r22082: merged the ctdb changes from bzr

added opendb ctdb backend from ronnie
(This used to be commit b0da25cb79f860bfa14ba7a8419c7996d936292b)

9 files changed, 753 insertions, 16 deletions

diff --git a/source4/cluster/ctdb/common/ctdb_call.c b/source4/cluster/ctdb/common/ctdb_call.c
index 75355f7ae7..0b4195140c 100644
--- a/source4/cluster/ctdb/common/ctdb_call.c
+++ b/source4/cluster/ctdb/common/ctdb_call.c
@@ -189,7 +189,7 @@ static void ctdb_call_send_dmaster(struct ctdb_db_context *ctdb_db,
 	r->hdr.srcnode   = ctdb->vnn;
 	r->hdr.reqid     = c->hdr.reqid;
 	r->db_id         = c->db_id;
-	r->dmaster       = header->laccessor;
+	r->dmaster       = c->hdr.srcnode;
 	r->keylen        = key->dsize;
 	r->datalen       = data->dsize;
 	memcpy(&r->data[0], key->dptr, key->dsize);
@@ -239,7 +239,7 @@ void ctdb_request_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr
 		ctdb_send_error(ctdb, hdr, ret, "Unknown database in request. db_id==0x%08x",c->db_id);
 		return;
 	}
-
+	
 	/* fetch the current record */
 	ret = ctdb_ltdb_fetch(ctdb_db, key, &header, &data2);
 	if (ret != 0) {
@@ -329,9 +329,12 @@ void ctdb_request_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 	}
 
 	/* if this nodes has done enough consecutive calls on the same record
-	   then give them the record */
-	if (header.laccessor == c->hdr.srcnode &&
-	    header.lacount >= ctdb->max_lacount) {
+	   then give them the record
+	   or if the node requested an immediate migration
+	*/
+	if ( (header.laccessor == c->hdr.srcnode
+	      && header.lacount >= ctdb->max_lacount)
+	   || c->flags&CTDB_IMMEDIATE_MIGRATION ) {
 		ctdb_call_send_dmaster(ctdb_db, c, &header, &call.key, &data);
 		talloc_free(data.dptr);
 		return;
@@ -373,6 +376,7 @@ struct ctdb_call_state {
 	struct ctdb_call call;
 	int redirect_count;
 	struct ctdb_ltdb_header header;
+	void *fetch_private;
 };
 
 
@@ -419,7 +423,6 @@ void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 	}
 	ctdb_db = state->ctdb_db;
 
-
 	data.dptr = c->data;
 	data.dsize = c->datalen;
 
@@ -578,6 +581,7 @@ struct ctdb_call_state *ctdb_call_send(struct ctdb_db_context *ctdb_db, struct c
 	state->c->hdr.srcnode   = ctdb->vnn;
 	/* this limits us to 16k outstanding messages - not unreasonable */
 	state->c->hdr.reqid     = idr_get_new(ctdb->idr, state, 0xFFFF);
+	state->c->flags         = call->flags;
 	state->c->db_id         = ctdb_db->db_id;
 	state->c->callid        = call->call_id;
 	state->c->keylen        = call->key.dsize;
@@ -604,6 +608,13 @@ struct ctdb_call_state *ctdb_call_send(struct ctdb_db_context *ctdb_db, struct c
 }
 
 
+
+struct ctdb_record_handle {
+	struct ctdb_db_context *ctdb_db;
+	TDB_DATA key;
+	TDB_DATA *data;
+};
+
 /*
   make a remote ctdb call - async recv. 
 
@@ -612,6 +623,8 @@ struct ctdb_call_state *ctdb_call_send(struct ctdb_db_context *ctdb_db, struct c
 */
 int ctdb_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
 {
+	struct ctdb_record_handle *rec;
+
 	while (state->state < CTDB_CALL_DONE) {
 		event_loop_once(state->node->ctdb->ev);
 	}
@@ -620,6 +633,18 @@ int ctdb_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
 		talloc_free(state);
 		return -1;
 	}
+
+	rec = state->fetch_private;
+
+	/* ugly hack to manage forced migration */
+	if (rec != NULL) {
+		rec->data->dptr = talloc_memdup(rec, state->call.reply_data.dptr,
+						     state->call.reply_data.dsize);
+		rec->data->dsize = state->call.reply_data.dsize;
+		talloc_free(state);
+		return 0;
+	}
+
 	if (state->call.reply_data.dsize) {
 		call->reply_data.dptr = talloc_memdup(state->node->ctdb,
 						      state->call.reply_data.dptr,
@@ -643,3 +668,59 @@ int ctdb_call(struct ctdb_db_context *ctdb_db, struct ctdb_call *call)
 	state = ctdb_call_send(ctdb_db, call);
 	return ctdb_call_recv(state, call);
 }
+
+
+
+
+
+
+struct ctdb_record_handle *ctdb_fetch_lock(struct ctdb_db_context *ctdb_db, TALLOC_CTX *mem_ctx, 
+					   TDB_DATA key, TDB_DATA *data)
+{
+	struct ctdb_call call;
+	struct ctdb_record_handle *rec;
+	struct ctdb_call_state *state;
+	int ret;
+
+	ZERO_STRUCT(call);
+	call.call_id = CTDB_FETCH_FUNC;
+	call.key = key;
+	call.flags = CTDB_IMMEDIATE_MIGRATION;
+
+	rec = talloc(mem_ctx, struct ctdb_record_handle);
+	CTDB_NO_MEMORY_NULL(ctdb_db->ctdb, rec);
+
+	rec->ctdb_db = ctdb_db;
+	rec->key = key;
+	rec->key.dptr = talloc_memdup(rec, key.dptr, key.dsize);
+	rec->data = data;
+
+	state = ctdb_call_send(ctdb_db, &call);
+	state->fetch_private = rec;
+
+	ret = ctdb_call_recv(state, &call);
+	if (ret != 0) {
+		talloc_free(rec);
+		return NULL;
+	}
+
+	return rec;
+}
+
+
+int ctdb_record_store(struct ctdb_record_handle *rec, TDB_DATA data)
+{
+	int ret;
+	struct ctdb_ltdb_header header;
+
+	/* should be avoided if possible    hang header off rec ? */
+	ret = ctdb_ltdb_fetch(rec->ctdb_db, rec->key, &header, NULL);
+	if (ret) {
+		ctdb_set_error(rec->ctdb_db->ctdb, "Fetch of locally held record failed");
+		return ret;
+	}
+
+	ret = ctdb_ltdb_store(rec->ctdb_db, rec->key, &header, data);
+		
+	return ret;
+}
diff --git a/source4/cluster/ctdb/common/ctdb_ltdb.c b/source4/cluster/ctdb/common/ctdb_ltdb.c
index 189816229f..ceedb6c5ce 100644
--- a/source4/cluster/ctdb/common/ctdb_ltdb.c
+++ b/source4/cluster/ctdb/common/ctdb_ltdb.c
@@ -41,6 +41,17 @@ struct ctdb_db_context *ctdb_db_handle(struct ctdb_context *ctdb, const char *na
 	return NULL;
 }
 
+
+/*
+  this is the dummy null procedure that all databases support
+*/
+static int ctdb_fetch_func(struct ctdb_call_info *call)
+{
+	call->reply_data = &call->record_data;
+	return 0;
+}
+
+
 /*
   attach to a specific database
 */
@@ -49,6 +60,7 @@ struct ctdb_db_context *ctdb_attach(struct ctdb_context *ctdb, const char *name,
 {
 	struct ctdb_db_context *ctdb_db, *tmp_db;
 	TDB_DATA data;
+	int ret;
 
 	ctdb_db = talloc_zero(ctdb, struct ctdb_db_context);
 	CTDB_NO_MEMORY_NULL(ctdb, ctdb_db);
@@ -80,7 +92,18 @@ struct ctdb_db_context *ctdb_attach(struct ctdb_context *ctdb, const char *name,
 		return NULL;
 	}
 
+
+	/* 
+	  all databases support the "fetch" function. we need this in order to do forced migration of records
+	 */
+	ret = ctdb_set_call(ctdb_db, ctdb_fetch_func, CTDB_FETCH_FUNC);
+	if (ret != 0) {
+		talloc_free(ctdb_db);
+		return NULL;
+	}
+
 	DLIST_ADD(ctdb->db_list, ctdb_db);
+
 	return ctdb_db;
 }
 
@@ -124,18 +147,25 @@ int ctdb_ltdb_fetch(struct ctdb_db_context *ctdb_db,
 		/* return an initial header */
 		free(rec.dptr);
 		ltdb_initial_header(ctdb_db, key, header);
-		data->dptr = NULL;
-		data->dsize = 0;
+		if (data) {
+			data->dptr = NULL;
+			data->dsize = 0;
+		}
 		return 0;
 	}
 
 	*header = *(struct ctdb_ltdb_header *)rec.dptr;
 
-	data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
-	data->dptr = talloc_memdup(ctdb_db, sizeof(struct ctdb_ltdb_header)+rec.dptr,
-				   data->dsize);
+	if (data) {
+		data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
+		data->dptr = talloc_memdup(ctdb_db, sizeof(struct ctdb_ltdb_header)+rec.dptr,
+					   data->dsize);
+	}
+
 	free(rec.dptr);
-	CTDB_NO_MEMORY(ctdb, data->dptr);
+	if (data) {
+		CTDB_NO_MEMORY(ctdb, data->dptr);
+	}
 
 	return 0;
 }
diff --git a/source4/cluster/ctdb/config.mk b/source4/cluster/ctdb/config.mk
index 89ac7aa7f4..893bd9f136 100644
--- a/source4/cluster/ctdb/config.mk
+++ b/source4/cluster/ctdb/config.mk
@@ -3,6 +3,10 @@
 OBJ_FILES = brlock_ctdb.o
 
 ##################
+[SUBSYSTEM::opendb_ctdb]
+OBJ_FILES = opendb_ctdb.o
+
+##################
 [SUBSYSTEM::ctdb_tcp]
 OBJ_FILES = \
 		tcp/tcp_init.o \
diff --git a/source4/cluster/ctdb/include/ctdb.h b/source4/cluster/ctdb/include/ctdb.h
index 51eeaebedb..9049314401 100644
--- a/source4/cluster/ctdb/include/ctdb.h
+++ b/source4/cluster/ctdb/include/ctdb.h
@@ -21,12 +21,14 @@
 #ifndef _CTDB_H
 #define _CTDB_H
 
+#define CTDB_IMMEDIATE_MIGRATION	0x00000001
 struct ctdb_call {
 	int call_id;
 	TDB_DATA key;
 	TDB_DATA call_data;
 	TDB_DATA reply_data;
 	uint32_t status;
+	uint32_t flags;
 };
 
 /*
@@ -148,6 +150,7 @@ int ctdb_set_message_handler(struct ctdb_context *ctdb, ctdb_message_fn_t handle
 int ctdb_send_message(struct ctdb_context *ctdb, uint32_t vnn,
 		      uint32_t srvid, TDB_DATA data);
 
+
 /* 
    fetch and lock a ctdb record. Underneath this will force the
    dmaster for the record to be moved to the local node. 
@@ -155,7 +158,7 @@ int ctdb_send_message(struct ctdb_context *ctdb, uint32_t vnn,
    The lock is released when is talloc_free() is called on the
    returned ctdb_record_handle. 
 */
-struct ctdb_record_handle *ctdb_fetch_lock(struct ctdb_db_context *ctdb_db, TDB_DATA key, TDB_DATA *data);
+struct ctdb_record_handle *ctdb_fetch_lock(struct ctdb_db_context *ctdb_db, TALLOC_CTX *mem_ctx, TDB_DATA key, TDB_DATA *data);
 
 /*
   change the data in a record held with a ctdb_record_handle
@@ -163,4 +166,5 @@ struct ctdb_record_handle *ctdb_fetch_lock(struct ctdb_db_context *ctdb_db, TDB_
  */
 int ctdb_record_store(struct ctdb_record_handle *rec, TDB_DATA data);
 
+
 #endif
diff --git a/source4/cluster/ctdb/include/ctdb_private.h b/source4/cluster/ctdb/include/ctdb_private.h
index 1e2244ff7d..66a9a6e244 100644
--- a/source4/cluster/ctdb/include/ctdb_private.h
+++ b/source4/cluster/ctdb/include/ctdb_private.h
@@ -23,6 +23,8 @@
 
 #include "ctdb.h"
 
+
+#define CTDB_FETCH_FUNC 0xf0000001
 /*
   an installed ctdb remote call
 */
@@ -166,6 +168,7 @@ struct ctdb_req_header {
 
 struct ctdb_req_call {
 	struct ctdb_req_header hdr;
+	uint32_t flags;
 	uint32_t db_id;
 	uint32_t callid;
 	uint32_t keylen;
diff --git a/source4/cluster/ctdb/opendb_ctdb.c b/source4/cluster/ctdb/opendb_ctdb.c
new file mode 100644
index 0000000000..e12bf3260b
--- /dev/null
+++ b/source4/cluster/ctdb/opendb_ctdb.c
@@ -0,0 +1,611 @@
+/* 
+   Unix SMB/CIFS implementation.
+
+   Copyright (C) Andrew Tridgell 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
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+   
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+   
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+*/
+
+/*
+  this is the open files database, ctdb backend. It implements shared
+  storage of what files are open between server instances, and
+  implements the rules of shared access to files.
+
+  The caller needs to provide a file_key, which specifies what file
+  they are talking about. This needs to be a unique key across all
+  filesystems, and is usually implemented in terms of a device/inode
+  pair.
+
+  Before any operations can be performed the caller needs to establish
+  a lock on the record associated with file_key. That is done by
+  calling odb_lock(). The caller releases this lock by calling
+  talloc_free() on the returned handle.
+
+  All other operations on a record are done by passing the odb_lock()
+  handle back to this module. The handle contains internal
+  information about what file_key is being operated on.
+*/
+
+#include "includes.h"
+#include "system/filesys.h"
+#include "lib/tdb/include/tdb.h"
+#include "messaging/messaging.h"
+#include "db_wrap.h"
+#include "lib/messaging/irpc.h"
+#include "librpc/gen_ndr/ndr_opendb.h"
+#include "ntvfs/ntvfs.h"
+#include "ntvfs/common/ntvfs_common.h"
+#include "cluster/cluster.h"
+#include "include/ctdb.h"
+
+struct odb_context {
+	struct ctdb_context *ctdb;
+	struct ctdb_db_context *ctdb_db;
+	struct ntvfs_context *ntvfs_ctx;
+	BOOL oplocks;
+};
+
+/*
+  an odb lock handle. You must obtain one of these using odb_lock() before doing
+  any other operations. 
+*/
+struct odb_lock {
+	struct odb_context *odb;
+	struct ctdb_record_handle *rec;
+	TDB_DATA key;
+	TDB_DATA data;
+};
+
+/*
+  Open up the openfiles.tdb database. Close it down using
+  talloc_free(). We need the messaging_ctx to allow for pending open
+  notifications.
+*/
+static struct odb_context *odb_ctdb_init(TALLOC_CTX *mem_ctx, 
+					struct ntvfs_context *ntvfs_ctx)
+{
+	struct odb_context *odb;
+	struct ctdb_context *ctdb = talloc_get_type(cluster_backend_handle(), 
+						    struct ctdb_context);
+
+	odb = talloc(mem_ctx, struct odb_context);
+	if (odb == NULL) {
+		return NULL;
+	}
+
+	odb->ctdb = ctdb;
+	odb->ctdb_db = ctdb_db_handle(ctdb, "opendb");
+	if (!odb->ctdb_db) {
+		DEBUG(0,("Failed to get attached ctdb db handle for opendb\n"));
+		talloc_free(odb);
+		return NULL;
+	}
+
+	odb->ntvfs_ctx = ntvfs_ctx;
+
+	/* leave oplocks disabled by default until the code is working */
+	odb->oplocks = lp_parm_bool(-1, "opendb", "oplocks", False);
+
+	return odb;
+}
+
+/*
+  get a lock on a entry in the odb. This call returns a lock handle,
+  which the caller should unlock using talloc_free().
+*/
+static struct odb_lock *odb_ctdb_lock(TALLOC_CTX *mem_ctx,
+				     struct odb_context *odb, DATA_BLOB *file_key)
+{
+	struct odb_lock *lck;
+
+	lck = talloc(mem_ctx, struct odb_lock);
+	if (lck == NULL) {
+		return NULL;
+	}
+
+	lck->odb = talloc_reference(lck, odb);
+	lck->key.dptr = talloc_memdup(lck, file_key->data, file_key->length);
+	lck->key.dsize = file_key->length;
+	if (lck->key.dptr == NULL) {
+		talloc_free(lck);
+		return NULL;
+	}
+
+	lck->rec = ctdb_fetch_lock(odb->ctdb_db, (TALLOC_CTX *)lck, lck->key, &lck->data);
+	if (!lck->rec) {
+		talloc_free(lck);
+		return NULL;
+	}
+
+	return lck;
+}
+
+/*
+  determine if two odb_entry structures conflict
+
+  return NT_STATUS_OK on no conflict
+*/
+static NTSTATUS share_conflict(struct opendb_entry *e1, struct opendb_entry *e2)
+{
+	/* if either open involves no read.write or delete access then
+	   it can't conflict */
+	if (!(e1->access_mask & (SEC_FILE_WRITE_DATA |
+				 SEC_FILE_APPEND_DATA |
+				 SEC_FILE_READ_DATA |
+				 SEC_FILE_EXECUTE |
+				 SEC_STD_DELETE))) {
+		return NT_STATUS_OK;
+	}
+	if (!(e2->access_mask & (SEC_FILE_WRITE_DATA |
+				 SEC_FILE_APPEND_DATA |
+				 SEC_FILE_READ_DATA |
+				 SEC_FILE_EXECUTE |
+				 SEC_STD_DELETE))) {
+		return NT_STATUS_OK;
+	}
+
+	/* data IO access masks. This is skipped if the two open handles
+	   are on different streams (as in that case the masks don't
+	   interact) */
+	if (e1->stream_id != e2->stream_id) {
+		return NT_STATUS_OK;
+	}
+
+#define CHECK_MASK(am, right, sa, share) \
+	if (((am) & (right)) && !((sa) & (share))) return NT_STATUS_SHARING_VIOLATION
+
+	CHECK_MASK(e1->access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA,
+		   e2->share_access, NTCREATEX_SHARE_ACCESS_WRITE);
+	CHECK_MASK(e2->access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA,
+		   e1->share_access, NTCREATEX_SHARE_ACCESS_WRITE);
+	
+	CHECK_MASK(e1->access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE,
+		   e2->share_access, NTCREATEX_SHARE_ACCESS_READ);
+	CHECK_MASK(e2->access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE,
+		   e1->share_access, NTCREATEX_SHARE_ACCESS_READ);
+
+	CHECK_MASK(e1->access_mask, SEC_STD_DELETE,
+		   e2->share_access, NTCREATEX_SHARE_ACCESS_DELETE);
+	CHECK_MASK(e2->access_mask, SEC_STD_DELETE,
+		   e1->share_access, NTCREATEX_SHARE_ACCESS_DELETE);
+
+	return NT_STATUS_OK;
+}
+
+/*
+  pull a record, translating from the db format to the opendb_file structure defined
+  in opendb.idl
+*/
+static NTSTATUS odb_pull_record(struct odb_lock *lck, struct opendb_file *file)
+{
+	TDB_DATA dbuf;
+	DATA_BLOB blob;
+	NTSTATUS status;
+		
+	dbuf = lck->data;
+
+	blob.data = dbuf.dptr;
+	blob.length = dbuf.dsize;
+
+	status = ndr_pull_struct_blob(&blob, lck, file, (ndr_pull_flags_fn_t)ndr_pull_opendb_file);
+
+	return status;
+}
+
+/*
+  push a record, translating from the opendb_file structure defined in opendb.idl
+*/
+static NTSTATUS odb_push_record(struct odb_lock *lck, struct opendb_file *file)
+{
+	TDB_DATA dbuf;
+	DATA_BLOB blob;
+	NTSTATUS status;
+	int ret;
+
+	if (!file->num_entries) {
+		dbuf.dptr  = NULL;
+		dbuf.dsize = 0;
+		ctdb_record_store(lck->rec, dbuf);
+	}
+
+	status = ndr_push_struct_blob(&blob, lck, file, (ndr_push_flags_fn_t)ndr_push_opendb_file);
+	NT_STATUS_NOT_OK_RETURN(status);
+
+	dbuf.dptr = blob.data;
+	dbuf.dsize = blob.length;
+		
+	ret = ctdb_record_store(lck->rec, dbuf);
+	data_blob_free(&blob);
+	if (ret != 0) {
+		return NT_STATUS_INTERNAL_DB_CORRUPTION;
+	}
+
+	return NT_STATUS_OK;
+}
+
+/*
+  send an oplock break to a client
+*/
+static NTSTATUS odb_oplock_break_send(struct odb_context *odb, struct opendb_entry *e)
+{
+	/* tell the server handling this open file about the need to send the client
+	   a break */
+	return messaging_send_ptr(odb->ntvfs_ctx->msg_ctx, e->server, 
+				  MSG_NTVFS_OPLOCK_BREAK, e->file_handle);
+}
+
+/*
+  register an open file in the open files database. This implements the share_access
+  rules
+
+  Note that the path is only used by the delete on close logic, not
+  for comparing with other filenames
+*/
+static NTSTATUS odb_ctdb_open_file(struct odb_lock *lck, void *file_handle,
+				  uint32_t stream_id, uint32_t share_access, 
+				  uint32_t access_mask, BOOL delete_on_close,
+				  const char *path, 
+				  uint32_t oplock_level, uint32_t *oplock_granted)
+{
+	struct odb_context *odb = lck->odb;
+	struct opendb_entry e;
+	int i;
+	struct opendb_file file;
+	NTSTATUS status;
+
+	if (odb->oplocks == False) {
+		oplock_level = OPLOCK_NONE;
+	}
+
+	status = odb_pull_record(lck, &file);
+	if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) {
+		/* initialise a blank structure */
+		ZERO_STRUCT(file);
+		file.path = path;
+	} else {
+		NT_STATUS_NOT_OK_RETURN(status);
+	}
+
+	/* see if it conflicts */
+	e.server          = odb->ntvfs_ctx->server_id;
+	e.file_handle     = file_handle;
+	e.stream_id       = stream_id;
+	e.share_access    = share_access;
+	e.access_mask     = access_mask;
+	e.delete_on_close = delete_on_close;
+	e.oplock_level    = OPLOCK_NONE;
+		
+	/* see if anyone has an oplock, which we need to break */
+	for (i=0;i<file.num_entries;i++) {
+		if (file.entries[i].oplock_level == OPLOCK_BATCH) {
+			/* a batch oplock caches close calls, which
+			   means the client application might have
+			   already closed the file. We have to allow
+			   this close to propogate by sending a oplock
+			   break request and suspending this call
+			   until the break is acknowledged or the file
+			   is closed */
+			odb_oplock_break_send(odb, &file.entries[i]);
+			return NT_STATUS_OPLOCK_NOT_GRANTED;
+		}
+	}
+
+	if (file.delete_on_close || 
+	    (file.num_entries != 0 && delete_on_close)) {
+		/* while delete on close is set, no new opens are allowed */
+		return NT_STATUS_DELETE_PENDING;
+	}
+
+	/* check for sharing violations */
+	for (i=0;i<file.num_entries;i++) {
+		status = share_conflict(&file.entries[i], &e);
+		NT_STATUS_NOT_OK_RETURN(status);
+	}
+
+	/* we now know the open could succeed, but we need to check
+	   for any exclusive oplocks. We can't grant a second open
+	   till these are broken. Note that we check for batch oplocks
+	   before checking for sharing violations, and check for
+	   exclusive oplocks afterwards. */
+	for (i=0;i<file.num_entries;i++) {
+		if (file.entries[i].oplock_level == OPLOCK_EXCLUSIVE) {
+			odb_oplock_break_send(odb, &file.entries[i]);
+			return NT_STATUS_OPLOCK_NOT_GRANTED;
+		}
+	}
+
+	/*
+	  possibly grant an exclusive or batch oplock if this is the only client
+	  with the file open. We don't yet grant levelII oplocks.
+	*/
+	if (oplock_granted != NULL) {
+		if ((oplock_level == OPLOCK_BATCH ||
+		     oplock_level == OPLOCK_EXCLUSIVE) &&
+		    file.num_entries == 0) {
+			(*oplock_granted) = oplock_level;
+		} else {
+			(*oplock_granted) = OPLOCK_NONE;
+		}
+		e.oplock_level = (*oplock_granted);
+	}
+
+	/* it doesn't conflict, so add it to the end */
+	file.entries = talloc_realloc(lck, file.entries, struct opendb_entry, 
+				      file.num_entries+1);
+	NT_STATUS_HAVE_NO_MEMORY(file.entries);
+
+	file.entries[file.num_entries] = e;
+	file.num_entries++;
+
+	return odb_push_record(lck, &file);
+}
+
+
+/*
+  register a pending open file in the open files database
+*/
+static NTSTATUS odb_ctdb_open_file_pending(struct odb_lock *lck, void *private)
+{
+	struct odb_context *odb = lck->odb;
+	struct opendb_file file;
+	NTSTATUS status;
+		
+	status = odb_pull_record(lck, &file);
+	NT_STATUS_NOT_OK_RETURN(status);
+
+	file.pending = talloc_realloc(lck, file.pending, struct opendb_pending, 
+				      file.num_pending+1);
+	NT_STATUS_HAVE_NO_MEMORY(file.pending);
+
+	file.pending[file.num_pending].server = odb->ntvfs_ctx->server_id;
+	file.pending[file.num_pending].notify_ptr = private;
+
+	file.num_pending++;
+
+	return odb_push_record(lck, &file);
+}
+
+
+/*
+  remove a opendb entry
+*/
+static NTSTATUS odb_ctdb_close_file(struct odb_lock *lck, void *file_handle)
+{
+	struct odb_context *odb = lck->odb;
+	struct opendb_file file;
+	int i;
+	NTSTATUS status;
+
+	status = odb_pull_record(lck, &file);
+	NT_STATUS_NOT_OK_RETURN(status);
+
+	/* find the entry, and delete it */
+	for (i=0;i<file.num_entries;i++) {
+		if (file_handle == file.entries[i].file_handle &&
+		    cluster_id_equal(&odb->ntvfs_ctx->server_id, &file.entries[i].server)) {
+			if (file.entries[i].delete_on_close) {
+				file.delete_on_close = True;
+			}
+			if (i < file.num_entries-1) {
+				memmove(file.entries+i, file.entries+i+1, 
+					(file.num_entries - (i+1)) * 
+					sizeof(struct opendb_entry));
+			}
+			break;
+		}
+	}
+
+	if (i == file.num_entries) {
+		return NT_STATUS_UNSUCCESSFUL;
+	}
+
+	/* send any pending notifications, removing them once sent */
+	for (i=0;i<file.num_pending;i++) {
+		messaging_send_ptr(odb->ntvfs_ctx->msg_ctx, file.pending[i].server, 
+				   MSG_PVFS_RETRY_OPEN, 
+				   file.pending[i].notify_ptr);
+	}
+	file.num_pending = 0;
+
+	file.num_entries--;
+	
+	return odb_push_record(lck, &file);
+}
+
+
+/*
+  remove a pending opendb entry
+*/
+static NTSTATUS odb_ctdb_remove_pending(struct odb_lock *lck, void *private)
+{
+	struct odb_context *odb = lck->odb;
+	int i;
+	NTSTATUS status;
+	struct opendb_file file;
+
+	status = odb_pull_record(lck, &file);
+	NT_STATUS_NOT_OK_RETURN(status);
+
+	/* find the entry, and delete it */
+	for (i=0;i<file.num_pending;i++) {
+		if (private == file.pending[i].notify_ptr &&
+		    cluster_id_equal(&odb->ntvfs_ctx->server_id, &file.pending[i].server)) {
+			if (i < file.num_pending-1) {
+				memmove(file.pending+i, file.pending+i+1, 
+					(file.num_pending - (i+1)) * 
+					sizeof(struct opendb_pending));
+			}
+			break;
+		}
+	}
+
+	if (i == file.num_pending) {
+		return NT_STATUS_UNSUCCESSFUL;
+	}
+
+	file.num_pending--;
+	
+	return odb_push_record(lck, &file);
+}
+
+
+/*
+  rename the path in a open file
+*/
+static NTSTATUS odb_ctdb_rename(struct odb_lock *lck, const char *path)
+{
+	struct opendb_file file;
+	NTSTATUS status;
+
+	status = odb_pull_record(lck, &file);
+	if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_NOT_FOUND, status)) {
+		/* not having the record at all is OK */
+		return NT_STATUS_OK;
+	}
+	NT_STATUS_NOT_OK_RETURN(status);
+
+	file.path = path;
+	return odb_push_record(lck, &file);
+}
+
+/*
+  update delete on close flag on an open file
+*/
+static NTSTATUS odb_ctdb_set_delete_on_close(struct odb_lock *lck, BOOL del_on_close)
+{
+	NTSTATUS status;
+	struct opendb_file file;
+
+	status = odb_pull_record(lck, &file);
+	NT_STATUS_NOT_OK_RETURN(status);
+
+	file.delete_on_close = del_on_close;
+
+	return odb_push_record(lck, &file);
+}
+
+/*
+  return the current value of the delete_on_close bit, and how many
+  people still have the file open
+*/
+static NTSTATUS odb_ctdb_get_delete_on_close(struct odb_context *odb, 
+					    DATA_BLOB *key, BOOL *del_on_close, 
+					    int *open_count, char **path)
+{
+	NTSTATUS status;
+	struct opendb_file file;
+	struct odb_lock *lck;
+
+	lck = odb_lock(odb, odb, key);
+	NT_STATUS_HAVE_NO_MEMORY(lck);
+
+	status = odb_pull_record(lck, &file);
+	if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_NOT_FOUND, status)) {
+		talloc_free(lck);
+		(*del_on_close) = False;
+		return NT_STATUS_OK;
+	}
+	if (!NT_STATUS_IS_OK(status)) {
+		talloc_free(lck);
+		return status;
+	}
+
+	(*del_on_close) = file.delete_on_close;
+	if (open_count != NULL) {
+		(*open_count) = file.num_entries;
+	}
+	if (path != NULL) {
+		*path = talloc_strdup(odb, file.path);
+		NT_STATUS_HAVE_NO_MEMORY(*path);
+		if (file.num_entries == 1 && file.entries[0].delete_on_close) {
+			(*del_on_close) = True;
+		}
+	}
+
+	talloc_free(lck);
+
+	return NT_STATUS_OK;
+}
+
+
+/*
+  determine if a file can be opened with the given share_access,
+  create_options and access_mask
+*/
+static NTSTATUS odb_ctdb_can_open(struct odb_lock *lck,
+				 uint32_t share_access, uint32_t create_options, 
+				 uint32_t access_mask)
+{
+	struct odb_context *odb = lck->odb;
+	NTSTATUS status;
+	struct opendb_file file;
+	struct opendb_entry e;
+	int i;
+
+	status = odb_pull_record(lck, &file);
+	if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) {
+		return NT_STATUS_OK;
+	}
+	NT_STATUS_NOT_OK_RETURN(status);
+
+	if ((create_options & NTCREATEX_OPTIONS_DELETE_ON_CLOSE) && 
+	    file.num_entries != 0) {
+		return NT_STATUS_SHARING_VIOLATION;
+	}
+
+	if (file.delete_on_close) {
+		return NT_STATUS_DELETE_PENDING;
+	}
+
+	e.server       = odb->ntvfs_ctx->server_id;
+	e.file_handle  = NULL;
+	e.stream_id    = 0;
+	e.share_access = share_access;
+	e.access_mask  = access_mask;
+		
+	for (i=0;i<file.num_entries;i++) {
+		status = share_conflict(&file.entries[i], &e);
+		if (!NT_STATUS_IS_OK(status)) {
+			/* note that we discard the error code
+			   here. We do this as unless we are actually
+			   doing an open (which comes via a different
+			   function), we need to return a sharing
+			   violation */
+			return NT_STATUS_SHARING_VIOLATION;
+		}
+	}
+
+	return NT_STATUS_OK;
+}
+
+
+static const struct opendb_ops opendb_ctdb_ops = {
+	.odb_init                = odb_ctdb_init,
+	.odb_lock                = odb_ctdb_lock,
+	.odb_open_file           = odb_ctdb_open_file,
+	.odb_open_file_pending   = odb_ctdb_open_file_pending,
+	.odb_close_file          = odb_ctdb_close_file,
+	.odb_remove_pending      = odb_ctdb_remove_pending,
+	.odb_rename              = odb_ctdb_rename,
+	.odb_set_delete_on_close = odb_ctdb_set_delete_on_close,
+	.odb_get_delete_on_close = odb_ctdb_get_delete_on_close,
+	.odb_can_open            = odb_ctdb_can_open
+};
+
+
+void odb_ctdb_init_ops(void)
+{
+	odb_set_ops(&opendb_ctdb_ops);
+}
diff --git a/source4/ntvfs/common/config.mk b/source4/ntvfs/common/config.mk
index 665e8e255b..5c744f9c90 100644
--- a/source4/ntvfs/common/config.mk
+++ b/source4/ntvfs/common/config.mk
@@ -10,6 +10,6 @@ OBJ_FILES = \
 		opendb_tdb.o \
 		notify.o
 PUBLIC_DEPENDENCIES = NDR_OPENDB NDR_NOTIFY sys_notify share
-PRIVATE_DEPENDENCIES = brlock_ctdb
+PRIVATE_DEPENDENCIES = brlock_ctdb opendb_ctdb
 # End LIBRARY ntvfs_common
 ################################################
diff --git a/source4/ntvfs/common/opendb.c b/source4/ntvfs/common/opendb.c
index c17b819c04..ea27efdf3b 100644
--- a/source4/ntvfs/common/opendb.c
+++ b/source4/ntvfs/common/opendb.c
@@ -62,7 +62,11 @@ _PUBLIC_ struct odb_context *odb_init(TALLOC_CTX *mem_ctx,
 				      struct ntvfs_context *ntvfs_ctx)
 {
 	if (ops == NULL) {
-		odb_tdb_init_ops();
+		if (lp_parm_bool(-1, "ctdb", "opendb", False)) {
+			odb_ctdb_init_ops();
+		} else {
+			odb_tdb_init_ops();
+		}
 	}
 	return ops->odb_init(mem_ctx, ntvfs_ctx);
 }
diff --git a/source4/ntvfs/common/opendb.h b/source4/ntvfs/common/opendb.h
index 9d6c18c471..85bb678d77 100644
--- a/source4/ntvfs/common/opendb.h
+++ b/source4/ntvfs/common/opendb.h
@@ -46,4 +46,4 @@ struct opendb_ops {
 
 void odb_set_ops(const struct opendb_ops *new_ops);
 void odb_tdb_init_ops(void);
-
+void odb_ctdb_init_ops(void);