[rebootcontrol] Adapt do new WakeOnLan task

2 files changed, 187 insertions, 200 deletions

diff --git a/modules-available/rebootcontrol/inc/rebootcontrol.inc.php b/modules-available/rebootcontrol/inc/rebootcontrol.inc.php
index 59d20641..99e8a4cf 100644
--- a/modules-available/rebootcontrol/inc/rebootcontrol.inc.php
+++ b/modules-available/rebootcontrol/inc/rebootcontrol.inc.php
@@ -48,7 +48,7 @@ class RebootControl
 			"port" => 9922, // Hard-coded, must match mgmt-sshd module
 		));
 		if (!Taskmanager::isFailed($task)) {
-			self::addTask($task['id'], self::TASK_REBOOTCTL, $list, $task['id'], ['action' => $mode]);
+			self::addTask($task['id'], self::TASK_REBOOTCTL, $list, ['action' => $mode]);
 			foreach ($list as $client) {
 				$client['mode'] = $mode;
 				$client['minutes'] = $minutes;
@@ -58,7 +58,7 @@ class RebootControl
 		return $task;
 	}
 
-	private static function addTask($id, $type, $clients, $taskIds, $other = false)
+	private static function addTask(string $taskId, string $type, array $clients, array $other = null)
 	{
 		$lids = ArrayUtil::flattenByKey($clients, 'locationid');
 		$lids = array_unique($lids);
@@ -70,15 +70,12 @@ class RebootControl
 			}
 			$newClients[] = $d;
 		}
-		if (!is_array($taskIds)) {
-			$taskIds = [$taskIds];
-		}
 		$data = [
-			'id' => $id,
+			'id' => $taskId,
 			'type' => $type,
 			'locations' => $lids,
 			'clients' => $newClients,
-			'tasks' => $taskIds,
+			'tasks' => [$taskId], // This did hold multiple tasks in the past; keep it in case we need this again
 		];
 		if (is_array($other)) {
 			$data += $other;
@@ -88,9 +85,10 @@ class RebootControl
 
 	/**
 	 * @param int[]|null $locations filter by these locations
+	 * @param ?string $id only with this TaskID
 	 * @return array|false list of active tasks for reboots/shutdowns.
 	 */
-	public static function getActiveTasks($locations = null, $id = null)
+	public static function getActiveTasks(array $locations = null, $id = null)
 	{
 		if (is_array($locations) && in_array(0, $locations)) {
 			$locations = null;
@@ -149,7 +147,7 @@ class RebootControl
 	{
 		$task = self::runScriptInternal($clients, $command, $timeout, $privkey);
 		if (!Taskmanager::isFailed($task)) {
-			self::addTask($task['id'], self::TASK_EXEC, $clients, $task['id']);
+			self::addTask($task['id'], self::TASK_EXEC, $clients);
 		}
 		return $task;
 	}
@@ -212,13 +210,12 @@ class RebootControl
 	}
 
 	/**
-	 * @param array $sourceMachines list of source machines. array of [clientip, machineuuid] entries
-	 * @param string $bcast directed broadcast address to send to
-	 * @param string|string[] $macaddr destination mac address(es)
-	 * @param string $passwd optional WOL password, mac address or ipv4 notation
-	 * @return array|false task struct, false on error
+	 * @param string|string[] $macs
+	 * @param $bcast
+	 * @param $passwd
+	 * @return string
 	 */
-	public static function wakeViaClient($sourceMachines, $macaddr, $bcast = false, $passwd = false)
+	private static function buildClientWakeCommand($macs, $bcast = false, $passwd = false): string
 	{
 		$command = 'jawol';
 		if (!empty($bcast)) {
@@ -229,22 +226,35 @@ class RebootControl
 		if (!empty($passwd)) {
 			$command .= " -p '$passwd'";
 		}
-		if (is_array($macaddr)) {
-			$macaddr = implode("' '", $macaddr);
+		if (is_array($macs)) {
+			$macs = implode(" ", $macs);
 		}
-		$command .= " '$macaddr'";
-		// Yes there is one zero missing from the usleep -- that's the whole point: we prefer 100ms sleeps
+		$command .= " $macs";
+		return $command;
+	}
+
+	/**
+	 * @param array $sourceMachines list of source machines. array of [clientip, machineuuid] entries
+	 * @param string $bcast directed broadcast address to send to
+	 * @param string|string[] $macaddr destination mac address(es)
+	 * @param string $passwd optional WOL password, mac address or ipv4 notation
+	 * @return array|false task struct, false on error
+	 */
+	public static function wakeViaClient($sourceMachines, $macaddr, string $bcast = null, string $passwd = null)
+	{
+		$command = self::buildClientWakeCommand($macaddr, $bcast, $passwd);
+		// Yes there is one zero "missing" from the usleep -- that's the whole point: we prefer 100ms sleeps
 		return self::runScriptInternal($sourceMachines,
 			"for i in 1 1 0; do $command; usleep \${i}00000 2> /dev/null || sleep \$i; done");
 	}
 
