Re: [RFC PATCH] samples/damon/mtier: error out for zero quota goal target values

SJ Park <[email protected]> Sat, 1 Aug 2026 14:30:26 -0700
Newsgroups dev.linux.lists.damon
Message-ID <[email protected]>
On Sat, 01 Aug 2026 21:25:50 +0000 [email protected] wrote:

> Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
> 
> Pre-existing issues:
> - [High] The patch addresses a division by zero in the mtier sample module but leaves the core vulnerability unpatched. In `mm/damon/lru_sort.c`, setting the `active_mem_bp` module parameter to 10002 results in a `target_value` of 0, triggering the same division by zero Oops in the core framework.
> - [High] `damos_new_quota_goal()` allocates `damos_quota_goal` using `kmalloc_obj()`, leaving `last_psi_total` uninitialized. This causes wildcard auto-tuning behavior for PSI-based quota goals on their first cycle.
> --
> 
> --- Patch [1]: [RFC PATCH] samples/damon/mtier: error out for zero quota goal target values ---
> commit 1b5b3916c23bf9b477a9c0308b9b7d82d8503f36
> Author: SJ Park <[email protected]>
> 
> samples/damon/mtier: error out for zero quota goal target values
> 
> This commit modifies the DAMON mtier sample to return an error when started
> with a zero value for node0_mem_used_bp or node0_mem_free_bp. This prevents a
> division by zero Oops in damos_quota_score() when target_value is zero.
> 
> Link: https://lore.kernel.org/[email protected]
> 
> > diff --git a/samples/damon/mtier.c b/samples/damon/mtier.c
> > index ac9c24b92ead8..d1123ebbfab90 100644
> > --- a/samples/damon/mtier.c
> > +++ b/samples/damon/mtier.c
> > @@ -156,6 +156,9 @@ static struct damon_ctx *damon_sample_mtier_build_ctx(bool promote)
> >  	if (!scheme)
> >  		goto free_out;
> >  	damon_set_schemes(ctx, &scheme, 1);
> > +	/* zero target value causes division by zero in damos_quota_store() */
> > +	if (!node0_mem_used_bp || !node0_mem_free_bp)
> > +		goto free_out;
> 
> [Severity: High]
> This isn't a bug introduced by this patch, but does this fix leave the core
> vulnerability unpatched?
> 
> Since the underlying issue is in damos_quota_score() when target_value is 0:
> 
> mm/damon/core.c:damos_quota_score() {
>     ...
>     mult_frac(goal->current_value, 10000, goal->target_value)
>     ...
> }
> 
> It appears other modules can still trigger this Oops.  For example, if a user
> writes 10002 to the active_mem_bp parameter in the damon_lru_sort module, the
> cold scheme quota goal calculation will result in a zero target value:
> 
> mm/damon/lru_sort.c:damon_lru_sort_add_quota_goals() {
>     ...
>     goal = damos_new_quota_goal(DAMOS_QUOTA_INACTIVE_MEM_BP,
>                                 10000 - active_mem_bp + 2);
>     ...
> }
> 
> Should the zero check be handled centrally in damos_new_quota_goal() or
> damos_quota_score() to prevent similar crashes from other modules?

Good catch.  Eventually I will fix this in core.  But let's make
easy-to-backport hotfixes for stable kernels at the moment.  I will add a fix
for lru_sort.

> 
> >  	quota_goal = damos_new_quota_goal(
> >  			promote ? DAMOS_QUOTA_NODE_MEM_USED_BP :
> >  			DAMOS_QUOTA_NODE_MEM_FREE_BP,
> >  			promote ? node0_mem_used_bp : node0_mem_free_bp);
> 
> [Severity: High]
> This is a pre-existing issue, but does damos_new_quota_goal() leave fields
> uninitialized?
> 
> Looking at damos_new_quota_goal(), it allocates the struct using kmalloc_obj()
> without zeroing the memory:
> 
> mm/damon/core.c:damos_new_quota_goal() {
>     ...
>     goal = kmalloc_obj(*goal);
>     if (!goal)
>         return NULL;
>     goal->metric = metric;
>     goal->target_value = target_value;
>     INIT_LIST_HEAD(&goal->list);
>     return goal;
> }
> 
> This leaves union fields like last_psi_total uninitialized.  When a PSI-based
> quota goal is evaluated for the first time, this uninitialized value is used
> in the feedback loop calculation:
> 
> mm/damon/core.c:damos_set_quota_goal_current_value() {
>     ...
>     goal->current_value = now_psi_total - goal->last_psi_total;
>     ...
> }
> 
> Could this uninitialized memory usage corrupt the auto-tuning feedback loop?

A fix [1] of the bug is under review.

[1] https://lore.kernel.org/[email protected]

> 
> >  	if (!quota_goal)
> >  		goto free_out;
> 
> -- 
> Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=1


Thanks,
SJ