xref: /illumos-gate/usr/src/cmd/ztest/ztest.c (revision 5bb0bdfe588c5df0f63ff8ac292cd608a5f4492a)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2017 Joyent, Inc.
28  * Copyright 2017 RackTop Systems.
29  */
30 
31 /*
32  * The objective of this program is to provide a DMU/ZAP/SPA stress test
33  * that runs entirely in userland, is easy to use, and easy to extend.
34  *
35  * The overall design of the ztest program is as follows:
36  *
37  * (1) For each major functional area (e.g. adding vdevs to a pool,
38  *     creating and destroying datasets, reading and writing objects, etc)
39  *     we have a simple routine to test that functionality.  These
40  *     individual routines do not have to do anything "stressful".
41  *
42  * (2) We turn these simple functionality tests into a stress test by
43  *     running them all in parallel, with as many threads as desired,
44  *     and spread across as many datasets, objects, and vdevs as desired.
45  *
46  * (3) While all this is happening, we inject faults into the pool to
47  *     verify that self-healing data really works.
48  *
49  * (4) Every time we open a dataset, we change its checksum and compression
50  *     functions.  Thus even individual objects vary from block to block
51  *     in which checksum they use and whether they're compressed.
52  *
53  * (5) To verify that we never lose on-disk consistency after a crash,
54  *     we run the entire test in a child of the main process.
55  *     At random times, the child self-immolates with a SIGKILL.
56  *     This is the software equivalent of pulling the power cord.
57  *     The parent then runs the test again, using the existing
58  *     storage pool, as many times as desired. If backwards compatibility
59  *     testing is enabled ztest will sometimes run the "older" version
60  *     of ztest after a SIGKILL.
61  *
62  * (6) To verify that we don't have future leaks or temporal incursions,
63  *     many of the functional tests record the transaction group number
64  *     as part of their data.  When reading old data, they verify that
65  *     the transaction group number is less than the current, open txg.
66  *     If you add a new test, please do this if applicable.
67  *
68  * When run with no arguments, ztest runs for about five minutes and
69  * produces no output if successful.  To get a little bit of information,
70  * specify -V.  To get more information, specify -VV, and so on.
71  *
72  * To turn this into an overnight stress test, use -T to specify run time.
73  *
74  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75  * to increase the pool capacity, fanout, and overall stress level.
76  *
77  * Use the -k option to set the desired frequency of kills.
78  *
79  * When ztest invokes itself it passes all relevant information through a
80  * temporary file which is mmap-ed in the child process. This allows shared
81  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82  * stored at offset 0 of this file and contains information on the size and
83  * number of shared structures in the file. The information stored in this file
84  * must remain backwards compatible with older versions of ztest so that
85  * ztest can invoke them during backwards compatibility testing (-B).
86  */
87 
88 #include <sys/zfs_context.h>
89 #include <sys/spa.h>
90 #include <sys/dmu.h>
91 #include <sys/txg.h>
92 #include <sys/dbuf.h>
93 #include <sys/zap.h>
94 #include <sys/dmu_objset.h>
95 #include <sys/poll.h>
96 #include <sys/stat.h>
97 #include <sys/time.h>
98 #include <sys/wait.h>
99 #include <sys/mman.h>
100 #include <sys/resource.h>
101 #include <sys/zio.h>
102 #include <sys/zil.h>
103 #include <sys/zil_impl.h>
104 #include <sys/vdev_impl.h>
105 #include <sys/vdev_file.h>
106 #include <sys/vdev_initialize.h>
107 #include <sys/spa_impl.h>
108 #include <sys/metaslab_impl.h>
109 #include <sys/dsl_prop.h>
110 #include <sys/dsl_dataset.h>
111 #include <sys/dsl_destroy.h>
112 #include <sys/dsl_scan.h>
113 #include <sys/zio_checksum.h>
114 #include <sys/refcount.h>
115 #include <sys/zfeature.h>
116 #include <sys/dsl_userhold.h>
117 #include <sys/abd.h>
118 #include <stdio.h>
119 #include <stdio_ext.h>
120 #include <stdlib.h>
121 #include <unistd.h>
122 #include <signal.h>
123 #include <umem.h>
124 #include <dlfcn.h>
125 #include <ctype.h>
126 #include <math.h>
127 #include <sys/fs/zfs.h>
128 #include <libnvpair.h>
129 #include <libcmdutils.h>
130 
131 static int ztest_fd_data = -1;
132 static int ztest_fd_rand = -1;
133 
134 typedef struct ztest_shared_hdr {
135 	uint64_t	zh_hdr_size;
136 	uint64_t	zh_opts_size;
137 	uint64_t	zh_size;
138 	uint64_t	zh_stats_size;
139 	uint64_t	zh_stats_count;
140 	uint64_t	zh_ds_size;
141 	uint64_t	zh_ds_count;
142 } ztest_shared_hdr_t;
143 
144 static ztest_shared_hdr_t *ztest_shared_hdr;
145 
146 typedef struct ztest_shared_opts {
147 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
148 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
149 	char zo_alt_ztest[MAXNAMELEN];
150 	char zo_alt_libpath[MAXNAMELEN];
151 	uint64_t zo_vdevs;
152 	uint64_t zo_vdevtime;
153 	size_t zo_vdev_size;
154 	int zo_ashift;
155 	int zo_mirrors;
156 	int zo_raidz;
157 	int zo_raidz_parity;
158 	int zo_datasets;
159 	int zo_threads;
160 	uint64_t zo_passtime;
161 	uint64_t zo_killrate;
162 	int zo_verbose;
163 	int zo_init;
164 	uint64_t zo_time;
165 	uint64_t zo_maxloops;
166 	uint64_t zo_metaslab_force_ganging;
167 } ztest_shared_opts_t;
168 
169 static const ztest_shared_opts_t ztest_opts_defaults = {
170 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
171 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
172 	.zo_alt_ztest = { '\0' },
173 	.zo_alt_libpath = { '\0' },
174 	.zo_vdevs = 5,
175 	.zo_ashift = SPA_MINBLOCKSHIFT,
176 	.zo_mirrors = 2,
177 	.zo_raidz = 4,
178 	.zo_raidz_parity = 1,
179 	.zo_vdev_size = SPA_MINDEVSIZE * 4,	/* 256m default size */
180 	.zo_datasets = 7,
181 	.zo_threads = 23,
182 	.zo_passtime = 60,		/* 60 seconds */
183 	.zo_killrate = 70,		/* 70% kill rate */
184 	.zo_verbose = 0,
185 	.zo_init = 1,
186 	.zo_time = 300,			/* 5 minutes */
187 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
188 	.zo_metaslab_force_ganging = 32 << 10
189 };
190 
191 extern uint64_t metaslab_force_ganging;
192 extern uint64_t metaslab_df_alloc_threshold;
193 extern uint64_t zfs_deadman_synctime_ms;
194 extern int metaslab_preload_limit;
195 extern boolean_t zfs_compressed_arc_enabled;
196 extern boolean_t zfs_abd_scatter_enabled;
197 extern boolean_t zfs_force_some_double_word_sm_entries;
198 
199 static ztest_shared_opts_t *ztest_shared_opts;
200 static ztest_shared_opts_t ztest_opts;
201 
202 typedef struct ztest_shared_ds {
203 	uint64_t	zd_seq;
204 } ztest_shared_ds_t;
205 
206 static ztest_shared_ds_t *ztest_shared_ds;
207 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
208 
209 #define	BT_MAGIC	0x123456789abcdefULL
210 #define	MAXFAULTS() \
211 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
212 
213 enum ztest_io_type {
214 	ZTEST_IO_WRITE_TAG,
215 	ZTEST_IO_WRITE_PATTERN,
216 	ZTEST_IO_WRITE_ZEROES,
217 	ZTEST_IO_TRUNCATE,
218 	ZTEST_IO_SETATTR,
219 	ZTEST_IO_REWRITE,
220 	ZTEST_IO_TYPES
221 };
222 
223 typedef struct ztest_block_tag {
224 	uint64_t	bt_magic;
225 	uint64_t	bt_objset;
226 	uint64_t	bt_object;
227 	uint64_t	bt_offset;
228 	uint64_t	bt_gen;
229 	uint64_t	bt_txg;
230 	uint64_t	bt_crtxg;
231 } ztest_block_tag_t;
232 
233 typedef struct bufwad {
234 	uint64_t	bw_index;
235 	uint64_t	bw_txg;
236 	uint64_t	bw_data;
237 } bufwad_t;
238 
239 /*
240  * XXX -- fix zfs range locks to be generic so we can use them here.
241  */
242 typedef enum {
243 	RL_READER,
244 	RL_WRITER,
245 	RL_APPEND
246 } rl_type_t;
247 
248 typedef struct rll {
249 	void		*rll_writer;
250 	int		rll_readers;
251 	kmutex_t	rll_lock;
252 	kcondvar_t	rll_cv;
253 } rll_t;
254 
255 typedef struct rl {
256 	uint64_t	rl_object;
257 	uint64_t	rl_offset;
258 	uint64_t	rl_size;
259 	rll_t		*rl_lock;
260 } rl_t;
261 
262 #define	ZTEST_RANGE_LOCKS	64
263 #define	ZTEST_OBJECT_LOCKS	64
264 
265 /*
266  * Object descriptor.  Used as a template for object lookup/create/remove.
267  */
268 typedef struct ztest_od {
269 	uint64_t	od_dir;
270 	uint64_t	od_object;
271 	dmu_object_type_t od_type;
272 	dmu_object_type_t od_crtype;
273 	uint64_t	od_blocksize;
274 	uint64_t	od_crblocksize;
275 	uint64_t	od_gen;
276 	uint64_t	od_crgen;
277 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
278 } ztest_od_t;
279 
280 /*
281  * Per-dataset state.
282  */
283 typedef struct ztest_ds {
284 	ztest_shared_ds_t *zd_shared;
285 	objset_t	*zd_os;
286 	krwlock_t	zd_zilog_lock;
287 	zilog_t		*zd_zilog;
288 	ztest_od_t	*zd_od;		/* debugging aid */
289 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
290 	kmutex_t	zd_dirobj_lock;
291 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
292 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
293 } ztest_ds_t;
294 
295 /*
296  * Per-iteration state.
297  */
298 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
299 
300 typedef struct ztest_info {
301 	ztest_func_t	*zi_func;	/* test function */
302 	uint64_t	zi_iters;	/* iterations per execution */
303 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
304 } ztest_info_t;
305 
306 typedef struct ztest_shared_callstate {
307 	uint64_t	zc_count;	/* per-pass count */
308 	uint64_t	zc_time;	/* per-pass time */
309 	uint64_t	zc_next;	/* next time to call this function */
310 } ztest_shared_callstate_t;
311 
312 static ztest_shared_callstate_t *ztest_shared_callstate;
313 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
314 
315 /*
316  * Note: these aren't static because we want dladdr() to work.
317  */
318 ztest_func_t ztest_dmu_read_write;
319 ztest_func_t ztest_dmu_write_parallel;
320 ztest_func_t ztest_dmu_object_alloc_free;
321 ztest_func_t ztest_dmu_commit_callbacks;
322 ztest_func_t ztest_zap;
323 ztest_func_t ztest_zap_parallel;
324 ztest_func_t ztest_zil_commit;
325 ztest_func_t ztest_zil_remount;
326 ztest_func_t ztest_dmu_read_write_zcopy;
327 ztest_func_t ztest_dmu_objset_create_destroy;
328 ztest_func_t ztest_dmu_prealloc;
329 ztest_func_t ztest_fzap;
330 ztest_func_t ztest_dmu_snapshot_create_destroy;
331 ztest_func_t ztest_dsl_prop_get_set;
332 ztest_func_t ztest_spa_prop_get_set;
333 ztest_func_t ztest_spa_create_destroy;
334 ztest_func_t ztest_fault_inject;
335 ztest_func_t ztest_ddt_repair;
336 ztest_func_t ztest_dmu_snapshot_hold;
337 ztest_func_t ztest_scrub;
338 ztest_func_t ztest_dsl_dataset_promote_busy;
339 ztest_func_t ztest_vdev_attach_detach;
340 ztest_func_t ztest_vdev_LUN_growth;
341 ztest_func_t ztest_vdev_add_remove;
342 ztest_func_t ztest_vdev_aux_add_remove;
343 ztest_func_t ztest_split_pool;
344 ztest_func_t ztest_reguid;
345 ztest_func_t ztest_spa_upgrade;
346 ztest_func_t ztest_device_removal;
347 ztest_func_t ztest_remap_blocks;
348 ztest_func_t ztest_spa_checkpoint_create_discard;
349 ztest_func_t ztest_initialize;
350 
351 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
352 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
353 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
354 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
355 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
356 
357 ztest_info_t ztest_info[] = {
358 	{ ztest_dmu_read_write,			1,	&zopt_always	},
359 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
360 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
361 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
362 	{ ztest_zap,				30,	&zopt_always	},
363 	{ ztest_zap_parallel,			100,	&zopt_always	},
364 	{ ztest_split_pool,			1,	&zopt_always	},
365 	{ ztest_zil_commit,			1,	&zopt_incessant	},
366 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
367 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
368 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
369 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
370 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
371 #if 0
372 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
373 #endif
374 	{ ztest_fzap,				1,	&zopt_sometimes	},
375 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
376 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
377 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
378 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
379 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
380 	{ ztest_reguid,				1,	&zopt_rarely	},
381 	{ ztest_scrub,				1,	&zopt_rarely	},
382 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
383 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
384 	{ ztest_vdev_attach_detach,		1,	&zopt_sometimes	},
385 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
386 	{ ztest_vdev_add_remove,		1,
387 	    &ztest_opts.zo_vdevtime				},
388 	{ ztest_vdev_aux_add_remove,		1,
389 	    &ztest_opts.zo_vdevtime				},
390 	{ ztest_device_removal,			1,	&zopt_sometimes	},
391 	{ ztest_remap_blocks,			1,	&zopt_sometimes },
392 	{ ztest_spa_checkpoint_create_discard,	1,	&zopt_rarely	},
393 	{ ztest_initialize,			1,	&zopt_sometimes }
394 };
395 
396 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
397 
398 /*
399  * The following struct is used to hold a list of uncalled commit callbacks.
400  * The callbacks are ordered by txg number.
401  */
402 typedef struct ztest_cb_list {
403 	kmutex_t zcl_callbacks_lock;
404 	list_t	zcl_callbacks;
405 } ztest_cb_list_t;
406 
407 /*
408  * Stuff we need to share writably between parent and child.
409  */
410 typedef struct ztest_shared {
411 	boolean_t	zs_do_init;
412 	hrtime_t	zs_proc_start;
413 	hrtime_t	zs_proc_stop;
414 	hrtime_t	zs_thread_start;
415 	hrtime_t	zs_thread_stop;
416 	hrtime_t	zs_thread_kill;
417 	uint64_t	zs_enospc_count;
418 	uint64_t	zs_vdev_next_leaf;
419 	uint64_t	zs_vdev_aux;
420 	uint64_t	zs_alloc;
421 	uint64_t	zs_space;
422 	uint64_t	zs_splits;
423 	uint64_t	zs_mirrors;
424 	uint64_t	zs_metaslab_sz;
425 	uint64_t	zs_metaslab_df_alloc_threshold;
426 	uint64_t	zs_guid;
427 } ztest_shared_t;
428 
429 #define	ID_PARALLEL	-1ULL
430 
431 static char ztest_dev_template[] = "%s/%s.%llua";
432 static char ztest_aux_template[] = "%s/%s.%s.%llu";
433 ztest_shared_t *ztest_shared;
434 
435 static spa_t *ztest_spa = NULL;
436 static ztest_ds_t *ztest_ds;
437 
438 static kmutex_t ztest_vdev_lock;
439 static kmutex_t ztest_checkpoint_lock;
440 static boolean_t ztest_device_removal_active = B_FALSE;
441 
442 /*
443  * The ztest_name_lock protects the pool and dataset namespace used by
444  * the individual tests. To modify the namespace, consumers must grab
445  * this lock as writer. Grabbing the lock as reader will ensure that the
446  * namespace does not change while the lock is held.
447  */
448 static krwlock_t ztest_name_lock;
449 
450 static boolean_t ztest_dump_core = B_TRUE;
451 static boolean_t ztest_exiting;
452 
453 /* Global commit callback list */
454 static ztest_cb_list_t zcl;
455 
456 enum ztest_object {
457 	ZTEST_META_DNODE = 0,
458 	ZTEST_DIROBJ,
459 	ZTEST_OBJECTS
460 };
461 
462 static void usage(boolean_t) __NORETURN;
463 
464 /*
465  * These libumem hooks provide a reasonable set of defaults for the allocator's
466  * debugging facilities.
467  */
468 const char *
469 _umem_debug_init()
470 {
471 	return ("default,verbose"); /* $UMEM_DEBUG setting */
472 }
473 
474 const char *
475 _umem_logging_init(void)
476 {
477 	return ("fail,contents"); /* $UMEM_LOGGING setting */
478 }
479 
480 #define	FATAL_MSG_SZ	1024
481 
482 char *fatal_msg;
483 
484 static void
485 fatal(int do_perror, char *message, ...)
486 {
487 	va_list args;
488 	int save_errno = errno;
489 	char buf[FATAL_MSG_SZ];
490 
491 	(void) fflush(stdout);
492 
493 	va_start(args, message);
494 	(void) sprintf(buf, "ztest: ");
495 	/* LINTED */
496 	(void) vsprintf(buf + strlen(buf), message, args);
497 	va_end(args);
498 	if (do_perror) {
499 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
500 		    ": %s", strerror(save_errno));
501 	}
502 	(void) fprintf(stderr, "%s\n", buf);
503 	fatal_msg = buf;			/* to ease debugging */
504 	if (ztest_dump_core)
505 		abort();
506 	exit(3);
507 }
508 
509 static int
510 str2shift(const char *buf)
511 {
512 	const char *ends = "BKMGTPEZ";
513 	int i;
514 
515 	if (buf[0] == '\0')
516 		return (0);
517 	for (i = 0; i < strlen(ends); i++) {
518 		if (toupper(buf[0]) == ends[i])
519 			break;
520 	}
521 	if (i == strlen(ends)) {
522 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
523 		    buf);
524 		usage(B_FALSE);
525 	}
526 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
527 		return (10*i);
528 	}
529 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
530 	usage(B_FALSE);
531 	/* NOTREACHED */
532 }
533 
534 static uint64_t
535 nicenumtoull(const char *buf)
536 {
537 	char *end;
538 	uint64_t val;
539 
540 	val = strtoull(buf, &end, 0);
541 	if (end == buf) {
542 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
543 		usage(B_FALSE);
544 	} else if (end[0] == '.') {
545 		double fval = strtod(buf, &end);
546 		fval *= pow(2, str2shift(end));
547 		if (fval > UINT64_MAX) {
548 			(void) fprintf(stderr, "ztest: value too large: %s\n",
549 			    buf);
550 			usage(B_FALSE);
551 		}
552 		val = (uint64_t)fval;
553 	} else {
554 		int shift = str2shift(end);
555 		if (shift >= 64 || (val << shift) >> shift != val) {
556 			(void) fprintf(stderr, "ztest: value too large: %s\n",
557 			    buf);
558 			usage(B_FALSE);
559 		}
560 		val <<= shift;
561 	}
562 	return (val);
563 }
564 
565 static void
566 usage(boolean_t requested)
567 {
568 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
569 
570 	char nice_vdev_size[NN_NUMBUF_SZ];
571 	char nice_force_ganging[NN_NUMBUF_SZ];
572 	FILE *fp = requested ? stdout : stderr;
573 
574 	nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
575 	nicenum(zo->zo_metaslab_force_ganging, nice_force_ganging,
576 	    sizeof (nice_force_ganging));
577 
578 	(void) fprintf(fp, "Usage: %s\n"
579 	    "\t[-v vdevs (default: %llu)]\n"
580 	    "\t[-s size_of_each_vdev (default: %s)]\n"
581 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
582 	    "\t[-m mirror_copies (default: %d)]\n"
583 	    "\t[-r raidz_disks (default: %d)]\n"
584 	    "\t[-R raidz_parity (default: %d)]\n"
585 	    "\t[-d datasets (default: %d)]\n"
586 	    "\t[-t threads (default: %d)]\n"
587 	    "\t[-g gang_block_threshold (default: %s)]\n"
588 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
589 	    "\t[-k kill_percentage (default: %llu%%)]\n"
590 	    "\t[-p pool_name (default: %s)]\n"
591 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
592 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
593 	    "\t[-E] use existing pool instead of creating new one\n"
594 	    "\t[-T time (default: %llu sec)] total run time\n"
595 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
596 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
597 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
598 	    "\t[-o variable=value] ... set global variable to an unsigned\n"
599 	    "\t    32-bit integer value\n"
600 	    "\t[-h] (print help)\n"
601 	    "",
602 	    zo->zo_pool,
603 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
604 	    nice_vdev_size,				/* -s */
605 	    zo->zo_ashift,				/* -a */
606 	    zo->zo_mirrors,				/* -m */
607 	    zo->zo_raidz,				/* -r */
608 	    zo->zo_raidz_parity,			/* -R */
609 	    zo->zo_datasets,				/* -d */
610 	    zo->zo_threads,				/* -t */
611 	    nice_force_ganging,				/* -g */
612 	    zo->zo_init,				/* -i */
613 	    (u_longlong_t)zo->zo_killrate,		/* -k */
614 	    zo->zo_pool,				/* -p */
615 	    zo->zo_dir,					/* -f */
616 	    (u_longlong_t)zo->zo_time,			/* -T */
617 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
618 	    (u_longlong_t)zo->zo_passtime);
619 	exit(requested ? 0 : 1);
620 }
621 
622 static void
623 process_options(int argc, char **argv)
624 {
625 	char *path;
626 	ztest_shared_opts_t *zo = &ztest_opts;
627 
628 	int opt;
629 	uint64_t value;
630 	char altdir[MAXNAMELEN] = { 0 };
631 
632 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
633 
634 	while ((opt = getopt(argc, argv,
635 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) {
636 		value = 0;
637 		switch (opt) {
638 		case 'v':
639 		case 's':
640 		case 'a':
641 		case 'm':
642 		case 'r':
643 		case 'R':
644 		case 'd':
645 		case 't':
646 		case 'g':
647 		case 'i':
648 		case 'k':
649 		case 'T':
650 		case 'P':
651 		case 'F':
652 			value = nicenumtoull(optarg);
653 		}
654 		switch (opt) {
655 		case 'v':
656 			zo->zo_vdevs = value;
657 			break;
658 		case 's':
659 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
660 			break;
661 		case 'a':
662 			zo->zo_ashift = value;
663 			break;
664 		case 'm':
665 			zo->zo_mirrors = value;
666 			break;
667 		case 'r':
668 			zo->zo_raidz = MAX(1, value);
669 			break;
670 		case 'R':
671 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
672 			break;
673 		case 'd':
674 			zo->zo_datasets = MAX(1, value);
675 			break;
676 		case 't':
677 			zo->zo_threads = MAX(1, value);
678 			break;
679 		case 'g':
680 			zo->zo_metaslab_force_ganging =
681 			    MAX(SPA_MINBLOCKSIZE << 1, value);
682 			break;
683 		case 'i':
684 			zo->zo_init = value;
685 			break;
686 		case 'k':
687 			zo->zo_killrate = value;
688 			break;
689 		case 'p':
690 			(void) strlcpy(zo->zo_pool, optarg,
691 			    sizeof (zo->zo_pool));
692 			break;
693 		case 'f':
694 			path = realpath(optarg, NULL);
695 			if (path == NULL) {
696 				(void) fprintf(stderr, "error: %s: %s\n",
697 				    optarg, strerror(errno));
698 				usage(B_FALSE);
699 			} else {
700 				(void) strlcpy(zo->zo_dir, path,
701 				    sizeof (zo->zo_dir));
702 			}
703 			break;
704 		case 'V':
705 			zo->zo_verbose++;
706 			break;
707 		case 'E':
708 			zo->zo_init = 0;
709 			break;
710 		case 'T':
711 			zo->zo_time = value;
712 			break;
713 		case 'P':
714 			zo->zo_passtime = MAX(1, value);
715 			break;
716 		case 'F':
717 			zo->zo_maxloops = MAX(1, value);
718 			break;
719 		case 'B':
720 			(void) strlcpy(altdir, optarg, sizeof (altdir));
721 			break;
722 		case 'o':
723 			if (set_global_var(optarg) != 0)
724 				usage(B_FALSE);
725 			break;
726 		case 'h':
727 			usage(B_TRUE);
728 			break;
729 		case '?':
730 		default:
731 			usage(B_FALSE);
732 			break;
733 		}
734 	}
735 
736 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
737 
738 	zo->zo_vdevtime =
739 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
740 	    UINT64_MAX >> 2);
741 
742 	if (strlen(altdir) > 0) {
743 		char *cmd;
744 		char *realaltdir;
745 		char *bin;
746 		char *ztest;
747 		char *isa;
748 		int isalen;
749 
750 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
751 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
752 
753 		VERIFY(NULL != realpath(getexecname(), cmd));
754 		if (0 != access(altdir, F_OK)) {
755 			ztest_dump_core = B_FALSE;
756 			fatal(B_TRUE, "invalid alternate ztest path: %s",
757 			    altdir);
758 		}
759 		VERIFY(NULL != realpath(altdir, realaltdir));
760 
761 		/*
762 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
763 		 * We want to extract <isa> to determine if we should use
764 		 * 32 or 64 bit binaries.
