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 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
98 struct buffer_head
**bh1
,
100 struct buffer_head
**bh2
,
101 struct inode
*inode2
,
104 static void ocfs2_double_unlock(struct inode
*inode1
, struct inode
*inode2
);
105 /* An orphan dir name is an 8 byte value, printed as a hex string */
106 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
108 static struct dentry
*ocfs2_lookup(struct inode
*dir
, struct dentry
*dentry
,
113 struct inode
*inode
= NULL
;
115 struct ocfs2_inode_info
*oi
;
117 trace_ocfs2_lookup(dir
, dentry
, dentry
->d_name
.len
,
119 (unsigned long long)OCFS2_I(dir
)->ip_blkno
, 0);
121 if (dentry
->d_name
.len
> OCFS2_MAX_FILENAME_LEN
) {
122 ret
= ERR_PTR(-ENAMETOOLONG
);
126 status
= ocfs2_inode_lock_nested(dir
, NULL
, 0, OI_LS_PARENT
);
128 if (status
!= -ENOENT
)
130 ret
= ERR_PTR(status
);
134 status
= ocfs2_lookup_ino_from_name(dir
, dentry
->d_name
.name
,
135 dentry
->d_name
.len
, &blkno
);
139 inode
= ocfs2_iget(OCFS2_SB(dir
->i_sb
), blkno
, 0, 0);
141 ret
= ERR_PTR(-EACCES
);
146 /* Clear any orphaned state... If we were able to look up the
147 * inode from a directory, it certainly can't be orphaned. We
148 * might have the bad state from a node which intended to
149 * orphan this inode but crashed before it could commit the
151 spin_lock(&oi
->ip_lock
);
152 oi
->ip_flags
&= ~OCFS2_INODE_MAYBE_ORPHANED
;
153 spin_unlock(&oi
->ip_lock
);
156 ret
= d_splice_alias(inode
, dentry
);
160 * If d_splice_alias() finds a DCACHE_DISCONNECTED
161 * dentry, it will d_move() it on top of ourse. The
162 * return value will indicate this however, so in
163 * those cases, we switch them around for the locking
166 * NOTE: This dentry already has ->d_op set from
167 * ocfs2_get_parent() and ocfs2_get_dentry()
169 if (!IS_ERR_OR_NULL(ret
))
172 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
173 OCFS2_I(dir
)->ip_blkno
);
176 ret
= ERR_PTR(status
);
180 ocfs2_dentry_attach_gen(dentry
);
183 /* Don't drop the cluster lock until *after* the d_add --
184 * unlink on another node will message us to remove that
185 * dentry under this lock so otherwise we can race this with
186 * the downconvert thread and have a stale dentry. */
187 ocfs2_inode_unlock(dir
, 0);
191 trace_ocfs2_lookup_ret(ret
);
196 static struct inode
*ocfs2_get_init_inode(struct inode
*dir
, umode_t mode
)
200 inode
= new_inode(dir
->i_sb
);
202 mlog(ML_ERROR
, "new_inode failed!\n");
206 /* populate as many fields early on as possible - many of
207 * these are used by the support functions here and in
211 inode_init_owner(inode
, dir
, mode
);
212 dquot_initialize(inode
);
216 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super
*osb
,
217 struct dentry
*dentry
, struct inode
*inode
)
219 struct ocfs2_dentry_lock
*dl
= dentry
->d_fsdata
;
221 ocfs2_simple_drop_lockres(osb
, &dl
->dl_lockres
);
222 ocfs2_lock_res_free(&dl
->dl_lockres
);
223 BUG_ON(dl
->dl_count
!= 1);
224 spin_lock(&dentry_attach_lock
);
225 dentry
->d_fsdata
= NULL
;
226 spin_unlock(&dentry_attach_lock
);
231 static int ocfs2_mknod(struct inode
*dir
,
232 struct dentry
*dentry
,
237 struct buffer_head
*parent_fe_bh
= NULL
;
238 handle_t
*handle
= NULL
;
239 struct ocfs2_super
*osb
;
240 struct ocfs2_dinode
*dirfe
;
241 struct buffer_head
*new_fe_bh
= NULL
;
242 struct inode
*inode
= NULL
;
243 struct ocfs2_alloc_context
*inode_ac
= NULL
;
244 struct ocfs2_alloc_context
*data_ac
= NULL
;
245 struct ocfs2_alloc_context
*meta_ac
= NULL
;
246 int want_clusters
= 0;
248 int xattr_credits
= 0;
249 struct ocfs2_security_xattr_info si
= {
252 int did_quota_inode
= 0;
253 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
255 int did_block_signals
= 0;
256 struct posix_acl
*default_acl
= NULL
, *acl
= NULL
;
257 struct ocfs2_dentry_lock
*dl
= NULL
;
259 trace_ocfs2_mknod(dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
,
260 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
261 (unsigned long)dev
, mode
);
263 dquot_initialize(dir
);
265 /* get our super block */
266 osb
= OCFS2_SB(dir
->i_sb
);
268 status
= ocfs2_inode_lock(dir
, &parent_fe_bh
, 1);
270 if (status
!= -ENOENT
)
275 if (S_ISDIR(mode
) && (dir
->i_nlink
>= ocfs2_link_max(osb
))) {
280 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
281 if (!ocfs2_read_links_count(dirfe
)) {
282 /* can't make a file in a deleted directory. */
287 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
292 /* get a spot inside the dir. */
293 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
295 dentry
->d_name
.len
, &lookup
);
301 /* reserve an inode spot */
302 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
304 if (status
!= -ENOSPC
)
309 inode
= ocfs2_get_init_inode(dir
, mode
);
316 /* get security xattr */
317 status
= ocfs2_init_security_get(inode
, dir
, &dentry
->d_name
, &si
);
319 if (status
== -EOPNOTSUPP
)
327 /* calculate meta data/clusters for setting security and acl xattr */
328 status
= ocfs2_calc_xattr_init(dir
, parent_fe_bh
, mode
,
330 &xattr_credits
, &want_meta
);
336 /* Reserve a cluster if creating an extent based directory. */
337 if (S_ISDIR(mode
) && !ocfs2_supports_inline_data(osb
)) {
340 /* Dir indexing requires extra space as well */
341 if (ocfs2_supports_indexed_dirs(osb
))
345 status
= ocfs2_reserve_new_metadata_blocks(osb
, want_meta
, &meta_ac
);
347 if (status
!= -ENOSPC
)
352 status
= ocfs2_reserve_clusters(osb
, want_clusters
, &data_ac
);
354 if (status
!= -ENOSPC
)
359 status
= posix_acl_create(dir
, &mode
, &default_acl
, &acl
);
365 handle
= ocfs2_start_trans(osb
, ocfs2_mknod_credits(osb
->sb
,
368 if (IS_ERR(handle
)) {
369 status
= PTR_ERR(handle
);
375 /* Starting to change things, restart is no longer possible. */
376 ocfs2_block_signals(&oldset
);
377 did_block_signals
= 1;
379 status
= dquot_alloc_inode(inode
);
384 /* do the real work now. */
385 status
= ocfs2_mknod_locked(osb
, dir
, inode
, dev
,
386 &new_fe_bh
, parent_fe_bh
, handle
,
394 status
= ocfs2_fill_new_dir(osb
, handle
, dir
, inode
,
395 new_fe_bh
, data_ac
, meta_ac
);
401 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(dir
),
403 OCFS2_JOURNAL_ACCESS_WRITE
);
408 ocfs2_add_links_count(dirfe
, 1);
409 ocfs2_journal_dirty(handle
, parent_fe_bh
);
414 status
= ocfs2_set_acl(handle
, inode
, new_fe_bh
,
415 ACL_TYPE_DEFAULT
, default_acl
,
418 if (!status
&& acl
) {
419 status
= ocfs2_set_acl(handle
, inode
, new_fe_bh
,
420 ACL_TYPE_ACCESS
, acl
,
430 status
= ocfs2_init_security_set(handle
, inode
, new_fe_bh
, &si
,
439 * Do this before adding the entry to the directory. We add
440 * also set d_op after success so that ->d_iput() will cleanup
441 * the dentry lock even if ocfs2_add_entry() fails below.
