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 = KERNEL_DS;
  63         newctxt->umask = current_umask();
  64         newctxt->dt = NULL;
  65
  66         push_ctxt(save, newctxt);
  67 }
  68
  69 struct dentry *osd_lookup_one_len_common(struct osd_device *dev,
  70                                          const char *name,
  71                                          struct dentry *base, int len,
  72                                          enum oi_check_flags flags)
  73 {
  74         struct dentry *dchild;
  75
  76         /*
  77          * We can't use inode_is_locked() directly since we can't know
  78          * if the current thread context took the lock earlier or if
  79          * another thread context took the lock. OI_LOCKED tells us
  80          * if the current thread context has already taken the lock.
  81          */
  82         if (!(flags & OI_LOCKED)) {
  83                 /* If another thread took this lock already we will
  84                  * just have to wait until the other thread is done.
  85                  */
  86                 inode_lock(base->d_inode);
  87                 dchild = lookup_one_len(name, base, len);
  88                 inode_unlock(base->d_inode);
  89         } else {
  90                 /* This thread context already has taken the lock.
  91                  * Other threads will have to wait until we are done.
  92                  */
  93                 dchild = lookup_one_len(name, base, len);
  94         }
  95         if (IS_ERR(dchild))
  96                 return dchild;
  97
  98         if (dchild->d_inode && unlikely(is_bad_inode(dchild->d_inode))) {
  99                 CERROR("%s: bad inode returned %lu/%u: rc = -ENOENT\n",
 100                        osd_name(dev), dchild->d_inode->i_ino,
 101                        dchild->d_inode->i_generation);
 102                 dput(dchild);
 103                 dchild = ERR_PTR(-ENOENT);
 104         }
 105
 106         return dchild;
 107 }
 108
 109 /**
 110  * osd_lookup_one_len_unlocked
 111  *
 112  * @dev:        obd device we are searching
 113  * @name:       pathname component to lookup
 114  * @base:       base directory to lookup from
 115  * @len:        maximum length @len should be interpreted to
 116  *
 117  * Unlike osd_lookup_one_len, this should be called without the parent
 118  * i_mutex held, and will take the i_mutex itself.
 119  */
 120 struct dentry *osd_lookup_one_len_unlocked(struct osd_device *dev,
 121                                            const char *name,
 122                                            struct dentry *base, int len)
 123 {
 124         return osd_lookup_one_len_common(dev, name, base, len, ~OI_LOCKED);
 125 }
 126
 127 /**
 128  * osd_lookup_one_len - lookup single pathname component
 129  *
 130  * @dev:        obd device we are searching
 131  * @name:       pathname component to lookup
 132  * @base:       base directory to lookup from
 133  * @len:        maximum length @len should be interpreted to
 134  *
 135  * The caller must hold inode lock
 136  */
 137 struct dentry *osd_lookup_one_len(struct osd_device *dev, const char *name,
 138                                   struct dentry *base, int len)
 139 {
 140         return osd_lookup_one_len_common(dev, name, base, len, OI_LOCKED);
 141 }
 142
 143 /* utility to make a directory */
 144 static struct dentry *
 145 simple_mkdir(const struct lu_env *env, struct osd_device *osd,
 146              struct dentry *dir, const struct lu_fid *fid,
 147              const char *name, __u32 compat, int mode, bool *created)
 148 {
 149         struct osd_thread_info *info = osd_oti_get(env);
 150         struct lu_fid *tfid = &info->oti_fid3;
 151         struct inode *inode;
 152         struct dentry *dchild;
 153         int err = 0;
 154
 155         ENTRY;
 156
 157         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 158         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
 159         dchild = osd_lookup_one_len_unlocked(osd, name, dir, strlen(name));
 160         if (IS_ERR(dchild))
 161                 RETURN(dchild);
 162
 163         inode = dchild->d_inode;
 164         if (inode) {
 165                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 166                 int old_mode = inode->i_mode;
 167
 168                 if (created)
 169                         *created = false;
 170
 171                 if (!S_ISDIR(old_mode)) {
 172                         CERROR("found %s (%lu/%u) is mode %o\n", name,
 173                                inode->i_ino, inode->i_generation, old_mode);
 174                         GOTO(out_err, err = -ENOTDIR);
 175                 }
 176
 177                 if (unlikely(osd->od_dt_dev.dd_rdonly))
 178                         RETURN(dchild);
 179
 180                 /* Fixup directory permissions if necessary */
 181                 if ((old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 182                         CDEBUG(D_CONFIG,
 183                                "fixing permissions on %s from %o to %o\n",
 184                                name, old_mode, mode);
 185                         inode->i_mode = (mode & S_IALLUGO) |
 186                                         (old_mode & ~S_IALLUGO);
 187                         mark_inode_dirty(inode);
 188                 }
 189
 190                 err = osd_get_lma(info, inode, &info->oti_obj_dentry,
 191                                   &info->oti_ost_attrs);
 192                 if (err == -ENODATA)
 193                         goto set_fid;
 194
 195                 if (err)
 196                         GOTO(out_err, err);
 197
 198                 if ((fid && !lu_fid_eq(fid, &lma->lma_self_fid)) ||
 199                     lma->lma_compat != compat)
 200                         goto set_fid;
 201
 202                 RETURN(dchild);
 203         }
 204
 205         err = vfs_mkdir(dir->d_inode, dchild, mode);
 206         if (err)
 207                 GOTO(out_err, err);
 208
 209         inode = dchild->d_inode;
 210         if (created)
 211                 *created = true;
 212
 213 set_fid:
 214         if (fid)
 215                 *tfid = *fid;
 216         else
 217                 lu_igif_build(tfid, inode->i_ino, inode->i_generation);
 218         err = osd_ea_fid_set(info, inode, tfid, compat, 0);
