xref: /freebsd/sys/contrib/openzfs/cmd/ztest.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or https://opensource.org/licenses/CDDL-1.0.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2017 Joyent, Inc.
28  * Copyright (c) 2017, Intel Corporation.
29  */
30 
31 /*
32  * The objective of this program is to provide a DMU/ZAP/SPA stress test
33  * that runs entirely in userland, is easy to use, and easy to extend.
34  *
35  * The overall design of the ztest program is as follows:
36  *
37  * (1) For each major functional area (e.g. adding vdevs to a pool,
38  *     creating and destroying datasets, reading and writing objects, etc)
39  *     we have a simple routine to test that functionality.  These
40  *     individual routines do not have to do anything "stressful".
41  *
42  * (2) We turn these simple functionality tests into a stress test by
43  *     running them all in parallel, with as many threads as desired,
44  *     and spread across as many datasets, objects, and vdevs as desired.
45  *
46  * (3) While all this is happening, we inject faults into the pool to
47  *     verify that self-healing data really works.
48  *
49  * (4) Every time we open a dataset, we change its checksum and compression
50  *     functions.  Thus even individual objects vary from block to block
51  *     in which checksum they use and whether they're compressed.
52  *
53  * (5) To verify that we never lose on-disk consistency after a crash,
54  *     we run the entire test in a child of the main process.
55  *     At random times, the child self-immolates with a SIGKILL.
56  *     This is the software equivalent of pulling the power cord.
57  *     The parent then runs the test again, using the existing
58  *     storage pool, as many times as desired. If backwards compatibility
59  *     testing is enabled ztest will sometimes run the "older" version
60  *     of ztest after a SIGKILL.
61  *
62  * (6) To verify that we don't have future leaks or temporal incursions,
63  *     many of the functional tests record the transaction group number
64  *     as part of their data.  When reading old data, they verify that
65  *     the transaction group number is less than the current, open txg.
66  *     If you add a new test, please do this if applicable.
67  *
68  * (7) Threads are created with a reduced stack size, for sanity checking.
69  *     Therefore, it's important not to allocate huge buffers on the stack.
70  *
71  * When run with no arguments, ztest runs for about five minutes and
72  * produces no output if successful.  To get a little bit of information,
73  * specify -V.  To get more information, specify -VV, and so on.
74  *
75  * To turn this into an overnight stress test, use -T to specify run time.
76  *
77  * You can ask more vdevs [-v], datasets [-d], or threads [-t]
78  * to increase the pool capacity, fanout, and overall stress level.
79  *
80  * Use the -k option to set the desired frequency of kills.
81  *
82  * When ztest invokes itself it passes all relevant information through a
83  * temporary file which is mmap-ed in the child process. This allows shared
84  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
85  * stored at offset 0 of this file and contains information on the size and
86  * number of shared structures in the file. The information stored in this file
87  * must remain backwards compatible with older versions of ztest so that
88  * ztest can invoke them during backwards compatibility testing (-B).
89  */
90 
91 #include <sys/zfs_context.h>
92 #include <sys/spa.h>
93 #include <sys/dmu.h>
94 #include <sys/txg.h>
95 #include <sys/dbuf.h>
96 #include <sys/zap.h>
97 #include <sys/dmu_objset.h>
98 #include <sys/poll.h>
99 #include <sys/stat.h>
100 #include <sys/time.h>
101 #include <sys/wait.h>
102 #include <sys/mman.h>
103 #include <sys/resource.h>
104 #include <sys/zio.h>
105 #include <sys/zil.h>
106 #include <sys/zil_impl.h>
107 #include <sys/vdev_draid.h>
108 #include <sys/vdev_impl.h>
109 #include <sys/vdev_file.h>
110 #include <sys/vdev_initialize.h>
111 #include <sys/vdev_raidz.h>
112 #include <sys/vdev_trim.h>
113 #include <sys/spa_impl.h>
114 #include <sys/metaslab_impl.h>
115 #include <sys/dsl_prop.h>
116 #include <sys/dsl_dataset.h>
117 #include <sys/dsl_destroy.h>
118 #include <sys/dsl_scan.h>
119 #include <sys/zio_checksum.h>
120 #include <sys/zfs_refcount.h>
121 #include <sys/zfeature.h>
122 #include <sys/dsl_userhold.h>
123 #include <sys/abd.h>
124 #include <sys/blake3.h>
125 #include <stdio.h>
126 #include <stdlib.h>
127 #include <unistd.h>
128 #include <getopt.h>
129 #include <signal.h>
130 #include <umem.h>
131 #include <ctype.h>
132 #include <math.h>
133 #include <sys/fs/zfs.h>
134 #include <zfs_fletcher.h>
135 #include <libnvpair.h>
136 #include <libzutil.h>
137 #include <sys/crypto/icp.h>
138 #include <sys/zfs_impl.h>
139 #if (__GLIBC__ && !__UCLIBC__)
140 #include <execinfo.h> /* for backtrace() */
141 #endif
142 
143 static int ztest_fd_data = -1;
144 static int ztest_fd_rand = -1;
145 
146 typedef struct ztest_shared_hdr {
147 	uint64_t	zh_hdr_size;
148 	uint64_t	zh_opts_size;
149 	uint64_t	zh_size;
150 	uint64_t	zh_stats_size;
151 	uint64_t	zh_stats_count;
152 	uint64_t	zh_ds_size;
153 	uint64_t	zh_ds_count;
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 #define	ZO_GVARS_MAX_ARGLEN	((size_t)64)
165 #define	ZO_GVARS_MAX_COUNT	((size_t)10)
166 
167 typedef struct ztest_shared_opts {
168 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
169 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
170 	char zo_alt_ztest[MAXNAMELEN];
171 	char zo_alt_libpath[MAXNAMELEN];
172 	uint64_t zo_vdevs;
173 	uint64_t zo_vdevtime;
174 	size_t zo_vdev_size;
175 	int zo_ashift;
176 	int zo_mirrors;
177 	int zo_raid_children;
178 	int zo_raid_parity;
179 	char zo_raid_type[8];
180 	int zo_draid_data;
181 	int zo_draid_spares;
182 	int zo_datasets;
183 	int zo_threads;
184 	uint64_t zo_passtime;
185 	uint64_t zo_killrate;
186 	int zo_verbose;
187 	int zo_init;
188 	uint64_t zo_time;
189 	uint64_t zo_maxloops;
190 	uint64_t zo_metaslab_force_ganging;
191 	int zo_mmp_test;
192 	int zo_special_vdevs;
193 	int zo_dump_dbgmsg;
194 	int zo_gvars_count;
195 	char zo_gvars[ZO_GVARS_MAX_COUNT][ZO_GVARS_MAX_ARGLEN];
196 } ztest_shared_opts_t;
197 
198 /* Default values for command line options. */
199 #define	DEFAULT_POOL "ztest"
200 #define	DEFAULT_VDEV_DIR "/tmp"
201 #define	DEFAULT_VDEV_COUNT 5
202 #define	DEFAULT_VDEV_SIZE (SPA_MINDEVSIZE * 4)	/* 256m default size */
203 #define	DEFAULT_VDEV_SIZE_STR "256M"
204 #define	DEFAULT_ASHIFT SPA_MINBLOCKSHIFT
205 #define	DEFAULT_MIRRORS 2
206 #define	DEFAULT_RAID_CHILDREN 4
207 #define	DEFAULT_RAID_PARITY 1
208 #define	DEFAULT_DRAID_DATA 4
209 #define	DEFAULT_DRAID_SPARES 1
210 #define	DEFAULT_DATASETS_COUNT 7
211 #define	DEFAULT_THREADS 23
212 #define	DEFAULT_RUN_TIME 300 /* 300 seconds */
213 #define	DEFAULT_RUN_TIME_STR "300 sec"
214 #define	DEFAULT_PASS_TIME 60 /* 60 seconds */
215 #define	DEFAULT_PASS_TIME_STR "60 sec"
216 #define	DEFAULT_KILL_RATE 70 /* 70% kill rate */
217 #define	DEFAULT_KILLRATE_STR "70%"
218 #define	DEFAULT_INITS 1
219 #define	DEFAULT_MAX_LOOPS 50 /* 5 minutes */
220 #define	DEFAULT_FORCE_GANGING (64 << 10)
221 #define	DEFAULT_FORCE_GANGING_STR "64K"
222 
223 /* Simplifying assumption: -1 is not a valid default. */
224 #define	NO_DEFAULT -1
225 
226 static const ztest_shared_opts_t ztest_opts_defaults = {
227 	.zo_pool = DEFAULT_POOL,
228 	.zo_dir = DEFAULT_VDEV_DIR,
229 	.zo_alt_ztest = { '\0' },
230 	.zo_alt_libpath = { '\0' },
231 	.zo_vdevs = DEFAULT_VDEV_COUNT,
232 	.zo_ashift = DEFAULT_ASHIFT,
233 	.zo_mirrors = DEFAULT_MIRRORS,
234 	.zo_raid_children = DEFAULT_RAID_CHILDREN,
235 	.zo_raid_parity = DEFAULT_RAID_PARITY,
236 	.zo_raid_type = VDEV_TYPE_RAIDZ,
237 	.zo_vdev_size = DEFAULT_VDEV_SIZE,
238 	.zo_draid_data = DEFAULT_DRAID_DATA,	/* data drives */
239 	.zo_draid_spares = DEFAULT_DRAID_SPARES, /* distributed spares */
240 	.zo_datasets = DEFAULT_DATASETS_COUNT,
241 	.zo_threads = DEFAULT_THREADS,
242 	.zo_passtime = DEFAULT_PASS_TIME,
243 	.zo_killrate = DEFAULT_KILL_RATE,
244 	.zo_verbose = 0,
245 	.zo_mmp_test = 0,
246 	.zo_init = DEFAULT_INITS,
247 	.zo_time = DEFAULT_RUN_TIME,
248 	.zo_maxloops = DEFAULT_MAX_LOOPS, /* max loops during spa_freeze() */
249 	.zo_metaslab_force_ganging = DEFAULT_FORCE_GANGING,
250 	.zo_special_vdevs = ZTEST_VDEV_CLASS_RND,
251 	.zo_gvars_count = 0,
252 };
253 
254 extern uint64_t metaslab_force_ganging;
255 extern uint64_t metaslab_df_alloc_threshold;
256 extern uint64_t zfs_deadman_synctime_ms;
257 extern uint_t metaslab_preload_limit;
258 extern int zfs_compressed_arc_enabled;
259 extern int zfs_abd_scatter_enabled;
260 extern uint_t dmu_object_alloc_chunk_shift;
261 extern boolean_t zfs_force_some_double_word_sm_entries;
262 extern unsigned long zio_decompress_fail_fraction;
263 extern unsigned long zfs_reconstruct_indirect_damage_fraction;
264 
265 
266 static ztest_shared_opts_t *ztest_shared_opts;
267 static ztest_shared_opts_t ztest_opts;
268 static const char *const ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
269 
270 typedef struct ztest_shared_ds {
271 	uint64_t	zd_seq;
272 } ztest_shared_ds_t;
273 
274 static ztest_shared_ds_t *ztest_shared_ds;
275 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
276 
277 #define	BT_MAGIC	0x123456789abcdefULL
278 #define	MAXFAULTS(zs) \
279 	(MAX((zs)->zs_mirrors, 1) * (ztest_opts.zo_raid_parity + 1) - 1)
280 
281 enum ztest_io_type {
282 	ZTEST_IO_WRITE_TAG,
283 	ZTEST_IO_WRITE_PATTERN,
284 	ZTEST_IO_WRITE_ZEROES,
285 	ZTEST_IO_TRUNCATE,
286 	ZTEST_IO_SETATTR,
287 	ZTEST_IO_REWRITE,
288 	ZTEST_IO_TYPES
289 };
290 
291 typedef struct ztest_block_tag {
292 	uint64_t	bt_magic;
293 	uint64_t	bt_objset;
294 	uint64_t	bt_object;
295 	uint64_t	bt_dnodesize;
296 	uint64_t	bt_offset;
297 	uint64_t	bt_gen;
298 	uint64_t	bt_txg;
299 	uint64_t	bt_crtxg;
300 } ztest_block_tag_t;
301 
302 typedef struct bufwad {
303 	uint64_t	bw_index;
304 	uint64_t	bw_txg;
305 	uint64_t	bw_data;
306 } bufwad_t;
307 
308 /*
309  * It would be better to use a rangelock_t per object.  Unfortunately
310  * the rangelock_t is not a drop-in replacement for rl_t, because we
311  * still need to map from object ID to rangelock_t.
312  */
313 typedef enum {
314 	RL_READER,
315 	RL_WRITER,
316 	RL_APPEND
317 } rl_type_t;
318 
319 typedef struct rll {
320 	void		*rll_writer;
321 	int		rll_readers;
322 	kmutex_t	rll_lock;
323 	kcondvar_t	rll_cv;
324 } rll_t;
325 
326 typedef struct rl {
327 	uint64_t	rl_object;
328 	uint64_t	rl_offset;
329 	uint64_t	rl_size;
330 	rll_t		*rl_lock;
331 } rl_t;
332 
333 #define	ZTEST_RANGE_LOCKS	64
334 #define	ZTEST_OBJECT_LOCKS	64
335 
336 /*
337  * Object descriptor.  Used as a template for object lookup/create/remove.
338  */
339 typedef struct ztest_od {
340 	uint64_t	od_dir;
341 	uint64_t	od_object;
342 	dmu_object_type_t od_type;
343 	dmu_object_type_t od_crtype;
344 	uint64_t	od_blocksize;
345 	uint64_t	od_crblocksize;
346 	uint64_t	od_crdnodesize;
347 	uint64_t	od_gen;
348 	uint64_t	od_crgen;
349 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
350 } ztest_od_t;
351 
352 /*
353  * Per-dataset state.
354  */
355 typedef struct ztest_ds {
356 	ztest_shared_ds_t *zd_shared;
357 	objset_t	*zd_os;
358 	pthread_rwlock_t zd_zilog_lock;
359 	zilog_t		*zd_zilog;
360 	ztest_od_t	*zd_od;		/* debugging aid */
361 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
362 	kmutex_t	zd_dirobj_lock;
363 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
364 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
365 } ztest_ds_t;
366 
367 /*
368  * Per-iteration state.
369  */
370 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
371 
372 typedef struct ztest_info {
373 	ztest_func_t	*zi_func;	/* test function */
374 	uint64_t	zi_iters;	/* iterations per execution */
375 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
376 	const char	*zi_funcname;	/* name of test function */
377 } ztest_info_t;
378 
379 typedef struct ztest_shared_callstate {
380 	uint64_t	zc_count;	/* per-pass count */
381 	uint64_t	zc_time;	/* per-pass time */
382 	uint64_t	zc_next;	/* next time to call this function */
383 } ztest_shared_callstate_t;
384 
385 static ztest_shared_callstate_t *ztest_shared_callstate;
386 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
387 
388 ztest_func_t ztest_dmu_read_write;
389 ztest_func_t ztest_dmu_write_parallel;
390 ztest_func_t ztest_dmu_object_alloc_free;
391 ztest_func_t ztest_dmu_object_next_chunk;
392 ztest_func_t ztest_dmu_commit_callbacks;
393 ztest_func_t ztest_zap;
394 ztest_func_t ztest_zap_parallel;
395 ztest_func_t ztest_zil_commit;
396 ztest_func_t ztest_zil_remount;
397 ztest_func_t ztest_dmu_read_write_zcopy;
398 ztest_func_t ztest_dmu_objset_create_destroy;
399 ztest_func_t ztest_dmu_prealloc;
400 ztest_func_t ztest_fzap;
401 ztest_func_t ztest_dmu_snapshot_create_destroy;
402 ztest_func_t ztest_dsl_prop_get_set;
403 ztest_func_t ztest_spa_prop_get_set;
404 ztest_func_t ztest_spa_create_destroy;
405 ztest_func_t ztest_fault_inject;
406 ztest_func_t ztest_dmu_snapshot_hold;
407 ztest_func_t ztest_mmp_enable_disable;
408 ztest_func_t ztest_scrub;
409 ztest_func_t ztest_dsl_dataset_promote_busy;
410 ztest_func_t ztest_vdev_attach_detach;
411 ztest_func_t ztest_vdev_LUN_growth;
412 ztest_func_t ztest_vdev_add_remove;
413 ztest_func_t ztest_vdev_class_add;
414 ztest_func_t ztest_vdev_aux_add_remove;
415 ztest_func_t ztest_split_pool;
416 ztest_func_t ztest_reguid;
417 ztest_func_t ztest_spa_upgrade;
418 ztest_func_t ztest_device_removal;
419 ztest_func_t ztest_spa_checkpoint_create_discard;
420 ztest_func_t ztest_initialize;
421 ztest_func_t ztest_trim;
422 ztest_func_t ztest_blake3;
423 ztest_func_t ztest_fletcher;
424 ztest_func_t ztest_fletcher_incr;
425 ztest_func_t ztest_verify_dnode_bt;
426 
427 static uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
428 static uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
429 static uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
430 static uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
431 static uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
432 
433 #define	ZTI_INIT(func, iters, interval) \
434 	{   .zi_func = (func), \
435 	    .zi_iters = (iters), \
436 	    .zi_interval = (interval), \
437 	    .zi_funcname = # func }
438 
439 static ztest_info_t ztest_info[] = {
440 	ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
441 	ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
442 	ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
443 	ZTI_INIT(ztest_dmu_object_next_chunk, 1, &zopt_sometimes),
444 	ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
445 	ZTI_INIT(ztest_zap, 30, &zopt_always),
446 	ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
447 	ZTI_INIT(ztest_split_pool, 1, &zopt_sometimes),
448 	ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
449 	ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
450 	ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
451 	ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
452 	ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
453 	ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
454 #if 0
455 	ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
456 #endif
457 	ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
458 	ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
459 	ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
460 	ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
461 	ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
462 	ZTI_INIT(ztest_mmp_enable_disable, 1, &zopt_sometimes),
463 	ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
464 	ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
465 	ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
466 	ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
467 	ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
468 	ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
469 	ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
470 	ZTI_INIT(ztest_vdev_class_add, 1, &ztest_opts.zo_vdevtime),
471 	ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
472 	ZTI_INIT(ztest_device_removal, 1, &zopt_sometimes),
473 	ZTI_INIT(ztest_spa_checkpoint_create_discard, 1, &zopt_rarely),
474 	ZTI_INIT(ztest_initialize, 1, &zopt_sometimes),
475 	ZTI_INIT(ztest_trim, 1, &zopt_sometimes),
476 	ZTI_INIT(ztest_blake3, 1, &zopt_rarely),
477 	ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
478 	ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
479 	ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
480 };
481 
482 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
483 
484 /*
485  * The following struct is used to hold a list of uncalled commit callbacks.
486  * The callbacks are ordered by txg number.
487  */
488 typedef struct ztest_cb_list {
489 	kmutex_t	zcl_callbacks_lock;
490 	list_t		zcl_callbacks;
491 } ztest_cb_list_t;
492 
493 /*
494  * Stuff we need to share writably between parent and child.
495  */
496 typedef struct ztest_shared {
497 	boolean_t	zs_do_init;
498 	hrtime_t	zs_proc_start;
499 	hrtime_t	zs_proc_stop;
500 	hrtime_t	zs_thread_start;
501 	hrtime_t	zs_thread_stop;
502 	hrtime_t	zs_thread_kill;
503 	uint64_t	zs_enospc_count;
504 	uint64_t	zs_vdev_next_leaf;
505 	uint64_t	zs_vdev_aux;
506 	uint64_t	zs_alloc;
507 	uint64_t	zs_space;
508 	uint64_t	zs_splits;
509 	uint64_t	zs_mirrors;
510 	uint64_t	zs_metaslab_sz;
511 	uint64_t	zs_metaslab_df_alloc_threshold;
512 	uint64_t	zs_guid;
513 } ztest_shared_t;
514 
515 #define	ID_PARALLEL	-1ULL
516 
517 static char ztest_dev_template[] = "%s/%s.%llua";
518 static char ztest_aux_template[] = "%s/%s.%s.%llu";
519 static ztest_shared_t *ztest_shared;
520 
521 static spa_t *ztest_spa = NULL;
522 static ztest_ds_t *ztest_ds;
523 
524 static kmutex_t ztest_vdev_lock;
525 static boolean_t ztest_device_removal_active = B_FALSE;
526 static boolean_t ztest_pool_scrubbed = B_FALSE;
527 static kmutex_t ztest_checkpoint_lock;
528 
529 /*
530  * The ztest_name_lock protects the pool and dataset namespace used by
531  * the individual tests. To modify the namespace, consumers must grab
532  * this lock as writer. Grabbing the lock as reader will ensure that the
533  * namespace does not change while the lock is held.
534  */
535 static pthread_rwlock_t ztest_name_lock;
536 
537 static boolean_t ztest_dump_core = B_TRUE;
538 static boolean_t ztest_exiting;
539 
540 /* Global commit callback list */
541 static ztest_cb_list_t zcl;
542 /* Commit cb delay */
543 static uint64_t zc_min_txg_delay = UINT64_MAX;
544 static int zc_cb_counter = 0;
545 
546 /*
547  * Minimum number of commit callbacks that need to be registered for us to check
548  * whether the minimum txg delay is acceptable.
549  */
550 #define	ZTEST_COMMIT_CB_MIN_REG	100
551 
552 /*
553  * If a number of txgs equal to this threshold have been created after a commit
554  * callback has been registered but not called, then we assume there is an
555  * implementation bug.
556  */
557 #define	ZTEST_COMMIT_CB_THRESH	(TXG_CONCURRENT_STATES + 1000)
558 
559 enum ztest_object {
560 	ZTEST_META_DNODE = 0,
561 	ZTEST_DIROBJ,
562 	ZTEST_OBJECTS
563 };
564 
565 static __attribute__((noreturn)) void usage(boolean_t requested);
566 static int ztest_scrub_impl(spa_t *spa);
567 
568 /*
569  * These libumem hooks provide a reasonable set of defaults for the allocator's
570  * debugging facilities.
571  */
572 const char *
573 _umem_debug_init(void)
574 {
575 	return ("default,verbose"); /* $UMEM_DEBUG setting */
576 }
577 
578 const char *
579 _umem_logging_init(void)
580 {
581 	return ("fail,contents"); /* $UMEM_LOGGING setting */
582 }
583 
584 static void
585 dump_debug_buffer(void)
586 {
587 	ssize_t ret __attribute__((unused));
588 
589 	if (!ztest_opts.zo_dump_dbgmsg)
590 		return;
591 
592 	/*
593 	 * We use write() instead of printf() so that this function
594 	 * is safe to call from a signal handler.
595 	 */
596 	ret = write(STDOUT_FILENO, "\n", 1);
597 	zfs_dbgmsg_print("ztest");
598 }
599 
600 #define	BACKTRACE_SZ	100
601 
602 static void sig_handler(int signo)
603 {
604 	struct sigaction action;
605 #if (__GLIBC__ && !__UCLIBC__) /* backtrace() is a GNU extension */
606 	int nptrs;
607 	void *buffer[BACKTRACE_SZ];
608 
609 	nptrs = backtrace(buffer, BACKTRACE_SZ);
610 	backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
611 #endif
612 	dump_debug_buffer();
613 
614 	/*
615 	 * Restore default action and re-raise signal so SIGSEGV and
616 	 * SIGABRT can trigger a core dump.
617 	 */
618 	action.sa_handler = SIG_DFL;
619 	sigemptyset(&action.sa_mask);
620 	action.sa_flags = 0;
621 	(void) sigaction(signo, &action, NULL);
622 	raise(signo);
623 }
624 
625 #define	FATAL_MSG_SZ	1024
626 
627 static const char *fatal_msg;
628 
629 static __attribute__((format(printf, 2, 3))) __attribute__((noreturn)) void
630 fatal(int do_perror, const char *message, ...)
631 {
632 	va_list args;
633 	int save_errno = errno;
634 	char *buf;
635 
636 	(void) fflush(stdout);
637 	buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
638 	if (buf == NULL)
639 		goto out;
640 
641 	va_start(args, message);
642 	(void) sprintf(buf, "ztest: ");
643 	/* LINTED */
644 	(void) vsprintf(buf + strlen(buf), message, args);
645 	va_end(args);
646 	if (do_perror) {
647 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
648 		    ": %s", strerror(save_errno));
649 	}
650 	(void) fprintf(stderr, "%s\n", buf);
651 	fatal_msg = buf;			/* to ease debugging */
652 
653 out:
654 	if (ztest_dump_core)
655 		abort();
656 	else
657 		dump_debug_buffer();
658 
659 	exit(3);
660 }
661 
662 static int
663 str2shift(const char *buf)
664 {
665 	const char *ends = "BKMGTPEZ";
666 	int i;
667 
668 	if (buf[0] == '\0')
669 		return (0);
670 	for (i = 0; i < strlen(ends); i++) {
671 		if (toupper(buf[0]) == ends[i])
672 			break;
673 	}
674 	if (i == strlen(ends)) {
675 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
676 		    buf);
677 		usage(B_FALSE);
678 	}
679 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
680 		return (10*i);
681 	}
682 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
683 	usage(B_FALSE);
684 }
685 
686 static uint64_t
687 nicenumtoull(const char *buf)
688 {
689 	char *end;
690 	uint64_t val;
691 
692 	val = strtoull(buf, &end, 0);
693 	if (end == buf) {
694 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
695 		usage(B_FALSE);
696 	} else if (end[0] == '.') {
697 		double fval = strtod(buf, &end);
698 		fval *= pow(2, str2shift(end));
699 		/*
700 		 * UINT64_MAX is not exactly representable as a double.
701 		 * The closest representation is UINT64_MAX + 1, so we
702 		 * use a >= comparison instead of > for the bounds check.
703 		 */
704 		if (fval >= (double)UINT64_MAX) {
705 			(void) fprintf(stderr, "ztest: value too large: %s\n",
706 			    buf);
707 			usage(B_FALSE);
708 		}
709 		val = (uint64_t)fval;
710 	} else {
711 		int shift = str2shift(end);
712 		if (shift >= 64 || (val << shift) >> shift != val) {
713 			(void) fprintf(stderr, "ztest: value too large: %s\n",
714 			    buf);
715 			usage(B_FALSE);
716 		}
717 		val <<= shift;
718 	}
719 	return (val);
720 }
721 
722 typedef struct ztest_option {
723 	const char	short_opt;
724 	const char	*long_opt;
725 	const char	*long_opt_param;
726 	const char	*comment;
727 	unsigned int	default_int;
728 	const char	*default_str;
729 } ztest_option_t;
730 
731 /*
732  * The following option_table is used for generating the usage info as well as
733  * the long and short option information for calling getopt_long().
734  */
735 static ztest_option_t option_table[] = {
736 	{ 'v',	"vdevs", "INTEGER", "Number of vdevs", DEFAULT_VDEV_COUNT,
737 	    NULL},
738 	{ 's',	"vdev-size", "INTEGER", "Size of each vdev",
739 	    NO_DEFAULT, DEFAULT_VDEV_SIZE_STR},
740 	{ 'a',	"alignment-shift", "INTEGER",
741 	    "Alignment shift; use 0 for random", DEFAULT_ASHIFT, NULL},
742 	{ 'm',	"mirror-copies", "INTEGER", "Number of mirror copies",
743 	    DEFAULT_MIRRORS, NULL},
744 	{ 'r',	"raid-disks", "INTEGER", "Number of raidz/draid disks",
745 	    DEFAULT_RAID_CHILDREN, NULL},
746 	{ 'R',	"raid-parity", "INTEGER", "Raid parity",
747 	    DEFAULT_RAID_PARITY, NULL},
748 	{ 'K',	"raid-kind", "raidz|draid|random", "Raid kind",
749 	    NO_DEFAULT, "random"},
750 	{ 'D',	"draid-data", "INTEGER", "Number of draid data drives",
751 	    DEFAULT_DRAID_DATA, NULL},
752 	{ 'S',	"draid-spares", "INTEGER", "Number of draid spares",
753 	    DEFAULT_DRAID_SPARES, NULL},
754 	{ 'd',	"datasets", "INTEGER", "Number of datasets",
755 	    DEFAULT_DATASETS_COUNT, NULL},
756 	{ 't',	"threads", "INTEGER", "Number of ztest threads",
757 	    DEFAULT_THREADS, NULL},
758 	{ 'g',	"gang-block-threshold", "INTEGER",
759 	    "Metaslab gang block threshold",
760 	    NO_DEFAULT, DEFAULT_FORCE_GANGING_STR},
761 	{ 'i',	"init-count", "INTEGER", "Number of times to initialize pool",
762 	    DEFAULT_INITS, NULL},
763 	{ 'k',	"kill-percentage", "INTEGER", "Kill percentage",
764 	    NO_DEFAULT, DEFAULT_KILLRATE_STR},
765 	{ 'p',	"pool-name", "STRING", "Pool name",
766 	    NO_DEFAULT, DEFAULT_POOL},
767 	{ 'f',	"vdev-file-directory", "PATH", "File directory for vdev files",
768 	    NO_DEFAULT, DEFAULT_VDEV_DIR},
769 	{ 'M',	"multi-host", NULL,
770 	    "Multi-host; simulate pool imported on remote host",
771 	    NO_DEFAULT, NULL},
772 	{ 'E',	"use-existing-pool", NULL,
773 	    "Use existing pool instead of creating new one", NO_DEFAULT, NULL},
774 	{ 'T',	"run-time", "INTEGER", "Total run time",
775 	    NO_DEFAULT, DEFAULT_RUN_TIME_STR},
776 	{ 'P',	"pass-time", "INTEGER", "Time per pass",
777 	    NO_DEFAULT, DEFAULT_PASS_TIME_STR},
778 	{ 'F',	"freeze-loops", "INTEGER", "Max loops in spa_freeze()",
779 	    DEFAULT_MAX_LOOPS, NULL},
780 	{ 'B',	"alt-ztest", "PATH", "Alternate ztest path",
781 	    NO_DEFAULT, NULL},
782 	{ 'C',	"vdev-class-state", "on|off|random", "vdev class state",
783 	    NO_DEFAULT, "random"},
784 	{ 'o',	"option", "\"OPTION=INTEGER\"",
785 	    "Set global variable to an unsigned 32-bit integer value",
786 	    NO_DEFAULT, NULL},
787 	{ 'G',	"dump-debug-msg", NULL,
788 	    "Dump zfs_dbgmsg buffer before exiting due to an error",
789 	    NO_DEFAULT, NULL},
790 	{ 'V',	"verbose", NULL,
791 	    "Verbose (use multiple times for ever more verbosity)",
792 	    NO_DEFAULT, NULL},
793 	{ 'h',	"help",	NULL, "Show this help",
794 	    NO_DEFAULT, NULL},
795 	{0, 0, 0, 0, 0, 0}
796 };
797 
798 static struct option *long_opts = NULL;
799 static char *short_opts = NULL;
800 
801 static void
802 init_options(void)
803 {
804 	ASSERT3P(long_opts, ==, NULL);
805 	ASSERT3P(short_opts, ==, NULL);
806 
807 	int count = sizeof (option_table) / sizeof (option_table[0]);
808 	long_opts = umem_alloc(sizeof (struct option) * count, UMEM_NOFAIL);
809 
810 	short_opts = umem_alloc(sizeof (char) * 2 * count, UMEM_NOFAIL);
811 	int short_opt_index = 0;
812 
813 	for (int i = 0; i < count; i++) {
814 		long_opts[i].val = option_table[i].short_opt;
815 		long_opts[i].name = option_table[i].long_opt;
816 		long_opts[i].has_arg = option_table[i].long_opt_param != NULL
817 		    ? required_argument : no_argument;
818 		long_opts[i].flag = NULL;
819 		short_opts[short_opt_index++] = option_table[i].short_opt;
820 		if (option_table[i].long_opt_param != NULL) {
821 			short_opts[short_opt_index++] = ':';
822 		}
823 	}
824 }
825 
826 static void
827 fini_options(void)
828 {
829 	int count = sizeof (option_table) / sizeof (option_table[0]);
830 
831 	umem_free(long_opts, sizeof (struct option) * count);
832 	umem_free(short_opts, sizeof (char) * 2 * count);
833 
834 	long_opts = NULL;
835 	short_opts = NULL;
836 }
837 
838 static __attribute__((noreturn)) void
839 usage(boolean_t requested)
840 {
841 	char option[80];
842 	FILE *fp = requested ? stdout : stderr;
843 
844 	(void) fprintf(fp, "Usage: %s [OPTIONS...]\n", DEFAULT_POOL);
845 	for (int i = 0; option_table[i].short_opt != 0; i++) {
846 		if (option_table[i].long_opt_param != NULL) {
847 			(void) sprintf(option, "  -%c --%s=%s",
848 			    option_table[i].short_opt,
849 			    option_table[i].long_opt,
850 			    option_table[i].long_opt_param);
851 		} else {
852 			(void) sprintf(option, "  -%c --%s",
853 			    option_table[i].short_opt,
854 			    option_table[i].long_opt);
855 		}
856 		(void) fprintf(fp, "  %-40s%s", option,
857 		    option_table[i].comment);
858 
859 		if (option_table[i].long_opt_param != NULL) {
860 			if (option_table[i].default_str != NULL) {
861 				(void) fprintf(fp, " (default: %s)",
862 				    option_table[i].default_str);
863 			} else if (option_table[i].default_int != NO_DEFAULT) {
864 				(void) fprintf(fp, " (default: %u)",
865 				    option_table[i].default_int);
866 			}
867 		}
868 		(void) fprintf(fp, "\n");
869 	}
870 	exit(requested ? 0 : 1);
871 }
872 
873 static uint64_t
874 ztest_random(uint64_t range)
875 {
876 	uint64_t r;
877 
878 	ASSERT3S(ztest_fd_rand, >=, 0);
879 
880 	if (range == 0)
881 		return (0);
882 
883 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
884 		fatal(B_TRUE, "short read from /dev/urandom");
885 
886 	return (r % range);
887 }
888 
889 static void
890 ztest_parse_name_value(const char *input, ztest_shared_opts_t *zo)
891 {
892 	char name[32];
893 	char *value;
894 	int state = ZTEST_VDEV_CLASS_RND;
895 
896 	(void) strlcpy(name, input, sizeof (name));
897 
898 	value = strchr(name, '=');
899 	if (value == NULL) {
900 		(void) fprintf(stderr, "missing value in property=value "
901 		    "'-C' argument (%s)\n", input);
902 		usage(B_FALSE);
903 	}
904 	*(value) = '\0';
905 	value++;
906 
907 	if (strcmp(value, "on") == 0) {
908 		state = ZTEST_VDEV_CLASS_ON;
909 	} else if (strcmp(value, "off") == 0) {
910 		state = ZTEST_VDEV_CLASS_OFF;
911 	} else if (strcmp(value, "random") == 0) {
912 		state = ZTEST_VDEV_CLASS_RND;
913 	} else {
914 		(void) fprintf(stderr, "invalid property value '%s'\n", value);
915 		usage(B_FALSE);
916 	}
917 
918 	if (strcmp(name, "special") == 0) {
919 		zo->zo_special_vdevs = state;
920 	} else {
921 		(void) fprintf(stderr, "invalid property name '%s'\n", name);
922 		usage(B_FALSE);
923 	}
924 	if (zo->zo_verbose >= 3)
925 		(void) printf("%s vdev state is '%s'\n", name, value);
926 }
927 
928 static void
929 process_options(int argc, char **argv)
930 {
931 	char *path;
932 	ztest_shared_opts_t *zo = &ztest_opts;
933 
934 	int opt;
935 	uint64_t value;
936 	const char *raid_kind = "random";
937 
938 	memcpy(zo, &ztest_opts_defaults, sizeof (*zo));
939 
940 	init_options();
941 
942 	while ((opt = getopt_long(argc, argv, short_opts, long_opts,
943 	    NULL)) != EOF) {
944 		value = 0;
945 		switch (opt) {
946 		case 'v':
947 		case 's':
948 		case 'a':
949 		case 'm':
950 		case 'r':
951 		case 'R':
952 		case 'D':
953 		case 'S':
954 		case 'd':
955 		case 't':
956 		case 'g':
957 		case 'i':
958 		case 'k':
959 		case 'T':
960 		case 'P':
961 		case 'F':
962 			value = nicenumtoull(optarg);
963 		}
964 		switch (opt) {
965 		case 'v':
966 			zo->zo_vdevs = value;
967 			break;
968 		case 's':
969 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
970 			break;
971 		case 'a':
972 			zo->zo_ashift = value;
973 			break;
974 		case 'm':
975 			zo->zo_mirrors = value;
976 			break;
977 		case 'r':
978 			zo->zo_raid_children = MAX(1, value);
979 			break;
980 		case 'R':
981 			zo->zo_raid_parity = MIN(MAX(value, 1), 3);
982 			break;
983 		case 'K':
984 			raid_kind = optarg;
985 			break;
986 		case 'D':
987 			zo->zo_draid_data = MAX(1, value);
988 			break;
989 		case 'S':
990 			zo->zo_draid_spares = MAX(1, value);
991 			break;
992 		case 'd':
993 			zo->zo_datasets = MAX(1, value);
994 			break;
995 		case 't':
996 			zo->zo_threads = MAX(1, value);
997 			break;
998 		case 'g':
999 			zo->zo_metaslab_force_ganging =
1000 			    MAX(SPA_MINBLOCKSIZE << 1, value);
1001 			break;
1002 		case 'i':
1003 			zo->zo_init = value;
1004 			break;
1005 		case 'k':
1006 			zo->zo_killrate = value;
1007 			break;
1008 		case 'p':
1009 			(void) strlcpy(zo->zo_pool, optarg,
1010 			    sizeof (zo->zo_pool));
1011 			break;
1012 		case 'f':
1013 			path = realpath(optarg, NULL);
1014 			if (path == NULL) {
1015 				(void) fprintf(stderr, "error: %s: %s\n",
1016 				    optarg, strerror(errno));
1017 				usage(B_FALSE);
1018 			} else {
1019 				(void) strlcpy(zo->zo_dir, path,
1020 				    sizeof (zo->zo_dir));
1021 				free(path);
1022 			}
1023 			break;
1024 		case 'M':
1025 			zo->zo_mmp_test = 1;
1026 			break;
1027 		case 'V':
1028 			zo->zo_verbose++;
1029 			break;
1030 		case 'E':
1031 			zo->zo_init = 0;
1032 			break;
1033 		case 'T':
1034 			zo->zo_time = value;
1035 			break;
1036 		case 'P':
1037 			zo->zo_passtime = MAX(1, value);
1038 			break;
1039 		case 'F':
1040 			zo->zo_maxloops = MAX(1, value);
1041 			break;
1042 		case 'B':
1043 			(void) strlcpy(zo->zo_alt_ztest, optarg,
1044 			    sizeof (zo->zo_alt_ztest));
1045 			break;
1046 		case 'C':
1047 			ztest_parse_name_value(optarg, zo);
1048 			break;
1049 		case 'o':
1050 			if (zo->zo_gvars_count >= ZO_GVARS_MAX_COUNT) {
1051 				(void) fprintf(stderr,
1052 				    "max global var count (%zu) exceeded\n",
1053 				    ZO_GVARS_MAX_COUNT);
1054 				usage(B_FALSE);
1055 			}
1056 			char *v = zo->zo_gvars[zo->zo_gvars_count];
1057 			if (strlcpy(v, optarg, ZO_GVARS_MAX_ARGLEN) >=
1058 			    ZO_GVARS_MAX_ARGLEN) {
1059 				(void) fprintf(stderr,
1060 				    "global var option '%s' is too long\n",
1061 				    optarg);
1062 				usage(B_FALSE);
1063 			}
1064 			zo->zo_gvars_count++;
1065 			break;
1066 		case 'G':
1067 			zo->zo_dump_dbgmsg = 1;
1068 			break;
1069 		case 'h':
1070 			usage(B_TRUE);
1071 			break;
1072 		case '?':
1073 		default:
1074 			usage(B_FALSE);
1075 			break;
1076 		}
1077 	}
1078 
1079 	fini_options();
1080 
1081 	/* When raid choice is 'random' add a draid pool 50% of the time */
1082 	if (strcmp(raid_kind, "random") == 0) {
1083 		raid_kind = (ztest_random(2) == 0) ? "draid" : "raidz";
1084 
1085 		if (ztest_opts.zo_verbose >= 3)
1086 			(void) printf("choosing RAID type '%s'\n", raid_kind);
1087 	}
1088 
1089 	if (strcmp(raid_kind, "draid") == 0) {
1090 		uint64_t min_devsize;
1091 
1092 		/* With fewer disk use 256M, otherwise 128M is OK */
1093 		min_devsize = (ztest_opts.zo_raid_children < 16) ?
