct lu_device *next)
5866 {
5867         struct osd_device *osd = osd_dev(d);
5868
5869         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5870             >= sizeof(osd->od_svname))
5871                 return -E2BIG;
5872         return osd_procfs_init(osd, name);
5873 }
5874
5875 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5876 {
5877         struct seq_server_site  *ss = osd_seq_site(osd);
5878         int                     rc;
5879         ENTRY;
5880
5881         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5882                 RETURN(0);
5883
5884         if (unlikely(ss == NULL))
5885                 RETURN(-ENODEV);
5886
5887         OBD_ALLOC_PTR(osd->od_cl_seq);
5888         if (osd->od_cl_seq == NULL)
5889                 RETURN(-ENOMEM);
5890
5891         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5892                              osd->od_svname, ss->ss_server_seq);
5893         if (rc != 0) {
5894                 OBD_FREE_PTR(osd->od_cl_seq);
5895                 osd->od_cl_seq = NULL;
5896                 RETURN(rc);
5897         }
5898
5899         if (ss->ss_node_id == 0) {
5900                 /* If the OSD on the sequence controller(MDT0), then allocate
5901                  * sequence here, otherwise allocate sequence after connected
5902                  * to MDT0 (see mdt_register_lwp_callback()). */
5903                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
5904                                    &osd->od_cl_seq->lcs_space, env);
5905         }
5906
5907         RETURN(rc);
5908 }
5909
5910 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5911 {
5912         if (osd->od_cl_seq == NULL)
5913                 return;
5914
5915         seq_client_fini(osd->od_cl_seq);
5916         OBD_FREE_PTR(osd->od_cl_seq);
5917         osd->od_cl_seq = NULL;
5918 }
5919
5920 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5921 {
5922         ENTRY;
5923
5924         /* shutdown quota slave instance associated with the device */
5925         if (o->od_quota_slave != NULL) {
5926                 qsd_fini(env, o->od_quota_slave);
5927                 o->od_quota_slave = NULL;
5928         }
5929
5930         osd_fid_fini(env, o);
5931
5932         RETURN(0);
5933 }
5934
5935 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5936 {
5937         ENTRY;
5938
5939         if (o->od_mnt != NULL) {
5940                 shrink_dcache_sb(osd_sb(o));
5941                 osd_sync(env, &o->od_dt_dev);
5942
5943                 mntput(o->od_mnt);
5944                 o->od_mnt = NULL;
5945         }
5946
5947         EXIT;
5948 }
5949
5950 static int osd_mount(const struct lu_env *env,
5951                      struct osd_device *o, struct lustre_cfg *cfg)
5952 {
5953         const char              *name  = lustre_cfg_string(cfg, 0);
5954         const char              *dev  = lustre_cfg_string(cfg, 1);
5955         const char              *opts;
5956         unsigned long            page, s_flags, lmd_flags = 0;
5957         struct page             *__page;
5958         struct file_system_type *type;
5959         char                    *options = NULL;
5960         char                    *str;
5961         struct osd_thread_info  *info = osd_oti_get(env);
5962         struct lu_fid           *fid = &info->oti_fid;
5963         struct inode            *inode;
5964         int                      rc = 0, force_over_128tb = 0;
5965         ENTRY;
5966
5967         if (o->od_mnt != NULL)
5968                 RETURN(0);
5969
5970         if (strlen(dev) >= sizeof(o->od_mntdev))
5971                 RETURN(-E2BIG);
5972         strcpy(o->od_mntdev, dev);
5973
5974         str = lustre_cfg_string(cfg, 2);
5975         s_flags = simple_strtoul(str, NULL, 0);
5976         str = strstr(str, ":");
5977         if (str)
5978                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5979         opts = lustre_cfg_string(cfg, 3);
5980 #ifdef __BIG_ENDIAN
5981         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
5982                 CERROR("%s: device %s extents feature is not guaranteed to "
5983                        "work on big-endian systems. Use \"bigendian_extents\" "
5984                        "mount option to override.\n", name, dev);
5985                 RETURN(-EINVAL);
5986         }
5987 #endif
5988         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL)
5989                 force_over_128tb = 1;
5990
5991         __page = alloc_page(GFP_IOFS);
5992         if (__page == NULL)
5993                 GOTO(out, rc = -ENOMEM);
5994         page = (unsigned long)page_address(__page);
5995         options = (char *)page;
5996         *options = '\0';
