xref: /titanic_52/usr/src/cmd/ztest/ztest.c (revision 210938ebfcf386cf4819f16c23979173b0d78097)
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 int
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 	return (0);
1102 }
1103 
1104 /*
1105  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
1106  */
1107 static uint64_t
1108 ztest_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t object, int mode)
1109 {
1110 	itx_t *itx;
1111 	lr_create_t *lr;
1112 	size_t namesize;
1113 	char name[24];
1114 
1115 	(void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object);
1116 	namesize = strlen(name) + 1;
1117 
1118 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize +
1119 	    ztest_random(ZIL_MAX_BLKSZ));
1120 	lr = (lr_create_t *)&itx->itx_lr;
1121 	bzero(lr + 1, lr->lr_common.lrc_reclen - sizeof (*lr));
1122 	lr->lr_doid = object;
1123 	lr->lr_foid = 0;
1124 	lr->lr_mode = mode;
1125 	lr->lr_uid = 0;
1126 	lr->lr_gid = 0;
1127 	lr->lr_gen = dmu_tx_get_txg(tx);
1128 	lr->lr_crtime[0] = time(NULL);
1129 	lr->lr_crtime[1] = 0;
1130 	lr->lr_rdev = 0;
1131 	bcopy(name, (char *)(lr + 1), namesize);
1132 
1133 	return (zil_itx_assign(zilog, itx, tx));
1134 }
1135 
1136 #ifndef lint
1137 static uint64_t
1138 ztest_log_remove(zilog_t *zilog, dmu_tx_t *tx, uint64_t object)
1139 {
1140 	itx_t *itx;
1141 	lr_remove_t *lr;
1142 	size_t namesize;
1143 	char name[24];
1144 
1145 	(void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object);
1146 	namesize = strlen(name) + 1;
1147 
1148 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize +
1149 	    ztest_random(8000));
1150 	lr = (lr_remove_t *)&itx->itx_lr;
1151 	lr->lr_doid = object;
1152 	bcopy(name, (char *)(lr + 1), namesize);
1153 
1154 	return (zil_itx_assign(zilog, itx, tx));
1155 }
1156 #endif /* lint */
1157 
1158 void
1159 ztest_dmu_objset_create_destroy(ztest_args_t *za)
1160 {
1161 	int error;
1162 	objset_t *os;
1163 	char name[100];
1164 	int mode, basemode, expected_error;
1165 	zilog_t *zilog;
1166 	uint64_t seq;
1167 	uint64_t objects;
1168 	ztest_replay_t zr;
1169 
1170 	(void) rw_rdlock(&ztest_shared->zs_name_lock);
1171 	(void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool,
1172 	    (u_longlong_t)za->za_instance);
1173 
1174 	basemode = DS_MODE_LEVEL(za->za_instance);
1175 	if (basemode == DS_MODE_NONE)
1176 		basemode++;
1177 
1178 	/*
1179 	 * If this dataset exists from a previous run, process its replay log
1180 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
1181 	 * (invoked from ztest_destroy_cb() below) should just throw it away.
1182 	 */
1183 	if (ztest_random(2) == 0 &&
1184 	    dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_PRIMARY, &os) == 0) {
1185 		zr.zr_os = os;
1186 		zil_replay(os, &zr, &zr.zr_assign, ztest_replay_vector, NULL);
1187 		dmu_objset_close(os);
1188 	}
1189 
1190 	/*
1191 	 * There may be an old instance of the dataset we're about to
1192 	 * create lying around from a previous run.  If so, destroy it
1193 	 * and all of its snapshots.
1194 	 */
1195 	(void) dmu_objset_find(name, ztest_destroy_cb, NULL,
1196 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1197 
1198 	/*
1199 	 * Verify that the destroyed dataset is no longer in the namespace.
1200 	 */
1201 	error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os);
1202 	if (error != ENOENT)
1203 		fatal(1, "dmu_objset_open(%s) found destroyed dataset %p",
1204 		    name, os);
1205 
1206 	/*
1207 	 * Verify that we can create a new dataset.
1208 	 */
1209 	error = dmu_objset_create(name, DMU_OST_OTHER, NULL, ztest_create_cb,
1210 	    NULL);
1211 	if (error) {
1212 		if (error == ENOSPC) {
1213 			ztest_record_enospc("dmu_objset_create");
1214 			(void) rw_unlock(&ztest_shared->zs_name_lock);
1215 			return;
1216 		}
1217 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
1218 	}
1219 
1220 	error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os);
1221 	if (error) {
1222 		fatal(0, "dmu_objset_open(%s) = %d", name, error);
1223 	}
1224 
1225 	/*
1226 	 * Open the intent log for it.
1227 	 */
1228 	zilog = zil_open(os, NULL);
1229 
1230 	/*
1231 	 * Put a random number of objects in there.
1232 	 */
1233 	objects = ztest_random(20);
1234 	seq = 0;
1235 	while (objects-- != 0) {
1236 		uint64_t object;
1237 		dmu_tx_t *tx = dmu_tx_create(os);
1238 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, sizeof (name));
1239 		error = dmu_tx_assign(tx, TXG_WAIT);
1240 		if (error) {
1241 			dmu_tx_abort(tx);
1242 		} else {
1243 			object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1244 			    DMU_OT_NONE, 0, tx);
1245 			ztest_set_random_blocksize(os, object, tx);
1246 			seq = ztest_log_create(zilog, tx, object,
1247 			    DMU_OT_UINT64_OTHER);
1248 			dmu_write(os, object, 0, sizeof (name), name, tx);
1249 			dmu_tx_commit(tx);
1250 		}
1251 		if (ztest_random(5) == 0) {
1252 			zil_commit(zilog, seq, object);
1253 		}
1254 		if (ztest_random(100) == 0) {
1255 			error = zil_suspend(zilog);
1256 			if (error == 0) {
1257 				zil_resume(zilog);
1258 			}
1259 		}
1260 	}
1261 
1262 	/*
1263 	 * Verify that we cannot create an existing dataset.
1264 	 */
1265 	error = dmu_objset_create(name, DMU_OST_OTHER, NULL, NULL, NULL);
1266 	if (error != EEXIST)
1267 		fatal(0, "created existing dataset, error = %d", error);
1268 
1269 	/*
1270 	 * Verify that multiple dataset opens are allowed, but only when
1271 	 * the new access mode is compatible with the base mode.
1272 	 * We use a mixture of typed and typeless opens, and when the
1273 	 * open succeeds, verify that the discovered type is correct.
1274 	 */
1275 	for (mode = DS_MODE_STANDARD; mode < DS_MODE_LEVELS; mode++) {
1276 		objset_t *os2;
1277 		error = dmu_objset_open(name, DMU_OST_OTHER, mode, &os2);
1278 		expected_error = (basemode + mode < DS_MODE_LEVELS) ? 0 : EBUSY;
1279 		if (error != expected_error)
1280 			fatal(0, "dmu_objset_open('%s') = %d, expected %d",
1281 			    name, error, expected_error);
1282 		if (error == 0)
1283 			dmu_objset_close(os2);
1284 	}
1285 
1286 	zil_close(zilog);
1287 	dmu_objset_close(os);
1288 
1289 	error = dmu_objset_destroy(name);
1290 	if (error)
1291 		fatal(0, "dmu_objset_destroy(%s) = %d", name, error);
1292 
1293 	(void) rw_unlock(&ztest_shared->zs_name_lock);
1294 }
1295 
1296 /*
1297  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
1298  */
1299 void
1300 ztest_dmu_snapshot_create_destroy(ztest_args_t *za)
1301 {
1302 	int error;
1303 	objset_t *os = za->za_os;
1304 	char snapname[100];
1305 	char osname[MAXNAMELEN];
1306 
1307 	(void) rw_rdlock(&ztest_shared->zs_name_lock);
1308 	dmu_objset_name(os, osname);
1309 	(void) snprintf(snapname, 100, "%s@%llu", osname,
1310 	    (u_longlong_t)za->za_instance);
1311 
1312 	error = dmu_objset_destroy(snapname);
1313 	if (error != 0 && error != ENOENT)
1314 		fatal(0, "dmu_objset_destroy() = %d", error);
1315 	error = dmu_objset_snapshot(osname, strchr(snapname, '@')+1, FALSE);
1316 	if (error == ENOSPC)
1317 		ztest_record_enospc("dmu_take_snapshot");
1318 	else if (error != 0 && error != EEXIST)
1319 		fatal(0, "dmu_take_snapshot() = %d", error);
1320 	(void) rw_unlock(&ztest_shared->zs_name_lock);
1321 }
1322 
1323 #define	ZTEST_TRAVERSE_BLOCKS	1000
1324 
1325 static int
1326 ztest_blk_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
1327 {
1328 	ztest_args_t *za = arg;
1329 	zbookmark_t *zb = &bc->bc_bookmark;
1330 	blkptr_t *bp = &bc->bc_blkptr;
1331 	dnode_phys_t *dnp = bc->bc_dnode;
1332 	traverse_handle_t *th = za->za_th;
1333 	uint64_t size = BP_GET_LSIZE(bp);
1334 
1335 	/*
1336 	 * Level -1 indicates the objset_phys_t or something in its intent log.
1337 	 */
1338 	if (zb->zb_level == -1) {
1339 		if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
1340 			ASSERT3U(zb->zb_object, ==, 0);
1341 			ASSERT3U(zb->zb_blkid, ==, 0);
1342 			ASSERT3U(size, ==, sizeof (objset_phys_t));
1343 			za->za_zil_seq = 0;
1344 		} else if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) {
1345 			ASSERT3U(zb->zb_object, ==, 0);
1346 			ASSERT3U(zb->zb_blkid, >, za->za_zil_seq);
1347 			za->za_zil_seq = zb->zb_blkid;
1348 		} else {
1349 			ASSERT3U(zb->zb_object, !=, 0);	/* lr_write_t */
1350 		}
1351 
1352 		return (0);
1353 	}
1354 
1355 	ASSERT(dnp != NULL);
1356 
1357 	if (bc->bc_errno)
1358 		return (ERESTART);
1359 
1360 	/*
1361 	 * Once in a while, abort the traverse.   We only do this to odd
1362 	 * instance numbers to ensure that even ones can run to completion.
1363 	 */
1364 	if ((za->za_instance & 1) && ztest_random(10000) == 0)
1365 		return (EINTR);
1366 
1367 	if (bp->blk_birth == 0) {
1368 		ASSERT(th->th_advance & ADVANCE_HOLES);
1369 		return (0);
1370 	}
1371 
1372 	if (zb->zb_level == 0 && !(th->th_advance & ADVANCE_DATA) &&
1373 	    bc == &th->th_cache[ZB_DN_CACHE][0]) {
1374 		ASSERT(bc->bc_data == NULL);
1375 		return (0);
1376 	}
1377 
1378 	ASSERT(bc->bc_data != NULL);
1379 
1380 	/*
1381 	 * This is an expensive question, so don't ask it too often.
1382 	 */
1383 	if (((za->za_random ^ th->th_callbacks) & 0xff) == 0) {
1384 		void *xbuf = umem_alloc(size, UMEM_NOFAIL);
1385 		if (arc_tryread(spa, bp, xbuf) == 0) {
1386 			ASSERT(bcmp(bc->bc_data, xbuf, size) == 0);
1387 		}
1388 		umem_free(xbuf, size);
1389 	}
1390 
1391 	if (zb->zb_level > 0) {
1392 		ASSERT3U(size, ==, 1ULL << dnp->dn_indblkshift);
1393 		return (0);
1394 	}
1395 
1396 	ASSERT(zb->zb_level == 0);
1397 	ASSERT3U(size, ==, dnp->dn_datablkszsec << DEV_BSHIFT);
1398 
1399 	return (0);
1400 }
1401 
1402 /*
1403  * Verify that live pool traversal works.
1404  */
1405 void
1406 ztest_traverse(ztest_args_t *za)
1407 {
1408 	spa_t *spa = dmu_objset_spa(za->za_os);
1409 	traverse_handle_t *th = za->za_th;
1410 	int rc, advance;
1411 	uint64_t cbstart, cblimit;
1412 
1413 	if (th == NULL) {
1414 		advance = 0;
1415 
1416 		if (ztest_random(2) == 0)
1417 			advance |= ADVANCE_PRE;
1418 
1419 		if (ztest_random(2) == 0)
1420 			advance |= ADVANCE_PRUNE;
1421 
1422 		if (ztest_random(2) == 0)
1423 			advance |= ADVANCE_DATA;
1424 
1425 		if (ztest_random(2) == 0)
1426 			advance |= ADVANCE_HOLES;
1427 
1428 		if (ztest_random(2) == 0)
1429 			advance |= ADVANCE_ZIL;
1430 
1431 		th = za->za_th = traverse_init(spa, ztest_blk_cb, za, advance,
1432 		    ZIO_FLAG_CANFAIL);
1433 
1434 		traverse_add_pool(th, 0, -1ULL);
1435 	}
1436 
1437 	advance = th->th_advance;
1438 	cbstart = th->th_callbacks;
1439 	cblimit = cbstart + ((advance & ADVANCE_DATA) ? 100 : 1000);
1440 
1441 	while ((rc = traverse_more(th)) == EAGAIN && th->th_callbacks < cblimit)
1442 		continue;
1443 
1444 	if (zopt_verbose >= 5)
1445 		(void) printf("traverse %s%s%s%s %llu blocks to "
1446 		    "<%llu, %llu, %lld, %llx>%s\n",
1447 		    (advance & ADVANCE_PRE) ? "pre" : "post",
1448 		    (advance & ADVANCE_PRUNE) ? "|prune" : "",
1449 		    (advance & ADVANCE_DATA) ? "|data" : "",
1450 		    (advance & ADVANCE_HOLES) ? "|holes" : "",
1451 		    (u_longlong_t)(th->th_callbacks - cbstart),
1452 		    (u_longlong_t)th->th_lastcb.zb_objset,
1453 		    (u_longlong_t)th->th_lastcb.zb_object,
1454 		    (u_longlong_t)th->th_lastcb.zb_level,
1455 		    (u_longlong_t)th->th_lastcb.zb_blkid,
1456 		    rc == 0 ? " [done]" :
1457 		    rc == EINTR ? " [aborted]" :
1458 		    rc == EAGAIN ? "" :
1459 		    strerror(rc));
1460 
1461 	if (rc != EAGAIN) {
1462 		if (rc != 0 && rc != EINTR)
1463 			fatal(0, "traverse_more(%p) = %d", th, rc);
1464 		traverse_fini(th);
1465 		za->za_th = NULL;
1466 	}
1467 }
1468 
1469 /*
1470  * Verify that dmu_object_{alloc,free} work as expected.
1471  */
1472 void
1473 ztest_dmu_object_alloc_free(ztest_args_t *za)
1474 {
1475 	objset_t *os = za->za_os;
1476 	dmu_buf_t *db;
1477 	dmu_tx_t *tx;
1478 	uint64_t batchobj, object, batchsize, endoff, temp;
1479 	int b, c, error, bonuslen;
1480 	dmu_object_info_t doi;
1481 	char osname[MAXNAMELEN];
1482 
1483 	dmu_objset_name(os, osname);
1484 
1485 	endoff = -8ULL;
1486 	batchsize = 2;
1487 
1488 	/*
1489 	 * Create a batch object if necessary, and record it in the directory.
1490 	 */
1491 	VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
1492 	    sizeof (uint64_t), &batchobj));
1493 	if (batchobj == 0) {
1494 		tx = dmu_tx_create(os);
1495 		dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff,
1496 		    sizeof (uint64_t));
1497 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1498 		error = dmu_tx_assign(tx, TXG_WAIT);
1499 		if (error) {
1500 			ztest_record_enospc("create a batch object");
1501 			dmu_tx_abort(tx);
1502 			return;
1503 		}
1504 		batchobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1505 		    DMU_OT_NONE, 0, tx);
1506 		ztest_set_random_blocksize(os, batchobj, tx);
1507 		dmu_write(os, ZTEST_DIROBJ, za->za_diroff,
1508 		    sizeof (uint64_t), &batchobj, tx);
1509 		dmu_tx_commit(tx);
1510 	}
1511 
1512 	/*
1513 	 * Destroy the previous batch of objects.
1514 	 */
1515 	for (b = 0; b < batchsize; b++) {
1516 		VERIFY(0 == dmu_read(os, batchobj, b * sizeof (uint64_t),
1517 		    sizeof (uint64_t), &object));
1518 		if (object == 0)
1519 			continue;
1520 		/*
1521 		 * Read and validate contents.
1522 		 * We expect the nth byte of the bonus buffer to be n.
1523 		 */
1524 		VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db));
1525 
1526 		dmu_object_info_from_db(db, &doi);
1527 		ASSERT(doi.doi_type == DMU_OT_UINT64_OTHER);
1528 		ASSERT(doi.doi_bonus_type == DMU_OT_PLAIN_OTHER);
1529 		ASSERT3S(doi.doi_physical_blks, >=, 0);
1530 
1531 		bonuslen = db->db_size;
1532 
1533 		for (c = 0; c < bonuslen; c++) {
1534 			if (((uint8_t *)db->db_data)[c] !=
1535 			    (uint8_t)(c + bonuslen)) {
1536 				fatal(0,
1537 				    "bad bonus: %s, obj %llu, off %d: %u != %u",
1538 				    osname, object, c,
1539 				    ((uint8_t *)db->db_data)[c],
1540 				    (uint8_t)(c + bonuslen));
1541 			}
1542 		}
1543 
1544 		dmu_buf_rele(db, FTAG);
1545 
1546 		/*
1547 		 * We expect the word at endoff to be our object number.
1548 		 */
1549 		VERIFY(0 == dmu_read(os, object, endoff,
1550 		    sizeof (uint64_t), &temp));
1551 
1552 		if (temp != object) {
1553 			fatal(0, "bad data in %s, got %llu, expected %llu",
1554 			    osname, temp, object);
1555 		}
1556 
1557 		/*
1558 		 * Destroy old object and clear batch entry.
1559 		 */
1560 		tx = dmu_tx_create(os);
1561 		dmu_tx_hold_write(tx, batchobj,
1562 		    b * sizeof (uint64_t), sizeof (uint64_t));
1563 		dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1564 		error = dmu_tx_assign(tx, TXG_WAIT);
1565 		if (error) {
1566 			ztest_record_enospc("free object");
1567 			dmu_tx_abort(tx);
1568 			return;
1569 		}
1570 		error = dmu_object_free(os, object, tx);
1571 		if (error) {
1572 			fatal(0, "dmu_object_free('%s', %llu) = %d",
1573 			    osname, object, error);
1574 		}
1575 		object = 0;
1576 
1577 		dmu_object_set_checksum(os, batchobj,
1578 		    ztest_random_checksum(), tx);
1579 		dmu_object_set_compress(os, batchobj,
1580 		    ztest_random_compress(), tx);
1581 
1582 		dmu_write(os, batchobj, b * sizeof (uint64_t),
1583 		    sizeof (uint64_t), &object, tx);
1584 
1585 		dmu_tx_commit(tx);
1586 	}
1587 
1588 	/*
1589 	 * Before creating the new batch of objects, generate a bunch of churn.
1590 	 */
1591 	for (b = ztest_random(100); b > 0; b--) {
1592 		tx = dmu_tx_create(os);
1593 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1594 		error = dmu_tx_assign(tx, TXG_WAIT);
1595 		if (error) {
1596 			ztest_record_enospc("churn objects");
1597 			dmu_tx_abort(tx);
1598 			return;
1599 		}
1600 		object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1601 		    DMU_OT_NONE, 0, tx);
1602 		ztest_set_random_blocksize(os, object, tx);
1603 		error = dmu_object_free(os, object, tx);
1604 		if (error) {
1605 			fatal(0, "dmu_object_free('%s', %llu) = %d",
1606 			    osname, object, error);
1607 		}
1608 		dmu_tx_commit(tx);
1609 	}
1610 
1611 	/*
1612 	 * Create a new batch of objects with randomly chosen
1613 	 * blocksizes and record them in the batch directory.
1614 	 */
1615 	for (b = 0; b < batchsize; b++) {
1616 		uint32_t va_blksize;
1617 		u_longlong_t va_nblocks;
1618 
1619 		tx = dmu_tx_create(os);
1620 		dmu_tx_hold_write(tx, batchobj, b * sizeof (uint64_t),
1621 		    sizeof (uint64_t));
1622 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1623 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, endoff,
1624 		    sizeof (uint64_t));
1625 		error = dmu_tx_assign(tx, TXG_WAIT);
1626 		if (error) {
1627 			ztest_record_enospc("create batchobj");
1628 			dmu_tx_abort(tx);
1629 			return;
1630 		}
1631 		bonuslen = (int)ztest_random(dmu_bonus_max()) + 1;
1632 
1633 		object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1634 		    DMU_OT_PLAIN_OTHER, bonuslen, tx);
1635 
1636 		ztest_set_random_blocksize(os, object, tx);
1637 
1638 		dmu_object_set_checksum(os, object,
1639 		    ztest_random_checksum(), tx);
1640 		dmu_object_set_compress(os, object,
1641 		    ztest_random_compress(), tx);
1642 
1643 		dmu_write(os, batchobj, b * sizeof (uint64_t),
1644 		    sizeof (uint64_t), &object, tx);
1645 
1646 		/*
1647 		 * Write to both the bonus buffer and the regular data.
