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, 2015, 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();
69 newctxt->umask = current_umask();
72 push_ctxt(save, newctxt);
75 /* utility to make a directory */
76 static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
77 const char *name, int mode, int fix)
79 struct dentry *dchild;
83 // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
84 CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
85 dchild = ll_lookup_one_len(name, dir, strlen(name));
89 if (dchild->d_inode) {
90 int old_mode = dchild->d_inode->i_mode;
91 if (!S_ISDIR(old_mode)) {
92 CERROR("found %s (%lu/%u) is mode %o\n", name,
93 dchild->d_inode->i_ino,
94 dchild->d_inode->i_generation, old_mode);
95 GOTO(out_err, err = -ENOTDIR);
98 /* Fixup directory permissions if necessary */
99 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
101 "fixing permissions on %s from %o to %o\n",
102 name, old_mode, mode);
103 dchild->d_inode->i_mode = (mode & S_IALLUGO) |
104 (old_mode & ~S_IALLUGO);
105 mark_inode_dirty(dchild->d_inode);
107 GOTO(out_up, dchild);
110 err = vfs_mkdir(dir->d_inode, dchild, mode);
118 dchild = ERR_PTR(err);
123 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
125 struct lr_server_data lsd;
126 struct dentry *dlast;
129 int count = OBJ_SUBDIR_COUNT;
133 dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
136 return PTR_ERR(dlast);
137 else if (dlast->d_inode == NULL)
141 rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
142 if (rc == sizeof(lsd)) {
143 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
144 "subdir count = %d\n", lsd.lsd_uuid,
145 lsd.lsd_subdir_count);
146 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
147 count = le16_to_cpu(lsd.lsd_subdir_count);
148 } else if (rc != 0) {
149 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
161 static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
162 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
164 struct lvfs_run_ctxt new;
165 struct lvfs_run_ctxt save;
166 struct dentry *parent;
167 struct osd_mdobj_map *omm;
169 struct osd_thread_info *info = osd_oti_get(env);
170 struct lu_fid *fid = &info->oti_fid3;
174 OBD_ALLOC_PTR(dev->od_mdt_map);
175 if (dev->od_mdt_map == NULL)
178 omm = dev->od_mdt_map;
180 parent = osd_sb(dev)->s_root;
181 osd_push_ctxt(dev, &new, &save);
183 d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
186 GOTO(cleanup, rc = PTR_ERR(d));
188 omm->omm_remote_parent = d;
190 /* Set LMA for remote parent inode */
191 lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
192 rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0);
197 pop_ctxt(&save, &new);
199 if (omm->omm_remote_parent != NULL)
200 dput(omm->omm_remote_parent);
202 dev->od_mdt_map = NULL;
207 static void osd_mdt_fini(struct osd_device *osd)
209 struct osd_mdobj_map *omm = osd->od_mdt_map;
214 if (omm->omm_remote_parent)
215 dput(omm->omm_remote_parent);
218 osd->od_ost_map = NULL;
221 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
222 struct osd_object *obj, struct osd_thandle *oh)
224 struct osd_mdobj_map *omm = osd->od_mdt_map;
225 struct osd_thread_info *oti = osd_oti_get(env);
226 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
227 char *name = oti->oti_name;
228 struct osd_thread_info *info = osd_oti_get(env);
229 struct dentry *dentry;
230 struct dentry *parent;
233 /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
234 * can identify this object quickly */
235 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
239 lma->lma_incompat |= LMAI_REMOTE_PARENT;
240 lustre_lma_swab(lma);
241 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
242 sizeof(*lma), XATTR_REPLACE);
246 parent = omm->omm_remote_parent;
