2 Unix SMB/CIFS implementation.
4 process model: prefork (n client connections per process)
6 Copyright (C) Andrew Tridgell 1992-2005
7 Copyright (C) James J Myers 2003 <myersjj@samba.org>
8 Copyright (C) Stefan (metze) Metzmacher 2004
9 Copyright (C) Andrew Bartlett 2008 <abartlet@samba.org>
10 Copyright (C) David Disseldorp 2008 <ddiss@sgi.com>
12 This program is free software; you can redistribute it and/or modify
13 it under the terms of the GNU General Public License as published by
14 the Free Software Foundation; either version 3 of the License, or
15 (at your option) any later version.
17 This program is distributed in the hope that it will be useful,
18 but WITHOUT ANY WARRANTY; without even the implied warranty of
19 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 GNU General Public License for more details.
22 You should have received a copy of the GNU General Public License
23 along with this program. If not, see <http://www.gnu.org/licenses/>.
26 * The pre-fork process model distributes the server workload amongst several
27 * designated worker threads (e.g. 'prefork-worker-ldap-0',
28 * 'prefork-worker-ldap-1', etc). The number of worker threads is controlled
29 * by the 'prefork children' conf setting. The worker threads are controlled
30 * by a prefork master process (e.g. 'prefork-master-ldap'). The prefork master
31 * doesn't handle the server workload (i.e. processing messages) itself, but is
32 * responsible for restarting workers if they exit unexpectedly. The top-level
33 * samba process is responsible for restarting the master process if it exits.
38 #include "lib/events/events.h"
39 #include "lib/messaging/messaging.h"
40 #include "lib/socket/socket.h"
41 #include "samba/process_model.h"
42 #include "cluster/cluster.h"
43 #include "param/param.h"
45 #include "lib/util/tfork.h"
46 #include "lib/messaging/irpc.h"
47 #include "lib/util/util_process.h"
48 #include "server_util.h"
50 #define min(a, b) (((a) < (b)) ? (a) : (b))
52 NTSTATUS
process_model_prefork_init(void);
53 static void prefork_new_task(
54 struct tevent_context
*ev
,
55 struct loadparm_context
*lp_ctx
,
56 const char *service_name
,
57 struct task_server
*(*new_task_fn
)(struct tevent_context
*,
58 struct loadparm_context
*lp_ctx
,
63 const struct service_details
*service_details
,
65 static void prefork_fork_worker(struct task_server
*task
,
66 struct tevent_context
*ev
,
67 struct tevent_context
*ev2
,
68 struct loadparm_context
*lp_ctx
,
69 const struct service_details
*service_details
,
70 const char *service_name
,
72 unsigned restart_delay
,
73 struct process_details
*pd
);
74 static void prefork_child_pipe_handler(struct tevent_context
*ev
,
75 struct tevent_fd
*fde
,
78 static void setup_handlers(struct tevent_context
*ev
,
79 struct loadparm_context
*lp_ctx
,
83 * State needed to restart the master process or a worker process if they
86 struct master_restart_context
{
87 struct task_server
*(*new_task_fn
)(struct tevent_context
*,
88 struct loadparm_context
*lp_ctx
,
95 struct worker_restart_context
{
96 unsigned int instance
;
97 struct task_server
*task
;
98 struct tevent_context
*ev2
;
102 struct restart_context
{
103 struct loadparm_context
*lp_ctx
;
106 const struct service_details
*service_details
;
107 const char *service_name
;
108 unsigned restart_delay
;
109 struct master_restart_context
*master
;
110 struct worker_restart_context
*worker
;
113 static void sighup_signal_handler(struct tevent_context
*ev
,
114 struct tevent_signal
*se
,
115 int signum
, int count
, void *siginfo
,
118 reopen_logs_internal();
121 static void sigterm_signal_handler(struct tevent_context
*ev
,
122 struct tevent_signal
*se
,
123 int signum
, int count
, void *siginfo
,
127 if (getpgrp() == getpid()) {
129 * We're the process group leader, send
130 * SIGTERM to our process group.
132 DBG_NOTICE("SIGTERM: killing children\n");
133 kill(-getpgrp(), SIGTERM
);
136 DBG_NOTICE("Exiting pid %d on SIGTERM\n", getpid());
142 called when the process model is selected
144 static void prefork_model_init(void)
148 static void prefork_reload_after_fork(void)
152 ldb_wrap_fork_hook();
153 /* Must be done after a fork() to reset messaging contexts. */
154 status
= imessaging_reinit_all();
155 if (!NT_STATUS_IS_OK(status
)) {
156 smb_panic("Failed to re-initialise imessaging after fork");
158 force_check_log_size();
162 * clean up any messaging associated with the old process.
