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"
25 #include "system/select.h"
27 static NTSTATUS g_lock_force_unlock(struct g_lock_ctx *ctx, const char *name,
28 struct server_id pid);
31 struct db_context *db;
32 struct messaging_context *msg;
36 * The "g_lock.tdb" file contains records, indexed by the 0-terminated
37 * lockname. The record contains an array of "struct g_lock_rec"
38 * structures. Waiters have the lock_type with G_LOCK_PENDING or'ed.
42 enum g_lock_type lock_type;
46 struct g_lock_ctx *g_lock_ctx_init(TALLOC_CTX *mem_ctx,
47 struct messaging_context *msg)
49 struct g_lock_ctx *result;
51 result = talloc(mem_ctx, struct g_lock_ctx);
57 result->db = db_open(result, lock_path("g_lock.tdb"), 0,
58 TDB_CLEAR_IF_FIRST|TDB_INCOMPATIBLE_HASH, O_RDWR|O_CREAT, 0700);
59 if (result->db == NULL) {
60 DEBUG(1, ("g_lock_init: Could not open g_lock.tdb"));
67 static bool g_lock_conflicts(enum g_lock_type lock_type,
68 const struct g_lock_rec *rec)
70 enum g_lock_type rec_lock = rec->lock_type;
72 if ((rec_lock & G_LOCK_PENDING) != 0) {
77 * Only tested write locks so far. Very likely this routine
78 * needs to be fixed for read locks....
80 if ((lock_type == G_LOCK_READ) && (rec_lock == G_LOCK_READ)) {
86 static bool g_lock_parse(TALLOC_CTX *mem_ctx, TDB_DATA data,
87 int *pnum_locks, struct g_lock_rec **plocks)
90 struct g_lock_rec *locks;
92 if ((data.dsize % sizeof(struct g_lock_rec)) != 0) {
93 DEBUG(1, ("invalid lock record length %d\n", (int)data.dsize));
97 num_locks = data.dsize / sizeof(struct g_lock_rec);
98 locks = talloc_array(mem_ctx, struct g_lock_rec, num_locks);
100 DEBUG(1, ("talloc failed\n"));
104 memcpy(locks, data.dptr, data.dsize);
106 DEBUG(10, ("locks:\n"));
107 for (i=0; i<num_locks; i++) {
108 DEBUGADD(10, ("%s: %s %s\n",
109 procid_str(talloc_tos(), &locks[i].pid),
110 ((locks[i].lock_type & 1) == G_LOCK_READ) ?
112 (locks[i].lock_type & G_LOCK_PENDING) ?
113 "(pending)" : "(owner)"));
115 if (((locks[i].lock_type & G_LOCK_PENDING) == 0)
116 && !process_exists(locks[i].pid)) {
118 DEBUGADD(10, ("lock owner %s died -- discarding\n",
119 procid_str(talloc_tos(),
122 if (i < (num_locks-1)) {
123 locks[i] = locks[num_locks-1];
130 *pnum_locks = num_locks;
134 static void g_lock_cleanup(int *pnum_locks, struct g_lock_rec *locks)
138 num_locks = *pnum_locks;
140 DEBUG(10, ("g_lock_cleanup: %d locks\n", num_locks));
142 for (i=0; i<num_locks; i++) {
143 if (process_exists(locks[i].pid)) {
146 DEBUGADD(10, ("%s does not exist -- discarding\n",
147 procid_str(talloc_tos(), &locks[i].pid)));
149 if (i < (num_locks-1)) {
150 locks[i] = locks[num_locks-1];
154 *pnum_locks = num_locks;
158 static struct g_lock_rec *g_lock_addrec(TALLOC_CTX *mem_ctx,
159 struct g_lock_rec *locks,
161 const struct server_id pid,
162 enum g_lock_type lock_type)
164 struct g_lock_rec *result;
165 int num_locks = *pnum_locks;
167 result = talloc_realloc(mem_ctx, locks, struct g_lock_rec,
169 if (result == NULL) {
173 result[num_locks].pid = pid;
174 result[num_locks].lock_type = lock_type;
179 static void g_lock_got_retry(struct messaging_context *msg,
182 struct server_id server_id,
185 static NTSTATUS g_lock_trylock(struct g_lock_ctx *ctx, const char *name,
186 enum g_lock_type lock_type)
