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>
44 #define MLOG_MASK_PREFIX ML_NAMEI
45 #include <cluster/masklog.h>
53 #include "extent_map.h"
65 #include "buffer_head_io.h"
67 #define NAMEI_RA_CHUNKS 2
68 #define NAMEI_RA_BLOCKS 4
69 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
70 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
72 static int inline ocfs2_search_dirblock(struct buffer_head
*bh
,
74 const char *name
, int namelen
,
76 struct ocfs2_dir_entry
**res_dir
);
78 static int ocfs2_delete_entry(struct ocfs2_journal_handle
*handle
,
80 struct ocfs2_dir_entry
*de_del
,
81 struct buffer_head
*bh
);
83 static int __ocfs2_add_entry(struct ocfs2_journal_handle
*handle
,
85 const char *name
, int namelen
,
86 struct inode
*inode
, u64 blkno
,
87 struct buffer_head
*parent_fe_bh
,
88 struct buffer_head
*insert_bh
);
90 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
92 struct dentry
*dentry
, int mode
,
94 struct buffer_head
**new_fe_bh
,
95 struct buffer_head
*parent_fe_bh
,
96 struct ocfs2_journal_handle
*handle
,
97 struct inode
**ret_inode
,
98 struct ocfs2_alloc_context
*inode_ac
);
100 static int ocfs2_fill_new_dir(struct ocfs2_super
*osb
,
101 struct ocfs2_journal_handle
*handle
,
102 struct inode
*parent
,
104 struct buffer_head
*fe_bh
,
105 struct ocfs2_alloc_context
*data_ac
);
107 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
108 struct ocfs2_journal_handle
*handle
,
109 struct buffer_head
**bh1
,
110 struct inode
*inode1
,
111 struct buffer_head
**bh2
,
112 struct inode
*inode2
);
114 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
115 struct ocfs2_journal_handle
*handle
,
118 struct buffer_head
**de_bh
);
120 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
121 struct ocfs2_journal_handle
*handle
,
123 struct ocfs2_dinode
*fe
,
125 struct buffer_head
*de_bh
);
127 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
128 struct ocfs2_journal_handle
*handle
,
130 const char *symname
);
132 static inline int ocfs2_add_entry(struct ocfs2_journal_handle
*handle
,
133 struct dentry
*dentry
,
134 struct inode
*inode
, u64 blkno
,
135 struct buffer_head
*parent_fe_bh
,
136 struct buffer_head
*insert_bh
)
138 return __ocfs2_add_entry(handle
, dentry
->d_parent
->d_inode
,
139 dentry
->d_name
.name
, dentry
->d_name
.len
,
140 inode
, blkno
, parent_fe_bh
, insert_bh
);
143 /* An orphan dir name is an 8 byte value, printed as a hex string */
144 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
146 static struct dentry
*ocfs2_lookup(struct inode
*dir
, struct dentry
*dentry
,
147 struct nameidata
*nd
)
151 struct buffer_head
*dirent_bh
= NULL
;
152 struct inode
*inode
= NULL
;
154 struct ocfs2_dir_entry
*dirent
;
155 struct ocfs2_inode_info
*oi
;
157 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
158 dentry
->d_name
.len
, dentry
->d_name
.name
);
160 if (dentry
->d_name
.len
> OCFS2_MAX_FILENAME_LEN
) {
161 ret
= ERR_PTR(-ENAMETOOLONG
);
165 mlog(0, "find name %.*s in directory %llu\n", dentry
->d_name
.len
,
166 dentry
->d_name
.name
, (unsigned long long)OCFS2_I(dir
)->ip_blkno
);
168 status
= ocfs2_meta_lock(dir
, NULL
, NULL
, 0);
170 if (status
!= -ENOENT
)
172 ret
= ERR_PTR(status
);
176 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
177 dentry
->d_name
.len
, &blkno
,
178 dir
, &dirent_bh
, &dirent
);
182 inode
= ocfs2_iget(OCFS2_SB(dir
->i_sb
), blkno
, 0);
184 mlog(ML_ERROR
, "Unable to create inode %llu\n",
185 (unsigned long long)blkno
);
186 ret
= ERR_PTR(-EACCES
);
191 /* Clear any orphaned state... If we were able to look up the
192 * inode from a directory, it certainly can't be orphaned. We
193 * might have the bad state from a node which intended to
194 * orphan this inode but crashed before it could commit the
196 spin_lock(&oi
->ip_lock
);
197 oi
->ip_flags
&= ~OCFS2_INODE_MAYBE_ORPHANED
;
198 oi
->ip_orphaned_slot
= OCFS2_INVALID_SLOT
;
199 spin_unlock(&oi
->ip_lock
);
202 dentry
->d_op
= &ocfs2_dentry_ops
;
203 ret
= d_splice_alias(inode
, dentry
);
207 * If d_splice_alias() finds a DCACHE_DISCONNECTED
208 * dentry, it will d_move() it on top of ourse. The
209 * return value will indicate this however, so in
210 * those cases, we switch them around for the locking
213 * NOTE: This dentry already has ->d_op set from
214 * ocfs2_get_parent() and ocfs2_get_dentry()
219 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
220 OCFS2_I(dir
)->ip_blkno
);
223 ret
= ERR_PTR(status
);
229 /* Don't drop the cluster lock until *after* the d_add --
230 * unlink on another node will message us to remove that
231 * dentry under this lock so otherwise we can race this with
232 * the vote thread and have a stale dentry. */
233 ocfs2_meta_unlock(dir
, 0);
244 static int ocfs2_fill_new_dir(struct ocfs2_super
*osb
,
245 struct ocfs2_journal_handle
*handle
,
246 struct inode
*parent
,
248 struct buffer_head
*fe_bh
,
249 struct ocfs2_alloc_context
*data_ac
)
252 struct buffer_head
*new_bh
= NULL
;
253 struct ocfs2_dir_entry
*de
= NULL
;
257 status
= ocfs2_do_extend_dir(osb
->sb
, handle
, inode
, fe_bh
,
258 data_ac
, NULL
, &new_bh
);
264 ocfs2_set_new_buffer_uptodate(inode
, new_bh
);
266 status
= ocfs2_journal_access(handle
, inode
, new_bh
,
267 OCFS2_JOURNAL_ACCESS_CREATE
);
272 memset(new_bh
->b_data
, 0, osb
->sb
->s_blocksize
);
274 de
= (struct ocfs2_dir_entry
*) new_bh
->b_data
;
275 de
->inode
= cpu_to_le64(OCFS2_I(inode
)->ip_blkno
);
278 cpu_to_le16(OCFS2_DIR_REC_LEN(de
->name_len
));
279 strcpy(de
->name
, ".");
280 ocfs2_set_de_type(de
, S_IFDIR
);
281 de
= (struct ocfs2_dir_entry
*) ((char *)de
+ le16_to_cpu(de
->rec_len
));
282 de
->inode
= cpu_to_le64(OCFS2_I(parent
)->ip_blkno
);
283 de
->rec_len
= cpu_to_le16(inode
->i_sb
->s_blocksize
-
284 OCFS2_DIR_REC_LEN(1));
286 strcpy(de
->name
