feat: range decryption API on top of the fee package (FIL-570)

README.md

@@ -46,8 +46,8 @@ import (
 
 | Package | Purpose |
 |---|---|
-| [`fee`](.) (root) | Composes the primitives below into a small encrypt/decrypt API for whole objects. Adds no cryptography of its own. |
-| [`aesstream`](./aesstream) | The chunked AES-256-GCM STREAM body cipher: streaming `Writer`/`Reader` plus the range-decryption API (`CiphertextRange`, `SpanReader`, `OpenSpan`). |
+| [`fee`](.) (root) | Composes the primitives below into a small API: whole-object `Encrypt`/`Decrypt` plus byte-range `DecryptRange`. Adds no cryptography of its own. |
+| [`aesstream`](./aesstream) | The chunked AES-256-GCM STREAM body cipher: streaming `Writer`/`Reader` plus the range primitives (`CiphertextRange`, `SpanReader`, `OpenSpan`) that `fee.DecryptRange` is built on. |
 | [`cose`](./cose) | Just enough of COSE (RFC 9052): `COSE_Encrypt` (tag 96) / `COSE_Encrypt0` (tag 16) with a detached payload, and the `Enc_structure` AAD. |
 | [`ecdhkw`](./ecdhkw) | ECDH-ES+A256KW key wrap over X25519 (COSE algorithm −31). |
 | [`aeskw`](./aeskw) | RFC 3394 AES Key Wrap / A256KW (COSE algorithm −5). |
@@ -254,73 +254,68 @@ func roundTripExternalCEK(data []byte) ([]byte, error) {
 ### Range (seekable) decryption
 
 Because chunks are sealed independently, any plaintext byte range can be
-decrypted from a single contiguous slice of the ciphertext — one HTTP range
-request against a remote blob. The root `fee` package covers whole-object
-decryption only; ranges use the `cose` and `aesstream` packages directly:
+decrypted without fetching or decrypting the rest of the object.
+`fee.DecryptRange` takes the stored blob as an `io.ReaderAt` plus its exact size,
+unwraps the CEK just as `fee.Decrypt` does, and returns a reader over exactly the
+requested bytes:
 
 ```go
 import (
-    "bytes"
-    "errors"
+    "fmt"
+    "io"
+    "net/http"
+    "os"
+    "strconv"
 
     fee "github.com/filecoin-project/go-fee"
-    "github.com/filecoin-project/go-fee/aesstream"
-    "github.com/filecoin-project/go-fee/cose"
 )
 
-// readRange decrypts plaintext[off : off+length] from a FEE blob without
-// decrypting the whole object. cek is the content-encryption key, obtained out
-// of band or unwrapped from a recipient entry (see ecdhkw / aeskw).
-func readRange(blob, cek []byte, off, length int64) ([]byte, error) {
-    const labelChunkSize = int64(-65790)
-
-    // Decode the envelope header: base nonce, chunk size, and the
-    // Enc_structure that every chunk is authenticated against.
-    env, ciphertext, err := cose.Decode(blob, cose.WithExpectedType(fee.EnvelopeType))
-    if err != nil {
-        return nil, err
-    }
-    baseNonce, ok := env.Headers.Unprotected.Bytes(cose.HeaderLabelIV)
-    if !ok {
-        return nil, errors.New("missing base nonce")
-    }
-    chunkSize := aesstream.DefaultChunkSize
-    if n, ok := env.Headers.Unprotected.Int(labelChunkSize); ok {
-        chunkSize = int(n)
-    }
-    aad, err := env.EncStructure(nil)
+// serveRange answers an HTTP range request straight from an encrypted object.
+func serveRange(w http.ResponseWriter, f *os.File, size int64, u fee.RecipientUnwrapper, off, length int64) error {
+    r, err := fee.DecryptRange(f, size, u, off, length)
     if err != nil {
-        return nil, err
+        return err // aesstream.ErrRange here means a 416
     }
 
-    // Which contiguous ciphertext bytes cover the requested plaintext range?
-    // The third result (ignored here) is the clamped plaintext length of the
-    // range — available before any fetch, e.g. for an HTTP Content-Length.
-    start, n, _, err := aesstream.CiphertextRange(int64(len(ciphertext)), chunkSize, off, length)
-    if err != nil {
-        return nil, err
+    // Len is the requested length clamped to the object; Size is the whole
+    // object's plaintext size. Both are known before any ciphertext is read.
+    if r.Len() == 0 {
+        w.Header().Set("Content-Range", fmt.Sprintf("bytes */%d", r.Size()))
+        w.WriteHeader(http.StatusRequestedRangeNotSatisfiable)
+        return nil
     }
+    w.Header().Set("Content-Length", strconv.FormatInt(r.Len(), 10))
+    w.Header().Set("Content-Range", fmt.Sprintf("bytes %d-%d/%d", off, off+r.Len()-1, r.Size()))
+    w.WriteHeader(http.StatusPartialContent)
 
-    // Fetch exactly that span. Against a remote blob this is one range
-    // request for bytes [headerLen+start, headerLen+start+n), where
-    // headerLen = len(blob) - len(ciphertext).
-    span := bytes.NewReader(ciphertext[start : start+n])
-
-    // Decrypt and trim to exactly the requested range. For large ranges,
-    // aesstream.NewSpanReader streams instead of buffering.
-    return aesstream.OpenSpan(aesstream.Config{
-        Key:       cek,
-        BaseNonce: baseNonce,
-        AAD:       aad,
-        ChunkSize: chunkSize,
-    }, span, int64(len(ciphertext)), off, length)
+    _, err = io.Copy(w, r) // decrypts chunk by chunk, O(chunk size) memory
+    return err
 }
 ```
 
