2 Unix SMB/CIFS implementation.
3 global locks based on dbwrap and messaging
4 Copyright (C) 2009 by Volker Lendecke
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program. If not, see <http://www.gnu.org/licenses/>.
22 #include "librpc/gen_ndr/messaging.h"
23 #include "ctdbd_conn.h"
24 #include "../lib/util/select.h"
26 static NTSTATUS
g_lock_force_unlock(struct g_lock_ctx
*ctx
, const char *name
,
27 struct server_id pid
);
30 struct db_context
*db
;
31 struct messaging_context
*msg
;
35 * The "g_lock.tdb" file contains records, indexed by the 0-terminated
36 * lockname. The record contains an array of "struct g_lock_rec"
37 * structures. Waiters have the lock_type with G_LOCK_PENDING or'ed.
41 enum g_lock_type lock_type
;
45 struct g_lock_ctx
*g_lock_ctx_init(TALLOC_CTX
*mem_ctx
,
46 struct messaging_context
*msg
)
48 struct g_lock_ctx
*result
;
50 result
= talloc(mem_ctx
, struct g_lock_ctx
);
56 result
->db
= db_open(result
, lock_path("g_lock.tdb"), 0,
57 TDB_CLEAR_IF_FIRST
|TDB_INCOMPATIBLE_HASH
, O_RDWR
|O_CREAT
, 0700);
58 if (result
->db
== NULL
) {
59 DEBUG(1, ("g_lock_init: Could not open g_lock.tdb"));
66 static bool g_lock_conflicts(enum g_lock_type lock_type
,
67 const struct g_lock_rec
*rec
)
69 enum g_lock_type rec_lock
= rec
->lock_type
;
71 if ((rec_lock
& G_LOCK_PENDING
) != 0) {
76 * Only tested write locks so far. Very likely this routine
77 * needs to be fixed for read locks....
79 if ((lock_type
== G_LOCK_READ
) && (rec_lock
== G_LOCK_READ
)) {
85 static bool g_lock_parse(TALLOC_CTX
*mem_ctx
, TDB_DATA data
,
86 int *pnum_locks
, struct g_lock_rec
**plocks
)
89 struct g_lock_rec
*locks
;
91 if ((data
.dsize
% sizeof(struct g_lock_rec
)) != 0) {
92 DEBUG(1, ("invalid lock record length %d\n", (int)data
.dsize
));
96 num_locks
= data
.dsize
/ sizeof(struct g_lock_rec
);
97 locks
= talloc_array(mem_ctx
, struct g_lock_rec
, num_locks
);
99 DEBUG(1, ("talloc failed\n"));
103 memcpy(locks
, data
.dptr
, data
.dsize
);
105 DEBUG(10, ("locks:\n"));
106 for (i
=0; i
<num_locks
; i
++) {
107 DEBUGADD(10, ("%s: %s %s\n",
108 procid_str(talloc_tos(), &locks
[i
].pid
),
109 ((locks
[i
].lock_type
& 1) == G_LOCK_READ
) ?
111 (locks
[i
].lock_type
& G_LOCK_PENDING
) ?
