Update ptrack to version ptrack_v1.5. Fix several race conditions which could lead to incorrect ptrack backups

Author: alubennikova

src/backend/access/heap/ptrack.c

@@ -3,10 +3,36 @@
  * ptrack.c
  *	  bitmap for tracking updates of relation's pages
  *
- * TODO Add description
- *
  * IDENTIFICATION
  *	  src/backend/access/heap/ptrack.c
+ *
+ * INTERFACE ROUTINES (PostgreSQL side)
+ *
+ *		ptrack_add_block - 		set a bit to track dirtied page
+ *		ptrack_add_block_redo - set a bit to track recovered page
+ *		create_ptrack_init_file - create PTRACK_INIT_FILE
+ *								  in the given database directory
+ *
+ * EXTERNAL INTERFACE ROUTINES (Backup utility side)
+ *		pg_ptrack_version() - Returns PTRACK version currently in use.
+ *   	pg_ptrack_control_lsn() - Gets LSN from ptrack_control file.
+ *   	pg_ptrack_clear() - Resets bits in all PTRACK files.
+ *   					    This function must be called for each database in the cluster.
+ *   	pg_ptrack_get_and_clear(Oid tablespace_oid, Oid table_oid) -
+ *   					Reads a PTRACK file for the given relation and resets it.
+ *   					Returns the PTRACK content as bytea. It is essential to receive and clear the map
+ *   					atomically in order to avoid losing PTRACK bits because of race conditions.
+ *   					(Imagine that your backup tool reads the map, then some blocks of the relation are
+ *   					updated and the ptrack bits are set, after that the backup tool cleans up the map
+ *    					and resets ptrack_clear_lsn. So, we may lose some of the updates).
+ *    					This function must be called for each database in the cluster.
+ *    	pg_ptrack_init_get_and_clear(Oid db_oid, Oid tablespace_oid) -
+ *    					Checks whether PTRACK_INIT_FILE exists in the given database and deletes it.
+ *    					Returns true if the file was found. This function is analogous to
+ *    					pg_ptrack_get_and_clear(), but it handles directory-level changes
+ *    					(i.e. CREATE DATABASE, ALTER DATABASE SET TABLESPACE).
+ *    					This function must be called for each database in the cluster.
+ *
  */
 
 #include "postgres.h"
@@ -31,6 +57,8 @@
 #include "utils/relfilenodemap.h"
 #include "utils/builtins.h"
 #include "utils/pg_lsn.h"
+#include "utils/lsyscache.h"
+#include "nodes/makefuncs.h"
 #include <unistd.h>
 #include <sys/stat.h>
 
@@ -53,15 +81,16 @@ bool ptrack_enable = false;
 
 static Buffer ptrack_readbuf(Relation rel, BlockNumber blkno, bool extend);
 static void ptrack_extend(Relation rel, BlockNumber nvmblocks);
+static void ptrack_set(BlockNumber heapBlk, Buffer ptrackBuf);
+
 void SetPtrackClearLSN(bool set_invalid);
-Datum pg_ptrack_test(PG_FUNCTION_ARGS);
+
 Datum pg_ptrack_clear(PG_FUNCTION_ARGS);
 Datum pg_ptrack_get_and_clear(PG_FUNCTION_ARGS);
 Datum pg_ptrack_get_and_clear_db(PG_FUNCTION_ARGS);
 Datum pg_ptrack_control_lsn(PG_FUNCTION_ARGS);
 
-void create_ptrack_init_file(char *dest_dir);
-void drop_ptrack_init_file(char *dest_dir);
+static void drop_ptrack_init_file(char *dest_dir);
 
 /*
  * Mark tracked memory block during recovery.
@@ -83,48 +112,43 @@ ptrack_add_block(Relation rel, BlockNumber heapBlk)
 {
 	Buffer ptrackbuf = InvalidBuffer;
 
+	/*
+	 * Do not track changes for unlogged and temp relations,
+	 * since we are not going to backup them anyway.
+	 */
+	if (rel->rd_rel->relpersistence != RELPERSISTENCE_PERMANENT)
+		return;
+
 	if (ptrack_enable)
 	{
-		ptrack_pin(rel, heapBlk, &ptrackbuf);
+		BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
+		ptrackbuf =  ptrack_readbuf(rel, mapBlock, true);
 		ptrack_set(heapBlk, ptrackbuf);
 		ReleaseBuffer(ptrackbuf);
 	}
 }
 
-/* Pin a ptrack map page for setting a bit */
-void
-ptrack_pin(Relation rel, BlockNumber heapBlk, Buffer *buf)
-{
-	BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
-
-	*buf = ptrack_readbuf(rel, mapBlock, true);
-}
-
-/* Set one bit to buffer  */
-void
+/* Set one bit to buffer */
+static void
 ptrack_set(BlockNumber heapBlk, Buffer ptrackBuf)
 {
-	BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
 	uint32		mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
 	uint8		mapOffset = HEAPBLK_TO_MAPBIT(heapBlk);
 	Page		page;
 	char	   *map;
 
-	/* Check that we have the right ptrack page pinned */
-	if (!BufferIsValid(ptrackBuf)
-		|| BufferGetBlockNumber(ptrackBuf) != mapBlock)
-		elog(ERROR, "wrong ptrack buffer passed to ptrack_set");
-
 	page = BufferGetPage(ptrackBuf);
 	map = PageGetContents(page);
 	LockBuffer(ptrackBuf, BUFFER_LOCK_SHARE);
 
+	/* Check if the bit already set */
 	if (!(map[mapByte] & (1 << mapOffset)))
 	{
 		/* Bad luck. Take an exclusive lock now after unlock share.*/
 		LockBuffer(ptrackBuf, BUFFER_LOCK_UNLOCK);
 		LockBuffer(ptrackBuf, BUFFER_LOCK_EXCLUSIVE);
 
+		/* The bit could have been set concurrently */
 		if (!(map[mapByte] & (1 << mapOffset)))
 		{
 			START_CRIT_SECTION();
@@ -133,8 +157,8 @@ ptrack_set(BlockNumber heapBlk, Buffer ptrackBuf)
 			MarkBufferDirty(ptrackBuf);
 
 			/*
-			 * We don't have Xlog entry for ptrack, but update pages
-			 * on recovery.
+			 * We don't have Xlog entry for ptrack, update pages
+			 * on recovery instead.
 			 */
 			END_CRIT_SECTION();
 		}
@@ -261,6 +285,8 @@ ptrack_extend(Relation rel, BlockNumber ptrack_nblocks)
 	pfree(pg);
 }
 
+
+
 /* Clear all blocks of relation's ptrack map */
 static void
 ptrack_clear_one_rel(Oid relid)
@@ -410,6 +436,7 @@ ptrack_get_and_clear(Oid tablespace_oid, Oid table_oid)
 	return result;
 }
 
+
 /*
  * Reset LSN in ptrack_control file.
  * If server started with ptrack_enable = off,
@@ -535,7 +562,11 @@ pg_ptrack_get_and_clear_db(PG_FUNCTION_ARGS)
 	}
 }
 
-/* create empty ptrack_init_file */
+/*
+ * Create PTRACK_INIT_FILE which allows to track changes
+ * on directory level made by operations which do not go
+ * through Shared Buffers.
+ */
 void
 create_ptrack_init_file(char *dest_dir)
 {
@@ -559,6 +590,7 @@ create_ptrack_init_file(char *dest_dir)
 				 errmsg("could not close file \"%s\": %m", ptrack_init_file_path)));
 }
 
+/* Delete PTRACK_INIT_FILE */
 void
 drop_ptrack_init_file(char *dest_dir)
 {
@@ -616,3 +648,138 @@ pg_ptrack_control_lsn(PG_FUNCTION_ARGS)
 
 	PG_RETURN_LSN(lsn);
 }
+
+
+static RangeVar *
+makeRangeVarFromRelid(Oid relid)
+{
+	char *relname = get_rel_name(relid);
+	char *nspname = get_namespace_name(get_rel_namespace(relid));
+
+	if (relname == NULL)
+		elog(ERROR, "Invalid relation OID");
+
+	return makeRangeVar(nspname, relname, -1);
+}
+
+PG_FUNCTION_INFO_V1(pg_ptrack_get_block);
+
+Datum
+pg_ptrack_get_block(PG_FUNCTION_ARGS)
+{
+	Oid			tablespace_oid = PG_GETARG_OID(0);
+	Oid			relfilenode = PG_GETARG_OID(1);
+	uint32		blkno = PG_GETARG_UINT32(2);
+	bytea	   *raw_page;
+	RangeVar   *relrv;
+	char	   *raw_page_data;
+	Buffer		buf;
+	Oid relationOid = RelidByRelfilenode(tablespace_oid, relfilenode);
+	Relation rel;
+
+	if (relationOid == InvalidOid)
+		elog(ERROR, "InvalidOid");
+
+	elog(DEBUG1, "pg_ptrack_get_block(%i, %i, %u). relationOid %i",
+		 tablespace_oid, relfilenode, blkno, relationOid);
+
+	rel = relation_open(relationOid, AccessShareLock);
+
+	if (rel == InvalidRelation)
+		elog(ERROR, "InvalidRelation");
+
+
+	/* Check that this relation has storage */
+	if (rel->rd_rel->relkind == RELKIND_VIEW)
+		ereport(ERROR,
+				(errcode(ERRCODE_WRONG_OBJECT_TYPE),
+				 errmsg("cannot get raw page from view \"%s\"",
+						RelationGetRelationName(rel))));
+	if (rel->rd_rel->relkind == RELKIND_COMPOSITE_TYPE)
+		ereport(ERROR,
+				(errcode(ERRCODE_WRONG_OBJECT_TYPE),
+				 errmsg("cannot get raw page from composite type \"%s\"",
+						RelationGetRelationName(rel))));
+	if (rel->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
+		ereport(ERROR,
+				(errcode(ERRCODE_WRONG_OBJECT_TYPE),
+				 errmsg("cannot get raw page from foreign table \"%s\"",
+						RelationGetRelationName(rel))));
+
+	/*
+	 * Reject attempts to read non-local temporary relations; we would be
+	 * likely to get wrong data since we have no visibility into the owning
+	 * session's local buffers.
+	 */
+	if (RELATION_IS_OTHER_TEMP(rel))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("cannot access temporary tables of other sessions")));
+
+	/* Initialize buffer to copy to */
+	raw_page = (bytea *) palloc(BLCKSZ + VARHDRSZ);
+	SET_VARSIZE(raw_page, BLCKSZ + VARHDRSZ);
+	raw_page_data = VARDATA(raw_page);
+
+	/* Take a verbatim copy of the page */
+
+	buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, NULL);
+	LockBuffer(buf, BUFFER_LOCK_SHARE);
+
+	memcpy(raw_page_data, BufferGetPage(buf), BLCKSZ);
+
+	LockBuffer(buf, BUFFER_LOCK_UNLOCK);
+	ReleaseBuffer(buf);
+
+	relation_close(rel, AccessShareLock);
+
+	PG_RETURN_BYTEA_P(raw_page);
+}
+
+PG_FUNCTION_INFO_V1(pg_ptrack_get_block_2);
+
+Datum
+pg_ptrack_get_block_2(PG_FUNCTION_ARGS)
+{
+	Oid			tablespace_oid = PG_GETARG_OID(0);
+	Oid			db_oid = PG_GETARG_OID(1);
+	Oid			relfilenode = PG_GETARG_OID(2);
+	BlockNumber blkno = PG_GETARG_UINT32(3);
+	bytea	   *raw_page;
+	char	   *raw_page_data;
+	Buffer		buf;
+	RelFileNode rnode;
+	SMgrRelation smgr;
+	BlockNumber nblocks;
+
+	rnode.dbNode = db_oid;
+	rnode.spcNode = tablespace_oid;
+	rnode.relNode = relfilenode;
+
+	elog(DEBUG1, "pg_ptrack_get_block_2(%i, %i, %i, %u)",
+		 tablespace_oid, db_oid, relfilenode, blkno);
+	smgr = smgropen(rnode, InvalidBackendId);
+	nblocks = smgrnblocks(smgr, MAIN_FORKNUM);
+
+	if (blkno >= nblocks)
+		PG_RETURN_NULL();
+
+	/* Initialize buffer to copy to */
+	raw_page = (bytea *) palloc0(BLCKSZ + VARHDRSZ);
+	SET_VARSIZE(raw_page, BLCKSZ + VARHDRSZ);
+	raw_page_data = VARDATA(raw_page);
+
+	buf = ReadBufferWithoutRelcache(rnode, MAIN_FORKNUM, blkno, RBM_NORMAL, NULL);
+
+	if (buf == InvalidBuffer)
+		elog(ERROR, "Block is not found in the buffer cache");
+
+	LockBuffer(buf, BUFFER_LOCK_SHARE);
+
+	memcpy(raw_page_data, BufferGetPage(buf), BLCKSZ);
+
+	LockBuffer(buf, BUFFER_LOCK_UNLOCK);
+	ReleaseBuffer(buf);
+
+	PG_RETURN_BYTEA_P(raw_page);
+}

src/include/access/ptrack.h

@@ -8,9 +8,9 @@
 #include "utils/relcache.h"
 
 /* Ptrack version as a string */
-#define PTRACK_VERSION "1.3"
+#define PTRACK_VERSION "1.5"
 /* Ptrack version as a number */
-#define PTRACK_VERSION_NUM 103
+#define PTRACK_VERSION_NUM 105
 
 /* Number of bits allocated for each heap block. */
 #define PTRACK_BITS_PER_HEAPBLOCK 1
@@ -21,10 +21,7 @@ extern PGDLLIMPORT bool ptrack_enable;
 
 extern void ptrack_add_block(Relation rel, BlockNumber heapBlk);
 extern void ptrack_add_block_redo(RelFileNode rnode, BlockNumber heapBlk);
-extern void ptrack_pin(Relation rel, BlockNumber heapBlk, Buffer *buf);
-extern void ptrack_set(BlockNumber heapBlk, Buffer ptrackBuf);
 extern void create_ptrack_init_file(char *dest_dir);
-extern void drop_ptrack_init_file(char *dest_dir);
 
 extern void ptrack_clear(void);
 extern bytea *ptrack_get_and_clear(Oid tablespace_oid, Oid table_oid);

src/include/catalog/pg_proc.h

@@ -5507,16 +5507,20 @@ DESCR("list of files in the WAL directory");
 #define PROPARALLEL_UNSAFE		'u' /* banned while in parallel mode */
 
 /* ptrack related functions*/
-DATA(insert OID = 6016 ( pg_ptrack_clear		PGNSP PGUID 12 1 0 0 0 f f f f t f v u 0 0 2278 "" _null_ _null_ _null_ _null_ _null_ pg_ptrack_clear _null_ _null_ _null_ ));
+DATA(insert OID = 6016 ( pg_ptrack_clear	PGNSP PGUID 12 1 0 0 0 f f f f t f v u 0 0 2278 "" _null_ _null_ _null_ _null_ _null_ pg_ptrack_clear _null_ _null_ _null_ ));
 DESCR("clear ptrack fork files");
-DATA(insert OID = 6018 ( pg_ptrack_get_and_clear		PGNSP PGUID 12 1 0 0 0 f f f f t f v s 2 0 17 "26 26" _null_ _null_ _null_ _null_ _null_ pg_ptrack_get_and_clear _null_ _null_ _null_ ));
+DATA(insert OID = 6018 ( pg_ptrack_get_and_clear	PGNSP PGUID 12 1 0 0 0 f f f f t f v s 2 0 17 "26 26" _null_ _null_ _null_ _null_ _null_ pg_ptrack_get_and_clear _null_ _null_ _null_ ));
 DESCR("get ptrack file as bytea and clear it");
 DATA(insert OID = 6022 ( pg_ptrack_get_and_clear_db		PGNSP PGUID 12 1 0 0 0 f f f f t f v s 2 0 16 "26 26" _null_ _null_ _null_ _null_ _null_ pg_ptrack_get_and_clear_db _null_ _null_ _null_ ));
 DESCR("check if ptrack_init_file exists in the given database");
 DATA(insert OID =  6021 ( ptrack_version	PGNSP PGUID 12 1 0 0 0 f f f f t f s s 0 0 25 "" _null_ _null_ _null_ _null_ _null_ ptrack_version _null_ _null_ _null_ ));
 DESCR("Ptrack version string");
 DATA(insert OID =  6023 ( pg_ptrack_control_lsn	PGNSP PGUID 12 1 0 0 0 f f f f t f s s 0 0 3220 "" _null_ _null_ _null_ _null_ _null_ pg_ptrack_control_lsn _null_ _null_ _null_ ));
 DESCR("read LSN from ptrack_control file");
+DATA(insert OID =  6024 ( pg_ptrack_get_block	PGNSP PGUID 12 1 0 0 0 f f f f t f s s 3 0 17 "26 26 20" _null_ _null_ _null_ _null_ _null_ pg_ptrack_get_block _null_ _null_ _null_ ));
+DESCR("get one block of the relation");
+DATA(insert OID =  6025 ( pg_ptrack_get_block_2	PGNSP PGUID 12 1 0 0 0 f f f f t f s s 4 0 17 "26 26 26 20" _null_ _null_ _null_ _null_ _null_ pg_ptrack_get_block_2 _null_ _null_ _null_ ));
+DESCR("get one block of the relation");
 
 /*
  * Symbolic values for proargmodes column.  Note that these must agree with