Skip to content

ocsp: replace stapling_mutex with std::shared_mutex and fix leaks#13226

Draft
c-taylor wants to merge 1 commit into
apache:masterfrom
c-taylor:ocsp-stapling-shared-mutex
Draft

ocsp: replace stapling_mutex with std::shared_mutex and fix leaks#13226
c-taylor wants to merge 1 commit into
apache:masterfrom
c-taylor:ocsp-stapling-shared-mutex

Conversation

@c-taylor
Copy link
Copy Markdown

@c-taylor c-taylor commented Jun 2, 2026

Replace the per-certinfo ink_mutex with a std::shared_mutex so readers on the TLS handshake hot path (ssl_callback_ocsp_stapling) and the refresh scan (ocsp_update) can run concurrently; only the cache update takes an exclusive lock. The handshake reader now copies the DER staple to a stack buffer under the shared lock and allocates outside the critical section.

Because the mutex is no longer a C type, certinfo is allocated with new/delete instead of OPENSSL_malloc.

Also fix three pre-existing bugs:

  • certinfo_map_free never freed cinf->cid (leak on every CTX teardown).
  • On BoringSSL the X509 key ref taken via X509_up_ref was never released; free it in certinfo_map_free.
  • ssl_stapling_init_cert's error path could delete a certinfo_map still owned by the SSL_CTX. Only delete a map this call created.

This is a rework of #13097 to separate into 2x independent PRs

std::shared_mutex vs. bravo

The lock pressure for new TLS connections is less than volume or other event/txn mutex. Favoring the reduced memory overhead in the implementation over perfect hypothetical performance. The majority of the benefit is realised from using ANY shared type.

I would suggest that should bravo be favored it become an isolated PR.

Replace the per-certinfo ink_mutex with a std::shared_mutex so readers
on the TLS handshake hot path (ssl_callback_ocsp_stapling) and the
refresh scan (ocsp_update) can run concurrently; only the cache update
takes an exclusive lock. The handshake reader now copies the DER staple
to a stack buffer under the shared lock and allocates outside the
critical section.

Because the mutex is no longer a C type, certinfo is allocated with
new/delete instead of OPENSSL_malloc.

Also fix three pre-existing bugs:
- certinfo_map_free never freed cinf->cid (leak on every CTX teardown).
- On BoringSSL the X509 key ref taken via X509_up_ref was never
  released; free it in certinfo_map_free.
- ssl_stapling_init_cert's error path could delete a certinfo_map still
  owned by the SSL_CTX. Only delete a map this call created.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant