@znight/http 0.0.0-stage → 0.1.0

package/src/multipart.ts

@@ -0,0 +1,597 @@
+/**
+ * Streaming `multipart/form-data` parser (RFC 7578, RFC 2046), shared by every adapter so that
+ * uploads behave the same on Node.js and Bun.
+ *
+ * The Web-standard `Request.formData()` is not used: it buffers the whole body before parsing, has
+ * no per-part limits, and its results differ between runtimes (default part type, `filename*`,
+ * escaped quotes, LF line endings, parts without name). This parser reads the body as it arrives
+ * and enforces the upload rules of the route (expected fields, file count, size and media type) on
+ * the fly, so an invalid upload is rejected as soon as the offending part shows up. Files are
+ * buffered in memory as standard `File` objects, within the limits.
+ *
+ * Status codes: malformed bodies, unexpected file fields and several files for a single-file field
+ * → 400; a file, a field, too many files or too many parts → 413; a refused media type → 415. Errors
+ * raised while the body is being read carry `connection: close` (the rest of the body is discarded).
+ */
+import type { HttpExceptionOptions } from './exceptions.ts'
+import type { QueryRecord } from './query.ts'
+
+import { Buffer } from 'node:buffer'
+
+import { BadRequestException, issue, PayloadTooLargeException, UnsupportedMediaTypeException } from './exceptions.ts'
+import { matchesMediaType, mediaTypeEssence, OCTET_STREAM } from './mime.ts'
+import { formatByteSize } from './size.ts'
+import { emptyRecord } from './utils.ts'
+
+/** Constraints of the files received by one upload parameter (`@UploadedFile` / `@UploadedFiles`). */
+export interface FileRule {
+  /** Field name; `undefined`: the files of every field. */
+  readonly field: string | undefined
+  /** `@UploadedFiles` (an array) rather than `@UploadedFile` (a single file). */
+  readonly multiple: boolean
+  /** Maximum size of each file, in bytes. */
+  readonly maxSize: number
+  /** Maximum number of files (1 for a single file). */
+  readonly maxFiles: number
+  /** Accepted media types (`image/png`, `image/*`); `undefined`: any. */
+  readonly mimeTypes: readonly string[] | undefined
+}
+
+/** How a multipart body is read. */
+export interface MultipartPolicy {
+  /**
+   * Upload rules of the route (files are only accepted in their fields: a file in another field is
+   * rejected with 400, unless a rule has no field). Without rules, every file is accepted without
+   * limit (but the body limit).
+   */
+  readonly rules: readonly FileRule[]
+  /** Maximum number of parts (text fields and files). */
+  readonly maxParts: number
+  /** Maximum size of a text field, in bytes. */
+  readonly maxFieldSize: number
+}
+
+/** A parsed multipart body. */
+export interface MultipartResult {
+  /** Text fields (repeated names → arrays), without prototype. */
+  readonly fields: QueryRecord
+  /** Files by field name, without prototype. */
+  readonly files: Record<string, File[]>
+  /** Every file, in body order. */
+  readonly list: readonly File[]
+  /** Every entry, in body order. */
+  readonly entries: readonly (readonly [string, string | File])[]
+}
+
+/** Default maximum size of the headers of one part. */
+const MAX_HEADER_SIZE = 16 * 1024
+/** Maximum transport padding (spaces) after a boundary delimiter. */
+const MAX_PADDING = 256
+/** Errors raised while the body is still being read ask the client to close the connection. */
+const CLOSE: HttpExceptionOptions = Object.freeze({ headers: Object.freeze({ connection: 'close' }) })
+
+const CR = 13
+const LF = 10
+const DASH = 45
+const SPACE = 32
+const TAB = 9
+const HEADER_END = Buffer.from('\r\n\r\n')
+const EMPTY: Buffer = Buffer.alloc(0)
+const DECODER = new TextDecoder()
+
+const PRINTABLE = /^[\u0020-\u007e]*$/
+
+/**
+ * A received file. Its `type` is the media type declared by the client, lower-cased (empty when it
+ * contains characters outside printable ASCII), identical on every runtime (Bun's `File` adds
+ * `;charset=utf-8` to some text types).
