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(handle_t
*handle
,
80 struct ocfs2_dir_entry
*de_del
,
81 struct buffer_head
*bh
);
83 static int __ocfs2_add_entry(handle_t
*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
,
97 struct inode
**ret_inode
,
98 struct ocfs2_alloc_context
*inode_ac
);
100 static int ocfs2_fill_new_dir(struct ocfs2_super
*osb
,
102 struct inode
*parent
,
104 struct buffer_head
*fe_bh
,
105 struct ocfs2_alloc_context
*data_ac
);
107 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
108 struct inode
**ret_orphan_dir
,
111 struct buffer_head
**de_bh
);
113 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
116 struct ocfs2_dinode
*fe
,
118 struct buffer_head
*de_bh
,
119 struct inode
*orphan_dir_inode
);
121 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
124 const char *symname
);
126 static inline int ocfs2_add_entry(handle_t
*handle
,
127 struct dentry
*dentry
,
128 struct inode
*inode
, u64 blkno
,
129 struct buffer_head
*parent_fe_bh
,
130 struct buffer_head
*insert_bh
)
132 return __ocfs2_add_entry(handle
, dentry
->d_parent
->d_inode
,
133 dentry
->d_name
.name
, dentry
->d_name
.len
,
134 inode
, blkno
, parent_fe_bh
, insert_bh
);
137 /* An orphan dir name is an 8 byte value, printed as a hex string */
138 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
140 static struct dentry
*ocfs2_lookup(struct inode
*dir
, struct dentry
*dentry
,
141 struct nameidata
*nd
)
145 struct buffer_head
*dirent_bh
= NULL
;
146 struct inode
*inode
= NULL
;
148 struct ocfs2_dir_entry
*dirent
;
149 struct ocfs2_inode_info
*oi
;
151 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
152 dentry
->d_name
.len
, dentry
->d_name
.name
);
154 if (dentry
->d_name
.len
> OCFS2_MAX_FILENAME_LEN
) {
155 ret
= ERR_PTR(-ENAMETOOLONG
);
159 mlog(0, "find name %.*s in directory %llu\n", dentry
->d_name
.len
,
160 dentry
->d_name
.name
, (unsigned long long)OCFS2_I(dir
)->ip_blkno
);
162 status
= ocfs2_meta_lock(dir
, NULL
, 0);
164 if (status
!= -ENOENT
)
166 ret
= ERR_PTR(status
);
170 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
171 dentry
->d_name
.len
, &blkno
,
172 dir
, &dirent_bh
, &dirent
);
176 inode
= ocfs2_iget(OCFS2_SB(dir
->i_sb
), blkno
, 0);
178 mlog(ML_ERROR
, "Unable to create inode %llu\n",
179 (unsigned long long)blkno
);
180 ret
= ERR_PTR(-EACCES
);
185 /* Clear any orphaned state... If we were able to look up the
186 * inode from a directory, it certainly can't be orphaned. We
187 * might have the bad state from a node which intended to
188 * orphan this inode but crashed before it could commit the
190 spin_lock(&oi
->ip_lock
);
191 oi
->ip_flags
&= ~OCFS2_INODE_MAYBE_ORPHANED
;
192 oi
->ip_orphaned_slot
= OCFS2_INVALID_SLOT
;
193 spin_unlock(&oi
->ip_lock
);
196 dentry
->d_op
= &ocfs2_dentry_ops
;
197 ret
= d_splice_alias(inode
, dentry
);
201 * If d_splice_alias() finds a DCACHE_DISCONNECTED
202 * dentry, it will d_move() it on top of ourse. The
203 * return value will indicate this however, so in
204 * those cases, we switch them around for the locking
207 * NOTE: This dentry already has ->d_op set from
208 * ocfs2_get_parent() and ocfs2_get_dentry()
213 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
214 OCFS2_I(dir
)->ip_blkno
);
217 ret
= ERR_PTR(status
);
223 /* Don't drop the cluster lock until *after* the d_add --
224 * unlink on another node will message us to remove that
225 * dentry under this lock so otherwise we can race this with
226 * the vote thread and have a stale dentry. */
227 ocfs2_meta_unlock(dir
, 0);
238 static int ocfs2_fill_new_dir(struct ocfs2_super
*osb
,
240 struct inode
*parent
,
242 struct buffer_head
*fe_bh
,
243 struct ocfs2_alloc_context
*data_ac
)
246 struct buffer_head
*new_bh
= NULL
;
247 struct ocfs2_dir_entry
*de
= NULL
;
251 status
= ocfs2_do_extend_dir(osb
->sb
, handle
, inode
, fe_bh
,
252 data_ac
, NULL
, &new_bh
);
258 ocfs2_set_new_buffer_uptodate(inode
, new_bh
);
260 status
= ocfs2_journal_access(handle
, inode
, new_bh
,
261 OCFS2_JOURNAL_ACCESS_CREATE
);
266 memset(new_bh
->b_data
, 0, osb
->sb
->s_blocksize
);
268 de
= (struct ocfs2_dir_entry
*) new_bh
->b_data
;
269 de
->inode
= cpu_to_le64(OCFS2_I(inode
)->ip_blkno
);
272 cpu_to_le16(OCFS2_DIR_REC_LEN(de
->name_len
));
273 strcpy(de
->name
, ".");
274 ocfs2_set_de_type(de
, S_IFDIR
);
275 de
= (struct ocfs2_dir_entry
*) ((char *)de
+ le16_to_cpu(de
->rec_len
));
276 de
->inode
= cpu_to_le64(OCFS2_I(parent
)->ip_blkno
);
277 de
->rec_len
= cpu_to_le16(inode
->i_sb
->s_blocksize
-
278 OCFS2_DIR_REC_LEN(1));
280 strcpy(de
->name
, "..");
281 ocfs2_set_de_type(de
, S_IFDIR
);
283 status
= ocfs2_journal_dirty(handle
, new_bh
);
289 i_size_write(inode
, inode
->i_sb
->s_blocksize
);
291 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(inode
->i_sb
->s_blocksize
);
292 status
= ocfs2_mark_inode_dirty(handle
, inode
, fe_bh
);
307 static int ocfs2_mknod(struct inode
*dir
,
308 struct dentry
*dentry
,
313 struct buffer_head
*parent_fe_bh
= NULL
;
314 handle_t
*handle
= NULL
;
315 struct ocfs2_super
*osb
;
316 struct ocfs2_dinode
*dirfe
;
317 struct buffer_head
*new_fe_bh
= NULL
;
318 struct buffer_head
*de_bh
= NULL
;
319 struct inode
*inode
= NULL
;
320 struct ocfs2_alloc_context
*inode_ac
= NULL
;
321 struct ocfs2_alloc_context
*data_ac
= NULL
;
323 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
, mode
,
324 (unsigned long)dev
, dentry
->d_name