, "..");
287 ocfs2_set_de_type(de
, S_IFDIR
);
289 status
= ocfs2_journal_dirty(handle
, new_bh
);
295 i_size_write(inode
, inode
->i_sb
->s_blocksize
);
297 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(inode
->i_sb
->s_blocksize
);
298 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
313 static int ocfs2_mknod(struct inode
*dir
,
314 struct dentry
*dentry
,
319 struct buffer_head
*parent_fe_bh
= NULL
;
320 struct ocfs2_journal_handle
*handle
= NULL
;
321 struct ocfs2_super
*osb
;
322 struct ocfs2_dinode
*dirfe
;
323 struct buffer_head
*new_fe_bh
= NULL
;
324 struct buffer_head
*de_bh
= NULL
;
325 struct inode
*inode
= NULL
;
326 struct ocfs2_alloc_context
*inode_ac
= NULL
;
327 struct ocfs2_alloc_context
*data_ac
= NULL
;
329 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
, mode
,
330 (unsigned long)dev
, dentry
->d_name
.len
,
331 dentry
->d_name
.name
);
333 /* get our super block */
334 osb
= OCFS2_SB(dir
->i_sb
);
336 handle
= ocfs2_alloc_handle(osb
);
337 if (handle
== NULL
) {
343 status
= ocfs2_meta_lock(dir
, handle
, &parent_fe_bh
, 1);
345 if (status
!= -ENOENT
)
350 if (S_ISDIR(mode
) && (dir
->i_nlink
>= OCFS2_LINK_MAX
)) {
355 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
356 if (!dirfe
->i_links_count
) {
357 /* can't make a file in a deleted directory. */
362 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
367 /* get a spot inside the dir. */
368 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
370 dentry
->d_name
.len
, &de_bh
);
376 /* reserve an inode spot */
377 status
= ocfs2_reserve_new_inode(osb
, handle
, &inode_ac
);
379 if (status
!= -ENOSPC
)
384 /* are we making a directory? If so, reserve a cluster for his
387 status
= ocfs2_reserve_clusters(osb
, handle
, 1, &data_ac
);
389 if (status
!= -ENOSPC
)
395 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_MKNOD_CREDITS
);
396 if (IS_ERR(handle
)) {
397 status
= PTR_ERR(handle
);
403 /* do the real work now. */
404 status
= ocfs2_mknod_locked(osb
, dir
, dentry
, mode
, dev
,
405 &new_fe_bh
, parent_fe_bh
, handle
,
413 status
= ocfs2_fill_new_dir(osb
, handle
, dir
, inode
,
420 status
= ocfs2_journal_access(handle
, dir
, parent_fe_bh
,
421 OCFS2_JOURNAL_ACCESS_WRITE
);
426 le16_add_cpu(&dirfe
->i_links_count
, 1);
427 status
= ocfs2_journal_dirty(handle
, parent_fe_bh
);
435 status
= ocfs2_add_entry(handle
, dentry
, inode
,
436 OCFS2_I(inode
)->ip_blkno
, parent_fe_bh
,
443 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
444 OCFS2_I(dir
)->ip_blkno
);
450 insert_inode_hash(inode
);
451 dentry
->d_op
= &ocfs2_dentry_ops
;
452 d_instantiate(dentry
, inode
);
456 ocfs2_commit_trans(handle
);
458 if (status
== -ENOSPC
)
459 mlog(0, "Disk is full\n");
468 brelse(parent_fe_bh
);
470 if ((status
< 0) && inode
)
474 ocfs2_free_alloc_context(inode_ac
);
477 ocfs2_free_alloc_context(data_ac
);
484 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
486 struct dentry
*dentry
, int mode
,
488 struct buffer_head
**new_fe_bh
,
489 struct buffer_head
*parent_fe_bh
,
490 struct ocfs2_journal_handle
*handle
,
491 struct inode
**ret_inode
,
492 struct ocfs2_alloc_context
*inode_ac
)
495 struct ocfs2_dinode
*fe
= NULL
;
496 struct ocfs2_extent_list
*fel
;
499 struct inode
*inode
= NULL
;
501 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
, mode
,
502 (unsigned long)dev
, dentry
->d_name
.len
,
503 dentry
->d_name
.name
);
508 status
= ocfs2_claim_new_inode(osb
, handle
, inode_ac
, &suballoc_bit
,
515 inode
= new_inode(dir
->i_sb
);
517 status
= PTR_ERR(inode
);
518 mlog(ML_ERROR
, "new_inode failed!\n");
522 /* populate as many fields early on as possible - many of
523 * these are used by the support functions here and in
525 inode
->i_ino
= ino_from_blkno(osb
->sb
, fe_blkno
);
526 OCFS2_I(inode
)->ip_blkno
= fe_blkno
;
531 inode
->i_mode
= mode
;
532 spin_lock(&osb
->osb_lock
);
533 inode
->i_generation
= osb
->s_next_generation
++;
534 spin_unlock(&osb
->osb_lock
);
536 *new_fe_bh
= sb_getblk(osb
->sb
, fe_blkno
);
542 ocfs2_set_new_buffer_uptodate(inode
, *new_fe_bh
);
544 status
= ocfs2_journal_access(handle
, inode
, *new_fe_bh
,
545 OCFS2_JOURNAL_ACCESS_CREATE
);
551 fe
= (struct ocfs2_dinode
*) (*new_fe_bh
)->b_data
;
552 memset(fe
, 0, osb
->sb
->s_blocksize
);
554 fe
->i_generation
= cpu_to_le32(inode
->i_generation
);
555 fe
->i_fs_generation
= cpu_to_le32(osb
->fs_generation
);
556 fe
->i_blkno
= cpu_to_le64(fe_blkno
);
557 fe
->i_suballoc_bit
= cpu_to_le16(suballoc_bit
);
558 fe
->i_suballoc_slot
= cpu_to_le16(osb
->slot_num
);
559 fe
->i_uid
= cpu_to_le32(current
->fsuid
);
560 if (dir
->i_mode
& S_ISGID
) {
561 fe
->i_gid
= cpu_to_le32(dir
->i_gid
);
565 fe
->i_gid
= cpu_to_le32(current
->fsgid
);
566 fe
->i_mode
= cpu_to_le16(mode
);
567 if (S_ISCHR(mode
) || S_ISBLK(mode
))
568 fe
->id1
.dev1
.i_rdev
= cpu_to_le64(huge_encode_dev(dev
));
570 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
572 fe
->i_last_eb_blk
= 0;
573 strcpy(fe
->i_signature
, OCFS2_INODE_SIGNATURE
);
574 le32_add_cpu(&fe
->i_flags
, 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
);
581 fel
= &fe
->id2
.i_list
;
582 fel
->l_tree_depth
= 0;
583 fel
->l_next_free_rec
= 0;
584 fel
->l_count
= cpu_to_le16(ocfs2_extent_recs_per_inode(osb
->sb
));
586 status
= ocfs2_journal_dirty(handle
, *new_fe_bh
);
592 if (ocfs2_populate_inode(inode
, fe
, 1) < 0) {
593 mlog(ML_ERROR
, "populate inode failed! bh->b_blocknr=%llu, "
594 "i_blkno=%llu, i_ino=%lu\n",
595 (unsigned long long) (*new_fe_bh
)->b_blocknr
,
596 (unsigned long long)fe
->i_blkno
, inode
->i_ino
);
600 ocfs2_inode_set_new(osb
, inode
);
601 status
= ocfs2_create_new_inode_locks(inode
);
605 status
= 0; /* error in ocfs2_create_new_inode_locks is not
623 static int ocfs2_mkdir(struct inode
*dir
,
624 struct dentry
