lustre/osd-ldiskfs/osd_compat.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_compat.c
  37  *
  38  * on-disk structure for managing /O
  39  *
  40  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  41  */
  42
  43 /* LUSTRE_VERSION_CODE */
  44 #include <lustre_ver.h>
  45 /* prerequisite for linux/xattr.h */
  46 #include <linux/types.h>
  47 /* prerequisite for linux/xattr.h */
  48 #include <linux/fs.h>
  49 /* XATTR_{REPLACE,CREATE} */
  50 #include <linux/xattr.h>
  51
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  * OBD_FAIL_CHECK
  55  */
  56 #include <obd_support.h>
  57
  58 #include "osd_internal.h"
  59 #include "osd_oi.h"
  60
  61 static void osd_push_ctxt(const struct osd_device *dev,
  62                           struct lvfs_run_ctxt *newctxt,
  63                           struct lvfs_run_ctxt *save)
  64 {
  65         OBD_SET_CTXT_MAGIC(newctxt);
  66         newctxt->pwdmnt = dev->od_mnt;
  67         newctxt->pwd = dev->od_mnt->mnt_root;
  68         newctxt->fs = get_ds();
  69         newctxt->umask = current_umask();
  70         newctxt->dt = NULL;
  71
  72         push_ctxt(save, newctxt);
  73 }
  74
  75 /* utility to make a directory */
  76 static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
  77                                    const char *name, int mode, int fix)
  78 {
  79         struct dentry *dchild;
  80         int err = 0;
  81         ENTRY;
  82
  83         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
  84         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
  85         dchild = ll_lookup_one_len(name, dir, strlen(name));
  86         if (IS_ERR(dchild))
  87                 GOTO(out_up, dchild);
  88
  89         if (dchild->d_inode) {
  90                 int old_mode = dchild->d_inode->i_mode;
  91                 if (!S_ISDIR(old_mode)) {
  92                         CERROR("found %s (%lu/%u) is mode %o\n", name,
  93                                dchild->d_inode->i_ino,
  94                                dchild->d_inode->i_generation, old_mode);
  95                         GOTO(out_err, err = -ENOTDIR);
  96                 }
  97
  98                 /* Fixup directory permissions if necessary */
  99                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 100                         CDEBUG(D_CONFIG,
 101                                "fixing permissions on %s from %o to %o\n",
 102                                name, old_mode, mode);
 103                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 104                                                   (old_mode & ~S_IALLUGO);
 105                         mark_inode_dirty(dchild->d_inode);
 106                 }
 107                 GOTO(out_up, dchild);
 108         }
 109
 110         err = vfs_mkdir(dir->d_inode, dchild, mode);
 111         if (err)
 112                 GOTO(out_err, err);
 113
 114         RETURN(dchild);
 115
 116 out_err:
 117         dput(dchild);
 118         dchild = ERR_PTR(err);
 119 out_up:
 120         return dchild;
 121 }
 122
 123 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
 124 {
 125         struct lr_server_data lsd;
 126         struct dentry        *dlast;
 127         loff_t                off;
 128         int                   rc = 0;
 129         int                   count = OBJ_SUBDIR_COUNT;
 130
 131         ENTRY;
 132
 133         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
 134                                   strlen(LAST_RCVD));
 135         if (IS_ERR(dlast))
 136                 return PTR_ERR(dlast);
 137         else if (dlast->d_inode == NULL)
 138                 goto out;
 139
 140         off = 0;
 141         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 142         if (rc == sizeof(lsd)) {
 143                 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
 144                        "subdir count = %d\n", lsd.lsd_uuid,
 145                        lsd.lsd_subdir_count);
 146                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 147                         count = le16_to_cpu(lsd.lsd_subdir_count);
 148         } else if (rc != 0) {
 149                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 150                 if (rc > 0)
 151                         rc = -EFAULT;
 152                 dput(dlast);
 153                 return rc;
 154         }
 155 out:
 156         dput(dlast);
 157         LASSERT(count > 0);
 158         return count;
 159 }
 160
 161 static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
 162 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
 163 {
 164         struct lvfs_run_ctxt    new;
 165         struct lvfs_run_ctxt    save;
 166         struct dentry           *parent;
 167         struct osd_mdobj_map    *omm;
 168         struct dentry           *d;
 169         struct osd_thread_info  *info = osd_oti_get(env);
 170         struct lu_fid           *fid = &info->oti_fid3;
 171         int                     rc = 0;
 172         ENTRY;
 173
 174         OBD_ALLOC_PTR(dev->od_mdt_map);
 175         if (dev->od_mdt_map == NULL)
 176                 RETURN(-ENOMEM);
 177
 178         omm = dev->od_mdt_map;
 179
 180         parent = osd_sb(dev)->s_root;
 181         osd_push_ctxt(dev, &new, &save);
 182
 183         d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
 184                          0755, 1);
 185         if (IS_ERR(d))
 186                 GOTO(cleanup, rc = PTR_ERR(d));
 187
 188         omm->omm_remote_parent = d;
 189
 190         /* Set LMA for remote parent inode */
 191         lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
 192         rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0);
 193
 194         GOTO(cleanup, rc);
 195
 196 cleanup:
 197         pop_ctxt(&save, &new);
 198         if (rc) {
 199                 if (omm->omm_remote_parent != NULL)
 200                         dput(omm->omm_remote_parent);
 201                 OBD_FREE_PTR(omm);
 202                 dev->od_mdt_map = NULL;
 203         }
 204         return rc;
 205 }
 206
 207 static void osd_mdt_fini(struct osd_device *osd)
 208 {
 209         struct osd_mdobj_map *omm = osd->od_mdt_map;
 210
 211         if (omm == NULL)
 212                 return;
 213
 214         if (omm->omm_remote_parent)
 215                 dput(omm->omm_remote_parent);
 216
 217         OBD_FREE_PTR(omm);
 218         osd->od_ost_map = NULL;
 219 }
 220
 221 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
 222                              struct osd_object *obj, struct osd_thandle *oh)
 223 {
 224         struct osd_mdobj_map    *omm = osd->od_mdt_map;
 225         struct osd_thread_info  *oti = osd_oti_get(env);
 226         struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
 227         char                    *name = oti->oti_name;