.len
,
325 dentry
->d_name
.name
);
327 /* get our super block */
328 osb
= OCFS2_SB(dir
->i_sb
);
330 status
= ocfs2_meta_lock(dir
, &parent_fe_bh
, 1);
332 if (status
!= -ENOENT
)
337 if (S_ISDIR(mode
) && (dir
->i_nlink
>= OCFS2_LINK_MAX
)) {
342 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
343 if (!dirfe
->i_links_count
) {
344 /* can't make a file in a deleted directory. */
349 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
354 /* get a spot inside the dir. */
355 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
357 dentry
->d_name
.len
, &de_bh
);
363 /* reserve an inode spot */
364 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
366 if (status
!= -ENOSPC
)
371 /* are we making a directory? If so, reserve a cluster for his
374 status
= ocfs2_reserve_clusters(osb
, 1, &data_ac
);
376 if (status
!= -ENOSPC
)
382 handle
= ocfs2_start_trans(osb
, OCFS2_MKNOD_CREDITS
);
383 if (IS_ERR(handle
)) {
384 status
= PTR_ERR(handle
);
390 /* do the real work now. */
391 status
= ocfs2_mknod_locked(osb
, dir
, dentry
, mode
, dev
,
392 &new_fe_bh
, parent_fe_bh
, handle
,
400 status
= ocfs2_fill_new_dir(osb
, handle
, dir
, inode
,
407 status
= ocfs2_journal_access(handle
, dir
, parent_fe_bh
,
408 OCFS2_JOURNAL_ACCESS_WRITE
);
413 le16_add_cpu(&dirfe
->i_links_count
, 1);
414 status
= ocfs2_journal_dirty(handle
, parent_fe_bh
);
422 status
= ocfs2_add_entry(handle
, dentry
, inode
,
423 OCFS2_I(inode
)->ip_blkno
, parent_fe_bh
,
430 status
= ocfs2_dentry_attach_lock(dentry
, inode
,
431 OCFS2_I(dir
)->ip_blkno
);
437 insert_inode_hash(inode
);
438 dentry
->d_op
= &ocfs2_dentry_ops
;
439 d_instantiate(dentry
, inode
);
443 ocfs2_commit_trans(osb
, handle
);
445 ocfs2_meta_unlock(dir
, 1);
447 if (status
== -ENOSPC
)
448 mlog(0, "Disk is full\n");
457 brelse(parent_fe_bh
);
459 if ((status
< 0) && inode
)
463 ocfs2_free_alloc_context(inode_ac
);
466 ocfs2_free_alloc_context(data_ac
);
473 static int ocfs2_mknod_locked(struct ocfs2_super
*osb
,
475 struct dentry
*dentry
, int mode
,
477 struct buffer_head
**new_fe_bh
,
478 struct buffer_head
*parent_fe_bh
,
480 struct inode
**ret_inode
,
481 struct ocfs2_alloc_context
*inode_ac
)
484 struct ocfs2_dinode
*fe
= NULL
;
485 struct ocfs2_extent_list
*fel
;
488 struct inode
*inode
= NULL
;
490 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir
, dentry
, mode
,
491 (unsigned long)dev
, dentry
->d_name
.len
,
492 dentry
->d_name
.name
);
497 status
= ocfs2_claim_new_inode(osb
, handle
, inode_ac
, &suballoc_bit
,
504 inode
= new_inode(dir
->i_sb
);
506 status
= PTR_ERR(inode
);
507 mlog(ML_ERROR
, "new_inode failed!\n");
511 /* populate as many fields early on as possible - many of
512 * these are used by the support functions here and in
514 inode
->i_ino
= ino_from_blkno(osb
->sb
, fe_blkno
);
515 OCFS2_I(inode
)->ip_blkno
= fe_blkno
;
520 inode
->i_mode
= mode
;
521 spin_lock(&osb
->osb_lock
);
522 inode
->i_generation
= osb
->s_next_generation
++;
523 spin_unlock(&osb
->osb_lock
);
525 *new_fe_bh
= sb_getblk(osb
->sb
, fe_blkno
);
531 ocfs2_set_new_buffer_uptodate(inode
, *new_fe_bh
);
533 status
= ocfs2_journal_access(handle
, inode
, *new_fe_bh
,
534 OCFS2_JOURNAL_ACCESS_CREATE
);
540 fe
= (struct ocfs2_dinode
*) (*new_fe_bh
)->b_data
;
541 memset(fe
, 0, osb
->sb
->s_blocksize
);
543 fe
->i_generation
= cpu_to_le32(inode
->i_generation
);
544 fe
->i_fs_generation
= cpu_to_le32(osb
->fs_generation
);
545 fe
->i_blkno
= cpu_to_le64(fe_blkno
);
546 fe
->i_suballoc_bit
= cpu_to_le16(suballoc_bit
);
547 fe
->i_suballoc_slot
= cpu_to_le16(osb
->slot_num
);
548 fe
->i_uid
= cpu_to_le32(current
->fsuid
);
549 if (dir
->i_mode
& S_ISGID
) {
550 fe
->i_gid
= cpu_to_le32(dir
->i_gid
);
554 fe
->i_gid
= cpu_to_le32(current
->fsgid
);
555 fe
->i_mode
= cpu_to_le16(mode
);
556 if (S_ISCHR(mode
) || S_ISBLK(mode
))
557 fe
->id1
.dev1
.i_rdev
= cpu_to_le64(huge_encode_dev(dev
));
559 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
561 fe
->i_last_eb_blk
= 0;
562 strcpy(fe
->i_signature
, OCFS2_INODE_SIGNATURE
);
563 le32_add_cpu(&fe
->i_flags
, OCFS2_VALID_FL
);
564 fe
->i_atime
= fe
->i_ctime
= fe
->i_mtime
=
565 cpu_to_le64(CURRENT_TIME
.tv_sec
);
566 fe
->i_mtime_nsec
= fe
->i_ctime_nsec
= fe
->i_atime_nsec
=
567 cpu_to_le32(CURRENT_TIME
.tv_nsec
);
570 fel
= &fe
->id2
.i_list
;
571 fel
->l_tree_depth
= 0;
572 fel
->l_next_free_rec
= 0;
573 fel
->l_count
= cpu_to_le16(ocfs2_extent_recs_per_inode(osb
->sb
));
575 status
= ocfs2_journal_dirty(handle
, *new_fe_bh
);
581 if (ocfs2_populate_inode(inode
, fe
, 1) < 0) {
582 mlog(ML_ERROR
, "populate inode failed! bh->b_blocknr=%llu, "
583 "i_blkno=%llu, i_ino=%lu\n",
584 (unsigned long long) (*new_fe_bh
)->b_blocknr
,
585 (unsigned long long)fe
->i_blkno
, inode
->i_ino
);
589 ocfs2_inode_set_new(osb
, inode
);
590 status
= ocfs2_create_new_inode_locks(inode
);
594 status
= 0; /* error in ocfs2_create_new_inode_locks is not
612 static int ocfs2_mkdir(struct inode
*dir
,
613 struct dentry
*dentry
,
618 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
619 dentry
->d_name
.len
, dentry
->d_name
.name
);
620 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
626 static int ocfs2_create(struct inode
*dir
,
627 struct dentry
*dentry
,
629 struct nameidata
*nd
)
633 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir
, dentry
, mode
,
634 dentry
->d_name
.len
, dentry
->d_name
.name
);
635 ret