1094 		    (256ULL << 20) : (128ULL << 20);
1095 
1096 		/* No top-level mirrors with dRAID for now */
1097 		zo->zo_mirrors = 0;
1098 
1099 		/* Use more appropriate defaults for dRAID */
1100 		if (zo->zo_vdevs == ztest_opts_defaults.zo_vdevs)
1101 			zo->zo_vdevs = 1;
1102 		if (zo->zo_raid_children ==
1103 		    ztest_opts_defaults.zo_raid_children)
1104 			zo->zo_raid_children = 16;
1105 		if (zo->zo_ashift < 12)
1106 			zo->zo_ashift = 12;
1107 		if (zo->zo_vdev_size < min_devsize)
1108 			zo->zo_vdev_size = min_devsize;
1109 
1110 		if (zo->zo_draid_data + zo->zo_raid_parity >
1111 		    zo->zo_raid_children - zo->zo_draid_spares) {
1112 			(void) fprintf(stderr, "error: too few draid "
1113 			    "children (%d) for stripe width (%d)\n",
1114 			    zo->zo_raid_children,
1115 			    zo->zo_draid_data + zo->zo_raid_parity);
1116 			usage(B_FALSE);
1117 		}
1118 
1119 		(void) strlcpy(zo->zo_raid_type, VDEV_TYPE_DRAID,
1120 		    sizeof (zo->zo_raid_type));
1121 
1122 	} else /* using raidz */ {
1123 		ASSERT0(strcmp(raid_kind, "raidz"));
1124 
1125 		zo->zo_raid_parity = MIN(zo->zo_raid_parity,
1126 		    zo->zo_raid_children - 1);
1127 	}
1128 
1129 	zo->zo_vdevtime =
1130 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
1131 	    UINT64_MAX >> 2);
1132 
1133 	if (*zo->zo_alt_ztest) {
1134 		const char *invalid_what = "ztest";
1135 		char *val = zo->zo_alt_ztest;
1136 		if (0 != access(val, X_OK) ||
1137 		    (strrchr(val, '/') == NULL && (errno == EINVAL)))
1138 			goto invalid;
1139 
1140 		int dirlen = strrchr(val, '/') - val;
1141 		strlcpy(zo->zo_alt_libpath, val,
1142 		    MIN(sizeof (zo->zo_alt_libpath), dirlen + 1));
1143 		invalid_what = "library path", val = zo->zo_alt_libpath;
1144 		if (strrchr(val, '/') == NULL && (errno == EINVAL))
1145 			goto invalid;
1146 		*strrchr(val, '/') = '\0';
1147 		strlcat(val, "/lib", sizeof (zo->zo_alt_libpath));
1148 
1149 		if (0 != access(zo->zo_alt_libpath, X_OK))
1150 			goto invalid;
1151 		return;
1152 
1153 invalid:
1154 		ztest_dump_core = B_FALSE;
1155 		fatal(B_TRUE, "invalid alternate %s %s", invalid_what, val);
1156 	}
1157 }
1158 
1159 static void
1160 ztest_kill(ztest_shared_t *zs)
1161 {
1162 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
1163 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
1164 
1165 	/*
1166 	 * Before we kill ourselves, make sure that the config is updated.
1167 	 * See comment above spa_write_cachefile().
1168 	 */
1169 	mutex_enter(&spa_namespace_lock);
1170 	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE, B_FALSE);
1171 	mutex_exit(&spa_namespace_lock);
1172 
1173 	(void) raise(SIGKILL);
1174 }
1175 
1176 static void
1177 ztest_record_enospc(const char *s)
1178 {
1179 	(void) s;
1180 	ztest_shared->zs_enospc_count++;
1181 }
1182 
1183 static uint64_t
1184 ztest_get_ashift(void)
1185 {
1186 	if (ztest_opts.zo_ashift == 0)
1187 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
1188 	return (ztest_opts.zo_ashift);
1189 }
1190 
1191 static boolean_t
1192 ztest_is_draid_spare(const char *name)
1193 {
1194 	uint64_t spare_id = 0, parity = 0, vdev_id = 0;
1195 
1196 	if (sscanf(name, VDEV_TYPE_DRAID "%"PRIu64"-%"PRIu64"-%"PRIu64"",
1197 	    &parity, &vdev_id, &spare_id) == 3) {
1198 		return (B_TRUE);
1199 	}
1200 
1201 	return (B_FALSE);
1202 }
1203 
1204 static nvlist_t *
1205 make_vdev_file(const char *path, const char *aux, const char *pool,
1206     size_t size, uint64_t ashift)
1207 {
1208 	char *pathbuf = NULL;
1209 	uint64_t vdev;
1210 	nvlist_t *file;
1211 	boolean_t draid_spare = B_FALSE;
1212 
1213 
1214 	if (ashift == 0)
1215 		ashift = ztest_get_ashift();
1216 
1217 	if (path == NULL) {
1218 		pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1219 		path = pathbuf;
1220 
1221 		if (aux != NULL) {
1222 			vdev = ztest_shared->zs_vdev_aux;
1223 			(void) snprintf(pathbuf, MAXPATHLEN,
1224 			    ztest_aux_template, ztest_opts.zo_dir,
1225 			    pool == NULL ? ztest_opts.zo_pool : pool,
1226 			    aux, vdev);
1227 		} else {
1228 			vdev = ztest_shared->zs_vdev_next_leaf++;
1229 			(void) snprintf(pathbuf, MAXPATHLEN,
1230 			    ztest_dev_template, ztest_opts.zo_dir,
1231 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
1232 		}
1233 	} else {
1234 		draid_spare = ztest_is_draid_spare(path);
1235 	}
1236 
1237 	if (size != 0 && !draid_spare) {
1238 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
1239 		if (fd == -1)
1240 			fatal(B_TRUE, "can't open %s", path);
1241 		if (ftruncate(fd, size) != 0)
1242 			fatal(B_TRUE, "can't ftruncate %s", path);
1243 		(void) close(fd);
1244 	}
1245 
1246 	file = fnvlist_alloc();
1247 	fnvlist_add_string(file, ZPOOL_CONFIG_TYPE,
1248 	    draid_spare ? VDEV_TYPE_DRAID_SPARE : VDEV_TYPE_FILE);
1249 	fnvlist_add_string(file, ZPOOL_CONFIG_PATH, path);
1250 	fnvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift);
1251 	umem_free(pathbuf, MAXPATHLEN);
1252 
1253 	return (file);
1254 }
1255 
1256 static nvlist_t *
1257 make_vdev_raid(const char *path, const char *aux, const char *pool, size_t size,
1258     uint64_t ashift, int r)
1259 {
1260 	nvlist_t *raid, **child;
1261 	int c;
1262 
1263 	if (r < 2)
1264 		return (make_vdev_file(path, aux, pool, size, ashift));
1265 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
1266 
1267 	for (c = 0; c < r; c++)
1268 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
1269 
1270 	raid = fnvlist_alloc();
1271 	fnvlist_add_string(raid, ZPOOL_CONFIG_TYPE,
1272 	    ztest_opts.zo_raid_type);
1273 	fnvlist_add_uint64(raid, ZPOOL_CONFIG_NPARITY,
1274 	    ztest_opts.zo_raid_parity);
1275 	fnvlist_add_nvlist_array(raid, ZPOOL_CONFIG_CHILDREN,
1276 	    (const nvlist_t **)child, r);
1277 
1278 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0) {
1279 		uint64_t ndata = ztest_opts.zo_draid_data;
1280 		uint64_t nparity = ztest_opts.zo_raid_parity;
1281 		uint64_t nspares = ztest_opts.zo_draid_spares;
1282 		uint64_t children = ztest_opts.zo_raid_children;
1283 		uint64_t ngroups = 1;
1284 
1285 		/*
1286 		 * Calculate the minimum number of groups required to fill a
1287 		 * slice. This is the LCM of the stripe width (data + parity)
1288 		 * and the number of data drives (children - spares).
1289 		 */
1290 		while (ngroups * (ndata + nparity) % (children - nspares) != 0)
1291 			ngroups++;
1292 
1293 		/* Store the basic dRAID configuration. */
1294 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NDATA, ndata);
1295 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NSPARES, nspares);
1296 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NGROUPS, ngroups);
1297 	}
1298 
1299 	for (c = 0; c < r; c++)
1300 		fnvlist_free(child[c]);
1301 
1302 	umem_free(child, r * sizeof (nvlist_t *));
1303 
1304 	return (raid);
1305 }
1306 
1307 static nvlist_t *
1308 make_vdev_mirror(const char *path, const char *aux, const char *pool,
1309     size_t size, uint64_t ashift, int r, int m)
1310 {
1311 	nvlist_t *mirror, **child;
1312 	int c;
1313 
1314 	if (m < 1)
1315 		return (make_vdev_raid(path, aux, pool, size, ashift, r));
1316 
1317 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
1318 
1319 	for (c = 0; c < m; c++)
1320 		child[c] = make_vdev_raid(path, aux, pool, size, ashift, r);
1321 
1322 	mirror = fnvlist_alloc();
1323 	fnvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, VDEV_TYPE_MIRROR);
1324 	fnvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
1325 	    (const nvlist_t **)child, m);
1326 
1327 	for (c = 0; c < m; c++)
1328 		fnvlist_free(child[c]);
1329 
1330 	umem_free(child, m * sizeof (nvlist_t *));
1331 
1332 	return (mirror);
1333 }
1334 
1335 static nvlist_t *
1336 make_vdev_root(const char *path, const char *aux, const char *pool, size_t size,
1337     uint64_t ashift, const char *class, int r, int m, int t)
1338 {
1339 	nvlist_t *root, **child;
1340 	int c;
1341 	boolean_t log;
1342 
1343 	ASSERT3S(t, >, 0);
1344 
1345 	log = (class != NULL && strcmp(class, "log") == 0);
1346 
1347 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1348 
1349 	for (c = 0; c < t; c++) {
1350 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1351 		    r, m);
1352 		fnvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, log);
1353 
1354 		if (class != NULL && class[0] != '\0') {
1355 			ASSERT(m > 1 || log);   /* expecting a mirror */
1356 			fnvlist_add_string(child[c],
1357 			    ZPOOL_CONFIG_ALLOCATION_BIAS, class);
1358 		}
1359 	}
1360 
1361 	root = fnvlist_alloc();
1362 	fnvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
1363 	fnvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1364 	    (const nvlist_t **)child, t);
1365 
1366 	for (c = 0; c < t; c++)
1367 		fnvlist_free(child[c]);
1368 
1369 	umem_free(child, t * sizeof (nvlist_t *));
1370 
1371 	return (root);
1372 }
1373 
1374 /*
1375  * Find a random spa version. Returns back a random spa version in the
1376  * range [initial_version, SPA_VERSION_FEATURES].
1377  */
1378 static uint64_t
1379 ztest_random_spa_version(uint64_t initial_version)
1380 {
1381 	uint64_t version = initial_version;
1382 
1383 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
1384 		version = version +
1385 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1386 	}
1387 
1388 	if (version > SPA_VERSION_BEFORE_FEATURES)
1389 		version = SPA_VERSION_FEATURES;
1390 
1391 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1392 	return (version);
1393 }
1394 
1395 static int
1396 ztest_random_blocksize(void)
1397 {
1398 	ASSERT3U(ztest_spa->spa_max_ashift, !=, 0);
1399 
1400 	/*
1401 	 * Choose a block size >= the ashift.
1402 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1403 	 */
1404 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1405 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1406 		maxbs = 20;
1407 	uint64_t block_shift =
1408 	    ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1409 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1410 }
1411 
1412 static int
1413 ztest_random_dnodesize(void)
1414 {
1415 	int slots;
1416 	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1417 
1418 	if (max_slots == DNODE_MIN_SLOTS)
1419 		return (DNODE_MIN_SIZE);
1420 
1421 	/*
1422 	 * Weight the random distribution more heavily toward smaller
1423 	 * dnode sizes since that is more likely to reflect real-world
1424 	 * usage.
1425 	 */
1426 	ASSERT3U(max_slots, >, 4);
1427 	switch (ztest_random(10)) {
1428 	case 0:
1429 		slots = 5 + ztest_random(max_slots - 4);
1430 		break;
1431 	case 1 ... 4:
1432 		slots = 2 + ztest_random(3);
1433 		break;
1434 	default:
1435 		slots = 1;
1436 		break;
1437 	}
1438 
1439 	return (slots << DNODE_SHIFT);
1440 }
1441 
1442 static int
1443 ztest_random_ibshift(void)
1444 {
1445 	return (DN_MIN_INDBLKSHIFT +
1446 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1447 }
1448 
1449 static uint64_t
1450 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1451 {
1452 	uint64_t top;
1453 	vdev_t *rvd = spa->spa_root_vdev;
1454 	vdev_t *tvd;
1455 
1456 	ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
1457 
1458 	do {
1459 		top = ztest_random(rvd->vdev_children);
1460 		tvd = rvd->vdev_child[top];
1461 	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1462 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1463 
1464 	return (top);
1465 }
1466 
1467 static uint64_t
1468 ztest_random_dsl_prop(zfs_prop_t prop)
1469 {
1470 	uint64_t value;
1471 
1472 	do {
1473 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1474 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1475 
1476 	return (value);
1477 }
1478 
1479 static int
1480 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1481     boolean_t inherit)
1482 {
1483 	const char *propname = zfs_prop_to_name(prop);
1484 	const char *valname;
1485 	char *setpoint;
1486 	uint64_t curval;
1487 	int error;
1488 
1489 	error = dsl_prop_set_int(osname, propname,
1490 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1491 
1492 	if (error == ENOSPC) {
1493 		ztest_record_enospc(FTAG);
1494 		return (error);
1495 	}
1496 	ASSERT0(error);
1497 
1498 	setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1499 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1500 
1501 	if (ztest_opts.zo_verbose >= 6) {
1502 		int err;
1503 
1504 		err = zfs_prop_index_to_string(prop, curval, &valname);
1505 		if (err)
1506 			(void) printf("%s %s = %llu at '%s'\n", osname,
1507 			    propname, (unsigned long long)curval, setpoint);
1508 		else
1509 			(void) printf("%s %s = %s at '%s'\n",
1510 			    osname, propname, valname, setpoint);
1511 	}
1512 	umem_free(setpoint, MAXPATHLEN);
1513 
1514 	return (error);
1515 }
1516 
1517 static int
1518 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1519 {
1520 	spa_t *spa = ztest_spa;
1521 	nvlist_t *props = NULL;
1522 	int error;
1523 
1524 	props = fnvlist_alloc();
1525 	fnvlist_add_uint64(props, zpool_prop_to_name(prop), value);
1526 
1527 	error = spa_prop_set(spa, props);
1528 
1529 	fnvlist_free(props);
1530 
1531 	if (error == ENOSPC) {
1532 		ztest_record_enospc(FTAG);
1533 		return (error);
1534 	}
1535 	ASSERT0(error);
1536 
1537 	return (error);
1538 }
1539 
1540 static int
1541 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1542     boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp)
1543 {
1544 	int err;
1545 	char *cp = NULL;
1546 	char ddname[ZFS_MAX_DATASET_NAME_LEN];
1547 
1548 	strlcpy(ddname, name, sizeof (ddname));
1549 	cp = strchr(ddname, '@');
1550 	if (cp != NULL)
1551 		*cp = '\0';
1552 
1553 	err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1554 	while (decrypt && err == EACCES) {
1555 		dsl_crypto_params_t *dcp;
1556 		nvlist_t *crypto_args = fnvlist_alloc();
1557 
1558 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
1559 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1560 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1561 		    crypto_args, &dcp));
1562 		err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1563 		/*
1564 		 * Note: if there was an error loading, the wkey was not
1565 		 * consumed, and needs to be freed.
1566 		 */
1567 		dsl_crypto_params_free(dcp, (err != 0));
1568 		fnvlist_free(crypto_args);
1569 
1570 		if (err == EINVAL) {
1571 			/*
1572 			 * We couldn't load a key for this dataset so try
1573 			 * the parent. This loop will eventually hit the
1574 			 * encryption root since ztest only makes clones
1575 			 * as children of their origin datasets.
1576 			 */
1577 			cp = strrchr(ddname, '/');
1578 			if (cp == NULL)
1579 				return (err);
1580 
1581 			*cp = '\0';
1582 			err = EACCES;
1583 			continue;
1584 		} else if (err != 0) {
1585 			break;
1586 		}
1587 
1588 		err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1589 		break;
1590 	}
1591 
1592 	return (err);
1593 }
1594 
1595 static void
1596 ztest_rll_init(rll_t *rll)
1597 {
1598 	rll->rll_writer = NULL;
1599 	rll->rll_readers = 0;
1600 	mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1601 	cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1602 }
1603 
1604 static void
1605 ztest_rll_destroy(rll_t *rll)
1606 {
1607 	ASSERT3P(rll->rll_writer, ==, NULL);
1608 	ASSERT0(rll->rll_readers);
1609 	mutex_destroy(&rll->rll_lock);
1610 	cv_destroy(&rll->rll_cv);
1611 }
1612 
1613 static void
1614 ztest_rll_lock(rll_t *rll, rl_type_t type)
1615 {
1616 	mutex_enter(&rll->rll_lock);
1617 
1618 	if (type == RL_READER) {
1619 		while (rll->rll_writer != NULL)
1620 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1621 		rll->rll_readers++;
1622 	} else {
1623 		while (rll->rll_writer != NULL || rll->rll_readers)
1624 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1625 		rll->rll_writer = curthread;
1626 	}
1627 
1628 	mutex_exit(&rll->rll_lock);
1629 }
1630 
1631 static void
1632 ztest_rll_unlock(rll_t *rll)
1633 {
1634 	mutex_enter(&rll->rll_lock);
1635 
1636 	if (rll->rll_writer) {
1637 		ASSERT0(rll->rll_readers);
1638 		rll->rll_writer = NULL;
1639 	} else {
1640 		ASSERT3S(rll->rll_readers, >, 0);
1641 		ASSERT3P(rll->rll_writer, ==, NULL);
1642 		rll->rll_readers--;
1643 	}
1644 
1645 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1646 		cv_broadcast(&rll->rll_cv);
1647 
1648 	mutex_exit(&rll->rll_lock);
1649 }
1650 
1651 static void
1652 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1653 {
1654 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1655 
1656 	ztest_rll_lock(rll, type);
1657 }
1658 
1659 static void
1660 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1661 {
1662 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1663 
1664 	ztest_rll_unlock(rll);
1665 }
1666 
1667 static rl_t *
1668 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1669     uint64_t size, rl_type_t type)
1670 {
1671 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1672 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1673 	rl_t *rl;
1674 
1675 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1676 	rl->rl_object = object;
1677 	rl->rl_offset = offset;
1678 	rl->rl_size = size;
1679 	rl->rl_lock = rll;
1680 
1681 	ztest_rll_lock(rll, type);
1682 
1683 	return (rl);
1684 }
1685 
1686 static void
1687 ztest_range_unlock(rl_t *rl)
1688 {
1689 	rll_t *rll = rl->rl_lock;
1690 
1691 	ztest_rll_unlock(rll);
1692 
1693 	umem_free(rl, sizeof (*rl));
1694 }
1695 
1696 static void
1697 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1698 {
1699 	zd->zd_os = os;
1700 	zd->zd_zilog = dmu_objset_zil(os);
1701 	zd->zd_shared = szd;
1702 	dmu_objset_name(os, zd->zd_name);
1703 	int l;
1704 
1705 	if (zd->zd_shared != NULL)
1706 		zd->zd_shared->zd_seq = 0;
1707 
1708 	VERIFY0(pthread_rwlock_init(&zd->zd_zilog_lock, NULL));
1709 	mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1710 
1711 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1712 		ztest_rll_init(&zd->zd_object_lock[l]);
1713 
1714 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1715 		ztest_rll_init(&zd->zd_range_lock[l]);
1716 }
1717 
1718 static void
1719 ztest_zd_fini(ztest_ds_t *zd)
1720 {
1721 	int l;
1722 
1723 	mutex_destroy(&zd->zd_dirobj_lock);
1724 	(void) pthread_rwlock_destroy(&zd->zd_zilog_lock);
1725 
1726 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1727 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1728 
1729 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1730 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1731 }
1732 
1733 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1734 
1735 static uint64_t
1736 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1737 {
1738 	uint64_t txg;
1739 	int error;
1740 
1741 	/*
1742 	 * Attempt to assign tx to some transaction group.
1743 	 */
1744 	error = dmu_tx_assign(tx, txg_how);
1745 	if (error) {
1746 		if (error == ERESTART) {
1747 			ASSERT3U(txg_how, ==, TXG_NOWAIT);
1748 			dmu_tx_wait(tx);
1749 		} else {
1750 			ASSERT3U(error, ==, ENOSPC);
1751 			ztest_record_enospc(tag);
1752 		}
1753 		dmu_tx_abort(tx);
1754 		return (0);
1755 	}
1756 	txg = dmu_tx_get_txg(tx);
1757 	ASSERT3U(txg, !=, 0);
1758 	return (txg);
1759 }
1760 
1761 static void
1762 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1763     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1764     uint64_t crtxg)
1765 {
1766 	bt->bt_magic = BT_MAGIC;
1767 	bt->bt_objset = dmu_objset_id(os);
1768 	bt->bt_object = object;
1769 	bt->bt_dnodesize = dnodesize;
1770 	bt->bt_offset = offset;
1771 	bt->bt_gen = gen;
1772 	bt->bt_txg = txg;
1773 	bt->bt_crtxg = crtxg;
1774 }
1775 
1776 static void
1777 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1778     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1779     uint64_t crtxg)
1780 {
1781 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1782 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1783 	ASSERT3U(bt->bt_object, ==, object);
1784 	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1785 	ASSERT3U(bt->bt_offset, ==, offset);
1786 	ASSERT3U(bt->bt_gen, <=, gen);
1787 	ASSERT3U(bt->bt_txg, <=, txg);
1788 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1789 }
1790 
1791 static ztest_block_tag_t *
1792 ztest_bt_bonus(dmu_buf_t *db)
1793 {
1794 	dmu_object_info_t doi;
1795 	ztest_block_tag_t *bt;
1796 
1797 	dmu_object_info_from_db(db, &doi);
1798 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1799 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1800 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1801 
1802 	return (bt);
1803 }
1804 
1805 /*
1806  * Generate a token to fill up unused bonus buffer space.  Try to make
1807  * it unique to the object, generation, and offset to verify that data
1808  * is not getting overwritten by data from other dnodes.
1809  */
1810 #define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1811 	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1812 
1813 /*
1814  * Fill up the unused bonus buffer region before the block tag with a
1815  * verifiable pattern. Filling the whole bonus area with non-zero data
1816  * helps ensure that all dnode traversal code properly skips the
1817  * interior regions of large dnodes.
1818  */
1819 static void
1820 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1821     objset_t *os, uint64_t gen)
1822 {
1823 	uint64_t *bonusp;
1824 
1825 	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1826 
1827 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1828 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1829 		    gen, bonusp - (uint64_t *)db->db_data);
1830 		*bonusp = token;
1831 	}
1832 }
1833 
1834 /*
1835  * Verify that the unused area of a bonus buffer is filled with the
1836  * expected tokens.
1837  */
1838 static void
1839 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1840     objset_t *os, uint64_t gen)
1841 {
1842 	uint64_t *bonusp;
1843 
1844 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1845 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1846 		    gen, bonusp - (uint64_t *)db->db_data);
1847 		VERIFY3U(*bonusp, ==, token);
1848 	}
1849 }
1850 
1851 /*
1852  * ZIL logging ops
1853  */
1854 
1855 #define	lrz_type	lr_mode
1856 #define	lrz_blocksize	lr_uid
1857 #define	lrz_ibshift	lr_gid
1858 #define	lrz_bonustype	lr_rdev
1859 #define	lrz_dnodesize	lr_crtime[1]
1860 
1861 static void
1862 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1863 {
1864 	char *name = (void *)(lr + 1);		/* name follows lr */
1865 	size_t namesize = strlen(name) + 1;
1866 	itx_t *itx;
1867 
1868 	if (zil_replaying(zd->zd_zilog, tx))
1869 		return;
1870 
1871 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1872 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1873 	    sizeof (*lr) + namesize - sizeof (lr_t));
1874 
1875 	zil_itx_assign(zd->zd_zilog, itx, tx);
1876 }
1877 
1878 static void
1879 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1880 {
1881 	char *name = (void *)(lr + 1);		/* name follows lr */
1882 	size_t namesize = strlen(name) + 1;
1883 	itx_t *itx;
1884 
1885 	if (zil_replaying(zd->zd_zilog, tx))
1886 		return;
1887 
1888 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1889 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1890 	    sizeof (*lr) + namesize - sizeof (lr_t));
1891 
1892 	itx->itx_oid = object;
1893 	zil_itx_assign(zd->zd_zilog, itx, tx);
1894 }
1895 
1896 static void
1897 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1898 {
1899 	itx_t *itx;
1900 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1901 
1902 	if (zil_replaying(zd->zd_zilog, tx))
1903 		return;
1904 
1905 	if (lr->lr_length > zil_max_log_data(zd->zd_zilog, sizeof (lr_write_t)))
1906 		write_state = WR_INDIRECT;
1907 
1908 	itx = zil_itx_create(TX_WRITE,
1909 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1910 
1911 	if (write_state == WR_COPIED &&
1912 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1913 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1914 		zil_itx_destroy(itx);
1915 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1916 		write_state = WR_NEED_COPY;
1917 	}
1918 	itx->itx_private = zd;
1919 	itx->itx_wr_state = write_state;
1920 	itx->itx_sync = (ztest_random(8) == 0);
1921 
1922 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1923 	    sizeof (*lr) - sizeof (lr_t));
1924 
1925 	zil_itx_assign(zd->zd_zilog, itx, tx);
1926 }
1927 
1928 static void
1929 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1930 {
1931 	itx_t *itx;
1932 
1933 	if (zil_replaying(zd->zd_zilog, tx))
1934 		return;
1935 
1936 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1937 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1938 	    sizeof (*lr) - sizeof (lr_t));
1939 
1940 	itx->itx_sync = B_FALSE;
1941 	zil_itx_assign(zd->zd_zilog, itx, tx);
1942 }
1943 
1944 static void
1945 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1946 {
1947 	itx_t *itx;
1948 
1949 	if (zil_replaying(zd->zd_zilog, tx))
1950 		return;
1951 
1952 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1953 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1954 	    sizeof (*lr) - sizeof (lr_t));
1955 
1956 	itx->itx_sync = B_FALSE;
1957 	zil_itx_assign(zd->zd_zilog, itx, tx);
1958 }
1959 
1960 /*
1961  * ZIL replay ops
1962  */
1963 static int
1964 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1965 {
1966 	ztest_ds_t *zd = arg1;
1967 	lr_create_t *lr = arg2;
1968 	char *name = (void *)(lr + 1);		/* name follows lr */
1969 	objset_t *os = zd->zd_os;
1970 	ztest_block_tag_t *bbt;
1971 	dmu_buf_t *db;
1972 	dmu_tx_t *tx;
1973 	uint64_t txg;
1974 	int error = 0;
1975 	int bonuslen;
1976 
1977 	if (byteswap)
1978 		byteswap_uint64_array(lr, sizeof (*lr));
1979 
1980 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
1981 	ASSERT3S(name[0], !=, '\0');
1982 
1983 	tx = dmu_tx_create(os);
1984 
1985 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1986 
1987 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1988 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1989 	} else {
1990 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1991 	}
1992 
1993 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1994 	if (txg == 0)
1995 		return (ENOSPC);
1996 
1997 	ASSERT3U(dmu_objset_zil(os)->zl_replay, ==, !!lr->lr_foid);
1998 	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1999 
2000 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
2001 		if (lr->lr_foid == 0) {
2002 			lr->lr_foid = zap_create_dnsize(os,
2003 			    lr->lrz_type, lr->lrz_bonustype,
2004 			    bonuslen, lr->lrz_dnodesize, tx);
2005 		} else {
2006 			error = zap_create_claim_dnsize(os, lr->lr_foid,
2007 			    lr->lrz_type, lr->lrz_bonustype,
2008 			    bonuslen, lr->lrz_dnodesize, tx);
2009 		}
2010 	} else {
2011 		if (lr->lr_foid == 0) {
2012 			lr->lr_foid = dmu_object_alloc_dnsize(os,
2013 			    lr->lrz_type, 0, lr->lrz_bonustype,
2014 			    bonuslen, lr->lrz_dnodesize, tx);
2015 		} else {
2016 			error = dmu_object_claim_dnsize(os, lr->lr_foid,
2017 			    lr->lrz_type, 0, lr->lrz_bonustype,
2018 			    bonuslen, lr->lrz_dnodesize, tx);
2019 		}
2020 	}
2021 
2022 	if (error) {
2023 		ASSERT3U(error, ==, EEXIST);
2024 		ASSERT(zd->zd_zilog->zl_replay);
2025 		dmu_tx_commit(tx);
2026 		return (error);
2027 	}
2028 
2029 	ASSERT3U(lr->lr_foid, !=, 0);
2030 
2031 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
2032 		VERIFY0(dmu_object_set_blocksize(os, lr->lr_foid,
2033 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
2034 
2035 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2036 	bbt = ztest_bt_bonus(db);
2037 	dmu_buf_will_dirty(db, tx);
2038 	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
2039 	    lr->lr_gen, txg, txg);
2040 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
2041 	dmu_buf_rele(db, FTAG);
2042 
2043 	VERIFY0(zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
2044 	    &lr->lr_foid, tx));
2045 
2046 	(void) ztest_log_create(zd, tx, lr);
2047 
2048 	dmu_tx_commit(tx);
2049 
2050 	return (0);
2051 }
2052 
2053 static int
2054 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
2055 {
2056 	ztest_ds_t *zd = arg1;
2057 	lr_remove_t *lr = arg2;
2058 	char *name = (void *)(lr + 1);		/* name follows lr */
2059 	objset_t *os = zd->zd_os;
2060 	dmu_object_info_t doi;
2061 	dmu_tx_t *tx;
2062 	uint64_t object, txg;
2063 
2064 	if (byteswap)
2065 		byteswap_uint64_array(lr, sizeof (*lr));
2066 
2067 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
2068 	ASSERT3S(name[0], !=, '\0');
2069 
2070 	VERIFY0(
2071 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
2072 	ASSERT3U(object, !=, 0);
2073 
2074 	ztest_object_lock(zd, object, RL_WRITER);
2075 
2076 	VERIFY0(dmu_object_info(os, object, &doi));
2077 
2078 	tx = dmu_tx_create(os);
2079 
2080 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
2081 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2082 
2083 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2084 	if (txg == 0) {
2085 		ztest_object_unlock(zd, object);
2086 		return (ENOSPC);
2087 	}
2088 
2089 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
2090 		VERIFY0(zap_destroy(os, object, tx));
2091 	} else {
2092 		VERIFY0(dmu_object_free(os, object, tx));
2093 	}
2094 
2095 	VERIFY0(zap_remove(os, lr->lr_doid, name, tx));
2096 
2097 	(void) ztest_log_remove(zd, tx, lr, object);
2098 
2099 	dmu_tx_commit(tx);
2100 
2101 	ztest_object_unlock(zd, object);
2102 
2103 	return (0);
2104 }
2105 
2106 static int
2107 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
2108 {
2109 	ztest_ds_t *zd = arg1;
2110 	lr_write_t *lr = arg2;
2111 	objset_t *os = zd->zd_os;
2112 	void *data = lr + 1;			/* data follows lr */
2113 	uint64_t offset, length;
2114 	ztest_block_tag_t *bt = data;
2115 	ztest_block_tag_t *bbt;
2116 	uint64_t gen, txg, lrtxg, crtxg;
2117 	dmu_object_info_t doi;
2118 	dmu_tx_t *tx;
2119 	dmu_buf_t *db;
2120 	arc_buf_t *abuf = NULL;
2121 	rl_t *rl;
2122 
2123 	if (byteswap)
2124 		byteswap_uint64_array(lr, sizeof (*lr));
2125 
2126 	offset = lr->lr_offset;
2127 	length = lr->lr_length;
2128 
2129 	/* If it's a dmu_sync() block, write the whole block */
2130 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
2131 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
2132 		if (length < blocksize) {
2133 			offset -= offset % blocksize;
2134 			length = blocksize;
2135 		}
2136 	}
2137 
2138 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
2139 		byteswap_uint64_array(bt, sizeof (*bt));
2140 
2141 	if (bt->bt_magic != BT_MAGIC)
2142 		bt = NULL;
2143 
2144 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
2145 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
2146 
2147 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2148 
2149 	dmu_object_info_from_db(db, &doi);
2150 
2151 	bbt = ztest_bt_bonus(db);
2152 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2153 	gen = bbt->bt_gen;
2154 	crtxg = bbt->bt_crtxg;
2155 	lrtxg = lr->lr_common.lrc_txg;
2156 
2157 	tx = dmu_tx_create(os);
2158 
2159 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
2160 
2161 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
2162 	    P2PHASE(offset, length) == 0)
2163 		abuf = dmu_request_arcbuf(db, length);
2164 
2165 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2166 	if (txg == 0) {
2167 		if (abuf != NULL)
2168 			dmu_return_arcbuf(abuf);
2169 		dmu_buf_rele(db, FTAG);
2170 		ztest_range_unlock(rl);
2171 		ztest_object_unlock(zd, lr->lr_foid);
2172 		return (ENOSPC);
2173 	}
2174 
2175 	if (bt != NULL) {
2176 		/*
2177 		 * Usually, verify the old data before writing new data --
2178 		 * but not always, because we also want to verify correct
2179 		 * behavior when the data was not recently read into cache.
