Fix incorrect order of lock file removal and failure to close() sockets.

Commit c9b0cbe98bd783e24a8c4d8d8ac472a494b81292 accidentally broke the
order of operations during postmaster shutdown: it resulted in removing
the per-socket lockfiles after, not before, postmaster.pid.  This creates
a race-condition hazard for a new postmaster that's started immediately
after observing that postmaster.pid has disappeared; if it sees the
socket lockfile still present, it will quite properly refuse to start.
This error appears to be the explanation for at least some of the
intermittent buildfarm failures we've seen in the pg_upgrade test.

Another problem, which has been there all along, is that the postmaster
has never bothered to close() its listen sockets, but has just allowed them
to close at process death.  This creates a different race condition for an
incoming postmaster: it might be unable to bind to the desired listen
address because the old postmaster is still incumbent.  This might explain
some odd failures we've seen in the past, too.  (Note: this is not related
to the fact that individual backends don't close their client communication
sockets.  That behavior is intentional and is not changed by this patch.)

Fix by adding an on_proc_exit function that closes the postmaster's ports
explicitly, and (in 9.3 and up) reshuffling the responsibility for where
to unlink the Unix socket files.  Lock file unlinking can stay where it
is, but teach it to unlink the lock files in reverse order of creation.

diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c
index 8c4a543bf4b7c652af3cbaf8e11b30e70f012f71..08188028198cd2c3387090061ac1cab27a0a330c 100644 (file)
--- a/src/backend/postmaster/postmaster.c
+++ b/src/backend/postmaster/postmaster.c
@@ -335,6 +335,7 @@ static DNSServiceRef bonjour_sdref = NULL;
 /*
  * postmaster.c - function prototypes
  */
+static void CloseServerPorts(int status, Datum arg);
 static void getInstallationPaths(const char *argv0);
 static void checkDataDir(void);
 static void pmdaemonize(void);
@@ -977,6 +978,14 @@ PostmasterMain(int argc, char *argv[])
                ereport(FATAL,
                                (errmsg("no socket created for listening")));
 
+       /*
+        * Set up an on_proc_exit function that's charged with closing the sockets
+        * again at postmaster shutdown.  You might think we should have done this
+        * earlier, but we want it to run before not after the proc_exit callback
+        * that will remove the Unix socket file.
+        */
+       on_proc_exit(CloseServerPorts, 0);
+
        /*
         * Set up shared memory and semaphores.
         */
@@ -1137,6 +1146,36 @@ PostmasterMain(int argc, char *argv[])
 }
 
 
+/*
+ * on_proc_exit callback to close server's listen sockets
+ */
+static void
+CloseServerPorts(int status, Datum arg)
+{
+       int                     i;
+
+       /*
+        * First, explicitly close all the socket FDs.  We used to just let this
+        * happen implicitly at postmaster exit, but it's better to close them
+        * before we remove the postmaster.pid lockfile; otherwise there's a race
+        * condition if a new postmaster wants to re-use the TCP port number.
+        */
+       for (i = 0; i < MAXLISTEN; i++)
+       {
+               if (ListenSocket[i] != PGINVALID_SOCKET)
+               {
+                       StreamClose(ListenSocket[i]);
+                       ListenSocket[i] = PGINVALID_SOCKET;
+               }
+       }
+
+       /*
+        * Removal of the Unix socket file and socket lockfile will happen in
+        * later on_proc_exit callbacks.
+        */
+}
+
+
 /*
  * Compute and check the directory paths to files that are part of the
  * installation (as deduced from the postgres executable's own location)