xref: /illumos-gate/usr/src/cmd/ztest/ztest.c (revision b5989ec7cd031c9bb066c68877fb5017e45f205b)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * The objective of this program is to provide a DMU/ZAP/SPA stress test
30  * that runs entirely in userland, is easy to use, and easy to extend.
31  *
32  * The overall design of the ztest program is as follows:
33  *
34  * (1) For each major functional area (e.g. adding vdevs to a pool,
35  *     creating and destroying datasets, reading and writing objects, etc)
36  *     we have a simple routine to test that functionality.  These
37  *     individual routines do not have to do anything "stressful".
38  *
39  * (2) We turn these simple functionality tests into a stress test by
40  *     running them all in parallel, with as many threads as desired,
41  *     and spread across as many datasets, objects, and vdevs as desired.
42  *
43  * (3) While all this is happening, we inject faults into the pool to
44  *     verify that self-healing data really works.
45  *
46  * (4) Every time we open a dataset, we change its checksum and compression
47  *     functions.  Thus even individual objects vary from block to block
48  *     in which checksum they use and whether they're compressed.
49  *
50  * (5) To verify that we never lose on-disk consistency after a crash,
51  *     we run the entire test in a child of the main process.
52  *     At random times, the child self-immolates with a SIGKILL.
53  *     This is the software equivalent of pulling the power cord.
54  *     The parent then runs the test again, using the existing
55  *     storage pool, as many times as desired.
56  *
57  * (6) To verify that we don't have future leaks or temporal incursions,
58  *     many of the functional tests record the transaction group number
59  *     as part of their data.  When reading old data, they verify that
60  *     the transaction group number is less than the current, open txg.
61  *     If you add a new test, please do this if applicable.
62  *
63  * When run with no arguments, ztest runs for about five minutes and
64  * produces no output if successful.  To get a little bit of information,
65  * specify -V.  To get more information, specify -VV, and so on.
66  *
67  * To turn this into an overnight stress test, use -T to specify run time.
68  *
69  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
70  * to increase the pool capacity, fanout, and overall stress level.
71  *
72  * The -N(okill) option will suppress kills, so each child runs to completion.
73  * This can be useful when you're trying to distinguish temporal incursions
74  * from plain old race conditions.
75  */
76 
77 #include <sys/zfs_context.h>
78 #include <sys/spa.h>
79 #include <sys/dmu.h>
80 #include <sys/txg.h>
81 #include <sys/zap.h>
82 #include <sys/dmu_traverse.h>
83 #include <sys/dmu_objset.h>
84 #include <sys/poll.h>
85 #include <sys/stat.h>
86 #include <sys/time.h>
87 #include <sys/wait.h>
88 #include <sys/mman.h>
89 #include <sys/resource.h>
90 #include <sys/zio.h>
91 #include <sys/zio_checksum.h>
92 #include <sys/zio_compress.h>
93 #include <sys/zil.h>
94 #include <sys/vdev_impl.h>
95 #include <sys/spa_impl.h>
96 #include <sys/dsl_prop.h>
97 #include <sys/refcount.h>
98 #include <stdio.h>
99 #include <stdio_ext.h>
100 #include <stdlib.h>
101 #include <unistd.h>
102 #include <signal.h>
103 #include <umem.h>
104 #include <dlfcn.h>
105 #include <ctype.h>
106 #include <math.h>
107 #include <sys/fs/zfs.h>
108 
109 static char cmdname[] = "ztest";
110 static char *zopt_pool = cmdname;
111 
112 static uint64_t zopt_vdevs = 5;
113 static uint64_t zopt_vdevtime;
114 static int zopt_ashift = SPA_MINBLOCKSHIFT;
115 static int zopt_mirrors = 2;
116 static int zopt_raidz = 4;
117 static int zopt_raidz_parity = 1;
118 static size_t zopt_vdev_size = SPA_MINDEVSIZE;
119 static int zopt_datasets = 7;
120 static int zopt_threads = 23;
121 static uint64_t zopt_passtime = 60;	/* 60 seconds */
122 static uint64_t zopt_killrate = 70;	/* 70% kill rate */
123 static int zopt_verbose = 0;
124 static int zopt_init = 1;
125 static char *zopt_dir = "/tmp";
126 static uint64_t zopt_time = 300;	/* 5 minutes */
127 static int zopt_maxfaults;
128 
129 typedef struct ztest_args {
130 	char		*za_pool;
131 	objset_t	*za_os;
132 	zilog_t		*za_zilog;
133 	thread_t	za_thread;
134 	uint64_t	za_instance;
135 	uint64_t	za_random;
136 	uint64_t	za_diroff;
137 	uint64_t	za_diroff_shared;
138 	uint64_t	za_zil_seq;
139 	hrtime_t	za_start;
140 	hrtime_t	za_stop;
141 	hrtime_t	za_kill;
142 	traverse_handle_t *za_th;
143 } ztest_args_t;
144 
145 typedef void ztest_func_t(ztest_args_t *);
146 
147 /*
148  * Note: these aren't static because we want dladdr() to work.
149  */
150 ztest_func_t ztest_dmu_read_write;
151 ztest_func_t ztest_dmu_write_parallel;
152 ztest_func_t ztest_dmu_object_alloc_free;
153 ztest_func_t ztest_zap;
154 ztest_func_t ztest_zap_parallel;
155 ztest_func_t ztest_traverse;
156 ztest_func_t ztest_dsl_prop_get_set;
157 ztest_func_t ztest_dmu_objset_create_destroy;
158 ztest_func_t ztest_dmu_snapshot_create_destroy;
159 ztest_func_t ztest_spa_create_destroy;
160 ztest_func_t ztest_fault_inject;
161 ztest_func_t ztest_vdev_attach_detach;
162 ztest_func_t ztest_vdev_LUN_growth;
163 ztest_func_t ztest_vdev_add_remove;
164 ztest_func_t ztest_scrub;
165 ztest_func_t ztest_spa_rename;
166 
167 typedef struct ztest_info {
168 	ztest_func_t	*zi_func;	/* test function */
169 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
170 	uint64_t	zi_calls;	/* per-pass count */
171 	uint64_t	zi_call_time;	/* per-pass time */
172 	uint64_t	zi_call_total;	/* cumulative total */
173 	uint64_t	zi_call_target;	/* target cumulative total */
174 } ztest_info_t;
175 
176 uint64_t zopt_always = 0;		/* all the time */
177 uint64_t zopt_often = 1;		/* every second */
178 uint64_t zopt_sometimes = 10;		/* every 10 seconds */
179 uint64_t zopt_rarely = 60;		/* every 60 seconds */
180 
181 ztest_info_t ztest_info[] = {
182 	{ ztest_dmu_read_write,			&zopt_always	},
183 	{ ztest_dmu_write_parallel,		&zopt_always	},
184 	{ ztest_dmu_object_alloc_free,		&zopt_always	},
185 	{ ztest_zap,				&zopt_always	},
186 	{ ztest_zap_parallel,			&zopt_always	},
187 	{ ztest_traverse,			&zopt_often	},
188 	{ ztest_dsl_prop_get_set,		&zopt_sometimes	},
189 	{ ztest_dmu_objset_create_destroy,	&zopt_sometimes	},
190 	{ ztest_dmu_snapshot_create_destroy,	&zopt_rarely	},
191 	{ ztest_spa_create_destroy,		&zopt_sometimes	},
192 	{ ztest_fault_inject,			&zopt_sometimes	},
193 	{ ztest_spa_rename,			&zopt_rarely	},
194 	{ ztest_vdev_attach_detach,		&zopt_rarely	},
195 	{ ztest_vdev_LUN_growth,		&zopt_rarely	},
196 	{ ztest_vdev_add_remove,		&zopt_vdevtime	},
197 	{ ztest_scrub,				&zopt_vdevtime	},
198 };
199 
200 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
201 
202 #define	ZTEST_SYNC_LOCKS	16
203 
204 /*
205  * Stuff we need to share writably between parent and child.
206  */
207 typedef struct ztest_shared {
208 	mutex_t		zs_vdev_lock;
209 	rwlock_t	zs_name_lock;
210 	uint64_t	zs_vdev_primaries;
211 	uint64_t	zs_enospc_count;
212 	hrtime_t	zs_start_time;
213 	hrtime_t	zs_stop_time;
214 	uint64_t	zs_alloc;
215 	uint64_t	zs_space;
216 	ztest_info_t	zs_info[ZTEST_FUNCS];
217 	mutex_t		zs_sync_lock[ZTEST_SYNC_LOCKS];
218 	uint64_t	zs_seq[ZTEST_SYNC_LOCKS];
219 } ztest_shared_t;
220 
221 typedef struct ztest_block_tag {
222 	uint64_t	bt_objset;
223 	uint64_t	bt_object;
224 	uint64_t	bt_offset;
225 	uint64_t	bt_txg;
226 	uint64_t	bt_thread;
227 	uint64_t	bt_seq;
228 } ztest_block_tag_t;
229 
230 static char ztest_dev_template[] = "%s/%s.%llua";
231 static ztest_shared_t *ztest_shared;
232 
233 static int ztest_random_fd;
234 static int ztest_dump_core = 1;
235 
236 extern uint64_t zio_gang_bang;
237 
238 #define	ZTEST_DIROBJ		1
239 #define	ZTEST_MICROZAP_OBJ	2
240 #define	ZTEST_FATZAP_OBJ	3
241 
242 #define	ZTEST_DIROBJ_BLOCKSIZE	(1 << 10)
243 #define	ZTEST_DIRSIZE		256
244 
245 /*
246  * These libumem hooks provide a reasonable set of defaults for the allocator's
247  * debugging facilities.
248  */
249 const char *
250 _umem_debug_init()
251 {
252 	return ("default,verbose"); /* $UMEM_DEBUG setting */
253 }
254 
255 const char *
256 _umem_logging_init(void)
257 {
258 	return ("fail,contents"); /* $UMEM_LOGGING setting */
259 }
260 
261 #define	FATAL_MSG_SZ	1024
262 
263 char *fatal_msg;
264 
265 static void
266 fatal(int do_perror, char *message, ...)
267 {
268 	va_list args;
269 	int save_errno = errno;
270 	char buf[FATAL_MSG_SZ];
271 
272 	(void) fflush(stdout);
273 
274 	va_start(args, message);
275 	(void) sprintf(buf, "ztest: ");
276 	/* LINTED */
277 	(void) vsprintf(buf + strlen(buf), message, args);
278 	va_end(args);
279 	if (do_perror) {
280 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
281 		    ": %s", strerror(save_errno));
282 	}
283 	(void) fprintf(stderr, "%s\n", buf);
284 	fatal_msg = buf;			/* to ease debugging */
285 	if (ztest_dump_core)
286 		abort();
287 	exit(3);
288 }
289 
290 static int
291 str2shift(const char *buf)
292 {
293 	const char *ends = "BKMGTPEZ";
294 	int i;
295 
296 	if (buf[0] == '\0')
297 		return (0);
298 	for (i = 0; i < strlen(ends); i++) {
299 		if (toupper(buf[0]) == ends[i])
300 			break;
301 	}
302 	if (i == strlen(ends))
303 		fatal(0, "invalid bytes suffix: %s", buf);
304 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
305 		return (10*i);
306 	}
307 	fatal(0, "invalid bytes suffix: %s", buf);
308 	return (-1);
309 }
310 
311 static uint64_t
312 nicenumtoull(const char *buf)
313 {
314 	char *end;
315 	uint64_t val;
316 
317 	val = strtoull(buf, &end, 0);
318 	if (end == buf) {
319 		fatal(0, "bad numeric value: %s", buf);
320 	} else if (end[0] == '.') {
321 		double fval = strtod(buf, &end);
322 		fval *= pow(2, str2shift(end));
323 		if (fval > UINT64_MAX)
324 			fatal(0, "value too large: %s", buf);
325 		val = (uint64_t)fval;
326 	} else {
327 		int shift = str2shift(end);
328 		if (shift >= 64 || (val << shift) >> shift != val)
329 			fatal(0, "value too large: %s", buf);
330 		val <<= shift;
331 	}
332 	return (val);
333 }
334 
335 static void
336 usage(void)
337 {
338 	char nice_vdev_size[10];
339 	char nice_gang_bang[10];
340 
341 	nicenum(zopt_vdev_size, nice_vdev_size);
342 	nicenum(zio_gang_bang, nice_gang_bang);
343 
344 	(void) printf("Usage: %s\n"
345 	    "\t[-v vdevs (default: %llu)]\n"
346 	    "\t[-s size_of_each_vdev (default: %s)]\n"
347 	    "\t[-a alignment_shift (default: %d) (use 0 for random)]\n"
348 	    "\t[-m mirror_copies (default: %d)]\n"
349 	    "\t[-r raidz_disks (default: %d)]\n"
350 	    "\t[-R raidz_parity (default: %d)]\n"
351 	    "\t[-d datasets (default: %d)]\n"
352 	    "\t[-t threads (default: %d)]\n"
353 	    "\t[-g gang_block_threshold (default: %s)]\n"
354 	    "\t[-i initialize pool i times (default: %d)]\n"
355 	    "\t[-k kill percentage (default: %llu%%)]\n"
356 	    "\t[-p pool_name (default: %s)]\n"
357 	    "\t[-f file directory for vdev files (default: %s)]\n"
358 	    "\t[-V(erbose)] (use multiple times for ever more blather)\n"
359 	    "\t[-E(xisting)] (use existing pool instead of creating new one)\n"
360 	    "\t[-T time] total run time (default: %llu sec)\n"
361 	    "\t[-P passtime] time per pass (default: %llu sec)\n"
362 	    "",
363 	    cmdname,
364 	    (u_longlong_t)zopt_vdevs,		/* -v */
365 	    nice_vdev_size,			/* -s */
366 	    zopt_ashift,			/* -a */
367 	    zopt_mirrors,			/* -m */
368 	    zopt_raidz,				/* -r */
369 	    zopt_raidz_parity,			/* -R */
370 	    zopt_datasets,			/* -d */
371 	    zopt_threads,			/* -t */
372 	    nice_gang_bang,			/* -g */
373 	    zopt_init,				/* -i */
374 	    (u_longlong_t)zopt_killrate,	/* -k */
375 	    zopt_pool,				/* -p */
376 	    zopt_dir,				/* -f */
377 	    (u_longlong_t)zopt_time,		/* -T */
378 	    (u_longlong_t)zopt_passtime);	/* -P */
379 	exit(1);
380 }
381 
382 static uint64_t
383 ztest_random(uint64_t range)
384 {
385 	uint64_t r;
386 
387 	if (range == 0)
388 		return (0);
389 
390 	if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r))
391 		fatal(1, "short read from /dev/urandom");
392 
393 	return (r % range);
394 }
395 
396 static void
397 ztest_record_enospc(char *s)
398 {
399 	dprintf("ENOSPC doing: %s\n", s ? s : "<unknown>");
400 	ztest_shared->zs_enospc_count++;
401 }
402 
403 static void
404 process_options(int argc, char **argv)
405 {
406 	int opt;
407 	uint64_t value;
408 
409 	/* By default, test gang blocks for blocks 32K and greater */
410 	zio_gang_bang = 32 << 10;
411 
412 	while ((opt = getopt(argc, argv,
413 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:")) != EOF) {
414 		value = 0;
415 		switch (opt) {
416 		    case 'v':
417 		    case 's':
418 		    case 'a':
419 		    case 'm':
420 		    case 'r':
421 		    case 'R':
422 		    case 'd':
423 		    case 't':
424 		    case 'g':
425 		    case 'i':
426 		    case 'k':
427 		    case 'T':
428 		    case 'P':
