Re: (sashiko review) [PATCH v2] mm/damon/core: validate goal nid before accessing node data

[SeongJae Park <sj@kernel.org>]

Forwarding Sashiko review for doing discussions via mails.

# review url: https://sashiko.dev/#/patchset/20260325155221.202700-1-objecting@objecting.org
# start of sashiko.dev inline review
commit 24e25058b8f508495d0b3db33bed16564df7fed4
Author: Josh Law <objecting@objecting.org>
Subject: mm/damon/core: validate goal nid before accessing node data

This patch adds bounds and node_state(N_MEMORY) checks before using
goal->nid from userspace to prevent out-of-bounds access in NODE_DATA().

Cc: stable@vger.kernel.org

> diff --git a/mm/damon/core.c b/mm/damon/core.c
> index db6c67e52d2b..ed45a7443abe 100644
> --- a/mm/damon/core.c
> +++ b/mm/damon/core.c
> @@ -2245,6 +2245,10 @@ static __kernel_ulong_t damos_get_node_mem_bp(
>  	struct sysinfo i;
>  	__kernel_ulong_t numerator;
>  
> +	if (goal->nid < 0 || goal->nid >= MAX_NUMNODES ||
> +	    !node_state(goal->nid, N_MEMORY))
> +		return 0;
> +

When goal->nid is invalid or offline, this unconditionally returns 0.
For the DAMOS_QUOTA_NODE_MEM_FREE_BP metric, wouldn't returning 0 imply
0% free memory?

Could this cause the DAMON feedback loop to interpret the goal as
severely under-achieved and drastically increase the scheme's
aggressiveness?

Should this fallback logic check the metric type and return 10000
(100% free) for FREE_BP metrics to fail gracefully, mirroring the
existing fallback logic used when a memory cgroup lookup fails?

>  	si_meminfo_node(&i, goal->nid);
>  	if (goal->metric == DAMOS_QUOTA_NODE_MEM_USED_BP)
>  		numerator = i.totalram - i.freeram;
>  	else	/* DAMOS_QUOTA_NODE_MEM_FREE_BP */
>  		numerator = i.freeram;
>  	return mult_frac(numerator, 10000, i.totalram);

While not introduced by this patch, since this function is being
modified to handle node edge cases, is it possible for a node to have
N_MEMORY set, but have 0 managed pages? For example, if all of its
memory is reserved for hugetlb or unmanaged ZONE_DEVICE memory.

If so, si_meminfo_node() would set i.totalram to 0. Would the call
to mult_frac() then cause a divide-by-zero panic when it evaluates
x_ / d_ ?

> @@ -2261,6 +2265,10 @@ static unsigned long damos_get_node_memcg_used_bp(
>  	unsigned long used_pages, numerator;
>  	struct sysinfo i;
>  
> +	if (goal->nid < 0 || goal->nid >= MAX_NUMNODES ||
> +	    !node_state(goal->nid, N_MEMORY))
> +		return 0;
> +

Similar to the function above, returning 0 unconditionally here for
DAMOS_QUOTA_NODE_MEMCG_FREE_BP might artificially trigger maximum
reclaim aggressiveness.


# end of sashiko.dev inline review
# review url: https://sashiko.dev/#/patchset/20260325155221.202700-1-objecting@objecting.org
#
# hkml [1] generated a draft of this mail.  It can be regenerated
# using below command:
#
#     hkml patch sashiko_dev --for_forwarding \
#             20260325155221.202700-1-objecting@objecting.org
#
# [1] https://github.com/sjp38/hackermail

Sent using hkml (https://github.com/sjp38/hackermail)