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"
64 #include "buffer_head_io.h"
66 #define NAMEI_RA_CHUNKS 2
67 #define NAMEI_RA_BLOCKS 4
68 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
69 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
71 static int inline ocfs2_search_dirblock(struct buffer_head
*bh
,
73 const char *name
, int namelen
,
75 struct ocfs2_dir_entry
**res_dir
);
77 static int ocfs2_delete_entry(struct ocfs2_journal_handle
*handle
,
79 struct ocfs2_dir_entry
*de_del
,
80 struct buffer_head
*bh
);
82 static int __ocfs2_add_entry(struct ocfs2_journal_handle
*handle
,
84 const char *name
, int namelen
,
85 struct inode
*inode
, u64 blkno
,
86 struct buffer_head
*parent_fe_bh
,
87 struct buffer_head
*insert_bh
);
89 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
91 struct dentry
*dentry
, int mode
,
93 struct buffer_head
**new_fe_bh
,
94 struct buffer_head
*parent_fe_bh
,
95 struct ocfs2_journal_handle
*handle
,
96 struct inode
**ret_inode
,
97 struct ocfs2_alloc_context
*inode_ac
);
99 static int ocfs2_fill_new_dir(struct ocfs2_super
*osb
,
100 struct ocfs2_journal_handle
*handle
,
101 struct inode
*parent
,
103 struct buffer_head
*fe_bh
,
104 struct ocfs2_alloc_context
*data_ac
);
106 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
107 struct ocfs2_journal_handle
*handle
,
108 struct buffer_head
**bh1
,
109 struct inode
*inode1
,
110 struct buffer_head
**bh2
,
111 struct inode
*inode2
);
113 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
114 struct ocfs2_journal_handle
*handle
,
117 struct buffer_head
**de_bh
);
119 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
120 struct ocfs2_journal_handle
*handle
,
122 struct ocfs2_dinode
*fe
,
124 struct buffer_head
*de_bh
);
126 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
127 struct ocfs2_journal_handle
*handle
,
129 const char *symname
);
131 static inline int ocfs2_add_entry(struct ocfs2_journal_handle
*handle
,
132 struct dentry
*dentry
,
133 struct inode
*inode
, u64 blkno
,
134 struct buffer_head
*parent_fe_bh
,
135 struct buffer_head
*insert_bh
)
137 return __ocfs2_add_entry(handle
, dentry
->d_parent
->d_inode
,
138 dentry
->d_name
.name
, dentry
->d_name
.len
,
139 inode
, blkno
, parent_fe_bh
, insert_bh
);
142 /* An orphan dir name is an 8 byte value, printed as a hex string */
143 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
145 static struct dentry
*ocfs2_lookup(struct inode
*dir
, struct dentry
*dentry
,
146 struct nameidata
*nd
)
150 struct buffer_head
*dirent_bh
= NULL
;
151 struct inode
*inode
= NULL
;
153 struct ocfs2_dir_entry
*dirent
;
154 struct ocfs2_inode_info
*oi
;
156 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
157 dentry
->d_name
.len
, dentry
->d_name
.name
);
159 if (dentry
->d_name
.len
> OCFS2_MAX_FILENAME_LEN
) {
160 ret
= ERR_PTR(-ENAMETOOLONG
);
164 mlog(0, "find name %.*s in directory %llu\n", dentry
->d_name
.len
,
165 dentry
->d_name
.name
, (unsigned long long)OCFS2_I(dir
)->ip_blkno
);
167 status
= ocfs2_meta_lock(dir
, NULL
, NULL
, 0);
169 if (status
!= -ENOENT
)
171 ret
= ERR_PTR(status
);
175 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
176 dentry
->d_name
.len
, &blkno
,
177 dir
, &dirent_bh
, &dirent
);
181 inode
= ocfs2_iget(OCFS2_SB(dir
->i_sb
), blkno
);
183 mlog(ML_ERROR
, "Unable to create inode %llu\n",
184 (unsigned long long)blkno
);
185 ret
= ERR_PTR(-EACCES
);
190 /* Clear any orphaned state... If we were able to look up the
191 * inode from a directory, it certainly can't be orphaned. We
192 * might have the bad state from a node which intended to
193 * orphan this inode but crashed before it could commit the
195 spin_lock(&oi
->ip_lock
);
196 oi
->ip_flags
&= ~OCFS2_INODE_MAYBE_ORPHANED
;
197 oi
->ip_orphaned_slot
= OCFS2_INVALID_SLOT
;
198 spin_unlock(&oi
->ip_lock
);
202 dentry
->d_op
= &ocfs2_dentry_ops
;
203 ret
= d_splice_alias(inode
, dentry
);
206 /* Don't drop the cluster lock until *after* the d_add --
207 * unlink on another node will message us to remove that
208 * dentry under this lock so otherwise we can race this with
209 * the vote thread and have a stale dentry. */
210 ocfs2_meta_unlock(dir
, 0);
221 static int ocfs2_fill_new_dir(struct ocfs2_super
*osb
,
222 struct ocfs2_journal_handle
*handle
,
223 struct inode
*parent
,
225 struct buffer_head
*fe_bh
,
226 struct ocfs2_alloc_context
*data_ac
)
229 struct buffer_head
*new_bh
= NULL
;
230 struct ocfs2_dir_entry
*de
= NULL
;
234 status
= ocfs2_do_extend_dir(osb
->sb
, handle
, inode
, fe_bh
,
235 data_ac
, NULL
, &new_bh
);
241 ocfs2_set_new_buffer_uptodate(inode
, new_bh
);
243 status
= ocfs2_journal_access(handle
, inode
, new_bh
,
244 OCFS2_JOURNAL_ACCESS_CREATE
);
249 memset(new_bh
->b_data
, 0, osb
->sb
->s_blocksize
);
251 de
= (struct ocfs2_dir_entry
*) new_bh
->b_data
;
252 de
->inode
= cpu_to_le64(OCFS2_I(inode
)->ip_blkno
);
255 cpu_to_le16(OCFS2_DIR_REC_LEN(de
->name_len
));
256 strcpy(de
->name
, ".");
257 ocfs2_set_de_type(de
, S_IFDIR
);
258 de
= (struct ocfs2_dir_entry
*) ((char *)de
+ le16_to_cpu(de
->rec_len
));
259 de
->inode
= cpu_to_le64(OCFS2_I(parent
)->ip_blkno
);
260 de
->rec_len
= cpu_to_le16(inode
->i_sb
->s_blocksize
-
261 OCFS2_DIR_REC_LEN(1));
263 strcpy(de
->name
, "..");
264 ocfs2_set_de_type(de
, S_IFDIR
);
266 status
= ocfs2_journal_dirty(handle
, new_bh
);
272 i_size_write(inode
, inode
->i_sb
->s_blocksize
);
274 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(inode
->i_sb
->s_blocksize
);
275 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