+ */
+function createFile(chunks: Uint8Array[], name: string, type: string): File {
+  const file = new File(chunks, name, { type })
+  const normalized = PRINTABLE.test(type) ? type.toLowerCase() : ''
+  if (file.type !== normalized) Object.defineProperty(file, 'type', { value: normalized, configurable: true })
+  return file
+}
+
+/** Description of a part, from its headers. */
+export interface PartInfo {
+  readonly name: string
+  /** `undefined` for text fields. */
+  readonly filename: string | undefined
+  readonly contentType: string | undefined
+}
+
+/** Receives the content of one part. */
+export interface PartSink {
+  data(chunk: Uint8Array): void
+  end(): void
+}
+
+/** Receives the parts found by {@link MultipartParser}. */
+export interface PartHandler {
+  start(part: PartInfo): PartSink
+}
+
+type State = 'preamble' | 'delimiter' | 'headers' | 'body' | 'done'
+
+/**
+ * Incremental parser: feed it with {@link write} as chunks arrive, then call {@link end}. Errors
+ * (malformed body, limits enforced by the handler) are thrown synchronously from `write`/`end`.
+ */
+export class MultipartParser {
+  readonly #delimiter: Buffer
+  readonly #handler: PartHandler
+  readonly #maxHeaderSize: number
+  /** The body is conceptually prefixed with CRLF so that every delimiter is `CRLF--boundary`. */
+  #buffer: Buffer = Buffer.from('\r\n')
+  #state: State = 'preamble'
+  #sink: PartSink | undefined
+  /** Bytes of the header block already searched for its end (avoids rescanning on every chunk). */
+  #scanned = 0
+
+  constructor(boundary: string, handler: PartHandler, maxHeaderSize: number = MAX_HEADER_SIZE) {
+    this.#delimiter = Buffer.from(`\r\n--${boundary}`, 'latin1')
+    this.#handler = handler
+    this.#maxHeaderSize = maxHeaderSize
+  }
+
+  /** `true` once the closing delimiter was read (what follows is the epilogue, ignored). */
+  get done(): boolean {
+    return this.#state === 'done'
+  }
+
+  write(chunk: Uint8Array): void {
+    if (this.#state === 'done' || chunk.byteLength === 0) return
+    const view = Buffer.from(chunk.buffer, chunk.byteOffset, chunk.byteLength)
+    this.#buffer = this.#buffer.length === 0 ? view : Buffer.concat([this.#buffer, view])
+    this.#process()
+  }
+
+  end(): void {
+    if (this.#state !== 'done') throw malformed('unexpected end of the body (missing closing boundary)')
+  }
+
+  #process(): void {
+    for (;;) {
+      switch (this.#state) {
+        case 'preamble': {
+          const index = this.#buffer.indexOf(this.#delimiter)
+          if (index === -1) {
+            this.#keepTail()
+            return
+          }
+          this.#buffer = this.#buffer.subarray(index + this.#delimiter.length)
+          this.#state = 'delimiter'
+          break
+        }
+        case 'delimiter':
+          if (!this.#afterDelimiter()) return
+          break
+        case 'headers':
+          if (!this.#headers()) return
+          break
+        case 'body':
+          if (!this.#body()) return
+          break
+        default:
+          return
+      }
+    }
+  }
+
+  /** After a delimiter: `--` (closing delimiter) or optional padding then CRLF (next part). */
+  #afterDelimiter(): boolean {
+    const buffer = this.#buffer
+    if (buffer.length < 2) return false
+    if (buffer[0] === DASH && buffer[1] === DASH) {
+      this.#state = 'done'
+      this.#buffer = EMPTY
+      return false
+    }
+    let index = 0
+    while (index < buffer.length && (buffer[index] === SPACE || buffer[index] === TAB)) index++
+    if (index > MAX_PADDING) throw malformed('invalid boundary delimiter')
+    if (index + 1 >= buffer.length) return false
+    if (buffer[index] !== CR || buffer[index + 1] !== LF) throw malformed('invalid boundary delimiter')
+    this.#buffer = buffer.subarray(index + 2)
+    this.#state = 'headers'
+    return true
+  }
+
+  #headers(): boolean {
