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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_compat.c
  33  *
  34  * on-disk structure for managing /O
  35  *
  36  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  37  */
  38
  39 /* LUSTRE_VERSION_CODE */
  40 #include <lustre_ver.h>
  41 /* prerequisite for linux/xattr.h */
  42 #include <linux/types.h>
  43 /* prerequisite for linux/xattr.h */
  44 #include <linux/fs.h>
  45 /* XATTR_{REPLACE,CREATE} */
  46 #include <linux/xattr.h>
  47
  48 /*
  49  * struct OBD_{ALLOC,FREE}*()
  50  * OBD_FAIL_CHECK
  51  */
  52 #include <obd_support.h>
  53
  54 #include "osd_internal.h"
  55 #include "osd_oi.h"
  56
  57 static void osd_push_ctxt(const struct osd_device *dev,
  58                           struct lvfs_run_ctxt *newctxt,
  59                           struct lvfs_run_ctxt *save)
  60 {
  61         OBD_SET_CTXT_MAGIC(newctxt);
  62         newctxt->pwdmnt = dev->od_mnt;
  63         newctxt->pwd = dev->od_mnt->mnt_root;
  64         newctxt->fs = get_ds();
  65         newctxt->umask = current_umask();
  66         newctxt->dt = NULL;
  67
  68         push_ctxt(save, newctxt);
  69 }
  70
  71 /* utility to make a directory */
  72 static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
  73                                    const char *name, int mode, int fix)
  74 {
  75         struct dentry *dchild;
  76         int err = 0;
  77         ENTRY;
  78
  79         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
  80         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
  81         dchild = ll_lookup_one_len(name, dir, strlen(name));
  82         if (IS_ERR(dchild))
  83                 GOTO(out_up, dchild);
  84
  85         if (dchild->d_inode) {
  86                 int old_mode = dchild->d_inode->i_mode;
  87                 if (!S_ISDIR(old_mode)) {
  88                         CERROR("found %s (%lu/%u) is mode %o\n", name,
  89                                dchild->d_inode->i_ino,
  90                                dchild->d_inode->i_generation, old_mode);
  91                         GOTO(out_err, err = -ENOTDIR);
  92                 }
  93
  94                 /* Fixup directory permissions if necessary */
  95                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
  96                         CDEBUG(D_CONFIG,
  97                                "fixing permissions on %s from %o to %o\n",
  98                                name, old_mode, mode);
  99                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 100                                                   (old_mode & ~S_IALLUGO);
 101                         mark_inode_dirty(dchild->d_inode);
 102                 }
 103                 GOTO(out_up, dchild);
 104         }
 105
 106         err = vfs_mkdir(dir->d_inode, dchild, mode);
 107         if (err)
 108                 GOTO(out_err, err);
 109
 110         RETURN(dchild);
 111
 112 out_err:
 113         dput(dchild);
 114         dchild = ERR_PTR(err);
 115 out_up:
 116         return dchild;
 117 }
 118
 119 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
 120 {
 121         struct lr_server_data lsd;
 122         struct dentry        *dlast;
 123         loff_t                off;
 124         int                   rc = 0;
 125         int                   count = OBJ_SUBDIR_COUNT;
 126
 127         ENTRY;
 128
 129         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
 130                                   strlen(LAST_RCVD));
 131         if (IS_ERR(dlast))
 132                 return PTR_ERR(dlast);
 133         else if (dlast->d_inode == NULL)
 134                 goto out;
 135
 136         off = 0;
 137         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 138         if (rc == sizeof(lsd)) {
 139                 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
 140                        "subdir count = %d\n", lsd.lsd_uuid,
 141                        lsd.lsd_subdir_count);
 142                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 143                         count = le16_to_cpu(lsd.lsd_subdir_count);
 144         } else if (rc != 0) {
 145                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 146                 if (rc > 0)
 147                         rc = -EFAULT;
 148                 dput(dlast);
 149                 return rc;
 150         }
 151 out:
 152         dput(dlast);
 153         LASSERT(count > 0);
 154         return count;
 155 }
 156
 157 static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
 158 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
 159 {
 160         struct lvfs_run_ctxt    new;
 161         struct lvfs_run_ctxt    save;
 162         struct dentry           *parent;
 163         struct osd_mdobj_map    *omm;
 164         struct dentry           *d;
 165         struct osd_thread_info  *info = osd_oti_get(env);
 166         struct lu_fid           *fid = &info->oti_fid3;
 167         int                     rc = 0;
 168         ENTRY;
 169
 170         OBD_ALLOC_PTR(dev->od_mdt_map);
 171         if (dev->od_mdt_map == NULL)
 172                 RETURN(-ENOMEM);
 173
 174         omm = dev->od_mdt_map;
 175
 176         parent = osd_sb(dev)->s_root;
 177         osd_push_ctxt(dev, &new, &save);
 178
 179         d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
 180                          0755, 1);
 181         if (IS_ERR(d))
 182                 GOTO(cleanup, rc = PTR_ERR(d));
 183
 184         omm->omm_remote_parent = d;
 185
 186         /* Set LMA for remote parent inode */
 187         lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
 188         rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0);
 189
 190         GOTO(cleanup, rc);
 191
 192 cleanup:
 193         pop_ctxt(&save, &new);
 194         if (rc) {
 195                 if (omm->omm_remote_parent != NULL)
 196                         dput(omm->omm_remote_parent);
 197                 OBD_FREE_PTR(omm);
 198                 dev->od_mdt_map = NULL;
 199         }
 200         return rc;
 201 }
 202
 203 static void osd_mdt_fini(struct osd_device *osd)
 204 {
 205         struct osd_mdobj_map *omm = osd->od_mdt_map;
 206
 207         if (omm == NULL)
 208                 return;
 209
 210         if (omm->omm_remote_parent)
 211                 dput(omm->omm_remote_parent);
 212
 213         OBD_FREE_PTR(omm);
 214         osd->od_ost_map = NULL;
 215 }
 216
 217 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
 218                              struct osd_object *obj, struct osd_thandle *oh)
 219 {
 220         struct osd_mdobj_map    *omm = osd->od_mdt_map;
 221         struct osd_thread_info  *oti = osd_oti_get(env);
 222         struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
 223         char                    *name = oti->oti_name;
 224         struct osd_thread_info  *info = osd_oti_get(env);
 225         struct dentry           *dentry;
 226         struct dentry           *parent;
 227         int                     rc;
 228
 229         /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
 230          * can identify this object quickly */
 231         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