443 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
444 OCFS2_I(dir
)->ip_blkno
);
450 dl
= dentry
->d_fsdata
;
452 status
= ocfs2_add_entry(handle
, dentry
, inode
,
453 OCFS2_I(inode
)->ip_blkno
, parent_fe_bh
,
460 insert_inode_hash(inode
);
461 d_instantiate(dentry
, inode
);
465 posix_acl_release(default_acl
);
467 posix_acl_release(acl
);
468 if (status
< 0 && did_quota_inode
)
469 dquot_free_inode(inode
);
471 ocfs2_commit_trans(osb
, handle
);
473 ocfs2_inode_unlock(dir
, 1);
474 if (did_block_signals
)
475 ocfs2_unblock_signals(&oldset
);
478 brelse(parent_fe_bh
);
481 ocfs2_free_dir_lookup_result(&lookup
);
484 ocfs2_free_alloc_context(inode_ac
);
487 ocfs2_free_alloc_context(data_ac
);
490 ocfs2_free_alloc_context(meta_ac
);
493 * We should call iput after the i_mutex of the bitmap been
494 * unlocked in ocfs2_free_alloc_context, or the
495 * ocfs2_delete_inode will mutex_lock again.
497 if ((status
< 0) && inode
) {
499 ocfs2_cleanup_add_entry_failure(osb
, dentry
, inode
);
501 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SKIP_ORPHAN_DIR
;
512 static int __ocfs2_mknod_locked(struct inode
*dir
,
515 struct buffer_head
**new_fe_bh
,
516 struct buffer_head
*parent_fe_bh
,
518 struct ocfs2_alloc_context
*inode_ac
,
519 u64 fe_blkno
, u64 suballoc_loc
, u16 suballoc_bit
)
522 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
523 struct ocfs2_dinode
*fe
= NULL
;
524 struct ocfs2_extent_list
*fel
;
526 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
530 /* populate as many fields early on as possible - many of
531 * these are used by the support functions here and in
533 inode
->i_ino
= ino_from_blkno(osb
->sb
, fe_blkno
);
534 OCFS2_I(inode
)->ip_blkno
= fe_blkno
;
535 spin_lock(&osb
->osb_lock
);
536 inode
->i_generation
= osb
->s_next_generation
++;
537 spin_unlock(&osb
->osb_lock
);
539 *new_fe_bh
= sb_getblk(osb
->sb
, fe_blkno
);
545 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
), *new_fe_bh
);
547 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
549 OCFS2_JOURNAL_ACCESS_CREATE
);
555 fe
= (struct ocfs2_dinode
*) (*new_fe_bh
)->b_data
;
556 memset(fe
, 0, osb
->sb
->s_blocksize
);
558 fe
->i_generation
= cpu_to_le32(inode
->i_generation
);
559 fe
->i_fs_generation
= cpu_to_le32(osb
->fs_generation
);
560 fe
->i_blkno
= cpu_to_le64(fe_blkno
);
561 fe
->i_suballoc_loc
= cpu_to_le64(suballoc_loc
);
562 fe
->i_suballoc_bit
= cpu_to_le16(suballoc_bit
);
563 fe
->i_suballoc_slot
= cpu_to_le16(inode_ac
->ac_alloc_slot
);
564 fe
->i_uid
= cpu_to_le32(i_uid_read(inode
));
565 fe
->i_gid
= cpu_to_le32(i_gid_read(inode
));
566 fe
->i_mode
= cpu_to_le16(inode
->i_mode
);
567 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
))
568 fe
->id1
.dev1
.i_rdev
= cpu_to_le64(huge_encode_dev(dev
));
570 ocfs2_set_links_count(fe
, inode
->i_nlink
);
572 fe
->i_last_eb_blk
= 0;
573 strcpy(fe
->i_signature
, OCFS2_INODE_SIGNATURE
);
574 fe
->i_flags
|= cpu_to_le32(OCFS2_VALID_FL
);
575 fe
->i_atime
= fe
->i_ctime
= fe
->i_mtime
=
576 cpu_to_le64(CURRENT_TIME
.tv_sec
);
577 fe
->i_mtime_nsec
= fe
->i_ctime_nsec
= fe
->i_atime_nsec
=
578 cpu_to_le32(CURRENT_TIME
.tv_nsec
);
582 * If supported, directories start with inline data. If inline
583 * isn't supported, but indexing is, we start them as indexed.
585 feat
= le16_to_cpu(fe
->i_dyn_features
);
586 if (S_ISDIR(inode
->i_mode
) && ocfs2_supports_inline_data(osb
)) {
587 fe
->i_dyn_features
= cpu_to_le16(feat
| OCFS2_INLINE_DATA_FL
);
589 fe
->id2
.i_data
.id_count
= cpu_to_le16(
590 ocfs2_max_inline_data_with_xattr(osb
->sb
, fe
));
592 fel
= &fe
->id2
.i_list
;
593 fel
->l_tree_depth
= 0;
594 fel
->l_next_free_rec
= 0;
595 fel
->l_count
= cpu_to_le16(ocfs2_extent_recs_per_inode(osb
->sb
));
598 ocfs2_journal_dirty(handle
, *new_fe_bh
);
600 ocfs2_populate_inode(inode
, fe
, 1);
601 ocfs2_ci_set_new(osb
, INODE_CACHE(inode
));
602 if (!ocfs2_mount_local(osb
)) {
603 status
= ocfs2_create_new_inode_locks(inode
);
608 oi
->i_sync_tid
= handle
->h_transaction
->t_tid
;
609 oi
->i_datasync_tid
= handle
->h_transaction
->t_tid
;
624 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
628 struct buffer_head
**new_fe_bh
,
629 struct buffer_head
*parent_fe_bh
,
631 struct ocfs2_alloc_context
*inode_ac
)
634 u64 suballoc_loc
, fe_blkno
= 0;
639 status
= ocfs2_claim_new_inode(handle
, dir
, parent_fe_bh
,
640 inode_ac
, &suballoc_loc
,
641 &suballoc_bit
, &fe_blkno
);
647 return __ocfs2_mknod_locked(dir
, inode
, dev
, new_fe_bh
,
648 parent_fe_bh
, handle
, inode_ac
,
649 fe_blkno
, suballoc_loc
, suballoc_bit
);
652 static int ocfs2_mkdir(struct inode
*dir
,
653 struct dentry
*dentry
,
658 trace_ocfs2_mkdir(dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
,
659 OCFS2_I(dir
)->ip_blkno
, mode
);
660 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
667 static int ocfs2_create(struct inode
*dir
,
668 struct dentry
*dentry
,
674 trace_ocfs2_create(dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
,
675 (unsigned long long)OCFS2_I(dir
)->ip_blkno
, mode
);
676 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
683 static int ocfs2_link(struct dentry
*old_dentry
,
685 struct dentry
*dentry
)
688 struct inode
*inode
= old_dentry
->d_inode
;
689 struct inode
*old_dir
= old_dentry
->d_parent
->d_inode
;
691 struct buffer_head
*fe_bh
= NULL
;
692 struct buffer_head
*old_dir_bh
= NULL
;
693 struct buffer_head
*parent_fe_bh
= NULL
;
694 struct ocfs2_dinode
*fe
= NULL
;
695 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
696 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
700 trace_ocfs2_link((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
701 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
702 dentry
->d_name
.len
, dentry
->d_name
.name
);
704 if (S_ISDIR(inode
->i_mode
))
707 dquot_initialize(dir
);
709 err
= ocfs2_double_lock(osb
, &old_dir_bh
, old_dir
,
710 &parent_fe_bh
, dir
, 0);
717 /* make sure both dirs have bhs
718 * get an extra ref on old_dir_bh if old==new */
721 parent_fe_bh
= old_dir_bh
;
722 get_bh(parent_fe_bh
);
724 mlog(ML_ERROR
, "%s: no old_dir_bh!\n", osb
->uuid_str
);
735 err
= ocfs2_lookup_ino_from_name(old_dir
, old_dentry
->d_name
.name
,
736 old_dentry
->d_name
.len
, &old_de_ino
);
743 * Check whether another node removed the source inode while we
746 if (old_de_ino
!= OCFS2_I(inode
)->ip_blkno
) {
751 err
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
756 err
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
758 dentry
->d_name
.len
, &lookup
);
764 err
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
771 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
772 if (ocfs2_read_links_count(fe
) >= ocfs2_link_max(osb
)) {
774 goto out_unlock_inode
;
777 handle
= ocfs2_start_trans(osb
, ocfs2_link_credits(osb
->sb
));
778 if (IS_ERR(handle
)) {
779 err
= PTR_ERR(handle
);
782 goto out_unlock_inode
;
785 /* Starting to change things, restart is no longer possible. */
786 ocfs2_block_signals(&oldset
);
788 err
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
789 OCFS2_JOURNAL_ACCESS_WRITE
);
796 inode
->i_ctime
= CURRENT_TIME
;
797 ocfs2_set_links_count(fe
, inode
->i_nlink
);
798 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
799 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
800 ocfs2_journal_dirty(handle
, fe_bh
);
802 err
= ocfs2_add_entry(handle
, dentry
, inode
,
803 OCFS2_I(inode
)->ip_blkno
,
804 parent_fe_bh
, &lookup
);
806 ocfs2_add_links_count(fe
, -1);
812 err
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
819 d_instantiate(dentry
, inode
);
822 ocfs2_commit_trans(osb
, handle
);
823 ocfs2_unblock_signals(&oldset
);
825 ocfs2_inode_unlock(inode
, 1);
828 ocfs2_double_unlock(old_dir
, dir
);
831 brelse(parent_fe_bh
);
834 ocfs2_free_dir_lookup_result(&lookup
);
843 * Takes and drops an exclusive lock on the given dentry. This will
844 * force other nodes to drop it.
