r1085: Now it's had some proper user testing, merge in the deferred open fix. I'm

still doing more testing, but it fixes a behaviour that we've been wrong
on ever since the start of Samba.
Jeremy.
(This used to be commit 894cc6d16296b934c112786eec896846156aee5d)

16 files changed, 974 insertions, 83 deletions

diff --git a/source3/client/client.c b/source3/client/client.c
index a0470315f82..63d73c2d4cf 100644
--- a/source3/client/client.c
+++ b/source3/client/client.c
@@ -1571,7 +1571,7 @@ static int cmd_open(void)
 	}
 	pstrcat(mask,buf);
 
-	cli_open(cli, mask, O_RDWR, DENY_ALL);
+	cli_nt_create(cli, mask, FILE_READ_DATA);
 
 	return 0;
 }
diff --git a/source3/include/local.h b/source3/include/local.h
index 540365047a2..ee8d6725535 100644
--- a/source3/include/local.h
+++ b/source3/include/local.h
@@ -230,4 +230,7 @@
 /* size of listen() backlog in smbd */
 #define SMBD_LISTEN_BACKLOG 50
 
+/* Number of microseconds to wait before a sharing violation. */
+#define SHARING_VIOLATION_USEC_WAIT 950000
+
 #endif
diff --git a/source3/include/smb.h b/source3/include/smb.h
index 54a69d1433a..9100701e212 100644
--- a/source3/include/smb.h
+++ b/source3/include/smb.h
@@ -584,6 +584,24 @@ struct interface
 	struct in_addr nmask;
 };
 
+/* struct used by share mode violation error processing */
+typedef struct {
+	pid_t pid;
+	uint16 mid;
+	struct timeval time;
+	SMB_DEV_T dev;
+	SMB_INO_T inode;
+	uint16 port;
+} deferred_open_entry;
+
+/* Internal message queue for deferred opens. */
+struct pending_message_list {
+	struct pending_message_list *next, *prev;
+	struct timeval msg_time; /* The timeout time */
+	DATA_BLOB buf;
+	DATA_BLOB private_data;
+};
+
 /* struct returned by get_share_modes */
 typedef struct {
 	pid_t pid;
@@ -663,28 +681,14 @@ struct locking_key {
 	SMB_INO_T inode;
 };
 
-struct locking_data {
-	union {
-		int num_share_mode_entries;
-		share_mode_entry dummy; /* Needed for alignment. */
-	} u;
-	/* the following two entries are implicit 
-	   share_mode_entry modes[num_share_mode_entries];
-           char file_name[];
-	*/
-};
-
-
 /* the following are used by loadparm for option lists */
-typedef enum
-{
-  P_BOOL,P_BOOLREV,P_CHAR,P_INTEGER,P_OCTAL,P_LIST,
-  P_STRING,P_USTRING,P_GSTRING,P_UGSTRING,P_ENUM,P_SEP
+typedef enum {
+	P_BOOL,P_BOOLREV,P_CHAR,P_INTEGER,P_OCTAL,P_LIST,
+	P_STRING,P_USTRING,P_GSTRING,P_UGSTRING,P_ENUM,P_SEP
 } parm_type;
 
-typedef enum
-{
-  P_LOCAL,P_GLOBAL,P_SEPARATOR,P_NONE
+typedef enum {
+	P_LOCAL,P_GLOBAL,P_SEPARATOR,P_NONE
 } parm_class;
 
 /* passed to br lock code */
@@ -1410,6 +1414,7 @@ extern int chain_size;
 #define EXCLUSIVE_OPLOCK 1
 #define BATCH_OPLOCK 2
 #define LEVEL_II_OPLOCK 4
+#define INTERNAL_OPEN_ONLY 8
 
 #define EXCLUSIVE_OPLOCK_TYPE(lck) ((lck) & (EXCLUSIVE_OPLOCK|BATCH_OPLOCK))
 #define BATCH_OPLOCK_TYPE(lck) ((lck) & BATCH_OPLOCK)
@@ -1461,6 +1466,25 @@ extern int chain_size;
 #define LEVEL_II_OPLOCK_BREAK_CMD 0x3
 #define ASYNC_LEVEL_II_OPLOCK_BREAK_CMD 0x4
 
+/* Add the "deferred open" message. */
+#define RETRY_DEFERRED_OPEN_CMD 0x5
+
+/*
+ * And the message format for it. Keep the same message length.
+ *
+ *  0     2       2+pid   2+pid+dev 2+pid+dev+ino
+ *  +----+--------+-------+--------+---------+
+ *  | cmd| pid    | dev   |  inode | mid     |
+ *  +----+--------+-------+--------+---------+
+ */
+
+#define DEFERRED_OPEN_CMD_OFFSET 0
+#define DEFERRED_OPEN_PID_OFFSET 2 /* pid we're *sending* from. */
+#define DEFERRED_OPEN_DEV_OFFSET (DEFERRED_OPEN_PID_OFFSET + sizeof(pid_t))
+#define DEFERRED_OPEN_INODE_OFFSET (DEFERRED_OPEN_DEV_OFFSET + sizeof(SMB_DEV_T))
+#define DEFERRED_OPEN_MID_OFFSET (DEFERRED_OPEN_INODE_OFFSET + sizeof(SMB_INO_T))
+#define DEFERRED_OPEN_MSG_LEN OPLOCK_BREAK_MSG_LEN
+
 /*
  * Capabilities abstracted for different systems.
  */
diff --git a/source3/lib/time.c b/source3/lib/time.c
index faca2cba879..e63e0b29659 100644
--- a/source3/lib/time.c
+++ b/source3/lib/time.c
@@ -754,3 +754,9 @@ BOOL nt_time_is_zero(NTTIME *nt)
 		return True;
 	return False;
 }
+
+SMB_BIG_INT usec_time_diff(struct timeval *larget, struct timeval *smallt)
+{
+	SMB_BIG_INT sec_diff = larget->tv_sec - smallt->tv_sec;
+	return (sec_diff * 1000000) + (SMB_BIG_INT)(larget->tv_usec - smallt->tv_usec);
+}
diff --git a/source3/locking/locking.c b/source3/locking/locking.c
index fd03a25940f..8f53b55fc54 100644
--- a/source3/locking/locking.c
+++ b/source3/locking/locking.c
@@ -40,6 +40,17 @@ uint16 global_smbpid;
 /* the locking database handle */
 static TDB_CONTEXT *tdb;
 
+struct locking_data {
+        union {
+                int num_share_mode_entries;
+                share_mode_entry dummy; /* Needed for alignment. */
+        } u;
+        /* the following two entries are implicit
+           share_mode_entry modes[num_share_mode_entries];
+           char file_name[];
+        */
+};
+
 /****************************************************************************
  Debugging aid :-).
 ****************************************************************************/
@@ -432,6 +443,7 @@ int get_share_modes(connection_struct *conn,
 	data = (struct locking_data *)dbuf.dptr;
 	num_share_modes = data->u.num_share_mode_entries;
 	if(num_share_modes) {
+		pstring fname;
 		int i;
 		int del_count = 0;
 
@@ -443,6 +455,9 @@ int get_share_modes(connection_struct *conn,
 			return 0;
 		}
 
+		/* Save off the associated filename. */
+		pstrcpy(fname, dbuf.dptr + sizeof(*data) + num_share_modes * sizeof(share_mode_entry));
+
 		/*
 		 * Ensure that each entry has a real process attached.
 		 */
@@ -467,17 +482,28 @@ int get_share_modes(connection_struct *conn,
 		if (del_count) {
 			data->u.num_share_mode_entries = num_share_modes;
 			
-			if (num_share_modes)
+			if (num_share_modes) {
 				memcpy(dbuf.dptr + sizeof(*data), shares,
 						num_share_modes * sizeof(share_mode_entry));
+				/* Append the filename. */
+				pstrcpy(dbuf.dptr + sizeof(*data) + num_share_modes * sizeof(share_mode_entry), fname);
+			}
 
 			/* The record has shrunk a bit */
 			dbuf.dsize -= del_count * sizeof(share_mode_entry);
 
