pre-allocate storage for metadata json files, see containers/podman#1…

…3967

Keeping a temporary file of at least the same size as the target file
for atomic writes helps reduce the probability of running out of space
when deleting entities from corresponding metadata in a disk full
scenario. This strategy is applied to writing containers.json,
layers.json, images.json and mountpoints.json.

Signed-off-by: Denys Knertser <denys@netze.io>

containers.go

@@ -541,13 +541,13 @@ func (r *containerStore) save(saveLocations containerLocations) error {
 		if err != nil {
 			return err
 		}
-		var opts *ioutils.AtomicFileWriterOptions
+		opts := ioutils.AtomicFileWriterOptions{
+			PreAllocate: true,
+		}
 		if location == volatileContainerLocation {
-			opts = &ioutils.AtomicFileWriterOptions{
-				NoSync: true,
-			}
+			opts.NoSync = true
 		}
-		if err := ioutils.AtomicWriteFileWithOpts(rpath, jdata, 0600, opts); err != nil {
+		if err := ioutils.AtomicWriteFileWithOpts(rpath, jdata, 0600, &opts); err != nil {
 			return err
 		}
 	}

images.go

@@ -536,7 +536,10 @@ func (r *imageStore) Save() error {
 	if err != nil {
 		return err
 	}
-	if err := ioutils.AtomicWriteFile(rpath, jdata, 0600); err != nil {
+	opts := ioutils.AtomicFileWriterOptions{
+		PreAllocate: true,
+	}
+	if err := ioutils.AtomicWriteFileWithOpts(rpath, jdata, 0600, &opts); err != nil {
 		return err
 	}
 	lw, err := r.lockfile.RecordWrite()

layers.go

@@ -942,7 +942,9 @@ func (r *layerStore) saveLayers(saveLocations layerLocations) error {
 		if err != nil {
 			return err
 		}
-		opts := ioutils.AtomicFileWriterOptions{}
+		opts := ioutils.AtomicFileWriterOptions{
+			PreAllocate: true,
+		}
 		if location == volatileLayerLocation {
 			opts.NoSync = true
 		}
@@ -984,7 +986,10 @@ func (r *layerStore) saveMounts() error {
 	if err != nil {
 		return err
 	}
-	if err = ioutils.AtomicWriteFile(mpath, jmdata, 0600); err != nil {
+	opts := ioutils.AtomicFileWriterOptions{
+		PreAllocate: true,
+	}
+	if err = ioutils.AtomicWriteFileWithOpts(mpath, jmdata, 0600, &opts); err != nil {
 		return err
 	}
 	lw, err := r.mountsLockfile.RecordWrite()

pkg/ioutils/fswriters.go

@@ -2,6 +2,8 @@ package ioutils
 
 import (
 	"io"
+	"strconv"
+	"math/rand"
 	"os"
 	"path/filepath"
 	"time"
@@ -17,6 +19,9 @@ type AtomicFileWriterOptions struct {
 	// On successful return from Close() this is set to the mtime of the
 	// newly written file.
 	ModTime time.Time
+	// Whenever an atomic file is successfully written create a temporary
+	// file of the same size in order to pre-allocate storage for the next write
+	PreAllocate bool
 }
 
 var defaultWriterOptions = AtomicFileWriterOptions{}
@@ -38,13 +43,23 @@ func NewAtomicFileWriterWithOpts(filename string, perm os.FileMode, opts *Atomic
 // temporary file and closing it atomically changes the temporary file to
 // destination path. Writing and closing concurrently is not allowed.
 func newAtomicFileWriter(filename string, perm os.FileMode, opts *AtomicFileWriterOptions) (*atomicFileWriter, error) {
-	f, err := os.CreateTemp(filepath.Dir(filename), ".tmp-"+filepath.Base(filename))
-	if err != nil {
-		return nil, err
-	}
+	dir := filepath.Dir(filename)
+	base := filepath.Base(filename)
 	if opts == nil {
 		opts = &defaultWriterOptions
 	}
+	random := ""
+	// need predictable name when pre-allocated
+	if ! opts.PreAllocate {
+		random = strconv.FormatUint(rand.Uint64(), 36)
+	}
+	tmp := filepath.Join(dir, ".tmp" + random + "-" + base)
+	// if pre-allocated the temporary file exists and contains some data
+	// do not truncate it here, instead truncate at Close()
+	f, err := os.OpenFile(tmp, os.O_WRONLY|os.O_CREATE, perm)
+	if err != nil {
+		return nil, err
+	}
 	abspath, err := filepath.Abs(filename)
 	if err != nil {
 		return nil, err
@@ -54,6 +69,7 @@ func newAtomicFileWriter(filename string, perm os.FileMode, opts *AtomicFileWrit
 		fn:     abspath,
 		perm:   perm,
 		noSync: opts.NoSync,
+		preAllocate: opts.PreAllocate,
 	}, nil
 }
 
