# OL-4256 — Update swap filesystem error
> The atomic binary swap failed at the filesystem layer. The running
> binary is left in place; the staging tempdir is dropped.
## When this fires
- Insufficient permissions to write at the running binary's path.
- ENOSPC / quota / read-only filesystem.
- Windows: AV / EDR locked the running PE while `self_replace` tried
to move-aside.
## How to fix it
1. Check the CLI's stderr — the underlying `std::io::Error` is
surfaced verbatim.
2. Linux/macOS: confirm the running user owns the binary path or has
write permission on its directory.
3. Windows: if AV is the cause, schedule the apply during a
maintenance window or temporarily allowlist the binary path.
4. If the `<exe>.bak` sibling is left on disk after the failure, run
`openlatch-provider --version` to confirm the running binary is
still the OLD one, then delete `<exe>.bak` manually if the apply
is being abandoned.
## Related
- `OL-4257` — supervisor-restart-loop rollback (a different fail-safe
for a different scenario).
- `.claude/rules/auto-update.md` — single-binary swap pipeline.