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 /*
 323  * directory structure on legacy OST:
 324  *
 325  * O/<seq>/d0-31/<objid>
 326  * O/<seq>/LAST_ID
 327  * last_rcvd
 328  * LAST_GROUP
 329  * CONFIGS
 330  *
 331  */
 332 static int osd_ost_init(struct osd_device *dev)
 333 {
 334         struct lvfs_run_ctxt  new;
 335         struct lvfs_run_ctxt  save;
 336         struct dentry        *rootd = osd_sb(dev)->s_root;
 337         struct dentry        *d;
 338         int                   rc;
 339         ENTRY;
 340
 341         OBD_ALLOC_PTR(dev->od_ost_map);
 342         if (dev->od_ost_map == NULL)
 343                 RETURN(-ENOMEM);
 344
 345         /* to get subdir count from last_rcvd */
 346         rc = osd_last_rcvd_subdir_count(dev);
 347         if (rc < 0) {
 348                 OBD_FREE_PTR(dev->od_ost_map);
 349                 RETURN(rc);
 350         }
 351
 352         dev->od_ost_map->om_subdir_count = rc;
 353         rc = 0;
 354
 355         CFS_INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 356         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
 357         sema_init(&dev->od_ost_map->om_dir_init_sem, 1);
 358
 359         LASSERT(dev->od_fsops);
 360         osd_push_ctxt(dev, &new, &save);
 361
 362         d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
 363         if (IS_ERR(d))
 364                 GOTO(cleanup, rc = PTR_ERR(d));
 365
 366         ldiskfs_set_inode_state(d->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 367         dev->od_ost_map->om_root = d;
 368
 369 cleanup:
 370         osd_pop_ctxt(dev, &new, &save);
 371         if (IS_ERR(d)) {
 372                 OBD_FREE_PTR(dev->od_ost_map);
 373                 RETURN(PTR_ERR(d));
 374         }
 375
 376         RETURN(rc);
 377 }
 378
 379 static void osd_seq_free(struct osd_obj_map *map,
 380                          struct osd_obj_seq *osd_seq)
 381 {
 382         int j;
 383
 384         cfs_list_del_init(&osd_seq->oos_seq_list);
 385
 386         if (osd_seq->oos_dirs) {
 387                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 388                         if (osd_seq->oos_dirs[j])
 389                                 dput(osd_seq->oos_dirs[j]);
 390                 }
 391                 OBD_FREE(osd_seq->oos_dirs,
 392                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
 393         }
 394
 395         if (osd_seq->oos_root)
 396                 dput(osd_seq->oos_root);
 397
 398         OBD_FREE_PTR(osd_seq);
 399 }
 400
 401 static void osd_ost_fini(struct osd_device *osd)
 402 {
 403         struct osd_obj_seq    *osd_seq;
 404         struct osd_obj_seq    *tmp;
 405         struct osd_obj_map    *map = osd->od_ost_map;
 406         ENTRY;
 407
 408         if (map == NULL)
 409                 return;
 410
 411         write_lock(&map->om_seq_list_lock);
 412         cfs_list_for_each_entry_safe(osd_seq, tmp,
 413                                      &map->om_seq_list,
 414                                      oos_seq_list) {
 415                 osd_seq_free(map, osd_seq);
 416         }
 417         write_unlock(&map->om_seq_list_lock);
 418         if (map->om_root)
 419                 dput(map->om_root);
 420         OBD_FREE_PTR(map);
 421         osd->od_ost_map = NULL;
 422         EXIT;
 423 }
 424
 425 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 426 {
 427         int rc;
 428         ENTRY;
 429
 430         /* prepare structures for OST */
 431         rc = osd_ost_init(dev);
 432         if (rc)
 433                 RETURN(rc);
 434
 435         /* prepare structures for MDS */
 436         rc = osd_mdt_init(env, dev);
 437
 438         RETURN(rc);
 439 }
 440
 441 struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq)