*dentry
,
629 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
630 dentry
->d_name
.len
, dentry
->d_name
.name
);
631 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
637 static int ocfs2_create(struct inode
*dir
,
638 struct dentry
*dentry
,
640 struct nameidata
*nd
)
644 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
645 dentry
->d_name
.len
, dentry
->d_name
.name
);
646 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
652 static int ocfs2_link(struct dentry
*old_dentry
,
654 struct dentry
*dentry
)
656 struct ocfs2_journal_handle
*handle
= NULL
;
657 struct inode
*inode
= old_dentry
->d_inode
;
659 struct buffer_head
*fe_bh
= NULL
;
660 struct buffer_head
*parent_fe_bh
= NULL
;
661 struct buffer_head
*de_bh
= NULL
;
662 struct ocfs2_dinode
*fe
= NULL
;
663 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
665 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode
->i_ino
,
666 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
667 dentry
->d_name
.len
, dentry
->d_name
.name
);
669 if (S_ISDIR(inode
->i_mode
)) {
674 handle
= ocfs2_alloc_handle(osb
);
675 if (handle
== NULL
) {
680 err
= ocfs2_meta_lock(dir
, handle
, &parent_fe_bh
, 1);
692 err
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
697 err
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
699 dentry
->d_name
.len
, &de_bh
);
705 err
= ocfs2_meta_lock(inode
, handle
, &fe_bh
, 1);
712 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
713 if (le16_to_cpu(fe
->i_links_count
) >= OCFS2_LINK_MAX
) {
718 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_LINK_CREDITS
);
719 if (IS_ERR(handle
)) {
720 err
= PTR_ERR(handle
);
726 err
= ocfs2_journal_access(handle
, inode
, fe_bh
,
727 OCFS2_JOURNAL_ACCESS_WRITE
);
734 inode
->i_ctime
= CURRENT_TIME
;
735 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
736 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
737 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
739 err
= ocfs2_journal_dirty(handle
, fe_bh
);
741 le16_add_cpu(&fe
->i_links_count
, -1);
747 err
= ocfs2_add_entry(handle
, dentry
, inode
,
748 OCFS2_I(inode
)->ip_blkno
,
749 parent_fe_bh
, de_bh
);
751 le16_add_cpu(&fe
->i_links_count
, -1);
757 err
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
763 atomic_inc(&inode
->i_count
);
764 dentry
->d_op
= &ocfs2_dentry_ops
;
765 d_instantiate(dentry
, inode
);
768 ocfs2_commit_trans(handle
);
774 brelse(parent_fe_bh
);
782 * Takes and drops an exclusive lock on the given dentry. This will
783 * force other nodes to drop it.
785 static int ocfs2_remote_dentry_delete(struct dentry
*dentry
)
789 ret
= ocfs2_dentry_lock(dentry
, 1);
793 ocfs2_dentry_unlock(dentry
, 1);
798 static int ocfs2_unlink(struct inode
*dir
,
799 struct dentry
*dentry
)
802 unsigned int saved_nlink
= 0;
803 struct inode
*inode
= dentry
->d_inode
;
804 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
806 struct ocfs2_dinode
*fe
= NULL
;
807 struct buffer_head
*fe_bh
= NULL
;
808 struct buffer_head
*parent_node_bh
= NULL
;
809 struct ocfs2_journal_handle
*handle
= NULL
;
810 struct ocfs2_dir_entry
*dirent
= NULL
;
811 struct buffer_head
*dirent_bh
= NULL
;
812 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
813 struct buffer_head
*orphan_entry_bh
= NULL
;
815 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
816 dentry
->d_name
.len
, dentry
->d_name
.name
);
818 BUG_ON(dentry
->d_parent
->d_inode
!= dir
);
820 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
822 if (inode
== osb
->root_inode
) {
823 mlog(0, "Cannot delete the root directory\n");
828 handle
= ocfs2_alloc_handle(osb
);
829 if (handle
== NULL
) {
835 status
= ocfs2_meta_lock(dir
, handle
, &parent_node_bh
, 1);
837 if (status
!= -ENOENT
)
842 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
843 dentry
->d_name
.len
, &blkno
,
844 dir
, &dirent_bh
, &dirent
);
846 if (status
!= -ENOENT
)
851 if (OCFS2_I(inode
)->ip_blkno
!= blkno
) {
854 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
855 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
856 (unsigned long long)blkno
, OCFS2_I(inode
)->ip_flags
);
860 status
= ocfs2_meta_lock(inode
, handle
, &fe_bh
, 1);
862 if (status
!= -ENOENT
)
867 if (S_ISDIR(inode
->i_mode
)) {
868 if (!ocfs2_empty_dir(inode
)) {
871 } else if (inode
->i_nlink
!= 2) {
877 /* There are still a few steps left until we can consider the
878 * unlink to have succeeded. Save off nlink here before
879 * modification so we can set it back in case we hit an issue
881 saved_nlink
= inode
->i_nlink
;
882 if (S_ISDIR(inode
->i_mode
))
887 status
= ocfs2_remote_dentry_delete(dentry
);
889 /* This vote should succeed under all normal
895 if (!inode
->i_nlink
) {
896 status
= ocfs2_prepare_orphan_dir(osb
, handle
, inode
,
905 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_UNLINK_CREDITS
);
906 if (IS_ERR(handle
)) {
907 status
= PTR_ERR(handle
);
913 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
914 OCFS2_JOURNAL_ACCESS_WRITE
);
920 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
922 if (!inode
->i_nlink
) {
923 status
= ocfs2_orphan_add(osb
, handle
, inode
, fe
, orphan_name
,
931 /* delete the name from the parent dir */
932 status
= ocfs2_delete_entry(handle
, dir
, dirent
, dirent_bh
);
938 /* We can set nlink on the dinode now. clear the saved version
939 * so that it doesn't get set later. */
940 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
943 status
= ocfs2_journal_dirty(handle
, fe_bh
);
949 if (S_ISDIR(inode
->i_mode
)) {
951 status
= ocfs2_mark_inode_dirty(handle
, dir
,
960 if (status
< 0 && saved_nlink
)
961 inode
->i_nlink
= saved_nlink
;
964 ocfs2_commit_trans(handle
);
973 brelse(parent_node_bh
);
976 brelse(orphan_entry_bh
);
984 * The only place this should be used is rename!
