1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Create and rename file, directory, symlinks
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * Portions of this code from linux/fs/ext3/dir.c
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise pascal
15 * Universite Pierre et Marie Curie (Paris VI)
19 * linux/fs/minix/dir.c
21 * Copyright (C) 1991, 1992 Linux Torvalds
23 * This program is free software; you can redistribute it and/or
24 * modify it under the terms of the GNU General Public
25 * License as published by the Free Software Foundation; either
26 * version 2 of the License, or (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
31 * General Public License for more details.
33 * You should have received a copy of the GNU General Public
34 * License along with this program; if not, write to the
35 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36 * Boston, MA 021110-1307, USA.
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/quotaops.h>
45 #define MLOG_MASK_PREFIX ML_NAMEI
46 #include <cluster/masklog.h>
54 #include "extent_map.h"
67 #include "buffer_head_io.h"
69 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
73 struct buffer_head
**new_fe_bh
,
74 struct buffer_head
*parent_fe_bh
,
76 struct ocfs2_alloc_context
*inode_ac
);
78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
79 struct inode
**ret_orphan_dir
,
82 struct ocfs2_dir_lookup_result
*lookup
);
84 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
87 struct ocfs2_dinode
*fe
,
89 struct ocfs2_dir_lookup_result
*lookup
,
90 struct inode
*orphan_dir_inode
);
92 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
97 /* An orphan dir name is an 8 byte value, printed as a hex string */
98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
100 static struct dentry
*ocfs2_lookup(struct inode
*dir
, struct dentry
*dentry
,
101 struct nameidata
*nd
)
105 struct inode
*inode
= NULL
;
107 struct ocfs2_inode_info
*oi
;
109 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
110 dentry
->d_name
.len
, dentry
->d_name
.name
);
112 if (dentry
->d_name
.len
> OCFS2_MAX_FILENAME_LEN
) {
113 ret
= ERR_PTR(-ENAMETOOLONG
);
117 mlog(0, "find name %.*s in directory %llu\n", dentry
->d_name
.len
,
118 dentry
->d_name
.name
, (unsigned long long)OCFS2_I(dir
)->ip_blkno
);
120 status
= ocfs2_inode_lock_nested(dir
, NULL
, 0, OI_LS_PARENT
);
122 if (status
!= -ENOENT
)
124 ret
= ERR_PTR(status
);
128 status
= ocfs2_lookup_ino_from_name(dir
, dentry
->d_name
.name
,
129 dentry
->d_name
.len
, &blkno
);
133 inode
= ocfs2_iget(OCFS2_SB(dir
->i_sb
), blkno
, 0, 0);
135 ret
= ERR_PTR(-EACCES
);
140 /* Clear any orphaned state... If we were able to look up the
141 * inode from a directory, it certainly can't be orphaned. We
142 * might have the bad state from a node which intended to
143 * orphan this inode but crashed before it could commit the
145 spin_lock(&oi
->ip_lock
);
146 oi
->ip_flags
&= ~OCFS2_INODE_MAYBE_ORPHANED
;
147 spin_unlock(&oi
->ip_lock
);
150 dentry
->d_op
= &ocfs2_dentry_ops
;
151 ret
= d_splice_alias(inode
, dentry
);
155 * If d_splice_alias() finds a DCACHE_DISCONNECTED
156 * dentry, it will d_move() it on top of ourse. The
157 * return value will indicate this however, so in
158 * those cases, we switch them around for the locking
161 * NOTE: This dentry already has ->d_op set from
162 * ocfs2_get_parent() and ocfs2_get_dentry()
167 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
168 OCFS2_I(dir
)->ip_blkno
);
171 ret
= ERR_PTR(status
);
177 /* Don't drop the cluster lock until *after* the d_add --
178 * unlink on another node will message us to remove that
179 * dentry under this lock so otherwise we can race this with
180 * the downconvert thread and have a stale dentry. */
181 ocfs2_inode_unlock(dir
, 0);
190 static struct inode
*ocfs2_get_init_inode(struct inode
*dir
, int mode
)
194 inode
= new_inode(dir
->i_sb
);
196 mlog(ML_ERROR
, "new_inode failed!\n");
200 /* populate as many fields early on as possible - many of
201 * these are used by the support functions here and in
207 inode
->i_uid
= current_fsuid();
208 if (dir
->i_mode
& S_ISGID
) {
209 inode
->i_gid
= dir
->i_gid
;
213 inode
->i_gid
= current_fsgid();
214 inode
->i_mode
= mode
;
219 static int ocfs2_mknod(struct inode
*dir
,
220 struct dentry
*dentry
,
225 struct buffer_head
*parent_fe_bh
= NULL
;
226 handle_t
*handle
= NULL
;
227 struct ocfs2_super
*osb
;
228 struct ocfs2_dinode
*dirfe
;
229 struct buffer_head
*new_fe_bh
= NULL
;
230 struct inode
*inode
= NULL
;
231 struct ocfs2_alloc_context
*inode_ac
= NULL
;
232 struct ocfs2_alloc_context
*data_ac
= NULL
;
233 struct ocfs2_alloc_context
*meta_ac
= NULL
;
234 int want_clusters
= 0;
236 int xattr_credits
= 0;
237 struct ocfs2_security_xattr_info si
= {
240 int did_quota_inode
= 0;
241 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
243 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
, mode
,
244 (unsigned long)dev
, dentry
->d_name
.len
,
245 dentry
->d_name
.name
);
247 /* get our super block */
248 osb
= OCFS2_SB(dir
->i_sb
);
250 status
= ocfs2_inode_lock(dir
, &parent_fe_bh
, 1);
252 if (status
!= -ENOENT
)
