Merge branch 'egil/fix-crashdump-epmd/OTP-12447' into maint-17

* egil/fix-crashdump-epmd/OTP-12447:
  erts: Check driver version before assigning callback
  erts: Don't lookup invalid port for crashdump handling
  erts: Reserve a file descriptor for the crashdump file
  erts: Use emergency close to close epmd
  erts: Extend driver interface with emergency_close

5 files changed, 78 insertions, 46 deletions

diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h
index f9938fc66c..e498ac70ec 100644
--- a/erts/emulator/beam/erl_driver.h
+++ b/erts/emulator/beam/erl_driver.h
@@ -133,7 +133,7 @@ typedef struct {
 
 #define ERL_DRV_EXTENDED_MARKER		(0xfeeeeeed)
 #define ERL_DRV_EXTENDED_MAJOR_VERSION	3
-#define ERL_DRV_EXTENDED_MINOR_VERSION	1
+#define ERL_DRV_EXTENDED_MINOR_VERSION	2
 
 /*
  * The emulator will refuse to load a driver with a major version
@@ -361,6 +361,9 @@ typedef struct erl_drv_entry {
     	                        /* Called on behalf of driver_select when
 				   it is safe to release 'event'. A typical
 				   unix driver would call close(event) */
+    void (*emergency_close)(ErlDrvData drv_data);
+                                /* called when the port is closed abruptly.
+				   specifically when erl_crash_dump is called. */
     /* When adding entries here, dont forget to pad in obsolete/driver.h */
 } ErlDrvEntry;
 
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index 891046a8b5..32a2dc43e8 100644
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -160,6 +160,7 @@ struct erts_driver_t_ {
     void (*ready_async)(ErlDrvData drv_data, ErlDrvThreadData thread_data); /* Might be NULL */ 
     void (*process_exit)(ErlDrvData drv_data, ErlDrvMonitor *monitor);
     void (*stop_select)(ErlDrvEvent event, void*); /* Might be NULL */
+    void (*emergency_close)(ErlDrvData drv_data);  /* Might be NULL */
 };
 
 extern erts_driver_t *driver_list;
@@ -852,6 +853,7 @@ Uint erts_port_ioq_size(Port *pp);
 void erts_stale_drv_select(Eterm, ErlDrvPort, ErlDrvEvent, int, int);
 
 Port *erts_get_heart_port(void);
+void erts_emergency_close_ports(void);
 
 #if defined(ERTS_SMP) && defined(ERTS_ENABLE_LOCK_COUNT)
 void erts_lcnt_enable_io_lock_count(int enable);
diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c
index 9ae973e108..012a7d1a4b 100644
--- a/erts/emulator/beam/io.c
+++ b/erts/emulator/beam/io.c
@@ -7349,6 +7349,8 @@ no_stop_select_callback(ErlDrvEvent event, void* private)
     erts_send_error_to_logger_nogl(dsbufp);
 }
 
+#define IS_DRIVER_VERSION_GE(DE,MAJOR,MINOR) \
+    ((DE)->major_version >= (MAJOR) && (DE)->minor_version >= (MINOR))
 
 static int
 init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle)
@@ -7396,6 +7398,7 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle)
     drv->timeout = de->timeout ? de->timeout : no_timeout_callback;
     drv->ready_async = de->ready_async;
     drv->process_exit = de->process_exit;
+    drv->emergency_close = IS_DRIVER_VERSION_GE(de,3,2) ? de->emergency_close : NULL;
     if (de->stop_select)
 	drv->stop_select = de->stop_select;
     else
@@ -7414,6 +7417,8 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle)
     }
 }
 
+#undef IS_DRIVER_VERSION_GE
+
 void
 erts_destroy_driver(erts_driver_t *drv)
 {
@@ -7557,7 +7562,7 @@ Port *erts_get_heart_port(void)
 	if (!port)
 	    continue;
 	/* only examine undead or alive ports */
-	if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_DEAD)
+	if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP)
 	    continue;
 	/* immediate atom compare */
 	reg = port->common.u.alive.reg;
@@ -7568,3 +7573,23 @@ Port *erts_get_heart_port(void)
 
