Re: [PATCH v2] dm dust: make badblock messages target-relative

[Benjamin Marzinski <bmarzins@redhat.com>]

On Mon, Aug 03, 2026 at 02:09:13PM +0000, Samuel Moelius wrote:
> dm-dust currently treats addbadblock, removebadblock and queryblock
> arguments as block numbers on the underlying device. That is surprising
> for a device-mapper target: a dm-dust table with a non-zero backing
> offset can add bad blocks that are outside the mapped target, and a
> badblock added for logical block 0 is missed because the I/O path checks
> the remapped backing-device block instead.
> 
> Interpret badblock message arguments as blocks relative to the start of
> the dm-dust target instead. Bound the arguments by the target length and
> perform badblock lookup using target-relative sectors before remapping
> the bio to the underlying device.
> 
> This intentionally changes the non-zero backing-offset behavior to make
> the badblock control interface match the mapped dm-dust device, rather
> than the underlying device.
> 
> Assisted-by: Codex:gpt-5.5-cyber-preview
> Signed-off-by: Samuel Moelius <sam.moelius@trailofbits.com>

Reviewed-by: Benjamin Marzinski <bmarzins@redhat.com>

> ---
> Changes in v2:
>   - Revise commit message
>   - Remove call to sector_div() in __dust_map_write()
> 
>  drivers/md/dm-dust.c | 16 +++++++---------
>  1 file changed, 7 insertions(+), 9 deletions(-)