1 /* Licensed to the Apache Software Foundation (ASF) under one or more
2 * contributor license agreements. See the NOTICE file distributed with
3 * this work for additional information regarding copyright ownership.
4 * The ASF licenses this file to You under the Apache License, Version 2.0
5 * (the "License"); you may not use this file except in compliance with
6 * the License. You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 /* The purpose of this file is to store the code that MOST mpm's will need
18 * this does not mean a function only goes into this file if every MPM needs
19 * it. It means that if a function is needed by more than one MPM, and
20 * future maintenance would be served by making the code common, then the
21 * function belongs here.
23 * This is going in src/main because it is not platform specific, it is
24 * specific to multi-process servers, but NOT to Unix. Which is why it
25 * does not belong in src/os/unix
29 #include "apr_thread_proc.h"
30 #include "apr_signal.h"
31 #include "apr_strings.h"
32 #define APR_WANT_STRFUNC
34 #include "apr_getopt.h"
35 #include "apr_optional.h"
36 #include "apr_allocator.h"
39 #include "http_config.h"
41 #include "http_main.h"
42 #include "mpm_common.h"
44 #include "ap_listen.h"
45 #include "scoreboard.h"
46 #include "util_mutex.h"
59 typedef enum {DO_NOTHING, SEND_SIGTERM, SEND_SIGKILL, GIVEUP} action_t;
61 typedef struct extra_process_t {
62 struct extra_process_t *next;
66 static extra_process_t *extras;
68 void ap_register_extra_mpm_process(pid_t pid)
70 extra_process_t *p = (extra_process_t *)malloc(sizeof(extra_process_t));
77 int ap_unregister_extra_mpm_process(pid_t pid)
79 extra_process_t *cur = extras;
80 extra_process_t *prev = NULL;
82 while (cur && cur->pid != pid) {
89 prev->next = cur->next;
98 /* we don't know about any such process */
103 static int reclaim_one_pid(pid_t pid, action_t action)
106 apr_status_t waitret;
110 /* Ensure pid sanity. */
116 waitret = apr_proc_wait(&proc, &status, &why, APR_NOWAIT);
117 if (waitret != APR_CHILD_NOTDONE) {
118 if (waitret == APR_CHILD_DONE)
119 ap_process_child_status(&proc, why, status);
128 /* ok, now it's being annoying */
129 ap_log_error(APLOG_MARK, APLOG_WARNING,
131 "child process %" APR_PID_T_FMT
132 " still did not exit, "
139 ap_log_error(APLOG_MARK, APLOG_ERR,
141 "child process %" APR_PID_T_FMT
142 " still did not exit, "
149 /* gave it our best shot, but alas... If this really
150 * is a child we are trying to kill and it really hasn't
151 * exited, we will likely fail to bind to the port
154 ap_log_error(APLOG_MARK, APLOG_ERR,
156 "could not make child process %" APR_PID_T_FMT
158 "attempting to continue anyway",
166 void ap_reclaim_child_processes(int terminate)
168 apr_time_t waittime = 1024 * 16;
170 extra_process_t *cur_extra;
173 apr_time_t starttime = apr_time_now();
174 /* this table of actions and elapsed times tells what action is taken
175 * at which elapsed time from starting the reclaim
179 apr_time_t action_time;
181 {DO_NOTHING, 0}, /* dummy entry for iterations where we reap
182 * children but take no action against
185 {SEND_SIGTERM, apr_time_from_sec(3)},
186 {SEND_SIGTERM, apr_time_from_sec(5)},
187 {SEND_SIGTERM, apr_time_from_sec(7)},
188 {SEND_SIGKILL, apr_time_from_sec(9)},
189 {GIVEUP, apr_time_from_sec(10)}
191 int cur_action; /* index of action we decided to take this
194 int next_action = 1; /* index of first real action */
196 ap_mpm_query(AP_MPMQ_MAX_DAEMON_USED, &max_daemons);
200 /* don't let waittime get longer than 1 second; otherwise, we don't
201 * react quickly to the last child exiting, and taking action can
204 waittime = waittime * 4;
205 if (waittime > apr_time_from_sec(1)) {
206 waittime = apr_time_from_sec(1);
209 /* see what action to take, if any */
210 if (action_table[next_action].action_time <= apr_time_now() - starttime) {
211 cur_action = next_action;
215 cur_action = 0; /* nothing to do */
218 /* now see who is done */
220 for (i = 0; i < max_daemons; ++i) {
221 process_score *ps = ap_get_scoreboard_process(i);
225 continue; /* not every scoreboard entry is in use */
228 if (reclaim_one_pid(pid, action_table[cur_action].action)) {
229 ap_mpm_note_child_killed(i);
238 extra_process_t *next = cur_extra->next;
240 if (reclaim_one_pid(cur_extra->pid, action_table[cur_action].action)) {
241 AP_DEBUG_ASSERT(1 == ap_unregister_extra_mpm_process(cur_extra->pid));
248 #if APR_HAS_OTHER_CHILD
249 apr_proc_other_child_refresh_all(APR_OC_REASON_RESTART);
252 } while (not_dead_yet > 0 &&
253 action_table[cur_action].action != GIVEUP);
256 void ap_relieve_child_processes(void)
259 extra_process_t *cur_extra;
262 ap_mpm_query(AP_MPMQ_MAX_DAEMON_USED, &max_daemons);
264 /* now see who is done */
265 for (i = 0; i < max_daemons; ++i) {
266 process_score *ps = ap_get_scoreboard_process(i);
270 continue; /* not every scoreboard entry is in use */
273 if (reclaim_one_pid(pid, DO_NOTHING)) {
274 ap_mpm_note_child_killed(i);
280 extra_process_t *next = cur_extra->next;
282 if (reclaim_one_pid(cur_extra->pid, DO_NOTHING)) {
283 AP_DEBUG_ASSERT(1 == ap_unregister_extra_mpm_process(cur_extra->pid));
289 /* Before sending the signal to the pid this function verifies that
290 * the pid is a member of the current process group; either using
291 * apr_proc_wait(), where waitpid() guarantees to fail for non-child
292 * processes; or by using getpgid() directly, if available. */
293 apr_status_t ap_mpm_safe_kill(pid_t pid, int sig)
301 /* Ensure pid sanity */
307 rv = apr_proc_wait(&proc, &status, &why, APR_NOWAIT);
308 if (rv == APR_CHILD_DONE) {
309 /* The child already died - log the termination status if
311 ap_process_child_status(&proc, why, status);
314 else if (rv != APR_CHILD_NOTDONE) {
315 /* The child is already dead and reaped, or was a bogus pid -
316 * log this either way. */
317 ap_log_error(APLOG_MARK, APLOG_NOTICE, rv, ap_server_conf,
318 "cannot send signal %d to pid %ld (non-child or "
319 "already dead)", sig, (long)pid);
325 /* Ensure pid sanity. */
332 /* Process already dead... */
336 if (pg != getpgrp()) {
337 ap_log_error(APLOG_MARK, APLOG_ALERT, 0, ap_server_conf,
338 "refusing to send signal %d to pid %ld outside "
339 "process group", sig, (long)pid);
344 return kill(pid, sig) ? errno : APR_SUCCESS;
348 int ap_process_child_status(apr_proc_t *pid, apr_exit_why_e why, int status)
353 /* Child died... if it died due to a fatal error,
354 * we should simply bail out. The caller needs to
355 * check for bad rc from us and exit, running any
356 * appropriate cleanups.
358 * If the child died due to a resource shortage,
359 * the parent should limit the rate of forking
361 if (APR_PROC_CHECK_EXIT(why)) {
362 if (status == APEXIT_CHILDSICK) {
366 if (status == APEXIT_CHILDFATAL) {
367 ap_log_error(APLOG_MARK, APLOG_ALERT,
369 "Child %" APR_PID_T_FMT
370 " returned a Fatal error... Apache is exiting!",
372 return APEXIT_CHILDFATAL;
378 if (APR_PROC_CHECK_SIGNALED(why)) {
379 sigdesc = apr_signal_description_get(signum);
384 case AP_SIG_GRACEFUL:
389 if (APR_PROC_CHECK_CORE_DUMP(why)) {
390 ap_log_error(APLOG_MARK, APLOG_NOTICE,
392 "child pid %ld exit signal %s (%d), "
393 "possible coredump in %s",
394 (long)pid->pid, sigdesc, signum,
398 ap_log_error(APLOG_MARK, APLOG_NOTICE,
400 "child pid %ld exit signal %s (%d)",
401 (long)pid->pid, sigdesc, signum);
408 AP_DECLARE(apr_status_t) ap_mpm_pod_open(apr_pool_t *p, ap_pod_t **pod)
412 *pod = apr_palloc(p, sizeof(**pod));
413 rv = apr_file_pipe_create_ex(&((*pod)->pod_in), &((*pod)->pod_out),
415 if (rv != APR_SUCCESS) {
419 apr_file_pipe_timeout_set((*pod)->pod_in, 0);
422 /* close these before exec. */
423 apr_file_inherit_unset((*pod)->pod_in);
424 apr_file_inherit_unset((*pod)->pod_out);
429 AP_DECLARE(apr_status_t) ap_mpm_pod_check(ap_pod_t *pod)
435 rv = apr_file_read(pod->pod_in, &c, &len);
437 if ((rv == APR_SUCCESS) && (len == 1)) {
441 if (rv != APR_SUCCESS) {
448 AP_DECLARE(apr_status_t) ap_mpm_pod_close(ap_pod_t *pod)
452 rv = apr_file_close(pod->pod_out);
453 if (rv != APR_SUCCESS) {
457 rv = apr_file_close(pod->pod_in);
458 if (rv != APR_SUCCESS) {
465 static apr_status_t pod_signal_internal(ap_pod_t *pod)
468 char char_of_death = '!';
471 rv = apr_file_write(pod->pod_out, &char_of_death, &one);
472 if (rv != APR_SUCCESS) {
473 ap_log_error(APLOG_MARK, APLOG_WARNING, rv, ap_server_conf,
474 "write pipe_of_death");
480 /* This function connects to the server, then immediately closes the connection.
481 * This permits the MPM to skip the poll when there is only one listening
482 * socket, because it provides a alternate way to unblock an accept() when
485 static apr_status_t dummy_connection(ap_pod_t *pod)
494 /* create a temporary pool for the socket. pconf stays around too long */
495 rv = apr_pool_create(&p, pod->p);
496 if (rv != APR_SUCCESS) {
500 /* If possible, find a listener which is configured for
501 * plain-HTTP, not SSL; using an SSL port would either be
502 * expensive to do correctly (performing a complete SSL handshake)
503 * or cause log spam by doing incorrectly (simply sending EOF). */
505 while (lp && lp->protocol && strcasecmp(lp->protocol, "http") != 0) {
512 rv = apr_socket_create(&sock, lp->bind_addr->family, SOCK_STREAM, 0, p);
513 if (rv != APR_SUCCESS) {
514 ap_log_error(APLOG_MARK, APLOG_WARNING, rv, ap_server_conf,
515 "get socket to connect to listener");
520 /* on some platforms (e.g., FreeBSD), the kernel won't accept many
521 * queued connections before it starts blocking local connects...
522 * we need to keep from blocking too long and instead return an error,
523 * because the MPM won't want to hold up a graceful restart for a
526 rv = apr_socket_timeout_set(sock, apr_time_from_sec(3));
527 if (rv != APR_SUCCESS) {
528 ap_log_error(APLOG_MARK, APLOG_WARNING, rv, ap_server_conf,
529 "set timeout on socket to connect to listener");
530 apr_socket_close(sock);
535 rv = apr_socket_connect(sock, lp->bind_addr);
536 if (rv != APR_SUCCESS) {
537 int log_level = APLOG_WARNING;
539 if (APR_STATUS_IS_TIMEUP(rv)) {
540 /* probably some server processes bailed out already and there
541 * is nobody around to call accept and clear out the kernel
542 * connection queue; usually this is not worth logging
544 log_level = APLOG_DEBUG;
547 ap_log_error(APLOG_MARK, log_level, rv, ap_server_conf,
548 "connect to listener on %pI", lp->bind_addr);
551 /* Create the request string. We include a User-Agent so that
552 * adminstrators can track down the cause of the odd-looking
553 * requests in their logs.
555 srequest = apr_pstrcat(p, "OPTIONS * HTTP/1.0\r\nUser-Agent: ",
556 ap_get_server_description(),
557 " (internal dummy connection)\r\n\r\n", NULL);
559 /* Since some operating systems support buffering of data or entire
560 * requests in the kernel, we send a simple request, to make sure
561 * the server pops out of a blocking accept().
