1
0
mirror of https://github.com/samba-team/samba.git synced 2024-12-24 21:34:56 +03:00
samba-mirror/lib/tevent/tevent_epoll.c

634 lines
16 KiB
C
Raw Normal View History

/*
Unix SMB/CIFS implementation.
main select loop and event handling - epoll implementation
Copyright (C) Andrew Tridgell 2003-2005
2009-02-16 10:52:06 +03:00
Copyright (C) Stefan Metzmacher 2005-2009
** NOTE! The following LGPL license applies to the tevent
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
2009-02-16 10:52:06 +03:00
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "replace.h"
#include "system/filesys.h"
#include "system/select.h"
#include "tevent.h"
#include "tevent_internal.h"
#include "tevent_util.h"
struct epoll_event_context {
/* a pointer back to the generic event_context */
struct tevent_context *ev;
/* when using epoll this is the handle from epoll_create */
int epoll_fd;
pid_t pid;
bool panic_force_replay;
bool *panic_state;
bool (*panic_fallback)(struct tevent_context *ev, bool replay);
};
#ifdef TEST_PANIC_FALLBACK
static int epoll_create_panic_fallback(struct epoll_event_context *epoll_ev,
int size)
{
if (epoll_ev->panic_fallback == NULL) {
return epoll_create(size);
}
/* 50% of the time, fail... */
if ((random() % 2) == 0) {
errno = EINVAL;
return -1;
}
return epoll_create(size);
}
static int epoll_ctl_panic_fallback(struct epoll_event_context *epoll_ev,
int epfd, int op, int fd,
struct epoll_event *event)
{
if (epoll_ev->panic_fallback == NULL) {
return epoll_ctl(epfd, op, fd, event);
}
/* 50% of the time, fail... */
if ((random() % 2) == 0) {
errno = EINVAL;
return -1;
}
return epoll_ctl(epfd, op, fd, event);
}
static int epoll_wait_panic_fallback(struct epoll_event_context *epoll_ev,
int epfd,
struct epoll_event *events,
int maxevents,
int timeout)
{
if (epoll_ev->panic_fallback == NULL) {
return epoll_wait(epfd, events, maxevents, timeout);
}
/* 50% of the time, fail... */
if ((random() % 2) == 0) {
errno = EINVAL;
return -1;
}
return epoll_wait(epfd, events, maxevents, timeout);
}
#define epoll_create(_size) \
epoll_create_panic_fallback(epoll_ev, _size)
#define epoll_ctl(_epfd, _op, _fd, _event) \
epoll_ctl_panic_fallback(epoll_ev,_epfd, _op, _fd, _event)
#define epoll_wait(_epfd, _events, _maxevents, _timeout) \
epoll_wait_panic_fallback(epoll_ev, _epfd, _events, _maxevents, _timeout)
#endif
/*
called to set the panic fallback function.
