Re: [BUG] selftests: af_unix: unix:diag.c does not compile on AlmaLinux 8.7

From: Mirsad Todorovac
Date: Wed May 24 2023 - 10:15:42 EST


Hi,

On 5/22/23 20:26, Kuniyuki Iwashima wrote:

I launched AlmaLinux/RockyLinux 8.7 and 9.2 with images listed in the pages
below.

https://wiki.almalinux.org/cloud/AWS.html#community-amis
https://rockylinux.org/cloud-images/

The kernel versions in each image were :

8.7:
Alma : 4.18.0-425.3.1.el8.x86_64
Rocky : 4.18.0-425.10.1.el8_7.x86_64

9.2:
Alma : 5.14.0-284.11.1.el9_2.x86_64
Rocky : 5.14.0-284.11.1.el9_2.x86_64

So, this is not a bug. It's just because v4.18 does not support
UNIX_DIAG_UID, which was introduced in v5.3.

You should install 5.3+ kernel if you want to build the test.

Thanks,
Kuniyuki

Hi, Kuniyuki,

Good point. However, newer kernel won't save me from old /usr/include
headers, will it?

Sorry, I meant kernel and kernel-headers package that should be
updated along with kernel.

You should use proper header files that match to the actual kernel
version running on the machine.


I was actually testing the 6.4-rc3 on AlmaLinux 8.7, as it is my only
RHEL-based box ...

What would then be the right action?

make headers_install ?

I would rather not to. For the installation to remain manageable,
preferably I'd have a kernel-devel RPM built.

If it was a #define instead of enum, I'd probably work around and
exclude the test that doesn't fit the kernel, or the system call
would return -EINVAL?

Including from the includes that came with the kernel might be
a solution:

../../../../../include/uapi/linux/unix_diag.h:44: UNIX_DIAG_UID,

Alas, when I try to include, I get these ugly errors:

[marvin@pc-mtodorov af_unix]$ gcc -I ../../../../../include/ diag_uid.c
In file included from ../../../../../include/linux/build_bug.h:5,
from ../../../../../include/linux/bits.h:21,
from ../../../../../include/linux/capability.h:18,
from ../../../../../include/linux/netlink.h:6,
from diag_uid.c:8:
../../../../../include/linux/compiler.h:246:10: fatal error:
asm/rwonce.h: No such file or directory
#include <asm/rwonce.h>
^~~~~~~~~~~~~~

FWIW, this is provided by kernel-devel package.

Actually, what is provided is essentially the same as before:

[root@pc-mtodorov kernel]# rpm -q --fileprovide kernel-devel-6.3.3 | grep rwonce.h
/usr/src/kernels/6.3.3-100.fc37.x86_64/arch/x86/include/generated/asm/rwonce.h
/usr/src/kernels/6.3.3-100.fc37.x86_64/include/asm-generic/rwonce.h
[root@pc-mtodorov kernel]#

compilation terminated.
[marvin@pc-mtodorov af_unix]$ vi +246
../../../../../include/linux/compiler.h
[marvin@pc-mtodorov af_unix]$ find ../../../../../include -name rwonce.h
../../../../../include/asm-generic/rwonce.h
[marvin@pc-mtodorov af_unix]$

Minimum reproducer is:

[marvin@pc-mtodorov af_unix]$ gcc -I ../../../../../include/ reproducer.c
In file included from ../../../../../include/linux/build_bug.h:5,
from ../../../../../include/linux/bits.h:21,
from ../../../../../include/linux/capability.h:18,
from ../../../../../include/linux/netlink.h:6,
from reproducer.c:5:
../../../../../include/linux/compiler.h:246:10: fatal error:
asm/rwonce.h: No such file or directory
#include <asm/rwonce.h>
^~~~~~~~~~~~~~
compilation terminated.
[marvin@pc-mtodorov af_unix]$

[marvin@pc-mtodorov af_unix]$ nl reproducer.c

1 #define _GNU_SOURCE
2 #include <linux/netlink.h>

[marvin@pc-mtodorov af_unix]$

Am I doing something very stupid right now, for actually I see

#include <asm/rwonce.h>

in "include/linux/compiler.h" 248L, 7843C

while actual rwonce.h is in <asm-generic/rwonce.h>

[marvin@pc-mtodorov af_unix]$ find ../../../../../include -name rwonce.h
../../../../../include/asm-generic/rwonce.h
[marvin@pc-mtodorov af_unix]$

I must be doing something wrong, for I see that the kernel compiled
despite not having include/asm ?

When looking at the invocations of rwonce.h in the kernel, they seem to
be equally spread between <asm-generic/rwonce.h> and <asm/rwonce.h> :

[marvin@pc-mtodorov af_unix]$ grep --include="*.[ch]" -n -w rwonce.h -r ../../../../.. 2> /dev/null | less
../../../../../arch/alpha/include/asm/rwonce.h:33:#include <asm-generic/rwonce.h>
../../../../../arch/arm64/include/asm/rwonce.h:71:#include <asm-generic/rwonce.h>
../../../../../arch/arm64/kvm/hyp/include/nvhe/spinlock.h:18:#include <asm/rwonce.h>
../../../../../arch/s390/include/asm/rwonce.h:29:#include <asm-generic/rwonce.h>
../../../../../arch/x86/include/generated/asm/rwonce.h:1:#include <asm-generic/rwonce.h>
../../../../../include/asm-generic/barrier.h:18:#include <asm/rwonce.h>
../../../../../include/kunit/test.h:29:#include <asm/rwonce.h>
../../../../../include/linux/compiler.h:246:#include <asm/rwonce.h>

I figured out I must be doing something wrong or the kernel otherwise
would not build for me.

Eventually, the UNIX_DIAG_UID enum is used in only one place:

ASSERT_EQ(attr->rta_type, UNIX_DIAG_UID);

That particular test should fail in case of kernel older than 5.3.

We don't expect it to be run on older kernels in the first place.

Certainly.

However, I fell into a terrible mess where one thing breaks the other.

I can't seem to make this work.

Thanks,
Mirsad

--
Mirsad Goran Todorovac
Sistem inženjer
Grafički fakultet | Akademija likovnih umjetnosti
Sveučilište u Zagrebu

System engineer
Faculty of Graphic Arts | Academy of Fine Arts
University of Zagreb, Republic of Croatia