765 		 */
766 		bin = strstr(cmd, "/usr/bin/");
767 		ztest = strstr(bin, "/ztest");
768 		isa = bin + 9;
769 		isalen = ztest - isa;
770 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
771 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
772 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
773 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
774 
775 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
776 			ztest_dump_core = B_FALSE;
777 			fatal(B_TRUE, "invalid alternate ztest: %s",
778 			    zo->zo_alt_ztest);
779 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
780 			ztest_dump_core = B_FALSE;
781 			fatal(B_TRUE, "invalid alternate lib directory %s",
782 			    zo->zo_alt_libpath);
783 		}
784 
785 		umem_free(cmd, MAXPATHLEN);
786 		umem_free(realaltdir, MAXPATHLEN);
787 	}
788 }
789 
790 static void
791 ztest_kill(ztest_shared_t *zs)
792 {
793 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
794 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
795 
796 	/*
797 	 * Before we kill off ztest, make sure that the config is updated.
798 	 * See comment above spa_write_cachefile().
799 	 */
800 	mutex_enter(&spa_namespace_lock);
801 	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
802 	mutex_exit(&spa_namespace_lock);
803 
804 	zfs_dbgmsg_print(FTAG);
805 	(void) kill(getpid(), SIGKILL);
806 }
807 
808 static uint64_t
809 ztest_random(uint64_t range)
810 {
811 	uint64_t r;
812 
813 	ASSERT3S(ztest_fd_rand, >=, 0);
814 
815 	if (range == 0)
816 		return (0);
817 
818 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
819 		fatal(1, "short read from /dev/urandom");
820 
821 	return (r % range);
822 }
823 
824 /* ARGSUSED */
825 static void
826 ztest_record_enospc(const char *s)
827 {
828 	ztest_shared->zs_enospc_count++;
829 }
830 
831 static uint64_t
832 ztest_get_ashift(void)
833 {
834 	if (ztest_opts.zo_ashift == 0)
835 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
836 	return (ztest_opts.zo_ashift);
837 }
838 
839 static nvlist_t *
840 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
841 {
842 	char pathbuf[MAXPATHLEN];
843 	uint64_t vdev;
844 	nvlist_t *file;
845 
846 	if (ashift == 0)
847 		ashift = ztest_get_ashift();
848 
849 	if (path == NULL) {
850 		path = pathbuf;
851 
852 		if (aux != NULL) {
853 			vdev = ztest_shared->zs_vdev_aux;
854 			(void) snprintf(path, sizeof (pathbuf),
855 			    ztest_aux_template, ztest_opts.zo_dir,
856 			    pool == NULL ? ztest_opts.zo_pool : pool,
857 			    aux, vdev);
858 		} else {
859 			vdev = ztest_shared->zs_vdev_next_leaf++;
860 			(void) snprintf(path, sizeof (pathbuf),
861 			    ztest_dev_template, ztest_opts.zo_dir,
862 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
863 		}
864 	}
865 
866 	if (size != 0) {
867 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
868 		if (fd == -1)
869 			fatal(1, "can't open %s", path);
870 		if (ftruncate(fd, size) != 0)
871 			fatal(1, "can't ftruncate %s", path);
872 		(void) close(fd);
873 	}
874 
875 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
876 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
877 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
878 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
879 
880 	return (file);
881 }
882 
883 static nvlist_t *
884 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
885     uint64_t ashift, int r)
886 {
887 	nvlist_t *raidz, **child;
888 	int c;
889 
890 	if (r < 2)
891 		return (make_vdev_file(path, aux, pool, size, ashift));
892 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
893 
894 	for (c = 0; c < r; c++)
895 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
896 
897 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
898 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
899 	    VDEV_TYPE_RAIDZ) == 0);
900 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
901 	    ztest_opts.zo_raidz_parity) == 0);
902 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
903 	    child, r) == 0);
904 
905 	for (c = 0; c < r; c++)
906 		nvlist_free(child[c]);
907 
908 	umem_free(child, r * sizeof (nvlist_t *));
909 
910 	return (raidz);
911 }
912 
913 static nvlist_t *
914 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
915     uint64_t ashift, int r, int m)
916 {
917 	nvlist_t *mirror, **child;
918 	int c;
919 
920 	if (m < 1)
921 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
922 
923 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
924 
925 	for (c = 0; c < m; c++)
926 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
927 
928 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
929 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
930 	    VDEV_TYPE_MIRROR) == 0);
931 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
932 	    child, m) == 0);
933 
934 	for (c = 0; c < m; c++)
935 		nvlist_free(child[c]);
936 
937 	umem_free(child, m * sizeof (nvlist_t *));
938 
939 	return (mirror);
940 }
941 
942 static nvlist_t *
943 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
944     int log, int r, int m, int t)
945 {
946 	nvlist_t *root, **child;
947 	int c;
948 
949 	ASSERT(t > 0);
950 
951 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
952 
953 	for (c = 0; c < t; c++) {
954 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
955 		    r, m);
956 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
957 		    log) == 0);
958 	}
959 
960 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
961 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
962 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
963 	    child, t) == 0);
964 
965 	for (c = 0; c < t; c++)
966 		nvlist_free(child[c]);
967 
968 	umem_free(child, t * sizeof (nvlist_t *));
969 
970 	return (root);
971 }
972 
973 /*
974  * Find a random spa version. Returns back a random spa version in the
975  * range [initial_version, SPA_VERSION_FEATURES].
976  */
977 static uint64_t
978 ztest_random_spa_version(uint64_t initial_version)
979 {
980 	uint64_t version = initial_version;
981 
982 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
983 		version = version +
984 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
985 	}
986 
987 	if (version > SPA_VERSION_BEFORE_FEATURES)
988 		version = SPA_VERSION_FEATURES;
989 
990 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
991 	return (version);
992 }
993 
994 static int
995 ztest_random_blocksize(void)
996 {
997 	uint64_t block_shift;
998 	/*
999 	 * Choose a block size >= the ashift.
1000 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1001 	 */
1002 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1003 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1004 		maxbs = 20;
1005 	block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1006 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1007 }
1008 
1009 static int
1010 ztest_random_ibshift(void)
1011 {
1012 	return (DN_MIN_INDBLKSHIFT +
1013 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1014 }
1015 
1016 static uint64_t
1017 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1018 {
1019 	uint64_t top;
1020 	vdev_t *rvd = spa->spa_root_vdev;
1021 	vdev_t *tvd;
1022 
1023 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1024 
1025 	do {
1026 		top = ztest_random(rvd->vdev_children);
1027 		tvd = rvd->vdev_child[top];
1028 	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1029 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1030 
1031 	return (top);
1032 }
1033 
1034 static uint64_t
1035 ztest_random_dsl_prop(zfs_prop_t prop)
1036 {
1037 	uint64_t value;
1038 
1039 	do {
1040 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1041 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1042 
1043 	return (value);
1044 }
1045 
1046 static int
1047 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1048     boolean_t inherit)
1049 {
1050 	const char *propname = zfs_prop_to_name(prop);
1051 	const char *valname;
1052 	char setpoint[MAXPATHLEN];
1053 	uint64_t curval;
1054 	int error;
1055 
1056 	error = dsl_prop_set_int(osname, propname,
1057 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1058 
1059 	if (error == ENOSPC) {
1060 		ztest_record_enospc(FTAG);
1061 		return (error);
1062 	}
1063 	ASSERT0(error);
1064 
1065 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1066 
1067 	if (ztest_opts.zo_verbose >= 6) {
1068 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1069 		(void) printf("%s %s = %s at '%s'\n",
1070 		    osname, propname, valname, setpoint);
1071 	}
1072 
1073 	return (error);
1074 }
1075 
1076 static int
1077 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1078 {
1079 	spa_t *spa = ztest_spa;
1080 	nvlist_t *props = NULL;
1081 	int error;
1082 
1083 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1084 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1085 
1086 	error = spa_prop_set(spa, props);
1087 
1088 	nvlist_free(props);
1089 
1090 	if (error == ENOSPC) {
1091 		ztest_record_enospc(FTAG);
1092 		return (error);
1093 	}
1094 	ASSERT0(error);
1095 
1096 	return (error);
1097 }
1098 
1099 static void
1100 ztest_rll_init(rll_t *rll)
1101 {
1102 	rll->rll_writer = NULL;
1103 	rll->rll_readers = 0;
1104 	mutex_init(&rll->rll_lock, NULL, USYNC_THREAD, NULL);
1105 	cv_init(&rll->rll_cv, NULL, USYNC_THREAD, NULL);
1106 }
1107 
1108 static void
1109 ztest_rll_destroy(rll_t *rll)
1110 {
1111 	ASSERT(rll->rll_writer == NULL);
1112 	ASSERT(rll->rll_readers == 0);
1113 	mutex_destroy(&rll->rll_lock);
1114 	cv_destroy(&rll->rll_cv);
1115 }
1116 
1117 static void
1118 ztest_rll_lock(rll_t *rll, rl_type_t type)
1119 {
1120 	mutex_enter(&rll->rll_lock);
1121 
1122 	if (type == RL_READER) {
1123 		while (rll->rll_writer != NULL)
1124 			cv_wait(&rll->rll_cv, &rll->rll_lock);
1125 		rll->rll_readers++;
1126 	} else {
1127 		while (rll->rll_writer != NULL || rll->rll_readers)
1128 			cv_wait(&rll->rll_cv, &rll->rll_lock);
1129 		rll->rll_writer = curthread;
1130 	}
1131 
1132 	mutex_exit(&rll->rll_lock);
1133 }
1134 
1135 static void
1136 ztest_rll_unlock(rll_t *rll)
1137 {
1138 	mutex_enter(&rll->rll_lock);
1139 
1140 	if (rll->rll_writer) {
1141 		ASSERT(rll->rll_readers == 0);
1142 		rll->rll_writer = NULL;
1143 	} else {
1144 		ASSERT(rll->rll_readers != 0);
1145 		ASSERT(rll->rll_writer == NULL);
1146 		rll->rll_readers--;
1147 	}
1148 
1149 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1150 		cv_broadcast(&rll->rll_cv);
1151 
1152 	mutex_exit(&rll->rll_lock);
1153 }
1154 
1155 static void
1156 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1157 {
1158 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1159 
1160 	ztest_rll_lock(rll, type);
1161 }
1162 
1163 static void
1164 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1165 {
1166 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1167 
1168 	ztest_rll_unlock(rll);
1169 }
1170 
1171 static rl_t *
1172 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1173     uint64_t size, rl_type_t type)
1174 {
1175 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1176 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1177 	rl_t *rl;
1178 
1179 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1180 	rl->rl_object = object;
1181 	rl->rl_offset = offset;
1182 	rl->rl_size = size;
1183 	rl->rl_lock = rll;
1184 
1185 	ztest_rll_lock(rll, type);
1186 
1187 	return (rl);
1188 }
1189 
1190 static void
1191 ztest_range_unlock(rl_t *rl)
1192 {
1193 	rll_t *rll = rl->rl_lock;
1194 
1195 	ztest_rll_unlock(rll);
1196 
1197 	umem_free(rl, sizeof (*rl));
1198 }
1199 
1200 static void
1201 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1202 {
1203 	zd->zd_os = os;
1204 	zd->zd_zilog = dmu_objset_zil(os);
1205 	zd->zd_shared = szd;
1206 	dmu_objset_name(os, zd->zd_name);
1207 
1208 	if (zd->zd_shared != NULL)
1209 		zd->zd_shared->zd_seq = 0;
1210 
1211 	rw_init(&zd->zd_zilog_lock, NULL, USYNC_THREAD, NULL);
1212 	mutex_init(&zd->zd_dirobj_lock, NULL, USYNC_THREAD, NULL);
1213 
1214 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1215 		ztest_rll_init(&zd->zd_object_lock[l]);
1216 
1217 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1218 		ztest_rll_init(&zd->zd_range_lock[l]);
1219 }
1220 
1221 static void
1222 ztest_zd_fini(ztest_ds_t *zd)
1223 {
1224 	mutex_destroy(&zd->zd_dirobj_lock);
1225 
1226 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1227 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1228 
1229 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1230 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1231 }
1232 
1233 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1234 
1235 static uint64_t
1236 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1237 {
1238 	uint64_t txg;
1239 	int error;
1240 
1241 	/*
1242 	 * Attempt to assign tx to some transaction group.
1243 	 */
1244 	error = dmu_tx_assign(tx, txg_how);
1245 	if (error) {
1246 		if (error == ERESTART) {
1247 			ASSERT(txg_how == TXG_NOWAIT);
1248 			dmu_tx_wait(tx);
1249 		} else {
1250 			ASSERT3U(error, ==, ENOSPC);
1251 			ztest_record_enospc(tag);
1252 		}
1253 		dmu_tx_abort(tx);
1254 		return (0);
1255 	}
1256 	txg = dmu_tx_get_txg(tx);
1257 	ASSERT(txg != 0);
1258 	return (txg);
1259 }
1260 
1261 static void
1262 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1263 {
1264 	uint64_t *ip = buf;
1265 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1266 
1267 	while (ip < ip_end)
1268 		*ip++ = value;
1269 }
1270 
1271 static boolean_t
1272 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1273 {
1274 	uint64_t *ip = buf;
1275 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1276 	uint64_t diff = 0;
1277 
1278 	while (ip < ip_end)
1279 		diff |= (value - *ip++);
1280 
1281 	return (diff == 0);
1282 }
1283 
1284 static void
1285 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1286     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1287 {
1288 	bt->bt_magic = BT_MAGIC;
1289 	bt->bt_objset = dmu_objset_id(os);
1290 	bt->bt_object = object;
1291 	bt->bt_offset = offset;
1292 	bt->bt_gen = gen;
1293 	bt->bt_txg = txg;
1294 	bt->bt_crtxg = crtxg;
1295 }
1296 
1297 static void
1298 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1299     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1300 {
1301 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1302 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1303 	ASSERT3U(bt->bt_object, ==, object);
1304 	ASSERT3U(bt->bt_offset, ==, offset);
1305 	ASSERT3U(bt->bt_gen, <=, gen);
1306 	ASSERT3U(bt->bt_txg, <=, txg);
1307 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1308 }
1309 
1310 static ztest_block_tag_t *
1311 ztest_bt_bonus(dmu_buf_t *db)
1312 {
1313 	dmu_object_info_t doi;
1314 	ztest_block_tag_t *bt;
1315 
1316 	dmu_object_info_from_db(db, &doi);
1317 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1318 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1319 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1320 
1321 	return (bt);
1322 }
1323 
1324 /*
1325  * ZIL logging ops
1326  */
1327 
1328 #define	lrz_type	lr_mode
1329 #define	lrz_blocksize	lr_uid
1330 #define	lrz_ibshift	lr_gid
1331 #define	lrz_bonustype	lr_rdev
1332 #define	lrz_bonuslen	lr_crtime[1]
1333 
1334 static void
1335 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1336 {
1337 	char *name = (void *)(lr + 1);		/* name follows lr */
1338 	size_t namesize = strlen(name) + 1;
1339 	itx_t *itx;
1340 
1341 	if (zil_replaying(zd->zd_zilog, tx))
1342 		return;
1343 
1344 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1345 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1346 	    sizeof (*lr) + namesize - sizeof (lr_t));
1347 
1348 	zil_itx_assign(zd->zd_zilog, itx, tx);
1349 }
1350 
1351 static void
1352 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1353 {
1354 	char *name = (void *)(lr + 1);		/* name follows lr */
1355 	size_t namesize = strlen(name) + 1;
1356 	itx_t *itx;
1357 
1358 	if (zil_replaying(zd->zd_zilog, tx))
1359 		return;
1360 
1361 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1362 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1363 	    sizeof (*lr) + namesize - sizeof (lr_t));
1364 
1365 	itx->itx_oid = object;
1366 	zil_itx_assign(zd->zd_zilog, itx, tx);
1367 }
1368 
1369 static void
1370 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1371 {
1372 	itx_t *itx;
1373 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1374 
1375 	if (zil_replaying(zd->zd_zilog, tx))
1376 		return;
1377 
1378 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1379 		write_state = WR_INDIRECT;
1380 
1381 	itx = zil_itx_create(TX_WRITE,
1382 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1383 
1384 	if (write_state == WR_COPIED &&
1385 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1386 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1387 		zil_itx_destroy(itx);
1388 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1389 		write_state = WR_NEED_COPY;
1390 	}
1391 	itx->itx_private = zd;
1392 	itx->itx_wr_state = write_state;
1393 	itx->itx_sync = (ztest_random(8) == 0);
1394 
1395 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1396 	    sizeof (*lr) - sizeof (lr_t));
1397 
1398 	zil_itx_assign(zd->zd_zilog, itx, tx);
1399 }
1400 
1401 static void
1402 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1403 {
1404 	itx_t *itx;
1405 
1406 	if (zil_replaying(zd->zd_zilog, tx))
1407 		return;
1408 
1409 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1410 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1411 	    sizeof (*lr) - sizeof (lr_t));
1412 
1413 	itx->itx_sync = B_FALSE;
1414 	zil_itx_assign(zd->zd_zilog, itx, tx);
1415 }
1416 
1417 static void
1418 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1419 {
1420 	itx_t *itx;
1421 
1422 	if (zil_replaying(zd->zd_zilog, tx))
1423 		return;
1424 
1425 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1426 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1427 	    sizeof (*lr) - sizeof (lr_t));
1428 
1429 	itx->itx_sync = B_FALSE;
1430 	zil_itx_assign(zd->zd_zilog, itx, tx);
1431 }
1432 
1433 /*
1434  * ZIL replay ops
1435  */
1436 static int
1437 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1438 {
1439 	ztest_ds_t *zd = arg1;
1440 	lr_create_t *lr = arg2;
1441 	char *name = (void *)(lr + 1);		/* name follows lr */
1442 	objset_t *os = zd->zd_os;
1443 	ztest_block_tag_t *bbt;
1444 	dmu_buf_t *db;
1445 	dmu_tx_t *tx;
1446 	uint64_t txg;
1447 	int error = 0;
1448 
1449 	if (byteswap)
1450 		byteswap_uint64_array(lr, sizeof (*lr));
1451 
1452 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1453 	ASSERT(name[0] != '\0');
1454 
1455 	tx = dmu_tx_create(os);
1456 
1457 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1458 
1459 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1460 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1461 	} else {
1462 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1463 	}
1464 
1465 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1466 	if (txg == 0)
1467 		return (ENOSPC);
1468 
1469 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1470 
1471 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1472 		if (lr->lr_foid == 0) {
1473 			lr->lr_foid = zap_create(os,
1474 			    lr->lrz_type, lr->lrz_bonustype,
1475 			    lr->lrz_bonuslen, tx);
1476 		} else {
1477 			error = zap_create_claim(os, lr->lr_foid,
1478 			    lr->lrz_type, lr->lrz_bonustype,
1479 			    lr->lrz_bonuslen, tx);
1480 		}
1481 	} else {
1482 		if (lr->lr_foid == 0) {
1483 			lr->lr_foid = dmu_object_alloc(os,
1484 			    lr->lrz_type, 0, lr->lrz_bonustype,
1485 			    lr->lrz_bonuslen, tx);
1486 		} else {
1487 			error = dmu_object_claim(os, lr->lr_foid,
1488 			    lr->lrz_type, 0, lr->lrz_bonustype,
1489 			    lr->lrz_bonuslen, tx);
1490 		}
1491 	}
1492 
1493 	if (error) {
1494 		ASSERT3U(error, ==, EEXIST);
1495 		ASSERT(zd->zd_zilog->zl_replay);
1496 		dmu_tx_commit(tx);
1497 		return (error);
1498 	}
1499 
1500 	ASSERT(lr->lr_foid != 0);
1501 
1502 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1503 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1504 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1505 
1506 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1507 	bbt = ztest_bt_bonus(db);
1508 	dmu_buf_will_dirty(db, tx);
1509 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1510 	dmu_buf_rele(db, FTAG);
1511 
1512 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1513 	    &lr->lr_foid, tx));
1514 
1515 	(void) ztest_log_create(zd, tx, lr);
1516 
1517 	dmu_tx_commit(tx);
1518 
1519 	return (0);
1520 }
1521 
1522 static int
1523 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
1524 {
1525 	ztest_ds_t *zd = arg1;
1526 	lr_remove_t *lr = arg2;
1527 	char *name = (void *)(lr + 1);		/* name follows lr */
1528 	objset_t *os = zd->zd_os;
1529 	dmu_object_info_t doi;
1530 	dmu_tx_t *tx;
1531 	uint64_t object, txg;
1532 
1533 	if (byteswap)
1534 		byteswap_uint64_array(lr, sizeof (*lr));
1535 
1536 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1537 	ASSERT(name[0] != '\0');
1538 
1539 	VERIFY3U(0, ==,
1540 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1541 	ASSERT(object != 0);
1542 
1543 	ztest_object_lock(zd, object, RL_WRITER);
1544 
1545 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1546 
1547 	tx = dmu_tx_create(os);
1548 
1549 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1550 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1551 
1552 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1553 	if (txg == 0) {
1554 		ztest_object_unlock(zd, object);
1555 		return (ENOSPC);
1556 	}
1557 
1558 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1559 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1560 	} else {
1561 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1562 	}
1563 
1564 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1565 
1566 	(void) ztest_log_remove(zd, tx, lr, object);
1567 
1568 	dmu_tx_commit(tx);
1569 
1570 	ztest_object_unlock(zd, object);
1571 
1572 	return (0);
1573 }
1574 
1575 static int
1576 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
1577 {
1578 	ztest_ds_t *zd = arg1;
1579 	lr_write_t *lr = arg2;
1580 	objset_t *os = zd->zd_os;
1581 	void *data = lr + 1;			/* data follows lr */
1582 	uint64_t offset, length;
1583 	ztest_block_tag_t *bt = data;
1584 	ztest_block_tag_t *bbt;
1585 	uint64_t gen, txg, lrtxg, crtxg;
1586 	dmu_object_info_t doi;
1587 	dmu_tx_t *tx;
1588 	dmu_buf_t *db;
1589 	arc_buf_t *abuf = NULL;
1590 	rl_t *rl;
1591 
1592 	if (byteswap)
1593 		byteswap_uint64_array(lr, sizeof (*lr));
1594 
1595 	offset = lr->lr_offset;
1596 	length = lr->lr_length;
1597 
1598 	/* If it's a dmu_sync() block, write the whole block */
1599 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1600 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1601 		if (length < blocksize) {
1602 			offset -= offset % blocksize;
1603 			length = blocksize;
1604 		}
1605 	}
1606 
1607 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1608 		byteswap_uint64_array(bt, sizeof (*bt));
1609 
1610 	if (bt->bt_magic != BT_MAGIC)
1611 		bt = NULL;
1612 
1613 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1614 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1615 
1616 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1617 
1618 	dmu_object_info_from_db(db, &doi);
1619 
1620 	bbt = ztest_bt_bonus(db);
1621 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1622 	gen = bbt->bt_gen;
1623 	crtxg = bbt->bt_crtxg;
1624 	lrtxg = lr->lr_common.lrc_txg;
1625 
1626 	tx = dmu_tx_create(os);
1627 
1628 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1629 
1630 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1631 	    P2PHASE(offset, length) == 0)
1632 		abuf = dmu_request_arcbuf(db, length);
1633 
1634 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1635 	if (txg == 0) {
1636 		if (abuf != NULL)
1637 			dmu_return_arcbuf(abuf);
1638 		dmu_buf_rele(db, FTAG);
1639 		ztest_range_unlock(rl);
1640 		ztest_object_unlock(zd, lr->lr_foid);
1641 		return (ENOSPC);
1642 	}
1643 
1644 	if (bt != NULL) {
1645 		/*
1646 		 * Usually, verify the old data before writing new data --
1647 		 * but not always, because we also want to verify correct
1648 		 * behavior when the data was not recently read into cache.
1649 		 */
1650 		ASSERT(offset % doi.doi_data_block_size == 0);
1651 		if (ztest_random(4) != 0) {
1652 			int prefetch = ztest_random(2) ?
1653 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1654 			ztest_block_tag_t rbt;
1655 
1656 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1657 			    sizeof (rbt), &rbt, prefetch) == 0);
1658 			if (rbt.bt_magic == BT_MAGIC) {
1659 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1660 				    offset, gen, txg, crtxg);
1661 			}
1662 		}
1663 
1664 		/*
1665 		 * Writes can appear to be newer than the bonus buffer because
1666 		 * the ztest_get_data() callback does a dmu_read() of the
1667 		 * open-context data, which may be different than the data
1668 		 * as it was when the write was generated.
1669 		 */
1670 		if (zd->zd_zilog->zl_replay) {
1671 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1672 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1673 			    bt->bt_crtxg);
1674 		}
1675 
1676 		/*
1677 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1678 		 * so that all of the usual ASSERTs will work.
