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_fid3;
 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, LMAC_NOT_IN_OI, 0);
 201
 202         GOTO(cleanup, rc);
 203
 204 cleanup:
 205         pop_ctxt(&save, &new, NULL);
 206         if (rc) {
 207                 if (omm->omm_remote_parent != NULL)
 208                         dput(omm->omm_remote_parent);
 209                 OBD_FREE_PTR(omm);
 210                 dev->od_mdt_map = NULL;
 211         }
 212         return rc;
 213 }
 214
 215 static void osd_mdt_fini(struct osd_device *osd)
 216 {
 217         struct osd_mdobj_map *omm = osd->od_mdt_map;
 218
 219         if (omm == NULL)
 220                 return;
 221
 222         if (omm->omm_remote_parent)
 223                 dput(omm->omm_remote_parent);
 224
 225         OBD_FREE_PTR(omm);
 226         osd->od_ost_map = NULL;
 227 }
 228
 229 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
 230                              struct osd_object *obj, struct osd_thandle *oh)
 231 {
 232         struct osd_mdobj_map    *omm = osd->od_mdt_map;
 233         struct osd_thread_info  *oti = osd_oti_get(env);
 234         struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
 235         char                    *name = oti->oti_name;
 236         struct dentry           *dentry;
 237         struct dentry           *parent;
 238         int                     rc;
 239
 240         /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
 241          * can identify this object quickly */
 242         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 243         if (rc != 0)
 244                 RETURN(rc);
 245
 246         lma->lma_incompat |= LMAI_REMOTE_PARENT;
 247         lustre_lma_swab(lma);
 248         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 249                              sizeof(*lma), XATTR_REPLACE);
 250         if (rc != 0)
 251                 RETURN(rc);
 252
 253         parent = omm->omm_remote_parent;
 254         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 255         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 256                                            name, strlen(name));
 257         mutex_lock(&parent->d_inode->i_mutex);
 258         rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
 259                                    NULL);
 260         CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
 261                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 262         LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
 263                  osd_name(osd), parent->d_inode->i_ino,
 264                  parent->d_inode->i_nlink);
 265         parent->d_inode->i_nlink++;
 266         mark_inode_dirty(parent->d_inode);
 267         mutex_unlock(&parent->d_inode->i_mutex);
 268         RETURN(rc);
 269 }
 270
 271 int osd_delete_from_remote_parent(const struct lu_env *env,
 272                                   struct osd_device *osd,
 273                                   struct osd_object *obj,
 274                                   struct osd_thandle *oh)
 275 {
 276         struct osd_mdobj_map       *omm = osd->od_mdt_map;
 277         struct osd_thread_info     *oti = osd_oti_get(env);
 278         struct lustre_mdt_attrs    *lma = &oti->oti_mdt_attrs;
 279         char                       *name = oti->oti_name;
 280         struct dentry              *dentry;
 281         struct dentry              *parent;
 282         struct ldiskfs_dir_entry_2 *de;
 283         struct buffer_head         *bh;
 284         int                        rc;
 285
 286         /* Check lma to see whether it is remote object */
 287         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
 288         if (rc != 0)
 289                 RETURN(rc);
 290
 291         if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
 292                 RETURN(0);
 293
 294         parent = omm->omm_remote_parent;
 295         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 296         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
 297                                            name, strlen(name));
 298         mutex_lock(&parent->d_inode->i_mutex);
 299         bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
 300         if (bh == NULL) {
 301                 mutex_unlock(&parent->d_inode->i_mutex);
 302                 RETURN(-ENOENT);
 303         }
 304         CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
