Markus posted a bug report about a bad 0-return from libssh2_channel_read:

http://libssh2.haxx.se/mail/libssh2-devel-archive-2009-04/0076.shtml

  And it was indeed a bad loop that terminated too early due to a receveived
  close packet.

NEWS

@@ -1,4 +1,12 @@
 
+* (Apr 30 2009) Daniel Stenberg:
+
+  Markus posted a bug report about a bad 0-return from libssh2_channel_read:
+  http://libssh2.haxx.se/mail/libssh2-devel-archive-2009-04/0076.shtml
+
+  And it was indeed a bad loop that terminated too early due to a receveived
+  close packet.
+
 * (Apr 14 2009) Daniel Stenberg:
 
   libssh2_poll() and libssh2_poll_channel_read() are now considered and

src/channel.c

@@ -1794,8 +1794,15 @@ static ssize_t channel_read(LIBSSH2_CHANNEL *channel, int stream_id,
 
     channel->read_packet = session->packets.head;
     while (channel->read_packet &&
-           !channel->remote.close &&
            (bytes_read < (int) buflen)) {
+        /* previously this loop condition also checked for
+           !channel->remote.close but we cannot let it do this:
+
+           We may have a series of packets to read that are still pending even
+           if a close has been received. Acknowledging the close too early
+           makes us flush buffers prematurely and loose data.
+        */
+
         LIBSSH2_PACKET *readpkt = channel->read_packet;
 
         /* In case packet gets destroyed during this iteration */