+Only the envelope header and the ciphertext chunks the range overlaps are read:
+one small `ReadAt` at offset 0 for the header, then one `ReadAt` per overlapping
+chunk as the result is read. Every chunk read is authenticated, so a tampered
+chunk fails rather than yielding corrupt plaintext.
+
+Nothing beyond the header is fetched until the first `Read`, so a caller backed
+by a remote store can prefetch the whole span in a single range request:
+
+```go
+r, err := fee.DecryptRange(blob, size, unwrapper, off, length)
+// ...
+spanOff, spanLen := r.CiphertextSpan() // blob-absolute; one range request
+```
+
 The span is chunk-aligned, so it over-fetches by at most the unused head of the
-first chunk and tail of the last (under 2 × chunk size total). For a local
-random-access source, wrap it with `io.NewSectionReader(src, start, n)` instead
-of a fetch.
+first chunk and tail of the last (under 2 × chunk size total).
+
+`fee.PlaintextSize` reports an object's decrypted size from the envelope header
+alone — no key material, no ciphertext — which is what a `HEAD` response or a
+suffix range (`bytes=-N`) needs. `fee.DecryptRangeWithCEK` is the external-CEK
+counterpart of `DecryptRange`. Callers holding raw ciphertext spans rather than a
+whole blob can use `aesstream.CiphertextRange` / `SpanReader` / `OpenSpan`
+directly.
 