-			if (tdb_store(tdb, key, dbuf, TDB_REPLACE) == -1) {
-				SAFE_FREE(shares);
-				SAFE_FREE(dbuf.dptr);
-				return 0;
+			if (data->u.num_share_mode_entries == 0) {
+				if (tdb_delete(tdb, key) == -1) {
+					SAFE_FREE(shares);
+					SAFE_FREE(dbuf.dptr);
+					return 0;
+				}
+			} else {
+				if (tdb_store(tdb, key, dbuf, TDB_REPLACE) == -1) {
+					SAFE_FREE(shares);
+					SAFE_FREE(dbuf.dptr);
+					return 0;
+				}
 			}
 		}
 	}
@@ -847,6 +873,358 @@ BOOL modify_delete_flag( SMB_DEV_T dev, SMB_INO_T inode, BOOL delete_on_close)
 	return True;
 }
 
+/*******************************************************************
+ Print out a deferred open entry.
+********************************************************************/
+
+char *deferred_open_str(int num, deferred_open_entry *e)
+{
+	static pstring de_str;
+
+	slprintf(de_str, sizeof(de_str)-1, "deferred_open_entry[%d]: \
+pid = %lu, mid = %u, dev = 0x%x, inode = %.0f, port = %u, time = [%u.%06u]",
+		num, (unsigned long)e->pid, (unsigned int)e->mid, (unsigned int)e->dev, (double)e->inode,
+		(unsigned int)e->port,
+		(unsigned int)e->time.tv_sec, (unsigned int)e->time.tv_usec );
+
+	return de_str;
+}
+
+/* Internal data structures for deferred opens... */
+
+struct de_locking_key {
+	char name[4];
+	SMB_DEV_T dev;
+	SMB_INO_T inode;
+};
+
+struct deferred_open_data {
+        union {
+                int num_deferred_open_entries;
+                deferred_open_entry dummy; /* Needed for alignment. */
+        } u;
+        /* the following two entries are implicit
+           deferred_open_entry de_entries[num_deferred_open_entries];
+           char file_name[];
+        */
+};
+
+/*******************************************************************
+ Print out a deferred open table.
+********************************************************************/
+
+static void print_deferred_open_table(struct deferred_open_data *data)
+{
+	int num_de_entries = data->u.num_deferred_open_entries;
+	deferred_open_entry *de_entries = (deferred_open_entry *)(data + 1);
+	int i;
+
+	for (i = 0; i < num_de_entries; i++) {
+		deferred_open_entry *entry_p = &de_entries[i];
+		DEBUG(10,("print_deferred_open_table: %s\n", deferred_open_str(i, entry_p) ));
+	}
+}
+
+
+/*******************************************************************
+ Form a static deferred open locking key for a dev/inode pair.
+******************************************************************/
+
+static TDB_DATA deferred_open_locking_key(SMB_DEV_T dev, SMB_INO_T inode)
+{
+	static struct de_locking_key key;
+	TDB_DATA kbuf;
+
+	memset(&key, '\0', sizeof(key));
+	memcpy(&key.name[0], "DOE", 4);
+	key.dev = dev;
+	key.inode = inode;
+	kbuf.dptr = (char *)&key;
+	kbuf.dsize = sizeof(key);
+	return kbuf;
+}
+
+/*******************************************************************
+ Get all deferred open entries for a dev/inode pair.
+********************************************************************/
+
+int get_deferred_opens(connection_struct *conn, 
+		    SMB_DEV_T dev, SMB_INO_T inode, 
+		    deferred_open_entry **pp_de_entries)
+{
+	TDB_DATA dbuf;
+	struct deferred_open_data *data;
+	int num_de_entries;
+	deferred_open_entry *de_entries = NULL;
+	TDB_DATA key = deferred_open_locking_key(dev, inode);
+
+	*pp_de_entries = NULL;
+
+	dbuf = tdb_fetch(tdb, key);
+	if (!dbuf.dptr)
+		return 0;
+
+	data = (struct deferred_open_data *)dbuf.dptr;
+	num_de_entries = data->u.num_deferred_open_entries;
+	if(num_de_entries) {
+		pstring fname;
+		int i;
+		int del_count = 0;
+
+		de_entries = (deferred_open_entry *)memdup(dbuf.dptr + sizeof(*data),	
+						num_de_entries * sizeof(deferred_open_entry));
+
+		if (!de_entries) {
+			SAFE_FREE(dbuf.dptr);
+			return 0;
+		}
+
+		/* Save off the associated filename. */
+		pstrcpy(fname, dbuf.dptr + sizeof(*data) + num_de_entries * sizeof(deferred_open_entry));
+
+		/*
+		 * Ensure that each entry has a real process attached.
+		 */
+
+		for (i = 0; i < num_de_entries; ) {
+			deferred_open_entry *entry_p = &de_entries[i];
+			if (process_exists(entry_p->pid)) {
+				DEBUG(10,("get_deferred_opens: %s\n", deferred_open_str(i, entry_p) ));
+				i++;
+			} else {
+				DEBUG(10,("get_deferred_opens: deleted %s\n", deferred_open_str(i, entry_p) ));
+				if (num_de_entries - i - 1 > 0) {
+					memcpy( &de_entries[i], &de_entries[i+1],
+						sizeof(deferred_open_entry) * (num_de_entries - i - 1));
+				}
+				num_de_entries--;
+				del_count++;
+			}
+		}
+
+		/* Did we delete any ? If so, re-store in tdb. */
+		if (del_count) {
+			data->u.num_deferred_open_entries = num_de_entries;
+			
+			if (num_de_entries) {
+				memcpy(dbuf.dptr + sizeof(*data), de_entries,
+						num_de_entries * sizeof(deferred_open_entry));
+				/* Append the filename. */
+				pstrcpy(dbuf.dptr + sizeof(*data) + num_de_entries * sizeof(deferred_open_entry), fname);
+			}
+
+			/* The record has shrunk a bit */
+			dbuf.dsize -= del_count * sizeof(deferred_open_entry);
+
+			if (data->u.num_deferred_open_entries == 0) {
+				if (tdb_delete(tdb, key) == -1) {
+					SAFE_FREE(de_entries);
+					SAFE_FREE(dbuf.dptr);
+					return 0;
+				}
+			} else {
+				if (tdb_store(tdb, key, dbuf, TDB_REPLACE) == -1) {
+					SAFE_FREE(de_entries);
+					SAFE_FREE(dbuf.dptr);
+					return 0;
+				}
+			}
+		}
+	}
+
+	SAFE_FREE(dbuf.dptr);
+	*pp_de_entries = de_entries;
+	return num_de_entries;
+}
+
+/*******************************************************************
+ Check if two deferred open entries are identical.
+********************************************************************/
+
+static BOOL deferred_open_entries_identical( deferred_open_entry *e1, deferred_open_entry *e2)
+{
+#if 1 /* JRA PARANOIA TEST - REMOVE LATER */
+	if (e1->pid == e2->pid &&
+		e1->port == e2->port &&
+		e1->dev == e2->dev &&
+		e1->inode == e2->inode &&
+		((e1->time.tv_sec != e2->time.tv_sec) ||
+		 (e1->time.tv_usec != e2->time.tv_usec) ||
+		 (e1->mid != e2->mid))) {
+		smb_panic("PANIC: deferred_open_entries_identical: logic error.\n");
+	}
+#endif
+
+	return (e1->pid == e2->pid &&
+		e1->mid == e2->mid &&
+		e1->port == e2->port &&
+		e1->dev == e2->dev &&
+		e1->inode == e2->inode &&
+		e1->time.tv_sec == e2->time.tv_sec &&
+		e1->time.tv_usec == e2->time.tv_usec);
+}
+
+
+/*******************************************************************
+ Delete a specific deferred open entry.
+ Ignore if no entry deleted.
+********************************************************************/
+
+BOOL delete_deferred_open_entry(deferred_open_entry *entry)
+{
+	TDB_DATA dbuf;
+	struct deferred_open_data *data;
+	int i, del_count=0;
+	deferred_open_entry *de_entries;
+	BOOL ret = True;
+	TDB_DATA key = deferred_open_locking_key(entry->dev, entry->inode);
+
+	/* read in the existing share modes */
+	dbuf = tdb_fetch(tdb, key);
+	if (!dbuf.dptr)
+		return -1;
+
+	data = (struct deferred_open_data *)dbuf.dptr;
+	de_entries = (deferred_open_entry *)(dbuf.dptr + sizeof(*data));
+
+	/*
+	 * Find any with this pid and delete it
+	 * by overwriting with the rest of the data 
+	 * from the record.
+	 */
+
+	DEBUG(10,("delete_deferred_open_entry: num_deferred_open_entries = %d\n",
+		data->u.num_deferred_open_entries ));
+
+	for (i=0;i<data->u.num_deferred_open_entries;) {
+		if (deferred_open_entries_identical(&de_entries[i], entry)) {
+			DEBUG(10,("delete_deferred_open_entry: deleted %s\n",
+				deferred_open_str(i, &de_entries[i]) ));
+
+			data->u.num_deferred_open_entries--;
+			if ((dbuf.dsize - (sizeof(*data) + (i+1)*sizeof(*de_entries))) > 0) {
+				memmove(&de_entries[i], &de_entries[i+1], 
+					dbuf.dsize - (sizeof(*data) + (i+1)*sizeof(*de_entries)));
+			}
+			del_count++;
+
+			DEBUG(10,("delete_deferred_open_entry: deleting entry %d\n", i ));
+
+		} else {
+			i++;
+		}
+	}
+
+	SMB_ASSERT(del_count == 0 || del_count == 1);
+
+	if (del_count) {
+		/* the record may have shrunk a bit */
+		dbuf.dsize -= del_count * sizeof(*de_entries);
+
+		/* store it back in the database */
+		if (data->u.num_deferred_open_entries == 0) {
+			if (tdb_delete(tdb, key) == -1)
+				ret = False;
+		} else {
+			if (tdb_store(tdb, key, dbuf, TDB_REPLACE) == -1)
+				ret = False;
+		}
+	}
+	DEBUG(10,("delete_deferred_open_entry: Remaining table.\n"));
+	print_deferred_open_table((struct deferred_open_data*)dbuf.dptr);
+	SAFE_FREE(dbuf.dptr);
+	return ret;
+}
+
+/*******************************************************************
+ Fill a deferred open entry.
+********************************************************************/
+
+static void fill_deferred_open(char *p, uint16 mid, struct timeval *ptv, SMB_DEV_T dev, SMB_INO_T inode, uint16 port)
+{
+	deferred_open_entry *e = (deferred_open_entry *)p;
+	void *x = &e->time; /* Needed to force alignment. p may not be aligned.... */
+
+	memset(e, '\0', sizeof(deferred_open_entry));
+	e->mid = mid;
+	e->pid = sys_getpid();
+	memcpy(x, ptv, sizeof(struct timeval));
+	e->dev = dev;
+	e->inode = inode;
+	e->port = port;
+}
+
+/*******************************************************************
+ Add a deferred open record. Return False on fail, True on success.
+********************************************************************/
+
+BOOL add_deferred_open(uint16 mid, struct timeval *ptv, SMB_DEV_T dev, SMB_INO_T inode, uint16 port, const char *fname)
+{
+	TDB_DATA dbuf;
+	struct deferred_open_data *data;
+	char *p=NULL;
+	int size;
+	TDB_DATA key = deferred_open_locking_key(dev, inode);
+	BOOL ret = True;
+		
+	/* read in the existing deferred open records if any */
+	dbuf = tdb_fetch(tdb, key);
+	if (!dbuf.dptr) {
+		size_t offset;
+		/* we'll need to create a new record */
+
+		size = sizeof(*data) + sizeof(deferred_open_entry) + strlen(fname) + 1;
+		p = (char *)malloc(size);
+		if (!p)
+			return False;
+		data = (struct deferred_open_data *)p;
+		data->u.num_deferred_open_entries = 1;
+	
+		DEBUG(10,("add_deferred_open: creating entry for file %s. num_deferred_open_entries = 1\n",
+			fname ));
+
+		offset = sizeof(*data) + sizeof(deferred_open_entry);
+		safe_strcpy(p + offset, fname, size - offset - 1);
+		fill_deferred_open(p + sizeof(*data), mid, ptv, dev, inode, port);
+		dbuf.dptr = p;
+		dbuf.dsize = size;
+		if (tdb_store(tdb, key, dbuf, TDB_REPLACE) == -1)
+			ret = False;
+
+		print_deferred_open_table((struct deferred_open_data *)p);
+
+		SAFE_FREE(p);
+		return ret;
+	}
+
+	/* we're adding to an existing entry - this is a bit fiddly */
+	data = (struct deferred_open_data *)dbuf.dptr;
+
+	data->u.num_deferred_open_entries++;
+	
+	DEBUG(10,("add_deferred_open: adding entry for file %s. new num_deferred_open_entries = %d\n",
+		fname, data->u.num_deferred_open_entries ));
+
+	size = dbuf.dsize + sizeof(deferred_open_entry);
+	p = malloc(size);
+	if (!p) {
+		SAFE_FREE(dbuf.dptr);
+		return False;
+	}
+	memcpy(p, dbuf.dptr, sizeof(*data));
+	fill_deferred_open(p + sizeof(*data), mid, ptv, dev, inode, port);
+	memcpy(p + sizeof(*data) + sizeof(deferred_open_entry), dbuf.dptr + sizeof(*data),
+	       dbuf.dsize - sizeof(*data));
+	SAFE_FREE(dbuf.dptr);
+	dbuf.dptr = p;
+	dbuf.dsize = size;
+	if (tdb_store(tdb, key, dbuf, TDB_REPLACE) == -1)
+		ret = False;
+	print_deferred_open_table((struct deferred_open_data *)p);
+	SAFE_FREE(p);
+	return ret;
+}
+
 /****************************************************************************
  Traverse the whole database with this function, calling traverse_callback
  on each share mode
@@ -862,6 +1240,10 @@ static int traverse_fn(TDB_CONTEXT *the_tdb, TDB_DATA kbuf, TDB_DATA dbuf,
 