257 if (S_ISDIR(mode
) && (dir
->i_nlink
>= ocfs2_link_max(osb
))) {
262 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
263 if (!ocfs2_read_links_count(dirfe
)) {
264 /* can't make a file in a deleted directory. */
269 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
274 /* get a spot inside the dir. */
275 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
277 dentry
->d_name
.len
, &lookup
);
283 /* reserve an inode spot */
284 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
286 if (status
!= -ENOSPC
)
291 inode
= ocfs2_get_init_inode(dir
, mode
);
298 /* get security xattr */
299 status
= ocfs2_init_security_get(inode
, dir
, &si
);
301 if (status
== -EOPNOTSUPP
)
309 /* calculate meta data/clusters for setting security and acl xattr */
310 status
= ocfs2_calc_xattr_init(dir
, parent_fe_bh
, mode
,
312 &xattr_credits
, &want_meta
);
318 /* Reserve a cluster if creating an extent based directory. */
319 if (S_ISDIR(mode
) && !ocfs2_supports_inline_data(osb
)) {
322 /* Dir indexing requires extra space as well */
323 if (ocfs2_supports_indexed_dirs(osb
))
327 status
= ocfs2_reserve_new_metadata_blocks(osb
, want_meta
, &meta_ac
);
329 if (status
!= -ENOSPC
)
334 status
= ocfs2_reserve_clusters(osb
, want_clusters
, &data_ac
);
336 if (status
!= -ENOSPC
)
341 handle
= ocfs2_start_trans(osb
, ocfs2_mknod_credits(osb
->sb
,
344 if (IS_ERR(handle
)) {
345 status
= PTR_ERR(handle
);
351 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
353 if (sb_any_quota_active(osb
->sb
) &&
354 osb
->sb
->dq_op
->alloc_inode(inode
, 1) == NO_QUOTA
) {
360 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
,
361 inode
->i_mode
, (unsigned long)dev
, dentry
->d_name
.len
,
362 dentry
->d_name
.name
);
364 /* do the real work now. */
365 status
= ocfs2_mknod_locked(osb
, dir
, inode
, dev
,
366 &new_fe_bh
, parent_fe_bh
, handle
,
374 status
= ocfs2_fill_new_dir(osb
, handle
, dir
, inode
,
375 new_fe_bh
, data_ac
, meta_ac
);
381 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(dir
),
383 OCFS2_JOURNAL_ACCESS_WRITE
);
388 ocfs2_add_links_count(dirfe
, 1);
389 status
= ocfs2_journal_dirty(handle
, parent_fe_bh
);
397 status
= ocfs2_init_acl(handle
, inode
, dir
, new_fe_bh
, parent_fe_bh
,
405 status
= ocfs2_init_security_set(handle
, inode
, new_fe_bh
, &si
,
413 status
= ocfs2_add_entry(handle
, dentry
, inode
,
414 OCFS2_I(inode
)->ip_blkno
, parent_fe_bh
,
421 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
422 OCFS2_I(dir
)->ip_blkno
);
428 insert_inode_hash(inode
);
429 dentry
->d_op
= &ocfs2_dentry_ops
;
430 d_instantiate(dentry
, inode
);
433 if (status
< 0 && did_quota_inode
)
434 vfs_dq_free_inode(inode
);
436 ocfs2_commit_trans(osb
, handle
);
438 ocfs2_inode_unlock(dir
, 1);
440 if (status
== -ENOSPC
)
441 mlog(0, "Disk is full\n");
444 brelse(parent_fe_bh
);
448 ocfs2_free_dir_lookup_result(&lookup
);
450 if ((status
< 0) && inode
) {
456 ocfs2_free_alloc_context(inode_ac
);
459 ocfs2_free_alloc_context(data_ac
);
462 ocfs2_free_alloc_context(meta_ac
);
469 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
473 struct buffer_head
**new_fe_bh
,
474 struct buffer_head
*parent_fe_bh
,
476 struct ocfs2_alloc_context
*inode_ac
)
479 struct ocfs2_dinode
*fe
= NULL
;
480 struct ocfs2_extent_list
*fel
;
487 status
= ocfs2_claim_new_inode(osb
, handle
, dir
, parent_fe_bh
,
488 inode_ac
, &suballoc_bit
, &fe_blkno
);
494 /* populate as many fields early on as possible - many of
495 * these are used by the support functions here and in
497 inode
->i_ino
= ino_from_blkno(osb
->sb
, fe_blkno
);
498 OCFS2_I(inode
)->ip_blkno
= fe_blkno
;
499 spin_lock(&osb
->osb_lock
);
500 inode
->i_generation
= osb
->s_next_generation
++;
501 spin_unlock(&osb
->osb_lock
);
503 *new_fe_bh
= sb_getblk(osb
->sb
, fe_blkno
);
509 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
), *new_fe_bh
);
511 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
513 OCFS2_JOURNAL_ACCESS_CREATE
);
519 fe
= (struct ocfs2_dinode
*) (*new_fe_bh
)->b_data
;
520 memset(fe
, 0, osb
->sb
->s_blocksize
);
522 fe
->i_generation
= cpu_to_le32(inode
->i_generation
);
523 fe
->i_fs_generation
= cpu_to_le32(osb
->fs_generation
);
524 fe
->i_blkno
= cpu_to_le64(fe_blkno
);
525 fe
->i_suballoc_bit
= cpu_to_le16(suballoc_bit
);
526 fe
->i_suballoc_slot
= cpu_to_le16(inode_ac
->ac_alloc_slot
);
527 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
528 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
529 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
530 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
))
531 fe
->id1
.dev1
.i_rdev
= cpu_to_le64(huge_encode_dev(dev
));
533 ocfs2_set_links_count(fe
, inode
->i_nlink
);
535 fe
->i_last_eb_blk
= 0;
536 strcpy(fe
->i_signature
, OCFS2_INODE_SIGNATURE
);
537 le32_add_cpu(&fe
->i_flags
, OCFS2_VALID_FL
);
538 fe
->i_atime
= fe
->i_ctime
= fe
->i_mtime
=
539 cpu_to_le64(CURRENT_TIME
.tv_sec
);
540 fe
->i_mtime_nsec
= fe
->i_ctime_nsec
= fe
->i_atime_nsec
=
541 cpu_to_le32(CURRENT_TIME
.tv_nsec
);
545 * If supported, directories start with inline data. If inline
546 * isn't supported, but indexing is, we start them as indexed.
