Bug 603316

Hmm, OOo crashes heavily on openSUSE-11.3-m6-x86_64 when the OpenOffice_org-LanguageTool package is installed. I do not see the problem when I remove it.

Thomas, does it help you to remove the OpenOffice_org-LanguageTool package?

Hmm, Donn's problems were somehow related to SELinux and too permissive access rights to ~/.ooo3.

Thomas, do you have something similar by chance?

After removing the languagetool rpms OOo saving works again.

I doubt that the problem is caused by SELinux because it also happens if SELinux is disable. But to be really sure we need hard facts.

The LanguageTool makes OOo very unstable on openSUSE-11.3-m6. I do not see this problem in SLED11. I am going to debug it.

Thanks a lot for feedback.

BTW: the packages junit4 and jgroups have not been rebuilt in Factory for 50 days, see https://build.opensuse.org/stage/project/status?project=openSUSE%3AFactory&filter_devel=All+Packages&ignore_pending=true&limit_to_fails=false&limit_to_fails=true&include_versions=false&commit=Filter+results

The build failed because of "Too many open files" => it is similar to this problem with too many open descriptors.

I wonder if these problems are really related and the updated openJDK has crazy regression.


Note that the LanguageTool was not installed by default before openSUSE-11.3-m6, so the problem was hidden. It might have been there the whole 50 days.

Also I do not see it on SLED11 with OOo and LanguageTool built from the same sources => I tend to thing that the problem is related to any other package (openJDK).

Michael, does it trigger any bell?

Hi Petr, I already know about the "Too many open files" problems, but was not able to reproduce locally - the local build (using chroot, or directly in unpacked source directory) works well. I never heard about some similar problem on upstream.

Adrian asked me to use the kvm based osc build, but I failed to establish that - it ends on weird qemu error. So don't have an idea what's going wrong.

I'll try to reproduce it (when an another 2GB zypper up will be finished).

Hi all, so I wrote a simple reproducer - not sure why the new openjdk opens a lot of /proc/PID/maps, but the previous one did not do that so extensively.

Thomas: can you attach the strace file please?

I was not able to reproduce it using 

$ rpm -q OpenOffice_org OpenOffice_org-LanguageTool java-1_6_0-openjdk
OpenOffice_org-3.2.0.99.3-1.1.x86_64
OpenOffice_org-LanguageTool-1.0.0-4.6.noarch
java-1_6_0-openjdk-1.6.0.0_b17-3.2.x86_64

Save and open of an arbitrary document work without a crash.

Created attachment 361376 [details]
strace from my crash

I opened writer, wrote few letters. OOo freezed for a bit and crashed.

Also this crash is caused by too many open files:

--- cut ---
the 32748 16:15:19.272428 open("/proc/self/maps", O_RDONLY) = -1 EMFILE (Too many open files)
32748 16:15:19.272587 mmap(0x7fa029e5b000, 12288, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_NORESERVE, -1, 0) = 0x7fa029e5b000
32748 16:15:19.272697 rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
32748 16:15:19.272935 sched_getaffinity(32748, 32, {1, 0, 0, 0}) = 32
32748 16:15:19.273044 sched_getaffinity(32748, 32, {1, 0, 0, 0}) = 32
32748 16:15:19.273146 gettid()          = 32748
32748 16:15:19.273234 rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
32748 16:15:19.273324 rt_sigprocmask(SIG_UNBLOCK, [HUP INT ILL BUS FPE SEGV USR2 TERM], NULL, 8) = 0
32748 16:15:19.273420 rt_sigprocmask(SIG_BLOCK, [QUIT], NULL, 8) = 0
32748 16:15:19.273591 open("/proc/self/maps", O_RDONLY) = -1 EMFILE (Too many open files)
32748 16:15:19.273704 mmap(0x7fa029e5b000, 12288, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x7fa029e5b000
32748 16:15:19.273802 mprotect(0x7fa029e5b000, 12288, PROT_NONE) = 0
32748 16:15:19.274327 --- SIGSEGV (Segmentation fault) @ 0 (0) ---
--- cut ---

So from the information I already have - new openjdk opens /proc/self/maps and not close it, which seems the root of all those problems.

Simple grep tells me, there are only two places which opened it

static bool find_vma(address addr, address* vma_low, address* vma_high) in hotspot/src/os/linux/vm/os_linux.cpp [1]

static bool read_lib_info(struct ps_prochandle* ph) in hotspot/agent/src/os/linux/ps_proc.c [2]

however those functions are 1) simple 2) call fclose after fopen properly.

[1] http://hg.openjdk.java.net/jdk6/jdk6-gate/hotspot/file/587f774a3e70/src/os/linux/vm/os_linux.cpp
[2] http://hg.openjdk.java.net/jdk6/jdk6-gate/hotspot/file/587f774a3e70/agent/src/os/linux/ps_proc.c

But it seems the bug is in JVM.

Petr: I see segfault in your strace. Does it means it crashing? And if so, can I have a stack trace?

Created attachment 361609 [details]
thomas strace file gzipped

Created attachment 361695 [details]
backtrace from the crash on hope.suse.cz

The stack of the main thread is somehow malformed. I hope that the stack from the other threads might be more useful.

So I see the same problem in your strace outputs as I have during a run of my simple test program. The problem is that JVM opens /proc/self/maps, but don't close it. Not sure why ...

So looking deeply on all patches are applied against openjdk sources I found the most probably place - the stack protector patch [1] requested by OpenOffice guys in bug#589021 seems to be wrong.

It opens the /proc/self/maps, but it returns false without a fclose call, so it's the most probably place, where it causes a problems. Lesson - never ever trust upstream and check their patches.

Fixed by sr#39886

[1] http://cr.openjdk.java.net/~aph/6929067-jdk7-webrev-4/

ok, the right sr is 39894, the previous fix was not complete, thanks Petr for a testing

*** Bug 604094 has been marked as a duplicate of this bug. ***

*** Bug 616991 has been marked as a duplicate of this bug. ***

JFYI, the fixed openJDK has appeared in the update channel for openSUSE 11.1 , 11.2, and 11.3. See the bug #623905 and the bug #601243.

*** Bug 628761 has been marked as a duplicate of this bug. ***

This is an autogenerated message for OBS integration:
This bug (603316) was mentioned in
https://build.opensuse.org/request/show/39894 Factory / java-1_6_0-openjdk