xref: /freebsd/usr.sbin/bhyve/mevent.c (revision dbd5678d) 
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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  * $FreeBSD$
29  */
30 
31 /*
32  * Micro event library for FreeBSD, designed for a single i/o thread
33  * using kqueue, and having events be persistent by default.
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <assert.h>
40 #ifndef WITHOUT_CAPSICUM
41 #include <capsicum_helpers.h>
42 #endif
43 #include <err.h>
44 #include <errno.h>
45 #include <stdbool.h>
46 #include <stdlib.h>
47 #include <stdio.h>
48 #include <string.h>
49 #include <sysexits.h>
50 #include <unistd.h>
51 
52 #include <sys/types.h>
53 #ifndef WITHOUT_CAPSICUM
54 #include <sys/capsicum.h>
55 #endif
56 #include <sys/event.h>
57 #include <sys/time.h>
58 
59 #include <pthread.h>
60 #include <pthread_np.h>
61 
62 #include "mevent.h"
63 
64 #define	MEVENT_MAX	64
65 
66 static pthread_t mevent_tid;
67 static pthread_once_t mevent_once = PTHREAD_ONCE_INIT;
68 static int mevent_timid = 43;
69 static int mevent_pipefd[2];
70 static int mfd;
71 static pthread_mutex_t mevent_lmutex = PTHREAD_MUTEX_INITIALIZER;
72 
73 struct mevent {
74 	void	(*me_func)(int, enum ev_type, void *);
75 #define me_msecs me_fd
76 	int	me_fd;
77 	int	me_timid;
78 	enum ev_type me_type;
79 	void    *me_param;
80 	int	me_cq;
81 	int	me_state; /* Desired kevent flags. */
82 	int	me_closefd;
83 	int	me_fflags;
84 	LIST_ENTRY(mevent) me_list;
85 };
86 
87 static LIST_HEAD(listhead, mevent) global_head, change_head;
88 
89 static void
90 mevent_qlock(void)
91 {
92 	pthread_mutex_lock(&mevent_lmutex);
93 }
94 
95 static void
96 mevent_qunlock(void)
97 {
98 	pthread_mutex_unlock(&mevent_lmutex);
99 }
100 
101 static void
102 mevent_pipe_read(int fd, enum ev_type type __unused, void *param __unused)
103 {
104 	char buf[MEVENT_MAX];
105 	int status;
106 
107 	/*
108 	 * Drain the pipe read side. The fd is non-blocking so this is
109 	 * safe to do.
110 	 */
111 	do {
112 		status = read(fd, buf, sizeof(buf));
113 	} while (status == MEVENT_MAX);
114 }
115 
116 static void
117 mevent_notify(void)
118 {
119 	char c = '\0';
120 
121 	/*
122 	 * If calling from outside the i/o thread, write a byte on the
123 	 * pipe to force the i/o thread to exit the blocking kevent call.
