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/>.
21 #include "system/filesys.h"
24 #include "ctdbd_conn.h"
25 #include "../lib/util/select.h"
26 #include "system/select.h"
29 static NTSTATUS g_lock_force_unlock(struct g_lock_ctx *ctx, const char *name,
30 struct server_id pid);
33 struct db_context *db;
34 struct messaging_context *msg;
38 * The "g_lock.tdb" file contains records, indexed by the 0-terminated
39 * lockname. The record contains an array of "struct g_lock_rec"
40 * structures. Waiters have the lock_type with G_LOCK_PENDING or'ed.
44 enum g_lock_type lock_type;
48 struct g_lock_ctx *g_lock_ctx_init(TALLOC_CTX *mem_ctx,
49 struct messaging_context *msg)
51 struct g_lock_ctx *result;
53 result = talloc(mem_ctx, struct g_lock_ctx);
59 result->db = db_open(result, lock_path("g_lock.tdb"), 0,
60 TDB_CLEAR_IF_FIRST|TDB_INCOMPATIBLE_HASH, O_RDWR|O_CREAT, 0700);
61 if (result->db == NULL) {
62 DEBUG(1, ("g_lock_init: Could not open g_lock.tdb"));
69 static bool g_lock_conflicts(enum g_lock_type lock_type,
70 const struct g_lock_rec *rec)
72 enum g_lock_type rec_lock = rec->lock_type;
74 if ((rec_lock & G_LOCK_PENDING) != 0) {
79 * Only tested write locks so far. Very likely this routine
80 * needs to be fixed for read locks....
82 if ((lock_type == G_LOCK_READ) && (rec_lock == G_LOCK_READ)) {
88 static bool g_lock_parse(TALLOC_CTX *mem_ctx, TDB_DATA data,
89 int *pnum_locks, struct g_lock_rec **plocks)
92 struct g_lock_rec *locks;
94 if ((data.dsize % sizeof(struct g_lock_rec)) != 0) {
95 DEBUG(1, ("invalid lock record length %d\n", (int)data.dsize));
99 num_locks = data.dsize / sizeof(struct g_lock_rec);
100 locks = talloc_array(mem_ctx, struct g_lock_rec, num_locks);
102 DEBUG(1, ("talloc failed\n"));
106 memcpy(locks, data.dptr, data.dsize);
108 DEBUG(10, ("locks:\n"));
109 for (i=0; i<num_locks; i++) {
110 DEBUGADD(10, ("%s: %s %s\n",
111 procid_str(talloc_tos(), &locks[i].pid),
112 ((locks[i].lock_type & 1) == G_LOCK_READ) ?
114 (locks[i].lock_type & G_LOCK_PENDING) ?
115 "(pending)" : "(owner)"));
117 if (((locks[i].lock_type & G_LOCK_PENDING) == 0)
118 && !process_exists(locks[i].pid)) {
120 DEBUGADD(10, ("lock owner %s died -- discarding\n",
121 procid_str(talloc_tos(),
124 if (i < (num_locks-1)) {
125 locks[i] = locks[num_locks-1];
132 *pnum_locks = num_locks;
136 static void g_lock_cleanup(int *pnum_locks, struct g_lock_rec *locks)
140 num_locks = *pnum_locks;
142 DEBUG(10, ("g_lock_cleanup: %d locks\n", num_locks));
144 for (i=0; i<num_locks; i++) {
145 if (process_exists(locks[i].pid)) {
148 DEBUGADD(10, ("%s does not exist -- discarding\n",
149 procid_str(talloc_tos(), &locks[i].pid)));
151 if (i < (num_locks-1)) {
152 locks[i] = locks[num_locks-1];
156 *pnum_locks = num_locks;
160 static struct g_lock_rec *g_lock_addrec(TALLOC_CTX *mem_ctx,
161 struct g_lock_rec *locks,
163 const struct server_id pid,
164 enum g_lock_type lock_type)
166 struct g_lock_rec *result;
167 int num_locks = *pnum_locks;
169 result = talloc_realloc(mem_ctx, locks, struct g_lock_rec,
171 if (result == NULL) {
175 result[num_locks].pid = pid;
176 result[num_locks].lock_type = lock_type;
181 static void g_lock_got_retry(struct messaging_context *msg,
184 struct server_id server_id,
187 static NTSTATUS g_lock_trylock(struct g_lock_ctx *ctx, const char *name,
188 enum g_lock_type lock_type)
190 struct db_record *rec = NULL;
191 struct g_lock_rec *locks = NULL;