5997         if (opts != NULL) {
5998                 /* strip out the options for back compatiblity */
5999                 static char *sout[] = {
6000                         "mballoc",
6001                         "iopen",
6002                         "noiopen",
6003                         "iopen_nopriv",
6004                         "extents",
6005                         "noextents",
6006                         /* strip out option we processed in osd */
6007                         "bigendian_extents",
6008                         "force_over_128tb",
6009                         NULL
6010                 };
6011                 strcat(options, opts);
6012                 for (rc = 0, str = options; sout[rc]; ) {
6013                         char *op = strstr(str, sout[rc]);
6014                         if (op == NULL) {
6015                                 rc++;
6016                                 str = options;
6017                                 continue;
6018                         }
6019                         if (op == options || *(op - 1) == ',') {
6020                                 str = op + strlen(sout[rc]);
6021                                 if (*str == ',' || *str == '\0') {
6022                                         *str == ',' ? str++ : str;
6023                                         memmove(op, str, strlen(str) + 1);
6024                                 }
6025                         }
6026                         for (str = op; *str != ',' && *str != '\0'; str++)
6027                                 ;
6028                 }
6029         } else {
6030                 strncat(options, "user_xattr,acl", 14);
6031         }
6032
6033         /* Glom up mount options */
6034         if (*options != '\0')
6035                 strcat(options, ",");
6036         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
6037
6038         type = get_fs_type("ldiskfs");
6039         if (!type) {
6040                 CERROR("%s: cannot find ldiskfs module\n", name);
6041                 GOTO(out, rc = -ENODEV);
6042         }
6043
6044         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6045         module_put(type->owner);
6046
6047         if (IS_ERR(o->od_mnt)) {
6048                 rc = PTR_ERR(o->od_mnt);
6049                 o->od_mnt = NULL;
6050                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6051                 GOTO(out, rc);
6052         }
6053
6054         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) &&
6055             force_over_128tb == 0) {
6056                 CERROR("%s: device %s LDISKFS does not support filesystems "
6057                        "greater than 128TB and can cause data corruption. "
6058                        "Use \"force_over_128tb\" mount option to override.\n",
6059                        name, dev);
6060                 GOTO(out, rc = -EINVAL);
6061         }
6062
6063 #ifdef HAVE_DEV_SET_RDONLY
6064         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
6065                 CERROR("%s: underlying device %s is marked as read-only. "
6066                        "Setup failed\n", name, dev);
6067                 GOTO(out_mnt, rc = -EROFS);
6068         }
6069 #endif
6070
6071         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6072                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6073                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6074                 GOTO(out_mnt, rc = -EINVAL);
6075         }
6076
6077 #ifdef LDISKFS_MOUNT_DIRDATA
6078         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6079                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6080                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6081         else if (!o->od_is_ost)
6082                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6083                       "enabling the dirdata feature. If you do not want to "
6084                       "downgrade to Lustre-1.x again, you can enable it via "
6085                       "'tune2fs -O dirdata device'\n", name, dev);
6086 #endif
6087         inode = osd_sb(o)->s_root->d_inode;
6088         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6089         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6090         if (rc != 0) {
6091                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6092                 GOTO(out_mnt, rc);
6093         }
6094
6095         if (lmd_flags & LMD_FLG_NOSCRUB)
6096                 o->od_noscrub = 1;
6097
6098         GOTO(out, rc = 0);
6099
6100 out_mnt:
6101         mntput(o->od_mnt);
6102         o->od_mnt = NULL;
6103
6104 out:
6105         if (__page)
