[ https://issues.apache.org/jira/browse/ARTEMIS-4340?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Robbie Gemmell updated ARTEMIS-4340: ------------------------------------ Fix Version/s: 2.30.0 Affects Version/s: 2.29.0 Description: Report was made (see below) about ByteUtil.uncheckedZeros throwing IndexOutOfBoundsException. Though some of ByteUtil's behaviour can perhaps be questioned, the problem ultimately starts earlier in the call tree with ThreadLocalByteBufferPool.borrow(int, boolean) from artemis-journal passing it invalid arguments. It passed in a buffer for zeroing, with its limit still set below the requested size to be zero'd, leading to the IndexOutOfBoundsException when trying to write beyond the limit. The reason this seemingly long-standing bug has perhaps not been hit so far is that it occurs in a less typical fallback case, where the buffer either isnt direct _and_ able to be updated with Netty's Unsafe bits, or also doesn't have an array. ThreadLocalByteBufferPool should clear position+limit before zeroing the buffer. Original Description: ================= In artemis-commons: ByteUtil.uncheckedZeros throws IndexOutOfBoundsException in the else part of the method. The code in the else part is looping through number of bytes specified by bytes input parameter which is often provided by buffer.capacity(), but then the code at buffer.put(i + offset, zero) where buffer is of DirectByteBuffer type calls checkIndex(int i) and it throws if i is >= limit. But limit is not the same as capacity in a Buffer. In our case capacity is 4096 and limit is 16 so method always throws IndexOutOfBoundsException if the else case is executed as soon as i ==16. This code is called when Artemis is launched and it validates/creates the various bindings files. This causes Artemis to fail to load properly. Note I believe all versions of artemis-commons has this bug but I currently am testing with 2.22.0. was: In artemis-commons: ByteUtil.uncheckedZeros throws IndexOutOfBoundsException in the else part of the method. The code in the else part is looping through number of bytes specified by bytes input parameter which is often provided by buffer.capacity(), but then the code at buffer.put(i + offset, zero) where buffer is of DirectByteBuffer type calls checkIndex(int i) and it throws if i is >= limit. But limit is not the same as capacity in a Buffer. In our case capacity is 4096 and limit is 16 so method always throws IndexOutOfBoundsException if the else case is executed as soon as i ==16. This code is called when Artemis is launched and it validates/creates the various bindings files. This causes Artemis to fail to load properly. Note I believe all versions of artemis-commons has this bug but I currently am testing with 2.22.0. Summary: fix artemis-journal ThreadLocalByteBufferPool.borrow zeroing (was: ByteUtil.uncheckedZeros throws IndexOutOfBoundsException in else condition) > fix artemis-journal ThreadLocalByteBufferPool.borrow zeroing > ------------------------------------------------------------ > > Key: ARTEMIS-4340 > URL: https://issues.apache.org/jira/browse/ARTEMIS-4340 > Project: ActiveMQ Artemis > Issue Type: Bug > Affects Versions: 2.22.0, 2.29.0 > Environment: Windows 10, Linux > Reporter: David Hoffer > Assignee: Robbie Gemmell > Priority: Major > Fix For: 2.30.0 > > Time Spent: 1h 20m > Remaining Estimate: 0h > > Report was made (see below) about ByteUtil.uncheckedZeros throwing > IndexOutOfBoundsException. Though some of ByteUtil's behaviour can perhaps be > questioned, the problem ultimately starts earlier in the call tree with > ThreadLocalByteBufferPool.borrow(int, boolean) from artemis-journal passing > it invalid arguments. It passed in a buffer for zeroing, with its limit still > set below the requested size to be zero'd, leading to the > IndexOutOfBoundsException when trying to write beyond the limit. > The reason this seemingly long-standing bug has perhaps not been hit so far > is that it occurs in a less typical fallback case, where the buffer either > isnt direct _and_ able to be updated with Netty's Unsafe bits, or also > doesn't have an array. > ThreadLocalByteBufferPool should clear position+limit before zeroing the > buffer. > > Original Description: > ================= > In artemis-commons: > ByteUtil.uncheckedZeros throws IndexOutOfBoundsException in the else part of > the method. > The code in the else part is looping through number of bytes specified by > bytes input parameter which is often provided by buffer.capacity(), but then > the code at > buffer.put(i + offset, zero) where buffer is of DirectByteBuffer type calls > checkIndex(int i) and it throws if i is >= limit. > But limit is not the same as capacity in a Buffer. In our case capacity is > 4096 and limit is 16 so method always throws IndexOutOfBoundsException if the > else case is executed as soon as i ==16. > This code is called when Artemis is launched and it validates/creates the > various bindings files. This causes Artemis to fail to load properly. > Note I believe all versions of artemis-commons has this bug but I currently > am testing with 2.22.0. -- This message was sent by Atlassian Jira (v8.20.10#820010)