Skip to contentSkip to conversation

"mount -o loop" lockup issue

DISCUSSION David E. Weekly 12 participants 17 messages Mar 27, 2001

lore
Share & tools
lore
#0 lore raw

On Linux 2.4.2, running a "mount -o loop" on a file properly created with "dd if=/dev/zero of=/path/to/my/file.img count=1024" seems to decide to freeze up my shell (not my system). An strace showed the lockup happening at the actual system "mount()" call, which never returns.

Since mount() is in glibc, it might be relevant to note that I'm running Mandrake's glibc 2.1.3-16mdk. I compiled the kernel with a gcc of 2.95.3 [1991030] (although oddly enough this binary seems to have come with the gcc-2.95.2 RPM and installed itself as /usr/bin/gcc-2.95.2) and binutils 2.10.0.24-4mdk.

I'm very sorry to post to this list, but several people independantly told me that there was a loopback mountpoint deadlocking issue with 2.4.2 and that I should check here. Of course, this could be a completely retarded system configuration issue, in which case please shut me up and I'll go away quietly. But if it is an issue with a known resolution I'd love to hear it - I wasn't able to find resolution on the web or with several rather knowledgeable people.

-david weekly [[email protected]]

#1 lore raw
Quoted 10 lines · David E. Weekly

On Linux 2.4.2, running a "mount -o loop" on a file properly created with "dd if=/dev/zero of=/path/to/my/file.img count=1024" seems to decide to freeze up my shell (not my system). An strace showed the lockup happening at the actual system "mount()" call, which never returns.

Since mount() is in glibc, it might be relevant to note that I'm running Mandrake's glibc 2.1.3-16mdk. I compiled the kernel with a gcc of 2.95.3 [1991030] (although oddly enough this binary seems to have come with the gcc-2.95.2 RPM and installed itself as /usr/bin/gcc-2.95.2) and binutils 2.10.0.24-4mdk.

I also experience this problem (using a floppy disk image created by dd if=/dev/fd0 of=floppy.img bs=1024, and then mount -o loop floppy.img /mnt/floppy ) with a different version
of glibc (RedHat's 2.1.92-5 rpm) and binutils (binutils-2.10.0.18-1). Loop is compiled into the kernel.

Once the mount command was executed, my load average shot up to a steady 1.0 on an idle system, and remained there until I rebooted. top et. al. showed no cpu utilization by the frozen mount.

#2 lore raw

It's a bug in Linux 2.4.2, fixed in later versions. Regression/quality control testing would
have caught this, but the developers usually just break things and wait for people to complain
as their "Regression" testers.

Quoted 27 lines · Jason Madden
Quoted 10 lines · David E. Weekly

On Linux 2.4.2, running a "mount -o loop" on a file properly created with "dd if=/dev/zero of=/path/to/my/file.img count=1024" seems to decide to freeze up my shell (not my system). An strace showed the lockup happening at the actual system "mount()" call, which never returns.

Since mount() is in glibc, it might be relevant to note that I'm running Mandrake's glibc 2.1.3-16mdk. I compiled the kernel with a gcc of 2.95.3 [1991030] (although oddly enough this binary seems to have come with the gcc-2.95.2 RPM and installed itself as /usr/bin/gcc-2.95.2) and binutils 2.10.0.24-4mdk.

I also experience this problem (using a floppy disk image created by dd if=/dev/fd0 of=floppy.img bs=1024, and then mount -o loop floppy.img /mnt/floppy ) with a different version
of glibc (RedHat's 2.1.92-5 rpm) and binutils (binutils-2.10.0.18-1). Loop is compiled into the kernel.

Once the mount command was executed, my load average shot up to a steady 1.0 on an idle system, and remained there until I rebooted. top et. al. showed no cpu utilization by the frozen mount.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#3 lore raw
Quoted 5 lines

It's a bug in Linux 2.4.2, fixed in later versions. Regression/quality control testing would
have caught this, but the developers usually just break things and wait for people to complain
as their "Regression" testers.

Hardly. We knew it was broken since well before 2.4.0. It just got a little interesting to fix.

