fix(auth): handle PermissionError on workload certificates to avoid startup hang and crash by nbayati · Pull Request #17568 · googleapis/google-cloud-python · GitHub

In sandboxed environments (such as when running the gcloud CLI within a snap package manager sandbox), AppArmor rules can block reading the workload credentials directory or certificate files, raising a PermissionError.

Previously, os.path.exists() caught this PermissionError and returned False. The library interpreted this as the files not being ready yet (e.g. due to a startup race) and entered a 30-second retry loop before failing with a RefreshError.

This change:

Updates _is_certificate_file_ready to use os.stat and propagate PermissionError.
Catches PermissionError immediately during certificate lookup, logging a warning and falling back to unbound tokens without retrying or crashing.
Adds corresponding unit tests.

Fixes b/522957573

Code Review

This pull request improves error handling when checking and accessing agent identity certificates. Specifically, it updates _is_certificate_file_ready to propagate PermissionError while catching other OSError exceptions, and handles PermissionError in get_agent_identity_certificate_path by logging a warning and falling back to unbound tokens. Corresponding unit tests have been added to verify these scenarios. There are no review comments, so I have no additional feedback to provide.

Thank you for this contribution!

System-level / Off-diff Finding:

Graceful OSError Handling in get_and_parse_agent_identity_certificate()

get_and_parse_agent_identity_certificate() (around line 208) does not handle exceptions when reading the certificate file. Because os.path.getsize only requires directory traversal / metadata search permissions, _is_certificate_file_ready can evaluate to True even for a file that the current process does not have read permissions to open (e.g., owned by root with 600). When open(cert_path, "rb") is called, it will crash with a raw PermissionError.

Wrap the file-read in a try...except OSError block to gracefully fail-closed with a controlled RefreshError:

    try:
        with open(cert_path, "rb") as cert_file:
            cert_bytes = cert_file.read()
    except OSError as e:
        raise exceptions.RefreshError(
            f"Failed to read agent identity certificate file at {cert_path}: {e}. "
            "Token binding protection is failing. You can turn off this protection by setting "
            f"{environment_vars.GOOGLE_API_PREVENT_AGENT_TOKEN_SHARING_FOR_GCP_SERVICES} to false "
            "to fall back to unbound tokens."
        ) from e

I think we may want to catch here too. I think it's possible for get_agent_identity_certificate_path to work correctly but then have .read to raise a PermissionError since .stat and .read require different permissions

Done! I have wrapped the file read block in a try...except OSError block to catch read-level permissions errors and fall back to unbound tokens.

nit: the PR description mentions using os.stat , but the code uses os.path.getsize . Functionally equivalent since getsize uses stat under the hood, but just pointing it out!

The new commit uses os.stat directly. I'll double check the PR description after making all the changes requested by reviewers to ensure it's still valid. Thanks for pointing it out :)

…tartup hang and crash In sandboxed environments (such as when running the gcloud CLI within a snap package manager sandbox), AppArmor rules can block reading the workload credentials directory or certificate files, raising a PermissionError. Previously, `os.path.exists()` caught this PermissionError and returned `False`. The library interpreted this as the files not being ready yet (e.g. due to a startup race) and entered a 30-second retry loop before failing with a RefreshError. This change: 1. Updates `_is_certificate_file_ready` to use `os.stat` and propagate `PermissionError`. 2. Catches `PermissionError` immediately during certificate lookup, logging a warning and falling back to unbound tokens without retrying or crashing. 3. Adds corresponding unit tests.

