-/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
-
/***
This file is part of systemd.
#include <sys/timerfd.h>
#include <sys/wait.h>
-#include "sd-id128.h"
#include "sd-daemon.h"
-#include "macro.h"
-#include "prioq.h"
+#include "sd-event.h"
+#include "sd-id128.h"
+
+#include "alloc-util.h"
+#include "fd-util.h"
#include "hashmap.h"
-#include "util.h"
-#include "time-util.h"
+#include "list.h"
+#include "macro.h"
#include "missing.h"
+#include "prioq.h"
+#include "process-util.h"
#include "set.h"
-#include "list.h"
#include "signal-util.h"
-
-#include "sd-event.h"
+#include "string-table.h"
+#include "string-util.h"
+#include "time-util.h"
+#include "util.h"
#define DEFAULT_ACCURACY_USEC (250 * USEC_PER_MSEC)
_SOURCE_EVENT_SOURCE_TYPE_INVALID = -1
} EventSourceType;
+static const char* const event_source_type_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = {
+ [SOURCE_IO] = "io",
+ [SOURCE_TIME_REALTIME] = "realtime",
+ [SOURCE_TIME_BOOTTIME] = "bootime",
+ [SOURCE_TIME_MONOTONIC] = "monotonic",
+ [SOURCE_TIME_REALTIME_ALARM] = "realtime-alarm",
+ [SOURCE_TIME_BOOTTIME_ALARM] = "boottime-alarm",
+ [SOURCE_SIGNAL] = "signal",
+ [SOURCE_CHILD] = "child",
+ [SOURCE_DEFER] = "defer",
+ [SOURCE_POST] = "post",
+ [SOURCE_EXIT] = "exit",
+ [SOURCE_WATCHDOG] = "watchdog",
+};
+
+DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(event_source_type, int);
+
/* All objects we use in epoll events start with this value, so that
* we know how to dispatch it */
typedef enum WakeupType {
bool exit_requested:1;
bool need_process_child:1;
bool watchdog:1;
+ bool profile_delays:1;
int exit_code;
unsigned n_sources;
LIST_HEAD(sd_event_source, sources);
+
+ usec_t last_run, last_log;
+ unsigned delays[sizeof(usec_t) * 8];
};
static void source_disconnect(sd_event_source *s);
if (x->pending_iteration > y->pending_iteration)
return 1;
- /* Stability for the rest */
- if (x < y)
- return -1;
- if (x > y)
- return 1;
-
return 0;
}
assert(x->prepare);
assert(y->prepare);
+ /* Enabled ones first */
+ if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF)
+ return -1;
+ if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF)
+ return 1;
+
/* Move most recently prepared ones last, so that we can stop
* preparing as soon as we hit one that has already been
* prepared in the current iteration */
if (x->prepare_iteration > y->prepare_iteration)
return 1;
- /* Enabled ones first */
- if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF)
- return -1;
- if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF)
- return 1;
-
/* Lower priority values first */
if (x->priority < y->priority)
return -1;
if (x->priority > y->priority)
return 1;
- /* Stability for the rest */
- if (x < y)
- return -1;
- if (x > y)
- return 1;
-
return 0;
}
if (x->time.next > y->time.next)
return 1;
- /* Stability for the rest */
- if (x < y)
- return -1;
- if (x > y)
- return 1;
-
return 0;
}
+static usec_t time_event_source_latest(const sd_event_source *s) {
+ return usec_add(s->time.next, s->time.accuracy);
+}
+
static int latest_time_prioq_compare(const void *a, const void *b) {
const sd_event_source *x = a, *y = b;
return 1;
/* Order by time */
- if (x->time.next + x->time.accuracy < y->time.next + y->time.accuracy)
- return -1;
- if (x->time.next + x->time.accuracy > y->time.next + y->time.accuracy)
- return 1;
-
- /* Stability for the rest */
- if (x < y)