1679 		 */
1680 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1681 	}
1682 
1683 	if (abuf == NULL) {
1684 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1685 	} else {
1686 		bcopy(data, abuf->b_data, length);
1687 		dmu_assign_arcbuf(db, offset, abuf, tx);
1688 	}
1689 
1690 	(void) ztest_log_write(zd, tx, lr);
1691 
1692 	dmu_buf_rele(db, FTAG);
1693 
1694 	dmu_tx_commit(tx);
1695 
1696 	ztest_range_unlock(rl);
1697 	ztest_object_unlock(zd, lr->lr_foid);
1698 
1699 	return (0);
1700 }
1701 
1702 static int
1703 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
1704 {
1705 	ztest_ds_t *zd = arg1;
1706 	lr_truncate_t *lr = arg2;
1707 	objset_t *os = zd->zd_os;
1708 	dmu_tx_t *tx;
1709 	uint64_t txg;
1710 	rl_t *rl;
1711 
1712 	if (byteswap)
1713 		byteswap_uint64_array(lr, sizeof (*lr));
1714 
1715 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1716 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1717 	    RL_WRITER);
1718 
1719 	tx = dmu_tx_create(os);
1720 
1721 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1722 
1723 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1724 	if (txg == 0) {
1725 		ztest_range_unlock(rl);
1726 		ztest_object_unlock(zd, lr->lr_foid);
1727 		return (ENOSPC);
1728 	}
1729 
1730 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1731 	    lr->lr_length, tx) == 0);
1732 
1733 	(void) ztest_log_truncate(zd, tx, lr);
1734 
1735 	dmu_tx_commit(tx);
1736 
1737 	ztest_range_unlock(rl);
1738 	ztest_object_unlock(zd, lr->lr_foid);
1739 
1740 	return (0);
1741 }
1742 
1743 static int
1744 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
1745 {
1746 	ztest_ds_t *zd = arg1;
1747 	lr_setattr_t *lr = arg2;
1748 	objset_t *os = zd->zd_os;
1749 	dmu_tx_t *tx;
1750 	dmu_buf_t *db;
1751 	ztest_block_tag_t *bbt;
1752 	uint64_t txg, lrtxg, crtxg;
1753 
1754 	if (byteswap)
1755 		byteswap_uint64_array(lr, sizeof (*lr));
1756 
1757 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1758 
1759 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1760 
1761 	tx = dmu_tx_create(os);
1762 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1763 
1764 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1765 	if (txg == 0) {
1766 		dmu_buf_rele(db, FTAG);
1767 		ztest_object_unlock(zd, lr->lr_foid);
1768 		return (ENOSPC);
1769 	}
1770 
1771 	bbt = ztest_bt_bonus(db);
1772 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1773 	crtxg = bbt->bt_crtxg;
1774 	lrtxg = lr->lr_common.lrc_txg;
1775 
1776 	if (zd->zd_zilog->zl_replay) {
1777 		ASSERT(lr->lr_size != 0);
1778 		ASSERT(lr->lr_mode != 0);
1779 		ASSERT(lrtxg != 0);
1780 	} else {
1781 		/*
1782 		 * Randomly change the size and increment the generation.
1783 		 */
1784 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1785 		    sizeof (*bbt);
1786 		lr->lr_mode = bbt->bt_gen + 1;
1787 		ASSERT(lrtxg == 0);
1788 	}
1789 
1790 	/*
1791 	 * Verify that the current bonus buffer is not newer than our txg.
1792 	 */
1793 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1794 	    MAX(txg, lrtxg), crtxg);
1795 
1796 	dmu_buf_will_dirty(db, tx);
1797 
1798 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1799 	ASSERT3U(lr->lr_size, <=, db->db_size);
1800 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1801 	bbt = ztest_bt_bonus(db);
1802 
1803 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1804 
1805 	dmu_buf_rele(db, FTAG);
1806 
1807 	(void) ztest_log_setattr(zd, tx, lr);
1808 
1809 	dmu_tx_commit(tx);
1810 
1811 	ztest_object_unlock(zd, lr->lr_foid);
1812 
1813 	return (0);
1814 }
1815 
1816 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1817 	NULL,			/* 0 no such transaction type */
1818 	ztest_replay_create,	/* TX_CREATE */
1819 	NULL,			/* TX_MKDIR */
1820 	NULL,			/* TX_MKXATTR */
1821 	NULL,			/* TX_SYMLINK */
1822 	ztest_replay_remove,	/* TX_REMOVE */
1823 	NULL,			/* TX_RMDIR */
1824 	NULL,			/* TX_LINK */
1825 	NULL,			/* TX_RENAME */
1826 	ztest_replay_write,	/* TX_WRITE */
1827 	ztest_replay_truncate,	/* TX_TRUNCATE */
1828 	ztest_replay_setattr,	/* TX_SETATTR */
1829 	NULL,			/* TX_ACL */
1830 	NULL,			/* TX_CREATE_ACL */
1831 	NULL,			/* TX_CREATE_ATTR */
1832 	NULL,			/* TX_CREATE_ACL_ATTR */
1833 	NULL,			/* TX_MKDIR_ACL */
1834 	NULL,			/* TX_MKDIR_ATTR */
1835 	NULL,			/* TX_MKDIR_ACL_ATTR */
1836 	NULL,			/* TX_WRITE2 */
1837 };
1838 
1839 /*
1840  * ZIL get_data callbacks
1841  */
1842 
1843 static void
1844 ztest_get_done(zgd_t *zgd, int error)
1845 {
1846 	ztest_ds_t *zd = zgd->zgd_private;
1847 	uint64_t object = zgd->zgd_rl->rl_object;
1848 
1849 	if (zgd->zgd_db)
1850 		dmu_buf_rele(zgd->zgd_db, zgd);
1851 
1852 	ztest_range_unlock(zgd->zgd_rl);
1853 	ztest_object_unlock(zd, object);
1854 
1855 	if (error == 0 && zgd->zgd_bp)
1856 		zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1857 
1858 	umem_free(zgd, sizeof (*zgd));
1859 }
1860 
1861 static int
1862 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
1863     zio_t *zio)
1864 {
1865 	ztest_ds_t *zd = arg;
1866 	objset_t *os = zd->zd_os;
1867 	uint64_t object = lr->lr_foid;
1868 	uint64_t offset = lr->lr_offset;
1869 	uint64_t size = lr->lr_length;
1870 	uint64_t txg = lr->lr_common.lrc_txg;
1871 	uint64_t crtxg;
1872 	dmu_object_info_t doi;
1873 	dmu_buf_t *db;
1874 	zgd_t *zgd;
1875 	int error;
1876 
1877 	ASSERT3P(lwb, !=, NULL);
1878 	ASSERT3P(zio, !=, NULL);
1879 	ASSERT3U(size, !=, 0);
1880 
1881 	ztest_object_lock(zd, object, RL_READER);
1882 	error = dmu_bonus_hold(os, object, FTAG, &db);
1883 	if (error) {
1884 		ztest_object_unlock(zd, object);
1885 		return (error);
1886 	}
1887 
1888 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1889 
1890 	if (crtxg == 0 || crtxg > txg) {
1891 		dmu_buf_rele(db, FTAG);
1892 		ztest_object_unlock(zd, object);
1893 		return (ENOENT);
1894 	}
1895 
1896 	dmu_object_info_from_db(db, &doi);
1897 	dmu_buf_rele(db, FTAG);
1898 	db = NULL;
1899 
1900 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1901 	zgd->zgd_lwb = lwb;
1902 	zgd->zgd_private = zd;
1903 
1904 	if (buf != NULL) {	/* immediate write */
1905 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1906 		    RL_READER);
1907 
1908 		error = dmu_read(os, object, offset, size, buf,
1909 		    DMU_READ_NO_PREFETCH);
1910 		ASSERT(error == 0);
1911 	} else {
1912 		size = doi.doi_data_block_size;
1913 		if (ISP2(size)) {
1914 			offset = P2ALIGN(offset, size);
1915 		} else {
1916 			ASSERT(offset < size);
1917 			offset = 0;
1918 		}
1919 
1920 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1921 		    RL_READER);
1922 
1923 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1924 		    DMU_READ_NO_PREFETCH);
1925 
1926 		if (error == 0) {
1927 			blkptr_t *bp = &lr->lr_blkptr;
1928 
1929 			zgd->zgd_db = db;
1930 			zgd->zgd_bp = bp;
1931 
1932 			ASSERT(db->db_offset == offset);
1933 			ASSERT(db->db_size == size);
1934 
1935 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1936 			    ztest_get_done, zgd);
1937 
1938 			if (error == 0)
1939 				return (0);
1940 		}
1941 	}
1942 
1943 	ztest_get_done(zgd, error);
1944 
1945 	return (error);
1946 }
1947 
1948 static void *
1949 ztest_lr_alloc(size_t lrsize, char *name)
1950 {
1951 	char *lr;
1952 	size_t namesize = name ? strlen(name) + 1 : 0;
1953 
1954 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1955 
1956 	if (name)
1957 		bcopy(name, lr + lrsize, namesize);
1958 
1959 	return (lr);
1960 }
1961 
1962 void
1963 ztest_lr_free(void *lr, size_t lrsize, char *name)
1964 {
1965 	size_t namesize = name ? strlen(name) + 1 : 0;
1966 
1967 	umem_free(lr, lrsize + namesize);
1968 }
1969 
1970 /*
1971  * Lookup a bunch of objects.  Returns the number of objects not found.
1972  */
1973 static int
1974 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1975 {
1976 	int missing = 0;
1977 	int error;
1978 
1979 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
1980 
1981 	for (int i = 0; i < count; i++, od++) {
1982 		od->od_object = 0;
1983 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1984 		    sizeof (uint64_t), 1, &od->od_object);
1985 		if (error) {
1986 			ASSERT(error == ENOENT);
1987 			ASSERT(od->od_object == 0);
1988 			missing++;
1989 		} else {
1990 			dmu_buf_t *db;
1991 			ztest_block_tag_t *bbt;
1992 			dmu_object_info_t doi;
1993 
1994 			ASSERT(od->od_object != 0);
1995 			ASSERT(missing == 0);	/* there should be no gaps */
1996 
1997 			ztest_object_lock(zd, od->od_object, RL_READER);
1998 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1999 			    od->od_object, FTAG, &db));
2000 			dmu_object_info_from_db(db, &doi);
2001 			bbt = ztest_bt_bonus(db);
2002 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2003 			od->od_type = doi.doi_type;
2004 			od->od_blocksize = doi.doi_data_block_size;
2005 			od->od_gen = bbt->bt_gen;
2006 			dmu_buf_rele(db, FTAG);
2007 			ztest_object_unlock(zd, od->od_object);
2008 		}
2009 	}
2010 
2011 	return (missing);
2012 }
2013 
2014 static int
2015 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2016 {
2017 	int missing = 0;
2018 
2019 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2020 
2021 	for (int i = 0; i < count; i++, od++) {
2022 		if (missing) {
2023 			od->od_object = 0;
2024 			missing++;
2025 			continue;
2026 		}
2027 
2028 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2029 
2030 		lr->lr_doid = od->od_dir;
2031 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2032 		lr->lrz_type = od->od_crtype;
2033 		lr->lrz_blocksize = od->od_crblocksize;
2034 		lr->lrz_ibshift = ztest_random_ibshift();
2035 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2036 		lr->lrz_bonuslen = dmu_bonus_max();
2037 		lr->lr_gen = od->od_crgen;
2038 		lr->lr_crtime[0] = time(NULL);
2039 
2040 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2041 			ASSERT(missing == 0);
2042 			od->od_object = 0;
2043 			missing++;
2044 		} else {
2045 			od->od_object = lr->lr_foid;
2046 			od->od_type = od->od_crtype;
2047 			od->od_blocksize = od->od_crblocksize;
2048 			od->od_gen = od->od_crgen;
2049 			ASSERT(od->od_object != 0);
2050 		}
2051 
2052 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2053 	}
2054 
2055 	return (missing);
2056 }
2057 
2058 static int
2059 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2060 {
2061 	int missing = 0;
2062 	int error;
2063 
2064 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2065 
2066 	od += count - 1;
2067 
2068 	for (int i = count - 1; i >= 0; i--, od--) {
2069 		if (missing) {
2070 			missing++;
2071 			continue;
2072 		}
2073 
2074 		/*
2075 		 * No object was found.
2076 		 */
2077 		if (od->od_object == 0)
2078 			continue;
2079 
2080 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2081 
2082 		lr->lr_doid = od->od_dir;
2083 
2084 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2085 			ASSERT3U(error, ==, ENOSPC);
2086 			missing++;
2087 		} else {
2088 			od->od_object = 0;
2089 		}
2090 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2091 	}
2092 
2093 	return (missing);
2094 }
2095 
2096 static int
2097 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2098     void *data)
2099 {
2100 	lr_write_t *lr;
2101 	int error;
2102 
2103 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2104 
2105 	lr->lr_foid = object;
2106 	lr->lr_offset = offset;
2107 	lr->lr_length = size;
2108 	lr->lr_blkoff = 0;
2109 	BP_ZERO(&lr->lr_blkptr);
2110 
2111 	bcopy(data, lr + 1, size);
2112 
2113 	error = ztest_replay_write(zd, lr, B_FALSE);
2114 
2115 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2116 
2117 	return (error);
2118 }
2119 
2120 static int
2121 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2122 {
2123 	lr_truncate_t *lr;
2124 	int error;
2125 
2126 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2127 
2128 	lr->lr_foid = object;
2129 	lr->lr_offset = offset;
2130 	lr->lr_length = size;
2131 
2132 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2133 
2134 	ztest_lr_free(lr, sizeof (*lr), NULL);
2135 
2136 	return (error);
2137 }
2138 
2139 static int
2140 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2141 {
2142 	lr_setattr_t *lr;
2143 	int error;
2144 
2145 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2146 
2147 	lr->lr_foid = object;
2148 	lr->lr_size = 0;
2149 	lr->lr_mode = 0;
2150 
2151 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2152 
2153 	ztest_lr_free(lr, sizeof (*lr), NULL);
2154 
2155 	return (error);
2156 }
2157 
2158 static void
2159 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2160 {
2161 	objset_t *os = zd->zd_os;
2162 	dmu_tx_t *tx;
2163 	uint64_t txg;
2164 	rl_t *rl;
2165 
2166 	txg_wait_synced(dmu_objset_pool(os), 0);
2167 
2168 	ztest_object_lock(zd, object, RL_READER);
2169 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2170 
2171 	tx = dmu_tx_create(os);
2172 
2173 	dmu_tx_hold_write(tx, object, offset, size);
2174 
2175 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2176 
2177 	if (txg != 0) {
2178 		dmu_prealloc(os, object, offset, size, tx);
2179 		dmu_tx_commit(tx);
2180 		txg_wait_synced(dmu_objset_pool(os), txg);
2181 	} else {
2182 		(void) dmu_free_long_range(os, object, offset, size);
2183 	}
2184 
2185 	ztest_range_unlock(rl);
2186 	ztest_object_unlock(zd, object);
2187 }
2188 
2189 static void
2190 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2191 {
2192 	int err;
2193 	ztest_block_tag_t wbt;
2194 	dmu_object_info_t doi;
2195 	enum ztest_io_type io_type;
2196 	uint64_t blocksize;
2197 	void *data;
2198 
2199 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2200 	blocksize = doi.doi_data_block_size;
2201 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2202 
2203 	/*
2204 	 * Pick an i/o type at random, biased toward writing block tags.
2205 	 */
2206 	io_type = ztest_random(ZTEST_IO_TYPES);
2207 	if (ztest_random(2) == 0)
2208 		io_type = ZTEST_IO_WRITE_TAG;
2209 
2210 	rw_enter(&zd->zd_zilog_lock, RW_READER);
2211 
2212 	switch (io_type) {
2213 
2214 	case ZTEST_IO_WRITE_TAG:
2215 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2216 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2217 		break;
2218 
2219 	case ZTEST_IO_WRITE_PATTERN:
2220 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2221 		if (ztest_random(2) == 0) {
2222 			/*
2223 			 * Induce fletcher2 collisions to ensure that
2224 			 * zio_ddt_collision() detects and resolves them
2225 			 * when using fletcher2-verify for deduplication.
2226 			 */
2227 			((uint64_t *)data)[0] ^= 1ULL << 63;
2228 			((uint64_t *)data)[4] ^= 1ULL << 63;
2229 		}
2230 		(void) ztest_write(zd, object, offset, blocksize, data);
2231 		break;
2232 
2233 	case ZTEST_IO_WRITE_ZEROES:
2234 		bzero(data, blocksize);
2235 		(void) ztest_write(zd, object, offset, blocksize, data);
2236 		break;
2237 
2238 	case ZTEST_IO_TRUNCATE:
2239 		(void) ztest_truncate(zd, object, offset, blocksize);
2240 		break;
2241 
2242 	case ZTEST_IO_SETATTR:
2243 		(void) ztest_setattr(zd, object);
2244 		break;
2245 
2246 	case ZTEST_IO_REWRITE:
2247 		rw_enter(&ztest_name_lock, RW_READER);
2248 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2249 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2250 		    B_FALSE);
2251 		VERIFY(err == 0 || err == ENOSPC);
2252 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2253 		    ZFS_PROP_COMPRESSION,
2254 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2255 		    B_FALSE);
2256 		VERIFY(err == 0 || err == ENOSPC);
2257 		rw_exit(&ztest_name_lock);
2258 
2259 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2260 		    DMU_READ_NO_PREFETCH));
2261 
2262 		(void) ztest_write(zd, object, offset, blocksize, data);
2263 		break;
2264 	}
2265 
2266 	rw_exit(&zd->zd_zilog_lock);
2267 
2268 	umem_free(data, blocksize);
2269 }
2270 
2271 /*
2272  * Initialize an object description template.
2273  */
2274 static void
2275 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2276     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2277 {
2278 	od->od_dir = ZTEST_DIROBJ;
2279 	od->od_object = 0;
2280 
2281 	od->od_crtype = type;
2282 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2283 	od->od_crgen = gen;
2284 
2285 	od->od_type = DMU_OT_NONE;
2286 	od->od_blocksize = 0;
2287 	od->od_gen = 0;
2288 
2289 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2290 	    tag, (int64_t)id, index);
2291 }
2292 
2293 /*
2294  * Lookup or create the objects for a test using the od template.
2295  * If the objects do not all exist, or if 'remove' is specified,
2296  * remove any existing objects and create new ones.  Otherwise,
2297  * use the existing objects.
2298  */
2299 static int
2300 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2301 {
2302 	int count = size / sizeof (*od);
2303 	int rv = 0;
2304 
2305 	mutex_enter(&zd->zd_dirobj_lock);
2306 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2307 	    (ztest_remove(zd, od, count) != 0 ||
2308 	    ztest_create(zd, od, count) != 0))
2309 		rv = -1;
2310 	zd->zd_od = od;
2311 	mutex_exit(&zd->zd_dirobj_lock);
2312 
2313 	return (rv);
2314 }
2315 
2316 /* ARGSUSED */
2317 void
2318 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2319 {
2320 	zilog_t *zilog = zd->zd_zilog;
2321 
2322 	rw_enter(&zd->zd_zilog_lock, RW_READER);
2323 
2324 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2325 
2326 	/*
2327 	 * Remember the committed values in zd, which is in parent/child
2328 	 * shared memory.  If we die, the next iteration of ztest_run()
2329 	 * will verify that the log really does contain this record.
2330 	 */
2331 	mutex_enter(&zilog->zl_lock);
2332 	ASSERT(zd->zd_shared != NULL);
2333 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2334 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2335 	mutex_exit(&zilog->zl_lock);
2336 
2337 	rw_exit(&zd->zd_zilog_lock);
2338 }
2339 
2340 /*
2341  * This function is designed to simulate the operations that occur during a
2342  * mount/unmount operation.  We hold the dataset across these operations in an
2343  * attempt to expose any implicit assumptions about ZIL management.
2344  */
2345 /* ARGSUSED */
2346 void
2347 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2348 {
2349 	objset_t *os = zd->zd_os;
2350 
2351 	/*
2352 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2353 	 * updating the zil (i.e. adding in-memory log records) and the
2354 	 * zd_zilog_lock to block any I/O.
2355 	 */
2356 	mutex_enter(&zd->zd_dirobj_lock);
2357 	rw_enter(&zd->zd_zilog_lock, RW_WRITER);
2358 
2359 	/* zfsvfs_teardown() */
2360 	zil_close(zd->zd_zilog);
2361 
2362 	/* zfsvfs_setup() */
2363 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2364 	zil_replay(os, zd, ztest_replay_vector);
2365 
2366 	rw_exit(&zd->zd_zilog_lock);
2367 	mutex_exit(&zd->zd_dirobj_lock);
2368 }
2369 
2370 /*
2371  * Verify that we can't destroy an active pool, create an existing pool,
2372  * or create a pool with a bad vdev spec.
2373  */
2374 /* ARGSUSED */
2375 void
2376 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2377 {
2378 	ztest_shared_opts_t *zo = &ztest_opts;
2379 	spa_t *spa;
2380 	nvlist_t *nvroot;
2381 
2382 	/*
2383 	 * Attempt to create using a bad file.
2384 	 */
2385 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2386 	VERIFY3U(ENOENT, ==,
2387 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2388 	nvlist_free(nvroot);
2389 
2390 	/*
2391 	 * Attempt to create using a bad mirror.
2392 	 */
2393 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2394 	VERIFY3U(ENOENT, ==,
2395 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2396 	nvlist_free(nvroot);
2397 
2398 	/*
2399 	 * Attempt to create an existing pool.  It shouldn't matter
2400 	 * what's in the nvroot; we should fail with EEXIST.
2401 	 */
2402 	rw_enter(&ztest_name_lock, RW_READER);
2403 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2404 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2405 	nvlist_free(nvroot);
2406 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2407 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2408 	spa_close(spa, FTAG);
2409 
2410 	rw_exit(&ztest_name_lock);
2411 }
2412 
2413 /* ARGSUSED */
2414 void
2415 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2416 {
2417 	spa_t *spa;
2418 	uint64_t initial_version = SPA_VERSION_INITIAL;
2419 	uint64_t version, newversion;
2420 	nvlist_t *nvroot, *props;
2421 	char *name;
2422 
2423 	mutex_enter(&ztest_vdev_lock);
2424 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2425 
2426 	/*
2427 	 * Clean up from previous runs.
2428 	 */
2429 	(void) spa_destroy(name);
2430 
2431 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2432 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2433 
2434 	/*
2435 	 * If we're configuring a RAIDZ device then make sure that the
2436 	 * the initial version is capable of supporting that feature.
2437 	 */
2438 	switch (ztest_opts.zo_raidz_parity) {
2439 	case 0:
2440 	case 1:
2441 		initial_version = SPA_VERSION_INITIAL;
2442 		break;
2443 	case 2:
2444 		initial_version = SPA_VERSION_RAIDZ2;
2445 		break;
2446 	case 3:
2447 		initial_version = SPA_VERSION_RAIDZ3;
2448 		break;
2449 	}
2450 
2451 	/*
2452 	 * Create a pool with a spa version that can be upgraded. Pick
2453 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2454 	 */
2455 	do {
2456 		version = ztest_random_spa_version(initial_version);
2457 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2458 
2459 	props = fnvlist_alloc();
2460 	fnvlist_add_uint64(props,
2461 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2462 	VERIFY0(spa_create(name, nvroot, props, NULL));
2463 	fnvlist_free(nvroot);
2464 	fnvlist_free(props);
2465 
2466 	VERIFY0(spa_open(name, &spa, FTAG));
2467 	VERIFY3U(spa_version(spa), ==, version);
2468 	newversion = ztest_random_spa_version(version + 1);
2469 
2470 	if (ztest_opts.zo_verbose >= 4) {
2471 		(void) printf("upgrading spa version from %llu to %llu\n",
2472 		    (u_longlong_t)version, (u_longlong_t)newversion);
2473 	}
2474 
2475 	spa_upgrade(spa, newversion);
2476 	VERIFY3U(spa_version(spa), >, version);
2477 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2478 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2479 	spa_close(spa, FTAG);
2480 
2481 	strfree(name);
2482 	mutex_exit(&ztest_vdev_lock);
2483 }
2484 
2485 static void
2486 ztest_spa_checkpoint(spa_t *spa)
2487 {
2488 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2489 
2490 	int error = spa_checkpoint(spa->spa_name);
2491 
2492 	switch (error) {
2493 	case 0:
2494 	case ZFS_ERR_DEVRM_IN_PROGRESS:
2495 	case ZFS_ERR_DISCARDING_CHECKPOINT:
2496 	case ZFS_ERR_CHECKPOINT_EXISTS:
2497 		break;
2498 	case ENOSPC:
2499 		ztest_record_enospc(FTAG);
2500 		break;
2501 	default:
2502 		fatal(0, "spa_checkpoint(%s) = %d", spa->spa_name, error);
2503 	}
2504 }
2505 
2506 static void
2507 ztest_spa_discard_checkpoint(spa_t *spa)
2508 {
2509 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2510 
2511 	int error = spa_checkpoint_discard(spa->spa_name);
2512 
2513 	switch (error) {
2514 	case 0:
2515 	case ZFS_ERR_DISCARDING_CHECKPOINT:
2516 	case ZFS_ERR_NO_CHECKPOINT:
2517 		break;
2518 	default:
2519 		fatal(0, "spa_discard_checkpoint(%s) = %d",
2520 		    spa->spa_name, error);
2521 	}
2522 
2523 }
2524 
2525 /* ARGSUSED */
2526 void
2527 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
2528 {
2529 	spa_t *spa = ztest_spa;
2530 
2531 	mutex_enter(&ztest_checkpoint_lock);
2532 	if (ztest_random(2) == 0) {
2533 		ztest_spa_checkpoint(spa);
2534 	} else {
2535 		ztest_spa_discard_checkpoint(spa);
2536 	}
2537 	mutex_exit(&ztest_checkpoint_lock);
2538 }
2539 
2540 
2541 static vdev_t *
2542 vdev_lookup_by_path(vdev_t *vd, const char *path)
2543 {
2544 	vdev_t *mvd;
2545 
2546 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2547 		return (vd);
2548 
2549 	for (int c = 0; c < vd->vdev_children; c++)
2550 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2551 		    NULL)
2552 			return (mvd);
2553 
2554 	return (NULL);
2555 }
2556 
2557 /*
2558  * Find the first available hole which can be used as a top-level.
