xref: /freebsd/usr.sbin/bhyve/mevent.c (revision 2a63c3be158216222d89a073dcbd6a72ee4aab5a) 
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011 NetApp, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 /*
30  * Micro event library for FreeBSD, designed for a single i/o thread
31  * using kqueue, and having events be persistent by default.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <assert.h>
38 #ifndef WITHOUT_CAPSICUM
39 #include <capsicum_helpers.h>
40 #endif
41 #include <err.h>
42 #include <errno.h>
43 #include <stdbool.h>
44 #include <stdlib.h>
45 #include <stdio.h>
46 #include <string.h>
47 #include <sysexits.h>
48 #include <unistd.h>
49 
50 #include <sys/types.h>
51 #ifndef WITHOUT_CAPSICUM
52 #include <sys/capsicum.h>
53 #endif
54 #include <sys/event.h>
55 #include <sys/time.h>
56 
57 #include <pthread.h>
58 #include <pthread_np.h>
59 
60 #include "mevent.h"
61 
62 #define	MEVENT_MAX	64
63 
64 static pthread_t mevent_tid;
65 static pthread_once_t mevent_once = PTHREAD_ONCE_INIT;
66 static int mevent_timid = 43;
67 static int mevent_pipefd[2];
68 static int mfd;
69 static pthread_mutex_t mevent_lmutex = PTHREAD_MUTEX_INITIALIZER;
70 
71 struct mevent {
72 	void	(*me_func)(int, enum ev_type, void *);
73 #define me_msecs me_fd
74 	int	me_fd;
75 	int	me_timid;
76 	enum ev_type me_type;
77 	void    *me_param;
78 	int	me_cq;
79 	int	me_state; /* Desired kevent flags. */
80 	int	me_closefd;
81 	int	me_fflags;
82 	LIST_ENTRY(mevent) me_list;
83 };
84 
85 static LIST_HEAD(listhead, mevent) global_head, change_head;
86 
87 static void
88 mevent_qlock(void)
89 {
90 	pthread_mutex_lock(&mevent_lmutex);
91 }
92 
93 static void
94 mevent_qunlock(void)
95 {
96 	pthread_mutex_unlock(&mevent_lmutex);
97 }
98 
99 static void
100 mevent_pipe_read(int fd, enum ev_type type __unused, void *param __unused)
101 {
102 	char buf[MEVENT_MAX];
103 	int status;
104 
105 	/*
106 	 * Drain the pipe read side. The fd is non-blocking so this is
107 	 * safe to do.
108 	 */
109 	do {
110 		status = read(fd, buf, sizeof(buf));
111 	} while (status == MEVENT_MAX);
112 }
113 
114 static void
115 mevent_notify(void)
116 {
117 	char c = '\0';
118 
119 	/*
120 	 * If calling from outside the i/o thread, write a byte on the
121 	 * pipe to force the i/o thread to exit the blocking kevent call.
