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, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_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
  70         push_ctxt(save, newctxt, NULL);
  71 }
  72
  73 static void osd_pop_ctxt(const struct osd_device *dev,
  74                          struct lvfs_run_ctxt *new,
  75                          struct lvfs_run_ctxt *save)
  76 {
  77         pop_ctxt(save, new, NULL);
  78 }
  79
  80 /* utility to make a directory */
  81 static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
  82                                    const char *name, int mode, int fix)
  83 {
  84         struct dentry *dchild;
  85         int err = 0;
  86         ENTRY;
  87
  88         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
  89         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
  90         dchild = ll_lookup_one_len(name, dir, strlen(name));
  91         if (IS_ERR(dchild))
  92                 GOTO(out_up, dchild);
  93
  94         if (dchild->d_inode) {
  95                 int old_mode = dchild->d_inode->i_mode;
  96                 if (!S_ISDIR(old_mode)) {
  97                         CERROR("found %s (%lu/%u) is mode %o\n", name,
  98                                dchild->d_inode->i_ino,
  99                                dchild->d_inode->i_generation, old_mode);
 100                         GOTO(out_err, err = -ENOTDIR);
 101                 }
 102
 103                 /* Fixup directory permissions if necessary */
 104                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 105                         CDEBUG(D_CONFIG,
 106                                "fixing permissions on %s from %o to %o\n",
 107                                name, old_mode, mode);
 108                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 109                                                   (old_mode & ~S_IALLUGO);
 110                         mark_inode_dirty(dchild->d_inode);
 111                 }
 112                 GOTO(out_up, dchild);
 113         }
 114
 115         err = vfs_mkdir(dir->d_inode, dchild, mode);
 116         if (err)
 117                 GOTO(out_err, err);
 118
 119         RETURN(dchild);
 120
 121 out_err:
 122         dput(dchild);
 123         dchild = ERR_PTR(err);
 124 out_up:
 125         return dchild;
 126 }
 127
 128 int osd_last_rcvd_subdir_count(struct osd_device *osd)
 129 {
 130         struct lr_server_data lsd;
 131         struct dentry        *dlast;
 132         loff_t                off;
 133         int                   rc = 0;
 134         int                   count = FILTER_SUBDIR_COUNT;
 135
 136         ENTRY;
 137
 138         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
 139                                   strlen(LAST_RCVD));
 140         if (IS_ERR(dlast))
 141                 return PTR_ERR(dlast);
 142         else if (dlast->d_inode == NULL)
 143                 goto out;
 144
 145         off = 0;
 146         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 147         if (rc == sizeof(lsd)) {
 148                 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
 149                        "subdir count = %d\n", lsd.lsd_uuid,
 150                        lsd.lsd_subdir_count);
 151                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 152                         count = le16_to_cpu(lsd.lsd_subdir_count);
 153         } else if (rc != 0) {
 154                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 155                 if (rc > 0)
 156                         rc = -EFAULT;
 157                 dput(dlast);
 158                 return rc;
 159         }
 160 out:
 161         dput(dlast);
 162         LASSERT(count > 0);
 163         return count;
 164 }
 165
 166 static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
 167 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
 168 {
 169         struct lvfs_run_ctxt    new;
 170         struct lvfs_run_ctxt    save;
 171         struct dentry           *parent;
 172         struct osd_mdobj_map    *omm;
 173         struct dentry           *d;
 174         struct osd_thread_info  *info = osd_oti_get(env);
 175         struct lu_fid           *fid = &info->oti_fid3;
 176         int                     rc = 0;
 177         ENTRY;
 178
 179         OBD_ALLOC_PTR(dev->od_mdt_map);
 180         if (dev->od_mdt_map == NULL)
 181                 RETURN(-ENOMEM);
 182
 183         omm = dev->od_mdt_map;
 184
 185         LASSERT(dev->od_fsops);
 186
 187         parent = osd_sb(dev)->s_root;
 188         osd_push_ctxt(dev, &new, &save);
 189
 190         d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
 191                          0755, 1);
 192         if (IS_ERR(d))
 193                 GOTO(cleanup, rc = PTR_ERR(d));
 194
 195         ldiskfs_set_inode_state(d->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 196         omm->omm_remote_parent = d;
 197
 198         /* Set LMA for remote parent inode */
 199         lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
 200         rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0);
 201
 202         GOTO(cleanup, rc);
 203
 204 cleanup:
 205         pop_ctxt(&save, &new, NULL);
 206         if (rc) {
 207                 if (omm->omm_remote_parent != NULL)
 208                         dput(omm->omm_remote_parent);
 209                 OBD_FREE_PTR(omm);
 210                 dev->od_mdt_map = NULL;
 211         }
 212         return rc;
 213 }
 214
 215 static void osd_mdt_fini(struct osd_device *osd)
 216 {
 217         struct osd_mdobj_map *omm = osd->od_mdt_map;
 218
 219         if (omm == NULL)
 220                 return;
 221
 222         if (omm->omm_remote_parent)
 223                 dput(omm->omm_remote_parent);
 224
 225         OBD_FREE_PTR(omm);
 226         osd->od_ost_map = NULL;
 227 }
 228
 229 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
 230                              struct osd_object *obj, struct osd_thandle *oh)
 231 {
 232         struct osd_mdobj_map    *omm = osd->od_mdt_map;
 233         struct osd_thread_info  *oti = osd_oti_get(env);
 234         struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
 235         char                    *name = oti->oti_name;
 236         struct dentry           *dentry;
 237         struct dentry           *parent;
 238         int                     rc;
 239
 240         /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
 241          * can identify this object quickly */
 242         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 243         if (rc != 0)
 244                 RETURN(rc);
 245
 246         lma->lma_incompat |= LMAI_REMOTE_PARENT;
