/*
- * Copyright (c) 1999-2006 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 1999-2008 Apple Computer, Inc. All rights reserved.
*
* @APPLE_APACHE_LICENSE_HEADER_START@
*
* @APPLE_APACHE_LICENSE_HEADER_END@
*/
-static const char *const __rcs_file_version__ = "$Revision: 23748 $";
+static const char *const __rcs_file_version__ = "$Revision: 24912 $";
#include "config.h"
#include "launchd_runtime.h"
#include <mach/mach_interface.h>
#include <mach/host_info.h>
#include <mach/mach_host.h>
+#include <mach/mach_time.h>
#include <mach/exception.h>
#include <sys/types.h>
#include <sys/stat.h>
+#include <sys/sysctl.h>
#include <sys/time.h>
#include <sys/proc.h>
#include <sys/event.h>
#include <sys/mount.h>
#include <sys/reboot.h>
#include <sys/fcntl.h>
+#include <sys/kdebug.h>
#include <bsm/libbsm.h>
#include <malloc/malloc.h>
#include <unistd.h>
#include "launchd_internalServer.h"
#include "launchd_internal.h"
#include "notifyServer.h"
-#include "excServer.h"
+#include "mach_excServer.h"
/* We shouldn't be including these */
#include "launch.h"
#include "launchd.h"
#include "launchd_core_logic.h"
+#include "vproc.h"
+#include "vproc_priv.h"
#include "vproc_internal.h"
+#include "protocol_vprocServer.h"
#include "protocol_job_reply.h"
+#if !TARGET_OS_EMBEDDED
+#include "domainServer.h"
+#endif
+#include "eventsServer.h"
+
static mach_port_t ipc_port_set;
static mach_port_t demand_port_set;
static mach_port_t launchd_internal_port;
static int bulk_kev_cnt;
static pthread_t kqueue_demand_thread;
-static pthread_t demand_thread;
-static void *mport_demand_loop(void *arg);
+static void mportset_callback(void);
+static kq_callback kqmportset_callback = (kq_callback)mportset_callback;
static void *kqueue_demand_loop(void *arg);
-static void log_kevent_struct(int level, struct kevent *kev, int indx);
-static boolean_t launchd_internal_demux(mach_msg_header_t *Request, mach_msg_header_t *Reply);
+boolean_t launchd_internal_demux(mach_msg_header_t *Request, mach_msg_header_t *Reply);
static void record_caller_creds(mach_msg_header_t *mh);
static void launchd_runtime2(mach_msg_size_t msg_size, mig_reply_error_t *bufRequest, mig_reply_error_t *bufReply);
static mach_msg_size_t max_msg_size;
static size_t mig_cb_table_sz;
static timeout_callback runtime_idle_callback;
static mach_msg_timeout_t runtime_idle_timeout;
-static audit_token_t *au_tok;
-static size_t runtime_busy_cnt;
-
+static struct ldcred ldc;
+static size_t runtime_standby_cnt;
static STAILQ_HEAD(, logmsg_s) logmsg_queue = STAILQ_HEAD_INITIALIZER(logmsg_queue);
static size_t logmsg_queue_sz;
static mach_port_t drain_reply_port;
static void runtime_log_uncork_pending_drain(void);
static kern_return_t runtime_log_pack(vm_offset_t *outval, mach_msg_type_number_t *outvalCnt);
-static void runtime_log_push(void);
static bool logmsg_add(struct runtime_syslog_attr *attr, int err_num, const char *msg);
static void logmsg_remove(struct logmsg_s *lm);
+static void do_file_init(void) __attribute__((constructor));
+static mach_timebase_info_data_t tbi;
+static uint64_t tbi_safe_math_max;
+static uint64_t time_of_mach_msg_return;
+static double tbi_float_val;
static const int sigigns[] = { SIGHUP, SIGINT, SIGPIPE, SIGALRM, SIGTERM,
SIGURG, SIGTSTP, SIGTSTP, SIGCONT, SIGTTIN, SIGTTOU, SIGIO, SIGXCPU,
SIGXFSZ, SIGVTALRM, SIGPROF, SIGWINCH, SIGINFO, SIGUSR1, SIGUSR2
};
static sigset_t sigign_set;
+static FILE *ourlogfile;
+bool pid1_magic;
+bool do_apple_internal_logging;
+bool low_level_debug;
+bool g_flat_mach_namespace = true;
+bool g_simulate_pid1_crash = false;
+bool g_malloc_log_stacks = false;
+bool g_use_gmalloc = false;
+bool g_log_per_user_shutdown = false;
+#if !TARGET_OS_EMBEDDED
+bool g_log_pid1_shutdown = true;
+#else
+bool g_log_pid1_shutdown = false;
+#endif
+bool g_log_strict_usage = false;
+bool g_trap_sigkill_bugs = false;
+pid_t g_wsp = 0;
+size_t runtime_busy_cnt;
mach_port_t
runtime_get_kernel_port(void)
return launchd_internal_port;
}
+// static const char *__crashreporter_info__ = "";
+
+static int internal_mask_pri = LOG_UPTO(LOG_NOTICE);
+
+
void
launchd_runtime_init(void)
{
mach_msg_size_t mxmsgsz;
- pthread_attr_t attr;
+ pid_t p = getpid();
launchd_assert((mainkq = kqueue()) != -1);
launchd_assert((errno = mach_port_allocate(mach_task_self(), MACH_PORT_RIGHT_PORT_SET, &demand_port_set)) == KERN_SUCCESS);
launchd_assert((errno = mach_port_allocate(mach_task_self(), MACH_PORT_RIGHT_PORT_SET, &ipc_port_set)) == KERN_SUCCESS);
+ launchd_assert(kevent_mod(demand_port_set, EVFILT_MACHPORT, EV_ADD, 0, 0, &kqmportset_callback) != -1);
+
launchd_assert(launchd_mport_create_recv(&launchd_internal_port) == KERN_SUCCESS);
launchd_assert(launchd_mport_make_send(launchd_internal_port) == KERN_SUCCESS);
}
launchd_assert(runtime_add_mport(launchd_internal_port, launchd_internal_demux, mxmsgsz) == KERN_SUCCESS);
+ launchd_assert(pthread_create(&kqueue_demand_thread, NULL, kqueue_demand_loop, NULL) == 0);
+ launchd_assert(pthread_detach(kqueue_demand_thread) == 0);
- pthread_attr_init(&attr);
- pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
- pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN);
- launchd_assert(pthread_create(&kqueue_demand_thread, &attr, kqueue_demand_loop, NULL) == 0);
- pthread_attr_destroy(&attr);
-
- pthread_attr_init(&attr);
- pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
- pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN);
- launchd_assert(pthread_create(&demand_thread, &attr, mport_demand_loop, NULL) == 0);
- pthread_attr_destroy(&attr);
+ (void)launchd_assumes(sysctlbyname("vfs.generic.noremotehang", NULL, NULL, &p, sizeof(p)) != -1);
}
void
for (i = 0; i < (sizeof(sigigns) / sizeof(int)); i++) {
sigaddset(&sigign_set, sigigns[i]);
