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, 2015, 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_mdt_attrs;
 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, lma);
 232         if (rc != 0)
 233                 RETURN(rc);
 234
 235         lma->lma_incompat |= LMAI_REMOTE_PARENT;
 236         lustre_lma_swab(lma);
 237         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 238                              sizeof(*lma), XATTR_REPLACE);
 239         if (rc != 0)
 240                 RETURN(rc);
 241
 242         parent = omm->omm_remote_parent;
 243         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 244         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 245                                            name, strlen(name));
 246         mutex_lock(&parent->d_inode->i_mutex);
 247         rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
 248                                    obj->oo_inode, NULL);
 249         CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
 250                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 251         ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
 252         mark_inode_dirty(parent->d_inode);
 253         mutex_unlock(&parent->d_inode->i_mutex);
 254         RETURN(rc);
 255 }
 256
 257 int osd_delete_from_remote_parent(const struct lu_env *env,
 258                                   struct osd_device *osd,
 259                                   struct osd_object *obj,
 260                                   struct osd_thandle *oh)
 261 {
 262         struct osd_mdobj_map       *omm = osd->od_mdt_map;
 263         struct osd_thread_info     *oti = osd_oti_get(env);
 264         struct lustre_mdt_attrs    *lma = &oti->oti_mdt_attrs;
 265         char                       *name = oti->oti_name;
 266         struct dentry              *dentry;
 267         struct dentry              *parent;
 268         struct ldiskfs_dir_entry_2 *de;
 269         struct buffer_head         *bh;
 270         int                        rc;
 271
 272         /* Check lma to see whether it is remote object */
 273         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 274         if (rc != 0) {
 275                 /* No LMA if the directory is created before 2.0 */
 276                 if (rc == -ENODATA)
 277                         rc = 0;
 278                 RETURN(rc);
 279         }
 280
 281         if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
 282                 RETURN(0);
 283
 284         parent = omm->omm_remote_parent;
 285         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 286         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 287                                            name, strlen(name));
 288         mutex_lock(&parent->d_inode->i_mutex);
 289         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 290                                     NULL, NULL);
 291         if (IS_ERR(bh)) {
 292                 mutex_unlock(&parent->d_inode->i_mutex);
 293                 RETURN(PTR_ERR(bh));
 294         }
 295         CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
 296                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 297         rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
 298         ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
 299         mark_inode_dirty(parent->d_inode);
 300         mutex_unlock(&parent->d_inode->i_mutex);
 301         brelse(bh);
 302
 303         /* Get rid of REMOTE_PARENT flag from incompat */
 304         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
 305         lustre_lma_swab(lma);
 306         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 307                              sizeof(*lma), XATTR_REPLACE);
 308         RETURN(rc);
 309 }
 310
 311 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 312                                 struct osd_device *osd,
 313                                 const struct lu_fid *fid,
 314                                 struct osd_inode_id *id)
 315 {
 316         struct osd_mdobj_map        *omm = osd->od_mdt_map;
 317         char                        *name = oti->oti_name;
 318         struct dentry               *parent;
 319         struct dentry               *dentry;
 320         struct ldiskfs_dir_entry_2 *de;
 321         struct buffer_head         *bh;
 322         int                         rc;
 323         ENTRY;
 324
 325         parent = omm->omm_remote_parent;
 326         sprintf(name, DFID_NOBRACE, PFID(fid));
 327         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
 328                                            name, strlen(name));
 329         mutex_lock(&parent->d_inode->i_mutex);
 330         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 331                                     NULL, NULL);
 332         if (IS_ERR(bh)) {
 333                 rc = PTR_ERR(bh);
 334         } else {
 335                 struct inode *inode;
 336
 337                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 338                 brelse(bh);
 339                 inode = osd_iget(oti, osd, id);
 340                 if (IS_ERR(inode)) {
 341                         rc = PTR_ERR(inode);
 342                         if (rc == -ESTALE)
 343                                 rc = -ENOENT;
 344                 } else {
 345                         iput(inode);
 346                         rc = 0;
 347                 }
 348         }
 349         mutex_unlock(&parent->d_inode->i_mutex);
 350         if (rc == 0)
 351                 osd_add_oi_cache(oti, osd, id, fid);
 352         RETURN(rc);
 353 }
 354
 355 /*
 356  * directory structure on legacy OST:
 357  *
 358  * O/<seq>/d0-31/<objid>
 359  * O/<seq>/LAST_ID
 360  * last_rcvd
 361  * LAST_GROUP
 362  * CONFIGS
 363  *
 364  */
 365 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
 366 {
 367         struct lvfs_run_ctxt     new;
 368         struct lvfs_run_ctxt     save;
 369         struct dentry           *rootd = osd_sb(dev)->s_root;
 370         struct dentry           *d;
 371         struct osd_thread_info  *info = osd_oti_get(env);
 372         struct inode            *inode;
 373         struct lu_fid           *fid = &info->oti_fid3;
 374         int                      rc;
 375         ENTRY;
 376
 377         OBD_ALLOC_PTR(dev->od_ost_map);
 378         if (dev->od_ost_map == NULL)
 379                 RETURN(-ENOMEM);
 380
 381         /* to get subdir count from last_rcvd */
 382         rc = osd_last_rcvd_subdir_count(dev);
 383         if (rc < 0)
 384                 GOTO(cleanup_alloc, rc);
 385
 386         dev->od_ost_map->om_subdir_count = rc;
 387         rc = 0;
 388
 389         INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 390         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 391         mutex_init(&dev->od_ost_map->om_dir_init_mutex);
 392
 393         osd_push_ctxt(dev, &new, &save);
 394
 395         d = ll_lookup_one_len("O", rootd, strlen("O"));
 396         if (IS_ERR(d))
 397                 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 398         if (d->d_inode == NULL) {
 399                 dput(d);
 400                 /* The lookup() may be called again inside simple_mkdir().
