xref: /freebsd/usr.sbin/bhyve/mevent.c (revision 48c779cdecb5f803e5fe5d761987e976ca9609db) 
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 <stdlib.h>
46 #include <stdio.h>
47 #include <string.h>
48 #include <sysexits.h>
49 #include <unistd.h>
50 
51 #include <sys/types.h>
52 #ifndef WITHOUT_CAPSICUM
53 #include <sys/capsicum.h>
54 #endif
55 #include <sys/event.h>
56 #include <sys/time.h>
57 
58 #include <pthread.h>
59 #include <pthread_np.h>
60 
61 #include "mevent.h"
62 
63 #define	MEVENT_MAX	64
64 
65 #define	MEV_ADD			1
66 #define	MEV_ENABLE		2
67 #define	MEV_DISABLE		3
68 #define	MEV_DEL_PENDING		4
69 #define	MEV_ADD_DISABLED	5
70 
71 extern char *vmname;
72 
73 static pthread_t mevent_tid;
74 static int mevent_timid = 43;
75 static int mevent_pipefd[2];
76 static pthread_mutex_t mevent_lmutex = PTHREAD_MUTEX_INITIALIZER;
77 
78 struct mevent {
79 	void	(*me_func)(int, enum ev_type, void *);
80 #define me_msecs me_fd
81 	int	me_fd;
82 	int	me_timid;
83 	enum ev_type me_type;
84 	void    *me_param;
85 	int	me_cq;
86 	int	me_state;
87 	int	me_closefd;
88 	LIST_ENTRY(mevent) me_list;
89 };
90 
91 static LIST_HEAD(listhead, mevent) global_head, change_head;
92 
93 static void
94 mevent_qlock(void)
95 {
96 	pthread_mutex_lock(&mevent_lmutex);
97 }
98 
99 static void
100 mevent_qunlock(void)
101 {
102 	pthread_mutex_unlock(&mevent_lmutex);
103 }
104 
105 static void
106 mevent_pipe_read(int fd, enum ev_type type, void *param)
107 {
108 	char buf[MEVENT_MAX];
109 	int status;
110 
111 	/*
112 	 * Drain the pipe read side. The fd is non-blocking so this is
113 	 * safe to do.
114 	 */
115 	do {
116 		status = read(fd, buf, sizeof(buf));
117 	} while (status == MEVENT_MAX);
118 }
119 
120 static void
121 mevent_notify(void)
122 {
123 	char c = '\0';
124 
125 	/*
126 	 * If calling from outside the i/o thread, write a byte on the
127 	 * pipe to force the i/o thread to exit the blocking kevent call.
128 	 */
129 	if (mevent_pipefd[1] != 0 && pthread_self() != mevent_tid) {
130 		write(mevent_pipefd[1], &c, 1);
131 	}
132 }
133 
134 static int
135 mevent_kq_filter(struct mevent *mevp)
136 {
137 	int retval;
138 
139 	retval = 0;
140 
141 	if (mevp->me_type == EVF_READ)
142 		retval = EVFILT_READ;
143 
144 	if (mevp->me_type == EVF_WRITE)
145 		retval = EVFILT_WRITE;
146 
147 	if (mevp->me_type == EVF_TIMER)
148 		retval = EVFILT_TIMER;
149 
150 	if (mevp->me_type == EVF_SIGNAL)
151 		retval = EVFILT_SIGNAL;
152 
153 	return (retval);
154 }
155 
156 static int
157 mevent_kq_flags(struct mevent *mevp)
158 {
159 	int ret;
160 
161 	switch (mevp->me_state) {
162 	case MEV_ADD:
163 		ret = EV_ADD;		/* implicitly enabled */
164 		break;
165 	case MEV_ADD_DISABLED:
166 		ret = EV_ADD | EV_DISABLE;
167 		break;
168 	case MEV_ENABLE:
169 		ret = EV_ENABLE;
170 		break;
171 	case MEV_DISABLE:
172 		ret = EV_DISABLE;
173 		break;
174 	case MEV_DEL_PENDING:
175 		ret = EV_DELETE;
176 		break;
177 	default:
178 		assert(0);
179 		break;
180 	}
181 
182 	return (ret);
183 }
184 
185 static int
186 mevent_kq_fflags(struct mevent *mevp)
187 {
188 	/* XXX nothing yet, perhaps EV_EOF for reads ? */
189 	return (0);
190 }
191 
192 static int
193 mevent_build(int mfd, struct kevent *kev)
194 {
195 	struct mevent *mevp, *tmpp;
196 	int i;
197 
198 	i = 0;
199 
200 	mevent_qlock();
201 
202 	LIST_FOREACH_SAFE(mevp, &change_head, me_list, tmpp) {
203 		if (mevp->me_closefd) {
