Re: [PATCH v4] jbd2: fix shrinker scan budget accounting in jbd2_journal_shrink_scan()

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] fs/jbd2: premature reclaim abort in jbd2_journal_shrink_scan()
--

commit ac994d3eed163d41361f281d91578f4538a2003f
Author: Qiliang Yuan <odys.yuan@gmail.com>

jbd2: fix shrinker scan budget accounting in jbd2_journal_shrink_scan()

This commit modifies jbd2_journal_shrink_scan() to return SHRINK_STOP
when jbd2_journal_shrink_checkpoint_list() frees zero items. The goal
is to prevent the shrinker from continually scanning a stale freeable count
when the checkpoint list only contains busy buffers.

> diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
> index 00f5a98f3d4fe..c921afce7d061 100644
> --- a/fs/jbd2/journal.c
> +++ b/fs/jbd2/journal.c
> @@ -1267,6 +1267,16 @@ static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
>  	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
>  	trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
>  
> +	/*
> +	 * A checkpoint list full of busy buffers can keep reporting a
> +	 * stale, positive freeable count after nothing more can be
> +	 * reclaimed. Stop this reclaim pass once a call frees nothing,
> +	 * instead of burning through do_shrink_slab()'s scan budget
> +	 * retrying against buffers whose writeback won't finish any sooner.
> +	 */
> +	if (nr_shrunk == 0)
> +		return SHRINK_STOP;

[Severity: High]
Could this unintentionally abort the entire reclaim pass when the code hits
a chunk of busy buffers or gets interrupted by the scheduler?

If do_shrink_slab() provides a scan budget and the first batch of buffers
happen to be busy writeback buffers, jbd2_journal_shrink_checkpoint_list()
will return 0. This change then translates that 0 into SHRINK_STOP, which
aborts the rest of the scanning budget for this reclaim pass.

Additionally, jbd2_journal_shrink_checkpoint_list() can return 0 early due
to scheduling or lock contention:

fs/jbd2/checkpoint.c:jbd2_journal_shrink_checkpoint_list()
    ...
		if (need_resched() || spin_needbreak(&journal->j_list_lock))
			break;
    ...

Will returning SHRINK_STOP in these scenarios prematurely skip remaining
transactions that might contain clean, freeable buffers?

> +
>  	return nr_shrunk;
>  }
>

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260930-fix-jbd2-shrink-scan-nr-scanned-v4-1-eb9a3ccc2524@gmail.com?part=1