[BUG 2.5.X] Hollow processes

[BUG 2.5.X] Hollow processes

[Witek Kręcicki]

(Had the same problem on 2.5.9 with poldek, now it appeared on 2.5.15dj1
while building glibc-2.5.4 to rpm)
Process is hanging. It's impossible to stop it (even SAK is just clearing
console). It's impossible to check what the process is (trying to read
/proc/{pid}/{anyting} causes reading process to hang in the same way (so we
have now 2 hanging processes). Trying to use ps/lsof/killall/{anything that
is using /proc/{pid} causes this software hang in the same way. What could
it be?
WK

Re: [BUG 2.5.X] Hollow processes

[Manfred Spraul]

 > It's impossible to check what the process is (trying to read
 > /proc/{pid}/{anyting} causes reading process to hang in the
 > same way (so we have now 2 hanging processes).

Have you tried SysRQ+showTasks? That dumps the kernel stack. You can 
convert the numbers to names with ksymoops, or often klogd will convert 
them and the result is in /var/log/messages.

What exactly hangs?
Could you run

	strace ls /proc/1234
	strace cat /proc/1234/maps
	strace ls /proc/1234/fd -l

Which syscall hangs?
SMP or UP?

-- 

	Manfred

Re: [BUG 2.5.X] Hollow processes

[Witek Krecicki]

----- Original Message -----
From: "Manfred Spraul" <manfred@colorfullife.com>
> > It's impossible to check what the process is (trying to read
>  > /proc/{pid}/{anyting} causes reading process to hang in the
>  > same way (so we have now 2 hanging processes).
>
> Have you tried SysRQ+showTasks? That dumps the kernel stack. You can
> convert the numbers to names with ksymoops, or often klogd will convert
> them and the result is in /var/log/messages.
>
> What exactly hangs?
> Could you run
>
> strace ls /proc/1234
> strace cat /proc/1234/maps
> strace ls /proc/1234/fd -l
>
> Which syscall hangs?
> SMP or UP?
It's UP machine with UP kernel. I haven't tried it now strace'ing 'cat
/proc/1234/cmdline' (on 2.5.15) but as I remember from earlier kernels it
hanged on file reading processes. I'll try to make some additional tests as
soon as I'll be back from short vacation
ls'ing /proc/1234 doesn't hangs but ls -al does.
BTW: from my earlier post:
<cut>
mremap(0x407c5000, 8192, 12288, MREMAP_MAYMOVE) = 0x407c1000
brk(0x82e5000)                          = 0x82e5000
brk(0x82e6000)                          = 0x82e6000
old_mmap(NULL, 135168, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS,
-1, 0) = 0x407dc000
mremap(0x407ce000, 8192, 12288, MREMAP_MAYMOVE
</cut>
This is strace poldek on 2.5.11 kernel (behaviour is the same)
WK
P.S. now I can't reproduce it (on 2.5.15) as it's glibc compilation and it
takes far too long and has many subprocesses, but poldek issue on 2.5.11/12
was reproductible.
WK