= ocfs2_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
641 static int ocfs2_link(struct dentry
*old_dentry
,
643 struct dentry
*dentry
)
646 struct inode
*inode
= old_dentry
->d_inode
;
648 struct buffer_head
*fe_bh
= NULL
;
649 struct buffer_head
*parent_fe_bh
= NULL
;
650 struct buffer_head
*de_bh
= NULL
;
651 struct ocfs2_dinode
*fe
= NULL
;
652 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
654 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode
->i_ino
,
655 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
656 dentry
->d_name
.len
, dentry
->d_name
.name
);
658 if (S_ISDIR(inode
->i_mode
))
661 err
= ocfs2_meta_lock(dir
, &parent_fe_bh
, 1);
673 err
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
678 err
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
680 dentry
->d_name
.len
, &de_bh
);
686 err
= ocfs2_meta_lock(inode
, &fe_bh
, 1);
693 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
694 if (le16_to_cpu(fe
->i_links_count
) >= OCFS2_LINK_MAX
) {
696 goto out_unlock_inode
;
699 handle
= ocfs2_start_trans(osb
, OCFS2_LINK_CREDITS
);
700 if (IS_ERR(handle
)) {
701 err
= PTR_ERR(handle
);
704 goto out_unlock_inode
;
707 err
= ocfs2_journal_access(handle
, inode
, fe_bh
,
708 OCFS2_JOURNAL_ACCESS_WRITE
);
715 inode
->i_ctime
= CURRENT_TIME
;
716 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
717 fe
->i_ctime
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
718 fe
->i_ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
720 err
= ocfs2_journal_dirty(handle
, fe_bh
);
722 le16_add_cpu(&fe
->i_links_count
, -1);
728 err
= ocfs2_add_entry(handle
, dentry
, inode
,
729 OCFS2_I(inode
)->ip_blkno
,
730 parent_fe_bh
, de_bh
);
732 le16_add_cpu(&fe
->i_links_count
, -1);
738 err
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
744 atomic_inc(&inode
->i_count
);
745 dentry
->d_op
= &ocfs2_dentry_ops
;
746 d_instantiate(dentry
, inode
);
749 ocfs2_commit_trans(osb
, handle
);
751 ocfs2_meta_unlock(inode
, 1);
754 ocfs2_meta_unlock(dir
, 1);
761 brelse(parent_fe_bh
);
769 * Takes and drops an exclusive lock on the given dentry. This will
770 * force other nodes to drop it.
772 static int ocfs2_remote_dentry_delete(struct dentry
*dentry
)
776 ret
= ocfs2_dentry_lock(dentry
, 1);
780 ocfs2_dentry_unlock(dentry
, 1);
785 static inline int inode_is_unlinkable(struct inode
*inode
)
787 if (S_ISDIR(inode
->i_mode
)) {
788 if (inode
->i_nlink
== 2)
793 if (inode
->i_nlink
== 1)
798 static int ocfs2_unlink(struct inode
*dir
,
799 struct dentry
*dentry
)
802 int child_locked
= 0;
803 struct inode
*inode
= dentry
->d_inode
;
804 struct inode
*orphan_dir
= NULL
;
805 struct ocfs2_super
*osb
= OCFS2_SB(dir
->i_sb
);
807 struct ocfs2_dinode
*fe
= NULL
;
808 struct buffer_head
*fe_bh
= NULL
;
809 struct buffer_head
*parent_node_bh
= NULL
;
810 handle_t
*handle
= NULL
;
811 struct ocfs2_dir_entry
*dirent
= NULL
;
812 struct buffer_head
*dirent_bh
= NULL
;
813 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
814 struct buffer_head
*orphan_entry_bh
= NULL
;
816 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir
, dentry
,
817 dentry
->d_name
.len
, dentry
->d_name
.name
);
819 BUG_ON(dentry
->d_parent
->d_inode
!= dir
);
821 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
823 if (inode
== osb
->root_inode
) {
824 mlog(0, "Cannot delete the root directory\n");
828 status
= ocfs2_meta_lock(dir
, &parent_node_bh
, 1);
830 if (status
!= -ENOENT
)
835 status
= ocfs2_find_files_on_disk(dentry
->d_name
.name
,
836 dentry
->d_name
.len
, &blkno
,
837 dir
, &dirent_bh
, &dirent
);
839 if (status
!= -ENOENT
)
844 if (OCFS2_I(inode
)->ip_blkno
!= blkno
) {
847 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
848 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
849 (unsigned long long)blkno
, OCFS2_I(inode
)->ip_flags
);
853 status
= ocfs2_meta_lock(inode
, &fe_bh
, 1);
855 if (status
!= -ENOENT
)
861 if (S_ISDIR(inode
->i_mode
)) {
862 if (!ocfs2_empty_dir(inode
)) {
865 } else if (inode
->i_nlink
!= 2) {
871 status
= ocfs2_remote_dentry_delete(dentry
);
873 /* This vote should succeed under all normal
879 if (inode_is_unlinkable(inode
)) {
880 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
, inode
,
889 handle
= ocfs2_start_trans(osb
, OCFS2_UNLINK_CREDITS
);
890 if (IS_ERR(handle
)) {
891 status
= PTR_ERR(handle
);
897 status
= ocfs2_journal_access(handle
, inode
, fe_bh
,
898 OCFS2_JOURNAL_ACCESS_WRITE
);
904 fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
906 if (inode_is_unlinkable(inode
)) {
907 status
= ocfs2_orphan_add(osb
, handle
, inode
, fe
, orphan_name
,
908 orphan_entry_bh
, orphan_dir
);
915 /* delete the name from the parent dir */
916 status
= ocfs2_delete_entry(handle
, dir
, dirent
, dirent_bh
);
922 if (S_ISDIR(inode
->i_mode
))
925 fe
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
927 status
= ocfs2_journal_dirty(handle
, fe_bh
);
933 if (S_ISDIR(inode
->i_mode
)) {
935 status
= ocfs2_mark_inode_dirty(handle
, dir
,
945 ocfs2_commit_trans(osb
, handle
);
948 ocfs2_meta_unlock(inode
, 1);
950 ocfs2_meta_unlock(dir
, 1);
953 /* This was locked for us in ocfs2_prepare_orphan_dir() */
954 ocfs2_meta_unlock(orphan_dir
, 1);
955 mutex_unlock(&orphan_dir
->i_mutex
);
966 brelse(parent_node_bh
);
969 brelse(orphan_entry_bh
);
977 * The only place this should be used is rename!
978 * if they have the same id, then the 1st one is the only one locked.
