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, 2017, 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 /* prerequisite for linux/xattr.h */
  40 #include <linux/types.h>
  41 /* prerequisite for linux/xattr.h */
  42 #include <linux/fs.h>
  43 /* XATTR_{REPLACE,CREATE} */
  44 #include <linux/xattr.h>
  45
  46 /*
  47  * struct OBD_{ALLOC,FREE}*()
  48  * OBD_FAIL_CHECK
  49  */
  50 #include <obd_support.h>
  51
  52 #include "osd_internal.h"
  53 #include "osd_oi.h"
  54
  55 static void osd_push_ctxt(const struct osd_device *dev,
  56                           struct lvfs_run_ctxt *newctxt,
  57                           struct lvfs_run_ctxt *save)
  58 {
  59         OBD_SET_CTXT_MAGIC(newctxt);
  60         newctxt->pwdmnt = dev->od_mnt;
  61         newctxt->pwd = dev->od_mnt->mnt_root;
  62         newctxt->fs = get_ds();
  63         newctxt->umask = current_umask();
  64         newctxt->dt = NULL;
  65
  66         push_ctxt(save, newctxt);
  67 }
  68
  69 /* utility to make a directory */
  70 static struct dentry *
  71 simple_mkdir(const struct lu_env *env, struct osd_device *osd,
  72              struct dentry *dir, const struct lu_fid *fid,
  73              const char *name, __u32 compat, int mode, bool *created)
  74 {
  75         struct osd_thread_info *info = osd_oti_get(env);
  76         struct lu_fid *tfid = &info->oti_fid3;
  77         struct inode *inode;
  78         struct dentry *dchild;
  79         int err = 0;
  80
  81         ENTRY;
  82
  83         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
  84         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
  85         dchild = ll_lookup_one_len(name, dir, strlen(name));
  86         if (IS_ERR(dchild))
  87                 RETURN(dchild);
  88
  89         inode = dchild->d_inode;
  90         if (inode) {
  91                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
  92                 int old_mode = inode->i_mode;
  93
  94                 if (created)
  95                         *created = false;
  96
  97                 if (!S_ISDIR(old_mode)) {
  98                         CERROR("found %s (%lu/%u) is mode %o\n", name,
  99                                inode->i_ino, inode->i_generation, old_mode);
 100                         GOTO(out_err, err = -ENOTDIR);
 101                 }
 102
 103                 if (unlikely(osd->od_dt_dev.dd_rdonly))
 104                         RETURN(dchild);
 105
 106                 /* Fixup directory permissions if necessary */
 107                 if ((old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 108                         CDEBUG(D_CONFIG,
 109                                "fixing permissions on %s from %o to %o\n",
 110                                name, old_mode, mode);
 111                         inode->i_mode = (mode & S_IALLUGO) |
 112                                         (old_mode & ~S_IALLUGO);
 113                         mark_inode_dirty(inode);
 114                 }
 115
 116                 err = osd_get_lma(info, inode, &info->oti_obj_dentry,
 117                                   &info->oti_ost_attrs);
 118                 if (err == -ENODATA)
 119                         goto set_fid;
 120
 121                 if (err)
 122                         GOTO(out_err, err);
 123
 124                 if ((fid && !lu_fid_eq(fid, &lma->lma_self_fid)) ||
 125                     lma->lma_compat != compat)
 126                         goto set_fid;
 127
 128                 RETURN(dchild);
 129         }
 130
 131         err = vfs_mkdir(dir->d_inode, dchild, mode);
 132         if (err)
 133                 GOTO(out_err, err);
 134
 135         inode = dchild->d_inode;
 136         if (created)
 137                 *created = true;
 138
 139 set_fid:
 140         if (fid)
 141                 *tfid = *fid;
 142         else
 143                 lu_igif_build(tfid, inode->i_ino, inode->i_generation);
 144         err = osd_ea_fid_set(info, inode, tfid, compat, 0);
 145         if (err)
 146                 GOTO(out_err, err);
 147
 148         RETURN(dchild);
 149
 150 out_err:
 151         dput(dchild);
 152         return ERR_PTR(err);
 153 }
 154
 155 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
 156 {
 157         struct lr_server_data lsd;
 158         struct dentry *dlast;
 159         loff_t off;
 160         int rc = 0;
 161         int count = OBJ_SUBDIR_COUNT;
 162
 163         ENTRY;
 164
 165         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
 166                                   strlen(LAST_RCVD));
 167         if (IS_ERR(dlast))
 168                 return PTR_ERR(dlast);
 169         else if (dlast->d_inode == NULL)
 170                 goto out;
 171
 172         off = 0;
 173         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 174         if (rc == sizeof(lsd)) {
 175                 CDEBUG(D_INFO,
 176                       "read last_rcvd header, uuid = %s, subdir count = %d\n",
 177                       lsd.lsd_uuid, lsd.lsd_subdir_count);
 178                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 179                         count = le16_to_cpu(lsd.lsd_subdir_count);
 180         } else if (rc != 0) {
 181                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 182                 if (rc > 0)
 183                         rc = -EFAULT;
 184                 dput(dlast);
 185                 return rc;
 186         }
 187 out:
 188         dput(dlast);
 189         LASSERT(count > 0);
 190         return count;
 191 }
 192
 193 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
 194 {
 195         struct lvfs_run_ctxt new;
 196         struct lvfs_run_ctxt save;
 197         struct dentry *parent;
 198         struct osd_mdobj_map *omm;
 199         struct dentry *d;
 200         struct osd_thread_info *info = osd_oti_get(env);
 201         struct lu_fid *fid = &info->oti_fid3;
 202         int rc = 0;
 203
 204         ENTRY;
 205
 206         OBD_ALLOC_PTR(dev->od_mdt_map);
 207         if (dev->od_mdt_map == NULL)
 208                 RETURN(-ENOMEM);
 209
 210         omm = dev->od_mdt_map;
 211
 212         parent = osd_sb(dev)->s_root;
 213         osd_push_ctxt(dev, &new, &save);
 214
 215         lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
 216         d = simple_mkdir(env, dev, parent, fid, REMOTE_PARENT_DIR,
 217                          LMAC_NOT_IN_OI, 0755, NULL);
 218         if (IS_ERR(d))
 219                 GOTO(cleanup, rc = PTR_ERR(d));
 220
 221         omm->omm_remote_parent = d;
 222
 223         GOTO(cleanup, rc = 0);
 224
 225 cleanup:
 226         pop_ctxt(&save, &new);
 227         if (rc) {
 228                 if (omm->omm_remote_parent != NULL)
