Horton

Thread

Commits

  1. Fix memory leak in connection string validation.

  2. Add new predefined role pg_create_subscription.

  1. Fix minor memory leak in connection string validation

    Jeff Davis <pgsql@j-davis.com> — 2024-01-12T23:06:26Z

    Introduced in commit c3afe8cf5a.

Someone issuing repeated "CREATE SUBSCRIPTION" commands where the
connection has no password and must_have_password is true will leak
malloc'd memory in the error path. Minor issue in practice, because I
suspect that a user privileged enough to create a subscription could
cause bigger problems.

It makes me wonder if we should use the resowner mechanism to track
pointers to malloc'd memory. Then we could use a standard pattern for
these kinds of cases, and it would also catch more remote issues, like
if a pstrdup() fails in an error path (which can happen a few lines up
if the parse fails).

Patch attached; intended for 16 and 17.

Regards,
	Jeff Davis

  2. Re: Fix minor memory leak in connection string validation

    Nathan Bossart <nathandbossart@gmail.com> — 2024-01-13T02:37:05Z

    On Fri, Jan 12, 2024 at 03:06:26PM -0800, Jeff Davis wrote:
> It makes me wonder if we should use the resowner mechanism to track
> pointers to malloc'd memory. Then we could use a standard pattern for
> these kinds of cases, and it would also catch more remote issues, like
> if a pstrdup() fails in an error path (which can happen a few lines up
> if the parse fails).

That seems worth exploring.

>  		if (!uses_password)
> +		{
> +			/* malloc'd, so we must free it explicitly */
> +			PQconninfoFree(opts);
> +
>  			ereport(ERROR,
>  					(errcode(ERRCODE_S_R_E_PROHIBITED_SQL_STATEMENT_ATTEMPTED),
>  					 errmsg("password is required"),
>  					 errdetail("Non-superusers must provide a password in the connection string.")));
> +		}
>  	}
>  
>  	PQconninfoFree(opts);

Another option could be to surround this with PG_TRY/PG_FINALLY, but your
patch seems sufficient, too.

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com

  3. Re: Fix minor memory leak in connection string validation

    Tom Lane <tgl@sss.pgh.pa.us> — 2024-01-13T03:18:55Z

    Nathan Bossart <nathandbossart@gmail.com> writes:
> On Fri, Jan 12, 2024 at 03:06:26PM -0800, Jeff Davis wrote:
>> It makes me wonder if we should use the resowner mechanism to track
>> pointers to malloc'd memory. Then we could use a standard pattern for
>> these kinds of cases, and it would also catch more remote issues, like
>> if a pstrdup() fails in an error path (which can happen a few lines up
>> if the parse fails).

> That seems worth exploring.

I'm pretty dubious about adding overhead for that, mainly because
most of the direct callers of malloc in a backend are going to be
code that's not under our control.  Modifying the callers that we
do control is not going to give a full solution, and could well be
outright misleading.

> Another option could be to surround this with PG_TRY/PG_FINALLY, but your
> patch seems sufficient, too.

Yeah, seems fine for now.  If that function grows any more complexity
then we could think about using PG_TRY.

			regards, tom lane