xref: /illumos-gate/usr/src/cmd/ztest/ztest.c (revision 46b592853d0f4f11781b6b0a7533f267c6aee132)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  * The objective of this program is to provide a DMU/ZAP/SPA stress test
28  * that runs entirely in userland, is easy to use, and easy to extend.
29  *
30  * The overall design of the ztest program is as follows:
31  *
32  * (1) For each major functional area (e.g. adding vdevs to a pool,
33  *     creating and destroying datasets, reading and writing objects, etc)
34  *     we have a simple routine to test that functionality.  These
35  *     individual routines do not have to do anything "stressful".
36  *
37  * (2) We turn these simple functionality tests into a stress test by
38  *     running them all in parallel, with as many threads as desired,
39  *     and spread across as many datasets, objects, and vdevs as desired.
40  *
41  * (3) While all this is happening, we inject faults into the pool to
42  *     verify that self-healing data really works.
43  *
44  * (4) Every time we open a dataset, we change its checksum and compression
45  *     functions.  Thus even individual objects vary from block to block
46  *     in which checksum they use and whether they're compressed.
47  *
48  * (5) To verify that we never lose on-disk consistency after a crash,
49  *     we run the entire test in a child of the main process.
50  *     At random times, the child self-immolates with a SIGKILL.
51  *     This is the software equivalent of pulling the power cord.
52  *     The parent then runs the test again, using the existing
53  *     storage pool, as many times as desired.
54  *
55  * (6) To verify that we don't have future leaks or temporal incursions,
56  *     many of the functional tests record the transaction group number
57  *     as part of their data.  When reading old data, they verify that
58  *     the transaction group number is less than the current, open txg.
59  *     If you add a new test, please do this if applicable.
60  *
61  * When run with no arguments, ztest runs for about five minutes and
62  * produces no output if successful.  To get a little bit of information,
63  * specify -V.  To get more information, specify -VV, and so on.
64  *
65  * To turn this into an overnight stress test, use -T to specify run time.
66  *
67  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
68  * to increase the pool capacity, fanout, and overall stress level.
69  *
70  * The -N(okill) option will suppress kills, so each child runs to completion.
71  * This can be useful when you're trying to distinguish temporal incursions
72  * from plain old race conditions.
73  */
74 
75 #include <sys/zfs_context.h>
76 #include <sys/spa.h>
77 #include <sys/dmu.h>
78 #include <sys/txg.h>
79 #include <sys/dbuf.h>
80 #include <sys/zap.h>
81 #include <sys/dmu_objset.h>
82 #include <sys/poll.h>
83 #include <sys/stat.h>
84 #include <sys/time.h>
85 #include <sys/wait.h>
86 #include <sys/mman.h>
87 #include <sys/resource.h>
88 #include <sys/zio.h>
89 #include <sys/zil.h>
90 #include <sys/zil_impl.h>
91 #include <sys/vdev_impl.h>
92 #include <sys/vdev_file.h>
93 #include <sys/spa_impl.h>
94 #include <sys/metaslab_impl.h>
95 #include <sys/dsl_prop.h>
96 #include <sys/dsl_dataset.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 #include <libnvpair.h>
109 
110 static char cmdname[] = "ztest";
111 static char *zopt_pool = cmdname;
112 
113 static uint64_t zopt_vdevs = 5;
114 static uint64_t zopt_vdevtime;
115 static int zopt_ashift = SPA_MINBLOCKSHIFT;
116 static int zopt_mirrors = 2;
117 static int zopt_raidz = 4;
118 static int zopt_raidz_parity = 1;
119 static size_t zopt_vdev_size = SPA_MINDEVSIZE;
120 static int zopt_datasets = 7;
121 static int zopt_threads = 23;
122 static uint64_t zopt_passtime = 60;	/* 60 seconds */
123 static uint64_t zopt_killrate = 70;	/* 70% kill rate */
124 static int zopt_verbose = 0;
125 static int zopt_init = 1;
126 static char *zopt_dir = "/tmp";
127 static uint64_t zopt_time = 300;	/* 5 minutes */
128 static int zopt_maxfaults;
129 
130 #define	BT_MAGIC	0x123456789abcdefULL
131 
132 enum ztest_io_type {
133 	ZTEST_IO_WRITE_TAG,
134 	ZTEST_IO_WRITE_PATTERN,
135 	ZTEST_IO_WRITE_ZEROES,
136 	ZTEST_IO_TRUNCATE,
137 	ZTEST_IO_SETATTR,
138 	ZTEST_IO_TYPES
139 };
140 
141 typedef struct ztest_block_tag {
142 	uint64_t	bt_magic;
143 	uint64_t	bt_objset;
144 	uint64_t	bt_object;
145 	uint64_t	bt_offset;
146 	uint64_t	bt_gen;
147 	uint64_t	bt_txg;
148 	uint64_t	bt_crtxg;
149 } ztest_block_tag_t;
150 
151 typedef struct bufwad {
152 	uint64_t	bw_index;
153 	uint64_t	bw_txg;
154 	uint64_t	bw_data;
155 } bufwad_t;
156 
157 /*
158  * XXX -- fix zfs range locks to be generic so we can use them here.
159  */
160 typedef enum {
161 	RL_READER,
162 	RL_WRITER,
163 	RL_APPEND
164 } rl_type_t;
165 
166 typedef struct rll {
167 	void		*rll_writer;
168 	int		rll_readers;
169 	mutex_t		rll_lock;
170 	cond_t		rll_cv;
171 } rll_t;
172 
173 typedef struct rl {
174 	uint64_t	rl_object;
175 	uint64_t	rl_offset;
176 	uint64_t	rl_size;
177 	rll_t		*rl_lock;
178 } rl_t;
179 
180 #define	ZTEST_RANGE_LOCKS	64
181 #define	ZTEST_OBJECT_LOCKS	64
182 
183 /*
184  * Object descriptor.  Used as a template for object lookup/create/remove.
185  */
186 typedef struct ztest_od {
187 	uint64_t	od_dir;
188 	uint64_t	od_object;
189 	dmu_object_type_t od_type;
190 	dmu_object_type_t od_crtype;
191 	uint64_t	od_blocksize;
192 	uint64_t	od_crblocksize;
193 	uint64_t	od_gen;
194 	uint64_t	od_crgen;
195 	char		od_name[MAXNAMELEN];
196 } ztest_od_t;
197 
198 /*
199  * Per-dataset state.
200  */
201 typedef struct ztest_ds {
202 	objset_t	*zd_os;
203 	zilog_t		*zd_zilog;
204 	uint64_t	zd_seq;
205 	ztest_od_t	*zd_od;		/* debugging aid */
206 	char		zd_name[MAXNAMELEN];
207 	mutex_t		zd_dirobj_lock;
208 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
209 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
210 } ztest_ds_t;
211 
212 /*
213  * Per-iteration state.
214  */
215 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
216 
217 typedef struct ztest_info {
218 	ztest_func_t	*zi_func;	/* test function */
219 	uint64_t	zi_iters;	/* iterations per execution */
220 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
221 	uint64_t	zi_call_count;	/* per-pass count */
222 	uint64_t	zi_call_time;	/* per-pass time */
223 	uint64_t	zi_call_next;	/* next time to call this function */
224 } ztest_info_t;
225 
226 /*
227  * Note: these aren't static because we want dladdr() to work.
228  */
229 ztest_func_t ztest_dmu_read_write;
230 ztest_func_t ztest_dmu_write_parallel;
231 ztest_func_t ztest_dmu_object_alloc_free;
232 ztest_func_t ztest_dmu_commit_callbacks;
233 ztest_func_t ztest_zap;
234 ztest_func_t ztest_zap_parallel;
235 ztest_func_t ztest_zil_commit;
236 ztest_func_t ztest_dmu_read_write_zcopy;
237 ztest_func_t ztest_dmu_objset_create_destroy;
238 ztest_func_t ztest_dmu_prealloc;
239 ztest_func_t ztest_fzap;
240 ztest_func_t ztest_dmu_snapshot_create_destroy;
241 ztest_func_t ztest_dsl_prop_get_set;
242 ztest_func_t ztest_spa_prop_get_set;
243 ztest_func_t ztest_spa_create_destroy;
244 ztest_func_t ztest_fault_inject;
245 ztest_func_t ztest_ddt_repair;
246 ztest_func_t ztest_dmu_snapshot_hold;
247 ztest_func_t ztest_spa_rename;
248 ztest_func_t ztest_scrub;
249 ztest_func_t ztest_dsl_dataset_promote_busy;
250 ztest_func_t ztest_vdev_attach_detach;
251 ztest_func_t ztest_vdev_LUN_growth;
252 ztest_func_t ztest_vdev_add_remove;
253 ztest_func_t ztest_vdev_aux_add_remove;
254 
255 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
256 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
257 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
258 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
259 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
260 
261 ztest_info_t ztest_info[] = {
262 	{ ztest_dmu_read_write,			1,	&zopt_always	},
263 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
264 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
265 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
266 	{ ztest_zap,				30,	&zopt_always	},
267 	{ ztest_zap_parallel,			100,	&zopt_always	},
268 	{ ztest_zil_commit,			1,	&zopt_incessant	},
269 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
270 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
271 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
272 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
273 #if 0
274 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
275 #endif
276 	{ ztest_fzap,				1,	&zopt_sometimes	},
277 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
278 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
279 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
280 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
281 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
282 	{ ztest_spa_rename,			1,	&zopt_rarely	},
283 	{ ztest_scrub,				1,	&zopt_rarely	},
284 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
285 	{ ztest_vdev_attach_detach,		1,	&zopt_rarely	},
286 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
287 	{ ztest_vdev_add_remove,		1,	&zopt_vdevtime	},
288 	{ ztest_vdev_aux_add_remove,		1,	&zopt_vdevtime	},
289 };
290 
291 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
292 
293 /*
294  * The following struct is used to hold a list of uncalled commit callbacks.
295  * The callbacks are ordered by txg number.
296  */
297 typedef struct ztest_cb_list {
298 	mutex_t	zcl_callbacks_lock;
299 	list_t	zcl_callbacks;
300 } ztest_cb_list_t;
301 
302 /*
303  * Stuff we need to share writably between parent and child.
304  */
305 typedef struct ztest_shared {
306 	char		*zs_pool;
307 	spa_t		*zs_spa;
308 	hrtime_t	zs_proc_start;
309 	hrtime_t	zs_proc_stop;
310 	hrtime_t	zs_thread_start;
311 	hrtime_t	zs_thread_stop;
312 	hrtime_t	zs_thread_kill;
313 	uint64_t	zs_enospc_count;
314 	uint64_t	zs_vdev_next_leaf;
315 	uint64_t	zs_vdev_aux;
316 	uint64_t	zs_alloc;
317 	uint64_t	zs_space;
318 	mutex_t		zs_vdev_lock;
319 	rwlock_t	zs_name_lock;
320 	ztest_info_t	zs_info[ZTEST_FUNCS];
321 	ztest_ds_t	zs_zd[];
322 } ztest_shared_t;
323 
324 #define	ID_PARALLEL	-1ULL
325 
326 static char ztest_dev_template[] = "%s/%s.%llua";
327 static char ztest_aux_template[] = "%s/%s.%s.%llu";
328 ztest_shared_t *ztest_shared;
329 uint64_t *ztest_seq;
330 
331 static int ztest_random_fd;
332 static int ztest_dump_core = 1;
333 
334 static boolean_t ztest_exiting;
335 
336 /* Global commit callback list */
337 static ztest_cb_list_t zcl;
338 
339 extern uint64_t metaslab_gang_bang;
340 extern uint64_t metaslab_df_alloc_threshold;
341 static uint64_t metaslab_sz;
342 
343 enum ztest_object {
344 	ZTEST_META_DNODE = 0,
345 	ZTEST_DIROBJ,
346 	ZTEST_OBJECTS
347 };
348 
349 static void usage(boolean_t) __NORETURN;
350 
351 /*
352  * These libumem hooks provide a reasonable set of defaults for the allocator's
353  * debugging facilities.
354  */
355 const char *
356 _umem_debug_init()
357 {
358 	return ("default,verbose"); /* $UMEM_DEBUG setting */
359 }
360 
361 const char *
362 _umem_logging_init(void)
363 {
364 	return ("fail,contents"); /* $UMEM_LOGGING setting */
365 }
366 
367 #define	FATAL_MSG_SZ	1024
368 
369 char *fatal_msg;
370 
371 static void
372 fatal(int do_perror, char *message, ...)
373 {
374 	va_list args;
375 	int save_errno = errno;
376 	char buf[FATAL_MSG_SZ];
377 
378 	(void) fflush(stdout);
379 
380 	va_start(args, message);
381 	(void) sprintf(buf, "ztest: ");
382 	/* LINTED */
383 	(void) vsprintf(buf + strlen(buf), message, args);
384 	va_end(args);
385 	if (do_perror) {
386 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
387 		    ": %s", strerror(save_errno));
388 	}
389 	(void) fprintf(stderr, "%s\n", buf);
390 	fatal_msg = buf;			/* to ease debugging */
391 	if (ztest_dump_core)
392 		abort();
393 	exit(3);
394 }
395 
396 static int
397 str2shift(const char *buf)
398 {
399 	const char *ends = "BKMGTPEZ";
400 	int i;
401 
402 	if (buf[0] == '\0')
403 		return (0);
404 	for (i = 0; i < strlen(ends); i++) {
405 		if (toupper(buf[0]) == ends[i])
406 			break;
407 	}
408 	if (i == strlen(ends)) {
409 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
410 		    buf);
411 		usage(B_FALSE);
412 	}
413 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
414 		return (10*i);
415 	}
416 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
417 	usage(B_FALSE);
418 	/* NOTREACHED */
419 }
420 
421 static uint64_t
422 nicenumtoull(const char *buf)
423 {
424 	char *end;
425 	uint64_t val;
426 
427 	val = strtoull(buf, &end, 0);
428 	if (end == buf) {
429 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
430 		usage(B_FALSE);
431 	} else if (end[0] == '.') {
432 		double fval = strtod(buf, &end);
433 		fval *= pow(2, str2shift(end));
434 		if (fval > UINT64_MAX) {
435 			(void) fprintf(stderr, "ztest: value too large: %s\n",
436 			    buf);
437 			usage(B_FALSE);
438 		}
439 		val = (uint64_t)fval;
440 	} else {
441 		int shift = str2shift(end);
442 		if (shift >= 64 || (val << shift) >> shift != val) {
443 			(void) fprintf(stderr, "ztest: value too large: %s\n",
444 			    buf);
445 			usage(B_FALSE);
446 		}
447 		val <<= shift;
448 	}
449 	return (val);
450 }
451 
452 static void
453 usage(boolean_t requested)
454 {
455 	char nice_vdev_size[10];
456 	char nice_gang_bang[10];
457 	FILE *fp = requested ? stdout : stderr;
458 
459 	nicenum(zopt_vdev_size, nice_vdev_size);
460 	nicenum(metaslab_gang_bang, nice_gang_bang);
461 
462 	(void) fprintf(fp, "Usage: %s\n"
463 	    "\t[-v vdevs (default: %llu)]\n"
464 	    "\t[-s size_of_each_vdev (default: %s)]\n"
465 	    "\t[-a alignment_shift (default: %d) (use 0 for random)]\n"
466 	    "\t[-m mirror_copies (default: %d)]\n"
467 	    "\t[-r raidz_disks (default: %d)]\n"
468 	    "\t[-R raidz_parity (default: %d)]\n"
469 	    "\t[-d datasets (default: %d)]\n"
470 	    "\t[-t threads (default: %d)]\n"
471 	    "\t[-g gang_block_threshold (default: %s)]\n"
472 	    "\t[-i initialize pool i times (default: %d)]\n"
473 	    "\t[-k kill percentage (default: %llu%%)]\n"
474 	    "\t[-p pool_name (default: %s)]\n"
475 	    "\t[-f file directory for vdev files (default: %s)]\n"
476 	    "\t[-V(erbose)] (use multiple times for ever more blather)\n"
477 	    "\t[-E(xisting)] (use existing pool instead of creating new one)\n"
478 	    "\t[-T time] total run time (default: %llu sec)\n"
479 	    "\t[-P passtime] time per pass (default: %llu sec)\n"
480 	    "\t[-h] (print help)\n"
481 	    "",
482 	    cmdname,
483 	    (u_longlong_t)zopt_vdevs,			/* -v */
484 	    nice_vdev_size,				/* -s */
485 	    zopt_ashift,				/* -a */
486 	    zopt_mirrors,				/* -m */
487 	    zopt_raidz,					/* -r */
488 	    zopt_raidz_parity,				/* -R */
489 	    zopt_datasets,				/* -d */
490 	    zopt_threads,				/* -t */
491 	    nice_gang_bang,				/* -g */
492 	    zopt_init,					/* -i */
493 	    (u_longlong_t)zopt_killrate,		/* -k */
494 	    zopt_pool,					/* -p */
495 	    zopt_dir,					/* -f */
496 	    (u_longlong_t)zopt_time,			/* -T */
497 	    (u_longlong_t)zopt_passtime);		/* -P */
498 	exit(requested ? 0 : 1);
499 }
500 
501 static void
502 process_options(int argc, char **argv)
503 {
504 	int opt;
505 	uint64_t value;
506 
507 	/* By default, test gang blocks for blocks 32K and greater */
508 	metaslab_gang_bang = 32 << 10;
509 
510 	while ((opt = getopt(argc, argv,
511 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:h")) != EOF) {
512 		value = 0;
513 		switch (opt) {
514 		case 'v':
515 		case 's':
516 		case 'a':
517 		case 'm':
518 		case 'r':
519 		case 'R':
520 		case 'd':
521 		case 't':
522 		case 'g':
523 		case 'i':
524 		case 'k':
525 		case 'T':
526 		case 'P':
527 			value = nicenumtoull(optarg);
528 		}
529 		switch (opt) {
530 		case 'v':
531 			zopt_vdevs = value;
532 			break;
533 		case 's':
534 			zopt_vdev_size = MAX(SPA_MINDEVSIZE, value);
535 			break;
536 		case 'a':
537 			zopt_ashift = value;
538 			break;
539 		case 'm':
540 			zopt_mirrors = value;
541 			break;
542 		case 'r':
543 			zopt_raidz = MAX(1, value);
544 			break;
545 		case 'R':
546 			zopt_raidz_parity = MIN(MAX(value, 1), 3);
547 			break;
548 		case 'd':
549 			zopt_datasets = MAX(1, value);
550 			break;
551 		case 't':
552 			zopt_threads = MAX(1, value);
553 			break;
554 		case 'g':
555 			metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value);
556 			break;
557 		case 'i':
558 			zopt_init = value;
559 			break;
560 		case 'k':
561 			zopt_killrate = value;
562 			break;
563 		case 'p':
564 			zopt_pool = strdup(optarg);
565 			break;
566 		case 'f':
567 			zopt_dir = strdup(optarg);
568 			break;
569 		case 'V':
570 			zopt_verbose++;
571 			break;
572 		case 'E':
573 			zopt_init = 0;
574 			break;
575 		case 'T':
576 			zopt_time = value;
577 			break;
578 		case 'P':
579 			zopt_passtime = MAX(1, value);
580 			break;
581 		case 'h':
582 			usage(B_TRUE);
583 			break;
584 		case '?':
585 		default:
586 			usage(B_FALSE);
587 			break;
588 		}
589 	}
590 
591 	zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1);
592 
593 	zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time * NANOSEC / zopt_vdevs :
594 	    UINT64_MAX >> 2);
595 	zopt_maxfaults = MAX(zopt_mirrors, 1) * (zopt_raidz_parity + 1) - 1;
596 }
597 
598 static void
599 ztest_kill(ztest_shared_t *zs)
600 {
601 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(zs->zs_spa));
602 	zs->zs_space = metaslab_class_get_space(spa_normal_class(zs->zs_spa));
603 	(void) kill(getpid(), SIGKILL);
604 }
605 
606 static uint64_t
607 ztest_random(uint64_t range)
608 {
609 	uint64_t r;
610 
611 	if (range == 0)
612 		return (0);
613 
614 	if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r))
615 		fatal(1, "short read from /dev/urandom");
616 
617 	return (r % range);
618 }
619 
620 /* ARGSUSED */
621 static void
622 ztest_record_enospc(const char *s)
623 {
624 	ztest_shared->zs_enospc_count++;
625 }
626 
627 static uint64_t
628 ztest_get_ashift(void)
629 {
630 	if (zopt_ashift == 0)
631 		return (SPA_MINBLOCKSHIFT + ztest_random(3));
632 	return (zopt_ashift);
633 }
634 
635 static nvlist_t *
636 make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift)
637 {
638 	char pathbuf[MAXPATHLEN];
639 	uint64_t vdev;
640 	nvlist_t *file;
641 
642 	if (ashift == 0)
643 		ashift = ztest_get_ashift();
644 
645 	if (path == NULL) {
646 		path = pathbuf;
647 
648 		if (aux != NULL) {
649 			vdev = ztest_shared->zs_vdev_aux;
650 			(void) sprintf(path, ztest_aux_template,
651 			    zopt_dir, zopt_pool, aux, vdev);
652 		} else {
653 			vdev = ztest_shared->zs_vdev_next_leaf++;
654 			(void) sprintf(path, ztest_dev_template,
655 			    zopt_dir, zopt_pool, vdev);
656 		}
657 	}
658 
659 	if (size != 0) {
660 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
661 		if (fd == -1)
662 			fatal(1, "can't open %s", path);
663 		if (ftruncate(fd, size) != 0)
664 			fatal(1, "can't ftruncate %s", path);
665 		(void) close(fd);
666 	}
667 
668 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
669 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
670 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
671 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
672 
673 	return (file);
674 }
675 
676 static nvlist_t *
677 make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r)
678 {
679 	nvlist_t *raidz, **child;
680 	int c;
681 
682 	if (r < 2)
683 		return (make_vdev_file(path, aux, size, ashift));
684 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
685 
686 	for (c = 0; c < r; c++)
687 		child[c] = make_vdev_file(path, aux, size, ashift);
688 
689 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
690 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
691 	    VDEV_TYPE_RAIDZ) == 0);
692 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
693 	    zopt_raidz_parity) == 0);
694 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
695 	    child, r) == 0);
696 
697 	for (c = 0; c < r; c++)
698 		nvlist_free(child[c]);
699 
700 	umem_free(child, r * sizeof (nvlist_t *));
701 
702 	return (raidz);
