4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_compat.c
38 * on-disk structure for managing /O
40 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
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 */
49 /* XATTR_{REPLACE,CREATE} */
50 #include <linux/xattr.h>
53 * struct OBD_{ALLOC,FREE}*()
56 #include <obd_support.h>
58 #include "osd_internal.h"
61 static void osd_push_ctxt(const struct osd_device *dev,
62 struct lvfs_run_ctxt *newctxt,
63 struct lvfs_run_ctxt *save)
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();
70 push_ctxt(save, newctxt, NULL);
73 static void osd_pop_ctxt(const struct osd_device *dev,
74 struct lvfs_run_ctxt *new,
75 struct lvfs_run_ctxt *save)
77 pop_ctxt(save, new, NULL);
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)
84 struct dentry *dchild;
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));
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);
103 /* Fixup directory permissions if necessary */
104 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
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);
112 GOTO(out_up, dchild);
115 err = ll_vfs_mkdir(dir->d_inode, dchild, mnt, mode);
123 dchild = ERR_PTR(err);
128 int osd_last_rcvd_subdir_count(struct osd_device *osd)
130 struct lr_server_data lsd;
131 struct dentry *dlast;
134 int count = FILTER_SUBDIR_COUNT;
138 dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
141 return PTR_ERR(dlast);
142 else if (dlast->d_inode == NULL)
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);
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)
169 struct lvfs_run_ctxt new;
170 struct lvfs_run_ctxt save;
171 struct dentry *parent;
172 struct osd_mdobj_map *omm;
174 struct osd_thread_info *info = osd_oti_get(env);
175 struct lu_fid *fid = &info->oti_fid3;
179 OBD_ALLOC_PTR(dev->od_mdt_map);
180 if (dev->od_mdt_map == NULL)
183 omm = dev->od_mdt_map;
185 LASSERT(dev->od_fsops);
187 parent = osd_sb(dev)->s_root;
188 osd_push_ctxt(dev, &new, &save);
190 d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
193 GOTO(cleanup, rc = PTR_ERR(d));
195 ldiskfs_set_inode_state(d->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
196 omm->omm_remote_parent = d;
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);
205 pop_ctxt(&save, &new, NULL);
207 if (omm->omm_remote_parent != NULL)
208 dput(omm->omm_remote_parent);
210 dev->od_mdt_map = NULL;
215 static void osd_mdt_fini(struct osd_device *osd)
217 struct osd_mdobj_map *omm = osd->od_mdt_map;
222 if (omm->omm_remote_parent)
223 dput(omm->omm_remote_parent);
226 osd->od_ost_map = NULL;
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)
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;
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);
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);
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,
257 mutex_lock(&parent->d_inode->i_mutex);
258 rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
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);
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)
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;
286 /* Check lma to see whether it is remote object */
287 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
291 if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
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,
298 mutex_lock(&parent->d_inode->i_mutex);
299 bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
301 mutex_unlock(&parent->d_inode->i_mutex);
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);
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);
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)
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;
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,
341 mutex_lock(&parent->d_inode->i_mutex);
342 bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
347 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
350 mutex_unlock(&parent->d_inode->i_mutex);
352 osd_add_oi_cache(oti, osd, id, fid);
357 * directory structure on legacy OST:
359 * O/<seq>/d0-31/<objid>
366 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
368 struct lvfs_run_ctxt new;
369 struct lvfs_run_ctxt save;
370 struct dentry *rootd = osd_sb(dev)->s_root;
372 struct osd_thread_info *info = osd_oti_get(env);
374 struct lu_fid *fid = &info->oti_fid3;
378 OBD_ALLOC_PTR(dev->od_ost_map);
379 if (dev->od_ost_map == NULL)
382 /* to get subdir count from last_rcvd */
383 rc = osd_last_rcvd_subdir_count(dev);
385 OBD_FREE_PTR(dev->od_ost_map);
389 dev->od_ost_map->om_subdir_count = rc;
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);
396 LASSERT(dev->od_fsops);
