Is "The Multiple SQLite Problem" An Issue for SQLCipher for Android?

SQLCipher
2

Hello @commonsguy,

SQLCipher doesn’t do anything specifically to mitigate this, so there is a theoretical risk. However, in practice, it is a lower risk with SQLCipher.

The possibility for data corruption really exists when you have two different SQLite libraries linked into a single application that are concurrently accessing the same database files. That is an important qualifier; the scenario that Erik describes in his post is a big problem for their Zumero library because it performs synchronization for arbitrary application data in SQLite databases. In that case, Zumero is presumably syncing and sharing access to the same database that the application is also managing. When a different library is used by the application than the Zumero sync engine, they risk corruption because they are acting on the same database files. This is the same scenario demonstrated in the sample application that causes a live corruption, i.e. two threads in the same process manipulating the same database using different instances of the library.

This is a much lower risk for applications using SQLCipher. Because the vast majority of SQLCipher deployments feature encrypted databases, it is not even possible to open or read them with a different version of SQLite. As a result, only SQLCipher can really operate on any given database, avoiding the issue completely.

There are a few edge cases where there could be a problem, but these should be rare:

  1. Using the SQLCipher library concurrently with another SQLite library to operate simultaneously on standard (plaintext) databases

  2. Cross-attaching standard databases to an encrypted session with SQLCipher while they are simultaneously being used by another SQLite library

  3. Using android.database.sqlite to open and close SQLCipher databases, even though its not able to read them

In summary, a separately linked SQLite library could be managing other databases, but as long as they don’t overlap the POSIX close bug and resulting corruptions should not occur.

Please let us know if this addresses your question, or if there were any other specific scenarios you’re considering that could cause problems.