openssl/INSTALL.md
slontis 250a7adbea Add "no-fips-post" configure option.
Using this option disables the OpenSSL FIPS provider
self tests.
This is intended for debugging purposes only,
as it breaks FIPS compliance.

Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/25063)
2024-08-09 09:12:45 +10:00

67 KiB

Build and Install

This document describes installation on all supported operating systems: the Unix/Linux family (including macOS), OpenVMS, and Windows.

Table of Contents

Prerequisites

To install OpenSSL, you will need:

  • A "make" implementation
  • Perl 5 with core modules (please read NOTES-PERL.md)
  • The Perl module Text::Template (please read NOTES-PERL.md)
  • an ANSI C compiler
  • POSIX C library (at least POSIX.1-2008), or compatible types and functionality.
  • a development environment in the form of development libraries and C header files
  • a supported operating system

For additional platform specific requirements, solutions to specific issues and other details, please read one of these:

Notational conventions

Throughout this document, we use the following conventions.

Commands

Any line starting with a dollar sign is a command line.

$ command

The dollar sign indicates the shell prompt and is not to be entered as part of the command.

Choices

Several words in curly braces separated by pipe characters indicate a mandatory choice, to be replaced with one of the given words. For example, the line

$ echo { WORD1 | WORD2 | WORD3 }

represents one of the following three commands

$ echo WORD1
- or -
$ echo WORD2
- or -
$ echo WORD3

One or several words in square brackets separated by pipe characters denote an optional choice. It is similar to the mandatory choice, but it can also be omitted entirely.

So the line

$ echo [ WORD1 | WORD2 | WORD3 ]

represents one of the four commands

$ echo WORD1
- or -
$ echo WORD2
- or -
$ echo WORD3
- or -
$ echo

Arguments

Optional Arguments are enclosed in square brackets.

[option...]

A trailing ellipsis means that more than one could be specified.

Quick Installation Guide

If you just want to get OpenSSL installed without bothering too much about the details, here is the short version of how to build and install OpenSSL. If any of the following steps fails, please consult the Installation in Detail section below.

Building OpenSSL

Use the following commands to configure, build and test OpenSSL. The testing is optional, but recommended if you intend to install OpenSSL for production use.

Unix / Linux / macOS / NonStop

$ ./Configure
$ make
$ make test

OpenVMS

Use the following commands to build OpenSSL:

$ perl Configure
$ mms
$ mms test

Windows

If you are using Visual Studio, open a Developer Command Prompt and issue the following commands to build OpenSSL.

$ perl Configure
$ nmake
$ nmake test

As mentioned in the Choices section, you need to pick one of the four Configure targets in the first command.

Most likely you will be using the VC-WIN64A/VC-WIN64A-HYBRIDCRT target for 64bit Windows binaries (AMD64) or VC-WIN32/VC-WIN32-HYBRIDCRT for 32bit Windows binaries (X86). The other two options are VC-WIN64I (Intel IA64, Itanium) and VC-CE (Windows CE) are rather uncommon nowadays.

Installing OpenSSL

The following commands will install OpenSSL to a default system location.

Danger Zone: even if you are impatient, please read the following two paragraphs carefully before you install OpenSSL.

For security reasons the default system location is by default not writable for unprivileged users. So for the final installation step administrative privileges are required. The default system location and the procedure to obtain administrative privileges depends on the operating system. It is recommended to compile and test OpenSSL with normal user privileges and use administrative privileges only for the final installation step.

On some platforms OpenSSL is preinstalled as part of the Operating System. In this case it is highly recommended not to overwrite the system versions, because other applications or libraries might depend on it. To avoid breaking other applications, install your copy of OpenSSL to a different location which is not in the global search path for system libraries.

Finally, if you plan on using the FIPS module, you need to read the Post-installation Notes further down.

Unix / Linux / macOS / NonStop

Depending on your distribution, you need to run the following command as root user or prepend sudo to the command:

$ make install

By default, OpenSSL will be installed to

/usr/local

More precisely, the files will be installed into the subdirectories

/usr/local/bin
/usr/local/lib
/usr/local/include
...

depending on the file type, as it is custom on Unix-like operating systems.

OpenVMS

Use the following command to install OpenSSL.

$ mms install

By default, OpenSSL will be installed to

SYS$COMMON:[OPENSSL]

Windows

If you are using Visual Studio, open the Developer Command Prompt elevated and issue the following command.

$ nmake install

The easiest way to elevate the Command Prompt is to press and hold down both the <CTRL> and <SHIFT> keys while clicking the menu item in the task menu.

The default installation location is

C:\Program Files\OpenSSL

for native binaries, or

C:\Program Files (x86)\OpenSSL

for 32bit binaries on 64bit Windows (WOW64).

Installing to a different location

To install OpenSSL to a different location (for example into your home directory for testing purposes) run Configure as shown in the following examples.

The options --prefix and --openssldir are explained in further detail in Directories below, and the values used here are mere examples.

On Unix:

$ ./Configure --prefix=/opt/openssl --openssldir=/usr/local/ssl

On OpenVMS:

$ perl Configure --prefix=PROGRAM:[INSTALLS] --openssldir=SYS$MANAGER:[OPENSSL]

Note: if you do add options to the configuration command, please make sure you've read more than just this Quick Start, such as relevant NOTES-* files, the options outline below, as configuration options may change the outcome in otherwise unexpected ways.

Configuration Options

There are several options to ./Configure to customize the build (note that for Windows, the defaults for --prefix and --openssldir depend on what configuration is used and what Windows implementation OpenSSL is built on. For more information, see the Notes for Windows platforms.

API Level

--api=x.y[.z]

Build the OpenSSL libraries to support the API for the specified version. If no-deprecated is also given, don't build with support for deprecated APIs in or below the specified version number. For example, adding

--api=1.1.0 no-deprecated

will remove support for all APIs that were deprecated in OpenSSL version 1.1.0 or below. This is a rather specialized option for developers. If you just intend to remove all deprecated APIs up to the current version entirely, just specify no-deprecated. If --api isn't given, it defaults to the current (minor) OpenSSL version.

Cross Compile Prefix

--cross-compile-prefix=<PREFIX>

The <PREFIX> to include in front of commands for your toolchain.