985 * if they have the same id, then the 1st one is the only one locked.
987 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
988 struct ocfs2_journal_handle
*handle
,
989 struct buffer_head
**bh1
,
990 struct inode
*inode1
,
991 struct buffer_head
**bh2
,
992 struct inode
*inode2
)
995 struct ocfs2_inode_info
*oi1
= OCFS2_I(inode1
);
996 struct ocfs2_inode_info
*oi2
= OCFS2_I(inode2
);
997 struct buffer_head
**tmpbh
;
998 struct inode
*tmpinode
;
1000 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
1001 (unsigned long long)oi1
->ip_blkno
,
1002 (unsigned long long)oi2
->ip_blkno
);
1011 /* we always want to lock the one with the lower lockid first. */
1012 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
1013 if (oi1
->ip_blkno
< oi2
->ip_blkno
) {
1014 /* switch id1 and id2 around */
1015 mlog(0, "switching them around...\n");
1025 status
= ocfs2_meta_lock(inode2
, handle
, bh2
, 1);
1027 if (status
!= -ENOENT
)
1033 status
= ocfs2_meta_lock(inode1
, handle
, bh1
, 1);
1035 if (status
!= -ENOENT
)
1044 #define PARENT_INO(buffer) \
1045 ((struct ocfs2_dir_entry *) \
1047 le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1049 static int ocfs2_rename(struct inode
*old_dir
,
1050 struct dentry
*old_dentry
,
1051 struct inode
*new_dir
,
1052 struct dentry
*new_dentry
)
1054 int status
= 0, rename_lock
= 0;
1055 struct inode
*old_inode
= old_dentry
->d_inode
;
1056 struct inode
*new_inode
= new_dentry
->d_inode
;
1057 struct ocfs2_dinode
*newfe
= NULL
;
1058 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1059 struct buffer_head
*orphan_entry_bh
= NULL
;
1060 struct buffer_head
*newfe_bh
= NULL
;
1061 struct buffer_head
*insert_entry_bh
= NULL
;
1062 struct ocfs2_super
*osb
= NULL
;
1064 struct ocfs2_journal_handle
*handle
= NULL
;
1065 struct buffer_head
*old_dir_bh
= NULL
;
1066 struct buffer_head
*new_dir_bh
= NULL
;
1067 struct ocfs2_dir_entry
*old_de
= NULL
, *new_de
= NULL
; // dirent for old_dentry
1069 struct buffer_head
*new_de_bh
= NULL
, *old_de_bh
= NULL
; // bhs for above
1070 struct buffer_head
*old_inode_de_bh
= NULL
; // if old_dentry is a dir,
1071 // this is the 1st dirent bh
1072 nlink_t old_dir_nlink
= old_dir
->i_nlink
, new_dir_nlink
= new_dir
->i_nlink
;
1074 /* At some point it might be nice to break this function up a
1077 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1078 old_dir
, old_dentry
, new_dir
, new_dentry
,
1079 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
1080 new_dentry
->d_name
.len
, new_dentry
->d_name
.name
);
1082 osb
= OCFS2_SB(old_dir
->i_sb
);
1085 if (!igrab(new_inode
))
1089 if (atomic_read(&old_dentry
->d_count
) > 2) {
1090 shrink_dcache_parent(old_dentry
);
1091 if (atomic_read(&old_dentry
->d_count
) > 2) {
1097 /* Assume a directory heirarchy thusly:
1100 * a,b,c, and d are all directories.
1102 * from cwd of 'a' on both nodes:
1106 * And that's why, just like the VFS, we need a file system
1108 if (old_dentry
!= new_dentry
) {
1109 status
= ocfs2_rename_lock(osb
);
1117 handle
= ocfs2_alloc_handle(osb
);
1118 if (handle
== NULL
) {
1124 /* if old and new are the same, this'll just do one lock. */
1125 status
= ocfs2_double_lock(osb
, handle
,
1126 &old_dir_bh
, old_dir
,
1127 &new_dir_bh
, new_dir
);
1133 /* make sure both dirs have bhs
1134 * get an extra ref on old_dir_bh if old==new */
1137 new_dir_bh
= old_dir_bh
;
1140 mlog(ML_ERROR
, "no old_dir_bh!\n");
1147 * Though we don't require an inode meta data update if
1148 * old_inode is not a directory, we lock anyway here to ensure
1149 * the vote thread on other nodes won't have to concurrently
1150 * downconvert the inode and the dentry locks.
1152 status
= ocfs2_meta_lock(old_inode
, handle
, NULL
, 1);
1154 if (status
!= -ENOENT
)
1159 status
= ocfs2_remote_dentry_delete(old_dentry
);
1165 if (S_ISDIR(old_inode
->i_mode
)) {
1167 old_inode_de_bh
= ocfs2_bread(old_inode
, 0, &status
, 0);
1168 if (!old_inode_de_bh
)
1172 if (le64_to_cpu(PARENT_INO(old_inode_de_bh
->b_data
)) !=
1173 OCFS2_I(old_dir
)->ip_blkno
)
1176 if (!new_inode
&& new_dir
!=old_dir
&&
1177 new_dir
->i_nlink
>= OCFS2_LINK_MAX
)
1182 old_de_bh
= ocfs2_find_entry(old_dentry
->d_name
.name
,
1183 old_dentry
->d_name
.len
,
1189 * Check for inode number is _not_ due to possible IO errors.
1190 * We might rmdir the source, keep it as pwd of some process
1191 * and merrily kill the link to whatever was created under the
1192 * same name. Goodbye sticky bit ;-<
1194 if (le64_to_cpu(old_de
->inode
) != OCFS2_I(old_inode
)->ip_blkno
)
1197 /* check if the target already exists (in which case we need
1199 status
= ocfs2_find_files_on_disk(new_dentry
->d_name
.name
,
1200 new_dentry
->d_name
.len
,
1201 &newfe_blkno
, new_dir
, &new_de_bh
,
1203 /* The only error we allow here is -ENOENT because the new
1204 * file not existing is perfectly valid. */
1205 if ((status
< 0) && (status
!= -ENOENT
)) {
1206 /* If we cannot find the file specified we should just */
1207 /* return the error... */
1212 if (!new_de
&& new_inode
)
1213 mlog(ML_ERROR
, "inode %lu does not exist in it's parent "
1214 "directory!", new_inode
->i_ino
);
1216 /* In case we need to overwrite an existing file, we blow it
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 mlog(0, "We found an inode for name %.*s but VFS "
1228 "didn't give us one.\n", new_dentry
->d_name
.len
,
1229 new_dentry
->d_name
.name
);
1233 if (OCFS2_I(new_inode
)->ip_blkno
!= newfe_blkno
) {
1236 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1237 (unsigned long long)OCFS2_I(new_inode
)->ip_blkno
,
1238 (unsigned long long)newfe_blkno
,
1239 OCFS2_I(new_inode
)->ip_flags
);
1243 status
= ocfs2_meta_lock(new_inode
, handle
, &newfe_bh
, 1);
1245 if (status
!= -ENOENT
)
1250 status
= ocfs2_remote_dentry_delete(new_dentry
);
1256 newfe
= (struct ocfs2_dinode
*) newfe_bh
->b_data
;
1258 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1259 "newfebh=%p bhblocknr=%llu\n", new_de
,
1260 (unsigned long long)newfe_blkno
, newfe_bh
, newfe_bh
?
