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(ntvfs_ctx->lp_ctx, 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);
136 static DATA_BLOB odb_ctdb_get_key(TALLOC_CTX *mem_ctx, struct odb_lock *lck)
139 * as this file will went away and isn't used yet,
140 * copy the implementation from the tdb backend
143 return data_blob_const(NULL, 0);
147 determine if two odb_entry structures conflict
149 return NT_STATUS_OK on no conflict
151 static NTSTATUS share_conflict(struct opendb_entry *e1, struct opendb_entry *e2)
153 /* if either open involves no read.write or delete access then
155 if (!(e1->access_mask & (SEC_FILE_WRITE_DATA |
156 SEC_FILE_APPEND_DATA |
162 if (!(e2->access_mask & (SEC_FILE_WRITE_DATA |
163 SEC_FILE_APPEND_DATA |
170 /* data IO access masks. This is skipped if the two open handles
171 are on different streams (as in that case the masks don't
173 if (e1->stream_id != e2->stream_id) {
177 #define CHECK_MASK(am, right, sa, share) \
178 if (((am) & (right)) && !((sa) & (share))) return NT_STATUS_SHARING_VIOLATION
180 CHECK_MASK(e1->access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA,
181 e2->share_access, NTCREATEX_SHARE_ACCESS_WRITE);
182 CHECK_MASK(e2->access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA,
183 e1->share_access, NTCREATEX_SHARE_ACCESS_WRITE);
185 CHECK_MASK(e1->access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE,
186 e2->share_access, NTCREATEX_SHARE_ACCESS_READ);
187 CHECK_MASK(e2->access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE,
188 e1->share_access, NTCREATEX_SHARE_ACCESS_READ);
190 CHECK_MASK(e1->access_mask, SEC_STD_DELETE,
191 e2->share_access, NTCREATEX_SHARE_ACCESS_DELETE);
192 CHECK_MASK(e2->access_mask, SEC_STD_DELETE,
193 e1->share_access, NTCREATEX_SHARE_ACCESS_DELETE);
199 pull a record, translating from the db format to the opendb_file structure defined
202 static NTSTATUS odb_pull_record(struct odb_lock *lck, struct opendb_file *file)
206 enum ndr_err_code ndr_err;
210 if (dbuf.dsize == 0) {
211 /* empty record in ctdb means the record isn't there */
212 return NT_STATUS_OBJECT_NAME_NOT_FOUND;
215 blob.data = dbuf.dptr;
216 blob.length = dbuf.dsize;
218 ndr_err = ndr_pull_struct_blob(&blob, lck, lp_iconv_convenience(lck->odb->ntvfs_ctx->lp_ctx), file, (ndr_pull_flags_fn_t)ndr_pull_opendb_file);
219 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
220 return ndr_map_error2ntstatus(ndr_err);
227 push a record, translating from the opendb_file structure defined in opendb.idl
229 static NTSTATUS odb_push_record(struct odb_lock *lck, struct opendb_file *file)
233 enum ndr_err_code ndr_err;
236 if (!file->num_entries) {
239 ctdb_record_store(lck->rec, dbuf);
240 talloc_free(lck->rec);
244 ndr_err = ndr_push_struct_blob(&blob, lck,
245 lp_iconv_convenience(lck->odb->ntvfs_ctx->lp_ctx),
246 file, (ndr_push_flags_fn_t)ndr_push_opendb_file);
247 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
248 return ndr_map_error2ntstatus(ndr_err);
251 dbuf.dptr = blob.data;
252 dbuf.dsize = blob.length;
254 ret = ctdb_record_store(lck->rec, dbuf);
255 talloc_free(lck->rec);
256 data_blob_free(&blob);
258 return NT_STATUS_INTERNAL_DB_CORRUPTION;
265 send an oplock break to a client
267 static NTSTATUS odb_oplock_break_send(struct odb_context *odb, struct opendb_entry *e)
269 /* tell the server handling this open file about the need to send the client
271 return messaging_send_ptr(odb->ntvfs_ctx->msg_ctx, e->server,
272 MSG_NTVFS_OPLOCK_BREAK, e->file_handle);
276 register an open file in the open files database. This implements the share_access
279 Note that the path is only used by the delete on close logic, not
280 for comparing with other filenames
282 static NTSTATUS odb_ctdb_open_file(struct odb_lock *lck, void *file_handle,
283 uint32_t stream_id, uint32_t share_access,
284 uint32_t access_mask, bool delete_on_close,
286 uint32_t oplock_level, uint32_t *oplock_granted)
288 struct odb_context *odb = lck->odb;
289 struct opendb_entry e;
291 struct opendb_file file;
294 if (odb->oplocks == false) {
295 oplock_level = OPLOCK_NONE;
298 status = odb_pull_record(lck, &file);
299 if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) {