It is likely to have to end with dash, e.g. a-b-c- would invoke GNU compiler as a-b-c-gcc, etc. Unfortunately cross-compiling is too case-specific to put together one-size-fits-all instructions. You might have to pass more flags or set up environment variables to actually make it work. Android and iOS cases are discussed in corresponding Configurations/15-*.conf files. But there are cases when this option alone is sufficient. For example to build the mingw64 target on Linux --cross-compile-prefix=x86_64-w64-mingw32- works. Naturally provided that mingw packages are installed. Today Debian and Ubuntu users have option to install a number of prepackaged cross-compilers along with corresponding run-time and development packages for "alien" hardware. To give another example --cross-compile-prefix=mipsel-linux-gnu- suffices in such case.

For cross compilation, you must configure manually. Also, note that --openssldir refers to target's file system, not one you are building on.

Build Type

--debug

Build OpenSSL with debugging symbols and zero optimization level.

--release

Build OpenSSL without debugging symbols. This is the default.

Directories

libdir

--libdir=DIR

The name of the directory under the top of the installation directory tree (see the --prefix option) where libraries will be installed. By default this is lib. Note that on Windows only static libraries (*.lib) will be stored in this location. Shared libraries (*.dll) will always be installed to the bin directory.

Some build targets have a multilib postfix set in the build configuration. For these targets the default libdir is lib<multilib-postfix>. Please use --libdir=lib to override the libdir if adding the postfix is undesirable.

openssldir

--openssldir=DIR

Directory for OpenSSL configuration files, and also the default certificate and key store. Defaults are:

Unix:           /usr/local/ssl
Windows:        C:\Program Files\Common Files\SSL
OpenVMS:        SYS$COMMON:[OPENSSL-COMMON]

For 32bit Windows applications on Windows 64bit (WOW64), always replace C:\Program Files by C:\Program Files (x86).

prefix

--prefix=DIR

The top of the installation directory tree. Defaults are:

Unix:           /usr/local
Windows:        C:\Program Files\OpenSSL
OpenVMS:        SYS$COMMON:[OPENSSL]

Compiler Warnings

--strict-warnings

This is a developer flag that switches on various compiler options recommended for OpenSSL development. It only works when using gcc or clang as the compiler. If you are developing a patch for OpenSSL then it is recommended that you use this option where possible.

Compression Algorithm Flags

with-brotli-include

--with-brotli-include=DIR

The directory for the location of the brotli include files (i.e. the location of the brotli include directory). This option is only necessary if enable-brotli is used and the include files are not already on the system include path.

with-brotli-lib

--with-brotli-lib=LIB

On Unix: this is the directory containing the brotli libraries. If not provided, the system library path will be used.

The names of the libraries are:

  • libbrotlicommon.a or libbrotlicommon.so
  • libbrotlidec.a or libbrotlidec.so
  • libbrotlienc.a or libbrotlienc.so

On Windows: this is the directory containing the brotli libraries. If not provided, the system library path will be used.

The names of the libraries are:

  • brotlicommon.lib
  • brotlidec.lib
  • brotlienc.lib

with-zlib-include

--with-zlib-include=DIR

The directory for the location of the zlib include file. This option is only necessary if zlib is used and the include file is not already on the system include path.

with-zlib-lib

--with-zlib-lib=LIB

On Unix: this is the directory containing the zlib library. If not provided the system library path will be used.

On Windows: this is the filename of the zlib library (with or without a path). This flag must be provided if the zlib-dynamic option is not also used. If zlib-dynamic is used then this flag is optional and defaults to ZLIB1 if not provided.

On VMS: this is the filename of the zlib library (with or without a path). This flag is optional and if not provided then GNV$LIBZSHR, GNV$LIBZSHR32 or GNV$LIBZSHR64 is used by default depending on the pointer size chosen.

with-zstd-include

--with-zstd-include=DIR

The directory for the location of the Zstd include file. This option is only necessary if enable-std is used and the include file is not already on the system include path.

OpenSSL requires Zstd 1.4 or greater. The Linux kernel source contains a zstd.h file that is not compatible with the 1.4.x Zstd distribution, the compilation will generate an error if the Linux zstd.h is included before (or instead of) the Zstd distribution header.

with-zstd-lib

--with-zstd-lib=LIB

On Unix: this is the directory containing the Zstd library. If not provided the system library path will be used.

On Windows: this is the filename of the Zstd library (with or without a path). This flag must be provided if the enable-zstd-dynamic option is not also used. If zstd-dynamic is used then this flag is optional and defaults to LIBZSTD if not provided.

Seeding the Random Generator

--with-rand-seed=seed1[,seed2,...]

A comma separated list of seeding methods which will be tried by OpenSSL in order to obtain random input (a.k.a "entropy") for seeding its cryptographically secure random number generator (CSPRNG). The current seeding methods are:

os

Use a trusted operating system entropy source. This is the default method if such an entropy source exists.

getrandom

Use the getrandom(2) or equivalent system call.

devrandom

Use the first device from the DEVRANDOM list which can be opened to read random bytes. The DEVRANDOM preprocessor constant expands to

"/dev/urandom","/dev/random","/dev/srandom"

on most unix-ish operating systems.

egd

Check for an entropy generating daemon. This source is ignored by the FIPS provider.

rdcpu

Use the RDSEED or RDRAND command on x86 or RNDRRS command on aarch64 if provided by the CPU.

none

Disable automatic seeding. This is the default on some operating systems where no suitable entropy source exists, or no support for it is implemented yet. This option is ignored by the FIPS provider.

For more information, see the section Notes on random number generation at the end of this document.

jitter

When configured with enable-jitter, a "JITTER" RNG is compiled that can provide an alternative software seed source. It can be configured by setting seed option in openssl.cnf. A minimal openssl.cnf is shown below:

openssl_conf = openssl_init

[openssl_init]
random = random

[random]
seed=JITTER

It uses a statically linked jitterentropy-library as the seed source.

Additional configuration flags available:

--with-jitter-include=DIR

The directory for the location of the jitterentropy.h include file, if it is outside the system include path.

--with-jitter-lib=DIR

This is the directory containing the static libjitterentropy.a library, if it is outside the system library path.

Setting the FIPS HMAC key

--fips-key=value

As part of its self-test validation, the FIPS module must verify itself by performing a SHA-256 HMAC computation on itself. The default key is the SHA256 value of "holy hand grenade of antioch" and is sufficient for meeting the FIPS requirements.

To change the key to a different value, use this flag. The value should be a hex string no more than 64 characters.

Enable and Disable Features

Feature options always come in pairs, an option to enable feature xxxx, and an option to disable it:

[ enable-xxxx | no-xxxx ]

Whether a feature is enabled or disabled by default, depends on the feature. In the following list, always the non-default variant is documented: if feature xxxx is disabled by default then enable-xxxx is documented and if feature xxxx is enabled by default then no-xxxx is documented.

no-afalgeng

Don't build the AFALG engine.