122 	 */
123 	if (mevent_pipefd[1] != 0 && pthread_self() != mevent_tid) {
124 		write(mevent_pipefd[1], &c, 1);
125 	}
126 }
127 
128 static void
129 mevent_init(void)
130 {
131 #ifndef WITHOUT_CAPSICUM
132 	cap_rights_t rights;
133 #endif
134 
135 	mfd = kqueue();
136 	assert(mfd > 0);
137 
138 #ifndef WITHOUT_CAPSICUM
139 	cap_rights_init(&rights, CAP_KQUEUE);
140 	if (caph_rights_limit(mfd, &rights) == -1)
141 		errx(EX_OSERR, "Unable to apply rights for sandbox");
142 #endif
143 
144 	LIST_INIT(&change_head);
145 	LIST_INIT(&global_head);
146 }
147 
148 static int
149 mevent_kq_filter(struct mevent *mevp)
150 {
151 	int retval;
152 
153 	retval = 0;
154 
155 	if (mevp->me_type == EVF_READ)
156 		retval = EVFILT_READ;
157 
158 	if (mevp->me_type == EVF_WRITE)
159 		retval = EVFILT_WRITE;
160 
161 	if (mevp->me_type == EVF_TIMER)
162 		retval = EVFILT_TIMER;
163 
164 	if (mevp->me_type == EVF_SIGNAL)
165 		retval = EVFILT_SIGNAL;
166 
167 	if (mevp->me_type == EVF_VNODE)
168 		retval = EVFILT_VNODE;
169 
170 	return (retval);
171 }
172 
173 static int
174 mevent_kq_flags(struct mevent *mevp)
175 {
176 	int retval;
177 
178 	retval = mevp->me_state;
179 
180 	if (mevp->me_type == EVF_VNODE)
181 		retval |= EV_CLEAR;
182 
183 	return (retval);
184 }
185 
186 static int
187 mevent_kq_fflags(struct mevent *mevp)
188 {
189 	int retval;
190 
191 	retval = 0;
192 
193 	switch (mevp->me_type) {
194 	case EVF_VNODE:
195 		if ((mevp->me_fflags & EVFF_ATTRIB) != 0)
196 			retval |= NOTE_ATTRIB;
197 		break;
198 	case EVF_READ:
199 	case EVF_WRITE:
200 	case EVF_TIMER:
201 	case EVF_SIGNAL:
202 		break;
203 	}
204 
205 	return (retval);
206 }
207 
208 static void
209 mevent_populate(struct mevent *mevp, struct kevent *kev)
210 {
211 	if (mevp->me_type == EVF_TIMER) {
212 		kev->ident = mevp->me_timid;
213 		kev->data = mevp->me_msecs;
214 	} else {
215 		kev->ident = mevp->me_fd;
216 		kev->data = 0;
217 	}
218 	kev->filter = mevent_kq_filter(mevp);
219 	kev->flags = mevent_kq_flags(mevp);
220 	kev->fflags = mevent_kq_fflags(mevp);
221 	kev->udata = mevp;
222 }
223 
224 static int
225 mevent_build(struct kevent *kev)
226 {
227 	struct mevent *mevp, *tmpp;
228 	int i;
229 
230 	i = 0;
231 
232 	mevent_qlock();
233 
234 	LIST_FOREACH_SAFE(mevp, &change_head, me_list, tmpp) {
235 		if (mevp->me_closefd) {
236 			/*
237 			 * A close of the file descriptor will remove the
238 			 * event
239 			 */
240 			close(mevp->me_fd);
241 		} else {
242 			assert((mevp->me_state & EV_ADD) == 0);
243 			mevent_populate(mevp, &kev[i]);
244 			i++;
245 		}
246 
247 		mevp->me_cq = 0;
248 		LIST_REMOVE(mevp, me_list);
249 
250 		if (mevp->me_state & EV_DELETE) {
251 			free(mevp);
252 		} else {
253 			LIST_INSERT_HEAD(&global_head, mevp, me_list);
254 		}
255 
256 		assert(i < MEVENT_MAX);
257 	}
258 
259 	mevent_qunlock();
260 
261 	return (i);
262 }
263 
264 static void
265 mevent_handle(struct kevent *kev, int numev)
266 {
267 	struct mevent *mevp;
268 	int i;
269 
270 	for (i = 0; i < numev; i++) {
271 		mevp = kev[i].udata;
272 
273 		/* XXX check for EV_ERROR ? */
274 
275 		(*mevp->me_func)(mevp->me_fd, mevp->me_type, mevp->me_param);
276 	}
277 }
278 
279 static struct mevent *
280 mevent_add_state(int tfd, enum ev_type type,
281 	   void (*func)(int, enum ev_type, void *), void *param,
282 	   int state, int fflags)
283 {
284 	struct kevent kev;
285 	struct mevent *lp, *mevp;
286 	int ret;
287 
288 	if (tfd < 0 || func == NULL) {
289 		return (NULL);
290 	}
291 
292 	mevp = NULL;
293 
294 	pthread_once(&mevent_once, mevent_init);
295 
296 	mevent_qlock();
297 
298 	/*
299 	 * Verify that the fd/type tuple is not present in any list
300 	 */
301 	LIST_FOREACH(lp, &global_head, me_list) {
302 		if (type != EVF_TIMER && lp->me_fd == tfd &&
303 		    lp->me_type == type) {
304 			goto exit;
305 		}
306 	}
307 
308 	LIST_FOREACH(lp, &change_head, me_list) {
309 		if (type != EVF_TIMER && lp->me_fd == tfd &&
310 		    lp->me_type == type) {
311 			goto exit;
312 		}
313 	}
314 
315 	/*
316 	 * Allocate an entry and populate it.
