Support adding a spare to a degraded array.

When signalled by the monitor, the manager will find spares and
add them to the array and initiate a recovery.

4 files changed, 96 insertions, 20 deletions

diff --git a/managemon.c b/managemon.c
index 167d176..7a96f36 100644
--- a/managemon.c
+++ b/managemon.c
@@ -26,7 +26,18 @@
  * - When a device is added to the container, we add it to the metadata
  *   as a spare.
  *
- * - assist with activating spares by opening relevant sysfs file.
+ * - Deal with degraded array
+ *    We only do this when first noticing the array is degraded.
+ *    This can be when we first see the array, when sync completes or
+ *    when recovery completes.
+ *
+ *    Check if number of failed devices suggests recovery is needed, and
+ *    skip if not.
+ *    Ask metadata to allocate a spare device
+ *    Add device as not in_sync and give a role
+ *    Update metadata.
+ *    Open sysfs files and pass to monitor.
+ *    Make sure that monitor Starts recovery....
  *
  * - Pass on metadata updates from external programs such as
  *   mdadm creating a new array.
@@ -104,6 +115,32 @@ static void free_aa(struct active_array *aa)
 	free(aa);
 }
 
+static struct active_array *duplicate_aa(struct active_array *aa)
+{
+	struct active_array *newa = malloc(sizeof(*newa));
+	struct mdinfo **dp1, **dp2;
+
+	*newa = *aa;
+	newa->next = NULL;
+	newa->replaces = NULL;
+	newa->info.next = NULL;
+
+	dp2 = &newa->info.devs;
+
+	for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) {
+		struct mdinfo *d;
+		if ((*dp1)->state_fd < 0)
+			continue;
+
+		d = malloc(sizeof(*d));
+		*d = **dp1;
+		*dp2 = d;
+		dp2 = & d->next;
+	}
+
+	return newa;
+}
+
 static void write_wakeup(struct supertype *c)
 {
 	static struct md_generic_cmd cmd = { .action = md_action_ping_monitor };
@@ -171,7 +208,7 @@ void check_update_queue(struct supertype *container)
 	}
 }
 
-void queue_metadata_update(struct metadata_update *mu)
+static void queue_metadata_update(struct metadata_update *mu)
 {
 	struct metadata_update **qp;
 
@@ -198,7 +235,6 @@ static void manage_container(struct mdstat_ent *mdstat,
 	 * array ignoring any metadata on it.
 	 * FIXME should we look for compatible metadata and take hints
 	 * about spare assignment.... probably not.
-	 *
 	 */
 	if (mdstat->devcnt != container->devcnt) {
 		/* read /sys/block/NAME/md/dev-??/block/dev to find out
@@ -222,12 +258,52 @@ static void manage_member(struct mdstat_ent *mdstat,
 	 * being requested.
 	 * Unfortunately decreases in raid_disks don't show up in
 	 * mdstat until the reshape completes FIXME.
+	 *
+	 * Actually, we also want to handle degraded arrays here by
+	 * trying to find and assign a spare.
+	 * We do that whenever the monitor tells us too.
 	 */
 	// FIXME
 	a->info.array.raid_disks = mdstat->raid_disks;
 	a->info.array.chunk_size = mdstat->chunk_size;
 	// MORE
 
+	if (a->check_degraded) {
+		struct metadata_update *updates = NULL;
+		struct mdinfo *newdev;
+		struct active_array *newa;
+		wait_update_handled();
+		a->check_degraded = 0;
+
+		/* The array may not be degraded, this is just a good time
+		 * to check.
+		 */
+		newdev = a->container->ss->activate_spare(a, &updates);
+		if (newdev) {
+			struct mdinfo *d;
+			/* Cool, we can add a device or several. */
+			newa = duplicate_aa(a);
+			/* suspend recovery - maybe not needed */
+
+			/* Add device to array and set offset/size/slot.
+			 * and open files for each newdev */
+			for (d = newdev; d ; d = d->next) {
+				struct mdinfo *newd;
+				if (sysfs_add_disk(&newa->info, d))
+					continue;
+				newd = newa->info.devs;
+				newd->state_fd = sysfs_open(a->devnum,
+							    newd->sys_name,
+							    "state");
+				newd->prev_state
+					= read_dev_state(newd->state_fd);
+				newd->curr_state = newd->prev_state;
+			}
+			queue_metadata_update(updates);
+			replace_array(a->container, a, newa);
+			sysfs_set_str(&a->info, NULL, "sync_action", "repair");
+		}
+	}
 }
 
