prepare-root: Add support for root.transient

Closes: https://github.com/ostreedev/ostree/issues/3113
It'd greatly improve compatibility with things like RPMs that install
in `/opt` if we supported a full "original docker" style model where
`/` is a transient overlayfs.  We'd still keep our semantics for `/etc`
and `/var` by default, but e.g. we'd stop recommending
`/opt` :arrow_right: `/var/opt`, in this model,
so `/opt` would be on the overlayfs.

Note this all aligns with composefs, where we'd actually be making
`/` a *read-only* overlayfs by default; it'd be really nice of course
to *implement* this by just making the composefs overlayfs writable,
but I am not sure we can hard require composefs for this right now.

So this change adds support for `root.transient = true`
in `/usr/lib/ostree/prepare-root.conf`.

The major downside is that people could be surprised if files they
write to e.g. `/opt` don't persist across upgrades.  But, that's
already again how it works since Docker started.

Note as part of the implementation of this, we need to add a whole
new "backing" directory distinct from the deployment directories.

(Tangentially related to this, it's tempting to switch to always
 using a *read-only* overlay mount by default.

diff --git a/src/libostree/ostree-sysroot-cleanup.c b/src/libostree/ostree-sysroot-cleanup.c
index ba9b3bacdaa18d976a40919eb0feb75147603405..e5ffc5a7542241b6a93a68ad75b5182005ffe8c4 100644 (file)
--- a/src/libostree/ostree-sysroot-cleanup.c
+++ b/src/libostree/ostree-sysroot-cleanup.c
@@ -218,6 +218,7 @@ gboolean
 _ostree_sysroot_rmrf_deployment (OstreeSysroot *self, OstreeDeployment *deployment,
                                  GCancellable *cancellable, GError **error)
 {
+  g_autofree char *backing_relpath = _ostree_sysroot_get_deployment_backing_relpath (deployment);
   g_autofree char *origin_relpath = ostree_deployment_get_origin_relpath (deployment);
   g_autofree char *deployment_path = ostree_sysroot_get_deployment_dirpath (self, deployment);
   struct stat stbuf;
@@ -238,6 +239,12 @@ _ostree_sysroot_rmrf_deployment (OstreeSysroot *self, OstreeDeployment *deployme
   /* This deployment wasn't referenced, so delete it */
   if (!_ostree_linuxfs_fd_alter_immutable_flag (deployment_fd, FALSE, cancellable, error))
     return FALSE;
+  /* Note we must delete the origin and backing directories first, as the "source of truth"
+   * is the deployment path.  We don't currently have code that detects "orphaned"
+   * origin files or work directories.
+   */
+  if (!glnx_shutil_rm_rf_at (self->sysroot_fd, backing_relpath, cancellable, error))
+    return FALSE;
   if (!glnx_shutil_rm_rf_at (self->sysroot_fd, origin_relpath, cancellable, error))
     return FALSE;
   if (!glnx_shutil_rm_rf_at (self->sysroot_fd, deployment_path, cancellable, error))

diff --git a/src/libostree/ostree-sysroot-deploy.c b/src/libostree/ostree-sysroot-deploy.c
index a809d560a7dd688ba58bee42455daa1a7be0b469..ebc0b4fcb298aad3ae8e058d94f9c0c08bb45e81 100644 (file)
--- a/src/libostree/ostree-sysroot-deploy.c
+++ b/src/libostree/ostree-sysroot-deploy.c
@@ -3078,6 +3078,10 @@ sysroot_initialize_deployment (OstreeSysroot *self, const char *osname, const ch
   if (!require_stateroot (self, osname, error))
     return FALSE;
 
+  g_autofree char *stateroot_backing = g_strdup_printf ("ostree/deploy/%s/backing", osname);
+  if (!glnx_shutil_mkdir_p_at (self->sysroot_fd, stateroot_backing, 0700, cancellable, error))
+    return glnx_prefix_error (error, "Creating backing directory");
+
   OstreeRepo *repo = ostree_sysroot_repo (self);
 
   gint new_deployserial;
@@ -3295,6 +3299,49 @@ sysroot_finalize_selinux_policy (int deployment_dfd, GError **error)
 }
 #endif /* HAVE_SELINUX */
 
