Potential use-after-free in partion related code - Mailing list pgsql-hackers

From Andres Freund
Subject Potential use-after-free in partion related code
Date
Msg-id 20231115165737.zeulb575cgrbqo74@awork3.anarazel.de
Whole thread Raw
Responses Re: Potential use-after-free in partion related code  (Alvaro Herrera <alvherre@alvh.no-ip.org>)
List pgsql-hackers
Hi,

A while back I had proposed annotations for palloc() et al that let the
compiler know about which allocators pair with what freeing functions. One
thing that allows the compiler to do is to detect use after free.

One such complaint is:

../../../../../home/andres/src/postgresql/src/backend/commands/tablecmds.c: In function ‘ATExecAttachPartition’:
../../../../../home/andres/src/postgresql/src/backend/commands/tablecmds.c:18758:25: warning: pointer
‘partBoundConstraint’may be used after ‘list_concat’ [-Wuse-after-free]
 
18758 |                         get_proposed_default_constraint(partBoundConstraint);
      |                         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../../../../../home/andres/src/postgresql/src/backend/commands/tablecmds.c:18711:26: note: call to ‘list_concat’ here
18711 |         partConstraint = list_concat(partBoundConstraint,
      |                          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
18712 |                                                                  RelationGetPartitionQual(rel));
      |                                                                  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


And it seems quite right:

    partConstraint = list_concat(partBoundConstraint,
                                 RelationGetPartitionQual(rel));

At this point partBoundConstraint may not be used anymore, because
list_concat() might have reallocated.

But then a few lines later:

        /* we already hold a lock on the default partition */
        defaultrel = table_open(defaultPartOid, NoLock);
        defPartConstraint =
            get_proposed_default_constraint(partBoundConstraint);

We use partBoundConstraint again.

I unfortunately can't quickly enough identify what partConstraint,
defPartConstraint, partBoundConstraint are, so I don't don't really know what
the fix here is.

Greetings,

Andres Freund



pgsql-hackers by date:

Previous
From: Andres Freund
Date:
Subject: Re: Explicitly skip TAP tests under Meson if disabled
Next
From: Andres Freund
Date:
Subject: Re: remaining sql/json patches