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 #include <cluster/masklog.h>
53 #include "extent_map.h"
65 #include "ocfs2_trace.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
,
105 struct inode
*inode
= NULL
;
107 struct ocfs2_inode_info
*oi
;
109 trace_ocfs2_lookup(dir
, dentry
, dentry
->d_name
.len
,
111 (unsigned long long)OCFS2_I(dir
)->ip_blkno
, 0);
113 if (dentry
->d_name
.len
> OCFS2_MAX_FILENAME_LEN
) {
114 ret
= ERR_PTR(-ENAMETOOLONG
);
118 status
= ocfs2_inode_lock_nested(dir
, NULL
, 0, OI_LS_PARENT
);
120 if (status
!= -ENOENT
)
122 ret
= ERR_PTR(status
);
126 status
= ocfs2_lookup_ino_from_name(dir
, dentry
->d_name
.name
,
127 dentry
->d_name
.len
, &blkno
);
131 inode
= ocfs2_iget(OCFS2_SB(dir
->i_sb
), blkno
, 0, 0);
133 ret
= ERR_PTR(-EACCES
);
138 /* Clear any orphaned state... If we were able to look up the
139 * inode from a directory, it certainly can't be orphaned. We
140 * might have the bad state from a node which intended to
141 * orphan this inode but crashed before it could commit the
143 spin_lock(&oi
->ip_lock
);
144 oi
->ip_flags
&= ~OCFS2_INODE_MAYBE_ORPHANED
;
145 spin_unlock(&oi
->ip_lock
);
148 ret
= d_splice_alias(inode
, dentry
);
152 * If d_splice_alias() finds a DCACHE_DISCONNECTED
153 * dentry, it will d_move() it on top of ourse. The
154 * return value will indicate this however, so in
155 * those cases, we switch them around for the locking
158 * NOTE: This dentry already has ->d_op set from
159 * ocfs2_get_parent() and ocfs2_get_dentry()
164 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
165 OCFS2_I(dir
)->ip_blkno
);
168 ret
= ERR_PTR(status
);
172 ocfs2_dentry_attach_gen(dentry
);
175 /* Don't drop the cluster lock until *after* the d_add --
176 * unlink on another node will message us to remove that
177 * dentry under this lock so otherwise we can race this with
178 * the downconvert thread and have a stale dentry. */
179 ocfs2_inode_unlock(dir
, 0);
183 trace_ocfs2_lookup_ret(ret
);
188 static struct inode
*ocfs2_get_init_inode(struct inode
*dir
, umode_t mode
)
192 inode
= new_inode(dir
->i_sb
);
194 mlog(ML_ERROR
, "new_inode failed!\n");
198 /* populate as many fields early on as possible - many of
199 * these are used by the support functions here and in
203 inode_init_owner(inode
, dir
, mode
);
204 dquot_initialize(inode
);
208 static int ocfs2_mknod(struct inode
*dir
,
209 struct dentry
*dentry
,
214 struct buffer_head
*parent_fe_bh
= NULL
;
215 handle_t
*handle
= NULL
;
216 struct ocfs2_super
*osb
;
217 struct ocfs2_dinode
*dirfe
;
218 struct buffer_head
*new_fe_bh
= NULL
;
219 struct inode
*inode
= NULL
;
220 struct ocfs2_alloc_context
*inode_ac
= NULL
;
221 struct ocfs2_alloc_context
*data_ac
= NULL
;
222 struct ocfs2_alloc_context
*meta_ac
= NULL
;
223 int want_clusters
= 0;
225 int xattr_credits
= 0;
226 struct ocfs2_security_xattr_info si
= {
229 int did_quota_inode
= 0;
230 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
232 int did_block_signals
= 0;
234 trace_ocfs2_mknod(dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
,
235 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
236 (unsigned long)dev
, mode
);
238 dquot_initialize(dir
);
240 /* get our super block */
241 osb
= OCFS2_SB(dir
->i_sb
);
243 status
= ocfs2_inode_lock(dir
, &parent_fe_bh
, 1);
245 if (status
!= -ENOENT
)
250 if (S_ISDIR(mode
) && (dir
->i_nlink
>= ocfs2_link_max(osb
))) {
255 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
256 if (!ocfs2_read_links_count(dirfe
)) {
257 /* can't make a file in a deleted directory. */
262 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
267 /* get a spot inside the dir. */
268 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
270 dentry
->d_name
.len
, &lookup
);
276 /* reserve an inode spot */
277 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
279 if (status
!= -ENOSPC
)
284 inode
= ocfs2_get_init_inode(dir
, mode
);
291 /* get security xattr */
292 status
= ocfs2_init_security_get(inode
, dir
, &dentry
->d_name
, &si
);
294 if (status
== -EOPNOTSUPP
)
302 /* calculate meta data/clusters for setting security and acl xattr */
303 status
= ocfs2_calc_xattr_init(dir
, parent_fe_bh
, mode
,
305 &xattr_credits
, &want_meta
);
311 /* Reserve a cluster if creating an extent based directory. */
312 if (S_ISDIR(mode
) && !ocfs2_supports_inline_data(osb
)) {
315 /* Dir indexing requires extra space as well */
316 if (ocfs2_supports_indexed_dirs(osb
))
320 status
= ocfs2_reserve_new_metadata_blocks(osb
, want_meta
, &meta_ac
);
322 if (status
!= -ENOSPC
)
327 status
= ocfs2_reserve_clusters(osb
, want_clusters
, &data_ac
);
329 if (status
!= -ENOSPC
)
334 handle
= ocfs2_start_trans(osb
, ocfs2_mknod_credits(osb
->sb
,
337 if (IS_ERR(handle
)) {
338 status
= PTR_ERR(handle
);
344 /* Starting to change things, restart is no longer possible. */
345 ocfs2_block_signals(&oldset
);
346 did_block_signals
= 1;
348 status
= dquot_alloc_inode(inode
);
353 /* do the real work now. */
354 status
= ocfs2_mknod_locked(osb
, dir
, inode
, dev
,
355 &new_fe_bh
, parent_fe_bh
, handle
,
363 status
= ocfs2_fill_new_dir(osb
, handle
, dir
, inode
,
364 new_fe_bh
, data_ac
, meta_ac
);
370 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(dir
),
372 OCFS2_JOURNAL_ACCESS_WRITE
);
377 ocfs2_add_links_count(dirfe
, 1);
378 ocfs2_journal_dirty(handle
, parent_fe_bh
);
382 status
= ocfs2_init_acl(handle
, inode
, dir
, new_fe_bh
, parent_fe_bh
,
390 status
= ocfs2_init_security_set(handle
, inode
, new_fe_bh
, &si
,
399 * Do this before adding the entry to the directory. We add
400 * also set d_op after success so that ->d_iput() will cleanup
401 * the dentry lock even if ocfs2_add_entry() fails below.
