xref: /illumos-gate/usr/src/cmd/ztest/ztest.c (revision cd277642e0bdc71a7f57c550df1279b0c091d6e2)
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, 2016 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  */
28 
29 /*
30  * The objective of this program is to provide a DMU/ZAP/SPA stress test
31  * that runs entirely in userland, is easy to use, and easy to extend.
32  *
33  * The overall design of the ztest program is as follows:
34  *
35  * (1) For each major functional area (e.g. adding vdevs to a pool,
36  *     creating and destroying datasets, reading and writing objects, etc)
37  *     we have a simple routine to test that functionality.  These
38  *     individual routines do not have to do anything "stressful".
39  *
40  * (2) We turn these simple functionality tests into a stress test by
41  *     running them all in parallel, with as many threads as desired,
42  *     and spread across as many datasets, objects, and vdevs as desired.
43  *
44  * (3) While all this is happening, we inject faults into the pool to
45  *     verify that self-healing data really works.
46  *
47  * (4) Every time we open a dataset, we change its checksum and compression
48  *     functions.  Thus even individual objects vary from block to block
49  *     in which checksum they use and whether they're compressed.
50  *
51  * (5) To verify that we never lose on-disk consistency after a crash,
52  *     we run the entire test in a child of the main process.
53  *     At random times, the child self-immolates with a SIGKILL.
54  *     This is the software equivalent of pulling the power cord.
55  *     The parent then runs the test again, using the existing
56  *     storage pool, as many times as desired. If backwards compatibility
57  *     testing is enabled ztest will sometimes run the "older" version
58  *     of ztest after a SIGKILL.
59  *
60  * (6) To verify that we don't have future leaks or temporal incursions,
61  *     many of the functional tests record the transaction group number
62  *     as part of their data.  When reading old data, they verify that
63  *     the transaction group number is less than the current, open txg.
64  *     If you add a new test, please do this if applicable.
65  *
66  * When run with no arguments, ztest runs for about five minutes and
67  * produces no output if successful.  To get a little bit of information,
68  * specify -V.  To get more information, specify -VV, and so on.
69  *
70  * To turn this into an overnight stress test, use -T to specify run time.
71  *
72  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
73  * to increase the pool capacity, fanout, and overall stress level.
74  *
75  * Use the -k option to set the desired frequency of kills.
76  *
77  * When ztest invokes itself it passes all relevant information through a
78  * temporary file which is mmap-ed in the child process. This allows shared
79  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
80  * stored at offset 0 of this file and contains information on the size and
81  * number of shared structures in the file. The information stored in this file
82  * must remain backwards compatible with older versions of ztest so that
83  * ztest can invoke them during backwards compatibility testing (-B).
84  */
85 
86 #include <sys/zfs_context.h>
87 #include <sys/spa.h>
88 #include <sys/dmu.h>
89 #include <sys/txg.h>
90 #include <sys/dbuf.h>
91 #include <sys/zap.h>
92 #include <sys/dmu_objset.h>
93 #include <sys/poll.h>
94 #include <sys/stat.h>
95 #include <sys/time.h>
96 #include <sys/wait.h>
97 #include <sys/mman.h>
98 #include <sys/resource.h>
99 #include <sys/zio.h>
100 #include <sys/zil.h>
101 #include <sys/zil_impl.h>
102 #include <sys/vdev_impl.h>
103 #include <sys/vdev_file.h>
104 #include <sys/spa_impl.h>
105 #include <sys/metaslab_impl.h>
106 #include <sys/dsl_prop.h>
107 #include <sys/dsl_dataset.h>
108 #include <sys/dsl_destroy.h>
109 #include <sys/dsl_scan.h>
110 #include <sys/zio_checksum.h>
111 #include <sys/refcount.h>
112 #include <sys/zfeature.h>
113 #include <sys/dsl_userhold.h>
114 #include <stdio.h>
115 #include <stdio_ext.h>
116 #include <stdlib.h>
117 #include <unistd.h>
118 #include <signal.h>
119 #include <umem.h>
120 #include <dlfcn.h>
121 #include <ctype.h>
122 #include <math.h>
123 #include <sys/fs/zfs.h>
124 #include <libnvpair.h>
125 
126 static int ztest_fd_data = -1;
127 static int ztest_fd_rand = -1;
128 
129 typedef struct ztest_shared_hdr {
130 	uint64_t	zh_hdr_size;
131 	uint64_t	zh_opts_size;
132 	uint64_t	zh_size;
133 	uint64_t	zh_stats_size;
134 	uint64_t	zh_stats_count;
135 	uint64_t	zh_ds_size;
136 	uint64_t	zh_ds_count;
137 } ztest_shared_hdr_t;
138 
139 static ztest_shared_hdr_t *ztest_shared_hdr;
140 
141 typedef struct ztest_shared_opts {
142 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
143 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
144 	char zo_alt_ztest[MAXNAMELEN];
145 	char zo_alt_libpath[MAXNAMELEN];
146 	uint64_t zo_vdevs;
147 	uint64_t zo_vdevtime;
148 	size_t zo_vdev_size;
149 	int zo_ashift;
150 	int zo_mirrors;
151 	int zo_raidz;
152 	int zo_raidz_parity;
153 	int zo_datasets;
154 	int zo_threads;
155 	uint64_t zo_passtime;
156 	uint64_t zo_killrate;
157 	int zo_verbose;
158 	int zo_init;
159 	uint64_t zo_time;
160 	uint64_t zo_maxloops;
161 	uint64_t zo_metaslab_gang_bang;
162 } ztest_shared_opts_t;
163 
164 static const ztest_shared_opts_t ztest_opts_defaults = {
165 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
166 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
167 	.zo_alt_ztest = { '\0' },
168 	.zo_alt_libpath = { '\0' },
169 	.zo_vdevs = 5,
170 	.zo_ashift = SPA_MINBLOCKSHIFT,
171 	.zo_mirrors = 2,
172 	.zo_raidz = 4,
173 	.zo_raidz_parity = 1,
174 	.zo_vdev_size = SPA_MINDEVSIZE * 4,	/* 256m default size */
175 	.zo_datasets = 7,
176 	.zo_threads = 23,
177 	.zo_passtime = 60,		/* 60 seconds */
178 	.zo_killrate = 70,		/* 70% kill rate */
179 	.zo_verbose = 0,
180 	.zo_init = 1,
181 	.zo_time = 300,			/* 5 minutes */
182 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
183 	.zo_metaslab_gang_bang = 32 << 10
184 };
185 
186 extern uint64_t metaslab_gang_bang;
187 extern uint64_t metaslab_df_alloc_threshold;
188 extern uint64_t zfs_deadman_synctime_ms;
189 extern int metaslab_preload_limit;
190 extern boolean_t zfs_compressed_arc_enabled;
191 
192 static ztest_shared_opts_t *ztest_shared_opts;
193 static ztest_shared_opts_t ztest_opts;
194 
195 typedef struct ztest_shared_ds {
196 	uint64_t	zd_seq;
197 } ztest_shared_ds_t;
198 
199 static ztest_shared_ds_t *ztest_shared_ds;
200 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
201 
202 #define	BT_MAGIC	0x123456789abcdefULL
203 #define	MAXFAULTS() \
204 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
205 
206 enum ztest_io_type {
207 	ZTEST_IO_WRITE_TAG,
208 	ZTEST_IO_WRITE_PATTERN,
209 	ZTEST_IO_WRITE_ZEROES,
210 	ZTEST_IO_TRUNCATE,
211 	ZTEST_IO_SETATTR,
212 	ZTEST_IO_REWRITE,
213 	ZTEST_IO_TYPES
214 };
215 
216 typedef struct ztest_block_tag {
217 	uint64_t	bt_magic;
218 	uint64_t	bt_objset;
219 	uint64_t	bt_object;
220 	uint64_t	bt_offset;
221 	uint64_t	bt_gen;
222 	uint64_t	bt_txg;
223 	uint64_t	bt_crtxg;
224 } ztest_block_tag_t;
225 
226 typedef struct bufwad {
227 	uint64_t	bw_index;
228 	uint64_t	bw_txg;
229 	uint64_t	bw_data;
230 } bufwad_t;
231 
232 /*
233  * XXX -- fix zfs range locks to be generic so we can use them here.
234  */
235 typedef enum {
236 	RL_READER,
237 	RL_WRITER,
238 	RL_APPEND
239 } rl_type_t;
240 
241 typedef struct rll {
242 	void		*rll_writer;
243 	int		rll_readers;
244 	mutex_t		rll_lock;
245 	cond_t		rll_cv;
246 } rll_t;
247 
248 typedef struct rl {
249 	uint64_t	rl_object;
250 	uint64_t	rl_offset;
251 	uint64_t	rl_size;
252 	rll_t		*rl_lock;
253 } rl_t;
254 
255 #define	ZTEST_RANGE_LOCKS	64
256 #define	ZTEST_OBJECT_LOCKS	64
257 
258 /*
259  * Object descriptor.  Used as a template for object lookup/create/remove.
260  */
261 typedef struct ztest_od {
262 	uint64_t	od_dir;
263 	uint64_t	od_object;
264 	dmu_object_type_t od_type;
265 	dmu_object_type_t od_crtype;
266 	uint64_t	od_blocksize;
267 	uint64_t	od_crblocksize;
268 	uint64_t	od_gen;
269 	uint64_t	od_crgen;
270 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
271 } ztest_od_t;
272 
273 /*
274  * Per-dataset state.
275  */
276 typedef struct ztest_ds {
277 	ztest_shared_ds_t *zd_shared;
278 	objset_t	*zd_os;
279 	rwlock_t	zd_zilog_lock;
280 	zilog_t		*zd_zilog;
281 	ztest_od_t	*zd_od;		/* debugging aid */
282 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
283 	mutex_t		zd_dirobj_lock;
284 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
285 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
286 } ztest_ds_t;
287 
288 /*
289  * Per-iteration state.
290  */
291 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
292 
293 typedef struct ztest_info {
294 	ztest_func_t	*zi_func;	/* test function */
295 	uint64_t	zi_iters;	/* iterations per execution */
296 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
297 } ztest_info_t;
298 
299 typedef struct ztest_shared_callstate {
300 	uint64_t	zc_count;	/* per-pass count */
301 	uint64_t	zc_time;	/* per-pass time */
302 	uint64_t	zc_next;	/* next time to call this function */
303 } ztest_shared_callstate_t;
304 
305 static ztest_shared_callstate_t *ztest_shared_callstate;
306 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
307 
308 /*
309  * Note: these aren't static because we want dladdr() to work.
310  */
311 ztest_func_t ztest_dmu_read_write;
312 ztest_func_t ztest_dmu_write_parallel;
313 ztest_func_t ztest_dmu_object_alloc_free;
314 ztest_func_t ztest_dmu_commit_callbacks;
315 ztest_func_t ztest_zap;
316 ztest_func_t ztest_zap_parallel;
317 ztest_func_t ztest_zil_commit;
318 ztest_func_t ztest_zil_remount;
319 ztest_func_t ztest_dmu_read_write_zcopy;
320 ztest_func_t ztest_dmu_objset_create_destroy;
321 ztest_func_t ztest_dmu_prealloc;
322 ztest_func_t ztest_fzap;
323 ztest_func_t ztest_dmu_snapshot_create_destroy;
324 ztest_func_t ztest_dsl_prop_get_set;
325 ztest_func_t ztest_spa_prop_get_set;
326 ztest_func_t ztest_spa_create_destroy;
327 ztest_func_t ztest_fault_inject;
328 ztest_func_t ztest_ddt_repair;
329 ztest_func_t ztest_dmu_snapshot_hold;
330 ztest_func_t ztest_spa_rename;
331 ztest_func_t ztest_scrub;
332 ztest_func_t ztest_dsl_dataset_promote_busy;
333 ztest_func_t ztest_vdev_attach_detach;
334 ztest_func_t ztest_vdev_LUN_growth;
335 ztest_func_t ztest_vdev_add_remove;
336 ztest_func_t ztest_vdev_aux_add_remove;
337 ztest_func_t ztest_split_pool;
338 ztest_func_t ztest_reguid;
339 ztest_func_t ztest_spa_upgrade;
340 
341 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
342 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
343 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
344 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
345 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
346 
347 ztest_info_t ztest_info[] = {
348 	{ ztest_dmu_read_write,			1,	&zopt_always	},
349 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
350 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
351 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
352 	{ ztest_zap,				30,	&zopt_always	},
353 	{ ztest_zap_parallel,			100,	&zopt_always	},
354 	{ ztest_split_pool,			1,	&zopt_always	},
355 	{ ztest_zil_commit,			1,	&zopt_incessant	},
356 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
357 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
358 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
359 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
360 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
361 #if 0
362 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
363 #endif
364 	{ ztest_fzap,				1,	&zopt_sometimes	},
365 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
366 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
367 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
368 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
369 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
370 	{ ztest_reguid,				1,	&zopt_rarely	},
371 	{ ztest_spa_rename,			1,	&zopt_rarely	},
372 	{ ztest_scrub,				1,	&zopt_rarely	},
373 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
374 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
375 	{ ztest_vdev_attach_detach,		1,	&zopt_sometimes	},
376 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
377 	{ ztest_vdev_add_remove,		1,
378 	    &ztest_opts.zo_vdevtime				},
379 	{ ztest_vdev_aux_add_remove,		1,
380 	    &ztest_opts.zo_vdevtime				},
381 };
382 
383 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
384 
385 /*
386  * The following struct is used to hold a list of uncalled commit callbacks.
387  * The callbacks are ordered by txg number.
388  */
389 typedef struct ztest_cb_list {
390 	mutex_t	zcl_callbacks_lock;
391 	list_t	zcl_callbacks;
392 } ztest_cb_list_t;
393 
394 /*
395  * Stuff we need to share writably between parent and child.
396  */
397 typedef struct ztest_shared {
398 	boolean_t	zs_do_init;
399 	hrtime_t	zs_proc_start;
400 	hrtime_t	zs_proc_stop;
401 	hrtime_t	zs_thread_start;
402 	hrtime_t	zs_thread_stop;
403 	hrtime_t	zs_thread_kill;
404 	uint64_t	zs_enospc_count;
405 	uint64_t	zs_vdev_next_leaf;
406 	uint64_t	zs_vdev_aux;
407 	uint64_t	zs_alloc;
408 	uint64_t	zs_space;
409 	uint64_t	zs_splits;
410 	uint64_t	zs_mirrors;
411 	uint64_t	zs_metaslab_sz;
412 	uint64_t	zs_metaslab_df_alloc_threshold;
413 	uint64_t	zs_guid;
414 } ztest_shared_t;
415 
416 #define	ID_PARALLEL	-1ULL
417 
418 static char ztest_dev_template[] = "%s/%s.%llua";
419 static char ztest_aux_template[] = "%s/%s.%s.%llu";
420 ztest_shared_t *ztest_shared;
421 
422 static spa_t *ztest_spa = NULL;
423 static ztest_ds_t *ztest_ds;
424 
425 static mutex_t ztest_vdev_lock;
426 
427 /*
428  * The ztest_name_lock protects the pool and dataset namespace used by
429  * the individual tests. To modify the namespace, consumers must grab
430  * this lock as writer. Grabbing the lock as reader will ensure that the
431  * namespace does not change while the lock is held.
432  */
433 static rwlock_t ztest_name_lock;
434 
435 static boolean_t ztest_dump_core = B_TRUE;
436 static boolean_t ztest_exiting;
437 
438 /* Global commit callback list */
439 static ztest_cb_list_t zcl;
440 
441 enum ztest_object {
442 	ZTEST_META_DNODE = 0,
443 	ZTEST_DIROBJ,
444 	ZTEST_OBJECTS
445 };
446 
447 static void usage(boolean_t) __NORETURN;
448 
449 /*
450  * These libumem hooks provide a reasonable set of defaults for the allocator's
451  * debugging facilities.
452  */
453 const char *
454 _umem_debug_init()
455 {
456 	return ("default,verbose"); /* $UMEM_DEBUG setting */
457 }
458 
459 const char *
460 _umem_logging_init(void)
461 {
462 	return ("fail,contents"); /* $UMEM_LOGGING setting */
463 }
464 
465 #define	FATAL_MSG_SZ	1024
466 
467 char *fatal_msg;
468 
469 static void
470 fatal(int do_perror, char *message, ...)
471 {
472 	va_list args;
473 	int save_errno = errno;
474 	char buf[FATAL_MSG_SZ];
475 
476 	(void) fflush(stdout);
477 
478 	va_start(args, message);
479 	(void) sprintf(buf, "ztest: ");
480 	/* LINTED */
481 	(void) vsprintf(buf + strlen(buf), message, args);
482 	va_end(args);
483 	if (do_perror) {
484 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
485 		    ": %s", strerror(save_errno));
486 	}
487 	(void) fprintf(stderr, "%s\n", buf);
488 	fatal_msg = buf;			/* to ease debugging */
489 	if (ztest_dump_core)
490 		abort();
491 	exit(3);
492 }
493 
494 static int
495 str2shift(const char *buf)
496 {
497 	const char *ends = "BKMGTPEZ";
498 	int i;
499 
500 	if (buf[0] == '\0')
501 		return (0);
502 	for (i = 0; i < strlen(ends); i++) {
503 		if (toupper(buf[0]) == ends[i])
504 			break;
505 	}
506 	if (i == strlen(ends)) {
507 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
508 		    buf);
509 		usage(B_FALSE);
510 	}
511 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
512 		return (10*i);
513 	}
514 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
515 	usage(B_FALSE);
516 	/* NOTREACHED */
517 }
518 
519 static uint64_t
520 nicenumtoull(const char *buf)
521 {
522 	char *end;
523 	uint64_t val;
524 
525 	val = strtoull(buf, &end, 0);
526 	if (end == buf) {
527 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
528 		usage(B_FALSE);
529 	} else if (end[0] == '.') {
530 		double fval = strtod(buf, &end);
531 		fval *= pow(2, str2shift(end));
532 		if (fval > UINT64_MAX) {
533 			(void) fprintf(stderr, "ztest: value too large: %s\n",
534 			    buf);
535 			usage(B_FALSE);
536 		}
537 		val = (uint64_t)fval;
538 	} else {
539 		int shift = str2shift(end);
540 		if (shift >= 64 || (val << shift) >> shift != val) {
541 			(void) fprintf(stderr, "ztest: value too large: %s\n",
542 			    buf);
543 			usage(B_FALSE);
544 		}
545 		val <<= shift;
546 	}
547 	return (val);
548 }
549 
550 static void
551 usage(boolean_t requested)
552 {
553 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
554 
555 	char nice_vdev_size[10];
556 	char nice_gang_bang[10];
557 	FILE *fp = requested ? stdout : stderr;
558 
559 	nicenum(zo->zo_vdev_size, nice_vdev_size);
560 	nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
561 
562 	(void) fprintf(fp, "Usage: %s\n"
563 	    "\t[-v vdevs (default: %llu)]\n"
564 	    "\t[-s size_of_each_vdev (default: %s)]\n"
565 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
566 	    "\t[-m mirror_copies (default: %d)]\n"
567 	    "\t[-r raidz_disks (default: %d)]\n"
568 	    "\t[-R raidz_parity (default: %d)]\n"
569 	    "\t[-d datasets (default: %d)]\n"
570 	    "\t[-t threads (default: %d)]\n"
571 	    "\t[-g gang_block_threshold (default: %s)]\n"
572 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
573 	    "\t[-k kill_percentage (default: %llu%%)]\n"
574 	    "\t[-p pool_name (default: %s)]\n"
575 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
576 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
577 	    "\t[-E] use existing pool instead of creating new one\n"
578 	    "\t[-T time (default: %llu sec)] total run time\n"
579 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
580 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
581 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
582 	    "\t[-o variable=value] ... set global variable to an unsigned\n"
583 	    "\t    32-bit integer value\n"
584 	    "\t[-h] (print help)\n"
585 	    "",
586 	    zo->zo_pool,
587 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
588 	    nice_vdev_size,				/* -s */
589 	    zo->zo_ashift,				/* -a */
590 	    zo->zo_mirrors,				/* -m */
591 	    zo->zo_raidz,				/* -r */
592 	    zo->zo_raidz_parity,			/* -R */
593 	    zo->zo_datasets,				/* -d */
594 	    zo->zo_threads,				/* -t */
595 	    nice_gang_bang,				/* -g */
596 	    zo->zo_init,				/* -i */
597 	    (u_longlong_t)zo->zo_killrate,		/* -k */
598 	    zo->zo_pool,				/* -p */
599 	    zo->zo_dir,					/* -f */
600 	    (u_longlong_t)zo->zo_time,			/* -T */
601 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
602 	    (u_longlong_t)zo->zo_passtime);
603 	exit(requested ? 0 : 1);
604 }
605 
606 static void
607 process_options(int argc, char **argv)
608 {
609 	char *path;
610 	ztest_shared_opts_t *zo = &ztest_opts;
611 
612 	int opt;
613 	uint64_t value;
614 	char altdir[MAXNAMELEN] = { 0 };
615 
616 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
617 
618 	while ((opt = getopt(argc, argv,
619 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) {
620 		value = 0;
621 		switch (opt) {
622 		case 'v':
623 		case 's':
624 		case 'a':
625 		case 'm':
626 		case 'r':
627 		case 'R':
628 		case 'd':
629 		case 't':
630 		case 'g':
631 		case 'i':
632 		case 'k':
633 		case 'T':
634 		case 'P':
635 		case 'F':
636 			value = nicenumtoull(optarg);
637 		}
638 		switch (opt) {
639 		case 'v':
640 			zo->zo_vdevs = value;
641 			break;
642 		case 's':
643 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
644 			break;
645 		case 'a':
646 			zo->zo_ashift = value;
647 			break;
648 		case 'm':
649 			zo->zo_mirrors = value;
650 			break;
651 		case 'r':
652 			zo->zo_raidz = MAX(1, value);
653 			break;
654 		case 'R':
655 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
656 			break;
657 		case 'd':
658 			zo->zo_datasets = MAX(1, value);
659 			break;
660 		case 't':
661 			zo->zo_threads = MAX(1, value);
662 			break;
663 		case 'g':
664 			zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
665 			    value);
666 			break;
667 		case 'i':
668 			zo->zo_init = value;
669 			break;
670 		case 'k':
671 			zo->zo_killrate = value;
672 			break;
673 		case 'p':
674 			(void) strlcpy(zo->zo_pool, optarg,
675 			    sizeof (zo->zo_pool));
676 			break;
677 		case 'f':
678 			path = realpath(optarg, NULL);
679 			if (path == NULL) {
680 				(void) fprintf(stderr, "error: %s: %s\n",
681 				    optarg, strerror(errno));
682 				usage(B_FALSE);
683 			} else {
684 				(void) strlcpy(zo->zo_dir, path,
685 				    sizeof (zo->zo_dir));
686 			}
687 			break;
688 		case 'V':
689 			zo->zo_verbose++;
690 			break;
691 		case 'E':
692 			zo->zo_init = 0;
693 			break;
694 		case 'T':
695 			zo->zo_time = value;
696 			break;
697 		case 'P':
698 			zo->zo_passtime = MAX(1, value);
699 			break;
700 		case 'F':
701 			zo->zo_maxloops = MAX(1, value);
702 			break;
703 		case 'B':
704 			(void) strlcpy(altdir, optarg, sizeof (altdir));
705 			break;
706 		case 'o':
707 			if (set_global_var(optarg) != 0)
708 				usage(B_FALSE);
709 			break;
710 		case 'h':
711 			usage(B_TRUE);
712 			break;
713 		case '?':
714 		default:
715 			usage(B_FALSE);
716 			break;
717 		}
718 	}
719 
720 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
721 
722 	zo->zo_vdevtime =
723 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
724 	    UINT64_MAX >> 2);
725 
726 	if (strlen(altdir) > 0) {
727 		char *cmd;
728 		char *realaltdir;
729 		char *bin;
730 		char *ztest;
731 		char *isa;
732 		int isalen;
733 
734 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
735 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
736 
737 		VERIFY(NULL != realpath(getexecname(), cmd));
738 		if (0 != access(altdir, F_OK)) {
739 			ztest_dump_core = B_FALSE;
740 			fatal(B_TRUE, "invalid alternate ztest path: %s",
741 			    altdir);
742 		}
743 		VERIFY(NULL != realpath(altdir, realaltdir));
744 
745 		/*
746 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
747 		 * We want to extract <isa> to determine if we should use
748 		 * 32 or 64 bit binaries.
