fix hang of remote logfile update

appending to logfile did not update copy_end_pos, so the copy
brick never switched off.

mars_aio.c

@@ -674,8 +674,11 @@ static int aio_submit_thread(void *data)
 	struct aio_threadinfo *tinfo = data;
 	struct aio_output *output = tinfo->output;
 	struct file *file = output->filp;
-	int err;
-	
+	mm_segment_t oldfs;
+	int err = -EINVAL;
+
+	CHECK_PTR_NULL(file, done);
+
 	/* TODO: this is provisionary. We only need it for sys_io_submit()
 	 * which uses userspace concepts like file handles.
 	 * This should be accompanied by a future kernelsapce vfs_submit() or
@@ -687,7 +690,7 @@ static int aio_submit_thread(void *data)
 	if (unlikely(err < 0))
 		goto done;
 	output->fd = err;
-	fd_install(err, output->filp);
+	fd_install(err, file);
 
 	MARS_INF("kthread has started.\n");
 	//set_user_nice(current, -20);
@@ -696,28 +699,18 @@ static int aio_submit_thread(void *data)
 
 	err = -ENOMEM;
 	if (unlikely(!current->mm))
-		goto done;
-
-#if 1
-	if (true) {
-		mm_segment_t oldfs;
-		if (!current->mm) {
-			MARS_ERR("mm = %p\n", current->mm);
-			err = -EINVAL;
-			goto done;
-		}
-		oldfs = get_fs();
-		set_fs(get_ds());
-		err = sys_io_setup(MARS_MAX_AIO, &output->ctxp);
-		set_fs(oldfs);
-		if (unlikely(err < 0))
-			goto done;
-	}
-#endif
+		goto cleanup_fd;
 
+	oldfs = get_fs();
+	set_fs(get_ds());
+	err = sys_io_setup(MARS_MAX_AIO, &output->ctxp);
+	set_fs(oldfs);
+	if (unlikely(err < 0))
+		goto cleanup_mm;
+	
 	err = aio_start_thread(output, 1, aio_event_thread);
 	if (unlikely(err < 0))
-		goto done;
+		goto cleanup_ctxp;
 
 	while (!kthread_should_stop()) {
 		struct aio_mref_aspect *mref_a;
@@ -802,24 +795,26 @@ static int aio_submit_thread(void *data)
 
 	aio_stop_thread(output, 1, true);
 
-#if 1
-	if (true) {
-		mm_segment_t oldfs;
-		MARS_INF("destroying ioctx.....\n");
-		oldfs = get_fs();
-		set_fs(get_ds());
-		sys_io_destroy(output->ctxp);
-		set_fs(oldfs);
-		output->ctxp = 0;
-	}
-#endif
-	MARS_INF("destroying fd %d\n", output->fd);
+cleanup_ctxp:
+	MARS_DBG("destroying ioctx.....\n");
+	oldfs = get_fs();
+	set_fs(get_ds());
+	sys_io_destroy(output->ctxp);
+	set_fs(oldfs);
+	output->ctxp = 0;
+
+cleanup_mm:
+	unuse_fake_mm();
+
+cleanup_fd:
+	MARS_DBG("destroying fd %d\n", output->fd);
 	fd_uninstall(output->fd);
 	put_unused_fd(output->fd);
-	unuse_fake_mm();
+
 	err = 0;
 
 done:
+	MARS_DBG("status = %d\n", err);
 	tinfo->terminated = true;
 	wake_up_interruptible_all(&tinfo->event);
 	return err;
@@ -925,7 +920,7 @@ static int aio_switch(struct aio_brick *brick)
 	if (!brick->power.button)
 		goto cleanup;
 
-	if (brick->power.led_on)
+	if (brick->power.led_on || output->filp)
 		goto done;
 
 	mars_power_led_off((void*)brick, false);
@@ -940,13 +935,16 @@ static int aio_switch(struct aio_brick *brick)
 	output->filp = filp_open(path, flags, prot);
 	set_fs(oldfs);
 	
