cgroup/cpuset: Prevent race between task attach and cpuset state change

Commit e44193d39e8d ("cpuset: let hotplug propagation work wait for
task attaching") was introduced to let hotplug operation to wait
until the completion of task attach operation. However, it is still
possible that the states of the source or destination cpuset can
be changed between the cpuset_can_attach() call and the subsequent
cpuset_attach()/cpuset_cancel_attach() call.

As a result, data gathered during cpuset_can_attach() cannot be reliably
used in the subsequent cpuset_attach()/cpuset_cancel_attach()
call at all. Make the task attach operation more robust
and allow the sharing of data between cpuset_can_attach() and
cpuset_attach()/cpuset_cancel_attach() by making cpuset_write_resmask()
and cpuset_partition_write() wait for the completion of task attach
as well.

Ideally, an ongoing task attach operation should block any cpuset write
operation that can change its internal state until the operation is
completed. However, the attach_in_progress flag is currently per cpuset
and only the destination cpuset will have this flag set. The flag is not
set in the source cpuset where the tasks will be moved from. Even if we
extend the scope to include the source cpuset, it will not block cpuset
operation that changes the state of one of its ancestor cpuset which may
indirectly impact the state of the source or destination cpuset. It may
be too costly to set the flag for the whole subtree, it is far easier
to just make the flag global and block all the cpuset write operation
whenever a task attach operation is in progress.

Make that change by creating a new cpuset attach context (attach_ctx)
structure to hold the global in_progress flag and use it for blocking
cpuset write operation if a cpuset attach operation is in progress. Also
add a new wait_attach_done_lock() helper to do the waiting for an
ongoing attach operation and acquire the cpuset_mutex.

The comments about validate_change() are no longer valid as it won't
be called at all if an attach operation is in progress. So the comments
can be removed.

The per-cpuset attach_in_progress flag is also currently used in
partition_is_populated() and cpuset_is_populated() to determine if
an empty cpuset will have incoming task. This check will no longer be
needed as this function will not be called when there is a task attach
in progress. So the flag check is now removed.

Reviewed-by: Ridong Chen <ridong.chen@linux.dev>
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

0 files changed, 0 insertions, 0 deletions