247 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
248 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
250 mutex_lock(&parent->d_inode->i_mutex);
251 rc = osd_ldiskfs_add_entry(info, oh->ot_handle, dentry,
252 obj->oo_inode, NULL);
253 CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
254 name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
255 ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
256 mark_inode_dirty(parent->d_inode);
257 mutex_unlock(&parent->d_inode->i_mutex);
261 int osd_delete_from_remote_parent(const struct lu_env *env,
262 struct osd_device *osd,
263 struct osd_object *obj,
264 struct osd_thandle *oh)
266 struct osd_mdobj_map *omm = osd->od_mdt_map;
267 struct osd_thread_info *oti = osd_oti_get(env);
268 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
269 char *name = oti->oti_name;
270 struct dentry *dentry;
271 struct dentry *parent;
272 struct ldiskfs_dir_entry_2 *de;
273 struct buffer_head *bh;
276 /* Check lma to see whether it is remote object */
277 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
281 if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
284 parent = omm->omm_remote_parent;
285 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
286 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
288 mutex_lock(&parent->d_inode->i_mutex);
289 bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
292 mutex_unlock(&parent->d_inode->i_mutex);
295 CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
296 name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
297 rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
298 ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
299 mark_inode_dirty(parent->d_inode);
300 mutex_unlock(&parent->d_inode->i_mutex);
303 /* Get rid of REMOTE_PARENT flag from incompat */
304 lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
305 lustre_lma_swab(lma);
306 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
307 sizeof(*lma), XATTR_REPLACE);
311 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
312 struct osd_device *osd,
313 const struct lu_fid *fid,
314 struct osd_inode_id *id)
316 struct osd_mdobj_map *omm = osd->od_mdt_map;
317 char *name = oti->oti_name;
318 struct dentry *parent;
319 struct dentry *dentry;
320 struct ldiskfs_dir_entry_2 *de;
321 struct buffer_head *bh;
325 parent = omm->omm_remote_parent;
326 sprintf(name, DFID_NOBRACE, PFID(fid));
327 dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
329 mutex_lock(&parent->d_inode->i_mutex);
330 bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
337 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
339 inode = osd_iget(oti, osd, id);
349 mutex_unlock(&parent->d_inode->i_mutex);
351 osd_add_oi_cache(oti, osd, id, fid);
356 * directory structure on legacy OST:
358 * O/<seq>/d0-31/<objid>
365 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
367 struct lvfs_run_ctxt new;
368 struct lvfs_run_ctxt save;
369 struct dentry *rootd = osd_sb(dev)->s_root;
371 struct osd_thread_info *info = osd_oti_get(env);
373 struct lu_fid *fid = &info->oti_fid3;
377 OBD_ALLOC_PTR(dev->od_ost_map);
378 if (dev->od_ost_map == NULL)
381 /* to get subdir count from last_rcvd */
382 rc = osd_last_rcvd_subdir_count(dev);
384 GOTO(cleanup_alloc, rc);
386 dev->od_ost_map->om_subdir_count = rc;
389 INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
390 rwlock_init(&dev->od_ost_map->om_seq_list_lock);
391 mutex_init(&dev->od_ost_map->om_dir_init_mutex);
393 osd_push_ctxt(dev, &new, &save);
395 d = ll_lookup_one_len("O", rootd, strlen("O"));
397 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
398 if (d->d_inode == NULL) {
400 /* The lookup() may be called again inside simple_mkdir().