-	if (unlikely(IS_ERR(output->filp))) {
+	MARS_DBG("opened file '%s' flags = %d prot = %d filp = %p\n", path, flags, prot, output->filp);
+
+	if (unlikely(!output->filp || IS_ERR(output->filp))) {
 		err = PTR_ERR(output->filp);
 		MARS_ERR("can't open file '%s' status=%d\n", path, err);
 		output->filp = NULL;
+		if (err >= 0)
+			err = -ENOENT;
 		return err;
 	}
-	MARS_DBG("opened file '%s'\n", path);
 #if 1
 	{
 		struct inode *inode = output->filp->f_mapping->host;

sy_old/mars_light.c

@@ -758,14 +758,15 @@ bool _is_peer_logfile(const char *name, const char *id)
 }
 
 static
-int _update_file(struct mars_global *global, const char *switch_path, const char *copy_path, const char *file, const char *peer, loff_t end_pos)
+int _update_file(struct mars_rotate *rot, const char *switch_path, const char *copy_path, const char *file, const char *peer, loff_t end_pos)
 {
+	struct mars_global *global = rot->global;
 	const char *tmp = path_make("%s@%s", file, peer);
 	const char *argv[2] = { tmp, file };
 	struct copy_brick *copy = NULL;
 	int status = -ENOMEM;
 
-	if (unlikely(!tmp))
+	if (unlikely(!tmp || !global))
 		goto done;
 
 	MARS_DBG("src = '%s' dst = '%s'\n", tmp, file);
@@ -774,6 +775,7 @@ int _update_file(struct mars_global *global, const char *switch_path, const char
 		if (end_pos > copy->copy_end) {
 			MARS_DBG("appending to '%s' %lld => %lld\n", copy_path, copy->copy_end, end_pos);
 			copy->copy_end = end_pos;
+			rot->copy_end_pos = end_pos;
 		}
 	}
 
@@ -820,7 +822,9 @@ int check_logfile(struct mars_peerinfo *peer, struct mars_dent *remote_dent, str
 		bool copy_is_done = (is_my_copy && copy_brick->copy_last == copy_brick->copy_end && local_dent != NULL && copy_brick->copy_end == rot->copy_end_pos);
 		bool is_next_copy = (remote_dent->d_serial == parent->d_logfile_serial + 1);
 
-		MARS_DBG("current copy brick '%s' copy_last = %lld copy_end = %lld dent '%s' serial = %d/%d is_done = %d is_my_copy = %d is_next_copy = %d\n", copy_brick->brick_path, copy_brick->copy_last, copy_brick->copy_end, remote_dent->d_path, remote_dent->d_serial, parent->d_logfile_serial, copy_is_done, is_my_copy, is_next_copy);
+		MARS_DBG("current copy brick '%s' copy_last = %lld copy_end = %lld dent '%s' serial = %d/%d local_dent = '%s' copy_end_pos = %lld | is_done = %d is_my_copy = %d is_next_copy = %d\n",
+			 copy_brick->brick_path, copy_brick->copy_last, copy_brick->copy_end, remote_dent->d_path, remote_dent->d_serial, parent->d_logfile_serial, local_dent ? local_dent->d_path : "", rot->copy_end_pos,
+			 copy_is_done, is_my_copy, is_next_copy);
 
 		if (is_my_copy) {
 			rot->copy_is_done = copy_is_done;
@@ -857,7 +861,7 @@ treat:
 	
 	// start / treat copy brick instance
 	rot->copy_end_pos = src_size;
-	status = _update_file(peer->global, switch_path, copy_path, remote_dent->d_path, peer->peer, src_size);
+	status = _update_file(rot, switch_path, copy_path, remote_dent->d_path, peer->peer, src_size);
 	MARS_DBG("update '%s' from peer '%s' status = %d\n", remote_dent->d_path, peer->peer, status);
 	if (status < 0) {
 		goto done;