Changeset d773bea9 in mainline

Timestamp:

2025-01-12T21:53:55Z (5 months ago)

Author:

Miroslav Cimerman <mc@…>

Children:

35f2a877

Parents:

dec4150

git-author:

Miroslav Cimerman <mc@…> (2025-01-12 20:12:46)

git-committer:

Miroslav Cimerman <mc@…> (2025-01-12 21:53:55)

Message:

hr: RAID1: handle state edge cases in a rebuild

File:

• uspace/srv/bd/hr/raid1.c (modified) (4 diffs)

uspace/srv/bd/hr/raid1.c

@@ -675 +675 @@
 
                 if (i + 1 >= vol->extent_no) {
-                        HR_ERROR("rebuild on \"%s\" (%lu), failed due to "
-                            "too many failed extents\n",
-                            vol->devname, vol->svc_id);
+                        if (rc != ENOMEM) {
+                                HR_ERROR("rebuild on \"%s\" (%lu), failed due "
+                                    "to too many failed extents\n",
+                                    vol->devname, vol->svc_id);
+                        }
+
+                        /* for now we have to invalidate the rebuild extent */
+                        if (rc == ENOMEM) {
+                                HR_ERROR("rebuild on \"%s\" (%lu), failed due "
+                                    "to too many failed reads, because of not "
+                                    "enough memory\n",
+                                    vol->devname, vol->svc_id);
+                                hr_raid1_ext_state_callback(vol, rebuild_idx,
+                                    ENOMEM);
+                        }
 
                         return rc;
@@ -685 +697 @@
         rc = block_write_direct(rebuild_ext->svc_id, ba, cnt, buf);
         if (rc != EOK) {
-                if (rc != ENOMEM)
-                        hr_raid1_ext_state_callback(vol, rebuild_idx, rc);
+                /*
+                 * Here we dont handle ENOMEM, because maybe in the
+                 * future, there is going to be M_WAITOK, or we are
+                 * going to wait for more memory, so that we don't
+                 * have to invalidate it...
+                 *
+                 * XXX: for now we do
+                 */
+                hr_raid1_ext_state_callback(vol, rebuild_idx, rc);
 
                 HR_ERROR("rebuild on \"%s\" (%lu), failed due to "
@@ -761 +780 @@
 
         fibril_rwlock_write_lock(&vol->states_lock);
+
         hr_update_ext_status(vol, rebuild_idx, HR_EXT_ONLINE);
+        /*
+         * We can be optimistic here, if some extents are
+         * still INVALID, FAULTY or MISSING, the update vol
+         * function will pick them up, and set the volume
+         * state accordingly.
+         */
+        hr_update_vol_status(vol, HR_VOL_ONLINE);
+        atomic_store(&vol->state_changed, true);
+
         fibril_rwlock_write_unlock(&vol->states_lock);
 
@@ -771 +800 @@
 end:
         if (rc != EOK) {
+                /*
+                 * We can fail either because:
+                 * - the rebuild extent failing or invalidation
+                 * - there is are no ONLINE extents (vol is FAULTY)
+                 * - we got ENOMEM on all READs (we also invalidate the
+                 *   rebuild extent here, for now)
+                 */
                 fibril_rwlock_write_lock(&vol->states_lock);
                 hr_update_vol_status(vol, HR_VOL_DEGRADED);
+                atomic_store(&vol->state_changed, true);
                 fibril_rwlock_write_unlock(&vol->states_lock);
         }