1999-07-22 17:09:15 +02:00
<!--
2010-09-20 22:08:53 +02:00
doc/src/sgml/ref/create_function.sgml
2020-05-15 14:52:24 +02:00
PostgreSQL documentation
1999-07-22 17:09:15 +02:00
-->
2017-10-20 03:16:39 +02:00
<refentry id="sql-createfunction">
2014-02-24 03:25:35 +01:00
<indexterm zone="sql-createfunction">
<primary>CREATE FUNCTION</primary>
</indexterm>
1999-06-14 09:37:05 +02:00
<refmeta>
2010-04-03 09:23:02 +02:00
<refentrytitle>CREATE FUNCTION</refentrytitle>
2008-11-14 11:22:48 +01:00
<manvolnum>7</manvolnum>
1999-06-14 09:37:05 +02:00
<refmiscinfo>SQL - Language Statements</refmiscinfo>
</refmeta>
2001-05-19 11:01:10 +02:00
1999-06-14 09:37:05 +02:00
<refnamediv>
2001-05-19 11:01:10 +02:00
<refname>CREATE FUNCTION</refname>
2001-09-03 14:57:50 +02:00
<refpurpose>define a new function</refpurpose>
1998-12-29 03:24:47 +01:00
</refnamediv>
2001-05-19 11:01:10 +02:00
1999-06-14 09:37:05 +02:00
<refsynopsisdiv>
2001-05-19 11:01:10 +02:00
<synopsis>
2005-04-01 00:46:33 +02:00
CREATE [ OR REPLACE ] FUNCTION
2009-09-19 12:23:27 +02:00
<replaceable class="parameter">name</replaceable> ( [ [ <replaceable class="parameter">argmode</replaceable> ] [ <replaceable class="parameter">argname</replaceable> ] <replaceable class="parameter">argtype</replaceable> [ { DEFAULT | = } <replaceable class="parameter">default_expr</replaceable> ] [, ...] ] )
2008-07-18 05:32:53 +02:00
[ RETURNS <replaceable class="parameter">rettype</replaceable>
2009-09-19 12:23:27 +02:00
| RETURNS TABLE ( <replaceable class="parameter">column_name</replaceable> <replaceable class="parameter">column_type</replaceable> [, ...] ) ]
{ LANGUAGE <replaceable class="parameter">lang_name</replaceable>
2015-04-26 16:33:14 +02:00
| TRANSFORM { FOR TYPE <replaceable class="parameter">type_name</replaceable> } [, ... ]
2008-12-31 03:25:06 +01:00
| WINDOW
2021-06-23 20:27:13 +02:00
| { IMMUTABLE | STABLE | VOLATILE }
| [ NOT ] LEAKPROOF
| { CALLED ON NULL INPUT | RETURNS NULL ON NULL INPUT | STRICT }
| { [ EXTERNAL ] SECURITY INVOKER | [ EXTERNAL ] SECURITY DEFINER }
2015-09-16 21:38:47 +02:00
| PARALLEL { UNSAFE | RESTRICTED | SAFE }
2007-01-22 02:35:23 +01:00
| COST <replaceable class="parameter">execution_cost</replaceable>
| ROWS <replaceable class="parameter">result_rows</replaceable>
2019-02-10 00:08:48 +01:00
| SUPPORT <replaceable class="parameter">support_function</replaceable>
2007-09-03 20:46:30 +02:00
| SET <replaceable class="parameter">configuration_parameter</replaceable> { TO <replaceable class="parameter">value</replaceable> | = <replaceable class="parameter">value</replaceable> | FROM CURRENT }
2002-05-17 20:32:52 +02:00
| AS '<replaceable class="parameter">definition</replaceable>'
| AS '<replaceable class="parameter">obj_file</replaceable>', '<replaceable class="parameter">link_symbol</replaceable>'
2021-04-07 21:30:08 +02:00
| <replaceable class="parameter">sql_body</replaceable>
2002-05-17 20:32:52 +02:00
} ...
2001-05-19 11:01:10 +02:00
</synopsis>
</refsynopsisdiv>
2008-12-18 19:20:35 +01:00
2001-05-19 11:01:10 +02:00
<refsect1 id="sql-createfunction-description">
<title>Description</title>
1999-07-06 19:16:42 +02:00
2001-05-19 11:01:10 +02:00
<para>
<command>CREATE FUNCTION</command> defines a new function.
2002-05-17 20:32:52 +02:00
<command>CREATE OR REPLACE FUNCTION</command> will either create a
new function, or replace an existing definition.
2010-03-03 04:14:08 +01:00
To be able to define a function, the user must have the
<literal>USAGE</literal> privilege on the language.
</para>
2003-04-22 12:08:08 +02:00
<para>
If a schema name is included, then the function is created in the
specified schema. Otherwise it is created in the current schema.
2017-11-30 14:46:13 +01:00
The name of the new function must not match any existing function or procedure
2008-07-18 05:32:53 +02:00
with the same input argument types in the same schema. However,
2017-11-30 14:46:13 +01:00
functions and procedures of different argument types can share a name (this is
2017-10-09 03:44:17 +02:00
called <firstterm>overloading</firstterm>).
2003-04-22 12:08:08 +02:00
</para>
<para>
2007-01-22 02:35:23 +01:00
To replace the current definition of an existing function, use
2003-04-22 12:08:08 +02:00
<command>CREATE OR REPLACE FUNCTION</command>. It is not possible
to change the name or argument types of a function this way (if you
2004-05-17 01:22:08 +02:00
tried, you would actually be creating a new, distinct function).
Also, <command>CREATE OR REPLACE FUNCTION</command> will not let
you change the return type of an existing function. To do that,
2017-10-09 03:44:17 +02:00
you must drop and recreate the function. (When using <literal>OUT</literal>
2009-10-08 04:39:25 +02:00
parameters, that means you cannot change the types of any
2017-10-09 03:44:17 +02:00
<literal>OUT</literal> parameters except by dropping the function.)
2003-04-22 12:08:08 +02:00
</para>
2010-03-03 04:14:08 +01:00
<para>
2017-10-09 03:44:17 +02:00
When <command>CREATE OR REPLACE FUNCTION</command> is used to replace an
2010-03-03 04:14:08 +01:00
existing function, the ownership and permissions of the function
do not change. All other function properties are assigned the
values specified or implied in the command. You must own the function
to replace it (this includes being a member of the owning role).
</para>
2003-04-22 12:08:08 +02:00
<para>
If you drop and then recreate a function, the new function is not
2005-01-04 01:39:53 +01:00
the same entity as the old; you will have to drop existing rules, views,
triggers, etc. that refer to the old function. Use
2003-04-22 12:08:08 +02:00
<command>CREATE OR REPLACE FUNCTION</command> to change a function
definition without breaking objects that refer to the function.
2017-10-09 03:44:17 +02:00
Also, <command>ALTER FUNCTION</command> can be used to change most of the
2007-09-03 02:39:26 +02:00
auxiliary properties of an existing function.
2003-04-22 12:08:08 +02:00
</para>
2002-05-17 20:32:52 +02:00
<para>
The user that creates the function becomes the owner of the function.