*/
_PRIVATE_ bool tevent_epoll_set_panic_fallback(struct tevent_context *ev,
bool (*panic_fallback)(struct tevent_context *ev,
bool replay))
{
struct epoll_event_context *epoll_ev;
if (ev->additional_data == NULL) {
return false;
}
epoll_ev = talloc_get_type(ev->additional_data,
struct epoll_event_context);
if (epoll_ev == NULL) {
return false;
}
epoll_ev->panic_fallback = panic_fallback;
return true;
}
/*
called when a epoll call fails
*/
static void epoll_panic(struct epoll_event_context *epoll_ev,
const char *reason, bool replay)
{
struct tevent_context *ev = epoll_ev->ev;
bool (*panic_fallback)(struct tevent_context *ev, bool replay);
panic_fallback = epoll_ev->panic_fallback;
if (epoll_ev->panic_state != NULL) {
*epoll_ev->panic_state = true;
}
if (epoll_ev->panic_force_replay) {
replay = true;
}
TALLOC_FREE(ev->additional_data);
if (panic_fallback == NULL) {
tevent_debug(ev, TEVENT_DEBUG_FATAL,
"%s (%s) replay[%u] - calling abort()\n",
reason, strerror(errno), (unsigned)replay);
abort();
}
tevent_debug(ev, TEVENT_DEBUG_WARNING,
"%s (%s) replay[%u] - calling panic_fallback\n",
reason, strerror(errno), (unsigned)replay);
if (!panic_fallback(ev, replay)) {
/* Fallback failed. */
tevent_debug(ev, TEVENT_DEBUG_FATAL,
"%s (%s) replay[%u] - calling abort()\n",
reason, strerror(errno), (unsigned)replay);
abort();
}
}
/*
map from TEVENT_FD_* to EPOLLIN/EPOLLOUT
*/
static uint32_t epoll_map_flags(uint16_t flags)
{
uint32_t ret = 0;
if (flags & TEVENT_FD_READ) ret |= (EPOLLIN | EPOLLERR | EPOLLHUP);
if (flags & TEVENT_FD_WRITE) ret |= (EPOLLOUT | EPOLLERR | EPOLLHUP);
return ret;
}
/*
free the epoll fd
*/
static int epoll_ctx_destructor(struct epoll_event_context *epoll_ev)
{
close(epoll_ev->epoll_fd);
epoll_ev->epoll_fd = -1;
return 0;
}
/*
init the epoll fd
*/
static int epoll_init_ctx(struct epoll_event_context *epoll_ev)
{
epoll_ev->epoll_fd = epoll_create(64);
if (epoll_ev->epoll_fd == -1) {
tevent_debug(epoll_ev->ev, TEVENT_DEBUG_FATAL,
"Failed to create epoll handle.\n");
return -1;
}
if (!ev_set_close_on_exec(epoll_ev->epoll_fd)) {
tevent_debug(epoll_ev->ev, TEVENT_DEBUG_WARNING,
"Failed to set close-on-exec, file descriptor may be leaked to children.\n");
}
epoll_ev->pid = getpid();
talloc_set_destructor(epoll_ev, epoll_ctx_destructor);
return 0;
}
static void epoll_add_event(struct epoll_event_context *epoll_ev, struct tevent_fd *fde);
/*
reopen the epoll handle when our pid changes
see http://junkcode.samba.org/ftp/unpacked/junkcode/epoll_fork.c for an
demonstration of why this is needed
*/
static void epoll_check_reopen(struct epoll_event_context *epoll_ev)
{
struct tevent_fd *fde;
bool *caller_panic_state = epoll_ev->panic_state;
bool panic_triggered = false;
if (epoll_ev->pid == getpid()) {
return;
}
close(epoll_ev->epoll_fd);
epoll_ev->epoll_fd = epoll_create(64);
if (epoll_ev->epoll_fd == -1) {
epoll_panic(epoll_ev, "epoll_create() failed", false);
return;
}
if (!ev_set_close_on_exec(epoll_ev->epoll_fd)) {
tevent_debug(epoll_ev->ev, TEVENT_DEBUG_WARNING,
"Failed to set close-on-exec, file descriptor may be leaked to children.\n");
}
epoll_ev->pid = getpid();
epoll_ev->panic_state = &panic_triggered;
for (fde=epoll_ev->ev->fd_events;fde;fde=fde->next) {
epoll_add_event(epoll_ev, fde);
if (panic_triggered) {
if (caller_panic_state != NULL) {
*caller_panic_state = true;
}
return;
}
}
epoll_ev->panic_state = NULL;
}
#define EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT (1<<0)
#define EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR (1<<1)
#define EPOLL_ADDITIONAL_FD_FLAG_GOT_ERROR (1<<2)
/*
add the epoll event to the given fd_event
*/
static void epoll_add_event(struct epoll_event_context *epoll_ev, struct tevent_fd *fde)
{
struct epoll_event event;
int ret;
fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
/* if we don't want events yet, don't add an epoll_event */
if (fde->flags == 0) return;
ZERO_STRUCT(event);
event.events = epoll_map_flags(fde->flags);
event.data.ptr = fde;
ret = epoll_ctl(epoll_ev->epoll_fd, EPOLL_CTL_ADD, fde->fd, &event);
if (ret != 0) {
epoll_panic(epoll_ev, "EPOLL_CTL_ADD failed", false);
return;
}
fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT;
/* only if we want to read we want to tell the event handler about errors */
if (fde->flags & TEVENT_FD_READ) {
fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
}
}
/*
delete the epoll event for given fd_event
*/
static void epoll_del_event(struct epoll_event_context *epoll_ev, struct tevent_fd *fde)
{
struct epoll_event event;
int ret;
fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
/* if there's no epoll_event, we don't need to delete it */
if (!(fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT)) return;
ZERO_STRUCT(event);
event.events = epoll_map_flags(fde->flags);
event.data.ptr = fde;
ret = epoll_ctl(epoll_ev->epoll_fd, EPOLL_CTL_DEL, fde->fd, &event);
if (ret != 0) {
tevent_debug(epoll_ev->ev, TEVENT_DEBUG_FATAL,
"epoll_del_event failed! probable early close bug (%s)\n",
strerror(errno));
}
fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT;
}
/*
change the epoll event to the given fd_event
*/
static void epoll_mod_event(struct epoll_event_context *epoll_ev, struct tevent_fd *fde)
{
struct epoll_event event;
int ret;
fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
ZERO_STRUCT(event);
event.events = epoll_map_flags(fde->flags);
event.data.ptr = fde;
ret = epoll_ctl(epoll_ev->epoll_fd, EPOLL_CTL_MOD, fde->fd, &event);
if (ret != 0) {
epoll_panic(epoll_ev, "EPOLL_CTL_MOD failed", false);
return;
}
/* only if we want to read we want to tell the event handler about errors */
if (fde->flags & TEVENT_FD_READ) {
fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
}
}
static void epoll_change_event(struct epoll_event_context *epoll_ev, struct tevent_fd *fde)
{
bool got_error = (fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_GOT_ERROR);
bool want_read = (fde->flags & TEVENT_FD_READ);
bool want_write= (fde->flags & TEVENT_FD_WRITE);
fde->additional_flags &= ~EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR;
/* there's already an event */
if (fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_HAS_EVENT) {
if (want_read || (want_write && !got_error)) {
epoll_mod_event(epoll_ev, fde);
return;
}
/*
* if we want to match the select behavior, we need to remove the epoll_event
* when the caller isn't interested in events.
*
* this is because epoll reports EPOLLERR and EPOLLHUP, even without asking for them
*/
epoll_del_event(epoll_ev, fde);
return;
}
/* there's no epoll_event attached to the fde */
if (want_read || (want_write && !got_error)) {
epoll_add_event(epoll_ev, fde);
return;
}
}
/*
event loop handling using epoll
*/
static int epoll_event_loop(struct epoll_event_context *epoll_ev, struct timeval *tvalp)
{
int ret, i;
#define MAXEVENTS 1
struct epoll_event events[MAXEVENTS];
int timeout = -1;
int wait_errno;
if (tvalp) {
/* it's better to trigger timed events a bit later than too early */
timeout = ((tvalp->tv_usec+999) / 1000) + (tvalp->tv_sec*1000);
}
if (epoll_ev->ev->signal_events &&
tevent_common_check_signal(epoll_ev->ev)) {
return 0;
}
tevent_trace_point_callback(epoll_ev->ev, TEVENT_TRACE_BEFORE_WAIT);
ret = epoll_wait(epoll_ev->epoll_fd, events, MAXEVENTS, timeout);
wait_errno = errno;
tevent_trace_point_callback(epoll_ev->ev, TEVENT_TRACE_AFTER_WAIT);
if (ret == -1 && wait_errno == EINTR && epoll_ev->ev->signal_events) {
if (tevent_common_check_signal(epoll_ev->ev)) {
return 0;
}
}
if (ret == -1 && wait_errno != EINTR) {
epoll_panic(epoll_ev, "epoll_wait() failed", true);
return -1;
}
if (ret == 0 && tvalp) {
r22661: optimize the handling of directly triggered timed events: - if someone adds a timed_event with a zero timeval we now avoid serval gettimeofday() calls and the event handler doesn't get the current time when it's called, instead we also pass a zero timeval - this also makes sure multiple timed events with a zero timeval are processed in the order there're added. the little benchmark shows that processing 2000000 directly timed events is now much faster, while avoiding syscalls at all! > time ./evtest (with the old code) real 0m6.388s user 0m1.740s sys 0m4.632s > time ./evtest (with the new code) real 0m1.498s user 0m1.496s sys 0m0.004s metze@SERNOX:~/devel/samba/4.0/samba4-ci/source> cat evtest.c #include <stdio.h> #include <stdint.h> #include <sys/time.h> #include <talloc.h> #include <events.h> static void dummy_fde_handler(struct event_context *ev_ctx, struct fd_event *fde, uint16_t flags, void *private_data) { } static void timeout_handler(struct event_context *ev, struct timed_event *te, struct timeval tval, void *private_data) { uint32_t *countp = (uint32_t *)private_data; (*countp)++; if (*countp > 2000000) exit(0); event_add_timed(ev, ev, tval, timeout_handler, countp); } int main(void) { struct event_context *ev; struct timeval tval = { 0, 0 }; uint32_t count = 0; ev = event_context_init(NULL); event_add_fd(ev, ev, 0, 0, dummy_fde_handler, NULL); event_add_timed(ev, ev, tval, timeout_handler, &count); return event_loop_wait(ev); } (This used to be commit 4db64b4ce2320b88d648078cbf86385f6fb44f1f)
2007-05-04 13:22:52 +04:00
/* we don't care about a possible delay here */
tevent_common_loop_timer_delay(epoll_ev->ev);
return 0;
}
for (i=0;i<ret;i++) {
struct tevent_fd *fde = talloc_get_type(events[i].data.ptr,
struct tevent_fd);
uint16_t flags = 0;
if (fde == NULL) {
epoll_panic(epoll_ev, "epoll_wait() gave bad data", true);
return -1;
}
if (events[i].events & (EPOLLHUP|EPOLLERR)) {
fde->additional_flags |= EPOLL_ADDITIONAL_FD_FLAG_GOT_ERROR;
/*
* if we only wait for TEVENT_FD_WRITE, we should not tell the
* event handler about it, and remove the epoll_event,
* as we only report errors when waiting for read events,
* to match the select() behavior
*/
if (!(fde->additional_flags & EPOLL_ADDITIONAL_FD_FLAG_REPORT_ERROR)) {
epoll_del_event(epoll_ev, fde);
continue;
}
flags |= TEVENT_FD_READ;
}
if (events[i].events & EPOLLIN) flags |= TEVENT_FD_READ;
if (events[i].events & EPOLLOUT) flags |= TEVENT_FD_WRITE;
if (flags) {
fde->handler(epoll_ev->ev, fde, flags, fde->private_data);
break;
}
}
return 0;
}
/*
create a epoll_event_context structure.
*/
static int epoll_event_context_init(struct tevent_context *ev)
{
int ret;
struct epoll_event_context *epoll_ev;
/*
* We might be called during tevent_re_initialise()
* which means we need to free our old additional_data.