 232                          &oti->oti_ost_attrs);
 233         if (rc)
 234                 RETURN(rc);
 235
 236         lma->lma_incompat |= LMAI_REMOTE_PARENT;
 237         lustre_lma_swab(lma);
 238         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 239                              sizeof(*lma), XATTR_REPLACE);
 240         if (rc)
 241                 RETURN(rc);
 242
 243         parent = omm->omm_remote_parent;
 244         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 245         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 246                                            name, strlen(name));
 247         mutex_lock(&parent->d_inode->i_mutex);
 248         rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
 249                                    obj->oo_inode, NULL);
 250         CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
 251                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 252         ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
 253         mark_inode_dirty(parent->d_inode);
 254         mutex_unlock(&parent->d_inode->i_mutex);
 255         RETURN(rc);
 256 }
 257
 258 int osd_delete_from_remote_parent(const struct lu_env *env,
 259                                   struct osd_device *osd,
 260                                   struct osd_object *obj,
 261                                   struct osd_thandle *oh)
 262 {
 263         struct osd_mdobj_map       *omm = osd->od_mdt_map;
 264         struct osd_thread_info     *oti = osd_oti_get(env);
 265         struct lustre_mdt_attrs    *lma = &oti->oti_ost_attrs.loa_lma;
 266         char                       *name = oti->oti_name;
 267         struct dentry              *dentry;
 268         struct dentry              *parent;
 269         struct ldiskfs_dir_entry_2 *de;
 270         struct buffer_head         *bh;
 271         int                        rc;
 272
 273         /* Check lma to see whether it is remote object */
 274         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
 275                          &oti->oti_ost_attrs);
 276         if (rc != 0) {
 277                 /* No LMA if the directory is created before 2.0 */
 278                 if (rc == -ENODATA)
 279                         rc = 0;
 280                 RETURN(rc);
 281         }
 282
 283         if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
 284                 RETURN(0);
 285
 286         parent = omm->omm_remote_parent;
 287         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 288         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 289                                            name, strlen(name));
 290         mutex_lock(&parent->d_inode->i_mutex);
 291         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 292                                     NULL, NULL);
 293         if (IS_ERR(bh)) {
 294                 mutex_unlock(&parent->d_inode->i_mutex);
 295                 RETURN(PTR_ERR(bh));
 296         }
 297         CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
 298                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 299         rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
 300         ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
 301         mark_inode_dirty(parent->d_inode);
 302         mutex_unlock(&parent->d_inode->i_mutex);
 303         brelse(bh);
 304
 305         /* Get rid of REMOTE_PARENT flag from incompat */
 306         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
 307         lustre_lma_swab(lma);
 308         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 309                              sizeof(*lma), XATTR_REPLACE);
 310         RETURN(rc);
 311 }
 312
 313 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 314                                 struct osd_device *osd,
 315                                 const struct lu_fid *fid,
 316                                 struct osd_inode_id *id)
 317 {
 318         struct osd_mdobj_map        *omm = osd->od_mdt_map;
 319         char                        *name = oti->oti_name;
 320         struct dentry               *parent;
 321         struct dentry               *dentry;
 322         struct ldiskfs_dir_entry_2 *de;
 323         struct buffer_head         *bh;
 324         int                         rc;
 325         ENTRY;
 326
 327         parent = omm->omm_remote_parent;
 328         sprintf(name, DFID_NOBRACE, PFID(fid));
 329         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
 330                                            name, strlen(name));
 331         mutex_lock(&parent->d_inode->i_mutex);
 332         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 333                                     NULL, NULL);
 334         if (IS_ERR(bh)) {
 335                 rc = PTR_ERR(bh);
 336         } else {
 337                 struct inode *inode;
 338
 339                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 340                 brelse(bh);
 341                 inode = osd_iget(oti, osd, id);
 342                 if (IS_ERR(inode)) {
 343                         rc = PTR_ERR(inode);
 344                         if (rc == -ESTALE)
 345                                 rc = -ENOENT;
 346                 } else {
 347                         iput(inode);
 348                         rc = 0;
 349                 }
 350         }
 351         mutex_unlock(&parent->d_inode->i_mutex);
 352         if (rc == 0)
 353                 osd_add_oi_cache(oti, osd, id, fid);
 354         RETURN(rc);
 355 }
 356
 357 /*
 358  * directory structure on legacy OST:
 359  *
 360  * O/<seq>/d0-31/<objid>
 361  * O/<seq>/LAST_ID
 362  * last_rcvd
 363  * LAST_GROUP
 364  * CONFIGS
 365  *
 366  */
 367 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
 368 {
 369         struct lvfs_run_ctxt     new;
 370         struct lvfs_run_ctxt     save;
 371         struct dentry           *rootd = osd_sb(dev)->s_root;
 372         struct dentry           *d;
 373         struct osd_thread_info  *info = osd_oti_get(env);
 374         struct inode            *inode;
 375         struct lu_fid           *fid = &info->oti_fid3;
 376         int                      rc;
 377         ENTRY;
 378
 379         OBD_ALLOC_PTR(dev->od_ost_map);
 380         if (dev->od_ost_map == NULL)
 381                 RETURN(-ENOMEM);
 382
 383         /* to get subdir count from last_rcvd */
 384         rc = osd_last_rcvd_subdir_count(dev);
 385         if (rc < 0)
 386                 GOTO(cleanup_alloc, rc);
 387
 388         dev->od_ost_map->om_subdir_count = rc;
 389         rc = 0;
 390
 391         INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 392         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 393         mutex_init(&dev->od_ost_map->om_dir_init_mutex);
 394
 395         osd_push_ctxt(dev, &new, &save);
 396
 397         d = ll_lookup_one_len("O", rootd, strlen("O"));
 398         if (IS_ERR(d))
 399                 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 400         if (d->d_inode == NULL) {
 401                 dput(d);
 402                 /* The lookup() may be called again inside simple_mkdir().
