Re: selftests: net/af_unix test_unix_oob [FAILED]

[Mirsad Todorovac]

On 8/8/23 10:53, Mirsad Todorovac wrote:
> On 8/8/23 01:09, Mirsad Todorovac wrote:
> > On 8/7/23 22:46, Kuniyuki Iwashima wrote:
> > > From: Mirsad Todorovac <mirsad.todorovac@xxxxxxxxxxxx>
> > > Date: Mon, 7 Aug 2023 21:44:41 +0200
> > > > Hi all,
> > > >
> > > > In the kernel 6.5-rc5 build on Ubuntu 22.04 LTS (jammy jellyfish) on a Ryzen 7950 assembled box,
> > > > vanilla torvalds tree kernel, the test test_unix_oob unexpectedly fails:
> > > >
> > > > # selftests: net/af_unix: test_unix_oob
> > > > # Test 2 failed, sigurg 23 len 63 OOB %
> > > >
> > > > It is this code:
> > > >
> > > >           /* Test 2:
> > > >            * Verify that the first OOB is over written by
> > > >            * the 2nd one and the first OOB is returned as
> > > >            * part of the read, and sigurg is received.
> > > >            */
> > > >           wait_for_data(pfd, POLLIN | POLLPRI);
> > > >           len = 0;
> > > >           while (len < 70)
> > > >                   len = recv(pfd, buf, 1024, MSG_PEEK);
> > > >           len = read_data(pfd, buf, 1024);
> > > >           read_oob(pfd, &oob);
> > > >           if (!signal_recvd || len != 127 || oob != '#') {
> > > >                   fprintf(stderr, "Test 2 failed, sigurg %d len %d OOB %c\n",
> > > >                   signal_recvd, len, oob);
> > > >                   die(1);
> > > >           }
> > > >
> > > > In 6.5-rc4, this test was OK, so it might mean we have a regression?
> > >
> > > Thanks for reporting.
> > >
> > > I confirmed the test doesn't fail on net-next at least, but it's based
> > > on v6.5-rc4.
> > >
> > >    ---8<---
> > >    [root@localhost ~]# ./test_unix_oob
> > >    [root@localhost ~]# echo $?
> > >    0
> > >    [root@localhost ~]# uname -r
> > >    6.5.0-rc4-01192-g66244337512f
> > >    ---8<---
> > >
> > > I'll check 6.5-rc5 later.
> >
> > Hi, Kuniyuki,
> >
> > It seems that there is a new development. I could reproduce the error with the failed test 2
> > as early as 6.0-rc1. However, the gotcha is that the error appears to be sporadically manifested
> > (possibly a race)?
> >
> > I am currently attempting a bisect.
>
> Bisect had shown that the condition existed already at 5.11 torvalds tree.
>
> It has to do with the configs chosen (I used the configs from seltests/*/config merged), but it
> is also present in the Ubuntu production build:
>
> marvin@defiant:~$ cd linux/kernel/linux_torvalds
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> Test 2 failed, sigurg 23 len 63 OOB %
> marvin@defiant:~/linux/kernel/linux_torvalds$ uname -rms
> Linux 6.4.8-060408-generic x86_64
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..1000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> Test 1 failed sigurg 0 len 63
> marvin@defiant:~/linux/kernel/linux_torvalds$
>
> It happens on rare occasions, so it seems to be a hard-to-spot race.
>
> Normal test running test_unix_oob once never noticed that, save by accident, which brought the problem to attention ...
>
> However, the problem seems to be config-driven rather than kernel-version-driven.
>
> marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..100000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
> Test 3.1 Inline failed, len 1 oob % atmark 0
> Test 1 Inline failed, sigurg 0 len 63
> Test 1 Inline failed, sigurg 0 len 63
> Test 1 Inline failed, sigurg 0 len 63
> Test 2 Inline failed, len 63 atmark 1
> Test 3 Inline failed, sigurg 23 len 63 data x
> Test 3 Inline failed, sigurg 23 len 63 data x
> Test 3 Inline failed, sigurg 23 len 63 data x
> Test 3 Inline failed, sigurg 23 len 63 data x
> Test 2 Inline failed, len 63 atmark 1
> Test 3.1 Inline failed, len 1 oob % atmark 0
> Test 2 failed, sigurg 23 len 63 OOB %
> marvin@defiant:~/linux/kernel/linux_torvalds$ uname -rms
> Linux 6.5.0-060500rc4-generic x86_64
> marvin@defiant:~/linux/kernel/linux_torvalds$
>
> At moments, I was able to reproduce with certain configs, but now something odd happens.
>
> I will keep investigating.

Please not that the bug persisted in 6.5-rc6:

marvin@defiant:~/linux/kernel/linux_torvalds$ for a in {0..100000}; do !!; done
for a in {0..100000}; do tools/testing/selftests/net/af_unix/test_unix_oob ; done
Test 2 failed, sigurg 23 len 63 OOB %
Test 2 Inline failed, len 63 atmark 1
Test 3 Inline failed, sigurg 23 len 63 data x
Test 2 failed, sigurg 23 len 63 OOB %
Test 3.1 Inline failed, len 1 oob % atmark 0
Test 3 Inline failed, sigurg 23 len 63 data x
Test 1 Inline failed, sigurg 0 len 63
Test 1 Inline failed, sigurg 0 len 63
Test 3.1 Inline failed, len 1 oob % atmark 0
Test 1 Inline failed, sigurg 0 len 63
Test 2 failed, sigurg 23 len 63 OOB %
Test 1 Inline failed, sigurg 0 len 63
Test 2 failed, sigurg 23 len 63 OOB %
Test 3.1 Inline failed, len 1 oob % atmark 0
Test 3.1 Inline failed, len 1 oob % atmark 0
marvin@defiant:~/linux/kernel/linux_torvalds$

The bug can be triggered as a non-privileged user, but is not clear whether it is exploitable to elevate privileges.

Best regards,
Mirsad Todorovac