# See the bottom of this file for the POD documentation. Search for the
# string '=head'.
#######################################################################
#
# Win32::API::Callback - Perl Win32 API Import Facility
#
# Author: Aldo Calpini <dada@perl.it>
# Author: Daniel Dragan <bulkdd@cpan.org>
# Maintainer: Cosimo Streppone <cosimo@cpan.org>
#
#######################################################################
package Win32::API::Callback;
use strict;
use warnings;
use vars qw( $VERSION @ISA $Stage2FuncPtrPkd );
$VERSION = '0.76_04';
require Exporter; # to export the constants to the main:: space
require DynaLoader; # to dynuhlode the module.
@ISA = qw( Exporter DynaLoader );
#use Data::Dumper;
sub DEBUG {
if ($WIN32::API::DEBUG) {
printf @_ if @_ or return 1;
}
else {
return 0;
}
}
use Win32::API qw ( WriteMemory ) ;
use Win32::API::Type;
use Config;
#use Win32::API::Struct; #not implemented
BEGIN {
#there is supposed to be 64 bit IVs on 32 bit perl compatibility here
#but it is untested
#Win64 added in 5.7.3
eval "sub IVSIZE () { ".length(pack($] >= 5.007003 ? 'J' : 'I' ,0))." }";
#what kind of stack processing/calling convention/machine code we needed
eval "sub ISX64 () { ".(index($Config{'archname'},"MSWin32-x64") == 0 ? 1 : 0)." }";
eval 'sub OPV () {'.$].'}';
sub OPV();
sub CONTEXT_XMM0();
sub CONTEXT_RAX();
*IsBadStringPtr = *Win32::API::IsBadStringPtr;
sub PTRSIZE ();
*PTRSIZE = *Win32::API::PTRSIZE;
sub PTRLET ();
*PTRLET = *Win32::API::Type::pointer_pack_type;
if(OPV <= 5.008000){ #don't have unpackstring in C
eval('sub _CallUnpack {return unpack($_[0], $_[1]);}');
}
}
#######################################################################
# dynamically load in the API extension module.
#
bootstrap Win32::API::Callback;
#######################################################################
# PUBLIC METHODS
#
sub new {
my ($class, $proc, $in, $out, $callconvention) = @_;
my $self = bless {}, $class; #about croak/die safety, can safely bless here,
#a ::Callback has no DESTROY, it has no resources to release, there is a HeapBlock obj
#stored in the ::Callback hash, but the HeapBlock destroys on its own
# printf "(PM)Callback::new: got proc='%s', in='%s', out='%s'\n", $proc, $in, $out;
$self->{intypes} = []; #XS requires this, do not remove
if (ref($in) eq 'ARRAY') {
foreach (@$in) {
push(@{$self->{intypes}}, $_);
}
}
else {
my @in = split '', $in;
foreach (@in) {
push(@{$self->{intypes}}, $_);
}
}
$self->{inbytes} = 0;
foreach(@{$self->{intypes}}){ #calc how long the c stack is
if($_ eq 'Q' or $_ eq 'q' or $_ eq 'D' or $_ eq 'd'){
$self->{inbytes} += 8; #always 8
}
else{
$self->{inbytes} += PTRSIZE; #4 or 8
}
}
$self->{outtype} = $out;
$self->{out} = Win32::API->type_to_num($out);
$self->{sub} = $proc;
$self->{cdecl} = Win32::API::calltype_to_num($callconvention);
DEBUG "(PM)Callback::new: calling CallbackCreate($self)...\n";
my $hproc = MakeCB($self);
DEBUG "(PM)Callback::new: hproc=$hproc\n";
$self->{code} = $hproc;
#### cast the spell
return $self;
}
sub MakeStruct {
my ($self, $n, $addr) = @_;
