BIND 10 trac2776, updated. abe72e9dff228269f889379f63228001e807982c [2776] Forward messages
BIND 10 source code commits
bind10-changes at lists.isc.org
Wed Feb 27 12:45:21 UTC 2013
The branch, trac2776 has been updated
via abe72e9dff228269f889379f63228001e807982c (commit)
from b859ed9695b1b4fb1a8861672a05e37065bc1593 (commit)
Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.
- Log -----------------------------------------------------------------
commit abe72e9dff228269f889379f63228001e807982c
Author: Michal 'vorner' Vaner <michal.vaner at nic.cz>
Date: Wed Feb 27 13:43:10 2013 +0100
[2776] Forward messages
Basics of the receptionists. Forward the messages to auth. It doesn't
seem to work properly, though. No idea if the error is on the auth or
receptionist side.
-----------------------------------------------------------------------
Summary of changes:
configure.ac | 1 +
src/bin/Makefile.am | 2 +-
src/bin/receptionist/Makefile.am | 9 +++
src/bin/receptionist/main.cc | 144 ++++++++++++++++++++++++++++++++++++++
4 files changed, 155 insertions(+), 1 deletion(-)
create mode 100644 src/bin/receptionist/Makefile.am
create mode 100644 src/bin/receptionist/main.cc
-----------------------------------------------------------------------
diff --git a/configure.ac b/configure.ac
index 5195c24..cebf105 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1180,6 +1180,7 @@ AC_CONFIG_FILES([Makefile
src/bin/auth/tests/Makefile
src/bin/auth/tests/testdata/Makefile
src/bin/auth/benchmarks/Makefile
+ src/bin/receptionist/Makefile
src/bin/ddns/Makefile
src/bin/ddns/tests/Makefile
src/bin/dhcp6/Makefile
diff --git a/src/bin/Makefile.am b/src/bin/Makefile.am
index 0b4c1ae..869903c 100644
--- a/src/bin/Makefile.am
+++ b/src/bin/Makefile.am
@@ -1,5 +1,5 @@
SUBDIRS = bind10 bindctl cfgmgr ddns loadzone msgq cmdctl auth xfrin \
xfrout usermgr zonemgr stats tests resolver sockcreator dhcp4 dhcp6 \
- dbutil sysinfo
+ dbutil sysinfo receptionist
check-recursive: all-recursive
diff --git a/src/bin/receptionist/Makefile.am b/src/bin/receptionist/Makefile.am
new file mode 100644
index 0000000..7c86a63
--- /dev/null
+++ b/src/bin/receptionist/Makefile.am
@@ -0,0 +1,9 @@
+if USE_STATIC_LINK
+AM_LDFLAGS = -static
+endif
+
+pkglibexecdir = $(libexecdir)/@PACKAGE@
+
+pkglibexec_PROGRAMS = b10-receptionist
+b10_receptionist_SOURCES = main.cc
+
diff --git a/src/bin/receptionist/main.cc b/src/bin/receptionist/main.cc
new file mode 100644
index 0000000..e9819e3
--- /dev/null
+++ b/src/bin/receptionist/main.cc
@@ -0,0 +1,144 @@
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <sys/epoll.h>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+
+// This is experimental code. There's only minimal error handling,
+// hardcoded values, no tests, etc. In short, it doesn't work :-P
+// We also avoid use of asio, etc. It doesn't allow us to use
+// low-level fancy functions like sendmmsg. Also, writing the code
+// fast is easier with the low-level (though unsafe, less portable
+// and harder to maintain) interface.
+//
+// Ah, also, many calls here are modern-linux specific.