+ if (time_event_source_latest(x) < time_event_source_latest(y))
return -1;
- if (x > y)
+ if (time_event_source_latest(x) > time_event_source_latest(y))
return 1;
return 0;
if (x->priority > y->priority)
return 1;
- /* Stability for the rest */
- if (x < y)
- return -1;
- if (x > y)
- return 1;
-
return 0;
}
e->original_pid = getpid();
e->perturb = USEC_INFINITY;
- e->pending = prioq_new(pending_prioq_compare);
- if (!e->pending) {
- r = -ENOMEM;
+ r = prioq_ensure_allocated(&e->pending, pending_prioq_compare);
+ if (r < 0)
goto fail;
- }
e->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
if (e->epoll_fd < 0) {
goto fail;
}
+ if (secure_getenv("SD_EVENT_PROFILE_DELAYS")) {
+ log_debug("Event loop profiling enabled. Logarithmic histogram of event loop iterations in the range 2^0 ... 2^63 us will be logged every 5s.");
+ e->profile_delays = true;
+ }
+
*ret = e;
return 0;
}
_public_ sd_event* sd_event_ref(sd_event *e) {
- assert_return(e, NULL);
+
+ if (!e)
+ return NULL;
assert(e->n_ref >= 1);
e->n_ref++;
r = epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->io.fd, NULL);
if (r < 0)
- log_debug_errno(errno, "Failed to remove source %s from epoll: %m", strna(s->description));
+ log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll: %m",
+ strna(s->description), event_source_type_to_string(s->type));
s->io.registered = false;
}
return 0;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
static clockid_t event_source_type_to_clock(EventSourceType t) {
switch (t) {
d->priority = priority;
r = hashmap_put(e->signal_data, &d->priority, d);
- if (r < 0)
+ if (r < 0) {
+ free(d);
return r;
+ }
added = true;
}
s->event->n_enabled_child_sources--;
}
- (void) hashmap_remove(s->event->child_sources, INT_TO_PTR(s->child.pid));
+ (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid));
event_gc_signal_data(s->event, &s->priority, SIGCHLD);
}
int r;
assert_return(e, -EINVAL);
- assert_return(usec != (uint64_t) -1, -EINVAL);
assert_return(accuracy != (uint64_t) -1, -EINVAL);
assert_return(e->state != SD_EVENT_FINISHED, -ESTALE);
assert_return(!event_pid_changed(e), -ECHILD);
+ if (IN_SET(clock, CLOCK_BOOTTIME, CLOCK_BOOTTIME_ALARM) &&
+ !clock_boottime_supported())
+ return -EOPNOTSUPP;
+
if (!callback)
callback = time_exit_callback;
d = event_get_clock_data(e, type);
assert(d);
- if (!d->earliest) {
- d->earliest = prioq_new(earliest_time_prioq_compare);
- if (!d->earliest)
- return -ENOMEM;
- }
+ r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare);
+ if (r < 0)
+ return r;
- if (!d->latest) {
- d->latest = prioq_new(latest_time_prioq_compare);
- if (!d->latest)
- return -ENOMEM;
- }
+ r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare);
+ if (r < 0)
+ return r;
if (d->fd < 0) {
r = event_setup_timer_fd(e, d, clock);
int r;
assert_return(e, -EINVAL);
- assert_return(sig > 0, -EINVAL);
- assert_return(sig < _NSIG, -EINVAL);
+ assert_return(SIGNAL_VALID(sig), -EINVAL);
assert_return(e->state != SD_EVENT_FINISHED, -ESTALE);
assert_return(!event_pid_changed(e), -ECHILD);
callback = signal_exit_callback;
r = pthread_sigmask(SIG_SETMASK, NULL, &ss);
- if (r < 0)
- return -errno;
+ if (r != 0)
+ return -r;
if (!sigismember(&ss, sig))
return -EBUSY;
return 0;
}
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_add_child(
sd_event *e,
sd_event_source **ret,
if (r < 0)
return r;
- if (hashmap_contains(e->child_sources, INT_TO_PTR(pid)))
+ if (hashmap_contains(e->child_sources, PID_TO_PTR(pid)))
return -EBUSY;
s = source_new(e, !ret, SOURCE_CHILD);
s->userdata = userdata;
s->enabled = SD_EVENT_ONESHOT;
- r = hashmap_put(e->child_sources, INT_TO_PTR(pid), s);
+ r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s);
if (r < 0) {
source_free(s);
return r;
return 0;
}
+#endif // 0
_public_ int sd_event_add_post(
sd_event *e,
assert_return(e->state != SD_EVENT_FINISHED, -ESTALE);
assert_return(!event_pid_changed(e), -ECHILD);
- if (!e->exit) {
- e->exit = prioq_new(exit_prioq_compare);
- if (!e->exit)
- return -ENOMEM;
- }
+ r = prioq_ensure_allocated(&e->exit, exit_prioq_compare);
+ if (r < 0)
+ return r;
s = source_new(e, !ret, SOURCE_EXIT);
if (!s)
return 0;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ sd_event_source* sd_event_source_ref(sd_event_source *s) {
- assert_return(s, NULL);
+
+ if (!s)
+ return NULL;
assert(s->n_ref >= 1);
s->n_ref++;
return free_and_strdup(&s->description, description);
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_source_get_description(sd_event_source *s, const char **description) {
assert_return(s, -EINVAL);
assert_return(description, -EINVAL);
return s->event;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_source_get_pending(sd_event_source *s) {
assert_return(s, -EINVAL);
assert_return(s->type != SOURCE_EXIT, -EDOM);
return 0;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) {
assert_return(s, -EINVAL);
assert_return(events, -EINVAL);
return 0;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) {
assert_return(s, -EINVAL);
assert_return(revents, -EINVAL);
return 0;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_source_get_enabled(sd_event_source *s, int *m) {
assert_return(s, -EINVAL);
assert_return(m, -EINVAL);
struct clock_data *d;
assert_return(s, -EINVAL);
- assert_return(usec != (uint64_t) -1, -EINVAL);
assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM);
assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE);
assert_return(!event_pid_changed(s->event), -ECHILD);
return 0;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) {
assert_return(s, -EINVAL);
assert_return(usec, -EINVAL);
return 0;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ void* sd_event_source_get_userdata(sd_event_source *s) {
assert_return(s, NULL);
if (a <= 0)
return 0;
+ if (a >= USEC_INFINITY)
+ return USEC_INFINITY;
if (b <= a + 1)
return a;
d->needs_rearm = false;
a = prioq_peek(d->earliest);
- if (!a || a->enabled == SD_EVENT_OFF) {
+ if (!a || a->enabled == SD_EVENT_OFF || a->time.next == USEC_INFINITY) {
if (d->fd < 0)
return 0;
b = prioq_peek(d->latest);
assert_se(b && b->enabled != SD_EVENT_OFF);
- t = sleep_between(e, a->time.next, b->time.next + b->time.accuracy);
+ t = sleep_between(e, a->time.next, time_event_source_latest(b));
if (d->next == t)
return 0;
if (_unlikely_(n != sizeof(si)))
return -EIO;
- assert(si.ssi_signo < _NSIG);
+ assert(SIGNAL_VALID(si.ssi_signo));
read_one = true;
s->dispatching = false;
- if (r < 0) {
- if (s->description)
- log_debug_errno(r, "Event source '%s' returned error, disabling: %m", s->description);
- else
- log_debug_errno(r, "Event source %p returned error, disabling: %m", s);
- }
+ if (r < 0)
+ log_debug_errno(r, "Event source %s (type %s) returned error, disabling: %m",
+ strna(s->description), event_source_type_to_string(s->type));
if (s->n_ref == 0)