204 			/*
205 			 * A close of the file descriptor will remove the
206 			 * event
207 			 */
208 			close(mevp->me_fd);
209 		} else {
210 			if (mevp->me_type == EVF_TIMER) {
211 				kev[i].ident = mevp->me_timid;
212 				kev[i].data = mevp->me_msecs;
213 			} else {
214 				kev[i].ident = mevp->me_fd;
215 				kev[i].data = 0;
216 			}
217 			kev[i].filter = mevent_kq_filter(mevp);
218 			kev[i].flags = mevent_kq_flags(mevp);
219 			kev[i].fflags = mevent_kq_fflags(mevp);
220 			kev[i].udata = mevp;
221 			i++;
222 		}
223 
224 		mevp->me_cq = 0;
225 		LIST_REMOVE(mevp, me_list);
226 
227 		if (mevp->me_state == MEV_DEL_PENDING) {
228 			free(mevp);
229 		} else {
230 			LIST_INSERT_HEAD(&global_head, mevp, me_list);
231 		}
232 
233 		assert(i < MEVENT_MAX);
234 	}
235 
236 	mevent_qunlock();
237 
238 	return (i);
239 }
240 
241 static void
242 mevent_handle(struct kevent *kev, int numev)
243 {
244 	struct mevent *mevp;
245 	int i;
246 
247 	for (i = 0; i < numev; i++) {
248 		mevp = kev[i].udata;
249 
250 		/* XXX check for EV_ERROR ? */
251 
252 		(*mevp->me_func)(mevp->me_fd, mevp->me_type, mevp->me_param);
253 	}
254 }
255 
256 static struct mevent *
257 mevent_add_state(int tfd, enum ev_type type,
258 	   void (*func)(int, enum ev_type, void *), void *param,
259 	   int state)
260 {
261 	struct mevent *lp, *mevp;
262 
263 	if (tfd < 0 || func == NULL) {
264 		return (NULL);
265 	}
266 
267 	mevp = NULL;
268 
269 	mevent_qlock();
270 
271 	/*
272 	 * Verify that the fd/type tuple is not present in any list
273 	 */
274 	LIST_FOREACH(lp, &global_head, me_list) {
275 		if (type != EVF_TIMER && lp->me_fd == tfd &&
276 		    lp->me_type == type) {
277 			goto exit;
278 		}
279 	}
280 
281 	LIST_FOREACH(lp, &change_head, me_list) {
282 		if (type != EVF_TIMER && lp->me_fd == tfd &&
283 		    lp->me_type == type) {
284 			goto exit;
285 		}
286 	}
287 
288 	/*
289 	 * Allocate an entry, populate it, and add it to the change list.
290 	 */
291 	mevp = calloc(1, sizeof(struct mevent));
292 	if (mevp == NULL) {
293 		goto exit;
294 	}
295 
296 	if (type == EVF_TIMER) {
297 		mevp->me_msecs = tfd;
298 		mevp->me_timid = mevent_timid++;
299 	} else
300 		mevp->me_fd = tfd;
301 	mevp->me_type = type;
302 	mevp->me_func = func;
303 	mevp->me_param = param;
304 
305 	LIST_INSERT_HEAD(&change_head, mevp, me_list);
306 	mevp->me_cq = 1;
307 	mevp->me_state = state;
308 	mevent_notify();
309 
310 exit:
311 	mevent_qunlock();
312 
313 	return (mevp);
314 }
315 
316 struct mevent *
317 mevent_add(int tfd, enum ev_type type,
318 	   void (*func)(int, enum ev_type, void *), void *param)
319 {
320 
321 	return mevent_add_state(tfd, type, func, param, MEV_ADD);
322 }
323 
324 struct mevent *
325 mevent_add_disabled(int tfd, enum ev_type type,
326 		    void (*func)(int, enum ev_type, void *), void *param)
327 {
328 
329 	return mevent_add_state(tfd, type, func, param, MEV_ADD_DISABLED);
330 }
331 
332 static int
333 mevent_update(struct mevent *evp, int newstate)
334 {
335 	/*
336 	 * It's not possible to enable/disable a deleted event
337 	 */
338 	if (evp->me_state == MEV_DEL_PENDING)
339 		return (EINVAL);
340 
341 	/*
342 	 * No update needed if state isn't changing
343 	 */
344 	if (evp->me_state == newstate)
345 		return (0);
346 
347 	mevent_qlock();
348 
349 	evp->me_state = newstate;
350 
351 	/*
352 	 * Place the entry onto the changed list if not already there.