193 struct server_id self;
196 NTSTATUS status = NT_STATUS_OK;
197 NTSTATUS store_status;
200 rec = ctx->db->fetch_locked(ctx->db, talloc_tos(),
201 string_term_tdb_data(name));
203 DEBUG(10, ("fetch_locked(\"%s\") failed\n", name));
204 status = NT_STATUS_LOCK_NOT_GRANTED;
208 if (!g_lock_parse(talloc_tos(), rec->value, &num_locks, &locks)) {
209 DEBUG(10, ("g_lock_parse for %s failed\n", name));
210 status = NT_STATUS_INTERNAL_ERROR;
214 self = messaging_server_id(ctx->msg);
217 for (i=0; i<num_locks; i++) {
218 if (procid_equal(&self, &locks[i].pid)) {
219 if (our_index != -1) {
220 DEBUG(1, ("g_lock_trylock: Added ourself "
222 status = NT_STATUS_INTERNAL_ERROR;
225 if ((locks[i].lock_type & G_LOCK_PENDING) == 0) {
226 DEBUG(1, ("g_lock_trylock: Found ourself not "
228 status = NT_STATUS_INTERNAL_ERROR;
234 /* never conflict with ourself */
237 if (g_lock_conflicts(lock_type, &locks[i])) {
238 struct server_id pid = locks[i].pid;
240 if (!process_exists(pid)) {
243 status = g_lock_force_unlock(ctx, name, pid);
244 if (!NT_STATUS_IS_OK(status)) {
245 DEBUG(1, ("Could not unlock dead lock "
251 lock_type |= G_LOCK_PENDING;
255 if (our_index == -1) {
256 /* First round, add ourself */
258 locks = g_lock_addrec(talloc_tos(), locks, &num_locks,
261 DEBUG(10, ("g_lock_addrec failed\n"));
262 status = NT_STATUS_NO_MEMORY;
267 * Retry. We were pending last time. Overwrite the
268 * stored lock_type with what we calculated, we might
269 * have acquired the lock this time.
271 locks[our_index].lock_type = lock_type;
274 if (NT_STATUS_IS_OK(status) && ((lock_type & G_LOCK_PENDING) == 0)) {
276 * Walk through the list of locks, search for dead entries
278 g_lock_cleanup(&num_locks, locks);
281 data = make_tdb_data((uint8_t *)locks, num_locks * sizeof(*locks));
282 store_status = rec->store(rec, data, 0);
283 if (!NT_STATUS_IS_OK(store_status)) {
284 DEBUG(1, ("rec->store failed: %s\n",
285 nt_errstr(store_status)));
286 status = store_status;
293 if (NT_STATUS_IS_OK(status) && (lock_type & G_LOCK_PENDING) != 0) {
294 return STATUS_PENDING;
300 NTSTATUS g_lock_lock(struct g_lock_ctx *ctx, const char *name,
301 enum g_lock_type lock_type, struct timeval timeout)
303 struct tevent_timer *te = NULL;
306 struct timeval timeout_end;
307 struct timeval time_now;
309 DEBUG(10, ("Trying to acquire lock %d for %s\n", (int)lock_type,
312 if (lock_type & ~1) {
313 DEBUG(1, ("Got invalid lock type %d for %s\n",
314 (int)lock_type, name));
315 return NT_STATUS_INVALID_PARAMETER;
318 #ifdef CLUSTER_SUPPORT
319 if (lp_clustering()) {
320 status = ctdb_watch_us(messaging_ctdbd_connection());
321 if (!NT_STATUS_IS_OK(status)) {
322 DEBUG(10, ("could not register retry with ctdb: %s\n",
329 status = messaging_register(ctx->msg, &retry, MSG_DBWRAP_G_LOCK_RETRY,
331 if (!NT_STATUS_IS_OK(status)) {
332 DEBUG(10, ("messaging_register failed: %s\n",
337 time_now = timeval_current();
338 timeout_end = timeval_sum(&time_now, &timeout);
341 struct pollfd *pollfds;
345 struct timeval timeout_remaining, select_timeout;
347 status = g_lock_trylock(ctx, name, lock_type);
348 if (NT_STATUS_IS_OK(status)) {
349 DEBUG(10, ("Got lock %s\n", name));
352 if (!NT_STATUS_EQUAL(status, STATUS_PENDING)) {
353 DEBUG(10, ("g_lock_trylock failed: %s\n",
358 DEBUG(10, ("g_lock_trylock: Did not get lock, waiting...\n"));
360 /* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
361 * !!! HACK ALERT --- FIX ME !!!
