https://bz.apache.org/bugzilla/show_bug.cgi?id=70260
--- Comment #13 from Eric Covener <[email protected]> --- @steffen can you see if defining HAVE_OPENSSL_INIT_CRYPTO in apu_config.hw solves it? Claude analysis: Likely culprit: r1908584 on the 1.6.x branch (ylavic, 2023-03-20, "apr_crypto_openssl: Compatibility with OpenSSL 3+", merging trunk r1908433/r1908448/r1908453/r1908455/r1908503 and 1.7.x r1862838/r1908581). It is the only change between the 1.6.3 and 1.6.5 tags that alters what the apr_crypto_t context pointer holds. Why this commit. The crash is a read fault inside the thin wrapper at crypto/apr_crypto.c:400, which does nothing but f->provider->get_block_key_types(types, f). So either f is NULL or garbage. Only three commits touch crypto in the range, and the other two are not candidates: - r1936805 / r1936828 / r1937013 (CVE-2025-49506 timing-safe compares): only change apr_crypto_equals and apr_password_validate. No pointer or lifecycle changes. - r1916306: adds (void) to prototypes in apu_dso.c. No behavior change. The struct apr_crypto_t layout that the wrapper peeks at (pool, provider) is identical in both tags, so a DLL mismatch between libaprutil-1.dll and apr_crypto_openssl-1.dll would not explain this fault either. What r1908584 changed in crypto_make() in crypto/apr_crypto_openssl.c: - It now sets *ff = NULL at entry and only assigns the context after everything succeeds. In 1.6.3 *ff was assigned early, so a caller got a fully usable context (types and modes populated, cleanup registered) even when the function returned APR_ENOENGINE or APR_EINITENGINE. - New failure modes were added: an engine= parameter returns APR_ENOTIMPL when the OpenSSL ENGINE API is unavailable, and engine=openssl is now silently skipped instead of looked up. - Initialization switched to OPENSSL_init_crypto() when HAVE_OPENSSL_INIT_CRYPTO is defined. On Windows that macro is never defined because apu_config.hw is static, so the legacy init path still runs there. -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