846 static int ocfs2_remote_dentry_delete(struct dentry
*dentry
)
850 ret
= ocfs2_dentry_lock(dentry
, 1);
854 ocfs2_dentry_unlock(dentry
, 1);
859 static inline int ocfs2_inode_is_unlinkable(struct inode
*inode
)
861 if (S_ISDIR(inode
->i_mode
)) {
862 if (inode
->i_nlink
== 2)
867 if (inode
->i_nlink
== 1)
872 static int ocfs2_unlink(struct inode
*dir
,
873 struct dentry
*dentry
)
876 int child_locked
= 0;
877 bool is_unlinkable
= false;
878 struct inode
*inode
= dentry
->d_inode
;
879 struct inode
*orphan_dir
= NULL
;
880 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
882 struct ocfs2_dinode
*fe
= NULL
;
883 struct buffer_head
*fe_bh
= NULL
;
884 struct buffer_head
*parent_node_bh
= NULL
;
885 handle_t
*handle
= NULL
;
886 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
887 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
888 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
890 trace_ocfs2_unlink(dir
, dentry
, dentry
->d_name
.len
,
892 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
893 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
895 dquot_initialize(dir
);
897 BUG_ON(dentry
->d_parent
->d_inode
!= dir
);
899 if (inode
== osb
->root_inode
)
902 status
= ocfs2_inode_lock_nested(dir
, &parent_node_bh
, 1,
905 if (status
!= -ENOENT
)
910 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
911 dentry
->d_name
.len
, &blkno
, dir
,
914 if (status
!= -ENOENT
)
919 if (OCFS2_I(inode
)->ip_blkno
!= blkno
) {
922 trace_ocfs2_unlink_noent(
923 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
924 (unsigned long long)blkno
,
925 OCFS2_I(inode
)->ip_flags
);
929 status
= ocfs2_inode_lock(inode
, &fe_bh
, 1);
931 if (status
!= -ENOENT
)
937 if (S_ISDIR(inode
->i_mode
)) {
938 if (inode
->i_nlink
!= 2 || !ocfs2_empty_dir(inode
)) {
944 status
= ocfs2_remote_dentry_delete(dentry
);
946 /* This remote delete should succeed under all normal
952 if (ocfs2_inode_is_unlinkable(inode
)) {
953 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
954 OCFS2_I(inode
)->ip_blkno
,
955 orphan_name
, &orphan_insert
);
960 is_unlinkable
= true;
963 handle
= ocfs2_start_trans(osb
, ocfs2_unlink_credits(osb
->sb
));
964 if (IS_ERR(handle
)) {
965 status
= PTR_ERR(handle
);
971 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), fe_bh
,
972 OCFS2_JOURNAL_ACCESS_WRITE
);
978 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
980 /* delete the name from the parent dir */
981 status
= ocfs2_delete_entry(handle
, dir
, &lookup
);
987 if (S_ISDIR(inode
->i_mode
))
990 ocfs2_set_links_count(fe
, inode
->i_nlink
);
991 ocfs2_journal_dirty(handle
, fe_bh
);
993 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
994 if (S_ISDIR(inode
->i_mode
))
997 status
= ocfs2_mark_inode_dirty(handle
, dir
, parent_node_bh
);
1000 if (S_ISDIR(inode
->i_mode
))
1005 if (is_unlinkable
) {
1006 status
= ocfs2_orphan_add(osb
, handle
, inode
, fe_bh
,
1007 orphan_name
, &orphan_insert
, orphan_dir
);
1014 ocfs2_commit_trans(osb
, handle
);
1017 ocfs2_inode_unlock(inode
, 1);
1019 ocfs2_inode_unlock(dir
, 1);
1022 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1023 ocfs2_inode_unlock(orphan_dir
, 1);
1024 mutex_unlock(&orphan_dir
->i_mutex
);
1029 brelse(parent_node_bh
);
1031 ocfs2_free_dir_lookup_result(&orphan_insert
);
1032 ocfs2_free_dir_lookup_result(&lookup
);
1034 if (status
&& (status
!= -ENOTEMPTY
) && (status
!= -ENOENT
))
1040 static int ocfs2_check_if_ancestor(struct ocfs2_super
*osb
,
1041 u64 src_inode_no
, u64 dest_inode_no
)
1044 u64 parent_inode_no
= 0;
1045 u64 child_inode_no
= src_inode_no
;
1046 struct inode
*child_inode
;
1048 #define MAX_LOOKUP_TIMES 32
1050 child_inode
= ocfs2_iget(osb
, child_inode_no
, 0, 0);
1051 if (IS_ERR(child_inode
)) {
1052 ret
= PTR_ERR(child_inode
);
1056 ret
= ocfs2_inode_lock(child_inode
, NULL
, 0);
1064 ret
= ocfs2_lookup_ino_from_name(child_inode
, "..", 2,
1066 ocfs2_inode_unlock(child_inode
, 0);
1073 if (parent_inode_no
== dest_inode_no
) {
1078 if (parent_inode_no
== osb
->root_inode
->i_ino
) {
1083 child_inode_no
= parent_inode_no
;
1085 if (++i
>= MAX_LOOKUP_TIMES
) {
1086 mlog(ML_NOTICE
, "max lookup times reached, filesystem "
1087 "may have nested directories, "
1088 "src inode: %llu, dest inode: %llu.\n",
1089 (unsigned long long)src_inode_no
,
1090 (unsigned long long)dest_inode_no
);
1100 * The only place this should be used is rename and link!
1101 * if they have the same id, then the 1st one is the only one locked.
1103 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
1104 struct buffer_head
**bh1
,
1105 struct inode
*inode1
,
1106 struct buffer_head
**bh2
,
1107 struct inode
*inode2
,
1111 int inode1_is_ancestor
, inode2_is_ancestor
;
1112 struct ocfs2_inode_info
*oi1
= OCFS2_I(inode1
);
1113 struct ocfs2_inode_info
*oi2
= OCFS2_I(inode2
);
1114 struct buffer_head
**tmpbh
;
1115 struct inode
*tmpinode
;
1117 trace_ocfs2_double_lock((unsigned long long)oi1
->ip_blkno
,
1118 (unsigned long long)oi2
->ip_blkno
);
1125 /* we always want to lock the one with the lower lockid first.
