Re: [PATCH 10/19] mm/damon/core: validate params for probe hits weighted sum overflow

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Pre-existing integer overflow in `damon_merge_two_regions()` during weighted average calculations.
- [High] Pre-existing integer overflows in `damon_sample_bp_to_count()` and `damon_sample_count_to_bp()`.
--

--- Patch [10]: [PATCH 10/19] mm/damon/core: validate params for probe hits weighted sum overflow ---
commit 74455514643a15cd1b70c8a9afa3237ba4226d34
Author: SJ Park <sj@kernel.org>
Date:   Fri Jul 10 13:46:39 2026 +0000

    mm/damon/core: validate params for probe hits weighted sum overflow

This commit adds parameter validation to prevent integer overflows in probe
hits weighted sum calculations. It ensures that user-configured probe weights
do not cause the weighted sum of probe hits to exceed the maximum possible
value during monitoring.

> diff --git a/mm/damon/core.c b/mm/damon/core.c
> index 4f1425e56950b..a0e6a5129fbf3 100644
> --- a/mm/damon/core.c
> +++ b/mm/damon/core.c
> @@ -1335,6 +1335,9 @@ static void damos_set_filters_default_reject(struct damos *s)
>  static bool damon_valid_probe_params(struct damon_ctx *ctx)
>  {
>  	unsigned long sample_interval;
> +	unsigned char max_probe_hits;
> +	struct damon_probe *probe;
> +	unsigned int wsum, wsum_to_add;
>  
>  	if (!damon_has_probe_weights(ctx))
>  		return true;
> @@ -1342,6 +1345,18 @@ static bool damon_valid_probe_params(struct damon_ctx *ctx)
>  	sample_interval = ctx->attrs.sample_interval ? : 1;
>  	if (ctx->attrs.aggr_interval / sample_interval > U8_MAX)
>  		return false;
> +
> +	/* invalid if probe hits weighted sum can overflow */
> +	max_probe_hits = damon_nr_samples_per_aggr(&ctx->attrs);
> +	wsum = 0;
> +	damon_for_each_probe(probe, ctx) {
> +		if (probe->weight > UINT_MAX / max_probe_hits)
> +			return false;
> +		wsum_to_add = probe->weight * max_probe_hits;
> +		if (UINT_MAX - wsum < wsum_to_add)
> +			return false;
> +		wsum += wsum_to_add;
> +	}
>  	return true;
>  }
>  

[Severity: High]
This is a pre-existing issue, but does damon_merge_two_regions() suffer from an
unsigned long overflow during average calculations?

mm/damon/core.c:damon_merge_two_regions() {
	...
	l->nr_accesses = (l->nr_accesses * sz_l + r->nr_accesses * sz_r) /
			(sz_l + sz_r);
	l->age = (l->age * sz_l + r->age * sz_r) / (sz_l + sz_r);
	...
}

During the calculation of the merged region metrics, if a region size reaches
1GB and nr_accesses is just 4, their product is 4,294,967,296. On 32-bit
architectures where unsigned long is 32 bits, this will instantly overflow.

Additionally, on 64-bit architectures, the age variable could grow indefinitely
and overflow if a region becomes very large over a long uptime. Can this lead
to corrupted statistics and misfiring of memory management actions?


[Severity: High]
This is a pre-existing issue, but I noticed a potential integer overflow in
damon_sample_bp_to_count() and damon_sample_count_to_bp():

mm/damon/core.c:damon_sample_bp_to_count() {
	return bp * damon_nr_samples_per_aggr(attrs) / 10000;
}

mm/damon/core.c:damon_sample_count_to_bp() {
	return mult_frac(count, 10000, damon_nr_samples_per_aggr(attrs));
}

Can these calculations overflow? The functions perform 32-bit multiplication of
bp (or the remainder in mult_frac) and the max samples using unsigned int
arithmetic.

If max_samples (aggr_interval / sample_interval) exceeds 429,496, which is
possible with a high aggregation interval compared to the sampling interval
(e.g., 5s aggregation and 10us sample intervals), the result could exceed
UINT_MAX (4,294,967,295) and silently overflow. Could this corrupt monitoring
region access counts during dynamic DAMON attribute updates?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260710134651.18084-1-sj@kernel.org?part=10