Skip to content

gh-136728: Refactor build.yml CI config and multissltests.py #136729

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 12 commits into
base: main
Choose a base branch
from

Conversation

WillChilds-Klein
Copy link
Contributor

@WillChilds-Klein WillChilds-Klein commented Jul 16, 2025

Notes

Please see #136728. Changes to build.yml are directly adapted from @hugovk's proof-of-concept here.

Testing

  • this PR's CI
  • PR CI on my fork.

@WillChilds-Klein WillChilds-Klein marked this pull request as ready for review July 16, 2025 20:39
Copy link
Member

@picnixz picnixz left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If possible, I want multissltests to be refactored separately.

Comment on lines +273 to +278
- { os: ubuntu-24.04, ssl: openssl, ssl_ver: 3.0.16 }
- { os: ubuntu-24.04, ssl: openssl, ssl_ver: 3.1.8 }
- { os: ubuntu-24.04, ssl: openssl, ssl_ver: 3.2.4 }
- { os: ubuntu-24.04, ssl: openssl, ssl_ver: 3.3.3 }
- { os: ubuntu-24.04, ssl: openssl, ssl_ver: 3.4.1 }
- { os: ubuntu-24.04, ssl: awslc, ssl_ver: 1.55.0 }
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can remove the os from the include and have:

matrix:
  os: [ubuntu-24.04]
  include:
    - { ssl: openssl, ssl_ver: 3.0.16 }
    - ...

It would be possible to merge two matrices but we'll need a pre-step and I don't think it's more maintainable.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could we do ssl_lib: openssl | awslc? It would be easier to follow later on than just matrix.ssl everywhere.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm sorry but what would ssl_lib: openssl | awslc in this case? does it run on both and if it doesn't work for one, it picks the other? (also, how would we use it below?)

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@picnixz I'm not sure I follow, the suggestion is just to rename matrix.ssl to matrix.ssl_lib. openssl | awslc are the valid values for either name.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh! I thought you suggested something like this:

matrix:
  os: [ubuntu-24.04]
  include:
    - { ssl_lib: openssl | awslc, ssl_ver: 3.0.16 }

and I didn't know what it would means. Yes, we can have matrix.ssl_lib

echo "LD_LIBRARY_PATH=${GITHUB_WORKSPACE}/multissl/aws-lc/${AWSLC_VER}/lib" >> "$GITHUB_ENV"
- name: 'Restore AWS-LC build'
id: cache-aws-lc
echo "SSL_DIR=${GITHUB_WORKSPACE}/multissl/${{ matrix.ssl }}/${SSL_VER}" >> "$GITHUB_ENV"
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can't we just use ${{ env.ssldir }}? I don't understand why we're defining it here and in the env section

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

zizmor says using ${{ env.ssldir }} can result in template injection:

help[template-injection]: code injection via template expansion
   --> ./.github/workflows/build.yml:324:134
    |
323 |       run: |
    |       --- help: this run block
324 |         CMD=(./configure CFLAGS="-fdiagnostics-format=json" --config-cache --enable-slower-safety --with-pydebug --with-openssl="${{ env.ssldir }...
    |                                                                                                                                      ---------- help: may expand into attacker-controllable code

Copy link
Member

@picnixz picnixz Aug 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Isn't env a procteced namespace that cannot be accessed from outside? but if zizmor flags it as possibly vulnerable, let's be verbose.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's ask zizmor author @woodruffw.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, so zizmor kwnows that it's not attackable but recommends against it for shell expansion purposes: https://github.com/woodruffw/gha-hazmat/blob/e4094ceabb0068a2a5b81ecc8e1ee919ba52a050/.github/workflows/template-injection.yml#L77.

