On Sat, November 25, 2017 21:27, Thomas Trepl wrote:

> Hi all,

>

> after digging around to get LFS booting on a mdadm-raid (based on
a

> fakeraid on C236 chipset) I've collected a few changes to some

> essentials files.

>

> LFS bootscripts

> Problem was that after every reboot, the array has been rebuild.
While

> rebuild, the system is nearly unusable slow and the rebuild takes
a

> good amount of time (here 1.5h). The reason seems to be that the

> initscripts - here especially sendsignals - do not really take care
of

> raid services. sendsignals sends SIGTERM/SIGKILL to any running

> process, including a mdmon process started earlier in initramfs.
This

> makes it impossible for mdmon to mark the array as clean. Booting
an

> dirty array forces a rebuild automatically. The patch to
sendsignals

> simply omits the pid of mdmon from being killed.

> Next, in mountfs I added a wait-clean after the root filesystem
has

> been remounted to be read-only on shutdown. Even though mdmon
should

> mark the array clean when no writes has occured for a certain amount
of

> time, it is kind of "forced" by calling mdadm to do so.

>

> Initramfs

> This patch belongs to BLFS. It simply adds readlink and basename
to

> initramsfs as some udev rules do refer to them. Unfortunatly, they
are

> called qualified in /usr/bin so it was not enough to simply add them
to

> $binfiles or $sbinfiles. A symlink in /usr to /bin makes those
pgms

> accessable via /usr/bin.

>

> Would you mind to have a look to the patches and discuss whether
they

> could applied?

>

> --

> Thomas

> --

> http://lists.linuxfromscratch.org/listinfo/lfs-dev

> FAQ: http://www.linuxfromscratch.org/faq/

> Unsubscribe: See the above information page


My initramfs needs to support (read-only) root filesystem being an
image on an NTFS.  Obviously I cannot let the root filesystem dies
with ntfs-3g being killed during sendsignal.  My solution was to
write down the pid of the ntfs-3g somewhere the sendsignal script may find
it later, because I do need any other instances of ntfs-3g to be
killed.  But I guess it is unlikely one would want to run more than
one mdadm so your patch should work.


Alex

 

-- 
http://lists.linuxfromscratch.org/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page

Reply via email to