 	SHAREMODE_FN(traverse_callback) = (SHAREMODE_FN_CAST())state;
 
+	/* Ensure this is a locking_key record. */
+	if (kbuf.dsize != sizeof(struct locking_key))
+		return 0;
+
 	data = (struct locking_data *)dbuf.dptr;
 	shares = (share_mode_entry *)(dbuf.dptr + sizeof(*data));
 	name = dbuf.dptr + sizeof(*data) + data->u.num_share_mode_entries*sizeof(*shares);
diff --git a/source3/printing/nt_printing.c b/source3/printing/nt_printing.c
index 909aed6c866..225ff20ec3e 100644
--- a/source3/printing/nt_printing.c
+++ b/source3/printing/nt_printing.c
@@ -1003,9 +1003,9 @@ static int file_version_is_newer(connection_struct *conn, fstring new_file, fstr
 	driver_unix_convert(filepath,conn,NULL,&bad_path,&stat_buf);
 
 	fsp = open_file_shared(conn, filepath, &stat_buf,
-						   SET_OPEN_MODE(DOS_OPEN_RDONLY),
+						   SET_DENY_MODE(DENY_NONE)|SET_OPEN_MODE(DOS_OPEN_RDONLY),
 						   (FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN),
-						   FILE_ATTRIBUTE_NORMAL, 0, &access_mode, &action);
+						   FILE_ATTRIBUTE_NORMAL, INTERNAL_OPEN_ONLY, &access_mode, &action);
 	if (!fsp) {
 		/* Old file not found, so by definition new file is in fact newer */
 		DEBUG(10,("file_version_is_newer: Can't open old file [%s], errno = %d\n",
@@ -1032,9 +1032,9 @@ static int file_version_is_newer(connection_struct *conn, fstring new_file, fstr
 	driver_unix_convert(filepath,conn,NULL,&bad_path,&stat_buf);
 
 	fsp = open_file_shared(conn, filepath, &stat_buf,
-						   SET_OPEN_MODE(DOS_OPEN_RDONLY),
+						   SET_DENY_MODE(DENY_NONE)|SET_OPEN_MODE(DOS_OPEN_RDONLY),
 						   (FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN),
-						   FILE_ATTRIBUTE_NORMAL, 0, &access_mode, &action);
+						   FILE_ATTRIBUTE_NORMAL, INTERNAL_OPEN_ONLY, &access_mode, &action);
 	if (!fsp) {
 		/* New file not found, this shouldn't occur if the caller did its job */
 		DEBUG(3,("file_version_is_newer: Can't open new file [%s], errno = %d\n",
@@ -1148,9 +1148,9 @@ static uint32 get_correct_cversion(const char *architecture, fstring driverpath_
 	driver_unix_convert(driverpath,conn,NULL,&bad_path,&st);
 