703 }
704 
705 static nvlist_t *
706 make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift,
707 	int r, int m)
708 {
709 	nvlist_t *mirror, **child;
710 	int c;
711 
712 	if (m < 1)
713 		return (make_vdev_raidz(path, aux, size, ashift, r));
714 
715 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
716 
717 	for (c = 0; c < m; c++)
718 		child[c] = make_vdev_raidz(path, aux, size, ashift, r);
719 
720 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
721 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
722 	    VDEV_TYPE_MIRROR) == 0);
723 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
724 	    child, m) == 0);
725 
726 	for (c = 0; c < m; c++)
727 		nvlist_free(child[c]);
728 
729 	umem_free(child, m * sizeof (nvlist_t *));
730 
731 	return (mirror);
732 }
733 
734 static nvlist_t *
735 make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift,
736 	int log, int r, int m, int t)
737 {
738 	nvlist_t *root, **child;
739 	int c;
740 
741 	ASSERT(t > 0);
742 
743 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
744 
745 	for (c = 0; c < t; c++) {
746 		child[c] = make_vdev_mirror(path, aux, size, ashift, r, m);
747 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
748 		    log) == 0);
749 	}
750 
751 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
752 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
753 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
754 	    child, t) == 0);
755 
756 	for (c = 0; c < t; c++)
757 		nvlist_free(child[c]);
758 
759 	umem_free(child, t * sizeof (nvlist_t *));
760 
761 	return (root);
762 }
763 
764 static int
765 ztest_random_blocksize(void)
766 {
767 	return (1 << (SPA_MINBLOCKSHIFT +
768 	    ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
769 }
770 
771 static int
772 ztest_random_ibshift(void)
773 {
774 	return (DN_MIN_INDBLKSHIFT +
775 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
776 }
777 
778 static uint64_t
779 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
780 {
781 	uint64_t top;
782 	vdev_t *rvd = spa->spa_root_vdev;
783 	vdev_t *tvd;
784 
785 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
786 
787 	do {
788 		top = ztest_random(rvd->vdev_children);
789 		tvd = rvd->vdev_child[top];
790 	} while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
791 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
792 
793 	return (top);
794 }
795 
796 static uint64_t
797 ztest_random_dsl_prop(zfs_prop_t prop)
798 {
799 	uint64_t value;
800 
801 	do {
802 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
803 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
804 
805 	return (value);
806 }
807 
808 static int
809 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
810     boolean_t inherit)
811 {
812 	const char *propname = zfs_prop_to_name(prop);
813 	const char *valname;
814 	char setpoint[MAXPATHLEN];
815 	uint64_t curval;
816 	int error;
817 
818 	error = dsl_prop_set(osname, propname, sizeof (value),
819 	    inherit ? 0 : 1, &value);
820 
821 	if (error == ENOSPC) {
822 		ztest_record_enospc(FTAG);
823 		return (error);
824 	}
825 	ASSERT3U(error, ==, 0);
826 
827 	VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval),
828 	    1, &curval, setpoint), ==, 0);
829 
830 	if (zopt_verbose >= 6) {
831 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
832 		(void) printf("%s %s = %s at '%s'\n",
833 		    osname, propname, valname, setpoint);
834 	}
835 
836 	return (error);
837 }
838 
839 #if 0
840 static int
841 ztest_spa_prop_set_uint64(ztest_shared_t *zs, zpool_prop_t prop, uint64_t value)
842 {
843 	spa_t *spa = zs->zs_spa;
844 	nvlist_t *props = NULL;
845 	int error;
846 
847 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
848 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
849 
850 	error = spa_prop_set(spa, props);
851 
852 	nvlist_free(props);
853 
854 	if (error == ENOSPC) {
855 		ztest_record_enospc(FTAG);
856 		return (error);
857 	}
858 	ASSERT3U(error, ==, 0);
859 
860 	return (error);
861 }
862 #endif
863 
864 static void
865 ztest_rll_init(rll_t *rll)
866 {
867 	rll->rll_writer = NULL;
868 	rll->rll_readers = 0;
869 	VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
870 	VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
871 }
872 
873 static void
874 ztest_rll_destroy(rll_t *rll)
875 {
876 	ASSERT(rll->rll_writer == NULL);
877 	ASSERT(rll->rll_readers == 0);
878 	VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
879 	VERIFY(cond_destroy(&rll->rll_cv) == 0);
880 }
881 
882 static void
883 ztest_rll_lock(rll_t *rll, rl_type_t type)
884 {
885 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
886 
887 	if (type == RL_READER) {
888 		while (rll->rll_writer != NULL)
889 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
890 		rll->rll_readers++;
891 	} else {
892 		while (rll->rll_writer != NULL || rll->rll_readers)
893 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
894 		rll->rll_writer = curthread;
895 	}
896 
897 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
898 }
899 
900 static void
901 ztest_rll_unlock(rll_t *rll)
902 {
903 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
904 
905 	if (rll->rll_writer) {
906 		ASSERT(rll->rll_readers == 0);
907 		rll->rll_writer = NULL;
908 	} else {
909 		ASSERT(rll->rll_readers != 0);
910 		ASSERT(rll->rll_writer == NULL);
911 		rll->rll_readers--;
912 	}
913 
914 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
915 		VERIFY(cond_broadcast(&rll->rll_cv) == 0);
916 
917 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
918 }
919 
920 static void
921 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
922 {
923 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
924 
925 	ztest_rll_lock(rll, type);
926 }
927 
928 static void
929 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
930 {
931 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
932 
933 	ztest_rll_unlock(rll);
934 }
935 
936 static rl_t *
937 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
938     uint64_t size, rl_type_t type)
939 {
940 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
941 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
942 	rl_t *rl;
943 
944 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
945 	rl->rl_object = object;
946 	rl->rl_offset = offset;
947 	rl->rl_size = size;
948 	rl->rl_lock = rll;
949 
950 	ztest_rll_lock(rll, type);
951 
952 	return (rl);
953 }
954 
955 static void
956 ztest_range_unlock(rl_t *rl)
957 {
958 	rll_t *rll = rl->rl_lock;
959 
960 	ztest_rll_unlock(rll);
961 
962 	umem_free(rl, sizeof (*rl));
963 }
964 
965 static void
966 ztest_zd_init(ztest_ds_t *zd, objset_t *os)
967 {
968 	zd->zd_os = os;
969 	zd->zd_zilog = dmu_objset_zil(os);
970 	zd->zd_seq = 0;
971 	dmu_objset_name(os, zd->zd_name);
972 
973 	VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
974 
975 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
976 		ztest_rll_init(&zd->zd_object_lock[l]);
977 
978 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
979 		ztest_rll_init(&zd->zd_range_lock[l]);
980 }
981 
982 static void
983 ztest_zd_fini(ztest_ds_t *zd)
984 {
985 	VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
986 
987 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
988 		ztest_rll_destroy(&zd->zd_object_lock[l]);
989 
990 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
991 		ztest_rll_destroy(&zd->zd_range_lock[l]);
992 }
993 
994 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
995 
996 static uint64_t
997 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
998 {
999 	uint64_t txg;
1000 	int error;
1001 
1002 	/*
1003 	 * Attempt to assign tx to some transaction group.
1004 	 */
1005 	error = dmu_tx_assign(tx, txg_how);
1006 	if (error) {
1007 		if (error == ERESTART) {
1008 			ASSERT(txg_how == TXG_NOWAIT);
1009 			dmu_tx_wait(tx);
1010 		} else {
1011 			ASSERT3U(error, ==, ENOSPC);
1012 			ztest_record_enospc(tag);
1013 		}
1014 		dmu_tx_abort(tx);
1015 		return (0);
1016 	}
1017 	txg = dmu_tx_get_txg(tx);
1018 	ASSERT(txg != 0);
1019 	return (txg);
1020 }
1021 
1022 static void
1023 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1024 {
1025 	uint64_t *ip = buf;
1026 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1027 
1028 	while (ip < ip_end)
1029 		*ip++ = value;
1030 }
1031 
1032 static boolean_t
1033 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1034 {
1035 	uint64_t *ip = buf;
1036 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1037 	uint64_t diff = 0;
1038 
1039 	while (ip < ip_end)
1040 		diff |= (value - *ip++);
1041 
1042 	return (diff == 0);
1043 }
1044 
1045 static void
1046 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1047     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1048 {
1049 	bt->bt_magic = BT_MAGIC;
1050 	bt->bt_objset = dmu_objset_id(os);
1051 	bt->bt_object = object;
1052 	bt->bt_offset = offset;
1053 	bt->bt_gen = gen;
1054 	bt->bt_txg = txg;
1055 	bt->bt_crtxg = crtxg;
1056 }
1057 
1058 static void
1059 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1060     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1061 {
1062 	ASSERT(bt->bt_magic == BT_MAGIC);
1063 	ASSERT(bt->bt_objset == dmu_objset_id(os));
1064 	ASSERT(bt->bt_object == object);
1065 	ASSERT(bt->bt_offset == offset);
1066 	ASSERT(bt->bt_gen <= gen);
1067 	ASSERT(bt->bt_txg <= txg);
1068 	ASSERT(bt->bt_crtxg == crtxg);
1069 }
1070 
1071 static ztest_block_tag_t *
1072 ztest_bt_bonus(dmu_buf_t *db)
1073 {
1074 	dmu_object_info_t doi;
1075 	ztest_block_tag_t *bt;
1076 
1077 	dmu_object_info_from_db(db, &doi);
1078 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1079 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1080 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1081 
1082 	return (bt);
1083 }
1084 
1085 /*
1086  * ZIL logging ops
1087  */
1088 
1089 #define	lrz_type	lr_mode
1090 #define	lrz_blocksize	lr_uid
1091 #define	lrz_ibshift	lr_gid
1092 #define	lrz_bonustype	lr_rdev
1093 #define	lrz_bonuslen	lr_crtime[1]
1094 
1095 static uint64_t
1096 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1097 {
1098 	char *name = (void *)(lr + 1);		/* name follows lr */
1099 	size_t namesize = strlen(name) + 1;
1100 	itx_t *itx;
1101 
1102 	if (zil_replaying(zd->zd_zilog, tx))
1103 		return (0);
1104 
1105 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1106 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1107 	    sizeof (*lr) + namesize - sizeof (lr_t));
1108 
1109 	return (zil_itx_assign(zd->zd_zilog, itx, tx));
1110 }
1111 
1112 static uint64_t
1113 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr)
1114 {
1115 	char *name = (void *)(lr + 1);		/* name follows lr */
1116 	size_t namesize = strlen(name) + 1;
1117 	itx_t *itx;
1118 
1119 	if (zil_replaying(zd->zd_zilog, tx))
1120 		return (0);
1121 
1122 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1123 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1124 	    sizeof (*lr) + namesize - sizeof (lr_t));
1125 
1126 	return (zil_itx_assign(zd->zd_zilog, itx, tx));
1127 }
1128 
1129 static uint64_t
1130 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1131 {
1132 	itx_t *itx;
1133 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1134 
1135 	if (zil_replaying(zd->zd_zilog, tx))
1136 		return (0);
1137 
1138 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1139 		write_state = WR_INDIRECT;
1140 
1141 	itx = zil_itx_create(TX_WRITE,
1142 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1143 
1144 	if (write_state == WR_COPIED &&
1145 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1146 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1147 		zil_itx_destroy(itx);
1148 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1149 		write_state = WR_NEED_COPY;
1150 	}
1151 	itx->itx_private = zd;
1152 	itx->itx_wr_state = write_state;
1153 	itx->itx_sync = (ztest_random(8) == 0);
1154 	itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1155 
1156 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1157 	    sizeof (*lr) - sizeof (lr_t));
1158 
1159 	return (zil_itx_assign(zd->zd_zilog, itx, tx));
1160 }
1161 
1162 static uint64_t
1163 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1164 {
1165 	itx_t *itx;
1166 
1167 	if (zil_replaying(zd->zd_zilog, tx))
1168 		return (0);
1169 
1170 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1171 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1172 	    sizeof (*lr) - sizeof (lr_t));
1173 
1174 	return (zil_itx_assign(zd->zd_zilog, itx, tx));
1175 }
1176 
1177 static uint64_t
1178 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1179 {
1180 	itx_t *itx;
1181 
1182 	if (zil_replaying(zd->zd_zilog, tx))
1183 		return (0);
1184 
1185 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1186 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1187 	    sizeof (*lr) - sizeof (lr_t));
1188 
1189 	return (zil_itx_assign(zd->zd_zilog, itx, tx));
1190 }
1191 
1192 /*
1193  * ZIL replay ops
1194  */
1195 static int
1196 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1197 {
1198 	char *name = (void *)(lr + 1);		/* name follows lr */
1199 	objset_t *os = zd->zd_os;
1200 	ztest_block_tag_t *bbt;
1201 	dmu_buf_t *db;
1202 	dmu_tx_t *tx;
1203 	uint64_t txg;
1204 	int error = 0;
1205 
1206 	if (byteswap)
1207 		byteswap_uint64_array(lr, sizeof (*lr));
1208 
1209 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1210 	ASSERT(name[0] != '\0');
1211 
1212 	tx = dmu_tx_create(os);
1213 
1214 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1215 
1216 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1217 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1218 	} else {
1219 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1220 	}
1221 
1222 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1223 	if (txg == 0)
1224 		return (ENOSPC);
1225 
1226 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1227 
1228 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1229 		if (lr->lr_foid == 0) {
1230 			lr->lr_foid = zap_create(os,
1231 			    lr->lrz_type, lr->lrz_bonustype,
1232 			    lr->lrz_bonuslen, tx);
1233 		} else {
1234 			error = zap_create_claim(os, lr->lr_foid,
1235 			    lr->lrz_type, lr->lrz_bonustype,
1236 			    lr->lrz_bonuslen, tx);
1237 		}
1238 	} else {
1239 		if (lr->lr_foid == 0) {
1240 			lr->lr_foid = dmu_object_alloc(os,
1241 			    lr->lrz_type, 0, lr->lrz_bonustype,
1242 			    lr->lrz_bonuslen, tx);
1243 		} else {
1244 			error = dmu_object_claim(os, lr->lr_foid,
1245 			    lr->lrz_type, 0, lr->lrz_bonustype,
1246 			    lr->lrz_bonuslen, tx);
1247 		}
1248 	}
1249 
1250 	if (error) {
1251 		ASSERT3U(error, ==, EEXIST);
1252 		ASSERT(zd->zd_zilog->zl_replay);
1253 		dmu_tx_commit(tx);
1254 		return (error);
1255 	}
1256 
1257 	ASSERT(lr->lr_foid != 0);
1258 
1259 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1260 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1261 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1262 
1263 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1264 	bbt = ztest_bt_bonus(db);
1265 	dmu_buf_will_dirty(db, tx);
1266 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1267 	dmu_buf_rele(db, FTAG);
1268 
1269 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1270 	    &lr->lr_foid, tx));
1271 
1272 	(void) ztest_log_create(zd, tx, lr);
1273 
1274 	dmu_tx_commit(tx);
1275 
1276 	return (0);
1277 }
1278 
1279 static int
1280 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1281 {
1282 	char *name = (void *)(lr + 1);		/* name follows lr */
1283 	objset_t *os = zd->zd_os;
1284 	dmu_object_info_t doi;
1285 	dmu_tx_t *tx;
1286 	uint64_t object, txg;
1287 
1288 	if (byteswap)
1289 		byteswap_uint64_array(lr, sizeof (*lr));
1290 
1291 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1292 	ASSERT(name[0] != '\0');
1293 
1294 	VERIFY3U(0, ==,
1295 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1296 	ASSERT(object != 0);
1297 
1298 	ztest_object_lock(zd, object, RL_WRITER);
1299 
1300 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1301 
1302 	tx = dmu_tx_create(os);
1303 
1304 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1305 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1306 
1307 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1308 	if (txg == 0) {
1309 		ztest_object_unlock(zd, object);
1310 		return (ENOSPC);
1311 	}
1312 
1313 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1314 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1315 	} else {
1316 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1317 	}
1318 
1319 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1320 
1321 	(void) ztest_log_remove(zd, tx, lr);
1322 
1323 	dmu_tx_commit(tx);
1324 
1325 	ztest_object_unlock(zd, object);
1326 
1327 	return (0);
1328 }
1329 
1330 static int
1331 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1332 {
1333 	objset_t *os = zd->zd_os;
1334 	void *data = lr + 1;			/* data follows lr */
1335 	uint64_t offset, length;
1336 	ztest_block_tag_t *bt = data;
1337 	ztest_block_tag_t *bbt;
1338 	uint64_t gen, txg, lrtxg, crtxg;
1339 	dmu_object_info_t doi;
1340 	dmu_tx_t *tx;
1341 	dmu_buf_t *db;
1342 	arc_buf_t *abuf = NULL;
1343 	rl_t *rl;
1344 
1345 	if (byteswap)
1346 		byteswap_uint64_array(lr, sizeof (*lr));
1347 
1348 	offset = lr->lr_offset;
1349 	length = lr->lr_length;
1350 
1351 	/* If it's a dmu_sync() block, write the whole block */
1352 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1353 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1354 		if (length < blocksize) {
1355 			offset -= offset % blocksize;
1356 			length = blocksize;
1357 		}
1358 	}
1359 
1360 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1361 		byteswap_uint64_array(bt, sizeof (*bt));
1362 
1363 	if (bt->bt_magic != BT_MAGIC)
1364 		bt = NULL;
1365 
1366 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1367 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1368 
1369 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1370 
1371 	dmu_object_info_from_db(db, &doi);
1372 
1373 	bbt = ztest_bt_bonus(db);
1374 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1375 	gen = bbt->bt_gen;
1376 	crtxg = bbt->bt_crtxg;
1377 	lrtxg = lr->lr_common.lrc_txg;
1378 
1379 	tx = dmu_tx_create(os);
1380 
1381 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1382 
1383 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1384 	    P2PHASE(offset, length) == 0)
1385 		abuf = dmu_request_arcbuf(db, length);
1386 
1387 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1388 	if (txg == 0) {
1389 		if (abuf != NULL)
1390 			dmu_return_arcbuf(abuf);
1391 		dmu_buf_rele(db, FTAG);
1392 		ztest_range_unlock(rl);
1393 		ztest_object_unlock(zd, lr->lr_foid);
1394 		return (ENOSPC);
1395 	}
1396 
1397 	if (bt != NULL) {
1398 		/*
1399 		 * Usually, verify the old data before writing new data --
1400 		 * but not always, because we also want to verify correct
1401 		 * behavior when the data was not recently read into cache.
1402 		 */
1403 		ASSERT(offset % doi.doi_data_block_size == 0);
1404 		if (ztest_random(4) != 0) {
1405 			int prefetch = ztest_random(2) ?
1406 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1407 			ztest_block_tag_t rbt;
1408 
1409 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1410 			    sizeof (rbt), &rbt, prefetch) == 0);
1411 			if (rbt.bt_magic == BT_MAGIC) {
1412 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1413 				    offset, gen, txg, crtxg);
1414 			}
1415 		}
1416 
1417 		/*
1418 		 * Writes can appear to be newer than the bonus buffer because
1419 		 * the ztest_get_data() callback does a dmu_read() of the
1420 		 * open-context data, which may be different than the data
1421 		 * as it was when the write was generated.
1422 		 */
1423 		if (zd->zd_zilog->zl_replay) {
1424 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1425 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1426 			    bt->bt_crtxg);
1427 		}
1428 
1429 		/*
1430 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1431 		 * so that all of the usual ASSERTs will work.
