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 buffer_head
*fe_bh
,
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 ret
= d_splice_alias(inode
, dentry
);
154 * If d_splice_alias() finds a DCACHE_DISCONNECTED
155 * dentry, it will d_move() it on top of ourse. The
156 * return value will indicate this however, so in
157 * those cases, we switch them around for the locking
160 * NOTE: This dentry already has ->d_op set from
161 * ocfs2_get_parent() and ocfs2_get_dentry()
166 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
167 OCFS2_I(dir
)->ip_blkno
);
170 ret
= ERR_PTR(status
);
174 ocfs2_dentry_attach_gen(dentry
);
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_init_owner(inode
, dir
, mode
);
208 dquot_initialize(inode
);
212 static int ocfs2_mknod(struct inode
*dir
,
213 struct dentry
*dentry
,
218 struct buffer_head
*parent_fe_bh
= NULL
;
219 handle_t
*handle
= NULL
;
220 struct ocfs2_super
*osb
;
221 struct ocfs2_dinode
*dirfe
;
222 struct buffer_head
*new_fe_bh
= NULL
;
223 struct inode
*inode
= NULL
;
224 struct ocfs2_alloc_context
*inode_ac
= NULL
;
225 struct ocfs2_alloc_context
*data_ac
= NULL
;
226 struct ocfs2_alloc_context
*meta_ac
= NULL
;
227 int want_clusters
= 0;
229 int xattr_credits
= 0;
230 struct ocfs2_security_xattr_info si
= {
233 int did_quota_inode
= 0;
234 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
236 int did_block_signals
= 0;
238 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
, mode
,
239 (unsigned long)dev
, dentry
->d_name
.len
,
240 dentry
->d_name
.name
);
242 dquot_initialize(dir
);
244 /* get our super block */
245 osb
= OCFS2_SB(dir
->i_sb
);
247 status
= ocfs2_inode_lock(dir
, &parent_fe_bh
, 1);
249 if (status
!= -ENOENT
)
254 if (S_ISDIR(mode
) && (dir
->i_nlink
>= ocfs2_link_max(osb
))) {
259 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
260 if (!ocfs2_read_links_count(dirfe
)) {
261 /* can't make a file in a deleted directory. */
266 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
271 /* get a spot inside the dir. */
272 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
274 dentry
->d_name
.len
, &lookup
);
280 /* reserve an inode spot */
281 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
283 if (status
!= -ENOSPC
)
288 inode
= ocfs2_get_init_inode(dir
, mode
);
295 /* get security xattr */
296 status
= ocfs2_init_security_get(inode
, dir
, &si
);
298 if (status
== -EOPNOTSUPP
)
306 /* calculate meta data/clusters for setting security and acl xattr */
307 status
= ocfs2_calc_xattr_init(dir
, parent_fe_bh
, mode
,
309 &xattr_credits
, &want_meta
);
315 /* Reserve a cluster if creating an extent based directory. */
316 if (S_ISDIR(mode
) && !ocfs2_supports_inline_data(osb
)) {
319 /* Dir indexing requires extra space as well */
320 if (ocfs2_supports_indexed_dirs(osb
))
324 status
= ocfs2_reserve_new_metadata_blocks(osb
, want_meta
, &meta_ac
);
326 if (status
!= -ENOSPC
)
331 status
= ocfs2_reserve_clusters(osb
, want_clusters
, &data_ac
);
333 if (status
!= -ENOSPC
)
338 handle
= ocfs2_start_trans(osb
, ocfs2_mknod_credits(osb
->sb
,
341 if (IS_ERR(handle
)) {
342 status
= PTR_ERR(handle
);
348 /* Starting to change things, restart is no longer possible. */
349 ocfs2_block_signals(&oldset
);
350 did_block_signals
= 1;
352 status
= dquot_alloc_inode(inode
);
357 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
,
358 inode
->i_mode
, (unsigned long)dev
, dentry
->d_name
.len
,
359 dentry
->d_name
.name
);
361 /* do the real work now. */
362 status
= ocfs2_mknod_locked(osb
, dir
, inode
, dev
,
363 &new_fe_bh
, parent_fe_bh
, handle
,
371 status
= ocfs2_fill_new_dir(osb
, handle
, dir
, inode
,
372 new_fe_bh
, data_ac
, meta_ac
);
378 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(dir
),
380 OCFS2_JOURNAL_ACCESS_WRITE
);
385 ocfs2_add_links_count(dirfe
, 1);
386 ocfs2_journal_dirty(handle
, parent_fe_bh
);
390 status
= ocfs2_init_acl(handle
, inode
, dir
, new_fe_bh
, parent_fe_bh
,
398 status
= ocfs2_init_security_set(handle
, inode
, new_fe_bh
, &si
,
407 * Do this before adding the entry to the directory. We add
408 * also set d_op after success so that ->d_iput() will cleanup
409 * the dentry lock even if ocfs2_add_entry() fails below.
411 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
412 OCFS2_I(dir
)->ip_blkno
);
418 status
= ocfs2_add_entry(handle
, dentry
, inode
,
419 OCFS2_I(inode
)->ip_blkno
, parent_fe_bh
,
426 insert_inode_hash(inode
);
427 d_instantiate(dentry
, inode
);
430 if (status
< 0 && did_quota_inode
)
431 dquot_free_inode(inode
);
433 ocfs2_commit_trans(osb
, handle
);
435 ocfs2_inode_unlock(dir
, 1);
436 if (did_block_signals
)
437 ocfs2_unblock_signals(&oldset
);
439 if (status
== -ENOSPC
)
440 mlog(0, "Disk is full\n");
443 brelse(parent_fe_bh
);
447 ocfs2_free_dir_lookup_result(&lookup
);
450 ocfs2_free_alloc_context(inode_ac
);
453 ocfs2_free_alloc_context(data_ac
);
456 ocfs2_free_alloc_context(meta_ac
);
459 * We should call iput after the i_mutex of the bitmap been
460 * unlocked in ocfs2_free_alloc_context, or the
461 * ocfs2_delete_inode will mutex_lock again.
