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"
54 struct ctdb_context *ctdb;
55 struct ctdb_db_context *ctdb_db;
56 struct ntvfs_context *ntvfs_ctx;
61 an odb lock handle. You must obtain one of these using odb_lock() before doing
65 struct odb_context *odb;
66 struct ctdb_record_handle *rec;
72 Open up the openfiles.tdb database. Close it down using
73 talloc_free(). We need the messaging_ctx to allow for pending open
76 static struct odb_context *odb_ctdb_init(TALLOC_CTX *mem_ctx,
77 struct ntvfs_context *ntvfs_ctx)
79 struct odb_context *odb;
80 struct ctdb_context *ctdb = talloc_get_type(cluster_backend_handle(),
83 odb = talloc(mem_ctx, struct odb_context);
89 odb->ctdb_db = ctdb_db_handle(ctdb, "opendb");
91 DEBUG(0,("Failed to get attached ctdb db handle for opendb\n"));
96 odb->ntvfs_ctx = ntvfs_ctx;
98 /* leave oplocks disabled by default until the code is working */
99 odb->oplocks = lp_parm_bool(-1, "opendb", "oplocks", False);
105 get a lock on a entry in the odb. This call returns a lock handle,
106 which the caller should unlock using talloc_free().
108 static struct odb_lock *odb_ctdb_lock(TALLOC_CTX *mem_ctx,
109 struct odb_context *odb, DATA_BLOB *file_key)
111 struct odb_lock *lck;
113 lck = talloc(mem_ctx, struct odb_lock);
118 lck->odb = talloc_reference(lck, odb);
119 lck->key.dptr = talloc_memdup(lck, file_key->data, file_key->length);
120 lck->key.dsize = file_key->length;
121 if (lck->key.dptr == NULL) {
126 lck->rec = ctdb_fetch_lock(odb->ctdb_db, (TALLOC_CTX *)lck, lck->key, &lck->data);
136 determine if two odb_entry structures conflict
138 return NT_STATUS_OK on no conflict
140 static NTSTATUS share_conflict(struct opendb_entry *e1, struct opendb_entry *e2)
142 /* if either open involves no read.write or delete access then
144 if (!(e1->access_mask & (SEC_FILE_WRITE_DATA |
145 SEC_FILE_APPEND_DATA |
151 if (!(e2->access_mask & (SEC_FILE_WRITE_DATA |
152 SEC_FILE_APPEND_DATA |
159 /* data IO access masks. This is skipped if the two open handles
160 are on different streams (as in that case the masks don't
162 if (e1->stream_id != e2->stream_id) {
166 #define CHECK_MASK(am, right, sa, share) \
167 if (((am) & (right)) && !((sa) & (share))) return NT_STATUS_SHARING_VIOLATION
169 CHECK_MASK(e1->access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA,
170 e2->share_access, NTCREATEX_SHARE_ACCESS_WRITE);
171 CHECK_MASK(e2->access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA,
172 e1->share_access, NTCREATEX_SHARE_ACCESS_WRITE);
174 CHECK_MASK(e1->access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE,
175 e2->share_access, NTCREATEX_SHARE_ACCESS_READ);
176 CHECK_MASK(e2->access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE,
177 e1->share_access, NTCREATEX_SHARE_ACCESS_READ);
179 CHECK_MASK(e1->access_mask, SEC_STD_DELETE,
180 e2->share_access, NTCREATEX_SHARE_ACCESS_DELETE);
181 CHECK_MASK(e2->access_mask, SEC_STD_DELETE,
182 e1->share_access, NTCREATEX_SHARE_ACCESS_DELETE);
188 pull a record, translating from the db format to the opendb_file structure defined
191 static NTSTATUS odb_pull_record(struct odb_lock *lck, struct opendb_file *file)
199 if (dbuf.dsize == 0) {
200 /* empty record in ctdb means the record isn't there */
201 return NT_STATUS_OBJECT_NAME_NOT_FOUND;
204 blob.data = dbuf.dptr;
205 blob.length = dbuf.dsize;
207 status = ndr_pull_struct_blob(&blob, lck, file, (ndr_pull_flags_fn_t)ndr_pull_opendb_file);