 229                         dput(omm->omm_remote_parent);
 230                 OBD_FREE_PTR(omm);
 231                 dev->od_mdt_map = NULL;
 232         }
 233         return rc;
 234 }
 235
 236 static void osd_mdt_fini(struct osd_device *osd)
 237 {
 238         struct osd_mdobj_map *omm = osd->od_mdt_map;
 239
 240         if (omm == NULL)
 241                 return;
 242
 243         if (omm->omm_remote_parent)
 244                 dput(omm->omm_remote_parent);
 245
 246         OBD_FREE_PTR(omm);
 247         osd->od_ost_map = NULL;
 248 }
 249
 250 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
 251                              struct osd_object *obj, struct osd_thandle *oh)
 252 {
 253         struct osd_mdobj_map *omm = osd->od_mdt_map;
 254         struct osd_thread_info *oti = osd_oti_get(env);
 255         struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
 256         char *name = oti->oti_name;
 257         struct osd_thread_info *info = osd_oti_get(env);
 258         struct dentry *dentry;
 259         struct dentry *parent;
 260         int rc;
 261
 262         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AGENTENT))
 263                 RETURN(0);
 264
 265         /*
 266          * Set REMOTE_PARENT in lma, so other process like unlink or lfsck
 267          * can identify this object quickly
 268          */
 269         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
 270                          &oti->oti_ost_attrs);
 271         if (rc)
 272                 RETURN(rc);
 273
 274         lma->lma_incompat |= LMAI_REMOTE_PARENT;
 275         lustre_lma_swab(lma);
 276         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 277                              sizeof(*lma), XATTR_REPLACE);
 278         if (rc)
 279                 RETURN(rc);
 280
 281         parent = omm->omm_remote_parent;
 282         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 283         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 284                                            name, strlen(name));
 285         inode_lock(parent->d_inode);
 286         rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
 287                                    obj->oo_inode, NULL);
 288         if (!rc && S_ISDIR(obj->oo_inode->i_mode))
 289                 ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
 290         else if (unlikely(rc == -EEXIST))
 291                 rc = 0;
 292         if (!rc)
 293                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 294         CDEBUG(D_INODE, "%s: create agent entry for %s: rc = %d\n",
 295                osd_name(osd), name, rc);
 296         mark_inode_dirty(parent->d_inode);
 297         inode_unlock(parent->d_inode);
 298         RETURN(rc);
 299 }
 300
 301 int osd_delete_from_remote_parent(const struct lu_env *env,
 302                                   struct osd_device *osd,
 303                                   struct osd_object *obj,
 304                                   struct osd_thandle *oh, bool destroy)
 305 {
 306         struct osd_mdobj_map *omm = osd->od_mdt_map;
 307         struct osd_thread_info *oti = osd_oti_get(env);
 308         struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
 309         char *name = oti->oti_name;
 310         struct dentry *dentry;
 311         struct dentry *parent;
 312         struct ldiskfs_dir_entry_2 *de;
 313         struct buffer_head *bh;
 314         int rc;
 315
 316         parent = omm->omm_remote_parent;
 317         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 318         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 319                                            name, strlen(name));
 320         inode_lock(parent->d_inode);
 321         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 322                                     NULL, NULL);
 323         if (IS_ERR(bh)) {
 324                 inode_unlock(parent->d_inode);
 325                 rc = PTR_ERR(bh);
 326                 if (unlikely(rc == -ENOENT))
 327                         rc = 0;
 328         } else {
 329                 rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode,
 330                                           de, bh);
 331                 if (!rc && S_ISDIR(obj->oo_inode->i_mode))
 332                         ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
 333                 mark_inode_dirty(parent->d_inode);
 334                 inode_unlock(parent->d_inode);
 335                 brelse(bh);
 336                 CDEBUG(D_INODE, "%s: remove agent entry for %s: rc = %d\n",
 337                        osd_name(osd), name, rc);
 338         }
 339
 340         if (destroy || rc) {
 341                 if (!rc)
 342                         lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
 343
 344                 RETURN(rc);
 345         }
 346
 347         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
 348                          &oti->oti_ost_attrs);
 349         if (rc)
 350                 RETURN(rc);
 351
 352         /* Get rid of REMOTE_PARENT flag from incompat */
 353         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
 354         lustre_lma_swab(lma);
 355         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 356                              sizeof(*lma), XATTR_REPLACE);
 357         if (!rc)
 358                 lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
 359         RETURN(rc);
 360 }
 361
 362 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 363                                 struct osd_device *osd,
 364                                 const struct lu_fid *fid,
 365                                 struct osd_inode_id *id)
 366 {
 367         struct osd_mdobj_map *omm = osd->od_mdt_map;
 368         char *name = oti->oti_name;
 369         struct dentry *parent;
 370         struct dentry *dentry;
 371         struct ldiskfs_dir_entry_2 *de;
 372         struct buffer_head *bh;
 373         int rc;
 374
 375         ENTRY;
 376
 377         if (unlikely(osd->od_is_ost))
 378                 RETURN(-ENOENT);
 379
 380         parent = omm->omm_remote_parent;
 381         sprintf(name, DFID_NOBRACE, PFID(fid));
 382         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
 383                                            name, strlen(name));
 384         inode_lock(parent->d_inode);
 385         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 386                                     NULL, NULL);
 387         if (IS_ERR(bh)) {
 388                 rc = PTR_ERR(bh);
 389         } else {
 390                 struct inode *inode;
 391
 392                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 393                 brelse(bh);
