Gil Tene created CASSANDRA-14284:
------------------------------------
Summary: Chunk checksum test needs to occur before uncompress to
avoid JVM crash
Key: CASSANDRA-14284
URL: https://issues.apache.org/jira/browse/CASSANDRA-14284
Project: Cassandra
Issue Type: Bug
Components: Core
Environment: The check-only-after-doing-the-decompress logic appears
to be in all current releases.
Here are some samples at different evolution points :
3.11.2:
https://github.com/apache/cassandra/blob/cassandra-3.5/src/java/org/apache/cassandra/io/compress/CompressedRandomAccessReader.java#L135
https://github.com/apache/cassandra/blob/cassandra-3.5/src/java/org/apache/cassandra/io/compress/CompressedRandomAccessReader.java#L198
3.5:
[https://github.com/apache/cassandra/blob/cassandra-3.5/src/java/org/apache/cassandra/io/compress/CompressedRandomAccessReader.java#L135]
https://github.com/apache/cassandra/blob/cassandra-3.5/src/java/org/apache/cassandra/io/compress/CompressedRandomAccessReader.java#L196
2.1.17:
[https://github.com/apache/cassandra/blob/cassandra-2.1.17/src/java/org/apache/cassandra/io/compress/CompressedRandomAccessReader.java#L122]
Reporter: Gil Tene
While checksums are (generally) performed on compressed data, the checksum test
when reading is currently (in all variants of C* 2.x, 3.x I've looked at) done
o[on the compressed data]nly after the uncompress operation has completed.
The issue here is that LZ4_decompress_fast (as documented in e.g.
[https://github.com/lz4/lz4/blob/dev/lib/lz4.h#L214)] can result in memory
overruns when provided with malformed source data. This in turn can (and does,
e.g. in CASSANDRA-13757) lead to JVM crashes during the uncompress of corrupted
chunks. The checksum operation would obviously detect the issue, but we'd never
get to it if the JVM crashes first.
Moving the checksum test of the compressed data to before the uncompress
operation (in cases where the checksum is done on compressed data) will resolve
this issue.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
