xref: /freebsd/tests/sys/aio/aio_test.c (revision c1d255d3ffdbe447de3ab875bf4e7d7accc5bfc5)
1 /*-
2  * Copyright (c) 2004 Robert N. M. Watson
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 THE AUTHOR AND CONTRIBUTORS ``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 THE AUTHOR 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  * Regression test to do some very basic AIO exercising on several types of
31  * file descriptors.  Currently, the tests consist of initializing a fixed
32  * size buffer with pseudo-random data, writing it to one fd using AIO, then
33  * reading it from a second descriptor using AIO.  For some targets, the same
34  * fd is used for write and read (i.e., file, md device), but for others the
35  * operation is performed on a peer (pty, socket, fifo, etc).  For each file
36  * descriptor type, several completion methods are tested.  This test program
37  * does not attempt to exercise error cases or more subtle asynchronous
38  * behavior, just make sure that the basic operations work on some basic object
39  * types.
40  */
41 
42 #include <sys/param.h>
43 #include <sys/module.h>
44 #include <sys/resource.h>
45 #include <sys/socket.h>
46 #include <sys/stat.h>
47 #include <sys/mdioctl.h>
48 
49 #include <aio.h>
50 #include <err.h>
51 #include <errno.h>
52 #include <fcntl.h>
53 #include <libutil.h>
54 #include <limits.h>
55 #include <semaphore.h>
56 #include <stdint.h>
57 #include <stdio.h>
58 #include <stdlib.h>
59 #include <string.h>
60 #include <termios.h>
61 #include <unistd.h>
62 
63 #include <atf-c.h>
64 
65 #include "freebsd_test_suite/macros.h"
66 #include "local.h"
67 
68 /*
69  * GLOBAL_MAX sets the largest usable buffer size to be read and written, as
70  * it sizes ac_buffer in the aio_context structure.  It is also the default
71  * size for file I/O.  For other types, we use smaller blocks or we risk
72  * blocking (and we run in a single process/thread so that would be bad).
73  */
74 #define	GLOBAL_MAX	16384
75 
76 #define	BUFFER_MAX	GLOBAL_MAX
77 
78 /*
79  * A completion function will block until the aio has completed, then return
80  * the result of the aio.  errno will be set appropriately.
81  */
82 typedef ssize_t (*completion)(struct aiocb*);
83 
84 struct aio_context {
85 	int		 ac_read_fd, ac_write_fd;
86 	long		 ac_seed;
87 	char		 ac_buffer[GLOBAL_MAX];
88 	int		 ac_buflen;
89 	int		 ac_seconds;
90 };
91 
92 static sem_t		completions;
93 
94 
95 /*
96  * Fill a buffer given a seed that can be fed into srandom() to initialize
97  * the PRNG in a repeatable manner.
98  */
99 static void
100 aio_fill_buffer(char *buffer, int len, long seed)
101 {
102 	char ch;
103 	int i;
104 
105 	srandom(seed);
106 	for (i = 0; i < len; i++) {
107 		ch = random() & 0xff;
108 		buffer[i] = ch;
109 	}
110 }
111 
112 /*
113  * Test that a buffer matches a given seed.  See aio_fill_buffer().  Return
114  * (1) on a match, (0) on a mismatch.
115  */
116 static int
117 aio_test_buffer(char *buffer, int len, long seed)
118 {
119 	char ch;
120 	int i;
121 
122 	srandom(seed);
123 	for (i = 0; i < len; i++) {
124 		ch = random() & 0xff;
125 		if (buffer[i] != ch)
126 			return (0);
127 	}
128 	return (1);
129 }
130 
131 /*
132  * Initialize a testing context given the file descriptors provided by the
133  * test setup.
134  */
135 static void
136 aio_context_init(struct aio_context *ac, int read_fd,
137     int write_fd, int buflen)
138 {
139 
140 	ATF_REQUIRE_MSG(buflen <= BUFFER_MAX,
141 	    "aio_context_init: buffer too large (%d > %d)",
142 	    buflen, BUFFER_MAX);
143 	bzero(ac, sizeof(*ac));
144 	ac->ac_read_fd = read_fd;
145 	ac->ac_write_fd = write_fd;
146 	ac->ac_buflen = buflen;
147 	srandomdev();
148 	ac->ac_seed = random();
149 	aio_fill_buffer(ac->ac_buffer, buflen, ac->ac_seed);
150 	ATF_REQUIRE_MSG(aio_test_buffer(ac->ac_buffer, buflen,
151 	    ac->ac_seed) != 0, "aio_test_buffer: internal error");
152 }
153 
154 static ssize_t
155 poll(struct aiocb *aio)
156 {
157 	int error;
158 
159 	while ((error = aio_error(aio)) == EINPROGRESS)
160 		usleep(25000);
161 	if (error)
162 		return (error);
163 	else
164 		return (aio_return(aio));
165 }
166 
167 static void
168 sigusr1_handler(int sig __unused)
169 {
170 	ATF_REQUIRE_EQ(0, sem_post(&completions));
171 }
172 
173 static void
174 thr_handler(union sigval sv __unused)
175 {
176 	ATF_REQUIRE_EQ(0, sem_post(&completions));
177 }
178 
179 static ssize_t
180 poll_signaled(struct aiocb *aio)
181 {
182 	int error;
183 
184 	ATF_REQUIRE_EQ(0, sem_wait(&completions));
185 	error = aio_error(aio);
186 	switch (error) {
187 		case EINPROGRESS:
188 			errno = EINTR;
189 			return (-1);
190 		case 0:
191 			return (aio_return(aio));
192 		default:
193 			return (error);
194 	}
195 }
196 
197 /*
198  * Setup a signal handler for signal delivery tests
199  * This isn't thread safe, but it's ok since ATF runs each testcase in a
200  * separate process
201  */
202 static struct sigevent*
203 setup_signal(void)
204 {
205 	static struct sigevent sev;
206 
207 	ATF_REQUIRE_EQ(0, sem_init(&completions, false, 0));
208 	sev.sigev_notify = SIGEV_SIGNAL;
209 	sev.sigev_signo = SIGUSR1;
210 	ATF_REQUIRE(SIG_ERR != signal(SIGUSR1, sigusr1_handler));
211 	return (&sev);
212 }
213 
214 /*
215  * Setup a thread for thread delivery tests
216  * This isn't thread safe, but it's ok since ATF runs each testcase in a
217  * separate process
218  */
219 static struct sigevent*
220 setup_thread(void)
221 {
222 	static struct sigevent sev;
223 
224 	ATF_REQUIRE_EQ(0, sem_init(&completions, false, 0));
225 	sev.sigev_notify = SIGEV_THREAD;
226 	sev.sigev_notify_function = thr_handler;
227 	sev.sigev_notify_attributes = NULL;
228 	return (&sev);
229 }
230 
231 static ssize_t
232 suspend(struct aiocb *aio)
233 {
234 	const struct aiocb *const iocbs[] = {aio};
235 	int error;
236 
237 	error = aio_suspend(iocbs, 1, NULL);
238 	if (error == 0)
239 		return (aio_return(aio));
240 	else
241 		return (error);
242 }
243 
244 static ssize_t
245 waitcomplete(struct aiocb *aio)
246 {
247 	struct aiocb *aiop;
248 	ssize_t ret;
249 
250 	ret = aio_waitcomplete(&aiop, NULL);
251 	ATF_REQUIRE_EQ(aio, aiop);
252 	return (ret);
253 }
254 
255 /*
256  * Perform a simple write test of our initialized data buffer to the provided
257  * file descriptor.
258  */
259 static void
260 aio_write_test(struct aio_context *ac, completion comp, struct sigevent *sev)
261 {
262 	struct aiocb aio;
263 	ssize_t len;
264 
265 	bzero(&aio, sizeof(aio));
266 	aio.aio_buf = ac->ac_buffer;
267 	aio.aio_nbytes = ac->ac_buflen;
268 	aio.aio_fildes = ac->ac_write_fd;
269 	aio.aio_offset = 0;
270 	if (sev)
271 		aio.aio_sigevent = *sev;
272 
273 	if (aio_write(&aio) < 0)
274 		atf_tc_fail("aio_write failed: %s", strerror(errno));
275 
276 	len = comp(&aio);
277 	if (len < 0)
278 		atf_tc_fail("aio failed: %s", strerror(errno));
279 
280 	if (len != ac->ac_buflen)
281 		atf_tc_fail("aio short write (%jd)", (intmax_t)len);
282 }
283 
284 /*
285  * Perform a vectored I/O test of our initialized data buffer to the provided
286  * file descriptor.
287  *
288  * To vectorize the linear buffer, chop it up into two pieces of dissimilar
289  * size, and swap their offsets.
