JuiceFS Local Filestore Backend Joins Object Keys onto the Storage Root Without a Containment Check, Allowing Writes Outside the Configured Directory

The filestore backend in pkg/object/file.go, used for file:// stores and as a common juicefs sync destination, derived every operation's target from path(key), which returned either filepath.Join(d.root, key) or filepath.Clean(d.root + key) with no check that the result stayed beneath the root. Put, Get, Head, Delete, Chmod, Chown, Symlink and Readlink all consumed that value directly. Object keys enumerated from a source object store during a sync are not constrained the way local filesystem names are, so a key containing traversal segments causes juicefs to write attacker-supplied content to a path outside the intended local destination, and no error is returned. An operator syncing from a bucket whose contents they do not fully control, such as a shared or public bucket or one an attacker can write to, is therefore exposed to a file write at an attacker-influenced location. The fix changes path() to return an error and rejects any key whose resolved path escapes the root.

Quoted source text, attributed separately from HOL analysis.