OpenSSL 3.0

https://wiki.openssl.org/index.php/OpenSSL_3.0

by ddb • 6 years ago

243 104 6 years ago

wiki.openssl.org

One of the major improvements here: this finalizes the license change to Apache 2.0, which makes OpenSSL finally GPL-compatible. That removes one of the major reasons people had to avoid it.

(Specifically, OpenSSL is now compatible with anything licensed "GPLv3", "GPLv3 or later", or "GPLv2 or later". It's not compatible with "GPLv2 only", but that's a relatively small amount of software.)

Other major improvements: TLS1.3 support, Linux kernel TLS support (hand off crypto to the kernel and then read/write as though you had a normal file descriptor and let the kernel handle the crypto), and opaque low-level structures (no more dependencies on OpenSSL internals).

LeoPanthera• 6 years ago

Isn't the Linux kernel "GPLv2 only"?

JoshTriplett5• 6 years ago

Yes, but it's unlikely to link in a userspace library. This does prevent adapting code from OpenSSL to put in the kernel, but there are plenty of other sources to adapt code from.

There are a few other notable codebases that are GPLv2-only, but most projects using the GPL use either "v2 or later" or "v3 or later".

wallacoloo• 6 years ago

Other major improvements: TLS1.3 support

OpenSSL 1.1.1 already supports TLS 1.3

NiekvdMaas• 6 years ago

What does the license change mean for LibreSSL (OpenSSL fork) and GnuTLS which have "a better license" as their main selling point?

throwaway2048• 6 years ago

libressl does not have a different license

jabl8• 6 years ago

The openbsd people behind libressl hate the apache 2.0 license, so they won't be able to copy such code from the openssl project any longer. I'm sure they're happy they have a TLS library with a license they like (new contributions are licensed under the openbsd license) and they can continue developing.

As for gnutls, apache 2.0 license is still incompatible with (L)GPL 2.x, so projects under those licenses without the "or later" clause can't use openssl 3.x->. And of course, the gnutls and openssl API's are different, users can't just link against one or the other without code changes.

bjoli• 6 years ago

A bit of a digression: the Apache license includes better patent language, which got me thinking of the patents on ocb mode. Shouldn't they expire sometime early in this decade? Like next year or something?

wbl• 6 years ago

Rogaway has licensed them to any TLS implementation for free.

ta177117712• 6 years ago

...wasn't the major reason to avoid it the shitton of vulns it has had compared to libreSSL!?

toyg• 6 years ago

The “shitton of vulnerabilities” is a recent phenomenon, it’s not what prevented people from using OpenSSL.

terrywang• 6 years ago

- Linux Kernel TLS support

Is this referring to kernel TLS offload (using linux kernel's TLS connection ooffload infrastrucuture)?

shoo• 6 years ago

i am unfamiliar with kernel TLS offload [1], but superficially, it looks like that may be supported:

https://github.com/openssl/openssl/blob/43a70f02022ebbc29aa7...

`SOL_TLS` referenced here

https://github.com/openssl/openssl/blob/f059e4cc435b7b850cfc...

as defined here:

https://github.com/torvalds/linux/commit/3c4d7559159bfe1e3b9...

[1] https://www.kernel.org/doc/html/latest/networking/tls-offloa...

aisio• 6 years ago

Yes it supports Kernel TLS offload support. Both in SW mode (where the kernel does the TLS operations) and HW mode https://www.kernel.org/doc/html/latest/networking/tls-offloa...

scottlamb• 6 years ago

Does anyone know if OpenSSL-derived projects like BoringSSL and ring will follow suit in switching to Apache 2.0? And if so, have a guess at how long that will take?

tpolzer• 6 years ago

Doesn't "GPLv2 or later" specifically allow you to fork into "GPLv2 only"? How can a license be incompatible with only the latter then?

CrLf4• 6 years ago

IANAL, but “GPLv2 or later” allows you (the recipient of the license) to choose either GPLv2 or GPLv3 (i.e. the one that’s more convenient to you) but does not allow you to prevent others (the recipients of your modified version) from having the same choice.