362 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
363 * What we really want to do here is to react to
364 * MSG_DBWRAP_G_LOCK_RETRY messages that are either sent
365 * by a client doing g_lock_unlock or by ourselves when
366 * we receive a CTDB_SRVID_SAMBA_NOTIFY or
367 * CTDB_SRVID_RECONFIGURE message from ctdbd, i.e. when
368 * either a client holding a lock or a complete node
371 * Doing this properly involves calling tevent_loop_once(),
372 * but doing this here with the main ctdbd messaging context
373 * creates a nested event loop when g_lock_lock() is called
374 * from the main event loop, e.g. in a tcon_and_X where the
375 * share_info.tdb needs to be initialized and is locked by
376 * another process, or when the remore registry is accessed
377 * for writing and some other process already holds a lock
378 * on the registry.tdb.
380 * So as a quick fix, we act a little coarsely here: we do
381 * a select on the ctdb connection fd and when it is readable
382 * or we get EINTR, then we retry without actually parsing
383 * any ctdb packages or dispatching messages. This means that
384 * we retry more often than intended by design, but this does
385 * not harm and it is unobtrusive. When we have finished,
386 * the main loop will pick up all the messages and ctdb
387 * packets. The only extra twist is that we cannot use timed
388 * events here but have to handcode a timeout.
392 * We allocate 2 entries here. One is needed anyway for
393 * sys_poll and in the clustering case we might have to add
394 * the ctdb fd. This avoids the realloc then.
396 pollfds = talloc_array(talloc_tos(), struct pollfd, 2);
397 if (pollfds == NULL) {
398 status = NT_STATUS_NO_MEMORY;
403 #ifdef CLUSTER_SUPPORT
404 if (lp_clustering()) {
405 struct ctdbd_connection *conn;
406 conn = messaging_ctdbd_connection();
408 pollfds[0].fd = ctdbd_conn_get_fd(conn);
409 pollfds[0].events = POLLIN|POLLHUP;
415 time_now = timeval_current();
416 timeout_remaining = timeval_until(&time_now, &timeout_end);
417 select_timeout = timeval_set(60, 0);
419 select_timeout = timeval_min(&select_timeout,
422 ret = sys_poll(pollfds, num_pollfds,
423 timeval_to_msec(select_timeout));
426 * We're not *really interested in the actual flags. We just
427 * need to retry this whole thing.
430 TALLOC_FREE(pollfds);
434 if (errno != EINTR) {
435 DEBUG(1, ("error calling select: %s\n",
437 status = NT_STATUS_INTERNAL_ERROR;
442 * This means a signal was received.
443 * It might have been a MSG_DBWRAP_G_LOCK_RETRY message.
