<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<div class="moz-cite-prefix">For convenience, here's a link below in
case people may wondering which patchset we are talking about.<br>
<pre wrap=""> <a class="moz-txt-link-freetext" href="http://git.pokylinux.org/cgit.cgi/poky-contrib/log/?h=ChenQi/read-only-rootfs-in-live-images">http://git.pokylinux.org/cgit.cgi/poky-contrib/log/?h=ChenQi/read-only-rootfs-in-live-images</a></pre>
<br>
//Chen Qi<br>
<br>
On 08/05/2013 10:18 AM, ChenQi wrote:<br>
</div>
<blockquote cite="mid:51FF0B8A.9030201@windriver.com" type="cite">
<meta http-equiv="Content-Type" content="text/html;
charset=ISO-8859-1">
<div class="moz-cite-prefix">Hi Chris,<br>
<br>
Thanks for your review and please see comments inline.<br>
(You might be reviewing the old patchset, please see the new one
if you have time.)<br>
<br>
On 08/02/2013 11:50 PM, Chris Larson wrote:<br>
</div>
<blockquote
cite="mid:CABcZAN=OC=DD4Sxvhf7P-SzmR8kObLceAm4TfKs90uiAqy9Uyg@mail.gmail.com"
type="cite">
<div dir="ltr">Greetings,
<div><br>
</div>
<div>Nice work on this branch. The bind based approach seems
clean. I do have some comments:</div>
<div><br>
</div>
<div>- I think the read-only-rootfs-hook stuff, at least the
bind mounting piece at a minimum, belongs in a separate
script in ${sbindir} rather than within the startup script
itself. This will ease adding support for r/o to systemd
images.</div>
</div>
</blockquote>
Although I still hold to the tenet we should make sysvinit and
systemd separate from each other whenever possible, this option is
acceptable if they do duplicate code and there's no side effect.<br>
But for now, I would suggest we leave it in the initscripts, and
when you add read-only-rootfs support for systemd , you could send
a patch to do the change.<br>
<br>
<blockquote
cite="mid:CABcZAN=OC=DD4Sxvhf7P-SzmR8kObLceAm4TfKs90uiAqy9Uyg@mail.gmail.com"
type="cite">
<div dir="ltr">
<div>- I think we should consider having an additional script
which unmounts the binds in reverse order, in case the binds
prevent unmounting of the partitions in which they live,
potentially causing an unclean shutdown.</div>
</div>
</blockquote>
<br>
I figure the bind mounts are essentially using volatile storage,
the data are lost between reboots after all. So could you please
give an example of unclean shutdown?<br>
<br>
Another problem with this unmounting mechanism is that it's hard
to determine whether a directory is bind mounted. Note that in the
read-only-rootfs-hook.sh script, a check has been added to see
whether the directory specified in config file is on a read-only
partition, and only if so, the directory is bind mounted with a
directory on volatile storage.<br>
<br>
<blockquote
cite="mid:CABcZAN=OC=DD4Sxvhf7P-SzmR8kObLceAm4TfKs90uiAqy9Uyg@mail.gmail.com"
type="cite">
<div dir="ltr">
<div>- The irda read only bits aren't necessary, it seems. The
startup script writes to /etc/sysconfig/ if its config file
doesn't already exist, as a mechanism of handling default
configuration. This is a terrible way to handle that, and
sysconfig is a redhatism, not the way we do things in our
distros. I have a commit I'll submit a patch for to resolve
this by fixing up the startup script.</div>
</div>
</blockquote>
This one has been fixed in the new patchset.<br>
<br>
<blockquote
cite="mid:CABcZAN=OC=DD4Sxvhf7P-SzmR8kObLceAm4TfKs90uiAqy9Uyg@mail.gmail.com"
type="cite">
<div dir="ltr">
<div>- I think we should enhance the copy part of the bind
script in two ways:</div>
<div> - cp from/. to/ -> this ensures it will copy
hidden directories as well, which the wildcard does not</div>