 ## Wire format
 

example_range_test.go

@@ -0,0 +1,66 @@
+package fee_test
+
+import (
+	"bytes"
+	"crypto/ecdh"
+	"crypto/rand"
+	"fmt"
+	"io"
+	"log"
+
+	"github.com/filecoin-project/go-fee"
+)
+
+// ExampleDecryptRange serves one byte range of an encrypted object the way an
+// HTTP handler would: it derives the response's Content-Length and Content-Range
+// from the reader before any ciphertext is fetched, then streams the range out.
+func ExampleDecryptRange() {
+	priv, err := ecdh.X25519().GenerateKey(rand.Reader)
+	if err != nil {
+		log.Fatal(err)
+	}
+	kid := []byte("did:key:zExampleRecipient#key-1")
+
+	plaintext := []byte("the quick brown fox jumps over the lazy dog")
+	enc, err := fee.Encrypt(bytes.NewReader(plaintext),
+		[]fee.Recipient{fee.NewECDHESRecipient(kid, priv.PublicKey())},
+		fee.WithContentLength(int64(len(plaintext))))
+	if err != nil {
+		log.Fatal(err)
+	}
+	blob, err := io.ReadAll(enc)
+	if err != nil {
+		log.Fatal(err)
+	}
+	if err := enc.Close(); err != nil {
+		log.Fatal(err)
+	}
+
+	// A store hands over random access to the blob and its exact size; only the
+	// envelope header and the chunks the range overlaps are ever read.
+	const off, length = 4, 15
+	r, err := fee.DecryptRange(bytes.NewReader(blob), int64(len(blob)),
+		fee.NewECDHESUnwrapper(kid, priv), off, length)
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	// Both are known up front, so a handler can write its headers before
+	// decrypting a single chunk.
+	fmt.Printf("Content-Length: %d\n", r.Len())
+	if r.Len() == 0 {
+		fmt.Printf("Content-Range: bytes */%d\n", r.Size())
+		return
+	}
+	fmt.Printf("Content-Range: bytes %d-%d/%d\n", off, off+r.Len()-1, r.Size())
+	got, err := io.ReadAll(r)
+	if err != nil {
+		log.Fatal(err)
+	}
+	fmt.Printf("range: %q\n", got)
+
+	// Output:
+	// Content-Length: 15
+	// Content-Range: bytes 4-18/43
+	// range: "quick brown fox"
+}

fee.go

@@ -62,12 +62,19 @@
 // detached ciphertext from its source on demand. Neither buffers the whole
 // object.
 //
+// # Byte ranges
+//
+// [DecryptRange] serves one plaintext byte range of a stored blob without
+// fetching or decrypting the rest of it: it decodes the envelope header, recovers
+// the CEK exactly as [Decrypt] does, and reads only the ciphertext chunks the
+// range overlaps. [DecryptRangeWithCEK] is its external-CEK counterpart, and
+// [PlaintextSize] answers an object's decrypted size from the header alone, with
+// no key material. Callers that hold raw ciphertext spans rather than a blob can
+// use the underlying primitives in fee/aesstream directly.
+//
 // # Scope
 //
-// This package covers full-object encrypt/decrypt only. Range-based decryption
-// is a separate primitive in fee/aesstream, keyed off the ciphertext length and
-// chunk size rather than the envelope's chunk count; a higher-level range API is
-// tracked separately. This package adds no cryptography of its own.
+// This package sequences the primitives and adds no cryptography of its own.
 package fee
 
