Re: BUG #19544: PostgreSQL 18.3 backend crashes with SIGSEGV when a PL/pgSQL function - Mailing list pgsql-bugs

From Ayush Tiwari
Subject Re: BUG #19544: PostgreSQL 18.3 backend crashes with SIGSEGV when a PL/pgSQL function
Date
Msg-id CAJTYsWXgcMm7YNMAvsGHmsLVJuBu3w5yjDSTXF8iHdJ=b7xn9g@mail.gmail.com
Whole thread
In response to BUG #19544: PostgreSQL 18.3 backend crashes with SIGSEGV when a PL/pgSQL function  (PG Bug reporting form <noreply@postgresql.org>)
List pgsql-bugs
Hi,

On Mon, 6 Jul 2026 at 22:21, PG Bug reporting form <noreply@postgresql.org> wrote:
The following bug has been logged on the website:

Bug reference:      19544
Logged by:          HaoGang Mao
Email address:      haogangmao@gmail.com
PostgreSQL version: 18.3
Operating system:   Linux
Description:       

A PL/pgSQL function declares a local variable of composite type foo where
attribute b is INT. Inside the function, the variable is initialized with
ROW(123, power(2, 30)). The function then changes foo.b from INT to TEXT
using ALTER TYPE and returns the previously initialized composite value.
When the caller executes SELECT bar(), PostgreSQL attempts to output the
returned record using the updated composite type metadata and the backend
segfaults in the text output path.

Reproduction steps:

```sql
DROP FUNCTION IF EXISTS bar();
DROP TYPE IF EXISTS foo CASCADE;

CREATE TYPE foo AS (a INT, b INT);

CREATE FUNCTION bar() RETURNS RECORD AS $$
DECLARE
    r foo := ROW(123, power(2, 30));
BEGIN
    ALTER TYPE foo ALTER ATTRIBUTE b TYPE TEXT;
    RETURN r;
END;
$$ LANGUAGE plpgsql;

SELECT bar();
```

Expected behavior:
PostgreSQL should reject the operation or report an error indicating that
the composite type changed while a value of the old layout is still being
returned.

Actual behavior:
The client connection is closed unexpectedly and the PostgreSQL backend
process terminates with SIGSEGV.

Server log:

```text
LOG:  client backend (PID 90) was terminated by signal 11: Segmentation
fault
DETAIL:  Failed process was running: SELECT bar();
LOG:  terminating any other active server processes
LOG:  all server processes terminated; reinitializing
```

GDB backtrace, trimmed:

```text
Program received signal SIGSEGV, Segmentation fault.
#0  0x0000762a0a47fc60 in ?? () from /lib/x86_64-linux-gnu/libc.so.6
#1  text_to_cstring(t=0x5c9ae669d6bc) at varlena.c:234
    len = 268435452
#2  textout(fcinfo=0x7fff0aabda20) at varlena.c:603
#3  FunctionCall1Coll(flinfo=0x5c9ae668c148, collation=0,
                      arg1=101820360414908) at fmgr.c:1139
#4  OutputFunctionCall(flinfo=0x5c9ae668c148,
                       val=101820360414908) at fmgr.c:1685
#5  record_out(fcinfo=0x7fff0aabdbc0) at rowtypes.c:435
    column_type = 25
    attr = 101820360414908
    tupdesc = 0x762a00b61fd8
    ncolumns = 2
    i = 1
#8  printtup(slot=0x5c9ae668bb88, self=0x5c9ae6699e18) at printtup.c:360
#9  ExecutePlan(queryDesc=0x5c9ae65c3ed0, operation=CMD_SELECT,
                sendTuples=true, numberTuples=0,
                direction=ForwardScanDirection, dest=0x5c9ae6699e18)
    at execMain.c:1742
#12 PortalRunSelect(portal=0x5c9ae6619740, forward=true,
                    count=0, dest=0x5c9ae6699e18) at pquery.c:921
#13 PortalRun(portal=0x5c9ae6619740,
              count=9223372036854775807, isTopLevel=true,
              dest=0x5c9ae6699e18, altdest=0x5c9ae6699e18,
              qc=0x7fff0aabdfe0) at pquery.c:765
#14 exec_simple_query(query_string=0x5c9ae6599b80 "SELECT bar();")
    at postgres.c:1273
#15 PostgresMain(dbname=0x5c9ae65d2f38 "postgres",
                 username=0x5c9ae65d2f20 "pguser") at postgres.c:4766
```

This looks the same as [1].

Could you please check that thread, patches are posted,
you may review those. I'll spend some time this week on that too.

Regards,
Ayush

[1] https://www.postgresql.org/message-id/19382-4c2060ffee72759b%40postgresql.org

pgsql-bugs by date:

Previous
From: PG Bug reporting form
Date:
Subject: BUG #19544: PostgreSQL 18.3 backend crashes with SIGSEGV when a PL/pgSQL function
Next
From: Robert Haas
Date:
Subject: Re: BUG #18876: HINT messages for mxid wrap-around say "drop stale slots", but that may not be appropriate