xref: /freebsd/sys/contrib/openzfs/cmd/ztest.c (revision b197d4b893974c9eb4d7b38704c6d5c486235d6f)
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 https://opensource.org/licenses/CDDL-1.0.
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, 2018 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2017 Joyent, Inc.
28  * Copyright (c) 2017, Intel Corporation.
29  */
30 
31 /*
32  * The objective of this program is to provide a DMU/ZAP/SPA stress test
33  * that runs entirely in userland, is easy to use, and easy to extend.
34  *
35  * The overall design of the ztest program is as follows:
36  *
37  * (1) For each major functional area (e.g. adding vdevs to a pool,
38  *     creating and destroying datasets, reading and writing objects, etc)
39  *     we have a simple routine to test that functionality.  These
40  *     individual routines do not have to do anything "stressful".
41  *
42  * (2) We turn these simple functionality tests into a stress test by
43  *     running them all in parallel, with as many threads as desired,
44  *     and spread across as many datasets, objects, and vdevs as desired.
45  *
46  * (3) While all this is happening, we inject faults into the pool to
47  *     verify that self-healing data really works.
48  *
49  * (4) Every time we open a dataset, we change its checksum and compression
50  *     functions.  Thus even individual objects vary from block to block
51  *     in which checksum they use and whether they're compressed.
52  *
53  * (5) To verify that we never lose on-disk consistency after a crash,
54  *     we run the entire test in a child of the main process.
55  *     At random times, the child self-immolates with a SIGKILL.
56  *     This is the software equivalent of pulling the power cord.
57  *     The parent then runs the test again, using the existing
58  *     storage pool, as many times as desired. If backwards compatibility
59  *     testing is enabled ztest will sometimes run the "older" version
60  *     of ztest after a SIGKILL.
61  *
62  * (6) To verify that we don't have future leaks or temporal incursions,
63  *     many of the functional tests record the transaction group number
64  *     as part of their data.  When reading old data, they verify that
65  *     the transaction group number is less than the current, open txg.
66  *     If you add a new test, please do this if applicable.
67  *
68  * (7) Threads are created with a reduced stack size, for sanity checking.
69  *     Therefore, it's important not to allocate huge buffers on the stack.
70  *
71  * When run with no arguments, ztest runs for about five minutes and
72  * produces no output if successful.  To get a little bit of information,
73  * specify -V.  To get more information, specify -VV, and so on.
74  *
75  * To turn this into an overnight stress test, use -T to specify run time.
76  *
77  * You can ask more vdevs [-v], datasets [-d], or threads [-t]
78  * to increase the pool capacity, fanout, and overall stress level.
79  *
80  * Use the -k option to set the desired frequency of kills.
81  *
82  * When ztest invokes itself it passes all relevant information through a
83  * temporary file which is mmap-ed in the child process. This allows shared
84  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
85  * stored at offset 0 of this file and contains information on the size and
86  * number of shared structures in the file. The information stored in this file
87  * must remain backwards compatible with older versions of ztest so that
88  * ztest can invoke them during backwards compatibility testing (-B).
89  */
90 
91 #include <sys/zfs_context.h>
92 #include <sys/spa.h>
93 #include <sys/dmu.h>
94 #include <sys/txg.h>
95 #include <sys/dbuf.h>
96 #include <sys/zap.h>
97 #include <sys/dmu_objset.h>
98 #include <sys/poll.h>
99 #include <sys/stat.h>
100 #include <sys/time.h>
101 #include <sys/wait.h>
102 #include <sys/mman.h>
103 #include <sys/resource.h>
104 #include <sys/zio.h>
105 #include <sys/zil.h>
106 #include <sys/zil_impl.h>
107 #include <sys/vdev_draid.h>
108 #include <sys/vdev_impl.h>
109 #include <sys/vdev_file.h>
110 #include <sys/vdev_initialize.h>
111 #include <sys/vdev_raidz.h>
112 #include <sys/vdev_trim.h>
113 #include <sys/spa_impl.h>
114 #include <sys/metaslab_impl.h>
115 #include <sys/dsl_prop.h>
116 #include <sys/dsl_dataset.h>
117 #include <sys/dsl_destroy.h>
118 #include <sys/dsl_scan.h>
119 #include <sys/zio_checksum.h>
120 #include <sys/zfs_refcount.h>
121 #include <sys/zfeature.h>
122 #include <sys/dsl_userhold.h>
123 #include <sys/abd.h>
124 #include <sys/blake3.h>
125 #include <stdio.h>
126 #include <stdlib.h>
127 #include <unistd.h>
128 #include <getopt.h>
129 #include <signal.h>
130 #include <umem.h>
131 #include <ctype.h>
132 #include <math.h>
133 #include <sys/fs/zfs.h>
134 #include <zfs_fletcher.h>
135 #include <libnvpair.h>
136 #include <libzutil.h>
137 #include <sys/crypto/icp.h>
138 #if (__GLIBC__ && !__UCLIBC__)
139 #include <execinfo.h> /* for backtrace() */
140 #endif
141 
142 static int ztest_fd_data = -1;
143 static int ztest_fd_rand = -1;
144 
145 typedef struct ztest_shared_hdr {
146 	uint64_t	zh_hdr_size;
147 	uint64_t	zh_opts_size;
148 	uint64_t	zh_size;
149 	uint64_t	zh_stats_size;
150 	uint64_t	zh_stats_count;
151 	uint64_t	zh_ds_size;
152 	uint64_t	zh_ds_count;
153 } ztest_shared_hdr_t;
154 
155 static ztest_shared_hdr_t *ztest_shared_hdr;
156 
157 enum ztest_class_state {
158 	ZTEST_VDEV_CLASS_OFF,
159 	ZTEST_VDEV_CLASS_ON,
160 	ZTEST_VDEV_CLASS_RND
161 };
162 
163 #define	ZO_GVARS_MAX_ARGLEN	((size_t)64)
164 #define	ZO_GVARS_MAX_COUNT	((size_t)10)
165 
166 typedef struct ztest_shared_opts {
167 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
168 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
169 	char zo_alt_ztest[MAXNAMELEN];
170 	char zo_alt_libpath[MAXNAMELEN];
171 	uint64_t zo_vdevs;
172 	uint64_t zo_vdevtime;
173 	size_t zo_vdev_size;
174 	int zo_ashift;
175 	int zo_mirrors;
176 	int zo_raid_children;
177 	int zo_raid_parity;
178 	char zo_raid_type[8];
179 	int zo_draid_data;
180 	int zo_draid_spares;
181 	int zo_datasets;
182 	int zo_threads;
183 	uint64_t zo_passtime;
184 	uint64_t zo_killrate;
185 	int zo_verbose;
186 	int zo_init;
187 	uint64_t zo_time;
188 	uint64_t zo_maxloops;
189 	uint64_t zo_metaslab_force_ganging;
190 	int zo_mmp_test;
191 	int zo_special_vdevs;
192 	int zo_dump_dbgmsg;
193 	int zo_gvars_count;
194 	char zo_gvars[ZO_GVARS_MAX_COUNT][ZO_GVARS_MAX_ARGLEN];
195 } ztest_shared_opts_t;
196 
197 /* Default values for command line options. */
198 #define	DEFAULT_POOL "ztest"
199 #define	DEFAULT_VDEV_DIR "/tmp"
200 #define	DEFAULT_VDEV_COUNT 5
201 #define	DEFAULT_VDEV_SIZE (SPA_MINDEVSIZE * 4)	/* 256m default size */
202 #define	DEFAULT_VDEV_SIZE_STR "256M"
203 #define	DEFAULT_ASHIFT SPA_MINBLOCKSHIFT
204 #define	DEFAULT_MIRRORS 2
205 #define	DEFAULT_RAID_CHILDREN 4
206 #define	DEFAULT_RAID_PARITY 1
207 #define	DEFAULT_DRAID_DATA 4
208 #define	DEFAULT_DRAID_SPARES 1
209 #define	DEFAULT_DATASETS_COUNT 7
210 #define	DEFAULT_THREADS 23
211 #define	DEFAULT_RUN_TIME 300 /* 300 seconds */
212 #define	DEFAULT_RUN_TIME_STR "300 sec"
213 #define	DEFAULT_PASS_TIME 60 /* 60 seconds */
214 #define	DEFAULT_PASS_TIME_STR "60 sec"
215 #define	DEFAULT_KILL_RATE 70 /* 70% kill rate */
216 #define	DEFAULT_KILLRATE_STR "70%"
217 #define	DEFAULT_INITS 1
218 #define	DEFAULT_MAX_LOOPS 50 /* 5 minutes */
219 #define	DEFAULT_FORCE_GANGING (64 << 10)
220 #define	DEFAULT_FORCE_GANGING_STR "64K"
221 
222 /* Simplifying assumption: -1 is not a valid default. */
223 #define	NO_DEFAULT -1
224 
225 static const ztest_shared_opts_t ztest_opts_defaults = {
226 	.zo_pool = DEFAULT_POOL,
227 	.zo_dir = DEFAULT_VDEV_DIR,
228 	.zo_alt_ztest = { '\0' },
229 	.zo_alt_libpath = { '\0' },
230 	.zo_vdevs = DEFAULT_VDEV_COUNT,
231 	.zo_ashift = DEFAULT_ASHIFT,
232 	.zo_mirrors = DEFAULT_MIRRORS,
233 	.zo_raid_children = DEFAULT_RAID_CHILDREN,
234 	.zo_raid_parity = DEFAULT_RAID_PARITY,
235 	.zo_raid_type = VDEV_TYPE_RAIDZ,
236 	.zo_vdev_size = DEFAULT_VDEV_SIZE,
237 	.zo_draid_data = DEFAULT_DRAID_DATA,	/* data drives */
238 	.zo_draid_spares = DEFAULT_DRAID_SPARES, /* distributed spares */
239 	.zo_datasets = DEFAULT_DATASETS_COUNT,
240 	.zo_threads = DEFAULT_THREADS,
241 	.zo_passtime = DEFAULT_PASS_TIME,
242 	.zo_killrate = DEFAULT_KILL_RATE,
243 	.zo_verbose = 0,
244 	.zo_mmp_test = 0,
245 	.zo_init = DEFAULT_INITS,
246 	.zo_time = DEFAULT_RUN_TIME,
247 	.zo_maxloops = DEFAULT_MAX_LOOPS, /* max loops during spa_freeze() */
248 	.zo_metaslab_force_ganging = DEFAULT_FORCE_GANGING,
249 	.zo_special_vdevs = ZTEST_VDEV_CLASS_RND,
250 	.zo_gvars_count = 0,
251 };
252 
253 extern uint64_t metaslab_force_ganging;
254 extern uint64_t metaslab_df_alloc_threshold;
255 extern unsigned long zfs_deadman_synctime_ms;
256 extern int metaslab_preload_limit;
257 extern int zfs_compressed_arc_enabled;
258 extern int zfs_abd_scatter_enabled;
259 extern int dmu_object_alloc_chunk_shift;
260 extern boolean_t zfs_force_some_double_word_sm_entries;
261 extern unsigned long zio_decompress_fail_fraction;
262 extern unsigned long zfs_reconstruct_indirect_damage_fraction;
263 
264 
265 static ztest_shared_opts_t *ztest_shared_opts;
266 static ztest_shared_opts_t ztest_opts;
267 static const char *const ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
268 
269 typedef struct ztest_shared_ds {
270 	uint64_t	zd_seq;
271 } ztest_shared_ds_t;
272 
273 static ztest_shared_ds_t *ztest_shared_ds;
274 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
275 
276 #define	BT_MAGIC	0x123456789abcdefULL
277 #define	MAXFAULTS(zs) \
278 	(MAX((zs)->zs_mirrors, 1) * (ztest_opts.zo_raid_parity + 1) - 1)
279 
280 enum ztest_io_type {
281 	ZTEST_IO_WRITE_TAG,
282 	ZTEST_IO_WRITE_PATTERN,
283 	ZTEST_IO_WRITE_ZEROES,
284 	ZTEST_IO_TRUNCATE,
285 	ZTEST_IO_SETATTR,
286 	ZTEST_IO_REWRITE,
287 	ZTEST_IO_TYPES
288 };
289 
290 typedef struct ztest_block_tag {
291 	uint64_t	bt_magic;
292 	uint64_t	bt_objset;
293 	uint64_t	bt_object;
294 	uint64_t	bt_dnodesize;
295 	uint64_t	bt_offset;
296 	uint64_t	bt_gen;
297 	uint64_t	bt_txg;
298 	uint64_t	bt_crtxg;
299 } ztest_block_tag_t;
300 
301 typedef struct bufwad {
302 	uint64_t	bw_index;
303 	uint64_t	bw_txg;
304 	uint64_t	bw_data;
305 } bufwad_t;
306 
307 /*
308  * It would be better to use a rangelock_t per object.  Unfortunately
309  * the rangelock_t is not a drop-in replacement for rl_t, because we
310  * still need to map from object ID to rangelock_t.
311  */
312 typedef enum {
313 	RL_READER,
314 	RL_WRITER,
315 	RL_APPEND
316 } rl_type_t;
317 
318 typedef struct rll {
319 	void		*rll_writer;
320 	int		rll_readers;
321 	kmutex_t	rll_lock;
322 	kcondvar_t	rll_cv;
323 } rll_t;
324 
325 typedef struct rl {
326 	uint64_t	rl_object;
327 	uint64_t	rl_offset;
328 	uint64_t	rl_size;
329 	rll_t		*rl_lock;
330 } rl_t;
331 
332 #define	ZTEST_RANGE_LOCKS	64
333 #define	ZTEST_OBJECT_LOCKS	64
334 
335 /*
336  * Object descriptor.  Used as a template for object lookup/create/remove.
337  */
338 typedef struct ztest_od {
339 	uint64_t	od_dir;
340 	uint64_t	od_object;
341 	dmu_object_type_t od_type;
342 	dmu_object_type_t od_crtype;
343 	uint64_t	od_blocksize;
344 	uint64_t	od_crblocksize;
345 	uint64_t	od_crdnodesize;
346 	uint64_t	od_gen;
347 	uint64_t	od_crgen;
348 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
349 } ztest_od_t;
350 
351 /*
352  * Per-dataset state.
353  */
354 typedef struct ztest_ds {
355 	ztest_shared_ds_t *zd_shared;
356 	objset_t	*zd_os;
357 	pthread_rwlock_t zd_zilog_lock;
358 	zilog_t		*zd_zilog;
359 	ztest_od_t	*zd_od;		/* debugging aid */
360 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
361 	kmutex_t	zd_dirobj_lock;
362 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
363 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
364 } ztest_ds_t;
365 
366 /*
367  * Per-iteration state.
368  */
369 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
370 
371 typedef struct ztest_info {
372 	ztest_func_t	*zi_func;	/* test function */
373 	uint64_t	zi_iters;	/* iterations per execution */
374 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
375 	const char	*zi_funcname;	/* name of test function */
376 } ztest_info_t;
377 
378 typedef struct ztest_shared_callstate {
379 	uint64_t	zc_count;	/* per-pass count */
380 	uint64_t	zc_time;	/* per-pass time */
381 	uint64_t	zc_next;	/* next time to call this function */
382 } ztest_shared_callstate_t;
383 
384 static ztest_shared_callstate_t *ztest_shared_callstate;
385 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
386 
387 ztest_func_t ztest_dmu_read_write;
388 ztest_func_t ztest_dmu_write_parallel;
389 ztest_func_t ztest_dmu_object_alloc_free;
390 ztest_func_t ztest_dmu_object_next_chunk;
391 ztest_func_t ztest_dmu_commit_callbacks;
392 ztest_func_t ztest_zap;
393 ztest_func_t ztest_zap_parallel;
394 ztest_func_t ztest_zil_commit;
395 ztest_func_t ztest_zil_remount;
396 ztest_func_t ztest_dmu_read_write_zcopy;
397 ztest_func_t ztest_dmu_objset_create_destroy;
398 ztest_func_t ztest_dmu_prealloc;
399 ztest_func_t ztest_fzap;
400 ztest_func_t ztest_dmu_snapshot_create_destroy;
401 ztest_func_t ztest_dsl_prop_get_set;
402 ztest_func_t ztest_spa_prop_get_set;
403 ztest_func_t ztest_spa_create_destroy;
404 ztest_func_t ztest_fault_inject;
405 ztest_func_t ztest_dmu_snapshot_hold;
406 ztest_func_t ztest_mmp_enable_disable;
407 ztest_func_t ztest_scrub;
408 ztest_func_t ztest_dsl_dataset_promote_busy;
409 ztest_func_t ztest_vdev_attach_detach;
410 ztest_func_t ztest_vdev_LUN_growth;
411 ztest_func_t ztest_vdev_add_remove;
412 ztest_func_t ztest_vdev_class_add;
413 ztest_func_t ztest_vdev_aux_add_remove;
414 ztest_func_t ztest_split_pool;
415 ztest_func_t ztest_reguid;
416 ztest_func_t ztest_spa_upgrade;
417 ztest_func_t ztest_device_removal;
418 ztest_func_t ztest_spa_checkpoint_create_discard;
419 ztest_func_t ztest_initialize;
420 ztest_func_t ztest_trim;
421 ztest_func_t ztest_blake3;
422 ztest_func_t ztest_fletcher;
423 ztest_func_t ztest_fletcher_incr;
424 ztest_func_t ztest_verify_dnode_bt;
425 
426 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
427 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
428 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
429 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
430 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
431 
432 #define	ZTI_INIT(func, iters, interval) \
433 	{   .zi_func = (func), \
434 	    .zi_iters = (iters), \
435 	    .zi_interval = (interval), \
436 	    .zi_funcname = # func }
437 
438 ztest_info_t ztest_info[] = {
439 	ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
440 	ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
441 	ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
442 	ZTI_INIT(ztest_dmu_object_next_chunk, 1, &zopt_sometimes),
443 	ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
444 	ZTI_INIT(ztest_zap, 30, &zopt_always),
445 	ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
446 	ZTI_INIT(ztest_split_pool, 1, &zopt_always),
447 	ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
448 	ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
449 	ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
450 	ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
451 	ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
452 	ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
453 #if 0
454 	ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
455 #endif
456 	ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
457 	ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
458 	ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
459 	ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
460 	ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
461 	ZTI_INIT(ztest_mmp_enable_disable, 1, &zopt_sometimes),
462 	ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
463 	ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
464 	ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
465 	ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
466 	ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
467 	ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
468 	ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
469 	ZTI_INIT(ztest_vdev_class_add, 1, &ztest_opts.zo_vdevtime),
470 	ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
471 	ZTI_INIT(ztest_device_removal, 1, &zopt_sometimes),
472 	ZTI_INIT(ztest_spa_checkpoint_create_discard, 1, &zopt_rarely),
473 	ZTI_INIT(ztest_initialize, 1, &zopt_sometimes),
474 	ZTI_INIT(ztest_trim, 1, &zopt_sometimes),
475 	ZTI_INIT(ztest_blake3, 1, &zopt_rarely),
476 	ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
477 	ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
478 	ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
479 };
480 
481 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
482 
483 /*
484  * The following struct is used to hold a list of uncalled commit callbacks.
485  * The callbacks are ordered by txg number.
486  */
487 typedef struct ztest_cb_list {
488 	kmutex_t	zcl_callbacks_lock;
489 	list_t		zcl_callbacks;
490 } ztest_cb_list_t;
491 
492 /*
493  * Stuff we need to share writably between parent and child.
494  */
495 typedef struct ztest_shared {
496 	boolean_t	zs_do_init;
497 	hrtime_t	zs_proc_start;
498 	hrtime_t	zs_proc_stop;
499 	hrtime_t	zs_thread_start;
500 	hrtime_t	zs_thread_stop;
501 	hrtime_t	zs_thread_kill;
502 	uint64_t	zs_enospc_count;
503 	uint64_t	zs_vdev_next_leaf;
504 	uint64_t	zs_vdev_aux;
505 	uint64_t	zs_alloc;
506 	uint64_t	zs_space;
507 	uint64_t	zs_splits;
508 	uint64_t	zs_mirrors;
509 	uint64_t	zs_metaslab_sz;
510 	uint64_t	zs_metaslab_df_alloc_threshold;
511 	uint64_t	zs_guid;
512 } ztest_shared_t;
513 
514 #define	ID_PARALLEL	-1ULL
515 
516 static char ztest_dev_template[] = "%s/%s.%llua";
517 static char ztest_aux_template[] = "%s/%s.%s.%llu";
518 ztest_shared_t *ztest_shared;
519 
520 static spa_t *ztest_spa = NULL;
521 static ztest_ds_t *ztest_ds;
522 
523 static kmutex_t ztest_vdev_lock;
524 static boolean_t ztest_device_removal_active = B_FALSE;
525 static boolean_t ztest_pool_scrubbed = B_FALSE;
526 static kmutex_t ztest_checkpoint_lock;
527 
528 /*
529  * The ztest_name_lock protects the pool and dataset namespace used by
530  * the individual tests. To modify the namespace, consumers must grab
531  * this lock as writer. Grabbing the lock as reader will ensure that the
532  * namespace does not change while the lock is held.
533  */
534 static pthread_rwlock_t ztest_name_lock;
535 
536 static boolean_t ztest_dump_core = B_TRUE;
537 static boolean_t ztest_exiting;
538 
539 /* Global commit callback list */
540 static ztest_cb_list_t zcl;
541 /* Commit cb delay */
542 static uint64_t zc_min_txg_delay = UINT64_MAX;
543 static int zc_cb_counter = 0;
544 
545 /*
546  * Minimum number of commit callbacks that need to be registered for us to check
547  * whether the minimum txg delay is acceptable.
548  */
549 #define	ZTEST_COMMIT_CB_MIN_REG	100
550 
551 /*
552  * If a number of txgs equal to this threshold have been created after a commit
553  * callback has been registered but not called, then we assume there is an
554  * implementation bug.
555  */
556 #define	ZTEST_COMMIT_CB_THRESH	(TXG_CONCURRENT_STATES + 1000)
557 
558 enum ztest_object {
559 	ZTEST_META_DNODE = 0,
560 	ZTEST_DIROBJ,
561 	ZTEST_OBJECTS
562 };
563 
564 static __attribute__((noreturn)) void usage(boolean_t requested);
565 static int ztest_scrub_impl(spa_t *spa);
566 
567 /*
568  * These libumem hooks provide a reasonable set of defaults for the allocator's
569  * debugging facilities.
570  */
571 const char *
572 _umem_debug_init(void)
573 {
574 	return ("default,verbose"); /* $UMEM_DEBUG setting */
575 }
576 
577 const char *
578 _umem_logging_init(void)
579 {
580 	return ("fail,contents"); /* $UMEM_LOGGING setting */
581 }
582 
583 static void
584 dump_debug_buffer(void)
585 {
586 	ssize_t ret __attribute__((unused));
587 
588 	if (!ztest_opts.zo_dump_dbgmsg)
589 		return;
590 
591 	/*
592 	 * We use write() instead of printf() so that this function
593 	 * is safe to call from a signal handler.
594 	 */
595 	ret = write(STDOUT_FILENO, "\n", 1);
596 	zfs_dbgmsg_print("ztest");
597 }
598 
599 #define	BACKTRACE_SZ	100
600 
601 static void sig_handler(int signo)
602 {
603 	struct sigaction action;
604 #if (__GLIBC__ && !__UCLIBC__) /* backtrace() is a GNU extension */
605 	int nptrs;
606 	void *buffer[BACKTRACE_SZ];
607 
608 	nptrs = backtrace(buffer, BACKTRACE_SZ);
609 	backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
610 #endif
611 	dump_debug_buffer();
612 
613 	/*
614 	 * Restore default action and re-raise signal so SIGSEGV and
615 	 * SIGABRT can trigger a core dump.
616 	 */
617 	action.sa_handler = SIG_DFL;
618 	sigemptyset(&action.sa_mask);
619 	action.sa_flags = 0;
620 	(void) sigaction(signo, &action, NULL);
621 	raise(signo);
622 }
623 
624 #define	FATAL_MSG_SZ	1024
625 
626 static const char *fatal_msg;
627 
628 static __attribute__((format(printf, 2, 3))) __attribute__((noreturn)) void
629 fatal(int do_perror, const char *message, ...)
630 {
631 	va_list args;
632 	int save_errno = errno;
633 	char *buf;
634 
635 	(void) fflush(stdout);
636 	buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
637 	if (buf == NULL)
638 		goto out;
639 
640 	va_start(args, message);
641 	(void) sprintf(buf, "ztest: ");
642 	/* LINTED */
643 	(void) vsprintf(buf + strlen(buf), message, args);
644 	va_end(args);
645 	if (do_perror) {
646 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
647 		    ": %s", strerror(save_errno));
648 	}
649 	(void) fprintf(stderr, "%s\n", buf);
650 	fatal_msg = buf;			/* to ease debugging */
651 
652 out:
653 	if (ztest_dump_core)
654 		abort();
655 	else
656 		dump_debug_buffer();
657 
658 	exit(3);
659 }
660 
661 static int
662 str2shift(const char *buf)
663 {
664 	const char *ends = "BKMGTPEZ";
665 	int i;
666 
667 	if (buf[0] == '\0')
668 		return (0);
669 	for (i = 0; i < strlen(ends); i++) {
670 		if (toupper(buf[0]) == ends[i])
671 			break;
672 	}
673 	if (i == strlen(ends)) {
674 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
675 		    buf);
676 		usage(B_FALSE);
677 	}
678 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
679 		return (10*i);
680 	}
681 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
682 	usage(B_FALSE);
683 }
684 
685 static uint64_t
686 nicenumtoull(const char *buf)
687 {
688 	char *end;
689 	uint64_t val;
690 
691 	val = strtoull(buf, &end, 0);
692 	if (end == buf) {
693 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
694 		usage(B_FALSE);
695 	} else if (end[0] == '.') {
696 		double fval = strtod(buf, &end);
697 		fval *= pow(2, str2shift(end));
698 		/*
699 		 * UINT64_MAX is not exactly representable as a double.
700 		 * The closest representation is UINT64_MAX + 1, so we
701 		 * use a >= comparison instead of > for the bounds check.
702 		 */
703 		if (fval >= (double)UINT64_MAX) {
704 			(void) fprintf(stderr, "ztest: value too large: %s\n",
705 			    buf);
706 			usage(B_FALSE);
707 		}
708 		val = (uint64_t)fval;
709 	} else {
710 		int shift = str2shift(end);
711 		if (shift >= 64 || (val << shift) >> shift != val) {
712 			(void) fprintf(stderr, "ztest: value too large: %s\n",
713 			    buf);
714 			usage(B_FALSE);
715 		}
716 		val <<= shift;
717 	}
718 	return (val);
719 }
720 
721 typedef struct ztest_option {
722 	const char	short_opt;
723 	const char	*long_opt;
724 	const char	*long_opt_param;
725 	const char	*comment;
726 	unsigned int	default_int;
727 	const char	*default_str;
728 } ztest_option_t;
729 
730 /*
731  * The following option_table is used for generating the usage info as well as
732  * the long and short option information for calling getopt_long().
733  */
734 static ztest_option_t option_table[] = {
735 	{ 'v',	"vdevs", "INTEGER", "Number of vdevs", DEFAULT_VDEV_COUNT,
736 	    NULL},
737 	{ 's',	"vdev-size", "INTEGER", "Size of each vdev",
738 	    NO_DEFAULT, DEFAULT_VDEV_SIZE_STR},
739 	{ 'a',	"alignment-shift", "INTEGER",
740 	    "Alignment shift; use 0 for random", DEFAULT_ASHIFT, NULL},
741 	{ 'm',	"mirror-copies", "INTEGER", "Number of mirror copies",
742 	    DEFAULT_MIRRORS, NULL},
743 	{ 'r',	"raid-disks", "INTEGER", "Number of raidz/draid disks",
744 	    DEFAULT_RAID_CHILDREN, NULL},
745 	{ 'R',	"raid-parity", "INTEGER", "Raid parity",
746 	    DEFAULT_RAID_PARITY, NULL},
747 	{ 'K',	"raid-kind", "raidz|draid|random", "Raid kind",
748 	    NO_DEFAULT, "random"},
749 	{ 'D',	"draid-data", "INTEGER", "Number of draid data drives",
750 	    DEFAULT_DRAID_DATA, NULL},
751 	{ 'S',	"draid-spares", "INTEGER", "Number of draid spares",
752 	    DEFAULT_DRAID_SPARES, NULL},
753 	{ 'd',	"datasets", "INTEGER", "Number of datasets",
754 	    DEFAULT_DATASETS_COUNT, NULL},
755 	{ 't',	"threads", "INTEGER", "Number of ztest threads",
756 	    DEFAULT_THREADS, NULL},
757 	{ 'g',	"gang-block-threshold", "INTEGER",
758 	    "Metaslab gang block threshold",
759 	    NO_DEFAULT, DEFAULT_FORCE_GANGING_STR},
760 	{ 'i',	"init-count", "INTEGER", "Number of times to initialize pool",
761 	    DEFAULT_INITS, NULL},
762 	{ 'k',	"kill-percentage", "INTEGER", "Kill percentage",
763 	    NO_DEFAULT, DEFAULT_KILLRATE_STR},
764 	{ 'p',	"pool-name", "STRING", "Pool name",
765 	    NO_DEFAULT, DEFAULT_POOL},
766 	{ 'f',	"vdev-file-directory", "PATH", "File directory for vdev files",
767 	    NO_DEFAULT, DEFAULT_VDEV_DIR},
768 	{ 'M',	"multi-host", NULL,
769 	    "Multi-host; simulate pool imported on remote host",
770 	    NO_DEFAULT, NULL},
771 	{ 'E',	"use-existing-pool", NULL,
772 	    "Use existing pool instead of creating new one", NO_DEFAULT, NULL},
773 	{ 'T',	"run-time", "INTEGER", "Total run time",
774 	    NO_DEFAULT, DEFAULT_RUN_TIME_STR},
775 	{ 'P',	"pass-time", "INTEGER", "Time per pass",
776 	    NO_DEFAULT, DEFAULT_PASS_TIME_STR},
777 	{ 'F',	"freeze-loops", "INTEGER", "Max loops in spa_freeze()",
778 	    DEFAULT_MAX_LOOPS, NULL},
779 	{ 'B',	"alt-ztest", "PATH", "Alternate ztest path",
780 	    NO_DEFAULT, NULL},
781 	{ 'C',	"vdev-class-state", "on|off|random", "vdev class state",
782 	    NO_DEFAULT, "random"},
783 	{ 'o',	"option", "\"OPTION=INTEGER\"",
784 	    "Set global variable to an unsigned 32-bit integer value",
785 	    NO_DEFAULT, NULL},
786 	{ 'G',	"dump-debug-msg", NULL,
787 	    "Dump zfs_dbgmsg buffer before exiting due to an error",
788 	    NO_DEFAULT, NULL},
789 	{ 'V',	"verbose", NULL,
790 	    "Verbose (use multiple times for ever more verbosity)",
791 	    NO_DEFAULT, NULL},
792 	{ 'h',	"help",	NULL, "Show this help",
793 	    NO_DEFAULT, NULL},
794 	{0, 0, 0, 0, 0, 0}
795 };
796 
797 static struct option *long_opts = NULL;
798 static char *short_opts = NULL;
799 
800 static void
801 init_options(void)
802 {
803 	ASSERT3P(long_opts, ==, NULL);
804 	ASSERT3P(short_opts, ==, NULL);
805 
806 	int count = sizeof (option_table) / sizeof (option_table[0]);
807 	long_opts = umem_alloc(sizeof (struct option) * count, UMEM_NOFAIL);
808 
809 	short_opts = umem_alloc(sizeof (char) * 2 * count, UMEM_NOFAIL);
810 	int short_opt_index = 0;
811 
812 	for (int i = 0; i < count; i++) {
813 		long_opts[i].val = option_table[i].short_opt;
814 		long_opts[i].name = option_table[i].long_opt;
815 		long_opts[i].has_arg = option_table[i].long_opt_param != NULL
816 		    ? required_argument : no_argument;
817 		long_opts[i].flag = NULL;
818 		short_opts[short_opt_index++] = option_table[i].short_opt;
819 		if (option_table[i].long_opt_param != NULL) {
820 			short_opts[short_opt_index++] = ':';
821 		}
822 	}
823 }
824 
825 static void
826 fini_options(void)
827 {
828 	int count = sizeof (option_table) / sizeof (option_table[0]);
829 
830 	umem_free(long_opts, sizeof (struct option) * count);
831 	umem_free(short_opts, sizeof (char) * 2 * count);
832 
833 	long_opts = NULL;
834 	short_opts = NULL;
835 }
836 
837 static __attribute__((noreturn)) void
838 usage(boolean_t requested)
839 {
840 	char option[80];
841 	FILE *fp = requested ? stdout : stderr;
842 
843 	(void) fprintf(fp, "Usage: %s [OPTIONS...]\n", DEFAULT_POOL);
844 	for (int i = 0; option_table[i].short_opt != 0; i++) {
845 		if (option_table[i].long_opt_param != NULL) {
846 			(void) sprintf(option, "  -%c --%s=%s",
847 			    option_table[i].short_opt,
848 			    option_table[i].long_opt,
849 			    option_table[i].long_opt_param);
850 		} else {
851 			(void) sprintf(option, "  -%c --%s",
852 			    option_table[i].short_opt,
853 			    option_table[i].long_opt);
854 		}
855 		(void) fprintf(fp, "  %-40s%s", option,
856 		    option_table[i].comment);
857 
858 		if (option_table[i].long_opt_param != NULL) {
859 			if (option_table[i].default_str != NULL) {
860 				(void) fprintf(fp, " (default: %s)",
861 				    option_table[i].default_str);
862 			} else if (option_table[i].default_int != NO_DEFAULT) {
863 				(void) fprintf(fp, " (default: %u)",
864 				    option_table[i].default_int);
865 			}
866 		}
867 		(void) fprintf(fp, "\n");
868 	}
869 	exit(requested ? 0 : 1);
870 }
871 
872 static uint64_t
873 ztest_random(uint64_t range)
874 {
875 	uint64_t r;
876 
877 	ASSERT3S(ztest_fd_rand, >=, 0);
878 
879 	if (range == 0)
880 		return (0);
881 
882 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
883 		fatal(B_TRUE, "short read from /dev/urandom");
884 
885 	return (r % range);
886 }
887 
888 static void
889 ztest_parse_name_value(const char *input, ztest_shared_opts_t *zo)
890 {
891 	char name[32];
892 	char *value;
893 	int state = ZTEST_VDEV_CLASS_RND;
894 
895 	(void) strlcpy(name, input, sizeof (name));
896 
897 	value = strchr(name, '=');
898 	if (value == NULL) {
899 		(void) fprintf(stderr, "missing value in property=value "
900 		    "'-C' argument (%s)\n", input);
901 		usage(B_FALSE);
902 	}
903 	*(value) = '\0';
904 	value++;
905 
906 	if (strcmp(value, "on") == 0) {
907 		state = ZTEST_VDEV_CLASS_ON;
908 	} else if (strcmp(value, "off") == 0) {
909 		state = ZTEST_VDEV_CLASS_OFF;
910 	} else if (strcmp(value, "random") == 0) {
911 		state = ZTEST_VDEV_CLASS_RND;
912 	} else {
913 		(void) fprintf(stderr, "invalid property value '%s'\n", value);
914 		usage(B_FALSE);
915 	}
916 
917 	if (strcmp(name, "special") == 0) {
918 		zo->zo_special_vdevs = state;
919 	} else {
920 		(void) fprintf(stderr, "invalid property name '%s'\n", name);
921 		usage(B_FALSE);
922 	}
923 	if (zo->zo_verbose >= 3)
924 		(void) printf("%s vdev state is '%s'\n", name, value);
925 }
926 
927 static void
928 process_options(int argc, char **argv)
929 {
930 	char *path;
931 	ztest_shared_opts_t *zo = &ztest_opts;
932 
933 	int opt;
934 	uint64_t value;
935 	const char *raid_kind = "random";
936 
937 	memcpy(zo, &ztest_opts_defaults, sizeof (*zo));
938 
939 	init_options();
940 
941 	while ((opt = getopt_long(argc, argv, short_opts, long_opts,
942 	    NULL)) != EOF) {
943 		value = 0;
944 		switch (opt) {
945 		case 'v':
946 		case 's':
947 		case 'a':
948 		case 'm':
949 		case 'r':
950 		case 'R':
951 		case 'D':
952 		case 'S':
953 		case 'd':
954 		case 't':
955 		case 'g':
956 		case 'i':
957 		case 'k':
958 		case 'T':
959 		case 'P':
960 		case 'F':
961 			value = nicenumtoull(optarg);
962 		}
963 		switch (opt) {
964 		case 'v':
965 			zo->zo_vdevs = value;
966 			break;
967 		case 's':
968 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
969 			break;
970 		case 'a':
971 			zo->zo_ashift = value;
972 			break;
973 		case 'm':
974 			zo->zo_mirrors = value;
975 			break;
976 		case 'r':
977 			zo->zo_raid_children = MAX(1, value);
978 			break;
979 		case 'R':
980 			zo->zo_raid_parity = MIN(MAX(value, 1), 3);
981 			break;
982 		case 'K':
983 			raid_kind = optarg;
984 			break;
985 		case 'D':
986 			zo->zo_draid_data = MAX(1, value);
987 			break;
988 		case 'S':
989 			zo->zo_draid_spares = MAX(1, value);
990 			break;
991 		case 'd':
992 			zo->zo_datasets = MAX(1, value);
993 			break;
994 		case 't':
995 			zo->zo_threads = MAX(1, value);
996 			break;
997 		case 'g':
998 			zo->zo_metaslab_force_ganging =
999 			    MAX(SPA_MINBLOCKSIZE << 1, value);
1000 			break;
1001 		case 'i':
1002 			zo->zo_init = value;
1003 			break;
1004 		case 'k':
1005 			zo->zo_killrate = value;
1006 			break;
1007 		case 'p':
1008 			(void) strlcpy(zo->zo_pool, optarg,
1009 			    sizeof (zo->zo_pool));
1010 			break;
1011 		case 'f':
1012 			path = realpath(optarg, NULL);
1013 			if (path == NULL) {
1014 				(void) fprintf(stderr, "error: %s: %s\n",
1015 				    optarg, strerror(errno));
1016 				usage(B_FALSE);
1017 			} else {
1018 				(void) strlcpy(zo->zo_dir, path,
1019 				    sizeof (zo->zo_dir));
1020 				free(path);
1021 			}
1022 			break;
1023 		case 'M':
1024 			zo->zo_mmp_test = 1;
1025 			break;
1026 		case 'V':
1027 			zo->zo_verbose++;
1028 			break;
1029 		case 'E':
1030 			zo->zo_init = 0;
1031 			break;
1032 		case 'T':
1033 			zo->zo_time = value;
1034 			break;
1035 		case 'P':
1036 			zo->zo_passtime = MAX(1, value);
1037 			break;
1038 		case 'F':
1039 			zo->zo_maxloops = MAX(1, value);
1040 			break;
1041 		case 'B':
1042 			(void) strlcpy(zo->zo_alt_ztest, optarg,
1043 			    sizeof (zo->zo_alt_ztest));
1044 			break;
1045 		case 'C':
1046 			ztest_parse_name_value(optarg, zo);
1047 			break;
1048 		case 'o':
1049 			if (zo->zo_gvars_count >= ZO_GVARS_MAX_COUNT) {
1050 				(void) fprintf(stderr,
1051 				    "max global var count (%zu) exceeded\n",
1052 				    ZO_GVARS_MAX_COUNT);
1053 				usage(B_FALSE);
1054 			}
1055 			char *v = zo->zo_gvars[zo->zo_gvars_count];
1056 			if (strlcpy(v, optarg, ZO_GVARS_MAX_ARGLEN) >=
1057 			    ZO_GVARS_MAX_ARGLEN) {
1058 				(void) fprintf(stderr,
1059 				    "global var option '%s' is too long\n",
1060 				    optarg);
1061 				usage(B_FALSE);
1062 			}
1063 			zo->zo_gvars_count++;
1064 			break;
1065 		case 'G':
1066 			zo->zo_dump_dbgmsg = 1;
1067 			break;
1068 		case 'h':
1069 			usage(B_TRUE);
1070 			break;
1071 		case '?':
1072 		default:
1073 			usage(B_FALSE);
1074 			break;
1075 		}
1076 	}
1077 
1078 	fini_options();
1079 
1080 	/* When raid choice is 'random' add a draid pool 50% of the time */
1081 	if (strcmp(raid_kind, "random") == 0) {
1082 		raid_kind = (ztest_random(2) == 0) ? "draid" : "raidz";
1083 
1084 		if (ztest_opts.zo_verbose >= 3)
1085 			(void) printf("choosing RAID type '%s'\n", raid_kind);
1086 	}
1087 
1088 	if (strcmp(raid_kind, "draid") == 0) {
1089 		uint64_t min_devsize;
1090 
1091 		/* With fewer disk use 256M, otherwise 128M is OK */
1092 		min_devsize = (ztest_opts.zo_raid_children < 16) ?
