Re: [PATCH] cgroup/cpuset: Fix partition recovery with unavailable CPUs

[Waiman Long <longman@redhat.com>]

On 10/20/26 9:33 AM, Guopeng Zhang wrote:
> From: Guopeng Zhang <zhangguopeng@kylinos.cn>
>
> An invalid local partition can become valid again even when its
> exclusive CPUs are no longer available from its parent.
>
> This can be reproduced on a cgroup v2 system with CPUs 0-7 online:
>
>      cd /sys/fs/cgroup
>      echo +cpuset > cgroup.subtree_control
>      mkdir A
>      echo 1-4 > A/cpuset.cpus
>      echo 1-4 > A/cpuset.cpus.exclusive
>      echo root > A/cpuset.cpus.partition
>      echo +cpuset > A/cgroup.subtree_control
>      mkdir A/B
>      echo 3-4 > A/B/cpuset.cpus
>      echo 3-4 > A/B/cpuset.cpus.exclusive
>      echo root > A/B/cpuset.cpus.partition
>      echo 1-2 > A/cpuset.cpus.exclusive
>      cat A/B/cpuset.cpus.partition
>      root
>      cat A/B/cpuset.cpus.exclusive.effective
>      cat A/cpuset.cpus.effective
>      1-2
>      cat A/B/cpuset.cpus.effective
>      1-2
>
> B is changed back to a valid root even though its exclusive CPUs are
> no longer available and cpuset.cpus.exclusive.effective is empty. B
> should remain invalid until A can provide its exclusive CPUs again.
>
> Shrinking A's exclusive CPUs invalidates B in
> compute_partition_effective_cpumask(). The subsequent walk in
> update_cpumasks_hier() tries to recover B, but
> update_parent_effective_cpumask() skips the recovery checks when xcpus
> is not a subset of parent->effective_xcpus. Since part_error is still
> zero, B is changed back to a valid partition.
>
> Set PERR_INVCPUS when the subset check fails so that the partition stays
> invalid. New test cases are added to test_cpuset_prs.sh for removing and
> restoring the parent's exclusive CPUs.
>
> Fixes: 0c7f293efc87 ("cgroup/cpuset: Add cpuset.cpus.exclusive.effective for v2")
> Signed-off-by: Guopeng Zhang <zhangguopeng@kylinos.cn>
> ---
>   kernel/cgroup/cpuset.c                            | 8 ++++++--
>   tools/testing/selftests/cgroup/test_cpuset_prs.sh | 4 ++++
>   2 files changed, 10 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> index 585a982249ad..be80763953ed 100644
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -1986,12 +1986,16 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd,
>   				cpumask_copy(tmp->addmask, cs->effective_xcpus);
>   				adding = true;
>   			}
> -		} else if (is_partition_invalid(cs) && !cpumask_empty(xcpus) &&
> -			   cpumask_subset(xcpus, parent->effective_xcpus)) {
> +		} else if (is_partition_invalid(cs) && !cpumask_empty(xcpus)) {
>   			struct cgroup_subsys_state *css;
>   			struct cpuset *child;
>   			bool exclusive = true;
>   
> +			if (!cpumask_subset(xcpus, parent->effective_xcpus)) {
> +				part_error = PERR_INVCPUS;
> +				goto write_error;
> +			}
> +
>   			/*
>   			 * Convert invalid partition to valid has to pass the
>   			 * cpu exclusivity test as well as the housekeeping
> diff --git a/tools/testing/selftests/cgroup/test_cpuset_prs.sh b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
> index d04aad697b96..f1d517647ade 100755
> --- a/tools/testing/selftests/cgroup/test_cpuset_prs.sh
> +++ b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
> @@ -427,6 +427,10 @@ TEST_MATRIX=(
>   	# cpuset.cpus is allowed, but those extra CPUs are ignored.
>   	"  C2-3:P1   C3:P1  .      .      .     C2-4    .      .     0 A1:|A2:2-3 A1:P1|A2:P1"
>   
> +	# A child cannot recover until its exclusive CPUs are available again.
> +	" CX1-4:P1 CX3-4:P1 .     .     X1-2    .      .      .     0 A1:1-2|A2:1-2|XA2: A1:P1|A2:P-1"
> +	" CX1-4:P1 CX3-4:P1 .     .  X1-2:X1-4 .      .      .     0 A1:1-2|A2:3-4|XA2:3-4 A1:P1|A2:P1"
> +
>   	# Taking away all CPUs from parent or itself if there are tasks
>   	# will make the partition invalid.
>   	"  C2-3:P1   C3:P1  .      .      T     C2-3    .      .     0 A1:2-3|A2:2-3 A1:P1|A2:P-1"

Patch 2 of my series should fix this problem too. I do borrow your test 
cases in my patch. I hope you don't mind about it.

Cheers,
Longman