165 static void irpc_cleanup(
166 struct loadparm_context
*lp_ctx
,
167 struct tevent_context
*ev
,
170 TALLOC_CTX
*mem_ctx
= talloc_new(NULL
);
171 struct imessaging_context
*msg_ctx
= NULL
;
172 NTSTATUS status
= NT_STATUS_OK
;
174 if (mem_ctx
== NULL
) {
175 DBG_ERR("OOM cleaning up irpc\n");
178 msg_ctx
= imessaging_client_init(mem_ctx
, lp_ctx
, ev
);
179 if (msg_ctx
== NULL
) {
180 DBG_ERR("Unable to create imessaging_context\n");
181 TALLOC_FREE(mem_ctx
);
184 status
= imessaging_process_cleanup(msg_ctx
, pid
);
185 if (!NT_STATUS_IS_OK(status
)) {
186 DBG_ERR("imessaging_process_cleanup returned (%s)\n",
188 TALLOC_FREE(mem_ctx
);
192 TALLOC_FREE(mem_ctx
);
196 * handle EOF on the parent-to-all-children pipe in the child, i.e.
197 * the parent has died and its end of the pipe has been closed.
198 * The child handles this by exiting as well.
200 static void prefork_pipe_handler(struct tevent_context
*event_ctx
,
201 struct tevent_fd
*fde
, uint16_t flags
,
204 struct loadparm_context
*lp_ctx
= NULL
;
208 * free the fde which removes the event and stops it firing again
213 * Clean up any irpc end points this process had.
216 lp_ctx
= talloc_get_type_abort(private_data
, struct loadparm_context
);
217 irpc_cleanup(lp_ctx
, event_ctx
, pid
);
219 DBG_NOTICE("Child %d exiting\n", getpid());
220 TALLOC_FREE(event_ctx
);
226 * Called by the top-level samba process to create a new prefork master process
228 static void prefork_fork_master(
229 struct tevent_context
*ev
,
230 struct loadparm_context
*lp_ctx
,
231 const char *service_name
,
232 struct task_server
*(*new_task_fn
)(struct tevent_context
*,
233 struct loadparm_context
*lp_ctx
,
238 const struct service_details
*service_details
,
239 unsigned restart_delay
,
243 struct tfork
* t
= NULL
;
246 struct tevent_context
*ev2
;
247 struct task_server
*task
= NULL
;
248 struct process_details pd
= initial_process_details
;
249 struct samba_tevent_trace_state
*samba_tevent_trace_state
= NULL
;
254 smb_panic("failure in tfork\n");
257 DBG_NOTICE("Forking [%s] pre-fork master process\n", service_name
);
258 pid
= tfork_child_pid(t
);
260 struct tevent_fd
*fde
= NULL
;
261 int fd
= tfork_event_fd(t
);
262 struct restart_context
*rc
= NULL
;
264 /* Register a pipe handler that gets called when the prefork
265 * master process terminates.
267 rc
= talloc_zero(ev
, struct restart_context
);
269 smb_panic("OOM allocating restart context\n");
273 rc
->service_name
= service_name
;
274 rc
->service_details
= service_details
;
275 rc
->from_parent_fd
= from_parent_fd
;
276 rc
->restart_delay
= restart_delay
;
277 rc
->master
= talloc_zero(rc
, struct master_restart_context
);
278 if (rc
->master
== NULL
) {
279 smb_panic("OOM allocating master restart context\n");
282 rc
->master
->new_task_fn
= new_task_fn
;
283 rc
->master
->private_data
= private_data
;
286 ev
, ev
, fd
, TEVENT_FD_READ
, prefork_child_pipe_handler
, rc
);
288 smb_panic("Failed to add child pipe handler, "
291 tevent_fd_set_auto_close(fde
);
297 process_set_title("%s[master]", "task[%s] pre-fork master", service_name
);
300 * this will free all the listening sockets and all state that
301 * is not associated with this new connection
303 if (tevent_re_initialise(ev
) != 0) {
304 smb_panic("Failed to re-initialise tevent after fork");
306 prefork_reload_after_fork();
307 setup_handlers(ev
, lp_ctx
, from_parent_fd
);
309 if (service_details
->inhibit_pre_fork
) {
311 ev
, lp_ctx
, cluster_id(pid
, 0), private_data
, NULL
);
313 * The task does not support pre-fork
315 if (task
!= NULL
&& service_details
->post_fork
!= NULL
) {
316 service_details
->post_fork(task
, &pd
);
318 tevent_loop_wait(ev
);
324 * This is now the child code. We need a completely new event_context
327 ev2
= s4_event_context_init(NULL
);
329 samba_tevent_trace_state
= create_samba_tevent_trace_state(ev2
);
330 if (samba_tevent_trace_state
== NULL
) {
336 tevent_set_trace_callback(ev2
,
337 samba_tevent_trace_callback
,
338 samba_tevent_trace_state
);
340 /* setup this new connection: process will bind to it's sockets etc
342 * While we can use ev for the child, which has been re-initialised
343 * above we must run the new task under ev2 otherwise the children would
344 * be listening on the sockets. Also we don't want the top level
345 * process accepting and handling requests, it's responsible for
346 * monitoring and controlling the child work processes.
348 task
= new_task_fn(ev2
, lp_ctx
, cluster_id(pid
, 0), private_data
, NULL
);
356 * Register an irpc name that can be used by the samba-tool processes
360 struct talloc_ctx
*ctx
= talloc_new(NULL
);
363 DBG_ERR("Out of memory");
366 name
= talloc_asprintf(ctx
, "prefork-master-%s", service_name
);
367 irpc_add_name(task
->msg_ctx
, name
);
372 int default_children
;
373 default_children
= lpcfg_prefork_children(lp_ctx
);
374 num_children
= lpcfg_parm_int(lp_ctx
, NULL
, "prefork children",
375 service_name
, default_children
);
377 if (num_children
== 0) {
378 DBG_WARNING("Number of pre-fork children for %s is zero, "
379 "NO worker processes will be started for %s\n",
380 service_name
, service_name
);
382 DBG_NOTICE("Forking %d %s worker processes\n",
383 num_children
, service_name
);
386 * the prefork master creates its own control pipe, so the prefork
387 * workers can detect if the master exits (in which case an EOF gets
388 * written). (Whereas from_parent_fd is the control pipe from the
389 * top-level process that the prefork master listens on)
393 ret
= pipe(control_pipe
);
395 smb_panic("Unable to create worker control pipe\n");
397 smb_set_close_on_exec(control_pipe
[0]);
398 smb_set_close_on_exec(control_pipe
[1]);
402 * We are now free to spawn some worker processes
404 for (i
=0; i
< num_children
; i
++) {
405 prefork_fork_worker(task
,
417 /* Don't listen on the sockets we just gave to the children */
418 tevent_loop_wait(ev
);
420 /* We need to keep ev2 until we're finished for the messaging to work */
425 static void prefork_restart_fn(struct tevent_context
*ev
,
426 struct tevent_timer
*te
,
431 * Restarts a child process if it exits unexpectedly
433 static bool prefork_restart(struct tevent_context
*ev
,
434 struct restart_context
*rc
)
436 struct tevent_timer
*te
= NULL
;
438 if (rc
->restart_delay
> 0) {
439 DBG_ERR("Restarting [%s] pre-fork %s in (%d) seconds\n",
441 (rc
->master
== NULL
) ? "worker" : "master",
446 * Always use an async timer event. If
447 * rc->restart_delay is zero this is the
448 * same as an immediate event and will be
449 * called immediately we go back into the
452 te
= tevent_add_timer(ev
,
454 tevent_timeval_current_ofs(rc
->restart_delay
, 0),
458 DBG_ERR("tevent_add_timer fail [%s] pre-fork event %s\n",
460 (rc
->master
== NULL
) ? "worker" : "master");
461 /* Caller needs to free rc. */
464 /* Caller must not free rc - it's in use. */
468 static void prefork_restart_fn(struct tevent_context
*ev
,
469 struct tevent_timer
*te
,
473 unsigned max_backoff
= 0;
474 unsigned backoff
= 0;
475 unsigned default_value
= 0;
476 struct restart_context
*rc
= talloc_get_type(private_data
,
477 struct restart_context
);
478 unsigned restart_delay
= rc
->restart_delay
;
483 * If the child process is constantly exiting, then restarting it can
484 * consume a lot of resources. In which case, we want to backoff a bit
485 * before respawning it
487 default_value
= lpcfg_prefork_backoff_increment(rc
->lp_ctx
);
488 backoff
= lpcfg_parm_int(rc
->lp_ctx
,
490 "prefork backoff increment",
494 default_value
= lpcfg_prefork_maximum_backoff(rc
->lp_ctx
);
495 max_backoff
= lpcfg_parm_int(rc
->lp_ctx
,
497 "prefork maximum backoff",
501 restart_delay
+= backoff
;
502 restart_delay
= min(restart_delay
, max_backoff
);
504 if (rc
->master
!= NULL
) {
505 DBG_ERR("Restarting [%s] pre-fork master\n", rc
->service_name
);
506 prefork_fork_master(ev
,
509 rc
->master
->new_task_fn
,
510 rc
->master
->private_data
,
514 } else if (rc
->worker
!= NULL
) {
515 struct process_details pd
= initial_process_details
;
516 DBG_ERR("Restarting [%s] pre-fork worker(%d)\n",
518 rc
->worker
->instance
);
519 pd
.instances
= rc
->worker
->instance
;
520 prefork_fork_worker(rc
->worker
->task
,
526 rc
->worker
->control_pipe
,
530 /* tfork allocates tfork structures with malloc */
531 tfork_destroy(&rc
->t
);
537 handle EOF on the child pipe in the parent, so we know when a
538 process terminates without using SIGCHLD or waiting on all possible pids.
