Finish multipart production hardening

Author: ubugeeei

.github/workflows/ci.yml

@@ -83,6 +83,27 @@ jobs:
           cargo test -p fastmulp-core --doc --locked
           cargo test -p fastmulp-wasm --doc --locked
 
+  bindings:
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v6
+      - uses: dtolnay/rust-toolchain@stable
+        with:
+          targets: wasm32-unknown-unknown
+      - uses: Swatinem/rust-cache@v2
+      - uses: actions/setup-node@v6
+        with:
+          node-version: "24"
+      - name: Enable pnpm
+        run: corepack enable
+      - name: Install wasm-bindgen CLI
+        run: bash scripts/install-wasm-bindgen.sh
+      - name: Node binding smoke test
+        run: pnpm test:node-binding
+      - name: Wasm binding smoke test
+        run: pnpm test:wasm-binding
+
   bench:
     if: github.event_name == 'workflow_dispatch'
     runs-on: ubuntu-latest

.github/workflows/release_bindings.yml

@@ -0,0 +1,73 @@
+name: release-bindings
+
+on:
+  push:
+    tags:
+      - "v*"
+
+permissions:
+  contents: write
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: false
+
+defaults:
+  run:
+    shell: bash
+
+env:
+  CARGO_TERM_COLOR: always
+
+jobs:
+  publish-binding-artifacts:
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    steps:
+      - uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+      - uses: dtolnay/rust-toolchain@stable
+        with:
+          targets: wasm32-unknown-unknown
+      - uses: Swatinem/rust-cache@v2
+      - uses: actions/setup-node@v6
+        with:
+          node-version: "24"
+      - name: Enable pnpm
+        run: corepack enable
+      - name: Verify tag is on main
+        run: git branch --remote --contains "${GITHUB_SHA}" | grep -qx '  origin/main'
+      - name: Verify tag matches workspace version
+        run: |
+          expected="${GITHUB_REF_NAME#v}"
+          actual="$(awk '
+            /^\[workspace\.package\]/ { in_section=1; next }
+            /^\[/ && in_section { exit }
+            in_section && /^version = "/ {
+              gsub(/^version = "/, "", $0)
+              gsub(/"$/, "", $0)
+              print
+              exit
+            }
+          ' Cargo.toml)"
+          test "${expected}" = "${actual}"
+      - name: Build and test Node binding
+        run: pnpm test:node-binding
+      - name: Install wasm-bindgen CLI
+        run: bash scripts/install-wasm-bindgen.sh
+      - name: Test wasm binding
+        run: pnpm test:wasm-binding
+      - name: Package binding artifacts
+        run: node scripts/package-bindings.mjs --pack
+      - name: Checksum artifacts
+        run: |
+          cd dist/bindings/artifacts
+          sha256sum *.tgz > SHA256SUMS
+      - name: Create or update GitHub release
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          gh release view "${GITHUB_REF_NAME}" >/dev/null 2>&1 \
+            || gh release create "${GITHUB_REF_NAME}" --verify-tag --title "${GITHUB_REF_NAME}"
+          gh release upload "${GITHUB_REF_NAME}" dist/bindings/artifacts/*.tgz dist/bindings/artifacts/SHA256SUMS --clobber

README.md

@@ -13,7 +13,7 @@ High-accuracy, low-allocation `multipart/form-data` parsing with a zero-copy Rus
 - The Rust core parses against a borrowed `&[u8]` and returns body ranges instead of copying payload bytes.
 - `Content-Disposition` is parsed eagerly, and `name` is enforced for `form-data` parts.
 - Header storage uses `SmallVec` so the common case stays stack-friendly.
-- Node.js and browser bindings return metadata plus `body_start` / `body_end`, so callers can slice the original buffer themselves.
+- Node.js and browser bindings return metadata plus body ranges, so callers can slice the original buffer themselves.
 - Boundary lines accept RFC 2046 transport padding, plus MIME-style preamble and epilogue.
 
