Re: [PATCH] befs: return BEFS_ERR if validation of ag_shift fails

[Luis de Bethencourt]

On 12/08/16 11:12, Salah Triki wrote:
> ag_shift is used by blockno2iaddr() to get allocation group number
> from block. If ag_shift is inconsistent with block_per_ag, an out of
> bounds allocation group may occur [1]. So add return BEFS_ERR and update
> comment and error message to reflect this change.
> 
> [1] https://lkml.org/lkml/2016/8/12/42
> 
> Signed-off-by: Salah Triki <salah.triki@xxxxxxxxx>
> ---
>  fs/befs/super.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/befs/super.c b/fs/befs/super.c
> index 7c50025..2e3a3fd 100644
> --- a/fs/befs/super.c
> +++ b/fs/befs/super.c
> @@ -101,10 +101,13 @@ befs_check_sb(struct super_block *sb)
>  
>  
>  	/* ag_shift also encodes the same information as blocks_per_ag in a
> -	 * different way, non-fatal consistency check
> +	 * different way as a consistency check.
>  	 */
> -	if ((1 << befs_sb->ag_shift) != befs_sb->blocks_per_ag)
> -		befs_error(sb, "ag_shift disagrees with blocks_per_ag.");
> +	if ((1 << befs_sb->ag_shift) != befs_sb->blocks_per_ag) {
> +		befs_error(sb, "ag_shift disagrees with blocks_per_ag. "
> +			   "Corruption likely.");
> +		return BEFS_ERR;
> +	}
>  
>  	if (befs_sb->log_start != befs_sb->log_end ||
>  	    befs_sb->flags == BEFS_DIRTY) {
> 

Hi Salah,

I initially also added the BEFS_ERR return for this check. I understand
it makes sense.

But as I mentioned in the patch adding this warning [0], a correct
blocks_per_ag isn't mandatory. I noticed this because BeFS images
created by Haiku OS just set blocks_per_ag to 1. Which is clearly not
what it should be.

The file systems created by Haiku OS can be read without issues since
sb->blocks_per_ag isn't actually used in lookups/reading file datastreams.

This is what I get in dmesg when loading a Haiku OS image with
CONFIG_BEFS_DEBUG on:
...
[  196.376651] befs: (loop1):   blocks_per_ag 1
[  196.376652] befs: (loop1):   ag_shift 14
...

With ag_shift of 14, blocks_per_ag should be 16,384.

If we return BEFS_ERR, the system will refuse to mount and users won't be
able to read these Haiku OS images they could read before. Unfortunate that
Haiku OS is not using blocks_per_ag properly, but users and avoiding
regressions go first.

Which BeFS images are you using to test the befs code? Since all the ones
I have are generated from Haiku OS, all of them have a bad blocks_per_ag.
LOL

Maybe we should contact Haiku OS developers and ask them to write a correct
blocks_per_ag. Though we should also support images created before they fix
this.

Sorry you spent time on this when the issue is out of our control :(
Nacked-by: Luis de Bethencourt <luisbg@xxxxxxxxxxxxxxx>

Thanks,
Luis

[0] https://lkml.org/lkml/2016/8/9/816