2180 		 */
2181 		ASSERT(doi.doi_data_block_size);
2182 		ASSERT0(offset % doi.doi_data_block_size);
2183 		if (ztest_random(4) != 0) {
2184 			int prefetch = ztest_random(2) ?
2185 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
2186 			ztest_block_tag_t rbt;
2187 
2188 			VERIFY(dmu_read(os, lr->lr_foid, offset,
2189 			    sizeof (rbt), &rbt, prefetch) == 0);
2190 			if (rbt.bt_magic == BT_MAGIC) {
2191 				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
2192 				    offset, gen, txg, crtxg);
2193 			}
2194 		}
2195 
2196 		/*
2197 		 * Writes can appear to be newer than the bonus buffer because
2198 		 * the ztest_get_data() callback does a dmu_read() of the
2199 		 * open-context data, which may be different than the data
2200 		 * as it was when the write was generated.
2201 		 */
2202 		if (zd->zd_zilog->zl_replay) {
2203 			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
2204 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
2205 			    bt->bt_crtxg);
2206 		}
2207 
2208 		/*
2209 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
2210 		 * so that all of the usual ASSERTs will work.
2211 		 */
2212 		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
2213 		    crtxg);
2214 	}
2215 
2216 	if (abuf == NULL) {
2217 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
2218 	} else {
2219 		memcpy(abuf->b_data, data, length);
2220 		VERIFY0(dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx));
2221 	}
2222 
2223 	(void) ztest_log_write(zd, tx, lr);
2224 
2225 	dmu_buf_rele(db, FTAG);
2226 
2227 	dmu_tx_commit(tx);
2228 
2229 	ztest_range_unlock(rl);
2230 	ztest_object_unlock(zd, lr->lr_foid);
2231 
2232 	return (0);
2233 }
2234 
2235 static int
2236 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
2237 {
2238 	ztest_ds_t *zd = arg1;
2239 	lr_truncate_t *lr = arg2;
2240 	objset_t *os = zd->zd_os;
2241 	dmu_tx_t *tx;
2242 	uint64_t txg;
2243 	rl_t *rl;
2244 
2245 	if (byteswap)
2246 		byteswap_uint64_array(lr, sizeof (*lr));
2247 
2248 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
2249 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
2250 	    RL_WRITER);
2251 
2252 	tx = dmu_tx_create(os);
2253 
2254 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
2255 
2256 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2257 	if (txg == 0) {
2258 		ztest_range_unlock(rl);
2259 		ztest_object_unlock(zd, lr->lr_foid);
2260 		return (ENOSPC);
2261 	}
2262 
2263 	VERIFY0(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
2264 	    lr->lr_length, tx));
2265 
2266 	(void) ztest_log_truncate(zd, tx, lr);
2267 
2268 	dmu_tx_commit(tx);
2269 
2270 	ztest_range_unlock(rl);
2271 	ztest_object_unlock(zd, lr->lr_foid);
2272 
2273 	return (0);
2274 }
2275 
2276 static int
2277 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
2278 {
2279 	ztest_ds_t *zd = arg1;
2280 	lr_setattr_t *lr = arg2;
2281 	objset_t *os = zd->zd_os;
2282 	dmu_tx_t *tx;
2283 	dmu_buf_t *db;
2284 	ztest_block_tag_t *bbt;
2285 	uint64_t txg, lrtxg, crtxg, dnodesize;
2286 
2287 	if (byteswap)
2288 		byteswap_uint64_array(lr, sizeof (*lr));
2289 
2290 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
2291 
2292 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2293 
2294 	tx = dmu_tx_create(os);
2295 	dmu_tx_hold_bonus(tx, lr->lr_foid);
2296 
2297 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2298 	if (txg == 0) {
2299 		dmu_buf_rele(db, FTAG);
2300 		ztest_object_unlock(zd, lr->lr_foid);
2301 		return (ENOSPC);
2302 	}
2303 
2304 	bbt = ztest_bt_bonus(db);
2305 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2306 	crtxg = bbt->bt_crtxg;
2307 	lrtxg = lr->lr_common.lrc_txg;
2308 	dnodesize = bbt->bt_dnodesize;
2309 
2310 	if (zd->zd_zilog->zl_replay) {
2311 		ASSERT3U(lr->lr_size, !=, 0);
2312 		ASSERT3U(lr->lr_mode, !=, 0);
2313 		ASSERT3U(lrtxg, !=, 0);
2314 	} else {
2315 		/*
2316 		 * Randomly change the size and increment the generation.
2317 		 */
2318 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2319 		    sizeof (*bbt);
2320 		lr->lr_mode = bbt->bt_gen + 1;
2321 		ASSERT0(lrtxg);
2322 	}
2323 
2324 	/*
2325 	 * Verify that the current bonus buffer is not newer than our txg.
2326 	 */
2327 	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2328 	    MAX(txg, lrtxg), crtxg);
2329 
2330 	dmu_buf_will_dirty(db, tx);
2331 
2332 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2333 	ASSERT3U(lr->lr_size, <=, db->db_size);
2334 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2335 	bbt = ztest_bt_bonus(db);
2336 
2337 	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2338 	    txg, crtxg);
2339 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2340 	dmu_buf_rele(db, FTAG);
2341 
2342 	(void) ztest_log_setattr(zd, tx, lr);
2343 
2344 	dmu_tx_commit(tx);
2345 
2346 	ztest_object_unlock(zd, lr->lr_foid);
2347 
2348 	return (0);
2349 }
2350 
2351 static zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
2352 	NULL,			/* 0 no such transaction type */
2353 	ztest_replay_create,	/* TX_CREATE */
2354 	NULL,			/* TX_MKDIR */
2355 	NULL,			/* TX_MKXATTR */
2356 	NULL,			/* TX_SYMLINK */
2357 	ztest_replay_remove,	/* TX_REMOVE */
2358 	NULL,			/* TX_RMDIR */
2359 	NULL,			/* TX_LINK */
2360 	NULL,			/* TX_RENAME */
2361 	ztest_replay_write,	/* TX_WRITE */
2362 	ztest_replay_truncate,	/* TX_TRUNCATE */
2363 	ztest_replay_setattr,	/* TX_SETATTR */
2364 	NULL,			/* TX_ACL */
2365 	NULL,			/* TX_CREATE_ACL */
2366 	NULL,			/* TX_CREATE_ATTR */
2367 	NULL,			/* TX_CREATE_ACL_ATTR */
2368 	NULL,			/* TX_MKDIR_ACL */
2369 	NULL,			/* TX_MKDIR_ATTR */
2370 	NULL,			/* TX_MKDIR_ACL_ATTR */
2371 	NULL,			/* TX_WRITE2 */
2372 	NULL,			/* TX_SETSAXATTR */
2373 	NULL,			/* TX_RENAME_EXCHANGE */
2374 	NULL,			/* TX_RENAME_WHITEOUT */
2375 };
2376 
2377 /*
2378  * ZIL get_data callbacks
2379  */
2380 
2381 static void
2382 ztest_get_done(zgd_t *zgd, int error)
2383 {
2384 	(void) error;
2385 	ztest_ds_t *zd = zgd->zgd_private;
2386 	uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object;
2387 
2388 	if (zgd->zgd_db)
2389 		dmu_buf_rele(zgd->zgd_db, zgd);
2390 
2391 	ztest_range_unlock((rl_t *)zgd->zgd_lr);
2392 	ztest_object_unlock(zd, object);
2393 
2394 	umem_free(zgd, sizeof (*zgd));
2395 }
2396 
2397 static int
2398 ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
2399     struct lwb *lwb, zio_t *zio)
2400 {
2401 	(void) arg2;
2402 	ztest_ds_t *zd = arg;
2403 	objset_t *os = zd->zd_os;
2404 	uint64_t object = lr->lr_foid;
2405 	uint64_t offset = lr->lr_offset;
2406 	uint64_t size = lr->lr_length;
2407 	uint64_t txg = lr->lr_common.lrc_txg;
2408 	uint64_t crtxg;
2409 	dmu_object_info_t doi;
2410 	dmu_buf_t *db;
2411 	zgd_t *zgd;
2412 	int error;
2413 
2414 	ASSERT3P(lwb, !=, NULL);
2415 	ASSERT3U(size, !=, 0);
2416 
2417 	ztest_object_lock(zd, object, RL_READER);
2418 	error = dmu_bonus_hold(os, object, FTAG, &db);
2419 	if (error) {
2420 		ztest_object_unlock(zd, object);
2421 		return (error);
2422 	}
2423 
2424 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
2425 
2426 	if (crtxg == 0 || crtxg > txg) {
2427 		dmu_buf_rele(db, FTAG);
2428 		ztest_object_unlock(zd, object);
2429 		return (ENOENT);
2430 	}
2431 
2432 	dmu_object_info_from_db(db, &doi);
2433 	dmu_buf_rele(db, FTAG);
2434 	db = NULL;
2435 
2436 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2437 	zgd->zgd_lwb = lwb;
2438 	zgd->zgd_private = zd;
2439 
2440 	if (buf != NULL) {	/* immediate write */
2441 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2442 		    object, offset, size, RL_READER);
2443 
2444 		error = dmu_read(os, object, offset, size, buf,
2445 		    DMU_READ_NO_PREFETCH);
2446 		ASSERT0(error);
2447 	} else {
2448 		ASSERT3P(zio, !=, NULL);
2449 		size = doi.doi_data_block_size;
2450 		if (ISP2(size)) {
2451 			offset = P2ALIGN(offset, size);
2452 		} else {
2453 			ASSERT3U(offset, <, size);
2454 			offset = 0;
2455 		}
2456 
2457 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2458 		    object, offset, size, RL_READER);
2459 
2460 		error = dmu_buf_hold_noread(os, object, offset, zgd, &db);
2461 
2462 		if (error == 0) {
2463 			blkptr_t *bp = &lr->lr_blkptr;
2464 
2465 			zgd->zgd_db = db;
2466 			zgd->zgd_bp = bp;
2467 
2468 			ASSERT3U(db->db_offset, ==, offset);
2469 			ASSERT3U(db->db_size, ==, size);
2470 
2471 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
2472 			    ztest_get_done, zgd);
2473 
2474 			if (error == 0)
2475 				return (0);
2476 		}
2477 	}
2478 
2479 	ztest_get_done(zgd, error);
2480 
2481 	return (error);
2482 }
2483 
2484 static void *
2485 ztest_lr_alloc(size_t lrsize, char *name)
2486 {
2487 	char *lr;
2488 	size_t namesize = name ? strlen(name) + 1 : 0;
2489 
2490 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2491 
2492 	if (name)
2493 		memcpy(lr + lrsize, name, namesize);
2494 
2495 	return (lr);
2496 }
2497 
2498 static void
2499 ztest_lr_free(void *lr, size_t lrsize, char *name)
2500 {
2501 	size_t namesize = name ? strlen(name) + 1 : 0;
2502 
2503 	umem_free(lr, lrsize + namesize);
2504 }
2505 
2506 /*
2507  * Lookup a bunch of objects.  Returns the number of objects not found.
2508  */
2509 static int
2510 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2511 {
2512 	int missing = 0;
2513 	int error;
2514 	int i;
2515 
2516 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2517 
2518 	for (i = 0; i < count; i++, od++) {
2519 		od->od_object = 0;
2520 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2521 		    sizeof (uint64_t), 1, &od->od_object);
2522 		if (error) {
2523 			ASSERT3S(error, ==, ENOENT);
2524 			ASSERT0(od->od_object);
2525 			missing++;
2526 		} else {
2527 			dmu_buf_t *db;
2528 			ztest_block_tag_t *bbt;
2529 			dmu_object_info_t doi;
2530 
2531 			ASSERT3U(od->od_object, !=, 0);
2532 			ASSERT0(missing);	/* there should be no gaps */
2533 
2534 			ztest_object_lock(zd, od->od_object, RL_READER);
2535 			VERIFY0(dmu_bonus_hold(zd->zd_os, od->od_object,
2536 			    FTAG, &db));
2537 			dmu_object_info_from_db(db, &doi);
2538 			bbt = ztest_bt_bonus(db);
2539 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2540 			od->od_type = doi.doi_type;
2541 			od->od_blocksize = doi.doi_data_block_size;
2542 			od->od_gen = bbt->bt_gen;
2543 			dmu_buf_rele(db, FTAG);
2544 			ztest_object_unlock(zd, od->od_object);
2545 		}
2546 	}
2547 
2548 	return (missing);
2549 }
2550 
2551 static int
2552 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2553 {
2554 	int missing = 0;
2555 	int i;
2556 
2557 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2558 
2559 	for (i = 0; i < count; i++, od++) {
2560 		if (missing) {
2561 			od->od_object = 0;
2562 			missing++;
2563 			continue;
2564 		}
2565 
2566 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2567 
2568 		lr->lr_doid = od->od_dir;
2569 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2570 		lr->lrz_type = od->od_crtype;
2571 		lr->lrz_blocksize = od->od_crblocksize;
2572 		lr->lrz_ibshift = ztest_random_ibshift();
2573 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2574 		lr->lrz_dnodesize = od->od_crdnodesize;
2575 		lr->lr_gen = od->od_crgen;
2576 		lr->lr_crtime[0] = time(NULL);
2577 
2578 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2579 			ASSERT0(missing);
2580 			od->od_object = 0;
2581 			missing++;
2582 		} else {
2583 			od->od_object = lr->lr_foid;
2584 			od->od_type = od->od_crtype;
2585 			od->od_blocksize = od->od_crblocksize;
2586 			od->od_gen = od->od_crgen;
2587 			ASSERT3U(od->od_object, !=, 0);
2588 		}
2589 
2590 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2591 	}
2592 
2593 	return (missing);
2594 }
2595 
2596 static int
2597 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2598 {
2599 	int missing = 0;
2600 	int error;
2601 	int i;
2602 
2603 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2604 
2605 	od += count - 1;
2606 
2607 	for (i = count - 1; i >= 0; i--, od--) {
2608 		if (missing) {
2609 			missing++;
2610 			continue;
2611 		}
2612 
2613 		/*
2614 		 * No object was found.
2615 		 */
2616 		if (od->od_object == 0)
2617 			continue;
2618 
2619 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2620 
2621 		lr->lr_doid = od->od_dir;
2622 
2623 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2624 			ASSERT3U(error, ==, ENOSPC);
2625 			missing++;
2626 		} else {
2627 			od->od_object = 0;
2628 		}
2629 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2630 	}
2631 
2632 	return (missing);
2633 }
2634 
2635 static int
2636 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2637     void *data)
2638 {
2639 	lr_write_t *lr;
2640 	int error;
2641 
2642 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2643 
2644 	lr->lr_foid = object;
2645 	lr->lr_offset = offset;
2646 	lr->lr_length = size;
2647 	lr->lr_blkoff = 0;
2648 	BP_ZERO(&lr->lr_blkptr);
2649 
2650 	memcpy(lr + 1, data, size);
2651 
2652 	error = ztest_replay_write(zd, lr, B_FALSE);
2653 
2654 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2655 
2656 	return (error);
2657 }
2658 
2659 static int
2660 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2661 {
2662 	lr_truncate_t *lr;
2663 	int error;
2664 
2665 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2666 
2667 	lr->lr_foid = object;
2668 	lr->lr_offset = offset;
2669 	lr->lr_length = size;
2670 
2671 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2672 
2673 	ztest_lr_free(lr, sizeof (*lr), NULL);
2674 
2675 	return (error);
2676 }
2677 
2678 static int
2679 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2680 {
2681 	lr_setattr_t *lr;
2682 	int error;
2683 
2684 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2685 
2686 	lr->lr_foid = object;
2687 	lr->lr_size = 0;
2688 	lr->lr_mode = 0;
2689 
2690 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2691 
2692 	ztest_lr_free(lr, sizeof (*lr), NULL);
2693 
2694 	return (error);
2695 }
2696 
2697 static void
2698 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2699 {
2700 	objset_t *os = zd->zd_os;
2701 	dmu_tx_t *tx;
2702 	uint64_t txg;
2703 	rl_t *rl;
2704 
2705 	txg_wait_synced(dmu_objset_pool(os), 0);
2706 
2707 	ztest_object_lock(zd, object, RL_READER);
2708 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2709 
2710 	tx = dmu_tx_create(os);
2711 
2712 	dmu_tx_hold_write(tx, object, offset, size);
2713 
2714 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2715 
2716 	if (txg != 0) {
2717 		dmu_prealloc(os, object, offset, size, tx);
2718 		dmu_tx_commit(tx);
2719 		txg_wait_synced(dmu_objset_pool(os), txg);
2720 	} else {
2721 		(void) dmu_free_long_range(os, object, offset, size);
2722 	}
2723 
2724 	ztest_range_unlock(rl);
2725 	ztest_object_unlock(zd, object);
2726 }
2727 
2728 static void
2729 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2730 {
2731 	int err;
2732 	ztest_block_tag_t wbt;
2733 	dmu_object_info_t doi;
2734 	enum ztest_io_type io_type;
2735 	uint64_t blocksize;
2736 	void *data;
2737 
2738 	VERIFY0(dmu_object_info(zd->zd_os, object, &doi));
2739 	blocksize = doi.doi_data_block_size;
2740 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2741 
2742 	/*
2743 	 * Pick an i/o type at random, biased toward writing block tags.
2744 	 */
2745 	io_type = ztest_random(ZTEST_IO_TYPES);
2746 	if (ztest_random(2) == 0)
2747 		io_type = ZTEST_IO_WRITE_TAG;
2748 
2749 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2750 
2751 	switch (io_type) {
2752 
2753 	case ZTEST_IO_WRITE_TAG:
2754 		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2755 		    offset, 0, 0, 0);
2756 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2757 		break;
2758 
2759 	case ZTEST_IO_WRITE_PATTERN:
2760 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2761 		if (ztest_random(2) == 0) {
2762 			/*
2763 			 * Induce fletcher2 collisions to ensure that
2764 			 * zio_ddt_collision() detects and resolves them
2765 			 * when using fletcher2-verify for deduplication.
2766 			 */
2767 			((uint64_t *)data)[0] ^= 1ULL << 63;
2768 			((uint64_t *)data)[4] ^= 1ULL << 63;
2769 		}
2770 		(void) ztest_write(zd, object, offset, blocksize, data);
2771 		break;
2772 
2773 	case ZTEST_IO_WRITE_ZEROES:
2774 		memset(data, 0, blocksize);
2775 		(void) ztest_write(zd, object, offset, blocksize, data);
2776 		break;
2777 
2778 	case ZTEST_IO_TRUNCATE:
2779 		(void) ztest_truncate(zd, object, offset, blocksize);
2780 		break;
2781 
2782 	case ZTEST_IO_SETATTR:
2783 		(void) ztest_setattr(zd, object);
2784 		break;
2785 	default:
2786 		break;
2787 
2788 	case ZTEST_IO_REWRITE:
2789 		(void) pthread_rwlock_rdlock(&ztest_name_lock);
2790 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2791 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2792 		    B_FALSE);
2793 		ASSERT(err == 0 || err == ENOSPC);
2794 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2795 		    ZFS_PROP_COMPRESSION,
2796 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2797 		    B_FALSE);
2798 		ASSERT(err == 0 || err == ENOSPC);
2799 		(void) pthread_rwlock_unlock(&ztest_name_lock);
2800 
2801 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2802 		    DMU_READ_NO_PREFETCH));
2803 
2804 		(void) ztest_write(zd, object, offset, blocksize, data);
2805 		break;
2806 	}
2807 
2808 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2809 
2810 	umem_free(data, blocksize);
2811 }
2812 
2813 /*
2814  * Initialize an object description template.
2815  */
2816 static void
2817 ztest_od_init(ztest_od_t *od, uint64_t id, const char *tag, uint64_t index,
2818     dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2819     uint64_t gen)
2820 {
2821 	od->od_dir = ZTEST_DIROBJ;
2822 	od->od_object = 0;
2823 
2824 	od->od_crtype = type;
2825 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2826 	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2827 	od->od_crgen = gen;
2828 
2829 	od->od_type = DMU_OT_NONE;
2830 	od->od_blocksize = 0;
2831 	od->od_gen = 0;
2832 
2833 	(void) snprintf(od->od_name, sizeof (od->od_name),
2834 	    "%s(%"PRId64")[%"PRIu64"]",
2835 	    tag, id, index);
2836 }
2837 
2838 /*
2839  * Lookup or create the objects for a test using the od template.
2840  * If the objects do not all exist, or if 'remove' is specified,
2841  * remove any existing objects and create new ones.  Otherwise,
2842  * use the existing objects.
2843  */
2844 static int
2845 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2846 {
2847 	int count = size / sizeof (*od);
2848 	int rv = 0;
2849 
2850 	mutex_enter(&zd->zd_dirobj_lock);
2851 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2852 	    (ztest_remove(zd, od, count) != 0 ||
2853 	    ztest_create(zd, od, count) != 0))
2854 		rv = -1;
2855 	zd->zd_od = od;
2856 	mutex_exit(&zd->zd_dirobj_lock);
2857 
2858 	return (rv);
2859 }
2860 
2861 void
2862 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2863 {
2864 	(void) id;
2865 	zilog_t *zilog = zd->zd_zilog;
2866 
2867 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2868 
2869 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2870 
2871 	/*
2872 	 * Remember the committed values in zd, which is in parent/child
2873 	 * shared memory.  If we die, the next iteration of ztest_run()
2874 	 * will verify that the log really does contain this record.
2875 	 */
2876 	mutex_enter(&zilog->zl_lock);
2877 	ASSERT3P(zd->zd_shared, !=, NULL);
2878 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2879 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2880 	mutex_exit(&zilog->zl_lock);
2881 
2882 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2883 }
2884 
2885 /*
2886  * This function is designed to simulate the operations that occur during a
2887  * mount/unmount operation.  We hold the dataset across these operations in an
2888  * attempt to expose any implicit assumptions about ZIL management.
2889  */
2890 void
2891 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2892 {
2893 	(void) id;
2894 	objset_t *os = zd->zd_os;
2895 
2896 	/*
2897 	 * We hold the ztest_vdev_lock so we don't cause problems with
2898 	 * other threads that wish to remove a log device, such as
2899 	 * ztest_device_removal().
2900 	 */
2901 	mutex_enter(&ztest_vdev_lock);
2902 
2903 	/*
2904 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2905 	 * updating the zil (i.e. adding in-memory log records) and the
2906 	 * zd_zilog_lock to block any I/O.
2907 	 */
2908 	mutex_enter(&zd->zd_dirobj_lock);
2909 	(void) pthread_rwlock_wrlock(&zd->zd_zilog_lock);
2910 
2911 	/* zfsvfs_teardown() */
2912 	zil_close(zd->zd_zilog);
2913 
2914 	/* zfsvfs_setup() */
2915 	VERIFY3P(zil_open(os, ztest_get_data, NULL), ==, zd->zd_zilog);
2916 	zil_replay(os, zd, ztest_replay_vector);
2917 
2918 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2919 	mutex_exit(&zd->zd_dirobj_lock);
2920 	mutex_exit(&ztest_vdev_lock);
2921 }
2922 
2923 /*
2924  * Verify that we can't destroy an active pool, create an existing pool,
2925  * or create a pool with a bad vdev spec.
2926  */
2927 void
2928 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2929 {
2930 	(void) zd, (void) id;
2931 	ztest_shared_opts_t *zo = &ztest_opts;
2932 	spa_t *spa;
2933 	nvlist_t *nvroot;
2934 
2935 	if (zo->zo_mmp_test)
2936 		return;
2937 
2938 	/*
2939 	 * Attempt to create using a bad file.
2940 	 */
2941 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2942 	VERIFY3U(ENOENT, ==,
2943 	    spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2944 	fnvlist_free(nvroot);
2945 
2946 	/*
2947 	 * Attempt to create using a bad mirror.
2948 	 */
2949 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 2, 1);
2950 	VERIFY3U(ENOENT, ==,
2951 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2952 	fnvlist_free(nvroot);
2953 
2954 	/*
2955 	 * Attempt to create an existing pool.  It shouldn't matter
2956 	 * what's in the nvroot; we should fail with EEXIST.
2957 	 */
2958 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
2959 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2960 	VERIFY3U(EEXIST, ==,
2961 	    spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2962 	fnvlist_free(nvroot);
2963 
2964 	/*
2965 	 * We open a reference to the spa and then we try to export it
2966 	 * expecting one of the following errors:
2967 	 *
2968 	 * EBUSY
2969 	 *	Because of the reference we just opened.
2970 	 *
2971 	 * ZFS_ERR_EXPORT_IN_PROGRESS
2972 	 *	For the case that there is another ztest thread doing
2973 	 *	an export concurrently.
2974 	 */
2975 	VERIFY0(spa_open(zo->zo_pool, &spa, FTAG));
2976 	int error = spa_destroy(zo->zo_pool);
2977 	if (error != EBUSY && error != ZFS_ERR_EXPORT_IN_PROGRESS) {
2978 		fatal(B_FALSE, "spa_destroy(%s) returned unexpected value %d",
2979 		    spa->spa_name, error);
2980 	}
2981 	spa_close(spa, FTAG);
2982 
2983 	(void) pthread_rwlock_unlock(&ztest_name_lock);
2984 }
2985 
2986 /*
2987  * Start and then stop the MMP threads to ensure the startup and shutdown code
2988  * works properly.  Actual protection and property-related code tested via ZTS.
2989  */
2990 void
2991 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2992 {
2993 	(void) zd, (void) id;
2994 	ztest_shared_opts_t *zo = &ztest_opts;
2995 	spa_t *spa = ztest_spa;
2996 
2997 	if (zo->zo_mmp_test)
2998 		return;
2999 
3000 	/*
3001 	 * Since enabling MMP involves setting a property, it could not be done
3002 	 * while the pool is suspended.
3003 	 */
3004 	if (spa_suspended(spa))
3005 		return;
3006 
3007 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3008 	mutex_enter(&spa->spa_props_lock);
3009 
3010 	zfs_multihost_fail_intervals = 0;
3011 
3012 	if (!spa_multihost(spa)) {
3013 		spa->spa_multihost = B_TRUE;
3014 		mmp_thread_start(spa);
3015 	}
3016 
3017 	mutex_exit(&spa->spa_props_lock);
3018 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3019 
3020 	txg_wait_synced(spa_get_dsl(spa), 0);
3021 	mmp_signal_all_threads();
3022 	txg_wait_synced(spa_get_dsl(spa), 0);
3023 
3024 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3025 	mutex_enter(&spa->spa_props_lock);
3026 
3027 	if (spa_multihost(spa)) {
3028 		mmp_thread_stop(spa);
3029 		spa->spa_multihost = B_FALSE;
3030 	}
3031 
3032 	mutex_exit(&spa->spa_props_lock);
3033 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3034 }
3035 
3036 void
3037 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
3038 {
3039 	(void) zd, (void) id;
3040 	spa_t *spa;
3041 	uint64_t initial_version = SPA_VERSION_INITIAL;
3042 	uint64_t version, newversion;
3043 	nvlist_t *nvroot, *props;
3044 	char *name;
3045 
3046 	if (ztest_opts.zo_mmp_test)
3047 		return;
3048 
3049 	/* dRAID added after feature flags, skip upgrade test. */
3050 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0)
3051 		return;
3052 
3053 	mutex_enter(&ztest_vdev_lock);
3054 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
3055 
3056 	/*
3057 	 * Clean up from previous runs.
3058 	 */
3059 	(void) spa_destroy(name);
3060 
3061 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
3062 	    NULL, ztest_opts.zo_raid_children, ztest_opts.zo_mirrors, 1);
3063 
3064 	/*
3065 	 * If we're configuring a RAIDZ device then make sure that the
3066 	 * initial version is capable of supporting that feature.
3067 	 */
3068 	switch (ztest_opts.zo_raid_parity) {
3069 	case 0:
3070 	case 1:
3071 		initial_version = SPA_VERSION_INITIAL;
3072 		break;
3073 	case 2:
3074 		initial_version = SPA_VERSION_RAIDZ2;
3075 		break;
3076 	case 3:
3077 		initial_version = SPA_VERSION_RAIDZ3;
3078 		break;
3079 	}
3080 
3081 	/*
3082 	 * Create a pool with a spa version that can be upgraded. Pick
3083 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
3084 	 */
3085 	do {
3086 		version = ztest_random_spa_version(initial_version);
3087 	} while (version > SPA_VERSION_BEFORE_FEATURES);
3088 
3089 	props = fnvlist_alloc();
3090 	fnvlist_add_uint64(props,
3091 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
3092 	VERIFY0(spa_create(name, nvroot, props, NULL, NULL));
3093 	fnvlist_free(nvroot);
3094 	fnvlist_free(props);
3095 
3096 	VERIFY0(spa_open(name, &spa, FTAG));
3097 	VERIFY3U(spa_version(spa), ==, version);
3098 	newversion = ztest_random_spa_version(version + 1);
3099 
3100 	if (ztest_opts.zo_verbose >= 4) {
3101 		(void) printf("upgrading spa version from "
3102 		    "%"PRIu64" to %"PRIu64"\n",
3103 		    version, newversion);
3104 	}
3105 
3106 	spa_upgrade(spa, newversion);
3107 	VERIFY3U(spa_version(spa), >, version);
3108 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
3109 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
3110 	spa_close(spa, FTAG);
3111 
3112 	kmem_strfree(name);
3113 	mutex_exit(&ztest_vdev_lock);
3114 }
3115 
3116 static void
3117 ztest_spa_checkpoint(spa_t *spa)
3118 {
3119 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3120 
3121 	int error = spa_checkpoint(spa->spa_name);
3122 
3123 	switch (error) {
3124 	case 0:
3125 	case ZFS_ERR_DEVRM_IN_PROGRESS:
3126 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3127 	case ZFS_ERR_CHECKPOINT_EXISTS:
3128 		break;
3129 	case ENOSPC:
3130 		ztest_record_enospc(FTAG);
3131 		break;
3132 	default:
3133 		fatal(B_FALSE, "spa_checkpoint(%s) = %d", spa->spa_name, error);
3134 	}
3135 }
3136 
3137 static void
3138 ztest_spa_discard_checkpoint(spa_t *spa)
3139 {
3140 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3141 
3142 	int error = spa_checkpoint_discard(spa->spa_name);
3143 
3144 	switch (error) {
3145 	case 0:
3146 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3147 	case ZFS_ERR_NO_CHECKPOINT:
3148 		break;
3149 	default:
3150 		fatal(B_FALSE, "spa_discard_checkpoint(%s) = %d",
3151 		    spa->spa_name, error);
3152 	}
3153 
3154 }
3155 
3156 void
3157 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
3158 {
3159 	(void) zd, (void) id;
3160 	spa_t *spa = ztest_spa;
3161 
3162 	mutex_enter(&ztest_checkpoint_lock);
3163 	if (ztest_random(2) == 0) {
3164 		ztest_spa_checkpoint(spa);
3165 	} else {
3166 		ztest_spa_discard_checkpoint(spa);
3167 	}
3168 	mutex_exit(&ztest_checkpoint_lock);
3169 }
3170 
3171 
3172 static vdev_t *
3173 vdev_lookup_by_path(vdev_t *vd, const char *path)
3174 {
3175 	vdev_t *mvd;
3176 	int c;
3177 
3178 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
3179 		return (vd);
3180 
3181 	for (c = 0; c < vd->vdev_children; c++)
3182 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
3183 		    NULL)
3184 			return (mvd);
3185 
3186 	return (NULL);
3187 }
3188 
3189 static int
3190 spa_num_top_vdevs(spa_t *spa)
3191 {
3192 	vdev_t *rvd = spa->spa_root_vdev;
3193 	ASSERT3U(spa_config_held(spa, SCL_VDEV, RW_READER), ==, SCL_VDEV);
3194 	return (rvd->vdev_children);
3195 }
3196 
3197 /*
3198  * Verify that vdev_add() works as expected.
