Not until recently I noticed that marble (http://edu.kde.org/marble/)
and nautilus was randomly crashing (sigsegv). Initially i thought this
was due to a glibc upgrade but even after downgrading the crashes
occurred, with occasional kernel oops on both 2.6.32.x and 2.6.33
kernels. I discovered that both these applications have a high chance
of crashing during mmap calls for example, marble will crash right
after this call:
if ((src = (unsigned char*) mmap (0, filelength, PROT_READ,
MAP_SHARED, fd, 0))
== (unsigned char*) (caddr_t) -1)
qDebug() << "mmap error for input";
where fd is a file on an aufs filesystem. The crash characteristic is
a segfault after using src without warning, despite the mmap call
never fails.
This is where i remembered that about the time when i noticed these
crashes, there was a change in aufs regarding mmap. The offending
commit happens to be:
commit 4b70e6f04d4292d8b5ce6cd7ac7371e68eab9175
Author: J. R. Okajima <[email protected]>
Date: Sat Jan 2 12:21:03 2010 +0900
aufs: bugfix, unlock mmap_sem temporary using BKL
The aufs rwsem should be acquired before mm->mmap_sem. But mm->mmap_sem
is acquired before ->mmap(). aufs_mmap() has to release it temporary
with using BKL.
I don't think it is a good approach, but I don't know other way.
Signed-off-by: J. R. Okajima <[email protected]>
Now, I am not good with kernel programming so i do not quite
understand the purpose of this commit, however, if i revert this in
the latest aufs (aufs 2-standalone.tree-20100222), no more crashes in
marble or nautilus, everything very stable.
I am using aufs as a live filesystem on tmpfs rw with squashfs.
uname -a:
Linux darkstar 2.6.33 #6 SMP PREEMPT Sun Feb 28 13:59:53 EST 2010 i686
AMD Phenom(tm) II X3 720 Processor AuthenticAMD GNU/Linu
If anyone could look into this that would be great. Aufs is such a
good project I use it religiously.
Thank you for your attention.
James.
------------------------------------------------------------------------------
Download Intel® Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
