OS2 World.Com Forum

Subject  :  ext2-os2
Author  :  Kim
Date  :  09 Feb, 2005 on 13:09
Noticed this, kind of dead, fs project at freshmeat, ext2-os2 by Matthieu Willm that allows OS/2 to seamlessly access Linux ext2 formatted partition under OS/2 as normal driver letters.

Has any one picked up where Matthieu left this project. Not that I have an urge need to read linux partition from OS2; but there might be other people.

The ext2-os2 drivers is available from Hobbes.


Subject  :  Re:ext2-os2
Author  :  obiwan
Date  :  03 Jan, 2007 on 22:56
I've had this on my list a very long time, and at the rate I am going on higher-priority projects I will likely never get to it. I have used this driver under Warp 3 and found it very useful. Here is what I know.

The package comes with two components: a generic 32-bit ifs support driver, intended to allow others to create 32-bit ifs drivers; and the ext2 ifs, ported from the Linux kernel driver. The documentation even describes how to install a bootable OS/2 system on an ext2 partition.

Though I have not tried, the word is this package cannot work under eComStation. I would assume that also applies to Warp 4.52.

If I were to revive development on this, I would not use the old codebase, but only study it. First, to discover why it won't work in eCS, and secondly, to observe the methodology for converting code intended for Linux into an OS/2 IFS.

The answer to the first question would help to decide whether to repair the incompatibility in the 32-bit compatibility driver, or scrap it. Somewhere recently I saw a Linux-driver compatibility library for OS/2 but I forget where. Called lx32.dll or similar? Might be a better choice, though I don't know for sure as I haven't looked closely at it yet.

The second study should be done, abandoning the old driver, because it is based on very old Linux code. Not only can newer Linux code be considered more mature and robust, but it now has support for journalling (ext3 is just ext2 with journalling) as well as extended attributes! I would fully expect OS/2 to be able to utilize the EA's seamlessly, once properly supported in the driver.

Besides providing the ability to read Linux volumes, a current native ext2/3 driver would provide a viable fully open source filesystem for OS/2, and open the door to other filesystems supported by Linux. An intriguing idea is a port of the Linux HPFS driver. Could it be faster as a 32-bit driver without the cache limitation?


Subject  :  Re:ext2-os2
Author  :  RobertM
Date  :  04 Jan, 2007 on 03:40

obiwan (03 Jan, 2007 22:56):
...An intriguing idea is a port of the Linux HPFS driver. Could it be faster as a 32-bit driver without the cache limitation?

I can personally tell you it will be noticeably faster on any system that does a lot of disk access or that opens the same file repeatedly. The reason I can tell you this is I am running HPFS386 which is 32bit and doesnt have the cache limitation. My cache varies depending on the machine from 64MB to a few hundred.

I'd love to see an updated HPFS driver - especially one that also gets rid of the (artificially imposed) 64GB partition limit.

The only things a Linux port would be missing are the parts of HPFS386 that are "tied" directly into the network stack, etc...

BUT wait... there is one other thing a Linux port would be missing... the $1400 price tag on a package that is near impossible to obtain.

If the Linux port handles things like (not) fragmenting as well as the current HPFS/HPFS386 models and includes no cache limitation, and no artificial partition size limit.

Just my one cent,
-Robert


Subject  :  Re:ext2-os2
Author  :  El Vato
Date  :  04 Jan, 2007 on 10:43

obiwan (03 Jan, 2007 22:56):
I've had this on my list a very long time, and at the rate I am going on higher-priority projects I will likely never get to it. I have used this driver under Warp 3 and found it very useful. Here is what I know.

The package comes with two components: a generic 32-bit ifs support driver, intended to allow others to create 32-bit ifs drivers; and the ext2 ifs, ported from the Linux kernel driver. The documentation even describes how to install a bootable OS/2 system on an ext2 partition.

Though I have not tried, the word is this package cannot work under eComStation. I would assume that also applies to Warp 4.52.

If I were to revive development on this, I would not use the old codebase, but only study it. First, to discover why it won't work in eCS, [...]


OS2LVM.dmd ?


Subject  :  Re:ext2-os2
Author  :  obiwan
Date  :  05 Jan, 2007 on 01:26

El Vato (04 Jan, 2007 10:43):
OS2LVM.dmd ?

That would be my first guess too. What I mean, though, is to investigate what it is about ext2-os2 or the 32-bit driver that causes it to break when used with it.