This goes both ways, “GPLv2 or later” cannot be changed to either GPLv2 or GPLv3 only without permission from everybody that has ever contributed to the codebase.

The FSF requires copyright attribution from cotributors, that’s why they were able to switch their projects to GPLv3-only.

yc123403• 6 years ago

How can a license be incompatible with only the latter then?

Your parent comment said no such thing. It says, that you need GPL v3 to be compatible with Apache 2. "Or later" clause on GPL 2-licensed project offers you a way to re-license the project under GPL v3.

andrewaylett2• 6 years ago

This is where some BSD folks get frustrated by some GPL folks. Compiling code under BSD with code under GPL into the same binary is not _relicensing_. Each section of code still has the same license, and any binaries must abide by the intersection of the licenses. It is indeed the case that the intersection of GPL and BSD is equivalent to GPL and therefore allowed by the GPL, but the BSD license remains on the portions (of code and of binary) that are BSD licensed.

chrisweekly• 6 years ago

Thanks Josh, that's a helpful summary!

gray_-_wolf3• 6 years ago

Specifically, OpenSSL is now compatible with anything licensed "GPLv3", "GPLv3 or later", or "GPLv2 or later". It's not compatible with "GPLv2 only", but that's a relatively small amount of software.

Does this mean I cannot use openssl in my gpl-2.0-only program? How does that work? Doesn't gpl-2.0-later imply that I can also take the code and use it in gpl-2.0-only?

Is there some tldr on this topic somewhere? I'm not good with these legal things to ELI5 would be nice :/

jabl8• 6 years ago

Does this mean I cannot use openssl in my gpl-2.0-only program?

Correct.

How does that work?

The GPL 2.0 and Apache-2.0 licenses contain terms which are incompatible with each other.

Doesn't gpl-2.0-later imply that I can also take the code and use it in gpl-2.0-only?

Yes, in that case you can choose whether you use the code subject to the gpl 2.0, 3.0 or any later version. If you combine that code with some gpl-2.0-only code then you're choosing to use the first code under 2.0, and the combined work is then gpl-2.0-only.

Is there some tldr on this topic somewhere? I'm not good with these legal things to ELI5 would be nice :/

The GNU project maintains a list of licenses and some comments about them, and conveniently categorizes them according to their GPL compatibility.

https://www.gnu.org/licenses/license-list.html

See also the chart at http://gplv3.fsf.org/dd3-faq for compatibility between the different GPL versions and variants.

NCommander• 6 years ago

If you are the copyright holder, you can put a OpenSSL exception clause in your license which allows this to work. Debian requires such a clause for GPL software to link with OpenSSL prior to this change.

2OEH8eoCRo0• 6 years ago

It's incredible how much time and energy is spent on what license things use.

flatiron• 6 years ago

But think of what the gpl has given to us!

wcoenen• 6 years ago

TLS1.3 support

TLS 1.3 support was already added in OpenSSL 1.1.1

https://www.openssl.org/news/openssl-1.1.1-notes.html

Mizza2• 6 years ago

I worry that this is going to break so many programs and scripts in the same way that the switch from Linux 2.6 did..

pilif• 6 years ago

That already happened once with OpenSSL 1.1 which also wasn’t really backwards compatible.

It was messy then, it will be messy now, though when you took the 1.1 opportunity to modernize your code to current best-practice as requested by the library rather than just fixing the minimum, you might be pretty ok this time around

Mizza2• 6 years ago

You know, I think it was a deep memory of the switch off of 0.9.8 that gave me the initial thought without realizing. Maybe the older programs will be more prepared after all. Although I hope they weren't expecting a single-digit increment..

jeltz• 6 years ago

This release actually seems less painful than 1.1 which broke a lot of APIs.

kroeckx2• 6 years ago

The plan at least is that at little as possible should break. If you do find a problem, please file an issue.

macintux• 6 years ago

From other threads it sounds like the licensing may turn out to be a larger pain point than any backwards incompatibility in the code.

saghm• 6 years ago

Obvious question that it don't see answered on the page: why not 2.0?

dharmab• 6 years ago

https://www.openssl.org/blog/blog/2018/11/28/version/

They want to use the same version numbers for OpenSSL and a FIPS module, and the FIPS module was already versioned at 2.X.X.

moonchild• 6 years ago

OpenSSL versions with the same major number are API and ABI compatible

Finally!