 403                  * Since the repeated lookup() only be called for "/O" at
 404                  * mount time, it will not affect the whole performance. */
 405                 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 406                 if (IS_ERR(d))
 407                         GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 408
 409                 /* It is quite probably that the device is new formatted. */
 410                 dev->od_maybe_new = 1;
 411         }
 412
 413         inode = d->d_inode;
 414         dev->od_ost_map->om_root = d;
 415
 416         /* 'What the @fid is' is not imporatant, because the object
 417          * has no OI mapping, and only is visible inside the OSD.*/
 418         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 419         rc = osd_ea_fid_set(info, inode, fid,
 420                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 421         if (rc)
 422                 GOTO(cleanup_dentry, rc);
 423
 424         pop_ctxt(&save, &new);
 425         RETURN(0);
 426
 427 cleanup_dentry:
 428         dput(d);
 429 cleanup_ctxt:
 430         pop_ctxt(&save, &new);
 431 cleanup_alloc:
 432         OBD_FREE_PTR(dev->od_ost_map);
 433         return rc;
 434 }
 435
 436 static void osd_seq_free(struct osd_obj_seq *osd_seq)
 437 {
 438         int j;
 439
 440         if (osd_seq->oos_dirs) {
 441                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 442                         if (osd_seq->oos_dirs[j])
 443                                 dput(osd_seq->oos_dirs[j]);
 444                 }
 445                 OBD_FREE(osd_seq->oos_dirs,
 446                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 447         }
 448
 449         if (osd_seq->oos_root)
 450                 dput(osd_seq->oos_root);
 451
 452         OBD_FREE_PTR(osd_seq);
 453 }
 454
 455 static void osd_ost_fini(struct osd_device *osd)
 456 {
 457         struct osd_obj_seq    *osd_seq;
 458         struct osd_obj_seq    *tmp;
 459         struct osd_obj_map    *map = osd->od_ost_map;
 460         ENTRY;
 461
 462         if (map == NULL)
 463                 return;
 464
 465         write_lock(&map->om_seq_list_lock);
 466         list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
 467                                  oos_seq_list) {
 468                 list_del_init(&osd_seq->oos_seq_list);
 469                 write_unlock(&map->om_seq_list_lock);
 470                 osd_seq_free(osd_seq);
 471                 write_lock(&map->om_seq_list_lock);
 472         }
 473         write_unlock(&map->om_seq_list_lock);
 474         if (map->om_root)
 475                 dput(map->om_root);
 476         OBD_FREE_PTR(map);
 477         osd->od_ost_map = NULL;
 478         EXIT;
 479 }
 480
 481 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 482 {
 483         int rc;
 484         ENTRY;
 485
 486         /* prepare structures for OST */
 487         rc = osd_ost_init(env, dev);
 488         if (rc)
 489                 RETURN(rc);
 490
 491         /* prepare structures for MDS */
 492         rc = osd_mdt_init(env, dev);
 493         if (rc)
 494                 osd_ost_fini(dev);
 495
 496         RETURN(rc);
 497 }
 498
 499 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
 500 {
 501         struct osd_obj_seq *osd_seq;
 502
 503         list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 504                 if (osd_seq->oos_seq == seq)
 505                         return osd_seq;
 506         }
 507         return NULL;
 508 }
 509
 510 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
 511 {
 512         struct osd_obj_seq *osd_seq;
 513
 514         read_lock(&map->om_seq_list_lock);
 515         osd_seq = osd_seq_find_locked(map, seq);
 516         read_unlock(&map->om_seq_list_lock);
 517         return osd_seq;
 518 }
 519
 520 void osd_obj_map_fini(struct osd_device *dev)
 521 {
 522         osd_ost_fini(dev);
 523         osd_mdt_fini(dev);
 524 }
 525
 526 /**
 527  * Update the specified OI mapping.
