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 if (S_ISDIR(mode
) && (dir
->i_nlink
>= OCFS2_LINK_MAX
)) {
314 mlog(ML_ERROR
, "inode %llu has i_nlink of %u\n",
315 (unsigned long long)OCFS2_I(dir
)->ip_blkno
, dir
->i_nlink
);
320 handle
= ocfs2_alloc_handle(osb
);
321 if (handle
== NULL
) {
327 status
= ocfs2_meta_lock(dir
, handle
, &parent_fe_bh
, 1);
329 if (status
!= -ENOENT
)
334 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
335 if (!dirfe
->i_links_count
) {
336 /* can't make a file in a deleted directory. */
341 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
346 /* get a spot inside the dir. */
347 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
349 dentry
->d_name
.len
, &de_bh
);
355 /* reserve an inode spot */
356 status
= ocfs2_reserve_new_inode(osb
, handle
, &inode_ac
);
358 if (status
!= -ENOSPC
)
363 /* are we making a directory? If so, reserve a cluster for his
366 status
= ocfs2_reserve_clusters(osb
, handle
, 1, &data_ac
);
368 if (status
!= -ENOSPC
)
374 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_MKNOD_CREDITS
);
375 if (IS_ERR(handle
)) {
376 status
= PTR_ERR(handle
);
382 /* do the real work now. */
383 status
= ocfs2_mknod_locked(osb
, dir
, dentry
, mode
, dev
,
384 &new_fe_bh
, parent_fe_bh
, handle
,
392 status
= ocfs2_fill_new_dir(osb
, handle
, dir
, inode
,
399 status
= ocfs2_journal_access(handle
, dir
, parent_fe_bh
,
400 OCFS2_JOURNAL_ACCESS_WRITE
);
405 le16_add_cpu(&dirfe
->i_links_count
, 1);
406 status
= ocfs2_journal_dirty(handle
, parent_fe_bh
);
414 status
= ocfs2_add_entry(handle
, dentry
, inode
,
415 OCFS2_I(inode
)->ip_blkno
, parent_fe_bh
,
422 insert_inode_hash(inode
);
423 dentry
->d_op
= &ocfs2_dentry_ops
;
424 d_instantiate(dentry
, inode
);
428 ocfs2_commit_trans(handle
);
430 if (status
== -ENOSPC
)
431 mlog(0, "Disk is full\n");
440 brelse(parent_fe_bh
);
442 if ((status
< 0) && inode
)
446 ocfs2_free_alloc_context(inode_ac
);
449 ocfs2_free_alloc_context(data_ac
);
456 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
458 struct dentry
*dentry
, int mode
,
460 struct buffer_head
**new_fe_bh
,
461 struct buffer_head
*parent_fe_bh
,
462 struct ocfs2_journal_handle
*handle
,
463 struct inode
**ret_inode
,
464 struct ocfs2_alloc_context
*inode_ac
)
467 struct ocfs2_dinode
*fe
= NULL
;
468 struct ocfs2_extent_list
*fel
;
471 struct inode
*inode
= NULL
;
473 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
, mode
,
474 (unsigned long)dev
, dentry
->d_name
.len
,
475 dentry
->d_name
.name
);
480 status
= ocfs2_claim_new_inode(osb
, handle
, inode_ac
, &suballoc_bit
,
487 inode
= new_inode(dir
->i_sb
);
489 status
= PTR_ERR(inode
);
490 mlog(ML_ERROR
, "new_inode failed!\n");
494 /* populate as many fields early on as possible - many of
495 * these are used by the support functions here and in
497 inode
->i_ino
= ino_from_blkno(osb
->sb
, fe_blkno
);
498 OCFS2_I(inode
)->ip_blkno
= fe_blkno
;
503 inode
->i_mode
= mode
;
504 spin_lock(&osb
->osb_lock
);
505 inode
->i_generation
= osb
->s_next_generation
++;
506 spin_unlock(&osb
->osb_lock
);
508 *new_fe_bh
= sb_getblk(osb
->sb
, fe_blkno
);
514 ocfs2_set_new_buffer_uptodate(inode
, *new_fe_bh
);
516 status
= ocfs2_journal_access(handle
, inode
, *new_fe_bh
,
517 OCFS2_JOURNAL_ACCESS_CREATE
);
523 fe
= (struct ocfs2_dinode
*) (*new_fe_bh
)->b_data
;
524 memset(fe
, 0, osb
->sb
->s_blocksize
);
526 fe
->i_generation
= cpu_to_le32(inode
->i_generation
);
527 fe
->i_fs_generation
= cpu_to_le32(osb
->fs_generation
);
528 fe
->i_blkno
= cpu_to_le64(fe_blkno
);
529 fe
->i_suballoc_bit
= cpu_to_le16(suballoc_bit
);
530 fe
->i_suballoc_slot
= cpu_to_le16(osb
->slot_num
);
531 fe
->i_uid
= cpu_to_le32(current
->fsuid
);
532 if (dir
->i_mode
& S_ISGID
) {
533 fe
->i_gid
= cpu_to_le32(dir
->i_gid
);
537 fe
->i_gid
= cpu_to_le32(current
->fsgid
);
538 fe
->i_mode
= cpu_to_le16(mode
);
539 if (S_ISCHR(mode
) || S_ISBLK(mode
))
540 fe
->id1
.dev1
.i_rdev
= cpu_to_le64(huge_encode_dev(dev
));
542 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
544 fe
->i_last_eb_blk
= 0;
545 strcpy(fe
->i_signature
, OCFS2_INODE_SIGNATURE
);
546 le32_add_cpu(&fe
->i_flags
, OCFS2_VALID_FL
);
547 fe
->i_atime
= fe
->i_ctime
= fe
->i_mtime
=
548 cpu_to_le64(CURRENT_TIME
.tv_sec
);
549 fe
->i_mtime_nsec
= fe
->i_ctime_nsec
= fe
->i_atime_nsec
=
550 cpu_to_le32(CURRENT_TIME
.tv_nsec
);
553 fel
= &fe
->id2
.i_list
;
554 fel
->l_tree_depth
= 0;
555 fel
->l_next_free_rec
= 0;
556 fel
->l_count
= cpu_to_le16(ocfs2_extent_recs_per_inode(osb
->sb
));
558 status
= ocfs2_journal_dirty(handle
, *new_fe_bh
);
564 if (ocfs2_populate_inode(inode
, fe
, 1) < 0) {
565 mlog(ML_ERROR
, "populate inode failed! bh->b_blocknr=%llu, "
566 "i_blkno=%llu, i_ino=%lu\n",
567 (unsigned long long) (*new_fe_bh
)->b_blocknr
,
568 (unsigned long long)fe
->i_blkno
, inode
->i_ino
);
572 ocfs2_inode_set_new(osb
, inode
);
573 status
= ocfs2_create_new_inode_locks(inode
);
577 status
= 0; /* error in ocfs2_create_new_inode_locks is not
595 static int ocfs2_mkdir(struct inode
*dir
,
596 struct dentry
*dentry
,
601 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
602 dentry
->d_name
.len
, dentry
->d_name
.name
);
603 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
609 static int ocfs2_create(struct inode
*dir
,
610 struct dentry