3199  */
3200 void
3201 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
3202 {
3203 	(void) zd, (void) id;
3204 	ztest_shared_t *zs = ztest_shared;
3205 	spa_t *spa = ztest_spa;
3206 	uint64_t leaves;
3207 	uint64_t guid;
3208 	nvlist_t *nvroot;
3209 	int error;
3210 
3211 	if (ztest_opts.zo_mmp_test)
3212 		return;
3213 
3214 	mutex_enter(&ztest_vdev_lock);
3215 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) *
3216 	    ztest_opts.zo_raid_children;
3217 
3218 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3219 
3220 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3221 
3222 	/*
3223 	 * If we have slogs then remove them 1/4 of the time.
3224 	 */
3225 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
3226 		metaslab_group_t *mg;
3227 
3228 		/*
3229 		 * find the first real slog in log allocation class
3230 		 */
3231 		mg =  spa_log_class(spa)->mc_allocator[0].mca_rotor;
3232 		while (!mg->mg_vd->vdev_islog)
3233 			mg = mg->mg_next;
3234 
3235 		guid = mg->mg_vd->vdev_guid;
3236 
3237 		spa_config_exit(spa, SCL_VDEV, FTAG);
3238 
3239 		/*
3240 		 * We have to grab the zs_name_lock as writer to
3241 		 * prevent a race between removing a slog (dmu_objset_find)
3242 		 * and destroying a dataset. Removing the slog will
3243 		 * grab a reference on the dataset which may cause
3244 		 * dsl_destroy_head() to fail with EBUSY thus
3245 		 * leaving the dataset in an inconsistent state.
3246 		 */
3247 		pthread_rwlock_wrlock(&ztest_name_lock);
3248 		error = spa_vdev_remove(spa, guid, B_FALSE);
3249 		pthread_rwlock_unlock(&ztest_name_lock);
3250 
3251 		switch (error) {
3252 		case 0:
3253 		case EEXIST:	/* Generic zil_reset() error */
3254 		case EBUSY:	/* Replay required */
3255 		case EACCES:	/* Crypto key not loaded */
3256 		case ZFS_ERR_CHECKPOINT_EXISTS:
3257 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3258 			break;
3259 		default:
3260 			fatal(B_FALSE, "spa_vdev_remove() = %d", error);
3261 		}
3262 	} else {
3263 		spa_config_exit(spa, SCL_VDEV, FTAG);
3264 
3265 		/*
3266 		 * Make 1/4 of the devices be log devices
3267 		 */
3268 		nvroot = make_vdev_root(NULL, NULL, NULL,
3269 		    ztest_opts.zo_vdev_size, 0, (ztest_random(4) == 0) ?
3270 		    "log" : NULL, ztest_opts.zo_raid_children, zs->zs_mirrors,
3271 		    1);
3272 
3273 		error = spa_vdev_add(spa, nvroot);
3274 		fnvlist_free(nvroot);
3275 
3276 		switch (error) {
3277 		case 0:
3278 			break;
3279 		case ENOSPC:
3280 			ztest_record_enospc("spa_vdev_add");
3281 			break;
3282 		default:
3283 			fatal(B_FALSE, "spa_vdev_add() = %d", error);
3284 		}
3285 	}
3286 
3287 	mutex_exit(&ztest_vdev_lock);
3288 }
3289 
3290 void
3291 ztest_vdev_class_add(ztest_ds_t *zd, uint64_t id)
3292 {
3293 	(void) zd, (void) id;
3294 	ztest_shared_t *zs = ztest_shared;
3295 	spa_t *spa = ztest_spa;
3296 	uint64_t leaves;
3297 	nvlist_t *nvroot;
3298 	const char *class = (ztest_random(2) == 0) ?
3299 	    VDEV_ALLOC_BIAS_SPECIAL : VDEV_ALLOC_BIAS_DEDUP;
3300 	int error;
3301 
3302 	/*
3303 	 * By default add a special vdev 50% of the time
3304 	 */
3305 	if ((ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_OFF) ||
3306 	    (ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_RND &&
3307 	    ztest_random(2) == 0)) {
3308 		return;
3309 	}
3310 
3311 	mutex_enter(&ztest_vdev_lock);
3312 
3313 	/* Only test with mirrors */
3314 	if (zs->zs_mirrors < 2) {
3315 		mutex_exit(&ztest_vdev_lock);
3316 		return;
3317 	}
3318 
3319 	/* requires feature@allocation_classes */
3320 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) {
3321 		mutex_exit(&ztest_vdev_lock);
3322 		return;
3323 	}
3324 
3325 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) *
3326 	    ztest_opts.zo_raid_children;
3327 
3328 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3329 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3330 	spa_config_exit(spa, SCL_VDEV, FTAG);
3331 
3332 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
3333 	    class, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
3334 
3335 	error = spa_vdev_add(spa, nvroot);
3336 	fnvlist_free(nvroot);
3337 
3338 	if (error == ENOSPC)
3339 		ztest_record_enospc("spa_vdev_add");
3340 	else if (error != 0)
3341 		fatal(B_FALSE, "spa_vdev_add() = %d", error);
3342 
3343 	/*
3344 	 * 50% of the time allow small blocks in the special class
3345 	 */
3346 	if (error == 0 &&
3347 	    spa_special_class(spa)->mc_groups == 1 && ztest_random(2) == 0) {
3348 		if (ztest_opts.zo_verbose >= 3)
3349 			(void) printf("Enabling special VDEV small blocks\n");
3350 		error = ztest_dsl_prop_set_uint64(zd->zd_name,
3351 		    ZFS_PROP_SPECIAL_SMALL_BLOCKS, 32768, B_FALSE);
3352 		ASSERT(error == 0 || error == ENOSPC);
3353 	}
3354 
3355 	mutex_exit(&ztest_vdev_lock);
3356 
3357 	if (ztest_opts.zo_verbose >= 3) {
3358 		metaslab_class_t *mc;
3359 
3360 		if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL) == 0)
3361 			mc = spa_special_class(spa);
3362 		else
3363 			mc = spa_dedup_class(spa);
3364 		(void) printf("Added a %s mirrored vdev (of %d)\n",
3365 		    class, (int)mc->mc_groups);
3366 	}
3367 }
3368 
3369 /*
3370  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
3371  */
3372 void
3373 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
3374 {
3375 	(void) zd, (void) id;
3376 	ztest_shared_t *zs = ztest_shared;
3377 	spa_t *spa = ztest_spa;
3378 	vdev_t *rvd = spa->spa_root_vdev;
3379 	spa_aux_vdev_t *sav;
3380 	const char *aux;
3381 	char *path;
3382 	uint64_t guid = 0;
3383 	int error, ignore_err = 0;
3384 
3385 	if (ztest_opts.zo_mmp_test)
3386 		return;
3387 
3388 	path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3389 
3390 	if (ztest_random(2) == 0) {
3391 		sav = &spa->spa_spares;
3392 		aux = ZPOOL_CONFIG_SPARES;
3393 	} else {
3394 		sav = &spa->spa_l2cache;
3395 		aux = ZPOOL_CONFIG_L2CACHE;
3396 	}
3397 
3398 	mutex_enter(&ztest_vdev_lock);
3399 
3400 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3401 
3402 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
3403 		/*
3404 		 * Pick a random device to remove.
3405 		 */
3406 		vdev_t *svd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3407 
3408 		/* dRAID spares cannot be removed; try anyways to see ENOTSUP */
3409 		if (strstr(svd->vdev_path, VDEV_TYPE_DRAID) != NULL)
3410 			ignore_err = ENOTSUP;
3411 
3412 		guid = svd->vdev_guid;
3413 	} else {
3414 		/*
3415 		 * Find an unused device we can add.
3416 		 */
3417 		zs->zs_vdev_aux = 0;
3418 		for (;;) {
3419 			int c;
3420 			(void) snprintf(path, MAXPATHLEN, ztest_aux_template,
3421 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
3422 			    zs->zs_vdev_aux);
3423 			for (c = 0; c < sav->sav_count; c++)
3424 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
3425 				    path) == 0)
3426 					break;
3427 			if (c == sav->sav_count &&
3428 			    vdev_lookup_by_path(rvd, path) == NULL)
3429 				break;
3430 			zs->zs_vdev_aux++;
3431 		}
3432 	}
3433 
3434 	spa_config_exit(spa, SCL_VDEV, FTAG);
3435 
3436 	if (guid == 0) {
3437 		/*
3438 		 * Add a new device.
3439 		 */
3440 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3441 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, NULL, 0, 0, 1);
3442 		error = spa_vdev_add(spa, nvroot);
3443 
3444 		switch (error) {
3445 		case 0:
3446 			break;
3447 		default:
3448 			fatal(B_FALSE, "spa_vdev_add(%p) = %d", nvroot, error);
3449 		}
3450 		fnvlist_free(nvroot);
3451 	} else {
3452 		/*
3453 		 * Remove an existing device.  Sometimes, dirty its
3454 		 * vdev state first to make sure we handle removal
3455 		 * of devices that have pending state changes.
3456 		 */
3457 		if (ztest_random(2) == 0)
3458 			(void) vdev_online(spa, guid, 0, NULL);
3459 
3460 		error = spa_vdev_remove(spa, guid, B_FALSE);
3461 
3462 		switch (error) {
3463 		case 0:
3464 		case EBUSY:
3465 		case ZFS_ERR_CHECKPOINT_EXISTS:
3466 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3467 			break;
3468 		default:
3469 			if (error != ignore_err)
3470 				fatal(B_FALSE,
3471 				    "spa_vdev_remove(%"PRIu64") = %d",
3472 				    guid, error);
3473 		}
3474 	}
3475 
3476 	mutex_exit(&ztest_vdev_lock);
3477 
3478 	umem_free(path, MAXPATHLEN);
3479 }
3480 
3481 /*
3482  * split a pool if it has mirror tlvdevs
3483  */
3484 void
3485 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3486 {
3487 	(void) zd, (void) id;
3488 	ztest_shared_t *zs = ztest_shared;
3489 	spa_t *spa = ztest_spa;
3490 	vdev_t *rvd = spa->spa_root_vdev;
3491 	nvlist_t *tree, **child, *config, *split, **schild;
3492 	uint_t c, children, schildren = 0, lastlogid = 0;
3493 	int error = 0;
3494 
3495 	if (ztest_opts.zo_mmp_test)
3496 		return;
3497 
3498 	mutex_enter(&ztest_vdev_lock);
3499 
3500 	/* ensure we have a usable config; mirrors of raidz aren't supported */
3501 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raid_children > 1) {
3502 		mutex_exit(&ztest_vdev_lock);
3503 		return;
3504 	}
3505 
3506 	/* clean up the old pool, if any */
3507 	(void) spa_destroy("splitp");
3508 
3509 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3510 
3511 	/* generate a config from the existing config */
3512 	mutex_enter(&spa->spa_props_lock);
3513 	tree = fnvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE);
3514 	mutex_exit(&spa->spa_props_lock);
3515 
3516 	VERIFY0(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN,
3517 	    &child, &children));
3518 
3519 	schild = umem_alloc(rvd->vdev_children * sizeof (nvlist_t *),
3520 	    UMEM_NOFAIL);
3521 	for (c = 0; c < children; c++) {
3522 		vdev_t *tvd = rvd->vdev_child[c];
3523 		nvlist_t **mchild;
3524 		uint_t mchildren;
3525 
3526 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3527 			schild[schildren] = fnvlist_alloc();
3528 			fnvlist_add_string(schild[schildren],
3529 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE);
3530 			fnvlist_add_uint64(schild[schildren],
3531 			    ZPOOL_CONFIG_IS_HOLE, 1);
3532 			if (lastlogid == 0)
3533 				lastlogid = schildren;
3534 			++schildren;
3535 			continue;
3536 		}
3537 		lastlogid = 0;
3538 		VERIFY0(nvlist_lookup_nvlist_array(child[c],
3539 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren));
3540 		schild[schildren++] = fnvlist_dup(mchild[0]);
3541 	}
3542 
3543 	/* OK, create a config that can be used to split */
3544 	split = fnvlist_alloc();
3545 	fnvlist_add_string(split, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
3546 	fnvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN,
3547 	    (const nvlist_t **)schild, lastlogid != 0 ? lastlogid : schildren);
3548 
3549 	config = fnvlist_alloc();
3550 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split);
3551 
3552 	for (c = 0; c < schildren; c++)
3553 		fnvlist_free(schild[c]);
3554 	umem_free(schild, rvd->vdev_children * sizeof (nvlist_t *));
3555 	fnvlist_free(split);
3556 
3557 	spa_config_exit(spa, SCL_VDEV, FTAG);
3558 
3559 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
3560 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3561 	(void) pthread_rwlock_unlock(&ztest_name_lock);
3562 
3563 	fnvlist_free(config);
3564 
3565 	if (error == 0) {
3566 		(void) printf("successful split - results:\n");
3567 		mutex_enter(&spa_namespace_lock);
3568 		show_pool_stats(spa);
3569 		show_pool_stats(spa_lookup("splitp"));
3570 		mutex_exit(&spa_namespace_lock);
3571 		++zs->zs_splits;
3572 		--zs->zs_mirrors;
3573 	}
3574 	mutex_exit(&ztest_vdev_lock);
3575 }
3576 
3577 /*
3578  * Verify that we can attach and detach devices.
3579  */
3580 void
3581 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3582 {
3583 	(void) zd, (void) id;
3584 	ztest_shared_t *zs = ztest_shared;
3585 	spa_t *spa = ztest_spa;
3586 	spa_aux_vdev_t *sav = &spa->spa_spares;
3587 	vdev_t *rvd = spa->spa_root_vdev;
3588 	vdev_t *oldvd, *newvd, *pvd;
3589 	nvlist_t *root;
3590 	uint64_t leaves;
3591 	uint64_t leaf, top;
3592 	uint64_t ashift = ztest_get_ashift();
3593 	uint64_t oldguid, pguid;
3594 	uint64_t oldsize, newsize;
3595 	char *oldpath, *newpath;
3596 	int replacing;
3597 	int oldvd_has_siblings = B_FALSE;
3598 	int newvd_is_spare = B_FALSE;
3599 	int newvd_is_dspare = B_FALSE;
3600 	int oldvd_is_log;
3601 	int oldvd_is_special;
3602 	int error, expected_error;
3603 
3604 	if (ztest_opts.zo_mmp_test)
3605 		return;
3606 
3607 	oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3608 	newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3609 
3610 	mutex_enter(&ztest_vdev_lock);
3611 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raid_children;
3612 
3613 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3614 
3615 	/*
3616 	 * If a vdev is in the process of being removed, its removal may
3617 	 * finish while we are in progress, leading to an unexpected error
3618 	 * value.  Don't bother trying to attach while we are in the middle
3619 	 * of removal.
3620 	 */
3621 	if (ztest_device_removal_active) {
3622 		spa_config_exit(spa, SCL_ALL, FTAG);
3623 		goto out;
3624 	}
3625 
3626 	/*
3627 	 * Decide whether to do an attach or a replace.
3628 	 */
3629 	replacing = ztest_random(2);
3630 
3631 	/*
3632 	 * Pick a random top-level vdev.
3633 	 */
3634 	top = ztest_random_vdev_top(spa, B_TRUE);
3635 
3636 	/*
3637 	 * Pick a random leaf within it.
3638 	 */
3639 	leaf = ztest_random(leaves);
3640 
3641 	/*
3642 	 * Locate this vdev.
3643 	 */
3644 	oldvd = rvd->vdev_child[top];
3645 
3646 	/* pick a child from the mirror */
3647 	if (zs->zs_mirrors >= 1) {
3648 		ASSERT3P(oldvd->vdev_ops, ==, &vdev_mirror_ops);
3649 		ASSERT3U(oldvd->vdev_children, >=, zs->zs_mirrors);
3650 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raid_children];
3651 	}
3652 
3653 	/* pick a child out of the raidz group */
3654 	if (ztest_opts.zo_raid_children > 1) {
3655 		if (strcmp(oldvd->vdev_ops->vdev_op_type, "raidz") == 0)
3656 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_raidz_ops);
3657 		else
3658 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_draid_ops);
3659 		ASSERT3U(oldvd->vdev_children, ==, ztest_opts.zo_raid_children);
3660 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raid_children];
3661 	}
3662 
3663 	/*
3664 	 * If we're already doing an attach or replace, oldvd may be a
3665 	 * mirror vdev -- in which case, pick a random child.
3666 	 */
3667 	while (oldvd->vdev_children != 0) {
3668 		oldvd_has_siblings = B_TRUE;
3669 		ASSERT3U(oldvd->vdev_children, >=, 2);
3670 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3671 	}
3672 
3673 	oldguid = oldvd->vdev_guid;
3674 	oldsize = vdev_get_min_asize(oldvd);
3675 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
3676 	oldvd_is_special =
3677 	    oldvd->vdev_top->vdev_alloc_bias == VDEV_BIAS_SPECIAL ||
3678 	    oldvd->vdev_top->vdev_alloc_bias == VDEV_BIAS_DEDUP;
3679 	(void) strlcpy(oldpath, oldvd->vdev_path, MAXPATHLEN);
3680 	pvd = oldvd->vdev_parent;
3681 	pguid = pvd->vdev_guid;
3682 
3683 	/*
3684 	 * If oldvd has siblings, then half of the time, detach it.  Prior
3685 	 * to the detach the pool is scrubbed in order to prevent creating
3686 	 * unrepairable blocks as a result of the data corruption injection.
3687 	 */
3688 	if (oldvd_has_siblings && ztest_random(2) == 0) {
3689 		spa_config_exit(spa, SCL_ALL, FTAG);
3690 
3691 		error = ztest_scrub_impl(spa);
3692 		if (error)
3693 			goto out;
3694 
3695 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3696 		if (error != 0 && error != ENODEV && error != EBUSY &&
3697 		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3698 		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
3699 			fatal(B_FALSE, "detach (%s) returned %d",
3700 			    oldpath, error);
3701 		goto out;
3702 	}
3703 
3704 	/*
3705 	 * For the new vdev, choose with equal probability between the two
3706 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3707 	 */
3708 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
3709 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3710 		newvd_is_spare = B_TRUE;
3711 
3712 		if (newvd->vdev_ops == &vdev_draid_spare_ops)
3713 			newvd_is_dspare = B_TRUE;
3714 
3715 		(void) strlcpy(newpath, newvd->vdev_path, MAXPATHLEN);
3716 	} else {
3717 		(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3718 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
3719 		    top * leaves + leaf);
3720 		if (ztest_random(2) == 0)
3721 			newpath[strlen(newpath) - 1] = 'b';
3722 		newvd = vdev_lookup_by_path(rvd, newpath);
3723 	}
3724 
3725 	if (newvd) {
3726 		/*
3727 		 * Reopen to ensure the vdev's asize field isn't stale.
3728 		 */
3729 		vdev_reopen(newvd);
3730 		newsize = vdev_get_min_asize(newvd);
3731 	} else {
3732 		/*
3733 		 * Make newsize a little bigger or smaller than oldsize.
3734 		 * If it's smaller, the attach should fail.
3735 		 * If it's larger, and we're doing a replace,
3736 		 * we should get dynamic LUN growth when we're done.
3737 		 */
3738 		newsize = 10 * oldsize / (9 + ztest_random(3));
3739 	}
3740 
3741 	/*
3742 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3743 	 * unless it's a replace; in that case any non-replacing parent is OK.
3744 	 *
3745 	 * If newvd is already part of the pool, it should fail with EBUSY.
3746 	 *
3747 	 * If newvd is too small, it should fail with EOVERFLOW.
3748 	 *
3749 	 * If newvd is a distributed spare and it's being attached to a
3750 	 * dRAID which is not its parent it should fail with EINVAL.
3751 	 */
3752 	if (pvd->vdev_ops != &vdev_mirror_ops &&
3753 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3754 	    pvd->vdev_ops == &vdev_replacing_ops ||
3755 	    pvd->vdev_ops == &vdev_spare_ops))
3756 		expected_error = ENOTSUP;
3757 	else if (newvd_is_spare &&
3758 	    (!replacing || oldvd_is_log || oldvd_is_special))
3759 		expected_error = ENOTSUP;
3760 	else if (newvd == oldvd)
3761 		expected_error = replacing ? 0 : EBUSY;
3762 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3763 		expected_error = EBUSY;
3764 	else if (!newvd_is_dspare && newsize < oldsize)
3765 		expected_error = EOVERFLOW;
3766 	else if (ashift > oldvd->vdev_top->vdev_ashift)
3767 		expected_error = EDOM;
3768 	else if (newvd_is_dspare && pvd != vdev_draid_spare_get_parent(newvd))
3769 		expected_error = EINVAL;
3770 	else
3771 		expected_error = 0;
3772 
3773 	spa_config_exit(spa, SCL_ALL, FTAG);
3774 
3775 	/*
3776 	 * Build the nvlist describing newpath.
3777 	 */
3778 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3779 	    ashift, NULL, 0, 0, 1);
3780 
3781 	/*
3782 	 * When supported select either a healing or sequential resilver.
3783 	 */
3784 	boolean_t rebuilding = B_FALSE;
3785 	if (pvd->vdev_ops == &vdev_mirror_ops ||
3786 	    pvd->vdev_ops ==  &vdev_root_ops) {
3787 		rebuilding = !!ztest_random(2);
3788 	}
3789 
3790 	error = spa_vdev_attach(spa, oldguid, root, replacing, rebuilding);
3791 
3792 	fnvlist_free(root);
3793 
3794 	/*
3795 	 * If our parent was the replacing vdev, but the replace completed,
3796 	 * then instead of failing with ENOTSUP we may either succeed,
3797 	 * fail with ENODEV, or fail with EOVERFLOW.
3798 	 */
3799 	if (expected_error == ENOTSUP &&
3800 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3801 		expected_error = error;
3802 
3803 	/*
3804 	 * If someone grew the LUN, the replacement may be too small.
3805 	 */
3806 	if (error == EOVERFLOW || error == EBUSY)
3807 		expected_error = error;
3808 
3809 	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3810 	    error == ZFS_ERR_DISCARDING_CHECKPOINT ||
3811 	    error == ZFS_ERR_RESILVER_IN_PROGRESS ||
3812 	    error == ZFS_ERR_REBUILD_IN_PROGRESS)
3813 		expected_error = error;
3814 
3815 	if (error != expected_error && expected_error != EBUSY) {
3816 		fatal(B_FALSE, "attach (%s %"PRIu64", %s %"PRIu64", %d) "
3817 		    "returned %d, expected %d",
3818 		    oldpath, oldsize, newpath,
3819 		    newsize, replacing, error, expected_error);
3820 	}
3821 out:
3822 	mutex_exit(&ztest_vdev_lock);
3823 
3824 	umem_free(oldpath, MAXPATHLEN);
3825 	umem_free(newpath, MAXPATHLEN);
3826 }
3827 
3828 void
3829 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3830 {
3831 	(void) zd, (void) id;
3832 	spa_t *spa = ztest_spa;
3833 	vdev_t *vd;
3834 	uint64_t guid;
3835 	int error;
3836 
3837 	mutex_enter(&ztest_vdev_lock);
3838 
3839 	if (ztest_device_removal_active) {
3840 		mutex_exit(&ztest_vdev_lock);
3841 		return;
3842 	}
3843 
3844 	/*
3845 	 * Remove a random top-level vdev and wait for removal to finish.
3846 	 */
3847 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3848 	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3849 	guid = vd->vdev_guid;
3850 	spa_config_exit(spa, SCL_VDEV, FTAG);
3851 
3852 	error = spa_vdev_remove(spa, guid, B_FALSE);
3853 	if (error == 0) {
3854 		ztest_device_removal_active = B_TRUE;
3855 		mutex_exit(&ztest_vdev_lock);
3856 
3857 		/*
3858 		 * spa->spa_vdev_removal is created in a sync task that
3859 		 * is initiated via dsl_sync_task_nowait(). Since the
3860 		 * task may not run before spa_vdev_remove() returns, we
3861 		 * must wait at least 1 txg to ensure that the removal
3862 		 * struct has been created.
3863 		 */
3864 		txg_wait_synced(spa_get_dsl(spa), 0);
3865 
3866 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
3867 			txg_wait_synced(spa_get_dsl(spa), 0);
3868 	} else {
3869 		mutex_exit(&ztest_vdev_lock);
3870 		return;
3871 	}
3872 
3873 	/*
3874 	 * The pool needs to be scrubbed after completing device removal.
3875 	 * Failure to do so may result in checksum errors due to the
3876 	 * strategy employed by ztest_fault_inject() when selecting which
3877 	 * offset are redundant and can be damaged.
3878 	 */
3879 	error = spa_scan(spa, POOL_SCAN_SCRUB);
3880 	if (error == 0) {
3881 		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
3882 			txg_wait_synced(spa_get_dsl(spa), 0);
3883 	}
3884 
3885 	mutex_enter(&ztest_vdev_lock);
3886 	ztest_device_removal_active = B_FALSE;
3887 	mutex_exit(&ztest_vdev_lock);
3888 }
3889 
3890 /*
3891  * Callback function which expands the physical size of the vdev.
3892  */
3893 static vdev_t *
3894 grow_vdev(vdev_t *vd, void *arg)
3895 {
3896 	spa_t *spa __maybe_unused = vd->vdev_spa;
3897 	size_t *newsize = arg;
3898 	size_t fsize;
3899 	int fd;
3900 
3901 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
3902 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3903 
3904 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3905 		return (vd);
3906 
3907 	fsize = lseek(fd, 0, SEEK_END);
3908 	VERIFY0(ftruncate(fd, *newsize));
3909 
3910 	if (ztest_opts.zo_verbose >= 6) {
3911 		(void) printf("%s grew from %lu to %lu bytes\n",
3912 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3913 	}
3914 	(void) close(fd);
3915 	return (NULL);
3916 }
3917 
3918 /*
3919  * Callback function which expands a given vdev by calling vdev_online().
3920  */
3921 static vdev_t *
3922 online_vdev(vdev_t *vd, void *arg)
3923 {
3924 	(void) arg;
3925 	spa_t *spa = vd->vdev_spa;
3926 	vdev_t *tvd = vd->vdev_top;
3927 	uint64_t guid = vd->vdev_guid;
3928 	uint64_t generation = spa->spa_config_generation + 1;
3929 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3930 	int error;
3931 
3932 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
3933 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3934 
3935 	/* Calling vdev_online will initialize the new metaslabs */
3936 	spa_config_exit(spa, SCL_STATE, spa);
3937 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3938 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3939 
3940 	/*
3941 	 * If vdev_online returned an error or the underlying vdev_open
3942 	 * failed then we abort the expand. The only way to know that
3943 	 * vdev_open fails is by checking the returned newstate.
3944 	 */
3945 	if (error || newstate != VDEV_STATE_HEALTHY) {
3946 		if (ztest_opts.zo_verbose >= 5) {
3947 			(void) printf("Unable to expand vdev, state %u, "
3948 			    "error %d\n", newstate, error);
3949 		}
3950 		return (vd);
3951 	}
3952 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3953 
3954 	/*
3955 	 * Since we dropped the lock we need to ensure that we're
3956 	 * still talking to the original vdev. It's possible this
3957 	 * vdev may have been detached/replaced while we were
3958 	 * trying to online it.
3959 	 */
3960 	if (generation != spa->spa_config_generation) {
3961 		if (ztest_opts.zo_verbose >= 5) {
3962 			(void) printf("vdev configuration has changed, "
3963 			    "guid %"PRIu64", state %"PRIu64", "
3964 			    "expected gen %"PRIu64", got gen %"PRIu64"\n",
3965 			    guid,
3966 			    tvd->vdev_state,
3967 			    generation,
3968 			    spa->spa_config_generation);
3969 		}
3970 		return (vd);
3971 	}
3972 	return (NULL);
3973 }
3974 
3975 /*
3976  * Traverse the vdev tree calling the supplied function.
3977  * We continue to walk the tree until we either have walked all
3978  * children or we receive a non-NULL return from the callback.
3979  * If a NULL callback is passed, then we just return back the first
3980  * leaf vdev we encounter.
3981  */
3982 static vdev_t *
3983 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3984 {
3985 	uint_t c;
3986 
3987 	if (vd->vdev_ops->vdev_op_leaf) {
3988 		if (func == NULL)
3989 			return (vd);
3990 		else
3991 			return (func(vd, arg));
3992 	}
3993 
3994 	for (c = 0; c < vd->vdev_children; c++) {
3995 		vdev_t *cvd = vd->vdev_child[c];
3996 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3997 			return (cvd);
3998 	}
3999 	return (NULL);
4000 }
4001 
4002 /*
4003  * Verify that dynamic LUN growth works as expected.
4004  */
4005 void
4006 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
4007 {
4008 	(void) zd, (void) id;
4009 	spa_t *spa = ztest_spa;
4010 	vdev_t *vd, *tvd;
4011 	metaslab_class_t *mc;
4012 	metaslab_group_t *mg;
4013 	size_t psize, newsize;
4014 	uint64_t top;
4015 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
4016 
4017 	mutex_enter(&ztest_checkpoint_lock);
4018 	mutex_enter(&ztest_vdev_lock);
4019 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4020 
4021 	/*
4022 	 * If there is a vdev removal in progress, it could complete while
4023 	 * we are running, in which case we would not be able to verify
4024 	 * that the metaslab_class space increased (because it decreases
4025 	 * when the device removal completes).
4026 	 */
4027 	if (ztest_device_removal_active) {
4028 		spa_config_exit(spa, SCL_STATE, spa);
4029 		mutex_exit(&ztest_vdev_lock);
4030 		mutex_exit(&ztest_checkpoint_lock);
4031 		return;
4032 	}
4033 
4034 	top = ztest_random_vdev_top(spa, B_TRUE);
4035 
4036 	tvd = spa->spa_root_vdev->vdev_child[top];
4037 	mg = tvd->vdev_mg;
4038 	mc = mg->mg_class;
4039 	old_ms_count = tvd->vdev_ms_count;
4040 	old_class_space = metaslab_class_get_space(mc);
4041 
4042 	/*
4043 	 * Determine the size of the first leaf vdev associated with
4044 	 * our top-level device.
4045 	 */
4046 	vd = vdev_walk_tree(tvd, NULL, NULL);
4047 	ASSERT3P(vd, !=, NULL);
4048 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4049 
4050 	psize = vd->vdev_psize;
4051 
4052 	/*
4053 	 * We only try to expand the vdev if it's healthy, less than 4x its
4054 	 * original size, and it has a valid psize.
4055 	 */
4056 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
4057 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
4058 		spa_config_exit(spa, SCL_STATE, spa);
4059 		mutex_exit(&ztest_vdev_lock);
4060 		mutex_exit(&ztest_checkpoint_lock);
4061 		return;
4062 	}
4063 	ASSERT3U(psize, >, 0);
4064 	newsize = psize + MAX(psize / 8, SPA_MAXBLOCKSIZE);
4065 	ASSERT3U(newsize, >, psize);
4066 
4067 	if (ztest_opts.zo_verbose >= 6) {
4068 		(void) printf("Expanding LUN %s from %lu to %lu\n",
4069 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
4070 	}
4071 
4072 	/*
4073 	 * Growing the vdev is a two step process:
4074 	 *	1). expand the physical size (i.e. relabel)
4075 	 *	2). online the vdev to create the new metaslabs
4076 	 */
4077 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
4078 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
4079 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
4080 		if (ztest_opts.zo_verbose >= 5) {
4081 			(void) printf("Could not expand LUN because "
4082 			    "the vdev configuration changed.\n");
4083 		}
4084 		spa_config_exit(spa, SCL_STATE, spa);
4085 		mutex_exit(&ztest_vdev_lock);
4086 		mutex_exit(&ztest_checkpoint_lock);
4087 		return;
4088 	}
4089 
4090 	spa_config_exit(spa, SCL_STATE, spa);
4091 
4092 	/*
4093 	 * Expanding the LUN will update the config asynchronously,
4094 	 * thus we must wait for the async thread to complete any
4095 	 * pending tasks before proceeding.
4096 	 */
4097 	for (;;) {
4098 		boolean_t done;
4099 		mutex_enter(&spa->spa_async_lock);
4100 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
4101 		mutex_exit(&spa->spa_async_lock);
4102 		if (done)
4103 			break;
4104 		txg_wait_synced(spa_get_dsl(spa), 0);
4105 		(void) poll(NULL, 0, 100);
4106 	}
4107 
4108 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4109 
4110 	tvd = spa->spa_root_vdev->vdev_child[top];
4111 	new_ms_count = tvd->vdev_ms_count;
4112 	new_class_space = metaslab_class_get_space(mc);
4113 
4114 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
4115 		if (ztest_opts.zo_verbose >= 5) {
4116 			(void) printf("Could not verify LUN expansion due to "
4117 			    "intervening vdev offline or remove.\n");
4118 		}
4119 		spa_config_exit(spa, SCL_STATE, spa);
4120 		mutex_exit(&ztest_vdev_lock);
4121 		mutex_exit(&ztest_checkpoint_lock);
4122 		return;
4123 	}
4124 
4125 	/*
4126 	 * Make sure we were able to grow the vdev.
4127 	 */
4128 	if (new_ms_count <= old_ms_count) {
4129 		fatal(B_FALSE,
4130 		    "LUN expansion failed: ms_count %"PRIu64" < %"PRIu64"\n",
4131 		    old_ms_count, new_ms_count);
4132 	}
4133 
4134 	/*
4135 	 * Make sure we were able to grow the pool.