290  */
291 static void
292 aio_writev_test(struct aio_context *ac, completion comp, struct sigevent *sev)
293 {
294 	struct aiocb aio;
295 	struct iovec iov[2];
296 	size_t len0, len1;
297 	ssize_t len;
298 
299 	bzero(&aio, sizeof(aio));
300 
301 	aio.aio_fildes = ac->ac_write_fd;
302 	aio.aio_offset = 0;
303 	len0 = ac->ac_buflen * 3 / 4;
304 	len1 = ac->ac_buflen / 4;
305 	iov[0].iov_base = ac->ac_buffer + len1;
306 	iov[0].iov_len = len0;
307 	iov[1].iov_base = ac->ac_buffer;
308 	iov[1].iov_len = len1;
309 	aio.aio_iov = iov;
310 	aio.aio_iovcnt = 2;
311 	if (sev)
312 		aio.aio_sigevent = *sev;
313 
314 	if (aio_writev(&aio) < 0)
315 		atf_tc_fail("aio_writev failed: %s", strerror(errno));
316 
317 	len = comp(&aio);
318 	if (len < 0)
319 		atf_tc_fail("aio failed: %s", strerror(errno));
320 
321 	if (len != ac->ac_buflen)
322 		atf_tc_fail("aio short write (%jd)", (intmax_t)len);
323 }
324 
325 /*
326  * Perform a simple read test of our initialized data buffer from the
327  * provided file descriptor.
328  */
329 static void
330 aio_read_test(struct aio_context *ac, completion comp, struct sigevent *sev)
331 {
332 	struct aiocb aio;
333 	ssize_t len;
334 
335 	bzero(ac->ac_buffer, ac->ac_buflen);
336 	bzero(&aio, sizeof(aio));
337 	aio.aio_buf = ac->ac_buffer;
338 	aio.aio_nbytes = ac->ac_buflen;
339 	aio.aio_fildes = ac->ac_read_fd;
340 	aio.aio_offset = 0;
341 	if (sev)
342 		aio.aio_sigevent = *sev;
343 
344 	if (aio_read(&aio) < 0)
345 		atf_tc_fail("aio_read failed: %s", strerror(errno));
346 
347 	len = comp(&aio);
348 	if (len < 0)
349 		atf_tc_fail("aio failed: %s", strerror(errno));
350 
351 	ATF_REQUIRE_EQ_MSG(len, ac->ac_buflen,
352 	    "aio short read (%jd)", (intmax_t)len);
353 
354 	if (aio_test_buffer(ac->ac_buffer, ac->ac_buflen, ac->ac_seed) == 0)
355 		atf_tc_fail("buffer mismatched");
356 }
357 
358 static void
359 aio_readv_test(struct aio_context *ac, completion comp, struct sigevent *sev)
360 {
361 	struct aiocb aio;
362 	struct iovec iov[2];
363 	size_t len0, len1;
364 	ssize_t len;
365 
366 	bzero(ac->ac_buffer, ac->ac_buflen);
367 	bzero(&aio, sizeof(aio));
368 	aio.aio_fildes = ac->ac_read_fd;
369 	aio.aio_offset = 0;
370 	len0 = ac->ac_buflen * 3 / 4;
371 	len1 = ac->ac_buflen / 4;
372 	iov[0].iov_base = ac->ac_buffer + len1;
373 	iov[0].iov_len = len0;
374 	iov[1].iov_base = ac->ac_buffer;
375 	iov[1].iov_len = len1;
376 	aio.aio_iov = iov;
377 	aio.aio_iovcnt = 2;
378 	if (sev)
379 		aio.aio_sigevent = *sev;
380 
381 	if (aio_readv(&aio) < 0)
382 		atf_tc_fail("aio_read failed: %s", strerror(errno));
383 
384 	len = comp(&aio);
385 	if (len < 0)
386 		atf_tc_fail("aio failed: %s", strerror(errno));
387 
388 	ATF_REQUIRE_EQ_MSG(len, ac->ac_buflen,
389 	    "aio short read (%jd)", (intmax_t)len);
390 
391 	if (aio_test_buffer(ac->ac_buffer, ac->ac_buflen, ac->ac_seed) == 0)
392 		atf_tc_fail("buffer mismatched");
393 }
394 
395 /*
396  * Series of type-specific tests for AIO.  For now, we just make sure we can
397  * issue a write and then a read to each type.  We assume that once a write
398  * is issued, a read can follow.
399  */
400 
401 /*
402  * Test with a classic file.  Assumes we can create a moderate size temporary
403  * file.
404  */
405 #define	FILE_LEN	GLOBAL_MAX
406 #define	FILE_PATHNAME	"testfile"
407 
408 static void
409 aio_file_test(completion comp, struct sigevent *sev, bool vectored)
410 {
411 	struct aio_context ac;
412 	int fd;
413 
414 	ATF_REQUIRE_KERNEL_MODULE("aio");
415 	ATF_REQUIRE_UNSAFE_AIO();
416 
417 	fd = open(FILE_PATHNAME, O_RDWR | O_CREAT, 0600);
418 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
419 
420 	aio_context_init(&ac, fd, fd, FILE_LEN);
421 	if (vectored) {
422 		aio_writev_test(&ac, comp, sev);
423 		aio_readv_test(&ac, comp, sev);
424 	} else {
425 		aio_write_test(&ac, comp, sev);
426 		aio_read_test(&ac, comp, sev);
427 	}
428 	close(fd);
429 }
430 
431 ATF_TC_WITHOUT_HEAD(file_poll);
432 ATF_TC_BODY(file_poll, tc)
433 {
434 	aio_file_test(poll, NULL, false);
435 }
436 
437 ATF_TC_WITHOUT_HEAD(file_signal);
438 ATF_TC_BODY(file_signal, tc)
439 {
440 	aio_file_test(poll_signaled, setup_signal(), false);
441 }
442 
443 ATF_TC_WITHOUT_HEAD(file_suspend);
444 ATF_TC_BODY(file_suspend, tc)
445 {
446 	aio_file_test(suspend, NULL, false);
447 }
448 
449 ATF_TC_WITHOUT_HEAD(file_thread);
450 ATF_TC_BODY(file_thread, tc)
451 {
452 	aio_file_test(poll_signaled, setup_thread(), false);
453 }
454 
455 ATF_TC_WITHOUT_HEAD(file_waitcomplete);
456 ATF_TC_BODY(file_waitcomplete, tc)
457 {
458 	aio_file_test(waitcomplete, NULL, false);
459 }
460 
461 #define	FIFO_LEN	256
462 #define	FIFO_PATHNAME	"testfifo"
463 
464 static void
465 aio_fifo_test(completion comp, struct sigevent *sev)
466 {
467 	int error, read_fd = -1, write_fd = -1;
468 	struct aio_context ac;
469 
470 	ATF_REQUIRE_KERNEL_MODULE("aio");
471 	ATF_REQUIRE_UNSAFE_AIO();
472 
473 	ATF_REQUIRE_MSG(mkfifo(FIFO_PATHNAME, 0600) != -1,
474 	    "mkfifo failed: %s", strerror(errno));
475 
476 	read_fd = open(FIFO_PATHNAME, O_RDONLY | O_NONBLOCK);
477 	if (read_fd == -1) {
478 		error = errno;
479 		errno = error;
480 		atf_tc_fail("read_fd open failed: %s",
481 		    strerror(errno));
482 	}
483 
484 	write_fd = open(FIFO_PATHNAME, O_WRONLY);
485 	if (write_fd == -1) {
486 		error = errno;
487 		errno = error;
488 		atf_tc_fail("write_fd open failed: %s",
489 		    strerror(errno));
490 	}
491 
492 	aio_context_init(&ac, read_fd, write_fd, FIFO_LEN);
493 	aio_write_test(&ac, comp, sev);
494 	aio_read_test(&ac, comp, sev);
495 
496 	close(read_fd);
497 	close(write_fd);
498 }
499 
500 ATF_TC_WITHOUT_HEAD(fifo_poll);
501 ATF_TC_BODY(fifo_poll, tc)
502 {
503 	aio_fifo_test(poll, NULL);
504 }
505 
506 ATF_TC_WITHOUT_HEAD(fifo_signal);
507 ATF_TC_BODY(fifo_signal, tc)
508 {
509 	aio_fifo_test(poll_signaled, setup_signal());
510 }
511 
512 ATF_TC_WITHOUT_HEAD(fifo_suspend);
513 ATF_TC_BODY(fifo_suspend, tc)
514 {
515 	aio_fifo_test(suspend, NULL);
516 }
517 
518 ATF_TC_WITHOUT_HEAD(fifo_thread);
519 ATF_TC_BODY(fifo_thread, tc)
520 {
521 	aio_fifo_test(poll_signaled, setup_thread());
522 }
523 
524 ATF_TC_WITHOUT_HEAD(fifo_waitcomplete);
525 ATF_TC_BODY(fifo_waitcomplete, tc)
526 {
527 	aio_fifo_test(waitcomplete, NULL);
528 }
529 
530 #define	UNIX_SOCKETPAIR_LEN	256
531 static void
532 aio_unix_socketpair_test(completion comp, struct sigevent *sev, bool vectored)
533 {
534 	struct aio_context ac;
