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/highmem.h>
30 #include <cluster/masklog.h>
38 #include "buffer_head_io.h"
39 #include "ocfs2_trace.h"
42 * Bits on bh->b_state used by ocfs2.
44 * These MUST be after the JBD2 bits. Hence, we use BH_JBDPrivateStart.
46 enum ocfs2_state_bits {
47 BH_NeedsValidate = BH_JBDPrivateStart,
50 /* Expand the magic b_state functions */
51 BUFFER_FNS(NeedsValidate, needs_validate);
53 int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
54 struct ocfs2_caching_info *ci)
58 trace_ocfs2_write_block((unsigned long long)bh->b_blocknr, ci);
60 BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
61 BUG_ON(buffer_jbd(bh));
63 /* No need to check for a soft readonly file system here. non
64 * journalled writes are only ever done on system files which
65 * can get modified during recovery even if read-only. */
66 if (ocfs2_is_hard_readonly(osb)) {
72 ocfs2_metadata_cache_io_lock(ci);
75 set_buffer_uptodate(bh);
77 /* remove from dirty list before I/O. */
78 clear_buffer_dirty(bh);
80 get_bh(bh); /* for end_buffer_write_sync() */
81 bh->b_end_io = end_buffer_write_sync;
86 if (buffer_uptodate(bh)) {
87 ocfs2_set_buffer_uptodate(ci, bh);
89 /* We don't need to remove the clustered uptodate
90 * information for this bh as it's not marked locally
96 ocfs2_metadata_cache_io_unlock(ci);
101 int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
102 unsigned int nr, struct buffer_head *bhs[])
106 struct buffer_head *bh;
108 trace_ocfs2_read_blocks_sync((unsigned long long)block, nr);
113 for (i = 0 ; i < nr ; i++) {
114 if (bhs[i] == NULL) {
115 bhs[i] = sb_getblk(osb->sb, block++);
116 if (bhs[i] == NULL) {
124 if (buffer_jbd(bh)) {
125 trace_ocfs2_read_blocks_sync_jbd(
126 (unsigned long long)bh->b_blocknr);
130 if (buffer_dirty(bh)) {
131 /* This should probably be a BUG, or
132 * at least return an error. */
134 "trying to sync read a dirty "
135 "buffer! (blocknr = %llu), skipping\n",
136 (unsigned long long)bh->b_blocknr);
141 if (buffer_jbd(bh)) {
143 "block %llu had the JBD bit set "
144 "while I was in lock_buffer!",
145 (unsigned long long)bh->b_blocknr);
149 clear_buffer_uptodate(bh);
150 get_bh(bh); /* for end_buffer_read_sync() */
151 bh->b_end_io = end_buffer_read_sync;
155 for (i = nr; i > 0; i--) {
158 /* No need to wait on the buffer if it's managed by JBD. */
162 if (!buffer_uptodate(bh)) {
163 /* Status won't be cleared from here on out,
164 * so we can safely record this and loop back
165 * to cleanup the other buffers. */
176 int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
177 struct buffer_head *bhs[], int flags,
178 int (*validate)(struct super_block *sb,
179 struct buffer_head *bh))
182 int i, ignore_cache = 0;
183 struct buffer_head *bh;
184 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
186 trace_ocfs2_read_blocks_begin(ci, (unsigned long long)block, nr, flags);
189 BUG_ON((flags & OCFS2_BH_READAHEAD) &&
190 (flags & OCFS2_BH_IGNORE_CACHE));
199 mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
210 ocfs2_metadata_cache_io_lock(ci);
211 for (i = 0 ; i < nr ; i++) {
212 if (bhs[i] == NULL) {
213 bhs[i] = sb_getblk(sb, block++);
214 if (bhs[i] == NULL) {
215 ocfs2_metadata_cache_io_unlock(ci);
222 ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE);
224 /* There are three read-ahead cases here which we need to
225 * be concerned with. All three assume a buffer has
226 * previously been submitted with OCFS2_BH_READAHEAD
227 * and it hasn't yet completed I/O.
229 * 1) The current request is sync to disk. This rarely
230 * happens these days, and never when performance
231 * matters - the code can just wait on the buffer
232 * lock and re-submit.