     return NULL;
 }
+
+void erts_emergency_close_ports(void)
+{
+    int ix, max = erts_ptab_max(&erts_port);
+
+    for (ix = 0; ix < max; ix++) {
+	Port *port = erts_pix2port(ix);
+
+	if (!port)
+	    continue;
+	/* only examine undead or alive ports */
+	if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP)
+	    continue;
+
+	/* emergency close socket */
+	if (port->drv_ptr->emergency_close) {
+	    port->drv_ptr->emergency_close((ErlDrvData) port->drv_data);
+	}
+    }
+}
diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c
index db8a251fdd..33d37f8a6a 100644
--- a/erts/emulator/drivers/common/inet_drv.c
+++ b/erts/emulator/drivers/common/inet_drv.c
@@ -268,14 +268,13 @@ static BOOL (WINAPI *fpSetHandleInformation)(HANDLE,DWORD,DWORD);
 #define sock_htonl(x)              htonl((x))
 #define sock_send(s,buf,len,flag)  send((s),(buf),(len),(flag))
 #define sock_sendv(s, vec, size, np, flag) \
-                WSASend((s),(WSABUF*)(vec),\
-				   (size),(np),(flag),NULL,NULL)
+            WSASend((s),(WSABUF*)(vec),(size),(np),(flag),NULL,NULL)
 #define sock_recv(s,buf,len,flag)  recv((s),(buf),(len),(flag))
 
 #define sock_recvfrom(s,buf,blen,flag,addr,alen) \
-                recvfrom((s),(buf),(blen),(flag),(addr),(alen))
+	    recvfrom((s),(buf),(blen),(flag),(addr),(alen))
 #define sock_sendto(s,buf,blen,flag,addr,alen) \
-                sendto((s),(buf),(blen),(flag),(addr),(alen))
+	    sendto((s),(buf),(blen),(flag),(addr),(alen))
 #define sock_hostname(buf, len)    gethostname((buf), (len))
 
 #define sock_getservbyname(name,proto) getservbyname((name),(proto))
@@ -360,9 +359,9 @@ static ssize_t writev_fallback(int fd, const struct iovec *iov, int iovcnt, int
 #define sock_accept(s, addr, len)   accept((s), (addr), (len))
 #define sock_send(s,buf,len,flag)   inet_send((s),(buf),(len),(flag))
 #define sock_sendto(s,buf,blen,flag,addr,alen) \
-                sendto((s),(buf),(blen),(flag),(addr),(alen))
+	    sendto((s),(buf),(blen),(flag),(addr),(alen))
 #define sock_sendv(s, vec, size, np, flag) \
-		(*(np) = writev_fallback((s), (struct iovec*)(vec), (size), (*(np))))
+	    (*(np) = writev_fallback((s), (struct iovec*)(vec), (size), (*(np))))
 #define sock_sendmsg(s,msghdr,flag) sendmsg((s),(msghdr),(flag))
 
 #define sock_open(af, type, proto)  socket((af), (type), (proto))
@@ -1178,6 +1177,7 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData, unsigned int,
 static void tcp_inet_timeout(ErlDrvData);
 static void tcp_inet_process_exit(ErlDrvData, ErlDrvMonitor *); 
 static void inet_stop_select(ErlDrvEvent, void*); 
+static void inet_emergency_close(ErlDrvData);
 #ifdef __WIN32__
 static void tcp_inet_event(ErlDrvData, ErlDrvEvent);
 static void find_dynamic_functions(void);
@@ -1288,7 +1288,8 @@ static struct erl_drv_entry tcp_inet_driver_entry =
     ERL_DRV_FLAG_USE_PORT_LOCKING|ERL_DRV_FLAG_SOFT_BUSY,
     NULL,
     tcp_inet_process_exit,
-    inet_stop_select
+    inet_stop_select,
+    inet_emergency_close
 };
 
 
@@ -1341,7 +1342,8 @@ static struct erl_drv_entry udp_inet_driver_entry =
     ERL_DRV_FLAG_USE_PORT_LOCKING,
     NULL,
     NULL,
-    inet_stop_select
+    inet_stop_select,
+    inet_emergency_close
 };
 #endif
 
@@ -1375,7 +1377,8 @@ static struct erl_drv_entry sctp_inet_driver_entry =
     ERL_DRV_FLAG_USE_PORT_LOCKING,
     NULL,
     NULL, /* process_exit */
-    inet_stop_select
+    inet_stop_select,
+    inet_emergency_close
 };
 #endif
 
@@ -1421,7 +1424,7 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event);
 static int packet_inet_output(udp_descriptor* udesc, HANDLE event);
 #endif
 
-/* convert descriptor poiner to inet_descriptor pointer */
+/* convert descriptor pointer to inet_descriptor pointer */
 #define INETP(d) (&(d)->inet)
 
 #ifdef __OSE__
@@ -8204,6 +8207,19 @@ static void inet_stop(inet_descriptor* desc)
     FREE(desc);
 }
 
