[net-next-2.6.git] / fs / ocfs2 / buffer_head_io.c

/* -*- mode: c; c-basic-offset: 8; -*-
 * vim: noexpandtab sw=8 ts=8 sts=0:
 *
 * io.c
 *
 * Buffer cache handling
 *
 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

#include <linux/fs.h>
#include <linux/types.h>
#include <linux/slab.h>
#include <linux/highmem.h>

#include <cluster/masklog.h>

#include "ocfs2.h"

#include "alloc.h"
#include "inode.h"
#include "journal.h"
#include "uptodate.h"

#include "buffer_head_io.h"

int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
		      struct inode *inode)
{
	int ret = 0;

	mlog_entry("(bh->b_blocknr = %llu, inode=%p)\n",
		   (unsigned long long)bh->b_blocknr, inode);

	BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
	BUG_ON(buffer_jbd(bh));

	/* No need to check for a soft readonly file system here. non
	 * journalled writes are only ever done on system files which
	 * can get modified during recovery even if read-only. */
	if (ocfs2_is_hard_readonly(osb)) {
		ret = -EROFS;
		goto out;
	}

	mutex_lock(&OCFS2_I(inode)->ip_io_mutex);

	lock_buffer(bh);
	set_buffer_uptodate(bh);

	/* remove from dirty list before I/O. */
	clear_buffer_dirty(bh);

	get_bh(bh); /* for end_buffer_write_sync() */                   
	bh->b_end_io = end_buffer_write_sync;
	submit_bh(WRITE, bh);

	wait_on_buffer(bh);

	if (buffer_uptodate(bh)) {
		ocfs2_set_buffer_uptodate(inode, bh);
	} else {
		/* We don't need to remove the clustered uptodate
		 * information for this bh as it's not marked locally
		 * uptodate. */
		ret = -EIO;
		brelse(bh);
	}

	mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
out:
	mlog_exit(ret);
	return ret;
}

int ocfs2_read_blocks(struct ocfs2_super *osb, u64 block, int nr,
		      struct buffer_head *bhs[], int flags,
		      struct inode *inode)
{
	int status = 0;
	struct super_block *sb;
	int i, ignore_cache = 0;
	struct buffer_head *bh;

	mlog_entry("(block=(%llu), nr=(%d), flags=%d, inode=%p)\n",
		   (unsigned long long)block, nr, flags, inode);

	BUG_ON((flags & OCFS2_BH_READAHEAD) &&
	       (!inode || !(flags & OCFS2_BH_CACHED)));

	if (osb == NULL || osb->sb == NULL || bhs == NULL) {
		status = -EINVAL;
		mlog_errno(status);
		goto bail;
	}

	if (nr < 0) {
		mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
		status = -EINVAL;
		mlog_errno(status);
		goto bail;
	}

	if (nr == 0) {
		mlog(ML_BH_IO, "No buffers will be read!\n");
		status = 0;
		goto bail;
	}

	sb = osb->sb;

	if (flags & OCFS2_BH_CACHED && !inode)
		flags &= ~OCFS2_BH_CACHED;

	if (inode)
		mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
	for (i = 0 ; i < nr ; i++) {
		if (bhs[i] == NULL) {
			bhs[i] = sb_getblk(sb, block++);
			if (bhs[i] == NULL) {
				if (inode)
					mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
				status = -EIO;
				mlog_errno(status);
				goto bail;
			}
		}
		bh = bhs[i];
		ignore_cache = 0;

		/* There are three read-ahead cases here which we need to
		 * be concerned with. All three assume a buffer has
		 * previously been submitted with OCFS2_BH_READAHEAD
		 * and it hasn't yet completed I/O.
		 *
		 * 1) The current request is sync to disk. This rarely
		 *    happens these days, and never when performance
		 *    matters - the code can just wait on the buffer
		 *    lock and re-submit.
		 *
		 * 2) The current request is cached, but not
		 *    readahead. ocfs2_buffer_uptodate() will return
		 *    false anyway, so we'll wind up waiting on the
		 *    buffer lock to do I/O. We re-check the request
		 *    with after getting the lock to avoid a re-submit.
		 *
		 * 3) The current request is readahead (and so must
		 *    also be a caching one). We short circuit if the
		 *    buffer is locked (under I/O) and if it's in the
		 *    uptodate cache. The re-check from #2 catches the
		 *    case that the previous read-ahead completes just
		 *    before our is-it-in-flight check.
		 */

		if (flags & OCFS2_BH_CACHED &&
		    !ocfs2_buffer_uptodate(inode, bh)) {
			mlog(ML_UPTODATE,
			     "bh (%llu), inode %llu not uptodate\n",
			     (unsigned long long)bh->b_blocknr,
			     (unsigned long long)OCFS2_I(inode)->ip_blkno);
			ignore_cache = 1;
		}