2559  */
2560 int
2561 find_vdev_hole(spa_t *spa)
2562 {
2563 	vdev_t *rvd = spa->spa_root_vdev;
2564 	int c;
2565 
2566 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2567 
2568 	for (c = 0; c < rvd->vdev_children; c++) {
2569 		vdev_t *cvd = rvd->vdev_child[c];
2570 
2571 		if (cvd->vdev_ishole)
2572 			break;
2573 	}
2574 	return (c);
2575 }
2576 
2577 /*
2578  * Verify that vdev_add() works as expected.
2579  */
2580 /* ARGSUSED */
2581 void
2582 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2583 {
2584 	ztest_shared_t *zs = ztest_shared;
2585 	spa_t *spa = ztest_spa;
2586 	uint64_t leaves;
2587 	uint64_t guid;
2588 	nvlist_t *nvroot;
2589 	int error;
2590 
2591 	mutex_enter(&ztest_vdev_lock);
2592 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2593 
2594 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2595 
2596 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2597 
2598 	/*
2599 	 * If we have slogs then remove them 1/4 of the time.
2600 	 */
2601 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2602 		/*
2603 		 * Grab the guid from the head of the log class rotor.
2604 		 */
2605 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2606 
2607 		spa_config_exit(spa, SCL_VDEV, FTAG);
2608 
2609 		/*
2610 		 * We have to grab the zs_name_lock as writer to
2611 		 * prevent a race between removing a slog (dmu_objset_find)
2612 		 * and destroying a dataset. Removing the slog will
2613 		 * grab a reference on the dataset which may cause
2614 		 * dmu_objset_destroy() to fail with EBUSY thus
2615 		 * leaving the dataset in an inconsistent state.
2616 		 */
2617 		rw_enter(&ztest_name_lock, RW_WRITER);
2618 		error = spa_vdev_remove(spa, guid, B_FALSE);
2619 		rw_exit(&ztest_name_lock);
2620 
2621 		switch (error) {
2622 		case 0:
2623 		case EEXIST:
2624 		case ZFS_ERR_CHECKPOINT_EXISTS:
2625 		case ZFS_ERR_DISCARDING_CHECKPOINT:
2626 			break;
2627 		default:
2628 			fatal(0, "spa_vdev_remove() = %d", error);
2629 		}
2630 	} else {
2631 		spa_config_exit(spa, SCL_VDEV, FTAG);
2632 
2633 		/*
2634 		 * Make 1/4 of the devices be log devices.
2635 		 */
2636 		nvroot = make_vdev_root(NULL, NULL, NULL,
2637 		    ztest_opts.zo_vdev_size, 0,
2638 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2639 		    zs->zs_mirrors, 1);
2640 
2641 		error = spa_vdev_add(spa, nvroot);
2642 		nvlist_free(nvroot);
2643 
2644 		switch (error) {
2645 		case 0:
2646 			break;
2647 		case ENOSPC:
2648 			ztest_record_enospc("spa_vdev_add");
2649 			break;
2650 		default:
2651 			fatal(0, "spa_vdev_add() = %d", error);
2652 		}
2653 	}
2654 
2655 	mutex_exit(&ztest_vdev_lock);
2656 }
2657 
2658 /*
2659  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2660  */
2661 /* ARGSUSED */
2662 void
2663 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2664 {
2665 	ztest_shared_t *zs = ztest_shared;
2666 	spa_t *spa = ztest_spa;
2667 	vdev_t *rvd = spa->spa_root_vdev;
2668 	spa_aux_vdev_t *sav;
2669 	char *aux;
2670 	uint64_t guid = 0;
2671 	int error;
2672 
2673 	if (ztest_random(2) == 0) {
2674 		sav = &spa->spa_spares;
2675 		aux = ZPOOL_CONFIG_SPARES;
2676 	} else {
2677 		sav = &spa->spa_l2cache;
2678 		aux = ZPOOL_CONFIG_L2CACHE;
2679 	}
2680 
2681 	mutex_enter(&ztest_vdev_lock);
2682 
2683 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2684 
2685 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2686 		/*
2687 		 * Pick a random device to remove.
2688 		 */
2689 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2690 	} else {
2691 		/*
2692 		 * Find an unused device we can add.
2693 		 */
2694 		zs->zs_vdev_aux = 0;
2695 		for (;;) {
2696 			char path[MAXPATHLEN];
2697 			int c;
2698 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2699 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2700 			    zs->zs_vdev_aux);
2701 			for (c = 0; c < sav->sav_count; c++)
2702 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2703 				    path) == 0)
2704 					break;
2705 			if (c == sav->sav_count &&
2706 			    vdev_lookup_by_path(rvd, path) == NULL)
2707 				break;
2708 			zs->zs_vdev_aux++;
2709 		}
2710 	}
2711 
2712 	spa_config_exit(spa, SCL_VDEV, FTAG);
2713 
2714 	if (guid == 0) {
2715 		/*
2716 		 * Add a new device.
2717 		 */
2718 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2719 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2720 		error = spa_vdev_add(spa, nvroot);
2721 
2722 		switch (error) {
2723 		case 0:
2724 			break;
2725 		default:
2726 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2727 		}
2728 		nvlist_free(nvroot);
2729 	} else {
2730 		/*
2731 		 * Remove an existing device.  Sometimes, dirty its
2732 		 * vdev state first to make sure we handle removal
2733 		 * of devices that have pending state changes.
2734 		 */
2735 		if (ztest_random(2) == 0)
2736 			(void) vdev_online(spa, guid, 0, NULL);
2737 
2738 		error = spa_vdev_remove(spa, guid, B_FALSE);
2739 
2740 		switch (error) {
2741 		case 0:
2742 		case EBUSY:
2743 		case ZFS_ERR_CHECKPOINT_EXISTS:
2744 		case ZFS_ERR_DISCARDING_CHECKPOINT:
2745 			break;
2746 		default:
2747 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2748 		}
2749 	}
2750 
2751 	mutex_exit(&ztest_vdev_lock);
2752 }
2753 
2754 /*
2755  * split a pool if it has mirror tlvdevs
2756  */
2757 /* ARGSUSED */
2758 void
2759 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2760 {
2761 	ztest_shared_t *zs = ztest_shared;
2762 	spa_t *spa = ztest_spa;
2763 	vdev_t *rvd = spa->spa_root_vdev;
2764 	nvlist_t *tree, **child, *config, *split, **schild;
2765 	uint_t c, children, schildren = 0, lastlogid = 0;
2766 	int error = 0;
2767 
2768 	mutex_enter(&ztest_vdev_lock);
2769 
2770 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2771 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2772 		mutex_exit(&ztest_vdev_lock);
2773 		return;
2774 	}
2775 
2776 	/* clean up the old pool, if any */
2777 	(void) spa_destroy("splitp");
2778 
2779 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2780 
2781 	/* generate a config from the existing config */
2782 	mutex_enter(&spa->spa_props_lock);
2783 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2784 	    &tree) == 0);
2785 	mutex_exit(&spa->spa_props_lock);
2786 
2787 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2788 	    &children) == 0);
2789 
2790 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2791 	for (c = 0; c < children; c++) {
2792 		vdev_t *tvd = rvd->vdev_child[c];
2793 		nvlist_t **mchild;
2794 		uint_t mchildren;
2795 
2796 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2797 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2798 			    0) == 0);
2799 			VERIFY(nvlist_add_string(schild[schildren],
2800 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2801 			VERIFY(nvlist_add_uint64(schild[schildren],
2802 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2803 			if (lastlogid == 0)
2804 				lastlogid = schildren;
2805 			++schildren;
2806 			continue;
2807 		}
2808 		lastlogid = 0;
2809 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2810 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2811 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2812 	}
2813 
2814 	/* OK, create a config that can be used to split */
2815 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2816 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2817 	    VDEV_TYPE_ROOT) == 0);
2818 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2819 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2820 
2821 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2822 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2823 
2824 	for (c = 0; c < schildren; c++)
2825 		nvlist_free(schild[c]);
2826 	free(schild);
2827 	nvlist_free(split);
2828 
2829 	spa_config_exit(spa, SCL_VDEV, FTAG);
2830 
2831 	rw_enter(&ztest_name_lock, RW_WRITER);
2832 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2833 	rw_exit(&ztest_name_lock);
2834 
2835 	nvlist_free(config);
2836 
2837 	if (error == 0) {
2838 		(void) printf("successful split - results:\n");
2839 		mutex_enter(&spa_namespace_lock);
2840 		show_pool_stats(spa);
2841 		show_pool_stats(spa_lookup("splitp"));
2842 		mutex_exit(&spa_namespace_lock);
2843 		++zs->zs_splits;
2844 		--zs->zs_mirrors;
2845 	}
2846 	mutex_exit(&ztest_vdev_lock);
2847 }
2848 
2849 /*
2850  * Verify that we can attach and detach devices.
2851  */
2852 /* ARGSUSED */
2853 void
2854 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2855 {
2856 	ztest_shared_t *zs = ztest_shared;
2857 	spa_t *spa = ztest_spa;
2858 	spa_aux_vdev_t *sav = &spa->spa_spares;
2859 	vdev_t *rvd = spa->spa_root_vdev;
2860 	vdev_t *oldvd, *newvd, *pvd;
2861 	nvlist_t *root;
2862 	uint64_t leaves;
2863 	uint64_t leaf, top;
2864 	uint64_t ashift = ztest_get_ashift();
2865 	uint64_t oldguid, pguid;
2866 	uint64_t oldsize, newsize;
2867 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2868 	int replacing;
2869 	int oldvd_has_siblings = B_FALSE;
2870 	int newvd_is_spare = B_FALSE;
2871 	int oldvd_is_log;
2872 	int error, expected_error;
2873 
2874 	mutex_enter(&ztest_vdev_lock);
2875 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2876 
2877 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2878 
2879 	/*
2880 	 * If a vdev is in the process of being removed, its removal may
2881 	 * finish while we are in progress, leading to an unexpected error
2882 	 * value.  Don't bother trying to attach while we are in the middle
2883 	 * of removal.
2884 	 */
2885 	if (ztest_device_removal_active) {
2886 		spa_config_exit(spa, SCL_ALL, FTAG);
2887 		mutex_exit(&ztest_vdev_lock);
2888 		return;
2889 	}
2890 
2891 	/*
2892 	 * Decide whether to do an attach or a replace.
2893 	 */
2894 	replacing = ztest_random(2);
2895 
2896 	/*
2897 	 * Pick a random top-level vdev.
2898 	 */
2899 	top = ztest_random_vdev_top(spa, B_TRUE);
2900 
2901 	/*
2902 	 * Pick a random leaf within it.
2903 	 */
2904 	leaf = ztest_random(leaves);
2905 
2906 	/*
2907 	 * Locate this vdev.
2908 	 */
2909 	oldvd = rvd->vdev_child[top];
2910 	if (zs->zs_mirrors >= 1) {
2911 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2912 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2913 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2914 	}
2915 	if (ztest_opts.zo_raidz > 1) {
2916 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2917 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2918 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2919 	}
2920 
2921 	/*
2922 	 * If we're already doing an attach or replace, oldvd may be a
2923 	 * mirror vdev -- in which case, pick a random child.
2924 	 */
2925 	while (oldvd->vdev_children != 0) {
2926 		oldvd_has_siblings = B_TRUE;
2927 		ASSERT(oldvd->vdev_children >= 2);
2928 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2929 	}
2930 
2931 	oldguid = oldvd->vdev_guid;
2932 	oldsize = vdev_get_min_asize(oldvd);
2933 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2934 	(void) strcpy(oldpath, oldvd->vdev_path);
2935 	pvd = oldvd->vdev_parent;
2936 	pguid = pvd->vdev_guid;
2937 
2938 	/*
2939 	 * If oldvd has siblings, then half of the time, detach it.
2940 	 */
2941 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2942 		spa_config_exit(spa, SCL_ALL, FTAG);
2943 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2944 		if (error != 0 && error != ENODEV && error != EBUSY &&
2945 		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
2946 		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
2947 			fatal(0, "detach (%s) returned %d", oldpath, error);
2948 		mutex_exit(&ztest_vdev_lock);
2949 		return;
2950 	}
2951 
2952 	/*
2953 	 * For the new vdev, choose with equal probability between the two
2954 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2955 	 */
2956 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2957 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2958 		newvd_is_spare = B_TRUE;
2959 		(void) strcpy(newpath, newvd->vdev_path);
2960 	} else {
2961 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2962 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
2963 		    top * leaves + leaf);
2964 		if (ztest_random(2) == 0)
2965 			newpath[strlen(newpath) - 1] = 'b';
2966 		newvd = vdev_lookup_by_path(rvd, newpath);
2967 	}
2968 
2969 	if (newvd) {
2970 		/*
2971 		 * Reopen to ensure the vdev's asize field isn't stale.
2972 		 */
2973 		vdev_reopen(newvd);
2974 		newsize = vdev_get_min_asize(newvd);
2975 	} else {
2976 		/*
2977 		 * Make newsize a little bigger or smaller than oldsize.
2978 		 * If it's smaller, the attach should fail.
2979 		 * If it's larger, and we're doing a replace,
2980 		 * we should get dynamic LUN growth when we're done.
2981 		 */
2982 		newsize = 10 * oldsize / (9 + ztest_random(3));
2983 	}
2984 
2985 	/*
2986 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2987 	 * unless it's a replace; in that case any non-replacing parent is OK.
2988 	 *
2989 	 * If newvd is already part of the pool, it should fail with EBUSY.
2990 	 *
2991 	 * If newvd is too small, it should fail with EOVERFLOW.
2992 	 */
2993 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2994 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2995 	    pvd->vdev_ops == &vdev_replacing_ops ||
2996 	    pvd->vdev_ops == &vdev_spare_ops))
2997 		expected_error = ENOTSUP;
2998 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2999 		expected_error = ENOTSUP;
3000 	else if (newvd == oldvd)
3001 		expected_error = replacing ? 0 : EBUSY;
3002 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3003 		expected_error = EBUSY;
3004 	else if (newsize < oldsize)
3005 		expected_error = EOVERFLOW;
3006 	else if (ashift > oldvd->vdev_top->vdev_ashift)
3007 		expected_error = EDOM;
3008 	else
3009 		expected_error = 0;
3010 
3011 	spa_config_exit(spa, SCL_ALL, FTAG);
3012 
3013 	/*
3014 	 * Build the nvlist describing newpath.
3015 	 */
3016 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3017 	    ashift, 0, 0, 0, 1);
3018 
3019 	error = spa_vdev_attach(spa, oldguid, root, replacing);
3020 
3021 	nvlist_free(root);
3022 
3023 	/*
3024 	 * If our parent was the replacing vdev, but the replace completed,
3025 	 * then instead of failing with ENOTSUP we may either succeed,
3026 	 * fail with ENODEV, or fail with EOVERFLOW.
3027 	 */
3028 	if (expected_error == ENOTSUP &&
3029 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3030 		expected_error = error;
3031 
3032 	/*
3033 	 * If someone grew the LUN, the replacement may be too small.
3034 	 */
3035 	if (error == EOVERFLOW || error == EBUSY)
3036 		expected_error = error;
3037 
3038 	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3039 	    error == ZFS_ERR_DISCARDING_CHECKPOINT)
3040 		expected_error = error;
3041 
3042 	/* XXX workaround 6690467 */
3043 	if (error != expected_error && expected_error != EBUSY) {
3044 		fatal(0, "attach (%s %llu, %s %llu, %d) "
3045 		    "returned %d, expected %d",
3046 		    oldpath, oldsize, newpath,
3047 		    newsize, replacing, error, expected_error);
3048 	}
3049 
3050 	mutex_exit(&ztest_vdev_lock);
3051 }
3052 
3053 /* ARGSUSED */
3054 void
3055 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3056 {
3057 	spa_t *spa = ztest_spa;
3058 	vdev_t *vd;
3059 	uint64_t guid;
3060 	int error;
3061 
3062 	mutex_enter(&ztest_vdev_lock);
3063 
3064 	if (ztest_device_removal_active) {
3065 		mutex_exit(&ztest_vdev_lock);
3066 		return;
3067 	}
3068 
3069 	/*
3070 	 * Remove a random top-level vdev and wait for removal to finish.
3071 	 */
3072 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3073 	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3074 	guid = vd->vdev_guid;
3075 	spa_config_exit(spa, SCL_VDEV, FTAG);
3076 
3077 	error = spa_vdev_remove(spa, guid, B_FALSE);
3078 	if (error == 0) {
3079 		ztest_device_removal_active = B_TRUE;
3080 		mutex_exit(&ztest_vdev_lock);
3081 
3082 		while (spa->spa_vdev_removal != NULL)
3083 			txg_wait_synced(spa_get_dsl(spa), 0);
3084 	} else {
3085 		mutex_exit(&ztest_vdev_lock);
3086 		return;
3087 	}
3088 
3089 	/*
3090 	 * The pool needs to be scrubbed after completing device removal.
3091 	 * Failure to do so may result in checksum errors due to the
3092 	 * strategy employed by ztest_fault_inject() when selecting which
3093 	 * offset are redundant and can be damaged.
3094 	 */
3095 	error = spa_scan(spa, POOL_SCAN_SCRUB);
3096 	if (error == 0) {
3097 		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
3098 			txg_wait_synced(spa_get_dsl(spa), 0);
3099 	}
3100 
3101 	mutex_enter(&ztest_vdev_lock);
3102 	ztest_device_removal_active = B_FALSE;
3103 	mutex_exit(&ztest_vdev_lock);
3104 }
3105 
3106 /*
3107  * Callback function which expands the physical size of the vdev.
3108  */
3109 vdev_t *
3110 grow_vdev(vdev_t *vd, void *arg)
3111 {
3112 	spa_t *spa = vd->vdev_spa;
3113 	size_t *newsize = arg;
3114 	size_t fsize;
3115 	int fd;
3116 
3117 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3118 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3119 
3120 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3121 		return (vd);
3122 
3123 	fsize = lseek(fd, 0, SEEK_END);
3124 	(void) ftruncate(fd, *newsize);
3125 
3126 	if (ztest_opts.zo_verbose >= 6) {
3127 		(void) printf("%s grew from %lu to %lu bytes\n",
3128 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3129 	}
3130 	(void) close(fd);
3131 	return (NULL);
3132 }
3133 
3134 /*
3135  * Callback function which expands a given vdev by calling vdev_online().
3136  */
3137 /* ARGSUSED */
3138 vdev_t *
3139 online_vdev(vdev_t *vd, void *arg)
3140 {
3141 	spa_t *spa = vd->vdev_spa;
3142 	vdev_t *tvd = vd->vdev_top;
3143 	uint64_t guid = vd->vdev_guid;
3144 	uint64_t generation = spa->spa_config_generation + 1;
3145 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3146 	int error;
3147 
3148 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3149 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3150 
3151 	/* Calling vdev_online will initialize the new metaslabs */
3152 	spa_config_exit(spa, SCL_STATE, spa);
3153 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3154 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3155 
3156 	/*
3157 	 * If vdev_online returned an error or the underlying vdev_open
3158 	 * failed then we abort the expand. The only way to know that
3159 	 * vdev_open fails is by checking the returned newstate.
3160 	 */
3161 	if (error || newstate != VDEV_STATE_HEALTHY) {
3162 		if (ztest_opts.zo_verbose >= 5) {
3163 			(void) printf("Unable to expand vdev, state %llu, "
3164 			    "error %d\n", (u_longlong_t)newstate, error);
3165 		}
3166 		return (vd);
3167 	}
3168 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3169 
3170 	/*
3171 	 * Since we dropped the lock we need to ensure that we're
3172 	 * still talking to the original vdev. It's possible this
3173 	 * vdev may have been detached/replaced while we were
3174 	 * trying to online it.
3175 	 */
3176 	if (generation != spa->spa_config_generation) {
3177 		if (ztest_opts.zo_verbose >= 5) {
3178 			(void) printf("vdev configuration has changed, "
3179 			    "guid %llu, state %llu, expected gen %llu, "
3180 			    "got gen %llu\n",
3181 			    (u_longlong_t)guid,
3182 			    (u_longlong_t)tvd->vdev_state,
3183 			    (u_longlong_t)generation,
3184 			    (u_longlong_t)spa->spa_config_generation);
3185 		}
3186 		return (vd);
3187 	}
3188 	return (NULL);
3189 }
3190 
3191 /*
3192  * Traverse the vdev tree calling the supplied function.
3193  * We continue to walk the tree until we either have walked all
3194  * children or we receive a non-NULL return from the callback.
3195  * If a NULL callback is passed, then we just return back the first
3196  * leaf vdev we encounter.
3197  */
3198 vdev_t *
3199 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3200 {
3201 	if (vd->vdev_ops->vdev_op_leaf) {
3202 		if (func == NULL)
3203 			return (vd);
3204 		else
3205 			return (func(vd, arg));
3206 	}
3207 
3208 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3209 		vdev_t *cvd = vd->vdev_child[c];
3210 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3211 			return (cvd);
3212 	}
3213 	return (NULL);
3214 }
3215 
3216 /*
3217  * Verify that dynamic LUN growth works as expected.
3218  */
3219 /* ARGSUSED */
3220 void
3221 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3222 {
3223 	spa_t *spa = ztest_spa;
3224 	vdev_t *vd, *tvd;
3225 	metaslab_class_t *mc;
3226 	metaslab_group_t *mg;
3227 	size_t psize, newsize;
3228 	uint64_t top;
3229 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3230 
3231 	mutex_enter(&ztest_checkpoint_lock);
3232 	mutex_enter(&ztest_vdev_lock);
3233 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3234 
3235 	/*
3236 	 * If there is a vdev removal in progress, it could complete while
3237 	 * we are running, in which case we would not be able to verify
3238 	 * that the metaslab_class space increased (because it decreases
3239 	 * when the device removal completes).
3240 	 */
3241 	if (ztest_device_removal_active) {
3242 		spa_config_exit(spa, SCL_STATE, spa);
3243 		mutex_exit(&ztest_vdev_lock);
3244 		mutex_exit(&ztest_checkpoint_lock);
3245 		return;
3246 	}
3247 
3248 	top = ztest_random_vdev_top(spa, B_TRUE);
3249 
3250 	tvd = spa->spa_root_vdev->vdev_child[top];
3251 	mg = tvd->vdev_mg;
3252 	mc = mg->mg_class;
3253 	old_ms_count = tvd->vdev_ms_count;
3254 	old_class_space = metaslab_class_get_space(mc);
3255 
3256 	/*
3257 	 * Determine the size of the first leaf vdev associated with
3258 	 * our top-level device.
3259 	 */
3260 	vd = vdev_walk_tree(tvd, NULL, NULL);
3261 	ASSERT3P(vd, !=, NULL);
3262 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3263 
3264 	psize = vd->vdev_psize;
3265 
3266 	/*
3267 	 * We only try to expand the vdev if it's healthy, less than 4x its
3268 	 * original size, and it has a valid psize.
3269 	 */
3270 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3271 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3272 		spa_config_exit(spa, SCL_STATE, spa);
3273 		mutex_exit(&ztest_vdev_lock);
3274 		mutex_exit(&ztest_checkpoint_lock);
3275 		return;
3276 	}
3277 	ASSERT(psize > 0);
3278 	newsize = psize + psize / 8;
3279 	ASSERT3U(newsize, >, psize);
3280 
3281 	if (ztest_opts.zo_verbose >= 6) {
3282 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3283 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3284 	}
3285 
3286 	/*
3287 	 * Growing the vdev is a two step process:
3288 	 *	1). expand the physical size (i.e. relabel)
3289 	 *	2). online the vdev to create the new metaslabs
3290 	 */
3291 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3292 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3293 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3294 		if (ztest_opts.zo_verbose >= 5) {
3295 			(void) printf("Could not expand LUN because "
3296 			    "the vdev configuration changed.\n");
3297 		}
3298 		spa_config_exit(spa, SCL_STATE, spa);
3299 		mutex_exit(&ztest_vdev_lock);
3300 		mutex_exit(&ztest_checkpoint_lock);
3301 		return;
3302 	}
3303 
3304 	spa_config_exit(spa, SCL_STATE, spa);
3305 
3306 	/*
3307 	 * Expanding the LUN will update the config asynchronously,
3308 	 * thus we must wait for the async thread to complete any
3309 	 * pending tasks before proceeding.
3310 	 */
3311 	for (;;) {
3312 		boolean_t done;
3313 		mutex_enter(&spa->spa_async_lock);
3314 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3315 		mutex_exit(&spa->spa_async_lock);
3316 		if (done)
3317 			break;
3318 		txg_wait_synced(spa_get_dsl(spa), 0);
3319 		(void) poll(NULL, 0, 100);
3320 	}
3321 
3322 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3323 
3324 	tvd = spa->spa_root_vdev->vdev_child[top];
3325 	new_ms_count = tvd->vdev_ms_count;
3326 	new_class_space = metaslab_class_get_space(mc);
3327 
3328 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3329 		if (ztest_opts.zo_verbose >= 5) {
3330 			(void) printf("Could not verify LUN expansion due to "
3331 			    "intervening vdev offline or remove.\n");
3332 		}
3333 		spa_config_exit(spa, SCL_STATE, spa);
3334 		mutex_exit(&ztest_vdev_lock);
3335 		mutex_exit(&ztest_checkpoint_lock);
3336 		return;
3337 	}
3338 
3339 	/*
3340 	 * Make sure we were able to grow the vdev.
