Input validation at trust boundaries

Goal

Ensure that only well-formed, expected data reaches business logic and storage, so that whole classes of injection and logic errors cannot occur.

Prerequisites

A clear map of trust boundaries: HTTP requests, message queues, files, environment, and data from other services all count as untrusted.

Steps

1. Define a schema for each input (types, required fields, lengths, patterns, enumerations) and validate against it before any other processing; typed models such as Pydantic or JSON Schema make the rules explicit.
2. Use allow-lists (what is permitted) rather than deny-lists (what is forbidden), as the OWASP guidance recommends.
3. Reject invalid input with a structured error that names the location and rule, not the offending value.
4. Validate semantically in the domain layer: referential existence, state transitions, quotas.
5. Encode on output for the target context (HTML, SQL parameters, shell arguments) instead of stripping characters on input; validation and output encoding are separate defences.
6. Enforce size limits at the transport layer to bound parsing cost.

Expected result

Malformed requests fail fast with clear errors; downstream code can assume shapes; logs and error messages do not echo attacker-controlled content.

Limits and test basis

Validation does not replace authorisation or output encoding. Free-text fields cannot be fully validated; they must be bounded and encoded. The steps follow the cited cheat sheet.