548 feat
= le16_to_cpu(fe
->i_dyn_features
);
549 if (S_ISDIR(inode
->i_mode
) && ocfs2_supports_inline_data(osb
)) {
550 fe
->i_dyn_features
= cpu_to_le16(feat
| OCFS2_INLINE_DATA_FL
);
552 fe
->id2
.i_data
.id_count
= cpu_to_le16(
553 ocfs2_max_inline_data_with_xattr(osb
->sb
, fe
));
555 fel
= &fe
->id2
.i_list
;
556 fel
->l_tree_depth
= 0;
557 fel
->l_next_free_rec
= 0;
558 fel
->l_count
= cpu_to_le16(ocfs2_extent_recs_per_inode(osb
->sb
));
561 status
= ocfs2_journal_dirty(handle
, *new_fe_bh
);
567 ocfs2_populate_inode(inode
, fe
, 1);
568 ocfs2_ci_set_new(osb
, INODE_CACHE(inode
));
569 if (!ocfs2_mount_local(osb
)) {
570 status
= ocfs2_create_new_inode_locks(inode
);
575 status
= 0; /* error in ocfs2_create_new_inode_locks is not
590 static int ocfs2_mkdir(struct inode
*dir
,
591 struct dentry
*dentry
,
596 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
597 dentry
->d_name
.len
, dentry
->d_name
.name
);
598 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
604 static int ocfs2_create(struct inode
*dir
,
605 struct dentry
*dentry
,
607 struct nameidata
*nd
)
611 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
612 dentry
->d_name
.len
, dentry
->d_name
.name
);
613 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
619 static int ocfs2_link(struct dentry
*old_dentry
,
621 struct dentry
*dentry
)
624 struct inode
*inode
= old_dentry
->d_inode
;
626 struct buffer_head
*fe_bh
= NULL
;
627 struct buffer_head
*parent_fe_bh
= NULL
;
628 struct ocfs2_dinode
*fe
= NULL
;
629 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
630 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
632 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode
->i_ino
,
633 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
634 dentry
->d_name
.len
, dentry
->d_name
.name
);
636 if (S_ISDIR(inode
->i_mode
))
639 err
= ocfs2_inode_lock_nested(dir
, &parent_fe_bh
, 1, OI_LS_PARENT
);
651 err
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
656 err
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
658 dentry
->d_name
.len
, &lookup
);
664 err
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
671 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
672 if (ocfs2_read_links_count(fe
) >= ocfs2_link_max(osb
)) {
674 goto out_unlock_inode
;
677 handle
= ocfs2_start_trans(osb
, ocfs2_link_credits(osb
->sb
));
678 if (IS_ERR(handle
)) {
679 err
= PTR_ERR(handle
);
682 goto out_unlock_inode
;
685 err
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
686 OCFS2_JOURNAL_ACCESS_WRITE
);
693 inode
->i_ctime
= CURRENT_TIME
;
694 ocfs2_set_links_count(fe
, inode
->i_nlink
);
695 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
696 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
698 err
= ocfs2_journal_dirty(handle
, fe_bh
);
700 ocfs2_add_links_count(fe
, -1);
706 err
= ocfs2_add_entry(handle
, dentry
, inode
,
707 OCFS2_I(inode
)->ip_blkno
,
708 parent_fe_bh
, &lookup
);
710 ocfs2_add_links_count(fe
, -1);
716 err
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
722 atomic_inc(&inode
->i_count
);
723 dentry
->d_op
= &ocfs2_dentry_ops
;
724 d_instantiate(dentry
, inode
);
727 ocfs2_commit_trans(osb
, handle
);
729 ocfs2_inode_unlock(inode
, 1);
732 ocfs2_inode_unlock(dir
, 1);
735 brelse(parent_fe_bh
);
737 ocfs2_free_dir_lookup_result(&lookup
);
745 * Takes and drops an exclusive lock on the given dentry. This will
746 * force other nodes to drop it.
748 static int ocfs2_remote_dentry_delete(struct dentry
*dentry
)
752 ret
= ocfs2_dentry_lock(dentry
, 1);
756 ocfs2_dentry_unlock(dentry
, 1);
761 static inline int inode_is_unlinkable(struct inode
*inode
)
763 if (S_ISDIR(inode
->i_mode
)) {
764 if (inode
->i_nlink
== 2)
769 if (inode
->i_nlink
== 1)
774 static int ocfs2_unlink(struct inode
*dir
,
775 struct dentry
*dentry
)
778 int child_locked
= 0;
779 struct inode
*inode
= dentry
->d_inode
;
780 struct inode
*orphan_dir
= NULL
;
781 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
783 struct ocfs2_dinode
*fe
= NULL
;
784 struct buffer_head
*fe_bh
= NULL
;
785 struct buffer_head
*parent_node_bh
= NULL
;
786 handle_t
*handle
= NULL
;
787 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
788 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
789 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
791 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
792 dentry
->d_name
.len
, dentry
->d_name
.name
);
794 BUG_ON(dentry
->d_parent
->d_inode
!= dir
);
796 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
798 if (inode
== osb
->root_inode
) {
799 mlog(0, "Cannot delete the root directory\n");
803 status
= ocfs2_inode_lock_nested(dir
, &parent_node_bh
, 1,
806 if (status
!= -ENOENT
)
811 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
812 dentry
->d_name
.len
, &blkno
, dir
,
815 if (status
!= -ENOENT
)
820 if (OCFS2_I(inode
)->ip_blkno
!= blkno
) {
823 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
824 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
825 (unsigned long long)blkno
, OCFS2_I(inode
)->ip_flags
);
829 status
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
831 if (status
!= -ENOENT
)
837 if (S_ISDIR(inode
->i_mode
)) {
838 if (inode
->i_nlink
!= 2 || !ocfs2_empty_dir(inode
)) {
844 status
= ocfs2_remote_dentry_delete(dentry
);
846 /* This remote delete should succeed under all normal
852 if (inode_is_unlinkable(inode
)) {
853 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
854 OCFS2_I(inode
)->ip_blkno
,
855 orphan_name
, &orphan_insert
);
862 handle
= ocfs2_start_trans(osb
, ocfs2_unlink_credits(osb
->sb
));
863 if (IS_ERR(handle
)) {
864 status
= PTR_ERR(handle
);
870 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
871 OCFS2_JOURNAL_ACCESS_WRITE
);
877 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
879 if (inode_is_unlinkable(inode
)) {
880 status
= ocfs2_orphan_add(osb
, handle
, inode
, fe
, orphan_name
,
881 &orphan_insert
, orphan_dir
);
888 /* delete the name from the parent dir */
889 status
= ocfs2_delete_entry(handle
, dir
, &lookup
);
895 if (S_ISDIR(inode
->i_mode
))
898 ocfs2_set_links_count(fe
, inode
->i_nlink
);
900 status
= ocfs2_journal_dirty(handle
, fe_bh
);
906 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
907 if (S_ISDIR(inode
->i_mode
))
910 status
= ocfs2_mark_inode_dirty(handle
, dir
, parent_node_bh
);
913 if (S_ISDIR(inode
->i_mode
))
919 ocfs2_commit_trans(osb
, handle
);
922 ocfs2_inode_unlock(inode
, 1);
924 ocfs2_inode_unlock(dir
, 1);
927 /* This was locked for us in ocfs2_prepare_orphan_dir() */
928 ocfs2_inode_unlock(orphan_dir
, 1);
929 mutex_unlock(&orphan_dir
->i_mutex
);
934 brelse(parent_node_bh
);
936 ocfs2_free_dir_lookup_result(&orphan_insert
);
937 ocfs2_free_dir_lookup_result(&lookup
);
945 * The only place this should be used is rename!
946 * if they have the same id, then the 1st one is the only one locked.