- launchd_assumes(signal(sigigns[i], SIG_IGN) != SIG_ERR);
- }
-}
-
-void *
-mport_demand_loop(void *arg __attribute__((unused)))
-{
- mach_msg_empty_rcv_t dummy;
- kern_return_t kr;
-
- for (;;) {
- kr = mach_msg(&dummy.header, MACH_RCV_MSG|MACH_RCV_LARGE, 0, 0, demand_port_set, 0, MACH_PORT_NULL);
- if (kr == MACH_RCV_PORT_CHANGED) {
- break;
- } else if (!launchd_assumes(kr == MACH_RCV_TOO_LARGE)) {
- continue;
- }
- launchd_assumes(handle_mport(launchd_internal_port) == 0);
+ (void)launchd_assumes(signal(sigigns[i], SIG_IGN) != SIG_ERR);
}
-
- return NULL;
}
-const char *
-proc_flags_to_C_names(unsigned int flags)
-{
-#define MAX_PFLAG_STR "P_ADVLOCK|P_CONTROLT|P_LP64|P_NOCLDSTOP|P_PPWAIT|P_PROFIL|P_SELECT|P_CONTINUED|P_SUGID|P_SYSTEM|P_TIMEOUT|P_TRACED|P_RESV3|P_WEXIT|P_EXEC|P_OWEUPC|P_AFFINITY|P_TRANSLATED|P_RESV5|P_CHECKOPENEVT|P_DEPENDENCY_CAPABLE|P_REBOOT|P_TBE|P_RESV7|P_THCWD|P_RESV9|P_RESV10|P_RESV11|P_NOSHLIB|P_FORCEQUOTA|P_NOCLDWAIT|P_NOREMOTEHANG|0xdeadbeeffeedface"
-
- static char flags_buf[sizeof(MAX_PFLAG_STR)];
- char *flags_off = NULL;
-
- if (!flags) {
- return "";
- }
-
- while (flags) {
- if (flags_off) {
- *flags_off = '|';
- flags_off++;
- *flags_off = '\0';
- } else {
- flags_off = flags_buf;
- }
-
#define FLAGIF(f) if (flags & f) { flags_off += sprintf(flags_off, #f); flags &= ~f; }
-
- FLAGIF(P_ADVLOCK)
- else FLAGIF(P_CONTROLT)
- else FLAGIF(P_LP64)
- else FLAGIF(P_NOCLDSTOP)
- else FLAGIF(P_PPWAIT)
- else FLAGIF(P_PROFIL)
- else FLAGIF(P_SELECT)
- else FLAGIF(P_CONTINUED)
- else FLAGIF(P_SUGID)
- else FLAGIF(P_SYSTEM)
- else FLAGIF(P_TIMEOUT)
- else FLAGIF(P_TRACED)
- else FLAGIF(P_RESV3)
- else FLAGIF(P_WEXIT)
- else FLAGIF(P_EXEC)
- else FLAGIF(P_OWEUPC)
- else FLAGIF(P_AFFINITY)
- else FLAGIF(P_TRANSLATED)
- else FLAGIF(P_RESV5)
- else FLAGIF(P_CHECKOPENEVT)
- else FLAGIF(P_DEPENDENCY_CAPABLE)
- else FLAGIF(P_REBOOT)
- else FLAGIF(P_TBE)
- else FLAGIF(P_RESV7)
- else FLAGIF(P_THCWD)
- else FLAGIF(P_RESV9)
- else FLAGIF(P_RESV10)
- else FLAGIF(P_RESV11)
- else FLAGIF(P_NOSHLIB)
- else FLAGIF(P_FORCEQUOTA)
- else FLAGIF(P_NOCLDWAIT)
- else FLAGIF(P_NOREMOTEHANG)
- else {
- flags_off += sprintf(flags_off, "0x%x", flags);
- flags = 0;
- }
- }
-
- return flags_buf;
-}
-
const char *
reboot_flags_to_C_names(unsigned int flags)
{
-#define MAX_RB_STR "RB_ASKNAME|RB_SINGLE|RB_NOSYNC|RB_KDB|RB_HALT|RB_INITNAME|RB_DFLTROOT|RB_ALTBOOT|RB_UNIPROC|RB_SAFEBOOT|RB_UPSDELAY|0xdeadbeeffeedface"
+#define MAX_RB_STR "RB_ASKNAME|RB_SINGLE|RB_NOSYNC|RB_HALT|RB_INITNAME|RB_DFLTROOT|RB_ALTBOOT|RB_UNIPROC|RB_SAFEBOOT|RB_UPSDELAY|0xdeadbeeffeedface"
static char flags_buf[sizeof(MAX_RB_STR)];
char *flags_off = NULL;
}
void
-log_kevent_struct(int level, struct kevent *kev, int indx)
+log_kevent_struct(int level, struct kevent *kev_base, int indx)
{
+ struct kevent *kev = &kev_base[indx];
const char *filter_str;
char ident_buf[100];
char filter_buf[100];
unsigned short flags = kev->flags;
unsigned int fflags = kev->fflags;
+ if (likely(!(LOG_MASK(level) & internal_mask_pri))) {
+ return;
+ }
+
if (flags) while (flags) {
if (flags_off) {
*flags_off = '|';
else FLAGIF(EV_ONESHOT)
else FLAGIF(EV_ERROR)
else {
- flags_off += sprintf(flags_off, "0x%x", flags);
+ flags_off += sprintf(flags_off, "0x%hx", flags);
flags = 0;
}
}
}
break;
default:
- snprintf(filter_buf, sizeof(filter_buf), "%d", kev->filter);
+ snprintf(filter_buf, sizeof(filter_buf), "%hd", kev->filter);
filter_str = filter_buf;
break;
}
indx, kev->udata, kev->data, ident_buf, filter_str, flags_buf, fflags_buf);
}
-kern_return_t
-x_handle_mport(mach_port_t junk __attribute__((unused)))
+void
+mportset_callback(void)
{
mach_port_name_array_t members;
mach_msg_type_number_t membersCnt;
unsigned int i;
if (!launchd_assumes((errno = mach_port_get_set_status(mach_task_self(), demand_port_set, &members, &membersCnt)) == KERN_SUCCESS)) {
- return 1;
+ return;
}
for (i = 0; i < membersCnt; i++) {
(*((kq_callback *)kev.udata))(kev.udata, &kev);
#if 0
} else {
- log_kevent_struct(LOG_ERR, &kev);
+ log_kevent_struct(LOG_ERR, &kev, 0);
}
#endif
/* the callback may have tainted our ability to continue this for loop */
}
}
- launchd_assumes(vm_deallocate(mach_task_self(), (vm_address_t)members,
+ (void)launchd_assumes(vm_deallocate(mach_task_self(), (vm_address_t)members,
(vm_size_t) membersCnt * sizeof(mach_port_name_t)) == KERN_SUCCESS);
-
- return 0;
}
void *
FD_ZERO(&rfds);
FD_SET(mainkq, &rfds);
if (launchd_assumes(select(mainkq + 1, &rfds, NULL, NULL, NULL) == 1)) {
- launchd_assumes(handle_kqueue(launchd_internal_port, mainkq) == 0);
+ (void)launchd_assumes(handle_kqueue(launchd_internal_port, mainkq) == 0);
}
}
x_handle_kqueue(mach_port_t junk __attribute__((unused)), integer_t fd)
{
struct timespec ts = { 0, 0 };
- struct kevent kev[BULK_KEV_MAX];
+ struct kevent *kevi, kev[BULK_KEV_MAX];
int i;
bulk_kev = kev;
- launchd_assumes((bulk_kev_cnt = kevent(fd, NULL, 0, kev, BULK_KEV_MAX, &ts)) != -1);
-
- if (bulk_kev_cnt > 0) {
-#if 0
- Dl_info dli;
-
- if (launchd_assumes(malloc_size(kev.udata) || dladdr(kev.udata, &dli))) {
-#endif
+ if (launchd_assumes((bulk_kev_cnt = kevent(fd, NULL, 0, kev, BULK_KEV_MAX, &ts)) != -1)) {
+#if 0
for (i = 0; i < bulk_kev_cnt; i++) {
- log_kevent_struct(LOG_DEBUG, &kev[i], i);
+ log_kevent_struct(LOG_DEBUG, &kev[0], i);
}
+#endif
for (i = 0; i < bulk_kev_cnt; i++) {
bulk_kev_i = i;
- if (kev[i].filter) {
- (*((kq_callback *)kev[i].udata))(kev[i].udata, &kev[i]);
- }
- }
+ kevi = &kev[i];
+
+ if (kevi->filter) {
+ runtime_syslog(LOG_DEBUG, "Dispatching kevent...");
+ log_kevent_struct(LOG_DEBUG, kev, i);
#if 0
- } else {
- log_kevent_struct(LOG_ERR, &kev);
- }
+ /* Check if kevi->udata was either malloc(3)ed or is a valid function pointer.
