Parse Lambda AppSec request bodies according to their content type

[claponcet]

What Does This Do

Dispatches Lambda AppSec request bodies on their declared Content-Type instead of handing every body to a JSON parser:

Declared type	Reported to the WAF as
JSON-ish (application/json, any +json suffix, application/x-amz-json-1.1, application/javascript) or no usable type at all	best-effort JSON parse
application/x-www-form-urlencoded	multimap, the same shape already used for query parameters
multipart/*	form fields, each part dispatched on its own Content-Type; file parts contribute their filename, not their content
text/* and everything else	raw string

A body is never dropped. Any type we cannot structure, and any parse failure, degrades to the raw string, which the WAF can still match string rules against. In particular a text/plain body of 12345 must reach the WAF as a String and not as the Double a JSON parse would produce.

The same rule decides what happens to a malformed multipart body. A part whose headers run into the next delimiter is not reported part-by-part: no conforming parser accepts such a body — Commons FileUpload and Netty both reject it outright — so reporting the parts that happen to survive would show the WAF less than the application receives. The whole body degrades to the raw string instead.

Parsing is bounded by three allowances, shared across nesting levels rather than re-satisfied at each one: 1 MiB of characters read, 256 parts or parameters, and 20 levels of nesting. The depth and part limits mirror the WAF's own (WAFModule.MAX_DEPTH / MAX_ELEMENTS) — structure beyond them is discarded by ObjectIntrospection before the WAF sees it, so producing it would be wasted work. Exceeding an allowance degrades the body to a raw string rather than truncating it, since a truncated map would hide the dropped parameter from every rule.

Behavior changes worth a reviewer's eye. The JSON-ish decision is now made in one place, ContentTypeBodyParser.isJsonOrUntyped, which the response path in LambdaEventParser shares with the request path. It matches on the subtype alone, where the response path previously looked for json or javascript anywhere in the raw header. Three consequences:

• A Content-Type holding nothing but parameters — ; charset=utf-8 — counts as declaring no type, and so gets the same best-effort JSON parse an absent header gets. Pinned by a case in dispatchesOnContentType.
• A response body with a blank content-type now gets that best-effort JSON parse too. Previously only an absent header did, since "" does not contain json.
• A response body whose type carries the word in a parameter — multipart/form-data; boundary=--json — no longer reaches the JSON parser. This is why the gate matches on the subtype only: a client-chosen boundary must not decide how the body is read.

Motivation

Peer tracers already structure urlencoded and multipart Lambda bodies. Without it, a form-encoded or multipart request reaching a Lambda is visible to the WAF only as one opaque string, so any rule addressing server.request.body by key cannot match — the same request routed through a non-Lambda entrypoint would be inspected properly.

Additional Notes

MultipartSplitter no longer collects a HashMap of every header per part. Only Content-Disposition and Content-Type are ever read, so the others are matched and dropped as the scan passes them: a body that is nothing but header lines would otherwise allocate a map entry per line, which is why the change is here at all. The naive version of the fix — matching header names inline with a case-insensitive regionMatches — measured slower on well-formed many-part forms, so the canonical spelling is tried with a case-sensitive compare first and the matching lives in its own small method rather than inlined into split.

Two known flaky failures were seen in dd-trace-core while validating, both unrelated to this diff and both passing on re-run in isolation: PendingTraceBufferTest.testingTracerFlareDumpWithMultipleTraces and DDAgentWriterCombinedTest.statsdCommFailure. Neither is annotated @Flaky.

File parts populate server.request.body.filenames, which comes almost free once the parts are split. The other two file addresses are left empty on purpose: the Lambda event delivers the body as one UTF-8-decoded string, so a file's bytes are already corrupted by the time we see them, and files_content would report that damage as if it were the upload. files_field_names waits with it.

The repeated callback-registration boilerplate in LambdaAppSecHandler is left for its own change.

Contributor Checklist

• Format the title according to the contribution guidelines
• Assign the type: and (comp: or inst:) labels in addition to any other useful labels
• Avoid using close, fix, or any linking keywords when referencing an issue
  Use solves instead, and assign the PR milestone to the issue
• Update the CODEOWNERS file on source file addition, migration, or deletion
• Update public documentation with any new configuration flags or behaviors
• Once approved, use merge queue to merge the PR

Jira ticket: [PROJ-IDENT]

🤖 Generated with Claude Code

[claponcet]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ebad5227cd

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[dd-octo-sts]

🟢 Java Benchmark SLOs — All performance SLOs passed

Suite	Status
Startup	🟢 pass

SLO thresholds are defined here based on automatically generated metrics. A warning is raised when results are within 5% of the threshold.