+static gboolean
+sysroot_initialize_deployment_backing (OstreeSysroot *self, OstreeDeployment *deployment,
+                                       OstreeSePolicy *sepolicy, GError **error)
+{
+  GLNX_AUTO_PREFIX_ERROR ("Preparing deployment backing dir", error);
+  g_autofree char *deployment_path = ostree_sysroot_get_deployment_dirpath (self, deployment);
+  g_autofree char *backing_relpath = _ostree_sysroot_get_deployment_backing_relpath (deployment);
+  struct stat stbuf;
+
+  if (!glnx_fstatat (self->sysroot_fd, deployment_path, &stbuf, AT_SYMLINK_NOFOLLOW, error))
+    return FALSE;
+
+  // Create the "backing" directory with additional data */
+  if (!glnx_ensure_dir (self->sysroot_fd, backing_relpath, 0700, error))
+    return glnx_prefix_error (error, "Creating backing dir");
+
+  // The root-transient holds overlayfs directories for the root
+  g_autofree char *rootovldir
+      = g_build_filename (backing_relpath, OSTREE_DEPLOYMENT_ROOT_TRANSIENT_DIR, NULL);
+  if (!glnx_ensure_dir (self->sysroot_fd, rootovldir, 0700, error))
+    return glnx_prefix_error (error, "Creating root ovldir");
+
+  // The overlayfs work (subdirectory of root-transient)
+  g_autofree char *workdir = g_build_filename (rootovldir, "work", NULL);
+  if (!glnx_ensure_dir (self->sysroot_fd, workdir, 0700, error))
+    return glnx_prefix_error (error, "Creating work dir");
+
+  // Create the overlayfs upper; this needs to have the same mode and SELinux label as the root
+  {
+    g_auto (OstreeSepolicyFsCreatecon) con = {
+      0,
+    };
+
+    if (!_ostree_sepolicy_preparefscreatecon (&con, sepolicy, "/", stbuf.st_mode, error))
+      return glnx_prefix_error (error, "Looking up SELinux label for /");
+    g_autofree char *upperdir = g_build_filename (rootovldir, "upper", NULL);
+    if (!glnx_ensure_dir (self->sysroot_fd, upperdir, stbuf.st_mode, error))
+      return glnx_prefix_error (error, "Creating upper dir");
+  }
+
+  return TRUE;
+}
+
 static gboolean
 sysroot_finalize_deployment (OstreeSysroot *self, OstreeDeployment *deployment,
                              OstreeDeployment *merge_deployment, GCancellable *cancellable,
@@ -3360,6 +3407,9 @@ sysroot_finalize_deployment (OstreeSysroot *self, OstreeDeployment *deployment,
   if (!selinux_relabel_var_if_needed (self, sepolicy, os_deploy_dfd, cancellable, error))
     return FALSE;
 
+  if (!sysroot_initialize_deployment_backing (self, deployment, sepolicy, error))
+    return FALSE;
+
   /* Rewrite the origin using the final merged selinux config, just to be
    * conservative about getting the right labels.
    */

diff --git a/src/libostree/ostree-sysroot-private.h b/src/libostree/ostree-sysroot-private.h
index 7373b19b63c9384cf8961233c333903fd218ffdd..5cbeae0ecb5bd63c33456c4808d8b7bdfe20418d 100644 (file)
--- a/src/libostree/ostree-sysroot-private.h
+++ b/src/libostree/ostree-sysroot-private.h
@@ -142,6 +142,8 @@ gboolean _ostree_sysroot_boot_complete (OstreeSysroot *self, GCancellable *cance
 
 OstreeDeployment *_ostree_sysroot_deserialize_deployment_from_variant (GVariant *v, GError **error);
 
