Horton

Re: PostgreSQL crashes with SIGSEGV

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Peter Geoghegan <pg@bowt.ie>
Cc: Aleksandr Parfenov <a.parfenov@postgrespro.ru>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2018-02-08T00:41:48Z
Lists: pgsql-bugs, pgsql-hackers 
Peter Geoghegan <pg@bowt.ie> writes:
> On Wed, Jan 17, 2018 at 2:23 PM, Peter Geoghegan <pg@bowt.ie> wrote:
>> A complicating factor for this fix of mine is that mode_final() seems
>> to have its own ideas about tuple memory lifetime, over and above what
>> tuplesort_getdatum() explicitly promises, as can be seen here:
>> /*
>> * Note: we *cannot* clean up the tuplesort object here, because the value
>> * to be returned is allocated inside its sortcontext.  We could use
>> * datumCopy to copy it out of there, but it doesn't seem worth the
>> * trouble, since the cleanup callback will clear the tuplesort later.
>> */

>> ISTM that either grouping sets or mode_final() must necessarily be
>> wrong, because each oversteps, and infers a different contract from
>> tuplesort tuple fetching routines (different assumptions about memory
>> contexts are made in each case). Only one can be right, unless it's
>> okay to have one rule for tuplesort_getdatum() and another for
>> tuplesort_gettupleslot() (which seems questionable to me). I still
>> think that grouping sets is right (and that mode_final() is wrong). Do
>> you?

> It would be nice to get an opinion on this mode_final() + tuplesort
> memory lifetime business from you, Tom.

I'm fairly sure that that bit in mode_final() was just a hack to make
it work.  If there's a better, more principled way, let's go for it.

> Note that you removed the quoted comment within be0ebb65, back in October.

There were multiple instances of basically that same comment before.
AFAICS I just consolidated them into one place, in the header comment for
ordered_set_shutdown.

			regards, tom lane

Commits

  1. Fix actual and potential double-frees around tuplesort usage.

  2. Allow the built-in ordered-set aggregates to share transition state.

  3. Avoid integer overflow while sifting-up a heap in tuplesort.c.

  4. Allow avoiding tuple copy within tuplesort_gettupleslot().

  5. Remove should_free arguments to tuplesort routines.

  6. Fix use-after-free around DISTINCT transition function calls.

  7. Reuse abbreviated keys in ordered [set] aggregates.