 247         lustre_lma_swab(lma);
 248         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 249                              sizeof(*lma), XATTR_REPLACE);
 250         if (rc != 0)
 251                 RETURN(rc);
 252
 253         parent = omm->omm_remote_parent;
 254         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 255         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 256                                            name, strlen(name));
 257         mutex_lock(&parent->d_inode->i_mutex);
 258         rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
 259                                    NULL);
 260         CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
 261                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 262         LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
 263                  osd_name(osd), parent->d_inode->i_ino,
 264                  parent->d_inode->i_nlink);
 265         parent->d_inode->i_nlink++;
 266         mark_inode_dirty(parent->d_inode);
 267         mutex_unlock(&parent->d_inode->i_mutex);
 268         RETURN(rc);
 269 }
 270
 271 int osd_delete_from_remote_parent(const struct lu_env *env,
 272                                   struct osd_device *osd,
 273                                   struct osd_object *obj,
 274                                   struct osd_thandle *oh)
 275 {
 276         struct osd_mdobj_map       *omm = osd->od_mdt_map;
 277         struct osd_thread_info     *oti = osd_oti_get(env);
 278         struct lustre_mdt_attrs    *lma = &oti->oti_mdt_attrs;
 279         char                       *name = oti->oti_name;
 280         struct dentry              *dentry;
 281         struct dentry              *parent;
 282         struct ldiskfs_dir_entry_2 *de;
 283         struct buffer_head         *bh;
 284         int                        rc;
 285
 286         /* Check lma to see whether it is remote object */
 287         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 288         if (rc != 0)
 289                 RETURN(rc);
 290
 291         if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
 292                 RETURN(0);
 293
 294         parent = omm->omm_remote_parent;
 295         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 296         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 297                                            name, strlen(name));
 298         mutex_lock(&parent->d_inode->i_mutex);
 299         bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
 300         if (bh == NULL) {
 301                 mutex_unlock(&parent->d_inode->i_mutex);
 302                 RETURN(-ENOENT);
 303         }
 304         CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
 305                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 306         rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
 307         LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
 308                  osd_name(osd), parent->d_inode->i_ino,
 309                  parent->d_inode->i_nlink);
 310         parent->d_inode->i_nlink--;
 311         mark_inode_dirty(parent->d_inode);
 312         mutex_unlock(&parent->d_inode->i_mutex);
 313         brelse(bh);
 314
 315         /* Get rid of REMOTE_PARENT flag from incompat */
 316         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
 317         lustre_lma_swab(lma);
 318         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 319                              sizeof(*lma), XATTR_REPLACE);
 320         RETURN(rc);
 321 }
 322
 323 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 324                                 struct osd_device *osd,
 325                                 const struct lu_fid *fid,
 326                                 struct osd_inode_id *id)
 327 {
 328         struct osd_mdobj_map        *omm = osd->od_mdt_map;
 329         char                        *name = oti->oti_name;
 330         struct dentry               *parent;
 331         struct dentry               *dentry;
 332         struct ldiskfs_dir_entry_2 *de;
 333         struct buffer_head         *bh;
 334         int                         rc;
 335         ENTRY;
 336
 337         parent = omm->omm_remote_parent;
 338         sprintf(name, DFID_NOBRACE, PFID(fid));
 339         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
 340                                            name, strlen(name));
 341         mutex_lock(&parent->d_inode->i_mutex);
 342         bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
 343         if (bh == NULL) {
 344                 rc = -ENOENT;
 345         } else {
 346                 rc = 0;
 347                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 348                 brelse(bh);
 349         }
 350         mutex_unlock(&parent->d_inode->i_mutex);
 351         if (rc == 0)
 352                 osd_add_oi_cache(oti, osd, id, fid);
 353         RETURN(rc);
 354 }
 355
 356 /*
 357  * directory structure on legacy OST:
 358  *
 359  * O/<seq>/d0-31/<objid>
 360  * O/<seq>/LAST_ID
 361  * last_rcvd
 362  * LAST_GROUP
 363  * CONFIGS
 364  *
 365  */
 366 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
 367 {
 368         struct lvfs_run_ctxt     new;
 369         struct lvfs_run_ctxt     save;
 370         struct dentry           *rootd = osd_sb(dev)->s_root;
 371         struct dentry           *d;
 372         struct osd_thread_info  *info = osd_oti_get(env);
 373         struct inode            *inode;
 374         struct lu_fid           *fid = &info->oti_fid3;
 375         int                      rc;
 376         ENTRY;
 377
 378         OBD_ALLOC_PTR(dev->od_ost_map);
 379         if (dev->od_ost_map == NULL)
 380                 RETURN(-ENOMEM);
 381
 382         /* to get subdir count from last_rcvd */
 383         rc = osd_last_rcvd_subdir_count(dev);
 384         if (rc < 0) {
 385                 OBD_FREE_PTR(dev->od_ost_map);
 386                 RETURN(rc);
 387         }
 388
 389         dev->od_ost_map->om_subdir_count = rc;
 390         rc = 0;
 391
 392         CFS_INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 393         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 394         sema_init(&dev->od_ost_map->om_dir_init_sem, 1);
 395
 396         LASSERT(dev->od_fsops);
 397         osd_push_ctxt(dev, &new, &save);
 398
 399         d = ll_lookup_one_len("O", rootd, strlen("O"));
 400         if (IS_ERR(d))
 401                 GOTO(cleanup, rc = PTR_ERR(d));
 402         if (d->d_inode == NULL) {
 403                 dput(d);
 404                 /* The lookup() may be called again inside simple_mkdir().
