[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#838328: marked as done (ghostscript: need to backport fix for Ghostscript bug 695862 (segfault on certain machines))

Your message dated Sun, 07 Jan 2024 22:45:54 -0600
with message-id <1911575.eGJsNajkDb@riemann>
and subject line Re: ghostscript: need to backport fix for Ghostscript bug 695862 (segfault on certain machines)
has caused the Debian Bug report #838328,
regarding ghostscript: need to backport fix for Ghostscript bug 695862 (segfault on certain machines)
to be marked as done.

This means that you claim that the problem has been dealt with.
If this is not the case it is now your responsibility to reopen the
Bug report if necessary, and/or fix the problem forthwith.

(NB: If you are a system administrator and have no idea what this
message is talking about, this may indicate a serious mail system
misconfiguration somewhere. Please contact owner@bugs.debian.org
immediately.)


-- 
838328: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=838328
Debian Bug Tracking System
Contact owner@bugs.debian.org with problems
--- Begin Message --- 
Package: ghostscript
Version: 9.06~dfsg-2+deb8u1

Ghostscript <=9.15 has a bug which manifests itself only on certain
machines, causing it to segfault on certain PDF files.  This is due to
the incorrect assumption that unlocking an already unlocked mutex is a
no-op, something which is not guaranteed by Single Unix, and may or
may not be true depending on the hardware / kernel / phase of the
moon.  See <URL:
http://bugs.ghostscript.com/show_bug.cgi?id=695862
 > for details.

I have been bitten by this bug (changed my processor and suddenly all
sorts of PDF files wouldn't print), so it is not theoretical.

I believe Debian should backport some fix for this.  One possible fix
(tested on my machine) is attached: it consists of using
error-checking mutexes instead of fast ones (error-checking mutexes
are documented to return EPERM when unlocked twice).  This fix is
called "hacky" by Chris Liddell in the above-mentioned bug report
page, but I think it is perfectly sensible.  Otherwise, backport the
patch made to the ghostscript upstream itself.

-- 
     David A. Madore
   ( http://www.madore.org/~david/ )

Description: fix ghostscript bug #695862 by using error-checking mutexes
Author: David Madore <david+bugs@madore.org>
Last-Update: 2016-09-20

Index: ghostscript-9.06~dfsg/base/gp_psync.c
===================================================================
--- ghostscript-9.06~dfsg.orig/base/gp_psync.c
+++ ghostscript-9.06~dfsg/base/gp_psync.c
@@ -20,6 +20,7 @@
 #include <pthread.h>
 #include "gserrors.h"
 #include "gpsync.h"
+#include <errno.h>
 
 /*
  * Thanks to Larry Jones <larry.jones@sdrc.com> for this revision of
@@ -64,10 +65,21 @@ gp_semaphore_open(gp_semaphore * sema)
     pt_semaphore_t * const sem = (pt_semaphore_t *)sema;
     int scode;
 
+    /* Idiotic threads API won't let you create an error-checking mutex
+     * without creating another "mutexattr" object which itself
+     * has to be initialized.  This is _absurdly_ verbose! */
+    static pthread_mutexattr_t mutexattr;
+    static char mutexattr_initialized = 0;
+
     if (!sema)
         return -1;		/* semaphores are not movable */
+    if ( ! mutexattr_initialized ) {
+	mutexattr_initialized = 1;
+	pthread_mutexattr_init(&mutexattr);
+	pthread_mutexattr_settype(&mutexattr, PTHREAD_MUTEX_ERRORCHECK);
+    }
     sem->count = 0;
-    scode = pthread_mutex_init(&sem->mutex, NULL);
+    scode = pthread_mutex_init(&sem->mutex, &mutexattr);
     if (scode == 0)
         scode = pthread_cond_init(&sem->cond, NULL);
     return SEM_ERROR_CODE(scode);
@@ -103,6 +115,8 @@ gp_semaphore_wait(gp_semaphore * sema)
     if (scode == 0)
         --sem->count;
     scode2 = pthread_mutex_unlock(&sem->mutex);
+    if ( scode2 == EPERM )
+	scode2 = 0;
     if (scode == 0)
         scode = scode2;
     return SEM_ERROR_CODE(scode);
@@ -120,6 +134,8 @@ gp_semaphore_signal(gp_semaphore * sema)
     if (sem->count++ == 0)
         scode = pthread_cond_signal(&sem->cond);
     scode2 = pthread_mutex_unlock(&sem->mutex);
+    if ( scode2 == EPERM )
+	scode2 = 0;
     if (scode == 0)
         scode = scode2;
     return SEM_ERROR_CODE(scode);
@@ -149,11 +165,22 @@ gp_monitor_open(gp_monitor * mona)
     pthread_mutex_t *mon;
     int scode;
 
+    /* Idiotic threads API won't let you create an error-checking mutex
+     * without creating another "mutexattr" object which itself
+     * has to be initialized.  This is _absurdly_ verbose! */
+    static pthread_mutexattr_t mutexattr;
+    static char mutexattr_initialized = 0;
+
     if (!mona)
         return -1;		/* monitors are not movable */
+    if ( ! mutexattr_initialized ) {
+	mutexattr_initialized = 1;
+	pthread_mutexattr_init(&mutexattr);
+	pthread_mutexattr_settype(&mutexattr, PTHREAD_MUTEX_ERRORCHECK);
+    }
     mon = &((gp_pthread_recursive_t *)mona)->mutex;
     ((gp_pthread_recursive_t *)mona)->self_id = 0;	/* Not valid unless mutex is locked */
-    scode = pthread_mutex_init(mon, NULL);
+    scode = pthread_mutex_init(mon, &mutexattr);
     return SEM_ERROR_CODE(scode);
 }
 
@@ -195,6 +222,8 @@ gp_monitor_leave(gp_monitor * mona)
     int scode;
 
     scode = pthread_mutex_unlock(mon);
+    if ( scode == EPERM )
+	scode = 0;
     ((gp_pthread_recursive_t *)mona)->self_id = 0;	/* Not valid unless mutex is locked */
     return SEM_ERROR_CODE(scode);
 }

--- End Message ---
--- Begin Message --- 
On Tue, 20 Sep 2016 00:44:30 +0200 David Madore <david+generic@madore.org> 
wrote:
> Package: ghostscript
> Version: 9.06~dfsg-2+deb8u1
> 
> Ghostscript <=9.15 has a bug which manifests itself only on certain
> machines, causing it to segfault on certain PDF files.  This is due to
> the incorrect assumption that unlocking an already unlocked mutex is a
> no-op, something which is not guaranteed by Single Unix, and may or
> may not be true depending on the hardware / kernel / phase of the
> moon.  See <URL:
> http://bugs.ghostscript.com/show_bug.cgi?id=695862
>  > for details.

I am sorry that no action was taken in 2016.  However, Debian is now shipping 
ghostscript verson 10, which includes this fix, so closing the bug.
-Steve

Attachment: signature.asc
Description: This is a digitally signed message part.


--- End Message ---
Reply to: