chiark / gitweb /
event: allow EPOLLET as event flag
[elogind.git] / src / libsystemd-bus / sd-event.c
index 9c641c9a8ec7e10c08608a0f8e2f74c68d6e9148..b5ddf71ebede89439e76adf1a3ba33fab830be05 100644 (file)
 #include <sys/timerfd.h>
 #include <sys/wait.h>
 
+#include "sd-id128.h"
 #include "macro.h"
 #include "prioq.h"
 #include "hashmap.h"
 #include "util.h"
 #include "time-util.h"
-#include "sd-id128.h"
+#include "missing.h"
 
 #include "sd-event.h"
 
@@ -50,7 +51,7 @@ struct sd_event_source {
 
         sd_event *event;
         void *userdata;
-        sd_prepare_handler_t prepare;
+        sd_event_handler_t prepare;
 
         EventSourceType type:4;
         int enabled:3;
@@ -64,34 +65,34 @@ struct sd_event_source {
 
         union {
                 struct {
-                        sd_io_handler_t callback;
+                        sd_event_io_handler_t callback;
                         int fd;
                         uint32_t events;
                         uint32_t revents;
                         bool registered:1;
                 } io;
                 struct {
-                        sd_time_handler_t callback;
+                        sd_event_time_handler_t callback;
                         usec_t next, accuracy;
                         unsigned earliest_index;
                         unsigned latest_index;
                 } time;
                 struct {
-                        sd_signal_handler_t callback;
+                        sd_event_signal_handler_t callback;
                         struct signalfd_siginfo siginfo;
                         int sig;
                 } signal;
                 struct {
-                        sd_child_handler_t callback;
+                        sd_event_child_handler_t callback;
                         siginfo_t siginfo;
                         pid_t pid;
                         int options;
                 } child;
                 struct {
-                        sd_defer_handler_t callback;
+                        sd_event_handler_t callback;
                 } defer;
                 struct {
-                        sd_quit_handler_t callback;
+                        sd_event_handler_t callback;
                         unsigned prioq_index;
                 } quit;
         };
@@ -138,6 +139,9 @@ struct sd_event {
 
         bool quit_requested:1;
         bool need_process_child:1;
+
+        pid_t tid;
+        sd_event **default_event_ptr;
 };
 
 static int pending_prioq_compare(const void *a, const void *b) {
@@ -230,7 +234,7 @@ static int earliest_time_prioq_compare(const void *a, const void *b) {
         if (x->time.next < y->time.next)
                 return -1;
         if (x->time.next > y->time.next)
-                return -1;
+                return 1;
 
         /* Stability for the rest */
         if (x < y)
@@ -263,7 +267,7 @@ static int latest_time_prioq_compare(const void *a, const void *b) {
         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;
+                return 1;
 
         /* Stability for the rest */
         if (x < y)
@@ -304,6 +308,9 @@ static int quit_prioq_compare(const void *a, const void *b) {
 static void event_free(sd_event *e) {
         assert(e);
 
+        if (e->default_event_ptr)
+                *(e->default_event_ptr) = NULL;
+
         if (e->epoll_fd >= 0)
                 close_nointr_nofail(e->epoll_fd);
 
@@ -536,6 +543,14 @@ static int source_set_pending(sd_event_source *s, bool b) {
         } else
                 assert_se(prioq_remove(s->event->pending, s, &s->pending_index));
 
+        if (s->type == SOURCE_REALTIME) {
+                prioq_reshuffle(s->event->realtime_earliest, s, &s->time.earliest_index);
+                prioq_reshuffle(s->event->realtime_latest, s, &s->time.latest_index);
+        } else if (s->type == SOURCE_MONOTONIC) {
+                prioq_reshuffle(s->event->monotonic_earliest, s, &s->time.earliest_index);
+                prioq_reshuffle(s->event->monotonic_latest, s, &s->time.latest_index);
+        }
+
         return 0;
 }
 
@@ -560,7 +575,7 @@ _public_ int sd_event_add_io(
                 sd_event *e,
                 int fd,
                 uint32_t events,
-                sd_io_handler_t callback,
+                sd_event_io_handler_t callback,
                 void *userdata,
                 sd_event_source **ret) {
 
@@ -569,7 +584,7 @@ _public_ int sd_event_add_io(
 
         assert_return(e, -EINVAL);
         assert_return(fd >= 0, -EINVAL);
-        assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP)), -EINVAL);
+        assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL);
         assert_return(callback, -EINVAL);
         assert_return(ret, -EINVAL);
         assert_return(e->state != SD_EVENT_FINISHED, -ESTALE);
@@ -625,15 +640,16 @@ static int event_setup_timer_fd(
         }
 