 405                  * Since the repeated lookup() only be called for "/O" at
 406                  * mount time, it will not affect the whole performance. */
 407                 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 408                 if (IS_ERR(d))
 409                         GOTO(cleanup, rc = PTR_ERR(d));
 410
 411                 /* It is quite probably that the device is new formatted. */
 412                 dev->od_maybe_new = 1;
 413         }
 414
 415         inode = d->d_inode;
 416         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
 417         dev->od_ost_map->om_root = d;
 418
 419         /* 'What the @fid is' is not imporatant, because the object
 420          * has no OI mapping, and only is visible inside the OSD.*/
 421         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 422         rc = osd_ea_fid_set(info, inode, fid,
 423                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 424
 425         GOTO(cleanup, rc);
 426
 427 cleanup:
 428         osd_pop_ctxt(dev, &new, &save);
 429         if (IS_ERR(d)) {
 430                 OBD_FREE_PTR(dev->od_ost_map);
 431                 RETURN(PTR_ERR(d));
 432         }
 433         return rc;
 434 }
 435
 436 static void osd_seq_free(struct osd_obj_map *map,
 437                          struct osd_obj_seq *osd_seq)
 438 {
 439         int j;
 440
 441         cfs_list_del_init(&osd_seq->oos_seq_list);
 442
 443         if (osd_seq->oos_dirs) {
 444                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 445                         if (osd_seq->oos_dirs[j])
 446                                 dput(osd_seq->oos_dirs[j]);
 447                 }
 448                 OBD_FREE(osd_seq->oos_dirs,
 449                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 450         }
 451
 452         if (osd_seq->oos_root)
 453                 dput(osd_seq->oos_root);
 454
 455         OBD_FREE_PTR(osd_seq);
 456 }
 457
 458 static void osd_ost_fini(struct osd_device *osd)
 459 {
 460         struct osd_obj_seq    *osd_seq;
 461         struct osd_obj_seq    *tmp;
 462         struct osd_obj_map    *map = osd->od_ost_map;
 463         ENTRY;
 464
 465         if (map == NULL)
 466                 return;
 467
 468         write_lock(&map->om_seq_list_lock);
 469         cfs_list_for_each_entry_safe(osd_seq, tmp,
 470                                      &map->om_seq_list,
 471                                      oos_seq_list) {
 472                 osd_seq_free(map, osd_seq);
 473         }
 474         write_unlock(&map->om_seq_list_lock);
 475         if (map->om_root)
 476                 dput(map->om_root);
 477         OBD_FREE_PTR(map);
 478         osd->od_ost_map = NULL;
 479         EXIT;
 480 }
 481
 482 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 483 {
 484         int rc;
 485         ENTRY;
 486
 487         /* prepare structures for OST */
 488         rc = osd_ost_init(env, dev);
 489         if (rc)
 490                 RETURN(rc);
 491
 492         /* prepare structures for MDS */
 493         rc = osd_mdt_init(env, dev);
 494
 495         RETURN(rc);
 496 }
 497
 498 struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq)
 499 {
 500         struct osd_obj_seq *osd_seq;
 501
 502         cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 503                 if (osd_seq->oos_seq == seq)
 504                         return osd_seq;
 505         }
 506         return NULL;
 507 }
 508
 509 struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
 510 {
 511         struct osd_obj_seq *osd_seq;
 512
 513         read_lock(&map->om_seq_list_lock);
 514         osd_seq = osd_seq_find_locked(map, seq);
 515         read_unlock(&map->om_seq_list_lock);
 516         return osd_seq;
 517 }
 518
 519 void osd_obj_map_fini(struct osd_device *dev)
 520 {
 521         osd_ost_fini(dev);
 522         osd_mdt_fini(dev);
 523 }
 524
 525 /**
 526  * Update the specified OI mapping.
