2 wait for a tdb chain lock
4 Copyright (C) Andrew Tridgell 2006
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 "lib/tevent/tevent.h"
22 #include "system/filesys.h"
23 #include "system/wait.h"
25 #include "lib/tdb/include/tdb.h"
26 #include "lib/util/dlinklist.h"
27 #include "../include/ctdb_private.h"
30 struct lockwait_handle {
31 struct lockwait_handle *next, *prev;
32 struct ctdb_context *ctdb;
33 struct ctdb_db_context *ctdb_db;
38 void (*callback)(void *);
40 struct timeval start_time;
43 /* If we managed to obtain a lock, find any overflow records which wanted the
44 * same one and do all the callbacks at once. */
45 static void do_overflow(struct ctdb_db_context *ctdb_db,
48 struct lockwait_handle *i, *next;
49 TALLOC_CTX *tmp_ctx = talloc_new(ctdb_db);
51 for (i = ctdb_db->lockwait_overflow; i; i = next) {
52 /* Careful: destructor removes it from list! */
54 if (key.dsize == i->key.dsize
55 && memcmp(key.dptr, i->key.dptr, key.dsize) == 0) {
56 /* Callback might free them, so reparent. */
57 talloc_steal(tmp_ctx, i);
58 i->callback(i->private_data);
62 /* This will free them if callback didn't. */
65 /* Remove one from the overflow queue if there is one. */
66 if (ctdb_db->lockwait_overflow) {
67 i = ctdb->lockwait_overflow;
68 ctdb_lockwait(ctdb_db, i->key, i->callback, i->private_data);
73 static void lockwait_handler(struct event_context *ev, struct fd_event *fde,
74 uint16_t flags, void *private_data)
76 struct lockwait_handle *h = talloc_get_type(private_data,
77 struct lockwait_handle);
78 void (*callback)(void *) = h->callback;
79 void *p = h->private_data;
80 pid_t child = h->child;
81 TDB_DATA key = h->key;
82 struct tdb_context *tdb = h->ctdb_db->ltdb->tdb;
83 TALLOC_CTX *tmp_ctx = talloc_new(ev);
85 key.dptr = talloc_memdup(tmp_ctx, key.dptr, key.dsize);
86 h->ctdb_db->pending_requests--;
88 talloc_set_destructor(h, NULL);
89 CTDB_UPDATE_LATENCY(h->ctdb, h->ctdb_db, "lockwait", lockwait_latency, h->start_time);
90 CTDB_DECREMENT_STAT(h->ctdb, pending_lockwait_calls);
92 /* the handle needs to go away when the context is gone - when
93 the handle goes away this implicitly closes the pipe, which
94 kills the child holding the lock */
95 talloc_steal(tmp_ctx, h);
97 if (h->ctdb->flags & CTDB_FLAG_TORTURE) {
98 if (tdb_chainlock_nonblock(tdb, key) == 0) {
99 ctdb_fatal(h->ctdb, "got chain lock while lockwait child active");
103 tdb_chainlock_mark(tdb, key);
105 if (h->ctdb_db->lockwait_overflow) {
106 do_overflow(h->ctdb_db, key);
108 tdb_chainlock_unmark(tdb, key);
110 kill(child, SIGKILL);
111 talloc_free(tmp_ctx);
114 static int lockwait_destructor(struct lockwait_handle *h)
116 CTDB_DECREMENT_STAT(h->ctdb, pending_lockwait_calls);
117 kill(h->child, SIGKILL);
118 h->ctdb_db->pending_requests--;
122 static int overflow_lockwait_destructor(struct lockwait_handle *h)
124 CTDB_DECREMENT_STAT(h->ctdb, pending_lockwait_calls);
125 DLIST_REMOVE(h->ctdb_db->lockwait_overflow, h);
130 setup a non-blocking chainlock on a tdb record. If this function
131 returns NULL then it could not get the chainlock. Otherwise it
132 returns a opaque handle, and will call callback() once it has
133 managed to get the chainlock. You can cancel it by using talloc_free
134 on the returned handle.
136 It is the callers responsibility to unlock the chainlock once
139 struct lockwait_handle *ctdb_lockwait(struct ctdb_db_context *ctdb_db,
141 void (*callback)(void *private_data),
144 struct lockwait_handle *result;
146 pid_t parent = getpid();
148 CTDB_INCREMENT_STAT(ctdb_db->ctdb, lockwait_calls);
149 CTDB_INCREMENT_STAT(ctdb_db->ctdb, pending_lockwait_calls);
151 if (!(result = talloc_zero(private_data, struct lockwait_handle))) {
152 CTDB_DECREMENT_STAT(ctdb_db->ctdb, pending_lockwait_calls);
156 result->callback = callback;
157 result->private_data = private_data;
158 result->ctdb = ctdb_db->ctdb;
159 result->ctdb_db = ctdb_db;
162 /* Don't fire off too many children at once! */
163 if (ctdb_db->pending_requests > 200) {
164 DLIST_ADD_END(ctdb_db->lockwait_overflow, result, NULL);
165 talloc_set_destructor(result, overflow_lockwait_destructor);
166 DEBUG(DEBUG_DEBUG, (__location__ " Created overflow for %s\n",
171 ret = pipe(result->fd);
175 CTDB_DECREMENT_STAT(ctdb_db->ctdb, pending_lockwait_calls);
179 result->child = ctdb_fork(ctdb_db->ctdb);
181 if (result->child == (pid_t)-1) {
182 close(result->fd[0]);
183 close(result->fd[1]);
185 CTDB_DECREMENT_STAT(ctdb_db->ctdb, pending_lockwait_calls);
189 if (result->child == 0) {
191 close(result->fd[0]);
192 debug_extra = talloc_asprintf(NULL, "chainlock-%s:", ctdb_db->db_name);
193 tdb_chainlock(ctdb_db->ltdb->tdb, key);
194 write(result->fd[1], &c, 1);
195 /* make sure we die when our parent dies */
196 while (kill(parent, 0) == 0 || errno != ESRCH) {
202 close(result->fd[1]);
203 set_close_on_exec(result->fd[0]);
205 DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d to child lockwait process\n", result->fd[0]));
207 ctdb_db->pending_requests++;
208 talloc_set_destructor(result, lockwait_destructor);
210 result->fde = event_add_fd(ctdb_db->ctdb->ev, result, result->fd[0],
211 EVENT_FD_READ, lockwait_handler,
213 if (result->fde == NULL) {
215 CTDB_DECREMENT_STAT(ctdb_db->ctdb, pending_lockwait_calls);
218 tevent_fd_set_auto_close(result->fde);
220 result->start_time = timeval_current();