1093 		    (256ULL << 20) : (128ULL << 20);
1094 
1095 		/* No top-level mirrors with dRAID for now */
1096 		zo->zo_mirrors = 0;
1097 
1098 		/* Use more appropriate defaults for dRAID */
1099 		if (zo->zo_vdevs == ztest_opts_defaults.zo_vdevs)
1100 			zo->zo_vdevs = 1;
1101 		if (zo->zo_raid_children ==
1102 		    ztest_opts_defaults.zo_raid_children)
1103 			zo->zo_raid_children = 16;
1104 		if (zo->zo_ashift < 12)
1105 			zo->zo_ashift = 12;
1106 		if (zo->zo_vdev_size < min_devsize)
1107 			zo->zo_vdev_size = min_devsize;
1108 
1109 		if (zo->zo_draid_data + zo->zo_raid_parity >
1110 		    zo->zo_raid_children - zo->zo_draid_spares) {
1111 			(void) fprintf(stderr, "error: too few draid "
1112 			    "children (%d) for stripe width (%d)\n",
1113 			    zo->zo_raid_children,
1114 			    zo->zo_draid_data + zo->zo_raid_parity);
1115 			usage(B_FALSE);
1116 		}
1117 
1118 		(void) strlcpy(zo->zo_raid_type, VDEV_TYPE_DRAID,
1119 		    sizeof (zo->zo_raid_type));
1120 
1121 	} else /* using raidz */ {
1122 		ASSERT0(strcmp(raid_kind, "raidz"));
1123 
1124 		zo->zo_raid_parity = MIN(zo->zo_raid_parity,
1125 		    zo->zo_raid_children - 1);
1126 	}
1127 
1128 	zo->zo_vdevtime =
1129 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
1130 	    UINT64_MAX >> 2);
1131 
1132 	if (*zo->zo_alt_ztest) {
1133 		const char *invalid_what = "ztest";
1134 		char *val = zo->zo_alt_ztest;
1135 		if (0 != access(val, X_OK) ||
1136 		    (strrchr(val, '/') == NULL && (errno = EINVAL)))
1137 			goto invalid;
1138 
1139 		int dirlen = strrchr(val, '/') - val;
1140 		strncpy(zo->zo_alt_libpath, val, dirlen);
1141 		invalid_what = "library path", val = zo->zo_alt_libpath;
1142 		if (strrchr(val, '/') == NULL && (errno = EINVAL))
1143 			goto invalid;
1144 		*strrchr(val, '/') = '\0';
1145 		strlcat(val, "/lib", sizeof (zo->zo_alt_libpath));
1146 
1147 		if (0 != access(zo->zo_alt_libpath, X_OK))
1148 			goto invalid;
1149 		return;
1150 
1151 invalid:
1152 		ztest_dump_core = B_FALSE;
1153 		fatal(B_TRUE, "invalid alternate %s %s", invalid_what, val);
1154 	}
1155 }
1156 
1157 static void
1158 ztest_kill(ztest_shared_t *zs)
1159 {
1160 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
1161 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
1162 
1163 	/*
1164 	 * Before we kill ourselves, make sure that the config is updated.
1165 	 * See comment above spa_write_cachefile().
1166 	 */
1167 	mutex_enter(&spa_namespace_lock);
1168 	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
1169 	mutex_exit(&spa_namespace_lock);
1170 
1171 	(void) raise(SIGKILL);
1172 }
1173 
1174 static void
1175 ztest_record_enospc(const char *s)
1176 {
1177 	(void) s;
1178 	ztest_shared->zs_enospc_count++;
1179 }
1180 
1181 static uint64_t
1182 ztest_get_ashift(void)
1183 {
1184 	if (ztest_opts.zo_ashift == 0)
1185 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
1186 	return (ztest_opts.zo_ashift);
1187 }
1188 
1189 static boolean_t
1190 ztest_is_draid_spare(const char *name)
1191 {
1192 	uint64_t spare_id = 0, parity = 0, vdev_id = 0;
1193 
1194 	if (sscanf(name, VDEV_TYPE_DRAID "%"PRIu64"-%"PRIu64"-%"PRIu64"",
1195 	    &parity, &vdev_id, &spare_id) == 3) {
1196 		return (B_TRUE);
1197 	}
1198 
1199 	return (B_FALSE);
1200 }
1201 
1202 static nvlist_t *
1203 make_vdev_file(const char *path, const char *aux, const char *pool,
1204     size_t size, uint64_t ashift)
1205 {
1206 	char *pathbuf = NULL;
1207 	uint64_t vdev;
1208 	nvlist_t *file;
1209 	boolean_t draid_spare = B_FALSE;
1210 
1211 
1212 	if (ashift == 0)
1213 		ashift = ztest_get_ashift();
1214 
1215 	if (path == NULL) {
1216 		pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1217 		path = pathbuf;
1218 
1219 		if (aux != NULL) {
1220 			vdev = ztest_shared->zs_vdev_aux;
1221 			(void) snprintf(pathbuf, MAXPATHLEN,
1222 			    ztest_aux_template, ztest_opts.zo_dir,
1223 			    pool == NULL ? ztest_opts.zo_pool : pool,
1224 			    aux, vdev);
1225 		} else {
1226 			vdev = ztest_shared->zs_vdev_next_leaf++;
1227 			(void) snprintf(pathbuf, MAXPATHLEN,
1228 			    ztest_dev_template, ztest_opts.zo_dir,
1229 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
1230 		}
1231 	} else {
1232 		draid_spare = ztest_is_draid_spare(path);
1233 	}
1234 
1235 	if (size != 0 && !draid_spare) {
1236 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
1237 		if (fd == -1)
1238 			fatal(B_TRUE, "can't open %s", path);
1239 		if (ftruncate(fd, size) != 0)
1240 			fatal(B_TRUE, "can't ftruncate %s", path);
1241 		(void) close(fd);
1242 	}
1243 
1244 	file = fnvlist_alloc();
1245 	fnvlist_add_string(file, ZPOOL_CONFIG_TYPE,
1246 	    draid_spare ? VDEV_TYPE_DRAID_SPARE : VDEV_TYPE_FILE);
1247 	fnvlist_add_string(file, ZPOOL_CONFIG_PATH, path);
1248 	fnvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift);
1249 	umem_free(pathbuf, MAXPATHLEN);
1250 
1251 	return (file);
1252 }
1253 
1254 static nvlist_t *
1255 make_vdev_raid(const char *path, const char *aux, const char *pool, size_t size,
1256     uint64_t ashift, int r)
1257 {
1258 	nvlist_t *raid, **child;
1259 	int c;
1260 
1261 	if (r < 2)
1262 		return (make_vdev_file(path, aux, pool, size, ashift));
1263 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
1264 
1265 	for (c = 0; c < r; c++)
1266 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
1267 
1268 	raid = fnvlist_alloc();
1269 	fnvlist_add_string(raid, ZPOOL_CONFIG_TYPE,
1270 	    ztest_opts.zo_raid_type);
1271 	fnvlist_add_uint64(raid, ZPOOL_CONFIG_NPARITY,
1272 	    ztest_opts.zo_raid_parity);
1273 	fnvlist_add_nvlist_array(raid, ZPOOL_CONFIG_CHILDREN,
1274 	    (const nvlist_t **)child, r);
1275 
1276 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0) {
1277 		uint64_t ndata = ztest_opts.zo_draid_data;
1278 		uint64_t nparity = ztest_opts.zo_raid_parity;
1279 		uint64_t nspares = ztest_opts.zo_draid_spares;
1280 		uint64_t children = ztest_opts.zo_raid_children;
1281 		uint64_t ngroups = 1;
1282 
1283 		/*
1284 		 * Calculate the minimum number of groups required to fill a
1285 		 * slice. This is the LCM of the stripe width (data + parity)
1286 		 * and the number of data drives (children - spares).
1287 		 */
1288 		while (ngroups * (ndata + nparity) % (children - nspares) != 0)
1289 			ngroups++;
1290 
1291 		/* Store the basic dRAID configuration. */
1292 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NDATA, ndata);
1293 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NSPARES, nspares);
1294 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NGROUPS, ngroups);
1295 	}
1296 
1297 	for (c = 0; c < r; c++)
1298 		fnvlist_free(child[c]);
1299 
1300 	umem_free(child, r * sizeof (nvlist_t *));
1301 
1302 	return (raid);
1303 }
1304 
1305 static nvlist_t *
1306 make_vdev_mirror(const char *path, const char *aux, const char *pool,
1307     size_t size, uint64_t ashift, int r, int m)
1308 {
1309 	nvlist_t *mirror, **child;
1310 	int c;
1311 
1312 	if (m < 1)
1313 		return (make_vdev_raid(path, aux, pool, size, ashift, r));
1314 
1315 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
1316 
1317 	for (c = 0; c < m; c++)
1318 		child[c] = make_vdev_raid(path, aux, pool, size, ashift, r);
1319 
1320 	mirror = fnvlist_alloc();
1321 	fnvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, VDEV_TYPE_MIRROR);
1322 	fnvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
1323 	    (const nvlist_t **)child, m);
1324 
1325 	for (c = 0; c < m; c++)
1326 		fnvlist_free(child[c]);
1327 
1328 	umem_free(child, m * sizeof (nvlist_t *));
1329 
1330 	return (mirror);
1331 }
1332 
1333 static nvlist_t *
1334 make_vdev_root(const char *path, const char *aux, const char *pool, size_t size,
1335     uint64_t ashift, const char *class, int r, int m, int t)
1336 {
1337 	nvlist_t *root, **child;
1338 	int c;
1339 	boolean_t log;
1340 
1341 	ASSERT3S(t, >, 0);
1342 
1343 	log = (class != NULL && strcmp(class, "log") == 0);
1344 
1345 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1346 
1347 	for (c = 0; c < t; c++) {
1348 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1349 		    r, m);
1350 		fnvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, log);
1351 
1352 		if (class != NULL && class[0] != '\0') {
1353 			ASSERT(m > 1 || log);   /* expecting a mirror */
1354 			fnvlist_add_string(child[c],
1355 			    ZPOOL_CONFIG_ALLOCATION_BIAS, class);
1356 		}
1357 	}
1358 
1359 	root = fnvlist_alloc();
1360 	fnvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
1361 	fnvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1362 	    (const nvlist_t **)child, t);
1363 
1364 	for (c = 0; c < t; c++)
1365 		fnvlist_free(child[c]);
1366 
1367 	umem_free(child, t * sizeof (nvlist_t *));
1368 
1369 	return (root);
1370 }
1371 
1372 /*
1373  * Find a random spa version. Returns back a random spa version in the
1374  * range [initial_version, SPA_VERSION_FEATURES].
1375  */
1376 static uint64_t
1377 ztest_random_spa_version(uint64_t initial_version)
1378 {
1379 	uint64_t version = initial_version;
1380 
1381 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
1382 		version = version +
1383 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1384 	}
1385 
1386 	if (version > SPA_VERSION_BEFORE_FEATURES)
1387 		version = SPA_VERSION_FEATURES;
1388 
1389 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1390 	return (version);
1391 }
1392 
1393 static int
1394 ztest_random_blocksize(void)
1395 {
1396 	ASSERT3U(ztest_spa->spa_max_ashift, !=, 0);
1397 
1398 	/*
1399 	 * Choose a block size >= the ashift.
1400 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1401 	 */
1402 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1403 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1404 		maxbs = 20;
1405 	uint64_t block_shift =
1406 	    ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1407 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1408 }
1409 
1410 static int
1411 ztest_random_dnodesize(void)
1412 {
1413 	int slots;
1414 	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1415 
1416 	if (max_slots == DNODE_MIN_SLOTS)
1417 		return (DNODE_MIN_SIZE);
1418 
1419 	/*
1420 	 * Weight the random distribution more heavily toward smaller
1421 	 * dnode sizes since that is more likely to reflect real-world
1422 	 * usage.
1423 	 */
1424 	ASSERT3U(max_slots, >, 4);
1425 	switch (ztest_random(10)) {
1426 	case 0:
1427 		slots = 5 + ztest_random(max_slots - 4);
1428 		break;
1429 	case 1 ... 4:
1430 		slots = 2 + ztest_random(3);
1431 		break;
1432 	default:
1433 		slots = 1;
1434 		break;
1435 	}
1436 
1437 	return (slots << DNODE_SHIFT);
1438 }
1439 
1440 static int
1441 ztest_random_ibshift(void)
1442 {
1443 	return (DN_MIN_INDBLKSHIFT +
1444 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1445 }
1446 
1447 static uint64_t
1448 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1449 {
1450 	uint64_t top;
1451 	vdev_t *rvd = spa->spa_root_vdev;
1452 	vdev_t *tvd;
1453 
1454 	ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
1455 
1456 	do {
1457 		top = ztest_random(rvd->vdev_children);
1458 		tvd = rvd->vdev_child[top];
1459 	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1460 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1461 
1462 	return (top);
1463 }
1464 
1465 static uint64_t
1466 ztest_random_dsl_prop(zfs_prop_t prop)
1467 {
1468 	uint64_t value;
1469 
1470 	do {
1471 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1472 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1473 
1474 	return (value);
1475 }
1476 
1477 static int
1478 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1479     boolean_t inherit)
1480 {
1481 	const char *propname = zfs_prop_to_name(prop);
1482 	const char *valname;
1483 	char *setpoint;
1484 	uint64_t curval;
1485 	int error;
1486 
1487 	error = dsl_prop_set_int(osname, propname,
1488 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1489 
1490 	if (error == ENOSPC) {
1491 		ztest_record_enospc(FTAG);
1492 		return (error);
1493 	}
1494 	ASSERT0(error);
1495 
1496 	setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1497 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1498 
1499 	if (ztest_opts.zo_verbose >= 6) {
1500 		int err;
1501 
1502 		err = zfs_prop_index_to_string(prop, curval, &valname);
1503 		if (err)
1504 			(void) printf("%s %s = %llu at '%s'\n", osname,
1505 			    propname, (unsigned long long)curval, setpoint);
1506 		else
1507 			(void) printf("%s %s = %s at '%s'\n",
1508 			    osname, propname, valname, setpoint);
1509 	}
1510 	umem_free(setpoint, MAXPATHLEN);
1511 
1512 	return (error);
1513 }
1514 
1515 static int
1516 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1517 {
1518 	spa_t *spa = ztest_spa;
1519 	nvlist_t *props = NULL;
1520 	int error;
1521 
1522 	props = fnvlist_alloc();
1523 	fnvlist_add_uint64(props, zpool_prop_to_name(prop), value);
1524 
1525 	error = spa_prop_set(spa, props);
1526 
1527 	fnvlist_free(props);
1528 
1529 	if (error == ENOSPC) {
1530 		ztest_record_enospc(FTAG);
1531 		return (error);
1532 	}
1533 	ASSERT0(error);
1534 
1535 	return (error);
1536 }
1537 
1538 static int
1539 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1540     boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp)
1541 {
1542 	int err;
1543 	char *cp = NULL;
1544 	char ddname[ZFS_MAX_DATASET_NAME_LEN];
1545 
1546 	strcpy(ddname, name);
1547 	cp = strchr(ddname, '@');
1548 	if (cp != NULL)
1549 		*cp = '\0';
1550 
1551 	err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1552 	while (decrypt && err == EACCES) {
1553 		dsl_crypto_params_t *dcp;
1554 		nvlist_t *crypto_args = fnvlist_alloc();
1555 
1556 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
1557 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1558 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1559 		    crypto_args, &dcp));
1560 		err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1561 		/*
1562 		 * Note: if there was an error loading, the wkey was not
1563 		 * consumed, and needs to be freed.
1564 		 */
1565 		dsl_crypto_params_free(dcp, (err != 0));
1566 		fnvlist_free(crypto_args);
1567 
1568 		if (err == EINVAL) {
1569 			/*
1570 			 * We couldn't load a key for this dataset so try
1571 			 * the parent. This loop will eventually hit the
1572 			 * encryption root since ztest only makes clones
1573 			 * as children of their origin datasets.
1574 			 */
1575 			cp = strrchr(ddname, '/');
1576 			if (cp == NULL)
1577 				return (err);
1578 
1579 			*cp = '\0';
1580 			err = EACCES;
1581 			continue;
1582 		} else if (err != 0) {
1583 			break;
1584 		}
1585 
1586 		err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1587 		break;
1588 	}
1589 
1590 	return (err);
1591 }
1592 
1593 static void
1594 ztest_rll_init(rll_t *rll)
1595 {
1596 	rll->rll_writer = NULL;
1597 	rll->rll_readers = 0;
1598 	mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1599 	cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1600 }
1601 
1602 static void
1603 ztest_rll_destroy(rll_t *rll)
1604 {
1605 	ASSERT3P(rll->rll_writer, ==, NULL);
1606 	ASSERT0(rll->rll_readers);
1607 	mutex_destroy(&rll->rll_lock);
1608 	cv_destroy(&rll->rll_cv);
1609 }
1610 
1611 static void
1612 ztest_rll_lock(rll_t *rll, rl_type_t type)
1613 {
1614 	mutex_enter(&rll->rll_lock);
1615 
1616 	if (type == RL_READER) {
1617 		while (rll->rll_writer != NULL)
1618 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1619 		rll->rll_readers++;
1620 	} else {
1621 		while (rll->rll_writer != NULL || rll->rll_readers)
1622 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1623 		rll->rll_writer = curthread;
1624 	}
1625 
1626 	mutex_exit(&rll->rll_lock);
1627 }
1628 
1629 static void
1630 ztest_rll_unlock(rll_t *rll)
1631 {
1632 	mutex_enter(&rll->rll_lock);
1633 
1634 	if (rll->rll_writer) {
1635 		ASSERT0(rll->rll_readers);
1636 		rll->rll_writer = NULL;
1637 	} else {
1638 		ASSERT3S(rll->rll_readers, >, 0);
1639 		ASSERT3P(rll->rll_writer, ==, NULL);
1640 		rll->rll_readers--;
1641 	}
1642 
1643 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1644 		cv_broadcast(&rll->rll_cv);
1645 
1646 	mutex_exit(&rll->rll_lock);
1647 }
1648 
1649 static void
1650 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1651 {
1652 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1653 
1654 	ztest_rll_lock(rll, type);
1655 }
1656 
1657 static void
1658 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1659 {
1660 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1661 
1662 	ztest_rll_unlock(rll);
1663 }
1664 
1665 static rl_t *
1666 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1667     uint64_t size, rl_type_t type)
1668 {
1669 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1670 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1671 	rl_t *rl;
1672 
1673 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1674 	rl->rl_object = object;
1675 	rl->rl_offset = offset;
1676 	rl->rl_size = size;
1677 	rl->rl_lock = rll;
1678 
1679 	ztest_rll_lock(rll, type);
1680 
1681 	return (rl);
1682 }
1683 
1684 static void
1685 ztest_range_unlock(rl_t *rl)
1686 {
1687 	rll_t *rll = rl->rl_lock;
1688 
1689 	ztest_rll_unlock(rll);
1690 
1691 	umem_free(rl, sizeof (*rl));
1692 }
1693 
1694 static void
1695 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1696 {
1697 	zd->zd_os = os;
1698 	zd->zd_zilog = dmu_objset_zil(os);
1699 	zd->zd_shared = szd;
1700 	dmu_objset_name(os, zd->zd_name);
1701 	int l;
1702 
1703 	if (zd->zd_shared != NULL)
1704 		zd->zd_shared->zd_seq = 0;
1705 
1706 	VERIFY0(pthread_rwlock_init(&zd->zd_zilog_lock, NULL));
1707 	mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1708 
1709 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1710 		ztest_rll_init(&zd->zd_object_lock[l]);
1711 
1712 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1713 		ztest_rll_init(&zd->zd_range_lock[l]);
1714 }
1715 
1716 static void
1717 ztest_zd_fini(ztest_ds_t *zd)
1718 {
1719 	int l;
1720 
1721 	mutex_destroy(&zd->zd_dirobj_lock);
1722 	(void) pthread_rwlock_destroy(&zd->zd_zilog_lock);
1723 
1724 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1725 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1726 
1727 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1728 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1729 }
1730 
1731 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1732 
1733 static uint64_t
1734 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1735 {
1736 	uint64_t txg;
1737 	int error;
1738 
1739 	/*
1740 	 * Attempt to assign tx to some transaction group.
1741 	 */
1742 	error = dmu_tx_assign(tx, txg_how);
1743 	if (error) {
1744 		if (error == ERESTART) {
1745 			ASSERT3U(txg_how, ==, TXG_NOWAIT);
1746 			dmu_tx_wait(tx);
1747 		} else {
1748 			ASSERT3U(error, ==, ENOSPC);
1749 			ztest_record_enospc(tag);
1750 		}
1751 		dmu_tx_abort(tx);
1752 		return (0);
1753 	}
1754 	txg = dmu_tx_get_txg(tx);
1755 	ASSERT3U(txg, !=, 0);
1756 	return (txg);
1757 }
1758 
1759 static void
1760 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1761     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1762     uint64_t crtxg)
1763 {
1764 	bt->bt_magic = BT_MAGIC;
1765 	bt->bt_objset = dmu_objset_id(os);
1766 	bt->bt_object = object;
1767 	bt->bt_dnodesize = dnodesize;
1768 	bt->bt_offset = offset;
1769 	bt->bt_gen = gen;
1770 	bt->bt_txg = txg;
1771 	bt->bt_crtxg = crtxg;
1772 }
1773 
1774 static void
1775 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1776     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1777     uint64_t crtxg)
1778 {
1779 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1780 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1781 	ASSERT3U(bt->bt_object, ==, object);
1782 	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1783 	ASSERT3U(bt->bt_offset, ==, offset);
1784 	ASSERT3U(bt->bt_gen, <=, gen);
1785 	ASSERT3U(bt->bt_txg, <=, txg);
1786 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1787 }
1788 
1789 static ztest_block_tag_t *
1790 ztest_bt_bonus(dmu_buf_t *db)
1791 {
1792 	dmu_object_info_t doi;
1793 	ztest_block_tag_t *bt;
1794 
1795 	dmu_object_info_from_db(db, &doi);
1796 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1797 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1798 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1799 
1800 	return (bt);
1801 }
1802 
1803 /*
1804  * Generate a token to fill up unused bonus buffer space.  Try to make
1805  * it unique to the object, generation, and offset to verify that data
1806  * is not getting overwritten by data from other dnodes.
1807  */
1808 #define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1809 	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1810 
1811 /*
1812  * Fill up the unused bonus buffer region before the block tag with a
1813  * verifiable pattern. Filling the whole bonus area with non-zero data
1814  * helps ensure that all dnode traversal code properly skips the
1815  * interior regions of large dnodes.
1816  */
1817 static void
1818 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1819     objset_t *os, uint64_t gen)
1820 {
1821 	uint64_t *bonusp;
1822 
1823 	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1824 
1825 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1826 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1827 		    gen, bonusp - (uint64_t *)db->db_data);
1828 		*bonusp = token;
1829 	}
1830 }
1831 
1832 /*
1833  * Verify that the unused area of a bonus buffer is filled with the
1834  * expected tokens.