234 * 2) The current request is cached, but not
235 * readahead. ocfs2_buffer_uptodate() will return
236 * false anyway, so we'll wind up waiting on the
237 * buffer lock to do I/O. We re-check the request
238 * with after getting the lock to avoid a re-submit.
240 * 3) The current request is readahead (and so must
241 * also be a caching one). We short circuit if the
242 * buffer is locked (under I/O) and if it's in the
243 * uptodate cache. The re-check from #2 catches the
244 * case that the previous read-ahead completes just
245 * before our is-it-in-flight check.
248 if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) {
249 trace_ocfs2_read_blocks_from_disk(
250 (unsigned long long)bh->b_blocknr,
251 (unsigned long long)ocfs2_metadata_cache_owner(ci));
252 /* We're using ignore_cache here to say
257 trace_ocfs2_read_blocks_bh((unsigned long long)bh->b_blocknr,
258 ignore_cache, buffer_jbd(bh), buffer_dirty(bh));
260 if (buffer_jbd(bh)) {
265 if (buffer_dirty(bh)) {
266 /* This should probably be a BUG, or
267 * at least return an error. */
271 /* A read-ahead request was made - if the
272 * buffer is already under read-ahead from a
273 * previously submitted request than we are
275 if ((flags & OCFS2_BH_READAHEAD)
276 && ocfs2_buffer_read_ahead(ci, bh))
280 if (buffer_jbd(bh)) {
281 #ifdef CATCH_BH_JBD_RACES
282 mlog(ML_ERROR, "block %llu had the JBD bit set "
283 "while I was in lock_buffer!",
284 (unsigned long long)bh->b_blocknr);
292 /* Re-check ocfs2_buffer_uptodate() as a
293 * previously read-ahead buffer may have
294 * completed I/O while we were waiting for the
296 if (!(flags & OCFS2_BH_IGNORE_CACHE)
297 && !(flags & OCFS2_BH_READAHEAD)
298 && ocfs2_buffer_uptodate(ci, bh)) {
303 clear_buffer_uptodate(bh);
304 get_bh(bh); /* for end_buffer_read_sync() */
306 set_buffer_needs_validate(bh);
307 bh->b_end_io = end_buffer_read_sync;
315 for (i = (nr - 1); i >= 0; i--) {
318 if (!(flags & OCFS2_BH_READAHEAD)) {
319 /* We know this can't have changed as we hold the
320 * owner sem. Avoid doing any work on the bh if the
325 if (!buffer_uptodate(bh)) {
326 /* Status won't be cleared from here on out,
327 * so we can safely record this and loop back
328 * to cleanup the other buffers. Don't need to
329 * remove the clustered uptodate information
330 * for this bh as it's not marked locally
338 if (buffer_needs_validate(bh)) {
339 /* We never set NeedsValidate if the
340 * buffer was held by the journal, so
341 * that better not have changed */
342 BUG_ON(buffer_jbd(bh));
343 clear_buffer_needs_validate(bh);
344 status = validate(sb, bh);
353 /* Always set the buffer in the cache, even if it was
354 * a forced read, or read-ahead which hasn't yet
356 ocfs2_set_buffer_uptodate(ci, bh);
358 ocfs2_metadata_cache_io_unlock(ci);
360 trace_ocfs2_read_blocks_end((unsigned long long)block, nr,
361 flags, ignore_cache);
368 /* Check whether the blkno is the super block or one of the backups. */
369 static void ocfs2_check_super_or_backup(struct super_block *sb,
375 if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
378 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
379 backup_blkno = ocfs2_backup_super_blkno(sb, i);
380 if (backup_blkno == blkno)
388 * Write super block and backups doesn't need to collaborate with journal,
389 * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed
390 * into this function.
392 int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
393 struct buffer_head *bh)
396 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
398 BUG_ON(buffer_jbd(bh));
399 ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
401 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
408 set_buffer_uptodate(bh);
410 /* remove from dirty list before I/O. */
411 clear_buffer_dirty(bh);
413 get_bh(bh); /* for end_buffer_write_sync() */
414 bh->b_end_io = end_buffer_write_sync;
415 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check);
416 submit_bh(WRITE, bh);
420 if (!buffer_uptodate(bh)) {