lustre/osd-ldiskfs/osd_compat.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) 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 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_compat.c
  37  *
  38  * on-disk structure for managing /O
  39  *
  40  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  41  */
  42
  43 /* LUSTRE_VERSION_CODE */
  44 #include <lustre_ver.h>
  45 /* prerequisite for linux/xattr.h */
  46 #include <linux/types.h>
  47 /* prerequisite for linux/xattr.h */
  48 #include <linux/fs.h>
  49 /* XATTR_{REPLACE,CREATE} */
  50 #include <linux/xattr.h>
  51
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  * OBD_FAIL_CHECK
  55  */
  56 #include <obd_support.h>
  57
  58 #include "osd_internal.h"
  59 #include "osd_oi.h"
  60
  61 static void osd_push_ctxt(const struct osd_device *dev,
  62                           struct lvfs_run_ctxt *newctxt,
  63                           struct lvfs_run_ctxt *save)
  64 {
  65         OBD_SET_CTXT_MAGIC(newctxt);
  66         newctxt->pwdmnt = dev->od_mnt;
  67         newctxt->pwd = dev->od_mnt->mnt_root;
  68         newctxt->fs = get_ds();
  69         newctxt->umask = current_umask();
  70         newctxt->dt = NULL;
  71
  72         push_ctxt(save, newctxt);
  73 }
  74
  75 /* utility to make a directory */
  76 static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
  77                                    const char *name, int mode, int fix)
  78 {
  79         struct dentry *dchild;
  80         int err = 0;
  81         ENTRY;
  82
  83         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
  84         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
  85         dchild = ll_lookup_one_len(name, dir, strlen(name));
  86         if (IS_ERR(dchild))
  87                 GOTO(out_up, dchild);
  88
  89         if (dchild->d_inode) {
  90                 int old_mode = dchild->d_inode->i_mode;
  91                 if (!S_ISDIR(old_mode)) {
  92                         CERROR("found %s (%lu/%u) is mode %o\n", name,
  93                                dchild->d_inode->i_ino,
  94                                dchild->d_inode->i_generation, old_mode);
  95                         GOTO(out_err, err = -ENOTDIR);
  96                 }
  97
  98                 /* Fixup directory permissions if necessary */
  99                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 100                         CDEBUG(D_CONFIG,
 101                                "fixing permissions on %s from %o to %o\n",
 102                                name, old_mode, mode);
 103                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 104                                                   (old_mode & ~S_IALLUGO);
 105                         mark_inode_dirty(dchild->d_inode);
 106                 }
 107                 GOTO(out_up, dchild);
 108         }
 109
 110         err = vfs_mkdir(dir->d_inode, dchild, mode);
 111         if (err)
 112                 GOTO(out_err, err);
 113
 114         RETURN(dchild);
 115
 116 out_err:
 117         dput(dchild);
 118         dchild = ERR_PTR(err);
 119 out_up:
 120         return dchild;
 121 }
 122
 123 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
 124 {
 125         struct lr_server_data lsd;
 126         struct dentry        *dlast;
 127         loff_t                off;
 128         int                   rc = 0;
 129         int                   count = OBJ_SUBDIR_COUNT;
 130
 131         ENTRY;
 132
 133         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
 134                                   strlen(LAST_RCVD));
 135         if (IS_ERR(dlast))
 136                 return PTR_ERR(dlast);
 137         else if (dlast->d_inode == NULL)
 138                 goto out;
 139
 140         off = 0;
 141         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 142         if (rc == sizeof(lsd)) {
 143                 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
 144                        "subdir count = %d\n", lsd.lsd_uuid,
 145                        lsd.lsd_subdir_count);
 146                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 147                         count = le16_to_cpu(lsd.lsd_subdir_count);
 148         } else if (rc != 0) {
 149                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 150                 if (rc > 0)
 151                         rc = -EFAULT;
 152                 dput(dlast);
 153                 return rc;
 154         }
 155 out:
 156         dput(dlast);
 157         LASSERT(count > 0);
 158         return count;
 159 }
 160
 161 static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
 162 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
 163 {
 164         struct lvfs_run_ctxt    new;
 165         struct lvfs_run_ctxt    save;
 166         struct dentry           *parent;
 167         struct osd_mdobj_map    *omm;
 168         struct dentry           *d;
 169         struct osd_thread_info  *info = osd_oti_get(env);
 170         struct lu_fid           *fid = &info->oti_fid3;
 171         int                     rc = 0;
 172         ENTRY;
 173
 174         OBD_ALLOC_PTR(dev->od_mdt_map);
 175         if (dev->od_mdt_map == NULL)
 176                 RETURN(-ENOMEM);
 177
 178         omm = dev->od_mdt_map;
 179
 180         parent = osd_sb(dev)->s_root;
 181         osd_push_ctxt(dev, &new, &save);
 182
 183         d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
 184                          0755, 1);
 185         if (IS_ERR(d))
 186                 GOTO(cleanup, rc = PTR_ERR(d));
 187
 188         omm->omm_remote_parent = d;
 189
 190         /* Set LMA for remote parent inode */
 191         lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
 192         rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0);
 193
 194         GOTO(cleanup, rc);
 195
 196 cleanup:
 197         pop_ctxt(&save, &new);
 198         if (rc) {
 199                 if (omm->omm_remote_parent != NULL)
 200                         dput(omm->omm_remote_parent);
 201                 OBD_FREE_PTR(omm);
 202                 dev->od_mdt_map = NULL;
 203         }
 204         return rc;
 205 }
 206
 207 static void osd_mdt_fini(struct osd_device *osd)
 208 {
 209         struct osd_mdobj_map *omm = osd->od_mdt_map;
 210
 211         if (omm == NULL)
 212                 return;
 213
 214         if (omm->omm_remote_parent)
 215                 dput(omm->omm_remote_parent);
 216
 217         OBD_FREE_PTR(omm);
 218         osd->od_ost_map = NULL;
 219 }
 220
 221 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
 222                              struct osd_object *obj, struct osd_thandle *oh)
 223 {
 224         struct osd_mdobj_map    *omm = osd->od_mdt_map;
 225         struct osd_thread_info  *oti = osd_oti_get(env);
 226         struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
