Changeset d84773a in mainline

Timestamp:

2024-10-28T21:46:47Z (7 months ago)

Author:

Miroslav Cimerman <mc@…>

Children:

cf28ffd3

Parents:

6124ee1

Message:

hr: RAID 1 state handling

Location:

uspace

Files:

• lib/device/include/hr.h (modified) (1 diff)
• lib/device/src/hr.c (modified) (1 diff)
• srv/bd/hr/raid1.c (modified) (5 diffs)

uspace/lib/device/include/hr.h

@@ -56 +56 @@
 typedef enum hr_vol_status {
         HR_VOL_ONLINE,  /* OK, OPTIMAL */
-        HR_VOL_FAULTY
+        HR_VOL_FAULTY,
+        HR_VOL_WEAKENED /* used for partial, but usable mirror */
 } hr_vol_status_t;
 

uspace/lib/device/src/hr.c

@@ -275 +275 @@
         case HR_VOL_FAULTY:
                 return "FAULTY";
+        case HR_VOL_WEAKENED:
+                return "WEAKENED";
         default:
                 return "UNKNOWN";

uspace/srv/bd/hr/raid1.c

@@ -73 +73 @@
 };
 
+static errno_t hr_raid1_check_vol_status(hr_volume_t *vol)
+{
+        if (vol->status == HR_VOL_ONLINE ||
+            vol->status == HR_VOL_WEAKENED)
+                return EOK;
+        return EINVAL;
+}
+
+/*
+ * Update vol->status and return EOK if volume
+ * is usable
+ */
+static errno_t hr_raid1_update_vol_status(hr_volume_t *vol)
+{
+        hr_vol_status_t old_state = vol->status;
+        size_t healthy = 0;
+        for (size_t i = 0; i < vol->dev_no; i++)
+                if (vol->extents[i].status == HR_EXT_ONLINE)
+                        healthy++;
+
+        if (healthy == 0) {
+                if (old_state != HR_VOL_FAULTY) {
+                        log_msg(LOG_DEFAULT, LVL_ERROR,
+                            "RAID 1 needs at least 1 extent to be ONLINE, "
+                            "marking \"%s\" (%lu) as FAULTY",
+                            vol->devname, vol->svc_id);
+                        vol->status = HR_VOL_FAULTY;
+                }
+                return EINVAL;
+        } else if (healthy < vol->dev_no) {
+                if (old_state != HR_VOL_WEAKENED) {
+                        log_msg(LOG_DEFAULT, LVL_ERROR,
+                            "RAID 1 array \"%s\" (%lu) has some inactive "
+                            "extents, marking as WEAKENED",
+                            vol->devname, vol->svc_id);
+                        vol->status = HR_VOL_WEAKENED;
+                }
+                return EOK;
+        } else {
+                if (old_state != HR_VOL_ONLINE) {
+                        log_msg(LOG_DEFAULT, LVL_ERROR,
+                            "RAID 1 array \"%s\" (%lu) has all extents active, "
+                            "marking as ONLINE",
+                            vol->devname, vol->svc_id);
+                        vol->status = HR_VOL_ONLINE;
+                }
+                return EOK;
+        }
+
+}
+
 static errno_t hr_raid1_bd_open(bd_srvs_t *bds, bd_srv_t *bd)
 {
@@ -83 +134 @@
         log_msg(LOG_DEFAULT, LVL_NOTE, "hr_bd_close()");
         return EOK;
+}
+
+static void handle_extent_error(hr_volume_t *vol, size_t extent,
+    errno_t rc)
+{
+        if (rc == ENOENT)
+                hr_update_ext_status(vol, extent, HR_EXT_MISSING);
+        else if (rc != EOK)
+                hr_update_ext_status(vol, extent, HR_EXT_FAILED);
 }
 
@@ -104 +164 @@
         fibril_mutex_lock(&vol->lock);
 
+        rc = hr_raid1_check_vol_status(vol);
+        if (rc != EOK) {
+                fibril_mutex_unlock(&vol->lock);
+                return EIO;
+        }
+
+        size_t successful = 0;
         switch (type) {
         case HR_BD_SYNC:
                 for (i = 0; i < vol->dev_no; i++) {
+                        if (vol->extents[i].status != HR_EXT_ONLINE)
+                                continue;
                         rc = block_sync_cache(vol->extents[i].svc_id, ba, cnt);
                         if (rc != EOK)
-                                goto error;
+                                handle_extent_error(vol, i, rc);
+                        else
+                                successful++;
                 }
                 break;
         case HR_BD_READ:
                 for (i = 0; i < vol->dev_no; i++) {
+                        if (vol->extents[i].status != HR_EXT_ONLINE)
+                                continue;
                         rc = block_read_direct(vol->extents[i].svc_id, ba, cnt,
                             data_read);
                         if (rc != EOK)
-                                goto error;
+                                handle_extent_error(vol, i, rc);
+                        else
+                                successful++;
                 }
                 break;
         case HR_BD_WRITE:
                 for (i = 0; i < vol->dev_no; i++) {
+                        if (vol->extents[i].status != HR_EXT_ONLINE)
+                                continue;
                         rc = block_write_direct(vol->extents[i].svc_id, ba, cnt,
                             data_write);
                         if (rc != EOK)
-                                goto error;
+                                handle_extent_error(vol, i, rc);
+                        else
+                                successful++;
                 }
                 break;
@@ -132 +211 @@
         }
 
-error:
+        if (successful > 0)
+                rc = EOK;
+        else
+                rc = EIO;
+
+        (void) hr_raid1_update_vol_status(vol);
         fibril_mutex_unlock(&vol->lock);
         return rc;
@@ -182 +266 @@
         }
 
+        rc = hr_raid1_update_vol_status(new_volume);
+        if (rc != EOK)
+                return rc;
+
         bd_srvs_init(&new_volume->hr_bds);
         new_volume->hr_bds.ops = &hr_raid1_bd_ops;