New contrib module, pg_surgery, with heap surgery functions.

Sometimes it happens that the visibility information for a tuple
becomes corrupted, either due to bugs in the database software or
external factors. Provide a function heap_force_kill() that can
be used to truncate such dead tuples to dead line pointers, and
a function heap_force_freeze() that can be used to overwrite the
visibility information in such a way that the tuple becomes
all-visible.

These functions are unsafe, in that you can easily use them to
corrupt a database that was not previously corrupted, and you can
use them to further corrupt an already-corrupted database or to
destroy data. The documentation accordingly cautions against
casual use. However, in some cases they permit recovery of data
that would otherwise be very difficult to recover, or to allow a
system to continue to function when it would otherwise be difficult
to do so.

Because we may want to add other functions for performing other
kinds of surgery in the future, the new contrib module is called
pg_surgery rather than something specific to these functions. I
proposed back-patching this so that it could be more easily used
by people running existing releases who are facing these kinds of
problems, but that proposal did not attract enough support, so
no back-patch for now.

Ashutosh Sharma, reviewed and tested by Andrey M. Borodin,
M. Beena Emerson, Masahiko Sawada, Rajkumar Raghuwanshi,
Asim Praveen, and Mark Dilger, and somewhat revised by me.

Discussion: http://postgr.es/m/CA+TgmoZW1fsU-QUNCRUQMGUygBDPVeOTLCqRdQZch=EYZnctSA@mail.gmail.com

7 files changed, 749 insertions, 0 deletions

diff --git a/contrib/pg_surgery/.gitignore b/contrib/pg_surgery/.gitignore
new file mode 100644
index 0000000000..5dcb3ff972
--- /dev/null
+++ b/contrib/pg_surgery/.gitignore
@@ -0,0 +1,4 @@
+# Generated subdirectories
+/log/
+/results/
+/tmp_check/
diff --git a/contrib/pg_surgery/Makefile b/contrib/pg_surgery/Makefile
new file mode 100644
index 0000000000..a66776c4c4
--- /dev/null
+++ b/contrib/pg_surgery/Makefile
@@ -0,0 +1,23 @@
+# contrib/pg_surgery/Makefile
+
+MODULE_big = pg_surgery
+OBJS = \
+	$(WIN32RES) \
+	heap_surgery.o
+
+EXTENSION = pg_surgery
+DATA = pg_surgery--1.0.sql
+PGFILEDESC = "pg_surgery - perform surgery on a damaged relation"
+
+REGRESS = heap_surgery
+
+ifdef USE_PGXS
+PG_CONFIG = pg_config
+PGXS := $(shell $(PG_CONFIG) --pgxs)
+include $(PGXS)
+else
+subdir = contrib/pg_surgery
+top_builddir = ../..
+include $(top_builddir)/src/Makefile.global
+include $(top_srcdir)/contrib/contrib-global.mk
+endif
diff --git a/contrib/pg_surgery/expected/heap_surgery.out b/contrib/pg_surgery/expected/heap_surgery.out
new file mode 100644
index 0000000000..9451c5747b
--- /dev/null
+++ b/contrib/pg_surgery/expected/heap_surgery.out
@@ -0,0 +1,180 @@
+create extension pg_surgery;
+-- create a normal heap table and insert some rows.
+-- note that we don't commit the transaction, so autovacuum can't interfere.