3341 	 */
3342 	if (new_ms_count <= old_ms_count) {
3343 		fatal(0, "LUN expansion failed: ms_count %llu < %llu\n",
3344 		    old_ms_count, new_ms_count);
3345 	}
3346 
3347 	/*
3348 	 * Make sure we were able to grow the pool.
3349 	 */
3350 	if (new_class_space <= old_class_space) {
3351 		fatal(0, "LUN expansion failed: class_space %llu < %llu\n",
3352 		    old_class_space, new_class_space);
3353 	}
3354 
3355 	if (ztest_opts.zo_verbose >= 5) {
3356 		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3357 
3358 		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3359 		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3360 		(void) printf("%s grew from %s to %s\n",
3361 		    spa->spa_name, oldnumbuf, newnumbuf);
3362 	}
3363 
3364 	spa_config_exit(spa, SCL_STATE, spa);
3365 	mutex_exit(&ztest_vdev_lock);
3366 	mutex_exit(&ztest_checkpoint_lock);
3367 }
3368 
3369 /*
3370  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3371  */
3372 /* ARGSUSED */
3373 static void
3374 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3375 {
3376 	/*
3377 	 * Create the objects common to all ztest datasets.
3378 	 */
3379 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3380 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3381 }
3382 
3383 static int
3384 ztest_dataset_create(char *dsname)
3385 {
3386 	uint64_t zilset = ztest_random(100);
3387 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3388 	    ztest_objset_create_cb, NULL);
3389 
3390 	if (err || zilset < 80)
3391 		return (err);
3392 
3393 	if (ztest_opts.zo_verbose >= 6)
3394 		(void) printf("Setting dataset %s to sync always\n", dsname);
3395 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3396 	    ZFS_SYNC_ALWAYS, B_FALSE));
3397 }
3398 
3399 /* ARGSUSED */
3400 static int
3401 ztest_objset_destroy_cb(const char *name, void *arg)
3402 {
3403 	objset_t *os;
3404 	dmu_object_info_t doi;
3405 	int error;
3406 
3407 	/*
3408 	 * Verify that the dataset contains a directory object.
3409 	 */
3410 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3411 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3412 	if (error != ENOENT) {
3413 		/* We could have crashed in the middle of destroying it */
3414 		ASSERT0(error);
3415 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3416 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3417 	}
3418 	dmu_objset_disown(os, FTAG);
3419 
3420 	/*
3421 	 * Destroy the dataset.
3422 	 */
3423 	if (strchr(name, '@') != NULL) {
3424 		VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3425 	} else {
3426 		error = dsl_destroy_head(name);
3427 		/* There could be a hold on this dataset */
3428 		if (error != EBUSY)
3429 			ASSERT0(error);
3430 	}
3431 	return (0);
3432 }
3433 
3434 static boolean_t
3435 ztest_snapshot_create(char *osname, uint64_t id)
3436 {
3437 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3438 	int error;
3439 
3440 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3441 
3442 	error = dmu_objset_snapshot_one(osname, snapname);
3443 	if (error == ENOSPC) {
3444 		ztest_record_enospc(FTAG);
3445 		return (B_FALSE);
3446 	}
3447 	if (error != 0 && error != EEXIST) {
3448 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3449 		    snapname, error);
3450 	}
3451 	return (B_TRUE);
3452 }
3453 
3454 static boolean_t
3455 ztest_snapshot_destroy(char *osname, uint64_t id)
3456 {
3457 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3458 	int error;
3459 
3460 	(void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3461 	    (u_longlong_t)id);
3462 
3463 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3464 	if (error != 0 && error != ENOENT)
3465 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3466 	return (B_TRUE);
3467 }
3468 
3469 /* ARGSUSED */
3470 void
3471 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3472 {
3473 	ztest_ds_t zdtmp;
3474 	int iters;
3475 	int error;
3476 	objset_t *os, *os2;
3477 	char name[ZFS_MAX_DATASET_NAME_LEN];
3478 	zilog_t *zilog;
3479 
3480 	rw_enter(&ztest_name_lock, RW_READER);
3481 
3482 	(void) snprintf(name, sizeof (name), "%s/temp_%llu",
3483 	    ztest_opts.zo_pool, (u_longlong_t)id);
3484 
3485 	/*
3486 	 * If this dataset exists from a previous run, process its replay log
3487 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3488 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3489 	 */
3490 	if (ztest_random(2) == 0 &&
3491 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3492 		ztest_zd_init(&zdtmp, NULL, os);
3493 		zil_replay(os, &zdtmp, ztest_replay_vector);
3494 		ztest_zd_fini(&zdtmp);
3495 		dmu_objset_disown(os, FTAG);
3496 	}
3497 
3498 	/*
3499 	 * There may be an old instance of the dataset we're about to
3500 	 * create lying around from a previous run.  If so, destroy it
3501 	 * and all of its snapshots.
3502 	 */
3503 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3504 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3505 
3506 	/*
3507 	 * Verify that the destroyed dataset is no longer in the namespace.
3508 	 */
3509 	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3510 	    FTAG, &os));
3511 
3512 	/*
3513 	 * Verify that we can create a new dataset.
3514 	 */
3515 	error = ztest_dataset_create(name);
3516 	if (error) {
3517 		if (error == ENOSPC) {
3518 			ztest_record_enospc(FTAG);
3519 			rw_exit(&ztest_name_lock);
3520 			return;
3521 		}
3522 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3523 	}
3524 
3525 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3526 
3527 	ztest_zd_init(&zdtmp, NULL, os);
3528 
3529 	/*
3530 	 * Open the intent log for it.
3531 	 */
3532 	zilog = zil_open(os, ztest_get_data);
3533 
3534 	/*
3535 	 * Put some objects in there, do a little I/O to them,
3536 	 * and randomly take a couple of snapshots along the way.
3537 	 */
3538 	iters = ztest_random(5);
3539 	for (int i = 0; i < iters; i++) {
3540 		ztest_dmu_object_alloc_free(&zdtmp, id);
3541 		if (ztest_random(iters) == 0)
3542 			(void) ztest_snapshot_create(name, i);
3543 	}
3544 
3545 	/*
3546 	 * Verify that we cannot create an existing dataset.
3547 	 */
3548 	VERIFY3U(EEXIST, ==,
3549 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3550 
3551 	/*
3552 	 * Verify that we can hold an objset that is also owned.
3553 	 */
3554 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3555 	dmu_objset_rele(os2, FTAG);
3556 
3557 	/*
3558 	 * Verify that we cannot own an objset that is already owned.
3559 	 */
3560 	VERIFY3U(EBUSY, ==,
3561 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3562 
3563 	zil_close(zilog);
3564 	dmu_objset_disown(os, FTAG);
3565 	ztest_zd_fini(&zdtmp);
3566 
3567 	rw_exit(&ztest_name_lock);
3568 }
3569 
3570 /*
3571  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3572  */
3573 void
3574 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3575 {
3576 	rw_enter(&ztest_name_lock, RW_READER);
3577 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3578 	(void) ztest_snapshot_create(zd->zd_name, id);
3579 	rw_exit(&ztest_name_lock);
3580 }
3581 
3582 /*
3583  * Cleanup non-standard snapshots and clones.
3584  */
3585 void
3586 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3587 {
3588 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3589 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3590 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3591 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3592 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3593 	int error;
3594 
3595 	(void) snprintf(snap1name, sizeof (snap1name),
3596 	    "%s@s1_%llu", osname, id);
3597 	(void) snprintf(clone1name, sizeof (clone1name),
3598 	    "%s/c1_%llu", osname, id);
3599 	(void) snprintf(snap2name, sizeof (snap2name),
3600 	    "%s@s2_%llu", clone1name, id);
3601 	(void) snprintf(clone2name, sizeof (clone2name),
3602 	    "%s/c2_%llu", osname, id);
3603 	(void) snprintf(snap3name, sizeof (snap3name),
3604 	    "%s@s3_%llu", clone1name, id);
3605 
3606 	error = dsl_destroy_head(clone2name);
3607 	if (error && error != ENOENT)
3608 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3609 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3610 	if (error && error != ENOENT)
3611 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3612 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3613 	if (error && error != ENOENT)
3614 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3615 	error = dsl_destroy_head(clone1name);
3616 	if (error && error != ENOENT)
3617 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3618 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3619 	if (error && error != ENOENT)
3620 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3621 }
3622 
3623 /*
3624  * Verify dsl_dataset_promote handles EBUSY
3625  */
3626 void
3627 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3628 {
3629 	objset_t *os;
3630 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3631 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3632 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3633 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3634 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3635 	char *osname = zd->zd_name;
3636 	int error;
3637 
3638 	rw_enter(&ztest_name_lock, RW_READER);
3639 
3640 	ztest_dsl_dataset_cleanup(osname, id);
3641 
3642 	(void) snprintf(snap1name, sizeof (snap1name),
3643 	    "%s@s1_%llu", osname, id);
3644 	(void) snprintf(clone1name, sizeof (clone1name),
3645 	    "%s/c1_%llu", osname, id);
3646 	(void) snprintf(snap2name, sizeof (snap2name),
3647 	    "%s@s2_%llu", clone1name, id);
3648 	(void) snprintf(clone2name, sizeof (clone2name),
3649 	    "%s/c2_%llu", osname, id);
3650 	(void) snprintf(snap3name, sizeof (snap3name),
3651 	    "%s@s3_%llu", clone1name, id);
3652 
3653 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3654 	if (error && error != EEXIST) {
3655 		if (error == ENOSPC) {
3656 			ztest_record_enospc(FTAG);
3657 			goto out;
3658 		}
3659 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3660 	}
3661 
3662 	error = dmu_objset_clone(clone1name, snap1name);
3663 	if (error) {
3664 		if (error == ENOSPC) {
3665 			ztest_record_enospc(FTAG);
3666 			goto out;
3667 		}
3668 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3669 	}
3670 
3671 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3672 	if (error && error != EEXIST) {
3673 		if (error == ENOSPC) {
3674 			ztest_record_enospc(FTAG);
3675 			goto out;
3676 		}
3677 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3678 	}
3679 
3680 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3681 	if (error && error != EEXIST) {
3682 		if (error == ENOSPC) {
3683 			ztest_record_enospc(FTAG);
3684 			goto out;
3685 		}
3686 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3687 	}
3688 
3689 	error = dmu_objset_clone(clone2name, snap3name);
3690 	if (error) {
3691 		if (error == ENOSPC) {
3692 			ztest_record_enospc(FTAG);
3693 			goto out;
3694 		}
3695 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3696 	}
3697 
3698 	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3699 	if (error)
3700 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3701 	error = dsl_dataset_promote(clone2name, NULL);
3702 	if (error == ENOSPC) {
3703 		dmu_objset_disown(os, FTAG);
3704 		ztest_record_enospc(FTAG);
3705 		goto out;
3706 	}
3707 	if (error != EBUSY)
3708 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3709 		    error);
3710 	dmu_objset_disown(os, FTAG);
3711 
3712 out:
3713 	ztest_dsl_dataset_cleanup(osname, id);
3714 
3715 	rw_exit(&ztest_name_lock);
3716 }
3717 
3718 /*
3719  * Verify that dmu_object_{alloc,free} work as expected.
3720  */
3721 void
3722 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3723 {
3724 	ztest_od_t od[4];
3725 	int batchsize = sizeof (od) / sizeof (od[0]);
3726 
3727 	for (int b = 0; b < batchsize; b++)
3728 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3729 
3730 	/*
3731 	 * Destroy the previous batch of objects, create a new batch,
3732 	 * and do some I/O on the new objects.
3733 	 */
3734 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3735 		return;
3736 
3737 	while (ztest_random(4 * batchsize) != 0)
3738 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3739 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3740 }
3741 
3742 /*
3743  * Verify that dmu_{read,write} work as expected.
3744  */
3745 void
3746 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3747 {
3748 	objset_t *os = zd->zd_os;
3749 	ztest_od_t od[2];
3750 	dmu_tx_t *tx;
3751 	int i, freeit, error;
3752 	uint64_t n, s, txg;
3753 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3754 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3755 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3756 	uint64_t regions = 997;
3757 	uint64_t stride = 123456789ULL;
3758 	uint64_t width = 40;
3759 	int free_percent = 5;
3760 
3761 	/*
3762 	 * This test uses two objects, packobj and bigobj, that are always
3763 	 * updated together (i.e. in the same tx) so that their contents are
3764 	 * in sync and can be compared.  Their contents relate to each other
3765 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3766 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3767 	 * for any index n, there are three bufwads that should be identical:
3768 	 *
3769 	 *	packobj, at offset n * sizeof (bufwad_t)
3770 	 *	bigobj, at the head of the nth chunk
3771 	 *	bigobj, at the tail of the nth chunk
3772 	 *
3773 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3774 	 * and it doesn't have any relation to the object blocksize.
3775 	 * The only requirement is that it can hold at least two bufwads.
3776 	 *
3777 	 * Normally, we write the bufwad to each of these locations.
3778 	 * However, free_percent of the time we instead write zeroes to
3779 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3780 	 * bigobj to packobj, we can verify that the DMU is correctly
3781 	 * tracking which parts of an object are allocated and free,
3782 	 * and that the contents of the allocated blocks are correct.
3783 	 */
3784 
3785 	/*
3786 	 * Read the directory info.  If it's the first time, set things up.
3787 	 */
3788 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3789 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3790 
3791 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3792 		return;
3793 
3794 	bigobj = od[0].od_object;
3795 	packobj = od[1].od_object;
3796 	chunksize = od[0].od_gen;
3797 	ASSERT(chunksize == od[1].od_gen);
3798 
3799 	/*
3800 	 * Prefetch a random chunk of the big object.
3801 	 * Our aim here is to get some async reads in flight
3802 	 * for blocks that we may free below; the DMU should
3803 	 * handle this race correctly.
3804 	 */
3805 	n = ztest_random(regions) * stride + ztest_random(width);
3806 	s = 1 + ztest_random(2 * width - 1);
3807 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3808 	    ZIO_PRIORITY_SYNC_READ);
3809 
3810 	/*
3811 	 * Pick a random index and compute the offsets into packobj and bigobj.
3812 	 */
3813 	n = ztest_random(regions) * stride + ztest_random(width);
3814 	s = 1 + ztest_random(width - 1);
3815 
3816 	packoff = n * sizeof (bufwad_t);
3817 	packsize = s * sizeof (bufwad_t);
3818 
3819 	bigoff = n * chunksize;
3820 	bigsize = s * chunksize;
3821 
3822 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3823 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3824 
3825 	/*
3826 	 * free_percent of the time, free a range of bigobj rather than
3827 	 * overwriting it.
3828 	 */
3829 	freeit = (ztest_random(100) < free_percent);
3830 
3831 	/*
3832 	 * Read the current contents of our objects.
3833 	 */
3834 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3835 	    DMU_READ_PREFETCH);
3836 	ASSERT0(error);
3837 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3838 	    DMU_READ_PREFETCH);
3839 	ASSERT0(error);
3840 
3841 	/*
3842 	 * Get a tx for the mods to both packobj and bigobj.
3843 	 */
3844 	tx = dmu_tx_create(os);
3845 
3846 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3847 
3848 	if (freeit)
3849 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3850 	else
3851 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3852 
3853 	/* This accounts for setting the checksum/compression. */
3854 	dmu_tx_hold_bonus(tx, bigobj);
3855 
3856 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3857 	if (txg == 0) {
3858 		umem_free(packbuf, packsize);
3859 		umem_free(bigbuf, bigsize);
3860 		return;
3861 	}
3862 
3863 	enum zio_checksum cksum;
3864 	do {
3865 		cksum = (enum zio_checksum)
3866 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3867 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3868 	dmu_object_set_checksum(os, bigobj, cksum, tx);
3869 
3870 	enum zio_compress comp;
3871 	do {
3872 		comp = (enum zio_compress)
3873 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3874 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3875 	dmu_object_set_compress(os, bigobj, comp, tx);
3876 
3877 	/*
3878 	 * For each index from n to n + s, verify that the existing bufwad
3879 	 * in packobj matches the bufwads at the head and tail of the
3880 	 * corresponding chunk in bigobj.  Then update all three bufwads
3881 	 * with the new values we want to write out.
3882 	 */
3883 	for (i = 0; i < s; i++) {
3884 		/* LINTED */
3885 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3886 		/* LINTED */
3887 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3888 		/* LINTED */
3889 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3890 
3891 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3892 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3893 
3894 		if (pack->bw_txg > txg)
3895 			fatal(0, "future leak: got %llx, open txg is %llx",
3896 			    pack->bw_txg, txg);
3897 
3898 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3899 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3900 			    pack->bw_index, n, i);
3901 
3902 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3903 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3904 
3905 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3906 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3907 
3908 		if (freeit) {
3909 			bzero(pack, sizeof (bufwad_t));
3910 		} else {
3911 			pack->bw_index = n + i;
3912 			pack->bw_txg = txg;
3913 			pack->bw_data = 1 + ztest_random(-2ULL);
3914 		}
3915 		*bigH = *pack;
3916 		*bigT = *pack;
3917 	}
3918 
3919 	/*
3920 	 * We've verified all the old bufwads, and made new ones.
3921 	 * Now write them out.
3922 	 */
3923 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3924 
3925 	if (freeit) {
3926 		if (ztest_opts.zo_verbose >= 7) {
3927 			(void) printf("freeing offset %llx size %llx"
3928 			    " txg %llx\n",
3929 			    (u_longlong_t)bigoff,
3930 			    (u_longlong_t)bigsize,
3931 			    (u_longlong_t)txg);
3932 		}
3933 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3934 	} else {
3935 		if (ztest_opts.zo_verbose >= 7) {
3936 			(void) printf("writing offset %llx size %llx"
3937 			    " txg %llx\n",
3938 			    (u_longlong_t)bigoff,
3939 			    (u_longlong_t)bigsize,
3940 			    (u_longlong_t)txg);
3941 		}
3942 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3943 	}
3944 
3945 	dmu_tx_commit(tx);
3946 
3947 	/*
3948 	 * Sanity check the stuff we just wrote.
3949 	 */
3950 	{
3951 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3952 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3953 
3954 		VERIFY(0 == dmu_read(os, packobj, packoff,
3955 		    packsize, packcheck, DMU_READ_PREFETCH));
3956 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3957 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3958 
3959 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3960 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3961 
3962 		umem_free(packcheck, packsize);
3963 		umem_free(bigcheck, bigsize);
3964 	}
3965 
3966 	umem_free(packbuf, packsize);
3967 	umem_free(bigbuf, bigsize);
3968 }
3969 
3970 void
3971 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3972     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3973 {
3974 	uint64_t i;
3975 	bufwad_t *pack;
3976 	bufwad_t *bigH;
3977 	bufwad_t *bigT;
3978 
3979 	/*
3980 	 * For each index from n to n + s, verify that the existing bufwad
3981 	 * in packobj matches the bufwads at the head and tail of the
3982 	 * corresponding chunk in bigobj.  Then update all three bufwads
3983 	 * with the new values we want to write out.
3984 	 */
3985 	for (i = 0; i < s; i++) {
3986 		/* LINTED */
3987 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3988 		/* LINTED */
3989 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3990 		/* LINTED */
3991 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3992 
3993 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3994 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3995 
3996 		if (pack->bw_txg > txg)
3997 			fatal(0, "future leak: got %llx, open txg is %llx",
3998 			    pack->bw_txg, txg);
3999 
4000 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4001 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4002 			    pack->bw_index, n, i);
4003 
4004 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4005 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4006 
4007 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4008 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4009 
4010 		pack->bw_index = n + i;
4011 		pack->bw_txg = txg;
4012 		pack->bw_data = 1 + ztest_random(-2ULL);
4013 
4014 		*bigH = *pack;
4015 		*bigT = *pack;
4016 	}
4017 }
4018 
4019 void
4020 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4021 {
4022 	objset_t *os = zd->zd_os;
4023 	ztest_od_t od[2];
4024 	dmu_tx_t *tx;
4025 	uint64_t i;
4026 	int error;
4027 	uint64_t n, s, txg;
4028 	bufwad_t *packbuf, *bigbuf;
4029 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4030 	uint64_t blocksize = ztest_random_blocksize();
4031 	uint64_t chunksize = blocksize;
4032 	uint64_t regions = 997;
4033 	uint64_t stride = 123456789ULL;
4034 	uint64_t width = 9;
4035 	dmu_buf_t *bonus_db;
4036 	arc_buf_t **bigbuf_arcbufs;
4037 	dmu_object_info_t doi;
4038 
4039 	/*
4040 	 * This test uses two objects, packobj and bigobj, that are always
4041 	 * updated together (i.e. in the same tx) so that their contents are
4042 	 * in sync and can be compared.  Their contents relate to each other
4043 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4044 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4045 	 * for any index n, there are three bufwads that should be identical:
4046 	 *
4047 	 *	packobj, at offset n * sizeof (bufwad_t)
4048 	 *	bigobj, at the head of the nth chunk
4049 	 *	bigobj, at the tail of the nth chunk
4050 	 *
4051 	 * The chunk size is set equal to bigobj block size so that
4052 	 * dmu_assign_arcbuf() can be tested for object updates.
4053 	 */
4054 
4055 	/*
4056 	 * Read the directory info.  If it's the first time, set things up.
4057 	 */
4058 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4059 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
4060 
4061 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4062 		return;
4063 
4064 	bigobj = od[0].od_object;
4065 	packobj = od[1].od_object;
4066 	blocksize = od[0].od_blocksize;
4067 	chunksize = blocksize;
4068 	ASSERT(chunksize == od[1].od_gen);
4069 
4070 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4071 	VERIFY(ISP2(doi.doi_data_block_size));
4072 	VERIFY(chunksize == doi.doi_data_block_size);
4073 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4074 
4075 	/*
4076 	 * Pick a random index and compute the offsets into packobj and bigobj.
4077 	 */
4078 	n = ztest_random(regions) * stride + ztest_random(width);
4079 	s = 1 + ztest_random(width - 1);
4080 
4081 	packoff = n * sizeof (bufwad_t);
4082 	packsize = s * sizeof (bufwad_t);
4083 
4084 	bigoff = n * chunksize;
4085 	bigsize = s * chunksize;
4086 
4087 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4088 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4089 
4090 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4091 
4092 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4093 
4094 	/*
4095 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4096 	 * Iteration 1 test zcopy to already referenced dbufs.
4097 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4098 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4099 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
4100 	 * Iteration 5 test zcopy when it can't be done.
4101 	 * Iteration 6 one more zcopy write.
4102 	 */
4103 	for (i = 0; i < 7; i++) {
4104 		uint64_t j;
4105 		uint64_t off;
4106 
4107 		/*
4108 		 * In iteration 5 (i == 5) use arcbufs
4109 		 * that don't match bigobj blksz to test
4110 		 * dmu_assign_arcbuf() when it can't directly
4111 		 * assign an arcbuf to a dbuf.
4112 		 */
4113 		for (j = 0; j < s; j++) {
4114 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4115 				bigbuf_arcbufs[j] =
4116 				    dmu_request_arcbuf(bonus_db, chunksize);
4117 			} else {
4118 				bigbuf_arcbufs[2 * j] =
4119 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
4120 				bigbuf_arcbufs[2 * j + 1] =
4121 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
4122 			}
4123 		}
4124 
4125 		/*
4126 		 * Get a tx for the mods to both packobj and bigobj.
4127 		 */
4128 		tx = dmu_tx_create(os);
4129 
4130 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
4131 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4132 
4133 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4134 		if (txg == 0) {
4135 			umem_free(packbuf, packsize);
4136 			umem_free(bigbuf, bigsize);
4137 			for (j = 0; j < s; j++) {
4138 				if (i != 5 ||
4139 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
4140 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
4141 				} else {
4142 					dmu_return_arcbuf(
4143 					    bigbuf_arcbufs[2 * j]);
4144 					dmu_return_arcbuf(
4145 					    bigbuf_arcbufs[2 * j + 1]);
4146 				}
4147 			}
4148 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4149 			dmu_buf_rele(bonus_db, FTAG);
4150 			return;
4151 		}
4152 
4153 		/*
4154 		 * 50% of the time don't read objects in the 1st iteration to
4155 		 * test dmu_assign_arcbuf() for the case when there're no
4156 		 * existing dbufs for the specified offsets.
4157 		 */
4158 		if (i != 0 || ztest_random(2) != 0) {
4159 			error = dmu_read(os, packobj, packoff,
4160 			    packsize, packbuf, DMU_READ_PREFETCH);
4161 			ASSERT0(error);
4162 			error = dmu_read(os, bigobj, bigoff, bigsize,
4163 			    bigbuf, DMU_READ_PREFETCH);
4164 			ASSERT0(error);
4165 		}
4166 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4167 		    n, chunksize, txg);
4168 
4169 		/*
4170 		 * We've verified all the old bufwads, and made new ones.
4171 		 * Now write them out.
