Win32: Fix some bugs in the restart logic. This is still rather adhoc. A more rigorous solution

is to maintain a list of overlapped io structures in the child_main control thread
and use HasOverLappedIoCompleted() to manage the connections and pending io.

We must resort to maintaining a list of pending io (i.e. overlapped structures) should
we ever begin using AcceptEx to receive data. Why? We need to be able to detect when
a connection is received but no data is being sent over that connection. There is a
Win32 function (whose name I forget at the moment) to handle this. Why can't Win32
just timeout the connection using setopt??? It would make life so much easier if AcceptEx
just returned TIMEOUT in this condition rather than requiring the application to
explicitly poll the socket to see if data is available.


git-svn-id: https://svn.apache.org/repos/asf/httpd/httpd/trunk@85131 13f79535-47bb-0310-9956-ffa450edef68

1 files changed, 47 insertions, 26 deletions

diff --git a/server/mpm/winnt/mpm_winnt.c b/server/mpm/winnt/mpm_winnt.c
index 71017ae7e1..860adaa309 100644
--- a/server/mpm/winnt/mpm_winnt.c
+++ b/server/mpm/winnt/mpm_winnt.c
@@ -935,7 +935,6 @@ static PCOMP_CONTEXT winnt_get_connection(PCOMP_CONTEXT context)
         /* If child shutdown has been signaled, clean-up the completion context */
         if (workers_may_exit) {
             CloseHandle(context->Overlapped.hEvent);
-            /* destroy pool */
         }
         else {
             context->accept_socket = INVALID_SOCKET; /* Don't reuse the accept_socket */
@@ -954,13 +953,14 @@ static PCOMP_CONTEXT winnt_get_connection(PCOMP_CONTEXT context)
                                        &CompKey,
                                        &pol,
                                        INFINITE);
+
         if (!rc) {
-            ap_log_error(APLOG_MARK,APLOG_ERR, GetLastError(), server_conf,
-                         "Child %d: - GetQueuedCompletionStatus() failed", my_pid);
             /* During a restart, the new child process can catch 
              * ERROR_OPERATION_ABORTED completion packets
              * posted by the old child process. Just continue...
              */
+            ap_log_error(APLOG_MARK,APLOG_ERR, GetLastError(), server_conf,
+                         "Child %d: - GetQueuedCompletionStatus() failed", my_pid);
             continue;
         }
 
@@ -973,8 +973,9 @@ static PCOMP_CONTEXT winnt_get_connection(PCOMP_CONTEXT context)
         if (CompKey == my_pid) {
             return NULL;
         }
-        if (CompKey != 0)
+        if (CompKey != 0) {
             continue;
+        }
 
         context = (PCOMP_CONTEXT) pol;
         break;
@@ -1212,20 +1213,13 @@ static void child_main()
             /* Something serious is wrong */
             workers_may_exit = 1;
             ap_log_error(APLOG_MARK, APLOG_CRIT, GetLastError(), server_conf,
-                         "Child %d: WaitForMultipeObjects WAIT_FAILED -- doing server shutdown");
-            /* Give a busy server the chance to drain AcceptEx completion contexts
-             * by servicing connections. Note that the setting of workers_may_exit
-             * prevents new AcceptEx completion contexts from being created.
-             */
-            Sleep(1000);
-
-            /* Drain any remaining contexts. May loose a few connections here. */
-            drain_acceptex_complport(AcceptExCompPort, FALSE);
+                         "Child %d: WAIT_FAILED -- shutting down server");
         }
         else if (rv == WAIT_TIMEOUT) {
             /* Hey, this cannot happen */
-            ap_log_error(APLOG_MARK, APLOG_INFO, APR_SUCCESS, server_conf,
-                         "Child %d: Server maintenance...", my_pid);
+            workers_may_exit = 1;
+            ap_log_error(APLOG_MARK, APLOG_CRIT, APR_SUCCESS, server_conf,
+                         "Child %d: WAIT_TIMEOUT -- shutting down server", my_pid);
         }
         else if (cld == 0) {
             /* Exit event was signaled */
@@ -1240,29 +1234,56 @@ static void child_main()
             }
             ResetEvent(maintenance_event);
             ap_log_error(APLOG_MARK, APLOG_INFO, APR_SUCCESS, server_conf,
-                         "Child %d: Child maintenance event signaled...", my_pid);
+                         "Child %d: Child maintenance event signaled.", my_pid);
         }
     }
 
     /* Shutdown the worker threads */
+
     if (osver.dwPlatformId == VER_PLATFORM_WIN32_WINDOWS) {
         for (i = 0; i < nthreads; i++) {
             add_job(-1);
         }
     }
     else { /* Windows NT/2000 */
-        /* Sometimes posted completion contexts intended for this process
-         * are consumed by the new process (during restart). Send a few 
-         * extra to compensate. At worst, this process may hang around
-         * for several minutes because a thread is not exiting because
-         * it is blocked on GetQueuedCompletionStatus.
+        SOCKET nsd;
+        ap_listen_rec *lr;
+        /*
+         * First thing to do is to drain all the completion contexts off the
+         * AcceptEx iocp. Give a busy server the chance to drain 
+         * the port by servicing connections (workers_may_exit prevents new 
+         * AcceptEx completion contexts from being queued to the port).
          */
-        for (i=0; i < 2*nthreads; i++) {
-            if (!PostQueuedCompletionStatus(AcceptExCompPort, 0, my_pid, NULL)) {
-                ap_log_error(APLOG_MARK,APLOG_INFO, APR_SUCCESS, server_conf, 
-                             "PostQueuedCompletionStatus failed");
-            }
+        Sleep(1000);
+
+        /* Cancel any remaining pending async i/o.
+         * This will cause io completion packets to be queued to the
+         * port for any remaining active contexts
+         */
+        for (lr = ap_listeners; lr != NULL; lr = lr->next) {
+            ap_get_os_sock(&nsd,lr->sd);
+            CancelIo(nsd);
         }
+
+        /* Drain the canceled contexts off the port */
+        drain_acceptex_complport(AcceptExCompPort, FALSE);
+
+        /* Hopefully by now, all the completion contexts should be drained 
+         * off the port. There could still be some cancel io completion packets
+         * flying around in the kernel... We will cover this possibility later..
+         * 
+         * Consider using HasOverlappedIoCompleted()...
+         *
+         * Next task is to unblock all the threads blocked on 
+         * GetQueuedCompletionStatus()
+         *
+         */
+        for (i=0; i < nthreads*2; i++) {
+            PostQueuedCompletionStatus(AcceptExCompPort, 0, my_pid, NULL);
+        }
+
+        /* Give the worker threads time to realize they've been posted */
+        Sleep(1000);
     }
 
     /* Release the start_mutex to let the new process (in the restart