opensync-users

[Opensync-users] opensync on x86_64

[Roland S. <ro...@xi...>]

Hi,

i tried opensync on an amd64 system. It dies with a segfault=20
in osync_member_initialize when the initialize function of the
plugin is called.
I tried this with gcc-3.2 -3.3 and -4.0 and always got the same=20
result. I don't know how to track this. It seems to me, that the=20
stack gets somehow overwritten during the call to fs_initialize,=20
but this is only a guess.

btw. compiling with -Werror fails because casting void* to int=20
gives a warning on x86_64.
opensync_debug.c:65: warning: cast from pointer to integer of different siz=
e


This is a run of msynctool --sync filefile in gdb

roland@alderaan:~/opensync$ gdb ./bin/msynctool
GNU gdb 6.3-debian
Copyright 2004 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and you ar=
e
welcome to change it and/or distribute copies of it under certain condition=
s.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB.  Type "show warranty" for details.
This GDB was configured as "x86_64-linux"...Using host libthread_db library=
 "/lib/libthread_db.so.1".

(gdb) break opensync_member.c:982
No source file named opensync_member.c.
Make breakpoint pending on future shared library load? (y or [n]) y

Breakpoint 1 (opensync_member.c:982) pending.
(gdb) run --sync filefile
Starting program: /home/roland/opensync/bin/msynctool --sync filefile
[Thread debugging using libthread_db enabled]
[New Thread 46912515451632 (LWP 32346)]
Breakpoint 2 at 0x2aaaaad56f94: file opensync_member.c, line 982.
Pending breakpoint "opensync_member.c:982" resolved
Synchronizing group "filefile"
The previous synchronization was unlean. Slow-syncing
[Switching to Thread 46912515451632 (LWP 32346)]

Breakpoint 2, osync_member_initialize (member=3D0x515750, error=3D0x7fffffc=
fb538)
    at opensync_member.c:982
982             if (!(member->plugindata =3D functions.initialize(member, e=
rror))) {
(gdb) print *member
$1 =3D {id =3D 1, configdir =3D 0x5157e0 "/home/roland/.opensync/group1/1",
  configdata =3D 0x0, configsize =3D 0, plugin =3D 0x50a2e0,
  memberfunctions =3D 0x5156f0, group =3D 0x506740, enginedata =3D 0x512a70=
,
  plugindata =3D 0x0, format_sinks =3D 0x50a170, objtype_sinks =3D 0x50a140=
,
  pluginname =3D 0x517760 "file-sync", accepted_objtypes =3D 0x0, filters =
=3D 0x0,
  extension =3D 0x0, loop =3D 0x512b80}
(gdb) s
fs_initialize (member=3D0x7fffffcfb538, error=3D0x7fffffcfb538) at file_syn=
c.c:66
66      {
(gdb) print *member
$2 =3D {id =3D 0, configdir =3D 0x0, configdata =3D 0x0, configsize =3D 0, =
plugin =3D 0x0,
  memberfunctions =3D 0x0, group =3D 0x0, enginedata =3D 0x0, plugindata =
=3D 0x0,
  format_sinks =3D 0x0, objtype_sinks =3D 0x0, pluginname =3D 0x0,
  accepted_objtypes =3D 0x0, filters =3D 0x0, extension =3D 0x0, loop =3D 0=
x0}
(gdb) finish
Run till exit from #0  fs_initialize (member=3D0x7fffffcfb538,
    error=3D0x7fffffcfb538) at file_sync.c:66
0x00002aaaaad56f9c in osync_member_initialize (member=3D0x0,
    error=3D0x7fffffcfb538) at opensync_member.c:982
982             if (!(member->plugindata =3D functions.initialize(member, e=
rror))) {
Value returned is $3 =3D (void *) 0x517670
(gdb) print *member
Cannot access memory at address 0x0
(gdb) n

Program received signal SIGSEGV, Segmentation fault.
0x00002aaaaad56f9f in osync_member_initialize (member=3D0x0,
    error=3D0x7fffffcfb538) at opensync_member.c:982
982             if (!(member->plugindata =3D functions.initialize(member, e=
rror))) {
(gdb)


Roland

Re: [Opensync-users] opensync on x86_64

[Stefan B. <be...@su...>]

Moin,

On Aug 16, 05 23:26:47 +0200, Roland Stoll wrote:
> Hi,
> 
> i tried opensync on an amd64 system. It dies with a segfault 
> in osync_member_initialize when the initialize function of the
> plugin is called.
> I tried this with gcc-3.2 -3.3 and -4.0 and always got the same 
> result. I don't know how to track this. It seems to me, that the 
> stack gets somehow overwritten during the call to fs_initialize, 
> but this is only a guess.
> 
> btw. compiling with -Werror fails because casting void* to int 
> gives a warning on x86_64.
> opensync_debug.c:65: warning: cast from pointer to integer of different size

I've appended a patch, please try to apply that and see if it's better
then.
(I currently don't have time to check the latest version, so the patch
might not be sufficient)

        Stefan


-- 
Stefan Behlert

Re: [Opensync-users] opensync on x86_64

[Roland S. <ro...@xi...>]

