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 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_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
);
417 dentry
->d_op
= &ocfs2_dentry_ops
;
419 status
= ocfs2_add_entry(handle
, dentry
, inode
,
420 OCFS2_I(inode
)->ip_blkno
, parent_fe_bh
,
427 insert_inode_hash(inode
);
428 d_instantiate(dentry
, inode
);
431 if (status
< 0 && did_quota_inode
)
432 dquot_free_inode(inode
);
434 ocfs2_commit_trans(osb
, handle
);
436 ocfs2_inode_unlock(dir
, 1);
437 if (did_block_signals
)
438 ocfs2_unblock_signals(&oldset
);
440 if (status
== -ENOSPC
)
441 mlog(0, "Disk is full\n");
444 brelse(parent_fe_bh
);
448 ocfs2_free_dir_lookup_result(&lookup
);
451 ocfs2_free_alloc_context(inode_ac
);
454 ocfs2_free_alloc_context(data_ac
);
457 ocfs2_free_alloc_context(meta_ac
);
460 * We should call iput after the i_mutex of the bitmap been
461 * unlocked in ocfs2_free_alloc_context, or the
462 * ocfs2_delete_inode will mutex_lock again.
464 if ((status
< 0) && inode
) {
465 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SKIP_ORPHAN_DIR
;
475 static int __ocfs2_mknod_locked(struct inode
*dir
,
478 struct buffer_head
**new_fe_bh
,
479 struct buffer_head
*parent_fe_bh
,
481 struct ocfs2_alloc_context
*inode_ac
,
482 u64 fe_blkno
, u64 suballoc_loc
, u16 suballoc_bit
)
485 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
486 struct ocfs2_dinode
*fe
= NULL
;
487 struct ocfs2_extent_list
*fel
;
492 /* populate as many fields early on as possible - many of
493 * these are used by the support functions here and in
495 inode
->i_ino
= ino_from_blkno(osb
->sb
, fe_blkno
);
496 OCFS2_I(inode
)->ip_blkno
= fe_blkno
;
497 spin_lock(&osb
->osb_lock
);
498 inode
->i_generation
= osb
->s_next_generation
++;
499 spin_unlock(&osb
->osb_lock
);
501 *new_fe_bh
= sb_getblk(osb
->sb
, fe_blkno
);
507 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
), *new_fe_bh
);
509 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
511 OCFS2_JOURNAL_ACCESS_CREATE
);
517 fe
= (struct ocfs2_dinode
*) (*new_fe_bh
)->b_data
;
518 memset(fe
, 0, osb
->sb
->s_blocksize
);
520 fe
->i_generation
= cpu_to_le32(inode
->i_generation
);
521 fe
->i_fs_generation
= cpu_to_le32(osb
->fs_generation
);
522 fe
->i_blkno
= cpu_to_le64(fe_blkno
);
523 fe
->i_suballoc_loc
= cpu_to_le64(suballoc_loc
);
524 fe
->i_suballoc_bit
= cpu_to_le16(suballoc_bit
);
525 fe
->i_suballoc_slot
= cpu_to_le16(inode_ac
->ac_alloc_slot
);
526 fe
->i_uid
= cpu_to_le32(inode
->i_uid
);
527 fe
->i_gid
= cpu_to_le32(inode
->i_gid
);
528 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
529 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
))
530 fe
->id1
.dev1
.i_rdev
= cpu_to_le64(huge_encode_dev(dev
));
532 ocfs2_set_links_count(fe
, inode
->i_nlink
);
534 fe
->i_last_eb_blk
= 0;
535 strcpy(fe
->i_signature
, OCFS2_INODE_SIGNATURE
);
536 le32_add_cpu(&fe
->i_flags
, OCFS2_VALID_FL
);
537 fe
->i_atime
= fe
->i_ctime
= fe
->i_mtime
=
538 cpu_to_le64(CURRENT_TIME
.tv_sec
);
539 fe
->i_mtime_nsec
= fe
->i_ctime_nsec
= fe
->i_atime_nsec
=
540 cpu_to_le32(CURRENT_TIME
.tv_nsec
);
544 * If supported, directories start with inline data. If inline
545 * isn't supported, but indexing is, we start them as indexed.