980 static int ocfs2_double_lock(struct ocfs2_super
*osb
,
981 struct buffer_head
**bh1
,
982 struct inode
*inode1
,
983 struct buffer_head
**bh2
,
984 struct inode
*inode2
)
987 struct ocfs2_inode_info
*oi1
= OCFS2_I(inode1
);
988 struct ocfs2_inode_info
*oi2
= OCFS2_I(inode2
);
989 struct buffer_head
**tmpbh
;
990 struct inode
*tmpinode
;
992 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
993 (unsigned long long)oi1
->ip_blkno
,
994 (unsigned long long)oi2
->ip_blkno
);
1001 /* we always want to lock the one with the lower lockid first. */
1002 if (oi1
->ip_blkno
!= oi2
->ip_blkno
) {
1003 if (oi1
->ip_blkno
< oi2
->ip_blkno
) {
1004 /* switch id1 and id2 around */
1005 mlog(0, "switching them around...\n");
1015 status
= ocfs2_meta_lock(inode2
, bh2
, 1);
1017 if (status
!= -ENOENT
)
1024 status
= ocfs2_meta_lock(inode1
, bh1
, 1);
1027 * An error return must mean that no cluster locks
1028 * were held on function exit.
1030 if (oi1
->ip_blkno
!= oi2
->ip_blkno
)
1031 ocfs2_meta_unlock(inode2
, 1);
1033 if (status
!= -ENOENT
)
1042 static void ocfs2_double_unlock(struct inode
*inode1
, struct inode
*inode2
)
1044 ocfs2_meta_unlock(inode1
, 1);
1046 if (inode1
!= inode2
)
1047 ocfs2_meta_unlock(inode2
, 1);
1050 #define PARENT_INO(buffer) \
1051 ((struct ocfs2_dir_entry *) \
1053 le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1055 static int ocfs2_rename(struct inode
*old_dir
,
1056 struct dentry
*old_dentry
,
1057 struct inode
*new_dir
,
1058 struct dentry
*new_dentry
)
1060 int status
= 0, rename_lock
= 0, parents_locked
= 0;
1061 int old_child_locked
= 0, new_child_locked
= 0;
1062 struct inode
*old_inode
= old_dentry
->d_inode
;
1063 struct inode
*new_inode
= new_dentry
->d_inode
;
1064 struct inode
*orphan_dir
= NULL
;
1065 struct ocfs2_dinode
*newfe
= NULL
;
1066 char orphan_name
[OCFS2_ORPHAN_NAMELEN
+ 1];
1067 struct buffer_head
*orphan_entry_bh
= NULL
;
1068 struct buffer_head
*newfe_bh
= NULL
;
1069 struct buffer_head
*insert_entry_bh
= NULL
;
1070 struct ocfs2_super
*osb
= NULL
;
1072 handle_t
*handle
= NULL
;
1073 struct buffer_head
*old_dir_bh
= NULL
;
1074 struct buffer_head
*new_dir_bh
= NULL
;
1075 struct ocfs2_dir_entry
*old_de
= NULL
, *new_de
= NULL
; // dirent for old_dentry
1077 struct buffer_head
*new_de_bh
= NULL
, *old_de_bh
= NULL
; // bhs for above
1078 struct buffer_head
*old_inode_de_bh
= NULL
; // if old_dentry is a dir,
1079 // this is the 1st dirent bh
1080 nlink_t old_dir_nlink
= old_dir
->i_nlink
, new_dir_nlink
= new_dir
->i_nlink
;
1082 /* At some point it might be nice to break this function up a
1085 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1086 old_dir
, old_dentry
, new_dir
, new_dentry
,
1087 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
1088 new_dentry
->d_name
.len
, new_dentry
->d_name
.name
);
1090 osb
= OCFS2_SB(old_dir
->i_sb
);
1093 if (!igrab(new_inode
))
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 /* if old and new are the same, this'll just do one lock. */
1118 status
= ocfs2_double_lock(osb
, &old_dir_bh
, old_dir
,
1119 &new_dir_bh
, new_dir
);
1126 /* make sure both dirs have bhs
1127 * get an extra ref on old_dir_bh if old==new */
1130 new_dir_bh
= old_dir_bh
;
1133 mlog(ML_ERROR
, "no old_dir_bh!\n");
1140 * Though we don't require an inode meta data update if
1141 * old_inode is not a directory, we lock anyway here to ensure
1142 * the vote thread on other nodes won't have to concurrently
1143 * downconvert the inode and the dentry locks.
1145 status
= ocfs2_meta_lock(old_inode
, NULL
, 1);
1147 if (status
!= -ENOENT
)
1151 old_child_locked
= 1;
1153 status
= ocfs2_remote_dentry_delete(old_dentry
);
1159 if (S_ISDIR(old_inode
->i_mode
)) {
1161 old_inode_de_bh
= ocfs2_bread(old_inode
, 0, &status
, 0);
1162 if (!old_inode_de_bh
)
1166 if (le64_to_cpu(PARENT_INO(old_inode_de_bh
->b_data
)) !=
1167 OCFS2_I(old_dir
)->ip_blkno
)
1170 if (!new_inode
&& new_dir
!=old_dir
&&
1171 new_dir
->i_nlink
>= OCFS2_LINK_MAX
)
1176 old_de_bh
= ocfs2_find_entry(old_dentry
->d_name
.name
,
1177 old_dentry
->d_name
.len
,
1183 * Check for inode number is _not_ due to possible IO errors.
1184 * We might rmdir the source, keep it as pwd of some process
1185 * and merrily kill the link to whatever was created under the
1186 * same name. Goodbye sticky bit ;-<
1188 if (le64_to_cpu(old_de
->inode
) != OCFS2_I(old_inode
)->ip_blkno
)
1191 /* check if the target already exists (in which case we need
1193 status
= ocfs2_find_files_on_disk(new_dentry
->d_name
.name
,
1194 new_dentry
->d_name
.len
,
1195 &newfe_blkno
, new_dir
, &new_de_bh
,
1197 /* The only error we allow here is -ENOENT because the new
1198 * file not existing is perfectly valid. */
1199 if ((status
< 0) && (status
!= -ENOENT
)) {
1200 /* If we cannot find the file specified we should just */
1201 /* return the error... */
1206 if (!new_de
&& new_inode
)
1207 mlog(ML_ERROR
, "inode %lu does not exist in it's parent "
1208 "directory!", new_inode
->i_ino
);
1210 /* In case we need to overwrite an existing file, we blow it
1213 /* VFS didn't think there existed an inode here, but
1214 * someone else in the cluster must have raced our
1215 * rename to create one. Today we error cleanly, in
1216 * the future we should consider calling iget to build
1217 * a new struct inode for this entry. */
1221 mlog(0, "We found an inode for name %.*s but VFS "
1222 "didn't give us one.\n", new_dentry
->d_name
.len
,
1223 new_dentry
->d_name
.name
);
1227 if (OCFS2_I(new_inode
)->ip_blkno
!= newfe_blkno
) {
1230 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1231 (unsigned long long)OCFS2_I(new_inode
)->ip_blkno
,
1232 (unsigned long long)newfe_blkno
,
1233 OCFS2_I(new_inode
)->ip_flags
);
1237 status
= ocfs2_meta_lock(new_inode
, &newfe_bh
, 1);
1239 if (status
!= -ENOENT
)
1243 new_child_locked
= 1;
1245 status
= ocfs2_remote_dentry_delete(new_dentry
);
1251 newfe
= (struct ocfs2_dinode
*) newfe_bh
->b_data
;
1253 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1254 "newfebh=%p bhblocknr=%llu\n", new_de
,
1255 (unsigned long long)newfe_blkno
, newfe_bh
, newfe_bh
?