403 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
404 OCFS2_I(dir
)->ip_blkno
);
410 status
= ocfs2_add_entry(handle
, dentry
, inode
,
411 OCFS2_I(inode
)->ip_blkno
, parent_fe_bh
,
418 insert_inode_hash(inode
);
419 d_instantiate(dentry
, inode
);
422 if (status
< 0 && did_quota_inode
)
423 dquot_free_inode(inode
);
425 ocfs2_commit_trans(osb
, handle
);
427 ocfs2_inode_unlock(dir
, 1);
428 if (did_block_signals
)
429 ocfs2_unblock_signals(&oldset
);
432 brelse(parent_fe_bh
);
436 ocfs2_free_dir_lookup_result(&lookup
);
439 ocfs2_free_alloc_context(inode_ac
);
442 ocfs2_free_alloc_context(data_ac
);
445 ocfs2_free_alloc_context(meta_ac
);
448 * We should call iput after the i_mutex of the bitmap been
449 * unlocked in ocfs2_free_alloc_context, or the
450 * ocfs2_delete_inode will mutex_lock again.
452 if ((status
< 0) && inode
) {
453 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SKIP_ORPHAN_DIR
;
464 static int __ocfs2_mknod_locked(struct inode
*dir
,
467 struct buffer_head
**new_fe_bh
,
468 struct buffer_head
*parent_fe_bh
,
470 struct ocfs2_alloc_context
*inode_ac
,
471 u64 fe_blkno
, u64 suballoc_loc
, u16 suballoc_bit
)
474 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
475 struct ocfs2_dinode
*fe
= NULL
;
476 struct ocfs2_extent_list
*fel
;
481 /* populate as many fields early on as possible - many of
482 * these are used by the support functions here and in
484 inode
->i_ino
= ino_from_blkno(osb
->sb
, fe_blkno
);
485 OCFS2_I(inode
)->ip_blkno
= fe_blkno
;
486 spin_lock(&osb
->osb_lock
);
487 inode
->i_generation
= osb
->s_next_generation
++;
488 spin_unlock(&osb
->osb_lock
);
490 *new_fe_bh
= sb_getblk(osb
->sb
, fe_blkno
);
496 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
), *new_fe_bh
);
498 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
500 OCFS2_JOURNAL_ACCESS_CREATE
);
506 fe
= (struct ocfs2_dinode
*) (*new_fe_bh
)->b_data
;
507 memset(fe
, 0, osb
->sb
->s_blocksize
);
509 fe
->i_generation
= cpu_to_le32(inode
->i_generation
);
510 fe
->i_fs_generation
= cpu_to_le32(osb
->fs_generation
);
511 fe
->i_blkno
= cpu_to_le64(fe_blkno
);
512 fe
->i_suballoc_loc
= cpu_to_le64(suballoc_loc
);
513 fe
->i_suballoc_bit
= cpu_to_le16(suballoc_bit
);
514 fe
->i_suballoc_slot
= cpu_to_le16(inode_ac
->ac_alloc_slot
);
515 fe
->i_uid
= cpu_to_le32(i_uid_read(inode
));
516 fe
->i_gid
= cpu_to_le32(i_gid_read(inode
));
517 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
518 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
))
519 fe
->id1
.dev1
.i_rdev
= cpu_to_le64(huge_encode_dev(dev
));
521 ocfs2_set_links_count(fe
, inode
->i_nlink
);
523 fe
->i_last_eb_blk
= 0;
524 strcpy(fe
->i_signature
, OCFS2_INODE_SIGNATURE
);
525 fe
->i_flags
|= cpu_to_le32(OCFS2_VALID_FL
);
526 fe
->i_atime
= fe
->i_ctime
= fe
->i_mtime
=
527 cpu_to_le64(CURRENT_TIME
.tv_sec
);
528 fe
->i_mtime_nsec
= fe
->i_ctime_nsec
= fe
->i_atime_nsec
=
529 cpu_to_le32(CURRENT_TIME
.tv_nsec
);
533 * If supported, directories start with inline data. If inline
534 * isn't supported, but indexing is, we start them as indexed.
536 feat
= le16_to_cpu(fe
->i_dyn_features
);
537 if (S_ISDIR(inode
->i_mode
) && ocfs2_supports_inline_data(osb
)) {
538 fe
->i_dyn_features
= cpu_to_le16(feat
| OCFS2_INLINE_DATA_FL
);
540 fe
->id2
.i_data
.id_count
= cpu_to_le16(
541 ocfs2_max_inline_data_with_xattr(osb
->sb
, fe
));
543 fel
= &fe
->id2
.i_list
;
544 fel
->l_tree_depth
= 0;
545 fel
->l_next_free_rec
= 0;
546 fel
->l_count
= cpu_to_le16(ocfs2_extent_recs_per_inode(osb
->sb
));
549 ocfs2_journal_dirty(handle
, *new_fe_bh
);
551 ocfs2_populate_inode(inode
, fe
, 1);
552 ocfs2_ci_set_new(osb
, INODE_CACHE(inode
));
553 if (!ocfs2_mount_local(osb
)) {
554 status
= ocfs2_create_new_inode_locks(inode
);
559 status
= 0; /* error in ocfs2_create_new_inode_locks is not
575 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
579 struct buffer_head
**new_fe_bh
,
580 struct buffer_head
*parent_fe_bh
,
582 struct ocfs2_alloc_context
*inode_ac
)
585 u64 suballoc_loc
, fe_blkno
= 0;
590 status
= ocfs2_claim_new_inode(handle
, dir
, parent_fe_bh
,
591 inode_ac
, &suballoc_loc
,
592 &suballoc_bit
, &fe_blkno
);
598 return __ocfs2_mknod_locked(dir
, inode
, dev
, new_fe_bh
,
599 parent_fe_bh
, handle
, inode_ac
,
600 fe_blkno
, suballoc_loc
, suballoc_bit
);
603 static int ocfs2_mkdir(struct inode
*dir
,
604 struct dentry
*dentry
,
609 trace_ocfs2_mkdir(dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
,
610 OCFS2_I(dir
)->ip_blkno
, mode
);
611 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
618 static int ocfs2_create(struct inode
*dir
,
619 struct dentry
*dentry
,
625 trace_ocfs2_create(dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
,
626 (unsigned long long)OCFS2_I(dir
)->ip_blkno
, mode
);
627 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
634 static int ocfs2_link(struct dentry
*old_dentry
,
636 struct dentry
*dentry
)
639 struct inode
*inode
= old_dentry
->d_inode
;
641 struct buffer_head
*fe_bh
= NULL
;
642 struct buffer_head
*parent_fe_bh
= NULL
;
643 struct ocfs2_dinode
*fe
= NULL
;
644 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
645 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
648 trace_ocfs2_link((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
649 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
650 dentry
->d_name
.len
, dentry
->d_name
.name
);
652 if (S_ISDIR(inode
->i_mode
))
655 dquot_initialize(dir
);
657 err
= ocfs2_inode_lock_nested(dir
, &parent_fe_bh
, 1, OI_LS_PARENT
);
669 err
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
674 err
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
676 dentry
->d_name
.len
, &lookup
);
682 err
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
689 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
690 if (ocfs2_read_links_count(fe
) >= ocfs2_link_max(osb
)) {
692 goto out_unlock_inode
;
695 handle
= ocfs2_start_trans(osb
, ocfs2_link_credits(osb
->sb
));
696 if (IS_ERR(handle
)) {
697 err
= PTR_ERR(handle
);
700 goto out_unlock_inode
;
703 /* Starting to change things, restart is no longer possible. */
704 ocfs2_block_signals(&oldset
);
706 err
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
707 OCFS2_JOURNAL_ACCESS_WRITE
);
714 inode
->i_ctime
= CURRENT_TIME
;
715 ocfs2_set_links_count(fe
, inode
->i_nlink
);
716 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
717 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
718 ocfs2_journal_dirty(handle
, fe_bh
);
720 err
= ocfs2_add_entry(handle
, dentry
, inode
,
721 OCFS2_I(inode
)->ip_blkno
,
722 parent_fe_bh
, &lookup
);
724 ocfs2_add_links_count(fe
, -1);
730 err
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
737 d_instantiate(dentry
, inode
);
740 ocfs2_commit_trans(osb
, handle
);
741 ocfs2_unblock_signals(&oldset
);
743 ocfs2_inode_unlock(inode
, 1);
746 ocfs2_inode_unlock(dir
, 1);
749 brelse(parent_fe_bh
);
751 ocfs2_free_dir_lookup_result(&lookup
);
760 * Takes and drops an exclusive lock on the given dentry. This will
761 * force other nodes to drop it.