1648 		 */
1649 		VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db));
1650 		ASSERT3U(bonuslen, ==, db->db_size);
1651 
1652 		dmu_object_size_from_db(db, &va_blksize, &va_nblocks);
1653 		ASSERT3S(va_nblocks, >=, 0);
1654 
1655 		dmu_buf_will_dirty(db, tx);
1656 
1657 		/*
1658 		 * See comments above regarding the contents of
1659 		 * the bonus buffer and the word at endoff.
1660 		 */
1661 		for (c = 0; c < db->db_size; c++)
1662 			((uint8_t *)db->db_data)[c] = (uint8_t)(c + bonuslen);
1663 
1664 		dmu_buf_rele(db, FTAG);
1665 
1666 		/*
1667 		 * Write to a large offset to increase indirection.
1668 		 */
1669 		dmu_write(os, object, endoff, sizeof (uint64_t), &object, tx);
1670 
1671 		dmu_tx_commit(tx);
1672 	}
1673 }
1674 
1675 /*
1676  * Verify that dmu_{read,write} work as expected.
1677  */
1678 typedef struct bufwad {
1679 	uint64_t	bw_index;
1680 	uint64_t	bw_txg;
1681 	uint64_t	bw_data;
1682 } bufwad_t;
1683 
1684 typedef struct dmu_read_write_dir {
1685 	uint64_t	dd_packobj;
1686 	uint64_t	dd_bigobj;
1687 	uint64_t	dd_chunk;
1688 } dmu_read_write_dir_t;
1689 
1690 void
1691 ztest_dmu_read_write(ztest_args_t *za)
1692 {
1693 	objset_t *os = za->za_os;
1694 	dmu_read_write_dir_t dd;
1695 	dmu_tx_t *tx;
1696 	int i, freeit, error;
1697 	uint64_t n, s, txg;
1698 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
1699 	uint64_t packoff, packsize, bigoff, bigsize;
1700 	uint64_t regions = 997;
1701 	uint64_t stride = 123456789ULL;
1702 	uint64_t width = 40;
1703 	int free_percent = 5;
1704 
1705 	/*
1706 	 * This test uses two objects, packobj and bigobj, that are always
1707 	 * updated together (i.e. in the same tx) so that their contents are
1708 	 * in sync and can be compared.  Their contents relate to each other
1709 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
1710 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
1711 	 * for any index n, there are three bufwads that should be identical:
1712 	 *
1713 	 *	packobj, at offset n * sizeof (bufwad_t)
1714 	 *	bigobj, at the head of the nth chunk
1715 	 *	bigobj, at the tail of the nth chunk
1716 	 *
1717 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
1718 	 * and it doesn't have any relation to the object blocksize.
1719 	 * The only requirement is that it can hold at least two bufwads.
1720 	 *
1721 	 * Normally, we write the bufwad to each of these locations.
1722 	 * However, free_percent of the time we instead write zeroes to
1723 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
1724 	 * bigobj to packobj, we can verify that the DMU is correctly
1725 	 * tracking which parts of an object are allocated and free,
1726 	 * and that the contents of the allocated blocks are correct.
1727 	 */
1728 
1729 	/*
1730 	 * Read the directory info.  If it's the first time, set things up.
1731 	 */
1732 	VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
1733 	    sizeof (dd), &dd));
1734 	if (dd.dd_chunk == 0) {
1735 		ASSERT(dd.dd_packobj == 0);
1736 		ASSERT(dd.dd_bigobj == 0);
1737 		tx = dmu_tx_create(os);
1738 		dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (dd));
1739 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1740 		error = dmu_tx_assign(tx, TXG_WAIT);
1741 		if (error) {
1742 			ztest_record_enospc("create r/w directory");
1743 			dmu_tx_abort(tx);
1744 			return;
1745 		}
1746 
1747 		dd.dd_packobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1748 		    DMU_OT_NONE, 0, tx);
1749 		dd.dd_bigobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
1750 		    DMU_OT_NONE, 0, tx);
1751 		dd.dd_chunk = (1000 + ztest_random(1000)) * sizeof (uint64_t);
1752 
1753 		ztest_set_random_blocksize(os, dd.dd_packobj, tx);
1754 		ztest_set_random_blocksize(os, dd.dd_bigobj, tx);
1755 
1756 		dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (dd), &dd,
1757 		    tx);
1758 		dmu_tx_commit(tx);
1759 	}
1760 
1761 	/*
1762 	 * Prefetch a random chunk of the big object.
1763 	 * Our aim here is to get some async reads in flight
1764 	 * for blocks that we may free below; the DMU should
1765 	 * handle this race correctly.
1766 	 */
1767 	n = ztest_random(regions) * stride + ztest_random(width);
1768 	s = 1 + ztest_random(2 * width - 1);
1769 	dmu_prefetch(os, dd.dd_bigobj, n * dd.dd_chunk, s * dd.dd_chunk);
1770 
1771 	/*
1772 	 * Pick a random index and compute the offsets into packobj and bigobj.
1773 	 */
1774 	n = ztest_random(regions) * stride + ztest_random(width);
1775 	s = 1 + ztest_random(width - 1);
1776 
1777 	packoff = n * sizeof (bufwad_t);
1778 	packsize = s * sizeof (bufwad_t);
1779 
1780 	bigoff = n * dd.dd_chunk;
1781 	bigsize = s * dd.dd_chunk;
1782 
1783 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
1784 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
1785 
1786 	/*
1787 	 * free_percent of the time, free a range of bigobj rather than
1788 	 * overwriting it.
1789 	 */
1790 	freeit = (ztest_random(100) < free_percent);
1791 
1792 	/*
1793 	 * Read the current contents of our objects.
1794 	 */
1795 	error = dmu_read(os, dd.dd_packobj, packoff, packsize, packbuf);
1796 	ASSERT3U(error, ==, 0);
1797 	error = dmu_read(os, dd.dd_bigobj, bigoff, bigsize, bigbuf);
1798 	ASSERT3U(error, ==, 0);
1799 
1800 	/*
1801 	 * Get a tx for the mods to both packobj and bigobj.
1802 	 */
1803 	tx = dmu_tx_create(os);
1804 
1805 	dmu_tx_hold_write(tx, dd.dd_packobj, packoff, packsize);
1806 
1807 	if (freeit)
1808 		dmu_tx_hold_free(tx, dd.dd_bigobj, bigoff, bigsize);
1809 	else
1810 		dmu_tx_hold_write(tx, dd.dd_bigobj, bigoff, bigsize);
1811 
1812 	error = dmu_tx_assign(tx, TXG_WAIT);
1813 
1814 	if (error) {
1815 		ztest_record_enospc("dmu r/w range");
1816 		dmu_tx_abort(tx);
1817 		umem_free(packbuf, packsize);
1818 		umem_free(bigbuf, bigsize);
1819 		return;
1820 	}
1821 
1822 	txg = dmu_tx_get_txg(tx);
1823 
1824 	/*
1825 	 * For each index from n to n + s, verify that the existing bufwad
1826 	 * in packobj matches the bufwads at the head and tail of the
1827 	 * corresponding chunk in bigobj.  Then update all three bufwads
1828 	 * with the new values we want to write out.
1829 	 */
1830 	for (i = 0; i < s; i++) {
1831 		/* LINTED */
1832 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
1833 		/* LINTED */
1834 		bigH = (bufwad_t *)((char *)bigbuf + i * dd.dd_chunk);
1835 		/* LINTED */
1836 		bigT = (bufwad_t *)((char *)bigH + dd.dd_chunk) - 1;
1837 
1838 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
1839 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
1840 
1841 		if (pack->bw_txg > txg)
1842 			fatal(0, "future leak: got %llx, open txg is %llx",
1843 			    pack->bw_txg, txg);
1844 
1845 		if (pack->bw_data != 0 && pack->bw_index != n + i)
1846 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
1847 			    pack->bw_index, n, i);
1848 
1849 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
1850 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
1851 
1852 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
1853 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
1854 
1855 		if (freeit) {
1856 			bzero(pack, sizeof (bufwad_t));
1857 		} else {
1858 			pack->bw_index = n + i;
1859 			pack->bw_txg = txg;
1860 			pack->bw_data = 1 + ztest_random(-2ULL);
1861 		}
1862 		*bigH = *pack;
1863 		*bigT = *pack;
1864 	}
1865 
1866 	/*
1867 	 * We've verified all the old bufwads, and made new ones.
1868 	 * Now write them out.
1869 	 */
1870 	dmu_write(os, dd.dd_packobj, packoff, packsize, packbuf, tx);
1871 
1872 	if (freeit) {
1873 		if (zopt_verbose >= 6) {
1874 			(void) printf("freeing offset %llx size %llx"
1875 			    " txg %llx\n",
1876 			    (u_longlong_t)bigoff,
1877 			    (u_longlong_t)bigsize,
1878 			    (u_longlong_t)txg);
1879 		}
1880 		VERIFY(0 == dmu_free_range(os, dd.dd_bigobj, bigoff,
1881 		    bigsize, tx));
1882 	} else {
1883 		if (zopt_verbose >= 6) {
1884 			(void) printf("writing offset %llx size %llx"
1885 			    " txg %llx\n",
1886 			    (u_longlong_t)bigoff,
1887 			    (u_longlong_t)bigsize,
1888 			    (u_longlong_t)txg);
1889 		}
1890 		dmu_write(os, dd.dd_bigobj, bigoff, bigsize, bigbuf, tx);
1891 	}
1892 
1893 	dmu_tx_commit(tx);
1894 
1895 	/*
1896 	 * Sanity check the stuff we just wrote.
1897 	 */
1898 	{
1899 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
1900 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
1901 
1902 		VERIFY(0 == dmu_read(os, dd.dd_packobj, packoff,
1903 		    packsize, packcheck));
1904 		VERIFY(0 == dmu_read(os, dd.dd_bigobj, bigoff,
1905 		    bigsize, bigcheck));
1906 
1907 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
1908 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
1909 
1910 		umem_free(packcheck, packsize);
1911 		umem_free(bigcheck, bigsize);
1912 	}
1913 
1914 	umem_free(packbuf, packsize);
1915 	umem_free(bigbuf, bigsize);
1916 }
1917 
1918 void
1919 ztest_dmu_write_parallel(ztest_args_t *za)
1920 {
1921 	objset_t *os = za->za_os;
1922 	dmu_tx_t *tx;
1923 	dmu_buf_t *db;
1924 	int i, b, error, do_free, bs;
1925 	uint64_t off, txg_how, txg;
1926 	mutex_t *lp;
1927 	char osname[MAXNAMELEN];
1928 	char iobuf[SPA_MAXBLOCKSIZE];
1929 	ztest_block_tag_t rbt, wbt;
1930 
1931 	dmu_objset_name(os, osname);
1932 	bs = ZTEST_DIROBJ_BLOCKSIZE;
1933 
1934 	/*
1935 	 * Have multiple threads write to large offsets in ZTEST_DIROBJ
1936 	 * to verify that having multiple threads writing to the same object
1937 	 * in parallel doesn't cause any trouble.
