Skip to contentSkip to conversation

improve qom debugging and fix a dangling MemoryRegion in rutabaga

PATCH Superseded Roman Kiryanov 3 participants 9 messages Apr 27 – May 11

lore
Share & tools
lore
#0 lore raw

cover improve qom debugging and fix a dangling MemoryRegion in rutabaga

Hi, this series of patches fixes a dangling MemoryRegion in rutabaga and also brings the debugging changes I used to locate it.

Roman Kiryanov (2):
qom: improve use-after-free debugging
display: rutabaga: unparent MemoryRegions in unmap

 hw/display/virtio-gpu-rutabaga.c | 1 +
 qom/object.c                     | 3 +++
 2 files changed, 4 insertions(+)
Signature · 1 line
2.54.0.rc2.544.gc7ae2d5bb8-goog
#1 lore raw

PATCH 1/2 qom: improve use-after-free debugging

This patch invalidates dead objects so their usage will lead to more predictable results (crashes or asserts).

Change-Id I1dc36dbc1030c2e41c937f08baa97b2e81d44c82
---
 qom/object.c | 3 +++
 1 file changed, 3 insertions(+)
qom/object.c +3 -0
index f981e27044..240e36c39e 100644
@@ -613,6 +613,7 @@ static void object_property_del_all(Object *obj)
     } while (released);
 
     g_hash_table_unref(obj->properties);
+    obj->properties = NULL;
 }
 
 static void object_property_del_child(Object *obj, Object *child)
@@ -670,6 +671,8 @@ static void object_finalize(void *data)
 
     g_assert(obj->ref == 0);
     g_assert(obj->parent == NULL);
+    g_assert(obj->properties == NULL);
+    obj->class = NULL;
     if (obj->free) {
         obj->free(obj);
     }
Signature · 1 line
2.54.0.rc2.544.gc7ae2d5bb8-goog
#2 lore raw

Re: [PATCH 1/2] qom: improve use-after-free debugging

Hi

Quoted 15 lines · Roman Kiryanov

This patch invalidates dead objects so their usage will lead to more predictable results (crashes or asserts).

Change-Id I1dc36dbc1030c2e41c937f08baa97b2e81d44c82
---
 qom/object.c | 3 +++
 1 file changed, 3 insertions(+)
qom/object.c +0 -0
index f981e27044..240e36c39e 100644

qom/object.c static void object_property_del_all(Object *obj) · line 616

Quoted 4 lines
qom/object.c +1 -0
@@ -613,6 +613,7 @@ static void object_property_del_all(Object *obj)
     } while (released);
 
     g_hash_table_unref(obj->properties);
+    obj->properties = NULL;

That makes sense (I'd g_clear_pointer for a safer one-liner instead)

qom/object.c static void object_finalize(void *data) · line 675

Quoted 9 lines
qom/object.c +2 -0
 }
 
 static void object_property_del_child(Object *obj, Object *child)
@@ -670,6 +671,8 @@ static void object_finalize(void *data)
 
     g_assert(obj->ref == 0);
     g_assert(obj->parent == NULL);
+    g_assert(obj->properties == NULL);
+    obj->class = NULL;

This one I am not sure is placed right. If the goal is to revert the object_initialize_with_type() effect, perhaps it should do more, in object_deinit() ?

Quoted 7 lines
qom/object.c +0 -1
     if (obj->free) {
         obj->free(obj);
     }
--

2.54.0.rc2.544.gc7ae2d5bb8-goog

Signature · 1 line
Marc-André Lureau
#3 lore raw

Re: [PATCH 1/2] qom: improve use-after-free debugging

Hi Marc, thank you for looking.

Quoted 3 lines · Marc-André Lureau

This one I am not sure is placed right. If the goal is to revert the object_initialize_with_type() effect, perhaps it should do more, in object_deinit() ?

I agree, I will send an update.

#4 lore raw

PATCH 2/2 display: rutabaga: unparent MemoryRegions in unmap

In reply to Roman Kiryanov

The virtio-gpu-rutabaga-device instance holds a hash table of child objects (Object::properties) and a memory region is added there every time memory_region_init_ram_ptr is called.

The unmap_blob call invalidates a MemoryRegion but does not remove it from the device, which makes pointers to the region dangling and eventually causes a crash when those pointers are dereferenced.