535 	struct rusage ru_before, ru_after;
536 	int sockets[2];
537 
538 	ATF_REQUIRE_KERNEL_MODULE("aio");
539 
540 	ATF_REQUIRE_MSG(socketpair(PF_UNIX, SOCK_STREAM, 0, sockets) != -1,
541 	    "socketpair failed: %s", strerror(errno));
542 
543 	aio_context_init(&ac, sockets[0], sockets[1], UNIX_SOCKETPAIR_LEN);
544 	ATF_REQUIRE_MSG(getrusage(RUSAGE_SELF, &ru_before) != -1,
545 	    "getrusage failed: %s", strerror(errno));
546 	if (vectored) {
547 		aio_writev_test(&ac, comp, sev);
548 		aio_readv_test(&ac, comp, sev);
549 	} else {
550 		aio_write_test(&ac, comp, sev);
551 		aio_read_test(&ac, comp, sev);
552 	}
553 	ATF_REQUIRE_MSG(getrusage(RUSAGE_SELF, &ru_after) != -1,
554 	    "getrusage failed: %s", strerror(errno));
555 	ATF_REQUIRE(ru_after.ru_msgsnd == ru_before.ru_msgsnd + 1);
556 	ATF_REQUIRE(ru_after.ru_msgrcv == ru_before.ru_msgrcv + 1);
557 
558 	close(sockets[0]);
559 	close(sockets[1]);
560 }
561 
562 ATF_TC_WITHOUT_HEAD(socket_poll);
563 ATF_TC_BODY(socket_poll, tc)
564 {
565 	aio_unix_socketpair_test(poll, NULL, false);
566 }
567 
568 ATF_TC_WITHOUT_HEAD(socket_signal);
569 ATF_TC_BODY(socket_signal, tc)
570 {
571 	aio_unix_socketpair_test(poll_signaled, setup_signal(), false);
572 }
573 
574 ATF_TC_WITHOUT_HEAD(socket_suspend);
575 ATF_TC_BODY(socket_suspend, tc)
576 {
577 	aio_unix_socketpair_test(suspend, NULL, false);
578 }
579 
580 ATF_TC_WITHOUT_HEAD(socket_thread);
581 ATF_TC_BODY(socket_thread, tc)
582 {
583 	aio_unix_socketpair_test(poll_signaled, setup_thread(), false);
584 }
585 
586 ATF_TC_WITHOUT_HEAD(socket_waitcomplete);
587 ATF_TC_BODY(socket_waitcomplete, tc)
588 {
589 	aio_unix_socketpair_test(waitcomplete, NULL, false);
590 }
591 
592 struct aio_pty_arg {
593 	int	apa_read_fd;
594 	int	apa_write_fd;
595 };
596 
597 #define	PTY_LEN		256
598 static void
599 aio_pty_test(completion comp, struct sigevent *sev)
600 {
601 	struct aio_context ac;
602 	int read_fd, write_fd;
603 	struct termios ts;
604 	int error;
605 
606 	ATF_REQUIRE_KERNEL_MODULE("aio");
607 	ATF_REQUIRE_UNSAFE_AIO();
608 
609 	ATF_REQUIRE_MSG(openpty(&read_fd, &write_fd, NULL, NULL, NULL) == 0,
610 	    "openpty failed: %s", strerror(errno));
611 
612 
613 	if (tcgetattr(write_fd, &ts) < 0) {
614 		error = errno;
615 		errno = error;
616 		atf_tc_fail("tcgetattr failed: %s", strerror(errno));
617 	}
618 	cfmakeraw(&ts);
619 	if (tcsetattr(write_fd, TCSANOW, &ts) < 0) {
620 		error = errno;
621 		errno = error;
622 		atf_tc_fail("tcsetattr failed: %s", strerror(errno));
623 	}
624 	aio_context_init(&ac, read_fd, write_fd, PTY_LEN);
625 
626 	aio_write_test(&ac, comp, sev);
627 	aio_read_test(&ac, comp, sev);
628 
629 	close(read_fd);
630 	close(write_fd);
631 }
632 
633 ATF_TC_WITHOUT_HEAD(pty_poll);
634 ATF_TC_BODY(pty_poll, tc)
635 {
636 	aio_pty_test(poll, NULL);
637 }
638 
639 ATF_TC_WITHOUT_HEAD(pty_signal);
640 ATF_TC_BODY(pty_signal, tc)
641 {
642 	aio_pty_test(poll_signaled, setup_signal());
643 }
644 
645 ATF_TC_WITHOUT_HEAD(pty_suspend);
646 ATF_TC_BODY(pty_suspend, tc)
647 {
648 	aio_pty_test(suspend, NULL);
649 }
650 
651 ATF_TC_WITHOUT_HEAD(pty_thread);
652 ATF_TC_BODY(pty_thread, tc)
653 {
654 	aio_pty_test(poll_signaled, setup_thread());
655 }
656 
657 ATF_TC_WITHOUT_HEAD(pty_waitcomplete);
658 ATF_TC_BODY(pty_waitcomplete, tc)
659 {
660 	aio_pty_test(waitcomplete, NULL);
661 }
662 
663 #define	PIPE_LEN	256
664 static void
665 aio_pipe_test(completion comp, struct sigevent *sev)
666 {
667 	struct aio_context ac;
668 	int pipes[2];
669 
670 	ATF_REQUIRE_KERNEL_MODULE("aio");
671 	ATF_REQUIRE_UNSAFE_AIO();
672 
673 	ATF_REQUIRE_MSG(pipe(pipes) != -1,
674 	    "pipe failed: %s", strerror(errno));
675 
676 	aio_context_init(&ac, pipes[0], pipes[1], PIPE_LEN);
677 	aio_write_test(&ac, comp, sev);
678 	aio_read_test(&ac, comp, sev);
679 
680 	close(pipes[0]);
681 	close(pipes[1]);
682 }
683 
684 ATF_TC_WITHOUT_HEAD(pipe_poll);
685 ATF_TC_BODY(pipe_poll, tc)
686 {
687 	aio_pipe_test(poll, NULL);
688 }
689 
690 ATF_TC_WITHOUT_HEAD(pipe_signal);
691 ATF_TC_BODY(pipe_signal, tc)
692 {
693 	aio_pipe_test(poll_signaled, setup_signal());
694 }
695 
696 ATF_TC_WITHOUT_HEAD(pipe_suspend);
697 ATF_TC_BODY(pipe_suspend, tc)
698 {
699 	aio_pipe_test(suspend, NULL);
700 }
701 
702 ATF_TC_WITHOUT_HEAD(pipe_thread);
703 ATF_TC_BODY(pipe_thread, tc)
704 {
705 	aio_pipe_test(poll_signaled, setup_thread());
706 }
707 
708 ATF_TC_WITHOUT_HEAD(pipe_waitcomplete);
709 ATF_TC_BODY(pipe_waitcomplete, tc)
710 {
711 	aio_pipe_test(waitcomplete, NULL);
712 }
713 
714 #define	MD_LEN		GLOBAL_MAX
715 #define	MDUNIT_LINK	"mdunit_link"
716 
717 static int
718 aio_md_setup(void)
719 {
720 	int error, fd, mdctl_fd, unit;
721 	char pathname[PATH_MAX];
722 	struct md_ioctl mdio;
723 	char buf[80];
724 
725 	ATF_REQUIRE_KERNEL_MODULE("aio");
726 
727 	mdctl_fd = open("/dev/" MDCTL_NAME, O_RDWR, 0);
728 	ATF_REQUIRE_MSG(mdctl_fd != -1,
729 	    "opening /dev/%s failed: %s", MDCTL_NAME, strerror(errno));
730 
731 	bzero(&mdio, sizeof(mdio));
732 	mdio.md_version = MDIOVERSION;
733 	mdio.md_type = MD_MALLOC;
734 	mdio.md_options = MD_AUTOUNIT | MD_COMPRESS;
735 	mdio.md_mediasize = GLOBAL_MAX;
736 	mdio.md_sectorsize = 512;
737 	strlcpy(buf, __func__, sizeof(buf));
738 	mdio.md_label = buf;
739 
740 	if (ioctl(mdctl_fd, MDIOCATTACH, &mdio) < 0) {
741 		error = errno;
742 		errno = error;
743 		atf_tc_fail("ioctl MDIOCATTACH failed: %s", strerror(errno));
744 	}
745 	close(mdctl_fd);
746 
747 	/* Store the md unit number in a symlink for future cleanup */
748 	unit = mdio.md_unit;
749 	snprintf(buf, sizeof(buf), "%d", unit);
750 	ATF_REQUIRE_EQ(0, symlink(buf, MDUNIT_LINK));
751 	snprintf(pathname, PATH_MAX, "/dev/md%d", unit);
752 	fd = open(pathname, O_RDWR);
753 	ATF_REQUIRE_MSG(fd != -1,
754 	    "opening %s failed: %s", pathname, strerror(errno));
755 
756 	return (fd);
757 }
758 
759 static void
760 aio_md_cleanup(void)
761 {
762 	struct md_ioctl mdio;
763 	int mdctl_fd, n, unit;
764 	char buf[80];
765 
766 	mdctl_fd = open("/dev/" MDCTL_NAME, O_RDWR, 0);
767 	if (mdctl_fd < 0) {
768 		fprintf(stderr, "opening /dev/%s failed: %s\n", MDCTL_NAME,
769 		    strerror(errno));
770 		return;
771 	}
772 	n = readlink(MDUNIT_LINK, buf, sizeof(buf) - 1);
773 	if (n > 0) {
774 		buf[n] = '\0';
775 		if (sscanf(buf, "%d", &unit) == 1 && unit >= 0) {
776 			bzero(&mdio, sizeof(mdio));
777 			mdio.md_version = MDIOVERSION;
778 			mdio.md_unit = unit;
779 			if (ioctl(mdctl_fd, MDIOCDETACH, &mdio) == -1) {
