netdev - Re: [PATCH 1/5] io_uring: Fix leak of async data when connect prep fails

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <31ae884f-9db9-47ff-b32f-f870386ba9f8@kernel.dk>
Date: Fri, 31 May 2024 17:07:15 -0600
From: Jens Axboe <axboe@...nel.dk>
To: Gabriel Krisman Bertazi <krisman@...e.de>
Cc: io-uring@...r.kernel.org, netdev@...r.kernel.org
Subject: Re: [PATCH 1/5] io_uring: Fix leak of async data when connect prep
 fails

On 5/31/24 5:01 PM, Gabriel Krisman Bertazi wrote:
> Jens Axboe <axboe@...nel.dk> writes:
> 
>> On 5/31/24 3:12 PM, Gabriel Krisman Bertazi wrote:
>>> move_addr_to_kernel can fail, like if the user provides a bad sockaddr
>>> pointer. In this case where the failure happens on ->prep() we don't
>>> have a chance to clean the request later, so handle it here.
>>
>> Hmm, that should still get freed in the cleanup path? It'll eventually
>> go on the compl_reqs list, and it has REQ_F_ASYNC_DATA set. Yes it'll
>> be slower than the recycling it, but that should not matter as it's
>> an erred request.
> 
> Hm right.  I actually managed to reproduce some kind of memory
> exhaustion yesterday that I thought was fixed by this patch.  But I see
> your point and I'm failing to trigger it today.
> 
> Please disregard this patch. I'll look further to figure out what I did
> there.

Maybe enable KMEMLEAK? It's pretty handy for testing. If there is a leak
there, you should be able to reliably get info by doing:

# ./reproducer (should be easy, just bogus addr)
# echo scan > /sys/kernel/debug/kmemleak
# sleep 5
# echo scan > /sys/kernel/debug/kmemleak

-- 
Jens Axboe