Support recursive read-only (RRO) mounts: `nerdctl run -v /foo:/bar:r…

…ro,rprivate`

`nerdctl run -v /foo:/bar:rro,rprivate` appends "rro" (in addition to "ro") to the OCI mount option string slice.

Requires opencontainers/runc PR 3272.
Older version of runc just ignores "rro" option.

The "rro" option string conforms to the proposal in util-linux/util-linux Issue 1501.

Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>

README.md

@@ -120,6 +120,8 @@ Also, `nerdctl` might be potentially useful for debugging Kubernetes clusters, b
 Major:
 - [On-demand image pulling (lazy-pulling) using Stargz Snapshotter](./docs/stargz.md): `nerdctl --snapshotter=stargz run IMAGE` .
 - [Image encryption and decryption using ocicrypt (imgcrypt)](./docs/ocicrypt.md): `nerdctl image (encrypt|decrypt) SRC DST`
+- Recursive read-only (RRO) bind-mount: `nerdctl run -v /mnt:/mnt:rro` (make children such as `/mnt/usb` to be read-only, too).
+  Requires kernel >= 5.12 and runc PR [#3272](https://github.com/opencontainers/runc/pull/3272).
 
 Minor:
 - Namespacing: `nerdctl --namespace=<NS> ps` .
@@ -332,7 +334,13 @@ Runtime flags:
 - :whale: `--sysctl`: Sysctl options, e.g \"net.ipv4.ip_forward=1\"
 
 Volume flags:
-- :whale: `-v, --volume`: Bind mount a volume
+- :whale: `-v, --volume <SRC>:<DST>[:<OPT>]`: Bind mount a volume, e.g., `-v /mnt:/mnt:rro,rprivate`
+  - :whale:     option `rw` : Read/Write (when writable)
+  - :whale:     option `ro` : Non-recursive read-only
+  - :nerd_face: option `rro`: Recursive read-only. Should be used in conjunction with `rprivate`. e.g., `-v /mnt:/mnt:rro,rprivate` makes children such as `/mnt/usb` to be read-only, too.
+    Requires kernel >= 5.12 and runc PR [#3272](https://github.com/opencontainers/runc/pull/3272). With older runc, `rro` just works as `ro`.
+  - :whale:     option `shared`, `slave`, `private`: Non-recursive "shared" / "slave" / "private" propagation
+  - :whale:     option `rshared`, `rslave`, `rprivate`: Recursive "shared" / "slave" / "private" propagation
 - :whale: `--tmpfs`: Mount a tmpfs directory
 
 Rootfs flags:

pkg/mountutil/mountutil_linux.go

@@ -94,9 +94,7 @@ func parseVolumeOptionsWithMountInfo(vType, src, optsRaw string, getMountInfoFun
 	)
 	for _, opt := range strings.Split(optsRaw, ",") {
 		switch opt {
-		case "rw":
-			writeModeRawOpts = append(writeModeRawOpts, opt)
-		case "ro":
+		case "rw", "ro", "rro":
 			writeModeRawOpts = append(writeModeRawOpts, opt)
 		case "private", "rprivate", "shared", "rshared", "slave", "rslave":
 			propagationRawOpts = append(propagationRawOpts, opt)
@@ -112,9 +110,24 @@ func parseVolumeOptionsWithMountInfo(vType, src, optsRaw string, getMountInfoFun
 
 	if len(writeModeRawOpts) > 1 {
 		return nil, nil, fmt.Errorf("duplicated read/write volume option: %+v", writeModeRawOpts)
-	} else if len(writeModeRawOpts) > 0 && writeModeRawOpts[0] == "ro" {
-		opts = append(opts, "ro")
-	} // No need to return option when "rw"
+	} else if len(writeModeRawOpts) > 0 {
+		switch writeModeRawOpts[0] {
+		case "ro":
+			opts = append(opts, "ro")
+		case "rro":
+			// Mount option "rro" is supported since runc vX.X (https://github.com/opencontainers/runc/pull/3272), with kernel >= 5.12.
+			// Older version of runc just ignores "rro", so we have to add "ro" too, to our best effort.
+			opts = append(opts, "ro", "rro")
+			if len(propagationRawOpts) != 1 || propagationRawOpts[0] != "rprivate" {
+				logrus.Warn("Mount option \"rro\" should be used in conjunction with \"rprivate\"")
+			}
+		case "rw":
+			// NOP
+		default:
+			// NOTREACHED
+			return nil, nil, fmt.Errorf("unexpected writeModeRawOpts[0]=%q", writeModeRawOpts[0])
+		}
+	}
 
 	if len(propagationRawOpts) > 1 {
 		return nil, nil, fmt.Errorf("duplicated volume propagation option: %+v", propagationRawOpts)