		/* XXX: Can we ever get this and *not* have the cached
		 * flag set? */
		if (buffer_jbd(bh)) {
			if (!(flags & OCFS2_BH_CACHED) || ignore_cache)
				mlog(ML_BH_IO, "trying to sync read a jbd "
					       "managed bh (blocknr = %llu)\n",
				     (unsigned long long)bh->b_blocknr);
			continue;
		}

		if (!(flags & OCFS2_BH_CACHED) || ignore_cache) {
			if (buffer_dirty(bh)) {
				/* This should probably be a BUG, or
				 * at least return an error. */
				mlog(ML_BH_IO, "asking me to sync read a dirty "
					       "buffer! (blocknr = %llu)\n",
				     (unsigned long long)bh->b_blocknr);
				continue;
			}

			/* A read-ahead request was made - if the
			 * buffer is already under read-ahead from a
			 * previously submitted request than we are
			 * done here. */
			if ((flags & OCFS2_BH_READAHEAD)
			    && ocfs2_buffer_read_ahead(inode, bh))
				continue;

			lock_buffer(bh);
			if (buffer_jbd(bh)) {
#ifdef CATCH_BH_JBD_RACES
				mlog(ML_ERROR, "block %llu had the JBD bit set "
					       "while I was in lock_buffer!",
				     (unsigned long long)bh->b_blocknr);
				BUG();
#else
				unlock_buffer(bh);
				continue;
#endif
			}

			/* Re-check ocfs2_buffer_uptodate() as a
			 * previously read-ahead buffer may have
			 * completed I/O while we were waiting for the
			 * buffer lock. */
			if ((flags & OCFS2_BH_CACHED)
			    && !(flags & OCFS2_BH_READAHEAD)
			    && ocfs2_buffer_uptodate(inode, bh)) {
				unlock_buffer(bh);
				continue;
			}

			clear_buffer_uptodate(bh);
			get_bh(bh); /* for end_buffer_read_sync() */
			bh->b_end_io = end_buffer_read_sync;
			submit_bh(READ, bh);
			continue;
		}
	}

	status = 0;

	for (i = (nr - 1); i >= 0; i--) {
		bh = bhs[i];

		if (!(flags & OCFS2_BH_READAHEAD)) {
			/* We know this can't have changed as we hold the
			 * inode sem. Avoid doing any work on the bh if the
			 * journal has it. */
			if (!buffer_jbd(bh))
				wait_on_buffer(bh);

			if (!buffer_uptodate(bh)) {
				/* Status won't be cleared from here on out,
				 * so we can safely record this and loop back
				 * to cleanup the other buffers. Don't need to
				 * remove the clustered uptodate information
				 * for this bh as it's not marked locally
				 * uptodate. */
				status = -EIO;
				brelse(bh);
				bhs[i] = NULL;
				continue;
			}
		}

		/* Always set the buffer in the cache, even if it was
		 * a forced read, or read-ahead which hasn't yet
		 * completed. */
		if (inode)
			ocfs2_set_buffer_uptodate(inode, bh);
	}
	if (inode)
		mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);

	mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n", 
	     (unsigned long long)block, nr,
	     (!(flags & OCFS2_BH_CACHED) || ignore_cache) ? "no" : "yes", flags);

bail:

	mlog_exit(status);
	return status;
}

/* Check whether the blkno is the super block or one of the backups. */
static void ocfs2_check_super_or_backup(struct super_block *sb,
					sector_t blkno)
{
	int i;
	u64 backup_blkno;

	if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
		return;

	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
		backup_blkno = ocfs2_backup_super_blkno(sb, i);
		if (backup_blkno == blkno)
			return;
	}

