FAQ Search Today's Posts Mark Forums Read
» Video Reviews

» Linux Archive

Linux-archive is a website aiming to archive linux email lists and to make them easily accessible for linux users/developers.


» Sponsor

» Partners

» Sponsor

Go Back   Linux Archive > Debian > Debian Development

 
 
LinkBack Thread Tools
 
Old 04-17-2011, 09:23 AM
Roger Leigh
 
Default /run in experimental

Hi folks,

An update on /run:

sysvinit/initscripts 2.88dsf-13.4 is now available in experimental.
If you haven't tested it yet, now would be a good time before it
goes into unstable. Additionally, if you wish to transition your
packages to use /run, it is now OK to upload to experimental with
a dependency on initscripts (>= 2.88dsf-13.3). More information is
available here:

http://wiki.debian.org/ReleaseGoals/RunDirectory

There is just one known outstanding thing pending, which is switching
the use of absolute symlinks to relative (/var/run→/run will be
/var/run→../run for example). This doesn't affect the functionality of
the packages though. It's just a nicety for chroot environments, so
you don't get redirected back to the host /run if accessing the chroot
from the host.


At the moment, udev is unfortunately still badly broken when the /run
directory (introduced with the new initscripts) is present:

udev 167-1:
- works with /run absent
- completely broken with /run present but nothing mounted
(it's read-only when it tries to use it)
- functional with /run present and with a tmpfs mounted
udev 167-2:
- works with /run absent
- broken with /run present and with a tmpfs mounted
(no networking, others have other non-working hardware)

For as yet unstated reasons, the udev maintainer has chosen not to use
a versioned dependency on initscripts, which guarantees /run to be
present and functional, and hence allows it to be used reliably and
unconditionally. Instead, a number of different (and broken) checks
have been used which have resulted in #622893 (closed but still
present), #622309 and #623060. This is due to the checks making
misplaced assumptions in both the postinst and init scripts. The logic
is broken, and this results in a partially functional udev.

Note that for the networking, an ifdown and ifup of the device does
work around the issue. For other devices, probably just a modprobe.

If there's anyone familiar with udev who could take a look at it, I'd
be very grateful.


Regards,
Roger

--
.'`. Roger Leigh
: :' : Debian GNU/Linux http://people.debian.org/~rleigh/
`. `' Printing on GNU/Linux? http://gutenprint.sourceforge.net/
`- GPG Public Key: 0x25BFB848 Please GPG sign your mail.
 
Old 04-17-2011, 09:33 PM
Tollef Fog Heen
 
Default /run in experimental

]] Roger Leigh

| sysvinit/initscripts 2.88dsf-13.4 is now available in experimental.
| If you haven't tested it yet, now would be a good time before it
| goes into unstable. Additionally, if you wish to transition your
| packages to use /run, it is now OK to upload to experimental with
| a dependency on initscripts (>= 2.88dsf-13.3). More information is
| available here:

Yay, this is great news. I'll be updating systemd to depend on the new
version. Thanks a lot for all your work, it's much appreciated. :-)

Regards,
--
Tollef Fog Heen
UNIX is user friendly, it's just picky about who its friends are


--
To UNSUBSCRIBE, email to debian-devel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 87hb9w4l6z.fsf@qurzaw.varnish-software.com">http://lists.debian.org/87hb9w4l6z.fsf@qurzaw.varnish-software.com
 
Old 04-18-2011, 06:25 AM
Christian PERRIER
 
Default /run in experimental

Quoting Tollef Fog Heen (tfheen@err.no):

> Yay, this is great news. I'll be updating systemd to depend on the new
> version. Thanks a lot for all your work, it's much appreciated. :-)

Yes, that had to be said. I'm witnessing Roger's work from quite far,
I just read the wiki page about the transition yesterday and I have to
say that this transition and release goal are handled really well. And,
from the level of "noise" generated in -devel, blogs or anywhere else,
I think it is a very well accepted one.

Of course, there will be breakage (unstable has to deserve its name
from time to time) but I'm somehow confident it will be handled well.

So, I think Roger is deserving a couple of "thumbs up" here or
there. Here's one.
 
Old 04-20-2011, 04:58 AM
Thomas Hood
 
Default /run in experimental

Resolvconf 1.49, which makes use of /run instead of /lib/init/rw for
storing run-time data, has just been uploaded to experimental. Those
testing the experimental initscripts package (2.88dsf-13.4) are
invited to test this experimental resolvconf package along with it.

--
Thomas Hood
resolvconf maintainers <resolvconf-devel@lists.alioth.debian.org>


--
To UNSUBSCRIBE, email to debian-devel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 4DAE67EE.507@gmail.com">http://lists.debian.org/4DAE67EE.507@gmail.com
 
Old 04-20-2011, 10:37 AM
 
Default /run in experimental

On Apr 17, Roger Leigh <rleigh@debian.org> wrote:

> udev 167-2:
> - works with /run absent
> - broken with /run present and with a tmpfs mounted
> (no networking, others have other non-working hardware)
Unsurprisingly, it turned out that udev works fine in both cases.
But ifupdown breaks if /etc/network/run/ is a symlink to /dev/shm/ (and
possibly in other situations yet to be understood), I will open a bug
later if nobody beats me to it.

