[uml-devel] bug in COW? - Re: [uml-user] uml 2.6.0-test9 crash

["roland" <for_spam@gmx.de>]

Hi!
I made some further investigation.
i forgot to mention, that i`m using reiserfs and also use the copy-on-write(cow)
feature, uml has. the system-freeze i experience seems to have to do with copy-on-write.
as a test, i made a copy of my rootfs, made it available under /dev/ubd/2 inside the
uml and mounted that under /mnt. i did an extra fsck before - no inconsistency was
reported. then i did some stresstests again - worked fine!

after making the rootfs-clone a cow-filesystem (by adding ubd2=rootfsclone.cow,rootfsclone.img),
the freeze happened again when doing some heavy writing. so - am i allowed to assume, that
there is a bug in the copy-on-write code?
btw: could anybody recommend a general filesystem stresstest-tool which does the worst to
filesytems, you can do (besides formatting or overwriting)? :D
it seems, that this bug happens under certain circumstances only.  if i "dd" to an empty filesystem,
it doesn`t seem to freeze the uml at once. but it happend again after a while, when i did some
additionial copying of directory-trees while dd`ìng to disk in the background.

regards
roland

ps:
accidentally, i opened a cow-file with vi - and i saw that the path to the original readonly
filesystem image seems to be stored with full path information. would that mean: if i want to
relocate the fs-image with it`s accociated cow-file, i`m out of luck? is there any reason why
there are absolute, not relative paths inside?

pps:
> > The UML did the frist two dd's without problems, then it started to
> > hang.. i checked outside and saw the process switching states from
> > R to D and then back to R... so after a few secs the dd's did all
> > succeed and the uml is still running. So it didn't crash or so
> > but it also didn't play very well with that.
>
> Sounds as if it started to get blocked by disk I/O on the host.. ubd
> device I/O is a latency killer for UML as a blocking disk I/O operation
> will block the whole UML kernel until the host finishes.
yes - i think what hendrik says is right! this is just normal behaviour. the first 2 dd`s
ran without "problems" because the host just cached that i/o`s. then it began writing to disk
and this may have caused some delay you recognized as a "hang". in general - i/o scheduling of 2.4.x
kernels is not the very best. 2.6 kernel series has new i/o schedulers which are improved a LOT!




----- Original Message ----- 
From: "Henrik Nordstrom" <hno@marasystems.com>
To: "Sven 'Darkman' Michels" <sven@darkman.de>
Cc: <user-mode-linux-devel@lists.sourceforge.net>
Sent: Friday, November 28, 2003 12:50 AM
Subject: Re: [uml-devel] Re: [uml-user] uml 2.6.0-test9 crash


> On Thu, 27 Nov 2003, Sven 'Darkman' Michels wrote:
>
> > The UML did the frist two dd's without problems, then it started to
> > hang.. i checked outside and saw the process switching states from
> > R to D and then back to R... so after a few secs the dd's did all
> > succeed and the uml is still running. So it didn't crash or so
> > but it also didn't play very well with that.
>
> Sounds as if it started to get blocked by disk I/O on the host.. ubd
> device I/O is a latency killer for UML as a blocking disk I/O operation
> will block the whole UML kernel until the host finishes.
>
> > BTW: did you had any problems with building an uml kernel on 9.0?
>
> I have no problem with building UML on RH9, but I have a memory of some
> singnal trick which was needed at some point.. Make sure the UML is up to
> date.
>
> Regards
> Henrik
>


hi !´
i think , i`m able to crash my uml. :(

i did some stress-testing with my uml-2.6.0-test9 on 2.6.0-test9-skas host and i
unfortunately seem able to crash it very easily.

this was the way i recognized it:
first i executed
while true;do find /;done >/dev/zero &
several times and all worked well.
watched that with "top" inside the uml and found dozens of "finds" sharing the
cpu as expected(spending most of their time on system calls indeed).
on host, cpu usage of uml went to ~99% - all fine so far.

then i thought: mhhhh - ok - eat THIS:
while true;do dd if=/dev/zero of=test.dat bs=1k count=10000;done &

after executing this, my uml got stuck. no output, no response to input - even a
ping to the uml didn`t´give back a sign of life. the uml didn`t crash or panic - it
just gets stuck and unresponsive - uml process remains at 99% on the host.

at a second, third, fourth try, i isolated the problem a little bit:
one or two non looped "dd if=/dev/zero of=test.dat bs=1k count=10000" just are enough
to produce the same result.(under 2.6.0test9-skas-host and ALSO under 2.4.22-99-default
suse9 host, rootfs inside uml is suse9)

is anybody able to reproduce this?
if so - what do do further for analyzing?
i could make the uml downloadable and could send some output from strace (if i attach strace
to the remaining "99% cpu hog", it exits very quickly with a SIGALARM)

maybe an uml bug or even an issue for lkml?

regards
roland



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?  SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel