e2fsck sigcatcher breaks valgrind

e2fsck sigcatcher breaks valgrind

[Andreas Dilger]

I was reading Eric's recent commit message in "next" about how he used
e2fsprogs "./test_script --valgrind" to find the uninitialized memory and
thought I would try it myself.  To my dismay it was failing on my system with:

==2806==    at 0x3566A0F5BE: __libc_sigaction (in /lib64/libpthread-2.12.2.so)
==2806==    by 0x4321A3: sigcatcher_setup (sigcatcher.c:224)
==2806==    by 0x40F4DE: main (unix.c:1027)
==2806==  Address 0x7fefffe18 is on thread 1's stack
==2806==
==2806== Syscall param rt_sigaction(act->sa_mask) points to uninitialised byte(s)

After a bunch of testing as to why this system was not working correctly,
it turns out that the problem is in the later commit 9b3018a82e843d.
Reverting that commit fixes the problem and allows --valgrind tests to pass.



I also noticed (for Lukas), "./test_script --valgrind i_e2image" fails with:

i_e2image: Create/convert raw and qcow2 disk images: ./i_e2image/script: line 2: test: too many arguments

That is due to:

   if test -x $E2IMAGE; then

expanding to:

   if test -x valgrind -q --sim-hints=lax-ioctls ../misc/e2image; then

It probably needs some parameter manipulation to check only the last parameter.

Cheers, Andreas
--
Andreas Dilger 
Principal Engineer
Whamcloud, Inc.

Re: e2fsck sigcatcher breaks valgrind

[Ted Ts'o]

On Fri, Sep 02, 2011 at 11:38:09AM -0600, Andreas Dilger wrote:
> I was reading Eric's recent commit message in "next" about how he used
> e2fsprogs "./test_script --valgrind" to find the uninitialized memory and
> thought I would try it myself.  To my dismay it was failing on my system with:
> 
> ==2806==    at 0x3566A0F5BE: __libc_sigaction (in /lib64/libpthread-2.12.2.so)
> ==2806==    by 0x4321A3: sigcatcher_setup (sigcatcher.c:224)
> ==2806==    by 0x40F4DE: main (unix.c:1027)
> ==2806==  Address 0x7fefffe18 is on thread 1's stack
> ==2806==
> ==2806== Syscall param rt_sigaction(act->sa_mask) points to uninitialised byte(s)
> 
> After a bunch of testing as to why this system was not working correctly,
> it turns out that the problem is in the later commit 9b3018a82e843d.
> Reverting that commit fixes the problem and allows --valgrind tests to pass.

Oops, thanks for the bug report, I'll fix it.

For context, the reason why I added this was I had a nearly impossible
to debug crash in e2fsck that happened extremely rarely, and only in
my virtualized xfstests environment.  I found the bug and fixed it in
commit a4aff9ca5.

> 
> I also noticed (for Lukas), "./test_script --valgrind i_e2image" fails with:
> 
> i_e2image: Create/convert raw and qcow2 disk images: ./i_e2image/script: line 2: test: too many arguments
> 
> That is due to:
> 
>    if test -x $E2IMAGE; then
> 
> expanding to:
> 
>    if test -x valgrind -q --sim-hints=lax-ioctls ../misc/e2image; then

Thanks, I'll look at fixing this as well.

	     	     	    	 - Ted