429 			value = nicenumtoull(optarg);
430 		}
431 		switch (opt) {
432 		    case 'v':
433 			zopt_vdevs = value;
434 			break;
435 		    case 's':
436 			zopt_vdev_size = MAX(SPA_MINDEVSIZE, value);
437 			break;
438 		    case 'a':
439 			zopt_ashift = value;
440 			break;
441 		    case 'm':
442 			zopt_mirrors = value;
443 			break;
444 		    case 'r':
445 			zopt_raidz = MAX(1, value);
446 			break;
447 		    case 'R':
448 			zopt_raidz_parity = MIN(MAX(value, 1), 2);
449 			break;
450 		    case 'd':
451 			zopt_datasets = MAX(1, value);
452 			break;
453 		    case 't':
454 			zopt_threads = MAX(1, value);
455 			break;
456 		    case 'g':
457 			zio_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value);
458 			break;
459 		    case 'i':
460 			zopt_init = value;
461 			break;
462 		    case 'k':
463 			zopt_killrate = value;
464 			break;
465 		    case 'p':
466 			zopt_pool = strdup(optarg);
467 			break;
468 		    case 'f':
469 			zopt_dir = strdup(optarg);
470 			break;
471 		    case 'V':
472 			zopt_verbose++;
473 			break;
474 		    case 'E':
475 			zopt_init = 0;
476 			break;
477 		    case 'T':
478 			zopt_time = value;
479 			break;
480 		    case 'P':
481 			zopt_passtime = MAX(1, value);
482 			break;
483 		    case '?':
484 		    default:
485 			usage();
486 			break;
487 		}
488 	}
489 
490 	zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1);
491 
492 	zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time / zopt_vdevs : UINT64_MAX);
493 	zopt_maxfaults = MAX(zopt_mirrors, 1) * (zopt_raidz_parity + 1) - 1;
494 }
495 
496 static uint64_t
497 ztest_get_ashift(void)
498 {
499 	if (zopt_ashift == 0)
500 		return (SPA_MINBLOCKSHIFT + ztest_random(3));
501 	return (zopt_ashift);
502 }
503 
504 static nvlist_t *
505 make_vdev_file(size_t size)
506 {
507 	char dev_name[MAXPATHLEN];
508 	uint64_t vdev;
509 	uint64_t ashift = ztest_get_ashift();
510 	int fd;
511 	nvlist_t *file;
512 
513 	if (size == 0) {
514 		(void) snprintf(dev_name, sizeof (dev_name), "%s",
515 		    "/dev/bogus");
516 	} else {
517 		vdev = ztest_shared->zs_vdev_primaries++;
518 		(void) sprintf(dev_name, ztest_dev_template,
519 		    zopt_dir, zopt_pool, vdev);
520 
521 		fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666);
522 		if (fd == -1)
523 			fatal(1, "can't open %s", dev_name);
524 		if (ftruncate(fd, size) != 0)
525 			fatal(1, "can't ftruncate %s", dev_name);
526 		(void) close(fd);
527 	}
528 
529 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
530 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
531 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0);
532 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
533 
534 	return (file);
535 }
536 
537 static nvlist_t *
538 make_vdev_raidz(size_t size, int r)
539 {
540 	nvlist_t *raidz, **child;
541 	int c;
542 
543 	if (r < 2)
544 		return (make_vdev_file(size));
545 
546 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
547 
548 	for (c = 0; c < r; c++)
549 		child[c] = make_vdev_file(size);
550 
551 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
552 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
553 	    VDEV_TYPE_RAIDZ) == 0);
554 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
555 	    zopt_raidz_parity) == 0);
556 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
557 	    child, r) == 0);
558 
559 	for (c = 0; c < r; c++)
560 		nvlist_free(child[c]);
561 
562 	umem_free(child, r * sizeof (nvlist_t *));
563 
564 	return (raidz);
565 }
566 
567 static nvlist_t *
568 make_vdev_mirror(size_t size, int r, int m)
569 {
570 	nvlist_t *mirror, **child;
571 	int c;
572 
573 	if (m < 1)
574 		return (make_vdev_raidz(size, r));
575 
576 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
577 
578 	for (c = 0; c < m; c++)
579 		child[c] = make_vdev_raidz(size, r);
580 
581 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
582 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
583 	    VDEV_TYPE_MIRROR) == 0);
584 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
585 	    child, m) == 0);
586 
587 	for (c = 0; c < m; c++)
588 		nvlist_free(child[c]);
589 
590 	umem_free(child, m * sizeof (nvlist_t *));
591 
592 	return (mirror);
593 }
594 
595 static nvlist_t *
596 make_vdev_root(size_t size, int r, int m, int t)
597 {
598 	nvlist_t *root, **child;
599 	int c;
600 
601 	ASSERT(t > 0);
602 
603 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
604 
605 	for (c = 0; c < t; c++)
606 		child[c] = make_vdev_mirror(size, r, m);
607 
608 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
609 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
610 	VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN,
611 	    child, t) == 0);
612 
613 	for (c = 0; c < t; c++)
614 		nvlist_free(child[c]);
615 
616 	umem_free(child, t * sizeof (nvlist_t *));
617 
618 	return (root);
619 }
620 
621 static void
622 ztest_set_random_blocksize(objset_t *os, uint64_t object, dmu_tx_t *tx)
623 {
624 	int bs = SPA_MINBLOCKSHIFT +
625 	    ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1);
626 	int ibs = DN_MIN_INDBLKSHIFT +
627 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1);
628 	int error;
629 
630 	error = dmu_object_set_blocksize(os, object, 1ULL << bs, ibs, tx);
631 	if (error) {
632 		char osname[300];
633 		dmu_objset_name(os, osname);
634 		fatal(0, "dmu_object_set_blocksize('%s', %llu, %d, %d) = %d",
635 		    osname, object, 1 << bs, ibs, error);
636 	}
637 }
638 
639 static uint8_t
640 ztest_random_checksum(void)
641 {
642 	uint8_t checksum;
643 
644 	do {
645 		checksum = ztest_random(ZIO_CHECKSUM_FUNCTIONS);
646 	} while (zio_checksum_table[checksum].ci_zbt);
647 
648 	if (checksum == ZIO_CHECKSUM_OFF)
649 		checksum = ZIO_CHECKSUM_ON;
650 
651 	return (checksum);
652 }
653 
654 static uint8_t
655 ztest_random_compress(void)
656 {
657 	return ((uint8_t)ztest_random(ZIO_COMPRESS_FUNCTIONS));
658 }
659 
660 typedef struct ztest_replay {
661 	objset_t	*zr_os;
662 	uint64_t	zr_assign;
663 } ztest_replay_t;
664 
665 static int
666 ztest_replay_create(ztest_replay_t *zr, lr_create_t *lr, boolean_t byteswap)
667 {
668 	objset_t *os = zr->zr_os;
669 	dmu_tx_t *tx;
670 	int error;
671 
672 	if (byteswap)
673 		byteswap_uint64_array(lr, sizeof (*lr));
674 
675 	tx = dmu_tx_create(os);
676 	dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
677 	error = dmu_tx_assign(tx, zr->zr_assign);
678 	if (error) {
679 		dmu_tx_abort(tx);
680 		return (error);
681 	}
682 
683 	error = dmu_object_claim(os, lr->lr_doid, lr->lr_mode, 0,
684 	    DMU_OT_NONE, 0, tx);
685 	ASSERT3U(error, ==, 0);
686 	dmu_tx_commit(tx);
687 
688 	if (zopt_verbose >= 5) {
689 		char osname[MAXNAMELEN];
690 		dmu_objset_name(os, osname);
691 		(void) printf("replay create of %s object %llu"
692 		    " in txg %llu = %d\n",
693 		    osname, (u_longlong_t)lr->lr_doid,
694 		    (u_longlong_t)zr->zr_assign, error);
695 	}
696 
697 	return (error);
698 }
699 
700 static int
701 ztest_replay_remove(ztest_replay_t *zr, lr_remove_t *lr, boolean_t byteswap)
702 {
703 	objset_t *os = zr->zr_os;
704 	dmu_tx_t *tx;
705 	int error;
706 
707 	if (byteswap)
708 		byteswap_uint64_array(lr, sizeof (*lr));
709 
710 	tx = dmu_tx_create(os);
711 	dmu_tx_hold_free(tx, lr->lr_doid, 0, DMU_OBJECT_END);
712 	error = dmu_tx_assign(tx, zr->zr_assign);
713 	if (error) {
714 		dmu_tx_abort(tx);
715 		return (error);
716 	}
717 
718 	error = dmu_object_free(os, lr->lr_doid, tx);
719 	dmu_tx_commit(tx);
720 
721 	return (error);
722 }
723 
724 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
725 	NULL,			/* 0 no such transaction type */
726 	ztest_replay_create,	/* TX_CREATE */
727 	NULL,			/* TX_MKDIR */
728 	NULL,			/* TX_MKXATTR */
729 	NULL,			/* TX_SYMLINK */
730 	ztest_replay_remove,	/* TX_REMOVE */
731 	NULL,			/* TX_RMDIR */
732 	NULL,			/* TX_LINK */
733 	NULL,			/* TX_RENAME */
734 	NULL,			/* TX_WRITE */
735 	NULL,			/* TX_TRUNCATE */
736 	NULL,			/* TX_SETATTR */
737 	NULL,			/* TX_ACL */
738 };
739 
740 /*
741  * Verify that we can't destroy an active pool, create an existing pool,
742  * or create a pool with a bad vdev spec.
743  */
744 void
745 ztest_spa_create_destroy(ztest_args_t *za)
746 {
747 	int error;
748 	spa_t *spa;
749 	nvlist_t *nvroot;
750 
751 	/*
752 	 * Attempt to create using a bad file.
753 	 */
754 	nvroot = make_vdev_root(0, 0, 0, 1);
755 	error = spa_create("ztest_bad_file", nvroot, NULL);
756 	nvlist_free(nvroot);
757 	if (error != ENOENT)
758 		fatal(0, "spa_create(bad_file) = %d", error);
759 
760 	/*
761 	 * Attempt to create using a bad mirror.
762 	 */
763 	nvroot = make_vdev_root(0, 0, 2, 1);
764 	error = spa_create("ztest_bad_mirror", nvroot, NULL);
765 	nvlist_free(nvroot);
766 	if (error != ENOENT)
767 		fatal(0, "spa_create(bad_mirror) = %d", error);
768 
769 	/*
770 	 * Attempt to create an existing pool.  It shouldn't matter
771 	 * what's in the nvroot; we should fail with EEXIST.
772 	 */
773 	(void) rw_rdlock(&ztest_shared->zs_name_lock);
774 	nvroot = make_vdev_root(0, 0, 0, 1);
775 	error = spa_create(za->za_pool, nvroot, NULL);
776 	nvlist_free(nvroot);
777 	if (error != EEXIST)
778 		fatal(0, "spa_create(whatever) = %d", error);
779 
780 	error = spa_open(za->za_pool, &spa, FTAG);
781 	if (error)
782 		fatal(0, "spa_open() = %d", error);
783 
784 	error = spa_destroy(za->za_pool);
785 	if (error != EBUSY)
786 		fatal(0, "spa_destroy() = %d", error);
787 
788 	spa_close(spa, FTAG);
789 	(void) rw_unlock(&ztest_shared->zs_name_lock);
790 }
791 
792 /*
793  * Verify that vdev_add() works as expected.
794  */
795 void
796 ztest_vdev_add_remove(ztest_args_t *za)
797 {
798 	spa_t *spa = dmu_objset_spa(za->za_os);
799 	uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
800 	nvlist_t *nvroot;
801 	int error;
802 
803 	if (zopt_verbose >= 6)
804 		(void) printf("adding vdev\n");
805 
806 	(void) mutex_lock(&ztest_shared->zs_vdev_lock);
807 
808 	spa_config_enter(spa, RW_READER, FTAG);
809 
810 	ztest_shared->zs_vdev_primaries =
811 	    spa->spa_root_vdev->vdev_children * leaves;
812 
813 	spa_config_exit(spa, FTAG);
814 
815 	nvroot = make_vdev_root(zopt_vdev_size, zopt_raidz, zopt_mirrors, 1);
816 	error = spa_vdev_add(spa, nvroot);
817 	nvlist_free(nvroot);
818 
819 	(void) mutex_unlock(&ztest_shared->zs_vdev_lock);
820 
821 	if (error == ENOSPC)
822 		ztest_record_enospc("spa_vdev_add");
823 	else if (error != 0)
824 		fatal(0, "spa_vdev_add() = %d", error);
825 
826 	if (zopt_verbose >= 6)
827 		(void) printf("spa_vdev_add = %d, as expected\n", error);
828 }
829 
830 static vdev_t *
831 vdev_lookup_by_path(vdev_t *vd, const char *path)
832 {
833 	int c;
834 	vdev_t *mvd;
835 
836 	if (vd->vdev_path != NULL) {
837 		if (vd->vdev_wholedisk == 1) {
838 			/*
839 			 * For whole disks, the internal path has 's0', but the
840 			 * path passed in by the user doesn't.
841 			 */
842 			if (strlen(path) == strlen(vd->vdev_path) - 2 &&
843 			    strncmp(path, vd->vdev_path, strlen(path)) == 0)
844 				return (vd);
845 		} else if (strcmp(path, vd->vdev_path) == 0) {
846 			return (vd);
847 		}
848 	}
849 
850 	for (c = 0; c < vd->vdev_children; c++)
851 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
852 		    NULL)
853 			return (mvd);
854 
855 	return (NULL);
856 }
857 
858 /*
859  * Verify that we can attach and detach devices.
860  */
861 void
862 ztest_vdev_attach_detach(ztest_args_t *za)
863 {
864 	spa_t *spa = dmu_objset_spa(za->za_os);
865 	vdev_t *rvd = spa->spa_root_vdev;
866 	vdev_t *oldvd, *newvd, *pvd;
867 	nvlist_t *root, *file;
868 	uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
869 	uint64_t leaf, top;
870 	uint64_t ashift = ztest_get_ashift();
871 	size_t oldsize, newsize;
872 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
873 	int replacing;
874 	int error, expected_error;
875 	int fd;
876 
877 	(void) mutex_lock(&ztest_shared->zs_vdev_lock);
878 
879 	spa_config_enter(spa, RW_READER, FTAG);
880 
881 	/*
882 	 * Decide whether to do an attach or a replace.
883 	 */
884 	replacing = ztest_random(2);
885 
886 	/*
887 	 * Pick a random top-level vdev.
888 	 */
889 	top = ztest_random(rvd->vdev_children);
890 
891 	/*
892 	 * Pick a random leaf within it.
893 	 */
894 	leaf = ztest_random(leaves);
895 
896 	/*
897 	 * Generate the path to this leaf.  The filename will end with 'a'.
898 	 * We'll alternate replacements with a filename that ends with 'b'.
899 	 */
900 	(void) snprintf(oldpath, sizeof (oldpath),
901 	    ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf);
902 
903 	bcopy(oldpath, newpath, MAXPATHLEN);
904 
905 	/*
906 	 * If the 'a' file isn't part of the pool, the 'b' file must be.
907 	 */
908 	if (vdev_lookup_by_path(rvd, oldpath) == NULL)
909 		oldpath[strlen(oldpath) - 1] = 'b';
910 	else
911 		newpath[strlen(newpath) - 1] = 'b';
912 
913 	/*
914 	 * Now oldpath represents something that's already in the pool,
915 	 * and newpath is the thing we'll try to attach.
