[Backport] Security bug 1161847

Partial cherry-pick of patch originally reviewed on
https://chromium-review.googlesource.com/c/v8/v8/+/2748077:
Merged: Squashed multiple commits.

Merged: [const-tracking] Mark const field as mutable when reconfiguring
Revision: 7535b91f7cb22274de734d5da7d0324d8653d626

Merged: [const-tracking] Fix incorrect DCHECK in MapUpdater
Revision: f95db8916a731e6e5ccc0282616bc907ce06012f

BUG=chromium:1161847,chromium:1185463,v8:9233
NOTRY=true
NOPRESUBMIT=true
NOTREECHECKS=true
[email protected]

Change-Id: I4a34bafb3b072f2e788b47949947c76110f1b85c
Reviewed-by: Igor Sheludko <[email protected]>
Commit-Queue: Georg Neis <[email protected]>
Cr-Commit-Position: refs/branch-heads/9.0@{#18}
Cr-Branched-From: bd0108b4c88e0d6f2350cb79b5f363fbd02f3eb7-refs/heads/9.0.257@{qt#1}
Cr-Branched-From: 349bcc6a075411f1a7ce2d866c3dfeefc2efa39d-refs/heads/master@{#73001}
Reviewed-by: Jüri Valdmann <[email protected]>

Author: GeorgNeis

chromium/v8/src/map-updater.cc

@@ -99,6 +99,41 @@ Handle<Map> MapUpdater::ReconfigureToDataField(int descriptor,
   PropertyDetails old_details =
       old_descriptors_->GetDetails(modified_descriptor_);
 
+  // If the {descriptor} was "const" data field so far, we need to update the
+  // {old_map_} here, otherwise we could get the constants wrong, i.e.
+  //
+  //   o.x = 1;
+  //   change o.x's attributes to something else
+  //   delete o.x;
+  //   o.x = 2;
+  //
+  // could trick V8 into thinking that `o.x` is still 1 even after the second
+  // assignment.
+  // This situation is similar to what might happen with property deletion.
+  if (old_details.constness() == PropertyConstness::kConst &&
+      old_details.location() == kField &&
+      old_details.attributes() != new_attributes_) {
+    Handle<FieldType> field_type(
+        old_descriptors_->GetFieldType(modified_descriptor_), isolate_);
+    Map::GeneralizeField(isolate_, old_map_, descriptor,
+                         PropertyConstness::kMutable,
+                         old_details.representation(), field_type);
+    // The old_map_'s property must become mutable.
+    // Note, that the {old_map_} and {old_descriptors_} are not expected to be
+    // updated by the generalization if the map is already deprecated.
+    DCHECK_IMPLIES(
+        !old_map_->is_deprecated(),
+        PropertyConstness::kMutable ==
+            old_descriptors_->GetDetails(modified_descriptor_).constness());
+    // Although the property in the old map is marked as mutable we still
+    // treat it as constant when merging with the new path in transition tree.
+    // This is fine because up until this reconfiguration the field was
+    // known to be constant, so it's fair to proceed treating it as such
+    // during this reconfiguration session. The issue is that after the
+    // reconfiguration the original field might become mutable (see the delete
+    // example above).
+  }
+
   // If property kind is not reconfigured merge the result with
   // representation/field type from the old descriptor.
   if (old_details.kind() == new_kind_) {