1432 		 */
1433 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1434 	}
1435 
1436 	if (abuf == NULL) {
1437 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1438 	} else {
1439 		bcopy(data, abuf->b_data, length);
1440 		dmu_assign_arcbuf(db, offset, abuf, tx);
1441 	}
1442 
1443 	(void) ztest_log_write(zd, tx, lr);
1444 
1445 	dmu_buf_rele(db, FTAG);
1446 
1447 	dmu_tx_commit(tx);
1448 
1449 	ztest_range_unlock(rl);
1450 	ztest_object_unlock(zd, lr->lr_foid);
1451 
1452 	return (0);
1453 }
1454 
1455 static int
1456 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1457 {
1458 	objset_t *os = zd->zd_os;
1459 	dmu_tx_t *tx;
1460 	uint64_t txg;
1461 	rl_t *rl;
1462 
1463 	if (byteswap)
1464 		byteswap_uint64_array(lr, sizeof (*lr));
1465 
1466 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1467 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1468 	    RL_WRITER);
1469 
1470 	tx = dmu_tx_create(os);
1471 
1472 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1473 
1474 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1475 	if (txg == 0) {
1476 		ztest_range_unlock(rl);
1477 		ztest_object_unlock(zd, lr->lr_foid);
1478 		return (ENOSPC);
1479 	}
1480 
1481 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1482 	    lr->lr_length, tx) == 0);
1483 
1484 	(void) ztest_log_truncate(zd, tx, lr);
1485 
1486 	dmu_tx_commit(tx);
1487 
1488 	ztest_range_unlock(rl);
1489 	ztest_object_unlock(zd, lr->lr_foid);
1490 
1491 	return (0);
1492 }
1493 
1494 static int
1495 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1496 {
1497 	objset_t *os = zd->zd_os;
1498 	dmu_tx_t *tx;
1499 	dmu_buf_t *db;
1500 	ztest_block_tag_t *bbt;
1501 	uint64_t txg, lrtxg, crtxg;
1502 
1503 	if (byteswap)
1504 		byteswap_uint64_array(lr, sizeof (*lr));
1505 
1506 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1507 
1508 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1509 
1510 	tx = dmu_tx_create(os);
1511 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1512 
1513 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1514 	if (txg == 0) {
1515 		dmu_buf_rele(db, FTAG);
1516 		ztest_object_unlock(zd, lr->lr_foid);
1517 		return (ENOSPC);
1518 	}
1519 
1520 	bbt = ztest_bt_bonus(db);
1521 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1522 	crtxg = bbt->bt_crtxg;
1523 	lrtxg = lr->lr_common.lrc_txg;
1524 
1525 	if (zd->zd_zilog->zl_replay) {
1526 		ASSERT(lr->lr_size != 0);
1527 		ASSERT(lr->lr_mode != 0);
1528 		ASSERT(lrtxg != 0);
1529 	} else {
1530 		/*
1531 		 * Randomly change the size and increment the generation.
1532 		 */
1533 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1534 		    sizeof (*bbt);
1535 		lr->lr_mode = bbt->bt_gen + 1;
1536 		ASSERT(lrtxg == 0);
1537 	}
1538 
1539 	/*
1540 	 * Verify that the current bonus buffer is not newer than our txg.
1541 	 */
1542 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1543 	    MAX(txg, lrtxg), crtxg);
1544 
1545 	dmu_buf_will_dirty(db, tx);
1546 
1547 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1548 	ASSERT3U(lr->lr_size, <=, db->db_size);
1549 	VERIFY3U(dmu_set_bonus(db, lr->lr_size, tx), ==, 0);
1550 	bbt = ztest_bt_bonus(db);
1551 
1552 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1553 
1554 	dmu_buf_rele(db, FTAG);
1555 
1556 	(void) ztest_log_setattr(zd, tx, lr);
1557 
1558 	dmu_tx_commit(tx);
1559 
1560 	ztest_object_unlock(zd, lr->lr_foid);
1561 
1562 	return (0);
1563 }
1564 
1565 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1566 	NULL,			/* 0 no such transaction type */
1567 	ztest_replay_create,	/* TX_CREATE */
1568 	NULL,			/* TX_MKDIR */
1569 	NULL,			/* TX_MKXATTR */
1570 	NULL,			/* TX_SYMLINK */
1571 	ztest_replay_remove,	/* TX_REMOVE */
1572 	NULL,			/* TX_RMDIR */
1573 	NULL,			/* TX_LINK */
1574 	NULL,			/* TX_RENAME */
1575 	ztest_replay_write,	/* TX_WRITE */
1576 	ztest_replay_truncate,	/* TX_TRUNCATE */
1577 	ztest_replay_setattr,	/* TX_SETATTR */
1578 	NULL,			/* TX_ACL */
1579 	NULL,			/* TX_CREATE_ACL */
1580 	NULL,			/* TX_CREATE_ATTR */
1581 	NULL,			/* TX_CREATE_ACL_ATTR */
1582 	NULL,			/* TX_MKDIR_ACL */
1583 	NULL,			/* TX_MKDIR_ATTR */
1584 	NULL,			/* TX_MKDIR_ACL_ATTR */
1585 	NULL,			/* TX_WRITE2 */
1586 };
1587 
1588 /*
1589  * ZIL get_data callbacks
1590  */
1591 
1592 static void
1593 ztest_get_done(zgd_t *zgd, int error)
1594 {
1595 	ztest_ds_t *zd = zgd->zgd_private;
1596 	uint64_t object = zgd->zgd_rl->rl_object;
1597 
1598 	if (zgd->zgd_db)
1599 		dmu_buf_rele(zgd->zgd_db, zgd);
1600 
1601 	ztest_range_unlock(zgd->zgd_rl);
1602 	ztest_object_unlock(zd, object);
1603 
1604 	if (error == 0 && zgd->zgd_bp)
1605 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1606 
1607 	umem_free(zgd, sizeof (*zgd));
1608 }
1609 
1610 static int
1611 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1612 {
1613 	ztest_ds_t *zd = arg;
1614 	objset_t *os = zd->zd_os;
1615 	uint64_t object = lr->lr_foid;
1616 	uint64_t offset = lr->lr_offset;
1617 	uint64_t size = lr->lr_length;
1618 	blkptr_t *bp = &lr->lr_blkptr;
1619 	uint64_t txg = lr->lr_common.lrc_txg;
1620 	uint64_t crtxg;
1621 	dmu_object_info_t doi;
1622 	dmu_buf_t *db;
1623 	zgd_t *zgd;
1624 	int error;
1625 
1626 	ztest_object_lock(zd, object, RL_READER);
1627 	error = dmu_bonus_hold(os, object, FTAG, &db);
1628 	if (error) {
1629 		ztest_object_unlock(zd, object);
1630 		return (error);
1631 	}
1632 
1633 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1634 
1635 	if (crtxg == 0 || crtxg > txg) {
1636 		dmu_buf_rele(db, FTAG);
1637 		ztest_object_unlock(zd, object);
1638 		return (ENOENT);
1639 	}
1640 
1641 	dmu_object_info_from_db(db, &doi);
1642 	dmu_buf_rele(db, FTAG);
1643 	db = NULL;
1644 
1645 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1646 	zgd->zgd_zilog = zd->zd_zilog;
1647 	zgd->zgd_private = zd;
1648 
1649 	if (buf != NULL) {	/* immediate write */
1650 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1651 		    RL_READER);
1652 
1653 		error = dmu_read(os, object, offset, size, buf,
1654 		    DMU_READ_NO_PREFETCH);
1655 		ASSERT(error == 0);
1656 	} else {
1657 		size = doi.doi_data_block_size;
1658 		if (ISP2(size))
1659 			offset = P2ALIGN(offset, size);
1660 
1661 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1662 		    RL_READER);
1663 
1664 		error = dmu_buf_hold(os, object, offset, zgd, &db);
1665 
1666 		if (error == 0) {
1667 			zgd->zgd_db = db;
1668 			zgd->zgd_bp = bp;
1669 
1670 			ASSERT(db->db_offset == offset);
1671 			ASSERT(db->db_size == size);
1672 
1673 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1674 			    ztest_get_done, zgd);
1675 
1676 			if (error == 0)
1677 				return (0);
1678 		}
1679 	}
1680 
1681 	ztest_get_done(zgd, error);
1682 
1683 	return (error);
1684 }
1685 
1686 static void *
1687 ztest_lr_alloc(size_t lrsize, char *name)
1688 {
1689 	char *lr;
1690 	size_t namesize = name ? strlen(name) + 1 : 0;
1691 
1692 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1693 
1694 	if (name)
1695 		bcopy(name, lr + lrsize, namesize);
1696 
1697 	return (lr);
1698 }
1699 
1700 void
1701 ztest_lr_free(void *lr, size_t lrsize, char *name)
1702 {
1703 	size_t namesize = name ? strlen(name) + 1 : 0;
1704 
1705 	umem_free(lr, lrsize + namesize);
1706 }
1707 
1708 /*
1709  * Lookup a bunch of objects.  Returns the number of objects not found.
1710  */
1711 static int
1712 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1713 {
1714 	int missing = 0;
1715 	int error;
1716 
1717 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1718 
1719 	for (int i = 0; i < count; i++, od++) {
1720 		od->od_object = 0;
1721 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1722 		    sizeof (uint64_t), 1, &od->od_object);
1723 		if (error) {
1724 			ASSERT(error == ENOENT);
1725 			ASSERT(od->od_object == 0);
1726 			missing++;
1727 		} else {
1728 			dmu_buf_t *db;
1729 			ztest_block_tag_t *bbt;
1730 			dmu_object_info_t doi;
1731 
1732 			ASSERT(od->od_object != 0);
1733 			ASSERT(missing == 0);	/* there should be no gaps */
1734 
1735 			ztest_object_lock(zd, od->od_object, RL_READER);
1736 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1737 			    od->od_object, FTAG, &db));
1738 			dmu_object_info_from_db(db, &doi);
1739 			bbt = ztest_bt_bonus(db);
1740 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1741 			od->od_type = doi.doi_type;
1742 			od->od_blocksize = doi.doi_data_block_size;
1743 			od->od_gen = bbt->bt_gen;
1744 			dmu_buf_rele(db, FTAG);
1745 			ztest_object_unlock(zd, od->od_object);
1746 		}
1747 	}
1748 
1749 	return (missing);
1750 }
1751 
1752 static int
1753 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1754 {
1755 	int missing = 0;
1756 
1757 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1758 
1759 	for (int i = 0; i < count; i++, od++) {
1760 		if (missing) {
1761 			od->od_object = 0;
1762 			missing++;
1763 			continue;
1764 		}
1765 
1766 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1767 
1768 		lr->lr_doid = od->od_dir;
1769 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
1770 		lr->lrz_type = od->od_crtype;
1771 		lr->lrz_blocksize = od->od_crblocksize;
1772 		lr->lrz_ibshift = ztest_random_ibshift();
1773 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1774 		lr->lrz_bonuslen = dmu_bonus_max();
1775 		lr->lr_gen = od->od_crgen;
1776 		lr->lr_crtime[0] = time(NULL);
1777 
1778 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1779 			ASSERT(missing == 0);
1780 			od->od_object = 0;
1781 			missing++;
1782 		} else {
1783 			od->od_object = lr->lr_foid;
1784 			od->od_type = od->od_crtype;
1785 			od->od_blocksize = od->od_crblocksize;
1786 			od->od_gen = od->od_crgen;
1787 			ASSERT(od->od_object != 0);
1788 		}
1789 
1790 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
1791 	}
1792 
1793 	return (missing);
1794 }
1795 
1796 static int
1797 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
1798 {
1799 	int missing = 0;
1800 	int error;
1801 
1802 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1803 
1804 	od += count - 1;
1805 
1806 	for (int i = count - 1; i >= 0; i--, od--) {
1807 		if (missing) {
1808 			missing++;
1809 			continue;
1810 		}
1811 
1812 		if (od->od_object == 0)
1813 			continue;
1814 
1815 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1816 
1817 		lr->lr_doid = od->od_dir;
1818 
1819 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
1820 			ASSERT3U(error, ==, ENOSPC);
1821 			missing++;
1822 		} else {
1823 			od->od_object = 0;
1824 		}
1825 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
1826 	}
1827 
1828 	return (missing);
1829 }
1830 
1831 static int
1832 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
1833     void *data)
1834 {
1835 	lr_write_t *lr;
1836 	int error;
1837 
1838 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
1839 
1840 	lr->lr_foid = object;
1841 	lr->lr_offset = offset;
1842 	lr->lr_length = size;
1843 	lr->lr_blkoff = 0;
1844 	BP_ZERO(&lr->lr_blkptr);
1845 
1846 	bcopy(data, lr + 1, size);
1847 
1848 	error = ztest_replay_write(zd, lr, B_FALSE);
1849 
1850 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
1851 
1852 	return (error);
1853 }
1854 
1855 static int
1856 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
1857 {
1858 	lr_truncate_t *lr;
1859 	int error;
1860 
1861 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
1862 
1863 	lr->lr_foid = object;
1864 	lr->lr_offset = offset;
1865 	lr->lr_length = size;
1866 
1867 	error = ztest_replay_truncate(zd, lr, B_FALSE);
1868 
1869 	ztest_lr_free(lr, sizeof (*lr), NULL);
1870 
1871 	return (error);
1872 }
1873 
1874 static int
1875 ztest_setattr(ztest_ds_t *zd, uint64_t object)
1876 {
1877 	lr_setattr_t *lr;
1878 	int error;
1879 
1880 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
1881 
1882 	lr->lr_foid = object;
1883 	lr->lr_size = 0;
1884 	lr->lr_mode = 0;
1885 
1886 	error = ztest_replay_setattr(zd, lr, B_FALSE);
1887 
1888 	ztest_lr_free(lr, sizeof (*lr), NULL);
1889 
1890 	return (error);
1891 }
1892 
1893 static void
1894 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
1895 {
1896 	objset_t *os = zd->zd_os;
1897 	dmu_tx_t *tx;
1898 	uint64_t txg;
1899 	rl_t *rl;
1900 
1901 	txg_wait_synced(dmu_objset_pool(os), 0);
1902 
1903 	ztest_object_lock(zd, object, RL_READER);
1904 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
1905 
1906 	tx = dmu_tx_create(os);
1907 
1908 	dmu_tx_hold_write(tx, object, offset, size);
1909 
1910 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1911 
1912 	if (txg != 0) {
1913 		dmu_prealloc(os, object, offset, size, tx);
1914 		dmu_tx_commit(tx);
1915 		txg_wait_synced(dmu_objset_pool(os), txg);
1916 	} else {
1917 		(void) dmu_free_long_range(os, object, offset, size);
1918 	}
1919 
1920 	ztest_range_unlock(rl);
1921 	ztest_object_unlock(zd, object);
1922 }
1923 
1924 static void
1925 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
1926 {
1927 	ztest_block_tag_t wbt;
1928 	dmu_object_info_t doi;
1929 	enum ztest_io_type io_type;
1930 	uint64_t blocksize;
1931 	void *data;
1932 
1933 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
1934 	blocksize = doi.doi_data_block_size;
1935 	data = umem_alloc(blocksize, UMEM_NOFAIL);
1936 
1937 	/*
1938 	 * Pick an i/o type at random, biased toward writing block tags.
1939 	 */
1940 	io_type = ztest_random(ZTEST_IO_TYPES);
1941 	if (ztest_random(2) == 0)
1942 		io_type = ZTEST_IO_WRITE_TAG;
1943 
1944 	switch (io_type) {
1945 
1946 	case ZTEST_IO_WRITE_TAG:
1947 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
1948 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
1949 		break;
1950 
1951 	case ZTEST_IO_WRITE_PATTERN:
1952 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
1953 		if (ztest_random(2) == 0) {
1954 			/*
1955 			 * Induce fletcher2 collisions to ensure that
1956 			 * zio_ddt_collision() detects and resolves them
1957 			 * when using fletcher2-verify for deduplication.
1958 			 */
1959 			((uint64_t *)data)[0] ^= 1ULL << 63;
1960 			((uint64_t *)data)[4] ^= 1ULL << 63;
1961 		}
1962 		(void) ztest_write(zd, object, offset, blocksize, data);
1963 		break;
1964 
1965 	case ZTEST_IO_WRITE_ZEROES:
1966 		bzero(data, blocksize);
1967 		(void) ztest_write(zd, object, offset, blocksize, data);
1968 		break;
1969 
1970 	case ZTEST_IO_TRUNCATE:
1971 		(void) ztest_truncate(zd, object, offset, blocksize);
1972 		break;
1973 
1974 	case ZTEST_IO_SETATTR:
1975 		(void) ztest_setattr(zd, object);
1976 		break;
1977 	}
1978 
1979 	umem_free(data, blocksize);
1980 }
1981 
1982 /*
1983  * Initialize an object description template.
1984  */
1985 static void
1986 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
1987     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
1988 {
1989 	od->od_dir = ZTEST_DIROBJ;
1990 	od->od_object = 0;
1991 
1992 	od->od_crtype = type;
1993 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
1994 	od->od_crgen = gen;
1995 
1996 	od->od_type = DMU_OT_NONE;
1997 	od->od_blocksize = 0;
1998 	od->od_gen = 0;
1999 
2000 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2001 	    tag, (int64_t)id, index);
2002 }
2003 
2004 /*
2005  * Lookup or create the objects for a test using the od template.
2006  * If the objects do not all exist, or if 'remove' is specified,
2007  * remove any existing objects and create new ones.  Otherwise,
2008  * use the existing objects.
2009  */
2010 static int
2011 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2012 {
2013 	int count = size / sizeof (*od);
2014 	int rv = 0;
2015 
2016 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2017 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2018 	    (ztest_remove(zd, od, count) != 0 ||
2019 	    ztest_create(zd, od, count) != 0))
2020 		rv = -1;
2021 	zd->zd_od = od;
2022 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2023 
2024 	return (rv);
2025 }
2026 
2027 /* ARGSUSED */
2028 void
2029 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2030 {
2031 	zilog_t *zilog = zd->zd_zilog;
2032 
2033 	zil_commit(zilog, UINT64_MAX, ztest_random(ZTEST_OBJECTS));
2034 
2035 	/*
2036 	 * Remember the committed values in zd, which is in parent/child
2037 	 * shared memory.  If we die, the next iteration of ztest_run()
2038 	 * will verify that the log really does contain this record.
2039 	 */
2040 	mutex_enter(&zilog->zl_lock);
2041 	ASSERT(zd->zd_seq <= zilog->zl_commit_lr_seq);
2042 	zd->zd_seq = zilog->zl_commit_lr_seq;
2043 	mutex_exit(&zilog->zl_lock);
2044 }
2045 
2046 /*
2047  * Verify that we can't destroy an active pool, create an existing pool,
2048  * or create a pool with a bad vdev spec.
2049  */
2050 /* ARGSUSED */
2051 void
2052 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2053 {
2054 	ztest_shared_t *zs = ztest_shared;
2055 	spa_t *spa;
2056 	nvlist_t *nvroot;
2057 
2058 	/*
2059 	 * Attempt to create using a bad file.
2060 	 */
2061 	nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2062 	VERIFY3U(ENOENT, ==,
2063 	    spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2064 	nvlist_free(nvroot);
2065 
2066 	/*
2067 	 * Attempt to create using a bad mirror.
2068 	 */
2069 	nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1);
2070 	VERIFY3U(ENOENT, ==,
2071 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2072 	nvlist_free(nvroot);
2073 
2074 	/*
2075 	 * Attempt to create an existing pool.  It shouldn't matter
2076 	 * what's in the nvroot; we should fail with EEXIST.
2077 	 */
2078 	(void) rw_rdlock(&zs->zs_name_lock);
2079 	nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2080 	VERIFY3U(EEXIST, ==, spa_create(zs->zs_pool, nvroot, NULL, NULL, NULL));
2081 	nvlist_free(nvroot);
2082 	VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
2083 	VERIFY3U(EBUSY, ==, spa_destroy(zs->zs_pool));
2084 	spa_close(spa, FTAG);
2085 
2086 	(void) rw_unlock(&zs->zs_name_lock);
2087 }
2088 
2089 static vdev_t *
2090 vdev_lookup_by_path(vdev_t *vd, const char *path)
2091 {
2092 	vdev_t *mvd;
2093 
2094 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2095 		return (vd);
2096 
2097 	for (int c = 0; c < vd->vdev_children; c++)
2098 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2099 		    NULL)
2100 			return (mvd);
2101 
2102 	return (NULL);
2103 }
2104 
2105 /*
2106  * Find the first available hole which can be used as a top-level.
2107  */
2108 int
2109 find_vdev_hole(spa_t *spa)
2110 {
2111 	vdev_t *rvd = spa->spa_root_vdev;
2112 	int c;
2113 
2114 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2115 
2116 	for (c = 0; c < rvd->vdev_children; c++) {
2117 		vdev_t *cvd = rvd->vdev_child[c];
2118 
2119 		if (cvd->vdev_ishole)
2120 			break;
2121 	}
2122 	return (c);
2123 }
2124 
2125 /*
2126  * Verify that vdev_add() works as expected.
2127  */
2128 /* ARGSUSED */
2129 void
2130 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2131 {
2132 	ztest_shared_t *zs = ztest_shared;
2133 	spa_t *spa = zs->zs_spa;
2134 	uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
2135 	uint64_t guid;
2136 	nvlist_t *nvroot;
2137 	int error;
2138 
2139 	VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2140 
2141 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2142 
2143 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2144 
2145 	/*
2146 	 * If we have slogs then remove them 1/4 of the time.