*dentry
,
612 struct nameidata
*nd
)
616 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
617 dentry
->d_name
.len
, dentry
->d_name
.name
);
618 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
624 static int ocfs2_link(struct dentry
*old_dentry
,
626 struct dentry
*dentry
)
628 struct ocfs2_journal_handle
*handle
= NULL
;
629 struct inode
*inode
= old_dentry
->d_inode
;
631 struct buffer_head
*fe_bh
= NULL
;
632 struct buffer_head
*parent_fe_bh
= NULL
;
633 struct buffer_head
*de_bh
= NULL
;
634 struct ocfs2_dinode
*fe
= NULL
;
635 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
637 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode
->i_ino
,
638 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
639 dentry
->d_name
.len
, dentry
->d_name
.name
);
641 if (S_ISDIR(inode
->i_mode
)) {
646 if (inode
->i_nlink
>= OCFS2_LINK_MAX
) {
651 handle
= ocfs2_alloc_handle(osb
);
652 if (handle
== NULL
) {
657 err
= ocfs2_meta_lock(dir
, handle
, &parent_fe_bh
, 1);
664 err
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
669 err
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
671 dentry
->d_name
.len
, &de_bh
);
677 err
= ocfs2_meta_lock(inode
, handle
, &fe_bh
, 1);
684 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
685 if (le16_to_cpu(fe
->i_links_count
) >= OCFS2_LINK_MAX
) {
690 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_LINK_CREDITS
);
691 if (IS_ERR(handle
)) {
692 err
= PTR_ERR(handle
);
698 err
= ocfs2_journal_access(handle
, inode
, fe_bh
,
699 OCFS2_JOURNAL_ACCESS_WRITE
);
706 inode
->i_ctime
= CURRENT_TIME
;
707 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
708 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
709 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
711 err
= ocfs2_journal_dirty(handle
, fe_bh
);
713 le16_add_cpu(&fe
->i_links_count
, -1);
719 err
= ocfs2_add_entry(handle
, dentry
, inode
,
720 OCFS2_I(inode
)->ip_blkno
,
721 parent_fe_bh
, de_bh
);
723 le16_add_cpu(&fe
->i_links_count
, -1);
729 atomic_inc(&inode
->i_count
);
730 dentry
->d_op
= &ocfs2_dentry_ops
;
731 d_instantiate(dentry
, inode
);
734 ocfs2_commit_trans(handle
);
740 brelse(parent_fe_bh
);
747 static int ocfs2_unlink(struct inode
*dir
,
748 struct dentry
*dentry
)
751 unsigned int saved_nlink
= 0;
752 struct inode
*inode
= dentry
->d_inode
;
753 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
755 struct ocfs2_dinode
*fe
= NULL
;
756 struct buffer_head
*fe_bh
= NULL
;
757 struct buffer_head
*parent_node_bh
= NULL
;
758 struct ocfs2_journal_handle
*handle
= NULL
;
759 struct ocfs2_dir_entry
*dirent
= NULL
;
760 struct buffer_head
*dirent_bh
= NULL
;
761 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
762 struct buffer_head
*orphan_entry_bh
= NULL
;
764 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
765 dentry
->d_name
.len
, dentry
->d_name
.name
);
767 BUG_ON(dentry
->d_parent
->d_inode
!= dir
);
769 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
771 if (inode
== osb
->root_inode
) {
772 mlog(0, "Cannot delete the root directory\n");
777 handle
= ocfs2_alloc_handle(osb
);
778 if (handle
== NULL
) {
784 status
= ocfs2_meta_lock(dir
, handle
, &parent_node_bh
, 1);
786 if (status
!= -ENOENT
)
791 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
792 dentry
->d_name
.len
, &blkno
,
793 dir
, &dirent_bh
, &dirent
);
795 if (status
!= -ENOENT
)
800 if (OCFS2_I(inode
)->ip_blkno
!= blkno
) {
803 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
804 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
805 (unsigned long long)blkno
, OCFS2_I(inode
)->ip_flags
);
809 status
= ocfs2_meta_lock(inode
, handle
, &fe_bh
, 1);
811 if (status
!= -ENOENT
)
816 if (S_ISDIR(inode
->i_mode
)) {
817 if (!ocfs2_empty_dir(inode
)) {
820 } else if (inode
->i_nlink
!= 2) {
826 /* There are still a few steps left until we can consider the
827 * unlink to have succeeded. Save off nlink here before
828 * modification so we can set it back in case we hit an issue
830 saved_nlink
= inode
->i_nlink
;
831 if (S_ISDIR(inode
->i_mode
))
836 status
= ocfs2_request_unlink_vote(inode
, dentry
,
837 (unsigned int) inode
->i_nlink
);
839 /* This vote should succeed under all normal
845 if (!inode
->i_nlink
) {
846 status
= ocfs2_prepare_orphan_dir(osb
, handle
, inode
,
855 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_UNLINK_CREDITS
);
856 if (IS_ERR(handle
)) {
857 status
= PTR_ERR(handle
);
863 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
864 OCFS2_JOURNAL_ACCESS_WRITE
);
870 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
872 if (!inode
->i_nlink
) {
873 status
= ocfs2_orphan_add(osb
, handle
, inode
, fe
, orphan_name
,
881 /* delete the name from the parent dir */
882 status
= ocfs2_delete_entry(handle
, dir
, dirent
, dirent_bh
);
888 /* We can set nlink on the dinode now. clear the saved version
889 * so that it doesn't get set later. */
890 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
893 status
= ocfs2_journal_dirty(handle
, fe_bh
);
899 if (S_ISDIR(inode
->i_mode
)) {
901 status
= ocfs2_mark_inode_dirty(handle
, dir
,
910 if (status
< 0 && saved_nlink
)
911 inode
->i_nlink
= saved_nlink
;
914 ocfs2_commit_trans(handle
);
923 brelse(parent_node_bh
);
926 brelse(orphan_entry_bh
);
934 * The only place this should be used is rename!