112 "(pending)" : "(owner)"));
114 if (((locks
[i
].lock_type
& G_LOCK_PENDING
) == 0)
115 && !process_exists(locks
[i
].pid
)) {
117 DEBUGADD(10, ("lock owner %s died -- discarding\n",
118 procid_str(talloc_tos(),
121 if (i
< (num_locks
-1)) {
122 locks
[i
] = locks
[num_locks
-1];
129 *pnum_locks
= num_locks
;
133 static void g_lock_cleanup(int *pnum_locks
, struct g_lock_rec
*locks
)
137 num_locks
= *pnum_locks
;
139 DEBUG(10, ("g_lock_cleanup: %d locks\n", num_locks
));
141 for (i
=0; i
<num_locks
; i
++) {
142 if (process_exists(locks
[i
].pid
)) {
145 DEBUGADD(10, ("%s does not exist -- discarding\n",
146 procid_str(talloc_tos(), &locks
[i
].pid
)));
148 if (i
< (num_locks
-1)) {
149 locks
[i
] = locks
[num_locks
-1];
153 *pnum_locks
= num_locks
;
157 static struct g_lock_rec
*g_lock_addrec(TALLOC_CTX
*mem_ctx
,
158 struct g_lock_rec
*locks
,
160 const struct server_id pid
,
161 enum g_lock_type lock_type
)
163 struct g_lock_rec
*result
;
164 int num_locks
= *pnum_locks
;
166 result
= talloc_realloc(mem_ctx
, locks
, struct g_lock_rec
,
168 if (result
== NULL
) {
172 result
[num_locks
].pid
= pid
;
173 result
[num_locks
].lock_type
= lock_type
;
178 static void g_lock_got_retry(struct messaging_context
*msg
,
181 struct server_id server_id
,
184 static NTSTATUS
g_lock_trylock(struct g_lock_ctx
*ctx
, const char *name
,
185 enum g_lock_type lock_type
)
187 struct db_record
*rec
= NULL
;
188 struct g_lock_rec
*locks
= NULL
;
190 struct server_id self
;
193 NTSTATUS status
= NT_STATUS_OK
;
194 NTSTATUS store_status
;
197 rec
= ctx
->db
->fetch_locked(ctx
->db
, talloc_tos(),
198 string_term_tdb_data(name
));
200 DEBUG(10, ("fetch_locked(\"%s\") failed\n", name
));
201 status
= NT_STATUS_LOCK_NOT_GRANTED
;
205 if (!g_lock_parse(talloc_tos(), rec
->value
, &num_locks
, &locks
)) {
206 DEBUG(10, ("g_lock_parse for %s failed\n", name
));
207 status
= NT_STATUS_INTERNAL_ERROR
;
211 self
= messaging_server_id(ctx
->msg
);
214 for (i
=0; i
<num_locks
; i
++) {
215 if (procid_equal(&self
, &locks
[i
].pid
)) {
216 if (our_index
!= -1) {
217 DEBUG(1, ("g_lock_trylock: Added ourself "
219 status
= NT_STATUS_INTERNAL_ERROR
;
222 if ((locks
[i
].lock_type
& G_LOCK_PENDING
) == 0) {
223 DEBUG(1, ("g_lock_trylock: Found ourself not "
225 status
= NT_STATUS_INTERNAL_ERROR
;
231 /* never conflict with ourself */
234 if (g_lock_conflicts(lock_type
, &locks
[i
])) {
235 struct server_id pid
= locks
[i
].pid
;
237 if (!process_exists(pid
)) {
240 status
= g_lock_force_unlock(ctx
, name
, pid
);
241 if (!NT_STATUS_IS_OK(status
)) {
242 DEBUG(1, ("Could not unlock dead lock "
248 lock_type
|= G_LOCK_PENDING
;
252 if (our_index
== -1) {
253 /* First round, add ourself */
255 locks
= g_lock_addrec(talloc_tos(), locks
, &num_locks
,
258 DEBUG(10, ("g_lock_addrec failed\n"));
259 status
= NT_STATUS_NO_MEMORY
;
264 * Retry. We were pending last time. Overwrite the
265 * stored lock_type with what we calculated, we might
266 * have acquired the lock this time.
268 locks
[our_index
].lock_type
= lock_type
;
271 if (NT_STATUS_IS_OK(status
) && ((lock_type
& G_LOCK_PENDING
) == 0)) {
273 * Walk through the list of locks, search for dead entries
275 g_lock_cleanup(&num_locks
, locks
);
278 data
= make_tdb_data((uint8_t *)locks
, num_locks
* sizeof(*locks
));
279 store_status
= rec
->store(rec
, data
, 0);
280 if (!NT_STATUS_IS_OK(store_status
)) {
281 DEBUG(1, ("rec->store failed: %s\n",
282 nt_errstr(store_status
)));
283 status
= store_status
;
290 if (NT_STATUS_IS_OK(status
) && (lock_type
& G_LOCK_PENDING
) != 0) {
291 return STATUS_PENDING
;
297 NTSTATUS
g_lock_lock(struct g_lock_ctx
*ctx
, const char *name
,
298 enum g_lock_type lock_type
, struct timeval timeout
)
300 struct tevent_timer
*te
= NULL
;
303 struct timeval timeout_end
;
304 struct timeval time_now
;
306 DEBUG(10, ("Trying to acquire lock %d for %s\n", (int)lock_type
,
309 if (lock_type
& ~1) {
310 DEBUG(1, ("Got invalid lock type %d for %s\n",
311 (int)lock_type
, name
));
312 return NT_STATUS_INVALID_PARAMETER
;
315 #ifdef CLUSTER_SUPPORT
316 if (lp_clustering()) {
317 status
= ctdb_watch_us(messaging_ctdbd_connection());
318 if (!NT_STATUS_IS_OK(status
)) {
319 DEBUG(10, ("could not register retry with ctdb: %s\n",
326 status
= messaging_register(ctx
->msg
, &retry
, MSG_DBWRAP_G_LOCK_RETRY
,
328 if (!NT_STATUS_IS_OK(status
)) {
329 DEBUG(10, ("messaging_register failed: %s\n",
334 time_now
= timeval_current();
335 timeout_end
= timeval_sum(&time_now
, &timeout
);
338 #ifdef CLUSTER_SUPPORT
341 fd_set
*r_fds
= NULL
;
344 struct timeval timeout_remaining
, select_timeout
;
346 status
= g_lock_trylock(ctx
, name
, lock_type
);
347 if (NT_STATUS_IS_OK(status
)) {
348 DEBUG(10, ("Got lock %s\n", name
));
351 if (!NT_STATUS_EQUAL(status
, STATUS_PENDING
)) {
352 DEBUG(10, ("g_lock_trylock failed: %s\n",
357 DEBUG(10, ("g_lock_trylock: Did not get lock, waiting...\n"));
359 /* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
360 * !!! HACK ALERT --- FIX ME !!!
361 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
362 * What we really want to do here is to react to
363 * MSG_DBWRAP_G_LOCK_RETRY messages that are either sent
364 * by a client doing g_lock_unlock or by ourselves when
365 * we receive a CTDB_SRVID_SAMBA_NOTIFY or
366 * CTDB_SRVID_RECONFIGURE message from ctdbd, i.e. when
367 * either a client holding a lock or a complete node
370 * Doing this properly involves calling tevent_loop_once(),
371 * but doing this here with the main ctdbd messaging context
372 * creates a nested event loop when g_lock_lock() is called
373 * from the main event loop, e.g. in a tcon_and_X where the
374 * share_info.tdb needs to be initialized and is locked by
375 * another process, or when the remore registry is accessed
376 * for writing and some other process already holds a lock
377 * on the registry.tdb.
379 * So as a quick fix, we act a little coarsely here: we do
380 * a select on the ctdb connection fd and when it is readable
381 * or we get EINTR, then we retry without actually parsing
382 * any ctdb packages or dispatching messages. This means that
383 * we retry more often than intended by design, but this does
384 * not harm and it is unobtrusive. When we have finished,
385 * the main loop will pick up all the messages and ctdb
386 * packets. The only extra twist is that we cannot use timed
387 * events here but have to handcode a timeout.
390 #ifdef CLUSTER_SUPPORT
391 if (lp_clustering()) {
392 struct ctdbd_connection
*conn
;
393 conn
= messaging_ctdbd_connection();
397 max_fd
= ctdbd_conn_get_fd(conn
);
398 FD_SET(max_fd
, r_fds
);
402 time_now
= timeval_current();
403 timeout_remaining
= timeval_until(&time_now
, &timeout_end
);
404 select_timeout
= timeval_set(60, 0);
406 select_timeout
= timeval_min(&select_timeout
,
409 ret
= sys_select(max_fd
+ 1, r_fds
, NULL
, NULL
,
412 if (errno
!= EINTR
) {
413 DEBUG(1, ("error calling select: %s\n",
415 status
= NT_STATUS_INTERNAL_ERROR
;
420 * This means a signal was received.
421 * It might have been a MSG_DBWRAP_G_LOCK_RETRY message.
424 } else if (ret
== 0) {
425 if (timeval_expired(&timeout_end
)) {
426 DEBUG(10, ("g_lock_lock timed out\n"));
427 status
= NT_STATUS_LOCK_NOT_GRANTED
;
430 DEBUG(10, ("select returned 0 but timeout not "
431 "not expired, retrying\n"));
433 } else if (ret
!= 1) {
434 DEBUG(1, ("invalid return code of select: %d\n", ret
));
435 status
= NT_STATUS_INTERNAL_ERROR
;
440 * This means ctdbd has sent us some data.
