diff options
author | hongbotian <hongbo.tianhongbo@huawei.com> | 2015-11-30 01:45:08 -0500 |
---|---|---|
committer | hongbotian <hongbo.tianhongbo@huawei.com> | 2015-11-30 01:45:08 -0500 |
commit | e8ec7aa8e38a93f5b034ac74cebce5de23710317 (patch) | |
tree | aa031937bf856c1f8d6ad7877b8d2cb0224da5ef /rubbos/app/httpd-2.0.64/server/mpm/experimental | |
parent | cc40af334e619bb549038238507407866f774f8f (diff) |
upload http
JIRA: BOTTLENECK-10
Change-Id: I7598427ff904df438ce77c2819ee48ac75ffa8da
Signed-off-by: hongbotian <hongbo.tianhongbo@huawei.com>
Diffstat (limited to 'rubbos/app/httpd-2.0.64/server/mpm/experimental')
19 files changed, 6837 insertions, 0 deletions
diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/Makefile.in b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/Makefile.in new file mode 100644 index 00000000..03f1765e --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/Makefile.in @@ -0,0 +1,5 @@ + +LTLIBRARY_NAME = libleader.la +LTLIBRARY_SOURCES = leader.c + +include $(top_srcdir)/build/ltlib.mk diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/README b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/README new file mode 100644 index 00000000..1981a5be --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/README @@ -0,0 +1,15 @@ +Leader MPM: +This is an experimental variant of the standard worker MPM. +It uses a Leader/Followers design pattern to coordinate work among threads: +http://deuce.doc.wustl.edu/doc/pspdfs/lf.pdf + +To use the leader MPM, add "--with-mpm=leader" to the configure +script's arguments when building the httpd. + +This MPM depends on APR's atomic compare-and-swap operations for +thread synchronization. If you are compiling for an x86 target +and you don't need to support 386s, or you're compiling for a +SPARC and you don't need to run on pre-UltraSPARC chips, add +"--enable-nonportable-atomics=yes" to the configure script's +arguments. This will cause APR to implement atomic operations +using efficient opcodes not available in older CPUs. diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/config5.m4 b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/config5.m4 new file mode 100644 index 00000000..9a915abf --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/config5.m4 @@ -0,0 +1,6 @@ +dnl ## XXX - Need a more thorough check of the proper flags to use + +if test "$MPM_NAME" = "leader" ; then + AC_CHECK_FUNCS(pthread_kill) + APACHE_FAST_OUTPUT(server/mpm/$MPM_SUBDIR_NAME/Makefile) +fi diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/leader.c b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/leader.c new file mode 100644 index 00000000..322c2286 --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/leader.c @@ -0,0 +1,1976 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "apr.h" +#include "apr_portable.h" +#include "apr_strings.h" +#include "apr_file_io.h" +#include "apr_thread_proc.h" +#include "apr_signal.h" +#include "apr_thread_cond.h" +#include "apr_thread_mutex.h" +#include "apr_proc_mutex.h" +#define APR_WANT_STRFUNC +#include "apr_want.h" + +#if APR_HAVE_UNISTD_H +#include <unistd.h> +#endif +#if APR_HAVE_SYS_SOCKET_H +#include <sys/socket.h> +#endif +#if APR_HAVE_SYS_WAIT_H +#include <sys/wait.h> +#endif +#ifdef HAVE_SYS_PROCESSOR_H +#include <sys/processor.h> /* for bindprocessor() */ +#endif + +#if !APR_HAS_THREADS +#error The Leader/Follower MPM requires APR threads, but they are unavailable. +#endif + +#define CORE_PRIVATE + +#include "ap_config.h" +#include "httpd.h" +#include "http_main.h" +#include "http_log.h" +#include "http_config.h" /* for read_config */ +#include "http_core.h" /* for get_remote_host */ +#include "http_connection.h" +#include "ap_mpm.h" +#include "mpm_common.h" +#include "ap_listen.h" +#include "scoreboard.h" +#include "mpm_default.h" +#include "apr_poll.h" + +#include <signal.h> +#include <limits.h> /* for INT_MAX */ + +#include "apr_atomic.h" + +/* Limit on the total --- clients will be locked out if more servers than + * this are needed. It is intended solely to keep the server from crashing + * when things get out of hand. + * + * We keep a hard maximum number of servers, for two reasons --- first off, + * in case something goes seriously wrong, we want to stop the fork bomb + * short of actually crashing the machine we're running on by filling some + * kernel table. Secondly, it keeps the size of the scoreboard file small + * enough that we can read the whole thing without worrying too much about + * the overhead. + */ +#ifndef DEFAULT_SERVER_LIMIT +#define DEFAULT_SERVER_LIMIT 16 +#endif + +/* Admin can't tune ServerLimit beyond MAX_SERVER_LIMIT. We want + * some sort of compile-time limit to help catch typos. + */ +#ifndef MAX_SERVER_LIMIT +#define MAX_SERVER_LIMIT 20000 +#endif + +/* Limit on the threads per process. Clients will be locked out if more than + * this * server_limit are needed. + * + * We keep this for one reason it keeps the size of the scoreboard file small + * enough that we can read the whole thing without worrying too much about + * the overhead. + */ +#ifndef DEFAULT_THREAD_LIMIT +#define DEFAULT_THREAD_LIMIT 64 +#endif + +/* Admin can't tune ThreadLimit beyond MAX_THREAD_LIMIT. We want + * some sort of compile-time limit to help catch typos. + */ +#ifndef MAX_THREAD_LIMIT +#define MAX_THREAD_LIMIT 20000 +#endif + +/* + * Actual definitions of config globals + */ + +int ap_threads_per_child = 0; /* Worker threads per child */ +static int ap_daemons_to_start = 0; +static int min_spare_threads = 0; +static int max_spare_threads = 0; +static int ap_daemons_limit = 0; +static int server_limit = DEFAULT_SERVER_LIMIT; +static int first_server_limit; +static int thread_limit = DEFAULT_THREAD_LIMIT; +static int first_thread_limit; +static int changed_limit_at_restart; +static int dying = 0; +static int workers_may_exit = 0; +static int start_thread_may_exit = 0; +static int requests_this_child; +static int num_listensocks = 0; +static int resource_shortage = 0; +static int mpm_state = AP_MPMQ_STARTING; + +typedef struct worker_wakeup_info worker_wakeup_info; + +/* The structure used to pass unique initialization info to each thread */ +typedef struct { + int pid; + int tid; + int sd; +} proc_info; + + +/* Structure used to pass information to the thread responsible for + * creating the rest of the threads. + */ +typedef struct { + apr_thread_t **threads; + int child_num_arg; + apr_threadattr_t *threadattr; +} thread_starter; + +#define ID_FROM_CHILD_THREAD(c, t) ((c * thread_limit) + t) + +/* + * The max child slot ever assigned, preserved across restarts. Necessary + * to deal with MaxClients changes across AP_SIG_GRACEFUL restarts. We + * use this value to optimize routines that have to scan the entire + * scoreboard. + */ +int ap_max_daemons_limit = -1; + +static ap_pod_t *pod; + +/* *Non*-shared http_main globals... */ + +server_rec *ap_server_conf; + +/* This MPM respects a couple of runtime flags that can aid in debugging. + * Setting the -DNO_DETACH flag will prevent the root process from + * detaching from its controlling terminal. Additionally, setting + * the -DONE_PROCESS flag (which implies -DNO_DETACH) will get you the + * child_main loop running in the process which originally started up. + * This gives you a pretty nice debugging environment. (You'll get a SIGHUP + * early in standalone_main; just continue through. This is the server + * trying to kill off any child processes which it might have lying + * around --- Apache doesn't keep track of their pids, it just sends + * SIGHUP to the process group, ignoring it in the root process. + * Continue through and you'll be fine.). + */ + +static int one_process = 0; + +#ifdef DEBUG_SIGSTOP +int raise_sigstop_flags; +#endif + +static apr_pool_t *pconf; /* Pool for config stuff */ +static apr_pool_t *pchild; /* Pool for httpd child stuff */ + +static pid_t ap_my_pid; /* Linux getpid() doesn't work except in main + thread. Use this instead */ +static pid_t parent_pid; + +/* Locks for accept serialization */ +static apr_proc_mutex_t *accept_mutex; + +#ifdef SINGLE_LISTEN_UNSERIALIZED_ACCEPT +#define SAFE_ACCEPT(stmt) (ap_listeners->next ? (stmt) : APR_SUCCESS) +#else +#define SAFE_ACCEPT(stmt) (stmt) +#endif + + +/* Structure used to wake up an idle worker thread + */ +struct worker_wakeup_info { + apr_uint32_t next; /* index into worker_wakeups array, + * used to build a linked list + */ + apr_thread_cond_t *cond; + apr_thread_mutex_t *mutex; +}; + +static worker_wakeup_info *worker_wakeup_create(apr_pool_t *pool) +{ + apr_status_t rv; + worker_wakeup_info *wakeup; + + wakeup = (worker_wakeup_info *)apr_palloc(pool, sizeof(*wakeup)); + if ((rv = apr_thread_cond_create(&wakeup->cond, pool)) != APR_SUCCESS) { + return NULL; + } + if ((rv = apr_thread_mutex_create(&wakeup->mutex, APR_THREAD_MUTEX_DEFAULT, + pool)) != APR_SUCCESS) { + return NULL; + } + /* The wakeup's mutex will be unlocked automatically when + * the worker blocks on the condition variable + */ + apr_thread_mutex_lock(wakeup->mutex); + return wakeup; +} + + +/* Structure used to hold a stack of idle worker threads + */ +typedef struct { + /* 'state' consists of several fields concatenated into a + * single 32-bit int for use with the apr_atomic_cas() API: + * state & STACK_FIRST is the thread ID of the first thread + * in a linked list of idle threads + * state & STACK_TERMINATED indicates whether the proc is shutting down + * state & STACK_NO_LISTENER indicates whether the process has + * no current listener thread + */ + apr_uint32_t state; +} worker_stack; + +#define STACK_FIRST 0xffff +#define STACK_LIST_END 0xffff +#define STACK_TERMINATED 0x10000 +#define STACK_NO_LISTENER 0x20000 + +static worker_wakeup_info **worker_wakeups = NULL; + +static worker_stack* worker_stack_create(apr_pool_t *pool, apr_size_t max) +{ + worker_stack *stack = (worker_stack *)apr_palloc(pool, sizeof(*stack)); + stack->state = STACK_NO_LISTENER | STACK_LIST_END; + return stack; +} + +static apr_status_t worker_stack_wait(worker_stack *stack, + apr_uint32_t worker_id) +{ + worker_wakeup_info *wakeup = worker_wakeups[worker_id]; + + while (1) { + apr_uint32_t state = stack->state; + if (state & (STACK_TERMINATED | STACK_NO_LISTENER)) { + if (state & STACK_TERMINATED) { + return APR_EINVAL; + } + if (apr_atomic_cas(&(stack->state), STACK_LIST_END, state) != + state) { + continue; + } + else { + return APR_SUCCESS; + } + } + wakeup->next = state; + if (apr_atomic_cas(&(stack->state), worker_id, state) != state) { + continue; + } + else { + return apr_thread_cond_wait(wakeup->cond, wakeup->mutex); + } + } +} + +static apr_status_t worker_stack_awaken_next(worker_stack *stack) +{ + + while (1) { + apr_uint32_t state = stack->state; + apr_uint32_t first = state & STACK_FIRST; + if (first == STACK_LIST_END) { + if (apr_atomic_cas(&(stack->state), state | STACK_NO_LISTENER, + state) != state) { + continue; + } + else { + return APR_SUCCESS; + } + } + else { + worker_wakeup_info *wakeup = worker_wakeups[first]; + if (apr_atomic_cas(&(stack->state), (state ^ first) | wakeup->next, + state) != state) { + continue; + } + else { + /* Acquire and release the idle worker's mutex to ensure + * that it's actually waiting on its condition variable + */ + apr_status_t rv; + if ((rv = apr_thread_mutex_lock(wakeup->mutex)) != + APR_SUCCESS) { + return rv; + } + if ((rv = apr_thread_mutex_unlock(wakeup->mutex)) != + APR_SUCCESS) { + return rv; + } + return apr_thread_cond_signal(wakeup->cond); + } + } + } +} + +static apr_status_t worker_stack_term(worker_stack *stack) +{ + int i; + apr_status_t rv; + + while (1) { + apr_uint32_t state = stack->state; + if (apr_atomic_cas(&(stack->state), state | STACK_TERMINATED, + state) == state) { + break; + } + } + for (i = 0; i < ap_threads_per_child; i++) { + if ((rv = worker_stack_awaken_next(stack)) != APR_SUCCESS) { + return rv; + } + } + return APR_SUCCESS; +} + +static worker_stack *idle_worker_stack; + +#define ST_INIT 0 +#define ST_GRACEFUL 1 +#define ST_UNGRACEFUL 2 + +static int terminate_mode = ST_INIT; + +static void signal_threads(int mode) +{ + if (terminate_mode == mode) { + return; + } + terminate_mode = mode; + mpm_state = AP_MPMQ_STOPPING; + workers_may_exit = 1; + + worker_stack_term(idle_worker_stack); +} + +AP_DECLARE(apr_status_t) ap_mpm_query(int query_code, int *result) +{ + switch(query_code){ + case AP_MPMQ_MAX_DAEMON_USED: + *result = ap_max_daemons_limit; + return APR_SUCCESS; + case AP_MPMQ_IS_THREADED: + *result = AP_MPMQ_STATIC; + return APR_SUCCESS; + case AP_MPMQ_IS_FORKED: + *result = AP_MPMQ_DYNAMIC; + return APR_SUCCESS; + case AP_MPMQ_HARD_LIMIT_DAEMONS: + *result = server_limit; + return APR_SUCCESS; + case AP_MPMQ_HARD_LIMIT_THREADS: + *result = thread_limit; + return APR_SUCCESS; + case AP_MPMQ_MAX_THREADS: + *result = ap_threads_per_child; + return APR_SUCCESS; + case AP_MPMQ_MIN_SPARE_DAEMONS: + *result = 0; + return APR_SUCCESS; + case AP_MPMQ_MIN_SPARE_THREADS: + *result = min_spare_threads; + return APR_SUCCESS; + case AP_MPMQ_MAX_SPARE_DAEMONS: + *result = 0; + return APR_SUCCESS; + case AP_MPMQ_MAX_SPARE_THREADS: + *result = max_spare_threads; + return APR_SUCCESS; + case AP_MPMQ_MAX_REQUESTS_DAEMON: + *result = ap_max_requests_per_child; + return APR_SUCCESS; + case AP_MPMQ_MAX_DAEMONS: + *result = ap_daemons_limit; + return APR_SUCCESS; + case AP_MPMQ_MPM_STATE: + *result = mpm_state; + return APR_SUCCESS; + } + return APR_ENOTIMPL; +} + +/* a clean exit from a child with proper cleanup */ +static void clean_child_exit(int code) __attribute__ ((noreturn)); +static void clean_child_exit(int code) +{ + mpm_state = AP_MPMQ_STOPPING; + if (pchild) { + apr_pool_destroy(pchild); + } + ap_mpm_pod_close(pod); + exit(code); +} + +static void just_die(int sig) +{ + clean_child_exit(0); +} + +/***************************************************************** + * Connection structures and accounting... + */ + +/* volatile just in case */ +static int volatile shutdown_pending; +static int volatile restart_pending; +static int volatile is_graceful; +static volatile int child_fatal; +ap_generation_t volatile ap_my_generation; + +/* + * ap_start_shutdown() and ap_start_restart(), below, are a first stab at + * functions to initiate shutdown or restart without relying on signals. + * Previously this was initiated in sig_term() and restart() signal handlers, + * but we want to be able to start a shutdown/restart from other sources -- + * e.g. on Win32, from the service manager. Now the service manager can + * call ap_start_shutdown() or ap_start_restart() as appropiate. Note that + * these functions can also be called by the child processes, since global + * variables are no longer used to pass on the required action to the parent. + * + * These should only be called from the parent process itself, since the + * parent process will use the shutdown_pending and restart_pending variables + * to determine whether to shutdown or restart. The child process should + * call signal_parent() directly to tell the parent to die -- this will + * cause neither of those variable to be set, which the parent will + * assume means something serious is wrong (which it will be, for the + * child to force an exit) and so do an exit anyway. + */ + +static void ap_start_shutdown(void) +{ + mpm_state = AP_MPMQ_STOPPING; + if (shutdown_pending == 1) { + /* Um, is this _probably_ not an error, if the user has + * tried to do a shutdown twice quickly, so we won't + * worry about reporting it. + */ + return; + } + shutdown_pending = 1; +} + +/* do a graceful restart if graceful == 1 */ +static void ap_start_restart(int graceful) +{ + mpm_state = AP_MPMQ_STOPPING; + if (restart_pending == 1) { + /* Probably not an error - don't bother reporting it */ + return; + } + restart_pending = 1; + is_graceful = graceful; +} + +static void sig_term(int sig) +{ + if (ap_my_pid == parent_pid) { + ap_start_shutdown(); + } + else { + signal_threads(ST_GRACEFUL); + } +} + +static void restart(int sig) +{ + ap_start_restart(sig == AP_SIG_GRACEFUL); +} + +static void set_signals(void) +{ +#ifndef NO_USE_SIGACTION + struct sigaction sa; +#endif + + if (!one_process) { + ap_fatal_signal_setup(ap_server_conf, pconf); + } + +#ifndef NO_USE_SIGACTION + sigemptyset(&sa.sa_mask); + sa.sa_flags = 0; + + sa.sa_handler = sig_term; + if (sigaction(SIGTERM, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGTERM)"); +#ifdef SIGINT + if (sigaction(SIGINT, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGINT)"); +#endif +#ifdef SIGXCPU + sa.sa_handler = SIG_DFL; + if (sigaction(SIGXCPU, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGXCPU)"); +#endif +#ifdef SIGXFSZ + sa.sa_handler = SIG_DFL; + if (sigaction(SIGXFSZ, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGXFSZ)"); +#endif +#ifdef SIGPIPE + sa.sa_handler = SIG_IGN; + if (sigaction(SIGPIPE, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGPIPE)"); +#endif + + /* we want to ignore HUPs and AP_SIG_GRACEFUL while we're busy + * processing one */ + sigaddset(&sa.sa_mask, SIGHUP); + sigaddset(&sa.sa_mask, AP_SIG_GRACEFUL); + sa.sa_handler = restart; + if (sigaction(SIGHUP, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGHUP)"); + if (sigaction(AP_SIG_GRACEFUL, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(" AP_SIG_GRACEFUL_STRING ")"); +#else + if (!one_process) { +#ifdef SIGXCPU + apr_signal(SIGXCPU, SIG_DFL); +#endif /* SIGXCPU */ +#ifdef SIGXFSZ + apr_signal(SIGXFSZ, SIG_DFL); +#endif /* SIGXFSZ */ + } + + apr_signal(SIGTERM, sig_term); +#ifdef SIGHUP + apr_signal(SIGHUP, restart); +#endif /* SIGHUP */ +#ifdef AP_SIG_GRACEFUL + apr_signal(AP_SIG_GRACEFUL, restart); +#endif /* AP_SIG_GRACEFUL */ +#ifdef SIGPIPE + apr_signal(SIGPIPE, SIG_IGN); +#endif /* SIGPIPE */ + +#endif +} + +/***************************************************************** + * Here follows a long bunch of generic server bookkeeping stuff... + */ + +int ap_graceful_stop_signalled(void) + /* XXX this is really a bad confusing obsolete name + * maybe it should be ap_mpm_process_exiting? + */ +{ + return workers_may_exit; +} + +/***************************************************************** + * Child process main loop. + */ + +static void process_socket(apr_pool_t *p, apr_socket_t *sock, int my_child_num, + int my_thread_num, apr_bucket_alloc_t *bucket_alloc) +{ + conn_rec *current_conn; + long conn_id = ID_FROM_CHILD_THREAD(my_child_num, my_thread_num); + int csd; + ap_sb_handle_t *sbh; + + ap_create_sb_handle(&sbh, p, my_child_num, my_thread_num); + apr_os_sock_get(&csd, sock); + + current_conn = ap_run_create_connection(p, ap_server_conf, sock, + conn_id, sbh, bucket_alloc); + if (current_conn) { + ap_process_connection(current_conn, sock); + ap_lingering_close(current_conn); + } +} + +/* requests_this_child has gone to zero or below. See if the admin coded + "MaxRequestsPerChild 0", and keep going in that case. Doing it this way + simplifies the hot path in worker_thread */ +static void check_infinite_requests(void) +{ + if (ap_max_requests_per_child) { + signal_threads(ST_GRACEFUL); + } + else { + /* wow! if you're executing this code, you may have set a record. + * either this child process has served over 2 billion requests, or + * you're running a threaded 2.0 on a 16 bit machine. + * + * I'll buy pizza and beers at Apachecon for the first person to do + * the former without cheating (dorking with INT_MAX, or running with + * uncommitted performance patches, for example). + * + * for the latter case, you probably deserve a beer too. Greg Ames + */ + + requests_this_child = INT_MAX; /* keep going */ + } +} + +static void unblock_signal(int sig) +{ + sigset_t sig_mask; + + sigemptyset(&sig_mask); + sigaddset(&sig_mask, sig); +#if defined(SIGPROCMASK_SETS_THREAD_MASK) + sigprocmask(SIG_UNBLOCK, &sig_mask, NULL); +#else + pthread_sigmask(SIG_UNBLOCK, &sig_mask, NULL); +#endif +} + +static void *worker_thread(apr_thread_t *thd, void * dummy) +{ + proc_info * ti = dummy; + int process_slot = ti->pid; + int thread_slot = ti->tid; + apr_uint32_t my_worker_num = (apr_uint32_t)(ti->tid); + apr_pool_t *tpool = apr_thread_pool_get(thd); + void *csd = NULL; + apr_allocator_t *allocator; + apr_pool_t *ptrans; /* Pool for per-transaction stuff */ + apr_bucket_alloc_t *bucket_alloc; + int n; + apr_pollfd_t *pollset; + apr_status_t rv; + ap_listen_rec *lr, *last_lr = ap_listeners; + int is_listener; + + ap_update_child_status_from_indexes(process_slot, thread_slot, SERVER_STARTING, NULL); + + free(ti); + + apr_allocator_create(&allocator); + apr_allocator_max_free_set(allocator, ap_max_mem_free); + /* XXX: why is ptrans's parent not tpool? --jcw 08/2003 */ + apr_pool_create_ex(&ptrans, NULL, NULL, allocator); + apr_allocator_owner_set(allocator, ptrans); + bucket_alloc = apr_bucket_alloc_create_ex(allocator); + + apr_poll_setup(&pollset, num_listensocks, tpool); + for(lr = ap_listeners ; lr != NULL ; lr = lr->next) + apr_poll_socket_add(pollset, lr->sd, APR_POLLIN); + + /* TODO: Switch to a system where threads reuse the results from earlier + poll calls - manoj */ + is_listener = 0; + while (!workers_may_exit) { + + ap_update_child_status_from_indexes(process_slot, thread_slot, + SERVER_READY, NULL); + if (!is_listener) { + /* Wait until it's our turn to become the listener */ + if ((rv = worker_stack_wait(idle_worker_stack, my_worker_num)) != + APR_SUCCESS) { + if (rv != APR_EINVAL) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "worker_stack_wait failed. Shutting down"); + } + break; + } + if (workers_may_exit) { + break; + } + is_listener = 1; + } + + /* TODO: requests_this_child should be synchronized - aaron */ + if (requests_this_child <= 0) { + check_infinite_requests(); + } + if (workers_may_exit) break; + + if ((rv = SAFE_ACCEPT(apr_proc_mutex_lock(accept_mutex))) + != APR_SUCCESS) { + int level = APLOG_EMERG; + + if (workers_may_exit) { + break; + } + if (ap_scoreboard_image->parent[process_slot].generation != + ap_scoreboard_image->global->running_generation) { + level = APLOG_DEBUG; /* common to get these at restart time */ + } + ap_log_error(APLOG_MARK, level, rv, ap_server_conf, + "apr_proc_mutex_lock failed. Attempting to shutdown " + "process gracefully."); + signal_threads(ST_GRACEFUL); + break; /* skip the lock release */ + } + + if (!ap_listeners->next) { + /* Only one listener, so skip the poll */ + lr = ap_listeners; + } + else { + while (!workers_may_exit) { + apr_status_t ret; + apr_int16_t event; + + ret = apr_poll(pollset, num_listensocks, &n, -1); + if (ret != APR_SUCCESS) { + if (APR_STATUS_IS_EINTR(ret)) { + continue; + } + + /* apr_poll() will only return errors in catastrophic + * circumstances. Let's try exiting gracefully, for now. */ + ap_log_error(APLOG_MARK, APLOG_ERR, ret, (const server_rec *) + ap_server_conf, "apr_poll: (listen)"); + signal_threads(ST_GRACEFUL); + } + + if (workers_may_exit) break; + + /* find a listener */ + lr = last_lr; + do { + lr = lr->next; + if (lr == NULL) { + lr = ap_listeners; + } + /* XXX: Should we check for POLLERR? */ + apr_poll_revents_get(&event, lr->sd, pollset); + if (event & APR_POLLIN) { + last_lr = lr; + goto got_fd; + } + } while (lr != last_lr); + } + } + got_fd: + if (!workers_may_exit) { + rv = lr->accept_func(&csd, lr, ptrans); + /* later we trash rv and rely on csd to indicate success/failure */ + AP_DEBUG_ASSERT(rv == APR_SUCCESS || !csd); + + if (rv == APR_EGENERAL) { + /* E[NM]FILE, ENOMEM, etc */ + resource_shortage = 1; + signal_threads(ST_GRACEFUL); + } + if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(accept_mutex))) + != APR_SUCCESS) { + int level = APLOG_EMERG; + + if (workers_may_exit) { + break; + } + if (ap_scoreboard_image->parent[process_slot].generation != + ap_scoreboard_image->global->running_generation) { + level = APLOG_DEBUG; /* common to get these at restart time */ + } + ap_log_error(APLOG_MARK, level, rv, ap_server_conf, + "apr_proc_mutex_unlock failed. Attempting to " + "shutdown process gracefully."); + signal_threads(ST_GRACEFUL); + } + if (csd != NULL) { + is_listener = 0; + worker_stack_awaken_next(idle_worker_stack); + process_socket(ptrans, csd, process_slot, + thread_slot, bucket_alloc); + apr_pool_clear(ptrans); + requests_this_child--; + } + if ((ap_mpm_pod_check(pod) == APR_SUCCESS) || + (ap_my_generation != + ap_scoreboard_image->global->running_generation)) { + signal_threads(ST_GRACEFUL); + break; + } + } + else { + if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(accept_mutex))) + != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "apr_proc_mutex_unlock failed. Attempting to " + "shutdown process gracefully."); + signal_threads(ST_GRACEFUL); + } + break; + } + } + + dying = 1; + ap_scoreboard_image->parent[process_slot].quiescing = 1; + + worker_stack_term(idle_worker_stack); + + ap_update_child_status_from_indexes(process_slot, thread_slot, + (dying) ? SERVER_DEAD : SERVER_GRACEFUL, (request_rec *) NULL); + + apr_bucket_alloc_destroy(bucket_alloc); + + apr_thread_exit(thd, APR_SUCCESS); + return NULL; +} + +static int check_signal(int signum) +{ + switch (signum) { + case SIGTERM: + case SIGINT: + return 1; + } + return 0; +} + +/* XXX under some circumstances not understood, children can get stuck + * in start_threads forever trying to take over slots which will + * never be cleaned up; for now there is an APLOG_DEBUG message issued + * every so often when this condition occurs + */ +static void * APR_THREAD_FUNC start_threads(apr_thread_t *thd, void *dummy) +{ + thread_starter *ts = dummy; + apr_thread_t **threads = ts->threads; + apr_threadattr_t *thread_attr = ts->threadattr; + int child_num_arg = ts->child_num_arg; + int my_child_num = child_num_arg; + proc_info *my_info; + apr_status_t rv; + int i; + int threads_created = 0; + int loops; + int prev_threads_created; + + idle_worker_stack = worker_stack_create(pchild, ap_threads_per_child); + if (idle_worker_stack == NULL) { + ap_log_error(APLOG_MARK, APLOG_ALERT, 0, ap_server_conf, + "worker_stack_create() failed"); + clean_child_exit(APEXIT_CHILDFATAL); + } + + worker_wakeups = (worker_wakeup_info **) + apr_palloc(pchild, sizeof(worker_wakeup_info *) * + ap_threads_per_child); + + loops = prev_threads_created = 0; + while (1) { + for (i = 0; i < ap_threads_per_child; i++) { + int status = ap_scoreboard_image->servers[child_num_arg][i].status; + worker_wakeup_info *wakeup; + + if (status != SERVER_GRACEFUL && status != SERVER_DEAD) { + continue; + } + + wakeup = worker_wakeup_create(pchild); + if (wakeup == NULL) { + ap_log_error(APLOG_MARK, APLOG_ALERT|APLOG_NOERRNO, 0, + ap_server_conf, "worker_wakeup_create failed"); + clean_child_exit(APEXIT_CHILDFATAL); + } + worker_wakeups[threads_created] = wakeup; + my_info = (proc_info *)malloc(sizeof(proc_info)); + if (my_info == NULL) { + ap_log_error(APLOG_MARK, APLOG_ALERT, errno, ap_server_conf, + "malloc: out of memory"); + clean_child_exit(APEXIT_CHILDFATAL); + } + my_info->pid = my_child_num; + my_info->tid = i; + my_info->sd = 0; + + /* We are creating threads right now */ + ap_update_child_status_from_indexes(my_child_num, i, + SERVER_STARTING, NULL); + /* We let each thread update its own scoreboard entry. This is + * done because it lets us deal with tid better. + */ + rv = apr_thread_create(&threads[i], thread_attr, + worker_thread, my_info, pchild); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf, + "apr_thread_create: unable to create worker thread"); + /* In case system resources are maxxed out, we don't want + Apache running away with the CPU trying to fork over and + over and over again if we exit. */ + apr_sleep(10 * APR_USEC_PER_SEC); + clean_child_exit(APEXIT_CHILDFATAL); + } + threads_created++; + } + if (start_thread_may_exit || threads_created == ap_threads_per_child) { + break; + } + /* wait for previous generation to clean up an entry */ + apr_sleep(1 * APR_USEC_PER_SEC); + ++loops; + if (loops % 120 == 0) { /* every couple of minutes */ + if (prev_threads_created == threads_created) { + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "child %" APR_PID_T_FMT " isn't taking over " + "slots very quickly (%d of %d)", + ap_my_pid, threads_created, ap_threads_per_child); + } + prev_threads_created = threads_created; + } + } + + /* What state should this child_main process be listed as in the + * scoreboard...? + * ap_update_child_status_from_indexes(my_child_num, i, SERVER_STARTING, + * (request_rec *) NULL); + * + * This state should be listed separately in the scoreboard, in some kind + * of process_status, not mixed in with the worker threads' status. + * "life_status" is almost right, but it's in the worker's structure, and + * the name could be clearer. gla + */ + apr_thread_exit(thd, APR_SUCCESS); + return NULL; +} + +static void join_workers(apr_thread_t **threads) +{ + int i; + apr_status_t rv, thread_rv; + + for (i = 0; i < ap_threads_per_child; i++) { + if (threads[i]) { /* if we ever created this thread */ + rv = apr_thread_join(&thread_rv, threads[i]); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, + "apr_thread_join: unable to join worker " + "thread %d", + i); + } + } + } +} + +static void join_start_thread(apr_thread_t *start_thread_id) +{ + apr_status_t rv, thread_rv; + + start_thread_may_exit = 1; /* tell it to give up in case it is still + * trying to take over slots from a + * previous generation + */ + rv = apr_thread_join(&thread_rv, start_thread_id); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, + "apr_thread_join: unable to join the start " + "thread"); + } +} + +static void child_main(int child_num_arg) +{ + apr_thread_t **threads; + apr_status_t rv; + thread_starter *ts; + apr_threadattr_t *thread_attr; + apr_thread_t *start_thread_id; + + mpm_state = AP_MPMQ_STARTING; /* for benefit of any hooks that run as this + * child initializes + */ + + ap_my_pid = getpid(); + ap_fatal_signal_child_setup(ap_server_conf); + apr_pool_create(&pchild, pconf); + + /*stuff to do before we switch id's, so we have permissions.*/ + ap_reopen_scoreboard(pchild, NULL, 0); + + rv = SAFE_ACCEPT(apr_proc_mutex_child_init(&accept_mutex, ap_lock_fname, + pchild)); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "Couldn't initialize cross-process lock in child"); + clean_child_exit(APEXIT_CHILDFATAL); + } + + if (unixd_setup_child()) { + clean_child_exit(APEXIT_CHILDFATAL); + } + + ap_run_child_init(pchild, ap_server_conf); + + /* done with init critical section */ + + /* Just use the standard apr_setup_signal_thread to block all signals + * from being received. The child processes no longer use signals for + * any communication with the parent process. + */ + rv = apr_setup_signal_thread(); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "Couldn't initialize signal thread"); + clean_child_exit(APEXIT_CHILDFATAL); + } + + if (ap_max_requests_per_child) { + requests_this_child = ap_max_requests_per_child; + } + else { + /* coding a value of zero means infinity */ + requests_this_child = INT_MAX; + } + + /* Setup worker threads */ + + /* clear the storage; we may not create all our threads immediately, + * and we want a 0 entry to indicate a thread which was not created + */ + threads = (apr_thread_t **)calloc(1, + sizeof(apr_thread_t *) * ap_threads_per_child); + if (threads == NULL) { + ap_log_error(APLOG_MARK, APLOG_ALERT, errno, ap_server_conf, + "malloc: out of memory"); + clean_child_exit(APEXIT_CHILDFATAL); + } + + ts = (thread_starter *)apr_palloc(pchild, sizeof(*ts)); + + apr_threadattr_create(&thread_attr, pchild); + /* 0 means PTHREAD_CREATE_JOINABLE */ + apr_threadattr_detach_set(thread_attr, 0); + + ts->threads = threads; + ts->child_num_arg = child_num_arg; + ts->threadattr = thread_attr; + + rv = apr_thread_create(&start_thread_id, thread_attr, start_threads, + ts, pchild); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf, + "apr_thread_create: unable to create worker thread"); + /* In case system resources are maxxed out, we don't want + Apache running away with the CPU trying to fork over and + over and over again if we exit. */ + apr_sleep(10 * APR_USEC_PER_SEC); + clean_child_exit(APEXIT_CHILDFATAL); + } + + mpm_state = AP_MPMQ_RUNNING; + + /* If we are only running in one_process mode, we will want to + * still handle signals. */ + if (one_process) { + /* Block until we get a terminating signal. */ + apr_signal_thread(check_signal); + /* make sure the start thread has finished; signal_threads() + * and join_workers() depend on that + */ + /* XXX join_start_thread() won't be awakened if one of our + * threads encounters a critical error and attempts to + * shutdown this child + */ + join_start_thread(start_thread_id); + signal_threads(ST_UNGRACEFUL); /* helps us terminate a little more + * quickly than the dispatch of the signal thread + * beats the Pipe of Death and the browsers + */ + /* A terminating signal was received. Now join each of the + * workers to clean them up. + * If the worker already exited, then the join frees + * their resources and returns. + * If the worker hasn't exited, then this blocks until + * they have (then cleans up). + */ + join_workers(threads); + } + else { /* !one_process */ + /* remove SIGTERM from the set of blocked signals... if one of + * the other threads in the process needs to take us down + * (e.g., for MaxRequestsPerChild) it will send us SIGTERM + */ + unblock_signal(SIGTERM); + join_start_thread(start_thread_id); + join_workers(threads); + } + + free(threads); + + clean_child_exit(resource_shortage ? APEXIT_CHILDSICK : 0); +} + +static int make_child(server_rec *s, int slot) +{ + int pid; + + if (slot + 1 > ap_max_daemons_limit) { + ap_max_daemons_limit = slot + 1; + } + + if (one_process) { + set_signals(); + ap_scoreboard_image->parent[slot].pid = getpid(); + child_main(slot); + } + + if ((pid = fork()) == -1) { + ap_log_error(APLOG_MARK, APLOG_ERR, errno, s, + "fork: Unable to fork new process"); + + /* fork didn't succeed. Fix the scoreboard or else + * it will say SERVER_STARTING forever and ever + */ + ap_update_child_status_from_indexes(slot, 0, SERVER_DEAD, NULL); + + /* In case system resources are maxxed out, we don't want + Apache running away with the CPU trying to fork over and + over and over again. */ + apr_sleep(10 * APR_USEC_PER_SEC); + + return -1; + } + + if (!pid) { +#ifdef HAVE_BINDPROCESSOR + /* By default, AIX binds to a single processor. This bit unbinds + * children which will then bind to another CPU. + */ + int status = bindprocessor(BINDPROCESS, (int)getpid(), + PROCESSOR_CLASS_ANY); + if (status != OK) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, + ap_server_conf, + "processor unbind failed %d", status); +#endif + RAISE_SIGSTOP(MAKE_CHILD); + + apr_signal(SIGTERM, just_die); + child_main(slot); + + clean_child_exit(0); + } + /* else */ + ap_scoreboard_image->parent[slot].quiescing = 0; + ap_scoreboard_image->parent[slot].pid = pid; + return 0; +} + +/* start up a bunch of children */ +static void startup_children(int number_to_start) +{ + int i; + + for (i = 0; number_to_start && i < ap_daemons_limit; ++i) { + if (ap_scoreboard_image->parent[i].pid != 0) { + continue; + } + if (make_child(ap_server_conf, i) < 0) { + break; + } + --number_to_start; + } +} + + +/* + * idle_spawn_rate is the number of children that will be spawned on the + * next maintenance cycle if there aren't enough idle servers. It is + * doubled up to MAX_SPAWN_RATE, and reset only when a cycle goes by + * without the need to spawn. + */ +static int idle_spawn_rate = 1; +#ifndef MAX_SPAWN_RATE +#define MAX_SPAWN_RATE (32) +#endif +static int hold_off_on_exponential_spawning; + +static void perform_idle_server_maintenance(void) +{ + int i, j; + int idle_thread_count; + worker_score *ws; + process_score *ps; + int free_length; + int totally_free_length = 0; + int free_slots[MAX_SPAWN_RATE]; + int last_non_dead; + int total_non_dead; + + /* initialize the free_list */ + free_length = 0; + + idle_thread_count = 0; + last_non_dead = -1; + total_non_dead = 0; + + for (i = 0; i < ap_daemons_limit; ++i) { + /* Initialization to satisfy the compiler. It doesn't know + * that ap_threads_per_child is always > 0 */ + int status = SERVER_DEAD; + int any_dying_threads = 0; + int any_dead_threads = 0; + int all_dead_threads = 1; + + if (i >= ap_max_daemons_limit && totally_free_length == idle_spawn_rate) + break; + ps = &ap_scoreboard_image->parent[i]; + for (j = 0; j < ap_threads_per_child; j++) { + ws = &ap_scoreboard_image->servers[i][j]; + status = ws->status; + + /* XXX any_dying_threads is probably no longer needed GLA */ + any_dying_threads = any_dying_threads || + (status == SERVER_GRACEFUL); + any_dead_threads = any_dead_threads || (status == SERVER_DEAD); + all_dead_threads = all_dead_threads && + (status == SERVER_DEAD || + status == SERVER_GRACEFUL); + + /* We consider a starting server as idle because we started it + * at least a cycle ago, and if it still hasn't finished starting + * then we're just going to swamp things worse by forking more. + * So we hopefully won't need to fork more if we count it. + * This depends on the ordering of SERVER_READY and SERVER_STARTING. + */ + if (status <= SERVER_READY && status != SERVER_DEAD && + !ps->quiescing && + ps->generation == ap_my_generation && + /* XXX the following shouldn't be necessary if we clean up + * properly after seg faults, but we're not yet GLA + */ + ps->pid != 0) { + ++idle_thread_count; + } + } + if (any_dead_threads && totally_free_length < idle_spawn_rate + && (!ps->pid /* no process in the slot */ + || ps->quiescing)) { /* or at least one is going away */ + if (all_dead_threads) { + /* great! we prefer these, because the new process can + * start more threads sooner. So prioritize this slot + * by putting it ahead of any slots with active threads. + * + * first, make room by moving a slot that's potentially still + * in use to the end of the array + */ + free_slots[free_length] = free_slots[totally_free_length]; + free_slots[totally_free_length++] = i; + } + else { + /* slot is still in use - back of the bus + */ + free_slots[free_length] = i; + } + ++free_length; + } + /* XXX if (!ps->quiescing) is probably more reliable GLA */ + if (!any_dying_threads) { + last_non_dead = i; + ++total_non_dead; + } + } + ap_max_daemons_limit = last_non_dead + 1; + + if (idle_thread_count > max_spare_threads) { + /* Kill off one child */ + ap_mpm_pod_signal(pod); + idle_spawn_rate = 1; + } + else if (idle_thread_count < min_spare_threads) { + /* terminate the free list */ + if (free_length == 0) { + /* only report this condition once */ + static int reported = 0; + + if (!reported) { + ap_log_error(APLOG_MARK, APLOG_ERR, 0, + ap_server_conf, + "server reached MaxClients setting, consider" + " raising the MaxClients setting"); + reported = 1; + } + idle_spawn_rate = 1; + } + else { + if (free_length > idle_spawn_rate) { + free_length = idle_spawn_rate; + } + if (idle_spawn_rate >= 8) { + ap_log_error(APLOG_MARK, APLOG_INFO, 0, + ap_server_conf, + "server seems busy, (you may need " + "to increase StartServers, ThreadsPerChild " + "or Min/MaxSpareThreads), " + "spawning %d children, there are around %d idle " + "threads, and %d total children", free_length, + idle_thread_count, total_non_dead); + } + for (i = 0; i < free_length; ++i) { + make_child(ap_server_conf, free_slots[i]); + } + /* the next time around we want to spawn twice as many if this + * wasn't good enough, but not if we've just done a graceful + */ + if (hold_off_on_exponential_spawning) { + --hold_off_on_exponential_spawning; + } + else if (idle_spawn_rate < MAX_SPAWN_RATE) { + idle_spawn_rate *= 2; + } + } + } + else { + idle_spawn_rate = 1; + } +} + +static void server_main_loop(int remaining_children_to_start) +{ + int child_slot; + apr_exit_why_e exitwhy; + int status, processed_status; + apr_proc_t pid; + int i; + + while (!restart_pending && !shutdown_pending) { + ap_wait_or_timeout(&exitwhy, &status, &pid, pconf); + + if (pid.pid != -1) { + processed_status = ap_process_child_status(&pid, exitwhy, status); + if (processed_status == APEXIT_CHILDFATAL) { + shutdown_pending = 1; + child_fatal = 1; + return; + } + /* non-fatal death... note that it's gone in the scoreboard. */ + child_slot = find_child_by_pid(&pid); + if (child_slot >= 0) { + for (i = 0; i < ap_threads_per_child; i++) + ap_update_child_status_from_indexes(child_slot, i, SERVER_DEAD, + (request_rec *) NULL); + + ap_scoreboard_image->parent[child_slot].pid = 0; + ap_scoreboard_image->parent[child_slot].quiescing = 0; + if (processed_status == APEXIT_CHILDSICK) { + /* resource shortage, minimize the fork rate */ + idle_spawn_rate = 1; + } + else if (remaining_children_to_start + && child_slot < ap_daemons_limit) { + /* we're still doing a 1-for-1 replacement of dead + * children with new children + */ + make_child(ap_server_conf, child_slot); + --remaining_children_to_start; + } +#if APR_HAS_OTHER_CHILD + } + else if (apr_proc_other_child_read(&pid, status) == 0) { + /* handled */ +#endif + } + else if (is_graceful) { + /* Great, we've probably just lost a slot in the + * scoreboard. Somehow we don't know about this child. + */ + ap_log_error(APLOG_MARK, APLOG_WARNING, 0, + ap_server_conf, + "long lost child came home! (pid %ld)", + (long)pid.pid); + } + /* Don't perform idle maintenance when a child dies, + * only do it when there's a timeout. Remember only a + * finite number of children can die, and it's pretty + * pathological for a lot to die suddenly. + */ + continue; + } + else if (remaining_children_to_start) { + /* we hit a 1 second timeout in which none of the previous + * generation of children needed to be reaped... so assume + * they're all done, and pick up the slack if any is left. + */ + startup_children(remaining_children_to_start); + remaining_children_to_start = 0; + /* In any event we really shouldn't do the code below because + * few of the servers we just started are in the IDLE state + * yet, so we'd mistakenly create an extra server. + */ + continue; + } + + perform_idle_server_maintenance(); + } +} + +int ap_mpm_run(apr_pool_t *_pconf, apr_pool_t *plog, server_rec *s) +{ + int remaining_children_to_start; + apr_status_t rv; + + ap_log_pid(pconf, ap_pid_fname); + + first_server_limit = server_limit; + first_thread_limit = thread_limit; + if (changed_limit_at_restart) { + ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, + "WARNING: Attempt to change ServerLimit or ThreadLimit " + "ignored during restart"); + changed_limit_at_restart = 0; + } + + /* Initialize cross-process accept lock */ + ap_lock_fname = apr_psprintf(_pconf, "%s.%" APR_PID_T_FMT, + ap_server_root_relative(_pconf, ap_lock_fname), + ap_my_pid); + + rv = apr_proc_mutex_create(&accept_mutex, ap_lock_fname, + ap_accept_lock_mech, _pconf); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, s, + "Couldn't create accept lock"); + mpm_state = AP_MPMQ_STOPPING; + return 1; + } + +#if APR_USE_SYSVSEM_SERIALIZE + if (ap_accept_lock_mech == APR_LOCK_DEFAULT || + ap_accept_lock_mech == APR_LOCK_SYSVSEM) { +#else + if (ap_accept_lock_mech == APR_LOCK_SYSVSEM) { +#endif + rv = unixd_set_proc_mutex_perms(accept_mutex); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, s, + "Couldn't set permissions on cross-process lock; " + "check User and Group directives"); + mpm_state = AP_MPMQ_STOPPING; + return 1; + } + } + + if (!is_graceful) { + if (ap_run_pre_mpm(s->process->pool, SB_SHARED) != OK) { + mpm_state = AP_MPMQ_STOPPING; + return 1; + } + /* fix the generation number in the global score; we just got a new, + * cleared scoreboard + */ + ap_scoreboard_image->global->running_generation = ap_my_generation; + } + + set_signals(); + /* Don't thrash... */ + if (max_spare_threads < min_spare_threads + ap_threads_per_child) + max_spare_threads = min_spare_threads + ap_threads_per_child; + + /* If we're doing a graceful_restart then we're going to see a lot + * of children exiting immediately when we get into the main loop + * below (because we just sent them AP_SIG_GRACEFUL). This happens pretty + * rapidly... and for each one that exits we'll start a new one until + * we reach at least daemons_min_free. But we may be permitted to + * start more than that, so we'll just keep track of how many we're + * supposed to start up without the 1 second penalty between each fork. + */ + remaining_children_to_start = ap_daemons_to_start; + if (remaining_children_to_start > ap_daemons_limit) { + remaining_children_to_start = ap_daemons_limit; + } + if (!is_graceful) { + startup_children(remaining_children_to_start); + remaining_children_to_start = 0; + } + else { + /* give the system some time to recover before kicking into + * exponential mode */ + hold_off_on_exponential_spawning = 10; + } + + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, + "%s configured -- resuming normal operations", + ap_get_server_version()); + ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf, + "Server built: %s", ap_get_server_built()); +#ifdef AP_MPM_WANT_SET_ACCEPT_LOCK_MECH + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "AcceptMutex: %s (default: %s)", + apr_proc_mutex_name(accept_mutex), + apr_proc_mutex_defname()); +#endif + restart_pending = shutdown_pending = 0; + mpm_state = AP_MPMQ_RUNNING; + + server_main_loop(remaining_children_to_start); + mpm_state = AP_MPMQ_STOPPING; + + if (shutdown_pending) { + /* Time to gracefully shut down: + * Kill child processes, tell them to call child_exit, etc... + * (By "gracefully" we don't mean graceful in the same sense as + * "apachectl graceful" where we allow old connections to finish.) + */ + if (unixd_killpg(getpgrp(), SIGTERM) < 0) { + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, "killpg SIGTERM"); + } + ap_reclaim_child_processes(1); /* Start with SIGTERM */ + + if (!child_fatal) { + /* cleanup pid file on normal shutdown */ + const char *pidfile = NULL; + pidfile = ap_server_root_relative (pconf, ap_pid_fname); + if ( pidfile != NULL && unlink(pidfile) == 0) + ap_log_error(APLOG_MARK, APLOG_INFO, 0, + ap_server_conf, + "removed PID file %s (pid=%ld)", + pidfile, (long)getpid()); + + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, + ap_server_conf, "caught SIGTERM, shutting down"); + } + return 1; + } + + /* we've been told to restart */ + apr_signal(SIGHUP, SIG_IGN); + + if (one_process) { + /* not worth thinking about */ + return 1; + } + + /* advance to the next generation */ + /* XXX: we really need to make sure this new generation number isn't in + * use by any of the children. + */ + ++ap_my_generation; + ap_scoreboard_image->global->running_generation = ap_my_generation; + + if (is_graceful) { + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, + AP_SIG_GRACEFUL_STRING " received. Doing graceful restart"); + /* wake up the children...time to die. But we'll have more soon */ + ap_mpm_pod_killpg(pod, ap_daemons_limit); + + + /* This is mostly for debugging... so that we know what is still + * gracefully dealing with existing request. + */ + + } + else { + /* Kill 'em all. Since the child acts the same on the parents SIGTERM + * and a SIGHUP, we may as well use the same signal, because some user + * pthreads are stealing signals from us left and right. + */ + ap_mpm_pod_killpg(pod, ap_daemons_limit); + + ap_reclaim_child_processes(1); /* Start with SIGTERM */ + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, + "SIGHUP received. Attempting to restart"); + } + + return 0; +} + +/* This really should be a post_config hook, but the error log is already + * redirected by that point, so we need to do this in the open_logs phase. + */ +static int leader_open_logs(apr_pool_t *p, apr_pool_t *plog, apr_pool_t *ptemp, server_rec *s) +{ + apr_status_t rv; + + pconf = p; + ap_server_conf = s; + + if ((num_listensocks = ap_setup_listeners(ap_server_conf)) < 1) { + ap_log_error(APLOG_MARK, APLOG_ALERT|APLOG_STARTUP, 0, + NULL, "no listening sockets available, shutting down"); + return DONE; + } + + if (!one_process) { + if ((rv = ap_mpm_pod_open(pconf, &pod))) { + ap_log_error(APLOG_MARK, APLOG_CRIT|APLOG_STARTUP, rv, NULL, + "Could not open pipe-of-death."); + return DONE; + } + } + return OK; +} + +static int leader_pre_config(apr_pool_t *pconf, apr_pool_t *plog, + apr_pool_t *ptemp) +{ + static int restart_num = 0; + int no_detach, debug, foreground; + ap_directive_t *pdir; + ap_directive_t *max_clients = NULL; + apr_status_t rv; + + mpm_state = AP_MPMQ_STARTING; + + /* make sure that "ThreadsPerChild" gets set before "MaxClients" */ + for (pdir = ap_conftree; pdir != NULL; pdir = pdir->next) { + if (strncasecmp(pdir->directive, "ThreadsPerChild", 15) == 0) { + if (!max_clients) { + break; /* we're in the clear, got ThreadsPerChild first */ + } + else { + /* now to swap the data */ + ap_directive_t temp; + + temp.directive = pdir->directive; + temp.args = pdir->args; + /* Make sure you don't change 'next', or you may get loops! */ + /* XXX: first_child, parent, and data can never be set + * for these directives, right? -aaron */ + temp.filename = pdir->filename; + temp.line_num = pdir->line_num; + + pdir->directive = max_clients->directive; + pdir->args = max_clients->args; + pdir->filename = max_clients->filename; + pdir->line_num = max_clients->line_num; + + max_clients->directive = temp.directive; + max_clients->args = temp.args; + max_clients->filename = temp.filename; + max_clients->line_num = temp.line_num; + break; + } + } + else if (!max_clients + && strncasecmp(pdir->directive, "MaxClients", 10) == 0) { + max_clients = pdir; + } + } + + debug = ap_exists_config_define("DEBUG"); + + if (debug) { + foreground = one_process = 1; + no_detach = 0; + } + else { + one_process = ap_exists_config_define("ONE_PROCESS"); + no_detach = ap_exists_config_define("NO_DETACH"); + foreground = ap_exists_config_define("FOREGROUND"); + } + + /* sigh, want this only the second time around */ + if (restart_num++ == 1) { + is_graceful = 0; + + if (!one_process && !foreground) { + rv = apr_proc_detach(no_detach ? APR_PROC_DETACH_FOREGROUND + : APR_PROC_DETACH_DAEMONIZE); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, NULL, + "apr_proc_detach failed"); + return HTTP_INTERNAL_SERVER_ERROR; + } + } + parent_pid = ap_my_pid = getpid(); + } + + unixd_pre_config(ptemp); + ap_listen_pre_config(); + ap_daemons_to_start = DEFAULT_START_DAEMON; + min_spare_threads = DEFAULT_MIN_FREE_DAEMON * DEFAULT_THREADS_PER_CHILD; + max_spare_threads = DEFAULT_MAX_FREE_DAEMON * DEFAULT_THREADS_PER_CHILD; + ap_daemons_limit = server_limit; + ap_threads_per_child = DEFAULT_THREADS_PER_CHILD; + ap_pid_fname = DEFAULT_PIDLOG; + ap_lock_fname = DEFAULT_LOCKFILE; + ap_max_requests_per_child = DEFAULT_MAX_REQUESTS_PER_CHILD; + ap_extended_status = 0; +#ifdef AP_MPM_WANT_SET_MAX_MEM_FREE + ap_max_mem_free = APR_ALLOCATOR_MAX_FREE_UNLIMITED; +#endif + + apr_cpystrn(ap_coredump_dir, ap_server_root, sizeof(ap_coredump_dir)); + + return OK; +} + +static void leader_hooks(apr_pool_t *p) +{ + /* The leader open_logs phase must run before the core's, or stderr + * will be redirected to a file, and the messages won't print to the + * console. + */ + static const char *const aszSucc[] = {"core.c", NULL}; + one_process = 0; + + ap_hook_open_logs(leader_open_logs, NULL, aszSucc, APR_HOOK_MIDDLE); + /* we need to set the MPM state before other pre-config hooks use MPM query + * to retrieve it, so register as REALLY_FIRST + */ + ap_hook_pre_config(leader_pre_config, NULL, NULL, APR_HOOK_REALLY_FIRST); +} + +static const char *set_daemons_to_start(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + ap_daemons_to_start = atoi(arg); + return NULL; +} + +static const char *set_min_spare_threads(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + min_spare_threads = atoi(arg); + if (min_spare_threads <= 0) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: detected MinSpareThreads set to non-positive."); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "Resetting to 1 to avoid almost certain Apache failure."); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "Please read the documentation."); + min_spare_threads = 1; + } + + return NULL; +} + +static const char *set_max_spare_threads(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + max_spare_threads = atoi(arg); + return NULL; +} + +static const char *set_max_clients (cmd_parms *cmd, void *dummy, + const char *arg) +{ + int max_clients; + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + /* It is ok to use ap_threads_per_child here because we are + * sure that it gets set before MaxClients in the pre_config stage. */ + max_clients = atoi(arg); + if (max_clients < ap_threads_per_child) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: MaxClients (%d) must be at least as large", + max_clients); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " large as ThreadsPerChild (%d). Automatically", + ap_threads_per_child); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " increasing MaxClients to %d.", + ap_threads_per_child); + max_clients = ap_threads_per_child; + } + ap_daemons_limit = max_clients / ap_threads_per_child; + if ((max_clients > 0) && (max_clients % ap_threads_per_child)) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: MaxClients (%d) is not an integer multiple", + max_clients); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " of ThreadsPerChild (%d), lowering MaxClients to %d", + ap_threads_per_child, + ap_daemons_limit * ap_threads_per_child); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " for a maximum of %d child processes,", + ap_daemons_limit); + max_clients = ap_daemons_limit * ap_threads_per_child; + } + if (ap_daemons_limit > server_limit) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: MaxClients of %d would require %d servers,", + max_clients, ap_daemons_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " and would exceed the ServerLimit value of %d.", + server_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " Automatically lowering MaxClients to %d. To increase,", + server_limit * ap_threads_per_child); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " please see the ServerLimit directive."); + ap_daemons_limit = server_limit; + } + else if (ap_daemons_limit < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require MaxClients > 0, setting to 1"); + ap_daemons_limit = 1; + } + return NULL; +} + +static const char *set_threads_per_child (cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + ap_threads_per_child = atoi(arg); + if (ap_threads_per_child > thread_limit) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: ThreadsPerChild of %d exceeds ThreadLimit " + "value of %d", ap_threads_per_child, + thread_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "threads, lowering ThreadsPerChild to %d. To increase, please" + " see the", thread_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " ThreadLimit directive."); + ap_threads_per_child = thread_limit; + } + else if (ap_threads_per_child < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require ThreadsPerChild > 0, setting to 1"); + ap_threads_per_child = 1; + } + return NULL; +} + +static const char *set_server_limit (cmd_parms *cmd, void *dummy, const char *arg) +{ + int tmp_server_limit; + + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + tmp_server_limit = atoi(arg); + /* you cannot change ServerLimit across a restart; ignore + * any such attempts + */ + if (first_server_limit && + tmp_server_limit != server_limit) { + /* how do we log a message? the error log is a bit bucket at this + * point; we'll just have to set a flag so that ap_mpm_run() + * logs a warning later + */ + changed_limit_at_restart = 1; + return NULL; + } + server_limit = tmp_server_limit; + + if (server_limit > MAX_SERVER_LIMIT) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: ServerLimit of %d exceeds compile time limit " + "of %d servers,", server_limit, MAX_SERVER_LIMIT); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " lowering ServerLimit to %d.", MAX_SERVER_LIMIT); + server_limit = MAX_SERVER_LIMIT; + } + else if (server_limit < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require ServerLimit > 0, setting to 1"); + server_limit = 1; + } + return NULL; +} + +static const char *set_thread_limit (cmd_parms *cmd, void *dummy, const char *arg) +{ + int tmp_thread_limit; + + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + tmp_thread_limit = atoi(arg); + /* you cannot change ThreadLimit across a restart; ignore + * any such attempts + */ + if (first_thread_limit && + tmp_thread_limit != thread_limit) { + /* how do we log a message? the error log is a bit bucket at this + * point; we'll just have to set a flag so that ap_mpm_run() + * logs a warning later + */ + changed_limit_at_restart = 1; + return NULL; + } + thread_limit = tmp_thread_limit; + + if (thread_limit > MAX_THREAD_LIMIT) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: ThreadLimit of %d exceeds compile time limit " + "of %d servers,", thread_limit, MAX_THREAD_LIMIT); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " lowering ThreadLimit to %d.", MAX_THREAD_LIMIT); + thread_limit = MAX_THREAD_LIMIT; + } + else if (thread_limit < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require ThreadLimit > 0, setting to 1"); + thread_limit = 1; + } + return NULL; +} + +static const command_rec leader_cmds[] = { +UNIX_DAEMON_COMMANDS, +LISTEN_COMMANDS, +AP_INIT_TAKE1("StartServers", set_daemons_to_start, NULL, RSRC_CONF, + "Number of child processes launched at server startup"), +AP_INIT_TAKE1("MinSpareThreads", set_min_spare_threads, NULL, RSRC_CONF, + "Minimum number of idle children, to handle request spikes"), +AP_INIT_TAKE1("MaxSpareThreads", set_max_spare_threads, NULL, RSRC_CONF, + "Maximum number of idle children"), +AP_INIT_TAKE1("MaxClients", set_max_clients, NULL, RSRC_CONF, + "Maximum number of children alive at the same time"), +AP_INIT_TAKE1("ThreadsPerChild", set_threads_per_child, NULL, RSRC_CONF, + "Number of threads each child creates"), +AP_INIT_TAKE1("ServerLimit", set_server_limit, NULL, RSRC_CONF, + "Maximum value of MaxClients for this run of Apache"), +AP_INIT_TAKE1("ThreadLimit", set_thread_limit, NULL, RSRC_CONF, + "Maximum worker threads in a server for this run of Apache"), +{ NULL } +}; + +module AP_MODULE_DECLARE_DATA mpm_leader_module = { + MPM20_MODULE_STUFF, + ap_mpm_rewrite_args, /* hook to run before apache parses args */ + NULL, /* create per-directory config structure */ + NULL, /* merge per-directory config structures */ + NULL, /* create per-server config structure */ + NULL, /* merge per-server config structures */ + leader_cmds, /* command apr_table_t */ + leader_hooks /* register_hooks */ +}; + diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/mpm.h b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/mpm.h new file mode 100644 index 00000000..1db1d1d0 --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/mpm.h @@ -0,0 +1,51 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "scoreboard.h" +#include "unixd.h" + +#ifndef APACHE_MPM_LEADER_H +#define APACHE_MPM_LEADER_H + +#define LEADER_MPM + +#define MPM_NAME "Leader-Follower" + +#define AP_MPM_WANT_RECLAIM_CHILD_PROCESSES +#define AP_MPM_WANT_WAIT_OR_TIMEOUT +#define AP_MPM_WANT_PROCESS_CHILD_STATUS +#define AP_MPM_WANT_SET_PIDFILE +#define AP_MPM_WANT_SET_SCOREBOARD +#define AP_MPM_WANT_SET_LOCKFILE +#define AP_MPM_WANT_SET_MAX_REQUESTS +#define AP_MPM_WANT_SET_COREDUMPDIR +#define AP_MPM_WANT_SET_ACCEPT_LOCK_MECH +#define AP_MPM_WANT_SIGNAL_SERVER +#define AP_MPM_WANT_SET_MAX_MEM_FREE +#define AP_MPM_WANT_FATAL_SIGNAL_HANDLER +#define AP_MPM_DISABLE_NAGLE_ACCEPTED_SOCK + +#define AP_MPM_USES_POD 1 +#define MPM_CHILD_PID(i) (ap_scoreboard_image->parent[i].pid) +#define MPM_NOTE_CHILD_KILLED(i) (MPM_CHILD_PID(i) = 0) +#define MPM_ACCEPT_FUNC unixd_accept + +extern int ap_threads_per_child; +extern int ap_max_daemons_limit; +extern server_rec *ap_server_conf; +extern char ap_coredump_dir[MAX_STRING_LEN]; + +#endif /* APACHE_MPM_LEADER_H */ diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/mpm_default.h b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/mpm_default.h new file mode 100644 index 00000000..d5a33989 --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/leader/mpm_default.h @@ -0,0 +1,69 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef APACHE_MPM_DEFAULT_H +#define APACHE_MPM_DEFAULT_H + +/* Number of servers to spawn off by default --- also, if fewer than + * this free when the caretaker checks, it will spawn more. + */ +#ifndef DEFAULT_START_DAEMON +#define DEFAULT_START_DAEMON 3 +#endif + +/* Maximum number of *free* server processes --- more than this, and + * they will die off. + */ + +#ifndef DEFAULT_MAX_FREE_DAEMON +#define DEFAULT_MAX_FREE_DAEMON 10 +#endif + +/* Minimum --- fewer than this, and more will be created */ + +#ifndef DEFAULT_MIN_FREE_DAEMON +#define DEFAULT_MIN_FREE_DAEMON 3 +#endif + +#ifndef DEFAULT_THREADS_PER_CHILD +#define DEFAULT_THREADS_PER_CHILD 25 +#endif + +/* File used for accept locking, when we use a file */ +#ifndef DEFAULT_LOCKFILE +#define DEFAULT_LOCKFILE DEFAULT_REL_RUNTIMEDIR "/accept.lock" +#endif + +/* Where the main/parent process's pid is logged */ +#ifndef DEFAULT_PIDLOG +#define DEFAULT_PIDLOG DEFAULT_REL_RUNTIMEDIR "/httpd.pid" +#endif + +/* + * Interval, in microseconds, between scoreboard maintenance. + */ +#ifndef SCOREBOARD_MAINTENANCE_INTERVAL +#define SCOREBOARD_MAINTENANCE_INTERVAL 1000000 +#endif + +/* Number of requests to try to handle in a single process. If <= 0, + * the children don't die off. + */ +#ifndef DEFAULT_MAX_REQUESTS_PER_CHILD +#define DEFAULT_MAX_REQUESTS_PER_CHILD 10000 +#endif + +#endif /* AP_MPM_DEFAULT_H */ diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/Makefile.in b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/Makefile.in new file mode 100644 index 00000000..374f1306 --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/Makefile.in @@ -0,0 +1,5 @@ + +LTLIBRARY_NAME = libperchild.la +LTLIBRARY_SOURCES = perchild.c + +include $(top_srcdir)/build/ltlib.mk diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/config5.m4 b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/config5.m4 new file mode 100644 index 00000000..368052f5 --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/config5.m4 @@ -0,0 +1,6 @@ +dnl ## XXX - Need a more thorough check of the proper flags to use + +if test "$MPM_NAME" = "perchild" ; then + AC_CHECK_FUNCS(pthread_kill) + APACHE_FAST_OUTPUT(server/mpm/$MPM_SUBDIR_NAME/Makefile) +fi diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/mpm.h b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/mpm.h new file mode 100644 index 00000000..84e808db --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/mpm.h @@ -0,0 +1,59 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "httpd.h" +#include "mpm_default.h" +#include "unixd.h" + +#ifndef APACHE_MPM_PERCHILD_H +#define APACHE_MPM_PERCHILD_H + +#define PERCHILD_MPM + +#define MPM_NAME "Perchild" + +#define AP_MPM_WANT_RECLAIM_CHILD_PROCESSES +#define AP_MPM_WANT_WAIT_OR_TIMEOUT +#define AP_MPM_WANT_PROCESS_CHILD_STATUS +#define AP_MPM_WANT_SET_PIDFILE +#define AP_MPM_WANT_SET_SCOREBOARD +#define AP_MPM_WANT_SET_LOCKFILE +#define AP_MPM_WANT_SET_MAX_REQUESTS +#define AP_MPM_WANT_SET_COREDUMPDIR +#define AP_MPM_WANT_SET_ACCEPT_LOCK_MECH +#define AP_MPM_WANT_SIGNAL_SERVER +#define AP_MPM_WANT_FATAL_SIGNAL_HANDLER +#define AP_MPM_USES_POD + +#define MPM_CHILD_PID(i) (ap_scoreboard_image->parent[i].pid) +#define MPM_NOTE_CHILD_KILLED(i) (MPM_CHILD_PID(i) = 0) +#define MPM_ACCEPT_FUNC unixd_accept + +/* Table of child status */ +#define SERVER_DEAD 0 +#define SERVER_DYING 1 +#define SERVER_ALIVE 2 + +typedef struct ap_ctable{ + pid_t pid; + unsigned char status; +} ap_ctable; + +extern int ap_threads_per_child; +extern int ap_max_daemons_limit; +extern server_rec *ap_server_conf; + +#endif /* APACHE_MPM_PERCHILD_H */ diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/mpm_default.h b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/mpm_default.h new file mode 100644 index 00000000..ece876b5 --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/mpm_default.h @@ -0,0 +1,71 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef APACHE_MPM_DEFAULT_H +#define APACHE_MPM_DEFAULT_H + +/* Number of threads to spawn off by default --- also, if fewer than + * this free when the caretaker checks, it will spawn more. + */ +#ifndef DEFAULT_START_THREAD +#define DEFAULT_START_THREAD 5 +#endif + +/* Maximum number of *free* server threads --- more than this, and + * they will die off. + */ + +#ifndef DEFAULT_MAX_SPARE_THREAD +#define DEFAULT_MAX_SPARE_THREAD 10 +#endif + +/* Minimum --- fewer than this, and more will be created */ + +#ifndef DEFAULT_MIN_SPARE_THREAD +#define DEFAULT_MIN_SPARE_THREAD 5 +#endif + +/* Number of servers to spawn off by default + */ +#ifndef DEFAULT_NUM_DAEMON +#define DEFAULT_NUM_DAEMON 2 +#endif + +/* File used for accept locking, when we use a file */ +#ifndef DEFAULT_LOCKFILE +#define DEFAULT_LOCKFILE DEFAULT_REL_RUNTIMEDIR "/accept.lock" +#endif + +/* Where the main/parent process's pid is logged */ +#ifndef DEFAULT_PIDLOG +#define DEFAULT_PIDLOG DEFAULT_REL_RUNTIMEDIR "/httpd.pid" +#endif + +/* + * Interval, in microseconds, between scoreboard maintenance. + */ +#ifndef SCOREBOARD_MAINTENANCE_INTERVAL +#define SCOREBOARD_MAINTENANCE_INTERVAL 1000000 +#endif + +/* Number of requests to try to handle in a single process. If <= 0, + * the children don't die off. + */ +#ifndef DEFAULT_MAX_REQUESTS_PER_CHILD +#define DEFAULT_MAX_REQUESTS_PER_CHILD 10000 +#endif + +#endif /* AP_MPM_DEFAULT_H */ diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/perchild.c b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/perchild.c new file mode 100644 index 00000000..8b205beb --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/perchild/perchild.c @@ -0,0 +1,2045 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "apr_hash.h" +#include "apr_strings.h" +#include "apr_pools.h" +#include "apr_portable.h" +#include "apr_file_io.h" +#include "apr_signal.h" + +#define APR_WANT_IOVEC +#include "apr_want.h" + +#if APR_HAVE_UNISTD_H +#include <unistd.h> +#endif +#if APR_HAVE_SYS_SOCKET_H +#include <sys/socket.h> +#endif + +#if !APR_HAS_THREADS +#error The perchild MPM requires APR threads, but they are unavailable. +#endif + +#define CORE_PRIVATE + +#include "ap_config.h" +#include "httpd.h" +#include "http_main.h" +#include "http_log.h" +#include "http_config.h" /* for read_config */ +#include "http_core.h" /* for get_remote_host */ +#include "http_protocol.h" +#include "http_connection.h" +#include "ap_mpm.h" +#include "unixd.h" +#include "mpm_common.h" +#include "ap_listen.h" +#include "mpm_default.h" +#include "mpm.h" +#include "scoreboard.h" +#include "util_filter.h" +#include "apr_poll.h" + +#ifdef HAVE_POLL_H +#include <poll.h> +#endif +#ifdef HAVE_SYS_POLL_H +#include <sys/poll.h> +#endif + +/* ### should be APR-ized */ +#include <grp.h> +#include <pwd.h> +#include <sys/stat.h> +#include <sys/un.h> +#include <setjmp.h> +#ifdef HAVE_SYS_PROCESSOR_H +#include <sys/processor.h> /* for bindprocessor() */ +#endif + +/* + * Define some magic numbers that we use for the state of the incomming + * request. These must be < 0 so they don't collide with a file descriptor. + */ +#define AP_PERCHILD_THISCHILD -1 +#define AP_PERCHILD_OTHERCHILD -2 + +/* Limit on the threads per process. Clients will be locked out if more than + * this * server_limit are needed. + * + * We keep this for one reason it keeps the size of the scoreboard file small + * enough that we can read the whole thing without worrying too much about + * the overhead. + */ +#ifndef DEFAULT_THREAD_LIMIT +#define DEFAULT_THREAD_LIMIT 64 +#endif + +/* Admin can't tune ThreadLimit beyond MAX_THREAD_LIMIT. We want + * some sort of compile-time limit to help catch typos. + */ +#ifndef MAX_THREAD_LIMIT +#define MAX_THREAD_LIMIT 20000 +#endif + +/* Limit on the total --- clients will be locked out if more servers than + * this are needed. It is intended solely to keep the server from crashing + * when things get out of hand. + * + * We keep a hard maximum number of servers, for two reasons --- first off, + * in case something goes seriously wrong, we want to stop the fork bomb + * short of actually crashing the machine we're running on by filling some + * kernel table. Secondly, it keeps the size of the scoreboard file small + * enough that we can read the whole thing without worrying too much about + * the overhead. + */ +#ifndef DEFAULT_SERVER_LIMIT +#define DEFAULT_SERVER_LIMIT 8 +#endif + +/* Admin can't tune ServerLimit beyond MAX_SERVER_LIMIT. We want + * some sort of compile-time limit to help catch typos. + */ +#ifndef MAX_SERVER_LIMIT +#define MAX_SERVER_LIMIT 20000 +#endif + +/* + * Actual definitions of config globals + */ + +static int threads_to_start = 0; /* Worker threads per child */ +static int min_spare_threads = 0; +static int max_spare_threads = 0; +static int max_threads = 0; +static int server_limit = DEFAULT_SERVER_LIMIT; +static int first_server_limit; +static int thread_limit = DEFAULT_THREAD_LIMIT; +static int first_thread_limit; +static int changed_limit_at_restart; +static int num_daemons = 0; +static int curr_child_num = 0; +static int workers_may_exit = 0; +static int requests_this_child; +static int num_listensocks = 0; +static ap_pod_t *pod; +static jmp_buf jmpbuffer; + +struct child_info_t { + uid_t uid; + gid_t gid; + int input; /* The socket descriptor */ + int output; /* The socket descriptor */ +}; + +typedef struct { + const char *sockname; /* The base name for the socket */ + const char *fullsockname; /* socket base name + extension */ + int input; /* The socket descriptor */ + int output; /* The socket descriptor */ +} perchild_server_conf; + +typedef struct child_info_t child_info_t; + +/* Tables used to determine the user and group each child process should + * run as. The hash table is used to correlate a server name with a child + * process. + */ +static child_info_t *child_info_table; +static int *thread_socket_table; +struct ap_ctable *ap_child_table; + +/* + * The max child slot ever assigned, preserved across restarts. Necessary + * to deal with NumServers changes across AP_SIG_GRACEFUL restarts. We + * use this value to optimize routines that have to scan the entire child + * table. + * + * XXX - It might not be worth keeping this code in. There aren't very + * many child processes in this MPM. + */ +int ap_max_daemons_limit = -1; +int ap_threads_per_child; /* XXX not part of API! axe it! */ + +module AP_MODULE_DECLARE_DATA mpm_perchild_module; + +static apr_file_t *pipe_of_death_in = NULL; +static apr_file_t *pipe_of_death_out = NULL; +static apr_thread_mutex_t *pipe_of_death_mutex; + +/* *Non*-shared http_main globals... */ + +server_rec *ap_server_conf; + +/* one_process --- debugging mode variable; can be set from the command line + * with the -X flag. If set, this gets you the child_main loop running + * in the process which originally started up (no detach, no make_child), + * which is a pretty nice debugging environment. (You'll get a SIGHUP + * early in standalone_main; just continue through. This is the server + * trying to kill off any child processes which it might have lying + * around --- Apache doesn't keep track of their pids, it just sends + * SIGHUP to the process group, ignoring it in the root process. + * Continue through and you'll be fine.). + */ + +static int one_process = 0; + +#ifdef DEBUG_SIGSTOP +int raise_sigstop_flags; +#endif + +static apr_pool_t *pconf; /* Pool for config stuff */ +static apr_pool_t *pchild; /* Pool for httpd child stuff */ +static apr_pool_t *thread_pool_parent; /* Parent of per-thread pools */ +static apr_thread_mutex_t *thread_pool_parent_mutex; + +static int child_num; +static unsigned int my_pid; /* Linux getpid() doesn't work except in + main thread. Use this instead */ +/* Keep track of the number of worker threads currently active */ +static int worker_thread_count; +static apr_thread_mutex_t *worker_thread_count_mutex; +static int *worker_thread_free_ids; +static apr_threadattr_t *worker_thread_attr; + +/* Keep track of the number of idle worker threads */ +static int idle_thread_count; +static apr_thread_mutex_t *idle_thread_count_mutex; + +/* Locks for accept serialization */ +#ifdef NO_SERIALIZED_ACCEPT +#define SAFE_ACCEPT(stmt) APR_SUCCESS +#else +#define SAFE_ACCEPT(stmt) (stmt) +static apr_proc_mutex_t *process_accept_mutex; +#endif /* NO_SERIALIZED_ACCEPT */ +static apr_thread_mutex_t *thread_accept_mutex; + +AP_DECLARE(apr_status_t) ap_mpm_query(int query_code, int *result) +{ + switch(query_code){ + case AP_MPMQ_MAX_DAEMON_USED: + *result = ap_max_daemons_limit; + return APR_SUCCESS; + case AP_MPMQ_IS_THREADED: + *result = AP_MPMQ_DYNAMIC; + return APR_SUCCESS; + case AP_MPMQ_IS_FORKED: + *result = AP_MPMQ_STATIC; + return APR_SUCCESS; + case AP_MPMQ_HARD_LIMIT_DAEMONS: + *result = server_limit; + return APR_SUCCESS; + case AP_MPMQ_HARD_LIMIT_THREADS: + *result = thread_limit; + return APR_SUCCESS; + case AP_MPMQ_MAX_THREADS: + *result = max_threads; + return APR_SUCCESS; + case AP_MPMQ_MIN_SPARE_DAEMONS: + *result = 0; + return APR_SUCCESS; + case AP_MPMQ_MIN_SPARE_THREADS: + *result = min_spare_threads; + return APR_SUCCESS; + case AP_MPMQ_MAX_SPARE_DAEMONS: + *result = 0; + return APR_SUCCESS; + case AP_MPMQ_MAX_SPARE_THREADS: + *result = max_spare_threads; + return APR_SUCCESS; + case AP_MPMQ_MAX_REQUESTS_DAEMON: + *result = ap_max_requests_per_child; + return APR_SUCCESS; + case AP_MPMQ_MAX_DAEMONS: + *result = num_daemons; + return APR_SUCCESS; + } + return APR_ENOTIMPL; +} + +/* a clean exit from a child with proper cleanup */ +static void clean_child_exit(int code) +{ + if (pchild) { + apr_pool_destroy(pchild); + } + exit(code); +} + +static void just_die(int sig) +{ + clean_child_exit(0); +} + +/***************************************************************** + * Connection structures and accounting... + */ + +/* volatile just in case */ +static int volatile shutdown_pending; +static int volatile restart_pending; +static int volatile is_graceful; +static int volatile child_fatal; +/* we don't currently track ap_my_generation, but mod_status + * references it so it must be defined */ +ap_generation_t volatile ap_my_generation=0; + +/* + * ap_start_shutdown() and ap_start_restart(), below, are a first stab at + * functions to initiate shutdown or restart without relying on signals. + * Previously this was initiated in sig_term() and restart() signal handlers, + * but we want to be able to start a shutdown/restart from other sources -- + * e.g. on Win32, from the service manager. Now the service manager can + * call ap_start_shutdown() or ap_start_restart() as appropiate. Note that + * these functions can also be called by the child processes, since global + * variables are no longer used to pass on the required action to the parent. + * + * These should only be called from the parent process itself, since the + * parent process will use the shutdown_pending and restart_pending variables + * to determine whether to shutdown or restart. The child process should + * call signal_parent() directly to tell the parent to die -- this will + * cause neither of those variable to be set, which the parent will + * assume means something serious is wrong (which it will be, for the + * child to force an exit) and so do an exit anyway. + */ + +static void ap_start_shutdown(void) +{ + if (shutdown_pending == 1) { + /* Um, is this _probably_ not an error, if the user has + * tried to do a shutdown twice quickly, so we won't + * worry about reporting it. + */ + return; + } + shutdown_pending = 1; +} + +/* do a graceful restart if graceful == 1 */ +static void ap_start_restart(int graceful) +{ + + if (restart_pending == 1) { + /* Probably not an error - don't bother reporting it */ + return; + } + restart_pending = 1; + is_graceful = graceful; +} + +static void sig_term(int sig) +{ + ap_start_shutdown(); +} + +static void restart(int sig) +{ +#ifndef WIN32 + ap_start_restart(sig == AP_SIG_GRACEFUL); +#else + ap_start_restart(1); +#endif +} + +static void set_signals(void) +{ +#ifndef NO_USE_SIGACTION + struct sigaction sa; +#endif + + if (!one_process) { + ap_fatal_signal_setup(ap_server_conf, pconf); + } + +#ifndef NO_USE_SIGACTION + sigemptyset(&sa.sa_mask); + sa.sa_flags = 0; + + sa.sa_handler = sig_term; + if (sigaction(SIGTERM, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGTERM)"); +#ifdef SIGINT + if (sigaction(SIGINT, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGINT)"); +#endif +#ifdef SIGXCPU + sa.sa_handler = SIG_DFL; + if (sigaction(SIGXCPU, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGXCPU)"); +#endif +#ifdef SIGXFSZ + sa.sa_handler = SIG_DFL; + if (sigaction(SIGXFSZ, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGXFSZ)"); +#endif +#ifdef SIGPIPE + sa.sa_handler = SIG_IGN; + if (sigaction(SIGPIPE, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGPIPE)"); +#endif + + /* we want to ignore HUPs and AP_SIG_GRACEFUL while we're busy + * processing one */ + sigaddset(&sa.sa_mask, SIGHUP); + sigaddset(&sa.sa_mask, AP_SIG_GRACEFUL); + sa.sa_handler = restart; + if (sigaction(SIGHUP, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGHUP)"); + if (sigaction(AP_SIG_GRACEFUL, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(" AP_SIG_GRACEFUL_STRING ")"); +#else + if (!one_process) { +#ifdef SIGXCPU + apr_signal(SIGXCPU, SIG_DFL); +#endif /* SIGXCPU */ +#ifdef SIGXFSZ + apr_signal(SIGXFSZ, SIG_DFL); +#endif /* SIGXFSZ */ + } + + apr_signal(SIGTERM, sig_term); +#ifdef SIGHUP + apr_signal(SIGHUP, restart); +#endif /* SIGHUP */ +#ifdef AP_SIG_GRACEFUL + apr_signal(AP_SIG_GRACEFUL, restart); +#endif /* AP_SIG_GRACEFUL */ +#ifdef SIGPIPE + apr_signal(SIGPIPE, SIG_IGN); +#endif /* SIGPIPE */ + +#endif +} + +/***************************************************************** + * Here follows a long bunch of generic server bookkeeping stuff... + */ + +int ap_graceful_stop_signalled(void) +{ + /* XXX - Does this really work? - Manoj */ + return is_graceful; +} + +/***************************************************************** + * Child process main loop. + */ + +static void process_socket(apr_pool_t *p, apr_socket_t *sock, long conn_id, + apr_bucket_alloc_t *bucket_alloc) +{ + conn_rec *current_conn; + int csd; + apr_status_t rv; + int thread_num = conn_id % thread_limit; + ap_sb_handle_t *sbh; + + if ((rv = apr_os_sock_get(&csd, sock)) != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ERR, rv, NULL, "apr_os_sock_get"); + } + + if (thread_socket_table[thread_num] < 0) { + ap_sock_disable_nagle(sock); + } + + ap_create_sb_handle(&sbh, p, conn_id / thread_limit, thread_num); + current_conn = ap_run_create_connection(p, ap_server_conf, sock, conn_id, + sbh, bucket_alloc); + if (current_conn) { + ap_process_connection(current_conn, sock); + ap_lingering_close(current_conn); + } +} + +static int perchild_process_connection(conn_rec *c) +{ + ap_filter_t *f; + apr_bucket_brigade *bb; + core_net_rec *net; + + apr_pool_userdata_get((void **)&bb, "PERCHILD_SOCKETS", c->pool); + if (bb != NULL) { + for (f = c->output_filters; f != NULL; f = f->next) { + if (!strcmp(f->frec->name, "core")) { + break; + } + } + if (f != NULL) { + net = f->ctx; + net->in_ctx = apr_palloc(c->pool, sizeof(*net->in_ctx)); + net->in_ctx->b = bb; + } + } + return DECLINED; +} + + +static void *worker_thread(apr_thread_t *, void *); + +/* Starts a thread as long as we're below max_threads */ +static int start_thread(void) +{ + apr_thread_t *thread; + int rc; + + apr_thread_mutex_lock(worker_thread_count_mutex); + if (worker_thread_count < max_threads - 1) { + rc = apr_thread_create(&thread, worker_thread_attr, worker_thread, + &worker_thread_free_ids[worker_thread_count], pchild); + if (rc != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ALERT, rc, ap_server_conf, + "apr_thread_create: unable to create worker thread"); + /* In case system resources are maxxed out, we don't want + Apache running away with the CPU trying to fork over and + over and over again if we exit. */ + sleep(10); + workers_may_exit = 1; + apr_thread_mutex_unlock(worker_thread_count_mutex); + return 0; + } + else { + worker_thread_count++; + } + } + else { + static int reported = 0; + + if (!reported) { + ap_log_error(APLOG_MARK, APLOG_ERR, 0, + ap_server_conf, + "server reached MaxThreadsPerChild setting, " + "consider raising the MaxThreadsPerChild or " + "NumServers settings"); + reported = 1; + } + apr_thread_mutex_unlock(worker_thread_count_mutex); + return 0; + } + apr_thread_mutex_unlock(worker_thread_count_mutex); + return 1; + +} + +/* Sets workers_may_exit if we received a character on the pipe_of_death */ +static apr_status_t check_pipe_of_death(void **csd, ap_listen_rec *lr, + apr_pool_t *ptrans) +{ + apr_thread_mutex_lock(pipe_of_death_mutex); + if (!workers_may_exit) { + int ret; + char pipe_read_char; + apr_size_t n = 1; + + ret = apr_recv(lr->sd, &pipe_read_char, &n); + if (APR_STATUS_IS_EAGAIN(ret)) { + /* It lost the lottery. It must continue to suffer + * through a life of servitude. */ + } + else { + /* It won the lottery (or something else is very + * wrong). Embrace death with open arms. */ + workers_may_exit = 1; + } + } + apr_thread_mutex_unlock(pipe_of_death_mutex); + return APR_SUCCESS; +} + +static apr_status_t receive_from_other_child(void **csd, ap_listen_rec *lr, + apr_pool_t *ptrans) +{ + struct msghdr msg; + struct cmsghdr *cmsg; + char buffer[HUGE_STRING_LEN * 2], *headers, *body; + int headerslen, bodylen; + struct iovec iov; + int ret, dp; + apr_os_sock_t sd; + apr_bucket_alloc_t *alloc = apr_bucket_alloc_create(ptrans); + apr_bucket_brigade *bb = apr_brigade_create(ptrans, alloc); + apr_bucket *bucket; + + apr_os_sock_get(&sd, lr->sd); + + iov.iov_base = buffer; + iov.iov_len = sizeof(buffer); + + msg.msg_name = NULL; + msg.msg_namelen = 0; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + + cmsg = apr_palloc(ptrans, sizeof(*cmsg) + sizeof(sd)); + cmsg->cmsg_len = sizeof(*cmsg) + sizeof(sd); + msg.msg_control = cmsg; + msg.msg_controllen = cmsg->cmsg_len; + + ret = recvmsg(sd, &msg, 0); + + memcpy(&dp, CMSG_DATA(cmsg), sizeof(dp)); + + *csd = NULL; /* tell apr_os_sock_put() to allocate new apr_socket_t */ + apr_os_sock_put((apr_socket_t **)csd, &dp, ptrans); + + bucket = apr_bucket_eos_create(alloc); + APR_BRIGADE_INSERT_HEAD(bb, bucket); + bucket = apr_bucket_socket_create(*csd, alloc); + APR_BRIGADE_INSERT_HEAD(bb, bucket); + + body = strchr(iov.iov_base, 0); + if (!body) { + return 1; + } + + body++; + bodylen = strlen(body); + + headers = iov.iov_base; + headerslen = body - headers; + + bucket = apr_bucket_heap_create(body, bodylen, NULL, alloc); + APR_BRIGADE_INSERT_HEAD(bb, bucket); + bucket = apr_bucket_heap_create(headers, headerslen, NULL, alloc); + APR_BRIGADE_INSERT_HEAD(bb, bucket); + + apr_pool_userdata_set(bb, "PERCHILD_SOCKETS", NULL, ptrans); + + return 0; +} + +/* idle_thread_count should be incremented before starting a worker_thread */ + +static void *worker_thread(apr_thread_t *thd, void *arg) +{ + void *csd; + apr_pool_t *tpool; /* Pool for this thread */ + apr_pool_t *ptrans; /* Pool for per-transaction stuff */ + volatile int thread_just_started = 1; + int srv; + int thread_num = *((int *) arg); + long conn_id = child_num * thread_limit + thread_num; + apr_pollfd_t *pollset; + apr_status_t rv; + ap_listen_rec *lr, *last_lr = ap_listeners; + int n; + apr_bucket_alloc_t *bucket_alloc; + + apr_thread_mutex_lock(thread_pool_parent_mutex); + apr_pool_create(&tpool, thread_pool_parent); + apr_thread_mutex_unlock(thread_pool_parent_mutex); + apr_pool_create(&ptrans, tpool); + + (void) ap_update_child_status_from_indexes(child_num, thread_num, + SERVER_STARTING, + (request_rec *) NULL); + + bucket_alloc = apr_bucket_alloc_create(apr_thread_pool_get(thd)); + + apr_poll_setup(&pollset, num_listensocks, tpool); + for(lr = ap_listeners; lr != NULL; lr = lr->next) { + int fd; + apr_poll_socket_add(pollset, lr->sd, APR_POLLIN); + + apr_os_sock_get(&fd, lr->sd); + } + + while (!workers_may_exit) { + workers_may_exit |= ((ap_max_requests_per_child != 0) + && (requests_this_child <= 0)); + if (workers_may_exit) break; + if (!thread_just_started) { + apr_thread_mutex_lock(idle_thread_count_mutex); + if (idle_thread_count < max_spare_threads) { + idle_thread_count++; + apr_thread_mutex_unlock(idle_thread_count_mutex); + } + else { + apr_thread_mutex_unlock(idle_thread_count_mutex); + break; + } + } + else { + thread_just_started = 0; + } + + (void) ap_update_child_status_from_indexes(child_num, thread_num, + SERVER_READY, + (request_rec *) NULL); + + apr_thread_mutex_lock(thread_accept_mutex); + if (workers_may_exit) { + apr_thread_mutex_unlock(thread_accept_mutex); + break; + } + if ((rv = SAFE_ACCEPT(apr_proc_mutex_lock(process_accept_mutex))) + != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "apr_proc_mutex_lock failed. Attempting to shutdown " + "process gracefully."); + workers_may_exit = 1; + } + + while (!workers_may_exit) { + apr_int16_t event; + srv = apr_poll(pollset, num_listensocks, &n, -1); + + if (srv != APR_SUCCESS) { + if (APR_STATUS_IS_EINTR(srv)) { + continue; + } + + /* apr_poll() will only return errors in catastrophic + * circumstances. Let's try exiting gracefully, for now. */ + ap_log_error(APLOG_MARK, APLOG_ERR, srv, (const server_rec *) + ap_server_conf, "apr_poll: (listen)"); + workers_may_exit = 1; + } + if (workers_may_exit) break; + + /* find a listener */ + lr = last_lr; + do { + lr = lr->next; + if (lr == NULL) { + lr = ap_listeners; + } + /* XXX: Should we check for POLLERR? */ + apr_poll_revents_get(&event, lr->sd, pollset); + if (event & (APR_POLLIN)) { + last_lr = lr; + goto got_fd; + } + } while (lr != last_lr); + } + got_fd: + if (!workers_may_exit) { + rv = lr->accept_func(&csd, lr, ptrans); + if (rv == APR_EGENERAL) { + /* E[NM]FILE, ENOMEM, etc */ + workers_may_exit = 1; + } + if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(process_accept_mutex))) + != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "apr_proc_mutex_unlock failed. Attempting to shutdown " + "process gracefully."); + workers_may_exit = 1; + } + apr_thread_mutex_unlock(thread_accept_mutex); + apr_thread_mutex_lock(idle_thread_count_mutex); + if (idle_thread_count > min_spare_threads) { + idle_thread_count--; + } + else { + if (!start_thread()) { + idle_thread_count--; + } + } + apr_thread_mutex_unlock(idle_thread_count_mutex); + if (setjmp(jmpbuffer) != 1) { + process_socket(ptrans, csd, conn_id, bucket_alloc); + } + else { + thread_socket_table[thread_num] = AP_PERCHILD_THISCHILD; + } + requests_this_child--; + } + else { + if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(process_accept_mutex))) + != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "apr_proc_mutex_unlock failed. Attempting to shutdown " + "process gracefully."); + workers_may_exit = 1; + } + apr_thread_mutex_unlock(thread_accept_mutex); + apr_thread_mutex_lock(idle_thread_count_mutex); + idle_thread_count--; + apr_thread_mutex_unlock(idle_thread_count_mutex); + break; + } + apr_pool_clear(ptrans); + } + + apr_thread_mutex_lock(thread_pool_parent_mutex); + ap_update_child_status_from_indexes(child_num, thread_num, SERVER_DEAD, + (request_rec *) NULL); + apr_pool_destroy(tpool); + apr_thread_mutex_unlock(thread_pool_parent_mutex); + apr_thread_mutex_lock(worker_thread_count_mutex); + worker_thread_count--; + worker_thread_free_ids[worker_thread_count] = thread_num; + if (worker_thread_count == 0) { + /* All the threads have exited, now finish the shutdown process + * by signalling the sigwait thread */ + kill(my_pid, SIGTERM); + } + apr_thread_mutex_unlock(worker_thread_count_mutex); + + apr_bucket_alloc_destroy(bucket_alloc); + + return NULL; +} + + + +/* Set group privileges. + * + * Note that we use the username as set in the config files, rather than + * the lookup of to uid --- the same uid may have multiple passwd entries, + * with different sets of groups for each. + */ + +static int set_group_privs(uid_t uid, gid_t gid) +{ + if (!geteuid()) { + const char *name; + + /* Get username if passed as a uid */ + + struct passwd *ent; + + if ((ent = getpwuid(uid)) == NULL) { + ap_log_error(APLOG_MARK, APLOG_ALERT, errno, NULL, + "getpwuid: couldn't determine user name from uid %u, " + "you probably need to modify the User directive", + (unsigned)uid); + return -1; + } + + name = ent->pw_name; + + /* + * Set the GID before initgroups(), since on some platforms + * setgid() is known to zap the group list. + */ + if (setgid(gid) == -1) { + ap_log_error(APLOG_MARK, APLOG_ALERT, errno, NULL, + "setgid: unable to set group id to Group %u", + (unsigned)gid); + return -1; + } + + /* Reset `groups' attributes. */ + + if (initgroups(name, gid) == -1) { + ap_log_error(APLOG_MARK, APLOG_ALERT, errno, NULL, + "initgroups: unable to set groups for User %s " + "and Group %u", name, (unsigned)gid); + return -1; + } + } + return 0; +} + + +static int perchild_setup_child(int childnum) +{ + child_info_t *ug = &child_info_table[childnum]; + + if (ug->uid == -1 && ug->gid == -1) { + return unixd_setup_child(); + } + if (set_group_privs(ug->uid, ug->gid)) { + return -1; + } + /* Only try to switch if we're running as root */ + if (!geteuid() + && ( +#ifdef _OSD_POSIX + os_init_job_environment(server_conf, unixd_config.user_name, + one_process) != 0 || +#endif + setuid(ug->uid) == -1)) { + ap_log_error(APLOG_MARK, APLOG_ALERT, errno, NULL, + "setuid: unable to change to uid: %ld", + (long) ug->uid); + return -1; + } + return 0; +} + +static int check_signal(int signum) +{ + switch (signum) { + case SIGTERM: + case SIGINT: + just_die(signum); + return 1; + } + return 0; +} + +typedef struct perchild_header { + char *headers; + apr_pool_t *p; +} perchild_header; + +/* Send a single HTTP header field to the client. Note that this function + * is used in calls to table_do(), so their interfaces are co-dependent. + * In other words, don't change this one without checking table_do in alloc.c. + * It returns true unless there was a write error of some kind. + */ +static int perchild_header_field(perchild_header *h, + const char *fieldname, const char *fieldval) +{ + apr_pstrcat(h->p, h->headers, fieldname, ": ", fieldval, CRLF, NULL); + return 1; +} + + +static void child_main(int child_num_arg) +{ + int i; + apr_status_t rv; + apr_socket_t *sock = NULL; + ap_listen_rec *lr; + + my_pid = getpid(); + ap_fatal_signal_child_setup(ap_server_conf); + child_num = child_num_arg; + apr_pool_create(&pchild, pconf); + + for (lr = ap_listeners ; lr->next != NULL; lr = lr->next) { + continue; + } + + apr_os_sock_put(&sock, &child_info_table[child_num].input, pconf); + lr->next = apr_palloc(pconf, sizeof(*lr)); + lr->next->sd = sock; + lr->next->active = 1; + lr->next->accept_func = receive_from_other_child; + lr->next->next = NULL; + lr = lr->next; + num_listensocks++; + + /*stuff to do before we switch id's, so we have permissions.*/ + + rv = SAFE_ACCEPT(apr_proc_mutex_child_init(&process_accept_mutex, + ap_lock_fname, pchild)); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "Couldn't initialize cross-process lock in child"); + clean_child_exit(APEXIT_CHILDFATAL); + } + + if (perchild_setup_child(child_num)) { + clean_child_exit(APEXIT_CHILDFATAL); + } + + ap_run_child_init(pchild, ap_server_conf); + + /*done with init critical section */ + + apr_setup_signal_thread(); + + requests_this_child = ap_max_requests_per_child; + + + /* Setup worker threads */ + + if (threads_to_start > max_threads) { + threads_to_start = max_threads; + } + idle_thread_count = threads_to_start; + worker_thread_count = 0; + worker_thread_free_ids = (int *)apr_pcalloc(pchild, thread_limit * sizeof(int)); + for (i = 0; i < max_threads; i++) { + worker_thread_free_ids[i] = i; + } + apr_pool_create(&thread_pool_parent, pchild); + apr_thread_mutex_create(&thread_pool_parent_mutex, + APR_THREAD_MUTEX_DEFAULT, pchild); + apr_thread_mutex_create(&idle_thread_count_mutex, + APR_THREAD_MUTEX_DEFAULT, pchild); + apr_thread_mutex_create(&worker_thread_count_mutex, + APR_THREAD_MUTEX_DEFAULT, pchild); + apr_thread_mutex_create(&pipe_of_death_mutex, + APR_THREAD_MUTEX_DEFAULT, pchild); + apr_thread_mutex_create(&thread_accept_mutex, + APR_THREAD_MUTEX_DEFAULT, pchild); + + apr_threadattr_create(&worker_thread_attr, pchild); + apr_threadattr_detach_set(worker_thread_attr, 1); + + /* We are creating worker threads right now */ + for (i=0; i < threads_to_start; i++) { + /* start_thread shouldn't fail here */ + if (!start_thread()) { + break; + } + } + + apr_signal_thread(check_signal); +} + +static int make_child(server_rec *s, int slot) +{ + int pid; + + if (slot + 1 > ap_max_daemons_limit) { + ap_max_daemons_limit = slot + 1; + } + + if (one_process) { + set_signals(); + ap_child_table[slot].pid = getpid(); + ap_child_table[slot].status = SERVER_ALIVE; + child_main(slot); + } + (void) ap_update_child_status_from_indexes(slot, 0, SERVER_STARTING, + (request_rec *) NULL); + + if ((pid = fork()) == -1) { + ap_log_error(APLOG_MARK, APLOG_ERR, errno, s, + "fork: Unable to fork new process"); + /* In case system resources are maxxed out, we don't want + * Apache running away with the CPU trying to fork over and + * over and over again. */ + sleep(10); + + return -1; + } + + if (!pid) { +#ifdef HAVE_BINDPROCESSOR + /* By default, AIX binds to a single processor. This bit unbinds + * children which will then bind to another CPU. + */ + int status = bindprocessor(BINDPROCESS, (int)getpid(), + PROCESSOR_CLASS_ANY); + if (status != OK) { + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, + ap_server_conf, "processor unbind failed %d", status); + } +#endif + + RAISE_SIGSTOP(MAKE_CHILD); + + /* XXX - For an unthreaded server, a signal handler will be necessary + * apr_signal(SIGTERM, just_die); + */ + child_main(slot); + clean_child_exit(0); + } + /* else */ + ap_child_table[slot].pid = pid; + ap_child_table[slot].status = SERVER_ALIVE; + + return 0; +} + +/* start up a bunch of children */ +static int startup_children(int number_to_start) +{ + int i; + + for (i = 0; number_to_start && i < num_daemons; ++i) { + if (ap_child_table[i].pid) { + continue; + } + if (make_child(ap_server_conf, i) < 0) { + break; + } + --number_to_start; + } + return number_to_start; +} + + +/* + * spawn_rate is the number of children that will be spawned on the + * next maintenance cycle if there aren't enough servers. It is + * doubled up to MAX_SPAWN_RATE, and reset only when a cycle goes by + * without the need to spawn. + */ +static int spawn_rate = 1; +#ifndef MAX_SPAWN_RATE +#define MAX_SPAWN_RATE (32) +#endif +static int hold_off_on_exponential_spawning; + +static void perform_child_maintenance(void) +{ + int i; + int free_length; + int free_slots[MAX_SPAWN_RATE]; + int last_non_dead = -1; + + /* initialize the free_list */ + free_length = 0; + + for (i = 0; i < num_daemons; ++i) { + if (ap_child_table[i].pid == 0) { + if (free_length < spawn_rate) { + free_slots[free_length] = i; + ++free_length; + } + } + else { + last_non_dead = i; + } + + if (i >= ap_max_daemons_limit && free_length >= spawn_rate) { + break; + } + } + ap_max_daemons_limit = last_non_dead + 1; + + if (free_length > 0) { + for (i = 0; i < free_length; ++i) { + make_child(ap_server_conf, free_slots[i]); + } + /* the next time around we want to spawn twice as many if this + * wasn't good enough, but not if we've just done a graceful + */ + if (hold_off_on_exponential_spawning) { + --hold_off_on_exponential_spawning; + } + else if (spawn_rate < MAX_SPAWN_RATE) { + spawn_rate *= 2; + } + } + else { + spawn_rate = 1; + } +} + +static void server_main_loop(int remaining_children_to_start) +{ + int child_slot; + apr_exit_why_e exitwhy; + int status; + apr_proc_t pid; + int i; + + while (!restart_pending && !shutdown_pending) { + ap_wait_or_timeout(&exitwhy, &status, &pid, pconf); + + if (pid.pid != -1) { + if (ap_process_child_status(&pid, exitwhy, status) + == APEXIT_CHILDFATAL) { + shutdown_pending = 1; + child_fatal = 1; + return; + } + /* non-fatal death... note that it's gone in the child table and + * clean out the status table. */ + child_slot = -1; + for (i = 0; i < ap_max_daemons_limit; ++i) { + if (ap_child_table[i].pid == pid.pid) { + child_slot = i; + break; + } + } + if (child_slot >= 0) { + ap_child_table[child_slot].pid = 0; + ap_update_child_status_from_indexes(child_slot, i, SERVER_DEAD, + (request_rec *) NULL); + + + if (remaining_children_to_start + && child_slot < num_daemons) { + /* we're still doing a 1-for-1 replacement of dead + * children with new children + */ + make_child(ap_server_conf, child_slot); + --remaining_children_to_start; + } +#if APR_HAS_OTHER_CHILD + } + else if (apr_proc_other_child_read(&pid, status) == 0) { + /* handled */ +#endif + } + else if (is_graceful) { + /* Great, we've probably just lost a slot in the + * child table. Somehow we don't know about this + * child. + */ + ap_log_error(APLOG_MARK, APLOG_WARNING, 0, + ap_server_conf, + "long lost child came home! (pid %ld)", + (long)pid.pid); + } + /* Don't perform idle maintenance when a child dies, + * only do it when there's a timeout. Remember only a + * finite number of children can die, and it's pretty + * pathological for a lot to die suddenly. + */ + continue; + } + else if (remaining_children_to_start) { + /* we hit a 1 second timeout in which none of the previous + * generation of children needed to be reaped... so assume + * they're all done, and pick up the slack if any is left. + */ + remaining_children_to_start = \ + startup_children(remaining_children_to_start); + /* In any event we really shouldn't do the code below because + * few of the servers we just started are in the IDLE state + * yet, so we'd mistakenly create an extra server. + */ + continue; + } + + perform_child_maintenance(); + } +} + +int ap_mpm_run(apr_pool_t *_pconf, apr_pool_t *plog, server_rec *s) +{ + int remaining_children_to_start; + int i; + apr_status_t rv; + apr_size_t one = 1; + ap_listen_rec *lr; + apr_socket_t *sock = NULL; + int fd; + + ap_log_pid(pconf, ap_pid_fname); + + first_server_limit = server_limit; + first_thread_limit = thread_limit; + if (changed_limit_at_restart) { + ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, + "WARNING: Attempt to change ServerLimit or ThreadLimit " + "ignored during restart"); + changed_limit_at_restart = 0; + } + + ap_server_conf = s; + + if ((ap_accept_lock_mech == APR_LOCK_SYSVSEM) || + (ap_accept_lock_mech == APR_LOCK_POSIXSEM)) { + ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, + "Server configured for an accept lock mechanism that " + "cannot be used with perchild. Falling back to FCNTL."); + ap_accept_lock_mech = APR_LOCK_FCNTL; + } + + /* Initialize cross-process accept lock */ + ap_lock_fname = apr_psprintf(_pconf, "%s.