935 * if they have the same id, then the 1st one is the only one locked.
937 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
938 struct ocfs2_journal_handle
*handle
,
939 struct buffer_head
**bh1
,
940 struct inode
*inode1
,
941 struct buffer_head
**bh2
,
942 struct inode
*inode2
)
945 struct ocfs2_inode_info
*oi1
= OCFS2_I(inode1
);
946 struct ocfs2_inode_info
*oi2
= OCFS2_I(inode2
);
947 struct buffer_head
**tmpbh
;
948 struct inode
*tmpinode
;
950 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
951 (unsigned long long)oi1
->ip_blkno
,
952 (unsigned long long)oi2
->ip_blkno
);
961 /* we always want to lock the one with the lower lockid first. */
962 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
963 if (oi1
->ip_blkno
< oi2
->ip_blkno
) {
964 /* switch id1 and id2 around */
965 mlog(0, "switching them around...\n");
975 status
= ocfs2_meta_lock(inode2
, handle
, bh2
, 1);
977 if (status
!= -ENOENT
)
983 status
= ocfs2_meta_lock(inode1
, handle
, bh1
, 1);
985 if (status
!= -ENOENT
)
994 #define PARENT_INO(buffer) \
995 ((struct ocfs2_dir_entry *) \
997 le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
999 static int ocfs2_rename(struct inode
*old_dir
,
1000 struct dentry
*old_dentry
,
1001 struct inode
*new_dir
,
1002 struct dentry
*new_dentry
)
1004 int status
= 0, rename_lock
= 0;
1005 struct inode
*old_inode
= old_dentry
->d_inode
;
1006 struct inode
*new_inode
= new_dentry
->d_inode
;
1007 struct ocfs2_dinode
*newfe
= NULL
;
1008 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1009 struct buffer_head
*orphan_entry_bh
= NULL
;
1010 struct buffer_head
*newfe_bh
= NULL
;
1011 struct buffer_head
*insert_entry_bh
= NULL
;
1012 struct ocfs2_super
*osb
= NULL
;
1014 struct ocfs2_journal_handle
*handle
= NULL
;
1015 struct buffer_head
*old_dir_bh
= NULL
;
1016 struct buffer_head
*new_dir_bh
= NULL
;
1017 struct ocfs2_dir_entry
*old_de
= NULL
, *new_de
= NULL
; // dirent for old_dentry
1019 struct buffer_head
*new_de_bh
= NULL
, *old_de_bh
= NULL
; // bhs for above
1020 struct buffer_head
*old_inode_de_bh
= NULL
; // if old_dentry is a dir,
1021 // this is the 1st dirent bh
1022 nlink_t old_dir_nlink
= old_dir
->i_nlink
, new_dir_nlink
= new_dir
->i_nlink
;
1023 unsigned int links_count
;
1025 /* At some point it might be nice to break this function up a
1028 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1029 old_dir
, old_dentry
, new_dir
, new_dentry
,
1030 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
1031 new_dentry
->d_name
.len
, new_dentry
->d_name
.name
);
1033 osb
= OCFS2_SB(old_dir
->i_sb
);
1036 if (!igrab(new_inode
))
1040 if (atomic_read(&old_dentry
->d_count
) > 2) {
1041 shrink_dcache_parent(old_dentry
);
1042 if (atomic_read(&old_dentry
->d_count
) > 2) {
1048 /* Assume a directory heirarchy thusly:
1051 * a,b,c, and d are all directories.
1053 * from cwd of 'a' on both nodes:
1057 * And that's why, just like the VFS, we need a file system
1059 if (old_dentry
!= new_dentry
) {
1060 status
= ocfs2_rename_lock(osb
);
1068 handle
= ocfs2_alloc_handle(osb
);
1069 if (handle
== NULL
) {
1075 /* if old and new are the same, this'll just do one lock. */
1076 status
= ocfs2_double_lock(osb
, handle
,
1077 &old_dir_bh
, old_dir
,
1078 &new_dir_bh
, new_dir
);
1084 /* make sure both dirs have bhs
1085 * get an extra ref on old_dir_bh if old==new */
1088 new_dir_bh
= old_dir_bh
;
1091 mlog(ML_ERROR
, "no old_dir_bh!\n");
1097 if (S_ISDIR(old_inode
->i_mode
)) {
1098 /* Directories actually require metadata updates to
1099 * the directory info so we can't get away with not
1100 * doing node locking on it. */
1101 status
= ocfs2_meta_lock(old_inode
, handle
, NULL
, 1);
1103 if (status
!= -ENOENT
)
1108 status
= ocfs2_request_rename_vote(old_inode
, old_dentry
);
1115 old_inode_de_bh
= ocfs2_bread(old_inode
, 0, &status
, 0);
1116 if (!old_inode_de_bh
)
1120 if (le64_to_cpu(PARENT_INO(old_inode_de_bh
->b_data
)) !=
1121 OCFS2_I(old_dir
)->ip_blkno
)
1124 if (!new_inode
&& new_dir
!=old_dir
&&
1125 new_dir
->i_nlink
>= OCFS2_LINK_MAX
)
1128 /* Ah, the simple case - we're a file so just send a
1130 status
= ocfs2_request_rename_vote(old_inode
, old_dentry
);
1138 old_de_bh
= ocfs2_find_entry(old_dentry
->d_name
.name
,
1139 old_dentry
->d_name
.len
,
1145 * Check for inode number is _not_ due to possible IO errors.