This option will be forced on a platform that does not support AFALG.

enable-ktls

Build with Kernel TLS support.

This option will enable the use of the Kernel TLS data-path, which can improve performance and allow for the use of sendfile and splice system calls on TLS sockets. The Kernel may use TLS accelerators if any are available on the system. This option will be forced off on systems that do not support the Kernel TLS data-path.

enable-asan

Build with the Address sanitiser.

This is a developer option only. It may not work on all platforms and should never be used in production environments. It will only work when used with gcc or clang and should be used in conjunction with the no-shared option.

enable-acvp-tests

Build support for Automated Cryptographic Validation Protocol (ACVP) tests.

This is required for FIPS validation purposes. Certain ACVP tests require access to algorithm internals that are not normally accessible. Additional information related to ACVP can be found at https://github.com/usnistgov/ACVP.

no-apps

Do not build apps, e.g. the openssl program. This is handy for minimization. This option also disables tests.

no-asm

Do not use assembler code.

This should be viewed as debugging/troubleshooting option rather than for production use. On some platforms a small amount of assembler code may still be used even with this option.

no-async

Do not build support for async operations.

no-atexit

Do not use atexit() in libcrypto builds.

atexit() has varied semantics between platforms and can cause SIGSEGV in some circumstances. This option disables the atexit registration of OPENSSL_cleanup. By default, NonStop configurations use no-atexit.

no-autoalginit

Don't automatically load all supported ciphers and digests.

Typically OpenSSL will make available all of its supported ciphers and digests. For a statically linked application this may be undesirable if small executable size is an objective. This only affects libcrypto. Ciphers and digests will have to be loaded manually using EVP_add_cipher() and EVP_add_digest() if this option is used. This option will force a non-shared build.

no-autoerrinit

Don't automatically load all libcrypto/libssl error strings.

Typically OpenSSL will automatically load human readable error strings. For a statically linked application this may be undesirable if small executable size is an objective.

enable-brotli

Build with support for brotli compression/decompression.

enable-brotli-dynamic

Like the enable-brotli option, but has OpenSSL load the brotli library dynamically when needed.

This is only supported on systems where loading of shared libraries is supported.

no-autoload-config

Don't automatically load the default openssl.cnf file.

Typically OpenSSL will automatically load a system config file which configures default SSL options.

enable-buildtest-c++

While testing, generate C++ buildtest files that simply check that the public OpenSSL header files are usable standalone with C++.

Enabling this option demands extra care. For any compiler flag given directly as configuration option, you must ensure that it's valid for both the C and the C++ compiler. If not, the C++ build test will most likely break. As an alternative, you can use the language specific variables, CFLAGS and CXXFLAGS.

--banner=text

Use the specified text instead of the default banner at the end of configuration.

--w

On platforms where the choice of 32-bit or 64-bit architecture is not explicitly specified, Configure will print a warning message and wait for a few seconds to let you interrupt the configuration. Using this flag skips the wait.

no-bulk

Build only some minimal set of features. This is a developer option used internally for CI build tests of the project.

no-cached-fetch

Never cache algorithms when they are fetched from a provider. Normally, a provider indicates if the algorithms it supplies can be cached or not. Using this option will reduce run-time memory usage but it also introduces a significant performance penalty. This option is primarily designed to help with detecting incorrect reference counting.

no-capieng

Don't build the CAPI engine.

This option will be forced if on a platform that does not support CAPI.

no-cmp

Don't build support for Certificate Management Protocol (CMP) and Certificate Request Message Format (CRMF).

no-cms

Don't build support for Cryptographic Message Syntax (CMS).

no-comp

Don't build support for SSL/TLS compression.

If this option is enabled (the default), then compression will only work if the zlib or zlib-dynamic options are also chosen.

enable-crypto-mdebug

This now only enables the failed-malloc feature.

enable-crypto-mdebug-backtrace

This is a no-op; the project uses the compiler's address/leak sanitizer instead.

no-ct

Don't build support for Certificate Transparency (CT).

no-deprecated

Don't build with support for deprecated APIs up until and including the version given with --api (or the current version, if --api wasn't specified).

no-dgram

Don't build support for datagram based BIOs.

Selecting this option will also force the disabling of DTLS.

no-docs

Don't build and install documentation, i.e. manual pages in various forms.

no-dso

Don't build support for loading Dynamic Shared Objects (DSO)

enable-devcryptoeng

Build the /dev/crypto engine.

This option is automatically selected on the BSD platform, in which case it can be disabled with no-devcryptoeng.

no-dynamic-engine

Don't build the dynamically loaded engines.

This only has an effect in a shared build.

no-ec

Don't build support for Elliptic Curves.

no-ec2m

Don't build support for binary Elliptic Curves

enable-ec_nistp_64_gcc_128

Enable support for optimised implementations of some commonly used NIST elliptic curves.

This option is only supported on platforms:

  • with little-endian storage of non-byte types
  • that tolerate misaligned memory references
  • where the compiler:
    • supports the non-standard type __uint128_t
    • defines the built-in macro __SIZEOF_INT128__

enable-egd

Build support for gathering entropy from the Entropy Gathering Daemon (EGD).

no-engine

Don't build support for loading engines.

no-err

Don't compile in any error strings.

enable-external-tests

Enable building of integration with external test suites.

This is a developer option and may not work on all platforms. The following external test suites are currently supported:

  • GOST engine test suite
  • Python PYCA/Cryptography test suite
  • krb5 test suite

See the file test/README-external.md for further details.

no-filenames

Don't compile in filename and line number information (e.g. for errors and memory allocation).

enable-fips

Build (and install) the FIPS provider

no-fips-securitychecks

Don't perform FIPS module run-time checks related to enforcement of security parameters such as minimum security strength of keys.

no-fips-post

Don't perform FIPS module Power On Self Tests.

This option MUST be used for debugging only as it makes the FIPS provider non-compliant. It is useful when setting breakpoints in FIPS algorithms.

enable-fuzz-libfuzzer, enable-fuzz-afl

Build with support for fuzzing using either libfuzzer or AFL.

These are developer options only. They may not work on all platforms and should never be used in production environments.

See the file fuzz/README.md for further details.

no-gost

Don't build support for GOST based ciphersuites.

Note that if this feature is enabled then GOST ciphersuites are only available if the GOST algorithms are also available through loading an externally supplied engine.

no-http

Disable HTTP support.

no-legacy

Don't build the legacy provider.

