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 = vfs_mkdir(dir->d_inode, dchild, 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 = OBJ_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 parent = osd_sb(dev)->s_root;
186 osd_push_ctxt(dev, &new, &save);
188 d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
191 GOTO(cleanup, rc = PTR_ERR(d));
193 ldiskfs_set_inode_state(d->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
194 omm->omm_remote_parent = d;
196 /* Set LMA for remote parent inode */
197 lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
198 rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0);
203 pop_ctxt(&save, &new, NULL);
205 if (omm->omm_remote_parent != NULL)
206 dput(omm->omm_remote_parent);
208 dev->od_mdt_map = NULL;
213 static void osd_mdt_fini(struct osd_device *osd)
215 struct osd_mdobj_map *omm = osd->od_mdt_map;
220 if (omm->omm_remote_parent)
221 dput(omm->omm_remote_parent);
224 osd->od_ost_map = NULL;
227 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
228 struct osd_object *obj, struct osd_thandle *oh)
230 struct osd_mdobj_map *omm = osd->od_mdt_map;
231 struct osd_thread_info *oti = osd_oti_get(env);
232 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
233 char *name = oti->oti_name;
234 struct dentry *dentry;
235 struct dentry *parent;
238 /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
239 * can identify this object quickly */
240 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
244 lma->lma_incompat |= LMAI_REMOTE_PARENT;
245 lustre_lma_swab(lma);
246 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
247 sizeof(*lma), XATTR_REPLACE);
251 parent = omm->omm_remote_parent;
252 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
253 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
255 mutex_lock(&parent->d_inode->i_mutex);
256 rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
258 CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
259 name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
260 LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
261 osd_name(osd), parent->d_inode->i_ino,
262 parent->d_inode->i_nlink);
263 parent->d_inode->i_nlink++;
264 mark_inode_dirty(parent->d_inode);
265 mutex_unlock(&parent->d_inode->i_mutex);
269 int osd_delete_from_remote_parent(const struct lu_env *env,
270 struct osd_device *osd,
271 struct osd_object *obj,
272 struct osd_thandle *oh)
274 struct osd_mdobj_map *omm = osd->od_mdt_map;
275 struct osd_thread_info *oti = osd_oti_get(env);
276 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
277 char *name = oti->oti_name;
278 struct dentry *dentry;
279 struct dentry *parent;
280 struct ldiskfs_dir_entry_2 *de;
281 struct buffer_head *bh;
284 /* Check lma to see whether it is remote object */
285 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
289 if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
292 parent = omm->omm_remote_parent;
293 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
294 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
296 mutex_lock(&parent->d_inode->i_mutex);
297 bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
299 mutex_unlock(&parent->d_inode->i_mutex);
302 CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
303 name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
304 rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
305 LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
306 osd_name(osd), parent->d_inode->i_ino,
307 parent->d_inode->i_nlink);
308 parent->d_inode->i_nlink--;
309 mark_inode_dirty(parent->d_inode);
310 mutex_unlock(&parent->d_inode->i_mutex);
313 /* Get rid of REMOTE_PARENT flag from incompat */
314 lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
315 lustre_lma_swab(lma);
316 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
317 sizeof(*lma), XATTR_REPLACE);
321 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
322 struct osd_device *osd,
323 const struct lu_fid *fid,
324 struct osd_inode_id *id)
326 struct osd_mdobj_map *omm = osd->od_mdt_map;
327 char *name = oti->oti_name;
328 struct dentry *parent;
329 struct dentry *dentry;
330 struct ldiskfs_dir_entry_2 *de;
331 struct buffer_head *bh;
335 parent = omm->omm_remote_parent;
336 sprintf(name, DFID_NOBRACE, PFID(fid));
337 dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
339 mutex_lock(&parent->d_inode->i_mutex);
340 bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
345 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
348 mutex_unlock(&parent->d_inode->i_mutex);
350 osd_add_oi_cache(oti, osd, id, fid);
355 * directory structure on legacy OST:
357 * O/<seq>/d0-31/<objid>
364 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
366 struct lvfs_run_ctxt new;
367 struct lvfs_run_ctxt save;
368 struct dentry *rootd = osd_sb(dev)->s_root;
370 struct osd_thread_info *info = osd_oti_get(env);
372 struct lu_fid *fid = &info->oti_fid3;
376 OBD_ALLOC_PTR(dev->od_ost_map);
377 if (dev->od_ost_map == NULL)
380 /* to get subdir count from last_rcvd */
381 rc = osd_last_rcvd_subdir_count(dev);
383 OBD_FREE_PTR(dev->od_ost_map);
387 dev->od_ost_map->om_subdir_count = rc;
390 CFS_INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
391 rwlock_init(&dev->od_ost_map->om_seq_list_lock);
392 sema_init(&dev->od_ost_map->om_dir_init_sem, 1);
394 osd_push_ctxt(dev, &new, &save);
396 d = ll_lookup_one_len("O", rootd, strlen("O"));
398 GOTO(cleanup, rc = PTR_ERR(d));
399 if (d->d_inode == NULL) {
401 /* The lookup() may be called again inside simple_mkdir().