</div>
</blockquote>
Agree. Thanks for reminding me of this. I'll fix this.<br>
<blockquote
cite="mid:CABcZAN=OC=DD4Sxvhf7P-SzmR8kObLceAm4TfKs90uiAqy9Uyg@mail.gmail.com"
type="cite">
<div dir="ltr">
<div> - handle bind mounts of files as well as directories</div>
</div>
</blockquote>
Agree. Thanks. I'll send out a new patchset.<br>
(Of course I'll first wait for your comments :) )<br>
<br>
<blockquote
cite="mid:CABcZAN=OC=DD4Sxvhf7P-SzmR8kObLceAm4TfKs90uiAqy9Uyg@mail.gmail.com"
type="cite">
<div dir="ltr">
<div style="">- I also think the mount script should support
mount options, making the config files essentially a subset
of the filesystem table in fstab. There are cases where
there's value in using options with bind mounts, and it'd
make the scripts potentially more generally useful.</div>
</div>
</blockquote>
I know it might be useful, but I can't think of a use case. Could
you please give an example where the mount options are useful?<br>
<br>
<blockquote
cite="mid:CABcZAN=OC=DD4Sxvhf7P-SzmR8kObLceAm4TfKs90uiAqy9Uyg@mail.gmail.com"
type="cite">
<div dir="ltr">
<div>- I think we should split out the mount and unmount
scripts, along with default configuration, into a separate
recipe which inherits update-rc.d, and which gets pulled
into IMAGE_INSTALL by image.bbclass only when
read-only-rootfs is in IMAGE_FEATURES. This means we'd need
less conditional "if this is r/o" code in the scripts, as we
could rely on it being installed in r/o images, and not in
others.</div>
</div>
</blockquote>
The r/o check is needed anyway, it's not dedicated to the
read-only-rootfs-hook.sh script. Other init scripts might also
need this. That's why I made it a common function in the functions
script both in initscripts and lsbinitscripts.<br>
<br>
<blockquote
cite="mid:CABcZAN=OC=DD4Sxvhf7P-SzmR8kObLceAm4TfKs90uiAqy9Uyg@mail.gmail.com"
type="cite">
<div dir="ltr">
<div><br>
</div>
<div>As an aside, I looked into libmount's previous support of
fstab.d, it appears using util-linux's libmount-based mount
is supposed to support passing a directory to its
-T/--fstab= argument, which would let us offload the work to
it if we use a real mount binary, but it seems that this
directory argument support doesn't work anyway, and of
course by doing that we'd suck in additional deps, so maybe
it's for the best :)</div>
<div><br>
</div>
<div style="">Thoughts? I have prototype mount-binds and
unmount-binds shell scripts here which implement the
aforementioned suggestions, but I wanted to throw my
thoughts out to the list for comments before proceeding any
further.</div>
<div><br>
</div>
</div>
</blockquote>
It would be great if you could share them, and let's have some
further discussions.<br>
<br>
Best Regards,<br>
Chen Qi<br>
<br>
<blockquote
cite="mid:CABcZAN=OC=DD4Sxvhf7P-SzmR8kObLceAm4TfKs90uiAqy9Uyg@mail.gmail.com"
type="cite">
<div dir="ltr">
<div style="">Thanks,</div>
<div>-- <br>
Christopher Larson<br>
clarson at kergoth dot com<br>
Founder - BitBake, OpenEmbedded, OpenZaurus<br>
Maintainer - Tslib<br>
Senior Software Engineer, Mentor Graphics </div>
</div>
</blockquote>
<br>
<br>
<fieldset class="mimeAttachmentHeader"></fieldset>
<br>
<pre wrap="">_______________________________________________
Openembedded-core mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Openembedded-core@lists.openembedded.org">Openembedded-core@lists.openembedded.org</a>
<a class="moz-txt-link-freetext" href="http://lists.openembedded.org/mailman/listinfo/openembedded-core">http://lists.openembedded.org/mailman/listinfo/openembedded-core</a>
</pre>
</blockquote>
<br>
</body>
</html>