2147 	 */
2148 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2149 		/*
2150 		 * Grab the guid from the head of the log class rotor.
2151 		 */
2152 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2153 
2154 		spa_config_exit(spa, SCL_VDEV, FTAG);
2155 
2156 		/*
2157 		 * We have to grab the zs_name_lock as writer to
2158 		 * prevent a race between removing a slog (dmu_objset_find)
2159 		 * and destroying a dataset. Removing the slog will
2160 		 * grab a reference on the dataset which may cause
2161 		 * dmu_objset_destroy() to fail with EBUSY thus
2162 		 * leaving the dataset in an inconsistent state.
2163 		 */
2164 		VERIFY(rw_wrlock(&ztest_shared->zs_name_lock) == 0);
2165 		error = spa_vdev_remove(spa, guid, B_FALSE);
2166 		VERIFY(rw_unlock(&ztest_shared->zs_name_lock) == 0);
2167 
2168 		if (error && error != EEXIST)
2169 			fatal(0, "spa_vdev_remove() = %d", error);
2170 	} else {
2171 		spa_config_exit(spa, SCL_VDEV, FTAG);
2172 
2173 		/*
2174 		 * Make 1/4 of the devices be log devices.
2175 		 */
2176 		nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
2177 		    ztest_random(4) == 0, zopt_raidz, zopt_mirrors, 1);
2178 
2179 		error = spa_vdev_add(spa, nvroot);
2180 		nvlist_free(nvroot);
2181 
2182 		if (error == ENOSPC)
2183 			ztest_record_enospc("spa_vdev_add");
2184 		else if (error != 0)
2185 			fatal(0, "spa_vdev_add() = %d", error);
2186 	}
2187 
2188 	VERIFY(mutex_unlock(&ztest_shared->zs_vdev_lock) == 0);
2189 }
2190 
2191 /*
2192  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2193  */
2194 /* ARGSUSED */
2195 void
2196 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2197 {
2198 	ztest_shared_t *zs = ztest_shared;
2199 	spa_t *spa = zs->zs_spa;
2200 	vdev_t *rvd = spa->spa_root_vdev;
2201 	spa_aux_vdev_t *sav;
2202 	char *aux;
2203 	uint64_t guid = 0;
2204 	int error;
2205 
2206 	if (ztest_random(2) == 0) {
2207 		sav = &spa->spa_spares;
2208 		aux = ZPOOL_CONFIG_SPARES;
2209 	} else {
2210 		sav = &spa->spa_l2cache;
2211 		aux = ZPOOL_CONFIG_L2CACHE;
2212 	}
2213 
2214 	VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2215 
2216 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2217 
2218 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2219 		/*
2220 		 * Pick a random device to remove.
2221 		 */
2222 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2223 	} else {
2224 		/*
2225 		 * Find an unused device we can add.
2226 		 */
2227 		zs->zs_vdev_aux = 0;
2228 		for (;;) {
2229 			char path[MAXPATHLEN];
2230 			int c;
2231 			(void) sprintf(path, ztest_aux_template, zopt_dir,
2232 			    zopt_pool, aux, zs->zs_vdev_aux);
2233 			for (c = 0; c < sav->sav_count; c++)
2234 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2235 				    path) == 0)
2236 					break;
2237 			if (c == sav->sav_count &&
2238 			    vdev_lookup_by_path(rvd, path) == NULL)
2239 				break;
2240 			zs->zs_vdev_aux++;
2241 		}
2242 	}
2243 
2244 	spa_config_exit(spa, SCL_VDEV, FTAG);
2245 
2246 	if (guid == 0) {
2247 		/*
2248 		 * Add a new device.
2249 		 */
2250 		nvlist_t *nvroot = make_vdev_root(NULL, aux,
2251 		    (zopt_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2252 		error = spa_vdev_add(spa, nvroot);
2253 		if (error != 0)
2254 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2255 		nvlist_free(nvroot);
2256 	} else {
2257 		/*
2258 		 * Remove an existing device.  Sometimes, dirty its
2259 		 * vdev state first to make sure we handle removal
2260 		 * of devices that have pending state changes.
2261 		 */
2262 		if (ztest_random(2) == 0)
2263 			(void) vdev_online(spa, guid, 0, NULL);
2264 
2265 		error = spa_vdev_remove(spa, guid, B_FALSE);
2266 		if (error != 0 && error != EBUSY)
2267 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2268 	}
2269 
2270 	VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2271 }
2272 
2273 /*
2274  * Verify that we can attach and detach devices.
2275  */
2276 /* ARGSUSED */
2277 void
2278 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2279 {
2280 	ztest_shared_t *zs = ztest_shared;
2281 	spa_t *spa = zs->zs_spa;
2282 	spa_aux_vdev_t *sav = &spa->spa_spares;
2283 	vdev_t *rvd = spa->spa_root_vdev;
2284 	vdev_t *oldvd, *newvd, *pvd;
2285 	nvlist_t *root;
2286 	uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
2287 	uint64_t leaf, top;
2288 	uint64_t ashift = ztest_get_ashift();
2289 	uint64_t oldguid, pguid;
2290 	size_t oldsize, newsize;
2291 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2292 	int replacing;
2293 	int oldvd_has_siblings = B_FALSE;
2294 	int newvd_is_spare = B_FALSE;
2295 	int oldvd_is_log;
2296 	int error, expected_error;
2297 
2298 	VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2299 
2300 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2301 
2302 	/*
2303 	 * Decide whether to do an attach or a replace.
2304 	 */
2305 	replacing = ztest_random(2);
2306 
2307 	/*
2308 	 * Pick a random top-level vdev.
2309 	 */
2310 	top = ztest_random_vdev_top(spa, B_TRUE);
2311 
2312 	/*
2313 	 * Pick a random leaf within it.
2314 	 */
2315 	leaf = ztest_random(leaves);
2316 
2317 	/*
2318 	 * Locate this vdev.
2319 	 */
2320 	oldvd = rvd->vdev_child[top];
2321 	if (zopt_mirrors >= 1) {
2322 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2323 		ASSERT(oldvd->vdev_children >= zopt_mirrors);
2324 		oldvd = oldvd->vdev_child[leaf / zopt_raidz];
2325 	}
2326 	if (zopt_raidz > 1) {
2327 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2328 		ASSERT(oldvd->vdev_children == zopt_raidz);
2329 		oldvd = oldvd->vdev_child[leaf % zopt_raidz];
2330 	}
2331 
2332 	/*
2333 	 * If we're already doing an attach or replace, oldvd may be a
2334 	 * mirror vdev -- in which case, pick a random child.
2335 	 */
2336 	while (oldvd->vdev_children != 0) {
2337 		oldvd_has_siblings = B_TRUE;
2338 		ASSERT(oldvd->vdev_children >= 2);
2339 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2340 	}
2341 
2342 	oldguid = oldvd->vdev_guid;
2343 	oldsize = vdev_get_min_asize(oldvd);
2344 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2345 	(void) strcpy(oldpath, oldvd->vdev_path);
2346 	pvd = oldvd->vdev_parent;
2347 	pguid = pvd->vdev_guid;
2348 
2349 	/*
2350 	 * If oldvd has siblings, then half of the time, detach it.
2351 	 */
2352 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2353 		spa_config_exit(spa, SCL_VDEV, FTAG);
2354 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2355 		if (error != 0 && error != ENODEV && error != EBUSY &&
2356 		    error != ENOTSUP)
2357 			fatal(0, "detach (%s) returned %d", oldpath, error);
2358 		VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2359 		return;
2360 	}
2361 
2362 	/*
2363 	 * For the new vdev, choose with equal probability between the two
2364 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2365 	 */
2366 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2367 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2368 		newvd_is_spare = B_TRUE;
2369 		(void) strcpy(newpath, newvd->vdev_path);
2370 	} else {
2371 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2372 		    zopt_dir, zopt_pool, top * leaves + leaf);
2373 		if (ztest_random(2) == 0)
2374 			newpath[strlen(newpath) - 1] = 'b';
2375 		newvd = vdev_lookup_by_path(rvd, newpath);
2376 	}
2377 
2378 	if (newvd) {
2379 		newsize = vdev_get_min_asize(newvd);
2380 	} else {
2381 		/*
2382 		 * Make newsize a little bigger or smaller than oldsize.
2383 		 * If it's smaller, the attach should fail.
2384 		 * If it's larger, and we're doing a replace,
2385 		 * we should get dynamic LUN growth when we're done.
2386 		 */
2387 		newsize = 10 * oldsize / (9 + ztest_random(3));
2388 	}
2389 
2390 	/*
2391 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2392 	 * unless it's a replace; in that case any non-replacing parent is OK.
2393 	 *
2394 	 * If newvd is already part of the pool, it should fail with EBUSY.
2395 	 *
2396 	 * If newvd is too small, it should fail with EOVERFLOW.
2397 	 */
2398 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2399 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2400 	    pvd->vdev_ops == &vdev_replacing_ops ||
2401 	    pvd->vdev_ops == &vdev_spare_ops))
2402 		expected_error = ENOTSUP;
2403 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2404 		expected_error = ENOTSUP;
2405 	else if (newvd == oldvd)
2406 		expected_error = replacing ? 0 : EBUSY;
2407 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2408 		expected_error = EBUSY;
2409 	else if (newsize < oldsize)
2410 		expected_error = EOVERFLOW;
2411 	else if (ashift > oldvd->vdev_top->vdev_ashift)
2412 		expected_error = EDOM;
2413 	else
2414 		expected_error = 0;
2415 
2416 	spa_config_exit(spa, SCL_VDEV, FTAG);
2417 
2418 	/*
2419 	 * Build the nvlist describing newpath.
2420 	 */
2421 	root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0,
2422 	    ashift, 0, 0, 0, 1);
2423 
2424 	error = spa_vdev_attach(spa, oldguid, root, replacing);
2425 
2426 	nvlist_free(root);
2427 
2428 	/*
2429 	 * If our parent was the replacing vdev, but the replace completed,
2430 	 * then instead of failing with ENOTSUP we may either succeed,
2431 	 * fail with ENODEV, or fail with EOVERFLOW.
2432 	 */
2433 	if (expected_error == ENOTSUP &&
2434 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
2435 		expected_error = error;
2436 
2437 	/*
2438 	 * If someone grew the LUN, the replacement may be too small.
2439 	 */
2440 	if (error == EOVERFLOW || error == EBUSY)
2441 		expected_error = error;
2442 
2443 	/* XXX workaround 6690467 */
2444 	if (error != expected_error && expected_error != EBUSY) {
2445 		fatal(0, "attach (%s %llu, %s %llu, %d) "
2446 		    "returned %d, expected %d",
2447 		    oldpath, (longlong_t)oldsize, newpath,
2448 		    (longlong_t)newsize, replacing, error, expected_error);
2449 	}
2450 
2451 	VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2452 }
2453 
2454 /*
2455  * Callback function which expands the physical size of the vdev.
2456  */
2457 vdev_t *
2458 grow_vdev(vdev_t *vd, void *arg)
2459 {
2460 	spa_t *spa = vd->vdev_spa;
2461 	size_t *newsize = arg;
2462 	size_t fsize;
2463 	int fd;
2464 
2465 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2466 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2467 
2468 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2469 		return (vd);
2470 
2471 	fsize = lseek(fd, 0, SEEK_END);
2472 	(void) ftruncate(fd, *newsize);
2473 
2474 	if (zopt_verbose >= 6) {
2475 		(void) printf("%s grew from %lu to %lu bytes\n",
2476 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2477 	}
2478 	(void) close(fd);
2479 	return (NULL);
2480 }
2481 
2482 /*
2483  * Callback function which expands a given vdev by calling vdev_online().
2484  */
2485 /* ARGSUSED */
2486 vdev_t *
2487 online_vdev(vdev_t *vd, void *arg)
2488 {
2489 	spa_t *spa = vd->vdev_spa;
2490 	vdev_t *tvd = vd->vdev_top;
2491 	uint64_t guid = vd->vdev_guid;
2492 	uint64_t generation = spa->spa_config_generation + 1;
2493 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2494 	int error;
2495 
2496 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2497 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2498 
2499 	/* Calling vdev_online will initialize the new metaslabs */
2500 	spa_config_exit(spa, SCL_STATE, spa);
2501 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2502 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2503 
2504 	/*
2505 	 * If vdev_online returned an error or the underlying vdev_open
2506 	 * failed then we abort the expand. The only way to know that
2507 	 * vdev_open fails is by checking the returned newstate.
2508 	 */
2509 	if (error || newstate != VDEV_STATE_HEALTHY) {
2510 		if (zopt_verbose >= 5) {
2511 			(void) printf("Unable to expand vdev, state %llu, "
2512 			    "error %d\n", (u_longlong_t)newstate, error);
2513 		}
2514 		return (vd);
2515 	}
2516 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2517 
2518 	/*
2519 	 * Since we dropped the lock we need to ensure that we're
2520 	 * still talking to the original vdev. It's possible this
2521 	 * vdev may have been detached/replaced while we were
2522 	 * trying to online it.
2523 	 */
2524 	if (generation != spa->spa_config_generation) {
2525 		if (zopt_verbose >= 5) {
2526 			(void) printf("vdev configuration has changed, "
2527 			    "guid %llu, state %llu, expected gen %llu, "
2528 			    "got gen %llu\n",
2529 			    (u_longlong_t)guid,
2530 			    (u_longlong_t)tvd->vdev_state,
2531 			    (u_longlong_t)generation,
2532 			    (u_longlong_t)spa->spa_config_generation);
2533 		}
2534 		return (vd);
2535 	}
2536 	return (NULL);
2537 }
2538 
2539 /*
2540  * Traverse the vdev tree calling the supplied function.
2541  * We continue to walk the tree until we either have walked all
2542  * children or we receive a non-NULL return from the callback.
2543  * If a NULL callback is passed, then we just return back the first
2544  * leaf vdev we encounter.
2545  */
2546 vdev_t *
2547 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2548 {
2549 	if (vd->vdev_ops->vdev_op_leaf) {
2550 		if (func == NULL)
2551 			return (vd);
2552 		else
2553 			return (func(vd, arg));
2554 	}
2555 
2556 	for (uint_t c = 0; c < vd->vdev_children; c++) {
2557 		vdev_t *cvd = vd->vdev_child[c];
2558 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
2559 			return (cvd);
2560 	}
2561 	return (NULL);
2562 }
2563 
2564 /*
2565  * Verify that dynamic LUN growth works as expected.
2566  */
2567 /* ARGSUSED */
2568 void
2569 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
2570 {
2571 	ztest_shared_t *zs = ztest_shared;
2572 	spa_t *spa = zs->zs_spa;
2573 	vdev_t *vd, *tvd;
2574 	metaslab_class_t *mc;
2575 	metaslab_group_t *mg;
2576 	size_t psize, newsize;
2577 	uint64_t top;
2578 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
2579 
2580 	VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2581 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2582 
2583 	top = ztest_random_vdev_top(spa, B_TRUE);
2584 
2585 	tvd = spa->spa_root_vdev->vdev_child[top];
2586 	mg = tvd->vdev_mg;
2587 	mc = mg->mg_class;
2588 	old_ms_count = tvd->vdev_ms_count;
2589 	old_class_space = metaslab_class_get_space(mc);
2590 
2591 	/*
2592 	 * Determine the size of the first leaf vdev associated with
2593 	 * our top-level device.
2594 	 */
2595 	vd = vdev_walk_tree(tvd, NULL, NULL);
2596 	ASSERT3P(vd, !=, NULL);
2597 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2598 
2599 	psize = vd->vdev_psize;
2600 
2601 	/*
2602 	 * We only try to expand the vdev if it's healthy, less than 4x its
2603 	 * original size, and it has a valid psize.
2604 	 */
2605 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
2606 	    psize == 0 || psize >= 4 * zopt_vdev_size) {
2607 		spa_config_exit(spa, SCL_STATE, spa);
2608 		VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2609 		return;
2610 	}
2611 	ASSERT(psize > 0);
2612 	newsize = psize + psize / 8;
2613 	ASSERT3U(newsize, >, psize);
2614 
2615 	if (zopt_verbose >= 6) {
2616 		(void) printf("Expanding LUN %s from %lu to %lu\n",
2617 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
2618 	}
2619 
2620 	/*
2621 	 * Growing the vdev is a two step process:
2622 	 *	1). expand the physical size (i.e. relabel)
2623 	 *	2). online the vdev to create the new metaslabs
2624 	 */
2625 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
2626 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
2627 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
2628 		if (zopt_verbose >= 5) {
2629 			(void) printf("Could not expand LUN because "
2630 			    "the vdev configuration changed.\n");
2631 		}
2632 		spa_config_exit(spa, SCL_STATE, spa);
2633 		VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2634 		return;
2635 	}
2636 
2637 	spa_config_exit(spa, SCL_STATE, spa);
2638 
2639 	/*
2640 	 * Expanding the LUN will update the config asynchronously,
2641 	 * thus we must wait for the async thread to complete any
2642 	 * pending tasks before proceeding.
2643 	 */
2644 	for (;;) {
2645 		boolean_t done;
2646 		mutex_enter(&spa->spa_async_lock);
2647 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
2648 		mutex_exit(&spa->spa_async_lock);
2649 		if (done)
2650 			break;
2651 		txg_wait_synced(spa_get_dsl(spa), 0);
2652 		(void) poll(NULL, 0, 100);
2653 	}
2654 
2655 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2656 
2657 	tvd = spa->spa_root_vdev->vdev_child[top];
2658 	new_ms_count = tvd->vdev_ms_count;
2659 	new_class_space = metaslab_class_get_space(mc);
2660 
2661 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
2662 		if (zopt_verbose >= 5) {
2663 			(void) printf("Could not verify LUN expansion due to "
2664 			    "intervening vdev offline or remove.\n");
2665 		}
2666 		spa_config_exit(spa, SCL_STATE, spa);
2667 		VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2668 		return;
2669 	}
2670 
2671 	/*
2672 	 * Make sure we were able to grow the vdev.
2673 	 */
2674 	if (new_ms_count <= old_ms_count)
2675 		fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
2676 		    old_ms_count, new_ms_count);
2677 
2678 	/*
2679 	 * Make sure we were able to grow the pool.
2680 	 */
2681 	if (new_class_space <= old_class_space)
2682 		fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
2683 		    old_class_space, new_class_space);
2684 
2685 	if (zopt_verbose >= 5) {
2686 		char oldnumbuf[6], newnumbuf[6];
2687 
2688 		nicenum(old_class_space, oldnumbuf);
2689 		nicenum(new_class_space, newnumbuf);
2690 		(void) printf("%s grew from %s to %s\n",
2691 		    spa->spa_name, oldnumbuf, newnumbuf);
2692 	}
2693 
2694 	spa_config_exit(spa, SCL_STATE, spa);
2695 	VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2696 }
2697 
2698 /*
2699  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
2700  */
2701 /* ARGSUSED */
2702 static void
2703 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
2704 {
2705 	/*
2706 	 * Create the objects common to all ztest datasets.
2707 	 */
2708 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
2709 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
2710 }
2711 
2712 /* ARGSUSED */
2713 static int
2714 ztest_objset_destroy_cb(char *name, void *arg)
2715 {
2716 	objset_t *os;
2717 	dmu_object_info_t doi;
2718 	int error;
2719 
2720 	/*
2721 	 * Verify that the dataset contains a directory object.
2722 	 */
2723 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os));
2724 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
2725 	if (error != ENOENT) {
2726 		/* We could have crashed in the middle of destroying it */
2727 		ASSERT3U(error, ==, 0);
2728 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
2729 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
2730 	}
2731 	dmu_objset_rele(os, FTAG);
2732 
2733 	/*
2734 	 * Destroy the dataset.
2735 	 */
2736 	VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE));
2737 	return (0);
2738 }
2739 
2740 static boolean_t
2741 ztest_snapshot_create(char *osname, uint64_t id)
2742 {
2743 	char snapname[MAXNAMELEN];
2744 	int error;
2745 
2746 	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
2747 	    (u_longlong_t)id);
2748 
2749 	error = dmu_objset_snapshot(osname, strchr(snapname, '@') + 1,
2750 	    NULL, B_FALSE);
2751 	if (error == ENOSPC) {
2752 		ztest_record_enospc(FTAG);
2753 		return (B_FALSE);
2754 	}
2755 	if (error != 0 && error != EEXIST)
2756 		fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error);
2757 	return (B_TRUE);
2758 }
2759 
2760 static boolean_t
2761 ztest_snapshot_destroy(char *osname, uint64_t id)
2762 {
2763 	char snapname[MAXNAMELEN];
2764 	int error;
2765 
2766 	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
2767 	    (u_longlong_t)id);
2768 
2769 	error = dmu_objset_destroy(snapname, B_FALSE);
2770 	if (error != 0 && error != ENOENT)
2771 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
2772 	return (B_TRUE);
2773 }
2774 
2775 /* ARGSUSED */
2776 void
2777 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
2778 {
2779 	ztest_shared_t *zs = ztest_shared;
2780 	ztest_ds_t zdtmp;
2781 	int iters;
2782 	int error;
2783 	objset_t *os, *os2;
2784 	char name[MAXNAMELEN];
2785 	zilog_t *zilog;
2786 
2787 	(void) rw_rdlock(&zs->zs_name_lock);
2788 
2789 	(void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
2790 	    zs->zs_pool, (u_longlong_t)id);
2791 
2792 	/*
2793 	 * If this dataset exists from a previous run, process its replay log
2794 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
2795 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
2796 	 */
2797 	if (ztest_random(2) == 0 &&
2798 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
2799 		ztest_zd_init(&zdtmp, os);
2800 		zil_replay(os, &zdtmp, ztest_replay_vector);
2801 		ztest_zd_fini(&zdtmp);
2802 		dmu_objset_disown(os, FTAG);
2803 	}
2804 
2805 	/*
2806 	 * There may be an old instance of the dataset we're about to
2807 	 * create lying around from a previous run.  If so, destroy it
2808 	 * and all of its snapshots.