+    const buffer = this.#buffer
+    if (buffer.length < 2) return false
+    let end: number
+    let next: number
+    if (buffer[0] === CR && buffer[1] === LF) {
+      end = 0
+      next = 2
+    } else {
+      end = buffer.indexOf(HEADER_END, Math.max(0, this.#scanned - (HEADER_END.length - 1)))
+      if (end === -1) {
+        if (buffer.length > this.#maxHeaderSize) throw malformed('the headers of a part are too large')
+        this.#scanned = buffer.length
+        return false
+      }
+      if (end > this.#maxHeaderSize) throw malformed('the headers of a part are too large')
+      next = end + 4
+    }
+    const part = parsePartHeaders(buffer.toString('utf8', 0, end))
+    this.#scanned = 0
+    this.#buffer = buffer.subarray(next)
+    this.#sink = this.#handler.start(part)
+    this.#state = 'body'
+    return true
+  }
+
+  #body(): boolean {
+    const buffer = this.#buffer
+    const sink = this.#sink as PartSink
+    const index = buffer.indexOf(this.#delimiter)
+    if (index === -1) {
+      // keep what may be the beginning of a delimiter
+      const safe = buffer.length - (this.#delimiter.length - 1)
+      if (safe > 0) {
+        sink.data(buffer.subarray(0, safe))
+        this.#buffer = buffer.subarray(safe)
+      }
+      return false
+    }
+    if (index > 0) sink.data(buffer.subarray(0, index))
+    this.#sink = undefined
+    sink.end()
+    this.#buffer = buffer.subarray(index + this.#delimiter.length)
+    this.#state = 'delimiter'
+    return true
+  }
+
+  #keepTail(): void {
+    const keep = this.#delimiter.length - 1
+    if (this.#buffer.length > keep) this.#buffer = this.#buffer.subarray(this.#buffer.length - keep)
+  }
+}
+
+// -------------------------------------------------------------------------------------------------
+// Headers
+// -------------------------------------------------------------------------------------------------
+
+function parsePartHeaders(block: string): PartInfo {
+  let disposition: string | undefined
+  let contentType: string | undefined
+  let current: 'content-disposition' | 'content-type' | undefined
+  if (block !== '') {
+    for (const line of block.split('\r\n')) {
+      if (line.startsWith(' ') || line.startsWith('\t')) {
+        // obsolete line folding: continuation of the previous header
+        if (current === 'content-disposition') disposition = `${disposition ?? ''} ${line.trim()}`
+        else if (current === 'content-type') contentType = `${contentType ?? ''} ${line.trim()}`
+        continue
+      }
+      const colon = line.indexOf(':')
+      if (colon <= 0) throw malformed('invalid part header')
+      const name = line.slice(0, colon).trim().toLowerCase()
+      const value = line.slice(colon + 1).trim()
+      current = undefined
+      if (name === 'content-disposition' && disposition === undefined) {
+        disposition = value
+        current = name
+      } else if (name === 'content-type' && contentType === undefined) {
+        contentType = value
+        current = name
+      }
+    }
+  }
+  if (disposition === undefined) throw malformed('a part has no Content-Disposition header')
+  const { type, params } = parseParameters(disposition)
+  if (type !== 'form-data') throw malformed('a part is not "form-data"')
+  const name = params.get('name')
+  if (name === undefined) throw malformed('a part has no name')
+  const extended = params.get('filename*')
+  const plain = params.get('filename')
+  const filename = (extended === undefined ? undefined : decodeExtendedValue(extended)) ?? plain
+  return {
+    name: decodeFormName(name),
+    filename: filename === undefined ? undefined : decodeFormName(filename),
+    contentType: contentType === '' ? undefined : contentType,
+  }
+}
+
+/**
+ * Parses `value; key=token; key="quoted \" string"` (Content-Disposition, Content-Type). Keys are
+ * lower-cased; the first occurrence of a key wins.