401 * Since the repeated lookup() only be called for "/O" at
402 * mount time, it will not affect the whole performance. */
403 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
405 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
407 /* It is quite probably that the device is new formatted. */
408 dev->od_maybe_new = 1;
412 dev->od_ost_map->om_root = d;
414 /* 'What the @fid is' is not imporatant, because the object
415 * has no OI mapping, and only is visible inside the OSD.*/
416 lu_igif_build(fid, inode->i_ino, inode->i_generation);
417 rc = osd_ea_fid_set(info, inode, fid,
418 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
420 GOTO(cleanup_dentry, rc);
422 pop_ctxt(&save, &new);
428 pop_ctxt(&save, &new);
430 OBD_FREE_PTR(dev->od_ost_map);
434 static void osd_seq_free(struct osd_obj_seq *osd_seq)
438 if (osd_seq->oos_dirs) {
439 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
440 if (osd_seq->oos_dirs[j])
441 dput(osd_seq->oos_dirs[j]);
443 OBD_FREE(osd_seq->oos_dirs,
444 sizeof(struct dentry *) * osd_seq->oos_subdir_count);
447 if (osd_seq->oos_root)
448 dput(osd_seq->oos_root);
450 OBD_FREE_PTR(osd_seq);
453 static void osd_ost_fini(struct osd_device *osd)
455 struct osd_obj_seq *osd_seq;
456 struct osd_obj_seq *tmp;
457 struct osd_obj_map *map = osd->od_ost_map;
463 write_lock(&map->om_seq_list_lock);
464 list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
466 list_del_init(&osd_seq->oos_seq_list);
467 write_unlock(&map->om_seq_list_lock);
468 osd_seq_free(osd_seq);
469 write_lock(&map->om_seq_list_lock);
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);
497 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
499 struct osd_obj_seq *osd_seq;
501 list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
502 if (osd_seq->oos_seq == seq)
508 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
510 struct osd_obj_seq *osd_seq;
512 read_lock(&map->om_seq_list_lock);
513 osd_seq = osd_seq_find_locked(map, seq);
514 read_unlock(&map->om_seq_list_lock);
518 void osd_obj_map_fini(struct osd_device *dev)
525 * Update the specified OI mapping.
527 * \retval 1, changed nothing
528 * \retval 0, changed successfully
529 * \retval -ve, on error
531 static int osd_obj_update_entry(struct osd_thread_info *info,
532 struct osd_device *osd,
533 struct dentry *dir, const char *name,
534 const struct lu_fid *fid,
535 const struct osd_inode_id *id,
538 struct inode *parent = dir->d_inode;
539 struct dentry *child;
540 struct ldiskfs_dir_entry_2 *de;
541 struct buffer_head *bh;
543 struct dentry *dentry = &info->oti_obj_dentry;
544 struct osd_inode_id *oi_id = &info->oti_id3;
545 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
546 struct lu_fid *oi_fid = &lma->lma_self_fid;
551 LASSERT(th->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->d_name, &de, NULL, 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 /* The EA inode should NOT be in OI, old OI scrub may added
578 * such OI mapping by wrong, replace it. */
579 if (unlikely(osd_is_ea_inode(inode))) {
584 rc = osd_get_lma(info, inode, dentry, lma);
585 if (rc == -ENODATA) {
586 rc = osd_get_idif(info, inode, dentry, oi_fid);
587 if (rc > 0 || rc == -ENODATA) {
597 /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
598 * either a crashed object or a uninitialized one. Replace it. */
599 if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
600 CERROR("%s: the FID "DFID" is used by two objects: "
601 "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
602 oi_id->oii_ino, oi_id->oii_gen,
603 id->oii_ino, id->oii_gen);
604 GOTO(out, rc = -EEXIST);
607 if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