 	/**
 	 * @param string|string[] $macaddr destination mac address(es)
-	 * @param string $bcast directed broadcast address to send to
-	 * @param string $passwd optional WOL password; mac address or ipv4 notation
+	 * @param ?string $bcast directed broadcast address to send to
+	 * @param ?string $passwd optional WOL password; mac address or ipv4 notation
 	 * @return array|false task struct, false on error
 	 */
-	public static function wakeDirectly($macaddr, $bcast = false, $passwd = false)
+	public static function wakeDirectly($macaddr, string $bcast = null, string $passwd = null)
 	{
 		if (!is_array($macaddr)) {
 			$macaddr = [$macaddr];
@@ -261,7 +271,14 @@ class RebootControl
 		]);
 	}
 
-	public static function wakeViaJumpHost($jumphost, $bcast, $clients)
+	/**
+	 * Explicitly wake given clients via jumphost
+	 * @param array $jumphost the according row from the database, representing the desired jumphost
+	 * @param string $bcast (directed) broadcast address for WOL packet, %IP% in command template
+	 * @param array $clients list of clients, must contain at least key 'macaddr' for every client
+	 * @return array|false task struct on successful submission to TM, false on error
+	 */
+	public static function wakeViaJumpHost(array $jumphost, string $bcast, array $clients)
 	{
 		$hostid = $jumphost['hostid'];
 		$macs = ArrayUtil::flattenByKey($clients, 'macaddr');
@@ -285,61 +302,38 @@ class RebootControl
 	}
 