547 feat
= le16_to_cpu(fe
->i_dyn_features
);
548 if (S_ISDIR(inode
->i_mode
) && ocfs2_supports_inline_data(osb
)) {
549 fe
->i_dyn_features
= cpu_to_le16(feat
| OCFS2_INLINE_DATA_FL
);
551 fe
->id2
.i_data
.id_count
= cpu_to_le16(
552 ocfs2_max_inline_data_with_xattr(osb
->sb
, fe
));
554 fel
= &fe
->id2
.i_list
;
555 fel
->l_tree_depth
= 0;
556 fel
->l_next_free_rec
= 0;
557 fel
->l_count
= cpu_to_le16(ocfs2_extent_recs_per_inode(osb
->sb
));
560 ocfs2_journal_dirty(handle
, *new_fe_bh
);
562 ocfs2_populate_inode(inode
, fe
, 1);
563 ocfs2_ci_set_new(osb
, INODE_CACHE(inode
));
564 if (!ocfs2_mount_local(osb
)) {
565 status
= ocfs2_create_new_inode_locks(inode
);
570 status
= 0; /* error in ocfs2_create_new_inode_locks is not
585 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
589 struct buffer_head
**new_fe_bh
,
590 struct buffer_head
*parent_fe_bh
,
592 struct ocfs2_alloc_context
*inode_ac
)
595 u64 suballoc_loc
, fe_blkno
= 0;
600 status
= ocfs2_claim_new_inode(handle
, dir
, parent_fe_bh
,
601 inode_ac
, &suballoc_loc
,
602 &suballoc_bit
, &fe_blkno
);
608 return __ocfs2_mknod_locked(dir
, inode
, dev
, new_fe_bh
,
609 parent_fe_bh
, handle
, inode_ac
,
610 fe_blkno
, suballoc_loc
, suballoc_bit
);
613 static int ocfs2_mkdir(struct inode
*dir
,
614 struct dentry
*dentry
,
619 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
620 dentry
->d_name
.len
, dentry
->d_name
.name
);
621 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
627 static int ocfs2_create(struct inode
*dir
,
628 struct dentry
*dentry
,
630 struct nameidata
*nd
)
634 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
635 dentry
->d_name
.len
, dentry
->d_name
.name
);
636 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
642 static int ocfs2_link(struct dentry
*old_dentry
,
644 struct dentry
*dentry
)
647 struct inode
*inode
= old_dentry
->d_inode
;
649 struct buffer_head
*fe_bh
= NULL
;
650 struct buffer_head
*parent_fe_bh
= NULL
;
651 struct ocfs2_dinode
*fe
= NULL
;
652 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
653 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
656 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode
->i_ino
,
657 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
658 dentry
->d_name
.len
, dentry
->d_name
.name
);
660 if (S_ISDIR(inode
->i_mode
))
663 dquot_initialize(dir
);
665 err
= ocfs2_inode_lock_nested(dir
, &parent_fe_bh
, 1, OI_LS_PARENT
);
677 err
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
682 err
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
684 dentry
->d_name
.len
, &lookup
);
690 err
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
697 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
698 if (ocfs2_read_links_count(fe
) >= ocfs2_link_max(osb
)) {
700 goto out_unlock_inode
;
703 handle
= ocfs2_start_trans(osb
, ocfs2_link_credits(osb
->sb
));
704 if (IS_ERR(handle
)) {
705 err
= PTR_ERR(handle
);
708 goto out_unlock_inode
;
711 /* Starting to change things, restart is no longer possible. */
712 ocfs2_block_signals(&oldset
);
714 err
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
715 OCFS2_JOURNAL_ACCESS_WRITE
);
722 inode
->i_ctime
= CURRENT_TIME
;
723 ocfs2_set_links_count(fe
, inode
->i_nlink
);
724 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
725 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
726 ocfs2_journal_dirty(handle
, fe_bh
);
728 err
= ocfs2_add_entry(handle
, dentry
, inode
,
729 OCFS2_I(inode
)->ip_blkno
,
730 parent_fe_bh
, &lookup
);
732 ocfs2_add_links_count(fe
, -1);
738 err
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
744 atomic_inc(&inode
->i_count
);
745 dentry
->d_op
= &ocfs2_dentry_ops
;
746 d_instantiate(dentry
, inode
);
749 ocfs2_commit_trans(osb
, handle
);
750 ocfs2_unblock_signals(&oldset
);
752 ocfs2_inode_unlock(inode
, 1);
755 ocfs2_inode_unlock(dir
, 1);
758 brelse(parent_fe_bh
);
760 ocfs2_free_dir_lookup_result(&lookup
);
768 * Takes and drops an exclusive lock on the given dentry. This will
769 * force other nodes to drop it.
