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;
82 submit_bh(REQ_OP_WRITE, 0, bh);
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 /* Caller must provide a bhs[] with all NULL or non-NULL entries, so it
102 * will be easier to handle read failure.
104 int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
105 unsigned int nr, struct buffer_head *bhs[])
109 struct buffer_head *bh;
112 trace_ocfs2_read_blocks_sync((unsigned long long)block, nr);
117 /* Don't put buffer head and re-assign it to NULL if it is allocated
118 * outside since the caller can't be aware of this alternation!
120 new_bh = (bhs[0] == NULL);
122 for (i = 0 ; i < nr ; i++) {
123 if (bhs[i] == NULL) {
124 bhs[i] = sb_getblk(osb->sb, block++);
125 if (bhs[i] == NULL) {
133 if (buffer_jbd(bh)) {
134 trace_ocfs2_read_blocks_sync_jbd(
135 (unsigned long long)bh->b_blocknr);
139 if (buffer_dirty(bh)) {
140 /* This should probably be a BUG, or
141 * at least return an error. */
143 "trying to sync read a dirty "
144 "buffer! (blocknr = %llu), skipping\n",
145 (unsigned long long)bh->b_blocknr);
150 if (buffer_jbd(bh)) {
151 #ifdef CATCH_BH_JBD_RACES
153 "block %llu had the JBD bit set "
154 "while I was in lock_buffer!",
155 (unsigned long long)bh->b_blocknr);
163 get_bh(bh); /* for end_buffer_read_sync() */
164 bh->b_end_io = end_buffer_read_sync;
165 submit_bh(REQ_OP_READ, 0, bh);
169 for (i = nr; i > 0; i--) {
172 if (unlikely(status)) {
174 /* If middle bh fails, let previous bh
175 * finish its read and then put it to
182 } else if (bh && buffer_uptodate(bh)) {
183 clear_buffer_uptodate(bh);
188 /* No need to wait on the buffer if it's managed by JBD. */
192 if (!buffer_uptodate(bh)) {
193 /* Status won't be cleared from here on out,
194 * so we can safely record this and loop back
195 * to cleanup the other buffers. */
205 /* Caller must provide a bhs[] with all NULL or non-NULL entries, so it
206 * will be easier to handle read failure.
208 int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
209 struct buffer_head *bhs[], int flags,
210 int (*validate)(struct super_block *sb,
211 struct buffer_head *bh))
214 int i, ignore_cache = 0;
215 struct buffer_head *bh;
216 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
219 trace_ocfs2_read_blocks_begin(ci, (unsigned long long)block, nr, flags);
222 BUG_ON((flags & OCFS2_BH_READAHEAD) &&
223 (flags & OCFS2_BH_IGNORE_CACHE));
232 mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
243 /* Don't put buffer head and re-assign it to NULL if it is allocated
244 * outside since the caller can't be aware of this alternation!
246 new_bh = (bhs[0] == NULL);
248 ocfs2_metadata_cache_io_lock(ci);
249 for (i = 0 ; i < nr ; i++) {
250 if (bhs[i] == NULL) {
251 bhs[i] = sb_getblk(sb, block++);
252 if (bhs[i] == NULL) {
253 ocfs2_metadata_cache_io_unlock(ci);
256 /* Don't forget to put previous bh! */
261 ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE);
263 /* There are three read-ahead cases here which we need to
264 * be concerned with. All three assume a buffer has
265 * previously been submitted with OCFS2_BH_READAHEAD
266 * and it hasn't yet completed I/O.
268 * 1) The current request is sync to disk. This rarely
269 * happens these days, and never when performance
270 * matters - the code can just wait on the buffer
271 * lock and re-submit.
273 * 2) The current request is cached, but not
274 * readahead. ocfs2_buffer_uptodate() will return
275 * false anyway, so we'll wind up waiting on the
276 * buffer lock to do I/O. We re-check the request
277 * with after getting the lock to avoid a re-submit.
279 * 3) The current request is readahead (and so must
280 * also be a caching one). We short circuit if the
281 * buffer is locked (under I/O) and if it's in the
282 * uptodate cache. The re-check from #2 catches the
283 * case that the previous read-ahead completes just
284 * before our is-it-in-flight check.