1261 (unsigned long long)newfe_bh
->b_blocknr
: 0ULL);
1263 if (S_ISDIR(new_inode
->i_mode
) || (new_inode
->i_nlink
== 1)) {
1264 status
= ocfs2_prepare_orphan_dir(osb
, handle
,
1274 BUG_ON(new_dentry
->d_parent
->d_inode
!= new_dir
);
1276 status
= ocfs2_check_dir_for_entry(new_dir
,
1277 new_dentry
->d_name
.name
,
1278 new_dentry
->d_name
.len
);
1282 status
= ocfs2_prepare_dir_for_insert(osb
, new_dir
, new_dir_bh
,
1283 new_dentry
->d_name
.name
,
1284 new_dentry
->d_name
.len
,
1292 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_RENAME_CREDITS
);
1293 if (IS_ERR(handle
)) {
1294 status
= PTR_ERR(handle
);
1301 if (S_ISDIR(new_inode
->i_mode
)) {
1302 if (!ocfs2_empty_dir(new_inode
) ||
1303 new_inode
->i_nlink
!= 2) {
1304 status
= -ENOTEMPTY
;
1308 status
= ocfs2_journal_access(handle
, new_inode
, newfe_bh
,
1309 OCFS2_JOURNAL_ACCESS_WRITE
);
1315 if (S_ISDIR(new_inode
->i_mode
) ||
1316 (newfe
->i_links_count
== cpu_to_le16(1))){
1317 status
= ocfs2_orphan_add(osb
, handle
, new_inode
,
1326 /* change the dirent to point to the correct inode */
1327 status
= ocfs2_journal_access(handle
, new_dir
, new_de_bh
,
1328 OCFS2_JOURNAL_ACCESS_WRITE
);
1333 new_de
->inode
= cpu_to_le64(OCFS2_I(old_inode
)->ip_blkno
);
1334 new_de
->file_type
= old_de
->file_type
;
1335 new_dir
->i_version
++;
1336 status
= ocfs2_journal_dirty(handle
, new_de_bh
);
1342 if (S_ISDIR(new_inode
->i_mode
))
1343 newfe
->i_links_count
= 0;
1345 le16_add_cpu(&newfe
->i_links_count
, -1);
1347 status
= ocfs2_journal_dirty(handle
, newfe_bh
);
1353 /* if the name was not found in new_dir, add it now */
1354 status
= ocfs2_add_entry(handle
, new_dentry
, old_inode
,
1355 OCFS2_I(old_inode
)->ip_blkno
,
1356 new_dir_bh
, insert_entry_bh
);
1359 old_inode
->i_ctime
= CURRENT_TIME
;
1360 mark_inode_dirty(old_inode
);
1362 /* now that the name has been added to new_dir, remove the old name */
1363 status
= ocfs2_delete_entry(handle
, old_dir
, old_de
, old_de_bh
);
1370 new_inode
->i_nlink
--;
1371 new_inode
->i_ctime
= CURRENT_TIME
;
1373 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1374 if (old_inode_de_bh
) {
1375 status
= ocfs2_journal_access(handle
, old_inode
,
1377 OCFS2_JOURNAL_ACCESS_WRITE
);
1378 PARENT_INO(old_inode_de_bh
->b_data
) =
1379 cpu_to_le64(OCFS2_I(new_dir
)->ip_blkno
);
1380 status
= ocfs2_journal_dirty(handle
, old_inode_de_bh
);
1383 new_inode
->i_nlink
--;
1386 mark_inode_dirty(new_dir
);
1389 mark_inode_dirty(old_dir
);
1391 mark_inode_dirty(new_inode
);
1393 if (old_dir
!= new_dir
)
1394 if (new_dir_nlink
!= new_dir
->i_nlink
) {
1396 mlog(ML_ERROR
, "need to change nlink for new "
1397 "dir %llu from %d to %d but bh is NULL\n",
1398 (unsigned long long)OCFS2_I(new_dir
)->ip_blkno
,
1399 (int)new_dir_nlink
, new_dir
->i_nlink
);
1401 struct ocfs2_dinode
*fe
;
1402 status
= ocfs2_journal_access(handle
,
1405 OCFS2_JOURNAL_ACCESS_WRITE
);
1406 fe
= (struct ocfs2_dinode
*) new_dir_bh
->b_data
;
1407 fe
->i_links_count
= cpu_to_le16(new_dir
->i_nlink
);
1408 status
= ocfs2_journal_dirty(handle
, new_dir_bh
);
1412 if (old_dir_nlink
!= old_dir
->i_nlink
) {
1414 mlog(ML_ERROR
, "need to change nlink for old dir "
1415 "%llu from %d to %d but bh is NULL!\n",
1416 (unsigned long long)OCFS2_I(old_dir
)->ip_blkno
,
1417 (int)old_dir_nlink
, old_dir
->i_nlink
);
1419 struct ocfs2_dinode
*fe
;
1420 status
= ocfs2_journal_access(handle
, old_dir
,
1422 OCFS2_JOURNAL_ACCESS_WRITE
);
1423 fe
= (struct ocfs2_dinode
*) old_dir_bh
->b_data
;
1424 fe
->i_links_count
= cpu_to_le16(old_dir
->i_nlink
);
1425 status
= ocfs2_journal_dirty(handle
, old_dir_bh
);
1429 ocfs2_dentry_move(old_dentry
, new_dentry
, old_dir
, new_dir
);
1433 ocfs2_rename_unlock(osb
);
1436 ocfs2_commit_trans(handle
);
1439 sync_mapping_buffers(old_inode
->i_mapping
);
1453 if (old_inode_de_bh
)
1454 brelse(old_inode_de_bh
);
1455 if (orphan_entry_bh
)
1456 brelse(orphan_entry_bh
);
1457 if (insert_entry_bh
)
1458 brelse(insert_entry_bh
);
1466 * we expect i_size = strlen(symname). Copy symname into the file
1467 * data, including the null terminator.
