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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_compat.c
  37  *
  38  * on-disk structure for managing /O
  39  *
  40  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  41  */
  42
  43 /* LUSTRE_VERSION_CODE */
  44 #include <lustre_ver.h>
  45 /* prerequisite for linux/xattr.h */
  46 #include <linux/types.h>
  47 /* prerequisite for linux/xattr.h */
  48 #include <linux/fs.h>
  49 /* XATTR_{REPLACE,CREATE} */
  50 #include <linux/xattr.h>
  51
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  * OBD_FAIL_CHECK
  55  */
  56 #include <obd_support.h>
  57
  58 #include "osd_internal.h"
  59 #include "osd_oi.h"
  60
  61 static void osd_push_ctxt(const struct osd_device *dev,
  62                           struct lvfs_run_ctxt *newctxt,
  63                           struct lvfs_run_ctxt *save)
  64 {
  65         OBD_SET_CTXT_MAGIC(newctxt);
  66         newctxt->pwdmnt = dev->od_mnt;
  67         newctxt->pwd = dev->od_mnt->mnt_root;
  68         newctxt->fs = get_ds();
  69
  70         push_ctxt(save, newctxt, NULL);
  71 }
  72
  73 static void osd_pop_ctxt(const struct osd_device *dev,
  74                          struct lvfs_run_ctxt *new,
  75                          struct lvfs_run_ctxt *save)
  76 {
  77         pop_ctxt(save, new, NULL);
  78 }
  79
  80 /* utility to make a directory */
  81 static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
  82                                    const char *name, int mode, int fix)
  83 {
  84         struct dentry *dchild;
  85         int err = 0;
  86         ENTRY;
  87
  88         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
  89         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
  90         dchild = ll_lookup_one_len(name, dir, strlen(name));
  91         if (IS_ERR(dchild))
  92                 GOTO(out_up, dchild);
  93
  94         if (dchild->d_inode) {
  95                 int old_mode = dchild->d_inode->i_mode;
  96                 if (!S_ISDIR(old_mode)) {
  97                         CERROR("found %s (%lu/%u) is mode %o\n", name,
  98                                dchild->d_inode->i_ino,
  99                                dchild->d_inode->i_generation, old_mode);
 100                         GOTO(out_err, err = -ENOTDIR);
 101                 }
 102
 103                 /* Fixup directory permissions if necessary */
 104                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 105                         CDEBUG(D_CONFIG,
 106                                "fixing permissions on %s from %o to %o\n",
 107                                name, old_mode, mode);
 108                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 109                                                   (old_mode & ~S_IALLUGO);
 110                         mark_inode_dirty(dchild->d_inode);
 111                 }
 112                 GOTO(out_up, dchild);
 113         }
 114
 115         err = ll_vfs_mkdir(dir->d_inode, dchild, mnt, mode);
 116         if (err)
 117                 GOTO(out_err, err);
 118
 119         RETURN(dchild);
 120
 121 out_err:
 122         dput(dchild);
 123         dchild = ERR_PTR(err);
 124 out_up:
 125         return dchild;
 126 }
 127
 128 int osd_last_rcvd_subdir_count(struct osd_device *osd)
 129 {
 130         struct lr_server_data lsd;
 131         struct dentry        *dlast;
 132         loff_t                off;
 133         int                   rc = 0;
 134         int                   count = FILTER_SUBDIR_COUNT;
 135
 136         ENTRY;
 137
 138         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
 139                                   strlen(LAST_RCVD));
 140         if (IS_ERR(dlast))
 141                 return PTR_ERR(dlast);
 142         else if (dlast->d_inode == NULL)
 143                 goto out;
 144
 145         off = 0;
 146         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
 147         if (rc == sizeof(lsd)) {
 148                 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
 149                        "subdir count = %d\n", lsd.lsd_uuid,
 150                        lsd.lsd_subdir_count);
 151                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 152                         count = le16_to_cpu(lsd.lsd_subdir_count);
 153         } else if (rc != 0) {
 154                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
 155                 if (rc > 0)
 156                         rc = -EFAULT;
 157                 dput(dlast);
 158                 return rc;
 159         }
 160 out:
 161         dput(dlast);
 162         LASSERT(count > 0);
 163         return count;
 164 }
 165
 166 static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
 167 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
 168 {
 169         struct lvfs_run_ctxt    new;
 170         struct lvfs_run_ctxt    save;
 171         struct dentry           *parent;
 172         struct osd_mdobj_map    *omm;