463 if ((status
< 0) && inode
) {
464 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SKIP_ORPHAN_DIR
;
474 static int __ocfs2_mknod_locked(struct inode
*dir
,
477 struct buffer_head
**new_fe_bh
,
478 struct buffer_head
*parent_fe_bh
,
480 struct ocfs2_alloc_context
*inode_ac
,
481 u64 fe_blkno
, u64 suballoc_loc
, u16 suballoc_bit
)
484 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
485 struct ocfs2_dinode
*fe
= NULL
;
486 struct ocfs2_extent_list
*fel
;
491 /* populate as many fields early on as possible - many of
492 * these are used by the support functions here and in
494 inode
->i_ino
= ino_from_blkno(osb
->sb
, fe_blkno
);
495 OCFS2_I(inode
)->ip_blkno
= fe_blkno
;
496 spin_lock(&osb
->osb_lock
);
497 inode
->i_generation
= osb
->s_next_generation
++;
498 spin_unlock(&osb
->osb_lock
);
500 *new_fe_bh
= sb_getblk(osb
->sb
, fe_blkno
);
506 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
), *new_fe_bh
);
508 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
510 OCFS2_JOURNAL_ACCESS_CREATE
);
516 fe
= (struct ocfs2_dinode
*) (*new_fe_bh
)->b_data
;
517 memset(fe
, 0, osb
->sb
->s_blocksize
);
519 fe
->i_generation
= cpu_to_le32(inode
->i_generation
);
520 fe
->i_fs_generation
= cpu_to_le32(osb
->fs_generation
);
521 fe
->i_blkno
= cpu_to_le64(fe_blkno
);
522 fe
->i_suballoc_loc
= cpu_to_le64(suballoc_loc
);
523 fe
->i_suballoc_bit
= cpu_to_le16(suballoc_bit
);
524 fe
->i_suballoc_slot
= cpu_to_le16(inode_ac
->ac_alloc_slot
);
525 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
526 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
527 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
528 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
))
529 fe
->id1
.dev1
.i_rdev
= cpu_to_le64(huge_encode_dev(dev
));
531 ocfs2_set_links_count(fe
, inode
->i_nlink
);
533 fe
->i_last_eb_blk
= 0;
534 strcpy(fe
->i_signature
, OCFS2_INODE_SIGNATURE
);
535 le32_add_cpu(&fe
->i_flags
, OCFS2_VALID_FL
);
536 fe
->i_atime
= fe
->i_ctime
= fe
->i_mtime
=
537 cpu_to_le64(CURRENT_TIME
.tv_sec
);
538 fe
->i_mtime_nsec
= fe
->i_ctime_nsec
= fe
->i_atime_nsec
=
539 cpu_to_le32(CURRENT_TIME
.tv_nsec
);
543 * If supported, directories start with inline data. If inline
544 * isn't supported, but indexing is, we start them as indexed.
546 feat
= le16_to_cpu(fe
->i_dyn_features
);
547 if (S_ISDIR(inode
->i_mode
) && ocfs2_supports_inline_data(osb
)) {
548 fe
->i_dyn_features
= cpu_to_le16(feat
| OCFS2_INLINE_DATA_FL
);
550 fe
->id2
.i_data
.id_count
= cpu_to_le16(
551 ocfs2_max_inline_data_with_xattr(osb
->sb
, fe
));
553 fel
= &fe
->id2
.i_list
;
554 fel
->l_tree_depth
= 0;
555 fel
->l_next_free_rec
= 0;
556 fel
->l_count
= cpu_to_le16(ocfs2_extent_recs_per_inode(osb
->sb
));
559 ocfs2_journal_dirty(handle
, *new_fe_bh
);
561 ocfs2_populate_inode(inode
, fe
, 1);
562 ocfs2_ci_set_new(osb
, INODE_CACHE(inode
));
563 if (!ocfs2_mount_local(osb
)) {
564 status
= ocfs2_create_new_inode_locks(inode
);
569 status
= 0; /* error in ocfs2_create_new_inode_locks is not
584 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
588 struct buffer_head
**new_fe_bh
,
589 struct buffer_head
*parent_fe_bh
,
591 struct ocfs2_alloc_context
*inode_ac
)
594 u64 suballoc_loc
, fe_blkno
= 0;
599 status
= ocfs2_claim_new_inode(handle
, dir
, parent_fe_bh
,
600 inode_ac
, &suballoc_loc
,
601 &suballoc_bit
, &fe_blkno
);
607 return __ocfs2_mknod_locked(dir
, inode
, dev
, new_fe_bh
,
608 parent_fe_bh
, handle
, inode_ac
,
609 fe_blkno
, suballoc_loc
, suballoc_bit
);
612 static int ocfs2_mkdir(struct inode
*dir
,
613 struct dentry
*dentry
,
618 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
619 dentry
->d_name
.len
, dentry
->d_name
.name
);
620 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
626 static int ocfs2_create(struct inode
*dir
,
627 struct dentry
*dentry
,
629 struct nameidata
*nd
)
633 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
634 dentry
->d_name
.len
, dentry
->d_name
.name
);
635 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
641 static int ocfs2_link(struct dentry
*old_dentry
,
643 struct dentry
*dentry
)
646 struct inode
*inode
= old_dentry
->d_inode
;
648 struct buffer_head
*fe_bh
= NULL
;
649 struct buffer_head
*parent_fe_bh
= NULL
;
650 struct ocfs2_dinode
*fe
= NULL
;
651 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
652 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
655 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode
->i_ino
,
656 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
657 dentry
->d_name
.len
, dentry
->d_name
.name
);
659 if (S_ISDIR(inode
->i_mode
))
662 dquot_initialize(dir
);
664 err
= ocfs2_inode_lock_nested(dir
, &parent_fe_bh
, 1, OI_LS_PARENT
);
676 err
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
681 err
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
683 dentry
->d_name
.len
, &lookup
);
689 err
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
696 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
697 if (ocfs2_read_links_count(fe
) >= ocfs2_link_max(osb
)) {
699 goto out_unlock_inode
;
702 handle
= ocfs2_start_trans(osb
, ocfs2_link_credits(osb
->sb
));
703 if (IS_ERR(handle
)) {
704 err
= PTR_ERR(handle
);
707 goto out_unlock_inode
;
710 /* Starting to change things, restart is no longer possible. */
711 ocfs2_block_signals(&oldset
);
713 err
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
714 OCFS2_JOURNAL_ACCESS_WRITE
);
721 inode
->i_ctime
= CURRENT_TIME
;
722 ocfs2_set_links_count(fe
, inode
->i_nlink
);
723 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
724 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
725 ocfs2_journal_dirty(handle
, fe_bh
);
727 err
= ocfs2_add_entry(handle
, dentry
, inode
,
728 OCFS2_I(inode
)->ip_blkno
,
729 parent_fe_bh
, &lookup
);
731 ocfs2_add_links_count(fe
, -1);
737 err
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
744 d_instantiate(dentry
, inode
);
747 ocfs2_commit_trans(osb
, handle
);
748 ocfs2_unblock_signals(&oldset
);
750 ocfs2_inode_unlock(inode
, 1);
753 ocfs2_inode_unlock(dir
, 1);
756 brelse(parent_fe_bh
);
758 ocfs2_free_dir_lookup_result(&lookup
);
766 * Takes and drops an exclusive lock on the given dentry. This will
767 * force other nodes to drop it.