353 	 */
354 	if (evp->me_cq == 0) {
355 		evp->me_cq = 1;
356 		LIST_REMOVE(evp, me_list);
357 		LIST_INSERT_HEAD(&change_head, evp, me_list);
358 		mevent_notify();
359 	}
360 
361 	mevent_qunlock();
362 
363 	return (0);
364 }
365 
366 int
367 mevent_enable(struct mevent *evp)
368 {
369 
370 	return (mevent_update(evp, MEV_ENABLE));
371 }
372 
373 int
374 mevent_disable(struct mevent *evp)
375 {
376 
377 	return (mevent_update(evp, MEV_DISABLE));
378 }
379 
380 static int
381 mevent_delete_event(struct mevent *evp, int closefd)
382 {
383 	mevent_qlock();
384 
385 	/*
386          * Place the entry onto the changed list if not already there, and
387 	 * mark as to be deleted.
388          */
389         if (evp->me_cq == 0) {
390 		evp->me_cq = 1;
391 		LIST_REMOVE(evp, me_list);
392 		LIST_INSERT_HEAD(&change_head, evp, me_list);
393 		mevent_notify();
394         }
395 	evp->me_state = MEV_DEL_PENDING;
396 
397 	if (closefd)
398 		evp->me_closefd = 1;
399 
400 	mevent_qunlock();
401 
402 	return (0);
403 }
404 
405 int
406 mevent_delete(struct mevent *evp)
407 {
408 
409 	return (mevent_delete_event(evp, 0));
410 }
411 
412 int
413 mevent_delete_close(struct mevent *evp)
414 {
415 
416 	return (mevent_delete_event(evp, 1));
417 }
418 
419 static void
420 mevent_set_name(void)
421 {
422 
423 	pthread_set_name_np(mevent_tid, "mevent");
424 }
425 
426 void
427 mevent_dispatch(void)
428 {
429 	struct kevent changelist[MEVENT_MAX];
430 	struct kevent eventlist[MEVENT_MAX];
431 	struct mevent *pipev;
432 	int mfd;
433 	int numev;
434 	int ret;
435 #ifndef WITHOUT_CAPSICUM
436 	cap_rights_t rights;
437 #endif
438 
439 	mevent_tid = pthread_self();
440 	mevent_set_name();
441 
442 	mfd = kqueue();
443 	assert(mfd > 0);
444 
445 #ifndef WITHOUT_CAPSICUM
446 	cap_rights_init(&rights, CAP_KQUEUE);
447 	if (caph_rights_limit(mfd, &rights) == -1)
448 		errx(EX_OSERR, "Unable to apply rights for sandbox");
449 #endif
450 
451 	/*
452 	 * Open the pipe that will be used for other threads to force
453 	 * the blocking kqueue call to exit by writing to it. Set the
454 	 * descriptor to non-blocking.
455 	 */
456 	ret = pipe(mevent_pipefd);
457 	if (ret < 0) {
458 		perror("pipe");
459 		exit(0);
460 	}
461 
462 #ifndef WITHOUT_CAPSICUM
463 	cap_rights_init(&rights, CAP_EVENT, CAP_READ, CAP_WRITE);
464 	if (caph_rights_limit(mevent_pipefd[0], &rights) == -1)
465 		errx(EX_OSERR, "Unable to apply rights for sandbox");
466 	if (caph_rights_limit(mevent_pipefd[1], &rights) == -1)
467 		errx(EX_OSERR, "Unable to apply rights for sandbox");
468 #endif
469 
470 	/*
471 	 * Add internal event handler for the pipe write fd
472 	 */
473 	pipev = mevent_add(mevent_pipefd[0], EVF_READ, mevent_pipe_read, NULL);
474 	assert(pipev != NULL);
475 
476 	for (;;) {
477 		/*
478 		 * Build changelist if required.
479 		 * XXX the changelist can be put into the blocking call
480 		 * to eliminate the extra syscall. Currently better for
481 		 * debug.
482 		 */
483 		numev = mevent_build(mfd, changelist);
484 		if (numev) {
485 			ret = kevent(mfd, changelist, numev, NULL, 0, NULL);
486 			if (ret == -1) {
487 				perror("Error return from kevent change");
488 			}
489 		}
490 
491 		/*
492 		 * Block awaiting events
493 		 */
494 		ret = kevent(mfd, NULL, 0, eventlist, MEVENT_MAX, NULL);
495 		if (ret == -1 && errno != EINTR) {
496 			perror("Error return from kevent monitor");
497 		}
498 
499 		/*
500 		 * Handle reported events
501 		 */
502 		mevent_handle(eventlist, ret);
503 	}
504 }
505