> For as yet unstated reasons, the udev maintainer has chosen not to use
> a versioned dependency on initscripts, which guarantees /run to be
> present and functional, and hence allows it to be used reliably and
> unconditionally.
For the records, the reason is that so far a dependency has not been
proven to be needed.

As usual, udev is more complex than people think it is (mostly because
we need to support upgrades in many different situations and people
want it to support annoying corner cases, let's make it required and it
will become much easier to deal with).

--
ciao,
Marco
 
Old 04-21-2011, 06:26 AM
"Andrew O. Shadoura"
 
Default /run in experimental

Hello,

On Wed, 20 Apr 2011 12:37:44 +0200
md@Linux.IT (Marco d'Itri) wrote:

> > udev 167-2:
> > - works with /run absent
> > - broken with /run present and with a tmpfs mounted
> > (no networking, others have other non-working hardware)
> Unsurprisingly, it turned out that udev works fine in both cases.
> But ifupdown breaks if /etc/network/run/ is a symlink to /dev/shm/
> (and possibly in other situations yet to be understood), I will open
> a bug later if nobody beats me to it.

I will Well, actually, no. I've ported ifupdown to use /run/network
instead of bunch of other variants already, so it's just a matter of
time and sponsorship.

--
WBR, Andrew
 
Old 04-27-2011, 04:40 PM
Roger Leigh
 
Default /run in experimental

On Sun, Apr 17, 2011 at 10:23:10AM +0100, Roger Leigh wrote:
> Hi folks,
>
> An update on /run:
>
> sysvinit/initscripts 2.88dsf-13.4 is now available in experimental.
> If you haven't tested it yet, now would be a good time before it
> goes into unstable. Additionally, if you wish to transition your
> packages to use /run, it is now OK to upload to experimental with
> a dependency on initscripts (>= 2.88dsf-13.3). More information is
> available here:
>
> http://wiki.debian.org/ReleaseGoals/RunDirectory
>
> There is just one known outstanding thing pending, which is switching
> the use of absolute symlinks to relative (/var/run→/run will be
> /var/run→../run for example). This doesn't affect the functionality of
> the packages though. It's just a nicety for chroot environments, so
> you don't get redirected back to the host /run if accessing the chroot
> from the host.

This issue has been addressed with the upload of
sysvinit/initscripts 2.88dsf-13.5 to experimental. Any further testing
of this would be appreciated. You might want the initramfs-tools
patch from #621803 as well if you're using LVM.

This version makes the early boot mount/mtab logic use a unified
single codepath to reduce the number of duplicated, almost
identical blocks of code which previously all had to be kept in
sync to avoid breakage. This also permits the mount options for
filesystems mounted in the initramfs (e.g. /dev, /run, /sys, /proc
etc.) to be set in /etc/fstab; the filesystems are remounted with
the options from /etc/fstab if already mounted during rcS. The
mount options for /run are also made stricter when possible.

/etc/default/tmpfs is deprecated in this version, since the same
settings (and more) can be stored in /etc/fstab directly. And
this also means that systemd can share the same settings. The
appropriate fstab entries are appended to /etc/fstab by the
initscripts postinst, which will supersede the /etc/default/tmpfs
settings. /etc/default/tmpfs isn't currently removed, but it may
be in a future version.


With these changes in place, initscripts has been tested under a
number of different systems and configurations and there are no
known issues remaining. In order to avoid udev causing breakage
when /run is introduced as in previous weeks, we're now waiting on
initramfs-tools to add /run (#621803) to allow for a seamless and
smooth transition. Once this is done, sysvinit/initscripts can
then go into unstable.


Regards,
Roger

--
.'`. Roger Leigh
: :' : Debian GNU/Linux http://people.debian.org/~rleigh/
`. `' Printing on GNU/Linux? http://gutenprint.sourceforge.net/
`- GPG Public Key: 0x25BFB848 Please GPG sign your mail.
 
Old 04-27-2011, 07:10 PM
Peter Samuelson
 
Default /run in experimental

[Roger Leigh]
> This also permits the mount options for filesystems mounted in the
> initramfs (e.g. /dev, /run, /sys, /proc etc.) to be set in
> /etc/fstab; the filesystems are remounted with the options from
> /etc/fstab if already mounted during rcS. The mount options for /run
> are also made stricter when possible.

I hope there are no traps for the unwary in remounting tmpfs
filesystems, like the one that hit me in ext3, Bug #520009. I put
"data=writeback" in the / entry of /etc/fstab. This caused the system
to become unbootable, and the initramfs-tools maintainer concluded that
it's my fault for being too stupid to notice that you have to duplicate
information between /etc/fstab and the boot loader config. He
WONTFIXed it because it is "much to big special case to start an fstab
parser on". ("fstab parser" meaning, I guess, getmntent(3).)

So, yeah, I hope tmpfs doesn't have any flags you're not allowed to
change in a remount.