402 * Since the repeated lookup() only be called for "/O" at
403 * mount time, it will not affect the whole performance. */
404 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
406 GOTO(cleanup, rc = PTR_ERR(d));
408 /* It is quite probably that the device is new formatted. */
409 dev->od_maybe_new = 1;
413 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
414 dev->od_ost_map->om_root = d;
416 /* 'What the @fid is' is not imporatant, because the object
417 * has no OI mapping, and only is visible inside the OSD.*/
418 lu_igif_build(fid, inode->i_ino, inode->i_generation);
419 rc = osd_ea_fid_set(info, inode, fid,
420 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
425 osd_pop_ctxt(dev, &new, &save);
427 OBD_FREE_PTR(dev->od_ost_map);
433 static void osd_seq_free(struct osd_obj_map *map,
434 struct osd_obj_seq *osd_seq)
438 cfs_list_del_init(&osd_seq->oos_seq_list);
440 if (osd_seq->oos_dirs) {
441 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
442 if (osd_seq->oos_dirs[j])
443 dput(osd_seq->oos_dirs[j]);
445 OBD_FREE(osd_seq->oos_dirs,
446 sizeof(struct dentry *) * osd_seq->oos_subdir_count);
449 if (osd_seq->oos_root)
450 dput(osd_seq->oos_root);
452 OBD_FREE_PTR(osd_seq);
455 static void osd_ost_fini(struct osd_device *osd)
457 struct osd_obj_seq *osd_seq;
458 struct osd_obj_seq *tmp;
459 struct osd_obj_map *map = osd->od_ost_map;
465 write_lock(&map->om_seq_list_lock);
466 cfs_list_for_each_entry_safe(osd_seq, tmp,
469 osd_seq_free(map, osd_seq);
471 write_unlock(&map->om_seq_list_lock);
475 osd->od_ost_map = NULL;
479 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
484 /* prepare structures for OST */
485 rc = osd_ost_init(env, dev);
489 /* prepare structures for MDS */
490 rc = osd_mdt_init(env, dev);
495 struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq)
497 struct osd_obj_seq *osd_seq;
499 cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
500 if (osd_seq->oos_seq == seq)
506 struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
508 struct osd_obj_seq *osd_seq;
510 read_lock(&map->om_seq_list_lock);
511 osd_seq = osd_seq_find_locked(map, seq);
512 read_unlock(&map->om_seq_list_lock);
516 void osd_obj_map_fini(struct osd_device *dev)
523 * Update the specified OI mapping.