 305                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
 306         rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
 307         LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
 308                  osd_name(osd), parent->d_inode->i_ino,
 309                  parent->d_inode->i_nlink);
 310         parent->d_inode->i_nlink--;
 311         mark_inode_dirty(parent->d_inode);
 312         mutex_unlock(&parent->d_inode->i_mutex);
 313         brelse(bh);
 314
 315         /* Get rid of REMOTE_PARENT flag from incompat */
 316         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
 317         lustre_lma_swab(lma);
 318         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
 319                              sizeof(*lma), XATTR_REPLACE);
 320         RETURN(rc);
 321 }
 322
 323 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
 324                                 struct osd_device *osd,
 325                                 const struct lu_fid *fid,
 326                                 struct osd_inode_id *id)
 327 {
 328         struct osd_mdobj_map        *omm = osd->od_mdt_map;
 329         char                        *name = oti->oti_name;
 330         struct dentry               *parent;
 331         struct dentry               *dentry;
 332         struct ldiskfs_dir_entry_2 *de;
 333         struct buffer_head         *bh;
 334         int                         rc;
 335         ENTRY;
 336
 337         parent = omm->omm_remote_parent;
 338         sprintf(name, DFID_NOBRACE, PFID(fid));
 339         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
 340                                            name, strlen(name));
 341         mutex_lock(&parent->d_inode->i_mutex);
 342         bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
 343         if (bh == NULL) {
 344                 rc = -ENOENT;
 345         } else {
 346                 rc = 0;
 347                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 348                 brelse(bh);
 349         }
 350         mutex_unlock(&parent->d_inode->i_mutex);
 351         if (rc == 0)
 352                 osd_add_oi_cache(oti, osd, id, fid);
 353         RETURN(rc);
 354 }
 355
 356 /*
 357  * directory structure on legacy OST:
 358  *
 359  * O/<seq>/d0-31/<objid>
 360  * O/<seq>/LAST_ID
 361  * last_rcvd
 362  * LAST_GROUP
 363  * CONFIGS
 364  *
 365  */
 366 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
 367 {
 368         struct lvfs_run_ctxt     new;
 369         struct lvfs_run_ctxt     save;
 370         struct dentry           *rootd = osd_sb(dev)->s_root;
 371         struct dentry           *d;
 372         struct osd_thread_info  *info = osd_oti_get(env);
 373         struct inode            *inode;
 374         struct lu_fid           *fid = &info->oti_fid3;
 375         int                      rc;
 376         ENTRY;
 377
 378         OBD_ALLOC_PTR(dev->od_ost_map);
 379         if (dev->od_ost_map == NULL)
 380                 RETURN(-ENOMEM);
 381
 382         /* to get subdir count from last_rcvd */
 383         rc = osd_last_rcvd_subdir_count(dev);
 384         if (rc < 0) {
 385                 OBD_FREE_PTR(dev->od_ost_map);
 386                 RETURN(rc);
 387         }
 388
 389         dev->od_ost_map->om_subdir_count = rc;
 390         rc = 0;
 391
 392         CFS_INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 393         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 394         sema_init(&dev->od_ost_map->om_dir_init_sem, 1);
 395
 396         LASSERT(dev->od_fsops);
 397         osd_push_ctxt(dev, &new, &save);
 398
 399         d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 400         if (IS_ERR(d))
 401                 GOTO(cleanup, rc = PTR_ERR(d));
 402
 403         inode = d->d_inode;
 404         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
 405         dev->od_ost_map->om_root = d;
 406
 407         /* 'What the @fid is' is not imporatant, because the object
 408          * has no OI mapping, and only is visible inside the OSD.*/
 409         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 410         rc = osd_ea_fid_set(info, inode, fid,
 411                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 412
 413         GOTO(cleanup, rc);
 414
 415 cleanup:
 416         osd_pop_ctxt(dev, &new, &save);
 417         if (IS_ERR(d)) {
 418                 OBD_FREE_PTR(dev->od_ost_map);
