rpcbind: systemd socket activation (v2)

From: Tom Gundersen <[email protected]>

Making rpcbind sockect activated will greatly simplify its integration in systemd systems. In essence, other services may now assume that rpcbind is always available, even during very early boot. This means that we no longer need to worry about any ordering dependencies.

Original-patch-by Lennart Poettering <[email protected]>

Cc [email protected]

Acked-by Cristian Rodríguez<[email protected]>

Signed-off-by Tom Gundersen <[email protected]>

Signed-off-by Steve Dickson <[email protected]>

---
 Makefile.am   |  6 +++++
 configure.ac  | 12 +++++++++
 src/rpcbind.c | 81 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-----
 3 files changed, 93 insertions(+), 6 deletions(-)

Makefile.am +6 -0

index 8715082..c99566d 100644
@@ -41,6 +41,12 @@ rpcbind_SOURCES = \
 	src/warmstart.c
 rpcbind_LDADD = $(TIRPC_LIBS)
 
+if SYSTEMD
+AM_CPPFLAGS += $(SYSTEMD_CFLAGS) -DSYSTEMD
+
+rpcbind_LDADD += $(SYSTEMD_LIBS)
+endif
+
 rpcinfo_SOURCES =       src/rpcinfo.c
 rpcinfo_LDADD   =       $(TIRPC_LIBS)
 

configure.ac +12 -0

index 5a88cc7..967eb05 100644
@@ -36,6 +36,18 @@ AC_SUBST([nss_modules], [$with_nss_modules])
 
 PKG_CHECK_MODULES([TIRPC], [libtirpc])
 
+PKG_PROG_PKG_CONFIG
+AC_ARG_WITH([systemdsystemunitdir],
+  AS_HELP_STRING([--with-systemdsystemunitdir=DIR], [Directory for systemd service files]),
+  [], [with_systemdsystemunitdir=$($PKG_CONFIG --variable=systemdsystemunitdir systemd)])
+  if test "x$with_systemdsystemunitdir" != xno; then
+    AC_SUBST([systemdsystemunitdir], [$with_systemdsystemunitdir])
+     PKG_CHECK_MODULES([SYSTEMD], [libsystemd], [],
+	   [PKG_CHECK_MODULES([SYSTEMD], [libsystemd-daemon], [],
+	   AC_MSG_ERROR([libsystemd support requested but found]))])
+  fi
+AM_CONDITIONAL(SYSTEMD, [test -n "$with_systemdsystemunitdir" -a "x$with_systemdsystemunitdir" != xno ])
+
 AS_IF([test x$enable_libwrap = xyes], [
 	AC_CHECK_LIB([wrap], [hosts_access], ,
 		AC_MSG_ERROR([libwrap support requested but unable to find libwrap]))

src/rpcbind.c +75 -6

index e3462e3..f7c71ee 100644
@@ -56,6 +56,9 @@
 #include <netinet/in.h>
 #endif
 #include <arpa/inet.h>
+#ifdef SYSTEMD
+#include <systemd/sd-daemon.h>
+#endif
 #include <fcntl.h>
 #include <netdb.h>
 #include <stdio.h>

@@ -296,6 +299,7 @@ init_transport(struct netconfig *nconf)
 	u_int32_t host_addr[4];  /* IPv4 or IPv6 */
 	struct sockaddr_un sun;
 	mode_t oldmask;
+	int n;
         res = NULL;
 
 	if ((nconf->nc_semantics != NC_TPI_CLTS) &&

@@ -314,6 +318,76 @@ init_transport(struct netconfig *nconf)
 			fprintf(stderr, "[%d] - %s\n", i, *s);
 	}
 #endif
+	if (!__rpc_nconf2sockinfo(nconf, &si)) {
+		syslog(LOG_ERR, "cannot get information for %s",
+		    nconf->nc_netid);
+		return (1);
+	}
+
+#ifdef SYSTEMD
+	n = sd_listen_fds(0);
+	if (n < 0) {
+		syslog(LOG_ERR, "failed to acquire systemd sockets: %s", strerror(-n));
+		return 1;
+	}
+
+	/* Try to find if one of the systemd sockets we were given match
+	 * our netconfig structure. */
+
+	for (fd = SD_LISTEN_FDS_START; fd < SD_LISTEN_FDS_START + n; fd++) {
+		struct __rpc_sockinfo si_other;
+		union {
+			struct sockaddr sa;
+			struct sockaddr_un un;
+			struct sockaddr_in in4;
+			struct sockaddr_in6 in6;
+			struct sockaddr_storage storage;
+		} sa;
+		socklen_t addrlen = sizeof(sa);
+
+		if (!__rpc_fd2sockinfo(fd, &si_other)) {
+			syslog(LOG_ERR, "cannot get information for fd %i", fd);
+			return 1;
+		}
+
+		if (si.si_af != si_other.si_af ||
+                    si.si_socktype != si_other.si_socktype ||
+                    si.si_proto != si_other.si_proto)
+			continue;
+
+		if (getsockname(fd, &sa.sa, &addrlen) < 0) {
+			syslog(LOG_ERR, "failed to query socket name: %s",
+                               strerror(errno));
+			goto error;
+		}
+
+		/* Copy the address */
+		taddr.addr.maxlen = taddr.addr.len = addrlen;
+		taddr.addr.buf = malloc(addrlen);
+		if (taddr.addr.buf == NULL) {
+			syslog(LOG_ERR,
+                               "cannot allocate memory for %s address",
+                               nconf->nc_netid);
+			goto error;
+		}
+		memcpy(taddr.addr.buf, &sa, addrlen);
+
+		my_xprt = (SVCXPRT *)svc_tli_create(fd, nconf, &taddr,
+                          RPC_MAXDATASIZE, RPC_MAXDATASIZE);
+		if (my_xprt == (SVCXPRT *)NULL) {
+			syslog(LOG_ERR, "%s: could not create service",
+                               nconf->nc_netid);
+			goto error;
+		}
+	}
+
+	/*
+	 * If none of the systemd sockets matched, we set up the socket in
+	 * the normal way:
+	 */
+#endif
+	if (my_xprt != NULL)
+		goto got_socket;
 
