diff --git a/updatehub/src/object/installer/mod.rs b/updatehub/src/object/installer/mod.rs
index 67e408d6..d00e4f8f 100644
--- a/updatehub/src/object/installer/mod.rs
+++ b/updatehub/src/object/installer/mod.rs
@@ -129,7 +129,7 @@ where
 }
 
 #[cfg(test)]
-mod tests {
+pub(crate) mod tests {
     use super::*;
     use lazy_static::lazy_static;
     use std::{
diff --git a/updatehub/src/utils/fs.rs b/updatehub/src/utils/fs.rs
index 887e9323..96300a60 100644
--- a/updatehub/src/utils/fs.rs
+++ b/updatehub/src/utils/fs.rs
@@ -9,7 +9,7 @@ use pkg_schema::definitions::{
     target_permissions::{Gid, Uid},
 };
 use slog_scope::trace;
-use std::{io, path::Path};
+use std::{io, io::Seek, os::unix::fs::FileTypeExt, path::Path};
 use sys_mount::{Mount, Unmount, UnmountDrop};
 
 pub(crate) struct MountGuard {
@@ -25,11 +25,7 @@ impl MountGuard {
 
 pub(crate) fn ensure_disk_space(target: &Path, required: u64) -> Result<()> {
     trace!("looking for {} free bytes on {:?}", required, target);
-    let stat = nix::sys::statvfs::statvfs(target)?;
-
-    // stat fields might be 32 or 64 bytes depending on host arch
-    #[allow(clippy::useless_conversion)]
-    let available = u64::from(stat.block_size() * stat.blocks_free());
+    let available = available_space(target)?;
 
     if required > available {
         return Err(Error::NotEnoughSpace { available, required });
@@ -37,6 +33,30 @@ pub(crate) fn ensure_disk_space(target: &Path, required: u64) -> Result<()> {
     Ok(())
 }
 
+/// Returns how many bytes may be written to `target`.
+///
+/// `statvfs` reports on the filesystem holding the given path, so pointing it
+/// at a device node measures the devtmpfs mounted on `/dev` -- which is sized
+/// after the available RAM -- instead of the device the objects are installed
+/// onto. Device nodes are therefore asked for their own capacity, which a seek
+/// to the end reports for block, MTD and UBI volume devices alike.
+fn available_space(target: &Path) -> Result<u64> {
+    let file_type = std::fs::metadata(target)?.file_type();
+
+    if file_type.is_block_device() || file_type.is_char_device() {
+        trace!("{:?} is a device node, asking it for its capacity", target);
+        return Ok(std::fs::File::open(target)?.seek(io::SeekFrom::End(0))?);
+    }
+
+    let stat = nix::sys::statvfs::statvfs(target)?;
+
+    // stat fields might be 32 or 64 bits wide depending on the host arch, so widen
+    // them before multiplying to keep filesystems bigger than 4 GiB from
+    // overflowing on 32 bit targets
+    #[allow(clippy::useless_conversion)]
+    Ok(u64::from(stat.fragment_size()) * u64::from(stat.blocks_free()))
+}
+
 pub(crate) fn is_executable_in_path(cmd: &str) -> Result<()> {
     trace!("checking if {} is executable", cmd);
     match quale::which(cmd) {
@@ -102,3 +122,86 @@ pub(crate) fn chown(path: &Path, uid: &Option<Uid>, gid: &Option<Gid>) -> Result
         gid.as_ref().map(|id| nix::unistd::Gid::from_raw(id.as_u32())),
     )?)
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::object::installer::tests::SERIALIZE;
+    use pretty_assertions::assert_eq;
+    use std::path::PathBuf;
+
+    // Big enough to tell a loop device apart from the devtmpfs on /dev, small
+    // enough to stay sparse on any builder.
+    const LOOP_DEVICE_SIZE: u64 = 64 * 1024 * 1024;
+
+    struct FakeLoopDevice {
+        loopdev: loopdev::LoopDevice,
+        device: PathBuf,
+        _backing_file: tempfile::NamedTempFile,
+    }
+
+    impl FakeLoopDevice {
+        fn new(size: u64) -> Result<FakeLoopDevice> {
+            let backing_file = tempfile::NamedTempFile::new()?;
+            backing_file.as_file().set_len(size)?;
+
+            // Loop device next_free is not thread safe
+            let mutex = SERIALIZE.clone();
+            let _mutex = mutex.lock().unwrap();
+            let loopdev = loopdev::LoopControl::open()?.next_free()?;
+            let device = loopdev.path().unwrap();
+            loopdev.attach_file(backing_file.path())?;
+
+            Ok(FakeLoopDevice { loopdev, device, _backing_file: backing_file })
+        }
+    }
+
+    impl Drop for FakeLoopDevice {
+        fn drop(&mut self) {
+            if let Err(e) = self.loopdev.detach() {
+                eprintln!("Failed to cleanup FakeLoopDevice, Error: {e}");
+            }
+        }
+    }
+
+    #[test]
+    fn regular_files_are_measured_by_their_filesystem() {
+        let dir = tempfile::tempdir().unwrap();
+        let file = dir.path().join("object");
+        std::fs::write(&file, [0; 16]).unwrap();
+
+        // The seek a device node is asked for would answer with the file's own
+        // 16 bytes instead.
+        assert!(available_space(&file).unwrap() > 16);
+    }
+
+    #[test]
+    fn a_requirement_beyond_the_available_space_is_rejected() {
+        let dir = tempfile::tempdir().unwrap();
+        let available = available_space(dir.path()).unwrap();
+
+        ensure_disk_space(dir.path(), available).unwrap();
+        assert!(matches!(
+            ensure_disk_space(dir.path(), available + 1),
+            Err(Error::NotEnoughSpace { .. })
+        ));
+    }
+
+    // Requires root, as creating a loop device does.
+    #[test]
+    #[ignore]
+    fn device_nodes_are_measured_by_their_own_capacity() {
+        let loop_device = FakeLoopDevice::new(LOOP_DEVICE_SIZE).unwrap();
+
+        // Asking for the directory holding the node is asking the devtmpfs mounted
+        // on /dev, which is the answer the check used to settle for.
+        assert_ne!(available_space(Path::new("/dev")).unwrap(), LOOP_DEVICE_SIZE);
+
+        assert_eq!(available_space(&loop_device.device).unwrap(), LOOP_DEVICE_SIZE);
+        ensure_disk_space(&loop_device.device, LOOP_DEVICE_SIZE).unwrap();
+        assert!(matches!(
+            ensure_disk_space(&loop_device.device, LOOP_DEVICE_SIZE + 1),
+            Err(Error::NotEnoughSpace { .. })
+        ));
+    }
+}