r21807: modularise the opendb code, so different backends can be
authorAndrew Tridgell <tridge@samba.org>
Tue, 13 Mar 2007 01:47:04 +0000 (01:47 +0000)
committerGerald (Jerry) Carter <jerry@samba.org>
Wed, 10 Oct 2007 19:49:29 +0000 (14:49 -0500)
implemented. This will allow for a ctdb opendb backend.
(This used to be commit 6b91ca0968209ad17489341c78d9610607cc8320)

source4/ntvfs/common/config.mk
source4/ntvfs/common/ntvfs_common.h
source4/ntvfs/common/opendb.c
source4/ntvfs/common/opendb.h [new file with mode: 0644]
source4/ntvfs/common/opendb_tdb.c [new file with mode: 0644]

index 99a46323e2ea274dd395a75c3ab2efd91d835952..665e8e255bf24deeff98b89b57a5c2648d550e36 100644 (file)
@@ -7,6 +7,7 @@ OBJ_FILES = \
                brlock.o \
                brlock_tdb.o \
                opendb.o \
+               opendb_tdb.o \
                notify.o
 PUBLIC_DEPENDENCIES = NDR_OPENDB NDR_NOTIFY sys_notify share
 PRIVATE_DEPENDENCIES = brlock_ctdb
index ab80bb08f004b2b4000455ad19e4a72d56ff4b93..4cc7ebe7528cb6ead9e40ccd9765d0b2c84c0b0e 100644 (file)
@@ -27,6 +27,7 @@ struct notify_event;
 struct notify_entry;
 
 #include "ntvfs/common/brlock.h"
+#include "ntvfs/common/opendb.h"
 #include "ntvfs/common/proto.h"
 
 #endif /* _NTVFS_COMMON_H_ */
index 875f90760cd63861afab21df1575f06fa3df248d..c17b819c0480acea19926b09e4e4fba86d6cb195 100644 (file)
 */
 
 #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"
 
-struct odb_context {
-       struct tdb_wrap *w;
-       struct ntvfs_context *ntvfs_ctx;
-       BOOL oplocks;
-};
+static const struct opendb_ops *ops;
 
 /*
-  an odb lock handle. You must obtain one of these using odb_lock() before doing
-  any other operations. 
+  set the odb backend ops
 */
-struct odb_lock {
-       struct odb_context *odb;
-       TDB_DATA key;
-};
+void odb_set_ops(const struct opendb_ops *new_ops)
+{
+       ops = new_ops;
+}
 
 /*
   Open up the openfiles.tdb database. Close it down using
@@ -72,34 +61,10 @@ struct odb_lock {
 _PUBLIC_ struct odb_context *odb_init(TALLOC_CTX *mem_ctx, 
                                      struct ntvfs_context *ntvfs_ctx)
 {
-       struct odb_context *odb;
-
-       odb = talloc(mem_ctx, struct odb_context);
-       if (odb == NULL) {
-               return NULL;
-       }
-
-       odb->w = cluster_tdb_tmp_open(odb, "openfiles.tdb", TDB_DEFAULT);
-       if (odb->w == NULL) {
-               talloc_free(odb);
-               return NULL;
+       if (ops == NULL) {
+               odb_tdb_init_ops();
        }
-
-       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;
-}
-
-/*
-  destroy a lock on the database
-*/
-static int odb_lock_destructor(struct odb_lock *lck)
-{
-       tdb_chainunlock(lck->odb->w->tdb, lck->key);
-       return 0;
+       return ops->odb_init(mem_ctx, ntvfs_ctx);
 }
 
 /*
@@ -109,153 +74,9 @@ static int odb_lock_destructor(struct odb_lock *lck)
 _PUBLIC_ struct odb_lock *odb_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;
-       }
-
-       if (tdb_chainlock(odb->w->tdb, lck->key) != 0) {
-               talloc_free(lck);
-               return NULL;
-       }
-
-       talloc_set_destructor(lck, odb_lock_destructor);
-       
-       return lck;
+       return ops->odb_lock(mem_ctx, odb, file_key);
 }
 
-/*
-  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)
-{
-       struct odb_context *odb = lck->odb;
-       TDB_DATA dbuf;
-       DATA_BLOB blob;
-       NTSTATUS status;
-               
-       dbuf = tdb_fetch(odb->w->tdb, lck->key);
-       if (dbuf.dptr == NULL) {
-               return NT_STATUS_OBJECT_NAME_NOT_FOUND;
-       }
-
-       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);
-
-       free(dbuf.dptr);
-
-       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)
-{
-       struct odb_context *odb = lck->odb;
-       TDB_DATA dbuf;
-       DATA_BLOB blob;
-       NTSTATUS status;
-       int ret;
-
-       if (file->num_entries == 0) {
-               ret = tdb_delete(odb->w->tdb, lck->key);
-               if (ret != 0) {
-                       return NT_STATUS_INTERNAL_DB_CORRUPTION;
-               }
-               return NT_STATUS_OK;
-       }
-
-       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 = tdb_store(odb->w->tdb, lck->key, dbuf, TDB_REPLACE);
-       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
@@ -270,97 +91,9 @@ _PUBLIC_ NTSTATUS odb_open_file(struct odb_lock *lck, void *file_handle,
                                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);
+       return ops->odb_open_file(lck, file_handle, stream_id, share_access,
+                                 access_mask, delete_on_close, path, oplock_level,
+                                 oplock_granted);
 }
 
 
@@ -369,23 +102,7 @@ _PUBLIC_ NTSTATUS odb_open_file(struct odb_lock *lck, void *file_handle,
 */
 _PUBLIC_ NTSTATUS odb_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);
+       return ops->odb_open_file_pending(lck, private);
 }
 
 
@@ -394,45 +111,7 @@ _PUBLIC_ NTSTATUS odb_open_file_pending(struct odb_lock *lck, void *private)
 */
 _PUBLIC_ NTSTATUS odb_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);
+       return ops->odb_close_file(lck, file_handle);
 }
 
 
@@ -441,34 +120,7 @@ _PUBLIC_ NTSTATUS odb_close_file(struct odb_lock *lck, void *file_handle)
 */
 _PUBLIC_ NTSTATUS odb_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);
+       return ops->odb_remove_pending(lck, private);
 }
 
 
@@ -477,18 +129,7 @@ _PUBLIC_ NTSTATUS odb_remove_pending(struct odb_lock *lck, void *private)
 */
 _PUBLIC_ NTSTATUS odb_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);
+       return ops->odb_rename(lck, path);
 }
 
 /*
@@ -496,15 +137,7 @@ _PUBLIC_ NTSTATUS odb_rename(struct odb_lock *lck, const char *path)
 */
 _PUBLIC_ NTSTATUS odb_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 ops->odb_set_delete_on_close(lck, del_on_close);
 }
 
 /*
@@ -515,39 +148,7 @@ _PUBLIC_ NTSTATUS odb_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;
+       return ops->odb_get_delete_on_close(odb, key, del_on_close, open_count, path);
 }
 