540 We need to ensure we do not ignore SIGCHLD because we need it to
541 work to get a valid error code from samba_runcmd_*().
543 static void prefork_child_pipe_handler(struct tevent_context
*ev
,
544 struct tevent_fd
*fde
,
548 struct restart_context
*rc
= NULL
;
551 bool rc_inuse
= false;
553 /* free the fde which removes the event and stops it firing again */
556 /* the child has closed the pipe, assume its dead */
558 rc
= talloc_get_type_abort(private_data
, struct restart_context
);
559 pid
= tfork_child_pid(rc
->t
);
562 irpc_cleanup(rc
->lp_ctx
, ev
, pid
);
563 status
= tfork_status(&rc
->t
, false);
565 DBG_ERR("Parent %d, Child %d terminated, "
566 "unable to get status code from tfork\n",
568 rc_inuse
= prefork_restart(ev
, rc
);
569 } else if (WIFEXITED(status
)) {
570 status
= WEXITSTATUS(status
);
571 DBG_ERR("Parent %d, Child %d exited with status %d\n",
572 getpid(), pid
, status
);
574 rc_inuse
= prefork_restart(ev
, rc
);
576 } else if (WIFSIGNALED(status
)) {
577 status
= WTERMSIG(status
);
578 DBG_ERR("Parent %d, Child %d terminated with signal %d\n",
579 getpid(), pid
, status
);
580 if (status
== SIGABRT
|| status
== SIGBUS
|| status
== SIGFPE
||
581 status
== SIGILL
|| status
== SIGSYS
|| status
== SIGSEGV
||
584 rc_inuse
= prefork_restart(ev
, rc
);
588 /* tfork allocates tfork structures with malloc */
589 tfork_destroy(&rc
->t
);
597 called when a listening socket becomes readable.
599 static void prefork_accept_connection(
600 struct tevent_context
*ev
,
601 struct loadparm_context
*lp_ctx
,
602 struct socket_context
*listen_socket
,
603 void (*new_conn
)(struct tevent_context
*,
604 struct loadparm_context
*,
605 struct socket_context
*,
610 void *process_context
)
613 struct socket_context
*connected_socket
;
614 pid_t pid
= getpid();
616 /* accept an incoming connection. */
617 status
= socket_accept(listen_socket
, &connected_socket
);
618 if (!NT_STATUS_IS_OK(status
)) {
620 * For prefork we can ignore STATUS_MORE_ENTRIES, as once a
621 * connection becomes available all waiting processes are
622 * woken, but only one gets work to process.
623 * AKA the thundering herd.