290 static int ocfs2_mknod(struct inode
*dir
,
291 struct dentry
*dentry
,
296 struct buffer_head
*parent_fe_bh
= NULL
;
297 struct ocfs2_journal_handle
*handle
= NULL
;
298 struct ocfs2_super
*osb
;
299 struct ocfs2_dinode
*dirfe
;
300 struct buffer_head
*new_fe_bh
= NULL
;
301 struct buffer_head
*de_bh
= NULL
;
302 struct inode
*inode
= NULL
;
303 struct ocfs2_alloc_context
*inode_ac
= NULL
;
304 struct ocfs2_alloc_context
*data_ac
= NULL
;
306 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
, mode
,
307 (unsigned long)dev
, dentry
->d_name
.len
,
308 dentry
->d_name
.name
);
310 /* get our super block */
311 osb
= OCFS2_SB(dir
->i_sb
);
313 handle
= ocfs2_alloc_handle(osb
);
314 if (handle
== NULL
) {
320 status
= ocfs2_meta_lock(dir
, handle
, &parent_fe_bh
, 1);
322 if (status
!= -ENOENT
)
327 if (S_ISDIR(mode
) && (dir
->i_nlink
>= OCFS2_LINK_MAX
)) {
332 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
333 if (!dirfe
->i_links_count
) {
334 /* can't make a file in a deleted directory. */
339 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
344 /* get a spot inside the dir. */
345 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
347 dentry
->d_name
.len
, &de_bh
);
353 /* reserve an inode spot */
354 status
= ocfs2_reserve_new_inode(osb
, handle
, &inode_ac
);
356 if (status
!= -ENOSPC
)
361 /* are we making a directory? If so, reserve a cluster for his
364 status
= ocfs2_reserve_clusters(osb
, handle
, 1, &data_ac
);
366 if (status
!= -ENOSPC
)
372 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_MKNOD_CREDITS
);
373 if (IS_ERR(handle
)) {
374 status
= PTR_ERR(handle
);
380 /* do the real work now. */
381 status
= ocfs2_mknod_locked(osb
, dir
, dentry
, mode
, dev
,
382 &new_fe_bh
, parent_fe_bh
, handle
,
390 status
= ocfs2_fill_new_dir(osb
, handle
, dir
, inode
,
397 status
= ocfs2_journal_access(handle
, dir
, parent_fe_bh
,
398 OCFS2_JOURNAL_ACCESS_WRITE
);
403 le16_add_cpu(&dirfe
->i_links_count
, 1);
404 status
= ocfs2_journal_dirty(handle
, parent_fe_bh
);
412 status
= ocfs2_add_entry(handle
, dentry
, inode
,
413 OCFS2_I(inode
)->ip_blkno
, parent_fe_bh
,
420 insert_inode_hash(inode
);
421 dentry
->d_op
= &ocfs2_dentry_ops
;
422 d_instantiate(dentry
, inode
);
426 ocfs2_commit_trans(handle
);
428 if (status
== -ENOSPC
)
429 mlog(0, "Disk is full\n");
438 brelse(parent_fe_bh
);
440 if ((status
< 0) && inode
)
444 ocfs2_free_alloc_context(inode_ac
);
447 ocfs2_free_alloc_context(data_ac
);
454 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
456 struct dentry
*dentry
, int mode
,
458 struct buffer_head
**new_fe_bh
,
459 struct buffer_head
*parent_fe_bh
,
460 struct ocfs2_journal_handle
*handle
,
461 struct inode
**ret_inode
,
462 struct ocfs2_alloc_context
*inode_ac
)
465 struct ocfs2_dinode
*fe
= NULL
;
466 struct ocfs2_extent_list
*fel
;
469 struct inode
*inode
= NULL
;
471 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
, mode
,
472 (unsigned long)dev
, dentry
->d_name
.len
,
473 dentry
->d_name
.name
);
478 status
= ocfs2_claim_new_inode(osb
, handle
, inode_ac
, &suballoc_bit
,
485 inode
= new_inode(dir
->i_sb
);
487 status
= PTR_ERR(inode
);
488 mlog(ML_ERROR
, "new_inode failed!\n");
492 /* populate as many fields early on as possible - many of
493 * these are used by the support functions here and in
495 inode
->i_ino
= ino_from_blkno(osb
->sb
, fe_blkno
);
496 OCFS2_I(inode
)->ip_blkno
= fe_blkno
;
501 inode
->i_mode
= mode
;
502 spin_lock(&osb
->osb_lock
);
503 inode
->i_generation
= osb
->s_next_generation
++;
504 spin_unlock(&osb
->osb_lock
);
506 *new_fe_bh
= sb_getblk(osb
->sb
, fe_blkno
);
512 ocfs2_set_new_buffer_uptodate(inode
, *new_fe_bh
);
514 status
= ocfs2_journal_access(handle
, inode
, *new_fe_bh
,
515 OCFS2_JOURNAL_ACCESS_CREATE
);
521 fe
= (struct ocfs2_dinode
*) (*new_fe_bh
)->b_data
;
522 memset(fe
, 0, osb
->sb
->s_blocksize
);
524 fe
->i_generation
= cpu_to_le32(inode
->i_generation
);
525 fe
->i_fs_generation
= cpu_to_le32(osb
->fs_generation
);
526 fe
->i_blkno
= cpu_to_le64(fe_blkno
);
527 fe
->i_suballoc_bit
= cpu_to_le16(suballoc_bit
);
528 fe
->i_suballoc_slot
= cpu_to_le16(osb
->slot_num
);
529 fe
->i_uid
= cpu_to_le32(current
->fsuid
);
530 if (dir
->i_mode
& S_ISGID
) {
531 fe
->i_gid
= cpu_to_le32(dir
->i_gid
);
535 fe
->i_gid
= cpu_to_le32(current
->fsgid
);
536 fe
->i_mode
= cpu_to_le16(mode
);
537 if (S_ISCHR(mode
) || S_ISBLK(mode
))
538 fe
->id1
.dev1
.i_rdev
= cpu_to_le64(huge_encode_dev(dev
));
540 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
542 fe
->i_last_eb_blk
= 0;
543 strcpy(fe
->i_signature
, OCFS2_INODE_SIGNATURE
);
544 le32_add_cpu(&fe
->i_flags
, OCFS2_VALID_FL
);
545 fe
->i_atime
= fe
->i_ctime
= fe
->i_mtime
=
546 cpu_to_le64(CURRENT_TIME
.tv_sec
);
547 fe
->i_mtime_nsec
= fe
->i_ctime_nsec
= fe
->i_atime_nsec
=
548 cpu_to_le32(CURRENT_TIME
.tv_nsec
);
551 fel
= &fe
->id2
.i_list
;
552 fel
->l_tree_depth
= 0;
553 fel
->l_next_free_rec
= 0;
554 fel
->l_count
= cpu_to_le16(ocfs2_extent_recs_per_inode(osb
->sb
));
556 status
= ocfs2_journal_dirty(handle
, *new_fe_bh
);
562 if (ocfs2_populate_inode(inode
, fe
, 1) < 0) {
563 mlog(ML_ERROR
, "populate inode failed! bh->b_blocknr=%llu, "
564 "i_blkno=%llu, i_ino=%lu\n",
565 (unsigned long long) (*new_fe_bh
)->b_blocknr
,
566 (unsigned long long)fe
->i_blkno
, inode
->i_ino
);
570 ocfs2_inode_set_new(osb
, inode
);
571 status
= ocfs2_create_new_inode_locks(inode
);
575 status
= 0; /* error in ocfs2_create_new_inode_locks is not
593 static int ocfs2_mkdir(struct inode
*dir
,
594 struct dentry
*dentry
