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