[Ocfs2-tools-devel] [patch 07/11] Modify file read and write.

[tao.ma]

Mark Fasheh wrote:
> On Thu, Aug 16, 2007 at 01:55:47AM +0800, tao.ma wrote:
>   
>> Change ocfs2_file_read and ocfs2_file_write accordingly.
>> ===================================================================
>> --- test.ocfs2-tools.orig/libocfs2/fileio.c	2007-08-16 00:27:42.000000000 -0400
>> +++ test.ocfs2-tools/libocfs2/fileio.c	2007-08-16 00:36:28.000000000 -0400
>> @@ -128,7 +128,7 @@ errcode_t ocfs2_file_read(ocfs2_cached_i
>>  	errcode_t	ret = 0;
>>  	char		*ptr = (char *) buf;
>>  	uint32_t	wanted_blocks;
>> -	uint32_t	contig_blocks;
>> +	uint64_t	contig_blocks;
>>  	uint64_t	v_blkno;
>>  	uint64_t	p_blkno;
>>  	uint32_t	tmp;
>> @@ -162,9 +162,15 @@ errcode_t ocfs2_file_read(ocfs2_cached_i
>>  		if (contig_blocks > wanted_blocks)
>>  			contig_blocks = wanted_blocks;
>>  
>> -		ret = io_read_block(fs->fs_io, p_blkno, contig_blocks, ptr);
>> -		if (ret)
>> -			return ret;
>> +		if (!p_blkno) {
>> +			/* we meet with a hole, just empty the content.*/
>> +			memset(ptr, 0, contig_blocks * fs->fs_blocksize);
>> +		} else {
>> +			ret = io_read_block(fs->fs_io, p_blkno,
>> +					    contig_blocks, ptr);
>> +			if (ret)
>> +				return ret;
>> +		}
>>  
>>  		*got += (contig_blocks <<
>>  			 OCFS2_RAW_SB(fs->fs_super)->s_blocksize_bits);
>> @@ -184,6 +190,42 @@ errcode_t ocfs2_file_read(ocfs2_cached_i
>>  	return ret;
>>  }
>>  
>> +/*
>> + * Emtpy the clusters on the disk.
>> + *
>> + * The "start_blk" is aligned with the cluster border since it is got by
>> + * ocfs2_new_clusters.
>> + */
>> +static errcode_t empty_clusters(ocfs2_filesys *fs,
>> +				uint64_t start_blk,
>> +				uint32_t num_clusters)
>> +{
>> +	errcode_t ret;
>> +	char *buf = NULL;
>> +	int bpc = fs->fs_clustersize / fs->fs_blocksize;
>> +
>> +	ret = ocfs2_malloc_blocks(fs->fs_io, bpc, &buf);
>> +	if (ret)
>> +		goto bail;
>> +
>> +	memset(buf, 0, fs->fs_clustersize);
>> +
>> +	while (num_clusters) {
>> +		ret = io_write_block(fs->fs_io, start_blk, bpc, buf);
>> +		if (ret)
>> +			goto bail;
>> +
>> +		num_clusters--;
>> +		start_blk += bpc;
>> +	}
>> +
>> +bail:
>> +	if (buf)
>> +		ocfs2_free(&buf);
>> +
>> +	return ret;
>> +}
>> +
>>  errcode_t ocfs2_file_write(ocfs2_cached_inode *ci, void *buf, uint32_t count,
>>  			   uint64_t offset, uint32_t *wrote)
>>  {
>> @@ -191,12 +233,14 @@ errcode_t ocfs2_file_write(ocfs2_cached_
>>  	errcode_t	ret = 0;
>>  	char		*ptr = (char *) buf;
>>  	uint32_t	wanted_blocks;
>> -	uint32_t	contig_blocks;
>> +	uint64_t	contig_blocks;
>>  	uint64_t	v_blkno;
>>  	uint64_t	p_blkno;
>>  	uint32_t	tmp;
>>  	uint64_t	num_blocks;
>>  	int		bs_bits = OCFS2_RAW_SB(fs->fs_super)->s_blocksize_bits;
>> +	uint64_t	ino = ci->ci_blkno;
>> +	uint32_t	n_clusters, cluster_begin, cluster_end;
>>  
>>  	/* o_direct requires aligned io */
>>  	tmp = fs->fs_blocksize - 1;
>> @@ -225,6 +269,50 @@ errcode_t ocfs2_file_write(ocfs2_cached_
>>  		if (contig_blocks > wanted_blocks)
>>  			contig_blocks = wanted_blocks;
>>  
>> +	 	if (!p_blkno) {
>> +			cluster_begin = ocfs2_blocks_to_clusters(fs, v_blkno);
>> +			cluster_end = ocfs2_blocks_to_clusters(fs,
>> +						v_blkno + contig_blocks -1);
>> +			n_clusters = cluster_end - cluster_begin + 1;
>> +			ret = ocfs2_new_clusters(fs, 1, n_clusters, &p_blkno,
>> +						 &n_clusters);
>> +			if (ret)
>> +				return ret;
>> +
>> +			ret = empty_clusters(fs, p_blkno, n_clusters);
>> +			if (ret) {
>> +				/* XXX: We don't wan't to overwrite the error
>> +				 * from insert_extent().  But we probably need
>> +				 * to BE LOUDLY UPSET. */
>> +				ocfs2_free_clusters(fs, n_clusters, p_blkno);
>> +				return ret;
>> +			}
>>     
>
> Ok, you're zeroing the entire cluster, and doing it before inserting the
> extent - this is great. I'm not so fond of the double write though - later
> on in this function, the users data will be re-written to some of the same
> area. Can you please write the data and only zero the blocks outside of the
> user request, but still preserve the ordering with respect to
> ocfs2_insert_extent?
>   
OK, I will modify it.