Hello everybody,
systemd v233 introduced code that makes use of the kernel keyring, initializing a private keyring for every service and adding a protected key named "invocation_id". This caused some trouble and we reverted it since then.
Things will change with systemd v235, which adds a new option "KeyringMode=" for units. The values are "inherit", "private" and "shared". The commit [0] message and changes give the details. Now cryptsetup units are generated with "KeyringMode=shared", which unbreaks this use case. Other services that use the kernel keyring and want to share secrets with other services have to add this as well.
However login sessions where user context is changed can not be handled by systemd. Looks like we have to update our PAM configurations and add a line for every service where session is expected to use the kernel keyring:
session optional pam_keyinit.so force revoke
This is required for eCryptfs to function properly. Any comments on this? Any concerns?
I would like to keep the upstream keyring behavior with release version 235. Would be nice to have this sorted before.
[0] https://github.com/systemd/systemd/commit/b1edf4456eabc5951d76b96bc7df2db3fe...
Christian Hesse list@eworm.de on Mon, 2017/09/18 14:29:
Hello everybody,
systemd v233 introduced code that makes use of the kernel keyring, initializing a private keyring for every service and adding a protected key named "invocation_id". This caused some trouble and we reverted it since then.
Things will change with systemd v235, which adds a new option "KeyringMode=" for units. The values are "inherit", "private" and "shared". The commit [0] message and changes give the details. Now cryptsetup units are generated with "KeyringMode=shared", which unbreaks this use case. Other services that use the kernel keyring and want to share secrets with other services have to add this as well.
However login sessions where user context is changed can not be handled by systemd. Looks like we have to update our PAM configurations and add a line for every service where session is expected to use the kernel keyring:
session optional pam_keyinit.so force revoke
This is required for eCryptfs to function properly. Any comments on this? Any concerns?
I would like to keep the upstream keyring behavior with release version 235. Would be nice to have this sorted before.
[0] https://github.com/systemd/systemd/commit/b1edf4456eabc5951d76b96bc7df2db3fe...
So we have a flyspray ticket requesting the same [1] and a report from Mantas who is already using a setup with pam_keyinit.
As systemd upstream started preparing a release and milestone items are being resolved [2] I would like to see some progress. Who will do this? Dave, do you update pambase? Do we add a todo-list containing all packages with pam configuration files so maintainers can decide on their own whether or not this is feasible for the package?
[1] https://bugs.archlinux.org/task/54915 [2] https://github.com/systemd/systemd/milestone/12
Christian Hesse list@eworm.de on Fri, 2017/09/29 16:30:
Christian Hesse list@eworm.de on Mon, 2017/09/18 14:29:
Hello everybody,
systemd v233 introduced code that makes use of the kernel keyring, initializing a private keyring for every service and adding a protected key named "invocation_id". This caused some trouble and we reverted it since then.
Things will change with systemd v235, which adds a new option "KeyringMode=" for units. The values are "inherit", "private" and "shared". The commit [0] message and changes give the details. Now cryptsetup units are generated with "KeyringMode=shared", which unbreaks this use case. Other services that use the kernel keyring and want to share secrets with other services have to add this as well.
However login sessions where user context is changed can not be handled by systemd. Looks like we have to update our PAM configurations and add a line for every service where session is expected to use the kernel keyring:
session optional pam_keyinit.so force revoke
This is required for eCryptfs to function properly. Any comments on this? Any concerns?
I would like to keep the upstream keyring behavior with release version 235. Would be nice to have this sorted before.
[0] https://github.com/systemd/systemd/commit/b1edf4456eabc5951d76b96bc7df2db3fe...
So we have a flyspray ticket requesting the same [1] and a report from Mantas who is already using a setup with pam_keyinit.
As systemd upstream started preparing a release and milestone items are being resolved [2] I would like to see some progress. Who will do this? Dave, do you update pambase? Do we add a todo-list containing all packages with pam configuration files so maintainers can decide on their own whether or not this is feasible for the package?
[1] https://bugs.archlinux.org/task/54915 [2] https://github.com/systemd/systemd/milestone/12
Pushed systemd 235.0-1 and pambase 20171006-1 to [testing]... Let's wait for people to complain. :-p
arch-dev-public@lists.archlinux.org