RegOpenKeyExW() Question

[Andrew Talbot]

Hi,

I want to patch msi:action.c:ITERATE_WriteEnvironmentString() so that it
only calls RegCloseKey(env), in the cleanup, if "env" has been initialized
(to fix Coverity report CID-562). I can bypass the call to RegCloseKey()
for any early exit that occurs prior to calling RegOpenKeyExW(), and I can
include the call to RegCloseKey() for exits that occur after a successful
call to RegOpenKeyExW(). However, the problem, for me, is what to do if the
call to RegOpenKeyExW() fails: does the occurrence of this initialization
depend on the reason for failure. In other words, do I need something like
the following code to cater for the reason failure occurred?

res = RegOpenKeyExW(root, environment, 0, KEY_ALL_ACCESS, &env);

if (res != ERROR_SUCCESS)
        if (res == ERROR_INVALID_HANDLE ||
            res == RtlNtStatusToDosError(STATUS_BUFFER_OVERFLOW ||
            res == RtlNtStatusToDosError(STATUS_INVALID_PARAMETER))
            goto done2;
        else
            goto done1;

...

done1:
    RegCloseKey(env);
done2:
    ...

Or will the following code suffice?

res = RegOpenKeyExW(root, environment, 0, KEY_ALL_ACCESS, &env);

if (res != ERROR_SUCCESS)
        goto done2;

...

done1:
    RegCloseKey(env);
done2:
    ...

Thanks,

[Tom Spear]

On 6/22/07, Andrew Talbot  wrote:
> Hi,
>
> I want to patch msi:action.c:ITERATE_WriteEnvironmentString() so that it
> only calls RegCloseKey(env), in the cleanup, if "env" has been initialized
> (to fix Coverity report CID-562). I can bypass the call to RegCloseKey()
> for any early exit that occurs prior to calling RegOpenKeyExW(), and I can
> include the call to RegCloseKey() for exits that occur after a successful
> call to RegOpenKeyExW(). However, the problem, for me, is what to do if the
> call to RegOpenKeyExW() fails: does the occurrence of this initialization
> depend on the reason for failure. In other words, do I need something like
> the following code to cater for the reason failure occurred?
>
> res = RegOpenKeyExW(root, environment, 0, KEY_ALL_ACCESS, &env);
>
> if (res != ERROR_SUCCESS)
>        if (res == ERROR_INVALID_HANDLE ||
>            res == RtlNtStatusToDosError(STATUS_BUFFER_OVERFLOW ||
>            res == RtlNtStatusToDosError(STATUS_INVALID_PARAMETER))
>            goto done2;
>        else
>            goto done1;
>
> ...
>
> done1:
>    RegCloseKey(env);
> done2:
>    ...
>
> Or will the following code suffice?
>
> res = RegOpenKeyExW(root, environment, 0, KEY_ALL_ACCESS, &env);
>
> if (res != ERROR_SUCCESS)
>        goto done2;
>
> ...
>
> done1:
>    RegCloseKey(env);
> done2:
>    ...

At the risk of looking like an idiot, I'll take a stab at this..  If
you RegCloseKey on env, and it is not initialized, then RegCloseKey
will generate ERROR_INVALID_HANDLE, so I personally think that
something like the first approach would be the better one to go with.
env should only be initialized (according to what I have seen) if
RegOpenKeyExW returns ERROR_SUCCESS.  But I haven't seen every case,
so there may be something that causes it to be initialized when the
return is not ERROR_SUCCESS.

[Andrew Talbot]

Tom Spear wrote:
> At the risk of looking like an idiot, I'll take a stab at this..  If
> you RegCloseKey on env, and it is not initialized, then RegCloseKey
> will generate ERROR_INVALID_HANDLE, so I personally think that
> something like the first approach would be the better one to go with.
> env should only be initialized (according to what I have seen) if
> RegOpenKeyExW returns ERROR_SUCCESS.  But I haven't seen every case,
> so there may be something that causes it to be initialized when the
> return is not ERROR_SUCCESS.

Thanks, Tom. I think I might submit a patch to fix the violation that
Coverity flags, which is prior to the RegOpenKeyW() call, and just insert a
FIXME: comment where the return value of RegOpenKeyW() is tested, without
changing the existing functionality.