+char *_ostree_sysroot_get_deployment_backing_relpath (OstreeDeployment *deployment);
+
 gboolean _ostree_sysroot_rmrf_deployment (OstreeSysroot *sysroot, OstreeDeployment *deployment,
                                           GCancellable *cancellable, GError **error);
 

diff --git a/src/libostree/ostree-sysroot.c b/src/libostree/ostree-sysroot.c
index 62adc6221c4c96debaba95219c1e72f23c6d78ba..5bcae1c79a0c7e7be60a649aa15f1cbffc456746 100644 (file)
--- a/src/libostree/ostree-sysroot.c
+++ b/src/libostree/ostree-sysroot.c
@@ -36,7 +36,7 @@
 #include "ostree-repo-private.h"
 #include "ostree-sepolicy-private.h"
 #include "ostree-sysroot-private.h"
-#include "ostree.h"
+#include "otcore.h"
 
 /**
  * SECTION:ostree-sysroot
@@ -1965,6 +1965,17 @@ ostree_sysroot_simple_write_deployment (OstreeSysroot *sysroot, const char *osna
   return TRUE;
 }
 
+/* Return the sysroot-relative path to the "backing" directory of a deployment
+ * which can hold additional data.
+ */
+char *
+_ostree_sysroot_get_deployment_backing_relpath (OstreeDeployment *deployment)
+{
+  return g_strdup_printf (
+      "ostree/deploy/%s/backing/%s.%d", ostree_deployment_get_osname (deployment),
+      ostree_deployment_get_csum (deployment), ostree_deployment_get_deployserial (deployment));
+}
+
 /* Deploy a copy of @target_deployment */
 static gboolean
 clone_deployment (OstreeSysroot *sysroot, OstreeDeployment *target_deployment,

diff --git a/src/libotcore/otcore.h b/src/libotcore/otcore.h
index 1593e7b77f325411262fbbd69729e88e728623d1..4d51398e6a686f2fa6d6ef8a33bf1f4e187dffac 100644 (file)
--- a/src/libotcore/otcore.h
+++ b/src/libotcore/otcore.h
@@ -55,6 +55,11 @@ GKeyFile *otcore_load_config (int rootfs, const char *filename, GError **error);
 // we make it with mode 0 (which requires CAP_DAC_OVERRIDE to pass through).
 #define OTCORE_RUN_OSTREE_PRIVATE "/run/ostree/.private"
 
+// The directory holding extra/backing data for a deployment, such as overlayfs workdirs
+#define OSTREE_DEPLOYMENT_BACKING_DIR "backing"
+// The directory holding the root overlayfs
+#define OSTREE_DEPLOYMENT_ROOT_TRANSIENT_DIR "root-transient"
+
 // The name of the composefs metadata root
 #define OSTREE_COMPOSEFS_NAME ".ostree.cfs"
 // The temporary directory used for the EROFS mount; it's in the .private directory
@@ -70,6 +75,8 @@ GKeyFile *otcore_load_config (int rootfs, const char *filename, GError **error);
 // This key if present contains the public key successfully used
 // to verify the signature.
 #define OTCORE_RUN_BOOTED_KEY_COMPOSEFS_SIGNATURE "composefs.signed"
+// This key will be present if the root is transient
+#define OTCORE_RUN_BOOTED_KEY_ROOT_TRANSIENT "root.transient"
 // This key will be present if the sysroot-ro flag was found
 #define OTCORE_RUN_BOOTED_KEY_SYSROOT_RO "sysroot-ro"
 

diff --git a/src/switchroot/ostree-prepare-root.c b/src/switchroot/ostree-prepare-root.c
index 41fc8f98c3ef40a70889b3a5edb8b6bbac67dcb7..8d0ab88e44527cc801bad78d609984b1b75201f6 100644 (file)
--- a/src/switchroot/ostree-prepare-root.c
+++ b/src/switchroot/ostree-prepare-root.c
@@ -87,6 +87,8 @@
 #define SYSROOT_KEY "sysroot"
 #define READONLY_KEY "readonly"
 
+/* This key configures the / mount in the deployment root */
+#define ROOT_KEY "root"
 #define ETC_KEY "etc"
 #define TRANSIENT_KEY "transient"
 