</para>
2011-12-19 23:05:19 +01:00
<para>
To be able to create a function, you must have <literal>USAGE</literal>
privilege on the argument types and the return type.
</para>
2021-03-10 17:33:50 +01:00
<para>
Refer to <xref linkend="xfunc"/> for further information on writing
functions.
</para>
2003-04-22 12:08:08 +02:00
</refsect1>
<refsect1>
<title>Parameters</title>
1999-07-06 19:16:42 +02:00
2001-05-19 11:01:10 +02:00
<variablelist>
1999-07-06 19:16:42 +02:00
2001-05-19 11:01:10 +02:00
<varlistentry>
<term><replaceable class="parameter">name</replaceable></term>
<listitem>
<para>
2004-09-16 06:16:08 +02:00
The name (optionally schema-qualified) of the function to create.
2001-05-19 11:01:10 +02:00
</para>
</listitem>
</varlistentry>
2005-04-01 00:46:33 +02:00
<varlistentry>
<term><replaceable class="parameter">argmode</replaceable></term>
<listitem>
<para>
2017-10-09 03:44:17 +02:00
The mode of an argument: <literal>IN</literal>, <literal>OUT</literal>,
<literal>INOUT</literal>, or <literal>VARIADIC</literal>.
If omitted, the default is <literal>IN</literal>.
Only <literal>OUT</literal> arguments can follow a <literal>VARIADIC</literal> one.
Also, <literal>OUT</literal> and <literal>INOUT</literal> arguments cannot be used
together with the <literal>RETURNS TABLE</literal> notation.
2005-04-01 00:46:33 +02:00
</para>
</listitem>
</varlistentry>
2004-06-25 09:05:34 +02:00
<varlistentry>
<term><replaceable class="parameter">argname</replaceable></term>
<listitem>
<para>
2013-06-20 04:25:13 +02:00
The name of an argument. Some languages (including SQL and PL/pgSQL)
2013-05-11 18:07:47 +02:00
let you use the name in the function body. For other languages the
2009-10-08 04:39:25 +02:00
name of an input argument is just extra documentation, so far as
the function itself is concerned; but you can use input argument names
when calling a function to improve readability (see <xref
2017-11-23 15:39:47 +01:00
linkend="sql-syntax-calling-funcs"/>). In any case, the name
2009-10-08 04:39:25 +02:00
of an output argument is significant, because it defines the column
2005-04-01 00:46:33 +02:00
name in the result row type. (If you omit the name for an output
argument, the system will choose a default column name.)
2004-06-25 09:05:34 +02:00
</para>
</listitem>
</varlistentry>
2001-05-19 11:01:10 +02:00
<varlistentry>
<term><replaceable class="parameter">argtype</replaceable></term>
<listitem>
<para>
2008-12-18 19:20:35 +01:00
The data type(s) of the function's arguments (optionally
Update reference documentation on may/can/might:
Standard English uses "may", "can", and "might" in different ways:
may - permission, "You may borrow my rake."
can - ability, "I can lift that log."
might - possibility, "It might rain today."
Unfortunately, in conversational English, their use is often mixed, as
in, "You may use this variable to do X", when in fact, "can" is a better
choice. Similarly, "It may crash" is better stated, "It might crash".
2007-02-01 00:26:05 +01:00
schema-qualified), if any. The argument types can be base, composite,
or domain types, or can reference the type of a table column.
2003-03-20 05:41:13 +01:00
</para>
<para>
Update reference documentation on may/can/might:
Standard English uses "may", "can", and "might" in different ways:
may - permission, "You may borrow my rake."
can - ability, "I can lift that log."
might - possibility, "It might rain today."
Unfortunately, in conversational English, their use is often mixed, as
in, "You may use this variable to do X", when in fact, "can" is a better
choice. Similarly, "It may crash" is better stated, "It might crash".
2007-02-01 00:26:05 +01:00
Depending on the implementation language it might also be allowed
2017-10-09 03:44:17 +02:00
to specify <quote>pseudo-types</quote> such as <type>cstring</type>.
2017-01-25 15:27:09 +01:00
Pseudo-types indicate that the actual argument type is either
2003-03-20 05:41:13 +01:00
incompletely specified, or outside the set of ordinary SQL data types.
2001-05-19 11:01:10 +02:00
</para>
2005-01-04 01:39:53 +01:00
<para>
The type of a column is referenced by writing
<literal><replaceable
2009-09-19 12:23:27 +02:00
class="parameter">table_name</replaceable>.<replaceable
class="parameter">column_name</replaceable>%TYPE</literal>.
2005-01-04 01:39:53 +01:00
Using this feature can sometimes help make a function independent of
changes to the definition of a table.
</para>
2001-05-19 11:01:10 +02:00
</listitem>
</varlistentry>
2008-12-04 18:51:28 +01:00
<varlistentry>
2009-09-19 12:23:27 +02:00
<term><replaceable class="parameter">default_expr</replaceable></term>
2008-12-04 18:51:28 +01:00
<listitem>
<para>
An expression to be used as default value if the parameter is
2008-12-18 19:20:35 +01:00
not specified. The expression has to be coercible to the
argument type of the parameter.
2017-10-09 03:44:17 +02:00
Only input (including <literal>INOUT</literal>) parameters can have a default
2008-12-18 19:20:35 +01:00
value. All input parameters following a
parameter with a default value must have default values as well.
2008-12-04 18:51:28 +01:00
</para>
</listitem>
</varlistentry>
2001-05-19 11:01:10 +02:00
<varlistentry>
<term><replaceable class="parameter">rettype</replaceable></term>
<listitem>
<para>
2008-12-18 19:20:35 +01:00
The return data type (optionally schema-qualified). The return type
Update reference documentation on may/can/might:
Standard English uses "may", "can", and "might" in different ways:
may - permission, "You may borrow my rake."
can - ability, "I can lift that log."
might - possibility, "It might rain today."
Unfortunately, in conversational English, their use is often mixed, as
in, "You may use this variable to do X", when in fact, "can" is a better
choice. Similarly, "It may crash" is better stated, "It might crash".
2007-02-01 00:26:05 +01:00
can be a base, composite, or domain type,
or can reference the type of a table column.
Depending on the implementation language it might also be allowed
2017-10-09 03:44:17 +02:00
to specify <quote>pseudo-types</quote> such as <type>cstring</type>.
2006-11-03 08:17:36 +01:00
If the function is not supposed to return a value, specify
2017-10-09 03:44:17 +02:00
<type>void</type> as the return type.
2005-01-04 01:39:53 +01:00
</para>
2005-04-01 00:46:33 +02:00
<para>
2017-10-09 03:44:17 +02:00
When there are <literal>OUT</literal> or <literal>INOUT</literal> parameters,
the <literal>RETURNS</literal> clause can be omitted. If present, it
2005-04-01 00:46:33 +02:00
must agree with the result type implied by the output parameters:
2017-10-09 03:44:17 +02:00
<literal>RECORD</literal> if there are multiple output parameters, or
2005-04-01 00:46:33 +02:00
the same type as the single output parameter.
</para>
2005-01-04 01:39:53 +01:00
<para>
2003-04-22 12:08:08 +02:00
The <literal>SETOF</literal>
2001-05-19 11:01:10 +02:00
modifier indicates that the function will return a set of
2002-08-22 02:01:51 +02:00
items, rather than a single item.
