Proper async reaping (or alarm(2)-based delooping at least) for ZED

cmd/zed/agents/zfs_agents.c

@@ -392,6 +392,7 @@ zfs_agent_init(libzfs_handle_t *zfs_hdl)
 		list_destroy(&agent_events);
 		zed_log_die("Failed to initialize agents");
 	}
+	pthread_setname_np(g_agents_tid, "agents");
 }
 
 void

cmd/zed/agents/zfs_mod.c

@@ -918,6 +918,7 @@ zfs_slm_init()
 		return (-1);
 	}
 
+	pthread_setname_np(g_zfs_tid, "enum-pools");
 	list_create(&g_device_list, sizeof (struct pendingdev),
 	    offsetof(struct pendingdev, pd_node));
 

cmd/zed/zed.c

@@ -60,8 +60,8 @@ _setup_sig_handlers(void)
 		zed_log_die("Failed to initialize sigset");
 
 	sa.sa_flags = SA_RESTART;
-	sa.sa_handler = SIG_IGN;
 
+	sa.sa_handler = SIG_IGN;
 	if (sigaction(SIGPIPE, &sa, NULL) < 0)
 		zed_log_die("Failed to ignore SIGPIPE");
 
@@ -75,6 +75,10 @@ _setup_sig_handlers(void)
 	sa.sa_handler = _hup_handler;
 	if (sigaction(SIGHUP, &sa, NULL) < 0)
 		zed_log_die("Failed to register SIGHUP handler");
+
+	(void) sigaddset(&sa.sa_mask, SIGCHLD);
+	if (pthread_sigmask(SIG_BLOCK, &sa.sa_mask, NULL) < 0)
+		zed_log_die("Failed to block SIGCHLD");
 }
 
 /*

cmd/zed/zed_conf.c

@@ -51,6 +51,7 @@ zed_conf_create(void)
 	zcp->state_fd = -1;		/* opened via zed_conf_open_state() */
 	zcp->zfs_hdl = NULL;		/* opened via zed_event_init() */
 	zcp->zevent_fd = -1;		/* opened via zed_event_init() */
+	zcp->max_jobs = 16;
 
 	if (!(zcp->conf_file = strdup(ZED_CONF_FILE)))
 		goto nomem;
@@ -172,6 +173,8 @@ _zed_conf_display_help(const char *prog, int got_err)
 	    "Write daemon's PID to FILE.", ZED_PID_FILE);
 	fprintf(fp, "%*c%*s %s [%s]\n", w1, 0x20, -w2, "-s FILE",
 	    "Write daemon's state to FILE.", ZED_STATE_FILE);
+	fprintf(fp, "%*c%*s %s [%d]\n", w1, 0x20, -w2, "-j JOBS",
+	    "Start at most JOBS at once.", 16);
 	fprintf(fp, "\n");
 
 	exit(got_err ? EXIT_FAILURE : EXIT_SUCCESS);
@@ -251,8 +254,9 @@ _zed_conf_parse_path(char **resultp, const char *path)
 void
 zed_conf_parse_opts(struct zed_conf *zcp, int argc, char **argv)
 {
-	const char * const opts = ":hLVc:d:p:P:s:vfFMZI";
+	const char * const opts = ":hLVc:d:p:P:s:vfFMZIj:";
 	int opt;
+	unsigned long raw;
 
 	if (!zcp || !argv || !argv[0])
 		zed_log_die("Failed to parse options: Internal error");
@@ -303,6 +307,17 @@ zed_conf_parse_opts(struct zed_conf *zcp, int argc, char **argv)
 		case 'Z':
 			zcp->do_zero = 1;
 			break;
+		case 'j':
+			errno = 0;
+			raw = strtoul(optarg, NULL, 0);
+			if (errno == ERANGE || raw > INT16_MAX) {
+				zed_log_die("%lu is too many jobs", raw);
+			} if (raw == 0) {
+				zed_log_die("0 jobs makes no sense");
+			} else {
+				zcp->max_jobs = raw;
+			}
+			break;
 		case '?':
 		default:
 			if (optopt == '?')

cmd/zed/zed_conf.h

@@ -39,6 +39,7 @@ struct zed_conf {
 	libzfs_handle_t	*zfs_hdl;		/* handle to libzfs */
 	int		zevent_fd;		/* fd for access to zevents */
 	char		*path;		/* custom $PATH for zedlets to use */
+	int16_t max_jobs;		/* max zedlets to run at one time */
 };
 
 struct zed_conf *zed_conf_create(void);

cmd/zed/zed_disk_event.c

@@ -379,6 +379,7 @@ zed_disk_event_init()
 		return (-1);
 	}
 
+	pthread_setname_np(g_mon_tid, "udev monitor");
 	zed_log_msg(LOG_INFO, "zed_disk_event_init");
 
 	return (0);

cmd/zed/zed_event.c

@@ -15,7 +15,7 @@
 #include <ctype.h>
 #include <errno.h>
 #include <fcntl.h>
-#include <libzfs.h>			/* FIXME: Replace with libzfs_core. */
+#include <libzfs_core.h>
 #include <paths.h>
 #include <stdarg.h>
 #include <stdio.h>
@@ -96,6 +96,8 @@ zed_event_fini(struct zed_conf *zcp)
 		libzfs_fini(zcp->zfs_hdl);
 		zcp->zfs_hdl = NULL;
 	}
+
+	zed_exec_fini();
 }
 
 /*
@@ -953,8 +955,7 @@ zed_event_service(struct zed_conf *zcp)
 
 		_zed_event_add_time_strings(eid, zsp, etime);
 
-		zed_exec_process(eid, class, subclass,
-		    zcp->zedlet_dir, zcp->zedlets, zsp, zcp->zevent_fd);
+		zed_exec_process(eid, class, subclass, zcp, zsp);
 
 		zed_conf_write_state(zcp, eid, etime);