[ruby-core:102139] [Ruby master Bug#17552] [PATCH] Fix a NULL pointer crash in ObjectSpace.dump_all

Issue #17552 has been updated by byroot (Jean Boussier).


So with some extra debug code, I've managed to identify the object that causes this, it's only one object on a multi-GiB dump:

```json
{"address":"0x7f32c8b8d6c8", "type":"IMEMO", "class":"0x8", "imemo_type":"ment", "generation":57, "memsize":48, "flags":{"wb_protected":true, "old":true, "uncollectible":true, "marked":true}}
```

I don't know wether this is expected or not, the `"class":"0x8"` is particularly surprising. 

----------------------------------------
Bug #17552: [PATCH] Fix a NULL pointer crash in ObjectSpace.dump_all
https://bugs.ruby-lang.org/issues/17552#change-89990

* Author: byroot (Jean Boussier)
* Status: Open
* Priority: Normal
* ruby -v: ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-darwin19]
* Backport: 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN
----------------------------------------
Patch: https://github.com/ruby/ruby/pull/4078

I wasn't able to reproduce the issue in isolation just yet, but I confirmed the patch fixes the issue for us.

What seem to happen in that some objects have an `allocation_info`, but `allocation_info->path == NULL`.

What is weird is that in 2.7.2, [there was no NULL check for `->path`, it was directly passed to `vfprintf`](https://github.com/ruby/ruby/blob/5445e0435260b449decf2ac16f9d09bae3cafe72/ext/objspace/objspace_dump.c#L322-L323), which from what I understand would have generated `"path": (null)`, which is invalid JSON.
So I suspect `allocation_info { path = NULL }` wasn't possible on 2.7.2?

Either way I'd like to write a test case for this, but I'm still unable to find a way to create an object with a NULL `path`.



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>