vmm: Fully adopt the getrlimit(2) machinery for limiting the number of VMM instances

[bnovkov]

Summary

The vmm module currently combines parts of the kernelspace
getrlimit(2) machinery and a sysctl value to limit the number
of VMM instances per-UID. The original rationale for this
approach was to sidestep the convoluted way in which per-UID
getrlimit(2) limit values are currently propagated.

However, this approach deviates from the expected workflow when dealing
getrlimit(2) limits.

Fix this by teaching login(1) about RLIMIT_VMM and changing
vmmdev_create to use lim_cur. While we're here, document the limit
in the getrlimit(2) manpage.

Fixes: 1092ec8b3375 ("kern: Introduce RLIMIT_VMM")

[des]

usr.bin/login/login.conf

47

[markj]

usr.bin/login/login.conf

47

So we're changing the existing limit to unlimited, I guess because there's no way to express a constant like 4 * mp_ncpus here?

[des]

usr.bin/login/login.conf

47

It's still named vms elsewhere, I think we should standardize on vmm to reflect RLIMIT_VMM.

[bnovkov]

usr.bin/login/login.conf

47

Yes, I'm not aware of way to dynamically populate login.conf with a machine-specific value.

[markj]

sys/dev/vmm/vmm_dev.c

1021

Do you mind adding an EXTERROR() here? Could be done in a separate patch.