763 static int ocfs2_remote_dentry_delete(struct dentry
*dentry
)
767 ret
= ocfs2_dentry_lock(dentry
, 1);
771 ocfs2_dentry_unlock(dentry
, 1);
776 static inline int ocfs2_inode_is_unlinkable(struct inode
*inode
)
778 if (S_ISDIR(inode
->i_mode
)) {
779 if (inode
->i_nlink
== 2)
784 if (inode
->i_nlink
== 1)
789 static int ocfs2_unlink(struct inode
*dir
,
790 struct dentry
*dentry
)
793 int child_locked
= 0;
794 bool is_unlinkable
= false;
795 struct inode
*inode
= dentry
->d_inode
;
796 struct inode
*orphan_dir
= NULL
;
797 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
799 struct ocfs2_dinode
*fe
= NULL
;
800 struct buffer_head
*fe_bh
= NULL
;
801 struct buffer_head
*parent_node_bh
= NULL
;
802 handle_t
*handle
= NULL
;
803 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
804 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
805 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
807 trace_ocfs2_unlink(dir
, dentry
, dentry
->d_name
.len
,
809 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
810 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
812 dquot_initialize(dir
);
814 BUG_ON(dentry
->d_parent
->d_inode
!= dir
);
816 if (inode
== osb
->root_inode
)
819 status
= ocfs2_inode_lock_nested(dir
, &parent_node_bh
, 1,
822 if (status
!= -ENOENT
)
827 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
828 dentry
->d_name
.len
, &blkno
, dir
,
831 if (status
!= -ENOENT
)
836 if (OCFS2_I(inode
)->ip_blkno
!= blkno
) {
839 trace_ocfs2_unlink_noent(
840 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
841 (unsigned long long)blkno
,
842 OCFS2_I(inode
)->ip_flags
);
846 status
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
848 if (status
!= -ENOENT
)
854 if (S_ISDIR(inode
->i_mode
)) {
855 if (inode
->i_nlink
!= 2 || !ocfs2_empty_dir(inode
)) {
861 status
= ocfs2_remote_dentry_delete(dentry
);
863 /* This remote delete should succeed under all normal
869 if (ocfs2_inode_is_unlinkable(inode
)) {
870 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
871 OCFS2_I(inode
)->ip_blkno
,
872 orphan_name
, &orphan_insert
);
877 is_unlinkable
= true;
880 handle
= ocfs2_start_trans(osb
, ocfs2_unlink_credits(osb
->sb
));
881 if (IS_ERR(handle
)) {
882 status
= PTR_ERR(handle
);
888 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
889 OCFS2_JOURNAL_ACCESS_WRITE
);
895 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
897 /* delete the name from the parent dir */
898 status
= ocfs2_delete_entry(handle
, dir
, &lookup
);
904 if (S_ISDIR(inode
->i_mode
))
907 ocfs2_set_links_count(fe
, inode
->i_nlink
);
908 ocfs2_journal_dirty(handle
, fe_bh
);
910 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
911 if (S_ISDIR(inode
->i_mode
))
914 status
= ocfs2_mark_inode_dirty(handle
, dir
, parent_node_bh
);
917 if (S_ISDIR(inode
->i_mode
))
923 status
= ocfs2_orphan_add(osb
, handle
, inode
, fe_bh
,
924 orphan_name
, &orphan_insert
, orphan_dir
);
931 ocfs2_commit_trans(osb
, handle
);
934 ocfs2_inode_unlock(inode
, 1);
936 ocfs2_inode_unlock(dir
, 1);
939 /* This was locked for us in ocfs2_prepare_orphan_dir() */
940 ocfs2_inode_unlock(orphan_dir
, 1);
941 mutex_unlock(&orphan_dir
->i_mutex
);
946 brelse(parent_node_bh
);
948 ocfs2_free_dir_lookup_result(&orphan_insert
);
949 ocfs2_free_dir_lookup_result(&lookup
);
951 if (status
&& (status
!= -ENOTEMPTY
))
958 * The only place this should be used is rename!
959 * if they have the same id, then the 1st one is the only one locked.
961 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
962 struct buffer_head
**bh1
,
963 struct inode
*inode1
,
964 struct buffer_head
**bh2
,
965 struct inode
*inode2
)
968 struct ocfs2_inode_info
*oi1
= OCFS2_I(inode1
);
969 struct ocfs2_inode_info
*oi2
= OCFS2_I(inode2
);
970 struct buffer_head
**tmpbh
;
971 struct inode
*tmpinode
;
973 trace_ocfs2_double_lock((unsigned long long)oi1
->ip_blkno
,
974 (unsigned long long)oi2
->ip_blkno
);
981 /* we always want to lock the one with the lower lockid first. */
982 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
983 if (oi1
->ip_blkno
< oi2
->ip_blkno
) {
984 /* switch id1 and id2 around */
994 status
= ocfs2_inode_lock_nested(inode2
, bh2
, 1,
997 if (status
!= -ENOENT
)
1004 status
= ocfs2_inode_lock_nested(inode1
, bh1
, 1, OI_LS_RENAME2
);
1007 * An error return must mean that no cluster locks
1008 * were held on function exit.
