[systemd-devel] Switch root slowness

[Lennart Poettering]

On Thu, 30.10.14 14:35, Lennart Poettering (lennart at poettering.net) wrote:

> On Thu, 30.10.14 13:18, Jan Synáček (jsynacek at redhat.com) wrote:
> 
> > Hello,
> > 
> > commit 539618a0ddc2dc7f0fbe28de2ae0e07b34c81e60
> > Author: Lennart Poettering <lennart at poettering.net>
> > Date:   Wed Oct 29 17:06:32 2014 +0100
> > 
> >     util: make use of the new getrandom() syscall if it is available when needing entropy
> > 
> >     Doesn't require an fd, and could be a bit faster, so let's make use of
> >     it, if it is available.
> > 
> > Oct 30 13:08:37 fedora-rawhide-systemd-virt kernel: EXT4-fs (sda3): mounted filesystem with ordered data mode. Opts: (null)
> > Oct 30 13:08:38 fedora-rawhide-systemd-virt systemd[1]: Switching root.
> > Oct 30 13:08:38 fedora-rawhide-systemd-virt systemd-journal[109]: Journal stopped
> > ^^^^
> > Hangs here for a while with no output.
> > vvvv
> > Oct 30 13:09:44 fedora-rawhide-systemd-virt systemd-journal[279]: Runtime journal is using 8.0M (max allowed 100.0M, trying to leave 150.1M free of 992.7M available → current limit 100.0M).
> > Oct 30 13:09:44 fedora-rawhide-systemd-virt systemd-journal[279]: Runtime journal is using 8.0M (max allowed 100.0M, trying to leave 150.1M free of 992.7M available → current limit 100.0M).
> > Oct 30 13:09:44 fedora-rawhide-systemd-virt systemd-journald[109]: Received SIGTERM from PID 1 (systemd).
> > <snip a lot of selinux related stuff>
> > Oct 30 13:09:44 fedora-rawhide-systemd-virt kernel: random: nonblocking pool is initialized
> > 
> > Is anyone else running into this?
> 
> This could be due to the blocking nature of getrandom() if little
> entropy is available. The call waits by default until the pool is
> initialized before returning, which is probably what your system is
> waiting on.
> 
> I wish there was a way how we could use getrandom() in a way like
> /dev/urandom, where we can pull out the non-initialized data
> anyway. In absence of that we can just fallback to /dev/urandom on
> EAGAIN I guess, and always pass GRND_NONBLOCK.

I have now implemented that. Please test!

Lennart

-- 
Lennart Poettering, Red Hat