 219         if (err)
 220                 GOTO(out_err, err);
 221
 222         RETURN(dchild);
 223
 224 out_err:
 225         dput(dchild);
 226         return ERR_PTR(err);
 227 }
 228
 229 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
 230 {
 231         struct lr_server_data lsd;
 232         struct dentry *dlast;
 233         loff_t off;
 234         int rc = 0;
 235         int count = OBJ_SUBDIR_COUNT;
 236
 237         ENTRY;
 238
 239         dlast = osd_lookup_one_len_unlocked(osd, LAST_RCVD, osd_sb(osd)->s_root,
 240                                             strlen(LAST_RCVD));
 241         if (IS_ERR(dlast))
 242                 return PTR_ERR(dlast);
 243         else if (dlast->d_inode == NULL)
 244                 goto out;
 245
 246         off = 0;
 247         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 248         if (rc == sizeof(lsd)) {
 249                 CDEBUG(D_INFO,
 250                       "read last_rcvd header, uuid = %s, subdir count = %d\n",
 251                       lsd.lsd_uuid, lsd.lsd_subdir_count);
 252                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 253                         count = le16_to_cpu(lsd.lsd_subdir_count);
 254         } else if (rc != 0) {
 255                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 256                 if (rc > 0)
 257                         rc = -EFAULT;
 258                 dput(dlast);
 259                 return rc;
 260         }
 261 out:
 262         dput(dlast);
 263         LASSERT(count > 0);
 264         return count;
 265 }
 266
 267 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
 268 {
 269         struct lvfs_run_ctxt new;
 270         struct lvfs_run_ctxt save;
 271         struct dentry *parent;
 272         struct osd_mdobj_map *omm;
 273         struct dentry *d;
 274         struct osd_thread_info *info = osd_oti_get(env);
 275         struct lu_fid *fid = &info->oti_fid3;
 276         int rc = 0;
 277
 278         ENTRY;
 279
 280         OBD_ALLOC_PTR(dev->od_mdt_map);
 281         if (dev->od_mdt_map == NULL)
 282                 RETURN(-ENOMEM);
 283
 284         omm = dev->od_mdt_map;
 285
 286         parent = osd_sb(dev)->s_root;
 287         osd_push_ctxt(dev, &new, &save);
 288
 289         lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
 290         d = simple_mkdir(env, dev, parent, fid, REMOTE_PARENT_DIR,
 291                          LMAC_NOT_IN_OI, 0755, NULL);
 292         if (IS_ERR(d))
 293                 GOTO(cleanup, rc = PTR_ERR(d));
 294
 295         omm->omm_remote_parent = d;
 296
 297         GOTO(cleanup, rc = 0);
 298
 299 cleanup:
 300         pop_ctxt(&save, &new);
 301         if (rc) {
 302                 if (omm->omm_remote_parent != NULL)
 303                         dput(omm->omm_remote_parent);
 304                 OBD_FREE_PTR(omm);
 305                 dev->od_mdt_map = NULL;
 306         }
 307         return rc;
 308 }
 309
 310 static void osd_mdt_fini(struct osd_device *osd)
 311 {
 312         struct osd_mdobj_map *omm = osd->od_mdt_map;
 313
 314         if (omm == NULL)
 315                 return;
 316
 317         if (omm->omm_remote_parent)
 318                 dput(omm->omm_remote_parent);
 319
 320         OBD_FREE_PTR(omm);
 321         osd->od_ost_map = NULL;
 322 }
 323
 324 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
 325                              struct osd_object *obj, struct osd_thandle *oh)
 326 {
 327         struct osd_mdobj_map *omm = osd->od_mdt_map;
 328         struct osd_thread_info *oti = osd_oti_get(env);
 329         struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
 330         char *name = oti->oti_name;
 331         struct osd_thread_info *info = osd_oti_get(env);
 332         struct dentry *dentry;
 333         struct dentry *parent;
 334         int rc;
 335
 336         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AGENTENT))
 337                 RETURN(0);
 338
 339         /*
 340          * Set REMOTE_PARENT in lma, so other process like unlink or lfsck
 341          * can identify this object quickly
 342          */
 343         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
 344                          &oti->oti_ost_attrs);
 345         if (rc)
 346                 RETURN(rc);
 347
 348         lma->lma_incompat |= LMAI_REMOTE_PARENT;
 349         lustre_lma_swab(lma);
 350         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 351                              sizeof(*lma), XATTR_REPLACE);
 352         if (rc)
 353                 RETURN(rc);
 354
 355         parent = omm->omm_remote_parent;
 356         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 357         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 358                                            name, strlen(name));
 359         inode_lock(parent->d_inode);
 360         rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
 361                                    obj->oo_inode, NULL);
 362         if (!rc && S_ISDIR(obj->oo_inode->i_mode))
 363                 ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
 364         else if (unlikely(rc == -EEXIST))
 365                 rc = 0;
 366         if (!rc)
 367                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 368         CDEBUG(D_INODE, "%s: create agent entry for %s: rc = %d\n",
 369                osd_name(osd), name, rc);
 370         mark_inode_dirty(parent->d_inode);
 371         inode_unlock(parent->d_inode);
 372         RETURN(rc);
 373 }
 374
 375 int osd_delete_from_remote_parent(const struct lu_env *env,
 376                                   struct osd_device *osd,
 377                                   struct osd_object *obj,
 378                                   struct osd_thandle *oh, bool destroy)
 379 {
 380         struct osd_mdobj_map *omm = osd->od_mdt_map;
 381         struct osd_thread_info *oti = osd_oti_get(env);
 382         struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
 383         char *name = oti->oti_name;
 384         struct dentry *dentry;
 385         struct dentry *parent;
 386         struct ldiskfs_dir_entry_2 *de;
 387         struct buffer_head *bh;
 388         int rc;
 389
 390         parent = omm->omm_remote_parent;
