The problem is that we need to support systems with older more. So we somehow need to check that more version that we're about to launch is new enough.

The problem is that we need to support systems with older more. So we somehow need to check that more version that we're about to launch is new enough.

That makes sense to me.

Now, is there a common way this is done in the systemd code base ?

I was thinking of a helper function in shared/pager.c that checks the --version of more with a combination of popen() and fopen() calls but not sure if thats sensible enough in this context. Any ideas ?

urks, i'd really prefer if we wouldn't have to add support for numerous XYZSECURE env vars for all the pagers around. Can we get util-linux to parse $PAGERSECURE in more in addition to MORESECURE? it would make a ton of set to augment $PAGER wit $PAGERSECURE, and if this indeed is a generic concept now, then the env var should be generic too.

i understand this would not really help with figuring out if a page supports the concept at all, but that's a separate issue.

our current baseline for util-linux is 2.41 right now (see README). 2.42 is not that far off, we might consider just bumping the dep to that whic would solve the MORESECURE feature check issue? that said, not sure how much that'd help if we push for PAGERSECURE instead...

@poettering when I added support for secure mode in more I also included PAGERSECURE as env, so yes this should be a good way to go.

btw, here's another issue. not sure how much i care, and i just wanted to mention that this exists, but buybox provides it's own crappy implementation of "more" with a very different feature set.

but hmm, i really don't care about busybox and we official do not offer compat with it anyway, so maybe let's ignore the busybox angle.

if you change the PR to use PAGERSECURE and drop MORESECURE and then bump the baseline i think we should be fine.

This sounds plausible! Fixing ...

Cutting off anything but util-linux 2.42 cuts off CentOS 9, CentOS 10, Fedora 42, Fedora 43 and the just released Debian 13 and Ubuntu 26.04. I don't think we can't really do this just yet...? It's basically everything but archlinux and rawhide...

Cutting off anything but util-linux 2.42 cuts off CentOS 9, CentOS 10, Fedora 42, Fedora 43 and the just released Debian 13 and Ubuntu 26.04. I don't think we can't really do this just yet...? It's basically everything but archlinux and rawhide...

ref: https://repology.org/project/util-linux/versions

I had just seen the doc commit, but looking at the diff it doesn't look like it changes the baseline at all? What is the failure mode if I run this with util-linux 2.41? What breaks precisely?

I had just seen the doc commit, but looking at the diff it doesn't look like it changes the baseline at all? What is the failure mode if I run this with util-linux 2.41? What breaks precisely?

The failure mode is that systemctl will open a version of more(1) that does not support secure mode, which could allow an unprivileged user to execute a shell as the EUID :(

Also, @bluca is there a place other than README where the baseline for runtime requirements is updated, I could not find anything obvious.

Either way, it might be a good idea to backport the secure mode to util-linux 2.41 at least. @karelzak what do you think ?

The failure mode is that systemctl will open a version of more(1) that does not support secure mode, which could allow an unprivileged user to execute a shell as the EUID :(

When you describe it like that, I really don't think this is a footgun we want to introduce. This needs at the very least to be enabled based on the version of util-linux used at build time, given we use libmount/libblkid there is a strong correlation anyway, so that should suffice.

Ok, so if I understand correctly this implies the introduction of something like HAVE_SECURE_MORE_PAGER or a better name.

more_pager = dependency('more',
                      version : '>= 2.42',
                      required : get_option('more_pager'))
have = more_pager.found()
conf.set10('HAVE_SECURE_MORE_PAGER', have)

Now, do we want to make PAGERSECURE env dependent on more(1)'s availability (probably not) ?

Either way, we can then use a macro to add more(1) to the secure pager list or if that is ugly an inline #ifdef HAVE_SECURE_MORE_PAGER.

#if HAVE_SECURE_MORE_PAGER
#define SECURE_MORE_PAGER "more",
#else
#define SECURE_MORE_PAGER
#endif /* HAVE_SECURE_MORE_PAGER */
...

if (use_secure_mode && !STR_IN_SET(pagers[i], "less", SECURE_MORE_PAGER "(built-in)"))
...

Makes sense ?

Something like that, can even use the existing libmount or libfdisk checks without checking for more, given it's the same sources

I was thinking of separating the check, because I was not too sure what side effects bumping the baseline for libmount to v2.42 could have.

Also, wouldn't it make sense to implicitly check for secure more(1), simply for HAVE_SECURE_MORE_PAGER and if its not the right version we build regardless just not with more in the list of secure pagers ? That means no need for required: get_option('more_pager').

sure

Cool! Preparing a new PR then :)

From a long-term point of view, maybe we can add util-linux.pc so we can ask for supported features with pkg-config --variable=<foo>, avoiding a strict dependency on a specific version.

From a long-term point of view, maybe we can add util-linux.pc so we can ask for supported features with pkg-config --variable=<foo>, avoiding a strict dependency on a specific version.

That would be nice, yes please!

I don't think this should have been merged. That util-linux release is only present in a few places right now, and this means that unless people carefully check, we're opening them up to a vulnerability. This is just the wrong way to approach this.