	BUG();
}

/*
 * Write super block and backups doesn't need to collaborate with journal,
 * so we don't need to lock ip_io_mutex and inode doesn't need to bea passed
 * into this function.
 */
int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
				struct buffer_head *bh)
{
	int ret = 0;

	mlog_entry_void();

	BUG_ON(buffer_jbd(bh));
	ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);

	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
		ret = -EROFS;
		goto out;
	}

	lock_buffer(bh);
	set_buffer_uptodate(bh);

	/* remove from dirty list before I/O. */
	clear_buffer_dirty(bh);

	get_bh(bh); /* for end_buffer_write_sync() */
	bh->b_end_io = end_buffer_write_sync;
	submit_bh(WRITE, bh);

	wait_on_buffer(bh);

	if (!buffer_uptodate(bh)) {
		ret = -EIO;
		brelse(bh);
	}

out:
	mlog_exit(ret);
	return ret;
}
Commit	Line	Data
ccd979bd MF	1	/* -- mode: c; c-basic-offset: 8; --
	2	* vim: noexpandtab sw=8 ts=8 sts=0:
	3	*
	4	* io.c
	5	*
	6	* Buffer cache handling
	7	*
	8	* Copyright (C) 2002, 2004 Oracle. All rights reserved.
	9	*
	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.
	14	*
	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.
	19	*
	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.
	24	*/
	25
	26	#include <linux/fs.h>
	27	#include <linux/types.h>
	28	#include <linux/slab.h>
	29	#include <linux/highmem.h>
	30
	31	#include <cluster/masklog.h>
	32
	33	#include "ocfs2.h"
	34
	35	#include "alloc.h"
	36	#include "inode.h"
	37	#include "journal.h"
	38	#include "uptodate.h"
	39
	40	#include "buffer_head_io.h"
	41
	42	int ocfs2_write_block(struct ocfs2_super osb, struct buffer_head bh,
	43	struct inode *inode)
	44	{
	45	int ret = 0;
	46
	47	mlog_entry("(bh->b_blocknr = %llu, inode=%p)\n",
	48	(unsigned long long)bh->b_blocknr, inode);
	49
	50	BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
	51	BUG_ON(buffer_jbd(bh));
	52
	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)) {
	57	ret = -EROFS;
	58	goto out;
	59	}
	60
251b6ecc	61	mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
ccd979bd MF	62
	63	lock_buffer(bh);
	64	set_buffer_uptodate(bh);
	65
	66	/* remove from dirty list before I/O. */
	67	clear_buffer_dirty(bh);
	68
	69	get_bh(bh); /* for end_buffer_write_sync() */
	70	bh->b_end_io = end_buffer_write_sync;
	71	submit_bh(WRITE, bh);
	72
	73	wait_on_buffer(bh);
	74
	75	if (buffer_uptodate(bh)) {
	76	ocfs2_set_buffer_uptodate(inode, bh);
	77	} else {
	78	/* We don't need to remove the clustered uptodate
	79	* information for this bh as it's not marked locally
	80	* uptodate. */
	81	ret = -EIO;
	82	brelse(bh);
	83	}
	84
251b6ecc	85	mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
ccd979bd MF	86	out:
	87	mlog_exit(ret);
	88	return ret;
	89	}
	90
	91	int ocfs2_read_blocks(struct ocfs2_super *osb, u64 block, int nr,
	92	struct buffer_head *bhs[], int flags,
	93	struct inode *inode)
	94	{
	95	int status = 0;
	96	struct super_block *sb;
	97	int i, ignore_cache = 0;
	98	struct buffer_head *bh;
	99
b0697053 MF	100	mlog_entry("(block=(%llu), nr=(%d), flags=%d, inode=%p)\n",
b0697053 MF	101	(unsigned long long)block, nr, flags, inode);
ccd979bd	102
aa958874 MF	103	BUG_ON((flags & OCFS2_BH_READAHEAD) &&
	104	(!inode \|\| !(flags & OCFS2_BH_CACHED)));
	105
ccd979bd MF	106	if (osb == NULL \|\| osb->sb == NULL \|\| bhs == NULL) {
	107	status = -EINVAL;
	108	mlog_errno(status);
	109	goto bail;
	110	}
	111