 391         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 392         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 393                                            name, strlen(name));
 394         inode_lock(parent->d_inode);
 395         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 396                                     NULL, NULL);
 397         if (IS_ERR(bh)) {
 398                 inode_unlock(parent->d_inode);
 399                 rc = PTR_ERR(bh);
 400                 if (unlikely(rc == -ENOENT))
 401                         rc = 0;
 402         } else {
 403                 rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode,
 404                                           de, bh);
 405                 if (!rc && S_ISDIR(obj->oo_inode->i_mode))
 406                         ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
 407                 mark_inode_dirty(parent->d_inode);
 408                 inode_unlock(parent->d_inode);
 409                 brelse(bh);
 410                 CDEBUG(D_INODE, "%s: remove agent entry for %s: rc = %d\n",
 411                        osd_name(osd), name, rc);
 412         }
 413
 414         if (destroy || rc) {
 415                 if (!rc)
 416                         lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
 417
 418                 RETURN(rc);
 419         }
 420
 421         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
 422                          &oti->oti_ost_attrs);
 423         if (rc)
 424                 RETURN(rc);
 425
 426         /* Get rid of REMOTE_PARENT flag from incompat */
 427         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
 428         lustre_lma_swab(lma);
 429         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 430                              sizeof(*lma), XATTR_REPLACE);
 431         if (!rc)
 432                 lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
 433         RETURN(rc);
 434 }
 435
 436 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 437                                 struct osd_device *osd,
 438                                 const struct lu_fid *fid,
 439                                 struct osd_inode_id *id)
 440 {
 441         struct osd_mdobj_map *omm = osd->od_mdt_map;
 442         char *name = oti->oti_name;
 443         struct dentry *parent;
 444         struct dentry *dentry;
 445         struct ldiskfs_dir_entry_2 *de;
 446         struct buffer_head *bh;
 447         int rc;
 448
 449         ENTRY;
 450
 451         if (unlikely(osd->od_is_ost))
 452                 RETURN(-ENOENT);
 453
 454         parent = omm->omm_remote_parent;
 455         sprintf(name, DFID_NOBRACE, PFID(fid));
 456         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
 457                                            name, strlen(name));
 458         inode_lock(parent->d_inode);
 459         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
 460                                     NULL, NULL);
 461         if (IS_ERR(bh)) {
 462                 rc = PTR_ERR(bh);
 463         } else {
 464                 struct inode *inode;
 465
 466                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 467                 brelse(bh);
 468                 inode = osd_iget(oti, osd, id);
 469                 if (IS_ERR(inode)) {
 470                         rc = PTR_ERR(inode);
 471                         if (rc == -ESTALE)
 472                                 rc = -ENOENT;
 473                 } else {
 474                         iput(inode);
 475                         rc = 0;
 476                 }
 477         }
 478         inode_unlock(parent->d_inode);
 479         if (rc == 0)
 480                 osd_add_oi_cache(oti, osd, id, fid);
 481         RETURN(rc);
 482 }
 483
 484 /*
 485  * directory structure on legacy OST:
 486  *
 487  * O/<seq>/d0-31/<objid>
 488  * O/<seq>/LAST_ID
 489  * last_rcvd
 490  * LAST_GROUP
 491  * CONFIGS
 492  *
 493  */
 494 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
 495 {
 496         struct lvfs_run_ctxt new;
 497         struct lvfs_run_ctxt save;
 498         struct dentry *d;
 499         int rc;
 500         bool created = false;
 501
 502         ENTRY;
 503
 504         OBD_ALLOC_PTR(dev->od_ost_map);
 505         if (dev->od_ost_map == NULL)
 506                 RETURN(-ENOMEM);
 507
 508         /* to get subdir count from last_rcvd */
 509         rc = osd_last_rcvd_subdir_count(dev);
 510         if (rc < 0)
 511                 GOTO(cleanup_alloc, rc);
 512
 513         dev->od_ost_map->om_subdir_count = rc;
 514         INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 515         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 516         mutex_init(&dev->od_ost_map->om_dir_init_mutex);
 517
 518         osd_push_ctxt(dev, &new, &save);
 519         d = simple_mkdir(env, dev, osd_sb(dev)->s_root, NULL, "O",
 520                          LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, &created);
 521         if (IS_ERR(d))
 522                 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
 523
 524         if (created)
 525                 /* It is quite probably that the device is new formatted. */
 526                 dev->od_maybe_new = 1;
 527
 528         dev->od_ost_map->om_root = d;
 529
 530         pop_ctxt(&save, &new);
 531         RETURN(0);
 532
 533 cleanup_ctxt:
 534         pop_ctxt(&save, &new);
 535 cleanup_alloc:
 536         OBD_FREE_PTR(dev->od_ost_map);
 537         return rc;
 538 }
 539
 540 static void osd_seq_free(struct osd_obj_seq *osd_seq)
 541 {
 542         int j;
 543
 544         if (osd_seq->oos_dirs) {
 545                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 546                         if (osd_seq->oos_dirs[j])
 547                                 dput(osd_seq->oos_dirs[j]);
 548                 }
 549                 OBD_FREE(osd_seq->oos_dirs,
 550                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 551         }
 552
 553         if (osd_seq->oos_root)
 554                 dput(osd_seq->oos_root);
 555
 556         OBD_FREE_PTR(osd_seq);
 557 }
 558
 559 static void osd_ost_fini(struct osd_device *osd)
 560 {
 561         struct osd_obj_seq *osd_seq;
 562         struct osd_obj_seq *tmp;
 563         struct osd_obj_map *map = osd->od_ost_map;
 564
 565         ENTRY;
 566
 567         if (map == NULL)
 568                 return;