771 static int ocfs2_remote_dentry_delete(struct dentry
*dentry
)
775 ret
= ocfs2_dentry_lock(dentry
, 1);
779 ocfs2_dentry_unlock(dentry
, 1);
784 static inline int inode_is_unlinkable(struct inode
*inode
)
786 if (S_ISDIR(inode
->i_mode
)) {
787 if (inode
->i_nlink
== 2)
792 if (inode
->i_nlink
== 1)
797 static int ocfs2_unlink(struct inode
*dir
,
798 struct dentry
*dentry
)
801 int child_locked
= 0;
802 struct inode
*inode
= dentry
->d_inode
;
803 struct inode
*orphan_dir
= NULL
;
804 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
806 struct ocfs2_dinode
*fe
= NULL
;
807 struct buffer_head
*fe_bh
= NULL
;
808 struct buffer_head
*parent_node_bh
= NULL
;
809 handle_t
*handle
= NULL
;
810 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
811 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
812 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
814 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
815 dentry
->d_name
.len
, dentry
->d_name
.name
);
817 dquot_initialize(dir
);
819 BUG_ON(dentry
->d_parent
->d_inode
!= dir
);
821 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
823 if (inode
== osb
->root_inode
) {
824 mlog(0, "Cannot delete the root directory\n");
828 status
= ocfs2_inode_lock_nested(dir
, &parent_node_bh
, 1,
831 if (status
!= -ENOENT
)
836 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
837 dentry
->d_name
.len
, &blkno
, dir
,
840 if (status
!= -ENOENT
)
845 if (OCFS2_I(inode
)->ip_blkno
!= blkno
) {
848 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
849 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
850 (unsigned long long)blkno
, OCFS2_I(inode
)->ip_flags
);
854 status
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
856 if (status
!= -ENOENT
)
862 if (S_ISDIR(inode
->i_mode
)) {
863 if (inode
->i_nlink
!= 2 || !ocfs2_empty_dir(inode
)) {
869 status
= ocfs2_remote_dentry_delete(dentry
);
871 /* This remote delete should succeed under all normal
877 if (inode_is_unlinkable(inode
)) {
878 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
879 OCFS2_I(inode
)->ip_blkno
,
880 orphan_name
, &orphan_insert
);
887 handle
= ocfs2_start_trans(osb
, ocfs2_unlink_credits(osb
->sb
));
888 if (IS_ERR(handle
)) {
889 status
= PTR_ERR(handle
);
895 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
896 OCFS2_JOURNAL_ACCESS_WRITE
);
902 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
904 if (inode_is_unlinkable(inode
)) {
905 status
= ocfs2_orphan_add(osb
, handle
, inode
, fe_bh
, orphan_name
,
906 &orphan_insert
, orphan_dir
);
913 /* delete the name from the parent dir */
914 status
= ocfs2_delete_entry(handle
, dir
, &lookup
);
920 if (S_ISDIR(inode
->i_mode
))
923 ocfs2_set_links_count(fe
, inode
->i_nlink
);
924 ocfs2_journal_dirty(handle
, fe_bh
);
926 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
927 if (S_ISDIR(inode
->i_mode
))
930 status
= ocfs2_mark_inode_dirty(handle
, dir
, parent_node_bh
);
933 if (S_ISDIR(inode
->i_mode
))
939 ocfs2_commit_trans(osb
, handle
);
942 ocfs2_inode_unlock(inode
, 1);
944 ocfs2_inode_unlock(dir
, 1);
947 /* This was locked for us in ocfs2_prepare_orphan_dir() */
948 ocfs2_inode_unlock(orphan_dir
, 1);
949 mutex_unlock(&orphan_dir
->i_mutex
);
954 brelse(parent_node_bh
);
956 ocfs2_free_dir_lookup_result(&orphan_insert
);
957 ocfs2_free_dir_lookup_result(&lookup
);
965 * The only place this should be used is rename!
966 * if they have the same id, then the 1st one is the only one locked.
968 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
969 struct buffer_head
**bh1
,
970 struct inode
*inode1
,
971 struct buffer_head
**bh2
,
972 struct inode
*inode2
)
975 struct ocfs2_inode_info
*oi1
= OCFS2_I(inode1
);
976 struct ocfs2_inode_info
*oi2
= OCFS2_I(inode2
);
977 struct buffer_head
**tmpbh
;
978 struct inode
*tmpinode
;
980 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
981 (unsigned long long)oi1
->ip_blkno
,
982 (unsigned long long)oi2
->ip_blkno
);
989 /* we always want to lock the one with the lower lockid first. */
990 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
991 if (oi1
->ip_blkno
< oi2
->ip_blkno
) {
992 /* switch id1 and id2 around */
993 mlog(0, "switching them around...\n");
1003 status
= ocfs2_inode_lock_nested(inode2
, bh2
, 1,
1006 if (status
!= -ENOENT
)
1013 status
= ocfs2_inode_lock_nested(inode1
, bh1
, 1, OI_LS_RENAME2
);
1016 * An error return must mean that no cluster locks
1017 * were held on function exit.
1019 if (oi1
->ip_blkno
!= oi2
->ip_blkno
)
1020 ocfs2_inode_unlock(inode2
, 1);
1022 if (status
!= -ENOENT
)
1031 static void ocfs2_double_unlock(struct inode
*inode1
, struct inode
*inode2
)
1033 ocfs2_inode_unlock(inode1
, 1);
1035 if (inode1
!= inode2
)
1036 ocfs2_inode_unlock(inode2
, 1);
1039 static int ocfs2_rename(struct inode
*old_dir
,
1040 struct dentry
*old_dentry
,
1041 struct inode
*new_dir
,
1042 struct dentry
*new_dentry
)
1044 int status
= 0, rename_lock
= 0, parents_locked
= 0, target_exists
= 0;
1045 int old_child_locked
= 0, new_child_locked
= 0, update_dot_dot
= 0;
1046 struct inode
*old_inode
= old_dentry
->d_inode
;
1047 struct inode
*new_inode
= new_dentry
->d_inode
;
1048 struct inode
*orphan_dir
= NULL
;
1049 struct ocfs2_dinode
*newfe
= NULL
;
1050 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1051 struct buffer_head
*newfe_bh
= NULL
;
1052 struct buffer_head
*old_inode_bh
= NULL
;
1053 struct ocfs2_super
*osb
= NULL
;
1054 u64 newfe_blkno
, old_de_ino
;
1055 handle_t
*handle
= NULL
;
1056 struct buffer_head
*old_dir_bh
= NULL
;
1057 struct buffer_head
*new_dir_bh
= NULL
;
1058 nlink_t old_dir_nlink
= old_dir
->i_nlink
;
1059 struct ocfs2_dinode
*old_di
;
1060 struct ocfs2_dir_lookup_result old_inode_dot_dot_res
= { NULL
, };
1061 struct ocfs2_dir_lookup_result target_lookup_res
= { NULL
, };
1062 struct ocfs2_dir_lookup_result old_entry_lookup
= { NULL
, };
1063 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
1064 struct ocfs2_dir_lookup_result target_insert
= { NULL
, };
1066 /* At some point it might be nice to break this function up a
1069 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1070 old_dir
, old_dentry
, new_dir
, new_dentry
,
1071 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
1072 new_dentry
->d_name
.len
, new_dentry
->d_name
.name
);
1074 dquot_initialize(old_dir
);
1075 dquot_initialize(new_dir
);
1077 osb
= OCFS2_SB(old_dir
->i_sb
);
1080 if (!igrab(new_inode
))
1084 /* Assume a directory hierarchy thusly:
1087 * a,b,c, and d are all directories.
1089 * from cwd of 'a' on both nodes:
1093 * And that's why, just like the VFS, we need a file system
1095 if (old_dir
!= new_dir
&& S_ISDIR(old_inode
->i_mode
)) {
1096 status
= ocfs2_rename_lock(osb
);
1104 /* if old and new are the same, this'll just do one lock. */
1105 status
= ocfs2_double_lock(osb
, &old_dir_bh
, old_dir
,
1106 &new_dir_bh
, new_dir
);
1113 /* make sure both dirs have bhs
1114 * get an extra ref on old_dir_bh if old==new */
1117 new_dir_bh
= old_dir_bh
;
1120 mlog(ML_ERROR
, "no old_dir_bh!\n");
1127 * Aside from allowing a meta data update, the locking here
1128 * also ensures that the downconvert thread on other nodes
1129 * won't have to concurrently downconvert the inode and the
1132 status
= ocfs2_inode_lock_nested(old_inode
, &old_inode_bh
, 1,
1135 if (status
!= -ENOENT
)
1139 old_child_locked
= 1;
1141 status
= ocfs2_remote_dentry_delete(old_dentry
);
1147 if (S_ISDIR(old_inode
->i_mode
)) {
1148 u64 old_inode_parent
;
1151 status
= ocfs2_find_files_on_disk("..", 2, &old_inode_parent
,
1153 &old_inode_dot_dot_res
);
1159 if (old_inode_parent
!= OCFS2_I(old_dir
)->ip_blkno
) {
1164 if (!new_inode
&& new_dir
!= old_dir
&&
1165 new_dir
->i_nlink
>= ocfs2_link_max(osb
)) {
1171 status
= ocfs2_lookup_ino_from_name(old_dir
, old_dentry
->d_name
.name
,
1172 old_dentry
->d_name
.len
,
1180 * Check for inode number is _not_ due to possible IO errors.