769 static int ocfs2_remote_dentry_delete(struct dentry
*dentry
)
773 ret
= ocfs2_dentry_lock(dentry
, 1);
777 ocfs2_dentry_unlock(dentry
, 1);
782 static inline int inode_is_unlinkable(struct inode
*inode
)
784 if (S_ISDIR(inode
->i_mode
)) {
785 if (inode
->i_nlink
== 2)
790 if (inode
->i_nlink
== 1)
795 static int ocfs2_unlink(struct inode
*dir
,
796 struct dentry
*dentry
)
799 int child_locked
= 0;
800 struct inode
*inode
= dentry
->d_inode
;
801 struct inode
*orphan_dir
= NULL
;
802 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
804 struct ocfs2_dinode
*fe
= NULL
;
805 struct buffer_head
*fe_bh
= NULL
;
806 struct buffer_head
*parent_node_bh
= NULL
;
807 handle_t
*handle
= NULL
;
808 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
809 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
810 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
812 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
813 dentry
->d_name
.len
, dentry
->d_name
.name
);
815 dquot_initialize(dir
);
817 BUG_ON(dentry
->d_parent
->d_inode
!= dir
);
819 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
821 if (inode
== osb
->root_inode
) {
822 mlog(0, "Cannot delete the root directory\n");
826 status
= ocfs2_inode_lock_nested(dir
, &parent_node_bh
, 1,
829 if (status
!= -ENOENT
)
834 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
835 dentry
->d_name
.len
, &blkno
, dir
,
838 if (status
!= -ENOENT
)
843 if (OCFS2_I(inode
)->ip_blkno
!= blkno
) {
846 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
847 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
848 (unsigned long long)blkno
, OCFS2_I(inode
)->ip_flags
);
852 status
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
854 if (status
!= -ENOENT
)
860 if (S_ISDIR(inode
->i_mode
)) {
861 if (inode
->i_nlink
!= 2 || !ocfs2_empty_dir(inode
)) {
867 status
= ocfs2_remote_dentry_delete(dentry
);
869 /* This remote delete should succeed under all normal
875 if (inode_is_unlinkable(inode
)) {
876 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
877 OCFS2_I(inode
)->ip_blkno
,
878 orphan_name
, &orphan_insert
);
885 handle
= ocfs2_start_trans(osb
, ocfs2_unlink_credits(osb
->sb
));
886 if (IS_ERR(handle
)) {
887 status
= PTR_ERR(handle
);
893 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
894 OCFS2_JOURNAL_ACCESS_WRITE
);
900 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
902 if (inode_is_unlinkable(inode
)) {
903 status
= ocfs2_orphan_add(osb
, handle
, inode
, fe_bh
, orphan_name
,
904 &orphan_insert
, orphan_dir
);
911 /* delete the name from the parent dir */
912 status
= ocfs2_delete_entry(handle
, dir
, &lookup
);
918 if (S_ISDIR(inode
->i_mode
))
921 ocfs2_set_links_count(fe
, inode
->i_nlink
);
922 ocfs2_journal_dirty(handle
, fe_bh
);
924 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
925 if (S_ISDIR(inode
->i_mode
))
928 status
= ocfs2_mark_inode_dirty(handle
, dir
, parent_node_bh
);
931 if (S_ISDIR(inode
->i_mode
))
937 ocfs2_commit_trans(osb
, handle
);
940 ocfs2_inode_unlock(inode
, 1);
942 ocfs2_inode_unlock(dir
, 1);
945 /* This was locked for us in ocfs2_prepare_orphan_dir() */
946 ocfs2_inode_unlock(orphan_dir
, 1);
947 mutex_unlock(&orphan_dir
->i_mutex
);
952 brelse(parent_node_bh
);
954 ocfs2_free_dir_lookup_result(&orphan_insert
);
955 ocfs2_free_dir_lookup_result(&lookup
);
963 * The only place this should be used is rename!
964 * if they have the same id, then the 1st one is the only one locked.
966 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
967 struct buffer_head
**bh1
,
968 struct inode
*inode1
,
969 struct buffer_head
**bh2
,
970 struct inode
*inode2
)
973 struct ocfs2_inode_info
*oi1
= OCFS2_I(inode1
);
974 struct ocfs2_inode_info
*oi2
= OCFS2_I(inode2
);
975 struct buffer_head
**tmpbh
;
976 struct inode
*tmpinode
;
978 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
979 (unsigned long long)oi1
->ip_blkno
,
980 (unsigned long long)oi2
->ip_blkno
);
987 /* we always want to lock the one with the lower lockid first. */
988 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
989 if (oi1
->ip_blkno
< oi2
->ip_blkno
) {
990 /* switch id1 and id2 around */
991 mlog(0, "switching them around...\n");
1001 status
= ocfs2_inode_lock_nested(inode2
, bh2
, 1,
1004 if (status
!= -ENOENT
)
1011 status
= ocfs2_inode_lock_nested(inode1
, bh1
, 1, OI_LS_RENAME2
);
1014 * An error return must mean that no cluster locks
1015 * were held on function exit.
1017 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
1018 ocfs2_inode_unlock(inode2
, 1);
1023 if (status
!= -ENOENT
)
1032 static void ocfs2_double_unlock(struct inode
*inode1
, struct inode
*inode2
)
1034 ocfs2_inode_unlock(inode1
, 1);
1036 if (inode1
!= inode2
)
1037 ocfs2_inode_unlock(inode2
, 1);
1040 static int ocfs2_rename(struct inode
*old_dir
,
1041 struct dentry
*old_dentry
,
1042 struct inode
*new_dir
,
1043 struct dentry
*new_dentry
)
1045 int status
= 0, rename_lock
= 0, parents_locked
= 0, target_exists
= 0;
1046 int old_child_locked
= 0, new_child_locked
= 0, update_dot_dot
= 0;
1047 struct inode
*old_inode
= old_dentry
->d_inode
;
1048 struct inode
*new_inode
= new_dentry
->d_inode
;
1049 struct inode
*orphan_dir
= NULL
;
1050 struct ocfs2_dinode
*newfe
= NULL
;
1051 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1052 struct buffer_head
*newfe_bh
= NULL
;
1053 struct buffer_head
*old_inode_bh
= NULL
;
1054 struct ocfs2_super
*osb
= NULL
;
1055 u64 newfe_blkno
, old_de_ino
;
1056 handle_t
*handle
= NULL
;
1057 struct buffer_head
*old_dir_bh
= NULL
;
1058 struct buffer_head
*new_dir_bh
= NULL
;
1059 nlink_t old_dir_nlink
= old_dir
->i_nlink
;
1060 struct ocfs2_dinode
*old_di
;
1061 struct ocfs2_dir_lookup_result old_inode_dot_dot_res
= { NULL
, };
1062 struct ocfs2_dir_lookup_result target_lookup_res
= { NULL
, };
1063 struct ocfs2_dir_lookup_result old_entry_lookup
= { NULL
, };
1064 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
1065 struct ocfs2_dir_lookup_result target_insert
= { NULL
, };
1067 /* At some point it might be nice to break this function up a
1070 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1071 old_dir
, old_dentry
, new_dir
, new_dentry
,
1072 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
1073 new_dentry
->d_name
.len
, new_dentry
->d_name
.name
);
1075 dquot_initialize(old_dir
);
1076 dquot_initialize(new_dir
);
1078 osb
= OCFS2_SB(old_dir
->i_sb
);
1081 if (!igrab(new_inode
))
1085 /* Assume a directory hierarchy thusly:
1088 * a,b,c, and d are all directories.
