[Bug 16666] New: wine segfaults on launch

http://bugs.winehq.org/show_bug.cgi?id=16666

Summary: wine segfaults on launch
           Product: Wine
           Version: 1.1.11
          Platform: PC
        OS/Version: OpenBSD
            Status: NEW
          Keywords: patch, source
          Severity: critical
          Priority: P2
         Component: build-env
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]

Created an attachment (id=18290)
 --> (http://bugs.winehq.org/attachment.cgi?id=18290)
core dump

Now for the biggest bug :-)

Wine segfaults on launch of any program. Even 'wine --version' crashes.

I've got this patch in place (from openbsd's port):
diff --git a/loader/pthread.c b/loader/pthread.c
index 4c0c892..e7f6479 100644
--- a/loader/pthread.c
+++ b/loader/pthread.c
@@ -96,6 +96,12 @@ static void init_thread( struct wine_pthread_thread_info
*info )
         /* if base is too large assume it's the top of the stack instead */
         if ((char *)info->stack_base > &dummy)
             info->stack_base = (char *)info->stack_base - info->stack_size;
+#elif defined(__OpenBSD__)
+        stack_t stack;
+        if (pthread_stackseg_np(pthread_self(), &stack) != 0)
+            abort ();
+        info->stack_base = (char *)stack.ss_sp - stack.ss_size;
+        info->stack_size = stack.ss_size;
 #else
         /* assume that the stack allocation is page aligned */
         char dummy;
@@ -163,6 +169,8 @@ static void init_current_teb( struct
wine_pthread_thread_info *info )
     info->pid = getpid();
 #ifdef __sun
     info->tid = pthread_self();  /* this should return the lwp id on solaris
*/
+#elif defined(__OpenBSD__)
+    info->tid = pthread_self();
 #elif defined(__APPLE__)
     info->tid = mach_thread_self();
 #elif defined(__FreeBSD__)

Though, it still segfaults without it.

wine-pthread.core attached

[Bug 16666] wine segfaults on launch

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #1 from Austin English   2009-01-02 18:41:22 ---
Created an attachment (id=18415)
 --> (http://bugs.winehq.org/attachment.cgi?id=18415)
new patch

This fixes 'wine --version', but anything more complex is still broken (even
notepad). There are several more bugs to work through, but wanted to upload for
anyone to test :-).

[Bug 16666] wine segfaults on launch

http://bugs.winehq.org/show_bug.cgi?id=16666

Tijl Coosemans  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #2 from Tijl Coosemans   2009-01-03 15:02:25 ---
Removing the --section-start flag is probably not correct. It's there to make
sure the wine executable is loaded where it doesn't interfere with the Windows
program.

On FreeBSD there used to be a crash when Wine failed to allocate
user_shared_data at 0x7ffe0000 and tried to initialise it
(dlls/ntdll/thread.c:thread_init()), so you might want to check if that address
is available. On FreeBSD loader/freebsd.c was needed to assure that.

Also, if you can't turn off the randomised mmap somehow, you'll have to port
loader/preloader.c I'm afraid.

[Bug 16666] wine segfaults on launch

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #3 from Austin English   2009-01-03 23:04:14 ---
http://www.openbsd.org/cgi-bin/man.cgi?query=mmap&apropos=0&sektion=...

There's a flag, MAP_FIXED, that prohibits mmap from selecting a different
address than specified.

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |blocker
            Summary|wine segfaults on launch    |wine segfaults on launch on
                   |                            |OpenBSD

--- Comment #4 from Austin English   2009-03-31 01:07:32 ---
Without any threading/loader patches, in git wine (1.1.18 + a couple days), on
OpenBSD out of the box, I get:

"Dynamic loader failure: malloc."

when attempting to run any wine process (even --version).

Still working on compiler bugs at the moment, and this is on a donated server,
so I'm working a bit slowly.

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #5 from Tijl Coosemans   2009-04-02 12:14:29 ---
In loader/main.c there's a pre_exec() function defined for FreeBSD. Maybe you
need that on OpenBSD as well.

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #6 from Austin English   2009-04-02 13:48:16 ---
(In reply to comment #5)
> In loader/main.c there's a pre_exec() function defined for FreeBSD. Maybe you
> need that on OpenBSD as well.

I'm sure it needs something, but it probably needs its own. OpenBSD is very
different from FreeBSD.

Just running './wine' gives the error. I'm attempting to contact OpenBSD
developers for more information, but knowing how that went last time, doubt
it'll get much done...

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #7 from Austin English   2009-04-02 13:51:34 ---
FWIW, here's the relevant OpenBSD code:
http://packages.e.kth.se/common/src/os/OpenBSD/current/src/libexec/ld.so/loa...

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #8 from Austin English   2009-04-11 03:38:40 ---
Apparently the regression may be on OpenBSD's end:

Assorted improvements and code cleanup:

* malloc(3) has gained new attack mitigation measures; critical bookkeeping
structures are protected at runtime using mprotect(2) and allocated at random
addresses where possible.
    * A new version of the gdtoa code has been integrated, bringing better C99
support to printf(3) and friends.
    * Vastly improved C99 support in libm, including complex math support.
    * Sorry, this list is still being worked on.

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #9 from Austin English   2009-04-11 12:13:00 ---
Here's the diff from 4.4 to 4.5. I'm going to test soon to see if that's the
cause.

http://www.openbsd.org/cgi-bin/cvsweb/src/lib/libc/stdlib/malloc.c.diff?r1=1...