 228         struct dentry           *dentry;
 229         struct dentry           *parent;
 230         int                     rc;
 231
 232         /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
 233          * can identify this object quickly */
 234         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 235         if (rc != 0)
 236                 RETURN(rc);
 237
 238         lma->lma_incompat |= LMAI_REMOTE_PARENT;
 239         lustre_lma_swab(lma);
 240         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 241                              sizeof(*lma), XATTR_REPLACE);
 242         if (rc != 0)
 243                 RETURN(rc);
 244
 245         parent = omm->omm_remote_parent;
 246         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 247         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 248                                            name, strlen(name));
 249         mutex_lock(&parent->d_inode->i_mutex);
 250         rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
 251                                    NULL);
 252         CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
 253                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 254         ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
 255         mark_inode_dirty(parent->d_inode);
 256         mutex_unlock(&parent->d_inode->i_mutex);
 257         RETURN(rc);
 258 }
 259
 260 int osd_delete_from_remote_parent(const struct lu_env *env,
 261                                   struct osd_device *osd,
 262                                   struct osd_object *obj,
 263                                   struct osd_thandle *oh)
 264 {
 265         struct osd_mdobj_map       *omm = osd->od_mdt_map;
 266         struct osd_thread_info     *oti = osd_oti_get(env);
 267         struct lustre_mdt_attrs    *lma = &oti->oti_mdt_attrs;
 268         char                       *name = oti->oti_name;
 269         struct dentry              *dentry;
 270         struct dentry              *parent;
 271         struct ldiskfs_dir_entry_2 *de;
 272         struct buffer_head         *bh;
 273         int                        rc;
 274
 275         /* Check lma to see whether it is remote object */
 276         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 277         if (rc != 0)
 278                 RETURN(rc);
 279
 280         if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
 281                 RETURN(0);
 282
 283         parent = omm->omm_remote_parent;
 284         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 285         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 286                                            name, strlen(name));
 287         mutex_lock(&parent->d_inode->i_mutex);
 288         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 289                                     NULL, NULL);
 290         if (bh == NULL) {
 291                 mutex_unlock(&parent->d_inode->i_mutex);
 292                 RETURN(-ENOENT);
 293         }
 294         CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
 295                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 296         rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
 297         ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
 298         mark_inode_dirty(parent->d_inode);
 299         mutex_unlock(&parent->d_inode->i_mutex);
 300         brelse(bh);
 301
 302         /* Get rid of REMOTE_PARENT flag from incompat */
 303         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
 304         lustre_lma_swab(lma);
 305         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 306                              sizeof(*lma), XATTR_REPLACE);
 307         RETURN(rc);
 308 }
 309
 310 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 311                                 struct osd_device *osd,
 312                                 const struct lu_fid *fid,
 313                                 struct osd_inode_id *id)
 314 {
 315         struct osd_mdobj_map        *omm = osd->od_mdt_map;
 316         char                        *name = oti->oti_name;
 317         struct dentry               *parent;
 318         struct dentry               *dentry;
 319         struct ldiskfs_dir_entry_2 *de;
 320         struct buffer_head         *bh;
 321         int                         rc;
 322         ENTRY;
 323
 324         parent = omm->omm_remote_parent;
 325         sprintf(name, DFID_NOBRACE, PFID(fid));
 326         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
 327                                            name, strlen(name));
 328         mutex_lock(&parent->d_inode->i_mutex);
 329         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 330                                     NULL, NULL);
 331         if (bh == NULL) {
 332                 rc = -ENOENT;
 333         } else {
 334                 rc = 0;
 335                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 336                 brelse(bh);
 337         }
 338         mutex_unlock(&parent->d_inode->i_mutex);
 339         if (rc == 0)
 340                 osd_add_oi_cache(oti, osd, id, fid);
 341         RETURN(rc);
 342 }
 343
 344 /*
 345  * directory structure on legacy OST:
 346  *
 347  * O/<seq>/d0-31/<objid>
 348  * O/<seq>/LAST_ID
 349  * last_rcvd
 350  * LAST_GROUP
 351  * CONFIGS
 352  *
 353  */
 354 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
 355 {
 356         struct lvfs_run_ctxt     new;
 357         struct lvfs_run_ctxt     save;
 358         struct dentry           *rootd = osd_sb(dev)->s_root;
 359         struct dentry           *d;
 360         struct osd_thread_info  *info = osd_oti_get(env);
 361         struct inode            *inode;
 362         struct lu_fid           *fid = &info->oti_fid3;
 363         int                      rc;
 364         ENTRY;
 365
 366         OBD_ALLOC_PTR(dev->od_ost_map);
 367         if (dev->od_ost_map == NULL)
 368                 RETURN(-ENOMEM);
 369
 370         /* to get subdir count from last_rcvd */
 371         rc = osd_last_rcvd_subdir_count(dev);
 372         if (rc < 0)
 373                 GOTO(cleanup_alloc, rc);
 374
 375         dev->od_ost_map->om_subdir_count = rc;
 376         rc = 0;
 377
 378         INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 379         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 380         mutex_init(&dev->od_ost_map->om_dir_init_mutex);
 381
 382         osd_push_ctxt(dev, &new, &save);
 383
 384         d = ll_lookup_one_len("O", rootd, strlen("O"));
 385         if (IS_ERR(d))
 386                 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 387         if (d->d_inode == NULL) {
 388                 dput(d);
 389                 /* The lookup() may be called again inside simple_mkdir().