608 __u32 idx1 = fid_idif_ost_idx(fid);
609 __u32 idx2 = fid_idif_ost_idx(oi_fid);
610 struct ost_id *ostid = &info->oti_ostid;
611 struct lu_fid *tfid = &info->oti_fid3;
613 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
614 "invalid given FID "DFID", not match the "
615 "device index %u\n", PFID(fid), osd->od_index);
618 if (idx1 == 0 && idx2 == osd->od_index) {
619 fid_to_ostid(fid, ostid);
620 ostid_to_fid(tfid, ostid, idx2);
621 if (lu_fid_eq(tfid, oi_fid)) {
622 CERROR("%s: the FID "DFID" is used by "
623 "two objects(2): %u/%u %u/%u\n",
624 osd_name(osd), PFID(fid),
625 oi_id->oii_ino, oi_id->oii_gen,
626 id->oii_ino, id->oii_gen);
628 GOTO(out, rc = -EEXIST);
630 } else if (idx2 == 0 && idx1 == osd->od_index) {
631 fid_to_ostid(oi_fid, ostid);
632 ostid_to_fid(tfid, ostid, idx1);
633 if (lu_fid_eq(tfid, fid)) {
634 CERROR("%s: the FID "DFID" is used by "
635 "two objects(2): %u/%u %u/%u\n",
636 osd_name(osd), PFID(fid),
637 oi_id->oii_ino, oi_id->oii_gen,
638 id->oii_ino, id->oii_gen);
640 GOTO(out, rc = -EEXIST);
647 /* There may be temporary inconsistency: On one hand, the new
648 * object may be referenced by multiple entries, which is out
649 * of our control unless we traverse the whole /O completely,
650 * which is non-flat order and inefficient, should be avoided;
651 * On the other hand, the old object may become orphan if it
652 * is still valid. Since it was referenced by an invalid entry,
653 * making it as invisible temporary may be not worse. OI scrub
654 * will process it later. */
655 rc = ldiskfs_journal_get_write_access(th, bh);
659 de->inode = cpu_to_le32(id->oii_ino);
660 rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
666 mutex_unlock(&parent->i_mutex);
670 static int osd_obj_del_entry(struct osd_thread_info *info,
671 struct osd_device *osd,
672 struct dentry *dird, char *name,
675 struct ldiskfs_dir_entry_2 *de;
676 struct buffer_head *bh;
677 struct dentry *child;
678 struct inode *dir = dird->d_inode;
683 LASSERT(th->h_transaction != NULL);
686 child = &info->oti_child_dentry;
687 child->d_name.hash = 0;
688 child->d_name.name = name;
689 child->d_name.len = strlen(name);
690 child->d_parent = dird;
691 child->d_inode = NULL;
694 mutex_lock(&dir->i_mutex);
696 bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
698 rc = ldiskfs_delete_entry(th, dir, de, bh);
701 mutex_unlock(&dir->i_mutex);
706 static int osd_obj_add_entry(struct osd_thread_info *info,
707 struct osd_device *osd,
708 struct dentry *dir, char *name,
709 const struct osd_inode_id *id,
712 struct dentry *child;
718 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
722 LASSERT(th->h_transaction != NULL);
724 inode = info->oti_inode;
725 if (unlikely(inode == NULL)) {
726 struct ldiskfs_inode_info *lii;
730 inode = info->oti_inode = &lii->vfs_inode;
733 inode->i_sb = osd_sb(osd);
734 osd_id_to_inode(inode, id);
735 inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
737 child = &info->oti_child_dentry;
738 child->d_name.hash = 0;
739 child->d_name.name = name;
740 child->d_name.len = strlen(name);
741 child->d_parent = dir;
742 child->d_inode = inode;
744 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
747 ll_vfs_dq_init(dir->d_inode);
748 mutex_lock(&dir->d_inode->i_mutex);
749 rc = osd_ldiskfs_add_entry(info, th, child, inode, NULL);
750 mutex_unlock(&dir->d_inode->i_mutex);
756 * Use LPU64 for legacy OST sequences, but use LPX64i for new
757 * sequences names, so that the O/{seq}/dN/{oid} more closely
758 * follows the DFID/PFID format. This makes it easier to map from
759 * debug messages to objects in the future, and the legacy space
760 * of FID_SEQ_OST_MDT0 will be unused in the future.