 static void manage_new(struct mdstat_ent *mdstat,
@@ -289,7 +365,7 @@ static void manage_new(struct mdstat_ent *mdstat,
 						    "state");
 
 			newd->prev_state = read_dev_state(newd->state_fd);
-			newd->curr_state = newd->curr_state;
+			newd->curr_state = newd->prev_state;
 		} else {
 			newd->state_fd = -1;
 		}
diff --git a/mdadm.h b/mdadm.h
index b97cd51..79ec5e2 100644
--- a/mdadm.h
+++ b/mdadm.h
@@ -438,6 +438,15 @@ extern struct superswitch {
 	void (*process_update)(struct supertype *st,
 			       struct metadata_update *update);
 
+	/* activate_spare will check if the array is degraded and, if it
+	 * is, try to find some spare space in the container.
+	 * On success, it add appropriate updates (For process_update) to
+	 * to the 'updates' list and returns a list of 'mdinfo' identifying
+	 * the device, or devices as there might be multiple missing
+	 * devices and multiple spares available.
+	 */
+	struct mdinfo *(*activate_spare)(struct active_array *a,
+					 struct metadata_update **updates);
 
 	int major;
 	int swapuuid; /* true if uuid is bigending rather than hostendian */
diff --git a/mdmon.h b/mdmon.h
index 965f643..11320ef 100644
--- a/mdmon.h
+++ b/mdmon.h
@@ -16,6 +16,8 @@ struct active_array {
 	enum array_state prev_state, curr_state, next_state;
 	enum sync_action prev_action, curr_action, next_action;
 
+	int check_degraded; /* flag set by mon, read by manage */
+
 	int devnum;
 
 	unsigned long long resync_start;
diff --git a/monitor.c b/monitor.c
index f46f7b1..690afa0 100644
--- a/monitor.c
+++ b/monitor.c
@@ -171,24 +171,10 @@ static void signal_manager(void)
  *    sync_action was 'resync' and becomes 'idle' and resync_start becomes
  *    MaxSector
  *    Notify metadata that sync is complete.
- *    "Deal with Degraded"
  *
  *  recovery completes
  *    sync_action changes from 'recover' to 'idle'
  *    Check each device state and mark metadata if 'faulty' or 'in_sync'.
- *    "Deal with Degraded"
- *
- *  deal with degraded array
- *    We only do this when first noticing the array is degraded.
- *    This can be when we first see the array, when sync completes or
- *    when recovery completes.
- *
- *    Check if number of failed devices suggests recovery is needed, and
- *    skip if not.
- *    Ask metadata for a spare device
- *    Add device as not in_sync and give a role
- *    Update metadata.
- *    Start recovery.
  *
  *  deal with resync
  *    This only happens on finding a new array... mdadm will have set
@@ -214,7 +200,7 @@ static void signal_manager(void)
 
 static int read_and_act(struct active_array *a)
 {
-	int check_degraded;
+	int check_degraded = 0;
 	int deactivate = 0;
 	struct mdinfo *mdi;
 
@@ -296,7 +282,9 @@ static int read_and_act(struct active_array *a)
 	}
 
 	if (check_degraded) {
-		// FIXME;
+		/* manager will do the actual check */
+		a->check_degraded = 1;
+		signal_manager();
 	}
 
 	a->container->ss->sync_metadata(a->container);
@@ -514,6 +502,7 @@ static int wait_and_act(struct supertype *container, int pfd,
 				*ap = (*ap)->next;
 			discard_this = a->replaces;
 			a->replaces = NULL;
+			/* FIXME check if device->state_fd need to be cleared?*/
 			signal_manager();
 		}
 		if (a->container)