 390                  * Since the repeated lookup() only be called for "/O" at
 391                  * mount time, it will not affect the whole performance. */
 392                 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 393                 if (IS_ERR(d))
 394                         GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 395
 396                 /* It is quite probably that the device is new formatted. */
 397                 dev->od_maybe_new = 1;
 398         }
 399
 400         inode = d->d_inode;
 401         dev->od_ost_map->om_root = d;
 402
 403         /* 'What the @fid is' is not imporatant, because the object
 404          * has no OI mapping, and only is visible inside the OSD.*/
 405         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 406         rc = osd_ea_fid_set(info, inode, fid,
 407                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 408         if (rc)
 409                 GOTO(cleanup_dentry, rc);
 410
 411         pop_ctxt(&save, &new);
 412         RETURN(0);
 413
 414 cleanup_dentry:
 415         dput(d);
 416 cleanup_ctxt:
 417         pop_ctxt(&save, &new);
 418 cleanup_alloc:
 419         OBD_FREE_PTR(dev->od_ost_map);
 420         return rc;
 421 }
 422
 423 static void osd_seq_free(struct osd_obj_map *map,
 424                          struct osd_obj_seq *osd_seq)
 425 {
 426         int j;
 427
 428         list_del_init(&osd_seq->oos_seq_list);
 429
 430         if (osd_seq->oos_dirs) {
 431                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 432                         if (osd_seq->oos_dirs[j])
 433                                 dput(osd_seq->oos_dirs[j]);
 434                 }
 435                 OBD_FREE(osd_seq->oos_dirs,
 436                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 437         }
 438
 439         if (osd_seq->oos_root)
 440                 dput(osd_seq->oos_root);
 441
 442         OBD_FREE_PTR(osd_seq);
 443 }
 444
 445 static void osd_ost_fini(struct osd_device *osd)
 446 {
 447         struct osd_obj_seq    *osd_seq;
 448         struct osd_obj_seq    *tmp;
 449         struct osd_obj_map    *map = osd->od_ost_map;
 450         ENTRY;
 451
 452         if (map == NULL)
 453                 return;
 454
 455         write_lock(&map->om_seq_list_lock);
 456         list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
 457                                  oos_seq_list) {
 458                 osd_seq_free(map, osd_seq);
 459         }
 460         write_unlock(&map->om_seq_list_lock);
 461         if (map->om_root)
 462                 dput(map->om_root);
 463         OBD_FREE_PTR(map);
 464         osd->od_ost_map = NULL;
 465         EXIT;
 466 }
 467
 468 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 469 {
 470         int rc;
 471         ENTRY;
 472
 473         /* prepare structures for OST */
 474         rc = osd_ost_init(env, dev);
 475         if (rc)
 476                 RETURN(rc);
 477
 478         /* prepare structures for MDS */
 479         rc = osd_mdt_init(env, dev);
 480         if (rc)
 481                 osd_ost_fini(dev);
 482
 483         RETURN(rc);
 484 }
 485
 486 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
 487 {
 488         struct osd_obj_seq *osd_seq;
 489
 490         list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 491                 if (osd_seq->oos_seq == seq)
 492                         return osd_seq;
 493         }
 494         return NULL;
 495 }
 496
 497 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
 498 {
 499         struct osd_obj_seq *osd_seq;
 500
 501         read_lock(&map->om_seq_list_lock);
 502         osd_seq = osd_seq_find_locked(map, seq);
 503         read_unlock(&map->om_seq_list_lock);
 504         return osd_seq;
 505 }
 506
 507 void osd_obj_map_fini(struct osd_device *dev)
 508 {
 509         osd_ost_fini(dev);
 510         osd_mdt_fini(dev);
 511 }
 512
 513 /**
 514  * Update the specified OI mapping.