+ */
+export function parseParameters(value: string): { readonly type: string; readonly params: Map<string, string> } {
+  const params = new Map<string, string>()
+  let index = value.indexOf(';')
+  const type = (index === -1 ? value : value.slice(0, index)).trim().toLowerCase()
+  while (index !== -1 && index < value.length) {
+    index++ // skip ';'
+    while (index < value.length && isSpace(value.charCodeAt(index))) index++
+    if (index >= value.length) break
+    if (value[index] === ';') continue // empty parameter
+    const equals = value.indexOf('=', index)
+    if (equals === -1) throw malformed(`invalid parameter in ${JSON.stringify(truncate(value))}`)
+    const key = value.slice(index, equals).trim().toLowerCase()
+    index = equals + 1
+    while (index < value.length && isSpace(value.charCodeAt(index))) index++
+    let parameter: string
+    if (value[index] === '"') {
+      parameter = ''
+      index++
+      for (;;) {
+        if (index >= value.length) throw malformed('unterminated quoted string in a part header')
+        const char = value[index] as string
+        if (char === '"') {
+          index++
+          break
+        }
+        if (char === '\\' && (value[index + 1] === '"' || value[index + 1] === '\\')) {
+          parameter += value[index + 1]
+          index += 2
+          continue
+        }
+        parameter += char
+        index++
+      }
+      while (index < value.length && isSpace(value.charCodeAt(index))) index++
+      if (index < value.length && value[index] !== ';') throw malformed('invalid quoted string in a part header')
+    } else {
+      const next = value.indexOf(';', index)
+      parameter = (next === -1 ? value.slice(index) : value.slice(index, next)).trim()
+      index = next === -1 ? value.length : next
+    }
+    if (key !== '' && !params.has(key)) params.set(key, parameter)
+    if (index >= value.length) break
+  }
+  return { type, params }
+}
+
+function isSpace(code: number): boolean {
+  return code === SPACE || code === TAB
+}
+
+/** Names and filenames: browsers escape `"`, CR and LF as `%22`, `%0D` and `%0A` (HTML standard). */
+function decodeFormName(value: string): string {
+  return value.includes('%')
+    ? value.replace(/%(22|0D|0A)/gi, (_match, hex: string) => String.fromCharCode(Number.parseInt(hex, 16)))
+    : value
+}
+
+/** RFC 5987 `charset'language'percent-encoded` (`filename*`); `undefined` when it cannot be decoded. */
+function decodeExtendedValue(value: string): string | undefined {
+  const match = /^([\w!#$%&+^`{}~-]+)'[\w-]*'(.*)$/.exec(value)
+  if (match === null) return undefined
+  const charset = (match[1] as string).toLowerCase()
+  const encoded = match[2] as string
+  try {
+    if (charset === 'utf-8') return decodeURIComponent(encoded)
+    if (charset === 'iso-8859-1') {
+      return encoded.replace(/%([0-9a-f]{2})/gi, (_match, hex: string) => String.fromCharCode(Number.parseInt(hex, 16)))
+    }
+  } catch {
+    return undefined
+  }
+  return undefined
+}
+
+/** The `boundary` parameter of a `multipart/form-data` content type (400 when missing or invalid). */
+export function multipartBoundary(contentType: string): string {
+  let boundary: string | undefined
+  try {
+    boundary = parseParameters(contentType).params.get('boundary')
+  } catch {
+    boundary = undefined
+  }
+  if (boundary === undefined || boundary === '') throw malformed('missing boundary in the content type')
+  // RFC 2046: 1 to 70 printable ASCII characters
+  if (boundary.length > 70 || [...boundary].some(char => char < ' ' || char > '~')) {
+    throw malformed('invalid boundary in the content type')
+  }
+  return boundary
+}
+
+// -------------------------------------------------------------------------------------------------
+// Collecting fields and files
+// -------------------------------------------------------------------------------------------------
+
+/** Applies a {@link MultipartPolicy} to the parts and collects fields and files. */
+class Collector implements PartHandler {
+  readonly fields: QueryRecord = emptyRecord()
