Thread: libpq endless loop if client_min_messages=debug1

libpq endless loop if client_min_messages=debug1

From
Andreas Pflug
Date:
My Deja wrote:

>
>> I am trying to get some query trees to appear in the PostgreSQL log
>> and in order to that I have set
>> client_min_messages = DEBUG1  in order to use the following settings
>> debug_print_parse, debug_print_rewritten, or debug_print_plan which
>> are required for the query tree to show up in the log.
>>
>> pgAdmin crashes whenever I set that option. I tried a few times and I
>> am sure of it.
>> I am using PostgreSQL 7.4 under Cygwin on a Windows 2000 machine, but
>> I don't think that has any relevance.
>
I reproduced this problem with 7.5 head backend and libpq under win32
and Linux, and found that the problem is pqParseInput3 expecting a
message length >= 30000 only for message types 'T', 'D' and 'd', but not
'N'. In the case tested above, the message will be 49336 bytes long,
causing an endless loop in PQexecFinish because PQgetResult will deliver
the same broken message forever.

The attached patch fixes this. I wonder if there are additional message
types that might be longer?

Regards,
Andreas
Index: fe-protocol3.c
===================================================================
RCS file: /projects/cvsroot/pgsql-server/src/interfaces/libpq/fe-protocol3.c,v
retrieving revision 1.10
diff -u -r1.10 fe-protocol3.c
--- fe-protocol3.c    29 Nov 2003 19:52:12 -0000    1.10
+++ fe-protocol3.c    28 Dec 2003 11:26:09 -0000
@@ -84,7 +84,7 @@
             return;
         }
         if (msgLength > 30000 &&
-            !(id == 'T' || id == 'D' || id == 'd'))
+            !(id == 'N' || id == 'T' || id == 'D' || id == 'd'))
         {
             handleSyncLoss(conn, id, msgLength);
             return;



Re: [PATCHES] libpq endless loop if client_min_messages=debug1

From
Tom Lane
Date:
Andreas Pflug <pgadmin@pse-consulting.de> writes:
> I reproduced this problem with 7.5 head backend and libpq under win32
> and Linux, and found that the problem is pqParseInput3 expecting a
> message length >= 30000 only for message types 'T', 'D' and 'd', but not
> 'N'.

Oops.  Good catch.

            regards, tom lane

Re: [PATCHES] libpq endless loop if client_min_messages=debug1

From
Tom Lane
Date:
Andreas Pflug <pgadmin@pse-consulting.de> writes:
> The attached patch fixes this. I wonder if there are additional message
> types that might be longer?

We should allow error messages to be long too, I guess, and I also added
NOTIFY messages since there's no telling what people might try to do
with the data payload string once we implement that.

I also fixed the underlying infinite-loop problem.

            regards, tom lane