scd: Fix problem with reader list becoming empty.

* scd/apdu.c (close_pcsc_reader): Do not decrement refcount if already
zero.  Always release context if or becomes zero.
(apdu_dev_list_start): Unlock prior to close_pcsc_reader.  For PC/SC
increment the count.  Always release the lock.
(apdu_dev_list_finish): No more unlocking.  Use close_pcsc_reader
instead of code duplication.

* scd/apdu.c (pcsc_error_string): Add an error code.
* scd/scdaemon.c (scd_kick_the_loop): Fix a diagnostic.
--

There was an obvious bug in that the pcsc.count could go below zero
and thus there was no chance to get the context release.  Releasing
and recreating the context is at least under Windows important to get
rit of the PCSC_E_SERVICE_STOPPED.

Also removes a potential problem in holding the reader_table_lock
between calls to apdu_dev_list_start apdu_dev_list_finish.  There is
no need for this.  Instead we bump the pcsc.count.

The reader_table_lock strategy should be reviewed; we may be able to
remove it.

Signed-off-by: Werner Koch <wk@gnupg.org>

1 files changed, 3 insertions, 1 deletions

diff --git a/scd/scdaemon.c b/scd/scdaemon.c
index 60d68c26a..da1f21d1f 100644
--- a/scd/scdaemon.c
+++ b/scd/scdaemon.c
@@ -1231,7 +1231,7 @@ scd_kick_the_loop (void)
 #else
   int ret = kill (main_thread_pid, SIGCONT);
   if (ret < 0)
-    log_error ("SetEvent for scd_kick_the_loop failed: %s\n",
+    log_error ("sending signal for scd_kick_the_loop failed: %s\n",
                gpg_strerror (gpg_error_from_syserror ()));
 #endif
 }
@@ -1287,6 +1287,8 @@ handle_connections (int listen_fd)
 
     events[0] = the_event = INVALID_HANDLE_VALUE;
     events[1] = INVALID_HANDLE_VALUE;
+    /* Create event for manual reset, initially non-signaled.  Make it
+     * waitable and inheritable.  */
     h = CreateEvent (&sa, TRUE, FALSE, NULL);
     if (!h)
       log_error ("can't create scd event: %s\n", w32_strerror (-1) );