 515  *
 516  * \retval   1, changed nothing
 517  * \retval   0, changed successfully
 518  * \retval -ve, on error
 519  */
 520 static int osd_obj_update_entry(struct osd_thread_info *info,
 521                                 struct osd_device *osd,
 522                                 struct dentry *dir, const char *name,
 523                                 const struct lu_fid *fid,
 524                                 const struct osd_inode_id *id,
 525                                 handle_t *th)
 526 {
 527         struct inode               *parent = dir->d_inode;
 528         struct dentry              *child;
 529         struct ldiskfs_dir_entry_2 *de;
 530         struct buffer_head         *bh;
 531         struct inode               *inode;
 532         struct dentry              *dentry = &info->oti_obj_dentry;
 533         struct osd_inode_id        *oi_id  = &info->oti_id3;
 534         struct lustre_mdt_attrs    *lma    = &info->oti_mdt_attrs;
 535         struct lu_fid              *oi_fid = &lma->lma_self_fid;
 536         int                         rc;
 537         ENTRY;
 538
 539         LASSERT(th != NULL);
 540         LASSERT(th->h_transaction != NULL);
 541
 542         child = &info->oti_child_dentry;
 543         child->d_parent = dir;
 544         child->d_name.hash = 0;
 545         child->d_name.name = name;
 546         child->d_name.len = strlen(name);
 547
 548         ll_vfs_dq_init(parent);
 549         mutex_lock(&parent->i_mutex);
 550         bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
 551         if (bh == NULL)
 552                 GOTO(out, rc = -ENOENT);
 553
 554         if (le32_to_cpu(de->inode) == id->oii_ino)
 555                 GOTO(out, rc = 1);
 556
 557         osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 558         inode = osd_iget(info, osd, oi_id);
 559         if (IS_ERR(inode)) {
 560                 rc = PTR_ERR(inode);
 561                 if (rc == -ENOENT || rc == -ESTALE)
 562                         goto update;
 563                 GOTO(out, rc);
 564         }
 565
 566         rc = osd_get_lma(info, inode, dentry, lma);
 567         if (rc == -ENODATA) {
 568                 rc = osd_get_idif(info, inode, dentry, oi_fid);
 569                 if (rc > 0 || rc == -ENODATA) {
 570                         oi_fid = NULL;
 571                         rc = 0;
 572                 }
 573         }
 574         iput(inode);
 575
 576         if (rc != 0)
 577                 GOTO(out, rc);
 578
 579         /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
 580          * either a crashed object or a uninitialized one. Replace it. */
 581         if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
 582                 CERROR("%s: the FID "DFID" is used by two objects: "
 583                        "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
 584                        oi_id->oii_ino, oi_id->oii_gen,
 585                        id->oii_ino, id->oii_gen);
 586                 GOTO(out, rc = -EEXIST);
 587         }
 588
 589         if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
 590                 __u32 idx1 = fid_idif_ost_idx(fid);
 591                 __u32 idx2 = fid_idif_ost_idx(oi_fid);
 592                 struct ost_id *ostid = &info->oti_ostid;
 593                 struct lu_fid *tfid = &info->oti_fid3;
 594
 595                 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
 596                          "invalid given FID "DFID", not match the "
 597                          "device index %u\n", PFID(fid), osd->od_index);
 598
 599                 if (idx1 != idx2) {
 600                         if (idx1 == 0 && idx2 == osd->od_index) {
 601                                 fid_to_ostid(fid, ostid);
 602                                 ostid_to_fid(tfid, ostid, idx2);
 603                                 if (lu_fid_eq(tfid, oi_fid)) {
 604                                         CERROR("%s: the FID "DFID" is used by "
 605                                                "two objects(2): %u/%u %u/%u\n",
 606                                                osd_name(osd), PFID(fid),
 607                                                oi_id->oii_ino, oi_id->oii_gen,
 608                                                id->oii_ino, id->oii_gen);
 609
 610                                         GOTO(out, rc = -EEXIST);
 611                                 }
 612                         } else if (idx2 == 0 && idx1 == osd->od_index) {
 613                                 fid_to_ostid(oi_fid, ostid);
 614                                 ostid_to_fid(tfid, ostid, idx1);
 615                                 if (lu_fid_eq(tfid, fid)) {
 616                                         CERROR("%s: the FID "DFID" is used by "
 617                                                "two objects(2): %u/%u %u/%u\n",
 618                                                osd_name(osd), PFID(fid),
 619                                                oi_id->oii_ino, oi_id->oii_gen,
 620                                                id->oii_ino, id->oii_gen);
 621
 622                                         GOTO(out, rc = -EEXIST);
 623                                 }
 624                         }
 625                 }
 626         }
 627
 628 update:
 629         /* There may be temporary inconsistency: On one hand, the new
 630          * object may be referenced by multiple entries, which is out
 631          * of our control unless we traverse the whole /O completely,
 632          * which is non-flat order and inefficient, should be avoided;
 633          * On the other hand, the old object may become orphan if it
 634          * is still valid. Since it was referenced by an invalid entry,
 635          * making it as invisible temporary may be not worse. OI scrub
 636          * will process it later. */
 637         rc = ldiskfs_journal_get_write_access(th, bh);
 638         if (rc != 0)
 639                 GOTO(out, rc);
 640
 641         de->inode = cpu_to_le32(id->oii_ino);
 642         rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
 643
 644         GOTO(out, rc);
 645
 646 out:
 647         brelse(bh);
 648         mutex_unlock(&parent->i_mutex);
 649         return rc;
 650 }
 651