1126 * and if they are nested, we lock ancestor first */
1127 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
1128 inode1_is_ancestor
= ocfs2_check_if_ancestor(osb
, oi2
->ip_blkno
,
1130 if (inode1_is_ancestor
< 0) {
1131 status
= inode1_is_ancestor
;
1135 inode2_is_ancestor
= ocfs2_check_if_ancestor(osb
, oi1
->ip_blkno
,
1137 if (inode2_is_ancestor
< 0) {
1138 status
= inode2_is_ancestor
;
1142 if ((inode1_is_ancestor
== 1) ||
1143 (oi1
->ip_blkno
< oi2
->ip_blkno
&&
1144 inode2_is_ancestor
== 0)) {
1145 /* switch id1 and id2 around */
1155 status
= ocfs2_inode_lock_nested(inode2
, bh2
, 1,
1156 rename
== 1 ? OI_LS_RENAME1
: OI_LS_PARENT
);
1158 if (status
!= -ENOENT
)
1165 status
= ocfs2_inode_lock_nested(inode1
, bh1
, 1,
1166 rename
== 1 ? OI_LS_RENAME2
: OI_LS_PARENT
);
1169 * An error return must mean that no cluster locks
1170 * were held on function exit.
1172 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
1173 ocfs2_inode_unlock(inode2
, 1);
1178 if (status
!= -ENOENT
)
1182 trace_ocfs2_double_lock_end(
1183 (unsigned long long)OCFS2_I(inode1
)->ip_blkno
,
1184 (unsigned long long)OCFS2_I(inode2
)->ip_blkno
);
1192 static void ocfs2_double_unlock(struct inode
*inode1
, struct inode
*inode2
)
1194 ocfs2_inode_unlock(inode1
, 1);
1196 if (inode1
!= inode2
)
1197 ocfs2_inode_unlock(inode2
, 1);
1200 static int ocfs2_rename(struct inode
*old_dir
,
1201 struct dentry
*old_dentry
,
1202 struct inode
*new_dir
,
1203 struct dentry
*new_dentry
)
1205 int status
= 0, rename_lock
= 0, parents_locked
= 0, target_exists
= 0;
1206 int old_child_locked
= 0, new_child_locked
= 0, update_dot_dot
= 0;
1207 struct inode
*old_inode
= old_dentry
->d_inode
;
1208 struct inode
*new_inode
= new_dentry
->d_inode
;
1209 struct inode
*orphan_dir
= NULL
;
1210 struct ocfs2_dinode
*newfe
= NULL
;
1211 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1212 struct buffer_head
*newfe_bh
= NULL
;
1213 struct buffer_head
*old_inode_bh
= NULL
;
1214 struct ocfs2_super
*osb
= NULL
;
1215 u64 newfe_blkno
, old_de_ino
;
1216 handle_t
*handle
= NULL
;
1217 struct buffer_head
*old_dir_bh
= NULL
;
1218 struct buffer_head
*new_dir_bh
= NULL
;
1219 u32 old_dir_nlink
= old_dir
->i_nlink
;
1220 struct ocfs2_dinode
*old_di
;
1221 struct ocfs2_dir_lookup_result old_inode_dot_dot_res
= { NULL
, };
1222 struct ocfs2_dir_lookup_result target_lookup_res
= { NULL
, };
1223 struct ocfs2_dir_lookup_result old_entry_lookup
= { NULL
, };
1224 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
1225 struct ocfs2_dir_lookup_result target_insert
= { NULL
, };
1226 bool should_add_orphan
= false;
1228 /* At some point it might be nice to break this function up a
1231 trace_ocfs2_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
1232 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
1233 new_dentry
->d_name
.len
, new_dentry
->d_name
.name
);
1235 dquot_initialize(old_dir
);
1236 dquot_initialize(new_dir
);
1238 osb
= OCFS2_SB(old_dir
->i_sb
);
1241 if (!igrab(new_inode
))
1245 /* Assume a directory hierarchy thusly:
1248 * a,b,c, and d are all directories.
1250 * from cwd of 'a' on both nodes:
1254 * And that's why, just like the VFS, we need a file system
1256 if (old_dir
!= new_dir
&& S_ISDIR(old_inode
->i_mode
)) {
1257 status
= ocfs2_rename_lock(osb
);
1264 /* here we cannot guarantee the inodes haven't just been
1265 * changed, so check if they are nested again */
1266 status
= ocfs2_check_if_ancestor(osb
, new_dir
->i_ino
,
1271 } else if (status
== 1) {
1273 trace_ocfs2_rename_not_permitted(
1274 (unsigned long long)old_inode
->i_ino
,
1275 (unsigned long long)new_dir
->i_ino
);
1280 /* if old and new are the same, this'll just do one lock. */
1281 status
= ocfs2_double_lock(osb
, &old_dir_bh
, old_dir
,
1282 &new_dir_bh
, new_dir
, 1);
1289 /* make sure both dirs have bhs
1290 * get an extra ref on old_dir_bh if old==new */
1293 new_dir_bh
= old_dir_bh
;
1296 mlog(ML_ERROR
, "no old_dir_bh!\n");
1303 * Aside from allowing a meta data update, the locking here
1304 * also ensures that the downconvert thread on other nodes
1305 * won't have to concurrently downconvert the inode and the
1308 status
= ocfs2_inode_lock_nested(old_inode
, &old_inode_bh
, 1,
1311 if (status
!= -ENOENT
)
1315 old_child_locked
= 1;
1317 status
= ocfs2_remote_dentry_delete(old_dentry
);
1323 if (S_ISDIR(old_inode
->i_mode
)) {
1324 u64 old_inode_parent
;
1327 status
= ocfs2_find_files_on_disk("..", 2, &old_inode_parent
,
1329 &old_inode_dot_dot_res
);
1335 if (old_inode_parent
!= OCFS2_I(old_dir
)->ip_blkno
) {
1340 if (!new_inode
&& new_dir
!= old_dir
&&
1341 new_dir
->i_nlink
>= ocfs2_link_max(osb
)) {
1347 status
= ocfs2_lookup_ino_from_name(old_dir
, old_dentry
->d_name
.name
,
1348 old_dentry
->d_name
.len
,
1356 * Check for inode number is _not_ due to possible IO errors.
1357 * We might rmdir the source, keep it as pwd of some process
1358 * and merrily kill the link to whatever was created under the
1359 * same name. Goodbye sticky bit ;-<
1361 if (old_de_ino
!= OCFS2_I(old_inode
)->ip_blkno
) {
1366 /* check if the target already exists (in which case we need
1368 status
= ocfs2_find_files_on_disk(new_dentry
->d_name
.name
,
1369 new_dentry
->d_name
.len
,
1370 &newfe_blkno
, new_dir
,
1371 &target_lookup_res
);
1372 /* The only error we allow here is -ENOENT because the new
1373 * file not existing is perfectly valid. */
1374 if ((status
< 0) && (status
!= -ENOENT
)) {
1375 /* If we cannot find the file specified we should just */
1376 /* return the error... */
1383 if (!target_exists
&& new_inode
) {
1385 * Target was unlinked by another node while we were
1386 * waiting to get to ocfs2_rename(). There isn't
1387 * anything we can do here to help the situation, so
1388 * bubble up the appropriate error.
1394 /* In case we need to overwrite an existing file, we blow it
1396 if (target_exists
) {
1397 /* VFS didn't think there existed an inode here, but
1398 * someone else in the cluster must have raced our
1399 * rename to create one. Today we error cleanly, in
1400 * the future we should consider calling iget to build
1401 * a new struct inode for this entry. */
1405 trace_ocfs2_rename_target_exists(new_dentry
->d_name
.len
,
1406 new_dentry
->d_name
.name
);
1410 if (OCFS2_I(new_inode
)->ip_blkno
!= newfe_blkno
) {
1413 trace_ocfs2_rename_disagree(
1414 (unsigned long long)OCFS2_I(new_inode
)->ip_blkno
,
1415 (unsigned long long)newfe_blkno
,
1416 OCFS2_I(new_inode
)->ip_flags
);
1420 status
= ocfs2_inode_lock(new_inode
, &newfe_bh
, 1);
1422 if (status
!= -ENOENT
)
1426 new_child_locked
= 1;
1428 status
= ocfs2_remote_dentry_delete(new_dentry
);
1434 newfe
= (struct ocfs2_dinode
*) newfe_bh
->b_data
;
1436 trace_ocfs2_rename_over_existing(
1437 (unsigned long long)newfe_blkno
, newfe_bh
, newfe_bh
?
