Skip to contentSkip to conversation

Better error message when script fails due to script-security setting

PATCH Superseded [email protected]... 3 participants 10 messages Feb 17, 2019 – Mar 7, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Better error message when script fails due to script-security setting

From: Selva Nair <[email protected]...>

- Add a new return value (-2) for openvpn_execve() when external program execution is not allowed due to a low script-security setting.

- Add a corresponding error message

Errors and warnings in such cases will now display as "WARNING: failed running command (<cmd>) :" followed by

"disallowed by script-security setting" on all platforms

instead of the current

"external program did not execute -- returned error code -1" on Windows and
"external program fork failed" on other platforms.

The error is FATAL for some scripts and that behaviour is unchanged.

This helps the Windows GUI to detect when a connection failure results from a safer script-security setting enforced by the GUI, and show a relevant message.

---
This is being presented as a better alternative for patch 684. A separate patch may be needed for 2.4 -- will do if this is found acceptable.

 src/openvpn/run_command.c | 25 +++++++++++++++++++++----
 src/openvpn/win32.c       | 10 +++++++---
 2 files changed, 28 insertions(+), 7 deletions(-)
src/openvpn/run_command.c +21 -4
index 2d75a3e..7df7576 100644
@@ -65,12 +65,23 @@ system_error_message(int stat, struct gc_arena *gc)
     {
         buf_printf(&out, "external program did not execute -- ");
     }
-    buf_printf(&out, "returned error code %d", stat);
+    if (stat == -2)
+    {
+        buf_printf(&out, "disallowed by script-security setting");
+    }
+    else
+    {
+        buf_printf(&out, "returned error code %d", stat);
+    }
 #else  /* ifdef _WIN32 */
     if (stat == -1)
     {
         buf_printf(&out, "external program fork failed");
     }
+    else if (stat == -2)
+    {
+        buf_printf(&out, "disallowed by script-security setting");
+    }
     else if (!WIFEXITED(stat))
     {
         buf_printf(&out, "external program did not exit normally");
@@ -114,6 +125,8 @@ openvpn_execve_allowed(const unsigned int flags)
  * Run execve() inside a fork().  Designed to replicate the semantics of system() but
  * in a safer way that doesn't require the invocation of a shell or the risks
  * associated with formatting and parsing a command line.
+ * Returns the exit status of child, -2 if openvpn_execve_allowed() returned false,
+ * or -1 on other errors.
  */
 int
 openvpn_execve(const struct argv *a, const struct env_set *es, const unsigned int flags)
@@ -150,10 +163,14 @@ openvpn_execve(const struct argv *a, const struct env_set *es, const unsigned in
                 }
             }
         }
-        else if (!warn_shown && (script_security() < SSEC_SCRIPTS))
+        else
         {
-            msg(M_WARN, SCRIPT_SECURITY_WARNING);
-            warn_shown = true;
+            ret = -2;
+            if (!warn_shown && (script_security() < SSEC_SCRIPTS))
+            {
+                msg(M_WARN, SCRIPT_SECURITY_WARNING);
+                warn_shown = true;
+            }
         }
 #else  /* if defined(ENABLE_FEATURE_EXECVE) */
         msg(M_WARN, "openvpn_execve: execve function not available");
src/openvpn/win32.c +7 -3
index 463ac07..4eb7c40 100644
@@ -1137,10 +1137,14 @@ openvpn_execve(const struct argv *a, const struct env_set *es, const unsigned in
             free(env);
             gc_free(&gc);
         }
-        else if (!exec_warn && (script_security() < SSEC_SCRIPTS))
+        else
         {
-            msg(M_WARN, SCRIPT_SECURITY_WARNING);
-            exec_warn = true;
+            ret = -2;
+            if (!exec_warn && (script_security() < SSEC_SCRIPTS))
+            {
+                msg(M_WARN, SCRIPT_SECURITY_WARNING);
+                exec_warn = true;
+            }
         }
     }
     else
Signature · 1 line
2.1.4
#1 lore raw

Hi,

Quoted 7 lines · On Sat, Feb 16, 2019 at 08:55:41PM -0500, [email protected]... wrote:

From: Selva Nair <[email protected]...>

- Add a new return value (-2) for openvpn_execve() when external program execution is not allowed due to a low script-security setting.

- Add a corresponding error message

I find this a useful way forward (because it will not introduce new FATALs where we had none before, just make the errors more clear).

Since David had reservations on the other patch, I want to have explicit agreement from him before going on - and will poke on IRC to make sure I get attention.

Nevertheless, as we agreed to tag 2.4.7 *today* to make the release tomorrow and meet the Debian deadline, I will not include this in 2.4.7.