749 		 */
750 		bin = strstr(cmd, "/usr/bin/");
751 		ztest = strstr(bin, "/ztest");
752 		isa = bin + 9;
753 		isalen = ztest - isa;
754 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
755 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
756 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
757 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
758 
759 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
760 			ztest_dump_core = B_FALSE;
761 			fatal(B_TRUE, "invalid alternate ztest: %s",
762 			    zo->zo_alt_ztest);
763 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
764 			ztest_dump_core = B_FALSE;
765 			fatal(B_TRUE, "invalid alternate lib directory %s",
766 			    zo->zo_alt_libpath);
767 		}
768 
769 		umem_free(cmd, MAXPATHLEN);
770 		umem_free(realaltdir, MAXPATHLEN);
771 	}
772 }
773 
774 static void
775 ztest_kill(ztest_shared_t *zs)
776 {
777 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
778 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
779 
780 	/*
781 	 * Before we kill off ztest, make sure that the config is updated.
782 	 * See comment above spa_config_sync().
783 	 */
784 	mutex_enter(&spa_namespace_lock);
785 	spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
786 	mutex_exit(&spa_namespace_lock);
787 
788 	zfs_dbgmsg_print(FTAG);
789 	(void) kill(getpid(), SIGKILL);
790 }
791 
792 static uint64_t
793 ztest_random(uint64_t range)
794 {
795 	uint64_t r;
796 
797 	ASSERT3S(ztest_fd_rand, >=, 0);
798 
799 	if (range == 0)
800 		return (0);
801 
802 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
803 		fatal(1, "short read from /dev/urandom");
804 
805 	return (r % range);
806 }
807 
808 /* ARGSUSED */
809 static void
810 ztest_record_enospc(const char *s)
811 {
812 	ztest_shared->zs_enospc_count++;
813 }
814 
815 static uint64_t
816 ztest_get_ashift(void)
817 {
818 	if (ztest_opts.zo_ashift == 0)
819 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
820 	return (ztest_opts.zo_ashift);
821 }
822 
823 static nvlist_t *
824 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
825 {
826 	char pathbuf[MAXPATHLEN];
827 	uint64_t vdev;
828 	nvlist_t *file;
829 
830 	if (ashift == 0)
831 		ashift = ztest_get_ashift();
832 
833 	if (path == NULL) {
834 		path = pathbuf;
835 
836 		if (aux != NULL) {
837 			vdev = ztest_shared->zs_vdev_aux;
838 			(void) snprintf(path, sizeof (pathbuf),
839 			    ztest_aux_template, ztest_opts.zo_dir,
840 			    pool == NULL ? ztest_opts.zo_pool : pool,
841 			    aux, vdev);
842 		} else {
843 			vdev = ztest_shared->zs_vdev_next_leaf++;
844 			(void) snprintf(path, sizeof (pathbuf),
845 			    ztest_dev_template, ztest_opts.zo_dir,
846 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
847 		}
848 	}
849 
850 	if (size != 0) {
851 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
852 		if (fd == -1)
853 			fatal(1, "can't open %s", path);
854 		if (ftruncate(fd, size) != 0)
855 			fatal(1, "can't ftruncate %s", path);
856 		(void) close(fd);
857 	}
858 
859 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
860 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
861 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
862 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
863 
864 	return (file);
865 }
866 
867 static nvlist_t *
868 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
869     uint64_t ashift, int r)
870 {
871 	nvlist_t *raidz, **child;
872 	int c;
873 
874 	if (r < 2)
875 		return (make_vdev_file(path, aux, pool, size, ashift));
876 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
877 
878 	for (c = 0; c < r; c++)
879 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
880 
881 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
882 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
883 	    VDEV_TYPE_RAIDZ) == 0);
884 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
885 	    ztest_opts.zo_raidz_parity) == 0);
886 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
887 	    child, r) == 0);
888 
889 	for (c = 0; c < r; c++)
890 		nvlist_free(child[c]);
891 
892 	umem_free(child, r * sizeof (nvlist_t *));
893 
894 	return (raidz);
895 }
896 
897 static nvlist_t *
898 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
899     uint64_t ashift, int r, int m)
900 {
901 	nvlist_t *mirror, **child;
902 	int c;
903 
904 	if (m < 1)
905 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
906 
907 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
908 
909 	for (c = 0; c < m; c++)
910 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
911 
912 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
913 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
914 	    VDEV_TYPE_MIRROR) == 0);
915 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
916 	    child, m) == 0);
917 
918 	for (c = 0; c < m; c++)
919 		nvlist_free(child[c]);
920 
921 	umem_free(child, m * sizeof (nvlist_t *));
922 
923 	return (mirror);
924 }
925 
926 static nvlist_t *
927 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
928     int log, int r, int m, int t)
929 {
930 	nvlist_t *root, **child;
931 	int c;
932 
933 	ASSERT(t > 0);
934 
935 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
936 
937 	for (c = 0; c < t; c++) {
938 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
939 		    r, m);
940 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
941 		    log) == 0);
942 	}
943 
944 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
945 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
946 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
947 	    child, t) == 0);
948 
949 	for (c = 0; c < t; c++)
950 		nvlist_free(child[c]);
951 
952 	umem_free(child, t * sizeof (nvlist_t *));
953 
954 	return (root);
955 }
956 
957 /*
958  * Find a random spa version. Returns back a random spa version in the
959  * range [initial_version, SPA_VERSION_FEATURES].
960  */
961 static uint64_t
962 ztest_random_spa_version(uint64_t initial_version)
963 {
964 	uint64_t version = initial_version;
965 
966 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
967 		version = version +
968 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
969 	}
970 
971 	if (version > SPA_VERSION_BEFORE_FEATURES)
972 		version = SPA_VERSION_FEATURES;
973 
974 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
975 	return (version);
976 }
977 
978 static int
979 ztest_random_blocksize(void)
980 {
981 	uint64_t block_shift;
982 	/*
983 	 * Choose a block size >= the ashift.
984 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
985 	 */
986 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
987 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
988 		maxbs = 20;
989 	block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
990 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
991 }
992 
993 static int
994 ztest_random_ibshift(void)
995 {
996 	return (DN_MIN_INDBLKSHIFT +
997 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
998 }
999 
1000 static uint64_t
1001 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1002 {
1003 	uint64_t top;
1004 	vdev_t *rvd = spa->spa_root_vdev;
1005 	vdev_t *tvd;
1006 
1007 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1008 
1009 	do {
1010 		top = ztest_random(rvd->vdev_children);
1011 		tvd = rvd->vdev_child[top];
1012 	} while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1013 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1014 
1015 	return (top);
1016 }
1017 
1018 static uint64_t
1019 ztest_random_dsl_prop(zfs_prop_t prop)
1020 {
1021 	uint64_t value;
1022 
1023 	do {
1024 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1025 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1026 
1027 	return (value);
1028 }
1029 
1030 static int
1031 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1032     boolean_t inherit)
1033 {
1034 	const char *propname = zfs_prop_to_name(prop);
1035 	const char *valname;
1036 	char setpoint[MAXPATHLEN];
1037 	uint64_t curval;
1038 	int error;
1039 
1040 	error = dsl_prop_set_int(osname, propname,
1041 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1042 
1043 	if (error == ENOSPC) {
1044 		ztest_record_enospc(FTAG);
1045 		return (error);
1046 	}
1047 	ASSERT0(error);
1048 
1049 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1050 
1051 	if (ztest_opts.zo_verbose >= 6) {
1052 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1053 		(void) printf("%s %s = %s at '%s'\n",
1054 		    osname, propname, valname, setpoint);
1055 	}
1056 
1057 	return (error);
1058 }
1059 
1060 static int
1061 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1062 {
1063 	spa_t *spa = ztest_spa;
1064 	nvlist_t *props = NULL;
1065 	int error;
1066 
1067 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1068 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1069 
1070 	error = spa_prop_set(spa, props);
1071 
1072 	nvlist_free(props);
1073 
1074 	if (error == ENOSPC) {
1075 		ztest_record_enospc(FTAG);
1076 		return (error);
1077 	}
1078 	ASSERT0(error);
1079 
1080 	return (error);
1081 }
1082 
1083 static void
1084 ztest_rll_init(rll_t *rll)
1085 {
1086 	rll->rll_writer = NULL;
1087 	rll->rll_readers = 0;
1088 	VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1089 	VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1090 }
1091 
1092 static void
1093 ztest_rll_destroy(rll_t *rll)
1094 {
1095 	ASSERT(rll->rll_writer == NULL);
1096 	ASSERT(rll->rll_readers == 0);
1097 	VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1098 	VERIFY(cond_destroy(&rll->rll_cv) == 0);
1099 }
1100 
1101 static void
1102 ztest_rll_lock(rll_t *rll, rl_type_t type)
1103 {
1104 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1105 
1106 	if (type == RL_READER) {
1107 		while (rll->rll_writer != NULL)
1108 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1109 		rll->rll_readers++;
1110 	} else {
1111 		while (rll->rll_writer != NULL || rll->rll_readers)
1112 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1113 		rll->rll_writer = curthread;
1114 	}
1115 
1116 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1117 }
1118 
1119 static void
1120 ztest_rll_unlock(rll_t *rll)
1121 {
1122 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1123 
1124 	if (rll->rll_writer) {
1125 		ASSERT(rll->rll_readers == 0);
1126 		rll->rll_writer = NULL;
1127 	} else {
1128 		ASSERT(rll->rll_readers != 0);
1129 		ASSERT(rll->rll_writer == NULL);
1130 		rll->rll_readers--;
1131 	}
1132 
1133 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1134 		VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1135 
1136 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1137 }
1138 
1139 static void
1140 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1141 {
1142 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1143 
1144 	ztest_rll_lock(rll, type);
1145 }
1146 
1147 static void
1148 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1149 {
1150 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1151 
1152 	ztest_rll_unlock(rll);
1153 }
1154 
1155 static rl_t *
1156 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1157     uint64_t size, rl_type_t type)
1158 {
1159 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1160 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1161 	rl_t *rl;
1162 
1163 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1164 	rl->rl_object = object;
1165 	rl->rl_offset = offset;
1166 	rl->rl_size = size;
1167 	rl->rl_lock = rll;
1168 
1169 	ztest_rll_lock(rll, type);
1170 
1171 	return (rl);
1172 }
1173 
1174 static void
1175 ztest_range_unlock(rl_t *rl)
1176 {
1177 	rll_t *rll = rl->rl_lock;
1178 
1179 	ztest_rll_unlock(rll);
1180 
1181 	umem_free(rl, sizeof (*rl));
1182 }
1183 
1184 static void
1185 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1186 {
1187 	zd->zd_os = os;
1188 	zd->zd_zilog = dmu_objset_zil(os);
1189 	zd->zd_shared = szd;
1190 	dmu_objset_name(os, zd->zd_name);
1191 
1192 	if (zd->zd_shared != NULL)
1193 		zd->zd_shared->zd_seq = 0;
1194 
1195 	VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1196 	VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1197 
1198 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1199 		ztest_rll_init(&zd->zd_object_lock[l]);
1200 
1201 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1202 		ztest_rll_init(&zd->zd_range_lock[l]);
1203 }
1204 
1205 static void
1206 ztest_zd_fini(ztest_ds_t *zd)
1207 {
1208 	VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1209 
1210 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1211 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1212 
1213 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1214 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1215 }
1216 
1217 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1218 
1219 static uint64_t
1220 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1221 {
1222 	uint64_t txg;
1223 	int error;
1224 
1225 	/*
1226 	 * Attempt to assign tx to some transaction group.
1227 	 */
1228 	error = dmu_tx_assign(tx, txg_how);
1229 	if (error) {
1230 		if (error == ERESTART) {
1231 			ASSERT(txg_how == TXG_NOWAIT);
1232 			dmu_tx_wait(tx);
1233 		} else {
1234 			ASSERT3U(error, ==, ENOSPC);
1235 			ztest_record_enospc(tag);
1236 		}
1237 		dmu_tx_abort(tx);
1238 		return (0);
1239 	}
1240 	txg = dmu_tx_get_txg(tx);
1241 	ASSERT(txg != 0);
1242 	return (txg);
1243 }
1244 
1245 static void
1246 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1247 {
1248 	uint64_t *ip = buf;
1249 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1250 
1251 	while (ip < ip_end)
1252 		*ip++ = value;
1253 }
1254 
1255 static boolean_t
1256 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1257 {
1258 	uint64_t *ip = buf;
1259 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1260 	uint64_t diff = 0;
1261 
1262 	while (ip < ip_end)
1263 		diff |= (value - *ip++);
1264 
1265 	return (diff == 0);
1266 }
1267 
1268 static void
1269 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1270     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1271 {
1272 	bt->bt_magic = BT_MAGIC;
1273 	bt->bt_objset = dmu_objset_id(os);
1274 	bt->bt_object = object;
1275 	bt->bt_offset = offset;
1276 	bt->bt_gen = gen;
1277 	bt->bt_txg = txg;
1278 	bt->bt_crtxg = crtxg;
1279 }
1280 
1281 static void
1282 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1283     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1284 {
1285 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1286 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1287 	ASSERT3U(bt->bt_object, ==, object);
1288 	ASSERT3U(bt->bt_offset, ==, offset);
1289 	ASSERT3U(bt->bt_gen, <=, gen);
1290 	ASSERT3U(bt->bt_txg, <=, txg);
1291 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1292 }
1293 
1294 static ztest_block_tag_t *
1295 ztest_bt_bonus(dmu_buf_t *db)
1296 {
1297 	dmu_object_info_t doi;
1298 	ztest_block_tag_t *bt;
1299 
1300 	dmu_object_info_from_db(db, &doi);
1301 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1302 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1303 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1304 
1305 	return (bt);
1306 }
1307 
1308 /*
1309  * ZIL logging ops
1310  */
1311 
1312 #define	lrz_type	lr_mode
1313 #define	lrz_blocksize	lr_uid
1314 #define	lrz_ibshift	lr_gid
1315 #define	lrz_bonustype	lr_rdev
1316 #define	lrz_bonuslen	lr_crtime[1]
1317 
1318 static void
1319 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1320 {
1321 	char *name = (void *)(lr + 1);		/* name follows lr */
1322 	size_t namesize = strlen(name) + 1;
1323 	itx_t *itx;
1324 
1325 	if (zil_replaying(zd->zd_zilog, tx))
1326 		return;
1327 
1328 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1329 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1330 	    sizeof (*lr) + namesize - sizeof (lr_t));
1331 
1332 	zil_itx_assign(zd->zd_zilog, itx, tx);
1333 }
1334 
1335 static void
1336 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1337 {
1338 	char *name = (void *)(lr + 1);		/* name follows lr */
1339 	size_t namesize = strlen(name) + 1;
1340 	itx_t *itx;
1341 
1342 	if (zil_replaying(zd->zd_zilog, tx))
1343 		return;
1344 
1345 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1346 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1347 	    sizeof (*lr) + namesize - sizeof (lr_t));
1348 
1349 	itx->itx_oid = object;
1350 	zil_itx_assign(zd->zd_zilog, itx, tx);
1351 }
1352 
1353 static void
1354 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1355 {
1356 	itx_t *itx;
1357 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1358 
1359 	if (zil_replaying(zd->zd_zilog, tx))
1360 		return;
1361 
1362 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1363 		write_state = WR_INDIRECT;
1364 
1365 	itx = zil_itx_create(TX_WRITE,
1366 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1367 
1368 	if (write_state == WR_COPIED &&
1369 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1370 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1371 		zil_itx_destroy(itx);
1372 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1373 		write_state = WR_NEED_COPY;
1374 	}
1375 	itx->itx_private = zd;
1376 	itx->itx_wr_state = write_state;
1377 	itx->itx_sync = (ztest_random(8) == 0);
1378 	itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1379 
1380 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1381 	    sizeof (*lr) - sizeof (lr_t));
1382 
1383 	zil_itx_assign(zd->zd_zilog, itx, tx);
1384 }
1385 
1386 static void
1387 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1388 {
1389 	itx_t *itx;
1390 
1391 	if (zil_replaying(zd->zd_zilog, tx))
1392 		return;
1393 
1394 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1395 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1396 	    sizeof (*lr) - sizeof (lr_t));
1397 
1398 	itx->itx_sync = B_FALSE;
1399 	zil_itx_assign(zd->zd_zilog, itx, tx);
1400 }
1401 
1402 static void
1403 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1404 {
1405 	itx_t *itx;
1406 
1407 	if (zil_replaying(zd->zd_zilog, tx))
1408 		return;
1409 
1410 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1411 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1412 	    sizeof (*lr) - sizeof (lr_t));
1413 
1414 	itx->itx_sync = B_FALSE;
1415 	zil_itx_assign(zd->zd_zilog, itx, tx);
1416 }
1417 
1418 /*
1419  * ZIL replay ops
1420  */
1421 static int
1422 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1423 {
1424 	char *name = (void *)(lr + 1);		/* name follows lr */
1425 	objset_t *os = zd->zd_os;
1426 	ztest_block_tag_t *bbt;
1427 	dmu_buf_t *db;
1428 	dmu_tx_t *tx;
1429 	uint64_t txg;
1430 	int error = 0;
1431 
1432 	if (byteswap)
1433 		byteswap_uint64_array(lr, sizeof (*lr));
1434 
1435 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1436 	ASSERT(name[0] != '\0');
1437 
1438 	tx = dmu_tx_create(os);
1439 
1440 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1441 
1442 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1443 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1444 	} else {
1445 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1446 	}
1447 
1448 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1449 	if (txg == 0)
1450 		return (ENOSPC);
1451 
1452 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1453 
1454 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1455 		if (lr->lr_foid == 0) {
1456 			lr->lr_foid = zap_create(os,
1457 			    lr->lrz_type, lr->lrz_bonustype,
1458 			    lr->lrz_bonuslen, tx);
1459 		} else {
1460 			error = zap_create_claim(os, lr->lr_foid,
1461 			    lr->lrz_type, lr->lrz_bonustype,
1462 			    lr->lrz_bonuslen, tx);
1463 		}
1464 	} else {
1465 		if (lr->lr_foid == 0) {
1466 			lr->lr_foid = dmu_object_alloc(os,
1467 			    lr->lrz_type, 0, lr->lrz_bonustype,
1468 			    lr->lrz_bonuslen, tx);
1469 		} else {
1470 			error = dmu_object_claim(os, lr->lr_foid,
1471 			    lr->lrz_type, 0, lr->lrz_bonustype,
1472 			    lr->lrz_bonuslen, tx);
1473 		}
1474 	}
1475 
1476 	if (error) {
1477 		ASSERT3U(error, ==, EEXIST);
1478 		ASSERT(zd->zd_zilog->zl_replay);
1479 		dmu_tx_commit(tx);
1480 		return (error);
1481 	}
1482 
1483 	ASSERT(lr->lr_foid != 0);
1484 
1485 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1486 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1487 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1488 
1489 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1490 	bbt = ztest_bt_bonus(db);
1491 	dmu_buf_will_dirty(db, tx);
1492 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1493 	dmu_buf_rele(db, FTAG);
1494 
1495 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1496 	    &lr->lr_foid, tx));
1497 
1498 	(void) ztest_log_create(zd, tx, lr);
1499 
1500 	dmu_tx_commit(tx);
1501 
1502 	return (0);
1503 }
1504 
1505 static int
1506 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1507 {
1508 	char *name = (void *)(lr + 1);		/* name follows lr */
1509 	objset_t *os = zd->zd_os;
1510 	dmu_object_info_t doi;
1511 	dmu_tx_t *tx;
1512 	uint64_t object, txg;
1513 
1514 	if (byteswap)
1515 		byteswap_uint64_array(lr, sizeof (*lr));
1516 
1517 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1518 	ASSERT(name[0] != '\0');
1519 
1520 	VERIFY3U(0, ==,
1521 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1522 	ASSERT(object != 0);
1523 
1524 	ztest_object_lock(zd, object, RL_WRITER);
1525 
1526 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1527 
1528 	tx = dmu_tx_create(os);
1529 
1530 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1531 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1532 
1533 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1534 	if (txg == 0) {
1535 		ztest_object_unlock(zd, object);
1536 		return (ENOSPC);
1537 	}
1538 
1539 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1540 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1541 	} else {
1542 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1543 	}
1544 
1545 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1546 
1547 	(void) ztest_log_remove(zd, tx, lr, object);
1548 
1549 	dmu_tx_commit(tx);
1550 
1551 	ztest_object_unlock(zd, object);
1552 
1553 	return (0);
1554 }
1555 
1556 static int
1557 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1558 {
1559 	objset_t *os = zd->zd_os;
1560 	void *data = lr + 1;			/* data follows lr */
1561 	uint64_t offset, length;
1562 	ztest_block_tag_t *bt = data;
1563 	ztest_block_tag_t *bbt;
1564 	uint64_t gen, txg, lrtxg, crtxg;
1565 	dmu_object_info_t doi;
1566 	dmu_tx_t *tx;
1567 	dmu_buf_t *db;
1568 	arc_buf_t *abuf = NULL;
1569 	rl_t *rl;
1570 
1571 	if (byteswap)
1572 		byteswap_uint64_array(lr, sizeof (*lr));
1573 
1574 	offset = lr->lr_offset;
1575 	length = lr->lr_length;
1576 
1577 	/* If it's a dmu_sync() block, write the whole block */
1578 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1579 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1580 		if (length < blocksize) {
1581 			offset -= offset % blocksize;
1582 			length = blocksize;
1583 		}
1584 	}
1585 
1586 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1587 		byteswap_uint64_array(bt, sizeof (*bt));
1588 
1589 	if (bt->bt_magic != BT_MAGIC)
1590 		bt = NULL;
1591 
1592 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1593 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1594 
1595 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1596 
1597 	dmu_object_info_from_db(db, &doi);
1598 
1599 	bbt = ztest_bt_bonus(db);
1600 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1601 	gen = bbt->bt_gen;
1602 	crtxg = bbt->bt_crtxg;
1603 	lrtxg = lr->lr_common.lrc_txg;
1604 
1605 	tx = dmu_tx_create(os);
1606 
1607 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1608 
1609 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1610 	    P2PHASE(offset, length) == 0)
1611 		abuf = dmu_request_arcbuf(db, length);
1612 
1613 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1614 	if (txg == 0) {
1615 		if (abuf != NULL)
1616 			dmu_return_arcbuf(abuf);
1617 		dmu_buf_rele(db, FTAG);
1618 		ztest_range_unlock(rl);
1619 		ztest_object_unlock(zd, lr->lr_foid);
1620 		return (ENOSPC);
1621 	}
1622 
1623 	if (bt != NULL) {
1624 		/*
1625 		 * Usually, verify the old data before writing new data --
1626 		 * but not always, because we also want to verify correct
1627 		 * behavior when the data was not recently read into cache.
1628 		 */
1629 		ASSERT(offset % doi.doi_data_block_size == 0);
1630 		if (ztest_random(4) != 0) {
1631 			int prefetch = ztest_random(2) ?
1632 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1633 			ztest_block_tag_t rbt;
1634 
1635 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1636 			    sizeof (rbt), &rbt, prefetch) == 0);
1637 			if (rbt.bt_magic == BT_MAGIC) {
1638 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1639 				    offset, gen, txg, crtxg);
1640 			}
1641 		}
1642 
1643 		/*
1644 		 * Writes can appear to be newer than the bonus buffer because
1645 		 * the ztest_get_data() callback does a dmu_read() of the
1646 		 * open-context data, which may be different than the data
1647 		 * as it was when the write was generated.