1010 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
1011 ocfs2_inode_unlock(inode2
, 1);
1016 if (status
!= -ENOENT
)
1020 trace_ocfs2_double_lock_end(
1021 (unsigned long long)OCFS2_I(inode1
)->ip_blkno
,
1022 (unsigned long long)OCFS2_I(inode2
)->ip_blkno
);
1030 static void ocfs2_double_unlock(struct inode
*inode1
, struct inode
*inode2
)
1032 ocfs2_inode_unlock(inode1
, 1);
1034 if (inode1
!= inode2
)
1035 ocfs2_inode_unlock(inode2
, 1);
1038 static int ocfs2_rename(struct inode
*old_dir
,
1039 struct dentry
*old_dentry
,
1040 struct inode
*new_dir
,
1041 struct dentry
*new_dentry
)
1043 int status
= 0, rename_lock
= 0, parents_locked
= 0, target_exists
= 0;
1044 int old_child_locked
= 0, new_child_locked
= 0, update_dot_dot
= 0;
1045 struct inode
*old_inode
= old_dentry
->d_inode
;
1046 struct inode
*new_inode
= new_dentry
->d_inode
;
1047 struct inode
*orphan_dir
= NULL
;
1048 struct ocfs2_dinode
*newfe
= NULL
;
1049 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1050 struct buffer_head
*newfe_bh
= NULL
;
1051 struct buffer_head
*old_inode_bh
= NULL
;
1052 struct ocfs2_super
*osb
= NULL
;
1053 u64 newfe_blkno
, old_de_ino
;
1054 handle_t
*handle
= NULL
;
1055 struct buffer_head
*old_dir_bh
= NULL
;
1056 struct buffer_head
*new_dir_bh
= NULL
;
1057 u32 old_dir_nlink
= old_dir
->i_nlink
;
1058 struct ocfs2_dinode
*old_di
;
1059 struct ocfs2_dir_lookup_result old_inode_dot_dot_res
= { NULL
, };
1060 struct ocfs2_dir_lookup_result target_lookup_res
= { NULL
, };
1061 struct ocfs2_dir_lookup_result old_entry_lookup
= { NULL
, };
1062 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
1063 struct ocfs2_dir_lookup_result target_insert
= { NULL
, };
1065 /* At some point it might be nice to break this function up a
1068 trace_ocfs2_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
1069 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
1070 new_dentry
->d_name
.len
, new_dentry
->d_name
.name
);
1072 dquot_initialize(old_dir
);
1073 dquot_initialize(new_dir
);
1075 osb
= OCFS2_SB(old_dir
->i_sb
);
1078 if (!igrab(new_inode
))
1082 /* Assume a directory hierarchy thusly:
1085 * a,b,c, and d are all directories.
1087 * from cwd of 'a' on both nodes:
1091 * And that's why, just like the VFS, we need a file system
1093 if (old_dir
!= new_dir
&& S_ISDIR(old_inode
->i_mode
)) {
1094 status
= ocfs2_rename_lock(osb
);
1102 /* if old and new are the same, this'll just do one lock. */
1103 status
= ocfs2_double_lock(osb
, &old_dir_bh
, old_dir
,
1104 &new_dir_bh
, new_dir
);
1111 /* make sure both dirs have bhs
1112 * get an extra ref on old_dir_bh if old==new */
1115 new_dir_bh
= old_dir_bh
;
1118 mlog(ML_ERROR
, "no old_dir_bh!\n");
1125 * Aside from allowing a meta data update, the locking here
1126 * also ensures that the downconvert thread on other nodes
1127 * won't have to concurrently downconvert the inode and the
1130 status
= ocfs2_inode_lock_nested(old_inode
, &old_inode_bh
, 1,
1133 if (status
!= -ENOENT
)
1137 old_child_locked
= 1;
1139 status
= ocfs2_remote_dentry_delete(old_dentry
);
1145 if (S_ISDIR(old_inode
->i_mode
)) {
1146 u64 old_inode_parent
;
1149 status
= ocfs2_find_files_on_disk("..", 2, &old_inode_parent
,
1151 &old_inode_dot_dot_res
);
1157 if (old_inode_parent
!= OCFS2_I(old_dir
)->ip_blkno
) {
1162 if (!new_inode
&& new_dir
!= old_dir
&&
1163 new_dir
->i_nlink
>= ocfs2_link_max(osb
)) {
1169 status
= ocfs2_lookup_ino_from_name(old_dir
, old_dentry
->d_name
.name
,
1170 old_dentry
->d_name
.len
,
1178 * Check for inode number is _not_ due to possible IO errors.
1179 * We might rmdir the source, keep it as pwd of some process
1180 * and merrily kill the link to whatever was created under the
1181 * same name. Goodbye sticky bit ;-<
1183 if (old_de_ino
!= OCFS2_I(old_inode
)->ip_blkno
) {
1188 /* check if the target already exists (in which case we need
1190 status
= ocfs2_find_files_on_disk(new_dentry
->d_name
.name
,
1191 new_dentry
->d_name
.len
,
1192 &newfe_blkno
, new_dir
,
1193 &target_lookup_res
);
1194 /* The only error we allow here is -ENOENT because the new
1195 * file not existing is perfectly valid. */
1196 if ((status
< 0) && (status
!= -ENOENT
)) {
1197 /* If we cannot find the file specified we should just */
1198 /* return the error... */
1205 if (!target_exists
&& new_inode
) {
1207 * Target was unlinked by another node while we were
1208 * waiting to get to ocfs2_rename(). There isn't
1209 * anything we can do here to help the situation, so
1210 * bubble up the appropriate error.
1216 /* In case we need to overwrite an existing file, we blow it
1218 if (target_exists
) {
1219 /* VFS didn't think there existed an inode here, but
1220 * someone else in the cluster must have raced our
1221 * rename to create one. Today we error cleanly, in
1222 * the future we should consider calling iget to build
1223 * a new struct inode for this entry. */
1227 trace_ocfs2_rename_target_exists(new_dentry
->d_name
.len
,
1228 new_dentry
->d_name
.name
);
1232 if (OCFS2_I(new_inode
)->ip_blkno
!= newfe_blkno
) {
1235 trace_ocfs2_rename_disagree(
1236 (unsigned long long)OCFS2_I(new_inode
)->ip_blkno
,
1237 (unsigned long long)newfe_blkno
,
1238 OCFS2_I(new_inode
)->ip_flags
);
1242 status
= ocfs2_inode_lock(new_inode
, &newfe_bh
, 1);
1244 if (status
!= -ENOENT
)
1248 new_child_locked
= 1;
1250 status
= ocfs2_remote_dentry_delete(new_dentry
);
1256 newfe
= (struct ocfs2_dinode
*) newfe_bh
->b_data
;
1258 trace_ocfs2_rename_over_existing(
1259 (unsigned long long)newfe_blkno
, newfe_bh
, newfe_bh
?
1260 (unsigned long long)newfe_bh
->b_blocknr
: 0ULL);
1262 if (S_ISDIR(new_inode
->i_mode
) || (new_inode
->i_nlink
== 1)) {
1263 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
1264 OCFS2_I(new_inode
)->ip_blkno
,
1265 orphan_name
, &orphan_insert
);
1272 BUG_ON(new_dentry
->d_parent
->d_inode
!= new_dir
);
1274 status
= ocfs2_check_dir_for_entry(new_dir
,
1275 new_dentry
->d_name
.name
,
1276 new_dentry
->d_name
.len
);
1280 status
= ocfs2_prepare_dir_for_insert(osb
, new_dir
, new_dir_bh
,
1281 new_dentry
->d_name
.name
,
1282 new_dentry
->d_name
.len
,
1290 handle
= ocfs2_start_trans(osb
, ocfs2_rename_credits(osb
->sb
));
1291 if (IS_ERR(handle
)) {
1292 status
= PTR_ERR(handle
);
1298 if (target_exists
) {
1299 if (S_ISDIR(new_inode
->i_mode
)) {
1300 if (new_inode
->i_nlink
!= 2 ||
1301 !ocfs2_empty_dir(new_inode
)) {
1302 status
= -ENOTEMPTY
;
1306 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(new_inode
),
1308 OCFS2_JOURNAL_ACCESS_WRITE
);
1314 if (S_ISDIR(new_inode
->i_mode
) ||
1315 (ocfs2_read_links_count(newfe
) == 1)) {
1316 status
= ocfs2_orphan_add(osb
, handle
, new_inode
,
1317 newfe_bh
, orphan_name
,
1318 &orphan_insert
, orphan_dir
);
1325 /* change the dirent to point to the correct inode */
1326 status
= ocfs2_update_entry(new_dir
, handle
, &target_lookup_res
,
1332 new_dir
->i_version
++;
1334 if (S_ISDIR(new_inode
->i_mode
))
1335 ocfs2_set_links_count(newfe
, 0);
1337 ocfs2_add_links_count(newfe
, -1);
1338 ocfs2_journal_dirty(handle
, newfe_bh
);
1340 /* if the name was not found in new_dir, add it now */
1341 status
= ocfs2_add_entry(handle
, new_dentry
, old_inode
,
1342 OCFS2_I(old_inode
)->ip_blkno
,
1343 new_dir_bh
, &target_insert
);
1346 old_inode
->i_ctime
= CURRENT_TIME
;
1347 mark_inode_dirty(old_inode
);
1349 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(old_inode
),
1351 OCFS2_JOURNAL_ACCESS_WRITE
);
1353 old_di
= (struct ocfs2_dinode
*) old_inode_bh
->b_data
;
1355 old_di
->i_ctime
= cpu_to_le64(old_inode
->i_ctime
.tv_sec
);
1356 old_di
->i_ctime_nsec
= cpu_to_le32(old_inode
->i_ctime
.tv_nsec
);
1357 ocfs2_journal_dirty(handle
, old_inode_bh
);
1362 * Now that the name has been added to new_dir, remove the old name.