+begin;
+create table htab(a int);
+insert into htab values (100), (200), (300), (400), (500);
+-- test empty TID array
+select heap_force_freeze('htab'::regclass, ARRAY[]::tid[]);
+ heap_force_freeze 
+-------------------
+ 
+(1 row)
+
+-- nothing should be frozen yet
+select * from htab where xmin = 2;
+ a 
+---
+(0 rows)
+
+-- freeze forcibly
+select heap_force_freeze('htab'::regclass, ARRAY['(0, 4)']::tid[]);
+ heap_force_freeze 
+-------------------
+ 
+(1 row)
+
+-- now we should have one frozen tuple
+select ctid, xmax from htab where xmin = 2;
+ ctid  | xmax 
+-------+------
+ (0,4) |    0
+(1 row)
+
+-- kill forcibly
+select heap_force_kill('htab'::regclass, ARRAY['(0, 4)']::tid[]);
+ heap_force_kill 
+-----------------
+ 
+(1 row)
+
+-- should be gone now
+select * from htab where ctid = '(0, 4)';
+ a 
+---
+(0 rows)
+
+-- should now be skipped because it's already dead
+select heap_force_kill('htab'::regclass, ARRAY['(0, 4)']::tid[]);
+NOTICE:  skipping tid (0, 4) for relation "htab" because it is marked dead
+ heap_force_kill 
+-----------------
+ 
+(1 row)
+
+select heap_force_freeze('htab'::regclass, ARRAY['(0, 4)']::tid[]);
+NOTICE:  skipping tid (0, 4) for relation "htab" because it is marked dead
+ heap_force_freeze 
+-------------------
+ 
+(1 row)
+
+-- freeze two TIDs at once while skipping an out-of-range block number
+select heap_force_freeze('htab'::regclass,
+						 ARRAY['(0, 1)', '(0, 3)', '(1, 1)']::tid[]);
+NOTICE:  skipping block 1 for relation "htab" because the block number is out of range
+ heap_force_freeze 
+-------------------
+ 
+(1 row)
+
+-- we should now have two frozen tuples
+select ctid, xmax from htab where xmin = 2;
+ ctid  | xmax 
+-------+------
+ (0,1) |    0
+ (0,3) |    0
+(2 rows)
+
+-- out-of-range TIDs should be skipped
+select heap_force_freeze('htab'::regclass, ARRAY['(0, 0)', '(0, 6)']::tid[]);
+NOTICE:  skipping tid (0, 0) for relation "htab" because the item number is out of range
+NOTICE:  skipping tid (0, 6) for relation "htab" because the item number is out of range
+ heap_force_freeze 
+-------------------
+ 
+(1 row)
+
+rollback;
+-- set up a new table with a redirected line pointer
+create table htab2(a int) with (autovacuum_enabled = off);
+insert into htab2 values (100);
+update htab2 set a = 200;
+vacuum htab2;
+-- redirected TIDs should be skipped
+select heap_force_kill('htab2'::regclass, ARRAY['(0, 1)']::tid[]);
+NOTICE:  skipping tid (0, 1) for relation "htab2" because it redirects to item 2
+ heap_force_kill 
+-----------------
+ 
+(1 row)
+
+-- now create an unused line pointer
+select ctid from htab2;
+ ctid  
+-------
+ (0,2)
+(1 row)
+
+update htab2 set a = 300;
+select ctid from htab2;
+ ctid  
+-------
+ (0,3)
+(1 row)
+
+vacuum freeze htab2;
+-- unused TIDs should be skipped
+select heap_force_kill('htab2'::regclass, ARRAY['(0, 2)']::tid[]);
+NOTICE:  skipping tid (0, 2) for relation "htab2" because it is marked unused
+ heap_force_kill 
+-----------------
+ 
+(1 row)
+
+-- multidimensional TID array should be rejected
+select heap_force_kill('htab2'::regclass, ARRAY[['(0, 2)']]::tid[]);
+ERROR:  argument must be empty or one-dimensional array
+-- TID array with nulls should be rejected
+select heap_force_kill('htab2'::regclass, ARRAY[NULL]::tid[]);
+ERROR:  array must not contain nulls
+-- but we should be able to kill the one tuple we have
+select heap_force_kill('htab2'::regclass, ARRAY['(0, 3)']::tid[]);
+ heap_force_kill 
+-----------------
+ 
+(1 row)
+
+-- materialized view.
+-- note that we don't commit the transaction, so autovacuum can't interfere.
+begin;
+create materialized view mvw as select a from generate_series(1, 3) a;
+select * from mvw where xmin = 2;
+ a 
+---
+(0 rows)
+
+select heap_force_freeze('mvw'::regclass, ARRAY['(0, 3)']::tid[]);
+ heap_force_freeze 
+-------------------
+ 
+(1 row)
+
+select * from mvw where xmin = 2;
+ a 
+---
+ 3
+(1 row)
+
+select heap_force_kill('mvw'::regclass, ARRAY['(0, 3)']::tid[]);
+ heap_force_kill 
+-----------------
+ 
+(1 row)
+
+select * from mvw where ctid = '(0, 3)';
+ a 
+---
+(0 rows)
+
+rollback;
+-- check that it fails on an unsupported relkind
+create view vw as select 1;
+select heap_force_kill('vw'::regclass, ARRAY['(0, 1)']::tid[]);
+ERROR:  "vw" is not a table, materialized view, or TOAST table
+select heap_force_freeze('vw'::regclass, ARRAY['(0, 1)']::tid[]);
+ERROR:  "vw" is not a table, materialized view, or TOAST table
+-- cleanup.