 394                 inode = osd_iget(oti, osd, id);
 395                 if (IS_ERR(inode)) {
 396                         rc = PTR_ERR(inode);
 397                         if (rc == -ESTALE)
 398                                 rc = -ENOENT;
 399                 } else {
 400                         iput(inode);
 401                         rc = 0;
 402                 }
 403         }
 404         inode_unlock(parent->d_inode);
 405         if (rc == 0)
 406                 osd_add_oi_cache(oti, osd, id, fid);
 407         RETURN(rc);
 408 }
 409
 410 /*
 411  * directory structure on legacy OST:
 412  *
 413  * O/<seq>/d0-31/<objid>
 414  * O/<seq>/LAST_ID
 415  * last_rcvd
 416  * LAST_GROUP
 417  * CONFIGS
 418  *
 419  */
 420 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
 421 {
 422         struct lvfs_run_ctxt new;
 423         struct lvfs_run_ctxt save;
 424         struct dentry *d;
 425         int rc;
 426         bool created = false;
 427
 428         ENTRY;
 429
 430         OBD_ALLOC_PTR(dev->od_ost_map);
 431         if (dev->od_ost_map == NULL)
 432                 RETURN(-ENOMEM);
 433
 434         /* to get subdir count from last_rcvd */
 435         rc = osd_last_rcvd_subdir_count(dev);
 436         if (rc < 0)
 437                 GOTO(cleanup_alloc, rc);
 438
 439         dev->od_ost_map->om_subdir_count = rc;
 440         INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 441         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 442         mutex_init(&dev->od_ost_map->om_dir_init_mutex);
 443
 444         osd_push_ctxt(dev, &new, &save);
 445         d = simple_mkdir(env, dev, osd_sb(dev)->s_root, NULL, "O",
 446                          LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, &created);
 447         if (IS_ERR(d))
 448                 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 449
 450         if (created)
 451                 /* It is quite probably that the device is new formatted. */
 452                 dev->od_maybe_new = 1;
 453
 454         dev->od_ost_map->om_root = d;
 455
 456         pop_ctxt(&save, &new);
 457         RETURN(0);
 458
 459 cleanup_ctxt:
 460         pop_ctxt(&save, &new);
 461 cleanup_alloc:
 462         OBD_FREE_PTR(dev->od_ost_map);
 463         return rc;
 464 }
 465
 466 static void osd_seq_free(struct osd_obj_seq *osd_seq)
 467 {
 468         int j;
 469
 470         if (osd_seq->oos_dirs) {
 471                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 472                         if (osd_seq->oos_dirs[j])
 473                                 dput(osd_seq->oos_dirs[j]);
 474                 }
 475                 OBD_FREE(osd_seq->oos_dirs,
 476                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 477         }
 478
 479         if (osd_seq->oos_root)
 480                 dput(osd_seq->oos_root);
 481
 482         OBD_FREE_PTR(osd_seq);
 483 }
 484
 485 static void osd_ost_fini(struct osd_device *osd)
 486 {
 487         struct osd_obj_seq *osd_seq;
 488         struct osd_obj_seq *tmp;
 489         struct osd_obj_map *map = osd->od_ost_map;
 490
 491         ENTRY;
 492
 493         if (map == NULL)
 494                 return;
 495
 496         write_lock(&map->om_seq_list_lock);
 497         list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
 498                                  oos_seq_list) {
 499                 list_del_init(&osd_seq->oos_seq_list);
 500                 write_unlock(&map->om_seq_list_lock);
 501                 osd_seq_free(osd_seq);
 502                 write_lock(&map->om_seq_list_lock);
 503         }
 504         write_unlock(&map->om_seq_list_lock);
 505         if (map->om_root)
 506                 dput(map->om_root);
 507         OBD_FREE_PTR(map);
 508         osd->od_ost_map = NULL;
 509         EXIT;
 510 }
 511
 512 static int osd_index_backup_dir_init(const struct lu_env *env,
 513                                      struct osd_device *dev)
 514 {
 515         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
 516         struct lvfs_run_ctxt new;
 517         struct lvfs_run_ctxt save;
 518         struct dentry *dentry;
 519         int rc = 0;
 520
 521         ENTRY;
 522
 523         lu_local_obj_fid(fid, INDEX_BACKUP_OID);
 524         osd_push_ctxt(dev, &new, &save);
 525         dentry = simple_mkdir(env, dev, osd_sb(dev)->s_root, fid,
 526                               INDEX_BACKUP_DIR, LMAC_NOT_IN_OI, 0755, NULL);
 527         if (IS_ERR(dentry)) {
 528                 rc = PTR_ERR(dentry);
 529         } else {
 530                 dev->od_index_backup_inode = igrab(dentry->d_inode);
 531                 dput(dentry);
 532         }
 533         pop_ctxt(&save, &new);
 534
 535         RETURN(rc);
 536 }
 537
 538 static void osd_index_backup_dir_fini(struct osd_device *dev)
 539 {
 540         if (dev->od_index_backup_inode) {
 541                 iput(dev->od_index_backup_inode);
 542                 dev->od_index_backup_inode = NULL;
 543         }
 544 }
 545
 546 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 547 {
 548         int rc;
 549         bool mdt_init = false;
 550
 551         ENTRY;
 552
 553         rc = osd_ost_init(env, dev);
 554         if (rc)
 555                 RETURN(rc);
 556
 557         if (!dev->od_is_ost) {
 558                 rc = osd_mdt_init(env, dev);
 559                 if (rc) {
 560                         osd_ost_fini(dev);
 561                         RETURN(rc);
 562                 }
 563
 564                 mdt_init = true;
 565         }
 566
 567         rc = osd_index_backup_dir_init(env, dev);
 568         if (rc) {
 569                 osd_ost_fini(dev);
 570                 if (mdt_init)
 571                         osd_mdt_fini(dev);
 572         }
 573
 574         RETURN(rc);
 575 }
 576
 577 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
 578 {
 579         struct osd_obj_seq *osd_seq;
 580
 581         list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 582                 if (osd_seq->oos_seq == seq)
 583                         return osd_seq;
 584         }
 585         return NULL;
 586 }
 587
 588 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
 589 {
 590         struct osd_obj_seq *osd_seq;
 591
 592         read_lock(&map->om_seq_list_lock);
 593         osd_seq = osd_seq_find_locked(map, seq);
 594         read_unlock(&map->om_seq_list_lock);
 595         return osd_seq;
 596 }
 597
 598 void osd_obj_map_fini(struct osd_device *dev)
 599 {
 600         osd_index_backup_dir_fini(dev);
 601         osd_ost_fini(dev);
 602         osd_mdt_fini(dev);
 603 }
 604
 605 /**
 606  * Update the specified OI mapping.
