Re: CRUX 3.7: who command shows no logged in users
Hello Markus,
I have made an update from CRUX 3.6 to 3.7. Now I have discovered that the output of the "who" command is empty.
For me the output of the "who" command has never listed all the users currently logged in to the system, because my login manager (greetd) does not write to /run/utmp. Since learning of this behaviour I stopped paying any attention to the output of "who". But your next observation reveals a more fundamental issue.
I found out that /var/run/utmp does not exist. But I have an empty file /run/utmp with a timestamp of the last boot time.
With a proper installation of the core/filesystem package from the ISO, /var/run in the former path should be a symlink to /run. Your report of a missing /var/run/utmp motivated me to inspect one of my own 3.6->3.7 upgrades, and I then discovered that /var/run had been left intact, not converted to a symlink during the upgrade! (This particular upgrade took place with the rc4 ISO, but I suspect that the September 26 release ISO would have behaved the same way.) As discussed on #crux-devel [1] the setup script calls pkgadd with the '-f' flag during upgrades. But if /var/run had been non-empty when you booted the installation media (as it was on the machine that I just inspected), then pkgadd might have quietly refused to recursively delete that directory even when called with the '-f' flag. As a result, you have a broken installation of the core/filesystem port, and misleading output from the "who" command. The volatile nature of the contents under /var/run distinguishes this run of 'pkgadd -u -f' from the corresponding attempt to force update the opt/linux-firmware port. Because everything under /lib/firmware/qcom/LENOVO/21BX was already known to be owned by the old version of linux-firmware, 'pkgadd -u -f' could proceed with the update even though this path is a symlink in the newer version. In contrast, it looks like pkgadd was unwilling to clean out the contents of /var/run left behind from the last boot of your 3.6 system, and so the symlink never got created properly. Tim has responded already with a simple fix. But the related issue of symlink handling in pkgutils [2] is still unresolved. Thanks Markus, for giving us another reason to revisit this long-standing bug in pkgutils. -- John [1] https://libera.irclog.whitequark.org/crux-devel/2022-09-23 [2] https://crux.nu/bugs/index.php?do=details&task_id=15
Hello John. On 02.10.2022 20:29, jmcquah@disroot.org wrote:
I found out that /var/run/utmp does not exist. But I have an empty file /run/utmp with a timestamp of the last boot time.
With a proper installation of the core/filesystem package from the ISO, /var/run in the former path should be a symlink to /run. Your report of a missing /var/run/utmp motivated me to inspect one of my own 3.6->3.7 upgrades, and I then discovered that /var/run had been left intact, not converted to a symlink during the upgrade! (This particular upgrade took place with the rc4 ISO, but I suspect that the September 26 release ISO would have behaved the same way.)
As discussed on #crux-devel [1] the setup script calls pkgadd with the '-f' flag during upgrades. But if /var/run had been non-empty when you booted the installation media (as it was on the machine that I just inspected), then pkgadd might have quietly refused to recursively delete that directory even when called with the '-f' flag. As a result, you have a broken installation of the core/filesystem port, and misleading output from the "who" command.
This has been exactly the issue on my system too: The old folder /var/run/ did still have some contents in it so it had not been deleted.
The volatile nature of the contents under /var/run distinguishes this run of 'pkgadd -u -f' from the corresponding attempt to force update the opt/linux-firmware port. Because everything under /lib/firmware/qcom/LENOVO/21BX was already known to be owned by the old version of linux-firmware, 'pkgadd -u -f' could proceed with the update even though this path is a symlink in the newer version. In contrast, it looks like pkgadd was unwilling to clean out the contents of /var/run left behind from the last boot of your 3.6 system, and so the symlink never got created properly.
Exactly. I have also noticed I had to use "pkgadd -f" while upgrading "linux-firmware" when still being on CRUX 3.6 because of a symlink.
Tim has responded already with a simple fix. But the related issue of symlink handling in pkgutils [2] is still unresolved. Thanks Markus, for giving us another reason to revisit this long-standing bug in pkgutils.
For the /var/run issue I am unsure if this case should have been handled by pkgutils: At least on my system there have been some leftovers which had been owned by old unrelated port versions or even by no port. I think generally deleting a folder which has some contents in it automatically is not a good idea. For the "linux-firmware" issue I think it could be right to be handled by pkgutils, also without "-f" parameter. If all files which need to be deleted belong to the same port which is about to be updated they can safely be removed.
-- John
[1] https://libera.irclog.whitequark.org/crux-devel/2022-09-23 [2] https://crux.nu/bugs/index.php?do=details&task_id=15
Best regards Markus Heinz
Hello Markus, Thanks for your feedback on the pkgutils symlink behaviour. I agree that the /var/run issue can be addressed on its own, without having to patch pkgutils for all subsequent instances when a port update converts a directory to a symlink.
For the /var/run issue I am unsure if this case should have been handled by pkgutils: At least on my system there have been some leftovers which had been owned by old unrelated port versions or even by no port.
Avoidance of data loss was the motivation for the following patch to iso/bin/setup-helper. In #crux-devel last week it was suggested that we add a test such as [ -d /var/run ] into rc.fix, but I think a one-time fix restricted to the ISO boot environment is cleaner than a function that runs every time a 3.7 system starts up. diff --git a/iso/bin/setup-helper b/iso/bin/setup-helper index b28789c..f202389 100755 --- a/iso/bin/setup-helper +++ b/iso/bin/setup-helper @@ -108,6 +108,14 @@ inject_packages jsoncpp libuv lzlib rhash # new deps of core/iptables inject_packages nftables libedit jansson libnftnl libmnl +# new filesystem layout +if [ $(find "$ROOT/var/run" 2>/dev/null | wc -l) -gt 1 ]; then + mv "$ROOT/var/run" "$ROOT/var/run36" + echo "Nonempty directory /var/run has been backed up to /var/run36." + echo "Inspect after reboot and feel free to delete." +else + rm -rf "$ROOT/var/run" +fi if [ `is_installed xfsprogs` ]; then inject_packages inih liburcu Best Regards, John
participants (3)
-
jmcquah＠disroot.org -
John McQuah
-
Markus Heinz