openssl/INSTALL.W32
2010-11-23 23:01:01 +00:00

325 lines
12 KiB
Text

INSTALLATION ON THE WIN32 PLATFORM
----------------------------------
[Instructions for building for Windows CE can be found in INSTALL.WCE]
[Instructions for building for Win64 can be found in INSTALL.W64]
Here are a few comments about building OpenSSL for Win32 environments,
such as Windows NT and Windows 9x. It should be noted though that
Windows 9x are not ordinarily tested. Its mention merely means that we
attempt to maintain certain programming discipline and pay attention
to backward compatibility issues, in other words it's kind of expected
to work on Windows 9x, but no regression tests are actually performed.
On additional note newer OpenSSL versions are compiled and linked with
Winsock 2. This means that minimum OS requirement was elevated to NT 4
and Windows 98 [there is Winsock 2 update for Windows 95 though].
- you need Perl for Win32. Unless you will build on Cygwin, you will need
ActiveState Perl, available from http://www.activestate.com/ActivePerl.
- one of the following C compilers:
* Visual C++
* Borland C
* GNU C (Cygwin or MinGW)
- Netwide Assembler, a.k.a. NASM, available from http://nasm.sourceforge.net/
is required if you intend to utilize assembler modules. Note that NASM
is now the only supported assembler.
If you are compiling from a tarball or a CVS snapshot then the Win32 files
may well be not up to date. This may mean that some "tweaking" is required to
get it all to work. See the trouble shooting section later on for if (when?)
it goes wrong.
Visual C++
----------
If you want to compile in the assembly language routines with Visual
C++, then you will need already mentioned Netwide Assembler binary,
nasmw.exe or nasm.exe, to be available on your %PATH%.
Firstly you should run Configure with platform VC-WIN32:
> perl Configure VC-WIN32 --prefix=c:\some\openssl\dir
Where the prefix argument specifies where OpenSSL will be installed to.
Next you need to build the Makefiles and optionally the assembly
language files:
- If you are using NASM then run:
> ms\do_nasm
- If you don't want to use the assembly language files at all then run:
> perl Configure VC-WIN32 no-asm --prefix=c:/some/openssl/dir
> ms\do_ms
If you get errors about things not having numbers assigned then check the
troubleshooting section: you probably won't be able to compile it as it
stands.
Then from the VC++ environment at a prompt do:
> nmake -f ms\ntdll.mak
If all is well it should compile and you will have some DLLs and
executables in out32dll. If you want to try the tests then do:
> nmake -f ms\ntdll.mak test
To install OpenSSL to the specified location do:
> nmake -f ms\ntdll.mak install
Tweaks:
There are various changes you can make to the Win32 compile
environment. By default the library is not compiled with debugging
symbols. If you use the platform debug-VC-WIN32 instead of VC-WIN32
then debugging symbols will be compiled in.
By default in 1.0.0 OpenSSL will compile builtin ENGINES into the
separate shared librariesy. If you specify the "enable-static-engine"
option on the command line to Configure the shared library build
(ms\ntdll.mak) will compile the engines into libeay32.dll instead.
The default Win32 environment is to leave out any Windows NT specific
features.
If you want to enable the NT specific features of OpenSSL (currently
only the logging BIO) follow the instructions above but call the batch
file do_nt.bat instead of do_ms.bat.
You can also build a static version of the library using the Makefile
ms\nt.mak
Borland C++ builder 5
---------------------
* Configure for building with Borland Builder:
> perl Configure BC-32
* Create the appropriate makefile
> ms\do_nasm
* Build
> make -f ms\bcb.mak
Borland C++ builder 3 and 4
---------------------------
* Setup PATH. First must be GNU make then bcb4/bin
* Run ms\bcb4.bat
* Run make:
> make -f bcb.mak
GNU C (Cygwin)
--------------
Cygwin implements a Posix/Unix runtime system (cygwin1.dll) on top of
Win32 subsystem and provides a bash shell and GNU tools environment.
Consequently, a make of OpenSSL with Cygwin is virtually identical to
Unix procedure. It is also possible to create Win32 binaries that only
use the Microsoft C runtime system (msvcrt.dll or crtdll.dll) using
MinGW. MinGW can be used in the Cygwin development environment or in a
standalone setup as described in the following section.
To build OpenSSL using Cygwin:
* Install Cygwin (see http://cygwin.com/)
* Install Perl and ensure it is in the path. Both Cygwin perl
(5.6.1-2 or newer) and ActivePerl work.
* Run the Cygwin bash shell
* $ tar zxvf openssl-x.x.x.tar.gz
$ cd openssl-x.x.x
To build the Cygwin version of OpenSSL:
$ ./config
[...]
$ make
[...]
$ make test
$ make install
This will create a default install in /usr/local/ssl.
To build the MinGW version (native Windows) in Cygwin:
$ ./Configure mingw
[...]
$ make
[...]
$ make test
$ make install
Cygwin Notes:
"make test" and normal file operations may fail in directories
mounted as text (i.e. mount -t c:\somewhere /home) due to Cygwin
stripping of carriage returns. To avoid this ensure that a binary
mount is used, e.g. mount -b c:\somewhere /home.
"bc" is not provided in older Cygwin distribution. This causes a
non-fatal error in "make test" but is otherwise harmless. If
desired and needed, GNU bc can be built with Cygwin without change.
