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"
25 static NTSTATUS g_lock_force_unlock(struct g_lock_ctx *ctx, const char *name,
26 struct server_id pid);
29 struct db_context *db;
30 struct messaging_context *msg;
34 * The "g_lock.tdb" file contains records, indexed by the 0-terminated
35 * lockname. The record contains an array of "struct g_lock_rec"
36 * structures. Waiters have the lock_type with G_LOCK_PENDING or'ed.
40 enum g_lock_type lock_type;
44 struct g_lock_ctx *g_lock_ctx_init(TALLOC_CTX *mem_ctx,
45 struct messaging_context *msg)
47 struct g_lock_ctx *result;
49 result = talloc(mem_ctx, struct g_lock_ctx);
55 result->db = db_open(result, lock_path("g_lock.tdb"), 0,
56 TDB_CLEAR_IF_FIRST|TDB_INCOMPATIBLE_HASH, O_RDWR|O_CREAT, 0700);
57 if (result->db == NULL) {
58 DEBUG(1, ("g_lock_init: Could not open g_lock.tdb"));
65 static bool g_lock_conflicts(enum g_lock_type lock_type,
66 const struct g_lock_rec *rec)
68 enum g_lock_type rec_lock = rec->lock_type;
70 if ((rec_lock & G_LOCK_PENDING) != 0) {
75 * Only tested write locks so far. Very likely this routine
76 * needs to be fixed for read locks....
78 if ((lock_type == G_LOCK_READ) && (rec_lock == G_LOCK_READ)) {
84 static bool g_lock_parse(TALLOC_CTX *mem_ctx, TDB_DATA data,
85 int *pnum_locks, struct g_lock_rec **plocks)
88 struct g_lock_rec *locks;
90 if ((data.dsize % sizeof(struct g_lock_rec)) != 0) {
91 DEBUG(1, ("invalid lock record length %d\n", (int)data.dsize));
95 num_locks = data.dsize / sizeof(struct g_lock_rec);
96 locks = talloc_array(mem_ctx, struct g_lock_rec, num_locks);
98 DEBUG(1, ("talloc failed\n"));
102 memcpy(locks, data.dptr, data.dsize);
104 DEBUG(10, ("locks:\n"));
105 for (i=0; i<num_locks; i++) {
106 DEBUGADD(10, ("%s: %s %s\n",
107 procid_str(talloc_tos(), &locks[i].pid),
108 ((locks[i].lock_type & 1) == G_LOCK_READ) ?
110 (locks[i].lock_type & G_LOCK_PENDING) ?
111 "(pending)" : "(owner)"));
113 if (((locks[i].lock_type & G_LOCK_PENDING) == 0)
114 && !process_exists(locks[i].pid)) {
116 DEBUGADD(10, ("lock owner %s died -- discarding\n",
117 procid_str(talloc_tos(),
120 if (i < (num_locks-1)) {
121 locks[i] = locks[num_locks-1];
128 *pnum_locks = num_locks;
132 static void g_lock_cleanup(int *pnum_locks, struct g_lock_rec *locks)
136 num_locks = *pnum_locks;
138 DEBUG(10, ("g_lock_cleanup: %d locks\n", num_locks));
140 for (i=0; i<num_locks; i++) {
141 if (process_exists(locks[i].pid)) {
144 DEBUGADD(10, ("%s does not exist -- discarding\n",
145 procid_str(talloc_tos(), &locks[i].pid)));
147 if (i < (num_locks-1)) {
148 locks[i] = locks[num_locks-1];
152 *pnum_locks = num_locks;
156 static struct g_lock_rec *g_lock_addrec(TALLOC_CTX *mem_ctx,
157 struct g_lock_rec *locks,
159 const struct server_id pid,
160 enum g_lock_type lock_type)
162 struct g_lock_rec *result;
163 int num_locks = *pnum_locks;
165 result = talloc_realloc(mem_ctx, locks, struct g_lock_rec,
167 if (result == NULL) {
171 result[num_locks].pid = pid;
172 result[num_locks].lock_type = lock_type;
177 static void g_lock_got_retry(struct messaging_context *msg,
180 struct server_id server_id,
183 static NTSTATUS g_lock_trylock(struct g_lock_ctx *ctx, const char *name,
184 enum g_lock_type lock_type)
186 struct db_record *rec = NULL;
187 struct g_lock_rec *locks = NULL;
189 struct server_id self;
192 NTSTATUS status = NT_STATUS_OK;
193 NTSTATUS store_status;
196 rec = ctx->db->fetch_locked(ctx->db, talloc_tos(),
197 string_term_tdb_data(name));
199 DEBUG(10, ("fetch_locked(\"%s\") failed\n", name));
200 status = NT_STATUS_LOCK_NOT_GRANTED;
204 if (!g_lock_parse(talloc_tos(), rec->value, &num_locks, &locks)) {
205 DEBUG(10, ("g_lock_parse for %s failed\n", name));
206 status = NT_STATUS_INTERNAL_ERROR;
210 self = messaging_server_id(ctx->msg);
213 for (i=0; i<num_locks; i++) {