1648 		 */
1649 		if (zd->zd_zilog->zl_replay) {
1650 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1651 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1652 			    bt->bt_crtxg);
1653 		}
1654 
1655 		/*
1656 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1657 		 * so that all of the usual ASSERTs will work.
1658 		 */
1659 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1660 	}
1661 
1662 	if (abuf == NULL) {
1663 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1664 	} else {
1665 		bcopy(data, abuf->b_data, length);
1666 		dmu_assign_arcbuf(db, offset, abuf, tx);
1667 	}
1668 
1669 	(void) ztest_log_write(zd, tx, lr);
1670 
1671 	dmu_buf_rele(db, FTAG);
1672 
1673 	dmu_tx_commit(tx);
1674 
1675 	ztest_range_unlock(rl);
1676 	ztest_object_unlock(zd, lr->lr_foid);
1677 
1678 	return (0);
1679 }
1680 
1681 static int
1682 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1683 {
1684 	objset_t *os = zd->zd_os;
1685 	dmu_tx_t *tx;
1686 	uint64_t txg;
1687 	rl_t *rl;
1688 
1689 	if (byteswap)
1690 		byteswap_uint64_array(lr, sizeof (*lr));
1691 
1692 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1693 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1694 	    RL_WRITER);
1695 
1696 	tx = dmu_tx_create(os);
1697 
1698 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1699 
1700 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1701 	if (txg == 0) {
1702 		ztest_range_unlock(rl);
1703 		ztest_object_unlock(zd, lr->lr_foid);
1704 		return (ENOSPC);
1705 	}
1706 
1707 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1708 	    lr->lr_length, tx) == 0);
1709 
1710 	(void) ztest_log_truncate(zd, tx, lr);
1711 
1712 	dmu_tx_commit(tx);
1713 
1714 	ztest_range_unlock(rl);
1715 	ztest_object_unlock(zd, lr->lr_foid);
1716 
1717 	return (0);
1718 }
1719 
1720 static int
1721 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1722 {
1723 	objset_t *os = zd->zd_os;
1724 	dmu_tx_t *tx;
1725 	dmu_buf_t *db;
1726 	ztest_block_tag_t *bbt;
1727 	uint64_t txg, lrtxg, crtxg;
1728 
1729 	if (byteswap)
1730 		byteswap_uint64_array(lr, sizeof (*lr));
1731 
1732 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1733 
1734 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1735 
1736 	tx = dmu_tx_create(os);
1737 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1738 
1739 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1740 	if (txg == 0) {
1741 		dmu_buf_rele(db, FTAG);
1742 		ztest_object_unlock(zd, lr->lr_foid);
1743 		return (ENOSPC);
1744 	}
1745 
1746 	bbt = ztest_bt_bonus(db);
1747 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1748 	crtxg = bbt->bt_crtxg;
1749 	lrtxg = lr->lr_common.lrc_txg;
1750 
1751 	if (zd->zd_zilog->zl_replay) {
1752 		ASSERT(lr->lr_size != 0);
1753 		ASSERT(lr->lr_mode != 0);
1754 		ASSERT(lrtxg != 0);
1755 	} else {
1756 		/*
1757 		 * Randomly change the size and increment the generation.
1758 		 */
1759 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1760 		    sizeof (*bbt);
1761 		lr->lr_mode = bbt->bt_gen + 1;
1762 		ASSERT(lrtxg == 0);
1763 	}
1764 
1765 	/*
1766 	 * Verify that the current bonus buffer is not newer than our txg.
1767 	 */
1768 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1769 	    MAX(txg, lrtxg), crtxg);
1770 
1771 	dmu_buf_will_dirty(db, tx);
1772 
1773 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1774 	ASSERT3U(lr->lr_size, <=, db->db_size);
1775 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1776 	bbt = ztest_bt_bonus(db);
1777 
1778 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1779 
1780 	dmu_buf_rele(db, FTAG);
1781 
1782 	(void) ztest_log_setattr(zd, tx, lr);
1783 
1784 	dmu_tx_commit(tx);
1785 
1786 	ztest_object_unlock(zd, lr->lr_foid);
1787 
1788 	return (0);
1789 }
1790 
1791 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1792 	NULL,			/* 0 no such transaction type */
1793 	ztest_replay_create,	/* TX_CREATE */
1794 	NULL,			/* TX_MKDIR */
1795 	NULL,			/* TX_MKXATTR */
1796 	NULL,			/* TX_SYMLINK */
1797 	ztest_replay_remove,	/* TX_REMOVE */
1798 	NULL,			/* TX_RMDIR */
1799 	NULL,			/* TX_LINK */
1800 	NULL,			/* TX_RENAME */
1801 	ztest_replay_write,	/* TX_WRITE */
1802 	ztest_replay_truncate,	/* TX_TRUNCATE */
1803 	ztest_replay_setattr,	/* TX_SETATTR */
1804 	NULL,			/* TX_ACL */
1805 	NULL,			/* TX_CREATE_ACL */
1806 	NULL,			/* TX_CREATE_ATTR */
1807 	NULL,			/* TX_CREATE_ACL_ATTR */
1808 	NULL,			/* TX_MKDIR_ACL */
1809 	NULL,			/* TX_MKDIR_ATTR */
1810 	NULL,			/* TX_MKDIR_ACL_ATTR */
1811 	NULL,			/* TX_WRITE2 */
1812 };
1813 
1814 /*
1815  * ZIL get_data callbacks
1816  */
1817 
1818 static void
1819 ztest_get_done(zgd_t *zgd, int error)
1820 {
1821 	ztest_ds_t *zd = zgd->zgd_private;
1822 	uint64_t object = zgd->zgd_rl->rl_object;
1823 
1824 	if (zgd->zgd_db)
1825 		dmu_buf_rele(zgd->zgd_db, zgd);
1826 
1827 	ztest_range_unlock(zgd->zgd_rl);
1828 	ztest_object_unlock(zd, object);
1829 
1830 	if (error == 0 && zgd->zgd_bp)
1831 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1832 
1833 	umem_free(zgd, sizeof (*zgd));
1834 }
1835 
1836 static int
1837 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1838 {
1839 	ztest_ds_t *zd = arg;
1840 	objset_t *os = zd->zd_os;
1841 	uint64_t object = lr->lr_foid;
1842 	uint64_t offset = lr->lr_offset;
1843 	uint64_t size = lr->lr_length;
1844 	blkptr_t *bp = &lr->lr_blkptr;
1845 	uint64_t txg = lr->lr_common.lrc_txg;
1846 	uint64_t crtxg;
1847 	dmu_object_info_t doi;
1848 	dmu_buf_t *db;
1849 	zgd_t *zgd;
1850 	int error;
1851 
1852 	ztest_object_lock(zd, object, RL_READER);
1853 	error = dmu_bonus_hold(os, object, FTAG, &db);
1854 	if (error) {
1855 		ztest_object_unlock(zd, object);
1856 		return (error);
1857 	}
1858 
1859 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1860 
1861 	if (crtxg == 0 || crtxg > txg) {
1862 		dmu_buf_rele(db, FTAG);
1863 		ztest_object_unlock(zd, object);
1864 		return (ENOENT);
1865 	}
1866 
1867 	dmu_object_info_from_db(db, &doi);
1868 	dmu_buf_rele(db, FTAG);
1869 	db = NULL;
1870 
1871 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1872 	zgd->zgd_zilog = zd->zd_zilog;
1873 	zgd->zgd_private = zd;
1874 
1875 	if (buf != NULL) {	/* immediate write */
1876 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1877 		    RL_READER);
1878 
1879 		error = dmu_read(os, object, offset, size, buf,
1880 		    DMU_READ_NO_PREFETCH);
1881 		ASSERT(error == 0);
1882 	} else {
1883 		size = doi.doi_data_block_size;
1884 		if (ISP2(size)) {
1885 			offset = P2ALIGN(offset, size);
1886 		} else {
1887 			ASSERT(offset < size);
1888 			offset = 0;
1889 		}
1890 
1891 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1892 		    RL_READER);
1893 
1894 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1895 		    DMU_READ_NO_PREFETCH);
1896 
1897 		if (error == 0) {
1898 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1899 			if (obp) {
1900 				ASSERT(BP_IS_HOLE(bp));
1901 				*bp = *obp;
1902 			}
1903 
1904 			zgd->zgd_db = db;
1905 			zgd->zgd_bp = bp;
1906 
1907 			ASSERT(db->db_offset == offset);
1908 			ASSERT(db->db_size == size);
1909 
1910 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1911 			    ztest_get_done, zgd);
1912 
1913 			if (error == 0)
1914 				return (0);
1915 		}
1916 	}
1917 
1918 	ztest_get_done(zgd, error);
1919 
1920 	return (error);
1921 }
1922 
1923 static void *
1924 ztest_lr_alloc(size_t lrsize, char *name)
1925 {
1926 	char *lr;
1927 	size_t namesize = name ? strlen(name) + 1 : 0;
1928 
1929 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1930 
1931 	if (name)
1932 		bcopy(name, lr + lrsize, namesize);
1933 
1934 	return (lr);
1935 }
1936 
1937 void
1938 ztest_lr_free(void *lr, size_t lrsize, char *name)
1939 {
1940 	size_t namesize = name ? strlen(name) + 1 : 0;
1941 
1942 	umem_free(lr, lrsize + namesize);
1943 }
1944 
1945 /*
1946  * Lookup a bunch of objects.  Returns the number of objects not found.
1947  */
1948 static int
1949 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1950 {
1951 	int missing = 0;
1952 	int error;
1953 
1954 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1955 
1956 	for (int i = 0; i < count; i++, od++) {
1957 		od->od_object = 0;
1958 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1959 		    sizeof (uint64_t), 1, &od->od_object);
1960 		if (error) {
1961 			ASSERT(error == ENOENT);
1962 			ASSERT(od->od_object == 0);
1963 			missing++;
1964 		} else {
1965 			dmu_buf_t *db;
1966 			ztest_block_tag_t *bbt;
1967 			dmu_object_info_t doi;
1968 
1969 			ASSERT(od->od_object != 0);
1970 			ASSERT(missing == 0);	/* there should be no gaps */
1971 
1972 			ztest_object_lock(zd, od->od_object, RL_READER);
1973 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1974 			    od->od_object, FTAG, &db));
1975 			dmu_object_info_from_db(db, &doi);
1976 			bbt = ztest_bt_bonus(db);
1977 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1978 			od->od_type = doi.doi_type;
1979 			od->od_blocksize = doi.doi_data_block_size;
1980 			od->od_gen = bbt->bt_gen;
1981 			dmu_buf_rele(db, FTAG);
1982 			ztest_object_unlock(zd, od->od_object);
1983 		}
1984 	}
1985 
1986 	return (missing);
1987 }
1988 
1989 static int
1990 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1991 {
1992 	int missing = 0;
1993 
1994 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1995 
1996 	for (int i = 0; i < count; i++, od++) {
1997 		if (missing) {
1998 			od->od_object = 0;
1999 			missing++;
2000 			continue;
2001 		}
2002 
2003 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2004 
2005 		lr->lr_doid = od->od_dir;
2006 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2007 		lr->lrz_type = od->od_crtype;
2008 		lr->lrz_blocksize = od->od_crblocksize;
2009 		lr->lrz_ibshift = ztest_random_ibshift();
2010 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2011 		lr->lrz_bonuslen = dmu_bonus_max();
2012 		lr->lr_gen = od->od_crgen;
2013 		lr->lr_crtime[0] = time(NULL);
2014 
2015 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2016 			ASSERT(missing == 0);
2017 			od->od_object = 0;
2018 			missing++;
2019 		} else {
2020 			od->od_object = lr->lr_foid;
2021 			od->od_type = od->od_crtype;
2022 			od->od_blocksize = od->od_crblocksize;
2023 			od->od_gen = od->od_crgen;
2024 			ASSERT(od->od_object != 0);
2025 		}
2026 
2027 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2028 	}
2029 
2030 	return (missing);
2031 }
2032 
2033 static int
2034 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2035 {
2036 	int missing = 0;
2037 	int error;
2038 
2039 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2040 
2041 	od += count - 1;
2042 
2043 	for (int i = count - 1; i >= 0; i--, od--) {
2044 		if (missing) {
2045 			missing++;
2046 			continue;
2047 		}
2048 
2049 		/*
2050 		 * No object was found.
2051 		 */
2052 		if (od->od_object == 0)
2053 			continue;
2054 
2055 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2056 
2057 		lr->lr_doid = od->od_dir;
2058 
2059 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2060 			ASSERT3U(error, ==, ENOSPC);
2061 			missing++;
2062 		} else {
2063 			od->od_object = 0;
2064 		}
2065 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2066 	}
2067 
2068 	return (missing);
2069 }
2070 
2071 static int
2072 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2073     void *data)
2074 {
2075 	lr_write_t *lr;
2076 	int error;
2077 
2078 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2079 
2080 	lr->lr_foid = object;
2081 	lr->lr_offset = offset;
2082 	lr->lr_length = size;
2083 	lr->lr_blkoff = 0;
2084 	BP_ZERO(&lr->lr_blkptr);
2085 
2086 	bcopy(data, lr + 1, size);
2087 
2088 	error = ztest_replay_write(zd, lr, B_FALSE);
2089 
2090 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2091 
2092 	return (error);
2093 }
2094 
2095 static int
2096 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2097 {
2098 	lr_truncate_t *lr;
2099 	int error;
2100 
2101 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2102 
2103 	lr->lr_foid = object;
2104 	lr->lr_offset = offset;
2105 	lr->lr_length = size;
2106 
2107 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2108 
2109 	ztest_lr_free(lr, sizeof (*lr), NULL);
2110 
2111 	return (error);
2112 }
2113 
2114 static int
2115 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2116 {
2117 	lr_setattr_t *lr;
2118 	int error;
2119 
2120 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2121 
2122 	lr->lr_foid = object;
2123 	lr->lr_size = 0;
2124 	lr->lr_mode = 0;
2125 
2126 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2127 
2128 	ztest_lr_free(lr, sizeof (*lr), NULL);
2129 
2130 	return (error);
2131 }
2132 
2133 static void
2134 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2135 {
2136 	objset_t *os = zd->zd_os;
2137 	dmu_tx_t *tx;
2138 	uint64_t txg;
2139 	rl_t *rl;
2140 
2141 	txg_wait_synced(dmu_objset_pool(os), 0);
2142 
2143 	ztest_object_lock(zd, object, RL_READER);
2144 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2145 
2146 	tx = dmu_tx_create(os);
2147 
2148 	dmu_tx_hold_write(tx, object, offset, size);
2149 
2150 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2151 
2152 	if (txg != 0) {
2153 		dmu_prealloc(os, object, offset, size, tx);
2154 		dmu_tx_commit(tx);
2155 		txg_wait_synced(dmu_objset_pool(os), txg);
2156 	} else {
2157 		(void) dmu_free_long_range(os, object, offset, size);
2158 	}
2159 
2160 	ztest_range_unlock(rl);
2161 	ztest_object_unlock(zd, object);
2162 }
2163 
2164 static void
2165 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2166 {
2167 	int err;
2168 	ztest_block_tag_t wbt;
2169 	dmu_object_info_t doi;
2170 	enum ztest_io_type io_type;
2171 	uint64_t blocksize;
2172 	void *data;
2173 
2174 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2175 	blocksize = doi.doi_data_block_size;
2176 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2177 
2178 	/*
2179 	 * Pick an i/o type at random, biased toward writing block tags.
2180 	 */
2181 	io_type = ztest_random(ZTEST_IO_TYPES);
2182 	if (ztest_random(2) == 0)
2183 		io_type = ZTEST_IO_WRITE_TAG;
2184 
2185 	(void) rw_rdlock(&zd->zd_zilog_lock);
2186 
2187 	switch (io_type) {
2188 
2189 	case ZTEST_IO_WRITE_TAG:
2190 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2191 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2192 		break;
2193 
2194 	case ZTEST_IO_WRITE_PATTERN:
2195 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2196 		if (ztest_random(2) == 0) {
2197 			/*
2198 			 * Induce fletcher2 collisions to ensure that
2199 			 * zio_ddt_collision() detects and resolves them
2200 			 * when using fletcher2-verify for deduplication.
2201 			 */
2202 			((uint64_t *)data)[0] ^= 1ULL << 63;
2203 			((uint64_t *)data)[4] ^= 1ULL << 63;
2204 		}
2205 		(void) ztest_write(zd, object, offset, blocksize, data);
2206 		break;
2207 
2208 	case ZTEST_IO_WRITE_ZEROES:
2209 		bzero(data, blocksize);
2210 		(void) ztest_write(zd, object, offset, blocksize, data);
2211 		break;
2212 
2213 	case ZTEST_IO_TRUNCATE:
2214 		(void) ztest_truncate(zd, object, offset, blocksize);
2215 		break;
2216 
2217 	case ZTEST_IO_SETATTR:
2218 		(void) ztest_setattr(zd, object);
2219 		break;
2220 
2221 	case ZTEST_IO_REWRITE:
2222 		(void) rw_rdlock(&ztest_name_lock);
2223 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2224 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2225 		    B_FALSE);
2226 		VERIFY(err == 0 || err == ENOSPC);
2227 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2228 		    ZFS_PROP_COMPRESSION,
2229 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2230 		    B_FALSE);
2231 		VERIFY(err == 0 || err == ENOSPC);
2232 		(void) rw_unlock(&ztest_name_lock);
2233 
2234 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2235 		    DMU_READ_NO_PREFETCH));
2236 
2237 		(void) ztest_write(zd, object, offset, blocksize, data);
2238 		break;
2239 	}
2240 
2241 	(void) rw_unlock(&zd->zd_zilog_lock);
2242 
2243 	umem_free(data, blocksize);
2244 }
2245 
2246 /*
2247  * Initialize an object description template.
2248  */
2249 static void
2250 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2251     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2252 {
2253 	od->od_dir = ZTEST_DIROBJ;
2254 	od->od_object = 0;
2255 
2256 	od->od_crtype = type;
2257 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2258 	od->od_crgen = gen;
2259 
2260 	od->od_type = DMU_OT_NONE;
2261 	od->od_blocksize = 0;
2262 	od->od_gen = 0;
2263 
2264 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2265 	    tag, (int64_t)id, index);
2266 }
2267 
2268 /*
2269  * Lookup or create the objects for a test using the od template.
2270  * If the objects do not all exist, or if 'remove' is specified,
2271  * remove any existing objects and create new ones.  Otherwise,
2272  * use the existing objects.
2273  */
2274 static int
2275 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2276 {
2277 	int count = size / sizeof (*od);
2278 	int rv = 0;
2279 
2280 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2281 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2282 	    (ztest_remove(zd, od, count) != 0 ||
2283 	    ztest_create(zd, od, count) != 0))
2284 		rv = -1;
2285 	zd->zd_od = od;
2286 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2287 
2288 	return (rv);
2289 }
2290 
2291 /* ARGSUSED */
2292 void
2293 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2294 {
2295 	zilog_t *zilog = zd->zd_zilog;
2296 
2297 	(void) rw_rdlock(&zd->zd_zilog_lock);
2298 
2299 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2300 
2301 	/*
2302 	 * Remember the committed values in zd, which is in parent/child
2303 	 * shared memory.  If we die, the next iteration of ztest_run()
2304 	 * will verify that the log really does contain this record.
2305 	 */
2306 	mutex_enter(&zilog->zl_lock);
2307 	ASSERT(zd->zd_shared != NULL);
2308 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2309 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2310 	mutex_exit(&zilog->zl_lock);
2311 
2312 	(void) rw_unlock(&zd->zd_zilog_lock);
2313 }
2314 
2315 /*
2316  * This function is designed to simulate the operations that occur during a
2317  * mount/unmount operation.  We hold the dataset across these operations in an
2318  * attempt to expose any implicit assumptions about ZIL management.
2319  */
2320 /* ARGSUSED */
2321 void
2322 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2323 {
2324 	objset_t *os = zd->zd_os;
2325 
2326 	/*
2327 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2328 	 * updating the zil (i.e. adding in-memory log records) and the
2329 	 * zd_zilog_lock to block any I/O.
2330 	 */
2331 	VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2332 	(void) rw_wrlock(&zd->zd_zilog_lock);
2333 
2334 	/* zfsvfs_teardown() */
2335 	zil_close(zd->zd_zilog);
2336 
2337 	/* zfsvfs_setup() */
2338 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2339 	zil_replay(os, zd, ztest_replay_vector);
2340 
2341 	(void) rw_unlock(&zd->zd_zilog_lock);
2342 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2343 }
2344 
2345 /*
2346  * Verify that we can't destroy an active pool, create an existing pool,
2347  * or create a pool with a bad vdev spec.
2348  */
2349 /* ARGSUSED */
2350 void
2351 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2352 {
2353 	ztest_shared_opts_t *zo = &ztest_opts;
2354 	spa_t *spa;
2355 	nvlist_t *nvroot;
2356 
2357 	/*
2358 	 * Attempt to create using a bad file.
2359 	 */
2360 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2361 	VERIFY3U(ENOENT, ==,
2362 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2363 	nvlist_free(nvroot);
2364 
2365 	/*
2366 	 * Attempt to create using a bad mirror.
2367 	 */
2368 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2369 	VERIFY3U(ENOENT, ==,
2370 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2371 	nvlist_free(nvroot);
2372 
2373 	/*
2374 	 * Attempt to create an existing pool.  It shouldn't matter
2375 	 * what's in the nvroot; we should fail with EEXIST.
2376 	 */
2377 	(void) rw_rdlock(&ztest_name_lock);
2378 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2379 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2380 	nvlist_free(nvroot);
2381 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2382 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2383 	spa_close(spa, FTAG);
2384 
2385 	(void) rw_unlock(&ztest_name_lock);
2386 }
2387 
2388 /* ARGSUSED */
2389 void
2390 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2391 {
2392 	spa_t *spa;
2393 	uint64_t initial_version = SPA_VERSION_INITIAL;
2394 	uint64_t version, newversion;
2395 	nvlist_t *nvroot, *props;
2396 	char *name;
2397 
2398 	VERIFY0(mutex_lock(&ztest_vdev_lock));
2399 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2400 
2401 	/*
2402 	 * Clean up from previous runs.
2403 	 */
2404 	(void) spa_destroy(name);
2405 
2406 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2407 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2408 
2409 	/*
2410 	 * If we're configuring a RAIDZ device then make sure that the
2411 	 * the initial version is capable of supporting that feature.
2412 	 */
2413 	switch (ztest_opts.zo_raidz_parity) {
2414 	case 0:
2415 	case 1:
2416 		initial_version = SPA_VERSION_INITIAL;
2417 		break;
2418 	case 2:
2419 		initial_version = SPA_VERSION_RAIDZ2;
2420 		break;
2421 	case 3:
2422 		initial_version = SPA_VERSION_RAIDZ3;
2423 		break;
2424 	}
2425 
2426 	/*
2427 	 * Create a pool with a spa version that can be upgraded. Pick
2428 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2429 	 */
2430 	do {
2431 		version = ztest_random_spa_version(initial_version);
2432 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2433 
2434 	props = fnvlist_alloc();
2435 	fnvlist_add_uint64(props,
2436 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2437 	VERIFY0(spa_create(name, nvroot, props, NULL));
2438 	fnvlist_free(nvroot);
2439 	fnvlist_free(props);
2440 
2441 	VERIFY0(spa_open(name, &spa, FTAG));
2442 	VERIFY3U(spa_version(spa), ==, version);
2443 	newversion = ztest_random_spa_version(version + 1);
2444 
2445 	if (ztest_opts.zo_verbose >= 4) {
2446 		(void) printf("upgrading spa version from %llu to %llu\n",
2447 		    (u_longlong_t)version, (u_longlong_t)newversion);
2448 	}
2449 
2450 	spa_upgrade(spa, newversion);
2451 	VERIFY3U(spa_version(spa), >, version);
2452 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2453 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2454 	spa_close(spa, FTAG);
2455 
2456 	strfree(name);
2457 	VERIFY0(mutex_unlock(&ztest_vdev_lock));
2458 }
2459 
2460 static vdev_t *
2461 vdev_lookup_by_path(vdev_t *vd, const char *path)
2462 {
2463 	vdev_t *mvd;
2464 
2465 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2466 		return (vd);
2467 
2468 	for (int c = 0; c < vd->vdev_children; c++)
2469 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2470 		    NULL)
2471 			return (mvd);
2472 
2473 	return (NULL);
2474 }
2475 
2476 /*
2477  * Find the first available hole which can be used as a top-level.