+static void inet_emergency_close(ErlDrvData data)
+{
+    /* valid for any (UDP, TCP or SCTP) descriptor */
+    tcp_descriptor* tcp_desc = (tcp_descriptor*)data;
+    inet_descriptor* desc = INETP(tcp_desc);
+    DEBUGF(("inet_emergency_close(%ld) {s=%d\r\n",
+	    (long)desc->port, desc->s));
+    if (desc->s != INVALID_SOCKET) {
+	sock_close(desc->s);
+    }
+}
+
+
 static void set_default_msgq_limits(ErlDrvPort port)
 {
     ErlDrvSizeT q_high = INET_HIGH_MSGQ_WATERMARK;
diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c
index 0d677d5f34..eff64b2c11 100644
--- a/erts/emulator/sys/unix/sys.c
+++ b/erts/emulator/sys/unix/sys.c
@@ -202,8 +202,6 @@ static erts_smp_atomic_t sys_misc_mem_sz;
 #if defined(ERTS_SMP)
 static void smp_sig_notify(char c);
 static int sig_notify_fds[2] = {-1, -1};
-#elif defined(USE_THREADS)
-static int async_fd[2];
 #endif
 
 #if CHLDWTHR || defined(ERTS_SMP)
@@ -246,6 +244,8 @@ static void note_child_death(int, int);
 static void* child_waiter(void *);
 #endif
 
+static int crashdump_companion_cube_fd = -1;
+
 /********************* General functions ****************************/
 
 /* This is used by both the drivers and general I/O, must be set early */
@@ -575,6 +575,14 @@ erts_sys_pre_init(void)
       close(fd);
     }
 
+    /* We need a file descriptor to close in the crashdump creation.
+     * We close this one to be sure we can get a fd for our real file ...
+     * so, we create one here ... a stone to carry all the way home.
+     */
+
+    crashdump_companion_cube_fd = open("/dev/null", O_RDONLY);
+
+    /* don't lose it, there will be cake */
 }
 
 void
@@ -719,14 +727,13 @@ static ERTS_INLINE int
 prepare_crash_dump(int secs)
 {
 #define NUFBUF (3)
-    int i, max;
+    int i;
     char env[21]; /* enough to hold any 64-bit integer */
     size_t envsz;
     DeclareTmpHeapNoproc(heap,NUFBUF);
     Port *heart_port;
     Eterm *hp = heap;
     Eterm list = NIL;
-    int heart_fd[2] = {-1,-1};
     int has_heart = 0;
 
     UseTmpHeapNoproc(NUFBUF);
@@ -749,43 +756,22 @@ prepare_crash_dump(int secs)
 	alarm((unsigned int)secs);
     }
 
-    if (heart_port) {
-	/* hearts input fd
-	 * We "know" drv_data is the in_fd since the port is started with read|write
-	 */
-	heart_fd[0] = (int)heart_port->drv_data;
-	heart_fd[1] = (int)driver_data[heart_fd[0]].ofd;
-	has_heart   = 1;
+    /* close all viable sockets via emergency close callbacks.
+     * Specifically we want to close epmd sockets.
+     */
 
-	list = CONS(hp, make_small(8), list); hp += 2;
+    erts_emergency_close_ports();
 
+    if (heart_port) {
+	has_heart = 1;
+	list = CONS(hp, make_small(8), list); hp += 2;
 	/* send to heart port, CMD = 8, i.e. prepare crash dump =o */
 	erts_port_output(NULL, ERTS_PORT_SIG_FLG_FORCE_IMM_CALL, heart_port,
 			 heart_port->common.id, list, NULL);
     }
 
-    /* Make sure we unregister at epmd (unknown fd) and get at least
-       one free filedescriptor (for erl_crash.dump) */
-
-    max = max_files;
-    if (max < 1024)
-	max = 1024;
-    for (i = 3; i < max; i++) {
-#if defined(ERTS_SMP)
-	/* We don't want to close the signal notification pipe... */
-	if (i == sig_notify_fds[0] || i == sig_notify_fds[1])
-	    continue;
-#elif defined(USE_THREADS)
-	/* We don't want to close the async notification pipe... */
-	if (i == async_fd[0] || i == async_fd[1])
-	    continue;
-#endif
-	/* We don't want to close our heart yet ... */
-	if (i == heart_fd[0] || i == heart_fd[1])
-	    continue;
-
-	close(i);
-    }
+    /* Make sure we have a fd for our crashdump file. */
+    close(crashdump_companion_cube_fd);
 
     envsz = sizeof(env);
     i = erts_sys_getenv__("ERL_CRASH_DUMP_NICE", env, &envsz);