397 osd_push_ctxt(dev, &new, &save);
399 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
401 GOTO(cleanup, rc = PTR_ERR(d));
404 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
405 dev->od_ost_map->om_root = d;
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);
416 osd_pop_ctxt(dev, &new, &save);
418 OBD_FREE_PTR(dev->od_ost_map);
424 static void osd_seq_free(struct osd_obj_map *map,
425 struct osd_obj_seq *osd_seq)
429 cfs_list_del_init(&osd_seq->oos_seq_list);
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]);
436 OBD_FREE(osd_seq->oos_dirs,
437 sizeof(struct dentry *) * osd_seq->oos_subdir_count);
440 if (osd_seq->oos_root)
441 dput(osd_seq->oos_root);
443 OBD_FREE_PTR(osd_seq);
446 static void osd_ost_fini(struct osd_device *osd)
448 struct osd_obj_seq *osd_seq;
449 struct osd_obj_seq *tmp;
450 struct osd_obj_map *map = osd->od_ost_map;
456 write_lock(&map->om_seq_list_lock);
457 cfs_list_for_each_entry_safe(osd_seq, tmp,
460 osd_seq_free(map, osd_seq);
462 write_unlock(&map->om_seq_list_lock);
466 osd->od_ost_map = NULL;
470 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
475 /* prepare structures for OST */
476 rc = osd_ost_init(env, dev);
480 /* prepare structures for MDS */
481 rc = osd_mdt_init(env, dev);
486 struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq)
488 struct osd_obj_seq *osd_seq;
490 cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
491 if (osd_seq->oos_seq == seq)
497 struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
499 struct osd_obj_seq *osd_seq;
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);
507 void osd_obj_map_fini(struct osd_device *dev)
514 * Update the specified OI mapping.
516 * \retval 1, changed nothing
517 * \retval 0, changed successfully
518 * \retval -ve, on error
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,
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;
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;
540 oh = container_of(th, struct osd_thandle, ot_super);
541 LASSERT(oh->ot_handle != NULL);
542 LASSERT(oh->ot_handle->h_transaction != NULL);
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);
550 ll_vfs_dq_init(parent);
551 mutex_lock(&parent->i_mutex);
552 bh = osd_ldiskfs_find_entry(parent, child, &de, NULL);
554 GOTO(out, rc = -ENOENT);
556 if (le32_to_cpu(de->inode) == id->oii_ino)
559 osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
560 inode = osd_iget(info, osd, oi_id);
563 if (rc == -ENOENT || rc == -ESTALE)
568 rc = osd_get_lma(info, inode, dentry, lma);
569 if (rc == -ENODATA) {
570 rc = osd_get_idif(info, inode, dentry, oi_fid);
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)
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);
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);
607 de->inode = cpu_to_le32(id->oii_ino);
608 rc = ldiskfs_journal_dirty_metadata(oh->ot_handle, bh);
614 mutex_unlock(&parent->i_mutex);
618 static int osd_obj_del_entry(struct osd_thread_info *info,
619 struct osd_device *osd,
620 struct dentry *dird, char *name,
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;
631 oh = container_of(th, struct osd_thandle, ot_super);
632 LASSERT(oh->ot_handle != NULL);
633 LASSERT(oh->ot_handle->h_transaction != NULL);
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;
644 mutex_lock(&dir->i_mutex);
646 bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
648 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
651 mutex_unlock(&dir->i_mutex);
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,
662 struct osd_thandle *oh;
663 struct dentry *child;
669 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
672 oh = container_of(th, struct osd_thandle, ot_super);
673 LASSERT(oh->ot_handle != NULL);
674 LASSERT(oh->ot_handle->h_transaction != NULL);
676 inode = &info->oti_inode;
677 inode->i_sb = osd_sb(osd);
678 osd_id_to_inode(inode, id);
679 inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
681 child = &info->oti_child_dentry;
682 child->d_name.hash = 0;
683 child->d_name.name = name;
684 child->d_name.len = strlen(name);
685 child->d_parent = dir;
686 child->d_inode = inode;
688 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
691 ll_vfs_dq_init(dir->d_inode);
692 mutex_lock(&dir->d_inode->i_mutex);
693 rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
694 mutex_unlock(&dir->d_inode->i_mutex);
700 * Use LPU64 for legacy OST sequences, but use LPX64i for new
701 * sequences names, so that the O/{seq}/dN/{oid} more closely
702 * follows the DFID/PFID format. This makes it easier to map from
703 * debug messages to objects in the future, and the legacy space
704 * of FID_SEQ_OST_MDT0 will be unused in the future.
