reiser4 in 2.6.14 - lockups with mmapped files

reiser4 in 2.6.14 - lockups with mmapped files

[Fionn Behrens]

[-- Attachment #1: Type: text/plain, Size: 1538 bytes --]


Hi all,

you may remember my recent "story from userland". Stubborn as I am, I
decided to give 2.6.14 another shot for it. Following a recommendation
found in the above mentioned thread, I tried the patch from 
http://iphitus.loudas.com/archck.php this time, which includes the
latest reiser4 from mm.

Unfortunately I once more hit the wall and was forced to go back to
2.6.11 (including another --build-fs run to regain compatibility with
the old but working reiser4 drivers). I wish there was a mount option to
exclude the new extensions so I can test new kernels without losing half
a day for fscking when things get whacky.

This time suspend seemingly works without reiser4 processes freaking
out.
But now it looks like any time any mmapped file is synced to a reiser4
partition, the process who tries to do that will freeze forever. In
addition, any process trying to access the open mmapped file afterwards
will freeze forever as well. I created a screenshot of an strace running
while this happened. Find it at http://fionn.de/reiser4again.jpg

Of course, tests on files on other file systems than reiser4 did not
show this behaviour. I tested on two different Systems, one Pentium M
and one Athlon, same result.

I wish I knew what I am doing wrong. I'd bet you guys all have reiser4
up and running with 2.6.14 and no problems at all. Or what? 

best regards,
		Fionn
-- 
 Taking away civil rights to protect a free democracy is like taking off
 the tires of a car to protect it from flats. 

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: reiser4 in 2.6.14 - lockups with mmapped files

[Vladimir V. Saveliev]

Hello

Fionn Behrens wrote:
> Hi all,
> 
> you may remember my recent "story from userland". Stubborn as I am, I
> decided to give 2.6.14 another shot for it. Following a recommendation
> found in the above mentioned thread, I tried the patch from 
> http://iphitus.loudas.com/archck.php this time, which includes the
> latest reiser4 from mm.
> 
> Unfortunately I once more hit the wall and was forced to go back to
> 2.6.11 (including another --build-fs run to regain compatibility with
> the old but working reiser4 drivers). I wish there was a mount option to
> exclude the new extensions so I can test new kernels without losing half
> a day for fscking when things get whacky.
> 

Did fsck.reiser4 say you that you have to --build-fs?

> This time suspend seemingly works without reiser4 processes freaking
> out.
> But now it looks like any time any mmapped file is synced to a reiser4
> partition, the process who tries to do that will freeze forever. In

Can you please check what fs/reiser4/reiser4.h:REISER4_USE_EFLUSH is set to?
If it is 0, please try your test having it set to 1.

> addition, any process trying to access the open mmapped file afterwards
> will freeze forever as well. I created a screenshot of an strace running
> while this happened. Find it at http://fionn.de/reiser4again.jpg
> 

If the above does not help  - would you please set up either sereial console or
netconsole and catch sysrq-T output with its help?

> Of course, tests on files on other file systems than reiser4 did not
> show this behaviour. I tested on two different Systems, one Pentium M
> and one Athlon, same result.
> 
> I wish I knew what I am doing wrong. I'd bet you guys all have reiser4
> up and running with 2.6.14 and no problems at all. Or what? 
> 
> best regards,
> 		Fionn

Re: reiser4 in 2.6.14 - lockups with mmapped files

[Fionn Behrens]

[-- Attachment #1: Type: text/plain, Size: 1521 bytes --]

On Di, 2005-11-01 at 19:25 +0300, Vladimir V. Saveliev wrote:

> Fionn Behrens wrote:
> > Unfortunately I once more hit the wall and was forced to go back to
> > 2.6.11 (including another --build-fs run to regain compatibility with
> > the old but working reiser4 drivers). I wish there was a mount option to
> > exclude the new extensions so I can test new kernels without losing half
> > a day for fscking when things get whacky.

> Did fsck.reiser4 say you that you have to --build-fs?

Yes, after I went back to 2.6.11 I tried "--fix" first but it said
indeed that I have to "--build-fs" and so I did.

> > But now it looks like any time any mmapped file is synced to a reiser4
> > partition, the process who tries to do that will freeze forever. In
> 
> Can you please check what fs/reiser4/reiser4.h:REISER4_USE_EFLUSH is set to?
> If it is 0, please try your test having it set to 1.

It is set to 0. I changed it to 1. What is the supposed result of this
change?
I am somewhat reluctant to try it and risk yet another two hours of
fschecking. Maybe tomorrow night. If it doesnt work again, I'll get you
some screenshots of the fscheck, too.

