Re: [PATCH bpf-next v3 1/2] mm/bpf: Add bpf_proactive_reclaim kfuncs

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Calling `try_to_free_mem_cgroup_pages` with `GFP_KERNEL` from arbitrary sleepable BPF programs can cause lock inversion deadlocks.
--

--- Patch [1]: [PATCH bpf-next v3 1/2] mm/bpf: Add bpf_proactive_reclaim kfuncs ---
commit 55d679bb7cd0ded2c29ec1fcf65b2396144cddad
Author: Hui Zhu <zhuhui@kylinos.cn>

mm/bpf: Add bpf_proactive_reclaim kfuncs

This commit introduces bpf_proactive_reclaim and
bpf_proactive_reclaim_swappiness kfuncs to expose memcg proactive reclaim
to sleepable BPF programs. They allow BPF programs to trigger one reclaim
pass on a given memcg with specific swappiness and size limits.

> diff --git a/mm/bpf_memcontrol.c b/mm/bpf_memcontrol.c
> index 716df49d76477..b0a0d4c55dc4e 100644
> --- a/mm/bpf_memcontrol.c
> +++ b/mm/bpf_memcontrol.c

[ ... ]

> +static unsigned long
> +bpf_proactive_reclaim_pages(struct mem_cgroup *memcg, unsigned long nr_pages,
> +			    int *swappiness)
> +{
> +	if (!nr_pages || unlikely(bpf_in_reclaim_context()))
> +		return 0;
> +
> +	return try_to_free_mem_cgroup_pages(memcg, nr_pages, GFP_KERNEL,
> +					    MEMCG_RECLAIM_MAY_SWAP |
> +					    MEMCG_RECLAIM_PROACTIVE,
> +					    swappiness);

[Severity: High]
Can calling try_to_free_mem_cgroup_pages() with GFP_KERNEL from arbitrary
sleepable BPF programs cause lock inversion deadlocks?

If a sleepable BPF program attaches to an LSM hook or fentry point where a
filesystem lock is already held (for example, bpf_lsm_inode_mkdir where
inode_lock is held), and the program calls bpf_proactive_reclaim(),
GFP_KERNEL will allow __GFP_FS reclaim.

The memory reclaim path could then enter filesystem shrinkers which attempt
to acquire the same lock, leading to a self-deadlock. 

Should this use a more restricted GFP mask, or perhaps dynamically inherit
the context's GFP constraints to avoid bypassing lock-context protections?

> +}

[ ... ]

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/cover.1787120833.git.zhuhui@kylinos.cn?part=1