#4 lore raw
Quoted 8 lines · Alan Cox
Quoted 5 lines

It's a bug in Linux 2.4.2, fixed in later versions. Regression/quality control testing would
have caught this, but the developers usually just break things and wait for people to complain
as their "Regression" testers.

Hardly. We knew it was broken since well before 2.4.0. It just got a little interesting to fix.

And this is described in what release notes? It worked just fine on Red Hat 7.0's 2.4 kernel.... oh wait, I see that they fixed it before they released it.

#5 lore raw
In reply to Alan Cox
Quoted 2 lines · David Konerding

And this is described in what release notes? It worked just fine on Red Hat 7.0's 2.4 kernel.... oh wait, I see that they fixed it before they released it.

And hmm..gee .. did they bother contributing back the code?

Signature · 8 lines
=====================================================================
Mohammad A. Haque                              http://www.haque.net/ 
                                               [email protected]

  "Alcohol and calculus don't mix.             Project Lead
   Don't drink and derive." --Unknown          http://wm.themes.org/
                                               [email protected]
=====================================================================
#6 lore raw
In reply to Alan Cox
Quoted 6 lines · Mohammad A. Haque
Quoted 2 lines · David Konerding

And this is described in what release notes? It worked just fine on Red Hat 7.0's 2.4 kernel.... oh wait, I see that they fixed it before they released it.

And hmm..gee .. did they bother contributing back the code?

Based on their track record that's a silly question.

jjs

#7 lore raw
Quoted 10 lines · J Sloan
Quoted 6 lines · Mohammad A. Haque
Quoted 2 lines · David Konerding

And this is described in what release notes? It worked just fine on Red Hat 7.0's 2.4 kernel.... oh wait, I see that they fixed it before they released it.

And hmm..gee .. did they bother contributing back the code?

Based on their track record that's a silly question.

Especially since patches in question had been written by Jens Axboe (who has nothing to RH) and announced (many times) on l-k.

I've fixed several races in Jens' patch and fed them back to him. His patch
+ these fixes were the only loop-related patches in RH tree[1]. Until fixes got merged into Jens' loop-6 which, in turn, was merged into -ac and into the main tree, that is.

I don't give a flying fsck through the rolling doughnut for "their" track record (whatever "their" means), but I'm somewhat partial to mine. Care to grep through l-k archives, check your facts and STFU?
Al

[1] there's also changeloop patch - adds an ioctl for switching the underlying file under opened /dev/loop; API is ugly and thing has so limited use that IMO it should die. Completely unrelated to the problems in question, anyway.

#8 lore raw
In reply to David Konerding
Quoted 4 lines · David Konerding

It's a bug in Linux 2.4.2, fixed in later versions. Regression/quality control testing would have caught this, but the developers usually just break things and wait for people to complain as their "Regression" testers.

As said before, we're interested in people willing to do regression tests on the kernel. Unfortunately, not all that many testers have stepped forward and not all that many artificial tests are being run.

Good thing we still have the beta-testers to catch these things, while running the kernel in real-world scenarios... ;)

regards,

Rik
--
Virtual memory is like a game you can't win; However, without VM there's truly nothing to lose...

		http://www.surriel.com/

http://www.conectiva.com/ http://distro.conectiva.com.br/

#9 lore raw
Quoted 10 lines · Rik van Riel
Quoted 4 lines · David Konerding

It's a bug in Linux 2.4.2, fixed in later versions. Regression/quality control testing would have caught this, but the developers usually just break things and wait for people to complain as their "Regression" testers.

As said before, we're interested in people willing to do regression tests on the kernel. Unfortunately, not all that many testers have stepped forward and not all that many artificial tests are being run.

No, the point is that the linux developers should regression test their code BEFORE
releasing it to the public as a version like "2.4.2". When I see a version like "2.4.2", I have an expectation that all the stupid little problems (like mounting loopback filesystem) have already been found.

It's even worse that these are obvious, simple bugs (like the "NFS doesn't work over reiserfs
because somebody changed the VFS layer and didn't fix any filesystems but ext2" that I reported a while ago) which would have been caught by a little testing.

