These two land together on purpose. Fixing the encoder alone changes nothing for the blobs already in the zone, because emit skips whatever is already present. #86 — klauspost/compress omits the zstd Frame_Content_Size field for inputs under 256 bytes, which the format permits. Reference libzstd never does, so the NWN client — which sizes its output buffer from ZSTD_getFrameContentSize and has therefore never met a frame without one — rejected roughly 6% of our blobs outright. Any single one stops a sync dead, so no client could complete a sync of the live manifest. No encoder option changes this, so emit re-headers the affected frames into the shape libzstd itself emits: Single_Segment_flag set, Window_Descriptor dropped, and the freed byte spent on a one-byte Frame_Content_Size. Same length in, same length out. Verified against the zstd CLI end to end: a real emitted 230-byte blob now reports its decompressed size where it previously reported none. emitter_version goes to 2, so assemble refuses to merge an index written by the encoder that omitted the field. #85 — a published blob was verified by exactly one thing, a player's client, at download time. `nwsync verify <manifest-sha1>` now reads a manifest and its blobs back through the public pull zone with no credential, decompresses and hashes every one, and reports missing / malformed framing / size mismatch / hash mismatch per blob. `--sample N` makes a routine check cheap against a manifest that is ~69,000 blobs and 15 GB. `emit --verify` applies the same check where emit would otherwise trust presence, and replaces a stored blob that is not what its name claims — which is what makes the #86 blobs repairable. Both new checks assert the frame property, not just a round trip. Round-tripping cannot see this class of fault: both the zstd CLI and Go's decoder stream such a frame happily, being more capable decoders than the client's, which is how the defect reached production and survived an audit. Refs #54, #55, #59, #75, sow-platform#79. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
171 lines
6.7 KiB
Go
171 lines
6.7 KiB
Go
package nwsync
|
|
|
|
import (
|
|
"bytes"
|
|
"encoding/binary"
|
|
"encoding/hex"
|
|
"fmt"
|
|
|
|
"github.com/klauspost/compress/zstd"
|
|
)
|
|
|
|
// NWCompressedBuffer framing, as upstream's neverwinter/compressedbuf.nim
|
|
// writes it for NWSync blobs. All fields are little-endian uint32:
|
|
//
|
|
// magic "NSYC", version 3, algorithm 2 (zstd), uncompressed size,
|
|
// zstd header version 1, dictionary 0, then the raw zstd frame.
|
|
const (
|
|
blobMagic = 0x4359534E // "NSYC" little-endian
|
|
blobVersion = 3
|
|
algorithmZstd = 2
|
|
zstdHeaderVer = 1
|
|
zstdDictionary = 0
|
|
blobHeaderBytes = 24
|
|
)
|
|
|
|
// EncodeAll/DecodeAll are single-threaded per call, so the default pool of one
|
|
// encoder per CPU only buys idle memory: each holds a window-sized history, so
|
|
// on a 24-core runner that is ~200 MB of live heap doing nothing. Concurrency 1
|
|
// produces byte-identical output.
|
|
var (
|
|
blobEncoder, _ = zstd.NewWriter(nil, zstd.WithEncoderConcurrency(1))
|
|
blobDecoder, _ = zstd.NewReader(nil, zstd.WithDecoderConcurrency(1))
|
|
)
|
|
|
|
// zstd frame header bits we care about. A frame starts with the magic, then a
|
|
// one-byte Frame_Header_Descriptor: bits 7-6 size the Frame_Content_Size field,
|
|
// bit 5 is Single_Segment_flag, bits 1-0 size the Dictionary_ID field.
|
|
const (
|
|
zstdFrameMagic = "\x28\xb5\x2f\xfd"
|
|
frameSingleSegment = 1 << 5
|
|
frameDictionaryMask = 0x03
|
|
// The size below which klauspost/compress writes no Frame_Content_Size,
|
|
// matching the field's own encoding threshold.
|
|
frameSizeThreshold = 256
|
|
)
|
|
|
|
// compressBlob wraps data in NWCompressedBuffer framing.
|
|
func compressBlob(data []byte) []byte {
|
|
var out bytes.Buffer
|
|
header := []uint32{blobMagic, blobVersion, algorithmZstd, uint32(len(data)), zstdHeaderVer, zstdDictionary}
|
|
for _, field := range header {
|
|
_ = binary.Write(&out, binary.LittleEndian, field)
|
|
}
|
|
out.Write(declareFrameContentSize(blobEncoder.EncodeAll(data, nil), len(data)))
|
|
return out.Bytes()
|
|
}
|
|
|
|
// inspectBlob unwraps a stored blob the way the game client reads it, and is the
|
|
// only reader that should be trusted to judge a published blob.
|
|
//
|
|
// It asserts the frame property on top of the round trip. Go's decoder — like
|
|
// the zstd CLI — streams a frame that declares no content size, so a check that
|
|
// only decompresses and hashes is a *more* capable decoder than the client's: it
|
|
// certifies exactly the blobs the client rejects, which is how #86 reached
|
|
// production and survived an audit.