300 /* initialise a blank structure */
304 NT_STATUS_NOT_OK_RETURN(status);
307 /* see if it conflicts */
308 e.server = odb->ntvfs_ctx->server_id;
309 e.file_handle = file_handle;
310 e.stream_id = stream_id;
311 e.share_access = share_access;
312 e.access_mask = access_mask;
313 e.delete_on_close = delete_on_close;
314 e.oplock_level = OPLOCK_NONE;
316 /* see if anyone has an oplock, which we need to break */
317 for (i=0;i<file.num_entries;i++) {
318 if (file.entries[i].oplock_level == OPLOCK_BATCH) {
319 /* a batch oplock caches close calls, which
320 means the client application might have
321 already closed the file. We have to allow
322 this close to propogate by sending a oplock
323 break request and suspending this call
324 until the break is acknowledged or the file
326 odb_oplock_break_send(odb, &file.entries[i]);
327 return NT_STATUS_OPLOCK_NOT_GRANTED;
331 if (file.delete_on_close ||
332 (file.num_entries != 0 && delete_on_close)) {
333 /* while delete on close is set, no new opens are allowed */
334 return NT_STATUS_DELETE_PENDING;
337 /* check for sharing violations */
338 for (i=0;i<file.num_entries;i++) {
339 status = share_conflict(&file.entries[i], &e);
340 NT_STATUS_NOT_OK_RETURN(status);
343 /* we now know the open could succeed, but we need to check
344 for any exclusive oplocks. We can't grant a second open
345 till these are broken. Note that we check for batch oplocks
346 before checking for sharing violations, and check for
347 exclusive oplocks afterwards. */
348 for (i=0;i<file.num_entries;i++) {
349 if (file.entries[i].oplock_level == OPLOCK_EXCLUSIVE) {
350 odb_oplock_break_send(odb, &file.entries[i]);
351 return NT_STATUS_OPLOCK_NOT_GRANTED;
356 possibly grant an exclusive or batch oplock if this is the only client
357 with the file open. We don't yet grant levelII oplocks.
359 if (oplock_granted != NULL) {
360 if ((oplock_level == OPLOCK_BATCH ||
361 oplock_level == OPLOCK_EXCLUSIVE) &&
362 file.num_entries == 0) {
363 (*oplock_granted) = oplock_level;
365 (*oplock_granted) = OPLOCK_NONE;
367 e.oplock_level = (*oplock_granted);
370 /* it doesn't conflict, so add it to the end */
371 file.entries = talloc_realloc(lck, file.entries, struct opendb_entry,
373 NT_STATUS_HAVE_NO_MEMORY(file.entries);
375 file.entries[file.num_entries] = e;
378 return odb_push_record(lck, &file);
383 register a pending open file in the open files database
385 static NTSTATUS odb_ctdb_open_file_pending(struct odb_lock *lck, void *private)
387 struct odb_context *odb = lck->odb;
388 struct opendb_file file;
391 status = odb_pull_record(lck, &file);
392 NT_STATUS_NOT_OK_RETURN(status);
394 file.pending = talloc_realloc(lck, file.pending, struct opendb_pending,
396 NT_STATUS_HAVE_NO_MEMORY(file.pending);
398 file.pending[file.num_pending].server = odb->ntvfs_ctx->server_id;
399 file.pending[file.num_pending].notify_ptr = private;
403 return odb_push_record(lck, &file);
408 remove a opendb entry
410 static NTSTATUS odb_ctdb_close_file(struct odb_lock *lck, void *file_handle)
412 struct odb_context *odb = lck->odb;
413 struct opendb_file file;
417 status = odb_pull_record(lck, &file);
418 NT_STATUS_NOT_OK_RETURN(status);
420 /* find the entry, and delete it */
421 for (i=0;i<file.num_entries;i++) {
422 if (file_handle == file.entries[i].file_handle &&
423 cluster_id_equal(&odb->ntvfs_ctx->server_id, &file.entries[i].server)) {
424 if (file.entries[i].delete_on_close) {
425 file.delete_on_close = true;
427 if (i < file.num_entries-1) {
428 memmove(file.entries+i, file.entries+i+1,
429 (file.num_entries - (i+1)) *
430 sizeof(struct opendb_entry));
436 if (i == file.num_entries) {
437 return NT_STATUS_UNSUCCESSFUL;
440 /* send any pending notifications, removing them once sent */
441 for (i=0;i<file.num_pending;i++) {
442 messaging_send_ptr(odb->ntvfs_ctx->msg_ctx, file.pending[i].server,
444 file.pending[i].notify_ptr);
446 file.num_pending = 0;
450 return odb_push_record(lck, &file);
454 update the oplock level of the client
456 static NTSTATUS odb_ctdb_update_oplock(struct odb_lock *lck, void *file_handle,
457 uint32_t oplock_level)
460 * as this file will went away and isn't used yet,