Now, don't even get me started on how the developers are fixing every legitimate bug found by CHECKER when they refused to put a debugger into the kernel "because a good programmer finds their bug by studying the code"-- well, obviously, you didn't find a lot of bugs by studying the code.

I've been using Linux for something like 6-7 years now, quite faithfully. I've been very impressed with
many of its facilities, and the improvements to the kernel (which I've compiled since 0.99) have been astounding. But the attitude that "many eyes make all bugs shallow" and "let the users test the code for us" just don't hold up. For the former, clearly, many eyes didn't find a lot of basically obvious bugs, for the latter, it's just impolite.

#11 lore raw
In reply to David Konerding
Quoted 5 lines

It's even worse that these are obvious, simple bugs (like the "NFS doesn't work over reiserfs
because somebody changed the VFS layer and didn't fix any filesystems but ext2" that I reported a while ago) which would have been caught by a little testing.

Again people knew about this. It was a chosen decision that 2.4.x shouldnt support NFS over reiserfs. If you want an extensively QA'd, signed off kernel tree then wait for vendors to release one.

Alan

#12 lore raw

Kernel QA

In reply to Rik van Riel
Quoted 5 lines · David Konerding

No, the point is that the linux developers should regression test their code BEFORE
releasing it to the public as a version like "2.4.2". When I see a version like "2.4.2", I have an expectation that all the stupid little problems (like mounting loopback filesystem) have already been found.

You bring up a good point. We call the even branches the stable branches and we do other things that promote the idea that people should be able to download a 2.even.X kernel, install it on their machine, and expect it to work. I think we need to back away from this idea. It seems to me that the real (perhaps not the intended) function of kernel releases is keeping kernel developers in sync. Promoting the idea that they are thought to be suitable for production use just gets us in trouble.

Instead I think we need to encourage people who want to use Linux, rather than develop it, to use kernels from a distribution. After all, the distributors put a lot of effort into doing QA and putting together a compatable system, we should leverage that. We need to ensure that people know that when they install the latest kernel from Linus, they are the QA.

Please note that I am not trying to say that we should not try and make the kernels we release as good as possible. It certainly makes things a lot better for everyone if bugs dont get introduced by new kernel versions. I do think we need to be more explicit about exactly what people should and should not be able to expect from a "Linus kernel".

Thanks,

Jim

#13 lore raw

Re: Kernel QA

I disagree, 2.4.x is "stable" and as such we need as many people to use the kernels to see whats wrong with them. 2.4 *DOES* Work, I've had very small problems (ok, the thread hanging issue was a big one) but other then that It's been solid.

It depends on the hardware.

Shawn.

Quoted 38 lines · James Lewis Nance
Quoted 5 lines · David Konerding

No, the point is that the linux developers should regression test their code BEFORE
releasing it to the public as a version like "2.4.2". When I see a version like "2.4.2", I have an expectation that all the stupid little problems (like mounting loopback filesystem) have already been found.

You bring up a good point. We call the even branches the stable branches and we do other things that promote the idea that people should be able to download a 2.even.X kernel, install it on their machine, and expect it to work. I think we need to back away from this idea. It seems to me that the real (perhaps not the intended) function of kernel releases is keeping kernel developers in sync. Promoting the idea that they are thought to be suitable for production use just gets us in trouble.

Instead I think we need to encourage people who want to use Linux, rather than develop it, to use kernels from a distribution. After all, the distributors put a lot of effort into doing QA and putting together a compatable system, we should leverage that. We need to ensure that people know that when they install the latest kernel from Linus, they are the QA.

Please note that I am not trying to say that we should not try and make the kernels we release as good as possible. It certainly makes things a lot better for everyone if bugs dont get introduced by new kernel versions. I do think we need to be more explicit about exactly what people should and should not be able to expect from a "Linus kernel".

Thanks,

Jim
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#14 lore raw

Re: Kernel QA

In reply to James Lewis Nance
Quoted 2 lines · James Lewis Nance

Instead I think we need to encourage people who want to use Linux, rather than develop it, to use kernels from a distribution.