…ects

…ad errors

🤖 I have created a release *beep* *boop* --- <details><summary>db-dtypes: 1.7.1</summary> ## [1.7.1](db-dtypes-v1.7.0...db-dtypes-v1.7.1) (2026-07-07) ### Bug Fixes * avoid deprecated unitless operations for NumPy 2.5 compatibility ([#17589](#17589)) ([d0b2abc](d0b2abc)) </details> <details><summary>gapic-generator: 1.37.0</summary> ## [1.37.0](gapic-generator-v1.36.0...gapic-generator-v1.37.0) (2026-07-07) ### Features * implement native PEP 0810 lazy loading ([#17591](#17591)) ([8a1270c](8a1270c)) ### Bug Fixes * **deps:** bump google-api-core to 2.25.0 ([#17599](#17599)) ([8b359e2](8b359e2)) * **tests:** add --cov-append to gapic-generator and proto-plus to preserve monorepo coverage ([#17603](#17603)) ([2ddcf4d](2ddcf4d)) </details> <details><summary>google-auth: 2.55.2</summary> ## [2.55.2](google-auth-v2.55.1...google-auth-v2.55.2) (2026-07-07) ### Bug Fixes * **auth:** Agentic Identites mTLS gaps fix _is_mtls and SslCredentials. ([#17387](#17387)) ([7bfa41a](7bfa41a)) * **auth:** align mTLS discovery and enforce fail-fast transport configuration. ([#17470](#17470)) ([f492d3d](f492d3d)) * **auth:** handle PermissionError on workload certificates to avoid startup hang and crash ([#17568](#17568)) ([f538ad8](f538ad8)) </details> <details><summary>google-cloud-agentregistry: 0.1.0</summary> ## 0.1.0 (2026-07-07) ### Features * **google/cloud/agentregistry/v1:** add google-cloud-agentregistry ([#17565](#17565)) ([f479800](f479800)) </details> <details><summary>google-cloud-biglake-hive: 0.3.1</summary> ## [0.3.1](google-cloud-biglake-hive-v0.3.0...google-cloud-biglake-hive-v0.3.1) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-bigquery: 3.42.2</summary> ## [3.42.2](google-cloud-bigquery-v3.42.1...google-cloud-bigquery-v3.42.2) (2026-07-07) ### Bug Fixes * **bigquery:** avoid SSLError retry loop ([#17489](#17489)) ([8248d8e](8248d8e)) * include amended user agent in read client ([#17592](#17592)) ([c43caee](c43caee)) </details> <details><summary>google-cloud-binary-authorization: 1.18.0</summary> ## [1.18.0](google-cloud-binary-authorization-v1.17.0...google-cloud-binary-authorization-v1.18.0) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-chronicle: 0.6.2</summary> ## [0.6.2](google-cloud-chronicle-v0.6.1...google-cloud-chronicle-v0.6.2) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-dataform: 0.11.2</summary> ## [0.11.2](google-cloud-dataform-v0.11.1...google-cloud-dataform-v0.11.2) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-dataproc: 5.30.0</summary> ## [5.30.0](google-cloud-dataproc-v5.29.0...google-cloud-dataproc-v5.30.0) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-datastore: 2.26.0</summary> ## [2.26.0](google-cloud-datastore-v2.25.0...google-cloud-datastore-v2.26.0) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-dialogflow: 2.50.0</summary> ## [2.50.0](google-cloud-dialogflow-v2.49.0...google-cloud-dialogflow-v2.50.0) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-kms: 3.15.0</summary> ## [3.15.0](google-cloud-kms-v3.14.0...google-cloud-kms-v3.15.0) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-logging: 3.16.1</summary> ## [3.16.1](google-cloud-logging-v3.16.0...google-cloud-logging-v3.16.1) (2026-07-07) ### Documentation * **logging:** fix StructuredLogHandler docstring parameter name ([#17625](#17625)) ([32862f0](32862f0)), closes [#17604](#17604) </details> <details><summary>google-cloud-modelarmor: 0.7.1</summary> ## [0.7.1](google-cloud-modelarmor-v0.7.0...google-cloud-modelarmor-v0.7.1) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-network-security: 0.13.3</summary> ## [0.13.3](google-cloud-network-security-v0.13.2...google-cloud-network-security-v0.13.3) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-saasplatform-saasservicemgmt: 0.7.1</summary> ## [0.7.1](google-cloud-saasplatform-saasservicemgmt-v0.7.0...google-cloud-saasplatform-saasservicemgmt-v0.7.1) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-securesourcemanager: 0.6.1</summary> ## [0.6.1](google-cloud-securesourcemanager-v0.6.0...google-cloud-securesourcemanager-v0.6.1) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>google-cloud-storage: 3.12.1</summary> ## [3.12.1](google-cloud-storage-v3.12.0...google-cloud-storage-v3.12.1) (2026-07-07) ### Bug Fixes * **storage:** log occasional (1 in 5 million) additional bytes received from GCS in read path ([#17423](#17423)) ([335c12f](335c12f)) </details> <details><summary>google-cloud-support: 0.5.1</summary> ## [0.5.1](google-cloud-support-v0.5.0...google-cloud-support-v0.5.1) (2026-07-07) ### Features * update googleapis and regenerate ([#17635](#17635)) ([9638879](9638879)) </details> <details><summary>proto-plus: 1.28.1</summary> ## [1.28.1](proto-plus-v1.28.0...proto-plus-v1.28.1) (2026-07-07) ### Bug Fixes * **tests:** add --cov-append to gapic-generator and proto-plus to preserve monorepo coverage ([#17603](#17603)) ([2ddcf4d](2ddcf4d)) </details> --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please). --------- Co-authored-by: release-please[bot] <55107282+release-please[bot]@users.noreply.github.com>

