container/initdev: mount tmpfs on shm for ro dev

Programs expect /dev/shm to be a writable tmpfs.

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

container/container_test.go

@@ -100,6 +100,7 @@ var containerTestCases = []struct {
 			ent("/tty", "/dev/tty", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
 			ent("/", "/dev/pts", "rw,nosuid,noexec,relatime", "devpts", "devpts", "rw,mode=620,ptmxmode=666"),
 			ent("/", "/dev/mqueue", "rw,nosuid,nodev,noexec,relatime", "mqueue", "mqueue", "rw"),
+			ent("/", "/dev/shm", "rw,nosuid,nodev,relatime", "tmpfs", "tmpfs", ignore),
 		),
 		1971, 100, nil, 0, seccomp.PresetStrict},
 
@@ -116,6 +117,7 @@ var containerTestCases = []struct {
 			ent("/urandom", "/dev/urandom", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
 			ent("/tty", "/dev/tty", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
 			ent("/", "/dev/pts", "rw,nosuid,noexec,relatime", "devpts", "devpts", "rw,mode=620,ptmxmode=666"),
+			ent("/", "/dev/shm", "rw,nosuid,nodev,relatime", "tmpfs", "tmpfs", ignore),
 		),
 		1971, 100, nil, 0, seccomp.PresetStrict},
 

container/initdev.go

@@ -72,8 +72,9 @@ func (d *MountDevOp) apply(state *setupState, k syscallDispatcher) error {
 		}
 	}
 
+	devShmPath := path.Join(target, "shm")
 	devPtsPath := path.Join(target, "pts")
-	for _, name := range []string{path.Join(target, "shm"), devPtsPath} {
+	for _, name := range []string{devShmPath, devPtsPath} {
 		if err := k.mkdir(name, state.ParentPerm); err != nil {
 			return wrapErrSelf(err)
 		}
@@ -117,8 +118,12 @@ func (d *MountDevOp) apply(state *setupState, k syscallDispatcher) error {
 	if d.Write {
 		return nil
 	}
-	return wrapErrSuffix(k.remount(target, MS_RDONLY),
-		fmt.Sprintf("cannot remount %q:", target))
+
+	if err := k.remount(target, MS_RDONLY); err != nil {
+		return wrapErrSuffix(k.remount(target, MS_RDONLY),
+			fmt.Sprintf("cannot remount %q:", target))
+	}
+	return k.mountTmpfs(SourceTmpfs, devShmPath, MS_NOSUID|MS_NODEV, 0, 01777)
 }
 
 func (d *MountDevOp) Is(op Op) bool {

container/initdev_test.go

@@ -645,6 +645,7 @@ func TestMountDevOp(t *testing.T) {
 			{"readlink", expectArgs{"/host/proc/self/fd/1"}, "/dev/pts/2", nil},
 			{"bindMount", expectArgs{"/host/dev/pts/2", "/sysroot/dev/console", uintptr(0), false}, nil, nil},
 			{"remount", expectArgs{"/sysroot/dev", uintptr(1)}, nil, nil},
+			{"mountTmpfs", expectArgs{"tmpfs", "/sysroot/dev/shm", uintptr(0x6), 0, os.FileMode(01777)}, nil, nil},
 		}, nil},
 
 		{"success rw", &Params{ParentPerm: 0750, RetainSession: true}, &MountDevOp{
@@ -715,6 +716,7 @@ func TestMountDevOp(t *testing.T) {
 			{"mkdir", expectArgs{"/sysroot/dev/mqueue", os.FileMode(0750)}, nil, nil},
 			{"mount", expectArgs{"mqueue", "/sysroot/dev/mqueue", "mqueue", uintptr(0xe), ""}, nil, nil},
 			{"remount", expectArgs{"/sysroot/dev", uintptr(1)}, nil, nil},
+			{"mountTmpfs", expectArgs{"tmpfs", "/sysroot/dev/shm", uintptr(0x6), 0, os.FileMode(01777)}, nil, nil},
 		}, nil},
 	})
 

container/mount.go

@@ -43,6 +43,8 @@ const (
 	// Note that any source value is allowed when fstype is [FstypeOverlay].
 	SourceOverlay = "overlay"
 
+	// SourceTmpfs is used when mounting tmpfs.
+	SourceTmpfs = "tmpfs"
 	// SourceTmpfsRootfs is used when mounting the tmpfs instance backing the intermediate root.
 	SourceTmpfsRootfs = "rootfs"
 	// SourceTmpfsDevtmpfs is used when mounting tmpfs representing a subset of host devtmpfs.