780 				fprintf(stderr,
781 				    "ioctl MDIOCDETACH unit %d failed: %s\n",
782 				    unit, strerror(errno));
783 			}
784 		}
785 	}
786 
787 	close(mdctl_fd);
788 }
789 
790 static void
791 aio_md_test(completion comp, struct sigevent *sev, bool vectored)
792 {
793 	struct aio_context ac;
794 	int fd;
795 
796 	fd = aio_md_setup();
797 	aio_context_init(&ac, fd, fd, MD_LEN);
798 	if (vectored) {
799 		aio_writev_test(&ac, comp, sev);
800 		aio_readv_test(&ac, comp, sev);
801 	} else {
802 		aio_write_test(&ac, comp, sev);
803 		aio_read_test(&ac, comp, sev);
804 	}
805 
806 	close(fd);
807 }
808 
809 ATF_TC_WITH_CLEANUP(md_poll);
810 ATF_TC_HEAD(md_poll, tc)
811 {
812 
813 	atf_tc_set_md_var(tc, "require.user", "root");
814 }
815 ATF_TC_BODY(md_poll, tc)
816 {
817 	aio_md_test(poll, NULL, false);
818 }
819 ATF_TC_CLEANUP(md_poll, tc)
820 {
821 	aio_md_cleanup();
822 }
823 
824 ATF_TC_WITH_CLEANUP(md_signal);
825 ATF_TC_HEAD(md_signal, tc)
826 {
827 
828 	atf_tc_set_md_var(tc, "require.user", "root");
829 }
830 ATF_TC_BODY(md_signal, tc)
831 {
832 	aio_md_test(poll_signaled, setup_signal(), false);
833 }
834 ATF_TC_CLEANUP(md_signal, tc)
835 {
836 	aio_md_cleanup();
837 }
838 
839 ATF_TC_WITH_CLEANUP(md_suspend);
840 ATF_TC_HEAD(md_suspend, tc)
841 {
842 
843 	atf_tc_set_md_var(tc, "require.user", "root");
844 }
845 ATF_TC_BODY(md_suspend, tc)
846 {
847 	aio_md_test(suspend, NULL, false);
848 }
849 ATF_TC_CLEANUP(md_suspend, tc)
850 {
851 	aio_md_cleanup();
852 }
853 
854 ATF_TC_WITH_CLEANUP(md_thread);
855 ATF_TC_HEAD(md_thread, tc)
856 {
857 
858 	atf_tc_set_md_var(tc, "require.user", "root");
859 }
860 ATF_TC_BODY(md_thread, tc)
861 {
862 	aio_md_test(poll_signaled, setup_thread(), false);
863 }
864 ATF_TC_CLEANUP(md_thread, tc)
865 {
866 	aio_md_cleanup();
867 }
868 
869 ATF_TC_WITH_CLEANUP(md_waitcomplete);
870 ATF_TC_HEAD(md_waitcomplete, tc)
871 {
872 
873 	atf_tc_set_md_var(tc, "require.user", "root");
874 }
875 ATF_TC_BODY(md_waitcomplete, tc)
876 {
877 	aio_md_test(waitcomplete, NULL, false);
878 }
879 ATF_TC_CLEANUP(md_waitcomplete, tc)
880 {
881 	aio_md_cleanup();
882 }
883 
884 #define	ZVOL_VDEV_PATHNAME	"test_vdev"
885 #define POOL_SIZE		(1 << 28)	/* 256 MB */
886 #define ZVOL_SIZE		"64m"
887 #define POOL_NAME		"aio_testpool"
888 #define ZVOL_NAME		"aio_testvol"
889 
890 static int
891 aio_zvol_setup(void)
892 {
893 	FILE *pidfile;
894 	int fd;
895 	pid_t pid;
896 	char pool_name[80];
897 	char cmd[160];
898 	char zvol_name[160];
899 	char devname[160];
900 
901 	ATF_REQUIRE_KERNEL_MODULE("aio");
902 	ATF_REQUIRE_KERNEL_MODULE("zfs");
903 
904 	fd = open(ZVOL_VDEV_PATHNAME, O_RDWR | O_CREAT, 0600);
905 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
906 	ATF_REQUIRE_EQ_MSG(0,
907 	    ftruncate(fd, POOL_SIZE), "ftruncate failed: %s", strerror(errno));
908 	close(fd);
909 
910 	pid = getpid();
911 	pidfile = fopen("pidfile", "w");
912 	ATF_REQUIRE_MSG(NULL != pidfile, "fopen: %s", strerror(errno));
913 	fprintf(pidfile, "%d", pid);
914 	fclose(pidfile);
915 
916 	snprintf(pool_name, sizeof(pool_name), POOL_NAME ".%d", pid);
917 	snprintf(zvol_name, sizeof(zvol_name), "%s/" ZVOL_NAME, pool_name);
918 	snprintf(cmd, sizeof(cmd), "zpool create %s $PWD/" ZVOL_VDEV_PATHNAME,
919 	    pool_name);
920 	ATF_REQUIRE_EQ_MSG(0, system(cmd),
921 	    "zpool create failed: %s", strerror(errno));
922 	snprintf(cmd, sizeof(cmd),
923 	    "zfs create -o volblocksize=8192 -o volmode=dev -V "
924 		ZVOL_SIZE " %s", zvol_name);
925 	ATF_REQUIRE_EQ_MSG(0, system(cmd),
926 	    "zfs create failed: %s", strerror(errno));
927 
928 	snprintf(devname, sizeof(devname), "/dev/zvol/%s", zvol_name);
929 	do {
930 		fd = open(devname, O_RDWR);
931 	} while (fd == -1 && errno == EINTR) ;
932 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
933 	return (fd);
934 }
935 
936 static void
937 aio_zvol_cleanup(void)
938 {
939 	FILE *pidfile;
940 	pid_t testpid;
941 	char cmd[160];
942 
943 	pidfile = fopen("pidfile", "r");
944 	if (pidfile == NULL && errno == ENOENT) {
945 		/* Setup probably failed */
946 		return;
947 	}
948 	ATF_REQUIRE_MSG(NULL != pidfile, "fopen: %s", strerror(errno));
949 	ATF_REQUIRE_EQ(1, fscanf(pidfile, "%d", &testpid));
950 	fclose(pidfile);
951 
952 	snprintf(cmd, sizeof(cmd), "zpool destroy " POOL_NAME ".%d", testpid);
953 	system(cmd);
954 }
955 
956 
957 ATF_TC_WITHOUT_HEAD(aio_large_read_test);
958 ATF_TC_BODY(aio_large_read_test, tc)
959 {
960 	struct aiocb cb, *cbp;
961 	ssize_t nread;
962 	size_t len;
963 	int fd;
964 #ifdef __LP64__
965 	int clamped;
966 #endif
967 
968 	ATF_REQUIRE_KERNEL_MODULE("aio");
969 	ATF_REQUIRE_UNSAFE_AIO();
970 
971 #ifdef __LP64__
972 	len = sizeof(clamped);
973 	if (sysctlbyname("debug.iosize_max_clamp", &clamped, &len, NULL, 0) ==
974 	    -1)
975 		atf_libc_error(errno, "Failed to read debug.iosize_max_clamp");
976 #endif
977 
978 	/* Determine the maximum supported read(2) size. */
979 	len = SSIZE_MAX;
980 #ifdef __LP64__
981 	if (clamped)
982 		len = INT_MAX;
983 #endif
984 
985 	fd = open(FILE_PATHNAME, O_RDWR | O_CREAT, 0600);
986 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
987 
988 	unlink(FILE_PATHNAME);
989 
990 	memset(&cb, 0, sizeof(cb));
991 	cb.aio_nbytes = len;
992 	cb.aio_fildes = fd;
993 	cb.aio_buf = NULL;
994 	if (aio_read(&cb) == -1)
995 		atf_tc_fail("aio_read() of maximum read size failed: %s",
996 		    strerror(errno));
997 
998 	nread = aio_waitcomplete(&cbp, NULL);
999 	if (nread == -1)
1000 		atf_tc_fail("aio_waitcomplete() failed: %s", strerror(errno));
1001 	if (nread != 0)
1002 		atf_tc_fail("aio_read() from empty file returned data: %zd",
1003 		    nread);
1004 
1005 	memset(&cb, 0, sizeof(cb));
1006 	cb.aio_nbytes = len + 1;
1007 	cb.aio_fildes = fd;
1008 	cb.aio_buf = NULL;
1009 	if (aio_read(&cb) == -1) {
1010 		if (errno == EINVAL)
1011 			goto finished;
1012 		atf_tc_fail("aio_read() of too large read size failed: %s",
1013 		    strerror(errno));
1014 	}
1015 
1016 	nread = aio_waitcomplete(&cbp, NULL);
1017 	if (nread == -1) {
1018 		if (errno == EINVAL)
1019 			goto finished;
1020 		atf_tc_fail("aio_waitcomplete() failed: %s", strerror(errno));
1021 	}
1022 	atf_tc_fail("aio_read() of too large read size returned: %zd", nread);
1023 
1024 finished:
1025 	close(fd);
1026 }
1027 
1028 /*
1029  * This tests for a bug where arriving socket data can wakeup multiple
1030  * AIO read requests resulting in an uncancellable request.