 401                  * Since the repeated lookup() only be called for "/O" at
 402                  * mount time, it will not affect the whole performance. */
 403                 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 404                 if (IS_ERR(d))
 405                         GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 406
 407                 /* It is quite probably that the device is new formatted. */
 408                 dev->od_maybe_new = 1;
 409         }
 410
 411         inode = d->d_inode;
 412         dev->od_ost_map->om_root = d;
 413
 414         /* 'What the @fid is' is not imporatant, because the object
 415          * has no OI mapping, and only is visible inside the OSD.*/
 416         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 417         rc = osd_ea_fid_set(info, inode, fid,
 418                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 419         if (rc)
 420                 GOTO(cleanup_dentry, rc);
 421
 422         pop_ctxt(&save, &new);
 423         RETURN(0);
 424
 425 cleanup_dentry:
 426         dput(d);
 427 cleanup_ctxt:
 428         pop_ctxt(&save, &new);
 429 cleanup_alloc:
 430         OBD_FREE_PTR(dev->od_ost_map);
 431         return rc;
 432 }
 433
 434 static void osd_seq_free(struct osd_obj_seq *osd_seq)
 435 {
 436         int j;
 437
 438         if (osd_seq->oos_dirs) {
 439                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 440                         if (osd_seq->oos_dirs[j])
 441                                 dput(osd_seq->oos_dirs[j]);
 442                 }
 443                 OBD_FREE(osd_seq->oos_dirs,
 444                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 445         }
 446
 447         if (osd_seq->oos_root)
 448                 dput(osd_seq->oos_root);
 449
 450         OBD_FREE_PTR(osd_seq);
 451 }
 452
 453 static void osd_ost_fini(struct osd_device *osd)
 454 {
 455         struct osd_obj_seq    *osd_seq;
 456         struct osd_obj_seq    *tmp;
 457         struct osd_obj_map    *map = osd->od_ost_map;
 458         ENTRY;
 459
 460         if (map == NULL)
 461                 return;
 462
 463         write_lock(&map->om_seq_list_lock);
 464         list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
 465                                  oos_seq_list) {
 466                 list_del_init(&osd_seq->oos_seq_list);
 467                 write_unlock(&map->om_seq_list_lock);
 468                 osd_seq_free(osd_seq);
 469                 write_lock(&map->om_seq_list_lock);
 470         }
 471         write_unlock(&map->om_seq_list_lock);
 472         if (map->om_root)
 473                 dput(map->om_root);
 474         OBD_FREE_PTR(map);
 475         osd->od_ost_map = NULL;
 476         EXIT;
 477 }
 478
 479 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 480 {
 481         int rc;
 482         ENTRY;
 483
 484         /* prepare structures for OST */
 485         rc = osd_ost_init(env, dev);
 486         if (rc)
 487                 RETURN(rc);
 488
 489         /* prepare structures for MDS */
 490         rc = osd_mdt_init(env, dev);
 491         if (rc)
 492                 osd_ost_fini(dev);
 493
 494         RETURN(rc);
 495 }
 496
 497 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
 498 {
 499         struct osd_obj_seq *osd_seq;
 500
 501         list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 502                 if (osd_seq->oos_seq == seq)
 503                         return osd_seq;
 504         }
 505         return NULL;
 506 }
 507
 508 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
 509 {
 510         struct osd_obj_seq *osd_seq;
 511
 512         read_lock(&map->om_seq_list_lock);
 513         osd_seq = osd_seq_find_locked(map, seq);
 514         read_unlock(&map->om_seq_list_lock);
 515         return osd_seq;
 516 }
 517
 518 void osd_obj_map_fini(struct osd_device *dev)
 519 {
 520         osd_ost_fini(dev);
 521         osd_mdt_fini(dev);
 522 }
 523
 524 /**
 525  * Update the specified OI mapping.