1090 * from cwd of 'a' on both nodes:
1094 * And that's why, just like the VFS, we need a file system
1096 if (old_dir
!= new_dir
&& S_ISDIR(old_inode
->i_mode
)) {
1097 status
= ocfs2_rename_lock(osb
);
1105 /* if old and new are the same, this'll just do one lock. */
1106 status
= ocfs2_double_lock(osb
, &old_dir_bh
, old_dir
,
1107 &new_dir_bh
, new_dir
);
1114 /* make sure both dirs have bhs
1115 * get an extra ref on old_dir_bh if old==new */
1118 new_dir_bh
= old_dir_bh
;
1121 mlog(ML_ERROR
, "no old_dir_bh!\n");
1128 * Aside from allowing a meta data update, the locking here
1129 * also ensures that the downconvert thread on other nodes
1130 * won't have to concurrently downconvert the inode and the
1133 status
= ocfs2_inode_lock_nested(old_inode
, &old_inode_bh
, 1,
1136 if (status
!= -ENOENT
)
1140 old_child_locked
= 1;
1142 status
= ocfs2_remote_dentry_delete(old_dentry
);
1148 if (S_ISDIR(old_inode
->i_mode
)) {
1149 u64 old_inode_parent
;
1152 status
= ocfs2_find_files_on_disk("..", 2, &old_inode_parent
,
1154 &old_inode_dot_dot_res
);
1160 if (old_inode_parent
!= OCFS2_I(old_dir
)->ip_blkno
) {
1165 if (!new_inode
&& new_dir
!= old_dir
&&
1166 new_dir
->i_nlink
>= ocfs2_link_max(osb
)) {
1172 status
= ocfs2_lookup_ino_from_name(old_dir
, old_dentry
->d_name
.name
,
1173 old_dentry
->d_name
.len
,
1181 * Check for inode number is _not_ due to possible IO errors.
1182 * We might rmdir the source, keep it as pwd of some process
1183 * and merrily kill the link to whatever was created under the
1184 * same name. Goodbye sticky bit ;-<
1186 if (old_de_ino
!= OCFS2_I(old_inode
)->ip_blkno
) {
1191 /* check if the target already exists (in which case we need
1193 status
= ocfs2_find_files_on_disk(new_dentry
->d_name
.name
,
1194 new_dentry
->d_name
.len
,
1195 &newfe_blkno
, new_dir
,
1196 &target_lookup_res
);
1197 /* The only error we allow here is -ENOENT because the new
1198 * file not existing is perfectly valid. */
1199 if ((status
< 0) && (status
!= -ENOENT
)) {
1200 /* If we cannot find the file specified we should just */
1201 /* return the error... */
1208 if (!target_exists
&& new_inode
) {
1210 * Target was unlinked by another node while we were
1211 * waiting to get to ocfs2_rename(). There isn't
1212 * anything we can do here to help the situation, so
1213 * bubble up the appropriate error.
1219 /* In case we need to overwrite an existing file, we blow it
1221 if (target_exists
) {
1222 /* VFS didn't think there existed an inode here, but
1223 * someone else in the cluster must have raced our
1224 * rename to create one. Today we error cleanly, in
1225 * the future we should consider calling iget to build
1226 * a new struct inode for this entry. */
1230 mlog(0, "We found an inode for name %.*s but VFS "
1231 "didn't give us one.\n", new_dentry
->d_name
.len
,
1232 new_dentry
->d_name
.name
);
1236 if (OCFS2_I(new_inode
)->ip_blkno
!= newfe_blkno
) {
1239 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1240 (unsigned long long)OCFS2_I(new_inode
)->ip_blkno
,
1241 (unsigned long long)newfe_blkno
,
1242 OCFS2_I(new_inode
)->ip_flags
);
1246 status
= ocfs2_inode_lock(new_inode
, &newfe_bh
, 1);
1248 if (status
!= -ENOENT
)
1252 new_child_locked
= 1;
1254 status
= ocfs2_remote_dentry_delete(new_dentry
);
1260 newfe
= (struct ocfs2_dinode
*) newfe_bh
->b_data
;
1262 mlog(0, "aha rename over existing... new_blkno=%llu "
1263 "newfebh=%p bhblocknr=%llu\n",
1264 (unsigned long long)newfe_blkno
, newfe_bh
, newfe_bh
?
1265 (unsigned long long)newfe_bh
->b_blocknr
: 0ULL);
1267 if (S_ISDIR(new_inode
->i_mode
) || (new_inode
->i_nlink
== 1)) {
1268 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
1269 OCFS2_I(new_inode
)->ip_blkno
,
1270 orphan_name
, &orphan_insert
);
1277 BUG_ON(new_dentry
->d_parent
->d_inode
!= new_dir
);
1279 status
= ocfs2_check_dir_for_entry(new_dir
,
1280 new_dentry
->d_name
.name
,
1281 new_dentry
->d_name
.len
);
1285 status
= ocfs2_prepare_dir_for_insert(osb
, new_dir
, new_dir_bh
,
1286 new_dentry
->d_name
.name
,
1287 new_dentry
->d_name
.len
,
1295 handle
= ocfs2_start_trans(osb
, ocfs2_rename_credits(osb
->sb
));
1296 if (IS_ERR(handle
)) {
1297 status
= PTR_ERR(handle
);
1303 if (target_exists
) {
1304 if (S_ISDIR(new_inode
->i_mode
)) {
1305 if (new_inode
->i_nlink
!= 2 ||
1306 !ocfs2_empty_dir(new_inode
)) {
1307 status
= -ENOTEMPTY
;
1311 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(new_inode
),
1313 OCFS2_JOURNAL_ACCESS_WRITE
);
1319 if (S_ISDIR(new_inode
->i_mode
) ||
1320 (ocfs2_read_links_count(newfe
) == 1)) {
1321 status
= ocfs2_orphan_add(osb
, handle
, new_inode
,
1322 newfe_bh
, orphan_name
,
1323 &orphan_insert
, orphan_dir
);
1330 /* change the dirent to point to the correct inode */
1331 status
= ocfs2_update_entry(new_dir
, handle
, &target_lookup_res
,
1337 new_dir
->i_version
++;
1339 if (S_ISDIR(new_inode
->i_mode
))
1340 ocfs2_set_links_count(newfe
, 0);
1342 ocfs2_add_links_count(newfe
, -1);
1343 ocfs2_journal_dirty(handle
, newfe_bh
);
1345 /* if the name was not found in new_dir, add it now */
1346 status
= ocfs2_add_entry(handle
, new_dentry
, old_inode
,
1347 OCFS2_I(old_inode
)->ip_blkno
,
1348 new_dir_bh
, &target_insert
);
1351 old_inode
->i_ctime
= CURRENT_TIME
;
1352 mark_inode_dirty(old_inode
);
1354 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(old_inode
),
1356 OCFS2_JOURNAL_ACCESS_WRITE
);
1358 old_di
= (struct ocfs2_dinode
*) old_inode_bh
->b_data
;
1360 old_di
->i_ctime
= cpu_to_le64(old_inode
->i_ctime
.tv_sec
);
1361 old_di
->i_ctime_nsec
= cpu_to_le32(old_inode
->i_ctime
.tv_nsec
);
1362 ocfs2_journal_dirty(handle
, old_inode_bh
);
1367 * Now that the name has been added to new_dir, remove the old name.