6106                 __free_page(__page);
6107
6108         return rc;
6109 }
6110
6111 static struct lu_device *osd_device_fini(const struct lu_env *env,
6112                                          struct lu_device *d)
6113 {
6114         struct osd_device *o = osd_dev(d);
6115         ENTRY;
6116
6117         osd_shutdown(env, o);
6118         osd_procfs_fini(o);
6119         osd_scrub_cleanup(env, o);
6120         osd_obj_map_fini(o);
6121         osd_umount(env, o);
6122
6123         RETURN(NULL);
6124 }
6125
6126 static int osd_device_init0(const struct lu_env *env,
6127                             struct osd_device *o,
6128                             struct lustre_cfg *cfg)
6129 {
6130         struct lu_device        *l = osd2lu_dev(o);
6131         struct osd_thread_info *info;
6132         int                     rc;
6133         int                     cplen = 0;
6134
6135         /* if the module was re-loaded, env can loose its keys */
6136         rc = lu_env_refill((struct lu_env *) env);
6137         if (rc)
6138                 GOTO(out, rc);
6139         info = osd_oti_get(env);
6140         LASSERT(info);
6141
6142         l->ld_ops = &osd_lu_ops;
6143         o->od_dt_dev.dd_ops = &osd_dt_ops;
6144
6145         spin_lock_init(&o->od_osfs_lock);
6146         mutex_init(&o->od_otable_mutex);
6147         INIT_LIST_HEAD(&o->od_orphan_list);
6148
6149         o->od_read_cache = 1;
6150         o->od_writethrough_cache = 1;
6151         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6152
6153         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6154                         sizeof(o->od_svname));
6155         if (cplen >= sizeof(o->od_svname)) {
6156                 rc = -E2BIG;
6157                 GOTO(out, rc);
6158         }
6159
6160         o->od_index = -1; /* -1 means index is invalid */
6161         rc = server_name2index(o->od_svname, &o->od_index, NULL);
6162         if (rc == LDD_F_SV_TYPE_OST)
6163                 o->od_is_ost = 1;
6164
6165         o->od_full_scrub_ratio = OFSR_DEFAULT;
6166         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
6167         rc = osd_mount(env, o, cfg);
6168         if (rc != 0)
6169                 GOTO(out, rc);
6170
6171         rc = osd_obj_map_init(env, o);
6172         if (rc != 0)
6173                 GOTO(out_mnt, rc);
6174
6175         rc = lu_site_init(&o->od_site, l);
6176         if (rc != 0)
6177                 GOTO(out_compat, rc);
6178         o->od_site.ls_bottom_dev = l;
6179
6180         rc = lu_site_init_finish(&o->od_site);
6181         if (rc != 0)
6182                 GOTO(out_site, rc);
6183
6184         INIT_LIST_HEAD(&o->od_ios_list);
6185         /* setup scrub, including OI files initialization */
6186         rc = osd_scrub_setup(env, o);
6187         if (rc < 0)
6188                 GOTO(out_site, rc);
6189
6190         rc = osd_procfs_init(o, o->od_svname);
6191         if (rc != 0) {
6192                 CERROR("%s: can't initialize procfs: rc = %d\n",
6193                        o->od_svname, rc);
6194                 GOTO(out_scrub, rc);
6195         }
6196
6197         LASSERT(l->ld_site->ls_linkage.next != NULL);
6198         LASSERT(l->ld_site->ls_linkage.prev != NULL);
6199
6200         /* initialize quota slave instance */
6201         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6202                                      o->od_proc_entry);
6203         if (IS_ERR(o->od_quota_slave)) {
6204                 rc = PTR_ERR(o->od_quota_slave);
6205                 o->od_quota_slave = NULL;
6206                 GOTO(out_procfs, rc);
6207         }
6208
6209         RETURN(0);
6210
6211 out_procfs:
6212         osd_procfs_fini(o);
6213 out_scrub:
6214         osd_scrub_cleanup(env, o);
6215 out_site:
6216         lu_site_fini(&o->od_site);
6217 out_compat:
6218         osd_obj_map_fini(o);
6219 out_mnt:
6220         osd_umount(env, o);
6221 out:
6222         return rc;
6223 }
6224
6225 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6226                                           struct lu_device_type *t,
6227                                           struct lustre_cfg *cfg)
6228 {
6229         struct osd_device *o;
6230         int                rc;
6231
6232         OBD_ALLOC_PTR(o);
6233         if (o == NULL)
6234                 return ERR_PTR(-ENOMEM);
6235
6236         rc = dt_device_init(&o->od_dt_dev, t);
6237         if (rc == 0) {
6238                 /* Because the ctx might be revived in dt_device_init,
6239                  * refill the env here */
6240                 lu_env_refill((struct lu_env *)env);
6241                 rc = osd_device_init0(env, o, cfg);
6242                 if (rc)