 526  *
 527  * \retval   1, changed nothing
 528  * \retval   0, changed successfully
 529  * \retval -ve, on error
 530  */
 531 static int osd_obj_update_entry(struct osd_thread_info *info,
 532                                 struct osd_device *osd,
 533                                 struct dentry *dir, const char *name,
 534                                 const struct lu_fid *fid,
 535                                 const struct osd_inode_id *id,
 536                                 handle_t *th)
 537 {
 538         struct inode               *parent = dir->d_inode;
 539         struct dentry              *child;
 540         struct ldiskfs_dir_entry_2 *de;
 541         struct buffer_head         *bh;
 542         struct inode               *inode;
 543         struct dentry              *dentry = &info->oti_obj_dentry;
 544         struct osd_inode_id        *oi_id  = &info->oti_id3;
 545         struct lustre_mdt_attrs    *lma    = &info->oti_mdt_attrs;
 546         struct lu_fid              *oi_fid = &lma->lma_self_fid;
 547         int                         rc;
 548         ENTRY;
 549
 550         LASSERT(th != NULL);
 551         LASSERT(th->h_transaction != NULL);
 552
 553         child = &info->oti_child_dentry;
 554         child->d_parent = dir;
 555         child->d_name.hash = 0;
 556         child->d_name.name = name;
 557         child->d_name.len = strlen(name);
 558
 559         ll_vfs_dq_init(parent);
 560         mutex_lock(&parent->i_mutex);
 561         bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
 562         if (IS_ERR(bh))
 563                 GOTO(out, rc = PTR_ERR(bh));
 564
 565         if (le32_to_cpu(de->inode) == id->oii_ino)
 566                 GOTO(out, rc = 1);
 567
 568         osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 569         inode = osd_iget(info, osd, oi_id);
 570         if (IS_ERR(inode)) {
 571                 rc = PTR_ERR(inode);
 572                 if (rc == -ENOENT || rc == -ESTALE)
 573                         goto update;
 574                 GOTO(out, rc);
 575         }
 576
 577         /* The EA inode should NOT be in OI, old OI scrub may added
 578          * such OI mapping by wrong, replace it. */
 579         if (unlikely(osd_is_ea_inode(inode))) {
 580                 iput(inode);
 581                 goto update;
 582         }
 583
 584         rc = osd_get_lma(info, inode, dentry, lma);
 585         if (rc == -ENODATA) {
 586                 rc = osd_get_idif(info, inode, dentry, oi_fid);
 587                 if (rc > 0 || rc == -ENODATA) {
 588                         oi_fid = NULL;
 589                         rc = 0;
 590                 }
 591         }
 592         iput(inode);
 593
 594         if (rc != 0)
 595                 GOTO(out, rc);
 596
 597         /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
 598          * either a crashed object or a uninitialized one. Replace it. */
 599         if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
 600                 CERROR("%s: the FID "DFID" is used by two objects: "
 601                        "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
 602                        oi_id->oii_ino, oi_id->oii_gen,
 603                        id->oii_ino, id->oii_gen);
 604                 GOTO(out, rc = -EEXIST);
 605         }
 606
 607         if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
 608                 __u32 idx1 = fid_idif_ost_idx(fid);
 609                 __u32 idx2 = fid_idif_ost_idx(oi_fid);
 610                 struct ost_id *ostid = &info->oti_ostid;
 611                 struct lu_fid *tfid = &info->oti_fid3;
 612
 613                 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
 614                          "invalid given FID "DFID", not match the "
 615                          "device index %u\n", PFID(fid), osd->od_index);
 616
 617                 if (idx1 != idx2) {
 618                         if (idx1 == 0 && idx2 == osd->od_index) {
 619                                 fid_to_ostid(fid, ostid);
 620                                 ostid_to_fid(tfid, ostid, idx2);
 621                                 if (lu_fid_eq(tfid, oi_fid)) {
 622                                         CERROR("%s: the FID "DFID" is used by "
 623                                                "two objects(2): %u/%u %u/%u\n",
 624                                                osd_name(osd), PFID(fid),
 625                                                oi_id->oii_ino, oi_id->oii_gen,
 626                                                id->oii_ino, id->oii_gen);
 627
 628                                         GOTO(out, rc = -EEXIST);
 629                                 }
 630                         } else if (idx2 == 0 && idx1 == osd->od_index) {
 631                                 fid_to_ostid(oi_fid, ostid);
 632                                 ostid_to_fid(tfid, ostid, idx1);
 633                                 if (lu_fid_eq(tfid, fid)) {
