unix-yy.test race fixed

[Dmitry V. Levin]

On Thu, Mar 05, 2015 at 05:56:26PM +0300, Dmitry V. Levin wrote:
> On Wed, Mar 04, 2015 at 10:55:45PM -0500, Mike Frysinger wrote:
> > vFAIL:  test;  s390x   64-bit/MSB  linux-3.19.0         kernel-headers-3.16.0  glibc-2.19  gcc-4.8.3
> 
> It's unix-yy.test again with the same issue that looks like a race
> condition: from time to time, print_sockaddr_by_inode fails to fetch
> netlink information.  I've run series of unix-yy.test on a x86_64 system
> and got about 0.4% chance of failure.  On these s390/s390x hosts, though,
> unix-yy.test fails with probability over 50%.

Yes, it was a genuine kernel race: NETLINK_SOCK_DIAG interface might send
a zero UNIX_DIAG_PEER message for a freshly connected socket if the accept
syscall hasn't been completed at this point.
Fixed with commit da66e25779635e59b42a4477ef1e8b6b6917dac6.


-- 
ldv
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 181 bytes
Desc: not available
URL: <http://lists.strace.io/pipermail/strace-devel/attachments/20150305/0976d117/attachment.bin>