 528  *
 529  * \retval   1, changed nothing
 530  * \retval   0, changed successfully
 531  * \retval -ve, on error
 532  */
 533 static int osd_obj_update_entry(struct osd_thread_info *info,
 534                                 struct osd_device *osd,
 535                                 struct dentry *dir, const char *name,
 536                                 const struct lu_fid *fid,
 537                                 const struct osd_inode_id *id,
 538                                 handle_t *th)
 539 {
 540         struct inode               *parent = dir->d_inode;
 541         struct dentry              *child;
 542         struct ldiskfs_dir_entry_2 *de;
 543         struct buffer_head         *bh;
 544         struct inode               *inode;
 545         struct dentry              *dentry = &info->oti_obj_dentry;
 546         struct osd_inode_id        *oi_id  = &info->oti_id3;
 547         struct lustre_mdt_attrs    *lma    = &info->oti_ost_attrs.loa_lma;
 548         struct lu_fid              *oi_fid = &lma->lma_self_fid;
 549         int                         rc;
 550         ENTRY;
 551
 552         LASSERT(th != NULL);
 553         LASSERT(th->h_transaction != NULL);
 554
 555         child = &info->oti_child_dentry;
 556         child->d_parent = dir;
 557         child->d_name.hash = 0;
 558         child->d_name.name = name;
 559         child->d_name.len = strlen(name);
 560
 561         ll_vfs_dq_init(parent);
 562         mutex_lock(&parent->i_mutex);
 563         bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
 564         if (IS_ERR(bh))
 565                 GOTO(out, rc = PTR_ERR(bh));
 566
 567         if (le32_to_cpu(de->inode) == id->oii_ino)
 568                 GOTO(out, rc = 1);
 569
 570         osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 571         inode = osd_iget(info, osd, oi_id);
 572         if (IS_ERR(inode)) {
 573                 rc = PTR_ERR(inode);
 574                 if (rc == -ENOENT || rc == -ESTALE)
 575                         goto update;
 576                 GOTO(out, rc);
 577         }
 578
 579         /* The EA inode should NOT be in OI, old OI scrub may added
 580          * such OI mapping by wrong, replace it. */
 581         if (unlikely(osd_is_ea_inode(inode))) {
 582                 iput(inode);
 583                 goto update;
 584         }
 585
 586         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
 587         if (rc == -ENODATA) {
 588                 rc = osd_get_idif(info, inode, dentry, oi_fid);
 589                 if (rc > 0 || rc == -ENODATA) {
 590                         oi_fid = NULL;
 591                         rc = 0;
 592                 }
 593         }
 594         iput(inode);
 595
 596         if (rc != 0)
 597                 GOTO(out, rc);
 598
 599         /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
 600          * either a crashed object or a uninitialized one. Replace it. */
 601         if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
 602                 CERROR("%s: the FID "DFID" is used by two objects: "
 603                        "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
 604                        oi_id->oii_ino, oi_id->oii_gen,
 605                        id->oii_ino, id->oii_gen);
 606                 GOTO(out, rc = -EEXIST);
 607         }
 608
 609         if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
 610                 __u32 idx1 = fid_idif_ost_idx(fid);
 611                 __u32 idx2 = fid_idif_ost_idx(oi_fid);
 612                 struct ost_id *ostid = &info->oti_ostid;
 613                 struct lu_fid *tfid = &info->oti_fid3;
 614
 615                 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
 616                          "invalid given FID "DFID", not match the "
 617                          "device index %u\n", PFID(fid), osd->od_index);
 618
 619                 if (idx1 != idx2) {
 620                         if (idx1 == 0 && idx2 == osd->od_index) {
 621                                 fid_to_ostid(fid, ostid);
 622                                 ostid_to_fid(tfid, ostid, idx2);
 623                                 if (lu_fid_eq(tfid, oi_fid)) {
 624                                         CERROR("%s: the FID "DFID" is used by "
 625                                                "two objects(2): %u/%u %u/%u\n",
 626                                                osd_name(osd), PFID(fid),
 627                                                oi_id->oii_ino, oi_id->oii_gen,
 628                                                id->oii_ino, id->oii_gen);
 629
 630                                         GOTO(out, rc = -EEXIST);
 631                                 }
 632                         } else if (idx2 == 0 && idx1 == osd->od_index) {
 633                                 fid_to_ostid(oi_fid, ostid);
 634                                 ostid_to_fid(tfid, ostid, idx1);
 635                                 if (lu_fid_eq(tfid, fid)) {
 636                                         CERROR("%s: the FID "DFID" is used by "
 637                                                "two objects(2): %u/%u %u/%u\n",
 638                                                osd_name(osd), PFID(fid),
 639                                                oi_id->oii_ino, oi_id->oii_gen,
 640                                                id->oii_ino, id->oii_gen);
 641
 642                                         GOTO(out, rc = -EEXIST);
 643                                 }
 644                         }
 645                 }
 646         }
 647
 648 update:
 649         /* There may be temporary inconsistency: On one hand, the new
 650          * object may be referenced by multiple entries, which is out
 651          * of our control unless we traverse the whole /O completely,
 652          * which is non-flat order and inefficient, should be avoided;
 653          * On the other hand, the old object may become orphan if it
 654          * is still valid. Since it was referenced by an invalid entry,
 655          * making it as invisible temporary may be not worse. OI scrub
 656          * will process it later. */
 657         rc = ldiskfs_journal_get_write_access(th, bh);
 658         if (rc != 0)
 659                 GOTO(out, rc);
 660
 661         de->inode = cpu_to_le32(id->oii_ino);