 	/**
-	 * @param array $list list of clients containing each keys 'macaddr' and 'clientip'
-	 * @return string id of this job
+	 * @param array $clientList list of clients containing each keys 'macaddr' and 'clientip'
+	 * @param array $failed list of failed clients from $clientList
+	 * @return ?string taskid of this job
 	 */
-	public static function wakeMachines($list, &$failed = [])
+	public static function wakeMachines(array $clientList, array &$failed = [])
 	{
-		/* TODO: Refactor mom's spaghetti
-		 * Now that I figured out what I want, do something like this:
-		 * 1) Group clients by subnet
-		 * 2) Only after step 1, start to collect possible ways to wake up clients for each subnet that's not empty
-		 * 3) Have some priority list for the methods, extend Taskmanager to have "negative dependency"
-		 *    i.e. submit task B with task A as parent task, but only launch task B if task A failed.
-		 *    If task A succeeded, mark task B as FINISHED immediately without actually running it.
-		 *    (or introduce new statusCode for this?)
-		 */
 		$errors = '';
-		$tasks = [];
-		$bad = $unknown = [];
+		$sent = $unknown = $unreachable = [];
 		// For event filtering by rule
-		$events = [];
 		// Need all subnets...
+		/* subnetid => [
+		 * subnetid => 1234,
+		 * start => 1234, (ip2long)
+		 * end => 5678, (ip2long)
+		 * jumphosts => [id1, id2, ...],
+		 */
 		$subnets = [];
 		$res = Database::simpleQuery('SELECT subnetid, start, end, isdirect FROM reboot_subnet');
 		foreach ($res as $row) {
 			$row += [
-				'jumphosts' => [],
-				'direct' => [],
-				'indirect' => [],
+				'djumphosts' => [],
+				'ijumphosts' => [],
 			];
 			$subnets[$row['subnetid']] = $row;
 		}
 		// Get all jump hosts
-		$jumphosts = [];
-		$res = Database::simpleQuery('SELECT jh.hostid, host, port, username, sshkey, script, jh.reachable,
-       	Group_Concat(jxs.subnetid) AS subnets1, Group_Concat(sxs.dstid) AS subnets2
-			FROM reboot_jumphost jh
-			LEFT JOIN reboot_jumphost_x_subnet jxs ON (jh.hostid = jxs.hostid)
-			LEFT JOIN reboot_subnet s ON (INET_ATON(jh.host) BETWEEN s.start AND s.end)
-			LEFT JOIN reboot_subnet_x_subnet sxs ON (sxs.srcid = s.subnetid AND sxs.reachable <> 0)
-			GROUP BY jh.hostid');
-		foreach ($res as $row) {
-			if ($row['subnets1'] === null && $row['subnets2'] === null)
-				continue;
-			$nets = explode(',', $row['subnets1'] . ',' . $row['subnets2']);
-			foreach ($nets as $net) {
-				if (empty($net) || !isset($subnets[$net]))
-					continue;
-				$subnets[$net]['jumphosts'][$row['hostid']] = $row['hostid'];
-			}
-			$row['jobs'] = [];
-			$jumphosts[$row['hostid']] = $row;
-		}
-		// Group by subnet
-		foreach ($list as $client) {
-			$ip = sprintf('%u', ip2long($client['clientip']));
-			//$client['numip'] = $ip;
+		self::addJumphostsToSubnets($subnets);
+		// Determine method for all clients
+		$taskClients = []; // array of arrays with keys [ip, mac, methods]
+		$taskSsh = []; // SSH configs for task, array of arrays with keys [username, sshkey, ip, port, command]
+		foreach ($clientList as $client) {
+			$ip = sprintf('%u', ip2long($client['clientip'])); // 32Bit snprintf
 			unset($subnet);
 			$subnet = false;
 			foreach ($subnets as &$sn) {
@@ -352,134 +346,63 @@ class RebootControl
 				$unknown[] = $client;
 				continue;
 			}
-			$ok = false;
-			if (!$ok && $subnet['isdirect']) {
-				// Directly reachable
-				$subnet['direct'][] = $client;
-				$ok = true;
-			}
-			if (!$ok && !empty($subnet['jumphosts'])) {
-				foreach ($subnet['jumphosts'] as $hostid) {
-					if ($jumphosts[$hostid]['reachable'] != 0) {
-						$jumphosts[$hostid]['jobs'][$subnet['end']][] = $client;
-						$ok = true;
-						break;
-					}
-				}
-			}
-			if (!$ok) {
-				// find clients in same subnet, or reachable ones
-				self::findMachinesForSubnet($subnet);
-				if (empty($subnet['dclients']) && empty($subnet['iclients'])) {
-					// Nothing found -- cannot wake this host
-					$bad[] = $client;
-				} else {
-					// Found suitable indirect host
-					$subnet['indirect'][] = $client;
-					$ok = true;
-				}
+			$c = [
+				'ip' => long2ip($subnet['end']),
+				'mac' => $client['macaddr'],
+				'methods' => [],
+			];
+			self::findMachinesForSubnet($subnet);
+			// Highest priority - clients in same subnet, no directed broadcast
+			// required, should be most reliable
+			self::addSshMethodUsingClient($subnet['dclients'], $c['methods'], $taskSsh);
+			// Jumphost - usually in same subnet
+			self::addSshMethodUsingJumphost($subnet['djumphosts'], true, $c['methods'], $taskSsh);
+			// Use clients in other subnets, known to be able to reach the destination net
+			self::addSshMethodUsingClient($subnet['iclients'], $c['methods'], $taskSsh);
+			// Jumphosts in other subnets, determined to be able to reach destination subnet
+			self::addSshMethodUsingJumphost($subnet['ijumphosts'], true, $c['methods'], $taskSsh);
+			// If directly reachable from server, prefer this now