1256 (unsigned long long)newfe_bh
->b_blocknr
: 0ULL);
1258 if (S_ISDIR(new_inode
->i_mode
) || (new_inode
->i_nlink
== 1)) {
1259 status
= ocfs2_prepare_orphan_dir(osb
, &orphan_dir
,
1269 BUG_ON(new_dentry
->d_parent
->d_inode
!= new_dir
);
1271 status
= ocfs2_check_dir_for_entry(new_dir
,
1272 new_dentry
->d_name
.name
,
1273 new_dentry
->d_name
.len
);
1277 status
= ocfs2_prepare_dir_for_insert(osb
, new_dir
, new_dir_bh
,
1278 new_dentry
->d_name
.name
,
1279 new_dentry
->d_name
.len
,
1287 handle
= ocfs2_start_trans(osb
, OCFS2_RENAME_CREDITS
);
1288 if (IS_ERR(handle
)) {
1289 status
= PTR_ERR(handle
);
1296 if (S_ISDIR(new_inode
->i_mode
)) {
1297 if (!ocfs2_empty_dir(new_inode
) ||
1298 new_inode
->i_nlink
!= 2) {
1299 status
= -ENOTEMPTY
;
1303 status
= ocfs2_journal_access(handle
, new_inode
, newfe_bh
,
1304 OCFS2_JOURNAL_ACCESS_WRITE
);
1310 if (S_ISDIR(new_inode
->i_mode
) ||
1311 (newfe
->i_links_count
== cpu_to_le16(1))){
1312 status
= ocfs2_orphan_add(osb
, handle
, new_inode
,
1314 orphan_entry_bh
, orphan_dir
);
1321 /* change the dirent to point to the correct inode */
1322 status
= ocfs2_journal_access(handle
, new_dir
, new_de_bh
,
1323 OCFS2_JOURNAL_ACCESS_WRITE
);
1328 new_de
->inode
= cpu_to_le64(OCFS2_I(old_inode
)->ip_blkno
);
1329 new_de
->file_type
= old_de
->file_type
;
1330 new_dir
->i_version
++;
1331 status
= ocfs2_journal_dirty(handle
, new_de_bh
);
1337 if (S_ISDIR(new_inode
->i_mode
))
1338 newfe
->i_links_count
= 0;
1340 le16_add_cpu(&newfe
->i_links_count
, -1);
1342 status
= ocfs2_journal_dirty(handle
, newfe_bh
);
1348 /* if the name was not found in new_dir, add it now */
1349 status
= ocfs2_add_entry(handle
, new_dentry
, old_inode
,
1350 OCFS2_I(old_inode
)->ip_blkno
,
1351 new_dir_bh
, insert_entry_bh
);
1354 old_inode
->i_ctime
= CURRENT_TIME
;
1355 mark_inode_dirty(old_inode
);
1357 /* now that the name has been added to new_dir, remove the old name */
1358 status
= ocfs2_delete_entry(handle
, old_dir
, old_de
, old_de_bh
);
1365 new_inode
->i_nlink
--;
1366 new_inode
->i_ctime
= CURRENT_TIME
;
1368 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME
;
1369 if (old_inode_de_bh
) {
1370 status
= ocfs2_journal_access(handle
, old_inode
,
1372 OCFS2_JOURNAL_ACCESS_WRITE
);
1373 PARENT_INO(old_inode_de_bh
->b_data
) =
1374 cpu_to_le64(OCFS2_I(new_dir
)->ip_blkno
);
1375 status
= ocfs2_journal_dirty(handle
, old_inode_de_bh
);
1378 new_inode
->i_nlink
--;
1381 mark_inode_dirty(new_dir
);
1384 mark_inode_dirty(old_dir
);
1386 mark_inode_dirty(new_inode
);
1388 if (old_dir
!= new_dir
)
1389 if (new_dir_nlink
!= new_dir
->i_nlink
) {
1391 mlog(ML_ERROR
, "need to change nlink for new "
1392 "dir %llu from %d to %d but bh is NULL\n",
1393 (unsigned long long)OCFS2_I(new_dir
)->ip_blkno
,
1394 (int)new_dir_nlink
, new_dir
->i_nlink
);
1396 struct ocfs2_dinode
*fe
;
1397 status
= ocfs2_journal_access(handle
,
1400 OCFS2_JOURNAL_ACCESS_WRITE
);
1401 fe
= (struct ocfs2_dinode
*) new_dir_bh
->b_data
;
1402 fe
->i_links_count
= cpu_to_le16(new_dir
->i_nlink
);
1403 status
= ocfs2_journal_dirty(handle
, new_dir_bh
);
1407 if (old_dir_nlink
!= old_dir
->i_nlink
) {
1409 mlog(ML_ERROR
, "need to change nlink for old dir "
1410 "%llu from %d to %d but bh is NULL!\n",
1411 (unsigned long long)OCFS2_I(old_dir
)->ip_blkno
,
1412 (int)old_dir_nlink
, old_dir
->i_nlink
);
1414 struct ocfs2_dinode
*fe
;
1415 status
= ocfs2_journal_access(handle
, old_dir
,
1417 OCFS2_JOURNAL_ACCESS_WRITE
);
1418 fe
= (struct ocfs2_dinode
*) old_dir_bh
->b_data
;
1419 fe
->i_links_count
= cpu_to_le16(old_dir
->i_nlink
);
1420 status
= ocfs2_journal_dirty(handle
, old_dir_bh
);
1424 ocfs2_dentry_move(old_dentry
, new_dentry
, old_dir
, new_dir
);
1428 ocfs2_rename_unlock(osb
);
1431 ocfs2_commit_trans(osb
, handle
);
1434 ocfs2_double_unlock(old_dir
, new_dir
);
1436 if (old_child_locked
)
1437 ocfs2_meta_unlock(old_inode
, 1);
1439 if (new_child_locked
)
1440 ocfs2_meta_unlock(new_inode
, 1);
1443 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1444 ocfs2_meta_unlock(orphan_dir
, 1);
1445 mutex_unlock(&orphan_dir
->i_mutex
);
1450 sync_mapping_buffers(old_inode
->i_mapping
);
1464 if (old_inode_de_bh
)
1465 brelse(old_inode_de_bh
);
1466 if (orphan_entry_bh
)
1467 brelse(orphan_entry_bh
);
1468 if (insert_entry_bh
)
1469 brelse(insert_entry_bh
);
1477 * we expect i_size = strlen(symname). Copy symname into the file
1478 * data, including the null terminator.