2809 	 */
2810 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
2811 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2812 
2813 	/*
2814 	 * Verify that the destroyed dataset is no longer in the namespace.
2815 	 */
2816 	VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os));
2817 
2818 	/*
2819 	 * Verify that we can create a new dataset.
2820 	 */
2821 	error = dmu_objset_create(name, DMU_OST_OTHER, 0,
2822 	    ztest_objset_create_cb, NULL);
2823 	if (error) {
2824 		if (error == ENOSPC) {
2825 			ztest_record_enospc(FTAG);
2826 			(void) rw_unlock(&zs->zs_name_lock);
2827 			return;
2828 		}
2829 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
2830 	}
2831 
2832 	VERIFY3U(0, ==,
2833 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
2834 
2835 	ztest_zd_init(&zdtmp, os);
2836 
2837 	/*
2838 	 * Open the intent log for it.
2839 	 */
2840 	zilog = zil_open(os, ztest_get_data);
2841 
2842 	/*
2843 	 * Put some objects in there, do a little I/O to them,
2844 	 * and randomly take a couple of snapshots along the way.
2845 	 */
2846 	iters = ztest_random(5);
2847 	for (int i = 0; i < iters; i++) {
2848 		ztest_dmu_object_alloc_free(&zdtmp, id);
2849 		if (ztest_random(iters) == 0)
2850 			(void) ztest_snapshot_create(name, i);
2851 	}
2852 
2853 	/*
2854 	 * Verify that we cannot create an existing dataset.
2855 	 */
2856 	VERIFY3U(EEXIST, ==,
2857 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
2858 
2859 	/*
2860 	 * Verify that we can hold an objset that is also owned.
2861 	 */
2862 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
2863 	dmu_objset_rele(os2, FTAG);
2864 
2865 	/*
2866 	 * Verify that we cannot own an objset that is already owned.
2867 	 */
2868 	VERIFY3U(EBUSY, ==,
2869 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
2870 
2871 	zil_close(zilog);
2872 	dmu_objset_disown(os, FTAG);
2873 	ztest_zd_fini(&zdtmp);
2874 
2875 	(void) rw_unlock(&zs->zs_name_lock);
2876 }
2877 
2878 /*
2879  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
2880  */
2881 void
2882 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
2883 {
2884 	ztest_shared_t *zs = ztest_shared;
2885 
2886 	(void) rw_rdlock(&zs->zs_name_lock);
2887 	(void) ztest_snapshot_destroy(zd->zd_name, id);
2888 	(void) ztest_snapshot_create(zd->zd_name, id);
2889 	(void) rw_unlock(&zs->zs_name_lock);
2890 }
2891 
2892 /*
2893  * Cleanup non-standard snapshots and clones.
2894  */
2895 void
2896 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
2897 {
2898 	char snap1name[MAXNAMELEN];
2899 	char clone1name[MAXNAMELEN];
2900 	char snap2name[MAXNAMELEN];
2901 	char clone2name[MAXNAMELEN];
2902 	char snap3name[MAXNAMELEN];
2903 	int error;
2904 
2905 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
2906 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
2907 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
2908 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
2909 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
2910 
2911 	error = dmu_objset_destroy(clone2name, B_FALSE);
2912 	if (error && error != ENOENT)
2913 		fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
2914 	error = dmu_objset_destroy(snap3name, B_FALSE);
2915 	if (error && error != ENOENT)
2916 		fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
2917 	error = dmu_objset_destroy(snap2name, B_FALSE);
2918 	if (error && error != ENOENT)
2919 		fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
2920 	error = dmu_objset_destroy(clone1name, B_FALSE);
2921 	if (error && error != ENOENT)
2922 		fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
2923 	error = dmu_objset_destroy(snap1name, B_FALSE);
2924 	if (error && error != ENOENT)
2925 		fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
2926 }
2927 
2928 /*
2929  * Verify dsl_dataset_promote handles EBUSY
2930  */
2931 void
2932 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
2933 {
2934 	ztest_shared_t *zs = ztest_shared;
2935 	objset_t *clone;
2936 	dsl_dataset_t *ds;
2937 	char snap1name[MAXNAMELEN];
2938 	char clone1name[MAXNAMELEN];
2939 	char snap2name[MAXNAMELEN];
2940 	char clone2name[MAXNAMELEN];
2941 	char snap3name[MAXNAMELEN];
2942 	char *osname = zd->zd_name;
2943 	int error;
2944 
2945 	(void) rw_rdlock(&zs->zs_name_lock);
2946 
2947 	ztest_dsl_dataset_cleanup(osname, id);
2948 
2949 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
2950 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
2951 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
2952 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
2953 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
2954 
2955 	error = dmu_objset_snapshot(osname, strchr(snap1name, '@')+1,
2956 	    NULL, B_FALSE);
2957 	if (error && error != EEXIST) {
2958 		if (error == ENOSPC) {
2959 			ztest_record_enospc(FTAG);
2960 			goto out;
2961 		}
2962 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
2963 	}
2964 
2965 	error = dmu_objset_hold(snap1name, FTAG, &clone);
2966 	if (error)
2967 		fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error);
2968 
2969 	error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0);
2970 	dmu_objset_rele(clone, FTAG);
2971 	if (error) {
2972 		if (error == ENOSPC) {
2973 			ztest_record_enospc(FTAG);
2974 			goto out;
2975 		}
2976 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
2977 	}
2978 
2979 	error = dmu_objset_snapshot(clone1name, strchr(snap2name, '@')+1,
2980 	    NULL, B_FALSE);
2981 	if (error && error != EEXIST) {
2982 		if (error == ENOSPC) {
2983 			ztest_record_enospc(FTAG);
2984 			goto out;
2985 		}
2986 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
2987 	}
2988 
2989 	error = dmu_objset_snapshot(clone1name, strchr(snap3name, '@')+1,
2990 	    NULL, B_FALSE);
2991 	if (error && error != EEXIST) {
2992 		if (error == ENOSPC) {
2993 			ztest_record_enospc(FTAG);
2994 			goto out;
2995 		}
2996 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
2997 	}
2998 
2999 	error = dmu_objset_hold(snap3name, FTAG, &clone);
3000 	if (error)
3001 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3002 
3003 	error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0);
3004 	dmu_objset_rele(clone, FTAG);
3005 	if (error) {
3006 		if (error == ENOSPC) {
3007 			ztest_record_enospc(FTAG);
3008 			goto out;
3009 		}
3010 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3011 	}
3012 
3013 	error = dsl_dataset_own(snap1name, B_FALSE, FTAG, &ds);
3014 	if (error)
3015 		fatal(0, "dsl_dataset_own(%s) = %d", snap1name, error);
3016 	error = dsl_dataset_promote(clone2name, NULL);
3017 	if (error != EBUSY)
3018 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3019 		    error);
3020 	dsl_dataset_disown(ds, FTAG);
3021 
3022 out:
3023 	ztest_dsl_dataset_cleanup(osname, id);
3024 
3025 	(void) rw_unlock(&zs->zs_name_lock);
3026 }
3027 
3028 /*
3029  * Verify that dmu_object_{alloc,free} work as expected.
3030  */
3031 void
3032 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3033 {
3034 	ztest_od_t od[4];
3035 	int batchsize = sizeof (od) / sizeof (od[0]);
3036 
3037 	for (int b = 0; b < batchsize; b++)
3038 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3039 
3040 	/*
3041 	 * Destroy the previous batch of objects, create a new batch,
3042 	 * and do some I/O on the new objects.
3043 	 */
3044 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3045 		return;
3046 
3047 	while (ztest_random(4 * batchsize) != 0)
3048 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3049 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3050 }
3051 
3052 /*
3053  * Verify that dmu_{read,write} work as expected.
3054  */
3055 void
3056 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3057 {
3058 	objset_t *os = zd->zd_os;
3059 	ztest_od_t od[2];
3060 	dmu_tx_t *tx;
3061 	int i, freeit, error;
3062 	uint64_t n, s, txg;
3063 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3064 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3065 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3066 	uint64_t regions = 997;
3067 	uint64_t stride = 123456789ULL;
3068 	uint64_t width = 40;
3069 	int free_percent = 5;
3070 
3071 	/*
3072 	 * This test uses two objects, packobj and bigobj, that are always
3073 	 * updated together (i.e. in the same tx) so that their contents are
3074 	 * in sync and can be compared.  Their contents relate to each other
3075 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3076 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3077 	 * for any index n, there are three bufwads that should be identical:
3078 	 *
3079 	 *	packobj, at offset n * sizeof (bufwad_t)
3080 	 *	bigobj, at the head of the nth chunk
3081 	 *	bigobj, at the tail of the nth chunk
3082 	 *
3083 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3084 	 * and it doesn't have any relation to the object blocksize.
3085 	 * The only requirement is that it can hold at least two bufwads.
3086 	 *
3087 	 * Normally, we write the bufwad to each of these locations.
3088 	 * However, free_percent of the time we instead write zeroes to
3089 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3090 	 * bigobj to packobj, we can verify that the DMU is correctly
3091 	 * tracking which parts of an object are allocated and free,
3092 	 * and that the contents of the allocated blocks are correct.
3093 	 */
3094 
3095 	/*
3096 	 * Read the directory info.  If it's the first time, set things up.
3097 	 */
3098 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3099 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3100 
3101 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3102 		return;
3103 
3104 	bigobj = od[0].od_object;
3105 	packobj = od[1].od_object;
3106 	chunksize = od[0].od_gen;
3107 	ASSERT(chunksize == od[1].od_gen);
3108 
3109 	/*
3110 	 * Prefetch a random chunk of the big object.
3111 	 * Our aim here is to get some async reads in flight
3112 	 * for blocks that we may free below; the DMU should
3113 	 * handle this race correctly.
3114 	 */
3115 	n = ztest_random(regions) * stride + ztest_random(width);
3116 	s = 1 + ztest_random(2 * width - 1);
3117 	dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3118 
3119 	/*
3120 	 * Pick a random index and compute the offsets into packobj and bigobj.
3121 	 */
3122 	n = ztest_random(regions) * stride + ztest_random(width);
3123 	s = 1 + ztest_random(width - 1);
3124 
3125 	packoff = n * sizeof (bufwad_t);
3126 	packsize = s * sizeof (bufwad_t);
3127 
3128 	bigoff = n * chunksize;
3129 	bigsize = s * chunksize;
3130 
3131 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3132 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3133 
3134 	/*
3135 	 * free_percent of the time, free a range of bigobj rather than
3136 	 * overwriting it.
3137 	 */
3138 	freeit = (ztest_random(100) < free_percent);
3139 
3140 	/*
3141 	 * Read the current contents of our objects.
3142 	 */
3143 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3144 	    DMU_READ_PREFETCH);
3145 	ASSERT3U(error, ==, 0);
3146 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3147 	    DMU_READ_PREFETCH);
3148 	ASSERT3U(error, ==, 0);
3149 
3150 	/*
3151 	 * Get a tx for the mods to both packobj and bigobj.
3152 	 */
3153 	tx = dmu_tx_create(os);
3154 
3155 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3156 
3157 	if (freeit)
3158 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3159 	else
3160 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3161 
3162 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3163 	if (txg == 0) {
3164 		umem_free(packbuf, packsize);
3165 		umem_free(bigbuf, bigsize);
3166 		return;
3167 	}
3168 
3169 	dmu_object_set_checksum(os, bigobj,
3170 	    (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3171 
3172 	dmu_object_set_compress(os, bigobj,
3173 	    (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3174 
3175 	/*
3176 	 * For each index from n to n + s, verify that the existing bufwad
3177 	 * in packobj matches the bufwads at the head and tail of the
3178 	 * corresponding chunk in bigobj.  Then update all three bufwads
3179 	 * with the new values we want to write out.
3180 	 */
3181 	for (i = 0; i < s; i++) {
3182 		/* LINTED */
3183 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3184 		/* LINTED */
3185 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3186 		/* LINTED */
3187 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3188 
3189 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3190 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3191 
3192 		if (pack->bw_txg > txg)
3193 			fatal(0, "future leak: got %llx, open txg is %llx",
3194 			    pack->bw_txg, txg);
3195 
3196 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3197 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3198 			    pack->bw_index, n, i);
3199 
3200 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3201 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3202 
3203 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3204 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3205 
3206 		if (freeit) {
3207 			bzero(pack, sizeof (bufwad_t));
3208 		} else {
3209 			pack->bw_index = n + i;
3210 			pack->bw_txg = txg;
3211 			pack->bw_data = 1 + ztest_random(-2ULL);
3212 		}
3213 		*bigH = *pack;
3214 		*bigT = *pack;
3215 	}
3216 
3217 	/*
3218 	 * We've verified all the old bufwads, and made new ones.
3219 	 * Now write them out.
3220 	 */
3221 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3222 
3223 	if (freeit) {
3224 		if (zopt_verbose >= 7) {
3225 			(void) printf("freeing offset %llx size %llx"
3226 			    " txg %llx\n",
3227 			    (u_longlong_t)bigoff,
3228 			    (u_longlong_t)bigsize,
3229 			    (u_longlong_t)txg);
3230 		}
3231 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3232 	} else {
3233 		if (zopt_verbose >= 7) {
3234 			(void) printf("writing offset %llx size %llx"
3235 			    " txg %llx\n",
3236 			    (u_longlong_t)bigoff,
3237 			    (u_longlong_t)bigsize,
3238 			    (u_longlong_t)txg);
3239 		}
3240 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3241 	}
3242 
3243 	dmu_tx_commit(tx);
3244 
3245 	/*
3246 	 * Sanity check the stuff we just wrote.
3247 	 */
3248 	{
3249 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3250 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3251 
3252 		VERIFY(0 == dmu_read(os, packobj, packoff,
3253 		    packsize, packcheck, DMU_READ_PREFETCH));
3254 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3255 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3256 
3257 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3258 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3259 
3260 		umem_free(packcheck, packsize);
3261 		umem_free(bigcheck, bigsize);
3262 	}
3263 
3264 	umem_free(packbuf, packsize);
3265 	umem_free(bigbuf, bigsize);
3266 }
3267 
3268 void
3269 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3270     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3271 {
3272 	uint64_t i;
3273 	bufwad_t *pack;
3274 	bufwad_t *bigH;
3275 	bufwad_t *bigT;
3276 
3277 	/*
3278 	 * For each index from n to n + s, verify that the existing bufwad
3279 	 * in packobj matches the bufwads at the head and tail of the
3280 	 * corresponding chunk in bigobj.  Then update all three bufwads
3281 	 * with the new values we want to write out.
3282 	 */
3283 	for (i = 0; i < s; i++) {
3284 		/* LINTED */
3285 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3286 		/* LINTED */
3287 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3288 		/* LINTED */
3289 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3290 
3291 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3292 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3293 
3294 		if (pack->bw_txg > txg)
3295 			fatal(0, "future leak: got %llx, open txg is %llx",
3296 			    pack->bw_txg, txg);
3297 
3298 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3299 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3300 			    pack->bw_index, n, i);
3301 
3302 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3303 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3304 
3305 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3306 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3307 
3308 		pack->bw_index = n + i;
3309 		pack->bw_txg = txg;
3310 		pack->bw_data = 1 + ztest_random(-2ULL);
3311 
3312 		*bigH = *pack;
3313 		*bigT = *pack;
3314 	}
3315 }
3316 
3317 void
3318 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3319 {
3320 	objset_t *os = zd->zd_os;
3321 	ztest_od_t od[2];
3322 	dmu_tx_t *tx;
3323 	uint64_t i;
3324 	int error;
3325 	uint64_t n, s, txg;
3326 	bufwad_t *packbuf, *bigbuf;
3327 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3328 	uint64_t blocksize = ztest_random_blocksize();
3329 	uint64_t chunksize = blocksize;
3330 	uint64_t regions = 997;
3331 	uint64_t stride = 123456789ULL;
3332 	uint64_t width = 9;
3333 	dmu_buf_t *bonus_db;
3334 	arc_buf_t **bigbuf_arcbufs;
3335 	dmu_object_info_t doi;
3336 
3337 	/*
3338 	 * This test uses two objects, packobj and bigobj, that are always
3339 	 * updated together (i.e. in the same tx) so that their contents are
3340 	 * in sync and can be compared.  Their contents relate to each other
3341 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3342 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3343 	 * for any index n, there are three bufwads that should be identical:
3344 	 *
3345 	 *	packobj, at offset n * sizeof (bufwad_t)
3346 	 *	bigobj, at the head of the nth chunk
3347 	 *	bigobj, at the tail of the nth chunk
3348 	 *
3349 	 * The chunk size is set equal to bigobj block size so that
3350 	 * dmu_assign_arcbuf() can be tested for object updates.
3351 	 */
3352 
3353 	/*
3354 	 * Read the directory info.  If it's the first time, set things up.
3355 	 */
3356 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3357 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3358 
3359 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3360 		return;
3361 
3362 	bigobj = od[0].od_object;
3363 	packobj = od[1].od_object;
3364 	blocksize = od[0].od_blocksize;
3365 	chunksize = blocksize;
3366 	ASSERT(chunksize == od[1].od_gen);
3367 
3368 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3369 	VERIFY(ISP2(doi.doi_data_block_size));
3370 	VERIFY(chunksize == doi.doi_data_block_size);
3371 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3372 
3373 	/*
3374 	 * Pick a random index and compute the offsets into packobj and bigobj.
3375 	 */
3376 	n = ztest_random(regions) * stride + ztest_random(width);
3377 	s = 1 + ztest_random(width - 1);
3378 
3379 	packoff = n * sizeof (bufwad_t);
3380 	packsize = s * sizeof (bufwad_t);
3381 
3382 	bigoff = n * chunksize;
3383 	bigsize = s * chunksize;
3384 
3385 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3386 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3387 
3388 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3389 
3390 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3391 
3392 	/*
3393 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3394 	 * Iteration 1 test zcopy to already referenced dbufs.
3395 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3396 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3397 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
3398 	 * Iteration 5 test zcopy when it can't be done.
3399 	 * Iteration 6 one more zcopy write.
3400 	 */
3401 	for (i = 0; i < 7; i++) {
3402 		uint64_t j;
3403 		uint64_t off;
3404 
3405 		/*
3406 		 * In iteration 5 (i == 5) use arcbufs
3407 		 * that don't match bigobj blksz to test
3408 		 * dmu_assign_arcbuf() when it can't directly
3409 		 * assign an arcbuf to a dbuf.
3410 		 */
3411 		for (j = 0; j < s; j++) {
3412 			if (i != 5) {
3413 				bigbuf_arcbufs[j] =
3414 				    dmu_request_arcbuf(bonus_db, chunksize);
3415 			} else {
3416 				bigbuf_arcbufs[2 * j] =
3417 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3418 				bigbuf_arcbufs[2 * j + 1] =
3419 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3420 			}
3421 		}
3422 
3423 		/*
3424 		 * Get a tx for the mods to both packobj and bigobj.
3425 		 */
3426 		tx = dmu_tx_create(os);
3427 
3428 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
3429 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3430 
3431 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3432 		if (txg == 0) {
3433 			umem_free(packbuf, packsize);
3434 			umem_free(bigbuf, bigsize);
3435 			for (j = 0; j < s; j++) {
3436 				if (i != 5) {
3437 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
3438 				} else {
3439 					dmu_return_arcbuf(
3440 					    bigbuf_arcbufs[2 * j]);
3441 					dmu_return_arcbuf(
3442 					    bigbuf_arcbufs[2 * j + 1]);
3443 				}
3444 			}
3445 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3446 			dmu_buf_rele(bonus_db, FTAG);
3447 			return;
3448 		}
3449 
3450 		/*
3451 		 * 50% of the time don't read objects in the 1st iteration to
3452 		 * test dmu_assign_arcbuf() for the case when there're no
3453 		 * existing dbufs for the specified offsets.
3454 		 */
3455 		if (i != 0 || ztest_random(2) != 0) {
3456 			error = dmu_read(os, packobj, packoff,
3457 			    packsize, packbuf, DMU_READ_PREFETCH);
3458 			ASSERT3U(error, ==, 0);
3459 			error = dmu_read(os, bigobj, bigoff, bigsize,
3460 			    bigbuf, DMU_READ_PREFETCH);
3461 			ASSERT3U(error, ==, 0);
3462 		}
3463 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3464 		    n, chunksize, txg);
3465 
3466 		/*
3467 		 * We've verified all the old bufwads, and made new ones.