@@ -352,6 +354,11 @@ main (int argc, char *argv[])
     errx (EXIT_FAILURE, "Failed to parse config: %s", error->message);
 
   gboolean sysroot_readonly = FALSE;
+  gboolean root_transient = FALSE;
+
+  if (!ot_keyfile_get_boolean_with_default (config, ROOT_KEY, TRANSIENT_KEY, FALSE, &root_transient,
+                                            &error))
+    return FALSE;
 
   // We always parse the composefs config, because we want to detect and error
   // out if it's enabled, but not supported at compile time.
@@ -375,6 +382,11 @@ main (int argc, char *argv[])
   if (root_mountpoint == NULL)
     err (EXIT_FAILURE, "realpath(\"%s\")", root_arg);
   g_autofree char *deploy_path = resolve_deploy_path (root_mountpoint);
+  const char *deploy_directory_name = glnx_basename (deploy_path);
+  // Note that realpath() should have stripped any trailing `/` which shouldn't
+  // be in the karg to start with, but we assert here to be sure we have a non-empty
+  // filename.
+  g_assert (deploy_directory_name && *deploy_directory_name);
 
   if (mkdirat (AT_FDCWD, OTCORE_RUN_OSTREE, 0755) < 0)
     err (EXIT_FAILURE, "Failed to create %s", OTCORE_RUN_OSTREE);
@@ -510,13 +522,36 @@ main (int argc, char *argv[])
     errx (EXIT_FAILURE, "composefs: enabled at runtime, but support is not compiled in");
 #endif
 
-  if (!using_composefs)
+  if (root_transient)
+    {
+      /* if (using_composefs)
+       * TODO: Add support to libcomposefs to mount writably; for now we end up with two overlayfs
+       * which is a bit silly.
+       */
+
+      g_autofree char *backingdir = g_strdup_printf ("../../backing/%s", deploy_directory_name);
+      g_autofree char *workdir
+          = g_build_filename (backingdir, OSTREE_DEPLOYMENT_ROOT_TRANSIENT_DIR, "work", NULL);
+      g_autofree char *upperdir
+          = g_build_filename (backingdir, OSTREE_DEPLOYMENT_ROOT_TRANSIENT_DIR, "upper", NULL);
+      g_autofree char *ovl_options
+          = g_strdup_printf ("lowerdir=.,upperdir=%s,workdir=%s", upperdir, workdir);
+      if (mount ("overlay", TMP_SYSROOT, "overlay", MS_SILENT, ovl_options) < 0)
+        err (EXIT_FAILURE, "failed to mount transient root overlayfs");
+      g_print ("Enabled transient /\n");
+    }
+  else if (!using_composefs)
     {
+      g_print ("Using legacy ostree bind mount for /\n");
       /* The deploy root starts out bind mounted to sysroot.tmp */
       if (mount (deploy_path, TMP_SYSROOT, NULL, MS_BIND | MS_SILENT, NULL) < 0)
         err (EXIT_FAILURE, "failed to make initial bind mount %s", deploy_path);
     }
 
+  /* Pass on the state  */
+  g_variant_builder_add (&metadata_builder, "{sv}", OTCORE_RUN_BOOTED_KEY_ROOT_TRANSIENT,
+                         g_variant_new_boolean (root_transient));
+
   /* This will result in a system with /sysroot read-only. Thus, two additional
    * writable bind-mounts (for /etc and /var) are required later on. */
   if (sysroot_readonly)
@@ -548,7 +583,7 @@ main (int argc, char *argv[])
   /* Prepare /etc.
    * No action required if sysroot is writable. Otherwise, a bind-mount for
    * the deployment needs to be created and remounted as read/write. */
-  if (sysroot_readonly || using_composefs)
+  if (sysroot_readonly || using_composefs || root_transient)
     {
       gboolean etc_transient = FALSE;
       if (!ot_keyfile_get_boolean_with_default (config, ETC_KEY, TRANSIENT_KEY, FALSE,