WIP: Add support for --replace-mode=alongside for ostree target

Ironically our support for `--replace-mode=alongside` breaks
when we're targeting an already extant ostree host, because when
we first blow away the `/boot` directory, this means the ostree
stack loses its knowledge that we're in a booted deployment,
and will attempt to GC it...

ostreedev/ostree-rs-ext@8fa019b
is a key part of the fix for that.

However, a notable improvement we can do here is to grow this
whole thing into a real "factory reset" mode, and this will
be a compelling answer to
coreos/fedora-coreos-tracker#399

To implement this though we need to support configuring the
stateroot and not just hardcode `default`.

Signed-off-by: Omer Tuchfeld <omer@tuchfeld.dev>

Author: cgwalters

lib/src/install.rs

@@ -564,26 +564,34 @@ pub(crate) fn print_configuration() -> Result<()> {
 }
 
 #[context("Creating ostree deployment")]
-async fn initialize_ostree_root(state: &State, root_setup: &RootSetup) -> Result<Storage> {
+async fn initialize_ostree_root(state: &State, root_setup: &RootSetup) -> Result<(Storage, bool)> {
     let sepolicy = state.load_policy()?;
     let sepolicy = sepolicy.as_ref();
     // Load a fd for the mounted target physical root
     let rootfs_dir = &root_setup.rootfs_fd;
     let rootfs = root_setup.rootfs.as_path();
     let cancellable = gio::Cancellable::NONE;
 
+    let stateroot = state.stateroot();
+
+    let has_ostree = rootfs_dir.try_exists("ostree/repo")?;
+    if !has_ostree {
+        Task::new_and_run(
+            "Initializing ostree layout",
+            "ostree",
+            ["admin", "init-fs", "--modern", rootfs.as_str()],
+        )?;
+    } else {
+        println!("Reusing extant ostree layout");
+        let path = ".".into();
+        let _ = crate::utils::open_dir_remount_rw(rootfs_dir, path)
+            .context("remounting sysroot as read-write")?;
+    }
+
     // Ensure that the physical root is labeled.
     // Another implementation: https://github.com/coreos/coreos-assembler/blob/3cd3307904593b3a131b81567b13a4d0b6fe7c90/src/create_disk.sh#L295
     crate::lsm::ensure_dir_labeled(rootfs_dir, "", Some("/".into()), 0o755.into(), sepolicy)?;
 
-    let stateroot = state.stateroot();
-
-    Task::new_and_run(
-        "Initializing ostree layout",
-        "ostree",
-        ["admin", "init-fs", "--modern", rootfs.as_str()],
-    )?;
-
     // And also label /boot AKA xbootldr, if it exists
     let bootdir = rootfs.join("boot");
     if bootdir.try_exists()? {
@@ -607,9 +615,14 @@ async fn initialize_ostree_root(state: &State, root_setup: &RootSetup) -> Result
     let sysroot = ostree::Sysroot::new(Some(&gio::File::for_path(rootfs)));
     sysroot.load(cancellable)?;
 
-    sysroot
-        .init_osname(stateroot, cancellable)
-        .context("initializing stateroot")?;
+    let stateroot_exists = rootfs_dir.try_exists(format!("ostree/deploy/{stateroot}"))?;
+    if stateroot_exists {
+        anyhow::bail!("Cannot redeploy over extant stateroot {stateroot}");
+    } else {
+        sysroot
+            .init_osname(stateroot, cancellable)
+            .context("initializing stateroot")?;
+    }
 
     let sysroot_dir = Dir::reopen_dir(&crate::utils::sysroot_fd(&sysroot))?;
 
@@ -637,14 +650,15 @@ async fn initialize_ostree_root(state: &State, root_setup: &RootSetup) -> Result
     let sysroot = ostree::Sysroot::new(Some(&gio::File::for_path(rootfs)));
     sysroot.load(cancellable)?;
     let sysroot = SysrootLock::new_from_sysroot(&sysroot).await?;
-    Storage::new(sysroot, &temp_run)
+    Ok((Storage::new(sysroot, &temp_run)?, has_ostree))
 }
 
 #[context("Creating ostree deployment")]
 async fn install_container(
     state: &State,
     root_setup: &RootSetup,
     sysroot: &ostree::Sysroot,
+    has_ostree: bool,
 ) -> Result<(ostree::Deployment, InstallAleph)> {
     let sepolicy = state.load_policy()?;
     let sepolicy = sepolicy.as_ref();
@@ -730,6 +744,7 @@ async fn install_container(
     options.kargs = Some(kargs.as_slice());
     options.target_imgref = Some(&state.target_imgref);
     options.proxy_cfg = proxy_cfg;
+    options.no_clean = has_ostree;
     let imgstate = crate::utils::async_task_with_spinner(
         "Deploying container image",
         ostree_container::deploy::deploy(&sysroot, stateroot, &src_imageref, Some(options)),
@@ -1272,10 +1287,11 @@ async fn install_with_sysroot(
     sysroot: &Storage,
     boot_uuid: &str,
     bound_images: &[crate::boundimage::ResolvedBoundImage],
+    has_ostree: bool,
 ) -> Result<()> {
     // And actually set up the container in that root, returning a deployment and
     // the aleph state (see below).
-    let (_deployment, aleph) = install_container(state, rootfs, &sysroot).await?;
+    let (_deployment, aleph) = install_container(state, rootfs, &sysroot, has_ostree).await?;
     // Write the aleph data that captures the system state at the time of provisioning for aid in future debugging.
     rootfs
         .rootfs_fd
@@ -1336,6 +1352,19 @@ async fn install_to_filesystem_impl(state: &State, rootfs: &mut RootSetup) -> Re
         .ok_or_else(|| anyhow!("No uuid for boot/root"))?;
     tracing::debug!("boot uuid={boot_uuid}");
 
+    // If we're doing an alongside install, then the /dev bootupd sees needs to be the host's.
+    // What we probably really want to do here is tunnel in the host's /dev properly, but for now
+    // just copy /dev/disk
+    if rootfs.skip_finalize {
+        if !Utf8Path::new("/dev/disk").try_exists()? {
+            Task::new_and_run(
+                "Copying host /dev/disk",
+                "cp",
+                ["-a", "/proc/1/root/dev/disk", "/dev/disk"],
+            )?;
+        }
+    }
+
     let bound_images = if state.config_opts.skip_bound_images {
         Vec::new()
     } else {
@@ -1356,8 +1385,16 @@ async fn install_to_filesystem_impl(state: &State, rootfs: &mut RootSetup) -> Re
 
     // Initialize the ostree sysroot (repo, stateroot, etc.)
     {
-        let sysroot = initialize_ostree_root(state, rootfs).await?;
-        install_with_sysroot(state, rootfs, &sysroot, &boot_uuid, &bound_images).await?;
+        let (sysroot, has_ostree) = initialize_ostree_root(state, rootfs).await?;
+        install_with_sysroot(
+            state,
+            rootfs,
+            &sysroot,
+            &boot_uuid,
+            &bound_images,
+            has_ostree,
+        )
+        .await?;
         // We must drop the sysroot here in order to close any open file
         // descriptors.
     }
@@ -1498,7 +1535,8 @@ fn remove_all_in_dir_no_xdev(d: &Dir) -> Result<()> {
 
 #[context("Removing boot directory content")]
 fn clean_boot_directories(rootfs: &Dir) -> Result<()> {
-    let bootdir = rootfs.open_dir(BOOT).context("Opening /boot")?;
+    let bootdir =
+        crate::utils::open_dir_remount_rw(rootfs, BOOT.into()).context("Opening /boot")?;
     // This should not remove /boot/efi note.
     remove_all_in_dir_no_xdev(&bootdir)?;
     if ARCH_USES_EFI {
@@ -1589,12 +1627,35 @@ pub(crate) async fn install_to_filesystem(
     if !st.is_dir() {
         anyhow::bail!("Not a directory: {root_path}");
     }
+
+    let inspect = crate::mount::inspect_filesystem(&fsopts.root_path)?;
+
+    let alternative_root = fsopts.root_path.join("sysroot");
+    let root_path = match inspect.source.as_str() {
+        // Our target filesystem is an overlay, the true root is in `/sysroot`
+        "overlay" => {
+            tracing::debug!(
+                "Overlay filesystem detected, using {alternative_root} instead of {root_path} as target root"
+            );
+            &alternative_root
+        }
+        _ => root_path,
+    };
     let rootfs_fd = Dir::open_ambient_dir(root_path, cap_std::ambient_authority())
         .with_context(|| format!("Opening target root directory {root_path}"))?;
+
+    tracing::debug!("Root filesystem: {root_path}");
+
     if let Some(false) = ostree_ext::mountutil::is_mountpoint(&rootfs_fd, ".")? {
         anyhow::bail!("Not a mountpoint: {root_path}");
     }
 
+    let fsopts = {
+        let mut fsopts = fsopts.clone();
+        fsopts.root_path = root_path.clone();
+        fsopts
+    };
+
     // Gather global state, destructuring the provided options.
     // IMPORTANT: We might re-execute the current process in this function (for SELinux among other things)
     // IMPORTANT: and hence anything that is done before MUST BE IDEMPOTENT.

lib/src/lsm.rs

@@ -333,7 +333,6 @@ pub(crate) fn ensure_dir_labeled_recurse(
     Ok(())
 }
 
-/// A wrapper for creating a directory, also optionally setting a SELinux label.
 #[cfg(feature = "install")]
 pub(crate) fn ensure_dir_labeled(
     root: &Dir,
@@ -365,6 +364,11 @@ pub(crate) fn ensure_dir_labeled(
 
     root.ensure_dir_with(local_destname, &DirBuilder::new())
         .with_context(|| format!("Opening {local_destname}"))?;
+    // tracing::trace!("Rooted at {}", root.canonicalize()
+    tracing::trace!(
+        "Going to open {} for fchmod",
+        local_destname.as_std_path().to_str().unwrap()
+    );
     let dirfd = cap_std_ext::cap_primitives::fs::open(
         &root.as_filelike_view(),
         local_destname.as_std_path(),

lib/src/utils.rs

@@ -5,10 +5,13 @@ use std::process::Command;
 use std::time::Duration;
 
 use anyhow::{Context, Result};
-use cap_std_ext::cap_std::fs::Dir;
+use camino::Utf8Path;
+use cap_std_ext::{cap_std::fs::Dir, prelude::CapStdExtCommandExt};
+use fn_error_context::context;
 use ostree::glib;
 use ostree_ext::container::SignatureSource;
 use ostree_ext::ostree;
+use std::os::fd::AsFd;
 
 /// Try to look for keys injected by e.g. rpm-ostree requesting machine-local
 /// changes; if any are present, return `true`.
@@ -55,6 +58,44 @@ pub(crate) fn find_mount_option<'a>(
         .next()
 }
 
+/// Try to (heuristically) determine if the provided path is a mount root.
+pub(crate) fn is_mountpoint(root: &Dir, path: &Utf8Path) -> Result<Option<bool>> {
+    // https://github.com/systemd/systemd/blob/8fbf0a214e2fe474655b17a4b663122943b55db0/src/basic/mountpoint-util.c#L176
+    use rustix::fs::{AtFlags, StatxFlags};
+
+    // SAFETY(unwrap): We can infallibly convert an i32 into a u64.
+    let mountroot_flag: u64 = libc::STATX_ATTR_MOUNT_ROOT.try_into().unwrap();
+    match rustix::fs::statx(
+        root.as_fd(),
+        path.as_std_path(),
+        AtFlags::NO_AUTOMOUNT | AtFlags::SYMLINK_NOFOLLOW,
+        StatxFlags::empty(),
+    ) {
+        Ok(r) => {
+            let present = (r.stx_attributes_mask & mountroot_flag) > 0;
+            Ok(present.then(|| r.stx_attributes & mountroot_flag > 0))
+        }
+        Err(e) if e == rustix::io::Errno::NOSYS => Ok(None),
+        Err(e) => Err(e.into()),
+    }
+}
+
+/// Given a target directory, if it's a read-only mount, then remount it writable
+#[context("Opening {target} with writable mount")]
+pub(crate) fn open_dir_remount_rw(root: &Dir, target: &Utf8Path) -> Result<Dir> {
+    if is_mountpoint(root, target)?.unwrap_or_default() {
+        tracing::debug!("Target {target} is a mountpoint, remounting rw");
+        let st = Command::new("mount")
+            .args(["-o", "remount,rw", target.as_str()])
+            .cwd_dir(root.try_clone()?)
+            .status()?;
+        if !st.success() {
+            anyhow::bail!("Failed to remount: {st:?}");
+        }
+    }
+    root.open_dir(target).map_err(anyhow::Error::new)
+}
+
 pub(crate) fn spawn_editor(tmpf: &tempfile::NamedTempFile) -> Result<()> {
     let editor_variables = ["EDITOR"];
     // These roughly match https://github.com/systemd/systemd/blob/769ca9ab557b19ee9fb5c5106995506cace4c68f/src/shared/edit-util.c#L275