 527  *
 528  * \retval   1, changed nothing
 529  * \retval   0, changed successfully
 530  * \retval -ve, on error
 531  */
 532 static int osd_obj_update_entry(struct osd_thread_info *info,
 533                                 struct osd_device *osd,
 534                                 struct dentry *dir, const char *name,
 535                                 const struct lu_fid *fid,
 536                                 const struct osd_inode_id *id,
 537                                 struct thandle *th)
 538 {
 539         struct inode               *parent = dir->d_inode;
 540         struct osd_thandle         *oh;
 541         struct dentry              *child;
 542         struct ldiskfs_dir_entry_2 *de;
 543         struct buffer_head         *bh;
 544         struct inode               *inode;
 545         struct dentry              *dentry = &info->oti_obj_dentry;
 546         struct osd_inode_id        *oi_id  = &info->oti_id3;
 547         struct lustre_mdt_attrs    *lma    = &info->oti_mdt_attrs;
 548         struct lu_fid              *oi_fid = &lma->lma_self_fid;
 549         int                         rc;
 550         ENTRY;
 551
 552         oh = container_of(th, struct osd_thandle, ot_super);
 553         LASSERT(oh->ot_handle != NULL);
 554         LASSERT(oh->ot_handle->h_transaction != NULL);
 555
 556         child = &info->oti_child_dentry;
 557         child->d_parent = dir;
 558         child->d_name.hash = 0;
 559         child->d_name.name = name;
 560         child->d_name.len = strlen(name);
 561
 562         ll_vfs_dq_init(parent);
 563         mutex_lock(&parent->i_mutex);
 564         bh = osd_ldiskfs_find_entry(parent, child, &de, NULL);
 565         if (bh == NULL)
 566                 GOTO(out, rc = -ENOENT);
 567
 568         if (le32_to_cpu(de->inode) == id->oii_ino)
 569                 GOTO(out, rc = 1);
 570
 571         osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 572         inode = osd_iget(info, osd, oi_id);
 573         if (IS_ERR(inode)) {
 574                 rc = PTR_ERR(inode);
 575                 if (rc == -ENOENT || rc == -ESTALE)
 576                         goto update;
 577                 GOTO(out, rc);
 578         }
 579
 580         rc = osd_get_lma(info, inode, dentry, lma);
 581         if (rc == -ENODATA) {
 582                 rc = osd_get_idif(info, inode, dentry, oi_fid);
 583                 if (rc > 0) {
 584                         oi_fid = NULL;
 585                         rc = 0;
 586                 }
 587         }
 588         iput(inode);
 589
 590         /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
 591          * either a crashed object or a uninitialized one. Replace it. */
 592         if (rc == -ENODATA || oi_fid == NULL)
 593                 goto update;
 594
 595         if (rc != 0)
 596                 GOTO(out, rc);
 597
 598         if (lu_fid_eq(fid, oi_fid)) {
 599                 CERROR("%s: the FID "DFID" is used by two objects: "
 600                        "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
 601                        oi_id->oii_ino, oi_id->oii_gen,
 602                        id->oii_ino, id->oii_gen);
 603                 GOTO(out, rc = -EEXIST);
 604         }
 605
 606 update:
 607         /* There may be temporary inconsistency: On one hand, the new
 608          * object may be referenced by multiple entries, which is out
 609          * of our control unless we traverse the whole /O completely,
 610          * which is non-flat order and inefficient, should be avoided;
 611          * On the other hand, the old object may become orphan if it
 612          * is still valid. Since it was referenced by an invalid entry,
 613          * making it as invisible temporary may be not worse. OI scrub
 614          * will process it later. */
 615         rc = ldiskfs_journal_get_write_access(oh->ot_handle, bh);
 616         if (rc != 0)
 617                 GOTO(out, rc);
 618
 619         de->inode = cpu_to_le32(id->oii_ino);
 620         rc = ldiskfs_journal_dirty_metadata(oh->ot_handle, bh);
 621
 622         GOTO(out, rc);
 623
 624 out:
 625         brelse(bh);
 626         mutex_unlock(&parent->i_mutex);
 627         return rc;
 628 }
 629
 630 static int osd_obj_del_entry(struct osd_thread_info *info,
 631                              struct osd_device *osd,
 632                              struct dentry *dird, char *name,
 633                              struct thandle *th)
 634 {
 635         struct ldiskfs_dir_entry_2 *de;
 636         struct buffer_head         *bh;
 637         struct osd_thandle         *oh;
 638         struct dentry              *child;
 639         struct inode               *dir = dird->d_inode;
 640         int                         rc;
 641         ENTRY;
 642
 643         oh = container_of(th, struct osd_thandle, ot_super);
 644         LASSERT(oh->ot_handle != NULL);
 645         LASSERT(oh->ot_handle->h_transaction != NULL);
 646
 647
 648         child = &info->oti_child_dentry;
 649         child->d_name.hash = 0;
 650         child->d_name.name = name;
 651         child->d_name.len = strlen(name);
 652         child->d_parent = dird;
 653         child->d_inode = NULL;
 654
 655         ll_vfs_dq_init(dir);
 656         mutex_lock(&dir->i_mutex);
 657         rc = -ENOENT;
 658         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 659         if (bh) {
 660                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
 661                 brelse(bh);
 662         }
 663         mutex_unlock(&dir->i_mutex);
 664
 665         RETURN(rc);
 666 }
 667
 668 int osd_obj_add_entry(struct osd_thread_info *info,
 669                       struct osd_device *osd,
 670                       struct dentry *dir, char *name,
 671                       const struct osd_inode_id *id,
 672                       struct thandle *th)
 673 {
 674         struct osd_thandle *oh;
 675         struct dentry *child;
 676         struct inode *inode;
 677         int rc;
 678
 679         ENTRY;
 680
 681         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
 682                 RETURN(0);
 683
 684         oh = container_of(th, struct osd_thandle, ot_super);
 685         LASSERT(oh->ot_handle != NULL);
 686         LASSERT(oh->ot_handle->h_transaction != NULL);
 687
 688         inode = &info->oti_inode;
 689         inode->i_sb = osd_sb(osd);
 690         osd_id_to_inode(inode, id);
 691         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 692
 693         child = &info->oti_child_dentry;
 694         child->d_name.hash = 0;
 695         child->d_name.name = name;
 696         child->d_name.len = strlen(name);
 697         child->d_parent = dir;
 698         child->d_inode = inode;
 699
 700         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
 701                 inode->i_ino++;
 702
 703         ll_vfs_dq_init(dir->d_inode);
 704         mutex_lock(&dir->d_inode->i_mutex);
 705         rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
 706         mutex_unlock(&dir->d_inode->i_mutex);
 707
 708         RETURN(rc);
 709 }
 710
 711 /**
 712  * Use LPU64 for legacy OST sequences, but use LPX64i for new
 713  * sequences names, so that the O/{seq}/dN/{oid} more closely
 714  * follows the DFID/PFID format. This makes it easier to map from
 715  * debug messages to objects in the future, and the legacy space
 716  * of FID_SEQ_OST_MDT0 will be unused in the future.