+ * If neither, it's probably an invalid pointer and we should log it.
+ */
+ Dl_info dli;
+ if (launchd_assumes(malloc_size(kevi->udata) || dladdr(kevi->udata, &dli))) {
+ runtime_ktrace(RTKT_LAUNCHD_BSD_KEVENT|DBG_FUNC_START, kevi->ident, kevi->filter, kevi->fflags);
+ (*((kq_callback *)kevi->udata))(kevi->udata, kevi);
+ runtime_ktrace0(RTKT_LAUNCHD_BSD_KEVENT|DBG_FUNC_END);
+ } else {
+ runtime_syslog(LOG_ERR, "The following kevent had invalid context data.");
+ log_kevent_struct(LOG_EMERG, &kev[0], i);
+ }
+#else
+ runtime_ktrace(RTKT_LAUNCHD_BSD_KEVENT|DBG_FUNC_START, kevi->ident, kevi->filter, kevi->fflags);
+ (*((kq_callback *)kevi->udata))(kevi->udata, kevi);
+ runtime_ktrace0(RTKT_LAUNCHD_BSD_KEVENT|DBG_FUNC_END);
#endif
+ }
+ }
}
bulk_kev = NULL;
return 0;
}
-
-
void
launchd_runtime(void)
{
int flags = VM_MAKE_TAG(VM_MEMORY_MACH_MSG)|TRUE;
for (;;) {
- if (req) {
- launchd_assumes(vm_deallocate(mach_task_self(), (vm_address_t)req, mz) == KERN_SUCCESS);
+ if (likely(req)) {
+ (void)launchd_assumes(vm_deallocate(mach_task_self(), (vm_address_t)req, mz) == KERN_SUCCESS);
req = NULL;
}
- if (resp) {
- launchd_assumes(vm_deallocate(mach_task_self(), (vm_address_t)resp, mz) == KERN_SUCCESS);
+ if (likely(resp)) {
+ (void)launchd_assumes(vm_deallocate(mach_task_self(), (vm_address_t)resp, mz) == KERN_SUCCESS);
resp = NULL;
}
if (which == MACH_NOTIFY_NO_SENDERS) {
/* Always make sure the send count is zero, in case a receive right is reused */
errno = mach_port_set_mscount(mach_task_self(), name, 0);
- if (errno != KERN_SUCCESS) {
+ if (unlikely(errno != KERN_SUCCESS)) {
return errno;
}
}
errno = mach_port_request_notification(mach_task_self(), name, which, msgc, where,
MACH_MSG_TYPE_MAKE_SEND_ONCE, &previous);
- if (errno == 0 && previous != MACH_PORT_NULL) {
- launchd_assumes(launchd_mport_deallocate(previous) == KERN_SUCCESS);
+ if (likely(errno == 0) && previous != MACH_PORT_NULL) {
+ (void)launchd_assumes(launchd_mport_deallocate(previous) == KERN_SUCCESS);
}
return errno;
sigemptyset(&emptyset);
- launchd_assumes(launchd_mport_make_send(bsport) == KERN_SUCCESS);
- launchd_assumes(launchd_set_bport(bsport) == KERN_SUCCESS);
- launchd_assumes(launchd_mport_deallocate(bsport) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_mport_make_send(bsport) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_set_bport(bsport) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_mport_deallocate(bsport) == KERN_SUCCESS);
- launchd_assumes(sigprocmask(SIG_BLOCK, &sigign_set, &oset) != -1);
+ (void)launchd_assumes(sigprocmask(SIG_BLOCK, &sigign_set, &oset) != -1);
for (i = 0; i < (sizeof(sigigns) / sizeof(int)); i++) {
- launchd_assumes(signal(sigigns[i], SIG_DFL) != SIG_ERR);
+ (void)launchd_assumes(signal(sigigns[i], SIG_DFL) != SIG_ERR);
}
r = fork();
if (r != 0) {
for (i = 0; i < (sizeof(sigigns) / sizeof(int)); i++) {
- launchd_assumes(signal(sigigns[i], SIG_IGN) != SIG_ERR);
+ (void)launchd_assumes(signal(sigigns[i], SIG_IGN) != SIG_ERR);
}
- launchd_assumes(sigprocmask(SIG_SETMASK, &oset, NULL) != -1);
- launchd_assumes(launchd_set_bport(MACH_PORT_NULL) == KERN_SUCCESS);
+ (void)launchd_assumes(sigprocmask(SIG_SETMASK, &oset, NULL) != -1);
+ (void)launchd_assumes(launchd_set_bport(MACH_PORT_NULL) == KERN_SUCCESS);
} else {
- launchd_assumes(sigprocmask(SIG_SETMASK, &emptyset, NULL) != -1);
+ pid_t p = -getpid();
+ (void)launchd_assumes(sysctlbyname("vfs.generic.noremotehang", NULL, NULL, &p, sizeof(p)) != -1);
+
+ (void)launchd_assumes(sigprocmask(SIG_SETMASK, &emptyset, NULL) != -1);
}
errno = saved_errno;
msg_size = round_page(msg_size + MAX_TRAILER_SIZE);
- if (needed_table_sz > mig_cb_table_sz) {
+ if (unlikely(needed_table_sz > mig_cb_table_sz)) {
needed_table_sz *= 2; /* Let's try and avoid realloc'ing for a while */
mig_callback *new_table = malloc(needed_table_sz);
return KERN_RESOURCE_SHORTAGE;
}
- if (mig_cb_table) {
+ if (likely(mig_cb_table)) {
memcpy(new_table, mig_cb_table, mig_cb_table_sz);
free(mig_cb_table);
}
return errno = mach_port_insert_right(mach_task_self(), name, name, MACH_MSG_TYPE_MAKE_SEND);
}
+kern_return_t
+launchd_mport_copy_send(mach_port_t name)
+{
+ return errno = mach_port_insert_right(mach_task_self(), name, name, MACH_MSG_TYPE_COPY_SEND);
+}
+
+kern_return_t
+launchd_mport_make_send_once(mach_port_t name, mach_port_t *so)
+{
+ mach_msg_type_name_t right = 0;
+ return errno = mach_port_extract_right(mach_task_self(), name, MACH_MSG_TYPE_MAKE_SEND_ONCE, so, &right);
+}
+
kern_return_t
launchd_mport_close_recv(mach_port_t name)
{
case EVFILT_READ:
case EVFILT_WRITE:
break;
+ case EVFILT_TIMER:
+ /* Workaround 5225889 */
+ if (flags & EV_ADD) {
+ kevent_mod(ident, EVFILT_TIMER, EV_DELETE, 0, 0, NULL);
+ }
+ /* fall through */
default:
flags |= EV_CLEAR;
break;
if (flags & EV_ADD && !launchd_assumes(udata != NULL)) {
errno = EINVAL;
return -1;
+ } else if ((flags & EV_DELETE) && bulk_kev) {
+ int i = 0;
+ for (i = bulk_kev_i + 1; i < bulk_kev_cnt; i++) {
+ if (bulk_kev[i].filter == filter && bulk_kev[i].ident == ident) {
+ runtime_syslog(LOG_DEBUG, "Pruning the following kevent:");
+ log_kevent_struct(LOG_DEBUG, &bulk_kev[0], i);
+ bulk_kev[i].filter = (short)0;
+ }
+ }
}
EV_SET(&kev, ident, filter, flags, fflags, data, udata);
} else if (notify_server_routine(Request)) {
return notify_server(Request, Reply);
} else {
- return exc_server(Request, Reply);
+ return mach_exc_server(Request, Reply);
}
}
kern_return_t
-do_mach_notify_port_destroyed(mach_port_t notify __attribute__((unused)),
- mach_port_t rights)
+do_mach_notify_port_destroyed(mach_port_t notify __attribute__((unused)), mach_port_t rights)
{
/* This message is sent to us when a receive right is returned to us. */
if (!launchd_assumes(job_ack_port_destruction(rights))) {
- launchd_assumes(launchd_mport_close_recv(rights) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_mport_close_recv(rights) == KERN_SUCCESS);
}
return KERN_SUCCESS;
}
kern_return_t
-do_mach_notify_port_deleted(mach_port_t notify __attribute__((unused)),
- mach_port_name_t name __attribute__((unused)))
+do_mach_notify_port_deleted(mach_port_t notify __attribute__((unused)), mach_port_name_t name __attribute__((unused)))
{
/* If we deallocate/destroy/mod_ref away a port with a pending
* notification, the original notification message is replaced with
}
kern_return_t
-do_mach_notify_no_senders(mach_port_t notify,
- mach_port_mscount_t mscount __attribute__((unused)))
+do_mach_notify_no_senders(mach_port_t notify, mach_port_mscount_t mscount __attribute__((unused)))
{
job_t j = job_mig_intran(notify);
kern_return_t
do_mach_notify_send_once(mach_port_t notify __attribute__((unused)))
{
- /* This message is sent to us every time we close a port that we have
- * outstanding Mach notification requests on. We can safely ignore this
- * message.