 419                 RETURN(PTR_ERR(d));
 420         }
 421         return rc;
 422 }
 423
 424 static void osd_seq_free(struct osd_obj_map *map,
 425                          struct osd_obj_seq *osd_seq)
 426 {
 427         int j;
 428
 429         cfs_list_del_init(&osd_seq->oos_seq_list);
 430
 431         if (osd_seq->oos_dirs) {
 432                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 433                         if (osd_seq->oos_dirs[j])
 434                                 dput(osd_seq->oos_dirs[j]);
 435                 }
 436                 OBD_FREE(osd_seq->oos_dirs,
 437                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 438         }
 439
 440         if (osd_seq->oos_root)
 441                 dput(osd_seq->oos_root);
 442
 443         OBD_FREE_PTR(osd_seq);
 444 }
 445
 446 static void osd_ost_fini(struct osd_device *osd)
 447 {
 448         struct osd_obj_seq    *osd_seq;
 449         struct osd_obj_seq    *tmp;
 450         struct osd_obj_map    *map = osd->od_ost_map;
 451         ENTRY;
 452
 453         if (map == NULL)
 454                 return;
 455
 456         write_lock(&map->om_seq_list_lock);
 457         cfs_list_for_each_entry_safe(osd_seq, tmp,
 458                                      &map->om_seq_list,
 459                                      oos_seq_list) {
 460                 osd_seq_free(map, osd_seq);
 461         }
 462         write_unlock(&map->om_seq_list_lock);
 463         if (map->om_root)
 464                 dput(map->om_root);
 465         OBD_FREE_PTR(map);
 466         osd->od_ost_map = NULL;
 467         EXIT;
 468 }
 469
 470 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 471 {
 472         int rc;
 473         ENTRY;
 474
 475         /* prepare structures for OST */
 476         rc = osd_ost_init(env, dev);
 477         if (rc)
 478                 RETURN(rc);
 479
 480         /* prepare structures for MDS */
 481         rc = osd_mdt_init(env, dev);
 482
 483         RETURN(rc);
 484 }
 485
 486 struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq)
 487 {
 488         struct osd_obj_seq *osd_seq;
 489
 490         cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 491                 if (osd_seq->oos_seq == seq)
 492                         return osd_seq;
 493         }
 494         return NULL;
 495 }
 496
 497 struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
 498 {
 499         struct osd_obj_seq *osd_seq;
 500
 501         read_lock(&map->om_seq_list_lock);
 502         osd_seq = osd_seq_find_locked(map, seq);
 503         read_unlock(&map->om_seq_list_lock);
 504         return osd_seq;
 505 }
 506
 507 void osd_obj_map_fini(struct osd_device *dev)
 508 {
 509         osd_ost_fini(dev);
 510         osd_mdt_fini(dev);
 511 }
 512
 513 /**
 514  * Update the specified OI mapping.
 515  *
 516  * \retval   1, changed nothing
 517  * \retval   0, changed successfully
 518  * \retval -ve, on error
 519  */
 520 static int osd_obj_update_entry(struct osd_thread_info *info,
 521                                 struct osd_device *osd,
 522                                 struct dentry *dir, const char *name,
 523                                 const struct lu_fid *fid,
 524                                 const struct osd_inode_id *id,
 525                                 struct thandle *th)
 526 {
 527         struct inode               *parent = dir->d_inode;
 528         struct osd_thandle         *oh;
 529         struct dentry              *child;
 530         struct ldiskfs_dir_entry_2 *de;
 531         struct buffer_head         *bh;
 532         struct inode               *inode;
 533         struct dentry              *dentry = &info->oti_obj_dentry;
 534         struct osd_inode_id        *oi_id  = &info->oti_id3;
 535         struct lustre_mdt_attrs    *lma    = &info->oti_mdt_attrs;
 536         struct lu_fid              *oi_fid = &lma->lma_self_fid;
 537         int                         rc;
 538         ENTRY;
 539
 540         oh = container_of(th, struct osd_thandle, ot_super);
 541         LASSERT(oh->ot_handle != NULL);
 542         LASSERT(oh->ot_handle->h_transaction != NULL);
 543
 544         child = &info->oti_child_dentry;
 545         child->d_parent = dir;
 546         child->d_name.hash = 0;
 547         child->d_name.name = name;
 548         child->d_name.len = strlen(name);
 549
 550         ll_vfs_dq_init(parent);