We can always do a 2.4.8 release "really soon", and include this (plus other potential 2.4 material which is important for Windows, like the OpenSSL 1.1.1 / cryptoapi stuff).

"Release early, release often" used to be the mantra... :-)

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
author
#2 lore raw

Hi

Quoted 9 lines · Gert Doering

Hi,

Quoted 6 lines · On Sat, Feb 16, 2019 at 08:55:41PM -0500, [email protected]... wrote:

From: Selva Nair <[email protected]...>

- Add a new return value (-2) for openvpn_execve() when external program execution is not allowed due to a low script-security setting.

Quoted 9 lines
Quoted 1 line

- Add a corresponding error message

I find this a useful way forward (because it will not introduce new FATALs where we had none before, just make the errors more clear).

Since David had reservations on the other patch, I want to have explicit agreement from him before going on - and will poke on IRC to make sure I get attention.

Just to be explicit:

Since David pointed out the "regressive" nature of the earlier patch, I too have
reservations about it, and prefer this approach.

Nevertheless, as we agreed to tag 2.4.7 *today* to make the release tomorrow

Quoted 1 line

and meet the Debian deadline, I will not include this in 2.4.7.

The planned script-security over-ride feature in the GUI is pretty intrusive. Taking
time to give it more thought, and preferably get some user feedback, is not a bad thing.

Selva

#3 lore raw
In reply to [email protected]...
Quoted 25 lines · On 17/02/2019 02:55, [email protected]... wrote:

From: Selva Nair <[email protected]...>

- Add a new return value (-2) for openvpn_execve() when external program execution is not allowed due to a low script-security setting.

- Add a corresponding error message

Errors and warnings in such cases will now display as "WARNING: failed running command (<cmd>) :" followed by

"disallowed by script-security setting" on all platforms

instead of the current

"external program did not execute -- returned error code -1" on Windows and
"external program fork failed" on other platforms.

The error is FATAL for some scripts and that behaviour is unchanged.

This helps the Windows GUI to detect when a connection failure results from a safer script-security setting enforced by the GUI, and show a relevant message.

Quoted 5 lines

---
This is being presented as a better alternative for patch 684. A separate patch may be needed for 2.4 -- will do if this is found acceptable.

Generally speaking, this looks much better and should avoid changing the behaviour. It just clarifies the error better. Feature-ACK from me.

[...]

Quoted 7 lines
src/openvpn/run_command.c +0 -0
index 2d75a3e..7df7576 100644
@@ -65,12 +65,23 @@ system_error_message(int stat, struct gc_arena *gc)
     {
         buf_printf(&out, "external program did not execute -- ");

src/openvpn/run_command.c system_error_message(int stat, struct gc_arena *gc) · line 75

Quoted 10 lines
src/openvpn/run_command.c +8 -1
     }
-    buf_printf(&out, "returned error code %d", stat);
+    if (stat == -2)
+    {
+        buf_printf(&out, "disallowed by script-security setting");
+    }
+    else
+    {
+        buf_printf(&out, "returned error code %d", stat);
+    }

Perhaps we could swap out the -1 and -2 values with macro constants, for code clarity? Also, perhaps it would look more structured if using a switch() instead of if()/else if().

Otherwise, the code looks good.

Another interesting are which most likely quite seldom fails, but we actually have yet another "undocumented" exit code ... This is not this patch's fault, but should probably be reviewed a bit more carefully.

In run_command.c:149-153:
--------------------------------------------------------------------

            pid = fork();
            if (pid == (pid_t)0) /* child side */
            {
                execve(cmd, argv, envp);
                exit(127);
            }

--------------------------------------------------------------------

If execve() fails, the exit code is 127. That would normally be caught by the waitpid() later on and this exit code would be returned by openvpn_execve().

This should be improved in a separate patch though, but is not of high urgency. I don't expect most setups having a high execve() failure rate. But catching it and reporting this error path would be good.

Signature · 4 lines
kind regards,

David Sommerseth
OpenVPN Inc
author
#4 lore raw

Hi,

Thanks for the review.

Quoted 51 lines · David Sommerseth
Quoted 25 lines · On 17/02/2019 02:55, [email protected]... wrote:

From: Selva Nair <[email protected]...>

- Add a new return value (-2) for openvpn_execve() when external program execution is not allowed due to a low script-security setting.

- Add a corresponding error message

Errors and warnings in such cases will now display as "WARNING: failed running command (<cmd>) :" followed by

"disallowed by script-security setting" on all platforms

instead of the current

"external program did not execute -- returned error code -1" on Windows and
"external program fork failed" on other platforms.