 173         struct dentry           *d;
 174         struct osd_thread_info  *info = osd_oti_get(env);
 175         struct lu_fid           *fid = &info->oti_fid;
 176         int                     rc = 0;
 177         ENTRY;
 178
 179         OBD_ALLOC_PTR(dev->od_mdt_map);
 180         if (dev->od_mdt_map == NULL)
 181                 RETURN(-ENOMEM);
 182
 183         omm = dev->od_mdt_map;
 184
 185         LASSERT(dev->od_fsops);
 186
 187         parent = osd_sb(dev)->s_root;
 188         osd_push_ctxt(dev, &new, &save);
 189
 190         d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
 191                          0755, 1);
 192         if (IS_ERR(d))
 193                 GOTO(cleanup, rc = PTR_ERR(d));
 194
 195         ldiskfs_set_inode_state(d->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 196         omm->omm_remote_parent = d;
 197
 198         /* Set LMA for remote parent inode */
 199         lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
 200         rc = osd_ea_fid_set(info, d->d_inode, fid, 0);
 201         if (rc != 0)
 202                 GOTO(cleanup, rc);
 203 cleanup:
 204         pop_ctxt(&save, &new, NULL);
 205         if (rc) {
 206                 if (omm->omm_remote_parent != NULL)
 207                         dput(omm->omm_remote_parent);
 208                 OBD_FREE_PTR(omm);
 209                 dev->od_mdt_map = NULL;
 210         }
 211         RETURN(rc);
 212 }
 213
 214 static void osd_mdt_fini(struct osd_device *osd)
 215 {
 216         struct osd_mdobj_map *omm = osd->od_mdt_map;
 217
 218         if (omm == NULL)
 219                 return;
 220
 221         if (omm->omm_remote_parent)
 222                 dput(omm->omm_remote_parent);
 223
 224         OBD_FREE_PTR(omm);
 225         osd->od_ost_map = NULL;
 226 }
 227
 228 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
 229                              struct osd_object *obj, struct osd_thandle *oh)
 230 {
 231         struct osd_mdobj_map    *omm = osd->od_mdt_map;
 232         struct osd_thread_info  *oti = osd_oti_get(env);
 233         struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
 234         char                    *name = oti->oti_name;
 235         struct dentry           *dentry;
 236         struct dentry           *parent;
 237         int                     rc;
 238
 239         /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
 240          * can identify this object quickly */
 241         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 242         if (rc != 0)
 243                 RETURN(rc);
 244
 245         lma->lma_incompat |= LMAI_REMOTE_PARENT;
 246         lustre_lma_swab(lma);
 247         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 248                              sizeof(*lma), XATTR_REPLACE);
 249         if (rc != 0)
 250                 RETURN(rc);
 251
 252         parent = omm->omm_remote_parent;
 253         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 254         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 255                                            name, strlen(name));
 256         mutex_lock(&parent->d_inode->i_mutex);
 257         rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
 258                                    NULL);
 259         CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
 260                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 261         LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
 262                  osd_name(osd), parent->d_inode->i_ino,
 263                  parent->d_inode->i_nlink);
 264         parent->d_inode->i_nlink++;
 265         mark_inode_dirty(parent->d_inode);
 266         mutex_unlock(&parent->d_inode->i_mutex);
 267         RETURN(rc);
 268 }
 269
 270 int osd_delete_from_remote_parent(const struct lu_env *env,
 271                                   struct osd_device *osd,
 272                                   struct osd_object *obj,
 273                                   struct osd_thandle *oh)
 274 {
 275         struct osd_mdobj_map       *omm = osd->od_mdt_map;
 276         struct osd_thread_info     *oti = osd_oti_get(env);
 277         struct lustre_mdt_attrs    *lma = &oti->oti_mdt_attrs;
 278         char                       *name = oti->oti_name;
 279         struct dentry              *dentry;
 280         struct dentry              *parent;
 281         struct ldiskfs_dir_entry_2 *de;
 282         struct buffer_head         *bh;
 283         int                        rc;
 284
 285         /* Check lma to see whether it is remote object */
 286         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 287         if (rc != 0)
 288                 RETURN(rc);
 289
 290         if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
 291                 RETURN(0);
 292
 293         parent = omm->omm_remote_parent;
 294         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 295         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 296                                            name, strlen(name));
