Failing tests in master branch as of commit a8c50391d0e264e56be494f54803b8b5e655549d

[Dmitry V. Levin]

On Fri, Dec 03, 2021 at 03:33:45PM +0100, Renaud Métrich wrote:
> Hello,
> 
> Not sure you are aware, but while resync my secontext PR 
> https://github.com/strace/strace/pull/197 I falled into 3 tests failing, 
> not related to my implementation:
> 
> FAIL: ioctl_block.gen.test
> FAIL: ioctl_block--pidns-translation.test
> FAIL: net-yy-unix.test
> 
> ioctl* ones are related to changes in the output for BLKGETDISKSEQ, e.g.:
> 
> -638534 ioctl(-1, BLKGETDISKSEQ, NULL) = -1 EBADF (Bad file descriptor)
> +638534 ioctl(-1, _IOC(_IOC_READ, 0x12, 0x80, 0x8), NULL) = -1 EBADF 
> (Bad file descriptor)

No, I haven't seen this anywhere yet.
What's the environment where you observe it?

> and net-yy-unix.test is related to changes in the output of unix 
> sockets, e.g.:
> 
> -connect(4<UNIX:[2474462]>, {sa_family=AF_UNIX, sun_path="net-yy-unix.socket"}, 21) = 0
> +connect(4<socket:[2474462]>, {sa_family=AF_UNIX, sun_path="net-yy-unix.socket"}, 21) = 0

Yes, I've started seeing this a few days before the latest release on the
following two hosts:

Linux aarch64-test01.fedorainfracloud.org 5.15.5-200.fc35.aarch64 #1 SMP Fri Nov 26 00:30:50 UTC 2021 aarch64 aarch64 aarch64 GNU/Linux
Linux rawhide-test.fedorainfracloud.org 5.15.0-0.rc3.20211001git4de593fb965f.30.fc36.x86_64 #1 SMP Fri Oct 1 16:47:19 UTC 2021 x86_64 x86_64 x86_64 GNU/Linux

It looks like an ABI change introduced in the latest fedora kernels so
that getxattr("/proc/PID/fd/FD", "system.sockprotoname") now returns
"UNIX-STREAM" instead of "UNIX".

The regression was introduced by Linux kernel commit
v5.15-rc1~157^2~22^2~30^2~3, and the subsequent attempt to fix
backwards compatibility v5.15-rc6~34^2~22 didn't help either.

Please report this kernel regression upstream.


-- 
ldv