On Wed, August 17, 2005 10:38, Stefan Behlert said:
> Moin,
>
> On Aug 16, 05 23:26:47 +0200, Roland Stoll wrote:
>> Hi,
>>
>> i tried opensync on an amd64 system. It dies with a segfault
>> in osync_member_initialize when the initialize function of the
>> plugin is called.
>> I tried this with gcc-3.2 -3.3 and -4.0 and always got the same
>> result. I don't know how to track this. It seems to me, that the
>> stack gets somehow overwritten during the call to fs_initialize,
>> but this is only a guess.
>>
>> btw. compiling with -Werror fails because casting void* to int
>> gives a warning on x86_64.
>> opensync_debug.c:65: warning: cast from pointer to integer of different
>> size
>
> I've appended a patch, please try to apply that and see if it's better
> then.
> (I currently don't have time to check the latest version, so the patch
> might not be sufficient)
>

Thank you. This solved the compilation issue.

running msynctool --sync on a group with a file-sync still segfaults. I
found out, that it works when file_sync.c is compiled without
optimizations.

Is this a gcc bug maybe?


Roland.

Re: [Opensync-users] opensync on x86_64

[Armin B. <arm...@de...>]

Roland Stoll wrote:
> On Wed, August 17, 2005 10:38, Stefan Behlert said:
> 
>>Moin,
>>
>>On Aug 16, 05 23:26:47 +0200, Roland Stoll wrote:
>>
>>>Hi,
>>>
>>>i tried opensync on an amd64 system. It dies with a segfault
>>>in osync_member_initialize when the initialize function of the
>>>plugin is called.
>>>I tried this with gcc-3.2 -3.3 and -4.0 and always got the same
>>>result. I don't know how to track this. It seems to me, that the
>>>stack gets somehow overwritten during the call to fs_initialize,
>>>but this is only a guess.
>>>
>>>btw. compiling with -Werror fails because casting void* to int
>>>gives a warning on x86_64.
>>>opensync_debug.c:65: warning: cast from pointer to integer of different
>>>size
>>
>>I've appended a patch, please try to apply that and see if it's better
>>then.
>>(I currently don't have time to check the latest version, so the patch
>>might not be sufficient)
>>
> 
> 
> Thank you. This solved the compilation issue.
> 
> running msynctool --sync on a group with a file-sync still segfaults. I
> found out, that it works when file_sync.c is compiled without
> optimizations.
> 
> Is this a gcc bug maybe?
> 

Maybe its just another assumption i made about the size of variables
that are not true any more on x64. Can you please show me a backtrace of
this new segfault?

Armin

> 
> Roland.
> 
> 
> 
> -------------------------------------------------------
> SF.Net email is Sponsored by the Better Software Conference & EXPO
> September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
> Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
> Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf
> _______________________________________________
> Opensync-users mailing list
> Ope...@li...
> https://lists.sourceforge.net/lists/listinfo/opensync-users

Re: [Opensync-users] opensync on x86_64

[Roland S. <ro...@xi...>]

On Wed, August 17, 2005 11:21, Armin Bauer said:
>
>
> Roland Stoll wrote:
>> running msynctool --sync on a group with a file-sync still segfaults. I
>> found out, that it works when file_sync.c is compiled without
>> optimizations.
>>
>> Is this a gcc bug maybe?
>>
>
> Maybe its just another assumption i made about the size of variables
> that are not true any more on x64. Can you please show me a backtrace of
> this new segfault?

i have attached a backtrace.
After fs_initialize() returns, the value of 'member' in
osync_member_initialize() is changed. This happens only if file_sync.c is
compiled with -O2 and osync_member_get_config() is called in
fs_initialize().

Roland

Re: [Opensync-users] opensync on x86_64

[Roland S. <ro...@xi...>]

On Wed, August 17, 2005 11:21, Armin Bauer said:
>
>
> Roland Stoll wrote:
>> On Wed, August 17, 2005 10:38, Stefan Behlert said:
>>
>>>Moin,
>>>
>>>On Aug 16, 05 23:26:47 +0200, Roland Stoll wrote:
>>>
>>>>Hi,
>>>>
>>>>i tried opensync on an amd64 system. It dies with a segfault
>>>>in osync_member_initialize when the initialize function of the
>>>>plugin is called.
>>>>I tried this with gcc-3.2 -3.3 and -4.0 and always got the same
>>>>result. I don't know how to track this. It seems to me, that the
>>>>stack gets somehow overwritten during the call to fs_initialize,
>>>>but this is only a guess.
>>>>
>>>>btw. compiling with -Werror fails because casting void* to int
>>>>gives a warning on x86_64.
>>>>opensync_debug.c:65: warning: cast from pointer to integer of different
>>>>size
>>>
>>>I've appended a patch, please try to apply that and see if it's better
>>>then.
>>>(I currently don't have time to check the latest version, so the patch
>>>might not be sufficient)
>>>
>>
>>
>> Thank you. This solved the compilation issue.
>>
>> running msynctool --sync on a group with a file-sync still segfaults. I
>> found out, that it works when file_sync.c is compiled without
>> optimizations.
>>
>> Is this a gcc bug maybe?
>>
>
> Maybe its just another assumption i made about the size of variables
> that are not true any more on x64. Can you please show me a backtrace of
> this new segfault?

You're right. attached is a patch that solves this problem.
gsize is 8 byte on x86_64 and casting (int*) to (gsize*) in 
osync_file_read() overwrites the stack in fs_initialize which leads to the
wrong value of member in osync_member_initialize().
I didn't want to change the interface, so i used a temp. variable.

Now i can try to sync something :-)

Roland.