1938 	 * Also do parallel writes to the bonus buffer on occasion.
1939 	 */
1940 	for (i = 0; i < 50; i++) {
1941 		b = ztest_random(ZTEST_SYNC_LOCKS);
1942 		lp = &ztest_shared->zs_sync_lock[b];
1943 
1944 		do_free = (ztest_random(4) == 0);
1945 
1946 		off = za->za_diroff_shared + ((uint64_t)b << SPA_MAXBLOCKSHIFT);
1947 
1948 		if (ztest_random(4) == 0) {
1949 			/*
1950 			 * Do the bonus buffer instead of a regular block.
1951 			 */
1952 			do_free = 0;
1953 			off = -1ULL;
1954 		}
1955 
1956 		tx = dmu_tx_create(os);
1957 
1958 		if (off == -1ULL)
1959 			dmu_tx_hold_bonus(tx, ZTEST_DIROBJ);
1960 		else if (do_free)
1961 			dmu_tx_hold_free(tx, ZTEST_DIROBJ, off, bs);
1962 		else
1963 			dmu_tx_hold_write(tx, ZTEST_DIROBJ, off, bs);
1964 
1965 		txg_how = ztest_random(2) == 0 ? TXG_WAIT : TXG_NOWAIT;
1966 		error = dmu_tx_assign(tx, txg_how);
1967 		if (error) {
1968 			if (error == ERESTART) {
1969 				ASSERT(txg_how == TXG_NOWAIT);
1970 				dmu_tx_wait(tx);
1971 				dmu_tx_abort(tx);
1972 				continue;
1973 			}
1974 			dmu_tx_abort(tx);
1975 			ztest_record_enospc("dmu write parallel");
1976 			return;
1977 		}
1978 		txg = dmu_tx_get_txg(tx);
1979 
1980 		if (do_free) {
1981 			(void) mutex_lock(lp);
1982 			VERIFY(0 == dmu_free_range(os, ZTEST_DIROBJ, off,
1983 			    bs, tx));
1984 			(void) mutex_unlock(lp);
1985 			dmu_tx_commit(tx);
1986 			continue;
1987 		}
1988 
1989 		wbt.bt_objset = dmu_objset_id(os);
1990 		wbt.bt_object = ZTEST_DIROBJ;
1991 		wbt.bt_offset = off;
1992 		wbt.bt_txg = txg;
1993 		wbt.bt_thread = za->za_instance;
1994 
1995 		if (off == -1ULL) {
1996 			wbt.bt_seq = 0;
1997 			VERIFY(0 == dmu_bonus_hold(os, ZTEST_DIROBJ,
1998 			    FTAG, &db));
1999 			ASSERT3U(db->db_size, ==, sizeof (wbt));
2000 			bcopy(db->db_data, &rbt, db->db_size);
2001 			if (rbt.bt_objset != 0) {
2002 				ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset);
2003 				ASSERT3U(rbt.bt_object, ==, wbt.bt_object);
2004 				ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset);
2005 				ASSERT3U(rbt.bt_txg, <=, wbt.bt_txg);
2006 			}
2007 			dmu_buf_will_dirty(db, tx);
2008 			bcopy(&wbt, db->db_data, db->db_size);
2009 			dmu_buf_rele(db, FTAG);
2010 			dmu_tx_commit(tx);
2011 			continue;
2012 		}
2013 
2014 		(void) mutex_lock(lp);
2015 
2016 		wbt.bt_seq = ztest_shared->zs_seq[b]++;
2017 
2018 		dmu_write(os, ZTEST_DIROBJ, off, sizeof (wbt), &wbt, tx);
2019 
2020 		(void) mutex_unlock(lp);
2021 
2022 		if (ztest_random(100) == 0)
2023 			(void) poll(NULL, 0, 1); /* open dn_notxholds window */
2024 
2025 		dmu_tx_commit(tx);
2026 
2027 		if (ztest_random(1000) == 0)
2028 			txg_wait_synced(dmu_objset_pool(os), txg);
2029 
2030 		if (ztest_random(2) == 0) {
2031 			blkptr_t blk = { 0 };
2032 			uint64_t blkoff;
2033 			zbookmark_t zb;
2034 
2035 			(void) mutex_lock(lp);
2036 			blkoff = P2ALIGN_TYPED(off, bs, uint64_t);
2037 			error = dmu_buf_hold(os,
2038 			    ZTEST_DIROBJ, blkoff, FTAG, &db);
2039 			if (error) {
2040 				dprintf("dmu_buf_hold(%s, %d, %llx) = %d\n",
2041 				    osname, ZTEST_DIROBJ, blkoff, error);
2042 				(void) mutex_unlock(lp);
2043 				continue;
2044 			}
2045 			blkoff = off - blkoff;
2046 			error = dmu_sync(NULL, db, &blk, txg, NULL, NULL);
2047 			dmu_buf_rele(db, FTAG);
2048 			(void) mutex_unlock(lp);
2049 			if (error) {
2050 				dprintf("dmu_sync(%s, %d, %llx) = %d\n",
2051 				    osname, ZTEST_DIROBJ, off, error);
2052 				continue;
2053 			}
2054 
2055 			if (blk.blk_birth == 0)	{	/* concurrent free */
2056 				continue;
2057 			}
2058 			txg_suspend(dmu_objset_pool(os));
2059 
2060 			ASSERT(blk.blk_fill == 1);
2061 			ASSERT3U(BP_GET_TYPE(&blk), ==, DMU_OT_UINT64_OTHER);
2062 			ASSERT3U(BP_GET_LEVEL(&blk), ==, 0);
2063 			ASSERT3U(BP_GET_LSIZE(&blk), ==, bs);
2064 
2065 			/*
2066 			 * Read the block that dmu_sync() returned to
2067 			 * make sure its contents match what we wrote.
2068 			 * We do this while still txg_suspend()ed to ensure
2069 			 * that the block can't be reused before we read it.
2070 			 */
2071 			zb.zb_objset = dmu_objset_id(os);
2072 			zb.zb_object = ZTEST_DIROBJ;
2073 			zb.zb_level = 0;
2074 			zb.zb_blkid = off / bs;
2075 			error = zio_wait(zio_read(NULL, dmu_objset_spa(os),
2076 			    &blk, iobuf, bs, NULL, NULL,
2077 			    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_MUSTSUCCEED, &zb));
2078 			ASSERT(error == 0);
2079 
2080 			txg_resume(dmu_objset_pool(os));
2081 
2082 			bcopy(&iobuf[blkoff], &rbt, sizeof (rbt));
2083 
2084 			if (rbt.bt_objset == 0)		/* concurrent free */
2085 				continue;
2086 
2087 			ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset);
2088 			ASSERT3U(rbt.bt_object, ==, wbt.bt_object);
2089 			ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset);
2090 
2091 			/*
2092 			 * The semantic of dmu_sync() is that we always
2093 			 * push the most recent version of the data,
2094 			 * so in the face of concurrent updates we may
2095 			 * see a newer version of the block.  That's OK.
2096 			 */
2097 			ASSERT3U(rbt.bt_txg, >=, wbt.bt_txg);
2098 			if (rbt.bt_thread == wbt.bt_thread)
2099 				ASSERT3U(rbt.bt_seq, ==, wbt.bt_seq);
2100 			else
2101 				ASSERT3U(rbt.bt_seq, >, wbt.bt_seq);
2102 		}
2103 	}
2104 }
2105 
2106 /*
2107  * Verify that zap_{create,destroy,add,remove,update} work as expected.
2108  */
2109 #define	ZTEST_ZAP_MIN_INTS	1
2110 #define	ZTEST_ZAP_MAX_INTS	4
2111 #define	ZTEST_ZAP_MAX_PROPS	1000
2112 
2113 void
2114 ztest_zap(ztest_args_t *za)
2115 {
2116 	objset_t *os = za->za_os;
2117 	uint64_t object;
2118 	uint64_t txg, last_txg;
2119 	uint64_t value[ZTEST_ZAP_MAX_INTS];
2120 	uint64_t zl_ints, zl_intsize, prop;
2121 	int i, ints;
2122 	int iters = 100;
2123 	dmu_tx_t *tx;
2124 	char propname[100], txgname[100];
2125 	int error;
2126 	char osname[MAXNAMELEN];
2127 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
2128 
2129 	dmu_objset_name(os, osname);
2130 
2131 	/*
2132 	 * Create a new object if necessary, and record it in the directory.
2133 	 */
2134 	VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff,
2135 	    sizeof (uint64_t), &object));
2136 
2137 	if (object == 0) {
2138 		tx = dmu_tx_create(os);
2139 		dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff,
2140 		    sizeof (uint64_t));
2141 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL);
2142 		error = dmu_tx_assign(tx, TXG_WAIT);
2143 		if (error) {
2144 			ztest_record_enospc("create zap test obj");
2145 			dmu_tx_abort(tx);
2146 			return;
2147 		}
2148 		object = zap_create(os, DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx);
2149 		if (error) {
2150 			fatal(0, "zap_create('%s', %llu) = %d",
2151 			    osname, object, error);
2152 		}
2153 		ASSERT(object != 0);
2154 		dmu_write(os, ZTEST_DIROBJ, za->za_diroff,
2155 		    sizeof (uint64_t), &object, tx);
2156 		/*
2157 		 * Generate a known hash collision, and verify that
2158 		 * we can lookup and remove both entries.
2159 		 */
2160 		for (i = 0; i < 2; i++) {
2161 			value[i] = i;
2162 			error = zap_add(os, object, hc[i], sizeof (uint64_t),
2163 			    1, &value[i], tx);
2164 			ASSERT3U(error, ==, 0);
2165 		}
2166 		for (i = 0; i < 2; i++) {
2167 			error = zap_add(os, object, hc[i], sizeof (uint64_t),
2168 			    1, &value[i], tx);
2169 			ASSERT3U(error, ==, EEXIST);
2170 			error = zap_length(os, object, hc[i],
2171 			    &zl_intsize, &zl_ints);
2172 			ASSERT3U(error, ==, 0);
2173 			ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2174 			ASSERT3U(zl_ints, ==, 1);
2175 		}
2176 		for (i = 0; i < 2; i++) {
2177 			error = zap_remove(os, object, hc[i], tx);
2178 			ASSERT3U(error, ==, 0);
2179 		}
2180 
2181 		dmu_tx_commit(tx);
2182 	}
2183 
2184 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
2185 
2186 	while (--iters >= 0) {
2187 		prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
2188 		(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
2189 		(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
2190 		bzero(value, sizeof (value));
2191 		last_txg = 0;
2192 
2193 		/*
2194 		 * If these zap entries already exist, validate their contents.