DEBUG "(PM)Win32::API::Callback::MakeStruct: got self='$self'\n";
my $struct = Win32::API::Struct->new($self->{intypes}->[$n]);
$struct->FromMemory($addr);
return $struct;
}
#this was rewritten in XS, and is broken b/c it doesn't work on 32bit Perl with Quads
#sub MakeParamArr { #on x64, never do "$i++; $packedparam .= $arr->[$i];"
# #on x86, structs and over word size params appears on the stack,
# #on x64 anything over the size of a "word" is passed by pointer
# #nothing takes more than 8 bytes per parameter on x64
# #there is no way to formally specify a pass by copy struct in ::Callback
# #this only matters on x86, a work around is a bunch of N/I parameters,
# #repack them as Js, then concat them, and you have the original pass by copy
# #x86 struct
# my ($self, $arr) = @_;
# my ($i, @pass_arr) = (0);
# for(@{$self->{intypes}}){ #elements of intypes are not 1 to 1 with stack params
# my ($typeletter, $packedparam, $finalParam, $unpackletter) = ($_, $arr->[$i]);
#
# #structs don't work, this is broken code from old version
# #$self->{intypes} is letters types not C prototype params
# #C prototype support would have to exist for MakeStruct to work
# if( $typeletter eq 'S' || $typeletter eq 's'){
# die "Win32::API::Callback::MakeParamArr type letter \"S\" and struct support not implemented";
# #push(@pass_arr, MakeStruct($self, $i, $packedparam));
# }elsif($typeletter eq 'I'){
# $unpackletter = 'I', goto UNPACK;
# }elsif($typeletter eq 'i'){
# $unpackletter = 'i', goto UNPACK;
# }elsif($typeletter eq 'f' || $typeletter eq 'F'){
# $unpackletter = 'f', goto UNPACK;
# }
# elsif($typeletter eq 'd' || $typeletter eq 'D'){
# if(IVSIZE == 4){ #need more data, 32 bit machine
# $packedparam .= $arr->[++$i];
# }
# $unpackletter = 'd', goto UNPACK;
# }
# elsif($typeletter eq 'N' || $typeletter eq 'L' #on x64, J is 8 bytes
# || (IVSIZE == 8 ? $typeletter eq 'Q': 0)){
# $unpackletter = 'J', goto UNPACK;
# }elsif($typeletter eq 'n' || $typeletter eq 'l'
# || (IVSIZE == 8 ? $typeletter eq 'q': 0)){
# $unpackletter = 'j', goto UNPACK;
# }elsif(IVSIZE == 4 && ($typeletter eq 'q' || $typeletter eq 'Q')){
# #need more data, 32 bit machine
# $finalParam = $packedparam . $arr->[++$i];
# }elsif($typeletter eq 'p' || $typeletter eq 'P'){
# if(!IsBadStringPtr($arr->[$i], ~0)){ #P letter is terrible design
# $unpackletter = 'p', goto UNPACK;
# }#else undef
# }
# else{ die "Win32::API::Callback::MakeParamArr unknown in type letter $typeletter";}
# goto GOTPARAM;
# UNPACK:
# $finalParam = unpack($unpackletter, $packedparam);
# GOTPARAM:
# $i++;
# push(@pass_arr, $finalParam);
# }
# return \@pass_arr;
#}
#on x64
#void RunCB($self, $EBP_ESP, $retval)
#on x86
#void RunCB($self, $EBP_ESP, $retval, $unwindcount, $F_or_D)
if(! ISX64 ) {
*RunCB = sub {#32 bits
my $self = $_[0];
my (@pass_arr, $return, $typeletter, $inbytes, @arr);
$inbytes = $self->{inbytes};
#first is ebp copy then ret address
$inbytes += PTRSIZE * 2;
my $paramcount = $inbytes / PTRSIZE ;
my $stackstr = unpack('P'.$inbytes, pack(PTRLET, $_[1]));
#pack () were added in 5.7.2
if (OPV > 5.007002) {