1469 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
1470 struct ocfs2_journal_handle
*handle
,
1471 struct inode
*inode
,
1472 const char *symname
)
1474 struct buffer_head
**bhs
= NULL
;
1476 struct super_block
*sb
= osb
->sb
;
1479 int virtual, blocks
, status
, i
, bytes_left
;
1481 bytes_left
= i_size_read(inode
) + 1;
1482 /* we can't trust i_blocks because we're actually going to
1483 * write i_size + 1 bytes. */
1484 blocks
= (bytes_left
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
1486 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1487 (unsigned long long)inode
->i_blocks
,
1488 i_size_read(inode
), blocks
);
1490 /* Sanity check -- make sure we're going to fit. */
1492 ocfs2_clusters_to_bytes(sb
, OCFS2_I(inode
)->ip_clusters
)) {
1498 bhs
= kcalloc(blocks
, sizeof(struct buffer_head
*), GFP_KERNEL
);
1505 status
= ocfs2_extent_map_get_blocks(inode
, 0, 1, &p_blkno
,
1512 /* links can never be larger than one cluster so we know this
1513 * is all going to be contiguous, but do a sanity check
1515 if ((p_blocks
<< sb
->s_blocksize_bits
) < bytes_left
) {
1522 while(bytes_left
> 0) {
1523 c
= &symname
[virtual * sb
->s_blocksize
];
1525 bhs
[virtual] = sb_getblk(sb
, p_blkno
);
1526 if (!bhs
[virtual]) {
1531 ocfs2_set_new_buffer_uptodate(inode
, bhs
[virtual]);
1533 status
= ocfs2_journal_access(handle
, inode
, bhs
[virtual],
1534 OCFS2_JOURNAL_ACCESS_CREATE
);
1540 memset(bhs
[virtual]->b_data
, 0, sb
->s_blocksize
);
1542 memcpy(bhs
[virtual]->b_data
, c
,
1543 (bytes_left
> sb
->s_blocksize
) ? sb
->s_blocksize
:
1546 status
= ocfs2_journal_dirty(handle
, bhs
[virtual]);
1554 bytes_left
-= sb
->s_blocksize
;
1561 for(i
= 0; i
< blocks
; i
++)
1571 static int ocfs2_symlink(struct inode
*dir
,
1572 struct dentry
*dentry
,
1573 const char *symname
)
1575 int status
, l
, credits
;
1577 struct ocfs2_super
*osb
= NULL
;
1578 struct inode
*inode
= NULL
;
1579 struct super_block
*sb
;
1580 struct buffer_head
*new_fe_bh
= NULL
;
1581 struct buffer_head
*de_bh
= NULL
;
1582 struct buffer_head
*parent_fe_bh
= NULL
;
1583 struct ocfs2_dinode
*fe
= NULL
;
1584 struct ocfs2_dinode
*dirfe
;
1585 struct ocfs2_journal_handle
*handle
= NULL
;
1586 struct ocfs2_alloc_context
*inode_ac
= NULL
;
1587 struct ocfs2_alloc_context
*data_ac
= NULL
;
1589 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir
,
1590 dentry
, symname
, dentry
->d_name
.len
, dentry
->d_name
.name
);
1595 l
= strlen(symname
) + 1;
1597 credits
= ocfs2_calc_symlink_credits(sb
);
1599 handle
= ocfs2_alloc_handle(osb
);
1600 if (handle
== NULL
) {
1606 /* lock the parent directory */
1607 status
= ocfs2_meta_lock(dir
, handle
, &parent_fe_bh
, 1);
1609 if (status
!= -ENOENT
)
1614 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
1615 if (!dirfe
->i_links_count
) {
1616 /* can't make a file in a deleted directory. */
1621 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
1622 dentry
->d_name
.len
);
1626 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
1627 dentry
->d_name
.name
,
1628 dentry
->d_name
.len
, &de_bh
);
1634 status
= ocfs2_reserve_new_inode(osb
, handle
, &inode_ac
);
1636 if (status
!= -ENOSPC
)
1641 /* don't reserve bitmap space for fast symlinks. */
1642 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1643 status
= ocfs2_reserve_clusters(osb
, handle
, 1, &data_ac
);
1645 if (status
!= -ENOSPC
)
1651 handle
= ocfs2_start_trans(osb
, handle
, credits
);
1652 if (IS_ERR(handle
)) {
1653 status
= PTR_ERR(handle
);
1659 status
= ocfs2_mknod_locked(osb
, dir
, dentry
,
1660 S_IFLNK
| S_IRWXUGO
, 0,
1661 &new_fe_bh
, parent_fe_bh
, handle
,
1668 fe
= (struct ocfs2_dinode
*) new_fe_bh
->b_data
;
1671 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1672 inode
->i_op
= &ocfs2_symlink_inode_operations
;
1673 status
= ocfs2_do_extend_allocation(osb
, inode
, 1, new_fe_bh
,
1674 handle
, data_ac
, NULL
,
1677 if (status
!= -ENOSPC
&& status
!= -EINTR
) {
1679 "Failed to extend file to %llu\n",
1680 (unsigned long long)newsize
);
1686 i_size_write(inode
, newsize
);
1687 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(newsize
);
1689 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
1690 memcpy((char *) fe
->id2
.i_symlink
, symname
, l
);
1691 i_size_write(inode
, newsize
);
1692 inode
->i_blocks
= 0;
1695 status
= ocfs2_mark_inode_dirty(handle
, inode
, new_fe_bh
);
1701 if (!ocfs2_inode_is_fast_symlink(inode
)) {
1702 status
= ocfs2_create_symlink_data(osb
, handle
, inode
,
1710 status
= ocfs2_add_entry(handle
, dentry
, inode
,
1711 le64_to_cpu(fe
->i_blkno
), parent_fe_bh
,
1718 status
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
1724 insert_inode_hash(inode
);
1725 dentry
->d_op
= &ocfs2_dentry_ops
;
1726 d_instantiate(dentry
, inode
);
1729 ocfs2_commit_trans(handle
);
1733 brelse(parent_fe_bh
);
1737 ocfs2_free_alloc_context(inode_ac
);
1739 ocfs2_free_alloc_context(data_ac
);
1740 if ((status
< 0) && inode
)
1748 int ocfs2_check_dir_entry(struct inode
* dir
,
1749 struct ocfs2_dir_entry
* de
,
1750 struct buffer_head
* bh
,
1751 unsigned long offset
)
1753 const char *error_msg
= NULL
;
1754 const int rlen
= le16_to_cpu(de
->rec_len
);
1756 if (rlen
< OCFS2_DIR_REC_LEN(1))
1757 error_msg
= "rec_len is smaller than minimal";
1758 else if (rlen
% 4 != 0)
1759 error_msg
= "rec_len % 4 != 0";
1760 else if (rlen
< OCFS2_DIR_REC_LEN(de
->name_len
))
1761 error_msg
= "rec_len is too small for name_len";
1762 else if (((char *) de
- bh
->b_data
) + rlen
> dir
->i_sb
->s_blocksize
)
1763 error_msg
= "directory entry across blocks";
1765 if (error_msg
!= NULL
)
1766 mlog(ML_ERROR
, "bad entry in directory #%llu: %s - "
1767 "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1768 (unsigned long long)OCFS2_I(dir
)->ip_blkno
, error_msg
,
1769 offset
, (unsigned long long)le64_to_cpu(de
->inode
), rlen
,
1771 return error_msg
== NULL
? 1 : 0;
1774 /* we don't always have a dentry for what we want to add, so people
1775 * like orphan dir can call this instead.