948 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
949 struct buffer_head
**bh1
,
950 struct inode
*inode1
,
951 struct buffer_head
**bh2
,
952 struct inode
*inode2
)
955 struct ocfs2_inode_info
*oi1
= OCFS2_I(inode1
);
956 struct ocfs2_inode_info
*oi2
= OCFS2_I(inode2
);
957 struct buffer_head
**tmpbh
;
958 struct inode
*tmpinode
;
960 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
961 (unsigned long long)oi1
->ip_blkno
,
962 (unsigned long long)oi2
->ip_blkno
);
969 /* we always want to lock the one with the lower lockid first. */
970 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
971 if (oi1
->ip_blkno
< oi2
->ip_blkno
) {
972 /* switch id1 and id2 around */
973 mlog(0, "switching them around...\n");
983 status
= ocfs2_inode_lock_nested(inode2
, bh2
, 1,
986 if (status
!= -ENOENT
)
993 status
= ocfs2_inode_lock_nested(inode1
, bh1
, 1, OI_LS_RENAME2
);
996 * An error return must mean that no cluster locks
997 * were held on function exit.
999 if (oi1
->ip_blkno
!= oi2
->ip_blkno
)
1000 ocfs2_inode_unlock(inode2
, 1);
1002 if (status
!= -ENOENT
)
1011 static void ocfs2_double_unlock(struct inode
*inode1
, struct inode
*inode2
)
1013 ocfs2_inode_unlock(inode1
, 1);
1015 if (inode1
!= inode2
)
1016 ocfs2_inode_unlock(inode2
, 1);
1019 static int ocfs2_rename(struct inode
*old_dir
,
1020 struct dentry
*old_dentry
,
1021 struct inode
*new_dir
,
1022 struct dentry
*new_dentry
)
1024 int status
= 0, rename_lock
= 0, parents_locked
= 0, target_exists
= 0;
1025 int old_child_locked
= 0, new_child_locked
= 0, update_dot_dot
= 0;
1026 struct inode
*old_inode
= old_dentry
->d_inode
;
1027 struct inode
*new_inode
= new_dentry
->d_inode
;
1028 struct inode
*orphan_dir
= NULL
;
1029 struct ocfs2_dinode
*newfe
= NULL
;
1030 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1031 struct buffer_head
*newfe_bh
= NULL
;
1032 struct buffer_head
*old_inode_bh
= NULL
;
1033 struct ocfs2_super
*osb
= NULL
;
1034 u64 newfe_blkno
, old_de_ino
;
1035 handle_t
*handle
= NULL
;
1036 struct buffer_head
*old_dir_bh
= NULL
;
1037 struct buffer_head
*new_dir_bh
= NULL
;
1038 nlink_t old_dir_nlink
= old_dir
->i_nlink
;
1039 struct ocfs2_dinode
*old_di
;
1040 struct ocfs2_dir_lookup_result old_inode_dot_dot_res
= { NULL
, };
1041 struct ocfs2_dir_lookup_result target_lookup_res
= { NULL
, };
1042 struct ocfs2_dir_lookup_result old_entry_lookup
= { NULL
, };
1043 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
1044 struct ocfs2_dir_lookup_result target_insert
= { NULL
, };
1046 /* At some point it might be nice to break this function up a
1049 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1050 old_dir
, old_dentry
, new_dir
, new_dentry
,
1051 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
1052 new_dentry
->d_name
.len
, new_dentry
->d_name
.name
);
1054 osb
= OCFS2_SB(old_dir
->i_sb
);
1057 if (!igrab(new_inode
))
1061 /* Assume a directory hierarchy thusly:
1064 * a,b,c, and d are all directories.
1066 * from cwd of 'a' on both nodes:
1070 * And that's why, just like the VFS, we need a file system
1072 if (old_dir
!= new_dir
&& S_ISDIR(old_inode
->i_mode
)) {
1073 status
= ocfs2_rename_lock(osb
);
1081 /* if old and new are the same, this'll just do one lock. */
1082 status
= ocfs2_double_lock(osb
, &old_dir_bh
, old_dir
,
1083 &new_dir_bh
, new_dir
);
1090 /* make sure both dirs have bhs
1091 * get an extra ref on old_dir_bh if old==new */
1094 new_dir_bh
= old_dir_bh
;
1097 mlog(ML_ERROR
, "no old_dir_bh!\n");
1104 * Aside from allowing a meta data update, the locking here
1105 * also ensures that the downconvert thread on other nodes
1106 * won't have to concurrently downconvert the inode and the
1109 status
= ocfs2_inode_lock_nested(old_inode
, &old_inode_bh
, 1,
1112 if (status
!= -ENOENT
)
1116 old_child_locked
= 1;
1118 status
= ocfs2_remote_dentry_delete(old_dentry
);
1124 if (S_ISDIR(old_inode
->i_mode
)) {
1125 u64 old_inode_parent
;
1128 status
= ocfs2_find_files_on_disk("..", 2, &old_inode_parent
,
1130 &old_inode_dot_dot_res
);
1136 if (old_inode_parent
!= OCFS2_I(old_dir
)->ip_blkno
) {
1141 if (!new_inode
&& new_dir
!= old_dir
&&
1142 new_dir
->i_nlink
>= ocfs2_link_max(osb
)) {
1148 status
= ocfs2_lookup_ino_from_name(old_dir
, old_dentry
->d_name
.name
,
1149 old_dentry
->d_name
.len
,
1157 * Check for inode number is _not_ due to possible IO errors.
1158 * We might rmdir the source, keep it as pwd of some process
1159 * and merrily kill the link to whatever was created under the
1160 * same name. Goodbye sticky bit ;-<
1162 if (old_de_ino
!= OCFS2_I(old_inode
)->ip_blkno
) {
1167 /* check if the target already exists (in which case we need
1169 status
= ocfs2_find_files_on_disk(new_dentry
->d_name
.name
,
1170 new_dentry
->d_name
.len
,
1171 &newfe_blkno
, new_dir
,
1172 &target_lookup_res
);
1173 /* The only error we allow here is -ENOENT because the new
1174 * file not existing is perfectly valid. */
1175 if ((status
< 0) && (status
!= -ENOENT
)) {
1176 /* If we cannot find the file specified we should just */
1177 /* return the error... */
1184 if (!target_exists
&& new_inode
) {
1186 * Target was unlinked by another node while we were
1187 * waiting to get to ocfs2_rename(). There isn't
1188 * anything we can do here to help the situation, so
1189 * bubble up the appropriate error.
