More deadlock code to check for escallation locks.

offsetof() addition to local socket size.

4 files changed, 111 insertions, 30 deletions

diff --git a/src/backend/storage/lmgr/README b/src/backend/storage/lmgr/README
index e382003f2a4..821f4a37451 100644
--- a/src/backend/storage/lmgr/README
+++ b/src/backend/storage/lmgr/README
@@ -1,4 +1,4 @@
-$Header: /cvsroot/pgsql/src/backend/storage/lmgr/README,v 1.1.1.1 1996/07/09 06:21:55 scrappy Exp $
+$Header: /cvsroot/pgsql/src/backend/storage/lmgr/README,v 1.2 1998/01/28 02:29:26 momjian Exp $
 
 This file is an attempt to save me (and future code maintainers) some
 time and a lot of headaches.  The existing lock manager code at the time
@@ -88,6 +88,12 @@ activeHolders -
     holders, summing the values of activeHolders should total to the value
     of nActive.
 
+---------------------------------------------------------------------------
 
-This is all I had the stomach for right now..... I will get back to this
-someday.	-mer 17 June 1992 12:00 am
+Locks are accessed in two ways.  Each PROC structure has a lockQueue,
+that is a circular linked list of LOCK pointers that this process holds
+or is waiting on.
+
+Second, there is a hash table that can do a lookup by combined LOCK
+address and transaction id(xid) which allows a process to see what
+type of locks it holds on that table.
diff --git a/src/backend/storage/lmgr/lock.c b/src/backend/storage/lmgr/lock.c
index a509050704a..e218a403425 100644
--- a/src/backend/storage/lmgr/lock.c
+++ b/src/backend/storage/lmgr/lock.c
@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/lock.c,v 1.23 1998/01/27 15:34:49 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/lock.c,v 1.24 1998/01/28 02:29:27 momjian Exp $
  *
  * NOTES
  *	  Outside modules can create a lock table and acquire/release
@@ -755,7 +755,7 @@ LockResolveConflicts(LOCKTAB *ltable,
 	tmpMask = 2;
 	for (i = 1; i <= nLockTypes; i++, tmpMask <<= 1)
 	{
-		if (lock->activeHolders[i] - myHolders[i])
+		if (lock->activeHolders[i] != myHolders[i])
 		{
 			bitmask |= tmpMask;
 		}
@@ -1429,14 +1429,38 @@ DeadLockCheck(SHM_QUEUE *lockQueue, LOCK *findlock, bool skip_check)
 	XIDLookupEnt *tmp = NULL;
 	SHMEM_OFFSET end = MAKE_OFFSET(lockQueue);
 	LOCK	   *lock;
-	static PROC*	checked_procs[MaxBackendId];
-	static int	nprocs;
 
+	LOCKTAB *ltable;
+	XIDLookupEnt *result,
+				item;
+	HTAB	   *xidTable;
+	bool		found;
+
+	static PROC*	checked_procs[MaxBackendId];
+	static int		nprocs;
+	static bool		MyNHolding;
+	
+	/* initialize at start of recursion */
 	if (skip_check)
 	{
-		/* initialize at start of recursion */
 		checked_procs[0] = MyProc;
 		nprocs = 1;
+
+		ltable = AllTables[1];
+		xidTable = ltable->xidHash;
+
+		MemSet(&item, 0, XID_TAGSIZE);
+		TransactionIdStore(MyProc->xid, &item.tag.xid);
+		item.tag.lock = MAKE_OFFSET(findlock);
+#if 0
+		item.tag.pid = pid;
+#endif
+	
+		if ((result = (XIDLookupEnt *)
+			  hash_search(xidTable, (Pointer) &item, HASH_FIND, &found)) && found)
+			MyNHolding = result->nHolding;
+		else
+			MyNHolding = 0;
 	}
 	
 	if (SHMQueueEmpty(lockQueue))
@@ -1469,13 +1493,7 @@ DeadLockCheck(SHM_QUEUE *lockQueue, LOCK *findlock, bool skip_check)
 		if (lock == findlock && !skip_check)
 			return true;
 