 227         char                    *name = oti->oti_name;
 228         struct dentry           *dentry;
 229         struct dentry           *parent;
 230         int                     rc;
 231
 232         /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
 233          * can identify this object quickly */
 234         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 235         if (rc != 0)
 236                 RETURN(rc);
 237
 238         lma->lma_incompat |= LMAI_REMOTE_PARENT;
 239         lustre_lma_swab(lma);
 240         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 241                              sizeof(*lma), XATTR_REPLACE);
 242         if (rc != 0)
 243                 RETURN(rc);
 244
 245         parent = omm->omm_remote_parent;
 246         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 247         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 248                                            name, strlen(name));
 249         mutex_lock(&parent->d_inode->i_mutex);
 250         rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
 251                                    NULL);
 252         CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
 253                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 254         ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
 255         mark_inode_dirty(parent->d_inode);
 256         mutex_unlock(&parent->d_inode->i_mutex);
 257         RETURN(rc);
 258 }
 259
 260 int osd_delete_from_remote_parent(const struct lu_env *env,
 261                                   struct osd_device *osd,
 262                                   struct osd_object *obj,
 263                                   struct osd_thandle *oh)
 264 {
 265         struct osd_mdobj_map       *omm = osd->od_mdt_map;
 266         struct osd_thread_info     *oti = osd_oti_get(env);
 267         struct lustre_mdt_attrs    *lma = &oti->oti_mdt_attrs;
 268         char                       *name = oti->oti_name;
 269         struct dentry              *dentry;
 270         struct dentry              *parent;
 271         struct ldiskfs_dir_entry_2 *de;
 272         struct buffer_head         *bh;
 273         int                        rc;
 274
 275         /* Check lma to see whether it is remote object */
 276         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 277         if (rc != 0)
 278                 RETURN(rc);
 279
 280         if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
 281                 RETURN(0);
 282
 283         parent = omm->omm_remote_parent;
 284         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 285         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 286                                            name, strlen(name));
 287         mutex_lock(&parent->d_inode->i_mutex);
 288         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 289                                     NULL, NULL);
 290         if (bh == NULL) {
 291                 mutex_unlock(&parent->d_inode->i_mutex);
 292                 RETURN(-ENOENT);
 293         }
 294         CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
 295                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 296         rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
 297         ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
 298         mark_inode_dirty(parent->d_inode);
 299         mutex_unlock(&parent->d_inode->i_mutex);
 300         brelse(bh);
 301
 302         /* Get rid of REMOTE_PARENT flag from incompat */
 303         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
 304         lustre_lma_swab(lma);
 305         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 306                              sizeof(*lma), XATTR_REPLACE);
 307         RETURN(rc);
 308 }
 309
 310 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 311                                 struct osd_device *osd,
 312                                 const struct lu_fid *fid,
 313                                 struct osd_inode_id *id)
 314 {
 315         struct osd_mdobj_map        *omm = osd->od_mdt_map;
 316         char                        *name = oti->oti_name;
 317         struct dentry               *parent;
 318         struct dentry               *dentry;
 319         struct ldiskfs_dir_entry_2 *de;
 320         struct buffer_head         *bh;
 321         int                         rc;
 322         ENTRY;
 323
 324         parent = omm->omm_remote_parent;
 325         sprintf(name, DFID_NOBRACE, PFID(fid));
 326         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
 327                                            name, strlen(name));
 328         mutex_lock(&parent->d_inode->i_mutex);
 329         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 330                                     NULL, NULL);
 331         if (bh == NULL) {
 332                 rc = -ENOENT;
 333         } else {
 334                 rc = 0;
 335                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 336                 brelse(bh);
 337         }
 338         mutex_unlock(&parent->d_inode->i_mutex);
 339         if (rc == 0)
 340                 osd_add_oi_cache(oti, osd, id, fid);
 341         RETURN(rc);
 342 }
 343
 344 /*
 345  * directory structure on legacy OST:
 346  *
 347  * O/<seq>/d0-31/<objid>
 348  * O/<seq>/LAST_ID
 349  * last_rcvd
 350  * LAST_GROUP
 351  * CONFIGS
 352  *
 353  */
 354 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
 355 {
 356         struct lvfs_run_ctxt     new;
 357         struct lvfs_run_ctxt     save;
 358         struct dentry           *rootd = osd_sb(dev)->s_root;
 359         struct dentry           *d;
 360         struct osd_thread_info  *info = osd_oti_get(env);
 361         struct inode            *inode;
 362         struct lu_fid           *fid = &info->oti_fid3;
 363         int                      rc;
 364         ENTRY;
 365
 366         OBD_ALLOC_PTR(dev->od_ost_map);
 367         if (dev->od_ost_map == NULL)
 368                 RETURN(-ENOMEM);
 369
 370         /* to get subdir count from last_rcvd */
 371         rc = osd_last_rcvd_subdir_count(dev);
 372         if (rc < 0)
 373                 GOTO(cleanup_alloc, rc);
 374
 375         dev->od_ost_map->om_subdir_count = rc;
 376         rc = 0;
 377
 378         INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 379         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 380         mutex_init(&dev->od_ost_map->om_dir_init_mutex);
 381
 382         osd_push_ctxt(dev, &new, &save);
 383
 384         d = ll_lookup_one_len("O", rootd, strlen("O"));
 385         if (IS_ERR(d))
 386                 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 387         if (d->d_inode == NULL) {
 388                 dput(d);
 389                 /* The lookup() may be called again inside simple_mkdir().
