internal/pkg: streaming archive reader/writer

This is much more robust and efficient than the simple buffering implementation for larger files. Allocations happen almost exclusively in WalkDir.

Signed-off-by: Ophestra <cat@gensokyo.uk>

1 files changed, 278 insertions, 0 deletions

diff --git a/internal/pkg/archive.go b/internal/pkg/archive.go
new file mode 100644
index 00000000..ebda6dbf
--- /dev/null
+++ b/internal/pkg/archive.go
@@ -0,0 +1,278 @@
+package pkg
+
+import (
+	"crypto/sha512"
+	"encoding/binary"
+	"errors"
+	"fmt"
+	"io"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"unsafe"
+
+	"hakurei.app/check"
+)
+
+/*
+| mode uint32 | path_sz uint32 |
+|        data_sz uint64        |
+|         path string          |
+|         data []byte          |
+*/
+
+// An ArchiveHeader represents a single header in an archive.
+type ArchiveHeader struct {
+	Mode fs.FileMode // file mode bits
+	Path string      // pathname of the file
+	Size uint64      // size of data segment
+}
+
+// Writer implements sequential writing of an archive. [Writer.WriteHeader]
+// begins a new file with the provided [ArchiveHeader], and then Writer can be
+// treated as an [io.Writer] to supply that file's data.
+//
+// It is the caller's responsibility to write entries in lexical order.
+type Writer struct {
+	// Underlying writer.
+	w io.Writer
+	// Current header.
+	h ArchiveHeader
+	// Fixed-size header segment.
+	buf [wordSize * 2]byte
+	// Current position in data segment.
+	n uint64
+}
+
+// NewWriter returns the address of a new [Writer] writing to w.
+func NewWriter(w io.Writer) *Writer { return &Writer{w: w} }
+
+var zero [wordSize]byte
+
+// padSize returns the padding size for aligning sz.
+func padSize[T int | uint64](sz T) T {
+	return (wordSize - (sz)%wordSize) % wordSize
+}
+
+// flush concludes writing to the current file and writes padding.
+func (aw *Writer) flush() error {
+	if aw.h.Size > aw.n {
+		return fmt.Errorf("missed writing %d bytes", aw.h.Size-aw.n)
+	} else if aw.h.Size < aw.n {
+		return fmt.Errorf("wrote %d bytes beyond end of file", aw.n-aw.h.Size)
+	}
+
+	if psz := padSize(aw.h.Size); psz != 0 {
+		if _, err := aw.w.Write(zero[:psz]); err != nil {
+			return err
+		}
+	}
+
+	aw.n = 0
+	return nil
+}
+
+// WriteHeader writes h and begins accepting its corresponding file.
+func (aw *Writer) WriteHeader(h *ArchiveHeader) error {
+	if err := aw.flush(); err != nil {
+		return err
+	}
+
+	aw.h = *h
+	binary.LittleEndian.PutUint32(aw.buf[:], uint32(aw.h.Mode))
+	binary.LittleEndian.PutUint32(aw.buf[wordSize/2:], uint32(len(aw.h.Path)))
+	binary.LittleEndian.PutUint64(aw.buf[wordSize:], aw.h.Size)
+	if _, err := aw.w.Write(aw.buf[:]); err != nil {
+		return err
+	} else if _, err = aw.w.Write(
+		unsafe.Slice(unsafe.StringData(aw.h.Path), len(aw.h.Path)),
+	); err != nil {
+		return err
+	} else if psz := padSize(len(aw.h.Path)); psz != 0 {
+		if _, err = aw.w.Write(zero[:psz]); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// Write writes p to the underlying writer and records the new position. Invalid
+// positions are reported by WriteHeader and Close.
+func (aw *Writer) Write(p []byte) (n int, err error) {
+	n, err = aw.w.Write(p)
+	aw.n += uint64(n)
+	return
+}
+
+// Close concludes writing to the archive stream.
+func (aw *Writer) Close() (err error) {
+	err = aw.flush()
+	aw.w = nil
+	return
+}
+
+// ErrInsecurePath is returned by [FlatEntry.Decode] if validation is requested
+// and a nonlocal path is encountered in the stream.
+var ErrInsecurePath = errors.New("insecure file path")
+
+// Reader implements sequential reading of an archive. [Reader.Next] advances to
+// the next file in the archive (including the first), and then Reader can be
+// treated as an [io.Reader] to access the file's data.
+type Reader struct {
