On 9/15/19 9:55 AM, Richard W.M. Jones wrote:
Works essentially just like calling getpeername(2), because that's how it is implemented. --- TODO | 6 ++++++ docs/nbdkit-plugin.pod | 23 +++++++++++++++++++++++ include/nbdkit-common.h | 2 ++ server/nbdkit.syms | 1 + server/public.c | 21 +++++++++++++++++++++ 5 files changed, 53 insertions(+)
diff --git a/TODO b/TODO index 04def3c..d2cf0ae 100644 --- a/TODO +++ b/TODO @@ -77,6 +77,12 @@ General ideas for improvements name(s) that a plugin might want to support. Probably we should deprecate the -e option entirely since it does nothing useful.
+* Add plugin "connect" method. This would be called on a connection + before handshaking or TLS negotiation, and could be used (with + nbdkit_peer_name) to accept or reject connections based on IP + address, rather like a poor man's TCP wrappers. See also commit + c05686f9577f.
Yes, you now have more justification for why a .connect would be a useful callback (and we would document that the plugin is responsible for NOT sticking a lot of code into .connect, so that it does not become an amplification attack).
+int +nbdkit_peer_name (struct sockaddr *addr, socklen_t *addrlen) +{ + struct connection *conn = threadlocal_get_conn (); + int s; + + if (!conn) { + nbdkit_error ("no connection in this thread"); + return -1; + } + + s = conn->sockin; + if (s == -1) { + nbdkit_error ("socket not open"); + return -1; + } + + return getpeername (s, addr, addrlen);
You need to call nbdkit_error() if getpeername() returns -1 Otherwise, looks reasonable. -- Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org