762 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
764 snprintf(seq_name, name_size,
765 (fid_seq_is_rsvd(seq) ||
766 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
767 fid_seq_is_idif(seq) ? 0 : seq);
770 static inline void osd_oid_name(char *name, size_t name_size,
771 const struct lu_fid *fid, u64 id)
773 snprintf(name, name_size,
774 (fid_seq_is_rsvd(fid_seq(fid)) ||
775 fid_seq_is_mdt0(fid_seq(fid)) ||
776 fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
779 /* external locking is required */
780 static int osd_seq_load_locked(struct osd_thread_info *info,
781 struct osd_device *osd,
782 struct osd_obj_seq *osd_seq)
784 struct osd_obj_map *map = osd->od_ost_map;
785 struct dentry *seq_dir;
787 struct lu_fid *fid = &info->oti_fid3;
793 if (osd_seq->oos_root != NULL)
797 LASSERT(map->om_root);
799 osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
801 seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
803 GOTO(out_err, rc = PTR_ERR(seq_dir));
804 else if (seq_dir->d_inode == NULL)
805 GOTO(out_put, rc = -EFAULT);
807 inode = seq_dir->d_inode;
808 osd_seq->oos_root = seq_dir;
810 /* 'What the @fid is' is not imporatant, because the object
811 * has no OI mapping, and only is visible inside the OSD.*/
812 lu_igif_build(fid, inode->i_ino, inode->i_generation);
813 rc = osd_ea_fid_set(info, inode, fid,
814 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
818 LASSERT(osd_seq->oos_dirs == NULL);
819 OBD_ALLOC(osd_seq->oos_dirs,
820 sizeof(seq_dir) * osd_seq->oos_subdir_count);
821 if (osd_seq->oos_dirs == NULL)
822 GOTO(out_put, rc = -ENOMEM);
824 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
827 snprintf(dir_name, sizeof(dir_name), "d%u", i);
828 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
831 GOTO(out_free, rc = PTR_ERR(dir));
832 } else if (dir->d_inode == NULL) {
834 GOTO(out_free, rc = -EFAULT);
837 inode = dir->d_inode;
838 osd_seq->oos_dirs[i] = dir;
840 /* 'What the @fid is' is not imporatant, because the object
841 * has no OI mapping, and only is visible inside the OSD.*/
842 lu_igif_build(fid, inode->i_ino, inode->i_generation);
843 rc = osd_ea_fid_set(info, inode, fid,
844 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
851 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
852 if (osd_seq->oos_dirs[i] != NULL)
853 dput(osd_seq->oos_dirs[i]);
855 OBD_FREE(osd_seq->oos_dirs,
856 sizeof(seq_dir) * osd_seq->oos_subdir_count);
859 osd_seq->oos_root = NULL;
865 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
866 struct osd_device *osd, u64 seq)
868 struct osd_obj_map *map;
869 struct osd_obj_seq *osd_seq;
873 map = osd->od_ost_map;
875 LASSERT(map->om_root);
877 osd_seq = osd_seq_find(map, seq);
878 if (likely(osd_seq != NULL))
881 /* Serializing init process */
882 mutex_lock(&map->om_dir_init_mutex);
884 /* Check whether the seq has been added */
885 read_lock(&map->om_seq_list_lock);
886 osd_seq = osd_seq_find_locked(map, seq);
887 if (osd_seq != NULL) {
888 read_unlock(&map->om_seq_list_lock);
889 GOTO(cleanup, rc = 0);
891 read_unlock(&map->om_seq_list_lock);
893 OBD_ALLOC_PTR(osd_seq);
895 GOTO(cleanup, rc = -ENOMEM);
897 INIT_LIST_HEAD(&osd_seq->oos_seq_list);
898 osd_seq->oos_seq = seq;
899 /* Init subdir count to be 32, but each seq can have
900 * different subdir count */
901 osd_seq->oos_subdir_count = map->om_subdir_count;
902 rc = osd_seq_load_locked(info, osd, osd_seq);
906 write_lock(&map->om_seq_list_lock);
907 list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
908 write_unlock(&map->om_seq_list_lock);
911 mutex_unlock(&map->om_dir_init_mutex);
914 OBD_FREE_PTR(osd_seq);
921 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
922 const struct lu_fid *fid, struct osd_inode_id *id)
924 struct osd_obj_map *map;
925 struct osd_obj_seq *osd_seq;
926 struct dentry *d_seq;
927 struct dentry *child;
928 struct ost_id *ostid = &info->oti_ostid;
931 struct ldiskfs_dir_entry_2 *de;
932 struct buffer_head *bh;
937 /* on the very first lookup we find and open directories */
939 map = dev->od_ost_map;
941 LASSERT(map->om_root);
943 fid_to_ostid(fid, ostid);
944 osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
946 RETURN(PTR_ERR(osd_seq));
948 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
949 d_seq = osd_seq->oos_dirs[dirn];
952 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
954 child = &info->oti_child_dentry;
955 child->d_parent = d_seq;
956 child->d_name.hash = 0;
957 child->d_name.name = name;
958 /* XXX: we can use rc from sprintf() instead of strlen() */