2195 		 */
2196 		error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
2197 		if (error == 0) {
2198 			ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2199 			ASSERT3U(zl_ints, ==, 1);
2200 
2201 			error = zap_lookup(os, object, txgname, zl_intsize,
2202 			    zl_ints, &last_txg);
2203 
2204 			ASSERT3U(error, ==, 0);
2205 
2206 			error = zap_length(os, object, propname, &zl_intsize,
2207 			    &zl_ints);
2208 
2209 			ASSERT3U(error, ==, 0);
2210 			ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
2211 			ASSERT3U(zl_ints, ==, ints);
2212 
2213 			error = zap_lookup(os, object, propname, zl_intsize,
2214 			    zl_ints, value);
2215 
2216 			ASSERT3U(error, ==, 0);
2217 
2218 			for (i = 0; i < ints; i++) {
2219 				ASSERT3U(value[i], ==, last_txg + object + i);
2220 			}
2221 		} else {
2222 			ASSERT3U(error, ==, ENOENT);
2223 		}
2224 
2225 		/*
2226 		 * Atomically update two entries in our zap object.
2227 		 * The first is named txg_%llu, and contains the txg
2228 		 * in which the property was last updated.  The second
2229 		 * is named prop_%llu, and the nth element of its value
2230 		 * should be txg + object + n.
2231 		 */
2232 		tx = dmu_tx_create(os);
2233 		dmu_tx_hold_zap(tx, object, TRUE, NULL);
2234 		error = dmu_tx_assign(tx, TXG_WAIT);
2235 		if (error) {
2236 			ztest_record_enospc("create zap entry");
2237 			dmu_tx_abort(tx);
2238 			return;
2239 		}
2240 		txg = dmu_tx_get_txg(tx);
2241 
2242 		if (last_txg > txg)
2243 			fatal(0, "zap future leak: old %llu new %llu",
2244 			    last_txg, txg);
2245 
2246 		for (i = 0; i < ints; i++)
2247 			value[i] = txg + object + i;
2248 
2249 		error = zap_update(os, object, txgname, sizeof (uint64_t),
2250 		    1, &txg, tx);
2251 		if (error)
2252 			fatal(0, "zap_update('%s', %llu, '%s') = %d",
2253 			    osname, object, txgname, error);
2254 
2255 		error = zap_update(os, object, propname, sizeof (uint64_t),
2256 		    ints, value, tx);
2257 		if (error)
2258 			fatal(0, "zap_update('%s', %llu, '%s') = %d",
2259 			    osname, object, propname, error);
2260 
2261 		dmu_tx_commit(tx);
2262 
2263 		/*
2264 		 * Remove a random pair of entries.
2265 		 */
2266 		prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
2267 		(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
2268 		(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
2269 
2270 		error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
2271 
2272 		if (error == ENOENT)
2273 			continue;
2274 
2275 		ASSERT3U(error, ==, 0);
2276 
2277 		tx = dmu_tx_create(os);
2278 		dmu_tx_hold_zap(tx, object, TRUE, NULL);
2279 		error = dmu_tx_assign(tx, TXG_WAIT);
2280 		if (error) {
2281 			ztest_record_enospc("remove zap entry");
2282 			dmu_tx_abort(tx);
2283 			return;
2284 		}
2285 		error = zap_remove(os, object, txgname, tx);
2286 		if (error)
2287 			fatal(0, "zap_remove('%s', %llu, '%s') = %d",
2288 			    osname, object, txgname, error);
2289 
2290 		error = zap_remove(os, object, propname, tx);
2291 		if (error)
2292 			fatal(0, "zap_remove('%s', %llu, '%s') = %d",
2293 			    osname, object, propname, error);
2294 
2295 		dmu_tx_commit(tx);
2296 	}
2297 
2298 	/*
2299 	 * Once in a while, destroy the object.
2300 	 */
2301 	if (ztest_random(100) != 0)
2302 		return;
2303 
2304 	tx = dmu_tx_create(os);
2305 	dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t));
2306 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2307 	error = dmu_tx_assign(tx, TXG_WAIT);
2308 	if (error) {
2309 		ztest_record_enospc("destroy zap object");
2310 		dmu_tx_abort(tx);
2311 		return;
2312 	}
2313 	error = zap_destroy(os, object, tx);
2314 	if (error)
2315 		fatal(0, "zap_destroy('%s', %llu) = %d",
2316 		    osname, object, error);
2317 	object = 0;
2318 	dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t),
2319 	    &object, tx);
2320 	dmu_tx_commit(tx);
2321 }
2322 
2323 void
2324 ztest_zap_parallel(ztest_args_t *za)
2325 {
2326 	objset_t *os = za->za_os;
2327 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
2328 	int iters = 100;
2329 	dmu_tx_t *tx;
2330 	int i, namelen, error;
2331 	char name[20], string_value[20];
2332 	void *data;
2333 
2334 	while (--iters >= 0) {
2335 		/*
2336 		 * Generate a random name of the form 'xxx.....' where each
2337 		 * x is a random printable character and the dots are dots.
2338 		 * There are 94 such characters, and the name length goes from
2339 		 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
2340 		 */
2341 		namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
2342 
2343 		for (i = 0; i < 3; i++)
2344 			name[i] = '!' + ztest_random('~' - '!' + 1);
2345 		for (; i < namelen - 1; i++)
2346 			name[i] = '.';
2347 		name[i] = '\0';
2348 
2349 		if (ztest_random(2) == 0)
2350 			object = ZTEST_MICROZAP_OBJ;
2351 		else
2352 			object = ZTEST_FATZAP_OBJ;
2353 
2354 		if ((namelen & 1) || object == ZTEST_MICROZAP_OBJ) {
2355 			wsize = sizeof (txg);
2356 			wc = 1;
2357 			data = &txg;
2358 		} else {
2359 			wsize = 1;
2360 			wc = namelen;
2361 			data = string_value;
2362 		}
2363 
2364 		count = -1ULL;
2365 		VERIFY(zap_count(os, object, &count) == 0);
2366 		ASSERT(count != -1ULL);
2367 
2368 		/*
2369 		 * Select an operation: length, lookup, add, update, remove.
2370 		 */
2371 		i = ztest_random(5);
2372 
2373 		if (i >= 2) {
2374 			tx = dmu_tx_create(os);
2375 			dmu_tx_hold_zap(tx, object, TRUE, NULL);
2376 			error = dmu_tx_assign(tx, TXG_WAIT);
2377 			if (error) {
2378 				ztest_record_enospc("zap parallel");
2379 				dmu_tx_abort(tx);
2380 				return;
2381 			}
2382 			txg = dmu_tx_get_txg(tx);
2383 			bcopy(name, string_value, namelen);
2384 		} else {
2385 			tx = NULL;
2386 			txg = 0;
2387 			bzero(string_value, namelen);
2388 		}
2389 
2390 		switch (i) {
2391 
2392 		case 0:
2393 			error = zap_length(os, object, name, &zl_wsize, &zl_wc);
2394 			if (error == 0) {
2395 				ASSERT3U(wsize, ==, zl_wsize);
2396 				ASSERT3U(wc, ==, zl_wc);
2397 			} else {
2398 				ASSERT3U(error, ==, ENOENT);
2399 			}
2400 			break;
2401 
2402 		case 1:
2403 			error = zap_lookup(os, object, name, wsize, wc, data);
2404 			if (error == 0) {
2405 				if (data == string_value &&
2406 				    bcmp(name, data, namelen) != 0)
2407 					fatal(0, "name '%s' != val '%s' len %d",
2408 					    name, data, namelen);
2409 			} else {
2410 				ASSERT3U(error, ==, ENOENT);
2411 			}
2412 			break;
2413 
2414 		case 2:
2415 			error = zap_add(os, object, name, wsize, wc, data, tx);
2416 			ASSERT(error == 0 || error == EEXIST);
2417 			break;
2418 
2419 		case 3:
2420 			VERIFY(zap_update(os, object, name, wsize, wc,
2421 			    data, tx) == 0);
2422 			break;
2423 
2424 		case 4:
2425 			error = zap_remove(os, object, name, tx);
2426 			ASSERT(error == 0 || error == ENOENT);
2427 			break;
2428 		}
2429 
2430 		if (tx != NULL)
2431 			dmu_tx_commit(tx);
2432 	}
2433 }
2434 
2435 void
2436 ztest_dsl_prop_get_set(ztest_args_t *za)
2437 {
2438 	objset_t *os = za->za_os;
2439 	int i, inherit;
2440 	uint64_t value;
2441 	const char *prop, *valname;
2442 	char setpoint[MAXPATHLEN];
2443 	char osname[MAXNAMELEN];
2444 	int error;
2445 
2446 	(void) rw_rdlock(&ztest_shared->zs_name_lock);
2447 
2448 	dmu_objset_name(os, osname);
2449 
2450 	for (i = 0; i < 2; i++) {
2451 		if (i == 0) {
2452 			prop = "checksum";
2453 			value = ztest_random_checksum();
2454 			inherit = (value == ZIO_CHECKSUM_INHERIT);
2455 		} else {
2456 			prop = "compression";
2457 			value = ztest_random_compress();
2458 			inherit = (value == ZIO_COMPRESS_INHERIT);
2459 		}
2460 
2461 		error = dsl_prop_set(osname, prop, sizeof (value),
2462 		    !inherit, &value);
2463 
2464 		if (error == ENOSPC) {
2465 			ztest_record_enospc("dsl_prop_set");
2466 			break;
2467 		}
2468 
2469 		ASSERT3U(error, ==, 0);
2470 
2471 		VERIFY3U(dsl_prop_get(osname, prop, sizeof (value),
2472 		    1, &value, setpoint), ==, 0);
2473 
2474 		if (i == 0)
2475 			valname = zio_checksum_table[value].ci_name;
2476 		else
2477 			valname = zio_compress_table[value].ci_name;
2478 
2479 		if (zopt_verbose >= 6) {
2480 			(void) printf("%s %s = %s for '%s'\n",
2481 			    osname, prop, valname, setpoint);
2482 		}
2483 	}
2484 
2485 	(void) rw_unlock(&ztest_shared->zs_name_lock);
2486 }
2487 
2488 static void
2489 ztest_error_setup(vdev_t *vd, int mode, int mask, uint64_t arg)
2490 {
2491 	int c;
2492 
2493 	for (c = 0; c < vd->vdev_children; c++)
2494 		ztest_error_setup(vd->vdev_child[c], mode, mask, arg);
2495 
2496 	if (vd->vdev_path != NULL) {
2497 		vd->vdev_fault_mode = mode;
2498 		vd->vdev_fault_mask = mask;
2499 		vd->vdev_fault_arg = arg;
2500 	}
2501 }
2502 
2503 /*
2504  * Inject random faults into the on-disk data.