I hope that's not the opinion of all the kernel developers - where does that leave distributions like slackware, debian, and the rest that don't have the time or resources to modify the kernel themselves? Every kernel release that is meant to keep developers "in sync", as you say, should be a 2.4.x-prex release, and the stable releases should actually be stable. If this means slowing the release schedule, so be it. You are proposing to release unfinished, buggy and unstable code and let the distributions pick up your slack. It sounds like something Microsoft would do.

#15 lore raw
In reply to Jason Madden
Quoted 7 lines · Jason Madden
Quoted 4 lines · David E. Weekly

On Linux 2.4.2, running a "mount -o loop" on a file properly created with "dd if=/dev/zero of=/path/to/my/file.img count=1024" seems to decide to freeze up my shell (not my system). An strace showed the lockup happening at the actual system "mount()" call, which never returns.

....

Quoted 5 lines

I also experience this problem (using a floppy disk image created by dd if=/dev/fd0 of=floppy.img bs=1024, and then mount -o loop floppy.img /mnt/floppy ) with a different version
of glibc (RedHat's 2.1.92-5 rpm) and binutils (binutils-2.10.0.18-1). Loop is compiled into the kernel.

Follow this thread -->
<http://marc.theaimsgroup.com/?l=linux-kernel&m=98289750805700&w=2>

Latest loop patch is available at <ftp://ftp.kernel.org/pub/linux/kernel/people/axboe/patches/2.4.3-pre1/>

Signature · 8 lines
=====================================================================
Mohammad A. Haque                              http://www.haque.net/ 
                                               [email protected]

  "Alcohol and calculus don't mix.             Project Lead
   Don't drink and derive." --Unknown          http://wm.themes.org/
                                               [email protected]
=====================================================================
#16 lore raw
In reply to Jason Madden

Good day, all,

Quoted 18 lines · Jason Madden
Quoted 10 lines · David E. Weekly

On Linux 2.4.2, running a "mount -o loop" on a file properly created with "dd if=/dev/zero of=/path/to/my/file.img count=1024" seems to decide to freeze up my shell (not my system). An strace showed the lockup happening at the actual system "mount()" call, which never returns.

Since mount() is in glibc, it might be relevant to note that I'm running Mandrake's glibc 2.1.3-16mdk. I compiled the kernel with a gcc of 2.95.3 [1991030] (although oddly enough this binary seems to have come with the gcc-2.95.2 RPM and installed itself as /usr/bin/gcc-2.95.2) and binutils 2.10.0.24-4mdk.

I also experience this problem (using a floppy disk image created by dd if=/dev/fd0 of=floppy.img bs=1024, and then mount -o loop floppy.img /mnt/floppy ) with a different version
of glibc (RedHat's 2.1.92-5 rpm) and binutils (binutils-2.10.0.18-1). Loop is compiled into the kernel.

Quoted 3 lines

Once the mount command was executed, my load average shot up to a steady 1.0 on an idle system, and remained there until I rebooted. top et. al. showed no cpu utilization by the frozen mount.

Jens Axboe, along with a number of other people, has put in a lot
of time coming up with a fix for the loop mount lockups. You can either get his patch directly from ftp://ftp.kernel.org/pub/linux/kernel/people/axboe/patches/
or simply use the most recent 2.4.2-ac patch (from
ftp://ftp.kernel.org/pub/linux/kernel/people/alan/
) to get this updated loop device code. I'm certain Jens would
like to hear from you if you find any problems with the updated code.
Cheers,
- Bill

---------------------------------------------------------------------------
The day Microsoft makes something that doesn't suck is
probably the day they start making vacuum cleaners.
-- Ernst Jan Plugge
(Courtesy of Christian Vogel <[email protected]>)
--------------------------------------------------------------------------
William Stearns ([email protected]). Mason, Buildkernel, named2hosts,
and ipfwadm2ipchains are at: http://www.pobox.com/~wstearns
LinuxMonth; articles for Linux Enthusiasts! http://www.linuxmonth.com
--------------------------------------------------------------------------