1480 static int ocfs2_create_symlink_data(struct ocfs2_super
*osb
,
1482 struct inode
*inode
,
1483 const char *symname
)
1485 struct buffer_head
**bhs
= NULL
;
1487 struct super_block
*sb
= osb
->sb
;
1490 int virtual, blocks
, status
, i
, bytes_left
;
1492 bytes_left
= i_size_read(inode
) + 1;
1493 /* we can't trust i_blocks because we're actually going to
1494 * write i_size + 1 bytes. */
1495 blocks
= (bytes_left
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
1497 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1498 (unsigned long long)inode
->i_blocks
,
1499 i_size_read(inode
), blocks
);
1501 /* Sanity check -- make sure we're going to fit. */
1503 ocfs2_clusters_to_bytes(sb
, OCFS2_I(inode
)->ip_clusters
)) {
1509 bhs
= kcalloc(blocks
, sizeof(struct buffer_head
*), GFP_KERNEL
);
1516 status
= ocfs2_extent_map_get_blocks(inode
, 0, 1, &p_blkno
,
1523 /* links can never be larger than one cluster so we know this
1524 * is all going to be contiguous, but do a sanity check
1526 if ((p_blocks
<< sb
->s_blocksize_bits
) < bytes_left
) {
1533 while(bytes_left
> 0) {
1534 c
= &symname
[virtual * sb
->s_blocksize
];
1536 bhs
[virtual] = sb_getblk(sb
, p_blkno
);
1537 if (!bhs
[virtual]) {
1542 ocfs2_set_new_buffer_uptodate(inode
, bhs
[virtual]);
1544 status
= ocfs2_journal_access(handle
, inode
, bhs
[virtual],
1545 OCFS2_JOURNAL_ACCESS_CREATE
);
1551 memset(bhs
[virtual]->b_data
, 0, sb
->s_blocksize
);
1553 memcpy(bhs
[virtual]->b_data
, c
,
1554 (bytes_left
> sb
->s_blocksize
) ? sb
->s_blocksize
:
1557 status
= ocfs2_journal_dirty(handle
, bhs
[virtual]);
1565 bytes_left
-= sb
->s_blocksize
;
1572 for(i
= 0; i
< blocks
; i
++)
1582 static int ocfs2_symlink(struct inode
*dir
,
1583 struct dentry
*dentry
,
1584 const char *symname
)
1586 int status
, l
, credits
;
1588 struct ocfs2_super
*osb
= NULL
;
1589 struct inode
*inode
= NULL
;
1590 struct super_block
*sb
;
1591 struct buffer_head
*new_fe_bh
= NULL
;
1592 struct buffer_head
*de_bh
= NULL
;
1593 struct buffer_head
*parent_fe_bh
= NULL
;
1594 struct ocfs2_dinode
*fe
= NULL
;
1595 struct ocfs2_dinode
*dirfe
;
1596 handle_t
*handle
= NULL
;
1597 struct ocfs2_alloc_context
*inode_ac
= NULL
;
1598 struct ocfs2_alloc_context
*data_ac
= NULL
;
1600 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir
,
1601 dentry
, symname
, dentry
->d_name
.len
, dentry
->d_name
.name
);
1606 l
= strlen(symname
) + 1;
1608 credits
= ocfs2_calc_symlink_credits(sb
);
1610 /* lock the parent directory */
1611 status
= ocfs2_meta_lock(dir
, &parent_fe_bh
, 1);
1613 if (status
!= -ENOENT
)
1618 dirfe
= (struct ocfs2_dinode
*) parent_fe_bh
->b_data
;
1619 if (!dirfe
->i_links_count
) {
1620 /* can't make a file in a deleted directory. */
1625 status
= ocfs2_check_dir_for_entry(dir
, dentry
->d_name
.name
,
1626 dentry
->d_name
.len
);
1630 status
= ocfs2_prepare_dir_for_insert(osb
, dir
, parent_fe_bh
,
1631 dentry
->d_name
.name
,
1632 dentry
->d_name
.len
, &de_bh
);
1638 status
= ocfs2_reserve_new_inode(osb
, &inode_ac
);
1640 if (status
!= -ENOSPC
)
1645 /* don't reserve bitmap space for fast symlinks. */
1646 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1647 status
= ocfs2_reserve_clusters(osb
, 1, &data_ac
);
1649 if (status
!= -ENOSPC
)
1655 handle
= ocfs2_start_trans(osb
, credits
);
1656 if (IS_ERR(handle
)) {
1657 status
= PTR_ERR(handle
);
1663 status
= ocfs2_mknod_locked(osb
, dir
, dentry
,
1664 S_IFLNK
| S_IRWXUGO
, 0,
1665 &new_fe_bh
, parent_fe_bh
, handle
,
1672 fe
= (struct ocfs2_dinode
*) new_fe_bh
->b_data
;
1675 if (l
> ocfs2_fast_symlink_chars(sb
)) {
1676 inode
->i_op
= &ocfs2_symlink_inode_operations
;
1677 status
= ocfs2_do_extend_allocation(osb
, inode
, 1, new_fe_bh
,
1678 handle
, data_ac
, NULL
,
1681 if (status
!= -ENOSPC
&& status
!= -EINTR
) {
1683 "Failed to extend file to %llu\n",
1684 (unsigned long long)newsize
);
1690 i_size_write(inode
, newsize
);
1691 inode
->i_blocks
= ocfs2_align_bytes_to_sectors(newsize
);
1693 inode
->i_op
= &ocfs2_fast_symlink_inode_operations
;
1694 memcpy((char *) fe
->id2
.i_symlink
, symname
, l
);
1695 i_size_write(inode
, newsize
);
1696 inode
->i_blocks
= 0;
1699 status
= ocfs2_mark_inode_dirty(handle
, inode
, new_fe_bh
);
1705 if (!ocfs2_inode_is_fast_symlink(inode
)) {
1706 status
= ocfs2_create_symlink_data(osb
, handle
, inode
,
1714 status
= ocfs2_add_entry(handle
, dentry
, inode
,
1715 le64_to_cpu(fe
->i_blkno
), parent_fe_bh
,
1722 status
= ocfs2_dentry_attach_lock(dentry
, inode
, OCFS2_I(dir
)->ip_blkno
);
1728 insert_inode_hash(inode
);
1729 dentry
->d_op
= &ocfs2_dentry_ops
;
1730 d_instantiate(dentry
, inode
);
1733 ocfs2_commit_trans(osb
, handle
);
1735 ocfs2_meta_unlock(dir
, 1);
1740 brelse(parent_fe_bh
);
1744 ocfs2_free_alloc_context(inode_ac
);
1746 ocfs2_free_alloc_context(data_ac
);
1747 if ((status
< 0) && inode
)
1755 int ocfs2_check_dir_entry(struct inode
* dir
,
1756 struct ocfs2_dir_entry
* de
,
1757 struct buffer_head
* bh
,
1758 unsigned long offset
)
1760 const char *error_msg
= NULL
;
1761 const int rlen
= le16_to_cpu(de
->rec_len
);
1763 if (rlen
< OCFS2_DIR_REC_LEN(1))
1764 error_msg
= "rec_len is smaller than minimal";
1765 else if (rlen
% 4 != 0)
1766 error_msg
= "rec_len % 4 != 0";
1767 else if (rlen
< OCFS2_DIR_REC_LEN(de
->name_len
))
1768 error_msg
= "rec_len is too small for name_len";
1769 else if (((char *) de
- bh
->b_data
) + rlen
> dir
->i_sb
->s_blocksize
)
1770 error_msg
= "directory entry across blocks";
1772 if (error_msg
!= NULL
)
1773 mlog(ML_ERROR
, "bad entry in directory #%llu: %s - "
1774 "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1775 (unsigned long long)OCFS2_I(dir
)->ip_blkno
, error_msg
,
1776 offset
, (unsigned long long)le64_to_cpu(de
->inode
), rlen
,
1778 return error_msg
== NULL
? 1 : 0;
1781 /* we don't always have a dentry for what we want to add, so people
1782 * like orphan dir can call this instead.