 662         rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
 663
 664         GOTO(out, rc);
 665
 666 out:
 667         if (!IS_ERR(bh))
 668                 brelse(bh);
 669         mutex_unlock(&parent->i_mutex);
 670         return rc;
 671 }
 672
 673 static int osd_obj_del_entry(struct osd_thread_info *info,
 674                              struct osd_device *osd,
 675                              struct dentry *dird, char *name,
 676                              handle_t *th)
 677 {
 678         struct ldiskfs_dir_entry_2 *de;
 679         struct buffer_head         *bh;
 680         struct dentry              *child;
 681         struct inode               *dir = dird->d_inode;
 682         int                         rc;
 683         ENTRY;
 684
 685         LASSERT(th != NULL);
 686         LASSERT(th->h_transaction != NULL);
 687
 688
 689         child = &info->oti_child_dentry;
 690         child->d_name.hash = 0;
 691         child->d_name.name = name;
 692         child->d_name.len = strlen(name);
 693         child->d_parent = dird;
 694         child->d_inode = NULL;
 695
 696         ll_vfs_dq_init(dir);
 697         mutex_lock(&dir->i_mutex);
 698         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 699         if (IS_ERR(bh)) {
 700                 rc = PTR_ERR(bh);
 701         } else {
 702                 rc = ldiskfs_delete_entry(th, dir, de, bh);
 703                 brelse(bh);
 704         }
 705         mutex_unlock(&dir->i_mutex);
 706
 707         RETURN(rc);
 708 }
 709
 710 static int osd_obj_add_entry(struct osd_thread_info *info,
 711                              struct osd_device *osd,
 712                              struct dentry *dir, char *name,
 713                              const struct osd_inode_id *id,
 714                              handle_t *th)
 715 {
 716         struct dentry *child;
 717         struct inode *inode;
 718         int rc;
 719
 720         ENTRY;
 721
 722         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
 723                 RETURN(0);
 724
 725         LASSERT(th != NULL);
 726         LASSERT(th->h_transaction != NULL);
 727
 728         inode = info->oti_inode;
 729         if (unlikely(inode == NULL)) {
 730                 struct ldiskfs_inode_info *lii;
 731                 OBD_ALLOC_PTR(lii);
 732                 if (lii == NULL)
 733                         RETURN(-ENOMEM);
 734                 inode = info->oti_inode = &lii->vfs_inode;
 735         }
 736
 737         inode->i_sb = osd_sb(osd);
 738         osd_id_to_inode(inode, id);
 739         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 740
 741         child = &info->oti_child_dentry;
 742         child->d_name.hash = 0;
 743         child->d_name.name = name;
 744         child->d_name.len = strlen(name);
 745         child->d_parent = dir;
 746         child->d_inode = inode;
 747
 748         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
 749                 inode->i_ino++;
 750
 751         ll_vfs_dq_init(dir->d_inode);
 752         mutex_lock(&dir->d_inode->i_mutex);
 753         rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
 754         mutex_unlock(&dir->d_inode->i_mutex);
 755
 756         RETURN(rc);
 757 }
 758
 759 /**
 760  * Use %llu for legacy OST sequences, but use %llx for new
 761  * sequences names, so that the O/{seq}/dN/{oid} more closely
 762  * follows the DFID/PFID format. This makes it easier to map from
 763  * debug messages to objects in the future, and the legacy space
 764  * of FID_SEQ_OST_MDT0 will be unused in the future.
 765  **/
 766 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
 767 {
 768         snprintf(seq_name, name_size,
 769                  (fid_seq_is_rsvd(seq) ||
 770                   fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
 771                  fid_seq_is_idif(seq) ? 0 : seq);
 772 }
 773
 774 static inline void osd_oid_name(char *name, size_t name_size,
 775                                 const struct lu_fid *fid, u64 id)
 776 {
 777         snprintf(name, name_size,
 778                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 779                   fid_seq_is_mdt0(fid_seq(fid)) ||
 780                   fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
 781 }
 782
 783 /* external locking is required */
 784 static int osd_seq_load_locked(struct osd_thread_info *info,
 785                                struct osd_device *osd,
 786                                struct osd_obj_seq *osd_seq)
 787 {
 788         struct osd_obj_map  *map = osd->od_ost_map;
 789         struct dentry       *seq_dir;
 790         struct inode        *inode;
 791         struct lu_fid       *fid = &info->oti_fid3;
 792         int                 rc = 0;
 793         int                 i;
 794         char                dir_name[32];
 795         ENTRY;
 796
 797         if (osd_seq->oos_root != NULL)
 798                 RETURN(0);
 799
 800         LASSERT(map);
 801         LASSERT(map->om_root);
 802
 803         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 804
 805         seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
 806         if (IS_ERR(seq_dir))
 807                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 808         else if (seq_dir->d_inode == NULL)
 809                 GOTO(out_put, rc = -EFAULT);
 810
 811         inode = seq_dir->d_inode;
 812         osd_seq->oos_root = seq_dir;
 813
 814         /* 'What the @fid is' is not imporatant, because the object
 815          * has no OI mapping, and only is visible inside the OSD.*/
 816         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 817         rc = osd_ea_fid_set(info, inode, fid,
 818                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 819         if (rc != 0)
 820                 GOTO(out_put, rc);
 821
 822         LASSERT(osd_seq->oos_dirs == NULL);
 823         OBD_ALLOC(osd_seq->oos_dirs,
 824                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 825         if (osd_seq->oos_dirs == NULL)
 826                 GOTO(out_put, rc = -ENOMEM);
 827
 828         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 829                 struct dentry   *dir;
 830
 831                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 832                 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
 833                                    0700, 1);