214 if (procid_equal(&self, &locks[i].pid)) {
215 if (our_index != -1) {
216 DEBUG(1, ("g_lock_trylock: Added ourself "
218 status = NT_STATUS_INTERNAL_ERROR;
221 if ((locks[i].lock_type & G_LOCK_PENDING) == 0) {
222 DEBUG(1, ("g_lock_trylock: Found ourself not "
224 status = NT_STATUS_INTERNAL_ERROR;
230 /* never conflict with ourself */
233 if (g_lock_conflicts(lock_type, &locks[i])) {
234 struct server_id pid = locks[i].pid;
236 if (!process_exists(pid)) {
239 status = g_lock_force_unlock(ctx, name, pid);
240 if (!NT_STATUS_IS_OK(status)) {
241 DEBUG(1, ("Could not unlock dead lock "
247 lock_type |= G_LOCK_PENDING;
251 if (our_index == -1) {
252 /* First round, add ourself */
254 locks = g_lock_addrec(talloc_tos(), locks, &num_locks,
257 DEBUG(10, ("g_lock_addrec failed\n"));
258 status = NT_STATUS_NO_MEMORY;
263 * Retry. We were pending last time. Overwrite the
264 * stored lock_type with what we calculated, we might
265 * have acquired the lock this time.
267 locks[our_index].lock_type = lock_type;
270 if (NT_STATUS_IS_OK(status) && ((lock_type & G_LOCK_PENDING) == 0)) {
272 * Walk through the list of locks, search for dead entries
274 g_lock_cleanup(&num_locks, locks);
277 data = make_tdb_data((uint8_t *)locks, num_locks * sizeof(*locks));
278 store_status = rec->store(rec, data, 0);
279 if (!NT_STATUS_IS_OK(store_status)) {
280 DEBUG(1, ("rec->store failed: %s\n",
281 nt_errstr(store_status)));
282 status = store_status;
289 if (NT_STATUS_IS_OK(status) && (lock_type & G_LOCK_PENDING) != 0) {
290 return STATUS_PENDING;
296 NTSTATUS g_lock_lock(struct g_lock_ctx *ctx, const char *name,
297 enum g_lock_type lock_type, struct timeval timeout)
299 struct tevent_timer *te = NULL;
302 struct timeval timeout_end;
303 struct timeval time_now;
305 DEBUG(10, ("Trying to acquire lock %d for %s\n", (int)lock_type,
308 if (lock_type & ~1) {
309 DEBUG(1, ("Got invalid lock type %d for %s\n",
310 (int)lock_type, name));
311 return NT_STATUS_INVALID_PARAMETER;
314 #ifdef CLUSTER_SUPPORT
315 if (lp_clustering()) {
316 status = ctdb_watch_us(messaging_ctdbd_connection());
317 if (!NT_STATUS_IS_OK(status)) {
318 DEBUG(10, ("could not register retry with ctdb: %s\n",
325 status = messaging_register(ctx->msg, &retry, MSG_DBWRAP_G_LOCK_RETRY,
327 if (!NT_STATUS_IS_OK(status)) {
328 DEBUG(10, ("messaging_register failed: %s\n",
333 time_now = timeval_current();
334 timeout_end = timeval_sum(&time_now, &timeout);
337 #ifdef CLUSTER_SUPPORT
340 fd_set *r_fds = NULL;
343 struct timeval timeout_remaining, select_timeout;
345 status = g_lock_trylock(ctx, name, lock_type);
346 if (NT_STATUS_IS_OK(status)) {
347 DEBUG(10, ("Got lock %s\n", name));
350 if (!NT_STATUS_EQUAL(status, STATUS_PENDING)) {
351 DEBUG(10, ("g_lock_trylock failed: %s\n",
356 DEBUG(10, ("g_lock_trylock: Did not get lock, waiting...\n"));
358 /* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
359 * !!! HACK ALERT --- FIX ME !!!
360 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
361 * What we really want to do here is to react to
362 * MSG_DBWRAP_G_LOCK_RETRY messages that are either sent
363 * by a client doing g_lock_unlock or by ourselves when
364 * we receive a CTDB_SRVID_SAMBA_NOTIFY or
365 * CTDB_SRVID_RECONFIGURE message from ctdbd, i.e. when
366 * either a client holding a lock or a complete node
369 * Doing this properly involves calling tevent_loop_once(),
370 * but doing this here with the main ctdbd messaging context
371 * creates a nested event loop when g_lock_lock() is called
372 * from the main event loop, e.g. in a tcon_and_X where the
373 * share_info.tdb needs to be initialized and is locked by
374 * another process, or when the remore registry is accessed
375 * for writing and some other process already holds a lock
376 * on the registry.tdb.
378 * So as a quick fix, we act a little coarsely here: we do
379 * a select on the ctdb connection fd and when it is readable
380 * or we get EINTR, then we retry without actually parsing
381 * any ctdb packages or dispatching messages. This means that
382 * we retry more often than intended by design, but this does
383 * not harm and it is unobtrusive. When we have finished,
384 * the main loop will pick up all the messages and ctdb
385 * packets. The only extra twist is that we cannot use timed
386 * events here but have to handcode a timeout.
389 #ifdef CLUSTER_SUPPORT
390 if (lp_clustering()) {
391 struct ctdbd_connection *conn;
392 conn = messaging_ctdbd_connection();
396 max_fd = ctdbd_conn_get_fd(conn);
397 FD_SET(max_fd, r_fds);
401 time_now = timeval_current();
402 timeout_remaining = timeval_until(&time_now, &timeout_end);
403 select_timeout = timeval_set(60, 0);
405 select_timeout = timeval_min(&select_timeout,
408 ret = sys_select(max_fd + 1, r_fds, NULL, NULL,
411 if (errno != EINTR) {
412 DEBUG(1, ("error calling select: %s\n",
414 status = NT_STATUS_INTERNAL_ERROR;
419 * This means a signal was received.
420 * It might have been a MSG_DBWRAP_G_LOCK_RETRY message.
423 } else if (ret == 0) {
424 if (timeval_expired(&timeout_end)) {
425 DEBUG(10, ("g_lock_lock timed out\n"));
426 status = NT_STATUS_LOCK_NOT_GRANTED;
429 DEBUG(10, ("select returned 0 but timeout not "
430 "not expired, retrying\n"));
432 } else if (ret != 1) {
433 DEBUG(1, ("invalid return code of select: %d\n", ret));
434 status = NT_STATUS_INTERNAL_ERROR;
439 * This means ctdbd has sent us some data.
440 * Might be a CTDB_SRVID_RECONFIGURE or a
441 * CTDB_SRVID_SAMBA_NOTIFY message.
446 #ifdef CLUSTER_SUPPORT
450 if (!NT_STATUS_IS_OK(status)) {
451 NTSTATUS unlock_status;
453 unlock_status = g_lock_unlock(ctx, name);
455 if (!NT_STATUS_IS_OK(unlock_status)) {
456 DEBUG(1, ("Could not remove ourself from the locking "
457 "db: %s\n", nt_errstr(status)));
461 messaging_deregister(ctx->msg, MSG_DBWRAP_G_LOCK_RETRY, &retry);
467 static void g_lock_got_retry(struct messaging_context *msg,
470 struct server_id server_id,
473 bool *pretry = (bool *)private_data;
475 DEBUG(10, ("Got retry message from pid %s\n",
476 procid_str(talloc_tos(), &server_id)));
481 static NTSTATUS g_lock_force_unlock(struct g_lock_ctx *ctx, const char *name,
482 struct server_id pid)
484 struct db_record *rec = NULL;
485 struct g_lock_rec *locks = NULL;
487 enum g_lock_type lock_type;
490 rec = ctx->db->fetch_locked(ctx->db, talloc_tos(),
491 string_term_tdb_data(name));
493 DEBUG(10, ("fetch_locked(\"%s\") failed\n", name));
494 status = NT_STATUS_INTERNAL_ERROR;
498 if (!g_lock_parse(talloc_tos(), rec->value, &num_locks, &locks)) {
499 DEBUG(10, ("g_lock_parse for %s failed\n", name));
500 status = NT_STATUS_INTERNAL_ERROR;
504 for (i=0; i<num_locks; i++) {
505 if (procid_equal(&pid, &locks[i].pid)) {
510 if (i == num_locks) {
511 DEBUG(10, ("g_lock_force_unlock: Lock not found\n"));
512 status = NT_STATUS_INTERNAL_ERROR;
516 lock_type = locks[i].lock_type;
518 if (i < (num_locks-1)) {
519 locks[i] = locks[num_locks-1];
523 if (num_locks == 0) {
524 status = rec->delete_rec(rec);
527 data = make_tdb_data((uint8_t *)locks,
528 sizeof(struct g_lock_rec) * num_locks);
529 status = rec->store(rec, data, 0);
532 if (!NT_STATUS_IS_OK(status)) {
533 DEBUG(1, ("g_lock_force_unlock: Could not store record: %s\n",
540 if ((lock_type & G_LOCK_PENDING) == 0) {
544 * We've been the lock holder. Others to retry. Don't
545 * tell all others to avoid a thundering herd. In case
546 * this leads to a complete stall because we miss some
547 * processes, the loop in g_lock_lock tries at least
551 for (i=0; i<num_locks; i++) {
552 if ((locks[i].lock_type & G_LOCK_PENDING) == 0) {
555 if (!process_exists(locks[i].pid)) {
560 * Ping all waiters to retry
562 status = messaging_send(ctx->msg, locks[i].pid,
563 MSG_DBWRAP_G_LOCK_RETRY,
565 if (!NT_STATUS_IS_OK(status)) {
566 DEBUG(1, ("sending retry to %s failed: %s\n",
567 procid_str(talloc_tos(),
573 if (num_wakeups > 5) {
580 * For the error path, TALLOC_FREE(rec) as well. In the good
581 * path we have already freed it.
