mdmon: fix removed disk handling

Use SKIP_GONE_DEVS when reading the container, and correct some confused
logic in manage_new().

Signed-off-by: Dan Williams <dan.j.williams@intel.com>

2 files changed, 11 insertions, 8 deletions

diff --git a/managemon.c b/managemon.c
index c9b054ff..c48c8517 100644
--- a/managemon.c
+++ b/managemon.c
@@ -283,7 +283,7 @@ static void manage_container(struct mdstat_ent *mdstat,
 		 * To see what is removed and what is added.
 		 * These need to be remove from, or added to, the array
 		 */
-		mdi = sysfs_read(-1, mdstat->devnum, GET_DEVS);
+		mdi = sysfs_read(-1, mdstat->devnum, GET_DEVS|SKIP_GONE_DEVS);
 		if (!mdi) {
 			/* invalidate the current count so we can try again */
 			container->devcnt = -1;
@@ -460,7 +460,7 @@ static void manage_new(struct mdstat_ent *mdstat,
 			if (i == di->disk.raid_disk)
 				break;
 
-		if (di) {
+		if (di && newd) {
 			memcpy(newd, di, sizeof(*newd));
 
 			newd->state_fd = sysfs_open(new->devnum,
@@ -469,13 +469,16 @@ static void manage_new(struct mdstat_ent *mdstat,
 
 			newd->prev_state = read_dev_state(newd->state_fd);
 			newd->curr_state = newd->prev_state;
-		} else if (failed + 1 > new->info.array.failed_disks) {
-			/* we cannot properly monitor without all working disks */
-			new->container = NULL;
-			break;
 		} else {
+			if (newd)
+				free(newd);
+
 			failed++;
-			free(newd);
+			if (failed > new->info.array.failed_disks) {
+				/* we cannot properly monitor without all working disks */
+				new->container = NULL;
+				break;
+			}
 			continue;
 		}
 		sprintf(newd->sys_name, "rd%d", i);
diff --git a/mdmon.c b/mdmon.c
index acb36a2d..7dc9120b 100644
--- a/mdmon.c
+++ b/mdmon.c
@@ -400,7 +400,7 @@ int mdmon(char *devname, int devnum, int scan, char *switchroot)
 	}
 
 	mdi = sysfs_read(mdfd, container->devnum,
-			 GET_VERSION|GET_LEVEL|GET_DEVS);
+			 GET_VERSION|GET_LEVEL|GET_DEVS|SKIP_GONE_DEVS);
 
 	if (!mdi) {
 		fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",