 import (
@@ -476,17 +483,51 @@ func DecryptWithCEK(src io.Reader, cek []byte) (io.Reader, error) {
 // lazily on later reads, which work from the internalized key, never the cek
 // slice.
 func openStream(env *cose.Envelope, ciphertext io.Reader, cek []byte) (io.Reader, error) {
+	body, err := validateBody(env)
+	if err != nil {
+		return nil, err
+	}
+	r, err := aesstream.NewReader(ciphertext, aesstream.Config{
+		Key:       cek,
+		BaseNonce: body.baseNonce,
+		AAD:       body.aad,
+		ChunkSize: body.chunkSize,
+	})
+	if err != nil {
+		return nil, fmt.Errorf("fee: initializing body cipher: %w", err)
+	}
+	return r, nil
+}
+
+// bodyParams is the validated STREAM configuration a FEE envelope's body header
+// carries: everything fee/aesstream needs to decrypt the detached ciphertext
+// apart from the content-encryption key.
+type bodyParams struct {
+	baseNonce []byte
+	chunkSize int
+	aad       []byte
+}
+
+// validateBody checks a decoded envelope's FEE body headers — the algorithm is
+// the chunked AES-256-GCM-STREAM cipher, the base nonce (iv) is present, and the
+// self-describing chunk size is in range — and rebuilds the Enc_structure AAD
+// that the encoder bound into every chunk.
+//
+// It is shared by the whole-object path ([openStream]) and the range path
+// ([newRangeReader]), so both accept exactly the same envelopes and report the
+// same errors for a body header they cannot honour.
+func validateBody(env *cose.Envelope) (bodyParams, error) {
 	alg, ok := env.Headers.Protected.Int(cose.HeaderLabelAlg)
 	if !ok {
-		return nil, fmt.Errorf("%w: body algorithm header missing or not an integer", ErrUnsupportedBodyAlg)
+		return bodyParams{}, fmt.Errorf("%w: body algorithm header missing or not an integer", ErrUnsupportedBodyAlg)
 	}
 	if alg != algChunkedAES256GCMStream {
-		return nil, fmt.Errorf("%w: body algorithm %d is not chunked AES-256-GCM-STREAM", ErrUnsupportedBodyAlg, alg)
+		return bodyParams{}, fmt.Errorf("%w: body algorithm %d is not chunked AES-256-GCM-STREAM", ErrUnsupportedBodyAlg, alg)
 	}
 
 	baseNonce, ok := env.Headers.Unprotected.Bytes(cose.HeaderLabelIV)
 	if !ok {
-		return nil, fmt.Errorf("%w: missing iv (base nonce)", ErrMalformedEnvelope)
+		return bodyParams{}, fmt.Errorf("%w: missing iv (base nonce)", ErrMalformedEnvelope)
 	}
 
 	// Self-describing chunk size; an envelope that omits it is read at the FEE
@@ -497,12 +538,12 @@ func openStream(env *cose.Envelope, ciphertext io.Reader, cek []byte) (io.Reader
 	if env.Headers.Unprotected.Has(labelChunkSize) {
 		n, ok := env.Headers.Unprotected.Int(labelChunkSize)
 		if !ok {
-			return nil, fmt.Errorf("%w: chunk-size header is present but not an integer", ErrMalformedEnvelope)
+			return bodyParams{}, fmt.Errorf("%w: chunk-size header is present but not an integer", ErrMalformedEnvelope)
 		}
 		chunkSize = n
 	}
 	if chunkSize < int64(aesstream.MinChunkSize) || chunkSize > int64(aesstream.MaxChunkSize) {
-		return nil, fmt.Errorf("%w: declared chunk size %d out of range [%d, %d]",
+		return bodyParams{}, fmt.Errorf("%w: declared chunk size %d out of range [%d, %d]",
 			ErrMalformedEnvelope, chunkSize, aesstream.MinChunkSize, aesstream.MaxChunkSize)
 	}
 
@@ -511,19 +552,10 @@ func openStream(env *cose.Envelope, ciphertext io.Reader, cek []byte) (io.Reader
 	// the encoder bound into every chunk.
 	aad, err := env.EncStructure(nil)
 	if err != nil {
-		return nil, fmt.Errorf("fee: building envelope AAD: %w", err)
+		return bodyParams{}, fmt.Errorf("fee: building envelope AAD: %w", err)
 	}
 
-	r, err := aesstream.NewReader(ciphertext, aesstream.Config{
-		Key:       cek,
-		BaseNonce: baseNonce,
-		AAD:       aad,
-		ChunkSize: int(chunkSize),
-	})
-	if err != nil {
-		return nil, fmt.Errorf("fee: initializing body cipher: %w", err)
-	}
-	return r, nil
+	return bodyParams{baseNonce: baseNonce, chunkSize: int(chunkSize), aad: aad}, nil
 }
 
 // matchRecipient returns the first recipient whose kid equals want. A recipient