Disabling this also disables the legacy algorithms: MD2 (already disabled by default).

no-makedepend

Don't generate dependencies.

no-module

Don't build any dynamically loadable engines.

This also implies no-dynamic-engine.

no-multiblock

Don't build support for writing multiple records in one go in libssl

Note: this is a different capability to the pipelining functionality.

no-nextprotoneg

Don't build support for the Next Protocol Negotiation (NPN) TLS extension.

no-ocsp

Don't build support for Online Certificate Status Protocol (OCSP).

no-padlockeng

Don't build the padlock engine.

no-hw-padlock

As synonym for no-padlockeng. Deprecated and should not be used.

no-pic

Don't build with support for Position Independent Code.

enable-pie

Build with support for Position Independent Execution.

no-pinshared

Don't pin the shared libraries.

By default OpenSSL will attempt to stay in memory until the process exits. This is so that libcrypto and libssl can be properly cleaned up automatically via an atexit() handler. The handler is registered by libcrypto and cleans up both libraries. On some platforms the atexit() handler will run on unload of libcrypto (if it has been dynamically loaded) rather than at process exit.

This option can be used to stop OpenSSL from attempting to stay in memory until the process exits. This could lead to crashes if either libcrypto or libssl have already been unloaded at the point that the atexit handler is invoked, e.g. on a platform which calls atexit() on unload of the library, and libssl is unloaded before libcrypto then a crash is likely to happen.

Note that shared library pinning is not automatically disabled for static builds, i.e., no-shared does not imply no-pinshared. This may come as a surprise when linking libcrypto statically into a shared third-party library, because in this case the shared library will be pinned. To prevent this behaviour, you need to configure the static build using no-shared and no-pinshared together.

Applications can suppress running of the atexit() handler at run time by using the OPENSSL_INIT_NO_ATEXIT option to OPENSSL_init_crypto(). See the man page for it for further details.

no-posix-io

Don't use POSIX IO capabilities.

no-psk

Don't build support for Pre-Shared Key based ciphersuites.

no-rdrand

Don't use hardware RDRAND capabilities.

no-rfc3779

Don't build support for RFC3779, "X.509 Extensions for IP Addresses and AS Identifiers".

sctp

Build support for Stream Control Transmission Protocol (SCTP).

no-shared

Do not create shared libraries, only static ones.

See Notes on shared libraries below.

no-sm2-precomp

Disable using the SM2 precomputed table on aarch64 to make the library smaller.

no-sock

Don't build support for socket BIOs.

no-srp

Don't build support for Secure Remote Password (SRP) protocol or SRP based ciphersuites.

no-srtp

Don't build Secure Real-Time Transport Protocol (SRTP) support.

no-sse2

Exclude SSE2 code paths from 32-bit x86 assembly modules.

Normally SSE2 extension is detected at run-time, but the decision whether or not the machine code will be executed is taken solely on CPU capability vector. This means that if you happen to run OS kernel which does not support SSE2 extension on Intel P4 processor, then your application might be exposed to "illegal instruction" exception. There might be a way to enable support in kernel, e.g. FreeBSD kernel can be compiled with CPU_ENABLE_SSE, and there is a way to disengage SSE2 code paths upon application start-up, but if you aim for wider "audience" running such kernel, consider no-sse2. Both the 386 and no-asm options imply no-sse2.

no-ssl-trace

Don't build with SSL Trace capabilities.

This removes the -trace option from s_client and s_server, and omits the SSL_trace() function from libssl.

Disabling ssl-trace may provide a small reduction in libssl binary size.

no-static-engine

Don't build the statically linked engines.

This only has an impact when not built "shared".

no-stdio

Don't use anything from the C header file stdio.h that makes use of the FILE type. Only libcrypto and libssl can be built in this way. Using this option will suppress building the command line applications. Additionally, since the OpenSSL tests also use the command line applications, the tests will also be skipped.

no-tests

Don't build test programs or run any tests.

enable-tfo

Build with support for TCP Fast Open (RFC7413). Supported on Linux, macOS and FreeBSD.

no-quic

Don't build with QUIC support.

no-threads

Don't build with support for multi-threaded applications.

threads

Build with support for multi-threaded applications. Most platforms will enable this by default. However, if on a platform where this is not the case then this will usually require additional system-dependent options!

See Notes on multi-threading below.

no-thread-pool

Don't build with support for thread pool functionality.

thread-pool

Build with thread pool functionality. If enabled, OpenSSL algorithms may use the thread pool to perform parallel computation. This option in itself does not enable OpenSSL to spawn new threads. Currently the only supported thread pool mechanism is the default thread pool.

no-default-thread-pool

Don't build with support for default thread pool functionality.

default-thread-pool

Build with default thread pool functionality. If enabled, OpenSSL may create and manage threads up to a maximum number of threads authorized by the application. Supported on POSIX compliant platforms and Windows.

enable-trace

Build with support for the integrated tracing api.

See manual pages OSSL_trace_set_channel(3) and OSSL_trace_enabled(3) for details.

no-ts

Don't build Time Stamping (TS) Authority support.

enable-ubsan

Build with the Undefined Behaviour sanitiser (UBSAN).

This is a developer option only. It may not work on all platforms and should never be used in production environments. It will only work when used with gcc or clang and should be used in conjunction with the -DPEDANTIC option (or the --strict-warnings option).

no-ui-console

Don't build with the User Interface (UI) console method

The User Interface console method enables text based console prompts.

enable-unit-test

Enable additional unit test APIs.

This should not typically be used in production deployments.

Don't build support for UPLINK interface.

enable-weak-ssl-ciphers

Build support for SSL/TLS ciphers that are considered "weak"

Enabling this includes for example the RC4 based ciphersuites.

zlib

Build with support for zlib compression/decompression.

zlib-dynamic

Like the zlib option, but has OpenSSL load the zlib library dynamically when needed.

This is only supported on systems where loading of shared libraries is supported.

enable-zstd

Build with support for Zstd compression/decompression.

enable-zstd-dynamic

Like the enable-zstd option, but has OpenSSL load the Zstd library dynamically when needed.

This is only supported on systems where loading of shared libraries is supported.

enable-unstable-qlog

Enables qlog output support for the QUIC protocol. This functionality is unstable and implements a draft version of the qlog specification. The qlog output from OpenSSL will change in incompatible ways in future, and is not subject to any format stability or compatibility guarantees at this time. See the manpage openssl-qlog(7) for details.