188 struct db_record *rec = NULL;
189 struct g_lock_rec *locks = NULL;
191 struct server_id self;
194 NTSTATUS status = NT_STATUS_OK;
195 NTSTATUS store_status;
198 rec = ctx->db->fetch_locked(ctx->db, talloc_tos(),
199 string_term_tdb_data(name));
201 DEBUG(10, ("fetch_locked(\"%s\") failed\n", name));
202 status = NT_STATUS_LOCK_NOT_GRANTED;
206 if (!g_lock_parse(talloc_tos(), rec->value, &num_locks, &locks)) {
207 DEBUG(10, ("g_lock_parse for %s failed\n", name));
208 status = NT_STATUS_INTERNAL_ERROR;
212 self = messaging_server_id(ctx->msg);
215 for (i=0; i<num_locks; i++) {
216 if (procid_equal(&self, &locks[i].pid)) {
217 if (our_index != -1) {
218 DEBUG(1, ("g_lock_trylock: Added ourself "
220 status = NT_STATUS_INTERNAL_ERROR;
223 if ((locks[i].lock_type & G_LOCK_PENDING) == 0) {
224 DEBUG(1, ("g_lock_trylock: Found ourself not "
226 status = NT_STATUS_INTERNAL_ERROR;
232 /* never conflict with ourself */
235 if (g_lock_conflicts(lock_type, &locks[i])) {
236 struct server_id pid = locks[i].pid;
238 if (!process_exists(pid)) {
241 status = g_lock_force_unlock(ctx, name, pid);
242 if (!NT_STATUS_IS_OK(status)) {
243 DEBUG(1, ("Could not unlock dead lock "
249 lock_type |= G_LOCK_PENDING;
253 if (our_index == -1) {
254 /* First round, add ourself */
256 locks = g_lock_addrec(talloc_tos(), locks, &num_locks,
259 DEBUG(10, ("g_lock_addrec failed\n"));
260 status = NT_STATUS_NO_MEMORY;
265 * Retry. We were pending last time. Overwrite the
266 * stored lock_type with what we calculated, we might
267 * have acquired the lock this time.
269 locks[our_index].lock_type = lock_type;
272 if (NT_STATUS_IS_OK(status) && ((lock_type & G_LOCK_PENDING) == 0)) {
274 * Walk through the list of locks, search for dead entries
276 g_lock_cleanup(&num_locks, locks);
279 data = make_tdb_data((uint8_t *)locks, num_locks * sizeof(*locks));
280 store_status = rec->store(rec, data, 0);
281 if (!NT_STATUS_IS_OK(store_status)) {
282 DEBUG(1, ("rec->store failed: %s\n",
283 nt_errstr(store_status)));
284 status = store_status;
291 if (NT_STATUS_IS_OK(status) && (lock_type & G_LOCK_PENDING) != 0) {
292 return STATUS_PENDING;
298 NTSTATUS g_lock_lock(struct g_lock_ctx *ctx, const char *name,
299 enum g_lock_type lock_type, struct timeval timeout)
301 struct tevent_timer *te = NULL;
304 struct timeval timeout_end;
305 struct timeval time_now;
307 DEBUG(10, ("Trying to acquire lock %d for %s\n", (int)lock_type,
310 if (lock_type & ~1) {
311 DEBUG(1, ("Got invalid lock type %d for %s\n",
312 (int)lock_type, name));
313 return NT_STATUS_INVALID_PARAMETER;
316 #ifdef CLUSTER_SUPPORT
317 if (lp_clustering()) {
318 status = ctdb_watch_us(messaging_ctdbd_connection());
319 if (!NT_STATUS_IS_OK(status)) {
320 DEBUG(10, ("could not register retry with ctdb: %s\n",
327 status = messaging_register(ctx->msg, &retry, MSG_DBWRAP_G_LOCK_RETRY,
329 if (!NT_STATUS_IS_OK(status)) {
330 DEBUG(10, ("messaging_register failed: %s\n",
335 time_now = timeval_current();
336 timeout_end = timeval_sum(&time_now, &timeout);
339 struct pollfd *pollfds;
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.
390 * We allocate 2 entries here. One is needed anyway for
391 * sys_poll and in the clustering case we might have to add
392 * the ctdb fd. This avoids the realloc then.