4172 		 */
4173 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4174 		if (ztest_opts.zo_verbose >= 7) {
4175 			(void) printf("writing offset %llx size %llx"
4176 			    " txg %llx\n",
4177 			    (u_longlong_t)bigoff,
4178 			    (u_longlong_t)bigsize,
4179 			    (u_longlong_t)txg);
4180 		}
4181 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4182 			dmu_buf_t *dbt;
4183 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4184 				bcopy((caddr_t)bigbuf + (off - bigoff),
4185 				    bigbuf_arcbufs[j]->b_data, chunksize);
4186 			} else {
4187 				bcopy((caddr_t)bigbuf + (off - bigoff),
4188 				    bigbuf_arcbufs[2 * j]->b_data,
4189 				    chunksize / 2);
4190 				bcopy((caddr_t)bigbuf + (off - bigoff) +
4191 				    chunksize / 2,
4192 				    bigbuf_arcbufs[2 * j + 1]->b_data,
4193 				    chunksize / 2);
4194 			}
4195 
4196 			if (i == 1) {
4197 				VERIFY(dmu_buf_hold(os, bigobj, off,
4198 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4199 			}
4200 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4201 				dmu_assign_arcbuf(bonus_db, off,
4202 				    bigbuf_arcbufs[j], tx);
4203 			} else {
4204 				dmu_assign_arcbuf(bonus_db, off,
4205 				    bigbuf_arcbufs[2 * j], tx);
4206 				dmu_assign_arcbuf(bonus_db,
4207 				    off + chunksize / 2,
4208 				    bigbuf_arcbufs[2 * j + 1], tx);
4209 			}
4210 			if (i == 1) {
4211 				dmu_buf_rele(dbt, FTAG);
4212 			}
4213 		}
4214 		dmu_tx_commit(tx);
4215 
4216 		/*
4217 		 * Sanity check the stuff we just wrote.
4218 		 */
4219 		{
4220 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4221 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4222 
4223 			VERIFY(0 == dmu_read(os, packobj, packoff,
4224 			    packsize, packcheck, DMU_READ_PREFETCH));
4225 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
4226 			    bigsize, bigcheck, DMU_READ_PREFETCH));
4227 
4228 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4229 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4230 
4231 			umem_free(packcheck, packsize);
4232 			umem_free(bigcheck, bigsize);
4233 		}
4234 		if (i == 2) {
4235 			txg_wait_open(dmu_objset_pool(os), 0);
4236 		} else if (i == 3) {
4237 			txg_wait_synced(dmu_objset_pool(os), 0);
4238 		}
4239 	}
4240 
4241 	dmu_buf_rele(bonus_db, FTAG);
4242 	umem_free(packbuf, packsize);
4243 	umem_free(bigbuf, bigsize);
4244 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4245 }
4246 
4247 /* ARGSUSED */
4248 void
4249 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4250 {
4251 	ztest_od_t od[1];
4252 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4253 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4254 
4255 	/*
4256 	 * Have multiple threads write to large offsets in an object
4257 	 * to verify that parallel writes to an object -- even to the
4258 	 * same blocks within the object -- doesn't cause any trouble.
4259 	 */
4260 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4261 
4262 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4263 		return;
4264 
4265 	while (ztest_random(10) != 0)
4266 		ztest_io(zd, od[0].od_object, offset);
4267 }
4268 
4269 void
4270 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4271 {
4272 	ztest_od_t od[1];
4273 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4274 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4275 	uint64_t count = ztest_random(20) + 1;
4276 	uint64_t blocksize = ztest_random_blocksize();
4277 	void *data;
4278 
4279 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4280 
4281 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4282 		return;
4283 
4284 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4285 		return;
4286 
4287 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4288 
4289 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4290 
4291 	while (ztest_random(count) != 0) {
4292 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4293 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4294 		    data) != 0)
4295 			break;
4296 		while (ztest_random(4) != 0)
4297 			ztest_io(zd, od[0].od_object, randoff);
4298 	}
4299 
4300 	umem_free(data, blocksize);
4301 }
4302 
4303 /*
4304  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4305  */
4306 #define	ZTEST_ZAP_MIN_INTS	1
4307 #define	ZTEST_ZAP_MAX_INTS	4
4308 #define	ZTEST_ZAP_MAX_PROPS	1000
4309 
4310 void
4311 ztest_zap(ztest_ds_t *zd, uint64_t id)
4312 {
4313 	objset_t *os = zd->zd_os;
4314 	ztest_od_t od[1];
4315 	uint64_t object;
4316 	uint64_t txg, last_txg;
4317 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4318 	uint64_t zl_ints, zl_intsize, prop;
4319 	int i, ints;
4320 	dmu_tx_t *tx;
4321 	char propname[100], txgname[100];
4322 	int error;
4323 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4324 
4325 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4326 
4327 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4328 		return;
4329 
4330 	object = od[0].od_object;
4331 
4332 	/*
4333 	 * Generate a known hash collision, and verify that
4334 	 * we can lookup and remove both entries.
4335 	 */
4336 	tx = dmu_tx_create(os);
4337 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4338 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4339 	if (txg == 0)
4340 		return;
4341 	for (i = 0; i < 2; i++) {
4342 		value[i] = i;
4343 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4344 		    1, &value[i], tx));
4345 	}
4346 	for (i = 0; i < 2; i++) {
4347 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4348 		    sizeof (uint64_t), 1, &value[i], tx));
4349 		VERIFY3U(0, ==,
4350 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4351 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4352 		ASSERT3U(zl_ints, ==, 1);
4353 	}
4354 	for (i = 0; i < 2; i++) {
4355 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4356 	}
4357 	dmu_tx_commit(tx);
4358 
4359 	/*
4360 	 * Generate a buch of random entries.
4361 	 */
4362 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4363 
4364 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4365 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4366 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4367 	bzero(value, sizeof (value));
4368 	last_txg = 0;
4369 
4370 	/*
4371 	 * If these zap entries already exist, validate their contents.
4372 	 */
4373 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4374 	if (error == 0) {
4375 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4376 		ASSERT3U(zl_ints, ==, 1);
4377 
4378 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4379 		    zl_ints, &last_txg) == 0);
4380 
4381 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4382 		    &zl_ints) == 0);
4383 
4384 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4385 		ASSERT3U(zl_ints, ==, ints);
4386 
4387 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4388 		    zl_ints, value) == 0);
4389 
4390 		for (i = 0; i < ints; i++) {
4391 			ASSERT3U(value[i], ==, last_txg + object + i);
4392 		}
4393 	} else {
4394 		ASSERT3U(error, ==, ENOENT);
4395 	}
4396 
4397 	/*
4398 	 * Atomically update two entries in our zap object.
4399 	 * The first is named txg_%llu, and contains the txg
4400 	 * in which the property was last updated.  The second
4401 	 * is named prop_%llu, and the nth element of its value
4402 	 * should be txg + object + n.
4403 	 */
4404 	tx = dmu_tx_create(os);
4405 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4406 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4407 	if (txg == 0)
4408 		return;
4409 
4410 	if (last_txg > txg)
4411 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4412 
4413 	for (i = 0; i < ints; i++)
4414 		value[i] = txg + object + i;
4415 
4416 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4417 	    1, &txg, tx));
4418 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4419 	    ints, value, tx));
4420 
4421 	dmu_tx_commit(tx);
4422 
4423 	/*
4424 	 * Remove a random pair of entries.
4425 	 */
4426 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4427 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4428 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4429 
4430 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4431 
4432 	if (error == ENOENT)
4433 		return;
4434 
4435 	ASSERT0(error);
4436 
4437 	tx = dmu_tx_create(os);
4438 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4439 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4440 	if (txg == 0)
4441 		return;
4442 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4443 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4444 	dmu_tx_commit(tx);
4445 }
4446 
4447 /*
4448  * Testcase to test the upgrading of a microzap to fatzap.
4449  */
4450 void
4451 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4452 {
4453 	objset_t *os = zd->zd_os;
4454 	ztest_od_t od[1];
4455 	uint64_t object, txg;
4456 
4457 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4458 
4459 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4460 		return;
4461 
4462 	object = od[0].od_object;
4463 
4464 	/*
4465 	 * Add entries to this ZAP and make sure it spills over
4466 	 * and gets upgraded to a fatzap. Also, since we are adding
4467 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4468 	 */
4469 	for (int i = 0; i < 2050; i++) {
4470 		char name[ZFS_MAX_DATASET_NAME_LEN];
4471 		uint64_t value = i;
4472 		dmu_tx_t *tx;
4473 		int error;
4474 
4475 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4476 		    id, value);
4477 
4478 		tx = dmu_tx_create(os);
4479 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4480 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4481 		if (txg == 0)
4482 			return;
4483 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4484 		    &value, tx);
4485 		ASSERT(error == 0 || error == EEXIST);
4486 		dmu_tx_commit(tx);
4487 	}
4488 }
4489 
4490 /* ARGSUSED */
4491 void
4492 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4493 {
4494 	objset_t *os = zd->zd_os;
4495 	ztest_od_t od[1];
4496 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4497 	dmu_tx_t *tx;
4498 	int i, namelen, error;
4499 	int micro = ztest_random(2);
4500 	char name[20], string_value[20];
4501 	void *data;
4502 
4503 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4504 
4505 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4506 		return;
4507 
4508 	object = od[0].od_object;
4509 
4510 	/*
4511 	 * Generate a random name of the form 'xxx.....' where each
4512 	 * x is a random printable character and the dots are dots.
4513 	 * There are 94 such characters, and the name length goes from
4514 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4515 	 */
4516 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4517 
4518 	for (i = 0; i < 3; i++)
4519 		name[i] = '!' + ztest_random('~' - '!' + 1);
4520 	for (; i < namelen - 1; i++)
4521 		name[i] = '.';
4522 	name[i] = '\0';
4523 
4524 	if ((namelen & 1) || micro) {
4525 		wsize = sizeof (txg);
4526 		wc = 1;
4527 		data = &txg;
4528 	} else {
4529 		wsize = 1;
4530 		wc = namelen;
4531 		data = string_value;
4532 	}
4533 
4534 	count = -1ULL;
4535 	VERIFY0(zap_count(os, object, &count));
4536 	ASSERT(count != -1ULL);
4537 
4538 	/*
4539 	 * Select an operation: length, lookup, add, update, remove.
4540 	 */
4541 	i = ztest_random(5);
4542 
4543 	if (i >= 2) {
4544 		tx = dmu_tx_create(os);
4545 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4546 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4547 		if (txg == 0)
4548 			return;
4549 		bcopy(name, string_value, namelen);
4550 	} else {
4551 		tx = NULL;
4552 		txg = 0;
4553 		bzero(string_value, namelen);
4554 	}
4555 
4556 	switch (i) {
4557 
4558 	case 0:
4559 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4560 		if (error == 0) {
4561 			ASSERT3U(wsize, ==, zl_wsize);
4562 			ASSERT3U(wc, ==, zl_wc);
4563 		} else {
4564 			ASSERT3U(error, ==, ENOENT);
4565 		}
4566 		break;
4567 
4568 	case 1:
4569 		error = zap_lookup(os, object, name, wsize, wc, data);
4570 		if (error == 0) {
4571 			if (data == string_value &&
4572 			    bcmp(name, data, namelen) != 0)
4573 				fatal(0, "name '%s' != val '%s' len %d",
4574 				    name, data, namelen);
4575 		} else {
4576 			ASSERT3U(error, ==, ENOENT);
4577 		}
4578 		break;
4579 
4580 	case 2:
4581 		error = zap_add(os, object, name, wsize, wc, data, tx);
4582 		ASSERT(error == 0 || error == EEXIST);
4583 		break;
4584 
4585 	case 3:
4586 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4587 		break;
4588 
4589 	case 4:
4590 		error = zap_remove(os, object, name, tx);
4591 		ASSERT(error == 0 || error == ENOENT);
4592 		break;
4593 	}
4594 
4595 	if (tx != NULL)
4596 		dmu_tx_commit(tx);
4597 }
4598 
4599 /*
4600  * Commit callback data.
4601  */
4602 typedef struct ztest_cb_data {
4603 	list_node_t		zcd_node;
4604 	uint64_t		zcd_txg;
4605 	int			zcd_expected_err;
4606 	boolean_t		zcd_added;
4607 	boolean_t		zcd_called;
4608 	spa_t			*zcd_spa;
4609 } ztest_cb_data_t;
4610 
4611 /* This is the actual commit callback function */
4612 static void
4613 ztest_commit_callback(void *arg, int error)
4614 {
4615 	ztest_cb_data_t *data = arg;
4616 	uint64_t synced_txg;
4617 
4618 	VERIFY(data != NULL);
4619 	VERIFY3S(data->zcd_expected_err, ==, error);
4620 	VERIFY(!data->zcd_called);
4621 
4622 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4623 	if (data->zcd_txg > synced_txg)
4624 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4625 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4626 		    synced_txg);
4627 
4628 	data->zcd_called = B_TRUE;
4629 
4630 	if (error == ECANCELED) {
4631 		ASSERT0(data->zcd_txg);
4632 		ASSERT(!data->zcd_added);
4633 
4634 		/*
4635 		 * The private callback data should be destroyed here, but
4636 		 * since we are going to check the zcd_called field after
4637 		 * dmu_tx_abort(), we will destroy it there.
4638 		 */
4639 		return;
4640 	}
4641 
4642 	/* Was this callback added to the global callback list? */
4643 	if (!data->zcd_added)
4644 		goto out;
4645 
4646 	ASSERT3U(data->zcd_txg, !=, 0);
4647 
4648 	/* Remove our callback from the list */
4649 	mutex_enter(&zcl.zcl_callbacks_lock);
4650 	list_remove(&zcl.zcl_callbacks, data);
4651 	mutex_exit(&zcl.zcl_callbacks_lock);
4652 
4653 out:
4654 	umem_free(data, sizeof (ztest_cb_data_t));
4655 }
4656 
4657 /* Allocate and initialize callback data structure */
4658 static ztest_cb_data_t *
4659 ztest_create_cb_data(objset_t *os, uint64_t txg)
4660 {
4661 	ztest_cb_data_t *cb_data;
4662 
4663 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4664 
4665 	cb_data->zcd_txg = txg;
4666 	cb_data->zcd_spa = dmu_objset_spa(os);
4667 
4668 	return (cb_data);
4669 }
4670 
4671 /*
4672  * If a number of txgs equal to this threshold have been created after a commit
4673  * callback has been registered but not called, then we assume there is an
4674  * implementation bug.
4675  */
4676 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4677 
4678 /*
4679  * Commit callback test.
4680  */
4681 void
4682 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4683 {
4684 	objset_t *os = zd->zd_os;
4685 	ztest_od_t od[1];
4686 	dmu_tx_t *tx;
4687 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4688 	uint64_t old_txg, txg;
4689 	int i, error;
4690 
4691 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4692 
4693 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4694 		return;
4695 
4696 	tx = dmu_tx_create(os);
4697 
4698 	cb_data[0] = ztest_create_cb_data(os, 0);
4699 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4700 
4701 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4702 
4703 	/* Every once in a while, abort the transaction on purpose */
4704 	if (ztest_random(100) == 0)
4705 		error = -1;
4706 
4707 	if (!error)
4708 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4709 
4710 	txg = error ? 0 : dmu_tx_get_txg(tx);
4711 
4712 	cb_data[0]->zcd_txg = txg;
4713 	cb_data[1] = ztest_create_cb_data(os, txg);
4714 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4715 
4716 	if (error) {
4717 		/*
4718 		 * It's not a strict requirement to call the registered
4719 		 * callbacks from inside dmu_tx_abort(), but that's what
4720 		 * it's supposed to happen in the current implementation
4721 		 * so we will check for that.
4722 		 */
4723 		for (i = 0; i < 2; i++) {
4724 			cb_data[i]->zcd_expected_err = ECANCELED;
4725 			VERIFY(!cb_data[i]->zcd_called);
4726 		}
4727 
4728 		dmu_tx_abort(tx);
4729 
4730 		for (i = 0; i < 2; i++) {
4731 			VERIFY(cb_data[i]->zcd_called);
4732 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4733 		}
4734 
4735 		return;
4736 	}
4737 
4738 	cb_data[2] = ztest_create_cb_data(os, txg);
4739 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4740 
4741 	/*
4742 	 * Read existing data to make sure there isn't a future leak.
4743 	 */
4744 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4745 	    &old_txg, DMU_READ_PREFETCH));
4746 
4747 	if (old_txg > txg)
4748 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4749 		    old_txg, txg);
4750 
4751 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4752 
4753 	mutex_enter(&zcl.zcl_callbacks_lock);
4754 
4755 	/*
4756 	 * Since commit callbacks don't have any ordering requirement and since
4757 	 * it is theoretically possible for a commit callback to be called
4758 	 * after an arbitrary amount of time has elapsed since its txg has been
4759 	 * synced, it is difficult to reliably determine whether a commit
4760 	 * callback hasn't been called due to high load or due to a flawed
4761 	 * implementation.
4762 	 *
4763 	 * In practice, we will assume that if after a certain number of txgs a
4764 	 * commit callback hasn't been called, then most likely there's an
4765 	 * implementation bug..
4766 	 */
4767 	tmp_cb = list_head(&zcl.zcl_callbacks);
4768 	if (tmp_cb != NULL &&
4769 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4770 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4771 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4772 	}
4773 
4774 	/*
4775 	 * Let's find the place to insert our callbacks.
4776 	 *
4777 	 * Even though the list is ordered by txg, it is possible for the
4778 	 * insertion point to not be the end because our txg may already be
4779 	 * quiescing at this point and other callbacks in the open txg
4780 	 * (from other objsets) may have sneaked in.
4781 	 */
4782 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4783 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4784 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4785 
4786 	/* Add the 3 callbacks to the list */
4787 	for (i = 0; i < 3; i++) {
4788 		if (tmp_cb == NULL)
4789 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4790 		else
4791 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4792 			    cb_data[i]);
4793 
4794 		cb_data[i]->zcd_added = B_TRUE;
4795 		VERIFY(!cb_data[i]->zcd_called);
4796 
4797 		tmp_cb = cb_data[i];
4798 	}
4799 
4800 	mutex_exit(&zcl.zcl_callbacks_lock);
4801 
4802 	dmu_tx_commit(tx);
4803 }
4804 
4805 /* ARGSUSED */
4806 void
4807 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4808 {
4809 	zfs_prop_t proplist[] = {
4810 		ZFS_PROP_CHECKSUM,
4811 		ZFS_PROP_COMPRESSION,
4812 		ZFS_PROP_COPIES,
4813 		ZFS_PROP_DEDUP
4814 	};
4815 
4816 	rw_enter(&ztest_name_lock, RW_READER);
4817 
4818 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4819 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4820 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4821 
4822 	rw_exit(&ztest_name_lock);
4823 }
4824 
4825 /* ARGSUSED */
4826 void
4827 ztest_remap_blocks(ztest_ds_t *zd, uint64_t id)
4828 {
4829 	rw_enter(&ztest_name_lock, RW_READER);
4830 
4831 	int error = dmu_objset_remap_indirects(zd->zd_name);
4832 	if (error == ENOSPC)
4833 		error = 0;
4834 	ASSERT0(error);
4835 
4836 	rw_exit(&ztest_name_lock);
4837 }
4838 
4839 /* ARGSUSED */
4840 void
4841 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4842 {
4843 	nvlist_t *props = NULL;
4844 
4845 	rw_enter(&ztest_name_lock, RW_READER);
4846 
4847 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4848 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4849 
4850 	VERIFY0(spa_prop_get(ztest_spa, &props));
4851 
4852 	if (ztest_opts.zo_verbose >= 6)
4853 		dump_nvlist(props, 4);
4854 
4855 	nvlist_free(props);
4856 
4857 	rw_exit(&ztest_name_lock);
4858 }
4859 
4860 static int
4861 user_release_one(const char *snapname, const char *holdname)
4862 {
4863 	nvlist_t *snaps, *holds;
4864 	int error;
4865 
4866 	snaps = fnvlist_alloc();
4867 	holds = fnvlist_alloc();
4868 	fnvlist_add_boolean(holds, holdname);
4869 	fnvlist_add_nvlist(snaps, snapname, holds);
4870 	fnvlist_free(holds);
4871 	error = dsl_dataset_user_release(snaps, NULL);
4872 	fnvlist_free(snaps);
4873 	return (error);
4874 }
4875 
4876 /*
4877  * Test snapshot hold/release and deferred destroy.
4878  */
4879 void
4880 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4881 {
4882 	int error;
4883 	objset_t *os = zd->zd_os;
4884 	objset_t *origin;
4885 	char snapname[100];
4886 	char fullname[100];
4887 	char clonename[100];
4888 	char tag[100];
4889 	char osname[ZFS_MAX_DATASET_NAME_LEN];
4890 	nvlist_t *holds;
4891 
4892 	rw_enter(&ztest_name_lock, RW_READER);
4893 
4894 	dmu_objset_name(os, osname);
4895 
4896 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4897 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4898 	(void) snprintf(clonename, sizeof (clonename),
4899 	    "%s/ch1_%llu", osname, id);
4900 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4901 
4902 	/*
4903 	 * Clean up from any previous run.
4904 	 */
4905 	error = dsl_destroy_head(clonename);
4906 	if (error != ENOENT)
4907 		ASSERT0(error);
4908 	error = user_release_one(fullname, tag);
4909 	if (error != ESRCH && error != ENOENT)
4910 		ASSERT0(error);
4911 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4912 	if (error != ENOENT)
4913 		ASSERT0(error);
4914 
4915 	/*
4916 	 * Create snapshot, clone it, mark snap for deferred destroy,
4917 	 * destroy clone, verify snap was also destroyed.
4918 	 */
4919 	error = dmu_objset_snapshot_one(osname, snapname);
4920 	if (error) {
4921 		if (error == ENOSPC) {
4922 			ztest_record_enospc("dmu_objset_snapshot");
4923 			goto out;
4924 		}
4925 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4926 	}
4927 
4928 	error = dmu_objset_clone(clonename, fullname);
4929 	if (error) {
4930 		if (error == ENOSPC) {
4931 			ztest_record_enospc("dmu_objset_clone");
4932 			goto out;
4933 		}
4934 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4935 	}
4936 
4937 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4938 	if (error) {
4939 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4940 		    fullname, error);
4941 	}
4942 
4943 	error = dsl_destroy_head(clonename);
4944 	if (error)
4945 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4946 
4947 	error = dmu_objset_hold(fullname, FTAG, &origin);
4948 	if (error != ENOENT)
4949 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4950 
4951 	/*
4952 	 * Create snapshot, add temporary hold, verify that we can't
4953 	 * destroy a held snapshot, mark for deferred destroy,
4954 	 * release hold, verify snapshot was destroyed.
4955 	 */
4956 	error = dmu_objset_snapshot_one(osname, snapname);
4957 	if (error) {
4958 		if (error == ENOSPC) {
4959 			ztest_record_enospc("dmu_objset_snapshot");
4960 			goto out;
4961 		}
4962 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4963 	}
4964 
4965 	holds = fnvlist_alloc();
4966 	fnvlist_add_string(holds, fullname, tag);
4967 	error = dsl_dataset_user_hold(holds, 0, NULL);
4968 	fnvlist_free(holds);
4969 
4970 	if (error == ENOSPC) {
4971 		ztest_record_enospc("dsl_dataset_user_hold");
4972 		goto out;
4973 	} else if (error) {
4974 		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4975 		    fullname, tag, error);
4976 	}
4977 
4978 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4979 	if (error != EBUSY) {
4980 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4981 		    fullname, error);
4982 	}
4983 
4984 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4985 	if (error) {
4986 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4987 		    fullname, error);
4988 	}
4989 
4990 	error = user_release_one(fullname, tag);
4991 	if (error)
4992 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4993 
4994 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4995 
4996 out:
4997 	rw_exit(&ztest_name_lock);
4998 }
4999 
5000 /*
5001  * Inject random faults into the on-disk data.
5002  */
5003 /* ARGSUSED */
5004 void
5005 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
5006 {
5007 	ztest_shared_t *zs = ztest_shared;
5008 	spa_t *spa = ztest_spa;
5009 	int fd;
5010 	uint64_t offset;
5011 	uint64_t leaves;
5012 	uint64_t bad = 0x1990c0ffeedecade;
5013 	uint64_t top, leaf;
5014 	char path0[MAXPATHLEN];
5015 	char pathrand[MAXPATHLEN];
5016 	size_t fsize;
5017 	int bshift = SPA_MAXBLOCKSHIFT + 2;
5018 	int iters = 1000;
5019 	int maxfaults;
5020 	int mirror_save;
5021 	vdev_t *vd0 = NULL;
5022 	uint64_t guid0 = 0;
5023 	boolean_t islog = B_FALSE;
5024 
5025 	mutex_enter(&ztest_vdev_lock);
5026 
5027 	/*
5028 	 * Device removal is in progress, fault injection must be disabled
5029 	 * until it completes and the pool is scrubbed.  The fault injection
5030 	 * strategy for damaging blocks does not take in to account evacuated
5031 	 * blocks which may have already been damaged.
5032 	 */
5033 	if (ztest_device_removal_active) {
5034 		mutex_exit(&ztest_vdev_lock);
5035 		return;
5036 	}
5037 
5038 	maxfaults = MAXFAULTS();
5039 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
5040 	mirror_save = zs->zs_mirrors;
5041 	mutex_exit(&ztest_vdev_lock);
5042 
5043 	ASSERT(leaves >= 1);
5044 
5045 	/*
5046 	 * Grab the name lock as reader. There are some operations
5047 	 * which don't like to have their vdevs changed while
5048 	 * they are in progress (i.e. spa_change_guid). Those
5049 	 * operations will have grabbed the name lock as writer.
5050 	 */
5051 	rw_enter(&ztest_name_lock, RW_READER);
5052 
5053 	/*
5054 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5055 	 */
5056 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5057 
5058 	if (ztest_random(2) == 0) {
5059 		/*
5060 		 * Inject errors on a normal data device or slog device.
5061 		 */
5062 		top = ztest_random_vdev_top(spa, B_TRUE);
5063 		leaf = ztest_random(leaves) + zs->zs_splits;
5064 
5065 		/*
5066 		 * Generate paths to the first leaf in this top-level vdev,
5067 		 * and to the random leaf we selected.  We'll induce transient
5068 		 * write failures and random online/offline activity on leaf 0,
5069 		 * and we'll write random garbage to the randomly chosen leaf.