461 * copy the implementation from the tdb backend
464 return NT_STATUS_FOOBAR;
468 remove a pending opendb entry
470 static NTSTATUS odb_ctdb_remove_pending(struct odb_lock *lck, void *private)
472 struct odb_context *odb = lck->odb;
475 struct opendb_file file;
477 status = odb_pull_record(lck, &file);
478 NT_STATUS_NOT_OK_RETURN(status);
480 /* find the entry, and delete it */
481 for (i=0;i<file.num_pending;i++) {
482 if (private == file.pending[i].notify_ptr &&
483 cluster_id_equal(&odb->ntvfs_ctx->server_id, &file.pending[i].server)) {
484 if (i < file.num_pending-1) {
485 memmove(file.pending+i, file.pending+i+1,
486 (file.num_pending - (i+1)) *
487 sizeof(struct opendb_pending));
493 if (i == file.num_pending) {
494 return NT_STATUS_UNSUCCESSFUL;
499 return odb_push_record(lck, &file);
504 rename the path in a open file
506 static NTSTATUS odb_ctdb_rename(struct odb_lock *lck, const char *path)
508 struct opendb_file file;
511 status = odb_pull_record(lck, &file);
512 if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_NOT_FOUND, status)) {
513 /* not having the record at all is OK */
516 NT_STATUS_NOT_OK_RETURN(status);
519 return odb_push_record(lck, &file);
523 update delete on close flag on an open file
525 static NTSTATUS odb_ctdb_set_delete_on_close(struct odb_lock *lck, bool del_on_close)
528 struct opendb_file file;
530 status = odb_pull_record(lck, &file);
531 NT_STATUS_NOT_OK_RETURN(status);
533 file.delete_on_close = del_on_close;
535 return odb_push_record(lck, &file);
539 return the current value of the delete_on_close bit, and how many
540 people still have the file open
542 static NTSTATUS odb_ctdb_get_delete_on_close(struct odb_context *odb,
543 DATA_BLOB *key, bool *del_on_close,
544 int *open_count, char **path)
547 struct opendb_file file;
548 struct odb_lock *lck;
550 lck = odb_lock(odb, odb, key);
551 NT_STATUS_HAVE_NO_MEMORY(lck);
553 status = odb_pull_record(lck, &file);
554 if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_NOT_FOUND, status)) {
556 (*del_on_close) = false;
559 if (!NT_STATUS_IS_OK(status)) {
564 (*del_on_close) = file.delete_on_close;
565 if (open_count != NULL) {
566 (*open_count) = file.num_entries;
569 *path = talloc_strdup(odb, file.path);
570 NT_STATUS_HAVE_NO_MEMORY(*path);
571 if (file.num_entries == 1 && file.entries[0].delete_on_close) {
572 (*del_on_close) = true;
583 determine if a file can be opened with the given share_access,
584 create_options and access_mask
586 static NTSTATUS odb_ctdb_can_open(struct odb_lock *lck,
587 uint32_t share_access, uint32_t create_options,
588 uint32_t access_mask)
590 struct odb_context *odb = lck->odb;
592 struct opendb_file file;
593 struct opendb_entry e;
596 status = odb_pull_record(lck, &file);
597 if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) {
600 NT_STATUS_NOT_OK_RETURN(status);
602 if ((create_options & NTCREATEX_OPTIONS_DELETE_ON_CLOSE) &&
603 file.num_entries != 0) {
604 return NT_STATUS_SHARING_VIOLATION;
607 if (file.delete_on_close) {
608 return NT_STATUS_DELETE_PENDING;
611 e.server = odb->ntvfs_ctx->server_id;
612 e.file_handle = NULL;
614 e.share_access = share_access;
615 e.access_mask = access_mask;
617 for (i=0;i<file.num_entries;i++) {
618 status = share_conflict(&file.entries[i], &e);
619 if (!NT_STATUS_IS_OK(status)) {
620 /* note that we discard the error code
621 here. We do this as unless we are actually
622 doing an open (which comes via a different
623 function), we need to return a sharing
625 return NT_STATUS_SHARING_VIOLATION;
633 static const struct opendb_ops opendb_ctdb_ops = {
634 .odb_init = odb_ctdb_init,
635 .odb_lock = odb_ctdb_lock,
636 .odb_get_key = odb_ctdb_get_key,
637 .odb_open_file = odb_ctdb_open_file,
638 .odb_open_file_pending = odb_ctdb_open_file_pending,
639 .odb_close_file = odb_ctdb_close_file,
640 .odb_remove_pending = odb_ctdb_remove_pending,
641 .odb_rename = odb_ctdb_rename,
642 .odb_set_delete_on_close = odb_ctdb_set_delete_on_close,
643 .odb_get_delete_on_close = odb_ctdb_get_delete_on_close,
644 .odb_can_open = odb_ctdb_can_open,
645 .odb_update_oplock = odb_ctdb_update_oplock
649 void odb_ctdb_init_ops(void)
651 odb_set_ops(&opendb_ctdb_ops);