 717  **/
 718 static inline void osd_seq_name(char *seq_name, size_t name_size, obd_seq seq)
 719 {
 720         snprintf(seq_name, name_size,
 721                  (fid_seq_is_rsvd(seq) ||
 722                   fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
 723                  fid_seq_is_idif(seq) ? 0 : seq);
 724 }
 725
 726 static inline void osd_oid_name(char *name, size_t name_size,
 727                                 const struct lu_fid *fid, obd_id id)
 728 {
 729         snprintf(name, name_size,
 730                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 731                   fid_seq_is_mdt0(fid_seq(fid)) ||
 732                   fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
 733 }
 734
 735 /* external locking is required */
 736 static int osd_seq_load_locked(struct osd_thread_info *info,
 737                                struct osd_device *osd,
 738                                struct osd_obj_seq *osd_seq)
 739 {
 740         struct osd_obj_map  *map = osd->od_ost_map;
 741         struct dentry       *seq_dir;
 742         struct inode        *inode;
 743         struct lu_fid       *fid = &info->oti_fid3;
 744         int                 rc = 0;
 745         int                 i;
 746         char                dir_name[32];
 747         ENTRY;
 748
 749         if (osd_seq->oos_root != NULL)
 750                 RETURN(0);
 751
 752         LASSERT(map);
 753         LASSERT(map->om_root);
 754
 755         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 756
 757         seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
 758         if (IS_ERR(seq_dir))
 759                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 760         else if (seq_dir->d_inode == NULL)
 761                 GOTO(out_put, rc = -EFAULT);
 762
 763         inode = seq_dir->d_inode;
 764         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
 765         osd_seq->oos_root = seq_dir;
 766
 767         /* 'What the @fid is' is not imporatant, because the object
 768          * has no OI mapping, and only is visible inside the OSD.*/
 769         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 770         rc = osd_ea_fid_set(info, inode, fid,
 771                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 772         if (rc != 0)
 773                 GOTO(out_put, rc);
 774
 775         LASSERT(osd_seq->oos_dirs == NULL);
 776         OBD_ALLOC(osd_seq->oos_dirs,
 777                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 778         if (osd_seq->oos_dirs == NULL)
 779                 GOTO(out_put, rc = -ENOMEM);
 780
 781         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 782                 struct dentry   *dir;
 783
 784                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 785                 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
 786                                    0700, 1);
 787                 if (IS_ERR(dir)) {
 788                         GOTO(out_free, rc = PTR_ERR(dir));
 789                 } else if (dir->d_inode == NULL) {
 790                         dput(dir);
 791                         GOTO(out_free, rc = -EFAULT);
 792                 }
 793
 794                 inode = dir->d_inode;
 795                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
 796                 osd_seq->oos_dirs[i] = dir;
 797
 798                 /* 'What the @fid is' is not imporatant, because the object
 799                  * has no OI mapping, and only is visible inside the OSD.*/
 800                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
 801                 rc = osd_ea_fid_set(info, inode, fid,
 802                                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 803                 if (rc != 0)
 804                         GOTO(out_free, rc);
 805         }
 806
 807         if (rc != 0) {
 808 out_free:
 809                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 810                         if (osd_seq->oos_dirs[i] != NULL)
 811                                 dput(osd_seq->oos_dirs[i]);
 812                 }
 813                 OBD_FREE(osd_seq->oos_dirs,
 814                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
 815 out_put:
 816                 dput(seq_dir);
 817                 osd_seq->oos_root = NULL;
 818         }
 819 out_err:
 820         RETURN(rc);
 821 }
 822
 823 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
 824                                         struct osd_device *osd, obd_seq seq)
 825 {
 826         struct osd_obj_map      *map;
 827         struct osd_obj_seq      *osd_seq;
 828         int                     rc = 0;
 829         ENTRY;
 830
 831         map = osd->od_ost_map;
 832         LASSERT(map);
 833         LASSERT(map->om_root);
 834
 835         osd_seq = osd_seq_find(map, seq);
 836         if (likely(osd_seq != NULL))
 837                 RETURN(osd_seq);
 838
 839         /* Serializing init process */
 840         down(&map->om_dir_init_sem);
 841
 842         /* Check whether the seq has been added */
 843         read_lock(&map->om_seq_list_lock);
 844         osd_seq = osd_seq_find_locked(map, seq);
 845         if (osd_seq != NULL) {
 846                 read_unlock(&map->om_seq_list_lock);
 847                 GOTO(cleanup, rc = 0);
 848         }
 849         read_unlock(&map->om_seq_list_lock);