 607  *
 608  * \retval   1, changed nothing
 609  * \retval   0, changed successfully
 610  * \retval -ve, on error
 611  */
 612 static int osd_obj_update_entry(struct osd_thread_info *info,
 613                                 struct osd_device *osd,
 614                                 struct dentry *dir, const char *name,
 615                                 const struct lu_fid *fid,
 616                                 const struct osd_inode_id *id,
 617                                 handle_t *th)
 618 {
 619         struct inode *parent = dir->d_inode;
 620         struct dentry *child;
 621         struct ldiskfs_dir_entry_2 *de;
 622         struct buffer_head *bh;
 623         struct inode *inode;
 624         struct dentry *dentry = &info->oti_obj_dentry;
 625         struct osd_inode_id *oi_id = &info->oti_id3;
 626         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 627         struct lu_fid *oi_fid = &lma->lma_self_fid;
 628         int rc;
 629
 630         ENTRY;
 631
 632         LASSERT(th != NULL);
 633         LASSERT(th->h_transaction != NULL);
 634
 635         child = &info->oti_child_dentry;
 636         child->d_parent = dir;
 637         child->d_name.hash = 0;
 638         child->d_name.name = name;
 639         child->d_name.len = strlen(name);
 640
 641         ll_vfs_dq_init(parent);
 642         inode_lock(parent);
 643         bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
 644         if (IS_ERR(bh))
 645                 GOTO(out, rc = PTR_ERR(bh));
 646
 647         if (le32_to_cpu(de->inode) == id->oii_ino)
 648                 GOTO(out, rc = 1);
 649
 650         osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 651         inode = osd_iget(info, osd, oi_id);
 652         if (IS_ERR(inode)) {
 653                 rc = PTR_ERR(inode);
 654                 if (rc == -ENOENT || rc == -ESTALE)
 655                         goto update;
 656                 GOTO(out, rc);
 657         }
 658
 659         /*
 660          * The EA inode should NOT be in OI, old OI scrub may added
 661          * such OI mapping by wrong, replace it.
 662          */
 663         if (unlikely(osd_is_ea_inode(inode))) {
 664                 iput(inode);
 665                 goto update;
 666         }
 667
 668         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
 669         if (rc == -ENODATA) {
 670                 rc = osd_get_idif(info, inode, dentry, oi_fid);
 671                 if (rc > 0 || rc == -ENODATA) {
 672                         oi_fid = NULL;
 673                         rc = 0;
 674                 }
 675         }
 676         iput(inode);
 677
 678         if (rc != 0)
 679                 GOTO(out, rc);
 680
 681         /*
 682          * If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
 683          * either a crashed object or a uninitialized one. Replace it.
 684          */
 685         if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
 686                 CERROR("%s: the FID "DFID" is used by two objects: "
 687                        "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
 688                        oi_id->oii_ino, oi_id->oii_gen,
 689                        id->oii_ino, id->oii_gen);
 690                 GOTO(out, rc = -EEXIST);
 691         }
 692
 693         if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
 694                 __u32 idx1 = fid_idif_ost_idx(fid);
 695                 __u32 idx2 = fid_idif_ost_idx(oi_fid);
 696                 struct ost_id *ostid = &info->oti_ostid;
 697                 struct lu_fid *tfid = &info->oti_fid3;
 698
 699                 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
 700                          "invalid given FID "DFID", not match the "
 701                          "device index %u\n", PFID(fid), osd->od_index);
 702
 703                 if (idx1 != idx2) {
 704                         if (idx1 == 0 && idx2 == osd->od_index) {
 705                                 fid_to_ostid(fid, ostid);
 706                                 ostid_to_fid(tfid, ostid, idx2);
 707                                 if (lu_fid_eq(tfid, oi_fid)) {
 708                                         CERROR("%s: the FID "DFID" is used by "
 709                                                "two objects(2): %u/%u %u/%u\n",
 710                                                osd_name(osd), PFID(fid),
 711                                                oi_id->oii_ino, oi_id->oii_gen,
 712                                                id->oii_ino, id->oii_gen);
 713
 714                                         GOTO(out, rc = -EEXIST);
 715                                 }
 716                         } else if (idx2 == 0 && idx1 == osd->od_index) {
 717                                 fid_to_ostid(oi_fid, ostid);
 718                                 ostid_to_fid(tfid, ostid, idx1);
 719                                 if (lu_fid_eq(tfid, fid)) {
 720                                         CERROR("%s: the FID "DFID" is used by "
 721                                                "two objects(2): %u/%u %u/%u\n",
 722                                                osd_name(osd), PFID(fid),
 723                                                oi_id->oii_ino, oi_id->oii_gen,
 724                                                id->oii_ino, id->oii_gen);
 725
 726                                         GOTO(out, rc = -EEXIST);
 727                                 }
 728                         }
 729                 }
 730         }
 731
 732 update:
 733         /*
 734          * There may be temporary inconsistency: On one hand, the new
 735          * object may be referenced by multiple entries, which is out
 736          * of our control unless we traverse the whole /O completely,
 737          * which is non-flat order and inefficient, should be avoided;
 738          * On the other hand, the old object may become orphan if it
 739          * is still valid. Since it was referenced by an invalid entry,
 740          * making it as invisible temporary may be not worse. OI scrub
 741          * will process it later.