2505  */
2506 void
2507 ztest_fault_inject(ztest_args_t *za)
2508 {
2509 	int fd;
2510 	uint64_t offset;
2511 	uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
2512 	uint64_t bad = 0x1990c0ffeedecade;
2513 	uint64_t top, leaf;
2514 	char path0[MAXPATHLEN];
2515 	char pathrand[MAXPATHLEN];
2516 	size_t fsize;
2517 	spa_t *spa = dmu_objset_spa(za->za_os);
2518 	int bshift = SPA_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
2519 	int iters = 1000;
2520 	vdev_t *vd0;
2521 	uint64_t guid0 = 0;
2522 
2523 	/*
2524 	 * We can't inject faults when we have no fault tolerance.
2525 	 */
2526 	if (zopt_maxfaults == 0)
2527 		return;
2528 
2529 	ASSERT(leaves >= 2);
2530 
2531 	/*
2532 	 * Pick a random top-level vdev.
2533 	 */
2534 	spa_config_enter(spa, RW_READER, FTAG);
2535 	top = ztest_random(spa->spa_root_vdev->vdev_children);
2536 	spa_config_exit(spa, FTAG);
2537 
2538 	/*
2539 	 * Pick a random leaf.
2540 	 */
2541 	leaf = ztest_random(leaves);
2542 
2543 	/*
2544 	 * Generate paths to the first two leaves in this top-level vdev,
2545 	 * and to the random leaf we selected.  We'll induce transient
2546 	 * I/O errors and random online/offline activity on leaf 0,
2547 	 * and we'll write random garbage to the randomly chosen leaf.
2548 	 */
2549 	(void) snprintf(path0, sizeof (path0),
2550 	    ztest_dev_template, zopt_dir, zopt_pool, top * leaves + 0);
2551 	(void) snprintf(pathrand, sizeof (pathrand),
2552 	    ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf);
2553 
2554 	dprintf("damaging %s and %s\n", path0, pathrand);
2555 
2556 	spa_config_enter(spa, RW_READER, FTAG);
2557 
2558 	/*
2559 	 * If we can tolerate two or more faults, make vd0 fail randomly.
2560 	 */
2561 	vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
2562 	if (vd0 != NULL && zopt_maxfaults >= 2) {
2563 		guid0 = vd0->vdev_guid;
2564 		ztest_error_setup(vd0, VDEV_FAULT_COUNT,
2565 		    (1U << ZIO_TYPE_READ) | (1U << ZIO_TYPE_WRITE), 100);
2566 	}
2567 
2568 	spa_config_exit(spa, FTAG);
2569 
2570 	/*
2571 	 * If we can tolerate two or more faults, randomly online/offline vd0.
2572 	 */
2573 	if (zopt_maxfaults >= 2 && guid0 != 0) {
2574 		if (ztest_random(10) < 6)
2575 			(void) vdev_offline(spa, guid0, B_TRUE);
2576 		else
2577 			(void) vdev_online(spa, guid0);
2578 	}
2579 
2580 	/*
2581 	 * We have at least single-fault tolerance, so inject data corruption.
2582 	 */
2583 	fd = open(pathrand, O_RDWR);
2584 
2585 	if (fd == -1)	/* we hit a gap in the device namespace */
2586 		return;
2587 
2588 	fsize = lseek(fd, 0, SEEK_END);
2589 
2590 	while (--iters != 0) {
2591 		offset = ztest_random(fsize / (leaves << bshift)) *
2592 		    (leaves << bshift) + (leaf << bshift) +
2593 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
2594 
2595 		if (offset >= fsize)
2596 			continue;
2597 
2598 		if (zopt_verbose >= 6)
2599 			(void) printf("injecting bad word into %s,"
2600 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
2601 
2602 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
2603 			fatal(1, "can't inject bad word at 0x%llx in %s",
2604 			    offset, pathrand);
2605 	}
2606 
2607 	(void) close(fd);
2608 }
2609 
2610 /*
2611  * Scrub the pool.
2612  */
2613 void
2614 ztest_scrub(ztest_args_t *za)
2615 {
2616 	spa_t *spa = dmu_objset_spa(za->za_os);
2617 
2618 	(void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE);
2619 	(void) poll(NULL, 0, 1000); /* wait a second, then force a restart */
2620 	(void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE);
2621 }
2622 
2623 /*
2624  * Rename the pool to a different name and then rename it back.
2625  */
2626 void
2627 ztest_spa_rename(ztest_args_t *za)
2628 {
2629 	char *oldname, *newname;
2630 	int error;
2631 	spa_t *spa;
2632 
2633 	(void) rw_wrlock(&ztest_shared->zs_name_lock);
2634 
2635 	oldname = za->za_pool;
2636 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
2637 	(void) strcpy(newname, oldname);
2638 	(void) strcat(newname, "_tmp");
2639 
2640 	/*
2641 	 * Do the rename
2642 	 */
2643 	error = spa_rename(oldname, newname);
2644 	if (error)
2645 		fatal(0, "spa_rename('%s', '%s') = %d", oldname,
2646 		    newname, error);
2647 
2648 	/*
2649 	 * Try to open it under the old name, which shouldn't exist
2650 	 */
2651 	error = spa_open(oldname, &spa, FTAG);
2652 	if (error != ENOENT)
2653 		fatal(0, "spa_open('%s') = %d", oldname, error);
2654 
2655 	/*
2656 	 * Open it under the new name and make sure it's still the same spa_t.
2657 	 */
2658 	error = spa_open(newname, &spa, FTAG);
2659 	if (error != 0)
2660 		fatal(0, "spa_open('%s') = %d", newname, error);
2661 
2662 	ASSERT(spa == dmu_objset_spa(za->za_os));
2663 	spa_close(spa, FTAG);
2664 
2665 	/*
2666 	 * Rename it back to the original
2667 	 */
2668 	error = spa_rename(newname, oldname);
2669 	if (error)
2670 		fatal(0, "spa_rename('%s', '%s') = %d", newname,
2671 		    oldname, error);
2672 
2673 	/*
2674 	 * Make sure it can still be opened
2675 	 */
2676 	error = spa_open(oldname, &spa, FTAG);
2677 	if (error != 0)
2678 		fatal(0, "spa_open('%s') = %d", oldname, error);
2679 
2680 	ASSERT(spa == dmu_objset_spa(za->za_os));
2681 	spa_close(spa, FTAG);
2682 
2683 	umem_free(newname, strlen(newname) + 1);
2684 
2685 	(void) rw_unlock(&ztest_shared->zs_name_lock);
2686 }
2687 
2688 
2689 /*
2690  * Completely obliterate one disk.
2691  */
2692 static void
2693 ztest_obliterate_one_disk(uint64_t vdev)
2694 {
2695 	int fd;
2696 	char dev_name[MAXPATHLEN], copy_name[MAXPATHLEN];
2697 	size_t fsize;
2698 
2699 	if (zopt_maxfaults < 2)
2700 		return;
2701 
2702 	(void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
2703 	(void) snprintf(copy_name, MAXPATHLEN, "%s.old", dev_name);
2704 
2705 	fd = open(dev_name, O_RDWR);
2706 
2707 	if (fd == -1)
2708 		fatal(1, "can't open %s", dev_name);
2709 
2710 	/*
2711 	 * Determine the size.
2712 	 */
2713 	fsize = lseek(fd, 0, SEEK_END);
2714 
2715 	(void) close(fd);
2716 
2717 	/*
2718 	 * Rename the old device to dev_name.old (useful for debugging).
2719 	 */
2720 	VERIFY(rename(dev_name, copy_name) == 0);
2721 
2722 	/*
2723 	 * Create a new one.
2724 	 */
2725 	VERIFY((fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666)) >= 0);
2726 	VERIFY(ftruncate(fd, fsize) == 0);
2727 	(void) close(fd);
2728 }
2729 
2730 static void
2731 ztest_replace_one_disk(spa_t *spa, uint64_t vdev)
2732 {
2733 	char dev_name[MAXPATHLEN];
2734 	nvlist_t *file, *root;
2735 	int error;
2736 	uint64_t guid;
2737 	uint64_t ashift = ztest_get_ashift();
2738 	vdev_t *vd;
2739 
2740 	(void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev);
2741 
2742 	/*
2743 	 * Build the nvlist describing dev_name.
2744 	 */
2745 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
2746 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
2747 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0);
2748 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
2749 
2750 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
2751 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
2752 	VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN,
2753 	    &file, 1) == 0);
2754 
2755 	spa_config_enter(spa, RW_READER, FTAG);
2756 	if ((vd = vdev_lookup_by_path(spa->spa_root_vdev, dev_name)) == NULL)
2757 		guid = 0;
2758 	else
2759 		guid = vd->vdev_guid;
2760 	spa_config_exit(spa, FTAG);
2761 	error = spa_vdev_attach(spa, guid, root, B_TRUE);
2762 	if (error != 0 &&
2763 	    error != EBUSY &&
2764 	    error != ENOTSUP &&
2765 	    error != ENODEV &&
2766 	    error != EDOM)
2767 		fatal(0, "spa_vdev_attach(in-place) = %d", error);
2768 
2769 	nvlist_free(file);
2770 	nvlist_free(root);
2771 }
2772 
2773 static void
2774 ztest_verify_blocks(char *pool)
2775 {
2776 	int status;
2777 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
2778 	char zbuf[1024];
2779 	char *bin;
2780 	FILE *fp;
2781 
2782 	(void) realpath(getexecname(), zdb);
2783 
2784 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
2785 	bin = strstr(zdb, "/usr/bin/");
2786 	/* LINTED */
2787 	(void) sprintf(bin, "/usr/sbin/zdb -bc%s%s -U -O %s %s",
2788 	    zopt_verbose >= 3 ? "s" : "",
2789 	    zopt_verbose >= 4 ? "v" : "",
2790 	    ztest_random(2) == 0 ? "pre" : "post", pool);
2791 
2792 	if (zopt_verbose >= 5)
2793 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
2794 
2795 	fp = popen(zdb, "r");
2796 
2797 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
2798 		if (zopt_verbose >= 3)
2799 			(void) printf("%s", zbuf);
2800 
2801 	status = pclose(fp);
2802 
2803 	if (status == 0)
2804 		return;
2805 
2806 	ztest_dump_core = 0;
2807 	if (WIFEXITED(status))
2808 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
2809 	else
2810 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
2811 }
2812 
2813 static void
2814 ztest_walk_pool_directory(char *header)
2815 {
2816 	spa_t *spa = NULL;
2817 
2818 	if (zopt_verbose >= 6)
2819 		(void) printf("%s\n", header);
2820 
2821 	mutex_enter(&spa_namespace_lock);
2822 	while ((spa = spa_next(spa)) != NULL)
2823 		if (zopt_verbose >= 6)
2824 			(void) printf("\t%s\n", spa_name(spa));
2825 	mutex_exit(&spa_namespace_lock);
2826 }
2827 
2828 static void
2829 ztest_spa_import_export(char *oldname, char *newname)
2830 {
2831 	nvlist_t *config;
2832 	uint64_t pool_guid;
2833 	spa_t *spa;
2834 	int error;
2835 
2836 	if (zopt_verbose >= 4) {
2837 		(void) printf("import/export: old = %s, new = %s\n",
2838 		    oldname, newname);
2839 	}
2840 
2841 	/*
2842 	 * Clean up from previous runs.