4136 	 */
4137 	if (new_class_space <= old_class_space) {
4138 		fatal(B_FALSE,
4139 		    "LUN expansion failed: class_space %"PRIu64" < %"PRIu64"\n",
4140 		    old_class_space, new_class_space);
4141 	}
4142 
4143 	if (ztest_opts.zo_verbose >= 5) {
4144 		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
4145 
4146 		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
4147 		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
4148 		(void) printf("%s grew from %s to %s\n",
4149 		    spa->spa_name, oldnumbuf, newnumbuf);
4150 	}
4151 
4152 	spa_config_exit(spa, SCL_STATE, spa);
4153 	mutex_exit(&ztest_vdev_lock);
4154 	mutex_exit(&ztest_checkpoint_lock);
4155 }
4156 
4157 /*
4158  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
4159  */
4160 static void
4161 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
4162 {
4163 	(void) arg, (void) cr;
4164 
4165 	/*
4166 	 * Create the objects common to all ztest datasets.
4167 	 */
4168 	VERIFY0(zap_create_claim(os, ZTEST_DIROBJ,
4169 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx));
4170 }
4171 
4172 static int
4173 ztest_dataset_create(char *dsname)
4174 {
4175 	int err;
4176 	uint64_t rand;
4177 	dsl_crypto_params_t *dcp = NULL;
4178 
4179 	/*
4180 	 * 50% of the time, we create encrypted datasets
4181 	 * using a random cipher suite and a hard-coded
4182 	 * wrapping key.
4183 	 */
4184 	rand = ztest_random(2);
4185 	if (rand != 0) {
4186 		nvlist_t *crypto_args = fnvlist_alloc();
4187 		nvlist_t *props = fnvlist_alloc();
4188 
4189 		/* slight bias towards the default cipher suite */
4190 		rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
4191 		if (rand < ZIO_CRYPT_AES_128_CCM)
4192 			rand = ZIO_CRYPT_ON;
4193 
4194 		fnvlist_add_uint64(props,
4195 		    zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
4196 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
4197 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
4198 
4199 		/*
4200 		 * These parameters aren't really used by the kernel. They
4201 		 * are simply stored so that userspace knows how to load
4202 		 * the wrapping key.
4203 		 */
4204 		fnvlist_add_uint64(props,
4205 		    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
4206 		fnvlist_add_string(props,
4207 		    zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
4208 		fnvlist_add_uint64(props,
4209 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
4210 		fnvlist_add_uint64(props,
4211 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
4212 
4213 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
4214 		    crypto_args, &dcp));
4215 
4216 		/*
4217 		 * Cycle through all available encryption implementations
4218 		 * to verify interoperability.
4219 		 */
4220 		VERIFY0(gcm_impl_set("cycle"));
4221 		VERIFY0(aes_impl_set("cycle"));
4222 
4223 		fnvlist_free(crypto_args);
4224 		fnvlist_free(props);
4225 	}
4226 
4227 	err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
4228 	    ztest_objset_create_cb, NULL);
4229 	dsl_crypto_params_free(dcp, !!err);
4230 
4231 	rand = ztest_random(100);
4232 	if (err || rand < 80)
4233 		return (err);
4234 
4235 	if (ztest_opts.zo_verbose >= 5)
4236 		(void) printf("Setting dataset %s to sync always\n", dsname);
4237 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
4238 	    ZFS_SYNC_ALWAYS, B_FALSE));
4239 }
4240 
4241 static int
4242 ztest_objset_destroy_cb(const char *name, void *arg)
4243 {
4244 	(void) arg;
4245 	objset_t *os;
4246 	dmu_object_info_t doi;
4247 	int error;
4248 
4249 	/*
4250 	 * Verify that the dataset contains a directory object.
4251 	 */
4252 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
4253 	    B_TRUE, FTAG, &os));
4254 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
4255 	if (error != ENOENT) {
4256 		/* We could have crashed in the middle of destroying it */
4257 		ASSERT0(error);
4258 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
4259 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
4260 	}
4261 	dmu_objset_disown(os, B_TRUE, FTAG);
4262 
4263 	/*
4264 	 * Destroy the dataset.
4265 	 */
4266 	if (strchr(name, '@') != NULL) {
4267 		error = dsl_destroy_snapshot(name, B_TRUE);
4268 		if (error != ECHRNG) {
4269 			/*
4270 			 * The program was executed, but encountered a runtime
4271 			 * error, such as insufficient slop, or a hold on the
4272 			 * dataset.
4273 			 */
4274 			ASSERT0(error);
4275 		}
4276 	} else {
4277 		error = dsl_destroy_head(name);
4278 		if (error == ENOSPC) {
4279 			/* There could be checkpoint or insufficient slop */
4280 			ztest_record_enospc(FTAG);
4281 		} else if (error != EBUSY) {
4282 			/* There could be a hold on this dataset */
4283 			ASSERT0(error);
4284 		}
4285 	}
4286 	return (0);
4287 }
4288 
4289 static boolean_t
4290 ztest_snapshot_create(char *osname, uint64_t id)
4291 {
4292 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4293 	int error;
4294 
4295 	(void) snprintf(snapname, sizeof (snapname), "%"PRIu64"", id);
4296 
4297 	error = dmu_objset_snapshot_one(osname, snapname);
4298 	if (error == ENOSPC) {
4299 		ztest_record_enospc(FTAG);
4300 		return (B_FALSE);
4301 	}
4302 	if (error != 0 && error != EEXIST && error != ECHRNG) {
4303 		fatal(B_FALSE, "ztest_snapshot_create(%s@%s) = %d", osname,
4304 		    snapname, error);
4305 	}
4306 	return (B_TRUE);
4307 }
4308 
4309 static boolean_t
4310 ztest_snapshot_destroy(char *osname, uint64_t id)
4311 {
4312 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4313 	int error;
4314 
4315 	(void) snprintf(snapname, sizeof (snapname), "%s@%"PRIu64"",
4316 	    osname, id);
4317 
4318 	error = dsl_destroy_snapshot(snapname, B_FALSE);
4319 	if (error != 0 && error != ENOENT && error != ECHRNG)
4320 		fatal(B_FALSE, "ztest_snapshot_destroy(%s) = %d",
4321 		    snapname, error);
4322 	return (B_TRUE);
4323 }
4324 
4325 void
4326 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
4327 {
4328 	(void) zd;
4329 	ztest_ds_t *zdtmp;
4330 	int iters;
4331 	int error;
4332 	objset_t *os, *os2;
4333 	char name[ZFS_MAX_DATASET_NAME_LEN];
4334 	zilog_t *zilog;
4335 	int i;
4336 
4337 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
4338 
4339 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4340 
4341 	(void) snprintf(name, sizeof (name), "%s/temp_%"PRIu64"",
4342 	    ztest_opts.zo_pool, id);
4343 
4344 	/*
4345 	 * If this dataset exists from a previous run, process its replay log
4346 	 * half of the time.  If we don't replay it, then dsl_destroy_head()
4347 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
4348 	 */
4349 	if (ztest_random(2) == 0 &&
4350 	    ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
4351 	    B_TRUE, FTAG, &os) == 0) {
4352 		ztest_zd_init(zdtmp, NULL, os);
4353 		zil_replay(os, zdtmp, ztest_replay_vector);
4354 		ztest_zd_fini(zdtmp);
4355 		dmu_objset_disown(os, B_TRUE, FTAG);
4356 	}
4357 
4358 	/*
4359 	 * There may be an old instance of the dataset we're about to
4360 	 * create lying around from a previous run.  If so, destroy it
4361 	 * and all of its snapshots.
4362 	 */
4363 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
4364 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
4365 
4366 	/*
4367 	 * Verify that the destroyed dataset is no longer in the namespace.
4368 	 * It may still be present if the destroy above fails with ENOSPC.
4369 	 */
4370 	error = ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, B_TRUE,
4371 	    FTAG, &os);
4372 	if (error == 0) {
4373 		dmu_objset_disown(os, B_TRUE, FTAG);
4374 		ztest_record_enospc(FTAG);
4375 		goto out;
4376 	}
4377 	VERIFY3U(ENOENT, ==, error);
4378 
4379 	/*
4380 	 * Verify that we can create a new dataset.
4381 	 */
4382 	error = ztest_dataset_create(name);
4383 	if (error) {
4384 		if (error == ENOSPC) {
4385 			ztest_record_enospc(FTAG);
4386 			goto out;
4387 		}
4388 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", name, error);
4389 	}
4390 
4391 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
4392 	    FTAG, &os));
4393 
4394 	ztest_zd_init(zdtmp, NULL, os);
4395 
4396 	/*
4397 	 * Open the intent log for it.
4398 	 */
4399 	zilog = zil_open(os, ztest_get_data, NULL);
4400 
4401 	/*
4402 	 * Put some objects in there, do a little I/O to them,
4403 	 * and randomly take a couple of snapshots along the way.
4404 	 */
4405 	iters = ztest_random(5);
4406 	for (i = 0; i < iters; i++) {
4407 		ztest_dmu_object_alloc_free(zdtmp, id);
4408 		if (ztest_random(iters) == 0)
4409 			(void) ztest_snapshot_create(name, i);
4410 	}
4411 
4412 	/*
4413 	 * Verify that we cannot create an existing dataset.
4414 	 */
4415 	VERIFY3U(EEXIST, ==,
4416 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
4417 
4418 	/*
4419 	 * Verify that we can hold an objset that is also owned.
4420 	 */
4421 	VERIFY0(dmu_objset_hold(name, FTAG, &os2));
4422 	dmu_objset_rele(os2, FTAG);
4423 
4424 	/*
4425 	 * Verify that we cannot own an objset that is already owned.
4426 	 */
4427 	VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
4428 	    B_FALSE, B_TRUE, FTAG, &os2));
4429 
4430 	zil_close(zilog);
4431 	dmu_objset_disown(os, B_TRUE, FTAG);
4432 	ztest_zd_fini(zdtmp);
4433 out:
4434 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4435 
4436 	umem_free(zdtmp, sizeof (ztest_ds_t));
4437 }
4438 
4439 /*
4440  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
4441  */
4442 void
4443 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
4444 {
4445 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4446 	(void) ztest_snapshot_destroy(zd->zd_name, id);
4447 	(void) ztest_snapshot_create(zd->zd_name, id);
4448 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4449 }
4450 
4451 /*
4452  * Cleanup non-standard snapshots and clones.
4453  */
4454 static void
4455 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
4456 {
4457 	char *snap1name;
4458 	char *clone1name;
4459 	char *snap2name;
4460 	char *clone2name;
4461 	char *snap3name;
4462 	int error;
4463 
4464 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4465 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4466 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4467 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4468 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4469 
4470 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4471 	    osname, id);
4472 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4473 	    osname, id);
4474 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4475 	    clone1name, id);
4476 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4477 	    osname, id);
4478 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4479 	    clone1name, id);
4480 
4481 	error = dsl_destroy_head(clone2name);
4482 	if (error && error != ENOENT)
4483 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone2name, error);
4484 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
4485 	if (error && error != ENOENT)
4486 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4487 		    snap3name, error);
4488 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
4489 	if (error && error != ENOENT)
4490 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4491 		    snap2name, error);
4492 	error = dsl_destroy_head(clone1name);
4493 	if (error && error != ENOENT)
4494 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone1name, error);
4495 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
4496 	if (error && error != ENOENT)
4497 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4498 		    snap1name, error);
4499 
4500 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4501 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4502 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4503 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4504 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4505 }
4506 
4507 /*
4508  * Verify dsl_dataset_promote handles EBUSY
4509  */
4510 void
4511 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
4512 {
4513 	objset_t *os;
4514 	char *snap1name;
4515 	char *clone1name;
4516 	char *snap2name;
4517 	char *clone2name;
4518 	char *snap3name;
4519 	char *osname = zd->zd_name;
4520 	int error;
4521 
4522 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4523 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4524 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4525 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4526 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4527 
4528 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4529 
4530 	ztest_dsl_dataset_cleanup(osname, id);
4531 
4532 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4533 	    osname, id);
4534 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4535 	    osname, id);
4536 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4537 	    clone1name, id);
4538 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4539 	    osname, id);
4540 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4541 	    clone1name, id);
4542 
4543 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
4544 	if (error && error != EEXIST) {
4545 		if (error == ENOSPC) {
4546 			ztest_record_enospc(FTAG);
4547 			goto out;
4548 		}
4549 		fatal(B_FALSE, "dmu_take_snapshot(%s) = %d", snap1name, error);
4550 	}
4551 
4552 	error = dmu_objset_clone(clone1name, snap1name);
4553 	if (error) {
4554 		if (error == ENOSPC) {
4555 			ztest_record_enospc(FTAG);
4556 			goto out;
4557 		}
4558 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone1name, error);
4559 	}
4560 
4561 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
4562 	if (error && error != EEXIST) {
4563 		if (error == ENOSPC) {
4564 			ztest_record_enospc(FTAG);
4565 			goto out;
4566 		}
4567 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap2name, error);
4568 	}
4569 
4570 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
4571 	if (error && error != EEXIST) {
4572 		if (error == ENOSPC) {
4573 			ztest_record_enospc(FTAG);
4574 			goto out;
4575 		}
4576 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap3name, error);
4577 	}
4578 
4579 	error = dmu_objset_clone(clone2name, snap3name);
4580 	if (error) {
4581 		if (error == ENOSPC) {
4582 			ztest_record_enospc(FTAG);
4583 			goto out;
4584 		}
4585 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone2name, error);
4586 	}
4587 
4588 	error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
4589 	    FTAG, &os);
4590 	if (error)
4591 		fatal(B_FALSE, "dmu_objset_own(%s) = %d", snap2name, error);
4592 	error = dsl_dataset_promote(clone2name, NULL);
4593 	if (error == ENOSPC) {
4594 		dmu_objset_disown(os, B_TRUE, FTAG);
4595 		ztest_record_enospc(FTAG);
4596 		goto out;
4597 	}
4598 	if (error != EBUSY)
4599 		fatal(B_FALSE, "dsl_dataset_promote(%s), %d, not EBUSY",
4600 		    clone2name, error);
4601 	dmu_objset_disown(os, B_TRUE, FTAG);
4602 
4603 out:
4604 	ztest_dsl_dataset_cleanup(osname, id);
4605 
4606 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4607 
4608 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4609 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4610 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4611 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4612 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4613 }
4614 
4615 #undef OD_ARRAY_SIZE
4616 #define	OD_ARRAY_SIZE	4
4617 
4618 /*
4619  * Verify that dmu_object_{alloc,free} work as expected.
4620  */
4621 void
4622 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4623 {
4624 	ztest_od_t *od;
4625 	int batchsize;
4626 	int size;
4627 	int b;
4628 
4629 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4630 	od = umem_alloc(size, UMEM_NOFAIL);
4631 	batchsize = OD_ARRAY_SIZE;
4632 
4633 	for (b = 0; b < batchsize; b++)
4634 		ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
4635 		    0, 0, 0);
4636 
4637 	/*
4638 	 * Destroy the previous batch of objects, create a new batch,
4639 	 * and do some I/O on the new objects.
4640 	 */
4641 	if (ztest_object_init(zd, od, size, B_TRUE) != 0) {
4642 		zd->zd_od = NULL;
4643 		umem_free(od, size);
4644 		return;
4645 	}
4646 
4647 	while (ztest_random(4 * batchsize) != 0)
4648 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
4649 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4650 
4651 	umem_free(od, size);
4652 }
4653 
4654 /*
4655  * Rewind the global allocator to verify object allocation backfilling.
4656  */
4657 void
4658 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
4659 {
4660 	(void) id;
4661 	objset_t *os = zd->zd_os;
4662 	uint_t dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
4663 	uint64_t object;
4664 
4665 	/*
4666 	 * Rewind the global allocator randomly back to a lower object number
4667 	 * to force backfilling and reclamation of recently freed dnodes.
4668 	 */
4669 	mutex_enter(&os->os_obj_lock);
4670 	object = ztest_random(os->os_obj_next_chunk);
4671 	os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk);
4672 	mutex_exit(&os->os_obj_lock);
4673 }
4674 
4675 #undef OD_ARRAY_SIZE
4676 #define	OD_ARRAY_SIZE	2
4677 
4678 /*
4679  * Verify that dmu_{read,write} work as expected.
4680  */
4681 void
4682 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
4683 {
4684 	int size;
4685 	ztest_od_t *od;
4686 
4687 	objset_t *os = zd->zd_os;
4688 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4689 	od = umem_alloc(size, UMEM_NOFAIL);
4690 	dmu_tx_t *tx;
4691 	int freeit, error;
4692 	uint64_t i, n, s, txg;
4693 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
4694 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4695 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
4696 	uint64_t regions = 997;
4697 	uint64_t stride = 123456789ULL;
4698 	uint64_t width = 40;
4699 	int free_percent = 5;
4700 
4701 	/*
4702 	 * This test uses two objects, packobj and bigobj, that are always
4703 	 * updated together (i.e. in the same tx) so that their contents are
4704 	 * in sync and can be compared.  Their contents relate to each other
4705 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4706 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4707 	 * for any index n, there are three bufwads that should be identical:
4708 	 *
4709 	 *	packobj, at offset n * sizeof (bufwad_t)
4710 	 *	bigobj, at the head of the nth chunk
4711 	 *	bigobj, at the tail of the nth chunk
4712 	 *
4713 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
4714 	 * and it doesn't have any relation to the object blocksize.
4715 	 * The only requirement is that it can hold at least two bufwads.
4716 	 *
4717 	 * Normally, we write the bufwad to each of these locations.
4718 	 * However, free_percent of the time we instead write zeroes to
4719 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
4720 	 * bigobj to packobj, we can verify that the DMU is correctly
4721 	 * tracking which parts of an object are allocated and free,
4722 	 * and that the contents of the allocated blocks are correct.
4723 	 */
4724 
4725 	/*
4726 	 * Read the directory info.  If it's the first time, set things up.
4727 	 */
4728 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
4729 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4730 	    chunksize);
4731 
4732 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4733 		umem_free(od, size);
4734 		return;
4735 	}
4736 
4737 	bigobj = od[0].od_object;
4738 	packobj = od[1].od_object;
4739 	chunksize = od[0].od_gen;
4740 	ASSERT3U(chunksize, ==, od[1].od_gen);
4741 
4742 	/*
4743 	 * Prefetch a random chunk of the big object.
4744 	 * Our aim here is to get some async reads in flight
4745 	 * for blocks that we may free below; the DMU should
4746 	 * handle this race correctly.
4747 	 */
4748 	n = ztest_random(regions) * stride + ztest_random(width);
4749 	s = 1 + ztest_random(2 * width - 1);
4750 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4751 	    ZIO_PRIORITY_SYNC_READ);
4752 
4753 	/*
4754 	 * Pick a random index and compute the offsets into packobj and bigobj.
4755 	 */
4756 	n = ztest_random(regions) * stride + ztest_random(width);
4757 	s = 1 + ztest_random(width - 1);
4758 
4759 	packoff = n * sizeof (bufwad_t);
4760 	packsize = s * sizeof (bufwad_t);
4761 
4762 	bigoff = n * chunksize;
4763 	bigsize = s * chunksize;
4764 
4765 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4766 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4767 
4768 	/*
4769 	 * free_percent of the time, free a range of bigobj rather than
4770 	 * overwriting it.
4771 	 */
4772 	freeit = (ztest_random(100) < free_percent);
4773 
4774 	/*
4775 	 * Read the current contents of our objects.
4776 	 */
4777 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
4778 	    DMU_READ_PREFETCH);
4779 	ASSERT0(error);
4780 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4781 	    DMU_READ_PREFETCH);
4782 	ASSERT0(error);
4783 
4784 	/*
4785 	 * Get a tx for the mods to both packobj and bigobj.
4786 	 */
4787 	tx = dmu_tx_create(os);
4788 
4789 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
4790 
4791 	if (freeit)
4792 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4793 	else
4794 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4795 
4796 	/* This accounts for setting the checksum/compression. */
4797 	dmu_tx_hold_bonus(tx, bigobj);
4798 
4799 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4800 	if (txg == 0) {
4801 		umem_free(packbuf, packsize);
4802 		umem_free(bigbuf, bigsize);
4803 		umem_free(od, size);
4804 		return;
4805 	}
4806 
4807 	enum zio_checksum cksum;
4808 	do {
4809 		cksum = (enum zio_checksum)
4810 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4811 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4812 	dmu_object_set_checksum(os, bigobj, cksum, tx);
4813 
4814 	enum zio_compress comp;
4815 	do {
4816 		comp = (enum zio_compress)
4817 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4818 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4819 	dmu_object_set_compress(os, bigobj, comp, tx);
4820 
4821 	/*
4822 	 * For each index from n to n + s, verify that the existing bufwad
4823 	 * in packobj matches the bufwads at the head and tail of the
4824 	 * corresponding chunk in bigobj.  Then update all three bufwads
4825 	 * with the new values we want to write out.
4826 	 */
4827 	for (i = 0; i < s; i++) {
4828 		/* LINTED */
4829 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4830 		/* LINTED */
4831 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4832 		/* LINTED */
4833 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4834 
4835 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
4836 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
4837 
4838 		if (pack->bw_txg > txg)
4839 			fatal(B_FALSE,
4840 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
4841 			    pack->bw_txg, txg);
4842 
4843 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4844 			fatal(B_FALSE, "wrong index: "
4845 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
4846 			    pack->bw_index, n, i);
4847 
4848 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4849 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
4850 			    pack, bigH);
4851 
4852 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4853 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
4854 			    pack, bigT);
4855 
4856 		if (freeit) {
4857 			memset(pack, 0, sizeof (bufwad_t));
4858 		} else {
4859 			pack->bw_index = n + i;
4860 			pack->bw_txg = txg;
4861 			pack->bw_data = 1 + ztest_random(-2ULL);
4862 		}
4863 		*bigH = *pack;
4864 		*bigT = *pack;
4865 	}
4866 
4867 	/*
4868 	 * We've verified all the old bufwads, and made new ones.
4869 	 * Now write them out.
4870 	 */
4871 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4872 
4873 	if (freeit) {
4874 		if (ztest_opts.zo_verbose >= 7) {
4875 			(void) printf("freeing offset %"PRIx64" size %"PRIx64""
4876 			    " txg %"PRIx64"\n",
4877 			    bigoff, bigsize, txg);
4878 		}
4879 		VERIFY0(dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4880 	} else {
4881 		if (ztest_opts.zo_verbose >= 7) {
4882 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
4883 			    " txg %"PRIx64"\n",
4884 			    bigoff, bigsize, txg);
4885 		}
4886 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4887 	}
4888 
4889 	dmu_tx_commit(tx);
4890 
4891 	/*
4892 	 * Sanity check the stuff we just wrote.
4893 	 */
4894 	{
4895 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4896 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4897 
4898 		VERIFY0(dmu_read(os, packobj, packoff,
4899 		    packsize, packcheck, DMU_READ_PREFETCH));
4900 		VERIFY0(dmu_read(os, bigobj, bigoff,
4901 		    bigsize, bigcheck, DMU_READ_PREFETCH));
4902 
4903 		ASSERT0(memcmp(packbuf, packcheck, packsize));
4904 		ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
4905 
4906 		umem_free(packcheck, packsize);
4907 		umem_free(bigcheck, bigsize);
4908 	}
4909 
4910 	umem_free(packbuf, packsize);
4911 	umem_free(bigbuf, bigsize);
4912 	umem_free(od, size);
4913 }
4914 
4915 static void
4916 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4917     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4918 {
4919 	uint64_t i;
4920 	bufwad_t *pack;
4921 	bufwad_t *bigH;
4922 	bufwad_t *bigT;
4923 
4924 	/*
4925 	 * For each index from n to n + s, verify that the existing bufwad
4926 	 * in packobj matches the bufwads at the head and tail of the
4927 	 * corresponding chunk in bigobj.  Then update all three bufwads
4928 	 * with the new values we want to write out.
4929 	 */
4930 	for (i = 0; i < s; i++) {
4931 		/* LINTED */
4932 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4933 		/* LINTED */
4934 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4935 		/* LINTED */
4936 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4937 
4938 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
4939 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
4940 
4941 		if (pack->bw_txg > txg)
4942 			fatal(B_FALSE,
4943 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
4944 			    pack->bw_txg, txg);
4945 
4946 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4947 			fatal(B_FALSE, "wrong index: "
4948 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
4949 			    pack->bw_index, n, i);
4950 
4951 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4952 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
4953 			    pack, bigH);
4954 
4955 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4956 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
4957 			    pack, bigT);
4958 
4959 		pack->bw_index = n + i;
4960 		pack->bw_txg = txg;
4961 		pack->bw_data = 1 + ztest_random(-2ULL);
4962 
4963 		*bigH = *pack;
4964 		*bigT = *pack;
4965 	}
4966 }
4967 
4968 #undef OD_ARRAY_SIZE
4969 #define	OD_ARRAY_SIZE	2
4970 
4971 void
4972 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4973 {
4974 	objset_t *os = zd->zd_os;
4975 	ztest_od_t *od;
4976 	dmu_tx_t *tx;
4977 	uint64_t i;
4978 	int error;
4979 	int size;
4980 	uint64_t n, s, txg;
4981 	bufwad_t *packbuf, *bigbuf;
4982 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4983 	uint64_t blocksize = ztest_random_blocksize();
4984 	uint64_t chunksize = blocksize;
4985 	uint64_t regions = 997;
4986 	uint64_t stride = 123456789ULL;
4987 	uint64_t width = 9;
4988 	dmu_buf_t *bonus_db;
4989 	arc_buf_t **bigbuf_arcbufs;
4990 	dmu_object_info_t doi;
4991 
4992 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4993 	od = umem_alloc(size, UMEM_NOFAIL);
4994 
4995 	/*
4996 	 * This test uses two objects, packobj and bigobj, that are always
4997 	 * updated together (i.e. in the same tx) so that their contents are
4998 	 * in sync and can be compared.  Their contents relate to each other
4999 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
5000 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
5001 	 * for any index n, there are three bufwads that should be identical:
5002 	 *
5003 	 *	packobj, at offset n * sizeof (bufwad_t)
5004 	 *	bigobj, at the head of the nth chunk
5005 	 *	bigobj, at the tail of the nth chunk
5006 	 *
5007 	 * The chunk size is set equal to bigobj block size so that
5008 	 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
5009 	 */
5010 
5011 	/*
5012 	 * Read the directory info.  If it's the first time, set things up.
5013 	 */
5014 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5015 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
5016 	    chunksize);
5017 
5018 
5019 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
5020 		umem_free(od, size);
5021 		return;
5022 	}
5023 
5024 	bigobj = od[0].od_object;
5025 	packobj = od[1].od_object;
5026 	blocksize = od[0].od_blocksize;
5027 	chunksize = blocksize;
5028 	ASSERT3U(chunksize, ==, od[1].od_gen);
5029 
5030 	VERIFY0(dmu_object_info(os, bigobj, &doi));
5031 	VERIFY(ISP2(doi.doi_data_block_size));
5032 	VERIFY3U(chunksize, ==, doi.doi_data_block_size);
5033 	VERIFY3U(chunksize, >=, 2 * sizeof (bufwad_t));
5034 
5035 	/*
5036 	 * Pick a random index and compute the offsets into packobj and bigobj.
5037 	 */
5038 	n = ztest_random(regions) * stride + ztest_random(width);
5039 	s = 1 + ztest_random(width - 1);
5040 
5041 	packoff = n * sizeof (bufwad_t);
5042 	packsize = s * sizeof (bufwad_t);
5043 
5044 	bigoff = n * chunksize;
5045 	bigsize = s * chunksize;
5046 
5047 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
5048 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
5049 
5050 	VERIFY0(dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
5051 
5052 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
5053 
5054 	/*
5055 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
5056 	 * Iteration 1 test zcopy to already referenced dbufs.
5057 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
5058 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
5059 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
5060 	 * Iteration 5 test zcopy when it can't be done.
5061 	 * Iteration 6 one more zcopy write.
5062 	 */
5063 	for (i = 0; i < 7; i++) {
5064 		uint64_t j;
5065 		uint64_t off;
5066 
5067 		/*
5068 		 * In iteration 5 (i == 5) use arcbufs
5069 		 * that don't match bigobj blksz to test
5070 		 * dmu_assign_arcbuf_by_dbuf() when it can't directly
5071 		 * assign an arcbuf to a dbuf.
5072 		 */
5073 		for (j = 0; j < s; j++) {
5074 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5075 				bigbuf_arcbufs[j] =
5076 				    dmu_request_arcbuf(bonus_db, chunksize);
5077 			} else {
5078 				bigbuf_arcbufs[2 * j] =
5079 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5080 				bigbuf_arcbufs[2 * j + 1] =
5081 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5082 			}
5083 		}
5084 
5085 		/*
5086 		 * Get a tx for the mods to both packobj and bigobj.
5087 		 */
5088 		tx = dmu_tx_create(os);
5089 
5090 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
5091 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
5092 
5093 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5094 		if (txg == 0) {
5095 			umem_free(packbuf, packsize);
5096 			umem_free(bigbuf, bigsize);
5097 			for (j = 0; j < s; j++) {
5098 				if (i != 5 ||
5099 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
5100 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
5101 				} else {
5102 					dmu_return_arcbuf(
5103 					    bigbuf_arcbufs[2 * j]);
5104 					dmu_return_arcbuf(
5105 					    bigbuf_arcbufs[2 * j + 1]);
5106 				}
5107 			}
5108 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5109 			umem_free(od, size);
5110 			dmu_buf_rele(bonus_db, FTAG);
5111 			return;
5112 		}
5113 
5114 		/*
5115 		 * 50% of the time don't read objects in the 1st iteration to
5116 		 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
5117 		 * no existing dbufs for the specified offsets.
5118 		 */
5119 		if (i != 0 || ztest_random(2) != 0) {
5120 			error = dmu_read(os, packobj, packoff,
5121 			    packsize, packbuf, DMU_READ_PREFETCH);
5122 			ASSERT0(error);
5123 			error = dmu_read(os, bigobj, bigoff, bigsize,
5124 			    bigbuf, DMU_READ_PREFETCH);
5125 			ASSERT0(error);
5126 		}
5127 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
5128 		    n, chunksize, txg);
5129 
5130 		/*
5131 		 * We've verified all the old bufwads, and made new ones.
5132 		 * Now write them out.
5133 		 */
5134 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
5135 		if (ztest_opts.zo_verbose >= 7) {
5136 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
5137 			    " txg %"PRIx64"\n",
5138 			    bigoff, bigsize, txg);
5139 		}
5140 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
5141 			dmu_buf_t *dbt;
5142 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5143 				memcpy(bigbuf_arcbufs[j]->b_data,
5144 				    (caddr_t)bigbuf + (off - bigoff),
5145 				    chunksize);
5146 			} else {
5147 				memcpy(bigbuf_arcbufs[2 * j]->b_data,
5148 				    (caddr_t)bigbuf + (off - bigoff),
5149 				    chunksize / 2);
5150 				memcpy(bigbuf_arcbufs[2 * j + 1]->b_data,
5151 				    (caddr_t)bigbuf + (off - bigoff) +
5152 				    chunksize / 2,
5153 				    chunksize / 2);
5154 			}
5155 
5156 			if (i == 1) {
5157 				VERIFY(dmu_buf_hold(os, bigobj, off,
5158 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
5159 			}
5160 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5161 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5162 				    off, bigbuf_arcbufs[j], tx));
5163 			} else {
5164 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5165 				    off, bigbuf_arcbufs[2 * j], tx));
5166 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5167 				    off + chunksize / 2,
5168 				    bigbuf_arcbufs[2 * j + 1], tx));
5169 			}
5170 			if (i == 1) {
5171 				dmu_buf_rele(dbt, FTAG);
5172 			}
5173 		}
5174 		dmu_tx_commit(tx);
5175 
5176 		/*
5177 		 * Sanity check the stuff we just wrote.
5178 		 */
5179 		{
5180 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
5181 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
5182 
5183 			VERIFY0(dmu_read(os, packobj, packoff,
5184 			    packsize, packcheck, DMU_READ_PREFETCH));
5185 			VERIFY0(dmu_read(os, bigobj, bigoff,
5186 			    bigsize, bigcheck, DMU_READ_PREFETCH));
5187 
5188 			ASSERT0(memcmp(packbuf, packcheck, packsize));
5189 			ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
5190 
5191 			umem_free(packcheck, packsize);
5192 			umem_free(bigcheck, bigsize);
5193 		}
5194 		if (i == 2) {
5195 			txg_wait_open(dmu_objset_pool(os), 0, B_TRUE);
5196 		} else if (i == 3) {
5197 			txg_wait_synced(dmu_objset_pool(os), 0);
5198 		}
5199 	}
5200 
5201 	dmu_buf_rele(bonus_db, FTAG);
5202 	umem_free(packbuf, packsize);
5203 	umem_free(bigbuf, bigsize);
5204 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5205 	umem_free(od, size);
5206 }
5207 
5208 void
5209 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
5210 {
5211 	(void) id;
5212 	ztest_od_t *od;
5213 
5214 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5215 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
5216 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5217 
5218 	/*
5219 	 * Have multiple threads write to large offsets in an object
5220 	 * to verify that parallel writes to an object -- even to the
5221 	 * same blocks within the object -- doesn't cause any trouble.
5222 	 */
5223 	ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5224 
5225 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
5226 		return;
5227 
5228 	while (ztest_random(10) != 0)
5229 		ztest_io(zd, od->od_object, offset);
5230 
5231 	umem_free(od, sizeof (ztest_od_t));
5232 }
5233 
5234 void
5235 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
5236 {
5237 	ztest_od_t *od;
5238 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
5239 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5240 	uint64_t count = ztest_random(20) + 1;
5241 	uint64_t blocksize = ztest_random_blocksize();
5242 	void *data;
5243 
5244 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5245 
5246 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5247 
5248 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5249 	    !ztest_random(2)) != 0) {
5250 		umem_free(od, sizeof (ztest_od_t));
5251 		return;
5252 	}
5253 
5254 	if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
5255 		umem_free(od, sizeof (ztest_od_t));
5256 		return;
5257 	}
5258 
5259 	ztest_prealloc(zd, od->od_object, offset, count * blocksize);
5260 
5261 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
5262 
5263 	while (ztest_random(count) != 0) {
5264 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
5265 		if (ztest_write(zd, od->od_object, randoff, blocksize,
5266 		    data) != 0)
5267 			break;
5268 		while (ztest_random(4) != 0)
5269 			ztest_io(zd, od->od_object, randoff);
5270 	}
5271 
5272 	umem_free(data, blocksize);
5273 	umem_free(od, sizeof (ztest_od_t));
5274 }
5275 
5276 /*
5277  * Verify that zap_{create,destroy,add,remove,update} work as expected.
