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"
43 * Bits on bh->b_state used by ocfs2.
45 * These MUST be after the JBD2 bits. Hence, we use BH_JBDPrivateStart.
47 enum ocfs2_state_bits {
48 BH_NeedsValidate = BH_JBDPrivateStart,
51 /* Expand the magic b_state functions */
52 BUFFER_FNS(NeedsValidate, needs_validate);
54 int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
59 mlog_entry("(bh->b_blocknr = %llu, inode=%p)\n",
60 (unsigned long long)bh->b_blocknr, inode);
62 BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
63 BUG_ON(buffer_jbd(bh));
65 /* No need to check for a soft readonly file system here. non
66 * journalled writes are only ever done on system files which
67 * can get modified during recovery even if read-only. */
68 if (ocfs2_is_hard_readonly(osb)) {
73 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
76 set_buffer_uptodate(bh);
78 /* remove from dirty list before I/O. */
79 clear_buffer_dirty(bh);
81 get_bh(bh); /* for end_buffer_write_sync() */
82 bh->b_end_io = end_buffer_write_sync;
87 if (buffer_uptodate(bh)) {
88 ocfs2_set_buffer_uptodate(inode, bh);
90 /* We don't need to remove the clustered uptodate
91 * information for this bh as it's not marked locally
97 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
103 int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
104 unsigned int nr, struct buffer_head *bhs[])
108 struct buffer_head *bh;
111 mlog(ML_BH_IO, "No buffers will be read!\n");
115 for (i = 0 ; i < nr ; i++) {
116 if (bhs[i] == NULL) {
117 bhs[i] = sb_getblk(osb->sb, block++);
118 if (bhs[i] == NULL) {
126 if (buffer_jbd(bh)) {
128 "trying to sync read a jbd "
129 "managed bh (blocknr = %llu), skipping\n",
130 (unsigned long long)bh->b_blocknr);
134 if (buffer_dirty(bh)) {
135 /* This should probably be a BUG, or
136 * at least return an error. */
138 "trying to sync read a dirty "
139 "buffer! (blocknr = %llu), skipping\n",
140 (unsigned long long)bh->b_blocknr);
145 if (buffer_jbd(bh)) {
147 "block %llu had the JBD bit set "
148 "while I was in lock_buffer!",
149 (unsigned long long)bh->b_blocknr);
153 clear_buffer_uptodate(bh);
154 get_bh(bh); /* for end_buffer_read_sync() */
155 bh->b_end_io = end_buffer_read_sync;
159 for (i = nr; i > 0; i--) {
162 /* No need to wait on the buffer if it's managed by JBD. */
166 if (!buffer_uptodate(bh)) {
167 /* Status won't be cleared from here on out,
168 * so we can safely record this and loop back
169 * to cleanup the other buffers. */
180 int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
181 struct buffer_head *bhs[], int flags,
182 int (*validate)(struct super_block *sb,
183 struct buffer_head *bh))
186 int i, ignore_cache = 0;
187 struct buffer_head *bh;
189 mlog_entry("(inode=%p, block=(%llu), nr=(%d), flags=%d)\n",
190 inode, (unsigned long long)block, nr, flags);
193 BUG_ON((flags & OCFS2_BH_READAHEAD) &&
194 (flags & OCFS2_BH_IGNORE_CACHE));
203 mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
210 mlog(ML_BH_IO, "No buffers will be read!\n");
215 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
216 for (i = 0 ; i < nr ; i++) {
217 if (bhs[i] == NULL) {
218 bhs[i] = sb_getblk(inode->i_sb, block++);
219 if (bhs[i] == NULL) {
220 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
227 ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE);
229 /* There are three read-ahead cases here which we need to
230 * be concerned with. All three assume a buffer has
231 * previously been submitted with OCFS2_BH_READAHEAD
232 * and it hasn't yet completed I/O.
234 * 1) The current request is sync to disk. This rarely
235 * happens these days, and never when performance
236 * matters - the code can just wait on the buffer
237 * lock and re-submit.
239 * 2) The current request is cached, but not
240 * readahead. ocfs2_buffer_uptodate() will return
241 * false anyway, so we'll wind up waiting on the
242 * buffer lock to do I/O. We re-check the request
243 * with after getting the lock to avoid a re-submit.
