diff --git a/webdav/file.go b/webdav/file.go
index 097fe6b..33ac017 100644
--- a/webdav/file.go
+++ b/webdav/file.go
@@ -11,6 +11,8 @@
 	"path"
 	"path/filepath"
 	"strings"
+	"sync"
+	"time"
 )
 
 // A FileSystem implements access to a collection of named files. The elements
@@ -88,4 +90,329 @@
 	return os.Stat(name)
 }
 
-// TODO: a MemFS implementation.
+// NewMemFS returns a new in-memory FileSystem implementation.
+func NewMemFS() FileSystem {
+	return &memFS{
+		root: memFSNode{
+			children: make(map[string]*memFSNode),
+			mode:     0660 | os.ModeDir,
+			modTime:  time.Now(),
+		},
+	}
+}
+
+// A memFS implements FileSystem, storing all metadata and actual file data
+// in-memory. No limits on filesystem size are used, so it is not recommended
+// this be used where the clients are untrusted.
+//
+// Concurrent access is permitted. The tree structure is protected by a mutex,
+// and each node's contents and metadata are protected by a per-node mutex.
+//
+// TODO: Enforce file permissions.
+type memFS struct {
+	mu   sync.Mutex
+	root memFSNode
+}
+
+// walk walks the directory tree for the fullname, calling f at each step. If f
+// returns an error, the walk will be aborted and return that same error.
+//
+// Each walk is atomic: fs's mutex is held for the entire operation, including
+// all calls to f.
+//
+// dir is the directory at that step, frag is the name fragment, and final is
+// whether it is the final step. For example, walking "/foo/bar/x" will result
+// in 3 calls to f:
+//   - "/", "foo", false
+//   - "/foo/", "bar", false
+//   - "/foo/bar/", "x", true
+func (fs *memFS) walk(op, fullname string, f func(dir *memFSNode, frag string, final bool) error) error {
+	fs.mu.Lock()
+	defer fs.mu.Unlock()
+
+	original := fullname
+	fullname = path.Clean("/" + fullname)
+
+	// Strip any leading "/"s to make fullname a relative path, as the walk
+	// starts at fs.root.
+	if fullname[0] == '/' {
+		fullname = fullname[1:]
+	}
+	dir := &fs.root
+
+	for {
+		frag, remaining := fullname, ""
+		i := strings.IndexRune(fullname, '/')
+		final := i < 0
+		if !final {
+			frag, remaining = fullname[:i], fullname[i+1:]
+		}
+		if err := f(dir, frag, final); err != nil {
+			return &os.PathError{
+				Op:   op,
+				Path: original,
+				Err:  err,
+			}
+		}
+		if final {
+			break
+		}
+		child := dir.children[frag]
+		if child == nil {
+			return &os.PathError{
+				Op:   op,
+				Path: original,
+				Err:  os.ErrNotExist,
+			}
+		}
+		if !child.IsDir() {
+			return &os.PathError{
+				Op:   op,
+				Path: original,
+				Err:  os.ErrInvalid,
+			}
+		}
+		dir, fullname = child, remaining
+	}
+	return nil
+}
+
+func (fs *memFS) Mkdir(name string, perm os.FileMode) error {
+	return fs.walk("mkdir", name, func(dir *memFSNode, frag string, final bool) error {
+		if !final {
+			return nil
+		}
+		if frag == "" {
+			return os.ErrInvalid
+		}
+		if _, ok := dir.children[frag]; ok {
+			return os.ErrExist
+		}
+		dir.children[frag] = &memFSNode{
+			name:     frag,
+			children: make(map[string]*memFSNode),
+			mode:     perm.Perm() | os.ModeDir,
+			modTime:  time.Now(),
+		}
+		return nil
+	})
+}
+
+func (fs *memFS) OpenFile(name string, flag int, perm os.FileMode) (File, error) {
+	var ret *memFile
+	err := fs.walk("open", name, func(dir *memFSNode, frag string, final bool) error {
+		if !final {
+			return nil
+		}
+		if frag == "" {
+			return os.ErrInvalid
+		}
+		if flag&(os.O_SYNC|os.O_APPEND) != 0 {
+			return os.ErrInvalid
+		}
+		n := dir.children[frag]
+		if flag&os.O_CREATE != 0 {
+			if flag&os.O_EXCL != 0 && n != nil {
+				return os.ErrExist
+			}
+			if n == nil {
+				n = &memFSNode{
+					name: frag,
+					mode: perm.Perm(),
+				}
+				dir.children[frag] = n
+			}
+		}
+		if n == nil {
+			return os.ErrNotExist
+		}
+		if flag&(os.O_WRONLY|os.O_RDWR) != 0 && flag&os.O_TRUNC != 0 {
+			n.mu.Lock()
+			n.data = nil
+			n.mu.Unlock()
+		}
+
+		children := make([]os.FileInfo, 0, len(n.children))
+		for _, c := range n.children {
+			children = append(children, c)
+		}
+		ret = &memFile{
+			n:        n,
+			children: children,
+		}
+		return nil
+	})
+	if err != nil {
+		return nil, err
+	}
+	return ret, nil
+}
+
+func (fs *memFS) RemoveAll(name string) error {
+	return fs.walk("remove", name, func(dir *memFSNode, frag string, final bool) error {
+		if !final {
+			return nil
+		}
+		if frag == "" {
+			return os.ErrInvalid
+		}
+		if _, ok := dir.children[frag]; !ok {
+			return os.ErrNotExist
+		}
+		delete(dir.children, frag)
+		return nil
+	})
+}
+
+func (fs *memFS) Stat(name string) (os.FileInfo, error) {
+	var n *memFSNode
+	err := fs.walk("stat", name, func(dir *memFSNode, frag string, final bool) error {
+		if !final {
+			return nil
+		}
+		if frag == "" {
+			return os.ErrInvalid
+		}
+		n = dir.children[frag]
+		if n == nil {
+			return os.ErrNotExist
+		}
+		return nil
+	})
+	if err != nil {
+		return nil, err
+	}
+	return n, nil
+}
+
+// A memFSNode represents a single entry in the in-memory filesystem and also
+// implements os.FileInfo.