 390                  * Since the repeated lookup() only be called for "/O" at
 391                  * mount time, it will not affect the whole performance. */
 392                 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 393                 if (IS_ERR(d))
 394                         GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 395
 396                 /* It is quite probably that the device is new formatted. */
 397                 dev->od_maybe_new = 1;
 398         }
 399
 400         inode = d->d_inode;
 401         dev->od_ost_map->om_root = d;
 402
 403         /* 'What the @fid is' is not imporatant, because the object
 404          * has no OI mapping, and only is visible inside the OSD.*/
 405         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 406         rc = osd_ea_fid_set(info, inode, fid,
 407                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 408         if (rc)
 409                 GOTO(cleanup_dentry, rc);
 410
 411         pop_ctxt(&save, &new);
 412         RETURN(0);
 413
 414 cleanup_dentry:
 415         dput(d);
 416 cleanup_ctxt:
 417         pop_ctxt(&save, &new);
 418 cleanup_alloc:
 419         OBD_FREE_PTR(dev->od_ost_map);
 420         return rc;
 421 }
 422
 423 static void osd_seq_free(struct osd_obj_map *map,
 424                          struct osd_obj_seq *osd_seq)
 425 {
 426         int j;
 427
 428         list_del_init(&osd_seq->oos_seq_list);
 429
 430         if (osd_seq->oos_dirs) {
 431                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 432                         if (osd_seq->oos_dirs[j])
 433                                 dput(osd_seq->oos_dirs[j]);
 434                 }
 435                 OBD_FREE(osd_seq->oos_dirs,
 436                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 437         }
 438
 439         if (osd_seq->oos_root)
 440                 dput(osd_seq->oos_root);
 441
 442         OBD_FREE_PTR(osd_seq);
 443 }
 444
 445 static void osd_ost_fini(struct osd_device *osd)
 446 {
 447         struct osd_obj_seq    *osd_seq;
 448         struct osd_obj_seq    *tmp;
 449         struct osd_obj_map    *map = osd->od_ost_map;
 450         ENTRY;
 451
 452         if (map == NULL)
 453                 return;
 454
 455         write_lock(&map->om_seq_list_lock);
 456         list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
 457                                  oos_seq_list) {
 458                 osd_seq_free(map, osd_seq);
 459         }
 460         write_unlock(&map->om_seq_list_lock);
 461         if (map->om_root)
 462                 dput(map->om_root);
 463         OBD_FREE_PTR(map);
 464         osd->od_ost_map = NULL;
 465         EXIT;
 466 }
 467
 468 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 469 {
 470         int rc;
 471         ENTRY;
 472
 473         /* prepare structures for OST */
 474         rc = osd_ost_init(env, dev);
 475         if (rc)
 476                 RETURN(rc);
 477
 478         /* prepare structures for MDS */
 479         rc = osd_mdt_init(env, dev);
 480         if (rc)
 481                 osd_ost_fini(dev);
 482
 483         RETURN(rc);
 484 }
 485
 486 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
 487 {
 488         struct osd_obj_seq *osd_seq;
 489
 490         list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 491                 if (osd_seq->oos_seq == seq)
 492                         return osd_seq;
 493         }
 494         return NULL;
 495 }
 496
 497 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
 498 {
 499         struct osd_obj_seq *osd_seq;
 500
 501         read_lock(&map->om_seq_list_lock);
 502         osd_seq = osd_seq_find_locked(map, seq);
 503         read_unlock(&map->om_seq_list_lock);
 504         return osd_seq;
 505 }
 506
 507 void osd_obj_map_fini(struct osd_device *dev)
 508 {
 509         osd_ost_fini(dev);
 510         osd_mdt_fini(dev);
 511 }
 512
 513 /**
 514  * Update the specified OI mapping.