internal/app/app_nixos_linux_test.go

@@ -148,6 +148,7 @@ var testCasesNixos = []sealTestCase{
 				Etc(m("/etc/"), "8e2c76b066dabe574cf073bdb46eb5c1").
 				Bind(m("/var/lib/persist/module/hakurei/0/1"), m("/var/lib/persist/module/hakurei/0/1"), container.BindWritable|container.BindEnsure).
 				Remount(m("/dev/"), syscall.MS_RDONLY).
+				Tmpfs(m("/dev/shm"), 0, 01777).
 				Tmpfs(m("/run/user/"), 4096, 0755).
 				Bind(m("/tmp/hakurei.1971/runtime/1"), m("/run/user/1971"), container.BindWritable).
 				Bind(m("/tmp/hakurei.1971/tmpdir/1"), m("/tmp/"), container.BindWritable).

internal/app/app_pd_linux_test.go

@@ -53,6 +53,7 @@ var testCasesPd = []sealTestCase{
 				Tmpfs(m("/run/user/1971"), 8192, 0755).
 				Tmpfs(m("/run/dbus"), 8192, 0755).
 				Remount(m("/dev/"), syscall.MS_RDONLY).
+				Tmpfs(m("/dev/shm"), 0, 01777).
 				Tmpfs(m("/run/user/"), 4096, 0755).
 				Bind(m("/tmp/hakurei.1971/runtime/0"), m("/run/user/65534"), container.BindWritable).
 				Bind(m("/tmp/hakurei.1971/tmpdir/0"), m("/tmp/"), container.BindWritable).
@@ -189,6 +190,7 @@ var testCasesPd = []sealTestCase{
 				Tmpfs(m("/run/user/1971"), 8192, 0755).
 				Tmpfs(m("/run/dbus"), 8192, 0755).
 				Remount(m("/dev/"), syscall.MS_RDONLY).
+				Tmpfs(m("/dev/shm"), 0, 01777).
 				Tmpfs(m("/run/user/"), 4096, 0755).
 				Bind(m("/tmp/hakurei.1971/runtime/9"), m("/run/user/65534"), container.BindWritable).
 				Bind(m("/tmp/hakurei.1971/tmpdir/9"), m("/tmp/"), container.BindWritable).

internal/app/container_linux.go

@@ -233,7 +233,9 @@ func newContainer(s *hst.ContainerConfig, os sys.State, prefix string, uid, gid
 
 	// no more ContainerConfig paths beyond this point
 	if !s.Device {
-		params.Remount(container.AbsFHSDev, syscall.MS_RDONLY)
+		params.
+			Remount(container.AbsFHSDev, syscall.MS_RDONLY).
+			Tmpfs(container.AbsFHSDev.Append("shm"), 0, 01777)
 	}
 
 	return params, maps.Clone(s.Env), nil

test/sandbox/case/mapuid.nix

@@ -72,7 +72,7 @@ in
         ptmx = fs "80001ff" null null;
         pts = fs "800001ed" { ptmx = fs "42001b6" null null; } null;
         random = fs "42001b6" null null;
-        shm = fs "800001ed" { } null;
+        shm = fs "801001ff" { } null;
         stderr = fs "80001ff" null null;
         stdin = fs "80001ff" null null;
         stdout = fs "80001ff" null null;
@@ -256,6 +256,7 @@ in
       (ent "/" "/.hakurei/store" "rw,relatime" "overlay" "overlay" "rw,lowerdir=/host/nix/.ro-store:/host/nix/.rw-store/upper,upperdir=/host/tmp/.hakurei-store-rw/upper,workdir=/host/tmp/.hakurei-store-rw/work,redirect_dir=nofollow,userxattr")
       (ent "/etc" ignore "ro,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/var/lib/hakurei/u0/a3" "/var/lib/hakurei/u0/a3" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
+      (ent "/" "/dev/shm" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,uid=1000003,gid=1000003")
       (ent "/" "/run/user" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,size=4k,mode=755,uid=1000003,gid=1000003")
       (ent "/tmp/hakurei.0/runtime/3" "/run/user/1000" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/tmp/hakurei.0/tmpdir/3" "/tmp" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")