Subject  :  Re:ext2-os2
Author  :  El Vato
Date  :  05 Jan, 2007 on 09:04

obiwan (05 Jan, 2007 01:26):

El Vato (04 Jan, 2007 10:43):
OS2LVM.dmd ?

That would be my first guess too. [...]


It is not a guess. If you had a Merlin with fixpak 15 applied, for instance, and you had the ext2 support and working, once you add the LVM support that we discussed elswhere, you will find that your now LVM enabled Merlin system becomes unbootable.

Moreover, in certain cases, in a multiboot environment that includes non-LVM enabled OS/2 such as Warp 3 -any flavour- and LVM enabled systems such as a retrofitted Merlin, and you decide to add a given Linux version, you may experience a similar behaviour with LVM enabled OS/2.

The fact is more prominent if you decide to boot your Warp 3 --it will boot!
Hence, when you put on your hacker suit and begin anlyzing the cause for which your more recent OS/2 will not boot, you zero it down to the driver that makes possible LVM.

But in the latter case, you solve the problem by deallocating a drive letter to Linux from the LVM enabled OS/2 Boot Manager. Note that it simply means, from the user perspective, that you simply hide your Linux Volume from OS/2 BUT if you previously had set the volume as bootable, it will retain that status. accordingly, your Linux volume will still live in harmony with your other OS/2.

Best regards !


Subject  :  Re:ext2-os2
Author  :  obiwan
Date  :  05 Jan, 2007 on 18:10

El Vato (05 Jan, 2007 09:04):

If you had a Merlin with fixpak 15 applied, for instance, and you had the ext2 support and working, once you add the LVM support that we discussed elswhere, you will find that your now LVM enabled Merlin system becomes unbootable.


That is what I would expect, guessing in my case as I have not tried, assuming you mean this system is on an ext2-formatted partition. Or you mean the presence of the driver in CONFIG.SYS halts everything anyway?



Moreover, in certain cases, in a multiboot environment that includes non-LVM enabled OS/2 such as Warp 3 -any flavour- and LVM enabled systems such as a retrofitted Merlin, and you decide to add a given Linux version, you may experience a similar behaviour with LVM enabled OS/2.

The fact is more prominent if you decide to boot your Warp 3 --it will boot!
Hence, when you put on your hacker suit and begin anlyzing the cause for which your more recent OS/2 will not boot, you zero it down to the driver that makes possible LVM.

But in the latter case, you solve the problem by deallocating a drive letter to Linux from the LVM enabled OS/2 Boot Manager. Note that it simply means, from the user perspective, that you simply hide your Linux Volume from OS/2 BUT if you previously had set the volume as bootable, it will retain that status. accordingly, your Linux volume will still live in harmony with your other OS/2.


I'm not sure I understand. I think what you are saying is that given a triple-boot system with an LVM OS/2 and a non-LVM OS/2, and Linux, with the ext2 driver installed on both OS/2's, the LVM OS/2 won't boot, but the non-LVM will, indicating that OS2LVM.DMD is the conflict.

Or are you saying that independent of the ext2 driver, LVM OS/2 always has problems if a drive letter is assigned to an ext2 partition?

Specifically how does it not boot? Where does it stop?

(I always have to be careful about looking at your screenshots at work, people will wonder what I'm doing!)


Subject  :  Re:ext2-os2
Author  :  obiwan
Date  :  06 Jan, 2007 on 02:21

obiwan (03 Jan, 2007 22:56):
Somewhere recently I saw a Linux-driver compatibility library for OS/2 but I forget where. Called lx32.dll or similar?

It was driving me nuts that I could not find this. Google is so useless for OS/2. It is called LXAPI32.SYS, by Stefan Milcke:

http://www.nord-com.net/s.milcke/lxapi_en.htm

Maybe not so applicable to filesystems though. ?


Subject  :  Re:ext2-os2
Author  :  obiwan
Date  :  06 Jan, 2007 on 07:46
Ok so I went ahead and gave this a try on a fresh install of eComStation.

I have a type 07 partition (OS/2 IFS) which I have formatted with ext2. That is, this is not a type 83 partition (Linux) but one just like OS/2 would create and use for HPFS. The ext2-os2 documentation says that should work.

I installed the ext2-os2 driver and the 32-bit support driver, and it loads. Not using the included ext2flt.flt, which is for assigning drive letters to otherwise unreadable type 83 partitions.

Now I go into LVM and assign a drive letter to this partition, and when I save, the program freezes. It can't be killed, so I had to reboot.

So I boot to the install CD, unassign the drive letter, and everything is ok again.

