Ben Hutchings <ben@xxxxxxxxxxxxxxx> writes:
> On Wed, 2013-07-03 at 12:20 +0100, Luis Henriques wrote:
>> Hi,
>>
>> While working on backporting commits to the 3.5 kernel, I came across
>> this commit which is tagged for stable kernels:
>>
>> 084457f284abf6789d90509ee11dae383842b23b cgroup: fix umount vs cgroup_cfts_commit() race
>>
>> This commit, however, seems to break the kernel build with the
>> following change:
>>
>> - if (cfts && ss->root != &rootnode) {
>> + if (cfts && ss->root != &rootnode &&
>> + atomic_inc_not_zero(sb->s_active)) {
>> ^^^^
>> It should be atomic_inc_not_zero(sb->s_active) instead, I believe.
>
> Failing to see a difference here...
>
>> This is fixed in a later commit:
>>
>> e8c82d20a9f729cf4b9f73043f7fd4e0872bebfd cgroup: convert cgroup_cft_commit() to use cgroup_for_each_descendant_pre()
>
> This adds an & before sb->s_active, which I guess is what you meant.
Yes, that's exactly what I meant and failed to show in my email :-)
Anyway, I'll probably just queue the patch for 3.5 modifying the patch
to include the fix from e8c82d20a9f729cf4b9f73043f7fd4e0872bebfd.
Cheers,
--
Luis
>
> Ben.
>
>> Which is not tagged for stable inclusion.
>>
>> Could you please advice? Should stable kernels just drop this patch?
>>
>> Cheers,
Attachment:
signature.asc
Description: PGP signature