 742          */
 743         rc = ldiskfs_journal_get_write_access(th, bh);
 744         if (rc != 0)
 745                 GOTO(out, rc);
 746
 747         de->inode = cpu_to_le32(id->oii_ino);
 748         rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
 749
 750         GOTO(out, rc);
 751
 752 out:
 753         if (!IS_ERR(bh))
 754                 brelse(bh);
 755         inode_unlock(parent);
 756         return rc;
 757 }
 758
 759 static int osd_obj_del_entry(struct osd_thread_info *info,
 760                              struct osd_device *osd,
 761                              struct dentry *dird, char *name,
 762                              handle_t *th)
 763 {
 764         struct ldiskfs_dir_entry_2 *de;
 765         struct buffer_head *bh;
 766         struct dentry *child;
 767         struct inode *dir = dird->d_inode;
 768         int rc;
 769
 770         ENTRY;
 771
 772         LASSERT(th != NULL);
 773         LASSERT(th->h_transaction != NULL);
 774
 775         child = &info->oti_child_dentry;
 776         child->d_name.hash = 0;
 777         child->d_name.name = name;
 778         child->d_name.len = strlen(name);
 779         child->d_parent = dird;
 780         child->d_inode = NULL;
 781
 782         ll_vfs_dq_init(dir);
 783         inode_lock(dir);
 784         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 785         if (IS_ERR(bh)) {
 786                 rc = PTR_ERR(bh);
 787         } else {
 788                 rc = ldiskfs_delete_entry(th, dir, de, bh);
 789                 brelse(bh);
 790         }
 791         inode_unlock(dir);
 792
 793         RETURN(rc);
 794 }
 795
 796 static int osd_obj_add_entry(struct osd_thread_info *info,
 797                              struct osd_device *osd,
 798                              struct dentry *dir, char *name,
 799                              const struct osd_inode_id *id,
 800                              handle_t *th)
 801 {
 802         struct dentry *child;
 803         struct inode *inode;
 804         int rc;
 805
 806         ENTRY;
 807
 808         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
 809                 RETURN(0);
 810
 811         LASSERT(th != NULL);
 812         LASSERT(th->h_transaction != NULL);
 813
 814         inode = info->oti_inode;
 815         if (unlikely(inode == NULL)) {
 816                 struct ldiskfs_inode_info *lii;
 817
 818                 OBD_ALLOC_PTR(lii);
 819                 if (lii == NULL)
 820                         RETURN(-ENOMEM);
 821                 inode = info->oti_inode = &lii->vfs_inode;
 822         }
 823
 824         inode->i_sb = osd_sb(osd);
 825         osd_id_to_inode(inode, id);
 826         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 827
 828         child = &info->oti_child_dentry;
 829         child->d_name.hash = 0;
 830         child->d_name.name = name;
 831         child->d_name.len = strlen(name);
 832         child->d_parent = dir;
 833         child->d_inode = inode;
 834
 835         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
 836                 inode->i_ino++;
 837
 838         ll_vfs_dq_init(dir->d_inode);
 839         inode_lock(dir->d_inode);
 840         rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
 841         inode_unlock(dir->d_inode);
 842
 843         RETURN(rc);
 844 }
 845
 846 /**
 847  * Use %llu for legacy OST sequences, but use %llx for new
 848  * sequences names, so that the O/{seq}/dN/{oid} more closely
 849  * follows the DFID/PFID format. This makes it easier to map from
 850  * debug messages to objects in the future, and the legacy space
 851  * of FID_SEQ_OST_MDT0 will be unused in the future.
