Hi Paulo, Steve,
On 08/24, Steve French wrote:
It does look like the first of these may fix a real leak
setup_ntlmv2_rsp when called from CIFS_SessSetup doesn't seem to free
ses->domainName
sesInfoFree does free it but it is not clear whether in all paths
sesInfoFree is called, but of course if it isn't called we have a much
worse leak.
Can you doublecheck about the memleak details.
On Tue, Aug 23, 2022 at 12:45 PM Paulo Alcantara <pc@xxxxxx> wrote:
Enzo Matsumiya <ematsumiya@xxxxxxx> writes:
>
> Fix memory leaks from some ses fields when cifs_negotiate_protocol() or
> cifs_setup_session() fails in cifs_get_smb_ses().
>
> A leak from ses->domainName has also been identified in setup_ntlmv2_rsp()
> when session setup fails.
Those fields are already freed by sesInfoFree().
> These has been reported by kmemleak.
Could you please include the report in commit message?
Indeed, I was hitting some uncommon path.
This was on a particular branch of mine (i.e. not for-next), and I think
I might have fixed it now(*).
This one can be dropped, I think.
(*) - my branch didn't directly touch that get ses -> negotiate -> setup
path, but only failing on tree connect. So, yes, there might be a bigger
leak involved and I'll try to reproduce it again, and be sure to save
the kmemleak report (which I didn't originally).
Cheers,
Enzo