 	/*
 	 * XXX - using RPC library internal functions. For NC_TPI_CLTS

@@ -327,12 +401,6 @@ init_transport(struct netconfig *nconf)
 		}
 	}
 
-	if (!__rpc_nconf2sockinfo(nconf, &si)) {
-		syslog(LOG_ERR, "cannot get information for %s",
-		    nconf->nc_netid);
-		return (1);
-	}
-
 	if ((strcmp(nconf->nc_netid, "local") == 0) ||
 	    (strcmp(nconf->nc_netid, "unix") == 0)) {
 		memset(&sun, 0, sizeof sun);

@@ -569,6 +637,7 @@ init_transport(struct netconfig *nconf)
 			goto error;
 		}
 	}
+got_socket:
 
 #ifdef PORTMAP
 	/*

This is based on a patch originally posted by Lennart Poettering:
<http://permalink.gmane.org/gmane.linux.nfs/33774>.

That patch was not merged due to the lack of a shared library and as systemd was seen to be too Fedora specific.

Systemd now provides a shared library, and it is (or very soon will be) the default init system on all the major Linux distributions.

This version of the patch has three changes from the original:

* It uses the shared library.
* It comes with unit files.
* It is rebased on top of master.

Please review the patch with "git show -b" or otherwise ignoring the whitespace changes, or it will be extremely difficult to read.

v5: incorporated comments on the PKG_CHECK_MODULES macro.

v4: reorganized the changes to make the diff easier to read
remove systemd scripts.

v3: rebase
fix typos
listen on /run/rpcbind.sock, rather than /var/run/rpcbind.sock (the latter is a symlink to the former, but this means the socket can be created before /var is mounted)
NB: this version has been compile-tested only as I no longer use rpcbind myself
v2: correctly enable systemd code at compile time
handle the case where not all the required sockets were supplied listen on udp/tcp port 111 in addition to /var/run/rpcbind.sock do not daemonize

Tom Gundersen (1):
rpcbind: add support for systemd socket activation

 Makefile.am   |  6 +++++
 configure.ac  | 12 +++++++++
 src/rpcbind.c | 81 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-----
 3 files changed, 93 insertions(+), 6 deletions(-)

Signature · 6 lines

1.9.3

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Quoted 2 lines · Steve Dickson

This is based on a patch originally posted by Lennart Poettering:
<http://permalink.gmane.org/gmane.linux.nfs/33774>.

Have you run this by the reporter
of https://bugzilla.redhat.com/show_bug.cgi?id=1158164 ?

Because he tried applying that old patch and found he was still having problems.

But they may well be problems that are fixed by your version, or he may have applied it incorrectly, I didn't try to figure it out.

--b.

Quoted 33 lines

That patch was not merged due to the lack of a shared library and as systemd was seen to be too Fedora specific.

Systemd now provides a shared library, and it is (or very soon will be) the default init system on all the major Linux distributions.

This version of the patch has three changes from the original:

* It uses the shared library.
* It comes with unit files.
* It is rebased on top of master.

Please review the patch with "git show -b" or otherwise ignoring the whitespace changes, or it will be extremely difficult to read.

v5: incorporated comments on the PKG_CHECK_MODULES macro.

v4: reorganized the changes to make the diff easier to read
remove systemd scripts.

v3: rebase
fix typos
listen on /run/rpcbind.sock, rather than /var/run/rpcbind.sock (the latter is a symlink to the former, but this means the socket can be created before /var is mounted)
NB: this version has been compile-tested only as I no longer use rpcbind myself
v2: correctly enable systemd code at compile time
handle the case where not all the required sockets were supplied listen on udp/tcp port 111 in addition to /var/run/rpcbind.sock do not daemonize

Quoted 15 lines

Tom Gundersen (1):
rpcbind: add support for systemd socket activation

 Makefile.am   |  6 +++++
 configure.ac  | 12 +++++++++
 src/rpcbind.c | 81 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-----
 3 files changed, 93 insertions(+), 6 deletions(-)

Signature · 6 lines

1.9.3

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html