525 * \retval 1, changed nothing
526 * \retval 0, changed successfully
527 * \retval -ve, on error
529 static int osd_obj_update_entry(struct osd_thread_info *info,
530 struct osd_device *osd,
531 struct dentry *dir, const char *name,
532 const struct lu_fid *fid,
533 const struct osd_inode_id *id,
536 struct inode *parent = dir->d_inode;
537 struct osd_thandle *oh;
538 struct dentry *child;
539 struct ldiskfs_dir_entry_2 *de;
540 struct buffer_head *bh;
542 struct dentry *dentry = &info->oti_obj_dentry;
543 struct osd_inode_id *oi_id = &info->oti_id3;
544 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
545 struct lu_fid *oi_fid = &lma->lma_self_fid;
549 oh = container_of(th, struct osd_thandle, ot_super);
550 LASSERT(oh->ot_handle != NULL);
551 LASSERT(oh->ot_handle->h_transaction != NULL);
553 child = &info->oti_child_dentry;
554 child->d_parent = dir;
555 child->d_name.hash = 0;
556 child->d_name.name = name;
557 child->d_name.len = strlen(name);
559 ll_vfs_dq_init(parent);
560 mutex_lock(&parent->i_mutex);
561 bh = osd_ldiskfs_find_entry(parent, child, &de, NULL);
563 GOTO(out, rc = -ENOENT);
565 if (le32_to_cpu(de->inode) == id->oii_ino)
568 osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
569 inode = osd_iget(info, osd, oi_id);
572 if (rc == -ENOENT || rc == -ESTALE)
577 rc = osd_get_lma(info, inode, dentry, lma);
578 if (rc == -ENODATA) {
579 rc = osd_get_idif(info, inode, dentry, oi_fid);
587 /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
588 * either a crashed object or a uninitialized one. Replace it. */
589 if (rc == -ENODATA || oi_fid == NULL)
595 if (lu_fid_eq(fid, oi_fid)) {
596 CERROR("%s: the FID "DFID" is used by two objects: "
597 "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
598 oi_id->oii_ino, oi_id->oii_gen,
599 id->oii_ino, id->oii_gen);
600 GOTO(out, rc = -EEXIST);
604 /* There may be temporary inconsistency: On one hand, the new
605 * object may be referenced by multiple entries, which is out
606 * of our control unless we traverse the whole /O completely,
607 * which is non-flat order and inefficient, should be avoided;
608 * On the other hand, the old object may become orphan if it
609 * is still valid. Since it was referenced by an invalid entry,
610 * making it as invisible temporary may be not worse. OI scrub
611 * will process it later. */
612 rc = ldiskfs_journal_get_write_access(oh->ot_handle, bh);
616 de->inode = cpu_to_le32(id->oii_ino);
617 rc = ldiskfs_journal_dirty_metadata(oh->ot_handle, bh);
623 mutex_unlock(&parent->i_mutex);
627 static int osd_obj_del_entry(struct osd_thread_info *info,
628 struct osd_device *osd,
629 struct dentry *dird, char *name,
632 struct ldiskfs_dir_entry_2 *de;
633 struct buffer_head *bh;
634 struct osd_thandle *oh;
635 struct dentry *child;
636 struct inode *dir = dird->d_inode;
640 oh = container_of(th, struct osd_thandle, ot_super);
641 LASSERT(oh->ot_handle != NULL);
642 LASSERT(oh->ot_handle->h_transaction != NULL);
645 child = &info->oti_child_dentry;
646 child->d_name.hash = 0;
647 child->d_name.name = name;
648 child->d_name.len = strlen(name);
649 child->d_parent = dird;
650 child->d_inode = NULL;
653 mutex_lock(&dir->i_mutex);
655 bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
657 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
660 mutex_unlock(&dir->i_mutex);
665 int osd_obj_add_entry(struct osd_thread_info *info,
666 struct osd_device *osd,
667 struct dentry *dir, char *name,
668 const struct osd_inode_id *id,
671 struct osd_thandle *oh;
672 struct dentry *child;
678 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
681 oh = container_of(th, struct osd_thandle, ot_super);
682 LASSERT(oh->ot_handle != NULL);
683 LASSERT(oh->ot_handle->h_transaction != NULL);
685 inode = &info->oti_inode;
686 inode->i_sb = osd_sb(osd);
687 osd_id_to_inode(inode, id);
688 inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
690 child = &info->oti_child_dentry;
691 child->d_name.hash = 0;
692 child->d_name.name = name;
693 child->d_name.len = strlen(name);
694 child->d_parent = dir;
695 child->d_inode = inode;
697 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
700 ll_vfs_dq_init(dir->d_inode);
701 mutex_lock(&dir->d_inode->i_mutex);
702 rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
703 mutex_unlock(&dir->d_inode->i_mutex);
709 * Use LPU64 for legacy OST sequences, but use LPX64i for new
710 * sequences names, so that the O/{seq}/dN/{oid} more closely
711 * follows the DFID/PFID format. This makes it easier to map from
712 * debug messages to objects in the future, and the legacy space
713 * of FID_SEQ_OST_MDT0 will be unused in the future.