1777 * If you pass me insert_bh, I'll skip the search of the other dir
1778 * blocks and put the record in there.
1780 static int __ocfs2_add_entry(struct ocfs2_journal_handle
*handle
,
1782 const char *name
, int namelen
,
1783 struct inode
*inode
, u64 blkno
,
1784 struct buffer_head
*parent_fe_bh
,
1785 struct buffer_head
*insert_bh
)
1787 unsigned long offset
;
1788 unsigned short rec_len
;
1789 struct ocfs2_dir_entry
*de
, *de1
;
1790 struct super_block
*sb
;
1800 rec_len
= OCFS2_DIR_REC_LEN(namelen
);
1802 de
= (struct ocfs2_dir_entry
*) insert_bh
->b_data
;
1804 BUG_ON((char *)de
>= sb
->s_blocksize
+ insert_bh
->b_data
);
1805 /* These checks should've already been passed by the
1806 * prepare function, but I guess we can leave them
1808 if (!ocfs2_check_dir_entry(dir
, de
, insert_bh
, offset
)) {
1812 if (ocfs2_match(namelen
, name
, de
)) {
1816 if (((le64_to_cpu(de
->inode
) == 0) &&
1817 (le16_to_cpu(de
->rec_len
) >= rec_len
)) ||
1818 (le16_to_cpu(de
->rec_len
) >=
1819 (OCFS2_DIR_REC_LEN(de
->name_len
) + rec_len
))) {
1820 status
= ocfs2_journal_access(handle
, dir
, insert_bh
,
1821 OCFS2_JOURNAL_ACCESS_WRITE
);
1822 /* By now the buffer is marked for journaling */
1823 offset
+= le16_to_cpu(de
->rec_len
);
1824 if (le64_to_cpu(de
->inode
)) {
1825 de1
= (struct ocfs2_dir_entry
*)((char *) de
+
1826 OCFS2_DIR_REC_LEN(de
->name_len
));
1828 cpu_to_le16(le16_to_cpu(de
->rec_len
) -
1829 OCFS2_DIR_REC_LEN(de
->name_len
));
1830 de
->rec_len
= cpu_to_le16(OCFS2_DIR_REC_LEN(de
->name_len
));
1833 de
->file_type
= OCFS2_FT_UNKNOWN
;
1835 de
->inode
= cpu_to_le64(blkno
);
1836 ocfs2_set_de_type(de
, inode
->i_mode
);
1839 de
->name_len
= namelen
;
1840 memcpy(de
->name
, name
, namelen
);
1842 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
1844 status
= ocfs2_journal_dirty(handle
, insert_bh
);
1848 offset
+= le16_to_cpu(de
->rec_len
);
1849 de
= (struct ocfs2_dir_entry
*) ((char *) de
+ le16_to_cpu(de
->rec_len
));
1852 /* when you think about it, the assert above should prevent us
1853 * from ever getting here. */
1863 * ocfs2_delete_entry deletes a directory entry by merging it with the
1866 static int ocfs2_delete_entry(struct ocfs2_journal_handle
*handle
,
1868 struct ocfs2_dir_entry
*de_del
,
1869 struct buffer_head
*bh
)
1871 struct ocfs2_dir_entry
*de
, *pde
;
1872 int i
, status
= -ENOENT
;
1874 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle
, dir
, de_del
, bh
);
1878 de
= (struct ocfs2_dir_entry
*) bh
->b_data
;
1879 while (i
< bh
->b_size
) {
1880 if (!ocfs2_check_dir_entry(dir
, de
, bh
, i
)) {
1886 status
= ocfs2_journal_access(handle
, dir
, bh
,
1887 OCFS2_JOURNAL_ACCESS_WRITE
);
1895 cpu_to_le16(le16_to_cpu(pde
->rec_len
) +
1896 le16_to_cpu(de
->rec_len
));
1900 status
= ocfs2_journal_dirty(handle
, bh
);
1903 i
+= le16_to_cpu(de
->rec_len
);
1905 de
= (struct ocfs2_dir_entry
*)((char *)de
+ le16_to_cpu(de
->rec_len
));
1913 * Returns 0 if not found, -1 on failure, and 1 on success
1915 static int inline ocfs2_search_dirblock(struct buffer_head
*bh
,
1917 const char *name
, int namelen
,
1918 unsigned long offset
,
1919 struct ocfs2_dir_entry
**res_dir
)
1921 struct ocfs2_dir_entry
*de
;
1922 char *dlimit
, *de_buf
;
1928 de_buf
= bh
->b_data
;
1929 dlimit
= de_buf
+ dir
->i_sb
->s_blocksize
;
1931 while (de_buf
< dlimit
) {
1932 /* this code is executed quadratically often */
1933 /* do minimal checking `by hand' */
1935 de
= (struct ocfs2_dir_entry
*) de_buf
;
1937 if (de_buf
+ namelen
<= dlimit
&&
1938 ocfs2_match(namelen
, name
, de
)) {
1939 /* found a match - just to be sure, do a full check */
1940 if (!ocfs2_check_dir_entry(dir
, de
, bh
, offset
)) {
1949 /* prevent looping on a bad block */
1950 de_len
= le16_to_cpu(de
->rec_len
);
1965 struct buffer_head
*ocfs2_find_entry(const char *name
, int namelen
,
1967 struct ocfs2_dir_entry
**res_dir
)
1969 struct super_block
*sb
;
1970 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1971 struct buffer_head
*bh
, *ret
= NULL
;
1972 unsigned long start
, block
, b
;
1973 int ra_max
= 0; /* Number of bh's in the readahead
1975 int ra_ptr
= 0; /* Current index into readahead
1978 int nblocks
, i
, err
;
1985 nblocks
= i_size_read(dir
) >> sb
->s_blocksize_bits
;
1986 start
= OCFS2_I(dir
)->ip_dir_start_lookup
;
1987 if (start
>= nblocks
)
1994 * We deal with the read-ahead logic here.
1996 if (ra_ptr
>= ra_max
) {
1997 /* Refill the readahead buffer */
2000 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
2002 * Terminate if we reach the end of the
2003 * directory and must wrap, or if our
2004 * search has finished at this block.