394 pollfds = TALLOC_ARRAY(talloc_tos(), struct pollfd, 2);
395 if (pollfds == NULL) {
396 status = NT_STATUS_NO_MEMORY;
401 #ifdef CLUSTER_SUPPORT
402 if (lp_clustering()) {
403 struct ctdbd_connection *conn;
404 conn = messaging_ctdbd_connection();
406 pollfds[0].fd = ctdbd_conn_get_fd(conn);
407 pollfds[0].events = POLLIN|POLLHUP;
413 time_now = timeval_current();
414 timeout_remaining = timeval_until(&time_now, &timeout_end);
415 select_timeout = timeval_set(60, 0);
417 select_timeout = timeval_min(&select_timeout,
420 ret = sys_poll(pollfds, num_pollfds,
421 timeval_to_msec(select_timeout));
424 * We're not *really interested in the actual flags. We just
425 * need to retry this whole thing.
428 TALLOC_FREE(pollfds);
432 if (errno != EINTR) {
433 DEBUG(1, ("error calling select: %s\n",
435 status = NT_STATUS_INTERNAL_ERROR;
440 * This means a signal was received.
441 * It might have been a MSG_DBWRAP_G_LOCK_RETRY message.
444 } else if (ret == 0) {
445 if (timeval_expired(&timeout_end)) {
446 DEBUG(10, ("g_lock_lock timed out\n"));
447 status = NT_STATUS_LOCK_NOT_GRANTED;
450 DEBUG(10, ("select returned 0 but timeout not "
451 "not expired, retrying\n"));
453 } else if (ret != 1) {
454 DEBUG(1, ("invalid return code of select: %d\n", ret));
455 status = NT_STATUS_INTERNAL_ERROR;
460 * This means ctdbd has sent us some data.
461 * Might be a CTDB_SRVID_RECONFIGURE or a
462 * CTDB_SRVID_SAMBA_NOTIFY message.
467 #ifdef CLUSTER_SUPPORT
471 if (!NT_STATUS_IS_OK(status)) {
472 NTSTATUS unlock_status;
474 unlock_status = g_lock_unlock(ctx, name);
476 if (!NT_STATUS_IS_OK(unlock_status)) {
477 DEBUG(1, ("Could not remove ourself from the locking "
478 "db: %s\n", nt_errstr(status)));
482 messaging_deregister(ctx->msg, MSG_DBWRAP_G_LOCK_RETRY, &retry);
488 static void g_lock_got_retry(struct messaging_context *msg,
491 struct server_id server_id,
494 bool *pretry = (bool *)private_data;
496 DEBUG(10, ("Got retry message from pid %s\n",
497 procid_str(talloc_tos(), &server_id)));
502 static NTSTATUS g_lock_force_unlock(struct g_lock_ctx *ctx, const char *name,
503 struct server_id pid)
505 struct db_record *rec = NULL;
506 struct g_lock_rec *locks = NULL;
508 enum g_lock_type lock_type;
511 rec = ctx->db->fetch_locked(ctx->db, talloc_tos(),
512 string_term_tdb_data(name));
514 DEBUG(10, ("fetch_locked(\"%s\") failed\n", name));
515 status = NT_STATUS_INTERNAL_ERROR;
519 if (!g_lock_parse(talloc_tos(), rec->value, &num_locks, &locks)) {
520 DEBUG(10, ("g_lock_parse for %s failed\n", name));
521 status = NT_STATUS_INTERNAL_ERROR;
525 for (i=0; i<num_locks; i++) {
526 if (procid_equal(&pid, &locks[i].pid)) {
531 if (i == num_locks) {
532 DEBUG(10, ("g_lock_force_unlock: Lock not found\n"));
533 status = NT_STATUS_INTERNAL_ERROR;
537 lock_type = locks[i].lock_type;
539 if (i < (num_locks-1)) {
540 locks[i] = locks[num_locks-1];
544 if (num_locks == 0) {
545 status = rec->delete_rec(rec);
548 data = make_tdb_data((uint8_t *)locks,
549 sizeof(struct g_lock_rec) * num_locks);
550 status = rec->store(rec, data, 0);
553 if (!NT_STATUS_IS_OK(status)) {
554 DEBUG(1, ("g_lock_force_unlock: Could not store record: %s\n",
561 if ((lock_type & G_LOCK_PENDING) == 0) {
565 * We've been the lock holder. Others to retry. Don't
566 * tell all others to avoid a thundering herd. In case
567 * this leads to a complete stall because we miss some
568 * processes, the loop in g_lock_lock tries at least
572 for (i=0; i<num_locks; i++) {
573 if ((locks[i].lock_type & G_LOCK_PENDING) == 0) {
576 if (!process_exists(locks[i].pid)) {
581 * Ping all waiters to retry
583 status = messaging_send(ctx->msg, locks[i].pid,
584 MSG_DBWRAP_G_LOCK_RETRY,
586 if (!NT_STATUS_IS_OK(status)) {
587 DEBUG(1, ("sending retry to %s failed: %s\n",
588 procid_str(talloc_tos(),
594 if (num_wakeups > 5) {
601 * For the error path, TALLOC_FREE(rec) as well. In the good
602 * path we have already freed it.