 442 {
 443         struct osd_obj_seq *osd_seq;
 444
 445         cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 446                 if (osd_seq->oos_seq == seq)
 447                         return osd_seq;
 448         }
 449         return NULL;
 450 }
 451
 452 struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
 453 {
 454         struct osd_obj_seq *osd_seq;
 455
 456         read_lock(&map->om_seq_list_lock);
 457         osd_seq = osd_seq_find_locked(map, seq);
 458         read_unlock(&map->om_seq_list_lock);
 459         return osd_seq;
 460 }
 461
 462 void osd_obj_map_fini(struct osd_device *dev)
 463 {
 464         osd_ost_fini(dev);
 465         osd_mdt_fini(dev);
 466 }
 467
 468 static int osd_obj_del_entry(struct osd_thread_info *info,
 469                              struct osd_device *osd,
 470                              struct dentry *dird, char *name,
 471                              struct thandle *th)
 472 {
 473         struct ldiskfs_dir_entry_2 *de;
 474         struct buffer_head         *bh;
 475         struct osd_thandle         *oh;
 476         struct dentry              *child;
 477         struct inode               *dir = dird->d_inode;
 478         int                         rc;
 479         ENTRY;
 480
 481         oh = container_of(th, struct osd_thandle, ot_super);
 482         LASSERT(oh->ot_handle != NULL);
 483         LASSERT(oh->ot_handle->h_transaction != NULL);
 484
 485
 486         child = &info->oti_child_dentry;
 487         child->d_name.hash = 0;
 488         child->d_name.name = name;
 489         child->d_name.len = strlen(name);
 490         child->d_parent = dird;
 491         child->d_inode = NULL;
 492
 493         ll_vfs_dq_init(dir);
 494         mutex_lock(&dir->i_mutex);
 495         rc = -ENOENT;
 496         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 497         if (bh) {
 498                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
 499                 brelse(bh);
 500         }
 501         mutex_unlock(&dir->i_mutex);
 502
 503         RETURN(rc);
 504 }
 505
 506 int osd_obj_add_entry(struct osd_thread_info *info,
 507                       struct osd_device *osd,
 508                       struct dentry *dir, char *name,
 509                       const struct osd_inode_id *id,
 510                       struct thandle *th)
 511 {
 512         struct osd_thandle *oh;
 513         struct dentry *child;
 514         struct inode *inode;
 515         int rc;
 516
 517         ENTRY;
 518
 519         oh = container_of(th, struct osd_thandle, ot_super);
 520         LASSERT(oh->ot_handle != NULL);
 521         LASSERT(oh->ot_handle->h_transaction != NULL);
 522
 523         inode = &info->oti_inode;
 524         inode->i_sb = osd_sb(osd);
 525         osd_id_to_inode(inode, id);
 526         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 527
 528         child = &info->oti_child_dentry;
 529         child->d_name.hash = 0;
 530         child->d_name.name = name;
 531         child->d_name.len = strlen(name);
 532         child->d_parent = dir;
 533         child->d_inode = inode;
 534
 535         ll_vfs_dq_init(dir->d_inode);
 536         mutex_lock(&dir->d_inode->i_mutex);
 537         rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
 538         mutex_unlock(&dir->d_inode->i_mutex);
 539
 540         RETURN(rc);
 541 }
 542
 543 /**
 544  * Use LPU64 for legacy OST sequences, but use LPX64i for new
 545  * sequences names, so that the O/{seq}/dN/{oid} more closely
 546  * follows the DFID/PFID format. This makes it easier to map from
 547  * debug messages to objects in the future, and the legacy space
 548  * of FID_SEQ_OST_MDT0 will be unused in the future.