706 static inline void osd_seq_name(char *seq_name, size_t name_size, obd_seq seq)
708 snprintf(seq_name, name_size,
709 (fid_seq_is_rsvd(seq) ||
710 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
711 fid_seq_is_idif(seq) ? 0 : seq);
714 static inline void osd_oid_name(char *name, size_t name_size,
715 const struct lu_fid *fid, obd_id id)
717 snprintf(name, name_size,
718 (fid_seq_is_rsvd(fid_seq(fid)) ||
719 fid_seq_is_mdt0(fid_seq(fid)) ||
720 fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
723 /* external locking is required */
724 static int osd_seq_load_locked(struct osd_thread_info *info,
725 struct osd_device *osd,
726 struct osd_obj_seq *osd_seq)
728 struct osd_obj_map *map = osd->od_ost_map;
729 struct dentry *seq_dir;
731 struct lu_fid *fid = &info->oti_fid3;
737 if (osd_seq->oos_root != NULL)
741 LASSERT(map->om_root);
743 osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
745 seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
747 GOTO(out_err, rc = PTR_ERR(seq_dir));
748 else if (seq_dir->d_inode == NULL)
749 GOTO(out_put, rc = -EFAULT);
751 inode = seq_dir->d_inode;
752 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
753 osd_seq->oos_root = seq_dir;
755 /* 'What the @fid is' is not imporatant, because the object
756 * has no OI mapping, and only is visible inside the OSD.*/
757 lu_igif_build(fid, inode->i_ino, inode->i_generation);
758 rc = osd_ea_fid_set(info, inode, fid,
759 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
763 LASSERT(osd_seq->oos_dirs == NULL);
764 OBD_ALLOC(osd_seq->oos_dirs,
765 sizeof(seq_dir) * osd_seq->oos_subdir_count);
766 if (osd_seq->oos_dirs == NULL)
767 GOTO(out_put, rc = -ENOMEM);
769 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
772 snprintf(dir_name, sizeof(dir_name), "d%u", i);
773 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
776 GOTO(out_free, rc = PTR_ERR(dir));
777 } else if (dir->d_inode == NULL) {
779 GOTO(out_free, rc = -EFAULT);
782 inode = dir->d_inode;
783 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
784 osd_seq->oos_dirs[i] = dir;
786 /* 'What the @fid is' is not imporatant, because the object
787 * has no OI mapping, and only is visible inside the OSD.*/
788 lu_igif_build(fid, inode->i_ino, inode->i_generation);
789 rc = osd_ea_fid_set(info, inode, fid,
790 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
797 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
798 if (osd_seq->oos_dirs[i] != NULL)
799 dput(osd_seq->oos_dirs[i]);
801 OBD_FREE(osd_seq->oos_dirs,
802 sizeof(seq_dir) * osd_seq->oos_subdir_count);
805 osd_seq->oos_root = NULL;
811 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
812 struct osd_device *osd, obd_seq seq)
814 struct osd_obj_map *map;
815 struct osd_obj_seq *osd_seq;
819 map = osd->od_ost_map;
821 LASSERT(map->om_root);
823 osd_seq = osd_seq_find(map, seq);
824 if (likely(osd_seq != NULL))
827 /* Serializing init process */
828 down(&map->om_dir_init_sem);
830 /* Check whether the seq has been added */
831 read_lock(&map->om_seq_list_lock);
832 osd_seq = osd_seq_find_locked(map, seq);
833 if (osd_seq != NULL) {
834 read_unlock(&map->om_seq_list_lock);
835 GOTO(cleanup, rc = 0);
837 read_unlock(&map->om_seq_list_lock);
839 OBD_ALLOC_PTR(osd_seq);
841 GOTO(cleanup, rc = -ENOMEM);
843 CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list);
844 osd_seq->oos_seq = seq;
845 /* Init subdir count to be 32, but each seq can have
846 * different subdir count */
847 osd_seq->oos_subdir_count = map->om_subdir_count;
848 rc = osd_seq_load_locked(info, osd, osd_seq);
852 write_lock(&map->om_seq_list_lock);
853 cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
854 write_unlock(&map->om_seq_list_lock);
857 up(&map->om_dir_init_sem);
860 OBD_FREE_PTR(osd_seq);
867 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
868 const struct lu_fid *fid, struct osd_inode_id *id)
870 struct osd_obj_map *map;
871 struct osd_obj_seq *osd_seq;
872 struct dentry *d_seq;
873 struct dentry *child;
874 struct ost_id *ostid = &info->oti_ostid;
877 struct ldiskfs_dir_entry_2 *de;
878 struct buffer_head *bh;
883 /* on the very first lookup we find and open directories */
885 map = dev->od_ost_map;
887 LASSERT(map->om_root);
889 fid_to_ostid(fid, ostid);
890 osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
892 RETURN(PTR_ERR(osd_seq));
894 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
895 d_seq = osd_seq->oos_dirs[dirn];
898 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
900 child = &info->oti_child_dentry;