1438 (unsigned long long)newfe_bh
->b_blocknr
: 0ULL);
1440 if (S_ISDIR(new_inode
->i_mode
) || (new_inode
->i_nlink
== 1)) {
1441 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
1442 OCFS2_I(new_inode
)->ip_blkno
,
1443 orphan_name
, &orphan_insert
);
1448 should_add_orphan
= true;
1451 BUG_ON(new_dentry
->d_parent
->d_inode
!= new_dir
);
1453 status
= ocfs2_check_dir_for_entry(new_dir
,
1454 new_dentry
->d_name
.name
,
1455 new_dentry
->d_name
.len
);
1459 status
= ocfs2_prepare_dir_for_insert(osb
, new_dir
, new_dir_bh
,
1460 new_dentry
->d_name
.name
,
1461 new_dentry
->d_name
.len
,
1469 handle
= ocfs2_start_trans(osb
, ocfs2_rename_credits(osb
->sb
));
1470 if (IS_ERR(handle
)) {
1471 status
= PTR_ERR(handle
);
1477 if (target_exists
) {
1478 if (S_ISDIR(new_inode
->i_mode
)) {
1479 if (new_inode
->i_nlink
!= 2 ||
1480 !ocfs2_empty_dir(new_inode
)) {
1481 status
= -ENOTEMPTY
;
1485 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(new_inode
),
1487 OCFS2_JOURNAL_ACCESS_WRITE
);
1493 /* change the dirent to point to the correct inode */
1494 status
= ocfs2_update_entry(new_dir
, handle
, &target_lookup_res
,
1500 new_dir
->i_version
++;
1502 if (S_ISDIR(new_inode
->i_mode
))
1503 ocfs2_set_links_count(newfe
, 0);
1505 ocfs2_add_links_count(newfe
, -1);
1506 ocfs2_journal_dirty(handle
, newfe_bh
);
1507 if (should_add_orphan
) {
1508 status
= ocfs2_orphan_add(osb
, handle
, new_inode
,
1509 newfe_bh
, orphan_name
,
1510 &orphan_insert
, orphan_dir
);
1517 /* if the name was not found in new_dir, add it now */
1518 status
= ocfs2_add_entry(handle
, new_dentry
, old_inode
,
1519 OCFS2_I(old_inode
)->ip_blkno
,
1520 new_dir_bh
, &target_insert
);
1523 old_inode
->i_ctime
= CURRENT_TIME
;
1524 mark_inode_dirty(old_inode
);
1526 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(old_inode
),
1528 OCFS2_JOURNAL_ACCESS_WRITE
);
1530 old_di
= (struct ocfs2_dinode
*) old_inode_bh
->b_data
;
1532 old_di
->i_ctime
= cpu_to_le64(old_inode
->i_ctime
.tv_sec
);
1533 old_di
->i_ctime_nsec
= cpu_to_le32(old_inode
->i_ctime
.tv_nsec
);
1534 ocfs2_journal_dirty(handle
, old_inode_bh
);
1539 * Now that the name has been added to new_dir, remove the old name.
1541 * We don't keep any directory entry context around until now
1542 * because the insert might have changed the type of directory
1543 * we're dealing with.
1545 status
= ocfs2_find_entry(old_dentry
->d_name
.name
,
1546 old_dentry
->d_name
.len
, old_dir
,
1551 status
= ocfs2_delete_entry(handle
, old_dir
, &old_entry_lookup
);
1558 drop_nlink(new_inode
);
1559 new_inode
->i_ctime
= CURRENT_TIME
;
1561 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1563 if (update_dot_dot
) {
1564 status
= ocfs2_update_entry(old_inode
, handle
,
1565 &old_inode_dot_dot_res
, new_dir
);
1566 drop_nlink(old_dir
);
1568 drop_nlink(new_inode
);
1571 mark_inode_dirty(new_dir
);
1574 mark_inode_dirty(old_dir
);
1575 ocfs2_mark_inode_dirty(handle
, old_dir
, old_dir_bh
);
1577 mark_inode_dirty(new_inode
);
1578 ocfs2_mark_inode_dirty(handle
, new_inode
, newfe_bh
);
1581 if (old_dir
!= new_dir
) {
1582 /* Keep the same times on both directories.*/
1583 new_dir
->i_ctime
= new_dir
->i_mtime
= old_dir
->i_ctime
;
1586 * This will also pick up the i_nlink change from the
1589 ocfs2_mark_inode_dirty(handle
, new_dir
, new_dir_bh
);
1592 if (old_dir_nlink
!= old_dir
->i_nlink
) {
1594 mlog(ML_ERROR
, "need to change nlink for old dir "
1595 "%llu from %d to %d but bh is NULL!\n",
1596 (unsigned long long)OCFS2_I(old_dir
)->ip_blkno
,
1597 (int)old_dir_nlink
, old_dir
->i_nlink
);
1599 struct ocfs2_dinode
*fe
;
1600 status
= ocfs2_journal_access_di(handle
,
1601 INODE_CACHE(old_dir
),
1603 OCFS2_JOURNAL_ACCESS_WRITE
);
1604 fe
= (struct ocfs2_dinode
*) old_dir_bh
->b_data
;
1605 ocfs2_set_links_count(fe
, old_dir
->i_nlink
);
1606 ocfs2_journal_dirty(handle
, old_dir_bh
);
1609 ocfs2_dentry_move(old_dentry
, new_dentry
, old_dir
, new_dir
);
1613 ocfs2_rename_unlock(osb
);
1616 ocfs2_commit_trans(osb
, handle
);
1619 ocfs2_double_unlock(old_dir
, new_dir
);
1621 if (old_child_locked
)
1622 ocfs2_inode_unlock(old_inode
, 1);
1624 if (new_child_locked
)
1625 ocfs2_inode_unlock(new_inode
, 1);
1628 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1629 ocfs2_inode_unlock(orphan_dir
, 1);
1630 mutex_unlock(&orphan_dir
->i_mutex
);
1635 sync_mapping_buffers(old_inode
->i_mapping
);
1640 ocfs2_free_dir_lookup_result(&target_lookup_res
);
1641 ocfs2_free_dir_lookup_result(&old_entry_lookup
);
1642 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res
);
1643 ocfs2_free_dir_lookup_result(&orphan_insert
);
1644 ocfs2_free_dir_lookup_result(&target_insert
);
1647 brelse(old_inode_bh
);
1658 * we expect i_size = strlen(symname). Copy symname into the file
1659 * data, including the null terminator.
