[Ocfs2-devel] [PATCH] Fix the nested PR lock calling issue

[Jiaju Zhang]

Hi Wengang,

On Wed, Jul 21, 2010 at 5:21 PM, Wengang Wang <wen.gang.wang at oracle.com> wrote:
> Hi Jiaju,
>
> On 10-07-21 16:28, Jiaju Zhang wrote:
>> Hi,
>>
>> This is an attempt to fix the bug
>> https://bugzilla.novell.com/show_bug.cgi?id=614332
>> (also http://oss.oracle.com/bugzilla/show_bug.cgi?id=1278)
>>
>> The full log is available at comment #54 of bug 614332 at Novell
>> bugzilla. Simply speaking, it should be nested PR lock calling issue.
>> The calling sequence should be:
>>
>> node1       node2
>>
>>             gr PR
>>               |
>>               V
>> PR(EX) -> BAST: OCFS2_LOCK_BLOCKED
>>               |
>>               V
>>             rq PR
>>               |
>>               V
>>             wait = 1
>>
>> Since we have already had the granted PR lock from the beginning, we
>> may not need to request the PR lock again.
>
> What's the real calling path?
> I guess you meant
> ocfs2_permission()(lock gothere)
>  ->ocfs2_check_acl()
>  ->ocfs2_get_acl()(require the lock again)
>
> But there is other ocfs2_get_acl() calling paths in which the lock is
> not taken before calling ocfs2_get_acl(). such as
> ocfs2_setattr()(the lock is taken and released before calling ocfs2_acl_chmod())
>  ->ocfs2_acl_chmod()
>  ->ocfs2_get_acl().
>
> So I think the patch is not safe.

Many thanks for pointing this out;)
I'll send an improved patch soon.

Thanks,
Jiaju