*/
TALLOC_FREE(ev->additional_data);
epoll_ev = talloc_zero(ev, struct epoll_event_context);
if (!epoll_ev) return -1;
epoll_ev->ev = ev;
epoll_ev->epoll_fd = -1;
ret = epoll_init_ctx(epoll_ev);
if (ret != 0) {
talloc_free(epoll_ev);
return ret;
}
ev->additional_data = epoll_ev;
return 0;
}
/*
destroy an fd_event
*/
static int epoll_event_fd_destructor(struct tevent_fd *fde)
{
struct tevent_context *ev = fde->event_ctx;
struct epoll_event_context *epoll_ev = NULL;
bool panic_triggered = false;
if (ev == NULL) {
return tevent_common_fd_destructor(fde);
}
epoll_ev = talloc_get_type_abort(ev->additional_data,
struct epoll_event_context);
/*
* we must remove the event from the list
* otherwise a panic fallback handler may
* reuse invalid memory
*/
DLIST_REMOVE(ev->fd_events, fde);
epoll_ev->panic_state = &panic_triggered;
epoll_check_reopen(epoll_ev);
if (panic_triggered) {
return tevent_common_fd_destructor(fde);
}
epoll_del_event(epoll_ev, fde);
if (panic_triggered) {
return tevent_common_fd_destructor(fde);
}
epoll_ev->panic_state = NULL;
return tevent_common_fd_destructor(fde);
}
/*
add a fd based event
return NULL on failure (memory allocation error)
*/
static struct tevent_fd *epoll_event_add_fd(struct tevent_context *ev, TALLOC_CTX *mem_ctx,
int fd, uint16_t flags,
tevent_fd_handler_t handler,
void *private_data,
const char *handler_name,
const char *location)
{
struct epoll_event_context *epoll_ev = talloc_get_type(ev->additional_data,
struct epoll_event_context);
struct tevent_fd *fde;
bool panic_triggered = false;
fde = tevent_common_add_fd(ev, mem_ctx, fd, flags,
handler, private_data,
handler_name, location);
if (!fde) return NULL;
talloc_set_destructor(fde, epoll_event_fd_destructor);
epoll_ev->panic_state = &panic_triggered;
epoll_check_reopen(epoll_ev);
if (panic_triggered) {
return fde;
}
epoll_ev->panic_state = NULL;
epoll_add_event(epoll_ev, fde);
return fde;
}
/*
set the fd event flags
*/
static void epoll_event_set_fd_flags(struct tevent_fd *fde, uint16_t flags)
{
struct tevent_context *ev;
struct epoll_event_context *epoll_ev;
bool panic_triggered = false;
if (fde->flags == flags) return;
ev = fde->event_ctx;
epoll_ev = talloc_get_type(ev->additional_data, struct epoll_event_context);
fde->flags = flags;
epoll_ev->panic_state = &panic_triggered;
epoll_check_reopen(epoll_ev);
if (panic_triggered) {
return;
}
epoll_ev->panic_state = NULL;
epoll_change_event(epoll_ev, fde);
}
/*
do a single event loop using the events defined in ev
*/
static int epoll_event_loop_once(struct tevent_context *ev, const char *location)
{
struct epoll_event_context *epoll_ev = talloc_get_type(ev->additional_data,
struct epoll_event_context);
struct timeval tval;
bool panic_triggered = false;
if (ev->signal_events &&
tevent_common_check_signal(ev)) {
return 0;
}
if (ev->immediate_events &&
tevent_common_loop_immediate(ev)) {
return 0;
}
tval = tevent_common_loop_timer_delay(ev);
if (tevent_timeval_is_zero(&tval)) {
return 0;
}
epoll_ev->panic_state = &panic_triggered;
epoll_ev->panic_force_replay = true;
epoll_check_reopen(epoll_ev);
if (panic_triggered) {
errno = EINVAL;
return -1;
}
epoll_ev->panic_force_replay = false;
epoll_ev->panic_state = NULL;
return epoll_event_loop(epoll_ev, &tval);
}
static const struct tevent_ops epoll_event_ops = {
.context_init = epoll_event_context_init,
.add_fd = epoll_event_add_fd,
.set_fd_close_fn = tevent_common_fd_set_close_fn,
.get_fd_flags = tevent_common_fd_get_flags,
.set_fd_flags = epoll_event_set_fd_flags,
.add_timer = tevent_common_add_timer,
.schedule_immediate = tevent_common_schedule_immediate,
.add_signal = tevent_common_add_signal,
.loop_once = epoll_event_loop_once,
.loop_wait = tevent_common_loop_wait,
};
_PRIVATE_ bool tevent_epoll_init(void)
{
return tevent_register_backend("epoll", &epoll_event_ops);
}