+
+using namespace std;
+
+namespace {
+
+// Configuration constants
+const uint16_t listen_port = 5310;
+const uint16_t up_port = 5311;
+const size_t conn_count = 4;
+const size_t event_cnt = 10;
+const size_t buffer_size = 6553600;
+const size_t msg_size = 65536;
+const size_t maxmsg_count = 50;
+
+struct epoll_handler {
+ void (*handler)(size_t param);
+ size_t param;
+};
+
+void udp_ready(size_t);
+
+int udp_socket;
+struct epoll_handler udp_handler = { udp_ready, 0 };
+
+struct upconn {
+ int sock;
+ struct epoll_handler handler;
+ uint8_t buff[buffer_size];
+ size_t buff_size;
+};
+struct upconn conns[conn_count];
+
+int epoll_socket;
+
+void
+check(int ecode, const char *what, size_t line) {
+ if (ecode == -1) {
+ fprintf(stderr, "%s on line %zu failed: %m\n", what, line);
+ abort();
+ }
+}
+
+#define CHECK(WHAT) check((WHAT), #WHAT, __LINE__)
+
+uint8_t udp_buffers[maxmsg_count][msg_size];
+struct mmsghdr udp_headers[maxmsg_count];
+struct iovec udp_vectors[maxmsg_count];
+
+void udp_ready(size_t) {
+ int result;
+ CHECK(result = recvmmsg(udp_socket, udp_headers, maxmsg_count, MSG_DONTWAIT, NULL));
+ for (size_t i = 0; i < result; i ++) {
+ size_t upstream = random() % conn_count;
+ uint16_t len = htons(udp_headers[i].msg_len);
+ memcpy(conns[upstream].buff + conns[upstream].buff_size, &len, 2);
+ conns[upstream].buff_size += 2;
+ memcpy(conns[upstream].buff + conns[upstream].buff_size, udp_buffers[i], udp_headers[i].msg_len);
+ conns[upstream].buff_size += udp_headers[i].msg_len;
+ }
+}
+
+void upstream_ready(size_t) {
+ abort(); // Not yet
+}
+
+}
+
+
+int main() {
+ // Initialize buffers
+ for (size_t i = 0; i < maxmsg_count; i ++) {
+ udp_headers[i].msg_hdr.msg_iovlen = 1;
+ udp_headers[i].msg_hdr.msg_iov = &udp_vectors[i];
+ udp_vectors[i].iov_base = udp_buffers[i];
+ udp_vectors[i].iov_len = msg_size;
+ }
+ // Initialize epoll
+ CHECK(epoll_socket = epoll_create(10));
+ // Initialize the listening UDP socket
+ CHECK(udp_socket = socket(AF_INET6, SOCK_DGRAM, 0));
+ struct sockaddr_in6 addr;
+ memset(&addr, 0, sizeof addr);
+ addr.sin6_family = AF_INET6;
+ addr.sin6_port = htons(listen_port);
+ addr.sin6_addr = in6addr_any;
+ CHECK(bind(udp_socket, (const struct sockaddr *) &addr, sizeof addr));
+ struct epoll_event event;
+ event.events = EPOLLIN;
+ event.data.ptr = &udp_handler;
+ CHECK(epoll_ctl(epoll_socket, EPOLL_CTL_ADD, udp_socket, &event));
+ // Initialize the upstream connections to the auth servers.
+ addr.sin6_port = htons(up_port);
+ addr.sin6_addr = in6addr_loopback;
+ for (size_t i = 0; i < conn_count; i ++) {
+ CHECK(conns[i].sock = socket(AF_INET6, SOCK_STREAM, 0));
+ CHECK(connect(conns[i].sock, (const struct sockaddr *) &addr, sizeof addr));
+ conns[i].handler.handler = upstream_ready;
+ conns[i].handler.param = i;
+ event.data.ptr = &conns[i].handler;
+ CHECK(epoll_ctl(epoll_socket, EPOLL_CTL_ADD, conns[i].sock, &event));
+ conns[i].buff_size = 4; // After the length header // After the length header
+ }
+ // Initialization complete. Handle events forever.
+ for (;;) {
+ struct epoll_event events[event_cnt];
+ int epoll_result;
+ CHECK(epoll_result = epoll_wait(epoll_socket, events, event_cnt, -1));
+ // Handle the events we got now, reading everything (up to some maximal size)
+ for (size_t i = 0; i < epoll_result; i ++) {
+ struct epoll_handler *handler = (struct epoll_handler *) events[i].data.ptr;
+ handler->handler(handler->param);
+ }
+ // Send all read data
+ for (size_t i = 0; i < conn_count; i ++) {
+ if (conns[i].buff_size == 4)
+ continue;
+ uint32_t size = htonl(conns[i].buff_size - 4);
+ memcpy(conns[i].buff, &size, 4);
+ size_t pos = 0;
+ ssize_t result;
+ while (pos < conns[i].buff_size) {
+ CHECK(result = send(conns[i].sock, conns[i].buff + pos, conns[i].buff_size - pos, 0));
+ pos += result;
+ }
+ conns[i].buff_size = 4;
+ }
+ }
+}
More information about the bind10-changes
mailing list