916 	 */
917 	oldvd = vdev_lookup_by_path(rvd, oldpath);
918 	newvd = vdev_lookup_by_path(rvd, newpath);
919 	ASSERT(oldvd != NULL);
920 	pvd = oldvd->vdev_parent;
921 
922 	/*
923 	 * Make newsize a little bigger or smaller than oldsize.
924 	 * If it's smaller, the attach should fail.
925 	 * If it's larger, and we're doing a replace,
926 	 * we should get dynamic LUN growth when we're done.
927 	 */
928 	oldsize = vdev_get_rsize(oldvd);
929 	newsize = 10 * oldsize / (9 + ztest_random(3));
930 
931 	/*
932 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
933 	 * unless it's a replace; in that case any non-replacing parent is OK.
934 	 *
935 	 * If newvd is already part of the pool, it should fail with EBUSY.
936 	 *
937 	 * If newvd is too small, it should fail with EOVERFLOW.
938 	 */
939 	if (newvd != NULL)
940 		expected_error = EBUSY;
941 	else if (pvd->vdev_ops != &vdev_mirror_ops &&
942 	    pvd->vdev_ops != &vdev_root_ops &&
943 	    (!replacing || pvd->vdev_ops == &vdev_replacing_ops))
944 		expected_error = ENOTSUP;
945 	else if (newsize < oldsize)
946 		expected_error = EOVERFLOW;
947 	else if (ashift > oldvd->vdev_top->vdev_ashift)
948 		expected_error = EDOM;
949 	else
950 		expected_error = 0;
951 
952 	/*
953 	 * If newvd isn't already part of the pool, create it.
954 	 */
955 	if (newvd == NULL) {
956 		fd = open(newpath, O_RDWR | O_CREAT | O_TRUNC, 0666);
957 		if (fd == -1)
958 			fatal(1, "can't open %s", newpath);
959 		if (ftruncate(fd, newsize) != 0)
960 			fatal(1, "can't ftruncate %s", newpath);
961 		(void) close(fd);
962 	}
963 
964 	spa_config_exit(spa, FTAG);
965 
966 	/*
967 	 * Build the nvlist describing newpath.
968 	 */
969 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
970 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
971 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, newpath) == 0);
972 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
973 
974 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
975 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
976 	VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN,
977 	    &file, 1) == 0);
978 
979 	error = spa_vdev_attach(spa, oldvd->vdev_guid, root, replacing);
980 
981 	nvlist_free(file);
982 	nvlist_free(root);
983 
984 	/*
985 	 * If our parent was the replacing vdev, but the replace completed,
986 	 * then instead of failing with ENOTSUP we may either succeed,
987 	 * fail with ENODEV, or fail with EOVERFLOW.
988 	 */
989 	if (expected_error == ENOTSUP &&
990 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
991 		expected_error = error;
992 
993 	/*
994 	 * If someone grew the LUN, the replacement may be too small.
995 	 */
996 	if (error == EOVERFLOW)
997 		expected_error = error;
998 
999 	if (error != expected_error) {
1000 		fatal(0, "attach (%s, %s, %d) returned %d, expected %d",
1001 		    oldpath, newpath, replacing, error, expected_error);
1002 	}
1003 
1004 	(void) mutex_unlock(&ztest_shared->zs_vdev_lock);
1005 }
1006 
1007 /*
1008  * Verify that dynamic LUN growth works as expected.
1009  */
1010 /* ARGSUSED */
1011 void
1012 ztest_vdev_LUN_growth(ztest_args_t *za)
1013 {
1014 	spa_t *spa = dmu_objset_spa(za->za_os);
1015 	char dev_name[MAXPATHLEN];
1016 	uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
1017 	uint64_t vdev;
1018 	size_t fsize;
1019 	int fd;
1020 
1021 	(void) mutex_lock(&ztest_shared->zs_vdev_lock);
1022 
1023 	/*
1024 	 * Pick a random leaf vdev.
1025 	 */
1026 	spa_config_enter(spa, RW_READER, FTAG);
1027 	vdev = ztest_random(spa->spa_root_vdev->vdev_children * leaves);
1028 	spa_config_exit(spa, FTAG);
1029 
1030 	(void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
1031 
1032 	if ((fd = open(dev_name, O_RDWR)) != -1) {
1033 		/*
1034 		 * Determine the size.
1035 		 */
1036 		fsize = lseek(fd, 0, SEEK_END);
1037 
1038 		/*
1039 		 * If it's less than 2x the original size, grow by around 3%.
1040 		 */
1041 		if (fsize < 2 * zopt_vdev_size) {
1042 			size_t newsize = fsize + ztest_random(fsize / 32);
1043 			(void) ftruncate(fd, newsize);
1044 			if (zopt_verbose >= 6) {
1045 				(void) printf("%s grew from %lu to %lu bytes\n",
1046 				    dev_name, (ulong_t)fsize, (ulong_t)newsize);
1047 			}
1048 		}
1049 		(void) close(fd);
1050 	}
1051 
1052 	(void) mutex_unlock(&ztest_shared->zs_vdev_lock);
1053 }
1054 
1055 /* ARGSUSED */
1056 static void
1057 ztest_create_cb(objset_t *os, void *arg, dmu_tx_t *tx)
1058 {
1059 	/*
1060 	 * Create the directory object.
1061 	 */
1062 	VERIFY(dmu_object_claim(os, ZTEST_DIROBJ,
1063 	    DMU_OT_UINT64_OTHER, ZTEST_DIROBJ_BLOCKSIZE,
1064 	    DMU_OT_UINT64_OTHER, sizeof (ztest_block_tag_t), tx) == 0);
1065 
1066 	VERIFY(zap_create_claim(os, ZTEST_MICROZAP_OBJ,
1067 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
1068 
1069 	VERIFY(zap_create_claim(os, ZTEST_FATZAP_OBJ,
1070 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
1071 }
1072 
1073 /* ARGSUSED */
1074 static void
1075 ztest_destroy_cb(char *name, void *arg)
1076 {
1077 	objset_t *os;
1078 	dmu_object_info_t doi;
1079 	int error;
1080 
1081 	/*
1082 	 * Verify that the dataset contains a directory object.
1083 	 */
1084 	error = dmu_objset_open(name, DMU_OST_OTHER,
1085 	    DS_MODE_STANDARD | DS_MODE_READONLY, &os);
1086 	ASSERT3U(error, ==, 0);
1087 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
1088 	if (error != ENOENT) {
1089 		/* We could have crashed in the middle of destroying it */
1090 		ASSERT3U(error, ==, 0);
1091 		ASSERT3U(doi.doi_type, ==, DMU_OT_UINT64_OTHER);
1092 		ASSERT3S(doi.doi_physical_blks, >=, 0);
1093 	}
1094 	dmu_objset_close(os);
1095 
1096 	/*
1097 	 * Destroy the dataset.
1098 	 */
1099 	error = dmu_objset_destroy(name);
1100 	ASSERT3U(error, ==, 0);
1101 }
1102 
1103 /*
1104  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
1105  */
1106 static uint64_t
1107 ztest_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t object, int mode)
1108 {
1109 	itx_t *itx;
1110 	lr_create_t *lr;
1111 	size_t namesize;
1112 	char name[24];
1113 
1114 	(void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object);
1115 	namesize = strlen(name) + 1;
1116 
1117 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize +
1118 	    ztest_random(ZIL_MAX_BLKSZ));
1119 	lr = (lr_create_t *)&itx->itx_lr;
1120 	bzero(lr + 1, lr->lr_common.lrc_reclen - sizeof (*lr));
1121 	lr->lr_doid = object;
1122 	lr->lr_foid = 0;
1123 	lr->lr_mode = mode;
1124 	lr->lr_uid = 0;
1125 	lr->lr_gid = 0;
1126 	lr->lr_gen = dmu_tx_get_txg(tx);
1127 	lr->lr_crtime[0] = time(NULL);
1128 	lr->lr_crtime[1] = 0;
1129 	lr->lr_rdev = 0;
1130 	bcopy(name, (char *)(lr + 1), namesize);
1131 
1132 	return (zil_itx_assign(zilog, itx, tx));
1133 }
1134 
1135 #ifndef lint
1136 static uint64_t
1137 ztest_log_remove(zilog_t *zilog, dmu_tx_t *tx, uint64_t object)
1138 {
1139 	itx_t *itx;
1140 	lr_remove_t *lr;
1141 	size_t namesize;
1142 	char name[24];
1143 
1144 	(void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object);
1145 	namesize = strlen(name) + 1;
1146 
1147 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize +
1148 	    ztest_random(8000));
1149 	lr = (lr_remove_t *)&itx->itx_lr;
1150 	lr->lr_doid = object;
1151 	bcopy(name, (char *)(lr + 1), namesize);
1152 
1153 	return (zil_itx_assign(zilog, itx, tx));
1154 }
1155 #endif /* lint */
1156 
1157 void
1158 ztest_dmu_objset_create_destroy(ztest_args_t *za)
1159 {
1160 	int error;
1161 	objset_t *os;
1162 	char name[100];
1163 	int mode, basemode, expected_error;
1164 	zilog_t *zilog;
1165 	uint64_t seq;
1166 	uint64_t objects;
1167 	ztest_replay_t zr;
1168 
1169 	(void) rw_rdlock(&ztest_shared->zs_name_lock);
1170 	(void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool,
1171 	    (u_longlong_t)za->za_instance);
1172 
1173 	basemode = DS_MODE_LEVEL(za->za_instance);
1174 	if (basemode == DS_MODE_NONE)
1175 		basemode++;
1176 
1177 	/*
1178 	 * If this dataset exists from a previous run, process its replay log
1179 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
1180 	 * (invoked from ztest_destroy_cb() below) should just throw it away.
1181 	 */
1182 	if (ztest_random(2) == 0 &&
1183 	    dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_PRIMARY, &os) == 0) {
1184 		zr.zr_os = os;
1185 		zil_replay(os, &zr, &zr.zr_assign, ztest_replay_vector, NULL);
1186 		dmu_objset_close(os);
1187 	}
1188 
1189 	/*
1190 	 * There may be an old instance of the dataset we're about to
1191 	 * create lying around from a previous run.  If so, destroy it
1192 	 * and all of its snapshots.
1193 	 */
1194 	dmu_objset_find(name, ztest_destroy_cb, NULL, DS_FIND_SNAPSHOTS);
1195 
1196 	/*
1197 	 * Verify that the destroyed dataset is no longer in the namespace.
1198 	 */
1199 	error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os);
1200 	if (error != ENOENT)
1201 		fatal(1, "dmu_objset_open(%s) found destroyed dataset %p",
1202 		    name, os);
1203 
1204 	/*
1205 	 * Verify that we can create a new dataset.
1206 	 */
1207 	error = dmu_objset_create(name, DMU_OST_OTHER, NULL, ztest_create_cb,
1208 	    NULL);
1209 	if (error) {
1210 		if (error == ENOSPC) {
1211 			ztest_record_enospc("dmu_objset_create");
1212 			(void) rw_unlock(&ztest_shared->zs_name_lock);
1213 			return;
1214 		}
1215 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
1216 	}
1217 
1218 	error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os);
1219 	if (error) {
1220 		fatal(0, "dmu_objset_open(%s) = %d", name, error);
1221 	}
1222 
1223 	/*
1224 	 * Open the intent log for it.
1225 	 */
1226 	zilog = zil_open(os, NULL);
1227 
1228 	/*
1229 	 * Put a random number of objects in there.
1230 	 */
1231 	objects = ztest_random(20);
1232 	seq = 0;
1233 	while (objects-- != 0) {
1234 		uint64_t object;
1235 		dmu_tx_t *tx = dmu_tx_create(os);
1236 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, sizeof (name));
1237 		error = dmu_tx_assign(tx, TXG_WAIT);
1238 		if (error) {
1239 			dmu_tx_abort(tx);
1240 		} else {
1241 			object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1242 			    DMU_OT_NONE, 0, tx);
1243 			ztest_set_random_blocksize(os, object, tx);
1244 			seq = ztest_log_create(zilog, tx, object,
1245 			    DMU_OT_UINT64_OTHER);
1246 			dmu_write(os, object, 0, sizeof (name), name, tx);
1247 			dmu_tx_commit(tx);
1248 		}
1249 		if (ztest_random(5) == 0) {
1250 			zil_commit(zilog, seq, FSYNC);
1251 		}
1252 		if (ztest_random(100) == 0) {
1253 			error = zil_suspend(zilog);
1254 			if (error == 0) {
1255 				zil_resume(zilog);
1256 			}
1257 		}
1258 	}
1259 
1260 	/*
1261 	 * Verify that we cannot create an existing dataset.
1262 	 */
1263 	error = dmu_objset_create(name, DMU_OST_OTHER, NULL, NULL, NULL);
1264 	if (error != EEXIST)
1265 		fatal(0, "created existing dataset, error = %d", error);
1266 
1267 	/*
1268 	 * Verify that multiple dataset opens are allowed, but only when
1269 	 * the new access mode is compatible with the base mode.
1270 	 * We use a mixture of typed and typeless opens, and when the
1271 	 * open succeeds, verify that the discovered type is correct.
1272 	 */
1273 	for (mode = DS_MODE_STANDARD; mode < DS_MODE_LEVELS; mode++) {
1274 		objset_t *os2;
1275 		error = dmu_objset_open(name, DMU_OST_OTHER, mode, &os2);
1276 		expected_error = (basemode + mode < DS_MODE_LEVELS) ? 0 : EBUSY;
1277 		if (error != expected_error)
1278 			fatal(0, "dmu_objset_open('%s') = %d, expected %d",
1279 			    name, error, expected_error);
1280 		if (error == 0)
1281 			dmu_objset_close(os2);
1282 	}
1283 
1284 	zil_close(zilog);
1285 	dmu_objset_close(os);
1286 
1287 	error = dmu_objset_destroy(name);
1288 	if (error)
1289 		fatal(0, "dmu_objset_destroy(%s) = %d", name, error);
1290 
1291 	(void) rw_unlock(&ztest_shared->zs_name_lock);
1292 }
1293 
1294 /*
1295  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
1296  */
1297 void
1298 ztest_dmu_snapshot_create_destroy(ztest_args_t *za)
1299 {
1300 	int error;
1301 	objset_t *os = za->za_os;
1302 	char snapname[100];
1303 	char osname[MAXNAMELEN];
1304 
1305 	(void) rw_rdlock(&ztest_shared->zs_name_lock);
1306 	dmu_objset_name(os, osname);
1307 	(void) snprintf(snapname, 100, "%s@%llu", osname,
1308 	    (u_longlong_t)za->za_instance);
1309 
1310 	error = dmu_objset_destroy(snapname);
1311 	if (error != 0 && error != ENOENT)
1312 		fatal(0, "dmu_objset_destroy() = %d", error);
1313 	error = dmu_objset_create(snapname, DMU_OST_OTHER, NULL, NULL, NULL);
1314 	if (error == ENOSPC)
1315 		ztest_record_enospc("dmu_take_snapshot");
1316 	else if (error != 0 && error != EEXIST)
1317 		fatal(0, "dmu_take_snapshot() = %d", error);
1318 	(void) rw_unlock(&ztest_shared->zs_name_lock);
1319 }
1320 
1321 #define	ZTEST_TRAVERSE_BLOCKS	1000
1322 
1323 static int
1324 ztest_blk_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
1325 {
1326 	ztest_args_t *za = arg;
1327 	zbookmark_t *zb = &bc->bc_bookmark;
1328 	blkptr_t *bp = &bc->bc_blkptr;
1329 	dnode_phys_t *dnp = bc->bc_dnode;
1330 	traverse_handle_t *th = za->za_th;
1331 	uint64_t size = BP_GET_LSIZE(bp);
1332 
1333 	/*
1334 	 * Level -1 indicates the objset_phys_t or something in its intent log.
1335 	 */
1336 	if (zb->zb_level == -1) {
1337 		if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
1338 			ASSERT3U(zb->zb_object, ==, 0);
1339 			ASSERT3U(zb->zb_blkid, ==, 0);
1340 			ASSERT3U(size, ==, sizeof (objset_phys_t));
1341 			za->za_zil_seq = 0;
1342 		} else if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) {
1343 			ASSERT3U(zb->zb_object, ==, 0);
1344 			ASSERT3U(zb->zb_blkid, >, za->za_zil_seq);
1345 			za->za_zil_seq = zb->zb_blkid;
1346 		} else {
1347 			ASSERT3U(zb->zb_object, !=, 0);	/* lr_write_t */
1348 		}
1349 
1350 		return (0);
1351 	}
1352 
1353 	ASSERT(dnp != NULL);
1354 
1355 	if (bc->bc_errno)
1356 		return (ERESTART);
1357 
1358 	/*
1359 	 * Once in a while, abort the traverse.   We only do this to odd
1360 	 * instance numbers to ensure that even ones can run to completion.
1361 	 */
1362 	if ((za->za_instance & 1) && ztest_random(10000) == 0)
1363 		return (EINTR);
1364 
1365 	if (bp->blk_birth == 0) {
1366 		ASSERT(th->th_advance & ADVANCE_HOLES);
1367 		return (0);
1368 	}
1369 
1370 	if (zb->zb_level == 0 && !(th->th_advance & ADVANCE_DATA) &&
1371 	    bc == &th->th_cache[ZB_DN_CACHE][0]) {
1372 		ASSERT(bc->bc_data == NULL);
1373 		return (0);
1374 	}
1375 
1376 	ASSERT(bc->bc_data != NULL);
1377 
1378 	/*
1379 	 * This is an expensive question, so don't ask it too often.
1380 	 */
1381 	if (((za->za_random ^ th->th_callbacks) & 0xff) == 0) {
1382 		void *xbuf = umem_alloc(size, UMEM_NOFAIL);
1383 		if (arc_tryread(spa, bp, xbuf) == 0) {
1384 			ASSERT(bcmp(bc->bc_data, xbuf, size) == 0);
1385 		}
1386 		umem_free(xbuf, size);
1387 	}
1388 
1389 	if (zb->zb_level > 0) {
1390 		ASSERT3U(size, ==, 1ULL << dnp->dn_indblkshift);
1391 		return (0);
1392 	}
1393 
1394 	ASSERT(zb->zb_level == 0);
1395 	ASSERT3U(size, ==, dnp->dn_datablkszsec << DEV_BSHIFT);
1396 
1397 	return (0);
1398 }
1399 
1400 /*
1401  * Verify that live pool traversal works.
1402  */
1403 void
1404 ztest_traverse(ztest_args_t *za)
1405 {
1406 	spa_t *spa = dmu_objset_spa(za->za_os);
1407 	traverse_handle_t *th = za->za_th;
1408 	int rc, advance;
1409 	uint64_t cbstart, cblimit;
1410 
1411 	if (th == NULL) {
1412 		advance = 0;
1413 
1414 		if (ztest_random(2) == 0)
1415 			advance |= ADVANCE_PRE;
1416 
1417 		if (ztest_random(2) == 0)
1418 			advance |= ADVANCE_PRUNE;
1419 
1420 		if (ztest_random(2) == 0)
1421 			advance |= ADVANCE_DATA;
1422 
1423 		if (ztest_random(2) == 0)
1424 			advance |= ADVANCE_HOLES;
1425 
1426 		if (ztest_random(2) == 0)
1427 			advance |= ADVANCE_ZIL;
1428 
1429 		th = za->za_th = traverse_init(spa, ztest_blk_cb, za, advance,
1430 		    ZIO_FLAG_CANFAIL);
1431 
1432 		traverse_add_pool(th, 0, -1ULL);
1433 	}
1434 
1435 	advance = th->th_advance;
1436 	cbstart = th->th_callbacks;
1437 	cblimit = cbstart + ((advance & ADVANCE_DATA) ? 100 : 1000);
1438 
1439 	while ((rc = traverse_more(th)) == EAGAIN && th->th_callbacks < cblimit)
1440 		continue;
1441 
1442 	if (zopt_verbose >= 5)
1443 		(void) printf("traverse %s%s%s%s %llu blocks to "
1444 		    "<%llu, %llu, %lld, %llx>%s\n",
1445 		    (advance & ADVANCE_PRE) ? "pre" : "post",
1446 		    (advance & ADVANCE_PRUNE) ? "|prune" : "",
1447 		    (advance & ADVANCE_DATA) ? "|data" : "",
1448 		    (advance & ADVANCE_HOLES) ? "|holes" : "",
1449 		    (u_longlong_t)(th->th_callbacks - cbstart),
1450 		    (u_longlong_t)th->th_lastcb.zb_objset,
1451 		    (u_longlong_t)th->th_lastcb.zb_object,
1452 		    (u_longlong_t)th->th_lastcb.zb_level,
1453 		    (u_longlong_t)th->th_lastcb.zb_blkid,
1454 		    rc == 0 ? " [done]" :
1455 		    rc == EINTR ? " [aborted]" :
1456 		    rc == EAGAIN ? "" :
1457 		    strerror(rc));
1458 
1459 	if (rc != EAGAIN) {
1460 		if (rc != 0 && rc != EINTR)
1461 			fatal(0, "traverse_more(%p) = %d", th, rc);
1462 		traverse_fini(th);
1463 		za->za_th = NULL;
1464 	}
1465 }
1466 
1467 /*
1468  * Verify that dmu_object_{alloc,free} work as expected.
1469  */
1470 void
1471 ztest_dmu_object_alloc_free(ztest_args_t *za)
1472 {
1473 	objset_t *os = za->za_os;
1474 	dmu_buf_t *db;
1475 	dmu_tx_t *tx;
1476 	uint64_t batchobj, object, batchsize, endoff, temp;
1477 	int b, c, error, bonuslen;
1478 	dmu_object_info_t doi;
1479 	char osname[MAXNAMELEN];
1480 
1481 	dmu_objset_name(os, osname);
1482 
1483 	endoff = -8ULL;
1484 	batchsize = 2;
1485 
1486 	/*
1487 	 * Create a batch object if necessary, and record it in the directory.
1488 	 */
1489 	VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
1490 	    sizeof (uint64_t), &batchobj));
1491 	if (batchobj == 0) {
1492 		tx = dmu_tx_create(os);
1493 		dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff,
1494 		    sizeof (uint64_t));
1495 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1496 		error = dmu_tx_assign(tx, TXG_WAIT);
1497 		if (error) {
1498 			ztest_record_enospc("create a batch object");
1499 			dmu_tx_abort(tx);
1500 			return;
1501 		}
1502 		batchobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1503 		    DMU_OT_NONE, 0, tx);
1504 		ztest_set_random_blocksize(os, batchobj, tx);
1505 		dmu_write(os, ZTEST_DIROBJ, za->za_diroff,
1506 		    sizeof (uint64_t), &batchobj, tx);
1507 		dmu_tx_commit(tx);
1508 	}
1509 
1510 	/*
1511 	 * Destroy the previous batch of objects.
1512 	 */
1513 	for (b = 0; b < batchsize; b++) {
1514 		VERIFY(0 == dmu_read(os, batchobj, b * sizeof (uint64_t),
1515 		    sizeof (uint64_t), &object));
1516 		if (object == 0)
1517 			continue;
1518 		/*
1519 		 * Read and validate contents.
1520 		 * We expect the nth byte of the bonus buffer to be n.
1521 		 */
1522 		VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db));
1523 
1524 		dmu_object_info_from_db(db, &doi);
1525 		ASSERT(doi.doi_type == DMU_OT_UINT64_OTHER);
1526 		ASSERT(doi.doi_bonus_type == DMU_OT_PLAIN_OTHER);
1527 		ASSERT3S(doi.doi_physical_blks, >=, 0);
1528 
1529 		bonuslen = db->db_size;
1530 
1531 		for (c = 0; c < bonuslen; c++) {
1532 			if (((uint8_t *)db->db_data)[c] !=
1533 			    (uint8_t)(c + bonuslen)) {
1534 				fatal(0,
1535 				    "bad bonus: %s, obj %llu, off %d: %u != %u",
1536 				    osname, object, c,
1537 				    ((uint8_t *)db->db_data)[c],
1538 				    (uint8_t)(c + bonuslen));
1539 			}
1540 		}
1541 
1542 		dmu_buf_rele(db, FTAG);
1543 
1544 		/*
1545 		 * We expect the word at endoff to be our object number.
1546 		 */
1547 		VERIFY(0 == dmu_read(os, object, endoff,
1548 		    sizeof (uint64_t), &temp));
1549 
1550 		if (temp != object) {
1551 			fatal(0, "bad data in %s, got %llu, expected %llu",
1552 			    osname, temp, object);
1553 		}
1554 
1555 		/*
1556 		 * Destroy old object and clear batch entry.
1557 		 */
1558 		tx = dmu_tx_create(os);
1559 		dmu_tx_hold_write(tx, batchobj,
1560 		    b * sizeof (uint64_t), sizeof (uint64_t));
1561 		dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1562 		error = dmu_tx_assign(tx, TXG_WAIT);
1563 		if (error) {
1564 			ztest_record_enospc("free object");
1565 			dmu_tx_abort(tx);
1566 			return;
1567 		}
1568 		error = dmu_object_free(os, object, tx);
1569 		if (error) {
1570 			fatal(0, "dmu_object_free('%s', %llu) = %d",
1571 			    osname, object, error);
1572 		}
1573 		object = 0;
1574 
1575 		dmu_object_set_checksum(os, batchobj,
1576 		    ztest_random_checksum(), tx);
1577 		dmu_object_set_compress(os, batchobj,
1578 		    ztest_random_compress(), tx);
1579 
1580 		dmu_write(os, batchobj, b * sizeof (uint64_t),
1581 		    sizeof (uint64_t), &object, tx);
1582 
1583 		dmu_tx_commit(tx);
1584 	}
1585 
1586 	/*
1587 	 * Before creating the new batch of objects, generate a bunch of churn.
1588 	 */
1589 	for (b = ztest_random(100); b > 0; b--) {
1590 		tx = dmu_tx_create(os);
1591 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1592 		error = dmu_tx_assign(tx, TXG_WAIT);
1593 		if (error) {
1594 			ztest_record_enospc("churn objects");
1595 			dmu_tx_abort(tx);
1596 			return;
1597 		}
1598 		object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1599 		    DMU_OT_NONE, 0, tx);
1600 		ztest_set_random_blocksize(os, object, tx);
1601 		error = dmu_object_free(os, object, tx);
1602 		if (error) {
1603 			fatal(0, "dmu_object_free('%s', %llu) = %d",
1604 			    osname, object, error);
1605 		}
1606 		dmu_tx_commit(tx);
1607 	}
1608 
1609 	/*
1610 	 * Create a new batch of objects with randomly chosen
1611 	 * blocksizes and record them in the batch directory.
1612 	 */
1613 	for (b = 0; b < batchsize; b++) {
1614 		uint32_t va_blksize;
1615 		u_longlong_t va_nblocks;
1616 
1617 		tx = dmu_tx_create(os);
1618 		dmu_tx_hold_write(tx, batchobj, b * sizeof (uint64_t),
1619 		    sizeof (uint64_t));
1620 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1621 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, endoff,
1622 		    sizeof (uint64_t));
1623 		error = dmu_tx_assign(tx, TXG_WAIT);
1624 		if (error) {
1625 			ztest_record_enospc("create batchobj");
1626 			dmu_tx_abort(tx);
1627 			return;
1628 		}
1629 		bonuslen = (int)ztest_random(dmu_bonus_max()) + 1;
1630 
1631 		object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1632 		    DMU_OT_PLAIN_OTHER, bonuslen, tx);
1633 
1634 		ztest_set_random_blocksize(os, object, tx);
1635 
1636 		dmu_object_set_checksum(os, object,
1637 		    ztest_random_checksum(), tx);
1638 		dmu_object_set_compress(os, object,
1639 		    ztest_random_compress(), tx);
1640 
1641 		dmu_write(os, batchobj, b * sizeof (uint64_t),
1642 		    sizeof (uint64_t), &object, tx);
1643 
1644 		/*
1645 		 * Write to both the bonus buffer and the regular data.
1646 		 */
1647 		VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db));
1648 		ASSERT3U(bonuslen, ==, db->db_size);
1649 
1650 		dmu_object_size_from_db(db, &va_blksize, &va_nblocks);
1651 		ASSERT3S(va_nblocks, >=, 0);
1652 
1653 		dmu_buf_will_dirty(db, tx);
1654 
1655 		/*
1656 		 * See comments above regarding the contents of
1657 		 * the bonus buffer and the word at endoff.
1658 		 */
1659 		for (c = 0; c < db->db_size; c++)
1660 			((uint8_t *)db->db_data)[c] = (uint8_t)(c + bonuslen);
1661 
1662 		dmu_buf_rele(db, FTAG);
1663 
1664 		/*
1665 		 * Write to a large offset to increase indirection.
1666 		 */
1667 		dmu_write(os, object, endoff, sizeof (uint64_t), &object, tx);
1668 
1669 		dmu_tx_commit(tx);
1670 	}
1671 }
1672 
1673 /*
1674  * Verify that dmu_{read,write} work as expected.
1675  */
1676 typedef struct bufwad {
1677 	uint64_t	bw_index;
1678 	uint64_t	bw_txg;
1679 	uint64_t	bw_data;
1680 } bufwad_t;
1681 
1682 typedef struct dmu_read_write_dir {
1683 	uint64_t	dd_packobj;
1684 	uint64_t	dd_bigobj;
1685 	uint64_t	dd_chunk;
1686 } dmu_read_write_dir_t;
1687 
1688 void
1689 ztest_dmu_read_write(ztest_args_t *za)
1690 {
1691 	objset_t *os = za->za_os;
1692 	dmu_read_write_dir_t dd;
1693 	dmu_tx_t *tx;
1694 	int i, freeit, error;
1695 	uint64_t n, s, txg;
1696 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
1697 	uint64_t packoff, packsize, bigoff, bigsize;
1698 	uint64_t regions = 997;
1699 	uint64_t stride = 123456789ULL;
1700 	uint64_t width = 40;
1701 	int free_percent = 5;
1702 
1703 	/*
1704 	 * This test uses two objects, packobj and bigobj, that are always
1705 	 * updated together (i.e. in the same tx) so that their contents are
1706 	 * in sync and can be compared.  Their contents relate to each other
1707 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
1708 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
1709 	 * for any index n, there are three bufwads that should be identical:
1710 	 *
1711 	 *	packobj, at offset n * sizeof (bufwad_t)
1712 	 *	bigobj, at the head of the nth chunk
1713 	 *	bigobj, at the tail of the nth chunk
1714 	 *
1715 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
1716 	 * and it doesn't have any relation to the object blocksize.
1717 	 * The only requirement is that it can hold at least two bufwads.
