char: fix handling of QIO_CHANNEL_ERR_BLOCK

If io_channel_send_full gets QIO_CHANNEL_ERR_BLOCK it
and has already sent some of the data, it should return
that amount of data, not EAGAIN, as that would cause
the caller to re-try already sent data.

Unfortunately due to a previous rebase conflict resolution
error, the code for dealing with this was in the wrong
part of the conditional, and so mistakenly ran on other
I/O errors.

This be seen running

   qemu-system-x86_64 -monitor stdio

and entering 'info mtree', when running on a slow console
(eg a slow remote ssh session). The monitor would get into
an indefinite loop writing the same data until it managed
to send it all without getting EAGAIN.
Reported-by: NIgor Mammedov <imammedo@redhat.com>
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>
Message-Id: <1455288410-27046-1-git-send-email-berrange@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

qemu-char.c

@@ -896,13 +896,13 @@ static int io_channel_send_full(QIOChannel *ioc,
             ioc, &iov, 1,
             fds, nfds, NULL);
         if (ret == QIO_CHANNEL_ERR_BLOCK) {
-            errno = EAGAIN;
-            return -1;
-        } else if (ret < 0) {
             if (offset) {
                 return offset;
             }
 
+            errno = EAGAIN;
+            return -1;
+        } else if (ret < 0) {
             errno = EINVAL;
             return -1;
         }