requirements: Add initial support for uploading PEP 740 attestations

Signed-off-by: William Woodruff <[email protected]>

Dockerfile

@@ -27,6 +27,7 @@ COPY LICENSE.md .
 COPY twine-upload.sh .
 COPY print-hash.py .
 COPY oidc-exchange.py .
+COPY attestations.py .
 
 RUN chmod +x twine-upload.sh
 ENTRYPOINT ["/app/twine-upload.sh"]

README.md

@@ -246,6 +246,25 @@ for example. See [Creating & using secrets]. While still secure,
 [trusted publishing] is now encouraged over API tokens as a best practice
 on supported platforms (like GitHub).
 
+### Generating and uploading attestations (EXPERIMENTAL)
+
+> [!NOTE]
+> Support for generating and uploading [PEP 740 attestations] is currently
+> experimental and limited only to Trusted Publishing flows using PyPI or TestPyPI.
+
+You can generate signed [PEP 740 attestations] for all the distribution files and
+upload them all together by enabling the `attestations` setting:
+
+```yml
+   with:
+     attestations: true
+```
+
+This will use `sigstore` to create attestation objects for each distribution package,
+signing them with the identity provided by the GitHub's OIDC token associated with the
+current workflow. This means both the trusted publishing authentication and the
+attestations are tied to the same identity.
+
 ## License
 
 The Dockerfile and associated scripts and documentation in this project
@@ -287,3 +306,5 @@ https://github.com/vshymanskyy/StandWithUkraine/blob/main/docs/README.md
 [configured on PyPI]: https://docs.pypi.org/trusted-publishers/adding-a-publisher/
 
 [how to specify username and password]: #specifying-a-different-username
+
+[PEP 740 attestations]: https://peps.python.org/pep-0740/

action.yml

@@ -80,6 +80,13 @@ inputs:
       Use `print-hash` instead.
     required: false
     default: 'false'
+  attestations:
+    description: >-
+      [EXPERIMENTAL]
+      Enable experimental support for PEP 740 attestations.
+      Only works with PyPI and TestPyPI via Trusted Publishing.
+    required: false
+    default: 'false'
 branding:
   color: yellow
   icon: upload-cloud
@@ -95,3 +102,4 @@ runs:
   - ${{ inputs.skip-existing }}
   - ${{ inputs.verbose }}
   - ${{ inputs.print-hash }}
+  - ${{ inputs.attestations }}

attestations.py

@@ -0,0 +1,102 @@
+import logging
+import os
+import sys
+from pathlib import Path
+from typing import NoReturn
+
+from pypi_attestation_models import AttestationPayload
+from sigstore.oidc import IdentityError, IdentityToken, detect_credential
+from sigstore.sign import Signer, SigningContext
+
+# Be very verbose.
+sigstore_logger = logging.getLogger("sigstore")
+sigstore_logger.setLevel(logging.DEBUG)
+sigstore_logger.addHandler(logging.StreamHandler())
+
+_GITHUB_STEP_SUMMARY = Path(os.getenv("GITHUB_STEP_SUMMARY"))
+
+# The top-level error message that gets rendered.
+# This message wraps one of the other templates/messages defined below.
+_ERROR_SUMMARY_MESSAGE = """
+Attestation generation failure:
+
+{message}
+
+You're seeing this because the action attempted to generated PEP 740
+attestations for its inputs, but failed to do so.
+"""
+
+# Rendered if OIDC identity token retrieval fails for any reason.
+_TOKEN_RETRIEVAL_FAILED_MESSAGE = """
+OpenID Connect token retrieval failed: {identity_error}
+
+This generally indicates a workflow configuration error, such as insufficient
+permissions. Make sure that your workflow has `id-token: write` configured
+at the job level, e.g.:
+
+```yaml
+permissions:
+  id-token: write
+```
+
+Learn more at https://docs.github.com/en/actions/deployment/security-hardening-your-deployments/about-security-hardening-with-openid-connect#adding-permissions-settings.
+"""
+
+
+def die(msg: str) -> NoReturn:
+    with _GITHUB_STEP_SUMMARY.open("a", encoding="utf-8") as io:
+        print(_ERROR_SUMMARY_MESSAGE.format(message=msg), file=io)
+
+    # HACK: GitHub Actions' annotations don't work across multiple lines naively;
+    # translating `\n` into `%0A` (i.e., HTML percent-encoding) is known to work.
+    # See: https://github.com/actions/toolkit/issues/193
+    msg = msg.replace("\n", "%0A")
+    print(f"::error::Attestation generation failure: {msg}", file=sys.stderr)
+    sys.exit(1)
+
+
+def debug(msg: str):
+    print(f"::debug::{msg}", file=sys.stderr)
+
+
+# pylint: disable=redefined-outer-name
+def attest_dist(dist: Path, signer: Signer) -> None:
+    # We are the publishing step, so there should be no pre-existing publish
+    # attestation. The presence of one indicates user confusion.
+    attestation_path = Path(f"{dist}.publish.attestation")
+    if attestation_path.is_file():
+        die(f"{dist} already has a publish attestation: {attestation_path}")
+
+    payload = AttestationPayload.from_dist(dist)
+    attestation = payload.sign(signer)
+
+    attestation_path.write_text(attestation.model_dump_json(), encoding="utf-8")
+    debug(f"saved publish attestation: {dist=} {attestation_path=}")
+
+
+packages_dir = Path(sys.argv[1])
+
+try:
+    # NOTE: audience is always sigstore.
+    oidc_token = detect_credential()
+    identity = IdentityToken(oidc_token)
+except IdentityError as identity_error:
+    # NOTE: We only perform attestations in trusted publishing flows, so we
+    # don't need to re-check for the "PR from fork" error mode, only
+    # generic token retrieval errors.
+    cause = _TOKEN_RETRIEVAL_FAILED_MESSAGE.format(identity_error=identity_error)
+    die(cause)
+
+# Collect all sdists and wheels.
+dists = [sdist.absolute() for sdist in packages_dir.glob("*.tar.gz")]
+dists.extend(whl.absolute() for whl in packages_dir.glob("*.whl"))
+
+with SigningContext.production().signer(identity, cache=True) as signer:
+    for dist in dists:
+        # This should never really happen, but some versions of GitHub's
+        # download-artifact will create a subdirectory with the same name
+        # as the artifact being downloaded, e.g. `dist/foo.whl/foo.whl`.
+        if not dist.is_file():
+            die(f"Path looks like a distribution but is not a file: {dist}")
+
+        attest_dist(dist, signer)