1718 	 *
1719 	 * Normally, we write the bufwad to each of these locations.
1720 	 * However, free_percent of the time we instead write zeroes to
1721 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
1722 	 * bigobj to packobj, we can verify that the DMU is correctly
1723 	 * tracking which parts of an object are allocated and free,
1724 	 * and that the contents of the allocated blocks are correct.
1725 	 */
1726 
1727 	/*
1728 	 * Read the directory info.  If it's the first time, set things up.
1729 	 */
1730 	VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
1731 	    sizeof (dd), &dd));
1732 	if (dd.dd_chunk == 0) {
1733 		ASSERT(dd.dd_packobj == 0);
1734 		ASSERT(dd.dd_bigobj == 0);
1735 		tx = dmu_tx_create(os);
1736 		dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (dd));
1737 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1738 		error = dmu_tx_assign(tx, TXG_WAIT);
1739 		if (error) {
1740 			ztest_record_enospc("create r/w directory");
1741 			dmu_tx_abort(tx);
1742 			return;
1743 		}
1744 
1745 		dd.dd_packobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1746 		    DMU_OT_NONE, 0, tx);
1747 		dd.dd_bigobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1748 		    DMU_OT_NONE, 0, tx);
1749 		dd.dd_chunk = (1000 + ztest_random(1000)) * sizeof (uint64_t);
1750 
1751 		ztest_set_random_blocksize(os, dd.dd_packobj, tx);
1752 		ztest_set_random_blocksize(os, dd.dd_bigobj, tx);
1753 
1754 		dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (dd), &dd,
1755 		    tx);
1756 		dmu_tx_commit(tx);
1757 	}
1758 
1759 	/*
1760 	 * Prefetch a random chunk of the big object.
1761 	 * Our aim here is to get some async reads in flight
1762 	 * for blocks that we may free below; the DMU should
1763 	 * handle this race correctly.
1764 	 */
1765 	n = ztest_random(regions) * stride + ztest_random(width);
1766 	s = 1 + ztest_random(2 * width - 1);
1767 	dmu_prefetch(os, dd.dd_bigobj, n * dd.dd_chunk, s * dd.dd_chunk);
1768 
1769 	/*
1770 	 * Pick a random index and compute the offsets into packobj and bigobj.
1771 	 */
1772 	n = ztest_random(regions) * stride + ztest_random(width);
1773 	s = 1 + ztest_random(width - 1);
1774 
1775 	packoff = n * sizeof (bufwad_t);
1776 	packsize = s * sizeof (bufwad_t);
1777 
1778 	bigoff = n * dd.dd_chunk;
1779 	bigsize = s * dd.dd_chunk;
1780 
1781 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
1782 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
1783 
1784 	/*
1785 	 * free_percent of the time, free a range of bigobj rather than
1786 	 * overwriting it.
1787 	 */
1788 	freeit = (ztest_random(100) < free_percent);
1789 
1790 	/*
1791 	 * Read the current contents of our objects.
1792 	 */
1793 	error = dmu_read(os, dd.dd_packobj, packoff, packsize, packbuf);
1794 	ASSERT3U(error, ==, 0);
1795 	error = dmu_read(os, dd.dd_bigobj, bigoff, bigsize, bigbuf);
1796 	ASSERT3U(error, ==, 0);
1797 
1798 	/*
1799 	 * Get a tx for the mods to both packobj and bigobj.
1800 	 */
1801 	tx = dmu_tx_create(os);
1802 
1803 	dmu_tx_hold_write(tx, dd.dd_packobj, packoff, packsize);
1804 
1805 	if (freeit)
1806 		dmu_tx_hold_free(tx, dd.dd_bigobj, bigoff, bigsize);
1807 	else
1808 		dmu_tx_hold_write(tx, dd.dd_bigobj, bigoff, bigsize);
1809 
1810 	error = dmu_tx_assign(tx, TXG_WAIT);
1811 
1812 	if (error) {
1813 		ztest_record_enospc("dmu r/w range");
1814 		dmu_tx_abort(tx);
1815 		umem_free(packbuf, packsize);
1816 		umem_free(bigbuf, bigsize);
1817 		return;
1818 	}
1819 
1820 	txg = dmu_tx_get_txg(tx);
1821 
1822 	/*
1823 	 * For each index from n to n + s, verify that the existing bufwad
1824 	 * in packobj matches the bufwads at the head and tail of the
1825 	 * corresponding chunk in bigobj.  Then update all three bufwads
1826 	 * with the new values we want to write out.
1827 	 */
1828 	for (i = 0; i < s; i++) {
1829 		/* LINTED */
1830 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
1831 		/* LINTED */
1832 		bigH = (bufwad_t *)((char *)bigbuf + i * dd.dd_chunk);
1833 		/* LINTED */
1834 		bigT = (bufwad_t *)((char *)bigH + dd.dd_chunk) - 1;
1835 
1836 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
1837 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
1838 
1839 		if (pack->bw_txg > txg)
1840 			fatal(0, "future leak: got %llx, open txg is %llx",
1841 			    pack->bw_txg, txg);
1842 
1843 		if (pack->bw_data != 0 && pack->bw_index != n + i)
1844 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
1845 			    pack->bw_index, n, i);
1846 
1847 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
1848 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
1849 
1850 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
1851 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
1852 
1853 		if (freeit) {
1854 			bzero(pack, sizeof (bufwad_t));
1855 		} else {
1856 			pack->bw_index = n + i;
1857 			pack->bw_txg = txg;
1858 			pack->bw_data = 1 + ztest_random(-2ULL);
1859 		}
1860 		*bigH = *pack;
1861 		*bigT = *pack;
1862 	}
1863 
1864 	/*
1865 	 * We've verified all the old bufwads, and made new ones.
1866 	 * Now write them out.
1867 	 */
1868 	dmu_write(os, dd.dd_packobj, packoff, packsize, packbuf, tx);
1869 
1870 	if (freeit) {
1871 		if (zopt_verbose >= 6) {
1872 			(void) printf("freeing offset %llx size %llx"
1873 			    " txg %llx\n",
1874 			    (u_longlong_t)bigoff,
1875 			    (u_longlong_t)bigsize,
1876 			    (u_longlong_t)txg);
1877 		}
1878 		VERIFY(0 == dmu_free_range(os, dd.dd_bigobj, bigoff,
1879 		    bigsize, tx));
1880 	} else {
1881 		if (zopt_verbose >= 6) {
1882 			(void) printf("writing offset %llx size %llx"
1883 			    " txg %llx\n",
1884 			    (u_longlong_t)bigoff,
1885 			    (u_longlong_t)bigsize,
1886 			    (u_longlong_t)txg);
1887 		}
1888 		dmu_write(os, dd.dd_bigobj, bigoff, bigsize, bigbuf, tx);
1889 	}
1890 
1891 	dmu_tx_commit(tx);
1892 
1893 	/*
1894 	 * Sanity check the stuff we just wrote.
1895 	 */
1896 	{
1897 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
1898 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
1899 
1900 		VERIFY(0 == dmu_read(os, dd.dd_packobj, packoff,
1901 		    packsize, packcheck));
1902 		VERIFY(0 == dmu_read(os, dd.dd_bigobj, bigoff,
1903 		    bigsize, bigcheck));
1904 
1905 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
1906 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
1907 
1908 		umem_free(packcheck, packsize);
1909 		umem_free(bigcheck, bigsize);
1910 	}
1911 
1912 	umem_free(packbuf, packsize);
1913 	umem_free(bigbuf, bigsize);
1914 }
1915 
1916 void
1917 ztest_dmu_write_parallel(ztest_args_t *za)
1918 {
1919 	objset_t *os = za->za_os;
1920 	dmu_tx_t *tx;
1921 	dmu_buf_t *db;
1922 	int i, b, error, do_free, bs;
1923 	uint64_t off, txg_how, txg;
1924 	mutex_t *lp;
1925 	char osname[MAXNAMELEN];
1926 	char iobuf[SPA_MAXBLOCKSIZE];
1927 	ztest_block_tag_t rbt, wbt;
1928 
1929 	dmu_objset_name(os, osname);
1930 	bs = ZTEST_DIROBJ_BLOCKSIZE;
1931 
1932 	/*
1933 	 * Have multiple threads write to large offsets in ZTEST_DIROBJ
1934 	 * to verify that having multiple threads writing to the same object
1935 	 * in parallel doesn't cause any trouble.
1936 	 * Also do parallel writes to the bonus buffer on occasion.
1937 	 */
1938 	for (i = 0; i < 50; i++) {
1939 		b = ztest_random(ZTEST_SYNC_LOCKS);
1940 		lp = &ztest_shared->zs_sync_lock[b];
1941 
1942 		do_free = (ztest_random(4) == 0);
1943 
1944 		off = za->za_diroff_shared + ((uint64_t)b << SPA_MAXBLOCKSHIFT);
1945 
1946 		if (ztest_random(4) == 0) {
1947 			/*
1948 			 * Do the bonus buffer instead of a regular block.
1949 			 */
1950 			do_free = 0;
1951 			off = -1ULL;
1952 		}
1953 
1954 		tx = dmu_tx_create(os);
1955 
1956 		if (off == -1ULL)
1957 			dmu_tx_hold_bonus(tx, ZTEST_DIROBJ);
1958 		else if (do_free)
1959 			dmu_tx_hold_free(tx, ZTEST_DIROBJ, off, bs);
1960 		else
1961 			dmu_tx_hold_write(tx, ZTEST_DIROBJ, off, bs);
1962 
1963 		txg_how = ztest_random(2) == 0 ? TXG_WAIT : TXG_NOWAIT;
1964 		error = dmu_tx_assign(tx, txg_how);
1965 		if (error) {
1966 			if (error == ERESTART) {
1967 				ASSERT(txg_how == TXG_NOWAIT);
1968 				dmu_tx_wait(tx);
1969 				dmu_tx_abort(tx);
1970 				continue;
1971 			}
1972 			dmu_tx_abort(tx);
1973 			ztest_record_enospc("dmu write parallel");
1974 			return;
1975 		}
1976 		txg = dmu_tx_get_txg(tx);
1977 
1978 		if (do_free) {
1979 			(void) mutex_lock(lp);
1980 			VERIFY(0 == dmu_free_range(os, ZTEST_DIROBJ, off,
1981 			    bs, tx));
1982 			(void) mutex_unlock(lp);
1983 			dmu_tx_commit(tx);
1984 			continue;
1985 		}
1986 
1987 		wbt.bt_objset = dmu_objset_id(os);
1988 		wbt.bt_object = ZTEST_DIROBJ;
1989 		wbt.bt_offset = off;
1990 		wbt.bt_txg = txg;
1991 		wbt.bt_thread = za->za_instance;
1992 
1993 		if (off == -1ULL) {
1994 			wbt.bt_seq = 0;
1995 			VERIFY(0 == dmu_bonus_hold(os, ZTEST_DIROBJ,
1996 			    FTAG, &db));
1997 			ASSERT3U(db->db_size, ==, sizeof (wbt));
1998 			bcopy(db->db_data, &rbt, db->db_size);
1999 			if (rbt.bt_objset != 0) {
2000 				ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset);
2001 				ASSERT3U(rbt.bt_object, ==, wbt.bt_object);
2002 				ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset);
2003 				ASSERT3U(rbt.bt_txg, <=, wbt.bt_txg);
2004 			}
2005 			dmu_buf_will_dirty(db, tx);
2006 			bcopy(&wbt, db->db_data, db->db_size);
2007 			dmu_buf_rele(db, FTAG);
2008 			dmu_tx_commit(tx);
2009 			continue;
2010 		}
2011 
2012 		(void) mutex_lock(lp);
2013 
2014 		wbt.bt_seq = ztest_shared->zs_seq[b]++;
2015 
2016 		dmu_write(os, ZTEST_DIROBJ, off, sizeof (wbt), &wbt, tx);
2017 
2018 		(void) mutex_unlock(lp);
2019 
2020 		if (ztest_random(100) == 0)
2021 			(void) poll(NULL, 0, 1); /* open dn_notxholds window */
2022 
2023 		dmu_tx_commit(tx);
2024 
2025 		if (ztest_random(1000) == 0)
2026 			txg_wait_synced(dmu_objset_pool(os), txg);
2027 
2028 		if (ztest_random(2) == 0) {
2029 			blkptr_t blk = { 0 };
2030 			uint64_t blkoff;
2031 			zbookmark_t zb;
2032 
2033 			txg_suspend(dmu_objset_pool(os));
2034 			(void) mutex_lock(lp);
2035 			error = dmu_sync(os, ZTEST_DIROBJ, off, &blkoff, &blk,
2036 			    txg);
2037 			(void) mutex_unlock(lp);
2038 			if (error) {
2039 				txg_resume(dmu_objset_pool(os));
2040 				dprintf("dmu_sync(%s, %d, %llx) = %d\n",
2041 				    osname, ZTEST_DIROBJ, off, error);
2042 				continue;
2043 			}
2044 
2045 			if (blk.blk_birth == 0)	{	/* concurrent free */
2046 				txg_resume(dmu_objset_pool(os));
2047 				continue;
2048 			}
2049 
2050 			ASSERT(blk.blk_fill == 1);
2051 			ASSERT3U(BP_GET_TYPE(&blk), ==, DMU_OT_UINT64_OTHER);
2052 			ASSERT3U(BP_GET_LEVEL(&blk), ==, 0);
2053 			ASSERT3U(BP_GET_LSIZE(&blk), ==, bs);
2054 
2055 			/*
2056 			 * Read the block that dmu_sync() returned to
2057 			 * make sure its contents match what we wrote.
2058 			 * We do this while still txg_suspend()ed to ensure
2059 			 * that the block can't be reused before we read it.
2060 			 */
2061 			zb.zb_objset = dmu_objset_id(os);
2062 			zb.zb_object = ZTEST_DIROBJ;
2063 			zb.zb_level = 0;
2064 			zb.zb_blkid = off / bs;
2065 			error = zio_wait(zio_read(NULL, dmu_objset_spa(os),
2066 			    &blk, iobuf, bs, NULL, NULL,
2067 			    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_MUSTSUCCEED, &zb));
2068 			ASSERT(error == 0);
2069 
2070 			txg_resume(dmu_objset_pool(os));
2071 
2072 			bcopy(&iobuf[blkoff], &rbt, sizeof (rbt));
2073 
2074 			if (rbt.bt_objset == 0)		/* concurrent free */
2075 				continue;
2076 
2077 			ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset);
2078 			ASSERT3U(rbt.bt_object, ==, wbt.bt_object);
2079 			ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset);
2080 
2081 			/*
2082 			 * The semantic of dmu_sync() is that we always
2083 			 * push the most recent version of the data,
2084 			 * so in the face of concurrent updates we may
2085 			 * see a newer version of the block.  That's OK.
2086 			 */
2087 			ASSERT3U(rbt.bt_txg, >=, wbt.bt_txg);
2088 			if (rbt.bt_thread == wbt.bt_thread)
2089 				ASSERT3U(rbt.bt_seq, ==, wbt.bt_seq);
2090 			else
2091 				ASSERT3U(rbt.bt_seq, >, wbt.bt_seq);
2092 		}
2093 	}
2094 }
2095 
2096 /*
2097  * Verify that zap_{create,destroy,add,remove,update} work as expected.