%u", + ap_server_root_relative(_pconf, ap_lock_fname), + my_pid); + rv = SAFE_ACCEPT(apr_proc_mutex_create(&process_accept_mutex, + ap_lock_fname, ap_accept_lock_mech, + _pconf)); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, s, + "Couldn't create cross-process lock"); + return 1; + } + + if (!is_graceful) { + if (ap_run_pre_mpm(s->process->pool, SB_SHARED) != OK) { + return 1; + } + } + /* Initialize the child table */ + if (!is_graceful) { + for (i = 0; i < server_limit; i++) { + ap_child_table[i].pid = 0; + } + } + + /* We need to put the new listeners at the end of the ap_listeners + * list. If we don't, then the pool will be cleared before the + * open_logs phase is called for the second time, and ap_listeners + * will have only invalid data. If that happens, then the sockets + * that we opened using make_sock() will be lost, and the server + * won't start. + */ + for (lr = ap_listeners ; lr->next != NULL; lr = lr->next) { + continue; + } + + apr_os_file_get(&fd, pipe_of_death_in); + apr_os_sock_put(&sock, &fd, pconf); + lr->next = apr_palloc(pconf, sizeof(*lr)); + lr->next->sd = sock; + lr->next->active = 1; + lr->next->accept_func = check_pipe_of_death; + lr->next->next = NULL; + lr = lr->next; + num_listensocks++; + + set_signals(); + + /* If we're doing a graceful_restart then we're going to see a lot + * of children exiting immediately when we get into the main loop + * below (because we just sent them AP_SIG_GRACEFUL). This happens + * pretty rapidly... and for each one that exits we'll start a new one + * until we reach at least daemons_min_free. But we may be permitted to + * start more than that, so we'll just keep track of how many we're + * supposed to start up without the 1 second penalty between each fork. + */ + remaining_children_to_start = num_daemons; + if (!is_graceful) { + remaining_children_to_start = \ + startup_children(remaining_children_to_start); + } + else { + /* give the system some time to recover before kicking into + * exponential mode */ + hold_off_on_exponential_spawning = 10; + } + + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, + "%s configured -- resuming normal operations", + ap_get_server_version()); + ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf, + "Server built: %s", ap_get_server_built()); +#ifdef AP_MPM_WANT_SET_ACCEPT_LOCK_MECH + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "AcceptMutex: %s (default: %s)", + apr_proc_mutex_name(process_accept_mutex), + apr_proc_mutex_defname()); +#endif + restart_pending = shutdown_pending = 0; + + server_main_loop(remaining_children_to_start); + + if (shutdown_pending) { + /* Time to gracefully shut down: + * Kill child processes, tell them to call child_exit, etc... + */ + if (unixd_killpg(getpgrp(), SIGTERM) < 0) { + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "killpg SIGTERM"); + } + ap_reclaim_child_processes(1); /* Start with SIGTERM */ + + if (!child_fatal) { + /* cleanup pid file on normal shutdown */ + const char *pidfile = NULL; + pidfile = ap_server_root_relative (pconf, ap_pid_fname); + if (pidfile != NULL && unlink(pidfile) == 0) { + ap_log_error(APLOG_MARK, APLOG_INFO, 0, + ap_server_conf, + "removed PID file %s (pid=%ld)", + pidfile, (long)getpid()); + } + + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, + ap_server_conf, "caught SIGTERM, shutting down"); + } + return 1; + } + + /* we've been told to restart */ + apr_signal(SIGHUP, SIG_IGN); + + if (one_process) { + /* not worth thinking about */ + return 1; + } + + if (is_graceful) { + char char_of_death = '!'; + + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, + ap_server_conf, AP_SIG_GRACEFUL_STRING " received. " + "Doing graceful restart"); + + /* This is mostly for debugging... so that we know what is still + * gracefully dealing with existing request. + */ + + for (i = 0; i < num_daemons; ++i) { + if (ap_child_table[i].pid) { + ap_child_table[i].status = SERVER_DYING; + } + } + /* give the children the signal to die */ + for (i = 0; i < num_daemons;) { + if ((rv = apr_file_write(pipe_of_death_out, &char_of_death, + &one)) != APR_SUCCESS) { + if (APR_STATUS_IS_EINTR(rv)) continue; + ap_log_error(APLOG_MARK, APLOG_WARNING, rv, ap_server_conf, + "write pipe_of_death"); + } + i++; + } + } + else { + /* Kill 'em all. Since the child acts the same on the parents SIGTERM + * and a SIGHUP, we may as well use the same signal, because some user + * pthreads are stealing signals from us left and right. + */ + if (unixd_killpg(getpgrp(), SIGTERM) < 0) { + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "killpg SIGTERM"); + } + ap_reclaim_child_processes(1); /* Start with SIGTERM */ + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, + ap_server_conf, "SIGHUP received. Attempting to restart"); + } + return 0; +} + +/* This really should be a post_config hook, but the error log is already + * redirected by that point, so we need to do this in the open_logs phase. + */ +static int perchild_open_logs(apr_pool_t *p, apr_pool_t *plog, apr_pool_t *ptemp, server_rec *s) +{ + apr_status_t rv; + + pconf = p; + ap_server_conf = s; + + if ((num_listensocks = ap_setup_listeners(ap_server_conf)) < 1) { + ap_log_error(APLOG_MARK, APLOG_ALERT|APLOG_STARTUP, 0, + NULL, "no listening sockets available, shutting down"); + return DONE; + } + + ap_log_pid(pconf, ap_pid_fname); + + if ((rv = ap_mpm_pod_open(pconf, &pod))) { + ap_log_error(APLOG_MARK, APLOG_CRIT|APLOG_STARTUP, rv, NULL, + "Could not open pipe-of-death."); + return DONE; + } + + if ((rv = apr_file_pipe_create(&pipe_of_death_in, &pipe_of_death_out, + pconf)) != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ERR, rv, + (const server_rec*) ap_server_conf, + "apr_file_pipe_create (pipe_of_death)"); + exit(1); + } + if ((rv = apr_file_pipe_timeout_set(pipe_of_death_in, 0)) != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ERR, rv, + (const server_rec*) ap_server_conf, + "apr_file_pipe_timeout_set (pipe_of_death)"); + exit(1); + } + + return OK; +} + +static int perchild_pre_config(apr_pool_t *p, apr_pool_t *plog, apr_pool_t *ptemp) +{ + static int restart_num = 0; + int no_detach, debug, foreground; + ap_directive_t *pdir; + int i; + int tmp_server_limit = DEFAULT_SERVER_LIMIT; + int tmp_thread_limit = DEFAULT_THREAD_LIMIT; + apr_status_t rv; + + debug = ap_exists_config_define("DEBUG"); + + if (debug) { + foreground = one_process = 1; + no_detach = 0; + } + else { + one_process = ap_exists_config_define("ONE_PROCESS"); + no_detach = ap_exists_config_define("NO_DETACH"); + foreground = ap_exists_config_define("FOREGROUND"); + } + + /* sigh, want this only the second time around */ + if (restart_num++ == 1) { + is_graceful = 0; + + if (!one_process && !foreground) { + rv = apr_proc_detach(no_detach ? APR_PROC_DETACH_FOREGROUND + : APR_PROC_DETACH_DAEMONIZE); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, NULL, + "apr_proc_detach failed"); + return HTTP_INTERNAL_SERVER_ERROR; + } + } + + my_pid = getpid(); + } + + unixd_pre_config(ptemp); + ap_listen_pre_config(); + num_daemons = DEFAULT_NUM_DAEMON; + threads_to_start = DEFAULT_START_THREAD; + min_spare_threads = DEFAULT_MIN_SPARE_THREAD; + max_spare_threads = DEFAULT_MAX_SPARE_THREAD; + max_threads = thread_limit; + ap_pid_fname = DEFAULT_PIDLOG; + ap_lock_fname = DEFAULT_LOCKFILE; + ap_max_requests_per_child = DEFAULT_MAX_REQUESTS_PER_CHILD; + curr_child_num = 0; +#ifdef AP_MPM_WANT_SET_MAX_MEM_FREE + ap_max_mem_free = APR_ALLOCATOR_MAX_FREE_UNLIMITED; +#endif + + apr_cpystrn(ap_coredump_dir, ap_server_root, sizeof(ap_coredump_dir)); + + /* we need to know ServerLimit and ThreadLimit before we start processing + * the tree because we need to already have allocated child_info_table + */ + for (pdir = ap_conftree; pdir != NULL; pdir = pdir->next) { + if (!strcasecmp(pdir->directive, "ServerLimit")) { + if (atoi(pdir->args) > tmp_server_limit) { + tmp_server_limit = atoi(pdir->args); + if (tmp_server_limit > MAX_SERVER_LIMIT) { + tmp_server_limit = MAX_SERVER_LIMIT; + } + } + } + else if (!strcasecmp(pdir->directive, "ThreadLimit")) { + if (atoi(pdir->args) > tmp_thread_limit) { + tmp_thread_limit = atoi(pdir->args); + if (tmp_thread_limit > MAX_THREAD_LIMIT) { + tmp_thread_limit = MAX_THREAD_LIMIT; + } + } + } + } + + child_info_table = (child_info_t *)apr_pcalloc(p, tmp_server_limit * sizeof(child_info_t)); + for (i = 0; i < tmp_server_limit; i++) { + child_info_table[i].uid = -1; + child_info_table[i].gid = -1; + child_info_table[i].input = -1; + child_info_table[i].output = -1; + } + + return OK; +} + +static int pass_request(request_rec *r) +{ + int rv; + apr_socket_t *thesock = ap_get_module_config(r->connection->conn_config, &core_module); + struct msghdr msg; + struct cmsghdr *cmsg; + int sfd; + struct iovec iov[2]; + conn_rec *c = r->connection; + apr_bucket_brigade *bb = apr_brigade_create(r->pool, c->bucket_alloc); + apr_bucket_brigade *sockbb; + char request_body[HUGE_STRING_LEN] = "\0"; + apr_size_t l = sizeof(request_body); + perchild_header h; + apr_bucket *sockbuck; + perchild_server_conf *sconf = (perchild_server_conf *) + ap_get_module_config(r->server->module_config, + &mpm_perchild_module); + + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "passing request to another child. Vhost: %s, child %d %d", + apr_table_get(r->headers_in, "Host"), child_num, sconf->output); + ap_get_brigade(r->connection->input_filters, bb, AP_MODE_EXHAUSTIVE, APR_NONBLOCK_READ, + 0); + + for (sockbuck = APR_BRIGADE_FIRST(bb); sockbuck != APR_BRIGADE_SENTINEL(bb); + sockbuck = APR_BUCKET_NEXT(sockbuck)) { + if (APR_BUCKET_IS_SOCKET(sockbuck)) { + break; + } + } + + if (!sockbuck) { + } + sockbb = apr_brigade_split(bb, sockbuck); + + if (apr_brigade_flatten(bb, request_body, &l) != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "Unable to flatten brigade, declining request"); + return DECLINED; + } + + apr_os_sock_get(&sfd, thesock); + + h.p = r->pool; + h.headers = apr_pstrcat(h.p, r->the_request, CRLF, "Host: ", r->hostname, + CRLF, NULL); + apr_table_do((int (*) (void *, const char *, const char *)) + perchild_header_field, (void *) &h, r->headers_in, NULL); + h.headers = apr_pstrcat(h.p, h.headers, CRLF, NULL); + + iov[0].iov_base = h.headers; + iov[0].iov_len = strlen(h.headers) + 1; + iov[1].iov_base = request_body; + iov[1].iov_len = l + 1; + + msg.msg_name = NULL; + msg.msg_namelen = 0; + msg.msg_iov = iov; + msg.msg_iovlen = 2; + + cmsg = apr_palloc(r->pool, sizeof(*cmsg) + sizeof(sfd)); + cmsg->cmsg_len = sizeof(*cmsg) + sizeof(sfd); + cmsg->cmsg_level = SOL_SOCKET; + cmsg->cmsg_type = SCM_RIGHTS; + + memcpy(CMSG_DATA(cmsg), &sfd, sizeof(sfd)); + + msg.msg_control = cmsg; + msg.msg_controllen = cmsg->cmsg_len; + + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "Writing message to %d, passing sd: %d", sconf->output, sfd); + + if ((rv = sendmsg(sconf->output, &msg, 0)) == -1) { + apr_pool_destroy(r->pool); + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "Writing message failed %d %d", rv, errno); + return -1; + } + + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "Writing message succeeded %d", rv); + + apr_pool_destroy(r->pool); + return 1; +} + +static char *make_perchild_socket(const char *fullsockname, int sd[2]) +{ + socketpair(PF_UNIX, SOCK_STREAM, 0, sd); + return NULL; +} + +static int perchild_post_config(apr_pool_t *p, apr_pool_t *plog, apr_pool_t *ptemp, server_rec *s) +{ + int i; + server_rec *sr; + perchild_server_conf *sconf; + int def_sd[2]; + + def_sd[0] = -1; + def_sd[1] = -1; + + for (sr = s; sr; sr = sr->next) { + sconf = (perchild_server_conf *)ap_get_module_config(sr->module_config, + &mpm_perchild_module); + + if (sconf->input == -1) { + sconf->fullsockname = apr_pstrcat(sr->process->pool, + sconf->sockname, ".DEFAULT", NULL); + if (def_sd[0] == -1) { + if (!make_perchild_socket(sconf->fullsockname, def_sd)) { + /* log error */ + } + } + sconf->input = def_sd[0]; + sconf->output = def_sd[1]; + } + } + + for (i = 0; i < num_daemons; i++) { + if (child_info_table[i].uid == -1) { + child_info_table[i].input = def_sd[0]; + child_info_table[i].output = def_sd[1]; + } + } + + thread_socket_table = (int *)apr_pcalloc(p, thread_limit * sizeof(int)); + for (i = 0; i < thread_limit; i++) { + thread_socket_table[i] = AP_PERCHILD_THISCHILD; + } + ap_child_table = (ap_ctable *)apr_pcalloc(p, server_limit * sizeof(ap_ctable)); + + return OK; +} + +static int perchild_post_read(request_rec *r) +{ + int thread_num = r->connection->id % thread_limit; + perchild_server_conf *sconf = (perchild_server_conf *) + ap_get_module_config(r->server->module_config, + &mpm_perchild_module); + + if (thread_socket_table[thread_num] != AP_PERCHILD_THISCHILD) { + apr_socket_t *csd = NULL; + + apr_os_sock_put(&csd, &thread_socket_table[thread_num], + r->connection->pool); + ap_sock_disable_nagle(csd); + ap_set_module_config(r->connection->conn_config, &core_module, csd); + return OK; + } + else { + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "Determining if request should be passed. " + "Child Num: %d, SD: %d, sd from table: %d, hostname from server: %s", child_num, + sconf->input, child_info_table[child_num].input, + r->server->server_hostname); + /* sconf is the server config for this vhost, so if our socket + * is not the same that was set in the config, then the request + * needs to be passed to another child. */ + if (sconf->input != child_info_table[child_num].input) { + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "Passing request."); + if (pass_request(r) == -1) { + ap_log_error(APLOG_MARK, APLOG_ERR, 0, + ap_server_conf, "Could not pass request to proper " + "child, request will not be honored."); + } + longjmp(jmpbuffer, 1); + } + return OK; + } + return OK; +} + +static void perchild_hooks(apr_pool_t *p) +{ + /* The perchild open_logs phase must run before the core's, or stderr + * will be redirected to a file, and the messages won't print to the + * console. + */ + static const char *const aszSucc[] = {"core.c", NULL}; + one_process = 0; + + ap_hook_open_logs(perchild_open_logs, NULL, aszSucc, APR_HOOK_MIDDLE); + ap_hook_pre_config(perchild_pre_config, NULL, NULL, APR_HOOK_MIDDLE); + ap_hook_post_config(perchild_post_config, NULL, NULL, APR_HOOK_MIDDLE); + + /* Both of these must be run absolutely first. If this request isn't for + * this server then we need to forward it to the proper child. No sense + * tying up this server running more post_read request hooks if it is + * just going to be forwarded along. The process_connection hook allows + * perchild to receive the passed request correctly, by automatically + * filling in the core_input_filter's ctx pointer. + */ + ap_hook_post_read_request(perchild_post_read, NULL, NULL, + APR_HOOK_REALLY_FIRST); + ap_hook_process_connection(perchild_process_connection, NULL, NULL, + APR_HOOK_REALLY_FIRST); +} + +static const char *set_num_daemons(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + num_daemons = atoi(arg); + if (num_daemons > server_limit) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: NumServers of %d exceeds ServerLimit value " + "of %d servers,", num_daemons, server_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " lowering NumServers to %d. To increase, please " + "see the", server_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " ServerLimit directive."); + num_daemons = server_limit; + } + else if (num_daemons < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require NumServers > 0, setting to 1"); + num_daemons = 1; + } + return NULL; +} + +static const char *set_threads_to_start(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + threads_to_start = atoi(arg); + if (threads_to_start > thread_limit) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: StartThreads of %d exceeds ThreadLimit value" + " of %d threads,", threads_to_start, + thread_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " lowering StartThreads to %d. To increase, please" + " see the", thread_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " ThreadLimit directive."); + } + else if (threads_to_start < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require StartThreads > 0, setting to 1"); + threads_to_start = 1; + } + return NULL; +} + +static const char *set_min_spare_threads(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + min_spare_threads = atoi(arg); + if (min_spare_threads <= 0) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: detected MinSpareThreads set to non-positive."); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "Resetting to 1 to avoid almost certain Apache failure."); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "Please read the documentation."); + min_spare_threads = 1; + } + + return NULL; +} + +static const char *set_max_spare_threads(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + max_spare_threads = atoi(arg); + if (max_spare_threads >= thread_limit) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: detected MinSpareThreads set higher than"); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "ThreadLimit. Resetting to %d", thread_limit); + max_spare_threads = thread_limit; + } + return NULL; +} + +static const char *set_max_threads(cmd_parms *cmd, void *dummy, const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + max_threads = atoi(arg); + if (max_threads > thread_limit) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: detected MaxThreadsPerChild set higher than"); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "ThreadLimit. Resetting to %d", thread_limit); + max_threads = thread_limit; + } + return NULL; +} + +static const char *set_child_per_uid(cmd_parms *cmd, void *dummy, const char *u, + const char *g, const char *num) +{ + int i; + int max_this_time = atoi(num) + curr_child_num; + + + for (i = curr_child_num; i < max_this_time; i++, curr_child_num++) { + if (i > num_daemons) { + return "Trying to use more child ID's than NumServers. Increase " + "NumServers in your config file."; + } + + child_info_table[i].uid = ap_uname2id(u); + child_info_table[i].gid = ap_gname2id(g); + +#ifndef BIG_SECURITY_HOLE + if (child_info_table[i].uid == 0 || child_info_table[i].gid == 0) { + return "Assigning root user/group to a child."; + } +#endif + } + return NULL; +} + +static const char *assign_childuid(cmd_parms *cmd, void *dummy, const char *uid, + const char *gid) +{ + int i; + int matching = 0; + int u = ap_uname2id(uid); + int g = ap_gname2id(gid); + const char *errstr; + int socks[2]; + perchild_server_conf *sconf = (perchild_server_conf *) + ap_get_module_config(cmd->server->module_config, + &mpm_perchild_module); + + sconf->fullsockname = apr_pstrcat(cmd->pool, sconf->sockname, ".", uid, + ":", gid, NULL); + + if ((errstr = make_perchild_socket(sconf->fullsockname, socks))) { + return errstr; + } + + sconf->input = socks[0]; + sconf->output = socks[1]; + + for (i = 0; i < num_daemons; i++) { + if (u == child_info_table[i].uid && g == child_info_table[i].gid) { + child_info_table[i].input = sconf->input; + child_info_table[i].output = sconf->output; + matching++; + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, cmd->server, + "filling out child_info_table; UID: %d, GID: %d, " + "SD: %d %d, OUTPUT: %d %d, Child Num: %d", + child_info_table[i].uid, child_info_table[i].gid, + sconf->input, child_info_table[i].input, sconf->output, + child_info_table[i].output, i); + } + } + + if (!matching) { + return "Unable to find process with matching uid/gid."; + } + return NULL; +} + +static const char *set_server_limit (cmd_parms *cmd, void *dummy, const char *arg) +{ + int tmp_server_limit; + + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + tmp_server_limit = atoi(arg); + /* you cannot change ServerLimit across a restart; ignore + * any such attempts + */ + if (first_server_limit && + tmp_server_limit != server_limit) { + /* how do we log a message? the error log is a bit bucket at this + * point; we'll just have to set a flag so that ap_mpm_run() + * logs a warning later + */ + changed_limit_at_restart = 1; + return NULL; + } + server_limit = tmp_server_limit; + + if (server_limit > MAX_SERVER_LIMIT) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: ServerLimit of %d exceeds compile time limit " + "of %d servers,", server_limit, MAX_SERVER_LIMIT); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " lowering ServerLimit to %d.", MAX_SERVER_LIMIT); + server_limit = MAX_SERVER_LIMIT; + } + else if (server_limit < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require ServerLimit > 0, setting to 1"); + server_limit = 1; + } + return NULL; +} + +static const char *set_thread_limit (cmd_parms *cmd, void *dummy, const char *arg) +{ + int tmp_thread_limit; + + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + tmp_thread_limit = atoi(arg); + /* you cannot change ThreadLimit across a restart; ignore + * any such attempts + */ + if (first_thread_limit && + tmp_thread_limit != thread_limit) { + /* how do we log a message? the error log is a bit bucket at this + * point; we'll just have to set a flag so that ap_mpm_run() + * logs a warning later + */ + changed_limit_at_restart = 1; + return NULL; + } + thread_limit = tmp_thread_limit; + + if (thread_limit > MAX_THREAD_LIMIT) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: ThreadLimit of %d exceeds compile time limit " + "of %d servers,", thread_limit, MAX_THREAD_LIMIT); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " lowering ThreadLimit to %d.", MAX_THREAD_LIMIT); + thread_limit = MAX_THREAD_LIMIT; + } + else if (thread_limit < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require ThreadLimit > 0, setting to 1"); + thread_limit = 1; + } + return NULL; +} + +static const command_rec perchild_cmds[] = { +UNIX_DAEMON_COMMANDS, +LISTEN_COMMANDS, +AP_INIT_TAKE1("NumServers", set_num_daemons, NULL, RSRC_CONF, + "Number of children alive at the same time"), +AP_INIT_TAKE1("StartThreads", set_threads_to_start, NULL, RSRC_CONF, + "Number of threads each child creates"), +AP_INIT_TAKE1("MinSpareThreads", set_min_spare_threads, NULL, RSRC_CONF, + "Minimum number of idle threads per child, to handle " + "request spikes"), +AP_INIT_TAKE1("MaxSpareThreads", set_max_spare_threads, NULL, RSRC_CONF, + "Maximum number of idle threads per child"), +AP_INIT_TAKE1("MaxThreadsPerChild", set_max_threads, NULL, RSRC_CONF, + "Maximum number of threads per child"), +AP_INIT_TAKE3("ChildperUserID", set_child_per_uid, NULL, RSRC_CONF, + "Specify a User and Group for a specific child process."), +AP_INIT_TAKE2("AssignUserID", assign_childuid, NULL, RSRC_CONF, + "Tie a virtual host to a specific child process."), +AP_INIT_TAKE1("ServerLimit", set_server_limit, NULL, RSRC_CONF, + "Maximum value of NumServers for this run of Apache"), +AP_INIT_TAKE1("ThreadLimit", set_thread_limit, NULL, RSRC_CONF, + "Maximum worker threads in a server for this run of Apache"), +{ NULL } +}; + +static void *perchild_create_config(apr_pool_t *p, server_rec *s) +{ + perchild_server_conf *c = (perchild_server_conf *) + apr_pcalloc(p, sizeof(perchild_server_conf)); + + c->input = -1; + c->output = -1; + return c; +} + +module AP_MODULE_DECLARE_DATA mpm_perchild_module = { + MPM20_MODULE_STUFF, + ap_mpm_rewrite_args, /* hook to run before