 551         mutex_lock(&parent->i_mutex);
 552         bh = osd_ldiskfs_find_entry(parent, child, &de, NULL);
 553         if (bh == NULL)
 554                 GOTO(out, rc = -ENOENT);
 555
 556         if (le32_to_cpu(de->inode) == id->oii_ino)
 557                 GOTO(out, rc = 1);
 558
 559         osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 560         inode = osd_iget(info, osd, oi_id);
 561         if (IS_ERR(inode)) {
 562                 rc = PTR_ERR(inode);
 563                 if (rc == -ENOENT || rc == -ESTALE)
 564                         goto update;
 565                 GOTO(out, rc);
 566         }
 567
 568         rc = osd_get_lma(info, inode, dentry, lma);
 569         if (rc == -ENODATA) {
 570                 rc = osd_get_idif(info, inode, dentry, oi_fid);
 571                 if (rc > 0) {
 572                         oi_fid = NULL;
 573                         rc = 0;
 574                 }
 575         }
 576         iput(inode);
 577
 578         /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
 579          * either a crashed object or a uninitialized one. Replace it. */
 580         if (rc == -ENODATA || oi_fid == NULL)
 581                 goto update;
 582
 583         if (rc != 0)
 584                 GOTO(out, rc);
 585
 586         if (lu_fid_eq(fid, oi_fid)) {
 587                 CERROR("%s: the FID "DFID" is used by two objects: "
 588                        "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
 589                        oi_id->oii_ino, oi_id->oii_gen,
 590                        id->oii_ino, id->oii_gen);
 591                 GOTO(out, rc = -EEXIST);
 592         }
 593
 594 update:
 595         /* There may be temporary inconsistency: On one hand, the new
 596          * object may be referenced by multiple entries, which is out
 597          * of our control unless we traverse the whole /O completely,
 598          * which is non-flat order and inefficient, should be avoided;
 599          * On the other hand, the old object may become orphan if it
 600          * is still valid. Since it was referenced by an invalid entry,
 601          * making it as invisible temporary may be not worse. OI scrub
 602          * will process it later. */
 603         rc = ldiskfs_journal_get_write_access(oh->ot_handle, bh);
 604         if (rc != 0)
 605                 GOTO(out, rc);
 606
 607         de->inode = cpu_to_le32(id->oii_ino);
 608         rc = ldiskfs_journal_dirty_metadata(oh->ot_handle, bh);
 609
 610         GOTO(out, rc);
 611
 612 out:
 613         brelse(bh);
 614         mutex_unlock(&parent->i_mutex);
 615         return rc;
 616 }
 617
 618 static int osd_obj_del_entry(struct osd_thread_info *info,
 619                              struct osd_device *osd,
 620                              struct dentry *dird, char *name,
 621                              struct thandle *th)
 622 {
 623         struct ldiskfs_dir_entry_2 *de;
 624         struct buffer_head         *bh;
 625         struct osd_thandle         *oh;
 626         struct dentry              *child;
 627         struct inode               *dir = dird->d_inode;
 628         int                         rc;
 629         ENTRY;
 630
 631         oh = container_of(th, struct osd_thandle, ot_super);
 632         LASSERT(oh->ot_handle != NULL);
 633         LASSERT(oh->ot_handle->h_transaction != NULL);
 634
 635
 636         child = &info->oti_child_dentry;
 637         child->d_name.hash = 0;
 638         child->d_name.name = name;
 639         child->d_name.len = strlen(name);
 640         child->d_parent = dird;
 641         child->d_inode = NULL;
 642
 643         ll_vfs_dq_init(dir);
 644         mutex_lock(&dir->i_mutex);
 645         rc = -ENOENT;
 646         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 647         if (bh) {
 648                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
 649                 brelse(bh);
 650         }
 651         mutex_unlock(&dir->i_mutex);
 652
 653         RETURN(rc);
 654 }
 655
 656 int osd_obj_add_entry(struct osd_thread_info *info,
 657                       struct osd_device *osd,
 658                       struct dentry *dir, char *name,
 659                       const struct osd_inode_id *id,
 660                       struct thandle *th)
 661 {
 662         struct osd_thandle *oh;
 663         struct dentry *child;
 664         struct inode *inode;
 665         int rc;
 666
 667         ENTRY;
 668