5070 		 */
5071 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
5072 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
5073 		    top * leaves + zs->zs_splits);
5074 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
5075 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
5076 		    top * leaves + leaf);
5077 
5078 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5079 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5080 			islog = B_TRUE;
5081 
5082 		/*
5083 		 * If the top-level vdev needs to be resilvered
5084 		 * then we only allow faults on the device that is
5085 		 * resilvering.
5086 		 */
5087 		if (vd0 != NULL && maxfaults != 1 &&
5088 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5089 		    vd0->vdev_resilver_txg != 0)) {
5090 			/*
5091 			 * Make vd0 explicitly claim to be unreadable,
5092 			 * or unwriteable, or reach behind its back
5093 			 * and close the underlying fd.  We can do this if
5094 			 * maxfaults == 0 because we'll fail and reexecute,
5095 			 * and we can do it if maxfaults >= 2 because we'll
5096 			 * have enough redundancy.  If maxfaults == 1, the
5097 			 * combination of this with injection of random data
5098 			 * corruption below exceeds the pool's fault tolerance.
5099 			 */
5100 			vdev_file_t *vf = vd0->vdev_tsd;
5101 
5102 			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
5103 			    (long long)vd0->vdev_id, (int)maxfaults);
5104 
5105 			if (vf != NULL && ztest_random(3) == 0) {
5106 				(void) close(vf->vf_vnode->v_fd);
5107 				vf->vf_vnode->v_fd = -1;
5108 			} else if (ztest_random(2) == 0) {
5109 				vd0->vdev_cant_read = B_TRUE;
5110 			} else {
5111 				vd0->vdev_cant_write = B_TRUE;
5112 			}
5113 			guid0 = vd0->vdev_guid;
5114 		}
5115 	} else {
5116 		/*
5117 		 * Inject errors on an l2cache device.
5118 		 */
5119 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
5120 
5121 		if (sav->sav_count == 0) {
5122 			spa_config_exit(spa, SCL_STATE, FTAG);
5123 			rw_exit(&ztest_name_lock);
5124 			return;
5125 		}
5126 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5127 		guid0 = vd0->vdev_guid;
5128 		(void) strcpy(path0, vd0->vdev_path);
5129 		(void) strcpy(pathrand, vd0->vdev_path);
5130 
5131 		leaf = 0;
5132 		leaves = 1;
5133 		maxfaults = INT_MAX;	/* no limit on cache devices */
5134 	}
5135 
5136 	spa_config_exit(spa, SCL_STATE, FTAG);
5137 	rw_exit(&ztest_name_lock);
5138 
5139 	/*
5140 	 * If we can tolerate two or more faults, or we're dealing
5141 	 * with a slog, randomly online/offline vd0.
5142 	 */
5143 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
5144 		if (ztest_random(10) < 6) {
5145 			int flags = (ztest_random(2) == 0 ?
5146 			    ZFS_OFFLINE_TEMPORARY : 0);
5147 
5148 			/*
5149 			 * We have to grab the zs_name_lock as writer to
5150 			 * prevent a race between offlining a slog and
5151 			 * destroying a dataset. Offlining the slog will
5152 			 * grab a reference on the dataset which may cause
5153 			 * dmu_objset_destroy() to fail with EBUSY thus
5154 			 * leaving the dataset in an inconsistent state.
5155 			 */
5156 			if (islog)
5157 				rw_enter(&ztest_name_lock, RW_WRITER);
5158 
5159 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5160 
5161 			if (islog)
5162 				rw_exit(&ztest_name_lock);
5163 		} else {
5164 			/*
5165 			 * Ideally we would like to be able to randomly
5166 			 * call vdev_[on|off]line without holding locks
5167 			 * to force unpredictable failures but the side
5168 			 * effects of vdev_[on|off]line prevent us from
5169 			 * doing so. We grab the ztest_vdev_lock here to
5170 			 * prevent a race between injection testing and
5171 			 * aux_vdev removal.
5172 			 */
5173 			mutex_enter(&ztest_vdev_lock);
5174 			(void) vdev_online(spa, guid0, 0, NULL);
5175 			mutex_exit(&ztest_vdev_lock);
5176 		}
5177 	}
5178 
5179 	if (maxfaults == 0)
5180 		return;
5181 
5182 	/*
5183 	 * We have at least single-fault tolerance, so inject data corruption.
5184 	 */
5185 	fd = open(pathrand, O_RDWR);
5186 
5187 	if (fd == -1) /* we hit a gap in the device namespace */
5188 		return;
5189 
5190 	fsize = lseek(fd, 0, SEEK_END);
5191 
5192 	while (--iters != 0) {
5193 		/*
5194 		 * The offset must be chosen carefully to ensure that
5195 		 * we do not inject a given logical block with errors
5196 		 * on two different leaf devices, because ZFS can not
5197 		 * tolerate that (if maxfaults==1).
5198 		 *
5199 		 * We divide each leaf into chunks of size
5200 		 * (# leaves * SPA_MAXBLOCKSIZE * 4).  Within each chunk
5201 		 * there is a series of ranges to which we can inject errors.
5202 		 * Each range can accept errors on only a single leaf vdev.
5203 		 * The error injection ranges are separated by ranges
5204 		 * which we will not inject errors on any device (DMZs).
5205 		 * Each DMZ must be large enough such that a single block
5206 		 * can not straddle it, so that a single block can not be
5207 		 * a target in two different injection ranges (on different
5208 		 * leaf vdevs).
5209 		 *
5210 		 * For example, with 3 leaves, each chunk looks like:
5211 		 *    0 to  32M: injection range for leaf 0
5212 		 *  32M to  64M: DMZ - no injection allowed
5213 		 *  64M to  96M: injection range for leaf 1
5214 		 *  96M to 128M: DMZ - no injection allowed
5215 		 * 128M to 160M: injection range for leaf 2
5216 		 * 160M to 192M: DMZ - no injection allowed
5217 		 */
5218 		offset = ztest_random(fsize / (leaves << bshift)) *
5219 		    (leaves << bshift) + (leaf << bshift) +
5220 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5221 
5222 		/*
5223 		 * Only allow damage to the labels at one end of the vdev.
5224 		 *
5225 		 * If all labels are damaged, the device will be totally
5226 		 * inaccessible, which will result in loss of data,
5227 		 * because we also damage (parts of) the other side of
5228 		 * the mirror/raidz.
5229 		 *
5230 		 * Additionally, we will always have both an even and an
5231 		 * odd label, so that we can handle crashes in the
5232 		 * middle of vdev_config_sync().
5233 		 */
5234 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5235 			continue;
5236 
5237 		/*
5238 		 * The two end labels are stored at the "end" of the disk, but
5239 		 * the end of the disk (vdev_psize) is aligned to
5240 		 * sizeof (vdev_label_t).
5241 		 */
5242 		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5243 		if ((leaf & 1) == 1 &&
5244 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5245 			continue;
5246 
5247 		mutex_enter(&ztest_vdev_lock);
5248 		if (mirror_save != zs->zs_mirrors) {
5249 			mutex_exit(&ztest_vdev_lock);
5250 			(void) close(fd);
5251 			return;
5252 		}
5253 
5254 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5255 			fatal(1, "can't inject bad word at 0x%llx in %s",
5256 			    offset, pathrand);
5257 
5258 		mutex_exit(&ztest_vdev_lock);
5259 
5260 		if (ztest_opts.zo_verbose >= 7)
5261 			(void) printf("injected bad word into %s,"
5262 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5263 	}
5264 
5265 	(void) close(fd);
5266 }
5267 
5268 /*
5269  * Verify that DDT repair works as expected.
5270  */
5271 void
5272 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5273 {
5274 	ztest_shared_t *zs = ztest_shared;
5275 	spa_t *spa = ztest_spa;
5276 	objset_t *os = zd->zd_os;
5277 	ztest_od_t od[1];
5278 	uint64_t object, blocksize, txg, pattern, psize;
5279 	enum zio_checksum checksum = spa_dedup_checksum(spa);
5280 	dmu_buf_t *db;
5281 	dmu_tx_t *tx;
5282 	abd_t *abd;
5283 	blkptr_t blk;
5284 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
5285 
5286 	blocksize = ztest_random_blocksize();
5287 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
5288 
5289 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5290 
5291 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5292 		return;
5293 
5294 	/*
5295 	 * Take the name lock as writer to prevent anyone else from changing
5296 	 * the pool and dataset properies we need to maintain during this test.
5297 	 */
5298 	rw_enter(&ztest_name_lock, RW_WRITER);
5299 
5300 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5301 	    B_FALSE) != 0 ||
5302 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5303 	    B_FALSE) != 0) {
5304 		rw_exit(&ztest_name_lock);
5305 		return;
5306 	}
5307 
5308 	dmu_objset_stats_t dds;
5309 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5310 	dmu_objset_fast_stat(os, &dds);
5311 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5312 
5313 	object = od[0].od_object;
5314 	blocksize = od[0].od_blocksize;
5315 	pattern = zs->zs_guid ^ dds.dds_guid;
5316 
5317 	ASSERT(object != 0);
5318 
5319 	tx = dmu_tx_create(os);
5320 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5321 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5322 	if (txg == 0) {
5323 		rw_exit(&ztest_name_lock);
5324 		return;
5325 	}
5326 
5327 	/*
5328 	 * Write all the copies of our block.
5329 	 */
5330 	for (int i = 0; i < copies; i++) {
5331 		uint64_t offset = i * blocksize;
5332 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5333 		    DMU_READ_NO_PREFETCH);
5334 		if (error != 0) {
5335 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5336 			    os, (long long)object, (long long) offset, error);
5337 		}
5338 		ASSERT(db->db_offset == offset);
5339 		ASSERT(db->db_size == blocksize);
5340 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5341 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5342 		dmu_buf_will_fill(db, tx);
5343 		ztest_pattern_set(db->db_data, db->db_size, pattern);
5344 		dmu_buf_rele(db, FTAG);
5345 	}
5346 
5347 	dmu_tx_commit(tx);
5348 	txg_wait_synced(spa_get_dsl(spa), txg);
5349 
5350 	/*
5351 	 * Find out what block we got.
5352 	 */
5353 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5354 	    DMU_READ_NO_PREFETCH));
5355 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5356 	dmu_buf_rele(db, FTAG);
5357 
5358 	/*
5359 	 * Damage the block.  Dedup-ditto will save us when we read it later.
5360 	 */
5361 	psize = BP_GET_PSIZE(&blk);
5362 	abd = abd_alloc_linear(psize, B_TRUE);
5363 	ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5364 
5365 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5366 	    abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5367 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5368 
5369 	abd_free(abd);
5370 
5371 	rw_exit(&ztest_name_lock);
5372 }
5373 
5374 /*
5375  * Scrub the pool.
5376  */
5377 /* ARGSUSED */
5378 void
5379 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5380 {
5381 	spa_t *spa = ztest_spa;
5382 
5383 	/*
5384 	 * Scrub in progress by device removal.
5385 	 */
5386 	if (ztest_device_removal_active)
5387 		return;
5388 
5389 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5390 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5391 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5392 }
5393 
5394 /*
5395  * Change the guid for the pool.
5396  */
5397 /* ARGSUSED */
5398 void
5399 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5400 {
5401 	spa_t *spa = ztest_spa;
5402 	uint64_t orig, load;
5403 	int error;
5404 
5405 	orig = spa_guid(spa);
5406 	load = spa_load_guid(spa);
5407 
5408 	rw_enter(&ztest_name_lock, RW_WRITER);
5409 	error = spa_change_guid(spa);
5410 	rw_exit(&ztest_name_lock);
5411 
5412 	if (error != 0)
5413 		return;
5414 
5415 	if (ztest_opts.zo_verbose >= 4) {
5416 		(void) printf("Changed guid old %llu -> %llu\n",
5417 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5418 	}
5419 
5420 	VERIFY3U(orig, !=, spa_guid(spa));
5421 	VERIFY3U(load, ==, spa_load_guid(spa));
5422 }
5423 
5424 static vdev_t *
5425 ztest_random_concrete_vdev_leaf(vdev_t *vd)
5426 {
5427 	if (vd == NULL)
5428 		return (NULL);
5429 
5430 	if (vd->vdev_children == 0)
5431 		return (vd);
5432 
5433 	vdev_t *eligible[vd->vdev_children];
5434 	int eligible_idx = 0, i;
5435 	for (i = 0; i < vd->vdev_children; i++) {
5436 		vdev_t *cvd = vd->vdev_child[i];
5437 		if (cvd->vdev_top->vdev_removing)
5438 			continue;
5439 		if (cvd->vdev_children > 0 ||
5440 		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
5441 			eligible[eligible_idx++] = cvd;
5442 		}
5443 	}
5444 	VERIFY(eligible_idx > 0);
5445 
5446 	uint64_t child_no = ztest_random(eligible_idx);
5447 	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
5448 }
5449 
5450 /* ARGSUSED */
5451 void
5452 ztest_initialize(ztest_ds_t *zd, uint64_t id)
5453 {
5454 	spa_t *spa = ztest_spa;
5455 	int error = 0;
5456 
5457 	mutex_enter(&ztest_vdev_lock);
5458 
5459 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
5460 
5461 	/* Random leaf vdev */
5462 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
5463 	if (rand_vd == NULL) {
5464 		spa_config_exit(spa, SCL_VDEV, FTAG);
5465 		mutex_exit(&ztest_vdev_lock);
5466 		return;
5467 	}
5468 
5469 	/*
5470 	 * The random vdev we've selected may change as soon as we
5471 	 * drop the spa_config_lock. We create local copies of things
5472 	 * we're interested in.
5473 	 */
5474 	uint64_t guid = rand_vd->vdev_guid;
5475 	char *path = strdup(rand_vd->vdev_path);
5476 	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
5477 
5478 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, guid);
5479 	spa_config_exit(spa, SCL_VDEV, FTAG);
5480 
5481 	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
5482 	error = spa_vdev_initialize(spa, guid, cmd);
5483 	switch (cmd) {
5484 	case POOL_INITIALIZE_CANCEL:
5485 		if (ztest_opts.zo_verbose >= 4) {
5486 			(void) printf("Cancel initialize %s", path);
5487 			if (!active)
5488 				(void) printf(" failed (no initialize active)");
5489 			(void) printf("\n");
5490 		}
5491 		break;
5492 	case POOL_INITIALIZE_DO:
5493 		if (ztest_opts.zo_verbose >= 4) {
5494 			(void) printf("Start initialize %s", path);
5495 			if (active && error == 0)
5496 				(void) printf(" failed (already active)");
5497 			else if (error != 0)
5498 				(void) printf(" failed (error %d)", error);
5499 			(void) printf("\n");
5500 		}
5501 		break;
5502 	case POOL_INITIALIZE_SUSPEND:
5503 		if (ztest_opts.zo_verbose >= 4) {
5504 			(void) printf("Suspend initialize %s", path);
5505 			if (!active)
5506 				(void) printf(" failed (no initialize active)");
5507 			(void) printf("\n");
5508 		}
5509 		break;
5510 	}
5511 	free(path);
5512 	mutex_exit(&ztest_vdev_lock);
5513 }
5514 
5515 /*
5516  * Verify pool integrity by running zdb.
5517  */
5518 static void
5519 ztest_run_zdb(char *pool)
5520 {
5521 	int status;
5522 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5523 	char zbuf[1024];
5524 	char *bin;
5525 	char *ztest;
5526 	char *isa;
5527 	int isalen;
5528 	FILE *fp;
5529 
5530 	(void) realpath(getexecname(), zdb);
5531 
5532 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5533 	bin = strstr(zdb, "/usr/bin/");
5534 	ztest = strstr(bin, "/ztest");
5535 	isa = bin + 8;
5536 	isalen = ztest - isa;
5537 	isa = strdup(isa);
5538 	/* LINTED */
5539 	(void) sprintf(bin,
5540 	    "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s %s",
5541 	    isalen,
5542 	    isa,
5543 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5544 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5545 	    spa_config_path,
5546 	    pool);
5547 	free(isa);
5548 
5549 	if (ztest_opts.zo_verbose >= 5)
5550 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5551 
5552 	fp = popen(zdb, "r");
5553 
5554 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5555 		if (ztest_opts.zo_verbose >= 3)
5556 			(void) printf("%s", zbuf);
5557 
5558 	status = pclose(fp);
5559 
5560 	if (status == 0)
5561 		return;
5562 
5563 	ztest_dump_core = 0;
5564 	if (WIFEXITED(status))
5565 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5566 	else
5567 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5568 }
5569 
5570 static void
5571 ztest_walk_pool_directory(char *header)
5572 {
5573 	spa_t *spa = NULL;
5574 
5575 	if (ztest_opts.zo_verbose >= 6)
5576 		(void) printf("%s\n", header);
5577 
5578 	mutex_enter(&spa_namespace_lock);
5579 	while ((spa = spa_next(spa)) != NULL)
5580 		if (ztest_opts.zo_verbose >= 6)
5581 			(void) printf("\t%s\n", spa_name(spa));
5582 	mutex_exit(&spa_namespace_lock);
5583 }
5584 
5585 static void
5586 ztest_spa_import_export(char *oldname, char *newname)
5587 {
5588 	nvlist_t *config, *newconfig;
5589 	uint64_t pool_guid;
5590 	spa_t *spa;
5591 	int error;
5592 
5593 	if (ztest_opts.zo_verbose >= 4) {
5594 		(void) printf("import/export: old = %s, new = %s\n",
5595 		    oldname, newname);
5596 	}
5597 
5598 	/*
5599 	 * Clean up from previous runs.
5600 	 */
5601 	(void) spa_destroy(newname);
5602 
5603 	/*
5604 	 * Get the pool's configuration and guid.
5605 	 */
5606 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5607 
5608 	/*
5609 	 * Kick off a scrub to tickle scrub/export races.
5610 	 */
5611 	if (ztest_random(2) == 0)
5612 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5613 
5614 	pool_guid = spa_guid(spa);
5615 	spa_close(spa, FTAG);
5616 
5617 	ztest_walk_pool_directory("pools before export");
5618 
5619 	/*
5620 	 * Export it.
5621 	 */
5622 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5623 
5624 	ztest_walk_pool_directory("pools after export");
5625 
5626 	/*
5627 	 * Try to import it.
5628 	 */
5629 	newconfig = spa_tryimport(config);
5630 	ASSERT(newconfig != NULL);
5631 	nvlist_free(newconfig);
5632 
5633 	/*
5634 	 * Import it under the new name.
5635 	 */
5636 	error = spa_import(newname, config, NULL, 0);
5637 	if (error != 0) {
5638 		dump_nvlist(config, 0);
5639 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5640 		    oldname, newname, error);
5641 	}
5642 
5643 	ztest_walk_pool_directory("pools after import");
5644 
5645 	/*
5646 	 * Try to import it again -- should fail with EEXIST.
5647 	 */
5648 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5649 
5650 	/*
5651 	 * Try to import it under a different name -- should fail with EEXIST.
5652 	 */
5653 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5654 
5655 	/*
5656 	 * Verify that the pool is no longer visible under the old name.
5657 	 */
5658 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5659 
5660 	/*
5661 	 * Verify that we can open and close the pool using the new name.
5662 	 */
5663 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5664 	ASSERT(pool_guid == spa_guid(spa));
5665 	spa_close(spa, FTAG);
5666 
5667 	nvlist_free(config);
5668 }
5669 
5670 static void
5671 ztest_resume(spa_t *spa)
5672 {
5673 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5674 		(void) printf("resuming from suspended state\n");
5675 	spa_vdev_state_enter(spa, SCL_NONE);
5676 	vdev_clear(spa, NULL);
5677 	(void) spa_vdev_state_exit(spa, NULL, 0);
5678 	(void) zio_resume(spa);
5679 }
5680 
5681 static void *
5682 ztest_resume_thread(void *arg)
5683 {
5684 	spa_t *spa = arg;
5685 
5686 	while (!ztest_exiting) {
5687 		if (spa_suspended(spa))
5688 			ztest_resume(spa);
5689 		(void) poll(NULL, 0, 100);
5690 
5691 		/*
5692 		 * Periodically change the zfs_compressed_arc_enabled setting.
5693 		 */
5694 		if (ztest_random(10) == 0)
5695 			zfs_compressed_arc_enabled = ztest_random(2);
5696 
5697 		/*
5698 		 * Periodically change the zfs_abd_scatter_enabled setting.
5699 		 */
5700 		if (ztest_random(10) == 0)
5701 			zfs_abd_scatter_enabled = ztest_random(2);
5702 	}
5703 	return (NULL);
5704 }
5705 
5706 static void *
5707 ztest_deadman_thread(void *arg)
5708 {
5709 	ztest_shared_t *zs = arg;
5710 	spa_t *spa = ztest_spa;
5711 	hrtime_t delta, total = 0;
5712 
5713 	for (;;) {
5714 		delta = zs->zs_thread_stop - zs->zs_thread_start +
5715 		    MSEC2NSEC(zfs_deadman_synctime_ms);
5716 
5717 		(void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5718 
5719 		/*
5720 		 * If the pool is suspended then fail immediately. Otherwise,
5721 		 * check to see if the pool is making any progress. If
5722 		 * vdev_deadman() discovers that there hasn't been any recent
5723 		 * I/Os then it will end up aborting the tests.
5724 		 */
5725 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5726 			fatal(0, "aborting test after %llu seconds because "
5727 			    "pool has transitioned to a suspended state.",
5728 			    zfs_deadman_synctime_ms / 1000);
5729 			return (NULL);
5730 		}
5731 		vdev_deadman(spa->spa_root_vdev);
5732 
5733 		total += zfs_deadman_synctime_ms/1000;
5734 		(void) printf("ztest has been running for %lld seconds\n",
5735 		    total);
5736 	}
5737 }
5738 
5739 static void
5740 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5741 {
5742 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5743 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5744 	hrtime_t functime = gethrtime();
5745 
5746 	for (int i = 0; i < zi->zi_iters; i++)
5747 		zi->zi_func(zd, id);
5748 
5749 	functime = gethrtime() - functime;
5750 
5751 	atomic_add_64(&zc->zc_count, 1);
5752 	atomic_add_64(&zc->zc_time, functime);
5753 
5754 	if (ztest_opts.zo_verbose >= 4) {
5755 		Dl_info dli;
5756 		(void) dladdr((void *)zi->zi_func, &dli);
5757 		(void) printf("%6.2f sec in %s\n",
5758 		    (double)functime / NANOSEC, dli.dli_sname);
5759 	}
5760 }
5761 
5762 static void *
5763 ztest_thread(void *arg)
5764 {
5765 	int rand;
5766 	uint64_t id = (uintptr_t)arg;
5767 	ztest_shared_t *zs = ztest_shared;
5768 	uint64_t call_next;
5769 	hrtime_t now;
5770 	ztest_info_t *zi;
5771 	ztest_shared_callstate_t *zc;
5772 
5773 	while ((now = gethrtime()) < zs->zs_thread_stop) {
5774 		/*
5775 		 * See if it's time to force a crash.
5776 		 */
5777 		if (now > zs->zs_thread_kill)
5778 			ztest_kill(zs);
5779 
5780 		/*
5781 		 * If we're getting ENOSPC with some regularity, stop.
5782 		 */
5783 		if (zs->zs_enospc_count > 10)
5784 			break;
5785 
5786 		/*
5787 		 * Pick a random function to execute.
5788 		 */
5789 		rand = ztest_random(ZTEST_FUNCS);
5790 		zi = &ztest_info[rand];
5791 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5792 		call_next = zc->zc_next;
5793 
5794 		if (now >= call_next &&
5795 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
5796 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5797 			ztest_execute(rand, zi, id);
5798 		}
5799 	}
5800 
5801 	return (NULL);
5802 }
5803 
5804 static void
5805 ztest_dataset_name(char *dsname, char *pool, int d)
5806 {
5807 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5808 }
5809 
5810 static void
5811 ztest_dataset_destroy(int d)
5812 {
5813 	char name[ZFS_MAX_DATASET_NAME_LEN];
5814 
5815 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5816 
5817 	if (ztest_opts.zo_verbose >= 3)
5818 		(void) printf("Destroying %s to free up space\n", name);
5819 
5820 	/*
5821 	 * Cleanup any non-standard clones and snapshots.  In general,
5822 	 * ztest thread t operates on dataset (t % zopt_datasets),
5823 	 * so there may be more than one thing to clean up.
5824 	 */
5825 	for (int t = d; t < ztest_opts.zo_threads;
5826 	    t += ztest_opts.zo_datasets) {
5827 		ztest_dsl_dataset_cleanup(name, t);
5828 	}
5829 
5830 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5831 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5832 }
5833 
5834 static void
5835 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5836 {
5837 	uint64_t usedobjs, dirobjs, scratch;
5838 
5839 	/*
5840 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
5841 	 * Therefore, the number of objects in use should equal the
5842 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5843 	 * If not, we have an object leak.
5844 	 *
5845 	 * Note that we can only check this in ztest_dataset_open(),
5846 	 * when the open-context and syncing-context values agree.
5847 	 * That's because zap_count() returns the open-context value,
5848 	 * while dmu_objset_space() returns the rootbp fill count.
