Aidan Keefe e0fd0c677f X509_verify_cert: honor explicit depth 0, don't count anchor against depth ...

Fenrir 12740: X509_STORE_CTX_set_depth(ctx, 0) was treated as unset
and silently got WOLFSSL_X509_STORE_DEFAULT_MAX_DEPTH.
- set_depth() now always marks the depth as explicit
  so 0 means no intermediate CAs are allowed.
- An explicit negative depth rejects every chain with
  CERT_CHAIN_TOO_LONG before any path is built, instead of falling
  back to the default.
- X509_STORE_CTX_init() sets the depth to the default chain max depth.
- A positive ctx->depth written directly to the struct (no setter) is
  still honored. The budget is clamped so INT_MAX cannot overflow it.

A trust anchor from X509_STORE_CTX_trusted_stack() or store->certs is
found by the same lookup as an untrusted intermediate and cost one extra
pass, so explicit depth N only allowed N-1 intermediates there (and
depth 0 rejected a leaf issued directly by the anchor). When the budget
runs out on a caller-trusted anchor that would end the path (self-issued
or PARTIAL_CHAIN), accept it instead of reporting CHAIN_TOO_LONG.

An issuer cycle among the untrusted certs (A <- B <- A) was followed
until the depth budget ran out, ~2^31 signature checks at INT_MAX. The
untrusted-issuer lookup now skips certs already on the path, so the
cycle ends the search with UNABLE_TO_GET_ISSUER_CERT_LOCALLY. Adds
loop-a/loop-b/loop-leaf fixtures to untrusted_anchor/ for the test.

Also fetch the verify callback once in X509_verify_cert() and pass it
to X509StoreVerifyCert(), X509StoreCheckPathLen() and the host/IP
checks, and factor the PARTIAL_CHAIN test into
X509StoreCertIsPartialChain().

2026-09-30 10:48:28 -06:00

3.7 KiB

Raw Permalink Blame History

View Raw