xref: /freebsd/usr.sbin/bhyve/mevent.c (revision ddc0daea20280c3a06a910b72b14ffe3f624df71) 
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 
70 extern char *vmname;
71 
72 static pthread_t mevent_tid;
73 static int mevent_timid = 43;
74 static int mevent_pipefd[2];
75 static pthread_mutex_t mevent_lmutex = PTHREAD_MUTEX_INITIALIZER;
76 
77 struct mevent {
78 	void	(*me_func)(int, enum ev_type, void *);
79 #define me_msecs me_fd
80 	int	me_fd;
81 	int	me_timid;
82 	enum ev_type me_type;
83 	void    *me_param;
84 	int	me_cq;
85 	int	me_state;
86 	int	me_closefd;
87 	LIST_ENTRY(mevent) me_list;
88 };
89 
90 static LIST_HEAD(listhead, mevent) global_head, change_head;
91 
92 static void
93 mevent_qlock(void)
94 {
95 	pthread_mutex_lock(&mevent_lmutex);
96 }
97 
98 static void
99 mevent_qunlock(void)
100 {
101 	pthread_mutex_unlock(&mevent_lmutex);
102 }
103 
104 static void
105 mevent_pipe_read(int fd, enum ev_type type, void *param)
106 {
107 	char buf[MEVENT_MAX];
108 	int status;
109 
110 	/*
111 	 * Drain the pipe read side. The fd is non-blocking so this is
112 	 * safe to do.
113 	 */
114 	do {
115 		status = read(fd, buf, sizeof(buf));
116 	} while (status == MEVENT_MAX);
117 }
118 
119 static void
120 mevent_notify(void)
121 {
122 	char c = '\0';
123 
124 	/*
125 	 * If calling from outside the i/o thread, write a byte on the
126 	 * pipe to force the i/o thread to exit the blocking kevent call.
127 	 */
128 	if (mevent_pipefd[1] != 0 && pthread_self() != mevent_tid) {
129 		write(mevent_pipefd[1], &c, 1);
130 	}
131 }
132 
133 static int
134 mevent_kq_filter(struct mevent *mevp)
135 {
136 	int retval;
137 
138 	retval = 0;
139 
140 	if (mevp->me_type == EVF_READ)
141 		retval = EVFILT_READ;
142 
143 	if (mevp->me_type == EVF_WRITE)
144 		retval = EVFILT_WRITE;
145 
146 	if (mevp->me_type == EVF_TIMER)
147 		retval = EVFILT_TIMER;
148 
149 	if (mevp->me_type == EVF_SIGNAL)
150 		retval = EVFILT_SIGNAL;
151 
152 	return (retval);
153 }
154 
155 static int
156 mevent_kq_flags(struct mevent *mevp)
157 {
158 	int ret;
159 
160 	switch (mevp->me_state) {
161 	case MEV_ADD:
162 		ret = EV_ADD;		/* implicitly enabled */
163 		break;
164 	case MEV_ENABLE:
165 		ret = EV_ENABLE;
166 		break;
167 	case MEV_DISABLE:
168 		ret = EV_DISABLE;
169 		break;
170 	case MEV_DEL_PENDING:
171 		ret = EV_DELETE;
172 		break;
173 	default:
174 		assert(0);
175 		break;
176 	}
177 
178 	return (ret);
179 }
180 
181 static int
182 mevent_kq_fflags(struct mevent *mevp)
183 {
184 	/* XXX nothing yet, perhaps EV_EOF for reads ? */
185 	return (0);
186 }
187 
188 static int
189 mevent_build(int mfd, struct kevent *kev)
190 {
191 	struct mevent *mevp, *tmpp;
192 	int i;
193 
194 	i = 0;
195 
196 	mevent_qlock();
197 
198 	LIST_FOREACH_SAFE(mevp, &change_head, me_list, tmpp) {
199 		if (mevp->me_closefd) {
200 			/*
201 			 * A close of the file descriptor will remove the
202 			 * event
203 			 */
204 			close(mevp->me_fd);
205 		} else {
206 			if (mevp->me_type == EVF_TIMER) {
207 				kev[i].ident = mevp->me_timid;
208 				kev[i].data = mevp->me_msecs;
209 			} else {
210 				kev[i].ident = mevp->me_fd;
211 				kev[i].data = 0;
212 			}
213 			kev[i].filter = mevent_kq_filter(mevp);