1369 * We don't keep any directory entry context around until now
1370 * because the insert might have changed the type of directory
1371 * we're dealing with.
1373 status
= ocfs2_find_entry(old_dentry
->d_name
.name
,
1374 old_dentry
->d_name
.len
, old_dir
,
1379 status
= ocfs2_delete_entry(handle
, old_dir
, &old_entry_lookup
);
1386 new_inode
->i_nlink
--;
1387 new_inode
->i_ctime
= CURRENT_TIME
;
1389 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1391 if (update_dot_dot
) {
1392 status
= ocfs2_update_entry(old_inode
, handle
,
1393 &old_inode_dot_dot_res
, new_dir
);
1396 new_inode
->i_nlink
--;
1399 mark_inode_dirty(new_dir
);
1402 mark_inode_dirty(old_dir
);
1403 ocfs2_mark_inode_dirty(handle
, old_dir
, old_dir_bh
);
1405 mark_inode_dirty(new_inode
);
1406 ocfs2_mark_inode_dirty(handle
, new_inode
, newfe_bh
);
1409 if (old_dir
!= new_dir
) {
1410 /* Keep the same times on both directories.*/
1411 new_dir
->i_ctime
= new_dir
->i_mtime
= old_dir
->i_ctime
;
1414 * This will also pick up the i_nlink change from the
1417 ocfs2_mark_inode_dirty(handle
, new_dir
, new_dir_bh
);
1420 if (old_dir_nlink
!= old_dir
->i_nlink
) {
1422 mlog(ML_ERROR
, "need to change nlink for old dir "
1423 "%llu from %d to %d but bh is NULL!\n",
1424 (unsigned long long)OCFS2_I(old_dir
)->ip_blkno
,
1425 (int)old_dir_nlink
, old_dir
->i_nlink
);
1427 struct ocfs2_dinode
*fe
;
1428 status
= ocfs2_journal_access_di(handle
,
1429 INODE_CACHE(old_dir
),
1431 OCFS2_JOURNAL_ACCESS_WRITE
);
1432 fe
= (struct ocfs2_dinode
*) old_dir_bh
->b_data
;
1433 ocfs2_set_links_count(fe
, old_dir
->i_nlink
);
1434 ocfs2_journal_dirty(handle
, old_dir_bh
);
1437 ocfs2_dentry_move(old_dentry
, new_dentry
, old_dir
, new_dir
);
1441 ocfs2_rename_unlock(osb
);
1444 ocfs2_commit_trans(osb
, handle
);
1447 ocfs2_double_unlock(old_dir
, new_dir
);
1449 if (old_child_locked
)
1450 ocfs2_inode_unlock(old_inode
, 1);
1452 if (new_child_locked
)
1453 ocfs2_inode_unlock(new_inode
, 1);
1456 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1457 ocfs2_inode_unlock(orphan_dir
, 1);
1458 mutex_unlock(&orphan_dir
->i_mutex
);
1463 sync_mapping_buffers(old_inode
->i_mapping
);
1468 ocfs2_free_dir_lookup_result(&target_lookup_res
);
1469 ocfs2_free_dir_lookup_result(&old_entry_lookup
);
1470 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res
);
1471 ocfs2_free_dir_lookup_result(&orphan_insert
);
1472 ocfs2_free_dir_lookup_result(&target_insert
);
1475 brelse(old_inode_bh
);
1485 * we expect i_size = strlen(symname). Copy symname into the file
1486 * data, including the null terminator.
1488 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
1490 struct inode
*inode
,
1491 const char *symname
)
1493 struct buffer_head
**bhs
= NULL
;
1495 struct super_block
*sb
= osb
->sb
;
1496 u64 p_blkno
, p_blocks
;
1497 int virtual, blocks
, status
, i
, bytes_left
;
1499 bytes_left
= i_size_read(inode
) + 1;
1500 /* we can't trust i_blocks because we're actually going to
1501 * write i_size + 1 bytes. */
1502 blocks
= (bytes_left
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
1504 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1505 (unsigned long long)inode
->i_blocks
,
1506 i_size_read(inode
), blocks
);
1508 /* Sanity check -- make sure we're going to fit. */
1510 ocfs2_clusters_to_bytes(sb
, OCFS2_I(inode
)->ip_clusters
)) {
1516 bhs
= kcalloc(blocks
, sizeof(struct buffer_head
*), GFP_KERNEL
);
1523 status
= ocfs2_extent_map_get_blocks(inode
, 0, &p_blkno
, &p_blocks
,
1530 /* links can never be larger than one cluster so we know this
1531 * is all going to be contiguous, but do a sanity check
1533 if ((p_blocks
<< sb
->s_blocksize_bits
) < bytes_left
) {
1540 while(bytes_left
> 0) {
1541 c
= &symname
[virtual * sb
->s_blocksize
];
1543 bhs
[virtual] = sb_getblk(sb
, p_blkno
);
1544 if (!bhs
[virtual]) {
1549 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
),
1552 status
= ocfs2_journal_access(handle
, INODE_CACHE(inode
),
1554 OCFS2_JOURNAL_ACCESS_CREATE
);
1560 memset(bhs
[virtual]->b_data
, 0, sb
->s_blocksize
);
1562 memcpy(bhs
[virtual]->b_data
, c
,
1563 (bytes_left
> sb
->s_blocksize
) ? sb
->s_blocksize
:
1566 ocfs2_journal_dirty(handle
, bhs
[virtual]);
1570 bytes_left
-= sb
->s_blocksize
;
1577 for(i
= 0; i
< blocks
; i
++)
1586 static int ocfs2_symlink(struct inode
*dir
,
1587 struct dentry
*dentry
,
1588 const char *symname
)
1590 int status
, l
, credits
;
1592 struct ocfs2_super
*osb
= NULL
;
1593 struct inode
*inode
= NULL
;
1594 struct super_block
*sb
;
1595 struct buffer_head
*new_fe_bh
= NULL
;
1596 struct buffer_head
*parent_fe_bh
= NULL
;
1597 struct ocfs2_dinode
*fe
= NULL
;
1598 struct ocfs2_dinode
*dirfe
;
1599 handle_t
*handle
= NULL
;
1600 struct ocfs2_alloc_context
*inode_ac
= NULL
;
1601 struct ocfs2_alloc_context
*data_ac
= NULL
;
1602 struct ocfs2_alloc_context
*xattr_ac
= NULL
;
1603 int want_clusters
= 0;
1604 int xattr_credits
= 0;
1605 struct ocfs2_security_xattr_info si
= {
1608 int did_quota
= 0, did_quota_inode
= 0;
1609 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
1611 int did_block_signals
= 0;
1613 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir
,
1614 dentry
, symname
, dentry
->d_name
.len
, dentry
->d_name
.name
);
1616 dquot_initialize(dir
);
1621 l
= strlen(symname
) + 1;
1623 credits