 852  **/
 853 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
 854 {
 855         snprintf(seq_name, name_size,
 856                  (fid_seq_is_rsvd(seq) ||
 857                   fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
 858                  fid_seq_is_idif(seq) ? 0 : seq);
 859 }
 860
 861 static inline void osd_oid_name(char *name, size_t name_size,
 862                                 const struct lu_fid *fid, u64 id)
 863 {
 864         snprintf(name, name_size,
 865                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 866                   fid_seq_is_mdt0(fid_seq(fid)) ||
 867                   fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
 868 }
 869
 870 /* external locking is required */
 871 static int osd_seq_load_locked(struct osd_thread_info *info,
 872                                struct osd_device *osd,
 873                                struct osd_obj_seq *osd_seq)
 874 {
 875         struct osd_obj_map *map = osd->od_ost_map;
 876         struct dentry *seq_dir;
 877         int rc = 0;
 878         int i;
 879         char dir_name[32];
 880
 881         ENTRY;
 882
 883         if (osd_seq->oos_root != NULL)
 884                 RETURN(0);
 885
 886         LASSERT(map);
 887         LASSERT(map->om_root);
 888
 889         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 890
 891         seq_dir = simple_mkdir(info->oti_env, osd, map->om_root, NULL, dir_name,
 892                                LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, NULL);
 893         if (IS_ERR(seq_dir))
 894                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 895         else if (seq_dir->d_inode == NULL)
 896                 GOTO(out_put, rc = -EFAULT);
 897
 898         osd_seq->oos_root = seq_dir;
 899
 900         LASSERT(osd_seq->oos_dirs == NULL);
 901         OBD_ALLOC(osd_seq->oos_dirs,
 902                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 903         if (osd_seq->oos_dirs == NULL)
 904                 GOTO(out_put, rc = -ENOMEM);
 905
 906         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 907                 struct dentry   *dir;
 908
 909                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 910                 dir = simple_mkdir(info->oti_env, osd, osd_seq->oos_root, NULL,
 911                                    dir_name, LMAC_NOT_IN_OI | LMAC_FID_ON_OST,
 912                                    0700, NULL);
 913                 if (IS_ERR(dir)) {
 914                         GOTO(out_free, rc = PTR_ERR(dir));
 915                 } else if (dir->d_inode == NULL) {
 916                         dput(dir);
 917                         GOTO(out_free, rc = -EFAULT);
 918                 }
 919
 920                 osd_seq->oos_dirs[i] = dir;
 921         }
 922
 923         if (rc != 0) {
 924 out_free:
 925                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 926                         if (osd_seq->oos_dirs[i] != NULL)
 927                                 dput(osd_seq->oos_dirs[i]);
 928                 }
 929                 OBD_FREE(osd_seq->oos_dirs,
 930                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
 931 out_put:
 932                 dput(seq_dir);
 933                 osd_seq->oos_root = NULL;
 934         }
 935 out_err:
 936         RETURN(rc);
 937 }
 938
 939 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
 940                                         struct osd_device *osd, u64 seq)
 941 {
 942         struct osd_obj_map *map;
 943         struct osd_obj_seq *osd_seq;
 944         int rc = 0;
 945
 946         ENTRY;
 947
 948         map = osd->od_ost_map;
 949         LASSERT(map);
 950         LASSERT(map->om_root);
 951
 952         osd_seq = osd_seq_find(map, seq);
 953         if (likely(osd_seq != NULL))
 954                 RETURN(osd_seq);
 955
 956         /* Serializing init process */
 957         mutex_lock(&map->om_dir_init_mutex);
 958
 959         /* Check whether the seq has been added */
 960         read_lock(&map->om_seq_list_lock);
 961         osd_seq = osd_seq_find_locked(map, seq);
 962         if (osd_seq != NULL) {
 963                 read_unlock(&map->om_seq_list_lock);
 964                 GOTO(cleanup, rc = 0);
 965         }
 966         read_unlock(&map->om_seq_list_lock);
 967
 968         OBD_ALLOC_PTR(osd_seq);
 969         if (osd_seq == NULL)
 970                 GOTO(cleanup, rc = -ENOMEM);
 971
 972         INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 973         osd_seq->oos_seq = seq;
 974         /*
 975          * Init subdir count to be 32, but each seq can have
 976          * different subdir count
 977          */
 978         osd_seq->oos_subdir_count = map->om_subdir_count;
 979         rc = osd_seq_load_locked(info, osd, osd_seq);
 980         if (rc != 0)
 981                 GOTO(cleanup, rc);
 982
 983         write_lock(&map->om_seq_list_lock);
 984         list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 985         write_unlock(&map->om_seq_list_lock);
 986
 987 cleanup:
 988         mutex_unlock(&map->om_dir_init_mutex);
 989         if (rc != 0) {
 990                 if (osd_seq != NULL)
 991                         OBD_FREE_PTR(osd_seq);
 992                 RETURN(ERR_PTR(rc));
 993         }
 994
 995         RETURN(osd_seq);
 996 }
 997
 998 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 999                        const struct lu_fid *fid, struct osd_inode_id *id)
1000 {
1001         struct osd_obj_map *map;
1002         struct osd_obj_seq *osd_seq;
1003         struct dentry *d_seq;
1004         struct dentry *child;
1005         struct ost_id *ostid = &info->oti_ostid;
1006         int dirn;
1007         char name[32];
1008         struct ldiskfs_dir_entry_2 *de;
1009         struct buffer_head *bh;
1010         struct inode *dir;
1011         struct inode *inode;
1012
1013         ENTRY;
1014
1015         /* on the very first lookup we find and open directories */
1016         map = dev->od_ost_map;
1017         LASSERT(map);
1018         LASSERT(map->om_root);
1019
1020         fid_to_ostid(fid, ostid);
1021         osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
1022         if (IS_ERR(osd_seq))
1023                 RETURN(PTR_ERR(osd_seq));
1024
1025         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1026         d_seq = osd_seq->oos_dirs[dirn];
1027         LASSERT(d_seq);
1028
1029         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1030
1031         child = &info->oti_child_dentry;
1032         child->d_parent = d_seq;
1033         child->d_name.hash = 0;
1034         child->d_name.name = name;
1035         /* XXX: we can use rc from sprintf() instead of strlen() */
1036         child->d_name.len = strlen(name);
1037
1038         dir = d_seq->d_inode;
1039         inode_lock(dir);
1040         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
1041         inode_unlock(dir);
1042
1043         if (IS_ERR(bh))
1044                 RETURN(PTR_ERR(bh));
1045
1046         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
1047         brelse(bh);
1048
1049         inode = osd_iget(info, dev, id);
1050         if (IS_ERR(inode)) {
1051                 int rc = PTR_ERR(inode);
1052
1053                 RETURN(rc == -ENOENT ? -ESTALE : rc);
1054         }
1055
1056         iput(inode);