6243                         dt_device_fini(&o->od_dt_dev);
6244         }
6245
6246         if (unlikely(rc != 0))
6247                 OBD_FREE_PTR(o);
6248
6249         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6250 }
6251
6252 static struct lu_device *osd_device_free(const struct lu_env *env,
6253                                          struct lu_device *d)
6254 {
6255         struct osd_device *o = osd_dev(d);
6256         ENTRY;
6257
6258         /* XXX: make osd top device in order to release reference */
6259         d->ld_site->ls_top_dev = d;
6260         lu_site_purge(env, d->ld_site, -1);
6261         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6262                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6263                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6264         }
6265         lu_site_fini(&o->od_site);
6266         dt_device_fini(&o->od_dt_dev);
6267         OBD_FREE_PTR(o);
6268         RETURN(NULL);
6269 }
6270
6271 static int osd_process_config(const struct lu_env *env,
6272                               struct lu_device *d, struct lustre_cfg *cfg)
6273 {
6274         struct osd_device               *o = osd_dev(d);
6275         int                             rc;
6276         ENTRY;
6277
6278         switch (cfg->lcfg_command) {
6279         case LCFG_SETUP:
6280                 rc = osd_mount(env, o, cfg);
6281                 break;
6282         case LCFG_CLEANUP:
6283                 lu_dev_del_linkage(d->ld_site, d);
6284                 rc = osd_shutdown(env, o);
6285                 break;
6286         case LCFG_PARAM:
6287                 LASSERT(&o->od_dt_dev);
6288                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6289                                               cfg, &o->od_dt_dev);
6290                 if (rc > 0 || rc == -ENOSYS)
6291                         rc = class_process_proc_param(PARAM_OST,
6292                                                       lprocfs_osd_obd_vars,
6293                                                       cfg, &o->od_dt_dev);
6294                 break;
6295         default:
6296                 rc = -ENOSYS;
6297         }
6298
6299         RETURN(rc);
6300 }
6301
6302 static int osd_recovery_complete(const struct lu_env *env,
6303                                  struct lu_device *d)
6304 {
6305         struct osd_device       *osd = osd_dev(d);
6306         int                      rc = 0;
6307         ENTRY;
6308
6309         if (osd->od_quota_slave == NULL)
6310                 RETURN(0);
6311
6312         /* start qsd instance on recovery completion, this notifies the quota
6313          * slave code that we are about to process new requests now */
6314         rc = qsd_start(env, osd->od_quota_slave);
6315         RETURN(rc);
6316 }
6317
6318 /*
6319  * we use exports to track all osd users
6320  */
6321 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6322                            struct obd_device *obd, struct obd_uuid *cluuid,
6323                            struct obd_connect_data *data, void *localdata)
6324 {
6325         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6326         struct lustre_handle  conn;
6327         int                   rc;
6328         ENTRY;
6329
6330         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6331
6332         rc = class_connect(&conn, obd, cluuid);
6333         if (rc)
6334                 RETURN(rc);
6335
6336         *exp = class_conn2export(&conn);
6337
6338         spin_lock(&osd->od_osfs_lock);
6339         osd->od_connects++;
6340         spin_unlock(&osd->od_osfs_lock);
6341
6342         RETURN(0);
6343 }
6344
6345 /*
6346  * once last export (we don't count self-export) disappeared
6347  * osd can be released
6348  */
6349 static int osd_obd_disconnect(struct obd_export *exp)
6350 {
6351         struct obd_device *obd = exp->exp_obd;
6352         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6353         int                rc, release = 0;
6354         ENTRY;
6355
6356         /* Only disconnect the underlying layers on the final disconnect. */
6357         spin_lock(&osd->od_osfs_lock);
6358         osd->od_connects--;
6359         if (osd->od_connects == 0)
6360                 release = 1;
6361         spin_unlock(&osd->od_osfs_lock);
6362
6363         rc = class_disconnect(exp); /* bz 9811 */
6364
6365         if (rc == 0 && release)
6366                 class_manual_cleanup(obd);
6367         RETURN(rc);
6368 }
6369
6370 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6371                        struct lu_device *dev)
6372 {
6373         struct osd_device       *osd    = osd_dev(dev);
6374         struct lr_server_data   *lsd    =
