xref: /freebsd/sys/contrib/openzfs/cmd/ztest.c (revision e47161e5f1f01ef300c6e7efdb9c92e3a6c497ff)
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, 2024 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  * Copyright (c) 2023, Klara, Inc.
30  */
31 
32 /*
33  * The objective of this program is to provide a DMU/ZAP/SPA stress test
34  * that runs entirely in userland, is easy to use, and easy to extend.
35  *
36  * The overall design of the ztest program is as follows:
37  *
38  * (1) For each major functional area (e.g. adding vdevs to a pool,
39  *     creating and destroying datasets, reading and writing objects, etc)
40  *     we have a simple routine to test that functionality.  These
41  *     individual routines do not have to do anything "stressful".
42  *
43  * (2) We turn these simple functionality tests into a stress test by
44  *     running them all in parallel, with as many threads as desired,
45  *     and spread across as many datasets, objects, and vdevs as desired.
46  *
47  * (3) While all this is happening, we inject faults into the pool to
48  *     verify that self-healing data really works.
49  *
50  * (4) Every time we open a dataset, we change its checksum and compression
51  *     functions.  Thus even individual objects vary from block to block
52  *     in which checksum they use and whether they're compressed.
53  *
54  * (5) To verify that we never lose on-disk consistency after a crash,
55  *     we run the entire test in a child of the main process.
56  *     At random times, the child self-immolates with a SIGKILL.
57  *     This is the software equivalent of pulling the power cord.
58  *     The parent then runs the test again, using the existing
59  *     storage pool, as many times as desired. If backwards compatibility
60  *     testing is enabled ztest will sometimes run the "older" version
61  *     of ztest after a SIGKILL.
62  *
63  * (6) To verify that we don't have future leaks or temporal incursions,
64  *     many of the functional tests record the transaction group number
65  *     as part of their data.  When reading old data, they verify that
66  *     the transaction group number is less than the current, open txg.
67  *     If you add a new test, please do this if applicable.
68  *
69  * (7) Threads are created with a reduced stack size, for sanity checking.
70  *     Therefore, it's important not to allocate huge buffers on the stack.
71  *
72  * When run with no arguments, ztest runs for about five minutes and
73  * produces no output if successful.  To get a little bit of information,
74  * specify -V.  To get more information, specify -VV, and so on.
75  *
76  * To turn this into an overnight stress test, use -T to specify run time.
77  *
78  * You can ask more vdevs [-v], datasets [-d], or threads [-t]
79  * to increase the pool capacity, fanout, and overall stress level.
80  *
81  * Use the -k option to set the desired frequency of kills.
82  *
83  * When ztest invokes itself it passes all relevant information through a
84  * temporary file which is mmap-ed in the child process. This allows shared
85  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
86  * stored at offset 0 of this file and contains information on the size and
87  * number of shared structures in the file. The information stored in this file
88  * must remain backwards compatible with older versions of ztest so that
89  * ztest can invoke them during backwards compatibility testing (-B).
90  */
91 
92 #include <sys/zfs_context.h>
93 #include <sys/spa.h>
94 #include <sys/dmu.h>
95 #include <sys/txg.h>
96 #include <sys/dbuf.h>
97 #include <sys/zap.h>
98 #include <sys/dmu_objset.h>
99 #include <sys/poll.h>
100 #include <sys/stat.h>
101 #include <sys/time.h>
102 #include <sys/wait.h>
103 #include <sys/mman.h>
104 #include <sys/resource.h>
105 #include <sys/zio.h>
106 #include <sys/zil.h>
107 #include <sys/zil_impl.h>
108 #include <sys/vdev_draid.h>
109 #include <sys/vdev_impl.h>
110 #include <sys/vdev_file.h>
111 #include <sys/vdev_initialize.h>
112 #include <sys/vdev_raidz.h>
113 #include <sys/vdev_trim.h>
114 #include <sys/spa_impl.h>
115 #include <sys/metaslab_impl.h>
116 #include <sys/dsl_prop.h>
117 #include <sys/dsl_dataset.h>
118 #include <sys/dsl_destroy.h>
119 #include <sys/dsl_scan.h>
120 #include <sys/zio_checksum.h>
121 #include <sys/zfs_refcount.h>
122 #include <sys/zfeature.h>
123 #include <sys/dsl_userhold.h>
124 #include <sys/abd.h>
125 #include <sys/blake3.h>
126 #include <stdio.h>
127 #include <stdlib.h>
128 #include <unistd.h>
129 #include <getopt.h>
130 #include <signal.h>
131 #include <umem.h>
132 #include <ctype.h>
133 #include <math.h>
134 #include <sys/fs/zfs.h>
135 #include <zfs_fletcher.h>
136 #include <libnvpair.h>
137 #include <libzutil.h>
138 #include <sys/crypto/icp.h>
139 #include <sys/zfs_impl.h>
140 #include <sys/backtrace.h>
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 	uint64_t	zh_scratch_state_size;
154 } ztest_shared_hdr_t;
155 
156 static ztest_shared_hdr_t *ztest_shared_hdr;
157 
158 enum ztest_class_state {
159 	ZTEST_VDEV_CLASS_OFF,
160 	ZTEST_VDEV_CLASS_ON,
161 	ZTEST_VDEV_CLASS_RND
162 };
163 
164 /* Dedicated RAIDZ Expansion test states */
165 typedef enum {
166 	RAIDZ_EXPAND_NONE,		/* Default is none, must opt-in	*/
167 	RAIDZ_EXPAND_REQUESTED,		/* The '-X' option was used	*/
168 	RAIDZ_EXPAND_STARTED,		/* Testing has commenced	*/
169 	RAIDZ_EXPAND_KILLED,		/* Reached the proccess kill	*/
170 	RAIDZ_EXPAND_CHECKED,		/* Pool scrub verification done	*/
171 } raidz_expand_test_state_t;
172 
173 
174 #define	ZO_GVARS_MAX_ARGLEN	((size_t)64)
175 #define	ZO_GVARS_MAX_COUNT	((size_t)10)
176 
177 typedef struct ztest_shared_opts {
178 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
179 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
180 	char zo_alt_ztest[MAXNAMELEN];
181 	char zo_alt_libpath[MAXNAMELEN];
182 	uint64_t zo_vdevs;
183 	uint64_t zo_vdevtime;
184 	size_t zo_vdev_size;
185 	int zo_ashift;
186 	int zo_mirrors;
187 	int zo_raid_do_expand;
188 	int zo_raid_children;
189 	int zo_raid_parity;
190 	char zo_raid_type[8];
191 	int zo_draid_data;
192 	int zo_draid_spares;
193 	int zo_datasets;
194 	int zo_threads;
195 	uint64_t zo_passtime;
196 	uint64_t zo_killrate;
197 	int zo_verbose;
198 	int zo_init;
199 	uint64_t zo_time;
200 	uint64_t zo_maxloops;
201 	uint64_t zo_metaslab_force_ganging;
202 	raidz_expand_test_state_t zo_raidz_expand_test;
203 	int zo_mmp_test;
204 	int zo_special_vdevs;
205 	int zo_dump_dbgmsg;
206 	int zo_gvars_count;
207 	char zo_gvars[ZO_GVARS_MAX_COUNT][ZO_GVARS_MAX_ARGLEN];
208 } ztest_shared_opts_t;
209 
210 /* Default values for command line options. */
211 #define	DEFAULT_POOL "ztest"
212 #define	DEFAULT_VDEV_DIR "/tmp"
213 #define	DEFAULT_VDEV_COUNT 5
214 #define	DEFAULT_VDEV_SIZE (SPA_MINDEVSIZE * 4)	/* 256m default size */
215 #define	DEFAULT_VDEV_SIZE_STR "256M"
216 #define	DEFAULT_ASHIFT SPA_MINBLOCKSHIFT
217 #define	DEFAULT_MIRRORS 2
218 #define	DEFAULT_RAID_CHILDREN 4
219 #define	DEFAULT_RAID_PARITY 1
220 #define	DEFAULT_DRAID_DATA 4
221 #define	DEFAULT_DRAID_SPARES 1
222 #define	DEFAULT_DATASETS_COUNT 7
223 #define	DEFAULT_THREADS 23
224 #define	DEFAULT_RUN_TIME 300 /* 300 seconds */
225 #define	DEFAULT_RUN_TIME_STR "300 sec"
226 #define	DEFAULT_PASS_TIME 60 /* 60 seconds */
227 #define	DEFAULT_PASS_TIME_STR "60 sec"
228 #define	DEFAULT_KILL_RATE 70 /* 70% kill rate */
229 #define	DEFAULT_KILLRATE_STR "70%"
230 #define	DEFAULT_INITS 1
231 #define	DEFAULT_MAX_LOOPS 50 /* 5 minutes */
232 #define	DEFAULT_FORCE_GANGING (64 << 10)
233 #define	DEFAULT_FORCE_GANGING_STR "64K"
234 
235 /* Simplifying assumption: -1 is not a valid default. */
236 #define	NO_DEFAULT -1
237 
238 static const ztest_shared_opts_t ztest_opts_defaults = {
239 	.zo_pool = DEFAULT_POOL,
240 	.zo_dir = DEFAULT_VDEV_DIR,
241 	.zo_alt_ztest = { '\0' },
242 	.zo_alt_libpath = { '\0' },
243 	.zo_vdevs = DEFAULT_VDEV_COUNT,
244 	.zo_ashift = DEFAULT_ASHIFT,
245 	.zo_mirrors = DEFAULT_MIRRORS,
246 	.zo_raid_children = DEFAULT_RAID_CHILDREN,
247 	.zo_raid_parity = DEFAULT_RAID_PARITY,
248 	.zo_raid_type = VDEV_TYPE_RAIDZ,
249 	.zo_vdev_size = DEFAULT_VDEV_SIZE,
250 	.zo_draid_data = DEFAULT_DRAID_DATA,	/* data drives */
251 	.zo_draid_spares = DEFAULT_DRAID_SPARES, /* distributed spares */
252 	.zo_datasets = DEFAULT_DATASETS_COUNT,
253 	.zo_threads = DEFAULT_THREADS,
254 	.zo_passtime = DEFAULT_PASS_TIME,
255 	.zo_killrate = DEFAULT_KILL_RATE,
256 	.zo_verbose = 0,
257 	.zo_mmp_test = 0,
258 	.zo_init = DEFAULT_INITS,
259 	.zo_time = DEFAULT_RUN_TIME,
260 	.zo_maxloops = DEFAULT_MAX_LOOPS, /* max loops during spa_freeze() */
261 	.zo_metaslab_force_ganging = DEFAULT_FORCE_GANGING,
262 	.zo_special_vdevs = ZTEST_VDEV_CLASS_RND,
263 	.zo_gvars_count = 0,
264 	.zo_raidz_expand_test = RAIDZ_EXPAND_NONE,
265 };
266 
267 extern uint64_t metaslab_force_ganging;
268 extern uint64_t metaslab_df_alloc_threshold;
269 extern uint64_t zfs_deadman_synctime_ms;
270 extern uint_t metaslab_preload_limit;
271 extern int zfs_compressed_arc_enabled;
272 extern int zfs_abd_scatter_enabled;
273 extern uint_t dmu_object_alloc_chunk_shift;
274 extern boolean_t zfs_force_some_double_word_sm_entries;
275 extern unsigned long zio_decompress_fail_fraction;
276 extern unsigned long zfs_reconstruct_indirect_damage_fraction;
277 extern uint64_t raidz_expand_max_reflow_bytes;
278 extern uint_t raidz_expand_pause_point;
279 extern boolean_t ddt_prune_artificial_age;
280 extern boolean_t ddt_dump_prune_histogram;
281 
282 
283 static ztest_shared_opts_t *ztest_shared_opts;
284 static ztest_shared_opts_t ztest_opts;
285 static const char *const ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
286 
287 typedef struct ztest_shared_ds {
288 	uint64_t	zd_seq;
289 } ztest_shared_ds_t;
290 
291 static ztest_shared_ds_t *ztest_shared_ds;
292 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
293 
294 typedef struct ztest_scratch_state {
295 	uint64_t	zs_raidz_scratch_verify_pause;
296 } ztest_shared_scratch_state_t;
297 
298 static ztest_shared_scratch_state_t *ztest_scratch_state;
299 
300 #define	BT_MAGIC	0x123456789abcdefULL
301 #define	MAXFAULTS(zs) \
302 	(MAX((zs)->zs_mirrors, 1) * (ztest_opts.zo_raid_parity + 1) - 1)
303 
304 enum ztest_io_type {
305 	ZTEST_IO_WRITE_TAG,
306 	ZTEST_IO_WRITE_PATTERN,
307 	ZTEST_IO_WRITE_ZEROES,
308 	ZTEST_IO_TRUNCATE,
309 	ZTEST_IO_SETATTR,
310 	ZTEST_IO_REWRITE,
311 	ZTEST_IO_TYPES
312 };
313 
314 typedef struct ztest_block_tag {
315 	uint64_t	bt_magic;
316 	uint64_t	bt_objset;
317 	uint64_t	bt_object;
318 	uint64_t	bt_dnodesize;
319 	uint64_t	bt_offset;
320 	uint64_t	bt_gen;
321 	uint64_t	bt_txg;
322 	uint64_t	bt_crtxg;
323 } ztest_block_tag_t;
324 
325 typedef struct bufwad {
326 	uint64_t	bw_index;
327 	uint64_t	bw_txg;
328 	uint64_t	bw_data;
329 } bufwad_t;
330 
331 /*
332  * It would be better to use a rangelock_t per object.  Unfortunately
333  * the rangelock_t is not a drop-in replacement for rl_t, because we
334  * still need to map from object ID to rangelock_t.
335  */
336 typedef enum {
337 	ZTRL_READER,
338 	ZTRL_WRITER,
339 	ZTRL_APPEND
340 } rl_type_t;
341 
342 typedef struct rll {
343 	void		*rll_writer;
344 	int		rll_readers;
345 	kmutex_t	rll_lock;
346 	kcondvar_t	rll_cv;
347 } rll_t;
348 
349 typedef struct rl {
350 	uint64_t	rl_object;
351 	uint64_t	rl_offset;
352 	uint64_t	rl_size;
353 	rll_t		*rl_lock;
354 } rl_t;
355 
356 #define	ZTEST_RANGE_LOCKS	64
357 #define	ZTEST_OBJECT_LOCKS	64
358 
359 /*
360  * Object descriptor.  Used as a template for object lookup/create/remove.
361  */
362 typedef struct ztest_od {
363 	uint64_t	od_dir;
364 	uint64_t	od_object;
365 	dmu_object_type_t od_type;
366 	dmu_object_type_t od_crtype;
367 	uint64_t	od_blocksize;
368 	uint64_t	od_crblocksize;
369 	uint64_t	od_crdnodesize;
370 	uint64_t	od_gen;
371 	uint64_t	od_crgen;
372 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
373 } ztest_od_t;
374 
375 /*
376  * Per-dataset state.
377  */
378 typedef struct ztest_ds {
379 	ztest_shared_ds_t *zd_shared;
380 	objset_t	*zd_os;
381 	pthread_rwlock_t zd_zilog_lock;
382 	zilog_t		*zd_zilog;
383 	ztest_od_t	*zd_od;		/* debugging aid */
384 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
385 	kmutex_t	zd_dirobj_lock;
386 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
387 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
388 } ztest_ds_t;
389 
390 /*
391  * Per-iteration state.
392  */
393 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
394 
395 typedef struct ztest_info {
396 	ztest_func_t	*zi_func;	/* test function */
397 	uint64_t	zi_iters;	/* iterations per execution */
398 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
399 	const char	*zi_funcname;	/* name of test function */
400 } ztest_info_t;
401 
402 typedef struct ztest_shared_callstate {
403 	uint64_t	zc_count;	/* per-pass count */
404 	uint64_t	zc_time;	/* per-pass time */
405 	uint64_t	zc_next;	/* next time to call this function */
406 } ztest_shared_callstate_t;
407 
408 static ztest_shared_callstate_t *ztest_shared_callstate;
409 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
410 
411 ztest_func_t ztest_dmu_read_write;
412 ztest_func_t ztest_dmu_write_parallel;
413 ztest_func_t ztest_dmu_object_alloc_free;
414 ztest_func_t ztest_dmu_object_next_chunk;
415 ztest_func_t ztest_dmu_commit_callbacks;
416 ztest_func_t ztest_zap;
417 ztest_func_t ztest_zap_parallel;
418 ztest_func_t ztest_zil_commit;
419 ztest_func_t ztest_zil_remount;
420 ztest_func_t ztest_dmu_read_write_zcopy;
421 ztest_func_t ztest_dmu_objset_create_destroy;
422 ztest_func_t ztest_dmu_prealloc;
423 ztest_func_t ztest_fzap;
424 ztest_func_t ztest_dmu_snapshot_create_destroy;
425 ztest_func_t ztest_dsl_prop_get_set;
426 ztest_func_t ztest_spa_prop_get_set;
427 ztest_func_t ztest_spa_create_destroy;
428 ztest_func_t ztest_fault_inject;
429 ztest_func_t ztest_dmu_snapshot_hold;
430 ztest_func_t ztest_mmp_enable_disable;
431 ztest_func_t ztest_scrub;
432 ztest_func_t ztest_dsl_dataset_promote_busy;
433 ztest_func_t ztest_vdev_attach_detach;
434 ztest_func_t ztest_vdev_raidz_attach;
435 ztest_func_t ztest_vdev_LUN_growth;
436 ztest_func_t ztest_vdev_add_remove;
437 ztest_func_t ztest_vdev_class_add;
438 ztest_func_t ztest_vdev_aux_add_remove;
439 ztest_func_t ztest_split_pool;
440 ztest_func_t ztest_reguid;
441 ztest_func_t ztest_spa_upgrade;
442 ztest_func_t ztest_device_removal;
443 ztest_func_t ztest_spa_checkpoint_create_discard;
444 ztest_func_t ztest_initialize;
445 ztest_func_t ztest_trim;
446 ztest_func_t ztest_blake3;
447 ztest_func_t ztest_fletcher;
448 ztest_func_t ztest_fletcher_incr;
449 ztest_func_t ztest_verify_dnode_bt;
450 ztest_func_t ztest_pool_prefetch_ddt;
451 ztest_func_t ztest_ddt_prune;
452 
453 static uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
454 static uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
455 static uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
456 static uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
457 static uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
458 
459 #define	ZTI_INIT(func, iters, interval) \
460 	{   .zi_func = (func), \
461 	    .zi_iters = (iters), \
462 	    .zi_interval = (interval), \
463 	    .zi_funcname = # func }
464 
465 static ztest_info_t ztest_info[] = {
466 	ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
467 	ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
468 	ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
469 	ZTI_INIT(ztest_dmu_object_next_chunk, 1, &zopt_sometimes),
470 	ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
471 	ZTI_INIT(ztest_zap, 30, &zopt_always),
472 	ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
473 	ZTI_INIT(ztest_split_pool, 1, &zopt_sometimes),
474 	ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
475 	ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
476 	ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
477 	ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
478 	ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
479 	ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
480 #if 0
481 	ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
482 #endif
483 	ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
484 	ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
485 	ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
486 	ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
487 	ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
488 	ZTI_INIT(ztest_mmp_enable_disable, 1, &zopt_sometimes),
489 	ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
490 	ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
491 	ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
492 	ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
493 	ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
494 	ZTI_INIT(ztest_vdev_raidz_attach, 1, &zopt_sometimes),
495 	ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
496 	ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
497 	ZTI_INIT(ztest_vdev_class_add, 1, &ztest_opts.zo_vdevtime),
498 	ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
499 	ZTI_INIT(ztest_device_removal, 1, &zopt_sometimes),
500 	ZTI_INIT(ztest_spa_checkpoint_create_discard, 1, &zopt_rarely),
501 	ZTI_INIT(ztest_initialize, 1, &zopt_sometimes),
502 	ZTI_INIT(ztest_trim, 1, &zopt_sometimes),
503 	ZTI_INIT(ztest_blake3, 1, &zopt_rarely),
504 	ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
505 	ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
506 	ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
507 	ZTI_INIT(ztest_pool_prefetch_ddt, 1, &zopt_rarely),
508 	ZTI_INIT(ztest_ddt_prune, 1, &zopt_rarely),
509 };
510 
511 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
512 
513 /*
514  * The following struct is used to hold a list of uncalled commit callbacks.
515  * The callbacks are ordered by txg number.
516  */
517 typedef struct ztest_cb_list {
518 	kmutex_t	zcl_callbacks_lock;
519 	list_t		zcl_callbacks;
520 } ztest_cb_list_t;
521 
522 /*
523  * Stuff we need to share writably between parent and child.
524  */
525 typedef struct ztest_shared {
526 	boolean_t	zs_do_init;
527 	hrtime_t	zs_proc_start;
528 	hrtime_t	zs_proc_stop;
529 	hrtime_t	zs_thread_start;
530 	hrtime_t	zs_thread_stop;
531 	hrtime_t	zs_thread_kill;
532 	uint64_t	zs_enospc_count;
533 	uint64_t	zs_vdev_next_leaf;
534 	uint64_t	zs_vdev_aux;
535 	uint64_t	zs_alloc;
536 	uint64_t	zs_space;
537 	uint64_t	zs_splits;
538 	uint64_t	zs_mirrors;
539 	uint64_t	zs_metaslab_sz;
540 	uint64_t	zs_metaslab_df_alloc_threshold;
541 	uint64_t	zs_guid;
542 } ztest_shared_t;
543 
544 #define	ID_PARALLEL	-1ULL
545 
546 static char ztest_dev_template[] = "%s/%s.%llua";
547 static char ztest_aux_template[] = "%s/%s.%s.%llu";
548 static ztest_shared_t *ztest_shared;
549 
550 static spa_t *ztest_spa = NULL;
551 static ztest_ds_t *ztest_ds;
552 
553 static kmutex_t ztest_vdev_lock;
554 static boolean_t ztest_device_removal_active = B_FALSE;
555 static boolean_t ztest_pool_scrubbed = B_FALSE;
556 static kmutex_t ztest_checkpoint_lock;
557 
558 /*
559  * The ztest_name_lock protects the pool and dataset namespace used by
560  * the individual tests. To modify the namespace, consumers must grab
561  * this lock as writer. Grabbing the lock as reader will ensure that the
562  * namespace does not change while the lock is held.
563  */
564 static pthread_rwlock_t ztest_name_lock;
565 
566 static boolean_t ztest_dump_core = B_TRUE;
567 static boolean_t ztest_exiting;
568 
569 /* Global commit callback list */
570 static ztest_cb_list_t zcl;
571 /* Commit cb delay */
572 static uint64_t zc_min_txg_delay = UINT64_MAX;
573 static int zc_cb_counter = 0;
574 
575 /*
576  * Minimum number of commit callbacks that need to be registered for us to check
577  * whether the minimum txg delay is acceptable.
578  */
579 #define	ZTEST_COMMIT_CB_MIN_REG	100
580 
581 /*
582  * If a number of txgs equal to this threshold have been created after a commit
583  * callback has been registered but not called, then we assume there is an
584  * implementation bug.
585  */
586 #define	ZTEST_COMMIT_CB_THRESH	(TXG_CONCURRENT_STATES + 1000)
587 
588 enum ztest_object {
589 	ZTEST_META_DNODE = 0,
590 	ZTEST_DIROBJ,
591 	ZTEST_OBJECTS
592 };
593 
594 static __attribute__((noreturn)) void usage(boolean_t requested);
595 static int ztest_scrub_impl(spa_t *spa);
596 
597 /*
598  * These libumem hooks provide a reasonable set of defaults for the allocator's
599  * debugging facilities.
600  */
601 const char *
602 _umem_debug_init(void)
603 {
604 	return ("default,verbose"); /* $UMEM_DEBUG setting */
605 }
606 
607 const char *
608 _umem_logging_init(void)
609 {
610 	return ("fail,contents"); /* $UMEM_LOGGING setting */
611 }
612 
613 static void
614 dump_debug_buffer(void)
615 {
616 	ssize_t ret __attribute__((unused));
617 
618 	if (!ztest_opts.zo_dump_dbgmsg)
619 		return;
620 
621 	/*
622 	 * We use write() instead of printf() so that this function
623 	 * is safe to call from a signal handler.
624 	 */
625 	ret = write(STDERR_FILENO, "\n", 1);
626 	zfs_dbgmsg_print(STDERR_FILENO, "ztest");
627 }
628 
629 static void sig_handler(int signo)
630 {
631 	struct sigaction action;
632 
633 	libspl_backtrace(STDERR_FILENO);
634 	dump_debug_buffer();
635 
636 	/*
637 	 * Restore default action and re-raise signal so SIGSEGV and
638 	 * SIGABRT can trigger a core dump.
639 	 */
640 	action.sa_handler = SIG_DFL;
641 	sigemptyset(&action.sa_mask);
642 	action.sa_flags = 0;
643 	(void) sigaction(signo, &action, NULL);
644 	raise(signo);
645 }
646 
647 #define	FATAL_MSG_SZ	1024
648 
649 static const char *fatal_msg;
650 
651 static __attribute__((format(printf, 2, 3))) __attribute__((noreturn)) void
652 fatal(int do_perror, const char *message, ...)
653 {
654 	va_list args;
655 	int save_errno = errno;
656 	char *buf;
657 
658 	(void) fflush(stdout);
659 	buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
660 	if (buf == NULL)
661 		goto out;
662 
663 	va_start(args, message);
664 	(void) sprintf(buf, "ztest: ");
665 	/* LINTED */
666 	(void) vsprintf(buf + strlen(buf), message, args);
667 	va_end(args);
668 	if (do_perror) {
669 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
670 		    ": %s", strerror(save_errno));
671 	}
672 	(void) fprintf(stderr, "%s\n", buf);
673 	fatal_msg = buf;			/* to ease debugging */
674 
675 out:
676 	if (ztest_dump_core)
677 		abort();
678 	else
679 		dump_debug_buffer();
680 
681 	exit(3);
682 }
683 
684 static int
685 str2shift(const char *buf)
686 {
687 	const char *ends = "BKMGTPEZ";
688 	int i;
689 
690 	if (buf[0] == '\0')
691 		return (0);
692 	for (i = 0; i < strlen(ends); i++) {
693 		if (toupper(buf[0]) == ends[i])
694 			break;
695 	}
696 	if (i == strlen(ends)) {
697 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
698 		    buf);
699 		usage(B_FALSE);
700 	}
701 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
702 		return (10*i);
703 	}
704 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
705 	usage(B_FALSE);
706 }
707 
708 static uint64_t
709 nicenumtoull(const char *buf)
710 {
711 	char *end;
712 	uint64_t val;
713 
714 	val = strtoull(buf, &end, 0);
715 	if (end == buf) {
716 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
717 		usage(B_FALSE);
718 	} else if (end[0] == '.') {
719 		double fval = strtod(buf, &end);
720 		fval *= pow(2, str2shift(end));
721 		/*
722 		 * UINT64_MAX is not exactly representable as a double.
723 		 * The closest representation is UINT64_MAX + 1, so we
724 		 * use a >= comparison instead of > for the bounds check.
725 		 */
726 		if (fval >= (double)UINT64_MAX) {
727 			(void) fprintf(stderr, "ztest: value too large: %s\n",
728 			    buf);
729 			usage(B_FALSE);
730 		}
731 		val = (uint64_t)fval;
732 	} else {
733 		int shift = str2shift(end);
734 		if (shift >= 64 || (val << shift) >> shift != val) {
735 			(void) fprintf(stderr, "ztest: value too large: %s\n",
736 			    buf);
737 			usage(B_FALSE);
738 		}
739 		val <<= shift;
740 	}
741 	return (val);
742 }
743 
744 typedef struct ztest_option {
745 	const char	short_opt;
746 	const char	*long_opt;
747 	const char	*long_opt_param;
748 	const char	*comment;
749 	unsigned int	default_int;
750 	const char	*default_str;
751 } ztest_option_t;
752 
753 /*
754  * The following option_table is used for generating the usage info as well as
755  * the long and short option information for calling getopt_long().
756  */
757 static ztest_option_t option_table[] = {
758 	{ 'v',	"vdevs", "INTEGER", "Number of vdevs", DEFAULT_VDEV_COUNT,
759 	    NULL},
760 	{ 's',	"vdev-size", "INTEGER", "Size of each vdev",
761 	    NO_DEFAULT, DEFAULT_VDEV_SIZE_STR},
762 	{ 'a',	"alignment-shift", "INTEGER",
763 	    "Alignment shift; use 0 for random", DEFAULT_ASHIFT, NULL},
764 	{ 'm',	"mirror-copies", "INTEGER", "Number of mirror copies",
765 	    DEFAULT_MIRRORS, NULL},
766 	{ 'r',	"raid-disks", "INTEGER", "Number of raidz/draid disks",
767 	    DEFAULT_RAID_CHILDREN, NULL},
768 	{ 'R',	"raid-parity", "INTEGER", "Raid parity",
769 	    DEFAULT_RAID_PARITY, NULL},
770 	{ 'K',  "raid-kind", "raidz|eraidz|draid|random", "Raid kind",
771 	    NO_DEFAULT, "random"},
772 	{ 'D',	"draid-data", "INTEGER", "Number of draid data drives",
773 	    DEFAULT_DRAID_DATA, NULL},
774 	{ 'S',	"draid-spares", "INTEGER", "Number of draid spares",
775 	    DEFAULT_DRAID_SPARES, NULL},
776 	{ 'd',	"datasets", "INTEGER", "Number of datasets",
777 	    DEFAULT_DATASETS_COUNT, NULL},
778 	{ 't',	"threads", "INTEGER", "Number of ztest threads",
779 	    DEFAULT_THREADS, NULL},
780 	{ 'g',	"gang-block-threshold", "INTEGER",
781 	    "Metaslab gang block threshold",
782 	    NO_DEFAULT, DEFAULT_FORCE_GANGING_STR},
783 	{ 'i',	"init-count", "INTEGER", "Number of times to initialize pool",
784 	    DEFAULT_INITS, NULL},
785 	{ 'k',	"kill-percentage", "INTEGER", "Kill percentage",
786 	    NO_DEFAULT, DEFAULT_KILLRATE_STR},
787 	{ 'p',	"pool-name", "STRING", "Pool name",
788 	    NO_DEFAULT, DEFAULT_POOL},
789 	{ 'f',	"vdev-file-directory", "PATH", "File directory for vdev files",
790 	    NO_DEFAULT, DEFAULT_VDEV_DIR},
791 	{ 'M',	"multi-host", NULL,
792 	    "Multi-host; simulate pool imported on remote host",
793 	    NO_DEFAULT, NULL},
794 	{ 'E',	"use-existing-pool", NULL,
795 	    "Use existing pool instead of creating new one", NO_DEFAULT, NULL},
796 	{ 'T',	"run-time", "INTEGER", "Total run time",
797 	    NO_DEFAULT, DEFAULT_RUN_TIME_STR},
798 	{ 'P',	"pass-time", "INTEGER", "Time per pass",
799 	    NO_DEFAULT, DEFAULT_PASS_TIME_STR},
800 	{ 'F',	"freeze-loops", "INTEGER", "Max loops in spa_freeze()",
801 	    DEFAULT_MAX_LOOPS, NULL},
802 	{ 'B',	"alt-ztest", "PATH", "Alternate ztest path",
803 	    NO_DEFAULT, NULL},
804 	{ 'C',	"vdev-class-state", "on|off|random", "vdev class state",
805 	    NO_DEFAULT, "random"},
806 	{ 'X', "raidz-expansion", NULL,
807 	    "Perform a dedicated raidz expansion test",
808 	    NO_DEFAULT, NULL},
809 	{ 'o',	"option", "\"OPTION=INTEGER\"",
810 	    "Set global variable to an unsigned 32-bit integer value",
811 	    NO_DEFAULT, NULL},
812 	{ 'G',	"dump-debug-msg", NULL,
813 	    "Dump zfs_dbgmsg buffer before exiting due to an error",
814 	    NO_DEFAULT, NULL},
815 	{ 'V',	"verbose", NULL,
816 	    "Verbose (use multiple times for ever more verbosity)",
817 	    NO_DEFAULT, NULL},
818 	{ 'h',	"help",	NULL, "Show this help",
819 	    NO_DEFAULT, NULL},
820 	{0, 0, 0, 0, 0, 0}
821 };
822 
823 static struct option *long_opts = NULL;
824 static char *short_opts = NULL;
825 
826 static void
827 init_options(void)
828 {
829 	ASSERT3P(long_opts, ==, NULL);
830 	ASSERT3P(short_opts, ==, NULL);
831 
832 	int count = sizeof (option_table) / sizeof (option_table[0]);
833 	long_opts = umem_alloc(sizeof (struct option) * count, UMEM_NOFAIL);
834 
835 	short_opts = umem_alloc(sizeof (char) * 2 * count, UMEM_NOFAIL);
836 	int short_opt_index = 0;
837 
838 	for (int i = 0; i < count; i++) {
839 		long_opts[i].val = option_table[i].short_opt;
840 		long_opts[i].name = option_table[i].long_opt;
841 		long_opts[i].has_arg = option_table[i].long_opt_param != NULL
842 		    ? required_argument : no_argument;
843 		long_opts[i].flag = NULL;
844 		short_opts[short_opt_index++] = option_table[i].short_opt;
845 		if (option_table[i].long_opt_param != NULL) {
846 			short_opts[short_opt_index++] = ':';
847 		}
848 	}
849 }
850 
851 static void
852 fini_options(void)
853 {
854 	int count = sizeof (option_table) / sizeof (option_table[0]);
855 
856 	umem_free(long_opts, sizeof (struct option) * count);
857 	umem_free(short_opts, sizeof (char) * 2 * count);
858 
859 	long_opts = NULL;
860 	short_opts = NULL;
861 }
862 
863 static __attribute__((noreturn)) void
864 usage(boolean_t requested)
865 {
866 	char option[80];
867 	FILE *fp = requested ? stdout : stderr;
868 
869 	(void) fprintf(fp, "Usage: %s [OPTIONS...]\n", DEFAULT_POOL);
870 	for (int i = 0; option_table[i].short_opt != 0; i++) {
871 		if (option_table[i].long_opt_param != NULL) {
872 			(void) sprintf(option, "  -%c --%s=%s",
873 			    option_table[i].short_opt,
874 			    option_table[i].long_opt,
875 			    option_table[i].long_opt_param);
876 		} else {
877 			(void) sprintf(option, "  -%c --%s",
878 			    option_table[i].short_opt,
879 			    option_table[i].long_opt);
880 		}
881 		(void) fprintf(fp, "  %-43s%s", option,
882 		    option_table[i].comment);
883 
884 		if (option_table[i].long_opt_param != NULL) {
885 			if (option_table[i].default_str != NULL) {
886 				(void) fprintf(fp, " (default: %s)",
887 				    option_table[i].default_str);
888 			} else if (option_table[i].default_int != NO_DEFAULT) {
889 				(void) fprintf(fp, " (default: %u)",
890 				    option_table[i].default_int);
891 			}
892 		}
893 		(void) fprintf(fp, "\n");
894 	}
895 	exit(requested ? 0 : 1);
896 }
897 
898 static uint64_t
899 ztest_random(uint64_t range)
900 {
901 	uint64_t r;
902 
903 	ASSERT3S(ztest_fd_rand, >=, 0);
904 
905 	if (range == 0)
906 		return (0);
907 
908 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
909 		fatal(B_TRUE, "short read from /dev/urandom");
910 
911 	return (r % range);
912 }
913 
914 static void
915 ztest_parse_name_value(const char *input, ztest_shared_opts_t *zo)
916 {
917 	char name[32];
918 	char *value;
919 	int state = ZTEST_VDEV_CLASS_RND;
920 
921 	(void) strlcpy(name, input, sizeof (name));
922 
923 	value = strchr(name, '=');
924 	if (value == NULL) {
925 		(void) fprintf(stderr, "missing value in property=value "
926 		    "'-C' argument (%s)\n", input);
927 		usage(B_FALSE);
928 	}
929 	*(value) = '\0';
930 	value++;
931 
932 	if (strcmp(value, "on") == 0) {
933 		state = ZTEST_VDEV_CLASS_ON;
934 	} else if (strcmp(value, "off") == 0) {
935 		state = ZTEST_VDEV_CLASS_OFF;
936 	} else if (strcmp(value, "random") == 0) {
937 		state = ZTEST_VDEV_CLASS_RND;
938 	} else {
939 		(void) fprintf(stderr, "invalid property value '%s'\n", value);
940 		usage(B_FALSE);
941 	}
942 
943 	if (strcmp(name, "special") == 0) {
944 		zo->zo_special_vdevs = state;
945 	} else {
946 		(void) fprintf(stderr, "invalid property name '%s'\n", name);
947 		usage(B_FALSE);
948 	}
949 	if (zo->zo_verbose >= 3)
950 		(void) printf("%s vdev state is '%s'\n", name, value);
951 }
952 
953 static void
954 process_options(int argc, char **argv)
955 {
956 	char *path;
957 	ztest_shared_opts_t *zo = &ztest_opts;
958 
959 	int opt;
960 	uint64_t value;
961 	const char *raid_kind = "random";
962 
963 	memcpy(zo, &ztest_opts_defaults, sizeof (*zo));
964 
965 	init_options();
966 
967 	while ((opt = getopt_long(argc, argv, short_opts, long_opts,
968 	    NULL)) != EOF) {
969 		value = 0;
970 		switch (opt) {
971 		case 'v':
972 		case 's':
973 		case 'a':
974 		case 'm':
975 		case 'r':
976 		case 'R':
977 		case 'D':
978 		case 'S':
979 		case 'd':
980 		case 't':
981 		case 'g':
982 		case 'i':
983 		case 'k':
984 		case 'T':
985 		case 'P':
986 		case 'F':
987 			value = nicenumtoull(optarg);
988 		}
989 		switch (opt) {
990 		case 'v':
991 			zo->zo_vdevs = value;
992 			break;
993 		case 's':
994 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
995 			break;
996 		case 'a':
997 			zo->zo_ashift = value;
998 			break;
999 		case 'm':
1000 			zo->zo_mirrors = value;
1001 			break;
1002 		case 'r':
1003 			zo->zo_raid_children = MAX(1, value);
1004 			break;
1005 		case 'R':
1006 			zo->zo_raid_parity = MIN(MAX(value, 1), 3);
1007 			break;
1008 		case 'K':
1009 			raid_kind = optarg;
1010 			break;
1011 		case 'D':
1012 			zo->zo_draid_data = MAX(1, value);
1013 			break;
1014 		case 'S':
1015 			zo->zo_draid_spares = MAX(1, value);
1016 			break;
1017 		case 'd':
1018 			zo->zo_datasets = MAX(1, value);
1019 			break;
1020 		case 't':
1021 			zo->zo_threads = MAX(1, value);
1022 			break;
1023 		case 'g':
1024 			zo->zo_metaslab_force_ganging =
1025 			    MAX(SPA_MINBLOCKSIZE << 1, value);
1026 			break;
1027 		case 'i':
1028 			zo->zo_init = value;
1029 			break;
1030 		case 'k':
1031 			zo->zo_killrate = value;
1032 			break;
1033 		case 'p':
1034 			(void) strlcpy(zo->zo_pool, optarg,
1035 			    sizeof (zo->zo_pool));
1036 			break;
1037 		case 'f':
1038 			path = realpath(optarg, NULL);
1039 			if (path == NULL) {
1040 				(void) fprintf(stderr, "error: %s: %s\n",
1041 				    optarg, strerror(errno));
1042 				usage(B_FALSE);
1043 			} else {
1044 				(void) strlcpy(zo->zo_dir, path,
1045 				    sizeof (zo->zo_dir));
1046 				free(path);
1047 			}
1048 			break;
1049 		case 'M':
1050 			zo->zo_mmp_test = 1;
1051 			break;
1052 		case 'V':
1053 			zo->zo_verbose++;
1054 			break;
1055 		case 'X':
1056 			zo->zo_raidz_expand_test = RAIDZ_EXPAND_REQUESTED;
1057 			break;
1058 		case 'E':
1059 			zo->zo_init = 0;
1060 			break;
1061 		case 'T':
1062 			zo->zo_time = value;
1063 			break;
1064 		case 'P':
1065 			zo->zo_passtime = MAX(1, value);
1066 			break;
1067 		case 'F':
1068 			zo->zo_maxloops = MAX(1, value);
1069 			break;
1070 		case 'B':
1071 			(void) strlcpy(zo->zo_alt_ztest, optarg,
1072 			    sizeof (zo->zo_alt_ztest));
1073 			break;
1074 		case 'C':
1075 			ztest_parse_name_value(optarg, zo);
1076 			break;
1077 		case 'o':
1078 			if (zo->zo_gvars_count >= ZO_GVARS_MAX_COUNT) {
1079 				(void) fprintf(stderr,
1080 				    "max global var count (%zu) exceeded\n",
1081 				    ZO_GVARS_MAX_COUNT);
1082 				usage(B_FALSE);
1083 			}
1084 			char *v = zo->zo_gvars[zo->zo_gvars_count];
1085 			if (strlcpy(v, optarg, ZO_GVARS_MAX_ARGLEN) >=
1086 			    ZO_GVARS_MAX_ARGLEN) {
1087 				(void) fprintf(stderr,
1088 				    "global var option '%s' is too long\n",
1089 				    optarg);
1090 				usage(B_FALSE);
1091 			}
1092 			zo->zo_gvars_count++;
1093 			break;
1094 		case 'G':
1095 			zo->zo_dump_dbgmsg = 1;
1096 			break;
1097 		case 'h':
1098 			usage(B_TRUE);
1099 			break;
1100 		case '?':
1101 		default:
1102 			usage(B_FALSE);
1103 			break;
1104 		}
1105 	}
1106 
1107 	fini_options();
1108 
1109 	/* Force compatible options for raidz expansion run */
1110 	if (zo->zo_raidz_expand_test == RAIDZ_EXPAND_REQUESTED) {
1111 		zo->zo_mmp_test = 0;
1112 		zo->zo_mirrors = 0;
1113 		zo->zo_vdevs = 1;
1114 		zo->zo_vdev_size = DEFAULT_VDEV_SIZE * 2;
1115 		zo->zo_raid_do_expand = B_FALSE;
1116 		raid_kind = "raidz";
1117 	}
1118 
1119 	if (strcmp(raid_kind, "random") == 0) {
1120 		switch (ztest_random(3)) {
1121 		case 0:
1122 			raid_kind = "raidz";
1123 			break;
1124 		case 1:
1125 			raid_kind = "eraidz";
1126 			break;
1127 		case 2:
1128 			raid_kind = "draid";
1129 			break;
1130 		}
1131 
1132 		if (ztest_opts.zo_verbose >= 3)
1133 			(void) printf("choosing RAID type '%s'\n", raid_kind);
1134 	}
1135 
1136 	if (strcmp(raid_kind, "draid") == 0) {
1137 		uint64_t min_devsize;
1138 
1139 		/* With fewer disk use 256M, otherwise 128M is OK */
1140 		min_devsize = (ztest_opts.zo_raid_children < 16) ?