1195 /* In case we need to overwrite an existing file, we blow it
1197 if (target_exists
) {
1198 /* VFS didn't think there existed an inode here, but
1199 * someone else in the cluster must have raced our
1200 * rename to create one. Today we error cleanly, in
1201 * the future we should consider calling iget to build
1202 * a new struct inode for this entry. */
1206 mlog(0, "We found an inode for name %.*s but VFS "
1207 "didn't give us one.\n", new_dentry
->d_name
.len
,
1208 new_dentry
->d_name
.name
);
1212 if (OCFS2_I(new_inode
)->ip_blkno
!= newfe_blkno
) {
1215 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1216 (unsigned long long)OCFS2_I(new_inode
)->ip_blkno
,
1217 (unsigned long long)newfe_blkno
,
1218 OCFS2_I(new_inode
)->ip_flags
);
1222 status
= ocfs2_inode_lock(new_inode
, &newfe_bh
, 1);
1224 if (status
!= -ENOENT
)
1228 new_child_locked
= 1;
1230 status
= ocfs2_remote_dentry_delete(new_dentry
);
1236 newfe
= (struct ocfs2_dinode
*) newfe_bh
->b_data
;
1238 mlog(0, "aha rename over existing... new_blkno=%llu "
1239 "newfebh=%p bhblocknr=%llu\n",
1240 (unsigned long long)newfe_blkno
, newfe_bh
, newfe_bh
?
1241 (unsigned long long)newfe_bh
->b_blocknr
: 0ULL);
1243 if (S_ISDIR(new_inode
->i_mode
) || (new_inode
->i_nlink
== 1)) {
1244 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
1245 OCFS2_I(new_inode
)->ip_blkno
,
1246 orphan_name
, &orphan_insert
);
1253 BUG_ON(new_dentry
->d_parent
->d_inode
!= new_dir
);
1255 status
= ocfs2_check_dir_for_entry(new_dir
,
1256 new_dentry
->d_name
.name
,
1257 new_dentry
->d_name
.len
);
1261 status
= ocfs2_prepare_dir_for_insert(osb
, new_dir
, new_dir_bh
,
1262 new_dentry
->d_name
.name
,
1263 new_dentry
->d_name
.len
,
1271 handle
= ocfs2_start_trans(osb
, ocfs2_rename_credits(osb
->sb
));
1272 if (IS_ERR(handle
)) {
1273 status
= PTR_ERR(handle
);
1279 if (target_exists
) {
1280 if (S_ISDIR(new_inode
->i_mode
)) {
1281 if (new_inode
->i_nlink
!= 2 ||
1282 !ocfs2_empty_dir(new_inode
)) {
1283 status
= -ENOTEMPTY
;
1287 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(new_inode
),
1289 OCFS2_JOURNAL_ACCESS_WRITE
);
1295 if (S_ISDIR(new_inode
->i_mode
) ||
1296 (ocfs2_read_links_count(newfe
) == 1)) {
1297 status
= ocfs2_orphan_add(osb
, handle
, new_inode
,
1299 &orphan_insert
, orphan_dir
);
1306 /* change the dirent to point to the correct inode */
1307 status
= ocfs2_update_entry(new_dir
, handle
, &target_lookup_res
,
1313 new_dir
->i_version
++;
1315 if (S_ISDIR(new_inode
->i_mode
))
1316 ocfs2_set_links_count(newfe
, 0);
1318 ocfs2_add_links_count(newfe
, -1);
1320 status
= ocfs2_journal_dirty(handle
, newfe_bh
);
1326 /* if the name was not found in new_dir, add it now */
1327 status
= ocfs2_add_entry(handle
, new_dentry
, old_inode
,
1328 OCFS2_I(old_inode
)->ip_blkno
,
1329 new_dir_bh
, &target_insert
);
1332 old_inode
->i_ctime
= CURRENT_TIME
;
1333 mark_inode_dirty(old_inode
);
1335 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(old_inode
),
1337 OCFS2_JOURNAL_ACCESS_WRITE
);
1339 old_di
= (struct ocfs2_dinode
*) old_inode_bh
->b_data
;
1341 old_di
->i_ctime
= cpu_to_le64(old_inode
->i_ctime
.tv_sec
);
1342 old_di
->i_ctime_nsec
= cpu_to_le32(old_inode
->i_ctime
.tv_nsec
);
1344 status
= ocfs2_journal_dirty(handle
, old_inode_bh
);
1351 * Now that the name has been added to new_dir, remove the old name.
1353 * We don't keep any directory entry context around until now
1354 * because the insert might have changed the type of directory
1355 * we're dealing with.
1357 status
= ocfs2_find_entry(old_dentry
->d_name
.name
,
1358 old_dentry
->d_name
.len
, old_dir
,
1363 status
= ocfs2_delete_entry(handle
, old_dir
, &old_entry_lookup
);
1370 new_inode
->i_nlink
--;
1371 new_inode
->i_ctime
= CURRENT_TIME
;
1373 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1375 if (update_dot_dot
) {
1376 status
= ocfs2_update_entry(old_inode
, handle
,
1377 &old_inode_dot_dot_res
, new_dir
);
1380 new_inode
->i_nlink
--;
1383 mark_inode_dirty(new_dir
);
1386 mark_inode_dirty(old_dir
);
1387 ocfs2_mark_inode_dirty(handle
, old_dir
, old_dir_bh
);
1389 mark_inode_dirty(new_inode
);
1390 ocfs2_mark_inode_dirty(handle
, new_inode
, newfe_bh
);
1393 if (old_dir
!= new_dir
) {
1394 /* Keep the same times on both directories.*/
1395 new_dir
->i_ctime
= new_dir
->i_mtime
= old_dir
->i_ctime
;
1398 * This will also pick up the i_nlink change from the
1401 ocfs2_mark_inode_dirty(handle
, new_dir
, new_dir_bh
);
1404 if (old_dir_nlink
!= old_dir
->i_nlink
) {
1406 mlog(ML_ERROR
, "need to change nlink for old dir "
1407 "%llu from %d to %d but bh is NULL!\n",
1408 (unsigned long long)OCFS2_I(old_dir
)->ip_blkno
,
1409 (int)old_dir_nlink
, old_dir
->i_nlink
);
1411 struct ocfs2_dinode
*fe
;
1412 status
= ocfs2_journal_access_di(handle
,
1413 INODE_CACHE(old_dir
),
1415 OCFS2_JOURNAL_ACCESS_WRITE
);
1416 fe
= (struct ocfs2_dinode
*) old_dir_bh
->b_data
;
1417 ocfs2_set_links_count(fe
, old_dir
->i_nlink
);
1418 status
= ocfs2_journal_dirty(handle
, old_dir_bh
);
1421 ocfs2_dentry_move(old_dentry
, new_dentry
, old_dir
, new_dir
);
1425 ocfs2_rename_unlock(osb
);
1428 ocfs2_commit_trans(osb
, handle
);
1431 ocfs2_double_unlock(old_dir
, new_dir
);
1433 if (old_child_locked
)
1434 ocfs2_inode_unlock(old_inode
, 1);
1436 if (new_child_locked
)
1437 ocfs2_inode_unlock(new_inode
, 1);
1440 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1441 ocfs2_inode_unlock(orphan_dir
, 1);
1442 mutex_unlock(&orphan_dir
->i_mutex
);
1447 sync_mapping_buffers(old_inode
->i_mapping
);
1452 ocfs2_free_dir_lookup_result(&target_lookup_res
);
1453 ocfs2_free_dir_lookup_result(&old_entry_lookup
);
1454 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res
);
1455 ocfs2_free_dir_lookup_result(&orphan_insert
);
1456 ocfs2_free_dir_lookup_result(&target_insert
);
1459 brelse(old_inode_bh
);
1469 * we expect i_size = strlen(symname). Copy symname into the file
1470 * data, including the null terminator.