 634                                         CERROR("%s: the FID "DFID" is used by "
 635                                                "two objects(2): %u/%u %u/%u\n",
 636                                                osd_name(osd), PFID(fid),
 637                                                oi_id->oii_ino, oi_id->oii_gen,
 638                                                id->oii_ino, id->oii_gen);
 639
 640                                         GOTO(out, rc = -EEXIST);
 641                                 }
 642                         }
 643                 }
 644         }
 645
 646 update:
 647         /* There may be temporary inconsistency: On one hand, the new
 648          * object may be referenced by multiple entries, which is out
 649          * of our control unless we traverse the whole /O completely,
 650          * which is non-flat order and inefficient, should be avoided;
 651          * On the other hand, the old object may become orphan if it
 652          * is still valid. Since it was referenced by an invalid entry,
 653          * making it as invisible temporary may be not worse. OI scrub
 654          * will process it later. */
 655         rc = ldiskfs_journal_get_write_access(th, bh);
 656         if (rc != 0)
 657                 GOTO(out, rc);
 658
 659         de->inode = cpu_to_le32(id->oii_ino);
 660         rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
 661
 662         GOTO(out, rc);
 663
 664 out:
 665         if (!IS_ERR(bh))
 666                 brelse(bh);
 667         mutex_unlock(&parent->i_mutex);
 668         return rc;
 669 }
 670
 671 static int osd_obj_del_entry(struct osd_thread_info *info,
 672                              struct osd_device *osd,
 673                              struct dentry *dird, char *name,
 674                              handle_t *th)
 675 {
 676         struct ldiskfs_dir_entry_2 *de;
 677         struct buffer_head         *bh;
 678         struct dentry              *child;
 679         struct inode               *dir = dird->d_inode;
 680         int                         rc;
 681         ENTRY;
 682
 683         LASSERT(th != NULL);
 684         LASSERT(th->h_transaction != NULL);
 685
 686
 687         child = &info->oti_child_dentry;
 688         child->d_name.hash = 0;
 689         child->d_name.name = name;
 690         child->d_name.len = strlen(name);
 691         child->d_parent = dird;
 692         child->d_inode = NULL;
 693
 694         ll_vfs_dq_init(dir);
 695         mutex_lock(&dir->i_mutex);
 696         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 697         if (IS_ERR(bh)) {
 698                 rc = PTR_ERR(bh);
 699         } else {
 700                 rc = ldiskfs_delete_entry(th, dir, de, bh);
 701                 brelse(bh);
 702         }
 703         mutex_unlock(&dir->i_mutex);
 704
 705         RETURN(rc);
 706 }
 707
 708 static int osd_obj_add_entry(struct osd_thread_info *info,
 709                              struct osd_device *osd,
 710                              struct dentry *dir, char *name,
 711                              const struct osd_inode_id *id,
 712                              handle_t *th)
 713 {
 714         struct dentry *child;
 715         struct inode *inode;
 716         int rc;
 717
 718         ENTRY;
 719
 720         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
 721                 RETURN(0);
 722
 723         LASSERT(th != NULL);
 724         LASSERT(th->h_transaction != NULL);
 725
 726         inode = info->oti_inode;
 727         if (unlikely(inode == NULL)) {
 728                 struct ldiskfs_inode_info *lii;
 729                 OBD_ALLOC_PTR(lii);
 730                 if (lii == NULL)
 731                         RETURN(-ENOMEM);
 732                 inode = info->oti_inode = &lii->vfs_inode;
 733         }
 734
 735         inode->i_sb = osd_sb(osd);
 736         osd_id_to_inode(inode, id);
 737         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 738
 739         child = &info->oti_child_dentry;
 740         child->d_name.hash = 0;
 741         child->d_name.name = name;
 742         child->d_name.len = strlen(name);
 743         child->d_parent = dir;
 744         child->d_inode = inode;
 745
 746         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
 747                 inode->i_ino++;
 748
 749         ll_vfs_dq_init(dir->d_inode);
 750         mutex_lock(&dir->d_inode->i_mutex);
 751         rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
 752         mutex_unlock(&dir->d_inode->i_mutex);
 753
 754         RETURN(rc);
 755 }
 756
 757 /**
 758  * Use %llu for legacy OST sequences, but use %llx for new
 759  * sequences names, so that the O/{seq}/dN/{oid} more closely
 760  * follows the DFID/PFID format. This makes it easier to map from
 761  * debug messages to objects in the future, and the legacy space
 762  * of FID_SEQ_OST_MDT0 will be unused in the future.
