So upgraded systems don't get the benefits of certain changes to the installer's defaults, or defaults in programs used by the installer.
I first thought of this when I noticed the change in the default tuning for ext2 partitions created by mke2fs. Notably, dir_index and filetype are turned on by default now; this was not always the case. You used to need to put /proc and /sys in your fstab, but it's unneeded and silly now. There are probably other changes like this floating around. For instance, if we switch to mounting a tmpfs over /tmp in the installer, that will probably end up being another one.
Some of these can have substantial impact -- particularly dir_index, which has never been mentioned in release notes. I think they're all appropriate topics for release notes: "things to do after rebooting" is probably the correct category. Perhaps people could comment on other things like this which they've noticed and we could get them into the next release notes, including anything which wasn't covered on previous major upgrades?
Nathanael Nerode <nero...@fastmail.fm> writes: > So upgraded systems don't get the benefits of certain changes to the installer's > defaults, or defaults in programs used by the installer.
> I first thought of this when I noticed the change in the default tuning for > ext2 partitions created by mke2fs. Notably, dir_index and filetype are turned > on by default now; this was not always the case. You used to need to put /proc > and /sys in your fstab, but it's unneeded and silly now. There are probably > other changes like this floating around. For instance, if we switch to mounting a > tmpfs over /tmp in the installer, that will probably end up being another one.
For dir_index you have to take the FS offline, tune2fs and fsck it or you totaly corrupt it.
As for /tmp being tmpfs that should be optional and the default should consider the amount of ram available. All low-mem cases will most likely not want tmpfs.
MfG Goswin
-- To UNSUBSCRIBE, email to debian-devel-REQU...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Anyobdy (Marco d'Itri?) can add some comments regarding the udev/dbus/hal (.../pcmcia-cs/hotplug/usbwhatever/...) situation to that page? Or does the normal upgrade from a default sarge installation (2.4 based!) to a default etch installation provide everything already through dependencies?
cheers -- vbi
-- Jesus is my duct tape. -- Robert Lindsay on alt.religion.kibology
On Sat, 05 Aug 2006 15:05:10 +0200, Adrian von Bidder wrote: > On Thursday 03 August 2006 06:24, Nathanael Nerode wrote: >> So upgraded systems don't get the benefits of certain changes to the >> installer's defaults, or defaults in programs used by the installer.
I added a link to a blog entry by Erich Schuberts [1], whose instructions I followed to turn on dir_index on my Debian. Maybe a more structured HowTo will be better, but in the meantime I think it could be useful.
On Thu, Aug 03, 2006 at 12:24:09AM -0400, Nathanael Nerode <nero...@fastmail.fm> wrote: > So upgraded systems don't get the benefits of certain changes to the installer's > defaults, or defaults in programs used by the installer.
> I first thought of this when I noticed the change in the default tuning for > ext2 partitions created by mke2fs. Notably, dir_index and filetype are turned > on by default now; this was not always the case.
I don't know about the installer, but all filesystems I created with mke2fs recently also have resize_inode, which isn't even in the tune2fs manpage.
Mike
-- To UNSUBSCRIBE, email to debian-devel-REQU...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
> I don't know about the installer, but all filesystems I created with > mke2fs recently also have resize_inode, which isn't even in the > tune2fs manpage.
The default was recently changed in /etc/mke2fs.conf. It make life with LVM a lot easier. :)
Friendly, -- Petter Reinholdtsen
-- To UNSUBSCRIBE, email to debian-devel-REQU...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
On Fri, Aug 04, 2006 at 03:02:34PM +0200, Goswin von Brederlow wrote: > For dir_index you have to take the FS offline, tune2fs and fsck it or > you totaly corrupt it.
Actually, that's not true. It's perfectly safe to run tune2fs on a mounted volume to enable the dir_index feature. All directories from that point on which grow beyond a single disk block will use the hashed-tree optimization. "e2fsck -fD" run off-line is only needed in order to upgrade existing large directories to use dir_tree.
(Of course, you can also do something like "cd ~/Maildir; mkdir cur.new; mv cur/* cur.new; rmdir cur; mv cur.new cur" if you want, as long as you are confident you don't screw up or confuse a currently running mail client.)