= ocfs2_calc_symlink_credits(sb
);
1625 /* lock the parent directory */
1626 status
= ocfs2_inode_lock(dir
, &parent_fe_bh
, 1);
1628 if (status
!= -ENOENT
)
1633 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
1634 if (!ocfs2_read_links_count(dirfe
)) {
1635 /* can't make a file in a deleted directory. */
1640 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
1641 dentry
->d_name
.len
);
1645 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
1646 dentry
->d_name
.name
,
1647 dentry
->d_name
.len
, &lookup
);
1653 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
1655 if (status
!= -ENOSPC
)
1660 inode
= ocfs2_get_init_inode(dir
, S_IFLNK
| S_IRWXUGO
);
1667 /* get security xattr */
1668 status
= ocfs2_init_security_get(inode
, dir
, &si
);
1670 if (status
== -EOPNOTSUPP
)
1678 /* calculate meta data/clusters for setting security xattr */
1680 status
= ocfs2_calc_security_init(dir
, &si
, &want_clusters
,
1681 &xattr_credits
, &xattr_ac
);
1688 /* don't reserve bitmap space for fast symlinks. */
1689 if (l
> ocfs2_fast_symlink_chars(sb
))
1692 status
= ocfs2_reserve_clusters(osb
, want_clusters
, &data_ac
);
1694 if (status
!= -ENOSPC
)
1699 handle
= ocfs2_start_trans(osb
, credits
+ xattr_credits
);
1700 if (IS_ERR(handle
)) {
1701 status
= PTR_ERR(handle
);
1707 /* Starting to change things, restart is no longer possible. */
1708 ocfs2_block_signals(&oldset
);
1709 did_block_signals
= 1;
1711 status
= dquot_alloc_inode(inode
);
1714 did_quota_inode
= 1;
1716 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
,
1717 inode
->i_mode
, dentry
->d_name
.len
,
1718 dentry
->d_name
.name
);
1720 status
= ocfs2_mknod_locked(osb
, dir
, inode
,
1721 0, &new_fe_bh
, parent_fe_bh
, handle
,
1728 fe
= (struct ocfs2_dinode
*) new_fe_bh
->b_data
;
1731 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1734 inode
->i_op
= &ocfs2_symlink_inode_operations
;
1735 status
= dquot_alloc_space_nodirty(inode
,
1736 ocfs2_clusters_to_bytes(osb
->sb
, 1));
1740 status
= ocfs2_add_inode_data(osb
, inode
, &offset
, 1, 0,
1742 handle
, data_ac
, NULL
,
1745 if (status
!= -ENOSPC
&& status
!= -EINTR
) {
1747 "Failed to extend file to %llu\n",
1748 (unsigned long long)newsize
);
1754 i_size_write(inode
, newsize
);
1755 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1757 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
1758 memcpy((char *) fe
->id2
.i_symlink
, symname
, l
);
1759 i_size_write(inode
, newsize
);
1760 inode
->i_blocks
= 0;
1763 status
= ocfs2_mark_inode_dirty(handle
, inode
, new_fe_bh
);
1769 if (!ocfs2_inode_is_fast_symlink(inode
)) {
1770 status
= ocfs2_create_symlink_data(osb
, handle
, inode
,
1779 status
= ocfs2_init_security_set(handle
, inode
, new_fe_bh
, &si
,
1788 * Do this before adding the entry to the directory. We add
1789 * also set d_op after success so that ->d_iput() will cleanup
1790 * the dentry lock even if ocfs2_add_entry() fails below.
1792 status
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
1798 status
= ocfs2_add_entry(handle
, dentry
, inode
,
1799 le64_to_cpu(fe
->i_blkno
), parent_fe_bh
,
1806 insert_inode_hash(inode
);
1807 d_instantiate(dentry
, inode
);
1809 if (status
< 0 && did_quota
)
1810 dquot_free_space_nodirty(inode
,
1811 ocfs2_clusters_to_bytes(osb
->sb
, 1));
1812 if (status
< 0 && did_quota_inode
)
1813 dquot_free_inode(inode
);
1815 ocfs2_commit_trans(osb
, handle
);
1817 ocfs2_inode_unlock(dir
, 1);
1818 if (did_block_signals
)
1819 ocfs2_unblock_signals(&oldset
);
1822 brelse(parent_fe_bh
);
1825 ocfs2_free_dir_lookup_result(&lookup
);
1827 ocfs2_free_alloc_context(inode_ac
);
1829 ocfs2_free_alloc_context(data_ac
);
1831 ocfs2_free_alloc_context(xattr_ac
);
1832 if ((status
< 0) && inode
) {
1833 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SKIP_ORPHAN_DIR
;
1843 static int ocfs2_blkno_stringify(u64 blkno
, char *name
)
1845 int status
, namelen
;
1849 namelen
= snprintf(name
, OCFS2_ORPHAN_NAMELEN
+ 1, "%016llx",
1859 if (namelen
!= OCFS2_ORPHAN_NAMELEN
) {
1865 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name
,
1874 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super
*osb
,
1875 struct inode
**ret_orphan_dir
,
1876 struct buffer_head
**ret_orphan_dir_bh
)
1878 struct inode
*orphan_dir_inode
;
1879 struct buffer_head
*orphan_dir_bh
= NULL
;
1882 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
1883 ORPHAN_DIR_SYSTEM_INODE
,
1885 if (!orphan_dir_inode
) {
1891 mutex_lock(&orphan_dir_inode
->i_mutex
);
1893 ret
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
1895 mutex_unlock(&orphan_dir_inode
->i_mutex
);
1896 iput(orphan_dir_inode
);
1902 *ret_orphan_dir
= orphan_dir_inode
;
1903 *ret_orphan_dir_bh
= orphan_dir_bh
;
1908 static int __ocfs2_prepare_orphan_dir(struct inode
*orphan_dir_inode
,
1909 struct buffer_head
*orphan_dir_bh
,
1912 struct ocfs2_dir_lookup_result
*lookup
)
1915 struct ocfs2_super
*osb
= OCFS2_SB(orphan_dir_inode
->i_sb
);
1917 ret
= ocfs2_blkno_stringify(blkno
, name
);
1923 ret
= ocfs2_prepare_dir_for_insert(osb
, orphan_dir_inode
,
1924 orphan_dir_bh
, name
,
1925 OCFS2_ORPHAN_NAMELEN
, lookup
);
1935 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1936 * insertion of an orphan.
1937 * @osb: ocfs2 file system
1938 * @ret_orphan_dir: Orphan dir inode - returned locked!