nbayati requested review from a team as code owners June 24, 2026 20:46

gemini-code-assist Bot reviewed Jun 24, 2026

View reviewed changes

lsirac reviewed Jun 24, 2026

View reviewed changes

macastelaz reviewed Jun 25, 2026

View reviewed changes

nbayati requested review from lsirac and macastelaz June 25, 2026 20:27

macastelaz approved these changes Jun 25, 2026

View reviewed changes

lsirac reviewed Jun 26, 2026

View reviewed changes

Comment thread packages/google-auth/google/auth/_agent_identity_utils.py Outdated

parthea approved these changes Jun 26, 2026

View reviewed changes

Comment thread packages/google-auth/google/auth/_agent_identity_utils.py Outdated

parthea assigned nbayati Jun 26, 2026

nbayati and others added 4 commits July 1, 2026 15:00

fix(auth): restrict certificate readiness check to regular files

f7628dd

test(auth): mock os.path.exists locally to avoid test runner side eff…

2f8f2bb

…ects

fix(auth): fallback to unbound tokens on certificate permission or re…

c9ecbd2

…ad errors

nbayati force-pushed the fix/auth-mtls-permission-denied branch from 022d2ae to cea7183 Compare July 1, 2026 22:24

fix(auth): change OSError to PermissionError and use warnings.warn

51b41f3

nbayati force-pushed the fix/auth-mtls-permission-denied branch from cea7183 to 51b41f3 Compare July 1, 2026 22:28

nbayati merged commit f538ad8 into googleapis:main Jul 2, 2026
30 checks passed

release-please Bot mentioned this pull request Jul 2, 2026

chore: release main #17600

Merged

nbayati mentioned this pull request Jul 14, 2026

feat: Enable Bound Token for Agentic Identities googleapis/google-cloud-java#13169

Closed

+                  if not path:
+                      return False
+                  try:
+                      return os.path.getsize(path) > 0

fix(auth): handle PermissionError on workload certificates to avoid startup hang and crash by nbayati · Pull Request #17568 · googleapis/google-cloud-python · GitHub

Uh oh!

fix(auth): handle PermissionError on workload certificates to avoid startup hang and crash - #17568

fix(auth): handle PermissionError on workload certificates to avoid startup hang and crash#17568
nbayati merged 5 commits into
googleapis:mainfrom
nbayati:fix/auth-mtls-permission-denied

nbayati commented Jun 24, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

lsirac left a comment •
edited
Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

macastelaz Jun 25, 2026

Uh oh!

nbayati Jun 25, 2026

Uh oh!

macastelaz Jun 25, 2026

Uh oh!

nbayati Jun 25, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

fix(auth): handle PermissionError on workload certificates to avoid startup hang and crash by nbayati · Pull Request #17568 · googleapis/google-cloud-python · GitHub

Uh oh!

fix(auth): handle PermissionError on workload certificates to avoid startup hang and crash - #17568

Conversation

nbayati commented Jun 24, 2026

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

lsirac left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

System-level / Off-diff Finding:

Graceful OSError Handling in get_and_parse_agent_identity_certificate()

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

macastelaz Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

nbayati Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

macastelaz Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

nbayati Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

lsirac left a comment •
edited
Loading