GNU C (MinGW/MSYS)
-------------
* Compiler and shell environment installation:
MinGW and MSYS are available from http://www.mingw.org/, both are
required. Run the installers and do whatever magic they say it takes
to start MSYS bash shell with GNU tools on its PATH.
N.B. Since source tar-ball can contain symbolic links, it's essential
that you use accompanying MSYS tar to unpack the source. It will
either handle them in one way or another or fail to extract them,
which does the trick too. Latter means that you may safely ignore all
"cannot create symlink" messages, as they will be "re-created" at
configure stage by copying corresponding files. Alternative programs
were observed to create empty files instead, which results in build
failure.
* Compile OpenSSL:
$ ./config
[...]
$ make
[...]
$ make test
This will create the library and binaries in root source directory
and openssl.exe application in apps directory.
It is also possible to cross-compile it on Linux by configuring
with './Configure --cross-compile-prefix=i386-mingw32- mingw ...'.
'make test' is naturally not applicable then.
libcrypto.a and libssl.a are the static libraries. To use the DLLs,
link with libeay32.a and libssl32.a instead.
See troubleshooting if you get error messages about functions not
having a number assigned.
Installation
------------
If you used the Cygwin procedure above, you have already installed and
can skip this section. For all other procedures, there's currently no real
installation procedure for Win32. There are, however, some suggestions:
- do nothing. The include files are found in the inc32/ subdirectory,
all binaries are found in out32dll/ or out32/ depending if you built
dynamic or static libraries.
- do as is written in INSTALL.Win32 that comes with modssl:
$ md c:\openssl
$ md c:\openssl\bin
$ md c:\openssl\lib
$ md c:\openssl\include
$ md c:\openssl\include\openssl
$ copy /b inc32\openssl\* c:\openssl\include\openssl
$ copy /b out32dll\ssleay32.lib c:\openssl\lib
$ copy /b out32dll\libeay32.lib c:\openssl\lib
$ copy /b out32dll\ssleay32.dll c:\openssl\bin
$ copy /b out32dll\libeay32.dll c:\openssl\bin
$ copy /b out32dll\openssl.exe c:\openssl\bin
Of course, you can choose another device than c:. C: is used here
because that's usually the first (and often only) harddisk device.
Note: in the modssl INSTALL.Win32, p: is used rather than c:.
Troubleshooting
---------------
Since the Win32 build is only occasionally tested it may not always compile
cleanly. If you get an error about functions not having numbers assigned
when you run ms\do_ms then this means the Win32 ordinal files are not up to
date. You can do:
> perl util\mkdef.pl crypto ssl update
then ms\do_XXX should not give a warning any more. However the numbers that
get assigned by this technique may not match those that eventually get
assigned in the CVS tree: so anything linked against this version of the
library may need to be recompiled.
If you get errors about unresolved symbols there are several possible
causes.
If this happens when the DLL is being linked and you have disabled some
ciphers then it is possible the DEF file generator hasn't removed all
the disabled symbols: the easiest solution is to edit the DEF files manually
to delete them. The DEF files are ms\libeay32.def ms\ssleay32.def.
Another cause is if you missed or ignored the errors about missing numbers
mentioned above.
If you get warnings in the code then the compilation will halt.
The default Makefile for Win32 halts whenever any warnings occur. Since VC++
has its own ideas about warnings which don't always match up to other
environments this can happen. The best fix is to edit the file with the
warning in and fix it. Alternatively you can turn off the halt on warnings by
editing the CFLAG line in the Makefile and deleting the /WX option.
You might get compilation errors. Again you will have to fix these or report
them.
One final comment about compiling applications linked to the OpenSSL library.
If you don't use the multithreaded DLL runtime library (/MD option) your
program will almost certainly crash because malloc gets confused -- the
OpenSSL DLLs are statically linked to one version, the application must
not use a different one. You might be able to work around such problems
by adding CRYPTO_malloc_init() to your program before any calls to the
OpenSSL libraries: This tells the OpenSSL libraries to use the same
malloc(), free() and realloc() as the application. However there are many
standard library functions used by OpenSSL that call malloc() internally
(e.g. fopen()), and OpenSSL cannot change these; so in general you cannot
rely on CRYPTO_malloc_init() solving your problem, and you should
consistently use the multithreaded library.
Linking your application
------------------------
If you link with static OpenSSL libraries [those built with ms/nt.mak],
then you're expected to additionally link your application with
WS2_32.LIB, ADVAPI32.LIB, GDI32.LIB and USER32.LIB. Those developing
non-interactive service applications might feel concerned about linking
with the latter two, as they are justly associated with interactive
desktop, which is not available to service processes. The toolkit is
designed to detect in which context it's currently executed, GUI,
console app or service, and act accordingly, namely whether or not to
actually make GUI calls. Additionally those who wish to
/DELAYLOAD:GDI32.DLL and /DELAYLOAD:USER32.DLL and actually keep them
off service process should consider implementing and exporting from
.exe image in question own _OPENSSL_isservice not relying on USER32.DLL.
E.g., on Windows Vista and later you could:
__declspec(dllexport) __cdecl BOOL _OPENSSL_isservice(void)
{ DWORD sess;
if (ProcessIdToSessionId(GetCurrentProcessId(),&sess))
return sess==0;
return FALSE;
}
If you link with OpenSSL .DLLs, then you're expected to include into
your application code small "shim" snippet, which provides glue between
OpenSSL BIO layer and your compiler run-time. Look up OPENSSL_Applink
reference page for further details.