@arr = unpack("(a[".PTRLET."])[$paramcount]",$stackstr);
} else {
#letter can not be used for size, must be numeric on 5.6
@arr = unpack(("a4") x $paramcount,$stackstr);
}
shift @arr, shift @arr; #remove ebp copy and ret address
$paramcount -= 2;
$return = &{$self->{sub}}(@{MakeParamArr($self, \@arr)});
#now the return type
$typeletter = $self->{outtype};
#float_or_double flag, its always used
#float is default for faster copy of probably unused value
$_[4] = 0;
#its all the same in memory
if($typeletter eq 'n' || $typeletter eq 'N'
|| $typeletter eq 'l' || $typeletter eq 'L'
|| $typeletter eq 'i' || $typeletter eq 'I'){
$_[2] = pack(PTRLET, $return);
}elsif($typeletter eq 'q' || $typeletter eq 'Q'){
if(IVSIZE == 4){
if($self->{'UseMI64'} || ref($return)){ #un/signed meaningless
$_[2] = Math::Int64::int64_to_native($return);
}
else{
warn("Win32::API::Callback::RunCB return value for return type Q is under 8 bytes long")
if length($return) < 8;
$_[2] = $return.''; #$return should be a 8 byte string
#will be garbage padded in XS if < 8, but must be a string, not a IV or under
}
}
else{
$_[2] = pack($typeletter, $return);
}
}elsif($typeletter eq 'f' || $typeletter eq 'F' ){
$_[2] = pack('f', $return);
}elsif($typeletter eq 'd' || $typeletter eq 'D' ){
$_[2] = pack('d', $return);
$_[4] = 1; #use double
}else { #return null
$_[2] = "\x00" x 8;
}
if(! $self->{cdecl}){
$_[3] = PTRSIZE * $paramcount; #stack rewind amount in bytes
}
else{$_[3] = 0;}
};
}
else{ #64 bits
*RunCB = sub {
my $self = $_[0];
my (@pass_arr, $return, $typeletter);
my $paramcount = $self->{inbytes} / IVSIZE;
my $stack_ptr = unpack('P[J]', pack('J', ($_[1]+CONTEXT_RAX())));
my $stack_str = unpack('P['.$self->{inbytes}.']', $stack_ptr);
my @stack_arr = unpack("(a[J])[$paramcount]",$stack_str);
#not very efficient, todo search for f/F/d/D in new() not here
my $XMMStr = unpack('P['.(4 * 16).']', pack('J', ($_[1]+CONTEXT_XMM0())));
#print Dumper([unpack('(H[32])[4]', $XMMStr)]);
my @XMM = unpack('(a[16])[4]', $XMMStr);
#assume registers are copied to shadow stack space already
#because of ... prototype, so only XMM registers need to be fetched.
#Limitation, vararg funcs on x64 get floating points in normal registers
#not XMMs, so a vararg function taking floats and doubles in the first 4
#parameters isn't supported
if($paramcount){
for(0..($paramcount > 4 ? 4 : $paramcount)-1){
my $typeletter = ${$self->{intypes}}[$_];
if($typeletter eq 'f' || $typeletter eq 'F' || $typeletter eq 'd'
|| $typeletter eq 'D'){
#x64 calling convention does not use the high 64 bits of a XMM register
#although right here the high 64 bits are in @XMM elements
#J on x64 is 8 bytes, a double will not corrupt, this is unreachable on x86
#note we are copying 16 bytes elements to @stack_arr, @stack_arr is
#normally 8 byte elements, unpack ignores the excess bytes later
$stack_arr[$_] = $XMM[$_];
}
}
}
#print Dumper(\@stack_arr);
#print Dumper(\@XMM);
$return = &{$self->{sub}}(@{MakeParamArr($self, \@stack_arr)});
#now the return type
$typeletter = $self->{outtype};
#its all the same in memory