A proper HTTP(S) client in libcrypto supporting GET and POST, redirection, plain and ASN.1-encoded contents, proxies, and timeouts

Is this really necessary? If you want a 'real' http client, you're probably using libcurl anyway (which is permissively licensed, more stable, and supports http/3).

JoshTriplett5• 6 years ago

Is this really necessary? If you want a 'real' http client, you're probably using libcurl anyway (which is permissively licensed, more stable, and supports http/3).

I'd guess it's there to support cryptographic protocols that require downloading/checking keys via HTTPS. Right before that line is "Implementation of the Certificate Management Protocol (CMP, RFC 4210) also covering CRMF (RFC 4211) and HTTP transfer (RFC 6712)".

And yes, use curl.

hannob• 6 years ago

Up until now OCSP checks were basically broken in openssl for all OCSP servers which speak only HTTP/1.1 (which is most of them) and only worked with a crude hack where you add an extra HTTP header via command line.

Not sure if this fixes it, but if you include functionality that relies on HTTP then you better support HTTP.

Tobu• 6 years ago

HTTPS is a really huge scope to tackle, and requires a lot of policy which OpenSSL traditionally hasn't encoded (how to root trust, for example — possibly involving system stores or custom ones, stapling, pinning…). Also, OpenSSL generally is embedded in HTTP clients, and having distinct implementations of HTTP calling into each other from the same library seems terrible, with potential for security issues related to policy or implementation mismatches.

Of course, there's all the traditional bug surface of any code that talks to the network to consider as well.

Defining an interface for calling back into another HTTP library might be doable, but there's still a question of scope creep.

The motivation seems to be this fork and pull request:

- https://github.com/mpeylo/cmpossl

- https://github.com/openssl/openssl/issues/5926

The use case is far away from what most users of OpenSSL need, and could quite easily be tackled outside OpenSSL.

woodruffw• 6 years ago

There's simply no way anybody will confuse OpenSSL 3.0, SSL 3, and TLS 1.3 :-)

tumetab1• 6 years ago

Came to the comments to try understand why SSL 3 was on HN.

This link should be named OpenSSL release 3.0

erwan• 6 years ago

I'm confused as to why they didn't skip a version release; for clarity's sake.

jamesog• 6 years ago

There was a proposal, which was declined: https://github.com/openssl/openssl/pull/8367

hyperman12• 6 years ago

Reading chapter 3, upgrading from 1.1.1, it seems strange for a security library to promote ignoring and even suppresing warnings. The best option, upgrading, is mentioned last.

I would prefer my security libraries to scream bloody murder if I am using them in a deprecated way.

saagarjha• 6 years ago

“Deprecated” need not mean “insecure”.

hyperman12• 6 years ago

Are you willing to take the risk? Deprecated might mean: There are known weaknesses. Or: Less reviewed code path.

ensiferum3• 6 years ago

As important as openssl is for many projects I find the engineering quality of it to be lackluster. My project is still on an older version and I wanted to upgrade. I tried to build approximately 7 versions for windows using msvs 2017 and msvs 2019. None even build but fail with compiler errors or linker errors!

On Linux if I go to system provided 1.2.x openssl version and try to build against that my code breaks with thousands of mysterious errors about undefined types (in openssl headers).

I mean Openssl solves a big problem but from engineering perspective the library is one giant problem. And I haven't even remarked on the horrible API design yet.

kakwa_• 6 years ago

I've mixed opinions over the recent breakages I've witnessed in OpenSSL.

On one hand, I absolutely hate a library which breaks its API (I think lack of stability in interfaces is one of the bigger bane of the software industry and causes numerous issues down the line).