287 if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) {
288 trace_ocfs2_read_blocks_from_disk(
289 (unsigned long long)bh->b_blocknr,
290 (unsigned long long)ocfs2_metadata_cache_owner(ci));
291 /* We're using ignore_cache here to say
296 trace_ocfs2_read_blocks_bh((unsigned long long)bh->b_blocknr,
297 ignore_cache, buffer_jbd(bh), buffer_dirty(bh));
299 if (buffer_jbd(bh)) {
304 if (buffer_dirty(bh)) {
305 /* This should probably be a BUG, or
306 * at least return an error. */
310 /* A read-ahead request was made - if the
311 * buffer is already under read-ahead from a
312 * previously submitted request than we are
314 if ((flags & OCFS2_BH_READAHEAD)
315 && ocfs2_buffer_read_ahead(ci, bh))
319 if (buffer_jbd(bh)) {
320 #ifdef CATCH_BH_JBD_RACES
321 mlog(ML_ERROR, "block %llu had the JBD bit set "
322 "while I was in lock_buffer!",
323 (unsigned long long)bh->b_blocknr);
331 /* Re-check ocfs2_buffer_uptodate() as a
332 * previously read-ahead buffer may have
333 * completed I/O while we were waiting for the
335 if (!(flags & OCFS2_BH_IGNORE_CACHE)
336 && !(flags & OCFS2_BH_READAHEAD)
337 && ocfs2_buffer_uptodate(ci, bh)) {
342 get_bh(bh); /* for end_buffer_read_sync() */
344 set_buffer_needs_validate(bh);
345 bh->b_end_io = end_buffer_read_sync;
346 submit_bh(REQ_OP_READ, 0, bh);
352 for (i = (nr - 1); i >= 0; i--) {
355 if (!(flags & OCFS2_BH_READAHEAD)) {
356 if (unlikely(status)) {
357 /* Clear the buffers on error including those
358 * ever succeeded in reading
361 /* If middle bh fails, let previous bh
362 * finish its read and then put it to
369 } else if (bh && buffer_uptodate(bh)) {
370 clear_buffer_uptodate(bh);
374 /* We know this can't have changed as we hold the
375 * owner sem. Avoid doing any work on the bh if the
380 if (!buffer_uptodate(bh)) {
381 /* Status won't be cleared from here on out,
382 * so we can safely record this and loop back
383 * to cleanup the other buffers. Don't need to
384 * remove the clustered uptodate information
385 * for this bh as it's not marked locally
388 clear_buffer_needs_validate(bh);
392 if (buffer_needs_validate(bh)) {
393 /* We never set NeedsValidate if the
394 * buffer was held by the journal, so
395 * that better not have changed */
396 BUG_ON(buffer_jbd(bh));
397 clear_buffer_needs_validate(bh);
398 status = validate(sb, bh);
404 /* Always set the buffer in the cache, even if it was
405 * a forced read, or read-ahead which hasn't yet
407 ocfs2_set_buffer_uptodate(ci, bh);
409 ocfs2_metadata_cache_io_unlock(ci);
411 trace_ocfs2_read_blocks_end((unsigned long long)block, nr,
412 flags, ignore_cache);
419 /* Check whether the blkno is the super block or one of the backups. */
420 static void ocfs2_check_super_or_backup(struct super_block *sb,
426 if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
429 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
430 backup_blkno = ocfs2_backup_super_blkno(sb, i);
431 if (backup_blkno == blkno)
439 * Write super block and backups doesn't need to collaborate with journal,
440 * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed
441 * into this function.
443 int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
444 struct buffer_head *bh)
447 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
449 BUG_ON(buffer_jbd(bh));
450 ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
452 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
459 set_buffer_uptodate(bh);
461 /* remove from dirty list before I/O. */
462 clear_buffer_dirty(bh);
464 get_bh(bh); /* for end_buffer_write_sync() */
465 bh->b_end_io = end_buffer_write_sync;
466 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check);
467 submit_bh(REQ_OP_WRITE, 0, bh);
471 if (!buffer_uptodate(bh)) {