 763  **/
 764 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
 765 {
 766         snprintf(seq_name, name_size,
 767                  (fid_seq_is_rsvd(seq) ||
 768                   fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
 769                  fid_seq_is_idif(seq) ? 0 : seq);
 770 }
 771
 772 static inline void osd_oid_name(char *name, size_t name_size,
 773                                 const struct lu_fid *fid, u64 id)
 774 {
 775         snprintf(name, name_size,
 776                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 777                   fid_seq_is_mdt0(fid_seq(fid)) ||
 778                   fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
 779 }
 780
 781 /* external locking is required */
 782 static int osd_seq_load_locked(struct osd_thread_info *info,
 783                                struct osd_device *osd,
 784                                struct osd_obj_seq *osd_seq)
 785 {
 786         struct osd_obj_map  *map = osd->od_ost_map;
 787         struct dentry       *seq_dir;
 788         struct inode        *inode;
 789         struct lu_fid       *fid = &info->oti_fid3;
 790         int                 rc = 0;
 791         int                 i;
 792         char                dir_name[32];
 793         ENTRY;
 794
 795         if (osd_seq->oos_root != NULL)
 796                 RETURN(0);
 797
 798         LASSERT(map);
 799         LASSERT(map->om_root);
 800
 801         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 802
 803         seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
 804         if (IS_ERR(seq_dir))
 805                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 806         else if (seq_dir->d_inode == NULL)
 807                 GOTO(out_put, rc = -EFAULT);
 808
 809         inode = seq_dir->d_inode;
 810         osd_seq->oos_root = seq_dir;
 811
 812         /* 'What the @fid is' is not imporatant, because the object
 813          * has no OI mapping, and only is visible inside the OSD.*/
 814         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 815         rc = osd_ea_fid_set(info, inode, fid,
 816                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 817         if (rc != 0)
 818                 GOTO(out_put, rc);
 819
 820         LASSERT(osd_seq->oos_dirs == NULL);
 821         OBD_ALLOC(osd_seq->oos_dirs,
 822                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 823         if (osd_seq->oos_dirs == NULL)
 824                 GOTO(out_put, rc = -ENOMEM);
 825
 826         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 827                 struct dentry   *dir;
 828
 829                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 830                 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
 831                                    0700, 1);
 832                 if (IS_ERR(dir)) {
 833                         GOTO(out_free, rc = PTR_ERR(dir));
 834                 } else if (dir->d_inode == NULL) {
 835                         dput(dir);
 836                         GOTO(out_free, rc = -EFAULT);
 837                 }
 838
 839                 inode = dir->d_inode;
 840                 osd_seq->oos_dirs[i] = dir;
 841
 842                 /* 'What the @fid is' is not imporatant, because the object
 843                  * has no OI mapping, and only is visible inside the OSD.*/
 844                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
 845                 rc = osd_ea_fid_set(info, inode, fid,
 846                                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 847                 if (rc != 0)
 848                         GOTO(out_free, rc);
 849         }
 850
 851         if (rc != 0) {
 852 out_free:
 853                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 854                         if (osd_seq->oos_dirs[i] != NULL)
 855                                 dput(osd_seq->oos_dirs[i]);
 856                 }
 857                 OBD_FREE(osd_seq->oos_dirs,
 858                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
 859 out_put:
 860                 dput(seq_dir);
 861                 osd_seq->oos_root = NULL;
 862         }
 863 out_err:
 864         RETURN(rc);
 865 }
 866
 867 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
 868                                         struct osd_device *osd, u64 seq)
 869 {
 870         struct osd_obj_map      *map;
 871         struct osd_obj_seq      *osd_seq;
 872         int                     rc = 0;
 873         ENTRY;
 874
 875         map = osd->od_ost_map;
 876         LASSERT(map);
 877         LASSERT(map->om_root);
 878
 879         osd_seq = osd_seq_find(map, seq);
 880         if (likely(osd_seq != NULL))
 881                 RETURN(osd_seq);
 882
 883         /* Serializing init process */
 884         mutex_lock(&map->om_dir_init_mutex);
 885
 886         /* Check whether the seq has been added */
 887         read_lock(&map->om_seq_list_lock);
 888         osd_seq = osd_seq_find_locked(map, seq);
 889         if (osd_seq != NULL) {
 890                 read_unlock(&map->om_seq_list_lock);
 891                 GOTO(cleanup, rc = 0);
 892         }
 893         read_unlock(&map->om_seq_list_lock);
 894
 895         OBD_ALLOC_PTR(osd_seq);
 896         if (osd_seq == NULL)
 897                 GOTO(cleanup, rc = -ENOMEM);
 898