 569
 570         write_lock(&map->om_seq_list_lock);
 571         list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
 572                                  oos_seq_list) {
 573                 list_del_init(&osd_seq->oos_seq_list);
 574                 write_unlock(&map->om_seq_list_lock);
 575                 osd_seq_free(osd_seq);
 576                 write_lock(&map->om_seq_list_lock);
 577         }
 578         write_unlock(&map->om_seq_list_lock);
 579         if (map->om_root)
 580                 dput(map->om_root);
 581         OBD_FREE_PTR(map);
 582         osd->od_ost_map = NULL;
 583         EXIT;
 584 }
 585
 586 static int osd_index_backup_dir_init(const struct lu_env *env,
 587                                      struct osd_device *dev)
 588 {
 589         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
 590         struct lvfs_run_ctxt new;
 591         struct lvfs_run_ctxt save;
 592         struct dentry *dentry;
 593         int rc = 0;
 594
 595         ENTRY;
 596
 597         lu_local_obj_fid(fid, INDEX_BACKUP_OID);
 598         osd_push_ctxt(dev, &new, &save);
 599         dentry = simple_mkdir(env, dev, osd_sb(dev)->s_root, fid,
 600                               INDEX_BACKUP_DIR, LMAC_NOT_IN_OI, 0755, NULL);
 601         if (IS_ERR(dentry)) {
 602                 rc = PTR_ERR(dentry);
 603         } else {
 604                 dev->od_index_backup_inode = igrab(dentry->d_inode);
 605                 dput(dentry);
 606         }
 607         pop_ctxt(&save, &new);
 608
 609         RETURN(rc);
 610 }
 611
 612 static void osd_index_backup_dir_fini(struct osd_device *dev)
 613 {
 614         if (dev->od_index_backup_inode) {
 615                 iput(dev->od_index_backup_inode);
 616                 dev->od_index_backup_inode = NULL;
 617         }
 618 }
 619
 620 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 621 {
 622         int rc;
 623         bool mdt_init = false;
 624
 625         ENTRY;
 626
 627         rc = osd_ost_init(env, dev);
 628         if (rc)
 629                 RETURN(rc);
 630
 631         if (!dev->od_is_ost) {
 632                 rc = osd_mdt_init(env, dev);
 633                 if (rc) {
 634                         osd_ost_fini(dev);
 635                         RETURN(rc);
 636                 }
 637
 638                 mdt_init = true;
 639         }
 640
 641         rc = osd_index_backup_dir_init(env, dev);
 642         if (rc) {
 643                 osd_ost_fini(dev);
 644                 if (mdt_init)
 645                         osd_mdt_fini(dev);
 646         }
 647
 648         RETURN(rc);
 649 }
 650
 651 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
 652 {
 653         struct osd_obj_seq *osd_seq;
 654
 655         list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 656                 if (osd_seq->oos_seq == seq)
 657                         return osd_seq;
 658         }
 659         return NULL;
 660 }
 661
 662 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
 663 {
 664         struct osd_obj_seq *osd_seq;
 665
 666         read_lock(&map->om_seq_list_lock);
 667         osd_seq = osd_seq_find_locked(map, seq);
 668         read_unlock(&map->om_seq_list_lock);
 669         return osd_seq;
 670 }
 671
 672 void osd_obj_map_fini(struct osd_device *dev)
 673 {
 674         osd_index_backup_dir_fini(dev);
 675         osd_ost_fini(dev);
 676         osd_mdt_fini(dev);
 677 }
 678
 679 /**
 680  * Update the specified OI mapping.
 681  *
 682  * \retval   1, changed nothing
 683  * \retval   0, changed successfully
 684  * \retval -ve, on error
 685  */
 686 static int osd_obj_update_entry(struct osd_thread_info *info,
 687                                 struct osd_device *osd,
 688                                 struct dentry *dir, const char *name,
 689                                 const struct lu_fid *fid,
 690                                 const struct osd_inode_id *id,
 691                                 handle_t *th)
 692 {
 693         struct inode *parent = dir->d_inode;
 694         struct dentry *child;
 695         struct ldiskfs_dir_entry_2 *de;
 696         struct buffer_head *bh;
 697         struct inode *inode;
 698         struct dentry *dentry = &info->oti_obj_dentry;
 699         struct osd_inode_id *oi_id = &info->oti_id3;
 700         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 701         struct lu_fid *oi_fid = &lma->lma_self_fid;
 702         int rc;
 703
 704         ENTRY;
 705
 706         LASSERT(th != NULL);
 707         LASSERT(th->h_transaction != NULL);
 708
 709         child = &info->oti_child_dentry;
 710         child->d_parent = dir;
 711         child->d_name.hash = 0;
 712         child->d_name.name = name;
 713         child->d_name.len = strlen(name);
 714
 715         dquot_initialize(parent);
 716         inode_lock(parent);
 717         bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
 718         if (IS_ERR(bh))
 719                 GOTO(out, rc = PTR_ERR(bh));
 720
 721         if (le32_to_cpu(de->inode) == id->oii_ino)
 722                 GOTO(out, rc = 1);
 723
 724         osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 725         inode = osd_iget(info, osd, oi_id);
 726         if (IS_ERR(inode)) {
 727                 rc = PTR_ERR(inode);
 728                 if (rc == -ENOENT || rc == -ESTALE)
 729                         goto update;
 730                 GOTO(out, rc);
 731         }
 732
 733         /*
 734          * The EA inode should NOT be in OI, old OI scrub may added
 735          * such OI mapping by wrong, replace it.
 736          */
 737         if (unlikely(osd_is_ea_inode(inode))) {
 738                 iput(inode);
 739                 goto update;
 740         }
 741
 742         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
 743         if (rc == -ENODATA) {
 744                 rc = osd_get_idif(info, inode, dentry, oi_fid);
 745                 if (rc > 0 || rc == -ENODATA) {
 746                         oi_fid = NULL;
 747                         rc = 0;
 748                 }
 749         }
 750         iput(inode);
 751
 752         if (rc != 0)
 753                 GOTO(out, rc);
 754
 755         /*
 756          * If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
 757          * either a crashed object or a uninitialized one. Replace it.