1835  */
1836 static void
1837 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1838     objset_t *os, uint64_t gen)
1839 {
1840 	uint64_t *bonusp;
1841 
1842 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1843 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1844 		    gen, bonusp - (uint64_t *)db->db_data);
1845 		VERIFY3U(*bonusp, ==, token);
1846 	}
1847 }
1848 
1849 /*
1850  * ZIL logging ops
1851  */
1852 
1853 #define	lrz_type	lr_mode
1854 #define	lrz_blocksize	lr_uid
1855 #define	lrz_ibshift	lr_gid
1856 #define	lrz_bonustype	lr_rdev
1857 #define	lrz_dnodesize	lr_crtime[1]
1858 
1859 static void
1860 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1861 {
1862 	char *name = (void *)(lr + 1);		/* name follows lr */
1863 	size_t namesize = strlen(name) + 1;
1864 	itx_t *itx;
1865 
1866 	if (zil_replaying(zd->zd_zilog, tx))
1867 		return;
1868 
1869 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1870 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1871 	    sizeof (*lr) + namesize - sizeof (lr_t));
1872 
1873 	zil_itx_assign(zd->zd_zilog, itx, tx);
1874 }
1875 
1876 static void
1877 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1878 {
1879 	char *name = (void *)(lr + 1);		/* name follows lr */
1880 	size_t namesize = strlen(name) + 1;
1881 	itx_t *itx;
1882 
1883 	if (zil_replaying(zd->zd_zilog, tx))
1884 		return;
1885 
1886 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1887 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1888 	    sizeof (*lr) + namesize - sizeof (lr_t));
1889 
1890 	itx->itx_oid = object;
1891 	zil_itx_assign(zd->zd_zilog, itx, tx);
1892 }
1893 
1894 static void
1895 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1896 {
1897 	itx_t *itx;
1898 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1899 
1900 	if (zil_replaying(zd->zd_zilog, tx))
1901 		return;
1902 
1903 	if (lr->lr_length > zil_max_log_data(zd->zd_zilog))
1904 		write_state = WR_INDIRECT;
1905 
1906 	itx = zil_itx_create(TX_WRITE,
1907 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1908 
1909 	if (write_state == WR_COPIED &&
1910 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1911 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1912 		zil_itx_destroy(itx);
1913 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1914 		write_state = WR_NEED_COPY;
1915 	}
1916 	itx->itx_private = zd;
1917 	itx->itx_wr_state = write_state;
1918 	itx->itx_sync = (ztest_random(8) == 0);
1919 
1920 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1921 	    sizeof (*lr) - sizeof (lr_t));
1922 
1923 	zil_itx_assign(zd->zd_zilog, itx, tx);
1924 }
1925 
1926 static void
1927 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1928 {
1929 	itx_t *itx;
1930 
1931 	if (zil_replaying(zd->zd_zilog, tx))
1932 		return;
1933 
1934 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1935 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1936 	    sizeof (*lr) - sizeof (lr_t));
1937 
1938 	itx->itx_sync = B_FALSE;
1939 	zil_itx_assign(zd->zd_zilog, itx, tx);
1940 }
1941 
1942 static void
1943 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1944 {
1945 	itx_t *itx;
1946 
1947 	if (zil_replaying(zd->zd_zilog, tx))
1948 		return;
1949 
1950 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1951 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1952 	    sizeof (*lr) - sizeof (lr_t));
1953 
1954 	itx->itx_sync = B_FALSE;
1955 	zil_itx_assign(zd->zd_zilog, itx, tx);
1956 }
1957 
1958 /*
1959  * ZIL replay ops
1960  */
1961 static int
1962 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1963 {
1964 	ztest_ds_t *zd = arg1;
1965 	lr_create_t *lr = arg2;
1966 	char *name = (void *)(lr + 1);		/* name follows lr */
1967 	objset_t *os = zd->zd_os;
1968 	ztest_block_tag_t *bbt;
1969 	dmu_buf_t *db;
1970 	dmu_tx_t *tx;
1971 	uint64_t txg;
1972 	int error = 0;
1973 	int bonuslen;
1974 
1975 	if (byteswap)
1976 		byteswap_uint64_array(lr, sizeof (*lr));
1977 
1978 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
1979 	ASSERT3S(name[0], !=, '\0');
1980 
1981 	tx = dmu_tx_create(os);
1982 
1983 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1984 
1985 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1986 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1987 	} else {
1988 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1989 	}
1990 
1991 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1992 	if (txg == 0)
1993 		return (ENOSPC);
1994 
1995 	ASSERT3U(dmu_objset_zil(os)->zl_replay, ==, !!lr->lr_foid);
1996 	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1997 
1998 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1999 		if (lr->lr_foid == 0) {
2000 			lr->lr_foid = zap_create_dnsize(os,
2001 			    lr->lrz_type, lr->lrz_bonustype,
2002 			    bonuslen, lr->lrz_dnodesize, tx);
2003 		} else {
2004 			error = zap_create_claim_dnsize(os, lr->lr_foid,
2005 			    lr->lrz_type, lr->lrz_bonustype,
2006 			    bonuslen, lr->lrz_dnodesize, tx);
2007 		}
2008 	} else {
2009 		if (lr->lr_foid == 0) {
2010 			lr->lr_foid = dmu_object_alloc_dnsize(os,
2011 			    lr->lrz_type, 0, lr->lrz_bonustype,
2012 			    bonuslen, lr->lrz_dnodesize, tx);
2013 		} else {
2014 			error = dmu_object_claim_dnsize(os, lr->lr_foid,
2015 			    lr->lrz_type, 0, lr->lrz_bonustype,
2016 			    bonuslen, lr->lrz_dnodesize, tx);
2017 		}
2018 	}
2019 
2020 	if (error) {
2021 		ASSERT3U(error, ==, EEXIST);
2022 		ASSERT(zd->zd_zilog->zl_replay);
2023 		dmu_tx_commit(tx);
2024 		return (error);
2025 	}
2026 
2027 	ASSERT3U(lr->lr_foid, !=, 0);
2028 
2029 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
2030 		VERIFY0(dmu_object_set_blocksize(os, lr->lr_foid,
2031 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
2032 
2033 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2034 	bbt = ztest_bt_bonus(db);
2035 	dmu_buf_will_dirty(db, tx);
2036 	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
2037 	    lr->lr_gen, txg, txg);
2038 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
2039 	dmu_buf_rele(db, FTAG);
2040 
2041 	VERIFY0(zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
2042 	    &lr->lr_foid, tx));
2043 
2044 	(void) ztest_log_create(zd, tx, lr);
2045 
2046 	dmu_tx_commit(tx);
2047 
2048 	return (0);
2049 }
2050 
2051 static int
2052 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
2053 {
2054 	ztest_ds_t *zd = arg1;
2055 	lr_remove_t *lr = arg2;
2056 	char *name = (void *)(lr + 1);		/* name follows lr */
2057 	objset_t *os = zd->zd_os;
2058 	dmu_object_info_t doi;
2059 	dmu_tx_t *tx;
2060 	uint64_t object, txg;
2061 
2062 	if (byteswap)
2063 		byteswap_uint64_array(lr, sizeof (*lr));
2064 
2065 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
2066 	ASSERT3S(name[0], !=, '\0');
2067 
2068 	VERIFY0(
2069 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
2070 	ASSERT3U(object, !=, 0);
2071 
2072 	ztest_object_lock(zd, object, RL_WRITER);
2073 
2074 	VERIFY0(dmu_object_info(os, object, &doi));
2075 
2076 	tx = dmu_tx_create(os);
2077 
2078 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
2079 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2080 
2081 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2082 	if (txg == 0) {
2083 		ztest_object_unlock(zd, object);
2084 		return (ENOSPC);
2085 	}
2086 
2087 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
2088 		VERIFY0(zap_destroy(os, object, tx));
2089 	} else {
2090 		VERIFY0(dmu_object_free(os, object, tx));
2091 	}
2092 
2093 	VERIFY0(zap_remove(os, lr->lr_doid, name, tx));
2094 
2095 	(void) ztest_log_remove(zd, tx, lr, object);
2096 
2097 	dmu_tx_commit(tx);
2098 
2099 	ztest_object_unlock(zd, object);
2100 
2101 	return (0);
2102 }
2103 
2104 static int
2105 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
2106 {
2107 	ztest_ds_t *zd = arg1;
2108 	lr_write_t *lr = arg2;
2109 	objset_t *os = zd->zd_os;
2110 	void *data = lr + 1;			/* data follows lr */
2111 	uint64_t offset, length;
2112 	ztest_block_tag_t *bt = data;
2113 	ztest_block_tag_t *bbt;
2114 	uint64_t gen, txg, lrtxg, crtxg;
2115 	dmu_object_info_t doi;
2116 	dmu_tx_t *tx;
2117 	dmu_buf_t *db;
2118 	arc_buf_t *abuf = NULL;
2119 	rl_t *rl;
2120 
2121 	if (byteswap)
2122 		byteswap_uint64_array(lr, sizeof (*lr));
2123 
2124 	offset = lr->lr_offset;
2125 	length = lr->lr_length;
2126 
2127 	/* If it's a dmu_sync() block, write the whole block */
2128 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
2129 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
2130 		if (length < blocksize) {
2131 			offset -= offset % blocksize;
2132 			length = blocksize;
2133 		}
2134 	}
2135 
2136 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
2137 		byteswap_uint64_array(bt, sizeof (*bt));
2138 
2139 	if (bt->bt_magic != BT_MAGIC)
2140 		bt = NULL;
2141 
2142 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
2143 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
2144 
2145 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2146 
2147 	dmu_object_info_from_db(db, &doi);
2148 
2149 	bbt = ztest_bt_bonus(db);
2150 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2151 	gen = bbt->bt_gen;
2152 	crtxg = bbt->bt_crtxg;
2153 	lrtxg = lr->lr_common.lrc_txg;
2154 
2155 	tx = dmu_tx_create(os);
2156 
2157 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
2158 
2159 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
2160 	    P2PHASE(offset, length) == 0)
2161 		abuf = dmu_request_arcbuf(db, length);
2162 
2163 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2164 	if (txg == 0) {
2165 		if (abuf != NULL)
2166 			dmu_return_arcbuf(abuf);
2167 		dmu_buf_rele(db, FTAG);
2168 		ztest_range_unlock(rl);
2169 		ztest_object_unlock(zd, lr->lr_foid);
2170 		return (ENOSPC);
2171 	}
2172 
2173 	if (bt != NULL) {
2174 		/*
2175 		 * Usually, verify the old data before writing new data --
2176 		 * but not always, because we also want to verify correct
2177 		 * behavior when the data was not recently read into cache.
2178 		 */
2179 		ASSERT0(offset % doi.doi_data_block_size);
2180 		if (ztest_random(4) != 0) {
2181 			int prefetch = ztest_random(2) ?
2182 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
2183 			ztest_block_tag_t rbt;
2184 
2185 			VERIFY(dmu_read(os, lr->lr_foid, offset,
2186 			    sizeof (rbt), &rbt, prefetch) == 0);
2187 			if (rbt.bt_magic == BT_MAGIC) {
2188 				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
2189 				    offset, gen, txg, crtxg);
2190 			}
2191 		}
2192 
2193 		/*
2194 		 * Writes can appear to be newer than the bonus buffer because
2195 		 * the ztest_get_data() callback does a dmu_read() of the
2196 		 * open-context data, which may be different than the data
2197 		 * as it was when the write was generated.
2198 		 */
2199 		if (zd->zd_zilog->zl_replay) {
2200 			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
2201 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
2202 			    bt->bt_crtxg);
2203 		}
2204 
2205 		/*
2206 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
2207 		 * so that all of the usual ASSERTs will work.
2208 		 */
2209 		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
2210 		    crtxg);
2211 	}
2212 
2213 	if (abuf == NULL) {
2214 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
2215 	} else {
2216 		memcpy(abuf->b_data, data, length);
2217 		dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx);
2218 	}
2219 
2220 	(void) ztest_log_write(zd, tx, lr);
2221 
2222 	dmu_buf_rele(db, FTAG);
2223 
2224 	dmu_tx_commit(tx);
2225 
2226 	ztest_range_unlock(rl);
2227 	ztest_object_unlock(zd, lr->lr_foid);
2228 
2229 	return (0);
2230 }
2231 
2232 static int
2233 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
2234 {
2235 	ztest_ds_t *zd = arg1;
2236 	lr_truncate_t *lr = arg2;
2237 	objset_t *os = zd->zd_os;
2238 	dmu_tx_t *tx;
2239 	uint64_t txg;
2240 	rl_t *rl;
2241 
2242 	if (byteswap)
2243 		byteswap_uint64_array(lr, sizeof (*lr));
2244 
2245 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
2246 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
2247 	    RL_WRITER);
2248 
2249 	tx = dmu_tx_create(os);
2250 
2251 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
2252 
2253 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2254 	if (txg == 0) {
2255 		ztest_range_unlock(rl);
2256 		ztest_object_unlock(zd, lr->lr_foid);
2257 		return (ENOSPC);
2258 	}
2259 
2260 	VERIFY0(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
2261 	    lr->lr_length, tx));
2262 
2263 	(void) ztest_log_truncate(zd, tx, lr);
2264 
2265 	dmu_tx_commit(tx);
2266 
2267 	ztest_range_unlock(rl);
2268 	ztest_object_unlock(zd, lr->lr_foid);
2269 
2270 	return (0);
2271 }
2272 
2273 static int
2274 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
2275 {
2276 	ztest_ds_t *zd = arg1;
2277 	lr_setattr_t *lr = arg2;
2278 	objset_t *os = zd->zd_os;
2279 	dmu_tx_t *tx;
2280 	dmu_buf_t *db;
2281 	ztest_block_tag_t *bbt;
2282 	uint64_t txg, lrtxg, crtxg, dnodesize;
2283 
2284 	if (byteswap)
2285 		byteswap_uint64_array(lr, sizeof (*lr));
2286 
2287 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
2288 
2289 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2290 
2291 	tx = dmu_tx_create(os);
2292 	dmu_tx_hold_bonus(tx, lr->lr_foid);
2293 
2294 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2295 	if (txg == 0) {
2296 		dmu_buf_rele(db, FTAG);
2297 		ztest_object_unlock(zd, lr->lr_foid);
2298 		return (ENOSPC);
2299 	}
2300 
2301 	bbt = ztest_bt_bonus(db);
2302 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2303 	crtxg = bbt->bt_crtxg;
2304 	lrtxg = lr->lr_common.lrc_txg;
2305 	dnodesize = bbt->bt_dnodesize;
2306 
2307 	if (zd->zd_zilog->zl_replay) {
2308 		ASSERT3U(lr->lr_size, !=, 0);
2309 		ASSERT3U(lr->lr_mode, !=, 0);
2310 		ASSERT3U(lrtxg, !=, 0);
2311 	} else {
2312 		/*
2313 		 * Randomly change the size and increment the generation.
2314 		 */
2315 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2316 		    sizeof (*bbt);
2317 		lr->lr_mode = bbt->bt_gen + 1;
2318 		ASSERT0(lrtxg);
2319 	}
2320 
2321 	/*
2322 	 * Verify that the current bonus buffer is not newer than our txg.
2323 	 */
2324 	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2325 	    MAX(txg, lrtxg), crtxg);
2326 
2327 	dmu_buf_will_dirty(db, tx);
2328 
2329 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2330 	ASSERT3U(lr->lr_size, <=, db->db_size);
2331 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2332 	bbt = ztest_bt_bonus(db);
2333 
2334 	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2335 	    txg, crtxg);
2336 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2337 	dmu_buf_rele(db, FTAG);
2338 
2339 	(void) ztest_log_setattr(zd, tx, lr);
2340 
2341 	dmu_tx_commit(tx);
2342 
2343 	ztest_object_unlock(zd, lr->lr_foid);
2344 
2345 	return (0);
2346 }
2347 
2348 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
2349 	NULL,			/* 0 no such transaction type */
2350 	ztest_replay_create,	/* TX_CREATE */
2351 	NULL,			/* TX_MKDIR */
2352 	NULL,			/* TX_MKXATTR */
2353 	NULL,			/* TX_SYMLINK */
2354 	ztest_replay_remove,	/* TX_REMOVE */
2355 	NULL,			/* TX_RMDIR */
2356 	NULL,			/* TX_LINK */
2357 	NULL,			/* TX_RENAME */
2358 	ztest_replay_write,	/* TX_WRITE */
2359 	ztest_replay_truncate,	/* TX_TRUNCATE */
2360 	ztest_replay_setattr,	/* TX_SETATTR */
2361 	NULL,			/* TX_ACL */
2362 	NULL,			/* TX_CREATE_ACL */
2363 	NULL,			/* TX_CREATE_ATTR */
2364 	NULL,			/* TX_CREATE_ACL_ATTR */
2365 	NULL,			/* TX_MKDIR_ACL */
2366 	NULL,			/* TX_MKDIR_ATTR */
2367 	NULL,			/* TX_MKDIR_ACL_ATTR */
2368 	NULL,			/* TX_WRITE2 */
2369 	NULL,			/* TX_SETSAXATTR */
2370 };
2371 
2372 /*
2373  * ZIL get_data callbacks
2374  */
2375 
2376 static void
2377 ztest_get_done(zgd_t *zgd, int error)
2378 {
2379 	(void) error;
2380 	ztest_ds_t *zd = zgd->zgd_private;
2381 	uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object;
2382 
2383 	if (zgd->zgd_db)
2384 		dmu_buf_rele(zgd->zgd_db, zgd);
2385 
2386 	ztest_range_unlock((rl_t *)zgd->zgd_lr);
2387 	ztest_object_unlock(zd, object);
2388 
2389 	umem_free(zgd, sizeof (*zgd));
2390 }
2391 
2392 static int
2393 ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
2394     struct lwb *lwb, zio_t *zio)
2395 {
2396 	(void) arg2;
2397 	ztest_ds_t *zd = arg;
2398 	objset_t *os = zd->zd_os;
2399 	uint64_t object = lr->lr_foid;
2400 	uint64_t offset = lr->lr_offset;
2401 	uint64_t size = lr->lr_length;
2402 	uint64_t txg = lr->lr_common.lrc_txg;
2403 	uint64_t crtxg;
2404 	dmu_object_info_t doi;
2405 	dmu_buf_t *db;
2406 	zgd_t *zgd;
2407 	int error;
2408 
2409 	ASSERT3P(lwb, !=, NULL);
2410 	ASSERT3P(zio, !=, NULL);
2411 	ASSERT3U(size, !=, 0);
2412 
2413 	ztest_object_lock(zd, object, RL_READER);
2414 	error = dmu_bonus_hold(os, object, FTAG, &db);
2415 	if (error) {
2416 		ztest_object_unlock(zd, object);
2417 		return (error);
2418 	}
2419 
2420 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
2421 
2422 	if (crtxg == 0 || crtxg > txg) {
2423 		dmu_buf_rele(db, FTAG);
2424 		ztest_object_unlock(zd, object);
2425 		return (ENOENT);
2426 	}
2427 
2428 	dmu_object_info_from_db(db, &doi);
2429 	dmu_buf_rele(db, FTAG);
2430 	db = NULL;
2431 
2432 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2433 	zgd->zgd_lwb = lwb;
2434 	zgd->zgd_private = zd;
2435 
2436 	if (buf != NULL) {	/* immediate write */
2437 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2438 		    object, offset, size, RL_READER);
2439 
2440 		error = dmu_read(os, object, offset, size, buf,
2441 		    DMU_READ_NO_PREFETCH);
2442 		ASSERT0(error);
2443 	} else {
2444 		size = doi.doi_data_block_size;
2445 		if (ISP2(size)) {
2446 			offset = P2ALIGN(offset, size);
2447 		} else {
2448 			ASSERT3U(offset, <, size);
2449 			offset = 0;
2450 		}
2451 
2452 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2453 		    object, offset, size, RL_READER);
2454 
2455 		error = dmu_buf_hold(os, object, offset, zgd, &db,
2456 		    DMU_READ_NO_PREFETCH);
2457 
2458 		if (error == 0) {
2459 			blkptr_t *bp = &lr->lr_blkptr;
2460 
2461 			zgd->zgd_db = db;
2462 			zgd->zgd_bp = bp;
2463 
2464 			ASSERT3U(db->db_offset, ==, offset);
2465 			ASSERT3U(db->db_size, ==, size);
2466 
2467 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
2468 			    ztest_get_done, zgd);
2469 
2470 			if (error == 0)
2471 				return (0);
2472 		}
2473 	}
2474 
2475 	ztest_get_done(zgd, error);
2476 
2477 	return (error);
2478 }
2479 
2480 static void *
2481 ztest_lr_alloc(size_t lrsize, char *name)
2482 {
2483 	char *lr;
2484 	size_t namesize = name ? strlen(name) + 1 : 0;
2485 
2486 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2487 
2488 	if (name)
2489 		memcpy(lr + lrsize, name, namesize);
2490 
2491 	return (lr);
2492 }
2493 
2494 static void
2495 ztest_lr_free(void *lr, size_t lrsize, char *name)
2496 {
2497 	size_t namesize = name ? strlen(name) + 1 : 0;
2498 
2499 	umem_free(lr, lrsize + namesize);
2500 }
2501 
2502 /*
2503  * Lookup a bunch of objects.  Returns the number of objects not found.
2504  */
2505 static int
2506 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2507 {
2508 	int missing = 0;
2509 	int error;
2510 	int i;
2511 
2512 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2513 
2514 	for (i = 0; i < count; i++, od++) {
2515 		od->od_object = 0;
2516 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2517 		    sizeof (uint64_t), 1, &od->od_object);
2518 		if (error) {
2519 			ASSERT3S(error, ==, ENOENT);
2520 			ASSERT0(od->od_object);
2521 			missing++;
2522 		} else {
2523 			dmu_buf_t *db;
2524 			ztest_block_tag_t *bbt;
2525 			dmu_object_info_t doi;
2526 
2527 			ASSERT3U(od->od_object, !=, 0);
2528 			ASSERT0(missing);	/* there should be no gaps */
2529 
2530 			ztest_object_lock(zd, od->od_object, RL_READER);
2531 			VERIFY0(dmu_bonus_hold(zd->zd_os, od->od_object,
2532 			    FTAG, &db));
2533 			dmu_object_info_from_db(db, &doi);
2534 			bbt = ztest_bt_bonus(db);
2535 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2536 			od->od_type = doi.doi_type;
2537 			od->od_blocksize = doi.doi_data_block_size;
2538 			od->od_gen = bbt->bt_gen;
2539 			dmu_buf_rele(db, FTAG);
2540 			ztest_object_unlock(zd, od->od_object);
2541 		}
2542 	}
2543 
2544 	return (missing);
2545 }
2546 
2547 static int
2548 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2549 {
2550 	int missing = 0;
2551 	int i;
2552 
2553 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2554 
2555 	for (i = 0; i < count; i++, od++) {
2556 		if (missing) {
2557 			od->od_object = 0;
2558 			missing++;
2559 			continue;
2560 		}
2561 
2562 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2563 
2564 		lr->lr_doid = od->od_dir;
2565 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2566 		lr->lrz_type = od->od_crtype;
2567 		lr->lrz_blocksize = od->od_crblocksize;
2568 		lr->lrz_ibshift = ztest_random_ibshift();
2569 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2570 		lr->lrz_dnodesize = od->od_crdnodesize;
2571 		lr->lr_gen = od->od_crgen;
2572 		lr->lr_crtime[0] = time(NULL);
2573 
2574 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2575 			ASSERT0(missing);
2576 			od->od_object = 0;
2577 			missing++;
2578 		} else {
2579 			od->od_object = lr->lr_foid;
2580 			od->od_type = od->od_crtype;
2581 			od->od_blocksize = od->od_crblocksize;
2582 			od->od_gen = od->od_crgen;
2583 			ASSERT3U(od->od_object, !=, 0);
2584 		}
2585 
2586 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2587 	}
2588 
2589 	return (missing);
2590 }
2591 
2592 static int
2593 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2594 {
2595 	int missing = 0;
2596 	int error;
2597 	int i;
2598 
2599 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2600 
2601 	od += count - 1;
2602 
2603 	for (i = count - 1; i >= 0; i--, od--) {
2604 		if (missing) {
2605 			missing++;
2606 			continue;
2607 		}
2608 
2609 		/*
2610 		 * No object was found.
2611 		 */
2612 		if (od->od_object == 0)
2613 			continue;
2614 
2615 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2616 
2617 		lr->lr_doid = od->od_dir;
2618 
2619 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2620 			ASSERT3U(error, ==, ENOSPC);
2621 			missing++;
2622 		} else {
2623 			od->od_object = 0;
2624 		}
2625 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2626 	}
2627 
2628 	return (missing);
2629 }
2630 
2631 static int
2632 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2633     void *data)
2634 {
2635 	lr_write_t *lr;
2636 	int error;
2637 
2638 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2639 
2640 	lr->lr_foid = object;
2641 	lr->lr_offset = offset;
2642 	lr->lr_length = size;
2643 	lr->lr_blkoff = 0;
2644 	BP_ZERO(&lr->lr_blkptr);
2645 
2646 	memcpy(lr + 1, data, size);
2647 
2648 	error = ztest_replay_write(zd, lr, B_FALSE);
2649 
2650 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2651 
2652 	return (error);
2653 }
2654 
2655 static int
2656 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2657 {
2658 	lr_truncate_t *lr;
2659 	int error;
2660 
2661 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2662 
2663 	lr->lr_foid = object;
2664 	lr->lr_offset = offset;
2665 	lr->lr_length = size;
2666 
2667 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2668 
2669 	ztest_lr_free(lr, sizeof (*lr), NULL);
2670 
2671 	return (error);
2672 }
2673 
2674 static int
2675 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2676 {
2677 	lr_setattr_t *lr;
2678 	int error;
2679 
2680 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2681 
2682 	lr->lr_foid = object;
2683 	lr->lr_size = 0;
2684 	lr->lr_mode = 0;
2685 
2686 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2687 
2688 	ztest_lr_free(lr, sizeof (*lr), NULL);
2689 
2690 	return (error);
2691 }
2692 
2693 static void
2694 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2695 {
2696 	objset_t *os = zd->zd_os;
2697 	dmu_tx_t *tx;
2698 	uint64_t txg;
2699 	rl_t *rl;
2700 
2701 	txg_wait_synced(dmu_objset_pool(os), 0);
2702 
2703 	ztest_object_lock(zd, object, RL_READER);
2704 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2705 
2706 	tx = dmu_tx_create(os);
2707 
2708 	dmu_tx_hold_write(tx, object, offset, size);
2709 
2710 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2711 
2712 	if (txg != 0) {
2713 		dmu_prealloc(os, object, offset, size, tx);
2714 		dmu_tx_commit(tx);
2715 		txg_wait_synced(dmu_objset_pool(os), txg);
2716 	} else {
2717 		(void) dmu_free_long_range(os, object, offset, size);
2718 	}
2719 
2720 	ztest_range_unlock(rl);
2721 	ztest_object_unlock(zd, object);
2722 }
2723 
2724 static void
2725 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2726 {
2727 	int err;
2728 	ztest_block_tag_t wbt;
2729 	dmu_object_info_t doi;
2730 	enum ztest_io_type io_type;
2731 	uint64_t blocksize;
2732 	void *data;
2733 
2734 	VERIFY0(dmu_object_info(zd->zd_os, object, &doi));
2735 	blocksize = doi.doi_data_block_size;
2736 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2737 
2738 	/*
2739 	 * Pick an i/o type at random, biased toward writing block tags.
2740 	 */
2741 	io_type = ztest_random(ZTEST_IO_TYPES);
2742 	if (ztest_random(2) == 0)
2743 		io_type = ZTEST_IO_WRITE_TAG;
2744 
2745 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2746 
2747 	switch (io_type) {
2748 
2749 	case ZTEST_IO_WRITE_TAG:
2750 		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2751 		    offset, 0, 0, 0);
2752 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2753 		break;
2754 
2755 	case ZTEST_IO_WRITE_PATTERN:
2756 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2757 		if (ztest_random(2) == 0) {
2758 			/*
2759 			 * Induce fletcher2 collisions to ensure that
2760 			 * zio_ddt_collision() detects and resolves them
2761 			 * when using fletcher2-verify for deduplication.
2762 			 */
2763 			((uint64_t *)data)[0] ^= 1ULL << 63;
2764 			((uint64_t *)data)[4] ^= 1ULL << 63;
2765 		}
2766 		(void) ztest_write(zd, object, offset, blocksize, data);
2767 		break;
2768 
2769 	case ZTEST_IO_WRITE_ZEROES:
2770 		memset(data, 0, blocksize);
2771 		(void) ztest_write(zd, object, offset, blocksize, data);
2772 		break;
2773 
2774 	case ZTEST_IO_TRUNCATE:
2775 		(void) ztest_truncate(zd, object, offset, blocksize);
2776 		break;
2777 
2778 	case ZTEST_IO_SETATTR:
2779 		(void) ztest_setattr(zd, object);
2780 		break;
2781 	default:
2782 		break;
2783 
2784 	case ZTEST_IO_REWRITE:
2785 		(void) pthread_rwlock_rdlock(&ztest_name_lock);
2786 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2787 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2788 		    B_FALSE);
2789 		VERIFY(err == 0 || err == ENOSPC);
2790 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2791 		    ZFS_PROP_COMPRESSION,
2792 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2793 		    B_FALSE);
2794 		VERIFY(err == 0 || err == ENOSPC);
2795 		(void) pthread_rwlock_unlock(&ztest_name_lock);
2796 
2797 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2798 		    DMU_READ_NO_PREFETCH));
2799 
2800 		(void) ztest_write(zd, object, offset, blocksize, data);
2801 		break;
2802 	}
2803 
2804 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2805 
2806 	umem_free(data, blocksize);
2807 }
2808 
2809 /*
2810  * Initialize an object description template.
2811  */
2812 static void
2813 ztest_od_init(ztest_od_t *od, uint64_t id, const char *tag, uint64_t index,
2814     dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2815     uint64_t gen)
2816 {
2817 	od->od_dir = ZTEST_DIROBJ;
2818 	od->od_object = 0;
2819 
2820 	od->od_crtype = type;
2821 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2822 	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2823 	od->od_crgen = gen;
2824 
2825 	od->od_type = DMU_OT_NONE;
2826 	od->od_blocksize = 0;
2827 	od->od_gen = 0;
2828 
2829 	(void) snprintf(od->od_name, sizeof (od->od_name),
2830 	    "%s(%"PRId64")[%"PRIu64"]",
2831 	    tag, id, index);
2832 }
2833 
2834 /*
2835  * Lookup or create the objects for a test using the od template.
2836  * If the objects do not all exist, or if 'remove' is specified,
2837  * remove any existing objects and create new ones.  Otherwise,
2838  * use the existing objects.
2839  */
2840 static int
2841 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2842 {
2843 	int count = size / sizeof (*od);
2844 	int rv = 0;
2845 
2846 	mutex_enter(&zd->zd_dirobj_lock);
2847 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2848 	    (ztest_remove(zd, od, count) != 0 ||
2849 	    ztest_create(zd, od, count) != 0))
2850 		rv = -1;
2851 	zd->zd_od = od;
2852 	mutex_exit(&zd->zd_dirobj_lock);
2853 
2854 	return (rv);
2855 }
2856 
2857 void
2858 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2859 {
2860 	(void) id;
2861 	zilog_t *zilog = zd->zd_zilog;
2862 
2863 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2864 
2865 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2866 
2867 	/*
2868 	 * Remember the committed values in zd, which is in parent/child
2869 	 * shared memory.  If we die, the next iteration of ztest_run()
2870 	 * will verify that the log really does contain this record.
2871 	 */
2872 	mutex_enter(&zilog->zl_lock);
2873 	ASSERT3P(zd->zd_shared, !=, NULL);
2874 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2875 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2876 	mutex_exit(&zilog->zl_lock);
2877 
2878 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2879 }
2880 
2881 /*
2882  * This function is designed to simulate the operations that occur during a
2883  * mount/unmount operation.  We hold the dataset across these operations in an
2884  * attempt to expose any implicit assumptions about ZIL management.
2885  */
2886 void
2887 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2888 {
2889 	(void) id;
2890 	objset_t *os = zd->zd_os;
2891 
2892 	/*
2893 	 * We hold the ztest_vdev_lock so we don't cause problems with
2894 	 * other threads that wish to remove a log device, such as
2895 	 * ztest_device_removal().
2896 	 */
2897 	mutex_enter(&ztest_vdev_lock);
2898 
2899 	/*
2900 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2901 	 * updating the zil (i.e. adding in-memory log records) and the
2902 	 * zd_zilog_lock to block any I/O.
2903 	 */
2904 	mutex_enter(&zd->zd_dirobj_lock);
2905 	(void) pthread_rwlock_wrlock(&zd->zd_zilog_lock);
2906 
2907 	/* zfsvfs_teardown() */
2908 	zil_close(zd->zd_zilog);
2909 
2910 	/* zfsvfs_setup() */
2911 	VERIFY3P(zil_open(os, ztest_get_data, NULL), ==, zd->zd_zilog);
2912 	zil_replay(os, zd, ztest_replay_vector);
2913 
2914 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2915 	mutex_exit(&zd->zd_dirobj_lock);
2916 	mutex_exit(&ztest_vdev_lock);
2917 }
2918 
2919 /*
2920  * Verify that we can't destroy an active pool, create an existing pool,
2921  * or create a pool with a bad vdev spec.
2922  */
2923 void
2924 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2925 {
2926 	(void) zd, (void) id;
2927 	ztest_shared_opts_t *zo = &ztest_opts;
2928 	spa_t *spa;
2929 	nvlist_t *nvroot;
2930 
2931 	if (zo->zo_mmp_test)
2932 		return;
2933 
2934 	/*
2935 	 * Attempt to create using a bad file.
2936 	 */
2937 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2938 	VERIFY3U(ENOENT, ==,
2939 	    spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2940 	fnvlist_free(nvroot);
2941 
2942 	/*
2943 	 * Attempt to create using a bad mirror.
2944 	 */
2945 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 2, 1);
2946 	VERIFY3U(ENOENT, ==,
2947 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2948 	fnvlist_free(nvroot);
2949 
2950 	/*
2951 	 * Attempt to create an existing pool.  It shouldn't matter
2952 	 * what's in the nvroot; we should fail with EEXIST.
2953 	 */
2954 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
2955 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2956 	VERIFY3U(EEXIST, ==,
2957 	    spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2958 	fnvlist_free(nvroot);
2959 
2960 	/*
2961 	 * We open a reference to the spa and then we try to export it
2962 	 * expecting one of the following errors:
2963 	 *
2964 	 * EBUSY
2965 	 *	Because of the reference we just opened.
2966 	 *
2967 	 * ZFS_ERR_EXPORT_IN_PROGRESS
2968 	 *	For the case that there is another ztest thread doing
2969 	 *	an export concurrently.
2970 	 */
2971 	VERIFY0(spa_open(zo->zo_pool, &spa, FTAG));
2972 	int error = spa_destroy(zo->zo_pool);
2973 	if (error != EBUSY && error != ZFS_ERR_EXPORT_IN_PROGRESS) {
2974 		fatal(B_FALSE, "spa_destroy(%s) returned unexpected value %d",
2975 		    spa->spa_name, error);
2976 	}
2977 	spa_close(spa, FTAG);
2978 
2979 	(void) pthread_rwlock_unlock(&ztest_name_lock);
2980 }
2981 
2982 /*
2983  * Start and then stop the MMP threads to ensure the startup and shutdown code
2984  * works properly.  Actual protection and property-related code tested via ZTS.
2985  */
2986 void
2987 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2988 {
2989 	(void) zd, (void) id;
2990 	ztest_shared_opts_t *zo = &ztest_opts;
2991 	spa_t *spa = ztest_spa;
2992 
2993 	if (zo->zo_mmp_test)
2994 		return;
2995 
2996 	/*
2997 	 * Since enabling MMP involves setting a property, it could not be done
2998 	 * while the pool is suspended.
2999 	 */
3000 	if (spa_suspended(spa))
3001 		return;
3002 
3003 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3004 	mutex_enter(&spa->spa_props_lock);
3005 
3006 	zfs_multihost_fail_intervals = 0;
3007 
3008 	if (!spa_multihost(spa)) {
3009 		spa->spa_multihost = B_TRUE;
3010 		mmp_thread_start(spa);
3011 	}
3012 
3013 	mutex_exit(&spa->spa_props_lock);
3014 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3015 
3016 	txg_wait_synced(spa_get_dsl(spa), 0);
3017 	mmp_signal_all_threads();
3018 	txg_wait_synced(spa_get_dsl(spa), 0);
3019 
3020 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3021 	mutex_enter(&spa->spa_props_lock);
3022 
3023 	if (spa_multihost(spa)) {
3024 		mmp_thread_stop(spa);
3025 		spa->spa_multihost = B_FALSE;
3026 	}
3027 
3028 	mutex_exit(&spa->spa_props_lock);
3029 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3030 }
3031 
3032 void
3033 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
3034 {
3035 	(void) zd, (void) id;
3036 	spa_t *spa;
3037 	uint64_t initial_version = SPA_VERSION_INITIAL;
3038 	uint64_t version, newversion;
3039 	nvlist_t *nvroot, *props;
3040 	char *name;
3041 
3042 	if (ztest_opts.zo_mmp_test)
3043 		return;
3044 
3045 	/* dRAID added after feature flags, skip upgrade test. */
3046 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0)
3047 		return;
3048 
3049 	mutex_enter(&ztest_vdev_lock);
3050 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
3051 
3052 	/*
3053 	 * Clean up from previous runs.
3054 	 */
3055 	(void) spa_destroy(name);
3056 
3057 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
3058 	    NULL, ztest_opts.zo_raid_children, ztest_opts.zo_mirrors, 1);
3059 
3060 	/*
3061 	 * If we're configuring a RAIDZ device then make sure that the
3062 	 * initial version is capable of supporting that feature.