if($typeletter eq 'n' || $typeletter eq 'N'
|| $typeletter eq 'l' || $typeletter eq 'L'
|| $typeletter eq 'i' || $typeletter eq 'I'
|| $typeletter eq 'q' || $typeletter eq 'Q'){
$_[2] = pack('J', $return);
}
elsif($typeletter eq 'f' || $typeletter eq 'F' ){
$_[2] = pack('f', $return);
}
elsif($typeletter eq 'd' || $typeletter eq 'D' ){
$_[2] = pack('d', $return);
}
else { #return null
$_[2] = pack('J', 0);
}
};
}
sub MakeCB{
my $self = $_[0];
#this x86 function does not corrupt the callstack in a debugger since it
#uses ebp and saves ebp on the stack, the function won't have a pretty
#name though
my $code = (!ISX64) ? ('' #parenthesis required to constant fold
."\x55" # push ebp
."\x8B\xEC" # mov ebp, esp
."\x83\xEC\x0C"# sub esp, 0Ch
."\x8D\x45\xFC" # lea eax, [ebp+FuncRtnCxtVar]
."\x50"# push eax
."\x8D\x45\xF4"# lea eax, [ebp+retval]
."\x50"# push eax
."\x8B\xC5"# mov eax,ebp
."\x50"# push eax
."\xB8").PackedRVTarget($self)#B8 mov imm32 to eax, a HV * winds up here
.("\x50"# push eax
."\xB8").$Stage2FuncPtrPkd # mov eax, 0C0DE0001h
.("\xFF\xD0"# call eax
#since ST(0) is volatile, we don't care if we fill it with garbage
."\x80\x7D\xFE\x00"#cmp [ebp+FuncRtnCxtVar.F_Or_D], 0
."\x74\x05"# jz 5 bytes
."\xDD\x45\xF4"# fld qword ptr [ebp+retval] (double)
."\xEB\x03"# jmp 3 bytes
."\xD9\x45\xF4"# fld dword ptr [ebp+retval] (float)
#rewind sp to entry sp, no pop push after this point
."\x83\xC4\x24"# add esp, 24h
."\x8B\x45\xF4"# mov eax, dword ptr [ebp+retval]
#edx might be garbage, we don't care, caller only looks at volatile
#registers that the caller's prototype says the caller does
."\x8B\x55\xF8"# mov edx, dword ptr [ebp+retval+4]
#can't use retn op, it requires a immediate count, our count is in a register
#only one register available now, this will be complicated
."\x0F\xB7\x4D\xFC"#movzx ecx, word ptr [ebp+FuncRtnCxtVar.unwind_len]
."\x01\xCC"# add esp, ecx , might be zero or more
."\x8B\x4D\x04"# mov ecx, dword ptr [ebp+4] ret address
."\x8B\x6D\x00"# mov ebp, dword ptr [ebp+0] restore BP
."\xFF\xE1")# jmp ecx
#begin x64 part
#these packs don't constant fold in < 5.17 :-(
#they are here for readability
:(''.pack('C', 0b01000000 #REX base
| 0b00001000 #REX.W
| 0b00000001 #REX.B
).pack('C', 0xB8+2) #mov to r10 register
.PackedRVTarget($self)
.pack('C', 0b01000000 #REX base
| 0b00001000 #REX.W
).pack('C', 0xB8) #mov to rax register
.$Stage2FuncPtrPkd
."\xFF\xE0");# jmp rax
#making a full function in Perl in x64 was removed because RtlAddFunctionTable
#has no effect on VS 2008 debugger, it is a bug in VS 2008, in WinDbg the C callstack
#is correct with RtlAddFunctionTable, and broken without RtlAddFunctionTable
#in VS 2008, the C callstack was always broken since WinDbg and VS 2008 both
#*only* use Unwind Tables on x64 to calculate C callstacks, they do not, I think,
#use 32 bit style EBP/RBP walking, x64 VC almost never uses BP addressing anyway.
#The easiest fix was to not have dynamic machine code in the callstack at all,
#which is what I did. Having good C callstacks in a debugger with ::API and
#::Callback are a good goal.