1141 		    (256ULL << 20) : (128ULL << 20);
1142 
1143 		/* No top-level mirrors with dRAID for now */
1144 		zo->zo_mirrors = 0;
1145 
1146 		/* Use more appropriate defaults for dRAID */
1147 		if (zo->zo_vdevs == ztest_opts_defaults.zo_vdevs)
1148 			zo->zo_vdevs = 1;
1149 		if (zo->zo_raid_children ==
1150 		    ztest_opts_defaults.zo_raid_children)
1151 			zo->zo_raid_children = 16;
1152 		if (zo->zo_ashift < 12)
1153 			zo->zo_ashift = 12;
1154 		if (zo->zo_vdev_size < min_devsize)
1155 			zo->zo_vdev_size = min_devsize;
1156 
1157 		if (zo->zo_draid_data + zo->zo_raid_parity >
1158 		    zo->zo_raid_children - zo->zo_draid_spares) {
1159 			(void) fprintf(stderr, "error: too few draid "
1160 			    "children (%d) for stripe width (%d)\n",
1161 			    zo->zo_raid_children,
1162 			    zo->zo_draid_data + zo->zo_raid_parity);
1163 			usage(B_FALSE);
1164 		}
1165 
1166 		(void) strlcpy(zo->zo_raid_type, VDEV_TYPE_DRAID,
1167 		    sizeof (zo->zo_raid_type));
1168 
1169 	} else if (strcmp(raid_kind, "eraidz") == 0) {
1170 		/* using eraidz (expandable raidz) */
1171 		zo->zo_raid_do_expand = B_TRUE;
1172 
1173 		/* tests expect top-level to be raidz */
1174 		zo->zo_mirrors = 0;
1175 		zo->zo_vdevs = 1;
1176 
1177 		/* Make sure parity is less than data columns */
1178 		zo->zo_raid_parity = MIN(zo->zo_raid_parity,
1179 		    zo->zo_raid_children - 1);
1180 
1181 	} else /* using raidz */ {
1182 		ASSERT0(strcmp(raid_kind, "raidz"));
1183 
1184 		zo->zo_raid_parity = MIN(zo->zo_raid_parity,
1185 		    zo->zo_raid_children - 1);
1186 	}
1187 
1188 	zo->zo_vdevtime =
1189 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
1190 	    UINT64_MAX >> 2);
1191 
1192 	if (*zo->zo_alt_ztest) {
1193 		const char *invalid_what = "ztest";
1194 		char *val = zo->zo_alt_ztest;
1195 		if (0 != access(val, X_OK) ||
1196 		    (strrchr(val, '/') == NULL && (errno == EINVAL)))
1197 			goto invalid;
1198 
1199 		int dirlen = strrchr(val, '/') - val;
1200 		strlcpy(zo->zo_alt_libpath, val,
1201 		    MIN(sizeof (zo->zo_alt_libpath), dirlen + 1));
1202 		invalid_what = "library path", val = zo->zo_alt_libpath;
1203 		if (strrchr(val, '/') == NULL && (errno == EINVAL))
1204 			goto invalid;
1205 		*strrchr(val, '/') = '\0';
1206 		strlcat(val, "/lib", sizeof (zo->zo_alt_libpath));
1207 
1208 		if (0 != access(zo->zo_alt_libpath, X_OK))
1209 			goto invalid;
1210 		return;
1211 
1212 invalid:
1213 		ztest_dump_core = B_FALSE;
1214 		fatal(B_TRUE, "invalid alternate %s %s", invalid_what, val);
1215 	}
1216 }
1217 
1218 static void
1219 ztest_kill(ztest_shared_t *zs)
1220 {
1221 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
1222 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
1223 
1224 	/*
1225 	 * Before we kill ourselves, make sure that the config is updated.
1226 	 * See comment above spa_write_cachefile().
1227 	 */
1228 	if (raidz_expand_pause_point != RAIDZ_EXPAND_PAUSE_NONE) {
1229 		if (mutex_tryenter(&spa_namespace_lock)) {
1230 			spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE,
1231 			    B_FALSE);
1232 			mutex_exit(&spa_namespace_lock);
1233 
1234 			ztest_scratch_state->zs_raidz_scratch_verify_pause =
1235 			    raidz_expand_pause_point;
1236 		} else {
1237 			/*
1238 			 * Do not verify scratch object in case if
1239 			 * spa_namespace_lock cannot be acquired,
1240 			 * it can cause deadlock in spa_config_update().
1241 			 */
1242 			raidz_expand_pause_point = RAIDZ_EXPAND_PAUSE_NONE;
1243 
1244 			return;
1245 		}
1246 	} else {
1247 		mutex_enter(&spa_namespace_lock);
1248 		spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE, B_FALSE);
1249 		mutex_exit(&spa_namespace_lock);
1250 	}
1251 
1252 	(void) raise(SIGKILL);
1253 }
1254 
1255 static void
1256 ztest_record_enospc(const char *s)
1257 {
1258 	(void) s;
1259 	ztest_shared->zs_enospc_count++;
1260 }
1261 
1262 static uint64_t
1263 ztest_get_ashift(void)
1264 {
1265 	if (ztest_opts.zo_ashift == 0)
1266 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
1267 	return (ztest_opts.zo_ashift);
1268 }
1269 
1270 static boolean_t
1271 ztest_is_draid_spare(const char *name)
1272 {
1273 	uint64_t spare_id = 0, parity = 0, vdev_id = 0;
1274 
1275 	if (sscanf(name, VDEV_TYPE_DRAID "%"PRIu64"-%"PRIu64"-%"PRIu64"",
1276 	    &parity, &vdev_id, &spare_id) == 3) {
1277 		return (B_TRUE);
1278 	}
1279 
1280 	return (B_FALSE);
1281 }
1282 
1283 static nvlist_t *
1284 make_vdev_file(const char *path, const char *aux, const char *pool,
1285     size_t size, uint64_t ashift)
1286 {
1287 	char *pathbuf = NULL;
1288 	uint64_t vdev;
1289 	nvlist_t *file;
1290 	boolean_t draid_spare = B_FALSE;
1291 
1292 
1293 	if (ashift == 0)
1294 		ashift = ztest_get_ashift();
1295 
1296 	if (path == NULL) {
1297 		pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1298 		path = pathbuf;
1299 
1300 		if (aux != NULL) {
1301 			vdev = ztest_shared->zs_vdev_aux;
1302 			(void) snprintf(pathbuf, MAXPATHLEN,
1303 			    ztest_aux_template, ztest_opts.zo_dir,
1304 			    pool == NULL ? ztest_opts.zo_pool : pool,
1305 			    aux, vdev);
1306 		} else {
1307 			vdev = ztest_shared->zs_vdev_next_leaf++;
1308 			(void) snprintf(pathbuf, MAXPATHLEN,
1309 			    ztest_dev_template, ztest_opts.zo_dir,
1310 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
1311 		}
1312 	} else {
1313 		draid_spare = ztest_is_draid_spare(path);
1314 	}
1315 
1316 	if (size != 0 && !draid_spare) {
1317 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
1318 		if (fd == -1)
1319 			fatal(B_TRUE, "can't open %s", path);
1320 		if (ftruncate(fd, size) != 0)
1321 			fatal(B_TRUE, "can't ftruncate %s", path);
1322 		(void) close(fd);
1323 	}
1324 
1325 	file = fnvlist_alloc();
1326 	fnvlist_add_string(file, ZPOOL_CONFIG_TYPE,
1327 	    draid_spare ? VDEV_TYPE_DRAID_SPARE : VDEV_TYPE_FILE);
1328 	fnvlist_add_string(file, ZPOOL_CONFIG_PATH, path);
1329 	fnvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift);
1330 	umem_free(pathbuf, MAXPATHLEN);
1331 
1332 	return (file);
1333 }
1334 
1335 static nvlist_t *
1336 make_vdev_raid(const char *path, const char *aux, const char *pool, size_t size,
1337     uint64_t ashift, int r)
1338 {
1339 	nvlist_t *raid, **child;
1340 	int c;
1341 
1342 	if (r < 2)
1343 		return (make_vdev_file(path, aux, pool, size, ashift));
1344 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
1345 
1346 	for (c = 0; c < r; c++)
1347 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
1348 
1349 	raid = fnvlist_alloc();
1350 	fnvlist_add_string(raid, ZPOOL_CONFIG_TYPE,
1351 	    ztest_opts.zo_raid_type);
1352 	fnvlist_add_uint64(raid, ZPOOL_CONFIG_NPARITY,
1353 	    ztest_opts.zo_raid_parity);
1354 	fnvlist_add_nvlist_array(raid, ZPOOL_CONFIG_CHILDREN,
1355 	    (const nvlist_t **)child, r);
1356 
1357 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0) {
1358 		uint64_t ndata = ztest_opts.zo_draid_data;
1359 		uint64_t nparity = ztest_opts.zo_raid_parity;
1360 		uint64_t nspares = ztest_opts.zo_draid_spares;
1361 		uint64_t children = ztest_opts.zo_raid_children;
1362 		uint64_t ngroups = 1;
1363 
1364 		/*
1365 		 * Calculate the minimum number of groups required to fill a
1366 		 * slice. This is the LCM of the stripe width (data + parity)
1367 		 * and the number of data drives (children - spares).
1368 		 */
1369 		while (ngroups * (ndata + nparity) % (children - nspares) != 0)
1370 			ngroups++;
1371 
1372 		/* Store the basic dRAID configuration. */
1373 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NDATA, ndata);
1374 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NSPARES, nspares);
1375 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NGROUPS, ngroups);
1376 	}
1377 
1378 	for (c = 0; c < r; c++)
1379 		fnvlist_free(child[c]);
1380 
1381 	umem_free(child, r * sizeof (nvlist_t *));
1382 
1383 	return (raid);
1384 }
1385 
1386 static nvlist_t *
1387 make_vdev_mirror(const char *path, const char *aux, const char *pool,
1388     size_t size, uint64_t ashift, int r, int m)
1389 {
1390 	nvlist_t *mirror, **child;
1391 	int c;
1392 
1393 	if (m < 1)
1394 		return (make_vdev_raid(path, aux, pool, size, ashift, r));
1395 
1396 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
1397 
1398 	for (c = 0; c < m; c++)
1399 		child[c] = make_vdev_raid(path, aux, pool, size, ashift, r);
1400 
1401 	mirror = fnvlist_alloc();
1402 	fnvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, VDEV_TYPE_MIRROR);
1403 	fnvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
1404 	    (const nvlist_t **)child, m);
1405 
1406 	for (c = 0; c < m; c++)
1407 		fnvlist_free(child[c]);
1408 
1409 	umem_free(child, m * sizeof (nvlist_t *));
1410 
1411 	return (mirror);
1412 }
1413 
1414 static nvlist_t *
1415 make_vdev_root(const char *path, const char *aux, const char *pool, size_t size,
1416     uint64_t ashift, const char *class, int r, int m, int t)
1417 {
1418 	nvlist_t *root, **child;
1419 	int c;
1420 	boolean_t log;
1421 
1422 	ASSERT3S(t, >, 0);
1423 
1424 	log = (class != NULL && strcmp(class, "log") == 0);
1425 
1426 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1427 
1428 	for (c = 0; c < t; c++) {
1429 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1430 		    r, m);
1431 		fnvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, log);
1432 
1433 		if (class != NULL && class[0] != '\0') {
1434 			ASSERT(m > 1 || log);   /* expecting a mirror */
1435 			fnvlist_add_string(child[c],
1436 			    ZPOOL_CONFIG_ALLOCATION_BIAS, class);
1437 		}
1438 	}
1439 
1440 	root = fnvlist_alloc();
1441 	fnvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
1442 	fnvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1443 	    (const nvlist_t **)child, t);
1444 
1445 	for (c = 0; c < t; c++)
1446 		fnvlist_free(child[c]);
1447 
1448 	umem_free(child, t * sizeof (nvlist_t *));
1449 
1450 	return (root);
1451 }
1452 
1453 /*
1454  * Find a random spa version. Returns back a random spa version in the
1455  * range [initial_version, SPA_VERSION_FEATURES].
1456  */
1457 static uint64_t
1458 ztest_random_spa_version(uint64_t initial_version)
1459 {
1460 	uint64_t version = initial_version;
1461 
1462 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
1463 		version = version +
1464 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1465 	}
1466 
1467 	if (version > SPA_VERSION_BEFORE_FEATURES)
1468 		version = SPA_VERSION_FEATURES;
1469 
1470 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1471 	return (version);
1472 }
1473 
1474 static int
1475 ztest_random_blocksize(void)
1476 {
1477 	ASSERT3U(ztest_spa->spa_max_ashift, !=, 0);
1478 
1479 	/*
1480 	 * Choose a block size >= the ashift.
1481 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1482 	 */
1483 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1484 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1485 		maxbs = 20;
1486 	uint64_t block_shift =
1487 	    ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1488 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1489 }
1490 
1491 static int
1492 ztest_random_dnodesize(void)
1493 {
1494 	int slots;
1495 	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1496 
1497 	if (max_slots == DNODE_MIN_SLOTS)
1498 		return (DNODE_MIN_SIZE);
1499 
1500 	/*
1501 	 * Weight the random distribution more heavily toward smaller
1502 	 * dnode sizes since that is more likely to reflect real-world
1503 	 * usage.
1504 	 */
1505 	ASSERT3U(max_slots, >, 4);
1506 	switch (ztest_random(10)) {
1507 	case 0:
1508 		slots = 5 + ztest_random(max_slots - 4);
1509 		break;
1510 	case 1 ... 4:
1511 		slots = 2 + ztest_random(3);
1512 		break;
1513 	default:
1514 		slots = 1;
1515 		break;
1516 	}
1517 
1518 	return (slots << DNODE_SHIFT);
1519 }
1520 
1521 static int
1522 ztest_random_ibshift(void)
1523 {
1524 	return (DN_MIN_INDBLKSHIFT +
1525 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1526 }
1527 
1528 static uint64_t
1529 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1530 {
1531 	uint64_t top;
1532 	vdev_t *rvd = spa->spa_root_vdev;
1533 	vdev_t *tvd;
1534 
1535 	ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
1536 
1537 	do {
1538 		top = ztest_random(rvd->vdev_children);
1539 		tvd = rvd->vdev_child[top];
1540 	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1541 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1542 
1543 	return (top);
1544 }
1545 
1546 static uint64_t
1547 ztest_random_dsl_prop(zfs_prop_t prop)
1548 {
1549 	uint64_t value;
1550 
1551 	do {
1552 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1553 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1554 
1555 	return (value);
1556 }
1557 
1558 static int
1559 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1560     boolean_t inherit)
1561 {
1562 	const char *propname = zfs_prop_to_name(prop);
1563 	const char *valname;
1564 	char *setpoint;
1565 	uint64_t curval;
1566 	int error;
1567 
1568 	error = dsl_prop_set_int(osname, propname,
1569 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1570 
1571 	if (error == ENOSPC) {
1572 		ztest_record_enospc(FTAG);
1573 		return (error);
1574 	}
1575 	ASSERT0(error);
1576 
1577 	setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1578 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1579 
1580 	if (ztest_opts.zo_verbose >= 6) {
1581 		int err;
1582 
1583 		err = zfs_prop_index_to_string(prop, curval, &valname);
1584 		if (err)
1585 			(void) printf("%s %s = %llu at '%s'\n", osname,
1586 			    propname, (unsigned long long)curval, setpoint);
1587 		else
1588 			(void) printf("%s %s = %s at '%s'\n",
1589 			    osname, propname, valname, setpoint);
1590 	}
1591 	umem_free(setpoint, MAXPATHLEN);
1592 
1593 	return (error);
1594 }
1595 
1596 static int
1597 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1598 {
1599 	spa_t *spa = ztest_spa;
1600 	nvlist_t *props = NULL;
1601 	int error;
1602 
1603 	props = fnvlist_alloc();
1604 	fnvlist_add_uint64(props, zpool_prop_to_name(prop), value);
1605 
1606 	error = spa_prop_set(spa, props);
1607 
1608 	fnvlist_free(props);
1609 
1610 	if (error == ENOSPC) {
1611 		ztest_record_enospc(FTAG);
1612 		return (error);
1613 	}
1614 	ASSERT0(error);
1615 
1616 	return (error);
1617 }
1618 
1619 static int
1620 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1621     boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp)
1622 {
1623 	int err;
1624 	char *cp = NULL;
1625 	char ddname[ZFS_MAX_DATASET_NAME_LEN];
1626 
1627 	strlcpy(ddname, name, sizeof (ddname));
1628 	cp = strchr(ddname, '@');
1629 	if (cp != NULL)
1630 		*cp = '\0';
1631 
1632 	err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1633 	while (decrypt && err == EACCES) {
1634 		dsl_crypto_params_t *dcp;
1635 		nvlist_t *crypto_args = fnvlist_alloc();
1636 
1637 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
1638 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1639 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1640 		    crypto_args, &dcp));
1641 		err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1642 		/*
1643 		 * Note: if there was an error loading, the wkey was not
1644 		 * consumed, and needs to be freed.
1645 		 */
1646 		dsl_crypto_params_free(dcp, (err != 0));
1647 		fnvlist_free(crypto_args);
1648 
1649 		if (err == EINVAL) {
1650 			/*
1651 			 * We couldn't load a key for this dataset so try
1652 			 * the parent. This loop will eventually hit the
1653 			 * encryption root since ztest only makes clones
1654 			 * as children of their origin datasets.
1655 			 */
1656 			cp = strrchr(ddname, '/');
1657 			if (cp == NULL)
1658 				return (err);
1659 
1660 			*cp = '\0';
1661 			err = EACCES;
1662 			continue;
1663 		} else if (err != 0) {
1664 			break;
1665 		}
1666 
1667 		err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1668 		break;
1669 	}
1670 
1671 	return (err);
1672 }
1673 
1674 static void
1675 ztest_rll_init(rll_t *rll)
1676 {
1677 	rll->rll_writer = NULL;
1678 	rll->rll_readers = 0;
1679 	mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1680 	cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1681 }
1682 
1683 static void
1684 ztest_rll_destroy(rll_t *rll)
1685 {
1686 	ASSERT3P(rll->rll_writer, ==, NULL);
1687 	ASSERT0(rll->rll_readers);
1688 	mutex_destroy(&rll->rll_lock);
1689 	cv_destroy(&rll->rll_cv);
1690 }
1691 
1692 static void
1693 ztest_rll_lock(rll_t *rll, rl_type_t type)
1694 {
1695 	mutex_enter(&rll->rll_lock);
1696 
1697 	if (type == ZTRL_READER) {
1698 		while (rll->rll_writer != NULL)
1699 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1700 		rll->rll_readers++;
1701 	} else {
1702 		while (rll->rll_writer != NULL || rll->rll_readers)
1703 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1704 		rll->rll_writer = curthread;
1705 	}
1706 
1707 	mutex_exit(&rll->rll_lock);
1708 }
1709 
1710 static void
1711 ztest_rll_unlock(rll_t *rll)
1712 {
1713 	mutex_enter(&rll->rll_lock);
1714 
1715 	if (rll->rll_writer) {
1716 		ASSERT0(rll->rll_readers);
1717 		rll->rll_writer = NULL;
1718 	} else {
1719 		ASSERT3S(rll->rll_readers, >, 0);
1720 		ASSERT3P(rll->rll_writer, ==, NULL);
1721 		rll->rll_readers--;
1722 	}
1723 
1724 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1725 		cv_broadcast(&rll->rll_cv);
1726 
1727 	mutex_exit(&rll->rll_lock);
1728 }
1729 
1730 static void
1731 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1732 {
1733 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1734 
1735 	ztest_rll_lock(rll, type);
1736 }
1737 
1738 static void
1739 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1740 {
1741 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1742 
1743 	ztest_rll_unlock(rll);
1744 }
1745 
1746 static rl_t *
1747 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1748     uint64_t size, rl_type_t type)
1749 {
1750 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1751 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1752 	rl_t *rl;
1753 
1754 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1755 	rl->rl_object = object;
1756 	rl->rl_offset = offset;
1757 	rl->rl_size = size;
1758 	rl->rl_lock = rll;
1759 
1760 	ztest_rll_lock(rll, type);
1761 
1762 	return (rl);
1763 }
1764 
1765 static void
1766 ztest_range_unlock(rl_t *rl)
1767 {
1768 	rll_t *rll = rl->rl_lock;
1769 
1770 	ztest_rll_unlock(rll);
1771 
1772 	umem_free(rl, sizeof (*rl));
1773 }
1774 
1775 static void
1776 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1777 {
1778 	zd->zd_os = os;
1779 	zd->zd_zilog = dmu_objset_zil(os);
1780 	zd->zd_shared = szd;
1781 	dmu_objset_name(os, zd->zd_name);
1782 	int l;
1783 
1784 	if (zd->zd_shared != NULL)
1785 		zd->zd_shared->zd_seq = 0;
1786 
1787 	VERIFY0(pthread_rwlock_init(&zd->zd_zilog_lock, NULL));
1788 	mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1789 
1790 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1791 		ztest_rll_init(&zd->zd_object_lock[l]);
1792 
1793 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1794 		ztest_rll_init(&zd->zd_range_lock[l]);
1795 }
1796 
1797 static void
1798 ztest_zd_fini(ztest_ds_t *zd)
1799 {
1800 	int l;
1801 
1802 	mutex_destroy(&zd->zd_dirobj_lock);
1803 	(void) pthread_rwlock_destroy(&zd->zd_zilog_lock);
1804 
1805 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1806 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1807 
1808 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1809 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1810 }
1811 
1812 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1813 
1814 static uint64_t
1815 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1816 {
1817 	uint64_t txg;
1818 	int error;
1819 
1820 	/*
1821 	 * Attempt to assign tx to some transaction group.
1822 	 */
1823 	error = dmu_tx_assign(tx, txg_how);
1824 	if (error) {
1825 		if (error == ERESTART) {
1826 			ASSERT3U(txg_how, ==, TXG_NOWAIT);
1827 			dmu_tx_wait(tx);
1828 		} else {
1829 			ASSERT3U(error, ==, ENOSPC);
1830 			ztest_record_enospc(tag);
1831 		}
1832 		dmu_tx_abort(tx);
1833 		return (0);
1834 	}
1835 	txg = dmu_tx_get_txg(tx);
1836 	ASSERT3U(txg, !=, 0);
1837 	return (txg);
1838 }
1839 
1840 static void
1841 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1842     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1843     uint64_t crtxg)
1844 {
1845 	bt->bt_magic = BT_MAGIC;
1846 	bt->bt_objset = dmu_objset_id(os);
1847 	bt->bt_object = object;
1848 	bt->bt_dnodesize = dnodesize;
1849 	bt->bt_offset = offset;
1850 	bt->bt_gen = gen;
1851 	bt->bt_txg = txg;
1852 	bt->bt_crtxg = crtxg;
1853 }
1854 
1855 static void
1856 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1857     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1858     uint64_t crtxg)
1859 {
1860 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1861 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1862 	ASSERT3U(bt->bt_object, ==, object);
1863 	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1864 	ASSERT3U(bt->bt_offset, ==, offset);
1865 	ASSERT3U(bt->bt_gen, <=, gen);
1866 	ASSERT3U(bt->bt_txg, <=, txg);
1867 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1868 }
1869 
1870 static ztest_block_tag_t *
1871 ztest_bt_bonus(dmu_buf_t *db)
1872 {
1873 	dmu_object_info_t doi;
1874 	ztest_block_tag_t *bt;
1875 
1876 	dmu_object_info_from_db(db, &doi);
1877 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1878 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1879 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1880 
1881 	return (bt);
1882 }
1883 
1884 /*
1885  * Generate a token to fill up unused bonus buffer space.  Try to make
1886  * it unique to the object, generation, and offset to verify that data
1887  * is not getting overwritten by data from other dnodes.
1888  */
1889 #define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1890 	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1891 
1892 /*
1893  * Fill up the unused bonus buffer region before the block tag with a
1894  * verifiable pattern. Filling the whole bonus area with non-zero data
1895  * helps ensure that all dnode traversal code properly skips the
1896  * interior regions of large dnodes.
1897  */
1898 static void
1899 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1900     objset_t *os, uint64_t gen)
1901 {
1902 	uint64_t *bonusp;
1903 
1904 	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1905 
1906 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1907 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1908 		    gen, bonusp - (uint64_t *)db->db_data);
1909 		*bonusp = token;
1910 	}
1911 }
1912 
1913 /*
1914  * Verify that the unused area of a bonus buffer is filled with the
1915  * expected tokens.
1916  */
1917 static void
1918 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1919     objset_t *os, uint64_t gen)
1920 {
1921 	uint64_t *bonusp;
1922 
1923 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1924 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1925 		    gen, bonusp - (uint64_t *)db->db_data);
1926 		VERIFY3U(*bonusp, ==, token);
1927 	}
1928 }
1929 
1930 /*
1931  * ZIL logging ops
1932  */
1933 
1934 #define	lrz_type	lr_mode
1935 #define	lrz_blocksize	lr_uid
1936 #define	lrz_ibshift	lr_gid
1937 #define	lrz_bonustype	lr_rdev
1938 #define	lrz_dnodesize	lr_crtime[1]
1939 
1940 static void
1941 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1942 {
1943 	char *name = (void *)(lr + 1);		/* name follows lr */
1944 	size_t namesize = strlen(name) + 1;
1945 	itx_t *itx;
1946 
1947 	if (zil_replaying(zd->zd_zilog, tx))
1948 		return;
1949 
1950 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1951 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1952 	    sizeof (*lr) + namesize - sizeof (lr_t));
1953 
1954 	zil_itx_assign(zd->zd_zilog, itx, tx);
1955 }
1956 
1957 static void
1958 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1959 {
1960 	char *name = (void *)(lr + 1);		/* name follows lr */
1961 	size_t namesize = strlen(name) + 1;
1962 	itx_t *itx;
1963 
1964 	if (zil_replaying(zd->zd_zilog, tx))
1965 		return;
1966 
1967 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1968 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1969 	    sizeof (*lr) + namesize - sizeof (lr_t));
1970 
1971 	itx->itx_oid = object;
1972 	zil_itx_assign(zd->zd_zilog, itx, tx);
1973 }
1974 
1975 static void
1976 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1977 {
1978 	itx_t *itx;
1979 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1980 
1981 	if (zil_replaying(zd->zd_zilog, tx))
1982 		return;
1983 
1984 	if (lr->lr_length > zil_max_log_data(zd->zd_zilog, sizeof (lr_write_t)))
1985 		write_state = WR_INDIRECT;
1986 
1987 	itx = zil_itx_create(TX_WRITE,
1988 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1989 
1990 	if (write_state == WR_COPIED &&
1991 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1992 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1993 		zil_itx_destroy(itx);
1994 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1995 		write_state = WR_NEED_COPY;
1996 	}
1997 	itx->itx_private = zd;
1998 	itx->itx_wr_state = write_state;
1999 	itx->itx_sync = (ztest_random(8) == 0);
2000 
2001 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
2002 	    sizeof (*lr) - sizeof (lr_t));
2003 
2004 	zil_itx_assign(zd->zd_zilog, itx, tx);
2005 }
2006 
2007 static void
2008 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
2009 {
2010 	itx_t *itx;
2011 
2012 	if (zil_replaying(zd->zd_zilog, tx))
2013 		return;
2014 
2015 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2016 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
2017 	    sizeof (*lr) - sizeof (lr_t));
2018 
2019 	itx->itx_sync = B_FALSE;
2020 	zil_itx_assign(zd->zd_zilog, itx, tx);
2021 }
2022 
2023 static void
2024 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
2025 {
2026 	itx_t *itx;
2027 
2028 	if (zil_replaying(zd->zd_zilog, tx))
2029 		return;
2030 
2031 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
2032 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
2033 	    sizeof (*lr) - sizeof (lr_t));
2034 
2035 	itx->itx_sync = B_FALSE;
2036 	zil_itx_assign(zd->zd_zilog, itx, tx);
2037 }
2038 
2039 /*
2040  * ZIL replay ops
2041  */
2042 static int
2043 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
2044 {
2045 	ztest_ds_t *zd = arg1;
2046 	lr_create_t *lr = arg2;
2047 	char *name = (void *)(lr + 1);		/* name follows lr */
2048 	objset_t *os = zd->zd_os;
2049 	ztest_block_tag_t *bbt;
2050 	dmu_buf_t *db;
2051 	dmu_tx_t *tx;
2052 	uint64_t txg;
2053 	int error = 0;
2054 	int bonuslen;
2055 
2056 	if (byteswap)
2057 		byteswap_uint64_array(lr, sizeof (*lr));
2058 
2059 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
2060 	ASSERT3S(name[0], !=, '\0');
2061 
2062 	tx = dmu_tx_create(os);
2063 
2064 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
2065 
2066 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
2067 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
2068 	} else {
2069 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2070 	}
2071 
2072 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2073 	if (txg == 0)
2074 		return (ENOSPC);
2075 
2076 	ASSERT3U(dmu_objset_zil(os)->zl_replay, ==, !!lr->lr_foid);
2077 	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
2078 
2079 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
2080 		if (lr->lr_foid == 0) {
2081 			lr->lr_foid = zap_create_dnsize(os,
2082 			    lr->lrz_type, lr->lrz_bonustype,
2083 			    bonuslen, lr->lrz_dnodesize, tx);
2084 		} else {
2085 			error = zap_create_claim_dnsize(os, lr->lr_foid,
2086 			    lr->lrz_type, lr->lrz_bonustype,
2087 			    bonuslen, lr->lrz_dnodesize, tx);
2088 		}
2089 	} else {
2090 		if (lr->lr_foid == 0) {
2091 			lr->lr_foid = dmu_object_alloc_dnsize(os,
2092 			    lr->lrz_type, 0, lr->lrz_bonustype,
2093 			    bonuslen, lr->lrz_dnodesize, tx);
2094 		} else {
2095 			error = dmu_object_claim_dnsize(os, lr->lr_foid,
2096 			    lr->lrz_type, 0, lr->lrz_bonustype,
2097 			    bonuslen, lr->lrz_dnodesize, tx);
2098 		}
2099 	}
2100 
2101 	if (error) {
2102 		ASSERT3U(error, ==, EEXIST);
2103 		ASSERT(zd->zd_zilog->zl_replay);
2104 		dmu_tx_commit(tx);
2105 		return (error);
2106 	}
2107 
2108 	ASSERT3U(lr->lr_foid, !=, 0);
2109 
2110 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
2111 		VERIFY0(dmu_object_set_blocksize(os, lr->lr_foid,
2112 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
2113 
2114 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2115 	bbt = ztest_bt_bonus(db);
2116 	dmu_buf_will_dirty(db, tx);
2117 	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
2118 	    lr->lr_gen, txg, txg);
2119 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
2120 	dmu_buf_rele(db, FTAG);
2121 
2122 	VERIFY0(zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
2123 	    &lr->lr_foid, tx));
2124 
2125 	(void) ztest_log_create(zd, tx, lr);
2126 
2127 	dmu_tx_commit(tx);
2128 
2129 	return (0);
2130 }
2131 
2132 static int
2133 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
2134 {
2135 	ztest_ds_t *zd = arg1;
2136 	lr_remove_t *lr = arg2;
2137 	char *name = (void *)(lr + 1);		/* name follows lr */
2138 	objset_t *os = zd->zd_os;
2139 	dmu_object_info_t doi;
2140 	dmu_tx_t *tx;
2141 	uint64_t object, txg;
2142 
2143 	if (byteswap)
2144 		byteswap_uint64_array(lr, sizeof (*lr));
2145 
2146 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
2147 	ASSERT3S(name[0], !=, '\0');
2148 
2149 	VERIFY0(
2150 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
2151 	ASSERT3U(object, !=, 0);
2152 
2153 	ztest_object_lock(zd, object, ZTRL_WRITER);
2154 
2155 	VERIFY0(dmu_object_info(os, object, &doi));
2156 
2157 	tx = dmu_tx_create(os);
2158 
2159 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
2160 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2161 
2162 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2163 	if (txg == 0) {
2164 		ztest_object_unlock(zd, object);
2165 		return (ENOSPC);
2166 	}
2167 
2168 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
2169 		VERIFY0(zap_destroy(os, object, tx));
2170 	} else {
2171 		VERIFY0(dmu_object_free(os, object, tx));
2172 	}
2173 
2174 	VERIFY0(zap_remove(os, lr->lr_doid, name, tx));
2175 
2176 	(void) ztest_log_remove(zd, tx, lr, object);
2177 
2178 	dmu_tx_commit(tx);
2179 
2180 	ztest_object_unlock(zd, object);
2181 
2182 	return (0);
2183 }
2184 
2185 static int
2186 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
2187 {
2188 	ztest_ds_t *zd = arg1;
2189 	lr_write_t *lr = arg2;
2190 	objset_t *os = zd->zd_os;
2191 	void *data = lr + 1;			/* data follows lr */
2192 	uint64_t offset, length;
2193 	ztest_block_tag_t *bt = data;
2194 	ztest_block_tag_t *bbt;
2195 	uint64_t gen, txg, lrtxg, crtxg;
2196 	dmu_object_info_t doi;
2197 	dmu_tx_t *tx;
2198 	dmu_buf_t *db;
2199 	arc_buf_t *abuf = NULL;
2200 	rl_t *rl;
2201 
2202 	if (byteswap)
2203 		byteswap_uint64_array(lr, sizeof (*lr));
2204 
2205 	offset = lr->lr_offset;
2206 	length = lr->lr_length;
2207 
2208 	/* If it's a dmu_sync() block, write the whole block */
2209 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
2210 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
2211 		if (length < blocksize) {
2212 			offset -= offset % blocksize;
2213 			length = blocksize;
2214 		}
2215 	}
2216 
2217 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
2218 		byteswap_uint64_array(bt, sizeof (*bt));
2219 
2220 	if (bt->bt_magic != BT_MAGIC)
2221 		bt = NULL;
2222 
2223 	ztest_object_lock(zd, lr->lr_foid, ZTRL_READER);
2224 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, ZTRL_WRITER);
2225 
2226 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2227 
2228 	dmu_object_info_from_db(db, &doi);
2229 
2230 	bbt = ztest_bt_bonus(db);
2231 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2232 	gen = bbt->bt_gen;
2233 	crtxg = bbt->bt_crtxg;
2234 	lrtxg = lr->lr_common.lrc_txg;
2235 
2236 	tx = dmu_tx_create(os);
2237 
2238 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
2239 
2240 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
2241 	    P2PHASE(offset, length) == 0)
2242 		abuf = dmu_request_arcbuf(db, length);
2243 
2244 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2245 	if (txg == 0) {
2246 		if (abuf != NULL)
2247 			dmu_return_arcbuf(abuf);
2248 		dmu_buf_rele(db, FTAG);
2249 		ztest_range_unlock(rl);
2250 		ztest_object_unlock(zd, lr->lr_foid);
2251 		return (ENOSPC);
2252 	}
2253 
2254 	if (bt != NULL) {
2255 		/*
2256 		 * Usually, verify the old data before writing new data --
2257 		 * but not always, because we also want to verify correct
2258 		 * behavior when the data was not recently read into cache.
2259 		 */
2260 		ASSERT(doi.doi_data_block_size);
2261 		ASSERT0(offset % doi.doi_data_block_size);
2262 		if (ztest_random(4) != 0) {
2263 			int prefetch = ztest_random(2) ?
2264 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
2265 			ztest_block_tag_t rbt;
2266 
2267 			VERIFY(dmu_read(os, lr->lr_foid, offset,
2268 			    sizeof (rbt), &rbt, prefetch) == 0);
2269 			if (rbt.bt_magic == BT_MAGIC) {
2270 				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
2271 				    offset, gen, txg, crtxg);
2272 			}
2273 		}
2274 
2275 		/*
2276 		 * Writes can appear to be newer than the bonus buffer because
2277 		 * the ztest_get_data() callback does a dmu_read() of the
2278 		 * open-context data, which may be different than the data
2279 		 * as it was when the write was generated.
2280 		 */
2281 		if (zd->zd_zilog->zl_replay) {
2282 			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
2283 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
2284 			    bt->bt_crtxg);
2285 		}
2286 
2287 		/*
2288 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
2289 		 * so that all of the usual ASSERTs will work.
2290 		 */
2291 		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
2292 		    crtxg);
2293 	}
2294 
2295 	if (abuf == NULL) {
2296 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
2297 	} else {
2298 		memcpy(abuf->b_data, data, length);
2299 		VERIFY0(dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx));
2300 	}
2301 
2302 	(void) ztest_log_write(zd, tx, lr);
2303 
2304 	dmu_buf_rele(db, FTAG);
2305 
2306 	dmu_tx_commit(tx);
2307 
2308 	ztest_range_unlock(rl);
2309 	ztest_object_unlock(zd, lr->lr_foid);
2310 
2311 	return (0);
2312 }
2313 
2314 static int
2315 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
2316 {
2317 	ztest_ds_t *zd = arg1;
2318 	lr_truncate_t *lr = arg2;
2319 	objset_t *os = zd->zd_os;
2320 	dmu_tx_t *tx;
2321 	uint64_t txg;
2322 	rl_t *rl;
2323 
2324 	if (byteswap)
2325 		byteswap_uint64_array(lr, sizeof (*lr));
2326 
2327 	ztest_object_lock(zd, lr->lr_foid, ZTRL_READER);
2328 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
2329 	    ZTRL_WRITER);
2330 
2331 	tx = dmu_tx_create(os);
2332 
2333 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
2334 
2335 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2336 	if (txg == 0) {
2337 		ztest_range_unlock(rl);
2338 		ztest_object_unlock(zd, lr->lr_foid);
2339 		return (ENOSPC);
2340 	}
2341 
2342 	VERIFY0(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
2343 	    lr->lr_length, tx));
2344 
2345 	(void) ztest_log_truncate(zd, tx, lr);
2346 
2347 	dmu_tx_commit(tx);
2348 
2349 	ztest_range_unlock(rl);
2350 	ztest_object_unlock(zd, lr->lr_foid);
2351 
2352 	return (0);
2353 }
2354 
2355 static int
2356 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
2357 {
2358 	ztest_ds_t *zd = arg1;
2359 	lr_setattr_t *lr = arg2;
2360 	objset_t *os = zd->zd_os;
2361 	dmu_tx_t *tx;
2362 	dmu_buf_t *db;
2363 	ztest_block_tag_t *bbt;
2364 	uint64_t txg, lrtxg, crtxg, dnodesize;
2365 
2366 	if (byteswap)
2367 		byteswap_uint64_array(lr, sizeof (*lr));
2368 
2369 	ztest_object_lock(zd, lr->lr_foid, ZTRL_WRITER);
2370 
2371 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2372 
2373 	tx = dmu_tx_create(os);
2374 	dmu_tx_hold_bonus(tx, lr->lr_foid);
2375 
2376 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2377 	if (txg == 0) {
2378 		dmu_buf_rele(db, FTAG);
2379 		ztest_object_unlock(zd, lr->lr_foid);
2380 		return (ENOSPC);
2381 	}
2382 
2383 	bbt = ztest_bt_bonus(db);
2384 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2385 	crtxg = bbt->bt_crtxg;
2386 	lrtxg = lr->lr_common.lrc_txg;
2387 	dnodesize = bbt->bt_dnodesize;
2388 
2389 	if (zd->zd_zilog->zl_replay) {
2390 		ASSERT3U(lr->lr_size, !=, 0);
2391 		ASSERT3U(lr->lr_mode, !=, 0);
2392 		ASSERT3U(lrtxg, !=, 0);
2393 	} else {
2394 		/*
2395 		 * Randomly change the size and increment the generation.
2396 		 */
2397 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2398 		    sizeof (*bbt);
2399 		lr->lr_mode = bbt->bt_gen + 1;
2400 		ASSERT0(lrtxg);
2401 	}
2402 
2403 	/*
2404 	 * Verify that the current bonus buffer is not newer than our txg.
2405 	 */
2406 	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2407 	    MAX(txg, lrtxg), crtxg);
2408 
2409 	dmu_buf_will_dirty(db, tx);
2410 
2411 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2412 	ASSERT3U(lr->lr_size, <=, db->db_size);
2413 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2414 	bbt = ztest_bt_bonus(db);
2415 
2416 	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2417 	    txg, crtxg);
2418 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2419 	dmu_buf_rele(db, FTAG);
2420 
2421 	(void) ztest_log_setattr(zd, tx, lr);
2422 
2423 	dmu_tx_commit(tx);
2424 
2425 	ztest_object_unlock(zd, lr->lr_foid);
2426 
2427 	return (0);
2428 }
2429 
2430 static zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
2431 	NULL,			/* 0 no such transaction type */
2432 	ztest_replay_create,	/* TX_CREATE */
2433 	NULL,			/* TX_MKDIR */
2434 	NULL,			/* TX_MKXATTR */
2435 	NULL,			/* TX_SYMLINK */
2436 	ztest_replay_remove,	/* TX_REMOVE */
2437 	NULL,			/* TX_RMDIR */
2438 	NULL,			/* TX_LINK */
2439 	NULL,			/* TX_RENAME */
2440 	ztest_replay_write,	/* TX_WRITE */
2441 	ztest_replay_truncate,	/* TX_TRUNCATE */
2442 	ztest_replay_setattr,	/* TX_SETATTR */
2443 	NULL,			/* TX_ACL */
2444 	NULL,			/* TX_CREATE_ACL */
2445 	NULL,			/* TX_CREATE_ATTR */
2446 	NULL,			/* TX_CREATE_ACL_ATTR */
2447 	NULL,			/* TX_MKDIR_ACL */
2448 	NULL,			/* TX_MKDIR_ATTR */
2449 	NULL,			/* TX_MKDIR_ACL_ATTR */
2450 	NULL,			/* TX_WRITE2 */
2451 	NULL,			/* TX_SETSAXATTR */
2452 	NULL,			/* TX_RENAME_EXCHANGE */
2453 	NULL,			/* TX_RENAME_WHITEOUT */
2454 };
2455 
2456 /*
2457  * ZIL get_data callbacks
2458  */
2459 
2460 static void
2461 ztest_get_done(zgd_t *zgd, int error)
2462 {
2463 	(void) error;
2464 	ztest_ds_t *zd = zgd->zgd_private;
2465 	uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object;
2466 
2467 	if (zgd->zgd_db)
2468 		dmu_buf_rele(zgd->zgd_db, zgd);
2469 
2470 	ztest_range_unlock((rl_t *)zgd->zgd_lr);
2471 	ztest_object_unlock(zd, object);
2472 
2473 	umem_free(zgd, sizeof (*zgd));
2474 }
2475 
2476 static int
2477 ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
2478     struct lwb *lwb, zio_t *zio)
2479 {
2480 	(void) arg2;
2481 	ztest_ds_t *zd = arg;
2482 	objset_t *os = zd->zd_os;
2483 	uint64_t object = lr->lr_foid;
2484 	uint64_t offset = lr->lr_offset;
2485 	uint64_t size = lr->lr_length;
2486 	uint64_t txg = lr->lr_common.lrc_txg;
2487 	uint64_t crtxg;
2488 	dmu_object_info_t doi;
2489 	dmu_buf_t *db;
2490 	zgd_t *zgd;
2491 	int error;
2492 
2493 	ASSERT3P(lwb, !=, NULL);
2494 	ASSERT3U(size, !=, 0);
2495 
2496 	ztest_object_lock(zd, object, ZTRL_READER);
2497 	error = dmu_bonus_hold(os, object, FTAG, &db);
2498 	if (error) {
2499 		ztest_object_unlock(zd, object);
2500 		return (error);
2501 	}
2502 
2503 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
2504 
2505 	if (crtxg == 0 || crtxg > txg) {
2506 		dmu_buf_rele(db, FTAG);
2507 		ztest_object_unlock(zd, object);
2508 		return (ENOENT);
2509 	}
2510 
2511 	dmu_object_info_from_db(db, &doi);
2512 	dmu_buf_rele(db, FTAG);
2513 	db = NULL;
2514 
2515 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2516 	zgd->zgd_lwb = lwb;
2517 	zgd->zgd_private = zd;
2518 
2519 	if (buf != NULL) {	/* immediate write */
2520 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2521 		    object, offset, size, ZTRL_READER);
2522 
2523 		error = dmu_read(os, object, offset, size, buf,
2524 		    DMU_READ_NO_PREFETCH);
2525 		ASSERT0(error);
2526 	} else {
2527 		ASSERT3P(zio, !=, NULL);
2528 		size = doi.doi_data_block_size;
2529 		if (ISP2(size)) {
2530 			offset = P2ALIGN_TYPED(offset, size, uint64_t);
2531 		} else {
2532 			ASSERT3U(offset, <, size);
2533 			offset = 0;
2534 		}
2535 
2536 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2537 		    object, offset, size, ZTRL_READER);
2538 
2539 		error = dmu_buf_hold_noread(os, object, offset, zgd, &db);
2540 
2541 		if (error == 0) {
2542 			blkptr_t *bp = &lr->lr_blkptr;
2543 
2544 			zgd->zgd_db = db;
2545 			zgd->zgd_bp = bp;
2546 
2547 			ASSERT3U(db->db_offset, ==, offset);
2548 			ASSERT3U(db->db_size, ==, size);
2549 
2550 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
2551 			    ztest_get_done, zgd);
2552 
2553 			if (error == 0)
2554 				return (0);
2555 		}
2556 	}
2557 
2558 	ztest_get_done(zgd, error);
2559 
2560 	return (error);
2561 }
2562 
2563 static void *
2564 ztest_lr_alloc(size_t lrsize, char *name)
2565 {
2566 	char *lr;
2567 	size_t namesize = name ? strlen(name) + 1 : 0;
2568 
2569 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2570 
2571 	if (name)
2572 		memcpy(lr + lrsize, name, namesize);
2573 
2574 	return (lr);
2575 }
2576 
2577 static void
2578 ztest_lr_free(void *lr, size_t lrsize, char *name)
2579 {
2580 	size_t namesize = name ? strlen(name) + 1 : 0;
2581 
2582 	umem_free(lr, lrsize + namesize);
2583 }
2584 
2585 /*
2586  * Lookup a bunch of objects.  Returns the number of objects not found.
2587  */
2588 static int
2589 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2590 {
2591 	int missing = 0;
2592 	int error;
2593 	int i;
2594 
2595 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2596 
2597 	for (i = 0; i < count; i++, od++) {
2598 		od->od_object = 0;
2599 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2600 		    sizeof (uint64_t), 1, &od->od_object);
2601 		if (error) {
2602 			ASSERT3S(error, ==, ENOENT);
2603 			ASSERT0(od->od_object);
2604 			missing++;
2605 		} else {
2606 			dmu_buf_t *db;
2607 			ztest_block_tag_t *bbt;
2608 			dmu_object_info_t doi;
2609 
2610 			ASSERT3U(od->od_object, !=, 0);
2611 			ASSERT0(missing);	/* there should be no gaps */
2612 
2613 			ztest_object_lock(zd, od->od_object, ZTRL_READER);
2614 			VERIFY0(dmu_bonus_hold(zd->zd_os, od->od_object,
2615 			    FTAG, &db));
2616 			dmu_object_info_from_db(db, &doi);
2617 			bbt = ztest_bt_bonus(db);
2618 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2619 			od->od_type = doi.doi_type;
2620 			od->od_blocksize = doi.doi_data_block_size;
2621 			od->od_gen = bbt->bt_gen;
2622 			dmu_buf_rele(db, FTAG);
2623 			ztest_object_unlock(zd, od->od_object);
2624 		}
2625 	}
2626 
2627 	return (missing);
2628 }
2629 
2630 static int
2631 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2632 {
2633 	int missing = 0;
2634 	int i;
2635 
2636 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2637 
2638 	for (i = 0; i < count; i++, od++) {
2639 		if (missing) {
2640 			od->od_object = 0;
2641 			missing++;
2642 			continue;
2643 		}
2644 
2645 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2646 
2647 		lr->lr_doid = od->od_dir;
2648 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2649 		lr->lrz_type = od->od_crtype;
2650 		lr->lrz_blocksize = od->od_crblocksize;
2651 		lr->lrz_ibshift = ztest_random_ibshift();
2652 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2653 		lr->lrz_dnodesize = od->od_crdnodesize;
2654 		lr->lr_gen = od->od_crgen;
2655 		lr->lr_crtime[0] = time(NULL);
2656 
2657 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2658 			ASSERT0(missing);
2659 			od->od_object = 0;
2660 			missing++;
2661 		} else {
2662 			od->od_object = lr->lr_foid;
2663 			od->od_type = od->od_crtype;
2664 			od->od_blocksize = od->od_crblocksize;
2665 			od->od_gen = od->od_crgen;
2666 			ASSERT3U(od->od_object, !=, 0);
2667 		}
2668 
2669 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2670 	}
2671 
2672 	return (missing);
2673 }
2674 
2675 static int
2676 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2677 {
2678 	int missing = 0;
2679 	int error;
2680 	int i;
2681 
2682 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2683 
2684 	od += count - 1;
2685 
2686 	for (i = count - 1; i >= 0; i--, od--) {
2687 		if (missing) {
2688 			missing++;
2689 			continue;
2690 		}
2691 
2692 		/*
2693 		 * No object was found.
2694 		 */
2695 		if (od->od_object == 0)
2696 			continue;
2697 
2698 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2699 
2700 		lr->lr_doid = od->od_dir;
2701 
2702 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2703 			ASSERT3U(error, ==, ENOSPC);
2704 			missing++;
2705 		} else {
2706 			od->od_object = 0;
2707 		}
2708 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2709 	}
2710 
2711 	return (missing);
2712 }
2713 
2714 static int
2715 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2716     const void *data)
2717 {
2718 	lr_write_t *lr;
2719 	int error;
2720 
2721 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2722 
2723 	lr->lr_foid = object;
2724 	lr->lr_offset = offset;
2725 	lr->lr_length = size;
2726 	lr->lr_blkoff = 0;
2727 	BP_ZERO(&lr->lr_blkptr);
2728 
2729 	memcpy(lr + 1, data, size);
2730 
2731 	error = ztest_replay_write(zd, lr, B_FALSE);
2732 
2733 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2734 
2735 	return (error);
2736 }
2737 
2738 static int
2739 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2740 {
2741 	lr_truncate_t *lr;
2742 	int error;
2743 
2744 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2745 
2746 	lr->lr_foid = object;
2747 	lr->lr_offset = offset;
2748 	lr->lr_length = size;
2749 
2750 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2751 
2752 	ztest_lr_free(lr, sizeof (*lr), NULL);
2753 
2754 	return (error);
2755 }
2756 
2757 static int
2758 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2759 {
2760 	lr_setattr_t *lr;
2761 	int error;
2762 
2763 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2764 
2765 	lr->lr_foid = object;
2766 	lr->lr_size = 0;
2767 	lr->lr_mode = 0;
2768 
2769 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2770 
2771 	ztest_lr_free(lr, sizeof (*lr), NULL);
2772 
2773 	return (error);
2774 }
2775 
2776 static void
2777 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2778 {
2779 	objset_t *os = zd->zd_os;
2780 	dmu_tx_t *tx;
2781 	uint64_t txg;
2782 	rl_t *rl;
2783 
2784 	txg_wait_synced(dmu_objset_pool(os), 0);
2785 
2786 	ztest_object_lock(zd, object, ZTRL_READER);
2787 	rl = ztest_range_lock(zd, object, offset, size, ZTRL_WRITER);
2788 
2789 	tx = dmu_tx_create(os);
2790 
2791 	dmu_tx_hold_write(tx, object, offset, size);
2792 
2793 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2794 
2795 	if (txg != 0) {
2796 		dmu_prealloc(os, object, offset, size, tx);
2797 		dmu_tx_commit(tx);
2798 		txg_wait_synced(dmu_objset_pool(os), txg);
2799 	} else {
2800 		(void) dmu_free_long_range(os, object, offset, size);
2801 	}
2802 
2803 	ztest_range_unlock(rl);
2804 	ztest_object_unlock(zd, object);
2805 }
2806 
2807 static void
2808 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2809 {
2810 	int err;
2811 	ztest_block_tag_t wbt;
2812 	dmu_object_info_t doi;
2813 	enum ztest_io_type io_type;
2814 	uint64_t blocksize;
2815 	void *data;
2816 
2817 	VERIFY0(dmu_object_info(zd->zd_os, object, &doi));
2818 	blocksize = doi.doi_data_block_size;
2819 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2820 
2821 	/*
2822 	 * Pick an i/o type at random, biased toward writing block tags.
2823 	 */
2824 	io_type = ztest_random(ZTEST_IO_TYPES);
2825 	if (ztest_random(2) == 0)
2826 		io_type = ZTEST_IO_WRITE_TAG;
2827 
2828 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2829 
2830 	switch (io_type) {
2831 
2832 	case ZTEST_IO_WRITE_TAG:
2833 		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2834 		    offset, 0, 0, 0);
2835 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2836 		break;
2837 
2838 	case ZTEST_IO_WRITE_PATTERN:
2839 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2840 		if (ztest_random(2) == 0) {
2841 			/*
2842 			 * Induce fletcher2 collisions to ensure that
2843 			 * zio_ddt_collision() detects and resolves them
2844 			 * when using fletcher2-verify for deduplication.
2845 			 */
2846 			((uint64_t *)data)[0] ^= 1ULL << 63;
2847 			((uint64_t *)data)[4] ^= 1ULL << 63;
2848 		}
2849 		(void) ztest_write(zd, object, offset, blocksize, data);
2850 		break;
2851 
2852 	case ZTEST_IO_WRITE_ZEROES:
2853 		memset(data, 0, blocksize);
2854 		(void) ztest_write(zd, object, offset, blocksize, data);
2855 		break;
2856 
2857 	case ZTEST_IO_TRUNCATE:
2858 		(void) ztest_truncate(zd, object, offset, blocksize);
2859 		break;
2860 
2861 	case ZTEST_IO_SETATTR:
2862 		(void) ztest_setattr(zd, object);
2863 		break;
2864 	default:
2865 		break;
2866 
2867 	case ZTEST_IO_REWRITE:
2868 		(void) pthread_rwlock_rdlock(&ztest_name_lock);
2869 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2870 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2871 		    B_FALSE);
2872 		ASSERT(err == 0 || err == ENOSPC);
2873 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2874 		    ZFS_PROP_COMPRESSION,
2875 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2876 		    B_FALSE);
2877 		ASSERT(err == 0 || err == ENOSPC);
2878 		(void) pthread_rwlock_unlock(&ztest_name_lock);
2879 
2880 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2881 		    DMU_READ_NO_PREFETCH));
2882 
2883 		(void) ztest_write(zd, object, offset, blocksize, data);
2884 		break;
2885 	}
2886 
2887 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2888 
2889 	umem_free(data, blocksize);
2890 }
2891 
2892 /*
2893  * Initialize an object description template.
2894  */
2895 static void
2896 ztest_od_init(ztest_od_t *od, uint64_t id, const char *tag, uint64_t index,
2897     dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2898     uint64_t gen)
2899 {
2900 	od->od_dir = ZTEST_DIROBJ;
2901 	od->od_object = 0;
2902 
2903 	od->od_crtype = type;
2904 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2905 	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2906 	od->od_crgen = gen;
2907 
2908 	od->od_type = DMU_OT_NONE;
2909 	od->od_blocksize = 0;
2910 	od->od_gen = 0;
2911 
2912 	(void) snprintf(od->od_name, sizeof (od->od_name),
2913 	    "%s(%"PRId64")[%"PRIu64"]",
2914 	    tag, id, index);
2915 }
2916 
2917 /*
2918  * Lookup or create the objects for a test using the od template.
2919  * If the objects do not all exist, or if 'remove' is specified,
2920  * remove any existing objects and create new ones.  Otherwise,
2921  * use the existing objects.
2922  */
2923 static int
2924 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2925 {
2926 	int count = size / sizeof (*od);
2927 	int rv = 0;
2928 
2929 	mutex_enter(&zd->zd_dirobj_lock);
2930 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2931 	    (ztest_remove(zd, od, count) != 0 ||
2932 	    ztest_create(zd, od, count) != 0))
2933 		rv = -1;
2934 	zd->zd_od = od;
2935 	mutex_exit(&zd->zd_dirobj_lock);
2936 
2937 	return (rv);
2938 }
2939 
2940 void
2941 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2942 {
2943 	(void) id;
2944 	zilog_t *zilog = zd->zd_zilog;
2945 
2946 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2947 
2948 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2949 
2950 	/*
2951 	 * Remember the committed values in zd, which is in parent/child
2952 	 * shared memory.  If we die, the next iteration of ztest_run()
2953 	 * will verify that the log really does contain this record.
2954 	 */
2955 	mutex_enter(&zilog->zl_lock);
2956 	ASSERT3P(zd->zd_shared, !=, NULL);
2957 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2958 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2959 	mutex_exit(&zilog->zl_lock);
2960 
2961 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2962 }
2963 
2964 /*
2965  * This function is designed to simulate the operations that occur during a
2966  * mount/unmount operation.  We hold the dataset across these operations in an
2967  * attempt to expose any implicit assumptions about ZIL management.
2968  */
2969 void
2970 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2971 {
2972 	(void) id;
2973 	objset_t *os = zd->zd_os;
2974 
2975 	/*
2976 	 * We hold the ztest_vdev_lock so we don't cause problems with
2977 	 * other threads that wish to remove a log device, such as
2978 	 * ztest_device_removal().
2979 	 */
2980 	mutex_enter(&ztest_vdev_lock);
2981 
2982 	/*
2983 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2984 	 * updating the zil (i.e. adding in-memory log records) and the
2985 	 * zd_zilog_lock to block any I/O.
2986 	 */
2987 	mutex_enter(&zd->zd_dirobj_lock);
2988 	(void) pthread_rwlock_wrlock(&zd->zd_zilog_lock);
2989 
2990 	/* zfsvfs_teardown() */
2991 	zil_close(zd->zd_zilog);
2992 
2993 	/* zfsvfs_setup() */
2994 	VERIFY3P(zil_open(os, ztest_get_data, NULL), ==, zd->zd_zilog);
2995 	zil_replay(os, zd, ztest_replay_vector);
2996 
2997 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2998 	mutex_exit(&zd->zd_dirobj_lock);
2999 	mutex_exit(&ztest_vdev_lock);
3000 }
3001 
3002 /*
3003  * Verify that we can't destroy an active pool, create an existing pool,
3004  * or create a pool with a bad vdev spec.
3005  */
3006 void
3007 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
3008 {
3009 	(void) zd, (void) id;
3010 	ztest_shared_opts_t *zo = &ztest_opts;
3011 	spa_t *spa;
3012 	nvlist_t *nvroot;
3013 
3014 	if (zo->zo_mmp_test)
3015 		return;
3016 
3017 	/*
3018 	 * Attempt to create using a bad file.
3019 	 */
3020 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
3021 	VERIFY3U(ENOENT, ==,
3022 	    spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
3023 	fnvlist_free(nvroot);
3024 
3025 	/*
3026 	 * Attempt to create using a bad mirror.
3027 	 */
3028 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 2, 1);
3029 	VERIFY3U(ENOENT, ==,
3030 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
3031 	fnvlist_free(nvroot);
3032 
3033 	/*
3034 	 * Attempt to create an existing pool.  It shouldn't matter
3035 	 * what's in the nvroot; we should fail with EEXIST.
3036 	 */
3037 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
3038 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
3039 	VERIFY3U(EEXIST, ==,
3040 	    spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
3041 	fnvlist_free(nvroot);
3042 
3043 	/*
3044 	 * We open a reference to the spa and then we try to export it
3045 	 * expecting one of the following errors:
3046 	 *
3047 	 * EBUSY
3048 	 *	Because of the reference we just opened.
3049 	 *
3050 	 * ZFS_ERR_EXPORT_IN_PROGRESS
3051 	 *	For the case that there is another ztest thread doing
3052 	 *	an export concurrently.
3053 	 */
3054 	VERIFY0(spa_open(zo->zo_pool, &spa, FTAG));
3055 	int error = spa_destroy(zo->zo_pool);
3056 	if (error != EBUSY && error != ZFS_ERR_EXPORT_IN_PROGRESS) {
3057 		fatal(B_FALSE, "spa_destroy(%s) returned unexpected value %d",
3058 		    spa->spa_name, error);
3059 	}
3060 	spa_close(spa, FTAG);
3061 
3062 	(void) pthread_rwlock_unlock(&ztest_name_lock);
3063 }
3064 
3065 /*
3066  * Start and then stop the MMP threads to ensure the startup and shutdown code
3067  * works properly.  Actual protection and property-related code tested via ZTS.
3068  */
3069 void
3070 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
3071 {
3072 	(void) zd, (void) id;
3073 	ztest_shared_opts_t *zo = &ztest_opts;
3074 	spa_t *spa = ztest_spa;
3075 
3076 	if (zo->zo_mmp_test)
3077 		return;
3078 
3079 	/*
3080 	 * Since enabling MMP involves setting a property, it could not be done
3081 	 * while the pool is suspended.
3082 	 */
3083 	if (spa_suspended(spa))
3084 		return;
3085 
3086 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3087 	mutex_enter(&spa->spa_props_lock);
3088 
3089 	zfs_multihost_fail_intervals = 0;
3090 
3091 	if (!spa_multihost(spa)) {
3092 		spa->spa_multihost = B_TRUE;
3093 		mmp_thread_start(spa);
3094 	}
3095 
3096 	mutex_exit(&spa->spa_props_lock);
3097 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3098 
3099 	txg_wait_synced(spa_get_dsl(spa), 0);
3100 	mmp_signal_all_threads();
3101 	txg_wait_synced(spa_get_dsl(spa), 0);
3102 
3103 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3104 	mutex_enter(&spa->spa_props_lock);
3105 
3106 	if (spa_multihost(spa)) {
3107 		mmp_thread_stop(spa);
3108 		spa->spa_multihost = B_FALSE;
3109 	}
3110 
3111 	mutex_exit(&spa->spa_props_lock);
3112 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3113 }
3114 
3115 static int
3116 ztest_get_raidz_children(spa_t *spa)
3117 {
3118 	(void) spa;
3119 	vdev_t *raidvd;
3120 
3121 	ASSERT(MUTEX_HELD(&ztest_vdev_lock));
3122 
3123 	if (ztest_opts.zo_raid_do_expand) {
3124 		raidvd = ztest_spa->spa_root_vdev->vdev_child[0];
3125 
3126 		ASSERT(raidvd->vdev_ops == &vdev_raidz_ops);
3127 
3128 		return (raidvd->vdev_children);
3129 	}
3130 
3131 	return (ztest_opts.zo_raid_children);
3132 }
3133 
3134 void
3135 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
3136 {
3137 	(void) zd, (void) id;
3138 	spa_t *spa;
3139 	uint64_t initial_version = SPA_VERSION_INITIAL;
3140 	uint64_t raidz_children, version, newversion;
3141 	nvlist_t *nvroot, *props;
3142 	char *name;
3143 
3144 	if (ztest_opts.zo_mmp_test)
3145 		return;
3146 
3147 	/* dRAID added after feature flags, skip upgrade test. */
3148 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0)
3149 		return;
3150 
3151 	mutex_enter(&ztest_vdev_lock);
3152 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
3153 
3154 	/*
3155 	 * Clean up from previous runs.
3156 	 */
3157 	(void) spa_destroy(name);
3158 
3159 	raidz_children = ztest_get_raidz_children(ztest_spa);
3160 
3161 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
3162 	    NULL, raidz_children, ztest_opts.zo_mirrors, 1);
3163 
3164 	/*
3165 	 * If we're configuring a RAIDZ device then make sure that the
3166 	 * initial version is capable of supporting that feature.
3167 	 */
3168 	switch (ztest_opts.zo_raid_parity) {
3169 	case 0:
3170 	case 1:
3171 		initial_version = SPA_VERSION_INITIAL;
3172 		break;
3173 	case 2:
3174 		initial_version = SPA_VERSION_RAIDZ2;
3175 		break;
3176 	case 3:
3177 		initial_version = SPA_VERSION_RAIDZ3;
3178 		break;
3179 	}
3180 
3181 	/*
3182 	 * Create a pool with a spa version that can be upgraded. Pick
3183 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
3184 	 */
3185 	do {
3186 		version = ztest_random_spa_version(initial_version);
3187 	} while (version > SPA_VERSION_BEFORE_FEATURES);
3188 
3189 	props = fnvlist_alloc();
3190 	fnvlist_add_uint64(props,
3191 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
3192 	VERIFY0(spa_create(name, nvroot, props, NULL, NULL));
3193 	fnvlist_free(nvroot);
3194 	fnvlist_free(props);
3195 
3196 	VERIFY0(spa_open(name, &spa, FTAG));
3197 	VERIFY3U(spa_version(spa), ==, version);
3198 	newversion = ztest_random_spa_version(version + 1);
3199 
3200 	if (ztest_opts.zo_verbose >= 4) {
3201 		(void) printf("upgrading spa version from "
3202 		    "%"PRIu64" to %"PRIu64"\n",
3203 		    version, newversion);
3204 	}
3205 
3206 	spa_upgrade(spa, newversion);
3207 	VERIFY3U(spa_version(spa), >, version);
3208 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
3209 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
3210 	spa_close(spa, FTAG);
3211 
3212 	kmem_strfree(name);
3213 	mutex_exit(&ztest_vdev_lock);
3214 }
3215 
3216 static void
3217 ztest_spa_checkpoint(spa_t *spa)
3218 {
3219 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3220 
3221 	int error = spa_checkpoint(spa->spa_name);
3222 
3223 	switch (error) {
3224 	case 0:
3225 	case ZFS_ERR_DEVRM_IN_PROGRESS:
3226 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3227 	case ZFS_ERR_CHECKPOINT_EXISTS:
3228 	case ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS:
3229 		break;
3230 	case ENOSPC:
3231 		ztest_record_enospc(FTAG);
3232 		break;
3233 	default:
3234 		fatal(B_FALSE, "spa_checkpoint(%s) = %d", spa->spa_name, error);
3235 	}
3236 }
3237 
3238 static void
3239 ztest_spa_discard_checkpoint(spa_t *spa)
3240 {
3241 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3242 
3243 	int error = spa_checkpoint_discard(spa->spa_name);
3244 
3245 	switch (error) {
3246 	case 0:
3247 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3248 	case ZFS_ERR_NO_CHECKPOINT:
3249 		break;
3250 	default:
3251 		fatal(B_FALSE, "spa_discard_checkpoint(%s) = %d",
3252 		    spa->spa_name, error);
3253 	}
3254 
3255 }
3256 
3257 void
3258 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
3259 {
3260 	(void) zd, (void) id;
3261 	spa_t *spa = ztest_spa;
3262 
3263 	mutex_enter(&ztest_checkpoint_lock);
3264 	if (ztest_random(2) == 0) {
3265 		ztest_spa_checkpoint(spa);
3266 	} else {
3267 		ztest_spa_discard_checkpoint(spa);
3268 	}
3269 	mutex_exit(&ztest_checkpoint_lock);
3270 }
3271 
3272 
3273 static vdev_t *
3274 vdev_lookup_by_path(vdev_t *vd, const char *path)
3275 {
3276 	vdev_t *mvd;
3277 	int c;
3278 
3279 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
3280 		return (vd);
3281 
3282 	for (c = 0; c < vd->vdev_children; c++)
3283 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
3284 		    NULL)
3285 			return (mvd);
3286 
3287 	return (NULL);
3288 }
3289 
3290 static int
3291 spa_num_top_vdevs(spa_t *spa)
3292 {
3293 	vdev_t *rvd = spa->spa_root_vdev;
3294 	ASSERT3U(spa_config_held(spa, SCL_VDEV, RW_READER), ==, SCL_VDEV);
3295 	return (rvd->vdev_children);
3296 }
3297 
3298 /*
3299  * Verify that vdev_add() works as expected.
3300  */
3301 void
3302 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
3303 {
3304 	(void) zd, (void) id;
3305 	ztest_shared_t *zs = ztest_shared;
3306 	spa_t *spa = ztest_spa;
3307 	uint64_t leaves;
3308 	uint64_t guid;
3309 	uint64_t raidz_children;
3310 
3311 	nvlist_t *nvroot;
3312 	int error;
3313 
3314 	if (ztest_opts.zo_mmp_test)
3315 		return;
3316 
3317 	mutex_enter(&ztest_vdev_lock);
3318 	raidz_children = ztest_get_raidz_children(spa);
3319 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * raidz_children;
3320 
3321 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3322 
3323 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3324 
3325 	/*
3326 	 * If we have slogs then remove them 1/4 of the time.
3327 	 */
3328 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
3329 		metaslab_group_t *mg;
3330 
3331 		/*
3332 		 * find the first real slog in log allocation class
3333 		 */
3334 		mg =  spa_log_class(spa)->mc_allocator[0].mca_rotor;
3335 		while (!mg->mg_vd->vdev_islog)
3336 			mg = mg->mg_next;
3337 
3338 		guid = mg->mg_vd->vdev_guid;
3339 
3340 		spa_config_exit(spa, SCL_VDEV, FTAG);
3341 
3342 		/*
3343 		 * We have to grab the zs_name_lock as writer to
3344 		 * prevent a race between removing a slog (dmu_objset_find)
3345 		 * and destroying a dataset. Removing the slog will
3346 		 * grab a reference on the dataset which may cause
3347 		 * dsl_destroy_head() to fail with EBUSY thus
3348 		 * leaving the dataset in an inconsistent state.
3349 		 */
3350 		pthread_rwlock_wrlock(&ztest_name_lock);
3351 		error = spa_vdev_remove(spa, guid, B_FALSE);
3352 		pthread_rwlock_unlock(&ztest_name_lock);
3353 
3354 		switch (error) {
3355 		case 0:
3356 		case EEXIST:	/* Generic zil_reset() error */
3357 		case EBUSY:	/* Replay required */
3358 		case EACCES:	/* Crypto key not loaded */
3359 		case ZFS_ERR_CHECKPOINT_EXISTS:
3360 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3361 			break;
3362 		default:
3363 			fatal(B_FALSE, "spa_vdev_remove() = %d", error);
3364 		}
3365 	} else {
3366 		spa_config_exit(spa, SCL_VDEV, FTAG);
3367 
3368 		/*
3369 		 * Make 1/4 of the devices be log devices
3370 		 */
3371 		nvroot = make_vdev_root(NULL, NULL, NULL,
3372 		    ztest_opts.zo_vdev_size, 0, (ztest_random(4) == 0) ?
3373 		    "log" : NULL, raidz_children, zs->zs_mirrors,
3374 		    1);
3375 
3376 		error = spa_vdev_add(spa, nvroot, B_FALSE);
3377 		fnvlist_free(nvroot);
3378 
3379 		switch (error) {
3380 		case 0:
3381 			break;
3382 		case ENOSPC:
3383 			ztest_record_enospc("spa_vdev_add");
3384 			break;
3385 		default:
3386 			fatal(B_FALSE, "spa_vdev_add() = %d", error);
3387 		}
3388 	}
3389 
3390 	mutex_exit(&ztest_vdev_lock);
3391 }
3392 
3393 void
3394 ztest_vdev_class_add(ztest_ds_t *zd, uint64_t id)
3395 {
3396 	(void) zd, (void) id;
3397 	ztest_shared_t *zs = ztest_shared;
3398 	spa_t *spa = ztest_spa;
3399 	uint64_t leaves;
3400 	nvlist_t *nvroot;
3401 	uint64_t raidz_children;
3402 	const char *class = (ztest_random(2) == 0) ?
3403 	    VDEV_ALLOC_BIAS_SPECIAL : VDEV_ALLOC_BIAS_DEDUP;
3404 	int error;
3405 
3406 	/*
3407 	 * By default add a special vdev 50% of the time
3408 	 */
3409 	if ((ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_OFF) ||
3410 	    (ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_RND &&
3411 	    ztest_random(2) == 0)) {
3412 		return;
3413 	}
3414 
3415 	mutex_enter(&ztest_vdev_lock);
3416 
3417 	/* Only test with mirrors */
3418 	if (zs->zs_mirrors < 2) {
3419 		mutex_exit(&ztest_vdev_lock);
3420 		return;
3421 	}
3422 
3423 	/* requires feature@allocation_classes */
3424 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) {
3425 		mutex_exit(&ztest_vdev_lock);
3426 		return;
3427 	}
3428 
3429 	raidz_children = ztest_get_raidz_children(spa);
3430 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * raidz_children;
3431 
3432 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3433 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3434 	spa_config_exit(spa, SCL_VDEV, FTAG);
3435 
3436 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
3437 	    class, raidz_children, zs->zs_mirrors, 1);
3438 
3439 	error = spa_vdev_add(spa, nvroot, B_FALSE);
3440 	fnvlist_free(nvroot);
3441 
3442 	if (error == ENOSPC)
3443 		ztest_record_enospc("spa_vdev_add");
3444 	else if (error != 0)
3445 		fatal(B_FALSE, "spa_vdev_add() = %d", error);
3446 
3447 	/*
3448 	 * 50% of the time allow small blocks in the special class
3449 	 */
3450 	if (error == 0 &&
3451 	    spa_special_class(spa)->mc_groups == 1 && ztest_random(2) == 0) {
3452 		if (ztest_opts.zo_verbose >= 3)
3453 			(void) printf("Enabling special VDEV small blocks\n");
3454 		error = ztest_dsl_prop_set_uint64(zd->zd_name,
3455 		    ZFS_PROP_SPECIAL_SMALL_BLOCKS, 32768, B_FALSE);
3456 		ASSERT(error == 0 || error == ENOSPC);
3457 	}
3458 
3459 	mutex_exit(&ztest_vdev_lock);
3460 
3461 	if (ztest_opts.zo_verbose >= 3) {
3462 		metaslab_class_t *mc;
3463 
3464 		if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL) == 0)
3465 			mc = spa_special_class(spa);
3466 		else
3467 			mc = spa_dedup_class(spa);
3468 		(void) printf("Added a %s mirrored vdev (of %d)\n",
3469 		    class, (int)mc->mc_groups);
3470 	}
3471 }
3472 
3473 /*
3474  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
3475  */
3476 void
3477 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
3478 {
3479 	(void) zd, (void) id;
3480 	ztest_shared_t *zs = ztest_shared;
3481 	spa_t *spa = ztest_spa;
3482 	vdev_t *rvd = spa->spa_root_vdev;
3483 	spa_aux_vdev_t *sav;
3484 	const char *aux;
3485 	char *path;
3486 	uint64_t guid = 0;
3487 	int error, ignore_err = 0;
3488 
3489 	if (ztest_opts.zo_mmp_test)
3490 		return;
3491 
3492 	path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3493 
3494 	if (ztest_random(2) == 0) {
3495 		sav = &spa->spa_spares;
3496 		aux = ZPOOL_CONFIG_SPARES;
3497 	} else {
3498 		sav = &spa->spa_l2cache;
3499 		aux = ZPOOL_CONFIG_L2CACHE;
3500 	}
3501 
3502 	mutex_enter(&ztest_vdev_lock);
3503 
3504 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3505 
3506 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
3507 		/*
3508 		 * Pick a random device to remove.
