| 1 |
/* |
|---|
| 2 |
* Copyright (c) 2007, OmniTI Computer Consulting, Inc. |
|---|
| 3 |
* All rights reserved. |
|---|
| 4 |
* |
|---|
| 5 |
* Redistribution and use in source and binary forms, with or without |
|---|
| 6 |
* modification, are permitted provided that the following conditions are |
|---|
| 7 |
* met: |
|---|
| 8 |
* |
|---|
| 9 |
* * Redistributions of source code must retain the above copyright |
|---|
| 10 |
* notice, this list of conditions and the following disclaimer. |
|---|
| 11 |
* * Redistributions in binary form must reproduce the above |
|---|
| 12 |
* copyright notice, this list of conditions and the following |
|---|
| 13 |
* disclaimer in the documentation and/or other materials provided |
|---|
| 14 |
* with the distribution. |
|---|
| 15 |
* * Neither the name OmniTI Computer Consulting, Inc. nor the names |
|---|
| 16 |
* of its contributors may be used to endorse or promote products |
|---|
| 17 |
* derived from this software without specific prior written |
|---|
| 18 |
* permission. |
|---|
| 19 |
* |
|---|
| 20 |
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
|---|
| 21 |
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
|---|
| 22 |
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
|---|
| 23 |
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
|---|
| 24 |
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
|---|
| 25 |
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
|---|
| 26 |
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
|---|
| 27 |
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
|---|
| 28 |
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
|---|
| 29 |
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
|---|
| 30 |
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
|---|
| 31 |
*/ |
|---|
| 32 |
|
|---|
| 33 |
#include "noit_defines.h" |
|---|
| 34 |
#include "eventer/eventer.h" |
|---|
| 35 |
#include "utils/noit_atomic.h" |
|---|
| 36 |
#include "utils/noit_skiplist.h" |
|---|
| 37 |
#include "utils/noit_log.h" |
|---|
| 38 |
#include "eventer/dtrace_probes.h" |
|---|
| 39 |
|
|---|
| 40 |
#include <errno.h> |
|---|
| 41 |
#include <stdio.h> |
|---|
| 42 |
#include <stdlib.h> |
|---|
| 43 |
#include <sys/epoll.h> |
|---|
| 44 |
#include <signal.h> |
|---|
| 45 |
#include <pthread.h> |
|---|
| 46 |
#include <assert.h> |
|---|
| 47 |
|
|---|
| 48 |
struct _eventer_impl eventer_epoll_impl; |
|---|
| 49 |
#define LOCAL_EVENTER eventer_epoll_impl |
|---|
| 50 |
#define LOCAL_EVENTER_foreach_fdevent eventer_epoll_impl_foreach_fdevent |
|---|
| 51 |
#define maxfds LOCAL_EVENTER.maxfds |
|---|
| 52 |
#define master_fds LOCAL_EVENTER.master_fds |
|---|
| 53 |
|
|---|
| 54 |
#include "eventer/eventer_impl_private.h" |
|---|
| 55 |
|
|---|
| 56 |
static int *masks; |
|---|
| 57 |
static int epoll_fd = -1; |
|---|
| 58 |
|
|---|
| 59 |
static int eventer_epoll_impl_init() { |
|---|
| 60 |
struct rlimit rlim; |
|---|
| 61 |
int rv; |
|---|
| 62 |
|
|---|
| 63 |
/* super init */ |
|---|
| 64 |
if((rv = eventer_impl_init()) != 0) return rv; |
|---|
| 65 |
|
|---|
| 66 |
