2 Unix SMB/CIFS implementation.
4 Copyright (C) Ronnie Sahlberg 2007
5 Copyright (C) Andrew Tridgell 2007
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>.
22 this is the open files database, ctdb backend. It implements shared
23 storage of what files are open between server instances, and
24 implements the rules of shared access to files.
26 The caller needs to provide a file_key, which specifies what file
27 they are talking about. This needs to be a unique key across all
28 filesystems, and is usually implemented in terms of a device/inode
31 Before any operations can be performed the caller needs to establish
32 a lock on the record associated with file_key. That is done by
33 calling odb_lock(). The caller releases this lock by calling
34 talloc_free() on the returned handle.
36 All other operations on a record are done by passing the odb_lock()
37 handle back to this module. The handle contains internal
38 information about what file_key is being operated on.
42 #include "system/filesys.h"
43 #include "lib/tdb/include/tdb.h"
44 #include "messaging/messaging.h"
46 #include "lib/messaging/irpc.h"
47 #include "librpc/gen_ndr/ndr_opendb.h"
48 #include "ntvfs/ntvfs.h"
49 #include "ntvfs/common/ntvfs_common.h"
50 #include "cluster/cluster.h"
51 #include "include/ctdb.h"
52 #include "param/param.h"
55 struct ctdb_context *ctdb;
56 struct ctdb_db_context *ctdb_db;
57 struct ntvfs_context *ntvfs_ctx;
62 an odb lock handle. You must obtain one of these using odb_lock() before doing
66 struct odb_context *odb;
67 struct ctdb_record_handle *rec;
73 Open up the openfiles.tdb database. Close it down using
74 talloc_free(). We need the messaging_ctx to allow for pending open
77 static struct odb_context *odb_ctdb_init(TALLOC_CTX *mem_ctx,
78 struct ntvfs_context *ntvfs_ctx)
80 struct odb_context *odb;
81 struct ctdb_context *ctdb = talloc_get_type(cluster_backend_handle(),
84 odb = talloc(mem_ctx, struct odb_context);
90 odb->ctdb_db = ctdb_attach(ctdb, "opendb");
92 DEBUG(0,("Failed to get attached ctdb db handle for opendb\n"));
97 odb->ntvfs_ctx = ntvfs_ctx;
99 /* leave oplocks disabled by default until the code is working */
100 odb->oplocks = lp_parm_bool(global_loadparm, NULL, "opendb", "oplocks", false);
106 get a lock on a entry in the odb. This call returns a lock handle,
107 which the caller should unlock using talloc_free().
109 static struct odb_lock *odb_ctdb_lock(TALLOC_CTX *mem_ctx,
110 struct odb_context *odb, DATA_BLOB *file_key)
112 struct odb_lock *lck;
114 lck = talloc(mem_ctx, struct odb_lock);
119 lck->odb = talloc_reference(lck, odb);
120 lck->key.dptr = talloc_memdup(lck, file_key->data, file_key->length);
121 lck->key.dsize = file_key->length;
122 if (lck->key.dptr == NULL) {
127 lck->rec = ctdb_fetch_lock(odb->ctdb_db, (TALLOC_CTX *)lck, lck->key, &lck->data);
137 determine if two odb_entry structures conflict
139 return NT_STATUS_OK on no conflict
141 static NTSTATUS share_conflict(struct opendb_entry *e1, struct opendb_entry *e2)
143 /* if either open involves no read.write or delete access then
145 if (!(e1->access_mask & (SEC_FILE_WRITE_DATA |
146 SEC_FILE_APPEND_DATA |
152 if (!(e2->access_mask & (SEC_FILE_WRITE_DATA |
153 SEC_FILE_APPEND_DATA |
160 /* data IO access masks. This is skipped if the two open handles
161 are on different streams (as in that case the masks don't
163 if (e1->stream_id != e2->stream_id) {
167 #define CHECK_MASK(am, right, sa, share) \
168 if (((am) & (right)) && !((sa) & (share))) return NT_STATUS_SHARING_VIOLATION
170 CHECK_MASK(e1->access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA,
171 e2->share_access, NTCREATEX_SHARE_ACCESS_WRITE);
172 CHECK_MASK(e2->access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA,
173 e1->share_access, NTCREATEX_SHARE_ACCESS_WRITE);
175 CHECK_MASK(e1->access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE,
176 e2->share_access, NTCREATEX_SHARE_ACCESS_READ);
177 CHECK_MASK(e2->access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE,
178 e1->share_access, NTCREATEX_SHARE_ACCESS_READ);
180 CHECK_MASK(e1->access_mask, SEC_STD_DELETE,
181 e2->share_access, NTCREATEX_SHARE_ACCESS_DELETE);
182 CHECK_MASK(e2->access_mask, SEC_STD_DELETE,
183 e1->share_access, NTCREATEX_SHARE_ACCESS_DELETE);
189 pull a record, translating from the db format to the opendb_file structure defined
192 static NTSTATUS odb_pull_record(struct odb_lock *lck, struct opendb_file *file)
200 if (dbuf.dsize == 0) {
201 /* empty record in ctdb means the record isn't there */
202 return NT_STATUS_OBJECT_NAME_NOT_FOUND;
205 blob.data = dbuf.dptr;
206 blob.length = dbuf.dsize;
208 status = ndr_pull_struct_blob(&blob, lck, file, (ndr_pull_flags_fn_t)ndr_pull_opendb_file);
214 push a record, translating from the opendb_file structure defined in opendb.idl
216 static NTSTATUS odb_push_record(struct odb_lock *lck, struct opendb_file *file)
223 if (!file->num_entries) {
226 ctdb_record_store(lck->rec, dbuf);
227 talloc_free(lck->rec);
231 status = ndr_push_struct_blob(&blob, lck, file, (ndr_push_flags_fn_t)ndr_push_opendb_file);
232 NT_STATUS_NOT_OK_RETURN(status);
234 dbuf.dptr = blob.data;
235 dbuf.dsize = blob.length;
237 ret = ctdb_record_store(lck->rec, dbuf);
238 talloc_free(lck->rec);
239 data_blob_free(&blob);
241 return NT_STATUS_INTERNAL_DB_CORRUPTION;
248 send an oplock break to a client
250 static NTSTATUS odb_oplock_break_send(struct odb_context *odb, struct opendb_entry *e)
252 /* tell the server handling this open file about the need to send the client
254 return messaging_send_ptr(odb->ntvfs_ctx->msg_ctx, e->server,
255 MSG_NTVFS_OPLOCK_BREAK, e->file_handle);
259 register an open file in the open files database. This implements the share_access
262 Note that the path is only used by the delete on close logic, not
263 for comparing with other filenames
265 static NTSTATUS odb_ctdb_open_file(struct odb_lock *lck, void *file_handle,
266 uint32_t stream_id, uint32_t share_access,
267 uint32_t access_mask, BOOL delete_on_close,
269 uint32_t oplock_level, uint32_t *oplock_granted)
271 struct odb_context *odb = lck->odb;
272 struct opendb_entry e;
274 struct opendb_file file;
277 if (odb->oplocks == False) {
278 oplock_level = OPLOCK_NONE;
281 status = odb_pull_record(lck, &file);
282 if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) {
283 /* initialise a blank structure */
287 NT_STATUS_NOT_OK_RETURN(status);
290 /* see if it conflicts */
291 e.server = odb->ntvfs_ctx->server_id;
292 e.file_handle = file_handle;
293 e.stream_id = stream_id;
294 e.share_access = share_access;
295 e.access_mask = access_mask;
296 e.delete_on_close = delete_on_close;
297 e.oplock_level = OPLOCK_NONE;
299 /* see if anyone has an oplock, which we need to break */
300 for (i=0;i<file.num_entries;i++) {
301 if (file.entries[i].oplock_level == OPLOCK_BATCH) {
302 /* a batch oplock caches close calls, which
303 means the client application might have
304 already closed the file. We have to allow
305 this close to propogate by sending a oplock
306 break request and suspending this call
307 until the break is acknowledged or the file
309 odb_oplock_break_send(odb, &file.entries[i]);
310 return NT_STATUS_OPLOCK_NOT_GRANTED;
314 if (file.delete_on_close ||
315 (file.num_entries != 0 && delete_on_close)) {
316 /* while delete on close is set, no new opens are allowed */
317 return NT_STATUS_DELETE_PENDING;
320 /* check for sharing violations */
321 for (i=0;i<file.num_entries;i++) {
322 status = share_conflict(&file.entries[i], &e);
323 NT_STATUS_NOT_OK_RETURN(status);
326 /* we now know the open could succeed, but we need to check
327 for any exclusive oplocks. We can't grant a second open
328 till these are broken. Note that we check for batch oplocks
329 before checking for sharing violations, and check for
330 exclusive oplocks afterwards. */
331 for (i=0;i<file.num_entries;i++) {
332 if (file.entries[i].oplock_level == OPLOCK_EXCLUSIVE) {
333 odb_oplock_break_send(odb, &file.entries[i]);
334 return NT_STATUS_OPLOCK_NOT_GRANTED;
339 possibly grant an exclusive or batch oplock if this is the only client
340 with the file open. We don't yet grant levelII oplocks.