Copy link
Member

@picnixz picnixz Aug 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

However! we can do another trick https://github.com/woodruffw/gha-hazmat/blob/e4094ceabb0068a2a5b81ecc8e1ee919ba52a050/.github/workflows/template-injection.yml#L67-L72. I guess we could directly use the environment variable name then without needing to make GH variable expansion.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, I'd recommend using the environment variable directly instead of intermediating it through a template (e.g. $FOO instead of ${{ env.FOO }}). That should be equivalent in terms of intended behavior without the injection risk 🙂

The reason zizmor flags these is because they can be exploitable in some subtle situations: most of the time env.* is static and not attacker-controllable, but sometimes it comes from GITHUB_ENV or another dynamic source that zizmor can't follow.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Some more docs here too (the first tip in that section mentions why env.* is not preferred versus direct expansion): https://docs.zizmor.sh/audits/#template-injection

--base-directory "$MULTISSL_DIR" \
--awslc ${{ matrix.awslc_ver }} \
--system Linux
python3 Tools/ssl/multissltests.py --steps=library --base-directory "$MULTISSL_DIR" --system Linux --ssl ${{ matrix.ssl }} --ssl-versions ${{ matrix.ssl_ver }}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you make it multiline as before? please

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note we can use run: > syntax to avoid the trailing backslashes

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, it would be better. We don't care about literal new lines here.

if [ "${{ matrix.ssl }}" = "openssl" ]; then
"${CMD[@]}"
else
"${CMD[@]}" --with-builtin-hashlib-hashes=blake2 --with-ssl-default-suites=openssl
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we populate a variable called "CONFIGURE_FLAGS" instead and use a switch/case so that we exactly match the SSL libname. It will become easier in the future if we add BoringSSL or LibreSSL tests.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

At this point I'd almost suggest using a small Python script to calculate flags. We should avoid complex shell logic in workflow files as much as possible, I don't want to read a bash switch-case in YAML! An if-statement is simpler to understand here for me, though the ${CMD[@]} syntax is arcane.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh yeah, a small Python script is also fine; here what matters to me is to be able to easily add a new library without having 3km long of options. As for ${CMD[@]}, it expands the array's content, so you end up with ./configure [...]. Usually, it's more common to expand the options and hardcode the command rather than expand the entire array containing the command as well.

@@ -567,7 +507,7 @@ jobs:
key: ${{ matrix.os }}-multissl-openssl-${{ env.OPENSSL_VER }}
- name: Install OpenSSL
if: steps.cache-openssl.outputs.cache-hit != 'true'
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

With steps.cache-openssl here, we won't be hitting the cache-ssl entry right? maybe it's better to store a cache hit with steps.cache-${{ matrix.ssl }} if it's possible?

Comment on lines +143 to +146
@property
@abstractmethod
def library(self=None):
pass
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please revert this. As we can't have abstract class methods, I prefer having real class variables where we would raise if they are not set.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yep, better to raise NotImplementedError. That's treated equivalently to an abstract method.

Copy link
Member

@picnixz picnixz Aug 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it's just better to have abstract class variables here. The reason is that I don't think we would ever need a non-static computation for most of those abstract properties. If they must be abstract class properties that are represented by dynamic values that need a bit more work to compute (but only needed to be computed once), just a @classmethod together with NotImplementedError.

)

versions = []
ssl_libs = AbstractBuilder.__subclasses__()
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please, don't use this. It's harder to maintain if we do it like this. Just hardcode the classes.

assert len(libs) == 1
cls = libs.pop()
if args.ssl_versions:
versions += [(cls, v) for v in args.ssl_versions]
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Use extend instead of += [...]

format="*** %(levelname)s %(message)s"
)

versions = []
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You could have a dict of classes to a list of versions instead of storing a tuple of (class, version)

"AWS-LC versions, defaults to '{}' if no crypto library versions are given."
).format(AWSLC_RECENT_VERSIONS)
default=None,
help="SSL lib version(s), default depends on value passed to --ssl",
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
help="SSL lib version(s), default depends on value passed to --ssl",
help="SSL lib versions, default depends on libs passed to --ssl",

@@ -78,6 +58,7 @@
"versions."
),
)
parser.color = True
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is now the default, so we can skip this:

Suggested change
parser.color = True

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(Also, it wouldn't be effective if we were to use an older Python version)

echo "LD_LIBRARY_PATH=${GITHUB_WORKSPACE}/multissl/aws-lc/${AWSLC_VER}/lib" >> "$GITHUB_ENV"
- name: 'Restore AWS-LC build'
id: cache-aws-lc
echo "SSL_DIR=${GITHUB_WORKSPACE}/multissl/${{ matrix.ssl }}/${SSL_VER}" >> "$GITHUB_ENV"
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

zizmor says using ${{ env.ssldir }} can result in template injection:

help[template-injection]: code injection via template expansion
   --> ./.github/workflows/build.yml:324:134
    |
323 |       run: |
    |       --- help: this run block
324 |         CMD=(./configure CFLAGS="-fdiagnostics-format=json" --config-cache --enable-slower-safety --with-pydebug --with-openssl="${{ env.ssldir }...
    |                                                                                                                                      ---------- help: may expand into attacker-controllable code

build-ubuntu-ssltests-openssl:
name: 'Ubuntu SSL tests with OpenSSL'
build-ubuntu-ssltests:
name: 'Ubuntu SSL tests'
runs-on: ${{ matrix.os }}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All jobs run on the same OS, this is simpler:

Suggested change
runs-on: ${{ matrix.os }}
runs-on: ubuntu-24.04

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We use matrix.os in the cache key so better not hardcode it here?

Comment on lines +21 to +22
Please keep this script compatible with all currently-maintained Python
versions.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do we care about Python 3.9 at this point? Frequently for in-tree tools we use python_for_regen as the base version, or the previous stable release (e.g. 3.12 at present).

Copy link
Member

@picnixz picnixz Aug 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, we care because of security branches I'd say. Also, there are still distros with Pyhton 3.9 as its default I think (I mean, my default Python in OpenSUSE was Python 3.6!). But generally, it's to limit to "simple" Python when possible.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

True, but e.g. for backports to 3.9 we would run against the version of Tools/ssl/multissltests.py in the 3.9 branch? We should add a comment explaining why we care about all non EOL versions, though.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

True, but e.g. for backports to 3.9 we would run against the version of Tools/ssl/multissltests.py in the 3.9 branch?

That's... a good question. I would have said "yes", but now I'm not entirely sure. Maybe not?

Copy link
Member

@picnixz picnixz Aug 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah actually, we do:

    def run_python_tests(self, tests, network=True):
        if not tests:
            cmd = [
                sys.executable,
                os.path.join(PYTHONROOT, 'Lib/test/ssltests.py'),
                '-j0'
            ]
        elif sys.version_info < (3, 3):
            cmd = [sys.executable, '-m', 'test.regrtest']
        else:
            cmd = [sys.executable, '-m', 'test', '-j0']
        if network:
            cmd.extend(['-u', 'network', '-u', 'urlfetch'])
        cmd.extend(['-w', '-r'])
        cmd.extend(tests)
self._subprocess_call(cmd, stdout=None)

So for executing the tests, we're really re-using the python interpreter that was used to generate the scripts, so we need to run the tests against the correct interpreter version.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@WillChilds-Klein could you update run_python_tests() in this PR to remove the 3.3 block & use long options to regrtest? -w -> --rerun & -r -> --randomize

Comment on lines +28 to +29
from abc import abstractmethod
from abc import ABCMeta
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
from abc import abstractmethod
from abc import ABCMeta
from abc import ABCMeta, abstractmethod

"OpenSSL and LibreSSL versions are given."
).format(LIBRESSL_RECENT_VERSIONS, LIBRESSL_OLD_VERSIONS)
'--ssl',
choices=['openssl', 'awslc', 'libressl'],
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
choices=['openssl', 'awslc', 'libressl'],
choices=('openssl', 'awslc', 'libressl'),

"LibreSSL versions, defaults to '{}' (ancient: '{}') if no "
"OpenSSL and LibreSSL versions are given."
).format(LIBRESSL_RECENT_VERSIONS, LIBRESSL_OLD_VERSIONS)
'--ssl',
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Again I'd suggest --ssl-library or etc

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

7 participants