213 push a record, translating from the opendb_file structure defined in opendb.idl
215 static NTSTATUS odb_push_record(struct odb_lock *lck, struct opendb_file *file)
222 if (!file->num_entries) {
225 ctdb_record_store(lck->rec, dbuf);
226 talloc_free(lck->rec);
230 status = ndr_push_struct_blob(&blob, lck, file, (ndr_push_flags_fn_t)ndr_push_opendb_file);
231 NT_STATUS_NOT_OK_RETURN(status);
233 dbuf.dptr = blob.data;
234 dbuf.dsize = blob.length;
236 ret = ctdb_record_store(lck->rec, dbuf);
237 talloc_free(lck->rec);
238 data_blob_free(&blob);
240 return NT_STATUS_INTERNAL_DB_CORRUPTION;
247 send an oplock break to a client
249 static NTSTATUS odb_oplock_break_send(struct odb_context *odb, struct opendb_entry *e)
251 /* tell the server handling this open file about the need to send the client
253 return messaging_send_ptr(odb->ntvfs_ctx->msg_ctx, e->server,
254 MSG_NTVFS_OPLOCK_BREAK, e->file_handle);
258 register an open file in the open files database. This implements the share_access
261 Note that the path is only used by the delete on close logic, not
262 for comparing with other filenames
264 static NTSTATUS odb_ctdb_open_file(struct odb_lock *lck, void *file_handle,
265 uint32_t stream_id, uint32_t share_access,
266 uint32_t access_mask, BOOL delete_on_close,
268 uint32_t oplock_level, uint32_t *oplock_granted)
270 struct odb_context *odb = lck->odb;
271 struct opendb_entry e;
273 struct opendb_file file;
276 if (odb->oplocks == False) {
277 oplock_level = OPLOCK_NONE;
280 status = odb_pull_record(lck, &file);
281 if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) {
282 /* initialise a blank structure */
286 NT_STATUS_NOT_OK_RETURN(status);
289 /* see if it conflicts */
290 e.server = odb->ntvfs_ctx->server_id;
291 e.file_handle = file_handle;
292 e.stream_id = stream_id;
293 e.share_access = share_access;
294 e.access_mask = access_mask;
295 e.delete_on_close = delete_on_close;
296 e.oplock_level = OPLOCK_NONE;
298 /* see if anyone has an oplock, which we need to break */
299 for (i=0;i<file.num_entries;i++) {
300 if (file.entries[i].oplock_level == OPLOCK_BATCH) {
301 /* a batch oplock caches close calls, which
302 means the client application might have
303 already closed the file. We have to allow
304 this close to propogate by sending a oplock
305 break request and suspending this call
306 until the break is acknowledged or the file
308 odb_oplock_break_send(odb, &file.entries[i]);
309 return NT_STATUS_OPLOCK_NOT_GRANTED;
313 if (file.delete_on_close ||
314 (file.num_entries != 0 && delete_on_close)) {
315 /* while delete on close is set, no new opens are allowed */
316 return NT_STATUS_DELETE_PENDING;
319 /* check for sharing violations */
320 for (i=0;i<file.num_entries;i++) {
321 status = share_conflict(&file.entries[i], &e);
322 NT_STATUS_NOT_OK_RETURN(status);
325 /* we now know the open could succeed, but we need to check
326 for any exclusive oplocks. We can't grant a second open
327 till these are broken. Note that we check for batch oplocks
328 before checking for sharing violations, and check for
329 exclusive oplocks afterwards. */
330 for (i=0;i<file.num_entries;i++) {
331 if (file.entries[i].oplock_level == OPLOCK_EXCLUSIVE) {
332 odb_oplock_break_send(odb, &file.entries[i]);
333 return NT_STATUS_OPLOCK_NOT_GRANTED;
338 possibly grant an exclusive or batch oplock if this is the only client
339 with the file open. We don't yet grant levelII oplocks.