 669         oh = container_of(th, struct osd_thandle, ot_super);
 670         LASSERT(oh->ot_handle != NULL);
 671         LASSERT(oh->ot_handle->h_transaction != NULL);
 672
 673         inode = &info->oti_inode;
 674         inode->i_sb = osd_sb(osd);
 675         osd_id_to_inode(inode, id);
 676         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 677
 678         child = &info->oti_child_dentry;
 679         child->d_name.hash = 0;
 680         child->d_name.name = name;
 681         child->d_name.len = strlen(name);
 682         child->d_parent = dir;
 683         child->d_inode = inode;
 684
 685         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
 686                 inode->i_ino++;
 687
 688         ll_vfs_dq_init(dir->d_inode);
 689         mutex_lock(&dir->d_inode->i_mutex);
 690         rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
 691         mutex_unlock(&dir->d_inode->i_mutex);
 692
 693         RETURN(rc);
 694 }
 695
 696 /**
 697  * Use LPU64 for legacy OST sequences, but use LPX64i for new
 698  * sequences names, so that the O/{seq}/dN/{oid} more closely
 699  * follows the DFID/PFID format. This makes it easier to map from
 700  * debug messages to objects in the future, and the legacy space
 701  * of FID_SEQ_OST_MDT0 will be unused in the future.
 702  **/
 703 static inline void osd_seq_name(char *seq_name, size_t name_size, obd_seq seq)
 704 {
 705         snprintf(seq_name, name_size,
 706                  (fid_seq_is_rsvd(seq) ||
 707                   fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
 708                  fid_seq_is_idif(seq) ? 0 : seq);
 709 }
 710
 711 static inline void osd_oid_name(char *name, size_t name_size,
 712                                 const struct lu_fid *fid, obd_id id)
 713 {
 714         snprintf(name, name_size,
 715                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 716                   fid_seq_is_mdt0(fid_seq(fid)) ||
 717                   fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
 718 }
 719
 720 /* external locking is required */
 721 static int osd_seq_load_locked(struct osd_thread_info *info,
 722                                struct osd_device *osd,
 723                                struct osd_obj_seq *osd_seq)
 724 {
 725         struct osd_obj_map  *map = osd->od_ost_map;
 726         struct dentry       *seq_dir;
 727         struct inode        *inode;
 728         struct lu_fid       *fid = &info->oti_fid3;
 729         int                 rc = 0;
 730         int                 i;
 731         char                dir_name[32];
 732         ENTRY;
 733
 734         if (osd_seq->oos_root != NULL)
 735                 RETURN(0);
 736
 737         LASSERT(map);
 738         LASSERT(map->om_root);
 739
 740         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 741
 742         seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
 743         if (IS_ERR(seq_dir))
 744                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 745         else if (seq_dir->d_inode == NULL)
 746                 GOTO(out_put, rc = -EFAULT);
 747
 748         inode = seq_dir->d_inode;
 749         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
 750         osd_seq->oos_root = seq_dir;
 751
 752         /* 'What the @fid is' is not imporatant, because the object
 753          * has no OI mapping, and only is visible inside the OSD.*/
 754         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 755         rc = osd_ea_fid_set(info, inode, fid,
 756                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 757         if (rc != 0)
 758                 GOTO(out_put, rc);
 759
 760         LASSERT(osd_seq->oos_dirs == NULL);
 761         OBD_ALLOC(osd_seq->oos_dirs,
 762                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 763         if (osd_seq->oos_dirs == NULL)
 764                 GOTO(out_put, rc = -ENOMEM);
 765
 766         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 767                 struct dentry   *dir;
 768
 769                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 770                 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
 771                                    0700, 1);
 772                 if (IS_ERR(dir)) {
 773                         GOTO(out_free, rc = PTR_ERR(dir));
 774                 } else if (dir->d_inode == NULL) {
 775                         dput(dir);