+ /*
+ * This message is sent for each send-once right that is deallocated
+ * without being used.
*/
return KERN_SUCCESS;
}
kern_return_t
-do_mach_notify_dead_name(mach_port_t notify __attribute__((unused)),
- mach_port_name_t name)
+do_mach_notify_dead_name(mach_port_t notify __attribute__((unused)), mach_port_name_t name)
{
/* This message is sent to us when one of our send rights no longer has
* a receiver somewhere else on the system.
*/
if (name == drain_reply_port) {
- launchd_assumes(launchd_mport_deallocate(name) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_mport_deallocate(name) == KERN_SUCCESS);
drain_reply_port = MACH_PORT_NULL;
}
* rights on said port. Let's deallocate it so that we don't leak
* dead-name ports.
*/
- launchd_assumes(launchd_mport_deallocate(name) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_mport_deallocate(name) == KERN_SUCCESS);
return KERN_SUCCESS;
}
trailer_size = tp->msgh_trailer_size - (mach_msg_size_t)(sizeof(mach_msg_trailer_type_t) - sizeof(mach_msg_trailer_size_t));
- if (trailer_size < (mach_msg_size_t)sizeof(audit_token_t)) {
- au_tok = NULL;
- return;
+ if (launchd_assumes(trailer_size >= (mach_msg_size_t)sizeof(audit_token_t))) {
+ audit_token_to_au32(tp->msgh_audit, /* audit UID */ NULL, &ldc.euid,
+ &ldc.egid, &ldc.uid, &ldc.gid, &ldc.pid,
+ &ldc.asid, /* au_tid_t */ NULL);
}
- au_tok = &tp->msgh_audit;
}
-bool
-runtime_get_caller_creds(struct ldcred *ldc)
+struct ldcred *
+runtime_get_caller_creds(void)
{
- if (!au_tok) {
- return false;
- }
-
- audit_token_to_au32(*au_tok, /* audit UID */ NULL, &ldc->euid,
- &ldc->egid, &ldc->uid, &ldc->gid, &ldc->pid,
- &ldc->asid, /* au_tid_t */ NULL);
+ return &ldc;
+}
- return true;
+mach_msg_return_t
+launchd_exc_runtime_once(mach_port_t port, mach_msg_size_t rcv_msg_size, mach_msg_size_t send_msg_size, mig_reply_error_t *bufRequest, mig_reply_error_t *bufReply, mach_msg_timeout_t to)
+{
+ mach_msg_return_t mr = ~MACH_MSG_SUCCESS;
+ mach_msg_option_t rcv_options = MACH_RCV_MSG |
+ MACH_RCV_TIMEOUT |
+ MACH_RCV_TRAILER_ELEMENTS(MACH_RCV_TRAILER_AUDIT) |
+ MACH_RCV_TRAILER_TYPE(MACH_MSG_TRAILER_FORMAT_0) ;
+
+ do {
+ mr = mach_msg(&bufRequest->Head, rcv_options, 0, rcv_msg_size, port, to, MACH_PORT_NULL);
+ switch (mr) {
+ case MACH_RCV_TIMED_OUT :
+ runtime_syslog(LOG_DEBUG, "Message queue is empty.");
+ break;
+ case MACH_RCV_TOO_LARGE :
+ runtime_syslog(LOG_INFO, "Message is larger than %u bytes.", rcv_msg_size);
+ break;
+ default :
+ (void)launchd_assumes(mr == MACH_MSG_SUCCESS);
+ }
+
+ if (mr == MACH_MSG_SUCCESS) {
+ if (!launchd_assumes(mach_exc_server(&bufRequest->Head, &bufReply->Head) == TRUE)) {
+ runtime_syslog(LOG_WARNING, "Exception server routine failed.");
+ break;
+ }
+
+ mach_msg_return_t smr = ~MACH_MSG_SUCCESS;
+ mach_msg_option_t send_options = MACH_SEND_MSG |
+ MACH_SEND_TIMEOUT ;
+
+ (void)launchd_assumes(bufReply->Head.msgh_size <= send_msg_size);
+ smr = mach_msg(&bufReply->Head, send_options, bufReply->Head.msgh_size, 0, MACH_PORT_NULL, to + 100, MACH_PORT_NULL);
+ switch (smr) {
+ case MACH_SEND_TIMED_OUT :
+ runtime_syslog(LOG_WARNING, "Timed out while trying to send reply to exception message.");
+ break;
+ case MACH_SEND_INVALID_DEST :
+ runtime_syslog(LOG_WARNING, "Tried sending a message to a port that we don't possess a send right to.");
+ break;
+ default :
+ if (!launchd_assumes(smr == MACH_MSG_SUCCESS)) {
+ runtime_syslog(LOG_WARNING, "Couldn't deliver exception reply: 0x%x", smr);
+ }
+ break;
+ }
+ }
+ } while (0);
+
+ return mr;
}
void
mig_callback the_demux;
mach_msg_timeout_t to;
mach_msg_return_t mr;
+ size_t busy_cnt;
options = MACH_RCV_MSG|MACH_RCV_TRAILER_ELEMENTS(MACH_RCV_TRAILER_AUDIT) |
MACH_RCV_TRAILER_TYPE(MACH_MSG_TRAILER_FORMAT_0);
tmp_options = options;
for (;;) {
+ busy_cnt = runtime_busy_cnt + runtime_standby_cnt;
to = MACH_MSG_TIMEOUT_NONE;
- if (msg_size != max_msg_size) {
- /* The buffer isn't big enougth to receive messages anymore... */
+ if (unlikely(msg_size != max_msg_size)) {
+ /* The buffer isn't big enough to receive messages anymore... */
tmp_options &= ~MACH_RCV_MSG;
options &= ~MACH_RCV_MSG;
if (!(tmp_options & MACH_SEND_MSG)) {
}
}
- if ((tmp_options & MACH_RCV_MSG) && (runtime_idle_callback || (runtime_busy_cnt == 0))) {
+ if ((tmp_options & MACH_RCV_MSG) && (runtime_idle_callback || (busy_cnt == 0))) {
tmp_options |= MACH_RCV_TIMEOUT;
if (!(tmp_options & MACH_SEND_TIMEOUT)) {
- to = runtime_busy_cnt ? runtime_idle_timeout : (RUNTIME_ADVISABLE_IDLE_TIMEOUT * 1000);
+#if !TARGET_OS_EMBEDDED
+ to = busy_cnt ? runtime_idle_timeout : (_vproc_standby_timeout() * 1000);
+#else
+ to = runtime_idle_timeout;
+#endif
}
}
mr = mach_msg(&bufReply->Head, tmp_options, bufReply->Head.msgh_size,
msg_size, ipc_port_set, to, MACH_PORT_NULL);
+ time_of_mach_msg_return = runtime_get_opaque_time();
+
tmp_options = options;
- if (mr == MACH_SEND_INVALID_DEST || mr == MACH_SEND_TIMED_OUT) {
+ /* It looks like the compiler doesn't optimize switch(unlikely(...)) See: 5691066 */
+ if (unlikely(mr)) switch (mr) {
+ case MACH_SEND_INVALID_DEST:
+ case MACH_SEND_TIMED_OUT:
/* We need to clean up and start over. */
if (bufReply->Head.msgh_bits & MACH_MSGH_BITS_COMPLEX) {
mach_msg_destroy(&bufReply->Head);
}
continue;
- } else if (mr == MACH_RCV_TIMED_OUT) {
+ case MACH_RCV_TIMED_OUT:
if (to != MACH_MSG_TIMEOUT_NONE) {
- if (runtime_busy_cnt == 0) {
+ if (busy_cnt == 0) {
+ runtime_syslog(LOG_INFO, "Idle exiting.");
launchd_shutdown();
} else if (runtime_idle_callback) {
runtime_idle_callback();
}
}
continue;
- } else if (!launchd_assumes(mr == MACH_MSG_SUCCESS)) {
+ default:
+ if (!launchd_assumes(mr == MACH_MSG_SUCCESS)) {
+ runtime_syslog(LOG_ERR, "mach_msg(): %u: %s", mr, mach_error_string(mr));
+ }
continue;
}
bufRequest = bufReply;
bufReply = bufTemp;
- if (!(tmp_options & MACH_RCV_MSG)) {
+ if (unlikely(!(tmp_options & MACH_RCV_MSG))) {
continue;
}
/* we have another request message */
-
+#if 0
if (!launchd_assumes(mig_cb_table != NULL)) {
break;
}
+#endif
the_demux = mig_cb_table[MACH_PORT_INDEX(bufRequest->Head.msgh_local_port)];
+#if 0
if (!launchd_assumes(the_demux != NULL)) {
break;
}
+#endif
record_caller_creds(&bufRequest->Head);
-
- /*
- * This is a total hack. We really need a bit in the kernel's proc
- * struct to declare our intent.