1364 * We don't keep any directory entry context around until now
1365 * because the insert might have changed the type of directory
1366 * we're dealing with.
1368 status
= ocfs2_find_entry(old_dentry
->d_name
.name
,
1369 old_dentry
->d_name
.len
, old_dir
,
1374 status
= ocfs2_delete_entry(handle
, old_dir
, &old_entry_lookup
);
1381 drop_nlink(new_inode
);
1382 new_inode
->i_ctime
= CURRENT_TIME
;
1384 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1386 if (update_dot_dot
) {
1387 status
= ocfs2_update_entry(old_inode
, handle
,
1388 &old_inode_dot_dot_res
, new_dir
);
1389 drop_nlink(old_dir
);
1391 drop_nlink(new_inode
);
1394 mark_inode_dirty(new_dir
);
1397 mark_inode_dirty(old_dir
);
1398 ocfs2_mark_inode_dirty(handle
, old_dir
, old_dir_bh
);
1400 mark_inode_dirty(new_inode
);
1401 ocfs2_mark_inode_dirty(handle
, new_inode
, newfe_bh
);
1404 if (old_dir
!= new_dir
) {
1405 /* Keep the same times on both directories.*/
1406 new_dir
->i_ctime
= new_dir
->i_mtime
= old_dir
->i_ctime
;
1409 * This will also pick up the i_nlink change from the
1412 ocfs2_mark_inode_dirty(handle
, new_dir
, new_dir_bh
);
1415 if (old_dir_nlink
!= old_dir
->i_nlink
) {
1417 mlog(ML_ERROR
, "need to change nlink for old dir "
1418 "%llu from %d to %d but bh is NULL!\n",
1419 (unsigned long long)OCFS2_I(old_dir
)->ip_blkno
,
1420 (int)old_dir_nlink
, old_dir
->i_nlink
);
1422 struct ocfs2_dinode
*fe
;
1423 status
= ocfs2_journal_access_di(handle
,
1424 INODE_CACHE(old_dir
),
1426 OCFS2_JOURNAL_ACCESS_WRITE
);
1427 fe
= (struct ocfs2_dinode
*) old_dir_bh
->b_data
;
1428 ocfs2_set_links_count(fe
, old_dir
->i_nlink
);
1429 ocfs2_journal_dirty(handle
, old_dir_bh
);
1432 ocfs2_dentry_move(old_dentry
, new_dentry
, old_dir
, new_dir
);
1436 ocfs2_rename_unlock(osb
);
1439 ocfs2_commit_trans(osb
, handle
);
1442 ocfs2_double_unlock(old_dir
, new_dir
);
1444 if (old_child_locked
)
1445 ocfs2_inode_unlock(old_inode
, 1);
1447 if (new_child_locked
)
1448 ocfs2_inode_unlock(new_inode
, 1);
1451 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1452 ocfs2_inode_unlock(orphan_dir
, 1);
1453 mutex_unlock(&orphan_dir
->i_mutex
);
1458 sync_mapping_buffers(old_inode
->i_mapping
);
1463 ocfs2_free_dir_lookup_result(&target_lookup_res
);
1464 ocfs2_free_dir_lookup_result(&old_entry_lookup
);
1465 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res
);
1466 ocfs2_free_dir_lookup_result(&orphan_insert
);
1467 ocfs2_free_dir_lookup_result(&target_insert
);
1470 brelse(old_inode_bh
);
1481 * we expect i_size = strlen(symname). Copy symname into the file
1482 * data, including the null terminator.
1484 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
1486 struct inode
*inode
,
1487 const char *symname
)
1489 struct buffer_head
**bhs
= NULL
;
1491 struct super_block
*sb
= osb
->sb
;
1492 u64 p_blkno
, p_blocks
;
1493 int virtual, blocks
, status
, i
, bytes_left
;
1495 bytes_left
= i_size_read(inode
) + 1;
1496 /* we can't trust i_blocks because we're actually going to
1497 * write i_size + 1 bytes. */
1498 blocks
= (bytes_left
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
1500 trace_ocfs2_create_symlink_data((unsigned long long)inode
->i_blocks
,
1501 i_size_read(inode
), blocks
);
1503 /* Sanity check -- make sure we're going to fit. */
1505 ocfs2_clusters_to_bytes(sb
, OCFS2_I(inode
)->ip_clusters
)) {
1511 bhs
= kcalloc(blocks
, sizeof(struct buffer_head
*), GFP_KERNEL
);
1518 status
= ocfs2_extent_map_get_blocks(inode
, 0, &p_blkno
, &p_blocks
,
1525 /* links can never be larger than one cluster so we know this
1526 * is all going to be contiguous, but do a sanity check
1528 if ((p_blocks
<< sb
->s_blocksize_bits
) < bytes_left
) {
1535 while(bytes_left
> 0) {
1536 c
= &symname
[virtual * sb
->s_blocksize
];
1538 bhs
[virtual] = sb_getblk(sb
, p_blkno
);
1539 if (!bhs
[virtual]) {
1544 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
),
1547 status
= ocfs2_journal_access(handle
, INODE_CACHE(inode
),
1549 OCFS2_JOURNAL_ACCESS_CREATE
);
1555 memset(bhs
[virtual]->b_data
, 0, sb
->s_blocksize
);
1557 memcpy(bhs
[virtual]->b_data
, c
,
1558 (bytes_left
> sb
->s_blocksize
) ? sb
->s_blocksize
:
1561 ocfs2_journal_dirty(handle
, bhs
[virtual]);
1565 bytes_left
-= sb
->s_blocksize
;
1572 for(i
= 0; i
< blocks
; i
++)
1582 static int ocfs2_symlink(struct inode
*dir
,
1583 struct dentry
*dentry
,
1584 const char *symname
)
1586 int status
, l
, credits
;
1588 struct ocfs2_super
*osb
= NULL
;
1589 struct inode
*inode
= NULL
;
1590 struct super_block
*sb
;
1591 struct buffer_head
*new_fe_bh
= NULL
;
1592 struct buffer_head
*parent_fe_bh
= NULL
;
1593 struct ocfs2_dinode
*fe
= NULL
;
1594 struct ocfs2_dinode
*dirfe
;
1595 handle_t
*handle
= NULL
;
1596 struct ocfs2_alloc_context
*inode_ac
= NULL
;
1597 struct ocfs2_alloc_context
*data_ac
= NULL
;
1598 struct ocfs2_alloc_context
*xattr_ac
= NULL
;
1599 int want_clusters
= 0;
1600 int xattr_credits
= 0;
1601 struct ocfs2_security_xattr_info si
= {
1604 int did_quota
= 0, did_quota_inode
= 0;
1605 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
1607 int did_block_signals
= 0;
1609 trace_ocfs2_symlink_begin(dir
, dentry
, symname
,
1610 dentry
->d_name
.len
, dentry
->d_name
.name
);
1612 dquot_initialize(dir
);
1617 l
= strlen(symname
) + 1;
1619 credits