 834                 if (IS_ERR(dir)) {
 835                         GOTO(out_free, rc = PTR_ERR(dir));
 836                 } else if (dir->d_inode == NULL) {
 837                         dput(dir);
 838                         GOTO(out_free, rc = -EFAULT);
 839                 }
 840
 841                 inode = dir->d_inode;
 842                 osd_seq->oos_dirs[i] = dir;
 843
 844                 /* 'What the @fid is' is not imporatant, because the object
 845                  * has no OI mapping, and only is visible inside the OSD.*/
 846                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
 847                 rc = osd_ea_fid_set(info, inode, fid,
 848                                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 849                 if (rc != 0)
 850                         GOTO(out_free, rc);
 851         }
 852
 853         if (rc != 0) {
 854 out_free:
 855                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 856                         if (osd_seq->oos_dirs[i] != NULL)
 857                                 dput(osd_seq->oos_dirs[i]);
 858                 }
 859                 OBD_FREE(osd_seq->oos_dirs,
 860                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
 861 out_put:
 862                 dput(seq_dir);
 863                 osd_seq->oos_root = NULL;
 864         }
 865 out_err:
 866         RETURN(rc);
 867 }
 868
 869 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
 870                                         struct osd_device *osd, u64 seq)
 871 {
 872         struct osd_obj_map      *map;
 873         struct osd_obj_seq      *osd_seq;
 874         int                     rc = 0;
 875         ENTRY;
 876
 877         map = osd->od_ost_map;
 878         LASSERT(map);
 879         LASSERT(map->om_root);
 880
 881         osd_seq = osd_seq_find(map, seq);
 882         if (likely(osd_seq != NULL))
 883                 RETURN(osd_seq);
 884
 885         /* Serializing init process */
 886         mutex_lock(&map->om_dir_init_mutex);
 887
 888         /* Check whether the seq has been added */
 889         read_lock(&map->om_seq_list_lock);
 890         osd_seq = osd_seq_find_locked(map, seq);
 891         if (osd_seq != NULL) {
 892                 read_unlock(&map->om_seq_list_lock);
 893                 GOTO(cleanup, rc = 0);
 894         }
 895         read_unlock(&map->om_seq_list_lock);
 896
 897         OBD_ALLOC_PTR(osd_seq);
 898         if (osd_seq == NULL)
 899                 GOTO(cleanup, rc = -ENOMEM);
 900
 901         INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 902         osd_seq->oos_seq = seq;
 903         /* Init subdir count to be 32, but each seq can have
 904          * different subdir count */
 905         osd_seq->oos_subdir_count = map->om_subdir_count;
 906         rc = osd_seq_load_locked(info, osd, osd_seq);
 907         if (rc != 0)
 908                 GOTO(cleanup, rc);
 909
 910         write_lock(&map->om_seq_list_lock);
 911         list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 912         write_unlock(&map->om_seq_list_lock);
 913
 914 cleanup:
 915         mutex_unlock(&map->om_dir_init_mutex);
 916         if (rc != 0) {
 917                 if (osd_seq != NULL)
 918                         OBD_FREE_PTR(osd_seq);
 919                 RETURN(ERR_PTR(rc));
 920         }
 921
 922         RETURN(osd_seq);
 923 }
 924
 925 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 926                        const struct lu_fid *fid, struct osd_inode_id *id)
 927 {
 928         struct osd_obj_map              *map;
 929         struct osd_obj_seq              *osd_seq;
 930         struct dentry                   *d_seq;
 931         struct dentry                   *child;
 932         struct ost_id                   *ostid = &info->oti_ostid;
 933         int                             dirn;
 934         char                            name[32];
 935         struct ldiskfs_dir_entry_2      *de;
 936         struct buffer_head              *bh;
 937         struct inode                    *dir;
 938         struct inode                    *inode;
 939         ENTRY;
 940
 941         /* on the very first lookup we find and open directories */
 942
 943         map = dev->od_ost_map;
 944         LASSERT(map);
 945         LASSERT(map->om_root);
 946
 947         fid_to_ostid(fid, ostid);
 948         osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
 949         if (IS_ERR(osd_seq))
 950                 RETURN(PTR_ERR(osd_seq));
 951
 952         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 953         d_seq = osd_seq->oos_dirs[dirn];
 954         LASSERT(d_seq);
 955
 956         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 957
 958         child = &info->oti_child_dentry;
 959         child->d_parent = d_seq;
 960         child->d_name.hash = 0;
 961         child->d_name.name = name;
 962         /* XXX: we can use rc from sprintf() instead of strlen() */
 963         child->d_name.len = strlen(name);
 964
 965         dir = d_seq->d_inode;
 966         mutex_lock(&dir->i_mutex);
 967         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 968         mutex_unlock(&dir->i_mutex);
 969
 970         if (IS_ERR(bh))
 971                 RETURN(PTR_ERR(bh));
 972
 973         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 974         brelse(bh);
 975
 976         inode = osd_iget(info, dev, id);
 977         if (IS_ERR(inode)) {
 978                 int rc = PTR_ERR(inode);
 979
 980                 RETURN(rc == -ENOENT ? -ESTALE : rc);
 981         }
 982
 983         iput(inode);
 984         RETURN(0);
 985 }
 986
 987 int osd_obj_map_insert(struct osd_thread_info *info,
 988                        struct osd_device *osd,
 989                        const struct lu_fid *fid,
 990                        const struct osd_inode_id *id,
 991                        handle_t *th)
 992 {
 993         struct osd_obj_map      *map;
 994         struct osd_obj_seq      *osd_seq;
 995         struct dentry           *d;
 996         struct ost_id           *ostid = &info->oti_ostid;
 997         u64                      oid;
 998         int                     dirn, rc = 0;
 999         char                    name[32];
1000         ENTRY;
1001
1002         map = osd->od_ost_map;
1003         LASSERT(map);
1004