#
##--- c:\documents and settings\administrator\desktop\w32api\callback\callback.c -
# $code .= "\x4C\x8B\xDC";# mov r11,rsp
# $code .= "\x49\x89\x4B\x08";# mov qword ptr [r11+8],rcx
# $code .= "\x49\x89\x53\x10";# mov qword ptr [r11+10h],rdx
# $code .= "\x4D\x89\x43\x18";# mov qword ptr [r11+18h],r8
# $code .= "\x4D\x89\x4B\x20";# mov qword ptr [r11+20h],r9
# $code .= "\x48\x83\xEC\x78";# sub rsp,78h
# #void (*LPerlCallback)(SV *, void *, unsigned __int64 *, void *) =
# #( void (*)(SV *, void *, unsigned __int64 *, void *)) 0xC0DE00FFFF000001;
# #__m128 arr [4];
# #__m128 retval;
## arr[0].m128_u64[0] = 0xFFFF00000000FF10;
##00000000022D1017 48 B8 10 FF 00 00 00 00 FF FF mov rax,0FFFF00000000FF10h
##arr[0].m128_u64[1] = 0xFFFF00000000FF11;
## arr[1].m128_u64[0] = 0xFFFF00000000FF20;
## arr[1].m128_u64[1] = 0xFFFF00000000FF21;
## arr[2].m128_u64[0] = 0xFFFF00000000FF30;
## arr[2].m128_u64[1] = 0xFFFF00000000FF31;
## arr[3].m128_u64[0] = 0xFFFF00000000FF40;
## arr[3].m128_u64[1] = 0xFFFF00000000FF41;
#
## LPerlCallback((SV *)0xC0DE00FFFF000002, (void*) arr, (unsigned __int64 *)&retval,
## (DWORD_PTR)&a);
##00000000022D1021 4D 8D 4B 08 lea r9,[r11+8] #no 4th param
# $code .= "\x4D\x8D\x43\xA8";# lea r8,[r11-58h] #&retval param
##00000000022D1029 49 89 43 B8 mov qword ptr [r11-48h],rax
##00000000022D102D 48 B8 11 FF 00 00 00 00 FF FF mov rax,0FFFF00000000FF11h
# $code .= "\x49\x8D\x53\xB8";# lea rdx,[r11-48h] #arr param
##00000000022D103B 49 89 43 C0 mov qword ptr [r11-40h],rax
##00000000022D103F 48 B8 20 FF 00 00 00 00 FF FF mov rax,0FFFF00000000FF20h
##00000000022D1049 48 B9 02 00 00 FF FF 00 DE C0 mov rcx,0C0DE00FFFF000002h
# $code .= "\x48\xB9".PackedRVTarget($self);# mov rcx, the HV *
##00000000022D1053 49 89 43 C8 mov qword ptr [r11-38h],rax
##00000000022D1057 48 B8 21 FF 00 00 00 00 FF FF mov rax,0FFFF00000000FF21h
##00000000022D1061 49 89 43 D0 mov qword ptr [r11-30h],rax
##00000000022D1065 48 B8 30 FF 00 00 00 00 FF FF mov rax,0FFFF00000000FF30h
##00000000022D106F 49 89 43 D8 mov qword ptr [r11-28h],rax
##00000000022D1073 48 B8 31 FF 00 00 00 00 FF FF mov rax,0FFFF00000000FF31h
##00000000022D107D 49 89 43 E0 mov qword ptr [r11-20h],rax
##00000000022D1081 48 B8 40 FF 00 00 00 00 FF FF mov rax,0FFFF00000000FF40h
##00000000022D108B 49 89 43 E8 mov qword ptr [r11-18h],rax
##00000000022D108F 48 B8 41 FF 00 00 00 00 FF FF mov rax,0FFFF00000000FF41h
##00000000022D1099 49 89 43 F0 mov qword ptr [r11-10h],rax
##00000000022D109D 48 B8 01 00 00 FF FF 00 DE C0 mov rax,0C0DE00FFFF000001h
# $code .= "\x48\xB8".$Stage2FuncPtrPkd; # mov rax,0C0DE00FFFF000001h
# $code .= "\xFF\xD0";# call rax
## return *(void **)&retval;
# $code .= "\x48\x8B\x44\x24\x20";# mov rax,qword ptr [retval]
##}
# $code .= "\x48\x83\xC4\x78";# add rsp,78h
# $code .= "\xC3";# ret
#$self->{codestr} = $code; #save memory
#32 bit perl doesn't use DEP in my testing, but use executable heap to be safe
#a Win32::API::Callback::HeapBlock is a ref to scalar, that scalar has the void *
my $ptr = ${($self->{codeExecAlloc} = Win32::API::Callback::HeapBlock->new(length($code)))};
WriteMemory($ptr, $code, length($code));
return $ptr;
}
1;
__END__
#######################################################################
# DOCUMENTATION
#
=head1 NAME
Win32::API::Callback - Callback support for Win32::API
=head1 SYNOPSIS
use Win32::API;
use Win32::API::Callback;
my $callback = Win32::API::Callback->new(
sub { my($a, $b) = @_; return $a+$b; },
"NN", "N",
);
Win32::API->Import(
'mydll', 'two_integers_cb', 'KNN', 'N',
);
$sum = two_integers_cb( $callback, 3, 2 );
=head1 FOREWORDS
=over 4
=item *
Support for this module is B<highly experimental> at this point.