1146 * We might rmdir the source, keep it as pwd of some process
1147 * and merrily kill the link to whatever was created under the
1148 * same name. Goodbye sticky bit ;-<
1150 if (le64_to_cpu(old_de
->inode
) != OCFS2_I(old_inode
)->ip_blkno
)
1153 /* check if the target already exists (in which case we need
1155 status
= ocfs2_find_files_on_disk(new_dentry
->d_name
.name
,
1156 new_dentry
->d_name
.len
,
1157 &newfe_blkno
, new_dir
, &new_de_bh
,
1159 /* The only error we allow here is -ENOENT because the new
1160 * file not existing is perfectly valid. */
1161 if ((status
< 0) && (status
!= -ENOENT
)) {
1162 /* If we cannot find the file specified we should just */
1163 /* return the error... */
1168 if (!new_de
&& new_inode
)
1169 mlog(ML_ERROR
, "inode %lu does not exist in it's parent "
1170 "directory!", new_inode
->i_ino
);
1172 /* In case we need to overwrite an existing file, we blow it
1175 /* VFS didn't think there existed an inode here, but
1176 * someone else in the cluster must have raced our
1177 * rename to create one. Today we error cleanly, in
1178 * the future we should consider calling iget to build
1179 * a new struct inode for this entry. */
1183 mlog(0, "We found an inode for name %.*s but VFS "
1184 "didn't give us one.\n", new_dentry
->d_name
.len
,
1185 new_dentry
->d_name
.name
);
1189 if (OCFS2_I(new_inode
)->ip_blkno
!= newfe_blkno
) {
1192 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1193 (unsigned long long)OCFS2_I(new_inode
)->ip_blkno
,
1194 (unsigned long long)newfe_blkno
,
1195 OCFS2_I(new_inode
)->ip_flags
);
1199 status
= ocfs2_meta_lock(new_inode
, handle
, &newfe_bh
, 1);
1201 if (status
!= -ENOENT
)
1206 if (S_ISDIR(new_inode
->i_mode
))
1209 links_count
= (unsigned int) (new_inode
->i_nlink
- 1);
1211 status
= ocfs2_request_unlink_vote(new_inode
, new_dentry
,
1218 newfe
= (struct ocfs2_dinode
*) newfe_bh
->b_data
;
1220 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1221 "newfebh=%p bhblocknr=%llu\n", new_de
,
1222 (unsigned long long)newfe_blkno
, newfe_bh
, newfe_bh
?
1223 (unsigned long long)newfe_bh
->b_blocknr
: 0ULL);
1225 if (S_ISDIR(new_inode
->i_mode
) || (new_inode
->i_nlink
== 1)) {
1226 status
= ocfs2_prepare_orphan_dir(osb
, handle
,
1236 BUG_ON(new_dentry
->d_parent
->d_inode
!= new_dir
);
1238 status
= ocfs2_check_dir_for_entry(new_dir
,
1239 new_dentry
->d_name
.name
,
1240 new_dentry
->d_name
.len
);
1244 status
= ocfs2_prepare_dir_for_insert(osb
, new_dir
, new_dir_bh
,
1245 new_dentry
->d_name
.name
,
1246 new_dentry
->d_name
.len
,
1254 handle
= ocfs2_start_trans(osb
, handle
, OCFS2_RENAME_CREDITS
);
1255 if (IS_ERR(handle
)) {
1256 status
= PTR_ERR(handle
);
1263 if (S_ISDIR(new_inode
->i_mode
)) {
1264 if (!ocfs2_empty_dir(new_inode
) ||
1265 new_inode
->i_nlink
!= 2) {
1266 status
= -ENOTEMPTY
;
1270 status
= ocfs2_journal_access(handle
, new_inode
, newfe_bh
,
1271 OCFS2_JOURNAL_ACCESS_WRITE
);
1277 if (S_ISDIR(new_inode
->i_mode
) ||
1278 (newfe
->i_links_count
== cpu_to_le16(1))){
1279 status
= ocfs2_orphan_add(osb
, handle
, new_inode
,
1288 /* change the dirent to point to the correct inode */
1289 status
= ocfs2_journal_access(handle
, new_dir
, new_de_bh
,
1290 OCFS2_JOURNAL_ACCESS_WRITE
);
1295 new_de
->inode
= cpu_to_le64(OCFS2_I(old_inode
)->ip_blkno
);
1296 new_de
->file_type
= old_de
->file_type
;
1297 new_dir
->i_version
++;
1298 status
= ocfs2_journal_dirty(handle
, new_de_bh
);
1304 if (S_ISDIR(new_inode
->i_mode
))
1305 newfe
->i_links_count
= 0;
1307 le16_add_cpu(&newfe
->i_links_count
, -1);
1309 status
= ocfs2_journal_dirty(handle
, newfe_bh
);
1315 /* if the name was not found in new_dir, add it now */
1316 status
= ocfs2_add_entry(handle
, new_dentry
, old_inode
,
1317 OCFS2_I(old_inode
)->ip_blkno
,
1318 new_dir_bh
, insert_entry_bh
);
1321 old_inode
->i_ctime
= CURRENT_TIME
;
1322 mark_inode_dirty(old_inode
);
1324 /* now that the name has been added to new_dir, remove the old name */
1325 status
= ocfs2_delete_entry(handle
, old_dir
, old_de
, old_de_bh
);
1332 new_inode
->i_nlink
--;
1333 new_inode
->i_ctime
= CURRENT_TIME
;
1335 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1336 if (old_inode_de_bh
) {
1337 status
= ocfs2_journal_access(handle
, old_inode
,
1339 OCFS2_JOURNAL_ACCESS_WRITE
);
1340 PARENT_INO(old_inode_de_bh
->b_data
) =
1341 cpu_to_le64(OCFS2_I(new_dir
)->ip_blkno
);
1342 status
= ocfs2_journal_dirty(handle
, old_inode_de_bh
);
1345 new_inode
->i_nlink
--;
1348 mark_inode_dirty(new_dir
);
1351 mark_inode_dirty(old_dir
);
1353 mark_inode_dirty(new_inode
);
1355 if (old_dir
!= new_dir
)
1356 if (new_dir_nlink
!= new_dir
->i_nlink
) {
1358 mlog(ML_ERROR
, "need to change nlink for new "
1359 "dir %llu from %d to %d but bh is NULL\n",
1360 (unsigned long long)OCFS2_I(new_dir
)->ip_blkno
,
1361 (int)new_dir_nlink
, new_dir
->i_nlink
);
1363 struct ocfs2_dinode
*fe
;
1364 status
= ocfs2_journal_access(handle
,
1367 OCFS2_JOURNAL_ACCESS_WRITE
);
1368 fe
= (struct ocfs2_dinode
*) new_dir_bh
->b_data
;
1369 fe
->i_links_count
= cpu_to_le16(new_dir
->i_nlink
);
1370 status
= ocfs2_journal_dirty(handle
, new_dir_bh
);
1374 if (old_dir_nlink
!= old_dir
->i_nlink
) {
1376 mlog(ML_ERROR
, "need to change nlink for old dir "
1377 "%llu from %d to %d but bh is NULL!\n",
1378 (unsigned long long)OCFS2_I(old_dir
)->ip_blkno
,
1379 (int)old_dir_nlink
, old_dir
->i_nlink
);
1381 struct ocfs2_dinode
*fe
;
1382 status
= ocfs2_journal_access(handle
, old_dir
,
1384 OCFS2_JOURNAL_ACCESS_WRITE
);
1385 fe
= (struct ocfs2_dinode
*) old_dir_bh
->b_data
;
1386 fe
->i_links_count
= cpu_to_le16(old_dir
->i_nlink
);
1387 status
= ocfs2_journal_dirty(handle
, old_dir_bh
);
1394 ocfs2_rename_unlock(osb
);
1397 ocfs2_commit_trans(handle
);
1400 sync_mapping_buffers(old_inode
->i_mapping
);
1414 if (old_inode_de_bh
)
1415 brelse(old_inode_de_bh
);
1416 if (orphan_entry_bh
)
1417 brelse(orphan_entry_bh
);
1418 if (insert_entry_bh
)
1419 brelse(insert_entry_bh
);
1427 * we expect i_size = strlen(symname). Copy symname into the file
1428 * data, including the null terminator.
1430 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
1431 struct ocfs2_journal_handle
*handle
,
1432 struct inode
*inode
,
1433 const char *symname
)
1435 struct buffer_head
**bhs
= NULL
;
1437 struct super_block
*sb
= osb
->sb
;
1440 int virtual, blocks
, status
, i
, bytes_left
;
1442 bytes_left
= i_size_read(inode
) + 1;
1443 /* we can't trust i_blocks because we're actually going to
1444 * write i_size + 1 bytes. */
1445 blocks
= (bytes_left
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
1447 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1448 (unsigned long long)inode
->i_blocks
,
1449 i_size_read(inode
), blocks
);
1451 /* Sanity check -- make sure we're going to fit. */
1453 ocfs2_clusters_to_bytes(sb
, OCFS2_I(inode
)->ip_clusters
)) {
1459 bhs
= kcalloc(blocks
, sizeof(struct buffer_head
*), GFP_KERNEL
);
1466 status
= ocfs2_extent_map_get_blocks(inode
, 0, 1, &p_blkno
,
1473 /* links can never be larger than one cluster so we know this
1474 * is all going to be contiguous, but do a sanity check
1476 if ((p_blocks
<< sb
->s_blocksize_bits
) < bytes_left
) {
1483 while(bytes_left
> 0) {
1484 c
= &symname
[virtual * sb
->s_blocksize
];
1486 bhs
[virtual] = sb_getblk(sb
, p_blkno
);
1487 if (!bhs
[virtual]) {
1492 ocfs2_set_new_buffer_uptodate(inode
, bhs
[virtual]);
1494 status
= ocfs2_journal_access(handle
, inode
, bhs
[virtual],
1495 OCFS2_JOURNAL_ACCESS_CREATE
);
1501 memset(bhs
[virtual]->b_data
, 0, sb
->s_blocksize
);
1503 memcpy(bhs
[virtual]->b_data
, c
,
1504 (bytes_left
> sb
->s_blocksize
) ? sb
->s_blocksize
:
1507 status
= ocfs2_journal_dirty(handle
, bhs
[virtual]);
1515 bytes_left
-= sb
->s_blocksize
;
1522 for(i
= 0; i
< blocks
; i
++)
1532 static int ocfs2_symlink(struct inode
*dir
,
1533 struct dentry
*dentry
,
1534 const char *symname
)
1536 int status
, l
, credits
;
1538 struct ocfs2_super
*osb
= NULL
;
1539 struct inode
*inode
= NULL
;
1540 struct super_block
*sb
;
1541 struct buffer_head
*new_fe_bh
= NULL
;
1542 struct buffer_head
