On So, 23.08.20 06:34, Andrii Zymohliad (azymohliad@xxxxxxxxxxxxxx) wrote: > Thanks for taking time to read this far! Is there anything obvious > here that I can do to fix it? Or any hints where to look? So as you figured out by now this is due to homed's "discard" handling. At logout we'll discard all unused blocks by default, and at login we'll reallocte them. If we can't reallocate them because the underlying fs doesn't have enough space we currently refuse login. (This appeared to be the right thing to do, given that the avoiding "discard" is really about guarantees on IO, and if we can't give those guarantees its pointless.) I want to make this a bit smarter: 1. We should probably never refuse login, but simply accept discarding in the worst case. 2. We should provide better errors and log messages explaining the situation. In particular if we cannot fallocate() the backing store, we should log about this loudly somehow. 3. It might be an option to only allow a read-only login when discard is off and we cannot fallocate(). 4. Most importantly though, we should elastically resize the home dir during login, logout and during runtime. Meaning: instead of configuring a fixed size for your $HOME (and otherwise only explicitly resizing on user request) you'd instead just declare how much space on the underlying fs shall stay free and how much inside your $HOME and then we'll try to keep things in that range by resizing automatically. Lennart -- Lennart Poettering, Berlin _______________________________________________ systemd-devel mailing list systemd-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/systemd-devel