 776                         GOTO(out_free, rc = -EFAULT);
 777                 }
 778
 779                 inode = dir->d_inode;
 780                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
 781                 osd_seq->oos_dirs[i] = dir;
 782
 783                 /* 'What the @fid is' is not imporatant, because the object
 784                  * has no OI mapping, and only is visible inside the OSD.*/
 785                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
 786                 rc = osd_ea_fid_set(info, inode, fid,
 787                                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
 788                 if (rc != 0)
 789                         GOTO(out_free, rc);
 790         }
 791
 792         if (rc != 0) {
 793 out_free:
 794                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 795                         if (osd_seq->oos_dirs[i] != NULL)
 796                                 dput(osd_seq->oos_dirs[i]);
 797                 }
 798                 OBD_FREE(osd_seq->oos_dirs,
 799                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
 800 out_put:
 801                 dput(seq_dir);
 802                 osd_seq->oos_root = NULL;
 803         }
 804 out_err:
 805         RETURN(rc);
 806 }
 807
 808 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
 809                                         struct osd_device *osd, obd_seq seq)
 810 {
 811         struct osd_obj_map      *map;
 812         struct osd_obj_seq      *osd_seq;
 813         int                     rc = 0;
 814         ENTRY;
 815
 816         map = osd->od_ost_map;
 817         LASSERT(map);
 818         LASSERT(map->om_root);
 819
 820         osd_seq = osd_seq_find(map, seq);
 821         if (likely(osd_seq != NULL))
 822                 RETURN(osd_seq);
 823
 824         /* Serializing init process */
 825         down(&map->om_dir_init_sem);
 826
 827         /* Check whether the seq has been added */
 828         read_lock(&map->om_seq_list_lock);
 829         osd_seq = osd_seq_find_locked(map, seq);
 830         if (osd_seq != NULL) {
 831                 read_unlock(&map->om_seq_list_lock);
 832                 GOTO(cleanup, rc = 0);
 833         }
 834         read_unlock(&map->om_seq_list_lock);
 835
 836         OBD_ALLOC_PTR(osd_seq);
 837         if (osd_seq == NULL)
 838                 GOTO(cleanup, rc = -ENOMEM);
 839
 840         CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 841         osd_seq->oos_seq = seq;
 842         /* Init subdir count to be 32, but each seq can have
 843          * different subdir count */
 844         osd_seq->oos_subdir_count = map->om_subdir_count;
 845         rc = osd_seq_load_locked(info, osd, osd_seq);
 846         if (rc != 0)
 847                 GOTO(cleanup, rc);
 848
 849         write_lock(&map->om_seq_list_lock);
 850         cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 851         write_unlock(&map->om_seq_list_lock);
 852
 853 cleanup:
 854         up(&map->om_dir_init_sem);
 855         if (rc != 0) {
 856                 if (osd_seq != NULL)
 857                         OBD_FREE_PTR(osd_seq);
 858                 RETURN(ERR_PTR(rc));
 859         }
 860
 861         RETURN(osd_seq);
 862 }
 863
 864 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 865                        const struct lu_fid *fid, struct osd_inode_id *id)
 866 {
 867         struct osd_obj_map              *map;
 868         struct osd_obj_seq              *osd_seq;
 869         struct dentry                   *d_seq;
 870         struct dentry                   *child;
 871         struct ost_id                   *ostid = &info->oti_ostid;
 872         int                             dirn;
 873         char                            name[32];
 874         struct ldiskfs_dir_entry_2      *de;
 875         struct buffer_head              *bh;
 876         struct inode                    *dir;
 877         struct inode                    *inode;
 878         ENTRY;
 879
 880         /* on the very first lookup we find and open directories */
 881
 882         map = dev->od_ost_map;
 883         LASSERT(map);
 884         LASSERT(map->om_root);
 885
 886         fid_to_ostid(fid, ostid);
 887         osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
 888         if (IS_ERR(osd_seq))
 889                 RETURN(PTR_ERR(osd_seq));
 890
 891         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 892         d_seq = osd_seq->oos_dirs[dirn];