 ## Spec Notes
@@ -51,15 +51,14 @@ Node.js:
 import { parse } from "./fastmulp.node";
 
 const parts = parse(bodyBuffer, boundary);
-const fileBytes = bodyBuffer.subarray(parts[0].body_start, parts[0].body_end);
+const fileBytes = bodyBuffer.subarray(parts[0].bodyStart, parts[0].bodyEnd);
 ```
 
 Browser:
 
 ```ts
-import init, { parse } from "./fastmulp_wasm.js";
+import { parse } from "fastmulp-wasm";
 
-await init();
 const parts = parse(formBytes, boundary);
 const fieldBytes = formBytes.subarray(parts[0].body_start, parts[0].body_end);
 ```
@@ -68,14 +67,36 @@ The wasm target still needs one JS-to-wasm copy at the ABI boundary, but it avoi
 
 Older nested `multipart/mixed` payloads can be handled by recursively calling `parse` on a part body after extracting the nested boundary from that part's `Content-Type`.
 
+### Parser Limits
+
+Use `parse_with_limits` or `MultipartParser::new_with_limits` when parsing untrusted uploads on a server boundary. `ParseLimits` can cap the number of parts, the number of headers per part, and the number of header bytes per part. The plain `parse` API stays unlimited for compatibility and small trusted payloads.
+
+### Security Notes
+
+Treat `filename` as display metadata only. Strip path separators and platform-specific path components before using it in storage, generate your own server-side object names, and keep the original value only as untrusted metadata. Preserve duplicate field names in order unless your application explicitly defines a merge rule. For nested `multipart/mixed`, extract the nested boundary from that part's `Content-Type` and parse the nested body with its own limits.
+
+### Binding Artifacts
+
+Tag pushes build npm-compatible release artifacts for the JS targets:
+
+- `fastmulp-node-linux-x64-*.tgz`: platform-specific Node.js native addon containing `fastmulp.node`, CommonJS entrypoint, and TypeScript declarations.
+- `fastmulp-wasm-*.tgz`: browser wasm package containing wasm-bindgen generated JavaScript glue, wasm, and TypeScript declarations.
+
+The Node binding uses camelCase object fields such as `bodyStart`, `bodyEnd`, `fileName`, and `contentType`. The wasm binding preserves the existing snake_case field names generated by `wasm-bindgen`.
+
+```ts
+import { parse as parseNode } from "fastmulp-node-linux-x64";
+import { parse as parseWasm } from "fastmulp-wasm";
+```
+
 ## Release
 
 - `vp run release:patch`
 - `vp run release:minor`
 - `vp run release:alpha`
 - `vp run release:beta`
 
-Each release command updates the workspace version in `Cargo.toml`, creates a release commit, and creates the matching `v...` git tag. Tag pushes publish `fastmulp-core` through GitHub Actions trusted publishing.
+Each release command updates the workspace version in `Cargo.toml`, creates a release commit, and creates the matching `v...` git tag. Tag pushes publish `fastmulp-core` through GitHub Actions trusted publishing and attach Node/wasm binding artifacts to the matching GitHub Release.
 
 ## Shared Tasks
 

crates/fastmulp_core/src/boundary.rs

@@ -1,3 +1,5 @@
+use std::borrow::Cow;
+
 use smallvec::SmallVec;
 
 use crate::{
@@ -11,8 +13,12 @@ pub struct Boundary<'a> {
     opening: SmallVec<[u8; 72]>,
 }
 
-pub fn boundary_from_content_type(_content_type: &str) -> Option<&str> {
-    let bytes = _content_type.as_bytes();
+/// Extracts the `boundary` parameter from a multipart `Content-Type` header value.
+///
+/// The returned value is borrowed for ordinary parameters and owned when a
+/// quoted-string contains MIME quoted-pair escapes that need to be decoded.
+pub fn boundary_from_content_type(content_type: &str) -> Option<Cow<'_, str>> {
+    let bytes = content_type.as_bytes();
     let mut cursor = skip_ascii_whitespace(bytes, 0);
     let media_start = cursor;
 
@@ -52,17 +58,13 @@ pub fn boundary_from_content_type(_content_type: &str) -> Option<&str> {
         cursor += 1;
         cursor = skip_ascii_whitespace(bytes, cursor);
 
-        let value = parse_content_type_value(_content_type, cursor)?;
+        let value = parse_content_type_value(content_type, cursor)?;
         cursor = skip_ascii_whitespace(bytes, value.next);
         if cursor < bytes.len() && bytes[cursor] != b';' {
             return None;
         }
 
         if eq_ignore_ascii_case(name, b"boundary") {