= ocfs2_calc_symlink_credits(sb
);
1621 /* lock the parent directory */
1622 status
= ocfs2_inode_lock(dir
, &parent_fe_bh
, 1);
1624 if (status
!= -ENOENT
)
1629 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
1630 if (!ocfs2_read_links_count(dirfe
)) {
1631 /* can't make a file in a deleted directory. */
1636 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
1637 dentry
->d_name
.len
);
1641 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
1642 dentry
->d_name
.name
,
1643 dentry
->d_name
.len
, &lookup
);
1649 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
1651 if (status
!= -ENOSPC
)
1656 inode
= ocfs2_get_init_inode(dir
, S_IFLNK
| S_IRWXUGO
);
1663 /* get security xattr */
1664 status
= ocfs2_init_security_get(inode
, dir
, &dentry
->d_name
, &si
);
1666 if (status
== -EOPNOTSUPP
)
1674 /* calculate meta data/clusters for setting security xattr */
1676 status
= ocfs2_calc_security_init(dir
, &si
, &want_clusters
,
1677 &xattr_credits
, &xattr_ac
);
1684 /* don't reserve bitmap space for fast symlinks. */
1685 if (l
> ocfs2_fast_symlink_chars(sb
))
1688 status
= ocfs2_reserve_clusters(osb
, want_clusters
, &data_ac
);
1690 if (status
!= -ENOSPC
)
1695 handle
= ocfs2_start_trans(osb
, credits
+ xattr_credits
);
1696 if (IS_ERR(handle
)) {
1697 status
= PTR_ERR(handle
);
1703 /* Starting to change things, restart is no longer possible. */
1704 ocfs2_block_signals(&oldset
);
1705 did_block_signals
= 1;
1707 status
= dquot_alloc_inode(inode
);
1710 did_quota_inode
= 1;
1712 trace_ocfs2_symlink_create(dir
, dentry
, dentry
->d_name
.len
,
1713 dentry
->d_name
.name
,
1714 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
1717 status
= ocfs2_mknod_locked(osb
, dir
, inode
,
1718 0, &new_fe_bh
, parent_fe_bh
, handle
,
1725 fe
= (struct ocfs2_dinode
*) new_fe_bh
->b_data
;
1728 inode
->i_op
= &ocfs2_symlink_inode_operations
;
1729 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1732 status
= dquot_alloc_space_nodirty(inode
,
1733 ocfs2_clusters_to_bytes(osb
->sb
, 1));
1737 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
1738 status
= ocfs2_add_inode_data(osb
, inode
, &offset
, 1, 0,
1740 handle
, data_ac
, NULL
,
1743 if (status
!= -ENOSPC
&& status
!= -EINTR
) {
1745 "Failed to extend file to %llu\n",
1746 (unsigned long long)newsize
);
1752 i_size_write(inode
, newsize
);
1753 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1755 inode
->i_mapping
->a_ops
= &ocfs2_fast_symlink_aops
;
1756 memcpy((char *) fe
->id2
.i_symlink
, symname
, l
);
1757 i_size_write(inode
, newsize
);
1758 inode
->i_blocks
= 0;
1761 status
= ocfs2_mark_inode_dirty(handle
, inode
, new_fe_bh
);
1767 if (!ocfs2_inode_is_fast_symlink(inode
)) {
1768 status
= ocfs2_create_symlink_data(osb
, handle
, inode
,
1777 status
= ocfs2_init_security_set(handle
, inode
, new_fe_bh
, &si
,
1786 * Do this before adding the entry to the directory. We add
1787 * also set d_op after success so that ->d_iput() will cleanup
1788 * the dentry lock even if ocfs2_add_entry() fails below.
1790 status
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
1796 status
= ocfs2_add_entry(handle
, dentry
, inode
,
1797 le64_to_cpu(fe
->i_blkno
), parent_fe_bh
,
1804 insert_inode_hash(inode
);
1805 d_instantiate(dentry
, inode
);
1807 if (status
< 0 && did_quota
)
1808 dquot_free_space_nodirty(inode
,
1809 ocfs2_clusters_to_bytes(osb
->sb
, 1));
1810 if (status
< 0 && did_quota_inode
)
1811 dquot_free_inode(inode
);
1813 ocfs2_commit_trans(osb
, handle
);
1815 ocfs2_inode_unlock(dir
, 1);
1816 if (did_block_signals
)
1817 ocfs2_unblock_signals(&oldset
);
1820 brelse(parent_fe_bh
);
1823 ocfs2_free_dir_lookup_result(&lookup
);
1825 ocfs2_free_alloc_context(inode_ac
);
1827 ocfs2_free_alloc_context(data_ac
);
1829 ocfs2_free_alloc_context(xattr_ac
);
1830 if ((status
< 0) && inode
) {
1831 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SKIP_ORPHAN_DIR
;
1842 static int ocfs2_blkno_stringify(u64 blkno
, char *name
)
1844 int status
, namelen
;
1846 namelen
= snprintf(name
, OCFS2_ORPHAN_NAMELEN
+ 1, "%016llx",
1856 if (namelen
!= OCFS2_ORPHAN_NAMELEN
) {
1862 trace_ocfs2_blkno_stringify(blkno
, name
, namelen
);
1871 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super
*osb
,
1872 struct inode
**ret_orphan_dir
,
1873 struct buffer_head
**ret_orphan_dir_bh
)
1875 struct inode
*orphan_dir_inode
;
1876 struct buffer_head
*orphan_dir_bh
= NULL
;
1879 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
1880 ORPHAN_DIR_SYSTEM_INODE
,
1882 if (!orphan_dir_inode
) {
1888 mutex_lock(&orphan_dir_inode
->i_mutex
);
1890 ret
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
1892 mutex_unlock(&orphan_dir_inode
->i_mutex
);
1893 iput(orphan_dir_inode
);
1899 *ret_orphan_dir
= orphan_dir_inode
;
1900 *ret_orphan_dir_bh
= orphan_dir_bh
;
1905 static int __ocfs2_prepare_orphan_dir(struct inode
*orphan_dir_inode
,
1906 struct buffer_head
*orphan_dir_bh
,
1909 struct ocfs2_dir_lookup_result
*lookup
)
1912 struct ocfs2_super
*osb
= OCFS2_SB(orphan_dir_inode
->i_sb
);
1914 ret
= ocfs2_blkno_stringify(blkno
, name
);
1920 ret
= ocfs2_prepare_dir_for_insert(osb
, orphan_dir_inode
,
1921 orphan_dir_bh
, name
,
1922 OCFS2_ORPHAN_NAMELEN
, lookup
);
1932 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1933 * insertion of an orphan.