610 NTSTATUS g_lock_unlock(struct g_lock_ctx *ctx, const char *name)
614 status = g_lock_force_unlock(ctx, name, messaging_server_id(ctx->msg));
616 #ifdef CLUSTER_SUPPORT
617 if (lp_clustering()) {
618 ctdb_unwatch(messaging_ctdbd_connection());
624 struct g_lock_locks_state {
625 int (*fn)(const char *name, void *private_data);
629 static int g_lock_locks_fn(struct db_record *rec, void *priv)
631 struct g_lock_locks_state *state = (struct g_lock_locks_state *)priv;
633 if ((rec->key.dsize == 0) || (rec->key.dptr[rec->key.dsize-1] != 0)) {
634 DEBUG(1, ("invalid key in g_lock.tdb, ignoring\n"));
637 return state->fn((char *)rec->key.dptr, state->private_data);
640 int g_lock_locks(struct g_lock_ctx *ctx,
641 int (*fn)(const char *name, void *private_data),
644 struct g_lock_locks_state state;
647 state.private_data = private_data;
649 return ctx->db->traverse_read(ctx->db, g_lock_locks_fn, &state);
652 NTSTATUS g_lock_dump(struct g_lock_ctx *ctx, const char *name,
653 int (*fn)(struct server_id pid,
654 enum g_lock_type lock_type,
660 struct g_lock_rec *locks = NULL;
663 if (ctx->db->fetch(ctx->db, talloc_tos(), string_term_tdb_data(name),
665 return NT_STATUS_NOT_FOUND;
668 if ((data.dsize == 0) || (data.dptr == NULL)) {
672 ret = g_lock_parse(talloc_tos(), data, &num_locks, &locks);
674 TALLOC_FREE(data.dptr);
677 DEBUG(10, ("g_lock_parse for %s failed\n", name));
678 return NT_STATUS_INTERNAL_ERROR;
681 for (i=0; i<num_locks; i++) {
682 if (fn(locks[i].pid, locks[i].lock_type, private_data) != 0) {
690 struct g_lock_get_state {
692 struct server_id *pid;
695 static int g_lock_get_fn(struct server_id pid, enum g_lock_type lock_type,
698 struct g_lock_get_state *state = (struct g_lock_get_state *)priv;
700 if ((lock_type & G_LOCK_PENDING) != 0) {
709 NTSTATUS g_lock_get(struct g_lock_ctx *ctx, const char *name,
710 struct server_id *pid)
712 struct g_lock_get_state state;
718 status = g_lock_dump(ctx, name, g_lock_get_fn, &state);
719 if (!NT_STATUS_IS_OK(status)) {
723 return NT_STATUS_NOT_FOUND;
728 static bool g_lock_init_all(TALLOC_CTX *mem_ctx,
729 struct tevent_context **pev,
730 struct messaging_context **pmsg,
731 const struct server_id self,
732 struct g_lock_ctx **pg_ctx)
734 struct tevent_context *ev = NULL;
735 struct messaging_context *msg = NULL;
736 struct g_lock_ctx *g_ctx = NULL;
738 ev = tevent_context_init(mem_ctx);
740 d_fprintf(stderr, "ERROR: could not init event context\n");
743 msg = messaging_init(mem_ctx, self, ev);
745 d_fprintf(stderr, "ERROR: could not init messaging context\n");
748 g_ctx = g_lock_ctx_init(mem_ctx, msg);
750 d_fprintf(stderr, "ERROR: could not init g_lock context\n");
765 NTSTATUS g_lock_do(const char *name, enum g_lock_type lock_type,
766 struct timeval timeout, const struct server_id self,
767 void (*fn)(void *private_data), void *private_data)
769 struct tevent_context *ev = NULL;
770 struct messaging_context *msg = NULL;
771 struct g_lock_ctx *g_ctx = NULL;
774 if (!g_lock_init_all(talloc_tos(), &ev, &msg, self, &g_ctx)) {
775 status = NT_STATUS_ACCESS_DENIED;
779 status = g_lock_lock(g_ctx, name, lock_type, timeout);
780 if (!NT_STATUS_IS_OK(status)) {
784 g_lock_unlock(g_ctx, name);