3468 		 * Now write them out.
3469 		 */
3470 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3471 		if (zopt_verbose >= 7) {
3472 			(void) printf("writing offset %llx size %llx"
3473 			    " txg %llx\n",
3474 			    (u_longlong_t)bigoff,
3475 			    (u_longlong_t)bigsize,
3476 			    (u_longlong_t)txg);
3477 		}
3478 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3479 			dmu_buf_t *dbt;
3480 			if (i != 5) {
3481 				bcopy((caddr_t)bigbuf + (off - bigoff),
3482 				    bigbuf_arcbufs[j]->b_data, chunksize);
3483 			} else {
3484 				bcopy((caddr_t)bigbuf + (off - bigoff),
3485 				    bigbuf_arcbufs[2 * j]->b_data,
3486 				    chunksize / 2);
3487 				bcopy((caddr_t)bigbuf + (off - bigoff) +
3488 				    chunksize / 2,
3489 				    bigbuf_arcbufs[2 * j + 1]->b_data,
3490 				    chunksize / 2);
3491 			}
3492 
3493 			if (i == 1) {
3494 				VERIFY(dmu_buf_hold(os, bigobj, off,
3495 				    FTAG, &dbt) == 0);
3496 			}
3497 			if (i != 5) {
3498 				dmu_assign_arcbuf(bonus_db, off,
3499 				    bigbuf_arcbufs[j], tx);
3500 			} else {
3501 				dmu_assign_arcbuf(bonus_db, off,
3502 				    bigbuf_arcbufs[2 * j], tx);
3503 				dmu_assign_arcbuf(bonus_db,
3504 				    off + chunksize / 2,
3505 				    bigbuf_arcbufs[2 * j + 1], tx);
3506 			}
3507 			if (i == 1) {
3508 				dmu_buf_rele(dbt, FTAG);
3509 			}
3510 		}
3511 		dmu_tx_commit(tx);
3512 
3513 		/*
3514 		 * Sanity check the stuff we just wrote.
3515 		 */
3516 		{
3517 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3518 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3519 
3520 			VERIFY(0 == dmu_read(os, packobj, packoff,
3521 			    packsize, packcheck, DMU_READ_PREFETCH));
3522 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
3523 			    bigsize, bigcheck, DMU_READ_PREFETCH));
3524 
3525 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3526 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3527 
3528 			umem_free(packcheck, packsize);
3529 			umem_free(bigcheck, bigsize);
3530 		}
3531 		if (i == 2) {
3532 			txg_wait_open(dmu_objset_pool(os), 0);
3533 		} else if (i == 3) {
3534 			txg_wait_synced(dmu_objset_pool(os), 0);
3535 		}
3536 	}
3537 
3538 	dmu_buf_rele(bonus_db, FTAG);
3539 	umem_free(packbuf, packsize);
3540 	umem_free(bigbuf, bigsize);
3541 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3542 }
3543 
3544 /* ARGSUSED */
3545 void
3546 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3547 {
3548 	ztest_od_t od[1];
3549 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3550 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3551 
3552 	/*
3553 	 * Have multiple threads write to large offsets in an object
3554 	 * to verify that parallel writes to an object -- even to the
3555 	 * same blocks within the object -- doesn't cause any trouble.
3556 	 */
3557 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
3558 
3559 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3560 		return;
3561 
3562 	while (ztest_random(10) != 0)
3563 		ztest_io(zd, od[0].od_object, offset);
3564 }
3565 
3566 void
3567 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
3568 {
3569 	ztest_od_t od[1];
3570 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
3571 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3572 	uint64_t count = ztest_random(20) + 1;
3573 	uint64_t blocksize = ztest_random_blocksize();
3574 	void *data;
3575 
3576 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3577 
3578 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3579 		return;
3580 
3581 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
3582 		return;
3583 
3584 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
3585 
3586 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
3587 
3588 	while (ztest_random(count) != 0) {
3589 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
3590 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
3591 		    data) != 0)
3592 			break;
3593 		while (ztest_random(4) != 0)
3594 			ztest_io(zd, od[0].od_object, randoff);
3595 	}
3596 
3597 	umem_free(data, blocksize);
3598 }
3599 
3600 /*
3601  * Verify that zap_{create,destroy,add,remove,update} work as expected.
3602  */
3603 #define	ZTEST_ZAP_MIN_INTS	1
3604 #define	ZTEST_ZAP_MAX_INTS	4
3605 #define	ZTEST_ZAP_MAX_PROPS	1000
3606 
3607 void
3608 ztest_zap(ztest_ds_t *zd, uint64_t id)
3609 {
3610 	objset_t *os = zd->zd_os;
3611 	ztest_od_t od[1];
3612 	uint64_t object;
3613 	uint64_t txg, last_txg;
3614 	uint64_t value[ZTEST_ZAP_MAX_INTS];
3615 	uint64_t zl_ints, zl_intsize, prop;
3616 	int i, ints;
3617 	dmu_tx_t *tx;
3618 	char propname[100], txgname[100];
3619 	int error;
3620 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
3621 
3622 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
3623 
3624 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3625 		return;
3626 
3627 	object = od[0].od_object;
3628 
3629 	/*
3630 	 * Generate a known hash collision, and verify that
3631 	 * we can lookup and remove both entries.
3632 	 */
3633 	tx = dmu_tx_create(os);
3634 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3635 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3636 	if (txg == 0)
3637 		return;
3638 	for (i = 0; i < 2; i++) {
3639 		value[i] = i;
3640 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
3641 		    1, &value[i], tx));
3642 	}
3643 	for (i = 0; i < 2; i++) {
3644 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
3645 		    sizeof (uint64_t), 1, &value[i], tx));
3646 		VERIFY3U(0, ==,
3647 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
3648 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3649 		ASSERT3U(zl_ints, ==, 1);
3650 	}
3651 	for (i = 0; i < 2; i++) {
3652 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
3653 	}
3654 	dmu_tx_commit(tx);
3655 
3656 	/*
3657 	 * Generate a buch of random entries.
3658 	 */
3659 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
3660 
3661 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
3662 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
3663 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
3664 	bzero(value, sizeof (value));
3665 	last_txg = 0;
3666 
3667 	/*
3668 	 * If these zap entries already exist, validate their contents.
3669 	 */
3670 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
3671 	if (error == 0) {
3672 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3673 		ASSERT3U(zl_ints, ==, 1);
3674 
3675 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
3676 		    zl_ints, &last_txg) == 0);
3677 
3678 		VERIFY(zap_length(os, object, propname, &zl_intsize,
3679 		    &zl_ints) == 0);
3680 
3681 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3682 		ASSERT3U(zl_ints, ==, ints);
3683 
3684 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
3685 		    zl_ints, value) == 0);
3686 
3687 		for (i = 0; i < ints; i++) {
3688 			ASSERT3U(value[i], ==, last_txg + object + i);
3689 		}
3690 	} else {
3691 		ASSERT3U(error, ==, ENOENT);
3692 	}
3693 
3694 	/*
3695 	 * Atomically update two entries in our zap object.
3696 	 * The first is named txg_%llu, and contains the txg
3697 	 * in which the property was last updated.  The second
3698 	 * is named prop_%llu, and the nth element of its value
3699 	 * should be txg + object + n.
3700 	 */
3701 	tx = dmu_tx_create(os);
3702 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3703 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3704 	if (txg == 0)
3705 		return;
3706 
3707 	if (last_txg > txg)
3708 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
3709 
3710 	for (i = 0; i < ints; i++)
3711 		value[i] = txg + object + i;
3712 
3713 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
3714 	    1, &txg, tx));
3715 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
3716 	    ints, value, tx));
3717 
3718 	dmu_tx_commit(tx);
3719 
3720 	/*
3721 	 * Remove a random pair of entries.
3722 	 */
3723 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
3724 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
3725 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
3726 
3727 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
3728 
3729 	if (error == ENOENT)
3730 		return;
3731 
3732 	ASSERT3U(error, ==, 0);
3733 
3734 	tx = dmu_tx_create(os);
3735 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3736 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3737 	if (txg == 0)
3738 		return;
3739 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
3740 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
3741 	dmu_tx_commit(tx);
3742 }
3743 
3744 /*
3745  * Testcase to test the upgrading of a microzap to fatzap.
3746  */
3747 void
3748 ztest_fzap(ztest_ds_t *zd, uint64_t id)
3749 {
3750 	objset_t *os = zd->zd_os;
3751 	ztest_od_t od[1];
3752 	uint64_t object, txg;
3753 
3754 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
3755 
3756 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3757 		return;
3758 
3759 	object = od[0].od_object;
3760 
3761 	/*
3762 	 * Add entries to this ZAP and make sure it spills over
3763 	 * and gets upgraded to a fatzap. Also, since we are adding
3764 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
3765 	 */
3766 	for (int i = 0; i < 2050; i++) {
3767 		char name[MAXNAMELEN];
3768 		uint64_t value = i;
3769 		dmu_tx_t *tx;
3770 		int error;
3771 
3772 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
3773 		    id, value);
3774 
3775 		tx = dmu_tx_create(os);
3776 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
3777 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3778 		if (txg == 0)
3779 			return;
3780 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
3781 		    &value, tx);
3782 		ASSERT(error == 0 || error == EEXIST);
3783 		dmu_tx_commit(tx);
3784 	}
3785 }
3786 
3787 /* ARGSUSED */
3788 void
3789 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
3790 {
3791 	objset_t *os = zd->zd_os;
3792 	ztest_od_t od[1];
3793 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
3794 	dmu_tx_t *tx;
3795 	int i, namelen, error;
3796 	int micro = ztest_random(2);
3797 	char name[20], string_value[20];
3798 	void *data;
3799 
3800 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
3801 
3802 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3803 		return;
3804 
3805 	object = od[0].od_object;
3806 
3807 	/*
3808 	 * Generate a random name of the form 'xxx.....' where each
3809 	 * x is a random printable character and the dots are dots.
3810 	 * There are 94 such characters, and the name length goes from
3811 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
3812 	 */
3813 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
3814 
3815 	for (i = 0; i < 3; i++)
3816 		name[i] = '!' + ztest_random('~' - '!' + 1);
3817 	for (; i < namelen - 1; i++)
3818 		name[i] = '.';
3819 	name[i] = '\0';
3820 
3821 	if ((namelen & 1) || micro) {
3822 		wsize = sizeof (txg);
3823 		wc = 1;
3824 		data = &txg;
3825 	} else {
3826 		wsize = 1;
3827 		wc = namelen;
3828 		data = string_value;
3829 	}
3830 
3831 	count = -1ULL;
3832 	VERIFY(zap_count(os, object, &count) == 0);
3833 	ASSERT(count != -1ULL);
3834 
3835 	/*
3836 	 * Select an operation: length, lookup, add, update, remove.
3837 	 */
3838 	i = ztest_random(5);
3839 
3840 	if (i >= 2) {
3841 		tx = dmu_tx_create(os);
3842 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3843 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3844 		if (txg == 0)
3845 			return;
3846 		bcopy(name, string_value, namelen);
3847 	} else {
3848 		tx = NULL;
3849 		txg = 0;
3850 		bzero(string_value, namelen);
3851 	}
3852 
3853 	switch (i) {
3854 
3855 	case 0:
3856 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
3857 		if (error == 0) {
3858 			ASSERT3U(wsize, ==, zl_wsize);
3859 			ASSERT3U(wc, ==, zl_wc);
3860 		} else {
3861 			ASSERT3U(error, ==, ENOENT);
3862 		}
3863 		break;
3864 
3865 	case 1:
3866 		error = zap_lookup(os, object, name, wsize, wc, data);
3867 		if (error == 0) {
3868 			if (data == string_value &&
3869 			    bcmp(name, data, namelen) != 0)
3870 				fatal(0, "name '%s' != val '%s' len %d",
3871 				    name, data, namelen);
3872 		} else {
3873 			ASSERT3U(error, ==, ENOENT);
3874 		}
3875 		break;
3876 
3877 	case 2:
3878 		error = zap_add(os, object, name, wsize, wc, data, tx);
3879 		ASSERT(error == 0 || error == EEXIST);
3880 		break;
3881 
3882 	case 3:
3883 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
3884 		break;
3885 
3886 	case 4:
3887 		error = zap_remove(os, object, name, tx);
3888 		ASSERT(error == 0 || error == ENOENT);
3889 		break;
3890 	}
3891 
3892 	if (tx != NULL)
3893 		dmu_tx_commit(tx);
3894 }
3895 
3896 /*
3897  * Commit callback data.
3898  */
3899 typedef struct ztest_cb_data {
3900 	list_node_t		zcd_node;
3901 	uint64_t		zcd_txg;
3902 	int			zcd_expected_err;
3903 	boolean_t		zcd_added;
3904 	boolean_t		zcd_called;
3905 	spa_t			*zcd_spa;
3906 } ztest_cb_data_t;
3907 
3908 /* This is the actual commit callback function */
3909 static void
3910 ztest_commit_callback(void *arg, int error)
3911 {
3912 	ztest_cb_data_t *data = arg;
3913 	uint64_t synced_txg;
3914 
3915 	VERIFY(data != NULL);
3916 	VERIFY3S(data->zcd_expected_err, ==, error);
3917 	VERIFY(!data->zcd_called);
3918 
3919 	synced_txg = spa_last_synced_txg(data->zcd_spa);
3920 	if (data->zcd_txg > synced_txg)
3921 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
3922 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
3923 		    synced_txg);
3924 
3925 	data->zcd_called = B_TRUE;
3926 
3927 	if (error == ECANCELED) {
3928 		ASSERT3U(data->zcd_txg, ==, 0);
3929 		ASSERT(!data->zcd_added);
3930 
3931 		/*
3932 		 * The private callback data should be destroyed here, but
3933 		 * since we are going to check the zcd_called field after
3934 		 * dmu_tx_abort(), we will destroy it there.
3935 		 */
3936 		return;
3937 	}
3938 
3939 	/* Was this callback added to the global callback list? */
3940 	if (!data->zcd_added)
3941 		goto out;
3942 
3943 	ASSERT3U(data->zcd_txg, !=, 0);
3944 
3945 	/* Remove our callback from the list */
3946 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
3947 	list_remove(&zcl.zcl_callbacks, data);
3948 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
3949 
3950 out:
3951 	umem_free(data, sizeof (ztest_cb_data_t));
3952 }
3953 
3954 /* Allocate and initialize callback data structure */
3955 static ztest_cb_data_t *
3956 ztest_create_cb_data(objset_t *os, uint64_t txg)
3957 {
3958 	ztest_cb_data_t *cb_data;
3959 
3960 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
3961 
3962 	cb_data->zcd_txg = txg;
3963 	cb_data->zcd_spa = dmu_objset_spa(os);
3964 
3965 	return (cb_data);
3966 }
3967 
3968 /*
3969  * If a number of txgs equal to this threshold have been created after a commit
3970  * callback has been registered but not called, then we assume there is an
3971  * implementation bug.
3972  */
3973 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
3974 
3975 /*
3976  * Commit callback test.
3977  */
3978 void
3979 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
3980 {
3981 	objset_t *os = zd->zd_os;
3982 	ztest_od_t od[1];
3983 	dmu_tx_t *tx;
3984 	ztest_cb_data_t *cb_data[3], *tmp_cb;
3985 	uint64_t old_txg, txg;
3986 	int i, error;
3987 
3988 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
3989 
3990 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3991 		return;
3992 
3993 	tx = dmu_tx_create(os);
3994 
3995 	cb_data[0] = ztest_create_cb_data(os, 0);
3996 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
3997 
3998 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
3999 
4000 	/* Every once in a while, abort the transaction on purpose */
4001 	if (ztest_random(100) == 0)
4002 		error = -1;
4003 
4004 	if (!error)
4005 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4006 
4007 	txg = error ? 0 : dmu_tx_get_txg(tx);
4008 
4009 	cb_data[0]->zcd_txg = txg;
4010 	cb_data[1] = ztest_create_cb_data(os, txg);
4011 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4012 
4013 	if (error) {
4014 		/*
4015 		 * It's not a strict requirement to call the registered
4016 		 * callbacks from inside dmu_tx_abort(), but that's what
4017 		 * it's supposed to happen in the current implementation
4018 		 * so we will check for that.
4019 		 */
4020 		for (i = 0; i < 2; i++) {
4021 			cb_data[i]->zcd_expected_err = ECANCELED;
4022 			VERIFY(!cb_data[i]->zcd_called);
4023 		}
4024 
4025 		dmu_tx_abort(tx);
4026 
4027 		for (i = 0; i < 2; i++) {
4028 			VERIFY(cb_data[i]->zcd_called);
4029 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4030 		}
4031 
4032 		return;
4033 	}
4034 
4035 	cb_data[2] = ztest_create_cb_data(os, txg);
4036 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4037 
4038 	/*
4039 	 * Read existing data to make sure there isn't a future leak.
4040 	 */
4041 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4042 	    &old_txg, DMU_READ_PREFETCH));
4043 
4044 	if (old_txg > txg)
4045 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4046 		    old_txg, txg);
4047 
4048 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4049 
4050 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4051 
4052 	/*
4053 	 * Since commit callbacks don't have any ordering requirement and since
4054 	 * it is theoretically possible for a commit callback to be called
4055 	 * after an arbitrary amount of time has elapsed since its txg has been
4056 	 * synced, it is difficult to reliably determine whether a commit
4057 	 * callback hasn't been called due to high load or due to a flawed
4058 	 * implementation.
4059 	 *
4060 	 * In practice, we will assume that if after a certain number of txgs a
4061 	 * commit callback hasn't been called, then most likely there's an
4062 	 * implementation bug..
4063 	 */
4064 	tmp_cb = list_head(&zcl.zcl_callbacks);
4065 	if (tmp_cb != NULL &&
4066 	    tmp_cb->zcd_txg > txg - ZTEST_COMMIT_CALLBACK_THRESH) {
4067 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4068 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4069 	}
4070 
4071 	/*
4072 	 * Let's find the place to insert our callbacks.
4073 	 *
4074 	 * Even though the list is ordered by txg, it is possible for the
4075 	 * insertion point to not be the end because our txg may already be
4076 	 * quiescing at this point and other callbacks in the open txg
4077 	 * (from other objsets) may have sneaked in.
4078 	 */
4079 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4080 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4081 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4082 
4083 	/* Add the 3 callbacks to the list */
4084 	for (i = 0; i < 3; i++) {
4085 		if (tmp_cb == NULL)
4086 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4087 		else
4088 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4089 			    cb_data[i]);
4090 
4091 		cb_data[i]->zcd_added = B_TRUE;
4092 		VERIFY(!cb_data[i]->zcd_called);
4093 
4094 		tmp_cb = cb_data[i];
4095 	}
4096 
4097 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4098 
4099 	dmu_tx_commit(tx);
4100 }
4101 
4102 /* ARGSUSED */
4103 void
4104 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4105 {
4106 	zfs_prop_t proplist[] = {
4107 		ZFS_PROP_CHECKSUM,
4108 		ZFS_PROP_COMPRESSION,
4109 		ZFS_PROP_COPIES,
4110 		ZFS_PROP_DEDUP
4111 	};
4112 	ztest_shared_t *zs = ztest_shared;
4113 
4114 	(void) rw_rdlock(&zs->zs_name_lock);
4115 
4116 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4117 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4118 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4119 
4120 	(void) rw_unlock(&zs->zs_name_lock);
4121 }
4122 
4123 /* ARGSUSED */
4124 void
4125 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4126 {
4127 	ztest_shared_t *zs = ztest_shared;
4128 	nvlist_t *props = NULL;
4129 
4130 	(void) rw_rdlock(&zs->zs_name_lock);
4131 
4132 #if 0
4133 	(void) ztest_spa_prop_set_uint64(zs, ZPOOL_PROP_DEDUPDITTO,
4134 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4135 #endif
4136 
4137 	VERIFY3U(spa_prop_get(zs->zs_spa, &props), ==, 0);
4138 
4139 	if (zopt_verbose >= 6)
4140 		dump_nvlist(props, 4);
4141 
4142 	nvlist_free(props);
4143 
4144 	(void) rw_unlock(&zs->zs_name_lock);
4145 }
4146 
4147 /*
4148  * Test snapshot hold/release and deferred destroy.
4149  */
4150 void
4151 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4152 {
4153 	int error;
4154 	objset_t *os = zd->zd_os;
4155 	objset_t *origin;
4156 	char snapname[100];
4157 	char fullname[100];
4158 	char clonename[100];
4159 	char tag[100];
4160 	char osname[MAXNAMELEN];
4161 
4162 	(void) rw_rdlock(&ztest_shared->zs_name_lock);
4163 
4164 	dmu_objset_name(os, osname);
4165 
4166 	(void) snprintf(snapname, 100, "sh1_%llu", id);
4167 	(void) snprintf(fullname, 100, "%s@%s", osname, snapname);
4168 	(void) snprintf(clonename, 100, "%s/ch1_%llu", osname, id);
4169 	(void) snprintf(tag, 100, "%tag_%llu", id);
4170 
4171 	/*
4172 	 * Clean up from any previous run.