The error is FATAL for some scripts and that behaviour is unchanged.

This helps the Windows GUI to detect when a connection failure results from a safer script-security setting enforced by the GUI, and show a relevant message.

Quoted 5 lines

---
This is being presented as a better alternative for patch 684. A separate patch may be needed for 2.4 -- will do if this is found acceptable.

Generally speaking, this looks much better and should avoid changing the behaviour. It just clarifies the error better. Feature-ACK from me.

[...]

Quoted 14 lines
src/openvpn/run_command.c +5 -1
index 2d75a3e..7df7576 100644
@@ -65,12 +65,23 @@ system_error_message(int stat, struct gc_arena *gc)
     {
         buf_printf(&out, "external program did not execute -- ");
     }
-    buf_printf(&out, "returned error code %d", stat);
+    if (stat == -2)
+    {
+        buf_printf(&out, "disallowed by script-security setting");
+    }
+    else
Quoted 8 lines

src/openvpn/run_command.c system_error_message(int stat, struct gc_arena *gc) · line 75

Quoted 3 lines
src/openvpn/run_command.c +3 -0
+    {
+        buf_printf(&out, "returned error code %d", stat);
+    }

Perhaps we could swap out the -1 and -2 values with macro constants, for code
clarity? Also, perhaps it would look more structured if using a switch() instead of if()/else if().

Because of branching based on if (WIFEXITED(stat)) etc., switch() may be only
marginally cleaner and cause some deep indentation. But will give it a go.

Quoted 23 lines

Otherwise, the code looks good.

Another interesting are which most likely quite seldom fails, but we actually
have yet another "undocumented" exit code ... This is not this patch's fault,
but should probably be reviewed a bit more carefully.

In run_command.c:149-153:
--------------------------------------------------------------------

            pid = fork();
            if (pid == (pid_t)0) /* child side */
            {
                execve(cmd, argv, envp);
                exit(127);
            }

--------------------------------------------------------------------

If execve() fails, the exit code is 127. That would normally be caught by the
waitpid() later on and this exit code would be returned by openvpn_execve().

Quoted 5 lines

This should be improved in a separate patch though, but is not of high urgency. I don't expect most setups having a high execve() failure rate. But
catching it and reporting this error path would be good.

In fact its easy to trigger this -- just have a shebang with a non-existent path like
#!/bin/sh1 and child will exit with 127. In system_error_message() we translate that as
"could not execute external program"

This "could not" instead of "did not" is a good choice of words, but would be better
if we somehow marshal the actual errno set by execve (in this case ENOENT) to the parent. Its lost in the current implementation. But that's beyond this patch,
as you say.

I'll just add a macro for this 127 as well and leave it at that.

Selva

#5 lore raw

PATCH v2 1/1 Better error message when script fails due to script-security setting

In reply to [email protected]...

From: Selva Nair <[email protected]...>

- Add a new return value (-2) for openvpn_execve() when external program execution is not allowed due to a low script-security setting.

- Add a corresponding error message

Errors and warnings in such cases will now display as "WARNING: failed running command (<cmd>) :" followed by

"disallowed by script-security setting" on all platforms

instead of the current

"external program did not execute -- returned error code -1" on Windows and
"external program fork failed" on other platforms.

The error is FATAL for some scripts and that behaviour is unchanged.

This helps the Windows GUI to detect when a connection failure results from a safer script-security setting enforced by the GUI, and show a relevant message.

v2 changes as suggested by <[email protected]...>

- define macros for return values of openvpn_execve()
- replace if/else by switch() in system_error_message()

---
 src/openvpn/run_command.c | 93 ++++++++++++++++++++++++++++-------------------
 src/openvpn/run_command.h |  4 ++
 src/openvpn/win32.c       | 12 ++++--
 3 files changed, 68 insertions(+), 41 deletions(-)
src/openvpn/run_command.c +56 -37
index 2d75a3e..4c4adf9 100644
@@ -54,44 +54,57 @@ script_security_set(int level)
 }
 