 297         mutex_lock(&parent->d_inode->i_mutex);
 298         bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
 299         if (bh == NULL) {
 300                 mutex_unlock(&parent->d_inode->i_mutex);
 301                 RETURN(-ENOENT);
 302         }
 303         CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
 304                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 305         rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
 306         LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
 307                  osd_name(osd), parent->d_inode->i_ino,
 308                  parent->d_inode->i_nlink);
 309         parent->d_inode->i_nlink--;
 310         mark_inode_dirty(parent->d_inode);
 311         mutex_unlock(&parent->d_inode->i_mutex);
 312         brelse(bh);
 313
 314         /* Get rid of REMOTE_PARENT flag from incompat */
 315         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
 316         lustre_lma_swab(lma);
 317         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 318                              sizeof(*lma), XATTR_REPLACE);
 319         RETURN(rc);
 320 }
 321
 322 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 323                                 struct osd_device *osd,
 324                                 const struct lu_fid *fid,
 325                                 struct osd_inode_id *id)
 326 {
 327         struct osd_mdobj_map        *omm = osd->od_mdt_map;
 328         char                        *name = oti->oti_name;
 329         struct dentry               *parent;
 330         struct dentry               *dentry;
 331         struct ldiskfs_dir_entry_2 *de;
 332         struct buffer_head         *bh;
 333         int                         rc;
 334         ENTRY;
 335
 336         parent = omm->omm_remote_parent;
 337         sprintf(name, DFID_NOBRACE, PFID(fid));
 338         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
 339                                            name, strlen(name));
 340         mutex_lock(&parent->d_inode->i_mutex);
 341         bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
 342         if (bh == NULL) {
 343                 rc = -ENOENT;
 344         } else {
 345                 rc = 0;
 346                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 347                 brelse(bh);
 348         }
 349         mutex_unlock(&parent->d_inode->i_mutex);
 350         if (rc == 0)
 351                 osd_add_oi_cache(oti, osd, id, fid);
 352         RETURN(rc);
 353 }
 354
 355 /*
 356  * directory structure on legacy OST:
 357  *
 358  * O/<seq>/d0-31/<objid>
 359  * O/<seq>/LAST_ID
 360  * last_rcvd
 361  * LAST_GROUP
 362  * CONFIGS
 363  *
 364  */
 365 static int osd_ost_init(struct osd_device *dev)
 366 {
 367         struct lvfs_run_ctxt  new;
 368         struct lvfs_run_ctxt  save;
 369         struct dentry        *rootd = osd_sb(dev)->s_root;
 370         struct dentry        *d;
 371         int                   rc;
 372         ENTRY;
 373
 374         OBD_ALLOC_PTR(dev->od_ost_map);
 375         if (dev->od_ost_map == NULL)
 376                 RETURN(-ENOMEM);
 377
 378         /* to get subdir count from last_rcvd */
 379         rc = osd_last_rcvd_subdir_count(dev);
 380         if (rc < 0) {
 381                 OBD_FREE_PTR(dev->od_ost_map);
 382                 RETURN(rc);
 383         }
 384
 385         dev->od_ost_map->om_subdir_count = rc;
 386         rc = 0;
 387
 388         CFS_INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 389         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 390         sema_init(&dev->od_ost_map->om_dir_init_sem, 1);
 391
 392         LASSERT(dev->od_fsops);
 393         osd_push_ctxt(dev, &new, &save);
 394
 395         d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 396         if (IS_ERR(d))
 397                 GOTO(cleanup, rc = PTR_ERR(d));
 398
 399         ldiskfs_set_inode_state(d->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 400         dev->od_ost_map->om_root = d;
 401
 402 cleanup:
 403         osd_pop_ctxt(dev, &new, &save);
 404         if (IS_ERR(d)) {
 405                 OBD_FREE_PTR(dev->od_ost_map);
 406                 RETURN(PTR_ERR(d));
 407         }
 408
 409         RETURN(rc);
 410 }
 411
 412 static void osd_seq_free(struct osd_obj_map *map,
 413                          struct osd_obj_seq *osd_seq)
 414 {
 415         int j;
 416
 417         cfs_list_del_init(&osd_seq->oos_seq_list);
 418
 419         if (osd_seq->oos_dirs) {
 420                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 421                         if (osd_seq->oos_dirs[j])
 422                                 dput(osd_seq->oos_dirs[j]);
 423                 }
 424                 OBD_FREE(osd_seq->oos_dirs,
 425                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 426         }
 427
 428         if (osd_seq->oos_root)
 429                 dput(osd_seq->oos_root);
 430
