Slink exits?

[jas4711]

Many thanks for adding slink! (#2)

Is anyone else getting exits from the slink image after running the first command? Shell builtins seems to work, and commands seems to run fine but as soon as they return, the image exits. Potato and later images works fine for me.

jas@kaka:~$ podman run -it --rm debian/eol:slink
a940937b7dd0:/# echo *
bin boot cdrom dev etc floppy home initrd lib mnt proc root run sbin sys tmp usr var
a940937b7dd0:/# cd etc
a940937b7dd0:/etc# echo *
adduser.conf adjtime ae ae.rc alternatives apt conf.modules console-tools cron.d cron.daily cron.weekly debian_version default dpkg fdmount.conf fstab gateways group host.conf hostname hosts hosts.allow hosts.deny inetd.conf init.d inittab inputrc isapnp.conf isapnp.gone issue issue.net ld.so.cache ld.so.conf lilo.conf limits locale.alias localtime login.access login.defs mediaprm modules modutils motd mtab nsswitch.conf pam.conf pam.d passwd porttime profile protocols pwdb.conf rc.boot rc0.d rc1.d rc2.d rc3.d rc4.d rc5.d rc6.d rcS.d resolv.conf rpc securetty security services shells skel syslog.conf terminfo updatedb.conf
a940937b7dd0:/etc# ls
adduser.conf	cron.d		gateways	init.d		ld.so.conf	modules		porttime	rc3.d		security
adjtime		cron.daily	group		inittab		lilo.conf	modutils	profile		rc4.d		services
ae		cron.weekly	host.conf	inputrc		limits		motd		protocols	rc5.d		shells
ae.rc		debian_version	hostname	isapnp.conf	locale.alias	mtab		pwdb.conf	rc6.d		skel
alternatives	default		hosts		isapnp.gone	localtime	nsswitch.conf	rc.boot		rcS.d		syslog.conf
apt		dpkg		hosts.allow	issue		login.access	pam.conf	rc0.d		resolv.conf	terminfo
conf.modules	fdmount.conf	hosts.deny	issue.net	login.defs	pam.d		rc1.d		rpc		updatedb.conf
console-tools	fstab		inetd.conf	ld.so.cache	mediaprm	passwd		rc2.d		securetty
a940937b7dd0:/etc# exit
jas@kaka:~$ 

[tianon]

When I was creating these, I recall running into a lot of issues thanks to 64bit inodes -- the kernel does its very best to shield older code like this from that complexity, but there's only so much it can do before things break (and strace confirmed it was the inodes breaking it, IIRC). Maybe that's what you're seeing?

I think I had more luck with using QEMU user-mode emulation to emulate x86 on my amd64, but maybe that's just something I tried and it didn't actually work better? (Sorry, it's been a while now!)

[mirabilos]

slink is tricky to invoke; compare e.g. https://github.com/MirBSD/int/blob/6b2fe241e23bc52e428eb317bd531ce9cee2aea6/.github/workflows/deb-eol.yml#L10 for slink to https://github.com/MirBSD/int/blob/6b2fe241e23bc52e428eb317bd531ce9cee2aea6/.github/workflows/deb-eol.yml#L26 for potato.

apt also segfaults installing, cf. https://github.com/MirBSD/int/blob/6b2fe241e23bc52e428eb317bd531ce9cee2aea6/.extra/debeol.sh#L48-L52

If this is indeed related to inodes, then it’s important that a 64-bit qemu-user is not used to run these. Either a 32-bit qemu-user or no qemu-user, just the 32-bit binaries directly, in which case the kernel supposedly translates the hashes to DTRT. Or use btrfs/xfs instead of ext4. (We have the same problem with qemu-user-based buildds for 32-bit Debian-Ports.)

But, having traced the problems a little, it also seems somehow that libc on slink passes smaller or not entirely initialised fields for some syscalls, compared to what recent kernels (5.10/bullseye) expect. So, treat with care.

With the same · workaround I got mksh to pass its testsuite under it, so it might just be something with apt in slink… which incidentally was my first Debian, and I never used apt with it because it was “new and highly experimental”, so we stuck to dselect back then.