2843 	 */
2844 	(void) spa_destroy(newname);
2845 
2846 	/*
2847 	 * Get the pool's configuration and guid.
2848 	 */
2849 	error = spa_open(oldname, &spa, FTAG);
2850 	if (error)
2851 		fatal(0, "spa_open('%s') = %d", oldname, error);
2852 
2853 	pool_guid = spa_guid(spa);
2854 	spa_close(spa, FTAG);
2855 
2856 	ztest_walk_pool_directory("pools before export");
2857 
2858 	/*
2859 	 * Export it.
2860 	 */
2861 	error = spa_export(oldname, &config);
2862 	if (error)
2863 		fatal(0, "spa_export('%s') = %d", oldname, error);
2864 
2865 	ztest_walk_pool_directory("pools after export");
2866 
2867 	/*
2868 	 * Import it under the new name.
2869 	 */
2870 	error = spa_import(newname, config, NULL);
2871 	if (error)
2872 		fatal(0, "spa_import('%s') = %d", newname, error);
2873 
2874 	ztest_walk_pool_directory("pools after import");
2875 
2876 	/*
2877 	 * Try to import it again -- should fail with EEXIST.
2878 	 */
2879 	error = spa_import(newname, config, NULL);
2880 	if (error != EEXIST)
2881 		fatal(0, "spa_import('%s') twice", newname);
2882 
2883 	/*
2884 	 * Try to import it under a different name -- should fail with EEXIST.
2885 	 */
2886 	error = spa_import(oldname, config, NULL);
2887 	if (error != EEXIST)
2888 		fatal(0, "spa_import('%s') under multiple names", newname);
2889 
2890 	/*
2891 	 * Verify that the pool is no longer visible under the old name.
2892 	 */
2893 	error = spa_open(oldname, &spa, FTAG);
2894 	if (error != ENOENT)
2895 		fatal(0, "spa_open('%s') = %d", newname, error);
2896 
2897 	/*
2898 	 * Verify that we can open and close the pool using the new name.
2899 	 */
2900 	error = spa_open(newname, &spa, FTAG);
2901 	if (error)
2902 		fatal(0, "spa_open('%s') = %d", newname, error);
2903 	ASSERT(pool_guid == spa_guid(spa));
2904 	spa_close(spa, FTAG);
2905 
2906 	nvlist_free(config);
2907 }
2908 
2909 static void *
2910 ztest_thread(void *arg)
2911 {
2912 	ztest_args_t *za = arg;
2913 	ztest_shared_t *zs = ztest_shared;
2914 	hrtime_t now, functime;
2915 	ztest_info_t *zi;
2916 	int f;
2917 
2918 	while ((now = gethrtime()) < za->za_stop) {
2919 		/*
2920 		 * See if it's time to force a crash.
2921 		 */
2922 		if (now > za->za_kill) {
2923 			zs->zs_alloc = spa_get_alloc(dmu_objset_spa(za->za_os));
2924 			zs->zs_space = spa_get_space(dmu_objset_spa(za->za_os));
2925 			(void) kill(getpid(), SIGKILL);
2926 		}
2927 
2928 		/*
2929 		 * Pick a random function.
2930 		 */
2931 		f = ztest_random(ZTEST_FUNCS);
2932 		zi = &zs->zs_info[f];
2933 
2934 		/*
2935 		 * Decide whether to call it, based on the requested frequency.
2936 		 */
2937 		if (zi->zi_call_target == 0 ||
2938 		    (double)zi->zi_call_total / zi->zi_call_target >
2939 		    (double)(now - zs->zs_start_time) / (zopt_time * NANOSEC))
2940 			continue;
2941 
2942 		atomic_add_64(&zi->zi_calls, 1);
2943 		atomic_add_64(&zi->zi_call_total, 1);
2944 
2945 		za->za_diroff = (za->za_instance * ZTEST_FUNCS + f) *
2946 		    ZTEST_DIRSIZE;
2947 		za->za_diroff_shared = (1ULL << 63);
2948 
2949 		ztest_dmu_write_parallel(za);
2950 
2951 		zi->zi_func(za);
2952 
2953 		functime = gethrtime() - now;
2954 
2955 		atomic_add_64(&zi->zi_call_time, functime);
2956 
2957 		if (zopt_verbose >= 4) {
2958 			Dl_info dli;
2959 			(void) dladdr((void *)zi->zi_func, &dli);
2960 			(void) printf("%6.2f sec in %s\n",
2961 			    (double)functime / NANOSEC, dli.dli_sname);
2962 		}
2963 
2964 		/*
2965 		 * If we're getting ENOSPC with some regularity, stop.
2966 		 */
2967 		if (zs->zs_enospc_count > 10)
2968 			break;
2969 	}
2970 
2971 	return (NULL);
2972 }
2973 
2974 /*
2975  * Kick off threads to run tests on all datasets in parallel.
2976  */
2977 static void
2978 ztest_run(char *pool)
2979 {
2980 	int t, d, error;
2981 	ztest_shared_t *zs = ztest_shared;
2982 	ztest_args_t *za;
2983 	spa_t *spa;
2984 	char name[100];
2985 
2986 	(void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL);
2987 	(void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL);
2988 
2989 	for (t = 0; t < ZTEST_SYNC_LOCKS; t++)
2990 		(void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL);
2991 
2992 	/*
2993 	 * Destroy one disk before we even start.
2994 	 * It's mirrored, so everything should work just fine.
2995 	 * This makes us exercise fault handling very early in spa_load().
2996 	 */
2997 	ztest_obliterate_one_disk(0);
2998 
2999 	/*
3000 	 * Verify that the sum of the sizes of all blocks in the pool
3001 	 * equals the SPA's allocated space total.
3002 	 */
3003 	ztest_verify_blocks(pool);
3004 
3005 	/*
3006 	 * Kick off a replacement of the disk we just obliterated.
3007 	 */
3008 	kernel_init(FREAD | FWRITE);
3009 	error = spa_open(pool, &spa, FTAG);
3010 	if (error)
3011 		fatal(0, "spa_open(%s) = %d", pool, error);
3012 	ztest_replace_one_disk(spa, 0);
3013 	if (zopt_verbose >= 5)
3014 		show_pool_stats(spa);
3015 	spa_close(spa, FTAG);
3016 	kernel_fini();
3017 
3018 	kernel_init(FREAD | FWRITE);
3019 
3020 	/*
3021 	 * Verify that we can export the pool and reimport it under a
3022 	 * different name.
3023 	 */
3024 	if (ztest_random(2) == 0) {
3025 		(void) snprintf(name, 100, "%s_import", pool);
3026 		ztest_spa_import_export(pool, name);
3027 		ztest_spa_import_export(name, pool);
3028 	}
3029 
3030 	/*
3031 	 * Verify that we can loop over all pools.
3032 	 */
3033 	mutex_enter(&spa_namespace_lock);
3034 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) {
3035 		if (zopt_verbose > 3) {
3036 			(void) printf("spa_next: found %s\n", spa_name(spa));
3037 		}
3038 	}
3039 	mutex_exit(&spa_namespace_lock);
3040 
3041 	/*
3042 	 * Open our pool.
3043 	 */
3044 	error = spa_open(pool, &spa, FTAG);
3045 	if (error)
3046 		fatal(0, "spa_open() = %d", error);
3047 
3048 	/*
3049 	 * Verify that we can safely inquire about about any object,
3050 	 * whether it's allocated or not.  To make it interesting,
3051 	 * we probe a 5-wide window around each power of two.
3052 	 * This hits all edge cases, including zero and the max.
3053 	 */
3054 	for (t = 0; t < 64; t++) {
3055 		for (d = -5; d <= 5; d++) {
3056 			error = dmu_object_info(spa->spa_meta_objset,
3057 			    (1ULL << t) + d, NULL);
3058 			ASSERT(error == 0 || error == ENOENT ||
3059 			    error == EINVAL);
3060 		}
3061 	}
3062 
3063 	/*
3064 	 * Now kick off all the tests that run in parallel.
3065 	 */
3066 	zs->zs_enospc_count = 0;
3067 
3068 	za = umem_zalloc(zopt_threads * sizeof (ztest_args_t), UMEM_NOFAIL);
3069 
3070 	if (zopt_verbose >= 4)
3071 		(void) printf("starting main threads...\n");
3072 
3073 	za[0].za_start = gethrtime();
3074 	za[0].za_stop = za[0].za_start + zopt_passtime * NANOSEC;
3075 	za[0].za_stop = MIN(za[0].za_stop, zs->zs_stop_time);
3076 	za[0].za_kill = za[0].za_stop;
3077 	if (ztest_random(100) < zopt_killrate)
3078 		za[0].za_kill -= ztest_random(zopt_passtime * NANOSEC);
3079 
3080 	for (t = 0; t < zopt_threads; t++) {
3081 		d = t % zopt_datasets;
3082 		if (t < zopt_datasets) {
3083 			ztest_replay_t zr;
3084 			(void) rw_rdlock(&ztest_shared->zs_name_lock);
3085 			(void) snprintf(name, 100, "%s/%s_%d", pool, pool, d);
3086 			error = dmu_objset_create(name, DMU_OST_OTHER, NULL,
3087 			    ztest_create_cb, NULL);
3088 			if (error != 0 && error != EEXIST) {
3089 				if (error == ENOSPC) {
3090 					zs->zs_enospc_count++;
3091 					(void) rw_unlock(
3092 					    &ztest_shared->zs_name_lock);
3093 					break;
3094 				}
3095 				fatal(0, "dmu_objset_create(%s) = %d",
3096 				    name, error);
3097 			}
3098 			error = dmu_objset_open(name, DMU_OST_OTHER,
3099 			    DS_MODE_STANDARD, &za[d].za_os);
3100 			if (error)
3101 				fatal(0, "dmu_objset_open('%s') = %d",
3102 				    name, error);
3103 			(void) rw_unlock(&ztest_shared->zs_name_lock);
3104 			zr.zr_os = za[d].za_os;
3105 			zil_replay(zr.zr_os, &zr, &zr.zr_assign,
3106 			    ztest_replay_vector, NULL);
3107 			za[d].za_zilog = zil_open(za[d].za_os, NULL);
3108 		}
3109 		za[t].za_pool = spa_strdup(pool);
3110 		za[t].za_os = za[d].za_os;
3111 		za[t].za_zilog = za[d].za_zilog;
3112 		za[t].za_instance = t;
3113 		za[t].za_random = ztest_random(-1ULL);
3114 		za[t].za_start = za[0].za_start;
3115 		za[t].za_stop = za[0].za_stop;
3116 		za[t].za_kill = za[0].za_kill;
3117 
3118 		error = thr_create(0, 0, ztest_thread, &za[t], THR_BOUND,
3119 		    &za[t].za_thread);
3120 		if (error)
3121 			fatal(0, "can't create thread %d: error %d",
3122 			    t, error);
3123 	}
3124 
3125 	while (--t >= 0) {
3126 		error = thr_join(za[t].za_thread, NULL, NULL);
3127 		if (error)
3128 			fatal(0, "thr_join(%d) = %d", t, error);
3129 		if (za[t].za_th)
3130 			traverse_fini(za[t].za_th);
3131 		if (t < zopt_datasets) {
3132 			zil_close(za[t].za_zilog);
3133 			dmu_objset_close(za[t].za_os);
3134 		}
3135 		spa_strfree(za[t].za_pool);
3136 	}
3137 
3138 	umem_free(za, zopt_threads * sizeof (ztest_args_t));
3139 
3140 	if (zopt_verbose >= 3)
3141 		show_pool_stats(spa);
3142 
3143 	txg_wait_synced(spa_get_dsl(spa), 0);
3144 
3145 	zs->zs_alloc = spa_get_alloc(spa);
3146 	zs->zs_space = spa_get_space(spa);
3147 
3148 	/*
3149 	 * Did we have out-of-space errors?  If so, destroy a random objset.