563 /* XXX: This is HTTP specific. We should look at the Protocol for each
564 * listener, and send the correct type of request to trigger any Accept
567 len = strlen(srequest);
568 apr_socket_send(sock, srequest, &len);
569 apr_socket_close(sock);
575 AP_DECLARE(apr_status_t) ap_mpm_pod_signal(ap_pod_t *pod)
579 rv = pod_signal_internal(pod);
580 if (rv != APR_SUCCESS) {
584 return dummy_connection(pod);
587 void ap_mpm_pod_killpg(ap_pod_t *pod, int num)
590 apr_status_t rv = APR_SUCCESS;
592 /* we don't write anything to the pod here... we assume
593 * that the would-be reader of the pod has another way to
594 * see that it is time to die once we wake it up
596 * writing lots of things to the pod at once is very
597 * problematic... we can fill the kernel pipe buffer and
598 * be blocked until somebody consumes some bytes or
599 * we hit a timeout... if we hit a timeout we can't just
600 * keep trying because maybe we'll never successfully
601 * write again... but then maybe we'll leave would-be
602 * readers stranded (a number of them could be tied up for
603 * a while serving time-consuming requests)
605 for (i = 0; i < num && rv == APR_SUCCESS; i++) {
606 rv = dummy_connection(pod);
610 static const char *dash_k_arg = NULL;
612 static int send_signal(pid_t pid, int sig)
614 if (kill(pid, sig) < 0) {
615 ap_log_error(APLOG_MARK, APLOG_STARTUP, errno, NULL,
616 "sending signal to server");
622 int ap_signal_server(int *exit_status, apr_pool_t *pconf)
627 int have_pid_file = 0;
632 rv = ap_read_pid(pconf, ap_pid_fname, &otherpid);
633 if (rv != APR_SUCCESS) {
634 if (rv != APR_ENOENT) {
635 ap_log_error(APLOG_MARK, APLOG_STARTUP, rv, NULL,
636 "Error retrieving pid file %s", ap_pid_fname);
637 ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL,
638 "Remove it before continuing if it is corrupted.");
642 status = "httpd (no pid file) not running";
646 if (kill(otherpid, 0) == 0) {
648 status = apr_psprintf(pconf,
649 "httpd (pid %" APR_PID_T_FMT ") already "
650 "running", otherpid);
653 status = apr_psprintf(pconf,
654 "httpd (pid %" APR_PID_T_FMT "?) not running",
659 if (!strcmp(dash_k_arg, "start")) {
661 printf("%s\n", status);
666 if (!strcmp(dash_k_arg, "stop")) {
668 printf("%s\n", status);
671 send_signal(otherpid, SIGTERM);
676 if (!strcmp(dash_k_arg, "restart")) {
678 printf("httpd not running, trying to start\n");
681 *exit_status = send_signal(otherpid, SIGHUP);
686 if (!strcmp(dash_k_arg, "graceful")) {
688 printf("httpd not running, trying to start\n");
691 *exit_status = send_signal(otherpid, AP_SIG_GRACEFUL);
696 if (!strcmp(dash_k_arg, "graceful-stop")) {
698 printf("%s\n", status);
701 *exit_status = send_signal(otherpid, AP_SIG_GRACEFUL_STOP);
709 void ap_mpm_rewrite_args(process_rec *process)
711 apr_array_header_t *mpm_new_argv;
718 mpm_new_argv = apr_array_make(process->pool, process->argc,
719 sizeof(const char **));
720 *(const char **)apr_array_push(mpm_new_argv) = process->argv[0];
721 fixed_args = mpm_new_argv->nelts;
722 apr_getopt_init(&opt, process->pool, process->argc, process->argv);
725 /* option char returned by apr_getopt() will be stored in optbuf[1] */
727 while ((rv = apr_getopt(opt, "k:" AP_SERVER_BASEARGS,
728 optbuf + 1, &optarg)) == APR_SUCCESS) {
732 if (!strcmp(optarg, "start") || !strcmp(optarg, "stop") ||
733 !strcmp(optarg, "restart") || !strcmp(optarg, "graceful") ||
734 !strcmp(optarg, "graceful-stop")) {
740 *(const char **)apr_array_push(mpm_new_argv) =
741 apr_pstrdup(process->pool, optbuf);
743 *(const char **)apr_array_push(mpm_new_argv) = optarg;
748 /* back up to capture the bad argument */
749 if (rv == APR_BADCH || rv == APR_BADARG) {
753 while (opt->ind < opt->argc) {
754 *(const char **)apr_array_push(mpm_new_argv) =
755 apr_pstrdup(process->pool, opt->argv[opt->ind++]);
758 process->argc = mpm_new_argv->nelts;
759 process->argv = (const char * const *)mpm_new_argv->elts;
762 APR_REGISTER_OPTIONAL_FN(ap_signal_server);
766 static pid_t parent_pid, my_pid;
767 static apr_pool_t *pconf;
769 #if AP_ENABLE_EXCEPTION_HOOK
771 static int exception_hook_enabled;
773 const char *ap_mpm_set_exception_hook(cmd_parms *cmd, void *dummy,
776 const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY);
781 if (cmd->server->is_virtual) {
782 return "EnableExceptionHook directive not allowed in <VirtualHost>";
785 if (strcasecmp(arg, "on") == 0) {
786 exception_hook_enabled = 1;
788 else if (strcasecmp(arg, "off") == 0) {
789 exception_hook_enabled = 0;
792 return "parameter must be 'on' or 'off'";
798 static void run_fatal_exception_hook(int sig)
800 ap_exception_info_t ei = {0};
802 if (exception_hook_enabled &&
804 my_pid != parent_pid) {
807 ap_run_fatal_exception(&ei);
810 #endif /* AP_ENABLE_EXCEPTION_HOOK */
812 /* handle all varieties of core dumping signals */
813 static void sig_coredump(int sig)
815 apr_filepath_set(ap_coredump_dir, pconf);
816 apr_signal(sig, SIG_DFL);
817 #if AP_ENABLE_EXCEPTION_HOOK
818 run_fatal_exception_hook(sig);
820 /* linuxthreads issue calling getpid() here:
821 * This comparison won't match if the crashing thread is
822 * some module's thread that runs in the parent process.