386

In 32-bit x86 builds, use the 80386 instruction set only in assembly modules

The default x86 code is more efficient, but requires at least an 486 processor. Note: This doesn't affect compiler generated code, so this option needs to be accompanied by a corresponding compiler-specific option.

no-{protocol}

no-{ssl|ssl3|tls|tls1|tls1_1|tls1_2|tls1_3|dtls|dtls1|dtls1_2}

Don't build support for negotiating the specified SSL/TLS protocol.

If no-tls is selected then all of tls1, tls1_1, tls1_2 and tls1_3 are disabled. Similarly no-dtls will disable dtls1 and dtls1_2. The no-ssl option is synonymous with no-ssl3. Note this only affects version negotiation. OpenSSL will still provide the methods for applications to explicitly select the individual protocol versions.

no-integrity-only-ciphers

Don't build support for integrity only ciphers in tls.

no-{protocol}-method

no-{ssl3|tls1|tls1_1|tls1_2|dtls1|dtls1_2}-method

Analogous to no-{protocol} but in addition do not build the methods for applications to explicitly select individual protocol versions. Note that there is no no-tls1_3-method option because there is no application method for TLSv1.3.

Using individual protocol methods directly is deprecated. Applications should use TLS_method() instead.

enable-{algorithm}

enable-{md2|rc5}

Build with support for the specified algorithm.

no-{algorithm}

no-{aria|bf|blake2|camellia|cast|chacha|cmac|
    des|dh|dsa|ecdh|ecdsa|idea|md4|mdc2|ocb|
    poly1305|rc2|rc4|rmd160|scrypt|seed|
    siphash|siv|sm2|sm3|sm4|whirlpool}

Build without support for the specified algorithm.

The ripemd algorithm is deprecated and if used is synonymous with rmd160.

Compiler-specific options

-Dxxx, -Ixxx, -Wp, -lxxx, -Lxxx, -Wl, -rpath, -R, -framework, -static

These system specific options will be recognised and passed through to the compiler to allow you to define preprocessor symbols, specify additional libraries, library directories or other compiler options. It might be worth noting that some compilers generate code specifically for processor the compiler currently executes on. This is not necessarily what you might have in mind, since it might be unsuitable for execution on other, typically older, processor. Consult your compiler documentation.

Take note of the Environment Variables documentation below and how these flags interact with those variables.

-xxx, +xxx, /xxx

Additional options that are not otherwise recognised are passed through as they are to the compiler as well. Unix-style options beginning with a - or + and Windows-style options beginning with a / are recognised. Again, consult your compiler documentation.

If the option contains arguments separated by spaces, then the URL-style notation %20 can be used for the space character in order to avoid having to quote the option. For example, -opt%20arg gets expanded to -opt arg. In fact, any ASCII character can be encoded as %xx using its hexadecimal encoding.

Take note of the Environment Variables documentation below and how these flags interact with those variables.

Environment Variables

VAR=value

Assign the given value to the environment variable VAR for Configure.

These work just like normal environment variable assignments, but are supported on all platforms and are confined to the configuration scripts only. These assignments override the corresponding value in the inherited environment, if there is one.

The following variables are used as "make variables" and can be used as an alternative to giving preprocessor, compiler and linker options directly as configuration. The following variables are supported:

AR              The static library archiver.
ARFLAGS         Flags for the static library archiver.
AS              The assembler compiler.
ASFLAGS         Flags for the assembler compiler.
CC              The C compiler.
CFLAGS          Flags for the C compiler.
CXX             The C++ compiler.
CXXFLAGS        Flags for the C++ compiler.
CPP             The C/C++ preprocessor.
CPPFLAGS        Flags for the C/C++ preprocessor.
CPPDEFINES      List of CPP macro definitions, separated
                by a platform specific character (':' or
                space for Unix, ';' for Windows, ',' for
                VMS).  This can be used instead of using
                -D (or what corresponds to that on your
                compiler) in CPPFLAGS.
CPPINCLUDES     List of CPP inclusion directories, separated
                the same way as for CPPDEFINES.  This can
                be used instead of -I (or what corresponds
                to that on your compiler) in CPPFLAGS.
HASHBANGPERL    Perl invocation to be inserted after '#!'
                in public perl scripts (only relevant on
                Unix).
LD              The program linker (not used on Unix, $(CC)
                is used there).
LDFLAGS         Flags for the shared library, DSO and
                program linker.
LDLIBS          Extra libraries to use when linking.
                Takes the form of a space separated list
                of library specifications on Unix and
                Windows, and as a comma separated list of
                libraries on VMS.
RANLIB          The library archive indexer.
RC              The Windows resource compiler.
RCFLAGS         Flags for the Windows resource compiler.
RM              The command to remove files and directories.

These cannot be mixed with compiling/linking flags given on the command line. In other words, something like this isn't permitted.

$ ./Configure -DFOO CPPFLAGS=-DBAR -DCOOKIE

Backward compatibility note:

To be compatible with older configuration scripts, the environment variables are ignored if compiling/linking flags are given on the command line, except for the following:

AR, CC, CXX, CROSS_COMPILE, HASHBANGPERL, PERL, RANLIB, RC, and WINDRES

For example, the following command will not see -DBAR:

$ CPPFLAGS=-DBAR ./Configure -DCOOKIE

However, the following will see both set variables:

$ CC=gcc CROSS_COMPILE=x86_64-w64-mingw32- ./Configure -DCOOKIE

If CC is set, it is advisable to also set CXX to ensure both the C and C++ compiler are in the same "family". This becomes relevant with enable-external-tests and enable-buildtest-c++.

Reconfigure

reconf
reconfigure

Reconfigure from earlier data.

This fetches the previous command line options and environment from data saved in configdata.pm and runs the configuration process again, using these options and environment. Note: NO other option is permitted together with reconf. Note: The original configuration saves away values for ALL environment variables that were used, and if they weren't defined, they are still saved away with information that they weren't originally defined. This information takes precedence over environment variables that are defined when reconfiguring.

Displaying configuration data

The configuration script itself will say very little, and finishes by creating configdata.pm. This perl module can be loaded by other scripts to find all the configuration data, and it can also be used as a script to display all sorts of configuration data in a human readable form.

For more information, please do:

$ ./configdata.pm --help                         # Unix

or

$ perl configdata.pm --help                      # Windows and VMS

Installation Steps in Detail

Configure OpenSSL

Automatic Configuration

In previous version, the config script determined the platform type and compiler and then called Configure. Starting with version 3.0, they are the same.