2006 if (b
>= nblocks
|| (num
&& block
== start
)) {
2007 bh_use
[ra_max
] = NULL
;
2012 bh
= ocfs2_bread(dir
, b
++, &err
, 1);
2013 bh_use
[ra_max
] = bh
;
2016 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
2019 if (!buffer_uptodate(bh
)) {
2020 /* read error, skip block & hope for the best */
2021 ocfs2_error(dir
->i_sb
, "reading directory %llu, "
2023 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
2028 i
= ocfs2_search_dirblock(bh
, dir
, name
, namelen
,
2029 block
<< sb
->s_blocksize_bits
,
2032 OCFS2_I(dir
)->ip_dir_start_lookup
= block
;
2034 goto cleanup_and_exit
;
2038 goto cleanup_and_exit
;
2041 if (++block
>= nblocks
)
2043 } while (block
!= start
);
2046 * If the directory has grown while we were searching, then
2047 * search the last part of the directory before giving up.
2050 nblocks
= i_size_read(dir
) >> sb
->s_blocksize_bits
;
2051 if (block
< nblocks
) {
2057 /* Clean up the read-ahead blocks */
2058 for (; ra_ptr
< ra_max
; ra_ptr
++)
2059 brelse(bh_use
[ra_ptr
]);
2065 static int ocfs2_blkno_stringify(u64 blkno
, char *name
)
2067 int status
, namelen
;
2071 namelen
= snprintf(name
, OCFS2_ORPHAN_NAMELEN
+ 1, "%016llx",
2081 if (namelen
!= OCFS2_ORPHAN_NAMELEN
) {
2087 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name
,
2096 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
2097 struct ocfs2_journal_handle
*handle
,
2098 struct inode
*inode
,
2100 struct buffer_head
**de_bh
)
2102 struct inode
*orphan_dir_inode
= NULL
;
2103 struct buffer_head
*orphan_dir_bh
= NULL
;
2106 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2112 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2113 ORPHAN_DIR_SYSTEM_INODE
,
2115 if (!orphan_dir_inode
) {
2121 ocfs2_handle_add_inode(handle
, orphan_dir_inode
);
2122 status
= ocfs2_meta_lock(orphan_dir_inode
, handle
, &orphan_dir_bh
, 1);
2128 status
= ocfs2_prepare_dir_for_insert(osb
, orphan_dir_inode
,
2129 orphan_dir_bh
, name
,
2130 OCFS2_ORPHAN_NAMELEN
, de_bh
);
2137 if (orphan_dir_inode
)
2138 iput(orphan_dir_inode
);
2141 brelse(orphan_dir_bh
);
2147 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
2148 struct ocfs2_journal_handle
*handle
,
2149 struct inode
*inode
,
2150 struct ocfs2_dinode
*fe
,
2152 struct buffer_head
*de_bh
)
2154 struct inode
*orphan_dir_inode
= NULL
;
2155 struct buffer_head
*orphan_dir_bh
= NULL
;
2157 struct ocfs2_dinode
*orphan_fe
;
2159 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
2161 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2162 ORPHAN_DIR_SYSTEM_INODE
,
2164 if (!orphan_dir_inode
) {
2170 status
= ocfs2_read_block(osb
,
2171 OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2172 &orphan_dir_bh
, OCFS2_BH_CACHED
,
2179 status
= ocfs2_journal_access(handle
, orphan_dir_inode
, orphan_dir_bh
,
2180 OCFS2_JOURNAL_ACCESS_WRITE
);
2186 /* we're a cluster, and nlink can change on disk from
2187 * underneath us... */
2188 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2189 if (S_ISDIR(inode
->i_mode
))
2190 le16_add_cpu(&orphan_fe
->i_links_count
, 1);
2191 orphan_dir_inode
->i_nlink
= le16_to_cpu(orphan_fe
->i_links_count
);
2193 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2199 status
= __ocfs2_add_entry(handle
, orphan_dir_inode
, name
,
2200 OCFS2_ORPHAN_NAMELEN
, inode
,
2201 OCFS2_I(inode
)->ip_blkno
,
2202 orphan_dir_bh
, de_bh
);
2208 le32_add_cpu(&fe
->i_flags
, OCFS2_ORPHANED_FL
);
2210 /* Record which orphan dir our inode now resides
2211 * in. delete_inode will use this to determine which orphan
2213 spin_lock(&OCFS2_I(inode
)->ip_lock
);
2214 OCFS2_I(inode
)->ip_orphaned_slot
= osb
->slot_num
;
2215 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
2217 mlog(0, "Inode %llu orphaned in slot %d\n",
2218 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, osb
->slot_num
);
2221 if (orphan_dir_inode
)
2222 iput(orphan_dir_inode
);
2225 brelse(orphan_dir_bh
);
2231 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2232 int ocfs2_orphan_del(struct ocfs2_super
*osb
,
2233 struct ocfs2_journal_handle
*handle
,
2234 struct inode
*orphan_dir_inode
,
2235 struct inode
*inode
,
2236 struct buffer_head
*orphan_dir_bh
)
2238 char name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2239 struct ocfs2_dinode
*orphan_fe
;
2241 struct buffer_head
*target_de_bh
= NULL
;
2242 struct ocfs2_dir_entry
*target_de
= NULL
;
2246 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2252 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2253 name
, (unsigned long long)OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2254 OCFS2_ORPHAN_NAMELEN
);
2256 /* find it's spot in the orphan directory */
2257 target_de_bh
= ocfs2_find_entry(name
, OCFS2_ORPHAN_NAMELEN
,
2258 orphan_dir_inode
, &target_de
);
2259 if (!target_de_bh
) {
2265 /* remove it from the orphan directory */
2266 status
= ocfs2_delete_entry(handle
, orphan_dir_inode
, target_de
,
2273 status
= ocfs2_journal_access(handle
,orphan_dir_inode
, orphan_dir_bh
,
2274 OCFS2_JOURNAL_ACCESS_WRITE
);
2280 /* do the i_nlink dance! :) */
2281 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2282 if (S_ISDIR(inode
->i_mode
))
2283 le16_add_cpu(&orphan_fe
->i_links_count
, -1);
2284 orphan_dir_inode
->i_nlink
= le16_to_cpu(orphan_fe
->i_links_count
);
2286 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2294 brelse(target_de_bh
);
2300 struct inode_operations ocfs2_dir_iops
= {
2301 .create
= ocfs2_create
,
2302 .lookup
= ocfs2_lookup
,
2304 .unlink
= ocfs2_unlink
,
2305 .rmdir
= ocfs2_unlink
,
2306 .symlink
= ocfs2_symlink
,
2307 .mkdir
= ocfs2_mkdir
,
2308 .mknod
= ocfs2_mknod
,
2309 .rename
= ocfs2_rename
,
2310 .setattr
= ocfs2_setattr
,
2311 .getattr
= ocfs2_getattr
,