214 			kev[i].flags = mevent_kq_flags(mevp);
215 			kev[i].fflags = mevent_kq_fflags(mevp);
216 			kev[i].udata = mevp;
217 			i++;
218 		}
219 
220 		mevp->me_cq = 0;
221 		LIST_REMOVE(mevp, me_list);
222 
223 		if (mevp->me_state == MEV_DEL_PENDING) {
224 			free(mevp);
225 		} else {
226 			LIST_INSERT_HEAD(&global_head, mevp, me_list);
227 		}
228 
229 		assert(i < MEVENT_MAX);
230 	}
231 
232 	mevent_qunlock();
233 
234 	return (i);
235 }
236 
237 static void
238 mevent_handle(struct kevent *kev, int numev)
239 {
240 	struct mevent *mevp;
241 	int i;
242 
243 	for (i = 0; i < numev; i++) {
244 		mevp = kev[i].udata;
245 
246 		/* XXX check for EV_ERROR ? */
247 
248 		(*mevp->me_func)(mevp->me_fd, mevp->me_type, mevp->me_param);
249 	}
250 }
251 
252 struct mevent *
253 mevent_add(int tfd, enum ev_type type,
254 	   void (*func)(int, enum ev_type, void *), void *param)
255 {
256 	struct mevent *lp, *mevp;
257 
258 	if (tfd < 0 || func == NULL) {
259 		return (NULL);
260 	}
261 
262 	mevp = NULL;
263 
264 	mevent_qlock();
265 
266 	/*
267 	 * Verify that the fd/type tuple is not present in any list
268 	 */
269 	LIST_FOREACH(lp, &global_head, me_list) {
270 		if (type != EVF_TIMER && lp->me_fd == tfd &&
271 		    lp->me_type == type) {
272 			goto exit;
273 		}
274 	}
275 
276 	LIST_FOREACH(lp, &change_head, me_list) {
277 		if (type != EVF_TIMER && lp->me_fd == tfd &&
278 		    lp->me_type == type) {
279 			goto exit;
280 		}
281 	}
282 
283 	/*
284 	 * Allocate an entry, populate it, and add it to the change list.
285 	 */
286 	mevp = calloc(1, sizeof(struct mevent));
287 	if (mevp == NULL) {
288 		goto exit;
289 	}
290 
291 	if (type == EVF_TIMER) {
292 		mevp->me_msecs = tfd;
293 		mevp->me_timid = mevent_timid++;
294 	} else
295 		mevp->me_fd = tfd;
296 	mevp->me_type = type;
297 	mevp->me_func = func;
298 	mevp->me_param = param;
299 
300 	LIST_INSERT_HEAD(&change_head, mevp, me_list);
301 	mevp->me_cq = 1;
302 	mevp->me_state = MEV_ADD;
303 	mevent_notify();
304 
305 exit:
306 	mevent_qunlock();
307 
308 	return (mevp);
309 }
310 
311 static int
312 mevent_update(struct mevent *evp, int newstate)
313 {
314 	/*
315 	 * It's not possible to enable/disable a deleted event
316 	 */
317 	if (evp->me_state == MEV_DEL_PENDING)
318 		return (EINVAL);
319 
320 	/*
321 	 * No update needed if state isn't changing
322 	 */
323 	if (evp->me_state == newstate)
324 		return (0);
325 
326 	mevent_qlock();
327 
328 	evp->me_state = newstate;
329 
330 	/*
331 	 * Place the entry onto the changed list if not already there.
332 	 */
333 	if (evp->me_cq == 0) {
334 		evp->me_cq = 1;
335 		LIST_REMOVE(evp, me_list);
336 		LIST_INSERT_HEAD(&change_head, evp, me_list);
337 		mevent_notify();
338 	}
339 
340 	mevent_qunlock();
341 
342 	return (0);
343 }
344 
345 int
346 mevent_enable(struct mevent *evp)
347 {
348 
349 	return (mevent_update(evp, MEV_ENABLE));
350 }
351 
352 int
353 mevent_disable(struct mevent *evp)
354 {
355 
356 	return (mevent_update(evp, MEV_DISABLE));
357 }
358 
359 static int
360 mevent_delete_event(struct mevent *evp, int closefd)
361 {
362 	mevent_qlock();
363 
364 	/*
365          * Place the entry onto the changed list if not already there, and
366 	 * mark as to be deleted.