715 static inline void osd_seq_name(char *seq_name, size_t name_size, obd_seq seq)
717 snprintf(seq_name, name_size,
718 (fid_seq_is_rsvd(seq) ||
719 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
720 fid_seq_is_idif(seq) ? 0 : seq);
723 static inline void osd_oid_name(char *name, size_t name_size,
724 const struct lu_fid *fid, obd_id id)
726 snprintf(name, name_size,
727 (fid_seq_is_rsvd(fid_seq(fid)) ||
728 fid_seq_is_mdt0(fid_seq(fid)) ||
729 fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
732 /* external locking is required */
733 static int osd_seq_load_locked(struct osd_thread_info *info,
734 struct osd_device *osd,
735 struct osd_obj_seq *osd_seq)
737 struct osd_obj_map *map = osd->od_ost_map;
738 struct dentry *seq_dir;
740 struct lu_fid *fid = &info->oti_fid3;
746 if (osd_seq->oos_root != NULL)
750 LASSERT(map->om_root);
752 osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
754 seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
756 GOTO(out_err, rc = PTR_ERR(seq_dir));
757 else if (seq_dir->d_inode == NULL)
758 GOTO(out_put, rc = -EFAULT);
760 inode = seq_dir->d_inode;
761 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
762 osd_seq->oos_root = seq_dir;
764 /* 'What the @fid is' is not imporatant, because the object
765 * has no OI mapping, and only is visible inside the OSD.*/
766 lu_igif_build(fid, inode->i_ino, inode->i_generation);
767 rc = osd_ea_fid_set(info, inode, fid,
768 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
772 LASSERT(osd_seq->oos_dirs == NULL);
773 OBD_ALLOC(osd_seq->oos_dirs,
774 sizeof(seq_dir) * osd_seq->oos_subdir_count);
775 if (osd_seq->oos_dirs == NULL)
776 GOTO(out_put, rc = -ENOMEM);
778 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
781 snprintf(dir_name, sizeof(dir_name), "d%u", i);
782 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
785 GOTO(out_free, rc = PTR_ERR(dir));
786 } else if (dir->d_inode == NULL) {
788 GOTO(out_free, rc = -EFAULT);
791 inode = dir->d_inode;
792 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
793 osd_seq->oos_dirs[i] = dir;
795 /* 'What the @fid is' is not imporatant, because the object
796 * has no OI mapping, and only is visible inside the OSD.*/
797 lu_igif_build(fid, inode->i_ino, inode->i_generation);
798 rc = osd_ea_fid_set(info, inode, fid,
799 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
806 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
807 if (osd_seq->oos_dirs[i] != NULL)
808 dput(osd_seq->oos_dirs[i]);
810 OBD_FREE(osd_seq->oos_dirs,
811 sizeof(seq_dir) * osd_seq->oos_subdir_count);
814 osd_seq->oos_root = NULL;
820 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
821 struct osd_device *osd, obd_seq seq)
823 struct osd_obj_map *map;
824 struct osd_obj_seq *osd_seq;
828 map = osd->od_ost_map;
830 LASSERT(map->om_root);
832 osd_seq = osd_seq_find(map, seq);
833 if (likely(osd_seq != NULL))
836 /* Serializing init process */
837 down(&map->om_dir_init_sem);
839 /* Check whether the seq has been added */
840 read_lock(&map->om_seq_list_lock);
841 osd_seq = osd_seq_find_locked(map, seq);
842 if (osd_seq != NULL) {
843 read_unlock(&map->om_seq_list_lock);
844 GOTO(cleanup, rc = 0);
846 read_unlock(&map->om_seq_list_lock);
848 OBD_ALLOC_PTR(osd_seq);
850 GOTO(cleanup, rc = -ENOMEM);
852 CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list);
853 osd_seq->oos_seq = seq;
854 /* Init subdir count to be 32, but each seq can have
855 * different subdir count */
856 osd_seq->oos_subdir_count = map->om_subdir_count;
857 rc = osd_seq_load_locked(info, osd, osd_seq);
861 write_lock(&map->om_seq_list_lock);
862 cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
863 write_unlock(&map->om_seq_list_lock);
866 up(&map->om_dir_init_sem);
869 OBD_FREE_PTR(osd_seq);
876 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
877 const struct lu_fid *fid, struct osd_inode_id *id)
879 struct osd_obj_map *map;
880 struct osd_obj_seq *osd_seq;
881 struct dentry *d_seq;
882 struct dentry *child;
883 struct ost_id *ostid = &info->oti_ostid;
886 struct ldiskfs_dir_entry_2 *de;
887 struct buffer_head *bh;
892 /* on the very first lookup we find and open directories */
894 map = dev->od_ost_map;
896 LASSERT(map->om_root);
898 fid_to_ostid(fid, ostid);
899 osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
901 RETURN(PTR_ERR(osd_seq));
903 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
904 d_seq = osd_seq->oos_dirs[dirn];
907 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
909 child = &info->oti_child_dentry;
910 child->d_parent = d_seq;
911 child->d_name.hash = 0;
912 child->d_name.name = name;
913 /* XXX: we can use rc from sprintf() instead of strlen() */
914 child->d_name.len = strlen(name);
916 dir = d_seq->d_inode;
917 mutex_lock(&dir->i_mutex);
918 bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
919 mutex_unlock(&dir->i_mutex);
924 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
927 inode = osd_iget(info, dev, id);
929 RETURN(PTR_ERR(inode));
935 int osd_obj_map_insert(struct osd_thread_info *info,
936 struct osd_device *osd,
937 const struct lu_fid *fid,
938 const struct osd_inode_id *id,
941 struct osd_obj_map *map;
942 struct osd_obj_seq *osd_seq;
944 struct ost_id *ostid = &info->oti_ostid;
950 map = osd->od_ost_map;
953 /* map fid to seq:objid */
954 fid_to_ostid(fid, ostid);
956 oid = ostid_id(ostid);
957 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
959 RETURN(PTR_ERR(osd_seq));
961 dirn = oid & (osd_seq->oos_subdir_count - 1);
962 d = osd_seq->oos_dirs[dirn];
965 osd_oid_name(name, sizeof(name), fid, oid);
968 rc = osd_obj_add_entry(info, osd, d, name, id, th);
970 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
971 if (unlikely(rc == -ENOENT))
974 if (unlikely(rc == 1))
981 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
982 const struct lu_fid *fid, struct thandle *th)
984 struct osd_obj_map *map;
985 struct osd_obj_seq *osd_seq;
987 struct ost_id *ostid = &info->oti_ostid;
992 map = osd->od_ost_map;
995 /* map fid to seq:objid */
996 fid_to_ostid(fid, ostid);
998 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1000 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1002 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1003 d = osd_seq->oos_dirs[dirn];
1006 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1007 rc = osd_obj_del_entry(info, osd, d, name, th);
1012 int osd_obj_map_update(struct osd_thread_info *info,
1013 struct osd_device *osd,
1014 const struct lu_fid *fid,
1015 const struct osd_inode_id *id,
1018 struct osd_obj_seq *osd_seq;
1020 struct ost_id *ostid = &info->oti_ostid;
1025 fid_to_ostid(fid, ostid);
1026 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1027 if (IS_ERR(osd_seq))
1028 RETURN(PTR_ERR(osd_seq));
1030 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1031 d = osd_seq->oos_dirs[dirn];
1034 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1035 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1040 int osd_obj_map_recover(struct osd_thread_info *info,
1041 struct osd_device *osd,
1042 struct inode *src_parent,
1043 struct dentry *src_child,
1044 const struct lu_fid *fid)
1046 struct osd_obj_seq *osd_seq;
1047 struct dentry *tgt_parent;
1048 struct dentry *tgt_child = &info->oti_child_dentry;
1050 struct inode *inode = src_child->d_inode;
1051 struct ost_id *ostid = &info->oti_ostid;
1053 struct ldiskfs_dir_entry_2 *de;
1054 struct buffer_head *bh;
1060 if (fid_is_last_id(fid)) {
1061 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1062 if (IS_ERR(osd_seq))
1063 RETURN(PTR_ERR(osd_seq));
1065 tgt_parent = osd_seq->oos_root;
1066 tgt_child->d_name.name = "LAST_ID";
1067 tgt_child->d_name.len = strlen("LAST_ID");
1069 fid_to_ostid(fid, ostid);
1070 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1071 if (IS_ERR(osd_seq))
1072 RETURN(PTR_ERR(osd_seq));
1074 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1075 tgt_parent = osd_seq->oos_dirs[dirn];
1076 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1077 tgt_child->d_name.name = name;
1078 tgt_child->d_name.len = strlen(name);
1080 LASSERT(tgt_parent != NULL);
1082 dir = tgt_parent->d_inode;
1083 tgt_child->d_name.hash = 0;
1084 tgt_child->d_parent = tgt_parent;
1085 tgt_child->d_inode = inode;
1087 /* The non-initialized src_child may be destroyed. */
1088 jh = ldiskfs_journal_start_sb(osd_sb(osd),
1089 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1090 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1091 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1093 RETURN(PTR_ERR(jh));
1095 ll_vfs_dq_init(src_parent);
1096 ll_vfs_dq_init(dir);
1098 mutex_lock(&src_parent->i_mutex);
1099 mutex_lock(&dir->i_mutex);
1100 bh = osd_ldiskfs_find_entry(dir, tgt_child, &de, NULL);
1102 /* XXX: If some other object occupied the same slot. And If such
1103 * inode is zero-sized and with SUID+SGID, then means it is
1104 * a new created one. Maybe we can remove it and insert the
1105 * original one back to the /O/<seq>/d<x>. But there are
1106 * something to be considered:
1108 * 1) The OST-object under /lost+found has crashed LMA.