2001-05-19 11:01:10 +02:00
</para>
2005-01-04 01:39:53 +01:00
<para>
The type of a column is referenced by writing
<literal><replaceable
2009-09-19 12:23:27 +02:00
class="parameter">table_name</replaceable>.<replaceable
class="parameter">column_name</replaceable>%TYPE</literal>.
2005-01-04 01:39:53 +01:00
</para>
2001-05-19 11:01:10 +02:00
</listitem>
</varlistentry>
2008-07-18 05:32:53 +02:00
<varlistentry>
2009-09-19 12:23:27 +02:00
<term><replaceable class="parameter">column_name</replaceable></term>
2008-07-18 05:32:53 +02:00
<listitem>
<para>
2017-10-09 03:44:17 +02:00
The name of an output column in the <literal>RETURNS TABLE</literal>
2008-07-18 05:32:53 +02:00
syntax. This is effectively another way of declaring a named
2017-10-09 03:44:17 +02:00
<literal>OUT</literal> parameter, except that <literal>RETURNS TABLE</literal>
also implies <literal>RETURNS SETOF</literal>.
2008-07-18 05:32:53 +02:00
</para>
</listitem>
</varlistentry>
<varlistentry>
2009-09-19 12:23:27 +02:00
<term><replaceable class="parameter">column_type</replaceable></term>
2008-07-18 05:32:53 +02:00
<listitem>
<para>
2017-10-09 03:44:17 +02:00
The data type of an output column in the <literal>RETURNS TABLE</literal>
2008-07-18 05:32:53 +02:00
syntax.
</para>
</listitem>
</varlistentry>
2002-05-17 20:32:52 +02:00
<varlistentry>
2009-09-19 12:23:27 +02:00
<term><replaceable class="parameter">lang_name</replaceable></term>
2002-05-17 20:32:52 +02:00
<listitem>
<para>
The name of the language that the function is implemented in.
2014-02-13 23:07:05 +01:00
It can be <literal>sql</literal>, <literal>c</literal>,
2002-05-17 20:32:52 +02:00
<literal>internal</literal>, or the name of a user-defined
2021-04-07 21:30:08 +02:00
procedural language, e.g., <literal>plpgsql</literal>. The default is
<literal>sql</literal> if <replaceable
class="parameter">sql_body</replaceable> is specified. Enclosing the
2014-02-13 23:07:05 +01:00
name in single quotes is deprecated and requires matching case.
2002-05-17 20:32:52 +02:00
</para>
</listitem>
2008-12-31 03:25:06 +01:00
</varlistentry>
2015-04-26 16:33:14 +02:00
<varlistentry>
<term><literal>TRANSFORM { FOR TYPE <replaceable class="parameter">type_name</replaceable> } [, ... ] }</literal></term>
<listitem>
<para>
Lists which transforms a call to the function should apply. Transforms
convert between SQL types and language-specific data types;
2017-11-23 15:39:47 +01:00
see <xref linkend="sql-createtransform"/>. Procedural language
2015-04-26 16:33:14 +02:00
implementations usually have hardcoded knowledge of the built-in types,
so those don't need to be listed here. If a procedural language
implementation does not know how to handle a type and no transform is
supplied, it will fall back to a default behavior for converting data
types, but this depends on the implementation.
</para>
</listitem>
</varlistentry>
2008-12-31 03:25:06 +01:00
<varlistentry>
<term><literal>WINDOW</literal></term>
<listitem>
2011-08-07 09:49:45 +02:00
<para><literal>WINDOW</literal> indicates that the function is a
2017-10-09 03:44:17 +02:00
<firstterm>window function</firstterm> rather than a plain function.
2008-12-31 03:25:06 +01:00
This is currently only useful for functions written in C.
2017-10-09 03:44:17 +02:00
The <literal>WINDOW</literal> attribute cannot be changed when
2008-12-31 03:25:06 +01:00
replacing an existing function definition.
</para>
</listitem>
2002-05-17 20:32:52 +02:00
</varlistentry>
<varlistentry>
2003-04-22 12:08:08 +02:00
<term><literal>IMMUTABLE</literal></term>
<term><literal>STABLE</literal></term>
<term><literal>VOLATILE</literal></term>
2002-05-17 20:32:52 +02:00
<listitem>
<para>
2006-11-10 21:52:18 +01:00
These attributes inform the query optimizer about the behavior
of the function. At most one choice
Update reference documentation on may/can/might:
Standard English uses "may", "can", and "might" in different ways:
may - permission, "You may borrow my rake."
can - ability, "I can lift that log."
might - possibility, "It might rain today."
Unfortunately, in conversational English, their use is often mixed, as
in, "You may use this variable to do X", when in fact, "can" is a better
choice. Similarly, "It may crash" is better stated, "It might crash".
2007-02-01 00:26:05 +01:00
can be specified. If none of these appear,
2002-05-17 20:32:52 +02:00
<literal>VOLATILE</literal> is the default assumption.
</para>
2011-08-07 09:49:45 +02:00
<para><literal>IMMUTABLE</literal> indicates that the function
2006-11-10 21:52:18 +01:00
cannot modify the database and always
2002-05-17 20:32:52 +02:00
returns the same result when given the same argument values; that
is, it does not do database lookups or otherwise use information not
2003-04-22 12:08:08 +02:00
directly present in its argument list. If this option is given,
2002-05-17 20:32:52 +02:00
any call of the function with all-constant arguments can be
immediately replaced with the function value.
</para>
2011-08-07 09:49:45 +02:00
<para><literal>STABLE</literal> indicates that the function
2006-11-10 21:52:18 +01:00
cannot modify the database,
and that within a single table scan it will consistently
2002-05-17 20:32:52 +02:00
return the same result for the same argument values, but that its
result could change across SQL statements. This is the appropriate
selection for functions whose results depend on database lookups,
2010-02-25 23:24:00 +01:00
parameter variables (such as the current time zone), etc. (It is
2017-10-09 03:44:17 +02:00
inappropriate for <literal>AFTER</literal> triggers that wish to
2010-02-25 23:24:00 +01:00
query rows modified by the current command.) Also note
2017-10-09 03:44:17 +02:00
that the <function>current_timestamp</function> family of functions qualify
2002-05-17 20:32:52 +02:00
as stable, since their values do not change within a transaction.
</para>
2011-08-07 09:49:45 +02:00
<para><literal>VOLATILE</literal> indicates that the function value can
2002-05-17 20:32:52 +02:00
change even within a single table scan, so no optimizations can be
made. Relatively few database functions are volatile in this sense;
2017-10-09 03:44:17 +02:00
some examples are <literal>random()</literal>, <literal>currval()</literal>,
<literal>timeofday()</literal>. But note that any function that has
2005-11-01 22:09:51 +01:00
side-effects must be classified volatile, even if its result is quite
predictable, to prevent calls from being optimized away; an example is
2017-10-09 03:44:17 +02:00
<literal>setval()</literal>.
2002-05-17 20:32:52 +02:00
</para>
2004-09-13 22:10:13 +02:00
<para>
2017-11-23 15:39:47 +01:00
For additional details see <xref linkend="xfunc-volatility"/>.