 431         OBD_FREE_PTR(osd_seq);
 432 }
 433
 434 static void osd_ost_fini(struct osd_device *osd)
 435 {
 436         struct osd_obj_seq    *osd_seq;
 437         struct osd_obj_seq    *tmp;
 438         struct osd_obj_map    *map = osd->od_ost_map;
 439         ENTRY;
 440
 441         if (map == NULL)
 442                 return;
 443
 444         write_lock(&map->om_seq_list_lock);
 445         cfs_list_for_each_entry_safe(osd_seq, tmp,
 446                                      &map->om_seq_list,
 447                                      oos_seq_list) {
 448                 osd_seq_free(map, osd_seq);
 449         }
 450         write_unlock(&map->om_seq_list_lock);
 451         if (map->om_root)
 452                 dput(map->om_root);
 453         OBD_FREE_PTR(map);
 454         osd->od_ost_map = NULL;
 455         EXIT;
 456 }
 457
 458 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 459 {
 460         int rc;
 461         ENTRY;
 462
 463         /* prepare structures for OST */
 464         rc = osd_ost_init(dev);
 465         if (rc)
 466                 RETURN(rc);
 467
 468         /* prepare structures for MDS */
 469         rc = osd_mdt_init(env, dev);
 470
 471         RETURN(rc);
 472 }
 473
 474 struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq)
 475 {
 476         struct osd_obj_seq *osd_seq;
 477
 478         cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 479                 if (osd_seq->oos_seq == seq)
 480                         return osd_seq;
 481         }
 482         return NULL;
 483 }
 484
 485 struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
 486 {
 487         struct osd_obj_seq *osd_seq;
 488
 489         read_lock(&map->om_seq_list_lock);
 490         osd_seq = osd_seq_find_locked(map, seq);
 491         read_unlock(&map->om_seq_list_lock);
 492         return osd_seq;
 493 }
 494
 495 void osd_obj_map_fini(struct osd_device *dev)
 496 {
 497         osd_ost_fini(dev);
 498         osd_mdt_fini(dev);
 499 }
 500
 501 static int osd_obj_del_entry(struct osd_thread_info *info,
 502                              struct osd_device *osd,
 503                              struct dentry *dird, char *name,
 504                              struct thandle *th)
 505 {
 506         struct ldiskfs_dir_entry_2 *de;
 507         struct buffer_head         *bh;
 508         struct osd_thandle         *oh;
 509         struct dentry              *child;
 510         struct inode               *dir = dird->d_inode;
 511         int                         rc;
 512         ENTRY;
 513
 514         oh = container_of(th, struct osd_thandle, ot_super);
 515         LASSERT(oh->ot_handle != NULL);
 516         LASSERT(oh->ot_handle->h_transaction != NULL);
 517
 518
 519         child = &info->oti_child_dentry;
 520         child->d_name.hash = 0;
 521         child->d_name.name = name;
 522         child->d_name.len = strlen(name);
 523         child->d_parent = dird;
 524         child->d_inode = NULL;
 525
 526         ll_vfs_dq_init(dir);
 527         mutex_lock(&dir->i_mutex);
 528         rc = -ENOENT;
 529         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 530         if (bh) {
 531                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
 532                 brelse(bh);
 533         }
 534         mutex_unlock(&dir->i_mutex);
 535
 536         RETURN(rc);
 537 }
 538
 539 int osd_obj_add_entry(struct osd_thread_info *info,
 540                       struct osd_device *osd,
 541                       struct dentry *dir, char *name,
 542                       const struct osd_inode_id *id,
 543                       struct thandle *th)
 544 {
 545         struct osd_thandle *oh;
 546         struct dentry *child;
 547         struct inode *inode;
 548         int rc;
 549
 550         ENTRY;
 551
 552         oh = container_of(th, struct osd_thandle, ot_super);
 553         LASSERT(oh->ot_handle != NULL);
 554         LASSERT(oh->ot_handle->h_transaction != NULL);
 555
 556         inode = &info->oti_inode;
 557         inode->i_sb = osd_sb(osd);
 558         osd_id_to_inode(inode, id);
 559         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 560
 561         child = &info->oti_child_dentry;
 562         child->d_name.hash = 0;
 563         child->d_name.name = name;
 564         child->d_name.len = strlen(name);
 565         child->d_parent = dir;
 566         child->d_inode = inode;
 567
 568         ll_vfs_dq_init(dir->d_inode);
 569         mutex_lock(&dir->d_inode->i_mutex);
 570         rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
 571         mutex_unlock(&dir->d_inode->i_mutex);
 572
 573         RETURN(rc);
 574 }
 575
 576 /**
 577  * Use LPU64 for legacy OST sequences, but use LPX64i for new
 578  * sequences names, so that the O/{seq}/dN/{oid} more closely
 579  * follows the DFID/PFID format. This makes it easier to map from
 580  * debug messages to objects in the future, and the legacy space
 581  * of FID_SEQ_OST_MDT0 will be unused in the future.