*de_bh
= NULL
;
1543 struct buffer_head
*parent_fe_bh
= NULL
;
1544 struct ocfs2_dinode
*fe
= NULL
;
1545 struct ocfs2_dinode
*dirfe
;
1546 struct ocfs2_journal_handle
*handle
= NULL
;
1547 struct ocfs2_alloc_context
*inode_ac
= NULL
;
1548 struct ocfs2_alloc_context
*data_ac
= NULL
;
1550 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir
,
1551 dentry
, symname
, dentry
->d_name
.len
, dentry
->d_name
.name
);
1556 l
= strlen(symname
) + 1;
1558 credits
= ocfs2_calc_symlink_credits(sb
);
1560 handle
= ocfs2_alloc_handle(osb
);
1561 if (handle
== NULL
) {
1567 /* lock the parent directory */
1568 status
= ocfs2_meta_lock(dir
, handle
, &parent_fe_bh
, 1);
1570 if (status
!= -ENOENT
)
1575 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
1576 if (!dirfe
->i_links_count
) {
1577 /* can't make a file in a deleted directory. */
1582 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
1583 dentry
->d_name
.len
);
1587 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
1588 dentry
->d_name
.name
,
1589 dentry
->d_name
.len
, &de_bh
);
1595 status
= ocfs2_reserve_new_inode(osb
, handle
, &inode_ac
);
1597 if (status
!= -ENOSPC
)
1602 /* don't reserve bitmap space for fast symlinks. */
1603 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1604 status
= ocfs2_reserve_clusters(osb
, handle
, 1, &data_ac
);
1606 if (status
!= -ENOSPC
)
1612 handle
= ocfs2_start_trans(osb
, handle
, credits
);
1613 if (IS_ERR(handle
)) {
1614 status
= PTR_ERR(handle
);
1620 status
= ocfs2_mknod_locked(osb
, dir
, dentry
,
1621 S_IFLNK
| S_IRWXUGO
, 0,
1622 &new_fe_bh
, parent_fe_bh
, handle
,
1629 fe
= (struct ocfs2_dinode
*) new_fe_bh
->b_data
;
1632 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1633 inode
->i_op
= &ocfs2_symlink_inode_operations
;
1634 status
= ocfs2_do_extend_allocation(osb
, inode
, 1, new_fe_bh
,
1635 handle
, data_ac
, NULL
,
1638 if (status
!= -ENOSPC
&& status
!= -EINTR
) {
1640 "Failed to extend file to %llu\n",
1641 (unsigned long long)newsize
);
1647 i_size_write(inode
, newsize
);
1648 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(newsize
);
1650 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
1651 memcpy((char *) fe
->id2
.i_symlink
, symname
, l
);
1652 i_size_write(inode
, newsize
);
1653 inode
->i_blocks
= 0;
1656 status
= ocfs2_mark_inode_dirty(handle
, inode
, new_fe_bh
);
1662 if (!ocfs2_inode_is_fast_symlink(inode
)) {
1663 status
= ocfs2_create_symlink_data(osb
, handle
, inode
,
1671 status
= ocfs2_add_entry(handle
, dentry
, inode
,
1672 le64_to_cpu(fe
->i_blkno
), parent_fe_bh
,
1679 insert_inode_hash(inode
);
1680 dentry
->d_op
= &ocfs2_dentry_ops
;
1681 d_instantiate(dentry
, inode
);
1684 ocfs2_commit_trans(handle
);
1688 brelse(parent_fe_bh
);
1692 ocfs2_free_alloc_context(inode_ac
);
1694 ocfs2_free_alloc_context(data_ac
);
1695 if ((status
< 0) && inode
)
1703 int ocfs2_check_dir_entry(struct inode
* dir
,
1704 struct ocfs2_dir_entry
* de
,
1705 struct buffer_head
* bh
,
1706 unsigned long offset
)
1708 const char *error_msg
= NULL
;
1709 const int rlen
= le16_to_cpu(de
->rec_len
);
1711 if (rlen
< OCFS2_DIR_REC_LEN(1))
1712 error_msg
= "rec_len is smaller than minimal";
1713 else if (rlen
% 4 != 0)
1714 error_msg
= "rec_len % 4 != 0";
1715 else if (rlen
< OCFS2_DIR_REC_LEN(de
->name_len
))
1716 error_msg
= "rec_len is too small for name_len";
1717 else if (((char *) de
- bh
->b_data
) + rlen
> dir
->i_sb
->s_blocksize
)
1718 error_msg
= "directory entry across blocks";
1720 if (error_msg
!= NULL
)
1721 mlog(ML_ERROR
, "bad entry in directory #%llu: %s - "
1722 "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1723 (unsigned long long)OCFS2_I(dir
)->ip_blkno
, error_msg
,
1724 offset
, (unsigned long long)le64_to_cpu(de
->inode
), rlen
,
1726 return error_msg
== NULL
? 1 : 0;
1729 /* we don't always have a dentry for what we want to add, so people
1730 * like orphan dir can call this instead.
1732 * If you pass me insert_bh, I'll skip the search of the other dir
1733 * blocks and put the record in there.