 652 static int osd_obj_del_entry(struct osd_thread_info *info,
 653                              struct osd_device *osd,
 654                              struct dentry *dird, char *name,
 655                              handle_t *th)
 656 {
 657         struct ldiskfs_dir_entry_2 *de;
 658         struct buffer_head         *bh;
 659         struct dentry              *child;
 660         struct inode               *dir = dird->d_inode;
 661         int                         rc;
 662         ENTRY;
 663
 664         LASSERT(th != NULL);
 665         LASSERT(th->h_transaction != NULL);
 666
 667
 668         child = &info->oti_child_dentry;
 669         child->d_name.hash = 0;
 670         child->d_name.name = name;
 671         child->d_name.len = strlen(name);
 672         child->d_parent = dird;
 673         child->d_inode = NULL;
 674
 675         ll_vfs_dq_init(dir);
 676         mutex_lock(&dir->i_mutex);
 677         rc = -ENOENT;
 678         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 679         if (bh) {
 680                 rc = ldiskfs_delete_entry(th, dir, de, bh);
 681                 brelse(bh);
 682         }
 683         mutex_unlock(&dir->i_mutex);
 684
 685         RETURN(rc);
 686 }
 687
 688 static int osd_obj_add_entry(struct osd_thread_info *info,
 689                              struct osd_device *osd,
 690                              struct dentry *dir, char *name,
 691                              const struct osd_inode_id *id,
 692                              handle_t *th)
 693 {
 694         struct dentry *child;
 695         struct inode *inode;
 696         int rc;
 697
 698         ENTRY;
 699
 700         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
 701                 RETURN(0);
 702
 703         LASSERT(th != NULL);
 704         LASSERT(th->h_transaction != NULL);
 705
 706         inode = info->oti_inode;
 707         if (unlikely(inode == NULL)) {
 708                 struct ldiskfs_inode_info *lii;
 709                 OBD_ALLOC_PTR(lii);
 710                 if (lii == NULL)
 711                         RETURN(-ENOMEM);
 712                 inode = info->oti_inode = &lii->vfs_inode;
 713         }
 714
 715         inode->i_sb = osd_sb(osd);
 716         osd_id_to_inode(inode, id);
 717         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 718
 719         child = &info->oti_child_dentry;
 720         child->d_name.hash = 0;
 721         child->d_name.name = name;
 722         child->d_name.len = strlen(name);
 723         child->d_parent = dir;
 724         child->d_inode = inode;
 725
 726         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
 727                 inode->i_ino++;
 728
 729         ll_vfs_dq_init(dir->d_inode);
 730         mutex_lock(&dir->d_inode->i_mutex);
 731         rc = osd_ldiskfs_add_entry(th, child, inode, NULL);
 732         mutex_unlock(&dir->d_inode->i_mutex);
 733
 734         RETURN(rc);
 735 }
 736
 737 /**
 738  * Use LPU64 for legacy OST sequences, but use LPX64i for new
 739  * sequences names, so that the O/{seq}/dN/{oid} more closely
 740  * follows the DFID/PFID format. This makes it easier to map from
 741  * debug messages to objects in the future, and the legacy space
 742  * of FID_SEQ_OST_MDT0 will be unused in the future.
 743  **/
 744 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
 745 {
 746         snprintf(seq_name, name_size,
 747                  (fid_seq_is_rsvd(seq) ||
 748                   fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
 749                  fid_seq_is_idif(seq) ? 0 : seq);
 750 }
 751
 752 static inline void osd_oid_name(char *name, size_t name_size,
 753                                 const struct lu_fid *fid, u64 id)
 754 {
 755         snprintf(name, name_size,
 756                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 757                   fid_seq_is_mdt0(fid_seq(fid)) ||
 758                   fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
 759 }
 760
 761 /* external locking is required */
 762 static int osd_seq_load_locked(struct osd_thread_info *info,
 763                                struct osd_device *osd,
 764                                struct osd_obj_seq *osd_seq)
 765 {
 766         struct osd_obj_map  *map = osd->od_ost_map;
 767         struct dentry       *seq_dir;
 768         struct inode        *inode;
 769         struct lu_fid       *fid = &info->oti_fid3;
 770         int                 rc = 0;
 771         int                 i;
 772         char                dir_name[32];
 773         ENTRY;
 774
 775         if (osd_seq->oos_root != NULL)
 776                 RETURN(0);
 777
 778         LASSERT(map);
 779         LASSERT(map->om_root);
 780
 781         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 782
 783         seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
 784         if (IS_ERR(seq_dir))
 785                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 786         else if (seq_dir->d_inode == NULL)
 787                 GOTO(out_put, rc = -EFAULT);
 788
 789         inode = seq_dir->d_inode;
 790         osd_seq->oos_root = seq_dir;
 791
 792         /* 'What the @fid is' is not imporatant, because the object
 793          * has no OI mapping, and only is visible inside the OSD.*/
 794         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 795         rc = osd_ea_fid_set(info, inode, fid,
 796                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 797         if (rc != 0)
 798                 GOTO(out_put, rc);
 799
 800         LASSERT(osd_seq->oos_dirs == NULL);
 801         OBD_ALLOC(osd_seq->oos_dirs,
 802                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 803         if (osd_seq->oos_dirs == NULL)
 804                 GOTO(out_put, rc = -ENOMEM);
 805
 806         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 807                 struct dentry   *dir;
 808
 809                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 810                 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
 811                                    0700, 1);
 812                 if (IS_ERR(dir)) {
 813                         GOTO(out_free, rc = PTR_ERR(dir));
 814                 } else if (dir->d_inode == NULL) {