For my second test, I rem out the drivers and reboot, and assign a drive letter. System remains stable. I reboot, and it comes up fine. I open the Drives folder, and the drive appears, but when I try to open it, it says it is not formatted. I un-rem the drivers and reboot. The system freezes during boot.

So from this we can conclude that the problem is not with the partition type, or the ext2 format, but with this ext2-os2 driver set - when it tries to mount an ext2 partition, in LVM-enabled OS/2.

The source is all here and quite tidy. Whether I have all that is required to build it is another question...


Subject  :  Re:ext2-os2
Author  :  El Vato
Date  :  07 Jan, 2007 on 22:59

obiwan (05 Jan, 2007 18:45):
(I always have to be careful about looking at your screenshots at work, people will wonder what I'm doing!)

I did not consider them offensive. Sharapova is among the best athletes in the world. She is competitive, and perhaps more important, she is a dynamic entity who likes to live on the edge fine tunning her skills to maintain her top position. In other words, her beauty is not only superficial.

If we were to construct analogies, albeit with some argument, I would state that she is a model (at least for the current time period) for the OS/2 operating system to emulate. I will make explicit what I mean.

The current incarnation of OS/2 is also competitive --but in a passive way. There needs to be an influx of dynamism to make the OS/2 competitive in an active way --for instance a 64-bit Kernel development effort.

The current incarnation of OS/2 is attempting to show some superficial beauty by the use of themes, etc. . Notwithstanding, that beauty is frivolous a la WinXX-- as opposed to the strength and dynamism of Linux; the latter albeit somewhat ugly by the "standards" imposed by MS.

It will be until OS/2 achieves a competitive level that I will begin showing some galactic backgrounds with the e trademark. Notwithstanding, judging from the current pattern of proprietary OS/2 development, Maria will stay on my client machine for a while. I apologize if her sight offends your coworkers --are they conservative, by the way?


obiwan (06 Jan, 2007 07:46):
Ok so I went ahead and gave this a try on a fresh install of eComStation.
[...]

So from this we can conclude that the problem is not with the partition type, or the ext2 format, but with this ext2-os2 driver set - when it tries to mount an ext2 partition, in LVM-enabled OS/2.


On the other hand, and although it has been a few years that I experimented with this stuff, the mere addition to the OS/2 LVM-enabled Boot Manager of an ext3 file system containing Red Hat Fedora 2 (I seem to recall) caused LVM-enabled OS/2 not to be able to boot.

Note that I am setting aside the ext2 OS/2 driver issue and I am talking about possible Ext-based file sytems conflicting with OS2LVM.DMD.

File systems like RaiserFS ans XFS do not appear to exhibit those issues under discussion. But Red Hat has a default Ext3 file system and, for those dynamic OS/2ers who install several operating systems because they challenge the status quo, the following may be of help.

Warp 3 flavours, did not appear to mind the ext3 addition even though those (Warp 3) were also bootable from the same LVM-enabled Boot Manager.

Further, a Warp 3 flavour will be able to make modifications to the LVM-enabled Boot Manager with the proper additions of the LVM support* --EXCEPT OS2LVM.DMD. The addition of that driver crashes Warp 3 with an error relating to DOSCALL1.DLL (I seem to recall).

Elaborating further and noting that the OS2LVM.DMD is what, among other functions, makes possible on-the-fly resizing and/or partition allocation of JFS volumes. It also, as we all know, makes possible for an OS/2 that supports the driver to boot --as opposed to merely reside-- beyond the traditional boudaries of a non LVM enabled OS/2.

Consequently, although Warp 3 can manipulate the LVM enabled Boot Manager partitions, it can only live and boot in an extended partition below some 8-9GB.

Hence although many others have ditched Warp 3 since a long time ago, some of us who boot multiple operating systems continue to appreciate the oldie OS/2 3 --even though it insists on not living too far away from the hard disk starting point.

After installing the files referenced below -those with *.exe extension under \OS2 whereas those with the *.dll under \OS2\DLL directories-- simply executing from the command line:

LVM.EXE /SI:FS/SIZE: 120

will bring up LVM-enabled Boot Manager under OS/2 Warp 3 prompting you to make a partition/volume of 120MB. It will also permit you to modify existing LVM-enabled OS/2 unable to boot (if that were the case). The above switch is taken, as you might have seen in the documentation, from the support for enabling Merlin with LVM that we discussed elsewhere.

Comming back to Ext3 and elaborating over the relationship that LVM and JFS have, it would not be wild to speculate that the infusion of JFS technology into the ext3 file system implementation created a few quirks that need to be worked out.