,
599 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
600 dentry
->d_name
.len
, dentry
->d_name
.name
);
601 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
607 static int ocfs2_create(struct inode
*dir
,
608 struct dentry
*dentry
,
610 struct nameidata
*nd
)
614 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
615 dentry
->d_name
.len
, dentry
->d_name
.name
);
616 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
622 static int ocfs2_link(struct dentry
*old_dentry
,
624 struct dentry
*dentry
)
626 struct ocfs2_journal_handle
*handle
= NULL
;
627 struct inode
*inode
= old_dentry
->d_inode
;
629 struct buffer_head
*fe_bh
= NULL
;
630 struct buffer_head
*parent_fe_bh
= NULL
;
631 struct buffer_head
*de_bh
= NULL
;
632 struct ocfs2_dinode
*fe
= NULL
;
633 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
635 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode
->i_ino
,
636 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
637 dentry
->d_name
.len
, dentry
->d_name
.name
);
639 if (S_ISDIR(inode
->i_mode
)) {
644 if (inode
->i_nlink
>= OCFS2_LINK_MAX
) {
649 handle
= ocfs2_alloc_handle(osb
);
650 if (handle
== NULL
) {
655 err
= ocfs2_meta_lock(dir
, handle
, &parent_fe_bh
, 1);
662 err
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
667 err
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
669 dentry
->d_name
.len
, &de_bh
);
675 err
= ocfs2_meta_lock(inode
, handle
, &fe_bh
, 1);
682 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
683 if (le16_to_cpu(fe
->i_links_count
) >= OCFS2_LINK_MAX
) {
688 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_LINK_CREDITS
);
689 if (IS_ERR(handle
)) {
690 err
= PTR_ERR(handle
);
696 err
= ocfs2_journal_access(handle
, inode
, fe_bh
,
697 OCFS2_JOURNAL_ACCESS_WRITE
);
704 inode
->i_ctime
= CURRENT_TIME
;
705 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
706 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
707 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
709 err
= ocfs2_journal_dirty(handle
, fe_bh
);
711 le16_add_cpu(&fe
->i_links_count
, -1);
717 err
= ocfs2_add_entry(handle
, dentry
, inode
,
718 OCFS2_I(inode
)->ip_blkno
,
719 parent_fe_bh
, de_bh
);
721 le16_add_cpu(&fe
->i_links_count
, -1);
727 atomic_inc(&inode
->i_count
);
728 dentry
->d_op
= &ocfs2_dentry_ops
;
729 d_instantiate(dentry
, inode
);
732 ocfs2_commit_trans(handle
);
738 brelse(parent_fe_bh
);
745 static int ocfs2_unlink(struct inode
*dir
,
746 struct dentry
*dentry
)
749 unsigned int saved_nlink
= 0;
750 struct inode
*inode
= dentry
->d_inode
;
751 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
753 struct ocfs2_dinode
*fe
= NULL
;
754 struct buffer_head
*fe_bh
= NULL
;
755 struct buffer_head
*parent_node_bh
= NULL
;
756 struct ocfs2_journal_handle
*handle
= NULL
;
757 struct ocfs2_dir_entry
*dirent
= NULL
;
758 struct buffer_head
*dirent_bh
= NULL
;
759 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
760 struct buffer_head
*orphan_entry_bh
= NULL
;
762 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
763 dentry
->d_name
.len
, dentry
->d_name
.name
);
765 BUG_ON(dentry
->d_parent
->d_inode
!= dir
);
767 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
769 if (inode
== osb
->root_inode
) {
770 mlog(0, "Cannot delete the root directory\n");
775 handle
= ocfs2_alloc_handle(osb
);
776 if (handle
== NULL
) {
782 status
= ocfs2_meta_lock(dir
, handle
, &parent_node_bh
, 1);
784 if (status
!= -ENOENT
)
789 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
790 dentry
->d_name
.len
, &blkno
,
791 dir
, &dirent_bh
, &dirent
);
793 if (status
!= -ENOENT
)
798 if (OCFS2_I(inode
)->ip_blkno
!= blkno
) {
801 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
802 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
803 (unsigned long long)blkno
, OCFS2_I(inode
)->ip_flags
);
807 status
= ocfs2_meta_lock(inode
, handle
, &fe_bh
, 1);
809 if (status
!= -ENOENT
)
814 if (S_ISDIR(inode
->i_mode
)) {
815 if (!ocfs2_empty_dir(inode
)) {
818 } else if (inode
->i_nlink
!= 2) {
824 /* There are still a few steps left until we can consider the
825 * unlink to have succeeded. Save off nlink here before
826 * modification so we can set it back in case we hit an issue
828 saved_nlink
= inode
->i_nlink
;
829 if (S_ISDIR(inode
->i_mode
))
834 status
= ocfs2_request_unlink_vote(inode
, dentry
,
835 (unsigned int) inode
->i_nlink
);
837 /* This vote should succeed under all normal
843 if (!inode
->i_nlink
) {
844 status
= ocfs2_prepare_orphan_dir(osb
, handle
, inode
,
853 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_UNLINK_CREDITS
);
854 if (IS_ERR(handle
)) {
855 status
= PTR_ERR(handle
);
861 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
862 OCFS2_JOURNAL_ACCESS_WRITE
);
868 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
870 if (!inode
->i_nlink
) {
871 status
= ocfs2_orphan_add(osb
, handle
, inode
, fe
, orphan_name
,
879 /* delete the name from the parent dir */
880 status
= ocfs2_delete_entry(handle
, dir
, dirent
, dirent_bh
);
886 /* We can set nlink on the dinode now. clear the saved version
887 * so that it doesn't get set later. */
888 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
891 status
= ocfs2_journal_dirty(handle
, fe_bh
);
897 if (S_ISDIR(inode
->i_mode
)) {
899 status
= ocfs2_mark_inode_dirty(handle
, dir
,
908 if (status
< 0 && saved_nlink
)
909 inode
->i_nlink
= saved_nlink
;
912 ocfs2_commit_trans(handle
);
921 brelse(parent_node_bh
);
924 brelse(orphan_entry_bh
);
932 * The only place this should be used is rename!