> If the above does not help  - would you please set up either sereial console or
> netconsole and catch sysrq-T output with its help?

Would another "screenshot" do the trick as well? 

br,
	Fionn
-- 
I believe we have been called by history to lead the world.
                                                 G.W. Bush, 2002-03-01


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: reiser4 in 2.6.14 - lockups with mmapped files

[Sander]

Fionn Behrens wrote (ao):
> On Di, 2005-11-01 at 19:25 +0300, Vladimir V. Saveliev wrote:
> > Can you please check what fs/reiser4/reiser4.h:REISER4_USE_EFLUSH is
> > set to? If it is 0, please try your test having it set to 1.
> 
> It is set to 0. I changed it to 1. What is the supposed result of this
> change?
> I am somewhat reluctant to try it and risk yet another two hours of
> fschecking. Maybe tomorrow night. If it doesnt work again, I'll get
> you some screenshots of the fscheck, too.

Maybe you can create a small partition with a minimal OS on which you
boot new these test kernels?

-- 
Humilis IT Services and Solutions
http://www.humilis.net

Re: reiser4 in 2.6.14 - lockups with mmapped files

[Fionn Behrens]

[-- Attachment #1: Type: text/plain, Size: 706 bytes --]


In the mean time I mailed the archck patch maintainer and he readily
acknowledged that his reiser4 patch is flawed. He recommended that I
should apply the "*-2" patches from
ftp.namesys.com/pub/reiser4-for-2.6/2.6.14-rc4-mm1 . Unfortunately the
spinlock-2 patch wont apply cleanly but has seven failed hunks. Since I
have no idea how relevant these patches are I wanted to ask again
whether I should try to fix that patch or just leave it for now and try
the USE_EFLUSH(1) solution first? What would be the most effective
approach?

br,
	Fionn
-- 
 New laws should have better reasons than "recommended by lawyers"
 and "they handle it the same way in the US".       *** Lawrence Lessig


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: reiser4 in 2.6.14 - lockups with mmapped files

[Vladimir V. Saveliev]

Hello

Fionn Behrens wrote:
> In the mean time I mailed the archck patch maintainer and he readily
> acknowledged that his reiser4 patch is flawed. He recommended that I
> should apply the "*-2" patches from
> ftp.namesys.com/pub/reiser4-for-2.6/2.6.14-rc4-mm1 . Unfortunately the
> spinlock-2 patch wont apply cleanly but has seven failed hunks. 

You should apply
ftp://ftp.namesys.com/pub/reiser4-for-2.6/2.6.14-rc4-mm1/reiser4-fix-livelock-2.patch
and then
ftp://ftp.namesys.com/pub/reiser4-for-2.6/2.6.14-rc4-mm1/reiser4-spinlock-cleanup-2.patch
to clean 2.6.14-rc4-mm1.
I just tried that and everything worked fine.

> Since I
> have no idea how relevant these patches are I wanted to ask again
> whether I should try to fix that patch or just leave it for now and try
> the USE_EFLUSH(1) solution first? 
> What would be the most effective
> approach?
> 

You should use 2.6.14-rc5-mm1. If your filesystem load encounteres lockup - you
should try to set REISER4_USE_EFLUSH to 1.
If that does not help - please report the bug.

Re: reiser4 in 2.6.14 - lockups with mmapped files

[jp]

Hi,

I have to begin my tests of reiser4 / 2.6.14 ASAP, but I can't use a 
"release candidate" version of the kernel for my project ;)

When will the vanilla patch be released  ?

Best regards

JP

Re: reiser4 in 2.6.14 - lockups with mmapped files

[Vladimir V. Saveliev]

Hello

jp wrote:
> Hi,
> 
> I have to begin my tests of reiser4 / 2.6.14 ASAP, but I can't use a
> "release candidate" version of the kernel for my project ;)
> 
> When will the vanilla patch be released  ?
> 

yes. When
ftp://ftp.namesys.com/pub/reiser4-for-2.6/2.6.14-rc5-mm1/reiser4-update-for-2.6.14-rc5-mm1-1.gz
has more testing - there will be made a patch for 2.6.14. I think it will happen
at the beginning of next week

> Best regards
> 
> JP
> 
> 
> 

Re: reiser4 in 2.6.14 - lockups with mmapped files

[jp]

cool :)

Vladimir V. Saveliev wrote:

>Hello
>
>jp wrote:
>  
>
>>Hi,
>>
>>I have to begin my tests of reiser4 / 2.6.14 ASAP, but I can't use a
>>"release candidate" version of the kernel for my project ;)
>>
>>When will the vanilla patch be released  ?
>>
>>    
>>
>
>yes. When
>ftp://ftp.namesys.com/pub/reiser4-for-2.6/2.6.14-rc5-mm1/reiser4-update-for-2.6.14-rc5-mm1-1.gz
>has more testing - there will be made a patch for 2.6.14. I think it will happen
>at the beginning of next week
>
>  
>
>>Best regards
>>
>>JP
>>
>>
>>
>>    
>>
>
>
>
>  
>

SOLVED: reiser4 in 2.6.14 - lockups with mmapped files

[Fionn Behrens]

[-- Attachment #1: Type: text/plain, Size: 569 bytes --]

On Di, 2005-11-01 at 16:50 +0100, Fionn Behrens wrote:
> Hi all,
> 
> it looks like any time any mmapped file is synced to a reiser4
> partition, the process who tries to do that will freeze forever.

The archck patch maintainer released an update which no longer exhibits
the problem. After several hours of run time and some stress tests I
didnt experience a problem so far.

thanks for your advice and best regards,
					Fionn
-- 
 Taking away civil rights to protect a free democracy is like taking off
 the tires of a car to protect it from flats. 

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: SOLVED: reiser4 in 2.6.14 - lockups with mmapped files

[rvalles]

[-- Attachment #1: Type: text/plain, Size: 674 bytes --]

On Fri, Nov 04, 2005 at 10:12:34PM +0100, Fionn Behrens wrote:
> > it looks like any time any mmapped file is synced to a reiser4
> > partition, the process who tries to do that will freeze forever.
> The archck patch maintainer released an update which no longer exhibits
> the problem. After several hours of run time and some stress tests I
> didnt experience a problem so far.
> 					Fionn
If you are refering to 2.6.14-arckck3, I tried it and it doesn't fix a
thing. Even more, it triggers it more often, probably related to it's
agressivity against cache (swap prefetch, less tendency to swap, etc.).

I'm back with 2.6.12, waiting for a 2.6.14 patch I can work with.

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: SOLVED: reiser4 in 2.6.14 - lockups with mmapped files

[Raymond A. Meijer]

On Wed 9 Nov 2005 2:21, rvalles wrote:

> If you are refering to 2.6.14-arckck3, I tried it and it doesn't fix a
> thing. Even more, it triggers it more often, probably related to it's
> agressivity against cache (swap prefetch, less tendency to swap, etc.).

I'm using 2.6.14-archck3.

Could one of the symptoms be that apt/dpkg gets stuck if its files are on a 
Reiser4 partition?

After installing 2.6.14-archck3 and rebooting I found that aptitude etc. 
didn't work any longer and I couldn't even kill the processes when they got 
stuck?!

When I reformatted the repository's partition (/var/cache/apt) to ReiserFS the 
problem disappeared.

The funny thing is that I don't experience problems with the rest of my 
Reiser4 partitions:

/dev/mapper/system-tmp on /tmp type reiser4 (rw)
/dev/mapper/system-squid on /var/spool/squid type reiser4 (rw)
/dev/mapper/extra-music on /extra/music type reiser4 (rw)

The apt repository looks like this now:

/dev/mapper/system-cache on /var/cache type reiserfs (rw)


For what it's worth,

Ray

Re: reiser4 in 2.6.14 - lockups with mmapped files

[rvalles]

[-- Attachment #1: Type: text/plain, Size: 250 bytes --]

Still having the same problem, with 2.6.14.2 patched with 2.6.14-1
reiser4 patch.

It's easy to trigger it while editing a file with vim, and it does take
a hell of a long wait (while it hits the disk for a minute or so,
sometimes) for it to unlock.

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: reiser4 in 2.6.14 - lockups with mmapped files

[Artur Makówka]

rvalles napisa³(a):
> Still having the same problem, with 2.6.14.2 patched with 2.6.14-1
> reiser4 patch.
>
> It's easy to trigger it while editing a file with vim, and it does take
> a hell of a long wait (while it hits the disk for a minute or so,
> sometimes) for it to unlock.
>   
or if  you try to run lilo... or  dpkg/apt-get its also very easy to 
trigger it this way.

it hangs my server from time to time ... is there any progress in fixing 
that?

Re: reiser4 in 2.6.14 - lockups with mmapped files

[Vladimir V. Saveliev]

Hello

rvalles wrote:
> Still having the same problem, with 2.6.14.2 patched with 2.6.14-1
> reiser4 patch.
> 
> It's easy to trigger it while editing a file with vim, and it does take
> a hell of a long wait (while it hits the disk for a minute or so,
> sometimes) for it to unlock.

I notice this behaviour some times. But it does not appear to happen very often.
If you know the way to reproduce it on demand - would  you please share your
knowledge?