---
 hw/display/virtio-gpu-rutabaga.c | 1 +
 1 file changed, 1 insertion(+)
hw/display/virtio-gpu-rutabaga.c +1 -0
index ebb6c783fb..5cd9b19336 100644
@@ -737,6 +737,7 @@ rutabaga_cmd_resource_unmap_blob(VirtIOGPU *g,
 
         MemoryRegion *mr = &(vr->memory_regions[slot].mr);
         memory_region_del_subregion(&vb->hostmem, mr);
+        object_unparent(OBJECT(mr));
 
         vr->memory_regions[slot].resource_id = 0;
         vr->memory_regions[slot].used = 0;
Signature · 1 line
2.54.0.rc2.544.gc7ae2d5bb8-goog
#5 lore raw

v2 cover improve qom debugging and fix a dangling MemoryRegion in rutabaga

In reply to Roman Kiryanov

Hi, this series of patches fixes a dangling MemoryRegion in rutabaga and also brings the debugging changes I used to locate it.

Changes in v2:
- Updated clearing of obj->properties using g_clear_pointer.
- Moved clearing of obj->class into object_deinit.
- obj->class is checked to be NULL before obj->free(obj).

Roman Kiryanov (2):
qom: improve use-after-free debugging
display: rutabaga: unparent MemoryRegions in unmap

 hw/display/virtio-gpu-rutabaga.c | 1 +
 qom/object.c                     | 6 +++++-
 2 files changed, 6 insertions(+), 1 deletion(-)
Signature · 1 line
2.54.0.545.g6539524ca2-goog
#6 lore raw

PATCH v2 1/2 qom: improve use-after-free debugging

This patch invalidates dead objects so their usage will lead to more predictable results (crashes or asserts).

---
Changes in v2:
- Updated clearing of obj->properties using g_clear_pointer.
- Moved clearing of obj->class into object_deinit.
- obj->class is checked to be NULL before obj->free(obj).

 qom/object.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
qom/object.c +5 -1
index f981e27044..b8ffb00976 100644
@@ -612,7 +612,7 @@ static void object_property_del_all(Object *obj)
         }
     } while (released);
 
-    g_hash_table_unref(obj->properties);
+    g_clear_pointer(&obj->properties, g_hash_table_unref);
 }
 
 static void object_property_del_child(Object *obj, Object *child)
@@ -658,6 +658,9 @@ static void object_deinit(Object *obj, TypeImpl *type)
     if (type_has_parent(type)) {
         object_deinit(obj, type_get_parent(type));
     }
+
+    g_assert(obj->properties == NULL);
+    obj->class = NULL;
 }
 
 static void object_finalize(void *data)
@@ -670,6 +673,7 @@ static void object_finalize(void *data)
 
     g_assert(obj->ref == 0);
     g_assert(obj->parent == NULL);
+    g_assert(obj->class == NULL);
     if (obj->free) {
         obj->free(obj);
     }
Signature · 1 line
2.54.0.545.g6539524ca2-goog
Reviewed-by on v2 1/2
#7 lore raw

Re: [PATCH v2 1/2] qom: improve use-after-free debugging

Quoted 4 lines · Roman Kiryanov

This patch invalidates dead objects so their usage will lead to more predictable results (crashes or asserts).

Quoted 8 lines

---
Changes in v2:
- Updated clearing of obj->properties using g_clear_pointer.
- Moved clearing of obj->class into object_deinit.
- obj->class is checked to be NULL before obj->free(obj).

  qom/object.c | 6 +++++-
  1 file changed, 5 insertions(+), 1 deletion(-)
#8 lore raw

PATCH v2 2/2 display: rutabaga: unparent MemoryRegions in unmap

In reply to Roman Kiryanov

The virtio-gpu-rutabaga-device instance holds a hash table of child objects (Object::properties) and a memory region is added there every time memory_region_init_ram_ptr is called.

The unmap_blob call invalidates a MemoryRegion but does not remove it from the device, which makes pointers to the region dangling and eventually causes a crash when those pointers are dereferenced.

---
 hw/display/virtio-gpu-rutabaga.c | 1 +
 1 file changed, 1 insertion(+)
hw/display/virtio-gpu-rutabaga.c +1 -0
index ebb6c783fb..5cd9b19336 100644
@@ -737,6 +737,7 @@ rutabaga_cmd_resource_unmap_blob(VirtIOGPU *g,
 
         MemoryRegion *mr = &(vr->memory_regions[slot].mr);
         memory_region_del_subregion(&vb->hostmem, mr);
+        object_unparent(OBJECT(mr));
 
         vr->memory_regions[slot].resource_id = 0;
         vr->memory_regions[slot].used = 0;
Signature · 1 line
2.54.0.545.g6539524ca2-goog