+			if ($subnet['isdirect']) {
+				$c['methods'][] = 'DIRECT';
 			}
-			if ($ok && isset($client['machineuuid'])) {
+			if (empty($c['methods'])) {
+				$unreachable[] = $client;
+			} else {
 				// TODO: Remember WOL was attempted
 			}
-		}
-		unset($subnet);
-		// Batch process
-		// First, via jump host
-		foreach ($jumphosts as $jh) {
-			foreach ($jh['jobs'] as $bcast => $clients) {
-				$errors .= 'Via jumphost ' . $jh['host'] . ': ' . implode(', ', ArrayUtil::flattenByKey($clients, 'clientip')) . "\n";
-				$task = self::wakeViaJumpHost($jh, $bcast, $clients);
-				if (Taskmanager::isFailed($task)) {
-					// TODO: Figure out $subnet from $bcast and queue as indirect
-					// (rather, overhaul this whole spaghetti code)
-					$errors .= ".... FAILED TO LAUNCH TASK ON JUMPHOST!\n";
-				} else {
-					self::addEventList($events, $clients, 'jumphost', $jh['host']);
-				}
-			}
-		}
-		// Server or client
-		foreach ($subnets as $subnet) {
-			if (!empty($subnet['direct'])) {
-				// Can wake directly
-				if (!self::wakeGroup('From server', $tasks, $errors, null, $subnet['direct'], $subnet['end'])) {
-					if (!empty($subnet['dclients']) || !empty($subnet['iclients'])) {
-						$errors .= "Re-queueing clients for indirect wakeup\n";
-						$subnet['indirect'] = array_merge($subnet['indirect'], $subnet['direct']);
-					}
-				} else {
-					self::addEventList($events, $subnet['direct'], 'satellite');
-				}
-			}
-			if (!empty($subnet['indirect'])) {
-				// Can wake indirectly
-				$ok = false;
-				if (!empty($subnet['dclients'])) {
-					$ok = true;
-					if (!self::wakeGroup('in same subnet', $tasks, $errors, $subnet['dclients'], $subnet['indirect'])) {
-						if (!empty($subnet['iclients'])) {
-							$errors .= "Re-re-queueing clients for indirect wakeup\n";
-							$ok = false;
-						}
-					} else {
-						self::addEventList($events, $subnet['indirect'], 'same-subnet', $subnet['dclients'][0]['clientip']);
-					}
-				}
-				if (!$ok && !empty($subnet['iclients'])) {
-					$ok = self::wakeGroup('across subnets', $tasks, $errors, $subnet['iclients'], $subnet['indirect'], $subnet['end']);
-					if ($ok) {
-						self::addEventList($events, $subnet['indirect'], 'other-subnet', $subnet['iclients'][0]['clientip']);
-					}
-				}
-				if (!$ok) {
-					$errors .= "I'm all out of ideas.\n";
-				}
+			// Only other fallback is jumphosts that were not reachable when last checked, this is really a last resort
+			self::addSshMethodUsingJumphost($subnet['djumphosts'], false, $c['methods'], $taskSsh);
+			self::addSshMethodUsingJumphost($subnet['ijumphosts'], false, $c['methods'], $taskSsh);
+
+			if (!empty($c['methods'])) {
+				$taskClients[] = $c;
+				$sent[] = $client;
 			}
 		}
-		if (!empty($bad)) {
-			$ips = ArrayUtil::flattenByKey($bad, 'clientip');
-			$errors .= "**** WARNING ****\nNo way to send WOL packets to the following machines:\n" . implode("\n", $ips) . "\n";
-		}
+		unset($subnet);
+
 		if (!empty($unknown)) {
 			$ips = ArrayUtil::flattenByKey($unknown, 'clientip');
 			$errors .= "**** WARNING ****\nThe following clients do not belong to a known subnet (bug?)\n" . implode("\n", $ips) . "\n";
 		}
-		$failed = array_merge($bad, $unknown);
-		$id = Util::randomUuid();
-		self::addTask($id, self::TASK_WOL, $list, $tasks, ['log' => $errors]);
-		foreach ($events as $event) {
-			EventLog::applyFilterRules('#action-wol', $event);
+		if (!empty($unreachable)) {
+			$ips = ArrayUtil::flattenByKey($unreachable, 'clientip');
+			$errors .= "**** WARNING ****\nThe following clients are not reachable with any method\n" . implode("\n", $ips) . "\n";
 		}
-		return $id;
-	}
-
-	private static function wakeGroup($type, &$tasks, &$errors, $via, $clients, $bcast = false)
-	{
-		$macs = ArrayUtil::flattenByKey($clients, 'macaddr');
-		$ips = ArrayUtil::flattenByKey($clients, 'clientip');
-		if ($via !== null) {
-			$srcips = ArrayUtil::flattenByKey($via, 'clientip');
-			$errors .= 'Via ' . implode(', ', $srcips) . ' ';
-		}
-		$errors .= $type . ': ' . implode(', ', $ips);
-		if ($bcast !== false) {
-			$errors .= ' (UDP to ' . long2ip($bcast) . ')';
-		}
-		$errors .= "\n";
-		if ($via === null) {
-			$task = self::wakeDirectly($macs, $bcast);
-		} else {
-			$task = self::wakeViaClient($via, $macs, $bcast);
-		}
-		if ($task !== false && isset($task['id'])) {
-			$tasks[] = $task['id'];
-		}
-		if (Taskmanager::isFailed($task)) {
-			$errors .= ".... FAILED TO START ACCORDING TASK!\n";
-			return false;
+		$failed = array_unique(array_merge($unknown, $unreachable));
+		$task = Taskmanager::submit('WakeOnLan', [
+			'clients' => $taskClients,
+			'ssh' => $taskSsh,
+		]);
+		if (isset($task['id'])) {
+			$id = $task['id'];
+			self::addTask($id, self::TASK_WOL, $clientList, ['log' => $errors]);
+			foreach ($sent as $client) {
+				EventLog::applyFilterRules('#action-wol', $client);
+			}
+			return $id;
 		}
-		return true;
+		return null;
 	}
 