446 } else if (ret == 0) {
447 if (timeval_expired(&timeout_end)) {
448 DEBUG(10, ("g_lock_lock timed out\n"));
449 status = NT_STATUS_LOCK_NOT_GRANTED;
452 DEBUG(10, ("select returned 0 but timeout not "
453 "not expired, retrying\n"));
455 } else if (ret != 1) {
456 DEBUG(1, ("invalid return code of select: %d\n", ret));
457 status = NT_STATUS_INTERNAL_ERROR;
462 * This means ctdbd has sent us some data.
463 * Might be a CTDB_SRVID_RECONFIGURE or a
464 * CTDB_SRVID_SAMBA_NOTIFY message.
469 #ifdef CLUSTER_SUPPORT
473 if (!NT_STATUS_IS_OK(status)) {
474 NTSTATUS unlock_status;
476 unlock_status = g_lock_unlock(ctx, name);
478 if (!NT_STATUS_IS_OK(unlock_status)) {
479 DEBUG(1, ("Could not remove ourself from the locking "
480 "db: %s\n", nt_errstr(status)));
484 messaging_deregister(ctx->msg, MSG_DBWRAP_G_LOCK_RETRY, &retry);
490 static void g_lock_got_retry(struct messaging_context *msg,
493 struct server_id server_id,
496 bool *pretry = (bool *)private_data;
498 DEBUG(10, ("Got retry message from pid %s\n",
499 procid_str(talloc_tos(), &server_id)));
504 static NTSTATUS g_lock_force_unlock(struct g_lock_ctx *ctx, const char *name,
505 struct server_id pid)
507 struct db_record *rec = NULL;
508 struct g_lock_rec *locks = NULL;
510 enum g_lock_type lock_type;
513 rec = ctx->db->fetch_locked(ctx->db, talloc_tos(),
514 string_term_tdb_data(name));
516 DEBUG(10, ("fetch_locked(\"%s\") failed\n", name));
517 status = NT_STATUS_INTERNAL_ERROR;
521 if (!g_lock_parse(talloc_tos(), rec->value, &num_locks, &locks)) {
522 DEBUG(10, ("g_lock_parse for %s failed\n", name));
523 status = NT_STATUS_INTERNAL_ERROR;
527 for (i=0; i<num_locks; i++) {
528 if (procid_equal(&pid, &locks[i].pid)) {
533 if (i == num_locks) {
534 DEBUG(10, ("g_lock_force_unlock: Lock not found\n"));
535 status = NT_STATUS_INTERNAL_ERROR;
539 lock_type = locks[i].lock_type;
541 if (i < (num_locks-1)) {
542 locks[i] = locks[num_locks-1];
546 if (num_locks == 0) {
547 status = rec->delete_rec(rec);
550 data = make_tdb_data((uint8_t *)locks,
551 sizeof(struct g_lock_rec) * num_locks);
552 status = rec->store(rec, data, 0);
555 if (!NT_STATUS_IS_OK(status)) {
556 DEBUG(1, ("g_lock_force_unlock: Could not store record: %s\n",
563 if ((lock_type & G_LOCK_PENDING) == 0) {
567 * We've been the lock holder. Others to retry. Don't
568 * tell all others to avoid a thundering herd. In case
569 * this leads to a complete stall because we miss some
570 * processes, the loop in g_lock_lock tries at least
574 for (i=0; i<num_locks; i++) {
575 if ((locks[i].lock_type & G_LOCK_PENDING) == 0) {
578 if (!process_exists(locks[i].pid)) {
583 * Ping all waiters to retry
585 status = messaging_send(ctx->msg, locks[i].pid,
586 MSG_DBWRAP_G_LOCK_RETRY,
588 if (!NT_STATUS_IS_OK(status)) {
589 DEBUG(1, ("sending retry to %s failed: %s\n",
590 procid_str(talloc_tos(),
596 if (num_wakeups > 5) {
603 * For the error path, TALLOC_FREE(rec) as well. In the good
604 * path we have already freed it.