1472 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
1474 struct inode
*inode
,
1475 const char *symname
)
1477 struct buffer_head
**bhs
= NULL
;
1479 struct super_block
*sb
= osb
->sb
;
1480 u64 p_blkno
, p_blocks
;
1481 int virtual, blocks
, status
, i
, bytes_left
;
1483 bytes_left
= i_size_read(inode
) + 1;
1484 /* we can't trust i_blocks because we're actually going to
1485 * write i_size + 1 bytes. */
1486 blocks
= (bytes_left
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
1488 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1489 (unsigned long long)inode
->i_blocks
,
1490 i_size_read(inode
), blocks
);
1492 /* Sanity check -- make sure we're going to fit. */
1494 ocfs2_clusters_to_bytes(sb
, OCFS2_I(inode
)->ip_clusters
)) {
1500 bhs
= kcalloc(blocks
, sizeof(struct buffer_head
*), GFP_KERNEL
);
1507 status
= ocfs2_extent_map_get_blocks(inode
, 0, &p_blkno
, &p_blocks
,
1514 /* links can never be larger than one cluster so we know this
1515 * is all going to be contiguous, but do a sanity check
1517 if ((p_blocks
<< sb
->s_blocksize_bits
) < bytes_left
) {
1524 while(bytes_left
> 0) {
1525 c
= &symname
[virtual * sb
->s_blocksize
];
1527 bhs
[virtual] = sb_getblk(sb
, p_blkno
);
1528 if (!bhs
[virtual]) {
1533 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
),
1536 status
= ocfs2_journal_access(handle
, INODE_CACHE(inode
),
1538 OCFS2_JOURNAL_ACCESS_CREATE
);
1544 memset(bhs
[virtual]->b_data
, 0, sb
->s_blocksize
);
1546 memcpy(bhs
[virtual]->b_data
, c
,
1547 (bytes_left
> sb
->s_blocksize
) ? sb
->s_blocksize
:
1550 status
= ocfs2_journal_dirty(handle
, bhs
[virtual]);
1558 bytes_left
-= sb
->s_blocksize
;
1565 for(i
= 0; i
< blocks
; i
++)
1574 static int ocfs2_symlink(struct inode
*dir
,
1575 struct dentry
*dentry
,
1576 const char *symname
)
1578 int status
, l
, credits
;
1580 struct ocfs2_super
*osb
= NULL
;
1581 struct inode
*inode
= NULL
;
1582 struct super_block
*sb
;
1583 struct buffer_head
*new_fe_bh
= NULL
;
1584 struct buffer_head
*parent_fe_bh
= NULL
;
1585 struct ocfs2_dinode
*fe
= NULL
;
1586 struct ocfs2_dinode
*dirfe
;
1587 handle_t
*handle
= NULL
;
1588 struct ocfs2_alloc_context
*inode_ac
= NULL
;
1589 struct ocfs2_alloc_context
*data_ac
= NULL
;
1590 struct ocfs2_alloc_context
*xattr_ac
= NULL
;
1591 int want_clusters
= 0;
1592 int xattr_credits
= 0;
1593 struct ocfs2_security_xattr_info si
= {
1596 int did_quota
= 0, did_quota_inode
= 0;
1597 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
1599 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir
,
1600 dentry
, symname
, dentry
->d_name
.len
, dentry
->d_name
.name
);
1605 l
= strlen(symname
) + 1;
1607 credits
= ocfs2_calc_symlink_credits(sb
);
1609 /* lock the parent directory */
1610 status
= ocfs2_inode_lock(dir
, &parent_fe_bh
, 1);
1612 if (status
!= -ENOENT
)
1617 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
1618 if (!ocfs2_read_links_count(dirfe
)) {
1619 /* can't make a file in a deleted directory. */
1624 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
1625 dentry
->d_name
.len
);
1629 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
1630 dentry
->d_name
.name
,
1631 dentry
->d_name
.len
, &lookup
);
1637 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
1639 if (status
!= -ENOSPC
)
1644 inode
= ocfs2_get_init_inode(dir
, S_IFLNK
| S_IRWXUGO
);
1651 /* get security xattr */
1652 status
= ocfs2_init_security_get(inode
, dir
, &si
);
1654 if (status
== -EOPNOTSUPP
)
1662 /* calculate meta data/clusters for setting security xattr */
1664 status
= ocfs2_calc_security_init(dir
, &si
, &want_clusters
,
1665 &xattr_credits
, &xattr_ac
);
1672 /* don't reserve bitmap space for fast symlinks. */
1673 if (l
> ocfs2_fast_symlink_chars(sb
))
1676 status
= ocfs2_reserve_clusters(osb
, want_clusters
, &data_ac
);
1678 if (status
!= -ENOSPC
)
1683 handle
= ocfs2_start_trans(osb
, credits
+ xattr_credits
);
1684 if (IS_ERR(handle
)) {
1685 status
= PTR_ERR(handle
);
1691 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1693 if (sb_any_quota_active(osb
->sb
) &&
1694 osb
->sb
->dq_op
->alloc_inode(inode
, 1) == NO_QUOTA
) {
1698 did_quota_inode
= 1;
1700 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
,
1701 inode
->i_mode
, dentry
->d_name
.len
,
1702 dentry
->d_name
.name
);
1704 status
= ocfs2_mknod_locked(osb
, dir
, inode
,
1705 0, &new_fe_bh
, parent_fe_bh
, handle
,
1712 fe
= (struct ocfs2_dinode
*) new_fe_bh
->b_data
;
1715 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1718 inode
->i_op
= &ocfs2_symlink_inode_operations
;
1719 if (vfs_dq_alloc_space_nodirty(inode
,
1720 ocfs2_clusters_to_bytes(osb
->sb
, 1))) {
1725 status
= ocfs2_add_inode_data(osb
, inode
, &offset
, 1, 0,
1727 handle
, data_ac
, NULL
,
1730 if (status
!= -ENOSPC
&& status
!= -EINTR
) {
1732 "Failed to extend file to %llu\n",
1733 (unsigned long long)newsize
);
1739 i_size_write(inode
, newsize
);
1740 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1742 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
1743 memcpy((char *) fe
->id2
.i_symlink
, symname
, l
);
1744 i_size_write(inode
, newsize
);
1745 inode
->i_blocks
= 0;
1748 status
= ocfs2_mark_inode_dirty(handle
, inode
, new_fe_bh
);
1754 if (!ocfs2_inode_is_fast_symlink(inode
)) {
1755 status
= ocfs2_create_symlink_data(osb
, handle
, inode
,
1764 status
= ocfs2_init_security_set(handle
, inode
, new_fe_bh
, &si
,
1772 status