 850
 851         OBD_ALLOC_PTR(osd_seq);
 852         if (osd_seq == NULL)
 853                 GOTO(cleanup, rc = -ENOMEM);
 854
 855         CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 856         osd_seq->oos_seq = seq;
 857         /* Init subdir count to be 32, but each seq can have
 858          * different subdir count */
 859         osd_seq->oos_subdir_count = map->om_subdir_count;
 860         rc = osd_seq_load_locked(info, osd, osd_seq);
 861         if (rc != 0)
 862                 GOTO(cleanup, rc);
 863
 864         write_lock(&map->om_seq_list_lock);
 865         cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 866         write_unlock(&map->om_seq_list_lock);
 867
 868 cleanup:
 869         up(&map->om_dir_init_sem);
 870         if (rc != 0) {
 871                 if (osd_seq != NULL)
 872                         OBD_FREE_PTR(osd_seq);
 873                 RETURN(ERR_PTR(rc));
 874         }
 875
 876         RETURN(osd_seq);
 877 }
 878
 879 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 880                        const struct lu_fid *fid, struct osd_inode_id *id)
 881 {
 882         struct osd_obj_map              *map;
 883         struct osd_obj_seq              *osd_seq;
 884         struct dentry                   *d_seq;
 885         struct dentry                   *child;
 886         struct ost_id                   *ostid = &info->oti_ostid;
 887         int                             dirn;
 888         char                            name[32];
 889         struct ldiskfs_dir_entry_2      *de;
 890         struct buffer_head              *bh;
 891         struct inode                    *dir;
 892         struct inode                    *inode;
 893         ENTRY;
 894
 895         /* on the very first lookup we find and open directories */
 896
 897         map = dev->od_ost_map;
 898         LASSERT(map);
 899         LASSERT(map->om_root);
 900
 901         fid_to_ostid(fid, ostid);
 902         osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
 903         if (IS_ERR(osd_seq))
 904                 RETURN(PTR_ERR(osd_seq));
 905
 906         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 907         d_seq = osd_seq->oos_dirs[dirn];
 908         LASSERT(d_seq);
 909
 910         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 911
 912         child = &info->oti_child_dentry;
 913         child->d_parent = d_seq;
 914         child->d_name.hash = 0;
 915         child->d_name.name = name;
 916         /* XXX: we can use rc from sprintf() instead of strlen() */
 917         child->d_name.len = strlen(name);
 918
 919         dir = d_seq->d_inode;
 920         mutex_lock(&dir->i_mutex);
 921         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 922         mutex_unlock(&dir->i_mutex);
 923
 924         if (bh == NULL)
 925                 RETURN(-ENOENT);
 926
 927         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 928         brelse(bh);
 929
 930         inode = osd_iget(info, dev, id);
 931         if (IS_ERR(inode))
 932                 RETURN(PTR_ERR(inode));
 933
 934         iput(inode);
 935         RETURN(0);
 936 }
 937
 938 int osd_obj_map_insert(struct osd_thread_info *info,
 939                        struct osd_device *osd,
 940                        const struct lu_fid *fid,
 941                        const struct osd_inode_id *id,
 942                        struct thandle *th)
 943 {
 944         struct osd_obj_map      *map;
 945         struct osd_obj_seq      *osd_seq;
 946         struct dentry           *d;
 947         struct ost_id           *ostid = &info->oti_ostid;
 948         obd_id                   oid;
 949         int                     dirn, rc = 0;
 950         char                    name[32];
 951         ENTRY;
 952
 953         map = osd->od_ost_map;
 954         LASSERT(map);
 955
 956         /* map fid to seq:objid */
 957         fid_to_ostid(fid, ostid);
 958
 959         oid = ostid_id(ostid);
 960         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
 961         if (IS_ERR(osd_seq))
 962                 RETURN(PTR_ERR(osd_seq));
 963
 964         dirn = oid & (osd_seq->oos_subdir_count - 1);
 965         d = osd_seq->oos_dirs[dirn];
 966         LASSERT(d);
 967
 968         osd_oid_name(name, sizeof(name), fid, oid);
 969
 970 again:
 971         rc = osd_obj_add_entry(info, osd, d, name, id, th);
 972         if (rc == -EEXIST) {
 973                 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
 974                 if (unlikely(rc == -ENOENT))
 975                         goto again;
 976
 977                 if (unlikely(rc == 1))
 978                         RETURN(0);
 979         }
 980
 981         RETURN(rc);
 982 }
 983
 984 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
 985                        const struct lu_fid *fid, struct thandle *th)
 986 {
 987         struct osd_obj_map      *map;
 988         struct osd_obj_seq      *osd_seq;
 989         struct dentry           *d;
 990         struct ost_id           *ostid = &info->oti_ostid;
 991         int                     dirn, rc = 0;
 992         char                    name[32];
 993         ENTRY;
 994
 995         map = osd->od_ost_map;
 996         LASSERT(map);
 997
 998         /* map fid to seq:objid */
 999         fid_to_ostid(fid, ostid);
1000
1001         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1002         if (IS_ERR(osd_seq))