1031  */
1032 ATF_TC_WITHOUT_HEAD(aio_socket_two_reads);
1033 ATF_TC_BODY(aio_socket_two_reads, tc)
1034 {
1035 	struct ioreq {
1036 		struct aiocb iocb;
1037 		char buffer[1024];
1038 	} ioreq[2];
1039 	struct aiocb *iocb;
1040 	unsigned i;
1041 	int s[2];
1042 	char c;
1043 
1044 	ATF_REQUIRE_KERNEL_MODULE("aio");
1045 #if __FreeBSD_version < 1100101
1046 	aft_tc_skip("kernel version %d is too old (%d required)",
1047 	    __FreeBSD_version, 1100101);
1048 #endif
1049 
1050 	ATF_REQUIRE(socketpair(PF_UNIX, SOCK_STREAM, 0, s) != -1);
1051 
1052 	/* Queue two read requests. */
1053 	memset(&ioreq, 0, sizeof(ioreq));
1054 	for (i = 0; i < nitems(ioreq); i++) {
1055 		ioreq[i].iocb.aio_nbytes = sizeof(ioreq[i].buffer);
1056 		ioreq[i].iocb.aio_fildes = s[0];
1057 		ioreq[i].iocb.aio_buf = ioreq[i].buffer;
1058 		ATF_REQUIRE(aio_read(&ioreq[i].iocb) == 0);
1059 	}
1060 
1061 	/* Send a single byte.  This should complete one request. */
1062 	c = 0xc3;
1063 	ATF_REQUIRE(write(s[1], &c, sizeof(c)) == 1);
1064 
1065 	ATF_REQUIRE(aio_waitcomplete(&iocb, NULL) == 1);
1066 
1067 	/* Determine which request completed and verify the data was read. */
1068 	if (iocb == &ioreq[0].iocb)
1069 		i = 0;
1070 	else
1071 		i = 1;
1072 	ATF_REQUIRE(ioreq[i].buffer[0] == c);
1073 
1074 	i ^= 1;
1075 
1076 	/*
1077 	 * Try to cancel the other request.  On broken systems this
1078 	 * will fail and the process will hang on exit.
1079 	 */
1080 	ATF_REQUIRE(aio_error(&ioreq[i].iocb) == EINPROGRESS);
1081 	ATF_REQUIRE(aio_cancel(s[0], &ioreq[i].iocb) == AIO_CANCELED);
1082 
1083 	close(s[1]);
1084 	close(s[0]);
1085 }
1086 
1087 static void
1088 aio_socket_blocking_short_write_test(bool vectored)
1089 {
1090 	struct aiocb iocb, *iocbp;
1091 	struct iovec iov[2];
1092 	char *buffer[2];
1093 	ssize_t done, r;
1094 	int buffer_size, sb_size;
1095 	socklen_t len;
1096 	int s[2];
1097 
1098 	ATF_REQUIRE_KERNEL_MODULE("aio");
1099 
1100 	ATF_REQUIRE(socketpair(PF_UNIX, SOCK_STREAM, 0, s) != -1);
1101 
1102 	len = sizeof(sb_size);
1103 	ATF_REQUIRE(getsockopt(s[0], SOL_SOCKET, SO_RCVBUF, &sb_size, &len) !=
1104 	    -1);
1105 	ATF_REQUIRE(len == sizeof(sb_size));
1106 	buffer_size = sb_size;
1107 
1108 	ATF_REQUIRE(getsockopt(s[1], SOL_SOCKET, SO_SNDBUF, &sb_size, &len) !=
1109 	    -1);
1110 	ATF_REQUIRE(len == sizeof(sb_size));
1111 	if (sb_size > buffer_size)
1112 		buffer_size = sb_size;
1113 
1114 	/*
1115 	 * Use twice the size of the MAX(receive buffer, send buffer)
1116 	 * to ensure that the write is split up into multiple writes
1117 	 * internally.
1118 	 */
1119 	buffer_size *= 2;
1120 
1121 	buffer[0] = malloc(buffer_size);
1122 	ATF_REQUIRE(buffer[0] != NULL);
1123 	buffer[1] = malloc(buffer_size);
1124 	ATF_REQUIRE(buffer[1] != NULL);
1125 
1126 	srandomdev();
1127 	aio_fill_buffer(buffer[1], buffer_size, random());
1128 
1129 	memset(&iocb, 0, sizeof(iocb));
1130 	iocb.aio_fildes = s[1];
1131 	if (vectored) {
1132 		iov[0].iov_base = buffer[1];
1133 		iov[0].iov_len = buffer_size / 2 + 1;
1134 		iov[1].iov_base = buffer[1] + buffer_size / 2 + 1;
1135 		iov[1].iov_len = buffer_size / 2 - 1;
1136 		iocb.aio_iov = iov;
1137 		iocb.aio_iovcnt = 2;
1138 		r = aio_writev(&iocb);
1139 		ATF_CHECK_EQ_MSG(0, r, "aio_writev returned %zd", r);
1140 	} else {
1141 		iocb.aio_buf = buffer[1];
1142 		iocb.aio_nbytes = buffer_size;
1143 		r = aio_write(&iocb);
1144 		ATF_CHECK_EQ_MSG(0, r, "aio_writev returned %zd", r);
1145 	}
1146 
1147 	done = recv(s[0], buffer[0], buffer_size, MSG_WAITALL);
1148 	ATF_REQUIRE(done == buffer_size);
1149 
1150 	done = aio_waitcomplete(&iocbp, NULL);
1151 	ATF_REQUIRE(iocbp == &iocb);
1152 	ATF_REQUIRE(done == buffer_size);
1153 
1154 	ATF_REQUIRE(memcmp(buffer[0], buffer[1], buffer_size) == 0);
1155 
1156 	close(s[1]);
1157 	close(s[0]);
1158 }
1159 
1160 /*
1161  * This test ensures that aio_write() on a blocking socket of a "large"
1162  * buffer does not return a short completion.
1163  */
1164 ATF_TC_WITHOUT_HEAD(aio_socket_blocking_short_write);
1165 ATF_TC_BODY(aio_socket_blocking_short_write, tc)
1166 {
1167 	aio_socket_blocking_short_write_test(false);
1168 }
1169 
1170 /*
1171  * Like aio_socket_blocking_short_write, but also tests that partially
1172  * completed vectored sends can be retried correctly.
1173  */
1174 ATF_TC_WITHOUT_HEAD(aio_socket_blocking_short_write_vectored);
1175 ATF_TC_BODY(aio_socket_blocking_short_write_vectored, tc)
1176 {
1177 	aio_socket_blocking_short_write_test(true);
1178 }
1179 
1180 /*
1181  * This test verifies that cancelling a partially completed socket write
1182  * returns a short write rather than ECANCELED.
1183  */
1184 ATF_TC_WITHOUT_HEAD(aio_socket_short_write_cancel);
1185 ATF_TC_BODY(aio_socket_short_write_cancel, tc)
1186 {
1187 	struct aiocb iocb, *iocbp;
1188 	char *buffer[2];
1189 	ssize_t done;
1190 	int buffer_size, sb_size;
1191 	socklen_t len;
1192 	int s[2];
1193 
1194 	ATF_REQUIRE_KERNEL_MODULE("aio");
1195 
1196 	ATF_REQUIRE(socketpair(PF_UNIX, SOCK_STREAM, 0, s) != -1);
1197 
1198 	len = sizeof(sb_size);
1199 	ATF_REQUIRE(getsockopt(s[0], SOL_SOCKET, SO_RCVBUF, &sb_size, &len) !=
1200 	    -1);
1201 	ATF_REQUIRE(len == sizeof(sb_size));
1202 	buffer_size = sb_size;
1203 
1204 	ATF_REQUIRE(getsockopt(s[1], SOL_SOCKET, SO_SNDBUF, &sb_size, &len) !=
1205 	    -1);
1206 	ATF_REQUIRE(len == sizeof(sb_size));
1207 	if (sb_size > buffer_size)
1208 		buffer_size = sb_size;
1209 
1210 	/*
1211 	 * Use three times the size of the MAX(receive buffer, send
1212 	 * buffer) for the write to ensure that the write is split up
1213 	 * into multiple writes internally.  The recv() ensures that
1214 	 * the write has partially completed, but a remaining size of
1215 	 * two buffers should ensure that the write has not completed
1216 	 * fully when it is cancelled.