317 	 */
318 	mevp = calloc(1, sizeof(struct mevent));
319 	if (mevp == NULL) {
320 		goto exit;
321 	}
322 
323 	if (type == EVF_TIMER) {
324 		mevp->me_msecs = tfd;
325 		mevp->me_timid = mevent_timid++;
326 	} else
327 		mevp->me_fd = tfd;
328 	mevp->me_type = type;
329 	mevp->me_func = func;
330 	mevp->me_param = param;
331 	mevp->me_state = state;
332 	mevp->me_fflags = fflags;
333 
334 	/*
335 	 * Try to add the event.  If this fails, report the failure to
336 	 * the caller.
337 	 */
338 	mevent_populate(mevp, &kev);
339 	ret = kevent(mfd, &kev, 1, NULL, 0, NULL);
340 	if (ret == -1) {
341 		free(mevp);
342 		mevp = NULL;
343 		goto exit;
344 	}
345 
346 	mevp->me_state &= ~EV_ADD;
347 	LIST_INSERT_HEAD(&global_head, mevp, me_list);
348 
349 exit:
350 	mevent_qunlock();
351 
352 	return (mevp);
353 }
354 
355 struct mevent *
356 mevent_add(int tfd, enum ev_type type,
357 	   void (*func)(int, enum ev_type, void *), void *param)
358 {
359 
360 	return (mevent_add_state(tfd, type, func, param, EV_ADD, 0));
361 }
362 
363 struct mevent *
364 mevent_add_flags(int tfd, enum ev_type type, int fflags,
365 		 void (*func)(int, enum ev_type, void *), void *param)
366 {
367 
368 	return (mevent_add_state(tfd, type, func, param, EV_ADD, fflags));
369 }
370 
371 struct mevent *
372 mevent_add_disabled(int tfd, enum ev_type type,
373 		    void (*func)(int, enum ev_type, void *), void *param)
374 {
375 
376 	return (mevent_add_state(tfd, type, func, param, EV_ADD | EV_DISABLE, 0));
377 }
378 
379 static int
380 mevent_update(struct mevent *evp, bool enable)
381 {
382 	int newstate;
383 
384 	mevent_qlock();
385 
386 	/*
387 	 * It's not possible to enable/disable a deleted event
388 	 */
389 	assert((evp->me_state & EV_DELETE) == 0);
390 
391 	newstate = evp->me_state;
392 	if (enable) {
393 		newstate |= EV_ENABLE;
394 		newstate &= ~EV_DISABLE;
395 	} else {
396 		newstate |= EV_DISABLE;
397 		newstate &= ~EV_ENABLE;
398 	}
399 
400 	/*
401 	 * No update needed if state isn't changing
402 	 */
403 	if (evp->me_state != newstate) {
404 		evp->me_state = newstate;
405 
406 		/*
407 		 * Place the entry onto the changed list if not
408 		 * already there.
409 		 */
410 		if (evp->me_cq == 0) {
411 			evp->me_cq = 1;
412 			LIST_REMOVE(evp, me_list);
413 			LIST_INSERT_HEAD(&change_head, evp, me_list);
414 			mevent_notify();
415 		}
416 	}
417 
418 	mevent_qunlock();
419 
420 	return (0);
421 }
422 
423 int
424 mevent_enable(struct mevent *evp)
425 {
426 
427 	return (mevent_update(evp, true));
428 }
429 
430 int
431 mevent_disable(struct mevent *evp)
432 {
433 
434 	return (mevent_update(evp, false));
435 }
436 
437 static int
438 mevent_delete_event(struct mevent *evp, int closefd)
439 {
440 	mevent_qlock();
441 
442 	/*
443          * Place the entry onto the changed list if not already there, and
444 	 * mark as to be deleted.
445          */
446         if (evp->me_cq == 0) {
447 		evp->me_cq = 1;
448 		LIST_REMOVE(evp, me_list);
449 		LIST_INSERT_HEAD(&change_head, evp, me_list);
450 		mevent_notify();
451         }
452 	evp->me_state = EV_DELETE;
453 
454 	if (closefd)
455 		evp->me_closefd = 1;
456 
457 	mevent_qunlock();
458 
459 	return (0);
460 }
461 
462 int
463 mevent_delete(struct mevent *evp)
464 {
465 
466 	return (mevent_delete_event(evp, 0));
467 }
468 
469 int
470 mevent_delete_close(struct mevent *evp)
471 {
472 
473 	return (mevent_delete_event(evp, 1));
474 }
475 
476 static void
477 mevent_set_name(void)
478 {
479 
480 	pthread_set_name_np(mevent_tid, "mevent");
481 }
482 
483 void
484 mevent_dispatch(void)
485 {
486 	struct kevent changelist[MEVENT_MAX];
487 	struct kevent eventlist[MEVENT_MAX];
488 	struct mevent *pipev;
489 	int numev;
490 	int ret;
491 #ifndef WITHOUT_CAPSICUM
492 	cap_rights_t rights;
493 #endif
494 
495 	mevent_tid = pthread_self();
496 	mevent_set_name();
497 
498 	pthread_once(&mevent_once, mevent_init);
499 
500 	/*
501 	 * Open the pipe that will be used for other threads to force
502 	 * the blocking kqueue call to exit by writing to it. Set the
503 	 * descriptor to non-blocking.
504 	 */
505 	ret = pipe(mevent_pipefd);
506 	if (ret < 0) {
507 		perror("pipe");
508 		exit(0);
509 	}
510 
511 #ifndef WITHOUT_CAPSICUM
512 	cap_rights_init(&rights, CAP_EVENT, CAP_READ, CAP_WRITE);
513 	if (caph_rights_limit(mevent_pipefd[0], &rights) == -1)
514 		errx(EX_OSERR, "Unable to apply rights for sandbox");
515 	if (caph_rights_limit(mevent_pipefd[1], &rights) == -1)
516 		errx(EX_OSERR, "Unable to apply rights for sandbox");
517 #endif
518 
519 	/*
520 	 * Add internal event handler for the pipe write fd
521 	 */
522 	pipev = mevent_add(mevent_pipefd[0], EVF_READ, mevent_pipe_read, NULL);
523 	assert(pipev != NULL);
524 
525 	for (;;) {
526 		/*
527 		 * Build changelist if required.
528 		 * XXX the changelist can be put into the blocking call
529 		 * to eliminate the extra syscall. Currently better for
530 		 * debug.
531 		 */
532 		numev = mevent_build(changelist);
533 		if (numev) {
534 			ret = kevent(mfd, changelist, numev, NULL, 0, NULL);
535 			if (ret == -1) {
536 				perror("Error return from kevent change");
537 			}
538 		}
539 
540 		/*
541 		 * Block awaiting events
542 		 */
543 		ret = kevent(mfd, NULL, 0, eventlist, MEVENT_MAX, NULL);
544 		if (ret == -1 && errno != EINTR) {
545 			perror("Error return from kevent monitor");
546 		}
547 
548 		/*
549 		 * Handle reported events
550 		 */
551 		mevent_handle(eventlist, ret);
552 	}
553 }
554