1003                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1004
1005         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1006         d = osd_seq->oos_dirs[dirn];
1007         LASSERT(d);
1008
1009         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1010         rc = osd_obj_del_entry(info, osd, d, name, th);
1011 cleanup:
1012         RETURN(rc);
1013 }
1014
1015 int osd_obj_map_update(struct osd_thread_info *info,
1016                        struct osd_device *osd,
1017                        const struct lu_fid *fid,
1018                        const struct osd_inode_id *id,
1019                        struct thandle *th)
1020 {
1021         struct osd_obj_seq      *osd_seq;
1022         struct dentry           *d;
1023         struct ost_id           *ostid = &info->oti_ostid;
1024         int                     dirn, rc = 0;
1025         char                    name[32];
1026         ENTRY;
1027
1028         fid_to_ostid(fid, ostid);
1029         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1030         if (IS_ERR(osd_seq))
1031                 RETURN(PTR_ERR(osd_seq));
1032
1033         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1034         d = osd_seq->oos_dirs[dirn];
1035         LASSERT(d);
1036
1037         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1038         rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1039
1040         RETURN(rc);
1041 }
1042
1043 int osd_obj_map_recover(struct osd_thread_info *info,
1044                         struct osd_device *osd,
1045                         struct inode *src_parent,
1046                         struct dentry *src_child,
1047                         const struct lu_fid *fid)
1048 {
1049         struct osd_obj_seq         *osd_seq;
1050         struct dentry              *tgt_parent;
1051         struct dentry              *tgt_child = &info->oti_child_dentry;
1052         struct inode               *dir;
1053         struct inode               *inode     = src_child->d_inode;
1054         struct ost_id              *ostid     = &info->oti_ostid;
1055         handle_t                   *jh;
1056         struct ldiskfs_dir_entry_2 *de;
1057         struct buffer_head         *bh;
1058         char                        name[32];
1059         int                         dirn;
1060         int                         rc        = 0;
1061         ENTRY;
1062
1063         if (fid_is_last_id(fid)) {
1064                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1065                 if (IS_ERR(osd_seq))
1066                         RETURN(PTR_ERR(osd_seq));
1067
1068                 tgt_parent = osd_seq->oos_root;
1069                 tgt_child->d_name.name = "LAST_ID";
1070                 tgt_child->d_name.len = strlen("LAST_ID");
1071         } else {
1072                 fid_to_ostid(fid, ostid);
1073                 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1074                 if (IS_ERR(osd_seq))
1075                         RETURN(PTR_ERR(osd_seq));
1076
1077                 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1078                 tgt_parent = osd_seq->oos_dirs[dirn];
1079                 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1080                 tgt_child->d_name.name = name;
1081                 tgt_child->d_name.len = strlen(name);
1082         }
1083         LASSERT(tgt_parent != NULL);
1084
1085         dir = tgt_parent->d_inode;
1086         tgt_child->d_name.hash = 0;
1087         tgt_child->d_parent = tgt_parent;
1088         tgt_child->d_inode = inode;
1089
1090         /* The non-initialized src_child may be destroyed. */
1091         jh = ldiskfs_journal_start_sb(osd_sb(osd),
1092                                 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1093                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1094                                 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1095         if (IS_ERR(jh))
1096                 RETURN(PTR_ERR(jh));
1097
1098         ll_vfs_dq_init(src_parent);
1099         ll_vfs_dq_init(dir);
1100
1101         mutex_lock(&src_parent->i_mutex);
1102         mutex_lock(&dir->i_mutex);
1103         bh = osd_ldiskfs_find_entry(dir, tgt_child, &de, NULL);
1104         if (bh != NULL) {
1105                 /* XXX: If some other object occupied the same slot. And If such
1106                  *      inode is zero-sized and with SUID+SGID, then means it is
1107                  *      a new created one. Maybe we can remove it and insert the
1108                  *      original one back to the /O/<seq>/d<x>. But there are
1109                  *      something to be considered:
1110                  *
1111                  *      1) The OST-object under /lost+found has crashed LMA.
1112                  *         So it should not conflict with the current one.
1113                  *
1114                  *      2) There are race conditions that: someone may just want
1115                  *         to modify the current one. Even if the OI scrub takes
1116                  *         the object lock when remove the current one, it still
1117                  *         cause the modification to be lost becasue the target
1118                  *         has been removed when the RPC service thread waiting
1119                  *         for the lock.