959 child->d_name.len = strlen(name);
961 dir = d_seq->d_inode;
962 mutex_lock(&dir->i_mutex);
963 bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
964 mutex_unlock(&dir->i_mutex);
969 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
972 inode = osd_iget(info, dev, id);
974 int rc = PTR_ERR(inode);
976 RETURN(rc == -ENOENT ? -ESTALE : rc);
983 int osd_obj_map_insert(struct osd_thread_info *info,
984 struct osd_device *osd,
985 const struct lu_fid *fid,
986 const struct osd_inode_id *id,
989 struct osd_obj_map *map;
990 struct osd_obj_seq *osd_seq;
992 struct ost_id *ostid = &info->oti_ostid;
998 map = osd->od_ost_map;
1001 /* map fid to seq:objid */
1002 fid_to_ostid(fid, ostid);
1004 oid = ostid_id(ostid);
1005 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1006 if (IS_ERR(osd_seq))
1007 RETURN(PTR_ERR(osd_seq));
1009 dirn = oid & (osd_seq->oos_subdir_count - 1);
1010 d = osd_seq->oos_dirs[dirn];
1013 osd_oid_name(name, sizeof(name), fid, oid);
1016 rc = osd_obj_add_entry(info, osd, d, name, id, th);
1017 if (rc == -EEXIST) {
1018 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1019 if (unlikely(rc == -ENOENT))
1022 if (unlikely(rc == 1))
1029 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1030 const struct lu_fid *fid, handle_t *th)
1032 struct osd_obj_map *map;
1033 struct osd_obj_seq *osd_seq;
1035 struct ost_id *ostid = &info->oti_ostid;
1040 map = osd->od_ost_map;
1043 /* map fid to seq:objid */
1044 fid_to_ostid(fid, ostid);
1046 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1047 if (IS_ERR(osd_seq))
1048 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1050 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1051 d = osd_seq->oos_dirs[dirn];
1054 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1055 rc = osd_obj_del_entry(info, osd, d, name, th);
1060 int osd_obj_map_update(struct osd_thread_info *info,
1061 struct osd_device *osd,
1062 const struct lu_fid *fid,
1063 const struct osd_inode_id *id,
1066 struct osd_obj_seq *osd_seq;
1068 struct ost_id *ostid = &info->oti_ostid;
1073 fid_to_ostid(fid, ostid);
1074 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1075 if (IS_ERR(osd_seq))
1076 RETURN(PTR_ERR(osd_seq));
1078 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1079 d = osd_seq->oos_dirs[dirn];
1082 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1083 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1088 int osd_obj_map_recover(struct osd_thread_info *info,
1089 struct osd_device *osd,
1090 struct inode *src_parent,
1091 struct dentry *src_child,
1092 const struct lu_fid *fid)
1094 struct osd_obj_seq *osd_seq;
1095 struct dentry *tgt_parent;
1096 struct dentry *tgt_child = &info->oti_child_dentry;
1098 struct inode *inode = src_child->d_inode;
1099 struct ost_id *ostid = &info->oti_ostid;
1101 struct ldiskfs_dir_entry_2 *de;
1102 struct buffer_head *bh;
1108 if (fid_is_last_id(fid)) {
1109 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1110 if (IS_ERR(osd_seq))
1111 RETURN(PTR_ERR(osd_seq));
1113 tgt_parent = osd_seq->oos_root;
1114 tgt_child->d_name.name = "LAST_ID";
1115 tgt_child->d_name.len = strlen("LAST_ID");
1117 fid_to_ostid(fid, ostid);
1118 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1119 if (IS_ERR(osd_seq))
1120 RETURN(PTR_ERR(osd_seq));
1122 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1123 tgt_parent = osd_seq->oos_dirs[dirn];
1124 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1125 tgt_child->d_name.name = name;
1126 tgt_child->d_name.len = strlen(name);
1128 LASSERT(tgt_parent != NULL);
1130 dir = tgt_parent->d_inode;
1131 tgt_child->d_name.hash = 0;
1132 tgt_child->d_parent = tgt_parent;
1133 tgt_child->d_inode = inode;
1135 /* The non-initialized src_child may be destroyed. */
1136 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1137 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1138 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1139 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1141 RETURN(PTR_ERR(jh));
1143 ll_vfs_dq_init(src_parent);
1144 ll_vfs_dq_init(dir);
1146 mutex_lock(&src_parent->i_mutex);
1147 mutex_lock(&dir->i_mutex);
1148 bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1150 /* XXX: If some other object occupied the same slot. And If such
1151 * inode is zero-sized and with SUID+SGID, then means it is
1152 * a new created one. Maybe we can remove it and insert the
1153 * original one back to the /O/<seq>/d<x>. But there are
1154 * something to be considered:
1156 * 1) The OST-object under /lost+found has crashed LMA.