 	fsp = open_file_shared(conn, driverpath, &st,
-						   SET_OPEN_MODE(DOS_OPEN_RDONLY),
+						   SET_DENY_MODE(DENY_NONE)|SET_OPEN_MODE(DOS_OPEN_RDONLY),
 						   (FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN),
-						   FILE_ATTRIBUTE_NORMAL, 0, &access_mode, &action);
+						   FILE_ATTRIBUTE_NORMAL, INTERNAL_OPEN_ONLY, &access_mode, &action);
 	if (!fsp) {
 		DEBUG(3,("get_correct_cversion: Can't open file [%s], errno = %d\n",
 				driverpath, errno));
diff --git a/source3/rpc_server/srv_srvsvc_nt.c b/source3/rpc_server/srv_srvsvc_nt.c
index 77b9be99660..087c50451e5 100644
--- a/source3/rpc_server/srv_srvsvc_nt.c
+++ b/source3/rpc_server/srv_srvsvc_nt.c
@@ -1886,8 +1886,9 @@ WERROR _srv_net_file_query_secdesc(pipes_struct *p, SRV_Q_NET_FILE_QUERY_SECDESC
 
 	unistr2_to_ascii(filename, &q_u->uni_file_name, sizeof(filename));
 	unix_convert(filename, conn, NULL, &bad_path, &st);
-	fsp = open_file_shared(conn, filename, &st, SET_OPEN_MODE(DOS_OPEN_RDONLY),
-				(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN), FILE_ATTRIBUTE_NORMAL, 0, &access_mode, &action);
+	fsp = open_file_shared(conn, filename, &st, SET_DENY_MODE(DENY_NONE)|SET_OPEN_MODE(DOS_OPEN_RDONLY),
+				(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN), FILE_ATTRIBUTE_NORMAL, INTERNAL_OPEN_ONLY,
+				&access_mode, &action);
 
 	if (!fsp) {
 		/* Perhaps it is a directory */
@@ -1990,8 +1991,9 @@ WERROR _srv_net_file_set_secdesc(pipes_struct *p, SRV_Q_NET_FILE_SET_SECDESC *q_
 	unistr2_to_ascii(filename, &q_u->uni_file_name, sizeof(filename));
 	unix_convert(filename, conn, NULL, &bad_path, &st);
 
-	fsp = open_file_shared(conn, filename, &st, SET_OPEN_MODE(DOS_OPEN_RDWR),
-			(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN), FILE_ATTRIBUTE_NORMAL, 0, &access_mode, &action);
+	fsp = open_file_shared(conn, filename, &st, SET_DENY_MODE(DENY_NONE)|SET_OPEN_MODE(DOS_OPEN_RDWR),
+			(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN), FILE_ATTRIBUTE_NORMAL, INTERNAL_OPEN_ONLY,
+			&access_mode, &action);
 
 	if (!fsp) {
 		/* Perhaps it is a directory */
diff --git a/source3/script/mkproto.awk b/source3/script/mkproto.awk
index b38f405e0da..fbe1bddf357 100644
--- a/source3/script/mkproto.awk
+++ b/source3/script/mkproto.awk
@@ -124,7 +124,7 @@ END {
     gotstart = 1;
   }
 
-  if( $0 ~ /^smb_iconv_t|^long|^char|^uint|^NTSTATUS|^WERROR|^CLI_POLICY_HND|^struct|^BOOL|^void|^time|^smb_shm_offset_t|^shm_offset_t|^FILE|^XFILE|^SMB_OFF_T|^size_t|^ssize_t|^SMB_BIG_UINT/ ) {
+  if( $0 ~ /^smb_iconv_t|^long|^char|^uint|^NTSTATUS|^WERROR|^CLI_POLICY_HND|^struct|^BOOL|^void|^time|^smb_shm_offset_t|^shm_offset_t|^FILE|^XFILE|^SMB_OFF_T|^size_t|^ssize_t|^SMB_BIG_UINT|^SMB_BIG_INT/ ) {
     gotstart = 1;
   }
 
diff --git a/source3/smbd/close.c b/source3/smbd/close.c
index 8b3010c1b2e..305377dfc63 100644
--- a/source3/smbd/close.c
+++ b/source3/smbd/close.c
@@ -2,6 +2,7 @@
    Unix SMB/CIFS implementation.
    file closing
    Copyright (C) Andrew Tridgell 1992-1998
+   Copyright (C) Jeremy Allison 1992-2004.
    
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -105,6 +106,32 @@ static int close_filestruct(files_struct *fsp)
 }    
 
 /****************************************************************************
+ If any deferred opens are waiting on this close, notify them.
+****************************************************************************/
+
+static void notify_deferred_opens(files_struct *fsp)
+{
+	deferred_open_entry *de_array = NULL;
+	int num_de_entries, i;
+	pid_t mypid = sys_getpid();
+
+	num_de_entries = get_deferred_opens(fsp->conn, fsp->dev, fsp->inode, &de_array);
+	for (i = 0; i < num_de_entries; i++) {
+		deferred_open_entry *entry = &de_array[i];
+		if (entry->pid == mypid) {
+			/*
+			 * We need to notify ourself to retry the open.
+			 * Do this by finding the queued SMB record, moving it
+			 * to the head of the queue and changing the wait time to zero.
+			 */
+			schedule_sharing_violation_open_smb_message(entry->mid);
+		} else {
+			send_deferred_open_retry_message(entry);
+		}
+	}
+}
+
+/****************************************************************************
  Close a file.
 
  If normal_close is 1 then this came from a normal SMBclose (or equivalent)
@@ -177,6 +204,9 @@ static int close_normal_file(files_struct *fsp, BOOL normal_close)
 
 	SAFE_FREE(share_entry);
 
+	/* Notify any deferred opens waiting on this close. */
+	notify_deferred_opens(fsp);
+
 	/*
 	 * NT can set delete_on_close of the last open
 	 * reference to a file.
diff --git a/source3/smbd/dir.c b/source3/smbd/dir.c
index 06ef23ab8cd..b88f687766d 100644
--- a/source3/smbd/dir.c
+++ b/source3/smbd/dir.c
@@ -763,7 +763,8 @@ static BOOL user_can_write_file(connection_struct *conn, char *name, SMB_STRUCT_
 		return True;
 	else
 		fsp = open_file_shared1(conn, name, pst, FILE_WRITE_ATTRIBUTES, SET_DENY_MODE(DENY_NONE),
-			(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN), FILE_ATTRIBUTE_NORMAL, 0, &access_mode, &smb_action);
+			(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN), FILE_ATTRIBUTE_NORMAL, INTERNAL_OPEN_ONLY,
+			&access_mode, &smb_action);
 
 	if (!fsp)
 		return False;
diff --git a/source3/smbd/nttrans.c b/source3/smbd/nttrans.c
index 26be4434fdb..e540db234a8 100644
--- a/source3/smbd/nttrans.c
+++ b/source3/smbd/nttrans.c
@@ -863,6 +863,10 @@ create_options = 0x%x root_dir_fid = 0x%x\n", flags, desired_access, file_attrib
 
 				restore_case_semantics(conn, file_attributes);
 				END_PROFILE(SMBntcreateX);
+				if (open_was_deferred(SVAL(inbuf,smb_mid))) {
+					/* We have re-scheduled this call. */
+					return -1;
+				}
 				return set_bad_path_error(errno, bad_path, outbuf, ERRDOS,ERRnoaccess);
 			}
 		} 
@@ -1347,6 +1351,10 @@ static int call_nt_transact_create(connection_struct *conn, char *inbuf, char *o
 				}
 			} else {
 				restore_case_semantics(conn, file_attributes);
+				if (open_was_deferred(SVAL(inbuf,smb_mid))) {
+					/* We have re-scheduled this call. */
+					return -1;
+				}
 				return set_bad_path_error(errno, bad_path, outbuf, ERRDOS,ERRnoaccess);
 			}
 		} 
diff --git a/source3/smbd/open.c b/source3/smbd/open.c
index ab5ea236faa..04e074d56eb 100644
--- a/source3/smbd/open.c
+++ b/source3/smbd/open.c
@@ -2,7 +2,7 @@
    Unix SMB/CIFS implementation.
    file opening and share modes
    Copyright (C) Andrew Tridgell 1992-1998
-   Copyright (C) Jeremy Allison 2001
+   Copyright (C) Jeremy Allison 2001-2004
    
