1 // Copyright (c) 2009 The Chromium Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style license that can be 3 // found in the LICENSE file. 4 5 #include "base/message_pump_libevent.h" 6 7 #include <errno.h> 8 #include <fcntl.h> 9 10 #include "eintr_wrapper.h" 11 #include "base/auto_reset.h" 12 #include "base/logging.h" 13 #include "base/scoped_nsautorelease_pool.h" 14 #include "base/scoped_ptr.h" 15 #include "base/time.h" 16 #if defined(USE_SYSTEM_LIBEVENT) 17 #include <event.h> 18 #else 19 #include "third_party/libevent/event.h" 20 #endif 21 22 // Lifecycle of struct event 23 // Libevent uses two main data structures: 24 // struct event_base (of which there is one per message pump), and 25 // struct event (of which there is roughly one per socket). 26 // The socket's struct event is created in 27 // MessagePumpLibevent::WatchFileDescriptor(), 28 // is owned by the FileDescriptorWatcher, and is destroyed in 29 // StopWatchingFileDescriptor(). 30 // It is moved into and out of lists in struct event_base by 31 // the libevent functions event_add() and event_del(). 32 // 33 // TODO(dkegel): 34 // At the moment bad things happen if a FileDescriptorWatcher 35 // is active after its MessagePumpLibevent has been destroyed. 36 // See MessageLoopTest.FileDescriptorWatcherOutlivesMessageLoop 37 // Not clear yet whether that situation occurs in practice, 38 // but if it does, we need to fix it. 39 40 namespace base { 41 42 // Return 0 on success 43 // Too small a function to bother putting in a library? 44 static int SetNonBlocking(int fd) { 45 int flags = fcntl(fd, F_GETFL, 0); 46 if (flags == -1) 47 flags = 0; 48 return fcntl(fd, F_SETFL, flags | O_NONBLOCK); 49 } 50 51 MessagePumpLibevent::FileDescriptorWatcher::FileDescriptorWatcher() 52 : is_persistent_(false), 53 event_(NULL) { 54 } 55 56 MessagePumpLibevent::FileDescriptorWatcher::~FileDescriptorWatcher() { 57 if (event_) { 58 StopWatchingFileDescriptor(); 59 } 60 } 61 62 void MessagePumpLibevent::FileDescriptorWatcher::Init(event *e, 63 bool is_persistent) { 64 DCHECK(e); 65 DCHECK(event_ == NULL); 66 67 is_persistent_ = is_persistent; 68 event_ = e; 69 } 70 71 event *MessagePumpLibevent::FileDescriptorWatcher::ReleaseEvent() { 72 struct event *e = event_; 73 event_ = NULL; 74 return e; 75 } 76 77 bool MessagePumpLibevent::FileDescriptorWatcher::StopWatchingFileDescriptor() { 78 event* e = ReleaseEvent(); 79 if (e == NULL) 80 return true; 81 82 // event_del() is a no-op if the event isn't active. 83 int rv = event_del(e); 84 delete e; 85 return (rv == 0); 86 } 87 88 // Called if a byte is received on the wakeup pipe. 89 void MessagePumpLibevent::OnWakeup(int socket, short flags, void* context) { 90 base::MessagePumpLibevent* that = 91 static_cast<base::MessagePumpLibevent*>(context); 92 DCHECK(that->wakeup_pipe_out_ == socket); 93 94 // Remove and discard the wakeup byte. 95 char buf; 96 int nread = HANDLE_EINTR(read(socket, &buf, 1)); 97 DCHECK_EQ(nread, 1); 98 // Tell libevent to break out of inner loop. 99 event_base_loopbreak(that->event_base_); 100 } 101 102 MessagePumpLibevent::MessagePumpLibevent() 103 : keep_running_(true), 104 in_run_(false), 105 event_base_(event_base_new()), 106 wakeup_pipe_in_(-1), 107 wakeup_pipe_out_(-1) { 108 if (!Init()) 109 NOTREACHED(); 110 } 111 112 bool MessagePumpLibevent::Init() { 113 int fds[2]; 114 if (pipe(fds)) { 115 DLOG(ERROR) << "pipe() failed, errno: " << errno; 116 return false; 117 } 118 if (SetNonBlocking(fds[0])) { 119 DLOG(ERROR) << "SetNonBlocking for pipe fd[0] failed, errno: " << errno; 120 return false; 121 } 122 if (SetNonBlocking(fds[1])) { 123 DLOG(ERROR) << "SetNonBlocking for pipe fd[1] failed, errno: " << errno; 124 return false; 125 } 126 wakeup_pipe_out_ = fds[0]; 127 wakeup_pipe_in_ = fds[1]; 128 129 wakeup_event_ = new event; 130 event_set(wakeup_event_, wakeup_pipe_out_, EV_READ | EV_PERSIST, 131 OnWakeup, this); 132 event_base_set(event_base_, wakeup_event_); 133 134 if (event_add(wakeup_event_, 0)) 135 return false; 136 return true; 137 } 138 139 MessagePumpLibevent::~MessagePumpLibevent() { 140 DCHECK(wakeup_event_); 141 DCHECK(event_base_); 142 event_del(wakeup_event_); 143 delete wakeup_event_; 144 if (wakeup_pipe_in_ >= 0) 145 close(wakeup_pipe_in_); 146 if (wakeup_pipe_out_ >= 0) 147 close(wakeup_pipe_out_); 148 event_base_free(event_base_); 149 } 150 151 bool MessagePumpLibevent::WatchFileDescriptor(int fd, 152 bool persistent, 153 Mode mode, 154 FileDescriptorWatcher *controller, 155 Watcher *delegate) { 156 DCHECK_GE(fd, 0); 157 DCHECK(controller); 158 DCHECK(delegate); 159 DCHECK(mode == WATCH_READ || mode == WATCH_WRITE || mode == WATCH_READ_WRITE); 160 161 int event_mask = persistent ? EV_PERSIST : 0; 162 if ((mode & WATCH_READ) != 0) { 163 event_mask |= EV_READ; 164 } 165 if ((mode & WATCH_WRITE) != 0) { 166 event_mask |= EV_WRITE; 167 } 168 169 scoped_ptr<event> evt(controller->ReleaseEvent()); 170 if (evt.get() == NULL) { 171 // Ownership is transferred to the controller. 