3063 	 */
3064 	switch (ztest_opts.zo_raid_parity) {
3065 	case 0:
3066 	case 1:
3067 		initial_version = SPA_VERSION_INITIAL;
3068 		break;
3069 	case 2:
3070 		initial_version = SPA_VERSION_RAIDZ2;
3071 		break;
3072 	case 3:
3073 		initial_version = SPA_VERSION_RAIDZ3;
3074 		break;
3075 	}
3076 
3077 	/*
3078 	 * Create a pool with a spa version that can be upgraded. Pick
3079 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
3080 	 */
3081 	do {
3082 		version = ztest_random_spa_version(initial_version);
3083 	} while (version > SPA_VERSION_BEFORE_FEATURES);
3084 
3085 	props = fnvlist_alloc();
3086 	fnvlist_add_uint64(props,
3087 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
3088 	VERIFY0(spa_create(name, nvroot, props, NULL, NULL));
3089 	fnvlist_free(nvroot);
3090 	fnvlist_free(props);
3091 
3092 	VERIFY0(spa_open(name, &spa, FTAG));
3093 	VERIFY3U(spa_version(spa), ==, version);
3094 	newversion = ztest_random_spa_version(version + 1);
3095 
3096 	if (ztest_opts.zo_verbose >= 4) {
3097 		(void) printf("upgrading spa version from "
3098 		    "%"PRIu64" to %"PRIu64"\n",
3099 		    version, newversion);
3100 	}
3101 
3102 	spa_upgrade(spa, newversion);
3103 	VERIFY3U(spa_version(spa), >, version);
3104 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
3105 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
3106 	spa_close(spa, FTAG);
3107 
3108 	kmem_strfree(name);
3109 	mutex_exit(&ztest_vdev_lock);
3110 }
3111 
3112 static void
3113 ztest_spa_checkpoint(spa_t *spa)
3114 {
3115 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3116 
3117 	int error = spa_checkpoint(spa->spa_name);
3118 
3119 	switch (error) {
3120 	case 0:
3121 	case ZFS_ERR_DEVRM_IN_PROGRESS:
3122 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3123 	case ZFS_ERR_CHECKPOINT_EXISTS:
3124 		break;
3125 	case ENOSPC:
3126 		ztest_record_enospc(FTAG);
3127 		break;
3128 	default:
3129 		fatal(B_FALSE, "spa_checkpoint(%s) = %d", spa->spa_name, error);
3130 	}
3131 }
3132 
3133 static void
3134 ztest_spa_discard_checkpoint(spa_t *spa)
3135 {
3136 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3137 
3138 	int error = spa_checkpoint_discard(spa->spa_name);
3139 
3140 	switch (error) {
3141 	case 0:
3142 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3143 	case ZFS_ERR_NO_CHECKPOINT:
3144 		break;
3145 	default:
3146 		fatal(B_FALSE, "spa_discard_checkpoint(%s) = %d",
3147 		    spa->spa_name, error);
3148 	}
3149 
3150 }
3151 
3152 void
3153 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
3154 {
3155 	(void) zd, (void) id;
3156 	spa_t *spa = ztest_spa;
3157 
3158 	mutex_enter(&ztest_checkpoint_lock);
3159 	if (ztest_random(2) == 0) {
3160 		ztest_spa_checkpoint(spa);
3161 	} else {
3162 		ztest_spa_discard_checkpoint(spa);
3163 	}
3164 	mutex_exit(&ztest_checkpoint_lock);
3165 }
3166 
3167 
3168 static vdev_t *
3169 vdev_lookup_by_path(vdev_t *vd, const char *path)
3170 {
3171 	vdev_t *mvd;
3172 	int c;
3173 
3174 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
3175 		return (vd);
3176 
3177 	for (c = 0; c < vd->vdev_children; c++)
3178 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
3179 		    NULL)
3180 			return (mvd);
3181 
3182 	return (NULL);
3183 }
3184 
3185 static int
3186 spa_num_top_vdevs(spa_t *spa)
3187 {
3188 	vdev_t *rvd = spa->spa_root_vdev;
3189 	ASSERT3U(spa_config_held(spa, SCL_VDEV, RW_READER), ==, SCL_VDEV);
3190 	return (rvd->vdev_children);
3191 }
3192 
3193 /*
3194  * Verify that vdev_add() works as expected.
3195  */
3196 void
3197 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
3198 {
3199 	(void) zd, (void) id;
3200 	ztest_shared_t *zs = ztest_shared;
3201 	spa_t *spa = ztest_spa;
3202 	uint64_t leaves;
3203 	uint64_t guid;
3204 	nvlist_t *nvroot;
3205 	int error;
3206 
3207 	if (ztest_opts.zo_mmp_test)
3208 		return;
3209 
3210 	mutex_enter(&ztest_vdev_lock);
3211 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) *
3212 	    ztest_opts.zo_raid_children;
3213 
3214 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3215 
3216 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3217 
3218 	/*
3219 	 * If we have slogs then remove them 1/4 of the time.
3220 	 */
3221 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
3222 		metaslab_group_t *mg;
3223 
3224 		/*
3225 		 * find the first real slog in log allocation class
3226 		 */
3227 		mg =  spa_log_class(spa)->mc_allocator[0].mca_rotor;
3228 		while (!mg->mg_vd->vdev_islog)
3229 			mg = mg->mg_next;
3230 
3231 		guid = mg->mg_vd->vdev_guid;
3232 
3233 		spa_config_exit(spa, SCL_VDEV, FTAG);
3234 
3235 		/*
3236 		 * We have to grab the zs_name_lock as writer to
3237 		 * prevent a race between removing a slog (dmu_objset_find)
3238 		 * and destroying a dataset. Removing the slog will
3239 		 * grab a reference on the dataset which may cause
3240 		 * dsl_destroy_head() to fail with EBUSY thus
3241 		 * leaving the dataset in an inconsistent state.
3242 		 */
3243 		pthread_rwlock_wrlock(&ztest_name_lock);
3244 		error = spa_vdev_remove(spa, guid, B_FALSE);
3245 		pthread_rwlock_unlock(&ztest_name_lock);
3246 
3247 		switch (error) {
3248 		case 0:
3249 		case EEXIST:	/* Generic zil_reset() error */
3250 		case EBUSY:	/* Replay required */
3251 		case EACCES:	/* Crypto key not loaded */
3252 		case ZFS_ERR_CHECKPOINT_EXISTS:
3253 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3254 			break;
3255 		default:
3256 			fatal(B_FALSE, "spa_vdev_remove() = %d", error);
3257 		}
3258 	} else {
3259 		spa_config_exit(spa, SCL_VDEV, FTAG);
3260 
3261 		/*
3262 		 * Make 1/4 of the devices be log devices
3263 		 */
3264 		nvroot = make_vdev_root(NULL, NULL, NULL,
3265 		    ztest_opts.zo_vdev_size, 0, (ztest_random(4) == 0) ?
3266 		    "log" : NULL, ztest_opts.zo_raid_children, zs->zs_mirrors,
3267 		    1);
3268 
3269 		error = spa_vdev_add(spa, nvroot);
3270 		fnvlist_free(nvroot);
3271 
3272 		switch (error) {
3273 		case 0:
3274 			break;
3275 		case ENOSPC:
3276 			ztest_record_enospc("spa_vdev_add");
3277 			break;
3278 		default:
3279 			fatal(B_FALSE, "spa_vdev_add() = %d", error);
3280 		}
3281 	}
3282 
3283 	mutex_exit(&ztest_vdev_lock);
3284 }
3285 
3286 void
3287 ztest_vdev_class_add(ztest_ds_t *zd, uint64_t id)
3288 {
3289 	(void) zd, (void) id;
3290 	ztest_shared_t *zs = ztest_shared;
3291 	spa_t *spa = ztest_spa;
3292 	uint64_t leaves;
3293 	nvlist_t *nvroot;
3294 	const char *class = (ztest_random(2) == 0) ?
3295 	    VDEV_ALLOC_BIAS_SPECIAL : VDEV_ALLOC_BIAS_DEDUP;
3296 	int error;
3297 
3298 	/*
3299 	 * By default add a special vdev 50% of the time
3300 	 */
3301 	if ((ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_OFF) ||
3302 	    (ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_RND &&
3303 	    ztest_random(2) == 0)) {
3304 		return;
3305 	}
3306 
3307 	mutex_enter(&ztest_vdev_lock);
3308 
3309 	/* Only test with mirrors */
3310 	if (zs->zs_mirrors < 2) {
3311 		mutex_exit(&ztest_vdev_lock);
3312 		return;
3313 	}
3314 
3315 	/* requires feature@allocation_classes */
3316 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) {
3317 		mutex_exit(&ztest_vdev_lock);
3318 		return;
3319 	}
3320 
3321 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) *
3322 	    ztest_opts.zo_raid_children;
3323 
3324 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3325 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3326 	spa_config_exit(spa, SCL_VDEV, FTAG);
3327 
3328 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
3329 	    class, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
3330 
3331 	error = spa_vdev_add(spa, nvroot);
3332 	fnvlist_free(nvroot);
3333 
3334 	if (error == ENOSPC)
3335 		ztest_record_enospc("spa_vdev_add");
3336 	else if (error != 0)
3337 		fatal(B_FALSE, "spa_vdev_add() = %d", error);
3338 
3339 	/*
3340 	 * 50% of the time allow small blocks in the special class
3341 	 */
3342 	if (error == 0 &&
3343 	    spa_special_class(spa)->mc_groups == 1 && ztest_random(2) == 0) {
3344 		if (ztest_opts.zo_verbose >= 3)
3345 			(void) printf("Enabling special VDEV small blocks\n");
3346 		(void) ztest_dsl_prop_set_uint64(zd->zd_name,
3347 		    ZFS_PROP_SPECIAL_SMALL_BLOCKS, 32768, B_FALSE);
3348 	}
3349 
3350 	mutex_exit(&ztest_vdev_lock);
3351 
3352 	if (ztest_opts.zo_verbose >= 3) {
3353 		metaslab_class_t *mc;
3354 
3355 		if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL) == 0)
3356 			mc = spa_special_class(spa);
3357 		else
3358 			mc = spa_dedup_class(spa);
3359 		(void) printf("Added a %s mirrored vdev (of %d)\n",
3360 		    class, (int)mc->mc_groups);
3361 	}
3362 }
3363 
3364 /*
3365  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
3366  */
3367 void
3368 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
3369 {
3370 	(void) zd, (void) id;
3371 	ztest_shared_t *zs = ztest_shared;
3372 	spa_t *spa = ztest_spa;
3373 	vdev_t *rvd = spa->spa_root_vdev;
3374 	spa_aux_vdev_t *sav;
3375 	const char *aux;
3376 	char *path;
3377 	uint64_t guid = 0;
3378 	int error, ignore_err = 0;
3379 
3380 	if (ztest_opts.zo_mmp_test)
3381 		return;
3382 
3383 	path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3384 
3385 	if (ztest_random(2) == 0) {
3386 		sav = &spa->spa_spares;
3387 		aux = ZPOOL_CONFIG_SPARES;
3388 	} else {
3389 		sav = &spa->spa_l2cache;
3390 		aux = ZPOOL_CONFIG_L2CACHE;
3391 	}
3392 
3393 	mutex_enter(&ztest_vdev_lock);
3394 
3395 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3396 
3397 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
3398 		/*
3399 		 * Pick a random device to remove.
3400 		 */
3401 		vdev_t *svd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3402 
3403 		/* dRAID spares cannot be removed; try anyways to see ENOTSUP */
3404 		if (strstr(svd->vdev_path, VDEV_TYPE_DRAID) != NULL)
3405 			ignore_err = ENOTSUP;
3406 
3407 		guid = svd->vdev_guid;
3408 	} else {
3409 		/*
3410 		 * Find an unused device we can add.
3411 		 */
3412 		zs->zs_vdev_aux = 0;
3413 		for (;;) {
3414 			int c;
3415 			(void) snprintf(path, MAXPATHLEN, ztest_aux_template,
3416 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
3417 			    zs->zs_vdev_aux);
3418 			for (c = 0; c < sav->sav_count; c++)
3419 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
3420 				    path) == 0)
3421 					break;
3422 			if (c == sav->sav_count &&
3423 			    vdev_lookup_by_path(rvd, path) == NULL)
3424 				break;
3425 			zs->zs_vdev_aux++;
3426 		}
3427 	}
3428 
3429 	spa_config_exit(spa, SCL_VDEV, FTAG);
3430 
3431 	if (guid == 0) {
3432 		/*
3433 		 * Add a new device.
3434 		 */
3435 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3436 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, NULL, 0, 0, 1);
3437 		error = spa_vdev_add(spa, nvroot);
3438 
3439 		switch (error) {
3440 		case 0:
3441 			break;
3442 		default:
3443 			fatal(B_FALSE, "spa_vdev_add(%p) = %d", nvroot, error);
3444 		}
3445 		fnvlist_free(nvroot);
3446 	} else {
3447 		/*
3448 		 * Remove an existing device.  Sometimes, dirty its
3449 		 * vdev state first to make sure we handle removal
3450 		 * of devices that have pending state changes.
3451 		 */
3452 		if (ztest_random(2) == 0)
3453 			(void) vdev_online(spa, guid, 0, NULL);
3454 
3455 		error = spa_vdev_remove(spa, guid, B_FALSE);
3456 
3457 		switch (error) {
3458 		case 0:
3459 		case EBUSY:
3460 		case ZFS_ERR_CHECKPOINT_EXISTS:
3461 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3462 			break;
3463 		default:
3464 			if (error != ignore_err)
3465 				fatal(B_FALSE,
3466 				    "spa_vdev_remove(%"PRIu64") = %d",
3467 				    guid, error);
3468 		}
3469 	}
3470 
3471 	mutex_exit(&ztest_vdev_lock);
3472 
3473 	umem_free(path, MAXPATHLEN);
3474 }
3475 
3476 /*
3477  * split a pool if it has mirror tlvdevs
3478  */
3479 void
3480 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3481 {
3482 	(void) zd, (void) id;
3483 	ztest_shared_t *zs = ztest_shared;
3484 	spa_t *spa = ztest_spa;
3485 	vdev_t *rvd = spa->spa_root_vdev;
3486 	nvlist_t *tree, **child, *config, *split, **schild;
3487 	uint_t c, children, schildren = 0, lastlogid = 0;
3488 	int error = 0;
3489 
3490 	if (ztest_opts.zo_mmp_test)
3491 		return;
3492 
3493 	mutex_enter(&ztest_vdev_lock);
3494 
3495 	/* ensure we have a usable config; mirrors of raidz aren't supported */
3496 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raid_children > 1) {
3497 		mutex_exit(&ztest_vdev_lock);
3498 		return;
3499 	}
3500 
3501 	/* clean up the old pool, if any */
3502 	(void) spa_destroy("splitp");
3503 
3504 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3505 
3506 	/* generate a config from the existing config */
3507 	mutex_enter(&spa->spa_props_lock);
3508 	tree = fnvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE);
3509 	mutex_exit(&spa->spa_props_lock);
3510 
3511 	VERIFY0(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN,
3512 	    &child, &children));
3513 
3514 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
3515 	for (c = 0; c < children; c++) {
3516 		vdev_t *tvd = rvd->vdev_child[c];
3517 		nvlist_t **mchild;
3518 		uint_t mchildren;
3519 
3520 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3521 			schild[schildren] = fnvlist_alloc();
3522 			fnvlist_add_string(schild[schildren],
3523 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE);
3524 			fnvlist_add_uint64(schild[schildren],
3525 			    ZPOOL_CONFIG_IS_HOLE, 1);
3526 			if (lastlogid == 0)
3527 				lastlogid = schildren;
3528 			++schildren;
3529 			continue;
3530 		}
3531 		lastlogid = 0;
3532 		VERIFY0(nvlist_lookup_nvlist_array(child[c],
3533 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren));
3534 		schild[schildren++] = fnvlist_dup(mchild[0]);
3535 	}
3536 
3537 	/* OK, create a config that can be used to split */
3538 	split = fnvlist_alloc();
3539 	fnvlist_add_string(split, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
3540 	fnvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN,
3541 	    (const nvlist_t **)schild, lastlogid != 0 ? lastlogid : schildren);
3542 
3543 	config = fnvlist_alloc();
3544 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split);
3545 
3546 	for (c = 0; c < schildren; c++)
3547 		fnvlist_free(schild[c]);
3548 	free(schild);
3549 	fnvlist_free(split);
3550 
3551 	spa_config_exit(spa, SCL_VDEV, FTAG);
3552 
3553 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
3554 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3555 	(void) pthread_rwlock_unlock(&ztest_name_lock);
3556 
3557 	fnvlist_free(config);
3558 
3559 	if (error == 0) {
3560 		(void) printf("successful split - results:\n");
3561 		mutex_enter(&spa_namespace_lock);
3562 		show_pool_stats(spa);
3563 		show_pool_stats(spa_lookup("splitp"));
3564 		mutex_exit(&spa_namespace_lock);
3565 		++zs->zs_splits;
3566 		--zs->zs_mirrors;
3567 	}
3568 	mutex_exit(&ztest_vdev_lock);
3569 }
3570 
3571 /*
3572  * Verify that we can attach and detach devices.
3573  */
3574 void
3575 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3576 {
3577 	(void) zd, (void) id;
3578 	ztest_shared_t *zs = ztest_shared;
3579 	spa_t *spa = ztest_spa;
3580 	spa_aux_vdev_t *sav = &spa->spa_spares;
3581 	vdev_t *rvd = spa->spa_root_vdev;
3582 	vdev_t *oldvd, *newvd, *pvd;
3583 	nvlist_t *root;
3584 	uint64_t leaves;
3585 	uint64_t leaf, top;
3586 	uint64_t ashift = ztest_get_ashift();
3587 	uint64_t oldguid, pguid;
3588 	uint64_t oldsize, newsize;
3589 	char *oldpath, *newpath;
3590 	int replacing;
3591 	int oldvd_has_siblings = B_FALSE;
3592 	int newvd_is_spare = B_FALSE;
3593 	int newvd_is_dspare = B_FALSE;
3594 	int oldvd_is_log;
3595 	int error, expected_error;
3596 
3597 	if (ztest_opts.zo_mmp_test)
3598 		return;
3599 
3600 	oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3601 	newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3602 
3603 	mutex_enter(&ztest_vdev_lock);
3604 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raid_children;
3605 
3606 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3607 
3608 	/*
3609 	 * If a vdev is in the process of being removed, its removal may
3610 	 * finish while we are in progress, leading to an unexpected error
3611 	 * value.  Don't bother trying to attach while we are in the middle
3612 	 * of removal.
3613 	 */
3614 	if (ztest_device_removal_active) {
3615 		spa_config_exit(spa, SCL_ALL, FTAG);
3616 		goto out;
3617 	}
3618 
3619 	/*
3620 	 * Decide whether to do an attach or a replace.
3621 	 */
3622 	replacing = ztest_random(2);
3623 
3624 	/*
3625 	 * Pick a random top-level vdev.
3626 	 */
3627 	top = ztest_random_vdev_top(spa, B_TRUE);
3628 
3629 	/*
3630 	 * Pick a random leaf within it.
3631 	 */
3632 	leaf = ztest_random(leaves);
3633 
3634 	/*
3635 	 * Locate this vdev.
3636 	 */
3637 	oldvd = rvd->vdev_child[top];
3638 
3639 	/* pick a child from the mirror */
3640 	if (zs->zs_mirrors >= 1) {
3641 		ASSERT3P(oldvd->vdev_ops, ==, &vdev_mirror_ops);
3642 		ASSERT3U(oldvd->vdev_children, >=, zs->zs_mirrors);
3643 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raid_children];
3644 	}
3645 
3646 	/* pick a child out of the raidz group */
3647 	if (ztest_opts.zo_raid_children > 1) {
3648 		if (strcmp(oldvd->vdev_ops->vdev_op_type, "raidz") == 0)
3649 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_raidz_ops);
3650 		else
3651 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_draid_ops);
3652 		ASSERT3U(oldvd->vdev_children, ==, ztest_opts.zo_raid_children);
3653 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raid_children];
3654 	}
3655 
3656 	/*
3657 	 * If we're already doing an attach or replace, oldvd may be a
3658 	 * mirror vdev -- in which case, pick a random child.
3659 	 */
3660 	while (oldvd->vdev_children != 0) {
3661 		oldvd_has_siblings = B_TRUE;
3662 		ASSERT3U(oldvd->vdev_children, >=, 2);
3663 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3664 	}
3665 
3666 	oldguid = oldvd->vdev_guid;
3667 	oldsize = vdev_get_min_asize(oldvd);
3668 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
3669 	(void) strcpy(oldpath, oldvd->vdev_path);
3670 	pvd = oldvd->vdev_parent;
3671 	pguid = pvd->vdev_guid;
3672 
3673 	/*
3674 	 * If oldvd has siblings, then half of the time, detach it.  Prior
3675 	 * to the detach the pool is scrubbed in order to prevent creating
3676 	 * unrepairable blocks as a result of the data corruption injection.
3677 	 */
3678 	if (oldvd_has_siblings && ztest_random(2) == 0) {
3679 		spa_config_exit(spa, SCL_ALL, FTAG);
3680 
3681 		error = ztest_scrub_impl(spa);
3682 		if (error)
3683 			goto out;
3684 
3685 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3686 		if (error != 0 && error != ENODEV && error != EBUSY &&
3687 		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3688 		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
3689 			fatal(B_FALSE, "detach (%s) returned %d",
3690 			    oldpath, error);
3691 		goto out;
3692 	}
3693 
3694 	/*
3695 	 * For the new vdev, choose with equal probability between the two
3696 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3697 	 */
3698 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
3699 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3700 		newvd_is_spare = B_TRUE;
3701 
3702 		if (newvd->vdev_ops == &vdev_draid_spare_ops)
3703 			newvd_is_dspare = B_TRUE;
3704 
3705 		(void) strcpy(newpath, newvd->vdev_path);
3706 	} else {
3707 		(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3708 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
3709 		    top * leaves + leaf);
3710 		if (ztest_random(2) == 0)
3711 			newpath[strlen(newpath) - 1] = 'b';
3712 		newvd = vdev_lookup_by_path(rvd, newpath);
3713 	}
3714 
3715 	if (newvd) {
3716 		/*
3717 		 * Reopen to ensure the vdev's asize field isn't stale.
3718 		 */
3719 		vdev_reopen(newvd);
3720 		newsize = vdev_get_min_asize(newvd);
3721 	} else {
3722 		/*
3723 		 * Make newsize a little bigger or smaller than oldsize.
3724 		 * If it's smaller, the attach should fail.
3725 		 * If it's larger, and we're doing a replace,
3726 		 * we should get dynamic LUN growth when we're done.
3727 		 */
3728 		newsize = 10 * oldsize / (9 + ztest_random(3));
3729 	}
3730 
3731 	/*
3732 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3733 	 * unless it's a replace; in that case any non-replacing parent is OK.
3734 	 *
3735 	 * If newvd is already part of the pool, it should fail with EBUSY.
3736 	 *
3737 	 * If newvd is too small, it should fail with EOVERFLOW.
3738 	 *
3739 	 * If newvd is a distributed spare and it's being attached to a
3740 	 * dRAID which is not its parent it should fail with EINVAL.
3741 	 */
3742 	if (pvd->vdev_ops != &vdev_mirror_ops &&
3743 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3744 	    pvd->vdev_ops == &vdev_replacing_ops ||
3745 	    pvd->vdev_ops == &vdev_spare_ops))
3746 		expected_error = ENOTSUP;
3747 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
3748 		expected_error = ENOTSUP;
3749 	else if (newvd == oldvd)
3750 		expected_error = replacing ? 0 : EBUSY;
3751 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3752 		expected_error = EBUSY;
3753 	else if (!newvd_is_dspare && newsize < oldsize)
3754 		expected_error = EOVERFLOW;
3755 	else if (ashift > oldvd->vdev_top->vdev_ashift)
3756 		expected_error = EDOM;
3757 	else if (newvd_is_dspare && pvd != vdev_draid_spare_get_parent(newvd))
3758 		expected_error = ENOTSUP;
3759 	else
3760 		expected_error = 0;
3761 
3762 	spa_config_exit(spa, SCL_ALL, FTAG);
3763 
3764 	/*
3765 	 * Build the nvlist describing newpath.
3766 	 */
3767 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3768 	    ashift, NULL, 0, 0, 1);
3769 
3770 	/*
3771 	 * When supported select either a healing or sequential resilver.
3772 	 */
3773 	boolean_t rebuilding = B_FALSE;
3774 	if (pvd->vdev_ops == &vdev_mirror_ops ||
3775 	    pvd->vdev_ops ==  &vdev_root_ops) {
3776 		rebuilding = !!ztest_random(2);
3777 	}
3778 
3779 	error = spa_vdev_attach(spa, oldguid, root, replacing, rebuilding);
3780 
3781 	fnvlist_free(root);
3782 
3783 	/*
3784 	 * If our parent was the replacing vdev, but the replace completed,
3785 	 * then instead of failing with ENOTSUP we may either succeed,
3786 	 * fail with ENODEV, or fail with EOVERFLOW.
3787 	 */
3788 	if (expected_error == ENOTSUP &&
3789 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3790 		expected_error = error;
3791 
3792 	/*
3793 	 * If someone grew the LUN, the replacement may be too small.
3794 	 */
3795 	if (error == EOVERFLOW || error == EBUSY)
3796 		expected_error = error;
3797 
3798 	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3799 	    error == ZFS_ERR_DISCARDING_CHECKPOINT ||
3800 	    error == ZFS_ERR_RESILVER_IN_PROGRESS ||
3801 	    error == ZFS_ERR_REBUILD_IN_PROGRESS)
3802 		expected_error = error;
3803 
3804 	if (error != expected_error && expected_error != EBUSY) {
3805 		fatal(B_FALSE, "attach (%s %"PRIu64", %s %"PRIu64", %d) "
3806 		    "returned %d, expected %d",
3807 		    oldpath, oldsize, newpath,
3808 		    newsize, replacing, error, expected_error);
3809 	}
3810 out:
3811 	mutex_exit(&ztest_vdev_lock);
3812 
3813 	umem_free(oldpath, MAXPATHLEN);
3814 	umem_free(newpath, MAXPATHLEN);
3815 }
3816 
3817 void
3818 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3819 {
3820 	(void) zd, (void) id;
3821 	spa_t *spa = ztest_spa;
3822 	vdev_t *vd;
3823 	uint64_t guid;
3824 	int error;
3825 
3826 	mutex_enter(&ztest_vdev_lock);
3827 
3828 	if (ztest_device_removal_active) {
3829 		mutex_exit(&ztest_vdev_lock);
3830 		return;
3831 	}
3832 
3833 	/*
3834 	 * Remove a random top-level vdev and wait for removal to finish.
3835 	 */
3836 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3837 	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3838 	guid = vd->vdev_guid;
3839 	spa_config_exit(spa, SCL_VDEV, FTAG);
3840 
3841 	error = spa_vdev_remove(spa, guid, B_FALSE);
3842 	if (error == 0) {
3843 		ztest_device_removal_active = B_TRUE;
3844 		mutex_exit(&ztest_vdev_lock);
3845 
3846 		/*
3847 		 * spa->spa_vdev_removal is created in a sync task that
3848 		 * is initiated via dsl_sync_task_nowait(). Since the
3849 		 * task may not run before spa_vdev_remove() returns, we
3850 		 * must wait at least 1 txg to ensure that the removal
3851 		 * struct has been created.
3852 		 */
3853 		txg_wait_synced(spa_get_dsl(spa), 0);
3854 
3855 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
3856 			txg_wait_synced(spa_get_dsl(spa), 0);
3857 	} else {
3858 		mutex_exit(&ztest_vdev_lock);
3859 		return;
3860 	}
3861 
3862 	/*
3863 	 * The pool needs to be scrubbed after completing device removal.
3864 	 * Failure to do so may result in checksum errors due to the
3865 	 * strategy employed by ztest_fault_inject() when selecting which
3866 	 * offset are redundant and can be damaged.
3867 	 */
3868 	error = spa_scan(spa, POOL_SCAN_SCRUB);
3869 	if (error == 0) {
3870 		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
3871 			txg_wait_synced(spa_get_dsl(spa), 0);
3872 	}
3873 
3874 	mutex_enter(&ztest_vdev_lock);
3875 	ztest_device_removal_active = B_FALSE;
3876 	mutex_exit(&ztest_vdev_lock);
3877 }
3878 
3879 /*
3880  * Callback function which expands the physical size of the vdev.
3881  */
3882 static vdev_t *
3883 grow_vdev(vdev_t *vd, void *arg)
3884 {
3885 	spa_t *spa __maybe_unused = vd->vdev_spa;
3886 	size_t *newsize = arg;
3887 	size_t fsize;
3888 	int fd;
3889 
3890 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
3891 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3892 
3893 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3894 		return (vd);
3895 
3896 	fsize = lseek(fd, 0, SEEK_END);
3897 	VERIFY0(ftruncate(fd, *newsize));
3898 
3899 	if (ztest_opts.zo_verbose >= 6) {
3900 		(void) printf("%s grew from %lu to %lu bytes\n",
3901 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3902 	}
3903 	(void) close(fd);
3904 	return (NULL);
3905 }
3906 
3907 /*
3908  * Callback function which expands a given vdev by calling vdev_online().
3909  */
3910 static vdev_t *
3911 online_vdev(vdev_t *vd, void *arg)
3912 {
3913 	(void) arg;
3914 	spa_t *spa = vd->vdev_spa;
3915 	vdev_t *tvd = vd->vdev_top;
3916 	uint64_t guid = vd->vdev_guid;
3917 	uint64_t generation = spa->spa_config_generation + 1;
3918 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3919 	int error;
3920 
3921 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
3922 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3923 
3924 	/* Calling vdev_online will initialize the new metaslabs */
3925 	spa_config_exit(spa, SCL_STATE, spa);
3926 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3927 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3928 
3929 	/*
3930 	 * If vdev_online returned an error or the underlying vdev_open
3931 	 * failed then we abort the expand. The only way to know that
3932 	 * vdev_open fails is by checking the returned newstate.
3933 	 */
3934 	if (error || newstate != VDEV_STATE_HEALTHY) {
3935 		if (ztest_opts.zo_verbose >= 5) {
3936 			(void) printf("Unable to expand vdev, state %u, "
3937 			    "error %d\n", newstate, error);
3938 		}
3939 		return (vd);
3940 	}
3941 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3942 
3943 	/*
3944 	 * Since we dropped the lock we need to ensure that we're
3945 	 * still talking to the original vdev. It's possible this
3946 	 * vdev may have been detached/replaced while we were
3947 	 * trying to online it.
3948 	 */
3949 	if (generation != spa->spa_config_generation) {
3950 		if (ztest_opts.zo_verbose >= 5) {
3951 			(void) printf("vdev configuration has changed, "
3952 			    "guid %"PRIu64", state %"PRIu64", "
3953 			    "expected gen %"PRIu64", got gen %"PRIu64"\n",
3954 			    guid,
3955 			    tvd->vdev_state,
3956 			    generation,
3957 			    spa->spa_config_generation);
3958 		}
3959 		return (vd);
3960 	}
3961 	return (NULL);
3962 }
3963 
3964 /*
3965  * Traverse the vdev tree calling the supplied function.
3966  * We continue to walk the tree until we either have walked all
3967  * children or we receive a non-NULL return from the callback.
3968  * If a NULL callback is passed, then we just return back the first
3969  * leaf vdev we encounter.
3970  */
3971 static vdev_t *
3972 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3973 {
3974 	uint_t c;
3975 
3976 	if (vd->vdev_ops->vdev_op_leaf) {
3977 		if (func == NULL)
3978 			return (vd);
3979 		else
3980 			return (func(vd, arg));
3981 	}
3982 
3983 	for (c = 0; c < vd->vdev_children; c++) {
3984 		vdev_t *cvd = vd->vdev_child[c];
3985 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3986 			return (cvd);
3987 	}
3988 	return (NULL);
3989 }
3990 
3991 /*
3992  * Verify that dynamic LUN growth works as expected.
3993  */
3994 void
3995 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3996 {
3997 	(void) zd, (void) id;
3998 	spa_t *spa = ztest_spa;
3999 	vdev_t *vd, *tvd;
4000 	metaslab_class_t *mc;
4001 	metaslab_group_t *mg;
4002 	size_t psize, newsize;
4003 	uint64_t top;
4004 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
4005 
4006 	mutex_enter(&ztest_checkpoint_lock);
4007 	mutex_enter(&ztest_vdev_lock);
4008 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4009 
4010 	/*
4011 	 * If there is a vdev removal in progress, it could complete while
4012 	 * we are running, in which case we would not be able to verify
4013 	 * that the metaslab_class space increased (because it decreases
4014 	 * when the device removal completes).
4015 	 */
4016 	if (ztest_device_removal_active) {
4017 		spa_config_exit(spa, SCL_STATE, spa);
4018 		mutex_exit(&ztest_vdev_lock);
4019 		mutex_exit(&ztest_checkpoint_lock);
4020 		return;
4021 	}
4022 
4023 	top = ztest_random_vdev_top(spa, B_TRUE);
4024 
4025 	tvd = spa->spa_root_vdev->vdev_child[top];
4026 	mg = tvd->vdev_mg;
4027 	mc = mg->mg_class;
4028 	old_ms_count = tvd->vdev_ms_count;
4029 	old_class_space = metaslab_class_get_space(mc);
4030 
4031 	/*
4032 	 * Determine the size of the first leaf vdev associated with
4033 	 * our top-level device.
4034 	 */
4035 	vd = vdev_walk_tree(tvd, NULL, NULL);
4036 	ASSERT3P(vd, !=, NULL);
4037 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4038 
4039 	psize = vd->vdev_psize;
4040 
4041 	/*
4042 	 * We only try to expand the vdev if it's healthy, less than 4x its
4043 	 * original size, and it has a valid psize.
4044 	 */
4045 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
4046 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
4047 		spa_config_exit(spa, SCL_STATE, spa);
4048 		mutex_exit(&ztest_vdev_lock);
4049 		mutex_exit(&ztest_checkpoint_lock);
4050 		return;
4051 	}
4052 	ASSERT3U(psize, >, 0);
4053 	newsize = psize + MAX(psize / 8, SPA_MAXBLOCKSIZE);
4054 	ASSERT3U(newsize, >, psize);
4055 
4056 	if (ztest_opts.zo_verbose >= 6) {
4057 		(void) printf("Expanding LUN %s from %lu to %lu\n",
4058 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
4059 	}
4060 
4061 	/*
4062 	 * Growing the vdev is a two step process:
4063 	 *	1). expand the physical size (i.e. relabel)
4064 	 *	2). online the vdev to create the new metaslabs
4065 	 */
4066 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
4067 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
4068 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
4069 		if (ztest_opts.zo_verbose >= 5) {
4070 			(void) printf("Could not expand LUN because "
4071 			    "the vdev configuration changed.\n");
4072 		}
4073 		spa_config_exit(spa, SCL_STATE, spa);
4074 		mutex_exit(&ztest_vdev_lock);
4075 		mutex_exit(&ztest_checkpoint_lock);
4076 		return;
4077 	}
4078 
4079 	spa_config_exit(spa, SCL_STATE, spa);
4080 
4081 	/*
4082 	 * Expanding the LUN will update the config asynchronously,
4083 	 * thus we must wait for the async thread to complete any
4084 	 * pending tasks before proceeding.
4085 	 */
4086 	for (;;) {
4087 		boolean_t done;
4088 		mutex_enter(&spa->spa_async_lock);
4089 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
4090 		mutex_exit(&spa->spa_async_lock);
4091 		if (done)
4092 			break;
4093 		txg_wait_synced(spa_get_dsl(spa), 0);
4094 		(void) poll(NULL, 0, 100);
4095 	}
4096 
4097 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4098 
4099 	tvd = spa->spa_root_vdev->vdev_child[top];
4100 	new_ms_count = tvd->vdev_ms_count;
4101 	new_class_space = metaslab_class_get_space(mc);
4102 
4103 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
4104 		if (ztest_opts.zo_verbose >= 5) {
4105 			(void) printf("Could not verify LUN expansion due to "
4106 			    "intervening vdev offline or remove.\n");
4107 		}
4108 		spa_config_exit(spa, SCL_STATE, spa);
4109 		mutex_exit(&ztest_vdev_lock);
4110 		mutex_exit(&ztest_checkpoint_lock);
4111 		return;
4112 	}
4113 
4114 	/*
4115 	 * Make sure we were able to grow the vdev.
4116 	 */
4117 	if (new_ms_count <= old_ms_count) {
4118 		fatal(B_FALSE,
4119 		    "LUN expansion failed: ms_count %"PRIu64" < %"PRIu64"\n",
4120 		    old_ms_count, new_ms_count);
4121 	}
4122 
4123 	/*
4124 	 * Make sure we were able to grow the pool.
4125 	 */
4126 	if (new_class_space <= old_class_space) {
4127 		fatal(B_FALSE,
4128 		    "LUN expansion failed: class_space %"PRIu64" < %"PRIu64"\n",
4129 		    old_class_space, new_class_space);
4130 	}
4131 
4132 	if (ztest_opts.zo_verbose >= 5) {
4133 		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
4134 
4135 		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
4136 		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
4137 		(void) printf("%s grew from %s to %s\n",
4138 		    spa->spa_name, oldnumbuf, newnumbuf);
4139 	}
4140 
4141 	spa_config_exit(spa, SCL_STATE, spa);
4142 	mutex_exit(&ztest_vdev_lock);
4143 	mutex_exit(&ztest_checkpoint_lock);
4144 }
4145 
4146 /*
4147  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
4148  */
4149 static void
4150 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
4151 {
4152 	(void) arg, (void) cr;
4153 
4154 	/*
4155 	 * Create the objects common to all ztest datasets.
4156 	 */
4157 	VERIFY0(zap_create_claim(os, ZTEST_DIROBJ,
4158 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx));
4159 }
4160 
4161 static int
4162 ztest_dataset_create(char *dsname)
4163 {
4164 	int err;
4165 	uint64_t rand;
4166 	dsl_crypto_params_t *dcp = NULL;
4167 
4168 	/*
4169 	 * 50% of the time, we create encrypted datasets
4170 	 * using a random cipher suite and a hard-coded
4171 	 * wrapping key.
4172 	 */
4173 	rand = ztest_random(2);
4174 	if (rand != 0) {
4175 		nvlist_t *crypto_args = fnvlist_alloc();
4176 		nvlist_t *props = fnvlist_alloc();
4177 
4178 		/* slight bias towards the default cipher suite */
4179 		rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
4180 		if (rand < ZIO_CRYPT_AES_128_CCM)
4181 			rand = ZIO_CRYPT_ON;
4182 
4183 		fnvlist_add_uint64(props,
4184 		    zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
4185 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
4186 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
4187 
4188 		/*
4189 		 * These parameters aren't really used by the kernel. They
4190 		 * are simply stored so that userspace knows how to load
4191 		 * the wrapping key.
4192 		 */
4193 		fnvlist_add_uint64(props,
4194 		    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
4195 		fnvlist_add_string(props,
4196 		    zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
4197 		fnvlist_add_uint64(props,
4198 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
4199 		fnvlist_add_uint64(props,
4200 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
4201 
4202 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
4203 		    crypto_args, &dcp));
4204 
4205 		/*
4206 		 * Cycle through all available encryption implementations
4207 		 * to verify interoperability.
4208 		 */
4209 		VERIFY0(gcm_impl_set("cycle"));
4210 		VERIFY0(aes_impl_set("cycle"));
4211 
4212 		fnvlist_free(crypto_args);
4213 		fnvlist_free(props);
4214 	}
4215 
4216 	err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
4217 	    ztest_objset_create_cb, NULL);
4218 	dsl_crypto_params_free(dcp, !!err);
4219 
4220 	rand = ztest_random(100);
4221 	if (err || rand < 80)
4222 		return (err);
4223 
4224 	if (ztest_opts.zo_verbose >= 5)
4225 		(void) printf("Setting dataset %s to sync always\n", dsname);
4226 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
4227 	    ZFS_SYNC_ALWAYS, B_FALSE));
4228 }
4229 
4230 static int
4231 ztest_objset_destroy_cb(const char *name, void *arg)
4232 {
4233 	(void) arg;
4234 	objset_t *os;
4235 	dmu_object_info_t doi;
4236 	int error;
4237 
4238 	/*
4239 	 * Verify that the dataset contains a directory object.
4240 	 */
4241 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
4242 	    B_TRUE, FTAG, &os));
4243 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
4244 	if (error != ENOENT) {
4245 		/* We could have crashed in the middle of destroying it */
4246 		ASSERT0(error);
4247 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
4248 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
4249 	}
4250 	dmu_objset_disown(os, B_TRUE, FTAG);
4251 
4252 	/*
4253 	 * Destroy the dataset.
4254 	 */
4255 	if (strchr(name, '@') != NULL) {
4256 		error = dsl_destroy_snapshot(name, B_TRUE);
4257 		if (error != ECHRNG) {
4258 			/*
4259 			 * The program was executed, but encountered a runtime
4260 			 * error, such as insufficient slop, or a hold on the
4261 			 * dataset.
4262 			 */
4263 			ASSERT0(error);
4264 		}
4265 	} else {
4266 		error = dsl_destroy_head(name);
4267 		if (error == ENOSPC) {
4268 			/* There could be checkpoint or insufficient slop */
4269 			ztest_record_enospc(FTAG);
4270 		} else if (error != EBUSY) {
4271 			/* There could be a hold on this dataset */
4272 			ASSERT0(error);
4273 		}
4274 	}
4275 	return (0);
4276 }
4277 
4278 static boolean_t
4279 ztest_snapshot_create(char *osname, uint64_t id)
4280 {
4281 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4282 	int error;
4283 
4284 	(void) snprintf(snapname, sizeof (snapname), "%"PRIu64"", id);
4285 
4286 	error = dmu_objset_snapshot_one(osname, snapname);
4287 	if (error == ENOSPC) {
4288 		ztest_record_enospc(FTAG);
4289 		return (B_FALSE);
4290 	}
4291 	if (error != 0 && error != EEXIST) {
4292 		fatal(B_FALSE, "ztest_snapshot_create(%s@%s) = %d", osname,
4293 		    snapname, error);
4294 	}
4295 	return (B_TRUE);
4296 }
4297 
4298 static boolean_t
4299 ztest_snapshot_destroy(char *osname, uint64_t id)
4300 {
4301 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4302 	int error;
4303 
4304 	(void) snprintf(snapname, sizeof (snapname), "%s@%"PRIu64"",
4305 	    osname, id);
4306 
4307 	error = dsl_destroy_snapshot(snapname, B_FALSE);
4308 	if (error != 0 && error != ENOENT)
4309 		fatal(B_FALSE, "ztest_snapshot_destroy(%s) = %d",
4310 		    snapname, error);
4311 	return (B_TRUE);
4312 }
4313 
4314 void
4315 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
4316 {
4317 	(void) zd;
4318 	ztest_ds_t *zdtmp;
4319 	int iters;
4320 	int error;
4321 	objset_t *os, *os2;
4322 	char name[ZFS_MAX_DATASET_NAME_LEN];
4323 	zilog_t *zilog;
4324 	int i;
4325 
4326 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
4327 
4328 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4329 
4330 	(void) snprintf(name, sizeof (name), "%s/temp_%"PRIu64"",
4331 	    ztest_opts.zo_pool, id);
4332 
4333 	/*
4334 	 * If this dataset exists from a previous run, process its replay log
4335 	 * half of the time.  If we don't replay it, then dsl_destroy_head()
4336 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
4337 	 */
4338 	if (ztest_random(2) == 0 &&
4339 	    ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
4340 	    B_TRUE, FTAG, &os) == 0) {
4341 		ztest_zd_init(zdtmp, NULL, os);
4342 		zil_replay(os, zdtmp, ztest_replay_vector);
4343 		ztest_zd_fini(zdtmp);
4344 		dmu_objset_disown(os, B_TRUE, FTAG);
4345 	}
4346 
4347 	/*
4348 	 * There may be an old instance of the dataset we're about to
4349 	 * create lying around from a previous run.  If so, destroy it
4350 	 * and all of its snapshots.
4351 	 */
4352 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
4353 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
4354 
4355 	/*
4356 	 * Verify that the destroyed dataset is no longer in the namespace.
4357 	 */
4358 	VERIFY3U(ENOENT, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
4359 	    B_TRUE, FTAG, &os));
4360 
4361 	/*
4362 	 * Verify that we can create a new dataset.
4363 	 */
4364 	error = ztest_dataset_create(name);
4365 	if (error) {
4366 		if (error == ENOSPC) {
4367 			ztest_record_enospc(FTAG);
4368 			goto out;
4369 		}
4370 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", name, error);
4371 	}
4372 
4373 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
4374 	    FTAG, &os));
4375 
4376 	ztest_zd_init(zdtmp, NULL, os);
4377 
4378 	/*
4379 	 * Open the intent log for it.
4380 	 */
4381 	zilog = zil_open(os, ztest_get_data, NULL);
4382 
4383 	/*
4384 	 * Put some objects in there, do a little I/O to them,
4385 	 * and randomly take a couple of snapshots along the way.
4386 	 */
4387 	iters = ztest_random(5);
4388 	for (i = 0; i < iters; i++) {
4389 		ztest_dmu_object_alloc_free(zdtmp, id);
4390 		if (ztest_random(iters) == 0)
4391 			(void) ztest_snapshot_create(name, i);
4392 	}
4393 
4394 	/*
4395 	 * Verify that we cannot create an existing dataset.
4396 	 */
4397 	VERIFY3U(EEXIST, ==,
4398 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
4399 
4400 	/*
4401 	 * Verify that we can hold an objset that is also owned.
4402 	 */
4403 	VERIFY0(dmu_objset_hold(name, FTAG, &os2));
4404 	dmu_objset_rele(os2, FTAG);
4405 
4406 	/*
4407 	 * Verify that we cannot own an objset that is already owned.
4408 	 */
4409 	VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
4410 	    B_FALSE, B_TRUE, FTAG, &os2));
4411 
4412 	zil_close(zilog);
4413 	dmu_objset_disown(os, B_TRUE, FTAG);
4414 	ztest_zd_fini(zdtmp);
4415 out:
4416 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4417 
4418 	umem_free(zdtmp, sizeof (ztest_ds_t));
4419 }
4420 
4421 /*
4422  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
4423  */
4424 void
4425 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
4426 {
4427 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4428 	(void) ztest_snapshot_destroy(zd->zd_name, id);
4429 	(void) ztest_snapshot_create(zd->zd_name, id);
4430 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4431 }
4432 
4433 /*
4434  * Cleanup non-standard snapshots and clones.
4435  */
4436 static void
4437 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
4438 {
4439 	char *snap1name;
4440 	char *clone1name;
4441 	char *snap2name;
4442 	char *clone2name;
4443 	char *snap3name;
4444 	int error;
4445 
4446 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4447 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4448 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4449 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4450 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4451 
4452 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4453 	    osname, id);
4454 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4455 	    osname, id);
4456 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4457 	    clone1name, id);
4458 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4459 	    osname, id);
4460 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4461 	    clone1name, id);
4462 
4463 	error = dsl_destroy_head(clone2name);
4464 	if (error && error != ENOENT)
4465 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone2name, error);
4466 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
4467 	if (error && error != ENOENT)
4468 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4469 		    snap3name, error);
4470 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
4471 	if (error && error != ENOENT)
4472 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4473 		    snap2name, error);
4474 	error = dsl_destroy_head(clone1name);
4475 	if (error && error != ENOENT)
4476 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone1name, error);
4477 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
4478 	if (error && error != ENOENT)
4479 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4480 		    snap1name, error);
4481 
4482 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4483 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4484 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4485 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4486 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4487 }
4488 
4489 /*
4490  * Verify dsl_dataset_promote handles EBUSY
4491  */
4492 void
4493 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
4494 {
4495 	objset_t *os;
4496 	char *snap1name;
4497 	char *clone1name;
4498 	char *snap2name;
4499 	char *clone2name;
4500 	char *snap3name;
4501 	char *osname = zd->zd_name;
4502 	int error;
4503 
4504 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4505 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4506 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4507 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4508 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4509 
4510 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4511 
4512 	ztest_dsl_dataset_cleanup(osname, id);
4513 
4514 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4515 	    osname, id);
4516 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4517 	    osname, id);
4518 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4519 	    clone1name, id);
4520 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4521 	    osname, id);
4522 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4523 	    clone1name, id);
4524 
4525 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
4526 	if (error && error != EEXIST) {
4527 		if (error == ENOSPC) {
4528 			ztest_record_enospc(FTAG);
4529 			goto out;
4530 		}
4531 		fatal(B_FALSE, "dmu_take_snapshot(%s) = %d", snap1name, error);
4532 	}
4533 
4534 	error = dmu_objset_clone(clone1name, snap1name);
4535 	if (error) {
4536 		if (error == ENOSPC) {
4537 			ztest_record_enospc(FTAG);
4538 			goto out;
4539 		}
4540 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone1name, error);
4541 	}
4542 
4543 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
4544 	if (error && error != EEXIST) {
4545 		if (error == ENOSPC) {
4546 			ztest_record_enospc(FTAG);
4547 			goto out;
4548 		}
4549 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap2name, error);
4550 	}
4551 
4552 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
4553 	if (error && error != EEXIST) {
4554 		if (error == ENOSPC) {
4555 			ztest_record_enospc(FTAG);
4556 			goto out;
4557 		}
4558 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap3name, error);
4559 	}
4560 
4561 	error = dmu_objset_clone(clone2name, snap3name);
4562 	if (error) {
4563 		if (error == ENOSPC) {
4564 			ztest_record_enospc(FTAG);
4565 			goto out;
4566 		}
4567 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone2name, error);
4568 	}
4569 
4570 	error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
4571 	    FTAG, &os);
4572 	if (error)
4573 		fatal(B_FALSE, "dmu_objset_own(%s) = %d", snap2name, error);
4574 	error = dsl_dataset_promote(clone2name, NULL);
4575 	if (error == ENOSPC) {
4576 		dmu_objset_disown(os, B_TRUE, FTAG);
4577 		ztest_record_enospc(FTAG);
4578 		goto out;
4579 	}
4580 	if (error != EBUSY)
4581 		fatal(B_FALSE, "dsl_dataset_promote(%s), %d, not EBUSY",
4582 		    clone2name, error);
4583 	dmu_objset_disown(os, B_TRUE, FTAG);
4584 
4585 out:
4586 	ztest_dsl_dataset_cleanup(osname, id);
4587 
4588 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4589 
4590 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4591 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4592 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4593 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4594 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4595 }
4596 
4597 #undef OD_ARRAY_SIZE
4598 #define	OD_ARRAY_SIZE	4
4599 
4600 /*
4601  * Verify that dmu_object_{alloc,free} work as expected.
4602  */
4603 void
4604 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4605 {
4606 	ztest_od_t *od;
4607 	int batchsize;
4608 	int size;
4609 	int b;
4610 
4611 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4612 	od = umem_alloc(size, UMEM_NOFAIL);
4613 	batchsize = OD_ARRAY_SIZE;
4614 
4615 	for (b = 0; b < batchsize; b++)
4616 		ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
4617 		    0, 0, 0);
4618 
4619 	/*
4620 	 * Destroy the previous batch of objects, create a new batch,
4621 	 * and do some I/O on the new objects.
4622 	 */
4623 	if (ztest_object_init(zd, od, size, B_TRUE) != 0)
4624 		return;
4625 
4626 	while (ztest_random(4 * batchsize) != 0)
4627 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
4628 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4629 
4630 	umem_free(od, size);
4631 }
4632 
4633 /*
4634  * Rewind the global allocator to verify object allocation backfilling.
4635  */
4636 void
4637 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
4638 {
4639 	(void) id;
4640 	objset_t *os = zd->zd_os;
4641 	int dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
4642 	uint64_t object;
4643 
4644 	/*
4645 	 * Rewind the global allocator randomly back to a lower object number
4646 	 * to force backfilling and reclamation of recently freed dnodes.
4647 	 */
4648 	mutex_enter(&os->os_obj_lock);
4649 	object = ztest_random(os->os_obj_next_chunk);
4650 	os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk);
4651 	mutex_exit(&os->os_obj_lock);
4652 }
4653 
4654 #undef OD_ARRAY_SIZE
4655 #define	OD_ARRAY_SIZE	2
4656 
4657 /*
4658  * Verify that dmu_{read,write} work as expected.
4659  */
4660 void
4661 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
4662 {
4663 	int size;
4664 	ztest_od_t *od;
4665 
4666 	objset_t *os = zd->zd_os;
4667 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4668 	od = umem_alloc(size, UMEM_NOFAIL);
4669 	dmu_tx_t *tx;
4670 	int freeit, error;
4671 	uint64_t i, n, s, txg;
4672 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
4673 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4674 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
4675 	uint64_t regions = 997;
4676 	uint64_t stride = 123456789ULL;
4677 	uint64_t width = 40;
4678 	int free_percent = 5;
4679 
4680 	/*
4681 	 * This test uses two objects, packobj and bigobj, that are always
4682 	 * updated together (i.e. in the same tx) so that their contents are
4683 	 * in sync and can be compared.  Their contents relate to each other
4684 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4685 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4686 	 * for any index n, there are three bufwads that should be identical:
4687 	 *
4688 	 *	packobj, at offset n * sizeof (bufwad_t)
4689 	 *	bigobj, at the head of the nth chunk
4690 	 *	bigobj, at the tail of the nth chunk
4691 	 *
4692 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
4693 	 * and it doesn't have any relation to the object blocksize.
4694 	 * The only requirement is that it can hold at least two bufwads.
4695 	 *
4696 	 * Normally, we write the bufwad to each of these locations.
4697 	 * However, free_percent of the time we instead write zeroes to
4698 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
4699 	 * bigobj to packobj, we can verify that the DMU is correctly
4700 	 * tracking which parts of an object are allocated and free,
4701 	 * and that the contents of the allocated blocks are correct.
4702 	 */
4703 
4704 	/*
4705 	 * Read the directory info.  If it's the first time, set things up.
4706 	 */
4707 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
4708 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4709 	    chunksize);
4710 
4711 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4712 		umem_free(od, size);
4713 		return;
4714 	}
4715 
4716 	bigobj = od[0].od_object;
4717 	packobj = od[1].od_object;
4718 	chunksize = od[0].od_gen;
4719 	ASSERT3U(chunksize, ==, od[1].od_gen);
4720 
4721 	/*
4722 	 * Prefetch a random chunk of the big object.
4723 	 * Our aim here is to get some async reads in flight
4724 	 * for blocks that we may free below; the DMU should
4725 	 * handle this race correctly.
4726 	 */
4727 	n = ztest_random(regions) * stride + ztest_random(width);
4728 	s = 1 + ztest_random(2 * width - 1);
4729 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4730 	    ZIO_PRIORITY_SYNC_READ);
4731 
4732 	/*
4733 	 * Pick a random index and compute the offsets into packobj and bigobj.
4734 	 */
4735 	n = ztest_random(regions) * stride + ztest_random(width);
4736 	s = 1 + ztest_random(width - 1);
4737 
4738 	packoff = n * sizeof (bufwad_t);
4739 	packsize = s * sizeof (bufwad_t);
4740 
4741 	bigoff = n * chunksize;
4742 	bigsize = s * chunksize;
4743 
4744 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4745 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4746 
4747 	/*
4748 	 * free_percent of the time, free a range of bigobj rather than
4749 	 * overwriting it.
4750 	 */
4751 	freeit = (ztest_random(100) < free_percent);
4752 
4753 	/*
4754 	 * Read the current contents of our objects.
4755 	 */
4756 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
4757 	    DMU_READ_PREFETCH);
4758 	ASSERT0(error);
4759 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4760 	    DMU_READ_PREFETCH);
4761 	ASSERT0(error);
4762 
4763 	/*
4764 	 * Get a tx for the mods to both packobj and bigobj.
4765 	 */
4766 	tx = dmu_tx_create(os);
4767 
4768 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
4769 
4770 	if (freeit)
4771 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4772 	else
4773 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4774 
4775 	/* This accounts for setting the checksum/compression. */
4776 	dmu_tx_hold_bonus(tx, bigobj);
4777 
4778 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4779 	if (txg == 0) {
4780 		umem_free(packbuf, packsize);
4781 		umem_free(bigbuf, bigsize);
4782 		umem_free(od, size);
4783 		return;
4784 	}
4785 
4786 	enum zio_checksum cksum;
4787 	do {
4788 		cksum = (enum zio_checksum)
4789 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4790 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4791 	dmu_object_set_checksum(os, bigobj, cksum, tx);
4792 
4793 	enum zio_compress comp;
4794 	do {
4795 		comp = (enum zio_compress)
4796 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4797 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4798 	dmu_object_set_compress(os, bigobj, comp, tx);
4799 
4800 	/*
4801 	 * For each index from n to n + s, verify that the existing bufwad
4802 	 * in packobj matches the bufwads at the head and tail of the
4803 	 * corresponding chunk in bigobj.  Then update all three bufwads
4804 	 * with the new values we want to write out.
4805 	 */
4806 	for (i = 0; i < s; i++) {
4807 		/* LINTED */
4808 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4809 		/* LINTED */
4810 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4811 		/* LINTED */
4812 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4813 
4814 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
4815 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
4816 
4817 		if (pack->bw_txg > txg)
4818 			fatal(B_FALSE,
4819 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
4820 			    pack->bw_txg, txg);
4821 
4822 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4823 			fatal(B_FALSE, "wrong index: "
4824 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
4825 			    pack->bw_index, n, i);
4826 
4827 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4828 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
4829 			    pack, bigH);
4830 
4831 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4832 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
4833 			    pack, bigT);
4834 
4835 		if (freeit) {
4836 			memset(pack, 0, sizeof (bufwad_t));
4837 		} else {
4838 			pack->bw_index = n + i;
4839 			pack->bw_txg = txg;
4840 			pack->bw_data = 1 + ztest_random(-2ULL);
4841 		}
4842 		*bigH = *pack;
4843 		*bigT = *pack;
4844 	}
4845 
4846 	/*
4847 	 * We've verified all the old bufwads, and made new ones.
4848 	 * Now write them out.
4849 	 */
4850 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4851 
4852 	if (freeit) {
4853 		if (ztest_opts.zo_verbose >= 7) {
4854 			(void) printf("freeing offset %"PRIx64" size %"PRIx64""
4855 			    " txg %"PRIx64"\n",
4856 			    bigoff, bigsize, txg);
4857 		}
4858 		VERIFY0(dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4859 	} else {
4860 		if (ztest_opts.zo_verbose >= 7) {
4861 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
4862 			    " txg %"PRIx64"\n",
4863 			    bigoff, bigsize, txg);
4864 		}
4865 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4866 	}
4867 
4868 	dmu_tx_commit(tx);
4869 
4870 	/*
4871 	 * Sanity check the stuff we just wrote.
4872 	 */
4873 	{
4874 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4875 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4876 
4877 		VERIFY0(dmu_read(os, packobj, packoff,
4878 		    packsize, packcheck, DMU_READ_PREFETCH));
4879 		VERIFY0(dmu_read(os, bigobj, bigoff,
4880 		    bigsize, bigcheck, DMU_READ_PREFETCH));
4881 
4882 		ASSERT0(memcmp(packbuf, packcheck, packsize));
4883 		ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
4884 
4885 		umem_free(packcheck, packsize);
4886 		umem_free(bigcheck, bigsize);
4887 	}
4888 
4889 	umem_free(packbuf, packsize);
4890 	umem_free(bigbuf, bigsize);
4891 	umem_free(od, size);
4892 }
4893 
4894 static void
4895 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4896     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4897 {
4898 	uint64_t i;
4899 	bufwad_t *pack;
4900 	bufwad_t *bigH;
4901 	bufwad_t *bigT;
4902 
4903 	/*
4904 	 * For each index from n to n + s, verify that the existing bufwad
4905 	 * in packobj matches the bufwads at the head and tail of the
4906 	 * corresponding chunk in bigobj.  Then update all three bufwads
4907 	 * with the new values we want to write out.
4908 	 */
4909 	for (i = 0; i < s; i++) {
4910 		/* LINTED */
4911 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4912 		/* LINTED */
4913 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4914 		/* LINTED */
4915 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4916 
4917 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
4918 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
4919 
4920 		if (pack->bw_txg > txg)
4921 			fatal(B_FALSE,
4922 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
4923 			    pack->bw_txg, txg);
4924 
4925 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4926 			fatal(B_FALSE, "wrong index: "
4927 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
4928 			    pack->bw_index, n, i);
4929 
4930 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4931 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
4932 			    pack, bigH);
4933 
4934 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4935 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
4936 			    pack, bigT);
4937 
4938 		pack->bw_index = n + i;
4939 		pack->bw_txg = txg;
4940 		pack->bw_data = 1 + ztest_random(-2ULL);
4941 
4942 		*bigH = *pack;
4943 		*bigT = *pack;
4944 	}
4945 }
4946 
4947 #undef OD_ARRAY_SIZE
4948 #define	OD_ARRAY_SIZE	2
4949 
4950 void
4951 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4952 {
4953 	objset_t *os = zd->zd_os;
4954 	ztest_od_t *od;
4955 	dmu_tx_t *tx;
4956 	uint64_t i;
4957 	int error;
4958 	int size;
4959 	uint64_t n, s, txg;
4960 	bufwad_t *packbuf, *bigbuf;
4961 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4962 	uint64_t blocksize = ztest_random_blocksize();
4963 	uint64_t chunksize = blocksize;
4964 	uint64_t regions = 997;
4965 	uint64_t stride = 123456789ULL;
4966 	uint64_t width = 9;
4967 	dmu_buf_t *bonus_db;
4968 	arc_buf_t **bigbuf_arcbufs;
4969 	dmu_object_info_t doi;
4970 
4971 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4972 	od = umem_alloc(size, UMEM_NOFAIL);
4973 
4974 	/*
4975 	 * This test uses two objects, packobj and bigobj, that are always
4976 	 * updated together (i.e. in the same tx) so that their contents are
4977 	 * in sync and can be compared.  Their contents relate to each other
4978 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4979 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4980 	 * for any index n, there are three bufwads that should be identical:
4981 	 *
4982 	 *	packobj, at offset n * sizeof (bufwad_t)
4983 	 *	bigobj, at the head of the nth chunk
4984 	 *	bigobj, at the tail of the nth chunk
4985 	 *
4986 	 * The chunk size is set equal to bigobj block size so that
4987 	 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
4988 	 */
4989 
4990 	/*
4991 	 * Read the directory info.  If it's the first time, set things up.
4992 	 */
4993 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4994 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4995 	    chunksize);
4996 
4997 
4998 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4999 		umem_free(od, size);
5000 		return;
5001 	}
5002 
5003 	bigobj = od[0].od_object;
5004 	packobj = od[1].od_object;
5005 	blocksize = od[0].od_blocksize;
5006 	chunksize = blocksize;
5007 	ASSERT3U(chunksize, ==, od[1].od_gen);
5008 
5009 	VERIFY0(dmu_object_info(os, bigobj, &doi));
5010 	VERIFY(ISP2(doi.doi_data_block_size));
5011 	VERIFY3U(chunksize, ==, doi.doi_data_block_size);
5012 	VERIFY3U(chunksize, >=, 2 * sizeof (bufwad_t));
5013 
5014 	/*
5015 	 * Pick a random index and compute the offsets into packobj and bigobj.
5016 	 */
5017 	n = ztest_random(regions) * stride + ztest_random(width);
5018 	s = 1 + ztest_random(width - 1);
5019 
5020 	packoff = n * sizeof (bufwad_t);
5021 	packsize = s * sizeof (bufwad_t);
5022 
5023 	bigoff = n * chunksize;
5024 	bigsize = s * chunksize;
5025 
5026 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
5027 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
5028 
5029 	VERIFY0(dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
5030 
5031 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
5032 
5033 	/*
5034 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
5035 	 * Iteration 1 test zcopy to already referenced dbufs.
5036 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
5037 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
5038 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
5039 	 * Iteration 5 test zcopy when it can't be done.
5040 	 * Iteration 6 one more zcopy write.
5041 	 */
5042 	for (i = 0; i < 7; i++) {
5043 		uint64_t j;
5044 		uint64_t off;
5045 
5046 		/*
5047 		 * In iteration 5 (i == 5) use arcbufs
5048 		 * that don't match bigobj blksz to test
5049 		 * dmu_assign_arcbuf_by_dbuf() when it can't directly
5050 		 * assign an arcbuf to a dbuf.
5051 		 */
5052 		for (j = 0; j < s; j++) {
5053 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5054 				bigbuf_arcbufs[j] =
5055 				    dmu_request_arcbuf(bonus_db, chunksize);
5056 			} else {
5057 				bigbuf_arcbufs[2 * j] =
5058 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5059 				bigbuf_arcbufs[2 * j + 1] =
5060 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5061 			}
5062 		}
5063 
5064 		/*
5065 		 * Get a tx for the mods to both packobj and bigobj.
5066 		 */
5067 		tx = dmu_tx_create(os);
5068 
5069 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
5070 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
5071 
5072 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5073 		if (txg == 0) {
5074 			umem_free(packbuf, packsize);
5075 			umem_free(bigbuf, bigsize);
5076 			for (j = 0; j < s; j++) {
5077 				if (i != 5 ||
5078 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
5079 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
5080 				} else {
5081 					dmu_return_arcbuf(
5082 					    bigbuf_arcbufs[2 * j]);
5083 					dmu_return_arcbuf(
5084 					    bigbuf_arcbufs[2 * j + 1]);
5085 				}
5086 			}
5087 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5088 			umem_free(od, size);
5089 			dmu_buf_rele(bonus_db, FTAG);
5090 			return;
5091 		}
5092 
5093 		/*
5094 		 * 50% of the time don't read objects in the 1st iteration to
5095 		 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
5096 		 * no existing dbufs for the specified offsets.
5097 		 */
5098 		if (i != 0 || ztest_random(2) != 0) {
5099 			error = dmu_read(os, packobj, packoff,
5100 			    packsize, packbuf, DMU_READ_PREFETCH);
5101 			ASSERT0(error);
5102 			error = dmu_read(os, bigobj, bigoff, bigsize,
5103 			    bigbuf, DMU_READ_PREFETCH);
5104 			ASSERT0(error);
5105 		}
5106 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
5107 		    n, chunksize, txg);
5108 
5109 		/*
5110 		 * We've verified all the old bufwads, and made new ones.
5111 		 * Now write them out.
5112 		 */
5113 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
5114 		if (ztest_opts.zo_verbose >= 7) {
5115 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
5116 			    " txg %"PRIx64"\n",
5117 			    bigoff, bigsize, txg);
5118 		}
5119 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
5120 			dmu_buf_t *dbt;
5121 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5122 				memcpy(bigbuf_arcbufs[j]->b_data,
5123 				    (caddr_t)bigbuf + (off - bigoff),
5124 				    chunksize);
5125 			} else {
5126 				memcpy(bigbuf_arcbufs[2 * j]->b_data,
5127 				    (caddr_t)bigbuf + (off - bigoff),
5128 				    chunksize / 2);
5129 				memcpy(bigbuf_arcbufs[2 * j + 1]->b_data,
5130 				    (caddr_t)bigbuf + (off - bigoff) +
5131 				    chunksize / 2,
5132 				    chunksize / 2);
5133 			}
5134 
5135 			if (i == 1) {
5136 				VERIFY(dmu_buf_hold(os, bigobj, off,
5137 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
5138 			}
5139 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5140 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5141 				    off, bigbuf_arcbufs[j], tx));
5142 			} else {
5143 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5144 				    off, bigbuf_arcbufs[2 * j], tx));
5145 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5146 				    off + chunksize / 2,
5147 				    bigbuf_arcbufs[2 * j + 1], tx));
5148 			}
5149 			if (i == 1) {
5150 				dmu_buf_rele(dbt, FTAG);
5151 			}
5152 		}
5153 		dmu_tx_commit(tx);
5154 
5155 		/*
5156 		 * Sanity check the stuff we just wrote.
5157 		 */
5158 		{
5159 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
5160 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
5161 
5162 			VERIFY0(dmu_read(os, packobj, packoff,
5163 			    packsize, packcheck, DMU_READ_PREFETCH));
5164 			VERIFY0(dmu_read(os, bigobj, bigoff,
5165 			    bigsize, bigcheck, DMU_READ_PREFETCH));
5166 
5167 			ASSERT0(memcmp(packbuf, packcheck, packsize));
5168 			ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
5169 
5170 			umem_free(packcheck, packsize);
5171 			umem_free(bigcheck, bigsize);
5172 		}
5173 		if (i == 2) {
5174 			txg_wait_open(dmu_objset_pool(os), 0, B_TRUE);
5175 		} else if (i == 3) {
5176 			txg_wait_synced(dmu_objset_pool(os), 0);
5177 		}
5178 	}
5179 
5180 	dmu_buf_rele(bonus_db, FTAG);
5181 	umem_free(packbuf, packsize);
5182 	umem_free(bigbuf, bigsize);
5183 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5184 	umem_free(od, size);
5185 }
5186 
5187 void
5188 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
5189 {
5190 	(void) id;
5191 	ztest_od_t *od;
5192 
5193 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5194 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
5195 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5196 
5197 	/*
5198 	 * Have multiple threads write to large offsets in an object
5199 	 * to verify that parallel writes to an object -- even to the
5200 	 * same blocks within the object -- doesn't cause any trouble.
5201 	 */
5202 	ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5203 
5204 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
5205 		return;
5206 
5207 	while (ztest_random(10) != 0)
5208 		ztest_io(zd, od->od_object, offset);
5209 
5210 	umem_free(od, sizeof (ztest_od_t));
5211 }
5212 
5213 void
5214 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
5215 {
5216 	ztest_od_t *od;
5217 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
5218 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5219 	uint64_t count = ztest_random(20) + 1;
5220 	uint64_t blocksize = ztest_random_blocksize();
5221 	void *data;
5222 
5223 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5224 
5225 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5226 
5227 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5228 	    !ztest_random(2)) != 0) {
5229 		umem_free(od, sizeof (ztest_od_t));
5230 		return;
5231 	}
5232 
5233 	if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
5234 		umem_free(od, sizeof (ztest_od_t));
5235 		return;
5236 	}
5237 
5238 	ztest_prealloc(zd, od->od_object, offset, count * blocksize);
5239 
5240 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
5241 
5242 	while (ztest_random(count) != 0) {
5243 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
5244 		if (ztest_write(zd, od->od_object, randoff, blocksize,
5245 		    data) != 0)
5246 			break;
5247 		while (ztest_random(4) != 0)
5248 			ztest_io(zd, od->od_object, randoff);
5249 	}
5250 
5251 	umem_free(data, blocksize);
5252 	umem_free(od, sizeof (ztest_od_t));
5253 }
5254 
5255 /*
5256  * Verify that zap_{create,destroy,add,remove,update} work as expected.
5257  */
5258 #define	ZTEST_ZAP_MIN_INTS	1
5259 #define	ZTEST_ZAP_MAX_INTS	4
5260 #define	ZTEST_ZAP_MAX_PROPS	1000
5261 
5262 void
5263 ztest_zap(ztest_ds_t *zd, uint64_t id)
5264 {
5265 	objset_t *os = zd->zd_os;
5266 	ztest_od_t *od;
5267 	uint64_t object;
5268 	uint64_t txg, last_txg;
5269 	uint64_t value[ZTEST_ZAP_MAX_INTS];
5270 	uint64_t zl_ints, zl_intsize, prop;
5271 	int i, ints;
5272 	dmu_tx_t *tx;
5273 	char propname[100], txgname[100];
5274 	int error;
5275 	const char *const hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
5276 
5277 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5278 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5279 
5280 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5281 	    !ztest_random(2)) != 0)
5282 		goto out;
5283 
5284 	object = od->od_object;
5285 
5286 	/*
5287 	 * Generate a known hash collision, and verify that
5288 	 * we can lookup and remove both entries.
5289 	 */
5290 	tx = dmu_tx_create(os);
5291 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5292 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5293 	if (txg == 0)
5294 		goto out;
5295 	for (i = 0; i < 2; i++) {
5296 		value[i] = i;
5297 		VERIFY0(zap_add(os, object, hc[i], sizeof (uint64_t),
5298 		    1, &value[i], tx));
5299 	}
5300 	for (i = 0; i < 2; i++) {
5301 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
5302 		    sizeof (uint64_t), 1, &value[i], tx));
5303 		VERIFY0(
5304 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
5305 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5306 		ASSERT3U(zl_ints, ==, 1);
5307 	}
5308 	for (i = 0; i < 2; i++) {
5309 		VERIFY0(zap_remove(os, object, hc[i], tx));
5310 	}
5311 	dmu_tx_commit(tx);
5312 
5313 	/*
5314 	 * Generate a bunch of random entries.
5315 	 */
5316 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
5317 
5318 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5319 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5320 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5321 	memset(value, 0, sizeof (value));
5322 	last_txg = 0;
5323 
5324 	/*
5325 	 * If these zap entries already exist, validate their contents.
5326 	 */
5327 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5328 	if (error == 0) {
5329 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5330 		ASSERT3U(zl_ints, ==, 1);
5331 
5332 		VERIFY0(zap_lookup(os, object, txgname, zl_intsize,
5333 		    zl_ints, &last_txg));
5334 
5335 		VERIFY0(zap_length(os, object, propname, &zl_intsize,
5336 		    &zl_ints));
5337 
5338 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5339 		ASSERT3U(zl_ints, ==, ints);
5340 
5341 		VERIFY0(zap_lookup(os, object, propname, zl_intsize,
5342 		    zl_ints, value));
5343 
5344 		for (i = 0; i < ints; i++) {
5345 			ASSERT3U(value[i], ==, last_txg + object + i);
5346 		}
5347 	} else {
5348 		ASSERT3U(error, ==, ENOENT);
5349 	}
5350 
5351 	/*
5352 	 * Atomically update two entries in our zap object.
5353 	 * The first is named txg_%llu, and contains the txg
5354 	 * in which the property was last updated.  The second
5355 	 * is named prop_%llu, and the nth element of its value
5356 	 * should be txg + object + n.
5357 	 */
5358 	tx = dmu_tx_create(os);
5359 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5360 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5361 	if (txg == 0)
5362 		goto out;
5363 
5364 	if (last_txg > txg)
5365 		fatal(B_FALSE, "zap future leak: old %"PRIu64" new %"PRIu64"",
5366 		    last_txg, txg);
5367 
5368 	for (i = 0; i < ints; i++)
5369 		value[i] = txg + object + i;
5370 
5371 	VERIFY0(zap_update(os, object, txgname, sizeof (uint64_t),
5372 	    1, &txg, tx));
5373 	VERIFY0(zap_update(os, object, propname, sizeof (uint64_t),
5374 	    ints, value, tx));
5375 
5376 	dmu_tx_commit(tx);
5377 
5378 	/*
5379 	 * Remove a random pair of entries.
5380 	 */
5381 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5382 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5383 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5384 
5385 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5386 
5387 	if (error == ENOENT)
5388 		goto out;
5389 
5390 	ASSERT0(error);
5391 
5392 	tx = dmu_tx_create(os);
5393 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5394 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5395 	if (txg == 0)
5396 		goto out;
5397 	VERIFY0(zap_remove(os, object, txgname, tx));
5398 	VERIFY0(zap_remove(os, object, propname, tx));
5399 	dmu_tx_commit(tx);
5400 out:
5401 	umem_free(od, sizeof (ztest_od_t));
5402 }
5403 
5404 /*
5405  * Test case to test the upgrading of a microzap to fatzap.
5406  */
5407 void
5408 ztest_fzap(ztest_ds_t *zd, uint64_t id)
5409 {
5410 	objset_t *os = zd->zd_os;
5411 	ztest_od_t *od;
5412 	uint64_t object, txg, value;
5413 
5414 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5415 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5416 
5417 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5418 	    !ztest_random(2)) != 0)
5419 		goto out;
5420 	object = od->od_object;
5421 
5422 	/*
5423 	 * Add entries to this ZAP and make sure it spills over
5424 	 * and gets upgraded to a fatzap. Also, since we are adding
5425 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
5426 	 */
5427 	for (value = 0; value < 2050; value++) {
5428 		char name[ZFS_MAX_DATASET_NAME_LEN];
5429 		dmu_tx_t *tx;
5430 		int error;
5431 
5432 		(void) snprintf(name, sizeof (name), "fzap-%"PRIu64"-%"PRIu64"",
5433 		    id, value);
5434 
5435 		tx = dmu_tx_create(os);
5436 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
5437 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5438 		if (txg == 0)
5439 			goto out;
5440 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
5441 		    &value, tx);
5442 		ASSERT(error == 0 || error == EEXIST);
5443 		dmu_tx_commit(tx);
5444 	}
5445 out:
5446 	umem_free(od, sizeof (ztest_od_t));
5447 }
5448 
5449 void
5450 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
5451 {
5452 	(void) id;
5453 	objset_t *os = zd->zd_os;
5454 	ztest_od_t *od;
5455 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
5456 	dmu_tx_t *tx;
5457 	int i, namelen, error;
5458 	int micro = ztest_random(2);
5459 	char name[20], string_value[20];
5460 	void *data;
5461 
5462 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5463 	ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
5464 
5465 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5466 		umem_free(od, sizeof (ztest_od_t));
5467 		return;
5468 	}
5469 
5470 	object = od->od_object;
5471 
5472 	/*
5473 	 * Generate a random name of the form 'xxx.....' where each
5474 	 * x is a random printable character and the dots are dots.
5475 	 * There are 94 such characters, and the name length goes from
5476 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
5477 	 */
5478 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
5479 
5480 	for (i = 0; i < 3; i++)
5481 		name[i] = '!' + ztest_random('~' - '!' + 1);
5482 	for (; i < namelen - 1; i++)
5483 		name[i] = '.';
5484 	name[i] = '\0';
5485 
5486 	if ((namelen & 1) || micro) {
5487 		wsize = sizeof (txg);
5488 		wc = 1;
5489 		data = &txg;
5490 	} else {
5491 		wsize = 1;
5492 		wc = namelen;
5493 		data = string_value;
5494 	}
5495 
5496 	count = -1ULL;
5497 	VERIFY0(zap_count(os, object, &count));
5498 	ASSERT3S(count, !=, -1ULL);
5499 
5500 	/*
5501 	 * Select an operation: length, lookup, add, update, remove.
5502 	 */
5503 	i = ztest_random(5);
5504 
5505 	if (i >= 2) {
5506 		tx = dmu_tx_create(os);
5507 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5508 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5509 		if (txg == 0) {
5510 			umem_free(od, sizeof (ztest_od_t));
5511 			return;
5512 		}
5513 		memcpy(string_value, name, namelen);
5514 	} else {
5515 		tx = NULL;
5516 		txg = 0;
5517 		memset(string_value, 0, namelen);
5518 	}
5519 
5520 	switch (i) {
5521 
5522 	case 0:
5523 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
5524 		if (error == 0) {
5525 			ASSERT3U(wsize, ==, zl_wsize);
5526 			ASSERT3U(wc, ==, zl_wc);
5527 		} else {
5528 			ASSERT3U(error, ==, ENOENT);
5529 		}
5530 		break;
5531 
5532 	case 1:
5533 		error = zap_lookup(os, object, name, wsize, wc, data);
5534 		if (error == 0) {
5535 			if (data == string_value &&
5536 			    memcmp(name, data, namelen) != 0)
5537 				fatal(B_FALSE, "name '%s' != val '%s' len %d",
5538 				    name, (char *)data, namelen);
5539 		} else {
5540 			ASSERT3U(error, ==, ENOENT);
5541 		}
5542 		break;
5543 
5544 	case 2:
5545 		error = zap_add(os, object, name, wsize, wc, data, tx);
5546 		ASSERT(error == 0 || error == EEXIST);
5547 		break;
5548 
5549 	case 3:
5550 		VERIFY0(zap_update(os, object, name, wsize, wc, data, tx));
5551 		break;
5552 
5553 	case 4:
5554 		error = zap_remove(os, object, name, tx);
5555 		ASSERT(error == 0 || error == ENOENT);
5556 		break;
5557 	}
5558 
5559 	if (tx != NULL)
5560 		dmu_tx_commit(tx);
5561 
5562 	umem_free(od, sizeof (ztest_od_t));
5563 }
5564 
5565 /*
5566  * Commit callback data.
5567  */
5568 typedef struct ztest_cb_data {
5569 	list_node_t		zcd_node;
5570 	uint64_t		zcd_txg;
5571 	int			zcd_expected_err;
5572 	boolean_t		zcd_added;
5573 	boolean_t		zcd_called;
5574 	spa_t			*zcd_spa;
5575 } ztest_cb_data_t;
5576 
5577 /* This is the actual commit callback function */
5578 static void
5579 ztest_commit_callback(void *arg, int error)
5580 {
5581 	ztest_cb_data_t *data = arg;
5582 	uint64_t synced_txg;
5583 
5584 	VERIFY3P(data, !=, NULL);
5585 	VERIFY3S(data->zcd_expected_err, ==, error);
5586 	VERIFY(!data->zcd_called);
5587 
5588 	synced_txg = spa_last_synced_txg(data->zcd_spa);
5589 	if (data->zcd_txg > synced_txg)
5590 		fatal(B_FALSE,
5591 		    "commit callback of txg %"PRIu64" called prematurely, "
5592 		    "last synced txg = %"PRIu64"\n",
5593 		    data->zcd_txg, synced_txg);
5594 
5595 	data->zcd_called = B_TRUE;
5596 
5597 	if (error == ECANCELED) {
5598 		ASSERT0(data->zcd_txg);
5599 		ASSERT(!data->zcd_added);
5600 
5601 		/*
5602 		 * The private callback data should be destroyed here, but
5603 		 * since we are going to check the zcd_called field after
5604 		 * dmu_tx_abort(), we will destroy it there.
5605 		 */
5606 		return;
5607 	}
5608 
5609 	ASSERT(data->zcd_added);
5610 	ASSERT3U(data->zcd_txg, !=, 0);
5611 
5612 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5613 
5614 	/* See if this cb was called more quickly */
5615 	if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
5616 		zc_min_txg_delay = synced_txg - data->zcd_txg;
5617 
5618 	/* Remove our callback from the list */
5619 	list_remove(&zcl.zcl_callbacks, data);
5620 
5621 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5622 
5623 	umem_free(data, sizeof (ztest_cb_data_t));
5624 }
5625 
5626 /* Allocate and initialize callback data structure */
5627 static ztest_cb_data_t *
5628 ztest_create_cb_data(objset_t *os, uint64_t txg)
5629 {
5630 	ztest_cb_data_t *cb_data;
5631 
5632 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
5633 
5634 	cb_data->zcd_txg = txg;
5635 	cb_data->zcd_spa = dmu_objset_spa(os);
5636 	list_link_init(&cb_data->zcd_node);
5637 
5638 	return (cb_data);
5639 }
5640 
5641 /*
5642  * Commit callback test.
5643  */
5644 void
5645 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
5646 {
5647 	objset_t *os = zd->zd_os;
5648 	ztest_od_t *od;
5649 	dmu_tx_t *tx;
5650 	ztest_cb_data_t *cb_data[3], *tmp_cb;
5651 	uint64_t old_txg, txg;
5652 	int i, error = 0;
5653 
5654 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5655 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5656 
5657 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5658 		umem_free(od, sizeof (ztest_od_t));
5659 		return;
5660 	}
5661 
5662 	tx = dmu_tx_create(os);
5663 
5664 	cb_data[0] = ztest_create_cb_data(os, 0);
5665 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
5666 
5667 	dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
5668 
5669 	/* Every once in a while, abort the transaction on purpose */
5670 	if (ztest_random(100) == 0)
5671 		error = -1;
5672 
5673 	if (!error)
5674 		error = dmu_tx_assign(tx, TXG_NOWAIT);
5675 
5676 	txg = error ? 0 : dmu_tx_get_txg(tx);
5677 
5678 	cb_data[0]->zcd_txg = txg;
5679 	cb_data[1] = ztest_create_cb_data(os, txg);
5680 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
5681 
5682 	if (error) {
5683 		/*
5684 		 * It's not a strict requirement to call the registered
5685 		 * callbacks from inside dmu_tx_abort(), but that's what
5686 		 * it's supposed to happen in the current implementation
5687 		 * so we will check for that.
5688 		 */
5689 		for (i = 0; i < 2; i++) {
5690 			cb_data[i]->zcd_expected_err = ECANCELED;
5691 			VERIFY(!cb_data[i]->zcd_called);
5692 		}
5693 
5694 		dmu_tx_abort(tx);
5695 
5696 		for (i = 0; i < 2; i++) {
5697 			VERIFY(cb_data[i]->zcd_called);
5698 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
5699 		}
5700 
5701 		umem_free(od, sizeof (ztest_od_t));
5702 		return;
5703 	}
5704 
5705 	cb_data[2] = ztest_create_cb_data(os, txg);
5706 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
5707 
5708 	/*
5709 	 * Read existing data to make sure there isn't a future leak.
5710 	 */
5711 	VERIFY0(dmu_read(os, od->od_object, 0, sizeof (uint64_t),
5712 	    &old_txg, DMU_READ_PREFETCH));
5713 
5714 	if (old_txg > txg)
5715 		fatal(B_FALSE,
5716 		    "future leak: got %"PRIu64", open txg is %"PRIu64"",
5717 		    old_txg, txg);
5718 
5719 	dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
5720 
5721 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5722 
5723 	/*
5724 	 * Since commit callbacks don't have any ordering requirement and since
5725 	 * it is theoretically possible for a commit callback to be called
5726 	 * after an arbitrary amount of time has elapsed since its txg has been
5727 	 * synced, it is difficult to reliably determine whether a commit
5728 	 * callback hasn't been called due to high load or due to a flawed
5729 	 * implementation.
5730 	 *
5731 	 * In practice, we will assume that if after a certain number of txgs a
5732 	 * commit callback hasn't been called, then most likely there's an
5733 	 * implementation bug..
5734 	 */
5735 	tmp_cb = list_head(&zcl.zcl_callbacks);
5736 	if (tmp_cb != NULL &&
5737 	    tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
5738 		fatal(B_FALSE,
5739 		    "Commit callback threshold exceeded, "
5740 		    "oldest txg: %"PRIu64", open txg: %"PRIu64"\n",
5741 		    tmp_cb->zcd_txg, txg);
5742 	}
5743 
5744 	/*
5745 	 * Let's find the place to insert our callbacks.
5746 	 *
5747 	 * Even though the list is ordered by txg, it is possible for the
5748 	 * insertion point to not be the end because our txg may already be
5749 	 * quiescing at this point and other callbacks in the open txg
5750 	 * (from other objsets) may have sneaked in.
5751 	 */
5752 	tmp_cb = list_tail(&zcl.zcl_callbacks);
5753 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
5754 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
5755 
5756 	/* Add the 3 callbacks to the list */
5757 	for (i = 0; i < 3; i++) {
5758 		if (tmp_cb == NULL)
5759 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
5760 		else
5761 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
5762 			    cb_data[i]);
5763 
5764 		cb_data[i]->zcd_added = B_TRUE;
5765 		VERIFY(!cb_data[i]->zcd_called);
5766 
5767 		tmp_cb = cb_data[i];
5768 	}
5769 
5770 	zc_cb_counter += 3;
5771 
5772 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5773 
5774 	dmu_tx_commit(tx);
5775 
5776 	umem_free(od, sizeof (ztest_od_t));
5777 }
5778 
5779 /*
5780  * Visit each object in the dataset. Verify that its properties
5781  * are consistent what was stored in the block tag when it was created,
5782  * and that its unused bonus buffer space has not been overwritten.
5783  */
5784 void
5785 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5786 {
5787 	(void) id;
5788 	objset_t *os = zd->zd_os;
5789 	uint64_t obj;
5790 	int err = 0;
5791 
5792 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5793 		ztest_block_tag_t *bt = NULL;
5794 		dmu_object_info_t doi;
5795 		dmu_buf_t *db;
5796 
5797 		ztest_object_lock(zd, obj, RL_READER);
5798 		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) {
5799 			ztest_object_unlock(zd, obj);
5800 			continue;
5801 		}
5802 
5803 		dmu_object_info_from_db(db, &doi);
5804 		if (doi.doi_bonus_size >= sizeof (*bt))
5805 			bt = ztest_bt_bonus(db);
5806 
5807 		if (bt && bt->bt_magic == BT_MAGIC) {
5808 			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5809 			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
5810 			    bt->bt_crtxg);
5811 			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5812 		}
5813 
5814 		dmu_buf_rele(db, FTAG);
5815 		ztest_object_unlock(zd, obj);
5816 	}
5817 }
5818 
5819 void
5820 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5821 {
5822 	(void) id;
5823 	zfs_prop_t proplist[] = {
5824 		ZFS_PROP_CHECKSUM,
5825 		ZFS_PROP_COMPRESSION,
5826 		ZFS_PROP_COPIES,
5827 		ZFS_PROP_DEDUP
5828 	};
5829 
5830 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5831 
5832 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5833 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5834 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5835 
5836 	VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
5837 	    ztest_random_blocksize(), (int)ztest_random(2)));
5838 
5839 	(void) pthread_rwlock_unlock(&ztest_name_lock);
5840 }
5841 
5842 void
5843 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5844 {
5845 	(void) zd, (void) id;
5846 	nvlist_t *props = NULL;
5847 
5848 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5849 
5850 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_AUTOTRIM, ztest_random(2));
5851 
5852 	VERIFY0(spa_prop_get(ztest_spa, &props));
5853 
5854 	if (ztest_opts.zo_verbose >= 6)
5855 		dump_nvlist(props, 4);
5856 
5857 	fnvlist_free(props);
5858 
5859 	(void) pthread_rwlock_unlock(&ztest_name_lock);
5860 }
5861 
5862 static int
5863 user_release_one(const char *snapname, const char *holdname)
5864 {
5865 	nvlist_t *snaps, *holds;
5866 	int error;
5867 
5868 	snaps = fnvlist_alloc();
5869 	holds = fnvlist_alloc();
5870 	fnvlist_add_boolean(holds, holdname);
5871 	fnvlist_add_nvlist(snaps, snapname, holds);
5872 	fnvlist_free(holds);
5873 	error = dsl_dataset_user_release(snaps, NULL);
5874 	fnvlist_free(snaps);
5875 	return (error);
5876 }
5877 
5878 /*
5879  * Test snapshot hold/release and deferred destroy.
5880  */
5881 void
5882 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5883 {
5884 	int error;
5885 	objset_t *os = zd->zd_os;
5886 	objset_t *origin;
5887 	char snapname[100];
5888 	char fullname[100];
5889 	char clonename[100];
5890 	char tag[100];
5891 	char osname[ZFS_MAX_DATASET_NAME_LEN];
5892 	nvlist_t *holds;
5893 
5894 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5895 
5896 	dmu_objset_name(os, osname);
5897 
5898 	(void) snprintf(snapname, sizeof (snapname), "sh1_%"PRIu64"", id);
5899 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5900 	(void) snprintf(clonename, sizeof (clonename), "%s/ch1_%"PRIu64"",
5901 	    osname, id);
5902 	(void) snprintf(tag, sizeof (tag), "tag_%"PRIu64"", id);
5903 
5904 	/*
5905 	 * Clean up from any previous run.
5906 	 */
5907 	error = dsl_destroy_head(clonename);
5908 	if (error != ENOENT)
5909 		ASSERT0(error);
5910 	error = user_release_one(fullname, tag);
5911 	if (error != ESRCH && error != ENOENT)
5912 		ASSERT0(error);
5913 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5914 	if (error != ENOENT)
5915 		ASSERT0(error);
5916 
5917 	/*
5918 	 * Create snapshot, clone it, mark snap for deferred destroy,
5919 	 * destroy clone, verify snap was also destroyed.
5920 	 */
5921 	error = dmu_objset_snapshot_one(osname, snapname);
5922 	if (error) {
5923 		if (error == ENOSPC) {
5924 			ztest_record_enospc("dmu_objset_snapshot");
5925 			goto out;
5926 		}
5927 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
5928 	}
5929 
5930 	error = dmu_objset_clone(clonename, fullname);
5931 	if (error) {
5932 		if (error == ENOSPC) {
5933 			ztest_record_enospc("dmu_objset_clone");
5934 			goto out;
5935 		}
5936 		fatal(B_FALSE, "dmu_objset_clone(%s) = %d", clonename, error);
5937 	}
5938 
5939 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5940 	if (error) {
5941 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5942 		    fullname, error);
5943 	}
5944 
5945 	error = dsl_destroy_head(clonename);
5946 	if (error)
5947 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clonename, error);
5948 
5949 	error = dmu_objset_hold(fullname, FTAG, &origin);
5950 	if (error != ENOENT)
5951 		fatal(B_FALSE, "dmu_objset_hold(%s) = %d", fullname, error);
5952 
5953 	/*
5954 	 * Create snapshot, add temporary hold, verify that we can't
5955 	 * destroy a held snapshot, mark for deferred destroy,
5956 	 * release hold, verify snapshot was destroyed.
5957 	 */
5958 	error = dmu_objset_snapshot_one(osname, snapname);
5959 	if (error) {
5960 		if (error == ENOSPC) {
5961 			ztest_record_enospc("dmu_objset_snapshot");
5962 			goto out;
5963 		}
5964 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
5965 	}
5966 
5967 	holds = fnvlist_alloc();
5968 	fnvlist_add_string(holds, fullname, tag);
5969 	error = dsl_dataset_user_hold(holds, 0, NULL);
5970 	fnvlist_free(holds);
5971 
5972 	if (error == ENOSPC) {
5973 		ztest_record_enospc("dsl_dataset_user_hold");
5974 		goto out;
5975 	} else if (error) {
5976 		fatal(B_FALSE, "dsl_dataset_user_hold(%s, %s) = %u",
5977 		    fullname, tag, error);
5978 	}
5979 
5980 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5981 	if (error != EBUSY) {
5982 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5983 		    fullname, error);
5984 	}
5985 
5986 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5987 	if (error) {
5988 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5989 		    fullname, error);
5990 	}
5991 
5992 	error = user_release_one(fullname, tag);
5993 	if (error)
5994 		fatal(B_FALSE, "user_release_one(%s, %s) = %d",
5995 		    fullname, tag, error);
5996 
5997 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5998 
5999 out:
6000 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6001 }
6002 
6003 /*
6004  * Inject random faults into the on-disk data.
6005  */
6006 void
6007 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
6008 {
6009 	(void) zd, (void) id;
6010 	ztest_shared_t *zs = ztest_shared;
6011 	spa_t *spa = ztest_spa;
6012 	int fd;
6013 	uint64_t offset;
6014 	uint64_t leaves;
6015 	uint64_t bad = 0x1990c0ffeedecadeull;
6016 	uint64_t top, leaf;
6017 	char *path0;
6018 	char *pathrand;
6019 	size_t fsize;
6020 	int bshift = SPA_MAXBLOCKSHIFT + 2;
6021 	int iters = 1000;
6022 	int maxfaults;
6023 	int mirror_save;
6024 	vdev_t *vd0 = NULL;
6025 	uint64_t guid0 = 0;
6026 	boolean_t islog = B_FALSE;
6027 
6028 	path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6029 	pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6030 
6031 	mutex_enter(&ztest_vdev_lock);
6032 
6033 	/*
6034 	 * Device removal is in progress, fault injection must be disabled
6035 	 * until it completes and the pool is scrubbed.  The fault injection
6036 	 * strategy for damaging blocks does not take in to account evacuated
6037 	 * blocks which may have already been damaged.
6038 	 */
6039 	if (ztest_device_removal_active) {
6040 		mutex_exit(&ztest_vdev_lock);
6041 		goto out;
6042 	}
6043 
6044 	maxfaults = MAXFAULTS(zs);
6045 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raid_children;
6046 	mirror_save = zs->zs_mirrors;
6047 	mutex_exit(&ztest_vdev_lock);
6048 
6049 	ASSERT3U(leaves, >=, 1);
6050 
6051 	/*
6052 	 * While ztest is running the number of leaves will not change.  This
6053 	 * is critical for the fault injection logic as it determines where
6054 	 * errors can be safely injected such that they are always repairable.
6055 	 *
6056 	 * When restarting ztest a different number of leaves may be requested
6057 	 * which will shift the regions to be damaged.  This is fine as long
6058 	 * as the pool has been scrubbed prior to using the new mapping.
6059 	 * Failure to do can result in non-repairable damage being injected.
6060 	 */
6061 	if (ztest_pool_scrubbed == B_FALSE)
6062 		goto out;
6063 
6064 	/*
6065 	 * Grab the name lock as reader. There are some operations
6066 	 * which don't like to have their vdevs changed while
6067 	 * they are in progress (i.e. spa_change_guid). Those
6068 	 * operations will have grabbed the name lock as writer.
6069 	 */
6070 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6071 
6072 	/*
6073 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
6074 	 */
6075 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6076 
6077 	if (ztest_random(2) == 0) {
6078 		/*
6079 		 * Inject errors on a normal data device or slog device.
6080 		 */
6081 		top = ztest_random_vdev_top(spa, B_TRUE);
6082 		leaf = ztest_random(leaves) + zs->zs_splits;
6083 
6084 		/*
6085 		 * Generate paths to the first leaf in this top-level vdev,
6086 		 * and to the random leaf we selected.  We'll induce transient
6087 		 * write failures and random online/offline activity on leaf 0,
6088 		 * and we'll write random garbage to the randomly chosen leaf.
6089 		 */
6090 		(void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
6091 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6092 		    top * leaves + zs->zs_splits);
6093 		(void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
6094 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6095 		    top * leaves + leaf);
6096 
6097 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
6098 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
6099 			islog = B_TRUE;
6100 
6101 		/*
6102 		 * If the top-level vdev needs to be resilvered
6103 		 * then we only allow faults on the device that is
6104 		 * resilvering.
6105 		 */
6106 		if (vd0 != NULL && maxfaults != 1 &&
6107 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
6108 		    vd0->vdev_resilver_txg != 0)) {
6109 			/*
6110 			 * Make vd0 explicitly claim to be unreadable,
6111 			 * or unwritable, or reach behind its back
6112 			 * and close the underlying fd.  We can do this if
6113 			 * maxfaults == 0 because we'll fail and reexecute,
6114 			 * and we can do it if maxfaults >= 2 because we'll
6115 			 * have enough redundancy.  If maxfaults == 1, the
6116 			 * combination of this with injection of random data
6117 			 * corruption below exceeds the pool's fault tolerance.
6118 			 */
6119 			vdev_file_t *vf = vd0->vdev_tsd;
6120 
6121 			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
6122 			    (long long)vd0->vdev_id, (int)maxfaults);
6123 
6124 			if (vf != NULL && ztest_random(3) == 0) {
6125 				(void) close(vf->vf_file->f_fd);
6126 				vf->vf_file->f_fd = -1;
6127 			} else if (ztest_random(2) == 0) {
6128 				vd0->vdev_cant_read = B_TRUE;
6129 			} else {
6130 				vd0->vdev_cant_write = B_TRUE;
6131 			}
6132 			guid0 = vd0->vdev_guid;
6133 		}
6134 	} else {
6135 		/*
6136 		 * Inject errors on an l2cache device.
6137 		 */
6138 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
6139 
6140 		if (sav->sav_count == 0) {
6141 			spa_config_exit(spa, SCL_STATE, FTAG);
6142 			(void) pthread_rwlock_unlock(&ztest_name_lock);
6143 			goto out;
6144 		}
6145 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
6146 		guid0 = vd0->vdev_guid;
6147 		(void) strcpy(path0, vd0->vdev_path);
6148 		(void) strcpy(pathrand, vd0->vdev_path);
6149 
6150 		leaf = 0;
6151 		leaves = 1;
6152 		maxfaults = INT_MAX;	/* no limit on cache devices */
6153 	}
6154 
6155 	spa_config_exit(spa, SCL_STATE, FTAG);
6156 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6157 
6158 	/*
6159 	 * If we can tolerate two or more faults, or we're dealing
6160 	 * with a slog, randomly online/offline vd0.
6161 	 */
6162 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
6163 		if (ztest_random(10) < 6) {
6164 			int flags = (ztest_random(2) == 0 ?
6165 			    ZFS_OFFLINE_TEMPORARY : 0);
6166 
6167 			/*
6168 			 * We have to grab the zs_name_lock as writer to
6169 			 * prevent a race between offlining a slog and
6170 			 * destroying a dataset. Offlining the slog will
6171 			 * grab a reference on the dataset which may cause
6172 			 * dsl_destroy_head() to fail with EBUSY thus
6173 			 * leaving the dataset in an inconsistent state.
6174 			 */
6175 			if (islog)
6176 				(void) pthread_rwlock_wrlock(&ztest_name_lock);
6177 
6178 			VERIFY3U(vdev_offline(spa, guid0, flags), !=, EBUSY);
6179 
6180 			if (islog)
6181 				(void) pthread_rwlock_unlock(&ztest_name_lock);
6182 		} else {
6183 			/*
6184 			 * Ideally we would like to be able to randomly
6185 			 * call vdev_[on|off]line without holding locks
6186 			 * to force unpredictable failures but the side
6187 			 * effects of vdev_[on|off]line prevent us from
6188 			 * doing so. We grab the ztest_vdev_lock here to
6189 			 * prevent a race between injection testing and
6190 			 * aux_vdev removal.
6191 			 */
6192 			mutex_enter(&ztest_vdev_lock);
6193 			(void) vdev_online(spa, guid0, 0, NULL);
6194 			mutex_exit(&ztest_vdev_lock);
6195 		}
6196 	}
6197 
6198 	if (maxfaults == 0)
6199 		goto out;
6200 
6201 	/*
6202 	 * We have at least single-fault tolerance, so inject data corruption.
6203 	 */
6204 	fd = open(pathrand, O_RDWR);
6205 
6206 	if (fd == -1) /* we hit a gap in the device namespace */
6207 		goto out;
6208 
6209 	fsize = lseek(fd, 0, SEEK_END);
6210 
6211 	while (--iters != 0) {
6212 		/*
6213 		 * The offset must be chosen carefully to ensure that
6214 		 * we do not inject a given logical block with errors
6215 		 * on two different leaf devices, because ZFS can not
6216 		 * tolerate that (if maxfaults==1).
6217 		 *
6218 		 * To achieve this we divide each leaf device into
6219 		 * chunks of size (# leaves * SPA_MAXBLOCKSIZE * 4).
6220 		 * Each chunk is further divided into error-injection
6221 		 * ranges (can accept errors) and clear ranges (we do
6222 		 * not inject errors in those). Each error-injection
6223 		 * range can accept errors only for a single leaf vdev.
6224 		 * Error-injection ranges are separated by clear ranges.
6225 		 *
6226 		 * For example, with 3 leaves, each chunk looks like:
6227 		 *    0 to  32M: injection range for leaf 0
6228 		 *  32M to  64M: clear range - no injection allowed
6229 		 *  64M to  96M: injection range for leaf 1
6230 		 *  96M to 128M: clear range - no injection allowed
6231 		 * 128M to 160M: injection range for leaf 2
6232 		 * 160M to 192M: clear range - no injection allowed
6233 		 *
6234 		 * Each clear range must be large enough such that a
6235 		 * single block cannot straddle it. This way a block
6236 		 * can't be a target in two different injection ranges
6237 		 * (on different leaf vdevs).
6238 		 */
6239 		offset = ztest_random(fsize / (leaves << bshift)) *
6240 		    (leaves << bshift) + (leaf << bshift) +
6241 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
6242 
6243 		/*
6244 		 * Only allow damage to the labels at one end of the vdev.
6245 		 *
6246 		 * If all labels are damaged, the device will be totally
6247 		 * inaccessible, which will result in loss of data,
6248 		 * because we also damage (parts of) the other side of
6249 		 * the mirror/raidz.
6250 		 *
6251 		 * Additionally, we will always have both an even and an
6252 		 * odd label, so that we can handle crashes in the
6253 		 * middle of vdev_config_sync().
6254 		 */
6255 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
6256 			continue;
6257 
6258 		/*
6259 		 * The two end labels are stored at the "end" of the disk, but
6260 		 * the end of the disk (vdev_psize) is aligned to
6261 		 * sizeof (vdev_label_t).
6262 		 */
6263 		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
6264 		if ((leaf & 1) == 1 &&
6265 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
6266 			continue;
6267 
6268 		mutex_enter(&ztest_vdev_lock);
6269 		if (mirror_save != zs->zs_mirrors) {
6270 			mutex_exit(&ztest_vdev_lock);
6271 			(void) close(fd);
6272 			goto out;
6273 		}
6274 
6275 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
6276 			fatal(B_TRUE,
6277 			    "can't inject bad word at 0x%"PRIx64" in %s",
6278 			    offset, pathrand);
6279 
6280 		mutex_exit(&ztest_vdev_lock);
6281 
6282 		if (ztest_opts.zo_verbose >= 7)
6283 			(void) printf("injected bad word into %s,"
6284 			    " offset 0x%"PRIx64"\n", pathrand, offset);
6285 	}
6286 
6287 	(void) close(fd);
6288 out:
6289 	umem_free(path0, MAXPATHLEN);
6290 	umem_free(pathrand, MAXPATHLEN);
6291 }
6292 
6293 /*
6294  * By design ztest will never inject uncorrectable damage in to the pool.
6295  * Issue a scrub, wait for it to complete, and verify there is never any
6296  * persistent damage.
6297  *
6298  * Only after a full scrub has been completed is it safe to start injecting
6299  * data corruption.  See the comment in zfs_fault_inject().
6300  */
6301 static int
6302 ztest_scrub_impl(spa_t *spa)
6303 {
6304 	int error = spa_scan(spa, POOL_SCAN_SCRUB);
6305 	if (error)
6306 		return (error);
6307 
6308 	while (dsl_scan_scrubbing(spa_get_dsl(spa)))
6309 		txg_wait_synced(spa_get_dsl(spa), 0);
6310 
6311 	if (spa_get_errlog_size(spa) > 0)
6312 		return (ECKSUM);
6313 
6314 	ztest_pool_scrubbed = B_TRUE;
6315 
6316 	return (0);
6317 }
6318 
6319 /*
6320  * Scrub the pool.
6321  */
6322 void
6323 ztest_scrub(ztest_ds_t *zd, uint64_t id)
6324 {
6325 	(void) zd, (void) id;
6326 	spa_t *spa = ztest_spa;
6327 	int error;
6328 
6329 	/*
6330 	 * Scrub in progress by device removal.
6331 	 */
6332 	if (ztest_device_removal_active)
6333 		return;
6334 
6335 	/*
6336 	 * Start a scrub, wait a moment, then force a restart.
6337 	 */
6338 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
6339 	(void) poll(NULL, 0, 100);
6340 
6341 	error = ztest_scrub_impl(spa);
6342 	if (error == EBUSY)
6343 		error = 0;
6344 	ASSERT0(error);
6345 }
6346 
6347 /*
6348  * Change the guid for the pool.
6349  */
6350 void
6351 ztest_reguid(ztest_ds_t *zd, uint64_t id)
6352 {
6353 	(void) zd, (void) id;
6354 	spa_t *spa = ztest_spa;
6355 	uint64_t orig, load;
6356 	int error;
6357 
6358 	if (ztest_opts.zo_mmp_test)
6359 		return;
6360 
6361 	orig = spa_guid(spa);
6362 	load = spa_load_guid(spa);
6363 
6364 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
6365 	error = spa_change_guid(spa);
6366 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6367 
6368 	if (error != 0)
6369 		return;
6370 
6371 	if (ztest_opts.zo_verbose >= 4) {
6372 		(void) printf("Changed guid old %"PRIu64" -> %"PRIu64"\n",
6373 		    orig, spa_guid(spa));
6374 	}
6375 
6376 	VERIFY3U(orig, !=, spa_guid(spa));
6377 	VERIFY3U(load, ==, spa_load_guid(spa));
6378 }
6379 
6380 void
6381 ztest_blake3(ztest_ds_t *zd, uint64_t id)
6382 {
6383 	(void) zd, (void) id;
6384 	hrtime_t end = gethrtime() + NANOSEC;
6385 	zio_cksum_salt_t salt;
6386 	void *salt_ptr = &salt.zcs_bytes;
6387 	struct abd *abd_data, *abd_meta;
6388 	void *buf, *templ;
6389 	int i, *ptr;
6390 	uint32_t size;
6391 	BLAKE3_CTX ctx;
6392 
6393 	size = ztest_random_blocksize();
6394 	buf = umem_alloc(size, UMEM_NOFAIL);
6395 	abd_data = abd_alloc(size, B_FALSE);
6396 	abd_meta = abd_alloc(size, B_TRUE);
6397 
6398 	for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6399 		*ptr = ztest_random(UINT_MAX);
6400 	memset(salt_ptr, 'A', 32);
6401 
6402 	abd_copy_from_buf_off(abd_data, buf, 0, size);
6403 	abd_copy_from_buf_off(abd_meta, buf, 0, size);
6404 
6405 	while (gethrtime() <= end) {
6406 		int run_count = 100;
6407 		zio_cksum_t zc_ref1, zc_ref2;
6408 		zio_cksum_t zc_res1, zc_res2;
6409 
6410 		void *ref1 = &zc_ref1;
6411 		void *ref2 = &zc_ref2;
6412 		void *res1 = &zc_res1;
6413 		void *res2 = &zc_res2;
6414 
6415 		/* BLAKE3_KEY_LEN = 32 */
6416 		VERIFY0(blake3_set_impl_name("generic"));
6417 		templ = abd_checksum_blake3_tmpl_init(&salt);
6418 		Blake3_InitKeyed(&ctx, salt_ptr);
6419 		Blake3_Update(&ctx, buf, size);
6420 		Blake3_Final(&ctx, ref1);
6421 		zc_ref2 = zc_ref1;
6422 		ZIO_CHECKSUM_BSWAP(&zc_ref2);
6423 		abd_checksum_blake3_tmpl_free(templ);
6424 
6425 		VERIFY0(blake3_set_impl_name("cycle"));
6426 		while (run_count-- > 0) {
6427 
6428 			/* Test current implementation */
6429 			Blake3_InitKeyed(&ctx, salt_ptr);
6430 			Blake3_Update(&ctx, buf, size);
6431 			Blake3_Final(&ctx, res1);
6432 			zc_res2 = zc_res1;
6433 			ZIO_CHECKSUM_BSWAP(&zc_res2);
6434 
6435 			VERIFY0(memcmp(ref1, res1, 32));
6436 			VERIFY0(memcmp(ref2, res2, 32));
6437 
6438 			/* Test ABD - data */
6439 			templ = abd_checksum_blake3_tmpl_init(&salt);
6440 			abd_checksum_blake3_native(abd_data, size,
6441 			    templ, &zc_res1);
6442 			abd_checksum_blake3_byteswap(abd_data, size,
6443 			    templ, &zc_res2);
6444 
6445 			VERIFY0(memcmp(ref1, res1, 32));
6446 			VERIFY0(memcmp(ref2, res2, 32));
6447 
6448 			/* Test ABD - metadata */
6449 			abd_checksum_blake3_native(abd_meta, size,
6450 			    templ, &zc_res1);
6451 			abd_checksum_blake3_byteswap(abd_meta, size,
6452 			    templ, &zc_res2);
6453 			abd_checksum_blake3_tmpl_free(templ);
6454 
6455 			VERIFY0(memcmp(ref1, res1, 32));
6456 			VERIFY0(memcmp(ref2, res2, 32));
6457 
6458 		}
6459 	}
6460 
6461 	abd_free(abd_data);
6462 	abd_free(abd_meta);
6463 	umem_free(buf, size);
6464 }
6465 
6466 void
6467 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
6468 {
6469 	(void) zd, (void) id;
6470 	hrtime_t end = gethrtime() + NANOSEC;
6471 
6472 	while (gethrtime() <= end) {
6473 		int run_count = 100;
6474 		void *buf;
6475 		struct abd *abd_data, *abd_meta;
6476 		uint32_t size;
6477 		int *ptr;
6478 		int i;
6479 		zio_cksum_t zc_ref;
6480 		zio_cksum_t zc_ref_byteswap;
6481 
6482 		size = ztest_random_blocksize();
6483 
6484 		buf = umem_alloc(size, UMEM_NOFAIL);
6485 		abd_data = abd_alloc(size, B_FALSE);
6486 		abd_meta = abd_alloc(size, B_TRUE);
6487 
6488 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6489 			*ptr = ztest_random(UINT_MAX);
6490 
6491 		abd_copy_from_buf_off(abd_data, buf, 0, size);
6492 		abd_copy_from_buf_off(abd_meta, buf, 0, size);
6493 
6494 		VERIFY0(fletcher_4_impl_set("scalar"));
6495 		fletcher_4_native(buf, size, NULL, &zc_ref);
6496 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
6497 
6498 		VERIFY0(fletcher_4_impl_set("cycle"));
6499 		while (run_count-- > 0) {
6500 			zio_cksum_t zc;
6501 			zio_cksum_t zc_byteswap;
6502 
6503 			fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
6504 			fletcher_4_native(buf, size, NULL, &zc);
6505 
6506 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6507 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6508 			    sizeof (zc_byteswap)));
6509 
6510 			/* Test ABD - data */
6511 			abd_fletcher_4_byteswap(abd_data, size, NULL,
6512 			    &zc_byteswap);
6513 			abd_fletcher_4_native(abd_data, size, NULL, &zc);
6514 
6515 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6516 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6517 			    sizeof (zc_byteswap)));
6518 
6519 			/* Test ABD - metadata */
6520 			abd_fletcher_4_byteswap(abd_meta, size, NULL,
6521 			    &zc_byteswap);
6522 			abd_fletcher_4_native(abd_meta, size, NULL, &zc);
6523 
6524 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6525 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6526 			    sizeof (zc_byteswap)));
6527 
6528 		}
6529 
6530 		umem_free(buf, size);
6531 		abd_free(abd_data);
6532 		abd_free(abd_meta);
6533 	}
6534 }
6535 
6536 void
6537 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
6538 {
6539 	(void) zd, (void) id;
6540 	void *buf;
6541 	size_t size;
6542 	int *ptr;
6543 	int i;
6544 	zio_cksum_t zc_ref;
6545 	zio_cksum_t zc_ref_bswap;
6546 
6547 	hrtime_t end = gethrtime() + NANOSEC;
6548 
6549 	while (gethrtime() <= end) {
6550 		int run_count = 100;
6551 
6552 		size = ztest_random_blocksize();
6553 		buf = umem_alloc(size, UMEM_NOFAIL);
6554 
6555 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6556 			*ptr = ztest_random(UINT_MAX);
6557 
6558 		VERIFY0(fletcher_4_impl_set("scalar"));
6559 		fletcher_4_native(buf, size, NULL, &zc_ref);
6560 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
6561 
6562 		VERIFY0(fletcher_4_impl_set("cycle"));
6563 
6564 		while (run_count-- > 0) {
6565 			zio_cksum_t zc;
6566 			zio_cksum_t zc_bswap;
6567 			size_t pos = 0;
6568 
6569 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6570 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6571 
6572 			while (pos < size) {
6573 				size_t inc = 64 * ztest_random(size / 67);
6574 				/* sometimes add few bytes to test non-simd */
6575 				if (ztest_random(100) < 10)
6576 					inc += P2ALIGN(ztest_random(64),
6577 					    sizeof (uint32_t));
6578 
6579 				if (inc > (size - pos))
6580 					inc = size - pos;
6581 
6582 				fletcher_4_incremental_native(buf + pos, inc,
6583 				    &zc);
6584 				fletcher_4_incremental_byteswap(buf + pos, inc,
6585 				    &zc_bswap);
6586 
6587 				pos += inc;
6588 			}
6589 
6590 			VERIFY3U(pos, ==, size);
6591 
6592 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6593 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6594 
6595 			/*
6596 			 * verify if incremental on the whole buffer is
6597 			 * equivalent to non-incremental version
6598 			 */
6599 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6600 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6601 
6602 			fletcher_4_incremental_native(buf, size, &zc);
6603 			fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
6604 
6605 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6606 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6607 		}
6608 
6609 		umem_free(buf, size);
6610 	}
6611 }
6612 
6613 static int
6614 ztest_set_global_vars(void)
6615 {
6616 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
6617 		char *kv = ztest_opts.zo_gvars[i];
6618 		VERIFY3U(strlen(kv), <=, ZO_GVARS_MAX_ARGLEN);
6619 		VERIFY3U(strlen(kv), >, 0);
6620 		int err = set_global_var(kv);
6621 		if (ztest_opts.zo_verbose > 0) {
6622 			(void) printf("setting global var %s ... %s\n", kv,
6623 			    err ? "failed" : "ok");
6624 		}
6625 		if (err != 0) {
6626 			(void) fprintf(stderr,
6627 			    "failed to set global var '%s'\n", kv);
6628 			return (err);
6629 		}
6630 	}
6631 	return (0);
6632 }
6633 
6634 static char **
6635 ztest_global_vars_to_zdb_args(void)
6636 {
6637 	char **args = calloc(2*ztest_opts.zo_gvars_count + 1, sizeof (char *));
6638 	char **cur = args;
6639 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
6640 		*cur++ = (char *)"-o";
6641 		*cur++ = ztest_opts.zo_gvars[i];
6642 	}
6643 	ASSERT3P(cur, ==, &args[2*ztest_opts.zo_gvars_count]);
6644 	*cur = NULL;
6645 	return (args);
6646 }
6647 
6648 /* The end of strings is indicated by a NULL element */
6649 static char *
6650 join_strings(char **strings, const char *sep)
6651 {
6652 	size_t totallen = 0;
6653 	for (char **sp = strings; *sp != NULL; sp++) {
6654 		totallen += strlen(*sp);
6655 		totallen += strlen(sep);
6656 	}
6657 	if (totallen > 0) {
6658 		ASSERT(totallen >= strlen(sep));
6659 		totallen -= strlen(sep);
6660 	}
6661 
6662 	size_t buflen = totallen + 1;
6663 	char *o = malloc(buflen); /* trailing 0 byte */
6664 	o[0] = '\0';
6665 	for (char **sp = strings; *sp != NULL; sp++) {
6666 		size_t would;
6667 		would = strlcat(o, *sp, buflen);
6668 		VERIFY3U(would, <, buflen);
6669 		if (*(sp+1) == NULL) {
6670 			break;
6671 		}
6672 		would = strlcat(o, sep, buflen);
6673 		VERIFY3U(would, <, buflen);
6674 	}
6675 	ASSERT3S(strlen(o), ==, totallen);
6676 	return (o);
6677 }
6678 
6679 static int
6680 ztest_check_path(char *path)
6681 {
6682 	struct stat s;
6683 	/* return true on success */
6684 	return (!stat(path, &s));
6685 }
6686 
6687 static void
6688 ztest_get_zdb_bin(char *bin, int len)
6689 {
6690 	char *zdb_path;
6691 	/*
6692 	 * Try to use $ZDB and in-tree zdb path. If not successful, just
6693 	 * let popen to search through PATH.
6694 	 */
6695 	if ((zdb_path = getenv("ZDB"))) {
6696 		strlcpy(bin, zdb_path, len); /* In env */
6697 		if (!ztest_check_path(bin)) {
6698 			ztest_dump_core = 0;
6699 			fatal(B_TRUE, "invalid ZDB '%s'", bin);
6700 		}
6701 		return;
6702 	}
6703 
6704 	VERIFY3P(realpath(getexecname(), bin), !=, NULL);
6705 	if (strstr(bin, ".libs/ztest")) {
6706 		strstr(bin, ".libs/ztest")[0] = '\0'; /* In-tree */
6707 		strcat(bin, "zdb");
6708 		if (ztest_check_path(bin))
6709 			return;
6710 	}
6711 	strcpy(bin, "zdb");
6712 }
6713 
6714 static vdev_t *
6715 ztest_random_concrete_vdev_leaf(vdev_t *vd)
6716 {
6717 	if (vd == NULL)
6718 		return (NULL);
6719 
6720 	if (vd->vdev_children == 0)
6721 		return (vd);
6722 
6723 	vdev_t *eligible[vd->vdev_children];
6724 	int eligible_idx = 0, i;
6725 	for (i = 0; i < vd->vdev_children; i++) {
6726 		vdev_t *cvd = vd->vdev_child[i];
6727 		if (cvd->vdev_top->vdev_removing)
6728 			continue;
6729 		if (cvd->vdev_children > 0 ||
6730 		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
6731 			eligible[eligible_idx++] = cvd;
6732 		}
6733 	}
6734 	VERIFY3S(eligible_idx, >, 0);
6735 
6736 	uint64_t child_no = ztest_random(eligible_idx);
6737 	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
6738 }
6739 
6740 void
6741 ztest_initialize(ztest_ds_t *zd, uint64_t id)
6742 {
6743 	(void) zd, (void) id;
6744 	spa_t *spa = ztest_spa;
6745 	int error = 0;
6746 
6747 	mutex_enter(&ztest_vdev_lock);
6748 
6749 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6750 
6751 	/* Random leaf vdev */
6752 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6753 	if (rand_vd == NULL) {
6754 		spa_config_exit(spa, SCL_VDEV, FTAG);
6755 		mutex_exit(&ztest_vdev_lock);
6756 		return;
6757 	}
6758 
6759 	/*
6760 	 * The random vdev we've selected may change as soon as we
6761 	 * drop the spa_config_lock. We create local copies of things
6762 	 * we're interested in.
6763 	 */
6764 	uint64_t guid = rand_vd->vdev_guid;
6765 	char *path = strdup(rand_vd->vdev_path);
6766 	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
6767 
6768 	zfs_dbgmsg("vd %px, guid %llu", rand_vd, (u_longlong_t)guid);
6769 	spa_config_exit(spa, SCL_VDEV, FTAG);
6770 
6771 	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
6772 
6773 	nvlist_t *vdev_guids = fnvlist_alloc();
6774 	nvlist_t *vdev_errlist = fnvlist_alloc();
6775 	fnvlist_add_uint64(vdev_guids, path, guid);
6776 	error = spa_vdev_initialize(spa, vdev_guids, cmd, vdev_errlist);
6777 	fnvlist_free(vdev_guids);
6778 	fnvlist_free(vdev_errlist);
6779 
6780 	switch (cmd) {
6781 	case POOL_INITIALIZE_CANCEL:
6782 		if (ztest_opts.zo_verbose >= 4) {
6783 			(void) printf("Cancel initialize %s", path);
6784 			if (!active)
6785 				(void) printf(" failed (no initialize active)");
6786 			(void) printf("\n");
6787 		}
6788 		break;
6789 	case POOL_INITIALIZE_START:
6790 		if (ztest_opts.zo_verbose >= 4) {
6791 			(void) printf("Start initialize %s", path);
6792 			if (active && error == 0)
6793 				(void) printf(" failed (already active)");
6794 			else if (error != 0)
6795 				(void) printf(" failed (error %d)", error);
6796 			(void) printf("\n");
6797 		}
6798 		break;
6799 	case POOL_INITIALIZE_SUSPEND:
6800 		if (ztest_opts.zo_verbose >= 4) {
6801 			(void) printf("Suspend initialize %s", path);
6802 			if (!active)
6803 				(void) printf(" failed (no initialize active)");
6804 			(void) printf("\n");
6805 		}
6806 		break;
6807 	}
6808 	free(path);
6809 	mutex_exit(&ztest_vdev_lock);
6810 }
6811 
6812 void
6813 ztest_trim(ztest_ds_t *zd, uint64_t id)
6814 {
6815 	(void) zd, (void) id;
6816 	spa_t *spa = ztest_spa;
6817 	int error = 0;
6818 
6819 	mutex_enter(&ztest_vdev_lock);
6820 
6821 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6822 
6823 	/* Random leaf vdev */
6824 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6825 	if (rand_vd == NULL) {
6826 		spa_config_exit(spa, SCL_VDEV, FTAG);
6827 		mutex_exit(&ztest_vdev_lock);
6828 		return;
6829 	}
6830 
6831 	/*
6832 	 * The random vdev we've selected may change as soon as we
6833 	 * drop the spa_config_lock. We create local copies of things
6834 	 * we're interested in.
6835 	 */
6836 	uint64_t guid = rand_vd->vdev_guid;
6837 	char *path = strdup(rand_vd->vdev_path);
6838 	boolean_t active = rand_vd->vdev_trim_thread != NULL;
6839 
6840 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, (u_longlong_t)guid);
6841 	spa_config_exit(spa, SCL_VDEV, FTAG);
6842 
6843 	uint64_t cmd = ztest_random(POOL_TRIM_FUNCS);
6844 	uint64_t rate = 1 << ztest_random(30);
6845 	boolean_t partial = (ztest_random(5) > 0);
6846 	boolean_t secure = (ztest_random(5) > 0);
6847 
6848 	nvlist_t *vdev_guids = fnvlist_alloc();
6849 	nvlist_t *vdev_errlist = fnvlist_alloc();
6850 	fnvlist_add_uint64(vdev_guids, path, guid);
6851 	error = spa_vdev_trim(spa, vdev_guids, cmd, rate, partial,
6852 	    secure, vdev_errlist);
6853 	fnvlist_free(vdev_guids);
6854 	fnvlist_free(vdev_errlist);
6855 
6856 	switch (cmd) {
6857 	case POOL_TRIM_CANCEL:
6858 		if (ztest_opts.zo_verbose >= 4) {
6859 			(void) printf("Cancel TRIM %s", path);
6860 			if (!active)
6861 				(void) printf(" failed (no TRIM active)");
6862 			(void) printf("\n");
6863 		}
6864 		break;
6865 	case POOL_TRIM_START:
6866 		if (ztest_opts.zo_verbose >= 4) {
6867 			(void) printf("Start TRIM %s", path);
6868 			if (active && error == 0)
6869 				(void) printf(" failed (already active)");
6870 			else if (error != 0)
6871 				(void) printf(" failed (error %d)", error);
6872 			(void) printf("\n");
6873 		}
6874 		break;
6875 	case POOL_TRIM_SUSPEND:
6876 		if (ztest_opts.zo_verbose >= 4) {
6877 			(void) printf("Suspend TRIM %s", path);
6878 			if (!active)
6879 				(void) printf(" failed (no TRIM active)");
6880 			(void) printf("\n");
6881 		}
6882 		break;
6883 	}
6884 	free(path);
6885 	mutex_exit(&ztest_vdev_lock);
6886 }
6887 
6888 /*
6889  * Verify pool integrity by running zdb.
6890  */
6891 static void
6892 ztest_run_zdb(const char *pool)
6893 {
6894 	int status;
6895 	char *bin;
6896 	char *zdb;
6897 	char *zbuf;
6898 	const int len = MAXPATHLEN + MAXNAMELEN + 20;
6899 	FILE *fp;
6900 
6901 	bin = umem_alloc(len, UMEM_NOFAIL);
6902 	zdb = umem_alloc(len, UMEM_NOFAIL);
6903 	zbuf = umem_alloc(1024, UMEM_NOFAIL);
6904 
6905 	ztest_get_zdb_bin(bin, len);
6906 
6907 	char **set_gvars_args = ztest_global_vars_to_zdb_args();
6908 	char *set_gvars_args_joined = join_strings(set_gvars_args, " ");
6909 	free(set_gvars_args);
6910 
6911 	size_t would = snprintf(zdb, len,
6912 	    "%s -bcc%s%s -G -d -Y -e -y %s -p %s %s",
6913 	    bin,
6914 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
6915 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
6916 	    set_gvars_args_joined,
6917 	    ztest_opts.zo_dir,
6918 	    pool);
6919 	ASSERT3U(would, <, len);
6920 
6921 	free(set_gvars_args_joined);
6922 
6923 	if (ztest_opts.zo_verbose >= 5)
6924 		(void) printf("Executing %s\n", zdb);
6925 
6926 	fp = popen(zdb, "r");
6927 
6928 	while (fgets(zbuf, 1024, fp) != NULL)
6929 		if (ztest_opts.zo_verbose >= 3)
6930 			(void) printf("%s", zbuf);
6931 
6932 	status = pclose(fp);
6933 
6934 	if (status == 0)
6935 		goto out;
6936 
6937 	ztest_dump_core = 0;
6938 	if (WIFEXITED(status))
6939 		fatal(B_FALSE, "'%s' exit code %d", zdb, WEXITSTATUS(status));
6940 	else
6941 		fatal(B_FALSE, "'%s' died with signal %d",
6942 		    zdb, WTERMSIG(status));
6943 out:
6944 	umem_free(bin, len);
6945 	umem_free(zdb, len);
6946 	umem_free(zbuf, 1024);
6947 }
6948 
6949 static void
6950 ztest_walk_pool_directory(const char *header)
6951 {
6952 	spa_t *spa = NULL;
6953 
6954 	if (ztest_opts.zo_verbose >= 6)
6955 		(void) puts(header);
6956 
6957 	mutex_enter(&spa_namespace_lock);
6958 	while ((spa = spa_next(spa)) != NULL)
6959 		if (ztest_opts.zo_verbose >= 6)
6960 			(void) printf("\t%s\n", spa_name(spa));
6961 	mutex_exit(&spa_namespace_lock);
6962 }
6963 
6964 static void
6965 ztest_spa_import_export(char *oldname, char *newname)
6966 {
6967 	nvlist_t *config, *newconfig;
6968 	uint64_t pool_guid;
6969 	spa_t *spa;
6970 	int error;
6971 
6972 	if (ztest_opts.zo_verbose >= 4) {
6973 		(void) printf("import/export: old = %s, new = %s\n",
6974 		    oldname, newname);
6975 	}
6976 
6977 	/*
6978 	 * Clean up from previous runs.
6979 	 */
6980 	(void) spa_destroy(newname);
6981 
6982 	/*
6983 	 * Get the pool's configuration and guid.
6984 	 */
6985 	VERIFY0(spa_open(oldname, &spa, FTAG));
6986 
6987 	/*
6988 	 * Kick off a scrub to tickle scrub/export races.
6989 	 */
6990 	if (ztest_random(2) == 0)
6991 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
6992 
6993 	pool_guid = spa_guid(spa);
6994 	spa_close(spa, FTAG);
6995 
6996 	ztest_walk_pool_directory("pools before export");
6997 
6998 	/*
6999 	 * Export it.
7000 	 */
7001 	VERIFY0(spa_export(oldname, &config, B_FALSE, B_FALSE));
7002 
7003 	ztest_walk_pool_directory("pools after export");
7004 
7005 	/*
7006 	 * Try to import it.
7007 	 */
7008 	newconfig = spa_tryimport(config);
7009 	ASSERT3P(newconfig, !=, NULL);
7010 	fnvlist_free(newconfig);
7011 
7012 	/*
7013 	 * Import it under the new name.
7014 	 */
7015 	error = spa_import(newname, config, NULL, 0);
7016 	if (error != 0) {
7017 		dump_nvlist(config, 0);
7018 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
7019 		    oldname, newname, error);
7020 	}
7021 
7022 	ztest_walk_pool_directory("pools after import");
7023 
7024 	/*
7025 	 * Try to import it again -- should fail with EEXIST.
7026 	 */
7027 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
7028 
7029 	/*
7030 	 * Try to import it under a different name -- should fail with EEXIST.
7031 	 */
7032 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
7033 
7034 	/*
7035 	 * Verify that the pool is no longer visible under the old name.
7036 	 */
7037 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
7038 
7039 	/*
7040 	 * Verify that we can open and close the pool using the new name.
7041 	 */
7042 	VERIFY0(spa_open(newname, &spa, FTAG));
7043 	ASSERT3U(pool_guid, ==, spa_guid(spa));
7044 	spa_close(spa, FTAG);
7045 
7046 	fnvlist_free(config);
7047 }
7048 
7049 static void
7050 ztest_resume(spa_t *spa)
7051 {
7052 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
7053 		(void) printf("resuming from suspended state\n");
7054 	spa_vdev_state_enter(spa, SCL_NONE);
7055 	vdev_clear(spa, NULL);
7056 	(void) spa_vdev_state_exit(spa, NULL, 0);
7057 	(void) zio_resume(spa);
7058 }
7059 
7060 static __attribute__((noreturn)) void
7061 ztest_resume_thread(void *arg)
7062 {
7063 	spa_t *spa = arg;
7064 
7065 	while (!ztest_exiting) {
7066 		if (spa_suspended(spa))
7067 			ztest_resume(spa);
7068 		(void) poll(NULL, 0, 100);
7069 
7070 		/*
7071 		 * Periodically change the zfs_compressed_arc_enabled setting.
7072 		 */
7073 		if (ztest_random(10) == 0)
7074 			zfs_compressed_arc_enabled = ztest_random(2);
7075 
7076 		/*
7077 		 * Periodically change the zfs_abd_scatter_enabled setting.
7078 		 */
7079 		if (ztest_random(10) == 0)
7080 			zfs_abd_scatter_enabled = ztest_random(2);
7081 	}
7082 
7083 	thread_exit();
7084 }
7085 
7086 static __attribute__((noreturn)) void
7087 ztest_deadman_thread(void *arg)
7088 {
7089 	ztest_shared_t *zs = arg;
7090 	spa_t *spa = ztest_spa;
7091 	hrtime_t delay, overdue, last_run = gethrtime();
7092 
7093 	delay = (zs->zs_thread_stop - zs->zs_thread_start) +
7094 	    MSEC2NSEC(zfs_deadman_synctime_ms);
7095 
7096 	while (!ztest_exiting) {
7097 		/*
7098 		 * Wait for the delay timer while checking occasionally
7099 		 * if we should stop.
7100 		 */
7101 		if (gethrtime() < last_run + delay) {
7102 			(void) poll(NULL, 0, 1000);
7103 			continue;
7104 		}
7105 
7106 		/*
7107 		 * If the pool is suspended then fail immediately. Otherwise,
7108 		 * check to see if the pool is making any progress. If
7109 		 * vdev_deadman() discovers that there hasn't been any recent
7110 		 * I/Os then it will end up aborting the tests.
7111 		 */
7112 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
7113 			fatal(B_FALSE,
7114 			    "aborting test after %lu seconds because "
7115 			    "pool has transitioned to a suspended state.",
7116 			    zfs_deadman_synctime_ms / 1000);
7117 		}
7118 		vdev_deadman(spa->spa_root_vdev, FTAG);
7119 
7120 		/*
7121 		 * If the process doesn't complete within a grace period of
7122 		 * zfs_deadman_synctime_ms over the expected finish time,
7123 		 * then it may be hung and is terminated.
7124 		 */
7125 		overdue = zs->zs_proc_stop + MSEC2NSEC(zfs_deadman_synctime_ms);
7126 		if (gethrtime() > overdue) {
7127 			fatal(B_FALSE,
7128 			    "aborting test after %llu seconds because "
7129 			    "the process is overdue for termination.",
7130 			    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7131 		}
7132 
7133 		(void) printf("ztest has been running for %lld seconds\n",
7134 		    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7135 
7136 		last_run = gethrtime();
7137 		delay = MSEC2NSEC(zfs_deadman_checktime_ms);
7138 	}
7139 
7140 	thread_exit();
7141 }
7142 
7143 static void
7144 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
7145 {
7146 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
7147 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
7148 	hrtime_t functime = gethrtime();
7149 	int i;
7150 
7151 	for (i = 0; i < zi->zi_iters; i++)
7152 		zi->zi_func(zd, id);
7153 
7154 	functime = gethrtime() - functime;
7155 
7156 	atomic_add_64(&zc->zc_count, 1);
7157 	atomic_add_64(&zc->zc_time, functime);
7158 
7159 	if (ztest_opts.zo_verbose >= 4)
7160 		(void) printf("%6.2f sec in %s\n",
7161 		    (double)functime / NANOSEC, zi->zi_funcname);
7162 }
7163 
7164 static __attribute__((noreturn)) void
7165 ztest_thread(void *arg)
7166 {
7167 	int rand;
7168 	uint64_t id = (uintptr_t)arg;
7169 	ztest_shared_t *zs = ztest_shared;
7170 	uint64_t call_next;
7171 	hrtime_t now;
7172 	ztest_info_t *zi;
7173 	ztest_shared_callstate_t *zc;
7174 
7175 	while ((now = gethrtime()) < zs->zs_thread_stop) {
7176 		/*
7177 		 * See if it's time to force a crash.
7178 		 */
7179 		if (now > zs->zs_thread_kill)
7180 			ztest_kill(zs);
7181 
7182 		/*
7183 		 * If we're getting ENOSPC with some regularity, stop.
7184 		 */
7185 		if (zs->zs_enospc_count > 10)
7186 			break;
7187 
7188 		/*
7189 		 * Pick a random function to execute.
7190 		 */
7191 		rand = ztest_random(ZTEST_FUNCS);
7192 		zi = &ztest_info[rand];
7193 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
7194 		call_next = zc->zc_next;
7195 
7196 		if (now >= call_next &&
7197 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
7198 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
7199 			ztest_execute(rand, zi, id);
7200 		}
7201 	}
7202 
7203 	thread_exit();
7204 }
7205 
7206 static void
7207 ztest_dataset_name(char *dsname, const char *pool, int d)
7208 {
7209 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
7210 }
7211 
7212 static void
7213 ztest_dataset_destroy(int d)
7214 {
7215 	char name[ZFS_MAX_DATASET_NAME_LEN];
7216 	int t;
7217 
7218 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7219 
7220 	if (ztest_opts.zo_verbose >= 3)
7221 		(void) printf("Destroying %s to free up space\n", name);
7222 
7223 	/*
7224 	 * Cleanup any non-standard clones and snapshots.  In general,
7225 	 * ztest thread t operates on dataset (t % zopt_datasets),
7226 	 * so there may be more than one thing to clean up.
7227 	 */
7228 	for (t = d; t < ztest_opts.zo_threads;
7229 	    t += ztest_opts.zo_datasets)
7230 		ztest_dsl_dataset_cleanup(name, t);
7231 
7232 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
7233 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
7234 }
7235 
7236 static void
7237 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
7238 {
7239 	uint64_t usedobjs, dirobjs, scratch;
7240 
7241 	/*
7242 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
7243 	 * Therefore, the number of objects in use should equal the
7244 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
7245 	 * If not, we have an object leak.
7246 	 *
7247 	 * Note that we can only check this in ztest_dataset_open(),
7248 	 * when the open-context and syncing-context values agree.
7249 	 * That's because zap_count() returns the open-context value,
7250 	 * while dmu_objset_space() returns the rootbp fill count.
7251 	 */
7252 	VERIFY0(zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
7253 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
7254 	ASSERT3U(dirobjs + 1, ==, usedobjs);
7255 }
7256 
7257 static int
7258 ztest_dataset_open(int d)
7259 {
7260 	ztest_ds_t *zd = &ztest_ds[d];
7261 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
7262 	objset_t *os;
7263 	zilog_t *zilog;
7264 	char name[ZFS_MAX_DATASET_NAME_LEN];
7265 	int error;
7266 
7267 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7268 
7269 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
7270 
7271 	error = ztest_dataset_create(name);
7272 	if (error == ENOSPC) {
7273 		(void) pthread_rwlock_unlock(&ztest_name_lock);
7274 		ztest_record_enospc(FTAG);
7275 		return (error);
7276 	}
7277 	ASSERT(error == 0 || error == EEXIST);
7278 
7279 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
7280 	    B_TRUE, zd, &os));
7281 	(void) pthread_rwlock_unlock(&ztest_name_lock);
7282 
7283 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
7284 
7285 	zilog = zd->zd_zilog;
7286 
7287 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
7288 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
7289 		fatal(B_FALSE, "missing log records: "
7290 		    "claimed %"PRIu64" < committed %"PRIu64"",
7291 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
7292 
7293 	ztest_dataset_dirobj_verify(zd);
7294 
7295 	zil_replay(os, zd, ztest_replay_vector);
7296 
7297 	ztest_dataset_dirobj_verify(zd);
7298 
7299 	if (ztest_opts.zo_verbose >= 6)
7300 		(void) printf("%s replay %"PRIu64" blocks, "
7301 		    "%"PRIu64" records, seq %"PRIu64"\n",
7302 		    zd->zd_name,
7303 		    zilog->zl_parse_blk_count,
7304 		    zilog->zl_parse_lr_count,
7305 		    zilog->zl_replaying_seq);
7306 
7307 	zilog = zil_open(os, ztest_get_data, NULL);
7308 
7309 	if (zilog->zl_replaying_seq != 0 &&
7310 	    zilog->zl_replaying_seq < committed_seq)
7311 		fatal(B_FALSE, "missing log records: "
7312 		    "replayed %"PRIu64" < committed %"PRIu64"",
7313 		    zilog->zl_replaying_seq, committed_seq);
7314 
7315 	return (0);
7316 }
7317 
7318 static void
7319 ztest_dataset_close(int d)
7320 {
7321 	ztest_ds_t *zd = &ztest_ds[d];
7322 
7323 	zil_close(zd->zd_zilog);
7324 	dmu_objset_disown(zd->zd_os, B_TRUE, zd);
7325 
7326 	ztest_zd_fini(zd);
7327 }
7328 
7329 static int
7330 ztest_replay_zil_cb(const char *name, void *arg)
7331 {
7332 	(void) arg;
7333 	objset_t *os;
7334 	ztest_ds_t *zdtmp;
7335 
7336 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_ANY, B_TRUE,
7337 	    B_TRUE, FTAG, &os));
7338 
7339 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
7340 
7341 	ztest_zd_init(zdtmp, NULL, os);
7342 	zil_replay(os, zdtmp, ztest_replay_vector);
7343 	ztest_zd_fini(zdtmp);
7344 
7345 	if (dmu_objset_zil(os)->zl_parse_lr_count != 0 &&
7346 	    ztest_opts.zo_verbose >= 6) {
7347 		zilog_t *zilog = dmu_objset_zil(os);
7348 
7349 		(void) printf("%s replay %"PRIu64" blocks, "
7350 		    "%"PRIu64" records, seq %"PRIu64"\n",
7351 		    name,
7352 		    zilog->zl_parse_blk_count,
7353 		    zilog->zl_parse_lr_count,
7354 		    zilog->zl_replaying_seq);
7355 	}
7356 
7357 	umem_free(zdtmp, sizeof (ztest_ds_t));
7358 
7359 	dmu_objset_disown(os, B_TRUE, FTAG);
7360 	return (0);
7361 }
7362 
7363 static void
7364 ztest_freeze(void)
7365 {
7366 	ztest_ds_t *zd = &ztest_ds[0];
7367 	spa_t *spa;
7368 	int numloops = 0;
7369 
7370 	if (ztest_opts.zo_verbose >= 3)
7371 		(void) printf("testing spa_freeze()...\n");
7372 
7373 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7374 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7375 	VERIFY0(ztest_dataset_open(0));
7376 	ztest_spa = spa;
7377 
7378 	/*
7379 	 * Force the first log block to be transactionally allocated.
7380 	 * We have to do this before we freeze the pool -- otherwise
7381 	 * the log chain won't be anchored.
7382 	 */
7383 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
7384 		ztest_dmu_object_alloc_free(zd, 0);
7385 		zil_commit(zd->zd_zilog, 0);
7386 	}
7387 
7388 	txg_wait_synced(spa_get_dsl(spa), 0);
7389 
7390 	/*
7391 	 * Freeze the pool.  This stops spa_sync() from doing anything,
7392 	 * so that the only way to record changes from now on is the ZIL.
7393 	 */
7394 	spa_freeze(spa);
7395 
7396 	/*
7397 	 * Because it is hard to predict how much space a write will actually
7398 	 * require beforehand, we leave ourselves some fudge space to write over
7399 	 * capacity.
7400 	 */
7401 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
7402 
7403 	/*
7404 	 * Run tests that generate log records but don't alter the pool config
7405 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
7406 	 * We do a txg_wait_synced() after each iteration to force the txg
7407 	 * to increase well beyond the last synced value in the uberblock.
7408 	 * The ZIL should be OK with that.
7409 	 *
7410 	 * Run a random number of times less than zo_maxloops and ensure we do
7411 	 * not run out of space on the pool.
7412 	 */
7413 	while (ztest_random(10) != 0 &&
7414 	    numloops++ < ztest_opts.zo_maxloops &&
7415 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
7416 		ztest_od_t od;
7417 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
7418 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
7419 		ztest_io(zd, od.od_object,
7420 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
7421 		txg_wait_synced(spa_get_dsl(spa), 0);
7422 	}
7423 
7424 	/*
7425 	 * Commit all of the changes we just generated.
7426 	 */
7427 	zil_commit(zd->zd_zilog, 0);
7428 	txg_wait_synced(spa_get_dsl(spa), 0);
7429 
7430 	/*
7431 	 * Close our dataset and close the pool.
7432 	 */
7433 	ztest_dataset_close(0);
7434 	spa_close(spa, FTAG);
7435 	kernel_fini();
7436 
7437 	/*
7438 	 * Open and close the pool and dataset to induce log replay.
7439 	 */
7440 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7441 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7442 	ASSERT3U(spa_freeze_txg(spa), ==, UINT64_MAX);
7443 	VERIFY0(ztest_dataset_open(0));
7444 	ztest_spa = spa;
7445 	txg_wait_synced(spa_get_dsl(spa), 0);
7446 	ztest_dataset_close(0);
7447 	ztest_reguid(NULL, 0);
7448 
7449 	spa_close(spa, FTAG);
7450 	kernel_fini();
7451 }
7452 
7453 static void
7454 ztest_import_impl(void)
7455 {
7456 	importargs_t args = { 0 };
7457 	nvlist_t *cfg = NULL;
7458 	int nsearch = 1;
7459 	char *searchdirs[nsearch];
7460 	int flags = ZFS_IMPORT_MISSING_LOG;
7461 
7462 	searchdirs[0] = ztest_opts.zo_dir;
7463 	args.paths = nsearch;
7464 	args.path = searchdirs;
7465 	args.can_be_active = B_FALSE;
7466 
7467 	VERIFY0(zpool_find_config(NULL, ztest_opts.zo_pool, &cfg, &args,
7468 	    &libzpool_config_ops));
7469 	VERIFY0(spa_import(ztest_opts.zo_pool, cfg, NULL, flags));
7470 	fnvlist_free(cfg);
7471 }
7472 
7473 /*
7474  * Import a storage pool with the given name.
7475  */
7476 static void
7477 ztest_import(ztest_shared_t *zs)
7478 {
7479 	spa_t *spa;
7480 
7481 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7482 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7483 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7484 
7485 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7486 
7487 	ztest_import_impl();
7488 
7489 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7490 	zs->zs_metaslab_sz =
7491 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7492 	spa_close(spa, FTAG);
7493 
7494 	kernel_fini();
7495 
7496 	if (!ztest_opts.zo_mmp_test) {
7497 		ztest_run_zdb(ztest_opts.zo_pool);
7498 		ztest_freeze();
7499 		ztest_run_zdb(ztest_opts.zo_pool);
7500 	}
7501 
7502 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7503 	mutex_destroy(&ztest_vdev_lock);
7504 	mutex_destroy(&ztest_checkpoint_lock);
7505 }
7506 
7507 /*
7508  * Kick off threads to run tests on all datasets in parallel.
7509  */
7510 static void
7511 ztest_run(ztest_shared_t *zs)
7512 {
7513 	spa_t *spa;
7514 	objset_t *os;
7515 	kthread_t *resume_thread, *deadman_thread;
7516 	kthread_t **run_threads;
7517 	uint64_t object;
7518 	int error;
7519 	int t, d;
7520 
7521 	ztest_exiting = B_FALSE;
7522 
7523 	/*
7524 	 * Initialize parent/child shared state.
7525 	 */
7526 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7527 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7528 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7529 
7530 	zs->zs_thread_start = gethrtime();
7531 	zs->zs_thread_stop =
7532 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
7533 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
7534 	zs->zs_thread_kill = zs->zs_thread_stop;
7535 	if (ztest_random(100) < ztest_opts.zo_killrate) {
7536 		zs->zs_thread_kill -=
7537 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
7538 	}
7539 
7540 	mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
7541 
7542 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
7543 	    offsetof(ztest_cb_data_t, zcd_node));
7544 
7545 	/*
7546 	 * Open our pool.  It may need to be imported first depending on
7547 	 * what tests were running when the previous pass was terminated.
7548 	 */
7549 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7550 	error = spa_open(ztest_opts.zo_pool, &spa, FTAG);
7551 	if (error) {
7552 		VERIFY3S(error, ==, ENOENT);
7553 		ztest_import_impl();
7554 		VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7555 		zs->zs_metaslab_sz =
7556 		    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7557 	}
7558 
7559 	metaslab_preload_limit = ztest_random(20) + 1;
7560 	ztest_spa = spa;
7561 
7562 	VERIFY0(vdev_raidz_impl_set("cycle"));
7563 
7564 	dmu_objset_stats_t dds;
7565 	VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
7566 	    DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
7567 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
7568 	dmu_objset_fast_stat(os, &dds);
7569 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
7570 	zs->zs_guid = dds.dds_guid;
7571 	dmu_objset_disown(os, B_TRUE, FTAG);
7572 
7573 	/*
7574 	 * Create a thread to periodically resume suspended I/O.
7575 	 */
7576 	resume_thread = thread_create(NULL, 0, ztest_resume_thread,
7577 	    spa, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
7578 
7579 	/*
7580 	 * Create a deadman thread and set to panic if we hang.
7581 	 */
7582 	deadman_thread = thread_create(NULL, 0, ztest_deadman_thread,
7583 	    zs, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
7584 
7585 	spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
7586 
7587 	/*
7588 	 * Verify that we can safely inquire about any object,
7589 	 * whether it's allocated or not.  To make it interesting,
7590 	 * we probe a 5-wide window around each power of two.
7591 	 * This hits all edge cases, including zero and the max.
7592 	 */
7593 	for (t = 0; t < 64; t++) {
7594 		for (d = -5; d <= 5; d++) {
7595 			error = dmu_object_info(spa->spa_meta_objset,
7596 			    (1ULL << t) + d, NULL);
7597 			ASSERT(error == 0 || error == ENOENT ||
7598 			    error == EINVAL);
7599 		}
7600 	}
7601 
7602 	/*
7603 	 * If we got any ENOSPC errors on the previous run, destroy something.
7604 	 */
7605 	if (zs->zs_enospc_count != 0) {
7606 		int d = ztest_random(ztest_opts.zo_datasets);
7607 		ztest_dataset_destroy(d);
7608 	}
7609 	zs->zs_enospc_count = 0;
7610 
7611 	/*
7612 	 * If we were in the middle of ztest_device_removal() and were killed
7613 	 * we need to ensure the removal and scrub complete before running
7614 	 * any tests that check ztest_device_removal_active. The removal will
7615 	 * be restarted automatically when the spa is opened, but we need to
7616 	 * initiate the scrub manually if it is not already in progress. Note
7617 	 * that we always run the scrub whenever an indirect vdev exists
7618 	 * because we have no way of knowing for sure if ztest_device_removal()
7619 	 * fully completed its scrub before the pool was reimported.
7620 	 */
7621 	if (spa->spa_removing_phys.sr_state == DSS_SCANNING ||
7622 	    spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
7623 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
7624 			txg_wait_synced(spa_get_dsl(spa), 0);
7625 
7626 		error = ztest_scrub_impl(spa);
7627 		if (error == EBUSY)
7628 			error = 0;
7629 		ASSERT0(error);
7630 	}
7631 
7632 	run_threads = umem_zalloc(ztest_opts.zo_threads * sizeof (kthread_t *),
7633 	    UMEM_NOFAIL);
7634 
7635 	if (ztest_opts.zo_verbose >= 4)
7636 		(void) printf("starting main threads...\n");
7637 
7638 	/*
7639 	 * Replay all logs of all datasets in the pool. This is primarily for
7640 	 * temporary datasets which wouldn't otherwise get replayed, which
7641 	 * can trigger failures when attempting to offline a SLOG in
7642 	 * ztest_fault_inject().
7643 	 */
7644 	(void) dmu_objset_find(ztest_opts.zo_pool, ztest_replay_zil_cb,
7645 	    NULL, DS_FIND_CHILDREN);
7646 
7647 	/*
7648 	 * Kick off all the tests that run in parallel.
7649 	 */
7650 	for (t = 0; t < ztest_opts.zo_threads; t++) {
7651 		if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
7652 			umem_free(run_threads, ztest_opts.zo_threads *
7653 			    sizeof (kthread_t *));
7654 			return;
7655 		}
7656 
7657 		run_threads[t] = thread_create(NULL, 0, ztest_thread,
7658 		    (void *)(uintptr_t)t, 0, NULL, TS_RUN | TS_JOINABLE,
7659 		    defclsyspri);
7660 	}
7661 
7662 	/*
7663 	 * Wait for all of the tests to complete.
7664 	 */
7665 	for (t = 0; t < ztest_opts.zo_threads; t++)
7666 		VERIFY0(thread_join(run_threads[t]));
7667 
7668 	/*
7669 	 * Close all datasets. This must be done after all the threads
7670 	 * are joined so we can be sure none of the datasets are in-use
7671 	 * by any of the threads.
7672 	 */
7673 	for (t = 0; t < ztest_opts.zo_threads; t++) {
7674 		if (t < ztest_opts.zo_datasets)
7675 			ztest_dataset_close(t);
7676 	}
7677 
7678 	txg_wait_synced(spa_get_dsl(spa), 0);
7679 
7680 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7681 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
7682 
7683 	umem_free(run_threads, ztest_opts.zo_threads * sizeof (kthread_t *));
7684 
7685 	/* Kill the resume and deadman threads */
7686 	ztest_exiting = B_TRUE;
7687 	VERIFY0(thread_join(resume_thread));
7688 	VERIFY0(thread_join(deadman_thread));
7689 	ztest_resume(spa);
7690 
7691 	/*
7692 	 * Right before closing the pool, kick off a bunch of async I/O;
7693 	 * spa_close() should wait for it to complete.
7694 	 */
7695 	for (object = 1; object < 50; object++) {
7696 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
7697 		    ZIO_PRIORITY_SYNC_READ);
7698 	}
7699 
7700 	/* Verify that at least one commit cb was called in a timely fashion */
7701 	if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
7702 		VERIFY0(zc_min_txg_delay);
7703 
7704 	spa_close(spa, FTAG);
7705 
7706 	/*
7707 	 * Verify that we can loop over all pools.
7708 	 */
7709 	mutex_enter(&spa_namespace_lock);
7710 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
7711 		if (ztest_opts.zo_verbose > 3)
7712 			(void) printf("spa_next: found %s\n", spa_name(spa));
7713 	mutex_exit(&spa_namespace_lock);
7714 
7715 	/*
7716 	 * Verify that we can export the pool and reimport it under a
7717 	 * different name.
7718 	 */
7719 	if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
7720 		char name[ZFS_MAX_DATASET_NAME_LEN];
7721 		(void) snprintf(name, sizeof (name), "%s_import",
7722 		    ztest_opts.zo_pool);
7723 		ztest_spa_import_export(ztest_opts.zo_pool, name);
7724 		ztest_spa_import_export(name, ztest_opts.zo_pool);
7725 	}
7726 
7727 	kernel_fini();
7728 
7729 	list_destroy(&zcl.zcl_callbacks);
7730 	mutex_destroy(&zcl.zcl_callbacks_lock);
7731 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7732 	mutex_destroy(&ztest_vdev_lock);
7733 	mutex_destroy(&ztest_checkpoint_lock);
7734 }
7735 
7736 static void
7737 print_time(hrtime_t t, char *timebuf)
7738 {
7739 	hrtime_t s = t / NANOSEC;
7740 	hrtime_t m = s / 60;
7741 	hrtime_t h = m / 60;
7742 	hrtime_t d = h / 24;
7743 
7744 	s -= m * 60;
7745 	m -= h * 60;
7746 	h -= d * 24;
7747 
7748 	timebuf[0] = '\0';
7749 
7750 	if (d)
7751 		(void) sprintf(timebuf,
7752 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
7753 	else if (h)
7754 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
7755 	else if (m)
7756 		(void) sprintf(timebuf, "%llum%02llus", m, s);
7757 	else
7758 		(void) sprintf(timebuf, "%llus", s);
7759 }
7760 
7761 static nvlist_t *
7762 make_random_props(void)
7763 {
7764 	nvlist_t *props;
7765 
7766 	props = fnvlist_alloc();
7767 
7768 	if (ztest_random(2) == 0)
7769 		return (props);
7770 
7771 	fnvlist_add_uint64(props,
7772 	    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 1);
7773 
7774 	return (props);
7775 }
7776 
7777 /*
7778  * Create a storage pool with the given name and initial vdev size.
7779  * Then test spa_freeze() functionality.
7780  */
7781 static void
7782 ztest_init(ztest_shared_t *zs)
7783 {
7784 	spa_t *spa;
7785 	nvlist_t *nvroot, *props;
7786 	int i;
7787 
7788 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7789 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7790 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7791 
7792 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7793 
7794 	/*
7795 	 * Create the storage pool.
7796 	 */
7797 	(void) spa_destroy(ztest_opts.zo_pool);
7798 	ztest_shared->zs_vdev_next_leaf = 0;
7799 	zs->zs_splits = 0;
7800 	zs->zs_mirrors = ztest_opts.zo_mirrors;
7801 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
7802 	    NULL, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
7803 	props = make_random_props();
7804 
7805 	/*
7806 	 * We don't expect the pool to suspend unless maxfaults == 0,
7807 	 * in which case ztest_fault_inject() temporarily takes away
7808 	 * the only valid replica.
7809 	 */
7810 	fnvlist_add_uint64(props,
7811 	    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
7812 	    MAXFAULTS(zs) ? ZIO_FAILURE_MODE_PANIC : ZIO_FAILURE_MODE_WAIT);
7813 
7814 	for (i = 0; i < SPA_FEATURES; i++) {
7815 		char *buf;
7816 
7817 		if (!spa_feature_table[i].fi_zfs_mod_supported)
7818 			continue;
7819 
7820 		/*
7821 		 * 75% chance of using the log space map feature. We want ztest
7822 		 * to exercise both the code paths that use the log space map
7823 		 * feature and the ones that don't.
7824 		 */
7825 		if (i == SPA_FEATURE_LOG_SPACEMAP && ztest_random(4) == 0)
7826 			continue;
7827 
7828 		VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
7829 		    spa_feature_table[i].fi_uname));
7830 		fnvlist_add_uint64(props, buf, 0);
7831 		free(buf);
7832 	}
7833 
7834 	VERIFY0(spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
7835 	fnvlist_free(nvroot);
7836 	fnvlist_free(props);
7837 
7838 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7839 	zs->zs_metaslab_sz =
7840 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7841 	spa_close(spa, FTAG);
7842 
7843 	kernel_fini();
7844 
7845 	if (!ztest_opts.zo_mmp_test) {
7846 		ztest_run_zdb(ztest_opts.zo_pool);
7847 		ztest_freeze();
7848 		ztest_run_zdb(ztest_opts.zo_pool);
7849 	}
7850 
7851 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7852 	mutex_destroy(&ztest_vdev_lock);
7853 	mutex_destroy(&ztest_checkpoint_lock);
7854 }
7855 
7856 static void
7857 setup_data_fd(void)
7858 {
7859 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
7860 
7861 	ztest_fd_data = mkstemp(ztest_name_data);
7862 	ASSERT3S(ztest_fd_data, >=, 0);
7863 	(void) unlink(ztest_name_data);
7864 }
7865 
7866 static int
7867 shared_data_size(ztest_shared_hdr_t *hdr)
7868 {
7869 	int size;
7870 
7871 	size = hdr->zh_hdr_size;
7872 	size += hdr->zh_opts_size;
7873 	size += hdr->zh_size;
7874 	size += hdr->zh_stats_size * hdr->zh_stats_count;
7875 	size += hdr->zh_ds_size * hdr->zh_ds_count;
7876 
7877 	return (size);
7878 }
7879 
7880 static void
7881 setup_hdr(void)
7882 {
7883 	int size;
7884 	ztest_shared_hdr_t *hdr;
7885 
7886 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7887 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7888 	ASSERT3P(hdr, !=, MAP_FAILED);
7889 
7890 	VERIFY0(ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
7891 
7892 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
7893 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
7894 	hdr->zh_size = sizeof (ztest_shared_t);
7895 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
7896 	hdr->zh_stats_count = ZTEST_FUNCS;
7897 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
7898 	hdr->zh_ds_count = ztest_opts.zo_datasets;
7899 
7900 	size = shared_data_size(hdr);
7901 	VERIFY0(ftruncate(ztest_fd_data, size));
7902 
7903 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7904 }
7905 
7906 static void
7907 setup_data(void)
7908 {
7909 	int size, offset;
7910 	ztest_shared_hdr_t *hdr;
7911 	uint8_t *buf;
7912 
7913 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7914 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
7915 	ASSERT3P(hdr, !=, MAP_FAILED);
7916 
7917 	size = shared_data_size(hdr);
7918 
7919 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7920 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
7921 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7922 	ASSERT3P(hdr, !=, MAP_FAILED);
7923 	buf = (uint8_t *)hdr;
7924 
7925 	offset = hdr->zh_hdr_size;
7926 	ztest_shared_opts = (void *)&buf[offset];
7927 	offset += hdr->zh_opts_size;
7928 	ztest_shared = (void *)&buf[offset];
7929 	offset += hdr->zh_size;
7930 	ztest_shared_callstate = (void *)&buf[offset];
7931 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
7932 	ztest_shared_ds = (void *)&buf[offset];
7933 }
7934 
7935 static boolean_t
7936 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
7937 {
7938 	pid_t pid;
7939 	int status;
7940 	char *cmdbuf = NULL;
7941 
7942 	pid = fork();
7943 
7944 	if (cmd == NULL) {
7945 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
7946 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
7947 		cmd = cmdbuf;
7948 	}
7949 
7950 	if (pid == -1)
7951 		fatal(B_TRUE, "fork failed");
7952 
7953 	if (pid == 0) {	/* child */
7954 		char fd_data_str[12];
7955 
7956 		VERIFY3S(11, >=,
7957 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
7958 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
7959 
7960 		if (libpath != NULL) {
7961 			const char *curlp = getenv("LD_LIBRARY_PATH");
7962 			if (curlp == NULL)
7963 				VERIFY0(setenv("LD_LIBRARY_PATH", libpath, 1));
7964 			else {
7965 				char *newlp = NULL;
7966 				VERIFY3S(-1, !=,
7967 				    asprintf(&newlp, "%s:%s", libpath, curlp));
7968 				VERIFY0(setenv("LD_LIBRARY_PATH", newlp, 1));
7969 			}
7970 		}
7971 		(void) execl(cmd, cmd, (char *)NULL);
7972 		ztest_dump_core = B_FALSE;
7973 		fatal(B_TRUE, "exec failed: %s", cmd);
7974 	}
7975 
7976 	if (cmdbuf != NULL) {
7977 		umem_free(cmdbuf, MAXPATHLEN);
7978 		cmd = NULL;
7979 	}
7980 
7981 	while (waitpid(pid, &status, 0) != pid)
7982 		continue;
7983 	if (statusp != NULL)
7984 		*statusp = status;
7985 
7986 	if (WIFEXITED(status)) {
7987 		if (WEXITSTATUS(status) != 0) {
7988 			(void) fprintf(stderr, "child exited with code %d\n",
7989 			    WEXITSTATUS(status));
7990 			exit(2);
7991 		}
7992 		return (B_FALSE);
7993 	} else if (WIFSIGNALED(status)) {
7994 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
7995 			(void) fprintf(stderr, "child died with signal %d\n",
7996 			    WTERMSIG(status));
7997 			exit(3);
7998 		}
7999 		return (B_TRUE);
8000 	} else {
8001 		(void) fprintf(stderr, "something strange happened to child\n");
8002 		exit(4);
8003 	}
8004 }
8005 
8006 static void
8007 ztest_run_init(void)
8008 {
8009 	int i;
8010 
8011 	ztest_shared_t *zs = ztest_shared;
8012 
8013 	/*
8014 	 * Blow away any existing copy of zpool.cache
8015 	 */
8016 	(void) remove(spa_config_path);
8017 
8018 	if (ztest_opts.zo_init == 0) {
8019 		if (ztest_opts.zo_verbose >= 1)
8020 			(void) printf("Importing pool %s\n",
8021 			    ztest_opts.zo_pool);
8022 		ztest_import(zs);
8023 		return;
8024 	}
8025 
8026 	/*
8027 	 * Create and initialize our storage pool.
8028 	 */
8029 	for (i = 1; i <= ztest_opts.zo_init; i++) {
8030 		memset(zs, 0, sizeof (*zs));
8031 		if (ztest_opts.zo_verbose >= 3 &&
8032 		    ztest_opts.zo_init != 1) {
8033 			(void) printf("ztest_init(), pass %d\n", i);
8034 		}
8035 		ztest_init(zs);
8036 	}
8037 }
8038 
8039 int
8040 main(int argc, char **argv)
8041 {
8042 	int kills = 0;
8043 	int iters = 0;
8044 	int older = 0;
8045 	int newer = 0;
8046 	ztest_shared_t *zs;
8047 	ztest_info_t *zi;
8048 	ztest_shared_callstate_t *zc;
8049 	char timebuf[100];
8050 	char numbuf[NN_NUMBUF_SZ];
8051 	char *cmd;
8052 	boolean_t hasalt;
8053 	int f, err;
8054 	char *fd_data_str = getenv("ZTEST_FD_DATA");
8055 	struct sigaction action;
8056 
8057 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
8058 
8059 	dprintf_setup(&argc, argv);
8060 	zfs_deadman_synctime_ms = 300000;
8061 	zfs_deadman_checktime_ms = 30000;
8062 	/*
8063 	 * As two-word space map entries may not come up often (especially
8064 	 * if pool and vdev sizes are small) we want to force at least some
8065 	 * of them so the feature get tested.
8066 	 */
8067 	zfs_force_some_double_word_sm_entries = B_TRUE;
8068 
8069 	/*
8070 	 * Verify that even extensively damaged split blocks with many
8071 	 * segments can be reconstructed in a reasonable amount of time
8072 	 * when reconstruction is known to be possible.
8073 	 *
8074 	 * Note: the lower this value is, the more damage we inflict, and
8075 	 * the more time ztest spends in recovering that damage. We chose
8076 	 * to induce damage 1/100th of the time so recovery is tested but
8077 	 * not so frequently that ztest doesn't get to test other code paths.
8078 	 */
8079 	zfs_reconstruct_indirect_damage_fraction = 100;
8080 
8081 	action.sa_handler = sig_handler;
8082 	sigemptyset(&action.sa_mask);
8083 	action.sa_flags = 0;
8084 
8085 	if (sigaction(SIGSEGV, &action, NULL) < 0) {
8086 		(void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
8087 		    strerror(errno));
8088 		exit(EXIT_FAILURE);
8089 	}
8090 
8091 	if (sigaction(SIGABRT, &action, NULL) < 0) {
8092 		(void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
8093 		    strerror(errno));
8094 		exit(EXIT_FAILURE);
8095 	}
8096 
8097 	/*
8098 	 * Force random_get_bytes() to use /dev/urandom in order to prevent
8099 	 * ztest from needlessly depleting the system entropy pool.
8100 	 */
8101 	random_path = "/dev/urandom";
8102 	ztest_fd_rand = open(random_path, O_RDONLY | O_CLOEXEC);
8103 	ASSERT3S(ztest_fd_rand, >=, 0);
8104 
8105 	if (!fd_data_str) {
8106 		process_options(argc, argv);
8107 
8108 		setup_data_fd();
8109 		setup_hdr();
8110 		setup_data();
8111 		memcpy(ztest_shared_opts, &ztest_opts,
8112 		    sizeof (*ztest_shared_opts));
8113 	} else {
8114 		ztest_fd_data = atoi(fd_data_str);
8115 		setup_data();
8116 		memcpy(&ztest_opts, ztest_shared_opts, sizeof (ztest_opts));
8117 	}
8118 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
8119 
8120 	err = ztest_set_global_vars();
8121 	if (err != 0 && !fd_data_str) {
8122 		/* error message done by ztest_set_global_vars */
8123 		exit(EXIT_FAILURE);
8124 	} else {
8125 		/* children should not be spawned if setting gvars fails */
8126 		VERIFY3S(err, ==, 0);
8127 	}
8128 
8129 	/* Override location of zpool.cache */
8130 	VERIFY3S(asprintf((char **)&spa_config_path, "%s/zpool.cache",
8131 	    ztest_opts.zo_dir), !=, -1);
8132 
8133 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
8134 	    UMEM_NOFAIL);
8135 	zs = ztest_shared;
8136 
8137 	if (fd_data_str) {
8138 		metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
8139 		metaslab_df_alloc_threshold =
8140 		    zs->zs_metaslab_df_alloc_threshold;
8141 
8142 		if (zs->zs_do_init)
8143 			ztest_run_init();
8144 		else
8145 			ztest_run(zs);
8146 		exit(0);
8147 	}
8148 
8149 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
8150 
8151 	if (ztest_opts.zo_verbose >= 1) {
8152 		(void) printf("%"PRIu64" vdevs, %d datasets, %d threads,"
8153 		    "%d %s disks, %"PRIu64" seconds...\n\n",
8154 		    ztest_opts.zo_vdevs,
8155 		    ztest_opts.zo_datasets,
8156 		    ztest_opts.zo_threads,
8157 		    ztest_opts.zo_raid_children,
8158 		    ztest_opts.zo_raid_type,
8159 		    ztest_opts.zo_time);
8160 	}
8161 
8162 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
8163 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
8164 
8165 	zs->zs_do_init = B_TRUE;
8166 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
8167 		if (ztest_opts.zo_verbose >= 1) {
8168 			(void) printf("Executing older ztest for "
8169 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
8170 		}
8171 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
8172 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
8173 	} else {
8174 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
8175 	}
8176 	zs->zs_do_init = B_FALSE;
8177 
8178 	zs->zs_proc_start = gethrtime();
8179 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
8180 
8181 	for (f = 0; f < ZTEST_FUNCS; f++) {
8182 		zi = &ztest_info[f];
8183 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
8184 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
8185 			zc->zc_next = UINT64_MAX;
8186 		else
8187 			zc->zc_next = zs->zs_proc_start +
8188 			    ztest_random(2 * zi->zi_interval[0] + 1);
8189 	}
8190 
8191 	/*
8192 	 * Run the tests in a loop.  These tests include fault injection
8193 	 * to verify that self-healing data works, and forced crashes
8194 	 * to verify that we never lose on-disk consistency.
8195 	 */
8196 	while (gethrtime() < zs->zs_proc_stop) {
8197 		int status;
8198 		boolean_t killed;
8199 
8200 		/*
8201 		 * Initialize the workload counters for each function.
8202 		 */
8203 		for (f = 0; f < ZTEST_FUNCS; f++) {
8204 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
8205 			zc->zc_count = 0;
8206 			zc->zc_time = 0;
8207 		}
8208 
8209 		/* Set the allocation switch size */
8210 		zs->zs_metaslab_df_alloc_threshold =
8211 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
8212 
8213 		if (!hasalt || ztest_random(2) == 0) {
8214 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8215 				(void) printf("Executing newer ztest: %s\n",
8216 				    cmd);
8217 			}
8218 			newer++;
8219 			killed = exec_child(cmd, NULL, B_TRUE, &status);
8220 		} else {
8221 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8222 				(void) printf("Executing older ztest: %s\n",
8223 				    ztest_opts.zo_alt_ztest);
8224 			}
8225 			older++;
8226 			killed = exec_child(ztest_opts.zo_alt_ztest,
8227 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
8228 		}
8229 
8230 		if (killed)
8231 			kills++;
8232 		iters++;
8233 
8234 		if (ztest_opts.zo_verbose >= 1) {
8235 			hrtime_t now = gethrtime();
8236 
8237 			now = MIN(now, zs->zs_proc_stop);
8238 			print_time(zs->zs_proc_stop - now, timebuf);
8239 			nicenum(zs->zs_space, numbuf, sizeof (numbuf));
8240 
8241 			(void) printf("Pass %3d, %8s, %3"PRIu64" ENOSPC, "
8242 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
8243 			    iters,
8244 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
8245 			    zs->zs_enospc_count,
8246 			    100.0 * zs->zs_alloc / zs->zs_space,
8247 			    numbuf,
8248 			    100.0 * (now - zs->zs_proc_start) /
8249 			    (ztest_opts.zo_time * NANOSEC), timebuf);
8250 		}
8251 
8252 		if (ztest_opts.zo_verbose >= 2) {
8253 			(void) printf("\nWorkload summary:\n\n");
8254 			(void) printf("%7s %9s   %s\n",
8255 			    "Calls", "Time", "Function");
8256 			(void) printf("%7s %9s   %s\n",
8257 			    "-----", "----", "--------");
8258 			for (f = 0; f < ZTEST_FUNCS; f++) {
8259 				zi = &ztest_info[f];
8260 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
8261 				print_time(zc->zc_time, timebuf);
8262 				(void) printf("%7"PRIu64" %9s   %s\n",
8263 				    zc->zc_count, timebuf,
8264 				    zi->zi_funcname);
8265 			}
8266 			(void) printf("\n");
8267 		}
8268 
8269 		if (!ztest_opts.zo_mmp_test)
8270 			ztest_run_zdb(ztest_opts.zo_pool);
8271 	}
8272 
8273 	if (ztest_opts.zo_verbose >= 1) {
8274 		if (hasalt) {
8275 			(void) printf("%d runs of older ztest: %s\n", older,
8276 			    ztest_opts.zo_alt_ztest);
8277 			(void) printf("%d runs of newer ztest: %s\n", newer,
8278 			    cmd);
8279 		}
8280 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
8281 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
8282 	}
8283 
8284 	umem_free(cmd, MAXNAMELEN);
8285 
8286 	return (0);
8287 }
8288