Fix NVMe hot-swap LED stuck in FAILURE under VMD controllers

doc/ledmon.conf.pod

@@ -63,6 +63,14 @@ Value also can be set by integer number - 0 means 'quiet' and 5 means 'all'.
 B<LOG_PATH> - Sets a path to local log file. If this option is specified the
 global log file F</var/log/ledmon.log> is not used.
 
+B<BLINK_PERSISTENT_FAIL_ON_READD> - Controls whether the failure LED remains
+on when a non-RAID drive is re-added after removal. If value is set to true,
+the failure indication persists until the drive is manually addressed. If value
+is set to false, non-RAID drives that reappear in the sysfs scan will
+automatically recover from the failure state and the LED will be cleared. RAID
+member drives always retain persistent failure regardless of this setting. The
+default value is true.
+
 B<RAID_MEMBERS_ONLY> - If flag is set to true ledmon will limit monitoring only
 to drives that are RAID members. The default value is false.
 

src/common/config_file.c

@@ -216,6 +216,11 @@ static int parse_next(FILE *fd, struct ledmon_conf *conf)
 		conf->raid_members_only = parse_bool(s);
 		if (conf->raid_members_only < 0)
 			return -1;
+	} else if (!strncmp(s, "BLINK_PERSISTENT_FAIL_ON_READD=", 31)) {
+		s += 31;
+		conf->blink_persistent_fail_on_readd = parse_bool(s);
+		if (conf->blink_persistent_fail_on_readd < 0)
+			return -1;
 	} else if (_parse_and_add_to_list(s, WHITELIST, WHITELIST_LEN, &conf->cntrls_allowlist)) {
 		/* Deprecated, provided for backwards compatibility */
 		return 0;
@@ -327,6 +332,9 @@ int ledmon_write_shared_conf(struct ledmon_conf *conf)
 		 "RAID_MEMBERS_ONLY=%d\n", conf->raid_members_only);
 	snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 		 "REBUILD_BLINK_ON_ALL=%d\n", conf->rebuild_blink_on_all);
+	snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
+		 "BLINK_PERSISTENT_FAIL_ON_READD=%d\n",
+		 conf->blink_persistent_fail_on_readd);
 	snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 		 "INTERVAL=%d\n", conf->scan_interval);
 	allowlist = conf_list_to_str(&conf->cntrls_allowlist);

src/common/config_file.h

@@ -35,6 +35,7 @@ struct ledmon_conf {
 	int blink_on_init;
 	int rebuild_blink_on_all;
 	int raid_members_only;
+	int blink_persistent_fail_on_readd;
 
 	/* allowlist and excludelist of controllers for blinking */
 	struct list cntrls_allowlist;

src/ledmon/ledmon.c

@@ -670,6 +670,10 @@ static void _add_block(struct block_device *block)
 			} else {
 				temp->ibpi = block->ibpi;
 			}
+		} else if (temp->ibpi == LED_IBPI_PATTERN_FAILED_DRIVE &&
+			   !temp->raid_dev &&
+			   !conf.blink_persistent_fail_on_readd) {
+			temp->ibpi = LED_IBPI_PATTERN_ADDED;
 		} else if (!(temp->ibpi == LED_IBPI_PATTERN_FAILED_DRIVE &&
 			block->ibpi == LED_IBPI_PATTERN_HOTSPARE) ||
 			(temp->ibpi == LED_IBPI_PATTERN_FAILED_DRIVE &&
@@ -884,6 +888,7 @@ static ledmon_status_code_t _init_ledmon_conf(void)
 	conf.blink_on_migration = 1;
 	conf.rebuild_blink_on_all = 0;
 	conf.raid_members_only = 0;
+	conf.blink_persistent_fail_on_readd = 1;
 	conf.scan_interval = LEDMON_DEF_SLEEP_INTERVAL;
 	return rc;
 }

src/ledmon/udev.c

@@ -27,24 +27,40 @@ static struct udev_monitor *udev_monitor;
 
 static int _compare(const struct block_device *bd, const char *syspath, struct led_ctx *ctx)
 {
+	struct block_device *bd_new;
+	const char *udev_name;
+	const char *dev_name;
+	int ret;
+
 	if (!bd || !syspath)
 		return 0;
 
-	if (strcmp(bd->sysfs_path, syspath) == 0) {
+	if (strcmp(bd->sysfs_path, syspath) == 0)
 		return 1;
-	} else {
-		struct block_device *bd_new;
-		int ret;
-
-		bd_new = block_device_init(sysfs_get_cntrl_devices(ctx), syspath);
-		if (!bd_new)
-			return 0;
 
+	bd_new = block_device_init(sysfs_get_cntrl_devices(ctx), syspath);
+	if (bd_new) {
 		ret = block_compare(bd, bd_new);
 		block_device_fini(bd_new);
-
-		return ret;
+		if (ret)
+			return 1;
 	}
+
+	/*
+	 * NVMe udev events may arrive with a virtual nvme-subsystem path
+	 * that cannot be resolved to a PCI controller. Fall back to matching
+	 * by device name so that add/remove events are not silently dropped.
+	 */
+	if (bd->devnode[0] == '\0')
+		return 0;
+
+	dev_name = strrchr(bd->devnode, '/');
+	dev_name = dev_name ? dev_name + 1 : bd->devnode;
+
+	udev_name = strrchr(syspath, '/');
+	udev_name = udev_name ? udev_name + 1 : syspath;
+
+	return strcmp(dev_name, udev_name) == 0;
 }
 
 static int create_udev_monitor(void)

src/lib/pci_slot.c

@@ -5,6 +5,7 @@
 #include <stdint.h>
 #include <stdlib.h>
 #include <string.h>
+#include <sys/stat.h>
 #include <unistd.h>
 #include "config.h"
 #include "led/libled.h"
@@ -71,6 +72,12 @@ struct slot_property *pci_slot_property_init(struct pci_slot *pci_slot)
 
 	result->bl_device = get_block_device_from_sysfs_path(pci_slot->ctx,
 							     pci_slot->address, true);
+	if (result->bl_device) {
+		struct stat st;
+
+		if (stat(result->bl_device->devnode, &st) != 0)
+			result->bl_device = NULL;
+	}
 	result->slot_spec.pci = pci_slot;
 	snprintf(result->slot_id, PATH_MAX, "%s", pci_slot->sysfs_path);
 	result->c = &pci_slot_common;