3509 		 */
3510 		vdev_t *svd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3511 
3512 		/* dRAID spares cannot be removed; try anyways to see ENOTSUP */
3513 		if (strstr(svd->vdev_path, VDEV_TYPE_DRAID) != NULL)
3514 			ignore_err = ENOTSUP;
3515 
3516 		guid = svd->vdev_guid;
3517 	} else {
3518 		/*
3519 		 * Find an unused device we can add.
3520 		 */
3521 		zs->zs_vdev_aux = 0;
3522 		for (;;) {
3523 			int c;
3524 			(void) snprintf(path, MAXPATHLEN, ztest_aux_template,
3525 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
3526 			    zs->zs_vdev_aux);
3527 			for (c = 0; c < sav->sav_count; c++)
3528 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
3529 				    path) == 0)
3530 					break;
3531 			if (c == sav->sav_count &&
3532 			    vdev_lookup_by_path(rvd, path) == NULL)
3533 				break;
3534 			zs->zs_vdev_aux++;
3535 		}
3536 	}
3537 
3538 	spa_config_exit(spa, SCL_VDEV, FTAG);
3539 
3540 	if (guid == 0) {
3541 		/*
3542 		 * Add a new device.
3543 		 */
3544 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3545 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, NULL, 0, 0, 1);
3546 		error = spa_vdev_add(spa, nvroot, B_FALSE);
3547 
3548 		switch (error) {
3549 		case 0:
3550 			break;
3551 		default:
3552 			fatal(B_FALSE, "spa_vdev_add(%p) = %d", nvroot, error);
3553 		}
3554 		fnvlist_free(nvroot);
3555 	} else {
3556 		/*
3557 		 * Remove an existing device.  Sometimes, dirty its
3558 		 * vdev state first to make sure we handle removal
3559 		 * of devices that have pending state changes.
3560 		 */
3561 		if (ztest_random(2) == 0)
3562 			(void) vdev_online(spa, guid, 0, NULL);
3563 
3564 		error = spa_vdev_remove(spa, guid, B_FALSE);
3565 
3566 		switch (error) {
3567 		case 0:
3568 		case EBUSY:
3569 		case ZFS_ERR_CHECKPOINT_EXISTS:
3570 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3571 			break;
3572 		default:
3573 			if (error != ignore_err)
3574 				fatal(B_FALSE,
3575 				    "spa_vdev_remove(%"PRIu64") = %d",
3576 				    guid, error);
3577 		}
3578 	}
3579 
3580 	mutex_exit(&ztest_vdev_lock);
3581 
3582 	umem_free(path, MAXPATHLEN);
3583 }
3584 
3585 /*
3586  * split a pool if it has mirror tlvdevs
3587  */
3588 void
3589 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3590 {
3591 	(void) zd, (void) id;
3592 	ztest_shared_t *zs = ztest_shared;
3593 	spa_t *spa = ztest_spa;
3594 	vdev_t *rvd = spa->spa_root_vdev;
3595 	nvlist_t *tree, **child, *config, *split, **schild;
3596 	uint_t c, children, schildren = 0, lastlogid = 0;
3597 	int error = 0;
3598 
3599 	if (ztest_opts.zo_mmp_test)
3600 		return;
3601 
3602 	mutex_enter(&ztest_vdev_lock);
3603 
3604 	/* ensure we have a usable config; mirrors of raidz aren't supported */
3605 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raid_children > 1) {
3606 		mutex_exit(&ztest_vdev_lock);
3607 		return;
3608 	}
3609 
3610 	/* clean up the old pool, if any */
3611 	(void) spa_destroy("splitp");
3612 
3613 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3614 
3615 	/* generate a config from the existing config */
3616 	mutex_enter(&spa->spa_props_lock);
3617 	tree = fnvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE);
3618 	mutex_exit(&spa->spa_props_lock);
3619 
3620 	VERIFY0(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN,
3621 	    &child, &children));
3622 
3623 	schild = umem_alloc(rvd->vdev_children * sizeof (nvlist_t *),
3624 	    UMEM_NOFAIL);
3625 	for (c = 0; c < children; c++) {
3626 		vdev_t *tvd = rvd->vdev_child[c];
3627 		nvlist_t **mchild;
3628 		uint_t mchildren;
3629 
3630 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3631 			schild[schildren] = fnvlist_alloc();
3632 			fnvlist_add_string(schild[schildren],
3633 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE);
3634 			fnvlist_add_uint64(schild[schildren],
3635 			    ZPOOL_CONFIG_IS_HOLE, 1);
3636 			if (lastlogid == 0)
3637 				lastlogid = schildren;
3638 			++schildren;
3639 			continue;
3640 		}
3641 		lastlogid = 0;
3642 		VERIFY0(nvlist_lookup_nvlist_array(child[c],
3643 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren));
3644 		schild[schildren++] = fnvlist_dup(mchild[0]);
3645 	}
3646 
3647 	/* OK, create a config that can be used to split */
3648 	split = fnvlist_alloc();
3649 	fnvlist_add_string(split, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
3650 	fnvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN,
3651 	    (const nvlist_t **)schild, lastlogid != 0 ? lastlogid : schildren);
3652 
3653 	config = fnvlist_alloc();
3654 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split);
3655 
3656 	for (c = 0; c < schildren; c++)
3657 		fnvlist_free(schild[c]);
3658 	umem_free(schild, rvd->vdev_children * sizeof (nvlist_t *));
3659 	fnvlist_free(split);
3660 
3661 	spa_config_exit(spa, SCL_VDEV, FTAG);
3662 
3663 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
3664 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3665 	(void) pthread_rwlock_unlock(&ztest_name_lock);
3666 
3667 	fnvlist_free(config);
3668 
3669 	if (error == 0) {
3670 		(void) printf("successful split - results:\n");
3671 		mutex_enter(&spa_namespace_lock);
3672 		show_pool_stats(spa);
3673 		show_pool_stats(spa_lookup("splitp"));
3674 		mutex_exit(&spa_namespace_lock);
3675 		++zs->zs_splits;
3676 		--zs->zs_mirrors;
3677 	}
3678 	mutex_exit(&ztest_vdev_lock);
3679 }
3680 
3681 /*
3682  * Verify that we can attach and detach devices.
3683  */
3684 void
3685 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3686 {
3687 	(void) zd, (void) id;
3688 	ztest_shared_t *zs = ztest_shared;
3689 	spa_t *spa = ztest_spa;
3690 	spa_aux_vdev_t *sav = &spa->spa_spares;
3691 	vdev_t *rvd = spa->spa_root_vdev;
3692 	vdev_t *oldvd, *newvd, *pvd;
3693 	nvlist_t *root;
3694 	uint64_t leaves;
3695 	uint64_t leaf, top;
3696 	uint64_t ashift = ztest_get_ashift();
3697 	uint64_t oldguid, pguid;
3698 	uint64_t oldsize, newsize;
3699 	uint64_t raidz_children;
3700 	char *oldpath, *newpath;
3701 	int replacing;
3702 	int oldvd_has_siblings = B_FALSE;
3703 	int newvd_is_spare = B_FALSE;
3704 	int newvd_is_dspare = B_FALSE;
3705 	int oldvd_is_log;
3706 	int oldvd_is_special;
3707 	int error, expected_error;
3708 
3709 	if (ztest_opts.zo_mmp_test)
3710 		return;
3711 
3712 	oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3713 	newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3714 
3715 	mutex_enter(&ztest_vdev_lock);
3716 	raidz_children = ztest_get_raidz_children(spa);
3717 	leaves = MAX(zs->zs_mirrors, 1) * raidz_children;
3718 
3719 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3720 
3721 	/*
3722 	 * If a vdev is in the process of being removed, its removal may
3723 	 * finish while we are in progress, leading to an unexpected error
3724 	 * value.  Don't bother trying to attach while we are in the middle
3725 	 * of removal.
3726 	 */
3727 	if (ztest_device_removal_active) {
3728 		spa_config_exit(spa, SCL_ALL, FTAG);
3729 		goto out;
3730 	}
3731 
3732 	/*
3733 	 * RAIDZ leaf VDEV mirrors are not currently supported while a
3734 	 * RAIDZ expansion is in progress.
3735 	 */
3736 	if (ztest_opts.zo_raid_do_expand) {
3737 		spa_config_exit(spa, SCL_ALL, FTAG);
3738 		goto out;
3739 	}
3740 
3741 	/*
3742 	 * Decide whether to do an attach or a replace.
3743 	 */
3744 	replacing = ztest_random(2);
3745 
3746 	/*
3747 	 * Pick a random top-level vdev.
3748 	 */
3749 	top = ztest_random_vdev_top(spa, B_TRUE);
3750 
3751 	/*
3752 	 * Pick a random leaf within it.
3753 	 */
3754 	leaf = ztest_random(leaves);
3755 
3756 	/*
3757 	 * Locate this vdev.
3758 	 */
3759 	oldvd = rvd->vdev_child[top];
3760 
3761 	/* pick a child from the mirror */
3762 	if (zs->zs_mirrors >= 1) {
3763 		ASSERT3P(oldvd->vdev_ops, ==, &vdev_mirror_ops);
3764 		ASSERT3U(oldvd->vdev_children, >=, zs->zs_mirrors);
3765 		oldvd = oldvd->vdev_child[leaf / raidz_children];
3766 	}
3767 
3768 	/* pick a child out of the raidz group */
3769 	if (ztest_opts.zo_raid_children > 1) {
3770 		if (strcmp(oldvd->vdev_ops->vdev_op_type, "raidz") == 0)
3771 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_raidz_ops);
3772 		else
3773 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_draid_ops);
3774 		oldvd = oldvd->vdev_child[leaf % raidz_children];
3775 	}
3776 
3777 	/*
3778 	 * If we're already doing an attach or replace, oldvd may be a
3779 	 * mirror vdev -- in which case, pick a random child.
3780 	 */
3781 	while (oldvd->vdev_children != 0) {
3782 		oldvd_has_siblings = B_TRUE;
3783 		ASSERT3U(oldvd->vdev_children, >=, 2);
3784 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3785 	}
3786 
3787 	oldguid = oldvd->vdev_guid;
3788 	oldsize = vdev_get_min_asize(oldvd);
3789 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
3790 	oldvd_is_special =
3791 	    oldvd->vdev_top->vdev_alloc_bias == VDEV_BIAS_SPECIAL ||
3792 	    oldvd->vdev_top->vdev_alloc_bias == VDEV_BIAS_DEDUP;
3793 	(void) strlcpy(oldpath, oldvd->vdev_path, MAXPATHLEN);
3794 	pvd = oldvd->vdev_parent;
3795 	pguid = pvd->vdev_guid;
3796 
3797 	/*
3798 	 * If oldvd has siblings, then half of the time, detach it.  Prior
3799 	 * to the detach the pool is scrubbed in order to prevent creating
3800 	 * unrepairable blocks as a result of the data corruption injection.
3801 	 */
3802 	if (oldvd_has_siblings && ztest_random(2) == 0) {
3803 		spa_config_exit(spa, SCL_ALL, FTAG);
3804 
3805 		error = ztest_scrub_impl(spa);
3806 		if (error)
3807 			goto out;
3808 
3809 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3810 		if (error != 0 && error != ENODEV && error != EBUSY &&
3811 		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3812 		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
3813 			fatal(B_FALSE, "detach (%s) returned %d",
3814 			    oldpath, error);
3815 		goto out;
3816 	}
3817 
3818 	/*
3819 	 * For the new vdev, choose with equal probability between the two
3820 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3821 	 */
3822 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
3823 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3824 		newvd_is_spare = B_TRUE;
3825 
3826 		if (newvd->vdev_ops == &vdev_draid_spare_ops)
3827 			newvd_is_dspare = B_TRUE;
3828 
3829 		(void) strlcpy(newpath, newvd->vdev_path, MAXPATHLEN);
3830 	} else {
3831 		(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3832 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
3833 		    top * leaves + leaf);
3834 		if (ztest_random(2) == 0)
3835 			newpath[strlen(newpath) - 1] = 'b';
3836 		newvd = vdev_lookup_by_path(rvd, newpath);
3837 	}
3838 
3839 	if (newvd) {
3840 		/*
3841 		 * Reopen to ensure the vdev's asize field isn't stale.
3842 		 */
3843 		vdev_reopen(newvd);
3844 		newsize = vdev_get_min_asize(newvd);
3845 	} else {
3846 		/*
3847 		 * Make newsize a little bigger or smaller than oldsize.
3848 		 * If it's smaller, the attach should fail.
3849 		 * If it's larger, and we're doing a replace,
3850 		 * we should get dynamic LUN growth when we're done.
3851 		 */
3852 		newsize = 10 * oldsize / (9 + ztest_random(3));
3853 	}
3854 
3855 	/*
3856 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3857 	 * unless it's a replace; in that case any non-replacing parent is OK.
3858 	 *
3859 	 * If newvd is already part of the pool, it should fail with EBUSY.
3860 	 *
3861 	 * If newvd is too small, it should fail with EOVERFLOW.
3862 	 *
3863 	 * If newvd is a distributed spare and it's being attached to a
3864 	 * dRAID which is not its parent it should fail with EINVAL.
3865 	 */
3866 	if (pvd->vdev_ops != &vdev_mirror_ops &&
3867 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3868 	    pvd->vdev_ops == &vdev_replacing_ops ||
3869 	    pvd->vdev_ops == &vdev_spare_ops))
3870 		expected_error = ENOTSUP;
3871 	else if (newvd_is_spare &&
3872 	    (!replacing || oldvd_is_log || oldvd_is_special))
3873 		expected_error = ENOTSUP;
3874 	else if (newvd == oldvd)
3875 		expected_error = replacing ? 0 : EBUSY;
3876 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3877 		expected_error = EBUSY;
3878 	else if (!newvd_is_dspare && newsize < oldsize)
3879 		expected_error = EOVERFLOW;
3880 	else if (ashift > oldvd->vdev_top->vdev_ashift)
3881 		expected_error = EDOM;
3882 	else if (newvd_is_dspare && pvd != vdev_draid_spare_get_parent(newvd))
3883 		expected_error = EINVAL;
3884 	else
3885 		expected_error = 0;
3886 
3887 	spa_config_exit(spa, SCL_ALL, FTAG);
3888 
3889 	/*
3890 	 * Build the nvlist describing newpath.
3891 	 */
3892 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3893 	    ashift, NULL, 0, 0, 1);
3894 
3895 	/*
3896 	 * When supported select either a healing or sequential resilver.
3897 	 */
3898 	boolean_t rebuilding = B_FALSE;
3899 	if (pvd->vdev_ops == &vdev_mirror_ops ||
3900 	    pvd->vdev_ops ==  &vdev_root_ops) {
3901 		rebuilding = !!ztest_random(2);
3902 	}
3903 
3904 	error = spa_vdev_attach(spa, oldguid, root, replacing, rebuilding);
3905 
3906 	fnvlist_free(root);
3907 
3908 	/*
3909 	 * If our parent was the replacing vdev, but the replace completed,
3910 	 * then instead of failing with ENOTSUP we may either succeed,
3911 	 * fail with ENODEV, or fail with EOVERFLOW.
3912 	 */
3913 	if (expected_error == ENOTSUP &&
3914 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3915 		expected_error = error;
3916 
3917 	/*
3918 	 * If someone grew the LUN, the replacement may be too small.
3919 	 */
3920 	if (error == EOVERFLOW || error == EBUSY)
3921 		expected_error = error;
3922 
3923 	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3924 	    error == ZFS_ERR_DISCARDING_CHECKPOINT ||
3925 	    error == ZFS_ERR_RESILVER_IN_PROGRESS ||
3926 	    error == ZFS_ERR_REBUILD_IN_PROGRESS)
3927 		expected_error = error;
3928 
3929 	if (error != expected_error && expected_error != EBUSY) {
3930 		fatal(B_FALSE, "attach (%s %"PRIu64", %s %"PRIu64", %d) "
3931 		    "returned %d, expected %d",
3932 		    oldpath, oldsize, newpath,
3933 		    newsize, replacing, error, expected_error);
3934 	}
3935 out:
3936 	mutex_exit(&ztest_vdev_lock);
3937 
3938 	umem_free(oldpath, MAXPATHLEN);
3939 	umem_free(newpath, MAXPATHLEN);
3940 }
3941 
3942 static void
3943 raidz_scratch_verify(void)
3944 {
3945 	spa_t *spa;
3946 	uint64_t write_size, logical_size, offset;
3947 	raidz_reflow_scratch_state_t state;
3948 	vdev_raidz_expand_t *vre;
3949 	vdev_t *raidvd;
3950 
3951 	ASSERT(raidz_expand_pause_point == RAIDZ_EXPAND_PAUSE_NONE);
3952 
3953 	if (ztest_scratch_state->zs_raidz_scratch_verify_pause == 0)
3954 		return;
3955 
3956 	kernel_init(SPA_MODE_READ);
3957 
3958 	mutex_enter(&spa_namespace_lock);
3959 	spa = spa_lookup(ztest_opts.zo_pool);
3960 	ASSERT(spa);
3961 	spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
3962 	mutex_exit(&spa_namespace_lock);
3963 
3964 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
3965 
3966 	ASSERT3U(RRSS_GET_OFFSET(&spa->spa_uberblock), !=, UINT64_MAX);
3967 
3968 	mutex_enter(&ztest_vdev_lock);
3969 
3970 	spa_config_enter(spa, SCL_ALL, FTAG, RW_READER);
3971 
3972 	vre = spa->spa_raidz_expand;
3973 	if (vre == NULL)
3974 		goto out;
3975 
3976 	raidvd = vdev_lookup_top(spa, vre->vre_vdev_id);
3977 	offset = RRSS_GET_OFFSET(&spa->spa_uberblock);
3978 	state = RRSS_GET_STATE(&spa->spa_uberblock);
3979 	write_size = P2ALIGN_TYPED(VDEV_BOOT_SIZE, 1 << raidvd->vdev_ashift,
3980 	    uint64_t);
3981 	logical_size = write_size * raidvd->vdev_children;
3982 
3983 	switch (state) {
3984 		/*
3985 		 * Initial state of reflow process.  RAIDZ expansion was
3986 		 * requested by user, but scratch object was not created.
3987 		 */
3988 		case RRSS_SCRATCH_NOT_IN_USE:
3989 			ASSERT3U(offset, ==, 0);
3990 			break;
3991 
3992 		/*
3993 		 * Scratch object was synced and stored in boot area.
3994 		 */
3995 		case RRSS_SCRATCH_VALID:
3996 
3997 		/*
3998 		 * Scratch object was synced back to raidz start offset,
3999 		 * raidz is ready for sector by sector reflow process.
4000 		 */
4001 		case RRSS_SCRATCH_INVALID_SYNCED:
4002 
4003 		/*
4004 		 * Scratch object was synced back to raidz start offset
4005 		 * on zpool importing, raidz is ready for sector by sector
4006 		 * reflow process.
4007 		 */
4008 		case RRSS_SCRATCH_INVALID_SYNCED_ON_IMPORT:
4009 			ASSERT3U(offset, ==, logical_size);
4010 			break;
4011 
4012 		/*
4013 		 * Sector by sector reflow process started.
4014 		 */
4015 		case RRSS_SCRATCH_INVALID_SYNCED_REFLOW:
4016 			ASSERT3U(offset, >=, logical_size);
4017 			break;
4018 	}
4019 
4020 out:
4021 	spa_config_exit(spa, SCL_ALL, FTAG);
4022 
4023 	mutex_exit(&ztest_vdev_lock);
4024 
4025 	ztest_scratch_state->zs_raidz_scratch_verify_pause = 0;
4026 
4027 	spa_close(spa, FTAG);
4028 	kernel_fini();
4029 }
4030 
4031 static void
4032 ztest_scratch_thread(void *arg)
4033 {
4034 	(void) arg;
4035 
4036 	/* wait up to 10 seconds */
4037 	for (int t = 100; t > 0; t -= 1) {
4038 		if (raidz_expand_pause_point == RAIDZ_EXPAND_PAUSE_NONE)
4039 			thread_exit();
4040 
4041 		(void) poll(NULL, 0, 100);
4042 	}
4043 
4044 	/* killed when the scratch area progress reached a certain point */
4045 	ztest_kill(ztest_shared);
4046 }
4047 
4048 /*
4049  * Verify that we can attach raidz device.
4050  */
4051 void
4052 ztest_vdev_raidz_attach(ztest_ds_t *zd, uint64_t id)
4053 {
4054 	(void) zd, (void) id;
4055 	ztest_shared_t *zs = ztest_shared;
4056 	spa_t *spa = ztest_spa;
4057 	uint64_t leaves, raidz_children, newsize, ashift = ztest_get_ashift();
4058 	kthread_t *scratch_thread = NULL;
4059 	vdev_t *newvd, *pvd;
4060 	nvlist_t *root;
4061 	char *newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
4062 	int error, expected_error = 0;
4063 
4064 	mutex_enter(&ztest_vdev_lock);
4065 
4066 	spa_config_enter(spa, SCL_ALL, FTAG, RW_READER);
4067 
4068 	/* Only allow attach when raid-kind = 'eraidz' */
4069 	if (!ztest_opts.zo_raid_do_expand) {
4070 		spa_config_exit(spa, SCL_ALL, FTAG);
4071 		goto out;
4072 	}
4073 
4074 	if (ztest_opts.zo_mmp_test) {
4075 		spa_config_exit(spa, SCL_ALL, FTAG);
4076 		goto out;
4077 	}
4078 
4079 	if (ztest_device_removal_active) {
4080 		spa_config_exit(spa, SCL_ALL, FTAG);
4081 		goto out;
4082 	}
4083 
4084 	pvd = vdev_lookup_top(spa, 0);
4085 
4086 	ASSERT(pvd->vdev_ops == &vdev_raidz_ops);
4087 
4088 	/*
4089 	 * Get size of a child of the raidz group,
4090 	 * make sure device is a bit bigger
4091 	 */
4092 	newvd = pvd->vdev_child[ztest_random(pvd->vdev_children)];
4093 	newsize = 10 * vdev_get_min_asize(newvd) / (9 + ztest_random(2));
4094 
4095 	/*
4096 	 * Get next attached leaf id
4097 	 */
4098 	raidz_children = ztest_get_raidz_children(spa);
4099 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * raidz_children;
4100 	zs->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
4101 
4102 	if (spa->spa_raidz_expand)
4103 		expected_error = ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS;
4104 
4105 	spa_config_exit(spa, SCL_ALL, FTAG);
4106 
4107 	/*
4108 	 * Path to vdev to be attached
4109 	 */
4110 	(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
4111 	    ztest_opts.zo_dir, ztest_opts.zo_pool, zs->zs_vdev_next_leaf);
4112 
4113 	/*
4114 	 * Build the nvlist describing newpath.
4115 	 */
4116 	root = make_vdev_root(newpath, NULL, NULL, newsize, ashift, NULL,
4117 	    0, 0, 1);
4118 
4119 	/*
4120 	 * 50% of the time, set raidz_expand_pause_point to cause
4121 	 * raidz_reflow_scratch_sync() to pause at a certain point and
4122 	 * then kill the test after 10 seconds so raidz_scratch_verify()
4123 	 * can confirm consistency when the pool is imported.
4124 	 */
4125 	if (ztest_random(2) == 0 && expected_error == 0) {
4126 		raidz_expand_pause_point =
4127 		    ztest_random(RAIDZ_EXPAND_PAUSE_SCRATCH_POST_REFLOW_2) + 1;
4128 		scratch_thread = thread_create(NULL, 0, ztest_scratch_thread,
4129 		    ztest_shared, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
4130 	}
4131 
4132 	error = spa_vdev_attach(spa, pvd->vdev_guid, root, B_FALSE, B_FALSE);
4133 
4134 	nvlist_free(root);
4135 
4136 	if (error == EOVERFLOW || error == ENXIO ||
4137 	    error == ZFS_ERR_CHECKPOINT_EXISTS ||
4138 	    error == ZFS_ERR_DISCARDING_CHECKPOINT)
4139 		expected_error = error;
4140 
4141 	if (error != 0 && error != expected_error) {
4142 		fatal(0, "raidz attach (%s %"PRIu64") returned %d, expected %d",
4143 		    newpath, newsize, error, expected_error);
4144 	}
4145 
4146 	if (raidz_expand_pause_point) {
4147 		if (error != 0) {
4148 			/*
4149 			 * Do not verify scratch object in case of error
4150 			 * returned by vdev attaching.
4151 			 */
4152 			raidz_expand_pause_point = RAIDZ_EXPAND_PAUSE_NONE;
4153 		}
4154 
4155 		VERIFY0(thread_join(scratch_thread));
4156 	}
4157 out:
4158 	mutex_exit(&ztest_vdev_lock);
4159 
4160 	umem_free(newpath, MAXPATHLEN);
4161 }
4162 
4163 void
4164 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
4165 {
4166 	(void) zd, (void) id;
4167 	spa_t *spa = ztest_spa;
4168 	vdev_t *vd;
4169 	uint64_t guid;
4170 	int error;
4171 
4172 	mutex_enter(&ztest_vdev_lock);
4173 
4174 	if (ztest_device_removal_active) {
4175 		mutex_exit(&ztest_vdev_lock);
4176 		return;
4177 	}
4178 
4179 	/*
4180 	 * Remove a random top-level vdev and wait for removal to finish.
4181 	 */
4182 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
4183 	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
4184 	guid = vd->vdev_guid;
4185 	spa_config_exit(spa, SCL_VDEV, FTAG);
4186 
4187 	error = spa_vdev_remove(spa, guid, B_FALSE);
4188 	if (error == 0) {
4189 		ztest_device_removal_active = B_TRUE;
4190 		mutex_exit(&ztest_vdev_lock);
4191 
4192 		/*
4193 		 * spa->spa_vdev_removal is created in a sync task that
4194 		 * is initiated via dsl_sync_task_nowait(). Since the
4195 		 * task may not run before spa_vdev_remove() returns, we
4196 		 * must wait at least 1 txg to ensure that the removal
4197 		 * struct has been created.
4198 		 */
4199 		txg_wait_synced(spa_get_dsl(spa), 0);
4200 
4201 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
4202 			txg_wait_synced(spa_get_dsl(spa), 0);
4203 	} else {
4204 		mutex_exit(&ztest_vdev_lock);
4205 		return;
4206 	}
4207 
4208 	/*
4209 	 * The pool needs to be scrubbed after completing device removal.
4210 	 * Failure to do so may result in checksum errors due to the
4211 	 * strategy employed by ztest_fault_inject() when selecting which
4212 	 * offset are redundant and can be damaged.
4213 	 */
4214 	error = spa_scan(spa, POOL_SCAN_SCRUB);
4215 	if (error == 0) {
4216 		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
4217 			txg_wait_synced(spa_get_dsl(spa), 0);
4218 	}
4219 
4220 	mutex_enter(&ztest_vdev_lock);
4221 	ztest_device_removal_active = B_FALSE;
4222 	mutex_exit(&ztest_vdev_lock);
4223 }
4224 
4225 /*
4226  * Callback function which expands the physical size of the vdev.
4227  */
4228 static vdev_t *
4229 grow_vdev(vdev_t *vd, void *arg)
4230 {
4231 	spa_t *spa __maybe_unused = vd->vdev_spa;
4232 	size_t *newsize = arg;
4233 	size_t fsize;
4234 	int fd;
4235 
4236 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
4237 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4238 
4239 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
4240 		return (vd);
4241 
4242 	fsize = lseek(fd, 0, SEEK_END);
4243 	VERIFY0(ftruncate(fd, *newsize));
4244 
4245 	if (ztest_opts.zo_verbose >= 6) {
4246 		(void) printf("%s grew from %lu to %lu bytes\n",
4247 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
4248 	}
4249 	(void) close(fd);
4250 	return (NULL);
4251 }
4252 
4253 /*
4254  * Callback function which expands a given vdev by calling vdev_online().
4255  */
4256 static vdev_t *
4257 online_vdev(vdev_t *vd, void *arg)
4258 {
4259 	(void) arg;
4260 	spa_t *spa = vd->vdev_spa;
4261 	vdev_t *tvd = vd->vdev_top;
4262 	uint64_t guid = vd->vdev_guid;
4263 	uint64_t generation = spa->spa_config_generation + 1;
4264 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
4265 	int error;
4266 
4267 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
4268 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4269 
4270 	/* Calling vdev_online will initialize the new metaslabs */
4271 	spa_config_exit(spa, SCL_STATE, spa);
4272 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
4273 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4274 
4275 	/*
4276 	 * If vdev_online returned an error or the underlying vdev_open
4277 	 * failed then we abort the expand. The only way to know that
4278 	 * vdev_open fails is by checking the returned newstate.
4279 	 */
4280 	if (error || newstate != VDEV_STATE_HEALTHY) {
4281 		if (ztest_opts.zo_verbose >= 5) {
4282 			(void) printf("Unable to expand vdev, state %u, "
4283 			    "error %d\n", newstate, error);
4284 		}
4285 		return (vd);
4286 	}
4287 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
4288 
4289 	/*
4290 	 * Since we dropped the lock we need to ensure that we're
4291 	 * still talking to the original vdev. It's possible this
4292 	 * vdev may have been detached/replaced while we were
4293 	 * trying to online it.
4294 	 */
4295 	if (generation != spa->spa_config_generation) {
4296 		if (ztest_opts.zo_verbose >= 5) {
4297 			(void) printf("vdev configuration has changed, "
4298 			    "guid %"PRIu64", state %"PRIu64", "
4299 			    "expected gen %"PRIu64", got gen %"PRIu64"\n",
4300 			    guid,
4301 			    tvd->vdev_state,
4302 			    generation,
4303 			    spa->spa_config_generation);
4304 		}
4305 		return (vd);
4306 	}
4307 	return (NULL);
4308 }
4309 
4310 /*
4311  * Traverse the vdev tree calling the supplied function.
4312  * We continue to walk the tree until we either have walked all
4313  * children or we receive a non-NULL return from the callback.
4314  * If a NULL callback is passed, then we just return back the first
4315  * leaf vdev we encounter.
4316  */
4317 static vdev_t *
4318 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
4319 {
4320 	uint_t c;
4321 
4322 	if (vd->vdev_ops->vdev_op_leaf) {
4323 		if (func == NULL)
4324 			return (vd);
4325 		else
4326 			return (func(vd, arg));
4327 	}
4328 
4329 	for (c = 0; c < vd->vdev_children; c++) {
4330 		vdev_t *cvd = vd->vdev_child[c];
4331 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
4332 			return (cvd);
4333 	}
4334 	return (NULL);
4335 }
4336 
4337 /*
4338  * Verify that dynamic LUN growth works as expected.
4339  */
4340 void
4341 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
4342 {
4343 	(void) zd, (void) id;
4344 	spa_t *spa = ztest_spa;
4345 	vdev_t *vd, *tvd;
4346 	metaslab_class_t *mc;
4347 	metaslab_group_t *mg;
4348 	size_t psize, newsize;
4349 	uint64_t top;
4350 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
4351 
4352 	mutex_enter(&ztest_checkpoint_lock);
4353 	mutex_enter(&ztest_vdev_lock);
4354 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4355 
4356 	/*
4357 	 * If there is a vdev removal in progress, it could complete while
4358 	 * we are running, in which case we would not be able to verify
4359 	 * that the metaslab_class space increased (because it decreases
4360 	 * when the device removal completes).
4361 	 */
4362 	if (ztest_device_removal_active) {
4363 		spa_config_exit(spa, SCL_STATE, spa);
4364 		mutex_exit(&ztest_vdev_lock);
4365 		mutex_exit(&ztest_checkpoint_lock);
4366 		return;
4367 	}
4368 
4369 	/*
4370 	 * If we are under raidz expansion, the test can failed because the
4371 	 * metaslabs count will not increase immediately after the vdev is
4372 	 * expanded. It will happen only after raidz expansion completion.
4373 	 */
4374 	if (spa->spa_raidz_expand) {
4375 		spa_config_exit(spa, SCL_STATE, spa);
4376 		mutex_exit(&ztest_vdev_lock);
4377 		mutex_exit(&ztest_checkpoint_lock);
4378 		return;
4379 	}
4380 
4381 	top = ztest_random_vdev_top(spa, B_TRUE);
4382 
4383 	tvd = spa->spa_root_vdev->vdev_child[top];
4384 	mg = tvd->vdev_mg;
4385 	mc = mg->mg_class;
4386 	old_ms_count = tvd->vdev_ms_count;
4387 	old_class_space = metaslab_class_get_space(mc);
4388 
4389 	/*
4390 	 * Determine the size of the first leaf vdev associated with
4391 	 * our top-level device.
4392 	 */
4393 	vd = vdev_walk_tree(tvd, NULL, NULL);
4394 	ASSERT3P(vd, !=, NULL);
4395 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4396 
4397 	psize = vd->vdev_psize;
4398 
4399 	/*
4400 	 * We only try to expand the vdev if it's healthy, less than 4x its
4401 	 * original size, and it has a valid psize.
4402 	 */
4403 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
4404 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
4405 		spa_config_exit(spa, SCL_STATE, spa);
4406 		mutex_exit(&ztest_vdev_lock);
4407 		mutex_exit(&ztest_checkpoint_lock);
4408 		return;
4409 	}
4410 	ASSERT3U(psize, >, 0);
4411 	newsize = psize + MAX(psize / 8, SPA_MAXBLOCKSIZE);
4412 	ASSERT3U(newsize, >, psize);
4413 
4414 	if (ztest_opts.zo_verbose >= 6) {
4415 		(void) printf("Expanding LUN %s from %lu to %lu\n",
4416 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
4417 	}
4418 
4419 	/*
4420 	 * Growing the vdev is a two step process:
4421 	 *	1). expand the physical size (i.e. relabel)
4422 	 *	2). online the vdev to create the new metaslabs
4423 	 */
4424 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
4425 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
4426 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
4427 		if (ztest_opts.zo_verbose >= 5) {
4428 			(void) printf("Could not expand LUN because "
4429 			    "the vdev configuration changed.\n");
4430 		}
4431 		spa_config_exit(spa, SCL_STATE, spa);
4432 		mutex_exit(&ztest_vdev_lock);
4433 		mutex_exit(&ztest_checkpoint_lock);
4434 		return;
4435 	}
4436 
4437 	spa_config_exit(spa, SCL_STATE, spa);
4438 
4439 	/*
4440 	 * Expanding the LUN will update the config asynchronously,
4441 	 * thus we must wait for the async thread to complete any
4442 	 * pending tasks before proceeding.
4443 	 */
4444 	for (;;) {
4445 		boolean_t done;
4446 		mutex_enter(&spa->spa_async_lock);
4447 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
4448 		mutex_exit(&spa->spa_async_lock);
4449 		if (done)
4450 			break;
4451 		txg_wait_synced(spa_get_dsl(spa), 0);
4452 		(void) poll(NULL, 0, 100);
4453 	}
4454 
4455 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4456 
4457 	tvd = spa->spa_root_vdev->vdev_child[top];
4458 	new_ms_count = tvd->vdev_ms_count;
4459 	new_class_space = metaslab_class_get_space(mc);
4460 
4461 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
4462 		if (ztest_opts.zo_verbose >= 5) {
4463 			(void) printf("Could not verify LUN expansion due to "
4464 			    "intervening vdev offline or remove.\n");
4465 		}
4466 		spa_config_exit(spa, SCL_STATE, spa);
4467 		mutex_exit(&ztest_vdev_lock);
4468 		mutex_exit(&ztest_checkpoint_lock);
4469 		return;
4470 	}
4471 
4472 	/*
4473 	 * Make sure we were able to grow the vdev.
4474 	 */
4475 	if (new_ms_count <= old_ms_count) {
4476 		fatal(B_FALSE,
4477 		    "LUN expansion failed: ms_count %"PRIu64" < %"PRIu64"\n",
4478 		    old_ms_count, new_ms_count);
4479 	}
4480 
4481 	/*
4482 	 * Make sure we were able to grow the pool.
4483 	 */
4484 	if (new_class_space <= old_class_space) {
4485 		fatal(B_FALSE,
4486 		    "LUN expansion failed: class_space %"PRIu64" < %"PRIu64"\n",
4487 		    old_class_space, new_class_space);
4488 	}
4489 
4490 	if (ztest_opts.zo_verbose >= 5) {
4491 		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
4492 
4493 		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
4494 		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
4495 		(void) printf("%s grew from %s to %s\n",
4496 		    spa->spa_name, oldnumbuf, newnumbuf);
4497 	}
4498 
4499 	spa_config_exit(spa, SCL_STATE, spa);
4500 	mutex_exit(&ztest_vdev_lock);
4501 	mutex_exit(&ztest_checkpoint_lock);
4502 }
4503 
4504 /*
4505  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
4506  */
4507 static void
4508 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
4509 {
4510 	(void) arg, (void) cr;
4511 
4512 	/*
4513 	 * Create the objects common to all ztest datasets.
4514 	 */
4515 	VERIFY0(zap_create_claim(os, ZTEST_DIROBJ,
4516 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx));
4517 }
4518 
4519 static int
4520 ztest_dataset_create(char *dsname)
4521 {
4522 	int err;
4523 	uint64_t rand;
4524 	dsl_crypto_params_t *dcp = NULL;
4525 
4526 	/*
4527 	 * 50% of the time, we create encrypted datasets
4528 	 * using a random cipher suite and a hard-coded
4529 	 * wrapping key.
4530 	 */
4531 	rand = ztest_random(2);
4532 	if (rand != 0) {
4533 		nvlist_t *crypto_args = fnvlist_alloc();
4534 		nvlist_t *props = fnvlist_alloc();
4535 
4536 		/* slight bias towards the default cipher suite */
4537 		rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
4538 		if (rand < ZIO_CRYPT_AES_128_CCM)
4539 			rand = ZIO_CRYPT_ON;
4540 
4541 		fnvlist_add_uint64(props,
4542 		    zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
4543 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
4544 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
4545 
4546 		/*
4547 		 * These parameters aren't really used by the kernel. They
4548 		 * are simply stored so that userspace knows how to load
4549 		 * the wrapping key.
4550 		 */
4551 		fnvlist_add_uint64(props,
4552 		    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
4553 		fnvlist_add_string(props,
4554 		    zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
4555 		fnvlist_add_uint64(props,
4556 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
4557 		fnvlist_add_uint64(props,
4558 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
4559 
4560 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
4561 		    crypto_args, &dcp));
4562 
4563 		/*
4564 		 * Cycle through all available encryption implementations
4565 		 * to verify interoperability.
4566 		 */
4567 		VERIFY0(gcm_impl_set("cycle"));
4568 		VERIFY0(aes_impl_set("cycle"));
4569 
4570 		fnvlist_free(crypto_args);
4571 		fnvlist_free(props);
4572 	}
4573 
4574 	err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
4575 	    ztest_objset_create_cb, NULL);
4576 	dsl_crypto_params_free(dcp, !!err);
4577 
4578 	rand = ztest_random(100);
4579 	if (err || rand < 80)
4580 		return (err);
4581 
4582 	if (ztest_opts.zo_verbose >= 5)
4583 		(void) printf("Setting dataset %s to sync always\n", dsname);
4584 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
4585 	    ZFS_SYNC_ALWAYS, B_FALSE));
4586 }
4587 
4588 static int
4589 ztest_objset_destroy_cb(const char *name, void *arg)
4590 {
4591 	(void) arg;
4592 	objset_t *os;
4593 	dmu_object_info_t doi;
4594 	int error;
4595 
4596 	/*
4597 	 * Verify that the dataset contains a directory object.
4598 	 */
4599 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
4600 	    B_TRUE, FTAG, &os));
4601 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
4602 	if (error != ENOENT) {
4603 		/* We could have crashed in the middle of destroying it */
4604 		ASSERT0(error);
4605 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
4606 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
4607 	}
4608 	dmu_objset_disown(os, B_TRUE, FTAG);
4609 
4610 	/*
4611 	 * Destroy the dataset.
4612 	 */
4613 	if (strchr(name, '@') != NULL) {
4614 		error = dsl_destroy_snapshot(name, B_TRUE);
4615 		if (error != ECHRNG) {
4616 			/*
4617 			 * The program was executed, but encountered a runtime
4618 			 * error, such as insufficient slop, or a hold on the
4619 			 * dataset.
4620 			 */
4621 			ASSERT0(error);
4622 		}
4623 	} else {
4624 		error = dsl_destroy_head(name);
4625 		if (error == ENOSPC) {
4626 			/* There could be checkpoint or insufficient slop */
4627 			ztest_record_enospc(FTAG);
4628 		} else if (error != EBUSY) {
4629 			/* There could be a hold on this dataset */
4630 			ASSERT0(error);
4631 		}
4632 	}
4633 	return (0);
4634 }
4635 
4636 static boolean_t
4637 ztest_snapshot_create(char *osname, uint64_t id)
4638 {
4639 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4640 	int error;
4641 
4642 	(void) snprintf(snapname, sizeof (snapname), "%"PRIu64"", id);
4643 
4644 	error = dmu_objset_snapshot_one(osname, snapname);
4645 	if (error == ENOSPC) {
4646 		ztest_record_enospc(FTAG);
4647 		return (B_FALSE);
4648 	}
4649 	if (error != 0 && error != EEXIST && error != ECHRNG) {
4650 		fatal(B_FALSE, "ztest_snapshot_create(%s@%s) = %d", osname,
4651 		    snapname, error);
4652 	}
4653 	return (B_TRUE);
4654 }
4655 
4656 static boolean_t
4657 ztest_snapshot_destroy(char *osname, uint64_t id)
4658 {
4659 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4660 	int error;
4661 
4662 	(void) snprintf(snapname, sizeof (snapname), "%s@%"PRIu64"",
4663 	    osname, id);
4664 
4665 	error = dsl_destroy_snapshot(snapname, B_FALSE);
4666 	if (error != 0 && error != ENOENT && error != ECHRNG)
4667 		fatal(B_FALSE, "ztest_snapshot_destroy(%s) = %d",
4668 		    snapname, error);
4669 	return (B_TRUE);
4670 }
4671 
4672 void
4673 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
4674 {
4675 	(void) zd;
4676 	ztest_ds_t *zdtmp;
4677 	int iters;
4678 	int error;
4679 	objset_t *os, *os2;
4680 	char name[ZFS_MAX_DATASET_NAME_LEN];
4681 	zilog_t *zilog;
4682 	int i;
4683 
4684 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
4685 
4686 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4687 
4688 	(void) snprintf(name, sizeof (name), "%s/temp_%"PRIu64"",
4689 	    ztest_opts.zo_pool, id);
4690 
4691 	/*
4692 	 * If this dataset exists from a previous run, process its replay log
4693 	 * half of the time.  If we don't replay it, then dsl_destroy_head()
4694 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
4695 	 */
4696 	if (ztest_random(2) == 0 &&
4697 	    ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
4698 	    B_TRUE, FTAG, &os) == 0) {
4699 		ztest_zd_init(zdtmp, NULL, os);
4700 		zil_replay(os, zdtmp, ztest_replay_vector);
4701 		ztest_zd_fini(zdtmp);
4702 		dmu_objset_disown(os, B_TRUE, FTAG);
4703 	}
4704 
4705 	/*
4706 	 * There may be an old instance of the dataset we're about to
4707 	 * create lying around from a previous run.  If so, destroy it
4708 	 * and all of its snapshots.
4709 	 */
4710 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
4711 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
4712 
4713 	/*
4714 	 * Verify that the destroyed dataset is no longer in the namespace.
4715 	 * It may still be present if the destroy above fails with ENOSPC.
4716 	 */
4717 	error = ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, B_TRUE,
4718 	    FTAG, &os);
4719 	if (error == 0) {
4720 		dmu_objset_disown(os, B_TRUE, FTAG);
4721 		ztest_record_enospc(FTAG);
4722 		goto out;
4723 	}
4724 	VERIFY3U(ENOENT, ==, error);
4725 
4726 	/*
4727 	 * Verify that we can create a new dataset.
4728 	 */
4729 	error = ztest_dataset_create(name);
4730 	if (error) {
4731 		if (error == ENOSPC) {
4732 			ztest_record_enospc(FTAG);
4733 			goto out;
4734 		}
4735 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", name, error);
4736 	}
4737 
4738 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
4739 	    FTAG, &os));
4740 
4741 	ztest_zd_init(zdtmp, NULL, os);
4742 
4743 	/*
4744 	 * Open the intent log for it.
4745 	 */
4746 	zilog = zil_open(os, ztest_get_data, NULL);
4747 
4748 	/*
4749 	 * Put some objects in there, do a little I/O to them,
4750 	 * and randomly take a couple of snapshots along the way.
4751 	 */
4752 	iters = ztest_random(5);
4753 	for (i = 0; i < iters; i++) {
4754 		ztest_dmu_object_alloc_free(zdtmp, id);
4755 		if (ztest_random(iters) == 0)
4756 			(void) ztest_snapshot_create(name, i);
4757 	}
4758 
4759 	/*
4760 	 * Verify that we cannot create an existing dataset.
4761 	 */
4762 	VERIFY3U(EEXIST, ==,
4763 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
4764 
4765 	/*
4766 	 * Verify that we can hold an objset that is also owned.
4767 	 */
4768 	VERIFY0(dmu_objset_hold(name, FTAG, &os2));
4769 	dmu_objset_rele(os2, FTAG);
4770 
4771 	/*
4772 	 * Verify that we cannot own an objset that is already owned.
4773 	 */
4774 	VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
4775 	    B_FALSE, B_TRUE, FTAG, &os2));
4776 
4777 	zil_close(zilog);
4778 	dmu_objset_disown(os, B_TRUE, FTAG);
4779 	ztest_zd_fini(zdtmp);
4780 out:
4781 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4782 
4783 	umem_free(zdtmp, sizeof (ztest_ds_t));
4784 }
4785 
4786 /*
4787  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
4788  */
4789 void
4790 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
4791 {
4792 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4793 	(void) ztest_snapshot_destroy(zd->zd_name, id);
4794 	(void) ztest_snapshot_create(zd->zd_name, id);
4795 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4796 }
4797 
4798 /*
4799  * Cleanup non-standard snapshots and clones.
4800  */
4801 static void
4802 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
4803 {
4804 	char *snap1name;
4805 	char *clone1name;
4806 	char *snap2name;
4807 	char *clone2name;
4808 	char *snap3name;
4809 	int error;
4810 
4811 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4812 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4813 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4814 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4815 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4816 
4817 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4818 	    osname, id);
4819 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4820 	    osname, id);
4821 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4822 	    clone1name, id);
4823 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4824 	    osname, id);
4825 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4826 	    clone1name, id);
4827 
4828 	error = dsl_destroy_head(clone2name);
4829 	if (error && error != ENOENT)
4830 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone2name, error);
4831 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
4832 	if (error && error != ENOENT)
4833 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4834 		    snap3name, error);
4835 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
4836 	if (error && error != ENOENT)
4837 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4838 		    snap2name, error);
4839 	error = dsl_destroy_head(clone1name);
4840 	if (error && error != ENOENT)
4841 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone1name, error);
4842 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
4843 	if (error && error != ENOENT)
4844 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4845 		    snap1name, error);
4846 
4847 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4848 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4849 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4850 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4851 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4852 }
4853 
4854 /*
4855  * Verify dsl_dataset_promote handles EBUSY
4856  */
4857 void
4858 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
4859 {
4860 	objset_t *os;
4861 	char *snap1name;
4862 	char *clone1name;
4863 	char *snap2name;
4864 	char *clone2name;
4865 	char *snap3name;
4866 	char *osname = zd->zd_name;
4867 	int error;
4868 
4869 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4870 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4871 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4872 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4873 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4874 
4875 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4876 
4877 	ztest_dsl_dataset_cleanup(osname, id);
4878 
4879 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4880 	    osname, id);
4881 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4882 	    osname, id);
4883 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4884 	    clone1name, id);
4885 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4886 	    osname, id);
4887 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4888 	    clone1name, id);
4889 
4890 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
4891 	if (error && error != EEXIST) {
4892 		if (error == ENOSPC) {
4893 			ztest_record_enospc(FTAG);
4894 			goto out;
4895 		}
4896 		fatal(B_FALSE, "dmu_take_snapshot(%s) = %d", snap1name, error);
4897 	}
4898 
4899 	error = dmu_objset_clone(clone1name, snap1name);
4900 	if (error) {
4901 		if (error == ENOSPC) {
4902 			ztest_record_enospc(FTAG);
4903 			goto out;
4904 		}
4905 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone1name, error);
4906 	}
4907 
4908 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
4909 	if (error && error != EEXIST) {
4910 		if (error == ENOSPC) {
4911 			ztest_record_enospc(FTAG);
4912 			goto out;
4913 		}
4914 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap2name, error);
4915 	}
4916 
4917 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
4918 	if (error && error != EEXIST) {
4919 		if (error == ENOSPC) {
4920 			ztest_record_enospc(FTAG);
4921 			goto out;
4922 		}
4923 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap3name, error);
4924 	}
4925 
4926 	error = dmu_objset_clone(clone2name, snap3name);
4927 	if (error) {
4928 		if (error == ENOSPC) {
4929 			ztest_record_enospc(FTAG);
4930 			goto out;
4931 		}
4932 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone2name, error);
4933 	}
4934 
4935 	error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
4936 	    FTAG, &os);
4937 	if (error)
4938 		fatal(B_FALSE, "dmu_objset_own(%s) = %d", snap2name, error);
4939 	error = dsl_dataset_promote(clone2name, NULL);
4940 	if (error == ENOSPC) {
4941 		dmu_objset_disown(os, B_TRUE, FTAG);
4942 		ztest_record_enospc(FTAG);
4943 		goto out;
4944 	}
4945 	if (error != EBUSY)
4946 		fatal(B_FALSE, "dsl_dataset_promote(%s), %d, not EBUSY",
4947 		    clone2name, error);
4948 	dmu_objset_disown(os, B_TRUE, FTAG);
4949 
4950 out:
4951 	ztest_dsl_dataset_cleanup(osname, id);
4952 
4953 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4954 
4955 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4956 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4957 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4958 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4959 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4960 }
4961 
4962 #undef OD_ARRAY_SIZE
4963 #define	OD_ARRAY_SIZE	4
4964 
4965 /*
4966  * Verify that dmu_object_{alloc,free} work as expected.
4967  */
4968 void
4969 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4970 {
4971 	ztest_od_t *od;
4972 	int batchsize;
4973 	int size;
4974 	int b;
4975 
4976 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4977 	od = umem_alloc(size, UMEM_NOFAIL);
4978 	batchsize = OD_ARRAY_SIZE;
4979 
4980 	for (b = 0; b < batchsize; b++)
4981 		ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
4982 		    0, 0, 0);
4983 
4984 	/*
4985 	 * Destroy the previous batch of objects, create a new batch,
4986 	 * and do some I/O on the new objects.
4987 	 */
4988 	if (ztest_object_init(zd, od, size, B_TRUE) != 0) {
4989 		zd->zd_od = NULL;
4990 		umem_free(od, size);
4991 		return;
4992 	}
4993 
4994 	while (ztest_random(4 * batchsize) != 0)
4995 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
4996 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4997 
4998 	umem_free(od, size);
4999 }
5000 
5001 /*
5002  * Rewind the global allocator to verify object allocation backfilling.
5003  */
5004 void
5005 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
5006 {
5007 	(void) id;
5008 	objset_t *os = zd->zd_os;
5009 	uint_t dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
5010 	uint64_t object;
5011 
5012 	/*
5013 	 * Rewind the global allocator randomly back to a lower object number
5014 	 * to force backfilling and reclamation of recently freed dnodes.
5015 	 */
5016 	mutex_enter(&os->os_obj_lock);
5017 	object = ztest_random(os->os_obj_next_chunk);
5018 	os->os_obj_next_chunk = P2ALIGN_TYPED(object, dnodes_per_chunk,
5019 	    uint64_t);
5020 	mutex_exit(&os->os_obj_lock);
5021 }
5022 
5023 #undef OD_ARRAY_SIZE
5024 #define	OD_ARRAY_SIZE	2
5025 
5026 /*
5027  * Verify that dmu_{read,write} work as expected.
5028  */
5029 void
5030 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
5031 {
5032 	int size;
5033 	ztest_od_t *od;
5034 
5035 	objset_t *os = zd->zd_os;
5036 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
5037 	od = umem_alloc(size, UMEM_NOFAIL);
5038 	dmu_tx_t *tx;
5039 	int freeit, error;
5040 	uint64_t i, n, s, txg;
5041 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
5042 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
5043 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
5044 	uint64_t regions = 997;
5045 	uint64_t stride = 123456789ULL;
5046 	uint64_t width = 40;
5047 	int free_percent = 5;
5048 
5049 	/*
5050 	 * This test uses two objects, packobj and bigobj, that are always
5051 	 * updated together (i.e. in the same tx) so that their contents are
5052 	 * in sync and can be compared.  Their contents relate to each other
5053 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
5054 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
5055 	 * for any index n, there are three bufwads that should be identical:
5056 	 *
5057 	 *	packobj, at offset n * sizeof (bufwad_t)
5058 	 *	bigobj, at the head of the nth chunk
5059 	 *	bigobj, at the tail of the nth chunk
5060 	 *
5061 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
5062 	 * and it doesn't have any relation to the object blocksize.
5063 	 * The only requirement is that it can hold at least two bufwads.
5064 	 *
5065 	 * Normally, we write the bufwad to each of these locations.
5066 	 * However, free_percent of the time we instead write zeroes to
5067 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
5068 	 * bigobj to packobj, we can verify that the DMU is correctly
5069 	 * tracking which parts of an object are allocated and free,
5070 	 * and that the contents of the allocated blocks are correct.
5071 	 */
5072 
5073 	/*
5074 	 * Read the directory info.  If it's the first time, set things up.
5075 	 */
5076 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
5077 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
5078 	    chunksize);
5079 
5080 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
5081 		umem_free(od, size);
5082 		return;
5083 	}
5084 
5085 	bigobj = od[0].od_object;
5086 	packobj = od[1].od_object;
5087 	chunksize = od[0].od_gen;
5088 	ASSERT3U(chunksize, ==, od[1].od_gen);
5089 
5090 	/*
5091 	 * Prefetch a random chunk of the big object.
5092 	 * Our aim here is to get some async reads in flight
5093 	 * for blocks that we may free below; the DMU should
5094 	 * handle this race correctly.
5095 	 */
5096 	n = ztest_random(regions) * stride + ztest_random(width);
5097 	s = 1 + ztest_random(2 * width - 1);
5098 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
5099 	    ZIO_PRIORITY_SYNC_READ);
5100 
5101 	/*
5102 	 * Pick a random index and compute the offsets into packobj and bigobj.
5103 	 */
5104 	n = ztest_random(regions) * stride + ztest_random(width);
5105 	s = 1 + ztest_random(width - 1);
5106 
5107 	packoff = n * sizeof (bufwad_t);
5108 	packsize = s * sizeof (bufwad_t);
5109 
5110 	bigoff = n * chunksize;
5111 	bigsize = s * chunksize;
5112 
5113 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
5114 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
5115 
5116 	/*
5117 	 * free_percent of the time, free a range of bigobj rather than
5118 	 * overwriting it.
5119 	 */
5120 	freeit = (ztest_random(100) < free_percent);
5121 
5122 	/*
5123 	 * Read the current contents of our objects.
5124 	 */
5125 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
5126 	    DMU_READ_PREFETCH);
5127 	ASSERT0(error);
5128 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
5129 	    DMU_READ_PREFETCH);
5130 	ASSERT0(error);
5131 
5132 	/*
5133 	 * Get a tx for the mods to both packobj and bigobj.
5134 	 */
5135 	tx = dmu_tx_create(os);
5136 
5137 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
5138 
5139 	if (freeit)
5140 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
5141 	else
5142 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
5143 
5144 	/* This accounts for setting the checksum/compression. */
5145 	dmu_tx_hold_bonus(tx, bigobj);
5146 
5147 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5148 	if (txg == 0) {
5149 		umem_free(packbuf, packsize);
5150 		umem_free(bigbuf, bigsize);
5151 		umem_free(od, size);
5152 		return;
5153 	}
5154 
5155 	enum zio_checksum cksum;
5156 	do {
5157 		cksum = (enum zio_checksum)
5158 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
5159 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
5160 	dmu_object_set_checksum(os, bigobj, cksum, tx);
5161 
5162 	enum zio_compress comp;
5163 	do {
5164 		comp = (enum zio_compress)
5165 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
5166 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
5167 	dmu_object_set_compress(os, bigobj, comp, tx);
5168 
5169 	/*
5170 	 * For each index from n to n + s, verify that the existing bufwad
5171 	 * in packobj matches the bufwads at the head and tail of the
5172 	 * corresponding chunk in bigobj.  Then update all three bufwads
5173 	 * with the new values we want to write out.
5174 	 */
5175 	for (i = 0; i < s; i++) {
5176 		/* LINTED */
5177 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
5178 		/* LINTED */
5179 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
5180 		/* LINTED */
5181 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
5182 
5183 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
5184 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
5185 
5186 		if (pack->bw_txg > txg)
5187 			fatal(B_FALSE,
5188 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
5189 			    pack->bw_txg, txg);
5190 
5191 		if (pack->bw_data != 0 && pack->bw_index != n + i)
5192 			fatal(B_FALSE, "wrong index: "
5193 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
5194 			    pack->bw_index, n, i);
5195 
5196 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
5197 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
5198 			    pack, bigH);
5199 
5200 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
5201 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
5202 			    pack, bigT);
5203 
5204 		if (freeit) {
5205 			memset(pack, 0, sizeof (bufwad_t));
5206 		} else {
5207 			pack->bw_index = n + i;
5208 			pack->bw_txg = txg;
5209 			pack->bw_data = 1 + ztest_random(-2ULL);
5210 		}
5211 		*bigH = *pack;
5212 		*bigT = *pack;
5213 	}
5214 
5215 	/*
5216 	 * We've verified all the old bufwads, and made new ones.
5217 	 * Now write them out.
5218 	 */
5219 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
5220 
5221 	if (freeit) {
5222 		if (ztest_opts.zo_verbose >= 7) {
5223 			(void) printf("freeing offset %"PRIx64" size %"PRIx64""
5224 			    " txg %"PRIx64"\n",
5225 			    bigoff, bigsize, txg);
5226 		}
5227 		VERIFY0(dmu_free_range(os, bigobj, bigoff, bigsize, tx));
5228 	} else {
5229 		if (ztest_opts.zo_verbose >= 7) {
5230 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
5231 			    " txg %"PRIx64"\n",
5232 			    bigoff, bigsize, txg);
5233 		}
5234 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
5235 	}
5236 
5237 	dmu_tx_commit(tx);
5238 
5239 	/*
5240 	 * Sanity check the stuff we just wrote.
5241 	 */
5242 	{
5243 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
5244 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
5245 
5246 		VERIFY0(dmu_read(os, packobj, packoff,
5247 		    packsize, packcheck, DMU_READ_PREFETCH));
5248 		VERIFY0(dmu_read(os, bigobj, bigoff,
5249 		    bigsize, bigcheck, DMU_READ_PREFETCH));
5250 
5251 		ASSERT0(memcmp(packbuf, packcheck, packsize));
5252 		ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
5253 
5254 		umem_free(packcheck, packsize);
5255 		umem_free(bigcheck, bigsize);
5256 	}
5257 
5258 	umem_free(packbuf, packsize);
5259 	umem_free(bigbuf, bigsize);
5260 	umem_free(od, size);
5261 }
5262 
5263 static void
5264 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
5265     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
5266 {
5267 	uint64_t i;
5268 	bufwad_t *pack;
5269 	bufwad_t *bigH;
5270 	bufwad_t *bigT;
5271 
5272 	/*
5273 	 * For each index from n to n + s, verify that the existing bufwad
5274 	 * in packobj matches the bufwads at the head and tail of the
5275 	 * corresponding chunk in bigobj.  Then update all three bufwads
5276 	 * with the new values we want to write out.
5277 	 */
5278 	for (i = 0; i < s; i++) {
5279 		/* LINTED */
5280 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
5281 		/* LINTED */
5282 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
5283 		/* LINTED */
5284 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
5285 
5286 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
5287 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
5288 
5289 		if (pack->bw_txg > txg)
5290 			fatal(B_FALSE,
5291 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
5292 			    pack->bw_txg, txg);
5293 
5294 		if (pack->bw_data != 0 && pack->bw_index != n + i)
5295 			fatal(B_FALSE, "wrong index: "
5296 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
5297 			    pack->bw_index, n, i);
5298 
5299 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
5300 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
5301 			    pack, bigH);
5302 
5303 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
5304 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
5305 			    pack, bigT);
5306 
5307 		pack->bw_index = n + i;
5308 		pack->bw_txg = txg;
5309 		pack->bw_data = 1 + ztest_random(-2ULL);
5310 
5311 		*bigH = *pack;
5312 		*bigT = *pack;
5313 	}
5314 }
5315 
5316 #undef OD_ARRAY_SIZE
5317 #define	OD_ARRAY_SIZE	2
5318 
5319 void
5320 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
5321 {
5322 	objset_t *os = zd->zd_os;
5323 	ztest_od_t *od;
5324 	dmu_tx_t *tx;
5325 	uint64_t i;
5326 	int error;
5327 	int size;
5328 	uint64_t n, s, txg;
5329 	bufwad_t *packbuf, *bigbuf;
5330 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
5331 	uint64_t blocksize = ztest_random_blocksize();
5332 	uint64_t chunksize = blocksize;
5333 	uint64_t regions = 997;
5334 	uint64_t stride = 123456789ULL;
5335 	uint64_t width = 9;
5336 	dmu_buf_t *bonus_db;
5337 	arc_buf_t **bigbuf_arcbufs;
5338 	dmu_object_info_t doi;
5339 
5340 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
5341 	od = umem_alloc(size, UMEM_NOFAIL);
5342 
5343 	/*
5344 	 * This test uses two objects, packobj and bigobj, that are always
5345 	 * updated together (i.e. in the same tx) so that their contents are
5346 	 * in sync and can be compared.  Their contents relate to each other
5347 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
5348 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
5349 	 * for any index n, there are three bufwads that should be identical:
5350 	 *
5351 	 *	packobj, at offset n * sizeof (bufwad_t)
5352 	 *	bigobj, at the head of the nth chunk
5353 	 *	bigobj, at the tail of the nth chunk
5354 	 *
5355 	 * The chunk size is set equal to bigobj block size so that
5356 	 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
5357 	 */
5358 
5359 	/*
5360 	 * Read the directory info.  If it's the first time, set things up.
5361 	 */
5362 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5363 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
5364 	    chunksize);
5365 
5366 
5367 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
5368 		umem_free(od, size);
5369 		return;
5370 	}
5371 
5372 	bigobj = od[0].od_object;
5373 	packobj = od[1].od_object;
5374 	blocksize = od[0].od_blocksize;
5375 	chunksize = blocksize;
5376 	ASSERT3U(chunksize, ==, od[1].od_gen);
5377 
5378 	VERIFY0(dmu_object_info(os, bigobj, &doi));
5379 	VERIFY(ISP2(doi.doi_data_block_size));
5380 	VERIFY3U(chunksize, ==, doi.doi_data_block_size);
5381 	VERIFY3U(chunksize, >=, 2 * sizeof (bufwad_t));
5382 
5383 	/*
5384 	 * Pick a random index and compute the offsets into packobj and bigobj.
5385 	 */
5386 	n = ztest_random(regions) * stride + ztest_random(width);
5387 	s = 1 + ztest_random(width - 1);
5388 
5389 	packoff = n * sizeof (bufwad_t);
5390 	packsize = s * sizeof (bufwad_t);
5391 
5392 	bigoff = n * chunksize;
5393 	bigsize = s * chunksize;
5394 
5395 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
5396 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
5397 
5398 	VERIFY0(dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
5399 
5400 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
5401 
5402 	/*
5403 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
5404 	 * Iteration 1 test zcopy to already referenced dbufs.
5405 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
5406 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
5407 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
5408 	 * Iteration 5 test zcopy when it can't be done.
5409 	 * Iteration 6 one more zcopy write.
5410 	 */
5411 	for (i = 0; i < 7; i++) {
5412 		uint64_t j;
5413 		uint64_t off;
5414 
5415 		/*
5416 		 * In iteration 5 (i == 5) use arcbufs
5417 		 * that don't match bigobj blksz to test
5418 		 * dmu_assign_arcbuf_by_dbuf() when it can't directly
5419 		 * assign an arcbuf to a dbuf.
5420 		 */
5421 		for (j = 0; j < s; j++) {
5422 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5423 				bigbuf_arcbufs[j] =
5424 				    dmu_request_arcbuf(bonus_db, chunksize);
5425 			} else {
5426 				bigbuf_arcbufs[2 * j] =
5427 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5428 				bigbuf_arcbufs[2 * j + 1] =
5429 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5430 			}
5431 		}
5432 
5433 		/*
5434 		 * Get a tx for the mods to both packobj and bigobj.
5435 		 */
5436 		tx = dmu_tx_create(os);
5437 
5438 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
5439 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
5440 
5441 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5442 		if (txg == 0) {
5443 			umem_free(packbuf, packsize);
5444 			umem_free(bigbuf, bigsize);
5445 			for (j = 0; j < s; j++) {
5446 				if (i != 5 ||
5447 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
5448 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
5449 				} else {
5450 					dmu_return_arcbuf(
5451 					    bigbuf_arcbufs[2 * j]);
5452 					dmu_return_arcbuf(
5453 					    bigbuf_arcbufs[2 * j + 1]);
5454 				}
5455 			}
5456 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5457 			umem_free(od, size);
5458 			dmu_buf_rele(bonus_db, FTAG);
5459 			return;
5460 		}
5461 
5462 		/*
5463 		 * 50% of the time don't read objects in the 1st iteration to
5464 		 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
5465 		 * no existing dbufs for the specified offsets.
5466 		 */
5467 		if (i != 0 || ztest_random(2) != 0) {
5468 			error = dmu_read(os, packobj, packoff,
5469 			    packsize, packbuf, DMU_READ_PREFETCH);
5470 			ASSERT0(error);
5471 			error = dmu_read(os, bigobj, bigoff, bigsize,
5472 			    bigbuf, DMU_READ_PREFETCH);
5473 			ASSERT0(error);
5474 		}
5475 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
5476 		    n, chunksize, txg);
5477 
5478 		/*
5479 		 * We've verified all the old bufwads, and made new ones.
5480 		 * Now write them out.
5481 		 */
5482 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
5483 		if (ztest_opts.zo_verbose >= 7) {
5484 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
5485 			    " txg %"PRIx64"\n",
5486 			    bigoff, bigsize, txg);
5487 		}
5488 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
5489 			dmu_buf_t *dbt;
5490 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5491 				memcpy(bigbuf_arcbufs[j]->b_data,
5492 				    (caddr_t)bigbuf + (off - bigoff),
5493 				    chunksize);
5494 			} else {
5495 				memcpy(bigbuf_arcbufs[2 * j]->b_data,
5496 				    (caddr_t)bigbuf + (off - bigoff),
5497 				    chunksize / 2);
5498 				memcpy(bigbuf_arcbufs[2 * j + 1]->b_data,
5499 				    (caddr_t)bigbuf + (off - bigoff) +
5500 				    chunksize / 2,
5501 				    chunksize / 2);
5502 			}
5503 
5504 			if (i == 1) {
5505 				VERIFY(dmu_buf_hold(os, bigobj, off,
5506 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
5507 			}
5508 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5509 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5510 				    off, bigbuf_arcbufs[j], tx));
5511 			} else {
5512 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5513 				    off, bigbuf_arcbufs[2 * j], tx));
5514 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5515 				    off + chunksize / 2,
5516 				    bigbuf_arcbufs[2 * j + 1], tx));
5517 			}
5518 			if (i == 1) {
5519 				dmu_buf_rele(dbt, FTAG);
5520 			}
5521 		}
5522 		dmu_tx_commit(tx);
5523 
5524 		/*
5525 		 * Sanity check the stuff we just wrote.
5526 		 */
5527 		{
5528 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
5529 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
5530 
5531 			VERIFY0(dmu_read(os, packobj, packoff,
5532 			    packsize, packcheck, DMU_READ_PREFETCH));
5533 			VERIFY0(dmu_read(os, bigobj, bigoff,
5534 			    bigsize, bigcheck, DMU_READ_PREFETCH));
5535 
5536 			ASSERT0(memcmp(packbuf, packcheck, packsize));
5537 			ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
5538 
5539 			umem_free(packcheck, packsize);
5540 			umem_free(bigcheck, bigsize);
5541 		}
5542 		if (i == 2) {
5543 			txg_wait_open(dmu_objset_pool(os), 0, B_TRUE);
5544 		} else if (i == 3) {
5545 			txg_wait_synced(dmu_objset_pool(os), 0);
5546 		}
5547 	}
5548 
5549 	dmu_buf_rele(bonus_db, FTAG);
5550 	umem_free(packbuf, packsize);
5551 	umem_free(bigbuf, bigsize);
5552 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5553 	umem_free(od, size);
5554 }
5555 
5556 void
5557 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
5558 {
5559 	(void) id;
5560 	ztest_od_t *od;
5561 
5562 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5563 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
5564 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5565 
5566 	/*
5567 	 * Have multiple threads write to large offsets in an object
5568 	 * to verify that parallel writes to an object -- even to the
5569 	 * same blocks within the object -- doesn't cause any trouble.
5570 	 */
5571 	ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5572 
5573 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
5574 		return;
5575 
5576 	while (ztest_random(10) != 0)
5577 		ztest_io(zd, od->od_object, offset);
5578 
5579 	umem_free(od, sizeof (ztest_od_t));
5580 }
5581 
5582 void
5583 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
5584 {
5585 	ztest_od_t *od;
5586 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
5587 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5588 	uint64_t count = ztest_random(20) + 1;
5589 	uint64_t blocksize = ztest_random_blocksize();
5590 	void *data;
5591 
5592 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5593 
5594 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5595 
5596 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5597 	    !ztest_random(2)) != 0) {
5598 		umem_free(od, sizeof (ztest_od_t));
5599 		return;
5600 	}
5601 
5602 	if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
5603 		umem_free(od, sizeof (ztest_od_t));
5604 		return;
5605 	}
5606 
5607 	ztest_prealloc(zd, od->od_object, offset, count * blocksize);
5608 
5609 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
5610 
5611 	while (ztest_random(count) != 0) {
5612 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
5613 		if (ztest_write(zd, od->od_object, randoff, blocksize,
5614 		    data) != 0)
5615 			break;
5616 		while (ztest_random(4) != 0)
5617 			ztest_io(zd, od->od_object, randoff);
5618 	}
5619 
5620 	umem_free(data, blocksize);
5621 	umem_free(od, sizeof (ztest_od_t));
5622 }
5623 
5624 /*
5625  * Verify that zap_{create,destroy,add,remove,update} work as expected.
5626  */
5627 #define	ZTEST_ZAP_MIN_INTS	1
5628 #define	ZTEST_ZAP_MAX_INTS	4
5629 #define	ZTEST_ZAP_MAX_PROPS	1000
5630 
5631 void
5632 ztest_zap(ztest_ds_t *zd, uint64_t id)
5633 {
5634 	objset_t *os = zd->zd_os;
5635 	ztest_od_t *od;
5636 	uint64_t object;
5637 	uint64_t txg, last_txg;
5638 	uint64_t value[ZTEST_ZAP_MAX_INTS];
5639 	uint64_t zl_ints, zl_intsize, prop;
5640 	int i, ints;
5641 	dmu_tx_t *tx;
5642 	char propname[100], txgname[100];
5643 	int error;
5644 	const char *const hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
5645 
5646 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5647 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5648 
5649 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5650 	    !ztest_random(2)) != 0)
5651 		goto out;
5652 
5653 	object = od->od_object;
5654 
5655 	/*
5656 	 * Generate a known hash collision, and verify that
5657 	 * we can lookup and remove both entries.
5658 	 */
5659 	tx = dmu_tx_create(os);
5660 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5661 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5662 	if (txg == 0)
5663 		goto out;
5664 	for (i = 0; i < 2; i++) {
5665 		value[i] = i;
5666 		VERIFY0(zap_add(os, object, hc[i], sizeof (uint64_t),
5667 		    1, &value[i], tx));
5668 	}
5669 	for (i = 0; i < 2; i++) {
5670 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
5671 		    sizeof (uint64_t), 1, &value[i], tx));
5672 		VERIFY0(
5673 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
5674 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5675 		ASSERT3U(zl_ints, ==, 1);
5676 	}
5677 	for (i = 0; i < 2; i++) {
5678 		VERIFY0(zap_remove(os, object, hc[i], tx));
5679 	}
5680 	dmu_tx_commit(tx);
5681 
5682 	/*
5683 	 * Generate a bunch of random entries.
5684 	 */
5685 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
5686 
5687 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5688 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5689 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5690 	memset(value, 0, sizeof (value));
5691 	last_txg = 0;
5692 
5693 	/*
5694 	 * If these zap entries already exist, validate their contents.
5695 	 */
5696 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5697 	if (error == 0) {
5698 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5699 		ASSERT3U(zl_ints, ==, 1);
5700 
5701 		VERIFY0(zap_lookup(os, object, txgname, zl_intsize,
5702 		    zl_ints, &last_txg));
5703 
5704 		VERIFY0(zap_length(os, object, propname, &zl_intsize,
5705 		    &zl_ints));
5706 
5707 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5708 		ASSERT3U(zl_ints, ==, ints);
5709 
5710 		VERIFY0(zap_lookup(os, object, propname, zl_intsize,
5711 		    zl_ints, value));
5712 
5713 		for (i = 0; i < ints; i++) {
5714 			ASSERT3U(value[i], ==, last_txg + object + i);
5715 		}
5716 	} else {
5717 		ASSERT3U(error, ==, ENOENT);
5718 	}
5719 
5720 	/*
5721 	 * Atomically update two entries in our zap object.
5722 	 * The first is named txg_%llu, and contains the txg
5723 	 * in which the property was last updated.  The second
5724 	 * is named prop_%llu, and the nth element of its value
5725 	 * should be txg + object + n.
5726 	 */
5727 	tx = dmu_tx_create(os);
5728 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5729 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5730 	if (txg == 0)
5731 		goto out;
5732 
5733 	if (last_txg > txg)
5734 		fatal(B_FALSE, "zap future leak: old %"PRIu64" new %"PRIu64"",
5735 		    last_txg, txg);
5736 
5737 	for (i = 0; i < ints; i++)
5738 		value[i] = txg + object + i;
5739 
5740 	VERIFY0(zap_update(os, object, txgname, sizeof (uint64_t),
5741 	    1, &txg, tx));
5742 	VERIFY0(zap_update(os, object, propname, sizeof (uint64_t),
5743 	    ints, value, tx));
5744 
5745 	dmu_tx_commit(tx);
5746 
5747 	/*
5748 	 * Remove a random pair of entries.
5749 	 */
5750 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5751 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5752 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5753 
5754 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5755 
5756 	if (error == ENOENT)
5757 		goto out;
5758 
5759 	ASSERT0(error);
5760 
5761 	tx = dmu_tx_create(os);
5762 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5763 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5764 	if (txg == 0)
5765 		goto out;
5766 	VERIFY0(zap_remove(os, object, txgname, tx));
5767 	VERIFY0(zap_remove(os, object, propname, tx));
5768 	dmu_tx_commit(tx);
5769 out:
5770 	umem_free(od, sizeof (ztest_od_t));
5771 }
5772 
5773 /*
5774  * Test case to test the upgrading of a microzap to fatzap.
5775  */
5776 void
5777 ztest_fzap(ztest_ds_t *zd, uint64_t id)
5778 {
5779 	objset_t *os = zd->zd_os;
5780 	ztest_od_t *od;
5781 	uint64_t object, txg, value;
5782 
5783 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5784 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5785 
5786 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5787 	    !ztest_random(2)) != 0)
5788 		goto out;
5789 	object = od->od_object;
5790 
5791 	/*
5792 	 * Add entries to this ZAP and make sure it spills over
5793 	 * and gets upgraded to a fatzap. Also, since we are adding
5794 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
5795 	 */
5796 	for (value = 0; value < 2050; value++) {
5797 		char name[ZFS_MAX_DATASET_NAME_LEN];
5798 		dmu_tx_t *tx;
5799 		int error;
5800 
5801 		(void) snprintf(name, sizeof (name), "fzap-%"PRIu64"-%"PRIu64"",
5802 		    id, value);
5803 
5804 		tx = dmu_tx_create(os);
5805 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
5806 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5807 		if (txg == 0)
5808 			goto out;
5809 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
5810 		    &value, tx);
5811 		ASSERT(error == 0 || error == EEXIST);
5812 		dmu_tx_commit(tx);
5813 	}
5814 out:
5815 	umem_free(od, sizeof (ztest_od_t));
5816 }
5817 
5818 void
5819 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
5820 {
5821 	(void) id;
5822 	objset_t *os = zd->zd_os;
5823 	ztest_od_t *od;
5824 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
5825 	dmu_tx_t *tx;
5826 	int i, namelen, error;
5827 	int micro = ztest_random(2);
5828 	char name[20], string_value[20];
5829 	void *data;
5830 
5831 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5832 	ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
5833 
5834 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5835 		umem_free(od, sizeof (ztest_od_t));
5836 		return;
5837 	}
5838 
5839 	object = od->od_object;
5840 
5841 	/*
5842 	 * Generate a random name of the form 'xxx.....' where each
5843 	 * x is a random printable character and the dots are dots.
5844 	 * There are 94 such characters, and the name length goes from
5845 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
5846 	 */
5847 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
5848 
5849 	for (i = 0; i < 3; i++)
5850 		name[i] = '!' + ztest_random('~' - '!' + 1);
5851 	for (; i < namelen - 1; i++)
5852 		name[i] = '.';
5853 	name[i] = '\0';
5854 
5855 	if ((namelen & 1) || micro) {
5856 		wsize = sizeof (txg);
5857 		wc = 1;
5858 		data = &txg;
5859 	} else {
5860 		wsize = 1;
5861 		wc = namelen;
5862 		data = string_value;
5863 	}
5864 
5865 	count = -1ULL;
5866 	VERIFY0(zap_count(os, object, &count));
5867 	ASSERT3S(count, !=, -1ULL);
5868 
5869 	/*
5870 	 * Select an operation: length, lookup, add, update, remove.
5871 	 */
5872 	i = ztest_random(5);
5873 
5874 	if (i >= 2) {
5875 		tx = dmu_tx_create(os);
5876 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5877 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5878 		if (txg == 0) {
5879 			umem_free(od, sizeof (ztest_od_t));
5880 			return;
5881 		}
5882 		memcpy(string_value, name, namelen);
5883 	} else {
5884 		tx = NULL;
5885 		txg = 0;
5886 		memset(string_value, 0, namelen);
5887 	}
5888 
5889 	switch (i) {
5890 
5891 	case 0:
5892 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
5893 		if (error == 0) {
5894 			ASSERT3U(wsize, ==, zl_wsize);
5895 			ASSERT3U(wc, ==, zl_wc);
5896 		} else {
5897 			ASSERT3U(error, ==, ENOENT);
5898 		}
5899 		break;
5900 
5901 	case 1:
5902 		error = zap_lookup(os, object, name, wsize, wc, data);
5903 		if (error == 0) {
5904 			if (data == string_value &&
5905 			    memcmp(name, data, namelen) != 0)
5906 				fatal(B_FALSE, "name '%s' != val '%s' len %d",
5907 				    name, (char *)data, namelen);
5908 		} else {
5909 			ASSERT3U(error, ==, ENOENT);
5910 		}
5911 		break;
5912 
5913 	case 2:
5914 		error = zap_add(os, object, name, wsize, wc, data, tx);
5915 		ASSERT(error == 0 || error == EEXIST);
5916 		break;
5917 
5918 	case 3:
5919 		VERIFY0(zap_update(os, object, name, wsize, wc, data, tx));
5920 		break;
5921 
5922 	case 4:
5923 		error = zap_remove(os, object, name, tx);
5924 		ASSERT(error == 0 || error == ENOENT);
5925 		break;
5926 	}
5927 
5928 	if (tx != NULL)
5929 		dmu_tx_commit(tx);
5930 
5931 	umem_free(od, sizeof (ztest_od_t));
5932 }
5933 
5934 /*
5935  * Commit callback data.
5936  */
5937 typedef struct ztest_cb_data {
5938 	list_node_t		zcd_node;
5939 	uint64_t		zcd_txg;
5940 	int			zcd_expected_err;
5941 	boolean_t		zcd_added;
5942 	boolean_t		zcd_called;
5943 	spa_t			*zcd_spa;
5944 } ztest_cb_data_t;
5945 
5946 /* This is the actual commit callback function */
5947 static void
5948 ztest_commit_callback(void *arg, int error)
5949 {
5950 	ztest_cb_data_t *data = arg;
5951 	uint64_t synced_txg;
5952 
5953 	VERIFY3P(data, !=, NULL);
5954 	VERIFY3S(data->zcd_expected_err, ==, error);
5955 	VERIFY(!data->zcd_called);
5956 
5957 	synced_txg = spa_last_synced_txg(data->zcd_spa);
5958 	if (data->zcd_txg > synced_txg)
5959 		fatal(B_FALSE,
5960 		    "commit callback of txg %"PRIu64" called prematurely, "
5961 		    "last synced txg = %"PRIu64"\n",
5962 		    data->zcd_txg, synced_txg);
5963 
5964 	data->zcd_called = B_TRUE;
5965 
5966 	if (error == ECANCELED) {
5967 		ASSERT0(data->zcd_txg);
5968 		ASSERT(!data->zcd_added);
5969 
5970 		/*
5971 		 * The private callback data should be destroyed here, but
5972 		 * since we are going to check the zcd_called field after
5973 		 * dmu_tx_abort(), we will destroy it there.
5974 		 */
5975 		return;
5976 	}
5977 
5978 	ASSERT(data->zcd_added);
5979 	ASSERT3U(data->zcd_txg, !=, 0);
5980 
5981 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5982 
5983 	/* See if this cb was called more quickly */
5984 	if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
5985 		zc_min_txg_delay = synced_txg - data->zcd_txg;
5986 
5987 	/* Remove our callback from the list */
5988 	list_remove(&zcl.zcl_callbacks, data);
5989 
5990 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5991 
5992 	umem_free(data, sizeof (ztest_cb_data_t));
5993 }
5994 
5995 /* Allocate and initialize callback data structure */
5996 static ztest_cb_data_t *
5997 ztest_create_cb_data(objset_t *os, uint64_t txg)
5998 {
5999 	ztest_cb_data_t *cb_data;
6000 
6001 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
6002 
6003 	cb_data->zcd_txg = txg;
6004 	cb_data->zcd_spa = dmu_objset_spa(os);
6005 	list_link_init(&cb_data->zcd_node);
6006 
6007 	return (cb_data);
6008 }
6009 
6010 /*
6011  * Commit callback test.
6012  */
6013 void
6014 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
6015 {
6016 	objset_t *os = zd->zd_os;
6017 	ztest_od_t *od;
6018 	dmu_tx_t *tx;
6019 	ztest_cb_data_t *cb_data[3], *tmp_cb;
6020 	uint64_t old_txg, txg;
6021 	int i, error = 0;
6022 
6023 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
6024 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
6025 
6026 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
6027 		umem_free(od, sizeof (ztest_od_t));
6028 		return;
6029 	}
6030 
6031 	tx = dmu_tx_create(os);
6032 
6033 	cb_data[0] = ztest_create_cb_data(os, 0);
6034 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
6035 
6036 	dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
6037 
6038 	/* Every once in a while, abort the transaction on purpose */
6039 	if (ztest_random(100) == 0)
6040 		error = -1;
6041 
6042 	if (!error)
6043 		error = dmu_tx_assign(tx, TXG_NOWAIT);
6044 
6045 	txg = error ? 0 : dmu_tx_get_txg(tx);
6046 
6047 	cb_data[0]->zcd_txg = txg;
6048 	cb_data[1] = ztest_create_cb_data(os, txg);
6049 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
6050 
6051 	if (error) {
6052 		/*
6053 		 * It's not a strict requirement to call the registered
6054 		 * callbacks from inside dmu_tx_abort(), but that's what
6055 		 * it's supposed to happen in the current implementation
6056 		 * so we will check for that.
6057 		 */
6058 		for (i = 0; i < 2; i++) {
6059 			cb_data[i]->zcd_expected_err = ECANCELED;
6060 			VERIFY(!cb_data[i]->zcd_called);
6061 		}
6062 
6063 		dmu_tx_abort(tx);
6064 
6065 		for (i = 0; i < 2; i++) {
6066 			VERIFY(cb_data[i]->zcd_called);
6067 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
6068 		}
6069 
6070 		umem_free(od, sizeof (ztest_od_t));
6071 		return;
6072 	}
6073 
6074 	cb_data[2] = ztest_create_cb_data(os, txg);
6075 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
6076 
6077 	/*
6078 	 * Read existing data to make sure there isn't a future leak.
6079 	 */
6080 	VERIFY0(dmu_read(os, od->od_object, 0, sizeof (uint64_t),
6081 	    &old_txg, DMU_READ_PREFETCH));
6082 
6083 	if (old_txg > txg)
6084 		fatal(B_FALSE,
6085 		    "future leak: got %"PRIu64", open txg is %"PRIu64"",
6086 		    old_txg, txg);
6087 
6088 	dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
6089 
6090 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
6091 
6092 	/*
6093 	 * Since commit callbacks don't have any ordering requirement and since
6094 	 * it is theoretically possible for a commit callback to be called
6095 	 * after an arbitrary amount of time has elapsed since its txg has been
6096 	 * synced, it is difficult to reliably determine whether a commit
6097 	 * callback hasn't been called due to high load or due to a flawed
6098 	 * implementation.
6099 	 *
6100 	 * In practice, we will assume that if after a certain number of txgs a
6101 	 * commit callback hasn't been called, then most likely there's an
6102 	 * implementation bug..
6103 	 */
6104 	tmp_cb = list_head(&zcl.zcl_callbacks);
6105 	if (tmp_cb != NULL &&
6106 	    tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
6107 		fatal(B_FALSE,
6108 		    "Commit callback threshold exceeded, "
6109 		    "oldest txg: %"PRIu64", open txg: %"PRIu64"\n",
6110 		    tmp_cb->zcd_txg, txg);
6111 	}
6112 
6113 	/*
6114 	 * Let's find the place to insert our callbacks.
6115 	 *
6116 	 * Even though the list is ordered by txg, it is possible for the
6117 	 * insertion point to not be the end because our txg may already be
6118 	 * quiescing at this point and other callbacks in the open txg
6119 	 * (from other objsets) may have sneaked in.
6120 	 */
6121 	tmp_cb = list_tail(&zcl.zcl_callbacks);
6122 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
6123 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
6124 
6125 	/* Add the 3 callbacks to the list */
6126 	for (i = 0; i < 3; i++) {
6127 		if (tmp_cb == NULL)
6128 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
6129 		else
6130 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
6131 			    cb_data[i]);
6132 
6133 		cb_data[i]->zcd_added = B_TRUE;
6134 		VERIFY(!cb_data[i]->zcd_called);
6135 
6136 		tmp_cb = cb_data[i];
6137 	}
6138 
6139 	zc_cb_counter += 3;
6140 
6141 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
6142 
6143 	dmu_tx_commit(tx);
6144 
6145 	umem_free(od, sizeof (ztest_od_t));
6146 }
6147 
6148 /*
6149  * Visit each object in the dataset. Verify that its properties
6150  * are consistent what was stored in the block tag when it was created,
6151  * and that its unused bonus buffer space has not been overwritten.
6152  */
6153 void
6154 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
6155 {
6156 	(void) id;
6157 	objset_t *os = zd->zd_os;
6158 	uint64_t obj;
6159 	int err = 0;
6160 
6161 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
6162 		ztest_block_tag_t *bt = NULL;
6163 		dmu_object_info_t doi;
6164 		dmu_buf_t *db;
6165 
6166 		ztest_object_lock(zd, obj, ZTRL_READER);
6167 		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) {
6168 			ztest_object_unlock(zd, obj);
6169 			continue;
6170 		}
6171 
6172 		dmu_object_info_from_db(db, &doi);
6173 		if (doi.doi_bonus_size >= sizeof (*bt))
6174 			bt = ztest_bt_bonus(db);
6175 
6176 		if (bt && bt->bt_magic == BT_MAGIC) {
6177 			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
6178 			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
6179 			    bt->bt_crtxg);
6180 			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
6181 		}
6182 
6183 		dmu_buf_rele(db, FTAG);
6184 		ztest_object_unlock(zd, obj);
6185 	}
6186 }
6187 
6188 void
6189 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
6190 {
6191 	(void) id;
6192 	zfs_prop_t proplist[] = {
6193 		ZFS_PROP_CHECKSUM,
6194 		ZFS_PROP_COMPRESSION,
6195 		ZFS_PROP_COPIES,
6196 		ZFS_PROP_DEDUP
6197 	};
6198 
6199 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6200 
6201 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) {
6202 		int error = ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
6203 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
6204 		ASSERT(error == 0 || error == ENOSPC);
6205 	}
6206 
6207 	int error = ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
6208 	    ztest_random_blocksize(), (int)ztest_random(2));
6209 	ASSERT(error == 0 || error == ENOSPC);
6210 
6211 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6212 }
6213 
6214 void
6215 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
6216 {
6217 	(void) zd, (void) id;
6218 
6219 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6220 
6221 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_AUTOTRIM, ztest_random(2));
6222 
6223 	nvlist_t *props = fnvlist_alloc();
6224 
6225 	VERIFY0(spa_prop_get(ztest_spa, props));
6226 
6227 	if (ztest_opts.zo_verbose >= 6)
6228 		dump_nvlist(props, 4);
6229 
6230 	fnvlist_free(props);
6231 
6232 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6233 }
6234 
6235 static int
6236 user_release_one(const char *snapname, const char *holdname)
6237 {
6238 	nvlist_t *snaps, *holds;
6239 	int error;
6240 
6241 	snaps = fnvlist_alloc();
6242 	holds = fnvlist_alloc();
6243 	fnvlist_add_boolean(holds, holdname);
6244 	fnvlist_add_nvlist(snaps, snapname, holds);
6245 	fnvlist_free(holds);
6246 	error = dsl_dataset_user_release(snaps, NULL);
6247 	fnvlist_free(snaps);
6248 	return (error);
6249 }
6250 
6251 /*
6252  * Test snapshot hold/release and deferred destroy.
6253  */
6254 void
6255 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
6256 {
6257 	int error;
6258 	objset_t *os = zd->zd_os;
6259 	objset_t *origin;
6260 	char snapname[100];
6261 	char fullname[100];
6262 	char clonename[100];
6263 	char tag[100];
6264 	char osname[ZFS_MAX_DATASET_NAME_LEN];
6265 	nvlist_t *holds;
6266 
6267 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6268 
6269 	dmu_objset_name(os, osname);
6270 
6271 	(void) snprintf(snapname, sizeof (snapname), "sh1_%"PRIu64"", id);
6272 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
6273 	(void) snprintf(clonename, sizeof (clonename), "%s/ch1_%"PRIu64"",
6274 	    osname, id);
6275 	(void) snprintf(tag, sizeof (tag), "tag_%"PRIu64"", id);
6276 
6277 	/*
6278 	 * Clean up from any previous run.
6279 	 */
6280 	error = dsl_destroy_head(clonename);
6281 	if (error != ENOENT)
6282 		ASSERT0(error);
6283 	error = user_release_one(fullname, tag);
6284 	if (error != ESRCH && error != ENOENT)
6285 		ASSERT0(error);
6286 	error = dsl_destroy_snapshot(fullname, B_FALSE);
6287 	if (error != ENOENT)
6288 		ASSERT0(error);
6289 
6290 	/*
6291 	 * Create snapshot, clone it, mark snap for deferred destroy,
6292 	 * destroy clone, verify snap was also destroyed.
6293 	 */
6294 	error = dmu_objset_snapshot_one(osname, snapname);
6295 	if (error) {
6296 		if (error == ENOSPC) {
6297 			ztest_record_enospc("dmu_objset_snapshot");
6298 			goto out;
6299 		}
6300 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
6301 	}
6302 
6303 	error = dmu_objset_clone(clonename, fullname);
6304 	if (error) {
6305 		if (error == ENOSPC) {
6306 			ztest_record_enospc("dmu_objset_clone");
6307 			goto out;
6308 		}
6309 		fatal(B_FALSE, "dmu_objset_clone(%s) = %d", clonename, error);
6310 	}
6311 
6312 	error = dsl_destroy_snapshot(fullname, B_TRUE);
6313 	if (error) {
6314 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
6315 		    fullname, error);
6316 	}
6317 
6318 	error = dsl_destroy_head(clonename);
6319 	if (error)
6320 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clonename, error);
6321 
6322 	error = dmu_objset_hold(fullname, FTAG, &origin);
6323 	if (error != ENOENT)
6324 		fatal(B_FALSE, "dmu_objset_hold(%s) = %d", fullname, error);
6325 
6326 	/*
6327 	 * Create snapshot, add temporary hold, verify that we can't
6328 	 * destroy a held snapshot, mark for deferred destroy,
6329 	 * release hold, verify snapshot was destroyed.
6330 	 */
6331 	error = dmu_objset_snapshot_one(osname, snapname);
6332 	if (error) {
6333 		if (error == ENOSPC) {
6334 			ztest_record_enospc("dmu_objset_snapshot");
6335 			goto out;
6336 		}
6337 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
6338 	}
6339 
6340 	holds = fnvlist_alloc();
6341 	fnvlist_add_string(holds, fullname, tag);
6342 	error = dsl_dataset_user_hold(holds, 0, NULL);
6343 	fnvlist_free(holds);
6344 
6345 	if (error == ENOSPC) {
6346 		ztest_record_enospc("dsl_dataset_user_hold");
6347 		goto out;
6348 	} else if (error) {
6349 		fatal(B_FALSE, "dsl_dataset_user_hold(%s, %s) = %u",
6350 		    fullname, tag, error);
6351 	}
6352 
6353 	error = dsl_destroy_snapshot(fullname, B_FALSE);
6354 	if (error != EBUSY) {
6355 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
6356 		    fullname, error);
6357 	}
6358 
6359 	error = dsl_destroy_snapshot(fullname, B_TRUE);
6360 	if (error) {
6361 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
6362 		    fullname, error);
6363 	}
6364 
6365 	error = user_release_one(fullname, tag);
6366 	if (error)
6367 		fatal(B_FALSE, "user_release_one(%s, %s) = %d",
6368 		    fullname, tag, error);
6369 
6370 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
6371 
6372 out:
6373 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6374 }
6375 
6376 /*
6377  * Inject random faults into the on-disk data.
6378  */
6379 void
6380 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
6381 {
6382 	(void) zd, (void) id;
6383 	ztest_shared_t *zs = ztest_shared;
6384 	spa_t *spa = ztest_spa;
6385 	int fd;
6386 	uint64_t offset;
6387 	uint64_t leaves;
6388 	uint64_t bad = 0x1990c0ffeedecadeull;
6389 	uint64_t top, leaf;
6390 	uint64_t raidz_children;
6391 	char *path0;
6392 	char *pathrand;
6393 	size_t fsize;
6394 	int bshift = SPA_MAXBLOCKSHIFT + 2;
6395 	int iters = 1000;
6396 	int maxfaults;
6397 	int mirror_save;
6398 	vdev_t *vd0 = NULL;
6399 	uint64_t guid0 = 0;
6400 	boolean_t islog = B_FALSE;
6401 	boolean_t injected = B_FALSE;
6402 
6403 	path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6404 	pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6405 
6406 	mutex_enter(&ztest_vdev_lock);
6407 
6408 	/*
6409 	 * Device removal is in progress, fault injection must be disabled
6410 	 * until it completes and the pool is scrubbed.  The fault injection
6411 	 * strategy for damaging blocks does not take in to account evacuated
6412 	 * blocks which may have already been damaged.
6413 	 */
6414 	if (ztest_device_removal_active)
6415 		goto out;
6416 
6417 	/*
6418 	 * The fault injection strategy for damaging blocks cannot be used
6419 	 * if raidz expansion is in progress. The leaves value
6420 	 * (attached raidz children) is variable and strategy for damaging
6421 	 * blocks will corrupt same data blocks on different child vdevs
6422 	 * because of the reflow process.
6423 	 */
6424 	if (spa->spa_raidz_expand != NULL)
6425 		goto out;
6426 
6427 	maxfaults = MAXFAULTS(zs);
6428 	raidz_children = ztest_get_raidz_children(spa);
6429 	leaves = MAX(zs->zs_mirrors, 1) * raidz_children;
6430 	mirror_save = zs->zs_mirrors;
6431 
6432 	ASSERT3U(leaves, >=, 1);
6433 
6434 	/*
6435 	 * While ztest is running the number of leaves will not change.  This
6436 	 * is critical for the fault injection logic as it determines where
6437 	 * errors can be safely injected such that they are always repairable.
6438 	 *
6439 	 * When restarting ztest a different number of leaves may be requested
6440 	 * which will shift the regions to be damaged.  This is fine as long
6441 	 * as the pool has been scrubbed prior to using the new mapping.
6442 	 * Failure to do can result in non-repairable damage being injected.
6443 	 */
6444 	if (ztest_pool_scrubbed == B_FALSE)
6445 		goto out;
6446 
6447 	/*
6448 	 * Grab the name lock as reader. There are some operations
6449 	 * which don't like to have their vdevs changed while
6450 	 * they are in progress (i.e. spa_change_guid). Those
6451 	 * operations will have grabbed the name lock as writer.
6452 	 */
6453 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6454 
6455 	/*
6456 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
6457 	 */
6458 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6459 
6460 	if (ztest_random(2) == 0) {
6461 		/*
6462 		 * Inject errors on a normal data device or slog device.
6463 		 */
6464 		top = ztest_random_vdev_top(spa, B_TRUE);
6465 		leaf = ztest_random(leaves) + zs->zs_splits;
6466 
6467 		/*
6468 		 * Generate paths to the first leaf in this top-level vdev,
6469 		 * and to the random leaf we selected.  We'll induce transient
6470 		 * write failures and random online/offline activity on leaf 0,
6471 		 * and we'll write random garbage to the randomly chosen leaf.
6472 		 */
6473 		(void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
6474 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6475 		    top * leaves + zs->zs_splits);
6476 		(void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
6477 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6478 		    top * leaves + leaf);
6479 
6480 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
6481 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
6482 			islog = B_TRUE;
6483 
6484 		/*
6485 		 * If the top-level vdev needs to be resilvered
6486 		 * then we only allow faults on the device that is
6487 		 * resilvering.
6488 		 */
6489 		if (vd0 != NULL && maxfaults != 1 &&
6490 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
6491 		    vd0->vdev_resilver_txg != 0)) {
6492 			/*
6493 			 * Make vd0 explicitly claim to be unreadable,
6494 			 * or unwritable, or reach behind its back
6495 			 * and close the underlying fd.  We can do this if
6496 			 * maxfaults == 0 because we'll fail and reexecute,
6497 			 * and we can do it if maxfaults >= 2 because we'll
6498 			 * have enough redundancy.  If maxfaults == 1, the
6499 			 * combination of this with injection of random data
6500 			 * corruption below exceeds the pool's fault tolerance.
6501 			 */
6502 			vdev_file_t *vf = vd0->vdev_tsd;
6503 
6504 			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
6505 			    (long long)vd0->vdev_id, (int)maxfaults);
6506 
6507 			if (vf != NULL && ztest_random(3) == 0) {
6508 				(void) close(vf->vf_file->f_fd);
6509 				vf->vf_file->f_fd = -1;
6510 			} else if (ztest_random(2) == 0) {
6511 				vd0->vdev_cant_read = B_TRUE;
6512 			} else {
6513 				vd0->vdev_cant_write = B_TRUE;
6514 			}
6515 			guid0 = vd0->vdev_guid;
6516 		}
6517 	} else {
6518 		/*
6519 		 * Inject errors on an l2cache device.
6520 		 */
6521 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
6522 
6523 		if (sav->sav_count == 0) {
6524 			spa_config_exit(spa, SCL_STATE, FTAG);
6525 			(void) pthread_rwlock_unlock(&ztest_name_lock);
6526 			goto out;
6527 		}
6528 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
6529 		guid0 = vd0->vdev_guid;
6530 		(void) strlcpy(path0, vd0->vdev_path, MAXPATHLEN);
6531 		(void) strlcpy(pathrand, vd0->vdev_path, MAXPATHLEN);
6532 
6533 		leaf = 0;
6534 		leaves = 1;
6535 		maxfaults = INT_MAX;	/* no limit on cache devices */
6536 	}
6537 
6538 	spa_config_exit(spa, SCL_STATE, FTAG);
6539 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6540 
6541 	/*
6542 	 * If we can tolerate two or more faults, or we're dealing
6543 	 * with a slog, randomly online/offline vd0.
6544 	 */
6545 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
6546 		if (ztest_random(10) < 6) {
6547 			int flags = (ztest_random(2) == 0 ?
6548 			    ZFS_OFFLINE_TEMPORARY : 0);
6549 
6550 			/*
6551 			 * We have to grab the zs_name_lock as writer to
6552 			 * prevent a race between offlining a slog and
6553 			 * destroying a dataset. Offlining the slog will
6554 			 * grab a reference on the dataset which may cause
6555 			 * dsl_destroy_head() to fail with EBUSY thus
6556 			 * leaving the dataset in an inconsistent state.
6557 			 */
6558 			if (islog)
6559 				(void) pthread_rwlock_wrlock(&ztest_name_lock);
6560 
6561 			VERIFY3U(vdev_offline(spa, guid0, flags), !=, EBUSY);
6562 
6563 			if (islog)
6564 				(void) pthread_rwlock_unlock(&ztest_name_lock);
6565 		} else {
6566 			/*
6567 			 * Ideally we would like to be able to randomly
6568 			 * call vdev_[on|off]line without holding locks
6569 			 * to force unpredictable failures but the side
6570 			 * effects of vdev_[on|off]line prevent us from
6571 			 * doing so.
6572 			 */
6573 			(void) vdev_online(spa, guid0, 0, NULL);
6574 		}
6575 	}
6576 
6577 	if (maxfaults == 0)
6578 		goto out;
6579 
6580 	/*
6581 	 * We have at least single-fault tolerance, so inject data corruption.
6582 	 */
6583 	fd = open(pathrand, O_RDWR);
6584 
6585 	if (fd == -1) /* we hit a gap in the device namespace */
6586 		goto out;
6587 
6588 	fsize = lseek(fd, 0, SEEK_END);
6589 
6590 	while (--iters != 0) {
6591 		/*
6592 		 * The offset must be chosen carefully to ensure that
6593 		 * we do not inject a given logical block with errors
6594 		 * on two different leaf devices, because ZFS can not
6595 		 * tolerate that (if maxfaults==1).
6596 		 *
6597 		 * To achieve this we divide each leaf device into
6598 		 * chunks of size (# leaves * SPA_MAXBLOCKSIZE * 4).
6599 		 * Each chunk is further divided into error-injection
6600 		 * ranges (can accept errors) and clear ranges (we do
6601 		 * not inject errors in those). Each error-injection
6602 		 * range can accept errors only for a single leaf vdev.
6603 		 * Error-injection ranges are separated by clear ranges.
6604 		 *
6605 		 * For example, with 3 leaves, each chunk looks like:
6606 		 *    0 to  32M: injection range for leaf 0
6607 		 *  32M to  64M: clear range - no injection allowed
6608 		 *  64M to  96M: injection range for leaf 1
6609 		 *  96M to 128M: clear range - no injection allowed
6610 		 * 128M to 160M: injection range for leaf 2
6611 		 * 160M to 192M: clear range - no injection allowed
6612 		 *
6613 		 * Each clear range must be large enough such that a
6614 		 * single block cannot straddle it. This way a block
6615 		 * can't be a target in two different injection ranges
6616 		 * (on different leaf vdevs).
6617 		 */
6618 		offset = ztest_random(fsize / (leaves << bshift)) *
6619 		    (leaves << bshift) + (leaf << bshift) +
6620 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
6621 
6622 		/*
6623 		 * Only allow damage to the labels at one end of the vdev.
6624 		 *
6625 		 * If all labels are damaged, the device will be totally
6626 		 * inaccessible, which will result in loss of data,
6627 		 * because we also damage (parts of) the other side of
6628 		 * the mirror/raidz.
6629 		 *
6630 		 * Additionally, we will always have both an even and an
6631 		 * odd label, so that we can handle crashes in the
6632 		 * middle of vdev_config_sync().
6633 		 */
6634 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
6635 			continue;
6636 
6637 		/*
6638 		 * The two end labels are stored at the "end" of the disk, but
6639 		 * the end of the disk (vdev_psize) is aligned to
6640 		 * sizeof (vdev_label_t).
6641 		 */
6642 		uint64_t psize = P2ALIGN_TYPED(fsize, sizeof (vdev_label_t),
6643 		    uint64_t);
6644 		if ((leaf & 1) == 1 &&
6645 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
6646 			continue;
6647 
6648 		if (mirror_save != zs->zs_mirrors) {
6649 			(void) close(fd);
6650 			goto out;
6651 		}
6652 
6653 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
6654 			fatal(B_TRUE,
6655 			    "can't inject bad word at 0x%"PRIx64" in %s",
6656 			    offset, pathrand);
6657 
6658 		if (ztest_opts.zo_verbose >= 7)
6659 			(void) printf("injected bad word into %s,"
6660 			    " offset 0x%"PRIx64"\n", pathrand, offset);
6661 
6662 		injected = B_TRUE;
6663 	}
6664 
6665 	(void) close(fd);
6666 out:
6667 	mutex_exit(&ztest_vdev_lock);
6668 
6669 	if (injected && ztest_opts.zo_raid_do_expand) {
6670 		int error = spa_scan(spa, POOL_SCAN_SCRUB);
6671 		if (error == 0) {
6672 			while (dsl_scan_scrubbing(spa_get_dsl(spa)))
6673 				txg_wait_synced(spa_get_dsl(spa), 0);
6674 		}
6675 	}
6676 
6677 	umem_free(path0, MAXPATHLEN);
6678 	umem_free(pathrand, MAXPATHLEN);
6679 }
6680 
6681 /*
6682  * By design ztest will never inject uncorrectable damage in to the pool.
6683  * Issue a scrub, wait for it to complete, and verify there is never any
6684  * persistent damage.
6685  *
6686  * Only after a full scrub has been completed is it safe to start injecting
6687  * data corruption.  See the comment in zfs_fault_inject().
6688  */
6689 static int
6690 ztest_scrub_impl(spa_t *spa)
6691 {
6692 	int error = spa_scan(spa, POOL_SCAN_SCRUB);
6693 	if (error)
6694 		return (error);
6695 
6696 	while (dsl_scan_scrubbing(spa_get_dsl(spa)))
6697 		txg_wait_synced(spa_get_dsl(spa), 0);
6698 
6699 	if (spa_approx_errlog_size(spa) > 0)
6700 		return (ECKSUM);
6701 
6702 	ztest_pool_scrubbed = B_TRUE;
6703 
6704 	return (0);
6705 }
6706 
6707 /*
6708  * Scrub the pool.
6709  */
6710 void
6711 ztest_scrub(ztest_ds_t *zd, uint64_t id)
6712 {
6713 	(void) zd, (void) id;
6714 	spa_t *spa = ztest_spa;
6715 	int error;
6716 
6717 	/*
6718 	 * Scrub in progress by device removal.
6719 	 */
6720 	if (ztest_device_removal_active)
6721 		return;
6722 
6723 	/*
6724 	 * Start a scrub, wait a moment, then force a restart.
6725 	 */
6726 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
6727 	(void) poll(NULL, 0, 100);
6728 
6729 	error = ztest_scrub_impl(spa);
6730 	if (error == EBUSY)
6731 		error = 0;
6732 	ASSERT0(error);
6733 }
6734 
6735 /*
6736  * Change the guid for the pool.
6737  */
6738 void
6739 ztest_reguid(ztest_ds_t *zd, uint64_t id)
6740 {
6741 	(void) zd, (void) id;
6742 	spa_t *spa = ztest_spa;
6743 	uint64_t orig, load;
6744 	int error;
6745 	ztest_shared_t *zs = ztest_shared;
6746 
6747 	if (ztest_opts.zo_mmp_test)
6748 		return;
6749 
6750 	orig = spa_guid(spa);
6751 	load = spa_load_guid(spa);
6752 
6753 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
6754 	error = spa_change_guid(spa, NULL);
6755 	zs->zs_guid = spa_guid(spa);
6756 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6757 
6758 	if (error != 0)
6759 		return;
6760 
6761 	if (ztest_opts.zo_verbose >= 4) {
6762 		(void) printf("Changed guid old %"PRIu64" -> %"PRIu64"\n",
6763 		    orig, spa_guid(spa));
6764 	}
6765 
6766 	VERIFY3U(orig, !=, spa_guid(spa));
6767 	VERIFY3U(load, ==, spa_load_guid(spa));
6768 }
6769 
6770 void
6771 ztest_blake3(ztest_ds_t *zd, uint64_t id)
6772 {
6773 	(void) zd, (void) id;
6774 	hrtime_t end = gethrtime() + NANOSEC;
6775 	zio_cksum_salt_t salt;
6776 	void *salt_ptr = &salt.zcs_bytes;
6777 	struct abd *abd_data, *abd_meta;
6778 	void *buf, *templ;
6779 	int i, *ptr;
6780 	uint32_t size;
6781 	BLAKE3_CTX ctx;
6782 	const zfs_impl_t *blake3 = zfs_impl_get_ops("blake3");
6783 
6784 	size = ztest_random_blocksize();
6785 	buf = umem_alloc(size, UMEM_NOFAIL);
6786 	abd_data = abd_alloc(size, B_FALSE);
6787 	abd_meta = abd_alloc(size, B_TRUE);
6788 
6789 	for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6790 		*ptr = ztest_random(UINT_MAX);
6791 	memset(salt_ptr, 'A', 32);
6792 
6793 	abd_copy_from_buf_off(abd_data, buf, 0, size);
6794 	abd_copy_from_buf_off(abd_meta, buf, 0, size);
6795 
6796 	while (gethrtime() <= end) {
6797 		int run_count = 100;
6798 		zio_cksum_t zc_ref1, zc_ref2;
6799 		zio_cksum_t zc_res1, zc_res2;
6800 
6801 		void *ref1 = &zc_ref1;
6802 		void *ref2 = &zc_ref2;
6803 		void *res1 = &zc_res1;
6804 		void *res2 = &zc_res2;
6805 
6806 		/* BLAKE3_KEY_LEN = 32 */
6807 		VERIFY0(blake3->setname("generic"));
6808 		templ = abd_checksum_blake3_tmpl_init(&salt);
6809 		Blake3_InitKeyed(&ctx, salt_ptr);
6810 		Blake3_Update(&ctx, buf, size);
6811 		Blake3_Final(&ctx, ref1);
6812 		zc_ref2 = zc_ref1;
6813 		ZIO_CHECKSUM_BSWAP(&zc_ref2);
6814 		abd_checksum_blake3_tmpl_free(templ);
6815 
6816 		VERIFY0(blake3->setname("cycle"));
6817 		while (run_count-- > 0) {
6818 
6819 			/* Test current implementation */
6820 			Blake3_InitKeyed(&ctx, salt_ptr);
6821 			Blake3_Update(&ctx, buf, size);
6822 			Blake3_Final(&ctx, res1);
6823 			zc_res2 = zc_res1;
6824 			ZIO_CHECKSUM_BSWAP(&zc_res2);
6825 
6826 			VERIFY0(memcmp(ref1, res1, 32));
6827 			VERIFY0(memcmp(ref2, res2, 32));
6828 
6829 			/* Test ABD - data */
6830 			templ = abd_checksum_blake3_tmpl_init(&salt);
6831 			abd_checksum_blake3_native(abd_data, size,
6832 			    templ, &zc_res1);
6833 			abd_checksum_blake3_byteswap(abd_data, size,
6834 			    templ, &zc_res2);
6835 
6836 			VERIFY0(memcmp(ref1, res1, 32));
6837 			VERIFY0(memcmp(ref2, res2, 32));
6838 
6839 			/* Test ABD - metadata */
6840 			abd_checksum_blake3_native(abd_meta, size,
6841 			    templ, &zc_res1);
6842 			abd_checksum_blake3_byteswap(abd_meta, size,
6843 			    templ, &zc_res2);
6844 			abd_checksum_blake3_tmpl_free(templ);
6845 
6846 			VERIFY0(memcmp(ref1, res1, 32));
6847 			VERIFY0(memcmp(ref2, res2, 32));
6848 
6849 		}
6850 	}
6851 
6852 	abd_free(abd_data);
6853 	abd_free(abd_meta);
6854 	umem_free(buf, size);
6855 }
6856 
6857 void
6858 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
6859 {
6860 	(void) zd, (void) id;
6861 	hrtime_t end = gethrtime() + NANOSEC;
6862 
6863 	while (gethrtime() <= end) {
6864 		int run_count = 100;
6865 		void *buf;
6866 		struct abd *abd_data, *abd_meta;
6867 		uint32_t size;
6868 		int *ptr;
6869 		int i;
6870 		zio_cksum_t zc_ref;
6871 		zio_cksum_t zc_ref_byteswap;
6872 
6873 		size = ztest_random_blocksize();
6874 
6875 		buf = umem_alloc(size, UMEM_NOFAIL);
6876 		abd_data = abd_alloc(size, B_FALSE);
6877 		abd_meta = abd_alloc(size, B_TRUE);
6878 
6879 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6880 			*ptr = ztest_random(UINT_MAX);
6881 
6882 		abd_copy_from_buf_off(abd_data, buf, 0, size);
6883 		abd_copy_from_buf_off(abd_meta, buf, 0, size);
6884 
6885 		VERIFY0(fletcher_4_impl_set("scalar"));
6886 		fletcher_4_native(buf, size, NULL, &zc_ref);
6887 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
6888 
6889 		VERIFY0(fletcher_4_impl_set("cycle"));
6890 		while (run_count-- > 0) {
6891 			zio_cksum_t zc;
6892 			zio_cksum_t zc_byteswap;
6893 
6894 			fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
6895 			fletcher_4_native(buf, size, NULL, &zc);
6896 
6897 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6898 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6899 			    sizeof (zc_byteswap)));
6900 
6901 			/* Test ABD - data */
6902 			abd_fletcher_4_byteswap(abd_data, size, NULL,
6903 			    &zc_byteswap);
6904 			abd_fletcher_4_native(abd_data, size, NULL, &zc);
6905 
6906 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6907 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6908 			    sizeof (zc_byteswap)));
6909 
6910 			/* Test ABD - metadata */
6911 			abd_fletcher_4_byteswap(abd_meta, size, NULL,
6912 			    &zc_byteswap);
6913 			abd_fletcher_4_native(abd_meta, size, NULL, &zc);
6914 
6915 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6916 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6917 			    sizeof (zc_byteswap)));
6918 
6919 		}
6920 
6921 		umem_free(buf, size);
6922 		abd_free(abd_data);
6923 		abd_free(abd_meta);
6924 	}
6925 }
6926 
6927 void
6928 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
6929 {
6930 	(void) zd, (void) id;
6931 	void *buf;
6932 	size_t size;
6933 	int *ptr;
6934 	int i;
6935 	zio_cksum_t zc_ref;
6936 	zio_cksum_t zc_ref_bswap;
6937 
6938 	hrtime_t end = gethrtime() + NANOSEC;
6939 
6940 	while (gethrtime() <= end) {
6941 		int run_count = 100;
6942 
6943 		size = ztest_random_blocksize();
6944 		buf = umem_alloc(size, UMEM_NOFAIL);
6945 
6946 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6947 			*ptr = ztest_random(UINT_MAX);
6948 
6949 		VERIFY0(fletcher_4_impl_set("scalar"));
6950 		fletcher_4_native(buf, size, NULL, &zc_ref);
6951 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
6952 
6953 		VERIFY0(fletcher_4_impl_set("cycle"));
6954 
6955 		while (run_count-- > 0) {
6956 			zio_cksum_t zc;
6957 			zio_cksum_t zc_bswap;
6958 			size_t pos = 0;
6959 
6960 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6961 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6962 
6963 			while (pos < size) {
6964 				size_t inc = 64 * ztest_random(size / 67);
6965 				/* sometimes add few bytes to test non-simd */
6966 				if (ztest_random(100) < 10)
6967 					inc += P2ALIGN_TYPED(ztest_random(64),
6968 					    sizeof (uint32_t), uint64_t);
6969 
6970 				if (inc > (size - pos))
6971 					inc = size - pos;
6972 
6973 				fletcher_4_incremental_native(buf + pos, inc,
6974 				    &zc);
6975 				fletcher_4_incremental_byteswap(buf + pos, inc,
6976 				    &zc_bswap);
6977 
6978 				pos += inc;
6979 			}
6980 
6981 			VERIFY3U(pos, ==, size);
6982 
6983 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6984 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6985 
6986 			/*
6987 			 * verify if incremental on the whole buffer is
6988 			 * equivalent to non-incremental version
6989 			 */
6990 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6991 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6992 
6993 			fletcher_4_incremental_native(buf, size, &zc);
6994 			fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
6995 
6996 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6997 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6998 		}
6999 
7000 		umem_free(buf, size);
7001 	}
7002 }
7003 
7004 void
7005 ztest_pool_prefetch_ddt(ztest_ds_t *zd, uint64_t id)
7006 {
7007 	(void) zd, (void) id;
7008 	spa_t *spa;
7009 
7010 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
7011 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7012 
7013 	ddt_prefetch_all(spa);
7014 
7015 	spa_close(spa, FTAG);
7016 	(void) pthread_rwlock_unlock(&ztest_name_lock);
7017 }
7018 
7019 static int
7020 ztest_set_global_vars(void)
7021 {
7022 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
7023 		char *kv = ztest_opts.zo_gvars[i];
7024 		VERIFY3U(strlen(kv), <=, ZO_GVARS_MAX_ARGLEN);
7025 		VERIFY3U(strlen(kv), >, 0);
7026 		int err = set_global_var(kv);
7027 		if (ztest_opts.zo_verbose > 0) {
7028 			(void) printf("setting global var %s ... %s\n", kv,
7029 			    err ? "failed" : "ok");
7030 		}
7031 		if (err != 0) {
7032 			(void) fprintf(stderr,
7033 			    "failed to set global var '%s'\n", kv);
7034 			return (err);
7035 		}
7036 	}
7037 	return (0);
7038 }
7039 
7040 static char **
7041 ztest_global_vars_to_zdb_args(void)
7042 {
7043 	char **args = calloc(2*ztest_opts.zo_gvars_count + 1, sizeof (char *));
7044 	char **cur = args;
7045 	if (args == NULL)
7046 		return (NULL);
7047 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
7048 		*cur++ = (char *)"-o";
7049 		*cur++ = ztest_opts.zo_gvars[i];
7050 	}
7051 	ASSERT3P(cur, ==, &args[2*ztest_opts.zo_gvars_count]);
7052 	*cur = NULL;
7053 	return (args);
7054 }
7055 
7056 /* The end of strings is indicated by a NULL element */
7057 static char *
7058 join_strings(char **strings, const char *sep)
7059 {
7060 	size_t totallen = 0;
7061 	for (char **sp = strings; *sp != NULL; sp++) {
7062 		totallen += strlen(*sp);
7063 		totallen += strlen(sep);
7064 	}
7065 	if (totallen > 0) {
7066 		ASSERT(totallen >= strlen(sep));
7067 		totallen -= strlen(sep);
7068 	}
7069 
7070 	size_t buflen = totallen + 1;
7071 	char *o = umem_alloc(buflen, UMEM_NOFAIL); /* trailing 0 byte */
7072 	o[0] = '\0';
7073 	for (char **sp = strings; *sp != NULL; sp++) {
7074 		size_t would;
7075 		would = strlcat(o, *sp, buflen);
7076 		VERIFY3U(would, <, buflen);
7077 		if (*(sp+1) == NULL) {
7078 			break;
7079 		}
7080 		would = strlcat(o, sep, buflen);
7081 		VERIFY3U(would, <, buflen);
7082 	}
7083 	ASSERT3S(strlen(o), ==, totallen);
7084 	return (o);
7085 }
7086 
7087 static int
7088 ztest_check_path(char *path)
7089 {
7090 	struct stat s;
7091 	/* return true on success */
7092 	return (!stat(path, &s));
7093 }
7094 
7095 static void
7096 ztest_get_zdb_bin(char *bin, int len)
7097 {
7098 	char *zdb_path;
7099 	/*
7100 	 * Try to use $ZDB and in-tree zdb path. If not successful, just
7101 	 * let popen to search through PATH.
7102 	 */
7103 	if ((zdb_path = getenv("ZDB"))) {
7104 		strlcpy(bin, zdb_path, len); /* In env */
7105 		if (!ztest_check_path(bin)) {
7106 			ztest_dump_core = 0;
7107 			fatal(B_TRUE, "invalid ZDB '%s'", bin);
7108 		}
7109 		return;
7110 	}
7111 
7112 	VERIFY3P(realpath(getexecname(), bin), !=, NULL);
7113 	if (strstr(bin, ".libs/ztest")) {
7114 		strstr(bin, ".libs/ztest")[0] = '\0'; /* In-tree */
7115 		strcat(bin, "zdb");
7116 		if (ztest_check_path(bin))
7117 			return;
7118 	}
7119 	strcpy(bin, "zdb");
7120 }
7121 
7122 static vdev_t *
7123 ztest_random_concrete_vdev_leaf(vdev_t *vd)
7124 {
7125 	if (vd == NULL)
7126 		return (NULL);
7127 
7128 	if (vd->vdev_children == 0)
7129 		return (vd);
7130 
7131 	vdev_t *eligible[vd->vdev_children];
7132 	int eligible_idx = 0, i;
7133 	for (i = 0; i < vd->vdev_children; i++) {
7134 		vdev_t *cvd = vd->vdev_child[i];
7135 		if (cvd->vdev_top->vdev_removing)
7136 			continue;
7137 		if (cvd->vdev_children > 0 ||
7138 		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
7139 			eligible[eligible_idx++] = cvd;
7140 		}
7141 	}
7142 	VERIFY3S(eligible_idx, >, 0);
7143 
7144 	uint64_t child_no = ztest_random(eligible_idx);
7145 	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
7146 }
7147 
7148 void
7149 ztest_initialize(ztest_ds_t *zd, uint64_t id)
7150 {
7151 	(void) zd, (void) id;
7152 	spa_t *spa = ztest_spa;
7153 	int error = 0;
7154 
7155 	mutex_enter(&ztest_vdev_lock);
7156 
7157 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
7158 
7159 	/* Random leaf vdev */
7160 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
7161 	if (rand_vd == NULL) {
7162 		spa_config_exit(spa, SCL_VDEV, FTAG);
7163 		mutex_exit(&ztest_vdev_lock);
7164 		return;
7165 	}
7166 
7167 	/*
7168 	 * The random vdev we've selected may change as soon as we
7169 	 * drop the spa_config_lock. We create local copies of things
7170 	 * we're interested in.
7171 	 */
7172 	uint64_t guid = rand_vd->vdev_guid;
7173 	char *path = strdup(rand_vd->vdev_path);
7174 	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
7175 
7176 	zfs_dbgmsg("vd %px, guid %llu", rand_vd, (u_longlong_t)guid);
7177 	spa_config_exit(spa, SCL_VDEV, FTAG);
7178 
7179 	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
7180 
7181 	nvlist_t *vdev_guids = fnvlist_alloc();
7182 	nvlist_t *vdev_errlist = fnvlist_alloc();
7183 	fnvlist_add_uint64(vdev_guids, path, guid);
7184 	error = spa_vdev_initialize(spa, vdev_guids, cmd, vdev_errlist);
7185 	fnvlist_free(vdev_guids);
7186 	fnvlist_free(vdev_errlist);
7187 
7188 	switch (cmd) {
7189 	case POOL_INITIALIZE_CANCEL:
7190 		if (ztest_opts.zo_verbose >= 4) {
7191 			(void) printf("Cancel initialize %s", path);
7192 			if (!active)
7193 				(void) printf(" failed (no initialize active)");
7194 			(void) printf("\n");
7195 		}
7196 		break;
7197 	case POOL_INITIALIZE_START:
7198 		if (ztest_opts.zo_verbose >= 4) {
7199 			(void) printf("Start initialize %s", path);
7200 			if (active && error == 0)
7201 				(void) printf(" failed (already active)");
7202 			else if (error != 0)
7203 				(void) printf(" failed (error %d)", error);
7204 			(void) printf("\n");
7205 		}
7206 		break;
7207 	case POOL_INITIALIZE_SUSPEND:
7208 		if (ztest_opts.zo_verbose >= 4) {
7209 			(void) printf("Suspend initialize %s", path);
7210 			if (!active)
7211 				(void) printf(" failed (no initialize active)");
7212 			(void) printf("\n");
7213 		}
7214 		break;
7215 	}
7216 	free(path);
7217 	mutex_exit(&ztest_vdev_lock);
7218 }
7219 
7220 void
7221 ztest_trim(ztest_ds_t *zd, uint64_t id)
7222 {
7223 	(void) zd, (void) id;
7224 	spa_t *spa = ztest_spa;
7225 	int error = 0;
7226 
7227 	mutex_enter(&ztest_vdev_lock);
7228 
7229 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
7230 
7231 	/* Random leaf vdev */
7232 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
7233 	if (rand_vd == NULL) {
7234 		spa_config_exit(spa, SCL_VDEV, FTAG);
7235 		mutex_exit(&ztest_vdev_lock);
7236 		return;
7237 	}
7238 
7239 	/*
7240 	 * The random vdev we've selected may change as soon as we
7241 	 * drop the spa_config_lock. We create local copies of things
7242 	 * we're interested in.
7243 	 */
7244 	uint64_t guid = rand_vd->vdev_guid;
7245 	char *path = strdup(rand_vd->vdev_path);
7246 	boolean_t active = rand_vd->vdev_trim_thread != NULL;
7247 
7248 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, (u_longlong_t)guid);
7249 	spa_config_exit(spa, SCL_VDEV, FTAG);
7250 
7251 	uint64_t cmd = ztest_random(POOL_TRIM_FUNCS);
7252 	uint64_t rate = 1 << ztest_random(30);
7253 	boolean_t partial = (ztest_random(5) > 0);
7254 	boolean_t secure = (ztest_random(5) > 0);
7255 
7256 	nvlist_t *vdev_guids = fnvlist_alloc();
7257 	nvlist_t *vdev_errlist = fnvlist_alloc();
7258 	fnvlist_add_uint64(vdev_guids, path, guid);
7259 	error = spa_vdev_trim(spa, vdev_guids, cmd, rate, partial,
7260 	    secure, vdev_errlist);
7261 	fnvlist_free(vdev_guids);
7262 	fnvlist_free(vdev_errlist);
7263 
7264 	switch (cmd) {
7265 	case POOL_TRIM_CANCEL:
7266 		if (ztest_opts.zo_verbose >= 4) {
7267 			(void) printf("Cancel TRIM %s", path);
7268 			if (!active)
7269 				(void) printf(" failed (no TRIM active)");
7270 			(void) printf("\n");
7271 		}
7272 		break;
7273 	case POOL_TRIM_START:
7274 		if (ztest_opts.zo_verbose >= 4) {
7275 			(void) printf("Start TRIM %s", path);
7276 			if (active && error == 0)
7277 				(void) printf(" failed (already active)");
7278 			else if (error != 0)
7279 				(void) printf(" failed (error %d)", error);
7280 			(void) printf("\n");
7281 		}
7282 		break;
7283 	case POOL_TRIM_SUSPEND:
7284 		if (ztest_opts.zo_verbose >= 4) {
7285 			(void) printf("Suspend TRIM %s", path);
7286 			if (!active)
7287 				(void) printf(" failed (no TRIM active)");
7288 			(void) printf("\n");
7289 		}
7290 		break;
7291 	}
7292 	free(path);
7293 	mutex_exit(&ztest_vdev_lock);
7294 }
7295 
7296 void
7297 ztest_ddt_prune(ztest_ds_t *zd, uint64_t id)
7298 {
7299 	(void) zd, (void) id;
7300 
7301 	spa_t *spa = ztest_spa;
7302 	uint64_t pct = ztest_random(15) + 1;
7303 
7304 	(void) ddt_prune_unique_entries(spa, ZPOOL_DDT_PRUNE_PERCENTAGE, pct);
7305 }
7306 
7307 /*
7308  * Verify pool integrity by running zdb.
7309  */
7310 static void
7311 ztest_run_zdb(uint64_t guid)
7312 {
7313 	int status;
7314 	char *bin;
7315 	char *zdb;
7316 	char *zbuf;
7317 	const int len = MAXPATHLEN + MAXNAMELEN + 20;
7318 	FILE *fp;
7319 
7320 	bin = umem_alloc(len, UMEM_NOFAIL);
7321 	zdb = umem_alloc(len, UMEM_NOFAIL);
7322 	zbuf = umem_alloc(1024, UMEM_NOFAIL);
7323 
7324 	ztest_get_zdb_bin(bin, len);
7325 
7326 	char **set_gvars_args = ztest_global_vars_to_zdb_args();
7327 	if (set_gvars_args == NULL) {
7328 		fatal(B_FALSE, "Failed to allocate memory in "
7329 		    "ztest_global_vars_to_zdb_args(). Cannot run zdb.\n");
7330 	}
7331 	char *set_gvars_args_joined = join_strings(set_gvars_args, " ");
7332 	free(set_gvars_args);
7333 
7334 	size_t would = snprintf(zdb, len,
7335 	    "%s -bcc%s%s -G -d -Y -e -y %s -p %s %"PRIu64,
7336 	    bin,
7337 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
7338 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
7339 	    set_gvars_args_joined,
7340 	    ztest_opts.zo_dir,
7341 	    guid);
7342 	ASSERT3U(would, <, len);
7343 
7344 	umem_free(set_gvars_args_joined, strlen(set_gvars_args_joined) + 1);
7345 
7346 	if (ztest_opts.zo_verbose >= 5)
7347 		(void) printf("Executing %s\n", zdb);
7348 
7349 	fp = popen(zdb, "r");
7350 
7351 	while (fgets(zbuf, 1024, fp) != NULL)
7352 		if (ztest_opts.zo_verbose >= 3)
7353 			(void) printf("%s", zbuf);
7354 
7355 	status = pclose(fp);
7356 
7357 	if (status == 0)
7358 		goto out;
7359 
7360 	ztest_dump_core = 0;
7361 	if (WIFEXITED(status))
7362 		fatal(B_FALSE, "'%s' exit code %d", zdb, WEXITSTATUS(status));
7363 	else
7364 		fatal(B_FALSE, "'%s' died with signal %d",
7365 		    zdb, WTERMSIG(status));
7366 out:
7367 	umem_free(bin, len);
7368 	umem_free(zdb, len);
7369 	umem_free(zbuf, 1024);
7370 }
7371 
7372 static void
7373 ztest_walk_pool_directory(const char *header)
7374 {
7375 	spa_t *spa = NULL;
7376 
7377 	if (ztest_opts.zo_verbose >= 6)
7378 		(void) puts(header);
7379 
7380 	mutex_enter(&spa_namespace_lock);
7381 	while ((spa = spa_next(spa)) != NULL)
7382 		if (ztest_opts.zo_verbose >= 6)
7383 			(void) printf("\t%s\n", spa_name(spa));
7384 	mutex_exit(&spa_namespace_lock);
7385 }
7386 
7387 static void
7388 ztest_spa_import_export(char *oldname, char *newname)
7389 {
7390 	nvlist_t *config, *newconfig;
7391 	uint64_t pool_guid;
7392 	spa_t *spa;
7393 	int error;
7394 
7395 	if (ztest_opts.zo_verbose >= 4) {
7396 		(void) printf("import/export: old = %s, new = %s\n",
7397 		    oldname, newname);
7398 	}
7399 
7400 	/*
7401 	 * Clean up from previous runs.
7402 	 */
7403 	(void) spa_destroy(newname);
7404 
7405 	/*
7406 	 * Get the pool's configuration and guid.
7407 	 */
7408 	VERIFY0(spa_open(oldname, &spa, FTAG));
7409 
7410 	/*
7411 	 * Kick off a scrub to tickle scrub/export races.
7412 	 */
7413 	if (ztest_random(2) == 0)
7414 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
7415 
7416 	pool_guid = spa_guid(spa);
7417 	spa_close(spa, FTAG);
7418 
7419 	ztest_walk_pool_directory("pools before export");
7420 
7421 	/*
7422 	 * Export it.
7423 	 */
7424 	VERIFY0(spa_export(oldname, &config, B_FALSE, B_FALSE));
7425 
7426 	ztest_walk_pool_directory("pools after export");
7427 
7428 	/*
7429 	 * Try to import it.
7430 	 */
7431 	newconfig = spa_tryimport(config);
7432 	ASSERT3P(newconfig, !=, NULL);
7433 	fnvlist_free(newconfig);
7434 
7435 	/*
7436 	 * Import it under the new name.
7437 	 */
7438 	error = spa_import(newname, config, NULL, 0);
7439 	if (error != 0) {
7440 		dump_nvlist(config, 0);
7441 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
7442 		    oldname, newname, error);
7443 	}
7444 
7445 	ztest_walk_pool_directory("pools after import");
7446 
7447 	/*
7448 	 * Try to import it again -- should fail with EEXIST.
7449 	 */
7450 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
7451 
7452 	/*
7453 	 * Try to import it under a different name -- should fail with EEXIST.
7454 	 */
7455 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
7456 
7457 	/*
7458 	 * Verify that the pool is no longer visible under the old name.
7459 	 */
7460 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
7461 
7462 	/*
7463 	 * Verify that we can open and close the pool using the new name.
7464 	 */
7465 	VERIFY0(spa_open(newname, &spa, FTAG));
7466 	ASSERT3U(pool_guid, ==, spa_guid(spa));
7467 	spa_close(spa, FTAG);
7468 
7469 	fnvlist_free(config);
7470 }
7471 
7472 static void
7473 ztest_resume(spa_t *spa)
7474 {
7475 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
7476 		(void) printf("resuming from suspended state\n");
7477 	spa_vdev_state_enter(spa, SCL_NONE);
7478 	vdev_clear(spa, NULL);
7479 	(void) spa_vdev_state_exit(spa, NULL, 0);
7480 	(void) zio_resume(spa);
7481 }
7482 
7483 static __attribute__((noreturn)) void
7484 ztest_resume_thread(void *arg)
7485 {
7486 	spa_t *spa = arg;
7487 
7488 	/*
7489 	 * Synthesize aged DDT entries for ddt prune testing
7490 	 */
7491 	ddt_prune_artificial_age = B_TRUE;
7492 	if (ztest_opts.zo_verbose >= 3)
7493 		ddt_dump_prune_histogram = B_TRUE;
7494 
7495 	while (!ztest_exiting) {
7496 		if (spa_suspended(spa))
7497 			ztest_resume(spa);
7498 		(void) poll(NULL, 0, 100);
7499 
7500 		/*
7501 		 * Periodically change the zfs_compressed_arc_enabled setting.
7502 		 */
7503 		if (ztest_random(10) == 0)
7504 			zfs_compressed_arc_enabled = ztest_random(2);
7505 
7506 		/*
7507 		 * Periodically change the zfs_abd_scatter_enabled setting.
7508 		 */
7509 		if (ztest_random(10) == 0)
7510 			zfs_abd_scatter_enabled = ztest_random(2);
7511 	}
7512 
7513 	thread_exit();
7514 }
7515 
7516 static __attribute__((noreturn)) void
7517 ztest_deadman_thread(void *arg)
7518 {
7519 	ztest_shared_t *zs = arg;
7520 	spa_t *spa = ztest_spa;
7521 	hrtime_t delay, overdue, last_run = gethrtime();
7522 
7523 	delay = (zs->zs_thread_stop - zs->zs_thread_start) +
7524 	    MSEC2NSEC(zfs_deadman_synctime_ms);
7525 
7526 	while (!ztest_exiting) {
7527 		/*
7528 		 * Wait for the delay timer while checking occasionally
7529 		 * if we should stop.
7530 		 */
7531 		if (gethrtime() < last_run + delay) {
7532 			(void) poll(NULL, 0, 1000);
7533 			continue;
7534 		}
7535 
7536 		/*
7537 		 * If the pool is suspended then fail immediately. Otherwise,
7538 		 * check to see if the pool is making any progress. If
7539 		 * vdev_deadman() discovers that there hasn't been any recent
7540 		 * I/Os then it will end up aborting the tests.
7541 		 */
7542 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
7543 			fatal(B_FALSE,
7544 			    "aborting test after %llu seconds because "
7545 			    "pool has transitioned to a suspended state.",
7546 			    (u_longlong_t)zfs_deadman_synctime_ms / 1000);
7547 		}
7548 		vdev_deadman(spa->spa_root_vdev, FTAG);
7549 
7550 		/*
7551 		 * If the process doesn't complete within a grace period of
7552 		 * zfs_deadman_synctime_ms over the expected finish time,
7553 		 * then it may be hung and is terminated.
7554 		 */
7555 		overdue = zs->zs_proc_stop + MSEC2NSEC(zfs_deadman_synctime_ms);
7556 		if (gethrtime() > overdue) {
7557 			fatal(B_FALSE,
7558 			    "aborting test after %llu seconds because "
7559 			    "the process is overdue for termination.",
7560 			    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7561 		}
7562 
7563 		(void) printf("ztest has been running for %lld seconds\n",
7564 		    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7565 
7566 		last_run = gethrtime();
7567 		delay = MSEC2NSEC(zfs_deadman_checktime_ms);
7568 	}
7569 
7570 	thread_exit();
7571 }
7572 
7573 static void
7574 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
7575 {
7576 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
7577 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
7578 	hrtime_t functime = gethrtime();
7579 	int i;
7580 
7581 	for (i = 0; i < zi->zi_iters; i++)
7582 		zi->zi_func(zd, id);
7583 
7584 	functime = gethrtime() - functime;
7585 
7586 	atomic_add_64(&zc->zc_count, 1);
7587 	atomic_add_64(&zc->zc_time, functime);
7588 
7589 	if (ztest_opts.zo_verbose >= 4)
7590 		(void) printf("%6.2f sec in %s\n",
7591 		    (double)functime / NANOSEC, zi->zi_funcname);
7592 }
7593 
7594 typedef struct ztest_raidz_expand_io {
7595 	uint64_t	rzx_id;
7596 	uint64_t	rzx_amount;
7597 	uint64_t	rzx_bufsize;
7598 	const void	*rzx_buffer;
7599 	uint64_t	rzx_alloc_max;
7600 	spa_t		*rzx_spa;
7601 } ztest_expand_io_t;
7602 
7603 #undef OD_ARRAY_SIZE
7604 #define	OD_ARRAY_SIZE	10
7605 
7606 /*
7607  * Write a request amount of data to some dataset objects.
7608  * There will be ztest_opts.zo_threads count of these running in parallel.
7609  */
7610 static __attribute__((noreturn)) void
7611 ztest_rzx_thread(void *arg)
7612 {
7613 	ztest_expand_io_t *info = (ztest_expand_io_t *)arg;
7614 	ztest_od_t *od;
7615 	int batchsize;
7616 	int od_size;
7617 	ztest_ds_t *zd = &ztest_ds[info->rzx_id % ztest_opts.zo_datasets];
7618 	spa_t *spa = info->rzx_spa;
7619 
7620 	od_size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
7621 	od = umem_alloc(od_size, UMEM_NOFAIL);
7622 	batchsize = OD_ARRAY_SIZE;
7623 
7624 	/* Create objects to write to */
7625 	for (int b = 0; b < batchsize; b++) {
7626 		ztest_od_init(od + b, info->rzx_id, FTAG, b,
7627 		    DMU_OT_UINT64_OTHER, 0, 0, 0);
7628 	}
7629 	if (ztest_object_init(zd, od, od_size, B_FALSE) != 0) {
7630 		umem_free(od, od_size);
7631 		thread_exit();
7632 	}
7633 
7634 	for (uint64_t offset = 0, written = 0; written < info->rzx_amount;
7635 	    offset += info->rzx_bufsize) {
7636 		/* write to 10 objects */
7637 		for (int i = 0; i < batchsize && written < info->rzx_amount;
7638 		    i++) {
7639 			(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
7640 			ztest_write(zd, od[i].od_object, offset,
7641 			    info->rzx_bufsize, info->rzx_buffer);
7642 			(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
7643 			written += info->rzx_bufsize;
7644 		}
7645 		txg_wait_synced(spa_get_dsl(spa), 0);
7646 		/* due to inflation, we'll typically bail here */
7647 		if (metaslab_class_get_alloc(spa_normal_class(spa)) >
7648 		    info->rzx_alloc_max) {
7649 			break;
7650 		}
7651 	}
7652 
7653 	/* Remove a few objects to leave some holes in allocation space */
7654 	mutex_enter(&zd->zd_dirobj_lock);
7655 	(void) ztest_remove(zd, od, 2);
7656 	mutex_exit(&zd->zd_dirobj_lock);
7657 
7658 	umem_free(od, od_size);
7659 
7660 	thread_exit();
7661 }
7662 
7663 static __attribute__((noreturn)) void
7664 ztest_thread(void *arg)
7665 {
7666 	int rand;
7667 	uint64_t id = (uintptr_t)arg;
7668 	ztest_shared_t *zs = ztest_shared;
7669 	uint64_t call_next;
7670 	hrtime_t now;
7671 	ztest_info_t *zi;
7672 	ztest_shared_callstate_t *zc;
7673 
7674 	while ((now = gethrtime()) < zs->zs_thread_stop) {
7675 		/*
7676 		 * See if it's time to force a crash.
7677 		 */
7678 		if (now > zs->zs_thread_kill &&
7679 		    raidz_expand_pause_point == RAIDZ_EXPAND_PAUSE_NONE) {
7680 			ztest_kill(zs);
7681 		}
7682 
7683 		/*
7684 		 * If we're getting ENOSPC with some regularity, stop.
7685 		 */
7686 		if (zs->zs_enospc_count > 10)
7687 			break;
7688 
7689 		/*
7690 		 * Pick a random function to execute.
7691 		 */
7692 		rand = ztest_random(ZTEST_FUNCS);
7693 		zi = &ztest_info[rand];
7694 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
7695 		call_next = zc->zc_next;
7696 
7697 		if (now >= call_next &&
7698 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
7699 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
7700 			ztest_execute(rand, zi, id);
7701 		}
7702 	}
7703 
7704 	thread_exit();
7705 }
7706 
7707 static void
7708 ztest_dataset_name(char *dsname, const char *pool, int d)
7709 {
7710 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
7711 }
7712 
7713 static void
7714 ztest_dataset_destroy(int d)
7715 {
7716 	char name[ZFS_MAX_DATASET_NAME_LEN];
7717 	int t;
7718 
7719 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7720 
7721 	if (ztest_opts.zo_verbose >= 3)
7722 		(void) printf("Destroying %s to free up space\n", name);
7723 
7724 	/*
7725 	 * Cleanup any non-standard clones and snapshots.  In general,
7726 	 * ztest thread t operates on dataset (t % zopt_datasets),
7727 	 * so there may be more than one thing to clean up.
7728 	 */
7729 	for (t = d; t < ztest_opts.zo_threads;
7730 	    t += ztest_opts.zo_datasets)
7731 		ztest_dsl_dataset_cleanup(name, t);
7732 
7733 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
7734 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
7735 }
7736 
7737 static void
7738 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
7739 {
7740 	uint64_t usedobjs, dirobjs, scratch;
7741 
7742 	/*
7743 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
7744 	 * Therefore, the number of objects in use should equal the
7745 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
7746 	 * If not, we have an object leak.
7747 	 *
7748 	 * Note that we can only check this in ztest_dataset_open(),
7749 	 * when the open-context and syncing-context values agree.
7750 	 * That's because zap_count() returns the open-context value,
7751 	 * while dmu_objset_space() returns the rootbp fill count.
7752 	 */
7753 	VERIFY0(zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
7754 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
7755 	ASSERT3U(dirobjs + 1, ==, usedobjs);
7756 }
7757 
7758 static int
7759 ztest_dataset_open(int d)
7760 {
7761 	ztest_ds_t *zd = &ztest_ds[d];
7762 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
7763 	objset_t *os;
7764 	zilog_t *zilog;
7765 	char name[ZFS_MAX_DATASET_NAME_LEN];
7766 	int error;
7767 
7768 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7769 
7770 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
7771 
7772 	error = ztest_dataset_create(name);
7773 	if (error == ENOSPC) {
7774 		(void) pthread_rwlock_unlock(&ztest_name_lock);
7775 		ztest_record_enospc(FTAG);
7776 		return (error);
7777 	}
7778 	ASSERT(error == 0 || error == EEXIST);
7779 
7780 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
7781 	    B_TRUE, zd, &os));
7782 	(void) pthread_rwlock_unlock(&ztest_name_lock);
7783 
7784 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
7785 
7786 	zilog = zd->zd_zilog;
7787 
7788 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
7789 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
7790 		fatal(B_FALSE, "missing log records: "
7791 		    "claimed %"PRIu64" < committed %"PRIu64"",
7792 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
7793 
7794 	ztest_dataset_dirobj_verify(zd);
7795 
7796 	zil_replay(os, zd, ztest_replay_vector);
7797 
7798 	ztest_dataset_dirobj_verify(zd);
7799 
7800 	if (ztest_opts.zo_verbose >= 6)
7801 		(void) printf("%s replay %"PRIu64" blocks, "
7802 		    "%"PRIu64" records, seq %"PRIu64"\n",
7803 		    zd->zd_name,
7804 		    zilog->zl_parse_blk_count,
7805 		    zilog->zl_parse_lr_count,
7806 		    zilog->zl_replaying_seq);
7807 
7808 	zilog = zil_open(os, ztest_get_data, NULL);
7809 
7810 	if (zilog->zl_replaying_seq != 0 &&
7811 	    zilog->zl_replaying_seq < committed_seq)
7812 		fatal(B_FALSE, "missing log records: "
7813 		    "replayed %"PRIu64" < committed %"PRIu64"",
7814 		    zilog->zl_replaying_seq, committed_seq);
7815 
7816 	return (0);
7817 }
7818 
7819 static void
7820 ztest_dataset_close(int d)
7821 {
7822 	ztest_ds_t *zd = &ztest_ds[d];
7823 
7824 	zil_close(zd->zd_zilog);
7825 	dmu_objset_disown(zd->zd_os, B_TRUE, zd);
7826 
7827 	ztest_zd_fini(zd);
7828 }
7829 
7830 static int
7831 ztest_replay_zil_cb(const char *name, void *arg)
7832 {
7833 	(void) arg;
7834 	objset_t *os;
7835 	ztest_ds_t *zdtmp;
7836 
7837 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_ANY, B_TRUE,
7838 	    B_TRUE, FTAG, &os));
7839 
7840 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
7841 
7842 	ztest_zd_init(zdtmp, NULL, os);
7843 	zil_replay(os, zdtmp, ztest_replay_vector);
7844 	ztest_zd_fini(zdtmp);
7845 
7846 	if (dmu_objset_zil(os)->zl_parse_lr_count != 0 &&
7847 	    ztest_opts.zo_verbose >= 6) {
7848 		zilog_t *zilog = dmu_objset_zil(os);
7849 
7850 		(void) printf("%s replay %"PRIu64" blocks, "
7851 		    "%"PRIu64" records, seq %"PRIu64"\n",
7852 		    name,
7853 		    zilog->zl_parse_blk_count,
7854 		    zilog->zl_parse_lr_count,
7855 		    zilog->zl_replaying_seq);
7856 	}
7857 
7858 	umem_free(zdtmp, sizeof (ztest_ds_t));
7859 
7860 	dmu_objset_disown(os, B_TRUE, FTAG);
7861 	return (0);
7862 }
7863 
7864 static void
7865 ztest_freeze(void)
7866 {
7867 	ztest_ds_t *zd = &ztest_ds[0];
7868 	spa_t *spa;
7869 	int numloops = 0;
7870 
7871 	/* freeze not supported during RAIDZ expansion */
7872 	if (ztest_opts.zo_raid_do_expand)
7873 		return;
7874 
7875 	if (ztest_opts.zo_verbose >= 3)
7876 		(void) printf("testing spa_freeze()...\n");
7877 
7878 	raidz_scratch_verify();
7879 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7880 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7881 	VERIFY0(ztest_dataset_open(0));
7882 	ztest_spa = spa;
7883 
7884 	/*
7885 	 * Force the first log block to be transactionally allocated.
7886 	 * We have to do this before we freeze the pool -- otherwise
7887 	 * the log chain won't be anchored.
7888 	 */
7889 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
7890 		ztest_dmu_object_alloc_free(zd, 0);
7891 		zil_commit(zd->zd_zilog, 0);
7892 	}
7893 
7894 	txg_wait_synced(spa_get_dsl(spa), 0);
7895 
7896 	/*
7897 	 * Freeze the pool.  This stops spa_sync() from doing anything,
7898 	 * so that the only way to record changes from now on is the ZIL.
7899 	 */
7900 	spa_freeze(spa);
7901 
7902 	/*
7903 	 * Because it is hard to predict how much space a write will actually
7904 	 * require beforehand, we leave ourselves some fudge space to write over
7905 	 * capacity.
7906 	 */
7907 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
7908 
7909 	/*
7910 	 * Run tests that generate log records but don't alter the pool config
7911 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
7912 	 * We do a txg_wait_synced() after each iteration to force the txg
7913 	 * to increase well beyond the last synced value in the uberblock.
7914 	 * The ZIL should be OK with that.
7915 	 *
7916 	 * Run a random number of times less than zo_maxloops and ensure we do
7917 	 * not run out of space on the pool.
7918 	 */
7919 	while (ztest_random(10) != 0 &&
7920 	    numloops++ < ztest_opts.zo_maxloops &&
7921 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
7922 		ztest_od_t od;
7923 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
7924 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
7925 		ztest_io(zd, od.od_object,
7926 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
7927 		txg_wait_synced(spa_get_dsl(spa), 0);
7928 	}
7929 
7930 	/*
7931 	 * Commit all of the changes we just generated.
7932 	 */
7933 	zil_commit(zd->zd_zilog, 0);
7934 	txg_wait_synced(spa_get_dsl(spa), 0);
7935 
7936 	/*
7937 	 * Close our dataset and close the pool.
7938 	 */
7939 	ztest_dataset_close(0);
7940 	spa_close(spa, FTAG);
7941 	kernel_fini();
7942 
7943 	/*
7944 	 * Open and close the pool and dataset to induce log replay.
7945 	 */
7946 	raidz_scratch_verify();
7947 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7948 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7949 	ASSERT3U(spa_freeze_txg(spa), ==, UINT64_MAX);
7950 	VERIFY0(ztest_dataset_open(0));
7951 	ztest_spa = spa;
7952 	txg_wait_synced(spa_get_dsl(spa), 0);
7953 	ztest_dataset_close(0);
7954 	ztest_reguid(NULL, 0);
7955 
7956 	spa_close(spa, FTAG);
7957 	kernel_fini();
7958 }
7959 
7960 static void
7961 ztest_import_impl(void)
7962 {
7963 	importargs_t args = { 0 };
7964 	nvlist_t *cfg = NULL;
7965 	int nsearch = 1;
7966 	char *searchdirs[nsearch];
7967 	int flags = ZFS_IMPORT_MISSING_LOG;
7968 
7969 	searchdirs[0] = ztest_opts.zo_dir;
7970 	args.paths = nsearch;
7971 	args.path = searchdirs;
7972 	args.can_be_active = B_FALSE;
7973 
7974 	libpc_handle_t lpch = {
7975 		.lpc_lib_handle = NULL,
7976 		.lpc_ops = &libzpool_config_ops,
7977 		.lpc_printerr = B_TRUE
7978 	};
7979 	VERIFY0(zpool_find_config(&lpch, ztest_opts.zo_pool, &cfg, &args));
7980 	VERIFY0(spa_import(ztest_opts.zo_pool, cfg, NULL, flags));
7981 	fnvlist_free(cfg);
7982 }
7983 
7984 /*
7985  * Import a storage pool with the given name.
7986  */
7987 static void
7988 ztest_import(ztest_shared_t *zs)
7989 {
7990 	spa_t *spa;
7991 
7992 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7993 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7994 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7995 
7996 	raidz_scratch_verify();
7997 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7998 
7999 	ztest_import_impl();
8000 
8001 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
8002 	zs->zs_metaslab_sz =
8003 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
8004 	zs->zs_guid = spa_guid(spa);
8005 	spa_close(spa, FTAG);
8006 
8007 	kernel_fini();
8008 
8009 	if (!ztest_opts.zo_mmp_test) {
8010 		ztest_run_zdb(zs->zs_guid);
8011 		ztest_freeze();
8012 		ztest_run_zdb(zs->zs_guid);
8013 	}
8014 
8015 	(void) pthread_rwlock_destroy(&ztest_name_lock);
8016 	mutex_destroy(&ztest_vdev_lock);
8017 	mutex_destroy(&ztest_checkpoint_lock);
8018 }
8019 
8020 /*
8021  * After the expansion was killed, check that the pool is healthy
8022  */
8023 static void
8024 ztest_raidz_expand_check(spa_t *spa)
8025 {
8026 	ASSERT3U(ztest_opts.zo_raidz_expand_test, ==, RAIDZ_EXPAND_KILLED);
8027 	/*
8028 	 * Set pool check done flag, main program will run a zdb check
8029 	 * of the pool when we exit.
8030 	 */
8031 	ztest_shared_opts->zo_raidz_expand_test = RAIDZ_EXPAND_CHECKED;
8032 
8033 	/* Wait for reflow to finish */
8034 	if (ztest_opts.zo_verbose >= 1) {
8035 		(void) printf("\nwaiting for reflow to finish ...\n");
8036 	}
8037 	pool_raidz_expand_stat_t rzx_stats;
8038 	pool_raidz_expand_stat_t *pres = &rzx_stats;
8039 	do {
8040 		txg_wait_synced(spa_get_dsl(spa), 0);
8041 		(void) poll(NULL, 0, 500); /* wait 1/2 second */
8042 
8043 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8044 		(void) spa_raidz_expand_get_stats(spa, pres);
8045 		spa_config_exit(spa, SCL_CONFIG, FTAG);
8046 	} while (pres->pres_state != DSS_FINISHED &&
8047 	    pres->pres_reflowed < pres->pres_to_reflow);
8048 
8049 	if (ztest_opts.zo_verbose >= 1) {
8050 		(void) printf("verifying an interrupted raidz "
8051 		    "expansion using a pool scrub ...\n");
8052 	}
8053 	/* Will fail here if there is non-recoverable corruption detected */
8054 	VERIFY0(ztest_scrub_impl(spa));
8055 	if (ztest_opts.zo_verbose >= 1) {
8056 		(void) printf("raidz expansion scrub check complete\n");
8057 	}
8058 }
8059 
8060 /*
8061  * Start a raidz expansion test.  We run some I/O on the pool for a while
8062  * to get some data in the pool.  Then we grow the raidz and
8063  * kill the test at the requested offset into the reflow, verifying that
8064  * doing such does not lead to pool corruption.
8065  */
8066 static void
8067 ztest_raidz_expand_run(ztest_shared_t *zs, spa_t *spa)
8068 {
8069 	nvlist_t *root;
8070 	pool_raidz_expand_stat_t rzx_stats;
8071 	pool_raidz_expand_stat_t *pres = &rzx_stats;
8072 	kthread_t **run_threads;
8073 	vdev_t *cvd, *rzvd = spa->spa_root_vdev->vdev_child[0];
8074 	int total_disks = rzvd->vdev_children;
8075 	int data_disks = total_disks - vdev_get_nparity(rzvd);
8076 	uint64_t alloc_goal;
8077 	uint64_t csize;
8078 	int error, t;
8079 	int threads = ztest_opts.zo_threads;
8080 	ztest_expand_io_t *thread_args;
8081 
8082 	ASSERT3U(ztest_opts.zo_raidz_expand_test, !=, RAIDZ_EXPAND_NONE);
8083 	ASSERT3P(rzvd->vdev_ops, ==, &vdev_raidz_ops);
8084 	ztest_opts.zo_raidz_expand_test = RAIDZ_EXPAND_STARTED;
8085 
8086 	/* Setup a 1 MiB buffer of random data */
8087 	uint64_t bufsize = 1024 * 1024;
8088 	void *buffer = umem_alloc(bufsize, UMEM_NOFAIL);
8089 
8090 	if (read(ztest_fd_rand, buffer, bufsize) != bufsize) {
8091 		fatal(B_TRUE, "short read from /dev/urandom");
8092 	}
8093 	/*
8094 	 * Put some data in the pool and then attach a vdev to initiate
8095 	 * reflow.
8096 	 */
8097 	run_threads = umem_zalloc(threads * sizeof (kthread_t *), UMEM_NOFAIL);
8098 	thread_args = umem_zalloc(threads * sizeof (ztest_expand_io_t),
8099 	    UMEM_NOFAIL);
8100 	/* Aim for roughly 25% of allocatable space up to 1GB */
8101 	alloc_goal = (vdev_get_min_asize(rzvd) * data_disks) / total_disks;
8102 	alloc_goal = MIN(alloc_goal >> 2, 1024*1024*1024);
8103 	if (ztest_opts.zo_verbose >= 1) {
8104 		(void) printf("adding data to pool '%s', goal %llu bytes\n",
8105 		    ztest_opts.zo_pool, (u_longlong_t)alloc_goal);
8106 	}
8107 
8108 	/*
8109 	 * Kick off all the I/O generators that run in parallel.
8110 	 */
8111 	for (t = 0; t < threads; t++) {
8112 		if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
8113 			umem_free(run_threads, threads * sizeof (kthread_t *));
8114 			umem_free(buffer, bufsize);
8115 			return;
8116 		}
8117 		thread_args[t].rzx_id = t;
8118 		thread_args[t].rzx_amount = alloc_goal / threads;
8119 		thread_args[t].rzx_bufsize = bufsize;
8120 		thread_args[t].rzx_buffer = buffer;
8121 		thread_args[t].rzx_alloc_max = alloc_goal;
8122 		thread_args[t].rzx_spa = spa;
8123 		run_threads[t] = thread_create(NULL, 0, ztest_rzx_thread,
8124 		    &thread_args[t], 0, NULL, TS_RUN | TS_JOINABLE,
8125 		    defclsyspri);
8126 	}
8127 
8128 	/*
8129 	 * Wait for all of the writers to complete.
8130 	 */
8131 	for (t = 0; t < threads; t++)
8132 		VERIFY0(thread_join(run_threads[t]));
8133 
8134 	/*
8135 	 * Close all datasets. This must be done after all the threads
8136 	 * are joined so we can be sure none of the datasets are in-use
8137 	 * by any of the threads.
8138 	 */
8139 	for (t = 0; t < ztest_opts.zo_threads; t++) {
8140 		if (t < ztest_opts.zo_datasets)
8141 			ztest_dataset_close(t);
8142 	}
8143 
8144 	txg_wait_synced(spa_get_dsl(spa), 0);
8145 
8146 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
8147 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
8148 
8149 	umem_free(buffer, bufsize);
8150 	umem_free(run_threads, threads * sizeof (kthread_t *));
8151 	umem_free(thread_args, threads * sizeof (ztest_expand_io_t));
8152 
8153 	/* Set our reflow target to 25%, 50% or 75% of allocated size */
8154 	uint_t multiple = ztest_random(3) + 1;
8155 	uint64_t reflow_max = (rzvd->vdev_stat.vs_alloc * multiple) / 4;
8156 	raidz_expand_max_reflow_bytes = reflow_max;
8157 
8158 	if (ztest_opts.zo_verbose >= 1) {
8159 		(void) printf("running raidz expansion test, killing when "
8160 		    "reflow reaches %llu bytes (%u/4 of allocated space)\n",
8161 		    (u_longlong_t)reflow_max, multiple);
8162 	}
8163 
8164 	/* XXX - do we want some I/O load during the reflow? */
8165 
8166 	/*
8167 	 * Use a disk size that is larger than existing ones
8168 	 */
8169 	cvd = rzvd->vdev_child[0];
8170 	csize = vdev_get_min_asize(cvd);
8171 	csize += csize / 10;
8172 	/*
8173 	 * Path to vdev to be attached
8174 	 */
8175 	char *newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
8176 	(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
8177 	    ztest_opts.zo_dir, ztest_opts.zo_pool, rzvd->vdev_children);
8178 	/*
8179 	 * Build the nvlist describing newpath.
8180 	 */
8181 	root = make_vdev_root(newpath, NULL, NULL, csize, ztest_get_ashift(),
8182 	    NULL, 0, 0, 1);
8183 	/*
8184 	 * Expand the raidz vdev by attaching the new disk
8185 	 */
8186 	if (ztest_opts.zo_verbose >= 1) {
8187 		(void) printf("expanding raidz: %d wide to %d wide with '%s'\n",
8188 		    (int)rzvd->vdev_children, (int)rzvd->vdev_children + 1,
8189 		    newpath);
8190 	}
8191 	error = spa_vdev_attach(spa, rzvd->vdev_guid, root, B_FALSE, B_FALSE);
8192 	nvlist_free(root);
8193 	if (error != 0) {
8194 		fatal(0, "raidz expand: attach (%s %llu) returned %d",
8195 		    newpath, (long long)csize, error);
8196 	}
8197 
8198 	/*
8199 	 * Wait for reflow to begin
8200 	 */
8201 	while (spa->spa_raidz_expand == NULL) {
8202 		txg_wait_synced(spa_get_dsl(spa), 0);
8203 		(void) poll(NULL, 0, 100); /* wait 1/10 second */
8204 	}
8205 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8206 	(void) spa_raidz_expand_get_stats(spa, pres);
8207 	spa_config_exit(spa, SCL_CONFIG, FTAG);
8208 	while (pres->pres_state != DSS_SCANNING) {
8209 		txg_wait_synced(spa_get_dsl(spa), 0);
8210 		(void) poll(NULL, 0, 100); /* wait 1/10 second */
8211 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8212 		(void) spa_raidz_expand_get_stats(spa, pres);
8213 		spa_config_exit(spa, SCL_CONFIG, FTAG);
8214 	}
8215 
8216 	ASSERT3U(pres->pres_state, ==, DSS_SCANNING);
8217 	ASSERT3U(pres->pres_to_reflow, !=, 0);
8218 	/*
8219 	 * Set so when we are killed we go to raidz checking rather than
8220 	 * restarting test.
8221 	 */
8222 	ztest_shared_opts->zo_raidz_expand_test = RAIDZ_EXPAND_KILLED;
8223 	if (ztest_opts.zo_verbose >= 1) {
8224 		(void) printf("raidz expansion reflow started, waiting for "
8225 		    "%llu bytes to be copied\n", (u_longlong_t)reflow_max);
8226 	}
8227 
8228 	/*
8229 	 * Wait for reflow maximum to be reached and then kill the test
8230 	 */
8231 	while (pres->pres_reflowed < reflow_max) {
8232 		txg_wait_synced(spa_get_dsl(spa), 0);
8233 		(void) poll(NULL, 0, 100); /* wait 1/10 second */
8234 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8235 		(void) spa_raidz_expand_get_stats(spa, pres);
8236 		spa_config_exit(spa, SCL_CONFIG, FTAG);
8237 	}
8238 
8239 	/* Reset the reflow pause before killing */
8240 	raidz_expand_max_reflow_bytes = 0;
8241 
8242 	if (ztest_opts.zo_verbose >= 1) {
8243 		(void) printf("killing raidz expansion test after reflow "
8244 		    "reached %llu bytes\n", (u_longlong_t)pres->pres_reflowed);
8245 	}
8246 
8247 	/*
8248 	 * Kill ourself to simulate a panic during a reflow.  Our parent will
8249 	 * restart the test and the changed flag value will drive the test
8250 	 * through the scrub/check code to verify the pool is not corrupted.
8251 	 */
8252 	ztest_kill(zs);
8253 }
8254 
8255 static void
8256 ztest_generic_run(ztest_shared_t *zs, spa_t *spa)
8257 {
8258 	kthread_t **run_threads;
8259 	int t;
8260 
8261 	run_threads = umem_zalloc(ztest_opts.zo_threads * sizeof (kthread_t *),
8262 	    UMEM_NOFAIL);
8263 
8264 	/*
8265 	 * Kick off all the tests that run in parallel.
8266 	 */
8267 	for (t = 0; t < ztest_opts.zo_threads; t++) {
8268 		if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
8269 			umem_free(run_threads, ztest_opts.zo_threads *
8270 			    sizeof (kthread_t *));
8271 			return;
8272 		}
8273 
8274 		run_threads[t] = thread_create(NULL, 0, ztest_thread,
8275 		    (void *)(uintptr_t)t, 0, NULL, TS_RUN | TS_JOINABLE,
8276 		    defclsyspri);
8277 	}
8278 
8279 	/*
8280 	 * Wait for all of the tests to complete.
8281 	 */
8282 	for (t = 0; t < ztest_opts.zo_threads; t++)
8283 		VERIFY0(thread_join(run_threads[t]));
8284 
8285 	/*
8286 	 * Close all datasets. This must be done after all the threads
8287 	 * are joined so we can be sure none of the datasets are in-use
8288 	 * by any of the threads.
8289 	 */
8290 	for (t = 0; t < ztest_opts.zo_threads; t++) {
8291 		if (t < ztest_opts.zo_datasets)
8292 			ztest_dataset_close(t);
8293 	}
8294 
8295 	txg_wait_synced(spa_get_dsl(spa), 0);
8296 
8297 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
8298 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
8299 
8300 	umem_free(run_threads, ztest_opts.zo_threads * sizeof (kthread_t *));
8301 }
8302 
8303 /*
8304  * Setup our test context and kick off threads to run tests on all datasets
8305  * in parallel.
8306  */
8307 static void
8308 ztest_run(ztest_shared_t *zs)
8309 {
8310 	spa_t *spa;
8311 	objset_t *os;
8312 	kthread_t *resume_thread, *deadman_thread;
8313 	uint64_t object;
8314 	int error;
8315 	int t, d;
8316 
8317 	ztest_exiting = B_FALSE;
8318 
8319 	/*
8320 	 * Initialize parent/child shared state.
8321 	 */
8322 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
8323 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
8324 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
8325 
8326 	zs->zs_thread_start = gethrtime();
8327 	zs->zs_thread_stop =
8328 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
8329 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
8330 	zs->zs_thread_kill = zs->zs_thread_stop;
8331 	if (ztest_random(100) < ztest_opts.zo_killrate) {
8332 		zs->zs_thread_kill -=
8333 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
8334 	}
8335 
8336 	mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
8337 
8338 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
8339 	    offsetof(ztest_cb_data_t, zcd_node));
8340 
8341 	/*
8342 	 * Open our pool.  It may need to be imported first depending on
8343 	 * what tests were running when the previous pass was terminated.
8344 	 */
8345 	raidz_scratch_verify();
8346 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
8347 	error = spa_open(ztest_opts.zo_pool, &spa, FTAG);
8348 	if (error) {
8349 		VERIFY3S(error, ==, ENOENT);
8350 		ztest_import_impl();
8351 		VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
8352 		zs->zs_metaslab_sz =
8353 		    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
8354 	}
8355 
8356 	metaslab_preload_limit = ztest_random(20) + 1;
8357 	ztest_spa = spa;
8358 
8359 	/*
8360 	 * XXX - BUGBUG raidz expansion do not run this for generic for now
8361 	 */
8362 	if (ztest_opts.zo_raidz_expand_test != RAIDZ_EXPAND_NONE)
8363 		VERIFY0(vdev_raidz_impl_set("cycle"));
8364 
8365 	dmu_objset_stats_t dds;
8366 	VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
8367 	    DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
8368 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
8369 	dmu_objset_fast_stat(os, &dds);
8370 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
8371 	dmu_objset_disown(os, B_TRUE, FTAG);
8372 
8373 	/* Give the dedicated raidz expansion test more grace time */
8374 	if (ztest_opts.zo_raidz_expand_test != RAIDZ_EXPAND_NONE)
8375 		zfs_deadman_synctime_ms *= 2;
8376 
8377 	/*
8378 	 * Create a thread to periodically resume suspended I/O.
8379 	 */
8380 	resume_thread = thread_create(NULL, 0, ztest_resume_thread,
8381 	    spa, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
8382 
8383 	/*
8384 	 * Create a deadman thread and set to panic if we hang.
8385 	 */
8386 	deadman_thread = thread_create(NULL, 0, ztest_deadman_thread,
8387 	    zs, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
8388 
8389 	spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
8390 
8391 	/*
8392 	 * Verify that we can safely inquire about any object,
8393 	 * whether it's allocated or not.  To make it interesting,
8394 	 * we probe a 5-wide window around each power of two.
8395 	 * This hits all edge cases, including zero and the max.
8396 	 */
8397 	for (t = 0; t < 64; t++) {
8398 		for (d = -5; d <= 5; d++) {
8399 			error = dmu_object_info(spa->spa_meta_objset,
8400 			    (1ULL << t) + d, NULL);
8401 			ASSERT(error == 0 || error == ENOENT ||
8402 			    error == EINVAL);
8403 		}
8404 	}
8405 
8406 	/*
8407 	 * If we got any ENOSPC errors on the previous run, destroy something.
8408 	 */
8409 	if (zs->zs_enospc_count != 0) {
8410 		/* Not expecting ENOSPC errors during raidz expansion tests */
8411 		ASSERT3U(ztest_opts.zo_raidz_expand_test, ==,
8412 		    RAIDZ_EXPAND_NONE);
8413 
8414 		int d = ztest_random(ztest_opts.zo_datasets);
8415 		ztest_dataset_destroy(d);
8416 	}
8417 	zs->zs_enospc_count = 0;
8418 
8419 	/*
8420 	 * If we were in the middle of ztest_device_removal() and were killed
8421 	 * we need to ensure the removal and scrub complete before running
8422 	 * any tests that check ztest_device_removal_active. The removal will
8423 	 * be restarted automatically when the spa is opened, but we need to
8424 	 * initiate the scrub manually if it is not already in progress. Note
8425 	 * that we always run the scrub whenever an indirect vdev exists
8426 	 * because we have no way of knowing for sure if ztest_device_removal()
8427 	 * fully completed its scrub before the pool was reimported.
8428 	 *
8429 	 * Does not apply for the RAIDZ expansion specific test runs
8430 	 */
8431 	if (ztest_opts.zo_raidz_expand_test == RAIDZ_EXPAND_NONE &&
8432 	    (spa->spa_removing_phys.sr_state == DSS_SCANNING ||
8433 	    spa->spa_removing_phys.sr_prev_indirect_vdev != -1)) {
8434 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
8435 			txg_wait_synced(spa_get_dsl(spa), 0);
8436 
8437 		error = ztest_scrub_impl(spa);
8438 		if (error == EBUSY)
8439 			error = 0;
8440 		ASSERT0(error);
8441 	}
8442 
8443 	if (ztest_opts.zo_verbose >= 4)
8444 		(void) printf("starting main threads...\n");
8445 
8446 	/*
8447 	 * Replay all logs of all datasets in the pool. This is primarily for
8448 	 * temporary datasets which wouldn't otherwise get replayed, which
8449 	 * can trigger failures when attempting to offline a SLOG in
8450 	 * ztest_fault_inject().
8451 	 */
8452 	(void) dmu_objset_find(ztest_opts.zo_pool, ztest_replay_zil_cb,
8453 	    NULL, DS_FIND_CHILDREN);
8454 
8455 	if (ztest_opts.zo_raidz_expand_test == RAIDZ_EXPAND_REQUESTED)
8456 		ztest_raidz_expand_run(zs, spa);
8457 	else if (ztest_opts.zo_raidz_expand_test == RAIDZ_EXPAND_KILLED)
8458 		ztest_raidz_expand_check(spa);
8459 	else
8460 		ztest_generic_run(zs, spa);
8461 
8462 	/* Kill the resume and deadman threads */
8463 	ztest_exiting = B_TRUE;
8464 	VERIFY0(thread_join(resume_thread));
8465 	VERIFY0(thread_join(deadman_thread));
8466 	ztest_resume(spa);
8467 
8468 	/*
8469 	 * Right before closing the pool, kick off a bunch of async I/O;
8470 	 * spa_close() should wait for it to complete.
8471 	 */
8472 	for (object = 1; object < 50; object++) {
8473 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
8474 		    ZIO_PRIORITY_SYNC_READ);
8475 	}
8476 
8477 	/* Verify that at least one commit cb was called in a timely fashion */
8478 	if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
8479 		VERIFY0(zc_min_txg_delay);
8480 
8481 	spa_close(spa, FTAG);
8482 
8483 	/*
8484 	 * Verify that we can loop over all pools.
8485 	 */
8486 	mutex_enter(&spa_namespace_lock);
8487 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
8488 		if (ztest_opts.zo_verbose > 3)
8489 			(void) printf("spa_next: found %s\n", spa_name(spa));
8490 	mutex_exit(&spa_namespace_lock);
8491 
8492 	/*
8493 	 * Verify that we can export the pool and reimport it under a
8494 	 * different name.
8495 	 */
8496 	if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
8497 		char name[ZFS_MAX_DATASET_NAME_LEN];
8498 		(void) snprintf(name, sizeof (name), "%s_import",
8499 		    ztest_opts.zo_pool);
8500 		ztest_spa_import_export(ztest_opts.zo_pool, name);
8501 		ztest_spa_import_export(name, ztest_opts.zo_pool);
8502 	}
8503 
8504 	kernel_fini();
8505 
8506 	list_destroy(&zcl.zcl_callbacks);
8507 	mutex_destroy(&zcl.zcl_callbacks_lock);
8508 	(void) pthread_rwlock_destroy(&ztest_name_lock);
8509 	mutex_destroy(&ztest_vdev_lock);
8510 	mutex_destroy(&ztest_checkpoint_lock);
8511 }
8512 
8513 static void
8514 print_time(hrtime_t t, char *timebuf)
8515 {
8516 	hrtime_t s = t / NANOSEC;
8517 	hrtime_t m = s / 60;
8518 	hrtime_t h = m / 60;
8519 	hrtime_t d = h / 24;
8520 
8521 	s -= m * 60;
8522 	m -= h * 60;
8523 	h -= d * 24;
8524 
8525 	timebuf[0] = '\0';
8526 
8527 	if (d)
8528 		(void) sprintf(timebuf,
8529 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
8530 	else if (h)
8531 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
8532 	else if (m)
8533 		(void) sprintf(timebuf, "%llum%02llus", m, s);
8534 	else
8535 		(void) sprintf(timebuf, "%llus", s);
8536 }
8537 
8538 static nvlist_t *
8539 make_random_pool_props(void)
8540 {
8541 	nvlist_t *props;
8542 
8543 	props = fnvlist_alloc();
8544 
8545 	/* Twenty percent of the time enable ZPOOL_PROP_DEDUP_TABLE_QUOTA */
8546 	if (ztest_random(5) == 0) {
8547 		fnvlist_add_uint64(props,
8548 		    zpool_prop_to_name(ZPOOL_PROP_DEDUP_TABLE_QUOTA),
8549 		    2 * 1024 * 1024);
8550 	}
8551 
8552 	/* Fifty percent of the time enable ZPOOL_PROP_AUTOREPLACE */
8553 	if (ztest_random(2) == 0) {
8554 		fnvlist_add_uint64(props,
8555 		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 1);
8556 	}
8557 
8558 	return (props);
8559 }
8560 
8561 /*
8562  * Create a storage pool with the given name and initial vdev size.
8563  * Then test spa_freeze() functionality.
8564  */
8565 static void
8566 ztest_init(ztest_shared_t *zs)
8567 {
8568 	spa_t *spa;
8569 	nvlist_t *nvroot, *props;
8570 	int i;
8571 
8572 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
8573 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
8574 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
8575 
8576 	raidz_scratch_verify();
8577 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
8578 
8579 	/*
8580 	 * Create the storage pool.
8581 	 */
8582 	(void) spa_destroy(ztest_opts.zo_pool);
8583 	ztest_shared->zs_vdev_next_leaf = 0;
8584 	zs->zs_splits = 0;
8585 	zs->zs_mirrors = ztest_opts.zo_mirrors;
8586 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
8587 	    NULL, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
8588 	props = make_random_pool_props();
8589 
8590 	/*
8591 	 * We don't expect the pool to suspend unless maxfaults == 0,
8592 	 * in which case ztest_fault_inject() temporarily takes away
8593 	 * the only valid replica.
8594 	 */
8595 	fnvlist_add_uint64(props,
8596 	    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
8597 	    MAXFAULTS(zs) ? ZIO_FAILURE_MODE_PANIC : ZIO_FAILURE_MODE_WAIT);
8598 
8599 	for (i = 0; i < SPA_FEATURES; i++) {
8600 		char *buf;
8601 
8602 		if (!spa_feature_table[i].fi_zfs_mod_supported)
8603 			continue;
8604 
8605 		/*
8606 		 * 75% chance of using the log space map feature. We want ztest
8607 		 * to exercise both the code paths that use the log space map
8608 		 * feature and the ones that don't.
8609 		 */
8610 		if (i == SPA_FEATURE_LOG_SPACEMAP && ztest_random(4) == 0)
8611 			continue;
8612 
8613 		/*
8614 		 * split 50/50 between legacy and fast dedup
8615 		 */
8616 		if (i == SPA_FEATURE_FAST_DEDUP && ztest_random(2) != 0)
8617 			continue;
8618 
8619 		VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
8620 		    spa_feature_table[i].fi_uname));
8621 		fnvlist_add_uint64(props, buf, 0);
8622 		free(buf);
8623 	}
8624 
8625 	VERIFY0(spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
8626 	fnvlist_free(nvroot);
8627 	fnvlist_free(props);
8628 
8629 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
8630 	zs->zs_metaslab_sz =
8631 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
8632 	zs->zs_guid = spa_guid(spa);
8633 	spa_close(spa, FTAG);
8634 
8635 	kernel_fini();
8636 
8637 	if (!ztest_opts.zo_mmp_test) {
8638 		ztest_run_zdb(zs->zs_guid);
8639 		ztest_freeze();
8640 		ztest_run_zdb(zs->zs_guid);
8641 	}
8642 
8643 	(void) pthread_rwlock_destroy(&ztest_name_lock);
8644 	mutex_destroy(&ztest_vdev_lock);
8645 	mutex_destroy(&ztest_checkpoint_lock);
8646 }
8647 
8648 static void
8649 setup_data_fd(void)
8650 {
8651 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
8652 
8653 	ztest_fd_data = mkstemp(ztest_name_data);
8654 	ASSERT3S(ztest_fd_data, >=, 0);
8655 	(void) unlink(ztest_name_data);
8656 }
8657 
8658 static int
8659 shared_data_size(ztest_shared_hdr_t *hdr)
8660 {
8661 	int size;
8662 
8663 	size = hdr->zh_hdr_size;
8664 	size += hdr->zh_opts_size;
8665 	size += hdr->zh_size;
8666 	size += hdr->zh_stats_size * hdr->zh_stats_count;
8667 	size += hdr->zh_ds_size * hdr->zh_ds_count;
8668 	size += hdr->zh_scratch_state_size;
8669 
8670 	return (size);
8671 }
8672 
8673 static void
8674 setup_hdr(void)
8675 {
8676 	int size;
8677 	ztest_shared_hdr_t *hdr;
8678 
8679 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
8680 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
8681 	ASSERT3P(hdr, !=, MAP_FAILED);
8682 
8683 	VERIFY0(ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
8684 
8685 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
8686 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
8687 	hdr->zh_size = sizeof (ztest_shared_t);
8688 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
8689 	hdr->zh_stats_count = ZTEST_FUNCS;
8690 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
8691 	hdr->zh_ds_count = ztest_opts.zo_datasets;
8692 	hdr->zh_scratch_state_size = sizeof (ztest_shared_scratch_state_t);
8693 
8694 	size = shared_data_size(hdr);
8695 	VERIFY0(ftruncate(ztest_fd_data, size));
8696 
8697 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
8698 }
8699 
8700 static void
8701 setup_data(void)
8702 {
8703 	int size, offset;
8704 	ztest_shared_hdr_t *hdr;
8705 	uint8_t *buf;
8706 
8707 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
8708 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
8709 	ASSERT3P(hdr, !=, MAP_FAILED);
8710 
8711 	size = shared_data_size(hdr);
8712 
8713 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
8714 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
8715 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
8716 	ASSERT3P(hdr, !=, MAP_FAILED);
8717 	buf = (uint8_t *)hdr;
8718 
8719 	offset = hdr->zh_hdr_size;
8720 	ztest_shared_opts = (void *)&buf[offset];
8721 	offset += hdr->zh_opts_size;
8722 	ztest_shared = (void *)&buf[offset];
8723 	offset += hdr->zh_size;
8724 	ztest_shared_callstate = (void *)&buf[offset];
8725 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
8726 	ztest_shared_ds = (void *)&buf[offset];
8727 	offset += hdr->zh_ds_size * hdr->zh_ds_count;
8728 	ztest_scratch_state = (void *)&buf[offset];
8729 }
8730 
8731 static boolean_t
8732 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
8733 {
8734 	pid_t pid;
8735 	int status;
8736 	char *cmdbuf = NULL;
8737 
8738 	pid = fork();
8739 
8740 	if (cmd == NULL) {
8741 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
8742 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
8743 		cmd = cmdbuf;
8744 	}
8745 
8746 	if (pid == -1)
8747 		fatal(B_TRUE, "fork failed");
8748 
8749 	if (pid == 0) {	/* child */
8750 		char fd_data_str[12];
8751 
8752 		VERIFY3S(11, >=,
8753 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
8754 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
8755 
8756 		if (libpath != NULL) {
8757 			const char *curlp = getenv("LD_LIBRARY_PATH");
8758 			if (curlp == NULL)
8759 				VERIFY0(setenv("LD_LIBRARY_PATH", libpath, 1));
8760 			else {
8761 				char *newlp = NULL;
8762 				VERIFY3S(-1, !=,
8763 				    asprintf(&newlp, "%s:%s", libpath, curlp));
8764 				VERIFY0(setenv("LD_LIBRARY_PATH", newlp, 1));
8765 				free(newlp);
8766 			}
8767 		}
8768 		(void) execl(cmd, cmd, (char *)NULL);
8769 		ztest_dump_core = B_FALSE;
8770 		fatal(B_TRUE, "exec failed: %s", cmd);
8771 	}
8772 
8773 	if (cmdbuf != NULL) {
8774 		umem_free(cmdbuf, MAXPATHLEN);
8775 		cmd = NULL;
8776 	}
8777 
8778 	while (waitpid(pid, &status, 0) != pid)
8779 		continue;
8780 	if (statusp != NULL)
8781 		*statusp = status;
8782 
8783 	if (WIFEXITED(status)) {
8784 		if (WEXITSTATUS(status) != 0) {
8785 			(void) fprintf(stderr, "child exited with code %d\n",
8786 			    WEXITSTATUS(status));
8787 			exit(2);
8788 		}
8789 		return (B_FALSE);
8790 	} else if (WIFSIGNALED(status)) {
8791 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
8792 			(void) fprintf(stderr, "child died with signal %d\n",
8793 			    WTERMSIG(status));
8794 			exit(3);
8795 		}
8796 		return (B_TRUE);
8797 	} else {
8798 		(void) fprintf(stderr, "something strange happened to child\n");
8799 		exit(4);
8800 	}
8801 }
8802 
8803 static void
8804 ztest_run_init(void)
8805 {
8806 	int i;
8807 
8808 	ztest_shared_t *zs = ztest_shared;
8809 
8810 	/*
8811 	 * Blow away any existing copy of zpool.cache
8812 	 */
8813 	(void) remove(spa_config_path);
8814 
8815 	if (ztest_opts.zo_init == 0) {
8816 		if (ztest_opts.zo_verbose >= 1)
8817 			(void) printf("Importing pool %s\n",
8818 			    ztest_opts.zo_pool);
8819 		ztest_import(zs);
8820 		return;
8821 	}
8822 
8823 	/*
8824 	 * Create and initialize our storage pool.
8825 	 */
8826 	for (i = 1; i <= ztest_opts.zo_init; i++) {
8827 		memset(zs, 0, sizeof (*zs));
8828 		if (ztest_opts.zo_verbose >= 3 &&
8829 		    ztest_opts.zo_init != 1) {
8830 			(void) printf("ztest_init(), pass %d\n", i);
8831 		}
8832 		ztest_init(zs);
8833 	}
8834 }
8835 
8836 int
8837 main(int argc, char **argv)
8838 {
8839 	int kills = 0;
8840 	int iters = 0;
8841 	int older = 0;
8842 	int newer = 0;
8843 	ztest_shared_t *zs;
8844 	ztest_info_t *zi;
8845 	ztest_shared_callstate_t *zc;
8846 	char timebuf[100];
8847 	char numbuf[NN_NUMBUF_SZ];
8848 	char *cmd;
8849 	boolean_t hasalt;
8850 	int f, err;
8851 	char *fd_data_str = getenv("ZTEST_FD_DATA");
8852 	struct sigaction action;
8853 
8854 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
8855 
8856 	dprintf_setup(&argc, argv);
8857 	zfs_deadman_synctime_ms = 300000;
8858 	zfs_deadman_checktime_ms = 30000;
8859 	/*
8860 	 * As two-word space map entries may not come up often (especially
8861 	 * if pool and vdev sizes are small) we want to force at least some
8862 	 * of them so the feature get tested.
8863 	 */
8864 	zfs_force_some_double_word_sm_entries = B_TRUE;
8865 
8866 	/*
8867 	 * Verify that even extensively damaged split blocks with many
8868 	 * segments can be reconstructed in a reasonable amount of time
8869 	 * when reconstruction is known to be possible.
8870 	 *
8871 	 * Note: the lower this value is, the more damage we inflict, and
8872 	 * the more time ztest spends in recovering that damage. We chose
8873 	 * to induce damage 1/100th of the time so recovery is tested but
8874 	 * not so frequently that ztest doesn't get to test other code paths.
8875 	 */
8876 	zfs_reconstruct_indirect_damage_fraction = 100;
8877 
8878 	action.sa_handler = sig_handler;
8879 	sigemptyset(&action.sa_mask);
8880 	action.sa_flags = 0;
8881 
8882 	if (sigaction(SIGSEGV, &action, NULL) < 0) {
8883 		(void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
8884 		    strerror(errno));
8885 		exit(EXIT_FAILURE);
8886 	}
8887 
8888 	if (sigaction(SIGABRT, &action, NULL) < 0) {
8889 		(void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
8890 		    strerror(errno));
8891 		exit(EXIT_FAILURE);
8892 	}
8893 
8894 	/*
8895 	 * Force random_get_bytes() to use /dev/urandom in order to prevent
8896 	 * ztest from needlessly depleting the system entropy pool.
8897 	 */
8898 	random_path = "/dev/urandom";
8899 	ztest_fd_rand = open(random_path, O_RDONLY | O_CLOEXEC);
8900 	ASSERT3S(ztest_fd_rand, >=, 0);
8901 
8902 	if (!fd_data_str) {
8903 		process_options(argc, argv);
8904 
8905 		setup_data_fd();
8906 		setup_hdr();
8907 		setup_data();
8908 		memcpy(ztest_shared_opts, &ztest_opts,
8909 		    sizeof (*ztest_shared_opts));
8910 	} else {
8911 		ztest_fd_data = atoi(fd_data_str);
8912 		setup_data();
8913 		memcpy(&ztest_opts, ztest_shared_opts, sizeof (ztest_opts));
8914 	}
8915 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
8916 
8917 	err = ztest_set_global_vars();
8918 	if (err != 0 && !fd_data_str) {
8919 		/* error message done by ztest_set_global_vars */
8920 		exit(EXIT_FAILURE);
8921 	} else {
8922 		/* children should not be spawned if setting gvars fails */
8923 		VERIFY3S(err, ==, 0);
8924 	}
8925 
8926 	/* Override location of zpool.cache */
8927 	VERIFY3S(asprintf((char **)&spa_config_path, "%s/zpool.cache",
8928 	    ztest_opts.zo_dir), !=, -1);
8929 
8930 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
8931 	    UMEM_NOFAIL);
8932 	zs = ztest_shared;
8933 
8934 	if (fd_data_str) {
8935 		metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
8936 		metaslab_df_alloc_threshold =
8937 		    zs->zs_metaslab_df_alloc_threshold;
8938 
8939 		if (zs->zs_do_init)
8940 			ztest_run_init();
8941 		else
8942 			ztest_run(zs);
8943 		exit(0);
8944 	}
8945 
8946 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
8947 
8948 	if (ztest_opts.zo_verbose >= 1) {
8949 		(void) printf("%"PRIu64" vdevs, %d datasets, %d threads, "
8950 		    "%d %s disks, parity %d, %"PRIu64" seconds...\n\n",
8951 		    ztest_opts.zo_vdevs,
8952 		    ztest_opts.zo_datasets,
8953 		    ztest_opts.zo_threads,
8954 		    ztest_opts.zo_raid_children,
8955 		    ztest_opts.zo_raid_type,
8956 		    ztest_opts.zo_raid_parity,
8957 		    ztest_opts.zo_time);
8958 	}
8959 
8960 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
8961 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
8962 
8963 	zs->zs_do_init = B_TRUE;
8964 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
8965 		if (ztest_opts.zo_verbose >= 1) {
8966 			(void) printf("Executing older ztest for "
8967 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
8968 		}
8969 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
8970 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
8971 	} else {
8972 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
8973 	}
8974 	zs->zs_do_init = B_FALSE;
8975 
8976 	zs->zs_proc_start = gethrtime();
8977 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
8978 
8979 	for (f = 0; f < ZTEST_FUNCS; f++) {
8980 		zi = &ztest_info[f];
8981 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
8982 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
8983 			zc->zc_next = UINT64_MAX;
8984 		else
8985 			zc->zc_next = zs->zs_proc_start +
8986 			    ztest_random(2 * zi->zi_interval[0] + 1);
8987 	}
8988 
8989 	/*
8990 	 * Run the tests in a loop.  These tests include fault injection
8991 	 * to verify that self-healing data works, and forced crashes
8992 	 * to verify that we never lose on-disk consistency.
8993 	 */
8994 	while (gethrtime() < zs->zs_proc_stop) {
8995 		int status;
8996 		boolean_t killed;
8997 
8998 		/*
8999 		 * Initialize the workload counters for each function.
9000 		 */
9001 		for (f = 0; f < ZTEST_FUNCS; f++) {
9002 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
9003 			zc->zc_count = 0;
9004 			zc->zc_time = 0;
9005 		}
9006 
9007 		/* Set the allocation switch size */
9008 		zs->zs_metaslab_df_alloc_threshold =
9009 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
9010 
9011 		if (!hasalt || ztest_random(2) == 0) {
9012 			if (hasalt && ztest_opts.zo_verbose >= 1) {
9013 				(void) printf("Executing newer ztest: %s\n",
9014 				    cmd);
9015 			}
9016 			newer++;
9017 			killed = exec_child(cmd, NULL, B_TRUE, &status);
9018 		} else {
9019 			if (hasalt && ztest_opts.zo_verbose >= 1) {
9020 				(void) printf("Executing older ztest: %s\n",
9021 				    ztest_opts.zo_alt_ztest);
9022 			}
9023 			older++;
9024 			killed = exec_child(ztest_opts.zo_alt_ztest,
9025 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
9026 		}
9027 
9028 		if (killed)
9029 			kills++;
9030 		iters++;
9031 
9032 		if (ztest_opts.zo_verbose >= 1) {
9033 			hrtime_t now = gethrtime();
9034 
9035 			now = MIN(now, zs->zs_proc_stop);
9036 			print_time(zs->zs_proc_stop - now, timebuf);
9037 			nicenum(zs->zs_space, numbuf, sizeof (numbuf));
9038 
9039 			(void) printf("Pass %3d, %8s, %3"PRIu64" ENOSPC, "
9040 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
9041 			    iters,
9042 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
9043 			    zs->zs_enospc_count,
9044 			    100.0 * zs->zs_alloc / zs->zs_space,
9045 			    numbuf,
9046 			    100.0 * (now - zs->zs_proc_start) /
9047 			    (ztest_opts.zo_time * NANOSEC), timebuf);
9048 		}
9049 
9050 		if (ztest_opts.zo_verbose >= 2) {
9051 			(void) printf("\nWorkload summary:\n\n");
9052 			(void) printf("%7s %9s   %s\n",
9053 			    "Calls", "Time", "Function");
9054 			(void) printf("%7s %9s   %s\n",
9055 			    "-----", "----", "--------");
9056 			for (f = 0; f < ZTEST_FUNCS; f++) {
9057 				zi = &ztest_info[f];
9058 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
9059 				print_time(zc->zc_time, timebuf);
9060 				(void) printf("%7"PRIu64" %9s   %s\n",
9061 				    zc->zc_count, timebuf,
9062 				    zi->zi_funcname);
9063 			}
9064 			(void) printf("\n");
9065 		}
9066 
9067 		if (!ztest_opts.zo_mmp_test)
9068 			ztest_run_zdb(zs->zs_guid);
9069 		if (ztest_shared_opts->zo_raidz_expand_test ==
9070 		    RAIDZ_EXPAND_CHECKED)
9071 			break; /* raidz expand test complete */
9072 	}
9073 
9074 	if (ztest_opts.zo_verbose >= 1) {
9075 		if (hasalt) {
9076 			(void) printf("%d runs of older ztest: %s\n", older,
9077 			    ztest_opts.zo_alt_ztest);
9078 			(void) printf("%d runs of newer ztest: %s\n", newer,
9079 			    cmd);
9080 		}
9081 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
9082 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
9083 	}
9084 
9085 	umem_free(cmd, MAXNAMELEN);
9086 
9087 	return (0);
9088 }
9089