requirements/runtime.in

@@ -1,9 +1,14 @@
 twine
 
-# NOTE: Used to detect an ambient OIDC credential for OIDC publishing.
+# NOTE: Used to detect an ambient OIDC credential for OIDC publishing,
+# as well as PEP 740 attestations.
 id ~= 1.0
 
 # NOTE: This is pulled in transitively through `twine`, but we also declare
 # NOTE: it explicitly here because `oidc-exchange.py` uses it.
 # Ref: https://github.com/di/id
 requests
+
+# NOTE: Used to generate attestations.
+pypi-attestation-models == 0.0.2
+sigstore ~= 3.0.0

requirements/runtime.txt

@@ -6,20 +6,41 @@
 #
 annotated-types==0.6.0
     # via pydantic
+betterproto==2.0.0b6
+    # via sigstore-protobuf-specs
 certifi==2024.2.2
     # via requests
 cffi==1.16.0
     # via cryptography
 charset-normalizer==3.3.2
     # via requests
 cryptography==42.0.7
-    # via secretstorage
+    # via
+    #   pyopenssl
+    #   pypi-attestation-models
+    #   sigstore
+dnspython==2.6.1
+    # via email-validator
 docutils==0.21.2
     # via readme-renderer
+email-validator==2.1.1
+    # via pydantic
+grpclib==0.4.7
+    # via betterproto
+h2==4.1.0
+    # via grpclib
+hpack==4.0.0
+    # via h2
+hyperframe==6.0.1
+    # via h2
 id==1.4.0
-    # via -r runtime.in
+    # via
+    #   -r runtime.in
+    #   sigstore
 idna==3.7
-    # via requests
+    # via
+    #   email-validator
+    #   requests
 importlib-metadata==7.1.0
     # via twine
 jaraco-classes==3.4.0
@@ -28,10 +49,6 @@ jaraco-context==5.3.0
     # via keyring
 jaraco-functools==4.0.1
     # via keyring
-jeepney==0.8.0
-    # via
-    #   keyring
-    #   secretstorage
 keyring==25.2.1
     # via twine
 markdown-it-py==3.0.0
@@ -42,36 +59,72 @@ more-itertools==10.2.0
     # via
     #   jaraco-classes
     #   jaraco-functools
+multidict==6.0.5
+    # via grpclib
 nh3==0.2.17
     # via readme-renderer
 pkginfo==1.10.0
     # via twine
+platformdirs==4.2.2
+    # via sigstore
+pyasn1==0.6.0
+    # via sigstore
 pycparser==2.22
     # via cffi
 pydantic==2.7.1
-    # via id
+    # via
+    #   id
+    #   pypi-attestation-models
+    #   sigstore
+    #   sigstore-rekor-types
 pydantic-core==2.18.2
     # via pydantic
 pygments==2.18.0
     # via
     #   readme-renderer
     #   rich
+pyjwt==2.8.0
+    # via sigstore
+pyopenssl==24.1.0
+    # via sigstore
+pypi-attestation-models==0.0.2
+    # via -r runtime.in
+python-dateutil==2.9.0.post0
+    # via betterproto
 readme-renderer==43.0
     # via twine
 requests==2.31.0
     # via
     #   -r runtime.in
     #   id
     #   requests-toolbelt
+    #   sigstore
+    #   tuf
     #   twine
 requests-toolbelt==1.0.0
     # via twine
 rfc3986==2.0.0
     # via twine
+rfc8785==0.1.2
+    # via sigstore
 rich==13.7.1
-    # via twine
-secretstorage==3.3.3
-    # via keyring
+    # via
+    #   sigstore
+    #   twine
+securesystemslib==1.0.0
+    # via tuf
+sigstore==3.0.0
+    # via
+    #   -r runtime.in
+    #   pypi-attestation-models
+sigstore-protobuf-specs==0.3.2
+    # via sigstore
+sigstore-rekor-types==0.0.13
+    # via sigstore
+six==1.16.0
+    # via python-dateutil
+tuf==5.0.0
+    # via sigstore
 twine==5.1.0
     # via -r runtime.in
 typing-extensions==4.11.0