1181 * We might rmdir the source, keep it as pwd of some process
1182 * and merrily kill the link to whatever was created under the
1183 * same name. Goodbye sticky bit ;-<
1185 if (old_de_ino
!= OCFS2_I(old_inode
)->ip_blkno
) {
1190 /* check if the target already exists (in which case we need
1192 status
= ocfs2_find_files_on_disk(new_dentry
->d_name
.name
,
1193 new_dentry
->d_name
.len
,
1194 &newfe_blkno
, new_dir
,
1195 &target_lookup_res
);
1196 /* The only error we allow here is -ENOENT because the new
1197 * file not existing is perfectly valid. */
1198 if ((status
< 0) && (status
!= -ENOENT
)) {
1199 /* If we cannot find the file specified we should just */
1200 /* return the error... */
1207 if (!target_exists
&& new_inode
) {
1209 * Target was unlinked by another node while we were
1210 * waiting to get to ocfs2_rename(). There isn't
1211 * anything we can do here to help the situation, so
1212 * bubble up the appropriate error.
1218 /* In case we need to overwrite an existing file, we blow it
1220 if (target_exists
) {
1221 /* VFS didn't think there existed an inode here, but
1222 * someone else in the cluster must have raced our
1223 * rename to create one. Today we error cleanly, in
1224 * the future we should consider calling iget to build
1225 * a new struct inode for this entry. */
1229 mlog(0, "We found an inode for name %.*s but VFS "
1230 "didn't give us one.\n", new_dentry
->d_name
.len
,
1231 new_dentry
->d_name
.name
);
1235 if (OCFS2_I(new_inode
)->ip_blkno
!= newfe_blkno
) {
1238 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1239 (unsigned long long)OCFS2_I(new_inode
)->ip_blkno
,
1240 (unsigned long long)newfe_blkno
,
1241 OCFS2_I(new_inode
)->ip_flags
);
1245 status
= ocfs2_inode_lock(new_inode
, &newfe_bh
, 1);
1247 if (status
!= -ENOENT
)
1251 new_child_locked
= 1;
1253 status
= ocfs2_remote_dentry_delete(new_dentry
);
1259 newfe
= (struct ocfs2_dinode
*) newfe_bh
->b_data
;
1261 mlog(0, "aha rename over existing... new_blkno=%llu "
1262 "newfebh=%p bhblocknr=%llu\n",
1263 (unsigned long long)newfe_blkno
, newfe_bh
, newfe_bh
?
1264 (unsigned long long)newfe_bh
->b_blocknr
: 0ULL);
1266 if (S_ISDIR(new_inode
->i_mode
) || (new_inode
->i_nlink
== 1)) {
1267 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
1268 OCFS2_I(new_inode
)->ip_blkno
,
1269 orphan_name
, &orphan_insert
);
1276 BUG_ON(new_dentry
->d_parent
->d_inode
!= new_dir
);
1278 status
= ocfs2_check_dir_for_entry(new_dir
,
1279 new_dentry
->d_name
.name
,
1280 new_dentry
->d_name
.len
);
1284 status
= ocfs2_prepare_dir_for_insert(osb
, new_dir
, new_dir_bh
,
1285 new_dentry
->d_name
.name
,
1286 new_dentry
->d_name
.len
,
1294 handle
= ocfs2_start_trans(osb
, ocfs2_rename_credits(osb
->sb
));
1295 if (IS_ERR(handle
)) {
1296 status
= PTR_ERR(handle
);
1302 if (target_exists
) {
1303 if (S_ISDIR(new_inode
->i_mode
)) {
1304 if (new_inode
->i_nlink
!= 2 ||
1305 !ocfs2_empty_dir(new_inode
)) {
1306 status
= -ENOTEMPTY
;
1310 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(new_inode
),
1312 OCFS2_JOURNAL_ACCESS_WRITE
);
1318 if (S_ISDIR(new_inode
->i_mode
) ||
1319 (ocfs2_read_links_count(newfe
) == 1)) {
1320 status
= ocfs2_orphan_add(osb
, handle
, new_inode
,
1321 newfe_bh
, orphan_name
,
1322 &orphan_insert
, orphan_dir
);
1329 /* change the dirent to point to the correct inode */
1330 status
= ocfs2_update_entry(new_dir
, handle
, &target_lookup_res
,
1336 new_dir
->i_version
++;
