1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Buffer cache handling
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
31 #include <cluster/masklog.h>
40 #include "buffer_head_io.h"
42 int ocfs2_write_block(struct ocfs2_super
*osb
, struct buffer_head
*bh
,
47 mlog_entry("(bh->b_blocknr = %llu, inode=%p)\n",
48 (unsigned long long)bh
->b_blocknr
, inode
);
50 BUG_ON(bh
->b_blocknr
< OCFS2_SUPER_BLOCK_BLKNO
);
51 BUG_ON(buffer_jbd(bh
));
53 /* No need to check for a soft readonly file system here. non
54 * journalled writes are only ever done on system files which
55 * can get modified during recovery even if read-only. */
56 if (ocfs2_is_hard_readonly(osb
)) {
61 mutex_lock(&OCFS2_I(inode
)->ip_io_mutex
);
64 set_buffer_uptodate(bh
);
66 /* remove from dirty list before I/O. */
67 clear_buffer_dirty(bh
);
69 get_bh(bh
); /* for end_buffer_write_sync() */
70 bh
->b_end_io
= end_buffer_write_sync
;
75 if (buffer_uptodate(bh
)) {
76 ocfs2_set_buffer_uptodate(inode
, bh
);
78 /* We don't need to remove the clustered uptodate
79 * information for this bh as it's not marked locally
85 mutex_unlock(&OCFS2_I(inode
)->ip_io_mutex
);
91 int ocfs2_read_blocks_sync(struct ocfs2_super
*osb
, u64 block
,
92 unsigned int nr
, struct buffer_head
*bhs
[])
96 struct buffer_head
*bh
;
99 mlog(ML_BH_IO
, "No buffers will be read!\n");
103 for (i
= 0 ; i
< nr
; i
++) {
104 if (bhs
[i
] == NULL
) {
105 bhs
[i
] = sb_getblk(osb
->sb
, block
++);
106 if (bhs
[i
] == NULL
) {
114 if (buffer_jbd(bh
)) {
116 "trying to sync read a jbd "
117 "managed bh (blocknr = %llu), skipping\n",
118 (unsigned long long)bh
->b_blocknr
);
122 if (buffer_dirty(bh
)) {
123 /* This should probably be a BUG, or
124 * at least return an error. */
126 "trying to sync read a dirty "
127 "buffer! (blocknr = %llu), skipping\n",
128 (unsigned long long)bh
->b_blocknr
);
133 if (buffer_jbd(bh
)) {
135 "block %llu had the JBD bit set "
136 "while I was in lock_buffer!",
137 (unsigned long long)bh
->b_blocknr
);
141 clear_buffer_uptodate(bh
);
142 get_bh(bh
); /* for end_buffer_read_sync() */
143 bh
->b_end_io
= end_buffer_read_sync
;
147 for (i
= nr
; i
> 0; i
--) {
150 /* No need to wait on the buffer if it's managed by JBD. */
154 if (!buffer_uptodate(bh
)) {
155 /* Status won't be cleared from here on out,
156 * so we can safely record this and loop back
157 * to cleanup the other buffers. */
168 int ocfs2_read_blocks(struct inode
*inode
, u64 block
, int nr
,
169 struct buffer_head
*bhs
[], int flags
)
172 int i
, ignore_cache
= 0;
173 struct buffer_head
*bh
;
175 mlog_entry("(inode=%p, block=(%llu), nr=(%d), flags=%d)\n",
176 inode
, (unsigned long long)block
, nr
, flags
);
179 BUG_ON((flags
& OCFS2_BH_READAHEAD
) &&
180 (flags
& OCFS2_BH_IGNORE_CACHE
));
189 mlog(ML_ERROR
, "asked to read %d blocks!\n", nr
);
196 mlog(ML_BH_IO
, "No buffers will be read!\n");
201 mutex_lock(&OCFS2_I(inode
)->ip_io_mutex
);
202 for (i
= 0 ; i
< nr
; i
++) {
203 if (bhs
[i
] == NULL
) {
204 bhs
[i
] = sb_getblk(inode
->i_sb
, block
++);
205 if (bhs
[i
] == NULL
) {
206 mutex_unlock(&OCFS2_I(inode
)->ip_io_mutex
);
213 ignore_cache
= (flags
& OCFS2_BH_IGNORE_CACHE
);
215 /* There are three read-ahead cases here which we need to
216 * be concerned with. All three assume a buffer has
217 * previously been submitted with OCFS2_BH_READAHEAD
218 * and it hasn't yet completed I/O.
220 * 1) The current request is sync to disk. This rarely
221 * happens these days, and never when performance
222 * matters - the code can just wait on the buffer
223 * lock and re-submit.
225 * 2) The current request is cached, but not
226 * readahead. ocfs2_buffer_uptodate() will return
227 * false anyway, so we'll wind up waiting on the
228 * buffer lock to do I/O. We re-check the request
229 * with after getting the lock to avoid a re-submit.
231 * 3) The current request is readahead (and so must
232 * also be a caching one). We short circuit if the
233 * buffer is locked (under I/O) and if it's in the
234 * uptodate cache. The re-check from #2 catches the
235 * case that the previous read-ahead completes just
236 * before our is-it-in-flight check.
239 if (!ignore_cache
&& !ocfs2_buffer_uptodate(inode
, bh
)) {
241 "bh (%llu), inode %llu not uptodate\n",
242 (unsigned long long)bh
->b_blocknr
,
243 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
244 /* We're using ignore_cache here to say
249 if (buffer_jbd(bh
)) {
251 mlog(ML_BH_IO
, "trying to sync read a jbd "
252 "managed bh (blocknr = %llu)\n",
253 (unsigned long long)bh
->b_blocknr
);
258 if (buffer_dirty(bh
)) {
259 /* This should probably be a BUG, or
260 * at least return an error. */
261 mlog(ML_BH_IO
, "asking me to sync read a dirty "
262 "buffer! (blocknr = %llu)\n",
263 (unsigned long long)bh
->b_blocknr
);
267 /* A read-ahead request was made - if the
268 * buffer is already under read-ahead from a
269 * previously submitted request than we are
271 if ((flags
& OCFS2_BH_READAHEAD
)
272 && ocfs2_buffer_read_ahead(inode
, bh
))
276 if (buffer_jbd(bh
)) {
277 #ifdef CATCH_BH_JBD_RACES
278 mlog(ML_ERROR
, "block %llu had the JBD bit set "
279 "while I was in lock_buffer!",
280 (unsigned long long)bh
->b_blocknr
);
288 /* Re-check ocfs2_buffer_uptodate() as a
289 * previously read-ahead buffer may have
290 * completed I/O while we were waiting for the
292 if (!(flags
& OCFS2_BH_IGNORE_CACHE
)
293 && !(flags
& OCFS2_BH_READAHEAD
)
294 && ocfs2_buffer_uptodate(inode
, bh
)) {
299 clear_buffer_uptodate(bh
);
300 get_bh(bh
); /* for end_buffer_read_sync() */
301 bh
->b_end_io
= end_buffer_read_sync
;
309 for (i
= (nr
- 1); i
>= 0; i
--) {
312 if (!(flags
& OCFS2_BH_READAHEAD
)) {
313 /* We know this can't have changed as we hold the
314 * inode sem. Avoid doing any work on the bh if the
319 if (!buffer_uptodate(bh
)) {
320 /* Status won't be cleared from here on out,
321 * so we can safely record this and loop back
322 * to cleanup the other buffers. Don't need to
323 * remove the clustered uptodate information
324 * for this bh as it's not marked locally
333 /* Always set the buffer in the cache, even if it was
334 * a forced read, or read-ahead which hasn't yet
336 ocfs2_set_buffer_uptodate(inode
, bh
);
338 mutex_unlock(&OCFS2_I(inode
)->ip_io_mutex
);
340 mlog(ML_BH_IO
, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n",
341 (unsigned long long)block
, nr
,
342 ((flags
& OCFS2_BH_IGNORE_CACHE
) || ignore_cache
) ? "no" : "yes",
351 /* Check whether the blkno is the super block or one of the backups. */
352 static void ocfs2_check_super_or_backup(struct super_block
*sb
,
358 if (blkno
== OCFS2_SUPER_BLOCK_BLKNO
)
361 for (i
= 0; i
< OCFS2_MAX_BACKUP_SUPERBLOCKS
; i
++) {
362 backup_blkno
= ocfs2_backup_super_blkno(sb
, i
);
363 if (backup_blkno
== blkno
)
371 * Write super block and backups doesn't need to collaborate with journal,
372 * so we don't need to lock ip_io_mutex and inode doesn't need to bea passed
373 * into this function.
375 int ocfs2_write_super_or_backup(struct ocfs2_super
*osb
,
376 struct buffer_head
*bh
)
382 BUG_ON(buffer_jbd(bh
));
383 ocfs2_check_super_or_backup(osb
->sb
, bh
->b_blocknr
);
385 if (ocfs2_is_hard_readonly(osb
) || ocfs2_is_soft_readonly(osb
)) {
391 set_buffer_uptodate(bh
);
393 /* remove from dirty list before I/O. */
394 clear_buffer_dirty(bh
);
396 get_bh(bh
); /* for end_buffer_write_sync() */
397 bh
->b_end_io
= end_buffer_write_sync
;
398 submit_bh(WRITE
, bh
);
402 if (!buffer_uptodate(bh
)) {