Re: Schema variables - new implementation for Postgres 15

po 14. 11. 2022 v 8:00 odesílatel Sergey Shinderuk <
s(dot)shinderuk(at)postgrespro(dot)ru> napsal:

> On 13.11.2022 20:59, Pavel Stehule wrote:
> > fresh rebase
>
> Hello,
>
> Sorry, I haven't been following this thread, but I'd like to report a
> memory management bug. I couldn't apply the latest patches, so I tested
> with v20221104-1-* patches applied atop of commit b0284bfb1db.
>
>
> postgres=# create variable s text default 'abc';
>
> create function f() returns text as $$
> begin
> return g(s);
> end;
> $$ language plpgsql;
>
> create function g(t text) returns text as $$
> begin
> let s = 'BOOM!';
> return t;
> end;
> $$ language plpgsql;
>
> select f();
> CREATE VARIABLE
> CREATE FUNCTION
> CREATE FUNCTION
> server closed the connection unexpectedly
> This probably means the server terminated abnormally
> before or while processing the request.
>
> LOG: server process (PID 55307) was terminated by signal 11:
> Segmentation fault
> DETAIL: Failed process was running: select f();
>
>
> I believe it's a use-after-free error, triggered by assigning a new
> value to s in g(), thus making t a dangling pointer.
>
> After reconnecting I get a scary error:
>
> postgres=# select f();
> ERROR: compressed pglz data is corrupt
>

I am able to reproduce it, and I have a quick fix, but I need to
investigate i this fix will be correct

It's a good example so I have to always return a copy of value.

Regards

Pavel

>
> Best regards,
>
> --
> Sergey Shinderuk https://postgrespro.com/
>
>