 758          */
 759         if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
 760                 CERROR("%s: the FID "DFID" is used by two objects: "
 761                        "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
 762                        oi_id->oii_ino, oi_id->oii_gen,
 763                        id->oii_ino, id->oii_gen);
 764                 GOTO(out, rc = -EEXIST);
 765         }
 766
 767         if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
 768                 __u32 idx1 = fid_idif_ost_idx(fid);
 769                 __u32 idx2 = fid_idif_ost_idx(oi_fid);
 770                 struct ost_id *ostid = &info->oti_ostid;
 771                 struct lu_fid *tfid = &info->oti_fid3;
 772
 773                 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
 774                          "invalid given FID "DFID", not match the "
 775                          "device index %u\n", PFID(fid), osd->od_index);
 776
 777                 if (idx1 != idx2) {
 778                         if (idx1 == 0 && idx2 == osd->od_index) {
 779                                 fid_to_ostid(fid, ostid);
 780                                 ostid_to_fid(tfid, ostid, idx2);
 781                                 if (lu_fid_eq(tfid, oi_fid)) {
 782                                         CERROR("%s: the FID "DFID" is used by "
 783                                                "two objects(2): %u/%u %u/%u\n",
 784                                                osd_name(osd), PFID(fid),
 785                                                oi_id->oii_ino, oi_id->oii_gen,
 786                                                id->oii_ino, id->oii_gen);
 787
 788                                         GOTO(out, rc = -EEXIST);
 789                                 }
 790                         } else if (idx2 == 0 && idx1 == osd->od_index) {
 791                                 fid_to_ostid(oi_fid, ostid);
 792                                 ostid_to_fid(tfid, ostid, idx1);
 793                                 if (lu_fid_eq(tfid, fid)) {
 794                                         CERROR("%s: the FID "DFID" is used by "
 795                                                "two objects(2): %u/%u %u/%u\n",
 796                                                osd_name(osd), PFID(fid),
 797                                                oi_id->oii_ino, oi_id->oii_gen,
 798                                                id->oii_ino, id->oii_gen);
 799
 800                                         GOTO(out, rc = -EEXIST);
 801                                 }
 802                         }
 803                 }
 804         }
 805
 806 update:
 807         /*
 808          * There may be temporary inconsistency: On one hand, the new
 809          * object may be referenced by multiple entries, which is out
 810          * of our control unless we traverse the whole /O completely,
 811          * which is non-flat order and inefficient, should be avoided;
 812          * On the other hand, the old object may become orphan if it
 813          * is still valid. Since it was referenced by an invalid entry,
 814          * making it as invisible temporary may be not worse. OI scrub
 815          * will process it later.
 816          */
 817         rc = ldiskfs_journal_get_write_access(th, bh);
 818         if (rc != 0)
 819                 GOTO(out, rc);
 820
 821         de->inode = cpu_to_le32(id->oii_ino);
 822         rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
 823
 824         GOTO(out, rc);
 825
 826 out:
 827         if (!IS_ERR(bh))
 828                 brelse(bh);
 829         inode_unlock(parent);
 830         return rc;
 831 }
 832
 833 static int osd_obj_del_entry(struct osd_thread_info *info,
 834                              struct osd_device *osd,
 835                              struct dentry *dird, char *name,
 836                              handle_t *th)
 837 {
 838         struct ldiskfs_dir_entry_2 *de;
 839         struct buffer_head *bh;
 840         struct dentry *child;
 841         struct inode *dir = dird->d_inode;
 842         int rc;
 843
 844         ENTRY;
 845
 846         LASSERT(th != NULL);
 847         LASSERT(th->h_transaction != NULL);
 848
 849         child = &info->oti_child_dentry;
 850         child->d_name.hash = 0;
 851         child->d_name.name = name;
 852         child->d_name.len = strlen(name);
 853         child->d_parent = dird;
 854         child->d_inode = NULL;
 855
 856         dquot_initialize(dir);
 857         inode_lock(dir);
 858         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
 859         if (IS_ERR(bh)) {
 860                 rc = PTR_ERR(bh);
 861         } else {
 862                 rc = ldiskfs_delete_entry(th, dir, de, bh);
 863                 brelse(bh);
 864         }
 865         inode_unlock(dir);
 866
 867         RETURN(rc);
 868 }
 869
 870 static int osd_obj_add_entry(struct osd_thread_info *info,
 871                              struct osd_device *osd,
 872                              struct dentry *dir, char *name,
 873                              const struct osd_inode_id *id,
 874                              handle_t *th)
 875 {
 876         struct dentry *child;
 877         struct inode *inode;
 878         int rc;
 879
 880         ENTRY;
 881
 882         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
 883                 RETURN(0);
 884
 885         LASSERT(th != NULL);
 886         LASSERT(th->h_transaction != NULL);
 887
 888         inode = info->oti_inode;
 889         if (unlikely(inode == NULL)) {
 890                 struct ldiskfs_inode_info *lii;
 891
 892                 OBD_ALLOC_PTR(lii);
 893                 if (lii == NULL)
 894                         RETURN(-ENOMEM);
 895                 inode = info->oti_inode = &lii->vfs_inode;
 896         }
 897
 898         inode->i_sb = osd_sb(osd);
 899         osd_id_to_inode(inode, id);
 900         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 901
 902         child = &info->oti_child_dentry;
 903         child->d_name.hash = 0;
 904         child->d_name.name = name;
 905         child->d_name.len = strlen(name);
 906         child->d_parent = dir;
 907         child->d_inode = inode;
 908
 909         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
 910                 inode->i_ino++;
 911
 912         dquot_initialize(dir->d_inode);
 913         inode_lock(dir->d_inode);
 914         rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
 915         inode_unlock(dir->d_inode);
 916
 917         RETURN(rc);
 918 }
 919
 920 /**
 921  * Use %llu for legacy OST sequences, but use %llx for new
 922  * sequences names, so that the O/{seq}/dN/{oid} more closely
 923  * follows the DFID/PFID format. This makes it easier to map from
 924  * debug messages to objects in the future, and the legacy space
 925  * of FID_SEQ_OST_MDT0 will be unused in the future.