1939 * @blkno: Actual block number of the inode to be inserted into orphan dir.
1940 * @lookup: dir lookup result, to be passed back into functions like
1943 * Returns zero on success and the ret_orphan_dir, name and lookup
1944 * fields will be populated.
1946 * Returns non-zero on failure.
1948 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
1949 struct inode
**ret_orphan_dir
,
1952 struct ocfs2_dir_lookup_result
*lookup
)
1954 struct inode
*orphan_dir_inode
= NULL
;
1955 struct buffer_head
*orphan_dir_bh
= NULL
;
1958 ret
= ocfs2_lookup_lock_orphan_dir(osb
, &orphan_dir_inode
,
1965 ret
= __ocfs2_prepare_orphan_dir(orphan_dir_inode
, orphan_dir_bh
,
1966 blkno
, name
, lookup
);
1972 *ret_orphan_dir
= orphan_dir_inode
;
1975 brelse(orphan_dir_bh
);
1978 ocfs2_inode_unlock(orphan_dir_inode
, 1);
1979 mutex_unlock(&orphan_dir_inode
->i_mutex
);
1980 iput(orphan_dir_inode
);
1987 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
1989 struct inode
*inode
,
1990 struct buffer_head
*fe_bh
,
1992 struct ocfs2_dir_lookup_result
*lookup
,
1993 struct inode
*orphan_dir_inode
)
1995 struct buffer_head
*orphan_dir_bh
= NULL
;
1997 struct ocfs2_dinode
*orphan_fe
;
1998 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
2000 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
2002 status
= ocfs2_read_inode_block(orphan_dir_inode
, &orphan_dir_bh
);
2008 status
= ocfs2_journal_access_di(handle
,
2009 INODE_CACHE(orphan_dir_inode
),
2011 OCFS2_JOURNAL_ACCESS_WRITE
);
2017 /* we're a cluster, and nlink can change on disk from
2018 * underneath us... */
2019 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2020 if (S_ISDIR(inode
->i_mode
))
2021 ocfs2_add_links_count(orphan_fe
, 1);
2022 orphan_dir_inode
->i_nlink
= ocfs2_read_links_count(orphan_fe
);
2023 ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2025 status
= __ocfs2_add_entry(handle
, orphan_dir_inode
, name
,
2026 OCFS2_ORPHAN_NAMELEN
, inode
,
2027 OCFS2_I(inode
)->ip_blkno
,
2028 orphan_dir_bh
, lookup
);
2035 * We're going to journal the change of i_flags and i_orphaned_slot.
2036 * It's safe anyway, though some callers may duplicate the journaling.
2037 * Journaling within the func just make the logic look more
2040 status
= ocfs2_journal_access_di(handle
,
2043 OCFS2_JOURNAL_ACCESS_WRITE
);
2049 le32_add_cpu(&fe
->i_flags
, OCFS2_ORPHANED_FL
);
2050 OCFS2_I(inode
)->ip_flags
&= ~OCFS2_INODE_SKIP_ORPHAN_DIR
;
2052 /* Record which orphan dir our inode now resides
2053 * in. delete_inode will use this to determine which orphan
2055 fe
->i_orphaned_slot
= cpu_to_le16(osb
->slot_num
);
2057 ocfs2_journal_dirty(handle
, fe_bh
);
2059 mlog(0, "Inode %llu orphaned in slot %d\n",
2060 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, osb
->slot_num
);
2063 brelse(orphan_dir_bh
);
2069 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2070 int ocfs2_orphan_del(struct ocfs2_super
*osb
,
2072 struct inode
*orphan_dir_inode
,
2073 struct inode
*inode
,
2074 struct buffer_head
*orphan_dir_bh
)
2076 char name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2077 struct ocfs2_dinode
*orphan_fe
;
2079 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
2083 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2089 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2090 name
, (unsigned long long)OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2091 OCFS2_ORPHAN_NAMELEN
);
2093 /* find it's spot in the orphan directory */
2094 status
= ocfs2_find_entry(name
, OCFS2_ORPHAN_NAMELEN
, orphan_dir_inode
,
2101 /* remove it from the orphan directory */
2102 status
= ocfs2_delete_entry(handle
, orphan_dir_inode
, &lookup
);
2108 status
= ocfs2_journal_access_di(handle
,
2109 INODE_CACHE(orphan_dir_inode
),
2111 OCFS2_JOURNAL_ACCESS_WRITE
);
2117 /* do the i_nlink dance! :) */
2118 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2119 if (S_ISDIR(inode
->i_mode
))
2120 ocfs2_add_links_count(orphan_fe
, -1);
2121 orphan_dir_inode
->i_nlink
= ocfs2_read_links_count(orphan_fe
);
2122 ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2125 ocfs2_free_dir_lookup_result(&lookup
);
2132 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to recieve a newly
2133 * allocated file. This is different from the typical 'add to orphan dir'
2134 * operation in that the inode does not yet exist. This is a problem because
2135 * the orphan dir stringifies the inode block number to come up with it's
2136 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2137 * problem. This function works around it by calling deeper into the orphan
2138 * and suballoc code than other callers. Use this only by necessity.
2139 * @dir: The directory which this inode will ultimately wind up under - not the
2141 * @dir_bh: buffer_head the @dir inode block
2142 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2143 * with the string to be used for orphan dirent. Pass back to the orphan dir
2145 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2147 * @ret_di_blkno: block number where the new inode will be allocated.
2148 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2149 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2151 * Returns zero on success and the ret_orphan_dir, name and lookup
2152 * fields will be populated.
2154 * Returns non-zero on failure.
