In a database you have to lock access at the file level, naturally. To do multi-process locks you need to store your locks either in shared memory on in an mmap block so that they're shared between processes and then you have to deal with the case of processes crashing or being killed without releasing their locks, which can then trigger a deadlock because the lock is held by a process that doesn't exist anymore.

To resolve that you can pop up to the process tree, assuming you saved the PID along with the lock and see if that process still exists (and hope its PID hasn't been reused) and if not iterate over all locks held by that process and free them to break the deadlock, but that sort of overhead at something as fundamental as locking code can be annoying.

Also note that your semaphores then have to usually be implemented in assembly or you have to have a library that implements atomic increments of reference counts. Qt's QAtomicInt will do the trick if abused a little (just because it has an assembly implementations for multiple architectures). pthread_* doesn't have support for this as far as I know.

So, yeah, it's doable and some DBs do handle multi-process concurrent access, but it can be a pain in the ass. I believe SQLite does it by just locking complete tables on write access rather than using more fine-grained locks. MySQL embedded may also support multi-process access -- not sure.