signal(SIGPIPE, SIG_IGN); |
|---|
| 67 |
epoll_fd = epoll_create(1024); |
|---|
| 68 |
if(epoll_fd == -1) { |
|---|
| 69 |
return -1; |
|---|
| 70 |
} |
|---|
| 71 |
getrlimit(RLIMIT_NOFILE, &rlim); |
|---|
| 72 |
maxfds = rlim.rlim_cur; |
|---|
| 73 |
master_fds = calloc(maxfds, sizeof(*master_fds)); |
|---|
| 74 |
masks = calloc(maxfds, sizeof(*masks)); |
|---|
| 75 |
return 0; |
|---|
| 76 |
} |
|---|
| 77 |
static int eventer_epoll_impl_propset(const char *key, const char *value) { |
|---|
| 78 |
if(eventer_impl_propset(key, value)) { |
|---|
| 79 |
/* Do our epoll local properties here */ |
|---|
| 80 |
return -1; |
|---|
| 81 |
} |
|---|
| 82 |
return 0; |
|---|
| 83 |
} |
|---|
| 84 |
static void eventer_epoll_impl_add(eventer_t e) { |
|---|
| 85 |
struct epoll_event _ev; |
|---|
| 86 |
ev_lock_state_t lockstate; |
|---|
| 87 |
assert(e->mask); |
|---|
| 88 |
|
|---|
| 89 |
if(e->mask & EVENTER_ASYNCH) { |
|---|
| 90 |
eventer_add_asynch(NULL, e); |
|---|
| 91 |
return; |
|---|
| 92 |
} |
|---|
| 93 |
|
|---|
| 94 |
/* Recurrent delegation */ |
|---|
| 95 |
if(e->mask & EVENTER_RECURRENT) { |
|---|
| 96 |
eventer_add_recurrent(e); |
|---|
| 97 |
return; |
|---|
| 98 |
} |
|---|
| 99 |
|
|---|
| 100 |
/* Timed events are simple */ |
|---|
| 101 |
if(e->mask & EVENTER_TIMER) { |
|---|
| 102 |
eventer_add_timed(e); |
|---|
| 103 |
return; |
|---|
| 104 |
} |
|---|
| 105 |
|
|---|
| 106 |
/* file descriptor event */ |
|---|
| 107 |
assert(e->whence.tv_sec == 0 && e->whence.tv_usec == 0); |
|---|
| 108 |
memset(&_ev, 0, sizeof(_ev)); |
|---|
| 109 |
_ev.data.fd = e->fd; |
|---|
| 110 |
if(e->mask & EVENTER_READ) _ev.events |= (EPOLLIN|EPOLLPRI); |
|---|
| 111 |
if(e->mask & EVENTER_WRITE) _ev.events |= (EPOLLOUT); |
|---|
| 112 |
if(e->mask & EVENTER_EXCEPTION) _ev.events |= (EPOLLERR|EPOLLHUP); |
|---|
| 113 |
|
|---|
| 114 |
lockstate = acquire_master_fd(e->fd); |
|---|
| 115 |
master_fds[e->fd].e = e; |
|---|
| 116 |
|
|---|
| 117 |
assert(epoll_ctl(epoll_fd, EPOLL_CTL_ADD, e->fd, &_ev) == 0); |
|---|
| 118 |
|
|---|
| 119 |
release_master_fd(e->fd, lockstate); |
|---|
| 120 |
} |
|---|
| 121 |
static eventer_t eventer_epoll_impl_remove(eventer_t e) { |
|---|
| 122 |
eventer_t removed = NULL; |
|---|
| 123 |
if(e->mask & EVENTER_ASYNCH) { |
|---|
| 124 |
abort(); |
|---|
| 125 |
} |
|---|
| 126 |
if(e->mask & (EVENTER_READ | EVENTER_WRITE | EVENTER_EXCEPTION)) { |
|---|
| 127 |
ev_lock_state_t lockstate; |
|---|
| 128 |
struct epoll_event _ev; |
|---|
| 129 |
memset(&_ev, 0, sizeof(_ev)); |
|---|
| 130 |
_ev.data.fd = e->fd; |
|---|
| 131 |
lockstate = acquire_master_fd(e->fd); |
|---|
| 132 |
if(e == master_fds[e->fd].e) { |
|---|
| 133 |
removed = e; |
|---|
| 134 |
master_fds[e->fd].e = NULL; |
|---|
| 135 |
assert(epoll_ctl(epoll_fd, EPOLL_CTL_DEL, e->fd, &_ev) == 0); |
|---|
| 136 |
} |
|---|
| 137 |
release_master_fd(e->fd, lockstate); |
|---|
| 138 |
} |
|---|
| 139 |
else if(e->mask & EVENTER_TIMER) { |
|---|
| 140 |
removed = eventer_remove_timed(e); |
|---|
| 141 |
} |
|---|
| 142 |
else if(e->mask & EVENTER_RECURRENT) { |
|---|
| 143 |
removed = eventer_remove_recurrent(e); |
|---|
| 144 |
} |
|---|
| 145 |
else { |
|---|
| 146 |
abort(); |
|---|
| 147 |
} |
|---|
| 148 |
return removed; |
|---|
| 149 |
} |
|---|
| 150 |
static void eventer_epoll_impl_update(eventer_t e, int mask) { |
|---|
| 151 |
struct epoll_event _ev; |
|---|
| 152 |
if(e->mask & EVENTER_TIMER) { |
|---|
| 153 |
eventer_update_timed(e,mask); |
|---|
| 154 |
return; |
|---|
| 155 |
} |
|---|
| 156 |
memset(&_ev, 0, sizeof(_ev)); |
|---|
| 157 |
_ev.data.fd = e->fd; |
|---|
| 158 |
e->mask = mask; |
|---|
| 159 |
if(e->mask & (EVENTER_READ | EVENTER_WRITE | EVENTER_EXCEPTION)) { |
|---|
| 160 |
if(e->mask & EVENTER_READ) _ev.events |= (EPOLLIN|EPOLLPRI); |
|---|
| 161 |
if(e->mask & EVENTER_WRITE) _ev.events |= (EPOLLOUT); |
|---|
| 162 |
if(e->mask & EVENTER_EXCEPTION) _ev.events |= (EPOLLERR|EPOLLHUP); |
|---|
| 163 |
assert(epoll_ctl(epoll_fd, EPOLL_CTL_MOD, e->fd, &_ev) == 0); |
|---|
| 164 |
} |
|---|
| 165 |
} |
|---|
| 166 |
static eventer_t eventer_epoll_impl_remove_fd(int fd) { |
|---|
| 167 |
eventer_t eiq = NULL; |
|---|
| 168 |
ev_lock_state_t lockstate; |
|---|
| 169 |
if(master_fds[fd].e) { |
|---|
| 170 |
struct epoll_event _ev; |
|---|
| 171 |
memset(&_ev, 0, sizeof(_ev)); |
|---|
| 172 |
_ev.data.fd = fd; |
|---|
| 173 |
lockstate = acquire_master_fd(fd); |
|---|
| 174 |
eiq = master_fds[fd].e; |
|---|
| 175 |
master_fds[fd].e = NULL; |
|---|
| 176 |
assert(epoll_ctl(epoll_fd, EPOLL_CTL_DEL, fd, &_ev) == 0); |
|---|
| 177 |
release_master_fd(fd, lockstate); |
|---|
| 178 |
} |
|---|
| 179 |
return eiq; |
|---|
| 180 |
} |
|---|
| 181 |
static eventer_t eventer_epoll_impl_find_fd(int fd) { |
|---|
| 182 |
return master_fds[fd].e; |
|---|
| 183 |
} |
|---|
| 184 |
|
|---|
| 185 |
static void eventer_epoll_impl_trigger(eventer_t e, int mask) { |
|---|
| 186 |
struct timeval __now; |
|---|
| 187 |
int fd, oldmask, newmask; |
|---|
| 188 |
const char *cbname; |
|---|
| 189 |
ev_lock_state_t lockstate; |
|---|
| 190 |
|
|---|
| 191 |
fd = e->fd; |
|---|
| 192 |
if(e != master_fds[fd].e) return; |
|---|
| 193 |
lockstate = acquire_master_fd(fd); |
|---|
| 194 |
if(lockstate == EV_ALREADY_OWNED) return; |
|---|
| 195 |
assert(lockstate == EV_OWNED); |
|---|
| 196 |
|
|---|
| 197 |
gettimeofday(&__now, NULL); |
|---|
| 198 |
oldmask = e->mask; |
|---|
| 199 |
cbname = eventer_name_for_callback(e->callback); |
|---|
| 200 |
noitLT(eventer_deb, &__now, "epoll: fire on %d/%x to %s(%p)\n", |
|---|
| 201 |
fd, mask, cbname?cbname:"???", e->callback); |
|---|
| 202 |
EVENTER_CALLBACK_ENTRY(e->callback, cbname, fd, e->mask, mask); |
|---|
| 203 |
newmask = e->callback(e, mask, e->closure, &__now); |
|---|
| 204 |
EVENTER_CALLBACK_RETURN(e->callback, cbname, newmask); |
|---|
| 205 |
|
|---|
| 206 |
if(newmask) { |
|---|
| 207 |
struct epoll_event _ev; |
|---|
| 208 |
memset(&_ev, 0, sizeof(_ev)); |
|---|
| 209 |
_ev.data.fd = fd; |
|---|
| 210 |
if(newmask & EVENTER_READ) _ev.events |= (EPOLLIN|EPOLLPRI); |
|---|
| 211 |
if(newmask & EVENTER_WRITE) _ev.events |= (EPOLLOUT); |
|---|
| 212 |
if(newmask & EVENTER_EXCEPTION) _ev.events |= (EPOLLERR|EPOLLHUP); |