+type memFSNode struct {
+	name string
+
+	mu       sync.Mutex
+	modTime  time.Time
+	mode     os.FileMode
+	children map[string]*memFSNode
+	data     []byte
+}
+
+func (n *memFSNode) Name() string {
+	return n.name
+}
+
+func (n *memFSNode) Size() int64 {
+	n.mu.Lock()
+	defer n.mu.Unlock()
+	return int64(len(n.data))
+}
+
+func (n *memFSNode) Mode() os.FileMode {
+	n.mu.Lock()
+	defer n.mu.Unlock()
+	return n.mode
+}
+
+func (n *memFSNode) ModTime() time.Time {
+	n.mu.Lock()
+	defer n.mu.Unlock()
+	return n.modTime
+}
+
+func (n *memFSNode) IsDir() bool {
+	return n.Mode().IsDir()
+}
+
+func (n *memFSNode) Sys() interface{} {
+	return nil
+}
+
+// A memFile is a File implementation for a memFSNode. It is a per-file (not
+// per-node) read/write position, and if the node is a directory, a snapshot of
+// that node's children.
+type memFile struct {
+	n        *memFSNode
+	children []os.FileInfo
+	// Changes to pos are guarded by n.mu.
+	pos int
+}
+
+func (f *memFile) Close() error {
+	return nil
+}
+
+func (f *memFile) Read(p []byte) (int, error) {
+	if f.n.IsDir() {
+		return 0, os.ErrInvalid
+	}
+	f.n.mu.Lock()
+	defer f.n.mu.Unlock()
+	if f.pos >= len(f.n.data) {
+		return 0, io.EOF
+	}
+	n := copy(p, f.n.data[f.pos:])
+	f.pos += n
+	return n, nil
+}
+
+func (f *memFile) Readdir(count int) ([]os.FileInfo, error) {
+	if !f.n.IsDir() {
+		return nil, os.ErrInvalid
+	}
+	f.n.mu.Lock()
+	defer f.n.mu.Unlock()
+	old := f.pos
+	if old >= len(f.children) {
+		return nil, io.EOF
+	}
+	if count > 0 {
+		f.pos += count
+		if f.pos > len(f.children) {
+			f.pos = len(f.children)
+		}
+	} else {
+		f.pos = len(f.children)
+		old = 0
+	}
+	return f.children[old:f.pos], nil
+}
+
+func (f *memFile) Seek(offset int64, whence int) (int64, error) {
+	f.n.mu.Lock()
+	defer f.n.mu.Unlock()
+	npos := f.pos
+	// TODO: How to handle offsets greater than the size of system int?
+	switch whence {
+	case os.SEEK_SET:
+		npos = int(offset)
+	case os.SEEK_CUR:
+		npos += int(offset)
+	case os.SEEK_END:
+		npos = len(f.n.data) + int(offset)
+	default:
+		npos = -1
+	}
+	if npos < 0 {
+		return 0, os.ErrInvalid
+	}
+	f.pos = npos
+	return int64(f.pos), nil
+}
+
+func (f *memFile) Stat() (os.FileInfo, error) {
+	return f.n, nil
+}
+
+func (f *memFile) Write(p []byte) (int, error) {
+	f.n.mu.Lock()
+	defer f.n.mu.Unlock()
+	epos := len(p) + f.pos
+	if epos > len(f.n.data) {
+		d := make([]byte, epos)
+		copy(d, f.n.data)
+		f.n.data = d
+	}
+	copy(f.n.data[f.pos:], p)
+	f.n.modTime = time.Now()
+	return len(p), nil
+}