2156 static int ocfs2_prep_new_orphaned_file(struct inode
*dir
,
2157 struct buffer_head
*dir_bh
,
2159 struct inode
**ret_orphan_dir
,
2161 struct ocfs2_dir_lookup_result
*orphan_insert
,
2162 struct ocfs2_alloc_context
**ret_inode_ac
)
2166 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2167 struct inode
*orphan_dir
= NULL
;
2168 struct buffer_head
*orphan_dir_bh
= NULL
;
2169 struct ocfs2_alloc_context
*inode_ac
= NULL
;
2171 ret
= ocfs2_lookup_lock_orphan_dir(osb
, &orphan_dir
, &orphan_dir_bh
);
2177 /* reserve an inode spot */
2178 ret
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
2185 ret
= ocfs2_find_new_inode_loc(dir
, dir_bh
, inode_ac
,
2192 ret
= __ocfs2_prepare_orphan_dir(orphan_dir
, orphan_dir_bh
,
2193 di_blkno
, orphan_name
, orphan_insert
);
2201 *ret_orphan_dir
= orphan_dir
;
2202 *ret_di_blkno
= di_blkno
;
2203 *ret_inode_ac
= inode_ac
;
2205 * orphan_name and orphan_insert are already up to
2206 * date via prepare_orphan_dir
2209 /* Unroll reserve_new_inode* */
2211 ocfs2_free_alloc_context(inode_ac
);
2213 /* Unroll orphan dir locking */
2214 mutex_unlock(&orphan_dir
->i_mutex
);
2215 ocfs2_inode_unlock(orphan_dir
, 1);
2219 brelse(orphan_dir_bh
);
2224 int ocfs2_create_inode_in_orphan(struct inode
*dir
,
2226 struct inode
**new_inode
)
2228 int status
, did_quota_inode
= 0;
2229 struct inode
*inode
= NULL
;
2230 struct inode
*orphan_dir
= NULL
;
2231 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2232 struct ocfs2_dinode
*di
= NULL
;
2233 handle_t
*handle
= NULL
;
2234 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2235 struct buffer_head
*parent_di_bh
= NULL
;
2236 struct buffer_head
*new_di_bh
= NULL
;
2237 struct ocfs2_alloc_context
*inode_ac
= NULL
;
2238 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
2239 u64
uninitialized_var(di_blkno
), suballoc_loc
;
2242 status
= ocfs2_inode_lock(dir
, &parent_di_bh
, 1);
2244 if (status
!= -ENOENT
)
2249 status
= ocfs2_prep_new_orphaned_file(dir
, parent_di_bh
,
2250 orphan_name
, &orphan_dir
,
2251 &di_blkno
, &orphan_insert
, &inode_ac
);
2253 if (status
!= -ENOSPC
)
2258 inode
= ocfs2_get_init_inode(dir
, mode
);
2265 handle
= ocfs2_start_trans(osb
, ocfs2_mknod_credits(osb
->sb
, 0, 0));
2266 if (IS_ERR(handle
)) {
2267 status
= PTR_ERR(handle
);
2273 status
= dquot_alloc_inode(inode
);
2276 did_quota_inode
= 1;
2278 status
= ocfs2_claim_new_inode_at_loc(handle
, dir
, inode_ac
,
2280 &suballoc_bit
, di_blkno
);
2287 /* do the real work now. */
2288 status
= __ocfs2_mknod_locked(dir
, inode
,
2289 0, &new_di_bh
, parent_di_bh
, handle
,
2290 inode_ac
, di_blkno
, suballoc_loc
,
2297 di
= (struct ocfs2_dinode
*)new_di_bh
->b_data
;
2298 status
= ocfs2_orphan_add(osb
, handle
, inode
, new_di_bh
, orphan_name
,
2299 &orphan_insert
, orphan_dir
);
2305 /* get open lock so that only nodes can't remove it from orphan dir. */
2306 status
= ocfs2_open_lock(inode
);
2310 insert_inode_hash(inode
);
2312 if (status
< 0 && did_quota_inode
)
2313 dquot_free_inode(inode
);
2315 ocfs2_commit_trans(osb
, handle
);
2318 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2319 ocfs2_inode_unlock(orphan_dir
, 1);
2320 mutex_unlock(&orphan_dir
->i_mutex
);
2324 if (status
== -ENOSPC
)
2325 mlog(0, "Disk is full\n");
2327 if ((status
< 0) && inode
) {
2333 ocfs2_free_alloc_context(inode_ac
);
2340 ocfs2_free_dir_lookup_result(&orphan_insert
);
2342 ocfs2_inode_unlock(dir
, 1);
2343 brelse(parent_di_bh
);
2347 int ocfs2_mv_orphaned_inode_to_new(struct inode
*dir
,
2348 struct inode
*inode
,
2349 struct dentry
*dentry
)
2352 struct buffer_head
*parent_di_bh
= NULL
;
2353 handle_t
*handle
= NULL
;
2354 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2355 struct ocfs2_dinode
*dir_di
, *di
;
2356 struct inode
*orphan_dir_inode
= NULL
;
2357 struct buffer_head
*orphan_dir_bh
= NULL
;
2358 struct buffer_head
*di_bh
= NULL
;
2359 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
2361 mlog_entry("(0x%p, 0x%p, %.*s')\n", dir
, dentry
,
2362 dentry
->d_name
.len
, dentry
->d_name
.name
);
2364 status
= ocfs2_inode_lock(dir
, &parent_di_bh
, 1);
2366 if (status
!= -ENOENT
)
2371 dir_di
= (struct ocfs2_dinode
*) parent_di_bh
->b_data
;
2372 if (!dir_di
->i_links_count
) {
2373 /* can't make a file in a deleted directory. */
2378 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
2379 dentry
->d_name
.len
);
2383 /* get a spot inside the dir. */
2384 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_di_bh
,
2385 dentry
->d_name
.name
,
2386 dentry
->d_name
.len
, &lookup
);
2392 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2393 ORPHAN_DIR_SYSTEM_INODE
,
2395 if (!orphan_dir_inode
) {
2401 mutex_lock(&orphan_dir_inode
->i_mutex
);
2403 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
2406 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2407 iput(orphan_dir_inode
);
2411 status
= ocfs2_read_inode_block(inode
, &di_bh
);
2417 handle
= ocfs2_start_trans(osb
, ocfs2_rename_credits(osb
->sb
));
2418 if (IS_ERR(handle
)) {
2419 status
= PTR_ERR(handle
);
2425 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
2426 di_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
2432 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
2439 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2440 le32_add_cpu(&di
->i_flags
, -OCFS2_ORPHANED_FL
);
2441 di
->i_orphaned_slot
= 0;
2443 ocfs2_set_links_count(di
, inode
->i_nlink
);
2444 ocfs2_journal_dirty(handle
, di_bh
);
2446 status
= ocfs2_add_entry(handle
, dentry
, inode
,
2447 OCFS2_I(inode
)->ip_blkno
, parent_di_bh
,
2454 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
2455 OCFS2_I(dir
)->ip_blkno
);
2461 d_instantiate(dentry
, inode
);
2464 ocfs2_commit_trans(osb
, handle
);
2466 ocfs2_inode_unlock(orphan_dir_inode
, 1);
2467 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2468 iput(orphan_dir_inode
);
2471 ocfs2_inode_unlock(dir
, 1);
2474 brelse(parent_di_bh
);
2475 brelse(orphan_dir_bh
);
2477 ocfs2_free_dir_lookup_result(&lookup
);
2484 const struct inode_operations ocfs2_dir_iops
= {
2485 .create
= ocfs2_create
,
2486 .lookup
= ocfs2_lookup
,
2488 .unlink
= ocfs2_unlink
,
2489 .rmdir
= ocfs2_unlink
,
2490 .symlink
= ocfs2_symlink
,
2491 .mkdir
= ocfs2_mkdir
,
2492 .mknod
= ocfs2_mknod
,
2493 .rename
= ocfs2_rename
,
2494 .setattr
= ocfs2_setattr
,
2495 .getattr
= ocfs2_getattr
,
2496 .permission
= ocfs2_permission
,
2497 .setxattr
= generic_setxattr
,
2498 .getxattr
= generic_getxattr
,
2499 .listxattr
= ocfs2_listxattr
,
2500 .removexattr
= generic_removexattr
,
2501 .fiemap
= ocfs2_fiemap
,