|---|
| 213 |
assert(epoll_ctl(epoll_fd, EPOLL_CTL_MOD, fd, &_ev) == 0); |
|---|
| 214 |
/* Set our mask */ |
|---|
| 215 |
e->mask = newmask; |
|---|
| 216 |
} |
|---|
| 217 |
else { |
|---|
| 218 |
/* see kqueue implementation for details on the next line */ |
|---|
| 219 |
if(master_fds[fd].e == e) master_fds[fd].e = NULL; |
|---|
| 220 |
eventer_free(e); |
|---|
| 221 |
} |
|---|
| 222 |
release_master_fd(fd, lockstate); |
|---|
| 223 |
} |
|---|
| 224 |
static int eventer_epoll_impl_loop() { |
|---|
| 225 |
struct epoll_event *epev; |
|---|
| 226 |
|
|---|
| 227 |
epev = malloc(sizeof(*epev) * maxfds); |
|---|
| 228 |
|
|---|
| 229 |
while(1) { |
|---|
| 230 |
struct timeval __now, __sleeptime; |
|---|
| 231 |
int fd_cnt = 0; |
|---|
| 232 |
|
|---|
| 233 |
__sleeptime = eventer_max_sleeptime; |
|---|
| 234 |
|
|---|
| 235 |
eventer_dispatch_timed(&__now, &__sleeptime); |
|---|
| 236 |
|
|---|
| 237 |
/* Handle recurrent events */ |
|---|
| 238 |
eventer_dispatch_recurrent(&__now); |
|---|
| 239 |
|
|---|
| 240 |
/* Now we move on to our fd-based events */ |
|---|
| 241 |
fd_cnt = epoll_wait(epoll_fd, epev, maxfds, |
|---|
| 242 |
__sleeptime.tv_sec * 1000 + __sleeptime.tv_usec / 1000); |
|---|
| 243 |
noitLT(eventer_deb, &__now, "debug: epoll_wait(%d, [], %d) => %d\n", epoll_fd, maxfds, fd_cnt); |
|---|
| 244 |
if(fd_cnt < 0) { |
|---|
| 245 |
noitLT(eventer_err, &__now, "epoll_wait: %s\n", strerror(errno)); |
|---|
| 246 |
} |
|---|
| 247 |
else { |
|---|
| 248 |
int idx; |
|---|
| 249 |
/* loop once to clear */ |
|---|
| 250 |
for(idx = 0; idx < fd_cnt; idx++) { |
|---|
| 251 |
struct epoll_event *ev; |
|---|
| 252 |
eventer_t e; |
|---|
| 253 |
int fd, mask = 0; |
|---|
| 254 |
|
|---|
| 255 |
ev = &epev[idx]; |
|---|
| 256 |
|
|---|
| 257 |
if(ev->events & (EPOLLIN | EPOLLPRI)) mask |= EVENTER_READ; |
|---|
| 258 |
if(ev->events & (EPOLLOUT)) mask |= EVENTER_WRITE; |
|---|
| 259 |
if(ev->events & (EPOLLERR|EPOLLHUP)) mask |= EVENTER_EXCEPTION; |
|---|
| 260 |
|
|---|
| 261 |
fd = ev->data.fd; |
|---|
| 262 |
|
|---|
| 263 |
e = master_fds[fd].e; |
|---|
| 264 |
/* It's possible that someone removed the event and freed it |
|---|
| 265 |
* before we got here. |
|---|
| 266 |
*/ |
|---|
| 267 |
if(!e) continue; |
|---|
| 268 |
|
|---|
| 269 |
eventer_epoll_impl_trigger(e, mask); |
|---|
| 270 |
} |
|---|
| 271 |
} |
|---|
| 272 |
} |
|---|
| 273 |
/* NOTREACHED */ |
|---|
| 274 |
return 0; |
|---|
| 275 |
} |
|---|
| 276 |
|
|---|
| 277 |
struct _eventer_impl eventer_epoll_impl = { |
|---|
| 278 |
"epoll", |
|---|
| 279 |
eventer_epoll_impl_init, |
|---|
| 280 |
eventer_epoll_impl_propset, |
|---|
| 281 |
eventer_epoll_impl_add, |
|---|
| 282 |
eventer_epoll_impl_remove, |
|---|
| 283 |
eventer_epoll_impl_update, |
|---|
| 284 |
eventer_epoll_impl_remove_fd, |
|---|
| 285 |
eventer_epoll_impl_find_fd, |
|---|
| 286 |
eventer_epoll_impl_trigger, |
|---|
| 287 |
eventer_epoll_impl_loop, |
|---|
| 288 |
eventer_epoll_impl_foreach_fdevent, |
|---|
| 289 |
{ 0, 200000 }, |
|---|
| 290 |
0, |
|---|
| 291 |
NULL |
|---|
| 292 |
}; |
|---|