         /* When we sleep for longer, we try to realign the wakeup to
-           the same time wihtin each second, so that events all across
-           the system can be coalesced into a single CPU
-           wakeup. However, let's take some system-specific randomness
-           for this value, so that in a network of systems with synced
-           clocks timer events are distributed a bit. Here, we
-           calculate a perturbation usec offset from the boot ID. */
+           the same time wihtin each minute/second/250ms, so that
+           events all across the system can be coalesced into a single
+           CPU wakeup. However, let's take some system-specific
+           randomness for this value, so that in a network of systems
+           with synced clocks timer events are distributed a
+           bit. Here, we calculate a perturbation usec offset from the
+           boot ID. */
 
         if (sd_id128_get_boot(&bootid) >= 0)
-                e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_SEC;
+                e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_MINUTE;
 
         *timer_fd = fd;
         return 0;
@@ -648,7 +664,7 @@ static int event_add_time_internal(
                 Prioq **latest,
                 uint64_t usec,
                 uint64_t accuracy,
-                sd_time_handler_t callback,
+                sd_event_time_handler_t callback,
                 void *userdata,
                 sd_event_source **ret) {
 
@@ -715,7 +731,7 @@ fail:
 _public_ int sd_event_add_monotonic(sd_event *e,
                                     uint64_t usec,
                                     uint64_t accuracy,
-                                    sd_time_handler_t callback,
+                                    sd_event_time_handler_t callback,
                                     void *userdata,
                                     sd_event_source **ret) {
 
@@ -725,7 +741,7 @@ _public_ int sd_event_add_monotonic(sd_event *e,
 _public_ int sd_event_add_realtime(sd_event *e,
                                    uint64_t usec,
                                    uint64_t accuracy,
-                                   sd_time_handler_t callback,
+                                   sd_event_time_handler_t callback,
                                    void *userdata,
                                    sd_event_source **ret) {
 
@@ -767,7 +783,7 @@ static int event_update_signal_fd(sd_event *e) {
 _public_ int sd_event_add_signal(
                 sd_event *e,
                 int sig,
-                sd_signal_handler_t callback,
+                sd_event_signal_handler_t callback,
                 void *userdata,
                 sd_event_source **ret) {
 
@@ -817,7 +833,7 @@ _public_ int sd_event_add_child(
                 sd_event *e,
                 pid_t pid,
                 int options,
-                sd_child_handler_t callback,
+                sd_event_child_handler_t callback,
                 void *userdata,
                 sd_event_source **ret) {
 
@@ -876,7 +892,7 @@ _public_ int sd_event_add_child(
 
 _public_ int sd_event_add_defer(
                 sd_event *e,
-                sd_defer_handler_t callback,
+                sd_event_handler_t callback,
                 void *userdata,
                 sd_event_source **ret) {
 
@@ -909,7 +925,7 @@ _public_ int sd_event_add_defer(
 
 _public_ int sd_event_add_quit(
                 sd_event *e,
-                sd_quit_handler_t callback,
+                sd_event_handler_t callback,
                 void *userdata,
                 sd_event_source **ret) {
 
@@ -968,7 +984,7 @@ _public_ sd_event_source* sd_event_source_unref(sd_event_source *s) {
         return NULL;
 }
 
-_public_ sd_event *sd_event_get(sd_event_source *s) {
+_public_ sd_event *sd_event_source_get_event(sd_event_source *s) {
         assert_return(s, NULL);
 
         return s->event;
@@ -1006,7 +1022,7 @@ _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events)
 
         assert_return(s, -EINVAL);
         assert_return(s->type == SOURCE_IO, -EDOM);
-        assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP)), -EINVAL);
+        assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL);
         assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE);
         assert_return(!event_pid_changed(s->event), -ECHILD);
 
@@ -1020,6 +1036,7 @@ _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events)
         }
 
         s->io.events = events;
+        source_set_pending(s, false);
 
         return 0;
 }
@@ -1230,11 +1247,10 @@ _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) {
         assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE);
         assert_return(!event_pid_changed(s->event), -ECHILD);
 
-        if (s->time.next == usec)
-                return 0;
-
         s->time.next = usec;
 
+        source_set_pending(s, false);
+
         if (s->type == SOURCE_REALTIME) {
                 prioq_reshuffle(s->event->realtime_earliest, s, &s->time.earliest_index);
                 prioq_reshuffle(s->event->realtime_latest, s, &s->time.latest_index);
@@ -1266,11 +1282,10 @@ _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec
         if (usec == 0)
                 usec = DEFAULT_ACCURACY_USEC;
 
-        if (s->time.accuracy == usec)
-                return 0;
-
         s->time.accuracy = usec;
 
+        source_set_pending(s, false);
+
         if (s->type == SOURCE_REALTIME)
                 prioq_reshuffle(s->event->realtime_latest, s, &s->time.latest_index);
         else
@@ -1289,7 +1304,7 @@ _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) {
         return 0;
 }
 
-_public_ int sd_event_source_set_prepare(sd_event_source *s, sd_prepare_handler_t callback) {
+_public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) {
         int r;
 
         assert_return(s, -EINVAL);
@@ -1349,13 +1364,24 @@ static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) {
              dispatch as much as possible on the entire system.
 