933 * if they have the same id, then the 1st one is the only one locked.
935 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
936 struct ocfs2_journal_handle
*handle
,
937 struct buffer_head
**bh1
,
938 struct inode
*inode1
,
939 struct buffer_head
**bh2
,
940 struct inode
*inode2
)
943 struct ocfs2_inode_info
*oi1
= OCFS2_I(inode1
);
944 struct ocfs2_inode_info
*oi2
= OCFS2_I(inode2
);
945 struct buffer_head
**tmpbh
;
946 struct inode
*tmpinode
;
948 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
949 (unsigned long long)oi1
->ip_blkno
,
950 (unsigned long long)oi2
->ip_blkno
);
959 /* we always want to lock the one with the lower lockid first. */
960 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
961 if (oi1
->ip_blkno
< oi2
->ip_blkno
) {
962 /* switch id1 and id2 around */
963 mlog(0, "switching them around...\n");
973 status
= ocfs2_meta_lock(inode2
, handle
, bh2
, 1);
975 if (status
!= -ENOENT
)
981 status
= ocfs2_meta_lock(inode1
, handle
, bh1
, 1);
983 if (status
!= -ENOENT
)
992 #define PARENT_INO(buffer) \
993 ((struct ocfs2_dir_entry *) \
995 le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
997 static int ocfs2_rename(struct inode
*old_dir
,
998 struct dentry
*old_dentry
,
999 struct inode
*new_dir
,
1000 struct dentry
*new_dentry
)
1002 int status
= 0, rename_lock
= 0;
1003 struct inode
*old_inode
= old_dentry
->d_inode
;
1004 struct inode
*new_inode
= new_dentry
->d_inode
;
1005 struct ocfs2_dinode
*newfe
= NULL
;
1006 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1007 struct buffer_head
*orphan_entry_bh
= NULL
;
1008 struct buffer_head
*newfe_bh
= NULL
;
1009 struct buffer_head
*insert_entry_bh
= NULL
;
1010 struct ocfs2_super
*osb
= NULL
;
1012 struct ocfs2_journal_handle
*handle
= NULL
;
1013 struct buffer_head
*old_dir_bh
= NULL
;
1014 struct buffer_head
*new_dir_bh
= NULL
;
1015 struct ocfs2_dir_entry
*old_de
= NULL
, *new_de
= NULL
; // dirent for old_dentry
1017 struct buffer_head
*new_de_bh
= NULL
, *old_de_bh
= NULL
; // bhs for above
1018 struct buffer_head
*old_inode_de_bh
= NULL
; // if old_dentry is a dir,
1019 // this is the 1st dirent bh
1020 nlink_t old_dir_nlink
= old_dir
->i_nlink
, new_dir_nlink
= new_dir
->i_nlink
;
1021 unsigned int links_count
;
1023 /* At some point it might be nice to break this function up a
1026 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1027 old_dir
, old_dentry
, new_dir
, new_dentry
,
1028 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
1029 new_dentry
->d_name
.len
, new_dentry
->d_name
.name
);
1031 osb
= OCFS2_SB(old_dir
->i_sb
);
1034 if (!igrab(new_inode
))
1038 if (atomic_read(&old_dentry
->d_count
) > 2) {
1039 shrink_dcache_parent(old_dentry
);
1040 if (atomic_read(&old_dentry
->d_count
) > 2) {
1046 /* Assume a directory heirarchy thusly:
1049 * a,b,c, and d are all directories.
1051 * from cwd of 'a' on both nodes:
1055 * And that's why, just like the VFS, we need a file system
1057 if (old_dentry
!= new_dentry
) {
1058 status
= ocfs2_rename_lock(osb
);
1066 handle
= ocfs2_alloc_handle(osb
);
1067 if (handle
== NULL
) {
1073 /* if old and new are the same, this'll just do one lock. */
1074 status
= ocfs2_double_lock(osb
, handle
,
1075 &old_dir_bh
, old_dir
,
1076 &new_dir_bh
, new_dir
);
1082 /* make sure both dirs have bhs
1083 * get an extra ref on old_dir_bh if old==new */
1086 new_dir_bh
= old_dir_bh
;
1089 mlog(ML_ERROR
, "no old_dir_bh!\n");
1095 if (S_ISDIR(old_inode
->i_mode
)) {
1096 /* Directories actually require metadata updates to
1097 * the directory info so we can't get away with not
1098 * doing node locking on it. */
1099 status
= ocfs2_meta_lock(old_inode
, handle
, NULL
, 1);
1101 if (status
!= -ENOENT
)
1106 status
= ocfs2_request_rename_vote(old_inode
, old_dentry
);
1113 old_inode_de_bh
= ocfs2_bread(old_inode
, 0, &status
, 0);
1114 if (!old_inode_de_bh
)
1118 if (le64_to_cpu(PARENT_INO(old_inode_de_bh
->b_data
)) !=
1119 OCFS2_I(old_dir
)->ip_blkno
)
1122 if (!new_inode
&& new_dir
!=old_dir
&&
1123 new_dir
->i_nlink
>= OCFS2_LINK_MAX
)
1126 /* Ah, the simple case - we're a file so just send a
1128 status
= ocfs2_request_rename_vote(old_inode
, old_dentry
);
1136 old_de_bh
= ocfs2_find_entry(old_dentry
->d_name
.name
,
1137 old_dentry
->d_name
.len
,
1143 * Check for inode number is _not_ due to possible IO errors.
1144 * We might rmdir the source, keep it as pwd of some process
1145 * and merrily kill the link to whatever was created under the
1146 * same name. Goodbye sticky bit ;-<
1148 if (le64_to_cpu(old_de
->inode
) != OCFS2_I(old_inode
)->ip_blkno
)
1151 /* check if the target already exists (in which case we need
1153 status
= ocfs2_find_files_on_disk(new_dentry
->d_name
.name
,
1154 new_dentry
->d_name
.len
,
1155 &newfe_blkno
, new_dir
, &new_de_bh
,
1157 /* The only error we allow here is -ENOENT because the new
1158 * file not existing is perfectly valid. */
1159 if ((status
< 0) && (status
!= -ENOENT
)) {
1160 /* If we cannot find the file specified we should just */
1161 /* return the error... */
1166 if (!new_de
&& new_inode
)
1167 mlog(ML_ERROR
, "inode %lu does not exist in it's parent "
1168 "directory!", new_inode
->i_ino
);
1170 /* In case we need to overwrite an existing file, we blow it
1173 /* VFS didn't think there existed an inode here, but
1174 * someone else in the cluster must have raced our
1175 * rename to create one. Today we error cleanly, in
1176 * the future we should consider calling iget to build
1177 * a new struct inode for this entry. */
1181 mlog(0, "We found an inode for name %.*s but VFS "
1182 "didn't give us one.\n", new_dentry
->d_name
.len
,
1183 new_dentry
->d_name
.name
);
1187 if (OCFS2_I(new_inode
)->ip_blkno
!= newfe_blkno
) {
1190 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1191 (unsigned long long)OCFS2_I(new_inode
)->ip_blkno
,
1192 (unsigned long long)newfe_blkno
,
1193 OCFS2_I(new_inode
)->ip_flags
);
1197 status
= ocfs2_meta_lock(new_inode
, handle
, &newfe_bh
, 1);
1199 if (status
!= -ENOENT
)
1204 if (S_ISDIR(new_inode
->i_mode
))
1207 links_count
= (unsigned int) (new_inode
->i_nlink
- 1);
1209 status
= ocfs2_request_unlink_vote(new_inode
, new_dentry
,
1216 newfe
= (struct ocfs2_dinode
*) newfe_bh
->b_data
;
1218 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1219 "newfebh=%p bhblocknr=%llu\n", new_de
,
1220 (unsigned long long)newfe_blkno
, newfe_bh
, newfe_bh
?