 /*
- * Print an error message based on the status code returned by system().
+ * Generate an error message based on the status code returned by openvpn_execve().
  */
 static const char *
 system_error_message(int stat, struct gc_arena *gc)
 {
     struct buffer out = alloc_buf_gc(256, gc);
-#ifdef _WIN32
-    if (stat == -1)
+
+    switch (stat)
     {
-        buf_printf(&out, "external program did not execute -- ");
-    }
-    buf_printf(&out, "returned error code %d", stat);
+        case OPENVPN_EXECVE_NOT_ALLOWED:
+            buf_printf(&out, "disallowed by script-security setting");
+            break;
+
+#ifdef _WIN32
+        case OPENVPN_EXECVE_ERROR:
+            buf_printf(&out, "external program did not execute -- ");
+        /* fall through */
+
+        default:
+            buf_printf(&out, "returned error code %d", stat);
+            break;
 #else  /* ifdef _WIN32 */
-    if (stat == -1)
-    {
-        buf_printf(&out, "external program fork failed");
-    }
-    else if (!WIFEXITED(stat))
-    {
-        buf_printf(&out, "external program did not exit normally");
-    }
-    else
-    {
-        const int cmd_ret = WEXITSTATUS(stat);
-        if (!cmd_ret)
-        {
-            buf_printf(&out, "external program exited normally");
-        }
-        else if (cmd_ret == 127)
-        {
-            buf_printf(&out, "could not execute external program");
-        }
-        else
-        {
-            buf_printf(&out, "external program exited with error status: %d", cmd_ret);
-        }
-    }
+
+        case OPENVPN_EXECVE_ERROR:
+            buf_printf(&out, "external program fork failed");
+            break;
+
+        default:
+            if (!WIFEXITED(stat))
+            {
+                buf_printf(&out, "external program did not exit normally");
+            }
+            else
+            {
+                const int cmd_ret = WEXITSTATUS(stat);
+                if (!cmd_ret)
+                {
+                    buf_printf(&out, "external program exited normally");
+                }
+                else if (cmd_ret == OPENVPN_EXECVE_FAILURE)
+                {
+                    buf_printf(&out, "could not execute external program");
+                }
+                else
+                {
+                    buf_printf(&out, "external program exited with error status: %d", cmd_ret);
+                }
+            }
+            break;
 #endif /* ifdef _WIN32 */
+    }
     return (const char *)out.data;
 }
 
@@ -114,12 +127,14 @@ openvpn_execve_allowed(const unsigned int flags)
  * Run execve() inside a fork().  Designed to replicate the semantics of system() but
  * in a safer way that doesn't require the invocation of a shell or the risks
  * associated with formatting and parsing a command line.
+ * Returns the exit status of child, OPENVPN_EXECVE_NOT_ALLOWED if openvpn_execve_allowed()
+ * returns false, or OPENVPN_EXECVE_ERROR on other errors.
  */
 int
 openvpn_execve(const struct argv *a, const struct env_set *es, const unsigned int flags)
 {
     struct gc_arena gc = gc_new();
-    int ret = -1;
+    int ret = OPENVPN_EXECVE_ERROR;
     static bool warn_shown = false;
 
     if (a && a->argv[0])
@@ -136,7 +151,7 @@ openvpn_execve(const struct argv *a, const struct env_set *es, const unsigned in
             if (pid == (pid_t)0) /* child side */
             {
                 execve(cmd, argv, envp);
-                exit(127);
+                exit(OPENVPN_EXECVE_FAILURE);
             }
             else if (pid < (pid_t)0) /* fork failed */
             {
@@ -146,14 +161,18 @@ openvpn_execve(const struct argv *a, const struct env_set *es, const unsigned in
             {
                 if (waitpid(pid, &ret, 0) != pid)
                 {
-                    ret = -1;
+                    ret = OPENVPN_EXECVE_ERROR;
                 }
             }
         }
-        else if (!warn_shown && (script_security() < SSEC_SCRIPTS))
+        else
         {
-            msg(M_WARN, SCRIPT_SECURITY_WARNING);
-            warn_shown = true;
+            ret = OPENVPN_EXECVE_NOT_ALLOWED;
+            if (!warn_shown && (script_security() < SSEC_SCRIPTS))
+            {
+                msg(M_WARN, SCRIPT_SECURITY_WARNING);
+                warn_shown = true;
+            }
         }
 #else  /* if defined(ENABLE_FEATURE_EXECVE) */
         msg(M_WARN, "openvpn_execve: execve function not available");
@@ -227,7 +246,7 @@ openvpn_popen(const struct argv *a,  const struct env_set *es)
                     close(pipe_stdout[0]);         /* Close read end */
                     dup2(pipe_stdout[1],1);
                     execve(cmd, argv, envp);
-                    exit(127);
+                    exit(OPENVPN_EXECVE_FAILURE);
                 }
                 else if (pid > (pid_t)0)       /* parent side */
                 {
src/openvpn/run_command.h +4 -0
index 4501a5c..7ccb13c 100644
@@ -33,6 +33,10 @@
 #define SSEC_SCRIPTS   2 /* allow calling of built-in programs and user-defined scripts */
 #define SSEC_PW_ENV    3 /* allow calling of built-in programs and user-defined scripts that may receive a password as an environmental variable */
 
