Linux file locks are process-wide, not tied to specific file descriptors. This means SQLite's multiple connections within one process can accidentally override each other's locks. Native locks work between processes but fail for intra-process concurrency.

SQLite solves this with an abstraction layer. It tracks locks per inode using a `unixinodeinfo` structure, grouping all file descriptors for the same file. This prevents the kernel from silently overwriting locks when the same process opens the same database multiple times.

The `unixinodeinfo` object is created once per inode per process, stored in a global list with a mutex for thread safety. Reference counting tracks open connections. Lock requests are managed internally first; only real transitions trigger a system call, minimizing overhead and preventing conflicts.

This design ensures serializable isolation for multi-connection, multi-threaded applications. Without it, SQLite would be unsafe. The `unixFile` object per connection ties back to the shared `unixinodeinfo`, providing a robust lock-tracking system that native OS locks alone cannot deliver.