5849 	 */
5850 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5851 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5852 	ASSERT3U(dirobjs + 1, ==, usedobjs);
5853 }
5854 
5855 static int
5856 ztest_dataset_open(int d)
5857 {
5858 	ztest_ds_t *zd = &ztest_ds[d];
5859 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5860 	objset_t *os;
5861 	zilog_t *zilog;
5862 	char name[ZFS_MAX_DATASET_NAME_LEN];
5863 	int error;
5864 
5865 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5866 
5867 	rw_enter(&ztest_name_lock, RW_READER);
5868 
5869 	error = ztest_dataset_create(name);
5870 	if (error == ENOSPC) {
5871 		rw_exit(&ztest_name_lock);
5872 		ztest_record_enospc(FTAG);
5873 		return (error);
5874 	}
5875 	ASSERT(error == 0 || error == EEXIST);
5876 
5877 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5878 	rw_exit(&ztest_name_lock);
5879 
5880 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5881 
5882 	zilog = zd->zd_zilog;
5883 
5884 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5885 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
5886 		fatal(0, "missing log records: claimed %llu < committed %llu",
5887 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
5888 
5889 	ztest_dataset_dirobj_verify(zd);
5890 
5891 	zil_replay(os, zd, ztest_replay_vector);
5892 
5893 	ztest_dataset_dirobj_verify(zd);
5894 
5895 	if (ztest_opts.zo_verbose >= 6)
5896 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5897 		    zd->zd_name,
5898 		    (u_longlong_t)zilog->zl_parse_blk_count,
5899 		    (u_longlong_t)zilog->zl_parse_lr_count,
5900 		    (u_longlong_t)zilog->zl_replaying_seq);
5901 
5902 	zilog = zil_open(os, ztest_get_data);
5903 
5904 	if (zilog->zl_replaying_seq != 0 &&
5905 	    zilog->zl_replaying_seq < committed_seq)
5906 		fatal(0, "missing log records: replayed %llu < committed %llu",
5907 		    zilog->zl_replaying_seq, committed_seq);
5908 
5909 	return (0);
5910 }
5911 
5912 static void
5913 ztest_dataset_close(int d)
5914 {
5915 	ztest_ds_t *zd = &ztest_ds[d];
5916 
5917 	zil_close(zd->zd_zilog);
5918 	dmu_objset_disown(zd->zd_os, zd);
5919 
5920 	ztest_zd_fini(zd);
5921 }
5922 
5923 /*
5924  * Kick off threads to run tests on all datasets in parallel.
5925  */
5926 static void
5927 ztest_run(ztest_shared_t *zs)
5928 {
5929 	thread_t *tid;
5930 	spa_t *spa;
5931 	objset_t *os;
5932 	thread_t resume_tid;
5933 	int error;
5934 
5935 	ztest_exiting = B_FALSE;
5936 
5937 	/*
5938 	 * Initialize parent/child shared state.
5939 	 */
5940 	mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL);
5941 	mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
5942 	rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
5943 
5944 	zs->zs_thread_start = gethrtime();
5945 	zs->zs_thread_stop =
5946 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5947 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5948 	zs->zs_thread_kill = zs->zs_thread_stop;
5949 	if (ztest_random(100) < ztest_opts.zo_killrate) {
5950 		zs->zs_thread_kill -=
5951 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
5952 	}
5953 
5954 	mutex_init(&zcl.zcl_callbacks_lock, NULL, USYNC_THREAD, NULL);
5955 
5956 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5957 	    offsetof(ztest_cb_data_t, zcd_node));
5958 
5959 	/*
5960 	 * Open our pool.
5961 	 */
5962 	kernel_init(FREAD | FWRITE);
5963 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5964 	metaslab_preload_limit = ztest_random(20) + 1;
5965 	ztest_spa = spa;
5966 
5967 	dmu_objset_stats_t dds;
5968 	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5969 	    DMU_OST_ANY, B_TRUE, FTAG, &os));
5970 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5971 	dmu_objset_fast_stat(os, &dds);
5972 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5973 	zs->zs_guid = dds.dds_guid;
5974 	dmu_objset_disown(os, FTAG);
5975 
5976 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5977 
5978 	/*
5979 	 * We don't expect the pool to suspend unless maxfaults == 0,
5980 	 * in which case ztest_fault_inject() temporarily takes away
5981 	 * the only valid replica.
5982 	 */
5983 	if (MAXFAULTS() == 0)
5984 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5985 	else
5986 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5987 
5988 	/*
5989 	 * Create a thread to periodically resume suspended I/O.
5990 	 */
5991 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5992 	    &resume_tid) == 0);
5993 
5994 	/*
5995 	 * Create a deadman thread to abort() if we hang.
5996 	 */
5997 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5998 	    NULL) == 0);
5999 
6000 	/*
6001 	 * Verify that we can safely inquire about any object,
6002 	 * whether it's allocated or not.  To make it interesting,
6003 	 * we probe a 5-wide window around each power of two.
6004 	 * This hits all edge cases, including zero and the max.
6005 	 */
6006 	for (int t = 0; t < 64; t++) {
6007 		for (int d = -5; d <= 5; d++) {
6008 			error = dmu_object_info(spa->spa_meta_objset,
6009 			    (1ULL << t) + d, NULL);
6010 			ASSERT(error == 0 || error == ENOENT ||
6011 			    error == EINVAL);
6012 		}
6013 	}
6014 
6015 	/*
6016 	 * If we got any ENOSPC errors on the previous run, destroy something.
6017 	 */
6018 	if (zs->zs_enospc_count != 0) {
6019 		int d = ztest_random(ztest_opts.zo_datasets);
6020 		ztest_dataset_destroy(d);
6021 	}
6022 	zs->zs_enospc_count = 0;
6023 
6024 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
6025 	    UMEM_NOFAIL);
6026 
6027 	if (ztest_opts.zo_verbose >= 4)
6028 		(void) printf("starting main threads...\n");
6029 
6030 	/*
6031 	 * Kick off all the tests that run in parallel.
6032 	 */
6033 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
6034 		if (t < ztest_opts.zo_datasets &&
6035 		    ztest_dataset_open(t) != 0)
6036 			return;
6037 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
6038 		    THR_BOUND, &tid[t]) == 0);
6039 	}
6040 
6041 	/*
6042 	 * Wait for all of the tests to complete.  We go in reverse order
6043 	 * so we don't close datasets while threads are still using them.
6044 	 */
6045 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
6046 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
6047 		if (t < ztest_opts.zo_datasets)
6048 			ztest_dataset_close(t);
6049 	}
6050 
6051 	txg_wait_synced(spa_get_dsl(spa), 0);
6052 
6053 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
6054 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
6055 	zfs_dbgmsg_print(FTAG);
6056 
6057 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
6058 
6059 	/* Kill the resume thread */
6060 	ztest_exiting = B_TRUE;
6061 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
6062 	ztest_resume(spa);
6063 
6064 	/*
6065 	 * Right before closing the pool, kick off a bunch of async I/O;
6066 	 * spa_close() should wait for it to complete.
6067 	 */
6068 	for (uint64_t object = 1; object < 50; object++) {
6069 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
6070 		    ZIO_PRIORITY_SYNC_READ);
6071 	}
6072 
6073 	spa_close(spa, FTAG);
6074 
6075 	/*
6076 	 * Verify that we can loop over all pools.
6077 	 */
6078 	mutex_enter(&spa_namespace_lock);
6079 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
6080 		if (ztest_opts.zo_verbose > 3)
6081 			(void) printf("spa_next: found %s\n", spa_name(spa));
6082 	mutex_exit(&spa_namespace_lock);
6083 
6084 	/*
6085 	 * Verify that we can export the pool and reimport it under a
6086 	 * different name.
6087 	 */
6088 	if (ztest_random(2) == 0) {
6089 		char name[ZFS_MAX_DATASET_NAME_LEN];
6090 		(void) snprintf(name, sizeof (name), "%s_import",
6091 		    ztest_opts.zo_pool);
6092 		ztest_spa_import_export(ztest_opts.zo_pool, name);
6093 		ztest_spa_import_export(name, ztest_opts.zo_pool);
6094 	}
6095 
6096 	kernel_fini();
6097 
6098 	list_destroy(&zcl.zcl_callbacks);
6099 
6100 	mutex_destroy(&zcl.zcl_callbacks_lock);
6101 
6102 	rw_destroy(&ztest_name_lock);
6103 	mutex_destroy(&ztest_vdev_lock);
6104 	mutex_destroy(&ztest_checkpoint_lock);
6105 }
6106 
6107 static void
6108 ztest_freeze(void)
6109 {
6110 	ztest_ds_t *zd = &ztest_ds[0];
6111 	spa_t *spa;
6112 	int numloops = 0;
6113 
6114 	if (ztest_opts.zo_verbose >= 3)
6115 		(void) printf("testing spa_freeze()...\n");
6116 
6117 	kernel_init(FREAD | FWRITE);
6118 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6119 	VERIFY3U(0, ==, ztest_dataset_open(0));
6120 	ztest_spa = spa;
6121 
6122 	/*
6123 	 * Force the first log block to be transactionally allocated.
6124 	 * We have to do this before we freeze the pool -- otherwise
6125 	 * the log chain won't be anchored.
6126 	 */
6127 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
6128 		ztest_dmu_object_alloc_free(zd, 0);
6129 		zil_commit(zd->zd_zilog, 0);
6130 	}
6131 
6132 	txg_wait_synced(spa_get_dsl(spa), 0);
6133 
6134 	/*
6135 	 * Freeze the pool.  This stops spa_sync() from doing anything,
6136 	 * so that the only way to record changes from now on is the ZIL.
6137 	 */
6138 	spa_freeze(spa);
6139 
6140 	/*
6141 	 * Because it is hard to predict how much space a write will actually
6142 	 * require beforehand, we leave ourselves some fudge space to write over
6143 	 * capacity.
6144 	 */
6145 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6146 
6147 	/*
6148 	 * Run tests that generate log records but don't alter the pool config
6149 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6150 	 * We do a txg_wait_synced() after each iteration to force the txg
6151 	 * to increase well beyond the last synced value in the uberblock.
6152 	 * The ZIL should be OK with that.
6153 	 *
6154 	 * Run a random number of times less than zo_maxloops and ensure we do
6155 	 * not run out of space on the pool.
6156 	 */
6157 	while (ztest_random(10) != 0 &&
6158 	    numloops++ < ztest_opts.zo_maxloops &&
6159 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6160 		ztest_od_t od;
6161 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
6162 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6163 		ztest_io(zd, od.od_object,
6164 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6165 		txg_wait_synced(spa_get_dsl(spa), 0);
6166 	}
6167 
6168 	/*
6169 	 * Commit all of the changes we just generated.
6170 	 */
6171 	zil_commit(zd->zd_zilog, 0);
6172 	txg_wait_synced(spa_get_dsl(spa), 0);
6173 
6174 	/*
6175 	 * Close our dataset and close the pool.
6176 	 */
6177 	ztest_dataset_close(0);
6178 	spa_close(spa, FTAG);
6179 	kernel_fini();
6180 
6181 	/*
6182 	 * Open and close the pool and dataset to induce log replay.
6183 	 */
6184 	kernel_init(FREAD | FWRITE);
6185 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6186 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6187 	VERIFY3U(0, ==, ztest_dataset_open(0));
6188 	ztest_dataset_close(0);
6189 
6190 	ztest_spa = spa;
6191 	txg_wait_synced(spa_get_dsl(spa), 0);
6192 	ztest_reguid(NULL, 0);
6193 
6194 	spa_close(spa, FTAG);
6195 	kernel_fini();
6196 }
6197 
6198 void
6199 print_time(hrtime_t t, char *timebuf)
6200 {
6201 	hrtime_t s = t / NANOSEC;
6202 	hrtime_t m = s / 60;
6203 	hrtime_t h = m / 60;
6204 	hrtime_t d = h / 24;
6205 
6206 	s -= m * 60;
6207 	m -= h * 60;
6208 	h -= d * 24;
6209 
6210 	timebuf[0] = '\0';
6211 
6212 	if (d)
6213 		(void) sprintf(timebuf,
6214 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
6215 	else if (h)
6216 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6217 	else if (m)
6218 		(void) sprintf(timebuf, "%llum%02llus", m, s);
6219 	else
6220 		(void) sprintf(timebuf, "%llus", s);
6221 }
6222 
6223 static nvlist_t *
6224 make_random_props()
6225 {
6226 	nvlist_t *props;
6227 
6228 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6229 	if (ztest_random(2) == 0)
6230 		return (props);
6231 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6232 
6233 	return (props);
6234 }
6235 
6236 /*
6237  * Create a storage pool with the given name and initial vdev size.
6238  * Then test spa_freeze() functionality.
6239  */
6240 static void
6241 ztest_init(ztest_shared_t *zs)
6242 {
6243 	spa_t *spa;
6244 	nvlist_t *nvroot, *props;
6245 
6246 	mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
6247 	mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL);
6248 	rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
6249 
6250 	kernel_init(FREAD | FWRITE);
6251 
6252 	/*
6253 	 * Create the storage pool.
6254 	 */
6255 	(void) spa_destroy(ztest_opts.zo_pool);
6256 	ztest_shared->zs_vdev_next_leaf = 0;
6257 	zs->zs_splits = 0;
6258 	zs->zs_mirrors = ztest_opts.zo_mirrors;
6259 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6260 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6261 	props = make_random_props();
6262 	for (int i = 0; i < SPA_FEATURES; i++) {
6263 		char buf[1024];
6264 		(void) snprintf(buf, sizeof (buf), "feature@%s",
6265 		    spa_feature_table[i].fi_uname);
6266 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6267 	}
6268 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6269 	nvlist_free(nvroot);
6270 	nvlist_free(props);
6271 
6272 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6273 	zs->zs_metaslab_sz =
6274 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6275 
6276 	spa_close(spa, FTAG);
6277 
6278 	kernel_fini();
6279 
6280 	ztest_run_zdb(ztest_opts.zo_pool);
6281 
6282 	ztest_freeze();
6283 
6284 	ztest_run_zdb(ztest_opts.zo_pool);
6285 
6286 	rw_destroy(&ztest_name_lock);
6287 	mutex_destroy(&ztest_vdev_lock);
6288 	mutex_destroy(&ztest_checkpoint_lock);
6289 }
6290 
6291 static void
6292 setup_data_fd(void)
6293 {
6294 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6295 
6296 	ztest_fd_data = mkstemp(ztest_name_data);
6297 	ASSERT3S(ztest_fd_data, >=, 0);
6298 	(void) unlink(ztest_name_data);
6299 }
6300 
6301 
6302 static int
6303 shared_data_size(ztest_shared_hdr_t *hdr)
6304 {
6305 	int size;
6306 
6307 	size = hdr->zh_hdr_size;
6308 	size += hdr->zh_opts_size;
6309 	size += hdr->zh_size;
6310 	size += hdr->zh_stats_size * hdr->zh_stats_count;
6311 	size += hdr->zh_ds_size * hdr->zh_ds_count;
6312 
6313 	return (size);
6314 }
6315 
6316 static void
6317 setup_hdr(void)
6318 {
6319 	int size;
6320 	ztest_shared_hdr_t *hdr;
6321 
6322 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6323 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6324 	ASSERT(hdr != MAP_FAILED);
6325 
6326 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6327 
6328 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6329 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6330 	hdr->zh_size = sizeof (ztest_shared_t);
6331 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6332 	hdr->zh_stats_count = ZTEST_FUNCS;
6333 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6334 	hdr->zh_ds_count = ztest_opts.zo_datasets;
6335 
6336 	size = shared_data_size(hdr);
6337 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6338 
6339 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6340 }
6341 
6342 static void
6343 setup_data(void)
6344 {
6345 	int size, offset;
6346 	ztest_shared_hdr_t *hdr;
6347 	uint8_t *buf;
6348 
6349 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6350 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6351 	ASSERT(hdr != MAP_FAILED);
6352 
6353 	size = shared_data_size(hdr);
6354 
6355 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6356 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6357 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6358 	ASSERT(hdr != MAP_FAILED);
6359 	buf = (uint8_t *)hdr;
6360 
6361 	offset = hdr->zh_hdr_size;
6362 	ztest_shared_opts = (void *)&buf[offset];
6363 	offset += hdr->zh_opts_size;
6364 	ztest_shared = (void *)&buf[offset];
6365 	offset += hdr->zh_size;
6366 	ztest_shared_callstate = (void *)&buf[offset];
6367 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
6368 	ztest_shared_ds = (void *)&buf[offset];
6369 }
6370 
6371 static boolean_t
6372 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6373 {
6374 	pid_t pid;
6375 	int status;
6376 	char *cmdbuf = NULL;
6377 
6378 	pid = fork();
6379 
6380 	if (cmd == NULL) {
6381 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6382 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6383 		cmd = cmdbuf;
6384 	}
6385 
6386 	if (pid == -1)
6387 		fatal(1, "fork failed");
6388 
6389 	if (pid == 0) {	/* child */
6390 		char *emptyargv[2] = { cmd, NULL };
6391 		char fd_data_str[12];
6392 
6393 		struct rlimit rl = { 1024, 1024 };
6394 		(void) setrlimit(RLIMIT_NOFILE, &rl);
6395 
6396 		(void) close(ztest_fd_rand);
6397 		VERIFY3U(11, >=,
6398 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6399 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6400 
6401 		(void) enable_extended_FILE_stdio(-1, -1);
6402 		if (libpath != NULL)
6403 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6404 		(void) execv(cmd, emptyargv);
6405 		ztest_dump_core = B_FALSE;
6406 		fatal(B_TRUE, "exec failed: %s", cmd);
6407 	}
6408 
6409 	if (cmdbuf != NULL) {
6410 		umem_free(cmdbuf, MAXPATHLEN);
6411 		cmd = NULL;
6412 	}
6413 
6414 	while (waitpid(pid, &status, 0) != pid)
6415 		continue;
6416 	if (statusp != NULL)
6417 		*statusp = status;
6418 
6419 	if (WIFEXITED(status)) {
6420 		if (WEXITSTATUS(status) != 0) {
6421 			(void) fprintf(stderr, "child exited with code %d\n",
6422 			    WEXITSTATUS(status));
6423 			exit(2);
6424 		}
6425 		return (B_FALSE);
6426 	} else if (WIFSIGNALED(status)) {
6427 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6428 			(void) fprintf(stderr, "child died with signal %d\n",
6429 			    WTERMSIG(status));
6430 			exit(3);
6431 		}
6432 		return (B_TRUE);
6433 	} else {
6434 		(void) fprintf(stderr, "something strange happened to child\n");
6435 		exit(4);
6436 		/* NOTREACHED */
6437 	}
6438 }
6439 
6440 static void
6441 ztest_run_init(void)
6442 {
6443 	ztest_shared_t *zs = ztest_shared;
6444 
6445 	ASSERT(ztest_opts.zo_init != 0);
6446 
6447 	/*
6448 	 * Blow away any existing copy of zpool.cache
6449 	 */
6450 	(void) remove(spa_config_path);
6451 
6452 	/*
6453 	 * Create and initialize our storage pool.
6454 	 */
6455 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
6456 		bzero(zs, sizeof (ztest_shared_t));
6457 		if (ztest_opts.zo_verbose >= 3 &&
6458 		    ztest_opts.zo_init != 1) {
6459 			(void) printf("ztest_init(), pass %d\n", i);
6460 		}
6461 		ztest_init(zs);
6462 	}
6463 }
6464 
6465 int
6466 main(int argc, char **argv)
6467 {
6468 	int kills = 0;
6469 	int iters = 0;
6470 	int older = 0;
6471 	int newer = 0;
6472 	ztest_shared_t *zs;
6473 	ztest_info_t *zi;
6474 	ztest_shared_callstate_t *zc;
6475 	char timebuf[100];
6476 	char numbuf[NN_NUMBUF_SZ];
6477 	char *cmd;
6478 	boolean_t hasalt;
6479 	char *fd_data_str = getenv("ZTEST_FD_DATA");
6480 
6481 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
6482 
6483 	dprintf_setup(&argc, argv);
6484 	zfs_deadman_synctime_ms = 300000;
6485 	/*
6486 	 * As two-word space map entries may not come up often (especially
6487 	 * if pool and vdev sizes are small) we want to force at least some
6488 	 * of them so the feature get tested.
6489 	 */
6490 	zfs_force_some_double_word_sm_entries = B_TRUE;
6491 
6492 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6493 	ASSERT3S(ztest_fd_rand, >=, 0);
6494 
6495 	if (!fd_data_str) {
6496 		process_options(argc, argv);
6497 
6498 		setup_data_fd();
6499 		setup_hdr();
6500 		setup_data();
6501 		bcopy(&ztest_opts, ztest_shared_opts,
6502 		    sizeof (*ztest_shared_opts));
6503 	} else {
6504 		ztest_fd_data = atoi(fd_data_str);
6505 		setup_data();
6506 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6507 	}
6508 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6509 
6510 	/* Override location of zpool.cache */
6511 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6512 	    ztest_opts.zo_dir), !=, -1);
6513 
6514 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6515 	    UMEM_NOFAIL);
6516 	zs = ztest_shared;
6517 
6518 	if (fd_data_str) {
6519 		metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
6520 		metaslab_df_alloc_threshold =
6521 		    zs->zs_metaslab_df_alloc_threshold;
6522 
6523 		if (zs->zs_do_init)
6524 			ztest_run_init();
6525 		else
6526 			ztest_run(zs);
6527 		exit(0);
6528 	}
6529 
6530 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6531 
6532 	if (ztest_opts.zo_verbose >= 1) {
6533 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6534 		    " %llu seconds...\n",
6535 		    (u_longlong_t)ztest_opts.zo_vdevs,
6536 		    ztest_opts.zo_datasets,
6537 		    ztest_opts.zo_threads,
6538 		    (u_longlong_t)ztest_opts.zo_time);
6539 	}
6540 
6541 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6542 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6543 
6544 	zs->zs_do_init = B_TRUE;
6545 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6546 		if (ztest_opts.zo_verbose >= 1) {
6547 			(void) printf("Executing older ztest for "
6548 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6549 		}
6550 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6551 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6552 	} else {
6553 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6554 	}
6555 	zs->zs_do_init = B_FALSE;
6556 
6557 	zs->zs_proc_start = gethrtime();
6558 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6559 
6560 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6561 		zi = &ztest_info[f];
6562 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6563 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6564 			zc->zc_next = UINT64_MAX;
6565 		else
6566 			zc->zc_next = zs->zs_proc_start +
6567 			    ztest_random(2 * zi->zi_interval[0] + 1);
6568 	}
6569 
6570 	/*
6571 	 * Run the tests in a loop.  These tests include fault injection
6572 	 * to verify that self-healing data works, and forced crashes
6573 	 * to verify that we never lose on-disk consistency.
6574 	 */
6575 	while (gethrtime() < zs->zs_proc_stop) {
6576 		int status;
6577 		boolean_t killed;
6578 
6579 		/*
6580 		 * Initialize the workload counters for each function.
6581 		 */
6582 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6583 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6584 			zc->zc_count = 0;
6585 			zc->zc_time = 0;
6586 		}
6587 
6588 		/* Set the allocation switch size */
6589 		zs->zs_metaslab_df_alloc_threshold =
6590 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6591 
6592 		if (!hasalt || ztest_random(2) == 0) {
6593 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6594 				(void) printf("Executing newer ztest: %s\n",
6595 				    cmd);
6596 			}
6597 			newer++;
6598 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6599 		} else {
6600 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6601 				(void) printf("Executing older ztest: %s\n",
6602 				    ztest_opts.zo_alt_ztest);
6603 			}
6604 			older++;
6605 			killed = exec_child(ztest_opts.zo_alt_ztest,
6606 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6607 		}
6608 
6609 		if (killed)
6610 			kills++;
6611 		iters++;
6612 
6613 		if (ztest_opts.zo_verbose >= 1) {
6614 			hrtime_t now = gethrtime();
6615 
6616 			now = MIN(now, zs->zs_proc_stop);
6617 			print_time(zs->zs_proc_stop - now, timebuf);
6618 			nicenum(zs->zs_space, numbuf, sizeof (numbuf));
6619 
6620 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6621 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6622 			    iters,
6623 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6624 			    (u_longlong_t)zs->zs_enospc_count,
6625 			    100.0 * zs->zs_alloc / zs->zs_space,
6626 			    numbuf,
6627 			    100.0 * (now - zs->zs_proc_start) /
6628 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6629 		}
6630 
6631 		if (ztest_opts.zo_verbose >= 2) {
6632 			(void) printf("\nWorkload summary:\n\n");
6633 			(void) printf("%7s %9s   %s\n",
6634 			    "Calls", "Time", "Function");
6635 			(void) printf("%7s %9s   %s\n",
6636 			    "-----", "----", "--------");
6637 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6638 				Dl_info dli;
6639 
6640 				zi = &ztest_info[f];
6641 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6642 				print_time(zc->zc_time, timebuf);
6643 				(void) dladdr((void *)zi->zi_func, &dli);
6644 				(void) printf("%7llu %9s   %s\n",
6645 				    (u_longlong_t)zc->zc_count, timebuf,
6646 				    dli.dli_sname);
6647 			}
6648 			(void) printf("\n");
6649 		}
6650 
6651 		ztest_run_zdb(ztest_opts.zo_pool);
6652 	}
6653 
6654 	if (ztest_opts.zo_verbose >= 1) {
6655 		if (hasalt) {
6656 			(void) printf("%d runs of older ztest: %s\n", older,
6657 			    ztest_opts.zo_alt_ztest);
6658 			(void) printf("%d runs of newer ztest: %s\n", newer,
6659 			    cmd);
6660 		}
6661 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6662 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6663 	}
6664 
6665 	umem_free(cmd, MAXNAMELEN);
6666 
6667 	return (0);
6668 }
6669