1338 if (S_ISDIR(new_inode
->i_mode
))
1339 ocfs2_set_links_count(newfe
, 0);
1341 ocfs2_add_links_count(newfe
, -1);
1342 ocfs2_journal_dirty(handle
, newfe_bh
);
1344 /* if the name was not found in new_dir, add it now */
1345 status
= ocfs2_add_entry(handle
, new_dentry
, old_inode
,
1346 OCFS2_I(old_inode
)->ip_blkno
,
1347 new_dir_bh
, &target_insert
);
1350 old_inode
->i_ctime
= CURRENT_TIME
;
1351 mark_inode_dirty(old_inode
);
1353 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(old_inode
),
1355 OCFS2_JOURNAL_ACCESS_WRITE
);
1357 old_di
= (struct ocfs2_dinode
*) old_inode_bh
->b_data
;
1359 old_di
->i_ctime
= cpu_to_le64(old_inode
->i_ctime
.tv_sec
);
1360 old_di
->i_ctime_nsec
= cpu_to_le32(old_inode
->i_ctime
.tv_nsec
);
1361 ocfs2_journal_dirty(handle
, old_inode_bh
);
1366 * Now that the name has been added to new_dir, remove the old name.
1368 * We don't keep any directory entry context around until now
1369 * because the insert might have changed the type of directory
1370 * we're dealing with.
1372 status
= ocfs2_find_entry(old_dentry
->d_name
.name
,
1373 old_dentry
->d_name
.len
, old_dir
,
1378 status
= ocfs2_delete_entry(handle
, old_dir
, &old_entry_lookup
);
1385 new_inode
->i_nlink
--;
1386 new_inode
->i_ctime
= CURRENT_TIME
;
1388 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1390 if (update_dot_dot
) {
1391 status
= ocfs2_update_entry(old_inode
, handle
,
1392 &old_inode_dot_dot_res
, new_dir
);
1395 new_inode
->i_nlink
--;
1398 mark_inode_dirty(new_dir
);
1401 mark_inode_dirty(old_dir
);
1402 ocfs2_mark_inode_dirty(handle
, old_dir
, old_dir_bh
);
1404 mark_inode_dirty(new_inode
);
1405 ocfs2_mark_inode_dirty(handle
, new_inode
, newfe_bh
);
1408 if (old_dir
!= new_dir
) {
1409 /* Keep the same times on both directories.*/
1410 new_dir
->i_ctime
= new_dir
->i_mtime
= old_dir
->i_ctime
;
1413 * This will also pick up the i_nlink change from the
1416 ocfs2_mark_inode_dirty(handle
, new_dir
, new_dir_bh
);
1419 if (old_dir_nlink
!= old_dir
->i_nlink
) {
1421 mlog(ML_ERROR
, "need to change nlink for old dir "
1422 "%llu from %d to %d but bh is NULL!\n",
1423 (unsigned long long)OCFS2_I(old_dir
)->ip_blkno
,
1424 (int)old_dir_nlink
, old_dir
->i_nlink
);
1426 struct ocfs2_dinode
*fe
;
1427 status
= ocfs2_journal_access_di(handle
,
1428 INODE_CACHE(old_dir
),
1430 OCFS2_JOURNAL_ACCESS_WRITE
);
1431 fe
= (struct ocfs2_dinode
*) old_dir_bh
->b_data
;
1432 ocfs2_set_links_count(fe
, old_dir
->i_nlink
);
1433 ocfs2_journal_dirty(handle
, old_dir_bh
);
1436 ocfs2_dentry_move(old_dentry
, new_dentry
, old_dir
, new_dir
);
1440 ocfs2_rename_unlock(osb
);
1443 ocfs2_commit_trans(osb
, handle
);
1446 ocfs2_double_unlock(old_dir
, new_dir
);
1448 if (old_child_locked
)
1449 ocfs2_inode_unlock(old_inode
, 1);
1451 if (new_child_locked
)
1452 ocfs2_inode_unlock(new_inode
, 1);
1455 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1456 ocfs2_inode_unlock(orphan_dir
, 1);
1457 mutex_unlock(&orphan_dir
->i_mutex
);
1462 sync_mapping_buffers(old_inode
->i_mapping
);
1467 ocfs2_free_dir_lookup_result(&target_lookup_res
);
1468 ocfs2_free_dir_lookup_result(&old_entry_lookup
);
1469 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res
);
1470 ocfs2_free_dir_lookup_result(&orphan_insert
);
1471 ocfs2_free_dir_lookup_result(&target_insert
);
1474 brelse(old_inode_bh
);
1484 * we expect i_size = strlen(symname). Copy symname into the file
1485 * data, including the null terminator.