test/sandbox/case/pd.nix

@@ -36,7 +36,7 @@
         ptmx = fs "80001ff" null null;
         pts = fs "800001ed" { ptmx = fs "42001b6" null null; } null;
         random = fs "42001b6" null null;
-        shm = fs "800001ed" { } null;
+        shm = fs "801001ff" { } null;
         stderr = fs "80001ff" null null;
         stdin = fs "80001ff" null null;
         stdout = fs "80001ff" null null;
@@ -185,6 +185,7 @@
       (ent "/etc" ignore "ro,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/" "/run/user/1000" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,size=8k,mode=755,uid=1000000,gid=1000000")
       (ent "/" "/run/dbus" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,size=8k,mode=755,uid=1000000,gid=1000000")
+      (ent "/" "/dev/shm" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,uid=1000000,gid=1000000")
       (ent "/" "/run/user" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,size=4k,mode=755,uid=1000000,gid=1000000")
       (ent "/tmp/hakurei.0/runtime/0" "/run/user/65534" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/tmp/hakurei.0/tmpdir/0" "/tmp" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")

test/sandbox/case/pdlike.nix

@@ -70,7 +70,7 @@ in
         ptmx = fs "80001ff" null null;
         pts = fs "800001ed" { ptmx = fs "42001b6" null null; } null;
         random = fs "42001b6" null null;
-        shm = fs "800001ed" { } null;
+        shm = fs "801001ff" { } null;
         stderr = fs "80001ff" null null;
         stdin = fs "80001ff" null null;
         stdout = fs "80001ff" null null;
@@ -251,6 +251,7 @@ in
       (ent "/dri" "/dev/dri" "rw,nosuid" "devtmpfs" "devtmpfs" ignore)
       (ent "/etc" ignore "ro,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/var/lib/hakurei/u0/a5" "/var/lib/hakurei/u0/a5" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
+      (ent "/" "/dev/shm" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,uid=1000005,gid=1000005")
       (ent "/" "/run/user" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,size=4k,mode=755,uid=1000005,gid=1000005")
       (ent "/tmp/hakurei.0/runtime/5" "/run/user/65534" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/tmp/hakurei.0/tmpdir/5" "/tmp" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")

test/sandbox/case/preset.nix

@@ -69,7 +69,7 @@ in
         ptmx = fs "80001ff" null null;
         pts = fs "800001ed" { ptmx = fs "42001b6" null null; } null;
         random = fs "42001b6" null null;
-        shm = fs "800001ed" { } null;
+        shm = fs "801001ff" { } null;
         stderr = fs "80001ff" null null;
         stdin = fs "80001ff" null null;
         stdout = fs "80001ff" null null;
@@ -249,6 +249,7 @@ in
       (ent "/dri" "/dev/dri" "rw,nosuid" "devtmpfs" "devtmpfs" ignore)
       (ent "/etc" ignore "ro,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/var/lib/hakurei/u0/a1" "/var/lib/hakurei/u0/a1" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
+      (ent "/" "/dev/shm" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,uid=1000001,gid=1000001")
       (ent "/" "/run/user" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,size=4k,mode=755,uid=1000001,gid=1000001")
       (ent "/tmp/hakurei.0/runtime/1" "/run/user/65534" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/tmp/hakurei.0/tmpdir/1" "/tmp" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")

test/sandbox/case/tty.nix

@@ -74,7 +74,7 @@ in
         ptmx = fs "80001ff" null null;
         pts = fs "800001ed" { ptmx = fs "42001b6" null null; } null;
         random = fs "42001b6" null null;
-        shm = fs "800001ed" { } null;
+        shm = fs "801001ff" { } null;
         stderr = fs "80001ff" null null;
         stdin = fs "80001ff" null null;
         stdout = fs "80001ff" null null;
@@ -261,6 +261,7 @@ in
       (ent "/" "/.hakurei/store" "rw,relatime" "overlay" "overlay" "rw,lowerdir=/host/nix/.ro-store:/host/nix/.rw-store/upper,upperdir=/host/tmp/.hakurei-store-rw/upper,workdir=/host/tmp/.hakurei-store-rw/work,redirect_dir=nofollow,uuid=on,userxattr")
       (ent "/etc" ignore "ro,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/var/lib/hakurei/u0/a2" "/var/lib/hakurei/u0/a2" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
+      (ent "/" "/dev/shm" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,uid=1000002,gid=1000002")
       (ent "/" "/run/user" "rw,nosuid,nodev,relatime" "tmpfs" "ephemeral" "rw,size=4k,mode=755,uid=1000002,gid=1000002")
       (ent "/tmp/hakurei.0/runtime/2" "/run/user/65534" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")
       (ent "/tmp/hakurei.0/tmpdir/2" "/tmp" "rw,nosuid,nodev,relatime" "ext4" "/dev/disk/by-label/nixos" "rw")