1221 (unsigned long long)newfe_bh
->b_blocknr
: 0ULL);
1223 if (S_ISDIR(new_inode
->i_mode
) || (new_inode
->i_nlink
== 1)) {
1224 status
= ocfs2_prepare_orphan_dir(osb
, handle
,
1234 BUG_ON(new_dentry
->d_parent
->d_inode
!= new_dir
);
1236 status
= ocfs2_check_dir_for_entry(new_dir
,
1237 new_dentry
->d_name
.name
,
1238 new_dentry
->d_name
.len
);
1242 status
= ocfs2_prepare_dir_for_insert(osb
, new_dir
, new_dir_bh
,
1243 new_dentry
->d_name
.name
,
1244 new_dentry
->d_name
.len
,
1252 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_RENAME_CREDITS
);
1253 if (IS_ERR(handle
)) {
1254 status
= PTR_ERR(handle
);
1261 if (S_ISDIR(new_inode
->i_mode
)) {
1262 if (!ocfs2_empty_dir(new_inode
) ||
1263 new_inode
->i_nlink
!= 2) {
1264 status
= -ENOTEMPTY
;
1268 status
= ocfs2_journal_access(handle
, new_inode
, newfe_bh
,
1269 OCFS2_JOURNAL_ACCESS_WRITE
);
1275 if (S_ISDIR(new_inode
->i_mode
) ||
1276 (newfe
->i_links_count
== cpu_to_le16(1))){
1277 status
= ocfs2_orphan_add(osb
, handle
, new_inode
,
1286 /* change the dirent to point to the correct inode */
1287 status
= ocfs2_journal_access(handle
, new_dir
, new_de_bh
,
1288 OCFS2_JOURNAL_ACCESS_WRITE
);
1293 new_de
->inode
= cpu_to_le64(OCFS2_I(old_inode
)->ip_blkno
);
1294 new_de
->file_type
= old_de
->file_type
;
1295 new_dir
->i_version
++;
1296 status
= ocfs2_journal_dirty(handle
, new_de_bh
);
1302 if (S_ISDIR(new_inode
->i_mode
))
1303 newfe
->i_links_count
= 0;
1305 le16_add_cpu(&newfe
->i_links_count
, -1);
1307 status
= ocfs2_journal_dirty(handle
, newfe_bh
);
1313 /* if the name was not found in new_dir, add it now */
1314 status
= ocfs2_add_entry(handle
, new_dentry
, old_inode
,
1315 OCFS2_I(old_inode
)->ip_blkno
,
1316 new_dir_bh
, insert_entry_bh
);
1319 old_inode
->i_ctime
= CURRENT_TIME
;
1320 mark_inode_dirty(old_inode
);
1322 /* now that the name has been added to new_dir, remove the old name */
1323 status
= ocfs2_delete_entry(handle
, old_dir
, old_de
, old_de_bh
);
1330 new_inode
->i_nlink
--;
1331 new_inode
->i_ctime
= CURRENT_TIME
;
1333 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1334 if (old_inode_de_bh
) {
1335 status
= ocfs2_journal_access(handle
, old_inode
,
1337 OCFS2_JOURNAL_ACCESS_WRITE
);
1338 PARENT_INO(old_inode_de_bh
->b_data
) =
1339 cpu_to_le64(OCFS2_I(new_dir
)->ip_blkno
);
1340 status
= ocfs2_journal_dirty(handle
, old_inode_de_bh
);
1343 new_inode
->i_nlink
--;
1346 mark_inode_dirty(new_dir
);
1349 mark_inode_dirty(old_dir
);
1351 mark_inode_dirty(new_inode
);
1353 if (old_dir
!= new_dir
)
1354 if (new_dir_nlink
!= new_dir
->i_nlink
) {
1356 mlog(ML_ERROR
, "need to change nlink for new "
1357 "dir %llu from %d to %d but bh is NULL\n",
1358 (unsigned long long)OCFS2_I(new_dir
)->ip_blkno
,
1359 (int)new_dir_nlink
, new_dir
->i_nlink
);
1361 struct ocfs2_dinode
*fe
;
1362 status
= ocfs2_journal_access(handle
,
1365 OCFS2_JOURNAL_ACCESS_WRITE
);
1366 fe
= (struct ocfs2_dinode
*) new_dir_bh
->b_data
;
1367 fe
->i_links_count
= cpu_to_le16(new_dir
->i_nlink
);
1368 status
= ocfs2_journal_dirty(handle
, new_dir_bh
);
1372 if (old_dir_nlink
!= old_dir
->i_nlink
) {
1374 mlog(ML_ERROR
, "need to change nlink for old dir "
1375 "%llu from %d to %d but bh is NULL!\n",
1376 (unsigned long long)OCFS2_I(old_dir
)->ip_blkno
,
1377 (int)old_dir_nlink
, old_dir
->i_nlink
);
1379 struct ocfs2_dinode
*fe
;
1380 status
= ocfs2_journal_access(handle
, old_dir
,
1382 OCFS2_JOURNAL_ACCESS_WRITE
);
1383 fe
= (struct ocfs2_dinode
*) old_dir_bh
->b_data
;
1384 fe
->i_links_count
= cpu_to_le16(old_dir
->i_nlink
);
1385 status
= ocfs2_journal_dirty(handle
, old_dir_bh
);
1392 ocfs2_rename_unlock(osb
);
1395 ocfs2_commit_trans(handle
);
1398 sync_mapping_buffers(old_inode
->i_mapping
);
1412 if (old_inode_de_bh
)
1413 brelse(old_inode_de_bh
);
1414 if (orphan_entry_bh
)
1415 brelse(orphan_entry_bh
);
1416 if (insert_entry_bh
)
1417 brelse(insert_entry_bh
);
1425 * we expect i_size = strlen(symname). Copy symname into the file
1426 * data, including the null terminator.