- */
- static int no_hang_fd = -1;
- if (no_hang_fd == -1) {
- no_hang_fd = _fd(open("/dev/autofs_nowait", 0));
- }
+ runtime_ktrace(RTKT_LAUNCHD_MACH_IPC|DBG_FUNC_START, bufRequest->Head.msgh_local_port, bufRequest->Head.msgh_id, (long)the_demux);
if (the_demux(&bufRequest->Head, &bufReply->Head) == FALSE) {
/* XXX - also gross */
- if (bufRequest->Head.msgh_id == MACH_NOTIFY_NO_SENDERS) {
+ if (likely(bufRequest->Head.msgh_id == MACH_NOTIFY_NO_SENDERS)) {
notify_server(&bufRequest->Head, &bufReply->Head);
+ } else if (the_demux == protocol_vproc_server) {
+
+#if !TARGET_OS_EMBEDDED
+ /* Similarly gross. */
+ if (xpc_domain_server(&bufRequest->Head, &bufReply->Head) == FALSE) {
+ (void)xpc_events_server(&bufRequest->Head, &bufReply->Head);
+ }
+#else
+ (void)xpc_events_server(&bufRequest->Head, &bufReply->Head);
+#endif
}
}
+ runtime_ktrace(RTKT_LAUNCHD_MACH_IPC|DBG_FUNC_END, bufReply->Head.msgh_remote_port, bufReply->Head.msgh_bits, bufReply->RetCode);
+
+ /* bufReply is a union. If MACH_MSGH_BITS_COMPLEX is set, then bufReply->RetCode is assumed to be zero. */
if (!(bufReply->Head.msgh_bits & MACH_MSGH_BITS_COMPLEX)) {
- if (bufReply->RetCode == MIG_NO_REPLY) {
- bufReply->Head.msgh_remote_port = MACH_PORT_NULL;
- } else if ((bufReply->RetCode != KERN_SUCCESS) && (bufRequest->Head.msgh_bits & MACH_MSGH_BITS_COMPLEX)) {
- /* destroy the request - but not the reply port */
- bufRequest->Head.msgh_remote_port = MACH_PORT_NULL;
- mach_msg_destroy(&bufRequest->Head);
+ if (unlikely(bufReply->RetCode != KERN_SUCCESS)) {
+ if (likely(bufReply->RetCode == MIG_NO_REPLY)) {
+ bufReply->Head.msgh_remote_port = MACH_PORT_NULL;
+ } else if (bufRequest->Head.msgh_bits & MACH_MSGH_BITS_COMPLEX) {
+ /* destroy the request - but not the reply port */
+ bufRequest->Head.msgh_remote_port = MACH_PORT_NULL;
+ mach_msg_destroy(&bufRequest->Head);
+ }
}
}
- if (bufReply->Head.msgh_remote_port != MACH_PORT_NULL) {
+ if (likely(bufReply->Head.msgh_remote_port != MACH_PORT_NULL)) {
tmp_options |= MACH_SEND_MSG;
- if (MACH_MSGH_BITS_REMOTE(bufReply->Head.msgh_bits) != MACH_MSG_TYPE_MOVE_SEND_ONCE) {
+ if (unlikely(MACH_MSGH_BITS_REMOTE(bufReply->Head.msgh_bits) != MACH_MSG_TYPE_MOVE_SEND_ONCE)) {
tmp_options |= MACH_SEND_TIMEOUT;
}
}
case EVFILT_VNODE:
case EVFILT_WRITE:
case EVFILT_READ:
- if ((int)bulk_kev[i].ident == fd) {
+ if (unlikely((int)bulk_kev[i].ident == fd)) {
runtime_syslog(LOG_DEBUG, "Skipping kevent index: %d", i);
bulk_kev[i].filter = 0;
}
return close(fd);
}
-static FILE *ourlogfile;
-
void
runtime_closelog(void)
{
+ runtime_log_push();
+
if (ourlogfile) {
- launchd_assumes(fflush(ourlogfile) == 0);
- launchd_assumes(runtime_fsync(fileno(ourlogfile)) != -1);
+ (void)launchd_assumes(fflush(ourlogfile) == 0);
+ (void)launchd_assumes(runtime_fsync(fileno(ourlogfile)) != -1);
}
}
int
runtime_fsync(int fd)
{
- if (debug_shutdown_hangs) {
+#if 0
+ if (do_apple_internal_logging) {
return fcntl(fd, F_FULLFSYNC, NULL);
} else {
return fsync(fd);
}
+#else
+ return fsync(fd);
+#endif
}
-static int internal_mask_pri = LOG_UPTO(LOG_NOTICE);
-//static int internal_mask_pri = LOG_UPTO(LOG_DEBUG);
-
int
runtime_setlogmask(int maskpri)
{
runtime_syslog(int pri, const char *message, ...)