901 child->d_parent = d_seq;
902 child->d_name.hash = 0;
903 child->d_name.name = name;
904 /* XXX: we can use rc from sprintf() instead of strlen() */
905 child->d_name.len = strlen(name);
907 dir = d_seq->d_inode;
908 mutex_lock(&dir->i_mutex);
909 bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
910 mutex_unlock(&dir->i_mutex);
915 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
918 inode = osd_iget(info, dev, id);
920 RETURN(PTR_ERR(inode));
926 int osd_obj_map_insert(struct osd_thread_info *info,
927 struct osd_device *osd,
928 const struct lu_fid *fid,
929 const struct osd_inode_id *id,
932 struct osd_obj_map *map;
933 struct osd_obj_seq *osd_seq;
935 struct ost_id *ostid = &info->oti_ostid;
941 map = osd->od_ost_map;
944 /* map fid to seq:objid */
945 fid_to_ostid(fid, ostid);
947 oid = ostid_id(ostid);
948 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
950 RETURN(PTR_ERR(osd_seq));
952 dirn = oid & (osd_seq->oos_subdir_count - 1);
953 d = osd_seq->oos_dirs[dirn];
956 osd_oid_name(name, sizeof(name), fid, oid);
959 rc = osd_obj_add_entry(info, osd, d, name, id, th);
961 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
962 if (unlikely(rc == -ENOENT))
965 if (unlikely(rc == 1))
972 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
973 const struct lu_fid *fid, struct thandle *th)
975 struct osd_obj_map *map;
976 struct osd_obj_seq *osd_seq;
978 struct ost_id *ostid = &info->oti_ostid;
983 map = osd->od_ost_map;
986 /* map fid to seq:objid */
987 fid_to_ostid(fid, ostid);
989 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
991 GOTO(cleanup, rc = PTR_ERR(osd_seq));
993 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
994 d = osd_seq->oos_dirs[dirn];
997 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
998 rc = osd_obj_del_entry(info, osd, d, name, th);
1003 int osd_obj_map_update(struct osd_thread_info *info,
1004 struct osd_device *osd,
1005 const struct lu_fid *fid,
1006 const struct osd_inode_id *id,
1009 struct osd_obj_seq *osd_seq;
1011 struct ost_id *ostid = &info->oti_ostid;
1016 fid_to_ostid(fid, ostid);
1017 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1018 if (IS_ERR(osd_seq))
1019 RETURN(PTR_ERR(osd_seq));
1021 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1022 d = osd_seq->oos_dirs[dirn];
1025 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1026 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1031 int osd_obj_map_recover(struct osd_thread_info *info,
1032 struct osd_device *osd,
1033 struct inode *src_parent,
1034 struct dentry *src_child,
1035 const struct lu_fid *fid)
1037 struct osd_obj_seq *osd_seq;
1038 struct dentry *tgt_parent;
1039 struct dentry *tgt_child = &info->oti_child_dentry;
1041 struct inode *inode = src_child->d_inode;
1042 struct ost_id *ostid = &info->oti_ostid;
1044 struct ldiskfs_dir_entry_2 *de;
1045 struct buffer_head *bh;
1051 if (fid_is_last_id(fid)) {
1052 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1053 if (IS_ERR(osd_seq))
1054 RETURN(PTR_ERR(osd_seq));
1056 tgt_parent = osd_seq->oos_root;
1057 tgt_child->d_name.name = "LAST_ID";
1058 tgt_child->d_name.len = strlen("LAST_ID");
1060 fid_to_ostid(fid, ostid);
1061 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1062 if (IS_ERR(osd_seq))
1063 RETURN(PTR_ERR(osd_seq));
1065 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1066 tgt_parent = osd_seq->oos_dirs[dirn];
1067 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1068 tgt_child->d_name.name = name;
1069 tgt_child->d_name.len = strlen(name);
1071 LASSERT(tgt_parent != NULL);
1073 dir = tgt_parent->d_inode;
1074 tgt_child->d_name.hash = 0;
1075 tgt_child->d_parent = tgt_parent;
1076 tgt_child->d_inode = inode;
1078 /* The non-initialized src_child may be destroyed. */
1079 jh = ldiskfs_journal_start_sb(osd_sb(osd),
1080 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1081 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1082 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1084 RETURN(PTR_ERR(jh));
1086 ll_vfs_dq_init(src_parent);
1087 ll_vfs_dq_init(dir);
1089 mutex_lock(&src_parent->i_mutex);
1090 mutex_lock(&dir->i_mutex);
1091 bh = osd_ldiskfs_find_entry(dir, tgt_child, &de, NULL);
1093 /* XXX: If some other object occupied the same slot. And If such
1094 * inode is zero-sized and with SUID+SGID, then means it is
1095 * a new created one. Maybe we can remove it and insert the
1096 * original one back to the /O/<seq>/d<x>. But there are
1097 * something to be considered:
1099 * 1) The OST-object under /lost+found has crashed LMA.