1005         /* map fid to seq:objid */
1006         fid_to_ostid(fid, ostid);
1007
1008         oid = ostid_id(ostid);
1009         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1010         if (IS_ERR(osd_seq))
1011                 RETURN(PTR_ERR(osd_seq));
1012
1013         dirn = oid & (osd_seq->oos_subdir_count - 1);
1014         d = osd_seq->oos_dirs[dirn];
1015         LASSERT(d);
1016
1017         osd_oid_name(name, sizeof(name), fid, oid);
1018
1019 again:
1020         rc = osd_obj_add_entry(info, osd, d, name, id, th);
1021         if (rc == -EEXIST) {
1022                 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1023                 if (unlikely(rc == -ENOENT))
1024                         goto again;
1025
1026                 if (unlikely(rc == 1))
1027                         RETURN(0);
1028         }
1029
1030         RETURN(rc);
1031 }
1032
1033 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1034                        const struct lu_fid *fid, handle_t *th)
1035 {
1036         struct osd_obj_map      *map;
1037         struct osd_obj_seq      *osd_seq;
1038         struct dentry           *d;
1039         struct ost_id           *ostid = &info->oti_ostid;
1040         int                     dirn, rc = 0;
1041         char                    name[32];
1042         ENTRY;
1043
1044         map = osd->od_ost_map;
1045         LASSERT(map);
1046
1047         /* map fid to seq:objid */
1048         fid_to_ostid(fid, ostid);
1049
1050         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1051         if (IS_ERR(osd_seq))
1052                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1053
1054         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1055         d = osd_seq->oos_dirs[dirn];
1056         LASSERT(d);
1057
1058         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1059         rc = osd_obj_del_entry(info, osd, d, name, th);
1060 cleanup:
1061         RETURN(rc);
1062 }
1063
1064 int osd_obj_map_update(struct osd_thread_info *info,
1065                        struct osd_device *osd,
1066                        const struct lu_fid *fid,
1067                        const struct osd_inode_id *id,
1068                        handle_t *th)
1069 {
1070         struct osd_obj_seq      *osd_seq;
1071         struct dentry           *d;
1072         struct ost_id           *ostid = &info->oti_ostid;
1073         int                     dirn, rc = 0;
1074         char                    name[32];
1075         ENTRY;
1076
1077         fid_to_ostid(fid, ostid);
1078         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1079         if (IS_ERR(osd_seq))
1080                 RETURN(PTR_ERR(osd_seq));
1081
1082         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1083         d = osd_seq->oos_dirs[dirn];
1084         LASSERT(d);
1085
1086         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1087         rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1088
1089         RETURN(rc);
1090 }
1091
1092 int osd_obj_map_recover(struct osd_thread_info *info,
1093                         struct osd_device *osd,
1094                         struct inode *src_parent,
1095                         struct dentry *src_child,
1096                         const struct lu_fid *fid)
1097 {
1098         struct osd_obj_seq         *osd_seq;
1099         struct dentry              *tgt_parent;
1100         struct dentry              *tgt_child = &info->oti_child_dentry;
1101         struct inode               *dir;
1102         struct inode               *inode     = src_child->d_inode;
1103         struct ost_id              *ostid     = &info->oti_ostid;
1104         handle_t                   *jh;
1105         struct ldiskfs_dir_entry_2 *de;
1106         struct buffer_head         *bh;
1107         char                        name[32];
1108         int                         dirn;
1109         int                         rc        = 0;
1110         ENTRY;
1111
1112         if (fid_is_last_id(fid)) {
1113                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1114                 if (IS_ERR(osd_seq))
1115                         RETURN(PTR_ERR(osd_seq));
1116
1117                 tgt_parent = osd_seq->oos_root;
1118                 tgt_child->d_name.name = "LAST_ID";
1119                 tgt_child->d_name.len = strlen("LAST_ID");
1120         } else {
1121                 fid_to_ostid(fid, ostid);
1122                 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1123                 if (IS_ERR(osd_seq))
1124                         RETURN(PTR_ERR(osd_seq));
1125
1126                 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1127                 tgt_parent = osd_seq->oos_dirs[dirn];
1128                 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1129                 tgt_child->d_name.name = name;
1130                 tgt_child->d_name.len = strlen(name);
1131         }
1132         LASSERT(tgt_parent != NULL);
1133
1134         dir = tgt_parent->d_inode;
1135         tgt_child->d_name.hash = 0;
1136         tgt_child->d_parent = tgt_parent;
1137         tgt_child->d_inode = inode;
1138
1139         /* The non-initialized src_child may be destroyed. */
1140         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1141                                 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1142                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1143                                 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1144         if (IS_ERR(jh))
1145                 RETURN(PTR_ERR(jh));
1146
1147         ll_vfs_dq_init(src_parent);
1148         ll_vfs_dq_init(dir);
1149
1150         mutex_lock(&src_parent->i_mutex);
1151         mutex_lock(&dir->i_mutex);
1152         bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1153         if (!IS_ERR(bh)) {
1154                 /* XXX: If some other object occupied the same slot. And If such
1155                  *      inode is zero-sized and with SUID+SGID, then means it is
1156                  *      a new created one. Maybe we can remove it and insert the
1157                  *      original one back to the /O/<seq>/d<x>. But there are
1158                  *      something to be considered:
1159                  *
1160                  *      1) The OST-object under /lost+found has crashed LMA.