5278  */
5279 #define	ZTEST_ZAP_MIN_INTS	1
5280 #define	ZTEST_ZAP_MAX_INTS	4
5281 #define	ZTEST_ZAP_MAX_PROPS	1000
5282 
5283 void
5284 ztest_zap(ztest_ds_t *zd, uint64_t id)
5285 {
5286 	objset_t *os = zd->zd_os;
5287 	ztest_od_t *od;
5288 	uint64_t object;
5289 	uint64_t txg, last_txg;
5290 	uint64_t value[ZTEST_ZAP_MAX_INTS];
5291 	uint64_t zl_ints, zl_intsize, prop;
5292 	int i, ints;
5293 	dmu_tx_t *tx;
5294 	char propname[100], txgname[100];
5295 	int error;
5296 	const char *const hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
5297 
5298 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5299 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5300 
5301 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5302 	    !ztest_random(2)) != 0)
5303 		goto out;
5304 
5305 	object = od->od_object;
5306 
5307 	/*
5308 	 * Generate a known hash collision, and verify that
5309 	 * we can lookup and remove both entries.
5310 	 */
5311 	tx = dmu_tx_create(os);
5312 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5313 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5314 	if (txg == 0)
5315 		goto out;
5316 	for (i = 0; i < 2; i++) {
5317 		value[i] = i;
5318 		VERIFY0(zap_add(os, object, hc[i], sizeof (uint64_t),
5319 		    1, &value[i], tx));
5320 	}
5321 	for (i = 0; i < 2; i++) {
5322 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
5323 		    sizeof (uint64_t), 1, &value[i], tx));
5324 		VERIFY0(
5325 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
5326 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5327 		ASSERT3U(zl_ints, ==, 1);
5328 	}
5329 	for (i = 0; i < 2; i++) {
5330 		VERIFY0(zap_remove(os, object, hc[i], tx));
5331 	}
5332 	dmu_tx_commit(tx);
5333 
5334 	/*
5335 	 * Generate a bunch of random entries.
5336 	 */
5337 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
5338 
5339 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5340 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5341 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5342 	memset(value, 0, sizeof (value));
5343 	last_txg = 0;
5344 
5345 	/*
5346 	 * If these zap entries already exist, validate their contents.
5347 	 */
5348 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5349 	if (error == 0) {
5350 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5351 		ASSERT3U(zl_ints, ==, 1);
5352 
5353 		VERIFY0(zap_lookup(os, object, txgname, zl_intsize,
5354 		    zl_ints, &last_txg));
5355 
5356 		VERIFY0(zap_length(os, object, propname, &zl_intsize,
5357 		    &zl_ints));
5358 
5359 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5360 		ASSERT3U(zl_ints, ==, ints);
5361 
5362 		VERIFY0(zap_lookup(os, object, propname, zl_intsize,
5363 		    zl_ints, value));
5364 
5365 		for (i = 0; i < ints; i++) {
5366 			ASSERT3U(value[i], ==, last_txg + object + i);
5367 		}
5368 	} else {
5369 		ASSERT3U(error, ==, ENOENT);
5370 	}
5371 
5372 	/*
5373 	 * Atomically update two entries in our zap object.
5374 	 * The first is named txg_%llu, and contains the txg
5375 	 * in which the property was last updated.  The second
5376 	 * is named prop_%llu, and the nth element of its value
5377 	 * should be txg + object + n.
5378 	 */
5379 	tx = dmu_tx_create(os);
5380 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5381 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5382 	if (txg == 0)
5383 		goto out;
5384 
5385 	if (last_txg > txg)
5386 		fatal(B_FALSE, "zap future leak: old %"PRIu64" new %"PRIu64"",
5387 		    last_txg, txg);
5388 
5389 	for (i = 0; i < ints; i++)
5390 		value[i] = txg + object + i;
5391 
5392 	VERIFY0(zap_update(os, object, txgname, sizeof (uint64_t),
5393 	    1, &txg, tx));
5394 	VERIFY0(zap_update(os, object, propname, sizeof (uint64_t),
5395 	    ints, value, tx));
5396 
5397 	dmu_tx_commit(tx);
5398 
5399 	/*
5400 	 * Remove a random pair of entries.
5401 	 */
5402 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5403 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5404 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5405 
5406 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5407 
5408 	if (error == ENOENT)
5409 		goto out;
5410 
5411 	ASSERT0(error);
5412 
5413 	tx = dmu_tx_create(os);
5414 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5415 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5416 	if (txg == 0)
5417 		goto out;
5418 	VERIFY0(zap_remove(os, object, txgname, tx));
5419 	VERIFY0(zap_remove(os, object, propname, tx));
5420 	dmu_tx_commit(tx);
5421 out:
5422 	umem_free(od, sizeof (ztest_od_t));
5423 }
5424 
5425 /*
5426  * Test case to test the upgrading of a microzap to fatzap.
5427  */
5428 void
5429 ztest_fzap(ztest_ds_t *zd, uint64_t id)
5430 {
5431 	objset_t *os = zd->zd_os;
5432 	ztest_od_t *od;
5433 	uint64_t object, txg, value;
5434 
5435 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5436 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5437 
5438 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5439 	    !ztest_random(2)) != 0)
5440 		goto out;
5441 	object = od->od_object;
5442 
5443 	/*
5444 	 * Add entries to this ZAP and make sure it spills over
5445 	 * and gets upgraded to a fatzap. Also, since we are adding
5446 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
5447 	 */
5448 	for (value = 0; value < 2050; value++) {
5449 		char name[ZFS_MAX_DATASET_NAME_LEN];
5450 		dmu_tx_t *tx;
5451 		int error;
5452 
5453 		(void) snprintf(name, sizeof (name), "fzap-%"PRIu64"-%"PRIu64"",
5454 		    id, value);
5455 
5456 		tx = dmu_tx_create(os);
5457 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
5458 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5459 		if (txg == 0)
5460 			goto out;
5461 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
5462 		    &value, tx);
5463 		ASSERT(error == 0 || error == EEXIST);
5464 		dmu_tx_commit(tx);
5465 	}
5466 out:
5467 	umem_free(od, sizeof (ztest_od_t));
5468 }
5469 
5470 void
5471 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
5472 {
5473 	(void) id;
5474 	objset_t *os = zd->zd_os;
5475 	ztest_od_t *od;
5476 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
5477 	dmu_tx_t *tx;
5478 	int i, namelen, error;
5479 	int micro = ztest_random(2);
5480 	char name[20], string_value[20];
5481 	void *data;
5482 
5483 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5484 	ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
5485 
5486 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5487 		umem_free(od, sizeof (ztest_od_t));
5488 		return;
5489 	}
5490 
5491 	object = od->od_object;
5492 
5493 	/*
5494 	 * Generate a random name of the form 'xxx.....' where each
5495 	 * x is a random printable character and the dots are dots.
5496 	 * There are 94 such characters, and the name length goes from
5497 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
5498 	 */
5499 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
5500 
5501 	for (i = 0; i < 3; i++)
5502 		name[i] = '!' + ztest_random('~' - '!' + 1);
5503 	for (; i < namelen - 1; i++)
5504 		name[i] = '.';
5505 	name[i] = '\0';
5506 
5507 	if ((namelen & 1) || micro) {
5508 		wsize = sizeof (txg);
5509 		wc = 1;
5510 		data = &txg;
5511 	} else {
5512 		wsize = 1;
5513 		wc = namelen;
5514 		data = string_value;
5515 	}
5516 
5517 	count = -1ULL;
5518 	VERIFY0(zap_count(os, object, &count));
5519 	ASSERT3S(count, !=, -1ULL);
5520 
5521 	/*
5522 	 * Select an operation: length, lookup, add, update, remove.
5523 	 */
5524 	i = ztest_random(5);
5525 
5526 	if (i >= 2) {
5527 		tx = dmu_tx_create(os);
5528 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5529 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5530 		if (txg == 0) {
5531 			umem_free(od, sizeof (ztest_od_t));
5532 			return;
5533 		}
5534 		memcpy(string_value, name, namelen);
5535 	} else {
5536 		tx = NULL;
5537 		txg = 0;
5538 		memset(string_value, 0, namelen);
5539 	}
5540 
5541 	switch (i) {
5542 
5543 	case 0:
5544 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
5545 		if (error == 0) {
5546 			ASSERT3U(wsize, ==, zl_wsize);
5547 			ASSERT3U(wc, ==, zl_wc);
5548 		} else {
5549 			ASSERT3U(error, ==, ENOENT);
5550 		}
5551 		break;
5552 
5553 	case 1:
5554 		error = zap_lookup(os, object, name, wsize, wc, data);
5555 		if (error == 0) {
5556 			if (data == string_value &&
5557 			    memcmp(name, data, namelen) != 0)
5558 				fatal(B_FALSE, "name '%s' != val '%s' len %d",
5559 				    name, (char *)data, namelen);
5560 		} else {
5561 			ASSERT3U(error, ==, ENOENT);
5562 		}
5563 		break;
5564 
5565 	case 2:
5566 		error = zap_add(os, object, name, wsize, wc, data, tx);
5567 		ASSERT(error == 0 || error == EEXIST);
5568 		break;
5569 
5570 	case 3:
5571 		VERIFY0(zap_update(os, object, name, wsize, wc, data, tx));
5572 		break;
5573 
5574 	case 4:
5575 		error = zap_remove(os, object, name, tx);
5576 		ASSERT(error == 0 || error == ENOENT);
5577 		break;
5578 	}
5579 
5580 	if (tx != NULL)
5581 		dmu_tx_commit(tx);
5582 
5583 	umem_free(od, sizeof (ztest_od_t));
5584 }
5585 
5586 /*
5587  * Commit callback data.
5588  */
5589 typedef struct ztest_cb_data {
5590 	list_node_t		zcd_node;
5591 	uint64_t		zcd_txg;
5592 	int			zcd_expected_err;
5593 	boolean_t		zcd_added;
5594 	boolean_t		zcd_called;
5595 	spa_t			*zcd_spa;
5596 } ztest_cb_data_t;
5597 
5598 /* This is the actual commit callback function */
5599 static void
5600 ztest_commit_callback(void *arg, int error)
5601 {
5602 	ztest_cb_data_t *data = arg;
5603 	uint64_t synced_txg;
5604 
5605 	VERIFY3P(data, !=, NULL);
5606 	VERIFY3S(data->zcd_expected_err, ==, error);
5607 	VERIFY(!data->zcd_called);
5608 
5609 	synced_txg = spa_last_synced_txg(data->zcd_spa);
5610 	if (data->zcd_txg > synced_txg)
5611 		fatal(B_FALSE,
5612 		    "commit callback of txg %"PRIu64" called prematurely, "
5613 		    "last synced txg = %"PRIu64"\n",
5614 		    data->zcd_txg, synced_txg);
5615 
5616 	data->zcd_called = B_TRUE;
5617 
5618 	if (error == ECANCELED) {
5619 		ASSERT0(data->zcd_txg);
5620 		ASSERT(!data->zcd_added);
5621 
5622 		/*
5623 		 * The private callback data should be destroyed here, but
5624 		 * since we are going to check the zcd_called field after
5625 		 * dmu_tx_abort(), we will destroy it there.
5626 		 */
5627 		return;
5628 	}
5629 
5630 	ASSERT(data->zcd_added);
5631 	ASSERT3U(data->zcd_txg, !=, 0);
5632 
5633 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5634 
5635 	/* See if this cb was called more quickly */
5636 	if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
5637 		zc_min_txg_delay = synced_txg - data->zcd_txg;
5638 
5639 	/* Remove our callback from the list */
5640 	list_remove(&zcl.zcl_callbacks, data);
5641 
5642 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5643 
5644 	umem_free(data, sizeof (ztest_cb_data_t));
5645 }
5646 
5647 /* Allocate and initialize callback data structure */
5648 static ztest_cb_data_t *
5649 ztest_create_cb_data(objset_t *os, uint64_t txg)
5650 {
5651 	ztest_cb_data_t *cb_data;
5652 
5653 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
5654 
5655 	cb_data->zcd_txg = txg;
5656 	cb_data->zcd_spa = dmu_objset_spa(os);
5657 	list_link_init(&cb_data->zcd_node);
5658 
5659 	return (cb_data);
5660 }
5661 
5662 /*
5663  * Commit callback test.
5664  */
5665 void
5666 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
5667 {
5668 	objset_t *os = zd->zd_os;
5669 	ztest_od_t *od;
5670 	dmu_tx_t *tx;
5671 	ztest_cb_data_t *cb_data[3], *tmp_cb;
5672 	uint64_t old_txg, txg;
5673 	int i, error = 0;
5674 
5675 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5676 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5677 
5678 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5679 		umem_free(od, sizeof (ztest_od_t));
5680 		return;
5681 	}
5682 
5683 	tx = dmu_tx_create(os);
5684 
5685 	cb_data[0] = ztest_create_cb_data(os, 0);
5686 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
5687 
5688 	dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
5689 
5690 	/* Every once in a while, abort the transaction on purpose */
5691 	if (ztest_random(100) == 0)
5692 		error = -1;
5693 
5694 	if (!error)
5695 		error = dmu_tx_assign(tx, TXG_NOWAIT);
5696 
5697 	txg = error ? 0 : dmu_tx_get_txg(tx);
5698 
5699 	cb_data[0]->zcd_txg = txg;
5700 	cb_data[1] = ztest_create_cb_data(os, txg);
5701 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
5702 
5703 	if (error) {
5704 		/*
5705 		 * It's not a strict requirement to call the registered
5706 		 * callbacks from inside dmu_tx_abort(), but that's what
5707 		 * it's supposed to happen in the current implementation
5708 		 * so we will check for that.
5709 		 */
5710 		for (i = 0; i < 2; i++) {
5711 			cb_data[i]->zcd_expected_err = ECANCELED;
5712 			VERIFY(!cb_data[i]->zcd_called);
5713 		}
5714 
5715 		dmu_tx_abort(tx);
5716 
5717 		for (i = 0; i < 2; i++) {
5718 			VERIFY(cb_data[i]->zcd_called);
5719 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
5720 		}
5721 
5722 		umem_free(od, sizeof (ztest_od_t));
5723 		return;
5724 	}
5725 
5726 	cb_data[2] = ztest_create_cb_data(os, txg);
5727 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
5728 
5729 	/*
5730 	 * Read existing data to make sure there isn't a future leak.
5731 	 */
5732 	VERIFY0(dmu_read(os, od->od_object, 0, sizeof (uint64_t),
5733 	    &old_txg, DMU_READ_PREFETCH));
5734 
5735 	if (old_txg > txg)
5736 		fatal(B_FALSE,
5737 		    "future leak: got %"PRIu64", open txg is %"PRIu64"",
5738 		    old_txg, txg);
5739 
5740 	dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
5741 
5742 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5743 
5744 	/*
5745 	 * Since commit callbacks don't have any ordering requirement and since
5746 	 * it is theoretically possible for a commit callback to be called
5747 	 * after an arbitrary amount of time has elapsed since its txg has been
5748 	 * synced, it is difficult to reliably determine whether a commit
5749 	 * callback hasn't been called due to high load or due to a flawed
5750 	 * implementation.
5751 	 *
5752 	 * In practice, we will assume that if after a certain number of txgs a
5753 	 * commit callback hasn't been called, then most likely there's an
5754 	 * implementation bug..
5755 	 */
5756 	tmp_cb = list_head(&zcl.zcl_callbacks);
5757 	if (tmp_cb != NULL &&
5758 	    tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
5759 		fatal(B_FALSE,
5760 		    "Commit callback threshold exceeded, "
5761 		    "oldest txg: %"PRIu64", open txg: %"PRIu64"\n",
5762 		    tmp_cb->zcd_txg, txg);
5763 	}
5764 
5765 	/*
5766 	 * Let's find the place to insert our callbacks.
5767 	 *
5768 	 * Even though the list is ordered by txg, it is possible for the
5769 	 * insertion point to not be the end because our txg may already be
5770 	 * quiescing at this point and other callbacks in the open txg
5771 	 * (from other objsets) may have sneaked in.
5772 	 */
5773 	tmp_cb = list_tail(&zcl.zcl_callbacks);
5774 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
5775 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
5776 
5777 	/* Add the 3 callbacks to the list */
5778 	for (i = 0; i < 3; i++) {
5779 		if (tmp_cb == NULL)
5780 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
5781 		else
5782 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
5783 			    cb_data[i]);
5784 
5785 		cb_data[i]->zcd_added = B_TRUE;
5786 		VERIFY(!cb_data[i]->zcd_called);
5787 
5788 		tmp_cb = cb_data[i];
5789 	}
5790 
5791 	zc_cb_counter += 3;
5792 
5793 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5794 
5795 	dmu_tx_commit(tx);
5796 
5797 	umem_free(od, sizeof (ztest_od_t));
5798 }
5799 
5800 /*
5801  * Visit each object in the dataset. Verify that its properties
5802  * are consistent what was stored in the block tag when it was created,
5803  * and that its unused bonus buffer space has not been overwritten.
5804  */
5805 void
5806 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5807 {
5808 	(void) id;
5809 	objset_t *os = zd->zd_os;
5810 	uint64_t obj;
5811 	int err = 0;
5812 
5813 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5814 		ztest_block_tag_t *bt = NULL;
5815 		dmu_object_info_t doi;
5816 		dmu_buf_t *db;
5817 
5818 		ztest_object_lock(zd, obj, RL_READER);
5819 		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) {
5820 			ztest_object_unlock(zd, obj);
5821 			continue;
5822 		}
5823 
5824 		dmu_object_info_from_db(db, &doi);
5825 		if (doi.doi_bonus_size >= sizeof (*bt))
5826 			bt = ztest_bt_bonus(db);
5827 
5828 		if (bt && bt->bt_magic == BT_MAGIC) {
5829 			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5830 			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
5831 			    bt->bt_crtxg);
5832 			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5833 		}
5834 
5835 		dmu_buf_rele(db, FTAG);
5836 		ztest_object_unlock(zd, obj);
5837 	}
5838 }
5839 
5840 void
5841 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5842 {
5843 	(void) id;
5844 	zfs_prop_t proplist[] = {
5845 		ZFS_PROP_CHECKSUM,
5846 		ZFS_PROP_COMPRESSION,
5847 		ZFS_PROP_COPIES,
5848 		ZFS_PROP_DEDUP
5849 	};
5850 
5851 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5852 
5853 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) {
5854 		int error = ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5855 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5856 		ASSERT(error == 0 || error == ENOSPC);
5857 	}
5858 
5859 	int error = ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
5860 	    ztest_random_blocksize(), (int)ztest_random(2));
5861 	ASSERT(error == 0 || error == ENOSPC);
5862 
5863 	(void) pthread_rwlock_unlock(&ztest_name_lock);
5864 }
5865 
5866 void
5867 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5868 {
5869 	(void) zd, (void) id;
5870 	nvlist_t *props = NULL;
5871 
5872 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5873 
5874 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_AUTOTRIM, ztest_random(2));
5875 
5876 	VERIFY0(spa_prop_get(ztest_spa, &props));
5877 
5878 	if (ztest_opts.zo_verbose >= 6)
5879 		dump_nvlist(props, 4);
5880 
5881 	fnvlist_free(props);
5882 
5883 	(void) pthread_rwlock_unlock(&ztest_name_lock);
5884 }
5885 
5886 static int
5887 user_release_one(const char *snapname, const char *holdname)
5888 {
5889 	nvlist_t *snaps, *holds;
5890 	int error;
5891 
5892 	snaps = fnvlist_alloc();
5893 	holds = fnvlist_alloc();
5894 	fnvlist_add_boolean(holds, holdname);
5895 	fnvlist_add_nvlist(snaps, snapname, holds);
5896 	fnvlist_free(holds);
5897 	error = dsl_dataset_user_release(snaps, NULL);
5898 	fnvlist_free(snaps);
5899 	return (error);
5900 }
5901 
5902 /*
5903  * Test snapshot hold/release and deferred destroy.
5904  */
5905 void
5906 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5907 {
5908 	int error;
5909 	objset_t *os = zd->zd_os;
5910 	objset_t *origin;
5911 	char snapname[100];
5912 	char fullname[100];
5913 	char clonename[100];
5914 	char tag[100];
5915 	char osname[ZFS_MAX_DATASET_NAME_LEN];
5916 	nvlist_t *holds;
5917 
5918 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5919 
5920 	dmu_objset_name(os, osname);
5921 
5922 	(void) snprintf(snapname, sizeof (snapname), "sh1_%"PRIu64"", id);
5923 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5924 	(void) snprintf(clonename, sizeof (clonename), "%s/ch1_%"PRIu64"",
5925 	    osname, id);
5926 	(void) snprintf(tag, sizeof (tag), "tag_%"PRIu64"", id);
5927 
5928 	/*
5929 	 * Clean up from any previous run.
5930 	 */
5931 	error = dsl_destroy_head(clonename);
5932 	if (error != ENOENT)
5933 		ASSERT0(error);
5934 	error = user_release_one(fullname, tag);
5935 	if (error != ESRCH && error != ENOENT)
5936 		ASSERT0(error);
5937 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5938 	if (error != ENOENT)
5939 		ASSERT0(error);
5940 
5941 	/*
5942 	 * Create snapshot, clone it, mark snap for deferred destroy,
5943 	 * destroy clone, verify snap was also destroyed.
5944 	 */
5945 	error = dmu_objset_snapshot_one(osname, snapname);
5946 	if (error) {
5947 		if (error == ENOSPC) {
5948 			ztest_record_enospc("dmu_objset_snapshot");
5949 			goto out;
5950 		}
5951 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
5952 	}
5953 
5954 	error = dmu_objset_clone(clonename, fullname);
5955 	if (error) {
5956 		if (error == ENOSPC) {
5957 			ztest_record_enospc("dmu_objset_clone");
5958 			goto out;
5959 		}
5960 		fatal(B_FALSE, "dmu_objset_clone(%s) = %d", clonename, error);
5961 	}
5962 
5963 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5964 	if (error) {
5965 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5966 		    fullname, error);
5967 	}
5968 
5969 	error = dsl_destroy_head(clonename);
5970 	if (error)
5971 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clonename, error);
5972 
5973 	error = dmu_objset_hold(fullname, FTAG, &origin);
5974 	if (error != ENOENT)
5975 		fatal(B_FALSE, "dmu_objset_hold(%s) = %d", fullname, error);
5976 
5977 	/*
5978 	 * Create snapshot, add temporary hold, verify that we can't
5979 	 * destroy a held snapshot, mark for deferred destroy,
5980 	 * release hold, verify snapshot was destroyed.
5981 	 */
5982 	error = dmu_objset_snapshot_one(osname, snapname);
5983 	if (error) {
5984 		if (error == ENOSPC) {
5985 			ztest_record_enospc("dmu_objset_snapshot");
5986 			goto out;
5987 		}
5988 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
5989 	}
5990 
5991 	holds = fnvlist_alloc();
5992 	fnvlist_add_string(holds, fullname, tag);
5993 	error = dsl_dataset_user_hold(holds, 0, NULL);
5994 	fnvlist_free(holds);
5995 
5996 	if (error == ENOSPC) {
5997 		ztest_record_enospc("dsl_dataset_user_hold");
5998 		goto out;
5999 	} else if (error) {
6000 		fatal(B_FALSE, "dsl_dataset_user_hold(%s, %s) = %u",
6001 		    fullname, tag, error);
6002 	}
6003 
6004 	error = dsl_destroy_snapshot(fullname, B_FALSE);
6005 	if (error != EBUSY) {
6006 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
6007 		    fullname, error);
6008 	}
6009 
6010 	error = dsl_destroy_snapshot(fullname, B_TRUE);
6011 	if (error) {
6012 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
6013 		    fullname, error);
6014 	}
6015 
6016 	error = user_release_one(fullname, tag);
6017 	if (error)
6018 		fatal(B_FALSE, "user_release_one(%s, %s) = %d",
6019 		    fullname, tag, error);
6020 
6021 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
6022 
6023 out:
6024 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6025 }
6026 
6027 /*
6028  * Inject random faults into the on-disk data.
6029  */
6030 void
6031 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
6032 {
6033 	(void) zd, (void) id;
6034 	ztest_shared_t *zs = ztest_shared;
6035 	spa_t *spa = ztest_spa;
6036 	int fd;
6037 	uint64_t offset;
6038 	uint64_t leaves;
6039 	uint64_t bad = 0x1990c0ffeedecadeull;
6040 	uint64_t top, leaf;
6041 	char *path0;
6042 	char *pathrand;
6043 	size_t fsize;
6044 	int bshift = SPA_MAXBLOCKSHIFT + 2;
6045 	int iters = 1000;
6046 	int maxfaults;
6047 	int mirror_save;
6048 	vdev_t *vd0 = NULL;
6049 	uint64_t guid0 = 0;
6050 	boolean_t islog = B_FALSE;
6051 
6052 	path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6053 	pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6054 
6055 	mutex_enter(&ztest_vdev_lock);
6056 
6057 	/*
6058 	 * Device removal is in progress, fault injection must be disabled
6059 	 * until it completes and the pool is scrubbed.  The fault injection
6060 	 * strategy for damaging blocks does not take in to account evacuated
6061 	 * blocks which may have already been damaged.
6062 	 */
6063 	if (ztest_device_removal_active) {
6064 		mutex_exit(&ztest_vdev_lock);
6065 		goto out;
6066 	}
6067 
6068 	maxfaults = MAXFAULTS(zs);
6069 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raid_children;
6070 	mirror_save = zs->zs_mirrors;
6071 	mutex_exit(&ztest_vdev_lock);
6072 
6073 	ASSERT3U(leaves, >=, 1);
6074 
6075 	/*
6076 	 * While ztest is running the number of leaves will not change.  This
6077 	 * is critical for the fault injection logic as it determines where
6078 	 * errors can be safely injected such that they are always repairable.
6079 	 *
6080 	 * When restarting ztest a different number of leaves may be requested
6081 	 * which will shift the regions to be damaged.  This is fine as long
6082 	 * as the pool has been scrubbed prior to using the new mapping.
6083 	 * Failure to do can result in non-repairable damage being injected.
6084 	 */
6085 	if (ztest_pool_scrubbed == B_FALSE)
6086 		goto out;
6087 
6088 	/*
6089 	 * Grab the name lock as reader. There are some operations
6090 	 * which don't like to have their vdevs changed while
6091 	 * they are in progress (i.e. spa_change_guid). Those
6092 	 * operations will have grabbed the name lock as writer.
6093 	 */
6094 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6095 
6096 	/*
6097 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
6098 	 */
6099 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6100 
6101 	if (ztest_random(2) == 0) {
6102 		/*
6103 		 * Inject errors on a normal data device or slog device.
6104 		 */
6105 		top = ztest_random_vdev_top(spa, B_TRUE);
6106 		leaf = ztest_random(leaves) + zs->zs_splits;
6107 
6108 		/*
6109 		 * Generate paths to the first leaf in this top-level vdev,
6110 		 * and to the random leaf we selected.  We'll induce transient
6111 		 * write failures and random online/offline activity on leaf 0,
6112 		 * and we'll write random garbage to the randomly chosen leaf.
6113 		 */
6114 		(void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
6115 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6116 		    top * leaves + zs->zs_splits);
6117 		(void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
6118 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6119 		    top * leaves + leaf);
6120 
6121 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
6122 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
6123 			islog = B_TRUE;
6124 
6125 		/*
6126 		 * If the top-level vdev needs to be resilvered
6127 		 * then we only allow faults on the device that is
6128 		 * resilvering.
6129 		 */
6130 		if (vd0 != NULL && maxfaults != 1 &&
6131 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
6132 		    vd0->vdev_resilver_txg != 0)) {
6133 			/*
6134 			 * Make vd0 explicitly claim to be unreadable,
6135 			 * or unwritable, or reach behind its back
6136 			 * and close the underlying fd.  We can do this if
6137 			 * maxfaults == 0 because we'll fail and reexecute,
6138 			 * and we can do it if maxfaults >= 2 because we'll
6139 			 * have enough redundancy.  If maxfaults == 1, the
6140 			 * combination of this with injection of random data
6141 			 * corruption below exceeds the pool's fault tolerance.
6142 			 */
6143 			vdev_file_t *vf = vd0->vdev_tsd;
6144 
6145 			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
6146 			    (long long)vd0->vdev_id, (int)maxfaults);
6147 
6148 			if (vf != NULL && ztest_random(3) == 0) {
6149 				(void) close(vf->vf_file->f_fd);
6150 				vf->vf_file->f_fd = -1;
6151 			} else if (ztest_random(2) == 0) {
6152 				vd0->vdev_cant_read = B_TRUE;
6153 			} else {
6154 				vd0->vdev_cant_write = B_TRUE;
6155 			}
6156 			guid0 = vd0->vdev_guid;
6157 		}
6158 	} else {
6159 		/*
6160 		 * Inject errors on an l2cache device.
6161 		 */
6162 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
6163 
6164 		if (sav->sav_count == 0) {
6165 			spa_config_exit(spa, SCL_STATE, FTAG);
6166 			(void) pthread_rwlock_unlock(&ztest_name_lock);
6167 			goto out;
6168 		}
6169 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
6170 		guid0 = vd0->vdev_guid;
6171 		(void) strlcpy(path0, vd0->vdev_path, MAXPATHLEN);
6172 		(void) strlcpy(pathrand, vd0->vdev_path, MAXPATHLEN);
6173 
6174 		leaf = 0;
6175 		leaves = 1;
6176 		maxfaults = INT_MAX;	/* no limit on cache devices */
6177 	}
6178 
6179 	spa_config_exit(spa, SCL_STATE, FTAG);
6180 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6181 
6182 	/*
6183 	 * If we can tolerate two or more faults, or we're dealing
6184 	 * with a slog, randomly online/offline vd0.
6185 	 */
6186 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
6187 		if (ztest_random(10) < 6) {
6188 			int flags = (ztest_random(2) == 0 ?
6189 			    ZFS_OFFLINE_TEMPORARY : 0);
6190 
6191 			/*
6192 			 * We have to grab the zs_name_lock as writer to
6193 			 * prevent a race between offlining a slog and
6194 			 * destroying a dataset. Offlining the slog will
6195 			 * grab a reference on the dataset which may cause
6196 			 * dsl_destroy_head() to fail with EBUSY thus
6197 			 * leaving the dataset in an inconsistent state.
6198 			 */
6199 			if (islog)
6200 				(void) pthread_rwlock_wrlock(&ztest_name_lock);
6201 
6202 			VERIFY3U(vdev_offline(spa, guid0, flags), !=, EBUSY);
6203 
6204 			if (islog)
6205 				(void) pthread_rwlock_unlock(&ztest_name_lock);
6206 		} else {
6207 			/*
6208 			 * Ideally we would like to be able to randomly
6209 			 * call vdev_[on|off]line without holding locks
6210 			 * to force unpredictable failures but the side
6211 			 * effects of vdev_[on|off]line prevent us from
6212 			 * doing so. We grab the ztest_vdev_lock here to
6213 			 * prevent a race between injection testing and
6214 			 * aux_vdev removal.
6215 			 */
6216 			mutex_enter(&ztest_vdev_lock);
6217 			(void) vdev_online(spa, guid0, 0, NULL);
6218 			mutex_exit(&ztest_vdev_lock);
6219 		}
6220 	}
6221 
6222 	if (maxfaults == 0)
6223 		goto out;
6224 
6225 	/*
6226 	 * We have at least single-fault tolerance, so inject data corruption.
6227 	 */
6228 	fd = open(pathrand, O_RDWR);
6229 
6230 	if (fd == -1) /* we hit a gap in the device namespace */
6231 		goto out;
6232 
6233 	fsize = lseek(fd, 0, SEEK_END);
6234 
6235 	while (--iters != 0) {
6236 		/*
6237 		 * The offset must be chosen carefully to ensure that
6238 		 * we do not inject a given logical block with errors
6239 		 * on two different leaf devices, because ZFS can not
6240 		 * tolerate that (if maxfaults==1).
6241 		 *
6242 		 * To achieve this we divide each leaf device into
6243 		 * chunks of size (# leaves * SPA_MAXBLOCKSIZE * 4).
6244 		 * Each chunk is further divided into error-injection
6245 		 * ranges (can accept errors) and clear ranges (we do
6246 		 * not inject errors in those). Each error-injection
6247 		 * range can accept errors only for a single leaf vdev.
6248 		 * Error-injection ranges are separated by clear ranges.
6249 		 *
6250 		 * For example, with 3 leaves, each chunk looks like:
6251 		 *    0 to  32M: injection range for leaf 0
6252 		 *  32M to  64M: clear range - no injection allowed
6253 		 *  64M to  96M: injection range for leaf 1
6254 		 *  96M to 128M: clear range - no injection allowed
6255 		 * 128M to 160M: injection range for leaf 2
6256 		 * 160M to 192M: clear range - no injection allowed
6257 		 *
6258 		 * Each clear range must be large enough such that a
6259 		 * single block cannot straddle it. This way a block
6260 		 * can't be a target in two different injection ranges
6261 		 * (on different leaf vdevs).
6262 		 */
6263 		offset = ztest_random(fsize / (leaves << bshift)) *
6264 		    (leaves << bshift) + (leaf << bshift) +
6265 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
6266 
6267 		/*
6268 		 * Only allow damage to the labels at one end of the vdev.
6269 		 *
6270 		 * If all labels are damaged, the device will be totally
6271 		 * inaccessible, which will result in loss of data,
6272 		 * because we also damage (parts of) the other side of
6273 		 * the mirror/raidz.
6274 		 *
6275 		 * Additionally, we will always have both an even and an
6276 		 * odd label, so that we can handle crashes in the
6277 		 * middle of vdev_config_sync().
6278 		 */
6279 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
6280 			continue;
6281 
6282 		/*
6283 		 * The two end labels are stored at the "end" of the disk, but
6284 		 * the end of the disk (vdev_psize) is aligned to
6285 		 * sizeof (vdev_label_t).
6286 		 */
6287 		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
6288 		if ((leaf & 1) == 1 &&
6289 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
6290 			continue;
6291 
6292 		mutex_enter(&ztest_vdev_lock);
6293 		if (mirror_save != zs->zs_mirrors) {
6294 			mutex_exit(&ztest_vdev_lock);
6295 			(void) close(fd);
6296 			goto out;
6297 		}
6298 
6299 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
6300 			fatal(B_TRUE,
6301 			    "can't inject bad word at 0x%"PRIx64" in %s",
6302 			    offset, pathrand);
6303 
6304 		mutex_exit(&ztest_vdev_lock);
6305 
6306 		if (ztest_opts.zo_verbose >= 7)
6307 			(void) printf("injected bad word into %s,"
6308 			    " offset 0x%"PRIx64"\n", pathrand, offset);
6309 	}
6310 
6311 	(void) close(fd);
6312 out:
6313 	umem_free(path0, MAXPATHLEN);
6314 	umem_free(pathrand, MAXPATHLEN);
6315 }
6316 
6317 /*
6318  * By design ztest will never inject uncorrectable damage in to the pool.
