Fix: Remote Denial of Service (Infinite Loop) in CharsetDecoder

[OxBat]

Summary

I identified a Remote Denial of Service vulnerability in CharsetDecoder (specifically APRCharsetDecoder).
When processing untrusted network input, malformed byte sequences trigger an infinite loop.

The Loop:
When apr_xlate_conv_buffer encounters an invalid byte (like 0xFF in UTF-8), it returns an error status (APR_BADCH) but consumes 0 bytes.
The previous loop logic exited on error, but without advancing the buffer pointer. If the caller (e.g., SocketNode) retries decoding the remaining buffer, it hits the same byte again indefinitely (100% CPU).

The Fix:
I updated the decode loop to detect when an error occurs without consuming input. In this case, it now:

1. Checks if it's just APR_INCOMPLETE (valid partial packet).
2. If not, manually advances the buffer position by 1 byte (skipping the bad byte).
3. Appends a Replacement Character (?) to the output.
4. Resets status to APR_SUCCESS to continue decoding the rest of the stream.

[rm5248]

this seems like a pretty easy thing to add a test for to prove that it is working correctly, could you add a test to the charsetdecodertestcase?

[OxBat]

Done! I added testInfiniteLoop to reproduce the hang with a bad byte (0xFF).

I purposely requested ISO-8859-2 in the test. If I used standard UTF-8, log4cxx would use its internal optimized decoder (skipping APR) and wouldn't hit the bug I fixed.

[rm5248]

your test does not appear to work properly. If I revert the changes to charsetdecoder.cpp the test fails and does not enter an infinite loop. What platform are you testing on and what version of APR are you using?

[swebb2066]

The test is too flaky as apr_xlate_conv_buffer relies on the iconv whose behaviour varies as versions change. #625 adds robustness checks to a number of decoder implementation including APRCharsetDecoder