1157 * So it should not conflict with the current one.
1159 * 2) There are race conditions that: someone may just want
1160 * to modify the current one. Even if the OI scrub takes
1161 * the object lock when remove the current one, it still
1162 * cause the modification to be lost becasue the target
1163 * has been removed when the RPC service thread waiting
1166 * So keep it there before we have suitable solution. */
1168 mutex_unlock(&dir->i_mutex);
1169 mutex_unlock(&src_parent->i_mutex);
1170 ldiskfs_journal_stop(jh);
1173 /* If the src object has never been modified, then remove it. */
1174 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1175 inode->i_mode & S_ISGID) {
1176 rc = ll_vfs_unlink(src_parent, src_child);
1177 if (unlikely(rc == -ENOENT))
1183 bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1185 if (unlikely(bh == NULL))
1186 GOTO(unlock, rc = -ENOENT);
1188 rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1193 rc = osd_ldiskfs_add_entry(info, jh, tgt_child, inode, NULL);
1198 mutex_unlock(&dir->i_mutex);
1199 mutex_unlock(&src_parent->i_mutex);
1200 ldiskfs_journal_stop(jh);
1204 static struct dentry *
1205 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1206 const struct lu_fid *fid, char **name)
1208 struct dentry *root = ERR_PTR(-ENOENT);
1210 if (fid_is_last_id(fid)) {
1211 struct osd_obj_seq *osd_seq;
1213 /* on creation of LAST_ID we create O/<seq> hierarchy */
1214 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1215 if (IS_ERR(osd_seq))
1216 RETURN((struct dentry *)osd_seq);
1219 root = osd_seq->oos_root;
1221 *name = osd_lf_fid2name(fid);
1223 CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1224 else if ((*name)[0])
1225 root = osd_sb(osd)->s_root;
1231 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1232 const struct lu_fid *fid, const struct osd_inode_id *id,
1235 struct dentry *root;
1240 root = osd_object_spec_find(info, osd, fid, &name);
1241 if (!IS_ERR(root)) {
1242 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1252 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1253 const struct lu_fid *fid, const struct osd_inode_id *id,
1256 struct dentry *root;
1261 root = osd_object_spec_find(info, osd, fid, &name);
1262 if (!IS_ERR(root)) {
1263 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1273 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1274 const struct lu_fid *fid, struct osd_inode_id *id)
1276 struct dentry *root;
1277 struct dentry *dentry;
1278 struct inode *inode;
1283 if (fid_is_last_id(fid)) {
1284 struct osd_obj_seq *osd_seq;
1286 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1287 if (IS_ERR(osd_seq))
1288 RETURN(PTR_ERR(osd_seq));
1289 root = osd_seq->oos_root;
1292 root = osd_sb(osd)->s_root;
1293 name = osd_lf_fid2name(fid);
1294 if (name == NULL || strlen(name) == 0)
1298 dentry = ll_lookup_one_len(name, root, strlen(name));
1299 if (!IS_ERR(dentry)) {
1300 inode = dentry->d_inode;
1302 if (is_bad_inode(inode)) {
1305 osd_id_gen(id, inode->i_ino,
1306 inode->i_generation);
1310 /* if dentry is accessible after osd_compat_spec_insert it
1311 * will still contain NULL inode, so don't keep it in cache */
1312 d_invalidate(dentry);