 	private static function findMachinesForSubnet(&$subnet)
@@ -488,15 +411,12 @@ class RebootControl
 			return;
 		$cutoff = time() - 320;
 		// Get clients from same subnet first
-		$subnet['dclients'] = Database::queryAll("SELECT machineuuid, clientip FROM machine
+		$subnet['dclients'] = Database::queryColumnArray("SELECT clientip FROM machine
 			WHERE state IN ('IDLE', 'OCCUPIED') AND INET_ATON(clientip) BETWEEN :start AND :end AND lastseen > :cutoff
 			LIMIT 3",
 			['start' => $subnet['start'], 'end' => $subnet['end'], 'cutoff' => $cutoff]);
-		$subnet['iclients'] = [];
-		if (!empty($subnet['dclients']))
-			return;
 		// If none, get clients from other subnets known to be able to reach this one
-		$subnet['iclients'] = Database::queryAll("SELECT m.machineuuid, m.clientip FROM reboot_subnet_x_subnet sxs
+		$subnet['iclients'] = Database::queryColumnArray("SELECT m.clientip FROM reboot_subnet_x_subnet sxs
     		INNER JOIN reboot_subnet s ON (s.subnetid = sxs.srcid AND sxs.dstid = :subnetid AND sxs.reachable = 1)
 			INNER JOIN machine m ON (INET_ATON(m.clientip) BETWEEN s.start AND s.end AND state IN ('IDLE', 'OCCUPIED') AND m.lastseen > :cutoff)
 			LIMIT 20", ['subnetid' => $subnet['subnetid'], 'cutoff' => $cutoff]);
@@ -517,20 +437,87 @@ class RebootControl
 	}
 