1161                  *         So it should not conflict with the current one.
1162                  *
1163                  *      2) There are race conditions that: someone may just want
1164                  *         to modify the current one. Even if the OI scrub takes
1165                  *         the object lock when remove the current one, it still
1166                  *         cause the modification to be lost becasue the target
1167                  *         has been removed when the RPC service thread waiting
1168                  *         for the lock.
1169                  *
1170                  *      So keep it there before we have suitable solution. */
1171                 brelse(bh);
1172                 mutex_unlock(&dir->i_mutex);
1173                 mutex_unlock(&src_parent->i_mutex);
1174                 ldiskfs_journal_stop(jh);
1175
1176                 rc = -EEXIST;
1177                 /* If the src object has never been modified, then remove it. */
1178                 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1179                     inode->i_mode & S_ISGID) {
1180                         rc = ll_vfs_unlink(src_parent, src_child);
1181                         if (unlikely(rc == -ENOENT))
1182                                 rc = 0;
1183                 }
1184                 RETURN(rc);
1185         }
1186
1187         bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1188                                     NULL, NULL);
1189         if (unlikely(IS_ERR(bh)))
1190                 GOTO(unlock, rc = PTR_ERR(bh));
1191
1192         rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1193         brelse(bh);
1194         if (rc != 0)
1195                 GOTO(unlock, rc);
1196
1197         rc = osd_ldiskfs_add_entry(info, osd, jh, tgt_child, inode, NULL);
1198
1199         GOTO(unlock, rc);
1200
1201 unlock:
1202         mutex_unlock(&dir->i_mutex);
1203         mutex_unlock(&src_parent->i_mutex);
1204         ldiskfs_journal_stop(jh);
1205         return rc;
1206 }
1207
1208 static struct dentry *
1209 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1210                      const struct lu_fid *fid, char **name)
1211 {
1212         struct dentry *root = ERR_PTR(-ENOENT);
1213
1214         if (fid_is_last_id(fid)) {
1215                 struct osd_obj_seq *osd_seq;
1216
1217                 /* on creation of LAST_ID we create O/<seq> hierarchy */
1218                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1219                 if (IS_ERR(osd_seq))
1220                         RETURN((struct dentry *)osd_seq);
1221
1222                 *name = "LAST_ID";
1223                 root = osd_seq->oos_root;
1224         } else {
1225                 *name = osd_lf_fid2name(fid);
1226                 if (*name == NULL)
1227                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1228                 else if ((*name)[0])
1229                         root = osd_sb(osd)->s_root;
1230         }
1231
1232         return root;
1233 }
1234
1235 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1236                         const struct lu_fid *fid, const struct osd_inode_id *id,
1237                         handle_t *th)
1238 {
1239         struct dentry   *root;
1240         char            *name = NULL;
1241         int              rc;
1242         ENTRY;
1243
1244         root = osd_object_spec_find(info, osd, fid, &name);
1245         if (!IS_ERR(root)) {
1246                 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1247         } else {
1248                 rc = PTR_ERR(root);
1249                 if (rc == -ENOENT)
1250                         rc = 1;
1251         }
1252
1253         RETURN(rc);
1254 }
1255
1256 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1257                         const struct lu_fid *fid, const struct osd_inode_id *id,
1258                         handle_t *th)
1259 {
1260         struct dentry   *root;
1261         char            *name = NULL;
1262         int              rc;
1263         ENTRY;
1264
1265         root = osd_object_spec_find(info, osd, fid, &name);
1266         if (!IS_ERR(root)) {
1267                 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1268         } else {
1269                 rc = PTR_ERR(root);
1270                 if (rc == -ENOENT)
1271                         rc = 0;
1272         }
1273
1274         RETURN(rc);
1275 }
1276
1277 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1278                         const struct lu_fid *fid, struct osd_inode_id *id)
1279 {
1280         struct dentry   *root;
1281         struct dentry   *dentry;
1282         struct inode    *inode;
1283         char            *name = NULL;
1284         int             rc = -ENOENT;
1285         ENTRY;
1286
1287         if (fid_is_last_id(fid)) {
1288                 struct osd_obj_seq *osd_seq;
1289
1290                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1291                 if (IS_ERR(osd_seq))
1292                         RETURN(PTR_ERR(osd_seq));
1293                 root = osd_seq->oos_root;
1294                 name = "LAST_ID";
1295         } else {
1296                 root = osd_sb(osd)->s_root;
1297                 name = osd_lf_fid2name(fid);
1298                 if (name == NULL || strlen(name) == 0)
1299                         RETURN(-ENOENT);
1300         }
1301
1302         dentry = ll_lookup_one_len(name, root, strlen(name));
1303         if (!IS_ERR(dentry)) {
1304                 inode = dentry->d_inode;
1305                 if (inode) {
1306                         if (is_bad_inode(inode)) {
1307                                 rc = -EIO;
1308                         } else {
1309                                 osd_id_gen(id, inode->i_ino,
1310                                            inode->i_generation);
1311                                 rc = 0;
1312                         }
1313                 }
1314                 /* if dentry is accessible after osd_compat_spec_insert it
1315                  * will still contain NULL inode, so don't keep it in cache */
1316                 d_invalidate(dentry);
1317                 dput(dentry);
1318         }
1319
1320         RETURN(rc);
1321 }