 899         INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 900         osd_seq->oos_seq = seq;
 901         /* Init subdir count to be 32, but each seq can have
 902          * different subdir count */
 903         osd_seq->oos_subdir_count = map->om_subdir_count;
 904         rc = osd_seq_load_locked(info, osd, osd_seq);
 905         if (rc != 0)
 906                 GOTO(cleanup, rc);
 907
 908         write_lock(&map->om_seq_list_lock);
 909         list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 910         write_unlock(&map->om_seq_list_lock);
 911
 912 cleanup:
 913         mutex_unlock(&map->om_dir_init_mutex);
 914         if (rc != 0) {
 915                 if (osd_seq != NULL)
 916                         OBD_FREE_PTR(osd_seq);
 917                 RETURN(ERR_PTR(rc));
 918         }
 919
 920         RETURN(osd_seq);
 921 }
 922
 923 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 924                        const struct lu_fid *fid, struct osd_inode_id *id)
 925 {
 926         struct osd_obj_map              *map;
 927         struct osd_obj_seq              *osd_seq;
 928         struct dentry                   *d_seq;
 929         struct dentry                   *child;
 930         struct ost_id                   *ostid = &info->oti_ostid;
 931         int                             dirn;
 932         char                            name[32];
 933         struct ldiskfs_dir_entry_2      *de;
 934         struct buffer_head              *bh;
 935         struct inode                    *dir;
 936         struct inode                    *inode;
 937         ENTRY;
 938
 939         /* on the very first lookup we find and open directories */
 940
 941         map = dev->od_ost_map;
 942         LASSERT(map);
 943         LASSERT(map->om_root);
 944
 945         fid_to_ostid(fid, ostid);
 946         osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
 947         if (IS_ERR(osd_seq))
 948                 RETURN(PTR_ERR(osd_seq));
 949
 950         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 951         d_seq = osd_seq->oos_dirs[dirn];
 952         LASSERT(d_seq);
 953
 954         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 955
 956         child = &info->oti_child_dentry;
 957         child->d_parent = d_seq;
 958         child->d_name.hash = 0;
 959         child->d_name.name = name;
 960         /* XXX: we can use rc from sprintf() instead of strlen() */
 961         child->d_name.len = strlen(name);
 962
 963         dir = d_seq->d_inode;
 964         mutex_lock(&dir->i_mutex);
 965         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 966         mutex_unlock(&dir->i_mutex);
 967
 968         if (IS_ERR(bh))
 969                 RETURN(PTR_ERR(bh));
 970
 971         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 972         brelse(bh);
 973
 974         inode = osd_iget(info, dev, id);
 975         if (IS_ERR(inode)) {
 976                 int rc = PTR_ERR(inode);
 977
 978                 RETURN(rc == -ENOENT ? -ESTALE : rc);
 979         }
 980
 981         iput(inode);
 982         RETURN(0);
 983 }
 984
 985 int osd_obj_map_insert(struct osd_thread_info *info,
 986                        struct osd_device *osd,
 987                        const struct lu_fid *fid,
 988                        const struct osd_inode_id *id,
 989                        handle_t *th)
 990 {
 991         struct osd_obj_map      *map;
 992         struct osd_obj_seq      *osd_seq;
 993         struct dentry           *d;
 994         struct ost_id           *ostid = &info->oti_ostid;
 995         u64                      oid;
 996         int                     dirn, rc = 0;
 997         char                    name[32];
 998         ENTRY;
 999
1000         map = osd->od_ost_map;
1001         LASSERT(map);
1002
1003         /* map fid to seq:objid */
1004         fid_to_ostid(fid, ostid);
1005
1006         oid = ostid_id(ostid);
1007         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1008         if (IS_ERR(osd_seq))
1009                 RETURN(PTR_ERR(osd_seq));
1010
1011         dirn = oid & (osd_seq->oos_subdir_count - 1);
1012         d = osd_seq->oos_dirs[dirn];
1013         LASSERT(d);
1014
1015         osd_oid_name(name, sizeof(name), fid, oid);
1016
1017 again:
1018         rc = osd_obj_add_entry(info, osd, d, name, id, th);
1019         if (rc == -EEXIST) {
1020                 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1021                 if (unlikely(rc == -ENOENT))
1022                         goto again;
1023
1024                 if (unlikely(rc == 1))
1025                         RETURN(0);
1026         }
1027
1028         RETURN(rc);
1029 }
1030
1031 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1032                        const struct lu_fid *fid, handle_t *th)
1033 {
1034         struct osd_obj_map      *map;
1035         struct osd_obj_seq      *osd_seq;
1036         struct dentry           *d;
1037         struct ost_id           *ostid = &info->oti_ostid;
1038         int                     dirn, rc = 0;
1039         char                    name[32];
1040         ENTRY;
1041
1042         map = osd->od_ost_map;
1043         LASSERT(map);
1044
1045         /* map fid to seq:objid */
1046         fid_to_ostid(fid, ostid);
1047
1048         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1049         if (IS_ERR(osd_seq))