 815                         dput(dir);
 816                         GOTO(out_free, rc = -EFAULT);
 817                 }
 818
 819                 inode = dir->d_inode;
 820                 osd_seq->oos_dirs[i] = dir;
 821
 822                 /* 'What the @fid is' is not imporatant, because the object
 823                  * has no OI mapping, and only is visible inside the OSD.*/
 824                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
 825                 rc = osd_ea_fid_set(info, inode, fid,
 826                                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 827                 if (rc != 0)
 828                         GOTO(out_free, rc);
 829         }
 830
 831         if (rc != 0) {
 832 out_free:
 833                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 834                         if (osd_seq->oos_dirs[i] != NULL)
 835                                 dput(osd_seq->oos_dirs[i]);
 836                 }
 837                 OBD_FREE(osd_seq->oos_dirs,
 838                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
 839 out_put:
 840                 dput(seq_dir);
 841                 osd_seq->oos_root = NULL;
 842         }
 843 out_err:
 844         RETURN(rc);
 845 }
 846
 847 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
 848                                         struct osd_device *osd, u64 seq)
 849 {
 850         struct osd_obj_map      *map;
 851         struct osd_obj_seq      *osd_seq;
 852         int                     rc = 0;
 853         ENTRY;
 854
 855         map = osd->od_ost_map;
 856         LASSERT(map);
 857         LASSERT(map->om_root);
 858
 859         osd_seq = osd_seq_find(map, seq);
 860         if (likely(osd_seq != NULL))
 861                 RETURN(osd_seq);
 862
 863         /* Serializing init process */
 864         mutex_lock(&map->om_dir_init_mutex);
 865
 866         /* Check whether the seq has been added */
 867         read_lock(&map->om_seq_list_lock);
 868         osd_seq = osd_seq_find_locked(map, seq);
 869         if (osd_seq != NULL) {
 870                 read_unlock(&map->om_seq_list_lock);
 871                 GOTO(cleanup, rc = 0);
 872         }
 873         read_unlock(&map->om_seq_list_lock);
 874
 875         OBD_ALLOC_PTR(osd_seq);
 876         if (osd_seq == NULL)
 877                 GOTO(cleanup, rc = -ENOMEM);
 878
 879         INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 880         osd_seq->oos_seq = seq;
 881         /* Init subdir count to be 32, but each seq can have
 882          * different subdir count */
 883         osd_seq->oos_subdir_count = map->om_subdir_count;
 884         rc = osd_seq_load_locked(info, osd, osd_seq);
 885         if (rc != 0)
 886                 GOTO(cleanup, rc);
 887
 888         write_lock(&map->om_seq_list_lock);
 889         list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 890         write_unlock(&map->om_seq_list_lock);
 891
 892 cleanup:
 893         mutex_unlock(&map->om_dir_init_mutex);
 894         if (rc != 0) {
 895                 if (osd_seq != NULL)
 896                         OBD_FREE_PTR(osd_seq);
 897                 RETURN(ERR_PTR(rc));
 898         }
 899
 900         RETURN(osd_seq);
 901 }
 902
 903 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 904                        const struct lu_fid *fid, struct osd_inode_id *id)
 905 {
 906         struct osd_obj_map              *map;
 907         struct osd_obj_seq              *osd_seq;
 908         struct dentry                   *d_seq;
 909         struct dentry                   *child;
 910         struct ost_id                   *ostid = &info->oti_ostid;
 911         int                             dirn;
 912         char                            name[32];
 913         struct ldiskfs_dir_entry_2      *de;
 914         struct buffer_head              *bh;
 915         struct inode                    *dir;
 916         struct inode                    *inode;
 917         ENTRY;
 918
 919         /* on the very first lookup we find and open directories */
 920
 921         map = dev->od_ost_map;
 922         LASSERT(map);
 923         LASSERT(map->om_root);
 924
 925         fid_to_ostid(fid, ostid);
 926         osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
 927         if (IS_ERR(osd_seq))
 928                 RETURN(PTR_ERR(osd_seq));
 929
 930         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 931         d_seq = osd_seq->oos_dirs[dirn];
 932         LASSERT(d_seq);
 933
 934         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 935
 936         child = &info->oti_child_dentry;
 937         child->d_parent = d_seq;
 938         child->d_name.hash = 0;
 939         child->d_name.name = name;
 940         /* XXX: we can use rc from sprintf() instead of strlen() */
 941         child->d_name.len = strlen(name);
 942
 943         dir = d_seq->d_inode;
 944         mutex_lock(&dir->i_mutex);
 945         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 946         mutex_unlock(&dir->i_mutex);
 947
 948         if (bh == NULL)
 949                 RETURN(-ENOENT);
 950
 951         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 952         brelse(bh);
 953
 954         inode = osd_iget(info, dev, id);
 955         if (IS_ERR(inode)) {
 956                 int rc = PTR_ERR(inode);
 957
 958                 RETURN(rc == -ENOENT ? -ESTALE : rc);
 959         }
 960
 961         iput(inode);
 962         RETURN(0);
 963 }
 964
 965 int osd_obj_map_insert(struct osd_thread_info *info,
 966                        struct osd_device *osd,
 967                        const struct lu_fid *fid,
 968                        const struct osd_inode_id *id,
 969                        handle_t *th)
 970 {
 971         struct osd_obj_map      *map;
 972         struct osd_obj_seq      *osd_seq;
 973         struct dentry           *d;
 974         struct ost_id           *ostid = &info->oti_ostid;
 975         u64                      oid;
 976         int                     dirn, rc = 0;
 977         char                    name[32];
 978         ENTRY;
 979
 980         map = osd->od_ost_map;
 981         LASSERT(map);
 982
 983         /* map fid to seq:objid */
 984         fid_to_ostid(fid, ostid);
 985