1217 	 */
1218 	buffer[0] = malloc(buffer_size);
1219 	ATF_REQUIRE(buffer[0] != NULL);
1220 	buffer[1] = malloc(buffer_size * 3);
1221 	ATF_REQUIRE(buffer[1] != NULL);
1222 
1223 	srandomdev();
1224 	aio_fill_buffer(buffer[1], buffer_size * 3, random());
1225 
1226 	memset(&iocb, 0, sizeof(iocb));
1227 	iocb.aio_fildes = s[1];
1228 	iocb.aio_buf = buffer[1];
1229 	iocb.aio_nbytes = buffer_size * 3;
1230 	ATF_REQUIRE(aio_write(&iocb) == 0);
1231 
1232 	done = recv(s[0], buffer[0], buffer_size, MSG_WAITALL);
1233 	ATF_REQUIRE(done == buffer_size);
1234 
1235 	ATF_REQUIRE(aio_error(&iocb) == EINPROGRESS);
1236 	ATF_REQUIRE(aio_cancel(s[1], &iocb) == AIO_NOTCANCELED);
1237 
1238 	done = aio_waitcomplete(&iocbp, NULL);
1239 	ATF_REQUIRE(iocbp == &iocb);
1240 	ATF_REQUIRE(done >= buffer_size && done <= buffer_size * 2);
1241 
1242 	ATF_REQUIRE(memcmp(buffer[0], buffer[1], buffer_size) == 0);
1243 
1244 	close(s[1]);
1245 	close(s[0]);
1246 }
1247 
1248 /*
1249  * test aio_fsync's behavior with bad inputs
1250  */
1251 ATF_TC_WITHOUT_HEAD(aio_fsync_errors);
1252 ATF_TC_BODY(aio_fsync_errors, tc)
1253 {
1254 	int fd;
1255 	struct aiocb iocb;
1256 
1257 	ATF_REQUIRE_KERNEL_MODULE("aio");
1258 	ATF_REQUIRE_UNSAFE_AIO();
1259 
1260 	fd = open(FILE_PATHNAME, O_RDWR | O_CREAT, 0600);
1261 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
1262 	unlink(FILE_PATHNAME);
1263 
1264 	/* aio_fsync should return EINVAL unless op is O_SYNC or O_DSYNC */
1265 	memset(&iocb, 0, sizeof(iocb));
1266 	iocb.aio_fildes = fd;
1267 	ATF_CHECK_EQ(-1, aio_fsync(666, &iocb));
1268 	ATF_CHECK_EQ(EINVAL, errno);
1269 
1270 	/* aio_fsync should return EBADF if fd is not a valid descriptor */
1271 	memset(&iocb, 0, sizeof(iocb));
1272 	iocb.aio_fildes = 666;
1273 	ATF_CHECK_EQ(-1, aio_fsync(O_SYNC, &iocb));
1274 	ATF_CHECK_EQ(EBADF, errno);
1275 
1276 	/* aio_fsync should return EINVAL if sigev_notify is invalid */
1277 	memset(&iocb, 0, sizeof(iocb));
1278 	iocb.aio_fildes = fd;
1279 	iocb.aio_sigevent.sigev_notify = 666;
1280 	ATF_CHECK_EQ(-1, aio_fsync(666, &iocb));
1281 	ATF_CHECK_EQ(EINVAL, errno);
1282 }
1283 
1284 /*
1285  * This test just performs a basic test of aio_fsync().
1286  */
1287 static void
1288 aio_fsync_test(int op)
1289 {
1290 	struct aiocb synccb, *iocbp;
1291 	struct {
1292 		struct aiocb iocb;
1293 		bool done;
1294 		char *buffer;
1295 	} buffers[16];
1296 	struct stat sb;
1297 	ssize_t rval;
1298 	unsigned i;
1299 	int fd;
1300 
1301 	ATF_REQUIRE_KERNEL_MODULE("aio");
1302 	ATF_REQUIRE_UNSAFE_AIO();
1303 
1304 	fd = open(FILE_PATHNAME, O_RDWR | O_CREAT, 0600);
1305 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
1306 	unlink(FILE_PATHNAME);
1307 
1308 	ATF_REQUIRE(fstat(fd, &sb) == 0);
1309 	ATF_REQUIRE(sb.st_blksize != 0);
1310 	ATF_REQUIRE(ftruncate(fd, sb.st_blksize * nitems(buffers)) == 0);
1311 
1312 	/*
1313 	 * Queue several asynchronous write requests.  Hopefully this
1314 	 * forces the aio_fsync() request to be deferred.  There is no
1315 	 * reliable way to guarantee that however.
1316 	 */
1317 	srandomdev();
1318 	for (i = 0; i < nitems(buffers); i++) {
1319 		buffers[i].done = false;
1320 		memset(&buffers[i].iocb, 0, sizeof(buffers[i].iocb));
1321 		buffers[i].buffer = malloc(sb.st_blksize);
1322 		aio_fill_buffer(buffers[i].buffer, sb.st_blksize, random());
1323 		buffers[i].iocb.aio_fildes = fd;
1324 		buffers[i].iocb.aio_buf = buffers[i].buffer;
1325 		buffers[i].iocb.aio_nbytes = sb.st_blksize;
1326 		buffers[i].iocb.aio_offset = sb.st_blksize * i;
1327 		ATF_REQUIRE(aio_write(&buffers[i].iocb) == 0);
1328 	}
1329 
1330 	/* Queue the aio_fsync request. */
1331 	memset(&synccb, 0, sizeof(synccb));
1332 	synccb.aio_fildes = fd;
1333 	ATF_REQUIRE(aio_fsync(op, &synccb) == 0);
1334 
1335 	/* Wait for requests to complete. */
1336 	for (;;) {
1337 	next:
1338 		rval = aio_waitcomplete(&iocbp, NULL);
1339 		ATF_REQUIRE(iocbp != NULL);
1340 		if (iocbp == &synccb) {
1341 			ATF_REQUIRE(rval == 0);
1342 			break;
1343 		}
1344 
1345 		for (i = 0; i < nitems(buffers); i++) {
1346 			if (iocbp == &buffers[i].iocb) {
1347 				ATF_REQUIRE(buffers[i].done == false);
1348 				ATF_REQUIRE(rval == sb.st_blksize);
1349 				buffers[i].done = true;
1350 				goto next;
1351 			}
1352 		}
1353 
1354 		ATF_REQUIRE_MSG(false, "unmatched AIO request");
1355 	}
1356 
1357 	for (i = 0; i < nitems(buffers); i++)
1358 		ATF_REQUIRE_MSG(buffers[i].done,
1359 		    "AIO request %u did not complete", i);
1360 
1361 	close(fd);
1362 }
1363 
1364 ATF_TC_WITHOUT_HEAD(aio_fsync_sync_test);
1365 ATF_TC_BODY(aio_fsync_sync_test, tc)
1366 {
1367 	aio_fsync_test(O_SYNC);
1368 }
1369 
1370 ATF_TC_WITHOUT_HEAD(aio_fsync_dsync_test);
1371 ATF_TC_BODY(aio_fsync_dsync_test, tc)
1372 {
1373 	aio_fsync_test(O_DSYNC);
1374 }
1375 
1376 /*
1377  * We shouldn't be able to DoS the system by setting iov_len to an insane
1378  * value
1379  */
1380 ATF_TC_WITHOUT_HEAD(aio_writev_dos_iov_len);
1381 ATF_TC_BODY(aio_writev_dos_iov_len, tc)
1382 {
1383 	struct aiocb aio;
1384 	const struct aiocb *const iocbs[] = {&aio};
1385 	const char *wbuf = "Hello, world!";
1386 	struct iovec iov[1];
1387 	ssize_t len, r;
1388 	int fd;
1389 
1390 	ATF_REQUIRE_KERNEL_MODULE("aio");
1391 	ATF_REQUIRE_UNSAFE_AIO();
1392 
1393 	fd = open("testfile", O_RDWR | O_CREAT, 0600);
1394 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
1395 
1396 	len = strlen(wbuf);
1397 	iov[0].iov_base = __DECONST(void*, wbuf);
1398 	iov[0].iov_len = 1 << 30;
1399 	bzero(&aio, sizeof(aio));
1400 	aio.aio_fildes = fd;
1401 	aio.aio_offset = 0;
1402 	aio.aio_iov = iov;
1403 	aio.aio_iovcnt = 1;
1404 
1405 	r = aio_writev(&aio);
1406 	ATF_CHECK_EQ_MSG(0, r, "aio_writev returned %zd", r);
1407 	ATF_REQUIRE_EQ(0, aio_suspend(iocbs, 1, NULL));
1408 	r = aio_return(&aio);
1409 	ATF_CHECK_EQ_MSG(-1, r, "aio_return returned %zd", r);
1410 	ATF_CHECK_MSG(errno == EFAULT || errno == EINVAL,
1411 	    "aio_writev: %s", strerror(errno));
1412 
1413 	close(fd);
1414 }
1415 
1416 /*
1417  * We shouldn't be able to DoS the system by setting aio_iovcnt to an insane
1418  * value
1419  */
1420 ATF_TC_WITHOUT_HEAD(aio_writev_dos_iovcnt);
1421 ATF_TC_BODY(aio_writev_dos_iovcnt, tc)
1422 {
1423 	struct aiocb aio;
1424 	const char *wbuf = "Hello, world!";
1425 	struct iovec iov[1];
1426 	ssize_t len;
1427 	int fd;
1428 
1429 	ATF_REQUIRE_KERNEL_MODULE("aio");
1430 	ATF_REQUIRE_UNSAFE_AIO();
