Merge pull request #12 from Percona-Lab/keys_tde_fork

Store and fetch relation keys from fork

src/access/pg_tde_tdemap.c

@@ -10,11 +10,19 @@
  *-------------------------------------------------------------------------
  */
 
+#define TDE_FORK_DEBUG 1
+
 #include "postgres.h"
 #include "access/pg_tde_tdemap.h"
 #include "transam/pg_tde_xact_handler.h"
 #include "storage/fd.h"
 #include "utils/wait_event.h"
+#include "utils/memutils.h"
+
+#include "access/pg_tde_tdemap.h"
+
+#include <openssl/rand.h>
+#include <openssl/err.h>
 
 /*
  * Creates a relation fork file relfilenode.tde that contains the
@@ -23,40 +31,179 @@
 void
 pg_tde_create_key_fork(const RelFileLocator *newrlocator, Relation rel)
 {
-	char 	*rel_file_path;
-	char 	*key_file_path;
-	File	file = -1;
-	char enc_key[256]; /* Dummy key */
+	/* TODO: should be a user defined */ 
+	static const char *MasterKeyName = "master-key";
+
+	char		*rel_file_path;
+	char		*key_file_path;
+	File		file = -1;
+	InternalKey int_key = {0};
+	RelKeysData *data;
+	size_t 		sz;
 
 	/* We get a relation name for MAIN fork and manually append the
 	 * .tde postfix to the file name
 	 */
 	rel_file_path = relpathperm(*newrlocator, MAIN_FORKNUM);
-	key_file_path = psprintf("%s.tde", rel_file_path);
+	key_file_path = psprintf("%s."TDE_FORK_EXT, rel_file_path);
 	pfree(rel_file_path);
 
 	file = PathNameOpenFile(key_file_path, O_RDWR | O_CREAT | PG_BINARY);
 	if (file < 0)
-	{
 		ereport(FATAL,
-                 (errcode_for_file_access(),
-                  errmsg("could not open tde key file %s", key_file_path)));
-	}
-	/* TODO:
-	 * For now just write a dummy data to the file. We will write the actual
-	 * key later.
+        		(errcode_for_file_access(),
+        		errmsg("could not open tde key file \"%s\": %m",
+				  		key_file_path)));
+
+
+	if (!RAND_bytes(int_key.key, INTERNAL_KEY_LEN))
+		ereport(FATAL,
+				(errcode(ERRCODE_INTERNAL_ERROR),
+				errmsg("could not generate internal key for relation \"%s\": %s",
+                		RelationGetRelationName(rel), ERR_error_string(ERR_get_error(), NULL))));
+
+#if TDE_FORK_DEBUG
+	ereport(DEBUG2,
+		(errmsg("internal_key: %s", tde_sprint_key(&int_key))));
+#endif
+
+	data = (RelKeysData *) palloc(SizeOfRelKeysData(2));
+
+	strcpy(data->master_key_name, MasterKeyName);
+	data->internal_key[0] = int_key;
+	data->internal_keys_len = 1;
+
+	sz = SizeOfRelKeysData(data->internal_keys_len);
+	/* 
+	 * TODO: internal key(s) should be encrypted
+	 */
+	if (FileWrite(file, data, sz, 0, WAIT_EVENT_DATA_FILE_WRITE) != sz)
+    	ereport(FATAL,
+				(errcode_for_file_access(),
+                errmsg("could not write key data to file \"%s\": %m",
+                		key_file_path)));
+
+	pfree(key_file_path);
+	pfree(data);
+	FileClose(file);
+}
+
+/*
+ * Reads tde keys for the relatoin from a fork file.
+ */
+RelKeysData *
+pg_tde_get_keys_from_fork(const RelFileLocator *rlocator)
+{
+	char		*rel_file_path;
+	char		*key_file_path;
+	File		file = -1;
+	Size		sz;
+	int			nbytes;
+	RelKeysData *keys;
+
+	rel_file_path = relpathperm(*rlocator, MAIN_FORKNUM);
+	key_file_path = psprintf("%s."TDE_FORK_EXT, rel_file_path);
+	pfree(rel_file_path);
+
+	file = PathNameOpenFile(key_file_path, O_RDONLY | PG_BINARY);
+	if (file < 0)
+		ereport(FATAL,
+                (errcode_for_file_access(),
+                errmsg("could not open tde key file \"%s\": %m",
+						key_file_path)));
+
+
+	sz = (Size) FileSize(file);
+	keys = (RelKeysData *) MemoryContextAlloc(TopMemoryContext, sz);
+
+	/* 
+	 * TODO: internal key(s) should be encrypted
 	 */
-	snprintf(enc_key, sizeof(enc_key), "Percona TDE Dummy key for relation:%s", RelationGetRelationName(rel));
-	if (FileWrite(file, enc_key, sizeof(enc_key),
-                    0, WAIT_EVENT_DATA_FILE_WRITE) != sizeof(enc_key))
-         ereport(FATAL, (errcode_for_file_access(),
-                         errmsg("Could not write key data to file: %s",
-                                key_file_path)));
+	nbytes = FileRead(file, keys, sz, 0, WAIT_EVENT_DATA_FILE_READ);
+	if (nbytes < 0)
+		ereport(FATAL,
+				(errcode_for_file_access(),
+				errmsg("could not read key data file \"%s\": %m",
+						key_file_path)));
+	else if (nbytes < SizeOfRelKeysData(1) || 
+				(nbytes - SizeOfRelKeysDataHeader) % sizeof(InternalKey) != 0)
+		ereport(FATAL,
+				(errcode(ERRCODE_DATA_CORRUPTED),
+				errmsg("corrupted key data in file \"%s\"",
+						key_file_path)));
+
+#if TDE_FORK_DEBUG
+	for (Size i = 0; i < keys->internal_keys_len; i++) 
+		ereport(DEBUG2,
+			(errmsg("fork file keys: [%lu] %s: %s", i+1, keys->master_key_name, tde_sprint_key(&keys->internal_key[i]))));
+#endif
 