1428 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
1429 struct ocfs2_journal_handle
*handle
,
1430 struct inode
*inode
,
1431 const char *symname
)
1433 struct buffer_head
**bhs
= NULL
;
1435 struct super_block
*sb
= osb
->sb
;
1438 int virtual, blocks
, status
, i
, bytes_left
;
1440 bytes_left
= i_size_read(inode
) + 1;
1441 /* we can't trust i_blocks because we're actually going to
1442 * write i_size + 1 bytes. */
1443 blocks
= (bytes_left
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
1445 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1446 (unsigned long long)inode
->i_blocks
,
1447 i_size_read(inode
), blocks
);
1449 /* Sanity check -- make sure we're going to fit. */
1451 ocfs2_clusters_to_bytes(sb
, OCFS2_I(inode
)->ip_clusters
)) {
1457 bhs
= kcalloc(blocks
, sizeof(struct buffer_head
*), GFP_KERNEL
);
1464 status
= ocfs2_extent_map_get_blocks(inode
, 0, 1, &p_blkno
,
1471 /* links can never be larger than one cluster so we know this
1472 * is all going to be contiguous, but do a sanity check
1474 if ((p_blocks
<< sb
->s_blocksize_bits
) < bytes_left
) {
1481 while(bytes_left
> 0) {
1482 c
= &symname
[virtual * sb
->s_blocksize
];
1484 bhs
[virtual] = sb_getblk(sb
, p_blkno
);
1485 if (!bhs
[virtual]) {
1490 ocfs2_set_new_buffer_uptodate(inode
, bhs
[virtual]);
1492 status
= ocfs2_journal_access(handle
, inode
, bhs
[virtual],
1493 OCFS2_JOURNAL_ACCESS_CREATE
);
1499 memset(bhs
[virtual]->b_data
, 0, sb
->s_blocksize
);
1501 memcpy(bhs
[virtual]->b_data
, c
,
1502 (bytes_left
> sb
->s_blocksize
) ? sb
->s_blocksize
:
1505 status
= ocfs2_journal_dirty(handle
, bhs
[virtual]);
1513 bytes_left
-= sb
->s_blocksize
;
1520 for(i
= 0; i
< blocks
; i
++)
1530 static int ocfs2_symlink(struct inode
*dir
,
1531 struct dentry
*dentry
,
1532 const char *symname
)
1534 int status
, l
, credits
;
1536 struct ocfs2_super
*osb
= NULL
;
1537 struct inode
*inode
= NULL
;
1538 struct super_block
*sb
;
1539 struct buffer_head
*new_fe_bh
= NULL
;
1540 struct buffer_head
*de_bh
= NULL
;
1541 struct buffer_head
*parent_fe_bh
= NULL
;
1542 struct ocfs2_dinode
*fe
= NULL
;
1543 struct ocfs2_dinode
*dirfe
;
1544 struct ocfs2_journal_handle
*handle
= NULL
;
1545 struct ocfs2_alloc_context
*inode_ac
= NULL
;
1546 struct ocfs2_alloc_context
*data_ac
= NULL
;
1548 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir
,
1549 dentry
, symname
, dentry
->d_name
.len
, dentry
->d_name
.name
);
1554 l
= strlen(symname
) + 1;
1556 credits
= ocfs2_calc_symlink_credits(sb
);
1558 handle
= ocfs2_alloc_handle(osb
);
1559 if (handle
== NULL
) {
1565 /* lock the parent directory */
1566 status
= ocfs2_meta_lock(dir
, handle
, &parent_fe_bh
, 1);
1568 if (status
!= -ENOENT
)
1573 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
1574 if (!dirfe
->i_links_count
) {
1575 /* can't make a file in a deleted directory. */
1580 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
1581 dentry
->d_name
.len
);
1585 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
1586 dentry
->d_name
.name
,
1587 dentry
->d_name
.len
, &de_bh
);
1593 status
= ocfs2_reserve_new_inode(osb
, handle
, &inode_ac
);
1595 if (status
!= -ENOSPC
)
1600 /* don't reserve bitmap space for fast symlinks. */
1601 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1602 status
= ocfs2_reserve_clusters(osb
, handle
, 1, &data_ac
);
1604 if (status
!= -ENOSPC
)
1610 handle
= ocfs2_start_trans(osb
, handle
, credits
);
1611 if (IS_ERR(handle
)) {
1612 status
= PTR_ERR(handle
);
1618 status
= ocfs2_mknod_locked(osb
, dir
, dentry
,
1619 S_IFLNK
| S_IRWXUGO
, 0,
1620 &new_fe_bh
, parent_fe_bh
, handle
,
1627 fe
= (struct ocfs2_dinode
*) new_fe_bh
->b_data
;
1630 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1631 inode
->i_op
= &ocfs2_symlink_inode_operations
;
1632 status
= ocfs2_do_extend_allocation(osb
, inode
, 1, new_fe_bh
,
1633 handle
, data_ac
, NULL
,
1636 if (status
!= -ENOSPC
&& status
!= -EINTR
) {
1638 "Failed to extend file to %llu\n",
1639 (unsigned long long)newsize
);
1645 i_size_write(inode
, newsize
);
1646 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(newsize
);
1648 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
1649 memcpy((char *) fe
->id2
.i_symlink
, symname
, l
);
1650 i_size_write(inode
, newsize
);
1651 inode
->i_blocks
= 0;
1654 status
= ocfs2_mark_inode_dirty(handle
, inode
, new_fe_bh
);
1660 if (!ocfs2_inode_is_fast_symlink(inode
)) {
1661 status
= ocfs2_create_symlink_data(osb
, handle
, inode
,
1669 status
= ocfs2_add_entry(handle
, dentry
, inode
,
1670 le64_to_cpu(fe
->i_blkno
), parent_fe_bh
,
1677 insert_inode_hash(inode
);
1678 dentry
->d_op
= &ocfs2_dentry_ops
;
1679 d_instantiate(dentry
, inode
);
1682 ocfs2_commit_trans(handle
);
1686 brelse(parent_fe_bh
);
1690 ocfs2_free_alloc_context(inode_ac
);
1692 ocfs2_free_alloc_context(data_ac
);
1693 if ((status
< 0) && inode
)
1701 int ocfs2_check_dir_entry(struct inode
* dir
,
1702 struct ocfs2_dir_entry
* de
,
1703 struct buffer_head
* bh
,
1704 unsigned long offset
)
1706 const char *error_msg
= NULL
;
1707 const int rlen
= le16_to_cpu(de
->rec_len
);
1709 if (rlen
< OCFS2_DIR_REC_LEN(1))
1710 error_msg
= "rec_len is smaller than minimal";
1711 else if (rlen
% 4 != 0)
1712 error_msg
= "rec_len % 4 != 0";
1713 else if (rlen
< OCFS2_DIR_REC_LEN(de
->name_len
))
1714 error_msg
= "rec_len is too small for name_len";
1715 else if (((char *) de
- bh
->b_data
) + rlen
> dir
->i_sb
->s_blocksize
)
1716 error_msg
= "directory entry across blocks";
1718 if (error_msg
!= NULL
)
1719 mlog(ML_ERROR
, "bad entry in directory #%llu: %s - "
1720 "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1721 (unsigned long long)OCFS2_I(dir
)->ip_blkno
, error_msg
,
1722 offset
, (unsigned long long)le64_to_cpu(de
->inode
), rlen
,
1724 return error_msg
== NULL
? 1 : 0;
1727 /* we don't always have a dentry for what we want to add, so people
1728 * like orphan dir can call this instead.