1784 * If you pass me insert_bh, I'll skip the search of the other dir
1785 * blocks and put the record in there.
1787 static int __ocfs2_add_entry(handle_t
*handle
,
1789 const char *name
, int namelen
,
1790 struct inode
*inode
, u64 blkno
,
1791 struct buffer_head
*parent_fe_bh
,
1792 struct buffer_head
*insert_bh
)
1794 unsigned long offset
;
1795 unsigned short rec_len
;
1796 struct ocfs2_dir_entry
*de
, *de1
;
1797 struct super_block
*sb
;
1807 rec_len
= OCFS2_DIR_REC_LEN(namelen
);
1809 de
= (struct ocfs2_dir_entry
*) insert_bh
->b_data
;
1811 BUG_ON((char *)de
>= sb
->s_blocksize
+ insert_bh
->b_data
);
1812 /* These checks should've already been passed by the
1813 * prepare function, but I guess we can leave them
1815 if (!ocfs2_check_dir_entry(dir
, de
, insert_bh
, offset
)) {
1819 if (ocfs2_match(namelen
, name
, de
)) {
1823 if (((le64_to_cpu(de
->inode
) == 0) &&
1824 (le16_to_cpu(de
->rec_len
) >= rec_len
)) ||
1825 (le16_to_cpu(de
->rec_len
) >=
1826 (OCFS2_DIR_REC_LEN(de
->name_len
) + rec_len
))) {
1827 status
= ocfs2_journal_access(handle
, dir
, insert_bh
,
1828 OCFS2_JOURNAL_ACCESS_WRITE
);
1829 /* By now the buffer is marked for journaling */
1830 offset
+= le16_to_cpu(de
->rec_len
);
1831 if (le64_to_cpu(de
->inode
)) {
1832 de1
= (struct ocfs2_dir_entry
*)((char *) de
+
1833 OCFS2_DIR_REC_LEN(de
->name_len
));
1835 cpu_to_le16(le16_to_cpu(de
->rec_len
) -
1836 OCFS2_DIR_REC_LEN(de
->name_len
));
1837 de
->rec_len
= cpu_to_le16(OCFS2_DIR_REC_LEN(de
->name_len
));
1840 de
->file_type
= OCFS2_FT_UNKNOWN
;
1842 de
->inode
= cpu_to_le64(blkno
);
1843 ocfs2_set_de_type(de
, inode
->i_mode
);
1846 de
->name_len
= namelen
;
1847 memcpy(de
->name
, name
, namelen
);
1849 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
1851 status
= ocfs2_journal_dirty(handle
, insert_bh
);
1855 offset
+= le16_to_cpu(de
->rec_len
);
1856 de
= (struct ocfs2_dir_entry
*) ((char *) de
+ le16_to_cpu(de
->rec_len
));
1859 /* when you think about it, the assert above should prevent us
1860 * from ever getting here. */
1870 * ocfs2_delete_entry deletes a directory entry by merging it with the
1873 static int ocfs2_delete_entry(handle_t
*handle
,
1875 struct ocfs2_dir_entry
*de_del
,
1876 struct buffer_head
*bh
)
1878 struct ocfs2_dir_entry
*de
, *pde
;
1879 int i
, status
= -ENOENT
;
1881 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle
, dir
, de_del
, bh
);
1885 de
= (struct ocfs2_dir_entry
*) bh
->b_data
;
1886 while (i
< bh
->b_size
) {
1887 if (!ocfs2_check_dir_entry(dir
, de
, bh
, i
)) {
1893 status
= ocfs2_journal_access(handle
, dir
, bh
,
1894 OCFS2_JOURNAL_ACCESS_WRITE
);
1902 cpu_to_le16(le16_to_cpu(pde
->rec_len
) +
1903 le16_to_cpu(de
->rec_len
));
1907 status
= ocfs2_journal_dirty(handle
, bh
);
1910 i
+= le16_to_cpu(de
->rec_len
);
1912 de
= (struct ocfs2_dir_entry
*)((char *)de
+ le16_to_cpu(de
->rec_len
));
1920 * Returns 0 if not found, -1 on failure, and 1 on success
1922 static int inline ocfs2_search_dirblock(struct buffer_head
*bh
,
1924 const char *name
, int namelen
,
1925 unsigned long offset
,
1926 struct ocfs2_dir_entry
**res_dir
)
1928 struct ocfs2_dir_entry
*de
;
1929 char *dlimit
, *de_buf
;
1935 de_buf
= bh
->b_data
;
1936 dlimit
= de_buf
+ dir
->i_sb
->s_blocksize
;
1938 while (de_buf
< dlimit
) {
1939 /* this code is executed quadratically often */
1940 /* do minimal checking `by hand' */
1942 de
= (struct ocfs2_dir_entry
*) de_buf
;
1944 if (de_buf
+ namelen
<= dlimit
&&
1945 ocfs2_match(namelen
, name
, de
)) {
1946 /* found a match - just to be sure, do a full check */
1947 if (!ocfs2_check_dir_entry(dir
, de
, bh
, offset
)) {
1956 /* prevent looping on a bad block */
1957 de_len
= le16_to_cpu(de
->rec_len
);
1972 struct buffer_head
*ocfs2_find_entry(const char *name
, int namelen
,
1974 struct ocfs2_dir_entry
**res_dir
)
1976 struct super_block
*sb
;
1977 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1978 struct buffer_head
*bh
, *ret
= NULL
;
1979 unsigned long start
, block
, b
;
1980 int ra_max
= 0; /* Number of bh's in the readahead
1982 int ra_ptr
= 0; /* Current index into readahead
1985 int nblocks
, i
, err
;
1992 nblocks
= i_size_read(dir
) >> sb
->s_blocksize_bits
;
1993 start
= OCFS2_I(dir
)->ip_dir_start_lookup
;
1994 if (start
>= nblocks
)
2001 * We deal with the read-ahead logic here.
2003 if (ra_ptr
>= ra_max
) {
2004 /* Refill the readahead buffer */
2007 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
2009 * Terminate if we reach the end of the
2010 * directory and must wrap, or if our
2011 * search has finished at this block.
2013 if (b
>= nblocks
|| (num
&& block
== start
)) {
2014 bh_use
[ra_max
] = NULL
;
2019 bh
= ocfs2_bread(dir
, b
++, &err
, 1);
2020 bh_use
[ra_max
] = bh
;
2023 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
2026 if (!buffer_uptodate(bh
)) {
2027 /* read error, skip block & hope for the best */
2028 ocfs2_error(dir
->i_sb
, "reading directory %llu, "
2030 (unsigned long long)OCFS2_I(dir
)->ip_blkno
,
2035 i
= ocfs2_search_dirblock(bh
, dir
, name
, namelen
,
2036 block
<< sb
->s_blocksize_bits
,
2039 OCFS2_I(dir
)->ip_dir_start_lookup
= block
;
2041 goto cleanup_and_exit
;
2045 goto cleanup_and_exit
;
2048 if (++block
>= nblocks
)
2050 } while (block
!= start
);
2053 * If the directory has grown while we were searching, then
2054 * search the last part of the directory before giving up.