1735 static int __ocfs2_add_entry(struct ocfs2_journal_handle
*handle
,
1737 const char *name
, int namelen
,
1738 struct inode
*inode
, u64 blkno
,
1739 struct buffer_head
*parent_fe_bh
,
1740 struct buffer_head
*insert_bh
)
1742 unsigned long offset
;
1743 unsigned short rec_len
;
1744 struct ocfs2_dir_entry
*de
, *de1
;
1745 struct super_block
*sb
;
1755 rec_len
= OCFS2_DIR_REC_LEN(namelen
);
1757 de
= (struct ocfs2_dir_entry
*) insert_bh
->b_data
;
1759 BUG_ON((char *)de
>= sb
->s_blocksize
+ insert_bh
->b_data
);
1760 /* These checks should've already been passed by the
1761 * prepare function, but I guess we can leave them
1763 if (!ocfs2_check_dir_entry(dir
, de
, insert_bh
, offset
)) {
1767 if (ocfs2_match(namelen
, name
, de
)) {
1771 if (((le64_to_cpu(de
->inode
) == 0) &&
1772 (le16_to_cpu(de
->rec_len
) >= rec_len
)) ||
1773 (le16_to_cpu(de
->rec_len
) >=
1774 (OCFS2_DIR_REC_LEN(de
->name_len
) + rec_len
))) {
1775 status
= ocfs2_journal_access(handle
, dir
, insert_bh
,
1776 OCFS2_JOURNAL_ACCESS_WRITE
);
1777 /* By now the buffer is marked for journaling */
1778 offset
+= le16_to_cpu(de
->rec_len
);
1779 if (le64_to_cpu(de
->inode
)) {
1780 de1
= (struct ocfs2_dir_entry
*)((char *) de
+
1781 OCFS2_DIR_REC_LEN(de
->name_len
));
1783 cpu_to_le16(le16_to_cpu(de
->rec_len
) -
1784 OCFS2_DIR_REC_LEN(de
->name_len
));
1785 de
->rec_len
= cpu_to_le16(OCFS2_DIR_REC_LEN(de
->name_len
));
1788 de
->file_type
= OCFS2_FT_UNKNOWN
;
1790 de
->inode
= cpu_to_le64(blkno
);
1791 ocfs2_set_de_type(de
, inode
->i_mode
);
1794 de
->name_len
= namelen
;
1795 memcpy(de
->name
, name
, namelen
);
1797 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
1799 status
= ocfs2_journal_dirty(handle
, insert_bh
);
1803 offset
+= le16_to_cpu(de
->rec_len
);
1804 de
= (struct ocfs2_dir_entry
*) ((char *) de
+ le16_to_cpu(de
->rec_len
));
1807 /* when you think about it, the assert above should prevent us
1808 * from ever getting here. */
1818 * ocfs2_delete_entry deletes a directory entry by merging it with the
1821 static int ocfs2_delete_entry(struct ocfs2_journal_handle
*handle
,
1823 struct ocfs2_dir_entry
*de_del
,
1824 struct buffer_head
*bh
)
1826 struct ocfs2_dir_entry
*de
, *pde
;
1827 int i
, status
= -ENOENT
;
1829 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle
, dir
, de_del
, bh
);
1833 de
= (struct ocfs2_dir_entry
*) bh
->b_data
;
1834 while (i
< bh
->b_size
) {
1835 if (!ocfs2_check_dir_entry(dir
, de
, bh
, i
)) {
1841 status
= ocfs2_journal_access(handle
, dir
, bh
,
1842 OCFS2_JOURNAL_ACCESS_WRITE
);
1850 cpu_to_le16(le16_to_cpu(pde
->rec_len
) +
1851 le16_to_cpu(de
->rec_len
));
1855 status
= ocfs2_journal_dirty(handle
, bh
);
1858 i
+= le16_to_cpu(de
->rec_len
);
1860 de
= (struct ocfs2_dir_entry
*)((char *)de
+ le16_to_cpu(de
->rec_len
));
1868 * Returns 0 if not found, -1 on failure, and 1 on success
1870 static int inline ocfs2_search_dirblock(struct buffer_head
*bh
,
1872 const char *name
, int namelen
,
1873 unsigned long offset
,
1874 struct ocfs2_dir_entry
**res_dir
)
1876 struct ocfs2_dir_entry
*de
;
1877 char *dlimit
, *de_buf
;
1883 de_buf
= bh
->b_data
;
1884 dlimit
= de_buf
+ dir
->i_sb
->s_blocksize
;
1886 while (de_buf
< dlimit
) {
1887 /* this code is executed quadratically often */
1888 /* do minimal checking `by hand' */
1890 de
= (struct ocfs2_dir_entry
*) de_buf
;
1892 if (de_buf
+ namelen
<= dlimit
&&
1893 ocfs2_match(namelen
, name
, de
)) {
1894 /* found a match - just to be sure, do a full check */
1895 if (!ocfs2_check_dir_entry(dir
, de
, bh
, offset
)) {
1904 /* prevent looping on a bad block */
1905 de_len
= le16_to_cpu(de
->rec_len
);
1920 struct buffer_head
*ocfs2_find_entry(const char *name
, int namelen
,
1922 struct ocfs2_dir_entry
**res_dir
)
1924 struct super_block
*sb
;
1925 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1926 struct buffer_head
*bh
, *ret
= NULL
;
1927 unsigned long start
, block
, b
;
1928 int ra_max
= 0; /* Number of bh's in the readahead
1930 int ra_ptr
= 0; /* Current index into readahead
1933 int nblocks
, i
, err
;
1940 nblocks
= i_size_read(dir
) >> sb
->s_blocksize_bits
;
1941 start
= OCFS2_I(dir
)->ip_dir_start_lookup
;
1942 if (start
>= nblocks
)
1949 * We deal with the read-ahead logic here.
1951 if (ra_ptr
>= ra_max
) {
1952 /* Refill the readahead buffer */
1955 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
1957 * Terminate if we reach the end of the
1958 * directory and must wrap, or if our
1959 * search has finished at this block.
1961 if (b
>= nblocks
|| (num
&& block
== start
)) {
1962 bh_use
[ra_max
] = NULL
;
1967 /* XXX: questionable readahead stuff here */
1968 bh
= ocfs2_bread(dir
, b
++, &err
, 1);
1969 bh_use
[ra_max
] = bh
;
1972 ll_rw_block(READ
, 1, &bh
);
1976 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1979 if (!buffer_uptodate(bh
)) {
1980 /* read error, skip block & hope for the best */
1984 i
= ocfs2_search_dirblock(bh
, dir
, name
, namelen
,
1985 block
<< sb
->s_blocksize_bits
,
1988 OCFS2_I(dir
)->ip_dir_start_lookup
= block
;
1990 goto cleanup_and_exit
;
1994 goto cleanup_and_exit
;
1997 if (++block
>= nblocks
)
1999 } while (block
!= start
);
2002 * If the directory has grown while we were searching, then
2003 * search the last part of the directory before giving up.