172 evt.reset(new event); 173 } else { 174 // Make sure we don't pick up any funky internal libevent masks. 175 int old_interest_mask = evt.get()->ev_events & 176 (EV_READ | EV_WRITE | EV_PERSIST); 177 178 // Combine old/new event masks. 179 event_mask |= old_interest_mask; 180 181 // Must disarm the event before we can reuse it. 182 event_del(evt.get()); 183 184 // It's illegal to use this function to listen on 2 separate fds with the 185 // same |controller|. 186 if (EVENT_FD(evt.get()) != fd) { 187 NOTREACHED() << "FDs don't match" << EVENT_FD(evt.get()) << "!=" << fd; 188 return false; 189 } 190 } 191 192 // Set current interest mask and message pump for this event. 193 event_set(evt.get(), fd, event_mask, OnLibeventNotification, delegate); 194 195 // Tell libevent which message pump this socket will belong to when we add it. 196 if (event_base_set(event_base_, evt.get()) != 0) { 197 return false; 198 } 199 200 // Add this socket to the list of monitored sockets. 201 if (event_add(evt.get(), NULL) != 0) { 202 return false; 203 } 204 205 // Transfer ownership of evt to controller. 206 controller->Init(evt.release(), persistent); 207 return true; 208 } 209 210 211 void MessagePumpLibevent::OnLibeventNotification(int fd, short flags, 212 void* context) { 213 Watcher* watcher = static_cast<Watcher*>(context); 214 215 if (flags & EV_WRITE) { 216 watcher->OnFileCanWriteWithoutBlocking(fd); 217 } 218 if (flags & EV_READ) { 219 watcher->OnFileCanReadWithoutBlocking(fd); 220 } 221 } 222 223 // Tell libevent to break out of inner loop. 224 static void timer_callback(int fd, short events, void *context) 225 { 226 event_base_loopbreak((struct event_base *)context); 227 } 228 229 // Reentrant! 230 void MessagePumpLibevent::Run(Delegate* delegate) { 231 DCHECK(keep_running_) << "Quit must have been called outside of Run!"; 232 AutoReset auto_reset_in_run(&in_run_, true); 233 234 // event_base_loopexit() + EVLOOP_ONCE is leaky, see http://crbug.com/25641. 235 // Instead, make our own timer and reuse it on each call to event_base_loop(). 236 scoped_ptr<event> timer_event(new event); 237 238 for (;;) { 239 ScopedNSAutoreleasePool autorelease_pool; 240 241 bool did_work = delegate->DoWork(); 242 if (!keep_running_) 243 break; 244 245 did_work |= delegate->DoDelayedWork(&delayed_work_time_); 246 if (!keep_running_) 247 break; 248 249 if (did_work) 250 continue; 251 252 did_work = delegate->DoIdleWork(); 253 if (!keep_running_) 254 break; 255 256 if (did_work) 257 continue; 258 259 // EVLOOP_ONCE tells libevent to only block once, 260 // but to service all pending events when it wakes up. 261 if (delayed_work_time_.is_null()) { 262 event_base_loop(event_base_, EVLOOP_ONCE); 263 } else { 264 TimeDelta delay = delayed_work_time_ - Time::Now(); 265 if (delay > TimeDelta()) { 266 struct timeval poll_tv; 267 poll_tv.tv_sec = delay.InSeconds(); 268 poll_tv.tv_usec = delay.InMicroseconds() % Time::kMicrosecondsPerSecond; 269 event_set(timer_event.get(), -1, 0, timer_callback, event_base_); 270 event_base_set(event_base_, timer_event.get()); 271 event_add(timer_event.get(), &poll_tv); 272 event_base_loop(event_base_, EVLOOP_ONCE); 273 event_del(timer_event.get()); 274 } else { 275 // It looks like delayed_work_time_ indicates a time in the past, so we 276 // need to call DoDelayedWork now. 277 delayed_work_time_ = Time(); 278 } 279 } 280 } 281 282 keep_running_ = true; 283 } 284 285 void MessagePumpLibevent::Quit() { 286 DCHECK(in_run_); 287 // Tell both libevent and Run that they should break out of their loops. 288 keep_running_ = false; 289 ScheduleWork(); 290 } 291 292 void MessagePumpLibevent::ScheduleWork() { 293 // Tell libevent (in a threadsafe way) that it should break out of its loop. 294 char buf = 0; 295 int nwrite = HANDLE_EINTR(write(wakeup_pipe_in_, &buf, 1)); 296 DCHECK(nwrite == 1 || errno == EAGAIN) 297 << "[nwrite:" << nwrite << "] [errno:" << errno << "]"; 298 } 299 300 void MessagePumpLibevent::ScheduleDelayedWork(const Time& delayed_work_time) { 301 // We know that we can't be blocked on Wait right now since this method can 302 // only be called on the same thread as Run, so we only need to update our 303 // record of how long to sleep when we do sleep. 304 delayed_work_time_ = delayed_work_time; 305 } 306 307 } // namespace base 308