lkml.org 
[lkml]   [2005]   [Jul]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC] [PATCH 2/4]delayed allocation for ext3
Mingming Cao <cmm@us.ibm.com> wrote:
>
> Here is the updated patch from Badari for delayed allocation for ext3.
> Delayed allocation defers block allocation from prepare-write time to
> page writeout time.

For data=writeback only, yes?

> ...
> --- linux-2.6.12/fs/ext3/inode.c~ext3-delalloc 2005-07-14 23:15:34.866752480 -0700
> +++ linux-2.6.12-ming/fs/ext3/inode.c 2005-07-14 23:15:34.889748984 -0700
> @@ -1340,6 +1340,9 @@ static int ext3_prepare_write(struct fil
> handle_t *handle;
> int retries = 0;
>
> +
> + if (test_opt(inode->i_sb, DELAYED_ALLOC))
> + return __nobh_prepare_write(page, from, to, ext3_get_block, 0);

Rather than performing this test on each ->prepare_write(), would it not be
better to set up a new set of address_space_operations for this mode?

__nobh_prepare_write() seems like a poor choice of name?

> retry:
> handle = ext3_journal_start(inode, needed_blocks);
> if (IS_ERR(handle)) {
> @@ -1439,6 +1442,9 @@ static int ext3_writeback_commit_write(s
> else
> ret = generic_commit_write(file, page, from, to);
>
> + if (test_opt(inode->i_sb, DELAYED_ALLOC))
> + return ret;
> +

Here too, perhaps.

> + }
> + }
> /*
> * The journal_load will have done any necessary log recovery,
> * so we can safely mount the rest of the filesystem now.
> diff -puN fs/buffer.c~ext3-delalloc fs/buffer.c
> --- linux-2.6.12/fs/buffer.c~ext3-delalloc 2005-07-14 23:15:34.875751112 -0700
> +++ linux-2.6.12-ming/fs/buffer.c 2005-07-14 23:15:34.903746856 -0700
> @@ -2337,8 +2337,8 @@ static void end_buffer_read_nobh(struct
> * On entry, the page is fully not uptodate.
> * On exit the page is fully uptodate in the areas outside (from,to)
> */
> -int nobh_prepare_write(struct page *page, unsigned from, unsigned to,
> - get_block_t *get_block)
> +int __nobh_prepare_write(struct page *page, unsigned from, unsigned to,
> + get_block_t *get_block, int create)

Suggest you make this static and update the comment.

> {
> struct inode *inode = page->mapping->host;
> const unsigned blkbits = inode->i_blkbits;
> @@ -2370,10 +2370,8 @@ int nobh_prepare_write(struct page *page
> block_start < PAGE_CACHE_SIZE;
> block_in_page++, block_start += blocksize) {
> unsigned block_end = block_start + blocksize;
> - int create;
>
> map_bh.b_state = 0;
> - create = 1;
> if (block_start >= to)
> create = 0;
> ret = get_block(inode, block_in_file + block_in_page,

What's going on here? Seems that we'll call get_block() with `create=0'.
Is there any point in doing that? For delayed allocation we shuld be able
to skip get_block() altogether here and, err, delay it.

> +int nobh_prepare_write(struct page *page, unsigned from, unsigned
> + get_block_t *get_block)
> +{
> + return __nobh_prepare_write(page, from, to, get_block, 1);
> +}
> +
> EXPORT_SYMBOL(nobh_prepare_write);

Here you add nobh_dalloc_prepare_write() and remember to export it to
modules this time ;)
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-07-27 00:56    [W:0.061 / U:0.268 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site