2006 nblocks
= i_size_read(dir
) >> sb
->s_blocksize_bits
;
2007 if (block
< nblocks
) {
2013 /* Clean up the read-ahead blocks */
2014 for (; ra_ptr
< ra_max
; ra_ptr
++)
2015 brelse(bh_use
[ra_ptr
]);
2021 static int ocfs2_blkno_stringify(u64 blkno
, char *name
)
2023 int status
, namelen
;
2027 namelen
= snprintf(name
, OCFS2_ORPHAN_NAMELEN
+ 1, "%016llx",
2037 if (namelen
!= OCFS2_ORPHAN_NAMELEN
) {
2043 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name
,
2052 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
2053 struct ocfs2_journal_handle
*handle
,
2054 struct inode
*inode
,
2056 struct buffer_head
**de_bh
)
2058 struct inode
*orphan_dir_inode
= NULL
;
2059 struct buffer_head
*orphan_dir_bh
= NULL
;
2062 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2068 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2069 ORPHAN_DIR_SYSTEM_INODE
,
2071 if (!orphan_dir_inode
) {
2077 ocfs2_handle_add_inode(handle
, orphan_dir_inode
);
2078 status
= ocfs2_meta_lock(orphan_dir_inode
, handle
, &orphan_dir_bh
, 1);
2084 status
= ocfs2_prepare_dir_for_insert(osb
, orphan_dir_inode
,
2085 orphan_dir_bh
, name
,
2086 OCFS2_ORPHAN_NAMELEN
, de_bh
);
2093 if (orphan_dir_inode
)
2094 iput(orphan_dir_inode
);
2097 brelse(orphan_dir_bh
);
2103 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
2104 struct ocfs2_journal_handle
*handle
,
2105 struct inode
*inode
,
2106 struct ocfs2_dinode
*fe
,
2108 struct buffer_head
*de_bh
)
2110 struct inode
*orphan_dir_inode
= NULL
;
2111 struct buffer_head
*orphan_dir_bh
= NULL
;
2113 struct ocfs2_dinode
*orphan_fe
;
2115 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
2117 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2118 ORPHAN_DIR_SYSTEM_INODE
,
2120 if (!orphan_dir_inode
) {
2126 status
= ocfs2_read_block(osb
,
2127 OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2128 &orphan_dir_bh
, OCFS2_BH_CACHED
,
2135 status
= ocfs2_journal_access(handle
, orphan_dir_inode
, orphan_dir_bh
,
2136 OCFS2_JOURNAL_ACCESS_WRITE
);
2142 /* we're a cluster, and nlink can change on disk from
2143 * underneath us... */
2144 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2145 if (S_ISDIR(inode
->i_mode
))
2146 le16_add_cpu(&orphan_fe
->i_links_count
, 1);
2147 orphan_dir_inode
->i_nlink
= le16_to_cpu(orphan_fe
->i_links_count
);
2149 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2155 status
= __ocfs2_add_entry(handle
, orphan_dir_inode
, name
,
2156 OCFS2_ORPHAN_NAMELEN
, inode
,
2157 OCFS2_I(inode
)->ip_blkno
,
2158 orphan_dir_bh
, de_bh
);
2164 le32_add_cpu(&fe
->i_flags
, OCFS2_ORPHANED_FL
);
2166 /* Record which orphan dir our inode now resides
2167 * in. delete_inode will use this to determine which orphan
2169 spin_lock(&OCFS2_I(inode
)->ip_lock
);
2170 OCFS2_I(inode
)->ip_orphaned_slot
= osb
->slot_num
;
2171 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
2173 mlog(0, "Inode %llu orphaned in slot %d\n",
2174 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, osb
->slot_num
);
2177 if (orphan_dir_inode
)
2178 iput(orphan_dir_inode
);
2181 brelse(orphan_dir_bh
);
2187 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2188 int ocfs2_orphan_del(struct ocfs2_super
*osb
,
2189 struct ocfs2_journal_handle
*handle
,
2190 struct inode
*orphan_dir_inode
,
2191 struct inode
*inode
,
2192 struct buffer_head
*orphan_dir_bh
)
2194 char name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2195 struct ocfs2_dinode
*orphan_fe
;
2197 struct buffer_head
*target_de_bh
= NULL
;
2198 struct ocfs2_dir_entry
*target_de
= NULL
;
2202 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2208 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2209 name
, (unsigned long long)OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2210 OCFS2_ORPHAN_NAMELEN
);
2212 /* find it's spot in the orphan directory */
2213 target_de_bh
= ocfs2_find_entry(name
, OCFS2_ORPHAN_NAMELEN
,
2214 orphan_dir_inode
, &target_de
);
2215 if (!target_de_bh
) {
2221 /* remove it from the orphan directory */
2222 status
= ocfs2_delete_entry(handle
, orphan_dir_inode
, target_de
,
2229 status
= ocfs2_journal_access(handle
,orphan_dir_inode
, orphan_dir_bh
,
2230 OCFS2_JOURNAL_ACCESS_WRITE
);
2236 /* do the i_nlink dance! :) */
2237 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2238 if (S_ISDIR(inode
->i_mode
))
2239 le16_add_cpu(&orphan_fe
->i_links_count
, -1);
2240 orphan_dir_inode
->i_nlink
= le16_to_cpu(orphan_fe
->i_links_count
);
2242 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2250 brelse(target_de_bh
);
2256 struct inode_operations ocfs2_dir_iops
= {
2257 .create
= ocfs2_create
,
2258 .lookup
= ocfs2_lookup
,
2260 .unlink
= ocfs2_unlink
,
2261 .rmdir
= ocfs2_unlink
,
2262 .symlink
= ocfs2_symlink
,
2263 .mkdir
= ocfs2_mkdir
,
2264 .mknod
= ocfs2_mknod
,
2265 .rename
= ocfs2_rename
,
2266 .setattr
= ocfs2_setattr
,
2267 .getattr
= ocfs2_getattr
,