1120                  *
1121                  *      So keep it there before we have suitable solution. */
1122                 brelse(bh);
1123                 mutex_unlock(&dir->i_mutex);
1124
1125                 rc = -EEXIST;
1126                 /* If the src object has never been modified, then remove it. */
1127                 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1128                     inode->i_mode & S_ISGID) {
1129                         bh = osd_ldiskfs_find_entry(src_parent, src_child, &de,
1130                                                     NULL);
1131                         if (unlikely(bh == NULL)) {
1132                                 mutex_unlock(&src_parent->i_mutex);
1133                                 ldiskfs_journal_stop(jh);
1134                                 RETURN(0);
1135                         }
1136
1137                         rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1138                         brelse(bh);
1139                         if (rc == 0) {
1140                                 drop_nlink(inode);
1141                                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1142                         }
1143                 }
1144                 mutex_unlock(&src_parent->i_mutex);
1145                 ldiskfs_journal_stop(jh);
1146                 RETURN(rc);
1147         }
1148
1149         bh = osd_ldiskfs_find_entry(src_parent, src_child, &de, NULL);
1150         if (unlikely(bh == NULL))
1151                 GOTO(unlock, rc = -ENOENT);
1152
1153         rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1154         brelse(bh);
1155         if (rc != 0)
1156                 GOTO(unlock, rc);
1157
1158         rc = osd_ldiskfs_add_entry(jh, tgt_child, inode, NULL);
1159
1160         GOTO(unlock, rc);
1161
1162 unlock:
1163         mutex_unlock(&dir->i_mutex);
1164         mutex_unlock(&src_parent->i_mutex);
1165         ldiskfs_journal_stop(jh);
1166         return rc;
1167 }
1168
1169 static struct dentry *
1170 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1171                      const struct lu_fid *fid, char **name)
1172 {
1173         struct dentry *root = ERR_PTR(-ENOENT);
1174
1175         if (fid_is_last_id(fid)) {
1176                 struct osd_obj_seq *osd_seq;
1177
1178                 /* on creation of LAST_ID we create O/<seq> hierarchy */
1179                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1180                 if (IS_ERR(osd_seq))
1181                         RETURN((struct dentry *)osd_seq);
1182
1183                 *name = "LAST_ID";
1184                 root = osd_seq->oos_root;
1185         } else {
1186                 *name = osd_lf_fid2name(fid);
1187                 if (*name == NULL)
1188                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1189                 else if ((*name)[0])
1190                         root = osd_sb(osd)->s_root;
1191         }
1192
1193         return root;
1194 }
1195
1196 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1197                         const struct lu_fid *fid, const struct osd_inode_id *id,
1198                         struct thandle *th)
1199 {
1200         struct dentry   *root;
1201         char            *name;
1202         int              rc;
1203         ENTRY;
1204
1205         root = osd_object_spec_find(info, osd, fid, &name);
1206         if (!IS_ERR(root)) {
1207                 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1208         } else {
1209                 rc = PTR_ERR(root);
1210                 if (rc == -ENOENT)
1211                         rc = 1;
1212         }
1213
1214         RETURN(rc);
1215 }
1216
1217 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1218                         const struct lu_fid *fid, const struct osd_inode_id *id,
1219                         struct thandle *th)
1220 {
1221         struct dentry   *root;
1222         char            *name;
1223         int              rc;
1224         ENTRY;
1225
1226         root = osd_object_spec_find(info, osd, fid, &name);
1227         if (!IS_ERR(root)) {
1228                 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1229         } else {
1230                 rc = PTR_ERR(root);
1231                 if (rc == -ENOENT)
1232                         rc = 0;
1233         }
1234
1235         RETURN(rc);
1236 }
1237
1238 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1239                         const struct lu_fid *fid, struct osd_inode_id *id)
1240 {
1241         struct dentry   *root;
1242         struct dentry   *dentry;
1243         struct inode    *inode;
1244         char            *name;
1245         int             rc = -ENOENT;
1246         ENTRY;
1247
1248         if (fid_is_last_id(fid)) {
1249                 struct osd_obj_seq *osd_seq;
1250
1251                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1252                 if (IS_ERR(osd_seq))
1253                         RETURN(PTR_ERR(osd_seq));
1254                 root = osd_seq->oos_root;
1255                 name = "LAST_ID";
1256         } else {
1257                 root = osd_sb(osd)->s_root;
1258                 name = osd_lf_fid2name(fid);
1259                 if (name == NULL || strlen(name) == 0)
1260                         RETURN(-ENOENT);
1261         }
1262
1263         dentry = ll_lookup_one_len(name, root, strlen(name));
1264         if (!IS_ERR(dentry)) {
1265                 inode = dentry->d_inode;
1266                 if (inode) {
1267                         if (is_bad_inode(inode)) {
1268                                 rc = -EIO;
1269                         } else {
1270                                 osd_id_gen(id, inode->i_ino,
1271                                            inode->i_generation);
1272                                 rc = 0;
1273                         }
1274                 }
1275                 /* if dentry is accessible after osd_compat_spec_insert it
1276                  * will still contain NULL inode, so don't keep it in cache */
1277                 d_invalidate(dentry);
1278                 dput(dentry);
1279         }
1280
1281         RETURN(rc);
1282 }