Lennart Poettering

oh, before you peek you need to wait for IO via poll() (or in our tree: fd_wait_for_event() which is a wrapper around it.

ah, this works on TCP but doesn't on AF_UNIX. kinda sucks. Seems like a bug/omission in the kernel. i think we can still save this thoug. After all we don't...

i think we are almost there, one more round

interesting, i like you did this without libdevmapper, it's a dep i would like to get rid of in general. i'd prefer if this PR could be split in multiple...

hmm, what's the rationale for shoehorning this into the mount options? shouldn't this be rather done like cryptsetup/veritysetup/integritysetup, i.e. via an /etc/clonetab?

hmm? the whole of userdb.c is written so that every thread can run their own event loop handling independently of any other. The only place where locking is necessary (and...

I ran your testcase here, it's absolutely rock solid: ``` === TEST COMPLETED SUCCESSFULLY === All 12 threads completed 13600 NSS operations each Total NSS calls: ~163200 No crashes detected...

This is highly unlikey to be a systemd problem. systemd just tells the kernel to suspend, and the kernel does all the complex work. If you run new systemd (i.e....

hmm, maybe this is caused by the cgroups freeze logic. i.e. do you have live nfs shares, some nvidia closed source crap driver, or are having live kvm machines while...

Does the issue go away if you set SYSTEMD_SLEEP_FREEZE_USER_SESSIONS=0 env var for the systemd-suspend.service unit? i.e.: Create a drop-in file /etc/systemd/system/systemd-suspend.service.d/something.conf with the following: ``` [Service] Environment="SYSTEMD_SLEEP_FREEZE_USER_SESSIONS=0 ```