2004-09-13 22:10:13 +02:00
</para>
2002-05-17 20:32:52 +02:00
</listitem>
</varlistentry>
2012-02-14 04:20:27 +01:00
<varlistentry>
<term><literal>LEAKPROOF</literal></term>
<listitem>
<para>
<literal>LEAKPROOF</literal> indicates that the function has no side
effects. It reveals no information about its arguments other than by
its return value. For example, a function which throws an error message
for some argument values but not others, or which includes the argument
2015-07-30 19:16:36 +02:00
values in any error message, is not leakproof. This affects how the
system executes queries against views created with the
<literal>security_barrier</literal> option or tables with row level
security enabled. The system will enforce conditions from security
policies and security barrier views before any user-supplied conditions
from the query itself that contain non-leakproof functions, in order to
prevent the inadvertent exposure of data. Functions and operators
marked as leakproof are assumed to be trustworthy, and may be executed
before conditions from security policies and security barrier views.
2015-09-11 03:22:21 +02:00
In addition, functions which do not take arguments or which are not
2015-07-30 19:16:36 +02:00
passed any arguments from the security barrier view or table do not have
to be marked as leakproof to be executed before security conditions. See
2017-11-23 15:39:47 +01:00
<xref linkend="sql-createview"/> and <xref linkend="rules-privileges"/>.
2012-02-14 04:20:27 +01:00
This option can only be set by the superuser.
</para>
</listitem>
</varlistentry>
2002-05-17 20:32:52 +02:00
<varlistentry>
2003-04-22 12:08:08 +02:00
<term><literal>CALLED ON NULL INPUT</literal></term>
<term><literal>RETURNS NULL ON NULL INPUT</literal></term>
<term><literal>STRICT</literal></term>
2002-05-17 20:32:52 +02:00
<listitem>
2011-08-07 09:49:45 +02:00
<para><literal>CALLED ON NULL INPUT</literal> (the default) indicates
2002-05-17 20:32:52 +02:00
that the function will be called normally when some of its
arguments are null. It is then the function author's
2002-09-21 20:32:54 +02:00
responsibility to check for null values if necessary and respond
2002-05-17 20:32:52 +02:00
appropriately.
</para>
2011-08-07 09:49:45 +02:00
<para><literal>RETURNS NULL ON NULL INPUT</literal> or
2002-05-17 20:32:52 +02:00
<literal>STRICT</literal> indicates that the function always
2003-04-22 12:08:08 +02:00
returns null whenever any of its arguments are null. If this
2002-05-17 20:32:52 +02:00
parameter is specified, the function is not executed when there
2003-04-22 12:08:08 +02:00
are null arguments; instead a null result is assumed
2002-05-17 20:32:52 +02:00
automatically.
</para>
</listitem>
</varlistentry>
2002-05-18 15:48:01 +02:00
<varlistentry>
2003-04-22 12:08:08 +02:00
<term><literal><optional>EXTERNAL</optional> SECURITY INVOKER</literal></term>
<term><literal><optional>EXTERNAL</optional> SECURITY DEFINER</literal></term>
2002-05-18 15:48:01 +02:00
<listitem>
2011-08-07 09:49:45 +02:00
<para><literal>SECURITY INVOKER</literal> indicates that the function
2002-05-18 15:48:01 +02:00
is to be executed with the privileges of the user that calls it.
That is the default. <literal>SECURITY DEFINER</literal>
specifies that the function is to be executed with the
2023-01-16 16:49:59 +01:00
privileges of the user that owns it. For information on how to
write <literal>SECURITY DEFINER</literal> functions safely,
<link linkend="sql-createfunction-security">see below</link>.
2002-05-18 15:48:01 +02:00
</para>
<para>
2005-11-01 22:09:51 +01:00
The key word <literal>EXTERNAL</literal> is allowed for SQL
conformance, but it is optional since, unlike in SQL, this feature
applies to all functions not only external ones.
2002-05-18 15:48:01 +02:00
</para>
</listitem>
</varlistentry>
2015-09-16 21:38:47 +02:00
<varlistentry>
<term><literal>PARALLEL</literal></term>
<listitem>
Allow "internal" subtransactions in parallel mode.
Allow use of BeginInternalSubTransaction() in parallel mode, so long
as the subtransaction doesn't attempt to acquire an XID or increment
the command counter. Given those restrictions, the other parallel
processes don't need to know about the subtransaction at all, so
this should be safe. The benefit is that it allows subtransactions
intended for error recovery, such as pl/pgsql exception blocks,
to be used in PARALLEL SAFE functions.
Another reason for doing this is that the API of
BeginInternalSubTransaction() doesn't allow reporting failure.
pl/python for one, and perhaps other PLs, copes very poorly with an
error longjmp out of BeginInternalSubTransaction(). The headline
feature of this patch removes the only easily-triggerable failure
case within that function. There remain some resource-exhaustion
and similar cases, which we now deal with by promoting them to FATAL
errors, so that callers need not try to clean up. (It is likely
that such errors would leave us with corrupted transaction state
inside xact.c, making recovery difficult if not impossible anyway.)
Although this work started because of a report of a pl/python crash,
we're not going to do anything about that in the back branches.
Back-patching this particular fix is obviously not very wise.
While we could contemplate some narrower band-aid, pl/python is
already an untrusted language, so it seems okay to classify this
as a "so don't do that" case.
Patch by me, per report from Hao Zhang. Thanks to Robert Haas for
review.
Discussion: https://postgr.es/m/CALY6Dr-2yLVeVPhNMhuBnRgOZo1UjoTETgtKBx1B2gUi8yy+3g@mail.gmail.com
2024-03-28 17:43:10 +01:00
<para>
<literal>PARALLEL UNSAFE</literal> indicates that the function
can't be executed in parallel mode; the presence of such a
2015-09-16 21:38:47 +02:00
function in an SQL statement forces a serial execution plan. This is
the default. <literal>PARALLEL RESTRICTED</literal> indicates that
Allow "internal" subtransactions in parallel mode.
Allow use of BeginInternalSubTransaction() in parallel mode, so long
as the subtransaction doesn't attempt to acquire an XID or increment
the command counter. Given those restrictions, the other parallel
processes don't need to know about the subtransaction at all, so
this should be safe. The benefit is that it allows subtransactions
intended for error recovery, such as pl/pgsql exception blocks,
to be used in PARALLEL SAFE functions.
Another reason for doing this is that the API of
BeginInternalSubTransaction() doesn't allow reporting failure.
pl/python for one, and perhaps other PLs, copes very poorly with an
error longjmp out of BeginInternalSubTransaction(). The headline
feature of this patch removes the only easily-triggerable failure
case within that function. There remain some resource-exhaustion
and similar cases, which we now deal with by promoting them to FATAL
errors, so that callers need not try to clean up. (It is likely
that such errors would leave us with corrupted transaction state
inside xact.c, making recovery difficult if not impossible anyway.)
Although this work started because of a report of a pl/python crash,
we're not going to do anything about that in the back branches.
Back-patching this particular fix is obviously not very wise.
While we could contemplate some narrower band-aid, pl/python is
already an untrusted language, so it seems okay to classify this
as a "so don't do that" case.
Patch by me, per report from Hao Zhang. Thanks to Robert Haas for
review.
Discussion: https://postgr.es/m/CALY6Dr-2yLVeVPhNMhuBnRgOZo1UjoTETgtKBx1B2gUi8yy+3g@mail.gmail.com
2024-03-28 17:43:10 +01:00
the function can be executed in parallel mode, but only in the parallel
group leader process. <literal>PARALLEL SAFE</literal>
2015-09-16 21:38:47 +02:00
indicates that the function is safe to run in parallel mode without
Allow "internal" subtransactions in parallel mode.
Allow use of BeginInternalSubTransaction() in parallel mode, so long
as the subtransaction doesn't attempt to acquire an XID or increment
the command counter. Given those restrictions, the other parallel
processes don't need to know about the subtransaction at all, so
this should be safe. The benefit is that it allows subtransactions
intended for error recovery, such as pl/pgsql exception blocks,
to be used in PARALLEL SAFE functions.
Another reason for doing this is that the API of
BeginInternalSubTransaction() doesn't allow reporting failure.
pl/python for one, and perhaps other PLs, copes very poorly with an
error longjmp out of BeginInternalSubTransaction(). The headline
feature of this patch removes the only easily-triggerable failure
case within that function. There remain some resource-exhaustion
and similar cases, which we now deal with by promoting them to FATAL
errors, so that callers need not try to clean up. (It is likely
that such errors would leave us with corrupted transaction state
inside xact.c, making recovery difficult if not impossible anyway.)
Although this work started because of a report of a pl/python crash,
we're not going to do anything about that in the back branches.
Back-patching this particular fix is obviously not very wise.
While we could contemplate some narrower band-aid, pl/python is
already an untrusted language, so it seems okay to classify this
as a "so don't do that" case.
Patch by me, per report from Hao Zhang. Thanks to Robert Haas for
review.
Discussion: https://postgr.es/m/CALY6Dr-2yLVeVPhNMhuBnRgOZo1UjoTETgtKBx1B2gUi8yy+3g@mail.gmail.com
2024-03-28 17:43:10 +01:00
restriction, including in parallel worker processes.
2015-09-16 21:38:47 +02:00
</para>
<para>
Functions should be labeled parallel unsafe if they modify any database
Allow "internal" subtransactions in parallel mode.
Allow use of BeginInternalSubTransaction() in parallel mode, so long
as the subtransaction doesn't attempt to acquire an XID or increment
the command counter. Given those restrictions, the other parallel
processes don't need to know about the subtransaction at all, so
this should be safe. The benefit is that it allows subtransactions
intended for error recovery, such as pl/pgsql exception blocks,
to be used in PARALLEL SAFE functions.
Another reason for doing this is that the API of
BeginInternalSubTransaction() doesn't allow reporting failure.
pl/python for one, and perhaps other PLs, copes very poorly with an
error longjmp out of BeginInternalSubTransaction(). The headline
feature of this patch removes the only easily-triggerable failure
case within that function. There remain some resource-exhaustion
and similar cases, which we now deal with by promoting them to FATAL
errors, so that callers need not try to clean up. (It is likely
that such errors would leave us with corrupted transaction state
inside xact.c, making recovery difficult if not impossible anyway.)
Although this work started because of a report of a pl/python crash,
we're not going to do anything about that in the back branches.
Back-patching this particular fix is obviously not very wise.
While we could contemplate some narrower band-aid, pl/python is
already an untrusted language, so it seems okay to classify this
as a "so don't do that" case.
Patch by me, per report from Hao Zhang. Thanks to Robert Haas for
review.
Discussion: https://postgr.es/m/CALY6Dr-2yLVeVPhNMhuBnRgOZo1UjoTETgtKBx1B2gUi8yy+3g@mail.gmail.com
2024-03-28 17:43:10 +01:00
state, change the transaction state (other than by using a
subtransaction for error recovery), access sequences (e.g., by
calling <literal>currval</literal>) or make persistent changes to
settings. They should
be labeled parallel restricted if they access temporary tables,
2015-09-16 21:38:47 +02:00
client connection state, cursors, prepared statements, or miscellaneous
backend-local state which the system cannot synchronize in parallel mode
2020-09-01 00:33:37 +02:00
(e.g., <literal>setseed</literal> cannot be executed other than by the group
2015-09-16 21:38:47 +02:00
leader because a change made by another process would not be reflected
in the leader). In general, if a function is labeled as being safe when
it is restricted or unsafe, or if it is labeled as being restricted when
it is in fact unsafe, it may throw errors or produce wrong answers
when used in a parallel query. C-language functions could in theory
exhibit totally undefined behavior if mislabeled, since there is no way
for the system to protect itself against arbitrary C code, but in most
likely cases the result will be no worse than for any other function.
2017-10-09 03:44:17 +02:00
If in doubt, functions should be labeled as <literal>UNSAFE</literal>, which is
2015-09-16 21:38:47 +02:00
the default.
</para>
</listitem>
</varlistentry>
2007-01-22 02:35:23 +01:00
<varlistentry>
2017-11-30 14:46:13 +01:00
<term><literal>COST</literal> <replaceable class="parameter">execution_cost</replaceable></term>
2007-01-22 02:35:23 +01:00
<listitem>
<para>
A positive number giving the estimated execution cost for the function,
2017-11-23 15:39:47 +01:00
in units of <xref linkend="guc-cpu-operator-cost"/>. If the function
2007-01-22 02:35:23 +01:00
returns a set, this is the cost per returned row. If the cost is
not specified, 1 unit is assumed for C-language and internal functions,
and 100 units for functions in all other languages. Larger values
cause the planner to try to avoid evaluating the function more often
than necessary.
</para>
</listitem>
</varlistentry>
<varlistentry>
2017-11-30 14:46:13 +01:00
<term><literal>ROWS</literal> <replaceable class="parameter">result_rows</replaceable></term>
2007-01-22 02:35:23 +01:00
<listitem>
<para>
A positive number giving the estimated number of rows that the planner
should expect the function to return. This is only allowed when the
function is declared to return a set. The default assumption is
1000 rows.
</para>
</listitem>
</varlistentry>
2019-02-10 00:08:48 +01:00
<varlistentry>
<term><literal>SUPPORT</literal> <replaceable class="parameter">support_function</replaceable></term>
<listitem>
<para>
The name (optionally schema-qualified) of a <firstterm>planner support
function</firstterm> to use for this function. See
<xref linkend="xfunc-optimization"/> for details.
You must be superuser to use this option.
</para>
</listitem>
</varlistentry>
2007-09-03 02:39:26 +02:00
<varlistentry>
2007-09-03 20:46:30 +02:00
<term><replaceable>configuration_parameter</replaceable></term>
2007-09-03 02:39:26 +02:00
<term><replaceable>value</replaceable></term>
<listitem>
<para>
2017-10-09 03:44:17 +02:00
The <literal>SET</literal> clause causes the specified configuration
2007-09-03 02:39:26 +02:00
parameter to be set to the specified value when the function is
entered, and then restored to its prior value when the function exits.
2017-10-09 03:44:17 +02:00
<literal>SET FROM CURRENT</literal> saves the value of the parameter that
is current when <command>CREATE FUNCTION</command> is executed as the value
2016-08-09 19:39:24 +02:00
to be applied when the function is entered.
2007-09-03 02:39:26 +02:00
</para>
2010-03-03 04:14:08 +01:00
<para>
2017-10-09 03:44:17 +02:00
If a <literal>SET</literal> clause is attached to a function, then
the effects of a <command>SET LOCAL</command> command executed inside the
2010-03-03 04:14:08 +01:00
function for the same variable are restricted to the function: the
configuration parameter's prior value is still restored at function exit.
However, an ordinary
2017-10-09 03:44:17 +02:00
<command>SET</command> command (without <literal>LOCAL</literal>) overrides the
<literal>SET</literal> clause, much as it would do for a previous <command>SET
LOCAL</command> command: the effects of such a command will persist after
2010-03-03 04:14:08 +01:00
function exit, unless the current transaction is rolled back.
</para>
2010-11-23 21:27:50 +01:00
2007-09-03 02:39:26 +02:00
<para>
2017-11-23 15:39:47 +01:00
See <xref linkend="sql-set"/> and
<xref linkend="runtime-config"/>
2007-09-03 02:39:26 +02:00
for more information about allowed parameter names and values.
</para>
</listitem>
</varlistentry>
2001-05-19 11:01:10 +02:00
<varlistentry>
<term><replaceable class="parameter">definition</replaceable></term>
<listitem>
<para>
2004-09-21 00:48:29 +02:00
A string constant defining the function; the meaning depends on the
Update reference documentation on may/can/might:
Standard English uses "may", "can", and "might" in different ways:
may - permission, "You may borrow my rake."
can - ability, "I can lift that log."
might - possibility, "It might rain today."
Unfortunately, in conversational English, their use is often mixed, as
in, "You may use this variable to do X", when in fact, "can" is a better
choice. Similarly, "It may crash" is better stated, "It might crash".
2007-02-01 00:26:05 +01:00
language. It can be an internal function name, the path to an
2004-09-21 00:48:29 +02:00
object file, an SQL command, or text in a procedural language.
2001-05-19 11:01:10 +02:00
</para>
2010-03-03 04:14:08 +01:00
<para>
It is often helpful to use dollar quoting (see <xref
2017-11-23 15:39:47 +01:00
linkend="sql-syntax-dollar-quoting"/>) to write the function definition
2010-03-03 04:14:08 +01:00
string, rather than the normal single quote syntax. Without dollar
quoting, any single quotes or backslashes in the function definition must
be escaped by doubling them.
</para>
2001-05-19 11:01:10 +02:00
</listitem>
</varlistentry>
<varlistentry>
2003-04-22 12:08:08 +02:00
<term><literal><replaceable class="parameter">obj_file</replaceable>, <replaceable class="parameter">link_symbol</replaceable></literal></term>
2001-05-19 11:01:10 +02:00
<listitem>
<para>
This form of the <literal>AS</literal> clause is used for
2003-04-22 12:08:08 +02:00
dynamically loadable C language functions when the function name
2001-05-19 11:01:10 +02:00
in the C language source code is not the same as the name of
the SQL function. The string <replaceable
2017-06-20 19:39:57 +02:00
class="parameter">obj_file</replaceable> is the name of the shared
library file containing the compiled C function, and is interpreted
Improve <xref> vs. <command> formatting in the documentation
SQL commands are generally marked up as <command>, except when a link
to a reference page is used using <xref>. But the latter doesn't
create monospace markup, so this looks strange especially when a
paragraph contains a mix of links and non-links.
We considered putting <command> in the <refentrytitle> on the target
side, but that creates some formatting side effects elsewhere.
Generally, it seems safer to solve this on the link source side.
We can't put the <xref> inside the <command>; the DTD doesn't allow
this. DocBook 5 would allow the <command> to have the linkend
attribute itself, but we are not there yet.
So to solve this for now, convert the <xref>s to <link> plus
<command>. This gives the correct look and also gives some more
flexibility what we can put into the link text (e.g., subcommands or
other clauses). In the future, these could then be converted to
DocBook 5 style.
I haven't converted absolutely all xrefs to SQL command reference
pages, only those where we care about the appearance of the link text
or where it was otherwise appropriate to make the appearance match a
bit better. Also in some cases, the links where repetitive, so in
those cases the links where just removed and replaced by a plain
<command>. In cases where we just want the link and don't
specifically care about the generated link text (typically phrased
"for further information see <xref ...>") the xref is kept.
Reported-by: Dagfinn Ilmari Mannsåker <ilmari@ilmari.org>
Discussion: https://www.postgresql.org/message-id/flat/87o8pco34z.fsf@wibble.ilmari.org
2020-10-03 16:16:51 +02:00
as for the <link linkend="sql-load"><command>LOAD</command></link> command. The string
2001-05-19 11:01:10 +02:00
<replaceable class="parameter">link_symbol</replaceable> is the
2003-04-22 12:08:08 +02:00
function's link symbol, that is, the name of the function in the C
2018-07-29 05:08:01 +02:00
language source code. If the link symbol is omitted, it is assumed to
be the same as the name of the SQL function being defined. The C names
of all functions must be different, so you must give overloaded C
functions different C names (for example, use the argument types as
part of the C names).
2001-05-19 11:01:10 +02:00
</para>
2010-03-03 04:14:08 +01:00
<para>
When repeated <command>CREATE FUNCTION</command> calls refer to
the same object file, the file is only loaded once per session.
To unload and
reload the file (perhaps during development), start a new session.
</para>
2001-05-19 11:01:10 +02:00
</listitem>
</varlistentry>
2021-04-07 21:30:08 +02:00
<varlistentry>
<term><replaceable class="parameter">sql_body</replaceable></term>
<listitem>
<para>
The body of a <literal>LANGUAGE SQL</literal> function. This can
either be a single statement
<programlisting>
RETURN <replaceable>expression</replaceable>
</programlisting>
or a block
<programlisting>
BEGIN ATOMIC
<replaceable>statement</replaceable>;
<replaceable>statement</replaceable>;
...
<replaceable>statement</replaceable>;
END
</programlisting>
</para>
<para>
This is similar to writing the text of the function body as a string
constant (see <replaceable>definition</replaceable> above), but there
are some differences: This form only works for <literal>LANGUAGE
SQL</literal>, the string constant form works for all languages. This
form is parsed at function definition time, the string constant form is
parsed at execution time; therefore this form cannot support
polymorphic argument types and other constructs that are not resolvable
at function definition time. This form tracks dependencies between the
function and objects used in the function body, so <literal>DROP
... CASCADE</literal> will work correctly, whereas the form using
string literals may leave dangling functions. Finally, this form is
more compatible with the SQL standard and other SQL implementations.
</para>
</listitem>
</varlistentry>
2001-05-19 11:01:10 +02:00
</variablelist>
2010-03-03 04:23:12 +01:00
</refsect1>
1999-07-22 17:09:15 +02:00
2010-03-03 04:14:08 +01:00
<refsect1 id="sql-createfunction-overloading">
<title>Overloading</title>
1999-05-20 04:57:15 +02:00
1999-07-22 17:09:15 +02:00
<para>
2003-04-22 12:08:08 +02:00
<productname>PostgreSQL</productname> allows function
<firstterm>overloading</firstterm>; that is, the same name can be
used for several different functions so long as they have distinct
2018-07-29 05:08:01 +02:00
input argument types. Whether or not you use it, this capability entails
security precautions when calling functions in databases where some users
mistrust other users; see <xref linkend="typeconv-func"/>.