4173 	 */
4174 	(void) dmu_objset_destroy(clonename, B_FALSE);
4175 	(void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4176 	(void) dmu_objset_destroy(fullname, B_FALSE);
4177 
4178 	/*
4179 	 * Create snapshot, clone it, mark snap for deferred destroy,
4180 	 * destroy clone, verify snap was also destroyed.
4181 	 */
4182 	error = dmu_objset_snapshot(osname, snapname, NULL, FALSE);
4183 	if (error) {
4184 		if (error == ENOSPC) {
4185 			ztest_record_enospc("dmu_objset_snapshot");
4186 			goto out;
4187 		}
4188 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4189 	}
4190 
4191 	error = dmu_objset_hold(fullname, FTAG, &origin);
4192 	if (error)
4193 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4194 
4195 	error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0);
4196 	dmu_objset_rele(origin, FTAG);
4197 	if (error) {
4198 		if (error == ENOSPC) {
4199 			ztest_record_enospc("dmu_objset_clone");
4200 			goto out;
4201 		}
4202 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4203 	}
4204 
4205 	error = dmu_objset_destroy(fullname, B_TRUE);
4206 	if (error) {
4207 		fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4208 		    fullname, error);
4209 	}
4210 
4211 	error = dmu_objset_destroy(clonename, B_FALSE);
4212 	if (error)
4213 		fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error);
4214 
4215 	error = dmu_objset_hold(fullname, FTAG, &origin);
4216 	if (error != ENOENT)
4217 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4218 
4219 	/*
4220 	 * Create snapshot, add temporary hold, verify that we can't
4221 	 * destroy a held snapshot, mark for deferred destroy,
4222 	 * release hold, verify snapshot was destroyed.
4223 	 */
4224 	error = dmu_objset_snapshot(osname, snapname, NULL, FALSE);
4225 	if (error) {
4226 		if (error == ENOSPC) {
4227 			ztest_record_enospc("dmu_objset_snapshot");
4228 			goto out;
4229 		}
4230 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4231 	}
4232 
4233 	error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE, B_TRUE);
4234 	if (error)
4235 		fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4236 
4237 	error = dmu_objset_destroy(fullname, B_FALSE);
4238 	if (error != EBUSY) {
4239 		fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4240 		    fullname, error);
4241 	}
4242 
4243 	error = dmu_objset_destroy(fullname, B_TRUE);
4244 	if (error) {
4245 		fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4246 		    fullname, error);
4247 	}
4248 
4249 	error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4250 	if (error)
4251 		fatal(0, "dsl_dataset_user_release(%s)", fullname, tag);
4252 
4253 	VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT);
4254 
4255 out:
4256 	(void) rw_unlock(&ztest_shared->zs_name_lock);
4257 }
4258 
4259 /*
4260  * Inject random faults into the on-disk data.
4261  */
4262 /* ARGSUSED */
4263 void
4264 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4265 {
4266 	ztest_shared_t *zs = ztest_shared;
4267 	spa_t *spa = zs->zs_spa;
4268 	int fd;
4269 	uint64_t offset;
4270 	uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz;
4271 	uint64_t bad = 0x1990c0ffeedecade;
4272 	uint64_t top, leaf;
4273 	char path0[MAXPATHLEN];
4274 	char pathrand[MAXPATHLEN];
4275 	size_t fsize;
4276 	int bshift = SPA_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
4277 	int iters = 1000;
4278 	int maxfaults = zopt_maxfaults;
4279 	vdev_t *vd0 = NULL;
4280 	uint64_t guid0 = 0;
4281 	boolean_t islog = B_FALSE;
4282 
4283 	ASSERT(leaves >= 1);
4284 
4285 	/*
4286 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4287 	 */
4288 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4289 
4290 	if (ztest_random(2) == 0) {
4291 		/*
4292 		 * Inject errors on a normal data device or slog device.
4293 		 */
4294 		top = ztest_random_vdev_top(spa, B_TRUE);
4295 		leaf = ztest_random(leaves);
4296 
4297 		/*
4298 		 * Generate paths to the first leaf in this top-level vdev,
4299 		 * and to the random leaf we selected.  We'll induce transient
4300 		 * write failures and random online/offline activity on leaf 0,
4301 		 * and we'll write random garbage to the randomly chosen leaf.
4302 		 */
4303 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
4304 		    zopt_dir, zopt_pool, top * leaves + 0);
4305 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4306 		    zopt_dir, zopt_pool, top * leaves + leaf);
4307 
4308 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4309 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4310 			islog = B_TRUE;
4311 
4312 		if (vd0 != NULL && maxfaults != 1) {
4313 			/*
4314 			 * Make vd0 explicitly claim to be unreadable,
4315 			 * or unwriteable, or reach behind its back
4316 			 * and close the underlying fd.  We can do this if
4317 			 * maxfaults == 0 because we'll fail and reexecute,
4318 			 * and we can do it if maxfaults >= 2 because we'll
4319 			 * have enough redundancy.  If maxfaults == 1, the
4320 			 * combination of this with injection of random data
4321 			 * corruption below exceeds the pool's fault tolerance.
4322 			 */
4323 			vdev_file_t *vf = vd0->vdev_tsd;
4324 
4325 			if (vf != NULL && ztest_random(3) == 0) {
4326 				(void) close(vf->vf_vnode->v_fd);
4327 				vf->vf_vnode->v_fd = -1;
4328 			} else if (ztest_random(2) == 0) {
4329 				vd0->vdev_cant_read = B_TRUE;
4330 			} else {
4331 				vd0->vdev_cant_write = B_TRUE;
4332 			}
4333 			guid0 = vd0->vdev_guid;
4334 		}
4335 	} else {
4336 		/*
4337 		 * Inject errors on an l2cache device.
4338 		 */
4339 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
4340 
4341 		if (sav->sav_count == 0) {
4342 			spa_config_exit(spa, SCL_STATE, FTAG);
4343 			return;
4344 		}
4345 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4346 		guid0 = vd0->vdev_guid;
4347 		(void) strcpy(path0, vd0->vdev_path);
4348 		(void) strcpy(pathrand, vd0->vdev_path);
4349 
4350 		leaf = 0;
4351 		leaves = 1;
4352 		maxfaults = INT_MAX;	/* no limit on cache devices */
4353 	}
4354 
4355 	spa_config_exit(spa, SCL_STATE, FTAG);
4356 
4357 	/*
4358 	 * If we can tolerate two or more faults, or we're dealing
4359 	 * with a slog, randomly online/offline vd0.
4360 	 */
4361 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
4362 		if (ztest_random(10) < 6) {
4363 			int flags = (ztest_random(2) == 0 ?
4364 			    ZFS_OFFLINE_TEMPORARY : 0);
4365 
4366 			/*
4367 			 * We have to grab the zs_name_lock as writer to
4368 			 * prevent a race between offlining a slog and
4369 			 * destroying a dataset. Offlining the slog will
4370 			 * grab a reference on the dataset which may cause
4371 			 * dmu_objset_destroy() to fail with EBUSY thus
4372 			 * leaving the dataset in an inconsistent state.
4373 			 */
4374 			if (islog)
4375 				(void) rw_wrlock(&ztest_shared->zs_name_lock);
4376 
4377 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4378 
4379 			if (islog)
4380 				(void) rw_unlock(&ztest_shared->zs_name_lock);
4381 		} else {
4382 			(void) vdev_online(spa, guid0, 0, NULL);
4383 		}
4384 	}
4385 
4386 	if (maxfaults == 0)
4387 		return;
4388 
4389 	/*
4390 	 * We have at least single-fault tolerance, so inject data corruption.
4391 	 */
4392 	fd = open(pathrand, O_RDWR);
4393 
4394 	if (fd == -1)	/* we hit a gap in the device namespace */
4395 		return;
4396 
4397 	fsize = lseek(fd, 0, SEEK_END);
4398 
4399 	while (--iters != 0) {
4400 		offset = ztest_random(fsize / (leaves << bshift)) *
4401 		    (leaves << bshift) + (leaf << bshift) +
4402 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4403 
4404 		if (offset >= fsize)
4405 			continue;
4406 
4407 		if (zopt_verbose >= 7)
4408 			(void) printf("injecting bad word into %s,"
4409 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4410 
4411 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4412 			fatal(1, "can't inject bad word at 0x%llx in %s",
4413 			    offset, pathrand);
4414 	}
4415 
4416 	(void) close(fd);
4417 }
4418 
4419 /*
4420  * Verify that DDT repair works as expected.
4421  */
4422 void
4423 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4424 {
4425 	ztest_shared_t *zs = ztest_shared;
4426 	spa_t *spa = zs->zs_spa;
4427 	objset_t *os = zd->zd_os;
4428 	ztest_od_t od[1];
4429 	uint64_t object, blocksize, txg, pattern, psize;
4430 	enum zio_checksum checksum = spa_dedup_checksum(spa);
4431 	dmu_buf_t *db;
4432 	dmu_tx_t *tx;
4433 	void *buf;
4434 	blkptr_t blk;
4435 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
4436 
4437 	blocksize = ztest_random_blocksize();
4438 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
4439 
4440 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4441 
4442 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4443 		return;
4444 
4445 	/*
4446 	 * Take the name lock as writer to prevent anyone else from changing
4447 	 * the pool and dataset properies we need to maintain during this test.
4448 	 */
4449 	(void) rw_wrlock(&zs->zs_name_lock);
4450 
4451 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4452 	    B_FALSE) != 0 ||
4453 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4454 	    B_FALSE) != 0) {
4455 		(void) rw_unlock(&zs->zs_name_lock);
4456 		return;
4457 	}
4458 
4459 	object = od[0].od_object;
4460 	blocksize = od[0].od_blocksize;
4461 	pattern = spa_guid(spa) ^ dmu_objset_fsid_guid(os);
4462 
4463 	ASSERT(object != 0);
4464 
4465 	tx = dmu_tx_create(os);
4466 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4467 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4468 	if (txg == 0) {
4469 		(void) rw_unlock(&zs->zs_name_lock);
4470 		return;
4471 	}
4472 
4473 	/*
4474 	 * Write all the copies of our block.
4475 	 */
4476 	for (int i = 0; i < copies; i++) {
4477 		uint64_t offset = i * blocksize;
4478 		VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db) == 0);
4479 		ASSERT(db->db_offset == offset);
4480 		ASSERT(db->db_size == blocksize);
4481 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4482 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4483 		dmu_buf_will_fill(db, tx);
4484 		ztest_pattern_set(db->db_data, db->db_size, pattern);
4485 		dmu_buf_rele(db, FTAG);
4486 	}
4487 
4488 	dmu_tx_commit(tx);
4489 	txg_wait_synced(spa_get_dsl(spa), txg);
4490 
4491 	/*
4492 	 * Find out what block we got.
4493 	 */
4494 	VERIFY(dmu_buf_hold(os, object, 0, FTAG, &db) == 0);
4495 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4496 	dmu_buf_rele(db, FTAG);
4497 
4498 	/*
4499 	 * Damage the block.  Dedup-ditto will save us when we read it later.
4500 	 */
4501 	psize = BP_GET_PSIZE(&blk);
4502 	buf = zio_buf_alloc(psize);
4503 	ztest_pattern_set(buf, psize, ~pattern);
4504 
4505 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
4506 	    buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
4507 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
4508 
4509 	zio_buf_free(buf, psize);
4510 
4511 	(void) rw_unlock(&zs->zs_name_lock);
4512 }
4513 
4514 /*
4515  * Scrub the pool.
4516  */
4517 /* ARGSUSED */
4518 void
4519 ztest_scrub(ztest_ds_t *zd, uint64_t id)
4520 {
4521 	ztest_shared_t *zs = ztest_shared;
4522 	spa_t *spa = zs->zs_spa;
4523 
4524 	(void) spa_scrub(spa, POOL_SCRUB_EVERYTHING);
4525 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
4526 	(void) spa_scrub(spa, POOL_SCRUB_EVERYTHING);
4527 }
4528 
4529 /*
4530  * Rename the pool to a different name and then rename it back.
4531  */
4532 /* ARGSUSED */
4533 void
4534 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
4535 {
4536 	ztest_shared_t *zs = ztest_shared;
4537 	char *oldname, *newname;
4538 	spa_t *spa;
4539 
4540 	(void) rw_wrlock(&zs->zs_name_lock);
4541 
4542 	oldname = zs->zs_pool;
4543 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
4544 	(void) strcpy(newname, oldname);
4545 	(void) strcat(newname, "_tmp");
4546 
4547 	/*
4548 	 * Do the rename
4549 	 */
4550 	VERIFY3U(0, ==, spa_rename(oldname, newname));
4551 
4552 	/*
4553 	 * Try to open it under the old name, which shouldn't exist
4554 	 */
4555 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
4556 
4557 	/*
4558 	 * Open it under the new name and make sure it's still the same spa_t.
4559 	 */
4560 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
4561 
4562 	ASSERT(spa == zs->zs_spa);
4563 	spa_close(spa, FTAG);
4564 
4565 	/*
4566 	 * Rename it back to the original
4567 	 */
4568 	VERIFY3U(0, ==, spa_rename(newname, oldname));
4569 
4570 	/*
4571 	 * Make sure it can still be opened
4572 	 */
4573 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
4574 
4575 	ASSERT(spa == zs->zs_spa);
4576 	spa_close(spa, FTAG);
4577 
4578 	umem_free(newname, strlen(newname) + 1);
4579 
4580 	(void) rw_unlock(&zs->zs_name_lock);
4581 }
4582 
4583 /*
4584  * Verify pool integrity by running zdb.
4585  */
4586 static void
4587 ztest_run_zdb(char *pool)
4588 {
4589 	int status;
4590 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
4591 	char zbuf[1024];
4592 	char *bin;
4593 	char *ztest;
4594 	char *isa;
4595 	int isalen;
4596 	FILE *fp;
4597 
4598 	(void) realpath(getexecname(), zdb);
4599 
4600 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
4601 	bin = strstr(zdb, "/usr/bin/");
4602 	ztest = strstr(bin, "/ztest");
4603 	isa = bin + 8;
4604 	isalen = ztest - isa;
4605 	isa = strdup(isa);
4606 	/* LINTED */
4607 	(void) sprintf(bin,
4608 	    "/usr/sbin%.*s/zdb -bcc%s%s -U /tmp/zpool.cache %s",
4609 	    isalen,
4610 	    isa,
4611 	    zopt_verbose >= 3 ? "s" : "",
4612 	    zopt_verbose >= 4 ? "v" : "",
4613 	    pool);
4614 	free(isa);
4615 
4616 	if (zopt_verbose >= 5)
4617 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
4618 
4619 	fp = popen(zdb, "r");
4620 
4621 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
4622 		if (zopt_verbose >= 3)
4623 			(void) printf("%s", zbuf);
4624 
4625 	status = pclose(fp);
4626 
4627 	if (status == 0)
4628 		return;
4629 
4630 	ztest_dump_core = 0;
4631 	if (WIFEXITED(status))
4632 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
4633 	else
4634 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
4635 }
4636 
4637 static void
4638 ztest_walk_pool_directory(char *header)
4639 {
4640 	spa_t *spa = NULL;
4641 
4642 	if (zopt_verbose >= 6)
4643 		(void) printf("%s\n", header);
4644 
4645 	mutex_enter(&spa_namespace_lock);
4646 	while ((spa = spa_next(spa)) != NULL)
4647 		if (zopt_verbose >= 6)
4648 			(void) printf("\t%s\n", spa_name(spa));
4649 	mutex_exit(&spa_namespace_lock);
4650 }
4651 
4652 static void
4653 ztest_spa_import_export(char *oldname, char *newname)
4654 {
4655 	nvlist_t *config, *newconfig;
4656 	uint64_t pool_guid;
4657 	spa_t *spa;
4658 
4659 	if (zopt_verbose >= 4) {
4660 		(void) printf("import/export: old = %s, new = %s\n",
4661 		    oldname, newname);
4662 	}
4663 
4664 	/*
4665 	 * Clean up from previous runs.
4666 	 */
4667 	(void) spa_destroy(newname);
4668 
4669 	/*
4670 	 * Get the pool's configuration and guid.
4671 	 */
4672 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
4673 
4674 	/*
4675 	 * Kick off a scrub to tickle scrub/export races.
4676 	 */
4677 	if (ztest_random(2) == 0)
4678 		(void) spa_scrub(spa, POOL_SCRUB_EVERYTHING);
4679 
4680 	pool_guid = spa_guid(spa);
4681 	spa_close(spa, FTAG);
4682 
4683 	ztest_walk_pool_directory("pools before export");
4684 
4685 	/*
4686 	 * Export it.
4687 	 */
4688 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
4689 
4690 	ztest_walk_pool_directory("pools after export");
4691 
4692 	/*
4693 	 * Try to import it.
4694 	 */
4695 	newconfig = spa_tryimport(config);
4696 	ASSERT(newconfig != NULL);
4697 	nvlist_free(newconfig);
4698 
4699 	/*
4700 	 * Import it under the new name.
4701 	 */
4702 	VERIFY3U(0, ==, spa_import(newname, config, NULL));
4703 
4704 	ztest_walk_pool_directory("pools after import");
4705 
4706 	/*
4707 	 * Try to import it again -- should fail with EEXIST.
4708 	 */
4709 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL));
4710 
4711 	/*
4712 	 * Try to import it under a different name -- should fail with EEXIST.
4713 	 */
4714 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL));
4715 
4716 	/*
4717 	 * Verify that the pool is no longer visible under the old name.
4718 	 */
4719 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
4720 
4721 	/*
4722 	 * Verify that we can open and close the pool using the new name.
4723 	 */
4724 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
4725 	ASSERT(pool_guid == spa_guid(spa));
4726 	spa_close(spa, FTAG);
4727 
4728 	nvlist_free(config);
4729 }
4730 
4731 static void
4732 ztest_resume(spa_t *spa)
4733 {
4734 	if (spa_suspended(spa) && zopt_verbose >= 6)
4735 		(void) printf("resuming from suspended state\n");
4736 	spa_vdev_state_enter(spa, SCL_NONE);
4737 	vdev_clear(spa, NULL);
4738 	(void) spa_vdev_state_exit(spa, NULL, 0);
4739 	(void) zio_resume(spa);
4740 }
4741 
4742 static void *
4743 ztest_resume_thread(void *arg)
4744 {
4745 	spa_t *spa = arg;
4746 
4747 	while (!ztest_exiting) {
4748 		if (spa_suspended(spa))
4749 			ztest_resume(spa);
4750 		(void) poll(NULL, 0, 100);
4751 	}
4752 	return (NULL);
4753 }
4754 
4755 static void *
4756 ztest_deadman_thread(void *arg)
4757 {
4758 	ztest_shared_t *zs = arg;
4759 	int grace = 300;
4760 	hrtime_t delta;
4761 
4762 	delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
4763 
4764 	(void) poll(NULL, 0, (int)(1000 * delta));
4765 
4766 	fatal(0, "failed to complete within %d seconds of deadline", grace);
4767 
4768 	return (NULL);
4769 }
4770 
4771 static void
4772 ztest_execute(ztest_info_t *zi, uint64_t id)
4773 {
4774 	ztest_shared_t *zs = ztest_shared;
4775 	ztest_ds_t *zd = &zs->zs_zd[id % zopt_datasets];
4776 	hrtime_t functime = gethrtime();
4777 
4778 	for (int i = 0; i < zi->zi_iters; i++)
4779 		zi->zi_func(zd, id);
4780 
4781 	functime = gethrtime() - functime;
4782 
4783 	atomic_add_64(&zi->zi_call_count, 1);
4784 	atomic_add_64(&zi->zi_call_time, functime);
4785 
4786 	if (zopt_verbose >= 4) {
4787 		Dl_info dli;
4788 		(void) dladdr((void *)zi->zi_func, &dli);
4789 		(void) printf("%6.2f sec in %s\n",
4790 		    (double)functime / NANOSEC, dli.dli_sname);
4791 	}
4792 }
4793 
4794 static void *
4795 ztest_thread(void *arg)
4796 {
4797 	uint64_t id = (uintptr_t)arg;
4798 	ztest_shared_t *zs = ztest_shared;
4799 	uint64_t call_next;
4800 	hrtime_t now;
4801 	ztest_info_t *zi;
4802 
4803 	while ((now = gethrtime()) < zs->zs_thread_stop) {
4804 		/*
4805 		 * See if it's time to force a crash.
4806 		 */
4807 		if (now > zs->zs_thread_kill)
4808 			ztest_kill(zs);
4809 
4810 		/*
4811 		 * If we're getting ENOSPC with some regularity, stop.
4812 		 */
4813 		if (zs->zs_enospc_count > 10)
4814 			break;
4815 
4816 		/*
4817 		 * Pick a random function to execute.