= ocfs2_add_entry(handle
, dentry
, inode
,
1773 le64_to_cpu(fe
->i_blkno
), parent_fe_bh
,
1780 status
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
1786 insert_inode_hash(inode
);
1787 dentry
->d_op
= &ocfs2_dentry_ops
;
1788 d_instantiate(dentry
, inode
);
1790 if (status
< 0 && did_quota
)
1791 vfs_dq_free_space_nodirty(inode
,
1792 ocfs2_clusters_to_bytes(osb
->sb
, 1));
1793 if (status
< 0 && did_quota_inode
)
1794 vfs_dq_free_inode(inode
);
1796 ocfs2_commit_trans(osb
, handle
);
1798 ocfs2_inode_unlock(dir
, 1);
1801 brelse(parent_fe_bh
);
1804 ocfs2_free_dir_lookup_result(&lookup
);
1806 ocfs2_free_alloc_context(inode_ac
);
1808 ocfs2_free_alloc_context(data_ac
);
1810 ocfs2_free_alloc_context(xattr_ac
);
1811 if ((status
< 0) && inode
) {
1821 static int ocfs2_blkno_stringify(u64 blkno
, char *name
)
1823 int status
, namelen
;
1827 namelen
= snprintf(name
, OCFS2_ORPHAN_NAMELEN
+ 1, "%016llx",
1837 if (namelen
!= OCFS2_ORPHAN_NAMELEN
) {
1843 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name
,
1852 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
1853 struct inode
**ret_orphan_dir
,
1856 struct ocfs2_dir_lookup_result
*lookup
)
1858 struct inode
*orphan_dir_inode
;
1859 struct buffer_head
*orphan_dir_bh
= NULL
;
1862 status
= ocfs2_blkno_stringify(blkno
, name
);
1868 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
1869 ORPHAN_DIR_SYSTEM_INODE
,
1871 if (!orphan_dir_inode
) {
1877 mutex_lock(&orphan_dir_inode
->i_mutex
);
1879 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
1885 status
= ocfs2_prepare_dir_for_insert(osb
, orphan_dir_inode
,
1886 orphan_dir_bh
, name
,
1887 OCFS2_ORPHAN_NAMELEN
, lookup
);
1889 ocfs2_inode_unlock(orphan_dir_inode
, 1);
1895 *ret_orphan_dir
= orphan_dir_inode
;
1899 mutex_unlock(&orphan_dir_inode
->i_mutex
);
1900 iput(orphan_dir_inode
);
1903 brelse(orphan_dir_bh
);
1909 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
1911 struct inode
*inode
,
1912 struct ocfs2_dinode
*fe
,
1914 struct ocfs2_dir_lookup_result
*lookup
,
1915 struct inode
*orphan_dir_inode
)
1917 struct buffer_head
*orphan_dir_bh
= NULL
;
1919 struct ocfs2_dinode
*orphan_fe
;
1921 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
1923 status
= ocfs2_read_inode_block(orphan_dir_inode
, &orphan_dir_bh
);
1929 status
= ocfs2_journal_access_di(handle
,
1930 INODE_CACHE(orphan_dir_inode
),
1932 OCFS2_JOURNAL_ACCESS_WRITE
);
1938 /* we're a cluster, and nlink can change on disk from
1939 * underneath us... */
1940 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
1941 if (S_ISDIR(inode
->i_mode
))
1942 ocfs2_add_links_count(orphan_fe
, 1);
1943 orphan_dir_inode
->i_nlink
= ocfs2_read_links_count(orphan_fe
);
1945 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
1951 status
= __ocfs2_add_entry(handle
, orphan_dir_inode
, name
,
1952 OCFS2_ORPHAN_NAMELEN
, inode
,
1953 OCFS2_I(inode
)->ip_blkno
,
1954 orphan_dir_bh
, lookup
);
1960 le32_add_cpu(&fe
->i_flags
, OCFS2_ORPHANED_FL
);
1962 /* Record which orphan dir our inode now resides
1963 * in. delete_inode will use this to determine which orphan
1965 fe
->i_orphaned_slot
= cpu_to_le16(osb
->slot_num
);
1967 mlog(0, "Inode %llu orphaned in slot %d\n",
1968 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, osb
->slot_num
);
1971 brelse(orphan_dir_bh
);
1977 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1978 int ocfs2_orphan_del(struct ocfs2_super
*osb
,
1980 struct inode
*orphan_dir_inode
,
1981 struct inode
*inode
,
1982 struct buffer_head
*orphan_dir_bh
)
1984 char name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1985 struct ocfs2_dinode
*orphan_fe
;
1987 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
1991 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
1997 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1998 name
, (unsigned long long)OCFS2_I(orphan_dir_inode
)->ip_blkno
,
1999 OCFS2_ORPHAN_NAMELEN
);
2001 /* find it's spot in the orphan directory */
2002 status
= ocfs2_find_entry(name
, OCFS2_ORPHAN_NAMELEN
, orphan_dir_inode
,
2009 /* remove it from the orphan directory */
2010 status
= ocfs2_delete_entry(handle
, orphan_dir_inode
, &lookup
);
2016 status
= ocfs2_journal_access_di(handle
,
2017 INODE_CACHE(orphan_dir_inode
),
2019 OCFS2_JOURNAL_ACCESS_WRITE
);
2025 /* do the i_nlink dance! :) */
2026 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2027 if (S_ISDIR(inode
->i_mode
))
2028 ocfs2_add_links_count(orphan_fe
, -1);
2029 orphan_dir_inode
->i_nlink
= ocfs2_read_links_count(orphan_fe
);
2031 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2038 ocfs2_free_dir_lookup_result(&lookup
);
2044 int ocfs2_create_inode_in_orphan(struct inode
*dir
,
2046 struct inode
**new_inode
)
2048 int status
, did_quota_inode
= 0;
2049 struct inode
*inode
= NULL
;
2050 struct inode
*orphan_dir
= NULL
;
2051 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2052 struct ocfs2_dinode
*di
= NULL
;
2053 handle_t
*handle
= NULL
;
2054 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2055 struct buffer_head
*parent_di_bh
= NULL
;
2056 struct buffer_head
*new_di_bh
= NULL
;
2057 struct ocfs2_alloc_context
*inode_ac
= NULL
;
2058 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
2060 status
= ocfs2_inode_lock(dir
, &parent_di_bh
, 1);
2062 if (status
!= -ENOENT
)
2068 * We give the orphan dir the root blkno to fake an orphan name,
2069 * and allocate enough space for our insertion.