 549  **/
 550 static inline void osd_seq_name(char *seq_name, size_t name_size, obd_seq seq)
 551 {
 552         snprintf(seq_name, name_size,
 553                  (fid_seq_is_rsvd(seq) ||
 554                   fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
 555                  fid_seq_is_idif(seq) ? 0 : seq);
 556 }
 557
 558 static inline void osd_oid_name(char *name, size_t name_size,
 559                                 const struct lu_fid *fid, obd_id id)
 560 {
 561         snprintf(name, name_size,
 562                  (fid_seq_is_rsvd(fid_seq(fid)) ||
 563                   fid_seq_is_mdt0(fid_seq(fid)) ||
 564                   fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
 565 }
 566
 567 /* external locking is required */
 568 static int osd_seq_load_locked(struct osd_device *osd,
 569                                struct osd_obj_seq *osd_seq)
 570 {
 571         struct osd_obj_map  *map = osd->od_ost_map;
 572         struct dentry       *seq_dir;
 573         int                 rc = 0;
 574         int                 i;
 575         char                dir_name[32];
 576         ENTRY;
 577
 578         if (osd_seq->oos_root != NULL)
 579                 RETURN(0);
 580
 581         LASSERT(map);
 582         LASSERT(map->om_root);
 583
 584         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 585
 586         seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
 587         if (IS_ERR(seq_dir))
 588                 GOTO(out_err, rc = PTR_ERR(seq_dir));
 589         else if (seq_dir->d_inode == NULL)
 590                 GOTO(out_put, rc = -EFAULT);
 591
 592         ldiskfs_set_inode_state(seq_dir->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 593         osd_seq->oos_root = seq_dir;
 594
 595         LASSERT(osd_seq->oos_dirs == NULL);
 596         OBD_ALLOC(osd_seq->oos_dirs,
 597                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
 598         if (osd_seq->oos_dirs == NULL)
 599                 GOTO(out_put, rc = -ENOMEM);
 600
 601         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 602                 struct dentry   *dir;
 603
 604                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
 605                 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
 606                                    0700, 1);
 607                 if (IS_ERR(dir)) {
 608                         GOTO(out_free, rc = PTR_ERR(dir));
 609                 } else if (dir->d_inode == NULL) {
 610                         dput(dir);
 611                         GOTO(out_free, rc = -EFAULT);
 612                 }
 613
 614                 ldiskfs_set_inode_state(dir->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 615                 osd_seq->oos_dirs[i] = dir;
 616         }
 617
 618         if (rc != 0) {
 619 out_free:
 620                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 621                         if (osd_seq->oos_dirs[i] != NULL)
 622                                 dput(osd_seq->oos_dirs[i]);
 623                 }
 624                 OBD_FREE(osd_seq->oos_dirs,
 625                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
 626 out_put:
 627                 dput(seq_dir);
 628                 osd_seq->oos_root = NULL;
 629         }
 630 out_err:
 631         RETURN(rc);
 632 }
 633
 634 struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq)
 635 {
 636         struct osd_obj_map      *map;
 637         struct osd_obj_seq      *osd_seq;
 638         int                     rc = 0;
 639         ENTRY;
 640
 641         map = osd->od_ost_map;
 642         LASSERT(map);
 643         LASSERT(map->om_root);
 644
 645         osd_seq = osd_seq_find(map, seq);
 646         if (likely(osd_seq != NULL))
 647                 RETURN(osd_seq);
 648
 649         /* Serializing init process */
 650         down(&map->om_dir_init_sem);
 651
 652         /* Check whether the seq has been added */
 653         read_lock(&map->om_seq_list_lock);
 654         osd_seq = osd_seq_find_locked(map, seq);
 655         if (osd_seq != NULL) {
 656                 read_unlock(&map->om_seq_list_lock);
 657                 GOTO(cleanup, rc = 0);
 658         }
 659         read_unlock(&map->om_seq_list_lock);
 660