Unix / Linux / macOS

$ ./Configure [options...]

OpenVMS

$ perl Configure [options...]

Windows

$ perl Configure [options...]

Manual Configuration

OpenSSL knows about a range of different operating system, hardware and compiler combinations. To see the ones it knows about, run

$ ./Configure LIST                               # Unix

or

$ perl Configure LIST                            # All other platforms

For the remainder of this text, the Unix form will be used in all examples. Please use the appropriate form for your platform.

Pick a suitable name from the list that matches your system. For most operating systems there is a choice between using cc or gcc. When you have identified your system (and if necessary compiler) use this name as the argument to Configure. For example, a linux-elf user would run:

$ ./Configure linux-elf [options...]

Creating your own Configuration

If your system isn't listed, you will have to create a configuration file named Configurations/YOURFILENAME.conf (replace YOURFILENAME with a filename of your choosing) and add the correct configuration for your system. See the available configs as examples and read Configurations/README.md and Configurations/README-design.md for more information.

The generic configurations cc or gcc should usually work on 32 bit Unix-like systems.

Configure creates a build file (Makefile on Unix, makefile on Windows and descrip.mms on OpenVMS) from a suitable template in Configurations/, and defines various macros in include/openssl/configuration.h (generated from include/openssl/configuration.h.in.

If none of the generated build files suit your purpose, it's possible to write your own build file template and give its name through the environment variable BUILDFILE. For example, Ninja build files could be supported by writing Configurations/build.ninja.tmpl and then configure with BUILDFILE set like this (Unix syntax shown, you'll have to adapt for other platforms):

$ BUILDFILE=build.ninja perl Configure [options...]

Out of Tree Builds

OpenSSL can be configured to build in a build directory separate from the source code directory. It's done by placing yourself in some other directory and invoking the configuration commands from there.

Unix example

$ mkdir /var/tmp/openssl-build
$ cd /var/tmp/openssl-build
$ /PATH/TO/OPENSSL/SOURCE/Configure [options...]

OpenVMS example

$ set default sys$login:
$ create/dir [.tmp.openssl-build]
$ set default [.tmp.openssl-build]
$ perl D:[PATH.TO.OPENSSL.SOURCE]Configure [options...]

Windows example

$ C:
$ mkdir \temp-openssl
$ cd \temp-openssl
$ perl d:\PATH\TO\OPENSSL\SOURCE\Configure [options...]

Paths can be relative just as well as absolute. Configure will do its best to translate them to relative paths whenever possible.

Build OpenSSL

Build OpenSSL by running:

$ make                                           # Unix
$ mms                                            ! (or mmk) OpenVMS
$ nmake                                          # Windows

This will build the OpenSSL libraries (libcrypto.a and libssl.a on Unix, corresponding on other platforms) and the OpenSSL binary (openssl). The libraries will be built in the top-level directory, and the binary will be in the apps/ subdirectory.

If the build fails, take a look at the Build Failures subsection of the Troubleshooting section.

Test OpenSSL

After a successful build, and before installing, the libraries should be tested. Run:

$ make test                                      # Unix
$ mms test                                       ! OpenVMS
$ nmake test                                     # Windows

Warning: you MUST run the tests from an unprivileged account (or disable your privileges temporarily if your platform allows it).

See test/README.md for further details how run tests.

See test/README-dev.md for guidelines on adding tests.

Install OpenSSL

If everything tests ok, install OpenSSL with

$ make install                                   # Unix
$ mms install                                    ! OpenVMS
$ nmake install                                  # Windows

Note that in order to perform the install step above you need to have appropriate permissions to write to the installation directory.

The above commands will install all the software components in this directory tree under <PREFIX> (the directory given with --prefix or its default):

Unix / Linux / macOS

bin/           Contains the openssl binary and a few other
               utility scripts.
include/openssl
               Contains the header files needed if you want
               to build your own programs that use libcrypto
               or libssl.
lib            Contains the OpenSSL library files.
lib/engines    Contains the OpenSSL dynamically loadable engines.

share/man/man1 Contains the OpenSSL command line man-pages.
share/man/man3 Contains the OpenSSL library calls man-pages.
share/man/man5 Contains the OpenSSL configuration format man-pages.
share/man/man7 Contains the OpenSSL other misc man-pages.

share/doc/openssl/html/man1
share/doc/openssl/html/man3
share/doc/openssl/html/man5
share/doc/openssl/html/man7
               Contains the HTML rendition of the man-pages.

OpenVMS

'arch' is replaced with the architecture name, ALPHA or IA64, 'sover' is replaced with the shared library version (0101 for 1.1), and 'pz' is replaced with the pointer size OpenSSL was built with:

[.EXE.'arch']  Contains the openssl binary.
[.EXE]         Contains a few utility scripts.
[.include.openssl]
               Contains the header files needed if you want
               to build your own programs that use libcrypto
               or libssl.
[.LIB.'arch']  Contains the OpenSSL library files.
[.ENGINES'sover''pz'.'arch']
               Contains the OpenSSL dynamically loadable engines.
[.SYS$STARTUP] Contains startup, login and shutdown scripts.
               These define appropriate logical names and
               command symbols.
[.SYSTEST]     Contains the installation verification procedure.
[.HTML]        Contains the HTML rendition of the manual pages.

Additional Directories

Additionally, install will add the following directories under OPENSSLDIR (the directory given with --openssldir or its default) for you convenience:

certs          Initially empty, this is the default location
               for certificate files.
private        Initially empty, this is the default location
               for private key files.
misc           Various scripts.

The installation directory should be appropriately protected to ensure unprivileged users cannot make changes to OpenSSL binaries or files, or install engines. If you already have a pre-installed version of OpenSSL as part of your Operating System it is recommended that you do not overwrite the system version and instead install to somewhere else.

Package builders who want to configure the library for standard locations, but have the package installed somewhere else so that it can easily be packaged, can use

$ make DESTDIR=/tmp/package-root install         # Unix
$ mms/macro="DESTDIR=TMP:[PACKAGE-ROOT]" install ! OpenVMS

The specified destination directory will be prepended to all installation target paths.

Compatibility issues with previous OpenSSL versions

COMPILING existing applications

Starting with version 1.1.0, OpenSSL hides a number of structures that were previously open. This includes all internal libssl structures and a number of EVP types. Accessor functions have been added to allow controlled access to the structures' data.

This means that some software needs to be rewritten to adapt to the new ways of doing things. This often amounts to allocating an instance of a structure explicitly where you could previously allocate them on the stack as automatic variables, and using the provided accessor functions where you would previously access a structure's field directly.