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -26,6 +26,11 @@ extern uint16 global_oplock_port;
 extern uint16 global_smbpid;
 extern BOOL global_client_failed_oplock_break;
 
+struct dev_inode_bundle {
+	SMB_DEV_T dev;
+	SMB_INO_T inode;
+};
+
 /****************************************************************************
  fd support routines - attempt to do a dos_open.
 ****************************************************************************/
@@ -243,7 +248,7 @@ static BOOL open_file(files_struct *fsp,connection_struct *conn,
 }
 
 /*******************************************************************
-return True if the filename is one of the special executable types
+ Return True if the filename is one of the special executable types.
 ********************************************************************/
 
 static BOOL is_executable(const char *fname)
@@ -262,12 +267,13 @@ static BOOL is_executable(const char *fname)
 enum {AFAIL,AREAD,AWRITE,AALL};
 
 /*******************************************************************
-reproduce the share mode access table
-this is horrendoously complex, and really can't be justified on any
-rational grounds except that this is _exactly_ what NT does. See
-the DENY1 and DENY2 tests in smbtorture for a comprehensive set of
-test routines.
+ Reproduce the share mode access table.
+ This is horrendoously complex, and really can't be justified on any
+ rational grounds except that this is _exactly_ what NT does. See
+ the DENY1 and DENY2 tests in smbtorture for a comprehensive set of
+ test routines.
 ********************************************************************/
+
 static int access_table(int new_deny,int old_deny,int old_mode,
 			BOOL same_pid, BOOL isexe)
 {
@@ -353,9 +359,8 @@ static int access_table(int new_deny,int old_deny,int old_mode,
 	  return(AFAIL);      
 }
 
-
 /****************************************************************************
-check if we can open a file with a share mode
+ Check if we can open a file with a share mode.
 ****************************************************************************/
 
 static BOOL check_share_mode(connection_struct *conn, share_mode_entry *share, int share_mode, uint32 desired_access,
@@ -770,9 +775,101 @@ after break ! For file %s, dev = %x, inode = %.0f. Deleting it to continue...\n"
 }
 
 /****************************************************************************
-set a kernel flock on a file for NFS interoperability
-this requires a patch to Linux
+ Delete the record for a handled deferred open entry.
+****************************************************************************/
+
+static void delete_defered_open_entry_record(connection_struct *conn, SMB_DEV_T dev, SMB_INO_T inode)
+{
+	uint16 mid = get_current_mid();
+	pid_t mypid = sys_getpid();
+	deferred_open_entry *de_array = NULL;
+	int num_de_entries, i;
+
+	num_de_entries = get_deferred_opens(conn, dev, inode, &de_array);
+	for (i = 0; i < num_de_entries; i++) {
+		deferred_open_entry *entry = &de_array[i];
+		if (entry->pid == mypid && entry->mid == mid && entry->dev == dev &&
+			entry->inode == inode) {
+
+			/* Remove the deferred open entry from the array. */
+			delete_deferred_open_entry(entry);
+			SAFE_FREE(de_array);
+			return;
+		}
+	}
+	SAFE_FREE(de_array);
+}
+
+/****************************************************************************
+ Handle the 1 second delay in returning a SHARING_VIOLATION error.
+****************************************************************************/
+
+void defer_open_sharing_error(connection_struct *conn, struct timeval *ptv,
+		char *fname, SMB_DEV_T dev, SMB_INO_T inode)
+{
+	uint16 mid = get_current_mid();
+	pid_t mypid = sys_getpid();
+	deferred_open_entry *de_array = NULL;
+	int num_de_entries, i;
+	struct dev_inode_bundle dib;
+
+	dib.dev = dev;
+	dib.inode = inode;
+
+	num_de_entries = get_deferred_opens(conn, dev, inode, &de_array);
+	for (i = 0; i < num_de_entries; i++) {
+		deferred_open_entry *entry = &de_array[i];
+		if (entry->pid == mypid && entry->mid == mid) {
+			/*
+			 * Check if a 1 second timeout has expired.
+			 */
+			if (usec_time_diff(ptv, &entry->time) > SHARING_VIOLATION_USEC_WAIT) {
+				DEBUG(10,("defer_open_sharing_error: Deleting deferred open entry for mid %u, \
+file %s\n",
+					(unsigned int)mid, fname ));
+
+				/* Expired, return a real error. */
+				/* Remove the deferred open entry from the array. */
+
+				delete_deferred_open_entry(entry);
+				SAFE_FREE(de_array);
+				return;
+			}
+			/*
+			 * If the timeout hasn't expired yet and we still have a sharing violation,
+			 * just leave the entry in the deferred open array alone. We do need to
+			 * reschedule this open call though (with the original created time).
+			 */
+			DEBUG(10,("defer_open_sharing_error: time [%u.%06u] updating \
+deferred open entry for mid %u, file %s\n",
+				(unsigned int)entry->time.tv_sec,
+				(unsigned int)entry->time.tv_usec,
+				(unsigned int)mid, fname ));
+
+			push_sharing_violation_open_smb_message(&entry->time, (char *)&dib, sizeof(dib));
+			SAFE_FREE(de_array);
+			return;
+		}
+	}
+
+	DEBUG(10,("defer_open_sharing_error: time [%u.%06u] adding deferred open entry for mid %u, file %s\n",
+		(unsigned int)ptv->tv_sec, (unsigned int)ptv->tv_usec, (unsigned int)mid, fname ));
+
+	if (!push_sharing_violation_open_smb_message(ptv, (char *)&dib, sizeof(dib))) {
+		SAFE_FREE(de_array);
+		return;
+	}
+	if (!add_deferred_open(mid, ptv, dev, inode, global_oplock_port, fname)) {
+		remove_sharing_violation_open_smb_message(mid);
+	}
+	SAFE_FREE(de_array);
+}
+
+/****************************************************************************
+ Set a kernel flock on a file for NFS interoperability.
+ This requires a patch to Linux.
 ****************************************************************************/
+
 static void kernel_flock(files_struct *fsp, int deny_mode)
 {
 #if HAVE_KERNEL_SHARE_MODES
@@ -847,6 +944,7 @@ files_struct *open_file_shared1(connection_struct *conn,char *fname, SMB_STRUCT_
 	BOOL fcbopen = False;
 	BOOL def_acl = False;
 	BOOL add_share_mode = True;
+	BOOL internal_only_open = False;
 	SMB_DEV_T dev = 0;
 	SMB_INO_T inode = 0;
 	int num_share_modes = 0;
@@ -858,9 +956,50 @@ files_struct *open_file_shared1(connection_struct *conn,char *fname, SMB_STRUCT_
 	mode_t new_mode = (mode_t)0;
 	int action;
 	uint32 existing_dos_mode = 0;
+	struct pending_message_list *pml = NULL;
+	uint16 mid = get_current_mid();
 	/* We add aARCH to this as this mode is only used if the file is created new. */
 	mode_t mode = unix_mode(conn,new_dos_mode | aARCH,fname);
 
+	if (oplock_request == INTERNAL_OPEN_ONLY) {
+		internal_only_open = True;
+		oplock_request = 0;
+	}
+
+	if ((pml = get_open_deferred_message(mid)) != NULL) {
+		struct dev_inode_bundle dib;
+
+		memcpy(&dib, pml->private_data.data, sizeof(dib));
+
+		/* There could be a race condition where the dev/inode pair
+			has changed since we deferred the message. If so, just
+			remove the deferred open entry and return sharing violation. */
+
+		/* If the timeout value is non-zero, we need to just
+			return sharing violation. Don't retry the open
+			as we were not notified of a close and we don't want to
+			trigger another spurious oplock break. */
+
+		if (!file_existed || dib.dev != psbuf->st_dev || dib.inode != psbuf->st_ino ||
+				pml->msg_time.tv_sec || pml->msg_time.tv_usec) {
+			/* Ensure we don't reprocess this message. */
+			remove_sharing_violation_open_smb_message(mid);
+
+			/* Now remove the deferred open entry under lock. */
+			lock_share_entry(conn, dib.dev, dib.inode);
+			delete_defered_open_entry_record(conn, dib.dev, dib.inode);
+			unlock_share_entry(conn, dib.dev, dib.inode);
+
+			unix_ERR_class = ERRDOS;
+			unix_ERR_code = ERRbadshare;
+			unix_ERR_ntstatus = NT_STATUS_SHARING_VIOLATION;
+			return NULL;
+		}
+		/* Ensure we don't reprocess this message. */
+		remove_sharing_violation_open_smb_message(mid);
+
+	}
+
 	if (conn->printer) {
 		/* printers are handled completely differently. Most of the passed parameters are
 			ignored */
@@ -1043,17 +1182,28 @@ flags=0x%X flags2=0x%X mode=0%o returned %d\n",
 				unix_ERR_ntstatus = NT_STATUS_ACCESS_DENIED;
 			}
 
+			/* 
+			 * If we're returning a share violation, ensure we cope with
+			 * the braindead 1 second delay.
+			 */
+
+			if (!internal_only_open && NT_STATUS_EQUAL(unix_ERR_ntstatus,NT_STATUS_SHARING_VIOLATION)) {
+				/* The fsp->open_time here represents the current time of day. */
+				defer_open_sharing_error(conn, &fsp->open_time, fname, dev, inode);
+			}
+
 			unlock_share_entry(conn, dev, inode);
-			if (fsp_open)
+			if (fsp_open) {
 				fd_close(conn, fsp);
+				/*
+				 * We have detected a sharing violation here
+				 * so return the correct error code
+				 */
+				unix_ERR_class = ERRDOS;
+				unix_ERR_code = ERRbadshare;
+				unix_ERR_ntstatus = NT_STATUS_SHARING_VIOLATION;
+			}
 			file_free(fsp);
-			/*
-			 * We have detected a sharing violation here
-			 * so return the correct error code
-			 */
-                        unix_ERR_class = ERRDOS;
-                        unix_ERR_code = ERRbadshare;
-                        unix_ERR_ntstatus = NT_STATUS_SHARING_VIOLATION;
 			return NULL;
 		}
 
@@ -1118,6 +1268,16 @@ flags=0x%X flags2=0x%X mode=0%o returned %d\n",
 						  &flags, &oplock_request, &all_current_opens_are_level_II);
 
 		if(num_share_modes == -1) {
+			/* 
+			 * If we're returning a share violation, ensure we cope with
+			 * the braindead 1 second delay.
+			 */
+
+			if (!internal_only_open && NT_STATUS_EQUAL(unix_ERR_ntstatus,NT_STATUS_SHARING_VIOLATION)) {
+				/* The fsp->open_time here represents the current time of day. */
+				defer_open_sharing_error(conn, &fsp->open_time, fname, dev, inode);
+			}
+
 			unlock_share_entry_fsp(fsp);
 			fd_close(conn,fsp);
 			file_free(fsp);
@@ -1286,6 +1446,8 @@ flags=0x%X flags2=0x%X mode=0%o returned %d\n",
 				fname, (int)new_mode));
 	}
 
+	/* If this is a successful open, we must remove any deferred open records. */
+	delete_defered_open_entry_record(conn, fsp->dev, fsp->inode);
 	unlock_share_entry_fsp(fsp);
 
 	conn->num_files_open++;
diff --git a/source3/smbd/oplock.c b/source3/smbd/oplock.c
index 19e6956d9ef..6739f5654f1 100644
--- a/source3/smbd/oplock.c
+++ b/source3/smbd/oplock.c
@@ -388,6 +388,30 @@ pid %d, port %d, dev = %x, inode = %.0f, file_id = %lu\n",
 				(int)remotepid, from_port, (unsigned int)dev, (double)inode, file_id));
 			break;
 
+		case RETRY_DEFERRED_OPEN_CMD:
+
+			/* Request to retry and open that would return SHARING_VIOLATION. */
+			if (msg_len != DEFERRED_OPEN_MSG_LEN) {
+				DEBUG(0,("process_local_message: incorrect length for RETRY_DEFERRED_OPEN_CMD (was %d, should be %d).\n",
+					(int)msg_len, (int)DEFERRED_OPEN_MSG_LEN));
+				return False;
+			}
+			{
+				uint16 mid;
+
+				memcpy((char *)&remotepid, msg_start+DEFERRED_OPEN_PID_OFFSET,sizeof(remotepid));
+				memcpy((char *)&inode, msg_start+DEFERRED_OPEN_INODE_OFFSET,sizeof(inode));
+				memcpy((char *)&dev, msg_start+DEFERRED_OPEN_DEV_OFFSET,sizeof(dev));
+				memcpy((char *)&mid, msg_start+DEFERRED_OPEN_MID_OFFSET,sizeof(mid));
+
+				DEBUG(5,("process_local_message: RETRY_DEFERRED_OPEN from \
+pid %d, port %d, dev = %x, inode = %.0f, mid = %u\n",
+					(int)remotepid, from_port, (unsigned int)dev, (double)inode, (unsigned int)mid));
+
+				schedule_sharing_violation_open_smb_message(mid);
+			}
+			break;
+
 		/* 
 		 * Keep this as a debug case - eventually we can remove it.
 		 */
@@ -1215,7 +1239,51 @@ void release_level_2_oplocks_on_change(files_struct *fsp)
 }
 
 /****************************************************************************
-setup oplocks for this process
+ Send a 'retry your open' message to a process with a deferred open entry.
+****************************************************************************/
+
+BOOL send_deferred_open_retry_message(deferred_open_entry *entry)
+{
+	char de_msg[DEFERRED_OPEN_MSG_LEN];
+	struct sockaddr_in addr_out;
+	pid_t pid = sys_getpid();
+
+	memset(de_msg, '\0', DEFERRED_OPEN_MSG_LEN);
+	SSVAL(de_msg,DEFERRED_OPEN_CMD_OFFSET,RETRY_DEFERRED_OPEN_CMD);
+	memcpy(de_msg+DEFERRED_OPEN_PID_OFFSET,(char *)&pid,sizeof(pid));
+	memcpy(de_msg+DEFERRED_OPEN_DEV_OFFSET,(char *)&entry->dev,sizeof(entry->dev));
+	memcpy(de_msg+DEFERRED_OPEN_INODE_OFFSET,(char *)&entry->inode,sizeof(entry->inode));
+	memcpy(de_msg+DEFERRED_OPEN_MID_OFFSET,(char *)&entry->mid,sizeof(entry->mid));
+
+	/* Set the address and port. */
+	memset((char *)&addr_out,'\0',sizeof(addr_out));
+	addr_out.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
+	addr_out.sin_port = htons( entry->port );
+	addr_out.sin_family = AF_INET;
+   
+	if( DEBUGLVL( 3 ) ) {
+		dbgtext( "send_deferred_open_retry_message: sending a message to ");
+		dbgtext( "pid %d on port %d ", (int)entry->pid, entry->port );
+		dbgtext( "for dev = %x, inode = %.0f, mid = %u\n",
+			(unsigned int)entry->dev, (double)entry->inode, (unsigned int)entry->mid );
+	}
+
+	if(sys_sendto(oplock_sock,de_msg,DEFERRED_OPEN_MSG_LEN,0,
+			(struct sockaddr *)&addr_out,sizeof(addr_out)) < 0) {
+		if( DEBUGLVL( 0 ) ) {
+			dbgtext( "send_deferred_open_retry_message: failed sending a message to ");
+			dbgtext( "pid %d on port %d ", (int)entry->pid, entry->port );
+			dbgtext( "for dev = %x, inode = %.0f, mid = %u\n",
+				(unsigned int)entry->dev, (double)entry->inode, (unsigned int)entry->mid );
+			dbgtext( "Error was %s\n", strerror(errno) );
+		}
+		return False;
+	}
+	return True;
+}
+
+/****************************************************************************
+ Setup oplocks for this process.
 ****************************************************************************/
 
 BOOL init_oplocks(void)
diff --git a/source3/smbd/process.c b/source3/smbd/process.c
index ccebd2b86c1..698c8475f79 100644
--- a/source3/smbd/process.c
+++ b/source3/smbd/process.c
@@ -61,20 +61,28 @@ uint16 get_current_mid(void)
  for processing.
 ****************************************************************************/
 
-struct pending_message_list {
-	struct pending_message_list *next, *prev;
-	char *msg_buf;
-	int msg_len;
-};
-
 static struct pending_message_list *smb_oplock_queue;
+static struct pending_message_list *smb_sharing_violation_queue;
+
+enum q_type { OPLOCK_QUEUE, SHARE_VIOLATION_QUEUE };
+
+/****************************************************************************
+ Free up a message.
+****************************************************************************/
+
+static void free_queued_message(struct pending_message_list *msg)
+{
+	data_blob_free(&msg->buf);
+	data_blob_free(&msg->private_data);
+	SAFE_FREE(msg);
+}
 