+#define OPENVPN_EXECVE_ERROR       -1 /* generic error while forking to run an external program */
+#define OPENVPN_EXECVE_NOT_ALLOWED -2 /* external program not run due to script security */
+#define OPENVPN_EXECVE_FAILURE    127 /* exit code passed back from child when execve fails */
+
 int script_security(void);
 
 void script_security_set(int level);
src/openvpn/win32.c +8 -4
index 463ac07..55d9843 100644
@@ -1088,7 +1088,7 @@ wide_cmd_line(const struct argv *a, struct gc_arena *gc)
 int
 openvpn_execve(const struct argv *a, const struct env_set *es, const unsigned int flags)
 {
-    int ret = -1;
+    int ret = OPENVPN_EXECVE_ERROR;
     static bool exec_warn = false;
 
     if (a && a->argv[0])
@@ -1137,10 +1137,14 @@ openvpn_execve(const struct argv *a, const struct env_set *es, const unsigned in
             free(env);
             gc_free(&gc);
         }
-        else if (!exec_warn && (script_security() < SSEC_SCRIPTS))
+        else
         {
-            msg(M_WARN, SCRIPT_SECURITY_WARNING);
-            exec_warn = true;
+            ret = OPENVPN_EXECVE_NOT_ALLOWED;
+            if (!exec_warn && (script_security() < SSEC_SCRIPTS))
+            {
+                msg(M_WARN, SCRIPT_SECURITY_WARNING);
+                exec_warn = true;
+            }
         }
     }
     else
Signature · 1 line
2.1.4
Acked-by on v2 1/1
#6 lore raw

Thanks. I have not tested the various error conditions, but the code makes sense to me and it passes compile and t_client tests (which use --up here), so it's not breaking stuff in fundamental and non-obvious ways.

(Since David was mostly happy with the v1 approach except for the "127" and "switch()", I assume he's totally fine with v2 now :-) )

Your patch has been applied to the master branch - it won't apply to 2.4 as we have no "run_command.*" there yet. Shall we cherrypick bf97c00f7dba4 and 253f015558 to get there? Or do you want to do a 2.4 version of the patch?

commit 01a3c876d4911ff7ea82b04edeb43defdebc69d7 (master)

Author: Selva Nair
Date:   Wed Feb 20 19:46:22 2019 -0500
     Better error message when script fails due to script-security setting

     Signed-off-by: Selva Nair <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg18223.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

author
#7 lore raw

Hi,

Quoted 7 lines · Gert Doering

Thanks. I have not tested the various error conditions, but the code makes sense to me and it passes compile and t_client tests (which use --up here), so it's not breaking stuff in fundamental and non-obvious ways.

Quoted 8 lines

(Since David was mostly happy with the v1 approach except for the "127" and "switch()", I assume he's totally fine with v2 now :-) )

Your patch has been applied to the master branch - it won't apply to 2.4 as we have no "run_command.*" there yet. Shall we cherrypick bf97c00f7dba4 and 253f015558 to get there? Or do you want to do a 2.4 version of the patch?

Would require more cherry-picks than that, or some manual edits, isn't it?

I can send a patch for 2.4 -- would you prefer the same as this -- i.e with switch(stat) in misc.c and and preprocessor macros in misc.h or a simpler, return code = -2 and error message as in version 1?

Selva

#8 lore raw

Hi,

Quoted 8 lines · Selva Nair
Quoted 4 lines · Gert Doering

Your patch has been applied to the master branch - it won't apply to 2.4 as we have no "run_command.*" there yet. Shall we cherrypick bf97c00f7dba4 and 253f015558 to get there? Or do you want to do a 2.4 version of the patch?

Would require more cherry-picks than that, or some manual edits, isn't it?

Yes, as we also have uncrustify patches that affect neighboring code.

Quoted 3 lines

I can send a patch for 2.4 -- would you prefer the same as this -- i.e with switch(stat) in misc.c and and preprocessor macros in misc.h or a simpler, return code = -2 and error message as in version 1?

I think we should have the same code (so, with the switch()), just in misc.c - when we come here next time, at least it will be the same code, not "something completely different".

thanks,

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#9 lore raw
In reply to Gert Doering
Quoted 3 lines · Gert Doering

(Since David was mostly happy with the v1 approach except for the "127" and "switch()", I assume he's totally fine with v2 now :-) )

Yeah, sorry I didn't manage to do the second review. I've done it post-apply, and it looks much better. It would get an ACK from me without much fuzz. Thanks!

Signature · 4 lines
kind regards,

David Sommerseth
OpenVPN Inc