6375                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6376         int                      result = 0;
6377         ENTRY;
6378
6379         if (osd->od_quota_slave != NULL) {
6380                 /* set up quota slave objects */
6381                 result = qsd_prepare(env, osd->od_quota_slave);
6382                 if (result != 0)
6383                         RETURN(result);
6384         }
6385
6386         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6387 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6388                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6389                         osd->od_index_in_idif = 1;
6390                 } else {
6391                         osd->od_index_in_idif = 0;
6392                         result = osd_register_proc_index_in_idif(osd);
6393                         if (result != 0)
6394                                 RETURN(result);
6395                 }
6396 #else
6397                 osd->od_index_in_idif = 1;
6398 #endif
6399         }
6400
6401         result = osd_fid_init(env, osd);
6402
6403         RETURN(result);
6404 }
6405
6406 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6407                          struct lu_fid *fid, struct md_op_data *op_data)
6408 {
6409         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6410
6411         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6412 }
6413
6414 static const struct lu_object_operations osd_lu_obj_ops = {
6415         .loo_object_init      = osd_object_init,
6416         .loo_object_delete    = osd_object_delete,
6417         .loo_object_release   = osd_object_release,
6418         .loo_object_free      = osd_object_free,
6419         .loo_object_print     = osd_object_print,
6420         .loo_object_invariant = osd_object_invariant
6421 };
6422
6423 const struct lu_device_operations osd_lu_ops = {
6424         .ldo_object_alloc      = osd_object_alloc,
6425         .ldo_process_config    = osd_process_config,
6426         .ldo_recovery_complete = osd_recovery_complete,
6427         .ldo_prepare           = osd_prepare,
6428 };
6429
6430 static const struct lu_device_type_operations osd_device_type_ops = {
6431         .ldto_init = osd_type_init,
6432         .ldto_fini = osd_type_fini,
6433
6434         .ldto_start = osd_type_start,
6435         .ldto_stop  = osd_type_stop,
6436
6437         .ldto_device_alloc = osd_device_alloc,
6438         .ldto_device_free  = osd_device_free,
6439
6440         .ldto_device_init    = osd_device_init,
6441         .ldto_device_fini    = osd_device_fini
6442 };
6443
6444 static struct lu_device_type osd_device_type = {
6445         .ldt_tags     = LU_DEVICE_DT,
6446         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6447         .ldt_ops      = &osd_device_type_ops,
6448         .ldt_ctx_tags = LCT_LOCAL,
6449 };
6450
6451 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
6452 {
6453         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6454         struct super_block *sb = osd_sb(osd);
6455
6456         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
6457 }
6458
6459 /*
6460  * lprocfs legacy support.
6461  */
6462 static struct obd_ops osd_obd_device_ops = {
6463         .o_owner = THIS_MODULE,
6464         .o_connect      = osd_obd_connect,
6465         .o_disconnect   = osd_obd_disconnect,
6466         .o_fid_alloc    = osd_fid_alloc,
6467         .o_health_check = osd_health_check,
6468 };
6469
6470 static int __init osd_init(void)
6471 {
6472         int rc;
6473
6474         LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
6475 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
6476         /* please, try to keep osd_thread_info smaller than a page */
6477         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
6478 #endif
6479
6480         osd_oi_mod_init();
6481
6482         rc = lu_kmem_init(ldiskfs_caches);
6483         if (rc)
6484                 return rc;
6485
6486         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6487                                  lprocfs_osd_module_vars,
6488                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6489         if (rc)
6490                 lu_kmem_fini(ldiskfs_caches);
6491         return rc;
6492 }
6493
6494 static void __exit osd_exit(void)
6495 {
6496         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6497         lu_kmem_fini(ldiskfs_caches);
6498 }
6499
6500 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6501 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6502 MODULE_VERSION(LUSTRE_VERSION_STRING);
6503 MODULE_LICENSE("GPL");
6504
6505 module_init(osd_init);
6506 module_exit(osd_exit);