4818 		 */
4819 		zi = &zs->zs_info[ztest_random(ZTEST_FUNCS)];
4820 		call_next = zi->zi_call_next;
4821 
4822 		if (now >= call_next &&
4823 		    atomic_cas_64(&zi->zi_call_next, call_next, call_next +
4824 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next)
4825 			ztest_execute(zi, id);
4826 	}
4827 
4828 	return (NULL);
4829 }
4830 
4831 static void
4832 ztest_dataset_name(char *dsname, char *pool, int d)
4833 {
4834 	(void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
4835 }
4836 
4837 static void
4838 ztest_dataset_destroy(ztest_shared_t *zs, int d)
4839 {
4840 	char name[MAXNAMELEN];
4841 
4842 	ztest_dataset_name(name, zs->zs_pool, d);
4843 
4844 	if (zopt_verbose >= 3)
4845 		(void) printf("Destroying %s to free up space\n", name);
4846 
4847 	/*
4848 	 * Cleanup any non-standard clones and snapshots.  In general,
4849 	 * ztest thread t operates on dataset (t % zopt_datasets),
4850 	 * so there may be more than one thing to clean up.
4851 	 */
4852 	for (int t = d; t < zopt_threads; t += zopt_datasets)
4853 		ztest_dsl_dataset_cleanup(name, t);
4854 
4855 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
4856 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
4857 }
4858 
4859 static void
4860 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
4861 {
4862 	uint64_t usedobjs, dirobjs, scratch;
4863 
4864 	/*
4865 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
4866 	 * Therefore, the number of objects in use should equal the
4867 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
4868 	 * If not, we have an object leak.
4869 	 *
4870 	 * Note that we can only check this in ztest_dataset_open(),
4871 	 * when the open-context and syncing-context values agree.
4872 	 * That's because zap_count() returns the open-context value,
4873 	 * while dmu_objset_space() returns the rootbp fill count.
4874 	 */
4875 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
4876 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
4877 	ASSERT3U(dirobjs + 1, ==, usedobjs);
4878 }
4879 
4880 static int
4881 ztest_dataset_open(ztest_shared_t *zs, int d)
4882 {
4883 	ztest_ds_t *zd = &zs->zs_zd[d];
4884 	uint64_t committed_seq = zd->zd_seq;
4885 	objset_t *os;
4886 	zilog_t *zilog;
4887 	char name[MAXNAMELEN];
4888 	int error;
4889 
4890 	ztest_dataset_name(name, zs->zs_pool, d);
4891 
4892 	(void) rw_rdlock(&zs->zs_name_lock);
4893 
4894 	error = dmu_objset_create(name, DMU_OST_OTHER, 0,
4895 	    ztest_objset_create_cb, NULL);
4896 	if (error == ENOSPC) {
4897 		(void) rw_unlock(&zs->zs_name_lock);
4898 		ztest_record_enospc(FTAG);
4899 		return (error);
4900 	}
4901 	ASSERT(error == 0 || error == EEXIST);
4902 
4903 	VERIFY3U(dmu_objset_hold(name, zd, &os), ==, 0);
4904 	(void) rw_unlock(&zs->zs_name_lock);
4905 
4906 	ztest_zd_init(zd, os);
4907 
4908 	zilog = zd->zd_zilog;
4909 
4910 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
4911 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
4912 		fatal(0, "missing log records: claimed %llu < committed %llu",
4913 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
4914 
4915 	ztest_dataset_dirobj_verify(zd);
4916 
4917 	zil_replay(os, zd, ztest_replay_vector);
4918 
4919 	ztest_dataset_dirobj_verify(zd);
4920 
4921 	if (zopt_verbose >= 6)
4922 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
4923 		    zd->zd_name,
4924 		    (u_longlong_t)zilog->zl_parse_blk_count,
4925 		    (u_longlong_t)zilog->zl_parse_lr_count,
4926 		    (u_longlong_t)zilog->zl_replaying_seq);
4927 
4928 	zilog = zil_open(os, ztest_get_data);
4929 
4930 	if (zilog->zl_replaying_seq != 0 &&
4931 	    zilog->zl_replaying_seq < committed_seq)
4932 		fatal(0, "missing log records: replayed %llu < committed %llu",
4933 		    zilog->zl_replaying_seq, committed_seq);
4934 
4935 	return (0);
4936 }
4937 
4938 static void
4939 ztest_dataset_close(ztest_shared_t *zs, int d)
4940 {
4941 	ztest_ds_t *zd = &zs->zs_zd[d];
4942 
4943 	zil_close(zd->zd_zilog);
4944 	dmu_objset_rele(zd->zd_os, zd);
4945 
4946 	ztest_zd_fini(zd);
4947 }
4948 
4949 /*
4950  * Kick off threads to run tests on all datasets in parallel.
4951  */
4952 static void
4953 ztest_run(ztest_shared_t *zs)
4954 {
4955 	thread_t *tid;
4956 	spa_t *spa;
4957 	thread_t resume_tid;
4958 	int error;
4959 
4960 	ztest_exiting = B_FALSE;
4961 
4962 	/*
4963 	 * Initialize parent/child shared state.
4964 	 */
4965 	VERIFY(_mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL) == 0);
4966 	VERIFY(rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL) == 0);
4967 
4968 	zs->zs_thread_start = gethrtime();
4969 	zs->zs_thread_stop = zs->zs_thread_start + zopt_passtime * NANOSEC;
4970 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
4971 	zs->zs_thread_kill = zs->zs_thread_stop;
4972 	if (ztest_random(100) < zopt_killrate)
4973 		zs->zs_thread_kill -= ztest_random(zopt_passtime * NANOSEC);
4974 
4975 	(void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
4976 
4977 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
4978 	    offsetof(ztest_cb_data_t, zcd_node));
4979 
4980 	/*
4981 	 * Open our pool.
4982 	 */
4983 	kernel_init(FREAD | FWRITE);
4984 	VERIFY(spa_open(zs->zs_pool, &spa, FTAG) == 0);
4985 	zs->zs_spa = spa;
4986 
4987 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
4988 
4989 	/*
4990 	 * We don't expect the pool to suspend unless maxfaults == 0,
4991 	 * in which case ztest_fault_inject() temporarily takes away
4992 	 * the only valid replica.
4993 	 */
4994 	if (zopt_maxfaults == 0)
4995 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
4996 	else
4997 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
4998 
4999 	/*
5000 	 * Create a thread to periodically resume suspended I/O.
5001 	 */
5002 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5003 	    &resume_tid) == 0);
5004 
5005 	/*
5006 	 * Create a deadman thread to abort() if we hang.
5007 	 */
5008 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5009 	    NULL) == 0);
5010 
5011 	/*
5012 	 * Verify that we can safely inquire about about any object,
5013 	 * whether it's allocated or not.  To make it interesting,
5014 	 * we probe a 5-wide window around each power of two.
5015 	 * This hits all edge cases, including zero and the max.
5016 	 */
5017 	for (int t = 0; t < 64; t++) {
5018 		for (int d = -5; d <= 5; d++) {
5019 			error = dmu_object_info(spa->spa_meta_objset,
5020 			    (1ULL << t) + d, NULL);
5021 			ASSERT(error == 0 || error == ENOENT ||
5022 			    error == EINVAL);
5023 		}
5024 	}
5025 
5026 	/*
5027 	 * If we got any ENOSPC errors on the previous run, destroy something.
5028 	 */
5029 	if (zs->zs_enospc_count != 0) {
5030 		int d = ztest_random(zopt_datasets);
5031 		ztest_dataset_destroy(zs, d);
5032 	}
5033 	zs->zs_enospc_count = 0;
5034 
5035 	tid = umem_zalloc(zopt_threads * sizeof (thread_t), UMEM_NOFAIL);
5036 
5037 	if (zopt_verbose >= 4)
5038 		(void) printf("starting main threads...\n");
5039 
5040 	/*
5041 	 * Kick off all the tests that run in parallel.
5042 	 */
5043 	for (int t = 0; t < zopt_threads; t++) {
5044 		if (t < zopt_datasets && ztest_dataset_open(zs, t) != 0)
5045 			return;
5046 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5047 		    THR_BOUND, &tid[t]) == 0);
5048 	}
5049 
5050 	/*
5051 	 * Wait for all of the tests to complete.  We go in reverse order
5052 	 * so we don't close datasets while threads are still using them.
5053 	 */
5054 	for (int t = zopt_threads - 1; t >= 0; t--) {
5055 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5056 		if (t < zopt_datasets)
5057 			ztest_dataset_close(zs, t);
5058 	}
5059 
5060 	txg_wait_synced(spa_get_dsl(spa), 0);
5061 
5062 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5063 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5064 
5065 	umem_free(tid, zopt_threads * sizeof (thread_t));
5066 
5067 	/* Kill the resume thread */
5068 	ztest_exiting = B_TRUE;
5069 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5070 	ztest_resume(spa);
5071 
5072 	/*
5073 	 * Right before closing the pool, kick off a bunch of async I/O;
5074 	 * spa_close() should wait for it to complete.
5075 	 */
5076 	for (uint64_t object = 1; object < 50; object++)
5077 		dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5078 
5079 	spa_close(spa, FTAG);
5080 
5081 	/*
5082 	 * Verify that we can loop over all pools.
5083 	 */
5084 	mutex_enter(&spa_namespace_lock);
5085 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5086 		if (zopt_verbose > 3)
5087 			(void) printf("spa_next: found %s\n", spa_name(spa));
5088 	mutex_exit(&spa_namespace_lock);
5089 
5090 	/*
5091 	 * Verify that we can export the pool and reimport it under a
5092 	 * different name.
5093 	 */
5094 	if (ztest_random(2) == 0) {
5095 		char name[MAXNAMELEN];
5096 		(void) snprintf(name, MAXNAMELEN, "%s_import", zs->zs_pool);
5097 		ztest_spa_import_export(zs->zs_pool, name);
5098 		ztest_spa_import_export(name, zs->zs_pool);
5099 	}
5100 
5101 	kernel_fini();
5102 }
5103 
5104 static void
5105 ztest_freeze(ztest_shared_t *zs)
5106 {
5107 	ztest_ds_t *zd = &zs->zs_zd[0];
5108 	spa_t *spa;
5109 
5110 	if (zopt_verbose >= 3)
5111 		(void) printf("testing spa_freeze()...\n");
5112 
5113 	kernel_init(FREAD | FWRITE);
5114 	VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
5115 	VERIFY3U(0, ==, ztest_dataset_open(zs, 0));
5116 
5117 	/*
5118 	 * Force the first log block to be transactionally allocated.
5119 	 * We have to do this before we freeze the pool -- otherwise
5120 	 * the log chain won't be anchored.
5121 	 */
5122 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5123 		ztest_dmu_object_alloc_free(zd, 0);
5124 		zil_commit(zd->zd_zilog, UINT64_MAX, 0);
5125 	}
5126 
5127 	txg_wait_synced(spa_get_dsl(spa), 0);
5128 
5129 	/*
5130 	 * Freeze the pool.  This stops spa_sync() from doing anything,
5131 	 * so that the only way to record changes from now on is the ZIL.
5132 	 */
5133 	spa_freeze(spa);
5134 
5135 	/*
5136 	 * Run tests that generate log records but don't alter the pool config
5137 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5138 	 * We do a txg_wait_synced() after each iteration to force the txg
5139 	 * to increase well beyond the last synced value in the uberblock.
5140 	 * The ZIL should be OK with that.
5141 	 */
5142 	while (ztest_random(20) != 0) {
5143 		ztest_dmu_write_parallel(zd, 0);
5144 		ztest_dmu_object_alloc_free(zd, 0);
5145 		txg_wait_synced(spa_get_dsl(spa), 0);
5146 	}
5147 
5148 	/*
5149 	 * Commit all of the changes we just generated.
5150 	 */
5151 	zil_commit(zd->zd_zilog, UINT64_MAX, 0);
5152 	txg_wait_synced(spa_get_dsl(spa), 0);
5153 
5154 	/*
5155 	 * Close our dataset and close the pool.
5156 	 */
5157 	ztest_dataset_close(zs, 0);
5158 	spa_close(spa, FTAG);
5159 	kernel_fini();
5160 
5161 	/*
5162 	 * Open and close the pool and dataset to induce log replay.
5163 	 */
5164 	kernel_init(FREAD | FWRITE);
5165 	VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
5166 	VERIFY3U(0, ==, ztest_dataset_open(zs, 0));
5167 	ztest_dataset_close(zs, 0);
5168 	spa_close(spa, FTAG);
5169 	kernel_fini();
5170 
5171 	list_destroy(&zcl.zcl_callbacks);
5172 
5173 	(void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5174 
5175 	(void) rwlock_destroy(&zs->zs_name_lock);
5176 	(void) _mutex_destroy(&zs->zs_vdev_lock);
5177 }
5178 
5179 void
5180 print_time(hrtime_t t, char *timebuf)
5181 {
5182 	hrtime_t s = t / NANOSEC;
5183 	hrtime_t m = s / 60;
5184 	hrtime_t h = m / 60;
5185 	hrtime_t d = h / 24;
5186 
5187 	s -= m * 60;
5188 	m -= h * 60;
5189 	h -= d * 24;
5190 
5191 	timebuf[0] = '\0';
5192 
5193 	if (d)
5194 		(void) sprintf(timebuf,
5195 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
5196 	else if (h)
5197 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5198 	else if (m)
5199 		(void) sprintf(timebuf, "%llum%02llus", m, s);
5200 	else
5201 		(void) sprintf(timebuf, "%llus", s);
5202 }
5203 
5204 /*
5205  * Create a storage pool with the given name and initial vdev size.
5206  * Then test spa_freeze() functionality.
5207  */
5208 static void
5209 ztest_init(ztest_shared_t *zs)
5210 {
5211 	spa_t *spa;
5212 	nvlist_t *nvroot;
5213 
5214 	VERIFY(_mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL) == 0);
5215 	VERIFY(rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL) == 0);
5216 
5217 	kernel_init(FREAD | FWRITE);
5218 
5219 	/*
5220 	 * Create the storage pool.
5221 	 */
5222 	(void) spa_destroy(zs->zs_pool);
5223 	ztest_shared->zs_vdev_next_leaf = 0;
5224 	nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
5225 	    0, zopt_raidz, zopt_mirrors, 1);
5226 	VERIFY3U(0, ==, spa_create(zs->zs_pool, nvroot, NULL, NULL, NULL));
5227 	nvlist_free(nvroot);
5228 
5229 	VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
5230 	metaslab_sz = 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5231 	spa_close(spa, FTAG);
5232 
5233 	kernel_fini();
5234 
5235 	ztest_run_zdb(zs->zs_pool);
5236 
5237 	ztest_freeze(zs);
5238 
5239 	ztest_run_zdb(zs->zs_pool);
5240 }
5241 
5242 int
5243 main(int argc, char **argv)
5244 {
5245 	int kills = 0;
5246 	int iters = 0;
5247 	ztest_shared_t *zs;
5248 	size_t shared_size;
5249 	ztest_info_t *zi;
5250 	char timebuf[100];
5251 	char numbuf[6];
5252 	spa_t *spa;
5253 
5254 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
5255 
5256 	/* Override location of zpool.cache */
5257 	spa_config_path = "/tmp/zpool.cache";
5258 
5259 	ztest_random_fd = open("/dev/urandom", O_RDONLY);
5260 
5261 	process_options(argc, argv);
5262 
5263 	/*
5264 	 * Blow away any existing copy of zpool.cache
5265 	 */
5266 	if (zopt_init != 0)
5267 		(void) remove("/tmp/zpool.cache");
5268 
5269 	shared_size = sizeof (*zs) + zopt_datasets * sizeof (ztest_ds_t);
5270 
5271 	zs = ztest_shared = (void *)mmap(0,
5272 	    P2ROUNDUP(shared_size, getpagesize()),
5273 	    PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
5274 
5275 	if (zopt_verbose >= 1) {
5276 		(void) printf("%llu vdevs, %d datasets, %d threads,"
5277 		    " %llu seconds...\n",
5278 		    (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads,
5279 		    (u_longlong_t)zopt_time);
5280 	}
5281 
5282 	/*
5283 	 * Create and initialize our storage pool.
5284 	 */
5285 	for (int i = 1; i <= zopt_init; i++) {
5286 		bzero(zs, sizeof (ztest_shared_t));
5287 		if (zopt_verbose >= 3 && zopt_init != 1)
5288 			(void) printf("ztest_init(), pass %d\n", i);
5289 		zs->zs_pool = zopt_pool;
5290 		ztest_init(zs);
5291 	}
5292 
5293 	zs->zs_pool = zopt_pool;
5294 	zs->zs_proc_start = gethrtime();
5295 	zs->zs_proc_stop = zs->zs_proc_start + zopt_time * NANOSEC;
5296 
5297 	for (int f = 0; f < ZTEST_FUNCS; f++) {
5298 		zi = &zs->zs_info[f];
5299 		*zi = ztest_info[f];
5300 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
5301 			zi->zi_call_next = UINT64_MAX;
5302 		else
5303 			zi->zi_call_next = zs->zs_proc_start +
5304 			    ztest_random(2 * zi->zi_interval[0] + 1);
5305 	}
5306 
5307 	/*
5308 	 * Run the tests in a loop.  These tests include fault injection
5309 	 * to verify that self-healing data works, and forced crashes
5310 	 * to verify that we never lose on-disk consistency.
5311 	 */
5312 	while (gethrtime() < zs->zs_proc_stop) {
5313 		int status;
5314 		pid_t pid;
5315 
5316 		/*
5317 		 * Initialize the workload counters for each function.
5318 		 */
5319 		for (int f = 0; f < ZTEST_FUNCS; f++) {
5320 			zi = &zs->zs_info[f];
5321 			zi->zi_call_count = 0;
5322 			zi->zi_call_time = 0;
5323 		}
5324 
5325 		/* Set the allocation switch size */
5326 		metaslab_df_alloc_threshold = ztest_random(metaslab_sz / 4) + 1;
5327 
5328 		pid = fork();
5329 
5330 		if (pid == -1)
5331 			fatal(1, "fork failed");
5332 
5333 		if (pid == 0) {	/* child */
5334 			struct rlimit rl = { 1024, 1024 };
5335 			(void) setrlimit(RLIMIT_NOFILE, &rl);
5336 			(void) enable_extended_FILE_stdio(-1, -1);
5337 			ztest_run(zs);
5338 			exit(0);
5339 		}
5340 
5341 		while (waitpid(pid, &status, 0) != pid)
5342 			continue;
5343 
5344 		if (WIFEXITED(status)) {
5345 			if (WEXITSTATUS(status) != 0) {
5346 				(void) fprintf(stderr,
5347 				    "child exited with code %d\n",
5348 				    WEXITSTATUS(status));
5349 				exit(2);
5350 			}
5351 		} else if (WIFSIGNALED(status)) {
5352 			if (WTERMSIG(status) != SIGKILL) {
5353 				(void) fprintf(stderr,
5354 				    "child died with signal %d\n",
5355 				    WTERMSIG(status));
5356 				exit(3);
5357 			}
5358 			kills++;
5359 		} else {
5360 			(void) fprintf(stderr, "something strange happened "
5361 			    "to child\n");
5362 			exit(4);
5363 		}
5364 
5365 		iters++;
5366 
5367 		if (zopt_verbose >= 1) {
5368 			hrtime_t now = gethrtime();
5369 
5370 			now = MIN(now, zs->zs_proc_stop);
5371 			print_time(zs->zs_proc_stop - now, timebuf);
5372 			nicenum(zs->zs_space, numbuf);
5373 
5374 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
5375 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
5376 			    iters,
5377 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
5378 			    (u_longlong_t)zs->zs_enospc_count,
5379 			    100.0 * zs->zs_alloc / zs->zs_space,
5380 			    numbuf,
5381 			    100.0 * (now - zs->zs_proc_start) /
5382 			    (zopt_time * NANOSEC), timebuf);
5383 		}
5384 
5385 		if (zopt_verbose >= 2) {
5386 			(void) printf("\nWorkload summary:\n\n");
5387 			(void) printf("%7s %9s   %s\n",
5388 			    "Calls", "Time", "Function");
5389 			(void) printf("%7s %9s   %s\n",
5390 			    "-----", "----", "--------");
5391 			for (int f = 0; f < ZTEST_FUNCS; f++) {
5392 				Dl_info dli;
5393 
5394 				zi = &zs->zs_info[f];
5395 				print_time(zi->zi_call_time, timebuf);
5396 				(void) dladdr((void *)zi->zi_func, &dli);
5397 				(void) printf("%7llu %9s   %s\n",
5398 				    (u_longlong_t)zi->zi_call_count, timebuf,
5399 				    dli.dli_sname);
5400 			}
5401 			(void) printf("\n");
5402 		}
5403 
5404 		/*
5405 		 * It's possible that we killed a child during a rename test,
5406 		 * in which case we'll have a 'ztest_tmp' pool lying around
5407 		 * instead of 'ztest'.  Do a blind rename in case this happened.
5408 		 */
5409 		kernel_init(FREAD);
5410 		if (spa_open(zopt_pool, &spa, FTAG) == 0) {
5411 			spa_close(spa, FTAG);
5412 		} else {
5413 			char tmpname[MAXNAMELEN];
5414 			kernel_fini();
5415 			kernel_init(FREAD | FWRITE);
5416 			(void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
5417 			    zopt_pool);
5418 			(void) spa_rename(tmpname, zopt_pool);
5419 		}
5420 		kernel_fini();
5421 
5422 		ztest_run_zdb(zopt_pool);
5423 	}
5424 
5425 	if (zopt_verbose >= 1) {
5426 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
5427 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
5428 	}
5429 
5430 	return (0);
5431 }
5432