apache parses args */ + NULL, /* create per-directory config structure */ + NULL, /* merge per-directory config structures */ + perchild_create_config, /* create per-server config structure */ + NULL, /* merge per-server config structures */ + perchild_cmds, /* command apr_table_t */ + perchild_hooks /* register_hooks */ +}; + diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/Makefile.in b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/Makefile.in new file mode 100644 index 00000000..ea0acb69 --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/Makefile.in @@ -0,0 +1,5 @@ + +LTLIBRARY_NAME = libthreadpool.la +LTLIBRARY_SOURCES = threadpool.c pod.c + +include $(top_srcdir)/build/ltlib.mk diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/README b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/README new file mode 100644 index 00000000..86e8524c --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/README @@ -0,0 +1,12 @@ +Threadpool MPM: +This is an experimental variant of the standard worker MPM. +Rather than queuing connections like the worker MPM, the threadpool +MPM queues idle worker threads and hands each accepted connection +to the next available worker. + +The threadpool MPM can't match the performance of the worker MPM +in benchmark testing. As of 2.0.39, some of the key load-throtting +concepts from the threadpool MPM have been incorporated into the +worker MPM. The threadpool code is useful primarily as a research +platform; for general-purpose use, and for any production environments, +use worker instead. diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/config5.m4 b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/config5.m4 new file mode 100644 index 00000000..667b534a --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/config5.m4 @@ -0,0 +1,6 @@ +dnl ## XXX - Need a more thorough check of the proper flags to use + +if test "$MPM_NAME" = "threadpool" ; then + AC_CHECK_FUNCS(pthread_kill) + APACHE_FAST_OUTPUT(server/mpm/$MPM_SUBDIR_NAME/Makefile) +fi diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/mpm.h b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/mpm.h new file mode 100644 index 00000000..222040bd --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/mpm.h @@ -0,0 +1,50 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "scoreboard.h" +#include "unixd.h" + +#ifndef APACHE_MPM_THREADPOOL_H +#define APACHE_MPM_THREADPOOL_H + +#define THREADPOOL_MPM + +#define MPM_NAME "ThreadPool" + +#define AP_MPM_WANT_RECLAIM_CHILD_PROCESSES +#define AP_MPM_WANT_WAIT_OR_TIMEOUT +#define AP_MPM_WANT_PROCESS_CHILD_STATUS +#define AP_MPM_WANT_SET_PIDFILE +#define AP_MPM_WANT_SET_SCOREBOARD +#define AP_MPM_WANT_SET_LOCKFILE +#define AP_MPM_WANT_SET_MAX_REQUESTS +#define AP_MPM_WANT_SET_COREDUMPDIR +#define AP_MPM_WANT_SET_ACCEPT_LOCK_MECH +#define AP_MPM_WANT_SIGNAL_SERVER +#define AP_MPM_WANT_SET_MAX_MEM_FREE +#define AP_MPM_WANT_FATAL_SIGNAL_HANDLER +#define AP_MPM_DISABLE_NAGLE_ACCEPTED_SOCK + +#define MPM_CHILD_PID(i) (ap_scoreboard_image->parent[i].pid) +#define MPM_NOTE_CHILD_KILLED(i) (MPM_CHILD_PID(i) = 0) +#define MPM_ACCEPT_FUNC unixd_accept + +extern int ap_threads_per_child; +extern int ap_max_daemons_limit; +extern server_rec *ap_server_conf; +extern char ap_coredump_dir[MAX_STRING_LEN]; + +#endif /* APACHE_MPM_THREADPOOL_H */ diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/mpm_default.h b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/mpm_default.h new file mode 100644 index 00000000..d5a33989 --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/mpm_default.h @@ -0,0 +1,69 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef APACHE_MPM_DEFAULT_H +#define APACHE_MPM_DEFAULT_H + +/* Number of servers to spawn off by default --- also, if fewer than + * this free when the caretaker checks, it will spawn more. + */ +#ifndef DEFAULT_START_DAEMON +#define DEFAULT_START_DAEMON 3 +#endif + +/* Maximum number of *free* server processes --- more than this, and + * they will die off. + */ + +#ifndef DEFAULT_MAX_FREE_DAEMON +#define DEFAULT_MAX_FREE_DAEMON 10 +#endif + +/* Minimum --- fewer than this, and more will be created */ + +#ifndef DEFAULT_MIN_FREE_DAEMON +#define DEFAULT_MIN_FREE_DAEMON 3 +#endif + +#ifndef DEFAULT_THREADS_PER_CHILD +#define DEFAULT_THREADS_PER_CHILD 25 +#endif + +/* File used for accept locking, when we use a file */ +#ifndef DEFAULT_LOCKFILE +#define DEFAULT_LOCKFILE DEFAULT_REL_RUNTIMEDIR "/accept.lock" +#endif + +/* Where the main/parent process's pid is logged */ +#ifndef DEFAULT_PIDLOG +#define DEFAULT_PIDLOG DEFAULT_REL_RUNTIMEDIR "/httpd.pid" +#endif + +/* + * Interval, in microseconds, between scoreboard maintenance. + */ +#ifndef SCOREBOARD_MAINTENANCE_INTERVAL +#define SCOREBOARD_MAINTENANCE_INTERVAL 1000000 +#endif + +/* Number of requests to try to handle in a single process. If <= 0, + * the children don't die off. + */ +#ifndef DEFAULT_MAX_REQUESTS_PER_CHILD +#define DEFAULT_MAX_REQUESTS_PER_CHILD 10000 +#endif + +#endif /* AP_MPM_DEFAULT_H */ diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/pod.c b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/pod.c new file mode 100644 index 00000000..2f26130c --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/pod.c @@ -0,0 +1,108 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "pod.h" + +#if APR_HAVE_UNISTD_H +#include <unistd.h> +#endif + +AP_DECLARE(apr_status_t) ap_mpm_pod_open(apr_pool_t *p, ap_pod_t **pod) +{ + apr_status_t rv; + + *pod = apr_palloc(p, sizeof(**pod)); + rv = apr_file_pipe_create(&((*pod)->pod_in), &((*pod)->pod_out), p); + if (rv != APR_SUCCESS) { + return rv; + } +/* + apr_file_pipe_timeout_set((*pod)->pod_in, 0); +*/ + (*pod)->p = p; + + return APR_SUCCESS; +} + +AP_DECLARE(int) ap_mpm_pod_check(ap_pod_t *pod) +{ + char c; + apr_os_file_t fd; + int rc; + + /* we need to surface EINTR so we'll have to grab the + * native file descriptor and do the OS read() ourselves + */ + apr_os_file_get(&fd, pod->pod_in); + rc = read(fd, &c, 1); + if (rc == 1) { + switch(c) { + case RESTART_CHAR: + return AP_RESTART; + case GRACEFUL_CHAR: + return AP_GRACEFUL; + } + } + return AP_NORESTART; +} + +AP_DECLARE(apr_status_t) ap_mpm_pod_close(ap_pod_t *pod) +{ + apr_status_t rv; + + rv = apr_file_close(pod->pod_out); + if (rv != APR_SUCCESS) { + return rv; + } + + rv = apr_file_close(pod->pod_in); + if (rv != APR_SUCCESS) { + return rv; + } + return rv; +} + +static apr_status_t pod_signal_internal(ap_pod_t *pod, int graceful) +{ + apr_status_t rv; + char char_of_death = graceful ? GRACEFUL_CHAR : RESTART_CHAR; + apr_size_t one = 1; + + do { + rv = apr_file_write(pod->pod_out, &char_of_death, &one); + } while (APR_STATUS_IS_EINTR(rv)); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_WARNING, rv, ap_server_conf, + "write pipe_of_death"); + } + return rv; +} + +AP_DECLARE(apr_status_t) ap_mpm_pod_signal(ap_pod_t *pod, int graceful) +{ + return pod_signal_internal(pod, graceful); +} + +AP_DECLARE(void) ap_mpm_pod_killpg(ap_pod_t *pod, int num, int graceful) +{ + int i; + apr_status_t rv = APR_SUCCESS; + + for (i = 0; i < num && rv == APR_SUCCESS; i++) { + rv = pod_signal_internal(pod, graceful); + } +} + diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/pod.h b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/pod.h new file mode 100644 index 00000000..21651e6f --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/pod.h @@ -0,0 +1,50 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "apr.h" +#include "apr_strings.h" +#define APR_WANT_STRFUNC +#include "apr_want.h" + +#include "httpd.h" +#include "http_config.h" +#include "http_log.h" +#include "http_main.h" +#include "mpm.h" +#include "mpm_common.h" +#include "ap_mpm.h" +#include "ap_listen.h" +#include "mpm_default.h" + +#define RESTART_CHAR '$' +#define GRACEFUL_CHAR '!' + +#define AP_RESTART 0 +#define AP_GRACEFUL 1 + +typedef struct ap_pod_t ap_pod_t; + +struct ap_pod_t { + apr_file_t *pod_in; + apr_file_t *pod_out; + apr_pool_t *p; +}; + +AP_DECLARE(apr_status_t) ap_mpm_pod_open(apr_pool_t *p, ap_pod_t **pod); +AP_DECLARE(int) ap_mpm_pod_check(ap_pod_t *pod); +AP_DECLARE(apr_status_t) ap_mpm_pod_close(ap_pod_t *pod); +AP_DECLARE(apr_status_t) ap_mpm_pod_signal(ap_pod_t *pod, int graceful); +AP_DECLARE(void) ap_mpm_pod_killpg(ap_pod_t *pod, int num, int graceful); diff --git a/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/threadpool.c b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/threadpool.c new file mode 100644 index 00000000..50de500e --- /dev/null +++ b/rubbos/app/httpd-2.0.64/server/mpm/experimental/threadpool/threadpool.c @@ -0,0 +1,2229 @@ +/* Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/* The purpose of this MPM is to fix the design flaws in the threaded + * model. Because of the way that pthreads and mutex locks interact, + * it is basically impossible to cleanly gracefully shutdown a child + * process if multiple threads are all blocked in accept. This model + * fixes those problems. + */ + +#include "apr.h" +#include "apr_portable.h" +#include "apr_strings.h" +#include "apr_file_io.h" +#include "apr_thread_proc.h" +#include "apr_signal.h" +#include "apr_poll.h" +#include "apr_thread_mutex.h" +#include "apr_thread_cond.h" +#include "apr_proc_mutex.h" +#define APR_WANT_STRFUNC +#include "apr_want.h" + +#if APR_HAVE_UNISTD_H +#include <unistd.h> +#endif +#if APR_HAVE_SYS_SOCKET_H +#include <sys/socket.h> +#endif +#if APR_HAVE_SYS_WAIT_H +#include <sys/wait.h> +#endif +#ifdef HAVE_SYS_PROCESSOR_H +#include <sys/processor.h> /* for bindprocessor() */ +#endif + +#if !APR_HAS_THREADS +#error The Worker MPM requires APR threads, but they are unavailable. +#endif + +#define CORE_PRIVATE + +#include "ap_config.h" +#include "httpd.h" +#include "http_main.h" +#include "http_log.h" +#include "http_config.h" /* for read_config */ +#include "http_core.h" /* for get_remote_host */ +#include "http_connection.h" +#include "ap_mpm.h" +#include "pod.h" +#include "mpm_common.h" +#include "ap_listen.h" +#include "scoreboard.h" +#include "mpm_default.h" + +#include <signal.h> +#include <limits.h> /* for INT_MAX */ + +/* Limit on the total --- clients will be locked out if more servers than + * this are needed. It is intended solely to keep the server from crashing + * when things get out of hand. + * + * We keep a hard maximum number of servers, for two reasons --- first off, + * in case something goes seriously wrong, we want to stop the fork bomb + * short of actually crashing the machine we're running on by filling some + * kernel table. Secondly, it keeps the size of the scoreboard file small + * enough that we can read the whole thing without worrying too much about + * the overhead. + */ +#ifndef DEFAULT_SERVER_LIMIT +#define DEFAULT_SERVER_LIMIT 16 +#endif + +/* Admin can't tune ServerLimit beyond MAX_SERVER_LIMIT. We want + * some sort of compile-time limit to help catch typos. + */ +#ifndef MAX_SERVER_LIMIT +#define MAX_SERVER_LIMIT 20000 +#endif + +/* Limit on the threads per process. Clients will be locked out if more than + * this * server_limit are needed. + * + * We keep this for one reason it keeps the size of the scoreboard file small + * enough that we can read the whole thing without worrying too much about + * the overhead. + */ +#ifndef DEFAULT_THREAD_LIMIT +#define DEFAULT_THREAD_LIMIT 64 +#endif + +/* Admin can't tune ThreadLimit beyond MAX_THREAD_LIMIT. We want + * some sort of compile-time limit to help catch typos. + */ +#ifndef MAX_THREAD_LIMIT +#define MAX_THREAD_LIMIT 20000 +#endif + +/* + * Actual definitions of config globals + */ + +int ap_threads_per_child = 0; /* Worker threads per child */ +static int ap_daemons_to_start = 0; +static int min_spare_threads = 0; +static int max_spare_threads = 0; +static int ap_daemons_limit = 0; +static int server_limit = DEFAULT_SERVER_LIMIT; +static int first_server_limit; +static int thread_limit = DEFAULT_THREAD_LIMIT; +static int first_thread_limit; +static int changed_limit_at_restart; +static int dying = 0; +static int workers_may_exit = 0; +static int start_thread_may_exit = 0; +static int listener_may_exit = 0; +static int requests_this_child; +static int num_listensocks = 0; +static int resource_shortage = 0; +static int mpm_state = AP_MPMQ_STARTING; + +/* The structure used to pass unique initialization info to each thread */ +typedef struct { + int pid; + int tid; + int sd; +} proc_info; + +/* Structure used to pass information to the thread responsible for + * creating the rest of the threads. + */ +typedef struct { + apr_thread_t **threads; + apr_thread_t *listener; + int child_num_arg; + apr_threadattr_t *threadattr; +} thread_starter; + +#define ID_FROM_CHILD_THREAD(c, t) ((c * thread_limit) + t) + +/* + * The max child slot ever assigned, preserved across restarts. Necessary + * to deal with MaxClients changes across AP_SIG_GRACEFUL restarts. We + * use this value to optimize routines that have to scan the entire + * scoreboard. + */ +int ap_max_daemons_limit = -1; + +static ap_pod_t *pod; + +/* *Non*-shared http_main globals... */ + +server_rec *ap_server_conf; + +/* The worker MPM respects a couple of runtime flags that can aid + * in debugging. Setting the -DNO_DETACH flag will prevent the root process + * from detaching from its controlling terminal. Additionally, setting + * the -DONE_PROCESS flag (which implies -DNO_DETACH) will get you the + * child_main loop running in the process which originally started up. + * This gives you a pretty nice debugging environment. (You'll get a SIGHUP + * early in standalone_main; just continue through. This is the server + * trying to kill off any child processes which it might have lying + * around --- Apache doesn't keep track of their pids, it just sends + * SIGHUP to the process group, ignoring it in the root process. + * Continue through and you'll be fine.). + */ + +static int one_process = 0; + +#ifdef DEBUG_SIGSTOP +int raise_sigstop_flags; +#endif + +static apr_pool_t *pconf; /* Pool for config stuff */ +static apr_pool_t *pchild; /* Pool for httpd child stuff */ + +static pid_t ap_my_pid; /* Linux getpid() doesn't work except in main + thread. Use this instead */ +static pid_t parent_pid; +static apr_os_thread_t *listener_os_thread; + +/* Locks for accept serialization */ +static apr_proc_mutex_t *accept_mutex; + +#if APR_O_NONBLOCK_INHERITED +#undef SINGLE_LISTEN_UNSERIALIZED_ACCEPT +#endif /* APR_O_NONBLOCK_INHERITED */ + +#ifdef SINGLE_LISTEN_UNSERIALIZED_ACCEPT +#define SAFE_ACCEPT(stmt) (ap_listeners->next ? (stmt) : APR_SUCCESS) +#else +#define SAFE_ACCEPT(stmt) (stmt) +#endif + +/* The LISTENER_SIGNAL signal will be sent from the main thread to the + * listener thread to wake it up for graceful termination (what a child + * process from an old generation does when the admin does "apachectl + * graceful"). This signal will be blocked in all threads of a child + * process except for the listener thread. + */ +#define LISTENER_SIGNAL SIGHUP + + +/* Possible states of a worker thread. */ +typedef enum { + WORKER_IDLE, + WORKER_BUSY, + WORKER_TERMINATED +} worker_state_e; + +/* Structure used to wake up an idle worker thread + */ +typedef struct { + apr_pool_t *pool; + apr_socket_t *csd; + worker_state_e state; + apr_thread_cond_t *cond; + apr_thread_mutex_t *mutex; +} worker_wakeup_info; + +/* Structure used to hold a stack of idle worker threads + */ +typedef struct { + apr_thread_mutex_t *mutex; + apr_thread_cond_t *cond; + worker_wakeup_info **stack; + apr_size_t nelts; + apr_size_t nalloc; + int terminated; +} worker_stack; + +static worker_stack* worker_stack_create(apr_pool_t *pool, apr_size_t max) +{ + apr_status_t rv; + worker_stack *stack = (worker_stack *)apr_palloc(pool, sizeof(*stack)); + + if ((rv = apr_thread_mutex_create(&stack->mutex, APR_THREAD_MUTEX_DEFAULT, + pool)) != APR_SUCCESS) { + return NULL; + } + if ((rv = apr_thread_cond_create(&stack->cond, pool)) != APR_SUCCESS) { + return NULL; + } + stack->nelts = 0; + stack->nalloc = max; + stack->stack = + (worker_wakeup_info **)apr_palloc(pool, stack->nalloc * + sizeof(worker_wakeup_info *)); + stack->terminated = 0; + return stack; +} + +static apr_status_t worker_stack_wait(worker_stack *stack, + worker_wakeup_info *wakeup) +{ + apr_status_t rv; + + wakeup->state = WORKER_IDLE; + + if ((rv = apr_thread_mutex_lock(stack->mutex)) != APR_SUCCESS) { + return rv; + } + if (stack->terminated) { + if ((rv = apr_thread_mutex_unlock(stack->mutex)) != APR_SUCCESS) { + return rv; + } + return APR_EOF; + } + if (stack->nelts == stack->nalloc) { + if ((rv = apr_thread_mutex_unlock(stack->mutex)) != APR_SUCCESS) { + return rv; + } + return APR_ENOSPC; + } + stack->stack[stack->nelts] = wakeup; + /* Signal a blocking listener thread only if we just made the + * stack non-empty. */ + if (stack->nelts++ == 0) { + (void)apr_thread_cond_signal(stack->cond); + } + if ((rv = apr_thread_mutex_unlock(stack->mutex)) != APR_SUCCESS) { + return rv; + } + + /* At this point we've already added this worker to the stack, now + * we just wait until the listener has accept()ed a connection + * for us. */ + if ((rv = apr_thread_mutex_lock(wakeup->mutex)) != APR_SUCCESS) { + return rv; + } + while (wakeup->state == WORKER_IDLE) { + if ((rv = apr_thread_cond_wait(wakeup->cond, wakeup->mutex)) != + APR_SUCCESS) { + return rv; + } + } + if ((rv = apr_thread_mutex_unlock(wakeup->mutex)) != APR_SUCCESS) { + return rv; + } + return APR_SUCCESS; +} + +static apr_status_t worker_stack_pop(worker_stack *stack, + worker_wakeup_info **worker) +{ + apr_status_t rv; + if ((rv = apr_thread_mutex_lock(stack->mutex)) != APR_SUCCESS) { + return rv; + } + AP_DEBUG_ASSERT(stack->nelts >= 0); + while ((stack->nelts == 0) && (!stack->terminated)) { + rv = apr_thread_cond_wait(stack->cond, stack->mutex); + if (rv != APR_SUCCESS) { + apr_status_t rv2; + rv2 = apr_thread_mutex_unlock(stack->mutex); + if (rv2 != APR_SUCCESS) { + return rv2; + } + return rv; + } + } + if (stack->terminated) { + if ((rv = apr_thread_mutex_unlock(stack->mutex)) != APR_SUCCESS) { + return rv; + } + return APR_EOF; + } + *worker = stack->stack[--stack->nelts]; + if ((rv = apr_thread_mutex_unlock(stack->mutex)) != APR_SUCCESS) { + return rv; + } + return APR_SUCCESS; +} + +static apr_status_t worker_stack_terminate(worker_stack *stack) +{ + apr_status_t rv; + worker_wakeup_info *worker; + + if ((rv = apr_thread_mutex_lock(stack->mutex)) != APR_SUCCESS) { + return rv; + } + stack->terminated = 1; + /* Wake up the listener thread. Although there will never be + * more than one thread blocking on this condition, broadcast + * just in case. */ + apr_thread_cond_broadcast(stack->cond); + while (stack->nelts) { + worker = stack->stack[--stack->nelts]; + apr_thread_mutex_lock(worker->mutex); + worker->csd = 0; + worker->state = WORKER_TERMINATED; + apr_thread_cond_signal(worker->cond); + apr_thread_mutex_unlock(worker->mutex); + } + if ((rv = apr_thread_mutex_unlock(stack->mutex)) != APR_SUCCESS) { + return rv; + } + return APR_SUCCESS; +} + +static worker_stack *idle_worker_stack; + +static void wakeup_listener(void) +{ + apr_status_t rv; + + listener_may_exit = 1; + if (!idle_worker_stack) { + return; + } + if ((rv = apr_thread_mutex_lock(idle_worker_stack->mutex)) != APR_SUCCESS) { + return; + } + if ((rv = apr_thread_cond_signal(idle_worker_stack->cond)) != + APR_SUCCESS) { + return; + } + if ((rv = apr_thread_mutex_unlock(idle_worker_stack->mutex)) != APR_SUCCESS) { + return; + } + if (!listener_os_thread) { + /* XXX there is an obscure path that this doesn't handle perfectly: + * right after listener thread is created but before + * listener_os_thread is set, the first worker thread hits an + * error and starts graceful termination + */ + return; + } + /* + * we should just be able to "kill(ap_my_pid, LISTENER_SIGNAL)" on all + * platforms and wake up the listener thread since it is the only thread + * with SIGHUP unblocked, but that doesn't work on Linux + */ +#ifdef HAVE_PTHREAD_KILL + pthread_kill(*listener_os_thread, LISTENER_SIGNAL); +#else + kill(ap_my_pid, LISTENER_SIGNAL); +#endif +} + +#define ST_INIT 0 +#define ST_GRACEFUL 1 +#define ST_UNGRACEFUL 2 + +static int terminate_mode = ST_INIT; + +static void signal_threads(int mode) +{ + if (terminate_mode == mode) { + return; + } + terminate_mode = mode; + mpm_state = AP_MPMQ_STOPPING; + + /* in case we weren't called from the listener thread, wake up the + * listener thread + */ + wakeup_listener(); + + /* for ungraceful termination, let the workers exit now; + * for graceful termination, the listener thread will notify the + * workers to exit once it has stopped accepting new connections + */ + if (mode == ST_UNGRACEFUL) { + workers_may_exit = 1; + worker_stack_terminate(idle_worker_stack); + } +} + +AP_DECLARE(apr_status_t) ap_mpm_query(int query_code, int *result) +{ + switch(query_code){ + case AP_MPMQ_MAX_DAEMON_USED: + *result = ap_max_daemons_limit; + return APR_SUCCESS; + case AP_MPMQ_IS_THREADED: + *result = AP_MPMQ_STATIC; + return APR_SUCCESS; + case AP_MPMQ_IS_FORKED: + *result = AP_MPMQ_DYNAMIC; + return APR_SUCCESS; + case AP_MPMQ_HARD_LIMIT_DAEMONS: + *result = server_limit; + return APR_SUCCESS; + case AP_MPMQ_HARD_LIMIT_THREADS: + *result = thread_limit; + return APR_SUCCESS; + case AP_MPMQ_MAX_THREADS: + *result = ap_threads_per_child; + return APR_SUCCESS; + case AP_MPMQ_MIN_SPARE_DAEMONS: + *result = 0; + return APR_SUCCESS; + case AP_MPMQ_MIN_SPARE_THREADS: + *result = min_spare_threads; + return APR_SUCCESS; + case AP_MPMQ_MAX_SPARE_DAEMONS: + *result = 0; + return APR_SUCCESS; + case AP_MPMQ_MAX_SPARE_THREADS: + *result = max_spare_threads; + return APR_SUCCESS; + case AP_MPMQ_MAX_REQUESTS_DAEMON: + *result = ap_max_requests_per_child; + return APR_SUCCESS; + case AP_MPMQ_MAX_DAEMONS: + *result = ap_daemons_limit; + return APR_SUCCESS; + case AP_MPMQ_MPM_STATE: + *result = mpm_state; + return APR_SUCCESS; + } + return APR_ENOTIMPL; +} + +/* a clean exit from a child with proper cleanup */ +static void clean_child_exit(int code) __attribute__ ((noreturn)); +static void clean_child_exit(int code) +{ + mpm_state = AP_MPMQ_STOPPING; + if (pchild) { + apr_pool_destroy(pchild); + } + exit(code); +} + +static void just_die(int sig) +{ + clean_child_exit(0); +} + +/***************************************************************** + * Connection structures and accounting... + */ + +/* volatile just in case */ +static int volatile shutdown_pending; +static int volatile restart_pending; +static int volatile is_graceful; +static volatile int child_fatal; +ap_generation_t volatile ap_my_generation; + +/* + * ap_start_shutdown() and ap_start_restart(), below, are a first stab at + * functions to initiate shutdown or restart without relying on signals. + * Previously this was initiated in sig_term() and restart() signal handlers, + * but we want to be able to start a shutdown/restart from other sources -- + * e.g. on Win32, from the service manager. Now the service manager can + * call ap_start_shutdown() or ap_start_restart() as appropiate. Note that + * these functions can also be called by the child processes, since global + * variables are no longer used to pass on the required action to the parent. + * + * These should only be called from the parent process itself, since the + * parent process will use the shutdown_pending and restart_pending variables + * to determine whether to shutdown or restart. The child process should + * call signal_parent() directly to tell the parent to die -- this will + * cause neither of those variable to be set, which the parent will + * assume means something serious is wrong (which it will be, for the + * child to force an exit) and so do an exit anyway. + */ + +static void ap_start_shutdown(void) +{ + mpm_state = AP_MPMQ_STOPPING; + if (shutdown_pending == 1) { + /* Um, is this _probably_ not an error, if the user has + * tried to do a shutdown twice quickly, so we won't + * worry about reporting it. + */ + return; + } + shutdown_pending = 1; +} + +/* do a graceful restart if graceful == 1 */ +static void ap_start_restart(int graceful) +{ + mpm_state = AP_MPMQ_STOPPING; + if (restart_pending == 1) { + /* Probably not an error - don't bother reporting it */ + return; + } + restart_pending = 1; + is_graceful = graceful; +} + +static void sig_term(int sig) +{ + ap_start_shutdown(); +} + +static void restart(int sig) +{ + ap_start_restart(sig == AP_SIG_GRACEFUL); +} + +static