Dmytro Shteflyuk created THRIFT-6099:
----------------------------------------

             Summary: Ruby MemoryBufferTransport should reject invalid read 
lengths
                 Key: THRIFT-6099
                 URL: https://issues.apache.org/jira/browse/THRIFT-6099
             Project: Thrift
          Issue Type: Bug
            Reporter: Dmytro Shteflyuk
            Assignee: Dmytro Shteflyuk


h3. Problem

{{Thrift::MemoryBufferTransport#read}} does not validate a negative length 
before passing it to Ruby string APIs in the native extension. When 
{{thrift_native}} is loaded, calling {{read(-1)}} can terminate the Ruby 
process instead of reporting an invalid request.

The pure-Ruby implementation also lacks the expected negative-size check. It 
raises an unrelated exception after changing the transport's read position, so 
the native and pure-Ruby implementations do not provide the same contract.

The native implementation additionally advances its internal index using signed 
integer addition before clamping it to the buffer size. A maximum-width read 
after the index has advanced can overflow that calculation.

h3. Client impact

This affects Ruby clients using {{MemoryBufferTransport}}. A negative length 
may terminate a process using the native extension, while the pure-Ruby 
implementation may raise the wrong exception and leave the transport in an 
unexpected state.

Normal reads with valid lengths are unaffected.

h3. Reproduction

With the native extension loaded:

{code:ruby}
require "thrift"

transport = Thrift::MemoryBufferTransport.new("abc")
transport.read(-1)
{code}

The Ruby process may terminate instead of raising a Thrift transport exception.

The boundary calculation can also be exercised with:

{code:ruby}
transport = Thrift::MemoryBufferTransport.new("abcd")
transport.read(1)
transport.read((2**31) - 1)
{code}

h3. Expected behavior

A negative length should raise {{Thrift::TransportException}} with type 
{{NEGATIVE_SIZE}} and should not consume any input.

Zero-length, exact-length, oversized, and maximum-width reads should update the 
transport position consistently without overflowing the native index 
calculation. Reads beyond the available data should continue to raise 
{{EOFError}}.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to