+  readonly files: Record<string, File[]> = emptyRecord()
+  readonly list: File[] = []
+  readonly entries: (readonly [string, string | File])[] = []
+  readonly #policy: MultipartPolicy
+  readonly #named: ReadonlyMap<string, FileRule>
+  readonly #any: FileRule | undefined
+  readonly #counts = new Map<FileRule, number>()
+  #parts = 0
+
+  constructor(policy: MultipartPolicy) {
+    this.#policy = policy
+    const named = new Map<string, FileRule>()
+    let any: FileRule | undefined
+    for (const rule of policy.rules) {
+      if (rule.field === undefined) any = rule
+      else named.set(rule.field, rule)
+    }
+    this.#named = named
+    this.#any = any
+  }
+
+  start(part: PartInfo): PartSink {
+    this.#parts++
+    if (this.#parts > this.#policy.maxParts) {
+      throw new PayloadTooLargeException(
+        `Too many parts in the multipart body (at most ${this.#policy.maxParts})`,
+        { maxParts: this.#policy.maxParts },
+        CLOSE,
+      )
+    }
+    return part.filename === undefined ? this.#field(part.name) : this.#file(part, part.filename)
+  }
+
+  #field(name: string): PartSink {
+    const limit = this.#policy.maxFieldSize
+    const chunks: Uint8Array[] = []
+    let size = 0
+    return {
+      data: chunk => {
+        size += chunk.byteLength
+        if (size > limit) {
+          throw new PayloadTooLargeException(
+            `Field ${JSON.stringify(truncate(name))} exceeds the maximum size of ${formatByteSize(limit)}`,
+            { field: name, maxSize: limit },
+            CLOSE,
+          )
+        }
+        chunks.push(chunk)
+      },
+      end: () => {
+        const value = DECODER.decode(chunks.length === 1 ? (chunks[0] as Uint8Array) : Buffer.concat(chunks))
+        const existing = this.fields[name]
+        if (existing === undefined) this.fields[name] = value
+        else if (typeof existing === 'string') this.fields[name] = [existing, value]
+        else existing.push(value)
+        this.entries.push([name, value])
+      },
+    }
+  }
+
+  #file(part: PartInfo, filename: string): PartSink {
+    const type = part.contentType === undefined ? OCTET_STREAM : part.contentType
+    const name = part.name
+    let rules: readonly FileRule[] = []
+    let admitted = false
+    // an empty <input type="file"> is sent as an empty part without filename: it is not a file
+    const admit = (): void => {
+      admitted = true
+      rules = this.#admit(name, type)
+    }
+    if (filename !== '') admit()
+    const chunks: Uint8Array[] = []
+    let size = 0
+    return {
+      data: chunk => {
+        if (!admitted) admit()
+        size += chunk.byteLength
+        for (const rule of rules) {
+          if (size > rule.maxSize) {
+            throw new PayloadTooLargeException(
+              `File in field ${JSON.stringify(truncate(name))} exceeds the maximum size of ${formatByteSize(rule.maxSize)}`,
+              { field: name, maxSize: rule.maxSize },
+              CLOSE,
+            )
+          }
+        }
+        chunks.push(chunk)
+      },
+      end: () => {
+        if (!admitted) return
+        const file = createFile(chunks, filename, type)
+        const list = this.files[name]
+        if (list === undefined) this.files[name] = [file]
+        else list.push(file)
+        this.list.push(file)
+        this.entries.push([name, file])
+      },
+    }
+  }
+
+  /** Rules applying to a file of `field` (400 for an unexpected field), with count and type checks. */
+  #admit(field: string, type: string): readonly FileRule[] {
+    if (this.#policy.rules.length === 0) return []
+    const rules: FileRule[] = []
+    const named = this.#named.get(field)
+    if (named !== undefined) rules.push(named)
+    if (this.#any !== undefined) rules.push(this.#any)
+    if (rules.length === 0) {
+      const expected = [...this.#named.keys()]
+      throw new BadRequestException(
+        `Unexpected file field ${JSON.stringify(truncate(field))}`,
+        issue(field, `Unexpected file field (expected ${expected.join(', ')})`),
+        CLOSE,
+      )
+    }
+    for (const rule of rules) {