3150 	 */
3151 	if (zs->zs_enospc_count != 0) {
3152 		(void) rw_rdlock(&ztest_shared->zs_name_lock);
3153 		(void) snprintf(name, 100, "%s/%s_%d", pool, pool,
3154 		    (int)ztest_random(zopt_datasets));
3155 		if (zopt_verbose >= 3)
3156 			(void) printf("Destroying %s to free up space\n", name);
3157 		(void) dmu_objset_find(name, ztest_destroy_cb, NULL,
3158 		    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
3159 		(void) rw_unlock(&ztest_shared->zs_name_lock);
3160 	}
3161 
3162 	txg_wait_synced(spa_get_dsl(spa), 0);
3163 
3164 	/*
3165 	 * Right before closing the pool, kick off a bunch of async I/O;
3166 	 * spa_close() should wait for it to complete.
3167 	 */
3168 	for (t = 1; t < 50; t++)
3169 		dmu_prefetch(spa->spa_meta_objset, t, 0, 1 << 15);
3170 
3171 	spa_close(spa, FTAG);
3172 
3173 	kernel_fini();
3174 }
3175 
3176 void
3177 print_time(hrtime_t t, char *timebuf)
3178 {
3179 	hrtime_t s = t / NANOSEC;
3180 	hrtime_t m = s / 60;
3181 	hrtime_t h = m / 60;
3182 	hrtime_t d = h / 24;
3183 
3184 	s -= m * 60;
3185 	m -= h * 60;
3186 	h -= d * 24;
3187 
3188 	timebuf[0] = '\0';
3189 
3190 	if (d)
3191 		(void) sprintf(timebuf,
3192 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
3193 	else if (h)
3194 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
3195 	else if (m)
3196 		(void) sprintf(timebuf, "%llum%02llus", m, s);
3197 	else
3198 		(void) sprintf(timebuf, "%llus", s);
3199 }
3200 
3201 /*
3202  * Create a storage pool with the given name and initial vdev size.
3203  * Then create the specified number of datasets in the pool.
3204  */
3205 static void
3206 ztest_init(char *pool)
3207 {
3208 	spa_t *spa;
3209 	int error;
3210 	nvlist_t *nvroot;
3211 
3212 	kernel_init(FREAD | FWRITE);
3213 
3214 	/*
3215 	 * Create the storage pool.
3216 	 */
3217 	(void) spa_destroy(pool);
3218 	ztest_shared->zs_vdev_primaries = 0;
3219 	nvroot = make_vdev_root(zopt_vdev_size, zopt_raidz, zopt_mirrors, 1);
3220 	error = spa_create(pool, nvroot, NULL);
3221 	nvlist_free(nvroot);
3222 
3223 	if (error)
3224 		fatal(0, "spa_create() = %d", error);
3225 	error = spa_open(pool, &spa, FTAG);
3226 	if (error)
3227 		fatal(0, "spa_open() = %d", error);
3228 
3229 	if (zopt_verbose >= 3)
3230 		show_pool_stats(spa);
3231 
3232 	spa_close(spa, FTAG);
3233 
3234 	kernel_fini();
3235 }
3236 
3237 int
3238 main(int argc, char **argv)
3239 {
3240 	int kills = 0;
3241 	int iters = 0;
3242 	int i, f;
3243 	ztest_shared_t *zs;
3244 	ztest_info_t *zi;
3245 	char timebuf[100];
3246 	char numbuf[6];
3247 
3248 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
3249 
3250 	/* Override location of zpool.cache */
3251 	spa_config_dir = "/tmp";
3252 
3253 	ztest_random_fd = open("/dev/urandom", O_RDONLY);
3254 
3255 	process_options(argc, argv);
3256 
3257 	argc -= optind;
3258 	argv += optind;
3259 
3260 	dprintf_setup(&argc, argv);
3261 
3262 	/*
3263 	 * Blow away any existing copy of zpool.cache
3264 	 */
3265 	if (zopt_init != 0)
3266 		(void) remove("/tmp/zpool.cache");
3267 
3268 	zs = ztest_shared = (void *)mmap(0,
3269 	    P2ROUNDUP(sizeof (ztest_shared_t), getpagesize()),
3270 	    PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
3271 
3272 	if (zopt_verbose >= 1) {
3273 		(void) printf("%llu vdevs, %d datasets, %d threads,"
3274 		    " %llu seconds...\n",
3275 		    (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads,
3276 		    (u_longlong_t)zopt_time);
3277 	}
3278 
3279 	/*
3280 	 * Create and initialize our storage pool.
3281 	 */
3282 	for (i = 1; i <= zopt_init; i++) {
3283 		bzero(zs, sizeof (ztest_shared_t));
3284 		if (zopt_verbose >= 3 && zopt_init != 1)
3285 			(void) printf("ztest_init(), pass %d\n", i);
3286 		ztest_init(zopt_pool);
3287 	}
3288 
3289 	/*
3290 	 * Initialize the call targets for each function.
3291 	 */
3292 	for (f = 0; f < ZTEST_FUNCS; f++) {
3293 		zi = &zs->zs_info[f];
3294 
3295 		*zi = ztest_info[f];
3296 
3297 		if (*zi->zi_interval == 0)
3298 			zi->zi_call_target = UINT64_MAX;
3299 		else
3300 			zi->zi_call_target = zopt_time / *zi->zi_interval;
3301 	}
3302 
3303 	zs->zs_start_time = gethrtime();
3304 	zs->zs_stop_time = zs->zs_start_time + zopt_time * NANOSEC;
3305 
3306 	/*
3307 	 * Run the tests in a loop.  These tests include fault injection
3308 	 * to verify that self-healing data works, and forced crashes
3309 	 * to verify that we never lose on-disk consistency.
3310 	 */
3311 	while (gethrtime() < zs->zs_stop_time) {
3312 		int status;
3313 		pid_t pid;
3314 		char *tmp;
3315 
3316 		/*
3317 		 * Initialize the workload counters for each function.
3318 		 */
3319 		for (f = 0; f < ZTEST_FUNCS; f++) {
3320 			zi = &zs->zs_info[f];
3321 			zi->zi_calls = 0;
3322 			zi->zi_call_time = 0;
3323 		}
3324 
3325 		pid = fork();
3326 
3327 		if (pid == -1)
3328 			fatal(1, "fork failed");
3329 
3330 		if (pid == 0) {	/* child */
3331 			struct rlimit rl = { 1024, 1024 };
3332 			(void) setrlimit(RLIMIT_NOFILE, &rl);
3333 			(void) enable_extended_FILE_stdio(-1, -1);
3334 			ztest_run(zopt_pool);
3335 			exit(0);
3336 		}
3337 
3338 		while (waitpid(pid, &status, WEXITED) != pid)
3339 			continue;
3340 
3341 		if (WIFEXITED(status)) {
3342 			if (WEXITSTATUS(status) != 0) {
3343 				(void) fprintf(stderr,
3344 				    "child exited with code %d\n",
3345 				    WEXITSTATUS(status));
3346 				exit(2);
3347 			}
3348 		} else {
3349 			if (WTERMSIG(status) != SIGKILL) {
3350 				(void) fprintf(stderr,
3351 				    "child died with signal %d\n",
3352 				    WTERMSIG(status));
3353 				exit(3);
3354 			}
3355 			kills++;
3356 		}
3357 
3358 		iters++;
3359 
3360 		if (zopt_verbose >= 1) {
3361 			hrtime_t now = gethrtime();
3362 
3363 			now = MIN(now, zs->zs_stop_time);
3364 			print_time(zs->zs_stop_time - now, timebuf);
3365 			nicenum(zs->zs_space, numbuf);
3366 
3367 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
3368 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
3369 			    iters,
3370 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
3371 			    (u_longlong_t)zs->zs_enospc_count,
3372 			    100.0 * zs->zs_alloc / zs->zs_space,
3373 			    numbuf,
3374 			    100.0 * (now - zs->zs_start_time) /
3375 			    (zopt_time * NANOSEC), timebuf);
3376 		}
3377 
3378 		if (zopt_verbose >= 2) {
3379 			(void) printf("\nWorkload summary:\n\n");
3380 			(void) printf("%7s %9s   %s\n",
3381 			    "Calls", "Time", "Function");
3382 			(void) printf("%7s %9s   %s\n",
3383 			    "-----", "----", "--------");
3384 			for (f = 0; f < ZTEST_FUNCS; f++) {
3385 				Dl_info dli;
3386 
3387 				zi = &zs->zs_info[f];
3388 				print_time(zi->zi_call_time, timebuf);
3389 				(void) dladdr((void *)zi->zi_func, &dli);
3390 				(void) printf("%7llu %9s   %s\n",
3391 				    (u_longlong_t)zi->zi_calls, timebuf,
3392 				    dli.dli_sname);
3393 			}
3394 			(void) printf("\n");
3395 		}
3396 
3397 		/*
3398 		 * It's possible that we killed a child during a rename test, in
3399 		 * which case we'll have a 'ztest_tmp' pool lying around instead
3400 		 * of 'ztest'.  Do a blind rename in case this happened.
3401 		 */
3402 		tmp = umem_alloc(strlen(zopt_pool) + 5, UMEM_NOFAIL);
3403 		(void) strcpy(tmp, zopt_pool);
3404 		(void) strcat(tmp, "_tmp");
3405 		kernel_init(FREAD | FWRITE);
3406 		(void) spa_rename(tmp, zopt_pool);
3407 		kernel_fini();
3408 		umem_free(tmp, strlen(tmp) + 1);
3409 	}
3410 
3411 	ztest_verify_blocks(zopt_pool);
3412 
3413 	if (zopt_verbose >= 1) {
3414 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
3415 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
3416 	}
3417 
3418 	return (0);
3419 }
3420