 926  **/
 927 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
 928 {
 929         snprintf(seq_name, name_size,
 930                  (fid_seq_is_rsvd(seq) ||
 931                   fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
 932                  fid_seq_is_idif(seq) ? 0 : seq);
 933 }
 934
 935 static inline void osd_oid_name(char *name, size_t name_size,
 936                                 const struct lu_fid *fid, u64 id)
 937 {
 938         snprintf(name, name_size,
 939                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 940                   fid_seq_is_mdt0(fid_seq(fid)) ||
 941                   fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
 942 }
 943
 944 /* external locking is required */
 945 static int osd_seq_load_locked(struct osd_thread_info *info,
 946                                struct osd_device *osd,
 947                                struct osd_obj_seq *osd_seq)
 948 {
 949         struct osd_obj_map *map = osd->od_ost_map;
 950         struct dentry *seq_dir;
 951         int rc = 0;
 952         int i;
 953         char dir_name[32];
 954
 955         ENTRY;
 956
 957         if (osd_seq->oos_root != NULL)
 958                 RETURN(0);
 959
 960         LASSERT(map);
 961         LASSERT(map->om_root);
 962
 963         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 964
 965         seq_dir = simple_mkdir(info->oti_env, osd, map->om_root, NULL, dir_name,
 966                                LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, NULL);
 967         if (IS_ERR(seq_dir))
 968                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 969         else if (seq_dir->d_inode == NULL)
 970                 GOTO(out_put, rc = -EFAULT);
 971
 972         osd_seq->oos_root = seq_dir;
 973
 974         LASSERT(osd_seq->oos_dirs == NULL);
 975         OBD_ALLOC(osd_seq->oos_dirs,
 976                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 977         if (osd_seq->oos_dirs == NULL)
 978                 GOTO(out_put, rc = -ENOMEM);
 979
 980         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 981                 struct dentry   *dir;
 982
 983                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 984                 dir = simple_mkdir(info->oti_env, osd, osd_seq->oos_root, NULL,
 985                                    dir_name, LMAC_NOT_IN_OI | LMAC_FID_ON_OST,
 986                                    0700, NULL);
 987                 if (IS_ERR(dir)) {
 988                         GOTO(out_free, rc = PTR_ERR(dir));
 989                 } else if (dir->d_inode == NULL) {
 990                         dput(dir);
 991                         GOTO(out_free, rc = -EFAULT);
 992                 }
 993
 994                 osd_seq->oos_dirs[i] = dir;
 995         }
 996
 997         if (rc != 0) {
 998 out_free:
 999                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
1000                         if (osd_seq->oos_dirs[i] != NULL)
1001                                 dput(osd_seq->oos_dirs[i]);
1002                 }
1003                 OBD_FREE(osd_seq->oos_dirs,
1004                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
1005 out_put:
1006                 dput(seq_dir);
1007                 osd_seq->oos_root = NULL;
1008         }
1009 out_err:
1010         RETURN(rc);
1011 }
1012
1013 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
1014                                         struct osd_device *osd, u64 seq)
1015 {
1016         struct osd_obj_map *map;
1017         struct osd_obj_seq *osd_seq;
1018         int rc = 0;
1019
1020         ENTRY;
1021
1022         map = osd->od_ost_map;
1023         LASSERT(map);
1024         LASSERT(map->om_root);
1025
1026         osd_seq = osd_seq_find(map, seq);
1027         if (likely(osd_seq != NULL))
1028                 RETURN(osd_seq);
1029
1030         /* Serializing init process */
1031         mutex_lock(&map->om_dir_init_mutex);
1032
1033         /* Check whether the seq has been added */
1034         read_lock(&map->om_seq_list_lock);
1035         osd_seq = osd_seq_find_locked(map, seq);
1036         if (osd_seq != NULL) {
1037                 read_unlock(&map->om_seq_list_lock);
1038                 GOTO(cleanup, rc = 0);
1039         }
1040         read_unlock(&map->om_seq_list_lock);
1041
1042         OBD_ALLOC_PTR(osd_seq);
1043         if (osd_seq == NULL)
1044                 GOTO(cleanup, rc = -ENOMEM);
1045
1046         INIT_LIST_HEAD(&osd_seq->oos_seq_list);
1047         osd_seq->oos_seq = seq;
1048         /*
1049          * Init subdir count to be 32, but each seq can have
1050          * different subdir count
1051          */
1052         osd_seq->oos_subdir_count = map->om_subdir_count;
1053         rc = osd_seq_load_locked(info, osd, osd_seq);
1054         if (rc != 0)
1055                 GOTO(cleanup, rc);
1056
1057         write_lock(&map->om_seq_list_lock);
1058         list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
1059         write_unlock(&map->om_seq_list_lock);
1060
1061 cleanup:
1062         mutex_unlock(&map->om_dir_init_mutex);
1063         if (rc != 0) {
1064                 if (osd_seq != NULL)
1065                         OBD_FREE_PTR(osd_seq);
1066                 RETURN(ERR_PTR(rc));
1067         }
1068
1069         RETURN(osd_seq);
1070 }
1071
1072 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
1073                        const struct lu_fid *fid, struct osd_inode_id *id)
1074 {
1075         struct osd_obj_map *map;
1076         struct osd_obj_seq *osd_seq;
1077         struct dentry *d_seq;
1078         struct dentry *child;
1079         struct ost_id *ostid = &info->oti_ostid;
1080         int dirn;
1081         char name[32];
1082         struct ldiskfs_dir_entry_2 *de;
1083         struct buffer_head *bh;
1084         struct inode *dir;
1085         struct inode *inode;
1086
1087         ENTRY;
1088
1089         /* on the very first lookup we find and open directories */
1090         map = dev->od_ost_map;
1091         LASSERT(map);
1092         LASSERT(map->om_root);
1093
1094         fid_to_ostid(fid, ostid);
1095         osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
1096         if (IS_ERR(osd_seq))
1097                 RETURN(PTR_ERR(osd_seq));
1098
1099         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1100         d_seq = osd_seq->oos_dirs[dirn];
1101         LASSERT(d_seq);
1102
1103         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1104
1105         child = &info->oti_child_dentry;
1106         child->d_parent = d_seq;
1107         child->d_name.hash = 0;
1108         child->d_name.name = name;
1109         /* XXX: we can use rc from sprintf() instead of strlen() */
1110         child->d_name.len = strlen(name);
1111
1112         dir = d_seq->d_inode;
1113         inode_lock(dir);
1114         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
1115         inode_unlock(dir);
1116
1117         if (IS_ERR(bh))
1118                 RETURN(PTR_ERR(bh));
1119
1120         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
1121         brelse(bh);
1122
1123         inode = osd_iget(info, dev, id);
1124         if (IS_ERR(inode)) {
1125                 int rc = PTR_ERR(inode);
1126
1127                 RETURN(rc == -ENOENT ? -ESTALE : rc);
1128         }