124 	 */
125 	if (mevent_pipefd[1] != 0 && pthread_self() != mevent_tid) {
126 		write(mevent_pipefd[1], &c, 1);
127 	}
128 }
129 
130 static void
131 mevent_init(void)
132 {
133 #ifndef WITHOUT_CAPSICUM
134 	cap_rights_t rights;
135 #endif
136 
137 	mfd = kqueue();
138 	assert(mfd > 0);
139 
140 #ifndef WITHOUT_CAPSICUM
141 	cap_rights_init(&rights, CAP_KQUEUE);
142 	if (caph_rights_limit(mfd, &rights) == -1)
143 		errx(EX_OSERR, "Unable to apply rights for sandbox");
144 #endif
145 
146 	LIST_INIT(&change_head);
147 	LIST_INIT(&global_head);
148 }
149 
150 static int
151 mevent_kq_filter(struct mevent *mevp)
152 {
153 	int retval;
154 
155 	retval = 0;
156 
157 	if (mevp->me_type == EVF_READ)
158 		retval = EVFILT_READ;
159 
160 	if (mevp->me_type == EVF_WRITE)
161 		retval = EVFILT_WRITE;
162 
163 	if (mevp->me_type == EVF_TIMER)
164 		retval = EVFILT_TIMER;
165 
166 	if (mevp->me_type == EVF_SIGNAL)
167 		retval = EVFILT_SIGNAL;
168 
169 	if (mevp->me_type == EVF_VNODE)
170 		retval = EVFILT_VNODE;
171 
172 	return (retval);
173 }
174 
175 static int
176 mevent_kq_flags(struct mevent *mevp)
177 {
178 	int retval;
179 
180 	retval = mevp->me_state;
181 
182 	if (mevp->me_type == EVF_VNODE)
183 		retval |= EV_CLEAR;
184 
185 	return (retval);
186 }
187 
188 static int
189 mevent_kq_fflags(struct mevent *mevp)
190 {
191 	int retval;
192 
193 	retval = 0;
194 
195 	switch (mevp->me_type) {
196 	case EVF_VNODE:
197 		if ((mevp->me_fflags & EVFF_ATTRIB) != 0)
198 			retval |= NOTE_ATTRIB;
199 		break;
200 	case EVF_READ:
201 	case EVF_WRITE:
202 	case EVF_TIMER:
203 	case EVF_SIGNAL:
204 		break;
205 	}
206 
207 	return (retval);
208 }
209 
210 static void
211 mevent_populate(struct mevent *mevp, struct kevent *kev)
212 {
213 	if (mevp->me_type == EVF_TIMER) {
214 		kev->ident = mevp->me_timid;
215 		kev->data = mevp->me_msecs;
216 	} else {
217 		kev->ident = mevp->me_fd;
218 		kev->data = 0;
219 	}
220 	kev->filter = mevent_kq_filter(mevp);
221 	kev->flags = mevent_kq_flags(mevp);
222 	kev->fflags = mevent_kq_fflags(mevp);
223 	kev->udata = mevp;
224 }
225 
226 static int
227 mevent_build(struct kevent *kev)
228 {
229 	struct mevent *mevp, *tmpp;
230 	int i;
231 
232 	i = 0;
233 
234 	mevent_qlock();
235 
236 	LIST_FOREACH_SAFE(mevp, &change_head, me_list, tmpp) {
237 		if (mevp->me_closefd) {
238 			/*
239 			 * A close of the file descriptor will remove the
240 			 * event
241 			 */
242 			close(mevp->me_fd);
243 		} else {
244 			assert((mevp->me_state & EV_ADD) == 0);
245 			mevent_populate(mevp, &kev[i]);
246 			i++;
247 		}
248 
249 		mevp->me_cq = 0;
250 		LIST_REMOVE(mevp, me_list);
251 
252 		if (mevp->me_state & EV_DELETE) {
253 			free(mevp);
254 		} else {
255 			LIST_INSERT_HEAD(&global_head, mevp, me_list);
256 		}
257 
258 		assert(i < MEVENT_MAX);
259 	}
260 
261 	mevent_qunlock();
262 
263 	return (i);
264 }
265 
266 static void
267 mevent_handle(struct kevent *kev, int numev)
268 {
269 	struct mevent *mevp;
270 	int i;
271 
272 	for (i = 0; i < numev; i++) {
273 		mevp = kev[i].udata;
274 
275 		/* XXX check for EV_ERROR ? */
276 
277 		(*mevp->me_func)(mevp->me_fd, mevp->me_type, mevp->me_param);
278 	}
279 }
280 
281 static struct mevent *
282 mevent_add_state(int tfd, enum ev_type type,
283 	   void (*func)(int, enum ev_type, void *), void *param,
284 	   int state, int fflags)
285 {
286 	struct kevent kev;
287 	struct mevent *lp, *mevp;
288 	int ret;
289 
290 	if (tfd < 0 || func == NULL) {
291 		return (NULL);
292 	}
293 
294 	mevp = NULL;
295 
296 	pthread_once(&mevent_once, mevent_init);
297 
298 	mevent_qlock();
299 
300 	/*
301 	 * Verify that the fd/type tuple is not present in any list
302 	 */
303 	LIST_FOREACH(lp, &global_head, me_list) {
304 		if (type != EVF_TIMER && lp->me_fd == tfd &&
305 		    lp->me_type == type) {
306 			goto exit;
307 		}
308 	}
309 
310 	LIST_FOREACH(lp, &change_head, me_list) {
311 		if (type != EVF_TIMER && lp->me_fd == tfd &&
312 		    lp->me_type == type) {
313 			goto exit;
314 		}
315 	}
316 
317 	/*
318 	 * Allocate an entry and populate it.