2057 nblocks
= i_size_read(dir
) >> sb
->s_blocksize_bits
;
2058 if (block
< nblocks
) {
2064 /* Clean up the read-ahead blocks */
2065 for (; ra_ptr
< ra_max
; ra_ptr
++)
2066 brelse(bh_use
[ra_ptr
]);
2072 static int ocfs2_blkno_stringify(u64 blkno
, char *name
)
2074 int status
, namelen
;
2078 namelen
= snprintf(name
, OCFS2_ORPHAN_NAMELEN
+ 1, "%016llx",
2088 if (namelen
!= OCFS2_ORPHAN_NAMELEN
) {
2094 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name
,
2103 static int ocfs2_prepare_orphan_dir(struct ocfs2_super
*osb
,
2104 struct inode
**ret_orphan_dir
,
2105 struct inode
*inode
,
2107 struct buffer_head
**de_bh
)
2109 struct inode
*orphan_dir_inode
;
2110 struct buffer_head
*orphan_dir_bh
= NULL
;
2113 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2119 orphan_dir_inode
= ocfs2_get_system_file_inode(osb
,
2120 ORPHAN_DIR_SYSTEM_INODE
,
2122 if (!orphan_dir_inode
) {
2128 mutex_lock(&orphan_dir_inode
->i_mutex
);
2130 status
= ocfs2_meta_lock(orphan_dir_inode
, &orphan_dir_bh
, 1);
2136 status
= ocfs2_prepare_dir_for_insert(osb
, orphan_dir_inode
,
2137 orphan_dir_bh
, name
,
2138 OCFS2_ORPHAN_NAMELEN
, de_bh
);
2140 ocfs2_meta_unlock(orphan_dir_inode
, 1);
2146 *ret_orphan_dir
= orphan_dir_inode
;
2150 mutex_unlock(&orphan_dir_inode
->i_mutex
);
2151 iput(orphan_dir_inode
);
2155 brelse(orphan_dir_bh
);
2161 static int ocfs2_orphan_add(struct ocfs2_super
*osb
,
2163 struct inode
*inode
,
2164 struct ocfs2_dinode
*fe
,
2166 struct buffer_head
*de_bh
,
2167 struct inode
*orphan_dir_inode
)
2169 struct buffer_head
*orphan_dir_bh
= NULL
;
2171 struct ocfs2_dinode
*orphan_fe
;
2173 mlog_entry("(inode->i_ino = %lu)\n", inode
->i_ino
);
2175 status
= ocfs2_read_block(osb
,
2176 OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2177 &orphan_dir_bh
, OCFS2_BH_CACHED
,
2184 status
= ocfs2_journal_access(handle
, orphan_dir_inode
, orphan_dir_bh
,
2185 OCFS2_JOURNAL_ACCESS_WRITE
);
2191 /* we're a cluster, and nlink can change on disk from
2192 * underneath us... */
2193 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2194 if (S_ISDIR(inode
->i_mode
))
2195 le16_add_cpu(&orphan_fe
->i_links_count
, 1);
2196 orphan_dir_inode
->i_nlink
= le16_to_cpu(orphan_fe
->i_links_count
);
2198 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2204 status
= __ocfs2_add_entry(handle
, orphan_dir_inode
, name
,
2205 OCFS2_ORPHAN_NAMELEN
, inode
,
2206 OCFS2_I(inode
)->ip_blkno
,
2207 orphan_dir_bh
, de_bh
);
2213 le32_add_cpu(&fe
->i_flags
, OCFS2_ORPHANED_FL
);
2215 /* Record which orphan dir our inode now resides
2216 * in. delete_inode will use this to determine which orphan
2218 spin_lock(&OCFS2_I(inode
)->ip_lock
);
2219 OCFS2_I(inode
)->ip_orphaned_slot
= osb
->slot_num
;
2220 spin_unlock(&OCFS2_I(inode
)->ip_lock
);
2222 mlog(0, "Inode %llu orphaned in slot %d\n",
2223 (unsigned long long)OCFS2_I(inode
)->ip_blkno
, osb
->slot_num
);
2227 brelse(orphan_dir_bh
);
2233 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2234 int ocfs2_orphan_del(struct ocfs2_super
*osb
,
2236 struct inode
*orphan_dir_inode
,
2237 struct inode
*inode
,
2238 struct buffer_head
*orphan_dir_bh
)
2240 char name
[OCFS2_ORPHAN_NAMELEN
+ 1];
2241 struct ocfs2_dinode
*orphan_fe
;
2243 struct buffer_head
*target_de_bh
= NULL
;
2244 struct ocfs2_dir_entry
*target_de
= NULL
;
2248 status
= ocfs2_blkno_stringify(OCFS2_I(inode
)->ip_blkno
, name
);
2254 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2255 name
, (unsigned long long)OCFS2_I(orphan_dir_inode
)->ip_blkno
,
2256 OCFS2_ORPHAN_NAMELEN
);
2258 /* find it's spot in the orphan directory */
2259 target_de_bh
= ocfs2_find_entry(name
, OCFS2_ORPHAN_NAMELEN
,
2260 orphan_dir_inode
, &target_de
);
2261 if (!target_de_bh
) {
2267 /* remove it from the orphan directory */
2268 status
= ocfs2_delete_entry(handle
, orphan_dir_inode
, target_de
,
2275 status
= ocfs2_journal_access(handle
,orphan_dir_inode
, orphan_dir_bh
,
2276 OCFS2_JOURNAL_ACCESS_WRITE
);
2282 /* do the i_nlink dance! :) */
2283 orphan_fe
= (struct ocfs2_dinode
*) orphan_dir_bh
->b_data
;
2284 if (S_ISDIR(inode
->i_mode
))
2285 le16_add_cpu(&orphan_fe
->i_links_count
, -1);
2286 orphan_dir_inode
->i_nlink
= le16_to_cpu(orphan_fe
->i_links_count
);
2288 status
= ocfs2_journal_dirty(handle
, orphan_dir_bh
);
2296 brelse(target_de_bh
);
2302 struct inode_operations ocfs2_dir_iops
= {
2303 .create
= ocfs2_create
,
2304 .lookup
= ocfs2_lookup
,
2306 .unlink
= ocfs2_unlink
,
2307 .rmdir
= ocfs2_unlink
,
2308 .symlink
= ocfs2_symlink
,
2309 .mkdir
= ocfs2_mkdir
,
2310 .mknod
= ocfs2_mknod
,
2311 .rename
= ocfs2_rename
,
2312 .setattr
= ocfs2_setattr
,
2313 .getattr
= ocfs2_getattr
,
2314 .permission
= ocfs2_permission
,