Hi.
On Mon, Nov 07, 2022 at 12:50:04PM +0600, Sergey Dolgov <palza00@xxxxxxxxx> wrote:
> we faced problem in removing cgroup in cgroup_addrm_files in kernel 5.4
Do you have a reproducer for this issue? Ideally, with a current kernel?
> crash> struct cgroup_subsys.cfts 0xffff88fc910d9f80
> cfts = {
> next = 0x703,
> prev = 0x0
> },
The cgroup_subsys pointed in css->ss doesn't look like a valid
cgroup_subsys, cfts.next looks like a rubbish.
> from (!css->ss) to (!(css->ss or css == css->cgroup->self)) it will
> be resolve problem
This doesn't seem correct to me. The condition in this place should
capture whether css represents a true subsys state or just a cgroup.
You made it effectively always false, regardless of css->ss is valid or
not. (Maybe there's a typo in your posting and the "fix" was making the
condition always true but that still doesn't seem correct to me, it's
just masking the garbled css->ss.)
> loaded Tainted: G OE 5.4.0-73-generic #82~18.04.1-Ubuntu
`(O)` externally-built ("out-of-tree") module was loaded
My bets are off.
HTH,
Michal
Attachment:
signature.asc
Description: Digital signature