	112	if (nr < 0) {
	113	mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
	114	status = -EINVAL;
	115	mlog_errno(status);
	116	goto bail;
	117	}
	118
	119	if (nr == 0) {
	120	mlog(ML_BH_IO, "No buffers will be read!\n");
	121	status = 0;
	122	goto bail;
	123	}
	124
	125	sb = osb->sb;
	126
	127	if (flags & OCFS2_BH_CACHED && !inode)
	128	flags &= ~OCFS2_BH_CACHED;
	129
	130	if (inode)
251b6ecc	131	mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
ccd979bd MF	132	for (i = 0 ; i < nr ; i++) {
	133	if (bhs[i] == NULL) {
	134	bhs[i] = sb_getblk(sb, block++);
	135	if (bhs[i] == NULL) {
	136	if (inode)
251b6ecc	137	mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
ccd979bd MF	138	status = -EIO;
	139	mlog_errno(status);
	140	goto bail;
	141	}
	142	}
	143	bh = bhs[i];
	144	ignore_cache = 0;
	145
aa958874 MF	146	/* There are three read-ahead cases here which we need to
	147	* be concerned with. All three assume a buffer has
	148	* previously been submitted with OCFS2_BH_READAHEAD
	149	* and it hasn't yet completed I/O.
	150	*
	151	* 1) The current request is sync to disk. This rarely
	152	* happens these days, and never when performance
	153	* matters - the code can just wait on the buffer
	154	* lock and re-submit.
	155	*
	156	* 2) The current request is cached, but not
	157	* readahead. ocfs2_buffer_uptodate() will return
	158	* false anyway, so we'll wind up waiting on the
	159	* buffer lock to do I/O. We re-check the request
	160	* with after getting the lock to avoid a re-submit.
	161	*
	162	* 3) The current request is readahead (and so must
	163	* also be a caching one). We short circuit if the
	164	* buffer is locked (under I/O) and if it's in the
	165	* uptodate cache. The re-check from #2 catches the
	166	* case that the previous read-ahead completes just
	167	* before our is-it-in-flight check.
	168	*/
	169
ccd979bd MF	170	if (flags & OCFS2_BH_CACHED &&
	171	!ocfs2_buffer_uptodate(inode, bh)) {
	172	mlog(ML_UPTODATE,
b0697053	173	"bh (%llu), inode %llu not uptodate\n",
ccd979bd	174	(unsigned long long)bh->b_blocknr,
b0697053	175	(unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd MF	176	ignore_cache = 1;
	177	}
	178
	179	/* XXX: Can we ever get this and not have the cached
	180	* flag set? */
	181	if (buffer_jbd(bh)) {
	182	if (!(flags & OCFS2_BH_CACHED) \|\| ignore_cache)
	183	mlog(ML_BH_IO, "trying to sync read a jbd "
	184	"managed bh (blocknr = %llu)\n",
	185	(unsigned long long)bh->b_blocknr);
	186	continue;
	187	}
	188
	189	if (!(flags & OCFS2_BH_CACHED) \|\| ignore_cache) {
	190	if (buffer_dirty(bh)) {
	191	/* This should probably be a BUG, or
	192	* at least return an error. */
	193	mlog(ML_BH_IO, "asking me to sync read a dirty "
	194	"buffer! (blocknr = %llu)\n",
	195	(unsigned long long)bh->b_blocknr);
	196	continue;
	197	}
	198
aa958874 MF	199	/* A read-ahead request was made - if the
	200	* buffer is already under read-ahead from a
	201	* previously submitted request than we are
	202	* done here. */
	203	if ((flags & OCFS2_BH_READAHEAD)
	204	&& ocfs2_buffer_read_ahead(inode, bh))
	205	continue;
	206
ccd979bd MF	207	lock_buffer(bh);
	208	if (buffer_jbd(bh)) {
	209	#ifdef CATCH_BH_JBD_RACES
	210	mlog(ML_ERROR, "block %llu had the JBD bit set "
	211	"while I was in lock_buffer!",
	212	(unsigned long long)bh->b_blocknr);
	213	BUG();
	214	#else
	215	unlock_buffer(bh);
	216	continue;
	217	#endif
	218	}