823 * The fallout, which is limited to linuxthreads:
824 * The special log message won't be written when such a
825 * thread in the parent causes the parent to crash.
827 if (getpid() == parent_pid) {
828 ap_log_error(APLOG_MARK, APLOG_NOTICE,
830 "seg fault or similar nasty error detected "
831 "in the parent process");
832 /* XXX we can probably add some rudimentary cleanup code here,
833 * like getting rid of the pid file. If any additional bad stuff
834 * happens, we are protected from recursive errors taking down the
835 * system since this function is no longer the signal handler GLA
839 /* At this point we've got sig blocked, because we're still inside
840 * the signal handler. When we leave the signal handler it will
841 * be unblocked, and we'll take the signal... and coredump or whatever
842 * is appropriate for this particular Unix. In addition the parent
843 * will see the real signal we received -- whereas if we called
844 * abort() here, the parent would only see SIGABRT.
848 apr_status_t ap_fatal_signal_child_setup(server_rec *s)
854 apr_status_t ap_fatal_signal_setup(server_rec *s, apr_pool_t *in_pconf)
856 #ifndef NO_USE_SIGACTION
859 sigemptyset(&sa.sa_mask);
861 #if defined(SA_ONESHOT)
862 sa.sa_flags = SA_ONESHOT;
863 #elif defined(SA_RESETHAND)
864 sa.sa_flags = SA_RESETHAND;
869 sa.sa_handler = sig_coredump;
870 if (sigaction(SIGSEGV, &sa, NULL) < 0)
871 ap_log_error(APLOG_MARK, APLOG_WARNING, errno, s, "sigaction(SIGSEGV)");
873 if (sigaction(SIGBUS, &sa, NULL) < 0)
874 ap_log_error(APLOG_MARK, APLOG_WARNING, errno, s, "sigaction(SIGBUS)");
877 if (sigaction(SIGABORT, &sa, NULL) < 0)
878 ap_log_error(APLOG_MARK, APLOG_WARNING, errno, s, "sigaction(SIGABORT)");
881 if (sigaction(SIGABRT, &sa, NULL) < 0)
882 ap_log_error(APLOG_MARK, APLOG_WARNING, errno, s, "sigaction(SIGABRT)");
885 if (sigaction(SIGILL, &sa, NULL) < 0)
886 ap_log_error(APLOG_MARK, APLOG_WARNING, errno, s, "sigaction(SIGILL)");
889 if (sigaction(SIGFPE, &sa, NULL) < 0)
890 ap_log_error(APLOG_MARK, APLOG_WARNING, errno, s, "sigaction(SIGFPE)");
893 #else /* NO_USE_SIGACTION */
895 apr_signal(SIGSEGV, sig_coredump);
897 apr_signal(SIGBUS, sig_coredump);
900 apr_signal(SIGABORT, sig_coredump);
901 #endif /* SIGABORT */
903 apr_signal(SIGABRT, sig_coredump);
906 apr_signal(SIGILL, sig_coredump);
909 apr_signal(SIGFPE, sig_coredump);
912 #endif /* NO_USE_SIGACTION */
915 parent_pid = my_pid = getpid();