2478  */
2479 int
2480 find_vdev_hole(spa_t *spa)
2481 {
2482 	vdev_t *rvd = spa->spa_root_vdev;
2483 	int c;
2484 
2485 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2486 
2487 	for (c = 0; c < rvd->vdev_children; c++) {
2488 		vdev_t *cvd = rvd->vdev_child[c];
2489 
2490 		if (cvd->vdev_ishole)
2491 			break;
2492 	}
2493 	return (c);
2494 }
2495 
2496 /*
2497  * Verify that vdev_add() works as expected.
2498  */
2499 /* ARGSUSED */
2500 void
2501 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2502 {
2503 	ztest_shared_t *zs = ztest_shared;
2504 	spa_t *spa = ztest_spa;
2505 	uint64_t leaves;
2506 	uint64_t guid;
2507 	nvlist_t *nvroot;
2508 	int error;
2509 
2510 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2511 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2512 
2513 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2514 
2515 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2516 
2517 	/*
2518 	 * If we have slogs then remove them 1/4 of the time.
2519 	 */
2520 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2521 		/*
2522 		 * Grab the guid from the head of the log class rotor.
2523 		 */
2524 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2525 
2526 		spa_config_exit(spa, SCL_VDEV, FTAG);
2527 
2528 		/*
2529 		 * We have to grab the zs_name_lock as writer to
2530 		 * prevent a race between removing a slog (dmu_objset_find)
2531 		 * and destroying a dataset. Removing the slog will
2532 		 * grab a reference on the dataset which may cause
2533 		 * dmu_objset_destroy() to fail with EBUSY thus
2534 		 * leaving the dataset in an inconsistent state.
2535 		 */
2536 		VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2537 		error = spa_vdev_remove(spa, guid, B_FALSE);
2538 		VERIFY(rw_unlock(&ztest_name_lock) == 0);
2539 
2540 		if (error && error != EEXIST)
2541 			fatal(0, "spa_vdev_remove() = %d", error);
2542 	} else {
2543 		spa_config_exit(spa, SCL_VDEV, FTAG);
2544 
2545 		/*
2546 		 * Make 1/4 of the devices be log devices.
2547 		 */
2548 		nvroot = make_vdev_root(NULL, NULL, NULL,
2549 		    ztest_opts.zo_vdev_size, 0,
2550 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2551 		    zs->zs_mirrors, 1);
2552 
2553 		error = spa_vdev_add(spa, nvroot);
2554 		nvlist_free(nvroot);
2555 
2556 		if (error == ENOSPC)
2557 			ztest_record_enospc("spa_vdev_add");
2558 		else if (error != 0)
2559 			fatal(0, "spa_vdev_add() = %d", error);
2560 	}
2561 
2562 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2563 }
2564 
2565 /*
2566  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2567  */
2568 /* ARGSUSED */
2569 void
2570 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2571 {
2572 	ztest_shared_t *zs = ztest_shared;
2573 	spa_t *spa = ztest_spa;
2574 	vdev_t *rvd = spa->spa_root_vdev;
2575 	spa_aux_vdev_t *sav;
2576 	char *aux;
2577 	uint64_t guid = 0;
2578 	int error;
2579 
2580 	if (ztest_random(2) == 0) {
2581 		sav = &spa->spa_spares;
2582 		aux = ZPOOL_CONFIG_SPARES;
2583 	} else {
2584 		sav = &spa->spa_l2cache;
2585 		aux = ZPOOL_CONFIG_L2CACHE;
2586 	}
2587 
2588 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2589 
2590 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2591 
2592 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2593 		/*
2594 		 * Pick a random device to remove.
2595 		 */
2596 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2597 	} else {
2598 		/*
2599 		 * Find an unused device we can add.
2600 		 */
2601 		zs->zs_vdev_aux = 0;
2602 		for (;;) {
2603 			char path[MAXPATHLEN];
2604 			int c;
2605 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2606 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2607 			    zs->zs_vdev_aux);
2608 			for (c = 0; c < sav->sav_count; c++)
2609 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2610 				    path) == 0)
2611 					break;
2612 			if (c == sav->sav_count &&
2613 			    vdev_lookup_by_path(rvd, path) == NULL)
2614 				break;
2615 			zs->zs_vdev_aux++;
2616 		}
2617 	}
2618 
2619 	spa_config_exit(spa, SCL_VDEV, FTAG);
2620 
2621 	if (guid == 0) {
2622 		/*
2623 		 * Add a new device.
2624 		 */
2625 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2626 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2627 		error = spa_vdev_add(spa, nvroot);
2628 		if (error != 0)
2629 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2630 		nvlist_free(nvroot);
2631 	} else {
2632 		/*
2633 		 * Remove an existing device.  Sometimes, dirty its
2634 		 * vdev state first to make sure we handle removal
2635 		 * of devices that have pending state changes.
2636 		 */
2637 		if (ztest_random(2) == 0)
2638 			(void) vdev_online(spa, guid, 0, NULL);
2639 
2640 		error = spa_vdev_remove(spa, guid, B_FALSE);
2641 		if (error != 0 && error != EBUSY)
2642 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2643 	}
2644 
2645 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2646 }
2647 
2648 /*
2649  * split a pool if it has mirror tlvdevs
2650  */
2651 /* ARGSUSED */
2652 void
2653 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2654 {
2655 	ztest_shared_t *zs = ztest_shared;
2656 	spa_t *spa = ztest_spa;
2657 	vdev_t *rvd = spa->spa_root_vdev;
2658 	nvlist_t *tree, **child, *config, *split, **schild;
2659 	uint_t c, children, schildren = 0, lastlogid = 0;
2660 	int error = 0;
2661 
2662 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2663 
2664 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2665 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2666 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2667 		return;
2668 	}
2669 
2670 	/* clean up the old pool, if any */
2671 	(void) spa_destroy("splitp");
2672 
2673 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2674 
2675 	/* generate a config from the existing config */
2676 	mutex_enter(&spa->spa_props_lock);
2677 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2678 	    &tree) == 0);
2679 	mutex_exit(&spa->spa_props_lock);
2680 
2681 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2682 	    &children) == 0);
2683 
2684 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2685 	for (c = 0; c < children; c++) {
2686 		vdev_t *tvd = rvd->vdev_child[c];
2687 		nvlist_t **mchild;
2688 		uint_t mchildren;
2689 
2690 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2691 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2692 			    0) == 0);
2693 			VERIFY(nvlist_add_string(schild[schildren],
2694 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2695 			VERIFY(nvlist_add_uint64(schild[schildren],
2696 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2697 			if (lastlogid == 0)
2698 				lastlogid = schildren;
2699 			++schildren;
2700 			continue;
2701 		}
2702 		lastlogid = 0;
2703 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2704 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2705 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2706 	}
2707 
2708 	/* OK, create a config that can be used to split */
2709 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2710 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2711 	    VDEV_TYPE_ROOT) == 0);
2712 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2713 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2714 
2715 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2716 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2717 
2718 	for (c = 0; c < schildren; c++)
2719 		nvlist_free(schild[c]);
2720 	free(schild);
2721 	nvlist_free(split);
2722 
2723 	spa_config_exit(spa, SCL_VDEV, FTAG);
2724 
2725 	(void) rw_wrlock(&ztest_name_lock);
2726 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2727 	(void) rw_unlock(&ztest_name_lock);
2728 
2729 	nvlist_free(config);
2730 
2731 	if (error == 0) {
2732 		(void) printf("successful split - results:\n");
2733 		mutex_enter(&spa_namespace_lock);
2734 		show_pool_stats(spa);
2735 		show_pool_stats(spa_lookup("splitp"));
2736 		mutex_exit(&spa_namespace_lock);
2737 		++zs->zs_splits;
2738 		--zs->zs_mirrors;
2739 	}
2740 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2741 
2742 }
2743 
2744 /*
2745  * Verify that we can attach and detach devices.
2746  */
2747 /* ARGSUSED */
2748 void
2749 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2750 {
2751 	ztest_shared_t *zs = ztest_shared;
2752 	spa_t *spa = ztest_spa;
2753 	spa_aux_vdev_t *sav = &spa->spa_spares;
2754 	vdev_t *rvd = spa->spa_root_vdev;
2755 	vdev_t *oldvd, *newvd, *pvd;
2756 	nvlist_t *root;
2757 	uint64_t leaves;
2758 	uint64_t leaf, top;
2759 	uint64_t ashift = ztest_get_ashift();
2760 	uint64_t oldguid, pguid;
2761 	uint64_t oldsize, newsize;
2762 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2763 	int replacing;
2764 	int oldvd_has_siblings = B_FALSE;
2765 	int newvd_is_spare = B_FALSE;
2766 	int oldvd_is_log;
2767 	int error, expected_error;
2768 
2769 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2770 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2771 
2772 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2773 
2774 	/*
2775 	 * Decide whether to do an attach or a replace.
2776 	 */
2777 	replacing = ztest_random(2);
2778 
2779 	/*
2780 	 * Pick a random top-level vdev.
2781 	 */
2782 	top = ztest_random_vdev_top(spa, B_TRUE);
2783 
2784 	/*
2785 	 * Pick a random leaf within it.
2786 	 */
2787 	leaf = ztest_random(leaves);
2788 
2789 	/*
2790 	 * Locate this vdev.
2791 	 */
2792 	oldvd = rvd->vdev_child[top];
2793 	if (zs->zs_mirrors >= 1) {
2794 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2795 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2796 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2797 	}
2798 	if (ztest_opts.zo_raidz > 1) {
2799 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2800 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2801 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2802 	}
2803 
2804 	/*
2805 	 * If we're already doing an attach or replace, oldvd may be a
2806 	 * mirror vdev -- in which case, pick a random child.
2807 	 */
2808 	while (oldvd->vdev_children != 0) {
2809 		oldvd_has_siblings = B_TRUE;
2810 		ASSERT(oldvd->vdev_children >= 2);
2811 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2812 	}
2813 
2814 	oldguid = oldvd->vdev_guid;
2815 	oldsize = vdev_get_min_asize(oldvd);
2816 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2817 	(void) strcpy(oldpath, oldvd->vdev_path);
2818 	pvd = oldvd->vdev_parent;
2819 	pguid = pvd->vdev_guid;
2820 
2821 	/*
2822 	 * If oldvd has siblings, then half of the time, detach it.
2823 	 */
2824 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2825 		spa_config_exit(spa, SCL_VDEV, FTAG);
2826 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2827 		if (error != 0 && error != ENODEV && error != EBUSY &&
2828 		    error != ENOTSUP)
2829 			fatal(0, "detach (%s) returned %d", oldpath, error);
2830 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2831 		return;
2832 	}
2833 
2834 	/*
2835 	 * For the new vdev, choose with equal probability between the two
2836 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2837 	 */
2838 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2839 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2840 		newvd_is_spare = B_TRUE;
2841 		(void) strcpy(newpath, newvd->vdev_path);
2842 	} else {
2843 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2844 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
2845 		    top * leaves + leaf);
2846 		if (ztest_random(2) == 0)
2847 			newpath[strlen(newpath) - 1] = 'b';
2848 		newvd = vdev_lookup_by_path(rvd, newpath);
2849 	}
2850 
2851 	if (newvd) {
2852 		newsize = vdev_get_min_asize(newvd);
2853 	} else {
2854 		/*
2855 		 * Make newsize a little bigger or smaller than oldsize.
2856 		 * If it's smaller, the attach should fail.
2857 		 * If it's larger, and we're doing a replace,
2858 		 * we should get dynamic LUN growth when we're done.
2859 		 */
2860 		newsize = 10 * oldsize / (9 + ztest_random(3));
2861 	}
2862 
2863 	/*
2864 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2865 	 * unless it's a replace; in that case any non-replacing parent is OK.
2866 	 *
2867 	 * If newvd is already part of the pool, it should fail with EBUSY.
2868 	 *
2869 	 * If newvd is too small, it should fail with EOVERFLOW.
2870 	 */
2871 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2872 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2873 	    pvd->vdev_ops == &vdev_replacing_ops ||
2874 	    pvd->vdev_ops == &vdev_spare_ops))
2875 		expected_error = ENOTSUP;
2876 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2877 		expected_error = ENOTSUP;
2878 	else if (newvd == oldvd)
2879 		expected_error = replacing ? 0 : EBUSY;
2880 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2881 		expected_error = EBUSY;
2882 	else if (newsize < oldsize)
2883 		expected_error = EOVERFLOW;
2884 	else if (ashift > oldvd->vdev_top->vdev_ashift)
2885 		expected_error = EDOM;
2886 	else
2887 		expected_error = 0;
2888 
2889 	spa_config_exit(spa, SCL_VDEV, FTAG);
2890 
2891 	/*
2892 	 * Build the nvlist describing newpath.
2893 	 */
2894 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2895 	    ashift, 0, 0, 0, 1);
2896 
2897 	error = spa_vdev_attach(spa, oldguid, root, replacing);
2898 
2899 	nvlist_free(root);
2900 
2901 	/*
2902 	 * If our parent was the replacing vdev, but the replace completed,
2903 	 * then instead of failing with ENOTSUP we may either succeed,
2904 	 * fail with ENODEV, or fail with EOVERFLOW.
2905 	 */
2906 	if (expected_error == ENOTSUP &&
2907 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
2908 		expected_error = error;
2909 
2910 	/*
2911 	 * If someone grew the LUN, the replacement may be too small.
2912 	 */
2913 	if (error == EOVERFLOW || error == EBUSY)
2914 		expected_error = error;
2915 
2916 	/* XXX workaround 6690467 */
2917 	if (error != expected_error && expected_error != EBUSY) {
2918 		fatal(0, "attach (%s %llu, %s %llu, %d) "
2919 		    "returned %d, expected %d",
2920 		    oldpath, oldsize, newpath,
2921 		    newsize, replacing, error, expected_error);
2922 	}
2923 
2924 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2925 }
2926 
2927 /*
2928  * Callback function which expands the physical size of the vdev.
2929  */
2930 vdev_t *
2931 grow_vdev(vdev_t *vd, void *arg)
2932 {
2933 	spa_t *spa = vd->vdev_spa;
2934 	size_t *newsize = arg;
2935 	size_t fsize;
2936 	int fd;
2937 
2938 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2939 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2940 
2941 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2942 		return (vd);
2943 
2944 	fsize = lseek(fd, 0, SEEK_END);
2945 	(void) ftruncate(fd, *newsize);
2946 
2947 	if (ztest_opts.zo_verbose >= 6) {
2948 		(void) printf("%s grew from %lu to %lu bytes\n",
2949 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2950 	}
2951 	(void) close(fd);
2952 	return (NULL);
2953 }
2954 
2955 /*
2956  * Callback function which expands a given vdev by calling vdev_online().
2957  */
2958 /* ARGSUSED */
2959 vdev_t *
2960 online_vdev(vdev_t *vd, void *arg)
2961 {
2962 	spa_t *spa = vd->vdev_spa;
2963 	vdev_t *tvd = vd->vdev_top;
2964 	uint64_t guid = vd->vdev_guid;
2965 	uint64_t generation = spa->spa_config_generation + 1;
2966 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2967 	int error;
2968 
2969 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2970 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2971 
2972 	/* Calling vdev_online will initialize the new metaslabs */
2973 	spa_config_exit(spa, SCL_STATE, spa);
2974 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2975 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2976 
2977 	/*
2978 	 * If vdev_online returned an error or the underlying vdev_open
2979 	 * failed then we abort the expand. The only way to know that
2980 	 * vdev_open fails is by checking the returned newstate.
2981 	 */
2982 	if (error || newstate != VDEV_STATE_HEALTHY) {
2983 		if (ztest_opts.zo_verbose >= 5) {
2984 			(void) printf("Unable to expand vdev, state %llu, "
2985 			    "error %d\n", (u_longlong_t)newstate, error);
2986 		}
2987 		return (vd);
2988 	}
2989 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2990 
2991 	/*
2992 	 * Since we dropped the lock we need to ensure that we're
2993 	 * still talking to the original vdev. It's possible this
2994 	 * vdev may have been detached/replaced while we were
2995 	 * trying to online it.
2996 	 */
2997 	if (generation != spa->spa_config_generation) {
2998 		if (ztest_opts.zo_verbose >= 5) {
2999 			(void) printf("vdev configuration has changed, "
3000 			    "guid %llu, state %llu, expected gen %llu, "
3001 			    "got gen %llu\n",
3002 			    (u_longlong_t)guid,
3003 			    (u_longlong_t)tvd->vdev_state,
3004 			    (u_longlong_t)generation,
3005 			    (u_longlong_t)spa->spa_config_generation);
3006 		}
3007 		return (vd);
3008 	}
3009 	return (NULL);
3010 }
3011 
3012 /*
3013  * Traverse the vdev tree calling the supplied function.
3014  * We continue to walk the tree until we either have walked all
3015  * children or we receive a non-NULL return from the callback.
3016  * If a NULL callback is passed, then we just return back the first
3017  * leaf vdev we encounter.
3018  */
3019 vdev_t *
3020 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3021 {
3022 	if (vd->vdev_ops->vdev_op_leaf) {
3023 		if (func == NULL)
3024 			return (vd);
3025 		else
3026 			return (func(vd, arg));
3027 	}
3028 
3029 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3030 		vdev_t *cvd = vd->vdev_child[c];
3031 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3032 			return (cvd);
3033 	}
3034 	return (NULL);
3035 }
3036 
3037 /*
3038  * Verify that dynamic LUN growth works as expected.
3039  */
3040 /* ARGSUSED */
3041 void
3042 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3043 {
3044 	spa_t *spa = ztest_spa;
3045 	vdev_t *vd, *tvd;
3046 	metaslab_class_t *mc;
3047 	metaslab_group_t *mg;
3048 	size_t psize, newsize;
3049 	uint64_t top;
3050 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3051 
3052 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3053 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3054 
3055 	top = ztest_random_vdev_top(spa, B_TRUE);
3056 
3057 	tvd = spa->spa_root_vdev->vdev_child[top];
3058 	mg = tvd->vdev_mg;
3059 	mc = mg->mg_class;
3060 	old_ms_count = tvd->vdev_ms_count;
3061 	old_class_space = metaslab_class_get_space(mc);
3062 
3063 	/*
3064 	 * Determine the size of the first leaf vdev associated with
3065 	 * our top-level device.
3066 	 */
3067 	vd = vdev_walk_tree(tvd, NULL, NULL);
3068 	ASSERT3P(vd, !=, NULL);
3069 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3070 
3071 	psize = vd->vdev_psize;
3072 
3073 	/*
3074 	 * We only try to expand the vdev if it's healthy, less than 4x its
3075 	 * original size, and it has a valid psize.
3076 	 */
3077 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3078 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3079 		spa_config_exit(spa, SCL_STATE, spa);
3080 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3081 		return;
3082 	}
3083 	ASSERT(psize > 0);
3084 	newsize = psize + psize / 8;
3085 	ASSERT3U(newsize, >, psize);
3086 
3087 	if (ztest_opts.zo_verbose >= 6) {
3088 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3089 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3090 	}
3091 
3092 	/*
3093 	 * Growing the vdev is a two step process:
3094 	 *	1). expand the physical size (i.e. relabel)
3095 	 *	2). online the vdev to create the new metaslabs
3096 	 */
3097 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3098 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3099 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3100 		if (ztest_opts.zo_verbose >= 5) {
3101 			(void) printf("Could not expand LUN because "
3102 			    "the vdev configuration changed.\n");
3103 		}
3104 		spa_config_exit(spa, SCL_STATE, spa);
3105 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3106 		return;
3107 	}
3108 
3109 	spa_config_exit(spa, SCL_STATE, spa);
3110 
3111 	/*
3112 	 * Expanding the LUN will update the config asynchronously,
3113 	 * thus we must wait for the async thread to complete any
3114 	 * pending tasks before proceeding.
3115 	 */
3116 	for (;;) {
3117 		boolean_t done;
3118 		mutex_enter(&spa->spa_async_lock);
3119 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3120 		mutex_exit(&spa->spa_async_lock);
3121 		if (done)
3122 			break;
3123 		txg_wait_synced(spa_get_dsl(spa), 0);
3124 		(void) poll(NULL, 0, 100);
3125 	}
3126 
3127 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3128 
3129 	tvd = spa->spa_root_vdev->vdev_child[top];
3130 	new_ms_count = tvd->vdev_ms_count;
3131 	new_class_space = metaslab_class_get_space(mc);
3132 
3133 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3134 		if (ztest_opts.zo_verbose >= 5) {
3135 			(void) printf("Could not verify LUN expansion due to "
3136 			    "intervening vdev offline or remove.\n");
3137 		}
3138 		spa_config_exit(spa, SCL_STATE, spa);
3139 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3140 		return;
3141 	}
3142 
3143 	/*
3144 	 * Make sure we were able to grow the vdev.
3145 	 */
3146 	if (new_ms_count <= old_ms_count)
3147 		fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3148 		    old_ms_count, new_ms_count);
3149 
3150 	/*
3151 	 * Make sure we were able to grow the pool.
3152 	 */
3153 	if (new_class_space <= old_class_space)
3154 		fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3155 		    old_class_space, new_class_space);
3156 
3157 	if (ztest_opts.zo_verbose >= 5) {
3158 		char oldnumbuf[6], newnumbuf[6];
3159 
3160 		nicenum(old_class_space, oldnumbuf);
3161 		nicenum(new_class_space, newnumbuf);
3162 		(void) printf("%s grew from %s to %s\n",
3163 		    spa->spa_name, oldnumbuf, newnumbuf);
3164 	}
3165 
3166 	spa_config_exit(spa, SCL_STATE, spa);
3167 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3168 }
3169 
3170 /*
3171  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3172  */
3173 /* ARGSUSED */
3174 static void
3175 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3176 {
3177 	/*
3178 	 * Create the objects common to all ztest datasets.
3179 	 */
3180 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3181 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3182 }
3183 
3184 static int
3185 ztest_dataset_create(char *dsname)
3186 {
3187 	uint64_t zilset = ztest_random(100);
3188 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3189 	    ztest_objset_create_cb, NULL);
3190 
3191 	if (err || zilset < 80)
3192 		return (err);
3193 
3194 	if (ztest_opts.zo_verbose >= 6)
3195 		(void) printf("Setting dataset %s to sync always\n", dsname);
3196 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3197 	    ZFS_SYNC_ALWAYS, B_FALSE));
3198 }
3199 
3200 /* ARGSUSED */
3201 static int
3202 ztest_objset_destroy_cb(const char *name, void *arg)
3203 {
3204 	objset_t *os;
3205 	dmu_object_info_t doi;
3206 	int error;
3207 
3208 	/*
3209 	 * Verify that the dataset contains a directory object.
3210 	 */
3211 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3212 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3213 	if (error != ENOENT) {
3214 		/* We could have crashed in the middle of destroying it */
3215 		ASSERT0(error);
3216 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3217 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3218 	}
3219 	dmu_objset_disown(os, FTAG);
3220 
3221 	/*
3222 	 * Destroy the dataset.
