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