342 if (oplock_granted != NULL) {
343 if ((oplock_level == OPLOCK_BATCH ||
344 oplock_level == OPLOCK_EXCLUSIVE) &&
345 file.num_entries == 0) {
346 (*oplock_granted) = oplock_level;
348 (*oplock_granted) = OPLOCK_NONE;
350 e.oplock_level = (*oplock_granted);
353 /* it doesn't conflict, so add it to the end */
354 file.entries = talloc_realloc(lck, file.entries, struct opendb_entry,
356 NT_STATUS_HAVE_NO_MEMORY(file.entries);
358 file.entries[file.num_entries] = e;
361 return odb_push_record(lck, &file);
366 register a pending open file in the open files database
368 static NTSTATUS odb_ctdb_open_file_pending(struct odb_lock *lck, void *private)
370 struct odb_context *odb = lck->odb;
371 struct opendb_file file;
374 status = odb_pull_record(lck, &file);
375 NT_STATUS_NOT_OK_RETURN(status);
377 file.pending = talloc_realloc(lck, file.pending, struct opendb_pending,
379 NT_STATUS_HAVE_NO_MEMORY(file.pending);
381 file.pending[file.num_pending].server = odb->ntvfs_ctx->server_id;
382 file.pending[file.num_pending].notify_ptr = private;
386 return odb_push_record(lck, &file);
391 remove a opendb entry
393 static NTSTATUS odb_ctdb_close_file(struct odb_lock *lck, void *file_handle)
395 struct odb_context *odb = lck->odb;
396 struct opendb_file file;
400 status = odb_pull_record(lck, &file);
401 NT_STATUS_NOT_OK_RETURN(status);
403 /* find the entry, and delete it */
404 for (i=0;i<file.num_entries;i++) {
405 if (file_handle == file.entries[i].file_handle &&
406 cluster_id_equal(&odb->ntvfs_ctx->server_id, &file.entries[i].server)) {
407 if (file.entries[i].delete_on_close) {
408 file.delete_on_close = True;
410 if (i < file.num_entries-1) {
411 memmove(file.entries+i, file.entries+i+1,
412 (file.num_entries - (i+1)) *
413 sizeof(struct opendb_entry));
419 if (i == file.num_entries) {
420 return NT_STATUS_UNSUCCESSFUL;
423 /* send any pending notifications, removing them once sent */
424 for (i=0;i<file.num_pending;i++) {
425 messaging_send_ptr(odb->ntvfs_ctx->msg_ctx, file.pending[i].server,
427 file.pending[i].notify_ptr);
429 file.num_pending = 0;
433 return odb_push_record(lck, &file);
438 remove a pending opendb entry
440 static NTSTATUS odb_ctdb_remove_pending(struct odb_lock *lck, void *private)
442 struct odb_context *odb = lck->odb;
445 struct opendb_file file;
447 status = odb_pull_record(lck, &file);
448 NT_STATUS_NOT_OK_RETURN(status);
450 /* find the entry, and delete it */
451 for (i=0;i<file.num_pending;i++) {
452 if (private == file.pending[i].notify_ptr &&
453 cluster_id_equal(&odb->ntvfs_ctx->server_id, &file.pending[i].server)) {
454 if (i < file.num_pending-1) {
455 memmove(file.pending+i, file.pending+i+1,
456 (file.num_pending - (i+1)) *
457 sizeof(struct opendb_pending));
463 if (i == file.num_pending) {
464 return NT_STATUS_UNSUCCESSFUL;
469 return odb_push_record(lck, &file);
474 rename the path in a open file
476 static NTSTATUS odb_ctdb_rename(struct odb_lock *lck, const char *path)
478 struct opendb_file file;
481 status = odb_pull_record(lck, &file);
482 if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_NOT_FOUND, status)) {
