Changeset 6aafb48 in mainline

Timestamp:

2025-07-04T10:14:23Z (2 weeks ago)

Author:

Miroslav Cimerman <mc@…>

Children:

974f9ba

Parents:

b247c714

Message:

hr: rebuild: fix deadlock on extents_lock

Location:

uspace/srv/bd/hr

Files:

• hr.c (modified) (3 diffs)
• raid1.c (modified) (3 diffs)
• raid5.c (modified) (1 diff)
• util.c (modified) (2 diffs)

uspace/srv/bd/hr/hr.c

@@ -358 +358 @@
         }
 
-        fibril_rwlock_read_lock(&vol->extents_lock);
+        fibril_rwlock_write_lock(&vol->extents_lock);
         fibril_rwlock_write_lock(&vol->states_lock);
 
@@ -368 +368 @@
         case HR_EXT_FAILED:
                 fibril_rwlock_write_unlock(&vol->states_lock);
-                fibril_rwlock_read_unlock(&vol->extents_lock);
+                fibril_rwlock_write_unlock(&vol->extents_lock);
                 async_answer_0(icall, EINVAL);
                 return;
@@ -374 +374 @@
                 hr_update_ext_state(vol, extent_idx_to_fail, HR_EXT_FAILED);
                 (void)vol->meta_ops->erase_block(ext->svc_id);
+                block_fini(ext->svc_id);
+                ext->svc_id = 0;
                 hr_mark_vol_state_dirty(vol);
         }
 
         fibril_rwlock_write_unlock(&vol->states_lock);
-        fibril_rwlock_read_unlock(&vol->extents_lock);
+        fibril_rwlock_write_unlock(&vol->extents_lock);
 
         vol->hr_ops.vol_state_eval(vol);

uspace/srv/bd/hr/raid1.c

@@ -185 +185 @@
         size_t rebuild_no = hr_count_extents(vol, HR_EXT_REBUILD);
 
+        fibril_rwlock_read_unlock(&vol->states_lock);
+        fibril_rwlock_read_unlock(&vol->extents_lock);
+
         fibril_mutex_lock(&vol->hotspare_lock);
         size_t hs_no = vol->hotspare_no;
         fibril_mutex_unlock(&vol->hotspare_lock);
-
-        fibril_rwlock_read_unlock(&vol->states_lock);
-        fibril_rwlock_read_unlock(&vol->extents_lock);
 
         if (healthy == 0) {
@@ -206 +206 @@
                 }
 
-                if (old_state != HR_VOL_REBUILD) {
-                        if (hs_no > 0 || invalid_no > 0 || rebuild_no > 0) {
-                                fid_t fib = fibril_create(hr_raid1_rebuild,
-                                    vol);
-                                if (fib == 0)
-                                        return;
-                                fibril_start(fib);
-                                fibril_detach(fib);
-                        }
+                if (hs_no > 0 || invalid_no > 0 || rebuild_no > 0) {
+                        fid_t fib = fibril_create(hr_raid1_rebuild, vol);
+                        if (fib == 0)
+                                return;
+                        fibril_start(fib);
+                        fibril_detach(fib);
                 }
         } else {
@@ -541 +538 @@
         hr_mark_vol_state_dirty(vol);
 
+        hr_update_vol_state(vol, HR_VOL_DEGRADED);
+
         fibril_rwlock_write_unlock(&vol->states_lock);
 end:

uspace/srv/bd/hr/raid5.c

@@ -846 +846 @@
         hr_mark_vol_state_dirty(vol);
 
+        hr_update_vol_state(vol, HR_VOL_DEGRADED);
+
         fibril_rwlock_write_unlock(&vol->states_lock);
 end:

uspace/srv/bd/hr/util.c

@@ -1110 +1110 @@
 errno_t hr_init_rebuild(hr_volume_t *vol, size_t *rebuild_idx)
 {
+        HR_DEBUG("%s()", __func__);
+
         errno_t rc = EOK;
         size_t bad = vol->extent_no;
@@ -1115 +1117 @@
         if (vol->level == HR_LVL_0)
                 return EINVAL;
+
+        fibril_rwlock_read_lock(&vol->states_lock);
+        if (vol->state != HR_VOL_DEGRADED) {
+                fibril_rwlock_read_unlock(&vol->states_lock);
+                return EINVAL;
+        }
+        fibril_rwlock_read_unlock(&vol->states_lock);
 
         fibril_rwlock_write_lock(&vol->extents_lock);
         fibril_rwlock_write_lock(&vol->states_lock);
         fibril_mutex_lock(&vol->hotspare_lock);
-
-        if (vol->state != HR_VOL_DEGRADED) {
-                rc = EINVAL;
-                goto error;
-        }
 
         size_t rebuild = vol->extent_no;