 	/**
-	 * Add given clients to given event array
-	 * @param array $events
-	 * @param array $clients
-	 * @param string $type
-	 * @param null $via
+	 * Append a "wake via client" WOL method for the given client. Append at least one, but stop
+	 * if there are at least two methods already.
+	 *
+	 * @param array $sshClients [in] list of online clients to use for waking
+	 * @param array $c [out] The client's methods array
+	 * @param array $taskSsh [out] add according task struct to this array, if not already exists
+	 * @return void
 	 */
-	private static function addEventList(array &$events, array $clients, string $type, $via = null)
+	private static function addSshMethodUsingClient(array $sshClients, array &$methods, array &$taskSsh)
 	{
-		foreach ($clients as $client) {
-			$client['type'] = $type;
-			if (!empty($via)) {
-				$client['via'] = $via;
+		foreach ($sshClients as $host) {
+			if (!isset($taskSsh[$host])) {
+				$taskSsh[$host] = [
+					'username' => 'root',
+					'sshkey' => SSHKey::getPrivateKey(),
+					'ip' => $host,
+					'port' => 9922,
+					'command' => self::buildClientWakeCommand('%MACS%', '%IP%'),
+				];
+			}
+			$methods[] = $host;
+			if (count($methods) >= 2)
+				break;
+		}
+	}
+
+	private static function addSshMethodUsingJumphost(array $jumpHosts, bool $reachable, array &$methods, array &$taskSsh)
+	{
+		// If it's the fallback to apparently unreachable jump-hosts, ignore if we already have two methods
+		if (!$reachable && count($methods) >= 2)
+			return;
+		// host, port, username, sshkey, script, jh.reachable
+		foreach ($jumpHosts as $jh) {
+			if ($reachable !== (bool)$jh['reachable'])
+				continue;
+			$key = substr(md5($jh['host'] . ':' . $jh['port'] . ':' . $jh['username']), 0, 10);
+			if (!isset($taskSsh[$key])) {
+				$taskSsh[$key] = [
+					'username' => $jh['username'],
+					'sshkey' => $jh['sshkey'],
+					'ip' => $jh['host'],
+					'port' => $jh['port'],
+					'command' => $jh['script'],
+				];
+			}
+			$methods[] = $key;
+			if (count($methods) >= 2)
+				break;
+		}
+	}
+
+	/**
+	 * Load all jumphosts from DB, sort into passed $subnets. Also split up
+	 * by directly assigned subnets, and indirectly dtermined, reachable subnets.
+	 * @param array $subnets [in]
+	 * @return void
+	 */
+	private static function addJumphostsToSubnets(array &$subnets)
+	{
+		$res = Database::simpleQuery('SELECT host, port, username, sshkey, script, jh.reachable,
+       	Group_Concat(jxs.subnetid) AS dsubnets, Group_Concat(sxs.dstid) AS isubnets
+			FROM reboot_jumphost jh
+			LEFT JOIN reboot_jumphost_x_subnet jxs ON (jh.hostid = jxs.hostid)
+			LEFT JOIN reboot_subnet s ON (INET_ATON(jh.host) BETWEEN s.start AND s.end)
+			LEFT JOIN reboot_subnet_x_subnet sxs ON (sxs.srcid = s.subnetid AND sxs.reachable <> 0)
+			GROUP BY jh.hostid');
+		foreach ($res as $row) {
+			$dnets = empty($row['dsubnets']) ? [] : explode(',', $row['dsubnets']);
+			$inets = empty($row['isubnets']) ? [] : explode(',', $row['isubnets']);
+			$inets = array_diff($inets, $dnets); // There might be duplicates if both joins match
+			foreach ($dnets as $net) {
+				if (empty($net) || !isset($subnets[$net]))
+					continue;
+				$subnets[$net]['djumphosts'][] =& $row;
+			}
+			foreach ($inets as $net) {
+				if (empty($net) || !isset($subnets[$net]))
+					continue;
+				$subnets[$net]['ijumphosts'][] =& $row;
 			}
-			$events[] = $client;
+			unset($row);
 		}
 	}
 
diff --git a/modules-available/rebootcontrol/templates/status-wol.html b/modules-available/rebootcontrol/templates/status-wol.html
index 5a53a6f8..38e581fe 100644
--- a/modules-available/rebootcontrol/templates/status-wol.html
+++ b/modules-available/rebootcontrol/templates/status-wol.html
@@ -4,7 +4,7 @@
 <div class="clearfix slx-space"></div>
 
 {{#tasks}}
-<div data-tm-id="{{.}}" data-tm-callback="wolCallback" data-tm-log="messages" data-tm-log-fail-only="true">{{lang_aWolJob}}</div>
+<div data-tm-id="{{.}}" data-tm-callback="wolCallback" data-tm-log="messages">{{lang_aWolJob}}</div>
 {{/tasks}}
 {{^tasks}}
 <div class="alert alert-warning">
@@ -71,4 +71,4 @@ function wolCallback(task) {
 		$do.text(txt);
 	}
 }
-//--></script>
\ No newline at end of file
+//--></script>