1999-07-22 17:09:15 +02:00
</para>
2005-04-01 00:46:33 +02:00
<para>
Two functions are considered the same if they have the same names and
2017-10-09 03:44:17 +02:00
<emphasis>input</emphasis> argument types, ignoring any <literal>OUT</literal>
2005-04-01 00:46:33 +02:00
parameters. Thus for example these declarations conflict:
<programlisting>
CREATE FUNCTION foo(int) ...
CREATE FUNCTION foo(int, out text) ...
</programlisting>
</para>
2008-12-18 19:20:35 +01:00
<para>
Functions that have different argument type lists will not be considered
to conflict at creation time, but if defaults are provided they might
conflict in use. For example, consider
<programlisting>
CREATE FUNCTION foo(int) ...
CREATE FUNCTION foo(int, int default 42) ...
</programlisting>
2017-10-09 03:44:17 +02:00
A call <literal>foo(10)</literal> will fail due to the ambiguity about which
2008-12-18 19:20:35 +01:00
function should be called.
</para>
2010-03-03 04:14:08 +01:00
</refsect1>
2007-09-03 02:39:26 +02:00
2010-03-03 04:14:08 +01:00
<refsect1 id="sql-createfunction-notes">
<title>Notes</title>
2003-10-26 05:34:05 +01:00
2009-10-02 20:13:04 +02:00
<para>
2010-03-03 04:14:08 +01:00
The full <acronym>SQL</acronym> type syntax is allowed for
2013-11-13 19:26:33 +01:00
declaring a function's arguments and return value. However,
parenthesized type modifiers (e.g., the precision field for
2017-10-09 03:44:17 +02:00
type <type>numeric</type>) are discarded by <command>CREATE FUNCTION</command>.
2013-11-13 19:26:33 +01:00
Thus for example
2017-10-09 03:44:17 +02:00
<literal>CREATE FUNCTION foo (varchar(10)) ...</literal>
2013-11-13 19:26:33 +01:00
is exactly the same as
2017-10-09 03:44:17 +02:00
<literal>CREATE FUNCTION foo (varchar) ...</literal>.
2009-10-02 20:13:04 +02:00
</para>
2009-10-08 04:39:25 +02:00
<para>
When replacing an existing function with <command>CREATE OR REPLACE
2017-10-09 03:44:17 +02:00
FUNCTION</command>, there are restrictions on changing parameter names.
2009-10-08 04:39:25 +02:00
You cannot change the name already assigned to any input parameter
(although you can add names to parameters that had none before).
If there is more than one output parameter, you cannot change the
names of the output parameters, because that would change the
column names of the anonymous composite type that describes the
function's result. These restrictions are made to ensure that
existing calls of the function do not stop working when it is replaced.
</para>
2010-02-14 01:48:12 +01:00
<para>
2017-10-09 03:44:17 +02:00
If a function is declared <literal>STRICT</literal> with a <literal>VARIADIC</literal>
2010-02-14 01:48:12 +01:00
argument, the strictness check tests that the variadic array <emphasis>as
2017-10-09 03:44:17 +02:00
a whole</emphasis> is non-null. The function will still be called if the
2010-02-14 02:01:35 +01:00
array has null elements.
2010-02-14 01:48:12 +01:00
</para>
1998-12-29 03:24:47 +01:00
</refsect1>
2001-05-19 11:01:10 +02:00
<refsect1 id="sql-createfunction-examples">
<title>Examples</title>
1999-06-14 09:37:05 +02:00
<para>
2021-06-11 03:38:04 +02:00
Add two integers using an SQL function:
2001-05-19 11:01:10 +02:00
<programlisting>
2003-04-22 12:08:08 +02:00
CREATE FUNCTION add(integer, integer) RETURNS integer
2005-01-04 01:39:53 +01:00
AS 'select $1 + $2;'
2003-04-22 12:08:08 +02:00
LANGUAGE SQL
IMMUTABLE
RETURNS NULL ON NULL INPUT;
2021-04-07 21:30:08 +02:00
</programlisting>
The same function written in a more SQL-conforming style, using argument
names and an unquoted body:
<programlisting>
CREATE FUNCTION add(a integer, b integer) RETURNS integer
LANGUAGE SQL
IMMUTABLE
RETURNS NULL ON NULL INPUT
RETURN a + b;
2004-07-12 01:23:43 +02:00
</programlisting>
</para>
<para>
2004-10-01 04:00:44 +02:00
Increment an integer, making use of an argument name, in
<application>PL/pgSQL</application>:
2004-07-12 01:23:43 +02:00
<programlisting>
2005-01-04 01:39:53 +01:00
CREATE OR REPLACE FUNCTION increment(i integer) RETURNS integer AS $$
2004-09-21 00:48:29 +02:00
BEGIN
RETURN i + 1;
2005-01-04 01:39:53 +01:00
END;
$$ LANGUAGE plpgsql;
2005-04-01 00:46:33 +02:00
</programlisting>
</para>
<para>
Return a record containing multiple output parameters:
<programlisting>
CREATE FUNCTION dup(in int, out f1 int, out f2 text)
AS $$ SELECT $1, CAST($1 AS text) || ' is text' $$
LANGUAGE SQL;
SELECT * FROM dup(42);
</programlisting>
You can do the same thing more verbosely with an explicitly named
composite type:
<programlisting>
CREATE TYPE dup_result AS (f1 int, f2 text);
CREATE FUNCTION dup(int) RETURNS dup_result
AS $$ SELECT $1, CAST($1 AS text) || ' is text' $$
LANGUAGE SQL;
SELECT * FROM dup(42);
2001-05-19 11:01:10 +02:00
</programlisting>
2017-10-09 03:44:17 +02:00
Another way to return multiple columns is to use a <literal>TABLE</literal>
2008-07-18 05:32:53 +02:00
function:
<programlisting>
CREATE FUNCTION dup(int) RETURNS TABLE(f1 int, f2 text)
AS $$ SELECT $1, CAST($1 AS text) || ' is text' $$
LANGUAGE SQL;
SELECT * FROM dup(42);
</programlisting>
2017-10-09 03:44:17 +02:00
However, a <literal>TABLE</literal> function is different from the
preceding examples, because it actually returns a <emphasis>set</emphasis>
2008-07-18 05:32:53 +02:00
of records, not just one record.