+      const count = (this.#counts.get(rule) ?? 0) + 1
+      this.#counts.set(rule, count)
+      if (count > rule.maxFiles) {
+        const where = rule.field === undefined ? '' : ` in field ${JSON.stringify(truncate(rule.field))}`
+        if (!rule.multiple) {
+          throw new BadRequestException(`Expected a single file${where}`, issue(field, 'Expected a single file'), CLOSE)
+        }
+        throw new PayloadTooLargeException(
+          `Too many files${where} (at most ${rule.maxFiles})`,
+          { field, maxFiles: rule.maxFiles },
+          CLOSE,
+        )
+      }
+      if (rule.mimeTypes !== undefined && !matchesMediaType(type, rule.mimeTypes)) {
+        throw new UnsupportedMediaTypeException(
+          `Unsupported file type ${JSON.stringify(truncate(mediaTypeEssence(type)))} in field ${JSON.stringify(truncate(field))} (accepted: ${rule.mimeTypes.join(', ')})`,
+          { field, type: mediaTypeEssence(type), accepted: rule.mimeTypes },
+          CLOSE,
+        )
+      }
+    }
+    return rules
+  }
+
+  result(): MultipartResult {
+    return { fields: this.fields, files: this.files, list: this.list, entries: this.entries }
+  }
+}
+
+/**
+ * Reads a multipart body from `stream` with `policy`. The stream is cancelled on error (the rest of
+ * the body is discarded).
+ *
+ * @throws BadRequestException (400) for a malformed body, too many parts / files or an unexpected
+ *   file field; PayloadTooLargeException (413) for a file or field above its limit;
+ *   UnsupportedMediaTypeException (415) for a file of a refused type.
+ */
+export async function readMultipart(
+  stream: ReadableStream<Uint8Array>,
+  contentType: string,
+  policy: MultipartPolicy,
+): Promise<MultipartResult> {
+  const reader = stream.getReader()
+  try {
+    const collector = new Collector(policy)
+    const parser = new MultipartParser(multipartBoundary(contentType), collector)
+    for (;;) {
+      const { done, value } = await reader.read()
+      if (done) break
+      parser.write(value)
+    }
+    parser.end()
+    return collector.result()
+  } catch (error) {
+    await reader.cancel(error).catch(() => undefined)
+    throw error
+  }
+}
+
+/** Parses a buffered multipart body (see {@link readMultipart}). */
+export function parseMultipart(bytes: Uint8Array, contentType: string, policy: MultipartPolicy): MultipartResult {
+  const collector = new Collector(policy)
+  const parser = new MultipartParser(multipartBoundary(contentType), collector)
+  parser.write(bytes)
+  parser.end()
+  return collector.result()
+}
+
+/** Converts a parse result to a `FormData` (entries in body order). */
+export function toFormData(result: MultipartResult): FormData {
+  const form = new FormData()
+  for (const [name, value] of result.entries) form.append(name, value)
+  return form
+}
+
+function malformed(reason: string): BadRequestException {
+  return new BadRequestException(`Invalid multipart body: ${reason}`, undefined, CLOSE)
+}
+
+function truncate(value: string): string {
+  return value.length > 100 ? `${value.slice(0, 100)}…` : value
+}

package/src/option-error.ts

@@ -0,0 +1,31 @@
+/** Errors of invalid `HttpPlugin` options (internal). */
+import { ErrorCode, ZnightError } from '@znight/core'
+
+/** `ZNIGHT005: HttpPlugin: option "<name>" must be <expected> (received <value>)`. */
+export function invalidOption(
+  name: string,
+  expected: string,
+  value: unknown,
+  cause?: unknown,
+  hints?: readonly string[],
+): ZnightError {
+  return new ZnightError(
+    ErrorCode.InvalidDecorator,
+    `HttpPlugin: option "${name}" must be ${expected} (received ${describeOption(value)})`,
+    {
+      ...(cause === undefined ? {} : { cause }),
+      ...(hints === undefined ? {} : { hints }),
+    },
+  )
+}
+
+/** A short description of an option value for messages. */
+export function describeOption(value: unknown): string {
+  if (typeof value === 'function') return 'a function'
+  if (value instanceof RegExp) return String(value)
+  try {
+    return JSON.stringify(value) ?? String(value)
+  } catch {
+    return typeof value
+  }
+}