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

Juan Lang  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #10 from Juan Lang   2009-09-08 17:22:09 ---
*** Bug 19987 has been marked as a duplicate of this bug. ***

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #11 from IronForge   2009-09-10 17:37:15 ---
Created an attachment (id=23546)
 --> (http://bugs.winehq.org/attachment.cgi?id=23546)
An ualloc document - OpenBSD Programming

Reference Doc 1

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #12 from IronForge   2009-09-10 17:39:36 ---
Created an attachment (id=23547)
 --> (http://bugs.winehq.org/attachment.cgi?id=23547)
Reference 2

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #13 from IronForge   2009-09-10 19:28:19 ---
Greetings:

I am not a system programmer; but have coded at tad for school and work -
dealing mostly with packaged commercial apps.

Looking into the Dynamic Load Failure / Segfault / malloc issue, I wanted to
post a few reference material for those interested in tackling this problem.

Correct me if I've read previous posts and the ref matl incorrectly; but from
what's "obvious to the most casual observer" (my Math Prof used to say that),
is that the Malloc Error we get when attempting to run Wine in recent versions
of OBSD is due to the fact that OBSD changed the way its malloc system function
handles memory pointer allocations back in OBSDv3.8.

Reasons given involve a few issues regarding security, efficiency, and the
enforcement of proper coding practices.

IMHO, this bug can most likely be fixed if an OBSD Port of, or a complete
revision of WINE Source Code become compliant with Dynamic, Randomized, and
Thoroughly 'Freed' (work with me, here) Memory Management by System Calls
employed by OBSD.

Openports.se, a repository of OBSD pkgs, has a maintainer designated for Wine;
and OBSD Collective itself has mailing lists. I've sent a msg out to the
Maintainer and the advocacy mailing list for assistance and review of WINE's
memory allocation and usage practice (not in this detail; but you'll get the
drift).  I've also argued that it's beneficial to have a current, working
version of WINE available for people curious and interested in trying OBSD.

My first impression on the 10-yro port was not positive; and IMHO, probably
would disinterest others who may wish to try the OS outside of firewall/router
roles. With various MMORGs, graphic intensive apps, and a good number of
Desktop Office / Application Suites being tested and used on Wine, and with the
improvement of Virtualization Software, more (and that includes new) *NIX users
can forgo the need to have a Windows Box or Dual Boot - at home and at work.

I posted a document; and will list a few URLs on OBSD Man Pages and other
malloc reference material here (attaching them one by one seems tedious).

My recommendation are to have someone from the WINE dev team:
   a) Contact Openports.se's WINE maintainer (and join their team)...
   b) Contact OBSD's camp and sign up to their misc, dev, and ports mailing
lists...
   c) Sign up with Daemonforums.org and start a thread in the OpenBSD portion
of the forum...
...to start a dialog and see if anyone's willing to take a look at your code
and help you tackle this malloc issue and others.  Maybe you may be able to
invite several of their developers to sign up here as well.

You'll run into people who are strict in procedure; but you actually do need
people like that on your team to write good code overall.

Without further delay, here are the links:
Old Articles
-malloc article for OBSD 3.8  
http://kerneltrap.org/node/5584
-an O'Reilly interview with several OBSD gurus on the (then) "new" malloc
http://onlamp.com/pub/a/bsd/2005/10/20/openbsd_3_8.html?page=last 
-older wiki article on malloc - citing OBSD segfaults...
http://en.wikipedia.org/wiki/Malloc#OpenBSD.27s_malloc
-a detailed explanation of Address Space Layout Randomization (ASLR)
http://en.wikipedia.org/wiki/Address_space_layout_randomization

OBSD Man Pages for 
Malloc(3) Subroutine ***
http://www.openbsd.org/cgi-bin/man.cgi?query=cfree&sektion=3&apropos...
free(3) Subroutine    *** 
http://www.openbsd.org/cgi-bin/man.cgi?query=free&apropos=0&sektion=...
mmap(2) System Call   *** 
http://www.openbsd.org/cgi-bin/man.cgi?query=mmap&apropos=0&sektion=...
munmap(2) System Call *** 
http://www.openbsd.org/cgi-bin/man.cgi?query=munmap&sektion=2&arch=i...
mprotect(2) System Call ***
http://www.openbsd.org/cgi-bin/man.cgi?query=mprotect&sektion=2

Hope these help.

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

IronForge  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #23547|0                           |1
        is obsolete|                            |

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

Sylvain Petreolle  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

André H.  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #14 from André H.  2012-06-12 13:12:25 CDT ---
Any update on this? I'd say the patches are both outdated...

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

--- Comment #15 from Austin English  2013-01-25 20:46:01 CST ---
(In reply to comment #14)
> Any update on this? I'd say the patches are both outdated...

In wine-1.5.22-158-g236b4da (no patches) and OpenBSD 5.2, this is the bug I get
when trying to run ./wine --version.

[Bug 16666] wine segfaults on launch on OpenBSD

http://bugs.winehq.org/show_bug.cgi?id=16666

Ken Sharp  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #23546|application/octet-stream    |application/pdf
          mime type|                            |

[Bug 16666] wine segfaults on launch on OpenBSD

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=16666

joaopa  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #16 from joaopa  ---
Does the bug still occur with wine-4.19?

[Bug 16666] wine segfaults on launch on OpenBSD

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=16666

Saulius K.  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]