1050                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1051
1052         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1053         d = osd_seq->oos_dirs[dirn];
1054         LASSERT(d);
1055
1056         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1057         rc = osd_obj_del_entry(info, osd, d, name, th);
1058 cleanup:
1059         RETURN(rc);
1060 }
1061
1062 int osd_obj_map_update(struct osd_thread_info *info,
1063                        struct osd_device *osd,
1064                        const struct lu_fid *fid,
1065                        const struct osd_inode_id *id,
1066                        handle_t *th)
1067 {
1068         struct osd_obj_seq      *osd_seq;
1069         struct dentry           *d;
1070         struct ost_id           *ostid = &info->oti_ostid;
1071         int                     dirn, rc = 0;
1072         char                    name[32];
1073         ENTRY;
1074
1075         fid_to_ostid(fid, ostid);
1076         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1077         if (IS_ERR(osd_seq))
1078                 RETURN(PTR_ERR(osd_seq));
1079
1080         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1081         d = osd_seq->oos_dirs[dirn];
1082         LASSERT(d);
1083
1084         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1085         rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1086
1087         RETURN(rc);
1088 }
1089
1090 int osd_obj_map_recover(struct osd_thread_info *info,
1091                         struct osd_device *osd,
1092                         struct inode *src_parent,
1093                         struct dentry *src_child,
1094                         const struct lu_fid *fid)
1095 {
1096         struct osd_obj_seq         *osd_seq;
1097         struct dentry              *tgt_parent;
1098         struct dentry              *tgt_child = &info->oti_child_dentry;
1099         struct inode               *dir;
1100         struct inode               *inode     = src_child->d_inode;
1101         struct ost_id              *ostid     = &info->oti_ostid;
1102         handle_t                   *jh;
1103         struct ldiskfs_dir_entry_2 *de;
1104         struct buffer_head         *bh;
1105         char                        name[32];
1106         int                         dirn;
1107         int                         rc        = 0;
1108         ENTRY;
1109
1110         if (fid_is_last_id(fid)) {
1111                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1112                 if (IS_ERR(osd_seq))
1113                         RETURN(PTR_ERR(osd_seq));
1114
1115                 tgt_parent = osd_seq->oos_root;
1116                 tgt_child->d_name.name = "LAST_ID";
1117                 tgt_child->d_name.len = strlen("LAST_ID");
1118         } else {
1119                 fid_to_ostid(fid, ostid);
1120                 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1121                 if (IS_ERR(osd_seq))
1122                         RETURN(PTR_ERR(osd_seq));
1123
1124                 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1125                 tgt_parent = osd_seq->oos_dirs[dirn];
1126                 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1127                 tgt_child->d_name.name = name;
1128                 tgt_child->d_name.len = strlen(name);
1129         }
1130         LASSERT(tgt_parent != NULL);
1131
1132         dir = tgt_parent->d_inode;
1133         tgt_child->d_name.hash = 0;
1134         tgt_child->d_parent = tgt_parent;
1135         tgt_child->d_inode = inode;
1136
1137         /* The non-initialized src_child may be destroyed. */
1138         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1139                                 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1140                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1141                                 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1142         if (IS_ERR(jh))
1143                 RETURN(PTR_ERR(jh));
1144
1145         ll_vfs_dq_init(src_parent);
1146         ll_vfs_dq_init(dir);
1147
1148         mutex_lock(&src_parent->i_mutex);
1149         mutex_lock(&dir->i_mutex);
1150         bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1151         if (!IS_ERR(bh)) {
1152                 /* XXX: If some other object occupied the same slot. And If such
1153                  *      inode is zero-sized and with SUID+SGID, then means it is
1154                  *      a new created one. Maybe we can remove it and insert the
1155                  *      original one back to the /O/<seq>/d<x>. But there are
1156                  *      something to be considered:
1157                  *
1158                  *      1) The OST-object under /lost+found has crashed LMA.
1159                  *         So it should not conflict with the current one.
1160                  *
1161                  *      2) There are race conditions that: someone may just want
1162                  *         to modify the current one. Even if the OI scrub takes
1163                  *         the object lock when remove the current one, it still
1164                  *         cause the modification to be lost becasue the target
1165                  *         has been removed when the RPC service thread waiting
1166                  *         for the lock.