The issue that I just mentioned, although probably not occurring in a uniform manner, is also "fixed" by deallocating a drive letter from LVM enabled Boot Manager.
[..]
...

*
lvm.dll
lvm.exe
lvm.msg
lvm.rep
lvmh.msg
vcu.exe
vcu.msg


Cheers !


Subject  :  Re:ext2-os2
Author  :  obiwan
Date  :  08 Jan, 2007 on 00:52

El Vato (07 Jan, 2007 23:41):
I apologize if her sight offends your coworkers

She isn't offensive to anyone. I like your analogy and motivation, although I think we're making progress. If the large picture of her on my screen showed only her face, one would still wonder what it is I am doing with my time.


On the other hand, and although it has been a few years that I experimented with this stuff, the mere presence of a an ext3 file system containing Red Hat Fedora 2 (I seem to recall) caused LVM-enabled OS/2 not to be able to boot.

Noting the relationship that LVM and JFS have, it would not be wild to speculate that the infusion of JFS technology into the ext3 file system implementation created a few quirks that need to be worked out.

The issue that I just mentioned, although probably not occurring in a uniform manner, is also "fixed" by deallocating a drive letter from LVM enabled Boot Manager.


Just tried it here. Formatted a partition for ext3 and gave it a drive letter. Boots fine, without the os2-ext2 driver. Changed the partition to type 83 (from 07) and formatted again, and it's still fine. Drive shows up, but opening it gives an error that it's unformatted.

I can't isolate the problem in the driver without being able to build it, and I don't have the DDK. But as I said in my first post, I have other things I need to finish first anyway.


Subject  :  Re:ext2-os2
Author  :  obiwan
Date  :  16 Jan, 2007 on 03:52
Curious about this Linux HPFS support, I ran some unscientific benchmarks under Linux to compare it with ext2. For more accurate results there is much that could have been done to isolate these tests from other factors, and better ways to test, but these Q&D tests are good enough for my purposes. I did, of course, reboot before any tests that read from data that had just been written or read, to avoid any caching benefit, and I didn't purposely start any other processes during the tests. The test filesystems were freshly-formatted partitions, HPFS formatted under OS/2, and ext2 formatted under Linux.


Write test
Copied about 1.4gb of mostly small files from an ext2 filesystem to the test filesystem.
HPFS: 7min 1s
EXT2: 6min 40s

Space Usage
Taken from output of 'df', in 1k "blocks"
HPFS: 1055451
EXT2: 1379444

Read Test
Time it took to execute 'tar -c . >/dev/null'
HPFS: 25s
EXT2: 34s

I was surprised at first that ext2 was notably faster at writing, with HPFS much faster at reading, but thinking about it that is probably what we should expect. More striking is the disk space savings of over 300mb with HPFS, one big reason we have grown to love HPFS over the years.

This says nothing about performance under OS/2, or as compared with JFS, but it answers my question, which was whether HPFS performs poorly in Linux, and the answer is no.

Some observations:

Linux has no ability to check or format HPFS partitions.

When comparing the files on the two test volumes, the case-insensitivity of HPFS revealed itself. Two files whose names differed only in case were present on the ext2 volume, but on the HPFS volume the second one that was copied overwrote the first, so only one was there - and it was wrong, because the case was preserved from the first file.

The documentation explains that extended attributes are respected, and Unix permissions and symlinks are implemented using them. I wonder whether they are compatible with those implemented in klibc. If not, that can be reconciled, and it really should be fixed in official Linux.

The documentation is also clear that HPFS386 features like ACL's are not supported.

There are some known minor issues pretty well described as well. Nothing I'm concerned about, but if the port is done and it becomes more widely used they should eventually be sorted out.

No mention of maximum size of files or volumes, or anti-fragmentation features.


Subject  :  Re:ext2-os2
Author  :  obiwan
Date  :  26 Mar, 2007 on 20:06
Discovered yesterday the Linux hpfs driver has a 1GB file size limit, and causes filesystem errors if you try to exceed it.

I wonder if it is still maintained. If not it could even be dropped eventually from the Linux kernel.


Subject  :  Re:ext2-os2
Author  :  Andrew
Date  :  29 Mar, 2007 on 01:34
Although it wouldn't be as useful as having a working version of ext2-os2, it might be easier to port LTOOLS (http://www.it.fht-esslingen.de/~zimmerma/software/ltools.html), a set of command-line programs for Windows and DOS that can read and write to ext2, ext3, and ReiserFS, to OS/2.

Powered by UltraBoard 2000 <www.ub2k.com>