1129
1130         iput(inode);
1131         RETURN(0);
1132 }
1133
1134 int osd_obj_map_insert(struct osd_thread_info *info,
1135                        struct osd_device *osd,
1136                        const struct lu_fid *fid,
1137                        const struct osd_inode_id *id,
1138                        handle_t *th)
1139 {
1140         struct osd_obj_map *map;
1141         struct osd_obj_seq *osd_seq;
1142         struct dentry *d;
1143         struct ost_id *ostid = &info->oti_ostid;
1144         u64 oid;
1145         int dirn, rc = 0;
1146         char name[32];
1147
1148         ENTRY;
1149
1150         map = osd->od_ost_map;
1151         LASSERT(map);
1152
1153         /* map fid to seq:objid */
1154         fid_to_ostid(fid, ostid);
1155
1156         oid = ostid_id(ostid);
1157         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1158         if (IS_ERR(osd_seq))
1159                 RETURN(PTR_ERR(osd_seq));
1160
1161         dirn = oid & (osd_seq->oos_subdir_count - 1);
1162         d = osd_seq->oos_dirs[dirn];
1163         LASSERT(d);
1164
1165         osd_oid_name(name, sizeof(name), fid, oid);
1166
1167 again:
1168         rc = osd_obj_add_entry(info, osd, d, name, id, th);
1169         if (rc == -EEXIST) {
1170                 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1171                 if (unlikely(rc == -ENOENT))
1172                         goto again;
1173
1174                 if (unlikely(rc == 1))
1175                         RETURN(0);
1176         }
1177
1178         RETURN(rc);
1179 }
1180
1181 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1182                        const struct lu_fid *fid, handle_t *th)
1183 {
1184         struct osd_obj_map *map;
1185         struct osd_obj_seq *osd_seq;
1186         struct dentry *d;
1187         struct ost_id *ostid = &info->oti_ostid;
1188         int dirn, rc = 0;
1189         char name[32];
1190
1191         ENTRY;
1192
1193         map = osd->od_ost_map;
1194         LASSERT(map);
1195
1196         /* map fid to seq:objid */
1197         fid_to_ostid(fid, ostid);
1198
1199         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1200         if (IS_ERR(osd_seq))
1201                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1202
1203         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1204         d = osd_seq->oos_dirs[dirn];
1205         LASSERT(d);
1206
1207         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1208         rc = osd_obj_del_entry(info, osd, d, name, th);
1209 cleanup:
1210         RETURN(rc);
1211 }
1212
1213 int osd_obj_map_update(struct osd_thread_info *info,
1214                        struct osd_device *osd,
1215                        const struct lu_fid *fid,
1216                        const struct osd_inode_id *id,
1217                        handle_t *th)
1218 {
1219         struct osd_obj_seq *osd_seq;
1220         struct dentry *d;
1221         struct ost_id *ostid = &info->oti_ostid;
1222         int dirn, rc = 0;
1223         char name[32];
1224
1225         ENTRY;
1226
1227         fid_to_ostid(fid, ostid);
1228         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1229         if (IS_ERR(osd_seq))
1230                 RETURN(PTR_ERR(osd_seq));
1231
1232         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1233         d = osd_seq->oos_dirs[dirn];
1234         LASSERT(d);
1235
1236         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1237         rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1238
1239         RETURN(rc);
1240 }
1241
1242 int osd_obj_map_recover(struct osd_thread_info *info,
1243                         struct osd_device *osd,
1244                         struct inode *src_parent,
1245                         struct dentry *src_child,
1246                         const struct lu_fid *fid)
1247 {
1248         struct osd_obj_seq *osd_seq;
1249         struct dentry *tgt_parent;
1250         struct dentry *tgt_child = &info->oti_child_dentry;
1251         struct inode *dir;
1252         struct inode *inode = src_child->d_inode;
1253         struct ost_id *ostid = &info->oti_ostid;
1254         handle_t *jh;
1255         struct ldiskfs_dir_entry_2 *de;
1256         struct buffer_head *bh;
1257         char name[32];
1258         int dirn;
1259         int rc = 0;
1260
1261         ENTRY;
1262
1263         if (fid_is_last_id(fid)) {
1264                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1265                 if (IS_ERR(osd_seq))
1266                         RETURN(PTR_ERR(osd_seq));
1267
1268                 tgt_parent = osd_seq->oos_root;
1269                 tgt_child->d_name.name = "LAST_ID";
1270                 tgt_child->d_name.len = strlen("LAST_ID");
1271         } else {
1272                 fid_to_ostid(fid, ostid);
1273                 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1274                 if (IS_ERR(osd_seq))
1275                         RETURN(PTR_ERR(osd_seq));
1276
1277                 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1278                 tgt_parent = osd_seq->oos_dirs[dirn];
1279                 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1280                 tgt_child->d_name.name = name;
1281                 tgt_child->d_name.len = strlen(name);
1282         }
1283         LASSERT(tgt_parent != NULL);
1284
1285         dir = tgt_parent->d_inode;
1286         tgt_child->d_name.hash = 0;
1287         tgt_child->d_parent = tgt_parent;
1288         tgt_child->d_inode = inode;
1289
1290         /* The non-initialized src_child may be destroyed. */
1291         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1292                                 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1293                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1294                                 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1295         if (IS_ERR(jh))
1296                 RETURN(PTR_ERR(jh));
1297
1298         dquot_initialize(src_parent);
1299         dquot_initialize(dir);
1300
1301         inode_lock(src_parent);
1302         inode_lock(dir);
1303         bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1304         if (!IS_ERR(bh)) {
1305                 /*
1306                  * XXX: If some other object occupied the same slot. And If such
1307                  *      inode is zero-sized and with SUID+SGID, then means it is
1308                  *      a new created one. Maybe we can remove it and insert the
1309                  *      original one back to the /O/<seq>/d<x>. But there are
1310                  *      something to be considered:
1311                  *
1312                  *      1) The OST-object under /lost+found has crashed LMA.
1313                  *         So it should not conflict with the current one.
1314                  *
1315                  *      2) There are race conditions that: someone may just want
1316                  *         to modify the current one. Even if the OI scrub takes
1317                  *         the object lock when remove the current one, it still
1318                  *         cause the modification to be lost becasue the target
1319                  *         has been removed when the RPC service thread waiting
1320                  *         for the lock.