319 	 */
320 	mevp = calloc(1, sizeof(struct mevent));
321 	if (mevp == NULL) {
322 		goto exit;
323 	}
324 
325 	if (type == EVF_TIMER) {
326 		mevp->me_msecs = tfd;
327 		mevp->me_timid = mevent_timid++;
328 	} else
329 		mevp->me_fd = tfd;
330 	mevp->me_type = type;
331 	mevp->me_func = func;
332 	mevp->me_param = param;
333 	mevp->me_state = state;
334 	mevp->me_fflags = fflags;
335 
336 	/*
337 	 * Try to add the event.  If this fails, report the failure to
338 	 * the caller.
339 	 */
340 	mevent_populate(mevp, &kev);
341 	ret = kevent(mfd, &kev, 1, NULL, 0, NULL);
342 	if (ret == -1) {
343 		free(mevp);
344 		mevp = NULL;
345 		goto exit;
346 	}
347 
348 	mevp->me_state &= ~EV_ADD;
349 	LIST_INSERT_HEAD(&global_head, mevp, me_list);
350 
351 exit:
352 	mevent_qunlock();
353 
354 	return (mevp);
355 }
356 
357 struct mevent *
358 mevent_add(int tfd, enum ev_type type,
359 	   void (*func)(int, enum ev_type, void *), void *param)
360 {
361 
362 	return (mevent_add_state(tfd, type, func, param, EV_ADD, 0));
363 }
364 
365 struct mevent *
366 mevent_add_flags(int tfd, enum ev_type type, int fflags,
367 		 void (*func)(int, enum ev_type, void *), void *param)
368 {
369 
370 	return (mevent_add_state(tfd, type, func, param, EV_ADD, fflags));
371 }
372 
373 struct mevent *
374 mevent_add_disabled(int tfd, enum ev_type type,
375 		    void (*func)(int, enum ev_type, void *), void *param)
376 {
377 
378 	return (mevent_add_state(tfd, type, func, param, EV_ADD | EV_DISABLE, 0));
379 }
380 
381 static int
382 mevent_update(struct mevent *evp, bool enable)
383 {
384 	int newstate;
385 
386 	mevent_qlock();
387 
388 	/*
389 	 * It's not possible to enable/disable a deleted event
390 	 */
391 	assert((evp->me_state & EV_DELETE) == 0);
392 
393 	newstate = evp->me_state;
394 	if (enable) {
395 		newstate |= EV_ENABLE;
396 		newstate &= ~EV_DISABLE;
397 	} else {
398 		newstate |= EV_DISABLE;
399 		newstate &= ~EV_ENABLE;
400 	}
401 
402 	/*
403 	 * No update needed if state isn't changing
404 	 */
405 	if (evp->me_state != newstate) {
406 		evp->me_state = newstate;
407 
408 		/*
409 		 * Place the entry onto the changed list if not
410 		 * already there.
