LockDIBSection problem

[Rein Klazes]

Hi,

Here is another problem in BigJig a jigsaw puzzle game (free download at
www.lenagames.com).

The program calls BitBlt with destination an in memory DIB. Inside
BitBlt, the destination bits are set to PAGE_NOACCESS:

| 0009:Call kernel32.VirtualProtect(727f0000,00123324,00000001,77a5f7bc) ret=770032ba
| 0009:Call ntdll.NtProtectVirtualMemory(ffffffff,77a5f69c,77a5f6a0,00000001,77a5f7bc) ret=77b6929e
| trace:virtual:NtProtectVirtualMemory 0xffffffff 0x727f0000 00123324 00000001
| trace:virtual:VIRTUAL_SetProt 0x727f0000-0x72913fff c----
| View: 0x727f0000 - 0x72913fff (valloc)
|       0x727f0000 - 0x72913fff c----
| 0009:Ret  ntdll.NtProtectVirtualMemory() retval=00000000 ret=77b6929e

This memory segment (727f0000,00123324) does not appear in the log
anymore until it is written to disk:

| 0009:Call kernel32.WriteFile(00000084,727f0000,00123324,77a5f360,00000000) ret=00408ac1
..
 The call fails ( not in the log because I fixed it) with error
STATUS_INVALID_USER_BUFFER and the program hangs.

I managed to fixed it in two ways:

1. put a 
    X11DRV_CoerceDIBSection( physDevDst, DIB_Status_InSync, FALSE );
at the end of X11DRV_BitBlt. This probably defeats the whole purpose of
these protections so:

2. Add a "IsBadReadPtr( buffer, bytesToWrite)" in the top of WriteFile
to force an exception and everything works.

Would that be an acceptable fix?

Rein.

[Alexandre Julliard]

Rein Klazes  writes:
> I managed to fixed it in two ways:
> - put a 
>  X11DRV_CoerceDIBSection( physDevDst, DIB_Status_InSync, FALSE );
>
> at the end of X11DRV_BitBlt. This probably defeats the whole purpose of
> these protections so:
> - Add a "IsBadReadPtr( buffer, bytesToWrite)" in the top of WriteFile
>
> to force an exception and everything works.
>
> Would that be an acceptable fix?

We don't want that at the top of WriteFile, but it could be OK to add
special handling of the INVALID_USER_BUFFER error, with a big FIXME
comment...

[Rein Klazes]

On 06 May 2005 16:01:06 +0200, you wrote:
> Rein Klazes  writes:
> > I managed to fixed it in two ways:
> > - put a 
> >  X11DRV_CoerceDIBSection( physDevDst, DIB_Status_InSync, FALSE );
> >
> > at the end of X11DRV_BitBlt. This probably defeats the whole purpose of
> > these protections so:
> > - Add a "IsBadReadPtr( buffer, bytesToWrite)" in the top of WriteFile
> >
> > to force an exception and everything works.
> >
> > Would that be an acceptable fix?
>
> We don't want that at the top of WriteFile, but it could be OK to add
> special handling of the INVALID_USER_BUFFER error, with a big FIXME
> comment...

Special handling: try again after an IsBadReadPtr call.

Changelog:

dlls/kernel	: file.c

Work around a problem where WriteFile is asked to write memory protected
by DIBSection code.

Rein.