2001-05-19 11:01:10 +02:00
</para>
1999-06-14 09:37:05 +02:00
</refsect1>
2001-05-19 11:01:10 +02:00
2007-04-20 04:37:38 +02:00
<refsect1 id="sql-createfunction-security">
<title>Writing <literal>SECURITY DEFINER</literal> Functions Safely</title>
2012-12-14 05:00:42 +01:00
<indexterm>
2017-10-09 03:44:17 +02:00
<primary><varname>search_path</varname> configuration parameter</primary>
<secondary>use in securing functions</secondary>
2012-12-14 05:00:42 +01:00
</indexterm>
2023-01-16 16:49:59 +01:00
<indexterm>
<primary><varname>createrole_self_grant</varname> configuration parameter</primary>
<secondary>use in securing functions</secondary>
</indexterm>
2007-04-20 04:37:38 +02:00
<para>
Because a <literal>SECURITY DEFINER</literal> function is executed
2017-03-21 03:33:26 +01:00
with the privileges of the user that owns it, care is needed to
2022-09-28 19:05:21 +02:00
ensure that the function cannot be misused. For security,
2017-11-23 15:39:47 +01:00
<xref linkend="guc-search-path"/> should be set to exclude any schemas
2007-04-20 04:37:38 +02:00
writable by untrusted users. This prevents
2016-07-15 16:58:39 +02:00
malicious users from creating objects (e.g., tables, functions, and
operators) that mask objects intended to be used by the function.
Particularly important in this regard is the
2007-04-20 04:37:38 +02:00
temporary-table schema, which is searched first by default, and
2016-05-12 14:32:12 +02:00
is normally writable by anyone. A secure arrangement can be obtained
2007-04-20 04:37:38 +02:00
by forcing the temporary schema to be searched last. To do this,
2017-10-09 03:44:17 +02:00
write <literal>pg_temp</literal><indexterm><primary>pg_temp</primary><secondary>securing functions</secondary></indexterm> as the last entry in <varname>search_path</varname>.
2007-04-20 04:37:38 +02:00
This function illustrates safe usage:
<programlisting>
CREATE FUNCTION check_password(uname TEXT, pass TEXT)
RETURNS BOOLEAN AS $$
DECLARE passed BOOLEAN;
BEGIN
SELECT (pwd = $2) INTO passed
FROM pwds
WHERE username = $1;
RETURN passed;
END;
2007-09-03 02:39:26 +02:00
$$ LANGUAGE plpgsql
SECURITY DEFINER
-- Set a secure search_path: trusted schema(s), then 'pg_temp'.
SET search_path = admin, pg_temp;
</programlisting>
2017-10-09 03:44:17 +02:00
This function's intention is to access a table <literal>admin.pwds</literal>.
But without the <literal>SET</literal> clause, or with a <literal>SET</literal> clause
mentioning only <literal>admin</literal>, the function could be subverted by
creating a temporary table named <literal>pwds</literal>.
2016-07-15 16:58:39 +02:00
</para>
2007-09-03 02:39:26 +02:00
<para>
2023-01-16 16:49:59 +01:00
If the security definer function intends to create roles, and if it
is running as a non-superuser, <varname>createrole_self_grant</varname>
should also be set to a known value using the <literal>SET</literal>
clause.
2007-09-03 02:39:26 +02:00
</para>
<para>
Another point to keep in mind is that by default, execute privilege
2017-10-09 03:44:17 +02:00
is granted to <literal>PUBLIC</literal> for newly created functions
2018-12-03 17:40:49 +01:00
(see <xref linkend="ddl-priv"/> for more
2007-09-03 02:39:26 +02:00
information). Frequently you will wish to restrict use of a security
definer function to only some users. To do that, you must revoke
2017-10-09 03:44:17 +02:00
the default <literal>PUBLIC</literal> privileges and then grant execute
2007-09-03 02:39:26 +02:00
privilege selectively. To avoid having a window where the new function
is accessible to all, create it and set the privileges within a single
transaction. For example:
</para>
<programlisting>
BEGIN;
CREATE FUNCTION check_password(uname TEXT, pass TEXT) ... SECURITY DEFINER;
REVOKE ALL ON FUNCTION check_password(uname TEXT, pass TEXT) FROM PUBLIC;
GRANT EXECUTE ON FUNCTION check_password(uname TEXT, pass TEXT) TO admins;
COMMIT;
2007-04-20 04:37:38 +02:00
</programlisting>
</refsect1>
2001-05-19 11:01:10 +02:00
<refsect1 id="sql-createfunction-compat">
<title>Compatibility</title>
1999-07-22 17:09:15 +02:00
2001-05-19 11:01:10 +02:00
<para>
2021-04-07 21:30:08 +02:00
A <command>CREATE FUNCTION</command> command is defined in the SQL
standard. The <productname>PostgreSQL</productname> implementation can be
used in a compatible way but has many extensions. Conversely, the SQL
standard specifies a number of optional features that are not implemented
in <productname>PostgreSQL</productname>.
2001-05-19 11:01:10 +02:00
</para>
2005-04-01 00:46:33 +02:00
<para>
2021-04-07 21:30:08 +02:00
The following are important compatibility issues:
<itemizedlist>
<listitem>
<para>
<literal>OR REPLACE</literal> is a PostgreSQL extension.
</para>
</listitem>
<listitem>
<para>
For compatibility with some other database systems, <replaceable
class="parameter">argmode</replaceable> can be written either before or
after <replaceable class="parameter">argname</replaceable>. But only
the first way is standard-compliant.
</para>
</listitem>
<listitem>
<para>
For parameter defaults, the SQL standard specifies only the syntax with
the <literal>DEFAULT</literal> key word. The syntax with
<literal>=</literal> is used in T-SQL and Firebird.
</para>
</listitem>
<listitem>
<para>
The <literal>SETOF</literal> modifier is a PostgreSQL extension.
</para>
</listitem>
<listitem>
<para>
Only <literal>SQL</literal> is standardized as a language.
</para>
</listitem>
<listitem>
<para>
All other attributes except <literal>CALLED ON NULL INPUT</literal> and
<literal>RETURNS NULL ON NULL INPUT</literal> are not standardized.
</para>
</listitem>
<listitem>
<para>
For the body of <literal>LANGUAGE SQL</literal> functions, the SQL
standard only specifies the <replaceable>sql_body</replaceable> form.
</para>
</listitem>
</itemizedlist>
2005-04-01 00:46:33 +02:00
</para>
2008-12-04 18:51:28 +01:00
<para>
2021-04-07 21:30:08 +02:00
Simple <literal>LANGUAGE SQL</literal> functions can be written in a way
that is both standard-conforming and portable to other implementations.
More complex functions using advanced features, optimization attributes, or
other languages will necessarily be specific to PostgreSQL in a significant
way.
2008-12-04 18:51:28 +01:00
</para>
2001-05-19 11:01:10 +02:00
</refsect1>
1998-09-16 16:43:12 +02:00
2001-05-19 11:01:10 +02:00
2005-01-04 01:39:53 +01:00
<refsect1>
2001-05-19 11:01:10 +02:00
<title>See Also</title>
2005-01-04 01:39:53 +01:00
<simplelist type="inline">
2017-11-23 15:39:47 +01:00
<member><xref linkend="sql-alterfunction"/></member>
<member><xref linkend="sql-dropfunction"/></member>
<member><xref linkend="sql-grant"/></member>
<member><xref linkend="sql-load"/></member>
<member><xref linkend="sql-revoke"/></member>
2005-01-04 01:39:53 +01:00
</simplelist>
1998-05-13 07:34:00 +02:00
</refsect1>
2001-05-19 11:01:10 +02:00
1999-06-14 09:37:05 +02:00
</refentry>