1 /*-
2  * Copyright (c) 2011 NetApp, 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
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 /*
30  * Micro event library for FreeBSD, designed for a single i/o thread
31  * using EPOLL, and having events be persistent by default.
32  */
33 #include <errno.h>
34 #include <stdlib.h>
35 #include <stdio.h>
36 #include <stdbool.h>
37 #include <fcntl.h>
38 #include <unistd.h>
39 #include <sys/epoll.h>
40 #include <sys/queue.h>
41 #include <pthread.h>
42 
43 #include "mevent.h"
44 #include "vmmapi.h"
45 #include "log.h"
46 
47 #define	MEVENT_MAX	64
48 
49 #define	MEV_ADD		1
50 #define	MEV_ENABLE	2
51 #define	MEV_DISABLE	3
52 #define	MEV_DEL_PENDING	4
53 
54 static int epoll_fd;
55 static pthread_t mevent_tid;
56 static int mevent_pipefd[2];
57 static pthread_mutex_t mevent_lmutex;
58 
59 struct mevent {
60 	void			(*run)(int, enum ev_type, void *);
61 	void			*run_param;
62 	void			(*teardown)(void *);
63 	void			*teardown_param;
64 
65 	int			me_fd;
66 	enum			ev_type me_type;
67 	int			me_cq;
68 	int			me_state;
69 
70 	int			closefd;
71 	LIST_ENTRY(mevent)	me_list;
72 };
73 
74 static LIST_HEAD(listhead, mevent) global_head;
75 /* List holds the mevent node which is requested to be deleted */
76 static LIST_HEAD(del_listhead, mevent) del_head;
77 
78 static void
mevent_qlock(void)79 mevent_qlock(void)
80 {
81 	pthread_mutex_lock(&mevent_lmutex);
82 }
83 
84 static void
mevent_qunlock(void)85 mevent_qunlock(void)
86 {
87 	pthread_mutex_unlock(&mevent_lmutex);
88 }
89 
90 static bool
is_dispatch_thread(void)91 is_dispatch_thread(void)
92 {
93 	return (pthread_self() == mevent_tid);
94 }
95 
96 static void
mevent_pipe_read(int fd,enum ev_type type,void * param)97 mevent_pipe_read(int fd, enum ev_type type, void *param)
98 {
99 	char buf[MEVENT_MAX];
100 	ssize_t status;
101 
102 	/*
103 	 * Drain the pipe read side. The fd is non-blocking so this is
104 	 * safe to do.
105 	 */
106 	do {
107 		status = read(fd, buf, sizeof(buf));
108 	} while (status == MEVENT_MAX);
109 }
110 
111 /* On error, -1 is returned, else return zero */
112 int
mevent_notify(void)113 mevent_notify(void)
114 {
115 	char c = 0;
116 
117 	/*
118 	 * If calling from outside the i/o thread, write a byte on the
119 	 * pipe to force the i/o thread to exit the blocking epoll call.
120 	 */
121 	if (mevent_pipefd[1] != 0 && !is_dispatch_thread())
122 		if (write(mevent_pipefd[1], &c, 1) <= 0)
123 			return -1;
124 	return 0;
125 }
126 
127 static int
mevent_kq_filter(struct mevent * mevp)128 mevent_kq_filter(struct mevent *mevp)
129 {
130 	int retval;
131 
132 	retval = 0;
133 
134 	if (mevp->me_type == EVF_READ)
135 		retval = EPOLLIN;
136 
137 	if (mevp->me_type == EVF_READ_ET)
138 		retval = EPOLLIN | EPOLLET;
139 
140 	if (mevp->me_type == EVF_WRITE)
141 		retval = EPOLLOUT;
142 
143 	if (mevp->me_type == EVF_WRITE_ET)
144 		retval = EPOLLOUT | EPOLLET;
145 
146 	return retval;
147 }
148 
149 static void
mevent_destroy(void)150 mevent_destroy(void)
151 {
152 	struct mevent *mevp, *tmpp;
153 
154 	mevent_qlock();
155 	list_foreach_safe(mevp, &global_head, me_list, tmpp) {
156 		LIST_REMOVE(mevp, me_list);
157 		epoll_ctl(epoll_fd, EPOLL_CTL_DEL, mevp->me_fd, NULL);
158 
159                if ((mevp->me_type == EVF_READ ||
160                     mevp->me_type == EVF_READ_ET ||
161                     mevp->me_type == EVF_WRITE ||
162                     mevp->me_type == EVF_WRITE_ET) &&
163                     mevp->me_fd != STDIN_FILENO)
164                        close(mevp->me_fd);
165 
166 		if (mevp->teardown)
167 			mevp->teardown(mevp->teardown_param);
168 
169 		free(mevp);
170 	}
171 
172 	/* the mevp in del_head was removed from epoll when add it
173 	 * to del_head already.
174 	 */
175 	list_foreach_safe(mevp, &del_head, me_list, tmpp) {
176 		LIST_REMOVE(mevp, me_list);
177 
178                if ((mevp->me_type == EVF_READ ||
179                     mevp->me_type == EVF_READ_ET ||
180                     mevp->me_type == EVF_WRITE ||
181                     mevp->me_type == EVF_WRITE_ET) &&
182                     mevp->me_fd != STDIN_FILENO)
183                        close(mevp->me_fd);
184 
185 		if (mevp->teardown)
186 			mevp->teardown(mevp->teardown_param);
187 
188 		free(mevp);
189 	}
190 	mevent_qunlock();
191 }
192 
193 static void
mevent_handle(struct epoll_event * kev,int numev)194 mevent_handle(struct epoll_event *kev, int numev)
195 {
196 	int i;
197 	struct mevent *mevp;
198 
199 	for (i = 0; i < numev; i++) {
200 		mevp = kev[i].data.ptr;
201 
202 		if (mevp->me_state)
203 			(*mevp->run)(mevp->me_fd, mevp->me_type, mevp->run_param);
204 	}
205 }
206 
207 struct mevent *
mevent_add(int tfd,enum ev_type type,void (* run)(int,enum ev_type,void *),void * run_param,void (* teardown)(void *),void * teardown_param)208 mevent_add(int tfd, enum ev_type type,
209 	   void (*run)(int, enum ev_type, void *), void *run_param,
210 	   void (*teardown)(void *), void *teardown_param)
211 {
212 	int ret;
213 	struct epoll_event ee;
214 	struct mevent *lp, *mevp;
215 
216 	if (tfd < 0 || run == NULL)
217 		return NULL;
218 
219 	if (type == EVF_TIMER)
220 		return NULL;
221 
222 	mevent_qlock();
223 	/* Verify that the fd/type tuple is not present in the list */
224 	LIST_FOREACH(lp, &global_head, me_list) {
225 		if (lp->me_fd == tfd && lp->me_type == type) {
226 			mevent_qunlock();
227 			return lp;
228 		}
229 	}
230 	mevent_qunlock();
231 
232 	/*
233 	 * Allocate an entry, populate it, and add it to the list.
234 	 */
235 	mevp = calloc(1, sizeof(struct mevent));
236 	if (mevp == NULL)
237 		return NULL;
238 
239 	mevp->me_fd = tfd;
240 	mevp->me_type = type;
241 	mevp->me_state = 1;
242 
243 	mevp->run = run;
244 	mevp->run_param = run_param;
245 	mevp->teardown = teardown;
246 	mevp->teardown_param = teardown_param;
247 
248 	ee.events = mevent_kq_filter(mevp);
249 	ee.data.ptr = mevp;
250 	ret = epoll_ctl(epoll_fd, EPOLL_CTL_ADD, mevp->me_fd, &ee);
251 
252 	if (ret == 0) {
253 		mevent_qlock();
254 		LIST_INSERT_HEAD(&global_head, mevp, me_list);
255 		mevent_qunlock();
256 
257 		return mevp;
258 	} else {
259 		free(mevp);
260 		return NULL;
261 	}
262 }
263 
264 int
mevent_enable(struct mevent * evp)265 mevent_enable(struct mevent *evp)
266 {
267 	int ret;
268 	struct epoll_event ee;
269 	struct mevent *lp, *mevp = NULL;
270 
271 	mevent_qlock();
272 	/* Verify that the fd/type tuple is not present in the list */
273 	LIST_FOREACH(lp, &global_head, me_list) {
274 		if (lp == evp) {
275 			mevp = lp;
276 			break;
277 		}
278 	}
279 	mevent_qunlock();
280 
281 	if (!mevp)
282 		return -1;
283 
284 	ee.events = mevent_kq_filter(mevp);
285 	ee.data.ptr = mevp;
286 	ret = epoll_ctl(epoll_fd, EPOLL_CTL_ADD, mevp->me_fd, &ee);
287 	if (ret < 0 && errno == EEXIST)
288 		ret = 0;
289 
290 	return ret;
291 }
292 
293 int
mevent_disable(struct mevent * evp)294 mevent_disable(struct mevent *evp)
295 {
296 	int ret;
297 
298 	ret = epoll_ctl(epoll_fd, EPOLL_CTL_DEL, evp->me_fd, NULL);
299 	if (ret < 0 && errno == ENOENT)
300 		ret = 0;
301 
302 	return ret;
303 }
304 
305 static void
mevent_add_to_del_list(struct mevent * evp,int closefd)306 mevent_add_to_del_list(struct mevent *evp, int closefd)
307 {
308 	mevent_qlock();
309 	LIST_INSERT_HEAD(&del_head, evp, me_list);
310 	mevent_qunlock();
311 
312 	mevent_notify();
313 }
314 
315 static void
mevent_drain_del_list(void)316 mevent_drain_del_list(void)
317 {
318 	struct mevent *evp, *tmpp;
319 
320 	mevent_qlock();
321 	list_foreach_safe(evp, &del_head, me_list, tmpp) {
322 		LIST_REMOVE(evp, me_list);
323 		if (evp->closefd) {
324 			close(evp->me_fd);
325 		}
326 
327 		if (evp->teardown)
328 			evp->teardown(evp->teardown_param);
329 		free(evp);
330 	}
331 	mevent_qunlock();
332 }
333 
334 static int
mevent_delete_event(struct mevent * evp,int closefd)335 mevent_delete_event(struct mevent *evp, int closefd)
336 {
337 	mevent_qlock();
338 	LIST_REMOVE(evp, me_list);
339 	mevent_qunlock();
340 	evp->me_state = 0;
341 	evp->closefd = closefd;
342 
343 	epoll_ctl(epoll_fd, EPOLL_CTL_DEL, evp->me_fd, NULL);
344 	if (!is_dispatch_thread()) {
345 		mevent_add_to_del_list(evp, closefd);
346 	} else {
347 		if (evp->closefd) {
348 			close(evp->me_fd);
349 		}
350 
351 		if (evp->teardown)
352 			evp->teardown(evp->teardown_param);
353 		free(evp);
354 	}
355 	return 0;
356 }
357 
358 int
mevent_delete(struct mevent * evp)359 mevent_delete(struct mevent *evp)
360 {
361 	return mevent_delete_event(evp, 0);
362 }
363 
364 int
mevent_delete_close(struct mevent * evp)365 mevent_delete_close(struct mevent *evp)
366 {
367 	return mevent_delete_event(evp, 1);
368 }
369 
370 static void
mevent_set_name(void)371 mevent_set_name(void)
372 {
373 	pthread_setname_np(mevent_tid, "mevent");
374 }
375 
376 int
mevent_init(void)377 mevent_init(void)
378 {
379 	pthread_mutexattr_t attr;
380 
381 	pthread_mutexattr_init(&attr);
382 	pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
383 	pthread_mutex_init(&mevent_lmutex, &attr);
384 	pthread_mutexattr_destroy(&attr);
385 
386 	epoll_fd = epoll_create1(0);
387 
388 	if (epoll_fd >= 0)
389 		return 0;
390 	else
391 		return -1;
392 }
393 
394 void
mevent_deinit(void)395 mevent_deinit(void)
396 {
397 	mevent_destroy();
398 	close(epoll_fd);
399 	if (mevent_pipefd[1] != 0)
400 		close(mevent_pipefd[1]);
401 
402 	pthread_mutex_destroy(&mevent_lmutex);
403 }
404 
405 void
mevent_dispatch(void)406 mevent_dispatch(void)
407 {
408 	struct epoll_event eventlist[MEVENT_MAX];
409 
410 	struct mevent *pipev;
411 	int ret;
412 
413 	mevent_tid = pthread_self();
414 	mevent_set_name();
415 
416 	/*
417 	 * Open the pipe that will be used for other threads to force
418 	 * the blocking kqueue call to exit by writing to it. Set the
419 	 * descriptor to non-blocking.
420 	 */
421 	ret = pipe2(mevent_pipefd, O_NONBLOCK);
422 	if (ret < 0) {
423 		pr_err("pipe");
424 		exit(0);
425 	}
426 
427 	/*
428 	 * Add internal event handler for the pipe write fd
429 	 */
430 	pipev = mevent_add(mevent_pipefd[0], EVF_READ, mevent_pipe_read, NULL, NULL, NULL);
431 	if (!pipev) {
432 		pr_err("pipefd mevent_add failed\n");
433 		exit(0);
434 	}
435 
436 	for (;;) {
437 		int suspend_mode;
438 
439 		/*
440 		 * Block awaiting events
441 		 */
442 		ret = epoll_wait(epoll_fd, eventlist, MEVENT_MAX, -1);
443 
444 		if (ret == -1 && errno != EINTR)
445 			pr_err("Error return from epoll_wait");
446 
447 		/*
448 		 * Handle reported events
449 		 */
450 		mevent_handle(eventlist, ret);
451 		mevent_drain_del_list();
452 
453 		suspend_mode = vm_get_suspend_mode();
454 		if ((suspend_mode != VM_SUSPEND_NONE) &&
455 		    (suspend_mode != VM_SUSPEND_SYSTEM_RESET) &&
456 		    (suspend_mode != VM_SUSPEND_SUSPEND))
457 			break;
458 	}
459 }
460