1321                  *
1322                  *      So keep it there before we have suitable solution.
1323                  */
1324                 brelse(bh);
1325                 inode_unlock(dir);
1326                 inode_unlock(src_parent);
1327                 ldiskfs_journal_stop(jh);
1328
1329                 rc = -EEXIST;
1330                 /* If the src object has never been modified, then remove it. */
1331                 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1332                     inode->i_mode & S_ISGID) {
1333                         rc = ll_vfs_unlink(src_parent, src_child);
1334                         if (unlikely(rc == -ENOENT))
1335                                 rc = 0;
1336                 }
1337                 if (rc)
1338                         RETURN(rc);
1339         }
1340
1341         bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1342                                     NULL, NULL);
1343         if (unlikely(IS_ERR(bh)))
1344                 GOTO(unlock, rc = PTR_ERR(bh));
1345
1346         rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1347         brelse(bh);
1348         if (rc != 0)
1349                 GOTO(unlock, rc);
1350
1351         rc = osd_ldiskfs_add_entry(info, osd, jh, tgt_child, inode, NULL);
1352
1353         GOTO(unlock, rc);
1354
1355 unlock:
1356         inode_unlock(dir);
1357         inode_unlock(src_parent);
1358         ldiskfs_journal_stop(jh);
1359         return rc;
1360 }
1361
1362 static struct dentry *
1363 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1364                      const struct lu_fid *fid, char **name)
1365 {
1366         struct dentry *root = ERR_PTR(-ENOENT);
1367
1368         if (fid_is_last_id(fid)) {
1369                 struct osd_obj_seq *osd_seq;
1370
1371                 /* on creation of LAST_ID we create O/<seq> hierarchy */
1372                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1373                 if (IS_ERR(osd_seq))
1374                         RETURN((struct dentry *)osd_seq);
1375
1376                 *name = "LAST_ID";
1377                 root = osd_seq->oos_root;
1378         } else {
1379                 *name = osd_lf_fid2name(fid);
1380                 if (*name == NULL)
1381                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1382                 else if ((*name)[0])
1383                         root = osd_sb(osd)->s_root;
1384         }
1385
1386         return root;
1387 }
1388
1389 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1390                         const struct lu_fid *fid, const struct osd_inode_id *id,
1391                         handle_t *th)
1392 {
1393         struct dentry *root;
1394         char *name = NULL;
1395         int rc;
1396
1397         ENTRY;
1398
1399         root = osd_object_spec_find(info, osd, fid, &name);
1400         if (!IS_ERR(root)) {
1401                 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1402         } else {
1403                 rc = PTR_ERR(root);
1404                 if (rc == -ENOENT)
1405                         rc = 1;
1406         }
1407
1408         RETURN(rc);
1409 }
1410
1411 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1412                         const struct lu_fid *fid, const struct osd_inode_id *id,
1413                         handle_t *th)
1414 {
1415         struct dentry *root;
1416         char *name = NULL;
1417         int rc;
1418
1419         ENTRY;
1420
1421         root = osd_object_spec_find(info, osd, fid, &name);
1422         if (!IS_ERR(root)) {
1423                 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1424         } else {
1425                 rc = PTR_ERR(root);
1426                 if (rc == -ENOENT)
1427                         rc = 0;
1428         }
1429
1430         RETURN(rc);
1431 }
1432
1433 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1434                         const struct lu_fid *fid, struct osd_inode_id *id)
1435 {
1436         struct dentry *root;
1437         struct dentry *dentry;
1438         struct inode *inode;
1439         char *name = NULL;
1440         int rc = -ENOENT;
1441
1442         ENTRY;
1443
1444         if (fid_is_last_id(fid)) {
1445                 struct osd_obj_seq *osd_seq;
1446
1447                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1448                 if (IS_ERR(osd_seq))
1449                         RETURN(PTR_ERR(osd_seq));
1450                 root = osd_seq->oos_root;
1451                 name = "LAST_ID";
1452         } else {
1453                 root = osd_sb(osd)->s_root;
1454                 name = osd_lf_fid2name(fid);
1455                 if (name == NULL || strlen(name) == 0)
1456                         RETURN(-ENOENT);
1457         }
1458
1459         dentry = osd_lookup_one_len_unlocked(osd, name, root, strlen(name));
1460         if (!IS_ERR(dentry)) {
1461                 inode = dentry->d_inode;
1462                 if (inode) {
1463                         if (is_bad_inode(inode)) {
1464                                 rc = -EIO;
1465                         } else {
1466                                 osd_id_gen(id, inode->i_ino,
1467                                            inode->i_generation);
1468                                 rc = 0;
1469                         }
1470                 }
1471                 /*
1472                  * if dentry is accessible after osd_compat_spec_insert it
1473                  * will still contain NULL inode, so don't keep it in cache
1474                  */
1475                 d_invalidate(dentry);
1476                 dput(dentry);
1477         }
1478
1479         RETURN(rc);
1480 }
1481
1482 #ifndef HAVE_BIO_INTEGRITY_ENABLED
1483 bool bio_integrity_enabled(struct bio *bio)
1484 {
1485         struct blk_integrity *bi = blk_get_integrity(bio_get_disk(bio));
1486
1487         if (bio_op(bio) != REQ_OP_READ && bio_op(bio) != REQ_OP_WRITE)
1488                 return false;
1489
1490         if (!bio_sectors(bio))
1491                 return false;
1492
1493          /* Already protected? */
1494         if (bio_integrity(bio))
1495                 return false;
1496
1497         if (bi == NULL)
1498                 return false;
1499
1500         if (bio_data_dir(bio) == READ && bi->profile->verify_fn != NULL &&
1501             (bi->flags & BLK_INTEGRITY_VERIFY))
1502                 return true;
1503
1504         if (bio_data_dir(bio) == WRITE && bi->profile->generate_fn != NULL &&
1505             (bi->flags & BLK_INTEGRITY_GENERATE))
1506                 return true;
1507
1508         return false;
1509 }
1510 #endif