1431 
1432 	fd = open("testfile", O_RDWR | O_CREAT, 0600);
1433 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
1434 
1435 	len = strlen(wbuf);
1436 	iov[0].iov_base = __DECONST(void*, wbuf);
1437 	iov[0].iov_len = len;
1438 	bzero(&aio, sizeof(aio));
1439 	aio.aio_fildes = fd;
1440 	aio.aio_offset = 0;
1441 	aio.aio_iov = iov;
1442 	aio.aio_iovcnt = 1 << 30;
1443 
1444 	ATF_REQUIRE_EQ(-1, aio_writev(&aio));
1445 	ATF_CHECK_EQ(EINVAL, errno);
1446 
1447 	close(fd);
1448 }
1449 
1450 ATF_TC_WITH_CLEANUP(aio_writev_efault);
1451 ATF_TC_HEAD(aio_writev_efault, tc)
1452 {
1453 	atf_tc_set_md_var(tc, "descr",
1454 	    "Vectored AIO should gracefully handle invalid addresses");
1455 	atf_tc_set_md_var(tc, "require.user", "root");
1456 }
1457 ATF_TC_BODY(aio_writev_efault, tc)
1458 {
1459 	struct aiocb aio;
1460 	ssize_t buflen;
1461 	char *buffer;
1462 	struct iovec iov[2];
1463 	long seed;
1464 	int fd;
1465 
1466 	ATF_REQUIRE_KERNEL_MODULE("aio");
1467 	ATF_REQUIRE_UNSAFE_AIO();
1468 
1469 	fd = aio_md_setup();
1470 
1471 	seed = random();
1472 	buflen = 4096;
1473 	buffer = malloc(buflen);
1474 	aio_fill_buffer(buffer, buflen, seed);
1475 	iov[0].iov_base = buffer;
1476 	iov[0].iov_len = buflen;
1477 	iov[1].iov_base = (void*)-1;	/* Invalid! */
1478 	iov[1].iov_len = buflen;
1479 	bzero(&aio, sizeof(aio));
1480 	aio.aio_fildes = fd;
1481 	aio.aio_offset = 0;
1482 	aio.aio_iov = iov;
1483 	aio.aio_iovcnt = nitems(iov);
1484 
1485 	ATF_REQUIRE_EQ(-1, aio_writev(&aio));
1486 	ATF_CHECK_EQ(EFAULT, errno);
1487 
1488 	close(fd);
1489 }
1490 ATF_TC_CLEANUP(aio_writev_efault, tc)
1491 {
1492 	aio_md_cleanup();
1493 }
1494 
1495 ATF_TC_WITHOUT_HEAD(aio_writev_empty_file_poll);
1496 ATF_TC_BODY(aio_writev_empty_file_poll, tc)
1497 {
1498 	struct aiocb aio;
1499 	int fd;
1500 
1501 	ATF_REQUIRE_KERNEL_MODULE("aio");
1502 	ATF_REQUIRE_UNSAFE_AIO();
1503 
1504 	fd = open("testfile", O_RDWR | O_CREAT, 0600);
1505 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
1506 
1507 	bzero(&aio, sizeof(aio));
1508 	aio.aio_fildes = fd;
1509 	aio.aio_offset = 0;
1510 	aio.aio_iovcnt = 0;
1511 
1512 	ATF_REQUIRE_EQ(0, aio_writev(&aio));
1513 	ATF_REQUIRE_EQ(0, suspend(&aio));
1514 
1515 	close(fd);
1516 }
1517 
1518 ATF_TC_WITHOUT_HEAD(aio_writev_empty_file_signal);
1519 ATF_TC_BODY(aio_writev_empty_file_signal, tc)
1520 {
1521 	struct aiocb aio;
1522 	int fd;
1523 
1524 	ATF_REQUIRE_KERNEL_MODULE("aio");
1525 	ATF_REQUIRE_UNSAFE_AIO();
1526 
1527 	fd = open("testfile", O_RDWR | O_CREAT, 0600);
1528 	ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno));
1529 
1530 	bzero(&aio, sizeof(aio));
1531 	aio.aio_fildes = fd;
1532 	aio.aio_offset = 0;
1533 	aio.aio_iovcnt = 0;
1534 	aio.aio_sigevent = *setup_signal();
1535 
1536 	ATF_REQUIRE_EQ(0, aio_writev(&aio));
1537 	ATF_REQUIRE_EQ(0, poll_signaled(&aio));
1538 
1539 	close(fd);
1540 }
1541 
1542 // aio_writev and aio_readv should still work even if the iovcnt is greater
1543 // than the number of buffered AIO operations permitted per process.
1544 ATF_TC_WITH_CLEANUP(vectored_big_iovcnt);
1545 ATF_TC_HEAD(vectored_big_iovcnt, tc)
1546 {
1547 	atf_tc_set_md_var(tc, "descr",
1548 	    "Vectored AIO should still work even if the iovcnt is greater than "
1549 	    "the number of buffered AIO operations permitted by the process");
1550 	atf_tc_set_md_var(tc, "require.user", "root");
1551 }
1552 ATF_TC_BODY(vectored_big_iovcnt, tc)
1553 {
1554 	struct aiocb aio;
1555 	struct iovec *iov;
1556 	ssize_t len, buflen;
1557 	char *buffer;
1558 	const char *oid = "vfs.aio.max_buf_aio";
1559 	long seed;
1560 	int max_buf_aio;
1561 	int fd, i;
1562 	ssize_t sysctl_len = sizeof(max_buf_aio);
1563 
1564 	ATF_REQUIRE_KERNEL_MODULE("aio");
1565 	ATF_REQUIRE_UNSAFE_AIO();
1566 
1567 	if (sysctlbyname(oid, &max_buf_aio, &sysctl_len, NULL, 0) == -1)
1568 		atf_libc_error(errno, "Failed to read %s", oid);
1569 
1570 	seed = random();
1571 	buflen = 512 * (max_buf_aio + 1);
1572 	buffer = malloc(buflen);
1573 	aio_fill_buffer(buffer, buflen, seed);
1574 	iov = calloc(max_buf_aio + 1, sizeof(struct iovec));
1575 
1576 	fd = aio_md_setup();
1577 
1578 	bzero(&aio, sizeof(aio));
1579 	aio.aio_fildes = fd;
1580 	aio.aio_offset = 0;
1581 	for (i = 0; i < max_buf_aio + 1; i++) {
1582 		iov[i].iov_base = &buffer[i * 512];
1583 		iov[i].iov_len = 512;
1584 	}
1585 	aio.aio_iov = iov;
1586 	aio.aio_iovcnt = max_buf_aio + 1;
1587 
1588 	if (aio_writev(&aio) < 0)
1589 		atf_tc_fail("aio_writev failed: %s", strerror(errno));
1590 
1591 	len = poll(&aio);
1592 	if (len < 0)
1593 		atf_tc_fail("aio failed: %s", strerror(errno));
1594 
1595 	if (len != buflen)
1596 		atf_tc_fail("aio short write (%jd)", (intmax_t)len);
1597 
1598 	bzero(&aio, sizeof(aio));
1599 	aio.aio_fildes = fd;
1600 	aio.aio_offset = 0;
1601 	aio.aio_iov = iov;
1602 	aio.aio_iovcnt = max_buf_aio + 1;
1603 
1604 	if (aio_readv(&aio) < 0)
1605 		atf_tc_fail("aio_readv failed: %s", strerror(errno));
1606 
1607 	len = poll(&aio);
1608 	if (len < 0)
1609 		atf_tc_fail("aio failed: %s", strerror(errno));
1610 
1611 	if (len != buflen)
1612 		atf_tc_fail("aio short read (%jd)", (intmax_t)len);
1613 
1614 	if (aio_test_buffer(buffer, buflen, seed) == 0)
1615 		atf_tc_fail("buffer mismatched");
1616 
1617 	close(fd);
1618 }
1619 ATF_TC_CLEANUP(vectored_big_iovcnt, tc)
1620 {
1621 	aio_md_cleanup();
1622 }
1623 
1624 ATF_TC_WITHOUT_HEAD(vectored_file_poll);
1625 ATF_TC_BODY(vectored_file_poll, tc)
1626 {
1627 	aio_file_test(poll, NULL, true);
1628 }
1629 
1630 ATF_TC_WITHOUT_HEAD(vectored_thread);
1631 ATF_TC_BODY(vectored_thread, tc)
1632 {
1633 	aio_file_test(poll_signaled, setup_thread(), true);
1634 }
1635 
1636 ATF_TC_WITH_CLEANUP(vectored_md_poll);
1637 ATF_TC_HEAD(vectored_md_poll, tc)
1638 {
1639 	atf_tc_set_md_var(tc, "require.user", "root");
1640 }
1641 ATF_TC_BODY(vectored_md_poll, tc)
1642 {
1643 	aio_md_test(poll, NULL, true);
1644 }
1645 ATF_TC_CLEANUP(vectored_md_poll, tc)
1646 {
1647 	aio_md_cleanup();
1648 }
1649 
1650 ATF_TC_WITHOUT_HEAD(vectored_socket_poll);
1651 ATF_TC_BODY(vectored_socket_poll, tc)
1652 {
1653 	aio_unix_socketpair_test(poll, NULL, true);
1654 }
1655 
1656 // aio_writev and aio_readv should still work even if the iov contains elements
1657 // that aren't a multiple of the device's sector size, and even if the total
1658 // amount if I/O _is_ a multiple of the device's sector size.