1934 * @osb: ocfs2 file system
1935 * @ret_orphan_dir: Orphan dir inode - returned locked!
1936 * @blkno: Actual block number of the inode to be inserted into orphan dir.
1937 * @lookup: dir lookup result, to be passed back into functions like
1940 * Returns zero on success and the ret_orphan_dir, name and lookup
1941 * fields will be populated.
1943 * Returns non-zero on failure.
1945 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
1946 struct inode
**ret_orphan_dir
,
1949 struct ocfs2_dir_lookup_result
*lookup
)
1951 struct inode
*orphan_dir_inode
= NULL
;
1952 struct buffer_head
*orphan_dir_bh
= NULL
;
1955 ret
= ocfs2_lookup_lock_orphan_dir(osb
, &orphan_dir_inode
,
1962 ret
= __ocfs2_prepare_orphan_dir(orphan_dir_inode
, orphan_dir_bh
,
1963 blkno
, name
, lookup
);
1969 *ret_orphan_dir
= orphan_dir_inode
;
1972 brelse(orphan_dir_bh
);
1975 ocfs2_inode_unlock(orphan_dir_inode
, 1);
1976 mutex_unlock(&orphan_dir_inode
->i_mutex
);
1977 iput(orphan_dir_inode
);
1985 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
1987 struct inode
*inode
,
1988 struct buffer_head
*fe_bh
,
1990 struct ocfs2_dir_lookup_result
*lookup
,
1991 struct inode
*orphan_dir_inode
)
1993 struct buffer_head
*orphan_dir_bh
= NULL
;
1995 struct ocfs2_dinode
*orphan_fe
;
1996 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
1998 trace_ocfs2_orphan_add_begin(
1999 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
2001 status
= ocfs2_read_inode_block(orphan_dir_inode
, &orphan_dir_bh
);
2007 status
= ocfs2_journal_access_di(handle
,
2008 INODE_CACHE(orphan_dir_inode
),
2010 OCFS2_JOURNAL_ACCESS_WRITE
);
2017 * We're going to journal the change of i_flags and i_orphaned_slot.
2018 * It's safe anyway, though some callers may duplicate the journaling.
2019 * Journaling within the func just make the logic look more
2022 status
= ocfs2_journal_access_di(handle
,
2025 OCFS2_JOURNAL_ACCESS_WRITE
);
2031 /* we're a cluster, and nlink can change on disk from
2032 * underneath us... */
2033 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2034 if (S_ISDIR(inode
->i_mode
))
2035 ocfs2_add_links_count(orphan_fe
, 1);
2036 set_nlink(orphan_dir_inode
, ocfs2_read_links_count(orphan_fe
));
2037 ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2039 status
= __ocfs2_add_entry(handle
, orphan_dir_inode
, name
,
2040 OCFS2_ORPHAN_NAMELEN
, inode
,
2041 OCFS2_I(inode
)->ip_blkno
,
2042 orphan_dir_bh
, lookup
);
2048 fe
->i_flags
|= cpu_to_le32(OCFS2_ORPHANED_FL
);
2049 OCFS2_I(inode
)->ip_flags
&= ~OCFS2_INODE_SKIP_ORPHAN_DIR
;
2051 /* Record which orphan dir our inode now resides
2052 * in. delete_inode will use this to determine which orphan
2054 fe
->i_orphaned_slot
= cpu_to_le16(osb
->slot_num
);
2056 ocfs2_journal_dirty(handle
, fe_bh
);
2058 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
2063 if (S_ISDIR(inode
->i_mode
))
2064 ocfs2_add_links_count(orphan_fe
, -1);
2065 set_nlink(orphan_dir_inode
, ocfs2_read_links_count(orphan_fe
));
2069 brelse(orphan_dir_bh
);
2074 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2075 int ocfs2_orphan_del(struct ocfs2_super
*osb
,
2077 struct inode
*orphan_dir_inode
,
2078 struct inode
*inode
,
2079 struct buffer_head
*orphan_dir_bh
)
2081 char name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2082 struct ocfs2_dinode
*orphan_fe
;
2084 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
2086 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2092 trace_ocfs2_orphan_del(
2093 (unsigned long long)OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2094 name
, OCFS2_ORPHAN_NAMELEN
);
2096 /* find it's spot in the orphan directory */
2097 status
= ocfs2_find_entry(name
, OCFS2_ORPHAN_NAMELEN
, orphan_dir_inode
,
2104 /* remove it from the orphan directory */
2105 status
= ocfs2_delete_entry(handle
, orphan_dir_inode
, &lookup
);
2111 status
= ocfs2_journal_access_di(handle
,
2112 INODE_CACHE(orphan_dir_inode
),
2114 OCFS2_JOURNAL_ACCESS_WRITE
);
2120 /* do the i_nlink dance! :) */
2121 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2122 if (S_ISDIR(inode
->i_mode
))
2123 ocfs2_add_links_count(orphan_fe
, -1);
2124 set_nlink(orphan_dir_inode
, ocfs2_read_links_count(orphan_fe
));
2125 ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2128 ocfs2_free_dir_lookup_result(&lookup
);
2136 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2137 * allocated file. This is different from the typical 'add to orphan dir'
2138 * operation in that the inode does not yet exist. This is a problem because
2139 * the orphan dir stringifies the inode block number to come up with it's
2140 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2141 * problem. This function works around it by calling deeper into the orphan
2142 * and suballoc code than other callers. Use this only by necessity.
2143 * @dir: The directory which this inode will ultimately wind up under - not the
2145 * @dir_bh: buffer_head the @dir inode block
2146 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2147 * with the string to be used for orphan dirent. Pass back to the orphan dir
2149 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2151 * @ret_di_blkno: block number where the new inode will be allocated.
2152 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2153 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2155 * Returns zero on success and the ret_orphan_dir, name and lookup
2156 * fields will be populated.
2158 * Returns non-zero on failure.