1109 * So it should not conflict with the current one.
1111 * 2) There are race conditions that: someone may just want
1112 * to modify the current one. Even if the OI scrub takes
1113 * the object lock when remove the current one, it still
1114 * cause the modification to be lost becasue the target
1115 * has been removed when the RPC service thread waiting
1118 * So keep it there before we have suitable solution. */
1120 mutex_unlock(&dir->i_mutex);
1123 /* If the src object has never been modified, then remove it. */
1124 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1125 inode->i_mode & S_ISGID) {
1126 bh = osd_ldiskfs_find_entry(src_parent, src_child, &de,
1128 if (unlikely(bh == NULL)) {
1129 mutex_unlock(&src_parent->i_mutex);
1130 ldiskfs_journal_stop(jh);
1134 rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1138 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1141 mutex_unlock(&src_parent->i_mutex);
1142 ldiskfs_journal_stop(jh);
1146 bh = osd_ldiskfs_find_entry(src_parent, src_child, &de, NULL);
1147 if (unlikely(bh == NULL))
1148 GOTO(unlock, rc = -ENOENT);
1150 rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1155 rc = osd_ldiskfs_add_entry(jh, tgt_child, inode, NULL);
1160 mutex_unlock(&dir->i_mutex);
1161 mutex_unlock(&src_parent->i_mutex);
1162 ldiskfs_journal_stop(jh);
1166 static struct dentry *
1167 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1168 const struct lu_fid *fid, char **name)
1170 struct dentry *root = ERR_PTR(-ENOENT);
1172 if (fid_is_last_id(fid)) {
1173 struct osd_obj_seq *osd_seq;
1175 /* on creation of LAST_ID we create O/<seq> hierarchy */
1176 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1177 if (IS_ERR(osd_seq))
1178 RETURN((struct dentry *)osd_seq);
1181 root = osd_seq->oos_root;
1183 *name = osd_lf_fid2name(fid);
1185 CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1186 else if ((*name)[0])
1187 root = osd_sb(osd)->s_root;
1193 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1194 const struct lu_fid *fid, const struct osd_inode_id *id,
1197 struct dentry *root;
1202 root = osd_object_spec_find(info, osd, fid, &name);
1203 if (!IS_ERR(root)) {
1204 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1214 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1215 const struct lu_fid *fid, const struct osd_inode_id *id,
1218 struct dentry *root;
1223 root = osd_object_spec_find(info, osd, fid, &name);
1224 if (!IS_ERR(root)) {
1225 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1235 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1236 const struct lu_fid *fid, struct osd_inode_id *id)
1238 struct dentry *root;
1239 struct dentry *dentry;
1240 struct inode *inode;
1245 if (fid_is_last_id(fid)) {
1246 struct osd_obj_seq *osd_seq;
1248 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1249 if (IS_ERR(osd_seq))
1250 RETURN(PTR_ERR(osd_seq));
1251 root = osd_seq->oos_root;
1254 root = osd_sb(osd)->s_root;
1255 name = osd_lf_fid2name(fid);
1256 if (name == NULL || strlen(name) == 0)
1260 dentry = ll_lookup_one_len(name, root, strlen(name));
1261 if (!IS_ERR(dentry)) {
1262 inode = dentry->d_inode;
1264 if (is_bad_inode(inode)) {
1267 osd_id_gen(id, inode->i_ino,
1268 inode->i_generation);
1272 /* if dentry is accessible after osd_compat_spec_insert it
1273 * will still contain NULL inode, so don't keep it in cache */
1274 d_invalidate(dentry);