To: djgpp AT delorie DOT com
From: "Stas Sergeev (stsp2 AT yandex DOT ru) [via djgpp AT delorie DOT com]" <djgpp AT delorie DOT com>
Subject: [PATCH] process: make descriptor leak work-around conditional
Message-ID: <6df014b9-33da-d336-7508-2034a8ef197a@yandex.ru>
Date: Sun, 18 Aug 2019 04:38:46 +0300
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.6.1
MIME-Version: 1.0
Content-Type: multipart/mixed;
 boundary="------------24E1DCF9B6393E59B4E308BC"
Content-Language: en-MW
Reply-To: djgpp AT delorie DOT com

This is a multi-part message in MIME format.
--------------24E1DCF9B6393E59B4E308BC
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 7bit

This patch replaces the patch I sent previously under the Subject
[PATCH] exec: fix inversions in leak detection logic

The logic was not inverted; rather that code tested
__dpmi_free_ldt_descriptor() to see if it works.
Since it always works, I would suggest to remove
the tests.
The attached patch doesn't mess with that and
simply adds a variable to control the behaviour.

--------------24E1DCF9B6393E59B4E308BC
Content-Type: text/x-patch;
 name="0001-process-make-descriptor-leak-work-around-conditional.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename*0="0001-process-make-descriptor-leak-work-around-conditional.pa";
 filename*1="tch"

From 12d00dab0e179b4ec1c5a6a7d6e93eb907ef606d Mon Sep 17 00:00:00 2001
From: Stas Sergeev <stsp AT users DOT sourceforge DOT net>
Date: Sat, 17 Aug 2019 19:00:10 +0300
Subject: [PATCH] process: make descriptor leak work-around conditional

This patch adds the __spawn_flags variable and __spawn_leak_workaround
flag to allow the software to control the leak work-around.
Previous behaviour was to always enable the work-around unless
the DPMI host is cwsdpmi.

Without this patch it is not possible to spawn a prot-mode TSR
program like 32rtm. djgpp treats it as a leak and wipes out of
memory. With this patch things work properly if the DPMI server
is smart enough to direct the control to prev client after 32rtm
TSRed. The problem is that 32rtm just jumps to the realmode exit
addr, so the DPMI server doesn't see the exit and may get confused
unless the special logic is implemented for that case (i.e. not all
DPMI servers treat that correctly even after that patch).
---
 include/process.h              | 4 ++++
 src/libc/dos/process/dosexec.c | 4 +++-
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/include/process.h b/include/process.h
index 2816ce25..e57467d4 100644
--- a/include/process.h
+++ b/include/process.h
@@ -52,6 +52,10 @@ int __djgpp_spawn(int _mode, const char *_path, char *const _argv[], char *const
 #define SPAWN_EXTENSION_SRCH    1
 #define SPAWN_NO_EXTENSION_SRCH 2
 
+#define __spawn_leak_workaround             0x0001 /* free descriptor leaks */
+
+extern int __spawn_flags;
+
 #endif /* !_POSIX_SOURCE */
 #endif /* !__STRICT_ANSI__ */
 #endif /* !__dj_ENFORCE_ANSI_FREESTANDING */
diff --git a/src/libc/dos/process/dosexec.c b/src/libc/dos/process/dosexec.c
index 87d7fbd9..d547a31f 100644
--- a/src/libc/dos/process/dosexec.c
+++ b/src/libc/dos/process/dosexec.c
@@ -45,6 +45,7 @@
 extern char **_environ;
 
 int __dosexec_in_system = 0;
+int __spawn_flags = __spawn_leak_workaround;
 
 typedef struct {
   unsigned short eseg;
@@ -492,7 +493,8 @@ static int direct_exec_tail (const char *program, const char *args,
     /* r5 as corresponding DPMI call is supported beginning with v5.  */
 
     ret = __dpmi_get_capabilities(&flags, dpmi_vendor);
-    if (ret == 0 && strcmp(dpmi_vendor + 2, "CWSDPMI") == 0)
+    if ((ret == 0 && strcmp(dpmi_vendor + 2, "CWSDPMI") == 0)
+        || (__spawn_flags & __spawn_leak_workaround) == 0)
       workaround_descriptor_leaks = 0;
     else
     {
-- 
2.20.1


--------------24E1DCF9B6393E59B4E308BC--