 515  *
 516  * \retval   1, changed nothing
 517  * \retval   0, changed successfully
 518  * \retval -ve, on error
 519  */
 520 static int osd_obj_update_entry(struct osd_thread_info *info,
 521                                 struct osd_device *osd,
 522                                 struct dentry *dir, const char *name,
 523                                 const struct lu_fid *fid,
 524                                 const struct osd_inode_id *id,
 525                                 handle_t *th)
 526 {
 527         struct inode               *parent = dir->d_inode;
 528         struct dentry              *child;
 529         struct ldiskfs_dir_entry_2 *de;
 530         struct buffer_head         *bh;
 531         struct inode               *inode;
 532         struct dentry              *dentry = &info->oti_obj_dentry;
 533         struct osd_inode_id        *oi_id  = &info->oti_id3;
 534         struct lustre_mdt_attrs    *lma    = &info->oti_mdt_attrs;
 535         struct lu_fid              *oi_fid = &lma->lma_self_fid;
 536         int                         rc;
 537         ENTRY;
 538
 539         LASSERT(th != NULL);
 540         LASSERT(th->h_transaction != NULL);
 541
 542         child = &info->oti_child_dentry;
 543         child->d_parent = dir;
 544         child->d_name.hash = 0;
 545         child->d_name.name = name;
 546         child->d_name.len = strlen(name);
 547
 548         ll_vfs_dq_init(parent);
 549         mutex_lock(&parent->i_mutex);
 550         bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
 551         if (bh == NULL)
 552                 GOTO(out, rc = -ENOENT);
 553
 554         if (le32_to_cpu(de->inode) == id->oii_ino)
 555                 GOTO(out, rc = 1);
 556
 557         osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 558         inode = osd_iget(info, osd, oi_id);
 559         if (IS_ERR(inode)) {
 560                 rc = PTR_ERR(inode);
 561                 if (rc == -ENOENT || rc == -ESTALE)
 562                         goto update;
 563                 GOTO(out, rc);
 564         }
 565
 566         /* The EA inode should NOT be in OI, old OI scrub may added
 567          * such OI mapping by wrong, replace it. */
 568         if (unlikely(osd_is_ea_inode(inode))) {
 569                 iput(inode);
 570                 goto update;
 571         }
 572
 573         rc = osd_get_lma(info, inode, dentry, lma);
 574         if (rc == -ENODATA) {
 575                 rc = osd_get_idif(info, inode, dentry, oi_fid);
 576                 if (rc > 0 || rc == -ENODATA) {
 577                         oi_fid = NULL;
 578                         rc = 0;
 579                 }
 580         }
 581         iput(inode);
 582
 583         if (rc != 0)
 584                 GOTO(out, rc);
 585
 586         /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
 587          * either a crashed object or a uninitialized one. Replace it. */
 588         if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
 589                 CERROR("%s: the FID "DFID" is used by two objects: "
 590                        "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
 591                        oi_id->oii_ino, oi_id->oii_gen,
 592                        id->oii_ino, id->oii_gen);
 593                 GOTO(out, rc = -EEXIST);
 594         }
 595
 596         if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
 597                 __u32 idx1 = fid_idif_ost_idx(fid);
 598                 __u32 idx2 = fid_idif_ost_idx(oi_fid);
 599                 struct ost_id *ostid = &info->oti_ostid;
 600                 struct lu_fid *tfid = &info->oti_fid3;
 601
 602                 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
 603                          "invalid given FID "DFID", not match the "
 604                          "device index %u\n", PFID(fid), osd->od_index);
 605
 606                 if (idx1 != idx2) {
 607                         if (idx1 == 0 && idx2 == osd->od_index) {
 608                                 fid_to_ostid(fid, ostid);
 609                                 ostid_to_fid(tfid, ostid, idx2);
 610                                 if (lu_fid_eq(tfid, oi_fid)) {
 611                                         CERROR("%s: the FID "DFID" is used by "
 612                                                "two objects(2): %u/%u %u/%u\n",
 613                                                osd_name(osd), PFID(fid),
 614                                                oi_id->oii_ino, oi_id->oii_gen,
 615                                                id->oii_ino, id->oii_gen);
 616
 617                                         GOTO(out, rc = -EEXIST);
 618                                 }
 619                         } else if (idx2 == 0 && idx1 == osd->od_index) {
 620                                 fid_to_ostid(oi_fid, ostid);
 621                                 ostid_to_fid(tfid, ostid, idx1);
 622                                 if (lu_fid_eq(tfid, fid)) {
 623                                         CERROR("%s: the FID "DFID" is used by "
 624                                                "two objects(2): %u/%u %u/%u\n",
 625                                                osd_name(osd), PFID(fid),
 626                                                oi_id->oii_ino, oi_id->oii_gen,
 627                                                id->oii_ino, id->oii_gen);
 628
 629                                         GOTO(out, rc = -EEXIST);
 630                                 }
 631                         }
 632                 }
 633         }
 634
 635 update:
 636         /* There may be temporary inconsistency: On one hand, the new
 637          * object may be referenced by multiple entries, which is out
 638          * of our control unless we traverse the whole /O completely,
 639          * which is non-flat order and inefficient, should be avoided;
 640          * On the other hand, the old object may become orphan if it
 641          * is still valid. Since it was referenced by an invalid entry,
 642          * making it as invisible temporary may be not worse. OI scrub
 643          * will process it later. */
 644         rc = ldiskfs_journal_get_write_access(th, bh);
 645         if (rc != 0)
 646                 GOTO(out, rc);
 647
 648         de->inode = cpu_to_le32(id->oii_ino);
 649         rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
 650
 651         GOTO(out, rc);
 652
 653 out:
 654         brelse(bh);
 655         mutex_unlock(&parent->i_mutex);
 656         return rc;
 657 }
 658
 659 static int osd_obj_del_entry(struct osd_thread_info *info,
 660                              struct osd_device *osd,
 661                              struct dentry *dird, char *name,
 662                              handle_t *th)
 663 {
 664         struct ldiskfs_dir_entry_2 *de;
 665         struct buffer_head         *bh;
 666         struct dentry              *child;
 667         struct inode               *dir = dird->d_inode;
 668         int                         rc;
 669         ENTRY;
 670
 671         LASSERT(th != NULL);
 672         LASSERT(th->h_transaction != NULL);
 673
 674
 675         child = &info->oti_child_dentry;
 676         child->d_name.hash = 0;
 677         child->d_name.name = name;
 678         child->d_name.len = strlen(name);
 679         child->d_parent = dird;
 680         child->d_inode = NULL;
 681
 682         ll_vfs_dq_init(dir);
 683         mutex_lock(&dir->i_mutex);
 684         rc = -ENOENT;
 685         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 686         if (bh) {
 687                 rc = ldiskfs_delete_entry(th, dir, de, bh);
 688                 brelse(bh);
 689         }
 690         mutex_unlock(&dir->i_mutex);
 691
 692         RETURN(rc);
 693 }
 694
 695 static int osd_obj_add_entry(struct osd_thread_info *info,
 696                              struct osd_device *osd,
 697                              struct dentry *dir, char *name,
 698                              const struct osd_inode_id *id,
 699                              handle_t *th)
 700 {
 701         struct dentry *child;
 702         struct inode *inode;
 703         int rc;
 704
 705         ENTRY;
 706
 707         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
 708                 RETURN(0);
 709
 710         LASSERT(th != NULL);
 711         LASSERT(th->h_transaction != NULL);
 712
 713         inode = info->oti_inode;
 714         if (unlikely(inode == NULL)) {
 715                 struct ldiskfs_inode_info *lii;
 716                 OBD_ALLOC_PTR(lii);
 717                 if (lii == NULL)
 718                         RETURN(-ENOMEM);
 719                 inode = info->oti_inode = &lii->vfs_inode;
 720         }
 721
 722         inode->i_sb = osd_sb(osd);
 723         osd_id_to_inode(inode, id);
 724         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 725
 726         child = &info->oti_child_dentry;
 727         child->d_name.hash = 0;
 728         child->d_name.name = name;
 729         child->d_name.len = strlen(name);
 730         child->d_parent = dir;
 731         child->d_inode = inode;
 732
 733         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
 734                 inode->i_ino++;
 735
 736         ll_vfs_dq_init(dir->d_inode);
 737         mutex_lock(&dir->d_inode->i_mutex);
 738         rc = osd_ldiskfs_add_entry(th, child, inode, NULL);
 739         mutex_unlock(&dir->d_inode->i_mutex);
 740
 741         RETURN(rc);
 742 }
 743
 744 /**
 745  * Use LPU64 for legacy OST sequences, but use LPX64i for new
 746  * sequences names, so that the O/{seq}/dN/{oid} more closely
 747  * follows the DFID/PFID format. This makes it easier to map from
 748  * debug messages to objects in the future, and the legacy space
 749  * of FID_SEQ_OST_MDT0 will be unused in the future.
 750  **/
 751 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
 752 {
 753         snprintf(seq_name, name_size,
 754                  (fid_seq_is_rsvd(seq) ||
 755                   fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
 756                  fid_seq_is_idif(seq) ? 0 : seq);
 757 }
 758
 759 static inline void osd_oid_name(char *name, size_t name_size,
 760                                 const struct lu_fid *fid, u64 id)
 761 {
 762         snprintf(name, name_size,
 763                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 764                   fid_seq_is_mdt0(fid_seq(fid)) ||
 765                   fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
 766 }
 767
 768 /* external locking is required */
 769 static int osd_seq_load_locked(struct osd_thread_info *info,
 770                                struct osd_device *osd,
 771                                struct osd_obj_seq *osd_seq)
 772 {
 773         struct osd_obj_map  *map = osd->od_ost_map;
 774         struct dentry       *seq_dir;
 775         struct inode        *inode;
 776         struct lu_fid       *fid = &info->oti_fid3;
 777         int                 rc = 0;
 778         int                 i;
 779         char                dir_name[32];
 780         ENTRY;
 781
 782         if (osd_seq->oos_root != NULL)
 783                 RETURN(0);
 784
 785         LASSERT(map);
 786         LASSERT(map->om_root);
 787
 788         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 789
 790         seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
 791         if (IS_ERR(seq_dir))
 792                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 793         else if (seq_dir->d_inode == NULL)
 794                 GOTO(out_put, rc = -EFAULT);
 795
 796         inode = seq_dir->d_inode;
 797         osd_seq->oos_root = seq_dir;
 798
 799         /* 'What the @fid is' is not imporatant, because the object
 800          * has no OI mapping, and only is visible inside the OSD.*/
 801         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 802         rc = osd_ea_fid_set(info, inode, fid,
 803                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 804         if (rc != 0)
 805                 GOTO(out_put, rc);
 806
 807         LASSERT(osd_seq->oos_dirs == NULL);
 808         OBD_ALLOC(osd_seq->oos_dirs,
 809                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 810         if (osd_seq->oos_dirs == NULL)
 811                 GOTO(out_put, rc = -ENOMEM);
 812
 813         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 814                 struct dentry   *dir;
 815
 816                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 817                 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
 818                                    0700, 1);
 819                 if (IS_ERR(dir)) {