1661 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
1663 struct inode
*inode
,
1664 const char *symname
)
1666 struct buffer_head
**bhs
= NULL
;
1668 struct super_block
*sb
= osb
->sb
;
1669 u64 p_blkno
, p_blocks
;
1670 int virtual, blocks
, status
, i
, bytes_left
;
1672 bytes_left
= i_size_read(inode
) + 1;
1673 /* we can't trust i_blocks because we're actually going to
1674 * write i_size + 1 bytes. */
1675 blocks
= (bytes_left
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
1677 trace_ocfs2_create_symlink_data((unsigned long long)inode
->i_blocks
,
1678 i_size_read(inode
), blocks
);
1680 /* Sanity check -- make sure we're going to fit. */
1682 ocfs2_clusters_to_bytes(sb
, OCFS2_I(inode
)->ip_clusters
)) {
1688 bhs
= kcalloc(blocks
, sizeof(struct buffer_head
*), GFP_KERNEL
);
1695 status
= ocfs2_extent_map_get_blocks(inode
, 0, &p_blkno
, &p_blocks
,
1702 /* links can never be larger than one cluster so we know this
1703 * is all going to be contiguous, but do a sanity check
1705 if ((p_blocks
<< sb
->s_blocksize_bits
) < bytes_left
) {
1712 while(bytes_left
> 0) {
1713 c
= &symname
[virtual * sb
->s_blocksize
];
1715 bhs
[virtual] = sb_getblk(sb
, p_blkno
);
1716 if (!bhs
[virtual]) {
1721 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode
),
1724 status
= ocfs2_journal_access(handle
, INODE_CACHE(inode
),
1726 OCFS2_JOURNAL_ACCESS_CREATE
);
1732 memset(bhs
[virtual]->b_data
, 0, sb
->s_blocksize
);
1734 memcpy(bhs
[virtual]->b_data
, c
,
1735 (bytes_left
> sb
->s_blocksize
) ? sb
->s_blocksize
:
1738 ocfs2_journal_dirty(handle
, bhs
[virtual]);
1742 bytes_left
-= sb
->s_blocksize
;
1749 for(i
= 0; i
< blocks
; i
++)
1759 static int ocfs2_symlink(struct inode
*dir
,
1760 struct dentry
*dentry
,
1761 const char *symname
)
1763 int status
, l
, credits
;
1765 struct ocfs2_super
*osb
= NULL
;
1766 struct inode
*inode
= NULL
;
1767 struct super_block
*sb
;
1768 struct buffer_head
*new_fe_bh
= NULL
;
1769 struct buffer_head
*parent_fe_bh
= NULL
;
1770 struct ocfs2_dinode
*fe
= NULL
;
1771 struct ocfs2_dinode
*dirfe
;
1772 handle_t
*handle
= NULL
;
1773 struct ocfs2_alloc_context
*inode_ac
= NULL
;
1774 struct ocfs2_alloc_context
*data_ac
= NULL
;
1775 struct ocfs2_alloc_context
*xattr_ac
= NULL
;
1776 int want_clusters
= 0;
1777 int xattr_credits
= 0;
1778 struct ocfs2_security_xattr_info si
= {
1781 int did_quota
= 0, did_quota_inode
= 0;
1782 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
1784 int did_block_signals
= 0;
1785 struct ocfs2_dentry_lock
*dl
= NULL
;
1787 trace_ocfs2_symlink_begin(dir
, dentry
, symname
,
1788 dentry
->d_name
.len
, dentry
->d_name
.name
);
1790 dquot_initialize(dir
);
1795 l
= strlen(symname
) + 1;
1797 credits
= ocfs2_calc_symlink_credits(sb
);
1799 /* lock the parent directory */
1800 status
= ocfs2_inode_lock(dir
, &parent_fe_bh
, 1);
1802 if (status
!= -ENOENT
)
1807 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
1808 if (!ocfs2_read_links_count(dirfe
)) {
1809 /* can't make a file in a deleted directory. */
1814 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
1815 dentry
->d_name
.len
);
1819 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
1820 dentry
->d_name
.name
,
1821 dentry
->d_name
.len
, &lookup
);
1827 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
1829 if (status
!= -ENOSPC
)
1834 inode
= ocfs2_get_init_inode(dir
, S_IFLNK
| S_IRWXUGO
);
1841 /* get security xattr */
1842 status
= ocfs2_init_security_get(inode
, dir
, &dentry
->d_name
, &si
);
1844 if (status
== -EOPNOTSUPP
)
1852 /* calculate meta data/clusters for setting security xattr */
1854 status
= ocfs2_calc_security_init(dir
, &si
, &want_clusters
,
1855 &xattr_credits
, &xattr_ac
);
1862 /* don't reserve bitmap space for fast symlinks. */
1863 if (l
> ocfs2_fast_symlink_chars(sb
))
1866 status
= ocfs2_reserve_clusters(osb
, want_clusters
, &data_ac
);
1868 if (status
!= -ENOSPC
)
1873 handle
= ocfs2_start_trans(osb
, credits
+ xattr_credits
);
1874 if (IS_ERR(handle
)) {
1875 status
= PTR_ERR(handle
);
1881 /* Starting to change things, restart is no longer possible. */
1882 ocfs2_block_signals(&oldset
);
1883 did_block_signals
= 1;
1885 status
= dquot_alloc_inode(inode
);
1888 did_quota_inode
= 1;
1890 trace_ocfs2_symlink_create(dir
, dentry
, dentry
->d_name
.len
,
1891 dentry
->d_name
.name
,
1892 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
1895 status
= ocfs2_mknod_locked(osb
, dir
, inode
,
1896 0, &new_fe_bh
, parent_fe_bh
, handle
,
1903 fe
= (struct ocfs2_dinode
*) new_fe_bh
->b_data
;
1906 inode
->i_op
= &ocfs2_symlink_inode_operations
;
1907 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1910 status
= dquot_alloc_space_nodirty(inode
,
1911 ocfs2_clusters_to_bytes(osb
->sb
, 1));
1915 inode
->i_mapping
->a_ops
= &ocfs2_aops
;
1916 status
= ocfs2_add_inode_data(osb
, inode
, &offset
, 1, 0,
1918 handle
, data_ac
, NULL
,
1921 if (status
!= -ENOSPC
&& status
!= -EINTR
) {
1923 "Failed to extend file to %llu\n",
1924 (unsigned long long)newsize
);
1930 i_size_write(inode
, newsize
);
1931 inode
->i_blocks
= ocfs2_inode_sector_count(inode
);
1933 inode
->i_mapping
->a_ops
= &ocfs2_fast_symlink_aops
;
1934 memcpy((char *) fe
->id2
.i_symlink
, symname
, l
);
1935 i_size_write(inode
, newsize
);
1936 inode
->i_blocks
= 0;
1939 status
= ocfs2_mark_inode_dirty(handle
, inode
, new_fe_bh
);
1945 if (!ocfs2_inode_is_fast_symlink(inode
)) {
1946 status
= ocfs2_create_symlink_data(osb
, handle
, inode
,
1955 status
= ocfs2_init_security_set(handle
, inode
, new_fe_bh
, &si
,
1964 * Do this before adding the entry to the directory. We add
1965 * also set d_op after success so that ->d_iput() will cleanup
1966 * the dentry lock even if ocfs2_add_entry() fails below.
1968 status
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
1974 dl
= dentry
->d_fsdata
;
1976 status
= ocfs2_add_entry(handle
, dentry
, inode
,
1977 le64_to_cpu(fe
->i_blkno
), parent_fe_bh
,
1984 insert_inode_hash(inode
);
1985 d_instantiate(dentry
, inode
);
1987 if (status
< 0 && did_quota
)
1988 dquot_free_space_nodirty(inode
,
1989 ocfs2_clusters_to_bytes(osb
->sb
, 1));
1990 if (status
< 0 && did_quota_inode
)
1991 dquot_free_inode(inode
);
1993 ocfs2_commit_trans(osb
, handle
);
1995 ocfs2_inode_unlock(dir
, 1);
1996 if (did_block_signals
)
1997 ocfs2_unblock_signals(&oldset
);
2000 brelse(parent_fe_bh
);
2002 ocfs2_free_dir_lookup_result(&lookup
);
2004 ocfs2_free_alloc_context(inode_ac
);
2006 ocfs2_free_alloc_context(data_ac
);
2008 ocfs2_free_alloc_context(xattr_ac
);
2009 if ((status
< 0) && inode
) {
2011 ocfs2_cleanup_add_entry_failure(osb
, dentry
, inode
);
2013 OCFS2_I(inode
)->ip_flags
|= OCFS2_INODE_SKIP_ORPHAN_DIR
;
2024 static int ocfs2_blkno_stringify(u64 blkno
, char *name
)
2026 int status
, namelen
;
2028 namelen
= snprintf(name
, OCFS2_ORPHAN_NAMELEN
+ 1, "%016llx",
2038 if (namelen
!= OCFS2_ORPHAN_NAMELEN
) {
2044 trace_ocfs2_blkno_stringify(blkno
, name
, namelen
);
2053 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super
*osb
,
2054 struct inode
**ret_orphan_dir
,
2055 struct buffer_head
**ret_orphan_dir_bh
)
2057 struct inode
*orphan_dir_inode
;
2058 struct buffer_head
*orphan_dir_bh
= NULL
;
2061 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2062 ORPHAN_DIR_SYSTEM_INODE
,
2064 if (!orphan_dir_inode
) {
2070 mutex_lock(&orphan_dir_inode
->i_mutex
);
2072 ret
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
2074 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2075 iput(orphan_dir_inode
);
2081 *ret_orphan_dir
= orphan_dir_inode
;
2082 *ret_orphan_dir_bh
= orphan_dir_bh
;
2087 static int __ocfs2_prepare_orphan_dir(struct inode
*orphan_dir_inode
,
2088 struct buffer_head
*orphan_dir_bh
,
2091 struct ocfs2_dir_lookup_result
*lookup
)
2094 struct ocfs2_super
*osb
= OCFS2_SB(orphan_dir_inode
->i_sb
);
2096 ret
= ocfs2_blkno_stringify(blkno
, name
);
2102 ret
= ocfs2_prepare_dir_for_insert(osb
, orphan_dir_inode
,
2103 orphan_dir_bh
, name
,
2104 OCFS2_ORPHAN_NAMELEN
, lookup
);
2114 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2115 * insertion of an orphan.