source_free(s);
r = s->prepare(s, s->userdata);
s->dispatching = false;
- if (r < 0) {
- if (s->description)
- log_debug_errno(r, "Prepare callback of event source '%s' returned error, disabling: %m", s->description);
- else
- log_debug_errno(r, "Prepare callback of event source %p returned error, disabling: %m", s);
- }
+ if (r < 0)
+ log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, disabling: %m",
+ strna(s->description), event_source_type_to_string(s->type));
if (s->n_ref == 0)
source_free(s);
e->iteration++;
+ e->state = SD_EVENT_PREPARING;
r = event_prepare(e);
+ e->state = SD_EVENT_INITIAL;
if (r < 0)
return r;
}
dual_timestamp_get(&e->timestamp);
- e->timestamp_boottime = now(CLOCK_BOOTTIME);
+ if (clock_boottime_supported())
+ e->timestamp_boottime = now(CLOCK_BOOTTIME);
for (i = 0; i < m; i++) {
return 1;
}
+static void event_log_delays(sd_event *e) {
+ char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1];
+ unsigned i;
+ int o;
+
+ for (i = o = 0; i < ELEMENTSOF(e->delays); i++) {
+ o += snprintf(&b[o], sizeof(b) - o, "%u ", e->delays[i]);
+ e->delays[i] = 0;
+ }
+ log_debug("Event loop iterations: %.*s", o, b);
+}
+
_public_ int sd_event_run(sd_event *e, uint64_t timeout) {
int r;
assert_return(e->state != SD_EVENT_FINISHED, -ESTALE);
assert_return(e->state == SD_EVENT_INITIAL, -EBUSY);
+ if (e->profile_delays && e->last_run) {
+ usec_t this_run;
+ unsigned l;
+
+ this_run = now(CLOCK_MONOTONIC);
+
+ l = u64log2(this_run - e->last_run);
+ assert(l < sizeof(e->delays));
+ e->delays[l]++;
+
+ if (this_run - e->last_log >= 5*USEC_PER_SEC) {
+ event_log_delays(e);
+ e->last_log = this_run;
+ }
+ }
+
r = sd_event_prepare(e);
if (r == 0)
/* There was nothing? Then wait... */
r = sd_event_wait(e, timeout);
+ if (e->profile_delays)
+ e->last_run = now(CLOCK_MONOTONIC);
+
if (r > 0) {
/* There's something now, then let's dispatch it */
r = sd_event_dispatch(e);
return r;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_loop(sd_event *e) {
int r;
return e->state;
}
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_get_exit_code(sd_event *e, int *code) {
assert_return(e, -EINVAL);
assert_return(code, -EINVAL);
*code = e->exit_code;
return 0;
}
+#endif // 0
_public_ int sd_event_exit(sd_event *e, int code) {
assert_return(e, -EINVAL);
return 0;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) {
assert_return(e, -EINVAL);
assert_return(usec, -EINVAL);
assert_return(!event_pid_changed(e), -ECHILD);
+ assert_return(IN_SET(clock,
+ CLOCK_REALTIME,
+ CLOCK_REALTIME_ALARM,
+ CLOCK_MONOTONIC,
+ CLOCK_BOOTTIME,
+ CLOCK_BOOTTIME_ALARM), -EOPNOTSUPP);
+
+ if (IN_SET(clock, CLOCK_BOOTTIME, CLOCK_BOOTTIME_ALARM) && !clock_boottime_supported())
+ return -EOPNOTSUPP;
if (!dual_timestamp_is_set(&e->timestamp)) {
/* Implicitly fall back to now() if we never ran
case CLOCK_BOOTTIME_ALARM:
*usec = e->timestamp_boottime;
break;
+
+ default:
+ assert_not_reached("Unknown clock?");
}
return 0;
return 1;
}
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_get_tid(sd_event *e, pid_t *tid) {
assert_return(e, -EINVAL);
assert_return(tid, -EINVAL);
return -ENXIO;
}
+#endif // 0
_public_ int sd_event_set_watchdog(sd_event *e, int b) {
int r;
return r;
}
-/// UNNEEDED by elogind
-#if 0
+#if 0 /// UNNEEDED by elogind
_public_ int sd_event_get_watchdog(sd_event *e) {
assert_return(e, -EINVAL);
assert_return(!event_pid_changed(e), -ECHILD);