PR vs. master results

Scenario	Candidate	master	Δ (95% CI of mean)
startup:insecure-bank:iast:Agent	13.99 s	13.90 s	[-0.1%; +1.4%] (no difference)
startup:insecure-bank:tracing:Agent	12.94 s	13.01 s	[-1.2%; +0.2%] (no difference)
startup:petclinic:appsec:Agent	17.00 s	16.83 s	[-0.0%; +2.0%] (no difference)
startup:petclinic:iast:Agent	16.95 s	17.02 s	[-1.1%; +0.3%] (no difference)
startup:petclinic:profiling:Agent	16.63 s	16.80 s	[-2.2%; +0.2%] (no difference)
startup:petclinic:sca:Agent	16.95 s	16.75 s	[+0.4%; +2.1%] (maybe worse)
startup:petclinic:tracing:Agent	16.18 s	16.18 s	[-1.0%; +1.1%] (no difference)

Commit: 693dfe0b · CI Pipeline · Benchmarking Platform UI

---

Load and DaCapo benchmarks can be triggered manually in the GitLab pipeline. Results will appear in the Benchmarking Platform UI after completion.

[claponcet]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 58ba1ec8bc

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 693dfe0be2

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Validate the initial delimiter before treating it as closing

When a multipart preamble begins with a line such as --x--not-a-close, this direct startsWith selects offset 0 without calling endsLine; the close check then stops parsing and never scans the later valid --x delimiter. The body consequently falls back to an opaque string, so key-addressed WAF rules can miss fields that a conforming multipart parser extracts after ignoring the preamble. The fresh evidence beyond the prior fix is this initial-boundary path, which still bypasses the validation now performed by nextDelimiter; validate its terminator as well.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve form values with empty parameter names

For a valid urlencoded body such as =payload&ok=1, this condition silently drops the first value but still returns the nonempty map containing ok, so the caller does not fall back to the raw body. The Lambda handler still receives and can decode the original empty-name parameter, while AppSec sees no trace of payload; this also differs from the existing Netty body collector, which retains attributes under data.getName() without rejecting an empty key. Preserve the empty-name entry or treat the whole parse as unusable so an attack value cannot disappear from WAF inspection.

Useful? React with 👍 / 👎.

[datadog-prod-us1-5]

The new body limit lets the WAF drop later form fields without a raw-body fallback. The multipart parser also treats a valid preamble near match as a close delimiter and misses later fields.

_{Open Bits AI session}

_{🤖 Datadog Autotest · Commit 693dfe0 · What is Autotest? · @DataDog review to ask questions · Any feedback? Reach out in #autotest}

[datadog-prod-us1-5]

WAF limit drops later form fields

An attacker can place a harmful value near the end and bypass key-based WAF rules.

Assertion details

• Input: A URL-encoded body with 85 to 256 one-value parameters, or a multipart body with more than 127 scalar fields.
• Expected: The parser should return the full raw body before the WAF can truncate any field.
• Actual: The parser accepts the structured map. ObjectIntrospection then counts the root, keys, lists, and values against its separate 256-element limit. It drops later fields.

Was this helpful? React 👍 or 👎
_{🤖 Datadog Autotest · What is Autotest? · @DataDog review to ask questions · Any feedback? Reach out in #autotest · Open Bits AI session}

[datadog-prod-us1-5]

Preamble near match stops multipart parsing

Key-based WAF rules can miss fields that the application parses.

Assertion details

• Input: A valid multipart body starts with a preamble line such as --boundary--not-a-close, followed by a real delimiter and form fields.
• Expected: The parser should ignore the preamble line and find the valid delimiter that follows it.
• Actual: The initial prefix check accepts the near match. The close-delimiter check then stops parsing. AppSec receives only the raw body.

Suggested change

	int position = body.startsWith(delimiter) ? 0 : nextDelimiter(body, delimiter, 0);
	int position = nextDelimiter(body, delimiter, 0);

Was this helpful? React 👍 or 👎
_{🤖 Datadog Autotest · What is Autotest? · @DataDog review to ask questions · Any feedback? Reach out in #autotest · Open Bits AI session}