Re: msvcrt: dont overwrite the standard streams (with test, try 5)
André Hentschel a écrit :
Some Apps e.g. mingw-apps use lpReserved2 to pass some data to the child-process, this data starts with 4 Byte of Zeros(unsinged 0) to tell msvcrt not to read that as an inherit-block.
Hi André a couple of comments to your patch
See also Bug 18479 --- dlls/msvcrt/file.c | 2 +- dlls/msvcrt/tests/Makefile.in | 2 +- dlls/msvcrt/tests/file.c | 50 +++++++++++++++++++++++++++++++++++++++++ 3 files changed, 52 insertions(+), 2 deletions(-)
diff --git a/dlls/msvcrt/file.c b/dlls/msvcrt/file.c index 6671f2f..8f63e8d 100644 --- a/dlls/msvcrt/file.c +++ b/dlls/msvcrt/file.c @@ -349,7 +349,7 @@ void msvcrt_init_io(void) InitializeCriticalSection(&MSVCRT_file_cs); MSVCRT_file_cs.DebugInfo->Spare[0] = (DWORD_PTR)(__FILE__ ": MSVCRT_file_cs"); GetStartupInfoA(&si); - if (si.cbReserved2 != 0 && si.lpReserved2 != NULL) + if (si.cbReserved2 != 0 && si.lpReserved2 != NULL && *(unsigned*)si.lpReserved2)
IMO, we should also test that cbReserved2 is >= sizeof(unsigned) otherwise we'd be in trouble
{ char* wxflag_ptr; HANDLE* handle_ptr; diff --git a/dlls/msvcrt/tests/Makefile.in b/dlls/msvcrt/tests/Makefile.in index 74199e1..8c71971 100644 --- a/dlls/msvcrt/tests/Makefile.in +++ b/dlls/msvcrt/tests/Makefile.in @@ -4,7 +4,7 @@ SRCDIR = @srcdir@ VPATH = @srcdir@ TESTDLL = msvcrt.dll APPMODE = -mno-cygwin -IMPORTS = kernel32 +IMPORTS = kernel32 user32
you should do it without user32. instead of doing this
+ while (MsgWaitForMultipleObjects( 1, &proc.hProcess, FALSE, INFINITE, QS_ALLINPUT ) != 0) + { + while (PeekMessage(&msg, 0, 0, 0, PM_REMOVE)) DispatchMessageA(&msg); + }
you could simply wait for the child process to terminate A+ -- Eric Pouech "The problem with designing something completely foolproof is to underestimate the ingenuity of a complete idiot." (Douglas Adams)
Eric Pouech schrieb:
Hi André a couple of comments to your patch always welcome!
IMO, we should also test that cbReserved2 is >= sizeof(unsigned) otherwise we'd be in trouble maybe you had something like that patch in mind: http://www.winehq.org/pipermail/wine-patches/2009-August/077766.html it checks if the size makes sense
you should do it without user32. instead of doing this
+ while (MsgWaitForMultipleObjects( 1, &proc.hProcess, FALSE, INFINITE, QS_ALLINPUT ) != 0) + { + while (PeekMessage(&msg, 0, 0, 0, PM_REMOVE)) DispatchMessageA(&msg); + }
you could simply wait for the child process to terminate A+
of course, your right, fixed that. -- Best Regards, André Hentschel
André Hentschel a écrit :
Eric Pouech schrieb:
Hi André a couple of comments to your patch
always welcome!
IMO, we should also test that cbReserved2 is >= sizeof(unsigned) otherwise we'd be in trouble
maybe you had something like that patch in mind: http://www.winehq.org/pipermail/wine-patches/2009-August/077766.html it checks if the size makes sense
no I'm just saying that since we do something like *(unsigned*)lpReserved, we'd better check that cbReserved2 >= sizeof(unsigned) I don't think that your patch is correct. If you want a coding of the "reserved" block, that you can extend, I think you could well have additional information *after* the block of handles+flags we're considering therefore the valid tests I see would be: cdReserved2 >= sizeof(unsigned) numH = *(unsigned*)lpReserved2 error if sizeof(unsigned) + numH*(1+sizeof(HANDLE)) > cbReserved2 A+ -- Eric Pouech "The problem with designing something completely foolproof is to underestimate the ingenuity of a complete idiot." (Douglas Adams)
Eric Pouech schrieb:
André Hentschel a écrit :
Eric Pouech schrieb:
Hi André a couple of comments to your patch
always welcome!
IMO, we should also test that cbReserved2 is >= sizeof(unsigned) otherwise we'd be in trouble
maybe you had something like that patch in mind: http://www.winehq.org/pipermail/wine-patches/2009-August/077766.html it checks if the size makes sense
no I'm just saying that since we do something like *(unsigned*)lpReserved, we'd better check that cbReserved2 >= sizeof(unsigned) I don't think that your patch is correct. If you want a coding of the "reserved" block, that you can extend, I think you could well have additional information *after* the block of handles+flags we're considering
therefore the valid tests I see would be: cdReserved2 >= sizeof(unsigned) numH = *(unsigned*)lpReserved2 error if sizeof(unsigned) + numH*(1+sizeof(HANDLE)) > cbReserved2
A+
I already understood the problem you mentioned and sent try 6: http://www.winehq.org/pipermail/wine-patches/2009-September/078158.html So for now this patch fixes a Problem, if we need it we also could check if the calculated size is bigger than the given one. (You were right with that, too) -- Best Regards, André Hentschel
participants (2)
-
André Hentschel -
Eric Pouech