1487 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
1489 struct inode
*inode
,
1490 const char *symname
)
1492 struct buffer_head
**bhs
= NULL
;
1494 struct super_block
*sb
= osb
->sb
;
1495 u64 p_blkno
, p_blocks
;
1496 int virtual, blocks
, status
, i
, bytes_left
;
1498 bytes_left
= i_size_read(inode
) + 1;
1499 /* we can't trust i_blocks because we're actually going to
1500 * write i_size + 1 bytes. */
1501 blocks
= (bytes_left
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
1503 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1504 (unsigned long long)inode
->i_blocks
,
1505 i_size_read(inode
), blocks
);
1507 /* Sanity check -- make sure we're going to fit. */
1509 ocfs2_clusters_to_bytes(sb
, OCFS2_I(inode
)->ip_clusters
)) {
1515 bhs
= kcalloc(blocks
, sizeof(struct buffer_head
*), GFP_KERNEL
);
1522 status
= ocfs2_extent_map_get_blocks(inode
, 0, &p_blkno
, &p_blocks
,
1529 /* links can never be larger than one cluster so we know this
1530 * is all going to be contiguous, but do a sanity check
1532 if ((p_blocks
<< sb
->s_blocksize_bits
) < bytes_left
) {
1539 while(bytes_left
> 0) {
1540 c
= &symname
[virtual * sb
->s_blocksize
];
1542 bhs
[virtual] = sb_getblk(sb
, p_blkno
);
1543 if (!bhs
[virtual]) {
1548 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
),
1551 status
= ocfs2_journal_access(handle
, INODE_CACHE(inode
),
1553 OCFS2_JOURNAL_ACCESS_CREATE
);
1559 memset(bhs
[virtual]->b_data
, 0, sb
->s_blocksize
);
1561 memcpy(bhs
[virtual]->b_data
, c
,
1562 (bytes_left
> sb
->s_blocksize
) ? sb
->s_blocksize
:
1565 ocfs2_journal_dirty(handle
, bhs
[virtual]);
1569 bytes_left
-= sb
->s_blocksize
;
1576 for(i
= 0; i
< blocks
; i
++)
1585 static int ocfs2_symlink(struct inode
*dir
,
1586 struct dentry
*dentry
,
1587 const char *symname
)
1589 int status
, l
, credits
;
1591 struct ocfs2_super
*osb
= NULL
;
1592 struct inode
*inode
= NULL
;
1593 struct super_block
*sb
;
1594 struct buffer_head
*new_fe_bh
= NULL
;
1595 struct buffer_head
*parent_fe_bh
= NULL
;
1596 struct ocfs2_dinode
*fe
= NULL
;
1597 struct ocfs2_dinode
*dirfe
;
1598 handle_t
*handle
= NULL
;
1599 struct ocfs2_alloc_context
*inode_ac
= NULL
;
1600 struct ocfs2_alloc_context
*data_ac
= NULL
;
1601 struct ocfs2_alloc_context
*xattr_ac
= NULL
;
1602 int want_clusters
= 0;
1603 int xattr_credits
= 0;
1604 struct ocfs2_security_xattr_info si
= {
1607 int did_quota
= 0, did_quota_inode
= 0;
1608 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
1610 int did_block_signals
= 0;
1612 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir
,
1613 dentry
, symname
, dentry
->d_name
.len
, dentry
->d_name
.name
);
1615 dquot_initialize(dir
);
1620 l
= strlen(symname
) + 1;
1622 credits
= ocfs2_calc_symlink_credits(sb
);
1624 /* lock the parent directory */
1625 status
= ocfs2_inode_lock(dir
, &parent_fe_bh
, 1);
1627 if (status
!= -ENOENT
)
1632 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
1633 if (!ocfs2_read_links_count(dirfe
)) {
1634 /* can't make a file in a deleted directory. */
1639 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
1640 dentry
->d_name
.len
);
1644 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
1645 dentry
->d_name
.name
,
1646 dentry
->d_name
.len
, &lookup
);
1652 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
1654 if (status
!= -ENOSPC
)
1659 inode
= ocfs2_get_init_inode(dir
, S_IFLNK
