[DERBY-1392] Corner case behaviour in RAFContainer#writePage() can cause invalid data to be written to data files

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 10.1.2.1, 10.2.1.6
• Fix Version/s: 10.1.3.1, 10.2.1.6
• Component/s: Store
• Environment: Platforms that throw an IOException when writing beyond the EOF, but permit the write to proceed if the file is padded, then written.
• Issue & fix info: Patch Available
• Assignee: Anders Morken
• Reporter: Anders Morken
• Created: 10/Jun/06 01:43
• Updated: 12/Jul/06 01:30
• Resolved: 16/Jun/06 22:58

Description

java/engine/org/apache/derby/impl/store/raw/data/RAFContainer.java#writePage(...) will
attempt to retry a write if an IOException is thrown on the first attempt. However, the next
attempt does not add container header data to the first page, nor does it encrypt the data
if the database is encrypted.

I'd expect this bug to be case silent corruption of encrypted databases if the code path
was actually exercised. The fact that this bug still lives and nobody has discovered it is
possibly an indication of how uncommon this code path is. Since the comment in the code
says nothing about exactly what platforms the workaround was intended for, I don't know if
these platforms are still supported for Derby. There's also a workaround for an EPOC Java
bug earlier in the code - EPOC only had a Java 1.1 VM, which is no longer supported.

I'll attach a patch for the issue, but I wonder if we might as well remove the "retry code path"
if it is never used?

(If you're paranoid, this could also be considered a security issue. If someone could
cause IO errors for Derby at will, they could make Derby write the database without
encryption - but there are far easier ways to attack Derby if you've got that kind of
access, so I'm discounting that. =)