 986         oid = ostid_id(ostid);
 987         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
 988         if (IS_ERR(osd_seq))
 989                 RETURN(PTR_ERR(osd_seq));
 990
 991         dirn = oid & (osd_seq->oos_subdir_count - 1);
 992         d = osd_seq->oos_dirs[dirn];
 993         LASSERT(d);
 994
 995         osd_oid_name(name, sizeof(name), fid, oid);
 996
 997 again:
 998         rc = osd_obj_add_entry(info, osd, d, name, id, th);
 999         if (rc == -EEXIST) {
1000                 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1001                 if (unlikely(rc == -ENOENT))
1002                         goto again;
1003
1004                 if (unlikely(rc == 1))
1005                         RETURN(0);
1006         }
1007
1008         RETURN(rc);
1009 }
1010
1011 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1012                        const struct lu_fid *fid, handle_t *th)
1013 {
1014         struct osd_obj_map      *map;
1015         struct osd_obj_seq      *osd_seq;
1016         struct dentry           *d;
1017         struct ost_id           *ostid = &info->oti_ostid;
1018         int                     dirn, rc = 0;
1019         char                    name[32];
1020         ENTRY;
1021
1022         map = osd->od_ost_map;
1023         LASSERT(map);
1024
1025         /* map fid to seq:objid */
1026         fid_to_ostid(fid, ostid);
1027
1028         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1029         if (IS_ERR(osd_seq))
1030                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1031
1032         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1033         d = osd_seq->oos_dirs[dirn];
1034         LASSERT(d);
1035
1036         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1037         rc = osd_obj_del_entry(info, osd, d, name, th);
1038 cleanup:
1039         RETURN(rc);
1040 }
1041
1042 int osd_obj_map_update(struct osd_thread_info *info,
1043                        struct osd_device *osd,
1044                        const struct lu_fid *fid,
1045                        const struct osd_inode_id *id,
1046                        handle_t *th)
1047 {
1048         struct osd_obj_seq      *osd_seq;
1049         struct dentry           *d;
1050         struct ost_id           *ostid = &info->oti_ostid;
1051         int                     dirn, rc = 0;
1052         char                    name[32];
1053         ENTRY;
1054
1055         fid_to_ostid(fid, ostid);
1056         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1057         if (IS_ERR(osd_seq))
1058                 RETURN(PTR_ERR(osd_seq));
1059
1060         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1061         d = osd_seq->oos_dirs[dirn];
1062         LASSERT(d);
1063
1064         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1065         rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1066
1067         RETURN(rc);
1068 }
1069
1070 int osd_obj_map_recover(struct osd_thread_info *info,
1071                         struct osd_device *osd,
1072                         struct inode *src_parent,
1073                         struct dentry *src_child,
1074                         const struct lu_fid *fid)
1075 {
1076         struct osd_obj_seq         *osd_seq;
1077         struct dentry              *tgt_parent;
1078         struct dentry              *tgt_child = &info->oti_child_dentry;
1079         struct inode               *dir;
1080         struct inode               *inode     = src_child->d_inode;
1081         struct ost_id              *ostid     = &info->oti_ostid;
1082         handle_t                   *jh;
1083         struct ldiskfs_dir_entry_2 *de;
1084         struct buffer_head         *bh;
1085         char                        name[32];
1086         int                         dirn;
1087         int                         rc        = 0;
1088         ENTRY;
1089
1090         if (fid_is_last_id(fid)) {
1091                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1092                 if (IS_ERR(osd_seq))
1093                         RETURN(PTR_ERR(osd_seq));
1094
1095                 tgt_parent = osd_seq->oos_root;
1096                 tgt_child->d_name.name = "LAST_ID";
1097                 tgt_child->d_name.len = strlen("LAST_ID");
1098         } else {
1099                 fid_to_ostid(fid, ostid);
1100                 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1101                 if (IS_ERR(osd_seq))
1102                         RETURN(PTR_ERR(osd_seq));
1103
1104                 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1105                 tgt_parent = osd_seq->oos_dirs[dirn];
1106                 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1107                 tgt_child->d_name.name = name;
1108                 tgt_child->d_name.len = strlen(name);
1109         }
1110         LASSERT(tgt_parent != NULL);
1111
1112         dir = tgt_parent->d_inode;
1113         tgt_child->d_name.hash = 0;
1114         tgt_child->d_parent = tgt_parent;
1115         tgt_child->d_inode = inode;
1116
1117         /* The non-initialized src_child may be destroyed. */
1118         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1119                                 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1120                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1121                                 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1122         if (IS_ERR(jh))
1123                 RETURN(PTR_ERR(jh));
1124
1125         ll_vfs_dq_init(src_parent);
1126         ll_vfs_dq_init(dir);
1127
1128         mutex_lock(&src_parent->i_mutex);
1129         mutex_lock(&dir->i_mutex);
1130         bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1131         if (bh != NULL) {
1132                 /* XXX: If some other object occupied the same slot. And If such
1133                  *      inode is zero-sized and with SUID+SGID, then means it is
1134                  *      a new created one. Maybe we can remove it and insert the
1135                  *      original one back to the /O/<seq>/d<x>. But there are
1136                  *      something to be considered:
1137                  *
1138                  *      1) The OST-object under /lost+found has crashed LMA.