1100 * So it should not conflict with the current one.
1102 * 2) There are race conditions that: someone may just want
1103 * to modify the current one. Even if the OI scrub takes
1104 * the object lock when remove the current one, it still
1105 * cause the modification to be lost becasue the target
1106 * has been removed when the RPC service thread waiting
1109 * So keep it there before we have suitable solution. */
1112 /* If the src object has never been modified, then remove it. */
1113 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1114 inode->i_mode & S_ISGID)
1115 ll_vfs_unlink(src_parent, src_child, osd->od_mnt);
1116 GOTO(unlock_src, rc = 0);
1119 bh = osd_ldiskfs_find_entry(src_parent, src_child, &de, NULL);
1120 if (unlikely(bh == NULL))
1121 GOTO(unlock, rc = -ENOENT);
1123 rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1128 rc = osd_ldiskfs_add_entry(jh, tgt_child, inode, NULL);
1133 mutex_unlock(&dir->i_mutex);
1136 mutex_unlock(&src_parent->i_mutex);
1137 ldiskfs_journal_stop(jh);
1141 static struct dentry *
1142 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1143 const struct lu_fid *fid, char **name)
1145 struct dentry *root = ERR_PTR(-ENOENT);
1147 if (fid_is_last_id(fid)) {
1148 struct osd_obj_seq *osd_seq;
1150 /* on creation of LAST_ID we create O/<seq> hierarchy */
1151 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1152 if (IS_ERR(osd_seq))
1153 RETURN((struct dentry *)osd_seq);
1156 root = osd_seq->oos_root;
1158 *name = osd_lf_fid2name(fid);
1160 CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1161 else if ((*name)[0])
1162 root = osd_sb(osd)->s_root;
1168 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1169 const struct lu_fid *fid, const struct osd_inode_id *id,
1172 struct dentry *root;
1177 root = osd_object_spec_find(info, osd, fid, &name);
1178 if (!IS_ERR(root)) {
1179 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1189 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1190 const struct lu_fid *fid, const struct osd_inode_id *id,
1193 struct dentry *root;
1198 root = osd_object_spec_find(info, osd, fid, &name);
1199 if (!IS_ERR(root)) {
1200 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1210 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1211 const struct lu_fid *fid, struct osd_inode_id *id)
1213 struct dentry *root;
1214 struct dentry *dentry;
1215 struct inode *inode;
1220 if (fid_is_last_id(fid)) {
1221 struct osd_obj_seq *osd_seq;
1223 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1224 if (IS_ERR(osd_seq))
1225 RETURN(PTR_ERR(osd_seq));
1226 root = osd_seq->oos_root;
1229 root = osd_sb(osd)->s_root;
1230 name = osd_lf_fid2name(fid);
1231 if (name == NULL || strlen(name) == 0)
1235 dentry = ll_lookup_one_len(name, root, strlen(name));
1236 if (!IS_ERR(dentry)) {
1237 inode = dentry->d_inode;
1239 if (is_bad_inode(inode)) {
1242 osd_id_gen(id, inode->i_ino,
1243 inode->i_generation);
1247 /* if dentry is accessible after osd_compat_spec_insert it
1248 * will still contain NULL inode, so don't keep it in cache */
1249 d_invalidate(dentry);