Thread: Finding memory corruption in an extension

Finding memory corruption in an extension

From
Jack Orenstein
Date:
An extension I'm creating is causing Postgres to crash, almost certainly due to memory corruption.  I am using palloc0/pfree, calling SET_VARSIZE, and generally following the procedures documented here: https://www.postgresql.org/docs/12/xfunc-c.html. I am also testing my code outside of Postgres (using alloc/free instead of palloc0/pfree), and valgrind is not finding any corruption or leaks.

The crash is not completely reproducible, but when it does happen, it's pretty fast -- create a table, insert a couple of rows, explain a query. (My goal is to create a GIN index on my datatype, but this crash occurs without the index.)

I'm interested in advice on how to go about hunting down my problem. Something along the lines of a debugging malloc, or valgrind, for Postgres.

Jack Orenstein

Re: Finding memory corruption in an extension

From
Pavel Stehule
Date:
Hi

pá 8. 1. 2021 v 18:48 odesílatel Jack Orenstein <jao@geophile.com> napsal:
An extension I'm creating is causing Postgres to crash, almost certainly due to memory corruption.  I am using palloc0/pfree, calling SET_VARSIZE, and generally following the procedures documented here: https://www.postgresql.org/docs/12/xfunc-c.html. I am also testing my code outside of Postgres (using alloc/free instead of palloc0/pfree), and valgrind is not finding any corruption or leaks.

The crash is not completely reproducible, but when it does happen, it's pretty fast -- create a table, insert a couple of rows, explain a query. (My goal is to create a GIN index on my datatype, but this crash occurs without the index.)

I'm interested in advice on how to go about hunting down my problem. Something along the lines of a debugging malloc, or valgrind, for Postgres.

The basic feature is using postgres compiled with --enable-cassert flag. It does lot of checks of memory corruptions

Regards

Pavel




Jack Orenstein

Re: Finding memory corruption in an extension

From
Tom Lane
Date:
Pavel Stehule <pavel.stehule@gmail.com> writes:
> pá 8. 1. 2021 v 18:48 odesílatel Jack Orenstein <jao@geophile.com> napsal:
>> I'm interested in advice on how to go about hunting down my problem.
>> Something along the lines of a debugging malloc, or valgrind, for Postgres.

> The basic feature is using postgres compiled with --enable-cassert flag. It
> does lot of checks of memory corruptions

Yeah, you should absolutely use --enable-cassert when working on C code.

If you need valgrind, there's advice about how to run it at

https://wiki.postgresql.org/wiki/Valgrind

            regards, tom lane