+drop table htab2;
+drop view vw;
+drop extension pg_surgery;
diff --git a/contrib/pg_surgery/heap_surgery.c b/contrib/pg_surgery/heap_surgery.c
new file mode 100644
index 0000000000..eb96b4bb36
--- /dev/null
+++ b/contrib/pg_surgery/heap_surgery.c
@@ -0,0 +1,428 @@
+/*-------------------------------------------------------------------------
+ *
+ * heap_surgery.c
+ *	  Functions to perform surgery on the damaged heap table.
+ *
+ * Copyright (c) 2020, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ *	  contrib/pg_surgery/heap_surgery.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#include "access/heapam.h"
+#include "access/visibilitymap.h"
+#include "catalog/pg_am_d.h"
+#include "catalog/pg_proc_d.h"
+#include "miscadmin.h"
+#include "storage/bufmgr.h"
+#include "utils/acl.h"
+#include "utils/rel.h"
+
+PG_MODULE_MAGIC;
+
+/* Options to forcefully change the state of a heap tuple. */
+typedef enum HeapTupleForceOption
+{
+	HEAP_FORCE_KILL,
+	HEAP_FORCE_FREEZE
+} HeapTupleForceOption;
+
+PG_FUNCTION_INFO_V1(heap_force_kill);
+PG_FUNCTION_INFO_V1(heap_force_freeze);
+
+static int32 tidcmp(const void *a, const void *b);
+static Datum heap_force_common(FunctionCallInfo fcinfo,
+							   HeapTupleForceOption heap_force_opt);
+static void sanity_check_tid_array(ArrayType *ta, int *ntids);
+static void sanity_check_relation(Relation rel);
+static BlockNumber find_tids_one_page(ItemPointer tids, int ntids,
+									  OffsetNumber *next_start_ptr);
+
+/*-------------------------------------------------------------------------
+ * heap_force_kill()
+ *
+ * Force kill the tuple(s) pointed to by the item pointer(s) stored in the
+ * given TID array.
+ *
+ * Usage: SELECT heap_force_kill(regclass, tid[]);
+ *-------------------------------------------------------------------------
+ */
+Datum
+heap_force_kill(PG_FUNCTION_ARGS)
+{
+	PG_RETURN_DATUM(heap_force_common(fcinfo, HEAP_FORCE_KILL));
+}
+
+/*-------------------------------------------------------------------------
+ * heap_force_freeze()
+ *
+ * Force freeze the tuple(s) pointed to by the item pointer(s) stored in the
+ * given TID array.
+ *
+ * Usage: SELECT heap_force_freeze(regclass, tid[]);
+ *-------------------------------------------------------------------------
+ */
+Datum
+heap_force_freeze(PG_FUNCTION_ARGS)
+{
+	PG_RETURN_DATUM(heap_force_common(fcinfo, HEAP_FORCE_FREEZE));
+}
+
+/*-------------------------------------------------------------------------
+ * heap_force_common()
+ *
+ * Common code for heap_force_kill and heap_force_freeze
+ *-------------------------------------------------------------------------
+ */
+static Datum
+heap_force_common(FunctionCallInfo fcinfo, HeapTupleForceOption heap_force_opt)
+{
+	Oid			relid = PG_GETARG_OID(0);
+	ArrayType  *ta = PG_GETARG_ARRAYTYPE_P_COPY(1);
+	ItemPointer tids;
+	int			ntids,
+				nblocks;
+	Relation	rel;
+	OffsetNumber curr_start_ptr,
+				next_start_ptr;
+	bool		include_this_tid[MaxHeapTuplesPerPage];
+
+	if (RecoveryInProgress())
+		ereport(ERROR,
+				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+				 errmsg("recovery is in progress"),
+				 errhint("heap surgery functions cannot be executed during recovery.")));
+
+	/* Check inputs. */
+	sanity_check_tid_array(ta, &ntids);
+
+	rel = relation_open(relid, RowExclusiveLock);
+
+	/* Check target relation. */
+	sanity_check_relation(rel);
+
+	tids = ((ItemPointer) ARR_DATA_PTR(ta));
+
+	/*
+	 * If there is more than one TID in the array, sort them so that we can
+	 * easily fetch all the TIDs belonging to one particular page from the
+	 * array.
+	 */
+	if (ntids > 1)
+		qsort((void *) tids, ntids, sizeof(ItemPointerData), tidcmp);
+
+	curr_start_ptr = next_start_ptr = 0;
+	nblocks = RelationGetNumberOfBlocks(rel);
+
+	/*
+	 * Loop, performing the necessary actions for each block.
+	 */
+	while (next_start_ptr != ntids)
+	{
+		Buffer		buf;
+		Buffer		vmbuf = InvalidBuffer;
+		Page		page;
+		BlockNumber blkno;
+		OffsetNumber curoff;
+		OffsetNumber maxoffset;
+		int			i;
+		bool		did_modify_page = false;
+		bool		did_modify_vm = false;
+
+		CHECK_FOR_INTERRUPTS();
+
+		/*
+		 * Find all the TIDs belonging to one particular page starting from
+		 * next_start_ptr and process them one by one.
+		 */
+		blkno = find_tids_one_page(tids, ntids, &next_start_ptr);
+
+		/* Check whether the block number is valid. */
+		if (blkno >= nblocks)
+		{
+			/* Update the current_start_ptr before moving to the next page. */
+			curr_start_ptr = next_start_ptr;
+
+			ereport(NOTICE,
+					(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+					 errmsg("skipping block %u for relation \"%s\" because the block number is out of range",
+							blkno, RelationGetRelationName(rel))));
+			continue;
+		}
+
+		buf = ReadBuffer(rel, blkno);
+		LockBufferForCleanup(buf);
+
+		page = BufferGetPage(buf);
+
+		maxoffset = PageGetMaxOffsetNumber(page);
+
+		/*
+		 * Figure out which TIDs we are going to process and which ones we are
+		 * going to skip.
+		 */
+		memset(include_this_tid, 0, sizeof(include_this_tid));
+		for (i = curr_start_ptr; i < next_start_ptr; i++)
+		{
+			OffsetNumber offno = ItemPointerGetOffsetNumberNoCheck(&tids[i]);
+			ItemId		itemid;
+
+			/* Check whether the offset number is valid. */
+			if (offno == InvalidOffsetNumber || offno > maxoffset)
+			{
+				ereport(NOTICE,
+						errmsg("skipping tid (%u, %u) for relation \"%s\" because the item number is out of range",
+							   blkno, offno, RelationGetRelationName(rel)));
+				continue;
+			}
+
+			itemid = PageGetItemId(page, offno);
+
+			/* Only accept an item ID that is used. */
+			if (ItemIdIsRedirected(itemid))
+			{
+				ereport(NOTICE,
+						errmsg("skipping tid (%u, %u) for relation \"%s\" because it redirects to item %u",
+							   blkno, offno, RelationGetRelationName(rel),
+							   ItemIdGetRedirect(itemid)));
+				continue;
+			}
+			else if (ItemIdIsDead(itemid))
+			{
+				ereport(NOTICE,
+						(errmsg("skipping tid (%u, %u) for relation \"%s\" because it is marked dead",
+								blkno, offno, RelationGetRelationName(rel))));
+				continue;
+			}
+			else if (!ItemIdIsUsed(itemid))
+			{
+				ereport(NOTICE,
+						(errmsg("skipping tid (%u, %u) for relation \"%s\" because it is marked unused",
+								blkno, offno, RelationGetRelationName(rel))));
+				continue;
+			}
+
+			/* Mark it for processing. */
+			Assert(offno < MaxHeapTuplesPerPage);
+			include_this_tid[offno] = true;
+		}
+
+		/*
+		 * Before entering the critical section, pin the visibility map page
+		 * if it appears to be necessary.
+		 */
+		if (heap_force_opt == HEAP_FORCE_KILL && PageIsAllVisible(page))
+			visibilitymap_pin(rel, blkno, &vmbuf);
+
+		/* No ereport(ERROR) from here until all the changes are logged. */
+		START_CRIT_SECTION();
+
+		for (curoff = FirstOffsetNumber; curoff <= maxoffset;
+			 curoff = OffsetNumberNext(curoff))
+		{
+			ItemId		itemid;
+
+			if (!include_this_tid[curoff])
+				continue;
+
+			itemid = PageGetItemId(page, curoff);
+			Assert(ItemIdIsNormal(itemid));
+
+			did_modify_page = true;
+
+			if (heap_force_opt == HEAP_FORCE_KILL)
+			{
+				ItemIdSetDead(itemid);
+
+				/*
+				 * If the page is marked all-visible, we must clear
+				 * PD_ALL_VISIBLE flag on the page header and an all-visible
+				 * bit on the visibility map corresponding to the page.
+				 */
+				if (PageIsAllVisible(page))
+				{
+					PageClearAllVisible(page);
+					visibilitymap_clear(rel, blkno, vmbuf,
+										VISIBILITYMAP_VALID_BITS);
+					did_modify_vm = true;
+				}
+			}
+			else
+			{
+				HeapTupleHeader htup;
+
+				Assert(heap_force_opt == HEAP_FORCE_FREEZE);
+
+				htup = (HeapTupleHeader) PageGetItem(page, itemid);
+
+				/*
+				 * Reset all visibility-related fields of the tuple. This
+				 * logic should mimic heap_execute_freeze_tuple(), but we
+				 * choose to reset xmin and ctid just to be sure that no
+				 * potentially-garbled data is left behind.
+				 */
+				ItemPointerSet(&htup->t_ctid, blkno, curoff);
+				HeapTupleHeaderSetXmin(htup, FrozenTransactionId);
+				HeapTupleHeaderSetXmax(htup, InvalidTransactionId);
+				if (htup->t_infomask & HEAP_MOVED)
+				{
+					if (htup->t_infomask & HEAP_MOVED_OFF)
+						HeapTupleHeaderSetXvac(htup, InvalidTransactionId);
+					else
+						HeapTupleHeaderSetXvac(htup, FrozenTransactionId);
+				}
+
+				/*
+				 * Clear all the visibility-related bits of this tuple and
+				 * mark it as frozen. Also, get rid of HOT_UPDATED and
+				 * KEYS_UPDATES bits.
+				 */
+				htup->t_infomask &= ~HEAP_XACT_MASK;
+				htup->t_infomask |= (HEAP_XMIN_FROZEN | HEAP_XMAX_INVALID);
+				htup->t_infomask2 &= ~HEAP_HOT_UPDATED;
+				htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
+			}
+		}
+
+		/*
+		 * If the page was modified, only then, we mark the buffer dirty or do
+		 * the WAL logging.
+		 */
+		if (did_modify_page)
+		{
+			/* Mark buffer dirty before we write WAL. */
+			MarkBufferDirty(buf);
+
+			/* XLOG stuff */
+			if (RelationNeedsWAL(rel))
+				log_newpage_buffer(buf, true);
+		}
+
+		/* WAL log the VM page if it was modified. */
+		if (did_modify_vm && RelationNeedsWAL(rel))
+			log_newpage_buffer(vmbuf, false);
+
+		END_CRIT_SECTION();
+
+		UnlockReleaseBuffer(buf);
+
+		if (vmbuf != InvalidBuffer)
+			ReleaseBuffer(vmbuf);
+
+		/* Update the current_start_ptr before moving to the next page. */
+		curr_start_ptr = next_start_ptr;
+	}
+
+	relation_close(rel, RowExclusiveLock);
+
+	pfree(ta);
+
+	PG_RETURN_VOID();
+}
+
+/*-------------------------------------------------------------------------
+ * tidcmp()
+ *
+ * Compare two item pointers, return -1, 0, or +1.
+ *
+ * See ItemPointerCompare for details.
+ * ------------------------------------------------------------------------
+ */
+static int32
+tidcmp(const void *a, const void *b)
+{
+	ItemPointer iptr1 = ((const ItemPointer) a);
+	ItemPointer iptr2 = ((const ItemPointer) b);
+
+	return ItemPointerCompare(iptr1, iptr2);
+}
+
+/*-------------------------------------------------------------------------
+ * sanity_check_tid_array()
+ *
+ * Perform sanity checks on the given tid array, and set *ntids to the
+ * number of items in the array.
+ * ------------------------------------------------------------------------
+ */
+static void
+sanity_check_tid_array(ArrayType *ta, int *ntids)
+{
+	if (ARR_HASNULL(ta) && array_contains_nulls(ta))
+		ereport(ERROR,
+				(errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
+				 errmsg("array must not contain nulls")));
+
+	if (ARR_NDIM(ta) > 1)
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("argument must be empty or one-dimensional array")));
+
+	*ntids = ArrayGetNItems(ARR_NDIM(ta), ARR_DIMS(ta));
+}
+
+/*-------------------------------------------------------------------------
+ * sanity_check_relation()
+ *
+ * Perform sanity checks on the given relation.
+ * ------------------------------------------------------------------------
+ */
+static void
+sanity_check_relation(Relation rel)
+{
+	if (rel->rd_rel->relkind != RELKIND_RELATION &&
+		rel->rd_rel->relkind != RELKIND_MATVIEW &&
+		rel->rd_rel->relkind != RELKIND_TOASTVALUE)
+		ereport(ERROR,
+				(errcode(ERRCODE_WRONG_OBJECT_TYPE),
+				 errmsg("\"%s\" is not a table, materialized view, or TOAST table",
+						RelationGetRelationName(rel))));
+
+	if (rel->rd_rel->relam != HEAP_TABLE_AM_OID)
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("only heap AM is supported")));
+
+	/* Must be owner of the table or superuser. */
+	if (!pg_class_ownercheck(RelationGetRelid(rel), GetUserId()))
+		aclcheck_error(ACLCHECK_NOT_OWNER,
+					   get_relkind_objtype(rel->rd_rel->relkind),
+					   RelationGetRelationName(rel));
+}
+
+/*-------------------------------------------------------------------------
+ * find_tids_one_page()
+ *
+ * Find all the tids residing in the same page as tids[next_start_ptr], and
+ * update next_start_ptr so that it points to the first tid in the next page.
+ *
+ * NOTE: The input tids[] array must be sorted.
+ * ------------------------------------------------------------------------
+ */
+static BlockNumber
+find_tids_one_page(ItemPointer tids, int ntids, OffsetNumber *next_start_ptr)
+{
+	int			i;
+	BlockNumber prev_blkno,
+				blkno;
+
+	prev_blkno = blkno = InvalidBlockNumber;
+
+	for (i = *next_start_ptr; i < ntids; i++)
+	{
+		ItemPointerData tid = tids[i];
+
+		blkno = ItemPointerGetBlockNumberNoCheck(&tid);
+
+		if (i == *next_start_ptr)
+			prev_blkno = blkno;
+
+		if (prev_blkno != blkno)
+			break;
+	}
+
+	*next_start_ptr = i;
+	return prev_blkno;
+}
diff --git a/contrib/pg_surgery/pg_surgery--1.0.sql b/contrib/pg_surgery/pg_surgery--1.0.sql
new file mode 100644
index 0000000000..2ae7f228c7
--- /dev/null
+++ b/contrib/pg_surgery/pg_surgery--1.0.sql
@@ -0,0 +1,18 @@
+/* contrib/pg_surgery/pg_surgery--1.0.sql */
+
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "CREATE EXTENSION pg_surgery" to load this file. \quit
+
+CREATE FUNCTION heap_force_kill(reloid regclass, tids tid[])
+RETURNS VOID
+AS 'MODULE_PATHNAME', 'heap_force_kill'
+LANGUAGE C STRICT;
+
+REVOKE EXECUTE ON FUNCTION heap_force_kill(regclass, tid[]) FROM PUBLIC;
+
+CREATE FUNCTION heap_force_freeze(reloid regclass, tids tid[])
+RETURNS VOID
+AS 'MODULE_PATHNAME', 'heap_force_freeze'
+LANGUAGE C STRICT;
+
+REVOKE EXECUTE ON FUNCTION heap_force_freeze(regclass, tid[]) FROM PUBLIC;
\ No newline at end of file
diff --git a/contrib/pg_surgery/pg_surgery.control b/contrib/pg_surgery/pg_surgery.control
new file mode 100644
index 0000000000..2bcdad1e3f
--- /dev/null
+++ b/contrib/pg_surgery/pg_surgery.control
@@ -0,0 +1,5 @@
+# pg_surgery extension
+comment = 'extension to perform surgery on a damaged relation'
+default_version = '1.0'
+module_pathname = '$libdir/pg_surgery'
+relocatable = true
diff --git a/contrib/pg_surgery/sql/heap_surgery.sql b/contrib/pg_surgery/sql/heap_surgery.sql
new file mode 100644
index 0000000000..8a27214e9c
--- /dev/null
+++ b/contrib/pg_surgery/sql/heap_surgery.sql
@@ -0,0 +1,91 @@
+create extension pg_surgery;
+
+-- create a normal heap table and insert some rows.
+-- note that we don't commit the transaction, so autovacuum can't interfere.
+begin;
+create table htab(a int);
+insert into htab values (100), (200), (300), (400), (500);
+
+-- test empty TID array
+select heap_force_freeze('htab'::regclass, ARRAY[]::tid[]);
+
+-- nothing should be frozen yet
+select * from htab where xmin = 2;
+
+-- freeze forcibly
+select heap_force_freeze('htab'::regclass, ARRAY['(0, 4)']::tid[]);
+
+-- now we should have one frozen tuple
+select ctid, xmax from htab where xmin = 2;
+
+-- kill forcibly
+select heap_force_kill('htab'::regclass, ARRAY['(0, 4)']::tid[]);
+
+-- should be gone now
+select * from htab where ctid = '(0, 4)';
+
+-- should now be skipped because it's already dead
+select heap_force_kill('htab'::regclass, ARRAY['(0, 4)']::tid[]);
+select heap_force_freeze('htab'::regclass, ARRAY['(0, 4)']::tid[]);
+
+-- freeze two TIDs at once while skipping an out-of-range block number
+select heap_force_freeze('htab'::regclass,
+						 ARRAY['(0, 1)', '(0, 3)', '(1, 1)']::tid[]);
+
+-- we should now have two frozen tuples
+select ctid, xmax from htab where xmin = 2;
+
+-- out-of-range TIDs should be skipped
+select heap_force_freeze('htab'::regclass, ARRAY['(0, 0)', '(0, 6)']::tid[]);
+
+rollback;
+
+-- set up a new table with a redirected line pointer
+create table htab2(a int) with (autovacuum_enabled = off);
+insert into htab2 values (100);
+update htab2 set a = 200;
+vacuum htab2;
+
+-- redirected TIDs should be skipped
+select heap_force_kill('htab2'::regclass, ARRAY['(0, 1)']::tid[]);
+
+-- now create an unused line pointer
+select ctid from htab2;
+update htab2 set a = 300;
+select ctid from htab2;
+vacuum freeze htab2;
+
+-- unused TIDs should be skipped
+select heap_force_kill('htab2'::regclass, ARRAY['(0, 2)']::tid[]);
+
+-- multidimensional TID array should be rejected
+select heap_force_kill('htab2'::regclass, ARRAY[['(0, 2)']]::tid[]);
+
+-- TID array with nulls should be rejected
+select heap_force_kill('htab2'::regclass, ARRAY[NULL]::tid[]);
+
+-- but we should be able to kill the one tuple we have
+select heap_force_kill('htab2'::regclass, ARRAY['(0, 3)']::tid[]);
+
+-- materialized view.
+-- note that we don't commit the transaction, so autovacuum can't interfere.
+begin;
+create materialized view mvw as select a from generate_series(1, 3) a;
+
+select * from mvw where xmin = 2;
+select heap_force_freeze('mvw'::regclass, ARRAY['(0, 3)']::tid[]);
+select * from mvw where xmin = 2;
+
+select heap_force_kill('mvw'::regclass, ARRAY['(0, 3)']::tid[]);
+select * from mvw where ctid = '(0, 3)';
+rollback;
+
+-- check that it fails on an unsupported relkind
+create view vw as select 1;
+select heap_force_kill('vw'::regclass, ARRAY['(0, 1)']::tid[]);
+select heap_force_freeze('vw'::regclass, ARRAY['(0, 1)']::tid[]);
+
+-- cleanup.
+drop table htab2;
+drop view vw;
+drop extension pg_surgery;