2116 * @osb: ocfs2 file system
2117 * @ret_orphan_dir: Orphan dir inode - returned locked!
2118 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2119 * @lookup: dir lookup result, to be passed back into functions like
2122 * Returns zero on success and the ret_orphan_dir, name and lookup
2123 * fields will be populated.
2125 * Returns non-zero on failure.
2127 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
2128 struct inode
**ret_orphan_dir
,
2131 struct ocfs2_dir_lookup_result
*lookup
)
2133 struct inode
*orphan_dir_inode
= NULL
;
2134 struct buffer_head
*orphan_dir_bh
= NULL
;
2137 ret
= ocfs2_lookup_lock_orphan_dir(osb
, &orphan_dir_inode
,
2144 ret
= __ocfs2_prepare_orphan_dir(orphan_dir_inode
, orphan_dir_bh
,
2145 blkno
, name
, lookup
);
2151 *ret_orphan_dir
= orphan_dir_inode
;
2154 brelse(orphan_dir_bh
);
2157 ocfs2_inode_unlock(orphan_dir_inode
, 1);
2158 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2159 iput(orphan_dir_inode
);
2167 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
2169 struct inode
*inode
,
2170 struct buffer_head
*fe_bh
,
2172 struct ocfs2_dir_lookup_result
*lookup
,
2173 struct inode
*orphan_dir_inode
)
2175 struct buffer_head
*orphan_dir_bh
= NULL
;
2177 struct ocfs2_dinode
*orphan_fe
;
2178 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
2180 trace_ocfs2_orphan_add_begin(
2181 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
2183 status
= ocfs2_read_inode_block(orphan_dir_inode
, &orphan_dir_bh
);
2189 status
= ocfs2_journal_access_di(handle
,
2190 INODE_CACHE(orphan_dir_inode
),
2192 OCFS2_JOURNAL_ACCESS_WRITE
);
2199 * We're going to journal the change of i_flags and i_orphaned_slot.
2200 * It's safe anyway, though some callers may duplicate the journaling.
2201 * Journaling within the func just make the logic look more
2204 status
= ocfs2_journal_access_di(handle
,
2207 OCFS2_JOURNAL_ACCESS_WRITE
);
2213 /* we're a cluster, and nlink can change on disk from
2214 * underneath us... */
2215 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2216 if (S_ISDIR(inode
->i_mode
))
2217 ocfs2_add_links_count(orphan_fe
, 1);
2218 set_nlink(orphan_dir_inode
, ocfs2_read_links_count(orphan_fe
));
2219 ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2221 status
= __ocfs2_add_entry(handle
, orphan_dir_inode
, name
,
2222 OCFS2_ORPHAN_NAMELEN
, inode
,
2223 OCFS2_I(inode
)->ip_blkno
,
2224 orphan_dir_bh
, lookup
);
2230 fe
->i_flags
|= cpu_to_le32(OCFS2_ORPHANED_FL
);
2231 OCFS2_I(inode
)->ip_flags
&= ~OCFS2_INODE_SKIP_ORPHAN_DIR
;
2233 /* Record which orphan dir our inode now resides
2234 * in. delete_inode will use this to determine which orphan
2236 fe
->i_orphaned_slot
= cpu_to_le16(osb
->slot_num
);
2238 ocfs2_journal_dirty(handle
, fe_bh
);
2240 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode
)->ip_blkno
,
2245 if (S_ISDIR(inode
->i_mode
))
2246 ocfs2_add_links_count(orphan_fe
, -1);
2247 set_nlink(orphan_dir_inode
, ocfs2_read_links_count(orphan_fe
));
2251 brelse(orphan_dir_bh
);
2256 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2257 int ocfs2_orphan_del(struct ocfs2_super
*osb
,
2259 struct inode
*orphan_dir_inode
,
2260 struct inode
*inode
,
2261 struct buffer_head
*orphan_dir_bh
)
2263 char name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2264 struct ocfs2_dinode
*orphan_fe
;
2266 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
2268 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2274 trace_ocfs2_orphan_del(
2275 (unsigned long long)OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2276 name
, OCFS2_ORPHAN_NAMELEN
);
2278 /* find it's spot in the orphan directory */
2279 status
= ocfs2_find_entry(name
, OCFS2_ORPHAN_NAMELEN
, orphan_dir_inode
,
2286 /* remove it from the orphan directory */
2287 status
= ocfs2_delete_entry(handle
, orphan_dir_inode
, &lookup
);
2293 status
= ocfs2_journal_access_di(handle
,
2294 INODE_CACHE(orphan_dir_inode
),
2296 OCFS2_JOURNAL_ACCESS_WRITE
);
2302 /* do the i_nlink dance! :) */
2303 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2304 if (S_ISDIR(inode
->i_mode
))
2305 ocfs2_add_links_count(orphan_fe
, -1);
2306 set_nlink(orphan_dir_inode
, ocfs2_read_links_count(orphan_fe
));
2307 ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2310 ocfs2_free_dir_lookup_result(&lookup
);
2318 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2319 * allocated file. This is different from the typical 'add to orphan dir'
2320 * operation in that the inode does not yet exist. This is a problem because
2321 * the orphan dir stringifies the inode block number to come up with it's
2322 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2323 * problem. This function works around it by calling deeper into the orphan
2324 * and suballoc code than other callers. Use this only by necessity.
2325 * @dir: The directory which this inode will ultimately wind up under - not the
2327 * @dir_bh: buffer_head the @dir inode block
2328 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2329 * with the string to be used for orphan dirent. Pass back to the orphan dir
2331 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2333 * @ret_di_blkno: block number where the new inode will be allocated.
2334 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2335 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2337 * Returns zero on success and the ret_orphan_dir, name and lookup
2338 * fields will be populated.
2340 * Returns non-zero on failure.