{
struct runtime_syslog_attr attr = {
- "com.apple.launchd", "com.apple.launchd",
- getpid() == 1 ? "System" : "Background",
- pri, getuid(), getpid(), getpid()
+ g_my_label,
+ g_my_label,
+ pid1_magic ? "System" : "Background",
+ pri,
+ getuid(),
+ getpid(),
+ getpid()
};
va_list ap;
va_start(ap, message);
-
runtime_vsyslog(&attr, message, ap);
va_end(ap);
void
runtime_vsyslog(struct runtime_syslog_attr *attr, const char *message, va_list args)
{
- static pthread_mutex_t ourlock = PTHREAD_MUTEX_INITIALIZER;
- static struct timeval shutdown_start;
- static struct timeval prev_msg;
- static int apple_internal_logging = 1;
- struct timeval tvnow, tvd_total, tvd_msg_delta = { 0, 0 };
- struct stat sb;
int saved_errno = errno;
char newmsg[10000];
- size_t i, j;
+ bool echo_to_console= false;
- if (!(LOG_MASK(attr->priority) & internal_mask_pri)) {
- goto out;
- }
-
- if (apple_internal_logging == 1) {
- apple_internal_logging = stat("/AppleInternal", &sb);
- }
-
-
- if (!(debug_shutdown_hangs && getpid() == 1)) {
- if (attr->priority == LOG_APPLEONLY) {
- if (apple_internal_logging == -1) {
- goto out;
- }
+ if (attr->priority == LOG_APPLEONLY) {
+ if (do_apple_internal_logging) {
attr->priority = LOG_NOTICE;
+ } else {
+ return;
}
- vsnprintf(newmsg, sizeof(newmsg), message, args);
- logmsg_add(attr, saved_errno, newmsg);
- goto out;
- }
-
- if (shutdown_start.tv_sec == 0) {
- gettimeofday(&shutdown_start, NULL);
- }
-
- if (gettimeofday(&tvnow, NULL) == -1) {
- tvnow.tv_sec = 0;
- tvnow.tv_usec = 0;
- }
-
- pthread_mutex_lock(&ourlock);
-
- if (ourlogfile == NULL) {
- rename("/var/log/launchd-shutdown.log", "/var/log/launchd-shutdown.log.1");
- ourlogfile = fopen("/var/log/launchd-shutdown.log", "a");
- }
-
- pthread_mutex_unlock(&ourlock);
-
- if (ourlogfile == NULL) {
- goto out;
+ } else if (attr->priority == LOG_SCOLDING) {
+ attr->priority = g_log_strict_usage ? LOG_NOTICE : LOG_DEBUG;
}
- if (message == NULL) {
- goto out;
+ if (attr->priority & LOG_CONSOLE) {
+ echo_to_console = true;
+ attr->priority &= ~LOG_CONSOLE;
}
- timersub(&tvnow, &shutdown_start, &tvd_total);
-
- if (prev_msg.tv_sec != 0) {
- timersub(&tvnow, &prev_msg, &tvd_msg_delta);
+ if (!(LOG_MASK(attr->priority) & internal_mask_pri)) {
+ return;
}
- prev_msg = tvnow;
-
- snprintf(newmsg, sizeof(newmsg), "%3ld.%06d%4ld.%06d%6u %-40s%6u %-40s ",
- tvd_total.tv_sec, tvd_total.tv_usec,
- tvd_msg_delta.tv_sec, tvd_msg_delta.tv_usec,
- attr->from_pid, attr->from_name,
- attr->about_pid, attr->about_name);
+ vsnprintf(newmsg, sizeof(newmsg), message, args);
- for (i = 0, j = strlen(newmsg); message[i];) {
- if (message[i] == '%' && message[i + 1] == 'm') {
- char *errs = strerror(saved_errno);
- strcpy(newmsg + j, errs ? errs : "unknown error");
- j += strlen(newmsg + j);
- i += 2;
- } else {
- newmsg[j] = message[i];
- j++;
- i++;
- }
+ if (g_console && (unlikely(low_level_debug) || echo_to_console)) {
+ fprintf(g_console, "%s %u\t%s %u\t%s\n", attr->from_name, attr->from_pid, attr->about_name, attr->about_pid, newmsg);
}
- strcpy(newmsg + j, "\n");
-
- vfprintf(ourlogfile, newmsg, args);
-
-out:
- runtime_log_uncork_pending_drain();
+ logmsg_add(attr, saved_errno, newmsg);
}
bool
/* we do this to make the unpacking for the log_drain cause unalignment faults */
lm_sz = ROUND_TO_64BIT_WORD_SIZE(lm_sz);
- if (!(lm = calloc(1, lm_sz))) {
+ if (unlikely((lm = calloc(1, lm_sz)) == NULL)) {
return false;
}
data_off = lm->data;
- launchd_assumes(gettimeofday(&lm->when, NULL) != -1);
+ lm->when = runtime_get_wall_time();
lm->from_pid = attr->from_pid;
lm->about_pid = attr->about_pid;
lm->err_num = err_num;
mig_allocate(outval, *outvalCnt);
- if (*outval == 0) {
+ if (unlikely(*outval == 0)) {
return 1;
}
offset = (void *)*outval;
+ if (g_log_per_user_shutdown && !ourlogfile && !pid1_magic && shutdown_in_progress) {
+ char logfile[NAME_MAX];
+ snprintf(logfile, sizeof(logfile), "/var/tmp/launchd-%s.shutdown.log", g_username);
+
+ char logfile1[NAME_MAX];
+ snprintf(logfile1, sizeof(logfile1), "/var/tmp/launchd-%s.shutdown.log.1", g_username);
+
+ rename(logfile, logfile1);
+ ourlogfile = fopen(logfile, "a");
+ }
+
+ static int64_t shutdown_start = 0;
+ if (shutdown_start == 0) {
+ shutdown_start = runtime_get_wall_time();
+ }
+
while ((lm = STAILQ_FIRST(&logmsg_queue))) {
- lm->from_name -= (size_t)lm;
- lm->about_name -= (size_t)lm;
- lm->msg -= (size_t)lm;
- lm->session_name -= (size_t)lm;
+ int64_t log_delta = lm->when - shutdown_start;
+ if (!pid1_magic && ourlogfile) {
+ fprintf(ourlogfile, "%8lld%6u %-40s%6u %-40s %s\n", log_delta,
+ lm->from_pid, lm->from_name, lm->about_pid, lm->about_name, lm->msg);
+ fflush(ourlogfile);
+ }
+
+ lm->from_name_offset = lm->from_name - (char *)lm;
+ lm->about_name_offset = lm->about_name - (char *)lm;
+ lm->msg_offset = lm->msg - (char *)lm;
+ lm->session_name_offset = lm->session_name - (char *)lm;
memcpy(offset, lm, lm->obj_sz);
logmsg_remove(lm);
}
+
+ if (ourlogfile) {
+ fflush(ourlogfile);
+ }
return 0;
}
tmp_port = drain_reply_port;
drain_reply_port = MACH_PORT_NULL;
- if ((errno = job_mig_log_drain_reply(tmp_port, 0, outval, outvalCnt))) {
- launchd_assumes(errno == MACH_SEND_INVALID_DEST);
- launchd_assumes(launchd_mport_deallocate(tmp_port) == KERN_SUCCESS);
+ if (unlikely(errno = job_mig_log_drain_reply(tmp_port, 0, outval, outvalCnt))) {
+ (void)launchd_assumes(errno == MACH_SEND_INVALID_DEST);
+ (void)launchd_assumes(launchd_mport_deallocate(tmp_port) == KERN_SUCCESS);
}
mig_deallocate(outval, outvalCnt);
void
runtime_log_push(void)
{
+ static pthread_mutex_t ourlock = PTHREAD_MUTEX_INITIALIZER;
+ static int64_t shutdown_start, log_delta;
mach_msg_type_number_t outvalCnt;
+ struct logmsg_s *lm;
vm_offset_t outval;
if (logmsg_queue_cnt == 0) {
- launchd_assumes(STAILQ_EMPTY(&logmsg_queue));
+ (void)launchd_assumes(STAILQ_EMPTY(&logmsg_queue));
return;
- } else if (getpid() == 1) {
+ } else if (!pid1_magic) {
+ if (runtime_log_pack(&outval, &outvalCnt) == 0) {
+ (void)launchd_assumes(_vprocmgr_log_forward(inherited_bootstrap_port, (void *)outval, outvalCnt) == NULL);
+ mig_deallocate(outval, outvalCnt);
+ }
return;
}
- if (runtime_log_pack(&outval, &outvalCnt) != 0) {
+ if (likely(!shutdown_in_progress && !fake_shutdown_in_progress)) {
+ runtime_log_uncork_pending_drain();
return;
}
- launchd_assumes(_vprocmgr_log_forward(inherited_bootstrap_port, (void *)outval, outvalCnt) == NULL);
+ if (unlikely(shutdown_start == 0)) {
+ shutdown_start = runtime_get_wall_time();
+ launchd_log_vm_stats();
+ }
- mig_deallocate(outval, outvalCnt);
+ pthread_mutex_lock(&ourlock);
+
+ if (unlikely(ourlogfile == NULL) && g_log_pid1_shutdown) {
+ rename("/var/log/launchd-shutdown.log", "/var/log/launchd-shutdown.log.1");
+ ourlogfile = fopen("/var/log/launchd-shutdown.log", "a");
+ }
+
+ pthread_mutex_unlock(&ourlock);
+
+ if (unlikely(!ourlogfile)) {
+ return;
+ }
+
+ while ((lm = STAILQ_FIRST(&logmsg_queue))) {
+ log_delta = lm->when - shutdown_start;
+
+ fprintf(ourlogfile, "%8lld%6u %-40s%6u %-40s %s\n", log_delta,
+ lm->from_pid, lm->from_name, lm->about_pid, lm->about_name, lm->msg);
+
+ logmsg_remove(lm);
+ }
+
+ fflush(ourlogfile);
}
kern_return_t
}
for (lm_walk = (struct logmsg_s *)inval; (data_left > 0) && (lm_walk->obj_sz <= data_left); lm_walk = ((void *)lm_walk + lm_walk->obj_sz)) {
+ /* malloc() does not return NULL if you ask it for an allocation of size 0.
+ * It will return a valid pointer that can be passed to free(). If we don't
+ * do this check, we'll wind up corrupting our heap in the subsequent
+ * assignments.
+ *
+ * We break out if this check fails because, obj_sz is supposed to include
+ * the size of the logmsg_s struct. If it claims to be of zero size, we
+ * can't safely increment our counter because something obviously got screwed
+ * up along the way, since this should always be at least sizeof(struct logmsg_s).
+ */
+ if (!launchd_assumes(lm_walk->obj_sz > 0)) {
+ runtime_syslog(LOG_WARNING, "Encountered a log message of size 0 with %u bytes left in forwarded data. Ignoring remaining messages.", data_left);
+ break;
+ }
+
+ /* If malloc() keeps failing, we shouldn't put additional pressure on the system
+ * by attempting to add more messages to the log until it returns success
+ * log a failure, hope pressure lets off, and move on.
+ */
if (!launchd_assumes(lm = malloc(lm_walk->obj_sz))) {
- continue;
+ runtime_syslog(LOG_WARNING, "Failed to allocate %llu bytes for log message with %u bytes left in forwarded data. Ignoring remaining messages.", lm_walk->obj_sz, data_left);
+ break;
}
memcpy(lm, lm_walk, lm_walk->obj_sz);
kern_return_t
runtime_log_drain(mach_port_t srp, vm_offset_t *outval, mach_msg_type_number_t *outvalCnt)
{
- if (logmsg_queue_cnt == 0) {
- launchd_assumes(STAILQ_EMPTY(&logmsg_queue));
- launchd_assumes(drain_reply_port == 0);
+ (void)launchd_assumes(drain_reply_port == 0);
+ if ((logmsg_queue_cnt == 0) || shutdown_in_progress || fake_shutdown_in_progress) {
drain_reply_port = srp;
- launchd_assumes(launchd_mport_notify_req(drain_reply_port, MACH_NOTIFY_DEAD_NAME) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_mport_notify_req(drain_reply_port, MACH_NOTIFY_DEAD_NAME) == KERN_SUCCESS);
return MIG_NO_REPLY;
}
void
runtime_add_ref(void)
{
+ if (!pid1_magic) {
+ #if !TARGET_OS_EMBEDDED
+ _vproc_transaction_begin();
+ #endif
+ }
+
runtime_busy_cnt++;
+ runtime_remove_timer();
}
void
runtime_del_ref(void)
{
+ if (!pid1_magic) {
+ #if !TARGET_OS_EMBEDDED
+ if (_vproc_transaction_count() == 0) {
+ runtime_syslog(LOG_INFO, "Exiting cleanly.");
+ }
+
+ runtime_closelog();
+ _vproc_transaction_end();
+ #endif
+ }
+
runtime_busy_cnt--;
+ runtime_install_timer();
+}
+
+void
+runtime_add_weak_ref(void)
+{
+ if (!pid1_magic) {
+ #if !TARGET_OS_EMBEDDED
+ _vproc_standby_begin();
+ #endif
+ }
+ runtime_standby_cnt++;
+}
+
+void
+runtime_del_weak_ref(void)
+{
+ if (!pid1_magic) {
+ #if !TARGET_OS_EMBEDDED
+ _vproc_standby_end();
+ #endif
+ }
+ runtime_standby_cnt--;
+}
+
+void
+runtime_install_timer(void)
+{
+ if (!pid1_magic && runtime_busy_cnt == 0) {
+ (void)launchd_assumes(kevent_mod((uintptr_t)&g_runtime_busy_time, EVFILT_TIMER, EV_ADD, NOTE_SECONDS, 30, root_jobmgr) != -1);
+ }
+}
+
+void
+runtime_remove_timer(void)
+{
+ if (!pid1_magic && runtime_busy_cnt > 0) {
+ (void)launchd_assumes(kevent_mod((uintptr_t)&g_runtime_busy_time, EVFILT_TIMER, EV_DELETE, 0, 0, NULL) != -1);
+ }
}
kern_return_t
-catch_exception_raise(mach_port_t exception_port __attribute__((unused)),
- mach_port_t thread, mach_port_t task,
- exception_type_t exception, exception_data_t code,
- mach_msg_type_number_t codeCnt)
+catch_mach_exception_raise(mach_port_t exception_port __attribute__((unused)), mach_port_t thread, mach_port_t task,
+ exception_type_t exception, mach_exception_data_t code, mach_msg_type_number_t codeCnt)
{
- runtime_syslog(LOG_NOTICE, "%s(): thread: 0x%x task: 0x%x type: 0x%x code: %p codeCnt: 0x%x",
- __func__, thread, task, exception, code, codeCnt);
+ pid_t p4t = -1;
- launchd_assumes(launchd_mport_deallocate(thread) == KERN_SUCCESS);
- launchd_assumes(launchd_mport_deallocate(task) == KERN_SUCCESS);
+ (void)launchd_assumes(pid_for_task(task, &p4t) == 0);
+
+ runtime_syslog(LOG_NOTICE, "%s(): PID: %u thread: 0x%x type: 0x%x code: %p codeCnt: 0x%x",
+ __func__, p4t, thread, exception, code, codeCnt);
+
+ (void)launchd_assumes(launchd_mport_deallocate(thread) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_mport_deallocate(task) == KERN_SUCCESS);
- return 0;
+ return KERN_SUCCESS;
}
kern_return_t
-catch_exception_raise_state(mach_port_t exception_port __attribute__((unused)),
- exception_type_t exception,
- const exception_data_t code, mach_msg_type_number_t codeCnt,
- int *flavor,
- const thread_state_t old_state, mach_msg_type_number_t old_stateCnt,
- thread_state_t new_state, mach_msg_type_number_t *new_stateCnt)
+catch_mach_exception_raise_state(mach_port_t exception_port __attribute__((unused)),
+ exception_type_t exception, const mach_exception_data_t code, mach_msg_type_number_t codeCnt,
+ int *flavor, const thread_state_t old_state, mach_msg_type_number_t old_stateCnt,
+ thread_state_t new_state, mach_msg_type_number_t *new_stateCnt)
{
runtime_syslog(LOG_NOTICE, "%s(): type: 0x%x code: %p codeCnt: 0x%x flavor: %p old_state: %p old_stateCnt: 0x%x new_state: %p new_stateCnt: %p",
__func__, exception, code, codeCnt, flavor, old_state, old_stateCnt, new_state, new_stateCnt);
-
+
memcpy(new_state, old_state, old_stateCnt * sizeof(old_state[0]));
*new_stateCnt = old_stateCnt;
- return 0;
+ return KERN_SUCCESS;
}
kern_return_t
-catch_exception_raise_state_identity(mach_port_t exception_port __attribute__((unused)),
- mach_port_t thread, mach_port_t task,
- exception_type_t exception,
- exception_data_t code, mach_msg_type_number_t codeCnt,
- int *flavor,
- thread_state_t old_state, mach_msg_type_number_t old_stateCnt,
- thread_state_t new_state, mach_msg_type_number_t *new_stateCnt)
