On Sun, Mar 21, 2021 at 8:54 PM Amit Kapila <amit.kapila16@gmail.com> wrote:
>
> On Sat, Mar 20, 2021 at 12:54 PM Peter Smith <smithpb2250@gmail.com> wrote:
> >
> > PSA my patch to correct this by firstly doing a HASH_FIND, then only
> > HASH_REMOVE after we've finished using the ent.
> >
>
> Why can't we keep using HASH_REMOVE as it is but get the output (entry
> found or not) in the last parameter of hash_search API and then
> perform Assert based on that? See similar usage in reorderbuffer.c and
> rewriteheap.c.
>
Changing the Assert doesn't do anything to fix the problem as
described, i.e. dereferencing of ent after the HASH_REMOVE.
The real problem isn't the Assert. It's all those other usages of ent
disobeying the API rule: "(NB: in the case of the REMOVE action, the
result is a dangling pointer that shouldn't be dereferenced!)"
e.g.
- SharedFileSetDeleteAll(ent->stream_fileset);
- pfree(ent->stream_fileset);
- ent->stream_fileset = NULL;
- if (ent->subxact_fileset)
- SharedFileSetDeleteAll(ent->subxact_fileset);
- pfree(ent->subxact_fileset);
- ent->subxact_fileset = NULL;
------
Kind Regards,
Peter Smith
Fujitsu Australia.
