[christianaurichzm]

An RSA-PSS public key can fail to decode when the first OpenSSL fetch of the process happens on a thread that Node.js started, and from then on every RSA-PSS decode in the process fails:

const { X509Certificate, pbkdf2 } = require('crypto');
const pem = require('fs').readFileSync('test/fixtures/keys/rsa_pss_cert_2048.pem');
pbkdf2('', '', 1, 32, 'sha256', () => {});
new X509Certificate(pem).publicKey;

$ for i in $(seq 30); do ./node repro.js; done 2>&1 | grep -c ERR_OSSL_EVP_DECODE_ERROR
11

The same happens with require('tls'), which loads the root certificates on another thread, and with a Worker that hashes something as it starts. It is also the intermittent parallel/test-crypto-x509 failure on GitHub Actions since 697ecce added require('tls') to that test (1, 2, 3). [2] ran on v26.x-staging.

On the first fetch, OpenSSL 3 copies its legacy algorithm names into the name map (ossl_namemap_stored()). The empty check and the copy are not under one lock, so another thread can look names up while the copy is still running. The SPKI decoder looks the key type up by the long name of its OID (x_pubkey.c), which for RSA-PSS is rsassaPss, and the 3.5 default provider doesn't register that name. In that window it resolves to nothing, and the empty decoder set gets cached. Until #65796 the CSPRNG check at startup did the first fetch before any other thread existed, and v22.23.1 doesn't fail.

This adds EnsureOpenSSLNameMap(), which does one fetch on the calling thread, once per process. It runs before a crypto job is scheduled, before the certificate loading threads start and before a Worker thread starts. Startup still does no fetch. The first pbkdf2(), require('tls') or new Worker() gets 0.3 to 0.5 ms slower, once.

Over 3000 runs, 16 at a time, failures go from 832 to 0 with pbkdf2(), from 44 to 0 with require('tls') and from 264 to 0 with a Worker. test/parallel/test-crypto-off-thread-first-use.js fails on main and passes with this change, and parallel, sequential and cctest pass locally. Threads that a native addon starts itself are still not covered.

Refs: #65796

[nodejs-github-bot]

Review requested:

• @nodejs/crypto

[christianaurichzm]

@panva @jasnell, since you reviewed #65796, could you take a look at this? If it looks good, could you also start a Jenkins CI run? Thanks!

[panva]

@christianaurichzm see #66903, i think this isn't needed when we revert the PRs that triggered this.

[christianaurichzm]

Thanks, I hadn't seen #66903. I tried main with the revert and all three cases here pass (0 of 3000), so I'm closing this one.

The revert doesn't add a test for this though. Want me to send test-crypto-off-thread-first-use.js as a test-only PR after it lands?

[panva]

That would be nice, i also have a repro script in that PR's comment, see if you can combine yours with mine (if that makes sense), up to you.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.58%. Comparing base (53f037e) to head (38c9121).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #66935      +/-   ##
==========================================
- Coverage   90.59%   90.58%   -0.02%     
==========================================
  Files         797      797              
  Lines      278475   278482       +7     
  Branches    53533    53539       +6     
==========================================
- Hits       252293   252255      -38     
- Misses      16532    16560      +28     
- Partials     9650     9667      +17     

Files with missing lines	Coverage Δ
src/crypto/crypto_context.cc	72.49% <100.00%> (+0.04%)	⬆️
src/crypto/crypto_util.cc	51.54% <100.00%> (+0.06%)	⬆️
src/crypto/crypto_util.h	66.75% <100.00%> (+0.09%)	⬆️
src/node_worker.cc	81.89% <100.00%> (-0.48%)	⬇️

... and 29 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.