1057         RETURN(0);
1058 }
1059
1060 int osd_obj_map_insert(struct osd_thread_info *info,
1061                        struct osd_device *osd,
1062                        const struct lu_fid *fid,
1063                        const struct osd_inode_id *id,
1064                        handle_t *th)
1065 {
1066         struct osd_obj_map *map;
1067         struct osd_obj_seq *osd_seq;
1068         struct dentry *d;
1069         struct ost_id *ostid = &info->oti_ostid;
1070         u64 oid;
1071         int dirn, rc = 0;
1072         char name[32];
1073
1074         ENTRY;
1075
1076         map = osd->od_ost_map;
1077         LASSERT(map);
1078
1079         /* map fid to seq:objid */
1080         fid_to_ostid(fid, ostid);
1081
1082         oid = ostid_id(ostid);
1083         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1084         if (IS_ERR(osd_seq))
1085                 RETURN(PTR_ERR(osd_seq));
1086
1087         dirn = oid & (osd_seq->oos_subdir_count - 1);
1088         d = osd_seq->oos_dirs[dirn];
1089         LASSERT(d);
1090
1091         osd_oid_name(name, sizeof(name), fid, oid);
1092
1093 again:
1094         rc = osd_obj_add_entry(info, osd, d, name, id, th);
1095         if (rc == -EEXIST) {
1096                 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1097                 if (unlikely(rc == -ENOENT))
1098                         goto again;
1099
1100                 if (unlikely(rc == 1))
1101                         RETURN(0);
1102         }
1103
1104         RETURN(rc);
1105 }
1106
1107 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1108                        const struct lu_fid *fid, handle_t *th)
1109 {
1110         struct osd_obj_map *map;
1111         struct osd_obj_seq *osd_seq;
1112         struct dentry *d;
1113         struct ost_id *ostid = &info->oti_ostid;
1114         int dirn, rc = 0;
1115         char name[32];
1116
1117         ENTRY;
1118
1119         map = osd->od_ost_map;
1120         LASSERT(map);
1121
1122         /* map fid to seq:objid */
1123         fid_to_ostid(fid, ostid);
1124
1125         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1126         if (IS_ERR(osd_seq))
1127                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1128
1129         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1130         d = osd_seq->oos_dirs[dirn];
1131         LASSERT(d);
1132
1133         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1134         rc = osd_obj_del_entry(info, osd, d, name, th);
1135 cleanup:
1136         RETURN(rc);
1137 }
1138
1139 int osd_obj_map_update(struct osd_thread_info *info,
1140                        struct osd_device *osd,
1141                        const struct lu_fid *fid,
1142                        const struct osd_inode_id *id,
1143                        handle_t *th)
1144 {
1145         struct osd_obj_seq *osd_seq;
1146         struct dentry *d;
1147         struct ost_id *ostid = &info->oti_ostid;
1148         int dirn, rc = 0;
1149         char name[32];
1150
1151         ENTRY;
1152
1153         fid_to_ostid(fid, ostid);
1154         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1155         if (IS_ERR(osd_seq))
1156                 RETURN(PTR_ERR(osd_seq));
1157
1158         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1159         d = osd_seq->oos_dirs[dirn];
1160         LASSERT(d);
1161
1162         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1163         rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1164
1165         RETURN(rc);
1166 }
1167
1168 int osd_obj_map_recover(struct osd_thread_info *info,
1169                         struct osd_device *osd,
1170                         struct inode *src_parent,
1171                         struct dentry *src_child,
1172                         const struct lu_fid *fid)
1173 {
1174         struct osd_obj_seq *osd_seq;
1175         struct dentry *tgt_parent;
1176         struct dentry *tgt_child = &info->oti_child_dentry;
1177         struct inode *dir;
1178         struct inode *inode = src_child->d_inode;
1179         struct ost_id *ostid = &info->oti_ostid;
1180         handle_t *jh;
1181         struct ldiskfs_dir_entry_2 *de;
1182         struct buffer_head *bh;
1183         char name[32];
1184         int dirn;
1185         int rc = 0;
1186
1187         ENTRY;
1188
1189         if (fid_is_last_id(fid)) {
1190                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1191                 if (IS_ERR(osd_seq))
1192                         RETURN(PTR_ERR(osd_seq));
1193
1194                 tgt_parent = osd_seq->oos_root;
1195                 tgt_child->d_name.name = "LAST_ID";
1196                 tgt_child->d_name.len = strlen("LAST_ID");
1197         } else {
1198                 fid_to_ostid(fid, ostid);
1199                 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1200                 if (IS_ERR(osd_seq))
1201                         RETURN(PTR_ERR(osd_seq));
1202
1203                 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1204                 tgt_parent = osd_seq->oos_dirs[dirn];
1205                 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1206                 tgt_child->d_name.name = name;
1207                 tgt_child->d_name.len = strlen(name);
1208         }
1209         LASSERT(tgt_parent != NULL);
1210
1211         dir = tgt_parent->d_inode;
1212         tgt_child->d_name.hash = 0;
1213         tgt_child->d_parent = tgt_parent;
1214         tgt_child->d_inode = inode;
1215
1216         /* The non-initialized src_child may be destroyed. */
1217         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1218                                 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1219                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1220                                 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1221         if (IS_ERR(jh))
1222                 RETURN(PTR_ERR(jh));
1223
1224         ll_vfs_dq_init(src_parent);
1225         ll_vfs_dq_init(dir);
1226
1227         inode_lock(src_parent);
1228         inode_lock(dir);
1229         bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1230         if (!IS_ERR(bh)) {
1231                 /*
1232                  * XXX: If some other object occupied the same slot. And If such
1233                  *      inode is zero-sized and with SUID+SGID, then means it is
1234                  *      a new created one. Maybe we can remove it and insert the
1235                  *      original one back to the /O/<seq>/d<x>. But there are
1236                  *      something to be considered:
1237                  *
1238                  *      1) The OST-object under /lost+found has crashed LMA.
1239                  *         So it should not conflict with the current one.
1240                  *
1241                  *      2) There are race conditions that: someone may just want
1242                  *         to modify the current one. Even if the OI scrub takes
1243                  *         the object lock when remove the current one, it still
1244                  *         cause the modification to be lost becasue the target
1245                  *         has been removed when the RPC service thread waiting
1246                  *         for the lock.
1247                  *
1248                  *      So keep it there before we have suitable solution.
1249                  */
1250                 brelse(bh);
1251                 inode_unlock(dir);
1252                 inode_unlock(src_parent);
1253                 ldiskfs_journal_stop(jh);
1254
1255                 rc = -EEXIST;
1256                 /* If the src object has never been modified, then remove it. */
1257                 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1258                     inode->i_mode & S_ISGID) {
1259                         rc = ll_vfs_unlink(src_parent, src_child);
1260                         if (unlikely(rc == -ENOENT))
1261                                 rc = 0;
1262                 }
1263                 RETURN(rc);
1264         }
1265
1266         bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1267                                     NULL, NULL);
1268         if (unlikely(IS_ERR(bh)))
1269                 GOTO(unlock, rc = PTR_ERR(bh));
1270
1271         rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1272         brelse(bh);
1273         if (rc != 0)
1274                 GOTO(unlock, rc);
1275
1276         rc = osd_ldiskfs_add_entry(info, osd, jh, tgt_child, inode, NULL);
1277
1278         GOTO(unlock, rc);
1279
1280 unlock:
1281         inode_unlock(dir);
1282         inode_unlock(src_parent);
1283         ldiskfs_journal_stop(jh);
1284         return rc;
1285 }
1286
1287 static struct dentry *
1288 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1289                      const struct lu_fid *fid, char **name)
1290 {
1291         struct dentry *root = ERR_PTR(-ENOENT);
1292
1293         if (fid_is_last_id(fid)) {
1294                 struct osd_obj_seq *osd_seq;
1295
1296                 /* on creation of LAST_ID we create O/<seq> hierarchy */
1297                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1298                 if (IS_ERR(osd_seq))
1299                         RETURN((struct dentry *)osd_seq);
1300
1301                 *name = "LAST_ID";
1302                 root = osd_seq->oos_root;
1303         } else {
1304                 *name = osd_lf_fid2name(fid);
1305                 if (*name == NULL)
1306                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1307                 else if ((*name)[0])
1308                         root = osd_sb(osd)->s_root;
1309         }
1310
1311         return root;
1312 }
1313
1314 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1315                         const struct lu_fid *fid, const struct osd_inode_id *id,
1316                         handle_t *th)
1317 {
1318         struct dentry *root;
1319         char *name = NULL;
1320         int rc;
1321
1322         ENTRY;
1323
1324         root = osd_object_spec_find(info, osd, fid, &name);
1325         if (!IS_ERR(root)) {
1326                 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1327         } else {
1328                 rc = PTR_ERR(root);
1329                 if (rc == -ENOENT)
1330                         rc = 1;
1331         }
1332
1333         RETURN(rc);
1334 }
1335
1336 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1337                         const struct lu_fid *fid, const struct osd_inode_id *id,
1338                         handle_t *th)
1339 {
1340         struct dentry *root;
1341         char *name = NULL;
1342         int rc;
1343
1344         ENTRY;
1345
1346         root = osd_object_spec_find(info, osd, fid, &name);
1347         if (!IS_ERR(root)) {
1348                 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1349         } else {
1350                 rc = PTR_ERR(root);
1351                 if (rc == -ENOENT)
1352                         rc = 0;
1353         }
1354
1355         RETURN(rc);
1356 }
1357
1358 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1359                         const struct lu_fid *fid, struct osd_inode_id *id)
1360 {
1361         struct dentry *root;
1362         struct dentry *dentry;
1363         struct inode *inode;
1364         char *name = NULL;
1365         int rc = -ENOENT;
1366
1367         ENTRY;
1368
1369         if (fid_is_last_id(fid)) {
1370                 struct osd_obj_seq *osd_seq;
1371
1372                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1373                 if (IS_ERR(osd_seq))
1374                         RETURN(PTR_ERR(osd_seq));
1375                 root = osd_seq->oos_root;
1376                 name = "LAST_ID";
1377         } else {
1378                 root = osd_sb(osd)->s_root;
1379                 name = osd_lf_fid2name(fid);
1380                 if (name == NULL || strlen(name) == 0)
1381                         RETURN(-ENOENT);
1382         }
1383
1384         dentry = ll_lookup_one_len(name, root, strlen(name));
1385         if (!IS_ERR(dentry)) {
1386                 inode = dentry->d_inode;
1387                 if (inode) {
1388                         if (is_bad_inode(inode)) {
1389                                 rc = -EIO;
1390                         } else {
1391                                 osd_id_gen(id, inode->i_ino,
1392                                            inode->i_generation);
1393                                 rc = 0;
1394                         }
1395                 }
1396                 /*
1397                  * if dentry is accessible after osd_compat_spec_insert it
1398                  * will still contain NULL inode, so don't keep it in cache
1399                  */
1400                 d_invalidate(dentry);
1401                 dput(dentry);
1402         }
1403
1404         RETURN(rc);
1405 }
1406
1407 #ifndef HAVE_BIO_INTEGRITY_ENABLED
1408 bool bio_integrity_enabled(struct bio *bio)
1409 {
1410         struct blk_integrity *bi = blk_get_integrity(bio_get_disk(bio));
1411
1412         if (bio_op(bio) != REQ_OP_READ && bio_op(bio) != REQ_OP_WRITE)
1413                 return false;
1414
1415         if (!bio_sectors(bio))
1416                 return false;
1417
1418          /* Already protected? */
1419         if (bio_integrity(bio))
1420                 return false;
1421
1422         if (bi == NULL)
1423                 return false;
1424
1425         if (bio_data_dir(bio) == READ && bi->profile->verify_fn != NULL &&
1426             (bi->flags & BLK_INTEGRITY_VERIFY))
1427                 return true;
1428
1429         if (bio_data_dir(bio) == WRITE && bi->profile->generate_fn != NULL &&
1430             (bi->flags & BLK_INTEGRITY_GENERATE))
1431                 return true;
1432
1433         return false;
1434 }
1435 #endif