1730 * If you pass me insert_bh, I'll skip the search of the other dir
1731 * blocks and put the record in there.
1733 static int __ocfs2_add_entry(struct ocfs2_journal_handle
*handle
,
1735 const char *name
, int namelen
,
1736 struct inode
*inode
, u64 blkno
,
1737 struct buffer_head
*parent_fe_bh
,
1738 struct buffer_head
*insert_bh
)
1740 unsigned long offset
;
1741 unsigned short rec_len
;
1742 struct ocfs2_dir_entry
*de
, *de1
;
1743 struct super_block
*sb
;
1753 rec_len
= OCFS2_DIR_REC_LEN(namelen
);
1755 de
= (struct ocfs2_dir_entry
*) insert_bh
->b_data
;
1757 BUG_ON((char *)de
>= sb
->s_blocksize
+ insert_bh
->b_data
);
1758 /* These checks should've already been passed by the
1759 * prepare function, but I guess we can leave them
1761 if (!ocfs2_check_dir_entry(dir
, de
, insert_bh
, offset
)) {
1765 if (ocfs2_match(namelen
, name
, de
)) {
1769 if (((le64_to_cpu(de
->inode
) == 0) &&
1770 (le16_to_cpu(de
->rec_len
) >= rec_len
)) ||
1771 (le16_to_cpu(de
->rec_len
) >=
1772 (OCFS2_DIR_REC_LEN(de
->name_len
) + rec_len
))) {
1773 status
= ocfs2_journal_access(handle
, dir
, insert_bh
,
1774 OCFS2_JOURNAL_ACCESS_WRITE
);
1775 /* By now the buffer is marked for journaling */
1776 offset
+= le16_to_cpu(de
->rec_len
);
1777 if (le64_to_cpu(de
->inode
)) {
1778 de1
= (struct ocfs2_dir_entry
*)((char *) de
+
1779 OCFS2_DIR_REC_LEN(de
->name_len
));
1781 cpu_to_le16(le16_to_cpu(de
->rec_len
) -
1782 OCFS2_DIR_REC_LEN(de
->name_len
));
1783 de
->rec_len
= cpu_to_le16(OCFS2_DIR_REC_LEN(de
->name_len
));
1786 de
->file_type
= OCFS2_FT_UNKNOWN
;
1788 de
->inode
= cpu_to_le64(blkno
);
1789 ocfs2_set_de_type(de
, inode
->i_mode
);
1792 de
->name_len
= namelen
;
1793 memcpy(de
->name
, name
, namelen
);
1795 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
1797 status
= ocfs2_journal_dirty(handle
, insert_bh
);
1801 offset
+= le16_to_cpu(de
->rec_len
);
1802 de
= (struct ocfs2_dir_entry
*) ((char *) de
+ le16_to_cpu(de
->rec_len
));
1805 /* when you think about it, the assert above should prevent us
1806 * from ever getting here. */
1816 * ocfs2_delete_entry deletes a directory entry by merging it with the
1819 static int ocfs2_delete_entry(struct ocfs2_journal_handle
*handle
,
1821 struct ocfs2_dir_entry
*de_del
,
1822 struct buffer_head
*bh
)
1824 struct ocfs2_dir_entry
*de
, *pde
;
1825 int i
, status
= -ENOENT
;
1827 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle
, dir
, de_del
, bh
);
1831 de
= (struct ocfs2_dir_entry
*) bh
->b_data
;
1832 while (i
< bh
->b_size
) {
1833 if (!ocfs2_check_dir_entry(dir
, de
, bh
, i
)) {
1839 status
= ocfs2_journal_access(handle
, dir
, bh
,
1840 OCFS2_JOURNAL_ACCESS_WRITE
);
1848 cpu_to_le16(le16_to_cpu(pde
->rec_len
) +
1849 le16_to_cpu(de
->rec_len
));
1853 status
= ocfs2_journal_dirty(handle
, bh
);
1856 i
+= le16_to_cpu(de
->rec_len
);
1858 de
= (struct ocfs2_dir_entry
*)((char *)de
+ le16_to_cpu(de
->rec_len
));
1866 * Returns 0 if not found, -1 on failure, and 1 on success
1868 static int inline ocfs2_search_dirblock(struct buffer_head
*bh
,
1870 const char *name
, int namelen
,
1871 unsigned long offset
,
1872 struct ocfs2_dir_entry
**res_dir
)
1874 struct ocfs2_dir_entry
*de
;
1875 char *dlimit
, *de_buf
;
1881 de_buf
= bh
->b_data
;
1882 dlimit
= de_buf
+ dir
->i_sb
->s_blocksize
;
1884 while (de_buf
< dlimit
) {
1885 /* this code is executed quadratically often */
1886 /* do minimal checking `by hand' */
1888 de
= (struct ocfs2_dir_entry
*) de_buf
;
1890 if (de_buf
+ namelen
<= dlimit
&&
1891 ocfs2_match(namelen
, name
, de
)) {
1892 /* found a match - just to be sure, do a full check */
1893 if (!ocfs2_check_dir_entry(dir
, de
, bh
, offset
)) {
1902 /* prevent looping on a bad block */
1903 de_len
= le16_to_cpu(de
->rec_len
);
1918 struct buffer_head
*ocfs2_find_entry(const char *name
, int namelen
,
1920 struct ocfs2_dir_entry
**res_dir
)
1922 struct super_block
*sb
;
1923 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1924 struct buffer_head
*bh
, *ret
= NULL
;
1925 unsigned long start
, block
, b
;
1926 int ra_max
= 0; /* Number of bh's in the readahead
1928 int ra_ptr
= 0; /* Current index into readahead
1931 int nblocks
, i
, err
;
1938 nblocks
= i_size_read(dir
) >> sb
->s_blocksize_bits
;
1939 start
= OCFS2_I(dir
)->ip_dir_start_lookup
;
1940 if (start
>= nblocks
)
1947 * We deal with the read-ahead logic here.
1949 if (ra_ptr
>= ra_max
) {
1950 /* Refill the readahead buffer */
1953 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
1955 * Terminate if we reach the end of the
1956 * directory and must wrap, or if our
1957 * search has finished at this block.