2160 static int ocfs2_prep_new_orphaned_file(struct inode
*dir
,
2161 struct buffer_head
*dir_bh
,
2163 struct inode
**ret_orphan_dir
,
2165 struct ocfs2_dir_lookup_result
*orphan_insert
,
2166 struct ocfs2_alloc_context
**ret_inode_ac
)
2170 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2171 struct inode
*orphan_dir
= NULL
;
2172 struct buffer_head
*orphan_dir_bh
= NULL
;
2173 struct ocfs2_alloc_context
*inode_ac
= NULL
;
2175 ret
= ocfs2_lookup_lock_orphan_dir(osb
, &orphan_dir
, &orphan_dir_bh
);
2181 /* reserve an inode spot */
2182 ret
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
2189 ret
= ocfs2_find_new_inode_loc(dir
, dir_bh
, inode_ac
,
2196 ret
= __ocfs2_prepare_orphan_dir(orphan_dir
, orphan_dir_bh
,
2197 di_blkno
, orphan_name
, orphan_insert
);
2205 *ret_orphan_dir
= orphan_dir
;
2206 *ret_di_blkno
= di_blkno
;
2207 *ret_inode_ac
= inode_ac
;
2209 * orphan_name and orphan_insert are already up to
2210 * date via prepare_orphan_dir
2213 /* Unroll reserve_new_inode* */
2215 ocfs2_free_alloc_context(inode_ac
);
2217 /* Unroll orphan dir locking */
2218 mutex_unlock(&orphan_dir
->i_mutex
);
2219 ocfs2_inode_unlock(orphan_dir
, 1);
2223 brelse(orphan_dir_bh
);
2228 int ocfs2_create_inode_in_orphan(struct inode
*dir
,
2230 struct inode
**new_inode
)
2232 int status
, did_quota_inode
= 0;
2233 struct inode
*inode
= NULL
;
2234 struct inode
*orphan_dir
= NULL
;
2235 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2236 struct ocfs2_dinode
*di
= NULL
;
2237 handle_t
*handle
= NULL
;
2238 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2239 struct buffer_head
*parent_di_bh
= NULL
;
2240 struct buffer_head
*new_di_bh
= NULL
;
2241 struct ocfs2_alloc_context
*inode_ac
= NULL
;
2242 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
2243 u64
uninitialized_var(di_blkno
), suballoc_loc
;
2246 status
= ocfs2_inode_lock(dir
, &parent_di_bh
, 1);
2248 if (status
!= -ENOENT
)
2253 status
= ocfs2_prep_new_orphaned_file(dir
, parent_di_bh
,
2254 orphan_name
, &orphan_dir
,
2255 &di_blkno
, &orphan_insert
, &inode_ac
);
2257 if (status
!= -ENOSPC
)
2262 inode
= ocfs2_get_init_inode(dir
, mode
);
2269 handle
= ocfs2_start_trans(osb
, ocfs2_mknod_credits(osb
->sb
, 0, 0));
2270 if (IS_ERR(handle
)) {
2271 status
= PTR_ERR(handle
);
2277 status
= dquot_alloc_inode(inode
);
2280 did_quota_inode
= 1;
2282 status
= ocfs2_claim_new_inode_at_loc(handle
, dir
, inode_ac
,
2284 &suballoc_bit
, di_blkno
);
2291 /* do the real work now. */
2292 status
= __ocfs2_mknod_locked(dir
, inode
,
2293 0, &new_di_bh
, parent_di_bh
, handle
,
2294 inode_ac
, di_blkno
, suballoc_loc
,
2301 di
= (struct ocfs2_dinode
*)new_di_bh
->b_data
;
2302 status
= ocfs2_orphan_add(osb
, handle
, inode
, new_di_bh
, orphan_name
,
2303 &orphan_insert
, orphan_dir
);
2309 /* get open lock so that only nodes can't remove it from orphan dir. */
2310 status
= ocfs2_open_lock(inode
);
2314 insert_inode_hash(inode
);
2316 if (status
< 0 && did_quota_inode
)
2317 dquot_free_inode(inode
);
2319 ocfs2_commit_trans(osb
, handle
);
2322 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2323 ocfs2_inode_unlock(orphan_dir
, 1);
2324 mutex_unlock(&orphan_dir
->i_mutex
);
2328 if ((status
< 0) && inode
) {
2334 ocfs2_free_alloc_context(inode_ac
);
2341 ocfs2_free_dir_lookup_result(&orphan_insert
);
2343 ocfs2_inode_unlock(dir
, 1);
2344 brelse(parent_di_bh
);
2348 int ocfs2_mv_orphaned_inode_to_new(struct inode
*dir
,
2349 struct inode
*inode
,
2350 struct dentry
*dentry
)
2353 struct buffer_head
*parent_di_bh
= NULL
;
2354 handle_t
*handle
= NULL
;
2355 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2356 struct ocfs2_dinode
*dir_di
, *di
;
2357 struct inode
*orphan_dir_inode
= NULL
;
2358 struct buffer_head
*orphan_dir_bh
= NULL
;
2359 struct buffer_head
*di_bh
= NULL
;
2360 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
2362 trace_ocfs2_mv_orphaned_inode_to_new(dir
, dentry
,
2363 dentry
->d_name
.len
, dentry
->d_name
.name
,
2364 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
2365 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
2367 status
= ocfs2_inode_lock(dir
, &parent_di_bh
, 1);
2369 if (status
!= -ENOENT
)
2374 dir_di
= (struct ocfs2_dinode
*) parent_di_bh
->b_data
;
2375 if (!dir_di
->i_links_count
) {
2376 /* can't make a file in a deleted directory. */
2381 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
2382 dentry
->d_name
.len
);
2386 /* get a spot inside the dir. */
2387 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_di_bh
,
2388 dentry
->d_name
.name
,
2389 dentry
->d_name
.len
, &lookup
);
2395 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2396 ORPHAN_DIR_SYSTEM_INODE
,
2398 if (!orphan_dir_inode
) {
2404 mutex_lock(&orphan_dir_inode
->i_mutex
);
2406 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
2409 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2410 iput(orphan_dir_inode
);
2414 status
= ocfs2_read_inode_block(inode
, &di_bh
);
2420 handle
= ocfs2_start_trans(osb
, ocfs2_rename_credits(osb
->sb
));
2421 if (IS_ERR(handle
)) {
2422 status
= PTR_ERR(handle
);
2428 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
2429 di_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
2435 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
2442 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2443 di
->i_flags
&= ~cpu_to_le32(OCFS2_ORPHANED_FL
);
2444 di
->i_orphaned_slot
= 0;
2445 set_nlink(inode
, 1);
2446 ocfs2_set_links_count(di
, inode
->i_nlink
);
2447 ocfs2_journal_dirty(handle
, di_bh
);
2449 status
= ocfs2_add_entry(handle
, dentry
, inode
,
2450 OCFS2_I(inode
)->ip_blkno
, parent_di_bh
,
2457 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
2458 OCFS2_I(dir
)->ip_blkno
);
2464 d_instantiate(dentry
, inode
);
2467 ocfs2_commit_trans(osb
, handle
);
2469 ocfs2_inode_unlock(orphan_dir_inode
, 1);
2470 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2471 iput(orphan_dir_inode
);
2474 ocfs2_inode_unlock(dir
, 1);
2477 brelse(parent_di_bh
);
2478 brelse(orphan_dir_bh
);
2480 ocfs2_free_dir_lookup_result(&lookup
);
2488 const struct inode_operations ocfs2_dir_iops
= {
2489 .create
= ocfs2_create
,
2490 .lookup
= ocfs2_lookup
,
2492 .unlink
= ocfs2_unlink
,
2493 .rmdir
= ocfs2_unlink
,
2494 .symlink
= ocfs2_symlink
,
2495 .mkdir
= ocfs2_mkdir
,
2496 .mknod
= ocfs2_mknod
,
2497 .rename
= ocfs2_rename
,
2498 .setattr
= ocfs2_setattr
,
2499 .getattr
= ocfs2_getattr
,
2500 .permission
= ocfs2_permission
,
2501 .setxattr
= generic_setxattr
,
2502 .getxattr
= generic_getxattr
,
2503 .listxattr
= ocfs2_listxattr
,
2504 .removexattr
= generic_removexattr
,
2505 .fiemap
= ocfs2_fiemap
,
2506 .get_acl
= ocfs2_iop_get_acl
,