Re: Observation of a memory leak with commit 314001f0bf92 ("af_unix: Add OOB support")

From: Jakub Kicinski <kuba@kernel.org>
To: Thorsten Leemhuis <regressions@leemhuis.info>
Cc: Lukas Bulwahn <lukas.bulwahn@gmail.com>,
	Rao Shoaib <rao.shoaib@oracle.com>,
	"David S. Miller" <davem@davemloft.net>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Netdev <netdev@vger.kernel.org>,
	Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>,
	regressions@lists.linux.dev
Subject: Re: Observation of a memory leak with commit 314001f0bf92 ("af_unix: Add OOB support")
Date: Mon, 10 Jan 2022 08:21:04 -0800	[thread overview]
Message-ID: <20220110082104.2292d7ad@kicinski-fedora-PC1C0HJN.hsd1.ca.comcast.net> (raw)
In-Reply-To: <a754b7d0-8a20-9730-c439-1660994005d0@leemhuis.info>

On Mon, 10 Jan 2022 15:02:23 +0100 Thorsten Leemhuis wrote:
> On 09.01.22 22:20, Jakub Kicinski wrote:
> > On Fri, 7 Jan 2022 07:48:46 +0100 Lukas Bulwahn wrote:  
> >> Dear Rao and David,
> >>
> >>
> >> In our syzkaller instance running on linux-next,
> >> https://elisa-builder-00.iol.unh.edu/syzkaller-next/, we have been
> >> observing a memory leak in prepare_creds,
> >> https://elisa-builder-00.iol.unh.edu/syzkaller-next/report?id=1dcac8539d69ad9eb94ab2c8c0d99c11a0b516a3,
> >> for quite some time.
> >>
> >> It is reproducible on v5.15-rc1, v5.15, v5.16-rc8 and next-20220104.
> >> So, it is in mainline, was released and has not been fixed in
> >> linux-next yet.
> >>
> >> As syzkaller also provides a reproducer, we bisected this memory leak
> >> to be introduced with  commit 314001f0bf92 ("af_unix: Add OOB
> >> support").
> >>
> >> We also tested that reverting this commit on torvalds' current tree
> >> made the memory leak with the reproducer go away.
> >>
> >> Could you please have a look how your commit introduces this memory
> >> leak? We will gladly support testing your fix in case help is needed.  
> > 
> > Let's test the regression/bug report tracking bot :)
> > 
> > #regzbot introduced: 314001f0bf92  
> 
> Great, thx for trying, you only did a small mistake: it lacked a caret
> (^) before the "introduced", which would have told regzbot that the
> parent mail (the one you quoted) is the one containing the report (which
> later is linked in patch descriptions of fixes and allows rezgbot to
> connect things). That's why regzbot now thinks you reported the issue
> and looks out for patches and commits that link to your mail. :-/
> 
> Don't worry, I just added it properly and now mark this as duplicate:
> 
> #regzbot dup-of:
> https://lore.kernel.org/lkml/CAKXUXMzZkQvHJ35nwVhcJe%2BDrtEXGw%2BeKGVD04=xRJkVUC2sPA@mail.gmail.com/
> 
> Thx again for trying.

Ah, thanks for the fix up, I copy/pasted the example with the hash and
forgot about the caret.

> I wonder if this mistake could be avoided. I came up with one idea while
> walking the dog:
> 
>  * if there is *no* parent mail, then "regzbot introduce" could consider
> the current mail as the report
> 
>  * if there *is* a parent mail, then "regzbot introduce" could consider
> the parent as the report
> 
> Then regzbot would have done the right thing in this case. But there is
> a "but": I wonder if such an approach would be too much black magic that
> confuses more than it helps. What do you think?

IMHO heuristics may do more harm than good. At least for maintainers.