367          */
368         if (evp->me_cq == 0) {
369 		evp->me_cq = 1;
370 		LIST_REMOVE(evp, me_list);
371 		LIST_INSERT_HEAD(&change_head, evp, me_list);
372 		mevent_notify();
373         }
374 	evp->me_state = MEV_DEL_PENDING;
375 
376 	if (closefd)
377 		evp->me_closefd = 1;
378 
379 	mevent_qunlock();
380 
381 	return (0);
382 }
383 
384 int
385 mevent_delete(struct mevent *evp)
386 {
387 
388 	return (mevent_delete_event(evp, 0));
389 }
390 
391 int
392 mevent_delete_close(struct mevent *evp)
393 {
394 
395 	return (mevent_delete_event(evp, 1));
396 }
397 
398 static void
399 mevent_set_name(void)
400 {
401 
402 	pthread_set_name_np(mevent_tid, "mevent");
403 }
404 
405 void
406 mevent_dispatch(void)
407 {
408 	struct kevent changelist[MEVENT_MAX];
409 	struct kevent eventlist[MEVENT_MAX];
410 	struct mevent *pipev;
411 	int mfd;
412 	int numev;
413 	int ret;
414 #ifndef WITHOUT_CAPSICUM
415 	cap_rights_t rights;
416 #endif
417 
418 	mevent_tid = pthread_self();
419 	mevent_set_name();
420 
421 	mfd = kqueue();
422 	assert(mfd > 0);
423 
424 #ifndef WITHOUT_CAPSICUM
425 	cap_rights_init(&rights, CAP_KQUEUE);
426 	if (caph_rights_limit(mfd, &rights) == -1)
427 		errx(EX_OSERR, "Unable to apply rights for sandbox");
428 #endif
429 
430 	/*
431 	 * Open the pipe that will be used for other threads to force
432 	 * the blocking kqueue call to exit by writing to it. Set the
433 	 * descriptor to non-blocking.
434 	 */
435 	ret = pipe(mevent_pipefd);
436 	if (ret < 0) {
437 		perror("pipe");
438 		exit(0);
439 	}
440 
441 #ifndef WITHOUT_CAPSICUM
442 	cap_rights_init(&rights, CAP_EVENT, CAP_READ, CAP_WRITE);
443 	if (caph_rights_limit(mevent_pipefd[0], &rights) == -1)
444 		errx(EX_OSERR, "Unable to apply rights for sandbox");
445 	if (caph_rights_limit(mevent_pipefd[1], &rights) == -1)
446 		errx(EX_OSERR, "Unable to apply rights for sandbox");
447 #endif
448 
449 	/*
450 	 * Add internal event handler for the pipe write fd
451 	 */
452 	pipev = mevent_add(mevent_pipefd[0], EVF_READ, mevent_pipe_read, NULL);
453 	assert(pipev != NULL);
454 
455 	for (;;) {
456 		/*
457 		 * Build changelist if required.
458 		 * XXX the changelist can be put into the blocking call
459 		 * to eliminate the extra syscall. Currently better for
460 		 * debug.
461 		 */
462 		numev = mevent_build(mfd, changelist);
463 		if (numev) {
464 			ret = kevent(mfd, changelist, numev, NULL, 0, NULL);
465 			if (ret == -1) {
466 				perror("Error return from kevent change");
467 			}
468 		}
469 
470 		/*
471 		 * Block awaiting events
472 		 */
473 		ret = kevent(mfd, NULL, 0, eventlist, MEVENT_MAX, NULL);
474 		if (ret == -1 && errno != EINTR) {
475 			perror("Error return from kevent monitor");
476 		}
477 
478 		/*
479 		 * Handle reported events
480 		 */
481 		mevent_handle(eventlist, ret);
482 	}
483 }
484