2098  */
2099 #define	ZTEST_ZAP_MIN_INTS	1
2100 #define	ZTEST_ZAP_MAX_INTS	4
2101 #define	ZTEST_ZAP_MAX_PROPS	1000
2102 
2103 void
2104 ztest_zap(ztest_args_t *za)
2105 {
2106 	objset_t *os = za->za_os;
2107 	uint64_t object;
2108 	uint64_t txg, last_txg;
2109 	uint64_t value[ZTEST_ZAP_MAX_INTS];
2110 	uint64_t zl_ints, zl_intsize, prop;
2111 	int i, ints;
2112 	int iters = 100;
2113 	dmu_tx_t *tx;
2114 	char propname[100], txgname[100];
2115 	int error;
2116 	char osname[MAXNAMELEN];
2117 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
2118 
2119 	dmu_objset_name(os, osname);
2120 
2121 	/*
2122 	 * Create a new object if necessary, and record it in the directory.
2123 	 */
2124 	VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
2125 	    sizeof (uint64_t), &object));
2126 
2127 	if (object == 0) {
2128 		tx = dmu_tx_create(os);
2129 		dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff,
2130 		    sizeof (uint64_t));
2131 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL);
2132 		error = dmu_tx_assign(tx, TXG_WAIT);
2133 		if (error) {
2134 			ztest_record_enospc("create zap test obj");
2135 			dmu_tx_abort(tx);
2136 			return;
2137 		}
2138 		object = zap_create(os, DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx);
2139 		if (error) {
2140 			fatal(0, "zap_create('%s', %llu) = %d",
2141 			    osname, object, error);
2142 		}
2143 		ASSERT(object != 0);
2144 		dmu_write(os, ZTEST_DIROBJ, za->za_diroff,
2145 		    sizeof (uint64_t), &object, tx);
2146 		/*
2147 		 * Generate a known hash collision, and verify that
2148 		 * we can lookup and remove both entries.
2149 		 */
2150 		for (i = 0; i < 2; i++) {
2151 			value[i] = i;
2152 			error = zap_add(os, object, hc[i], sizeof (uint64_t),
2153 			    1, &value[i], tx);
2154 			ASSERT3U(error, ==, 0);
2155 		}
2156 		for (i = 0; i < 2; i++) {
2157 			error = zap_add(os, object, hc[i], sizeof (uint64_t),
2158 			    1, &value[i], tx);
2159 			ASSERT3U(error, ==, EEXIST);
2160 			error = zap_length(os, object, hc[i],
2161 			    &zl_intsize, &zl_ints);
2162 			ASSERT3U(error, ==, 0);
2163 			ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2164 			ASSERT3U(zl_ints, ==, 1);
2165 		}
2166 		for (i = 0; i < 2; i++) {
2167 			error = zap_remove(os, object, hc[i], tx);
2168 			ASSERT3U(error, ==, 0);
2169 		}
2170 
2171 		dmu_tx_commit(tx);
2172 	}
2173 
2174 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
2175 
2176 	while (--iters >= 0) {
2177 		prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
2178 		(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
2179 		(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
2180 		bzero(value, sizeof (value));
2181 		last_txg = 0;
2182 
2183 		/*
2184 		 * If these zap entries already exist, validate their contents.
2185 		 */
2186 		error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
2187 		if (error == 0) {
2188 			ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2189 			ASSERT3U(zl_ints, ==, 1);
2190 
2191 			error = zap_lookup(os, object, txgname, zl_intsize,
2192 			    zl_ints, &last_txg);
2193 
2194 			ASSERT3U(error, ==, 0);
2195 
2196 			error = zap_length(os, object, propname, &zl_intsize,
2197 			    &zl_ints);
2198 
2199 			ASSERT3U(error, ==, 0);
2200 			ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2201 			ASSERT3U(zl_ints, ==, ints);
2202 
2203 			error = zap_lookup(os, object, propname, zl_intsize,
2204 			    zl_ints, value);
2205 
2206 			ASSERT3U(error, ==, 0);
2207 
2208 			for (i = 0; i < ints; i++) {
2209 				ASSERT3U(value[i], ==, last_txg + object + i);
2210 			}
2211 		} else {
2212 			ASSERT3U(error, ==, ENOENT);
2213 		}
2214 
2215 		/*
2216 		 * Atomically update two entries in our zap object.
2217 		 * The first is named txg_%llu, and contains the txg
2218 		 * in which the property was last updated.  The second
2219 		 * is named prop_%llu, and the nth element of its value
2220 		 * should be txg + object + n.
2221 		 */
2222 		tx = dmu_tx_create(os);
2223 		dmu_tx_hold_zap(tx, object, TRUE, NULL);
2224 		error = dmu_tx_assign(tx, TXG_WAIT);
2225 		if (error) {
2226 			ztest_record_enospc("create zap entry");
2227 			dmu_tx_abort(tx);
2228 			return;
2229 		}
2230 		txg = dmu_tx_get_txg(tx);
2231 
2232 		if (last_txg > txg)
2233 			fatal(0, "zap future leak: old %llu new %llu",
2234 			    last_txg, txg);
2235 
2236 		for (i = 0; i < ints; i++)
2237 			value[i] = txg + object + i;
2238 
2239 		error = zap_update(os, object, txgname, sizeof (uint64_t),
2240 		    1, &txg, tx);
2241 		if (error)
2242 			fatal(0, "zap_update('%s', %llu, '%s') = %d",
2243 			    osname, object, txgname, error);
2244 
2245 		error = zap_update(os, object, propname, sizeof (uint64_t),
2246 		    ints, value, tx);
2247 		if (error)
2248 			fatal(0, "zap_update('%s', %llu, '%s') = %d",
2249 			    osname, object, propname, error);
2250 
2251 		dmu_tx_commit(tx);
2252 
2253 		/*
2254 		 * Remove a random pair of entries.
2255 		 */
2256 		prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
2257 		(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
2258 		(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
2259 
2260 		error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
2261 
2262 		if (error == ENOENT)
2263 			continue;
2264 
2265 		ASSERT3U(error, ==, 0);
2266 
2267 		tx = dmu_tx_create(os);
2268 		dmu_tx_hold_zap(tx, object, TRUE, NULL);
2269 		error = dmu_tx_assign(tx, TXG_WAIT);
2270 		if (error) {
2271 			ztest_record_enospc("remove zap entry");
2272 			dmu_tx_abort(tx);
2273 			return;
2274 		}
2275 		error = zap_remove(os, object, txgname, tx);
2276 		if (error)
2277 			fatal(0, "zap_remove('%s', %llu, '%s') = %d",
2278 			    osname, object, txgname, error);
2279 
2280 		error = zap_remove(os, object, propname, tx);
2281 		if (error)
2282 			fatal(0, "zap_remove('%s', %llu, '%s') = %d",
2283 			    osname, object, propname, error);
2284 
2285 		dmu_tx_commit(tx);
2286 	}
2287 
2288 	/*
2289 	 * Once in a while, destroy the object.
2290 	 */
2291 	if (ztest_random(100) != 0)
2292 		return;
2293 
2294 	tx = dmu_tx_create(os);
2295 	dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t));
2296 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2297 	error = dmu_tx_assign(tx, TXG_WAIT);
2298 	if (error) {
2299 		ztest_record_enospc("destroy zap object");
2300 		dmu_tx_abort(tx);
2301 		return;
2302 	}
2303 	error = zap_destroy(os, object, tx);
2304 	if (error)
2305 		fatal(0, "zap_destroy('%s', %llu) = %d",
2306 		    osname, object, error);
2307 	object = 0;
2308 	dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t),
2309 	    &object, tx);
2310 	dmu_tx_commit(tx);
2311 }
2312 
2313 void
2314 ztest_zap_parallel(ztest_args_t *za)
2315 {
2316 	objset_t *os = za->za_os;
2317 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
2318 	int iters = 100;
2319 	dmu_tx_t *tx;
2320 	int i, namelen, error;
2321 	char name[20], string_value[20];
2322 	void *data;
2323 
2324 	while (--iters >= 0) {
2325 		/*
2326 		 * Generate a random name of the form 'xxx.....' where each
2327 		 * x is a random printable character and the dots are dots.
2328 		 * There are 94 such characters, and the name length goes from
2329 		 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
2330 		 */
2331 		namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
2332 
2333 		for (i = 0; i < 3; i++)
2334 			name[i] = '!' + ztest_random('~' - '!' + 1);
2335 		for (; i < namelen - 1; i++)
2336 			name[i] = '.';
2337 		name[i] = '\0';
2338 
2339 		if (ztest_random(2) == 0)
2340 			object = ZTEST_MICROZAP_OBJ;
2341 		else
2342 			object = ZTEST_FATZAP_OBJ;
2343 
2344 		if ((namelen & 1) || object == ZTEST_MICROZAP_OBJ) {
2345 			wsize = sizeof (txg);
2346 			wc = 1;
2347 			data = &txg;
2348 		} else {
2349 			wsize = 1;
2350 			wc = namelen;
2351 			data = string_value;
2352 		}
2353 
2354 		count = -1ULL;
2355 		VERIFY(zap_count(os, object, &count) == 0);
2356 		ASSERT(count != -1ULL);
2357 
2358 		/*
2359 		 * Select an operation: length, lookup, add, update, remove.
2360 		 */
2361 		i = ztest_random(5);
2362 
2363 		if (i >= 2) {
2364 			tx = dmu_tx_create(os);
2365 			dmu_tx_hold_zap(tx, object, TRUE, NULL);
2366 			error = dmu_tx_assign(tx, TXG_WAIT);
2367 			if (error) {
2368 				ztest_record_enospc("zap parallel");
2369 				dmu_tx_abort(tx);
2370 				return;
2371 			}
2372 			txg = dmu_tx_get_txg(tx);
2373 			bcopy(name, string_value, namelen);
2374 		} else {
2375 			tx = NULL;
2376 			txg = 0;
2377 			bzero(string_value, namelen);
2378 		}
2379 
2380 		switch (i) {
2381 
2382 		case 0:
2383 			error = zap_length(os, object, name, &zl_wsize, &zl_wc);
2384 			if (error == 0) {
2385 				ASSERT3U(wsize, ==, zl_wsize);
2386 				ASSERT3U(wc, ==, zl_wc);
2387 			} else {
2388 				ASSERT3U(error, ==, ENOENT);
2389 			}
2390 			break;
2391 
2392 		case 1:
2393 			error = zap_lookup(os, object, name, wsize, wc, data);
2394 			if (error == 0) {
2395 				if (data == string_value &&
2396 				    bcmp(name, data, namelen) != 0)
2397 					fatal(0, "name '%s' != val '%s' len %d",
2398 					    name, data, namelen);
2399 			} else {
2400 				ASSERT3U(error, ==, ENOENT);
2401 			}
2402 			break;
2403 
2404 		case 2:
2405 			error = zap_add(os, object, name, wsize, wc, data, tx);
2406 			ASSERT(error == 0 || error == EEXIST);
2407 			break;
2408 
2409 		case 3:
2410 			VERIFY(zap_update(os, object, name, wsize, wc,
2411 			    data, tx) == 0);
2412 			break;
2413 
2414 		case 4:
2415 			error = zap_remove(os, object, name, tx);
2416 			ASSERT(error == 0 || error == ENOENT);
2417 			break;
2418 		}
2419 
2420 		if (tx != NULL)
2421 			dmu_tx_commit(tx);
2422 	}
2423 }
2424 
2425 void
2426 ztest_dsl_prop_get_set(ztest_args_t *za)
2427 {
2428 	objset_t *os = za->za_os;
2429 	int i, inherit;
2430 	uint64_t value;
2431 	const char *prop, *valname;
2432 	char setpoint[MAXPATHLEN];
2433 	char osname[MAXNAMELEN];
2434 	int error;
2435 
2436 	(void) rw_rdlock(&ztest_shared->zs_name_lock);
2437 
2438 	dmu_objset_name(os, osname);
2439 
2440 	for (i = 0; i < 2; i++) {
2441 		if (i == 0) {
2442 			prop = "checksum";
2443 			value = ztest_random_checksum();
2444 			inherit = (value == ZIO_CHECKSUM_INHERIT);
2445 		} else {
2446 			prop = "compression";
2447 			value = ztest_random_compress();
2448 			inherit = (value == ZIO_COMPRESS_INHERIT);
2449 		}
2450 
2451 		error = dsl_prop_set(osname, prop, sizeof (value),
2452 		    !inherit, &value);
2453 
2454 		if (error == ENOSPC) {
2455 			ztest_record_enospc("dsl_prop_set");
2456 			break;
2457 		}
2458 
2459 		ASSERT3U(error, ==, 0);
2460 
2461 		VERIFY3U(dsl_prop_get(osname, prop, sizeof (value),
2462 		    1, &value, setpoint), ==, 0);
2463 
2464 		if (i == 0)
2465 			valname = zio_checksum_table[value].ci_name;
2466 		else
2467 			valname = zio_compress_table[value].ci_name;
2468 
2469 		if (zopt_verbose >= 6) {
2470 			(void) printf("%s %s = %s for '%s'\n",
2471 			    osname, prop, valname, setpoint);
2472 		}
2473 	}
2474 
2475 	(void) rw_unlock(&ztest_shared->zs_name_lock);
2476 }
2477 
2478 static void
2479 ztest_error_setup(vdev_t *vd, int mode, int mask, uint64_t arg)
2480 {
2481 	int c;
2482 
2483 	for (c = 0; c < vd->vdev_children; c++)
2484 		ztest_error_setup(vd->vdev_child[c], mode, mask, arg);
2485 
2486 	if (vd->vdev_path != NULL) {
2487 		vd->vdev_fault_mode = mode;
2488 		vd->vdev_fault_mask = mask;
2489 		vd->vdev_fault_arg = arg;
2490 	}
2491 }
2492 
2493 /*
2494  * Inject random faults into the on-disk data.
2495  */
2496 void
2497 ztest_fault_inject(ztest_args_t *za)
2498 {
2499 	int fd;
2500 	uint64_t offset;
2501 	uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
2502 	uint64_t bad = 0x1990c0ffeedecade;
2503 	uint64_t top, leaf;
2504 	char path0[MAXPATHLEN];
2505 	char pathrand[MAXPATHLEN];
2506 	size_t fsize;
2507 	spa_t *spa = dmu_objset_spa(za->za_os);
2508 	int bshift = SPA_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
2509 	int iters = 1000;
2510 	vdev_t *vd0;
2511 	uint64_t guid0 = 0;
2512 
2513 	/*
2514 	 * We can't inject faults when we have no fault tolerance.
2515 	 */
2516 	if (zopt_maxfaults == 0)
2517 		return;
2518 
2519 	ASSERT(leaves >= 2);
2520 
2521 	/*
2522 	 * Pick a random top-level vdev.
2523 	 */
2524 	spa_config_enter(spa, RW_READER, FTAG);
2525 	top = ztest_random(spa->spa_root_vdev->vdev_children);
2526 	spa_config_exit(spa, FTAG);
2527 
2528 	/*
2529 	 * Pick a random leaf.
2530 	 */
2531 	leaf = ztest_random(leaves);
2532 
2533 	/*
2534 	 * Generate paths to the first two leaves in this top-level vdev,
2535 	 * and to the random leaf we selected.  We'll induce transient
2536 	 * I/O errors and random online/offline activity on leaf 0,
2537 	 * and we'll write random garbage to the randomly chosen leaf.
2538 	 */
2539 	(void) snprintf(path0, sizeof (path0),
2540 	    ztest_dev_template, zopt_dir, zopt_pool, top * leaves + 0);
2541 	(void) snprintf(pathrand, sizeof (pathrand),
2542 	    ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf);
2543 
2544 	dprintf("damaging %s and %s\n", path0, pathrand);
2545 
2546 	spa_config_enter(spa, RW_READER, FTAG);
2547 
2548 	/*
2549 	 * If we can tolerate two or more faults, make vd0 fail randomly.
2550 	 */
2551 	vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
2552 	if (vd0 != NULL && zopt_maxfaults >= 2) {
2553 		guid0 = vd0->vdev_guid;
2554 		ztest_error_setup(vd0, VDEV_FAULT_COUNT,
2555 		    (1U << ZIO_TYPE_READ) | (1U << ZIO_TYPE_WRITE), 100);
2556 	}
2557 
2558 	spa_config_exit(spa, FTAG);
2559 
2560 	/*
2561 	 * If we can tolerate two or more faults, randomly online/offline vd0.
2562 	 */
2563 	if (zopt_maxfaults >= 2 && guid0 != 0) {
2564 		if (ztest_random(10) < 6)
2565 			(void) vdev_offline(spa, guid0, B_TRUE);
2566 		else
2567 			(void) vdev_online(spa, guid0);
2568 	}
2569 
2570 	/*
2571 	 * We have at least single-fault tolerance, so inject data corruption.
2572 	 */
2573 	fd = open(pathrand, O_RDWR);
2574 
2575 	if (fd == -1)	/* we hit a gap in the device namespace */
2576 		return;
2577 
2578 	fsize = lseek(fd, 0, SEEK_END);
2579 
2580 	while (--iters != 0) {
2581 		offset = ztest_random(fsize / (leaves << bshift)) *
2582 		    (leaves << bshift) + (leaf << bshift) +
2583 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
2584 
2585 		if (offset >= fsize)
2586 			continue;
2587 
2588 		if (zopt_verbose >= 6)
2589 			(void) printf("injecting bad word into %s,"
2590 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
2591 
2592 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
2593 			fatal(1, "can't inject bad word at 0x%llx in %s",
2594 			    offset, pathrand);
2595 	}
2596 
2597 	(void) close(fd);
2598 }
2599 
2600 /*
2601  * Scrub the pool.
2602  */
2603 void
2604 ztest_scrub(ztest_args_t *za)
2605 {
2606 	spa_t *spa = dmu_objset_spa(za->za_os);
2607 
2608 	(void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE);
2609 	(void) poll(NULL, 0, 1000); /* wait a second, then force a restart */
2610 	(void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE);
2611 }
2612 
2613 /*
2614  * Rename the pool to a different name and then rename it back.
2615  */
2616 void
2617 ztest_spa_rename(ztest_args_t *za)
2618 {
2619 	char *oldname, *newname;
2620 	int error;
2621 	spa_t *spa;
2622 
2623 	(void) rw_wrlock(&ztest_shared->zs_name_lock);
2624 
2625 	oldname = za->za_pool;
2626 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
2627 	(void) strcpy(newname, oldname);
2628 	(void) strcat(newname, "_tmp");
2629 
2630 	/*
2631 	 * Do the rename
2632 	 */
2633 	error = spa_rename(oldname, newname);
2634 	if (error)
2635 		fatal(0, "spa_rename('%s', '%s') = %d", oldname,
2636 		    newname, error);
2637 
2638 	/*
2639 	 * Try to open it under the old name, which shouldn't exist
2640 	 */
2641 	error = spa_open(oldname, &spa, FTAG);
2642 	if (error != ENOENT)
2643 		fatal(0, "spa_open('%s') = %d", oldname, error);
2644 
2645 	/*
2646 	 * Open it under the new name and make sure it's still the same spa_t.
2647 	 */
2648 	error = spa_open(newname, &spa, FTAG);
2649 	if (error != 0)
2650 		fatal(0, "spa_open('%s') = %d", newname, error);
2651 
2652 	ASSERT(spa == dmu_objset_spa(za->za_os));
2653 	spa_close(spa, FTAG);
2654 
2655 	/*
2656 	 * Rename it back to the original
2657 	 */
2658 	error = spa_rename(newname, oldname);
2659 	if (error)
2660 		fatal(0, "spa_rename('%s', '%s') = %d", newname,
2661 		    oldname, error);
2662 
2663 	/*
2664 	 * Make sure it can still be opened
2665 	 */
2666 	error = spa_open(oldname, &spa, FTAG);
2667 	if (error != 0)
2668 		fatal(0, "spa_open('%s') = %d", oldname, error);
2669 
2670 	ASSERT(spa == dmu_objset_spa(za->za_os));
2671 	spa_close(spa, FTAG);
2672 
2673 	umem_free(newname, strlen(newname) + 1);
2674 
2675 	(void) rw_unlock(&ztest_shared->zs_name_lock);
2676 }
2677 
2678 
2679 /*
2680  * Completely obliterate one disk.
2681  */
2682 static void
2683 ztest_obliterate_one_disk(uint64_t vdev)
2684 {
2685 	int fd;
2686 	char dev_name[MAXPATHLEN], copy_name[MAXPATHLEN];
2687 	size_t fsize;
2688 
2689 	if (zopt_maxfaults < 2)
2690 		return;
2691 
2692 	(void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
2693 	(void) snprintf(copy_name, MAXPATHLEN, "%s.old", dev_name);
2694 
2695 	fd = open(dev_name, O_RDWR);
2696 
2697 	if (fd == -1)
2698 		fatal(1, "can't open %s", dev_name);
2699 
2700 	/*
2701 	 * Determine the size.
2702 	 */
2703 	fsize = lseek(fd, 0, SEEK_END);
2704 
2705 	(void) close(fd);
2706 
2707 	/*
2708 	 * Rename the old device to dev_name.old (useful for debugging).
2709 	 */
2710 	VERIFY(rename(dev_name, copy_name) == 0);
2711 
2712 	/*
2713 	 * Create a new one.
2714 	 */
2715 	VERIFY((fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666)) >= 0);
2716 	VERIFY(ftruncate(fd, fsize) == 0);
2717 	(void) close(fd);
2718 }
2719 
2720 static void
2721 ztest_replace_one_disk(spa_t *spa, uint64_t vdev)
2722 {
2723 	char dev_name[MAXPATHLEN];
2724 	nvlist_t *file, *root;
2725 	int error;
2726 	uint64_t guid;
2727 	uint64_t ashift = ztest_get_ashift();
2728 	vdev_t *vd;
2729 
2730 	(void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
2731 
2732 	/*
2733 	 * Build the nvlist describing dev_name.
2734 	 */
2735 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
2736 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
2737 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0);
2738 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
2739 
2740 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
2741 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
2742 	VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN,
2743 	    &file, 1) == 0);
2744 
2745 	spa_config_enter(spa, RW_READER, FTAG);
2746 	if ((vd = vdev_lookup_by_path(spa->spa_root_vdev, dev_name)) == NULL)
2747 		guid = 0;
2748 	else
2749 		guid = vd->vdev_guid;
2750 	spa_config_exit(spa, FTAG);
2751 	error = spa_vdev_attach(spa, guid, root, B_TRUE);
2752 	if (error != 0 &&
2753 	    error != EBUSY &&
2754 	    error != ENOTSUP &&
2755 	    error != ENODEV &&
2756 	    error != EDOM)
2757 		fatal(0, "spa_vdev_attach(in-place) = %d", error);
2758 
2759 	nvlist_free(file);
2760 	nvlist_free(root);
2761 }
2762 
2763 static void
2764 ztest_verify_blocks(char *pool)
2765 {
2766 	int status;
2767 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
2768 	char zbuf[1024];
2769 	char *bin;
2770 	FILE *fp;
2771 
2772 	(void) realpath(getexecname(), zdb);
2773 
2774 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
2775 	bin = strstr(zdb, "/usr/bin/");
2776 	/* LINTED */
2777 	(void) sprintf(bin, "/usr/sbin/zdb -bc%s%s -U -O %s %s",
2778 	    zopt_verbose >= 3 ? "s" : "",
2779 	    zopt_verbose >= 4 ? "v" : "",
2780 	    ztest_random(2) == 0 ? "pre" : "post", pool);
2781 
2782 	if (zopt_verbose >= 5)
2783 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
2784 
2785 	fp = popen(zdb, "r");
2786 
2787 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
2788 		if (zopt_verbose >= 3)
2789 			(void) printf("%s", zbuf);
2790 
2791 	status = pclose(fp);
2792 
2793 	if (status == 0)
2794 		return;
2795 
2796 	ztest_dump_core = 0;
2797 	if (WIFEXITED(status))
2798 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
2799 	else
2800 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
2801 }
2802 
2803 static void
2804 ztest_walk_pool_directory(char *header)
2805 {
2806 	spa_t *spa = NULL;
2807 
2808 	if (zopt_verbose >= 6)
2809 		(void) printf("%s\n", header);
2810 
2811 	mutex_enter(&spa_namespace_lock);
2812 	while ((spa = spa_next(spa)) != NULL)
2813 		if (zopt_verbose >= 6)
2814 			(void) printf("\t%s\n", spa_name(spa));
2815 	mutex_exit(&spa_namespace_lock);
2816 }
2817 
2818 static void
2819 ztest_spa_import_export(char *oldname, char *newname)
2820 {
2821 	nvlist_t *config;
2822 	uint64_t pool_guid;
2823 	spa_t *spa;
2824 	int error;
2825 
2826 	if (zopt_verbose >= 4) {
2827 		(void) printf("import/export: old = %s, new = %s\n",
2828 		    oldname, newname);
2829 	}
2830 
2831 	/*
2832 	 * Clean up from previous runs.
2833 	 */
2834 	(void) spa_destroy(newname);
2835 
2836 	/*
2837 	 * Get the pool's configuration and guid.
2838 	 */
2839 	error = spa_open(oldname, &spa, FTAG);
2840 	if (error)
2841 		fatal(0, "spa_open('%s') = %d", oldname, error);
2842 
2843 	pool_guid = spa_guid(spa);
2844 	spa_close(spa, FTAG);
2845 
2846 	ztest_walk_pool_directory("pools before export");
2847 
2848 	/*
2849 	 * Export it.
2850 	 */
2851 	error = spa_export(oldname, &config);
2852 	if (error)
2853 		fatal(0, "spa_export('%s') = %d", oldname, error);
2854 
2855 	ztest_walk_pool_directory("pools after export");
2856 
2857 	/*
2858 	 * Import it under the new name.
2859 	 */
2860 	error = spa_import(newname, config, NULL);
2861 	if (error)
2862 		fatal(0, "spa_import('%s') = %d", newname, error);
2863 
2864 	ztest_walk_pool_directory("pools after import");
2865 
2866 	/*
2867 	 * Try to import it again -- should fail with EEXIST.
2868 	 */
2869 	error = spa_import(newname, config, NULL);
2870 	if (error != EEXIST)
2871 		fatal(0, "spa_import('%s') twice", newname);
2872 
2873 	/*
2874 	 * Try to import it under a different name -- should fail with EEXIST.
2875 	 */
2876 	error = spa_import(oldname, config, NULL);
2877 	if (error != EEXIST)
2878 		fatal(0, "spa_import('%s') under multiple names", newname);
2879 
2880 	/*
2881 	 * Verify that the pool is no longer visible under the old name.
2882 	 */
2883 	error = spa_open(oldname, &spa, FTAG);
2884 	if (error != ENOENT)
2885 		fatal(0, "spa_open('%s') = %d", newname, error);
2886 
2887 	/*
2888 	 * Verify that we can open and close the pool using the new name.
2889 	 */
2890 	error = spa_open(newname, &spa, FTAG);
2891 	if (error)
2892 		fatal(0, "spa_open('%s') = %d", newname, error);
2893 	ASSERT(pool_guid == spa_guid(spa));
2894 	spa_close(spa, FTAG);
2895 
2896 	nvlist_free(config);
2897 }
2898 
2899 static void *
2900 ztest_thread(void *arg)
2901 {
2902 	ztest_args_t *za = arg;
2903 	ztest_shared_t *zs = ztest_shared;
2904 	hrtime_t now, functime;
2905 	ztest_info_t *zi;
2906 	int f;
2907 
2908 	while ((now = gethrtime()) < za->za_stop) {
2909 		/*
2910 		 * See if it's time to force a crash.
2911 		 */
2912 		if (now > za->za_kill) {
2913 			zs->zs_alloc = spa_get_alloc(dmu_objset_spa(za->za_os));
2914 			zs->zs_space = spa_get_space(dmu_objset_spa(za->za_os));
2915 			(void) kill(getpid(), SIGKILL);
2916 		}
2917 
2918 		/*
2919 		 * Pick a random function.
2920 		 */
2921 		f = ztest_random(ZTEST_FUNCS);
2922 		zi = &zs->zs_info[f];
2923 
2924 		/*
2925 		 * Decide whether to call it, based on the requested frequency.
2926 		 */
2927 		if (zi->zi_call_target == 0 ||
2928 		    (double)zi->zi_call_total / zi->zi_call_target >
2929 		    (double)(now - zs->zs_start_time) / (zopt_time * NANOSEC))
2930 			continue;
2931 
2932 		atomic_add_64(&zi->zi_calls, 1);
2933 		atomic_add_64(&zi->zi_call_total, 1);
2934 
2935 		za->za_diroff = (za->za_instance * ZTEST_FUNCS + f) *
2936 		    ZTEST_DIRSIZE;
2937 		za->za_diroff_shared = (1ULL << 63);
2938 
2939 		ztest_dmu_write_parallel(za);
2940 
2941 		zi->zi_func(za);
2942 
2943 		functime = gethrtime() - now;
2944 
2945 		atomic_add_64(&zi->zi_call_time, functime);
2946 
2947 		if (zopt_verbose >= 4) {
2948 			Dl_info dli;
2949 			(void) dladdr((void *)zi->zi_func, &dli);
2950 			(void) printf("%6.2f sec in %s\n",
2951 			    (double)functime / NANOSEC, dli.dli_sname);
2952 		}
2953 
2954 		/*
2955 		 * If we're getting ENOSPC with some regularity, stop.
2956 		 */
2957 		if (zs->zs_enospc_count > 10)
2958 			break;
2959 	}
2960 
2961 	return (NULL);
2962 }
2963 
2964 /*
2965  * Kick off threads to run tests on all datasets in parallel.
2966  */
2967 static void
2968 ztest_run(char *pool)
2969 {
2970 	int t, d, error;
2971 	ztest_shared_t *zs = ztest_shared;
2972 	ztest_args_t *za;
2973 	spa_t *spa;
2974 	char name[100];
2975 
2976 	(void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL);
2977 	(void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL);
2978 
2979 	for (t = 0; t < ZTEST_SYNC_LOCKS; t++)
2980 		(void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL);
2981 
2982 	/*
2983 	 * Destroy one disk before we even start.
2984 	 * It's mirrored, so everything should work just fine.
2985 	 * This makes us exercise fault handling very early in spa_load().
2986 	 */
2987 	ztest_obliterate_one_disk(0);
2988 
2989 	/*
2990 	 * Verify that the sum of the sizes of all blocks in the pool
2991 	 * equals the SPA's allocated space total.
2992 	 */
2993 	ztest_verify_blocks(pool);
2994 
2995 	/*
2996 	 * Kick off a replacement of the disk we just obliterated.