 820                         GOTO(out_free, rc = PTR_ERR(dir));
 821                 } else if (dir->d_inode == NULL) {
 822                         dput(dir);
 823                         GOTO(out_free, rc = -EFAULT);
 824                 }
 825
 826                 inode = dir->d_inode;
 827                 osd_seq->oos_dirs[i] = dir;
 828
 829                 /* 'What the @fid is' is not imporatant, because the object
 830                  * has no OI mapping, and only is visible inside the OSD.*/
 831                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
 832                 rc = osd_ea_fid_set(info, inode, fid,
 833                                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 834                 if (rc != 0)
 835                         GOTO(out_free, rc);
 836         }
 837
 838         if (rc != 0) {
 839 out_free:
 840                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 841                         if (osd_seq->oos_dirs[i] != NULL)
 842                                 dput(osd_seq->oos_dirs[i]);
 843                 }
 844                 OBD_FREE(osd_seq->oos_dirs,
 845                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
 846 out_put:
 847                 dput(seq_dir);
 848                 osd_seq->oos_root = NULL;
 849         }
 850 out_err:
 851         RETURN(rc);
 852 }
 853
 854 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
 855                                         struct osd_device *osd, u64 seq)
 856 {
 857         struct osd_obj_map      *map;
 858         struct osd_obj_seq      *osd_seq;
 859         int                     rc = 0;
 860         ENTRY;
 861
 862         map = osd->od_ost_map;
 863         LASSERT(map);
 864         LASSERT(map->om_root);
 865
 866         osd_seq = osd_seq_find(map, seq);
 867         if (likely(osd_seq != NULL))
 868                 RETURN(osd_seq);
 869
 870         /* Serializing init process */
 871         mutex_lock(&map->om_dir_init_mutex);
 872
 873         /* Check whether the seq has been added */
 874         read_lock(&map->om_seq_list_lock);
 875         osd_seq = osd_seq_find_locked(map, seq);
 876         if (osd_seq != NULL) {
 877                 read_unlock(&map->om_seq_list_lock);
 878                 GOTO(cleanup, rc = 0);
 879         }
 880         read_unlock(&map->om_seq_list_lock);
 881
 882         OBD_ALLOC_PTR(osd_seq);
 883         if (osd_seq == NULL)
 884                 GOTO(cleanup, rc = -ENOMEM);
 885
 886         INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 887         osd_seq->oos_seq = seq;
 888         /* Init subdir count to be 32, but each seq can have
 889          * different subdir count */
 890         osd_seq->oos_subdir_count = map->om_subdir_count;
 891         rc = osd_seq_load_locked(info, osd, osd_seq);
 892         if (rc != 0)
 893                 GOTO(cleanup, rc);
 894
 895         write_lock(&map->om_seq_list_lock);
 896         list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 897         write_unlock(&map->om_seq_list_lock);
 898
 899 cleanup:
 900         mutex_unlock(&map->om_dir_init_mutex);
 901         if (rc != 0) {
 902                 if (osd_seq != NULL)
 903                         OBD_FREE_PTR(osd_seq);
 904                 RETURN(ERR_PTR(rc));
 905         }
 906
 907         RETURN(osd_seq);
 908 }
 909
 910 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 911                        const struct lu_fid *fid, struct osd_inode_id *id)
 912 {
 913         struct osd_obj_map              *map;
 914         struct osd_obj_seq              *osd_seq;
 915         struct dentry                   *d_seq;
 916         struct dentry                   *child;
 917         struct ost_id                   *ostid = &info->oti_ostid;
 918         int                             dirn;
 919         char                            name[32];
 920         struct ldiskfs_dir_entry_2      *de;
 921         struct buffer_head              *bh;
 922         struct inode                    *dir;
 923         struct inode                    *inode;
 924         ENTRY;
 925
 926         /* on the very first lookup we find and open directories */
 927
 928         map = dev->od_ost_map;
 929         LASSERT(map);
 930         LASSERT(map->om_root);
 931
 932         fid_to_ostid(fid, ostid);
 933         osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
 934         if (IS_ERR(osd_seq))
 935                 RETURN(PTR_ERR(osd_seq));
 936
 937         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 938         d_seq = osd_seq->oos_dirs[dirn];
 939         LASSERT(d_seq);
 940
 941         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 942
 943         child = &info->oti_child_dentry;
 944         child->d_parent = d_seq;
 945         child->d_name.hash = 0;
 946         child->d_name.name = name;
 947         /* XXX: we can use rc from sprintf() instead of strlen() */
 948         child->d_name.len = strlen(name);
 949
 950         dir = d_seq->d_inode;
 951         mutex_lock(&dir->i_mutex);
 952         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 953         mutex_unlock(&dir->i_mutex);
 954
 955         if (bh == NULL)
 956                 RETURN(-ENOENT);
 957
 958         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 959         brelse(bh);
 960
 961         inode = osd_iget(info, dev, id);
 962         if (IS_ERR(inode)) {
 963                 int rc = PTR_ERR(inode);
 964
 965                 RETURN(rc == -ENOENT ? -ESTALE : rc);
 966         }
 967
 968         iput(inode);
 969         RETURN(0);
 970 }
 971
 972 int osd_obj_map_insert(struct osd_thread_info *info,
 973                        struct osd_device *osd,
 974                        const struct lu_fid *fid,
 975                        const struct osd_inode_id *id,
 976                        handle_t *th)
 977 {
 978         struct osd_obj_map      *map;
 979         struct osd_obj_seq      *osd_seq;
 980         struct dentry           *d;
 981         struct ost_id           *ostid = &info->oti_ostid;
 982         u64                      oid;
 983         int                     dirn, rc = 0;
 984         char                    name[32];
 985         ENTRY;
 986
 987         map = osd->od_ost_map;
 988         LASSERT(map);
 989
 990         /* map fid to seq:objid */