 582  **/
 583 static inline void osd_seq_name(char *seq_name, size_t name_size, obd_seq seq)
 584 {
 585         snprintf(seq_name, name_size,
 586                  (fid_seq_is_rsvd(seq) ||
 587                   fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
 588                  fid_seq_is_idif(seq) ? 0 : seq);
 589 }
 590
 591 static inline void osd_oid_name(char *name, size_t name_size,
 592                                 const struct lu_fid *fid, obd_id id)
 593 {
 594         snprintf(name, name_size,
 595                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 596                   fid_seq_is_mdt0(fid_seq(fid)) ||
 597                   fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
 598 }
 599
 600 /* external locking is required */
 601 static int osd_seq_load_locked(struct osd_device *osd,
 602                                struct osd_obj_seq *osd_seq)
 603 {
 604         struct osd_obj_map  *map = osd->od_ost_map;
 605         struct dentry       *seq_dir;
 606         int                 rc = 0;
 607         int                 i;
 608         char                dir_name[32];
 609         ENTRY;
 610
 611         if (osd_seq->oos_root != NULL)
 612                 RETURN(0);
 613
 614         LASSERT(map);
 615         LASSERT(map->om_root);
 616
 617         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 618
 619         seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
 620         if (IS_ERR(seq_dir))
 621                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 622         else if (seq_dir->d_inode == NULL)
 623                 GOTO(out_put, rc = -EFAULT);
 624
 625         ldiskfs_set_inode_state(seq_dir->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 626         osd_seq->oos_root = seq_dir;
 627
 628         LASSERT(osd_seq->oos_dirs == NULL);
 629         OBD_ALLOC(osd_seq->oos_dirs,
 630                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 631         if (osd_seq->oos_dirs == NULL)
 632                 GOTO(out_put, rc = -ENOMEM);
 633
 634         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 635                 struct dentry   *dir;
 636
 637                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 638                 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
 639                                    0700, 1);
 640                 if (IS_ERR(dir)) {
 641                         GOTO(out_free, rc = PTR_ERR(dir));
 642                 } else if (dir->d_inode == NULL) {
 643                         dput(dir);
 644                         GOTO(out_free, rc = -EFAULT);
 645                 }
 646
 647                 ldiskfs_set_inode_state(dir->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 648                 osd_seq->oos_dirs[i] = dir;
 649         }
 650
 651         if (rc != 0) {
 652 out_free:
 653                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 654                         if (osd_seq->oos_dirs[i] != NULL)
 655                                 dput(osd_seq->oos_dirs[i]);
 656                 }
 657                 OBD_FREE(osd_seq->oos_dirs,
 658                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
 659 out_put:
 660                 dput(seq_dir);
 661                 osd_seq->oos_root = NULL;
 662         }
 663 out_err:
 664         RETURN(rc);
 665 }
 666
 667 struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq)
 668 {
 669         struct osd_obj_map      *map;
 670         struct osd_obj_seq      *osd_seq;
 671         int                     rc = 0;
 672         ENTRY;
 673
 674         map = osd->od_ost_map;
 675         LASSERT(map);
 676         LASSERT(map->om_root);
 677
 678         osd_seq = osd_seq_find(map, seq);
 679         if (likely(osd_seq != NULL))
 680                 RETURN(osd_seq);
 681
 682         /* Serializing init process */
 683         down(&map->om_dir_init_sem);
 684
 685         /* Check whether the seq has been added */
 686         read_lock(&map->om_seq_list_lock);
 687         osd_seq = osd_seq_find_locked(map, seq);
 688         if (osd_seq != NULL) {
 689                 read_unlock(&map->om_seq_list_lock);
 690                 GOTO(cleanup, rc = 0);
 691         }
 692         read_unlock(&map->om_seq_list_lock);
 693
 694         OBD_ALLOC_PTR(osd_seq);
 695         if (osd_seq == NULL)
 696                 GOTO(cleanup, rc = -ENOMEM);
 697
 698         CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 699         osd_seq->oos_seq = seq;
 700         /* Init subdir count to be 32, but each seq can have
 701          * different subdir count */
 702         osd_seq->oos_subdir_count = map->om_subdir_count;
 703         rc = osd_seq_load_locked(osd, osd_seq);
 704         if (rc != 0)
 705                 GOTO(cleanup, rc);
 706
 707         write_lock(&map->om_seq_list_lock);
 708         cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 709         write_unlock(&map->om_seq_list_lock);
 710
 711 cleanup:
 712         up(&map->om_dir_init_sem);
 713         if (rc != 0) {
 714                 if (osd_seq != NULL)
 715                         OBD_FREE_PTR(osd_seq);
 716                 RETURN(ERR_PTR(rc));
 717         }
 718
 719         RETURN(osd_seq);
 720 }
 721
 722 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 723                        const struct lu_fid *fid, struct osd_inode_id *id)
 724 {
 725         struct osd_obj_map              *map;
 726         struct osd_obj_seq              *osd_seq;
 727         struct dentry                   *d_seq;
 728         struct dentry                   *child;
 729         struct ost_id                   *ostid = &info->oti_ostid;
 730         int                             dirn;
 731         char                            name[32];
 732         struct ldiskfs_dir_entry_2      *de;
 733         struct buffer_head              *bh;
 734         struct inode                    *dir;
 735         struct inode                    *inode;
 736         ENTRY;
 737
 738         /* on the very first lookup we find and open directories */
 739
 740         map = dev->od_ost_map;
 741         LASSERT(map);
 742         LASSERT(map->om_root);
 743
 744         fid_to_ostid(fid, ostid);
 745         osd_seq = osd_seq_load(dev, ostid_seq(ostid));
 746         if (IS_ERR(osd_seq))
 747                 RETURN(PTR_ERR(osd_seq));
 748
 749         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 750         d_seq = osd_seq->oos_dirs[dirn];
 751         LASSERT(d_seq);
 752
 753         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 754
 755         child = &info->oti_child_dentry;
 756         child->d_parent = d_seq;
 757         child->d_name.hash = 0;
 758         child->d_name.name = name;
 759         /* XXX: we can use rc from sprintf() instead of strlen() */
 760         child->d_name.len = strlen(name);
 761
 762         dir = d_seq->d_inode;
 763         mutex_lock(&dir->i_mutex);
 764         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 765         mutex_unlock(&dir->i_mutex);
 766
 767         if (bh == NULL)
 768                 RETURN(-ENOENT);
 769
 770         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 771         brelse(bh);
 772
 773         inode = osd_iget(info, dev, id);
 774         if (IS_ERR(inode))
 775                 RETURN(PTR_ERR(inode));
 776
 777         iput(inode);
 778         RETURN(0);
 779 }
 780
 781 int osd_obj_map_insert(struct osd_thread_info *info,
 782                        struct osd_device *osd,
 783                        const struct lu_fid *fid,
 784                        const struct osd_inode_id *id,
 785                        struct thandle *th)
 786 {
 787         struct osd_obj_map      *map;
 788         struct osd_obj_seq      *osd_seq;
 789         struct dentry           *d;
 790         struct ost_id           *ostid = &info->oti_ostid;
 791         int                     dirn, rc = 0;
 792         char                    name[32];
 793         ENTRY;
 794
 795         map = osd->od_ost_map;
 796         LASSERT(map);
 797
 798         /* map fid to seq:objid */
 799         fid_to_ostid(fid, ostid);
 800
 801         osd_seq = osd_seq_load(osd, ostid_seq(ostid));
 802         if (IS_ERR(osd_seq))
 803                 RETURN(PTR_ERR(osd_seq));
 804
 805         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 806         d = osd_seq->oos_dirs[dirn];
 807         LASSERT(d);
 808
 809         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 810         rc = osd_obj_add_entry(info, osd, d, name, id, th);