1659 ATF_TC_WITH_CLEANUP(vectored_unaligned);
1660 ATF_TC_HEAD(vectored_unaligned, tc)
1661 {
1662 	atf_tc_set_md_var(tc, "descr",
1663 	    "Vectored AIO should still work even if the iov contains elements "
1664 	    "that aren't a multiple of the sector size.");
1665 	atf_tc_set_md_var(tc, "require.user", "root");
1666 }
1667 ATF_TC_BODY(vectored_unaligned, tc)
1668 {
1669 	struct aio_context ac;
1670 	struct aiocb aio;
1671 	struct iovec iov[3];
1672 	ssize_t len, total_len;
1673 	int fd;
1674 
1675 	ATF_REQUIRE_KERNEL_MODULE("aio");
1676 	ATF_REQUIRE_UNSAFE_AIO();
1677 
1678 	/*
1679 	 * Use a zvol with volmode=dev, so it will allow .d_write with
1680 	 * unaligned uio.  geom devices use physio, which doesn't allow that.
1681 	 */
1682 	fd = aio_zvol_setup();
1683 	aio_context_init(&ac, fd, fd, FILE_LEN);
1684 
1685 	/* Break the buffer into 3 parts:
1686 	 * * A 4kB part, aligned to 4kB
1687 	 * * Two other parts that add up to 4kB:
1688 	 *   - 256B
1689 	 *   - 4kB - 256B
1690 	 */
1691 	iov[0].iov_base = ac.ac_buffer;
1692 	iov[0].iov_len = 4096;
1693 	iov[1].iov_base = (void*)((uintptr_t)iov[0].iov_base + iov[0].iov_len);
1694 	iov[1].iov_len = 256;
1695 	iov[2].iov_base = (void*)((uintptr_t)iov[1].iov_base + iov[1].iov_len);
1696 	iov[2].iov_len = 4096 - iov[1].iov_len;
1697 	total_len = iov[0].iov_len + iov[1].iov_len + iov[2].iov_len;
1698 	bzero(&aio, sizeof(aio));
1699 	aio.aio_fildes = ac.ac_write_fd;
1700 	aio.aio_offset = 0;
1701 	aio.aio_iov = iov;
1702 	aio.aio_iovcnt = 3;
1703 
1704 	if (aio_writev(&aio) < 0)
1705 		atf_tc_fail("aio_writev failed: %s", strerror(errno));
1706 
1707 	len = poll(&aio);
1708 	if (len < 0)
1709 		atf_tc_fail("aio failed: %s", strerror(errno));
1710 
1711 	if (len != total_len)
1712 		atf_tc_fail("aio short write (%jd)", (intmax_t)len);
1713 
1714 	bzero(&aio, sizeof(aio));
1715 	aio.aio_fildes = ac.ac_read_fd;
1716 	aio.aio_offset = 0;
1717 	aio.aio_iov = iov;
1718 	aio.aio_iovcnt = 3;
1719 
1720 	if (aio_readv(&aio) < 0)
1721 		atf_tc_fail("aio_readv failed: %s", strerror(errno));
1722 	len = poll(&aio);
1723 
1724 	ATF_REQUIRE_MSG(aio_test_buffer(ac.ac_buffer, total_len,
1725 	    ac.ac_seed) != 0, "aio_test_buffer: internal error");
1726 
1727 	close(fd);
1728 }
1729 ATF_TC_CLEANUP(vectored_unaligned, tc)
1730 {
1731 	aio_zvol_cleanup();
1732 }
1733 
1734 static void
1735 aio_zvol_test(completion comp, struct sigevent *sev, bool vectored)
1736 {
1737 	struct aio_context ac;
1738 	int fd;
1739 
1740 	fd = aio_zvol_setup();
1741 	aio_context_init(&ac, fd, fd, MD_LEN);
1742 	if (vectored) {
1743 		aio_writev_test(&ac, comp, sev);
1744 		aio_readv_test(&ac, comp, sev);
1745 	} else {
1746 		aio_write_test(&ac, comp, sev);
1747 		aio_read_test(&ac, comp, sev);
1748 	}
1749 
1750 	close(fd);
1751 }
1752 
1753 /*
1754  * Note that unlike md, the zvol is not a geom device, does not allow unmapped
1755  * buffers, and does not use physio.
1756  */
1757 ATF_TC_WITH_CLEANUP(vectored_zvol_poll);
1758 ATF_TC_HEAD(vectored_zvol_poll, tc)
1759 {
1760 	atf_tc_set_md_var(tc, "require.user", "root");
1761 }
1762 ATF_TC_BODY(vectored_zvol_poll, tc)
1763 {
1764 	aio_zvol_test(poll, NULL, true);
1765 }
1766 ATF_TC_CLEANUP(vectored_zvol_poll, tc)
1767 {
1768 	aio_zvol_cleanup();
1769 }
1770 
1771 ATF_TP_ADD_TCS(tp)
1772 {
1773 
1774 	ATF_TP_ADD_TC(tp, file_poll);
1775 	ATF_TP_ADD_TC(tp, file_signal);
1776 	ATF_TP_ADD_TC(tp, file_suspend);
1777 	ATF_TP_ADD_TC(tp, file_thread);
1778 	ATF_TP_ADD_TC(tp, file_waitcomplete);
1779 	ATF_TP_ADD_TC(tp, fifo_poll);
1780 	ATF_TP_ADD_TC(tp, fifo_signal);
1781 	ATF_TP_ADD_TC(tp, fifo_suspend);
1782 	ATF_TP_ADD_TC(tp, fifo_thread);
1783 	ATF_TP_ADD_TC(tp, fifo_waitcomplete);
1784 	ATF_TP_ADD_TC(tp, socket_poll);
1785 	ATF_TP_ADD_TC(tp, socket_signal);
1786 	ATF_TP_ADD_TC(tp, socket_suspend);
1787 	ATF_TP_ADD_TC(tp, socket_thread);
1788 	ATF_TP_ADD_TC(tp, socket_waitcomplete);
1789 	ATF_TP_ADD_TC(tp, pty_poll);
1790 	ATF_TP_ADD_TC(tp, pty_signal);
1791 	ATF_TP_ADD_TC(tp, pty_suspend);
1792 	ATF_TP_ADD_TC(tp, pty_thread);
1793 	ATF_TP_ADD_TC(tp, pty_waitcomplete);
1794 	ATF_TP_ADD_TC(tp, pipe_poll);
1795 	ATF_TP_ADD_TC(tp, pipe_signal);
1796 	ATF_TP_ADD_TC(tp, pipe_suspend);
1797 	ATF_TP_ADD_TC(tp, pipe_thread);
1798 	ATF_TP_ADD_TC(tp, pipe_waitcomplete);
1799 	ATF_TP_ADD_TC(tp, md_poll);
1800 	ATF_TP_ADD_TC(tp, md_signal);
1801 	ATF_TP_ADD_TC(tp, md_suspend);
1802 	ATF_TP_ADD_TC(tp, md_thread);
1803 	ATF_TP_ADD_TC(tp, md_waitcomplete);
1804 	ATF_TP_ADD_TC(tp, aio_fsync_errors);
1805 	ATF_TP_ADD_TC(tp, aio_fsync_sync_test);
1806 	ATF_TP_ADD_TC(tp, aio_fsync_dsync_test);
1807 	ATF_TP_ADD_TC(tp, aio_large_read_test);
1808 	ATF_TP_ADD_TC(tp, aio_socket_two_reads);
1809 	ATF_TP_ADD_TC(tp, aio_socket_blocking_short_write);
1810 	ATF_TP_ADD_TC(tp, aio_socket_blocking_short_write_vectored);
1811 	ATF_TP_ADD_TC(tp, aio_socket_short_write_cancel);
1812 	ATF_TP_ADD_TC(tp, aio_writev_dos_iov_len);
1813 	ATF_TP_ADD_TC(tp, aio_writev_dos_iovcnt);
1814 	ATF_TP_ADD_TC(tp, aio_writev_efault);
1815 	ATF_TP_ADD_TC(tp, aio_writev_empty_file_poll);
1816 	ATF_TP_ADD_TC(tp, aio_writev_empty_file_signal);
1817 	ATF_TP_ADD_TC(tp, vectored_big_iovcnt);
1818 	ATF_TP_ADD_TC(tp, vectored_file_poll);
1819 	ATF_TP_ADD_TC(tp, vectored_md_poll);
1820 	ATF_TP_ADD_TC(tp, vectored_zvol_poll);
1821 	ATF_TP_ADD_TC(tp, vectored_unaligned);
1822 	ATF_TP_ADD_TC(tp, vectored_socket_poll);
1823 	ATF_TP_ADD_TC(tp, vectored_thread);
1824 
1825 	return (atf_no_error());
1826 }
1827