245 * 3) The current request is readahead (and so must
246 * also be a caching one). We short circuit if the
247 * buffer is locked (under I/O) and if it's in the
248 * uptodate cache. The re-check from #2 catches the
249 * case that the previous read-ahead completes just
250 * before our is-it-in-flight check.
253 if (!ignore_cache && !ocfs2_buffer_uptodate(inode, bh)) {
255 "bh (%llu), inode %llu not uptodate\n",
256 (unsigned long long)bh->b_blocknr,
257 (unsigned long long)OCFS2_I(inode)->ip_blkno);
258 /* We're using ignore_cache here to say
263 if (buffer_jbd(bh)) {
265 mlog(ML_BH_IO, "trying to sync read a jbd "
266 "managed bh (blocknr = %llu)\n",
267 (unsigned long long)bh->b_blocknr);
272 if (buffer_dirty(bh)) {
273 /* This should probably be a BUG, or
274 * at least return an error. */
275 mlog(ML_BH_IO, "asking me to sync read a dirty "
276 "buffer! (blocknr = %llu)\n",
277 (unsigned long long)bh->b_blocknr);
281 /* A read-ahead request was made - if the
282 * buffer is already under read-ahead from a
283 * previously submitted request than we are
285 if ((flags & OCFS2_BH_READAHEAD)
286 && ocfs2_buffer_read_ahead(inode, bh))
290 if (buffer_jbd(bh)) {
291 #ifdef CATCH_BH_JBD_RACES
292 mlog(ML_ERROR, "block %llu had the JBD bit set "
293 "while I was in lock_buffer!",
294 (unsigned long long)bh->b_blocknr);
302 /* Re-check ocfs2_buffer_uptodate() as a
303 * previously read-ahead buffer may have
304 * completed I/O while we were waiting for the
306 if (!(flags & OCFS2_BH_IGNORE_CACHE)
307 && !(flags & OCFS2_BH_READAHEAD)
308 && ocfs2_buffer_uptodate(inode, bh)) {
313 clear_buffer_uptodate(bh);
314 get_bh(bh); /* for end_buffer_read_sync() */
316 set_buffer_needs_validate(bh);
317 bh->b_end_io = end_buffer_read_sync;
325 for (i = (nr - 1); i >= 0; i--) {
328 if (!(flags & OCFS2_BH_READAHEAD)) {
329 /* We know this can't have changed as we hold the
330 * inode sem. Avoid doing any work on the bh if the
335 if (!buffer_uptodate(bh)) {
336 /* Status won't be cleared from here on out,
337 * so we can safely record this and loop back
338 * to cleanup the other buffers. Don't need to
339 * remove the clustered uptodate information
340 * for this bh as it's not marked locally
348 if (buffer_needs_validate(bh)) {
349 /* We never set NeedsValidate if the
350 * buffer was held by the journal, so
351 * that better not have changed */
352 BUG_ON(buffer_jbd(bh));
353 clear_buffer_needs_validate(bh);
354 status = validate(inode->i_sb, bh);
363 /* Always set the buffer in the cache, even if it was
364 * a forced read, or read-ahead which hasn't yet
366 ocfs2_set_buffer_uptodate(inode, bh);
368 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
370 mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n",
371 (unsigned long long)block, nr,
372 ((flags & OCFS2_BH_IGNORE_CACHE) || ignore_cache) ? "no" : "yes",
381 /* Check whether the blkno is the super block or one of the backups. */
382 static void ocfs2_check_super_or_backup(struct super_block *sb,
388 if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
391 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
392 backup_blkno = ocfs2_backup_super_blkno(sb, i);
393 if (backup_blkno == blkno)
401 * Write super block and backups doesn't need to collaborate with journal,
402 * so we don't need to lock ip_io_mutex and inode doesn't need to bea passed
403 * into this function.
405 int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
406 struct buffer_head *bh)
412 BUG_ON(buffer_jbd(bh));
413 ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
415 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
421 set_buffer_uptodate(bh);
423 /* remove from dirty list before I/O. */
424 clear_buffer_dirty(bh);
426 get_bh(bh); /* for end_buffer_write_sync() */
427 bh->b_end_io = end_buffer_write_sync;
428 submit_bh(WRITE, bh);
432 if (!buffer_uptodate(bh)) {