2997 	 */
2998 	kernel_init(FREAD | FWRITE);
2999 	error = spa_open(pool, &spa, FTAG);
3000 	if (error)
3001 		fatal(0, "spa_open(%s) = %d", pool, error);
3002 	ztest_replace_one_disk(spa, 0);
3003 	if (zopt_verbose >= 5)
3004 		show_pool_stats(spa);
3005 	spa_close(spa, FTAG);
3006 	kernel_fini();
3007 
3008 	kernel_init(FREAD | FWRITE);
3009 
3010 	/*
3011 	 * Verify that we can export the pool and reimport it under a
3012 	 * different name.
3013 	 */
3014 	if (ztest_random(2) == 0) {
3015 		(void) snprintf(name, 100, "%s_import", pool);
3016 		ztest_spa_import_export(pool, name);
3017 		ztest_spa_import_export(name, pool);
3018 	}
3019 
3020 	/*
3021 	 * Verify that we can loop over all pools.
3022 	 */
3023 	mutex_enter(&spa_namespace_lock);
3024 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) {
3025 		if (zopt_verbose > 3) {
3026 			(void) printf("spa_next: found %s\n", spa_name(spa));
3027 		}
3028 	}
3029 	mutex_exit(&spa_namespace_lock);
3030 
3031 	/*
3032 	 * Open our pool.
3033 	 */
3034 	error = spa_open(pool, &spa, FTAG);
3035 	if (error)
3036 		fatal(0, "spa_open() = %d", error);
3037 
3038 	/*
3039 	 * Verify that we can safely inquire about about any object,
3040 	 * whether it's allocated or not.  To make it interesting,
3041 	 * we probe a 5-wide window around each power of two.
3042 	 * This hits all edge cases, including zero and the max.
3043 	 */
3044 	for (t = 0; t < 64; t++) {
3045 		for (d = -5; d <= 5; d++) {
3046 			error = dmu_object_info(spa->spa_meta_objset,
3047 			    (1ULL << t) + d, NULL);
3048 			ASSERT(error == 0 || error == ENOENT ||
3049 			    error == EINVAL);
3050 		}
3051 	}
3052 
3053 	/*
3054 	 * Now kick off all the tests that run in parallel.
3055 	 */
3056 	zs->zs_enospc_count = 0;
3057 
3058 	za = umem_zalloc(zopt_threads * sizeof (ztest_args_t), UMEM_NOFAIL);
3059 
3060 	if (zopt_verbose >= 4)
3061 		(void) printf("starting main threads...\n");
3062 
3063 	za[0].za_start = gethrtime();
3064 	za[0].za_stop = za[0].za_start + zopt_passtime * NANOSEC;
3065 	za[0].za_stop = MIN(za[0].za_stop, zs->zs_stop_time);
3066 	za[0].za_kill = za[0].za_stop;
3067 	if (ztest_random(100) < zopt_killrate)
3068 		za[0].za_kill -= ztest_random(zopt_passtime * NANOSEC);
3069 
3070 	for (t = 0; t < zopt_threads; t++) {
3071 		d = t % zopt_datasets;
3072 		if (t < zopt_datasets) {
3073 			ztest_replay_t zr;
3074 			(void) rw_rdlock(&ztest_shared->zs_name_lock);
3075 			(void) snprintf(name, 100, "%s/%s_%d", pool, pool, d);
3076 			error = dmu_objset_create(name, DMU_OST_OTHER, NULL,
3077 			    ztest_create_cb, NULL);
3078 			if (error != 0 && error != EEXIST) {
3079 				if (error == ENOSPC) {
3080 					zs->zs_enospc_count++;
3081 					(void) rw_unlock(
3082 					    &ztest_shared->zs_name_lock);
3083 					break;
3084 				}
3085 				fatal(0, "dmu_objset_create(%s) = %d",
3086 				    name, error);
3087 			}
3088 			error = dmu_objset_open(name, DMU_OST_OTHER,
3089 			    DS_MODE_STANDARD, &za[d].za_os);
3090 			if (error)
3091 				fatal(0, "dmu_objset_open('%s') = %d",
3092 				    name, error);
3093 			(void) rw_unlock(&ztest_shared->zs_name_lock);
3094 			zr.zr_os = za[d].za_os;
3095 			zil_replay(zr.zr_os, &zr, &zr.zr_assign,
3096 			    ztest_replay_vector, NULL);
3097 			za[d].za_zilog = zil_open(za[d].za_os, NULL);
3098 		}
3099 		za[t].za_pool = spa_strdup(pool);
3100 		za[t].za_os = za[d].za_os;
3101 		za[t].za_zilog = za[d].za_zilog;
3102 		za[t].za_instance = t;
3103 		za[t].za_random = ztest_random(-1ULL);
3104 		za[t].za_start = za[0].za_start;
3105 		za[t].za_stop = za[0].za_stop;
3106 		za[t].za_kill = za[0].za_kill;
3107 
3108 		error = thr_create(0, 0, ztest_thread, &za[t], THR_BOUND,
3109 		    &za[t].za_thread);
3110 		if (error)
3111 			fatal(0, "can't create thread %d: error %d",
3112 			    t, error);
3113 	}
3114 
3115 	while (--t >= 0) {
3116 		error = thr_join(za[t].za_thread, NULL, NULL);
3117 		if (error)
3118 			fatal(0, "thr_join(%d) = %d", t, error);
3119 		if (za[t].za_th)
3120 			traverse_fini(za[t].za_th);
3121 		if (t < zopt_datasets) {
3122 			zil_close(za[t].za_zilog);
3123 			dmu_objset_close(za[t].za_os);
3124 		}
3125 		spa_strfree(za[t].za_pool);
3126 	}
3127 
3128 	umem_free(za, zopt_threads * sizeof (ztest_args_t));
3129 
3130 	if (zopt_verbose >= 3)
3131 		show_pool_stats(spa);
3132 
3133 	txg_wait_synced(spa_get_dsl(spa), 0);
3134 
3135 	zs->zs_alloc = spa_get_alloc(spa);
3136 	zs->zs_space = spa_get_space(spa);
3137 
3138 	/*
3139 	 * Did we have out-of-space errors?  If so, destroy a random objset.
3140 	 */
3141 	if (zs->zs_enospc_count != 0) {
3142 		(void) rw_rdlock(&ztest_shared->zs_name_lock);
3143 		(void) snprintf(name, 100, "%s/%s_%d", pool, pool,
3144 		    (int)ztest_random(zopt_datasets));
3145 		if (zopt_verbose >= 3)
3146 			(void) printf("Destroying %s to free up space\n", name);
3147 		dmu_objset_find(name, ztest_destroy_cb, NULL,
3148 		    DS_FIND_SNAPSHOTS);
3149 		(void) rw_unlock(&ztest_shared->zs_name_lock);
3150 	}
3151 
3152 	txg_wait_synced(spa_get_dsl(spa), 0);
3153 
3154 	/*
3155 	 * Right before closing the pool, kick off a bunch of async I/O;
3156 	 * spa_close() should wait for it to complete.
3157 	 */
3158 	for (t = 1; t < 50; t++)
3159 		dmu_prefetch(spa->spa_meta_objset, t, 0, 1 << 15);
3160 
3161 	spa_close(spa, FTAG);
3162 
3163 	kernel_fini();
3164 }
3165 
3166 void
3167 print_time(hrtime_t t, char *timebuf)
3168 {
3169 	hrtime_t s = t / NANOSEC;
3170 	hrtime_t m = s / 60;
3171 	hrtime_t h = m / 60;
3172 	hrtime_t d = h / 24;
3173 
3174 	s -= m * 60;
3175 	m -= h * 60;
3176 	h -= d * 24;
3177 
3178 	timebuf[0] = '\0';
3179 
3180 	if (d)
3181 		(void) sprintf(timebuf,
3182 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
3183 	else if (h)
3184 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
3185 	else if (m)
3186 		(void) sprintf(timebuf, "%llum%02llus", m, s);
3187 	else
3188 		(void) sprintf(timebuf, "%llus", s);
3189 }
3190 
3191 /*
3192  * Create a storage pool with the given name and initial vdev size.
3193  * Then create the specified number of datasets in the pool.
3194  */
3195 static void
3196 ztest_init(char *pool)
3197 {
3198 	spa_t *spa;
3199 	int error;
3200 	nvlist_t *nvroot;
3201 
3202 	kernel_init(FREAD | FWRITE);
3203 
3204 	/*
3205 	 * Create the storage pool.
3206 	 */
3207 	(void) spa_destroy(pool);
3208 	ztest_shared->zs_vdev_primaries = 0;
3209 	nvroot = make_vdev_root(zopt_vdev_size, zopt_raidz, zopt_mirrors, 1);
3210 	error = spa_create(pool, nvroot, NULL);
3211 	nvlist_free(nvroot);
3212 
3213 	if (error)
3214 		fatal(0, "spa_create() = %d", error);
3215 	error = spa_open(pool, &spa, FTAG);
3216 	if (error)
3217 		fatal(0, "spa_open() = %d", error);
3218 
3219 	if (zopt_verbose >= 3)
3220 		show_pool_stats(spa);
3221 
3222 	spa_close(spa, FTAG);
3223 
3224 	kernel_fini();
3225 }
3226 
3227 int
3228 main(int argc, char **argv)
3229 {
3230 	int kills = 0;
3231 	int iters = 0;
3232 	int i, f;
3233 	ztest_shared_t *zs;
3234 	ztest_info_t *zi;
3235 	char timebuf[100];
3236 	char numbuf[6];
3237 
3238 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
3239 
3240 	/* Override location of zpool.cache */
3241 	spa_config_dir = "/tmp";
3242 
3243 	ztest_random_fd = open("/dev/urandom", O_RDONLY);
3244 
3245 	process_options(argc, argv);
3246 
3247 	argc -= optind;
3248 	argv += optind;
3249 
3250 	dprintf_setup(&argc, argv);
3251 
3252 	/*
3253 	 * Blow away any existing copy of zpool.cache
3254 	 */
3255 	if (zopt_init != 0)
3256 		(void) remove("/tmp/zpool.cache");
3257 
3258 	zs = ztest_shared = (void *)mmap(0,
3259 	    P2ROUNDUP(sizeof (ztest_shared_t), getpagesize()),
3260 	    PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
3261 
3262 	if (zopt_verbose >= 1) {
3263 		(void) printf("%llu vdevs, %d datasets, %d threads,"
3264 		    " %llu seconds...\n",
3265 		    (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads,
3266 		    (u_longlong_t)zopt_time);
3267 	}
3268 
3269 	/*
3270 	 * Create and initialize our storage pool.
3271 	 */
3272 	for (i = 1; i <= zopt_init; i++) {
3273 		bzero(zs, sizeof (ztest_shared_t));
3274 		if (zopt_verbose >= 3 && zopt_init != 1)
3275 			(void) printf("ztest_init(), pass %d\n", i);
3276 		ztest_init(zopt_pool);
3277 	}
3278 
3279 	/*
3280 	 * Initialize the call targets for each function.
3281 	 */
3282 	for (f = 0; f < ZTEST_FUNCS; f++) {
3283 		zi = &zs->zs_info[f];
3284 
3285 		*zi = ztest_info[f];
3286 
3287 		if (*zi->zi_interval == 0)
3288 			zi->zi_call_target = UINT64_MAX;
3289 		else
3290 			zi->zi_call_target = zopt_time / *zi->zi_interval;
3291 	}
3292 
3293 	zs->zs_start_time = gethrtime();
3294 	zs->zs_stop_time = zs->zs_start_time + zopt_time * NANOSEC;
3295 
3296 	/*
3297 	 * Run the tests in a loop.  These tests include fault injection
3298 	 * to verify that self-healing data works, and forced crashes
3299 	 * to verify that we never lose on-disk consistency.
3300 	 */
3301 	while (gethrtime() < zs->zs_stop_time) {
3302 		int status;
3303 		pid_t pid;
3304 		char *tmp;
3305 
3306 		/*
3307 		 * Initialize the workload counters for each function.
3308 		 */
3309 		for (f = 0; f < ZTEST_FUNCS; f++) {
3310 			zi = &zs->zs_info[f];
3311 			zi->zi_calls = 0;
3312 			zi->zi_call_time = 0;
3313 		}
3314 
3315 		pid = fork();
3316 
3317 		if (pid == -1)
3318 			fatal(1, "fork failed");
3319 
3320 		if (pid == 0) {	/* child */
3321 			struct rlimit rl = { 1024, 1024 };
3322 			(void) setrlimit(RLIMIT_NOFILE, &rl);
3323 			(void) enable_extended_FILE_stdio(-1, -1);
3324 			ztest_run(zopt_pool);
3325 			exit(0);
3326 		}
3327 
3328 		while (waitpid(pid, &status, WEXITED) != pid)
3329 			continue;
3330 
3331 		if (WIFEXITED(status)) {
3332 			if (WEXITSTATUS(status) != 0) {
3333 				(void) fprintf(stderr,
3334 				    "child exited with code %d\n",
3335 				    WEXITSTATUS(status));
3336 				exit(2);
3337 			}
3338 		} else {
3339 			if (WTERMSIG(status) != SIGKILL) {
3340 				(void) fprintf(stderr,
3341 				    "child died with signal %d\n",
3342 				    WTERMSIG(status));
3343 				exit(3);
3344 			}
3345 			kills++;
3346 		}
3347 
3348 		iters++;
3349 
3350 		if (zopt_verbose >= 1) {
3351 			hrtime_t now = gethrtime();
3352 
3353 			now = MIN(now, zs->zs_stop_time);
3354 			print_time(zs->zs_stop_time - now, timebuf);
3355 			nicenum(zs->zs_space, numbuf);
3356 
3357 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
3358 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
3359 			    iters,
3360 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
3361 			    (u_longlong_t)zs->zs_enospc_count,
3362 			    100.0 * zs->zs_alloc / zs->zs_space,
3363 			    numbuf,
3364 			    100.0 * (now - zs->zs_start_time) /
3365 			    (zopt_time * NANOSEC), timebuf);
3366 		}
3367 
3368 		if (zopt_verbose >= 2) {
3369 			(void) printf("\nWorkload summary:\n\n");
3370 			(void) printf("%7s %9s   %s\n",
3371 			    "Calls", "Time", "Function");
3372 			(void) printf("%7s %9s   %s\n",
3373 			    "-----", "----", "--------");
3374 			for (f = 0; f < ZTEST_FUNCS; f++) {
3375 				Dl_info dli;
3376 
3377 				zi = &zs->zs_info[f];
3378 				print_time(zi->zi_call_time, timebuf);
3379 				(void) dladdr((void *)zi->zi_func, &dli);
3380 				(void) printf("%7llu %9s   %s\n",
3381 				    (u_longlong_t)zi->zi_calls, timebuf,
3382 				    dli.dli_sname);
3383 			}
3384 			(void) printf("\n");
3385 		}
3386 
3387 		/*
3388 		 * It's possible that we killed a child during a rename test, in
3389 		 * which case we'll have a 'ztest_tmp' pool lying around instead
3390 		 * of 'ztest'.  Do a blind rename in case this happened.
3391 		 */
3392 		tmp = umem_alloc(strlen(zopt_pool) + 5, UMEM_NOFAIL);
3393 		(void) strcpy(tmp, zopt_pool);
3394 		(void) strcat(tmp, "_tmp");
3395 		kernel_init(FREAD | FWRITE);
3396 		(void) spa_rename(tmp, zopt_pool);
3397 		kernel_fini();
3398 		umem_free(tmp, strlen(tmp) + 1);
3399 	}
3400 
3401 	ztest_verify_blocks(zopt_pool);
3402 
3403 	if (zopt_verbose >= 1) {
3404 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
3405 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
3406 	}
3407 
3408 	return (0);
3409 }
3410