void set_signals(void) +{ +#ifndef NO_USE_SIGACTION + struct sigaction sa; +#endif + + if (!one_process) { + ap_fatal_signal_setup(ap_server_conf, pconf); + } + +#ifndef NO_USE_SIGACTION + sigemptyset(&sa.sa_mask); + sa.sa_flags = 0; + + sa.sa_handler = sig_term; + if (sigaction(SIGTERM, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGTERM)"); +#ifdef SIGINT + if (sigaction(SIGINT, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGINT)"); +#endif +#ifdef SIGXCPU + sa.sa_handler = SIG_DFL; + if (sigaction(SIGXCPU, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGXCPU)"); +#endif +#ifdef SIGXFSZ + sa.sa_handler = SIG_DFL; + if (sigaction(SIGXFSZ, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGXFSZ)"); +#endif +#ifdef SIGPIPE + sa.sa_handler = SIG_IGN; + if (sigaction(SIGPIPE, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGPIPE)"); +#endif + + /* we want to ignore HUPs and AP_SIG_GRACEFUL while we're busy + * processing one */ + sigaddset(&sa.sa_mask, SIGHUP); + sigaddset(&sa.sa_mask, AP_SIG_GRACEFUL); + sa.sa_handler = restart; + if (sigaction(SIGHUP, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(SIGHUP)"); + if (sigaction(AP_SIG_GRACEFUL, &sa, NULL) < 0) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, ap_server_conf, + "sigaction(" AP_SIG_GRACEFUL_STRING ")"); +#else + if (!one_process) { +#ifdef SIGXCPU + apr_signal(SIGXCPU, SIG_DFL); +#endif /* SIGXCPU */ +#ifdef SIGXFSZ + apr_signal(SIGXFSZ, SIG_DFL); +#endif /* SIGXFSZ */ + } + + apr_signal(SIGTERM, sig_term); +#ifdef SIGHUP + apr_signal(SIGHUP, restart); +#endif /* SIGHUP */ +#ifdef AP_SIG_GRACEFUL + apr_signal(AP_SIG_GRACEFUL, restart); +#endif /* AP_SIG_GRACEFUL */ +#ifdef SIGPIPE + apr_signal(SIGPIPE, SIG_IGN); +#endif /* SIGPIPE */ + +#endif +} + +/***************************************************************** + * Here follows a long bunch of generic server bookkeeping stuff... + */ + +int ap_graceful_stop_signalled(void) + /* XXX this is really a bad confusing obsolete name + * maybe it should be ap_mpm_process_exiting? + */ +{ + /* note: for a graceful termination, listener_may_exit will be set before + * workers_may_exit, so check listener_may_exit + */ + return listener_may_exit; +} + +/***************************************************************** + * Child process main loop. + */ + +static void process_socket(apr_pool_t *p, apr_socket_t *sock, int my_child_num, + int my_thread_num, apr_bucket_alloc_t *bucket_alloc) +{ + conn_rec *current_conn; + long conn_id = ID_FROM_CHILD_THREAD(my_child_num, my_thread_num); + int csd; + ap_sb_handle_t *sbh; + + ap_create_sb_handle(&sbh, p, my_child_num, my_thread_num); + apr_os_sock_get(&csd, sock); + + current_conn = ap_run_create_connection(p, ap_server_conf, sock, + conn_id, sbh, bucket_alloc); + if (current_conn) { + ap_process_connection(current_conn, sock); + ap_lingering_close(current_conn); + } +} + +/* requests_this_child has gone to zero or below. See if the admin coded + "MaxRequestsPerChild 0", and keep going in that case. Doing it this way + simplifies the hot path in worker_thread */ +static void check_infinite_requests(void) +{ + if (ap_max_requests_per_child) { + signal_threads(ST_GRACEFUL); + } + else { + /* wow! if you're executing this code, you may have set a record. + * either this child process has served over 2 billion requests, or + * you're running a threaded 2.0 on a 16 bit machine. + * + * I'll buy pizza and beers at Apachecon for the first person to do + * the former without cheating (dorking with INT_MAX, or running with + * uncommitted performance patches, for example). + * + * for the latter case, you probably deserve a beer too. Greg Ames + */ + + requests_this_child = INT_MAX; /* keep going */ + } +} + +static void unblock_signal(int sig) +{ + sigset_t sig_mask; + + sigemptyset(&sig_mask); + sigaddset(&sig_mask, sig); +#if defined(SIGPROCMASK_SETS_THREAD_MASK) + sigprocmask(SIG_UNBLOCK, &sig_mask, NULL); +#else + pthread_sigmask(SIG_UNBLOCK, &sig_mask, NULL); +#endif +} + +static void dummy_signal_handler(int sig) +{ + /* XXX If specifying SIG_IGN is guaranteed to unblock a syscall, + * then we don't need this goofy function. + */ +} + +static void *listener_thread(apr_thread_t *thd, void * dummy) +{ + proc_info * ti = dummy; + int process_slot = ti->pid; + apr_pool_t *tpool = apr_thread_pool_get(thd); + void *csd = NULL; + apr_pool_t *ptrans; /* Pool for per-transaction stuff */ + int n; + apr_pollfd_t *pollset; + apr_status_t rv; + ap_listen_rec *lr, *last_lr = ap_listeners; + worker_wakeup_info *worker = NULL; + + free(ti); + + apr_poll_setup(&pollset, num_listensocks, tpool); + for(lr = ap_listeners ; lr != NULL ; lr = lr->next) + apr_poll_socket_add(pollset, lr->sd, APR_POLLIN); + + /* Unblock the signal used to wake this thread up, and set a handler for + * it. + */ + unblock_signal(LISTENER_SIGNAL); + apr_signal(LISTENER_SIGNAL, dummy_signal_handler); + + /* TODO: Switch to a system where threads reuse the results from earlier + poll calls - manoj */ + while (1) { + /* TODO: requests_this_child should be synchronized - aaron */ + if (requests_this_child <= 0) { + check_infinite_requests(); + } + if (listener_may_exit) break; + + if (worker == NULL) { + rv = worker_stack_pop(idle_worker_stack, &worker); + if (APR_STATUS_IS_EOF(rv)) { + break; + } + else if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, + "worker_stack_pop failed"); + break; + } + ptrans = worker->pool; + } + AP_DEBUG_ASSERT(worker->state == WORKER_IDLE); + + if ((rv = SAFE_ACCEPT(apr_proc_mutex_lock(accept_mutex))) + != APR_SUCCESS) { + int level = APLOG_EMERG; + + if (listener_may_exit) { + break; + } + if (ap_scoreboard_image->parent[process_slot].generation != + ap_scoreboard_image->global->running_generation) { + level = APLOG_DEBUG; /* common to get these at restart time */ + } + ap_log_error(APLOG_MARK, level, rv, ap_server_conf, + "apr_proc_mutex_lock failed. Attempting to shutdown " + "process gracefully."); + signal_threads(ST_GRACEFUL); + break; /* skip the lock release */ + } + + if (!APR_O_NONBLOCK_INHERITED && !ap_listeners->next) { + /* Only one listener, so skip the poll */ + lr = ap_listeners; + } + else { + while (!listener_may_exit) { + apr_status_t ret; + apr_int16_t event; + + ret = apr_poll(pollset, num_listensocks, &n, -1); + if (ret != APR_SUCCESS) { + if (APR_STATUS_IS_EINTR(ret)) { + continue; + } + + /* apr_poll() will only return errors in catastrophic + * circumstances. Let's try exiting gracefully, for now. */ + ap_log_error(APLOG_MARK, APLOG_ERR, ret, (const server_rec *) + ap_server_conf, "apr_poll: (listen)"); + signal_threads(ST_GRACEFUL); + } + + if (listener_may_exit) break; + + /* find a listener */ + lr = last_lr; + do { + lr = lr->next; + if (lr == NULL) { + lr = ap_listeners; + } + /* XXX: Should we check for POLLERR? */ + apr_poll_revents_get(&event, lr->sd, pollset); + if (event & APR_POLLIN) { + last_lr = lr; + goto got_fd; + } + } while (lr != last_lr); + } + } + got_fd: + if (!listener_may_exit) { + rv = lr->accept_func(&csd, lr, ptrans); + /* later we trash rv and rely on csd to indicate success/failure */ + AP_DEBUG_ASSERT(rv == APR_SUCCESS || !csd); + + if (rv == APR_EGENERAL) { + /* E[NM]FILE, ENOMEM, etc */ + resource_shortage = 1; + signal_threads(ST_GRACEFUL); + } + if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(accept_mutex))) + != APR_SUCCESS) { + int level = APLOG_EMERG; + + if (listener_may_exit) { + break; + } + if (ap_scoreboard_image->parent[process_slot].generation != + ap_scoreboard_image->global->running_generation) { + level = APLOG_DEBUG; /* common to get these at restart time */ + } + ap_log_error(APLOG_MARK, level, rv, ap_server_conf, + "apr_proc_mutex_unlock failed. Attempting to " + "shutdown process gracefully."); + signal_threads(ST_GRACEFUL); + } + if (csd != NULL) { + /* Wake up the sleeping worker. */ + apr_thread_mutex_lock(worker->mutex); + worker->csd = (apr_socket_t *)csd; + worker->state = WORKER_BUSY; + /* Posix allows us to signal this condition without + * owning the associated mutex, but in that case it can + * not guarantee predictable scheduling. See + * _UNIX Network Programming: Interprocess Communication_ + * by W. Richard Stevens, Vol 2, 2nd Ed, pp. 170-171. */ + apr_thread_cond_signal(worker->cond); + apr_thread_mutex_unlock(worker->mutex); + worker = NULL; + } + } + else { + if ((rv = SAFE_ACCEPT(apr_proc_mutex_unlock(accept_mutex))) + != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "apr_proc_mutex_unlock failed. Attempting to " + "shutdown process gracefully."); + signal_threads(ST_GRACEFUL); + } + break; + } + } + + workers_may_exit = 1; + if (worker) { + apr_thread_mutex_lock(worker->mutex); + worker->state = WORKER_TERMINATED; + /* Posix allows us to signal this condition without + * owning the associated mutex, but in that case it can + * not guarantee predictable scheduling. See + * _UNIX Network Programming: Interprocess Communication_ + * by W. Richard Stevens, Vol 2, 2nd Ed, pp. 170-171. */ + apr_thread_cond_signal(worker->cond); + apr_thread_mutex_unlock(worker->mutex); + } + worker_stack_terminate(idle_worker_stack); + dying = 1; + ap_scoreboard_image->parent[process_slot].quiescing = 1; + + /* wake up the main thread */ + kill(ap_my_pid, SIGTERM); + + apr_thread_exit(thd, APR_SUCCESS); + return NULL; +} + +/* XXX For ungraceful termination/restart, we definitely don't want to + * wait for active connections to finish but we may want to wait + * for idle workers to get out of the queue code and release mutexes, + * since those mutexes are cleaned up pretty soon and some systems + * may not react favorably (i.e., segfault) if operations are attempted + * on cleaned-up mutexes. + */ +static void * APR_THREAD_FUNC worker_thread(apr_thread_t *thd, void * dummy) +{ + proc_info * ti = dummy; + int process_slot = ti->pid; + int thread_slot = ti->tid; + apr_bucket_alloc_t *bucket_alloc; + apr_pool_t *tpool = apr_thread_pool_get(thd); + apr_pool_t *ptrans; /* Pool for per-transaction stuff */ + apr_allocator_t *allocator; + apr_status_t rv; + worker_wakeup_info *wakeup; + + free(ti); + + ap_update_child_status_from_indexes(process_slot, thread_slot, SERVER_STARTING, NULL); + + apr_allocator_create(&allocator); + apr_allocator_max_free_set(allocator, ap_max_mem_free); + /* XXX: why is ptrans's parent not tpool? --jcw 08/2003 */ + apr_pool_create_ex(&ptrans, NULL, NULL, allocator); + apr_allocator_owner_set(allocator, ptrans); + bucket_alloc = apr_bucket_alloc_create_ex(allocator); + + wakeup = (worker_wakeup_info *)apr_palloc(tpool, sizeof(*wakeup)); + wakeup->pool = ptrans; + if ((rv = apr_thread_cond_create(&wakeup->cond, tpool)) != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "apr_thread_cond_create failed. Attempting to shutdown " + "process gracefully."); + signal_threads(ST_GRACEFUL); + apr_thread_exit(thd, rv); + } + if ((rv = apr_thread_mutex_create(&wakeup->mutex, APR_THREAD_MUTEX_DEFAULT, + tpool)) != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "apr_thread_mutex_create failed. Attempting to shutdown " + "process gracefully."); + signal_threads(ST_GRACEFUL); + apr_thread_exit(thd, rv); + } + + while (!workers_may_exit) { + ap_update_child_status_from_indexes(process_slot, thread_slot, SERVER_READY, NULL); + rv = worker_stack_wait(idle_worker_stack, wakeup); + if (APR_STATUS_IS_EOF(rv)) { + break; /* The queue has been terminated. */ + } + else if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, + "worker_stack_wait failed"); + break; /* Treat all other errors as fatal. */ + } + else if (wakeup->state == WORKER_TERMINATED) { + break; /* They told us to quit. */ + } + AP_DEBUG_ASSERT(wakeup->state != WORKER_IDLE); + process_socket(ptrans, wakeup->csd, + process_slot, thread_slot, bucket_alloc); + requests_this_child--; /* FIXME: should be synchronized - aaron */ + apr_pool_clear(ptrans); + } + + ap_update_child_status_from_indexes(process_slot, thread_slot, + (dying) ? SERVER_DEAD : SERVER_GRACEFUL, (request_rec *) NULL); + + apr_bucket_alloc_destroy(bucket_alloc); + + apr_thread_exit(thd, APR_SUCCESS); + return NULL; +} + +static int check_signal(int signum) +{ + switch (signum) { + case SIGTERM: + case SIGINT: + return 1; + } + return 0; +} + +static void create_listener_thread(thread_starter *ts) +{ + int my_child_num = ts->child_num_arg; + apr_threadattr_t *thread_attr = ts->threadattr; + proc_info *my_info; + apr_status_t rv; + + my_info = (proc_info *)malloc(sizeof(proc_info)); + my_info->pid = my_child_num; + my_info->tid = -1; /* listener thread doesn't have a thread slot */ + my_info->sd = 0; + rv = apr_thread_create(&ts->listener, thread_attr, listener_thread, + my_info, pchild); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf, + "apr_thread_create: unable to create listener thread"); + /* In case system resources are maxxed out, we don't want + * Apache running away with the CPU trying to fork over and + * over and over again if we exit. + * XXX Jeff doesn't see how Apache is going to try to fork again since + * the exit code is APEXIT_CHILDFATAL + */ + apr_sleep(10 * APR_USEC_PER_SEC); + clean_child_exit(APEXIT_CHILDFATAL); + } + apr_os_thread_get(&listener_os_thread, ts->listener); +} + +/* XXX under some circumstances not understood, children can get stuck + * in start_threads forever trying to take over slots which will + * never be cleaned up; for now there is an APLOG_DEBUG message issued + * every so often when this condition occurs + */ +static void * APR_THREAD_FUNC start_threads(apr_thread_t *thd, void *dummy) +{ + thread_starter *ts = dummy; + apr_thread_t **threads = ts->threads; + apr_threadattr_t *thread_attr = ts->threadattr; + int child_num_arg = ts->child_num_arg; + int my_child_num = child_num_arg; + proc_info *my_info; + apr_status_t rv; + int i; + int threads_created = 0; + int loops; + int prev_threads_created; + + idle_worker_stack = worker_stack_create(pchild, ap_threads_per_child); + if (idle_worker_stack == NULL) { + ap_log_error(APLOG_MARK, APLOG_ALERT, 0, ap_server_conf, + "worker_stack_create() failed"); + clean_child_exit(APEXIT_CHILDFATAL); + } + + loops = prev_threads_created = 0; + while (1) { + /* ap_threads_per_child does not include the listener thread */ + for (i = 0; i < ap_threads_per_child; i++) { + int status = ap_scoreboard_image->servers[child_num_arg][i].status; + + if (status != SERVER_GRACEFUL && status != SERVER_DEAD) { + continue; + } + + my_info = (proc_info *)malloc(sizeof(proc_info)); + if (my_info == NULL) { + ap_log_error(APLOG_MARK, APLOG_ALERT, errno, ap_server_conf, + "malloc: out of memory"); + clean_child_exit(APEXIT_CHILDFATAL); + } + my_info->pid = my_child_num; + my_info->tid = i; + my_info->sd = 0; + + /* We are creating threads right now */ + ap_update_child_status_from_indexes(my_child_num, i, + SERVER_STARTING, NULL); + /* We let each thread update its own scoreboard entry. This is + * done because it lets us deal with tid better. + */ + rv = apr_thread_create(&threads[i], thread_attr, + worker_thread, my_info, pchild); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf, + "apr_thread_create: unable to create worker thread"); + /* In case system resources are maxxed out, we don't want + Apache running away with the CPU trying to fork over and + over and over again if we exit. */ + apr_sleep(10 * APR_USEC_PER_SEC); + clean_child_exit(APEXIT_CHILDFATAL); + } + threads_created++; + if (threads_created == 1) { + /* now that we have a worker thread, it makes sense to create + * a listener thread (we don't want a listener without a worker!) + */ + create_listener_thread(ts); + } + } + if (start_thread_may_exit || threads_created == ap_threads_per_child) { + break; + } + /* wait for previous generation to clean up an entry */ + apr_sleep(1 * APR_USEC_PER_SEC); + ++loops; + if (loops % 120 == 0) { /* every couple of minutes */ + if (prev_threads_created == threads_created) { + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "child %" APR_PID_T_FMT " isn't taking over " + "slots very quickly (%d of %d)", + ap_my_pid, threads_created, ap_threads_per_child); + } + prev_threads_created = threads_created; + } + } + + /* What state should this child_main process be listed as in the + * scoreboard...? + * ap_update_child_status_from_indexes(my_child_num, i, SERVER_STARTING, + * (request_rec *) NULL); + * + * This state should be listed separately in the scoreboard, in some kind + * of process_status, not mixed in with the worker threads' status. + * "life_status" is almost right, but it's in the worker's structure, and + * the name could be clearer. gla + */ + apr_thread_exit(thd, APR_SUCCESS); + return NULL; +} + +static void join_workers(apr_thread_t *listener, apr_thread_t **threads) +{ + int i; + apr_status_t rv, thread_rv; + + if (listener) { + int iter; + + /* deal with a rare timing window which affects waking up the + * listener thread... if the signal sent to the listener thread + * is delivered between the time it verifies that the + * listener_may_exit flag is clear and the time it enters a + * blocking syscall, the signal didn't do any good... work around + * that by sleeping briefly and sending it again + */ + + iter = 0; + while (iter < 10 && +#ifdef HAVE_PTHREAD_KILL + pthread_kill(*listener_os_thread, 0) +#else + kill(ap_my_pid, 0) +#endif + == 0) { + /* listener not dead yet */ + apr_sleep(APR_USEC_PER_SEC / 2); + wakeup_listener(); + ++iter; + } + if (iter >= 10) { + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "the listener thread didn't exit"); + } + else { + rv = apr_thread_join(&thread_rv, listener); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, + "apr_thread_join: unable to join listener thread"); + } + } + } + + for (i = 0; i < ap_threads_per_child; i++) { + if (threads[i]) { /* if we ever created this thread */ + rv = apr_thread_join(&thread_rv, threads[i]); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, + "apr_thread_join: unable to join worker " + "thread %d", + i); + } + } + } +} + +static void join_start_thread(apr_thread_t *start_thread_id) +{ + apr_status_t rv, thread_rv; + + start_thread_may_exit = 1; /* tell it to give up in case it is still + * trying to take over slots from a + * previous generation + */ + rv = apr_thread_join(&thread_rv, start_thread_id); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, ap_server_conf, + "apr_thread_join: unable to join the start " + "thread"); + } +} + +static void child_main(int child_num_arg) +{ + apr_thread_t **threads; + apr_status_t rv; + thread_starter *ts; + apr_threadattr_t *thread_attr; + apr_thread_t *start_thread_id; + + mpm_state = AP_MPMQ_STARTING; /* for benefit of any hooks that run as this + * child initializes + */ + ap_my_pid = getpid(); + ap_fatal_signal_child_setup(ap_server_conf); + apr_pool_create(&pchild, pconf); + + /*stuff to do before we switch id's, so we have permissions.*/ + ap_reopen_scoreboard(pchild, NULL, 0); + + rv = SAFE_ACCEPT(apr_proc_mutex_child_init(&accept_mutex, ap_lock_fname, + pchild)); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "Couldn't initialize cross-process lock in child"); + clean_child_exit(APEXIT_CHILDFATAL); + } + + if (unixd_setup_child()) { + clean_child_exit(APEXIT_CHILDFATAL); + } + + ap_run_child_init(pchild, ap_server_conf); + + /* done with init critical section */ + + /* Just use the standard apr_setup_signal_thread to block all signals + * from being received. The child processes no longer use signals for + * any communication with the parent process. + */ + rv = apr_setup_signal_thread(); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, ap_server_conf, + "Couldn't initialize signal thread"); + clean_child_exit(APEXIT_CHILDFATAL); + } + + if (ap_max_requests_per_child) { + requests_this_child = ap_max_requests_per_child; + } + else { + /* coding a value of zero means infinity */ + requests_this_child = INT_MAX; + } + + /* Setup worker threads */ + + /* clear the storage; we may not create all our threads immediately, + * and we want a 0 entry to indicate a thread which was not created + */ + threads = (apr_thread_t **)calloc(1, + sizeof(apr_thread_t *) * ap_threads_per_child); + if (threads == NULL) { + ap_log_error(APLOG_MARK, APLOG_ALERT, errno, ap_server_conf, + "malloc: out of memory"); + clean_child_exit(APEXIT_CHILDFATAL); + } + + ts = (thread_starter *)apr_palloc(pchild, sizeof(*ts)); + + apr_threadattr_create(&thread_attr, pchild); + /* 0 means PTHREAD_CREATE_JOINABLE */ + apr_threadattr_detach_set(thread_attr, 0); + + ts->threads = threads; + ts->listener = NULL; + ts->child_num_arg = child_num_arg; + ts->threadattr = thread_attr; + + rv = apr_thread_create(&start_thread_id, thread_attr, start_threads, + ts, pchild); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_ALERT, rv, ap_server_conf, + "apr_thread_create: unable to create worker thread"); + /* In case system resources are maxxed out, we don't want + Apache running away with the CPU trying to fork over and + over and over again if we exit. */ + apr_sleep(10 * APR_USEC_PER_SEC); + clean_child_exit(APEXIT_CHILDFATAL); + } + + mpm_state = AP_MPMQ_RUNNING; + + /* If we are only running in one_process mode, we will want to + * still handle signals. */ + if (one_process) { + /* Block until we get a terminating signal. */ + apr_signal_thread(check_signal); + /* make sure the start thread has finished; signal_threads() + * and join_workers() depend on that + */ + /* XXX join_start_thread() won't be awakened if one of our + * threads encounters a critical error and attempts to + * shutdown this child + */ + join_start_thread(start_thread_id); + signal_threads(ST_UNGRACEFUL); /* helps us terminate a little more + * quickly than the dispatch of the signal thread + * beats the Pipe of Death and the browsers + */ + /* A terminating signal was received. Now join each of the + * workers to clean them up. + * If the worker already exited, then the join frees + * their resources and returns. + * If the worker hasn't exited, then this blocks until + * they have (then cleans up). + */ + join_workers(ts->listener, threads); + } + else { /* !one_process */ + /* remove SIGTERM from the set of blocked signals... if one of + * the other threads in the process needs to take us down + * (e.g., for MaxRequestsPerChild) it will send us SIGTERM + */ + unblock_signal(SIGTERM); + apr_signal(SIGTERM, dummy_signal_handler); + /* Watch for any messages from the parent over the POD */ + while (1) { + rv = ap_mpm_pod_check(pod); + if (rv == AP_NORESTART) { + /* see if termination was triggered while we slept */ + switch(terminate_mode) { + case ST_GRACEFUL: + rv = AP_GRACEFUL; + break; + case ST_UNGRACEFUL: + rv = AP_RESTART; + break; + } + } + if (rv == AP_GRACEFUL || rv == AP_RESTART) { + /* make sure the start thread has finished; + * signal_threads() and join_workers depend on that + */ + join_start_thread(start_thread_id); + signal_threads(rv == AP_GRACEFUL ? ST_GRACEFUL : ST_UNGRACEFUL); + break; + } + } + + if (rv == AP_GRACEFUL) { + /* A terminating signal was received. Now join each of the + * workers to clean them up. + * If the worker already exited, then the join frees + * their resources and returns. + * If the worker hasn't exited, then this blocks until + * they have (then cleans up). + */ + join_workers(ts->listener, threads); + } + } + + free(threads); + + clean_child_exit(resource_shortage ? APEXIT_CHILDSICK : 0); +} + +static int make_child(server_rec *s, int slot) +{ + int pid; + + if (slot + 1 > ap_max_daemons_limit) { + ap_max_daemons_limit = slot + 1; + } + + if (one_process) { + set_signals(); + ap_scoreboard_image->parent[slot].pid = getpid(); + child_main(slot); + } + + if ((pid = fork()) == -1) { + ap_log_error(APLOG_MARK, APLOG_ERR, errno, s, + "fork: Unable to fork new process"); + + /* fork didn't succeed. Fix the scoreboard or else + * it will say SERVER_STARTING forever and ever + */ + ap_update_child_status_from_indexes(slot, 0, SERVER_DEAD, NULL); + + /* In case system resources are maxxed out, we don't want + Apache running away with the CPU trying to fork over and + over and over again. */ + apr_sleep(10 * APR_USEC_PER_SEC); + + return -1; + } + + if (!pid) { +#ifdef HAVE_BINDPROCESSOR + /* By default, AIX binds to a single processor. This bit unbinds + * children which will then bind to another CPU. + */ + int status = bindprocessor(BINDPROCESS, (int)getpid(), + PROCESSOR_CLASS_ANY); + if (status != OK) + ap_log_error(APLOG_MARK, APLOG_WARNING, errno, + ap_server_conf, + "processor unbind failed %d", status); +#endif + RAISE_SIGSTOP(MAKE_CHILD); + + apr_signal(SIGTERM, just_die); + child_main(slot); + + clean_child_exit(0); + } + /* else */ + ap_scoreboard_image->parent[slot].quiescing = 0; + ap_scoreboard_image->parent[slot].pid = pid; + return 0; +} + +/* start up a bunch of children */ +static void startup_children(int number_to_start) +{ + int i; + + for (i = 0; number_to_start && i < ap_daemons_limit; ++i) { + if (ap_scoreboard_image->parent[i].pid != 0) { + continue; + } + if (make_child(ap_server_conf, i) < 0) { + break; + } + --number_to_start; + } +} + + +/* + * idle_spawn_rate is the number of children that will be spawned on the + * next maintenance cycle if there aren't enough idle servers. It is + * doubled up to MAX_SPAWN_RATE, and reset only when a cycle goes by + * without the need to spawn. + */ +static int idle_spawn_rate = 1; +#ifndef MAX_SPAWN_RATE +#define MAX_SPAWN_RATE (32) +#endif +static int hold_off_on_exponential_spawning; + +static void perform_idle_server_maintenance(void) +{ + int i, j; + int idle_thread_count; + worker_score *ws; + process_score *ps; + int free_length; + int totally_free_length = 0; + int free_slots[MAX_SPAWN_RATE]; + int last_non_dead; + int total_non_dead; + + /* initialize the free_list */ + free_length = 0; + + idle_thread_count = 0; + last_non_dead = -1; + total_non_dead = 0; + + for (i = 0; i < ap_daemons_limit; ++i) { + /* Initialization to satisfy the compiler. It doesn't know + * that ap_threads_per_child is always > 0 */ + int status = SERVER_DEAD; + int any_dying_threads = 0; + int any_dead_threads = 0; + int all_dead_threads = 1; + + if (i >= ap_max_daemons_limit && totally_free_length == idle_spawn_rate) + break; + ps = &ap_scoreboard_image->parent[i]; + for (j = 0; j < ap_threads_per_child; j++) { + ws = &ap_scoreboard_image->servers[i][j]; + status = ws->status; + + /* XXX any_dying_threads is probably no longer needed GLA */ + any_dying_threads = any_dying_threads || + (status == SERVER_GRACEFUL); + any_dead_threads = any_dead_threads || (status == SERVER_DEAD); + all_dead_threads = all_dead_threads && + (status == SERVER_DEAD || + status == SERVER_GRACEFUL); + + /* We consider a starting server as idle because we started it + * at least a cycle ago, and if it still hasn't finished starting + * then we're just going to swamp things worse by forking more. + * So we hopefully won't need to fork more if we count it. + * This depends on the ordering of SERVER_READY and SERVER_STARTING. + */ + if (status <= SERVER_READY && status != SERVER_DEAD && + !ps->quiescing && + ps->generation == ap_my_generation && + /* XXX the following shouldn't be necessary if we clean up + * properly after seg faults, but we're not yet GLA + */ + ps->pid != 0) { + ++idle_thread_count; + } + } + if (any_dead_threads && totally_free_length < idle_spawn_rate + && (!ps->pid /* no process in the slot */ + || ps->quiescing)) { /* or at least one is going away */ + if (all_dead_threads) { + /* great! we prefer these, because the new process can + * start more threads sooner. So prioritize this slot + * by putting it ahead of any slots with active threads. + * + * first, make room by moving a slot that's potentially still + * in use to the end of the array + */ + free_slots[free_length] = free_slots[totally_free_length]; + free_slots[totally_free_length++] = i; + } + else { + /* slot is still in use - back of the bus + */ + free_slots[free_length] = i; + } + ++free_length; + } + /* XXX if (!ps->quiescing) is probably more reliable GLA */ + if (!any_dying_threads) { + last_non_dead = i; + ++total_non_dead; + } + } + ap_max_daemons_limit = last_non_dead + 1; + + if (idle_thread_count > max_spare_threads) { + /* Kill off one child */ + ap_mpm_pod_signal(pod, TRUE); + idle_spawn_rate = 1; + } + else if (idle_thread_count < min_spare_threads) { + /* terminate the free list */ + if (free_length == 0) { + /* only report this condition once */ + static int reported = 0; + + if (!reported) { + ap_log_error(APLOG_MARK, APLOG_ERR, 0, + ap_server_conf, + "server reached MaxClients setting, consider" + " raising the MaxClients setting"); + reported = 1; + } + idle_spawn_rate = 1; + } + else { + if (free_length > idle_spawn_rate) { + free_length = idle_spawn_rate; + } + if (idle_spawn_rate >= 8) { + ap_log_error(APLOG_MARK, APLOG_INFO, 0, + ap_server_conf, + "server seems busy, (you may need " + "to increase StartServers, ThreadsPerChild " + "or Min/MaxSpareThreads), " + "spawning %d children, there are around %d idle " + "threads, and %d total children", free_length, + idle_thread_count, total_non_dead); + } + for (i = 0; i < free_length; ++i) { + make_child(ap_server_conf, free_slots[i]); + } + /* the next time around we want to spawn twice as many if this + * wasn't good enough, but not if we've just done a graceful + */ + if (hold_off_on_exponential_spawning) { + --hold_off_on_exponential_spawning; + } + else if (idle_spawn_rate < MAX_SPAWN_RATE) { + idle_spawn_rate *= 2; + } + } + } + else { + idle_spawn_rate = 1; + } +} + +static void server_main_loop(int remaining_children_to_start) +{ + int child_slot; + apr_exit_why_e exitwhy; + int status, processed_status; + apr_proc_t pid; + int i; + + while (!restart_pending && !shutdown_pending) { + ap_wait_or_timeout(&exitwhy, &status, &pid, pconf); + + if (pid.pid != -1) { + processed_status = ap_process_child_status(&pid, exitwhy, status); + if (processed_status == APEXIT_CHILDFATAL) { + shutdown_pending = 1; + child_fatal = 1; + return; + } + /* non-fatal death... note that it's gone in the scoreboard. */ + child_slot = find_child_by_pid(&pid); + if (child_slot >= 0) { + for (i = 0; i < ap_threads_per_child; i++) + ap_update_child_status_from_indexes(child_slot, i, SERVER_DEAD, + (request_rec *) NULL); + + ap_scoreboard_image->parent[child_slot].pid = 0; + ap_scoreboard_image->parent[child_slot].quiescing = 0; + if (processed_status == APEXIT_CHILDSICK) { + /* resource shortage, minimize the fork rate */ + idle_spawn_rate = 1; + } + else if (remaining_children_to_start + && child_slot < ap_daemons_limit) { + /* we're still doing a 1-for-1 replacement of dead + * children with new children + */ + make_child(ap_server_conf, child_slot); + --remaining_children_to_start; + } +#if APR_HAS_OTHER_CHILD + } + else if (apr_proc_other_child_read(&pid, status) == 0) { + /* handled */ +#endif + } + else if (is_graceful) { + /* Great, we've probably just lost a slot in the + * scoreboard. Somehow we don't know about this child. + */ + ap_log_error(APLOG_MARK, APLOG_WARNING, 0, + ap_server_conf, + "long lost child came home! (pid %ld)", + (long)pid.pid); + } + /* Don't perform idle maintenance when a child dies, + * only do it when there's a timeout. Remember only a + * finite number of children can die, and it's pretty + * pathological for a lot to die suddenly. + */ + continue; + } + else if (remaining_children_to_start) { + /* we hit a 1 second timeout in which none of the previous + * generation of children needed to be reaped... so assume + * they're all done, and pick up the slack if any is left. + */ + startup_children(remaining_children_to_start); + remaining_children_to_start = 0; + /* In any event we really shouldn't do the code below because + * few of the servers we just started are in the IDLE state + * yet, so we'd mistakenly create an extra server. + */ + continue; + } + + perform_idle_server_maintenance(); + } +} + +int ap_mpm_run(apr_pool_t *_pconf, apr_pool_t *plog, server_rec *s) +{ + int remaining_children_to_start; + apr_status_t rv; + + ap_log_pid(pconf, ap_pid_fname); + + first_server_limit = server_limit; + first_thread_limit = thread_limit; + if (changed_limit_at_restart) { + ap_log_error(APLOG_MARK, APLOG_WARNING, 0, s, + "WARNING: Attempt to change ServerLimit or ThreadLimit " + "ignored during restart"); + changed_limit_at_restart = 0; + } + + /* Initialize cross-process accept lock */ + ap_lock_fname = apr_psprintf(_pconf, "%s.%" APR_PID_T_FMT, + ap_server_root_relative(_pconf, ap_lock_fname), + ap_my_pid); + + rv = apr_proc_mutex_create(&accept_mutex, ap_lock_fname, + ap_accept_lock_mech, _pconf); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, s, + "Couldn't create accept lock"); + mpm_state = AP_MPMQ_STOPPING; + return 1; + } + +#if APR_USE_SYSVSEM_SERIALIZE + if (ap_accept_lock_mech == APR_LOCK_DEFAULT || + ap_accept_lock_mech == APR_LOCK_SYSVSEM) { +#else + if (ap_accept_lock_mech == APR_LOCK_SYSVSEM) { +#endif + rv = unixd_set_proc_mutex_perms(accept_mutex); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_EMERG, rv, s, + "Couldn't set permissions on cross-process lock; " + "check User and Group directives"); + mpm_state = AP_MPMQ_STOPPING; + return 1; + } + } + + if (!is_graceful) { + if (ap_run_pre_mpm(s->process->pool, SB_SHARED) != OK) { + mpm_state = AP_MPMQ_STOPPING; + return 1; + } + /* fix the generation number in the global score; we just got a new, + * cleared scoreboard + */ + ap_scoreboard_image->global->running_generation = ap_my_generation; + } + + set_signals(); + /* Don't thrash... */ + if (max_spare_threads < min_spare_threads + ap_threads_per_child) + max_spare_threads = min_spare_threads + ap_threads_per_child; + + /* If we're doing a graceful_restart then we're going to see a lot + * of children exiting immediately when we get into the main loop + * below (because we just sent them AP_SIG_GRACEFUL). This happens pretty + * rapidly... and for each one that exits we'll start a new one until + * we reach at least daemons_min_free. But we may be permitted to + * start more than that, so we'll just keep track of how many we're + * supposed to start up without the 1 second penalty between each fork. + */ + remaining_children_to_start = ap_daemons_to_start; + if (remaining_children_to_start > ap_daemons_limit) { + remaining_children_to_start = ap_daemons_limit; + } + if (!is_graceful) { + startup_children(remaining_children_to_start); + remaining_children_to_start = 0; + } + else { + /* give the system some time to recover before kicking into + * exponential mode */ + hold_off_on_exponential_spawning = 10; + } + + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, + "%s configured -- resuming normal operations", + ap_get_server_version()); + ap_log_error(APLOG_MARK, APLOG_INFO, 0, ap_server_conf, + "Server built: %s", ap_get_server_built()); +#ifdef AP_MPM_WANT_SET_ACCEPT_LOCK_MECH + ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, ap_server_conf, + "AcceptMutex: %s (default: %s)", + apr_proc_mutex_name(accept_mutex), + apr_proc_mutex_defname()); +#endif + restart_pending = shutdown_pending = 0; + mpm_state = AP_MPMQ_RUNNING; + + server_main_loop(remaining_children_to_start); + mpm_state = AP_MPMQ_STOPPING; + + if (shutdown_pending) { + /* Time to gracefully shut down: + * Kill child processes, tell them to call child_exit, etc... + * (By "gracefully" we don't mean graceful in the same sense as + * "apachectl graceful" where we allow old connections to finish.) + */ + ap_mpm_pod_killpg(pod, ap_daemons_limit, FALSE); + ap_reclaim_child_processes(1); /* Start with SIGTERM */ + + if (!child_fatal) { + /* cleanup pid file on normal shutdown */ + const char *pidfile = NULL; + pidfile = ap_server_root_relative (pconf, ap_pid_fname); + if ( pidfile != NULL && unlink(pidfile) == 0) + ap_log_error(APLOG_MARK, APLOG_INFO, 0, + ap_server_conf, + "removed PID file %s (pid=%ld)", + pidfile, (long)getpid()); + + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, + ap_server_conf, "caught SIGTERM, shutting down"); + } + return 1; + } + + /* we've been told to restart */ + apr_signal(SIGHUP, SIG_IGN); + + if (one_process) { + /* not worth thinking about */ + return 1; + } + + /* advance to the next generation */ + /* XXX: we really need to make sure this new generation number isn't in + * use by any of the children. + */ + ++ap_my_generation; + ap_scoreboard_image->global->running_generation = ap_my_generation; + + if (is_graceful) { + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, + AP_SIG_GRACEFUL_STRING " received. Doing graceful restart"); + /* wake up the children...time to die. But we'll have more soon */ + ap_mpm_pod_killpg(pod, ap_daemons_limit, TRUE); + + + /* This is mostly for debugging... so that we know what is still + * gracefully dealing with existing request. + */ + + } + else { + /* Kill 'em all. Since the child acts the same on the parents SIGTERM + * and a SIGHUP, we may as well use the same signal, because some user + * pthreads are stealing signals from us left and right. + */ + ap_mpm_pod_killpg(pod, ap_daemons_limit, FALSE); + + ap_reclaim_child_processes(1); /* Start with SIGTERM */ + ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, ap_server_conf, + "SIGHUP received. Attempting to restart"); + } + + return 0; +} + +/* This really should be a post_config hook, but the error log is already + * redirected by that point, so we need to do this in the open_logs phase. + */ +static int worker_open_logs(apr_pool_t *p, apr_pool_t *plog, apr_pool_t *ptemp, server_rec *s) +{ + apr_status_t rv; + ap_listen_rec *lr; + + pconf = p; + ap_server_conf = s; + + if ((num_listensocks = ap_setup_listeners(ap_server_conf)) < 1) { + ap_log_error(APLOG_MARK, APLOG_ALERT|APLOG_STARTUP, 0, + NULL, "no listening sockets available, shutting down"); + return DONE; + } + +#if APR_O_NONBLOCK_INHERITED + for(lr = ap_listeners ; lr != NULL ; lr = lr->next) { + apr_socket_opt_set(lr->sd, APR_SO_NONBLOCK, 1); + } +#endif /* APR_O_NONBLOCK_INHERITED */ + + if (!one_process) { + if ((rv = ap_mpm_pod_open(pconf, &pod))) { + ap_log_error(APLOG_MARK, APLOG_CRIT|APLOG_STARTUP, rv, NULL, + "Could not open pipe-of-death."); + return DONE; + } + } + return OK; +} + +static int worker_pre_config(apr_pool_t *pconf, apr_pool_t *plog, + apr_pool_t *ptemp) +{ + static int restart_num = 0; + int no_detach, debug, foreground; + ap_directive_t *pdir; + ap_directive_t *max_clients = NULL; + apr_status_t rv; + + mpm_state = AP_MPMQ_STARTING; + + /* make sure that "ThreadsPerChild" gets set before "MaxClients" */ + for (pdir = ap_conftree; pdir != NULL; pdir = pdir->next) { + if (strncasecmp(pdir->directive, "ThreadsPerChild", 15) == 0) { + if (!max_clients) { + break; /* we're in the clear, got ThreadsPerChild first */ + } + else { + /* now to swap the data */ + ap_directive_t temp; + + temp.directive = pdir->directive; + temp.args = pdir->args; + /* Make sure you don't change 'next', or you may get loops! */ + /* XXX: first_child, parent, and data can never be set + * for these directives, right? -aaron */ + temp.filename = pdir->filename; + temp.line_num = pdir->line_num; + + pdir->directive = max_clients->directive; + pdir->args = max_clients->args; + pdir->filename = max_clients->filename; + pdir->line_num = max_clients->line_num; + + max_clients->directive = temp.directive; + max_clients->args = temp.args; + max_clients->filename = temp.filename; + max_clients->line_num = temp.line_num; + break; + } + } + else if (!max_clients + && strncasecmp(pdir->directive, "MaxClients", 10) == 0) { + max_clients = pdir; + } + } + + debug = ap_exists_config_define("DEBUG"); + + if (debug) { + foreground = one_process = 1; + no_detach = 0; + } + else { + one_process = ap_exists_config_define("ONE_PROCESS"); + no_detach = ap_exists_config_define("NO_DETACH"); + foreground = ap_exists_config_define("FOREGROUND"); + } + + /* sigh, want this only the second time around */ + if (restart_num++ == 1) { + is_graceful = 0; + + if (!one_process && !foreground) { + rv = apr_proc_detach(no_detach ? APR_PROC_DETACH_FOREGROUND + : APR_PROC_DETACH_DAEMONIZE); + if (rv != APR_SUCCESS) { + ap_log_error(APLOG_MARK, APLOG_CRIT, rv, NULL, + "apr_proc_detach failed"); + return HTTP_INTERNAL_SERVER_ERROR; + } + } + parent_pid = ap_my_pid = getpid(); + } + + unixd_pre_config(ptemp); + ap_listen_pre_config(); + ap_daemons_to_start = DEFAULT_START_DAEMON; + min_spare_threads = DEFAULT_MIN_FREE_DAEMON * DEFAULT_THREADS_PER_CHILD; + max_spare_threads = DEFAULT_MAX_FREE_DAEMON * DEFAULT_THREADS_PER_CHILD; + ap_daemons_limit = server_limit; + ap_threads_per_child = DEFAULT_THREADS_PER_CHILD; + ap_pid_fname = DEFAULT_PIDLOG; + ap_lock_fname = DEFAULT_LOCKFILE; + ap_max_requests_per_child = DEFAULT_MAX_REQUESTS_PER_CHILD; + ap_extended_status = 0; +#ifdef AP_MPM_WANT_SET_MAX_MEM_FREE + ap_max_mem_free = APR_ALLOCATOR_MAX_FREE_UNLIMITED; +#endif + + apr_cpystrn(ap_coredump_dir, ap_server_root, sizeof(ap_coredump_dir)); + + return OK; +} + +static void threadpool_hooks(apr_pool_t *p) +{ + /* The worker open_logs phase must run before the core's, or stderr + * will be redirected to a file, and the messages won't print to the + * console. + */ + static const char *const aszSucc[] = {"core.c", NULL}; + one_process = 0; + + ap_hook_open_logs(worker_open_logs, NULL, aszSucc, APR_HOOK_MIDDLE); + /* we need to set the MPM state before other pre-config hooks use MPM query + * to retrieve it, so register as REALLY_FIRST + */ + ap_hook_pre_config(worker_pre_config, NULL, NULL, APR_HOOK_REALLY_FIRST); +} + +static const char *set_daemons_to_start(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + ap_daemons_to_start = atoi(arg); + return NULL; +} + +static const char *set_min_spare_threads(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + min_spare_threads = atoi(arg); + if (min_spare_threads <= 0) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: detected MinSpareThreads set to non-positive."); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "Resetting to 1 to avoid almost certain Apache failure."); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "Please read the documentation."); + min_spare_threads = 1; + } + + return NULL; +} + +static const char *set_max_spare_threads(cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + max_spare_threads = atoi(arg); + return NULL; +} + +static const char *set_max_clients (cmd_parms *cmd, void *dummy, + const char *arg) +{ + int max_clients; + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + /* It is ok to use ap_threads_per_child here because we are + * sure that it gets set before MaxClients in the pre_config stage. */ + max_clients = atoi(arg); + if (max_clients < ap_threads_per_child) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: MaxClients (%d) must be at least as large", + max_clients); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " large as ThreadsPerChild (%d). Automatically", + ap_threads_per_child); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " increasing MaxClients to %d.", + ap_threads_per_child); + max_clients = ap_threads_per_child; + } + ap_daemons_limit = max_clients / ap_threads_per_child; + if ((max_clients > 0) && (max_clients % ap_threads_per_child)) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: MaxClients (%d) is not an integer multiple", + max_clients); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " of ThreadsPerChild (%d), lowering MaxClients to %d", + ap_threads_per_child, + ap_daemons_limit * ap_threads_per_child); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " for a maximum of %d child processes,", + ap_daemons_limit); + max_clients = ap_daemons_limit * ap_threads_per_child; + } + if (ap_daemons_limit > server_limit) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: MaxClients of %d would require %d servers,", + max_clients, ap_daemons_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " and would exceed the ServerLimit value of %d.", + server_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " Automatically lowering MaxClients to %d. To increase,", + server_limit * ap_threads_per_child); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " please see the ServerLimit directive."); + ap_daemons_limit = server_limit; + } + else if (ap_daemons_limit < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require MaxClients > 0, setting to 1"); + ap_daemons_limit = 1; + } + return NULL; +} + +static const char *set_threads_per_child (cmd_parms *cmd, void *dummy, + const char *arg) +{ + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + ap_threads_per_child = atoi(arg); + if (ap_threads_per_child > thread_limit) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: ThreadsPerChild of %d exceeds ThreadLimit " + "value of %d", ap_threads_per_child, + thread_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "threads, lowering ThreadsPerChild to %d. To increase, please" + " see the", thread_limit); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " ThreadLimit directive."); + ap_threads_per_child = thread_limit; + } + else if (ap_threads_per_child < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require ThreadsPerChild > 0, setting to 1"); + ap_threads_per_child = 1; + } + return NULL; +} + +static const char *set_server_limit (cmd_parms *cmd, void *dummy, const char *arg) +{ + int tmp_server_limit; + + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + tmp_server_limit = atoi(arg); + /* you cannot change ServerLimit across a restart; ignore + * any such attempts + */ + if (first_server_limit && + tmp_server_limit != server_limit) { + /* how do we log a message? the error log is a bit bucket at this + * point; we'll just have to set a flag so that ap_mpm_run() + * logs a warning later + */ + changed_limit_at_restart = 1; + return NULL; + } + server_limit = tmp_server_limit; + + if (server_limit > MAX_SERVER_LIMIT) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: ServerLimit of %d exceeds compile time limit " + "of %d servers,", server_limit, MAX_SERVER_LIMIT); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " lowering ServerLimit to %d.", MAX_SERVER_LIMIT); + server_limit = MAX_SERVER_LIMIT; + } + else if (server_limit < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require ServerLimit > 0, setting to 1"); + server_limit = 1; + } + return NULL; +} + +static const char *set_thread_limit (cmd_parms *cmd, void *dummy, const char *arg) +{ + int tmp_thread_limit; + + const char *err = ap_check_cmd_context(cmd, GLOBAL_ONLY); + if (err != NULL) { + return err; + } + + tmp_thread_limit = atoi(arg); + /* you cannot change ThreadLimit across a restart; ignore + * any such attempts + */ + if (first_thread_limit && + tmp_thread_limit != thread_limit) { + /* how do we log a message? the error log is a bit bucket at this + * point; we'll just have to set a flag so that ap_mpm_run() + * logs a warning later + */ + changed_limit_at_restart = 1; + return NULL; + } + thread_limit = tmp_thread_limit; + + if (thread_limit > MAX_THREAD_LIMIT) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: ThreadLimit of %d exceeds compile time limit " + "of %d servers,", thread_limit, MAX_THREAD_LIMIT); + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + " lowering ThreadLimit to %d.", MAX_THREAD_LIMIT); + thread_limit = MAX_THREAD_LIMIT; + } + else if (thread_limit < 1) { + ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, + "WARNING: Require ThreadLimit > 0, setting to 1"); + thread_limit = 1; + } + return NULL; +} + +static const command_rec threadpool_cmds[] = { +UNIX_DAEMON_COMMANDS, +LISTEN_COMMANDS, +AP_INIT_TAKE1("StartServers", set_daemons_to_start, NULL, RSRC_CONF, + "Number of child processes launched at server startup"), +AP_INIT_TAKE1("MinSpareThreads", set_min_spare_threads, NULL, RSRC_CONF, + "Minimum number of idle children, to handle request spikes"), +AP_INIT_TAKE1("MaxSpareThreads", set_max_spare_threads, NULL, RSRC_CONF, + "Maximum number of idle children"), +AP_INIT_TAKE1("MaxClients", set_max_clients, NULL, RSRC_CONF, + "Maximum number of children alive at the same time"), +AP_INIT_TAKE1("ThreadsPerChild", set_threads_per_child, NULL, RSRC_CONF, + "Number of threads each child creates"), +AP_INIT_TAKE1("ServerLimit", set_server_limit, NULL, RSRC_CONF, + "Maximum value of MaxClients for this run of Apache"), +AP_INIT_TAKE1("ThreadLimit", set_thread_limit, NULL, RSRC_CONF, + "Maximum worker threads in a server for this run of Apache"), +{ NULL } +}; + +module AP_MODULE_DECLARE_DATA mpm_threadpool_module = { + MPM20_MODULE_STUFF, + ap_mpm_rewrite_args, /* hook to run before apache parses args */ + NULL, /* create per-directory config structure */ + NULL, /* merge per-directory config structures */ + NULL, /* create per-server config structure */ + NULL, /* merge per-server config structures */ + threadpool_cmds, /* command apr_table_t */ + threadpool_hooks /* register_hooks */ +}; + |