svn commit: r218603 - head/sbin/tunefs

Bruce Evans brde at
Sun Feb 13 10:51:35 UTC 2011

On Sat, 12 Feb 2011, Konstantin Belousov wrote:

> Log:
>  When creating a directory entry for the journal, always read at least
>  the fragment, and write the full block. Reading less might not work
>  due to device sector size bigger then size of direntries in the
>  last directory fragment.

I think it should always write full fragments too (and the kernel should
always read/write in units of fragments, not sectors of any size).

> Modified: head/sbin/tunefs/tunefs.c
> ==============================================================================
> --- head/sbin/tunefs/tunefs.c	Sat Feb 12 12:52:12 2011	(r218602)
> +++ head/sbin/tunefs/tunefs.c	Sat Feb 12 13:12:45 2011	(r218603)
> @@ -733,16 +740,19 @@ dir_extend(ufs2_daddr_t blk, ufs2_daddr_
> {
> 	char block[MAXBSIZE];
> -	if (bread(&disk, fsbtodb(&sblock, blk), block, size) <= 0) {
> +	if (bread(&disk, fsbtodb(&sblock, blk), block,
> +	    roundup(size, sblock.fs_fsize)) <= 0) {

Rounding up to a fragment boundary is spelled fragroundup(fs, size) in
ffs.  This use fs->fs_qfmask and fs->fs_fmask for optimality.  It is
unclear if the kernel macros work in userland, but here we already
use fsbtodb() which uses fs->fsbtodb for optimality.

[I've just learned again about fragroundup() after trying to fix rounding
in cluster_read().  See blksize().  Since cluster_read() doesn't know
about the fragment size or the fs dependencies in blksize(), it cannot
read ahead correctly across i/o size boundaries (exactly one of which
occurs near EOF for almost all files with fragments in ffs).  This bug is
missing in old breadn() -- it can even handle multiple i/o size boundaries,
since it is passed a separate size for every block.]


More information about the svn-src-all mailing list