3223 	 */
3224 	if (strchr(name, '@') != NULL) {
3225 		VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3226 	} else {
3227 		error = dsl_destroy_head(name);
3228 		/* There could be a hold on this dataset */
3229 		if (error != EBUSY)
3230 			ASSERT0(error);
3231 	}
3232 	return (0);
3233 }
3234 
3235 static boolean_t
3236 ztest_snapshot_create(char *osname, uint64_t id)
3237 {
3238 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3239 	int error;
3240 
3241 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3242 
3243 	error = dmu_objset_snapshot_one(osname, snapname);
3244 	if (error == ENOSPC) {
3245 		ztest_record_enospc(FTAG);
3246 		return (B_FALSE);
3247 	}
3248 	if (error != 0 && error != EEXIST) {
3249 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3250 		    snapname, error);
3251 	}
3252 	return (B_TRUE);
3253 }
3254 
3255 static boolean_t
3256 ztest_snapshot_destroy(char *osname, uint64_t id)
3257 {
3258 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3259 	int error;
3260 
3261 	(void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3262 	    (u_longlong_t)id);
3263 
3264 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3265 	if (error != 0 && error != ENOENT)
3266 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3267 	return (B_TRUE);
3268 }
3269 
3270 /* ARGSUSED */
3271 void
3272 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3273 {
3274 	ztest_ds_t zdtmp;
3275 	int iters;
3276 	int error;
3277 	objset_t *os, *os2;
3278 	char name[ZFS_MAX_DATASET_NAME_LEN];
3279 	zilog_t *zilog;
3280 
3281 	(void) rw_rdlock(&ztest_name_lock);
3282 
3283 	(void) snprintf(name, sizeof (name), "%s/temp_%llu",
3284 	    ztest_opts.zo_pool, (u_longlong_t)id);
3285 
3286 	/*
3287 	 * If this dataset exists from a previous run, process its replay log
3288 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3289 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3290 	 */
3291 	if (ztest_random(2) == 0 &&
3292 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3293 		ztest_zd_init(&zdtmp, NULL, os);
3294 		zil_replay(os, &zdtmp, ztest_replay_vector);
3295 		ztest_zd_fini(&zdtmp);
3296 		dmu_objset_disown(os, FTAG);
3297 	}
3298 
3299 	/*
3300 	 * There may be an old instance of the dataset we're about to
3301 	 * create lying around from a previous run.  If so, destroy it
3302 	 * and all of its snapshots.
3303 	 */
3304 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3305 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3306 
3307 	/*
3308 	 * Verify that the destroyed dataset is no longer in the namespace.
3309 	 */
3310 	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3311 	    FTAG, &os));
3312 
3313 	/*
3314 	 * Verify that we can create a new dataset.
3315 	 */
3316 	error = ztest_dataset_create(name);
3317 	if (error) {
3318 		if (error == ENOSPC) {
3319 			ztest_record_enospc(FTAG);
3320 			(void) rw_unlock(&ztest_name_lock);
3321 			return;
3322 		}
3323 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3324 	}
3325 
3326 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3327 
3328 	ztest_zd_init(&zdtmp, NULL, os);
3329 
3330 	/*
3331 	 * Open the intent log for it.
3332 	 */
3333 	zilog = zil_open(os, ztest_get_data);
3334 
3335 	/*
3336 	 * Put some objects in there, do a little I/O to them,
3337 	 * and randomly take a couple of snapshots along the way.
3338 	 */
3339 	iters = ztest_random(5);
3340 	for (int i = 0; i < iters; i++) {
3341 		ztest_dmu_object_alloc_free(&zdtmp, id);
3342 		if (ztest_random(iters) == 0)
3343 			(void) ztest_snapshot_create(name, i);
3344 	}
3345 
3346 	/*
3347 	 * Verify that we cannot create an existing dataset.
3348 	 */
3349 	VERIFY3U(EEXIST, ==,
3350 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3351 
3352 	/*
3353 	 * Verify that we can hold an objset that is also owned.
3354 	 */
3355 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3356 	dmu_objset_rele(os2, FTAG);
3357 
3358 	/*
3359 	 * Verify that we cannot own an objset that is already owned.
3360 	 */
3361 	VERIFY3U(EBUSY, ==,
3362 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3363 
3364 	zil_close(zilog);
3365 	dmu_objset_disown(os, FTAG);
3366 	ztest_zd_fini(&zdtmp);
3367 
3368 	(void) rw_unlock(&ztest_name_lock);
3369 }
3370 
3371 /*
3372  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3373  */
3374 void
3375 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3376 {
3377 	(void) rw_rdlock(&ztest_name_lock);
3378 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3379 	(void) ztest_snapshot_create(zd->zd_name, id);
3380 	(void) rw_unlock(&ztest_name_lock);
3381 }
3382 
3383 /*
3384  * Cleanup non-standard snapshots and clones.
3385  */
3386 void
3387 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3388 {
3389 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3390 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3391 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3392 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3393 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3394 	int error;
3395 
3396 	(void) snprintf(snap1name, sizeof (snap1name),
3397 	    "%s@s1_%llu", osname, id);
3398 	(void) snprintf(clone1name, sizeof (clone1name),
3399 	    "%s/c1_%llu", osname, id);
3400 	(void) snprintf(snap2name, sizeof (snap2name),
3401 	    "%s@s2_%llu", clone1name, id);
3402 	(void) snprintf(clone2name, sizeof (clone2name),
3403 	    "%s/c2_%llu", osname, id);
3404 	(void) snprintf(snap3name, sizeof (snap3name),
3405 	    "%s@s3_%llu", clone1name, id);
3406 
3407 	error = dsl_destroy_head(clone2name);
3408 	if (error && error != ENOENT)
3409 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3410 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3411 	if (error && error != ENOENT)
3412 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3413 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3414 	if (error && error != ENOENT)
3415 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3416 	error = dsl_destroy_head(clone1name);
3417 	if (error && error != ENOENT)
3418 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3419 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3420 	if (error && error != ENOENT)
3421 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3422 }
3423 
3424 /*
3425  * Verify dsl_dataset_promote handles EBUSY
3426  */
3427 void
3428 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3429 {
3430 	objset_t *os;
3431 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3432 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3433 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3434 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3435 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3436 	char *osname = zd->zd_name;
3437 	int error;
3438 
3439 	(void) rw_rdlock(&ztest_name_lock);
3440 
3441 	ztest_dsl_dataset_cleanup(osname, id);
3442 
3443 	(void) snprintf(snap1name, sizeof (snap1name),
3444 	    "%s@s1_%llu", osname, id);
3445 	(void) snprintf(clone1name, sizeof (clone1name),
3446 	    "%s/c1_%llu", osname, id);
3447 	(void) snprintf(snap2name, sizeof (snap2name),
3448 	    "%s@s2_%llu", clone1name, id);
3449 	(void) snprintf(clone2name, sizeof (clone2name),
3450 	    "%s/c2_%llu", osname, id);
3451 	(void) snprintf(snap3name, sizeof (snap3name),
3452 	    "%s@s3_%llu", clone1name, id);
3453 
3454 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3455 	if (error && error != EEXIST) {
3456 		if (error == ENOSPC) {
3457 			ztest_record_enospc(FTAG);
3458 			goto out;
3459 		}
3460 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3461 	}
3462 
3463 	error = dmu_objset_clone(clone1name, snap1name);
3464 	if (error) {
3465 		if (error == ENOSPC) {
3466 			ztest_record_enospc(FTAG);
3467 			goto out;
3468 		}
3469 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3470 	}
3471 
3472 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3473 	if (error && error != EEXIST) {
3474 		if (error == ENOSPC) {
3475 			ztest_record_enospc(FTAG);
3476 			goto out;
3477 		}
3478 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3479 	}
3480 
3481 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3482 	if (error && error != EEXIST) {
3483 		if (error == ENOSPC) {
3484 			ztest_record_enospc(FTAG);
3485 			goto out;
3486 		}
3487 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3488 	}
3489 
3490 	error = dmu_objset_clone(clone2name, snap3name);
3491 	if (error) {
3492 		if (error == ENOSPC) {
3493 			ztest_record_enospc(FTAG);
3494 			goto out;
3495 		}
3496 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3497 	}
3498 
3499 	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3500 	if (error)
3501 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3502 	error = dsl_dataset_promote(clone2name, NULL);
3503 	if (error == ENOSPC) {
3504 		dmu_objset_disown(os, FTAG);
3505 		ztest_record_enospc(FTAG);
3506 		goto out;
3507 	}
3508 	if (error != EBUSY)
3509 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3510 		    error);
3511 	dmu_objset_disown(os, FTAG);
3512 
3513 out:
3514 	ztest_dsl_dataset_cleanup(osname, id);
3515 
3516 	(void) rw_unlock(&ztest_name_lock);
3517 }
3518 
3519 /*
3520  * Verify that dmu_object_{alloc,free} work as expected.
3521  */
3522 void
3523 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3524 {
3525 	ztest_od_t od[4];
3526 	int batchsize = sizeof (od) / sizeof (od[0]);
3527 
3528 	for (int b = 0; b < batchsize; b++)
3529 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3530 
3531 	/*
3532 	 * Destroy the previous batch of objects, create a new batch,
3533 	 * and do some I/O on the new objects.
3534 	 */
3535 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3536 		return;
3537 
3538 	while (ztest_random(4 * batchsize) != 0)
3539 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3540 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3541 }
3542 
3543 /*
3544  * Verify that dmu_{read,write} work as expected.
3545  */
3546 void
3547 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3548 {
3549 	objset_t *os = zd->zd_os;
3550 	ztest_od_t od[2];
3551 	dmu_tx_t *tx;
3552 	int i, freeit, error;
3553 	uint64_t n, s, txg;
3554 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3555 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3556 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3557 	uint64_t regions = 997;
3558 	uint64_t stride = 123456789ULL;
3559 	uint64_t width = 40;
3560 	int free_percent = 5;
3561 
3562 	/*
3563 	 * This test uses two objects, packobj and bigobj, that are always
3564 	 * updated together (i.e. in the same tx) so that their contents are
3565 	 * in sync and can be compared.  Their contents relate to each other
3566 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3567 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3568 	 * for any index n, there are three bufwads that should be identical:
3569 	 *
3570 	 *	packobj, at offset n * sizeof (bufwad_t)
3571 	 *	bigobj, at the head of the nth chunk
3572 	 *	bigobj, at the tail of the nth chunk
3573 	 *
3574 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3575 	 * and it doesn't have any relation to the object blocksize.
3576 	 * The only requirement is that it can hold at least two bufwads.
3577 	 *
3578 	 * Normally, we write the bufwad to each of these locations.
3579 	 * However, free_percent of the time we instead write zeroes to
3580 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3581 	 * bigobj to packobj, we can verify that the DMU is correctly
3582 	 * tracking which parts of an object are allocated and free,
3583 	 * and that the contents of the allocated blocks are correct.
3584 	 */
3585 
3586 	/*
3587 	 * Read the directory info.  If it's the first time, set things up.
3588 	 */
3589 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3590 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3591 
3592 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3593 		return;
3594 
3595 	bigobj = od[0].od_object;
3596 	packobj = od[1].od_object;
3597 	chunksize = od[0].od_gen;
3598 	ASSERT(chunksize == od[1].od_gen);
3599 
3600 	/*
3601 	 * Prefetch a random chunk of the big object.
3602 	 * Our aim here is to get some async reads in flight
3603 	 * for blocks that we may free below; the DMU should
3604 	 * handle this race correctly.
3605 	 */
3606 	n = ztest_random(regions) * stride + ztest_random(width);
3607 	s = 1 + ztest_random(2 * width - 1);
3608 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3609 	    ZIO_PRIORITY_SYNC_READ);
3610 
3611 	/*
3612 	 * Pick a random index and compute the offsets into packobj and bigobj.
3613 	 */
3614 	n = ztest_random(regions) * stride + ztest_random(width);
3615 	s = 1 + ztest_random(width - 1);
3616 
3617 	packoff = n * sizeof (bufwad_t);
3618 	packsize = s * sizeof (bufwad_t);
3619 
3620 	bigoff = n * chunksize;
3621 	bigsize = s * chunksize;
3622 
3623 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3624 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3625 
3626 	/*
3627 	 * free_percent of the time, free a range of bigobj rather than
3628 	 * overwriting it.
3629 	 */
3630 	freeit = (ztest_random(100) < free_percent);
3631 
3632 	/*
3633 	 * Read the current contents of our objects.
3634 	 */
3635 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3636 	    DMU_READ_PREFETCH);
3637 	ASSERT0(error);
3638 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3639 	    DMU_READ_PREFETCH);
3640 	ASSERT0(error);
3641 
3642 	/*
3643 	 * Get a tx for the mods to both packobj and bigobj.
3644 	 */
3645 	tx = dmu_tx_create(os);
3646 
3647 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3648 
3649 	if (freeit)
3650 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3651 	else
3652 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3653 
3654 	/* This accounts for setting the checksum/compression. */
3655 	dmu_tx_hold_bonus(tx, bigobj);
3656 
3657 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3658 	if (txg == 0) {
3659 		umem_free(packbuf, packsize);
3660 		umem_free(bigbuf, bigsize);
3661 		return;
3662 	}
3663 
3664 	enum zio_checksum cksum;
3665 	do {
3666 		cksum = (enum zio_checksum)
3667 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3668 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3669 	dmu_object_set_checksum(os, bigobj, cksum, tx);
3670 
3671 	enum zio_compress comp;
3672 	do {
3673 		comp = (enum zio_compress)
3674 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3675 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3676 	dmu_object_set_compress(os, bigobj, comp, tx);
3677 
3678 	/*
3679 	 * For each index from n to n + s, verify that the existing bufwad
3680 	 * in packobj matches the bufwads at the head and tail of the
3681 	 * corresponding chunk in bigobj.  Then update all three bufwads
3682 	 * with the new values we want to write out.
3683 	 */
3684 	for (i = 0; i < s; i++) {
3685 		/* LINTED */
3686 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3687 		/* LINTED */
3688 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3689 		/* LINTED */
3690 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3691 
3692 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3693 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3694 
3695 		if (pack->bw_txg > txg)
3696 			fatal(0, "future leak: got %llx, open txg is %llx",
3697 			    pack->bw_txg, txg);
3698 
3699 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3700 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3701 			    pack->bw_index, n, i);
3702 
3703 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3704 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3705 
3706 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3707 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3708 
3709 		if (freeit) {
3710 			bzero(pack, sizeof (bufwad_t));
3711 		} else {
3712 			pack->bw_index = n + i;
3713 			pack->bw_txg = txg;
3714 			pack->bw_data = 1 + ztest_random(-2ULL);
3715 		}
3716 		*bigH = *pack;
3717 		*bigT = *pack;
3718 	}
3719 
3720 	/*
3721 	 * We've verified all the old bufwads, and made new ones.
3722 	 * Now write them out.
3723 	 */
3724 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3725 
3726 	if (freeit) {
3727 		if (ztest_opts.zo_verbose >= 7) {
3728 			(void) printf("freeing offset %llx size %llx"
3729 			    " txg %llx\n",
3730 			    (u_longlong_t)bigoff,
3731 			    (u_longlong_t)bigsize,
3732 			    (u_longlong_t)txg);
3733 		}
3734 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3735 	} else {
3736 		if (ztest_opts.zo_verbose >= 7) {
3737 			(void) printf("writing offset %llx size %llx"
3738 			    " txg %llx\n",
3739 			    (u_longlong_t)bigoff,
3740 			    (u_longlong_t)bigsize,
3741 			    (u_longlong_t)txg);
3742 		}
3743 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3744 	}
3745 
3746 	dmu_tx_commit(tx);
3747 
3748 	/*
3749 	 * Sanity check the stuff we just wrote.
3750 	 */
3751 	{
3752 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3753 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3754 
3755 		VERIFY(0 == dmu_read(os, packobj, packoff,
3756 		    packsize, packcheck, DMU_READ_PREFETCH));
3757 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3758 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3759 
3760 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3761 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3762 
3763 		umem_free(packcheck, packsize);
3764 		umem_free(bigcheck, bigsize);
3765 	}
3766 
3767 	umem_free(packbuf, packsize);
3768 	umem_free(bigbuf, bigsize);
3769 }
3770 
3771 void
3772 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3773     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3774 {
3775 	uint64_t i;
3776 	bufwad_t *pack;
3777 	bufwad_t *bigH;
3778 	bufwad_t *bigT;
3779 
3780 	/*
3781 	 * For each index from n to n + s, verify that the existing bufwad
3782 	 * in packobj matches the bufwads at the head and tail of the
3783 	 * corresponding chunk in bigobj.  Then update all three bufwads
3784 	 * with the new values we want to write out.
3785 	 */
3786 	for (i = 0; i < s; i++) {
3787 		/* LINTED */
3788 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3789 		/* LINTED */
3790 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3791 		/* LINTED */
3792 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3793 
3794 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3795 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3796 
3797 		if (pack->bw_txg > txg)
3798 			fatal(0, "future leak: got %llx, open txg is %llx",
3799 			    pack->bw_txg, txg);
3800 
3801 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3802 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3803 			    pack->bw_index, n, i);
3804 
3805 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3806 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3807 
3808 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3809 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3810 
3811 		pack->bw_index = n + i;
3812 		pack->bw_txg = txg;
3813 		pack->bw_data = 1 + ztest_random(-2ULL);
3814 
3815 		*bigH = *pack;
3816 		*bigT = *pack;
3817 	}
3818 }
3819 
3820 void
3821 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3822 {
3823 	objset_t *os = zd->zd_os;
3824 	ztest_od_t od[2];
3825 	dmu_tx_t *tx;
3826 	uint64_t i;
3827 	int error;
3828 	uint64_t n, s, txg;
3829 	bufwad_t *packbuf, *bigbuf;
3830 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3831 	uint64_t blocksize = ztest_random_blocksize();
3832 	uint64_t chunksize = blocksize;
3833 	uint64_t regions = 997;
3834 	uint64_t stride = 123456789ULL;
3835 	uint64_t width = 9;
3836 	dmu_buf_t *bonus_db;
3837 	arc_buf_t **bigbuf_arcbufs;
3838 	dmu_object_info_t doi;
3839 
3840 	/*
3841 	 * This test uses two objects, packobj and bigobj, that are always
3842 	 * updated together (i.e. in the same tx) so that their contents are
3843 	 * in sync and can be compared.  Their contents relate to each other
3844 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3845 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3846 	 * for any index n, there are three bufwads that should be identical:
3847 	 *
3848 	 *	packobj, at offset n * sizeof (bufwad_t)
3849 	 *	bigobj, at the head of the nth chunk
3850 	 *	bigobj, at the tail of the nth chunk
3851 	 *
3852 	 * The chunk size is set equal to bigobj block size so that
3853 	 * dmu_assign_arcbuf() can be tested for object updates.
3854 	 */
3855 
3856 	/*
3857 	 * Read the directory info.  If it's the first time, set things up.
3858 	 */
3859 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3860 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3861 
3862 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3863 		return;
3864 
3865 	bigobj = od[0].od_object;
3866 	packobj = od[1].od_object;
3867 	blocksize = od[0].od_blocksize;
3868 	chunksize = blocksize;
3869 	ASSERT(chunksize == od[1].od_gen);
3870 
3871 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3872 	VERIFY(ISP2(doi.doi_data_block_size));
3873 	VERIFY(chunksize == doi.doi_data_block_size);
3874 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3875 
3876 	/*
3877 	 * Pick a random index and compute the offsets into packobj and bigobj.
3878 	 */
3879 	n = ztest_random(regions) * stride + ztest_random(width);
3880 	s = 1 + ztest_random(width - 1);
3881 
3882 	packoff = n * sizeof (bufwad_t);
3883 	packsize = s * sizeof (bufwad_t);
3884 
3885 	bigoff = n * chunksize;
3886 	bigsize = s * chunksize;
3887 
3888 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3889 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3890 
3891 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3892 
3893 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3894 
3895 	/*
3896 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3897 	 * Iteration 1 test zcopy to already referenced dbufs.
3898 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3899 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3900 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
3901 	 * Iteration 5 test zcopy when it can't be done.
3902 	 * Iteration 6 one more zcopy write.
3903 	 */
3904 	for (i = 0; i < 7; i++) {
3905 		uint64_t j;
3906 		uint64_t off;
3907 
3908 		/*
3909 		 * In iteration 5 (i == 5) use arcbufs
3910 		 * that don't match bigobj blksz to test
3911 		 * dmu_assign_arcbuf() when it can't directly
3912 		 * assign an arcbuf to a dbuf.
3913 		 */
3914 		for (j = 0; j < s; j++) {
3915 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3916 				bigbuf_arcbufs[j] =
3917 				    dmu_request_arcbuf(bonus_db, chunksize);
3918 			} else {
3919 				bigbuf_arcbufs[2 * j] =
3920 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3921 				bigbuf_arcbufs[2 * j + 1] =
3922 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3923 			}
3924 		}
3925 
3926 		/*
3927 		 * Get a tx for the mods to both packobj and bigobj.
3928 		 */
3929 		tx = dmu_tx_create(os);
3930 
3931 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
3932 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3933 
3934 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3935 		if (txg == 0) {
3936 			umem_free(packbuf, packsize);
3937 			umem_free(bigbuf, bigsize);
3938 			for (j = 0; j < s; j++) {
3939 				if (i != 5 ||
3940 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
3941 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
3942 				} else {
3943 					dmu_return_arcbuf(
3944 					    bigbuf_arcbufs[2 * j]);
3945 					dmu_return_arcbuf(
3946 					    bigbuf_arcbufs[2 * j + 1]);
3947 				}
3948 			}
3949 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3950 			dmu_buf_rele(bonus_db, FTAG);
3951 			return;
3952 		}
3953 
3954 		/*
3955 		 * 50% of the time don't read objects in the 1st iteration to
3956 		 * test dmu_assign_arcbuf() for the case when there're no
3957 		 * existing dbufs for the specified offsets.
3958 		 */
3959 		if (i != 0 || ztest_random(2) != 0) {
3960 			error = dmu_read(os, packobj, packoff,
3961 			    packsize, packbuf, DMU_READ_PREFETCH);
3962 			ASSERT0(error);
3963 			error = dmu_read(os, bigobj, bigoff, bigsize,
3964 			    bigbuf, DMU_READ_PREFETCH);
3965 			ASSERT0(error);
3966 		}
3967 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3968 		    n, chunksize, txg);
3969 
3970 		/*
3971 		 * We've verified all the old bufwads, and made new ones.
3972 		 * Now write them out.
3973 		 */
3974 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3975 		if (ztest_opts.zo_verbose >= 7) {
3976 			(void) printf("writing offset %llx size %llx"
3977 			    " txg %llx\n",
3978 			    (u_longlong_t)bigoff,
3979 			    (u_longlong_t)bigsize,
3980 			    (u_longlong_t)txg);
3981 		}
3982 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3983 			dmu_buf_t *dbt;
3984 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3985 				bcopy((caddr_t)bigbuf + (off - bigoff),
3986 				    bigbuf_arcbufs[j]->b_data, chunksize);
3987 			} else {
3988 				bcopy((caddr_t)bigbuf + (off - bigoff),
3989 				    bigbuf_arcbufs[2 * j]->b_data,
3990 				    chunksize / 2);
3991 				bcopy((caddr_t)bigbuf + (off - bigoff) +
3992 				    chunksize / 2,
3993 				    bigbuf_arcbufs[2 * j + 1]->b_data,
3994 				    chunksize / 2);
3995 			}
3996 
3997 			if (i == 1) {
3998 				VERIFY(dmu_buf_hold(os, bigobj, off,
3999 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4000 			}
4001 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4002 				dmu_assign_arcbuf(bonus_db, off,
4003 				    bigbuf_arcbufs[j], tx);
4004 			} else {
4005 				dmu_assign_arcbuf(bonus_db, off,
4006 				    bigbuf_arcbufs[2 * j], tx);
4007 				dmu_assign_arcbuf(bonus_db,
4008 				    off + chunksize / 2,
4009 				    bigbuf_arcbufs[2 * j + 1], tx);
4010 			}
4011 			if (i == 1) {
4012 				dmu_buf_rele(dbt, FTAG);
4013 			}
4014 		}
4015 		dmu_tx_commit(tx);
4016 
4017 		/*
4018 		 * Sanity check the stuff we just wrote.