-            if value.requires_unescape {
-                return None;
-            }
-
             return Some(value.raw);
         }
     }
@@ -106,9 +108,8 @@ impl<'a> Boundary<'a> {
 }
 
 struct ContentTypeValue<'a> {
-    raw: &'a str,
+    raw: Cow<'a, str>,
     next: usize,
-    requires_unescape: bool,
 }
 
 fn parse_content_type_value(content_type: &str, start: usize) -> Option<ContentTypeValue<'_>> {
@@ -120,26 +121,37 @@ fn parse_content_type_value(content_type: &str, start: usize) -> Option<ContentT
     if bytes[start] == b'"' {
         let inner_start = start + 1;
         let mut cursor = inner_start;
-        let mut requires_unescape = false;
+        let mut decoded = None::<Vec<u8>>;
+        let mut copied_from = inner_start;
         while cursor < bytes.len() {
             match bytes[cursor] {
                 b'"' => {
+                    if let Some(mut decoded) = decoded {
+                        decoded.extend_from_slice(&bytes[copied_from..cursor]);
+                        let raw = String::from_utf8(decoded).ok()?;
+                        return Some(ContentTypeValue {
+                            raw: Cow::Owned(raw),
+                            next: cursor + 1,
+                        });
+                    }
+
                     return content_type
                         .get(inner_start..cursor)
                         .map(|raw| ContentTypeValue {
-                            raw,
+                            raw: Cow::Borrowed(raw),
                             next: cursor + 1,
-                            requires_unescape,
                         });
                 }
                 b'\\' => {
-                    requires_unescape = true;
-                    cursor += 1;
-                    if cursor == bytes.len() {
+                    if cursor + 1 == bytes.len() {
                         return None;
                     }
 
-                    cursor += 1;
+                    let decoded = decoded.get_or_insert_with(|| Vec::with_capacity(bytes.len()));
+                    decoded.extend_from_slice(&bytes[copied_from..cursor]);
+                    decoded.push(bytes[cursor + 1]);
+                    cursor += 2;
+                    copied_from = cursor;
                 }
                 _ => {
                     cursor += 1;
@@ -157,9 +169,8 @@ fn parse_content_type_value(content_type: &str, start: usize) -> Option<ContentT
 
     let end = trim_ascii_end(bytes, cursor);
     content_type.get(start..end).map(|raw| ContentTypeValue {
-        raw,
+        raw: Cow::Borrowed(raw),
         next: cursor,
-        requires_unescape: false,
     })
 }
 

crates/fastmulp_core/src/error.rs

@@ -1,5 +1,6 @@
 use core::fmt;
 
+#[non_exhaustive]
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub enum Error {
     EmptyBoundary,
@@ -17,6 +18,9 @@ pub enum Error {
     InvalidContentDisposition { offset: usize },
     MissingClosingBoundary { offset: usize },
     TrailingData { offset: usize },
+    PartLimitExceeded { limit: usize },
+    HeaderCountLimitExceeded { limit: usize, offset: usize },
+    HeaderBytesLimitExceeded { limit: usize, offset: usize },
 }
 
 impl fmt::Display for Error {
@@ -87,6 +91,17 @@ impl fmt::Display for Error {
                 f,
                 "multipart body has trailing data after byte offset {offset}"
             ),
+            Self::PartLimitExceeded { limit } => {
+                write!(f, "multipart part count exceeded configured limit {limit}")
+            }
+            Self::HeaderCountLimitExceeded { limit, offset } => write!(
+                f,
+                "multipart header count exceeded configured limit {limit} at byte offset {offset}"
+            ),
+            Self::HeaderBytesLimitExceeded { limit, offset } => write!(
+                f,
+                "multipart header bytes exceeded configured limit {limit} at byte offset {offset}"
+            ),
         }
     }
 }

crates/fastmulp_core/src/lib.rs

@@ -26,7 +26,7 @@ mod util;
 pub use boundary::{Boundary, boundary_from_content_type};
 pub use error::{Error, Result};
 pub use header::Header;
-pub use parser::{Multipart, MultipartParser, parse};
+pub use parser::{Multipart, MultipartParser, ParseLimits, parse, parse_with_limits};
 pub use part::Part;
 pub use text::TextValue;