+	// Underlying reader.
+	r io.Reader
+	// Fixed-size header segment.
+	buf [wordSize * 2]byte
+	// Remaining bytes in current data segment.
+	n, pad uint64
+}
+
+// NewReader returns the address of a new [Reader] reading from r.
+func NewReader(r io.Reader) *Reader { return &Reader{r: r} }
+
+// Next advances ar to the next entry. Remaining bytes of the current data
+// segment are discarded. Advancing beyond the final entry returns [io.EOF].
+func (ar *Reader) Next() (*ArchiveHeader, error) {
+	if dsz := int64(ar.n + ar.pad); dsz > 0 {
+		if n, err := io.CopyN(io.Discard, ar.r, dsz); err != nil {
+			if errors.Is(err, io.EOF) && n != dsz {
+				err = io.ErrUnexpectedEOF
+			}
+			return nil, err
+		}
+	}
+
+	if n, err := ar.r.Read(ar.buf[:]); err != nil {
+		if errors.Is(err, io.EOF) && n != 0 {
+			err = io.ErrUnexpectedEOF
+		}
+		return nil, err
+	}
+
+	h := ArchiveHeader{
+		Mode: fs.FileMode(binary.LittleEndian.Uint32(ar.buf[:])),
+		Size: binary.LittleEndian.Uint64(ar.buf[wordSize:]),
+	}
+	pathSize := int(binary.LittleEndian.Uint32(ar.buf[wordSize/2:]))
+	pPathSize := alignSize(pathSize)
+
+	buf := make([]byte, pPathSize)
+	if n, err := ar.r.Read(buf); err != nil {
+		if errors.Is(err, io.EOF) {
+			if n != len(buf) {
+				return nil, io.ErrUnexpectedEOF
+			}
+			h.Path = unsafe.String(unsafe.SliceData(buf), pathSize)
+			return &h, err
+		}
+		return nil, err
+	}
+
+	h.Path = unsafe.String(unsafe.SliceData(buf), pathSize)
+	if !filepath.IsLocal(h.Path) {
+		return &h, ErrInsecurePath
+	}
+
+	ar.n = h.Size
+	ar.pad = padSize(h.Size)
+	return &h, nil
+}
+
+// Read implements [io.Reader] for the data segment of the current entry.
+func (ar *Reader) Read(p []byte) (n int, err error) {
+	if uint64(len(p)) > ar.n {
+		p = p[:ar.n]
+	}
+
+	if len(p) > 0 {
+		n, err = ar.r.Read(p)
+		ar.n -= uint64(n)
+	}
+
+	switch err {
+	case io.EOF:
+		if ar.n > 0 {
+			return n, io.ErrUnexpectedEOF
+		}
+
+	case nil:
+		if ar.n == 0 {
+			return n, io.EOF
+		}
+	}
+	return
+}
+
+// Write writes a deterministic representation of the contents of fsys to w.
+// The resulting data can be hashed to produce a deterministic checksum for the
+// directory.
+func Write(fsys fs.FS, root string, w io.Writer) error {
+	aw := NewWriter(w)
+	if err := fs.WalkDir(fsys, root, func(path string, d fs.DirEntry, err error) error {
+		if err != nil {
+			return err
+		}
+
+		var fi fs.FileInfo
+		fi, err = d.Info()
+		if err != nil {
+			return err
+		}
+
+		h := ArchiveHeader{
+			Path: path,
+			Mode: fi.Mode(),
+		}
+		if h.Mode.IsRegular() {
+			h.Size = uint64(fi.Size())
+			if err = aw.WriteHeader(&h); err != nil {
+				return err
+			}
+
+			var r fs.File
+			r, err = fsys.Open(path)
+			if err != nil {
+				return err
+			}
+			_, err = io.Copy(aw, r)
+			if _err := r.Close(); err == nil {
+				err = _err
+			}
+			return err
+		} else if h.Mode&fs.ModeSymlink != 0 {
+			var newpath string
+			if newpath, err = fs.ReadLink(fsys, path); err != nil {
+				return err
+			}
+
+			h.Size = uint64(len(newpath))
+			if err = aw.WriteHeader(&h); err != nil {
+				return err
+			}
+
+			_, err = aw.Write(unsafe.Slice(unsafe.StringData(newpath), len(newpath)))
+			return err
+		} else if !h.Mode.IsDir() {
+			return InvalidFileModeError(h.Mode)
+		}
+		return aw.WriteHeader(&h)
+	}); err != nil {
+		return err
+	}
+	return aw.Close()
+}
+
+// HashFS returns a checksum produced by hashing the result of [Flatten].
+func HashFS(buf *Checksum, fsys fs.FS, root string) error {
+	h := sha512.New384()
+	if err := Write(fsys, root, h); err != nil {
+		return err
+	}
+	h.Sum(buf[:0])
+	return nil
+}
+
+// HashDir returns a checksum produced by hashing the result of [Flatten].
+func HashDir(buf *Checksum, pathname *check.Absolute) error {
+	return HashFS(buf, os.DirFS(pathname.String()), ".")
+}