|
|
func inspectBlob(blob []byte) ([]byte, error) {
|
|
data, err := decompressBlob(blob)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("malformed framing: %w", err)
|
|
}
|
|
if len(data) > 0 && !frameDeclaresContentSize(blob[blobHeaderBytes:]) {
|
|
return nil, fmt.Errorf("malformed framing: the zstd frame declares no content size, which the game client cannot decode")
|
|
}
|
|
return data, nil
|
|
}
|
|
|
|
// blobMatchesName holds a stored blob to its own file name: a blob is named
|
|
// after the sha1 of its uncompressed bytes, so the name is a complete statement
|
|
// about the contents and nothing else is needed to check it.
|
|
func blobMatchesName(blob []byte, sha1Hex string) error {
|
|
data, err := inspectBlob(blob)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if got := hex.EncodeToString(sha1Sum(data)); got != sha1Hex {
|
|
return fmt.Errorf("blob %s holds the contents of %s", sha1Hex, got)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// frameDeclaresContentSize reports whether a zstd frame states how many bytes it
|
|
// decompresses to. A frame with a zero-sized Frame_Content_Size field declares
|
|
// one only when Single_Segment_flag is set; otherwise the size is unknown.
|
|
func frameDeclaresContentSize(frame []byte) bool {
|
|
if len(frame) < 5 || string(frame[:4]) != zstdFrameMagic {
|
|
return false
|
|
}
|
|
descriptor := frame[4]
|
|
return descriptor>>6 != 0 || descriptor&frameSingleSegment != 0
|
|
}
|
|
|
|
// declareFrameContentSize rewrites a frame that does not declare its
|
|
// Frame_Content_Size so that it does, and returns any other frame unchanged.
|
|
//
|
|
// klauspost/compress omits the field for inputs under 256 bytes, which the spec
|
|
// permits. Reference libzstd never does, so the NWN client — which sizes its
|
|
// output buffer from ZSTD_getFrameContentSize and has therefore never met a
|
|
// frame without one — rejects the blob outright with an empty "potential
|
|
// compression error" (#86). No encoder option changes this, so the frame is
|
|
// re-headered here.
|
|
//
|
|
// The result is the shape libzstd itself emits for the same input: setting
|
|
// Single_Segment_flag drops the Window_Descriptor byte, and the freed byte pays
|
|
// for a one-byte Frame_Content_Size. Window_Size then equals the content size,
|
|
// which is sound because the content is under 256 bytes and every match in it
|
|
// therefore falls inside that window. Same length in, same length out.
|
|
func declareFrameContentSize(frame []byte, size int) []byte {
|
|
if size <= 0 || size >= frameSizeThreshold || len(frame) < 6 || string(frame[:4]) != zstdFrameMagic {
|
|
return frame
|
|
}
|
|
descriptor := frame[4]
|
|
// Rewrite only the exact shape a small input produces: no declared size, no
|
|
// single segment, no dictionary. Anything else either declares a size
|
|
// already or is not a frame this reinterpretation is safe on.
|
|
if descriptor>>6 != 0 || descriptor&frameSingleSegment != 0 || descriptor&frameDictionaryMask != 0 {
|
|
return frame
|
|
}
|
|
reframed := make([]byte, len(frame))
|
|
copy(reframed, frame)
|
|
reframed[4] = descriptor | frameSingleSegment
|
|
reframed[5] = byte(size) // replaces Window_Descriptor
|
|
return reframed
|
|
}
|
|
|
|
// decompressBlob unwraps NWCompressedBuffer framing. It exists so a blob this
|
|
// package wrote — or one upstream wrote — can be compared by its uncompressed
|
|
// bytes, which is the only comparison that is meaningful across zstd
|
|
// implementations.
|
|
func decompressBlob(blob []byte) ([]byte, error) {
|
|
if len(blob) < blobHeaderBytes {
|
|
return nil, fmt.Errorf("blob too small: %d bytes", len(blob))
|
|
}
|
|
header := make([]uint32, 6)
|
|
if err := binary.Read(bytes.NewReader(blob[:blobHeaderBytes]), binary.LittleEndian, header); err != nil {
|
|
return nil, fmt.Errorf("decode blob header: %w", err)
|
|
}
|
|
switch {
|
|
case header[0] != blobMagic:
|
|
return nil, fmt.Errorf("invalid blob magic: %#x", header[0])
|
|
case header[1] != blobVersion:
|
|
return nil, fmt.Errorf("unsupported blob version: %d", header[1])
|
|
case header[2] != algorithmZstd:
|
|
return nil, fmt.Errorf("unsupported compression algorithm: %d", header[2])
|
|
case header[4] != zstdHeaderVer:
|
|
return nil, fmt.Errorf("unsupported zstd header version: %d", header[4])
|
|
case header[5] != zstdDictionary:
|
|
return nil, fmt.Errorf("zstd dictionaries are not supported")
|
|
}
|
|
if header[3] == 0 {
|
|
return nil, nil
|
|
}
|
|
data, err := blobDecoder.DecodeAll(blob[blobHeaderBytes:], nil)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("decompress blob: %w", err)
|
|
}
|
|
if uint32(len(data)) != header[3] {
|
|
return nil, fmt.Errorf("blob size mismatch: header says %d, got %d", header[3], len(data))
|
|
}
|
|
return data, nil
|
|
}
|