 /****************************************************************************
  Function to push a message onto the tail of a linked list of smb messages ready
  for processing.
 ****************************************************************************/
 
-static BOOL push_message(char *buf, int msg_len)
+static BOOL push_queued_message(enum q_type qt, char *buf, int msg_len, struct timeval *ptv, char *private, size_t private_len)
 {
 	struct pending_message_list *tmp_msg;
 	struct pending_message_list *msg = (struct pending_message_list *)
@@ -85,32 +93,159 @@ static BOOL push_message(char *buf, int msg_len)
 		return False;
 	}
 
-	msg->msg_buf = (char *)malloc(msg_len);
-	if(msg->msg_buf == NULL) {
+	memset(msg,'\0',sizeof(*msg));
+
+	msg->buf = data_blob(buf, msg_len);
+	if(msg->buf.data == NULL) {
 		DEBUG(0,("push_message: malloc fail (2)\n"));
 		SAFE_FREE(msg);
 		return False;
 	}
 
-	memcpy(msg->msg_buf, buf, msg_len);
-	msg->msg_len = msg_len;
+	if (ptv) {
+		msg->msg_time = *ptv;
+	}
 
-	DLIST_ADD_END(smb_oplock_queue, msg, tmp_msg);
+	if (private) {
+		msg->private_data = data_blob(private, private_len);
+		if (msg->private_data.data == NULL) {
+			DEBUG(0,("push_message: malloc fail (3)\n"));
+			data_blob_free(&msg->buf);
+			SAFE_FREE(msg);
+		}
+	}
+
+	if (qt == OPLOCK_QUEUE) {
+		DLIST_ADD_END(smb_oplock_queue, msg, tmp_msg);
+	} else {
+		DLIST_ADD_END(smb_sharing_violation_queue, msg, tmp_msg);
+	}
 
 	/* Push the MID of this packet on the signing queue. */
 	srv_defer_sign_response(SVAL(buf,smb_mid));
 
+	DEBUG(10,("push_message: pushed message length %u on queue %s\n",
+		(unsigned int)msg_len,
+		qt == OPLOCK_QUEUE ? "smb_oplock_queue" : "smb_sharing_violation_queue" ));
+
 	return True;
 }
 
 /****************************************************************************
- Function to push a smb message onto a linked list of local smb messages ready
+ Function to push an oplock smb message onto a linked list of local smb messages ready
  for processing.
 ****************************************************************************/
 
 BOOL push_oplock_pending_smb_message(char *buf, int msg_len)
 {
-	return push_message(buf, msg_len);
+	return push_queued_message(OPLOCK_QUEUE, buf, msg_len, NULL, NULL, 0);
+}
+
+/****************************************************************************
+ Function to delete a sharing violation open message by mid.
+****************************************************************************/
+
+void remove_sharing_violation_open_smb_message(uint16 mid)
+{
+	struct pending_message_list *pml;
+
+	for (pml = smb_sharing_violation_queue; pml; pml = pml->next) {
+		if (mid == SVAL(pml->buf.data,smb_mid)) {
+			DEBUG(10,("remove_sharing_violation_open_smb_message: deleting mid %u len %u\n",
+				(unsigned int)mid, (unsigned int)pml->buf.length ));
+			DLIST_REMOVE(smb_sharing_violation_queue, pml);
+			free_queued_message(pml);
+			return;
+		}
+	}
+}
+
+/****************************************************************************
+ Move a sharing violation open retry message to the front of the list and
+ schedule it for immediate processing.
+****************************************************************************/
+
+void schedule_sharing_violation_open_smb_message(uint16 mid)
+{
+	struct pending_message_list *pml;
+	int i = 0;
+
+	for (pml = smb_sharing_violation_queue; pml; pml = pml->next) {
+		uint16 msg_mid = SVAL(pml->buf.data,smb_mid);
+		DEBUG(10,("schedule_sharing_violation_open_smb_message: [%d] msg_mid = %u\n", i++,
+			(unsigned int)msg_mid ));
+		if (mid == msg_mid) {
+			DEBUG(10,("schedule_sharing_violation_open_smb_message: scheduling mid %u\n",
+				mid ));
+			pml->msg_time.tv_sec = 0;
+			pml->msg_time.tv_usec = 0;
+			DLIST_PROMOTE(smb_sharing_violation_queue, pml);
+			return;
+		}
+	}
+
+	DEBUG(10,("schedule_sharing_violation_open_smb_message: failed to find message mid %u\n",
+		mid ));
+}
+
+/****************************************************************************
+ Return true if this mid is on the deferred queue.
+****************************************************************************/
+
+BOOL open_was_deferred(uint16 mid)
+{
+	struct pending_message_list *pml;
+	for (pml = smb_sharing_violation_queue; pml; pml = pml->next) {
+		if (SVAL(pml->buf.data,smb_mid) == mid) {
+			return True;
+		}
+	}
+	return False;
+}
+
+/****************************************************************************
+ Return the message queued by this mid.
+****************************************************************************/
+
+struct pending_message_list *get_open_deferred_message(uint16 mid)
+{
+	struct pending_message_list *pml;
+	for (pml = smb_sharing_violation_queue; pml; pml = pml->next) {
+		if (SVAL(pml->buf.data,smb_mid) == mid) {
+			return pml;
+		}
+	}
+	return NULL;
+}
+
+/****************************************************************************
+ Function to push a sharing violation open smb message onto a linked list of local smb messages ready
+ for processing.
+****************************************************************************/
+
+BOOL push_sharing_violation_open_smb_message(struct timeval *ptv, char *private, size_t priv_len)
+{
+	uint16 mid = SVAL(InBuffer,smb_mid);
+	struct timeval tv;
+	SMB_BIG_INT tdif;
+
+	tv = *ptv;
+	tdif = tv.tv_sec;
+	tdif *= 1000000;
+	tdif += tv.tv_usec;
+
+	/* Add on the timeout. */
+	tdif += SHARING_VIOLATION_USEC_WAIT;
+	
+	tv.tv_sec = tdif / 1000000;
+	tv.tv_usec = tdif % 1000000;
+	
+	DEBUG(10,("push_sharing_violation_open_smb_message: pushing message len %u mid %u\
+ timeout time [%u.%06u]\n", (unsigned int) smb_len(InBuffer)+4, (unsigned int)mid,
+		(unsigned int)tv.tv_sec, (unsigned int)tv.tv_usec));
+
+	return push_queued_message(SHARE_VIOLATION_QUEUE, InBuffer,
+			smb_len(InBuffer)+4, &tv, private, priv_len);
 }
 
 /****************************************************************************
@@ -169,12 +304,17 @@ static BOOL receive_message_or_smb(char *buffer, int buffer_len, int timeout)
 	fd_set fds;
 	int selrtn;
 	struct timeval to;
+	struct timeval *pto;
 	int maxfd;
 
 	smb_read_error = 0;
 
  again:
 
+	to.tv_sec = timeout / 1000;
+	to.tv_usec = (timeout % 1000) * 1000;
+	pto = timeout > 0 ? &to : NULL;
+
 	/*
 	 * Note that this call must be before processing any SMB
 	 * messages as we need to synchronously process any messages
@@ -188,17 +328,55 @@ static BOOL receive_message_or_smb(char *buffer, int buffer_len, int timeout)
 	 */
   	if(smb_oplock_queue != NULL) {
 		struct pending_message_list *msg = smb_oplock_queue;
-		memcpy(buffer, msg->msg_buf, MIN(buffer_len, msg->msg_len));
+		memcpy(buffer, msg->buf.data, MIN(buffer_len, msg->buf.length));
   
 		/* Free the message we just copied. */
 		DLIST_REMOVE(smb_oplock_queue, msg);
-		SAFE_FREE(msg->msg_buf);
-		SAFE_FREE(msg);
+		free_queued_message(msg);
 		
 		DEBUG(5,("receive_message_or_smb: returning queued smb message.\n"));
 		return True;
 	}
 