twine-upload.sh

@@ -39,6 +39,7 @@ INPUT_PACKAGES_DIR="$(get-normalized-input 'packages-dir')"
 INPUT_VERIFY_METADATA="$(get-normalized-input 'verify-metadata')"
 INPUT_SKIP_EXISTING="$(get-normalized-input 'skip-existing')"
 INPUT_PRINT_HASH="$(get-normalized-input 'print-hash')"
+INPUT_ATTESTATIONS="$(get-normalized-input 'attestations')"
 
 PASSWORD_DEPRECATION_NUDGE="::error title=Password-based uploads disabled::\
 As of 2024, PyPI requires all users to enable Two-Factor \
@@ -53,6 +54,33 @@ environments like GitHub Actions without needing to use username/password \
 combinations or API tokens to authenticate with PyPI. Read more: \
 https://docs.pypi.org/trusted-publishers"
 
+ATTESTATIONS_WITHOUT_TP_WARNING="::warning title=attestations setting ignored::\
+The workflow was run with 'attestations: true', but an explicit password was \
+also supplied, disabling Trusted Publishing. As a result, the attestations \
+setting is ignored."
+
+ATTESTATIONS_WRONG_INDEX_WARNING="::warning title=attestations setting ignored::\
+The workflow was run with 'attestations: true', but the specified repository URL \
+does not support PEP 740 attestations. As a result, the attestations setting \
+is ignored."
+
+if [[ "${INPUT_ATTESTATIONS}" != "false" ]] ; then
+    # Setting `attestations: true` and explicitly passing a password indicates
+    # user confusion, since attestations (currently) require Trusted Publishing.
+    if [[ -n "${INPUT_PASSWORD}" ]] ; then
+        echo "${ATTESTATIONS_WITHOUT_TP_WARNING}"
+        INPUT_ATTESTATIONS="false"
+    fi
+
+    # Setting `attestations: true` with an index other than PyPI or TestPyPI
+    # indicates user confusion, since attestations are not supported on other
+    # indices presently.
+    if [[ ! "${INPUT_REPOSITORY_URL}" =~ pypi\.org ]] ; then
+        echo "${ATTESTATIONS_WRONG_INDEX_WARNING}"
+        INPUT_ATTESTATIONS="false"
+    fi
+fi
+
 if [[ "${INPUT_USER}" == "__token__" && -z "${INPUT_PASSWORD}" ]] ; then
     # No password supplied by the user implies that we're in the OIDC flow;
     # retrieve the OIDC credential and exchange it for a PyPI API token.
@@ -130,6 +158,15 @@ if [[ ${INPUT_VERBOSE,,} != "false" ]] ; then
     TWINE_EXTRA_ARGS="--verbose $TWINE_EXTRA_ARGS"
 fi
 
+if [[ ${INPUT_ATTESTATIONS,,} != "false" ]] ; then
+    # NOTE: Intentionally placed after `twine check`, to prevent attestation
+    # generation on distributions with invalid metadata.
+    echo "::debug::Generating and uploading PEP 740 attestations"
+    python /app/attestations.py "${INPUT_PACKAGES_DIR%%/}"
+
+    TWINE_EXTRA_ARGS="--attestations $TWINE_EXTRA_ARGS"
+fi
+
 if [[ ${INPUT_PRINT_HASH,,} != "false" || ${INPUT_VERBOSE,,} != "false" ]] ; then
     python /app/print-hash.py ${INPUT_PACKAGES_DIR%%/}
 fi