 991         fid_to_ostid(fid, ostid);
 992
 993         oid = ostid_id(ostid);
 994         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
 995         if (IS_ERR(osd_seq))
 996                 RETURN(PTR_ERR(osd_seq));
 997
 998         dirn = oid & (osd_seq->oos_subdir_count - 1);
 999         d = osd_seq->oos_dirs[dirn];
1000         LASSERT(d);
1001
1002         osd_oid_name(name, sizeof(name), fid, oid);
1003
1004 again:
1005         rc = osd_obj_add_entry(info, osd, d, name, id, th);
1006         if (rc == -EEXIST) {
1007                 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1008                 if (unlikely(rc == -ENOENT))
1009                         goto again;
1010
1011                 if (unlikely(rc == 1))
1012                         RETURN(0);
1013         }
1014
1015         RETURN(rc);
1016 }
1017
1018 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1019                        const struct lu_fid *fid, handle_t *th)
1020 {
1021         struct osd_obj_map      *map;
1022         struct osd_obj_seq      *osd_seq;
1023         struct dentry           *d;
1024         struct ost_id           *ostid = &info->oti_ostid;
1025         int                     dirn, rc = 0;
1026         char                    name[32];
1027         ENTRY;
1028
1029         map = osd->od_ost_map;
1030         LASSERT(map);
1031
1032         /* map fid to seq:objid */
1033         fid_to_ostid(fid, ostid);
1034
1035         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1036         if (IS_ERR(osd_seq))
1037                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1038
1039         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1040         d = osd_seq->oos_dirs[dirn];
1041         LASSERT(d);
1042
1043         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1044         rc = osd_obj_del_entry(info, osd, d, name, th);
1045 cleanup:
1046         RETURN(rc);
1047 }
1048
1049 int osd_obj_map_update(struct osd_thread_info *info,
1050                        struct osd_device *osd,
1051                        const struct lu_fid *fid,
1052                        const struct osd_inode_id *id,
1053                        handle_t *th)
1054 {
1055         struct osd_obj_seq      *osd_seq;
1056         struct dentry           *d;
1057         struct ost_id           *ostid = &info->oti_ostid;
1058         int                     dirn, rc = 0;
1059         char                    name[32];
1060         ENTRY;
1061
1062         fid_to_ostid(fid, ostid);
1063         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1064         if (IS_ERR(osd_seq))
1065                 RETURN(PTR_ERR(osd_seq));
1066
1067         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1068         d = osd_seq->oos_dirs[dirn];
1069         LASSERT(d);
1070
1071         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1072         rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1073
1074         RETURN(rc);
1075 }
1076
1077 int osd_obj_map_recover(struct osd_thread_info *info,
1078                         struct osd_device *osd,
1079                         struct inode *src_parent,
1080                         struct dentry *src_child,
1081                         const struct lu_fid *fid)
1082 {
1083         struct osd_obj_seq         *osd_seq;
1084         struct dentry              *tgt_parent;
1085         struct dentry              *tgt_child = &info->oti_child_dentry;
1086         struct inode               *dir;
1087         struct inode               *inode     = src_child->d_inode;
1088         struct ost_id              *ostid     = &info->oti_ostid;
1089         handle_t                   *jh;
1090         struct ldiskfs_dir_entry_2 *de;
1091         struct buffer_head         *bh;
1092         char                        name[32];
1093         int                         dirn;
1094         int                         rc        = 0;
1095         ENTRY;
1096
1097         if (fid_is_last_id(fid)) {
1098                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1099                 if (IS_ERR(osd_seq))
1100                         RETURN(PTR_ERR(osd_seq));
1101
1102                 tgt_parent = osd_seq->oos_root;
1103                 tgt_child->d_name.name = "LAST_ID";
1104                 tgt_child->d_name.len = strlen("LAST_ID");
1105         } else {
1106                 fid_to_ostid(fid, ostid);
1107                 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1108                 if (IS_ERR(osd_seq))
1109                         RETURN(PTR_ERR(osd_seq));
1110
1111                 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1112                 tgt_parent = osd_seq->oos_dirs[dirn];
1113                 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1114                 tgt_child->d_name.name = name;
1115                 tgt_child->d_name.len = strlen(name);
1116         }
1117         LASSERT(tgt_parent != NULL);
1118
1119         dir = tgt_parent->d_inode;
1120         tgt_child->d_name.hash = 0;
1121         tgt_child->d_parent = tgt_parent;
1122         tgt_child->d_inode = inode;
1123
1124         /* The non-initialized src_child may be destroyed. */
1125         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1126                                 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1127                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1128                                 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1129         if (IS_ERR(jh))
1130                 RETURN(PTR_ERR(jh));
1131
1132         ll_vfs_dq_init(src_parent);
1133         ll_vfs_dq_init(dir);
1134
1135         mutex_lock(&src_parent->i_mutex);
1136         mutex_lock(&dir->i_mutex);
1137         bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1138         if (bh != NULL) {
1139                 /* XXX: If some other object occupied the same slot. And If such
1140                  *      inode is zero-sized and with SUID+SGID, then means it is
1141                  *      a new created one. Maybe we can remove it and insert the
1142                  *      original one back to the /O/<seq>/d<x>. But there are
1143                  *      something to be considered:
1144                  *
1145                  *      1) The OST-object under /lost+found has crashed LMA.
1146                  *         So it should not conflict with the current one.
1147                  *
1148                  *      2) There are race conditions that: someone may just want
1149                  *         to modify the current one. Even if the OI scrub takes
1150                  *         the object lock when remove the current one, it still
1151                  *         cause the modification to be lost becasue the target
1152                  *         has been removed when the RPC service thread waiting
1153                  *         for the lock.
1154                  *
1155                  *      So keep it there before we have suitable solution. */
1156                 brelse(bh);
1157                 mutex_unlock(&dir->i_mutex);
1158                 mutex_unlock(&src_parent->i_mutex);
1159                 ldiskfs_journal_stop(jh);
1160
1161                 rc = -EEXIST;
1162                 /* If the src object has never been modified, then remove it. */
1163                 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1164                     inode->i_mode & S_ISGID) {
1165                         rc = ll_vfs_unlink(src_parent, src_child);
1166                         if (unlikely(rc == -ENOENT))
1167                                 rc = 0;
1168                 }
1169                 RETURN(rc);
1170         }
1171
1172         bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1173                                     NULL, NULL);
1174         if (unlikely(bh == NULL))
1175                 GOTO(unlock, rc = -ENOENT);
1176
1177         rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1178         brelse(bh);
1179         if (rc != 0)
1180                 GOTO(unlock, rc);
1181
1182         rc = osd_ldiskfs_add_entry(jh, tgt_child, inode, NULL);
1183
1184         GOTO(unlock, rc);
1185
1186 unlock:
1187         mutex_unlock(&dir->i_mutex);
1188         mutex_unlock(&src_parent->i_mutex);
1189         ldiskfs_journal_stop(jh);
1190         return rc;
1191 }
1192
1193 static struct dentry *
1194 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1195                      const struct lu_fid *fid, char **name)
1196 {
1197         struct dentry *root = ERR_PTR(-ENOENT);
1198
1199         if (fid_is_last_id(fid)) {
1200                 struct osd_obj_seq *osd_seq;
1201
1202                 /* on creation of LAST_ID we create O/<seq> hierarchy */
1203                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1204                 if (IS_ERR(osd_seq))
1205                         RETURN((struct dentry *)osd_seq);
1206
1207                 *name = "LAST_ID";
1208                 root = osd_seq->oos_root;
1209         } else {
1210                 *name = osd_lf_fid2name(fid);
1211                 if (*name == NULL)
1212                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1213                 else if ((*name)[0])
1214                         root = osd_sb(osd)->s_root;
1215         }
1216
1217         return root;
1218 }
1219
1220 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1221                         const struct lu_fid *fid, const struct osd_inode_id *id,
1222                         handle_t *th)
1223 {
1224         struct dentry   *root;
1225         char            *name = NULL;
1226         int              rc;
1227         ENTRY;
1228
1229         root = osd_object_spec_find(info, osd, fid, &name);
1230         if (!IS_ERR(root)) {
1231                 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1232         } else {
1233                 rc = PTR_ERR(root);
1234                 if (rc == -ENOENT)
1235                         rc = 1;
1236         }
1237
1238         RETURN(rc);
1239 }
1240
1241 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1242                         const struct lu_fid *fid, const struct osd_inode_id *id,
1243                         handle_t *th)
1244 {
1245         struct dentry   *root;
1246         char            *name = NULL;
1247         int              rc;
1248         ENTRY;
1249
1250         root = osd_object_spec_find(info, osd, fid, &name);
1251         if (!IS_ERR(root)) {
1252                 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1253         } else {
1254                 rc = PTR_ERR(root);
1255                 if (rc == -ENOENT)
1256                         rc = 0;
1257         }
1258
1259         RETURN(rc);
1260 }
1261
1262 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1263                         const struct lu_fid *fid, struct osd_inode_id *id)
1264 {
1265         struct dentry   *root;
1266         struct dentry   *dentry;
1267         struct inode    *inode;
1268         char            *name = NULL;
1269         int             rc = -ENOENT;
1270         ENTRY;
1271
1272         if (fid_is_last_id(fid)) {
1273                 struct osd_obj_seq *osd_seq;
1274
1275                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1276                 if (IS_ERR(osd_seq))
1277                         RETURN(PTR_ERR(osd_seq));
1278                 root = osd_seq->oos_root;
1279                 name = "LAST_ID";
1280         } else {
1281                 root = osd_sb(osd)->s_root;
1282                 name = osd_lf_fid2name(fid);
1283                 if (name == NULL || strlen(name) == 0)
1284                         RETURN(-ENOENT);
1285         }
1286
1287         dentry = ll_lookup_one_len(name, root, strlen(name));
1288         if (!IS_ERR(dentry)) {
1289                 inode = dentry->d_inode;
1290                 if (inode) {
1291                         if (is_bad_inode(inode)) {
1292                                 rc = -EIO;
1293                         } else {
1294                                 osd_id_gen(id, inode->i_ino,
1295                                            inode->i_generation);
1296                                 rc = 0;
1297                         }
1298                 }
1299                 /* if dentry is accessible after osd_compat_spec_insert it
1300                  * will still contain NULL inode, so don't keep it in cache */
1301                 d_invalidate(dentry);
1302                 dput(dentry);
1303         }
1304
1305         RETURN(rc);
1306 }