- Ted
-- To UNSUBSCRIBE, email to debian-devel-REQU...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
On Sat, Aug 05, 2006 at 07:58:20PM +0200, Mike Hommey wrote: > I don't know about the installer, but all filesystems I created with > mke2fs recently also have resize_inode, which isn't even in the tune2fs > manpage.
It's not in the tune2fs man page because e2fsprogs doesn't currently support adding the resize_inode feature after the fact; this is one you have to do at mke2fs time. (There is an ext2prepare program that will do this, but the code has been too scary for me to just integrate into e2fsprogs with a complete rewrite, and I haven't had time to do this.)
The e2fsprogs-udeb package does include the mke2fs.conf file, so if the installer is using the latest e2fsprogs-udeb, it should be creating filesystems with the dir_index and resize_inode features.
- Ted
-- To UNSUBSCRIBE, email to debian-devel-REQU...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Theodore Tso <ty...@mit.edu> writes: > On Fri, Aug 04, 2006 at 03:02:34PM +0200, Goswin von Brederlow wrote: >> For dir_index you have to take the FS offline, tune2fs and fsck it or >> you totaly corrupt it.
> Actually, that's not true. It's perfectly safe to run tune2fs on a > mounted volume to enable the dir_index feature. All directories from > that point on which grow beyond a single disk block will use the > hashed-tree optimization. "e2fsck -fD" run off-line is only needed in > order to upgrade existing large directories to use dir_tree.
Did they fix that? When I first looked into dir_index it was said that it would corrupt the directories since it would search the old linear dirs via hash and insert new entries by hash into linear dirs.
That would totaly confuse the filesystem for obvious reasons.
> (Of course, you can also do something like "cd ~/Maildir; mkdir > cur.new; mv cur/* cur.new; rmdir cur; mv cur.new cur" if you want, as > long as you are confident you don't screw up or confuse a currently > running mail client.)
> - Ted
MfG Goswin
-- To UNSUBSCRIBE, email to debian-devel-REQU...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
On Tue, Aug 15, 2006 at 10:53:14AM +0200, Goswin von Brederlow wrote: > Did they fix that? When I first looked into dir_index it was said that > it would corrupt the directories since it would search the old linear > dirs via hash and insert new entries by hash into linear dirs.
Who said that? I was the one who merged the dir_index feature into Linux 2.5/2.6, and that was never the case; it was always safe to enable dir_index on the fly, by desgin.
- Ted
-- To UNSUBSCRIBE, email to debian-devel-REQU...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Theodore Tso <ty...@mit.edu> writes: > On Tue, Aug 15, 2006 at 10:53:14AM +0200, Goswin von Brederlow wrote: >> Did they fix that? When I first looked into dir_index it was said that >> it would corrupt the directories since it would search the old linear >> dirs via hash and insert new entries by hash into linear dirs.
> Who said that? I was the one who merged the dir_index feature into > Linux 2.5/2.6, and that was never the case; it was always safe to > enable dir_index on the fly, by desgin.
> - Ted
Must have been misinformation then. Sorry.
MfG Goswin
-- To UNSUBSCRIBE, email to debian-devel-REQU...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
On Thu, Aug 03, 2006 at 12:24:09AM -0400, Nathanael Nerode wrote: > So upgraded systems don't get the benefits of certain changes to the installer's > defaults, or defaults in programs used by the installer.
> [...] Perhaps people could comment on other things like this which > they've noticed and we could get them into the next release notes, > including anything which wasn't covered on previous major upgrades?
* Added /usr/local/etc, required by FHS (Closes: #383493). [ As usual, please note that this will only affect new installs ]. ... ----------------------------------------------------------------------
> The e2fsprogs-udeb package does include the mke2fs.conf file, so if > the installer is using the latest e2fsprogs-udeb, it should be > creating filesystems with the dir_index and resize_inode features.
Soon the etch installer is using the latest mkfs.ext3 to create the file system. debian-installer beta3 use libparted + tune2fs -j, and this fail to enable both these features.
I really happy to know that d-i in sid now creates online re-sizable file systems.
Friendly, -- Petter Reinholdtsen
-- To UNSUBSCRIBE, email to debian-devel-REQU...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
|