@@ -96,6 +112,7 @@ type atomicFileWriter struct {
 	writeErr error
 	perm     os.FileMode
 	noSync   bool
+	preAllocate bool
 	modTime  time.Time
 }
 
@@ -108,22 +125,35 @@ func (w *atomicFileWriter) Write(dt []byte) (int, error) {
 }
 
 func (w *atomicFileWriter) Close() (retErr error) {
-	defer func() {
-		if retErr != nil || w.writeErr != nil {
-			os.Remove(w.f.Name())
+	if ! w.preAllocate {
+		defer func() {
+			if retErr != nil || w.writeErr != nil {
+				os.Remove(w.f.Name())
+			}
+		}()
+	} else {
+		truncateAt, err := w.f.Seek(0, io.SeekCurrent);
+		if err != nil {
+			return err
+		}
+		err = w.f.Truncate(truncateAt);
+		if err != nil {
+			return err
 		}
-	}()
+	}
 	if !w.noSync {
 		if err := fdatasync(w.f); err != nil {
 			w.f.Close()
 			return err
 		}
 	}
 
+	var size int64 = 0
 	// fstat before closing the fd
 	info, statErr := w.f.Stat()
 	if statErr == nil {
 		w.modTime = info.ModTime()
+		size = info.Size()
 	}
 	// We delay error reporting until after the real call to close()
 	// to match the traditional linux close() behaviour that an fd
@@ -138,11 +168,51 @@ func (w *atomicFileWriter) Close() (retErr error) {
 		return statErr
 	}
 
-	if err := os.Chmod(w.f.Name(), w.perm); err != nil {
+	tmpName := w.f.Name()
+	if err := os.Chmod(tmpName, w.perm); err != nil {
 		return err
 	}
 	if w.writeErr == nil {
-		return os.Rename(w.f.Name(), w.fn)
+		if w.preAllocate {
+			err := swapOrMove(tmpName, w.fn)
+			if err != nil {
+				return err
+			}
+			// ignore errors, this is a best effort operation
+			preAllocate(tmpName, size, w.perm)
+			return nil
+		} else {
+			return os.Rename(tmpName, w.fn)
+		}
+	}
+	return nil
+}
+
+// ensure that the file is of at least indicated size
+func preAllocate(filename string, size int64, perm os.FileMode) error {
+	f, err := os.OpenFile(filename, os.O_WRONLY|os.O_CREATE|os.O_APPEND, perm)
+	if err != nil {
+		return err
+	}
+	defer f.Close()
+	info, err := f.Stat()
+	if err != nil {
+		return err
+	}
+	extendBytes := size - info.Size()
+	if extendBytes > 0 {
+		var blockSize int64 = 65536
+		block := make([]byte, blockSize)
+		for extendBytes > 0 {
+			if blockSize > extendBytes {
+				blockSize = extendBytes
+			}
+			_, err := f.Write(block[:blockSize])
+			if err != nil {
+				return err
+			}
+			extendBytes -= blockSize
+		}
 	}
 	return nil
 }

pkg/ioutils/fswriters_linux.go

@@ -9,3 +9,12 @@ import (
 func fdatasync(f *os.File) error {
 	return unix.Fdatasync(int(f.Fd()))
 }
+
+func swapOrMove(oldpath string, newpath string) error {
+	err := unix.Renameat2(unix.AT_FDCWD, oldpath, unix.AT_FDCWD, newpath, unix.RENAME_EXCHANGE)
+	if err != nil {
+		// unlikely that rename will succeed if renameat2 failed, but just in case
+		err = os.Rename(oldpath, newpath)
+	}
+	return err
+}

pkg/ioutils/fswriters_unsupported.go

@@ -10,3 +10,7 @@ import (
 func fdatasync(f *os.File) error {
 	return f.Sync()
 }
+
+func swapOrMove(oldpath string, newpath string) error {
+	return os.Rename(oldpath, newpath)
+}