[Boost-users] [asio] How socket::shutdown() safe is within multithreaded usage?

Subject: [Boost-users] [asio] How socket::shutdown() safe is within multithreaded usage?
From: Slav (slavmfm_at_[hidden])
Date: 2011-05-14 08:47:32

Hello, my server runs single io_service and asynchronosly accepts incoming
connections (creating new sockets for each incoming connection).
Usage of connection happens in such ways:
     - calls to socket::async_write() from multiple threads (embraced to
mutex) at any time
     - handling socket::acync_read() (which runs within io_service' thread
as I understand the documentation)
     - single call to socket::shutdown() at any moment from some other
thread
Previusly I used multiple of mutexes to cleanly wait for all current
asynchronous read/write handlers: write function incremented "writes"
variable which then was used on socket shutdown to wait the same amount of
write handlers - but deadlock sometimes happend at moment of waiting write
handlers. So I referred to socket::shutdown()'s documentation which say "Any
asynchronous send, receive or connect operations will be cancelled
immediately" - and don't wait any asynchronous operations for now and
everything seems to work fine for now (what scares me).

Is it safe to do so? How must I safely handle all currenlty running
asynchronous read/write handlers?
In case of boost.asio waits for all asynchronous operations to stop inside
socket::shutdown/close so, then, if to call it within one of such operations
- deadlock will (theoretically must) happen - so, socket::close/shutdown
isn't safe?

Thank you.