sysroot: Support for directories instead of symbolic links in boot part

Allow manipulating and updating /boot/loader entries under a normal
directory, as well as using symbolic links.

For directories this uses `renameat2` to do atomic swap of the loader
directory in the boot partition. It fallsback to non-atomic rename.
This stays atomic on filesystems supporting links but also provide
a non-atomic behavior when filesystem does not provide any atomic
alternative.

/boot/loader as a normal directory is needed by systemd-boot support,
and can be stored under the EFI ESP vfat partition.

Based on the original implementation done by Valentin David [1].

[1] ostreedev#1967

Signed-off-by: Ricardo Salveti <ricardo@foundries.io>
Signed-off-by: Jose Quaresma <jose.quaresma@foundries.io>
Signed-off-by: Igor Opaniuk <igor.opaniuk@foundries.io>

Author: ricardosalveti

src/libostree/ostree-sysroot-deploy.c

@@ -102,6 +102,21 @@ sysroot_flags_to_copy_flags (GLnxFileCopyFlags defaults, OstreeSysrootDebugFlags
   return defaults;
 }
 
+static gboolean
+create_bootself_link (OstreeSysroot *self, GCancellable *cancellable, GError **error)
+{
+  /* When there is no loader, check if the fs supports symlink or not */
+  if (TEMP_FAILURE_RETRY (symlinkat (".", self->sysroot_fd, "boot/boot")) < 0)
+    {
+      if (errno == EPERM)
+        return FALSE;
+      else if (errno != EEXIST)
+        return glnx_throw_errno_prefix (error, "symlinkat");
+    }
+
+  return TRUE;
+}
+
 /* Try a hardlink if we can, otherwise fall back to copying.  Used
  * right now for kernels/initramfs/device trees in /boot, where we can just
  * hardlink if we're on the same partition.
@@ -2190,9 +2205,7 @@ prepare_new_bootloader_link (OstreeSysroot *sysroot, int current_bootversion, in
 
   /* This allows us to support both /boot on a seperate filesystem to / as well
    * as on the same filesystem. */
-  if (TEMP_FAILURE_RETRY (symlinkat (".", sysroot->sysroot_fd, "boot/boot")) < 0)
-    if (errno != EEXIST)
-      return glnx_throw_errno_prefix (error, "symlinkat");
+  (void)create_bootself_link (sysroot, cancellable, error);
 
   g_autofree char *new_target = g_strdup_printf ("loader.%d", new_bootversion);
 
@@ -2204,10 +2217,57 @@ prepare_new_bootloader_link (OstreeSysroot *sysroot, int current_bootversion, in
   return TRUE;
 }
 
+/* We generate the directory on disk, then potentially do a syncfs() to ensure
+ * that it (and everything else we wrote) has hit disk. Only after that do we
+ * rename it into place (via renameat2 RENAME_EXCHANGE).
+ */
+static gboolean
+prepare_new_bootloader_dir (OstreeSysroot *sysroot, int current_bootversion, int new_bootversion,
+                            GCancellable *cancellable, GError **error)
+{
+  GLNX_AUTO_PREFIX_ERROR ("Preparing bootloader directory", error);
+  g_assert ((current_bootversion == 0 && new_bootversion == 1)
+            || (current_bootversion == 1 && new_bootversion == 0));
+
+  if (!_ostree_sysroot_ensure_boot_fd (sysroot, error))
+    return FALSE;
+
+  /* This allows us to support both /boot on a seperate filesystem to / as well
+   * as on the same filesystem. Allowed to fail with EPERM on ESP/vfat.
+   */
+  if (TEMP_FAILURE_RETRY (symlinkat (".", sysroot->sysroot_fd, "boot/boot")) < 0)
+    if (errno != EPERM && errno != EEXIST)
+      return glnx_throw_errno_prefix (error, "symlinkat");
+
+  /* As the directory gets swapped with glnx_renameat2_exchange, the new bootloader
+   * deployment needs to first be moved to the 'old' path, as the 'current' one will
+   * become the older deployment after the exchange.
+   */
+  g_autofree char *loader_new = g_strdup_printf ("loader.%d", new_bootversion);
+  g_autofree char *loader_old = g_strdup_printf ("loader.%d", current_bootversion);
+
+  /* Tag boot version under an ostree specific file */
+  g_autofree char *version_name = g_strdup_printf ("%s/ostree_bootversion", loader_new);
+  if (!glnx_file_replace_contents_at (sysroot->boot_fd, version_name, (guint8 *)loader_new,
+                                      strlen (loader_new), 0, cancellable, error))
+    return FALSE;
+
+  /* It is safe to remove older loader version as it wasn't really deployed */
+  if (!glnx_shutil_rm_rf_at (sysroot->boot_fd, loader_old, cancellable, error))
+    return FALSE;
+
+  /* Rename new deployment to the older path before the exchange */
+  if (!glnx_renameat2_noreplace (sysroot->boot_fd, loader_new, sysroot->boot_fd, loader_old))
+    return FALSE;
+
+  return TRUE;
+}
+
 /* Update the /boot/loader symlink to point to /boot/loader.$new_bootversion */
 static gboolean
-swap_bootloader (OstreeSysroot *sysroot, OstreeBootloader *bootloader, int current_bootversion,
-                 int new_bootversion, GCancellable *cancellable, GError **error)
+swap_bootloader (OstreeSysroot *sysroot, OstreeBootloader *bootloader, gboolean loader_link,
+                 int current_bootversion, int new_bootversion, GCancellable *cancellable,
+                 GError **error)
 {
   GLNX_AUTO_PREFIX_ERROR ("Final bootloader swap", error);
 
@@ -2217,14 +2277,24 @@ swap_bootloader (OstreeSysroot *sysroot, OstreeBootloader *bootloader, int curre
   if (!_ostree_sysroot_ensure_boot_fd (sysroot, error))
     return FALSE;
 
-  g_assert_cmpint (sysroot->boot_fd, !=, -1);
+  if (loader_link)
+    {
+      g_assert_cmpint (sysroot->boot_fd, !=, -1);
 
-  /* The symlink was already written, and we used syncfs() to ensure
-   * its data is in place.  Renaming now should give us atomic semantics;
-   * see https://bugzilla.gnome.org/show_bug.cgi?id=755595
-   */
-  if (!glnx_renameat (sysroot->boot_fd, "loader.tmp", sysroot->boot_fd, "loader", error))
-    return FALSE;
+      /* The symlink was already written, and we used syncfs() to ensure
+       * its data is in place.  Renaming now should give us atomic semantics;
+       * see https://bugzilla.gnome.org/show_bug.cgi?id=755595
+       */
+      if (!glnx_renameat (sysroot->boot_fd, "loader.tmp", sysroot->boot_fd, "loader", error))
+        return FALSE;
+    }
+  else
+    {
+      /* New target is currently under the old/current version */
+      g_autofree char *new_target = g_strdup_printf ("loader.%d", current_bootversion);
+      if (glnx_renameat2_exchange (sysroot->boot_fd, new_target, sysroot->boot_fd, "loader") != 0)
+        return FALSE;
+    }
 
   /* As grub doesn't support replaying XFS journal,
    * we must fsfreeze/thaw again here:
@@ -2447,13 +2517,43 @@ write_deployments_bootswap (OstreeSysroot *self, GPtrArray *new_deployments,
         return glnx_prefix_error (error, "Bootloader write config");
     }
 
-  if (!prepare_new_bootloader_link (self, self->bootversion, new_bootversion, cancellable, error))
+  /* Handle when boot/loader is a symlink (original ostree model) or is a directory (e.g. EFI/vfat)
+   * for recommended systemd type1 BLS
+   */
+  struct stat stbuf;
+  gboolean loader_link = FALSE;
+  if (!glnx_fstatat_allow_noent (self->sysroot_fd, "boot/loader", &stbuf, AT_SYMLINK_NOFOLLOW,
+                                 error))
     return FALSE;
+  if (errno == ENOENT)
+    loader_link = create_bootself_link (self, cancellable, error);
+  else if (S_ISLNK (stbuf.st_mode))
+    loader_link = TRUE;
+  else if (S_ISDIR (stbuf.st_mode))
+    loader_link = FALSE;
+  else
+    return FALSE;
+
+  if (loader_link)
+    {
+      /* Default and when loader is a link is to swap links */
+      if (!prepare_new_bootloader_link (self, self->bootversion, new_bootversion, cancellable,
+                                        error))
+        return FALSE;
+    }
+  else
+    {
+      /* Handle boot/loader as a directory, and swap with renameat2 RENAME_EXCHANGE */
+      if (!prepare_new_bootloader_dir (self, self->bootversion, new_bootversion, cancellable,
+                                       error))
+        return FALSE;
+    }
 
   if (!full_system_sync (self, out_syncstats, cancellable, error))
     return FALSE;
 
-  if (!swap_bootloader (self, bootloader, self->bootversion, new_bootversion, cancellable, error))
+  if (!swap_bootloader (self, bootloader, loader_link, self->bootversion, new_bootversion,
+                        cancellable, error))
     return FALSE;
 
   if (out_subbootdir)

src/libostree/ostree-sysroot.c

@@ -38,6 +38,8 @@
 #include "ostree-sysroot-private.h"
 #include "otcore.h"
 
+#define BOOTVERSION_FILE "boot/loader/ostree_bootversion"
+
 /**
  * SECTION:ostree-sysroot
  * @title: Root partition mount point
@@ -601,6 +603,9 @@ compare_loader_configs_for_sorting (gconstpointer a_pp, gconstpointer b_pp)
   return compare_boot_loader_configs (a, b);
 }
 
+static gboolean read_current_bootversion (OstreeSysroot *self, int *out_bootversion,
+                                          GCancellable *cancellable, GError **error);
+
 /* Read all the bootconfigs from `/boot/loader/`. */
 gboolean
 _ostree_sysroot_read_boot_loader_configs (OstreeSysroot *self, int bootversion,
@@ -613,7 +618,16 @@ _ostree_sysroot_read_boot_loader_configs (OstreeSysroot *self, int bootversion,
   g_autoptr (GPtrArray) ret_loader_configs
       = g_ptr_array_new_with_free_func ((GDestroyNotify)g_object_unref);
 
-  g_autofree char *entries_path = g_strdup_printf ("boot/loader.%d/entries", bootversion);
+  g_autofree char *entries_path = NULL;
+  int current_version;
+  if (!read_current_bootversion (self, &current_version, cancellable, error))
+    return FALSE;
+
+  if (current_version == bootversion)
+    entries_path = g_strdup ("boot/loader/entries");
+  else
+    entries_path = g_strdup_printf ("boot/loader.%d/entries", bootversion);
+
   gboolean entries_exists;
   g_auto (GLnxDirFdIterator) dfd_iter = {
     0,
@@ -660,7 +674,7 @@ _ostree_sysroot_read_boot_loader_configs (OstreeSysroot *self, int bootversion,
   return TRUE;
 }
 
-/* Get the bootversion from the `/boot/loader` symlink. */
+/* Get the bootversion from the `/boot/loader` directory or symlink. */
 static gboolean
 read_current_bootversion (OstreeSysroot *self, int *out_bootversion, GCancellable *cancellable,
                           GError **error)
@@ -673,24 +687,50 @@ read_current_bootversion (OstreeSysroot *self, int *out_bootversion, GCancellabl
     return FALSE;
   if (errno == ENOENT)
     {
-      g_debug ("Didn't find $sysroot/boot/loader symlink; assuming bootversion 0");
+      g_debug ("Didn't find $sysroot/boot/loader directory or symlink; assuming bootversion 0");
       ret_bootversion = 0;
     }
   else
     {
-      if (!S_ISLNK (stbuf.st_mode))
-        return glnx_throw (error, "Not a symbolic link: boot/loader");
-
-      g_autofree char *target
-          = glnx_readlinkat_malloc (self->sysroot_fd, "boot/loader", cancellable, error);
-      if (!target)
-        return FALSE;
-      if (g_strcmp0 (target, "loader.0") == 0)
-        ret_bootversion = 0;
-      else if (g_strcmp0 (target, "loader.1") == 0)
-        ret_bootversion = 1;
+      if (S_ISLNK (stbuf.st_mode))
+        {
+          /* Traditional link, check version by reading link name */
+          g_autofree char *target
+              = glnx_readlinkat_malloc (self->sysroot_fd, "boot/loader", cancellable, error);
+          if (!target)
+            return FALSE;
+          if (g_strcmp0 (target, "loader.0") == 0)
+            ret_bootversion = 0;
+          else if (g_strcmp0 (target, "loader.1") == 0)
+            ret_bootversion = 1;
+          else
+            return glnx_throw (error, "Invalid target '%s' in boot/loader", target);
+        }
       else
-        return glnx_throw (error, "Invalid target '%s' in boot/loader", target);
+        {
+          /* Loader is a directory, check version by reading ostree_bootversion */
+          glnx_autofd int bversion_fd = -1;
+          if (!ot_openat_ignore_enoent (self->sysroot_fd, BOOTVERSION_FILE,
+                                        &bversion_fd, error))
+            return FALSE;
+
+          if (bversion_fd == -1)
+            {
+              g_debug ("File " BOOTVERSION_FILE " is not available, assuming bootversion 0");
+              ret_bootversion = 0;
+            }
+          else
+            {
+              g_autofree char *version
+                  = glnx_fd_readall_utf8 (bversion_fd, NULL, cancellable, error);
+              if (g_strcmp0 (version, "loader.0") == 0)
+                ret_bootversion = 0;
+              else if (g_strcmp0 (version, "loader.1") == 0)
+                ret_bootversion = 1;
+              else
+                return glnx_throw (error, "Invalid version '%s' in " BOOTVERSION_FILE, version);
+            }
+        }
     }
 
   *out_bootversion = ret_bootversion;

src/switchroot/ostree-prepare-root.c

@@ -519,7 +519,7 @@ main (int argc, char *argv[])
    * at /boot inside the deployment. */
   if (snprintf (srcpath, sizeof (srcpath), "%s/boot/loader", root_mountpoint) < 0)
     err (EXIT_FAILURE, "failed to assemble /boot/loader path");
-  if (lstat (srcpath, &stbuf) == 0 && S_ISLNK (stbuf.st_mode))
+  if (lstat (srcpath, &stbuf) == 0 && (S_ISLNK (stbuf.st_mode) || S_ISDIR (stbuf.st_mode)))
     {
       if (lstat ("boot", &stbuf) == 0 && S_ISDIR (stbuf.st_mode))
         {