--
To UNSUBSCRIBE, email to debian-devel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20110427191026.GC20906@p12n.org">http://lists.debian.org/20110427191026.GC20906@p12n.org
 
Old 04-27-2011, 08:02 PM
Roger Leigh
 
Default /run in experimental

On Wed, Apr 27, 2011 at 02:10:27PM -0500, Peter Samuelson wrote:
>
> [Roger Leigh]
> > This also permits the mount options for filesystems mounted in the
> > initramfs (e.g. /dev, /run, /sys, /proc etc.) to be set in
> > /etc/fstab; the filesystems are remounted with the options from
> > /etc/fstab if already mounted during rcS. The mount options for /run
> > are also made stricter when possible.
>
> I hope there are no traps for the unwary in remounting tmpfs
> filesystems, like the one that hit me in ext3, Bug #520009. I put
> "data=writeback" in the / entry of /etc/fstab. This caused the system
> to become unbootable, and the initramfs-tools maintainer concluded that
> it's my fault for being too stupid to notice that you have to duplicate
> information between /etc/fstab and the boot loader config. He
> WONTFIXed it because it is "much to big special case to start an fstab
> parser on". ("fstab parser" meaning, I guess, getmntent(3).)
>
> So, yeah, I hope tmpfs doesn't have any flags you're not allowed to
> change in a remount.

I've not seen any issues so far. Remounting with the various size options
and noexec/nodev etc. all appear to work.

However, the code which does the remounting is a general routine used for
all filesystems; it gets run for each filesystem mounted by mountkernfs/
mountdevsubfs, so it will include procfs and sysfs and devpts. Not seen
any issues so far though--they all appear to cope with remounting just
fine; I haven't tried any particularly exotic options though...

I had a look at #520009. Not too impressed with the response given that
parsing fstab, even in shell where there isn't a convenient interface
to getmntent(), is simple:

# Find a specific fstab entry
# $1=mountpoint
# $2=fstype (optional)
# returns 0 on success, 1 on failure (not found or no fstab)
read_fstab_entry () {
# Not found by default.
found=1

if [ -f /etc/fstab ]; then
exec 9<&0 </etc/fstab

while read MNT_FSNAME MNT_DIR MNT_TYPE MNT_OPTS MNT_FREQ MNT_PASS MNT_JUNK
do
case "$MNT_FSNAME" in
""|#*)
continue;
;;
esac
if [ "$MNT_DIR" = "$1" ]; then
if [ -n "$2" ]; then
[ "$MNT_TYPE" = "$2" ] || continue;
fi
# Found
found=0;
break
fi
done

exec 0<&9 9<&-
fi

return $found
}

So to get the mount options, you just
read_fstab_entry / && echo "Mount options for rootfs: $MNT_OPTS"

So grabbing the root mount options to feed to update-grub/
update-initramfs or whatever needs to have the options is hardly
difficult. If that's something which needs implementing, feel free
to use the above if it helps.


Regards,
Roger

--
.'`. Roger Leigh
: :' : Debian GNU/Linux http://people.debian.org/~rleigh/
`. `' Printing on GNU/Linux? http://gutenprint.sourceforge.net/
`- GPG Public Key: 0x25BFB848 Please GPG sign your mail.
 
Old 05-13-2011, 06:28 PM
Roger Leigh
 
Default /run in experimental

On Wed, Apr 27, 2011 at 05:40:13PM +0100, Roger Leigh wrote:
> On Sun, Apr 17, 2011 at 10:23:10AM +0100, Roger Leigh wrote:
> > Hi folks,
> >
> > An update on /run:
> >
> > sysvinit/initscripts 2.88dsf-13.4 is now available in experimental.
> > If you haven't tested it yet, now would be a good time before it
> > goes into unstable. Additionally, if you wish to transition your
> > packages to use /run, it is now OK to upload to experimental with
> > a dependency on initscripts (>= 2.88dsf-13.3). More information is
> > available here:
> >
> > http://wiki.debian.org/ReleaseGoals/RunDirectory
> >
> > There is just one known outstanding thing pending, which is switching
> > the use of absolute symlinks to relative (/var/run→/run will be
> > /var/run→../run for example). This doesn't affect the functionality of
> > the packages though. It's just a nicety for chroot environments, so
> > you don't get redirected back to the host /run if accessing the chroot
> > from the host.
>
> This issue has been addressed with the upload of
> sysvinit/initscripts 2.88dsf-13.5 to experimental.

With the upload of initramfs-tools (0.99) to unstable, we're now
ready to upload sysvinit 2.88dsf-13.6 to unstable, available at
http://people.debian.org/~rleigh/sysvinit_2.88dsf-13.6.dsc
if you would like to test it. I'm just waiting for confirmation
from the maintainers before I NMU it.


Regards,
Roger

--
.'`. Roger Leigh
: :' : Debian GNU/Linux http://people.debian.org/~rleigh/
`. `' Printing on GNU/Linux? http://gutenprint.sourceforge.net/
`- GPG Public Key: 0x25BFB848 Please GPG sign your mail.
 

Thread Tools




All times are GMT. The time now is 09:38 PM.

VBulletin, Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Content Relevant URLs by vBSEO ©2007, Crawlability, Inc.
Copyright 2007 - 2008, www.linux-archive.org