4019 		 */
4020 		{
4021 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4022 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4023 
4024 			VERIFY(0 == dmu_read(os, packobj, packoff,
4025 			    packsize, packcheck, DMU_READ_PREFETCH));
4026 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
4027 			    bigsize, bigcheck, DMU_READ_PREFETCH));
4028 
4029 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4030 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4031 
4032 			umem_free(packcheck, packsize);
4033 			umem_free(bigcheck, bigsize);
4034 		}
4035 		if (i == 2) {
4036 			txg_wait_open(dmu_objset_pool(os), 0);
4037 		} else if (i == 3) {
4038 			txg_wait_synced(dmu_objset_pool(os), 0);
4039 		}
4040 	}
4041 
4042 	dmu_buf_rele(bonus_db, FTAG);
4043 	umem_free(packbuf, packsize);
4044 	umem_free(bigbuf, bigsize);
4045 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4046 }
4047 
4048 /* ARGSUSED */
4049 void
4050 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4051 {
4052 	ztest_od_t od[1];
4053 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4054 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4055 
4056 	/*
4057 	 * Have multiple threads write to large offsets in an object
4058 	 * to verify that parallel writes to an object -- even to the
4059 	 * same blocks within the object -- doesn't cause any trouble.
4060 	 */
4061 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4062 
4063 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4064 		return;
4065 
4066 	while (ztest_random(10) != 0)
4067 		ztest_io(zd, od[0].od_object, offset);
4068 }
4069 
4070 void
4071 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4072 {
4073 	ztest_od_t od[1];
4074 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4075 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4076 	uint64_t count = ztest_random(20) + 1;
4077 	uint64_t blocksize = ztest_random_blocksize();
4078 	void *data;
4079 
4080 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4081 
4082 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4083 		return;
4084 
4085 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4086 		return;
4087 
4088 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4089 
4090 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4091 
4092 	while (ztest_random(count) != 0) {
4093 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4094 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4095 		    data) != 0)
4096 			break;
4097 		while (ztest_random(4) != 0)
4098 			ztest_io(zd, od[0].od_object, randoff);
4099 	}
4100 
4101 	umem_free(data, blocksize);
4102 }
4103 
4104 /*
4105  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4106  */
4107 #define	ZTEST_ZAP_MIN_INTS	1
4108 #define	ZTEST_ZAP_MAX_INTS	4
4109 #define	ZTEST_ZAP_MAX_PROPS	1000
4110 
4111 void
4112 ztest_zap(ztest_ds_t *zd, uint64_t id)
4113 {
4114 	objset_t *os = zd->zd_os;
4115 	ztest_od_t od[1];
4116 	uint64_t object;
4117 	uint64_t txg, last_txg;
4118 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4119 	uint64_t zl_ints, zl_intsize, prop;
4120 	int i, ints;
4121 	dmu_tx_t *tx;
4122 	char propname[100], txgname[100];
4123 	int error;
4124 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4125 
4126 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4127 
4128 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4129 		return;
4130 
4131 	object = od[0].od_object;
4132 
4133 	/*
4134 	 * Generate a known hash collision, and verify that
4135 	 * we can lookup and remove both entries.
4136 	 */
4137 	tx = dmu_tx_create(os);
4138 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4139 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4140 	if (txg == 0)
4141 		return;
4142 	for (i = 0; i < 2; i++) {
4143 		value[i] = i;
4144 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4145 		    1, &value[i], tx));
4146 	}
4147 	for (i = 0; i < 2; i++) {
4148 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4149 		    sizeof (uint64_t), 1, &value[i], tx));
4150 		VERIFY3U(0, ==,
4151 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4152 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4153 		ASSERT3U(zl_ints, ==, 1);
4154 	}
4155 	for (i = 0; i < 2; i++) {
4156 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4157 	}
4158 	dmu_tx_commit(tx);
4159 
4160 	/*
4161 	 * Generate a buch of random entries.
4162 	 */
4163 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4164 
4165 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4166 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4167 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4168 	bzero(value, sizeof (value));
4169 	last_txg = 0;
4170 
4171 	/*
4172 	 * If these zap entries already exist, validate their contents.
4173 	 */
4174 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4175 	if (error == 0) {
4176 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4177 		ASSERT3U(zl_ints, ==, 1);
4178 
4179 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4180 		    zl_ints, &last_txg) == 0);
4181 
4182 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4183 		    &zl_ints) == 0);
4184 
4185 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4186 		ASSERT3U(zl_ints, ==, ints);
4187 
4188 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4189 		    zl_ints, value) == 0);
4190 
4191 		for (i = 0; i < ints; i++) {
4192 			ASSERT3U(value[i], ==, last_txg + object + i);
4193 		}
4194 	} else {
4195 		ASSERT3U(error, ==, ENOENT);
4196 	}
4197 
4198 	/*
4199 	 * Atomically update two entries in our zap object.
4200 	 * The first is named txg_%llu, and contains the txg
4201 	 * in which the property was last updated.  The second
4202 	 * is named prop_%llu, and the nth element of its value
4203 	 * should be txg + object + n.
4204 	 */
4205 	tx = dmu_tx_create(os);
4206 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4207 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4208 	if (txg == 0)
4209 		return;
4210 
4211 	if (last_txg > txg)
4212 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4213 
4214 	for (i = 0; i < ints; i++)
4215 		value[i] = txg + object + i;
4216 
4217 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4218 	    1, &txg, tx));
4219 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4220 	    ints, value, tx));
4221 
4222 	dmu_tx_commit(tx);
4223 
4224 	/*
4225 	 * Remove a random pair of entries.
4226 	 */
4227 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4228 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4229 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4230 
4231 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4232 
4233 	if (error == ENOENT)
4234 		return;
4235 
4236 	ASSERT0(error);
4237 
4238 	tx = dmu_tx_create(os);
4239 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4240 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4241 	if (txg == 0)
4242 		return;
4243 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4244 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4245 	dmu_tx_commit(tx);
4246 }
4247 
4248 /*
4249  * Testcase to test the upgrading of a microzap to fatzap.
4250  */
4251 void
4252 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4253 {
4254 	objset_t *os = zd->zd_os;
4255 	ztest_od_t od[1];
4256 	uint64_t object, txg;
4257 
4258 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4259 
4260 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4261 		return;
4262 
4263 	object = od[0].od_object;
4264 
4265 	/*
4266 	 * Add entries to this ZAP and make sure it spills over
4267 	 * and gets upgraded to a fatzap. Also, since we are adding
4268 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4269 	 */
4270 	for (int i = 0; i < 2050; i++) {
4271 		char name[ZFS_MAX_DATASET_NAME_LEN];
4272 		uint64_t value = i;
4273 		dmu_tx_t *tx;
4274 		int error;
4275 
4276 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4277 		    id, value);
4278 
4279 		tx = dmu_tx_create(os);
4280 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4281 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4282 		if (txg == 0)
4283 			return;
4284 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4285 		    &value, tx);
4286 		ASSERT(error == 0 || error == EEXIST);
4287 		dmu_tx_commit(tx);
4288 	}
4289 }
4290 
4291 /* ARGSUSED */
4292 void
4293 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4294 {
4295 	objset_t *os = zd->zd_os;
4296 	ztest_od_t od[1];
4297 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4298 	dmu_tx_t *tx;
4299 	int i, namelen, error;
4300 	int micro = ztest_random(2);
4301 	char name[20], string_value[20];
4302 	void *data;
4303 
4304 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4305 
4306 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4307 		return;
4308 
4309 	object = od[0].od_object;
4310 
4311 	/*
4312 	 * Generate a random name of the form 'xxx.....' where each
4313 	 * x is a random printable character and the dots are dots.
4314 	 * There are 94 such characters, and the name length goes from
4315 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4316 	 */
4317 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4318 
4319 	for (i = 0; i < 3; i++)
4320 		name[i] = '!' + ztest_random('~' - '!' + 1);
4321 	for (; i < namelen - 1; i++)
4322 		name[i] = '.';
4323 	name[i] = '\0';
4324 
4325 	if ((namelen & 1) || micro) {
4326 		wsize = sizeof (txg);
4327 		wc = 1;
4328 		data = &txg;
4329 	} else {
4330 		wsize = 1;
4331 		wc = namelen;
4332 		data = string_value;
4333 	}
4334 
4335 	count = -1ULL;
4336 	VERIFY0(zap_count(os, object, &count));
4337 	ASSERT(count != -1ULL);
4338 
4339 	/*
4340 	 * Select an operation: length, lookup, add, update, remove.
4341 	 */
4342 	i = ztest_random(5);
4343 
4344 	if (i >= 2) {
4345 		tx = dmu_tx_create(os);
4346 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4347 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4348 		if (txg == 0)
4349 			return;
4350 		bcopy(name, string_value, namelen);
4351 	} else {
4352 		tx = NULL;
4353 		txg = 0;
4354 		bzero(string_value, namelen);
4355 	}
4356 
4357 	switch (i) {
4358 
4359 	case 0:
4360 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4361 		if (error == 0) {
4362 			ASSERT3U(wsize, ==, zl_wsize);
4363 			ASSERT3U(wc, ==, zl_wc);
4364 		} else {
4365 			ASSERT3U(error, ==, ENOENT);
4366 		}
4367 		break;
4368 
4369 	case 1:
4370 		error = zap_lookup(os, object, name, wsize, wc, data);
4371 		if (error == 0) {
4372 			if (data == string_value &&
4373 			    bcmp(name, data, namelen) != 0)
4374 				fatal(0, "name '%s' != val '%s' len %d",
4375 				    name, data, namelen);
4376 		} else {
4377 			ASSERT3U(error, ==, ENOENT);
4378 		}
4379 		break;
4380 
4381 	case 2:
4382 		error = zap_add(os, object, name, wsize, wc, data, tx);
4383 		ASSERT(error == 0 || error == EEXIST);
4384 		break;
4385 
4386 	case 3:
4387 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4388 		break;
4389 
4390 	case 4:
4391 		error = zap_remove(os, object, name, tx);
4392 		ASSERT(error == 0 || error == ENOENT);
4393 		break;
4394 	}
4395 
4396 	if (tx != NULL)
4397 		dmu_tx_commit(tx);
4398 }
4399 
4400 /*
4401  * Commit callback data.
4402  */
4403 typedef struct ztest_cb_data {
4404 	list_node_t		zcd_node;
4405 	uint64_t		zcd_txg;
4406 	int			zcd_expected_err;
4407 	boolean_t		zcd_added;
4408 	boolean_t		zcd_called;
4409 	spa_t			*zcd_spa;
4410 } ztest_cb_data_t;
4411 
4412 /* This is the actual commit callback function */
4413 static void
4414 ztest_commit_callback(void *arg, int error)
4415 {
4416 	ztest_cb_data_t *data = arg;
4417 	uint64_t synced_txg;
4418 
4419 	VERIFY(data != NULL);
4420 	VERIFY3S(data->zcd_expected_err, ==, error);
4421 	VERIFY(!data->zcd_called);
4422 
4423 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4424 	if (data->zcd_txg > synced_txg)
4425 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4426 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4427 		    synced_txg);
4428 
4429 	data->zcd_called = B_TRUE;
4430 
4431 	if (error == ECANCELED) {
4432 		ASSERT0(data->zcd_txg);
4433 		ASSERT(!data->zcd_added);
4434 
4435 		/*
4436 		 * The private callback data should be destroyed here, but
4437 		 * since we are going to check the zcd_called field after
4438 		 * dmu_tx_abort(), we will destroy it there.
4439 		 */
4440 		return;
4441 	}
4442 
4443 	/* Was this callback added to the global callback list? */
4444 	if (!data->zcd_added)
4445 		goto out;
4446 
4447 	ASSERT3U(data->zcd_txg, !=, 0);
4448 
4449 	/* Remove our callback from the list */
4450 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4451 	list_remove(&zcl.zcl_callbacks, data);
4452 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4453 
4454 out:
4455 	umem_free(data, sizeof (ztest_cb_data_t));
4456 }
4457 
4458 /* Allocate and initialize callback data structure */
4459 static ztest_cb_data_t *
4460 ztest_create_cb_data(objset_t *os, uint64_t txg)
4461 {
4462 	ztest_cb_data_t *cb_data;
4463 
4464 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4465 
4466 	cb_data->zcd_txg = txg;
4467 	cb_data->zcd_spa = dmu_objset_spa(os);
4468 
4469 	return (cb_data);
4470 }
4471 
4472 /*
4473  * If a number of txgs equal to this threshold have been created after a commit
4474  * callback has been registered but not called, then we assume there is an
4475  * implementation bug.
4476  */
4477 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4478 
4479 /*
4480  * Commit callback test.
4481  */
4482 void
4483 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4484 {
4485 	objset_t *os = zd->zd_os;
4486 	ztest_od_t od[1];
4487 	dmu_tx_t *tx;
4488 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4489 	uint64_t old_txg, txg;
4490 	int i, error;
4491 
4492 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4493 
4494 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4495 		return;
4496 
4497 	tx = dmu_tx_create(os);
4498 
4499 	cb_data[0] = ztest_create_cb_data(os, 0);
4500 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4501 
4502 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4503 
4504 	/* Every once in a while, abort the transaction on purpose */
4505 	if (ztest_random(100) == 0)
4506 		error = -1;
4507 
4508 	if (!error)
4509 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4510 
4511 	txg = error ? 0 : dmu_tx_get_txg(tx);
4512 
4513 	cb_data[0]->zcd_txg = txg;
4514 	cb_data[1] = ztest_create_cb_data(os, txg);
4515 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4516 
4517 	if (error) {
4518 		/*
4519 		 * It's not a strict requirement to call the registered
4520 		 * callbacks from inside dmu_tx_abort(), but that's what
4521 		 * it's supposed to happen in the current implementation
4522 		 * so we will check for that.
4523 		 */
4524 		for (i = 0; i < 2; i++) {
4525 			cb_data[i]->zcd_expected_err = ECANCELED;
4526 			VERIFY(!cb_data[i]->zcd_called);
4527 		}
4528 
4529 		dmu_tx_abort(tx);
4530 
4531 		for (i = 0; i < 2; i++) {
4532 			VERIFY(cb_data[i]->zcd_called);
4533 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4534 		}
4535 
4536 		return;
4537 	}
4538 
4539 	cb_data[2] = ztest_create_cb_data(os, txg);
4540 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4541 
4542 	/*
4543 	 * Read existing data to make sure there isn't a future leak.
4544 	 */
4545 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4546 	    &old_txg, DMU_READ_PREFETCH));
4547 
4548 	if (old_txg > txg)
4549 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4550 		    old_txg, txg);
4551 
4552 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4553 
4554 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4555 
4556 	/*
4557 	 * Since commit callbacks don't have any ordering requirement and since
4558 	 * it is theoretically possible for a commit callback to be called
4559 	 * after an arbitrary amount of time has elapsed since its txg has been
4560 	 * synced, it is difficult to reliably determine whether a commit
4561 	 * callback hasn't been called due to high load or due to a flawed
4562 	 * implementation.
4563 	 *
4564 	 * In practice, we will assume that if after a certain number of txgs a
4565 	 * commit callback hasn't been called, then most likely there's an
4566 	 * implementation bug..
4567 	 */
4568 	tmp_cb = list_head(&zcl.zcl_callbacks);
4569 	if (tmp_cb != NULL &&
4570 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4571 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4572 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4573 	}
4574 
4575 	/*
4576 	 * Let's find the place to insert our callbacks.
4577 	 *
4578 	 * Even though the list is ordered by txg, it is possible for the
4579 	 * insertion point to not be the end because our txg may already be
4580 	 * quiescing at this point and other callbacks in the open txg
4581 	 * (from other objsets) may have sneaked in.
4582 	 */
4583 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4584 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4585 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4586 
4587 	/* Add the 3 callbacks to the list */
4588 	for (i = 0; i < 3; i++) {
4589 		if (tmp_cb == NULL)
4590 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4591 		else
4592 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4593 			    cb_data[i]);
4594 
4595 		cb_data[i]->zcd_added = B_TRUE;
4596 		VERIFY(!cb_data[i]->zcd_called);
4597 
4598 		tmp_cb = cb_data[i];
4599 	}
4600 
4601 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4602 
4603 	dmu_tx_commit(tx);
4604 }
4605 
4606 /* ARGSUSED */
4607 void
4608 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4609 {
4610 	zfs_prop_t proplist[] = {
4611 		ZFS_PROP_CHECKSUM,
4612 		ZFS_PROP_COMPRESSION,
4613 		ZFS_PROP_COPIES,
4614 		ZFS_PROP_DEDUP
4615 	};
4616 
4617 	(void) rw_rdlock(&ztest_name_lock);
4618 
4619 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4620 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4621 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4622 
4623 	(void) rw_unlock(&ztest_name_lock);
4624 }
4625 
4626 /* ARGSUSED */
4627 void
4628 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4629 {
4630 	nvlist_t *props = NULL;
4631 
4632 	(void) rw_rdlock(&ztest_name_lock);
4633 
4634 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4635 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4636 
4637 	VERIFY0(spa_prop_get(ztest_spa, &props));
4638 
4639 	if (ztest_opts.zo_verbose >= 6)
4640 		dump_nvlist(props, 4);
4641 
4642 	nvlist_free(props);
4643 
4644 	(void) rw_unlock(&ztest_name_lock);
4645 }
4646 
4647 static int
4648 user_release_one(const char *snapname, const char *holdname)
4649 {
4650 	nvlist_t *snaps, *holds;
4651 	int error;
4652 
4653 	snaps = fnvlist_alloc();
4654 	holds = fnvlist_alloc();
4655 	fnvlist_add_boolean(holds, holdname);
4656 	fnvlist_add_nvlist(snaps, snapname, holds);
4657 	fnvlist_free(holds);
4658 	error = dsl_dataset_user_release(snaps, NULL);
4659 	fnvlist_free(snaps);
4660 	return (error);
4661 }
4662 
4663 /*
4664  * Test snapshot hold/release and deferred destroy.
4665  */
4666 void
4667 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4668 {
4669 	int error;
4670 	objset_t *os = zd->zd_os;
4671 	objset_t *origin;
4672 	char snapname[100];
4673 	char fullname[100];
4674 	char clonename[100];
4675 	char tag[100];
4676 	char osname[ZFS_MAX_DATASET_NAME_LEN];
4677 	nvlist_t *holds;
4678 
4679 	(void) rw_rdlock(&ztest_name_lock);
4680 
4681 	dmu_objset_name(os, osname);
4682 
4683 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4684 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4685 	(void) snprintf(clonename, sizeof (clonename),
4686 	    "%s/ch1_%llu", osname, id);
4687 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4688 
4689 	/*
4690 	 * Clean up from any previous run.
4691 	 */
4692 	error = dsl_destroy_head(clonename);
4693 	if (error != ENOENT)
4694 		ASSERT0(error);
4695 	error = user_release_one(fullname, tag);
4696 	if (error != ESRCH && error != ENOENT)
4697 		ASSERT0(error);
4698 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4699 	if (error != ENOENT)
4700 		ASSERT0(error);
4701 
4702 	/*
4703 	 * Create snapshot, clone it, mark snap for deferred destroy,
4704 	 * destroy clone, verify snap was also destroyed.
4705 	 */
4706 	error = dmu_objset_snapshot_one(osname, snapname);
4707 	if (error) {
4708 		if (error == ENOSPC) {
4709 			ztest_record_enospc("dmu_objset_snapshot");
4710 			goto out;
4711 		}
4712 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4713 	}
4714 
4715 	error = dmu_objset_clone(clonename, fullname);
4716 	if (error) {
4717 		if (error == ENOSPC) {
4718 			ztest_record_enospc("dmu_objset_clone");
4719 			goto out;
4720 		}
4721 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4722 	}
4723 
4724 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4725 	if (error) {
4726 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4727 		    fullname, error);
4728 	}
4729 
4730 	error = dsl_destroy_head(clonename);
4731 	if (error)
4732 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4733 
4734 	error = dmu_objset_hold(fullname, FTAG, &origin);
4735 	if (error != ENOENT)
4736 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4737 
4738 	/*
4739 	 * Create snapshot, add temporary hold, verify that we can't
4740 	 * destroy a held snapshot, mark for deferred destroy,
4741 	 * release hold, verify snapshot was destroyed.
4742 	 */
4743 	error = dmu_objset_snapshot_one(osname, snapname);
4744 	if (error) {
4745 		if (error == ENOSPC) {
4746 			ztest_record_enospc("dmu_objset_snapshot");
4747 			goto out;
4748 		}
4749 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4750 	}
4751 
4752 	holds = fnvlist_alloc();
4753 	fnvlist_add_string(holds, fullname, tag);
4754 	error = dsl_dataset_user_hold(holds, 0, NULL);
4755 	fnvlist_free(holds);
4756 
4757 	if (error == ENOSPC) {
4758 		ztest_record_enospc("dsl_dataset_user_hold");
4759 		goto out;
4760 	} else if (error) {
4761 		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4762 		    fullname, tag, error);
4763 	}
4764 
4765 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4766 	if (error != EBUSY) {
4767 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4768 		    fullname, error);
4769 	}
4770 
4771 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4772 	if (error) {
4773 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4774 		    fullname, error);
4775 	}
4776 
4777 	error = user_release_one(fullname, tag);
4778 	if (error)
4779 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4780 
4781 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4782 
4783 out:
4784 	(void) rw_unlock(&ztest_name_lock);
4785 }
4786 
4787 /*
4788  * Inject random faults into the on-disk data.
4789  */
4790 /* ARGSUSED */
4791 void
4792 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4793 {
4794 	ztest_shared_t *zs = ztest_shared;
4795 	spa_t *spa = ztest_spa;
4796 	int fd;
4797 	uint64_t offset;
4798 	uint64_t leaves;
4799 	uint64_t bad = 0x1990c0ffeedecade;
4800 	uint64_t top, leaf;
4801 	char path0[MAXPATHLEN];
4802 	char pathrand[MAXPATHLEN];
4803 	size_t fsize;
4804 	int bshift = SPA_MAXBLOCKSHIFT + 2;
4805 	int iters = 1000;
4806 	int maxfaults;
4807 	int mirror_save;
4808 	vdev_t *vd0 = NULL;
4809 	uint64_t guid0 = 0;
4810 	boolean_t islog = B_FALSE;
4811 
4812 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4813 	maxfaults = MAXFAULTS();
4814 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4815 	mirror_save = zs->zs_mirrors;
4816 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4817 
4818 	ASSERT(leaves >= 1);
4819 
4820 	/*
4821 	 * Grab the name lock as reader. There are some operations
4822 	 * which don't like to have their vdevs changed while
4823 	 * they are in progress (i.e. spa_change_guid). Those
4824 	 * operations will have grabbed the name lock as writer.
4825 	 */
4826 	(void) rw_rdlock(&ztest_name_lock);
4827 
4828 	/*
4829 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4830 	 */
4831 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4832 
4833 	if (ztest_random(2) == 0) {
4834 		/*
4835 		 * Inject errors on a normal data device or slog device.
4836 		 */
4837 		top = ztest_random_vdev_top(spa, B_TRUE);
4838 		leaf = ztest_random(leaves) + zs->zs_splits;
4839 
4840 		/*
4841 		 * Generate paths to the first leaf in this top-level vdev,
4842 		 * and to the random leaf we selected.  We'll induce transient
4843 		 * write failures and random online/offline activity on leaf 0,
4844 		 * and we'll write random garbage to the randomly chosen leaf.
4845 		 */
4846 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
4847 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4848 		    top * leaves + zs->zs_splits);
4849 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4850 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4851 		    top * leaves + leaf);
4852 
4853 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4854 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4855 			islog = B_TRUE;
4856 
4857 		/*
4858 		 * If the top-level vdev needs to be resilvered
4859 		 * then we only allow faults on the device that is
4860 		 * resilvering.
4861 		 */
4862 		if (vd0 != NULL && maxfaults != 1 &&
4863 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4864 		    vd0->vdev_resilver_txg != 0)) {
4865 			/*
4866 			 * Make vd0 explicitly claim to be unreadable,
4867 			 * or unwriteable, or reach behind its back
4868 			 * and close the underlying fd.  We can do this if
4869 			 * maxfaults == 0 because we'll fail and reexecute,
4870 			 * and we can do it if maxfaults >= 2 because we'll
4871 			 * have enough redundancy.  If maxfaults == 1, the
4872 			 * combination of this with injection of random data
4873 			 * corruption below exceeds the pool's fault tolerance.