aa958874 MF	219
	220	/* Re-check ocfs2_buffer_uptodate() as a
	221	* previously read-ahead buffer may have
	222	* completed I/O while we were waiting for the
	223	* buffer lock. */
	224	if ((flags & OCFS2_BH_CACHED)
	225	&& !(flags & OCFS2_BH_READAHEAD)
	226	&& ocfs2_buffer_uptodate(inode, bh)) {
	227	unlock_buffer(bh);
	228	continue;
	229	}
	230
ccd979bd MF	231	clear_buffer_uptodate(bh);
	232	get_bh(bh); /* for end_buffer_read_sync() */
	233	bh->b_end_io = end_buffer_read_sync;
aa958874	234	submit_bh(READ, bh);
ccd979bd MF	235	continue;
	236	}
	237	}
	238
	239	status = 0;
	240
	241	for (i = (nr - 1); i >= 0; i--) {
	242	bh = bhs[i];
	243
aa958874 MF	244	if (!(flags & OCFS2_BH_READAHEAD)) {
	245	/* We know this can't have changed as we hold the
	246	* inode sem. Avoid doing any work on the bh if the
	247	* journal has it. */
	248	if (!buffer_jbd(bh))
	249	wait_on_buffer(bh);
	250
	251	if (!buffer_uptodate(bh)) {
	252	/* Status won't be cleared from here on out,
	253	* so we can safely record this and loop back
	254	* to cleanup the other buffers. Don't need to
	255	* remove the clustered uptodate information
	256	* for this bh as it's not marked locally
	257	* uptodate. */
	258	status = -EIO;
	259	brelse(bh);
	260	bhs[i] = NULL;
	261	continue;
	262	}
ccd979bd MF	263	}
ccd979bd MF	264
aa958874 MF	265	/* Always set the buffer in the cache, even if it was
	266	* a forced read, or read-ahead which hasn't yet
	267	* completed. */
ccd979bd MF	268	if (inode)
	269	ocfs2_set_buffer_uptodate(inode, bh);
	270	}
	271	if (inode)
251b6ecc	272	mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
ccd979bd	273
aa958874	274	mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n",
b0697053	275	(unsigned long long)block, nr,
aa958874	276	(!(flags & OCFS2_BH_CACHED) \|\| ignore_cache) ? "no" : "yes", flags);
ccd979bd MF	277
	278	bail:
	279
	280	mlog_exit(status);
	281	return status;
	282	}
d659072f TM	283
	284	/* Check whether the blkno is the super block or one of the backups. */
	285	static void ocfs2_check_super_or_backup(struct super_block *sb,
	286	sector_t blkno)
	287	{
	288	int i;
	289	u64 backup_blkno;
	290
	291	if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
	292	return;
	293
	294	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
	295	backup_blkno = ocfs2_backup_super_blkno(sb, i);
	296	if (backup_blkno == blkno)
	297	return;
	298	}
	299
	300	BUG();
	301	}
	302
	303	/*
	304	* Write super block and backups doesn't need to collaborate with journal,
	305	* so we don't need to lock ip_io_mutex and inode doesn't need to bea passed
	306	* into this function.
	307	*/
	308	int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
	309	struct buffer_head *bh)
	310	{
	311	int ret = 0;
	312
	313	mlog_entry_void();
	314
	315	BUG_ON(buffer_jbd(bh));
	316	ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
	317
	318	if (ocfs2_is_hard_readonly(osb) \|\| ocfs2_is_soft_readonly(osb)) {
	319	ret = -EROFS;
	320	goto out;
	321	}
	322
	323	lock_buffer(bh);
	324	set_buffer_uptodate(bh);
	325
	326	/* remove from dirty list before I/O. */
	327	clear_buffer_dirty(bh);
	328
	329	get_bh(bh); /* for end_buffer_write_sync() */
	330	bh->b_end_io = end_buffer_write_sync;
	331	submit_bh(WRITE, bh);
	332
	333	wait_on_buffer(bh);
	334
	335	if (!buffer_uptodate(bh)) {
	336	ret = -EIO;
	337	brelse(bh);
	338	}
	339
	340	out:
	341	mlog_exit(ret);
	342	return ret;
	343	}