-		/*
-		 *	No sense in looking at the wait queue of the lock we are
-		 *	looking for as it is MyProc's lock entry.
-		 *  If lock == findlock, and I got here, skip_check must be true.
-		 */
-		if (lock != findlock)
-		{
+		 {
 			PROC_QUEUE  *waitQueue = &(lock->waitProcs);
 			PROC		*proc;
 			int			i;
@@ -1484,16 +1502,70 @@ DeadLockCheck(SHM_QUEUE *lockQueue, LOCK *findlock, bool skip_check)
 			proc = (PROC *) MAKE_PTR(waitQueue->links.prev);
 			for (i = 0; i < waitQueue->size; i++)
 			{
-				for (j = 0; j < nprocs; j++)
-					if (checked_procs[j] == proc)
-						break;
-				if (j >= nprocs)
+				if (proc != MyProc &&
+					lock == findlock && /* skip_check also true */
+					MyNHolding) /* I already hold some lock on it */
+				{
+					/*
+					 *	For findlock's wait queue, we are interested in
+					 *	procs who are blocked waiting for a write-lock on the
+					 *	table we are waiting on, and already hold a lock on it.
+					 *	We first check to see if there is an escalation
+					 *	deadlock, where we hold a readlock and want a
+					 *	writelock, and someone else holds readlock on
+					 *	the same table, and wants a writelock.
+					 *
+					 *	Basically, the test is, "Do we both hold some lock
+					 *	on findlock, and we are both waiting in the lock
+					 *	queue?"
+					 */
+
+					Assert(skip_check);
+					Assert(MyProc->prio == 2);
+					
+					ltable = AllTables[1];
+					xidTable = ltable->xidHash;
+
+					MemSet(&item, 0, XID_TAGSIZE);
+					TransactionIdStore(proc->xid, &item.tag.xid);
+					item.tag.lock = MAKE_OFFSET(findlock);
+#if 0
+					item.tag.pid = pid;
+#endif
+				
+					if ((result = (XIDLookupEnt *)
+						  hash_search(xidTable, (Pointer) &item, HASH_FIND, &found)) && found)
+					{
+						if (result->nHolding)
+								return true;
+					}
+				}
+				/*
+				 *	No sense in looking at the wait queue of the lock we are
+				 *	looking for.
+				 *  If lock == findlock, and I got here, skip_check must be
+				 *	true too.
+				 */
+				if (lock != findlock)
 				{
-					checked_procs[nprocs++] = proc;
-					Assert(nprocs <= MaxBackendId);
-					/* If we found a deadlock, we can stop right now */
-					if (DeadLockCheck(&(proc->lockQueue), findlock, false))
-						return true;
+					for (j = 0; j < nprocs; j++)
+						if (checked_procs[j] == proc)
+							break;
+					if (j >= nprocs && lock != findlock)
+					{
+						checked_procs[nprocs++] = proc;
+						Assert(nprocs <= MaxBackendId);
+						/*
+						 *	For non-MyProc entries, we are looking only waiters,
+						 *	not necessarily people who already hold locks and are
+						 *	waiting.
+						 *	Now we check for cases where we have two or more
+						 *	tables in a deadlock.  We do this by continuing
+						 *	to search for someone holding a lock
+						 */
+						if (DeadLockCheck(&(proc->lockQueue), findlock, false))
+							return true;
+					}
 				}
 				proc = (PROC *) MAKE_PTR(proc->links.prev);
 			}
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index 893ea41833d..1ab096ac527 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.29 1998/01/27 03:00:29 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.30 1998/01/28 02:29:29 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -46,7 +46,7 @@
  *		This is so that we can support more backends. (system-wide semaphore
  *		sets run out pretty fast.)				  -ay 4/95
  *
- * $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.29 1998/01/27 03:00:29 momjian Exp $
+ * $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.30 1998/01/28 02:29:29 momjian Exp $
  */
 #include <sys/time.h>
 #include <unistd.h>
@@ -249,7 +249,10 @@ InitProcess(IPCKey key)
 	 */
 	SpinRelease(ProcStructLock);
 
+	MyProc->pid = 0;
+#if 0
 	MyProc->pid = MyProcPid;
+#endif
 	MyProc->xid = InvalidTransactionId;
 
 	/* ----------------
diff --git a/src/include/libpq/pqcomm.h b/src/include/libpq/pqcomm.h
index b3b67f8b73b..0145f262702 100644
--- a/src/include/libpq/pqcomm.h
+++ b/src/include/libpq/pqcomm.h
@@ -6,7 +6,7 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- * $Id: pqcomm.h,v 1.21 1998/01/27 15:35:22 momjian Exp $
+ * $Id: pqcomm.h,v 1.22 1998/01/28 02:29:40 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -35,10 +35,10 @@ typedef union SockAddr {
 
 #define	UNIXSOCK_PATH(sun,port) \
 	(sprintf((sun).sun_path, "/tmp/.s.PGSQL.%d", (port)) + \
-		+ 1 + sizeof ((sun).sun_family))
+		offsetof(struct sockaddr_un, sun_path))
 /*
- *		+ 1 is for BSD-specific sizeof((sun).sun_len)
- *		We never actually set sun_len, and I can't think of a
+ *		We do this because sun_len is in BSD's struct, while others don't.
+ *		We never actually set BSD's sun_len, and I can't think of a
  *		platform-safe way of doing it, but the code still works. bjm
  */