624 * In the short term this should not be an issue as the number
625 * of workers should be a small multiple of the number of cpus
626 * In the longer term socket_accept needs to implement a
627 * mutex/semaphore (like apache does) to serialise the accepts
629 if (!NT_STATUS_EQUAL(status
, STATUS_MORE_ENTRIES
)) {
630 DBG_ERR("Worker process (%d), error in accept [%s]\n",
631 getpid(), nt_errstr(status
));
636 talloc_steal(private_data
, connected_socket
);
638 new_conn(ev
, lp_ctx
, connected_socket
,
639 cluster_id(pid
, socket_get_fd(connected_socket
)),
640 private_data
, process_context
);
643 static void setup_handlers(
644 struct tevent_context
*ev
,
645 struct loadparm_context
*lp_ctx
,
648 struct tevent_fd
*fde
= NULL
;
649 struct tevent_signal
*se
= NULL
;
651 fde
= tevent_add_fd(ev
, ev
, from_parent_fd
, TEVENT_FD_READ
,
652 prefork_pipe_handler
, lp_ctx
);
654 smb_panic("Failed to add fd handler after fork");
657 se
= tevent_add_signal(ev
,
661 sighup_signal_handler
,
664 smb_panic("Failed to add SIGHUP handler after fork");
667 se
= tevent_add_signal(ev
,
671 sigterm_signal_handler
,
674 smb_panic("Failed to add SIGTERM handler after fork");
679 * Called by the prefork master to create a new prefork worker process
681 static void prefork_fork_worker(struct task_server
*task
,
682 struct tevent_context
*ev
,
683 struct tevent_context
*ev2
,
684 struct loadparm_context
*lp_ctx
,
685 const struct service_details
*service_details
,
686 const char *service_name
,
688 unsigned restart_delay
,
689 struct process_details
*pd
)
691 struct tfork
*w
= NULL
;
696 smb_panic("failure in tfork\n");
699 pid
= tfork_child_pid(w
);
701 struct tevent_fd
*fde
= NULL
;
702 int fd
= tfork_event_fd(w
);
703 struct restart_context
*rc
= NULL
;
706 * we're the parent (prefork master), so store enough info to
707 * restart the worker/child if it exits unexpectedly
709 rc
= talloc_zero(ev
, struct restart_context
);
711 smb_panic("OOM allocating restart context\n");
715 rc
->service_name
= service_name
;
716 rc
->service_details
= service_details
;
717 rc
->restart_delay
= restart_delay
;
719 rc
->worker
= talloc_zero(rc
, struct worker_restart_context
);
720 if (rc
->worker
== NULL
) {
721 smb_panic("OOM allocating master restart context\n");
723 rc
->worker
->ev2
= ev2
;
724 rc
->worker
->instance
= pd
->instances
;
725 rc
->worker
->task
= task
;
726 rc
->worker
->control_pipe
[0] = control_pipe
[0];
727 rc
->worker
->control_pipe
[1] = control_pipe
[1];
730 ev
, ev
, fd
, TEVENT_FD_READ
, prefork_child_pipe_handler
, rc
);
732 smb_panic("Failed to add child pipe handler, "
735 tevent_fd_set_auto_close(fde
);
739 * we're the child (prefork-worker). We never write to the
740 * control pipe, but listen on the read end in case our parent
741 * (the pre-fork master) exits
743 close(control_pipe
[1]);
744 setup_handlers(ev2
, lp_ctx
, control_pipe
[0]);
753 process_set_title("%s(%d)",
754 "task[%s] pre-forked worker(%d)",
758 prefork_reload_after_fork();
759 if (service_details
->post_fork
!= NULL
) {
760 service_details
->post_fork(task
, pd
);
763 struct talloc_ctx
*ctx
= talloc_new(NULL
);
766 smb_panic("OOM allocating talloc context\n");
768 name
= talloc_asprintf(ctx
,
769 "prefork-worker-%s-%d",
772 irpc_add_name(task
->msg_ctx
, name
);
775 tevent_loop_wait(ev2
);
781 * called to create a new server task
783 static void prefork_new_task(
784 struct tevent_context
*ev
,
785 struct loadparm_context
*lp_ctx
,
786 const char *service_name
,
787 struct task_server
*(*new_task_fn
)(struct tevent_context
*,
788 struct loadparm_context
*lp_ctx
,
789 struct server_id
, void *, void *),
791 const struct service_details
*service_details
,
794 prefork_fork_master(ev
,
806 * called when a task terminates
808 static void prefork_terminate_task(struct tevent_context
*ev
,
809 struct loadparm_context
*lp_ctx
,
812 void *process_context
)
814 DBG_DEBUG("called with reason[%s]\n", reason
);
824 * called when a connection completes
826 static void prefork_terminate_connection(struct tevent_context
*ev
,
827 struct loadparm_context
*lp_ctx
,
829 void *process_context
)
833 /* called to set a title of a task or connection */
834 static void prefork_set_title(struct tevent_context
*ev
, const char *title
)
838 static const struct model_ops prefork_ops
= {
840 .model_init
= prefork_model_init
,
841 .accept_connection
= prefork_accept_connection
,
842 .new_task
= prefork_new_task
,
843 .terminate_task
= prefork_terminate_task
,
844 .terminate_connection
= prefork_terminate_connection
,
845 .set_title
= prefork_set_title
,
849 * initialise the prefork process model, registering ourselves with the
850 * process model subsystem
852 NTSTATUS
process_model_prefork_init(void)
854 return register_process_model(&prefork_ops
);