411 		 */
412 		if (evp->me_cq == 0) {
413 			evp->me_cq = 1;
414 			LIST_REMOVE(evp, me_list);
415 			LIST_INSERT_HEAD(&change_head, evp, me_list);
416 			mevent_notify();
417 		}
418 	}
419 
420 	mevent_qunlock();
421 
422 	return (0);
423 }
424 
425 int
426 mevent_enable(struct mevent *evp)
427 {
428 
429 	return (mevent_update(evp, true));
430 }
431 
432 int
433 mevent_disable(struct mevent *evp)
434 {
435 
436 	return (mevent_update(evp, false));
437 }
438 
439 static int
440 mevent_delete_event(struct mevent *evp, int closefd)
441 {
442 	mevent_qlock();
443 
444 	/*
445          * Place the entry onto the changed list if not already there, and
446 	 * mark as to be deleted.
447          */
448         if (evp->me_cq == 0) {
449 		evp->me_cq = 1;
450 		LIST_REMOVE(evp, me_list);
451 		LIST_INSERT_HEAD(&change_head, evp, me_list);
452 		mevent_notify();
453         }
454 	evp->me_state = EV_DELETE;
455 
456 	if (closefd)
457 		evp->me_closefd = 1;
458 
459 	mevent_qunlock();
460 
461 	return (0);
462 }
463 
464 int
465 mevent_delete(struct mevent *evp)
466 {
467 
468 	return (mevent_delete_event(evp, 0));
469 }
470 
471 int
472 mevent_delete_close(struct mevent *evp)
473 {
474 
475 	return (mevent_delete_event(evp, 1));
476 }
477 
478 static void
479 mevent_set_name(void)
480 {
481 
482 	pthread_set_name_np(mevent_tid, "mevent");
483 }
484 
485 void
486 mevent_dispatch(void)
487 {
488 	struct kevent changelist[MEVENT_MAX];
489 	struct kevent eventlist[MEVENT_MAX];
490 	struct mevent *pipev;
491 	int numev;
492 	int ret;
493 #ifndef WITHOUT_CAPSICUM
494 	cap_rights_t rights;
495 #endif
496 
497 	mevent_tid = pthread_self();
498 	mevent_set_name();
499 
500 	pthread_once(&mevent_once, mevent_init);
501 
502 	/*
503 	 * Open the pipe that will be used for other threads to force
504 	 * the blocking kqueue call to exit by writing to it. Set the
505 	 * descriptor to non-blocking.
506 	 */
507 	ret = pipe(mevent_pipefd);
508 	if (ret < 0) {
509 		perror("pipe");
510 		exit(0);
511 	}
512 
513 #ifndef WITHOUT_CAPSICUM
514 	cap_rights_init(&rights, CAP_EVENT, CAP_READ, CAP_WRITE);
515 	if (caph_rights_limit(mevent_pipefd[0], &rights) == -1)
516 		errx(EX_OSERR, "Unable to apply rights for sandbox");
517 	if (caph_rights_limit(mevent_pipefd[1], &rights) == -1)
518 		errx(EX_OSERR, "Unable to apply rights for sandbox");
519 #endif
520 
521 	/*
522 	 * Add internal event handler for the pipe write fd
523 	 */
524 	pipev = mevent_add(mevent_pipefd[0], EVF_READ, mevent_pipe_read, NULL);
525 	assert(pipev != NULL);
526 
527 	for (;;) {
528 		/*
529 		 * Build changelist if required.
530 		 * XXX the changelist can be put into the blocking call
531 		 * to eliminate the extra syscall. Currently better for
532 		 * debug.
533 		 */
534 		numev = mevent_build(changelist);
535 		if (numev) {
536 			ret = kevent(mfd, changelist, numev, NULL, 0, NULL);
537 			if (ret == -1) {
538 				perror("Error return from kevent change");
539 			}
540 		}
541 
542 		/*
543 		 * Block awaiting events
544 		 */
545 		ret = kevent(mfd, NULL, 0, eventlist, MEVENT_MAX, NULL);
546 		if (ret == -1 && errno != EINTR) {
547 			perror("Error return from kevent monitor");
548 		}
549 
550 		/*
551 		 * Handle reported events
552 		 */
553 		mevent_handle(eventlist, ret);
554 	}
555 }
556