=item *
I won't be surprised if it doesn't work for you.
=item *
Feedback is very appreciated.
=item *
Documentation is in the work. Either see the SYNOPSIS above
or the samples in the F<samples> or the tests in the F<t> directory.
=back
=head1 USAGE
Win32::API::Callback uses a subset of the type letters of Win32::API. C Prototype
interface isn't supported. Not all the type letters of Win32::API are supported
in Win32::API::Callback.
=over 4
=item C<I>:
value is an unsigned integer (unsigned int)
=item C<i>:
value is an signed integer (signed int or int)
=item C<N>:
value is a unsigned pointer sized number (unsigned long)
=item C<n>:
value is a signed pointer sized number (signed long or long)
=item C<Q>:
value is a unsigned 64 bit integer number (unsigned long long, unsigned __int64)
See next item for details.
=item C<q>:
value is a signed 64 bit integer number (long long, __int64)
If your perl has 'Q'/'q' quads support for L<pack> then Win32::API's 'q'
is a normal perl numeric scalar. All 64 bit Perls have quad support. Almost no
32 bit Perls have quad support. On 32 bit Perls, without quad support,
Win32::API::Callback's 'q'/'Q' letter is a packed 8 byte string.
So C<0x8000000050000000> from a perl with native Quad support
would be written as C<"\x00\x00\x00\x50\x00\x00\x00\x80"> on a 32 bit
Perl without Quad support. To improve the use of 64 bit integers with
Win32::API::Callback on a 32 bit Perl without Quad support, there is
a per Win32::API::Callback object setting called L<Win32::API/UseMI64>
that causes all quads to be accepted as, and returned as L<Math::Int64>
objects. 4 to 8 byte long pass by copy/return type C aggregate types
are very rare in Windows, but they are supported as "in" and return
types by using 'q'/'Q' on 32 and 64 bits. Converting between the C aggregate
and its representation as a quad is up to the reader. For "out" in
Win32::API::Callback (not "in"), if the argument is a reference, it will
automatically be treated as a Math::Int64 object without having to
previously call this function.
=item C<F>:
value is a floating point number (float)
=item C<D>:
value is a double precision number (double)
=item C<Unimplemented types>:
Unimplemented in Win32::API::Callback types such as shorts, chars, and
smaller than "machine word size" (32/64bit) numbers can be processed
by specifying N, then masking off the high bytes.
For example, to get a char, specify N, then do C<$numeric_char = $_[2] & 0xFF;>
in your Perl callback sub. To get a short, specify N, then do
C<$numeric_char = $_[2] & 0xFFFF;> in your Perl callback sub.
=back
=head2 FUNCTIONS
=head3 new
$CallbackObj = Win32::API::Callback->new( sub { print "hello world";},
'NDF', 'Q', '__cdecl');
$CallbackObj = Win32::API::Callback->new( sub { print "hello world";},
$in, $out);
Creates and returns a new Win32::API::Callback object. Calling convention
parameter is optional. Calling convention parameter has same behaviour as
Win32::API's calling convention parameter. C prototype parsing of Win32::API
is not available with Win32::API::Callback. If the C caller assumes the
callback has vararg parameters, and the platform is 64 bits/x64, in the first 4
parameters, if they are floats or doubles they will be garbage. Note there is
no way to create a Win32::API::Callback callback with a vararg prototype.
A workaround is to put "enough" Ns as the in types, and stop looking at the @_
slices in your Perl sub callback after a certain count. Usually the first
parameter will somehow indicate how many additional stack parameters you are
receiving. The Ns in @_ will eventually become garbage, technically they are
the return address, saved registers, and C stack allocated variables of the
caller. They are effectivly garbage for your vararg callback. All vararg
callbacks on 32 bits must supply a calling convention, and it must be '__cdecl'
or 'WINAPIV'.
=head2 METHODS
=head3 UseMI64
See L<Win32::API/UseMI64>.
=head1 SEE ALSO
L<Win32::API::Callback::IATPatch>
=head1 AUTHOR
Aldo Calpini ( I<dada@perl.it> ).
Daniel Dragan ( I<bulkdd@cpan.org> ).
=head1 MAINTAINER
Cosimo Streppone ( I<cosimo@cpan.org> ).
=cut