 893         LASSERT(d_seq);
 894
 895         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 896
 897         child = &info->oti_child_dentry;
 898         child->d_parent = d_seq;
 899         child->d_name.hash = 0;
 900         child->d_name.name = name;
 901         /* XXX: we can use rc from sprintf() instead of strlen() */
 902         child->d_name.len = strlen(name);
 903
 904         dir = d_seq->d_inode;
 905         mutex_lock(&dir->i_mutex);
 906         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 907         mutex_unlock(&dir->i_mutex);
 908
 909         if (bh == NULL)
 910                 RETURN(-ENOENT);
 911
 912         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 913         brelse(bh);
 914
 915         inode = osd_iget(info, dev, id);
 916         if (IS_ERR(inode))
 917                 RETURN(PTR_ERR(inode));
 918
 919         iput(inode);
 920         RETURN(0);
 921 }
 922
 923 int osd_obj_map_insert(struct osd_thread_info *info,
 924                        struct osd_device *osd,
 925                        const struct lu_fid *fid,
 926                        const struct osd_inode_id *id,
 927                        struct thandle *th)
 928 {
 929         struct osd_obj_map      *map;
 930         struct osd_obj_seq      *osd_seq;
 931         struct dentry           *d;
 932         struct ost_id           *ostid = &info->oti_ostid;
 933         obd_id                   oid;
 934         int                     dirn, rc = 0;
 935         char                    name[32];
 936         ENTRY;
 937
 938         map = osd->od_ost_map;
 939         LASSERT(map);
 940
 941         /* map fid to seq:objid */
 942         fid_to_ostid(fid, ostid);
 943
 944         oid = ostid_id(ostid);
 945         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
 946         if (IS_ERR(osd_seq))
 947                 RETURN(PTR_ERR(osd_seq));
 948
 949         dirn = oid & (osd_seq->oos_subdir_count - 1);
 950         d = osd_seq->oos_dirs[dirn];
 951         LASSERT(d);
 952
 953         osd_oid_name(name, sizeof(name), fid, oid);
 954
 955 again:
 956         rc = osd_obj_add_entry(info, osd, d, name, id, th);
 957         if (rc == -EEXIST) {
 958                 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
 959                 if (unlikely(rc == -ENOENT))
 960                         goto again;
 961
 962                 if (unlikely(rc == 1))
 963                         RETURN(0);
 964         }
 965
 966         RETURN(rc);
 967 }
 968
 969 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
 970                        const struct lu_fid *fid, struct thandle *th)
 971 {
 972         struct osd_obj_map      *map;
 973         struct osd_obj_seq      *osd_seq;
 974         struct dentry           *d;
 975         struct ost_id           *ostid = &info->oti_ostid;
 976         int                     dirn, rc = 0;
 977         char                    name[32];
 978         ENTRY;
 979
 980         map = osd->od_ost_map;
 981         LASSERT(map);
 982
 983         /* map fid to seq:objid */
 984         fid_to_ostid(fid, ostid);
 985
 986         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
 987         if (IS_ERR(osd_seq))
 988                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
 989
 990         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 991         d = osd_seq->oos_dirs[dirn];
 992         LASSERT(d);
 993
 994         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 995         rc = osd_obj_del_entry(info, osd, d, name, th);
 996 cleanup:
 997         RETURN(rc);
 998 }
 999
1000 int osd_obj_map_update(struct osd_thread_info *info,
1001                        struct osd_device *osd,
1002                        const struct lu_fid *fid,
1003                        const struct osd_inode_id *id,
1004                        struct thandle *th)
1005 {
1006         struct osd_obj_seq      *osd_seq;
1007         struct dentry           *d;
1008         struct ost_id           *ostid = &info->oti_ostid;
1009         int                     dirn, rc = 0;
1010         char                    name[32];
1011         ENTRY;
1012
1013         fid_to_ostid(fid, ostid);
1014         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1015         if (IS_ERR(osd_seq))
1016                 RETURN(PTR_ERR(osd_seq));
1017