1139                  *         So it should not conflict with the current one.
1140                  *
1141                  *      2) There are race conditions that: someone may just want
1142                  *         to modify the current one. Even if the OI scrub takes
1143                  *         the object lock when remove the current one, it still
1144                  *         cause the modification to be lost becasue the target
1145                  *         has been removed when the RPC service thread waiting
1146                  *         for the lock.
1147                  *
1148                  *      So keep it there before we have suitable solution. */
1149                 brelse(bh);
1150                 mutex_unlock(&dir->i_mutex);
1151                 mutex_unlock(&src_parent->i_mutex);
1152                 ldiskfs_journal_stop(jh);
1153
1154                 rc = -EEXIST;
1155                 /* If the src object has never been modified, then remove it. */
1156                 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1157                     inode->i_mode & S_ISGID) {
1158                         rc = ll_vfs_unlink(src_parent, src_child);
1159                         if (unlikely(rc == -ENOENT))
1160                                 rc = 0;
1161                 }
1162                 RETURN(rc);
1163         }
1164
1165         bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1166                                     NULL, NULL);
1167         if (unlikely(bh == NULL))
1168                 GOTO(unlock, rc = -ENOENT);
1169
1170         rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1171         brelse(bh);
1172         if (rc != 0)
1173                 GOTO(unlock, rc);
1174
1175         rc = osd_ldiskfs_add_entry(jh, tgt_child, inode, NULL);
1176
1177         GOTO(unlock, rc);
1178
1179 unlock:
1180         mutex_unlock(&dir->i_mutex);
1181         mutex_unlock(&src_parent->i_mutex);
1182         ldiskfs_journal_stop(jh);
1183         return rc;
1184 }
1185
1186 static struct dentry *
1187 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1188                      const struct lu_fid *fid, char **name)
1189 {
1190         struct dentry *root = ERR_PTR(-ENOENT);
1191
1192         if (fid_is_last_id(fid)) {
1193                 struct osd_obj_seq *osd_seq;
1194
1195                 /* on creation of LAST_ID we create O/<seq> hierarchy */
1196                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1197                 if (IS_ERR(osd_seq))
1198                         RETURN((struct dentry *)osd_seq);
1199
1200                 *name = "LAST_ID";
1201                 root = osd_seq->oos_root;
1202         } else {
1203                 *name = osd_lf_fid2name(fid);
1204                 if (*name == NULL)
1205                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1206                 else if ((*name)[0])
1207                         root = osd_sb(osd)->s_root;
1208         }
1209
1210         return root;
1211 }
1212
1213 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1214                         const struct lu_fid *fid, const struct osd_inode_id *id,
1215                         handle_t *th)
1216 {
1217         struct dentry   *root;
1218         char            *name = NULL;
1219         int              rc;
1220         ENTRY;
1221
1222         root = osd_object_spec_find(info, osd, fid, &name);
1223         if (!IS_ERR(root)) {
1224                 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1225         } else {
1226                 rc = PTR_ERR(root);
1227                 if (rc == -ENOENT)
1228                         rc = 1;
1229         }
1230
1231         RETURN(rc);
1232 }
1233
1234 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1235                         const struct lu_fid *fid, const struct osd_inode_id *id,
1236                         handle_t *th)
1237 {
1238         struct dentry   *root;
1239         char            *name = NULL;
1240         int              rc;
1241         ENTRY;
1242
1243         root = osd_object_spec_find(info, osd, fid, &name);
1244         if (!IS_ERR(root)) {
1245                 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1246         } else {
1247                 rc = PTR_ERR(root);
1248                 if (rc == -ENOENT)
1249                         rc = 0;
1250         }
1251
1252         RETURN(rc);
1253 }
1254
1255 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1256                         const struct lu_fid *fid, struct osd_inode_id *id)
1257 {
1258         struct dentry   *root;
1259         struct dentry   *dentry;
1260         struct inode    *inode;
1261         char            *name = NULL;
1262         int             rc = -ENOENT;
1263         ENTRY;
1264
1265         if (fid_is_last_id(fid)) {
1266                 struct osd_obj_seq *osd_seq;
1267
1268                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1269                 if (IS_ERR(osd_seq))
1270                         RETURN(PTR_ERR(osd_seq));
1271                 root = osd_seq->oos_root;
1272                 name = "LAST_ID";
1273         } else {
1274                 root = osd_sb(osd)->s_root;
1275                 name = osd_lf_fid2name(fid);
1276                 if (name == NULL || strlen(name) == 0)
1277                         RETURN(-ENOENT);
1278         }
1279
1280         dentry = ll_lookup_one_len(name, root, strlen(name));
1281         if (!IS_ERR(dentry)) {
1282                 inode = dentry->d_inode;
1283                 if (inode) {
1284                         if (is_bad_inode(inode)) {
1285                                 rc = -EIO;
1286                         } else {
1287                                 osd_id_gen(id, inode->i_ino,
1288                                            inode->i_generation);
1289                                 rc = 0;
1290                         }
1291                 }
1292                 /* if dentry is accessible after osd_compat_spec_insert it
1293                  * will still contain NULL inode, so don't keep it in cache */
1294                 d_invalidate(dentry);
1295                 dput(dentry);
1296         }
1297
1298         RETURN(rc);
1299 }