 661         OBD_ALLOC_PTR(osd_seq);
 662         if (osd_seq == NULL)
 663                 GOTO(cleanup, rc = -ENOMEM);
 664
 665         CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 666         osd_seq->oos_seq = seq;
 667         /* Init subdir count to be 32, but each seq can have
 668          * different subdir count */
 669         osd_seq->oos_subdir_count = map->om_subdir_count;
 670         rc = osd_seq_load_locked(osd, osd_seq);
 671         if (rc != 0)
 672                 GOTO(cleanup, rc);
 673
 674         write_lock(&map->om_seq_list_lock);
 675         cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 676         write_unlock(&map->om_seq_list_lock);
 677
 678 cleanup:
 679         up(&map->om_dir_init_sem);
 680         if (rc != 0) {
 681                 if (osd_seq != NULL)
 682                         OBD_FREE_PTR(osd_seq);
 683                 RETURN(ERR_PTR(rc));
 684         }
 685
 686         RETURN(osd_seq);
 687 }
 688
 689 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 690                        const struct lu_fid *fid, struct osd_inode_id *id)
 691 {
 692         struct osd_obj_map              *map;
 693         struct osd_obj_seq              *osd_seq;
 694         struct dentry                   *d_seq;
 695         struct dentry                   *child;
 696         struct ost_id                   *ostid = &info->oti_ostid;
 697         int                             dirn;
 698         char                            name[32];
 699         struct ldiskfs_dir_entry_2      *de;
 700         struct buffer_head              *bh;
 701         struct inode                    *dir;
 702         struct inode                    *inode;
 703         ENTRY;
 704
 705         /* on the very first lookup we find and open directories */
 706
 707         map = dev->od_ost_map;
 708         LASSERT(map);
 709         LASSERT(map->om_root);
 710
 711         fid_to_ostid(fid, ostid);
 712         osd_seq = osd_seq_load(dev, ostid_seq(ostid));
 713         if (IS_ERR(osd_seq))
 714                 RETURN(PTR_ERR(osd_seq));
 715
 716         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 717         d_seq = osd_seq->oos_dirs[dirn];
 718         LASSERT(d_seq);
 719
 720         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 721
 722         child = &info->oti_child_dentry;
 723         child->d_parent = d_seq;
 724         child->d_name.hash = 0;
 725         child->d_name.name = name;
 726         /* XXX: we can use rc from sprintf() instead of strlen() */
 727         child->d_name.len = strlen(name);
 728
 729         dir = d_seq->d_inode;
 730         mutex_lock(&dir->i_mutex);
 731         bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
 732         mutex_unlock(&dir->i_mutex);
 733
 734         if (bh == NULL)
 735                 RETURN(-ENOENT);
 736
 737         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 738         brelse(bh);
 739
 740         inode = osd_iget(info, dev, id);
 741         if (IS_ERR(inode))
 742                 RETURN(PTR_ERR(inode));
 743
 744         iput(inode);
 745         RETURN(0);
 746 }
 747
 748 int osd_obj_map_insert(struct osd_thread_info *info,
 749                        struct osd_device *osd,
 750                        const struct lu_fid *fid,
 751                        const struct osd_inode_id *id,
 752                        struct thandle *th)
 753 {
 754         struct osd_obj_map      *map;
 755         struct osd_obj_seq      *osd_seq;
 756         struct dentry           *d;
 757         struct ost_id           *ostid = &info->oti_ostid;
 758         int                     dirn, rc = 0;
 759         char                    name[32];
 760         ENTRY;
 761
 762         map = osd->od_ost_map;
 763         LASSERT(map);
 764
 765         /* map fid to seq:objid */
 766         fid_to_ostid(fid, ostid);
 767
 768         osd_seq = osd_seq_load(osd, ostid_seq(ostid));
 769         if (IS_ERR(osd_seq))
 770                 RETURN(PTR_ERR(osd_seq));
 771
 772         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 773         d = osd_seq->oos_dirs[dirn];
 774         LASSERT(d);
 775
 776         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 777         rc = osd_obj_add_entry(info, osd, d, name, id, th);