2342 static int ocfs2_prep_new_orphaned_file(struct inode
*dir
,
2343 struct buffer_head
*dir_bh
,
2345 struct inode
**ret_orphan_dir
,
2347 struct ocfs2_dir_lookup_result
*orphan_insert
,
2348 struct ocfs2_alloc_context
**ret_inode_ac
)
2352 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2353 struct inode
*orphan_dir
= NULL
;
2354 struct buffer_head
*orphan_dir_bh
= NULL
;
2355 struct ocfs2_alloc_context
*inode_ac
= NULL
;
2357 ret
= ocfs2_lookup_lock_orphan_dir(osb
, &orphan_dir
, &orphan_dir_bh
);
2363 /* reserve an inode spot */
2364 ret
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
2371 ret
= ocfs2_find_new_inode_loc(dir
, dir_bh
, inode_ac
,
2378 ret
= __ocfs2_prepare_orphan_dir(orphan_dir
, orphan_dir_bh
,
2379 di_blkno
, orphan_name
, orphan_insert
);
2387 *ret_orphan_dir
= orphan_dir
;
2388 *ret_di_blkno
= di_blkno
;
2389 *ret_inode_ac
= inode_ac
;
2391 * orphan_name and orphan_insert are already up to
2392 * date via prepare_orphan_dir
2395 /* Unroll reserve_new_inode* */
2397 ocfs2_free_alloc_context(inode_ac
);
2399 /* Unroll orphan dir locking */
2400 mutex_unlock(&orphan_dir
->i_mutex
);
2401 ocfs2_inode_unlock(orphan_dir
, 1);
2405 brelse(orphan_dir_bh
);
2410 int ocfs2_create_inode_in_orphan(struct inode
*dir
,
2412 struct inode
**new_inode
)
2414 int status
, did_quota_inode
= 0;
2415 struct inode
*inode
= NULL
;
2416 struct inode
*orphan_dir
= NULL
;
2417 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2418 struct ocfs2_dinode
*di
= NULL
;
2419 handle_t
*handle
= NULL
;
2420 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2421 struct buffer_head
*parent_di_bh
= NULL
;
2422 struct buffer_head
*new_di_bh
= NULL
;
2423 struct ocfs2_alloc_context
*inode_ac
= NULL
;
2424 struct ocfs2_dir_lookup_result orphan_insert
= { NULL
, };
2425 u64
uninitialized_var(di_blkno
), suballoc_loc
;
2428 status
= ocfs2_inode_lock(dir
, &parent_di_bh
, 1);
2430 if (status
!= -ENOENT
)
2435 status
= ocfs2_prep_new_orphaned_file(dir
, parent_di_bh
,
2436 orphan_name
, &orphan_dir
,
2437 &di_blkno
, &orphan_insert
, &inode_ac
);
2439 if (status
!= -ENOSPC
)
2444 inode
= ocfs2_get_init_inode(dir
, mode
);
2451 handle
= ocfs2_start_trans(osb
, ocfs2_mknod_credits(osb
->sb
, 0, 0));
2452 if (IS_ERR(handle
)) {
2453 status
= PTR_ERR(handle
);
2459 status
= dquot_alloc_inode(inode
);
2462 did_quota_inode
= 1;
2464 status
= ocfs2_claim_new_inode_at_loc(handle
, dir
, inode_ac
,
2466 &suballoc_bit
, di_blkno
);
2473 /* do the real work now. */
2474 status
= __ocfs2_mknod_locked(dir
, inode
,
2475 0, &new_di_bh
, parent_di_bh
, handle
,
2476 inode_ac
, di_blkno
, suballoc_loc
,
2483 di
= (struct ocfs2_dinode
*)new_di_bh
->b_data
;
2484 status
= ocfs2_orphan_add(osb
, handle
, inode
, new_di_bh
, orphan_name
,
2485 &orphan_insert
, orphan_dir
);
2491 /* get open lock so that only nodes can't remove it from orphan dir. */
2492 status
= ocfs2_open_lock(inode
);
2496 insert_inode_hash(inode
);
2498 if (status
< 0 && did_quota_inode
)
2499 dquot_free_inode(inode
);
2501 ocfs2_commit_trans(osb
, handle
);
2504 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2505 ocfs2_inode_unlock(orphan_dir
, 1);
2506 mutex_unlock(&orphan_dir
->i_mutex
);
2510 if ((status
< 0) && inode
) {
2516 ocfs2_free_alloc_context(inode_ac
);
2523 ocfs2_free_dir_lookup_result(&orphan_insert
);
2525 ocfs2_inode_unlock(dir
, 1);
2526 brelse(parent_di_bh
);
2530 int ocfs2_mv_orphaned_inode_to_new(struct inode
*dir
,
2531 struct inode
*inode
,
2532 struct dentry
*dentry
)
2535 struct buffer_head
*parent_di_bh
= NULL
;
2536 handle_t
*handle
= NULL
;
2537 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
2538 struct ocfs2_dinode
*dir_di
, *di
;
2539 struct inode
*orphan_dir_inode
= NULL
;
2540 struct buffer_head
*orphan_dir_bh
= NULL
;
2541 struct buffer_head
*di_bh
= NULL
;
2542 struct ocfs2_dir_lookup_result lookup
= { NULL
, };
2544 trace_ocfs2_mv_orphaned_inode_to_new(dir
, dentry
,
2545 dentry
->d_name
.len
, dentry
->d_name
.name
,
2546 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
2547 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
2549 status
= ocfs2_inode_lock(dir
, &parent_di_bh
, 1);
2551 if (status
!= -ENOENT
)
2556 dir_di
= (struct ocfs2_dinode
*) parent_di_bh
->b_data
;
2557 if (!dir_di
->i_links_count
) {
2558 /* can't make a file in a deleted directory. */
2563 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
2564 dentry
->d_name
.len
);
2568 /* get a spot inside the dir. */
2569 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_di_bh
,
2570 dentry
->d_name
.name
,
2571 dentry
->d_name
.len
, &lookup
);
2577 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2578 ORPHAN_DIR_SYSTEM_INODE
,
2580 if (!orphan_dir_inode
) {
2586 mutex_lock(&orphan_dir_inode
->i_mutex
);
2588 status
= ocfs2_inode_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
2591 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2592 iput(orphan_dir_inode
);
2596 status
= ocfs2_read_inode_block(inode
, &di_bh
);
2602 handle
= ocfs2_start_trans(osb
, ocfs2_rename_credits(osb
->sb
));
2603 if (IS_ERR(handle
)) {
2604 status
= PTR_ERR(handle
);
2610 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
),
2611 di_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
2617 status
= ocfs2_orphan_del(osb
, handle
, orphan_dir_inode
, inode
,
2624 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
2625 di
->i_flags
&= ~cpu_to_le32(OCFS2_ORPHANED_FL
);
2626 di
->i_orphaned_slot
= 0;
2627 set_nlink(inode
, 1);
2628 ocfs2_set_links_count(di
, inode
->i_nlink
);
2629 ocfs2_update_inode_fsync_trans(handle
, inode
, 1);
2630 ocfs2_journal_dirty(handle
, di_bh
);
2632 status
= ocfs2_add_entry(handle
, dentry
, inode
,
2633 OCFS2_I(inode
)->ip_blkno
, parent_di_bh
,
2640 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
2641 OCFS2_I(dir
)->ip_blkno
);
2647 d_instantiate(dentry
, inode
);
2650 ocfs2_commit_trans(osb
, handle
);
2652 ocfs2_inode_unlock(orphan_dir_inode
, 1);
2653 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2654 iput(orphan_dir_inode
);
2657 ocfs2_inode_unlock(dir
, 1);
2660 brelse(parent_di_bh
);
2661 brelse(orphan_dir_bh
);
2663 ocfs2_free_dir_lookup_result(&lookup
);
2671 const struct inode_operations ocfs2_dir_iops
= {
2672 .create
= ocfs2_create
,
2673 .lookup
= ocfs2_lookup
,
2675 .unlink
= ocfs2_unlink
,
2676 .rmdir
= ocfs2_unlink
,
2677 .symlink
= ocfs2_symlink
,
2678 .mkdir
= ocfs2_mkdir
,
2679 .mknod
= ocfs2_mknod
,
2680 .rename
= ocfs2_rename
,
2681 .setattr
= ocfs2_setattr
,
2682 .getattr
= ocfs2_getattr
,
2683 .permission
= ocfs2_permission
,
2684 .setxattr
= generic_setxattr
,
2685 .getxattr
= generic_getxattr
,
2686 .listxattr
= ocfs2_listxattr
,
2687 .removexattr
= generic_removexattr
,
2688 .fiemap
= ocfs2_fiemap
,
2689 .get_acl
= ocfs2_iop_get_acl
,
2690 .set_acl
= ocfs2_iop_set_acl
,