589 NTSTATUS g_lock_unlock(struct g_lock_ctx *ctx, const char *name)
593 status = g_lock_force_unlock(ctx, name, messaging_server_id(ctx->msg));
595 #ifdef CLUSTER_SUPPORT
596 if (lp_clustering()) {
597 ctdb_unwatch(messaging_ctdbd_connection());
603 struct g_lock_locks_state {
604 int (*fn)(const char *name, void *private_data);
608 static int g_lock_locks_fn(struct db_record *rec, void *priv)
610 struct g_lock_locks_state *state = (struct g_lock_locks_state *)priv;
612 if ((rec->key.dsize == 0) || (rec->key.dptr[rec->key.dsize-1] != 0)) {
613 DEBUG(1, ("invalid key in g_lock.tdb, ignoring\n"));
616 return state->fn((char *)rec->key.dptr, state->private_data);
619 int g_lock_locks(struct g_lock_ctx *ctx,
620 int (*fn)(const char *name, void *private_data),
623 struct g_lock_locks_state state;
626 state.private_data = private_data;
628 return ctx->db->traverse_read(ctx->db, g_lock_locks_fn, &state);
631 NTSTATUS g_lock_dump(struct g_lock_ctx *ctx, const char *name,
632 int (*fn)(struct server_id pid,
633 enum g_lock_type lock_type,
639 struct g_lock_rec *locks = NULL;
642 if (ctx->db->fetch(ctx->db, talloc_tos(), string_term_tdb_data(name),
644 return NT_STATUS_NOT_FOUND;
647 if ((data.dsize == 0) || (data.dptr == NULL)) {
651 ret = g_lock_parse(talloc_tos(), data, &num_locks, &locks);
653 TALLOC_FREE(data.dptr);
656 DEBUG(10, ("g_lock_parse for %s failed\n", name));
657 return NT_STATUS_INTERNAL_ERROR;
660 for (i=0; i<num_locks; i++) {
661 if (fn(locks[i].pid, locks[i].lock_type, private_data) != 0) {
669 struct g_lock_get_state {
671 struct server_id *pid;
674 static int g_lock_get_fn(struct server_id pid, enum g_lock_type lock_type,
677 struct g_lock_get_state *state = (struct g_lock_get_state *)priv;
679 if ((lock_type & G_LOCK_PENDING) != 0) {
688 NTSTATUS g_lock_get(struct g_lock_ctx *ctx, const char *name,
689 struct server_id *pid)
691 struct g_lock_get_state state;
697 status = g_lock_dump(ctx, name, g_lock_get_fn, &state);
698 if (!NT_STATUS_IS_OK(status)) {
702 return NT_STATUS_NOT_FOUND;
707 static bool g_lock_init_all(TALLOC_CTX *mem_ctx,
708 struct tevent_context **pev,
709 struct messaging_context **pmsg,
710 const struct server_id self,
711 struct g_lock_ctx **pg_ctx)
713 struct tevent_context *ev = NULL;
714 struct messaging_context *msg = NULL;
715 struct g_lock_ctx *g_ctx = NULL;
717 ev = tevent_context_init(mem_ctx);
719 d_fprintf(stderr, "ERROR: could not init event context\n");
722 msg = messaging_init(mem_ctx, self, ev);
724 d_fprintf(stderr, "ERROR: could not init messaging context\n");
727 g_ctx = g_lock_ctx_init(mem_ctx, msg);
729 d_fprintf(stderr, "ERROR: could not init g_lock context\n");
744 NTSTATUS g_lock_do(const char *name, enum g_lock_type lock_type,
745 struct timeval timeout, const struct server_id self,
746 void (*fn)(void *private_data), void *private_data)
748 struct tevent_context *ev = NULL;
749 struct messaging_context *msg = NULL;
750 struct g_lock_ctx *g_ctx = NULL;
753 if (!g_lock_init_all(talloc_tos(), &ev, &msg, self, &g_ctx)) {
754 status = NT_STATUS_ACCESS_DENIED;
758 status = g_lock_lock(g_ctx, name, lock_type, timeout);
759 if (!NT_STATUS_IS_OK(status)) {
763 g_lock_unlock(g_ctx, name);