| S_IRWXUGO
);
1666 /* get security xattr */
1667 status
= ocfs2_init_security_get(inode
, dir
, &si
);
1669 if (status
== -EOPNOTSUPP
)
1677 /* calculate meta data/clusters for setting security xattr */
1679 status
= ocfs2_calc_security_init(dir
, &si
, &want_clusters
,
1680 &xattr_credits
, &xattr_ac
);
1687 /* don't reserve bitmap space for fast symlinks. */
1688 if (l
> ocfs2_fast_symlink_chars(sb
))
1691 status
= ocfs2_reserve_clusters(osb
, want_clusters
, &data_ac
);
1693 if (status
!= -ENOSPC
)
1698 handle
= ocfs2_start_trans(osb
, credits
+ xattr_credits
);
1699 if (IS_ERR(handle
)) {
1700 status
= PTR_ERR(handle
);
1706 /* Starting to change things, restart is no longer possible. */
1707 ocfs2_block_signals(&oldset
);
1708 did_block_signals
= 1;
1710 status
= dquot_alloc_inode(inode
);
1713 did_quota_inode
= 1;
1715 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
,
1716 inode
->i_mode
, dentry
->d_name
.len
,
1717 dentry
->d_name
.name
);
1719 status
= ocfs2_mknod_locked(osb
, dir
, inode
,
1720 0, &new_fe_bh
, parent_fe_bh
, handle
,
1727 fe
= (struct ocfs2_dinode
*) new_fe_bh
->b_data
;
1730 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1733 inode
->i_op
= &ocfs2_symlink_inode_operations
;
1734 status
= dquot_alloc_space_nodirty(inode
,
1735 ocfs2_clusters_to_bytes(osb
->sb
, 1));
1739 status
= ocfs2_add_inode_data(osb
, inode
, &offset
, 1, 0,
1741 handle
, data_ac
, NULL
,
1744 if (status
!= -ENOSPC
&& status
!= -EINTR
) {
1746 "Failed to extend file to %llu\n",
1747 (unsigned long long)newsize
);
1753 i_size_write(inode
, newsize
);
1754 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1756 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
1757 memcpy((char *) fe
->id2
.i_symlink
, symname
, l
);
1758 i_size_write(inode
, newsize
);
1759 inode
->i_blocks
= 0;
1762 status
= ocfs2_mark_inode_dirty(handle
, inode
, new_fe_bh
);
1768 if (!ocfs2_inode_is_fast_symlink(inode
)) {
1769 status
= ocfs2_create_symlink_data(osb
, handle
, inode
,
1778 status
= ocfs2_init_security_set(handle
, inode
, new_fe_bh
, &si
,
1787 * Do this before adding the entry to the directory. We add
1788 * also set d_op after success so that ->d_iput() will cleanup
1789 * the dentry lock even if ocfs2_add_entry() fails below.
1791 status
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
1796 dentry
->d_op
= &ocfs2_dentry_ops
;
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 dentry
->d_op
= &ocfs2_dentry_ops
;
2462 d_instantiate(dentry
, inode
);
2465 ocfs2_commit_trans(osb
, handle
);
2467 ocfs2_inode_unlock(orphan_dir_inode
, 1);
2468 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2469 iput(orphan_dir_inode
);
2472 ocfs2_inode_unlock(dir
, 1);
2475 brelse(parent_di_bh
);
2476 brelse(orphan_dir_bh
);
2478 ocfs2_free_dir_lookup_result(&lookup
);
2485 const struct inode_operations ocfs2_dir_iops
= {
2486 .create
= ocfs2_create
,
2487 .lookup
= ocfs2_lookup
,
2489 .unlink
= ocfs2_unlink
,
2490 .rmdir
= ocfs2_unlink
,
2491 .symlink
= ocfs2_symlink
,
2492 .mkdir
= ocfs2_mkdir
,
2493 .mknod
= ocfs2_mknod
,
2494 .rename
= ocfs2_rename
,
2495 .setattr
= ocfs2_setattr
,
2496 .getattr
= ocfs2_getattr
,
2497 .permission
= ocfs2_permission
,
2498 .setxattr
= generic_setxattr
,
2499 .getxattr
= generic_getxattr
,
2500 .listxattr
= ocfs2_listxattr
,
2501 .removexattr
= generic_removexattr
,
2502 .fiemap
= ocfs2_fiemap
,