6319  * Issue a scrub, wait for it to complete, and verify there is never any
6320  * persistent damage.
6321  *
6322  * Only after a full scrub has been completed is it safe to start injecting
6323  * data corruption.  See the comment in zfs_fault_inject().
6324  */
6325 static int
6326 ztest_scrub_impl(spa_t *spa)
6327 {
6328 	int error = spa_scan(spa, POOL_SCAN_SCRUB);
6329 	if (error)
6330 		return (error);
6331 
6332 	while (dsl_scan_scrubbing(spa_get_dsl(spa)))
6333 		txg_wait_synced(spa_get_dsl(spa), 0);
6334 
6335 	if (spa_approx_errlog_size(spa) > 0)
6336 		return (ECKSUM);
6337 
6338 	ztest_pool_scrubbed = B_TRUE;
6339 
6340 	return (0);
6341 }
6342 
6343 /*
6344  * Scrub the pool.
6345  */
6346 void
6347 ztest_scrub(ztest_ds_t *zd, uint64_t id)
6348 {
6349 	(void) zd, (void) id;
6350 	spa_t *spa = ztest_spa;
6351 	int error;
6352 
6353 	/*
6354 	 * Scrub in progress by device removal.
6355 	 */
6356 	if (ztest_device_removal_active)
6357 		return;
6358 
6359 	/*
6360 	 * Start a scrub, wait a moment, then force a restart.
6361 	 */
6362 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
6363 	(void) poll(NULL, 0, 100);
6364 
6365 	error = ztest_scrub_impl(spa);
6366 	if (error == EBUSY)
6367 		error = 0;
6368 	ASSERT0(error);
6369 }
6370 
6371 /*
6372  * Change the guid for the pool.
6373  */
6374 void
6375 ztest_reguid(ztest_ds_t *zd, uint64_t id)
6376 {
6377 	(void) zd, (void) id;
6378 	spa_t *spa = ztest_spa;
6379 	uint64_t orig, load;
6380 	int error;
6381 	ztest_shared_t *zs = ztest_shared;
6382 
6383 	if (ztest_opts.zo_mmp_test)
6384 		return;
6385 
6386 	orig = spa_guid(spa);
6387 	load = spa_load_guid(spa);
6388 
6389 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
6390 	error = spa_change_guid(spa);
6391 	zs->zs_guid = spa_guid(spa);
6392 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6393 
6394 	if (error != 0)
6395 		return;
6396 
6397 	if (ztest_opts.zo_verbose >= 4) {
6398 		(void) printf("Changed guid old %"PRIu64" -> %"PRIu64"\n",
6399 		    orig, spa_guid(spa));
6400 	}
6401 
6402 	VERIFY3U(orig, !=, spa_guid(spa));
6403 	VERIFY3U(load, ==, spa_load_guid(spa));
6404 }
6405 
6406 void
6407 ztest_blake3(ztest_ds_t *zd, uint64_t id)
6408 {
6409 	(void) zd, (void) id;
6410 	hrtime_t end = gethrtime() + NANOSEC;
6411 	zio_cksum_salt_t salt;
6412 	void *salt_ptr = &salt.zcs_bytes;
6413 	struct abd *abd_data, *abd_meta;
6414 	void *buf, *templ;
6415 	int i, *ptr;
6416 	uint32_t size;
6417 	BLAKE3_CTX ctx;
6418 	const zfs_impl_t *blake3 = zfs_impl_get_ops("blake3");
6419 
6420 	size = ztest_random_blocksize();
6421 	buf = umem_alloc(size, UMEM_NOFAIL);
6422 	abd_data = abd_alloc(size, B_FALSE);
6423 	abd_meta = abd_alloc(size, B_TRUE);
6424 
6425 	for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6426 		*ptr = ztest_random(UINT_MAX);
6427 	memset(salt_ptr, 'A', 32);
6428 
6429 	abd_copy_from_buf_off(abd_data, buf, 0, size);
6430 	abd_copy_from_buf_off(abd_meta, buf, 0, size);
6431 
6432 	while (gethrtime() <= end) {
6433 		int run_count = 100;
6434 		zio_cksum_t zc_ref1, zc_ref2;
6435 		zio_cksum_t zc_res1, zc_res2;
6436 
6437 		void *ref1 = &zc_ref1;
6438 		void *ref2 = &zc_ref2;
6439 		void *res1 = &zc_res1;
6440 		void *res2 = &zc_res2;
6441 
6442 		/* BLAKE3_KEY_LEN = 32 */
6443 		VERIFY0(blake3->setname("generic"));
6444 		templ = abd_checksum_blake3_tmpl_init(&salt);
6445 		Blake3_InitKeyed(&ctx, salt_ptr);
6446 		Blake3_Update(&ctx, buf, size);
6447 		Blake3_Final(&ctx, ref1);
6448 		zc_ref2 = zc_ref1;
6449 		ZIO_CHECKSUM_BSWAP(&zc_ref2);
6450 		abd_checksum_blake3_tmpl_free(templ);
6451 
6452 		VERIFY0(blake3->setname("cycle"));
6453 		while (run_count-- > 0) {
6454 
6455 			/* Test current implementation */
6456 			Blake3_InitKeyed(&ctx, salt_ptr);
6457 			Blake3_Update(&ctx, buf, size);
6458 			Blake3_Final(&ctx, res1);
6459 			zc_res2 = zc_res1;
6460 			ZIO_CHECKSUM_BSWAP(&zc_res2);
6461 
6462 			VERIFY0(memcmp(ref1, res1, 32));
6463 			VERIFY0(memcmp(ref2, res2, 32));
6464 
6465 			/* Test ABD - data */
6466 			templ = abd_checksum_blake3_tmpl_init(&salt);
6467 			abd_checksum_blake3_native(abd_data, size,
6468 			    templ, &zc_res1);
6469 			abd_checksum_blake3_byteswap(abd_data, size,
6470 			    templ, &zc_res2);
6471 
6472 			VERIFY0(memcmp(ref1, res1, 32));
6473 			VERIFY0(memcmp(ref2, res2, 32));
6474 
6475 			/* Test ABD - metadata */
6476 			abd_checksum_blake3_native(abd_meta, size,
6477 			    templ, &zc_res1);
6478 			abd_checksum_blake3_byteswap(abd_meta, size,
6479 			    templ, &zc_res2);
6480 			abd_checksum_blake3_tmpl_free(templ);
6481 
6482 			VERIFY0(memcmp(ref1, res1, 32));
6483 			VERIFY0(memcmp(ref2, res2, 32));
6484 
6485 		}
6486 	}
6487 
6488 	abd_free(abd_data);
6489 	abd_free(abd_meta);
6490 	umem_free(buf, size);
6491 }
6492 
6493 void
6494 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
6495 {
6496 	(void) zd, (void) id;
6497 	hrtime_t end = gethrtime() + NANOSEC;
6498 
6499 	while (gethrtime() <= end) {
6500 		int run_count = 100;
6501 		void *buf;
6502 		struct abd *abd_data, *abd_meta;
6503 		uint32_t size;
6504 		int *ptr;
6505 		int i;
6506 		zio_cksum_t zc_ref;
6507 		zio_cksum_t zc_ref_byteswap;
6508 
6509 		size = ztest_random_blocksize();
6510 
6511 		buf = umem_alloc(size, UMEM_NOFAIL);
6512 		abd_data = abd_alloc(size, B_FALSE);
6513 		abd_meta = abd_alloc(size, B_TRUE);
6514 
6515 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6516 			*ptr = ztest_random(UINT_MAX);
6517 
6518 		abd_copy_from_buf_off(abd_data, buf, 0, size);
6519 		abd_copy_from_buf_off(abd_meta, buf, 0, size);
6520 
6521 		VERIFY0(fletcher_4_impl_set("scalar"));
6522 		fletcher_4_native(buf, size, NULL, &zc_ref);
6523 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
6524 
6525 		VERIFY0(fletcher_4_impl_set("cycle"));
6526 		while (run_count-- > 0) {
6527 			zio_cksum_t zc;
6528 			zio_cksum_t zc_byteswap;
6529 
6530 			fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
6531 			fletcher_4_native(buf, size, NULL, &zc);
6532 
6533 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6534 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6535 			    sizeof (zc_byteswap)));
6536 
6537 			/* Test ABD - data */
6538 			abd_fletcher_4_byteswap(abd_data, size, NULL,
6539 			    &zc_byteswap);
6540 			abd_fletcher_4_native(abd_data, size, NULL, &zc);
6541 
6542 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6543 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6544 			    sizeof (zc_byteswap)));
6545 
6546 			/* Test ABD - metadata */
6547 			abd_fletcher_4_byteswap(abd_meta, size, NULL,
6548 			    &zc_byteswap);
6549 			abd_fletcher_4_native(abd_meta, size, NULL, &zc);
6550 
6551 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6552 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6553 			    sizeof (zc_byteswap)));
6554 
6555 		}
6556 
6557 		umem_free(buf, size);
6558 		abd_free(abd_data);
6559 		abd_free(abd_meta);
6560 	}
6561 }
6562 
6563 void
6564 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
6565 {
6566 	(void) zd, (void) id;
6567 	void *buf;
6568 	size_t size;
6569 	int *ptr;
6570 	int i;
6571 	zio_cksum_t zc_ref;
6572 	zio_cksum_t zc_ref_bswap;
6573 
6574 	hrtime_t end = gethrtime() + NANOSEC;
6575 
6576 	while (gethrtime() <= end) {
6577 		int run_count = 100;
6578 
6579 		size = ztest_random_blocksize();
6580 		buf = umem_alloc(size, UMEM_NOFAIL);
6581 
6582 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6583 			*ptr = ztest_random(UINT_MAX);
6584 
6585 		VERIFY0(fletcher_4_impl_set("scalar"));
6586 		fletcher_4_native(buf, size, NULL, &zc_ref);
6587 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
6588 
6589 		VERIFY0(fletcher_4_impl_set("cycle"));
6590 
6591 		while (run_count-- > 0) {
6592 			zio_cksum_t zc;
6593 			zio_cksum_t zc_bswap;
6594 			size_t pos = 0;
6595 
6596 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6597 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6598 
6599 			while (pos < size) {
6600 				size_t inc = 64 * ztest_random(size / 67);
6601 				/* sometimes add few bytes to test non-simd */
6602 				if (ztest_random(100) < 10)
6603 					inc += P2ALIGN(ztest_random(64),
6604 					    sizeof (uint32_t));
6605 
6606 				if (inc > (size - pos))
6607 					inc = size - pos;
6608 
6609 				fletcher_4_incremental_native(buf + pos, inc,
6610 				    &zc);
6611 				fletcher_4_incremental_byteswap(buf + pos, inc,
6612 				    &zc_bswap);
6613 
6614 				pos += inc;
6615 			}
6616 
6617 			VERIFY3U(pos, ==, size);
6618 
6619 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6620 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6621 
6622 			/*
6623 			 * verify if incremental on the whole buffer is
6624 			 * equivalent to non-incremental version
6625 			 */
6626 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6627 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6628 
6629 			fletcher_4_incremental_native(buf, size, &zc);
6630 			fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
6631 
6632 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6633 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6634 		}
6635 
6636 		umem_free(buf, size);
6637 	}
6638 }
6639 
6640 static int
6641 ztest_set_global_vars(void)
6642 {
6643 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
6644 		char *kv = ztest_opts.zo_gvars[i];
6645 		VERIFY3U(strlen(kv), <=, ZO_GVARS_MAX_ARGLEN);
6646 		VERIFY3U(strlen(kv), >, 0);
6647 		int err = set_global_var(kv);
6648 		if (ztest_opts.zo_verbose > 0) {
6649 			(void) printf("setting global var %s ... %s\n", kv,
6650 			    err ? "failed" : "ok");
6651 		}
6652 		if (err != 0) {
6653 			(void) fprintf(stderr,
6654 			    "failed to set global var '%s'\n", kv);
6655 			return (err);
6656 		}
6657 	}
6658 	return (0);
6659 }
6660 
6661 static char **
6662 ztest_global_vars_to_zdb_args(void)
6663 {
6664 	char **args = calloc(2*ztest_opts.zo_gvars_count + 1, sizeof (char *));
6665 	char **cur = args;
6666 	if (args == NULL)
6667 		return (NULL);
6668 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
6669 		*cur++ = (char *)"-o";
6670 		*cur++ = ztest_opts.zo_gvars[i];
6671 	}
6672 	ASSERT3P(cur, ==, &args[2*ztest_opts.zo_gvars_count]);
6673 	*cur = NULL;
6674 	return (args);
6675 }
6676 
6677 /* The end of strings is indicated by a NULL element */
6678 static char *
6679 join_strings(char **strings, const char *sep)
6680 {
6681 	size_t totallen = 0;
6682 	for (char **sp = strings; *sp != NULL; sp++) {
6683 		totallen += strlen(*sp);
6684 		totallen += strlen(sep);
6685 	}
6686 	if (totallen > 0) {
6687 		ASSERT(totallen >= strlen(sep));
6688 		totallen -= strlen(sep);
6689 	}
6690 
6691 	size_t buflen = totallen + 1;
6692 	char *o = umem_alloc(buflen, UMEM_NOFAIL); /* trailing 0 byte */
6693 	o[0] = '\0';
6694 	for (char **sp = strings; *sp != NULL; sp++) {
6695 		size_t would;
6696 		would = strlcat(o, *sp, buflen);
6697 		VERIFY3U(would, <, buflen);
6698 		if (*(sp+1) == NULL) {
6699 			break;
6700 		}
6701 		would = strlcat(o, sep, buflen);
6702 		VERIFY3U(would, <, buflen);
6703 	}
6704 	ASSERT3S(strlen(o), ==, totallen);
6705 	return (o);
6706 }
6707 
6708 static int
6709 ztest_check_path(char *path)
6710 {
6711 	struct stat s;
6712 	/* return true on success */
6713 	return (!stat(path, &s));
6714 }
6715 
6716 static void
6717 ztest_get_zdb_bin(char *bin, int len)
6718 {
6719 	char *zdb_path;
6720 	/*
6721 	 * Try to use $ZDB and in-tree zdb path. If not successful, just
6722 	 * let popen to search through PATH.
6723 	 */
6724 	if ((zdb_path = getenv("ZDB"))) {
6725 		strlcpy(bin, zdb_path, len); /* In env */
6726 		if (!ztest_check_path(bin)) {
6727 			ztest_dump_core = 0;
6728 			fatal(B_TRUE, "invalid ZDB '%s'", bin);
6729 		}
6730 		return;
6731 	}
6732 
6733 	VERIFY3P(realpath(getexecname(), bin), !=, NULL);
6734 	if (strstr(bin, ".libs/ztest")) {
6735 		strstr(bin, ".libs/ztest")[0] = '\0'; /* In-tree */
6736 		strcat(bin, "zdb");
6737 		if (ztest_check_path(bin))
6738 			return;
6739 	}
6740 	strcpy(bin, "zdb");
6741 }
6742 
6743 static vdev_t *
6744 ztest_random_concrete_vdev_leaf(vdev_t *vd)
6745 {
6746 	if (vd == NULL)
6747 		return (NULL);
6748 
6749 	if (vd->vdev_children == 0)
6750 		return (vd);
6751 
6752 	vdev_t *eligible[vd->vdev_children];
6753 	int eligible_idx = 0, i;
6754 	for (i = 0; i < vd->vdev_children; i++) {
6755 		vdev_t *cvd = vd->vdev_child[i];
6756 		if (cvd->vdev_top->vdev_removing)
6757 			continue;
6758 		if (cvd->vdev_children > 0 ||
6759 		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
6760 			eligible[eligible_idx++] = cvd;
6761 		}
6762 	}
6763 	VERIFY3S(eligible_idx, >, 0);
6764 
6765 	uint64_t child_no = ztest_random(eligible_idx);
6766 	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
6767 }
6768 
6769 void
6770 ztest_initialize(ztest_ds_t *zd, uint64_t id)
6771 {
6772 	(void) zd, (void) id;
6773 	spa_t *spa = ztest_spa;
6774 	int error = 0;
6775 
6776 	mutex_enter(&ztest_vdev_lock);
6777 
6778 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6779 
6780 	/* Random leaf vdev */
6781 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6782 	if (rand_vd == NULL) {
6783 		spa_config_exit(spa, SCL_VDEV, FTAG);
6784 		mutex_exit(&ztest_vdev_lock);
6785 		return;
6786 	}
6787 
6788 	/*
6789 	 * The random vdev we've selected may change as soon as we
6790 	 * drop the spa_config_lock. We create local copies of things
6791 	 * we're interested in.
6792 	 */
6793 	uint64_t guid = rand_vd->vdev_guid;
6794 	char *path = strdup(rand_vd->vdev_path);
6795 	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
6796 
6797 	zfs_dbgmsg("vd %px, guid %llu", rand_vd, (u_longlong_t)guid);
6798 	spa_config_exit(spa, SCL_VDEV, FTAG);
6799 
6800 	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
6801 
6802 	nvlist_t *vdev_guids = fnvlist_alloc();
6803 	nvlist_t *vdev_errlist = fnvlist_alloc();
6804 	fnvlist_add_uint64(vdev_guids, path, guid);
6805 	error = spa_vdev_initialize(spa, vdev_guids, cmd, vdev_errlist);
6806 	fnvlist_free(vdev_guids);
6807 	fnvlist_free(vdev_errlist);
6808 
6809 	switch (cmd) {
6810 	case POOL_INITIALIZE_CANCEL:
6811 		if (ztest_opts.zo_verbose >= 4) {
6812 			(void) printf("Cancel initialize %s", path);
6813 			if (!active)
6814 				(void) printf(" failed (no initialize active)");
6815 			(void) printf("\n");
6816 		}
6817 		break;
6818 	case POOL_INITIALIZE_START:
6819 		if (ztest_opts.zo_verbose >= 4) {
6820 			(void) printf("Start initialize %s", path);
6821 			if (active && error == 0)
6822 				(void) printf(" failed (already active)");
6823 			else if (error != 0)
6824 				(void) printf(" failed (error %d)", error);
6825 			(void) printf("\n");
6826 		}
6827 		break;
6828 	case POOL_INITIALIZE_SUSPEND:
6829 		if (ztest_opts.zo_verbose >= 4) {
6830 			(void) printf("Suspend initialize %s", path);
6831 			if (!active)
6832 				(void) printf(" failed (no initialize active)");
6833 			(void) printf("\n");
6834 		}
6835 		break;
6836 	}
6837 	free(path);
6838 	mutex_exit(&ztest_vdev_lock);
6839 }
6840 
6841 void
6842 ztest_trim(ztest_ds_t *zd, uint64_t id)
6843 {
6844 	(void) zd, (void) id;
6845 	spa_t *spa = ztest_spa;
6846 	int error = 0;
6847 
6848 	mutex_enter(&ztest_vdev_lock);
6849 
6850 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6851 
6852 	/* Random leaf vdev */
6853 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6854 	if (rand_vd == NULL) {
6855 		spa_config_exit(spa, SCL_VDEV, FTAG);
6856 		mutex_exit(&ztest_vdev_lock);
6857 		return;
6858 	}
6859 
6860 	/*
6861 	 * The random vdev we've selected may change as soon as we
6862 	 * drop the spa_config_lock. We create local copies of things
6863 	 * we're interested in.
6864 	 */
6865 	uint64_t guid = rand_vd->vdev_guid;
6866 	char *path = strdup(rand_vd->vdev_path);
6867 	boolean_t active = rand_vd->vdev_trim_thread != NULL;
6868 
6869 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, (u_longlong_t)guid);
6870 	spa_config_exit(spa, SCL_VDEV, FTAG);
6871 
6872 	uint64_t cmd = ztest_random(POOL_TRIM_FUNCS);
6873 	uint64_t rate = 1 << ztest_random(30);
6874 	boolean_t partial = (ztest_random(5) > 0);
6875 	boolean_t secure = (ztest_random(5) > 0);
6876 
6877 	nvlist_t *vdev_guids = fnvlist_alloc();
6878 	nvlist_t *vdev_errlist = fnvlist_alloc();
6879 	fnvlist_add_uint64(vdev_guids, path, guid);
6880 	error = spa_vdev_trim(spa, vdev_guids, cmd, rate, partial,
6881 	    secure, vdev_errlist);
6882 	fnvlist_free(vdev_guids);
6883 	fnvlist_free(vdev_errlist);
6884 
6885 	switch (cmd) {
6886 	case POOL_TRIM_CANCEL:
6887 		if (ztest_opts.zo_verbose >= 4) {
6888 			(void) printf("Cancel TRIM %s", path);
6889 			if (!active)
6890 				(void) printf(" failed (no TRIM active)");
6891 			(void) printf("\n");
6892 		}
6893 		break;
6894 	case POOL_TRIM_START:
6895 		if (ztest_opts.zo_verbose >= 4) {
6896 			(void) printf("Start TRIM %s", path);
6897 			if (active && error == 0)
6898 				(void) printf(" failed (already active)");
6899 			else if (error != 0)
6900 				(void) printf(" failed (error %d)", error);
6901 			(void) printf("\n");
6902 		}
6903 		break;
6904 	case POOL_TRIM_SUSPEND:
6905 		if (ztest_opts.zo_verbose >= 4) {
6906 			(void) printf("Suspend TRIM %s", path);
6907 			if (!active)
6908 				(void) printf(" failed (no TRIM active)");
6909 			(void) printf("\n");
6910 		}
6911 		break;
6912 	}
6913 	free(path);
6914 	mutex_exit(&ztest_vdev_lock);
6915 }
6916 
6917 /*
6918  * Verify pool integrity by running zdb.
6919  */
6920 static void
6921 ztest_run_zdb(uint64_t guid)
6922 {
6923 	int status;
6924 	char *bin;
6925 	char *zdb;
6926 	char *zbuf;
6927 	const int len = MAXPATHLEN + MAXNAMELEN + 20;
6928 	FILE *fp;
6929 
6930 	bin = umem_alloc(len, UMEM_NOFAIL);
6931 	zdb = umem_alloc(len, UMEM_NOFAIL);
6932 	zbuf = umem_alloc(1024, UMEM_NOFAIL);
6933 
6934 	ztest_get_zdb_bin(bin, len);
6935 
6936 	char **set_gvars_args = ztest_global_vars_to_zdb_args();
6937 	if (set_gvars_args == NULL) {
6938 		fatal(B_FALSE, "Failed to allocate memory in "
6939 		    "ztest_global_vars_to_zdb_args(). Cannot run zdb.\n");
6940 	}
6941 	char *set_gvars_args_joined = join_strings(set_gvars_args, " ");
6942 	free(set_gvars_args);
6943 
6944 	size_t would = snprintf(zdb, len,
6945 	    "%s -bcc%s%s -G -d -Y -e -y %s -p %s %"PRIu64,
6946 	    bin,
6947 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
6948 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
6949 	    set_gvars_args_joined,
6950 	    ztest_opts.zo_dir,
6951 	    guid);
6952 	ASSERT3U(would, <, len);
6953 
6954 	umem_free(set_gvars_args_joined, strlen(set_gvars_args_joined) + 1);
6955 
6956 	if (ztest_opts.zo_verbose >= 5)
6957 		(void) printf("Executing %s\n", zdb);
6958 
6959 	fp = popen(zdb, "r");
6960 
6961 	while (fgets(zbuf, 1024, fp) != NULL)
6962 		if (ztest_opts.zo_verbose >= 3)
6963 			(void) printf("%s", zbuf);
6964 
6965 	status = pclose(fp);
6966 
6967 	if (status == 0)
6968 		goto out;
6969 
6970 	ztest_dump_core = 0;
6971 	if (WIFEXITED(status))
6972 		fatal(B_FALSE, "'%s' exit code %d", zdb, WEXITSTATUS(status));
6973 	else
6974 		fatal(B_FALSE, "'%s' died with signal %d",
6975 		    zdb, WTERMSIG(status));
6976 out:
6977 	umem_free(bin, len);
6978 	umem_free(zdb, len);
6979 	umem_free(zbuf, 1024);
6980 }
6981 
6982 static void
6983 ztest_walk_pool_directory(const char *header)
6984 {
6985 	spa_t *spa = NULL;
6986 
6987 	if (ztest_opts.zo_verbose >= 6)
6988 		(void) puts(header);
6989 
6990 	mutex_enter(&spa_namespace_lock);
6991 	while ((spa = spa_next(spa)) != NULL)
6992 		if (ztest_opts.zo_verbose >= 6)
6993 			(void) printf("\t%s\n", spa_name(spa));
6994 	mutex_exit(&spa_namespace_lock);
6995 }
6996 
6997 static void
6998 ztest_spa_import_export(char *oldname, char *newname)
6999 {
7000 	nvlist_t *config, *newconfig;
7001 	uint64_t pool_guid;
7002 	spa_t *spa;
7003 	int error;
7004 
7005 	if (ztest_opts.zo_verbose >= 4) {
7006 		(void) printf("import/export: old = %s, new = %s\n",
7007 		    oldname, newname);
7008 	}
7009 
7010 	/*
7011 	 * Clean up from previous runs.
7012 	 */
7013 	(void) spa_destroy(newname);
7014 
7015 	/*
7016 	 * Get the pool's configuration and guid.
7017 	 */
7018 	VERIFY0(spa_open(oldname, &spa, FTAG));
7019 
7020 	/*
7021 	 * Kick off a scrub to tickle scrub/export races.
7022 	 */
7023 	if (ztest_random(2) == 0)
7024 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
7025 
7026 	pool_guid = spa_guid(spa);
7027 	spa_close(spa, FTAG);
7028 
7029 	ztest_walk_pool_directory("pools before export");
7030 
7031 	/*
7032 	 * Export it.
7033 	 */
7034 	VERIFY0(spa_export(oldname, &config, B_FALSE, B_FALSE));
7035 
7036 	ztest_walk_pool_directory("pools after export");
7037 
7038 	/*
7039 	 * Try to import it.
7040 	 */
7041 	newconfig = spa_tryimport(config);
7042 	ASSERT3P(newconfig, !=, NULL);
7043 	fnvlist_free(newconfig);
7044 
7045 	/*
7046 	 * Import it under the new name.
7047 	 */
7048 	error = spa_import(newname, config, NULL, 0);
7049 	if (error != 0) {
7050 		dump_nvlist(config, 0);
7051 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
7052 		    oldname, newname, error);
7053 	}
7054 
7055 	ztest_walk_pool_directory("pools after import");
7056 
7057 	/*
7058 	 * Try to import it again -- should fail with EEXIST.
7059 	 */
7060 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
7061 
7062 	/*
7063 	 * Try to import it under a different name -- should fail with EEXIST.
7064 	 */
7065 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
7066 
7067 	/*
7068 	 * Verify that the pool is no longer visible under the old name.
7069 	 */
7070 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
7071 
7072 	/*
7073 	 * Verify that we can open and close the pool using the new name.
7074 	 */
7075 	VERIFY0(spa_open(newname, &spa, FTAG));
7076 	ASSERT3U(pool_guid, ==, spa_guid(spa));
7077 	spa_close(spa, FTAG);
7078 
7079 	fnvlist_free(config);
7080 }
7081 
7082 static void
7083 ztest_resume(spa_t *spa)
7084 {
7085 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
7086 		(void) printf("resuming from suspended state\n");
7087 	spa_vdev_state_enter(spa, SCL_NONE);
7088 	vdev_clear(spa, NULL);
7089 	(void) spa_vdev_state_exit(spa, NULL, 0);
7090 	(void) zio_resume(spa);
7091 }
7092 
7093 static __attribute__((noreturn)) void
7094 ztest_resume_thread(void *arg)
7095 {
7096 	spa_t *spa = arg;
7097 
7098 	while (!ztest_exiting) {
7099 		if (spa_suspended(spa))
7100 			ztest_resume(spa);
7101 		(void) poll(NULL, 0, 100);
7102 
7103 		/*
7104 		 * Periodically change the zfs_compressed_arc_enabled setting.
7105 		 */
7106 		if (ztest_random(10) == 0)
7107 			zfs_compressed_arc_enabled = ztest_random(2);
7108 
7109 		/*
7110 		 * Periodically change the zfs_abd_scatter_enabled setting.
7111 		 */
7112 		if (ztest_random(10) == 0)
7113 			zfs_abd_scatter_enabled = ztest_random(2);
7114 	}
7115 
7116 	thread_exit();
7117 }
7118 
7119 static __attribute__((noreturn)) void
7120 ztest_deadman_thread(void *arg)
7121 {
7122 	ztest_shared_t *zs = arg;
7123 	spa_t *spa = ztest_spa;
7124 	hrtime_t delay, overdue, last_run = gethrtime();
7125 
7126 	delay = (zs->zs_thread_stop - zs->zs_thread_start) +
7127 	    MSEC2NSEC(zfs_deadman_synctime_ms);
7128 
7129 	while (!ztest_exiting) {
7130 		/*
7131 		 * Wait for the delay timer while checking occasionally
7132 		 * if we should stop.
7133 		 */
7134 		if (gethrtime() < last_run + delay) {
7135 			(void) poll(NULL, 0, 1000);
7136 			continue;
7137 		}
7138 
7139 		/*
7140 		 * If the pool is suspended then fail immediately. Otherwise,
7141 		 * check to see if the pool is making any progress. If
7142 		 * vdev_deadman() discovers that there hasn't been any recent
7143 		 * I/Os then it will end up aborting the tests.
7144 		 */
7145 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
7146 			fatal(B_FALSE,
7147 			    "aborting test after %llu seconds because "
7148 			    "pool has transitioned to a suspended state.",
7149 			    (u_longlong_t)zfs_deadman_synctime_ms / 1000);
7150 		}
7151 		vdev_deadman(spa->spa_root_vdev, FTAG);
7152 
7153 		/*
7154 		 * If the process doesn't complete within a grace period of
7155 		 * zfs_deadman_synctime_ms over the expected finish time,
7156 		 * then it may be hung and is terminated.
7157 		 */
7158 		overdue = zs->zs_proc_stop + MSEC2NSEC(zfs_deadman_synctime_ms);
7159 		if (gethrtime() > overdue) {
7160 			fatal(B_FALSE,
7161 			    "aborting test after %llu seconds because "
7162 			    "the process is overdue for termination.",
7163 			    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7164 		}
7165 
7166 		(void) printf("ztest has been running for %lld seconds\n",
7167 		    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7168 
7169 		last_run = gethrtime();
7170 		delay = MSEC2NSEC(zfs_deadman_checktime_ms);
7171 	}
7172 
7173 	thread_exit();
7174 }
7175 
7176 static void
7177 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
7178 {
7179 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
7180 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
7181 	hrtime_t functime = gethrtime();
7182 	int i;
7183 
7184 	for (i = 0; i < zi->zi_iters; i++)
7185 		zi->zi_func(zd, id);
7186 
7187 	functime = gethrtime() - functime;
7188 
7189 	atomic_add_64(&zc->zc_count, 1);
7190 	atomic_add_64(&zc->zc_time, functime);
7191 
7192 	if (ztest_opts.zo_verbose >= 4)
7193 		(void) printf("%6.2f sec in %s\n",
7194 		    (double)functime / NANOSEC, zi->zi_funcname);
7195 }
7196 
7197 static __attribute__((noreturn)) void
7198 ztest_thread(void *arg)
7199 {
7200 	int rand;
7201 	uint64_t id = (uintptr_t)arg;
7202 	ztest_shared_t *zs = ztest_shared;
7203 	uint64_t call_next;
7204 	hrtime_t now;
7205 	ztest_info_t *zi;
7206 	ztest_shared_callstate_t *zc;
7207 
7208 	while ((now = gethrtime()) < zs->zs_thread_stop) {
7209 		/*
7210 		 * See if it's time to force a crash.
7211 		 */
7212 		if (now > zs->zs_thread_kill)
7213 			ztest_kill(zs);
7214 
7215 		/*
7216 		 * If we're getting ENOSPC with some regularity, stop.
7217 		 */
7218 		if (zs->zs_enospc_count > 10)
7219 			break;
7220 
7221 		/*
7222 		 * Pick a random function to execute.
7223 		 */
7224 		rand = ztest_random(ZTEST_FUNCS);
7225 		zi = &ztest_info[rand];
7226 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
7227 		call_next = zc->zc_next;
7228 
7229 		if (now >= call_next &&
7230 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
7231 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
7232 			ztest_execute(rand, zi, id);
7233 		}
7234 	}
7235 
7236 	thread_exit();
7237 }
7238 
7239 static void
7240 ztest_dataset_name(char *dsname, const char *pool, int d)
7241 {
7242 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
7243 }
7244 
7245 static void
7246 ztest_dataset_destroy(int d)
7247 {
7248 	char name[ZFS_MAX_DATASET_NAME_LEN];
7249 	int t;
7250 
7251 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7252 
7253 	if (ztest_opts.zo_verbose >= 3)
7254 		(void) printf("Destroying %s to free up space\n", name);
7255 
7256 	/*
7257 	 * Cleanup any non-standard clones and snapshots.  In general,
7258 	 * ztest thread t operates on dataset (t % zopt_datasets),
7259 	 * so there may be more than one thing to clean up.
7260 	 */
7261 	for (t = d; t < ztest_opts.zo_threads;
7262 	    t += ztest_opts.zo_datasets)
7263 		ztest_dsl_dataset_cleanup(name, t);
7264 
7265 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
7266 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
7267 }
7268 
7269 static void
7270 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
7271 {
7272 	uint64_t usedobjs, dirobjs, scratch;
7273 
7274 	/*
7275 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
7276 	 * Therefore, the number of objects in use should equal the
7277 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
7278 	 * If not, we have an object leak.
7279 	 *
7280 	 * Note that we can only check this in ztest_dataset_open(),
7281 	 * when the open-context and syncing-context values agree.
7282 	 * That's because zap_count() returns the open-context value,
7283 	 * while dmu_objset_space() returns the rootbp fill count.
