[
https://issues.apache.org/jira/browse/HDDS-15480?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated HDDS-15480:
----------------------------------
Labels: pull-request-available (was: )
> Potential for NPE / infinite loop in StreamBlockReader
> ------------------------------------------------------
>
> Key: HDDS-15480
> URL: https://issues.apache.org/jira/browse/HDDS-15480
> Project: Apache Ozone
> Issue Type: Sub-task
> Reporter: Stephen O'Donnell
> Assignee: Stephen O'Donnell
> Priority: Major
> Labels: pull-request-available
>
> There is potential for an NPE at stream shutdown in the StreamBlockReader:
> poll() explicitly returns null when future.isDone() is true (stream
> completed). However readFromQueue() uses the return value without a null
> check:
> {code}
> // StreamingReader.readFromQueue(), line 455-459
> ByteBuffer readFromQueue() throws IOException {
> final ReadBlockResponseProto readBlock = poll(); // can return null
> final ByteString data = readBlock.getData(); // NPE!
> {code}
> This fires in the normal shutdown path: after the server sends the last chunk
> and closes the stream, the next poll() call returns null and this line throws
> NullPointerException instead of letting the caller handle EOF gracefully.
> Fixing this, can then cause an infinite loop in StreamingReader.read() (lines
> 446-451)
>
> The inner read loop has no exit when readFromQueue() returns null or an empty
> buffer:
> {code}
> while (true) {
> final ByteBuffer buf = readFromQueue();
> if (buf != null && buf.hasRemaining()) {
> return buf; // only exit
> }
> // else: loop forever
> }
> {code}
> This could happen when:
> - Stream ends: once Bug 1 is fixed so readFromQueue() can return null, this
> loop spins forever on null.
> - Full offset skip: readFromQueue() adjusts the buffer position by pos -
> blockOffset. If the entire response arrived before the current position (e.g.
> after a seek, or when reading near the end of a small block), the adjusted
> buffer has 0 remaining bytes. The loop spins indefinitely without requesting
> new data, because readBlock() is only called once before the loop starts.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