4874 			 */
4875 			vdev_file_t *vf = vd0->vdev_tsd;
4876 
4877 			if (vf != NULL && ztest_random(3) == 0) {
4878 				(void) close(vf->vf_vnode->v_fd);
4879 				vf->vf_vnode->v_fd = -1;
4880 			} else if (ztest_random(2) == 0) {
4881 				vd0->vdev_cant_read = B_TRUE;
4882 			} else {
4883 				vd0->vdev_cant_write = B_TRUE;
4884 			}
4885 			guid0 = vd0->vdev_guid;
4886 		}
4887 	} else {
4888 		/*
4889 		 * Inject errors on an l2cache device.
4890 		 */
4891 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
4892 
4893 		if (sav->sav_count == 0) {
4894 			spa_config_exit(spa, SCL_STATE, FTAG);
4895 			(void) rw_unlock(&ztest_name_lock);
4896 			return;
4897 		}
4898 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4899 		guid0 = vd0->vdev_guid;
4900 		(void) strcpy(path0, vd0->vdev_path);
4901 		(void) strcpy(pathrand, vd0->vdev_path);
4902 
4903 		leaf = 0;
4904 		leaves = 1;
4905 		maxfaults = INT_MAX;	/* no limit on cache devices */
4906 	}
4907 
4908 	spa_config_exit(spa, SCL_STATE, FTAG);
4909 	(void) rw_unlock(&ztest_name_lock);
4910 
4911 	/*
4912 	 * If we can tolerate two or more faults, or we're dealing
4913 	 * with a slog, randomly online/offline vd0.
4914 	 */
4915 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
4916 		if (ztest_random(10) < 6) {
4917 			int flags = (ztest_random(2) == 0 ?
4918 			    ZFS_OFFLINE_TEMPORARY : 0);
4919 
4920 			/*
4921 			 * We have to grab the zs_name_lock as writer to
4922 			 * prevent a race between offlining a slog and
4923 			 * destroying a dataset. Offlining the slog will
4924 			 * grab a reference on the dataset which may cause
4925 			 * dmu_objset_destroy() to fail with EBUSY thus
4926 			 * leaving the dataset in an inconsistent state.
4927 			 */
4928 			if (islog)
4929 				(void) rw_wrlock(&ztest_name_lock);
4930 
4931 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4932 
4933 			if (islog)
4934 				(void) rw_unlock(&ztest_name_lock);
4935 		} else {
4936 			/*
4937 			 * Ideally we would like to be able to randomly
4938 			 * call vdev_[on|off]line without holding locks
4939 			 * to force unpredictable failures but the side
4940 			 * effects of vdev_[on|off]line prevent us from
4941 			 * doing so. We grab the ztest_vdev_lock here to
4942 			 * prevent a race between injection testing and
4943 			 * aux_vdev removal.
4944 			 */
4945 			VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4946 			(void) vdev_online(spa, guid0, 0, NULL);
4947 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4948 		}
4949 	}
4950 
4951 	if (maxfaults == 0)
4952 		return;
4953 
4954 	/*
4955 	 * We have at least single-fault tolerance, so inject data corruption.
4956 	 */
4957 	fd = open(pathrand, O_RDWR);
4958 
4959 	if (fd == -1)	/* we hit a gap in the device namespace */
4960 		return;
4961 
4962 	fsize = lseek(fd, 0, SEEK_END);
4963 
4964 	while (--iters != 0) {
4965 		/*
4966 		 * The offset must be chosen carefully to ensure that
4967 		 * we do not inject a given logical block with errors
4968 		 * on two different leaf devices, because ZFS can not
4969 		 * tolerate that (if maxfaults==1).
4970 		 *
4971 		 * We divide each leaf into chunks of size
4972 		 * (# leaves * SPA_MAXBLOCKSIZE * 4).  Within each chunk
4973 		 * there is a series of ranges to which we can inject errors.
4974 		 * Each range can accept errors on only a single leaf vdev.
4975 		 * The error injection ranges are separated by ranges
4976 		 * which we will not inject errors on any device (DMZs).
4977 		 * Each DMZ must be large enough such that a single block
4978 		 * can not straddle it, so that a single block can not be
4979 		 * a target in two different injection ranges (on different
4980 		 * leaf vdevs).
4981 		 *
4982 		 * For example, with 3 leaves, each chunk looks like:
4983 		 *    0 to  32M: injection range for leaf 0
4984 		 *  32M to  64M: DMZ - no injection allowed
4985 		 *  64M to  96M: injection range for leaf 1
4986 		 *  96M to 128M: DMZ - no injection allowed
4987 		 * 128M to 160M: injection range for leaf 2
4988 		 * 160M to 192M: DMZ - no injection allowed
4989 		 */
4990 		offset = ztest_random(fsize / (leaves << bshift)) *
4991 		    (leaves << bshift) + (leaf << bshift) +
4992 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4993 
4994 		/*
4995 		 * Only allow damage to the labels at one end of the vdev.
4996 		 *
4997 		 * If all labels are damaged, the device will be totally
4998 		 * inaccessible, which will result in loss of data,
4999 		 * because we also damage (parts of) the other side of
5000 		 * the mirror/raidz.
5001 		 *
5002 		 * Additionally, we will always have both an even and an
5003 		 * odd label, so that we can handle crashes in the
5004 		 * middle of vdev_config_sync().
5005 		 */
5006 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5007 			continue;
5008 
5009 		/*
5010 		 * The two end labels are stored at the "end" of the disk, but
5011 		 * the end of the disk (vdev_psize) is aligned to
5012 		 * sizeof (vdev_label_t).
5013 		 */
5014 		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5015 		if ((leaf & 1) == 1 &&
5016 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5017 			continue;
5018 
5019 		VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
5020 		if (mirror_save != zs->zs_mirrors) {
5021 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5022 			(void) close(fd);
5023 			return;
5024 		}
5025 
5026 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5027 			fatal(1, "can't inject bad word at 0x%llx in %s",
5028 			    offset, pathrand);
5029 
5030 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5031 
5032 		if (ztest_opts.zo_verbose >= 7)
5033 			(void) printf("injected bad word into %s,"
5034 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5035 	}
5036 
5037 	(void) close(fd);
5038 }
5039 
5040 /*
5041  * Verify that DDT repair works as expected.
5042  */
5043 void
5044 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5045 {
5046 	ztest_shared_t *zs = ztest_shared;
5047 	spa_t *spa = ztest_spa;
5048 	objset_t *os = zd->zd_os;
5049 	ztest_od_t od[1];
5050 	uint64_t object, blocksize, txg, pattern, psize;
5051 	enum zio_checksum checksum = spa_dedup_checksum(spa);
5052 	dmu_buf_t *db;
5053 	dmu_tx_t *tx;
5054 	void *buf;
5055 	blkptr_t blk;
5056 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
5057 
5058 	blocksize = ztest_random_blocksize();
5059 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
5060 
5061 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5062 
5063 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5064 		return;
5065 
5066 	/*
5067 	 * Take the name lock as writer to prevent anyone else from changing
5068 	 * the pool and dataset properies we need to maintain during this test.
5069 	 */
5070 	(void) rw_wrlock(&ztest_name_lock);
5071 
5072 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5073 	    B_FALSE) != 0 ||
5074 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5075 	    B_FALSE) != 0) {
5076 		(void) rw_unlock(&ztest_name_lock);
5077 		return;
5078 	}
5079 
5080 	dmu_objset_stats_t dds;
5081 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5082 	dmu_objset_fast_stat(os, &dds);
5083 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5084 
5085 	object = od[0].od_object;
5086 	blocksize = od[0].od_blocksize;
5087 	pattern = zs->zs_guid ^ dds.dds_guid;
5088 
5089 	ASSERT(object != 0);
5090 
5091 	tx = dmu_tx_create(os);
5092 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5093 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5094 	if (txg == 0) {
5095 		(void) rw_unlock(&ztest_name_lock);
5096 		return;
5097 	}
5098 
5099 	/*
5100 	 * Write all the copies of our block.
5101 	 */
5102 	for (int i = 0; i < copies; i++) {
5103 		uint64_t offset = i * blocksize;
5104 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5105 		    DMU_READ_NO_PREFETCH);
5106 		if (error != 0) {
5107 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5108 			    os, (long long)object, (long long) offset, error);
5109 		}
5110 		ASSERT(db->db_offset == offset);
5111 		ASSERT(db->db_size == blocksize);
5112 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5113 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5114 		dmu_buf_will_fill(db, tx);
5115 		ztest_pattern_set(db->db_data, db->db_size, pattern);
5116 		dmu_buf_rele(db, FTAG);
5117 	}
5118 
5119 	dmu_tx_commit(tx);
5120 	txg_wait_synced(spa_get_dsl(spa), txg);
5121 
5122 	/*
5123 	 * Find out what block we got.
5124 	 */
5125 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5126 	    DMU_READ_NO_PREFETCH));
5127 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5128 	dmu_buf_rele(db, FTAG);
5129 
5130 	/*
5131 	 * Damage the block.  Dedup-ditto will save us when we read it later.
5132 	 */
5133 	psize = BP_GET_PSIZE(&blk);
5134 	buf = zio_buf_alloc(psize);
5135 	ztest_pattern_set(buf, psize, ~pattern);
5136 
5137 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5138 	    buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5139 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5140 
5141 	zio_buf_free(buf, psize);
5142 
5143 	(void) rw_unlock(&ztest_name_lock);
5144 }
5145 
5146 /*
5147  * Scrub the pool.
5148  */
5149 /* ARGSUSED */
5150 void
5151 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5152 {
5153 	spa_t *spa = ztest_spa;
5154 
5155 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5156 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5157 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5158 }
5159 
5160 /*
5161  * Change the guid for the pool.
5162  */
5163 /* ARGSUSED */
5164 void
5165 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5166 {
5167 	spa_t *spa = ztest_spa;
5168 	uint64_t orig, load;
5169 	int error;
5170 
5171 	orig = spa_guid(spa);
5172 	load = spa_load_guid(spa);
5173 
5174 	(void) rw_wrlock(&ztest_name_lock);
5175 	error = spa_change_guid(spa);
5176 	(void) rw_unlock(&ztest_name_lock);
5177 
5178 	if (error != 0)
5179 		return;
5180 
5181 	if (ztest_opts.zo_verbose >= 4) {
5182 		(void) printf("Changed guid old %llu -> %llu\n",
5183 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5184 	}
5185 
5186 	VERIFY3U(orig, !=, spa_guid(spa));
5187 	VERIFY3U(load, ==, spa_load_guid(spa));
5188 }
5189 
5190 /*
5191  * Rename the pool to a different name and then rename it back.
5192  */
5193 /* ARGSUSED */
5194 void
5195 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5196 {
5197 	char *oldname, *newname;
5198 	spa_t *spa;
5199 
5200 	(void) rw_wrlock(&ztest_name_lock);
5201 
5202 	oldname = ztest_opts.zo_pool;
5203 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5204 	(void) strcpy(newname, oldname);
5205 	(void) strcat(newname, "_tmp");
5206 
5207 	/*
5208 	 * Do the rename
5209 	 */
5210 	VERIFY3U(0, ==, spa_rename(oldname, newname));
5211 
5212 	/*
5213 	 * Try to open it under the old name, which shouldn't exist
5214 	 */
5215 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5216 
5217 	/*
5218 	 * Open it under the new name and make sure it's still the same spa_t.
5219 	 */
5220 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5221 
5222 	ASSERT(spa == ztest_spa);
5223 	spa_close(spa, FTAG);
5224 
5225 	/*
5226 	 * Rename it back to the original
5227 	 */
5228 	VERIFY3U(0, ==, spa_rename(newname, oldname));
5229 
5230 	/*
5231 	 * Make sure it can still be opened
5232 	 */
5233 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5234 
5235 	ASSERT(spa == ztest_spa);
5236 	spa_close(spa, FTAG);
5237 
5238 	umem_free(newname, strlen(newname) + 1);
5239 
5240 	(void) rw_unlock(&ztest_name_lock);
5241 }
5242 
5243 /*
5244  * Verify pool integrity by running zdb.
5245  */
5246 static void
5247 ztest_run_zdb(char *pool)
5248 {
5249 	int status;
5250 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5251 	char zbuf[1024];
5252 	char *bin;
5253 	char *ztest;
5254 	char *isa;
5255 	int isalen;
5256 	FILE *fp;
5257 
5258 	(void) realpath(getexecname(), zdb);
5259 
5260 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5261 	bin = strstr(zdb, "/usr/bin/");
5262 	ztest = strstr(bin, "/ztest");
5263 	isa = bin + 8;
5264 	isalen = ztest - isa;
5265 	isa = strdup(isa);
5266 	/* LINTED */
5267 	(void) sprintf(bin,
5268 	    "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s %s",
5269 	    isalen,
5270 	    isa,
5271 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5272 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5273 	    spa_config_path,
5274 	    pool);
5275 	free(isa);
5276 
5277 	if (ztest_opts.zo_verbose >= 5)
5278 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5279 
5280 	fp = popen(zdb, "r");
5281 
5282 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5283 		if (ztest_opts.zo_verbose >= 3)
5284 			(void) printf("%s", zbuf);
5285 
5286 	status = pclose(fp);
5287 
5288 	if (status == 0)
5289 		return;
5290 
5291 	ztest_dump_core = 0;
5292 	if (WIFEXITED(status))
5293 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5294 	else
5295 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5296 }
5297 
5298 static void
5299 ztest_walk_pool_directory(char *header)
5300 {
5301 	spa_t *spa = NULL;
5302 
5303 	if (ztest_opts.zo_verbose >= 6)
5304 		(void) printf("%s\n", header);
5305 
5306 	mutex_enter(&spa_namespace_lock);
5307 	while ((spa = spa_next(spa)) != NULL)
5308 		if (ztest_opts.zo_verbose >= 6)
5309 			(void) printf("\t%s\n", spa_name(spa));
5310 	mutex_exit(&spa_namespace_lock);
5311 }
5312 
5313 static void
5314 ztest_spa_import_export(char *oldname, char *newname)
5315 {
5316 	nvlist_t *config, *newconfig;
5317 	uint64_t pool_guid;
5318 	spa_t *spa;
5319 	int error;
5320 
5321 	if (ztest_opts.zo_verbose >= 4) {
5322 		(void) printf("import/export: old = %s, new = %s\n",
5323 		    oldname, newname);
5324 	}
5325 
5326 	/*
5327 	 * Clean up from previous runs.
5328 	 */
5329 	(void) spa_destroy(newname);
5330 
5331 	/*
5332 	 * Get the pool's configuration and guid.
5333 	 */
5334 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5335 
5336 	/*
5337 	 * Kick off a scrub to tickle scrub/export races.
5338 	 */
5339 	if (ztest_random(2) == 0)
5340 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5341 
5342 	pool_guid = spa_guid(spa);
5343 	spa_close(spa, FTAG);
5344 
5345 	ztest_walk_pool_directory("pools before export");
5346 
5347 	/*
5348 	 * Export it.
5349 	 */
5350 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5351 
5352 	ztest_walk_pool_directory("pools after export");
5353 
5354 	/*
5355 	 * Try to import it.
5356 	 */
5357 	newconfig = spa_tryimport(config);
5358 	ASSERT(newconfig != NULL);
5359 	nvlist_free(newconfig);
5360 
5361 	/*
5362 	 * Import it under the new name.
5363 	 */
5364 	error = spa_import(newname, config, NULL, 0);
5365 	if (error != 0) {
5366 		dump_nvlist(config, 0);
5367 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5368 		    oldname, newname, error);
5369 	}
5370 
5371 	ztest_walk_pool_directory("pools after import");
5372 
5373 	/*
5374 	 * Try to import it again -- should fail with EEXIST.
5375 	 */
5376 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5377 
5378 	/*
5379 	 * Try to import it under a different name -- should fail with EEXIST.
5380 	 */
5381 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5382 
5383 	/*
5384 	 * Verify that the pool is no longer visible under the old name.
5385 	 */
5386 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5387 
5388 	/*
5389 	 * Verify that we can open and close the pool using the new name.
5390 	 */
5391 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5392 	ASSERT(pool_guid == spa_guid(spa));
5393 	spa_close(spa, FTAG);
5394 
5395 	nvlist_free(config);
5396 }
5397 
5398 static void
5399 ztest_resume(spa_t *spa)
5400 {
5401 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5402 		(void) printf("resuming from suspended state\n");
5403 	spa_vdev_state_enter(spa, SCL_NONE);
5404 	vdev_clear(spa, NULL);
5405 	(void) spa_vdev_state_exit(spa, NULL, 0);
5406 	(void) zio_resume(spa);
5407 }
5408 
5409 static void *
5410 ztest_resume_thread(void *arg)
5411 {
5412 	spa_t *spa = arg;
5413 
5414 	while (!ztest_exiting) {
5415 		if (spa_suspended(spa))
5416 			ztest_resume(spa);
5417 		(void) poll(NULL, 0, 100);
5418 
5419 		/*
5420 		 * Periodically change the zfs_compressed_arc_enabled setting.
5421 		 */
5422 		if (ztest_random(10) == 0)
5423 			zfs_compressed_arc_enabled = ztest_random(2);
5424 	}
5425 	return (NULL);
5426 }
5427 
5428 static void *
5429 ztest_deadman_thread(void *arg)
5430 {
5431 	ztest_shared_t *zs = arg;
5432 	spa_t *spa = ztest_spa;
5433 	hrtime_t delta, total = 0;
5434 
5435 	for (;;) {
5436 		delta = zs->zs_thread_stop - zs->zs_thread_start +
5437 		    MSEC2NSEC(zfs_deadman_synctime_ms);
5438 
5439 		(void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5440 
5441 		/*
5442 		 * If the pool is suspended then fail immediately. Otherwise,
5443 		 * check to see if the pool is making any progress. If
5444 		 * vdev_deadman() discovers that there hasn't been any recent
5445 		 * I/Os then it will end up aborting the tests.
5446 		 */
5447 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5448 			fatal(0, "aborting test after %llu seconds because "
5449 			    "pool has transitioned to a suspended state.",
5450 			    zfs_deadman_synctime_ms / 1000);
5451 			return (NULL);
5452 		}
5453 		vdev_deadman(spa->spa_root_vdev);
5454 
5455 		total += zfs_deadman_synctime_ms/1000;
5456 		(void) printf("ztest has been running for %lld seconds\n",
5457 		    total);
5458 	}
5459 }
5460 
5461 static void
5462 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5463 {
5464 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5465 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5466 	hrtime_t functime = gethrtime();
5467 
5468 	for (int i = 0; i < zi->zi_iters; i++)
5469 		zi->zi_func(zd, id);
5470 
5471 	functime = gethrtime() - functime;
5472 
5473 	atomic_add_64(&zc->zc_count, 1);
5474 	atomic_add_64(&zc->zc_time, functime);
5475 
5476 	if (ztest_opts.zo_verbose >= 4) {
5477 		Dl_info dli;
5478 		(void) dladdr((void *)zi->zi_func, &dli);
5479 		(void) printf("%6.2f sec in %s\n",
5480 		    (double)functime / NANOSEC, dli.dli_sname);
5481 	}
5482 }
5483 
5484 static void *
5485 ztest_thread(void *arg)
5486 {
5487 	int rand;
5488 	uint64_t id = (uintptr_t)arg;
5489 	ztest_shared_t *zs = ztest_shared;
5490 	uint64_t call_next;
5491 	hrtime_t now;
5492 	ztest_info_t *zi;
5493 	ztest_shared_callstate_t *zc;
5494 
5495 	while ((now = gethrtime()) < zs->zs_thread_stop) {
5496 		/*
5497 		 * See if it's time to force a crash.
5498 		 */
5499 		if (now > zs->zs_thread_kill)
5500 			ztest_kill(zs);
5501 
5502 		/*
5503 		 * If we're getting ENOSPC with some regularity, stop.
5504 		 */
5505 		if (zs->zs_enospc_count > 10)
5506 			break;
5507 
5508 		/*
5509 		 * Pick a random function to execute.
5510 		 */
5511 		rand = ztest_random(ZTEST_FUNCS);
5512 		zi = &ztest_info[rand];
5513 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5514 		call_next = zc->zc_next;
5515 
5516 		if (now >= call_next &&
5517 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
5518 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5519 			ztest_execute(rand, zi, id);
5520 		}
5521 	}
5522 
5523 	return (NULL);
5524 }
5525 
5526 static void
5527 ztest_dataset_name(char *dsname, char *pool, int d)
5528 {
5529 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5530 }
5531 
5532 static void
5533 ztest_dataset_destroy(int d)
5534 {
5535 	char name[ZFS_MAX_DATASET_NAME_LEN];
5536 
5537 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5538 
5539 	if (ztest_opts.zo_verbose >= 3)
5540 		(void) printf("Destroying %s to free up space\n", name);
5541 
5542 	/*
5543 	 * Cleanup any non-standard clones and snapshots.  In general,
5544 	 * ztest thread t operates on dataset (t % zopt_datasets),
5545 	 * so there may be more than one thing to clean up.
5546 	 */
5547 	for (int t = d; t < ztest_opts.zo_threads;
5548 	    t += ztest_opts.zo_datasets) {
5549 		ztest_dsl_dataset_cleanup(name, t);
5550 	}
5551 
5552 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5553 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5554 }
5555 
5556 static void
5557 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5558 {
5559 	uint64_t usedobjs, dirobjs, scratch;
5560 
5561 	/*
5562 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
5563 	 * Therefore, the number of objects in use should equal the
5564 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5565 	 * If not, we have an object leak.
5566 	 *
5567 	 * Note that we can only check this in ztest_dataset_open(),
5568 	 * when the open-context and syncing-context values agree.
5569 	 * That's because zap_count() returns the open-context value,
5570 	 * while dmu_objset_space() returns the rootbp fill count.
5571 	 */
5572 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5573 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5574 	ASSERT3U(dirobjs + 1, ==, usedobjs);
5575 }
5576 
5577 static int
5578 ztest_dataset_open(int d)
5579 {
5580 	ztest_ds_t *zd = &ztest_ds[d];
5581 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5582 	objset_t *os;
5583 	zilog_t *zilog;
5584 	char name[ZFS_MAX_DATASET_NAME_LEN];
5585 	int error;
5586 
5587 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5588 
5589 	(void) rw_rdlock(&ztest_name_lock);
5590 
5591 	error = ztest_dataset_create(name);
5592 	if (error == ENOSPC) {
5593 		(void) rw_unlock(&ztest_name_lock);
5594 		ztest_record_enospc(FTAG);
5595 		return (error);
5596 	}
5597 	ASSERT(error == 0 || error == EEXIST);
5598 
5599 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5600 	(void) rw_unlock(&ztest_name_lock);
5601 
5602 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5603 
5604 	zilog = zd->zd_zilog;
5605 
5606 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5607 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
5608 		fatal(0, "missing log records: claimed %llu < committed %llu",
5609 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
5610 
5611 	ztest_dataset_dirobj_verify(zd);
5612 
5613 	zil_replay(os, zd, ztest_replay_vector);
5614 
5615 	ztest_dataset_dirobj_verify(zd);
5616 
5617 	if (ztest_opts.zo_verbose >= 6)
5618 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5619 		    zd->zd_name,
5620 		    (u_longlong_t)zilog->zl_parse_blk_count,
5621 		    (u_longlong_t)zilog->zl_parse_lr_count,
5622 		    (u_longlong_t)zilog->zl_replaying_seq);
5623 
5624 	zilog = zil_open(os, ztest_get_data);
5625 
5626 	if (zilog->zl_replaying_seq != 0 &&
5627 	    zilog->zl_replaying_seq < committed_seq)
5628 		fatal(0, "missing log records: replayed %llu < committed %llu",
5629 		    zilog->zl_replaying_seq, committed_seq);
5630 
5631 	return (0);
5632 }
5633 
5634 static void
5635 ztest_dataset_close(int d)
5636 {
5637 	ztest_ds_t *zd = &ztest_ds[d];
5638 
5639 	zil_close(zd->zd_zilog);
5640 	dmu_objset_disown(zd->zd_os, zd);
5641 
5642 	ztest_zd_fini(zd);
5643 }
5644 
5645 /*
5646  * Kick off threads to run tests on all datasets in parallel.