7284 	 */
7285 	VERIFY0(zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
7286 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
7287 	ASSERT3U(dirobjs + 1, ==, usedobjs);
7288 }
7289 
7290 static int
7291 ztest_dataset_open(int d)
7292 {
7293 	ztest_ds_t *zd = &ztest_ds[d];
7294 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
7295 	objset_t *os;
7296 	zilog_t *zilog;
7297 	char name[ZFS_MAX_DATASET_NAME_LEN];
7298 	int error;
7299 
7300 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7301 
7302 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
7303 
7304 	error = ztest_dataset_create(name);
7305 	if (error == ENOSPC) {
7306 		(void) pthread_rwlock_unlock(&ztest_name_lock);
7307 		ztest_record_enospc(FTAG);
7308 		return (error);
7309 	}
7310 	ASSERT(error == 0 || error == EEXIST);
7311 
7312 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
7313 	    B_TRUE, zd, &os));
7314 	(void) pthread_rwlock_unlock(&ztest_name_lock);
7315 
7316 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
7317 
7318 	zilog = zd->zd_zilog;
7319 
7320 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
7321 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
7322 		fatal(B_FALSE, "missing log records: "
7323 		    "claimed %"PRIu64" < committed %"PRIu64"",
7324 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
7325 
7326 	ztest_dataset_dirobj_verify(zd);
7327 
7328 	zil_replay(os, zd, ztest_replay_vector);
7329 
7330 	ztest_dataset_dirobj_verify(zd);
7331 
7332 	if (ztest_opts.zo_verbose >= 6)
7333 		(void) printf("%s replay %"PRIu64" blocks, "
7334 		    "%"PRIu64" records, seq %"PRIu64"\n",
7335 		    zd->zd_name,
7336 		    zilog->zl_parse_blk_count,
7337 		    zilog->zl_parse_lr_count,
7338 		    zilog->zl_replaying_seq);
7339 
7340 	zilog = zil_open(os, ztest_get_data, NULL);
7341 
7342 	if (zilog->zl_replaying_seq != 0 &&
7343 	    zilog->zl_replaying_seq < committed_seq)
7344 		fatal(B_FALSE, "missing log records: "
7345 		    "replayed %"PRIu64" < committed %"PRIu64"",
7346 		    zilog->zl_replaying_seq, committed_seq);
7347 
7348 	return (0);
7349 }
7350 
7351 static void
7352 ztest_dataset_close(int d)
7353 {
7354 	ztest_ds_t *zd = &ztest_ds[d];
7355 
7356 	zil_close(zd->zd_zilog);
7357 	dmu_objset_disown(zd->zd_os, B_TRUE, zd);
7358 
7359 	ztest_zd_fini(zd);
7360 }
7361 
7362 static int
7363 ztest_replay_zil_cb(const char *name, void *arg)
7364 {
7365 	(void) arg;
7366 	objset_t *os;
7367 	ztest_ds_t *zdtmp;
7368 
7369 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_ANY, B_TRUE,
7370 	    B_TRUE, FTAG, &os));
7371 
7372 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
7373 
7374 	ztest_zd_init(zdtmp, NULL, os);
7375 	zil_replay(os, zdtmp, ztest_replay_vector);
7376 	ztest_zd_fini(zdtmp);
7377 
7378 	if (dmu_objset_zil(os)->zl_parse_lr_count != 0 &&
7379 	    ztest_opts.zo_verbose >= 6) {
7380 		zilog_t *zilog = dmu_objset_zil(os);
7381 
7382 		(void) printf("%s replay %"PRIu64" blocks, "
7383 		    "%"PRIu64" records, seq %"PRIu64"\n",
7384 		    name,
7385 		    zilog->zl_parse_blk_count,
7386 		    zilog->zl_parse_lr_count,
7387 		    zilog->zl_replaying_seq);
7388 	}
7389 
7390 	umem_free(zdtmp, sizeof (ztest_ds_t));
7391 
7392 	dmu_objset_disown(os, B_TRUE, FTAG);
7393 	return (0);
7394 }
7395 
7396 static void
7397 ztest_freeze(void)
7398 {
7399 	ztest_ds_t *zd = &ztest_ds[0];
7400 	spa_t *spa;
7401 	int numloops = 0;
7402 
7403 	if (ztest_opts.zo_verbose >= 3)
7404 		(void) printf("testing spa_freeze()...\n");
7405 
7406 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7407 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7408 	VERIFY0(ztest_dataset_open(0));
7409 	ztest_spa = spa;
7410 
7411 	/*
7412 	 * Force the first log block to be transactionally allocated.
7413 	 * We have to do this before we freeze the pool -- otherwise
7414 	 * the log chain won't be anchored.
7415 	 */
7416 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
7417 		ztest_dmu_object_alloc_free(zd, 0);
7418 		zil_commit(zd->zd_zilog, 0);
7419 	}
7420 
7421 	txg_wait_synced(spa_get_dsl(spa), 0);
7422 
7423 	/*
7424 	 * Freeze the pool.  This stops spa_sync() from doing anything,
7425 	 * so that the only way to record changes from now on is the ZIL.
7426 	 */
7427 	spa_freeze(spa);
7428 
7429 	/*
7430 	 * Because it is hard to predict how much space a write will actually
7431 	 * require beforehand, we leave ourselves some fudge space to write over
7432 	 * capacity.
7433 	 */
7434 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
7435 
7436 	/*
7437 	 * Run tests that generate log records but don't alter the pool config
7438 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
7439 	 * We do a txg_wait_synced() after each iteration to force the txg
7440 	 * to increase well beyond the last synced value in the uberblock.
7441 	 * The ZIL should be OK with that.
7442 	 *
7443 	 * Run a random number of times less than zo_maxloops and ensure we do
7444 	 * not run out of space on the pool.
7445 	 */
7446 	while (ztest_random(10) != 0 &&
7447 	    numloops++ < ztest_opts.zo_maxloops &&
7448 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
7449 		ztest_od_t od;
7450 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
7451 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
7452 		ztest_io(zd, od.od_object,
7453 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
7454 		txg_wait_synced(spa_get_dsl(spa), 0);
7455 	}
7456 
7457 	/*
7458 	 * Commit all of the changes we just generated.
7459 	 */
7460 	zil_commit(zd->zd_zilog, 0);
7461 	txg_wait_synced(spa_get_dsl(spa), 0);
7462 
7463 	/*
7464 	 * Close our dataset and close the pool.
7465 	 */
7466 	ztest_dataset_close(0);
7467 	spa_close(spa, FTAG);
7468 	kernel_fini();
7469 
7470 	/*
7471 	 * Open and close the pool and dataset to induce log replay.
7472 	 */
7473 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7474 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7475 	ASSERT3U(spa_freeze_txg(spa), ==, UINT64_MAX);
7476 	VERIFY0(ztest_dataset_open(0));
7477 	ztest_spa = spa;
7478 	txg_wait_synced(spa_get_dsl(spa), 0);
7479 	ztest_dataset_close(0);
7480 	ztest_reguid(NULL, 0);
7481 
7482 	spa_close(spa, FTAG);
7483 	kernel_fini();
7484 }
7485 
7486 static void
7487 ztest_import_impl(void)
7488 {
7489 	importargs_t args = { 0 };
7490 	nvlist_t *cfg = NULL;
7491 	int nsearch = 1;
7492 	char *searchdirs[nsearch];
7493 	int flags = ZFS_IMPORT_MISSING_LOG;
7494 
7495 	searchdirs[0] = ztest_opts.zo_dir;
7496 	args.paths = nsearch;
7497 	args.path = searchdirs;
7498 	args.can_be_active = B_FALSE;
7499 
7500 	libpc_handle_t lpch = {
7501 		.lpc_lib_handle = NULL,
7502 		.lpc_ops = &libzpool_config_ops,
7503 		.lpc_printerr = B_TRUE
7504 	};
7505 	VERIFY0(zpool_find_config(&lpch, ztest_opts.zo_pool, &cfg, &args));
7506 	VERIFY0(spa_import(ztest_opts.zo_pool, cfg, NULL, flags));
7507 	fnvlist_free(cfg);
7508 }
7509 
7510 /*
7511  * Import a storage pool with the given name.
7512  */
7513 static void
7514 ztest_import(ztest_shared_t *zs)
7515 {
7516 	spa_t *spa;
7517 
7518 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7519 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7520 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7521 
7522 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7523 
7524 	ztest_import_impl();
7525 
7526 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7527 	zs->zs_metaslab_sz =
7528 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7529 	zs->zs_guid = spa_guid(spa);
7530 	spa_close(spa, FTAG);
7531 
7532 	kernel_fini();
7533 
7534 	if (!ztest_opts.zo_mmp_test) {
7535 		ztest_run_zdb(zs->zs_guid);
7536 		ztest_freeze();
7537 		ztest_run_zdb(zs->zs_guid);
7538 	}
7539 
7540 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7541 	mutex_destroy(&ztest_vdev_lock);
7542 	mutex_destroy(&ztest_checkpoint_lock);
7543 }
7544 
7545 /*
7546  * Kick off threads to run tests on all datasets in parallel.
7547  */
7548 static void
7549 ztest_run(ztest_shared_t *zs)
7550 {
7551 	spa_t *spa;
7552 	objset_t *os;
7553 	kthread_t *resume_thread, *deadman_thread;
7554 	kthread_t **run_threads;
7555 	uint64_t object;
7556 	int error;
7557 	int t, d;
7558 
7559 	ztest_exiting = B_FALSE;
7560 
7561 	/*
7562 	 * Initialize parent/child shared state.
7563 	 */
7564 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7565 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7566 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7567 
7568 	zs->zs_thread_start = gethrtime();
7569 	zs->zs_thread_stop =
7570 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
7571 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
7572 	zs->zs_thread_kill = zs->zs_thread_stop;
7573 	if (ztest_random(100) < ztest_opts.zo_killrate) {
7574 		zs->zs_thread_kill -=
7575 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
7576 	}
7577 
7578 	mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
7579 
7580 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
7581 	    offsetof(ztest_cb_data_t, zcd_node));
7582 
7583 	/*
7584 	 * Open our pool.  It may need to be imported first depending on
7585 	 * what tests were running when the previous pass was terminated.
7586 	 */
7587 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7588 	error = spa_open(ztest_opts.zo_pool, &spa, FTAG);
7589 	if (error) {
7590 		VERIFY3S(error, ==, ENOENT);
7591 		ztest_import_impl();
7592 		VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7593 		zs->zs_metaslab_sz =
7594 		    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7595 	}
7596 
7597 	metaslab_preload_limit = ztest_random(20) + 1;
7598 	ztest_spa = spa;
7599 
7600 	VERIFY0(vdev_raidz_impl_set("cycle"));
7601 
7602 	dmu_objset_stats_t dds;
7603 	VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
7604 	    DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
7605 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
7606 	dmu_objset_fast_stat(os, &dds);
7607 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
7608 	dmu_objset_disown(os, B_TRUE, FTAG);
7609 
7610 	/*
7611 	 * Create a thread to periodically resume suspended I/O.
7612 	 */
7613 	resume_thread = thread_create(NULL, 0, ztest_resume_thread,
7614 	    spa, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
7615 
7616 	/*
7617 	 * Create a deadman thread and set to panic if we hang.
7618 	 */
7619 	deadman_thread = thread_create(NULL, 0, ztest_deadman_thread,
7620 	    zs, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
7621 
7622 	spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
7623 
7624 	/*
7625 	 * Verify that we can safely inquire about any object,
7626 	 * whether it's allocated or not.  To make it interesting,
7627 	 * we probe a 5-wide window around each power of two.
7628 	 * This hits all edge cases, including zero and the max.
7629 	 */
7630 	for (t = 0; t < 64; t++) {
7631 		for (d = -5; d <= 5; d++) {
7632 			error = dmu_object_info(spa->spa_meta_objset,
7633 			    (1ULL << t) + d, NULL);
7634 			ASSERT(error == 0 || error == ENOENT ||
7635 			    error == EINVAL);
7636 		}
7637 	}
7638 
7639 	/*
7640 	 * If we got any ENOSPC errors on the previous run, destroy something.
7641 	 */
7642 	if (zs->zs_enospc_count != 0) {
7643 		int d = ztest_random(ztest_opts.zo_datasets);
7644 		ztest_dataset_destroy(d);
7645 	}
7646 	zs->zs_enospc_count = 0;
7647 
7648 	/*
7649 	 * If we were in the middle of ztest_device_removal() and were killed
7650 	 * we need to ensure the removal and scrub complete before running
7651 	 * any tests that check ztest_device_removal_active. The removal will
7652 	 * be restarted automatically when the spa is opened, but we need to
7653 	 * initiate the scrub manually if it is not already in progress. Note
7654 	 * that we always run the scrub whenever an indirect vdev exists
7655 	 * because we have no way of knowing for sure if ztest_device_removal()
7656 	 * fully completed its scrub before the pool was reimported.
7657 	 */
7658 	if (spa->spa_removing_phys.sr_state == DSS_SCANNING ||
7659 	    spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
7660 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
7661 			txg_wait_synced(spa_get_dsl(spa), 0);
7662 
7663 		error = ztest_scrub_impl(spa);
7664 		if (error == EBUSY)
7665 			error = 0;
7666 		ASSERT0(error);
7667 	}
7668 
7669 	run_threads = umem_zalloc(ztest_opts.zo_threads * sizeof (kthread_t *),
7670 	    UMEM_NOFAIL);
7671 
7672 	if (ztest_opts.zo_verbose >= 4)
7673 		(void) printf("starting main threads...\n");
7674 
7675 	/*
7676 	 * Replay all logs of all datasets in the pool. This is primarily for
7677 	 * temporary datasets which wouldn't otherwise get replayed, which
7678 	 * can trigger failures when attempting to offline a SLOG in
7679 	 * ztest_fault_inject().
7680 	 */
7681 	(void) dmu_objset_find(ztest_opts.zo_pool, ztest_replay_zil_cb,
7682 	    NULL, DS_FIND_CHILDREN);
7683 
7684 	/*
7685 	 * Kick off all the tests that run in parallel.
7686 	 */
7687 	for (t = 0; t < ztest_opts.zo_threads; t++) {
7688 		if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
7689 			umem_free(run_threads, ztest_opts.zo_threads *
7690 			    sizeof (kthread_t *));
7691 			return;
7692 		}
7693 
7694 		run_threads[t] = thread_create(NULL, 0, ztest_thread,
7695 		    (void *)(uintptr_t)t, 0, NULL, TS_RUN | TS_JOINABLE,
7696 		    defclsyspri);
7697 	}
7698 
7699 	/*
7700 	 * Wait for all of the tests to complete.
7701 	 */
7702 	for (t = 0; t < ztest_opts.zo_threads; t++)
7703 		VERIFY0(thread_join(run_threads[t]));
7704 
7705 	/*
7706 	 * Close all datasets. This must be done after all the threads
7707 	 * are joined so we can be sure none of the datasets are in-use
7708 	 * by any of the threads.
7709 	 */
7710 	for (t = 0; t < ztest_opts.zo_threads; t++) {
7711 		if (t < ztest_opts.zo_datasets)
7712 			ztest_dataset_close(t);
7713 	}
7714 
7715 	txg_wait_synced(spa_get_dsl(spa), 0);
7716 
7717 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7718 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
7719 
7720 	umem_free(run_threads, ztest_opts.zo_threads * sizeof (kthread_t *));
7721 
7722 	/* Kill the resume and deadman threads */
7723 	ztest_exiting = B_TRUE;
7724 	VERIFY0(thread_join(resume_thread));
7725 	VERIFY0(thread_join(deadman_thread));
7726 	ztest_resume(spa);
7727 
7728 	/*
7729 	 * Right before closing the pool, kick off a bunch of async I/O;
7730 	 * spa_close() should wait for it to complete.
7731 	 */
7732 	for (object = 1; object < 50; object++) {
7733 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
7734 		    ZIO_PRIORITY_SYNC_READ);
7735 	}
7736 
7737 	/* Verify that at least one commit cb was called in a timely fashion */
7738 	if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
7739 		VERIFY0(zc_min_txg_delay);
7740 
7741 	spa_close(spa, FTAG);
7742 
7743 	/*
7744 	 * Verify that we can loop over all pools.
7745 	 */
7746 	mutex_enter(&spa_namespace_lock);
7747 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
7748 		if (ztest_opts.zo_verbose > 3)
7749 			(void) printf("spa_next: found %s\n", spa_name(spa));
7750 	mutex_exit(&spa_namespace_lock);
7751 
7752 	/*
7753 	 * Verify that we can export the pool and reimport it under a
7754 	 * different name.
7755 	 */
7756 	if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
7757 		char name[ZFS_MAX_DATASET_NAME_LEN];
7758 		(void) snprintf(name, sizeof (name), "%s_import",
7759 		    ztest_opts.zo_pool);
7760 		ztest_spa_import_export(ztest_opts.zo_pool, name);
7761 		ztest_spa_import_export(name, ztest_opts.zo_pool);
7762 	}
7763 
7764 	kernel_fini();
7765 
7766 	list_destroy(&zcl.zcl_callbacks);
7767 	mutex_destroy(&zcl.zcl_callbacks_lock);
7768 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7769 	mutex_destroy(&ztest_vdev_lock);
7770 	mutex_destroy(&ztest_checkpoint_lock);
7771 }
7772 
7773 static void
7774 print_time(hrtime_t t, char *timebuf)
7775 {
7776 	hrtime_t s = t / NANOSEC;
7777 	hrtime_t m = s / 60;
7778 	hrtime_t h = m / 60;
7779 	hrtime_t d = h / 24;
7780 
7781 	s -= m * 60;
7782 	m -= h * 60;
7783 	h -= d * 24;
7784 
7785 	timebuf[0] = '\0';
7786 
7787 	if (d)
7788 		(void) sprintf(timebuf,
7789 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
7790 	else if (h)
7791 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
7792 	else if (m)
7793 		(void) sprintf(timebuf, "%llum%02llus", m, s);
7794 	else
7795 		(void) sprintf(timebuf, "%llus", s);
7796 }
7797 
7798 static nvlist_t *
7799 make_random_props(void)
7800 {
7801 	nvlist_t *props;
7802 
7803 	props = fnvlist_alloc();
7804 
7805 	if (ztest_random(2) == 0)
7806 		return (props);
7807 
7808 	fnvlist_add_uint64(props,
7809 	    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 1);
7810 
7811 	return (props);
7812 }
7813 
7814 /*
7815  * Create a storage pool with the given name and initial vdev size.
7816  * Then test spa_freeze() functionality.
7817  */
7818 static void
7819 ztest_init(ztest_shared_t *zs)
7820 {
7821 	spa_t *spa;
7822 	nvlist_t *nvroot, *props;
7823 	int i;
7824 
7825 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7826 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7827 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7828 
7829 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7830 
7831 	/*
7832 	 * Create the storage pool.
7833 	 */
7834 	(void) spa_destroy(ztest_opts.zo_pool);
7835 	ztest_shared->zs_vdev_next_leaf = 0;
7836 	zs->zs_splits = 0;
7837 	zs->zs_mirrors = ztest_opts.zo_mirrors;
7838 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
7839 	    NULL, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
7840 	props = make_random_props();
7841 
7842 	/*
7843 	 * We don't expect the pool to suspend unless maxfaults == 0,
7844 	 * in which case ztest_fault_inject() temporarily takes away
7845 	 * the only valid replica.
7846 	 */
7847 	fnvlist_add_uint64(props,
7848 	    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
7849 	    MAXFAULTS(zs) ? ZIO_FAILURE_MODE_PANIC : ZIO_FAILURE_MODE_WAIT);
7850 
7851 	for (i = 0; i < SPA_FEATURES; i++) {
7852 		char *buf;
7853 
7854 		if (!spa_feature_table[i].fi_zfs_mod_supported)
7855 			continue;
7856 
7857 		/*
7858 		 * 75% chance of using the log space map feature. We want ztest
7859 		 * to exercise both the code paths that use the log space map
7860 		 * feature and the ones that don't.
7861 		 */
7862 		if (i == SPA_FEATURE_LOG_SPACEMAP && ztest_random(4) == 0)
7863 			continue;
7864 
7865 		VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
7866 		    spa_feature_table[i].fi_uname));
7867 		fnvlist_add_uint64(props, buf, 0);
7868 		free(buf);
7869 	}
7870 
7871 	VERIFY0(spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
7872 	fnvlist_free(nvroot);
7873 	fnvlist_free(props);
7874 
7875 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7876 	zs->zs_metaslab_sz =
7877 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7878 	zs->zs_guid = spa_guid(spa);
7879 	spa_close(spa, FTAG);
7880 
7881 	kernel_fini();
7882 
7883 	if (!ztest_opts.zo_mmp_test) {
7884 		ztest_run_zdb(zs->zs_guid);
7885 		ztest_freeze();
7886 		ztest_run_zdb(zs->zs_guid);
7887 	}
7888 
7889 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7890 	mutex_destroy(&ztest_vdev_lock);
7891 	mutex_destroy(&ztest_checkpoint_lock);
7892 }
7893 
7894 static void
7895 setup_data_fd(void)
7896 {
7897 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
7898 
7899 	ztest_fd_data = mkstemp(ztest_name_data);
7900 	ASSERT3S(ztest_fd_data, >=, 0);
7901 	(void) unlink(ztest_name_data);
7902 }
7903 
7904 static int
7905 shared_data_size(ztest_shared_hdr_t *hdr)
7906 {
7907 	int size;
7908 
7909 	size = hdr->zh_hdr_size;
7910 	size += hdr->zh_opts_size;
7911 	size += hdr->zh_size;
7912 	size += hdr->zh_stats_size * hdr->zh_stats_count;
7913 	size += hdr->zh_ds_size * hdr->zh_ds_count;
7914 
7915 	return (size);
7916 }
7917 
7918 static void
7919 setup_hdr(void)
7920 {
7921 	int size;
7922 	ztest_shared_hdr_t *hdr;
7923 
7924 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7925 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7926 	ASSERT3P(hdr, !=, MAP_FAILED);
7927 
7928 	VERIFY0(ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
7929 
7930 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
7931 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
7932 	hdr->zh_size = sizeof (ztest_shared_t);
7933 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
7934 	hdr->zh_stats_count = ZTEST_FUNCS;
7935 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
7936 	hdr->zh_ds_count = ztest_opts.zo_datasets;
7937 
7938 	size = shared_data_size(hdr);
7939 	VERIFY0(ftruncate(ztest_fd_data, size));
7940 
7941 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7942 }
7943 
7944 static void
7945 setup_data(void)
7946 {
7947 	int size, offset;
7948 	ztest_shared_hdr_t *hdr;
7949 	uint8_t *buf;
7950 
7951 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7952 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
7953 	ASSERT3P(hdr, !=, MAP_FAILED);
7954 
7955 	size = shared_data_size(hdr);
7956 
7957 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7958 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
7959 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7960 	ASSERT3P(hdr, !=, MAP_FAILED);
7961 	buf = (uint8_t *)hdr;
7962 
7963 	offset = hdr->zh_hdr_size;
7964 	ztest_shared_opts = (void *)&buf[offset];
7965 	offset += hdr->zh_opts_size;
7966 	ztest_shared = (void *)&buf[offset];
7967 	offset += hdr->zh_size;
7968 	ztest_shared_callstate = (void *)&buf[offset];
7969 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
7970 	ztest_shared_ds = (void *)&buf[offset];
7971 }
7972 
7973 static boolean_t
7974 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
7975 {
7976 	pid_t pid;
7977 	int status;
7978 	char *cmdbuf = NULL;
7979 
7980 	pid = fork();
7981 
7982 	if (cmd == NULL) {
7983 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
7984 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
7985 		cmd = cmdbuf;
7986 	}
7987 
7988 	if (pid == -1)
7989 		fatal(B_TRUE, "fork failed");
7990 
7991 	if (pid == 0) {	/* child */
7992 		char fd_data_str[12];
7993 
7994 		VERIFY3S(11, >=,
7995 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
7996 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
7997 
7998 		if (libpath != NULL) {
7999 			const char *curlp = getenv("LD_LIBRARY_PATH");
8000 			if (curlp == NULL)
8001 				VERIFY0(setenv("LD_LIBRARY_PATH", libpath, 1));
8002 			else {
8003 				char *newlp = NULL;
8004 				VERIFY3S(-1, !=,
8005 				    asprintf(&newlp, "%s:%s", libpath, curlp));
8006 				VERIFY0(setenv("LD_LIBRARY_PATH", newlp, 1));
8007 				free(newlp);
8008 			}
8009 		}
8010 		(void) execl(cmd, cmd, (char *)NULL);
8011 		ztest_dump_core = B_FALSE;
8012 		fatal(B_TRUE, "exec failed: %s", cmd);
8013 	}
8014 
8015 	if (cmdbuf != NULL) {
8016 		umem_free(cmdbuf, MAXPATHLEN);
8017 		cmd = NULL;
8018 	}
8019 
8020 	while (waitpid(pid, &status, 0) != pid)
8021 		continue;
8022 	if (statusp != NULL)
8023 		*statusp = status;
8024 
8025 	if (WIFEXITED(status)) {
8026 		if (WEXITSTATUS(status) != 0) {
8027 			(void) fprintf(stderr, "child exited with code %d\n",
8028 			    WEXITSTATUS(status));
8029 			exit(2);
8030 		}
8031 		return (B_FALSE);
8032 	} else if (WIFSIGNALED(status)) {
8033 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
8034 			(void) fprintf(stderr, "child died with signal %d\n",
8035 			    WTERMSIG(status));
8036 			exit(3);
8037 		}
8038 		return (B_TRUE);
8039 	} else {
8040 		(void) fprintf(stderr, "something strange happened to child\n");
8041 		exit(4);
8042 	}
8043 }
8044 
8045 static void
8046 ztest_run_init(void)
8047 {
8048 	int i;
8049 
8050 	ztest_shared_t *zs = ztest_shared;
8051 
8052 	/*
8053 	 * Blow away any existing copy of zpool.cache
8054 	 */
8055 	(void) remove(spa_config_path);
8056 
8057 	if (ztest_opts.zo_init == 0) {
8058 		if (ztest_opts.zo_verbose >= 1)
8059 			(void) printf("Importing pool %s\n",
8060 			    ztest_opts.zo_pool);
8061 		ztest_import(zs);
8062 		return;
8063 	}
8064 
8065 	/*
8066 	 * Create and initialize our storage pool.
8067 	 */
8068 	for (i = 1; i <= ztest_opts.zo_init; i++) {
8069 		memset(zs, 0, sizeof (*zs));
8070 		if (ztest_opts.zo_verbose >= 3 &&
8071 		    ztest_opts.zo_init != 1) {
8072 			(void) printf("ztest_init(), pass %d\n", i);
8073 		}
8074 		ztest_init(zs);
8075 	}
8076 }
8077 
8078 int
8079 main(int argc, char **argv)
8080 {
8081 	int kills = 0;
8082 	int iters = 0;
8083 	int older = 0;
8084 	int newer = 0;
8085 	ztest_shared_t *zs;
8086 	ztest_info_t *zi;
8087 	ztest_shared_callstate_t *zc;
8088 	char timebuf[100];
8089 	char numbuf[NN_NUMBUF_SZ];
8090 	char *cmd;
8091 	boolean_t hasalt;
8092 	int f, err;
8093 	char *fd_data_str = getenv("ZTEST_FD_DATA");
8094 	struct sigaction action;
8095 
8096 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
8097 
8098 	dprintf_setup(&argc, argv);
8099 	zfs_deadman_synctime_ms = 300000;
8100 	zfs_deadman_checktime_ms = 30000;
8101 	/*
8102 	 * As two-word space map entries may not come up often (especially
8103 	 * if pool and vdev sizes are small) we want to force at least some
8104 	 * of them so the feature get tested.
8105 	 */
8106 	zfs_force_some_double_word_sm_entries = B_TRUE;
8107 
8108 	/*
8109 	 * Verify that even extensively damaged split blocks with many
8110 	 * segments can be reconstructed in a reasonable amount of time
8111 	 * when reconstruction is known to be possible.
8112 	 *
8113 	 * Note: the lower this value is, the more damage we inflict, and
8114 	 * the more time ztest spends in recovering that damage. We chose
8115 	 * to induce damage 1/100th of the time so recovery is tested but
8116 	 * not so frequently that ztest doesn't get to test other code paths.
8117 	 */
8118 	zfs_reconstruct_indirect_damage_fraction = 100;
8119 
8120 	action.sa_handler = sig_handler;
8121 	sigemptyset(&action.sa_mask);
8122 	action.sa_flags = 0;
8123 
8124 	if (sigaction(SIGSEGV, &action, NULL) < 0) {
8125 		(void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
8126 		    strerror(errno));
8127 		exit(EXIT_FAILURE);
8128 	}
8129 
8130 	if (sigaction(SIGABRT, &action, NULL) < 0) {
8131 		(void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
8132 		    strerror(errno));
8133 		exit(EXIT_FAILURE);
8134 	}
8135 
8136 	/*
8137 	 * Force random_get_bytes() to use /dev/urandom in order to prevent
8138 	 * ztest from needlessly depleting the system entropy pool.
8139 	 */
8140 	random_path = "/dev/urandom";
8141 	ztest_fd_rand = open(random_path, O_RDONLY | O_CLOEXEC);
8142 	ASSERT3S(ztest_fd_rand, >=, 0);
8143 
8144 	if (!fd_data_str) {
8145 		process_options(argc, argv);
8146 
8147 		setup_data_fd();
8148 		setup_hdr();
8149 		setup_data();
8150 		memcpy(ztest_shared_opts, &ztest_opts,
8151 		    sizeof (*ztest_shared_opts));
8152 	} else {
8153 		ztest_fd_data = atoi(fd_data_str);
8154 		setup_data();
8155 		memcpy(&ztest_opts, ztest_shared_opts, sizeof (ztest_opts));
8156 	}
8157 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
8158 
8159 	err = ztest_set_global_vars();
8160 	if (err != 0 && !fd_data_str) {
8161 		/* error message done by ztest_set_global_vars */
8162 		exit(EXIT_FAILURE);
8163 	} else {
8164 		/* children should not be spawned if setting gvars fails */
8165 		VERIFY3S(err, ==, 0);
8166 	}
8167 
8168 	/* Override location of zpool.cache */
8169 	VERIFY3S(asprintf((char **)&spa_config_path, "%s/zpool.cache",
8170 	    ztest_opts.zo_dir), !=, -1);
8171 
8172 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
8173 	    UMEM_NOFAIL);
8174 	zs = ztest_shared;
8175 
8176 	if (fd_data_str) {
8177 		metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
8178 		metaslab_df_alloc_threshold =
8179 		    zs->zs_metaslab_df_alloc_threshold;
8180 
8181 		if (zs->zs_do_init)
8182 			ztest_run_init();
8183 		else
8184 			ztest_run(zs);
8185 		exit(0);
8186 	}
8187 
8188 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
8189 
8190 	if (ztest_opts.zo_verbose >= 1) {
8191 		(void) printf("%"PRIu64" vdevs, %d datasets, %d threads,"
8192 		    "%d %s disks, %"PRIu64" seconds...\n\n",
8193 		    ztest_opts.zo_vdevs,
8194 		    ztest_opts.zo_datasets,
8195 		    ztest_opts.zo_threads,
8196 		    ztest_opts.zo_raid_children,
8197 		    ztest_opts.zo_raid_type,
8198 		    ztest_opts.zo_time);
8199 	}
8200 
8201 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
8202 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
8203 
8204 	zs->zs_do_init = B_TRUE;
8205 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
8206 		if (ztest_opts.zo_verbose >= 1) {
8207 			(void) printf("Executing older ztest for "
8208 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
8209 		}
8210 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
8211 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
8212 	} else {
8213 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
8214 	}
8215 	zs->zs_do_init = B_FALSE;
8216 
8217 	zs->zs_proc_start = gethrtime();
8218 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
8219 
8220 	for (f = 0; f < ZTEST_FUNCS; f++) {
8221 		zi = &ztest_info[f];
8222 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
8223 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
8224 			zc->zc_next = UINT64_MAX;
8225 		else
8226 			zc->zc_next = zs->zs_proc_start +
8227 			    ztest_random(2 * zi->zi_interval[0] + 1);
8228 	}
8229 
8230 	/*
8231 	 * Run the tests in a loop.  These tests include fault injection
8232 	 * to verify that self-healing data works, and forced crashes
8233 	 * to verify that we never lose on-disk consistency.
8234 	 */
8235 	while (gethrtime() < zs->zs_proc_stop) {
8236 		int status;
8237 		boolean_t killed;
8238 
8239 		/*
8240 		 * Initialize the workload counters for each function.
8241 		 */
8242 		for (f = 0; f < ZTEST_FUNCS; f++) {
8243 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
8244 			zc->zc_count = 0;
8245 			zc->zc_time = 0;
8246 		}
8247 
8248 		/* Set the allocation switch size */
8249 		zs->zs_metaslab_df_alloc_threshold =
8250 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
8251 
8252 		if (!hasalt || ztest_random(2) == 0) {
8253 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8254 				(void) printf("Executing newer ztest: %s\n",
8255 				    cmd);
8256 			}
8257 			newer++;
8258 			killed = exec_child(cmd, NULL, B_TRUE, &status);
8259 		} else {
8260 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8261 				(void) printf("Executing older ztest: %s\n",
8262 				    ztest_opts.zo_alt_ztest);
8263 			}
8264 			older++;
8265 			killed = exec_child(ztest_opts.zo_alt_ztest,
8266 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
8267 		}
8268 
8269 		if (killed)
8270 			kills++;
8271 		iters++;
8272 
8273 		if (ztest_opts.zo_verbose >= 1) {
8274 			hrtime_t now = gethrtime();
8275 
8276 			now = MIN(now, zs->zs_proc_stop);
8277 			print_time(zs->zs_proc_stop - now, timebuf);
8278 			nicenum(zs->zs_space, numbuf, sizeof (numbuf));
8279 
8280 			(void) printf("Pass %3d, %8s, %3"PRIu64" ENOSPC, "
8281 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
8282 			    iters,
8283 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
8284 			    zs->zs_enospc_count,
8285 			    100.0 * zs->zs_alloc / zs->zs_space,
8286 			    numbuf,
8287 			    100.0 * (now - zs->zs_proc_start) /
8288 			    (ztest_opts.zo_time * NANOSEC), timebuf);
8289 		}
8290 
8291 		if (ztest_opts.zo_verbose >= 2) {
8292 			(void) printf("\nWorkload summary:\n\n");
8293 			(void) printf("%7s %9s   %s\n",
8294 			    "Calls", "Time", "Function");
8295 			(void) printf("%7s %9s   %s\n",
8296 			    "-----", "----", "--------");
8297 			for (f = 0; f < ZTEST_FUNCS; f++) {
8298 				zi = &ztest_info[f];
8299 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
8300 				print_time(zc->zc_time, timebuf);
8301 				(void) printf("%7"PRIu64" %9s   %s\n",
8302 				    zc->zc_count, timebuf,
8303 				    zi->zi_funcname);
8304 			}
8305 			(void) printf("\n");
8306 		}
8307 
8308 		if (!ztest_opts.zo_mmp_test)
8309 			ztest_run_zdb(zs->zs_guid);
8310 	}
8311 
8312 	if (ztest_opts.zo_verbose >= 1) {
8313 		if (hasalt) {
8314 			(void) printf("%d runs of older ztest: %s\n", older,
8315 			    ztest_opts.zo_alt_ztest);
8316 			(void) printf("%d runs of newer ztest: %s\n", newer,
8317 			    cmd);
8318 		}
8319 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
8320 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
8321 	}
8322 
8323 	umem_free(cmd, MAXNAMELEN);
8324 
8325 	return (0);
8326 }
8327