 
@@ -559,44 +160,5 @@ _PUBLIC_ NTSTATUS odb_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;
+       return ops->odb_can_open(lck, share_access, create_options, access_mask);
 }
diff --git a/source4/ntvfs/common/opendb.h b/source4/ntvfs/common/opendb.h
new file mode 100644 (file)
index 0000000..9d6c18c
--- /dev/null
@@ -0,0 +1,49 @@
+/* 
+   Unix SMB/CIFS implementation.
+
+   open database code - common include
+
+   Copyright (C) Andrew Tridgell 2007
+   
+   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.
+*/
+
+struct opendb_ops {
+       struct odb_context *(*odb_init)(TALLOC_CTX *mem_ctx, 
+                                       struct ntvfs_context *ntvfs_ctx);
+       struct odb_lock *(*odb_lock)(TALLOC_CTX *mem_ctx,
+                                    struct odb_context *odb, DATA_BLOB *file_key);
+       NTSTATUS (*odb_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);
+       NTSTATUS (*odb_open_file_pending)(struct odb_lock *lck, void *private);
+       NTSTATUS (*odb_close_file)(struct odb_lock *lck, void *file_handle);
+       NTSTATUS (*odb_remove_pending)(struct odb_lock *lck, void *private);
+       NTSTATUS (*odb_rename)(struct odb_lock *lck, const char *path);
+       NTSTATUS (*odb_set_delete_on_close)(struct odb_lock *lck, BOOL del_on_close);
+       NTSTATUS (*odb_get_delete_on_close)(struct odb_context *odb, 
+                                           DATA_BLOB *key, BOOL *del_on_close, 
+                                           int *open_count, char **path);
+       NTSTATUS (*odb_can_open)(struct odb_lock *lck,
+                                uint32_t share_access, uint32_t create_options, 
+                                uint32_t access_mask);
+};
+
+
+void odb_set_ops(const struct opendb_ops *new_ops);
+void odb_tdb_init_ops(void);
+
diff --git a/source4/ntvfs/common/opendb_tdb.c b/source4/ntvfs/common/opendb_tdb.c
new file mode 100644 (file)
index 0000000..c181b03
--- /dev/null
@@ -0,0 +1,622 @@
+/* 
+   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, tdb 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"
+
+struct odb_context {
+       struct tdb_wrap *w;
+       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;
+       TDB_DATA key;
+};
+
+/*
+  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_tdb_init(TALLOC_CTX *mem_ctx, 
+                                       struct ntvfs_context *ntvfs_ctx)
+{
+       struct odb_context *odb;
+
+       odb = talloc(mem_ctx, struct odb_context);
+       if (odb == NULL) {
+               return NULL;
+       }
+
+       odb->w = cluster_tdb_tmp_open(odb, "openfiles.tdb", TDB_DEFAULT);
+       if (odb->w == NULL) {
+               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;
+}
+
+/*
+  destroy a lock on the database
+*/
+static int odb_lock_destructor(struct odb_lock *lck)
+{
+       tdb_chainunlock(lck->odb->w->tdb, lck->key);
+       return 0;
+}
+
+/*
+  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_tdb_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;
+       }
+
+       if (tdb_chainlock(odb->w->tdb, lck->key) != 0) {
+               talloc_free(lck);
+               return NULL;
+       }
+
+       talloc_set_destructor(lck, odb_lock_destructor);
+       
+       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)
+{
+       struct odb_context *odb = lck->odb;
+       TDB_DATA dbuf;
+       DATA_BLOB blob;
+       NTSTATUS status;
+               
+       dbuf = tdb_fetch(odb->w->tdb, lck->key);
+       if (dbuf.dptr == NULL) {
+               return NT_STATUS_OBJECT_NAME_NOT_FOUND;
+       }
+
+       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);
+
+       free(dbuf.dptr);
+
+       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)
+{
+       struct odb_context *odb = lck->odb;
+       TDB_DATA dbuf;
+       DATA_BLOB blob;
+       NTSTATUS status;
+       int ret;
+
+       if (file->num_entries == 0) {
+               ret = tdb_delete(odb->w->tdb, lck->key);
+               if (ret != 0) {
+                       return NT_STATUS_INTERNAL_DB_CORRUPTION;
+               }
+               return NT_STATUS_OK;
+       }
+
+       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 = tdb_store(odb->w->tdb, lck->key, dbuf, TDB_REPLACE);
+       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_tdb_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_tdb_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_tdb_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_tdb_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_tdb_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_tdb_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_tdb_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_tdb_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_tdb_ops = {
+       .odb_init                = odb_tdb_init,
+       .odb_lock                = odb_tdb_lock,
+       .odb_open_file           = odb_tdb_open_file,
+       .odb_open_file_pending   = odb_tdb_open_file_pending,
+       .odb_close_file          = odb_tdb_close_file,
+       .odb_remove_pending      = odb_tdb_remove_pending,
+       .odb_rename              = odb_tdb_rename,
+       .odb_set_delete_on_close = odb_tdb_set_delete_on_close,
+       .odb_get_delete_on_close = odb_tdb_get_delete_on_close,
+       .odb_can_open            = odb_tdb_can_open
+};
+
+
+void odb_tdb_init_ops(void)
+{
+       odb_set_ops(&opendb_tdb_ops);
+}