5647  */
5648 static void
5649 ztest_run(ztest_shared_t *zs)
5650 {
5651 	thread_t *tid;
5652 	spa_t *spa;
5653 	objset_t *os;
5654 	thread_t resume_tid;
5655 	int error;
5656 
5657 	ztest_exiting = B_FALSE;
5658 
5659 	/*
5660 	 * Initialize parent/child shared state.
5661 	 */
5662 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5663 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5664 
5665 	zs->zs_thread_start = gethrtime();
5666 	zs->zs_thread_stop =
5667 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5668 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5669 	zs->zs_thread_kill = zs->zs_thread_stop;
5670 	if (ztest_random(100) < ztest_opts.zo_killrate) {
5671 		zs->zs_thread_kill -=
5672 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
5673 	}
5674 
5675 	(void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5676 
5677 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5678 	    offsetof(ztest_cb_data_t, zcd_node));
5679 
5680 	/*
5681 	 * Open our pool.
5682 	 */
5683 	kernel_init(FREAD | FWRITE);
5684 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5685 	spa->spa_debug = B_TRUE;
5686 	metaslab_preload_limit = ztest_random(20) + 1;
5687 	ztest_spa = spa;
5688 
5689 	dmu_objset_stats_t dds;
5690 	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5691 	    DMU_OST_ANY, B_TRUE, FTAG, &os));
5692 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5693 	dmu_objset_fast_stat(os, &dds);
5694 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5695 	zs->zs_guid = dds.dds_guid;
5696 	dmu_objset_disown(os, FTAG);
5697 
5698 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5699 
5700 	/*
5701 	 * We don't expect the pool to suspend unless maxfaults == 0,
5702 	 * in which case ztest_fault_inject() temporarily takes away
5703 	 * the only valid replica.
5704 	 */
5705 	if (MAXFAULTS() == 0)
5706 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5707 	else
5708 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5709 
5710 	/*
5711 	 * Create a thread to periodically resume suspended I/O.
5712 	 */
5713 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5714 	    &resume_tid) == 0);
5715 
5716 	/*
5717 	 * Create a deadman thread to abort() if we hang.
5718 	 */
5719 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5720 	    NULL) == 0);
5721 
5722 	/*
5723 	 * Verify that we can safely inquire about about any object,
5724 	 * whether it's allocated or not.  To make it interesting,
5725 	 * we probe a 5-wide window around each power of two.
5726 	 * This hits all edge cases, including zero and the max.
5727 	 */
5728 	for (int t = 0; t < 64; t++) {
5729 		for (int d = -5; d <= 5; d++) {
5730 			error = dmu_object_info(spa->spa_meta_objset,
5731 			    (1ULL << t) + d, NULL);
5732 			ASSERT(error == 0 || error == ENOENT ||
5733 			    error == EINVAL);
5734 		}
5735 	}
5736 
5737 	/*
5738 	 * If we got any ENOSPC errors on the previous run, destroy something.
5739 	 */
5740 	if (zs->zs_enospc_count != 0) {
5741 		int d = ztest_random(ztest_opts.zo_datasets);
5742 		ztest_dataset_destroy(d);
5743 	}
5744 	zs->zs_enospc_count = 0;
5745 
5746 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5747 	    UMEM_NOFAIL);
5748 
5749 	if (ztest_opts.zo_verbose >= 4)
5750 		(void) printf("starting main threads...\n");
5751 
5752 	/*
5753 	 * Kick off all the tests that run in parallel.
5754 	 */
5755 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
5756 		if (t < ztest_opts.zo_datasets &&
5757 		    ztest_dataset_open(t) != 0)
5758 			return;
5759 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5760 		    THR_BOUND, &tid[t]) == 0);
5761 	}
5762 
5763 	/*
5764 	 * Wait for all of the tests to complete.  We go in reverse order
5765 	 * so we don't close datasets while threads are still using them.
5766 	 */
5767 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5768 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5769 		if (t < ztest_opts.zo_datasets)
5770 			ztest_dataset_close(t);
5771 	}
5772 
5773 	txg_wait_synced(spa_get_dsl(spa), 0);
5774 
5775 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5776 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5777 	zfs_dbgmsg_print(FTAG);
5778 
5779 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5780 
5781 	/* Kill the resume thread */
5782 	ztest_exiting = B_TRUE;
5783 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5784 	ztest_resume(spa);
5785 
5786 	/*
5787 	 * Right before closing the pool, kick off a bunch of async I/O;
5788 	 * spa_close() should wait for it to complete.
5789 	 */
5790 	for (uint64_t object = 1; object < 50; object++) {
5791 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5792 		    ZIO_PRIORITY_SYNC_READ);
5793 	}
5794 
5795 	spa_close(spa, FTAG);
5796 
5797 	/*
5798 	 * Verify that we can loop over all pools.
5799 	 */
5800 	mutex_enter(&spa_namespace_lock);
5801 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5802 		if (ztest_opts.zo_verbose > 3)
5803 			(void) printf("spa_next: found %s\n", spa_name(spa));
5804 	mutex_exit(&spa_namespace_lock);
5805 
5806 	/*
5807 	 * Verify that we can export the pool and reimport it under a
5808 	 * different name.
5809 	 */
5810 	if (ztest_random(2) == 0) {
5811 		char name[ZFS_MAX_DATASET_NAME_LEN];
5812 		(void) snprintf(name, sizeof (name), "%s_import",
5813 		    ztest_opts.zo_pool);
5814 		ztest_spa_import_export(ztest_opts.zo_pool, name);
5815 		ztest_spa_import_export(name, ztest_opts.zo_pool);
5816 	}
5817 
5818 	kernel_fini();
5819 
5820 	list_destroy(&zcl.zcl_callbacks);
5821 
5822 	(void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5823 
5824 	(void) rwlock_destroy(&ztest_name_lock);
5825 	(void) _mutex_destroy(&ztest_vdev_lock);
5826 }
5827 
5828 static void
5829 ztest_freeze(void)
5830 {
5831 	ztest_ds_t *zd = &ztest_ds[0];
5832 	spa_t *spa;
5833 	int numloops = 0;
5834 
5835 	if (ztest_opts.zo_verbose >= 3)
5836 		(void) printf("testing spa_freeze()...\n");
5837 
5838 	kernel_init(FREAD | FWRITE);
5839 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5840 	VERIFY3U(0, ==, ztest_dataset_open(0));
5841 	spa->spa_debug = B_TRUE;
5842 	ztest_spa = spa;
5843 
5844 	/*
5845 	 * Force the first log block to be transactionally allocated.
5846 	 * We have to do this before we freeze the pool -- otherwise
5847 	 * the log chain won't be anchored.
5848 	 */
5849 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5850 		ztest_dmu_object_alloc_free(zd, 0);
5851 		zil_commit(zd->zd_zilog, 0);
5852 	}
5853 
5854 	txg_wait_synced(spa_get_dsl(spa), 0);
5855 
5856 	/*
5857 	 * Freeze the pool.  This stops spa_sync() from doing anything,
5858 	 * so that the only way to record changes from now on is the ZIL.
5859 	 */
5860 	spa_freeze(spa);
5861 
5862 	/*
5863 	 * Because it is hard to predict how much space a write will actually
5864 	 * require beforehand, we leave ourselves some fudge space to write over
5865 	 * capacity.
5866 	 */
5867 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5868 
5869 	/*
5870 	 * Run tests that generate log records but don't alter the pool config
5871 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5872 	 * We do a txg_wait_synced() after each iteration to force the txg
5873 	 * to increase well beyond the last synced value in the uberblock.
5874 	 * The ZIL should be OK with that.
5875 	 *
5876 	 * Run a random number of times less than zo_maxloops and ensure we do
5877 	 * not run out of space on the pool.
5878 	 */
5879 	while (ztest_random(10) != 0 &&
5880 	    numloops++ < ztest_opts.zo_maxloops &&
5881 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5882 		ztest_od_t od;
5883 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5884 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5885 		ztest_io(zd, od.od_object,
5886 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5887 		txg_wait_synced(spa_get_dsl(spa), 0);
5888 	}
5889 
5890 	/*
5891 	 * Commit all of the changes we just generated.
5892 	 */
5893 	zil_commit(zd->zd_zilog, 0);
5894 	txg_wait_synced(spa_get_dsl(spa), 0);
5895 
5896 	/*
5897 	 * Close our dataset and close the pool.
5898 	 */
5899 	ztest_dataset_close(0);
5900 	spa_close(spa, FTAG);
5901 	kernel_fini();
5902 
5903 	/*
5904 	 * Open and close the pool and dataset to induce log replay.
5905 	 */
5906 	kernel_init(FREAD | FWRITE);
5907 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5908 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5909 	VERIFY3U(0, ==, ztest_dataset_open(0));
5910 	ztest_dataset_close(0);
5911 
5912 	spa->spa_debug = B_TRUE;
5913 	ztest_spa = spa;
5914 	txg_wait_synced(spa_get_dsl(spa), 0);
5915 	ztest_reguid(NULL, 0);
5916 
5917 	spa_close(spa, FTAG);
5918 	kernel_fini();
5919 }
5920 
5921 void
5922 print_time(hrtime_t t, char *timebuf)
5923 {
5924 	hrtime_t s = t / NANOSEC;
5925 	hrtime_t m = s / 60;
5926 	hrtime_t h = m / 60;
5927 	hrtime_t d = h / 24;
5928 
5929 	s -= m * 60;
5930 	m -= h * 60;
5931 	h -= d * 24;
5932 
5933 	timebuf[0] = '\0';
5934 
5935 	if (d)
5936 		(void) sprintf(timebuf,
5937 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
5938 	else if (h)
5939 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5940 	else if (m)
5941 		(void) sprintf(timebuf, "%llum%02llus", m, s);
5942 	else
5943 		(void) sprintf(timebuf, "%llus", s);
5944 }
5945 
5946 static nvlist_t *
5947 make_random_props()
5948 {
5949 	nvlist_t *props;
5950 
5951 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5952 	if (ztest_random(2) == 0)
5953 		return (props);
5954 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5955 
5956 	return (props);
5957 }
5958 
5959 /*
5960  * Create a storage pool with the given name and initial vdev size.
5961  * Then test spa_freeze() functionality.
5962  */
5963 static void
5964 ztest_init(ztest_shared_t *zs)
5965 {
5966 	spa_t *spa;
5967 	nvlist_t *nvroot, *props;
5968 
5969 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5970 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5971 
5972 	kernel_init(FREAD | FWRITE);
5973 
5974 	/*
5975 	 * Create the storage pool.
5976 	 */
5977 	(void) spa_destroy(ztest_opts.zo_pool);
5978 	ztest_shared->zs_vdev_next_leaf = 0;
5979 	zs->zs_splits = 0;
5980 	zs->zs_mirrors = ztest_opts.zo_mirrors;
5981 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5982 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5983 	props = make_random_props();
5984 	for (int i = 0; i < SPA_FEATURES; i++) {
5985 		char buf[1024];
5986 		(void) snprintf(buf, sizeof (buf), "feature@%s",
5987 		    spa_feature_table[i].fi_uname);
5988 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5989 	}
5990 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5991 	nvlist_free(nvroot);
5992 	nvlist_free(props);
5993 
5994 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5995 	zs->zs_metaslab_sz =
5996 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5997 
5998 	spa_close(spa, FTAG);
5999 
6000 	kernel_fini();
6001 
6002 	ztest_run_zdb(ztest_opts.zo_pool);
6003 
6004 	ztest_freeze();
6005 
6006 	ztest_run_zdb(ztest_opts.zo_pool);
6007 
6008 	(void) rwlock_destroy(&ztest_name_lock);
6009 	(void) _mutex_destroy(&ztest_vdev_lock);
6010 }
6011 
6012 static void
6013 setup_data_fd(void)
6014 {
6015 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6016 
6017 	ztest_fd_data = mkstemp(ztest_name_data);
6018 	ASSERT3S(ztest_fd_data, >=, 0);
6019 	(void) unlink(ztest_name_data);
6020 }
6021 
6022 
6023 static int
6024 shared_data_size(ztest_shared_hdr_t *hdr)
6025 {
6026 	int size;
6027 
6028 	size = hdr->zh_hdr_size;
6029 	size += hdr->zh_opts_size;
6030 	size += hdr->zh_size;
6031 	size += hdr->zh_stats_size * hdr->zh_stats_count;
6032 	size += hdr->zh_ds_size * hdr->zh_ds_count;
6033 
6034 	return (size);
6035 }
6036 
6037 static void
6038 setup_hdr(void)
6039 {
6040 	int size;
6041 	ztest_shared_hdr_t *hdr;
6042 
6043 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6044 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6045 	ASSERT(hdr != MAP_FAILED);
6046 
6047 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6048 
6049 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6050 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6051 	hdr->zh_size = sizeof (ztest_shared_t);
6052 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6053 	hdr->zh_stats_count = ZTEST_FUNCS;
6054 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6055 	hdr->zh_ds_count = ztest_opts.zo_datasets;
6056 
6057 	size = shared_data_size(hdr);
6058 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6059 
6060 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6061 }
6062 
6063 static void
6064 setup_data(void)
6065 {
6066 	int size, offset;
6067 	ztest_shared_hdr_t *hdr;
6068 	uint8_t *buf;
6069 
6070 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6071 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6072 	ASSERT(hdr != MAP_FAILED);
6073 
6074 	size = shared_data_size(hdr);
6075 
6076 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6077 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6078 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6079 	ASSERT(hdr != MAP_FAILED);
6080 	buf = (uint8_t *)hdr;
6081 
6082 	offset = hdr->zh_hdr_size;
6083 	ztest_shared_opts = (void *)&buf[offset];
6084 	offset += hdr->zh_opts_size;
6085 	ztest_shared = (void *)&buf[offset];
6086 	offset += hdr->zh_size;
6087 	ztest_shared_callstate = (void *)&buf[offset];
6088 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
6089 	ztest_shared_ds = (void *)&buf[offset];
6090 }
6091 
6092 static boolean_t
6093 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6094 {
6095 	pid_t pid;
6096 	int status;
6097 	char *cmdbuf = NULL;
6098 
6099 	pid = fork();
6100 
6101 	if (cmd == NULL) {
6102 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6103 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6104 		cmd = cmdbuf;
6105 	}
6106 
6107 	if (pid == -1)
6108 		fatal(1, "fork failed");
6109 
6110 	if (pid == 0) {	/* child */
6111 		char *emptyargv[2] = { cmd, NULL };
6112 		char fd_data_str[12];
6113 
6114 		struct rlimit rl = { 1024, 1024 };
6115 		(void) setrlimit(RLIMIT_NOFILE, &rl);
6116 
6117 		(void) close(ztest_fd_rand);
6118 		VERIFY3U(11, >=,
6119 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6120 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6121 
6122 		(void) enable_extended_FILE_stdio(-1, -1);
6123 		if (libpath != NULL)
6124 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6125 		(void) execv(cmd, emptyargv);
6126 		ztest_dump_core = B_FALSE;
6127 		fatal(B_TRUE, "exec failed: %s", cmd);
6128 	}
6129 
6130 	if (cmdbuf != NULL) {
6131 		umem_free(cmdbuf, MAXPATHLEN);
6132 		cmd = NULL;
6133 	}
6134 
6135 	while (waitpid(pid, &status, 0) != pid)
6136 		continue;
6137 	if (statusp != NULL)
6138 		*statusp = status;
6139 
6140 	if (WIFEXITED(status)) {
6141 		if (WEXITSTATUS(status) != 0) {
6142 			(void) fprintf(stderr, "child exited with code %d\n",
6143 			    WEXITSTATUS(status));
6144 			exit(2);
6145 		}
6146 		return (B_FALSE);
6147 	} else if (WIFSIGNALED(status)) {
6148 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6149 			(void) fprintf(stderr, "child died with signal %d\n",
6150 			    WTERMSIG(status));
6151 			exit(3);
6152 		}
6153 		return (B_TRUE);
6154 	} else {
6155 		(void) fprintf(stderr, "something strange happened to child\n");
6156 		exit(4);
6157 		/* NOTREACHED */
6158 	}
6159 }
6160 
6161 static void
6162 ztest_run_init(void)
6163 {
6164 	ztest_shared_t *zs = ztest_shared;
6165 
6166 	ASSERT(ztest_opts.zo_init != 0);
6167 
6168 	/*
6169 	 * Blow away any existing copy of zpool.cache
6170 	 */
6171 	(void) remove(spa_config_path);
6172 
6173 	/*
6174 	 * Create and initialize our storage pool.
6175 	 */
6176 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
6177 		bzero(zs, sizeof (ztest_shared_t));
6178 		if (ztest_opts.zo_verbose >= 3 &&
6179 		    ztest_opts.zo_init != 1) {
6180 			(void) printf("ztest_init(), pass %d\n", i);
6181 		}
6182 		ztest_init(zs);
6183 	}
6184 }
6185 
6186 int
6187 main(int argc, char **argv)
6188 {
6189 	int kills = 0;
6190 	int iters = 0;
6191 	int older = 0;
6192 	int newer = 0;
6193 	ztest_shared_t *zs;
6194 	ztest_info_t *zi;
6195 	ztest_shared_callstate_t *zc;
6196 	char timebuf[100];
6197 	char numbuf[6];
6198 	spa_t *spa;
6199 	char *cmd;
6200 	boolean_t hasalt;
6201 	char *fd_data_str = getenv("ZTEST_FD_DATA");
6202 
6203 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
6204 
6205 	dprintf_setup(&argc, argv);
6206 	zfs_deadman_synctime_ms = 300000;
6207 
6208 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6209 	ASSERT3S(ztest_fd_rand, >=, 0);
6210 
6211 	if (!fd_data_str) {
6212 		process_options(argc, argv);
6213 
6214 		setup_data_fd();
6215 		setup_hdr();
6216 		setup_data();
6217 		bcopy(&ztest_opts, ztest_shared_opts,
6218 		    sizeof (*ztest_shared_opts));
6219 	} else {
6220 		ztest_fd_data = atoi(fd_data_str);
6221 		setup_data();
6222 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6223 	}
6224 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6225 
6226 	/* Override location of zpool.cache */
6227 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6228 	    ztest_opts.zo_dir), !=, -1);
6229 
6230 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6231 	    UMEM_NOFAIL);
6232 	zs = ztest_shared;
6233 
6234 	if (fd_data_str) {
6235 		metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6236 		metaslab_df_alloc_threshold =
6237 		    zs->zs_metaslab_df_alloc_threshold;
6238 
6239 		if (zs->zs_do_init)
6240 			ztest_run_init();
6241 		else
6242 			ztest_run(zs);
6243 		exit(0);
6244 	}
6245 
6246 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6247 
6248 	if (ztest_opts.zo_verbose >= 1) {
6249 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6250 		    " %llu seconds...\n",
6251 		    (u_longlong_t)ztest_opts.zo_vdevs,
6252 		    ztest_opts.zo_datasets,
6253 		    ztest_opts.zo_threads,
6254 		    (u_longlong_t)ztest_opts.zo_time);
6255 	}
6256 
6257 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6258 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6259 
6260 	zs->zs_do_init = B_TRUE;
6261 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6262 		if (ztest_opts.zo_verbose >= 1) {
6263 			(void) printf("Executing older ztest for "
6264 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6265 		}
6266 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6267 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6268 	} else {
6269 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6270 	}
6271 	zs->zs_do_init = B_FALSE;
6272 
6273 	zs->zs_proc_start = gethrtime();
6274 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6275 
6276 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6277 		zi = &ztest_info[f];
6278 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6279 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6280 			zc->zc_next = UINT64_MAX;
6281 		else
6282 			zc->zc_next = zs->zs_proc_start +
6283 			    ztest_random(2 * zi->zi_interval[0] + 1);
6284 	}
6285 
6286 	/*
6287 	 * Run the tests in a loop.  These tests include fault injection
6288 	 * to verify that self-healing data works, and forced crashes
6289 	 * to verify that we never lose on-disk consistency.
6290 	 */
6291 	while (gethrtime() < zs->zs_proc_stop) {
6292 		int status;
6293 		boolean_t killed;
6294 
6295 		/*
6296 		 * Initialize the workload counters for each function.
6297 		 */
6298 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6299 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6300 			zc->zc_count = 0;
6301 			zc->zc_time = 0;
6302 		}
6303 
6304 		/* Set the allocation switch size */
6305 		zs->zs_metaslab_df_alloc_threshold =
6306 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6307 
6308 		if (!hasalt || ztest_random(2) == 0) {
6309 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6310 				(void) printf("Executing newer ztest: %s\n",
6311 				    cmd);
6312 			}
6313 			newer++;
6314 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6315 		} else {
6316 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6317 				(void) printf("Executing older ztest: %s\n",
6318 				    ztest_opts.zo_alt_ztest);
6319 			}
6320 			older++;
6321 			killed = exec_child(ztest_opts.zo_alt_ztest,
6322 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6323 		}
6324 
6325 		if (killed)
6326 			kills++;
6327 		iters++;
6328 
6329 		if (ztest_opts.zo_verbose >= 1) {
6330 			hrtime_t now = gethrtime();
6331 
6332 			now = MIN(now, zs->zs_proc_stop);
6333 			print_time(zs->zs_proc_stop - now, timebuf);
6334 			nicenum(zs->zs_space, numbuf);
6335 
6336 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6337 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6338 			    iters,
6339 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6340 			    (u_longlong_t)zs->zs_enospc_count,
6341 			    100.0 * zs->zs_alloc / zs->zs_space,
6342 			    numbuf,
6343 			    100.0 * (now - zs->zs_proc_start) /
6344 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6345 		}
6346 
6347 		if (ztest_opts.zo_verbose >= 2) {
6348 			(void) printf("\nWorkload summary:\n\n");
6349 			(void) printf("%7s %9s   %s\n",
6350 			    "Calls", "Time", "Function");
6351 			(void) printf("%7s %9s   %s\n",
6352 			    "-----", "----", "--------");
6353 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6354 				Dl_info dli;
6355 
6356 				zi = &ztest_info[f];
6357 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6358 				print_time(zc->zc_time, timebuf);
6359 				(void) dladdr((void *)zi->zi_func, &dli);
6360 				(void) printf("%7llu %9s   %s\n",
6361 				    (u_longlong_t)zc->zc_count, timebuf,
6362 				    dli.dli_sname);
6363 			}
6364 			(void) printf("\n");
6365 		}
6366 
6367 		/*
6368 		 * It's possible that we killed a child during a rename test,
6369 		 * in which case we'll have a 'ztest_tmp' pool lying around
6370 		 * instead of 'ztest'.  Do a blind rename in case this happened.
6371 		 */
6372 		kernel_init(FREAD);
6373 		if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6374 			spa_close(spa, FTAG);
6375 		} else {
6376 			char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6377 			kernel_fini();
6378 			kernel_init(FREAD | FWRITE);
6379 			(void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6380 			    ztest_opts.zo_pool);
6381 			(void) spa_rename(tmpname, ztest_opts.zo_pool);
6382 		}
6383 		kernel_fini();
6384 
6385 		ztest_run_zdb(ztest_opts.zo_pool);
6386 	}
6387 
6388 	if (ztest_opts.zo_verbose >= 1) {
6389 		if (hasalt) {
6390 			(void) printf("%d runs of older ztest: %s\n", older,
6391 			    ztest_opts.zo_alt_ztest);
6392 			(void) printf("%d runs of newer ztest: %s\n", newer,
6393 			    cmd);
6394 		}
6395 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6396 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6397 	}
6398 
6399 	umem_free(cmd, MAXNAMELEN);
6400 
6401 	return (0);
6402 }
6403