2071 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
2073 orphan_name
, &orphan_insert
);
2079 /* reserve an inode spot */
2080 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
2082 if (status
!= -ENOSPC
)
2087 inode
= ocfs2_get_init_inode(dir
, mode
);
2094 handle
= ocfs2_start_trans(osb
, ocfs2_mknod_credits(osb
->sb
, 0, 0));
2095 if (IS_ERR(handle
)) {
2096 status
= PTR_ERR(handle
);
2102 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
2104 if (sb_any_quota_active(osb
->sb
) &&
2105 osb
->sb
->dq_op
->alloc_inode(inode
, 1) == NO_QUOTA
) {
2109 did_quota_inode
= 1;
2111 /* do the real work now. */
2112 status
= ocfs2_mknod_locked(osb
, dir
, inode
,
2113 0, &new_di_bh
, parent_di_bh
, handle
,
2120 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, orphan_name
);
2126 di
= (struct ocfs2_dinode
*)new_di_bh
->b_data
;
2127 status
= ocfs2_orphan_add(osb
, handle
, inode
, di
, orphan_name
,
2128 &orphan_insert
, orphan_dir
);
2134 /* get open lock so that only nodes can't remove it from orphan dir. */
2135 status
= ocfs2_open_lock(inode
);
2140 if (status
< 0 && did_quota_inode
)
2141 vfs_dq_free_inode(inode
);
2143 ocfs2_commit_trans(osb
, handle
);
2146 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2147 ocfs2_inode_unlock(orphan_dir
, 1);
2148 mutex_unlock(&orphan_dir
->i_mutex
);
2152 if (status
== -ENOSPC
)
2153 mlog(0, "Disk is full\n");
2155 if ((status
< 0) && inode
) {
2161 ocfs2_free_alloc_context(inode_ac
);
2168 ocfs2_free_dir_lookup_result(&orphan_insert
);
2170 ocfs2_inode_unlock(dir
, 1);
2171 brelse(parent_di_bh
);
2175 int ocfs2_mv_orphaned_inode_to_new(struct inode
*dir
,
2176 struct inode
*inode
,
2177 struct dentry
*dentry
)
2180 struct buffer_head
*parent_di_bh
= NULL
;
2181 handle_t
*handle
= NULL
;
2182 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2183 struct ocfs2_dinode
*dir_di
, *di
;
2184 struct inode
*orphan_dir_inode
= NULL
;
2185 struct buffer_head
*orphan_dir_bh
= NULL
;
2186 struct buffer_head
*di_bh
= NULL
;
2187 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
2189 mlog_entry("(0x%p, 0x%p, %.*s')\n", dir
, dentry
,
2190 dentry
->d_name
.len
, dentry
->d_name
.name
);
2192 status
= ocfs2_inode_lock(dir
, &parent_di_bh
, 1);
2194 if (status
!= -ENOENT
)
2199 dir_di
= (struct ocfs2_dinode
*) parent_di_bh
->b_data
;
2200 if (!dir_di
->i_links_count
) {
2201 /* can't make a file in a deleted directory. */
2206 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
2207 dentry
->d_name
.len
);
2211 /* get a spot inside the dir. */
2212 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_di_bh
,
2213 dentry
->d_name
.name
,
2214 dentry
->d_name
.len
, &lookup
);
2220 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2221 ORPHAN_DIR_SYSTEM_INODE
,
2223 if (!orphan_dir_inode
) {
2229 mutex_lock(&orphan_dir_inode
->i_mutex
);
2231 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
2234 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2235 iput(orphan_dir_inode
);
2239 status
= ocfs2_read_inode_block(inode
, &di_bh
);
2245 handle
= ocfs2_start_trans(osb
, ocfs2_rename_credits(osb
->sb
));
2246 if (IS_ERR(handle
)) {
2247 status
= PTR_ERR(handle
);
2253 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
2254 di_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
2260 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
2267 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2268 le32_add_cpu(&di
->i_flags
, -OCFS2_ORPHANED_FL
);
2269 di
->i_orphaned_slot
= 0;
2270 ocfs2_journal_dirty(handle
, di_bh
);
2272 status
= ocfs2_add_entry(handle
, dentry
, inode
,
2273 OCFS2_I(inode
)->ip_blkno
, parent_di_bh
,
2280 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
2281 OCFS2_I(dir
)->ip_blkno
);
2287 insert_inode_hash(inode
);
2288 dentry
->d_op
= &ocfs2_dentry_ops
;
2289 d_instantiate(dentry
, inode
);
2292 ocfs2_commit_trans(osb
, handle
);
2294 ocfs2_inode_unlock(orphan_dir_inode
, 1);
2295 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2296 iput(orphan_dir_inode
);
2299 ocfs2_inode_unlock(dir
, 1);
2302 brelse(parent_di_bh
);
2303 brelse(orphan_dir_bh
);
2305 ocfs2_free_dir_lookup_result(&lookup
);
2312 const struct inode_operations ocfs2_dir_iops
= {
2313 .create
= ocfs2_create
,
2314 .lookup
= ocfs2_lookup
,
2316 .unlink
= ocfs2_unlink
,
2317 .rmdir
= ocfs2_unlink
,
2318 .symlink
= ocfs2_symlink
,
2319 .mkdir
= ocfs2_mkdir
,
2320 .mknod
= ocfs2_mknod
,
2321 .rename
= ocfs2_rename
,
2322 .setattr
= ocfs2_setattr
,
2323 .getattr
= ocfs2_getattr
,
2324 .permission
= ocfs2_permission
,
2325 .setxattr
= generic_setxattr
,
2326 .getxattr
= generic_getxattr
,
2327 .listxattr
= ocfs2_listxattr
,
2328 .removexattr
= generic_removexattr
,