+	/*
+	 * Check to see if we already have a message on the deferred open queue
+	 * and it's time to schedule.
+	 */
+  	if(smb_sharing_violation_queue != NULL) {
+		BOOL pop_message = False;
+		struct pending_message_list *msg = smb_sharing_violation_queue;
+
+		if (msg->msg_time.tv_sec == 0 && msg->msg_time.tv_usec == 0) {
+			pop_message = True;
+		} else {
+			struct timeval tv;
+			SMB_BIG_INT tdif;
+
+			GetTimeOfDay(&tv);
+			tdif = usec_time_diff(&msg->msg_time, &tv);
+			if (tdif <= 0) {
+				/* Timed out. Schedule...*/
+				pop_message = True;
+				DEBUG(10,("receive_message_or_smb: queued message timed out.\n"));
+			} else {
+				/* Make a more accurate select timeout. */
+				to.tv_sec = tdif / 1000000;
+				to.tv_usec = tdif % 1000000;
+				pto = &to;
+				DEBUG(10,("receive_message_or_smb: select with timeout of [%u.%06u]\n",
+					(unsigned int)pto->tv_sec, (unsigned int)pto->tv_usec ));
+			}
+		}
+
+		if (pop_message) {
+			memcpy(buffer, msg->buf.data, MIN(buffer_len, msg->buf.length));
+  
+			/* We leave this message on the queue so the open code can
+			   know this is a retry. */
+			DEBUG(5,("receive_message_or_smb: returning deferred open smb message.\n"));
+			return True;
+		}
+	}
 
 	/*
 	 * Setup the select read fd set.
@@ -229,10 +407,7 @@ static BOOL receive_message_or_smb(char *buffer, int buffer_len, int timeout)
 	FD_SET(smbd_server_fd(),&fds);
 	maxfd = setup_oplock_select_set(&fds);
 
-	to.tv_sec = timeout / 1000;
-	to.tv_usec = (timeout % 1000) * 1000;
-
-	selrtn = sys_select(MAX(maxfd,smbd_server_fd())+1,&fds,NULL,NULL,timeout>0?&to:NULL);
+	selrtn = sys_select(MAX(maxfd,smbd_server_fd())+1,&fds,NULL,NULL,pto);
 
 	/* if we get EINTR then maybe we have received an oplock
 	   signal - treat this as select returning 1. This is ugly, but
diff --git a/source3/smbd/reply.c b/source3/smbd/reply.c
index fff5385171d..bf208b0fa4e 100644
--- a/source3/smbd/reply.c
+++ b/source3/smbd/reply.c
@@ -1043,6 +1043,10 @@ int reply_open(connection_struct *conn, char *inbuf,char *outbuf, int dum_size,
 
 	if (!fsp) {
 		END_PROFILE(SMBopen);
+		if (open_was_deferred(SVAL(inbuf,smb_mid))) {
+			/* We have re-scheduled this call. */
+			return -1;
+		}
 		return set_bad_path_error(errno, bad_path, outbuf, ERRDOS, ERRnoaccess);
 	}
 
@@ -1132,6 +1136,10 @@ int reply_open_and_X(connection_struct *conn, char *inbuf,char *outbuf,int lengt
       
 	if (!fsp) {
 		END_PROFILE(SMBopenX);
+		if (open_was_deferred(SVAL(inbuf,smb_mid))) {
+			/* We have re-scheduled this call. */
+			return -1;
+		}
 		return set_bad_path_error(errno, bad_path, outbuf, ERRDOS, ERRnoaccess);
 	}
 
@@ -1257,6 +1265,10 @@ int reply_mknew(connection_struct *conn, char *inbuf,char *outbuf, int dum_size,
   
 	if (!fsp) {
 		END_PROFILE(SMBcreate);
+		if (open_was_deferred(SVAL(inbuf,smb_mid))) {
+			/* We have re-scheduled this call. */
+			return -1;
+		}
 		return set_bad_path_error(errno, bad_path, outbuf, ERRDOS, ERRnoaccess);
 	}
  
@@ -1332,6 +1344,10 @@ int reply_ctemp(connection_struct *conn, char *inbuf,char *outbuf, int dum_size,
 
 	if (!fsp) {
 		END_PROFILE(SMBctemp);
+		if (open_was_deferred(SVAL(inbuf,smb_mid))) {
+			/* We have re-scheduled this call. */
+			return -1;
+		}
 		return set_bad_path_error(errno, bad_path, outbuf, ERRDOS, ERRnoaccess);
 	}
 
@@ -1623,8 +1639,13 @@ int reply_unlink(connection_struct *conn, char *inbuf,char *outbuf, int dum_size
 	DEBUG(3,("reply_unlink : %s\n",name));
 	
 	status = unlink_internals(conn, dirtype, name);
-	if (!NT_STATUS_IS_OK(status))
+	if (!NT_STATUS_IS_OK(status)) {
+		if (open_was_deferred(SVAL(inbuf,smb_mid))) {
+			/* We have re-scheduled this call. */
+			return -1;
+		}
 		return ERROR_NT(status);
+	}
 
 	/*
 	 * Win2k needs a changenotify request response before it will
@@ -3944,6 +3965,10 @@ int reply_mv(connection_struct *conn, char *inbuf,char *outbuf, int dum_size,
 	status = rename_internals(conn, name, newname, attrs, False);
 	if (!NT_STATUS_IS_OK(status)) {
 		END_PROFILE(SMBmv);
+		if (open_was_deferred(SVAL(inbuf,smb_mid))) {
+			/* We have re-scheduled this call. */
+			return -1;
+		}
 		return ERROR_NT(status);
 	}
 
@@ -3989,7 +4014,8 @@ static BOOL copy_file(char *src,char *dest1,connection_struct *conn, int ofun,
 		return(False);
 
 	fsp1 = open_file_shared(conn,src,&src_sbuf,SET_DENY_MODE(DENY_NONE)|SET_OPEN_MODE(DOS_OPEN_RDONLY),
-					(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN),FILE_ATTRIBUTE_NORMAL,0,&Access,&action);
+					(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN),FILE_ATTRIBUTE_NORMAL,INTERNAL_OPEN_ONLY,
+					&Access,&action);
 
 	if (!fsp1)
 		return(False);
@@ -4002,7 +4028,7 @@ static BOOL copy_file(char *src,char *dest1,connection_struct *conn, int ofun,
 		ZERO_STRUCTP(&sbuf2);
 
 	fsp2 = open_file_shared(conn,dest,&sbuf2,SET_DENY_MODE(DENY_NONE)|SET_OPEN_MODE(DOS_OPEN_WRONLY),
-			ofun,dosattrs,0,&Access,&action);
+			ofun,dosattrs,INTERNAL_OPEN_ONLY,&Access,&action);
 
 	if (!fsp2) {
 		close_file(fsp1,False);
diff --git a/source3/smbd/trans2.c b/source3/smbd/trans2.c
index 02a6bf6e4bb..0ba26a91479 100644
--- a/source3/smbd/trans2.c
+++ b/source3/smbd/trans2.c
@@ -627,6 +627,10 @@ static int call_trans2open(connection_struct *conn, char *inbuf, char *outbuf, i
 		oplock_request, &rmode,&smb_action);
       
 	if (!fsp) {
+		if (open_was_deferred(SVAL(inbuf,smb_mid))) {
+			/* We have re-scheduled this call. */
+			return -1;
+		}
 		return set_bad_path_error(errno, bad_path, outbuf, ERRDOS,ERRnoaccess);
 	}
 
@@ -3205,7 +3209,7 @@ static int call_trans2setfilepathinfo(connection_struct *conn,
 									SET_OPEN_MODE(DOS_OPEN_RDWR),
 									(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN),
 									FILE_ATTRIBUTE_NORMAL,
-									0, &access_mode, &action);
+									INTERNAL_OPEN_ONLY, &access_mode, &action);
  
 					if (new_fsp == NULL)
 						return(UNIXERROR(ERRDOS,ERRbadpath));
@@ -3655,7 +3659,7 @@ size = %.0f, uid = %u, gid = %u, raw perms = 0%o\n",
 						SET_OPEN_MODE(DOS_OPEN_RDWR),
 						(FILE_FAIL_IF_NOT_EXIST|FILE_EXISTS_OPEN),
 						FILE_ATTRIBUTE_NORMAL,
-						0, &access_mode, &action);
+						INTERNAL_OPEN_ONLY, &access_mode, &action);
 	
 			if (new_fsp == NULL)
 				return(UNIXERROR(ERRDOS,ERRbadpath));