441 * Might be a CTDB_SRVID_RECONFIGURE or a
442 * CTDB_SRVID_SAMBA_NOTIFY message.
447 #ifdef CLUSTER_SUPPORT
451 if (!NT_STATUS_IS_OK(status
)) {
452 NTSTATUS unlock_status
;
454 unlock_status
= g_lock_unlock(ctx
, name
);
456 if (!NT_STATUS_IS_OK(unlock_status
)) {
457 DEBUG(1, ("Could not remove ourself from the locking "
458 "db: %s\n", nt_errstr(status
)));
462 messaging_deregister(ctx
->msg
, MSG_DBWRAP_G_LOCK_RETRY
, &retry
);
468 static void g_lock_got_retry(struct messaging_context
*msg
,
471 struct server_id server_id
,
474 bool *pretry
= (bool *)private_data
;
476 DEBUG(10, ("Got retry message from pid %s\n",
477 procid_str(talloc_tos(), &server_id
)));
482 static NTSTATUS
g_lock_force_unlock(struct g_lock_ctx
*ctx
, const char *name
,
483 struct server_id pid
)
485 struct db_record
*rec
= NULL
;
486 struct g_lock_rec
*locks
= NULL
;
488 enum g_lock_type lock_type
;
491 rec
= ctx
->db
->fetch_locked(ctx
->db
, talloc_tos(),
492 string_term_tdb_data(name
));
494 DEBUG(10, ("fetch_locked(\"%s\") failed\n", name
));
495 status
= NT_STATUS_INTERNAL_ERROR
;
499 if (!g_lock_parse(talloc_tos(), rec
->value
, &num_locks
, &locks
)) {
500 DEBUG(10, ("g_lock_parse for %s failed\n", name
));
501 status
= NT_STATUS_INTERNAL_ERROR
;
505 for (i
=0; i
<num_locks
; i
++) {
506 if (procid_equal(&pid
, &locks
[i
].pid
)) {
511 if (i
== num_locks
) {
512 DEBUG(10, ("g_lock_force_unlock: Lock not found\n"));
513 status
= NT_STATUS_INTERNAL_ERROR
;
517 lock_type
= locks
[i
].lock_type
;
519 if (i
< (num_locks
-1)) {
520 locks
[i
] = locks
[num_locks
-1];
524 if (num_locks
== 0) {
525 status
= rec
->delete_rec(rec
);
528 data
= make_tdb_data((uint8_t *)locks
,
529 sizeof(struct g_lock_rec
) * num_locks
);
530 status
= rec
->store(rec
, data
, 0);
533 if (!NT_STATUS_IS_OK(status
)) {
534 DEBUG(1, ("g_lock_force_unlock: Could not store record: %s\n",
541 if ((lock_type
& G_LOCK_PENDING
) == 0) {
545 * We've been the lock holder. Others to retry. Don't
546 * tell all others to avoid a thundering herd. In case
547 * this leads to a complete stall because we miss some
548 * processes, the loop in g_lock_lock tries at least
552 for (i
=0; i
<num_locks
; i
++) {
553 if ((locks
[i
].lock_type
& G_LOCK_PENDING
) == 0) {
556 if (!process_exists(locks
[i
].pid
)) {
561 * Ping all waiters to retry
563 status
= messaging_send(ctx
->msg
, locks
[i
].pid
,
564 MSG_DBWRAP_G_LOCK_RETRY
,
566 if (!NT_STATUS_IS_OK(status
)) {
567 DEBUG(1, ("sending retry to %s failed: %s\n",
568 procid_str(talloc_tos(),
574 if (num_wakeups
> 5) {
581 * For the error path, TALLOC_FREE(rec) as well. In the good
582 * path we have already freed it.