 	/* Register the file for delete in case transaction Aborts */
 	RegisterFileForDeletion(key_file_path, false);
 
 	pfree(key_file_path);
 	/* For now just close the key file.*/
 	FileClose(file);
+
+	return keys;
+}
+
+/* Head of the keys cache (linked list) */
+RelKeys *tde_rel_keys_map = NULL;
+
+/*
+ * Returns TDE keys for a given relation.
+ * First it looks in a cache. If nothing found in the cache, it reads data from
+ * the tde fork file and populates cache.
+ */
+RelKeysData *
+GetRelationKeys(Relation rel)
+{
+	RelKeys		*curr;
+	RelKeys		*prev;
+	RelKeys		*new;
+	RelKeysData *keys;
+
+	Oid rel_id = RelationGetRelid(rel);
+	for (curr = tde_rel_keys_map; curr != NULL; curr = curr->next)
+	{
+		if (curr->rel_id == rel_id) {
+#if TDE_FORK_DEBUG
+			ereport(DEBUG2,
+					(errmsg("TDE: cache hit, \"%s\" %s |  rel %s (%d)",
+							curr->keys->master_key_name,
+							tde_sprint_key(&curr->keys->internal_key[0]), 
+							RelationGetRelationName(rel), 
+							rel_id)));
+#endif
+			return curr->keys;
+		}
+		prev = curr;
+	}
+
+	keys = pg_tde_get_keys_from_fork(&rel->rd_locator);
+	new = (RelKeys *) MemoryContextAlloc(TopMemoryContext, sizeof(RelKeys));
+	new->rel_id = rel_id;
+	new->keys = keys;
+
+	if (tde_rel_keys_map == NULL)
+		tde_rel_keys_map = new;
+	else
+		prev->next = new;
+
+	return keys;
+}
+
+const char *
+tde_sprint_key(InternalKey *k)
+{
+	static char buf[256];
+	int 	i;
+
+	for (i = 0; i < sizeof(k->key); i++)
+		sprintf(buf+i, "%02X", k->key[i]);
+
+	sprintf(buf+i, "[%lu, %lu]", k->start_loc, k->end_loc);
+
+	return buf;
 }

src/access/pg_tdeam.c

@@ -1168,7 +1168,7 @@ pg_tde_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot
 
 	pgstat_count_pg_tde_getnext(scan->rs_base.rs_rd);
 
-	PGTdeExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
+	PGTdeExecStoreBufferHeapTuple(sscan->rs_rd, &scan->rs_ctup, slot,
 							 scan->rs_cbuf);
 	return true;
 }
@@ -1316,7 +1316,7 @@ pg_tde_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
 	 */
 	pgstat_count_pg_tde_getnext(scan->rs_base.rs_rd);
 
-	PGTdeExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
+	PGTdeExecStoreBufferHeapTuple(sscan->rs_rd, &scan->rs_ctup, slot, scan->rs_cbuf);
 	return true;
 }
 

src/access/pg_tdeam_handler.c

@@ -169,7 +169,7 @@ pg_tdeam_index_fetch_tuple(struct IndexFetchTableData *scan,
 		*call_again = !IsMVCCSnapshot(snapshot);
 
 		slot->tts_tableOid = RelationGetRelid(scan->rel);
-		PGTdeExecStoreBufferHeapTuple(&bslot->base.tupdata, slot, hscan->xs_cbuf);
+		PGTdeExecStoreBufferHeapTuple(scan->rel, &bslot->base.tupdata, slot, hscan->xs_cbuf);
 	}
 	else
 	{
@@ -201,7 +201,7 @@ pg_tdeam_fetch_row_version(Relation relation,
 	if (pg_tde_fetch(relation, snapshot, &bslot->base.tupdata, &buffer, false))
 	{
 		/* store in slot, transferring existing pin */
-		PGTdeExecStorePinnedBufferHeapTuple(&bslot->base.tupdata, slot, buffer);
+		PGTdeExecStorePinnedBufferHeapTuple(relation, &bslot->base.tupdata, slot, buffer);
 		slot->tts_tableOid = RelationGetRelid(relation);
 
 		return true;
@@ -575,7 +575,7 @@ pg_tdeam_tuple_lock(Relation relation, ItemPointer tid, Snapshot snapshot,
 	tuple->t_tableOid = slot->tts_tableOid;
 
 	/* store in slot, transferring existing pin */
-	PGTdeExecStorePinnedBufferHeapTuple(tuple, slot, buffer);
+	PGTdeExecStorePinnedBufferHeapTuple(relation, tuple, slot, buffer);
 
 	return result;
 }
@@ -1165,7 +1165,7 @@ pg_tdeam_scan_analyze_next_tuple(TableScanDesc scan, TransactionId OldestXmin,
 
 		if (sample_it)
 		{
-			PGTdeExecStoreBufferHeapTuple(targtuple, slot, hscan->rs_cbuf);
+			PGTdeExecStoreBufferHeapTuple(scan->rs_rd, targtuple, slot, hscan->rs_cbuf);
 			hscan->rs_cindex++;
 
 			/* note that we leave the buffer locked here! */
@@ -1645,7 +1645,7 @@ pg_tdeam_index_build_range_scan(Relation heapRelation,
 		MemoryContextReset(econtext->ecxt_per_tuple_memory);
 
 		/* Set up for predicate or expression evaluation */
-		PGTdeExecStoreBufferHeapTuple(heapTuple, slot, hscan->rs_cbuf);
+		PGTdeExecStoreBufferHeapTuple(heapRelation, heapTuple, slot, hscan->rs_cbuf);
 
 		/*
 		 * In a partial index, discard tuples that don't satisfy the
@@ -2279,7 +2279,7 @@ pg_tdeam_scan_bitmap_next_tuple(TableScanDesc scan,
 	 * Set up the result slot to point to this tuple.  Note that the slot
 	 * acquires a pin on the buffer.
 	 */
-	PGTdeExecStoreBufferHeapTuple(&hscan->rs_ctup,
+	PGTdeExecStoreBufferHeapTuple(scan->rs_rd, &hscan->rs_ctup,
 							 slot,
 							 hscan->rs_cbuf);
 
@@ -2433,7 +2433,7 @@ pg_tdeam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
 			if (!pagemode)
 				LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
 
-			PGTdeExecStoreBufferHeapTuple(tuple, slot, hscan->rs_cbuf);
+			PGTdeExecStoreBufferHeapTuple(scan->rs_rd, tuple, slot, hscan->rs_cbuf);
 
 			/* Count successfully-fetched tuples as heap fetches */
 			pgstat_count_pg_tde_getnext(scan->rs_rd);

src/encryption/enc_tuple.c

@@ -4,6 +4,7 @@
 #include "postgres.h"
 #include "utils/memutils.h"
 
+#include "access/pg_tde_tdemap.h"
 #include "encryption/enc_tuple.h"
 #include "encryption/enc_aes.h"
 #include "storage/bufmgr.h"
@@ -130,21 +131,27 @@ PGTdePageAddItemExtended(Oid oid,
 }
 
 TupleTableSlot *
-PGTdeExecStoreBufferHeapTuple(HeapTuple tuple, TupleTableSlot *slot, Buffer buffer)
+PGTdeExecStoreBufferHeapTuple(Relation rel, HeapTuple tuple, TupleTableSlot *slot, Buffer buffer)
 {
 	Page pageHeader;
 
 	pageHeader = BufferGetPage(buffer);
 	PGTdeDecryptTupData(BufferGetBlockNumber(buffer), pageHeader, tuple);
 
+	/* TODO: use the keys in approprate place */
+	RelKeysData *keys = GetRelationKeys(rel);
+
 	return  ExecStoreBufferHeapTuple(tuple, slot, buffer);
 }
 
 TupleTableSlot *
-PGTdeExecStorePinnedBufferHeapTuple(HeapTuple tuple, TupleTableSlot *slot, Buffer buffer)
+PGTdeExecStorePinnedBufferHeapTuple(Relation rel, HeapTuple tuple, TupleTableSlot *slot, Buffer buffer)
 {
 	Page pageHeader;
 
+	/* TODO: use the keys in approprate place */
+	RelKeysData *keys = GetRelationKeys(rel);
+
 	pageHeader = BufferGetPage(buffer);
 	PGTdeDecryptTupData(BufferGetBlockNumber(buffer), pageHeader, tuple);
 

src/include/access/pg_tde_tdemap.h

@@ -11,6 +11,47 @@
 #include "utils/rel.h"
 #include "storage/relfilelocator.h"
 
-extern void pg_tde_create_key_fork(const RelFileLocator *newrlocator, Relation rel);
+#define TDE_FORK_EXT "tde"
+
+#define INTERNAL_KEY_LEN 16
+typedef struct InternalKey
+{
+    uint8   key[INTERNAL_KEY_LEN];
+    /* a start and end range of the key
+     * (start_loc == 0 && end_loc == 0) -> the key is for the whole file
+     */
+    Size    start_loc; 
+    Size    end_loc;
+} InternalKey;
+
+#define MASTER_KEY_NAME_LEN 256
+typedef struct RelKeysData
+{
+    char        master_key_name[MASTER_KEY_NAME_LEN];
+    Size        internal_keys_len;
+    InternalKey internal_key[FLEXIBLE_ARRAY_MEMBER];
+} RelKeysData;
+
+#define SizeOfRelKeysDataHeader offsetof(RelKeysData, internal_key)
+#define SizeOfRelKeysData(keys_num) \
+    (SizeOfRelKeysDataHeader + sizeof(InternalKey) * keys_num)
 
+/* Relation keys cache.
+ * 
+ * TODO: For now it is just a linked list. Data can only be added w/o any
+ * ability to remove or change it. Also consider usage of more efficient data
+ * struct (hash map) in the shared memory(?) - currently allocated in the
+ * TopMemoryContext of the process. 
+ */
+typedef struct RelKeys
+{
+    Oid     rel_id;
+    RelKeysData    *keys;
+    struct RelKeys *next;
+} RelKeys;
+
+extern void pg_tde_create_key_fork(const RelFileLocator *newrlocator, Relation rel);
+extern RelKeysData *pg_tde_get_keys_from_fork(const RelFileLocator *rlocator);
+extern RelKeysData *GetRelationKeys(Relation rel);
+const char * tde_sprint_key(InternalKey *k);
 #endif                            /* PG_TDE_MAP_H */