           We implement this by waking up everywhere at the same time
-          within any given second if we can, synchronised via the
+          within any given minute if we can, synchronised via the
           perturbation value determined from the boot ID. If we can't,
-          then we try to find the same spot in every a 250ms
+          then we try to find the same spot in every 1s and then 250ms
           step. Otherwise, we pick the last possible time to wake up.
         */
 
-        c = (b / USEC_PER_SEC) * USEC_PER_SEC + e->perturb;
+        c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb;
+        if (c >= b) {
+                if (_unlikely_(c < USEC_PER_MINUTE))
+                        return b;
+
+                c -= USEC_PER_MINUTE;
+        }
+
+        if (c >= a)
+                return c;
+
+        c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC);
         if (c >= b) {
                 if (_unlikely_(c < USEC_PER_SEC))
                         return b;
@@ -1398,6 +1424,9 @@ static int event_arm_timer(
         a = prioq_peek(earliest);
         if (!a || a->enabled == SD_EVENT_OFF) {
 
+                if (timer_fd < 0)
+                        return 0;
+
                 if (*next == (usec_t) -1)
                         return 0;
 
@@ -1614,6 +1643,8 @@ static int source_dispatch(sd_event_source *s) {
                         return r;
         }
 
+        sd_event_source_ref(s);
+
         switch (s->type) {
 
         case SOURCE_IO:
@@ -1645,6 +1676,8 @@ static int source_dispatch(sd_event_source *s) {
                 break;
         }
 
+        sd_event_source_unref(s);
+
         return r;
 }
 
@@ -1735,18 +1768,16 @@ _public_ int sd_event_run(sd_event *e, uint64_t timeout) {
         if (r < 0)
                 goto finish;
 
-        if (event_next_pending(e) || e->need_process_child)
-                timeout = 0;
+        r = event_arm_timer(e, e->monotonic_fd, e->monotonic_earliest, e->monotonic_latest, &e->monotonic_next);
+        if (r < 0)
+                goto finish;
 
-        if (timeout > 0) {
-                r = event_arm_timer(e, e->monotonic_fd, e->monotonic_earliest, e->monotonic_latest, &e->monotonic_next);
-                if (r < 0)
-                        goto finish;
+        r = event_arm_timer(e, e->realtime_fd, e->realtime_earliest, e->realtime_latest, &e->realtime_next);
+        if (r < 0)
+                goto finish;
 
-                r = event_arm_timer(e, e->realtime_fd, e->realtime_earliest, e->realtime_latest, &e->realtime_next);
-                if (r < 0)
-                        goto finish;
-        }
+        if (event_next_pending(e) || e->need_process_child)
+                timeout = 0;
 
         m = epoll_wait(e->epoll_fd, ev_queue, EPOLL_QUEUE_MAX,
                        timeout == (uint64_t) -1 ? -1 : (int) ((timeout + USEC_PER_MSEC - 1) / USEC_PER_MSEC));
@@ -1865,3 +1896,42 @@ _public_ int sd_event_get_now_monotonic(sd_event *e, uint64_t *usec) {
         *usec = e->timestamp.monotonic;
         return 0;
 }
+
+_public_ int sd_event_default(sd_event **ret) {
+
+        static __thread sd_event *default_event = NULL;
+        sd_event *e;
+        int r;
+
+        if (!ret)
+                return !!default_event;
+
+        if (default_event) {
+                *ret = sd_event_ref(default_event);
+                return 0;
+        }
+
+        r = sd_event_new(&e);
+        if (r < 0)
+                return r;
+
+        e->default_event_ptr = &default_event;
+        e->tid = gettid();
+        default_event = e;
+
+        *ret = e;
+        return 1;
+}
+
+_public_ int sd_event_get_tid(sd_event *e, pid_t *tid) {
+        assert_return(e, -EINVAL);
+        assert_return(tid, -EINVAL);
+        assert_return(!event_pid_changed(e), -ECHILD);
+
+        if (e->tid != 0) {
+                *tid = e->tid;
+                return 0;
+        }
+
+        return -ENXIO;
+}