Some APIs have changed as well. However, older APIs have been preserved when possible.

Post-installation Notes

With the default OpenSSL installation comes a FIPS provider module, which needs some post-installation attention, without which it will not be usable. This involves using the following command:

$ openssl fipsinstall

See the openssl-fipsinstall(1) manual for details and examples.

Advanced Build Options

Environment Variables

A number of environment variables can be used to provide additional control over the build process. Typically these should be defined prior to running Configure. Not all environment variables are relevant to all platforms.

AR
               The name of the ar executable to use.

BUILDFILE
               Use a different build file name than the platform default
               ("Makefile" on Unix-like platforms, "makefile" on native Windows,
               "descrip.mms" on OpenVMS).  This requires that there is a
               corresponding build file template.
               See [Configurations/README.md](Configurations/README.md)
               for further information.

CC
               The compiler to use. Configure will attempt to pick a default
               compiler for your platform but this choice can be overridden
               using this variable. Set it to the compiler executable you wish
               to use, e.g. gcc or clang.

CROSS_COMPILE
               This environment variable has the same meaning as for the
               "--cross-compile-prefix" Configure flag described above. If both
               are set then the Configure flag takes precedence.

HASHBANGPERL
               The command string for the Perl executable to insert in the
               #! line of perl scripts that will be publicly installed.
               Default: /usr/bin/env perl
               Note: the value of this variable is added to the same scripts
               on all platforms, but it's only relevant on Unix-like platforms.

KERNEL_BITS
               This can be the value `32` or `64` to specify the architecture
               when it is not "obvious" to the configuration. It should generally
               not be necessary to specify this environment variable.

NM
               The name of the nm executable to use.

OPENSSL_LOCAL_CONFIG_DIR
               OpenSSL comes with a database of information about how it
               should be built on different platforms as well as build file
               templates for those platforms. The database is comprised of
               ".conf" files in the Configurations directory.  The build
               file templates reside there as well as ".tmpl" files. See the
               file [Configurations/README.md](Configurations/README.md)
               for further information about the format of ".conf" files
               as well as information on the ".tmpl" files.
               In addition to the standard ".conf" and ".tmpl" files, it is
               possible to create your own ".conf" and ".tmpl" files and
               store them locally, outside the OpenSSL source tree.
               This environment variable can be set to the directory where
               these files are held and will be considered by Configure
               before it looks in the standard directories.

PERL
               The name of the Perl executable to use when building OpenSSL.
               Only needed if building should use a different Perl executable
               than what is used to run the Configure script.

RANLIB
               The name of the ranlib executable to use.

RC
               The name of the rc executable to use. The default will be as
               defined for the target platform in the ".conf" file. If not
               defined then "windres" will be used. The WINDRES environment
               variable is synonymous to this. If both are defined then RC
               takes precedence.

WINDRES
               See RC.

Makefile Targets

The Configure script generates a Makefile in a format relevant to the specific platform. The Makefiles provide a number of targets that can be used. Not all targets may be available on all platforms. Only the most common targets are described here. Examine the Makefiles themselves for the full list.

all
               The target to build all the software components and
               documentation.

build_sw
               Build all the software components.
               THIS IS THE DEFAULT TARGET.

build_docs
               Build all documentation components.

clean
               Remove all build artefacts and return the directory to a "clean"
               state.

depend
               Rebuild the dependencies in the Makefiles. This is a legacy
               option that no longer needs to be used since OpenSSL 1.1.0.

install
               Install all OpenSSL components.

install_sw
               Only install the OpenSSL software components.

install_docs
               Only install the OpenSSL documentation components.

install_man_docs
               Only install the OpenSSL man pages (Unix only).

install_html_docs
               Only install the OpenSSL HTML documentation.

install_fips
               Install the FIPS provider module configuration file.

list-tests
               Prints a list of all the self test names.

test
               Build and run the OpenSSL self tests.

uninstall
               Uninstall all OpenSSL components.

reconfigure
reconf
               Re-run the configuration process, as exactly as the last time
               as possible.

update
               This is a developer option. If you are developing a patch for
               OpenSSL you may need to use this if you want to update
               automatically generated files; add new error codes or add new
               (or change the visibility of) public API functions. (Unix only).

Running Selected Tests

You can specify a set of tests to be performed using the make variable TESTS.

See the section Running Selected Tests of test/README.md.

Troubleshooting

Configuration Problems

Selecting the correct target

The ./Configure script tries hard to guess your operating system, but in some cases it does not succeed. You will see a message like the following:

$ ./Configure
Operating system: x86-whatever-minix
This system (minix) is not supported. See file INSTALL.md for details.

Even if the automatic target selection by the ./Configure script fails, chances are that you still might find a suitable target in the Configurations directory, which you can supply to the ./Configure command, possibly after some adjustment.

The Configurations/ directory contains a lot of examples of such targets. The main configuration file is 10-main.conf, which contains all targets that are officially supported by the OpenSSL team. Other configuration files contain targets contributed by other OpenSSL users. The list of targets can be found in a Perl list my %targets = ( ... ).

my %targets = (
...
"target-name" => {
    inherit_from     => [ "base-target" ],
    CC               => "...",
    cflags           => add("..."),
    asm_arch         => '...',
    perlasm_scheme   => "...",
},
...
)

If you call ./Configure without arguments, it will give you a list of all known targets. Using grep, you can lookup the target definition in the Configurations/ directory. For example the android-x86_64 can be found in Configurations/15-android.conf.

The directory contains two README files, which explain the general syntax and design of the configuration files.

If you need further help, try to search the openssl-users mailing list or the GitHub Issues for existing solutions. If you don't find anything, you can raise an issue to ask a question yourself.

More about our support resources can be found in the SUPPORT file.

Configuration Errors

If the ./config or ./Configure command fails with an error message, read the error message carefully and try to figure out whether you made a mistake (e.g., by providing a wrong option), or whether the script is working incorrectly. If you think you encountered a bug, please raise an issue on GitHub to file a bug report.

Along with a short description of the bug, please provide the complete configure command line and the relevant output including the error message.

Note: To make the output readable, please add a 'code fence' (three backquotes ``` on a separate line) before and after your output:

 ```
 ./Configure [your arguments...]

 [output...]

 ```

Build Failures

If the build fails, look carefully at the output. Try to locate and understand the error message. It might be that the compiler is already telling you exactly what you need to do to fix your problem.

