Re: [RFC PATCH] mm/damon/core: validate ranges in damon_set_regions()

[sashiko-bot@kernel.org]

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

Pre-existing issues:
- [High] The validation check to prevent empty regions is bypassed due to integer overflow in ALIGN(), leaving the divide-by-zero vulnerability unpatched.
- [High] Memory allocation failure during region updates leaves the running DAMON context with a corrupted, partially updated region list.
--

commit e67b2388e5d46f77944a06f1bd1d3d550638fade
Author: SeongJae Park <sj@kernel.org>

mm/damon/core: validate ranges in damon_set_regions()

This patch updates damon_set_regions() to validate ranges and reject
inputs where the start address is greater than or equal to the end
address. This aims to prevent zero-length regions which can cause
divide-by-zero errors in damon_merge_two_regions().

> diff --git a/mm/damon/core.c b/mm/damon/core.c
> index 7e4b9affc5b06..b3100d7fa5596 100644
> --- a/mm/damon/core.c
> +++ b/mm/damon/core.c
> @@ -358,6 +358,11 @@ int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges,
>  	unsigned int i;
>  	int err;
>  
> +	for (i = 0; i < nr_ranges; i++) {
> +		if (ranges[i].start >= ranges[i].end)
> +			return -EINVAL;
> +	}
> +

[Severity: High]
This isn't a bug introduced by this patch, but does this validation check 
sufficiently protect against overflow when the regions are aligned later?

If a user supplies a region where the end address is ULONG_MAX,
ALIGN(ULONG_MAX, min_region_sz) will wrap around to 0.

Since this validation check uses unaligned inputs, ULONG_MAX passes
the check here. Later, during allocation:

damon_set_regions() {
    ...
    r = damon_new_region(
            ALIGN_DOWN(ranges[i].start, min_region_sz),
            ALIGN(ranges[i].end, min_region_sz));
    ...
}

This creates a region with a wrapped and inverted size. If the adjacent
regions sum to a size of 0, for example 4096 + (ULONG_MAX - 4095),
can this lead to a divide-by-zero crash in damon_merge_two_regions()
when the sizes are added together?

damon_merge_two_regions() {
    ...
    l->nr_accesses = (l->nr_accesses * sz_l + r->nr_accesses * sz_r) /
            (sz_l + sz_r);
    ...
}

[Severity: High]
This is also a pre-existing issue, but what happens to the region list if
memory allocation fails after existing regions are destroyed?

In damon_set_regions(), existing regions not in the new ranges are
destroyed. If a subsequent damon_new_region() fails and returns -ENOMEM:

damon_set_regions() {
    ...
    if (!r)
        return -ENOMEM;
    damon_add_region(r, t);
    ...
}

The error rollback path in damon_revert_target_commits() does not restore
the regions_list state:

damon_revert_target_commits() {
    ...
    damon_for_each_target(target, dst) {
        if (target == failed)
            return;
        put_pid(target->pid);
    }
}

Will this transient allocation failure permanently leave the running damon
context with a corrupted or partially updated region list, silencing
reclamation daemons?

>  	/* Remove regions which are not in the new ranges */
>  	damon_for_each_region_safe(r, next, t) {
[ ... ]

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260627170057.1867-1-sj@kernel.org?part=1