fix: retry getting xattrs on concurrent write error (#700)

Author: LarsJurgensen

pkg/storage/utils/decomposedfs/metadata/errors.go

@@ -45,6 +45,19 @@ func IsNotExist(err error) bool {
 	return false
 }
 
+// IsErrRange checks for a syscall.ERANGE buried inside an xattr error. listxattr
+// returns ERANGE when the set of attribute names grows between the size-probe call
+// and the read call (a concurrent setxattr landed in between). This is transient and
+// self-clearing once the competing writer finishes its batch.
+func IsErrRange(err error) bool {
+	if xerr, ok := errors.Cause(err).(*xattr.Error); ok {
+		if serr, ok2 := xerr.Err.(syscall.Errno); ok2 {
+			return serr == syscall.ERANGE
+		}
+	}
+	return false
+}
+
 // IsAttrUnset checks the xattr.ENOATTR from the xattr package which redifines it as ENODATA on platforms that do not natively support it (eg. linux)
 // see https://github.com/pkg/xattr/blob/8725d4ccc0fcef59c8d9f0eaf606b3c6f962467a/xattr_linux.go#L19-L22
 func IsAttrUnset(err error) bool {

pkg/storage/utils/decomposedfs/metadata/xattrs_backend.go

@@ -26,6 +26,7 @@ import (
 	"strconv"
 	"strings"
 
+	"github.com/owncloud/reva/v2/pkg/appctx"
 	"github.com/owncloud/reva/v2/pkg/storage/cache"
 	"github.com/owncloud/reva/v2/pkg/storage/utils/decomposedfs/metadata/prefixes"
 	"github.com/owncloud/reva/v2/pkg/storage/utils/filelocks"
@@ -82,12 +83,32 @@ func (b XattrsBackend) List(ctx context.Context, filePath string) (attribs []str
 	return b.list(ctx, filePath, true)
 }
 
+const listMaxRetries = 10
+
 func (b XattrsBackend) list(ctx context.Context, filePath string, acquireLock bool) (attribs []string, err error) {
 	attrs, err := xattr.List(filePath)
 	if err == nil {
 		return attrs, nil
 	}
 
+	// ERANGE means a concurrent writer (e.g. the posix inotify assimilate worker)
+	// mutated the attribute set mid-read. It is transient and self-clearing: retry
+	// until the set stabilises.
+	if IsErrRange(err) {
+		for attempt := 1; attempt <= listMaxRetries; attempt++ {
+			appctx.GetLogger(ctx).Debug().Err(err).Str("path", filePath).Int("attempt", attempt).Msg("xattr.List ERANGE, retrying")
+			attrs, err = xattr.List(filePath)
+			if err == nil {
+				appctx.GetLogger(ctx).Debug().Str("path", filePath).Int("attempt", attempt).Msg("xattr.List ERANGE recovered after retry")
+				return attrs, nil
+			}
+			if !IsErrRange(err) {
+				break
+			}
+		}
+		appctx.GetLogger(ctx).Error().Err(err).Str("path", filePath).Msg("xattr.List still failing after ERANGE retries")
+	}
+
 	// listing xattrs failed, try again, either with lock or without
 	if acquireLock {
 		f, err := lockedfile.OpenFile(filePath+filelocks.LockFileSuffix, os.O_CREATE|os.O_WRONLY, 0600)