1959 if (b
>= nblocks
|| (num
&& block
== start
)) {
1960 bh_use
[ra_max
] = NULL
;
1965 /* XXX: questionable readahead stuff here */
1966 bh
= ocfs2_bread(dir
, b
++, &err
, 1);
1967 bh_use
[ra_max
] = bh
;
1970 ll_rw_block(READ
, 1, &bh
);
1974 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1977 if (!buffer_uptodate(bh
)) {
1978 /* read error, skip block & hope for the best */
1982 i
= ocfs2_search_dirblock(bh
, dir
, name
, namelen
,
1983 block
<< sb
->s_blocksize_bits
,
1986 OCFS2_I(dir
)->ip_dir_start_lookup
= block
;
1988 goto cleanup_and_exit
;
1992 goto cleanup_and_exit
;
1995 if (++block
>= nblocks
)
1997 } while (block
!= start
);
2000 * If the directory has grown while we were searching, then
2001 * search the last part of the directory before giving up.
2004 nblocks
= i_size_read(dir
) >> sb
->s_blocksize_bits
;
2005 if (block
< nblocks
) {
2011 /* Clean up the read-ahead blocks */
2012 for (; ra_ptr
< ra_max
; ra_ptr
++)
2013 brelse(bh_use
[ra_ptr
]);
2019 static int ocfs2_blkno_stringify(u64 blkno
, char *name
)
2021 int status
, namelen
;
2025 namelen
= snprintf(name
, OCFS2_ORPHAN_NAMELEN
+ 1, "%016llx",
2035 if (namelen
!= OCFS2_ORPHAN_NAMELEN
) {
2041 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name
,
2050 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
2051 struct ocfs2_journal_handle
*handle
,
2052 struct inode
*inode
,
2054 struct buffer_head
**de_bh
)
2056 struct inode
*orphan_dir_inode
= NULL
;
2057 struct buffer_head
*orphan_dir_bh
= NULL
;
2060 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2066 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2067 ORPHAN_DIR_SYSTEM_INODE
,
2069 if (!orphan_dir_inode
) {
2075 ocfs2_handle_add_inode(handle
, orphan_dir_inode
);
2076 status
= ocfs2_meta_lock(orphan_dir_inode
, handle
, &orphan_dir_bh
, 1);
2082 status
= ocfs2_prepare_dir_for_insert(osb
, orphan_dir_inode
,
2083 orphan_dir_bh
, name
,
2084 OCFS2_ORPHAN_NAMELEN
, de_bh
);
2091 if (orphan_dir_inode
)
2092 iput(orphan_dir_inode
);
2095 brelse(orphan_dir_bh
);
2101 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
2102 struct ocfs2_journal_handle
*handle
,
2103 struct inode
*inode
,
2104 struct ocfs2_dinode
*fe
,
2106 struct buffer_head
*de_bh
)
2108 struct inode
*orphan_dir_inode
= NULL
;
2109 struct buffer_head
*orphan_dir_bh
= NULL
;
2111 struct ocfs2_dinode
*orphan_fe
;
2113 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
2115 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2116 ORPHAN_DIR_SYSTEM_INODE
,
2118 if (!orphan_dir_inode
) {
2124 status
= ocfs2_read_block(osb
,
2125 OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2126 &orphan_dir_bh
, OCFS2_BH_CACHED
,
2133 status
= ocfs2_journal_access(handle
, orphan_dir_inode
, orphan_dir_bh
,
2134 OCFS2_JOURNAL_ACCESS_WRITE
);
2140 /* we're a cluster, and nlink can change on disk from
2141 * underneath us... */
2142 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2143 if (S_ISDIR(inode
->i_mode
))
2144 le16_add_cpu(&orphan_fe
->i_links_count
, 1);
2145 orphan_dir_inode
->i_nlink
= le16_to_cpu(orphan_fe
->i_links_count
);
2147 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2153 status
= __ocfs2_add_entry(handle
, orphan_dir_inode
, name
,
2154 OCFS2_ORPHAN_NAMELEN
, inode
,
2155 OCFS2_I(inode
)->ip_blkno
,
2156 orphan_dir_bh
, de_bh
);
2162 le32_add_cpu(&fe
->i_flags
, OCFS2_ORPHANED_FL
);
2164 /* Record which orphan dir our inode now resides
2165 * in. delete_inode will use this to determine which orphan
2167 spin_lock(&OCFS2_I(inode
)->ip_lock
);
2168 OCFS2_I(inode
)->ip_orphaned_slot
= osb
->slot_num
;
2169 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
2171 mlog(0, "Inode %llu orphaned in slot %d\n",
2172 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, osb
->slot_num
);
2175 if (orphan_dir_inode
)
2176 iput(orphan_dir_inode
);
2179 brelse(orphan_dir_bh
);
2185 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2186 int ocfs2_orphan_del(struct ocfs2_super
*osb
,
2187 struct ocfs2_journal_handle
*handle
,
2188 struct inode
*orphan_dir_inode
,
2189 struct inode
*inode
,
2190 struct buffer_head
*orphan_dir_bh
)
2192 char name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2193 struct ocfs2_dinode
*orphan_fe
;
2195 struct buffer_head
*target_de_bh
= NULL
;
2196 struct ocfs2_dir_entry
*target_de
= NULL
;
2200 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2206 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2207 name
, (unsigned long long)OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2208 OCFS2_ORPHAN_NAMELEN
);
2210 /* find it's spot in the orphan directory */
2211 target_de_bh
= ocfs2_find_entry(name
, OCFS2_ORPHAN_NAMELEN
,
2212 orphan_dir_inode
, &target_de
);
2213 if (!target_de_bh
) {
2219 /* remove it from the orphan directory */
2220 status
= ocfs2_delete_entry(handle
, orphan_dir_inode
, target_de
,
2227 status
= ocfs2_journal_access(handle
,orphan_dir_inode
, orphan_dir_bh
,
2228 OCFS2_JOURNAL_ACCESS_WRITE
);
2234 /* do the i_nlink dance! :) */
2235 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2236 if (S_ISDIR(inode
->i_mode
))
2237 le16_add_cpu(&orphan_fe
->i_links_count
, -1);
2238 orphan_dir_inode
->i_nlink
= le16_to_cpu(orphan_fe
->i_links_count
);
2240 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2248 brelse(target_de_bh
);
2254 struct inode_operations ocfs2_dir_iops
= {
2255 .create
= ocfs2_create
,
2256 .lookup
= ocfs2_lookup
,
2258 .unlink
= ocfs2_unlink
,
2259 .rmdir
= ocfs2_unlink
,
2260 .symlink
= ocfs2_symlink
,
2261 .mkdir
= ocfs2_mkdir
,
2262 .mknod
= ocfs2_mknod
,
2263 .rename
= ocfs2_rename
,
2264 .setattr
= ocfs2_setattr
,
2265 .getattr
= ocfs2_getattr
,