+catch_mach_exception_raise_state_identity(mach_port_t exception_port __attribute__((unused)), mach_port_t thread, mach_port_t task,
+ exception_type_t exception, mach_exception_data_t code, mach_msg_type_number_t codeCnt,
+ int *flavor, thread_state_t old_state, mach_msg_type_number_t old_stateCnt,
+ thread_state_t new_state, mach_msg_type_number_t *new_stateCnt)
{
- runtime_syslog(LOG_NOTICE, "%s(): thread: 0x%x task: 0x%x type: 0x%x code: %p codeCnt: 0x%x flavor: %p old_state: %p old_stateCnt: 0x%x new_state: %p new_stateCnt: %p",
- __func__, thread, task, exception, code, codeCnt, flavor, old_state, old_stateCnt, new_state, new_stateCnt);
+ pid_t p4t = -1;
+
+ (void)launchd_assumes(pid_for_task(task, &p4t) == 0);
+ runtime_syslog(LOG_NOTICE, "%s(): PID: %u thread: 0x%x type: 0x%x code: %p codeCnt: 0x%x flavor: %p old_state: %p old_stateCnt: 0x%x new_state: %p new_stateCnt: %p",
+ __func__, p4t, thread, exception, code, codeCnt, flavor, old_state, old_stateCnt, new_state, new_stateCnt);
+
memcpy(new_state, old_state, old_stateCnt * sizeof(old_state[0]));
*new_stateCnt = old_stateCnt;
- launchd_assumes(launchd_mport_deallocate(thread) == KERN_SUCCESS);
- launchd_assumes(launchd_mport_deallocate(task) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_mport_deallocate(thread) == KERN_SUCCESS);
+ (void)launchd_assumes(launchd_mport_deallocate(task) == KERN_SUCCESS);
- return 0;
+ return KERN_SUCCESS;
+}
+
+void
+launchd_log_vm_stats(void)
+{
+ static struct vm_statistics orig_stats;
+ static bool did_first_pass;
+ unsigned int count = HOST_VM_INFO_COUNT;
+ struct vm_statistics stats, *statsp;
+ mach_port_t mhs = mach_host_self();
+
+ statsp = did_first_pass ? &stats : &orig_stats;
+
+ if (!launchd_assumes(host_statistics(mhs, HOST_VM_INFO, (host_info_t)statsp, &count) == KERN_SUCCESS)) {
+ return;
+ }
+
+ (void)launchd_assumes(count == HOST_VM_INFO_COUNT);
+
+ if (did_first_pass) {
+ runtime_syslog(LOG_DEBUG, "VM statistics (now - orig): Free: %d Active: %d Inactive: %d Reactivations: %d PageIns: %d PageOuts: %d Faults: %d COW-Faults: %d Purgeable: %d Purges: %d",
+ stats.free_count - orig_stats.free_count,
+ stats.active_count - orig_stats.active_count,
+ stats.inactive_count - orig_stats.inactive_count,
+ stats.reactivations - orig_stats.reactivations,
+ stats.pageins - orig_stats.pageins,
+ stats.pageouts - orig_stats.pageouts,
+ stats.faults - orig_stats.faults,
+ stats.cow_faults - orig_stats.cow_faults,
+ stats.purgeable_count - orig_stats.purgeable_count,
+ stats.purges - orig_stats.purges);
+ } else {
+ runtime_syslog(LOG_DEBUG, "VM statistics (now): Free: %d Active: %d Inactive: %d Reactivations: %d PageIns: %d PageOuts: %d Faults: %d COW-Faults: %d Purgeable: %d Purges: %d",
+ orig_stats.free_count,
+ orig_stats.active_count,
+ orig_stats.inactive_count,
+ orig_stats.reactivations,
+ orig_stats.pageins,
+ orig_stats.pageouts,
+ orig_stats.faults,
+ orig_stats.cow_faults,
+ orig_stats.purgeable_count,
+ orig_stats.purges);
+
+ did_first_pass = true;
+ }
+
+ launchd_mport_deallocate(mhs);
+}
+
+int64_t
+runtime_get_wall_time(void)
+{
+ struct timeval tv;
+ int64_t r;
+
+ (void)launchd_assumes(gettimeofday(&tv, NULL) != -1);
+
+ r = tv.tv_sec;
+ r *= USEC_PER_SEC;
+ r += tv.tv_usec;
+
+ return r;
+}
+
+uint64_t
+runtime_get_opaque_time(void)
+{
+ return mach_absolute_time();
+}
+
+uint64_t
+runtime_get_opaque_time_of_event(void)
+{
+ return time_of_mach_msg_return;
+}
+
+uint64_t
+runtime_get_nanoseconds_since(uint64_t o)
+{
+ return runtime_opaque_time_to_nano(runtime_get_opaque_time_of_event() - o);
+}
+
+uint64_t
+runtime_opaque_time_to_nano(uint64_t o)
+{
+#if defined(__i386__) || defined(__x86_64__)
+ if (unlikely(tbi.numer != tbi.denom)) {
+#elif defined(__ppc__) || defined(__ppc64__)
+ if (likely(tbi.numer != tbi.denom)) {
+#else
+ if (tbi.numer != tbi.denom) {
+#endif
+#ifdef __LP64__
+ __uint128_t tmp = o;
+ tmp *= tbi.numer;
+ tmp /= tbi.denom;
+ o = tmp;
+#else
+ if (o <= tbi_safe_math_max) {
+ o *= tbi.numer;
+ o /= tbi.denom;
+ } else {
+ double d = o;
+ d *= tbi_float_val;
+ o = d;
+ }
+#endif
+ }
+
+ return o;
+}
+
+void
+do_file_init(void)
+{
+ struct stat sb;
+
+ launchd_assert(mach_timebase_info(&tbi) == 0);
+ tbi_float_val = tbi.numer;
+ tbi_float_val /= tbi.denom;
+ tbi_safe_math_max = UINT64_MAX / tbi.numer;
+
+ if (getpid() == 1) {
+ pid1_magic = true;
+ }
+
+ if (stat("/AppleInternal", &sb) == 0 && stat("/var/db/disableAppleInternal", &sb) == -1) {
+ do_apple_internal_logging = true;
+ }
+
+ if (stat("/var/db/.debug_launchd", &sb) == 0) {
+ internal_mask_pri = LOG_UPTO(LOG_DEBUG);
+ low_level_debug = true;
+ }
+
+ if (stat("/var/db/.launchd_log_per_user_shutdown", &sb) == 0) {
+ g_log_per_user_shutdown = true;
+ }
+
+ if (stat("/var/db/.launchd_use_gmalloc", &sb) == 0) {
+ g_use_gmalloc = true;
+ }
+
+ if (stat("/var/db/.launchd_malloc_log_stacks", &sb) == 0) {
+ g_malloc_log_stacks = true;
+ g_use_gmalloc = false;
+ }
+
+ if (pid1_magic && stat("/var/db/.launchd_log_pid1_shutdown", &sb) == 0) {
+ g_log_pid1_shutdown = true;
+ }
+
+ char bootargs[128];
+ size_t len = sizeof(bootargs) - 1;
+ int r = pid1_magic ? sysctlbyname("kern.bootargs", bootargs, &len, NULL, 0) : -1;
+ if (r == 0) {
+ if (strnstr(bootargs, "-v", len)) {
+ g_verbose_boot = true;
+ }
+ if (strnstr(bootargs, "launchd_trap_sigkill_bugs", len)) {
+ g_trap_sigkill_bugs = true;
+ }
+ }
+
+ if (pid1_magic && g_verbose_boot && stat("/var/db/.launchd_shutdown_debugging", &sb) == 0) {
+ g_shutdown_debugging = true;
+ }
+
+ if (stat("/var/db/.launchd_log_strict_usage", &sb) == 0) {
+ g_log_strict_usage = true;
+ }
}