612 NTSTATUS g_lock_unlock(struct g_lock_ctx *ctx, const char *name)
616 status = g_lock_force_unlock(ctx, name, messaging_server_id(ctx->msg));
618 #ifdef CLUSTER_SUPPORT
619 if (lp_clustering()) {
620 ctdb_unwatch(messaging_ctdbd_connection());
626 struct g_lock_locks_state {
627 int (*fn)(const char *name, void *private_data);
631 static int g_lock_locks_fn(struct db_record *rec, void *priv)
633 struct g_lock_locks_state *state = (struct g_lock_locks_state *)priv;
635 if ((rec->key.dsize == 0) || (rec->key.dptr[rec->key.dsize-1] != 0)) {
636 DEBUG(1, ("invalid key in g_lock.tdb, ignoring\n"));
639 return state->fn((char *)rec->key.dptr, state->private_data);
642 int g_lock_locks(struct g_lock_ctx *ctx,
643 int (*fn)(const char *name, void *private_data),
646 struct g_lock_locks_state state;
649 state.private_data = private_data;
651 return ctx->db->traverse_read(ctx->db, g_lock_locks_fn, &state);
654 NTSTATUS g_lock_dump(struct g_lock_ctx *ctx, const char *name,
655 int (*fn)(struct server_id pid,
656 enum g_lock_type lock_type,
662 struct g_lock_rec *locks = NULL;
665 if (ctx->db->fetch(ctx->db, talloc_tos(), string_term_tdb_data(name),
667 return NT_STATUS_NOT_FOUND;
670 if ((data.dsize == 0) || (data.dptr == NULL)) {
674 ret = g_lock_parse(talloc_tos(), data, &num_locks, &locks);
676 TALLOC_FREE(data.dptr);
679 DEBUG(10, ("g_lock_parse for %s failed\n", name));
680 return NT_STATUS_INTERNAL_ERROR;
683 for (i=0; i<num_locks; i++) {
684 if (fn(locks[i].pid, locks[i].lock_type, private_data) != 0) {
692 struct g_lock_get_state {
694 struct server_id *pid;
697 static int g_lock_get_fn(struct server_id pid, enum g_lock_type lock_type,
700 struct g_lock_get_state *state = (struct g_lock_get_state *)priv;
702 if ((lock_type & G_LOCK_PENDING) != 0) {
711 NTSTATUS g_lock_get(struct g_lock_ctx *ctx, const char *name,
712 struct server_id *pid)
714 struct g_lock_get_state state;
720 status = g_lock_dump(ctx, name, g_lock_get_fn, &state);
721 if (!NT_STATUS_IS_OK(status)) {
725 return NT_STATUS_NOT_FOUND;
730 static bool g_lock_init_all(TALLOC_CTX *mem_ctx,
731 struct tevent_context **pev,
732 struct messaging_context **pmsg,
733 const struct server_id self,
734 struct g_lock_ctx **pg_ctx)
736 struct tevent_context *ev = NULL;
737 struct messaging_context *msg = NULL;
738 struct g_lock_ctx *g_ctx = NULL;
740 ev = tevent_context_init(mem_ctx);
742 d_fprintf(stderr, "ERROR: could not init event context\n");
745 msg = messaging_init(mem_ctx, self, ev);
747 d_fprintf(stderr, "ERROR: could not init messaging context\n");
750 g_ctx = g_lock_ctx_init(mem_ctx, msg);
752 d_fprintf(stderr, "ERROR: could not init g_lock context\n");
767 NTSTATUS g_lock_do(const char *name, enum g_lock_type lock_type,
768 struct timeval timeout, const struct server_id self,
769 void (*fn)(void *private_data), void *private_data)
771 struct tevent_context *ev = NULL;
772 struct messaging_context *msg = NULL;
773 struct g_lock_ctx *g_ctx = NULL;
776 if (!g_lock_init_all(talloc_tos(), &ev, &msg, self, &g_ctx)) {
777 status = NT_STATUS_ACCESS_DENIED;
781 status = g_lock_lock(g_ctx, name, lock_type, timeout);
782 if (!NT_STATUS_IS_OK(status)) {
786 g_lock_unlock(g_ctx, name);