341 if (oplock_granted != NULL) {
342 if ((oplock_level == OPLOCK_BATCH ||
343 oplock_level == OPLOCK_EXCLUSIVE) &&
344 file.num_entries == 0) {
345 (*oplock_granted) = oplock_level;
347 (*oplock_granted) = OPLOCK_NONE;
349 e.oplock_level = (*oplock_granted);
352 /* it doesn't conflict, so add it to the end */
353 file.entries = talloc_realloc(lck, file.entries, struct opendb_entry,
355 NT_STATUS_HAVE_NO_MEMORY(file.entries);
357 file.entries[file.num_entries] = e;
360 return odb_push_record(lck, &file);
365 register a pending open file in the open files database
367 static NTSTATUS odb_ctdb_open_file_pending(struct odb_lock *lck, void *private)
369 struct odb_context *odb = lck->odb;
370 struct opendb_file file;
373 status = odb_pull_record(lck, &file);
374 NT_STATUS_NOT_OK_RETURN(status);
376 file.pending = talloc_realloc(lck, file.pending, struct opendb_pending,
378 NT_STATUS_HAVE_NO_MEMORY(file.pending);
380 file.pending[file.num_pending].server = odb->ntvfs_ctx->server_id;
381 file.pending[file.num_pending].notify_ptr = private;
385 return odb_push_record(lck, &file);
390 remove a opendb entry
392 static NTSTATUS odb_ctdb_close_file(struct odb_lock *lck, void *file_handle)
394 struct odb_context *odb = lck->odb;
395 struct opendb_file file;
399 status = odb_pull_record(lck, &file);
400 NT_STATUS_NOT_OK_RETURN(status);
402 /* find the entry, and delete it */
403 for (i=0;i<file.num_entries;i++) {
404 if (file_handle == file.entries[i].file_handle &&
405 cluster_id_equal(&odb->ntvfs_ctx->server_id, &file.entries[i].server)) {
406 if (file.entries[i].delete_on_close) {
407 file.delete_on_close = True;
409 if (i < file.num_entries-1) {
410 memmove(file.entries+i, file.entries+i+1,
411 (file.num_entries - (i+1)) *
412 sizeof(struct opendb_entry));
418 if (i == file.num_entries) {
419 return NT_STATUS_UNSUCCESSFUL;
422 /* send any pending notifications, removing them once sent */
423 for (i=0;i<file.num_pending;i++) {
424 messaging_send_ptr(odb->ntvfs_ctx->msg_ctx, file.pending[i].server,
426 file.pending[i].notify_ptr);
428 file.num_pending = 0;
432 return odb_push_record(lck, &file);
437 remove a pending opendb entry
439 static NTSTATUS odb_ctdb_remove_pending(struct odb_lock *lck, void *private)
441 struct odb_context *odb = lck->odb;
444 struct opendb_file file;
446 status = odb_pull_record(lck, &file);
447 NT_STATUS_NOT_OK_RETURN(status);
449 /* find the entry, and delete it */
450 for (i=0;i<file.num_pending;i++) {
451 if (private == file.pending[i].notify_ptr &&
452 cluster_id_equal(&odb->ntvfs_ctx->server_id, &file.pending[i].server)) {
453 if (i < file.num_pending-1) {
454 memmove(file.pending+i, file.pending+i+1,
455 (file.num_pending - (i+1)) *
456 sizeof(struct opendb_pending));
462 if (i == file.num_pending) {
463 return NT_STATUS_UNSUCCESSFUL;
468 return odb_push_record(lck, &file);
473 rename the path in a open file
475 static NTSTATUS odb_ctdb_rename(struct odb_lock *lck, const char *path)
477 struct opendb_file file;
480 status = odb_pull_record(lck, &file);
481 if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_NOT_FOUND, status)) {