 811
 812         RETURN(rc);
 813 }
 814
 815 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
 816                        const struct lu_fid *fid, struct thandle *th)
 817 {
 818         struct osd_obj_map      *map;
 819         struct osd_obj_seq      *osd_seq;
 820         struct dentry           *d;
 821         struct ost_id           *ostid = &info->oti_ostid;
 822         int                     dirn, rc = 0;
 823         char                    name[32];
 824         ENTRY;
 825
 826         map = osd->od_ost_map;
 827         LASSERT(map);
 828
 829         /* map fid to seq:objid */
 830         fid_to_ostid(fid, ostid);
 831
 832         osd_seq = osd_seq_load(osd, ostid_seq(ostid));
 833         if (IS_ERR(osd_seq))
 834                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
 835
 836         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 837         d = osd_seq->oos_dirs[dirn];
 838         LASSERT(d);
 839
 840         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 841         rc = osd_obj_del_entry(info, osd, d, name, th);
 842 cleanup:
 843         RETURN(rc);
 844 }
 845
 846 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
 847                         const struct lu_fid *fid,
 848                         const struct osd_inode_id *id,
 849                         struct thandle *th)
 850 {
 851         struct osd_obj_map      *map = osd->od_ost_map;
 852         struct dentry           *root = osd_sb(osd)->s_root;
 853         char                    *name;
 854         int                     rc = 0;
 855         ENTRY;
 856
 857         if (fid_is_last_id(fid)) {
 858                 struct osd_obj_seq      *osd_seq;
 859
 860                 /* on creation of LAST_ID we create O/<seq> hierarchy */
 861                 LASSERT(map);
 862                 osd_seq = osd_seq_load(osd, fid_seq(fid));
 863                 if (IS_ERR(osd_seq))
 864                         RETURN(PTR_ERR(osd_seq));
 865                 rc = osd_obj_add_entry(info, osd, osd_seq->oos_root,
 866                                        "LAST_ID", id, th);
 867         } else {
 868                 name = osd_lf_fid2name(fid);
 869                 if (name == NULL)
 870                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
 871                 else if (name[0])
 872                         rc = osd_obj_add_entry(info, osd, root, name, id, th);
 873         }
 874
 875         RETURN(rc);
 876 }
 877
 878 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
 879                         const struct lu_fid *fid, struct osd_inode_id *id)
 880 {
 881         struct dentry   *root;
 882         struct dentry   *dentry;
 883         struct inode    *inode;
 884         char            *name;
 885         int             rc = -ENOENT;
 886         ENTRY;
 887
 888         if (fid_is_last_id(fid)) {
 889                 struct osd_obj_seq *osd_seq;
 890
 891                 osd_seq = osd_seq_load(osd, fid_seq(fid));
 892                 if (IS_ERR(osd_seq))
 893                         RETURN(PTR_ERR(osd_seq));
 894                 root = osd_seq->oos_root;
 895                 name = "LAST_ID";
 896         } else {
 897                 root = osd_sb(osd)->s_root;
 898                 name = osd_lf_fid2name(fid);
 899                 if (name == NULL || strlen(name) == 0)
 900                         RETURN(-ENOENT);
 901         }
 902
 903         dentry = ll_lookup_one_len(name, root, strlen(name));
 904         if (!IS_ERR(dentry)) {
 905                 inode = dentry->d_inode;
 906                 if (inode) {
 907                         if (is_bad_inode(inode)) {
 908                                 rc = -EIO;
 909                         } else {
 910                                 osd_id_gen(id, inode->i_ino,
 911                                            inode->i_generation);
 912                                 rc = 0;
 913                         }
 914                 }
 915                 /* if dentry is accessible after osd_compat_spec_insert it
 916                  * will still contain NULL inode, so don't keep it in cache */
 917                 d_invalidate(dentry);
 918                 dput(dentry);
 919         }
 920
 921         RETURN(rc);
 922 }