1018         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1019         d = osd_seq->oos_dirs[dirn];
1020         LASSERT(d);
1021
1022         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1023         rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1024
1025         RETURN(rc);
1026 }
1027
1028 static struct dentry *
1029 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1030                      const struct lu_fid *fid, char **name)
1031 {
1032         struct dentry *root = ERR_PTR(-ENOENT);
1033
1034         if (fid_is_last_id(fid)) {
1035                 struct osd_obj_seq *osd_seq;
1036
1037                 /* on creation of LAST_ID we create O/<seq> hierarchy */
1038                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1039                 if (IS_ERR(osd_seq))
1040                         RETURN((struct dentry *)osd_seq);
1041
1042                 *name = "LAST_ID";
1043                 root = osd_seq->oos_root;
1044         } else {
1045                 *name = osd_lf_fid2name(fid);
1046                 if (*name == NULL)
1047                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1048                 else if ((*name)[0])
1049                         root = osd_sb(osd)->s_root;
1050         }
1051
1052         return root;
1053 }
1054
1055 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1056                         const struct lu_fid *fid, const struct osd_inode_id *id,
1057                         struct thandle *th)
1058 {
1059         struct dentry   *root;
1060         char            *name;
1061         int              rc;
1062         ENTRY;
1063
1064         root = osd_object_spec_find(info, osd, fid, &name);
1065         if (!IS_ERR(root)) {
1066                 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1067         } else {
1068                 rc = PTR_ERR(root);
1069                 if (rc == -ENOENT)
1070                         rc = 1;
1071         }
1072
1073         RETURN(rc);
1074 }
1075
1076 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1077                         const struct lu_fid *fid, const struct osd_inode_id *id,
1078                         struct thandle *th)
1079 {
1080         struct dentry   *root;
1081         char            *name;
1082         int              rc;
1083         ENTRY;
1084
1085         root = osd_object_spec_find(info, osd, fid, &name);
1086         if (!IS_ERR(root)) {
1087                 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1088         } else {
1089                 rc = PTR_ERR(root);
1090                 if (rc == -ENOENT)
1091                         rc = 0;
1092         }
1093
1094         RETURN(rc);
1095 }
1096
1097 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1098                         const struct lu_fid *fid, struct osd_inode_id *id)
1099 {
1100         struct dentry   *root;
1101         struct dentry   *dentry;
1102         struct inode    *inode;
1103         char            *name;
1104         int             rc = -ENOENT;
1105         ENTRY;
1106
1107         if (fid_is_last_id(fid)) {
1108                 struct osd_obj_seq *osd_seq;
1109
1110                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1111                 if (IS_ERR(osd_seq))
1112                         RETURN(PTR_ERR(osd_seq));
1113                 root = osd_seq->oos_root;
1114                 name = "LAST_ID";
1115         } else {
1116                 root = osd_sb(osd)->s_root;
1117                 name = osd_lf_fid2name(fid);
1118                 if (name == NULL || strlen(name) == 0)
1119                         RETURN(-ENOENT);
1120         }
1121
1122         dentry = ll_lookup_one_len(name, root, strlen(name));
1123         if (!IS_ERR(dentry)) {
1124                 inode = dentry->d_inode;
1125                 if (inode) {
1126                         if (is_bad_inode(inode)) {
1127                                 rc = -EIO;
1128                         } else {
1129                                 osd_id_gen(id, inode->i_ino,
1130                                            inode->i_generation);
1131                                 rc = 0;
1132                         }
1133                 }
1134                 /* if dentry is accessible after osd_compat_spec_insert it
1135                  * will still contain NULL inode, so don't keep it in cache */
1136                 d_invalidate(dentry);
1137                 dput(dentry);
1138         }
1139
1140         RETURN(rc);
1141 }