On Mon, Jun 02, 2008 at 08:42:24AM -0500, Eric Sandeen wrote:
> Theodore Tso wrote:
> > On Thu, May 15, 2008 at 12:17:11AM +0530, Aneesh Kumar K.V wrote:
> >> @@ -3134,8 +3135,7 @@ static void ext4_mb_use_inode_pa(struct ext4_allocation_context *ac,
> >> static void ext4_mb_use_group_pa(struct ext4_allocation_context *ac,
> >> struct ext4_prealloc_space *pa)
> >> {
> >> - unsigned len = ac->ac_o_ex.fe_len;
> >> -
> >> + unsigned int len = ac->ac_o_ex.fe_len;
> >> ext4_get_group_no_and_offset(ac->ac_sb, pa->pa_pstart,
> >> &ac->ac_b_ex.fe_group,
> >> &ac->ac_b_ex.fe_start);
> >> --
> >
> > This change had nothing to do with fixing the use of unitialized data,
> > but when I started looking more closely, it raised a potential signed
> > vs. unsigned issue: ac_o_ex is a struct ext4_free_extent, and fe_len
> > is an int.
> >
> > So here we are assigning an int to an unsigned int. Later, len is
> > assigned to ac_b_ex.len, which means assigning an unsigned int to an
> > int. In other places, fe_len (an int) is compared against pa_free
> > (which is an unsigned short), and fe_len gets assined to pa_free, once
> > again mixing signed and unsigned.
> >
> > Can someone who is really familiar with this code check this out? I
> > think the following pseudo-patch to mballoc.h might be in order:
> >
> > struct ext4_free_extent {
> > ext4_lblk_t fe_logical;
> > ext4_grpblk_t fe_start;
> > ext4_group_t fe_group;
> > - int fe_len;
> > + unsigned int fe_len;
> > };
>
> Hm, ok, so what's going on here:
>
> ext4_mb_normalize_group_request()
> {
> ...
> if (EXT4_SB(sb)->s_stripe)
> ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_stripe;
> else
> ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_mb_group_prealloc;
> ...
> }
>
> and that's a long:
>
> unsigned long s_mb_group_prealloc;
>
> Oh, but that's only ever assigned as
>
> sbi->s_mb_group_prealloc = MB_DEFAULT_GROUP_PREALLOC;
>
> which is
>
> /*
> * default group prealloc size 512 blocks
> */
> #define MB_DEFAULT_GROUP_PREALLOC 512
>
>
> so it's fine... but why are we carrying around a field in the sbi to
> hold a constant that cannot be changed runtime?
We can tune that via MB_PROC_FOPS(group_prealloc);
-aneesh
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