482 /* not having the record at all is OK */
485 NT_STATUS_NOT_OK_RETURN(status);
488 return odb_push_record(lck, &file);
492 update delete on close flag on an open file
494 static NTSTATUS odb_ctdb_set_delete_on_close(struct odb_lock *lck, BOOL del_on_close)
497 struct opendb_file file;
499 status = odb_pull_record(lck, &file);
500 NT_STATUS_NOT_OK_RETURN(status);
502 file.delete_on_close = del_on_close;
504 return odb_push_record(lck, &file);
508 return the current value of the delete_on_close bit, and how many
509 people still have the file open
511 static NTSTATUS odb_ctdb_get_delete_on_close(struct odb_context *odb,
512 DATA_BLOB *key, BOOL *del_on_close,
513 int *open_count, char **path)
516 struct opendb_file file;
517 struct odb_lock *lck;
519 lck = odb_lock(odb, odb, key);
520 NT_STATUS_HAVE_NO_MEMORY(lck);
522 status = odb_pull_record(lck, &file);
523 if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_NOT_FOUND, status)) {
525 (*del_on_close) = False;
528 if (!NT_STATUS_IS_OK(status)) {
533 (*del_on_close) = file.delete_on_close;
534 if (open_count != NULL) {
535 (*open_count) = file.num_entries;
538 *path = talloc_strdup(odb, file.path);
539 NT_STATUS_HAVE_NO_MEMORY(*path);
540 if (file.num_entries == 1 && file.entries[0].delete_on_close) {
541 (*del_on_close) = True;
552 determine if a file can be opened with the given share_access,
553 create_options and access_mask
555 static NTSTATUS odb_ctdb_can_open(struct odb_lock *lck,
556 uint32_t share_access, uint32_t create_options,
557 uint32_t access_mask)
559 struct odb_context *odb = lck->odb;
561 struct opendb_file file;
562 struct opendb_entry e;
565 status = odb_pull_record(lck, &file);
566 if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) {
569 NT_STATUS_NOT_OK_RETURN(status);
571 if ((create_options & NTCREATEX_OPTIONS_DELETE_ON_CLOSE) &&
572 file.num_entries != 0) {
573 return NT_STATUS_SHARING_VIOLATION;
576 if (file.delete_on_close) {
577 return NT_STATUS_DELETE_PENDING;
580 e.server = odb->ntvfs_ctx->server_id;
581 e.file_handle = NULL;
583 e.share_access = share_access;
584 e.access_mask = access_mask;
586 for (i=0;i<file.num_entries;i++) {
587 status = share_conflict(&file.entries[i], &e);
588 if (!NT_STATUS_IS_OK(status)) {
589 /* note that we discard the error code
590 here. We do this as unless we are actually
591 doing an open (which comes via a different
592 function), we need to return a sharing
594 return NT_STATUS_SHARING_VIOLATION;
602 static const struct opendb_ops opendb_ctdb_ops = {
603 .odb_init = odb_ctdb_init,
604 .odb_lock = odb_ctdb_lock,
605 .odb_open_file = odb_ctdb_open_file,
606 .odb_open_file_pending = odb_ctdb_open_file_pending,
607 .odb_close_file = odb_ctdb_close_file,
608 .odb_remove_pending = odb_ctdb_remove_pending,
609 .odb_rename = odb_ctdb_rename,
610 .odb_set_delete_on_close = odb_ctdb_set_delete_on_close,
611 .odb_get_delete_on_close = odb_ctdb_get_delete_on_close,
612 .odb_can_open = odb_ctdb_can_open
616 void odb_ctdb_init_ops(void)
618 odb_set_ops(&opendb_ctdb_ops);