 778
 779         RETURN(rc);
 780 }
 781
 782 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
 783                        const struct lu_fid *fid, struct thandle *th)
 784 {
 785         struct osd_obj_map      *map;
 786         struct osd_obj_seq      *osd_seq;
 787         struct dentry           *d;
 788         struct ost_id           *ostid = &info->oti_ostid;
 789         int                     dirn, rc = 0;
 790         char                    name[32];
 791         ENTRY;
 792
 793         map = osd->od_ost_map;
 794         LASSERT(map);
 795
 796         /* map fid to seq:objid */
 797         fid_to_ostid(fid, ostid);
 798
 799         osd_seq = osd_seq_load(osd, ostid_seq(ostid));
 800         if (IS_ERR(osd_seq))
 801                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
 802
 803         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 804         d = osd_seq->oos_dirs[dirn];
 805         LASSERT(d);
 806
 807         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 808         rc = osd_obj_del_entry(info, osd, d, name, th);
 809 cleanup:
 810         RETURN(rc);
 811 }
 812
 813 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
 814                         const struct lu_fid *fid,
 815                         const struct osd_inode_id *id,
 816                         struct thandle *th)
 817 {
 818         struct osd_obj_map      *map = osd->od_ost_map;
 819         struct dentry           *root = osd_sb(osd)->s_root;
 820         char                    *name;
 821         int                     rc = 0;
 822         ENTRY;
 823
 824         if (fid_is_last_id(fid)) {
 825                 struct osd_obj_seq      *osd_seq;
 826
 827                 /* on creation of LAST_ID we create O/<seq> hierarchy */
 828                 LASSERT(map);
 829                 osd_seq = osd_seq_load(osd, fid_seq(fid));
 830                 if (IS_ERR(osd_seq))
 831                         RETURN(PTR_ERR(osd_seq));
 832                 rc = osd_obj_add_entry(info, osd, osd_seq->oos_root,
 833                                        "LAST_ID", id, th);
 834         } else {
 835                 name = osd_lf_fid2name(fid);
 836                 if (name == NULL)
 837                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
 838                 else if (name[0])
 839                         rc = osd_obj_add_entry(info, osd, root, name, id, th);
 840         }
 841
 842         RETURN(rc);
 843 }
 844
 845 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
 846                         const struct lu_fid *fid, struct osd_inode_id *id)
 847 {
 848         struct dentry   *root;
 849         struct dentry   *dentry;
 850         struct inode    *inode;
 851         char            *name;
 852         int             rc = -ENOENT;
 853         ENTRY;
 854
 855         if (fid_is_last_id(fid)) {
 856                 struct osd_obj_seq *osd_seq;
 857
 858                 osd_seq = osd_seq_load(osd, fid_seq(fid));
 859                 if (IS_ERR(osd_seq))
 860                         RETURN(PTR_ERR(osd_seq));
 861                 root = osd_seq->oos_root;
 862                 name = "LAST_ID";
 863         } else {
 864                 root = osd_sb(osd)->s_root;
 865                 name = osd_lf_fid2name(fid);
 866                 if (name == NULL || strlen(name) == 0)
 867                         RETURN(-ENOENT);
 868         }
 869
 870         dentry = ll_lookup_one_len(name, root, strlen(name));
 871         if (!IS_ERR(dentry)) {
 872                 inode = dentry->d_inode;
 873                 if (inode) {
 874                         if (is_bad_inode(inode)) {
 875                                 rc = -EIO;
 876                         } else {
 877                                 osd_id_gen(id, inode->i_ino,
 878                                            inode->i_generation);
 879                                 rc = 0;
 880                         }
 881                 }
 882                 /* if dentry is accessible after osd_compat_spec_insert it
 883                  * will still contain NULL inode, so don't keep it in cache */
 884                 d_invalidate(dentry);
 885                 dput(dentry);
 886         }
 887
 888         RETURN(rc);
 889 }