I wrote:
> 3. In release 1.4.3, src/channel.c, _libssh2_channel_write(), line 2017:
>
> return (rc==LIBSSH2_ERROR_EAGAIN?rc:0);
>
>
> Here rc is the result from _libssh2_transport_read(session).
>
> When the output buffer is full, but the input buffer wasn't, this
> returns 0. It seems to me (and please correct me if I am mistaken)
> that libssh2_channel_write_ex() should only return 0 if the buflen
> argument is itself zero. I think that the correct line here would
> be simply
>
> return LIBSSH2_ERROR_EAGAIN;
>
> This fixes an infinite loop in typical fwrite()-like code that
> assumes that (the equivalent of) write() will prefer reporting
> EAGAIN over returning zero.
On a related note... if the output socket gets shut down or otherwise
blocked when the send window is exactly full, it's not clear to me how
libssh2_channel_write_ex() will ever return anything other than 0 or
LIBSSH2_ERROR_EAGAIN. Presumably calling send() would return an error,
allowing a flag (local.eof?) to be set, but how can it ever get called?
Shouldn't _ssh_transport_read() call send_existing()? Would that break
the logjam?
Nathan Myers
ncm_at_cantrip.org
_______________________________________________
libssh2-devel http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-devel
Received on 2012-12-05