483 /* not having the record at all is OK */
486 NT_STATUS_NOT_OK_RETURN(status);
489 return odb_push_record(lck, &file);
493 update delete on close flag on an open file
495 static NTSTATUS odb_ctdb_set_delete_on_close(struct odb_lock *lck, BOOL del_on_close)
498 struct opendb_file file;
500 status = odb_pull_record(lck, &file);
501 NT_STATUS_NOT_OK_RETURN(status);
503 file.delete_on_close = del_on_close;
505 return odb_push_record(lck, &file);
509 return the current value of the delete_on_close bit, and how many
510 people still have the file open
512 static NTSTATUS odb_ctdb_get_delete_on_close(struct odb_context *odb,
513 DATA_BLOB *key, BOOL *del_on_close,
514 int *open_count, char **path)
517 struct opendb_file file;
518 struct odb_lock *lck;
520 lck = odb_lock(odb, odb, key);
521 NT_STATUS_HAVE_NO_MEMORY(lck);
523 status = odb_pull_record(lck, &file);
524 if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_NOT_FOUND, status)) {
526 (*del_on_close) = False;
529 if (!NT_STATUS_IS_OK(status)) {
534 (*del_on_close) = file.delete_on_close;
535 if (open_count != NULL) {
536 (*open_count) = file.num_entries;
539 *path = talloc_strdup(odb, file.path);
540 NT_STATUS_HAVE_NO_MEMORY(*path);
541 if (file.num_entries == 1 && file.entries[0].delete_on_close) {
542 (*del_on_close) = True;
553 determine if a file can be opened with the given share_access,
554 create_options and access_mask
556 static NTSTATUS odb_ctdb_can_open(struct odb_lock *lck,
557 uint32_t share_access, uint32_t create_options,
558 uint32_t access_mask)
560 struct odb_context *odb = lck->odb;
562 struct opendb_file file;
563 struct opendb_entry e;
566 status = odb_pull_record(lck, &file);
567 if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) {
570 NT_STATUS_NOT_OK_RETURN(status);
572 if ((create_options & NTCREATEX_OPTIONS_DELETE_ON_CLOSE) &&
573 file.num_entries != 0) {
574 return NT_STATUS_SHARING_VIOLATION;
577 if (file.delete_on_close) {
578 return NT_STATUS_DELETE_PENDING;
581 e.server = odb->ntvfs_ctx->server_id;
582 e.file_handle = NULL;
584 e.share_access = share_access;
585 e.access_mask = access_mask;
587 for (i=0;i<file.num_entries;i++) {
588 status = share_conflict(&file.entries[i], &e);
589 if (!NT_STATUS_IS_OK(status)) {
590 /* note that we discard the error code
591 here. We do this as unless we are actually
592 doing an open (which comes via a different
593 function), we need to return a sharing
595 return NT_STATUS_SHARING_VIOLATION;
603 static const struct opendb_ops opendb_ctdb_ops = {
604 .odb_init = odb_ctdb_init,
605 .odb_lock = odb_ctdb_lock,
606 .odb_open_file = odb_ctdb_open_file,
607 .odb_open_file_pending = odb_ctdb_open_file_pending,
608 .odb_close_file = odb_ctdb_close_file,
609 .odb_remove_pending = odb_ctdb_remove_pending,
610 .odb_rename = odb_ctdb_rename,
611 .odb_set_delete_on_close = odb_ctdb_set_delete_on_close,
612 .odb_get_delete_on_close = odb_ctdb_get_delete_on_close,
613 .odb_can_open = odb_ctdb_can_open
617 void odb_ctdb_init_ops(void)
619 odb_set_ops(&opendb_ctdb_ops);