There may be reasons for the failure that aren't problems in OpenSSL itself, for example if the compiler reports missing standard or third party headers.

If the build succeeded previously, but fails after a source or configuration change, it might be helpful to clean the build tree before attempting another build. Use this command:

$ make clean                                     # Unix
$ mms clean                                      ! (or mmk) OpenVMS
$ nmake clean                                    # Windows

Assembler error messages can sometimes be sidestepped by using the no-asm configuration option. See also notes.

Compiling parts of OpenSSL with gcc and others with the system compiler will result in unresolved symbols on some systems.

If you are still having problems, try to search the openssl-users mailing list or the GitHub Issues for existing solutions. If you think you encountered an OpenSSL bug, please raise an issue to file a bug report. Please take the time to review the existing issues first; maybe the bug was already reported or has already been fixed.

Test Failures

If some tests fail, look at the output. There may be reasons for the failure that isn't a problem in OpenSSL itself (like an OS malfunction or a Perl issue).

You may want increased verbosity, that can be accomplished as described in section Test Failures of test/README.md.

You may also want to selectively specify which test(s) to perform. This can be done using the make variable TESTS as described in section Running Selected Tests of test/README.md.

If you find a problem with OpenSSL itself, try removing any compiler optimization flags from the CFLAGS line in the Makefile and run make clean; make or corresponding.

To report a bug please open an issue on GitHub, at https://github.com/openssl/openssl/issues.

Notes

Notes on multi-threading

For some systems, the OpenSSL Configure script knows what compiler options are needed to generate a library that is suitable for multi-threaded applications. On these systems, support for multi-threading is enabled by default; use the no-threads option to disable (this should never be necessary).

On other systems, to enable support for multi-threading, you will have to specify at least two options: threads, and a system-dependent option. (The latter is -D_REENTRANT on various systems.) The default in this case, obviously, is not to include support for multi-threading (but you can still use no-threads to suppress an annoying warning message from the Configure script.)

OpenSSL provides built-in support for two threading models: pthreads (found on most UNIX/Linux systems), and Windows threads. No other threading models are supported. If your platform does not provide pthreads or Windows threads then you should use Configure with the no-threads option.

For pthreads, all locks are non-recursive. In addition, in a debug build, the mutex attribute PTHREAD_MUTEX_ERRORCHECK is used. If this is not available on your platform, you might have to add -DOPENSSL_NO_MUTEX_ERRORCHECK to your Configure invocation. (On Linux PTHREAD_MUTEX_ERRORCHECK is an enum value, so a built-in ifdef test cannot be used.)

Notes on shared libraries

For most systems the OpenSSL Configure script knows what is needed to build shared libraries for libcrypto and libssl. On these systems the shared libraries will be created by default. This can be suppressed and only static libraries created by using the no-shared option. On systems where OpenSSL does not know how to build shared libraries the no-shared option will be forced and only static libraries will be created.

Shared libraries are named a little differently on different platforms. One way or another, they all have the major OpenSSL version number as part of the file name, i.e. for OpenSSL 1.1.x, 1.1 is somehow part of the name.

On most POSIX platforms, shared libraries are named libcrypto.so.1.1 and libssl.so.1.1.

on Cygwin, shared libraries are named cygcrypto-1.1.dll and cygssl-1.1.dll with import libraries libcrypto.dll.a and libssl.dll.a.

On Windows build with MSVC or using MingW, shared libraries are named libcrypto-1_1.dll and libssl-1_1.dll for 32-bit Windows, libcrypto-1_1-x64.dll and libssl-1_1-x64.dll for 64-bit x86_64 Windows, and libcrypto-1_1-ia64.dll and libssl-1_1-ia64.dll for IA64 Windows. With MSVC, the import libraries are named libcrypto.lib and libssl.lib, while with MingW, they are named libcrypto.dll.a and libssl.dll.a.

On VMS, shareable images (VMS speak for shared libraries) are named ossl$libcrypto0101_shr.exe and ossl$libssl0101_shr.exe. However, when OpenSSL is specifically built for 32-bit pointers, the shareable images are named ossl$libcrypto0101_shr32.exe and ossl$libssl0101_shr32.exe instead, and when built for 64-bit pointers, they are named ossl$libcrypto0101_shr64.exe and ossl$libssl0101_shr64.exe.

Notes on random number generation

Availability of cryptographically secure random numbers is required for secret key generation. OpenSSL provides several options to seed the internal CSPRNG. If not properly seeded, the internal CSPRNG will refuse to deliver random bytes and a "PRNG not seeded error" will occur.

The seeding method can be configured using the --with-rand-seed option, which can be used to specify a comma separated list of seed methods. However, in most cases OpenSSL will choose a suitable default method, so it is not necessary to explicitly provide this option. Note also that not all methods are available on all platforms. The FIPS provider will silently ignore seed sources that were not validated.

I) On operating systems which provide a suitable randomness source (in form of a system call or system device), OpenSSL will use the optimal available method to seed the CSPRNG from the operating system's randomness sources. This corresponds to the option --with-rand-seed=os.

II) On systems without such a suitable randomness source, automatic seeding and reseeding is disabled (--with-rand-seed=none) and it may be necessary to install additional support software to obtain a random seed and reseed the CSPRNG manually. Please check out the manual pages for RAND_add(), RAND_bytes(), RAND_egd(), and the FAQ for more information.

Notes on assembler modules compilation

Compilation of some code paths in assembler modules might depend on whether the current assembler version supports certain ISA extensions or not. Code paths that use the AES-NI, PCLMULQDQ, SSSE3, and SHA extensions are always assembled. Apart from that, the minimum requirements for the assembler versions are shown in the table below:

ISA extension GNU as nasm llvm
AVX 2.19 2.09 3.0
AVX2 2.22 2.10 3.1
ADCX/ADOX 2.23 2.10 3.3
AVX512 2.25 2.11.8 3.6 (*)
AVX512IFMA 2.26 2.11.8 6.0 (*)
VAES 2.30 2.13.3 6.0 (*)

(*) Even though AVX512 support was implemented in llvm 3.6, prior to version 7.0 an explicit -march flag was apparently required to compile assembly modules. But then the compiler generates processor-specific code, which in turn contradicts the idea of performing dispatch at run-time, which is facilitated by the special variable OPENSSL_ia32cap. For versions older than 7.0, it is possible to work around the problem by forcing the build procedure to use the following script:

#!/bin/sh
exec clang -no-integrated-as "$@"

instead of the real clang. In which case it doesn't matter what clang version is used, as it is the version of the GNU assembler that will be checked.