590 NTSTATUS
g_lock_unlock(struct g_lock_ctx
*ctx
, const char *name
)
594 status
= g_lock_force_unlock(ctx
, name
, messaging_server_id(ctx
->msg
));
596 #ifdef CLUSTER_SUPPORT
597 if (lp_clustering()) {
598 ctdb_unwatch(messaging_ctdbd_connection());
604 struct g_lock_locks_state
{
605 int (*fn
)(const char *name
, void *private_data
);
609 static int g_lock_locks_fn(struct db_record
*rec
, void *priv
)
611 struct g_lock_locks_state
*state
= (struct g_lock_locks_state
*)priv
;
613 if ((rec
->key
.dsize
== 0) || (rec
->key
.dptr
[rec
->key
.dsize
-1] != 0)) {
614 DEBUG(1, ("invalid key in g_lock.tdb, ignoring\n"));
617 return state
->fn((char *)rec
->key
.dptr
, state
->private_data
);
620 int g_lock_locks(struct g_lock_ctx
*ctx
,
621 int (*fn
)(const char *name
, void *private_data
),
624 struct g_lock_locks_state state
;
627 state
.private_data
= private_data
;
629 return ctx
->db
->traverse_read(ctx
->db
, g_lock_locks_fn
, &state
);
632 NTSTATUS
g_lock_dump(struct g_lock_ctx
*ctx
, const char *name
,
633 int (*fn
)(struct server_id pid
,
634 enum g_lock_type lock_type
,
640 struct g_lock_rec
*locks
= NULL
;
643 if (ctx
->db
->fetch(ctx
->db
, talloc_tos(), string_term_tdb_data(name
),
645 return NT_STATUS_NOT_FOUND
;
648 if ((data
.dsize
== 0) || (data
.dptr
== NULL
)) {
652 ret
= g_lock_parse(talloc_tos(), data
, &num_locks
, &locks
);
654 TALLOC_FREE(data
.dptr
);
657 DEBUG(10, ("g_lock_parse for %s failed\n", name
));
658 return NT_STATUS_INTERNAL_ERROR
;
661 for (i
=0; i
<num_locks
; i
++) {
662 if (fn(locks
[i
].pid
, locks
[i
].lock_type
, private_data
) != 0) {
670 struct g_lock_get_state
{
672 struct server_id
*pid
;
675 static int g_lock_get_fn(struct server_id pid
, enum g_lock_type lock_type
,
678 struct g_lock_get_state
*state
= (struct g_lock_get_state
*)priv
;
680 if ((lock_type
& G_LOCK_PENDING
) != 0) {
689 NTSTATUS
g_lock_get(struct g_lock_ctx
*ctx
, const char *name
,
690 struct server_id
*pid
)
692 struct g_lock_get_state state
;
698 status
= g_lock_dump(ctx
, name
, g_lock_get_fn
, &state
);
699 if (!NT_STATUS_IS_OK(status
)) {
703 return NT_STATUS_NOT_FOUND
;
708 static bool g_lock_init_all(TALLOC_CTX
*mem_ctx
,
709 struct tevent_context
**pev
,
710 struct messaging_context
**pmsg
,
711 const struct server_id self
,
712 struct g_lock_ctx
**pg_ctx
)
714 struct tevent_context
*ev
= NULL
;
715 struct messaging_context
*msg
= NULL
;
716 struct g_lock_ctx
*g_ctx
= NULL
;
718 ev
= tevent_context_init(mem_ctx
);
720 d_fprintf(stderr
, "ERROR: could not init event context\n");
723 msg
= messaging_init(mem_ctx
, self
, ev
);
725 d_fprintf(stderr
, "ERROR: could not init messaging context\n");
728 g_ctx
= g_lock_ctx_init(mem_ctx
, msg
);
730 d_fprintf(stderr
, "ERROR: could not init g_lock context\n");
745 NTSTATUS
g_lock_do(const char *name
, enum g_lock_type lock_type
,
746 struct timeval timeout
, const struct server_id self
,
747 void (*fn
)(void *private_data
), void *private_data
)
749 struct tevent_context
*ev
= NULL
;
750 struct messaging_context
*msg
= NULL
;
751 struct g_lock_ctx
*g_ctx
= NULL
;
754 if (!g_lock_init_all(talloc_tos(), &ev
, &msg
, self
, &g_ctx
)) {
755 status
= NT_STATUS_ACCESS_DENIED
;
759 status
= g_lock_lock(g_ctx
, name
, lock_type
, timeout
);
760 if (!NT_STATUS_IS_OK(status
)) {
764 g_lock_unlock(g_ctx
, name
);