Reading is pull-based and works like this: %php_stream_read()% or %fread()% requests ~n~ bytes to be placed into a buffer. The topmost filter on the read stack returns the data from its own buffered brigade, if any, splitting buckets if required. If it doesn't have enough data, it asks the next filter in the stack for some more data. This process repeats until the actual underlying stream is hit. The stream reads a chunk of data and returns it to the filter. The data moves along the brigades, perhaps being transformed along the way. Topmost filter gets the data and stores it in the buffer. Process repeats until the buffer is filled or there is no more data available from the stream.