But in the specific case with OpenSSL, the breakages are legitimate imho. Most of the breakages I have seen are to clean-up interfaces and have a clear boundary between OpenSSL internals and items controlled by third party apps.

Concretely, they are switching from an API which exposed the internal structures directly to a getter/setter pattern.

Example in from one of my project:

old:

  ts_response->tst_info->serial

new:

  TS_TST_INFO_get_serial(TS_RESP_get_tst_info(ts_response))

Breaking APIs is always a balancing act, it's not something which must happen willy-nilly, but it's sometimes a necessary evil, and in the specific case of OpenSSL, if the project wants to go forward and improve in quality and stability, they are kind of forced to do it, it's just a bit regrettable that better design decisions had not be made from the start.

ensiferum3• 6 years ago

I'm not even talking about clear cut API breakage. I'm talking about that if I do #include <openssl.h> and build against the 1.2.x system installed OpenSSL I will get thousands of errors from headers internal to OpenSSL.

blitmap3• 6 years ago

I don't have much familiarity with OpenSSL and crypto scares me away from reading the sources. I wish someone could give a full run-down of everything that is in OpenSSL, an overview. You hear all the time about it being bloated and supporting too many things. I wish I better understood that. It's why people turn to wolfssl and mbedtls, right?

Smaller projects that aim for minimalism and robustness probably suffer from a lack of peer review with a more niche community backing them. Trade-offs trade-offs. I also wish I understood where they compete:

OpenSSL <-> WolfSSL <-> mbedtls

als0• 6 years ago

It's why people turn to wolfssl and mbedtls, right?

I tried to use a single algorithm from OpenSSL for an embedded project and seems like it needs hacking for all the dependencies to be met. I gave up. With mbedTLS it was done within a minute (simpler to build and read IMO).

There aren't many differences between mbedTLS and WolfSSL. Both are small libraries designed for embedded use. The latter supports TLS 1.3.

Today OpenSSL probably has the best support for hardware acceleration and secure elements.

blitmap3• 6 years ago

I had not considered that it may have support for hardware accelerated crypto.

oefrha• 6 years ago

The documentation already has an overview: https://www.openssl.org/docs/manmaster/man7/

blitmap3• 6 years ago

That goes a long way, thank you :-) One of the other comments was saying OpenSSL probably has the best support for secure elements (hardware accelerated crypto?).

stock_toaster2• 6 years ago

don't forget BearSSL[1]!

[1] https://bearssl.org

mrweasel• 6 years ago

There is a fascinating and funny talk by Bob Beck from OpenBSD on the first 30 days after forking OpenSSL. One of the things he addresses is the scary code and how is prevents community involvement. He also talks a great deal about all that stuff that's not really relevant anymore and how much they removed from the LibreSSL source code.

Apparently the crypto modules are pretty well made according to the OpenBSD developers.

LibreSSL: The first 30 days, and what the Future Holds: https://www.youtube.com/watch?v=oM6S7FEUfkU

aorth2• 6 years ago

1.7 Versioning Scheme
The OpenSSL versioning scheme has changed with the 3.0 release. The new versioning scheme has this format:
MAJOR.MINOR.PATCH

Yay. OpenSSL finally switches to semantic versioning. \o/

NikolaeVarius2• 6 years ago

That isn't semver

aorth2• 6 years ago

What do you mean? According to https://semver.org/:

Given a version number MAJOR.MINOR.PATCH, increment the:
MAJOR version when you make incompatible API changes, MINOR version when you add functionality in a backwards compatible manner, and PATCH version when you make backwards compatible bug fixes.
Additional labels for pre-release and build metadata are available as extensions to the MAJOR.MINOR.PATCH format.

Seems like semantic versioning to me! At least the major, minor, and patch will mean something now.

saghul• 6 years ago

Pity that the bits required for QUIC didn't make the cut: https://github.com/openssl/openssl/pull/8797

OpenSSL 3.0

Related Stories