EVP_DigestUpdate chunked hashing or EVP_DigestEx single-call for paginated datasets

Direct Answer

Use a single EVP_MD_CTX for the entire dataset. Call EVP_DigestInit_ex once, then EVP_DigestUpdate for each page, then EVP_DigestFinal_ex once. Do not create a fresh context per page and do not call EVP_MD_CTX_reset between pages of the same digest operation. EVP_DigestEx is a single-shot convenience function that requires the full input buffer; it cannot process paginated input incrementally.

Why This Works

• Streaming design: The EVP API is built for exactly this pattern — one context, many Update calls, one Final. Each Update processes the page buffer in place with no extra copies.
• Constant memory: Only the context (a few hundred bytes) and one page buffer are live at any time.
• Minimal CPU overhead: Each byte is hashed once. No per-page allocation, no context re-initialization, no reset overhead.

OpenSSL 3.x vs 1.1.1 Reset Semantics

EVP_MD_CTX_reset() (3.x) and EVP_MD_CTX_cleanup() (1.1.1) clear algorithm-specific state and any keyed-hash keys but keep the allocated context structure. They are intended only when you finish one complete digest and want to start a completely new, unrelated digest with the same context object. Between Update calls of the same digest stream, do not call either function.

When to Use Separate Contexts

If your pagination logic interleaves multiple independent digest operations (e.g., hashing several files concurrently), allocate one EVP_MD_CTX per digest stream. Do not share a single context and reset it between files — separate contexts are simpler and avoid any cross-contamination risk.

Minimal Verification Checklist

1. Compile a test that hashes a known multi-page input (e.g., 10 MiB in 64 KiB chunks) using one context with repeated Update.
2. Compare the final digest to openssl dgst -sha256 <file> or a single-shot EVP_DigestEx call on the full buffer.
3. Run under Valgrind or ASAN to confirm no leaks from missing EVP_MD_CTX_free.

One Diagnostic Detail Needed

Are you hashing a single logical dataset across pages, or multiple independent datasets that happen to be read with the same pagination loop? The answer determines whether you need one context or several — but in either case, you never reset between pages of the same dataset.