1.48.5: test_file_architecture_26 and test_xfs_growfs_0 failing

[dvzrv]

Hi! While packaging 1.48.5 for Arch Linux I ran into these two tests failing:
test_file_architecture_26 and test_xfs_growfs_0.

libguestfs-1.48.5-1-x86_64-build.log
libguestfs-1.48.5-1-x86_64-check.log
test-suite.log

[rwmjones]

libguestfs: trace: file_architecture "/lib-i586.so.zst"
libguestfs: trace: file_architecture = NULL (error)
libguestfs: error: file_architecture: external command ‘file’ killed by signal -
24
FAIL: test_file_architecture_26

signal 24 may be SIGXCPU, indicating some sort of crash. Not sure what to say about this, but it might be significant that the test involves a zstd file which we only added support for recently. Is it reproducible if you run file ./test-data/files/lib-i586.so.zst on the host?

537/550 test_xfs_growfs_0
libguestfs: error: mkfs: xfs: /dev/VG/LV: Filesystem must be larger than 300MB.

This is a bug (of sorts) caused by changing requirements again in xfs. Our test disk was only 40MB. I will see if I can fix this.

[dvzrv]

Is it reproducible if you run file ./test-data/files/lib-i586.so.zst on the host?

That seems fine (outside of the build environment):

file ./test-data/files/lib-i586.so.zst
./test-data/files/lib-i586.so.zst: Zstandard compressed data (v0.8+), Dictionary ID: None

[rwmjones]

Thanks. Another thing to try would be something like:

$ ./run virt-rescue --ro -a test-data/test.iso -m /dev/sda

><rescue> find / -name '*.zst'
/sysroot/lib-i586.so.zst
><rescue> file -bsL /sysroot/lib-i586.so.zst
Zstandard compressed data (v0.8+), Dictionary ID: None

I expect that the last command may segfault for you inside the appliance, although I don't know why.

[dvzrv]

Hmm, that actually also went fine (in the build environment):

><rescue> file -bsL /sysroot/lib-i586.so.zst
Zstandard compressed data (v0.8+), Dictionary ID: None

[dvzrv]

During the build the test still fails though.
Is there a way for me to disable it for now?

btw: Your patch for test_xfs_growfs_0 works as intended! Thanks!

[dvzrv]

Ah, I guess it would be export SKIP_TEST_FILE_ARCHITECTURE_26=1

[rwmjones]

Right, although I'm confused/worried why that test is failing ...

[dvzrv]

Let me know if I can test further things. Build/test always takes quite a while (even on my machine) but I can try again tomorrow.

[Toolybird]

It's probably the 'z' switch:

><rescue> file -zbsL /sysroot/lib-i586.so.zst
Bad system call

Arch compiles its "file" utility with seccomp support whereas it appears Fedora does not. At least that's what it looks like after checking the koji build log:

checking for seccomp_init in -lseccomp... no

Is that intentional by Fedora?

[rwmjones]

Oh, well spotted! This could well be the case. Let me compile the Fedora package with seccomp support to see if I can reproduce the problem ...

[rwmjones]

Yes, I can reproduce this. In fact this is both a bug upstream (in the seccomp support) and in Fedora. I will file bugs for both.

[rwmjones]

Fedora bug: https://bugzilla.redhat.com/show_bug.cgi?id=2148753

[rwmjones]

Upstream bug: https://bugs.astron.com/view.php?id=406

[Toolybird]

re: the upstream report...not sure that'll fly as it's kinda expected when the sandbox is in effect. See the man page [1].

The libguestfs test can be made to pass if the "file" call on line 46 of daemon/file.ml is changed from "-zb" to "-zbS", but I'm not sure if that's what you really want.

[1] https://man.archlinux.org/man/file.1.en#SECURITY