1167                  *
1168                  *      So keep it there before we have suitable solution. */
1169                 brelse(bh);
1170                 mutex_unlock(&dir->i_mutex);
1171                 mutex_unlock(&src_parent->i_mutex);
1172                 ldiskfs_journal_stop(jh);
1173
1174                 rc = -EEXIST;
1175                 /* If the src object has never been modified, then remove it. */
1176                 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1177                     inode->i_mode & S_ISGID) {
1178                         rc = ll_vfs_unlink(src_parent, src_child);
1179                         if (unlikely(rc == -ENOENT))
1180                                 rc = 0;
1181                 }
1182                 RETURN(rc);
1183         }
1184
1185         bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1186                                     NULL, NULL);
1187         if (unlikely(IS_ERR(bh)))
1188                 GOTO(unlock, rc = PTR_ERR(bh));
1189
1190         rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1191         brelse(bh);
1192         if (rc != 0)
1193                 GOTO(unlock, rc);
1194
1195         rc = osd_ldiskfs_add_entry(info, osd, jh, tgt_child, inode, NULL);
1196
1197         GOTO(unlock, rc);
1198
1199 unlock:
1200         mutex_unlock(&dir->i_mutex);
1201         mutex_unlock(&src_parent->i_mutex);
1202         ldiskfs_journal_stop(jh);
1203         return rc;
1204 }
1205
1206 static struct dentry *
1207 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1208                      const struct lu_fid *fid, char **name)
1209 {
1210         struct dentry *root = ERR_PTR(-ENOENT);
1211
1212         if (fid_is_last_id(fid)) {
1213                 struct osd_obj_seq *osd_seq;
1214
1215                 /* on creation of LAST_ID we create O/<seq> hierarchy */
1216                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1217                 if (IS_ERR(osd_seq))
1218                         RETURN((struct dentry *)osd_seq);
1219
1220                 *name = "LAST_ID";
1221                 root = osd_seq->oos_root;
1222         } else {
1223                 *name = osd_lf_fid2name(fid);
1224                 if (*name == NULL)
1225                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1226                 else if ((*name)[0])
1227                         root = osd_sb(osd)->s_root;
1228         }
1229
1230         return root;
1231 }
1232
1233 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1234                         const struct lu_fid *fid, const struct osd_inode_id *id,
1235                         handle_t *th)
1236 {
1237         struct dentry   *root;
1238         char            *name = NULL;
1239         int              rc;
1240         ENTRY;
1241
1242         root = osd_object_spec_find(info, osd, fid, &name);
1243         if (!IS_ERR(root)) {
1244                 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1245         } else {
1246                 rc = PTR_ERR(root);
1247                 if (rc == -ENOENT)
1248                         rc = 1;
1249         }
1250
1251         RETURN(rc);
1252 }
1253
1254 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1255                         const struct lu_fid *fid, const struct osd_inode_id *id,
1256                         handle_t *th)
1257 {
1258         struct dentry   *root;
1259         char            *name = NULL;
1260         int              rc;
1261         ENTRY;
1262
1263         root = osd_object_spec_find(info, osd, fid, &name);
1264         if (!IS_ERR(root)) {
1265                 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1266         } else {
1267                 rc = PTR_ERR(root);
1268                 if (rc == -ENOENT)
1269                         rc = 0;
1270         }
1271
1272         RETURN(rc);
1273 }
1274
1275 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1276                         const struct lu_fid *fid, struct osd_inode_id *id)
1277 {
1278         struct dentry   *root;
1279         struct dentry   *dentry;
1280         struct inode    *inode;
1281         char            *name = NULL;
1282         int             rc = -ENOENT;
1283         ENTRY;
1284
1285         if (fid_is_last_id(fid)) {
1286                 struct osd_obj_seq *osd_seq;
1287
1288                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1289                 if (IS_ERR(osd_seq))
1290                         RETURN(PTR_ERR(osd_seq));
1291                 root = osd_seq->oos_root;
1292                 name = "LAST_ID";
1293         } else {
1294                 root = osd_sb(osd)->s_root;
1295                 name = osd_lf_fid2name(fid);
1296                 if (name == NULL || strlen(name) == 0)
1297                         RETURN(-ENOENT);
1298         }
1299
1300         dentry = ll_lookup_one_len(name, root, strlen(name));
1301         if (!IS_ERR(dentry)) {
1302                 inode = dentry->d_inode;
1303                 if (inode) {
1304                         if (is_bad_inode(inode)) {
1305                                 rc = -EIO;
1306                         } else {
1307                                 osd_id_gen(id, inode->i_ino,
1308                                            inode->i_generation);
1309                                 rc = 0;
1310                         }
1311                 }
1312                 /* if dentry is accessible after osd_compat_spec_insert it
1313                  * will still contain NULL inode, so don't keep it in cache */
1314                 d_invalidate(dentry);
1315                 dput(dentry);
1316         }
1317
1318         RETURN(rc);
1319 }